Myeloma-Modified Adipocytes Exhibit Metabolic Dysfunction and a Senescence-Associated Secretory Phenotype

Multiple myeloma (MM) is a malignant B-cell neoplasm characterized by uncontrolled growth of mutated plasma cells within the bone marrow leading to disturbed bone and marrow homeostasis. Overt (symptomatic) myeloma disconnects the normal equilibrium between osteoblastic (bone building) and osteoclastic (bone resorbing) activities leading to net bone loss, osteolysis, and pathologic fractures. Properties of the bone marrow allow for migration and proliferation of multiple myeloma cells, while also providing an environment for quiescent, drug-resistant multiple myeloma cells (1). Recently, targeting this tumor-supportive microenvironment, rather than the cancer cells directly, has proven to be an effective and innovative therapeutic strategy to inhibit tumor growth and osteolysis. The effectiveness of this approach is based, in part, on the pathologic manipulation (i.e., hijacking) of the microenvironment by multiple myeloma cells to create an even more tumor-supportive environment via a positive feedback system (2, 3).

The pathogenesis of multiple myeloma involves complex, bidirectional interactions of multiple myeloma cells with bone marrow resident cells, including osteoblasts, osteoclasts, osteocytes, mesenchymal stem/stromal cells (MSC), and bone marrow adipocytes (BMAd; refs. 4, 5). BMAds have recently been shown to contribute to systemic metabolism via enhanced circulating adipokines (6), and can regulate both bone remodeling, via the production of RANKL (7), and hematopoiesis, via stem cell factor production (8). Interestingly, bone marrow adipose tissue (BMAT) expands with aging and obesity, two key risk factors for myeloma (9), suggesting that BMAds may contribute to multiple myeloma disease and progression.

Previously, we demonstrated that myeloma-associated adipocytes exhibit reduced adipogenic gene expression and lipid loss (delipidation; ref. 10), and that adipocytes support multiple myeloma cell growth (11). Indeed, adipocyte-derived factors, such as MCP-1/CCL2 and SDF1α/CXCL12, are chemotactic factors for myeloma cells (4, 12), while other adipokines promote myeloma proliferation (e.g., leptin/LEP; ref. 13) and resistance to chemotherapies (e.g., leptin/LEP and adipsin/CFD; ref. 5). In 2019, two main studies examined the relationship between BMAds and myeloma cells, demonstrating that multiple myeloma–reprogrammed BMAds contribute to myeloma-induced bone disease (14) and that adipocytes are modulated by multiple myeloma cells in terms of their lipid droplet and adipocyte marker gene expression (15).

Despite these publications, the characteristics of these “multiple myeloma–modified” adipocytes (MM-adipocyte) remain relatively unexplored and their ability to feedback on tumor cells has not been systematically characterized. In this study, we hypothesized that MM-adipocytes regulated multiple myeloma growth and progression through a secretory profile consistent with senescence. We report an in-depth analysis of MM-adipocytes and detail the ways in which BMAT is altered during myeloma disease progression and treatment, highlighting their unique, tumor-supportive phenotypes. Our findings suggest that targeting these MM-adipocytes, or factors derived from them, may represent a new path forward for multiple myeloma therapy.

The control, MGUS, and multiple myeloma cohorts used in this study have been described previously (16, 17). All biopsies were collected from Danish pathologic biobanks according to the Declaration of Helsinki and with approvals by The Regional Scientific Ethical Committees for Southern Denmark (ID S-20070121, S-20110155, and S-20190110). According to these approvals, we were allowed to include diagnostic 3-mm trephine needle bone marrow biopsies from control, MGUS, and patients with multiple myeloma in Danish pathologic biobanks without their written informed consent, as long as they had not declined the use of their tissues for research by registering on the National Tissue Application Register. Adjacent 3.5 μm sections from the biopsies were Masson trichrome stained or immunostained for CD138 as described previously (16).

Adipocyte volume per marrow volume (AV/MV) was estimated as the percentage of points hitting adipocytes versus not in the marrow cavity using the point grid. An average of 691 (range, 579–1,276) points were evaluated within each biopsy. The adipocyte density was estimated as the number of adipocyte profiles per marrow area (#/mm²) within the boxes of the box grid, and the mean adipocyte profile size (mm²) was estimated by measuring the area of the adipocytes profiled within the boxes of the box grid. The tumor load (multiple myeloma tumor area per marrow area, MM.Ar/M.Ar) was estimated as the percentage of points hitting CD138⁺ myeloma cells versus not in the marrow cavity using the point grid. Quantitation was performed blinded to the clinical data.

All experimental studies and procedures involving mice were performed in accordance with approved protocols from the Garvan Institute/St. Vincent's Hospital Animal Ethics Committee ARA 14/03 (Sydney, New South Wales, Australia) and the Maine Medical Center Research Institute's (Scarborough, Maine) Institutional Animal Care and Use Committee. All mice were weaned at 21 days after birth and fed a diet of normal chow and autoclaved water. Six-week-old C57BL/KaLwRiJHsd (BKAL) male mice were injected with vehicle or 5TGM1 cells as described previously (2). In the MM.1S study, 12-week-old female Fox Chase SCID-Beige Mice (Charles River Laboratories; n = 3–6) were inoculated via tail vein injections with 5 × 10⁶ MM.1S^gfp+luc+ cells. “Survival endpoints” were euthanasia following multiple myeloma–induced hind limb paralysis; mice from a predetermined endpoint (21 days following tumor cell inoculation) were used for measuring whole-marrow gene expression (n = 6). Tumor burden was measured in the MM.1S studies weekly or biweekly utilizing in vivo bioluminescence imaging in an IVIS Lumina LT (PerkinElmer) as described previously (18). Following sacrifice, mouse tibia and femora were fixed and stained with hematoxylin and eosin (H&E) prior to sectioning and imaging as described previously (19).

