Crosstalk with bone marrow adipocytes, but not osteoblasts, drives a cortical bone resorption phenotype in female mice with adult-onset deletion of the glucocorticoid receptor in Osterix-expressing cells

Abstract

Mesenchymal stem cells (MSCs) and heterogeneous bone marrow stromal cells (BMSCs) are multipotent progenitors that can differentiate into osteoblasts and bone marrow adipocytes (BMAd). The role of BMAd in skeletal homeostasis is not yet fully understood, in part due to a lack of reproducible in vitro models that faithfully mimic the biology and molecular signatures of BMAds to study their behavior. Here, we report the in vitro generation of murine BMSC- and MSC-derived Osterix-expressing BMAd-like cells via a trans-differentiation model (TD-BMAd), the development of a semi-automated analysis platform for quantification of lipid-laden cells, and the use of these models to interrogate the role of the glucocorticoid receptor (GR) in bone marrow adipose tissue (BMAT) as a regulator of osteoclastic bone resorption. The TD-BMAd cells stored intracellular lipids and robustly expressed BMAd-associated genes, such as Sp7/Osx, Pparg, and Adipoq. The pro-osteoclastogenic gene Tnfsf11/Rankl was comparably expressed between TD-BMAd and osteoblasts. We previously reported that female mice with adult-onset conditional KO (CKO) of the GR in Osx-expressing cells exhibited a low cortical bone mass and high BMAT phenotype. Here, we demonstrate that this phenotype was associated with an increase in the abundance of cortical bone osteoclasts in female but not male GR-CKO mice, but that surprisingly, GR-deficient osteoblasts from these mice did not express higher levels of pro-osteoclastogenic Rankl. Instead, TD-BMAd from female (but not male) GR-CKO mice expressed significantly higher levels of Rankl as compared to cells from GR-WT mice, suggesting that GR-deficient BMAT may be the source of the sexually dimorphic osteoclastogenic phenotype seen in vivo. Although this methodology is not intricate in nature, this simple culture technique and subsequent quantification platform may have broader utility for further studies of BMAd biology and lipid-laden cells.

Introduction

Derived from the more tightly defined skeletal stem cell population,¹ mesenchymal stem cells (MSCs) and heterogeneous bone marrow stromal cells (BMSCs) are multipotent cells that can differentiate into osteoblasts and bone marrow adipocytes (BMAd). Bone marrow adipose tissue (BMAT) is a unique fat depot that is transcriptionally distinct from other fat depots such as white and brown adipose tissue.² Our understanding of the role of BMAd in skeletal homeostasis is rapidly evolving; once thought of as an inert “filler” population in the bone marrow, recent studies demonstrate critical roles for BMAd within the skeletal niche, including serving as an important fuel source for bone cells,³ a driver of bone remodeling activity,⁴ a regulator of hematopoiesis and hematopoietic-lineage cells,^5,6 a factor in tumor metastasis to bone,^7,8 and a source of important cytokines, such as adiponectin.^9,10 Major advances in understanding of the role of BMAd and BMAT have been facilitated through the introduction of new tools, such as a BMAd-targeting Cre driver for in vivo genetic studies and lineage tracing,^3,11 and establishment of protocols that permit harvest of primary BMAd directly from bone marrow for molecular characterizations.^12,13 However, despite these recent advances, much remains to be learned regarding the role of BMAd in skeletal homeostasis, in part due to a lack of reproducible in vitro models that faithfully mimic the molecular signatures of BMAd that can be used to study their behavior in response to genetic and pharmaceutical-based manipulations.

BMAT frequently, but not always, demonstrates an inverse relationship with bone mass. It is well established that BMAT is elevated in conditions such as aging,^14–16 diabetes,¹⁷ caloric restriction,¹⁸ and sex steroid depletion,^19,20 which are associated with decreased cortical bone mass. Some controversy exists here, as depletion of BMAT has been reported both to be beneficial²¹ or to have no impact²² on age-related cortical bone loss, and likewise did not protect against ovariectomy-induced bone loss.²³ We previously reported that conditional constitutive or adult-onset deletion of the glucocorticoid receptor (GR; gene name: Nr3c1) in female mice using Osterix (Osx)-Cre resulted in a phenotype of decreased bone mass accompanied by a concomitant increase in BMAT.^24,25 Although we reported bone remodeling-based changes (eg, histology) that explained the trabecular bone phenotype in this model, we did not interrogate the cellular changes that promoted the low cortical bone mass phenotype. Here, we report that the adult-onset conditional deletion of GR in Osx-expressing cells promoted a sexually dimorphic increase in cortical bone osteoclast burden that was not explained by pro-resorptive signals from osteoblasts. Accordingly, we generated an easy and mechanistically useful tissue culture-adherent, Osx-expressing BMAd model which revealed novel insights into the phenotype of mice with skeletal deficiency of the GR. This model demonstrated increased pro-resorptive signals from female BMAd-like cells derived from mice with conditional deletion of GR in Osx-expressing cells. We also report the development of a semi-automated imaging and analysis platform, utilizing a commercially available imaging system, that facilitated unbiased quantification of lipid-containing cells such as BMAd and osteoblasts.

Materials and methods

Animal models

Research was conducted according to guidelines provided by the National Institutes of Health (NIH) under protocols approved by the Augusta University Institutional Animal Care and Use Committee (IACUC). Mice were group-housed in an accredited facility, maintained on a 12-h light/dark cycle, and fed a standard rodent diet (Teklad 2018, Envigo). For in vitro studies focused on developing the TD-BMAd tissue culture model and subsequent analysis platform, forty 6-mo-old (“young”) and 22-mo-old (“old”) C57BL/6 male and female mice (n = 10 mice/sex/age) were obtained from the NIA rodent colony (NIA). In addition, 10 12-wk-old female CD-1 mice were obtained from a commercial supplier (Envigo). All mice were provided access to food and water ad libitum while being maintained on a 12/12-h light-dark cycle in a standard temperature and humidity environment. Mice were euthanized using CO₂ asphyxiation followed by cervical dislocation, and hindlimb long bones were collected at sacrifice for isolation of primary BMSC and MSC. As we previously reported on the phenomenon of glucocorticoid-dependent osteoblastic lipid storage by BMSC-derived osteoblasts obtained from C57BL/6 mice,²⁶ the CD-1 mouse model was utilized here to test whether this phenomenon was reproducible across different mouse strains. Additional experiments with C57BL/6 mice were conducted to ensure consistency with the genetic background of the GR-CKO models and to enhance reproducibility.

