Doxorubicin enhances adipogenesis in an FGF2-dependent manner and induces a tumour-promoting secretory phenotype

Graphical abstract

Low-dose chemotherapy promotes enhanced adipogenesis in part via upregulation of FGF2, which results in enhanced secretion of factors that promote tumor cell growth.

Highlights

• •Chemotherapy treatment of differentiating mesenchymal stem cells resulted in increased numbers of mature lipid-containing adipocytes.
• •RNA-seq identified alterations in key pathways including the fibroblast growth factor (FGF) signaling pathway.
• •Inhibition of FGF2 ameliorated doxorubicin-enhanced adipogenesis.
• •Chemotherapy treated adipocytes released factors including FGF2 that promoted tumour cell growth.

Abstract

The bone marrow microenvironment is highly saturated with bone marrow adipocytes (BMA), which differentiate from their precursor, mesenchymal stem cells (MSC). Evidence from patient trials suggests that bone marrow adiposity is increased in patients following some forms of chemotherapy. Moreover, it has been suggested that BMA can confer chemotherapeutic resistance to tumour cells, thereby ascribing a tumour-supportive role to BMA. We investigated the effect of chemotherapy on adipogenesis of human MSC in vitro, as well as potential underlying mechanisms leading to altered adipogenesis, and the effects in turn on tumour cell proliferation. Doxorubicin or carboplatin treatment of adipogenic differentiating MSC led to an increased percentage of mature BMA confirmed by increased gene expression of the adipocyte marker, PPARG. RNA-seq analysis identified significant increases in fibroblast growth factor (FGF) pathway genes in doxorubicin treated adipogenic differentiated MSC, which were validated at the mRNA and protein level. Notably, endogenous and secreted FGF2 was significantly increased with doxorubicin treatment. Furthermore, siRNA-mediated targeting of FGF2 impeded the doxorubicin-enhanced formation of lipid-containing mature BMA returning it to levels similar to vehicle control treated BMA. As FGF2 is a secreted protein we tested and confirmed that transfer of conditioned media from doxorubicin-treated BMA enhanced proliferation of tumour cells in vitro, a phenotype that was partially abrogated when FGF2 was depleted from adipogenic differentiating MSC. Our findings suggest that chemotherapy actively promotes adipogenesis, in part by alteration of FGF2 in the context of doxorubicin treatment, which directly enhances adipogenesis and in turn leads to enhanced tumour cell growth as a result.

1. Introduction

Patients with some forms of advanced cancer, including breast, lung and prostate cancer, commonly develop bone metastases [1]. Following their dissemination to the bone, tumour cells can exist in a dormant state within the bone for long periods of time (months to years), supported in their growth and resistance to chemotherapy by the local bone marrow microenvironment [2], [3], [4], [5], [6]. Bone metastases cause an increase in patient skeletal-related events which can include severe pain, fracture, spinal cord compression or hypercalcemia [1]. Current approaches to treat bone metastasis, namely radiation or bisphosphonate therapy, can reduce tumour size and symptoms however have a limited period of efficacy and do not cure the patient of bone metastasis [7], [8], [9], [10]. As such, a better understanding of the processes that promote tumour cell growth in the bone is necessary to offer preventative or curative treatments to patients at risk of developing bone metastases.

The bone marrow microenvironment is highly saturated with bone marrow adipocytes (BMA), which represent between 75 and 90% of cells in the adult bone marrow [11]. BMA differentiate from their precursor, bone-resident mesenchymal stem cells (MSC), in a process termed adipogenesis. In the early phase of adipogenesis, activity of two transcription factors are invoked, CCAAT/enhancer-binding protein-beta (CEBPB) and CCAAT/enhancer-binding protein-delta (CEBPD), which then bind to the promoter regions and drive expression of the two “master regulators” of adipogenesis: peroxisome proliferator-activated receptor gamma (PPARG) and CCAAT/enhancer-binding protein-alpha (CEBPA) [12].