Mouse 3T3-L1 preadipocytes, as well as mesenchymal stromal cells (MSC) derived from human or mouse bone marrow, or mouse external ears, were cultured and differentiated into adipocytes as described previously (19). MM.1S and 5TGM1 cells were cultured as described previously (20); OPM-2 and RPMI-8226 cells were similarly grown in RPMI1640 basal medium supplemented with 10% FBS (VWR) and 1% penicillin–streptomycin (VWR). For transwell coculture experiments, preadipocytes or MSCs were seeded into 24- or 6-well plates prior to adipogenic differentiation. Following differentiation, myeloma cells were seeded above mature adipocytes on 0.4-μm transwell membranes (Corning). Lipid droplets from adipocytes in vitro were labeled with Oil Red O and analyzed (19). For beta-galactosidase (β-gal) staining of senescent cells, differentiated human BMAT samples were exposed to multiple myeloma cells for 72 hours, incubated in normal conditions for 10 days, and then fixed and stained for β-gal using the Senescence β-galactosidase Staining Kit (Cell Signaling Technology; ref. 21). For characterization of myeloma cell phenotypes, cells were collected and stained with APC-Annexin V (BioLegend) and DAPI (Thermo Fisher Scientific) for apoptosis. For proliferation and cell cycle, cells were fixed in 70% ethanol prior to washing and staining with Alexa Fluor 647 anti-human Ki-67 antibody (BioLegend) and DAPI (0.5 μg/mL), respectively, prior to flow cytometry via MACSQuant Analyzer (Miltenyi Biotec). A minimum of 10,000 events were captured and analyzed using FlowJo v.10 (Becton, Dickinson & Company). To compare the effects of multiple myeloma–associated BMAT (MM-BMAT) and healthy BMAT–derived soluble factors on myeloma cells, conditioned media (CM) were collected from the same BMAT donors (n = 7) after 72-hour transwell coculture experiments (as described above) with/without tumor cells and used with 50/50 fresh media for experiments. For senolytic experiments, BMAT was treated with 500 nmol/L dasatinib (Selleck Chemical, S1021) and 100 μmol/L quercetin (Selleck Chemical, S2391) for 48 hours approximately 10 days following irradiation as described previously (22). BMAT was washed prior to seeding and collection of conditioned media (48-hour incubation period). Four BMAT donors were used for irradiation experiments, with control, nonirradiation plates paired with irradiation plates for each donor.

Total RNA was harvested in QIAzol and prepared via Qiagen miRNEASY Kit with DNase On-column Digestion (Qiagen) according to the manufacturer's protocol. RNA was quantified and tested for quality and contamination using NanoDrop 2000 (Thermo Fisher Scientific) and subjected to quality control minimum standards of 260/230 > 2 and 260/280 > 1.8 prior to downstream applications. Quantitative PCR (qPCR) was executed as described previously (19).

Total RNA (100 ng) from 3T3-L1 cells was used for cRNA synthesis, prepared, and purified as described previously (20). Briefly, 5.5 μg of fragmented single-strand cDNA (GeneChip WT PLUS Reagent Kit) was purified, labeled, and hybridized prior to injection into Mouse Clariom S arrays. Arrays were placed in the Affymetrix GeneChip Hybridization Oven 645 and stained with the Affymetrix GeneChip Fluidics Station 450 prior to scanning (7G Affymetrix GeneChip Scanner 3000). Raw data (Affymetrix CEL files) were imported into the Gene Expression Workflow (Partek Genomics Suite v. 6.17.0918, Partek; GSE143269) and normalized prior to log₂ transformation and differential expression (DE) analysis, as described previously (20), except here one-way ANOVA was utilized in the DE analysis and DE genes were defined on the basis of an absolute fold change (FC) > 1.5 in combination with an unadjusted P ≤ 0.05.

Cell culture conditioned media were collected from mature-naïve adipocytes or MM-adipocytes in culture and frozen at −20°C. Secreted cytokines in the conditioned media were quantified with either the Mouse Adipokine Array (R&D System) or the Human Cytokine Array (R&D System) per the manufacturer's instructions. 3T3-L1 IL6 protein secretion was also measured using a mouse-specific IL6 ELISA (R&D System) with or without MM.1S coculture [direct or indirect (using a 0.4-μm transwell)].

A Seahorse XF cell mito stress test kit was run utilizing a Seahorse XFe96 Analyzer (Agilent Technologies). 3T3-L1 cells were seeded in a 96-well Seahorse XF cell culture microplate at density of 5 × 10³ cells per well and adipogenesis was induced. At day 7 of differentiation, cells were transferred to maintenance media with either 50% MM-CM or basal media for 3 days, and on day 10, the test was run according to the manufacturer's instructions.