For studies focused on defining the role of the GR in skeletal biology, mice with loxp sites flanking exon 2 of the GR^25,27,28 were crossed with Osx-Cre+ mice (The Jackson Laboratory #006361) for several generations, as described in our previous studies,²⁵ to generate GR-conditional knockout (GR-CKO_Osx; GRfl/fl: Osx-Cre+) mice along with WT (GR-WT; GRfl/fl: Osx-Cre−) littermates. This GR-floxed allele deletes ~50% of the mature GR protein, encompassing the translation start site and the tau 1 transactivation domain, and has previously been shown to effectively disrupt GR action.^27,28 While fully appreciating the importance of Osx-Cre+ WT control mice when using the Osx-Cre model,^29–31 we previously demonstrated that Osx-Cre+ WT control mice present with a phenotype similar to Osx-Cre− mice in this GR-floxed model,²⁴ and therefore used Cre− WT mice as our control group to remain consistent with our previous aging-related studies using this model.²⁵ Furthermore, to address known concerns with the Osx-Cre driver, including expression of Cre outside the skeleton,³⁰ we employed the Tet-OFF feature of the Osx-Cre as we described previously.²⁵ Mice were maintained on doxycycline-supplemented chow (625 mg/kg; Envigo Teklad) until 3 mo of age to suppress Cre activation, followed by standard rodent chow afterwards to induce Cre activation and genetic deletion of the GR before sacrifice after 6 mo of age. We previously demonstrated that this adult-onset GR-CKO model effectively maintains GR expression in osteoblast-lineage cells until 3 mo of age, followed by successful deletion of GR in the skeleton (but not in extra-skeletal tissues) at 6 mo of age after withdrawal of the doxycycline treatment.²⁵ The GR-CKO_Osx and GR-WT mice were euthanized under isoflurane anesthesia to minimize the confounding effects of stress, and hindlimb bones were collected at sacrifice.

Histology

Tibias from 6-mo-old GR-CKO and WT littermates (n = 7 female WT, n = 6 female GR-CKO, n = 5 male WT, and n = 8 male GR-CKO), were fixed in 10% neutral buffered formalin and decalcified in 15% EDTA prior to paraffin embedding and sectioning using a Leica Microtome (7 μm sections; Leica Biosystems). H&E and tartrate-resistant acid phosphatase (TRAP)/fast green staining procedures were used, and histological images were acquired using a ZEISS Axioscan 7 slide scanner. Cortical bone osteoclast number and surface (N.Oc/Ct.B.Pm, and Oc.Pm/Ct.B.Pm), and bone marrow adipocyte density and area fraction (N.BMAd/M.Ar and BMAd.Ar/M.Ar) were quantified with Bioquant Osteo.³¹

Osmium tetroxide micro-CT

Ten-month-old female GR-CKO and WT littermates (n = 4/genotype) were sacrificed and tibias were fixed in 10% neutral buffered formalin. Fixed bones were decalcified with 15% EDTA solution for 15 d, then stained with osmium tetroxide and embedded in 1% agarose prior to micro-CT for BMAT quantification.³² To measure BMAT volume, tibias were scanned in a Bruker SkyScan 1272 μCT instrument (Bruker MicroCT). Projection images were acquired using an image pixel size of 9.5 μm² at a camera resolution of 1224 × 820 pixels. The X-ray source was set to a voltage of 60 kVp at 166 μA. A 0.25 mm aluminum filter was used for beam hardening correction. Each specimen was rotated 180° in 0.5° steps, with 4 averaged frames acquired per step using an integration time of 390 ms. A random movement setting of 10 was applied. 386 projections were reconstructed into cross-sectional images using Bruker’s NRecon software (ver. 1.7.4.6). A software beam hardening correction of 30% and a ring artifact reduction of 3 was applied across all specimens. Reconstructions were analyzed with Bruker’s CT-Analyzer (CTAn) program for 3D morphometry. The tibial regulated BMAT (rBMAT) volume-of-interest (VOI) was selected for quantification using the proximal growth plate as a reference point. VOIs were 100 slices in length (950 μm) beginning immediately distal to the proximal tibia growth plate.

MRI scanning

Ten-month-old female GR-CKO and WT littermates (n = 3/genotype) were subjected to MRI scanning for quantification of BMAT in vivo. Mice were anesthetized with isoflurane (5% for induction and 1.5%-2.5% for maintenance) and placed in a custom cradle heated with circulating warm water. Temperature and respiration were monitored continuously using an MR compatible animal monitoring system (SA Instruments). A body coil with built-in temperature control facilitated maintenance of stable physiological conditions. MRI was performed using a Bruker Biospec 7.0 Tesla 30 cm horizontal bore scanner (Bruker Biospin MRI GmbH). Each animal was imaged in the inguinal region using the following settings: Multislice Multiecho Gradient Echo (MGE) sequence, TR = 3000 ms, TE = 40.5 ms, rear factor = 16, echo spacing = 4.5 ms, FOV = 72 × 56 mm, matrix size = 180 × 140, and slice thickness = 0.4 mm. A water-fat decomposition of the images was performed to separate fat and water signals for fat fraction (FF) quantification. For FF analysis, ImageJ³³ was used to create region-specific masks on T2-weighted images. These masks were precisely co-registered to match the corresponding locations in the fat and water fraction for each sample, ensuring accurate comparison and analysis of the same anatomical area.

