Bone marrow adipose tissue is a unique adipose subtype with distinct roles in systemic glucose homeostasis

Bone marrow adipose tissue (BMAT) represents >10% of total adipose mass, yet unlike white or brown adipose tissues (WAT or BAT), its role in systemic metabolism remains unclear. Using transcriptomics, we reveal that BMAT is molecularly distinct to WAT but is not enriched for brown or beige adipocyte markers. Instead, pathway analysis indicated altered glucose metabolism and decreased insulin responsiveness in BMAT. We therefore tested these functions in mice and humans using positron emission tomography–computed tomography (PET/CT) with 18F-fluorodeoxyglucose, including establishing a new method for BMAT identification from clinical CT scans. This revealed that BMAT resists insulin- and cold-stimulated glucose uptake and is thus functionally distinct to WAT and BAT. However, BMAT displayed greater basal glucose uptake than axial bones or subcutaneous WAT, underscoring its potential to influence systemic glucose homeostasis. These PET/CT studies are the first to characterise BMAT function in vivo and identify BMAT as a distinct, major subtype of adipose tissue. HIGHLIGHTS Bone marrow adipose tissue (BMAT) is molecularly distinct to other adipose subtypes. BMAT is less insulin responsive than WAT and, unlike BAT, is not cold-responsive. Human BMAT has greater basal glucose uptake than axial bone or subcutaneous WAT. We establish a PET/CT method for BMAT localisation and functional analysis in vivo.

not enriched for brown or beige adipocyte markers. Instead, pathway analysis indicated 48 altered glucose metabolism and decreased insulin responsiveness in BMAT. We therefore 49 tested these functions in mice and humans using positron emission tomography-computed 50 tomography (PET/CT) with 18 F-fluorodeoxyglucose, including establishing a new method for 51 BMAT identification from clinical CT scans. This revealed that BMAT resists insulin-and 52 cold-stimulated glucose uptake and is thus functionally distinct to WAT and BAT. However, 53 BMAT displayed greater basal glucose uptake than axial bones or subcutaneous WAT, 54 underscoring its potential to influence systemic glucose homeostasis. These PET/CT 55 studies are the first to characterise BMAT function in vivo and identify BMAT as a distinct, 56 major subtype of adipose tissue. is transcriptionally and functionally distinct to WAT, BAT and beige adipose tissue, 132 identifying BMAT as a unique class of adipose tissue. We show that BMAT has greater basal 133 glucose uptake than WAT and establish methods for BMAT characterisation by PET/CT. 134 Together, this knowledge underscores the potential for BMAT to influence metabolic 135 homeostasis and sets a foundation for future research to reveal further roles of BMAT in 136 normal physiology and disease. 137 140 The functional hallmarks of WAT, BAT and beige adipose tissue are reflected on a molecular 141 level, with each class having distinct transcriptomic profiles and characteristic marker genes 142 (Rosell et al., 2014;Svensson et al., 2011;Wu et al., 2012). Thus, to test if BMAT has distinct 143 metabolic functions, we first compared the transcriptomes of whole BMAT and WAT from 144 two cohorts of rabbits. Principle component analysis of both cohorts identified BMAT as a 145 distinct depot compared to gonadal WAT (gWAT) and inguinal WAT (iWAT) (Fig. 1A); 146 however, BMAT from either rabbit cohort was not uniformly enriched for markers of brown 147 or beige adipocytes (Fig. 1B, S1A): although SLC27A2 was significantly higher in BMAT 148 than WAT from both cohorts, and PPARGC1A in BMAT from cohort 1, several other brown 149 and/or beige markers were more highly expressed in WAT, while most such markers were 150 not differentially expressed between BMAT and WAT in either cohort (Fig. 1B, S1A). Thus, 151 the transcriptomic distinction with WAT is not a result of BMAT being more brown-or beige-152 like. Instead, gene set enrichment analysis (GSEA) highlighted the potential for BMAT to 153 have altered glucose metabolism and decreased insulin responsiveness compared to WAT 154 ( Fig. 1C,1D, S1B).