High-throughput RNA sequencing (RNA-seq) was performed on human BMAT, differentiated from human MSCs from three donors and three human multiple myeloma cell lines. BMAT from each donor was either cultured alone or exposed to one of the three human multiple myeloma cell lines via transwell coculture for 72 hours. Similarly, each multiple myeloma cell line was either cultured alone or exposed to one of three human BMAT from donors prior to RNA collection and purification as outlined above. RNA libraries were prepared and high-throughput read mapping and bioinformatics analyses were executed as described previously (GSE140374; ref. 23).

Gene expression data were downloaded from the Gene Expression Omnibus (GEO; GSE6477; ref. 24), log-transformed, and analyzed with an one-way ANOVA model using the aov() function in R. Heatmaps were generated using the heatmap.2() function in R.

Data generated from RNA-seq were used to validate the MM.1S, OPM-2, and RPMI-8226 cells (Table 1). Base calling and variant determination were performed using Illumina's RTA version 1.17.21.3 and the MAPRSeq bioinformatics pipeline (http://bioinformaticstools.mayo.edu/research/maprseq) as described previously (23). Variants were compared with RNA-seq variants present in the Cancer Cell Line Encyclopedia (CCLE; portals.broadinstitute.org/ccle; ref. 25) utilizing publicly available tools from Galaxy (use.galaxy.org). Known translocations were confirmed using gene fusion predictions resulting from the MAPRSeq pipeline, and also compared against CCLE gene fusion predictions. 3T3-L1 cells were from the ATCC (www.atcc.org). 5TGM1 cells have not been validated at this time.

All graphs were created with GraphPad Prism (version 5 or 7); statistical significance was determined using one-way or two-way ANOVA, or Student t test, unless otherwise stated. Nonlinear regression using Gaussian model with least squares fitting method was used to present and compare BMAd size distribution profiles. Data represent mean ± SEM, unless otherwise noted. Significance is indicated as: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Additional details on methods can be found in the Supplementary Data.

To gain insight into the relationship between BMAds and myeloma cells, we explored BMAT changes in bone marrow from patients with newly diagnosed multiple myeloma. Iliac crest biopsies from patients with multiple myeloma contained a large number of BMAds, many of which were flanked by CD138⁺ myeloma cells (Fig. 1A), suggesting bidirectional signaling between these cells. Quantification of BMAds indicated a significant decrease in adipocyte volume fraction (AV/MV; Fig. 1B) and mean adipocyte size (Fig. 1C), as well as significantly altered adipocyte size profiles (Supplementary Fig. S1A) in newly diagnosed multiple myeloma patient samples compared with controls, but no change in adipocyte density (Fig. 1D). Interestingly, the reduced adipocyte volume fraction and size in multiple myeloma samples were independent of the tumor load in the bone marrow, indicating that decreased BMAT in multiple myeloma is not simply due to crowding of the marrow by multiple myeloma cells (Fig. 1E–G). Interestingly, treatment of multiple myeloma increased the adipocyte volume fraction (Fig. 1H) and density (Fig. 1J) in patients. Moreover, we observed significant alteration of the adipocyte size distribution profile after treatment, mainly because of increased frequency of medium-sized adipocytes, and decreased frequency of larger or smaller adipocytes (Supplementary Fig. S1B), such that the mean adipocyte size was unchanged (Fig. 1I). Thus, BMAT is altered in patients with multiple myeloma on the basis of treatment or stage of disease.

To model what occurs in patients and explore the BMAT–multiple myeloma relationship more deeply, we next examined the bone marrow of BKAL mice harboring 5TGM1^gfp+luc+ multiple myeloma cells (Fig. 2A; ref. 2). Histologic analysis of femurs harvested 21 days after tail vein injection of multiple myeloma cells (Fig. 2B and C) indicated that bone marrow adiposity was decreased in terms of AV/MV (Fig. 2D) and adipocyte density (Fig. 2F), but not adipocyte size (Fig. 2E). Furthermore, bone marrow adiposity was not correlated with tumor burden as assessed via bone marrow GFP⁺ flow cytometry (2). We next employed a SCID-Beige human xenograft bone-homing MM.1S^gfp+luc+ mouse model (Fig. 2G; ref. 18; femoral H&E images, Fig. 2H and I). MM.1S^gfp+luc+ proliferation in vivo was confirmed via luciferase activity at 14, 17, and 21 days after injection (Fig. 2J). Histologic analysis of femurs harvested at the terminal endpoint (determined by hind limb paralysis), indicated presence of a large number of BMAds in proximal femurs of control animals (Fig. 2H), which were largely absent in tumor-bearing mice (Fig. 2I). Quantification of BMAds revealed that MM.1S-bearing mice had significantly decreased AV/MV (Fig. 2K), BMA size (Fig. 2L), and density (Fig. 2M). Overall, BMAT in multiple myeloma mouse models and human multiple myeloma patient samples was decreased, demonstrating that multiple myeloma infiltration alters BMAds, a phenomenon observed across mammalian species that may contribute to disease progression.