Primary BMSC isolation and MSC sorting

Primary BMSC were harvested from long bones by flushing with a syringe containing basal cell culture medium that consisted of minimum essential medium (MEM)-α (Gibco #12561-072; Thermo Fisher Scientific) + 20% fetal bovine serum (FBS; Corning #35010CV) + 1% antibiotic-antimycotic (Gibco #15240-062) + 1% MEM non-essential amino acids (Gibco #11140-050). Murine MSCs derived from male and female Osx-Cre GR-CKO and WT littermates were isolated and sorted via flow cytometry as previously described using the following cell surface markers: Sca1+ (Miltenyi Biotec)/CD29+/CD44+/CD45−/CD11b− (BD Pharmingen).³⁴ The MSC were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% FBS and 1% penicillin/streptomycin (Corning #MT30002CI).

Cellular differentiation models

Primary BMSC were seeded at 4 × 10⁶ cells per well in 12-well plates, and sorted MSC were seeded at 1 × 10⁴ cells per well in 12-well plates. For osteoblastic cultures, cells were cultured in osteogenic media containing ascorbic acid (50 μg/mL; Sigma-Aldrich #A4544) and β-glycerophosphate (10 mM; Sigma-Aldrich #G9422) for 7, 14, or 21 d to induce the expression of osteogenic genes.

In an attempt to develop a tissue culture-adherent model that mimicked expression patterns of BMAd (which express both Sp7/Osterix and Adipoq/adiponectin),³ cells were initially seeded with osteogenic medium as described above to promote expression of early osteoblast genes such as Sp7/Osterix. At day 7, media were switched to adipogenic media containing rosiglitazone (1 μM; R&D systems #5325/10) to induce transdifferentiation (TD) into BMAd (TD-BMAd). At day 14, cells were either fixed in 10% neutral buffered formalin or lysed in TRIzol for harvest of RNA.

Lipid staining and quantification

To visualize and quantify lipid storage, cells were stained with 4,4-difluoro-4-bora-3a,4a-diaza-s-indacene (BODIPY; MedChemExpress #HY-W090090) or Oil Red O (ORO; Sigma-Aldrich #O1391) to detect lipids and Hoechst (1 μg; Fisher Scientific #H3570) to highlight the nucleus and permit accurate quantification of total cell number. A Cytation 5 plate reader (Agilent Biotek) was used to collect a montage of six fluorescent images at 20× original magnification in the center of each well. Gen5 software (Agilent Biotek) was used to evaluate the fluorescence intensity of the images, yielding the maximum fluorescence intensity of each channel detected and the object size range for that image. These data were used to determine a range of threshold values that the software used to quantify the image. After setting the threshold values, the Gen5 software performed quantification utilizing a primary mask approach, where the primary mask was used to count one population (total number of nuclei, approximating total cell number), and then within that population, a secondary population (lipid-containing cells) was identified. Briefly, Gen5 software quantified the number of nuclei with a DAPI channel intensity exceeding 2500 AU (AU: arbitrary units, intensity per pixel), yielding a total cell count for the corresponding image. Following the cell counting process above the primary mask threshold, the software further identified cells that stored lipids by counting those that exhibited fluorescence intensity above the minimum threshold values as specified that overlapped the Hoechst-stained nuclear signal (Figure S1). The software provided the total cell count and the count of cells positive for lipid staining (ORO+ or BODIPY+) within a timeframe of 10 s or less per image. A subset of images from the Cytation 5 were subsequently manually quantified using ImageJ for automated technique verification.

Gene expression analyses

Primary BMSC-derived osteoblasts or TD-BMAds were cultured for 14 d in the indicated treatments. Cells were washed with PBS and lysed in TRIzol. Total RNA was isolated and reverse transcribed as previously described.²⁵ Expression levels of mRNA were quantified by subjecting cDNA to real-time PCR amplification (37.5 ng cDNA per 15 μL reaction volume reaction, run in technical triplicates) using a Bio-Rad CFX Connect system and SYBR green reagent (ThermoFisher #A25780). Gene expression levels were quantified using the comparative threshold cycle (2^−ΔΔCt) method. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the internal control (housekeeping gene) for normalization. Primer sequences are listed in Table 1.

Table 1.

Primer sequences for RT-qPCR reactions.

Primer	Direction	Sequence
Gapdh	Forward	5′-GGGAAGCCCATCACCATC-3′
Gapdh	Reverse	5′-GCCTCACCCCATTTGATGTT-3′
Adipoq	Forward	5′-AAAGGAGAGCCTGGAGAAG-3′
Adipoq	Reverse	5′-CGAATGGGTACATTGGGAA-3′
Hsd11b1	Forward	5′-GCTCCCTACTCTGCAAGCAA-3′
Hsd11b1	Reverse	5′-ACACCTCGCTTTTGCGTAGA-3′
Tnfrsf11b/Opg	Forward	5′-CCAAGAGCCCAGTGTTTCTT-3′
Tnfrsf11b/Opg	Reverse	5′-CCAAGCCAGCCATTGTTAAT-3′
Sp7/Osx	Forward	5′-GGAGGTTTCACTCCATTCCA-3′
Sp7/Osx	Reverse	5′-TAGAAGGAGCAGGGGACAGA-3′
Pparg/Ppar-γ	Forward	5′-GGGTCAGCTCTTGTGAATGG-3′
Pparg/Ppar-γ	Reverse	5′-CTGATGCACTTGCCTATGAGC-3′
Tnfsf11/Rankl	Forward	5′-GCTGGGACCTGCAAATAAGT-3′
Tnfsf11/Rankl	Reverse	5′-TTGCACAGAAAACATTACACCTG-3′
Cre	Forward	5′-ACCAGCCAGCTATCAACTCG-3′
Cre	Reverse	5′-TTACATTGGTCCAGCCACC-3′