Multiple in vivo studies have shown that rat tibial bone marrow adiposity increases in conjunction with increased bone marrow expression of PPARG following treatment with doxorubicin and cyclophosphamide [13], paclitaxel and lapatinib [14], or methotrexate [15]. Similar findings have been observed in chemotherapy treated patients receiving combined carboplatin and paclitaxel, where an increase in bone marrow adiposity in the lower vertebrae of the spine 6–12-months post-chemotherapy as compared to pre-treatment was observed [16]. Similarly, in a small cohort of patients that received chemotherapy or radiation, the marrow fat fraction of the L4 vertebral body and the femoral neck were significantly increased at 6-months post-treatment as compared to baseline [17]. Elevated bone marrow adiposity following chemotherapy may even persist long-term, as a retrospective PET/MRI study found that in patients who had previously received chemotherapy at a median time interval of 8 ± 34 months (range of 0–228 months) there was a significantly higher marrow fat fraction of the lumbar spine [18].

The effect of this increased bone marrow adiposity on the tumour-promoting ability of the bone marrow niche is unknown. However, in vitro data suggests BMA can directly confer chemotherapeutic resistance to tumour cells in part via their ability to sequester drug [19], [20], [21]. Bone marrow adipocytes may also drive tumour cell growth and progression by providing them with a source of lipids [22], promoting tumour cell migration [23] and promoting a metabolic shift towards aerobic glycolysis that favours accelerated growth, also known as the Warburg effect [24]. This suggests that primary tumour growth and response to treatment in bone is likely influenced by BMA. In addition, as many patients harbour disseminated tumour cells in the bone at early stages of disease, including at initial diagnosis of primary tumour [25], [26], there is a concern that increased bone marrow adiposity enhanced by neoadjuvant or adjuvant chemotherapy for primary tumours with a propensity to spread to the bone may create a “tumour-supportive” microenvironment within the bone, potentially contributing to growth of bone metastatic disseminated tumour cells.

While evidence shows that chemotherapy can increase bone marrow adiposity, some investigators attribute this to increased adipocyte hypertrophy. However, if de novo adipogenesis is promoted, increased numbers of adipocytes could support the survival of bone resident tumours or disseminated quiescent tumour cells, thereby priming a tumour-supportive bone microenvironment and contributing to progressive primary cancers in the bone or bone metastases [27], [28]. We hypothesized that chemotherapy promotes adipogenesis; thus, we investigated whether chemotherapy directly affected de novo adipogenesis of MSC and the possible contribution of this to tumour cell growth. We chose to study the effects of the DNA-damaging agents carboplatin and doxorubicin [29], [30]. Doxorubicin and carboplatin are also known to cause bone loss and increased bone turnover in preclinical models and patients [31], [32] which contributes to forming a favourable bone metastatic microenvironment [33].

To determine the potential effects of chemotherapy on the adipogenic compartment in the bone, we used human bone marrow-derived MSC and assessed the effect of doxorubicin or carboplatin on adipogenic differentiation in vitro. We further assessed the effects of doxorubicin treatment on alterations in gene expression in adipogenic differentiating MSC. We found that doxorubicin increased de novo adipogenesis and identified the fibroblast growth factor (FGF) signaling pathway as highly relevant in this process. Using both direct co-culture and conditioned supernatant transfer co-culture assays, we show that doxorubicin treatment of differentiating MSC results in secretion of factors which promote the proliferation of tumour cells.