We next characterized how multiple myeloma cells affect BMAds using in vitro cocultures. We first utilized 3T3-L1 adipocytes to increase reproducibility of our results by eliminating the potential confounding effects of a mixed stromal population or donor variability present in primary samples. 3T3-L1 preadipocytes were differentiated into adipocytes and exposed to myeloma cells via indirect transwell coculture for 72 hours (Fig. 3A). 3T3-L1 adipocytes exposed to MM.1S cells had decreased lipid content (Fig. 3B) and significantly reduced expression of the early adipogenic transcription factor Cebpa, as well as mature adipocyte genes, Adipoq and Fabp4, when cocultured with MM.1S or 5TGM1 cells (Fig. 3C). Similarly, 3T3-L1 adipocytes exposed to MM-CM had reduced lipid content (Supplementary Fig. S2A) and adipogenic gene expression (Supplementary Fig. S2B). Furthermore, primary adipocytes derived from mouse bone marrow MSCs (Fig. 3D) and mouse ear MSCs (Supplementary Fig. S2C) cocultured with multiple myeloma cells using transwells also exhibited reduced adipogenic mRNA transcripts. Hence, multiple myeloma cells suppress normal adipogenic functions in differentiated adipocytes.

Microarray analysis of 3T3-L1 adipocytes with or without transwell coculture with MM.1S cells revealed 71 differentially expressed genes (Fig. 3E; Supplementary Table S1). In total, 28 genes were significantly downregulated (FC ≤ 1.5) and 43 genes were upregulated (FC > 1.5) in the adipocytes cocultured with multiple myeloma cells. Pathway ANOVA analysis of significantly altered transcripts indicated that multiple myeloma cells affect vital cellular pathways and significantly alter oxidative phosphorylation and cellular processes (e.g., ribosome, cytokine–cytokine receptor interactions, cell cycle, and spliceosome; Table 2). These transcriptome changes reflect the loss of the adipogenic phenotype in 3T3-L1 cells.

We next investigated the relationship between BMAT and myeloma cells utilizing RNA-seq. Human MSCs from three similarly aged donors were differentiated into BMAT and then cocultured with or without multiple myeloma cells (MM.1S, OPM-2, or RPMI-8226) for 72 hours using transwells. RNA-seq revealed significant changes in BMAT gene expression profiles in response to multiple myeloma (MM-BMAT), and many of these genes were tightly connected with nodes and grouped into functions, as evidenced by STRING-db (Fig. 4A) and GeneMania analyses (Supplementary Fig. S3A and S3B). The largest significant increase was found to be in the gene podoplanin (PDPN), a gene also expressed in other cancer-associated adipocytes (26), and cancer-associated fibroblasts (27), where its expression levels predict less favorable clinical outcomes (28).

BMAT exposed to MM.1S cells (BMAT_MM.1S) exhibited the greatest response (number of genes with large expression FCs), followed by BMAT exposed to OPM2 (BMAT_OPM2) and then BMAT exposed to RPMI-8226 (Supplementary Fig. S3C). A total of 102 significantly upregulated genes were common to all three multiple myeloma–associated BMAT samples; these fell into Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, including “pathways in cancer,” “proteoglycans in cancer,” and “cellular senescence” (Supplementary Table S2). Indeed, when genes encoding senescence-associated secretory phenotype (SASP; ref. 29) proteins were examined in MM-BMAT, compared with BMAT alone, many were upregulated including, SERPINB2 (204×), IL6 (175×), ICAM1 (33×), CXCL1 (58×), CXCL2 (39×), and VEGFA (5×; Fig. 4B). These results suggest that myeloma cells induce senescence in adipocytes, with genes encoding SASP proteins (e.g., IL6, IL8, CXCL1, and CXCL2) among the top 10 genes increased in response to MM.1S and OPM-2 cell lines individually (Supplementary Table S3). The top 10 up- and downregulated genes in BMAT in response to each multiple myeloma cell line are shown in Supplementary Tables S3 and S4, respectively. Importantly, MM.1S and OPM-2 significantly induced IL6 expression in BMAT (Supplementary Table S3), while no IL6 expression was observed in the myeloma cells themselves, either basally or after exposure to BMAT.

Only a few genes (16) were commonly downregulated in response to myeloma (Supplementary Fig. S3D; Supplementary Table S4), but all three multiple myeloma cell lines appeared to decrease important processes in BMAT, such as metabolism and signaling by receptor tyrosine kinases (Supplementary Table S5), according to pathway analysis of RNA-seq data. In agreement with our previous in vitro data, BMAT also exhibited marked reduction in adipogenic transcripts in response to multiple myeloma cell coculture (Fig. 4B). A noteworthy decrease across MM-BMAT (Supplementary Fig. S3E) was PI3K regulatory subunit 3 (PIK3R3/p55; red box), which interacts with PPARγ/PPARG to increase adipogenesis, while dominant negative p55 inhibits adipogenesis (30). SOX18 was the most downregulated molecule in MM-BMAT (Supplementary Fig. S3F), decreased on average approximately 4-fold. These changes suggest multiple myeloma–associated BMAT is indeed distinct from healthy BMAT, and likely plays a unique functional role in influencing myeloma cells.