156
To determine if similar differences occur in humans, we next analysed the transcriptomes of 157 adipocytes isolated from human femoral BMAT and subcutaneous WAT, which our previous 158 analyses revealed to be globally distinct (Mattiucci et al., 2018). Consistent with our findings 159 in rabbits, human BMAds were not enriched for brown or beige markers and had decreased 160 expression of genes relating to glucose metabolism and insulin responsiveness (Fig. 1E, 161 S2A, S2B). To further address this we next pursued targeted analysis of adipocytes isolated 162 from BM and WAT of a second cohort of subjects; we also isolated adipocytes from 163 trabecular bone (Bone Ads) to assess potential site-specific differences in BMAd function 164 (Craft et al., 2018). Adipocyte purity was confirmed histologically (data not shown) and by 165 qPCR (Fig. S2C). In situ, these BM and bone adipocytes resembled unilocular white 166 adipocytes (Fig. 1F); however, qPCR revealed significant differences in transcript 167 expression of INSR, IRS1, IRS2, SLC2A4, SLC2A1 and SLC2A3 between WAT adipocytes 168 and those from BM or bone (Fig. 1G). Notably, compared to white adipocytes, each BMAd 169 subtype had decreased SLC2A4 and increased SLC2A1 and SLC2A3, suggesting that 170 BMAds may have higher basal glucose uptake that is less insulin responsive. In contrast, 171 there were no differences in expression of UCP1, and most other brown or beige adipocyte 172 markers were not enriched in either BMAd subtype (Fig. 1G, S2D). 173 174 Taken together, these data demonstrate that BMAds in animal models and humans are 175 transcriptionally distinct to white, brown and beige adipocytes, and suggest altered roles in 176 systemic glucose homeostasis and insulin responsiveness. 179 To test the metabolic functions of BMAT in vivo we used 18 F-FDG PET/CT in mice to 180 determine if, like WAT, BMAT is insulin-responsive. As expected, insulin decreased blood 181 glucose ( Fig. 2A) and increased 18 F -FDG uptake in the heart, iWAT and gWAT (Fig. 2B, C 182 and F). To assess 18 F-FDG uptake separately within bone and BMAT, we first applied 183 thresholding to the PET/CT data to separate bone from BM based on their different tissue 184 densities (data not shown). This revealed that insulin significantly increased 18 F -FDG uptake 185 in femoral bone, whereas humoral bone uptake decreased; uptake in proximal or distal tibial 186 bone was unaffected (Fig. 2F). To assess BMAT-specific 18 F-FDG uptake we took 187 advantage of the regional differences in BMAT content around the mouse skeleton. Thus, 188 adipocytes comprise only a small percentage of total BM volume of humeri, femurs and 189 proximal tibiae, but predominate in distal tibiae (Fig. 2E). To address the contribution of 190 BMAT to skeletal 18 F-FDG uptake, we therefore quantified 18 F-FDG in a bone-region-specific 191 manner to distinguish between areas of low BMAT (humerus, femur, proximal tibia) and high 192 BMAT (distal tibia). This revealed that insulin did not significantly affect glucose uptake in 193 any of the BM regions analysed (Fig. 2F). Thus, compared to WAT, BM and BMAT resist 194 insulin-stimulated glucose uptake. 