2. Materials and methods

2.1. Cell lines and cell culture

Human MSC were obtained from Lonza (Walkersville, MD, USA) (Table 1). MSC were maintained in MSCGM Mesenchymal Stem Cell Growth Medium (cat #PT-3001) obtained from Lonza. MSC were differentiated using hMSC Adipogenic Differentiation Medium (cat #PT-3004) from Lonza or MesenCult™ Adipogenic Differentiation Kit (Human) (cat #05412) from StemCell Technologies (Vancouver, BC). PC3 prostate tumour cells (ATCC CRL-1435), 4T1 breast tumour cells (CRL-2539) and K562 leukemia tumour cells (CCL-243) were obtained from the American Type Culture Collection (ATCC, Manassas, VA). PC3 cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) (cat #11995123, ThermoFisher Scientific, Ottawa, ON) with 10% fetal bovine serum (FBS, cat# 35–077-CV, Corning, Ottawa ON). 4T1 cells were cultured in DMEM with 10% FBS, 10uM HEPES, 1 mM sodium pyruvate and 1.5 mg/ml sodium bicarbonate. K562 cells were cultured in RPMI 1640 Medium (cat #11875093, Thermo Fisher Scientific, Ottawa, ON) with 10% FBS. All cells were maintained in culture conditions of 5% CO₂ and 37°C. Doxorubicin and carboplatin were supplied by the Ottawa Hospital General Campus Pharmacy.

Table 1.

Human mesenchymal stem cell donor characteristics.

Lot #	Sex	Age (years)	Race
21TL046615	Female	23	White
19TL329433	Male	23	Black
20TL293908	Male	23	Black
22TL238331	Male	21	Black
22TL202827	Male	37	White
22TL290855	Male	31	Asian

2.2. Mesenchymal stem cell adipogenic differentiation and doxorubicin treatment

MSC were seeded in tissue culture plates at a density of 2.1x10⁴ cells/cm². At 80–90% confluency, cells were differentiated according to manufacturers’ protocols. When using media supplied by Lonza, adipogenesis was induced by culture with Adipogenic Induction Media (IM) and Adipogenic Maintenance Media (MM) on the following schedule: 72 h IM, 24 h MM, 72 h IM, 24 h MM, 72 h IM. When using adipogenic media supplied by StemCell Technologies, media was replenished on an identical schedule. The vehicle control, doxorubicin (25 nM or 50 nM), or carboplatin (25 μM or 50 μM) were added to cells beginning on day 5 or day 9 of the differentiation process and maintained throughout thereafter.

2.3. siRNA transfection

Non-targeting siRNA (cat #D-001210–02, Dharmacon, Lafayette, CO) or ON-TARGETplus siRNA targeting FGF2 (cat #LQ-006695–00-0010, Dharmacon, Lafayette, CO) (Table 2) were diluted in Opti-MEM I Reduced Serum Medium (#31985–070, Thermo Fisher Scientific, Ottawa, ON) and combined with Lipofectamine RNAiMAX Transfection Reagent (#13778075, Thermo Fisher Scientific, Ottawa, ON) according to the manufacturer’s protocol. Differentiating MSC were treated at day 9 of adipogenic differentiation with adipogenic media containing vehicle control or doxorubicin, followed by addition of the siRNA-containing transfection reagent at a final siRNA concentration of 100 nM. Cells were subsequently assessed on day 12 of differentiation (i.e. 72 h post-siRNA transfection) by BODIPY (see section 2.4) and RT-qPCR (see section 2.5).

Table 2.

Sirna sequences.

Gene Target	Reference Number	Target Sequence
Non-targeting	#2	UAAGGCUAUGAAGAGAUAC
FGF2	#6	UCAAAGGAGUGUGUGCUAA
FGF2	#7	GCUAAGAGCUGAUUUUAAU
FGF2	#8	GAUGGAAGAUUACUGGCUU

2.4. Lipid droplet detection

At day 12 of adipogenic differentiation, media was removed and cells washed 3x with phosphate-buffered saline (PBS) prior to fixation in 3% paraformaldehyde for 30 min. Fixed cells were then washed 3x with PBS and stained for 10 min with BODIPY 493/503 lipid dye (#D3922, Thermo Fisher Scientific, Ottawa, ON). The stock solution of BODIPY 493/503 dye was made at a concentration of 1.9 mM (0.5 mg/mL) in ethanol and the working solution was prepared immediately before use by diluting in 150 mM sodium chloride solution to a final concentration of 3.8 nM. Stained cells were washed 3x with PBS and then imaged using EVOS M5000 Imaging System (Invitrogen, Waltham, MA). All steps were performed at room temperature. Percentage of BODIPY positive cells was calculated following enumeration of total number of cells and number of cells positive for BODIPY staining, and average diameter of BODIPY positive cells calculated in 3 images from each of 3 wells per condition using ImageJ software [34].