To validate that multiple myeloma cells functionally induce a SASP in BMAds, we conducted 72-hour indirect cocultures utilizing five additional bone marrow donors and MM.1S cells. Human MSC–derived BMAT was exposed to MM.1S cells for 72 hours, and then fixed and stained for the senescence marker, β-gal (21). Significantly fewer lipid-containing adipocytes were identified in the coculture versus controls and the majority of multiple myeloma–associated adipocytes at this timepoint were β-gal positive (Fig. 4C; Supplementary Fig. S4A and S4B;). We also confirmed highly elevated expression levels of four SASP genes, IL6 (Fig. 4D), CSF2 (Supplementary Fig. S4C), CXCL1 (Supplementary Fig. S4D), and CXCL2 (Fig. 4E), in response to MM.1S and OPM-2 via qPCR and observed slight increases in the protein secretion of the SASP proteins IL-6, IL-8, and CCL2/MCP1 in MM-BMAT (Fig. 4F). To test whether senescence can occur in vivo, we examined SASP gene expression in whole-bone marrow samples from SCID-Beige mice. We detected elevated Il6 (Supplementary Fig. S4E) and Cxcl1 (Supplementary Fig. S4F) expression in MM.1S tumor–bearing mice compared with naïve mice 21 days after tumor cell injection. In a study by Liu and colleagues, BMAds isolated from patients with multiple myeloma in complete remission were shown to exhibit reduced adipogenesis (14); we reanalyzed this publicly available data utilizing GEO2R, and found that these adipocytes also exhibit marked increases in the SASP gene signature (Supplementary Fig. S4G; ref. 29), although not all results reached significance because of a small sample size. To explore IL6 protein changes in MM-adipocytes, we cocultured 3T3-L1 adipocytes with MM.1S cells or alone and measured IL6 secretion into the conditioned media of these cultures by mouse-specific ELISA. 3T3-L1 adipocytes exposed to MM.1S cells exhibited increased IL6 secretion after only 4 hours in direct coculture (MM D), and significant increases in both direct and indirect (MM ID) transwell coculture systems at 24 and 48 hours (Fig. 5A). Seventy-two-hour exposure to MM.1S-CM also increased IL6 secretion in these adipocytes and revealed significant increases in other cytokines, including the melanocortin receptor antagonist Agouti-related neuropeptide (AgRP/Agrp), lipocalin-2 (Lcn2), RAGE/Ager, RANTES/Ccl5, and RBP4/Rbp4, and reduced IGF-2/Igf2 (Fig. 5B).

Because senescent cells often have altered metabolisms, and our RNA-seq, microarray, and prior data suggested aberrations in MM-BMAT metabolism, we performed bioenergetic studies using the Seahorse XF cell mito stress test on 3T3-L1 adipocytes exposed to MM-CM (Fig. 5C). MM-CM–treated adipocytes exhibited significant increases in basal oxygen consumption rates (OCR; Fig. 5D) and nonmitochondrial respiration (Fig. 5E), the latter being indicative of increased reactive oxygen species generation, which damages mitochondria. Furthermore, MM-CM–treated cells exhibited trends toward increased proton leak (Fig. 5F), further indicating damage from increased oxidative stress (31). These damaged mitochondria result in metabolically dysfunctional cells, requiring higher levels of ATP for maintaining organelle integrity, resulting in the increased basal OCR noted above (32). Taken together, these results suggest that MM-CM treatment is damaging to the mitochondria of adipocytes, such that they produce an aberrant cytokine profile, which includes elevated release of IL6, a cytokine known to support myeloma cells in vitro and in vivo.

We next interrogated our RNA-seq data for common, highly expressed genes in the three multiple myeloma cell lines that could explain their ability to induce senescence in BMAT. Under normal maintenance conditions, the three multiple myeloma cell lines expressed 10,662 common, highly expressed genes (Supplementary Fig. S5A), including HMGB1 and HMGB2 (Supplementary Fig. S5B). These two soluble proteins can trigger NF-κB (NFKB1) and P38/MAPK pathways (which were increased in MM-BMAT; Supplementary Table S2) to initiate the production of IL6/IL6 (33) or directly modulate chromatin opening at SASP gene loci (34). Interestingly, HMGB3 was also increased in 3T3-L1 adipocytes exposed to multiple myeloma cells (Fig. 3E). Thus, although we cannot be certain which proteins from multiple myeloma cells induce the senescent phenotype in BMAds, our data suggest that the HMGB family may be involved.

We then investigated how multiple myeloma cells alter adipogenesis. Our analysis of publicly available data (24) found that the expression of adipogenesis regulators is altered in myeloma cells (Supplementary Fig. S6A and S6B). Notably, the gene expression levels of WNT5A, TGFB1, and TNF (Supplementary Fig. S6A; orange arrows), which encode secreted, soluble negative regulators of adipocyte differentiation (Supplementary Fig. S6C), were elevated in all stages of myeloma disease: MGUS, smoldering, overt, and relapsed (Supplementary Fig. S6A). Indeed, all three of our cell lines exhibit basal expression of one or more of these genes (Supplementary Fig. S6D), suggesting that these proteins are likely factoring into the reduced adipogenesis of MM-BMAds.

The three multiple myeloma cell lines used in our study exhibited vastly different expression profile changes (Supplementary Fig. S7A–S7C) in response to adipocytes, reflecting their distinct biological origins from different donors and subsequent adaptation to culture in different laboratories. Yet, six genes were commonly upregulated (>2 ×) among all three lines (Fig. 6A; Supplementary Fig. S8A) and no genes were commonly downregulated (>2×; Supplementary Fig. S8B). The upregulated genes (Supplementary Fig. S8C) were analyzed with STRING analysis (Fig. 6A; Supplementary Fig. S7, red boxes), and three of these (KLF9, TSC22D3, and FKBP5) are implicated in glucocorticoid receptor signaling (35). Increased expression of FKBP5 (Fig. 6B), KLF9 (Fig. 6C), and PARP9 (Supplementary Fig. S8D) was confirmed in MM.1S and OPM-2 lines cocultured with three BMAT donors via qPCR. Hence, irrespective of major distinctions in transcriptomes of different multiple myeloma cell lines, there is a select number of upregulated genes that may contribute to prosurvival effects of adipocytes on multiple myeloma cells.