Statistical analysis

Investigators were blinded to study groups during all analyses. All in vitro experiments were performed at least three times (ie, 3 independent biological replicates). Statistical analyses for gene expression data were assessed by one-sample t-test on the delta-delta Ct values as compared to a control group and fold changes were plotted in graphs for visualization purposes. Data from in vitro studies are shown as the mean ± SEM of all biological replicates, where each biological replicate sample is indicated by a separate data point. The effects of genotype or age and condition were compared between mice using two-factor ANOVA with interaction (factor 1 = genotype or age; factor 2 = condition). When a significant interaction effect was detected, pairwise post-hoc comparisons were made between groups using Fisher’s least significant difference tests. Linear regression analysis was used to evaluate the accuracy of different fluorescence intensities for comparing the automated quantification of lipid+ cells vs manual quantification. A p-value of <.05 was considered to be statistically significant. Data from in vivo studies are represented in box plots showing median, quartiles, and outlier fences for each dataset, where outlier fences represent first quartile −1.5^*(interquartile range) and third quartile +1.5^*(interquartile range), and each data point shown represents one mouse.

Results

Female GR-CKO mice exhibit increased cortical osteoclasts and high BMAT

Our previous studies, investigating the effects of loss of GR function in Osx-expressing cells, demonstrated that GR regulates bone mass and bone marrow adipogenesis in female mice.^24,25 Although we reported bone remodeling-based changes (quantified via histomorphometry) that explained the trabecular bone phenotype in this model, we did not investigate the cellular changes that promoted the low cortical bone mass phenotype. In the current study, we further characterized loss of GR function in our GR-WT and adult-onset GR-CKO_Osx model by quantifying osteoclasts in the cortical bone compartment of TRAP-stained tibial sections. Interestingly, an interaction effect between sex and genotype was observed in cortical osteoclast surface area and number, with female (but not male) GR-CKO_Osx mice exhibiting a significant increase in both osteoclast parameters (p_{sex^*genotype} = .006, p_{sex^*genotype} = .002, respectively; Figure 1A-C). This finding suggests a potential mechanism for the decreased cortical bone area and cortical bone thickness we previously observed via micro-CT in the female GR-CKO_Osx mice from this model.²⁵

Figure 1.

Female GR-CKO_Osx mice exhibited increased cortical osteoclasts and high BMAT as compared to female WT mice. (A) Cortical bone osteoclast surface (Oc.Pm/Ct.B.Pm) and osteoclast number (N.Oc/Ct.B.Pm). (B) were quantified in sections labeled by tartrate-resistant acid phosphatase (TRAP) staining. (C) Representative TRAP-stained tibial sections from GR-WT and adult-onset GR-CKO_Osx mice at 6-mo of age. (D) Bone marrow adipocyte area fraction (BMAd.Ar/BM.Ar) was quantified in H&E-stained sections. (E) Representative H&E-stained tibial sections from GR-WT and adult-onset GR-CKO_Osx mice at 6-mo of age. (F). BMAT volume in the proximal tibia (regulated BMAT region) is presented as a fraction of marrow volume (BMAd.V/BM.V%). Representative micro-CT scans of osmium tetroxide-stained bones are shown. rBMAT, regulated BMAT; cBMAT, constitutive BMAT. (G). BMAT proton density fat fraction for rBMAT and cBMAT regions was quantified from MRI scans. Representative MRI scans are shown. Boxes show median, quartiles and outlier fences (where appropriate) for each dataset, and each data point represents one mouse. For properties where a significant interaction was detected between genotype and age, bars with different letters are significantly (p < .05) different from one another, as indicated by Fisher’s LSD post-hoc testing.

Given the increased bone marrow adiposity phenotype of GR-deficient mice in our previously studied constitutive GR-CKO_Osx model,^24,25 a high BMAT phenotype was anticipated in the adult-onset model. Histological analyses revealed that female, but not male, GR-CKO_Osx mice had a significantly greater bone marrow adipocyte area fraction when compared to WT littermates, indicated by an interaction effect of sex and genotype and post-hoc testing (p_{sex^*genotype} = .019; Figure 1D and E). Recognizing that BMAT is sensitive to a variety of factors within the bone marrow niche^35,36 and tends to show high variability when measured histologically, we also applied osmium tetroxide staining followed by micro-CT as a more precise method for measuring BMAT in the female mice. Comparable to the histological measurements, the female GR-CKO_Osx mice also demonstrated high BMAT as compared to WT littermates using this technique (p_Genotype = .044; Figure 1F). Finally, a subset of female GR-CKO_Osx mice and their WT littermates underwent MRI scanning to quantify BMAT in vivo. In agreement with both the histological and osmium tetroxide methods, GR-CKO_Osx mice exhibited a significant increase in rBMAT, but not cBMAT, of the proximal tibia as measured by MRI (p_interaction = .01; Figure 1G). Together, these findings demonstrate that 3 independent methodologies; histology, osmium tetroxide micro-CT, and MRI, consistently identified increased BMAT in GR-CKO_Osx mice.

GR-deficient osteoblasts produced the same magnitude of pro-osteoclastogenic signals as WT osteoblasts

As the Osx-Cre driver is typically utilized to target osteoblast progenitor cells, we initially hypothesized that osteoblasts derived from our adult-onset GR-CKO model would demonstrate pro-osteoclastogenic signaling patterns that would explain the high abundance of cortical bone osteoclasts in female GR-CKO mice. Surprisingly, we did not detect statistically significant changes in the expression of Tnfsf11/Rankl or Tnfrsf11b/Osteoprotegerin (Opg) in GR-CKO_Osx as compared to WT BMSC-derived osteoblasts isolated from either male or female mice after 7, 14, or 21 d in osteogenic culture conditions (Figure 2).

Figure 2.