2.5. RNA extraction, cDNA synthesis and quantitative RT-PCR

At day 12 of adipogenic differentiation, cells were collected in QIAzol Lysis Reagent (cat #79306, Qiagen, Germantown, MD), extracted using the RNeasy Mini Kit (cat #74104, Qiagen, Germantown, MD), according to the manufacturer’s instructions, and stored at –80°C. cDNA was synthesized by combining experimental RNA and nuclease-free water with the GoScript Reverse Transcription Mix, Oligo(dT) (cat #A2791, Promega, Madison, WI) on ice, followed by incubation in the T100 Thermal Cycler (Bio-Rad, Mississauga, ON) at 25°C for 5 min, 42°C for 60 min and 70°C for 15 min. cDNA was diluted 1:10 in ultra-pure nuclease-free water (cat #10977015, Invitrogen, Waltham, MA). RT-qPCR was performed with the RT² SYBR Green ROX qPCR Mastermix (cat # 330523, Qiagen, Germantown, MD) and primer/cDNA mixes were prepared according to manufacturer’s instructions. qPCR reactions were performed on the 7500 Fast Real-Time PCR System (Applied Biosystems, Waltham, MA) with a holding stage of 50°C for 20 s, denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Primer sequences are listed in Table 3 and all primers were acquired from Thermo Fisher Scientific (Ottawa, ON). All RT-qPCR results were normalized to gene expression of the controls RPS13 and HSPCB, which were determined to be the least changed with doxorubicin treatment as compared to other commonly used housekeeping genes (Fig. S1), and have been previously found to be ideal housekeeping genes for use with human adipose tissue [35].

Table 3.

Primer sequences (human).

Gene	Forward Primer (5′ to 3′)	Reverse Primer (5′ to 3′)
CEBPA	GGTGGATAAGAACAGCAACGA	TCAACTCCAACACCTTCTGCT
PPARG	AGGCGAGGGCGATCTTG	CCCATCATTAAGGAATTCATGTCAT
FGF2	AGAAGAGCGACCCTCACATCA	CGGTTAGCACACACTCCTTTG
FGFR1	GGCTGTATGAAAAGGGTGGGAATG	GGTGCGTCGTGAGGTCTGG
FGFR2	ATCTGCCTGGTCGTGGTC	GCTCTAATGTGGTATCCTCAAC
GPC1	GCCAGATCTACGGAGCCAAG	AGGTTCTCCTCCATCTCGCT
DUSP6	CAGTGACTGAGCGGCTAATG	TGTCCCAGTTTTTCCCTGAG
RPS13	CGAAAGCATCTTGAGAGGAACA	TCGAGCCAAACGGTGAATC
HSPCB	TCTGGGTATCGGAAAGCAAGCC	GTGCACTTCCTCAGGCATCTTG
ACTB	CCAACCGCGAGAAGATGA	CCAGAGGCGTACAGGGATAG
GAPDH	GGAGCGAGATCCCTCCAAAAT	GGCTGTTGTCATACTTCTCATGG
R18S	AGAAACGGCTACCACATCCA	CACCAGACTTGCCCTCCA
MDM2	TCTAGGAGATTTGTTTGGCGT	TCACAGATGTACCTGAGTCC
CDKN1A	CCGAAGTCAGTTCCTTGTGG	CATGGGTTCTGACGGACAT

2.6. Western blot analysis

For Western blot analysis, the following primary antibodies were used: rabbit anti-basic FGF clone E5Y6M (cat #46879, Cell Signaling Technology, Beverly, MA) and mouse anti-GAPDH (cat #ab8245, Abcam Limited, Cambridge, UK). Secondary antibodies used for Western blot analysis were obtained from Jackson ImmunoResearch Inc. (West Grove, PA) and include peroxidase-conjugated AffiniPure goat anti-mouse IgG H + L (cat #115–035-146) and AffiniPure goat anti-rabbit IgG H + L (cat #11–035-144).