To explore whether the observed changes in glucocorticoid receptor signaling–related transcripts translated to differences in dexamethasone responsiveness, we cocultured MM.1S in transwell with human MSC–derived BMAT and examined apoptosis, proliferation, and cell cycle with or without dexamethasone. When exposed to dexamethasone, MM.1S cells exhibited a nearly 3-fold increase in total apoptosis, which was severely diminished when cocultured with BMAT (Fig. 6D). Similarly, the dexamethasone-induced reduction in proliferation, as measured by the percent of Ki67 (MKI67)⁺ cells, was also rescued by BMAT coculture (Fig. 6E). Dexamethasone also altered MM.1S progression through cell cycle (Supplementary Fig. S8E); however, in the presence of BMAT, the percent of MM.1S cells in sub-G₁-phase was partially rescued and a complete rescue of S-phase populations was observed (Fig. 6F). To investigate the clinical relevance of these findings, we next examined BMAT in multiple myeloma patient iliac crest biopsies prior to their treatment with high-dose melphalan and dexamethasone followed by autologous bone marrow transplantation. Interestingly, we found that individuals who showed a complete response to treatment had significantly lower BMAT volume fraction (AV/MV) before treatment (−54%; P < 0.05) compared with patients with partial/very good partial response (Fig. 6G). These findings support our in vitro data, suggesting that higher BMAT prior to treatment protects multiple myeloma cells against chemotherapy. Thus, interactions between multiple myeloma cells and adipocytes protect multiple myeloma cells against the apoptotic effects of dexamethasone and add a new layer to the “vicious cycle” of myeloma progression.

Next, to clearly delineate effects of MM-BMAT versus naïve BMAT, which is difficult in transwell cocultures, we tested conditioned media from MM-BMAT (BMAT that had been cocultured in transwell with MM.1S cells for 72 hours) and naïve BMAT from matched donors for their effects on multiple myeloma cells (Fig. 6H). MM.1S cells exhibited a significant approximately 40% decrease in cell number in response to dexamethasone alone. Naïve BMAT conditioned media elicited a proliferative effect on multiple myeloma cell number, as well as dexamethasone resistance. Conversely, MM-BMAT conditioned media induced a slight, nonsignificant decrease in multiple myeloma cell number, but similar to naïve BMAT conditioned media, also induced dexamethasone resistance. The dexamethasone resistance induced by naïve or MM-BMAT was similar (Supplementary Fig. S8F).

Finally, we created senescent human BMAT in vitro by administering 10 Gy irradiation to differentiated BMAT and evaluated its effects on multiple myeloma cells. We also investigated whether senescent BMAT could be targeted by senolytics [dasatinib and quercetin (D+Q); Supplementary Fig. S9A and S9B]. Conditioned media from these samples were applied to MM.1S cells. Similar to MM-BMAT conditioned media, we observed a significant reduction in multiple myeloma cell number in response to irradiated BMAT, but not control BMAT (Supplementary Fig. S9B), suggesting that senescent BMAT differentially effects multiple myeloma cells. We also observed dexamethasone resistance in multiple myeloma cells treated with conditioned media from irradiated BMAT (but not control BMAT). These data suggest that irradiated BMAT may induce a drug-resistant and potentially more dormant phenotype in multiple myeloma cells. Importantly, in conditioned media collected from irradiated BMAT following treatment with D+Q for 48 hours, we observed increased sensitivity to dexamethasone, and no restoration of multiple myeloma cell numbers. Collectively, these data indicate novel functions of BMAT in multiple myeloma and suggest that healthy and senescent (myeloma-associated or irradiated) BMAT provide resistance to multiple myeloma treatment (Fig. 6I), although potentially through different mechanisms.

Our study demonstrates that multiple myeloma cells disrupt the normal bone marrow microenvironment by decreasing adiposity and inducing a SASP in BMAds. These observations were observed in both patient-derived samples and murine models, reflecting conservation of the pathologic mechanism in mammalian species. Our patient data demonstrated no direct correlation between bone marrow tumor burden and bone marrow adiposity in overt multiple myeloma, arguing against a “space-constricted” mechanism causing decreased BMAT. The mechanism of decreased marrow adiposity varied between our patient data and in vivo models, demonstrating the need for multiple methods of interrogating BMAT response to tumor cells. Overall, these data support the interpretation that multiple myeloma cells inhibit bone marrow adiposity in mice and humans and induce senescence in BMAds. Hence, senescent BMAds may be a new biological target for senolytic strategies to treat multiple myeloma, which may be even more relevant clinically because aging is a major MGUS and multiple myeloma risk factor.