GR-deficient osteoblasts did not differentially express pro-osteoclastogenic signals as compared to WT cells. RNA was harvested from BMSC-derived osteoblasts after 7, 14, or 21 d of culture in osteogenic medium, after which the mRNA expression levels of (A) Tnfsf11/Rankl or (B) Tnfrsf11b/Opg were quantified via qPCR. Means ± SE are shown. Each data point represents one biological replicate (ie, cells isolated from different groups of mice). p-values from one sample t-tests comparing GR-CKO_Osx to GR-WT groups for each sex at each time point of culture are shown.

Development of a platform for intracellular lipid droplet quantification

The lack of a pro-osteoclastogenic signaling phenotype in female GR-CKO osteoblasts despite the increased numbers of cortical bone osteoclasts led us to begin investigating potential contributions of BMAds in this model. We first sought to establish a robust and unbiased methodology for quantification of intracellular lipid abundance. In our previous studies of age-related changes in the GR-deficient mouse model, we detected high intracellular lipid burden in GR-deficient osteoblasts as visualized by Oil Red O (ORO), a dye that stains neutral lipids, cholesterol esters, and lipoproteins.²⁵ In these previous studies, lipid-containing cells were quantified in a blinded fashion manually with a handheld cell counter or with the assistance of a software package such as ImageJ. While effective for demonstrating differences between GR-CKO_Osx and WT cells,²⁵ the imaging and manual quantification steps were labor-intensive and introduced a source of intra- and inter-observer variability. To mitigate these limitations, we developed a semi-automated imaging and quantification procedure using a Cytation 5 plate reader and Gen5 software to measure total cell count, fluorescence intensity, and the relative fraction of lipid positive (ORO+ or BODIPY+) cells within a total cell population (Figure 3A).

Figure 3.

Validation of automated cell quantification. (A) BMSCs were isolated from 12-wk-old CD-1 mice (n = 3 biological replicates per group), seeded, and treated with dexamethasone for 14 d. At day 14, cells were fixed, stained with Oil Red O (ORO) and counterstained with Hoechst. Quantification was performed using a Cytation 5 plate reader and Gen5 software. (B) Linear regression model evaluating the impact of different fluorescence intensities for comparing automated quantification of ORO positive cells vs manual quantification. (C) Manual and automated total cell counts of osteogenic cultures treated with, and without dexamethasone. There were no statistically significant differences between the quantification methods. (D) Manual and automated ratio of ORO positive cells to total cells. Both methods were able to detect glucocorticoid-induced lipid storage. (E and F) GR-CKO BMSC-derived osteoblasts had increased lipid storage compared to osteoblasts derived from WT littermates (n = 3 biological replicates per genotype and sex). Bar charts show group mean ± SE. Each data point represents one biological replicate. p-values for variables assessed by 2-factor analysis of variance (ANOVA), Fisher’s LSD post-hoc testing, and t-tests are shown.

To establish this method, in vitro BMSC cultures from 12-wk-old CD-1 mice were differentiated in osteogenic media and treated with the GR ligand dexamethasone (Dex) to induce lipid storage in osteoblastic cells as, previously reported by our laboratory for C57BL/6-derived mice.²⁵ The amount of fluorescent signal emitted by ORO in response to excitation per pixel, or fluorescence intensity, was used to detect lipids. To determine an appropriate fluorescence intensity threshold, the relative abundance of lipid-containing cells in each condition was manually quantified using our previously established methodology,²⁵ and values ranging from 1000 to 3000 AU (Arbitrary Units, intensity per pixel) were tested in a linear regression model to evaluate the different intensity threshold of the ORO+ cells vs manual quantification (Figure 3B). The fluorescence intensity threshold of 1500 AU had the highest R² (R² = 0.828) value when regressing manual quantification against automated quantification and was, therefore, chosen for later analyses. We compared results obtained from automated quantification (at a fluorescence intensity threshold of 1500 AU) to manual quantification and found that there was no statistical difference in total cell counts, ORO+ cells, or the relative fraction of ORO+ cells (Figure 3C and D). Additionally, both quantification methods identified a significant increase in osteoblastic lipid storage in dexamethasone-treated BMSCs, consistent with our published findings^25,26 (Figure 3D).

After establishing this methodology, we isolated BMSC from 6-mo-old GR-WT and adult onset GR-CKO_Osx littermate mice. Cells were cultured in osteogenic media for 14 d, fixed and stained with ORO, and then imaged and analyzed with the cell quantification platform (Figure 3E and F). Consistent with our earlier findings in female cells,²⁵ both female and male GR-CKO BMSC-derived osteoblast cultures demonstrated a greater abundance of lipid-storing cells as compared to cultures from WT mice (p = .042 and p = .027, respectively; Figure 3F).

Creation and validation of an in vitro tissue culture adherent BMAd model

As mentioned above, the lack of a pro-osteoclastogenic signaling phenotype in female GR-CKO_Osx osteoblasts despite their increased burden of cortical bone osteoclasts, led us to begin investigating potential contributions of BMAds in this model. We sought to test whether BMAd,³⁰ known to be targeted by Osx-Cre,^11,37 contributed to the regulation of cortical osteoclast activity in a GR-dependent manner. Initial attempts to isolate and culture primary BMAd from the long bones of the mice were unsuccessful as the cells were fragile and non-adherent in nature (data not shown). This difficulty led us to develop an alternative approach for studying BMAd in vitro.

To first establish this model in WT cells, BMSC were isolated from 6-mo-old or 22-mo-old C57Bl/6 mice and were cultured in osteogenic media for 7 d. On day 7, cells were either switched to adipogenic media until day 14 of culture to induce BMAd TD or were allowed to remain in osteogenic media until day 14 of culture to continue osteoblastic differentiation (Figure 4A). On day 14, cells were fixed, stained with Hoechst, and counterstained with the fluorescent lipid droplet dye BODIPY 493/503. While no differences were observed with age (Figure 4B; p > .253). Staining analysis showed that TD-BMAd cultures had more abundant lipid droplets than osteoblast cultures, and automated quantification analysis using the Cytation 5/Gen5 software approach showed that both female and male TD-BMAd had a significantly higher ratio of BODIPY-positive cells to total cells as compared to osteoblasts (p_condition = .004 and p_condition = .018, respectively, Figure 4B).