195 196 Cold exposure in mice does not induce glucose uptake or beiging in BMAT 197 To test if BMAT is BAT-or beige-like in vivo, we next analysed 18 F-FDG uptake following 198 acute or chronic cold exposure in mice (Fig. S3A). Acute (4 h) or chronic cold (72 h) 199 increased energy expenditure without causing weight loss or hypoglycaemia (Fig. S3B-D), 200 likely due to increased food consumption in chronic cold mice (Fig. S3E). BAT 18 F-FDG 201 uptake increased after either duration of cold ( Fig. 3A-C). Chronic cold also increased iWAT 202 18 F-FDG uptake, suggesting beiging of this depot (Fig. 3C). However, neither acute nor 203 chronic cold exposure increased 18 F-FDG uptake into bone or BM (Fig. 3B). Indeed, cold 204 exposure decreased 18 F-FDG uptake into distal tibial BMAT, highlighting fundamental 205 differences with iWAT and BAT. Cold exposure also decreased BAT lipid content and 206 promoted beiging of iWAT, as indicated by formation of multilocular adipocytes, but these 207 effects did not occur in BMAT (Fig. 3D). Consistent with this, cold exposure induced brown 208 and beige transcripts in BAT and iWAT, but not within bone ( Fig. 3E 216 We next tested if these distinct metabolic properties extend to BMAT in humans. First, we 217 established a method to identify BMAT from clinical PET/CT scans. To determine Hounsfield 218 Units (HU) for BMAT, we identified BMAT-rich and BMAT-deficient BM regions by magnetic 219 resonance imaging (MRI). This revealed that sternal BM is BMAT-enriched while vertebral 220 BM is BMAT-deficient (Fig. 4A); WAT was also analysed as an adipose-rich control region. 221 We then co-registered the MRI data with paired CT scans of the same subjects (Fig. 4A). 222 This revealed a distinct HU distribution for BMAT-rich sternal BM, intermediate between 223 WAT and red marrow (RM) of BMAT-deficient vertebrae (Fig. 4B).

225
Using these distinct HU distributions, we generated a receiver operating characteristic 226 (ROC) curve to identify optimal diagnostic HU thresholds to distinguish BMAT from RM ( Fig.   227   S4A). This revealed that BMAT-rich BM has HU <115, whereas RM is mostly within 115-228 300 HU (Fig. 4B); bone was defined as >300 HU. To test the validity of these thresholds we 229 applied them to clinical CT data to determine BMAT volume as % BM volume. We found 230 that BMAT predominated in the arms, legs and sternum but was markedly lower in the 231 clavicle, ribs and vertebrae (Fig. 4C, Fig. S4B). Moreover, %BMAT showed age-associated 232 increases in the axial skeleton but not in long bones (Fig. S4B) Human BMAT is functionally distinct to BAT and is a major site of basal glucose 241 uptake 242 We then applied these thresholds to human co-registered PET/CT data to assess 18 F-FDG 243 uptake in BMAT, RM and bone. To test if BMAT is BAT-like we first compared BMAT 18 F-244 FDG uptake between three groups: subjects with no detectable supraclavicular BAT at room 245 temperature (No BAT), subjects with active BAT at room temperature (Active BAT), and 246 cold-exposed subjects (16 ºC for 2 h; Cold). PET/CT confirmed BAT 18 F-FDG uptake in the 247 latter two groups but not in the No BAT group (Fig. 4D-E, Fig. S4C). Cold exposure did not 248 alter 18 F-FDG uptake in scWAT but was associated with increased uptake in sternal and 249 clavicular bone tissue; however, these were the only skeletal sites at which 18 F-FDG uptake 250 significantly differed between the No BAT, Active BAT and Cold subjects (Fig. 4E, Fig. S4D). 251 Indeed, the Active BAT and Cold subjects did not have increased glucose uptake in BMAT 252 or RM of any bones analysed (Fig. 4E, Fig. S4D). Thus, consistent with our findings in mice, 253 BMAT glucose uptake in humans is not cold-responsive.  (Ramage et al., 2016a) to determine if glucocorticoids also influence glucose 260 uptake in human BMAT. We found that prednisolone significantly influenced 18 F-FDG uptake 261 only in vertebrae, in which there was a trend for increased uptake into RM but not BMAT or 262 bone (Fig. S4E). However, prednisolone did not influence 18 F-FDG uptake at any other site. 263 Thus, unlike BAT, BMAT glucose uptake is not glucocorticoid-responsive.