One of the hallmarks of cancer-associated senescent cells is the presence of a SASP, which involves the secretion of molecules used by cancer cells to promote their growth and survival. Senescence in tumor-related stromal cells has been observed in myeloma patient MSCs and other cancer-associated cells (3, 36–38). Senescent cells also contribute to age-related bone loss (39). Importantly, we observed increases in adipocyte IL6/IL6 and IL8/CXCL1 gene expression and secreted protein in adipocytes upon multiple myeloma coculture. These findings were recapitulated in whole-bone marrow samples from tumor-bearing mice compared with naïve controls, providing in vivo evidence of a myeloma-driven SASP phenotype. Moreover, other CXCL family members were also upregulated in MM-adipocytes, such as CXCL5, CXCL6, and CXCL7. Although these CXCL members are not considered typical senescence-related chemokines, these ligands can bind the same CXCR2 receptor as known SASP proteins, CXCL1, CXCL2, and CXCL3 (40). Upregulation of these senescence-associated genes strengthens the emerging concept that MM-BMAds are senescent and supports the notion that MM-BMAT inflammatory SASP proteins contribute to myeloma progression or related bone disease. Indeed, Liu and colleagues recently reported that MM-adipocytes release soluble factors that inhibit osteoblastogenesis and stimulate osteoclastogenesis (14). Reduced levels of adiponectin (ADIPOQ), adipsin (CFD), and visfatin (NAMPT), as well as increased TNFα (TNF) in MM-adipocytes contributed to this imbalance. Our reanalysis of the data (14) demonstrated increases of other SASP-related transcripts (Supplementary Fig. S4E), which could also contribute to multiple myeloma bone disease. Ozcan and colleagues investigated the relationship between senescent MSCs and myeloma cells in vitro and observed a substantial effect on myeloma cell cycle and proliferation in cocultures, and differences in effects between naïve cells and myeloma “primed” or preexposed cells (41). These findings demonstrate that myeloma cell proliferation is modulated by senescence-related factors, and that myeloma cells may reprogram neighboring cells to support their survival. Overall, our findings integrate well with data suggesting that myeloma induces senescence in the marrow, and that this may further contribute to multiple myeloma disease.

We also observed that multiple myeloma cells consistently suppressed Pparg/PPARG and other adipokines in adipocytes. It has been reported that MM-BMAds have reduced PPARγ signaling (14) and that reprogramming by multiple myeloma cells occurs with direct contact via integrin-α6, despite the fact that identical changes were observed with indirect contact, although to a lesser extent. Our data argue that indirect (conditioned media and transwell) coculture, not via integrin signaling, is also sufficient to induce this PPARγ/PPARG change in MM-BMAds. Our study suggests that, in addition to basal expression of WNT5A, TGFB1, and TNF, high HMGB1 and HMGB2 expression by myeloma cells contributes to this multiple myeloma–associated adipocyte phenotype. HMGB1 signals via toll-like receptor 4 (TLR4) to trigger NF-κB and P38/MAPK pathways, activating the production of IL6/IL6 and MCP1/CCL2 in adipocytes (33). Myeloma-derived HMGB1 may elicit similar effects from BMAds, as both the TLR4 cascade and NF-κB signaling pathway were significantly upregulated in the multiple myeloma–associated adipocyte dataset. HMGB2 is also highly expressed in myeloma cells, and has been shown to regulate SASP gene expression by aiding in chromatin opening at SASP gene loci (34). Thus, high HMGB1 and HMGB2 expression may contribute to the multiple myeloma–associated adipocyte phenotype. Cell line–specific expression differences, including high expression of S100A4 (42) in RPMI-8226, IL32 (43) in OPM-2, and CCL3 (44) in MM.1S, have also been shown to modulate expression of either adipogenic genes or inflammatory factors, and may modulate the MM-adipocyte phenotype. Together, this complex cocktail of multiple myeloma cell–secreted signaling molecules contributes to the MM-adipocyte phenotype by suppressing metabolism and driving a senescent phenotype.

Despite advances in myeloma treatment, patients continue to develop drug resistance and relapse, in part, due to interactions between multiple myeloma cells and microenvironmental cells, such as preadipocytes and adipocytes (4). Mouse preadipocyte conditioned media can enhance myeloma chemotaxis via activation of the Wnt signaling cascade, while mature adipocyte conditioned media can promote myeloma growth through the ERK signaling cascade (4). Similarly, human adipocyte conditioned media confer increased viability (by increased STAT3) to myeloma cells and stimulate angiogenesis in vitro (12). BMAd-derived leptin and adipsin also support myeloma cell chemotherapy resistance through the induction of autophagy and subsequent reduction of apoptosis (5). While these mechanisms may be involved in the dexamethasone resistance reported here, we also demonstrate downregulation of these adipokines in MM-BMAds, and specific increases in the genes known to be involved in glucocorticoid receptor trafficking in cocultured multiple myeloma cells. Thus, our findings reveal previously underappreciated driving mechanisms through which BMAds support multiple myeloma disease progression.

We also observed consistently increased expression of PARP family member 9 (PARP9) in all three multiple myeloma cell lines in response to BMAT coculture. PARP9, also known as B-aggressive lymphoma (or BAL/BAL1), is highly expressed in high-risk diffuse large B-cell lymphomas (45) and breast cancer (46), and can promote migration in these cancer cells. PARP9 stimulates phosphorylation of STAT1 to ultimately suppresses p53 via IRF1, thereby enhancing proliferation and reducing apoptosis (47), two phenotypes that we observed in our multiple myeloma cells cultures with BMAds when challenged with dexamethasone. Thus, while we demonstrated upregulation of genes intimately involved in glucocorticoid receptor trafficking and subsequent dexamethasone resistance, BMAT also induces expression of other survival pathways (Supplementary Fig. S7; MAPK/BCL2, negative regulation of apoptosis, JAK/STAT signaling, and IFN signaling), which may be related to PARP9; this will be investigated in future studies.