Protein lysates were collected by washing cells in PBS and performing cell scraping in cold radioimmunoprecipitation assay buffer (RIPA) containing Protease Inhibitor Cocktail (cat #P2714, MilliporeSigma, Burlington, MA), 80 mM sodium pyrophosphate and 0.5 M ammonium vanadate. Protein lysates were vortexed, rotated for 30 min at 4°C and centrifuged for 10 min at 10,000xg. Supernatants were stored at −80°C. Protein concentration was quantified by adding Protein Assay Reagent Concentrate Dye (Bio-Rad, Mississauga, ON) to equal volumes of protein lysate between samples, and absorbance measured using the BioMate 3 (Thermo Fisher Scientific, Ottawa, ON). Absorbance was used to calculate protein concentration following interpolation of a standard curve generated from a serial dilution of bovine serum albumin (BSA). Laemmli SDS 4x sample buffer was added to protein lysates which were then denatured at 100°C for 5 min prior to loading onto 10% tris–glycine polyacrylamide gels for electrophoresis at 50 mA using the PowerPac 300 Electrophoresis Power Supply (Bio-Rad, Mississauga, ON). BLUeye Prestained Protein Ladder (cat # PM007-0500, FroggaBio Inc, Toronto, ON) was added in a separate lane for molecular weight mapping. Transfer of proteins to Immobilon-P PVDF membrane (cat #IPVH00010, Millipore Ltd., Etobicoke, ON) was facilitated by electrophoresis for 90 min at 100 V. Membranes were blocked for 30 min at room temperature on an orbital shaker in 5% milk or 5% BSA (cat #ALB001, BioShop Canada Inc., Burlington, ON) in tris-buffered saline with Tween 20® reagent (TBST) (20 mM Tris, 150 mM NaCl, 0.1% (w/v) Tween 20® reagent). Primary antibodies were incubated with membrane overnight at 4°C on a rocking platform in 5% BSA. Membranes were subsequently washed 3x 5 min in TBST, incubated for 1 h at room temperature in secondary antibody in TBST, and washed 3x 5 min in TBST. Membranes were imaged following 30–60 s in Clarity Western ECL Substrate (cat #1705061, Bio-Rad, Mississauga, ON) using the ChemiDoc Imaging System (Bio-Rad, Mississauga, ON).

2.7. RNA-sequencing

RNA-sequencing (RNA-seq) was performed by the Biomedical Research Core at the University of British Columbia (Vancouver, BC). RNA quality was assessed by the Agilent 2100 Bioanalyzer. cDNA libraries were generated using the NEBnext Ultra ii Stranded mRNA (New England Biolabs, Whitby, ON) with sequencing performed on the Illumina NextSeq 500 (Illumina, San Diego, CA) with Paired End 42 bp x 42 bp reads. Transcript quantification was executed using Kallisto (v0.45.0) [36] with the GRCh38 build of the human transcriptome and the −b 50 bootstrap option. Following alignment, raw abundance counts were imported into DESeq 2 (1.40.2) using TXImport (1.28.0) and analyzed using the standard published workflow [37] (library size correction, normalization). Genes with fewer than 10 counts across samples were filtered out. Differential gene expression was done using the ‘results’ function of DESeq2, contrasting the relevant control-treatment pairs for each cell line. Genes with significantly altered expression (adjusted p-value < 0.05) were assessed by GO term analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID) [38], [39]. PCA and Jaccard distances were calculated using variance stabilization method within the DESeq2 package.