265
The above findings confirm that, in humans, BMAT is functionally distinct to BAT. However, 266 while analyzing these data, two other phenomena became apparent. Firstly, within axial 267 bones of each subject, BMAT had significantly higher glucose uptake than bone (Fig. 4F). 268 In the sternum, glucose uptake was also greater in BMAT than in RM (Fig. 4F). Secondly, 269 BMAT at each skeletal site had higher glucose uptake than scWAT (Fig. 4G). Thus, despite 270 being unresponsive to insulin or activators of BAT, BMAT has high basal glucose uptake, 271 highlighting its potential to influence systemic glucose homeostasis.  (Liu et al., 2011). These data are strikingly consistent with our results for transcript 301 expression in rabbits and humans (Fig. 1, Fig. S1-2) and further support the conclusion that, 302 compared to WAT, BMAT resists insulin-stimulated glucose uptake. 303 304 In addition to BMAT, we also found that insulin responsiveness varies among different 305 bones: in insulin-treated mice, bone glucose uptake increases in femurs, decreases in 306 humeri and is unaltered in tibiae. In contrast, Zoch et al report that insulin stimulates 18 F-307 FDG uptake into whole femurs and tibiae (Zoch et al., 2016). This discrepancy may relate 308 to technical differences: Zoch et al analysed whole bones (including BM) of anesthetised 309 mice, whereas we distinguished between bone and BM and avoided anaesthesia. It is 310 unclear why insulin is associated with decreased glucose uptake in humeral bone and BM; 311 this is unlikely to be a technical issue given that we see expected insulin-stimulated glucose 312 uptake in the heart, WAT and femur. Thus, the lack of increases in humeri and tibiae 313 suggests that there are site-specific differences in skeletal insulin responsiveness.

315
Another major finding is that BMAT is molecularly and functionally distinct to brown and 316 beige adipose tissues, both for cBMAT of mice and rabbits, and for rBMAT of humans at 317 multiple skeletal sites. These molecular distinctions are consistent with several other 318 studies. We and others previously found that tibial Ucp1 expression is over 10,000-fold lower lines of evidence to the contrary, the concept that BMAT may be BAT-or beige-like has 326 persisted. Thus, our in vivo functional analyses of mice and humans are a key advance 327 because they confirm that cold exposure does not induce glucose uptake or beiging in 328 BMAT. This demonstrates, conclusively, that BMAT is not BAT-or beige-like.

330
Our glucocorticoid studies provide further insights. Unlike in BAT, acute glucocorticoid 331 treatment in humans does not stimulate glucose uptake in BMAT; however, it does influence 332 uptake across lumbar vertebrae, with a trend for increases in RM (Fig. S4E). It is notable 333 that this occurs only in vertebrae, because these are also the bones in which glucocorticoids 334 drive the greatest increases in fracture risk (Briot and Roux, 2015). This raises the possibility 335 that glucocorticoids modulate BM and bone metabolism in a site-specific manner and that Although BMAT glucose uptake is not stimulated by insulin at physiological concentrations, 342 cold exposure or glucocorticoids, a major finding is that BMAT in humans has high basal 343 glucose uptake, exceeding that of WAT and greater than that for bone or RM in the axial 344 skeleton. Superficially, this seems at odds with two studies reporting that BM 18 F-FDG In summary, this study is the first to dissect BMAT glucose metabolism in vivo and identifies 394 BMAT as a distinct, major subtype of adipose tissue.  