In MM.1S cells, we observed significant increases in cells in S-phase and dexamethasone resistance, when cocultured with MM-BMAT in transwells (Fig. 6D–F). We also observed that conditioned media from naïve BMAT increased multiple myeloma cell numbers, while MM-BMAT conditioned media decreased cell numbers (Fig. 6H), while both conditioned media types induced dexamethasone resistance, suggesting that MM-BMAT supports tumor cells by mechanisms distinct from healthy BMAT. Irradiated BMAT was similar to MM-BMAT in that both decreased multiple myeloma cell numbers over time (suggesting dormancy in tumor cells) while inducing dexamethasone resistance. It is possible that senescent BMAT induces a dormant or senescent phenotype in multiple myeloma cells, as suggest with a potential S-phase arrest in multiple myeloma cells in transwell coculture with BMAT, although other factors may also be at play. Interestingly, senescence can be induced in cells from nearby senescent cells, which may be occurring here (41). This potential phenomenon and the potential for dormancy-driven drug resistance to be occurring in multiple myeloma cells necessitate further investigation. Future directions should also compare BMAds induced to undergo senescence through different mechanisms (e.g., aging, irradiation, or culture with multiple myeloma cells), and assess differences their effects on multiple myeloma cells.

In terms of translational relevancy, irradiated BMAT treated with senolytics (D+Q) restored sensitivity to dexamethasone and resulted in the lowest tumor cell number after 72-hour treatment. Combined, these results suggest that senescent and senescent-like adipocytes support dexamethasone resistance, and that this can be partially rescued with senolytic treatment. Targeting senescent BMAds may reflect a new therapeutic option for targeting microenvironment-induced drug resistance in myeloma without inducing proliferation of multiple myeloma cells (as an aside, BMAT used to make conditioned media for experiments in Supplementary Figs. S6H and S9B was from different donors and cultured for different times, due to different requirements for the irradiation protocol; these differences likely explain differences in the BMAT conditioned media effects observed. This highlights that within each experiment, the same BMAT donors should always be used, and as many primary human samples should be used as possible, to capture human donor variability).

In conclusion, our studies demonstrate that BMAT changes dramatically in murine myeloma models and human patients. Adipocytes exposed to myeloma-derived factors exhibited decreased lipid content, decreased adipogenic transcripts, and increased expression of SASP transcripts and proteins including IL6/IL6. Signals from MM-adipocytes affected myeloma cell number, response to dexamethasone, cell-cycle states, and expression of genes related to glucocorticoid receptor signaling, which could affect the availability of the glucocorticoid receptor for ligand binding and nuclear localization (48–50). Senescent, MM-BMAT, and naïve BMAT have different effects on multiple myeloma cells, and all contribute to dexamethasone resistance. Finally, it is plausible that SASP proteins from multiple myeloma–associated BMAds contribute to the vicious cycle of multiple myeloma–induced osteolysis, beyond other established functions, such as autocrine inhibition of adipogenesis and generating a tumor-supportive microniche (14). Therefore, removal of these cells via senolytic therapies could have a 2-fold benefit of alleviating bone damage and diminishing myeloma drug resistance.

M. Hinge reports personal fees from Celgene Aps outside the submitted work. M.R. Reagan reports grants from NIH, American Cancer Society, and Kane Foundation during the conduct of the study and NIH, Onkimmune, UCB Biopharma SRL, Biopact, and National University of Ireland, Galway outside the submitted work. No disclosures were reported by the other authors.

H. Fairfield: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A. Dudakovic: Resources, data curation, software, formal analysis, investigation, methodology, writing-original draft, writing-review and editing. C.M. Khatib: Formal analysis, investigation, project administration. M. Farrell: Resources, data curation, validation, investigation, writing-original draft, writing-review and editing. S. Costa: Resources, data curation, software, formal analysis, methodology, writing-original draft. C. Falank: Resources, methodology. M. Hinge: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. C.S. Murphy: Software, investigation, methodology. V. DeMambro: Resources, data curation, formal analysis, writing-original draft, writing-review and editing. J.A. Pettitt: Conceptualization, resources, software, validation, investigation, methodology. C.W. Lary: Resources, writing-original draft. H.E. Driscoll: Resources, data curation, software, methodology, writing-original draft, project administration. M.M. McDonald: Resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M. Kassem: Resources, funding acquisition. C. Rosen: Conceptualization, resources, software, formal analysis, supervision, funding acquisition, visualization, methodology, writing-original draft, project administration, writing-review and editing. T.L. Andersen: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A.J. van Wijnen: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A. Jafari: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M.R. Reagan: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

We thank Drs. Calvin Vary and Anyonya Guntur for their intellectual contributions, Dr. Anne Breggia and the Maine Medical Center Biobank, and Dr. Asha Nair from Biomedical Statistics and Informatics Core at Mayo Clinic. We also thank the caregivers taking care of the authors' children so the authors could perform this work. This work was supported by the American Cancer Society (research grant #IRG-16-191-33 and RSG-19-037-01-LIB to M.R. Reagan), the NIH (R24 DK092759-01, U54GM115516, R01AR049069, R37CA245330, and P20GM121301 to M.R. Reagan), the Danish Southern Region Research grant (20/14282 to principal investigator T.L. Andersen), and the Kane Foundation. Microarray and functional analysis were supported by the Vermont Biomedical Research Network Bioinformatics Core, which is supported by an Institutional Development Award from the National Institute of General Medical Sciences of the NIH under grant number P20GM103449.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.