Figure 4.

Creation and validation of in vitro TD-BMAd model. (A) BMSC were isolated from 6- and 21-mo-old male and female C57BL/6 mice and cultured in osteogenic media for 7 d. On day 7, cells in the plate designated for trans-differentiation into BMAd were cultured in adipogenic media until day 14, whereas osteoblasts remained in osteogenic media until day 14. (B) Representative images and ratio of BODIPY-positive cells via automated quantification of BODIPY 493/503 and Hoechst-stained osteoblast and TD-BMAd cultures. Analysis showed that TD-BMAd of both sexes had significantly higher ratios of BODIPY-positive cells to total cells. (C) mRNA expression of Sp7/Osx, PPAR-γ, Adipoq, Hsd11b1, Tnfsf11/Rankl, and Tnfrsf11b/Opg in osteoblast and TD-BMAd cultures were analyzed via RT-qPCR. Bar charts show group mean ± SE. Each data point represents one biological replicate (ie, cells isolated from different groups of mice). p-values assessed by one sample t-tests are shown.

We next investigated whether cells in the TD-BMAd cultures expressed genes that reflect the transcriptional profile of BMAd. Gene expression analysis showed that Sp7/Osx and the hallmark adipocyte genes Pparg/Ppar-gamma and Adipoq/adiponectin were robustly expressed in TD-BMAd cultures. The TD-BMAd expressed lower levels of Sp7/Osx as compared to osteoblasts in female cultures (p = .001 for young, p = .029 for old), but were not significantly different between TD-BMAd and osteoblasts in male cultures (p = .648 for young, p = .302 for old; Figure 4C). TD-BMAd cultures expressed significantly higher levels of Pparg and Adipoq as compared to osteoblasts across age and sex (Figure 4C). Expression levels of the glucocorticoid-activating enzyme Hsd11b1 (11-beta-hydroxysteroid dehydrogenase type 1) were elevated in young and old male TD-BMAd (p = .033, p = .041 respectively) and aged female TD-BMAd cultures (p = .026; Figure 4C) as compared to osteoblasts. As both osteoblast-lineage cells³⁸ and BMAd³⁹ have been reported to represent a critical source of the pro-osteoclastogenic signaling molecule Rankl, we quantified gene expression levels of Tnfsf11/Rankl and Tnfrsf11b/Opg in the cell cultures. Tnfsf11/Rankl was robustly expressed, and its expression did not change across age or culture condition in either sex (p > .253; Figure 4C). Expression of Tnfrsf11b/Opg, in contrast, was significantly lower in aged female TD-BMAd as compared to aged female osteoblast cultures (p = .029; Figure 4C). Together, these data suggest that TD-BMAds are robust sources of pro-osteoclastogenic signals.

Female GR-CKO TD-BMAd express pro-osteoclastogenic factors

Bone marrow stromal cell isolated from 6-mo-old male and female adult onset Osx-Cre-mediated GR-CKO and GR-WT mice were magnetically sorted to isolate a homogenous population of MSC capable of differentiating into osteoblasts and BMAd (Figure 5A). Osteoblast and TD-BMAd cultures were maintained for 14 d, and endpoints were measured as described above. Both osteoblast and TD-BMAd cultures differentiated from the MSC derived from GR-CKO mice robustly expressed Cre (Figure S2), confirming the efficacy of the in vitro models. Similar to the cultures derived from more heterogeneous BMSC, automated quantification analysis confirmed that female TD-BMAd cultures had a significantly higher fraction of BODIPY-positive cells as compared to female osteoblasts (p_condition = .001), although male TD-BMAds did not show a difference in lipid storage as compared to male osteoblasts (Figure 5B and C). Additionally, the TD-BMAd and osteoblast cultures from female GR-CKO mice had increased lipid-storing cell abundance compared to female GR-WT cultures (p_genotype = .005, Figure 5C), although this trend was also absent in males (p_genotype = .190, Figure 5C). No statistically significant interactions between genotype and culture condition were observed (p_interaction ≥ .104; Figure 5C).

Figure 5.

Glucocorticoid receptor (GR) deficient TD-BMAd stored more lipids than osteoblasts (A) BMSCs isolated from 6-mo-old female and male GR-WT and adult-onset GR-CKO mice were magnetically sorted to isolate the MSC population. Cells were cultured in osteogenic media for 7 d. On day 7, cells designated for transdifferentiation (TD) into BMAd were cultured in adipogenic media until day 14, whereas osteoblasts remained in osteogenic media until day 14. (B) Representative images of BODIPY 493/503 and Hoechst-stained osteoblast and TD-BMAd cultures. (C) Ratio of BODIPY positive cells via automated quantification of BODIPY 493/503- and Hoechst-stained osteoblast and TD-BMAd cultures. Analysis showed female TD-BMAd cultures had a significantly higher ratio of BODIPY-positive cells to total cells. Bar charts show group mean ± SE. Each data point represents one biological replicate experiment. p-values for variables and interactions assessed by two-factor analysis of variance (ANOVA) are shown.