2.8. TUNEL

Cells were seeded onto sterile coverslips, washed with PBS and fixed with 3% paraformaldehyde, pH 7.4, for 1 h at room temperature prior to TUNEL staining. Following fixation, coverslips were washed with PBS and incubated in a permeabilization solution (0.1% Triton X-100 in 0.1% sodium citrate) for 2 min on ice. The positive control was incubated with 3000 U recombinant DNase 1 in 50 mM Tris-HCl, pH 7.5, for 10 min at room temperature. Coverslips were washed twice with PBS and then TUNEL reaction mixture (cat # 11684795910, In Situ Cell Death Detection Kit, Fluorescein, Roche, Mannheim, Germany) was added to the sample and incubated for 1 h at 37°C in the dark. Coverslips were then washed three times and mounted onto slides with DAPI mounting media. Samples were visualized using the EVOS M5000 Imaging System with images taken at 10x magnification.

2.9. ELISA

ELISA to measure protein concentration of FGF2 in conditioned supernatants was performed using the Quantikine ELISA kit (cat #DFB50, R&D Systems, Minneapolis, MN), according to manufacturer’s instructions. Absorbance was measured using the Biotek Cytation5 Cell Imaging Multimode Reader with the Gen5 Image Prime 3.10 software (Agilent Technologies Canada Inc., Mississauga, ON). FGF2 concentration was determined following interpolation of a standard curve generated from known concentrations of FGF2.

2.10. Tumour cell proliferation

In the conditioned supernatant transfer model, PC3, 4T1 or K562 cells were seeded at 2500 cells per well in 96-well plate format. Cells were imaged 24 h later using the Biotek Cytation5 Cell Imaging Multimode Reader with the Gen5 Image Prime 3.10 software (Agilent Technologies Canada Inc., Mississauga, ON) to establish cell numbers at time 0. Conditioned media was then added to cells in their normal culture media at a 30:70 volumetric media ratio. Conditioned media were previously collected from adipogenic differentiated MSC at day 10 or day 12 of adipogenic differentiation following treatment with vehicle control or 25 nM doxorubicin at day 9. Non-conditioned media controls consisted of adipogenic media incubated for 24 h or 72 h with vehicle or 25 nM doxorubicin in tissue culture conditions with no cells. Both the conditioned media and the non-conditioned media controls were collected and stored at −20°C until thawing for use. Cells were imaged as before at 24 h, 48 h and 72 h after addition of conditioned media. Cell counting was performed using ImageJ software [34] for each well with 3 wells per condition, and mean cell count for each time point was normalized to cell count at 0 h for PC3 cell experiments. For 4T1 and K562 experiments, cells were counted at 48 hr and mean counts normalized to vehicle controls at the same time point.

In the direct co-culture assay, differentiating MSC were treated at day 9 of adipogenic differentiation with vehicle or 25 nM or 50 nM doxorubicin. At day 12 of differentiation, the media was removed and fresh adipogenic media in the absence of drug containing 10,000 PC3 cells was added directly on top of the day 12 adipogenic differentiated MSC in the adipogenic media. Proliferation of PC3 was assessed by cell counting using ImageJ software [34] from images taken at 24 h, 48 h and 72 h time points using the EVOS XL Core AMEX 1200 (Advanced Microscopy Group, Bothell, WA).

2.11. Statistical analysis

Statistical analyses were performed using GraphPad Prism v9. All data shown represent the mean ± SEM. All data were analyzed with t-test, one-way ANOVA or two-way ANOVA as specified in each figure, and p < 0.05 was considered statistically significant. For one-way and two-way ANOVA tests, Tukey's Honestly Significant Difference post-hoc test was performed.