For human subjects in cohort 1 (Fig. 1E, Supplemental Fig. 2A-B), ethical approval and 501 subject characteristics are as described previously (Mattiucci et al., 2018). For human 502 subjects in cohort 2 ( Fig. 1F-G, Supplemental Fig. 2C-D) and those undergoing MRI or 503 PET/CT (Fig. 4, Supplemental Fig. 5-6 Hounsfield Units for BMAT), and those without or with detectable BAT at room 512 temperature (Fig. 4, Supplemental Fig. 5-6 After weighing, each tissue was minced in the petri dish using a sterile scalpel and scissors,   Mouse cold-exposure studies 654 The protocol is adapted from (Wang et al., 2012), with a summary depicted in Supplemental 655 Figure 3A. For the acute and chronic cold exposure studies (Fig. 3, Supplemental Fig. 4B RPTOR  IRS2  PHKB  ADIPOQ  PRKAB1  SORT1  TBL1XR1  MAPK9  PPP1CC  PRKAR2B  PYGL  CBL  PRKAR2A  SOS1  CREBL2  TSC1  INSR  PTPN11  ARPP19  PRKACB  NUP43  CRK  NUP155  GSK3B  MTOR  MAPK1  GRB2  EXOC7  FLOT1  NUP88  PIK3R3  SLC2A6  SLC2A1  SLC2A3  RFX6  SLC2A2  PIK3R5  SLC2A12  HNF1A  DRD1  SLC2A7  G6PC  FBP2  SHC4  GCKR  AKT3  SLC2A13   FGF10  SLC2A10  PIK3R2  SLC2A8  SREBF1  PCK2  FASN  IRS1  SLC2A4  CEBPA  AACS  CAV2  RARRES2  AKT2  PCK1  PPP1CB  PPARG  PIK3R1  EIF4EBP2  PDE3B  DENND4C  FABP4  PIK3CA  SIRT1  PIK3CB  FLOT2  SESN2  PRKAG1  TSC2  POM121  MAPK3  MKNK2  CALM2  PRKAG2  TRIP10  SLC25A33  VAMP2  ARAF  ADIPOR2  INPP5K  LIPE  LEP  ACACB  SIK2  PYGB  CALM1  RAB4B Figures 1 and S1. (C,D) qPCR to validate purity of adipocytes isolated from each tissue (C) and showing that BM adipocytes generally do not have increased expression of brown or beige adipocyte markers (D). Transcript expression was normalized to expression of IPO8 (C) or RNA18SN5 (D). Data in (C) are mean ± SEM of the following numbers per group: BM Ads, n = 8 (ADIPOQ) or 9 (PPARG); BM SVF, n = 3 (ADIPOQ) or 5 (PPARG); WAT Ads, n = 10 (ADIPOQ and PPARG); WAT SVF, n = 5 (ADIPOQ and PPARG); Bone Ads and SVF, n = 7 (PPARG) or 3 (ADIPOQ). Data in (D) are mean ± SEM of the following numbers per group: BM Ads, n = 3-10; WAT Ads, n = 6-10; Bone Ads, n = 3-7. For each transcript, significant differences are indicated as for Figure 1. (B) Quantification of BMAT in CT scans of male and female subjects aged <60 or >60 years. A HU threshold of <115 was used to identify BMAT voxels in BM of the indicated bones, and total BM volume was also determined. The proportion of the BM cavity corresponding to BMAT (Ad.V/Ma.V) was then calculated. Data are shown as box-and-whisker plots of the following numbers of subjects for each group: <60 years, n = 28 (humerus), 9 (femur), or 27 (clavicle, sternum and vertebrae); >60 years, n = 7 for each bone. Significant differences between <60 and >60 groups are indicated by ** (P <0.01). (C) Representative coronal PET/CT images of No BAT, Active BAT and Cold subjects. FDG uptake in BAT is evident in the Active BAT and Cold subjects (arrows). Femurs were not included in scans of the Cold group. (D) FDG uptake in bone tissue, RM and BMAT of the indicated bones. Data are shown as mean ± SEM of 8 (No BAT) or 7 (7 Active BAT, Cold) subjects per group. Significant differences between these groups No BAT, Active BAT and Cold groups are indicated by # (P <0.05). (E) Subjects were treated with prednisolone or placebo control prior to analysis of FDG uptake by PET/CT. Data are shown as paired individual values for each subject. For each skeletal site, the influence of treatment or tissue (bone, RM, BMAT), and interactions between these, were determined by 2-way ANOVA; P values are shown beneath the graph.