While the MSC-derived GR-WT and GR-CKO osteoblast and TD-BMAd cultures both expressed Sp7/Osx, the female GR-CKO TD-BMAd interestingly expressed greater levels of Sp7/Osx as compared to GR-WT TD-BMAds (p = .034, Figure 6A). Female GR-CKO TD-BMAd also had significantly increased expression of Adipoq compared to female GR-WT TD-BMAd (p < .0001, Figure 6A). In contrast, expression levels of Sp7/Osx and Adipoq were not significantly affected by genotype in the cultures derived from male mice, although male GR-CKO TD-BMAd tended to have decreased expression of Adipoq compared to osteoblasts (p = .063, Figure 6A). Critically, the expression of Tnfsf11/Rankl was significantly greater in female GR-CKO TD-BMAd compared to GR-WT TD-BMAd (p = .043), while in contrast the female GR-CKO osteoblasts had significantly lower expression of Tnfsf11/Rankl compared to female GR-WT osteoblasts (p = .037; Figure 6B). Tnfrsf11b/Opg expression was not different between GR-CKO and GR-WT cultures for either osteoblasts or TD-BMAds from either sex (Figure 6B). These data support the hypothesis that female GR-CKO TD-BMAds, and not GR-CKO osteoblasts, may drive a sexually dimorphic osteoclastogenesis and bone resorption phenotype causing decreased cortical bone mass in GR-CKO_Osx female mice.

Figure 6.

Female GR-CKO_Osx TD-BMAd might support osteoclastogenesis by expressing higher Rankl. (A) The mRNA expression levels of Sp7/Osx and Adipoq and (B) Tnfsf11/Rankl, and Tnfrsf11b/Opg in osteoblast and TD-BMAd cultures were analyzed via RT-qPCR. Bar charts show group mean ± SE. Inset images in panels A and B are provided to better show expression levels of biological replicate experiments from datasets with a wide spread in data values. Each data point represents one biological replicate. p-values assessed by one-sample t-test are shown.

Discussion

A wealth of protocols have recently been reported to facilitate in-depth study of BMAT⁴⁰ and BMAd,¹² including protocols to isolate BMAd from human⁴¹ and rodent⁴² tissues. While useful for endpoint analyses, such as RNAseq and lipidomics, these direct-isolation protocols yield fragile, non-adherent cell populations which are difficult to further manipulate or visualize.¹³ To that end, we sought to establish a straightforward model with which to study tissue culture-adherent BMAd in our Osx-Cre:GR-CKO model. Such a model necessitated robust expression of the Sp7/Osx gene to induce functional recombination of the floxed GR site, as the Osx-Cre driver is operated under the control of the endogenous mouse Osx locus.⁴³ Subjecting either heterogeneous BMSC or more homogenous MSC to osteogenic culture for 7 d, followed by adipogenic culture for 7 d, produced a cell population that robustly expressed both Sp7/Osx and Adipoq/adiponectin while also promoting abundant intracellular lipid storage. Moreover, the establishment of a semi-automated and unbiased analysis platform to quantify the relative abundance of these lipid-storing cells increased the throughput for analyses while also faithfully reproducing genotype-related trends previously obtained via manual quantification.²⁵ Therefore, we believe this simple to perform culture technique and subsequent quantification analysis will have broader utility for the study of BMAd biology and lipid-laden cells.

We were initially surprised to discover that the low cortical bone mass phenotype of female Osx-Cre:GR-CKO mice was associated with increased abundance of cortical bone osteoclasts, as the trabecular bone phenotype in this model appeared to be driven by decreased osteoblastic bone formation.^24,25 In trabecular bone, Osx-Cre-mediated constitutive deletion of GR in adult female mice (~6-mo-old) was associated with a decrease in bone mineralizing surface (MS/BS, %) and bone formation rate (BFR/BS, microns/day), whereas trabecular osteoclast surface (Oc.S/BS, %) and osteoclast number (N.Oc/BS, #/mm) also tended to be reduced (rather than increased).²⁴ Serum markers of bone formation (P1NP) and bone resorption (TRAcP5b) were likewise reduced in these GR-CKO as compared to GR-WT mice.²⁴ Similarly, trabecular bone mineralizing surface was reduced in 3- and 21-mo-old female GR-CKO_Osx as compared to GR-WT mice, whereas osteoclast-related metrics were unchanged.²⁵ We were, therefore, surprised to observe increased cortical bone osteoclast abundance in the adult-onset female GR-CKO mice, and even more so to find that GR-CKO osteoblasts did not show significant changes in expression of the pro-osteoclastogenic gene Rankl that could drive this phenotype, as the Osx-Cre driver does not target osteoclasts.³⁰ Expression of Rankl is required for osteoclastogenesis.⁴⁴ While several reports have shown that Rankl is expressed by a variety of osteoblast-lineage cells from uncommitted mesenchymal progenitors⁴⁵ to fully differentiated osteocytes,³⁸ it is intriguing that older literature showed that ablation of osteoblasts did not negatively affect osteoclastic bone resorption.⁴⁶ This finding is consistent with recent publications highlighting that a BMAd progenitor population, marrow adipogenic lineage progenitor (MALP) cells, highly express Rankl⁴ and may represent a major regulator of bone resorption activity in adult mice,⁴⁷ although phenotypes attributed to MALP Rankl expression were largely confined to trabecular rather than cortical bone. While MALP cells may be a predominant source of Rankl in the bone niche, the broader population of mesenchymal-lineage cells also robustly expresses the pro-osteoclastogenic molecule colony stimulating factor 1 (Csf1)⁴⁷; future studies should investigate the potential contributions of osteoblast-lineage Csf-1, and its regulation by GR-mediated signaling, in the cortical and trabecular bone phenotypes of GR-deficient mice.