3. Results

3.1. Doxorubicin treatment of differentiating MSC enhances adipogenesis in vitro

We employed an in vitro model to investigate the effects of low-dose doxorubicin on adipogenesis of MSC to best model the low doses of chemotherapy which are present in the bone marrow of patients following treatment. Human MSC were treated with varying doses of doxorubicin beginning at mid- and late- time points during the adipogenic differentiation process and maturity of the adipogenic differentiated MSC was assessed by BODIPY staining of lipid droplets (Fig. 1A). We found that with treatment beginning at day 5 or day 9 of adipogenic differentiation using three unique MSC cell donors, doxorubicin at 25 nM and 50 nM resulted in an increased percentage of lipid droplet-containing cells, with all conditions reaching statistical significance with exception of the 25 nM dose beginning at day 5 (Fig. 1B&C). Despite the increased number of BODIPY+ adipocytes with doxorubicin treatment observed, we saw no significant changes in the average diameter of BODIPY+ adipocytes with doxorubicin compared to vehicle control treatment (Fig. 1D). The percentage of lipid-droplet containing cells was also assessed in differentiated MSC that were treated with a 100 nM dose of doxorubicin at day 9 of adipogenic differentiation and was found to be significantly decreased compared to vehicle control, indicating that higher doses negatively affect adipogenesis within this model (Fig. S2). We assessed gene expression of two known markers of adipogenesis, PPARG and CEBPA, and found that in the cells treated at day 9 of adipogenic differentiation with 25 nM doxorubicin, gene expression of PPARG was significantly increased compared to the vehicle control treated cells with a trend for increased CEBPA (Fig. 1F). TUNEL assessment confirmed that cells did not appear to be undergoing apoptosis resulting from DNA damage by treatment with 25 nM, 50 nM or 100 nM doxorubicin (Fig. 1G). We additionally examined the effect of another DNA damaging agent on adipogenesis, namely carboplatin. We found similar dose-dependent increases in adipogenesis when differentiating MSC cells were treated with 25 or 50 μM of carboplatin on day 9 of differentiation (Fig. 1E). These findings suggest adipogenic differentiating MSC treated with low-dose chemotherapy displayed enhanced adipogenic capacity.

Fig. 1.

Doxorubicin treatment of differentiating human MSC enhances adipogenic differentiation in vitro. A) MSC undergoing adipogenic differentiation were treated with doxorubicin beginning at day 5 or day 9 with cells collected at day 12 for subsequent analysis with magnified image of BODIPY staining of lipid droplets in adipogenic differentiated MSC shown below B) Representative images of BODIPY stained cells from (A), with C) percentage of BODIPY+ cells quantified from images at day 12 in adipogenic differentiated MSC treated with 25 nM or 50 nM doxorubicin beginning at day 5 or day 9 of adipogenic differentiation. D) Mean diameter of BODIPY+ cells from same images enumerated in (C). E) MSC undergoing adipogenic differentiation were treated with vehicle, 25 μM or 50 μM carboplatin at day 9 of adipogenic differentiation and cells were stained with BODIPY to identify lipid droplets. The percentage of BIODIPY+ cells was assessed and normalized to the vehicle control. F) Gene expression as measured by RT-qPCR of the adipogenic factors, PPARG and CEBPA, in adipogenic differentiated MSC treated with 25 nM or 50 nM doxorubicin beginning at day 5 or day 9 of adipogenic differentiation. Plots illustrate relative quantification following normalization to endogenous control gene expression for RPS13 and HSPCB, and normalized to the vehicle control treated cell levels. G) Representative photos of TUNEL staining in cells at day 12 of differentiation following treatment at day 9 with 25 nM, 50 nM or 100 nM doxorubicin. A positive control consisted of DNase I treatment. Graphs show the mean ± SEM, using the statistical test one-way ANOVA, (*p < 0.05; **p < 0.01), n = 3 biological replicates each with 3 technical replicates.