Bone marrow adipose tissue shows high glucose uptake and high rates of de novo lipogenesis despite being fairly insulin resistant.² Although glucocorticoid treatment promotes BMAT accumulation,^2,48 acute glucocorticoid treatment did not stimulate glucose uptake by BMAT in the human appendicular skeleton.² It is interesting, therefore, that deficiency in GR induced by Osx-Cre would promote BMAT expansion. In contrast to our findings, targeted deletion of GR (floxed exon 3) with BMAd-Cre (Adiponectin-Cre:Osterix-FLPo) in mice that were between 7 and 9 mo of age surprisingly did not alter BMAT volume in the tibia or caudal vertebrae, and also did not impact cortical bone in the mid-cortical region of the tibia.⁴⁹ Female BMAd-Cre: GR-CKO mice also surprisingly exhibited a small but significant increase in both trabecular bone of the proximal tibia and cortical bone volume in the distal tibia, whereas no differences were seen in males,⁴⁹ in contrast to our finding of decreased cortical mass in female mice following conditional deletion of the GR with Osx-Cre. At present, neither our previous nor current findings can explain the discrepancy in these models, which may have been influenced by the GR floxed model used (exons 1c and 2 in our Osx-Cre model vs exon 3 in the BMAd-Cre model) or differences attributable to the Cre drivers themselves. We note that an earlier publication,⁵⁰ in which GR was conditionally deleted via a Runx2-Cre driver, also produced a phenotype of low trabecular bone mass (comparable to the impact of GR deletion with Osx-Cre²⁵), although neither cortical bone nor BMAT phenotypes were reported in this model.⁵⁰ When co-cultured with osteoclast progenitor cells, primary osteoblasts derived from these Runx2-Cre:GR-CKO mice produced fewer, rather than more, tartrate-resistant acid phosphatase-positive cells.⁵⁰ While further work is needed to fully elucidate the role of GR in these cell populations, it is clear from studies using Runx2-Cre (which targets osteoblast-lineage progenitor cells), Osx-Cre (which targets osteoblast-lineage progenitor cells and BMAd progenitor cells) and BMAd-Cre (which targets MALP and mature BMAd) that GR is a critical regulator of bone cell activity in the marrow niche.

We are clearly not the first to use a TD (also known as lineage reprogramming) model to study MSC-derived adipocytes.⁵¹ This approach has been used in the past to interrogate MSC fate decisions between adipocytic and osteogenic lineages (as extensively reviewed in⁵²). For example, more than 20 yr ago, Song and Tuan cultured human MSC in osteogenic medium for up to 30 d, followed by trypsinization and culture in adipogenic medium for an additional 21 d; these studies demonstrated that even fully differentiated osteoblasts retained the ability to trans-differentiate into lipid-laden adipocytes, and vice versa.⁵¹ More recently, co-culture experiments between human BMSC differentiated into osteoblast or adipocyte lineages demonstrated the ability of osteoblasts to trans-differentiate into an adipocyte lineage upon co-culture with adipocytes,⁵³ and suggested that this process may be mediated via signaling through the PI3K-AKT, JAK2-STAT3, or SMAD pathways.⁵⁴ The current studies, in contrast, were not designed to test BMSC or MSC fate decisions between downstream lineages; rather, the goal of our studies described here was to produce an in vitro, tissue culture-adherent cell population that could be used to model the biology of BMAd and the potential contributions of these cells to skeletal phenotypes observed in vivo. Future studies, ranging from simple time course studies of osteoblast as compared to BMAd maturation to more modern techniques already being applied such as single cell RNAseq, single nucleus RNAseq, and spatial transcriptomics⁴⁷ will undoubtedly advance the field’s understanding of the contributions of BMAd to skeletal homeostasis and metabolic adaptation. A limitation of the current studies, beyond the simplistic nature of our in vitro TD-BMAd model, is the relatively low sample sizes used for our in vivo studies. When combined with the inherent biological variability of bone and BMAT phenotypes, we recognize that we likely had limited ability to detect subtle phenotypic changes in our mouse model. BMAT, in particular, is known to exhibit regional and age-dependent heterogeneity, which may further contribute to variability in our analyses. In addition, while our studies here included static histomorphometric assessments, we did not perform dynamic histomorphometry of cortical bone. Future studies incorporating larger sample cohorts and fluorochrome-based dynamic histomorphometry will be important to more fully define the contribution of GR signaling to compartment-specific bone remodeling and BMAd-associated skeletal phenotypes.

In conclusion, studies described here report the in vitro generation of BMSC- and MSC-derived Osterix-expressing BMAd-like cells via a trans-differentiation model (TD-BMAd), the development of a semi-automated analysis platform for quantification of lipid-laden cells, and the use of these models to interrogate the role of the GR in BMAT as a regulator of osteoclastic bone resorption. This model revealed that the pro-resorptive, low cortical bone mass phenotype observed in adult female Osx-Cre GR-CKO mice may be attributable to BMAd, as BMAT and cortical bone osteoclasts were elevated in female GR CKO mice in vivo and TD-BMAd (but not osteoblasts) derived from these female mice expressed significantly higher levels of Rankl. While not technically complex, we believe the methodology for this simple culture technique and subsequent quantification may have broader utility for the further study of BMAd biology and lipid-laden cells.

Acknowledgments

The authors would like to acknowledge the Augusta University Electron Microscopy and Histology Core facility for assistance with histological specimen preparation, the Augusta University Cell Imaging Core for assistance with specimen imaging, and the Augusta University Small Animal Imaging Core for assistance with MRI scans.

Author contributions

Jaeshia Lindsay (Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing), Husam Bensreti (Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing), Colby Gross (Investigation, Methodology, Writing—review & editing), Alok Tripathi (Data curation, Investigation, Writing—review & editing), David Maridas (Investigation, Writing—review & editing), Maribeth Johnson (Investigation, Formal analysis), Xingming Shi (Investigation, Methodology, Writing—review & editing), Wendy B. Bollag (Funding acquisition, Writing—review & editing), Carlos M. Isales (Funding acquisition, Writing—review & editing), and Meghan E. McGee-Lawrence (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing)

Funding

This study was supported by the National Institutes of Health (NIH) (U01 AG086158, P01-AG036675, S10 OD025177). W.B.B. is supported in part by a VA Research Career Scientist Award (#IK6 BX005691). The authors would like to acknowledge the Augusta University Electron Microscopy and Histology Core facility (RRID:SCR_026810) for assistance with histological specimen preparation, the Augusta University Cell Imaging Core (RRID:SCR_026799) for assistance with specimen imaging, and the Augusta University Small Animal Imaging Core (RRID:SCR_027047) for assistance with MRI scans.

Conflicts of interest

The authors state that they have no conflicts of interest.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.