3.2. Gene expression is altered with doxorubicin treatment of adipogenic differentiating MSC

To investigate possible mechanisms underlying doxorubicin-enhanced adipogenesis, we performed RNA-seq of samples obtained from day 12 cells treated at day 9 of adipogenic differentiation with vehicle control or doxorubicin. Global gene expression was altered in the doxorubicin-treated cells, with 378 significantly altered genes with a log₂ fold-change < -0.5 or > 0.5 (Fig. 2A). GO term analysis using DAVID software with the genes from Fig. 2A revealed significant GO terms (Fig. 2B) (the full list of GO terms is shared as supplementary material). Of note, pathways associated with cellular differentiation and osteoblastic differentiation were significantly altered, with the latter being downregulated. As MSC are also the precursor cell for osteoblasts, which is mutually exclusive to differentiation into adipocytes [40], [41], these results are in line with our hypothesis that doxorubicin enhances adipogenesis. Upregulation of genes related to the GO terms “response to toxic substance” and “response to drug” were anticipated, as doxorubicin is toxic to cells. Genes linked to these GO terms are among the most significantly upregulated in the doxorubicin-treated cells and also have known roles in DNA repair and replication: e.g. TYMS, MDM2 and CDKN1A [42], [43], [44]. We further validated two of these genes, MDM2 and CDKN1A, by RT-qPCR using MSC from three unique donors and similarly found that gene expression of both were significantly increased with doxorubicin treatment in adipogenic differentiated MSC (Fig. S3).

Fig. 2.

RNA-sequencing of day 12 differentiated MSC treated at day 9 of adipogenic differentiation with doxorubicin showed altered gene expression compared to vehicle control treated cells. A) Volcano plot displaying significant alterations in gene expression with 100 nM doxorubicin treatment of the adipogenic differentiating MSC compared to vehicle control treated adipogenic differentiating MSC. Only values with adjusted p-value < 0.05 and log₂ fold-change < -0.5 or > 0.5 are displayed. B) GO terms with significant p-values were generated using DAVID and indicate alterations in gene expression of multiple cellular pathways with doxorubicin treatment. C) Heat map of logarithmic gene counts of members of the FGF pathway within each sample. D) Gene expression of FGF2 by RT-qPCR in human MSC at day 12 following treatment with vehicle control or 25 nM doxorubicin beginning at day 5 or day 9 of adipogenic differentiation. E) Gene expression of GPC1, SOS1, DUSP6, FGFR1 and FGFR2 by RT-qPCR in MSC at day 12 following treatment with vehicle control or 50 nM doxorubicin beginning at day 9 of adipogenic differentiation.

The FGF pathway was also indicated in the DAVID GO term analysis, and upon further analysis, multiple members of the FGF pathway had altered gene expression (Fig. 2C). For example, gene expression of FGF2 ligand was significantly increased in the doxorubicin-treated samples, as was GPC1, which encodes a transmembrane protein that plays a role in secretion of FGF2 [45]. DUSP6, a negative feedback regulator of FGF signaling and SOS1, a protein involved in FGFR signaling to MAPK pathway [46], were also upregulated supporting overall increased FGF pathway activity in doxorubicin treated cells. Using RT-qPCR, we validated that FGF2 gene expression is increased in adipogenic differentiated MSC treated with 25 nM doxorubicin beginning at day 5 or day 9 of adipogenic differentiation, using MSC isolated from three unique donors (Fig. 2D). We further validated the gene expression of other genes involved in the FGF pathway using RT-qPCR (Fig. 2E). Consistent with results from RNA-seq, gene expression of GPC1 and DUSP6 were significantly increased in the doxorubicin-treated cells. SOS1 gene expression also showed a pattern of increased gene expression with doxorubicin treatment, however this comparison did not reach statistical significance. In contrast to the RNA-seq analysis, we found by performing RT-qPCR using MSC isolated from three unique donors that FGFR1 gene expression is significantly increased in adipogenic differentiated MSC treated with doxorubicin beginning at day 9 of adipogenic differentiation compared to vehicle control, whereas FGFR2 gene expression is unaltered.

3.3. Doxorubicin-enhanced adipogenesis is mediated by FGF2 and involves enhanced FGF2 secretion

Given validation of significant alterations in gene expression of FGF pathway members identified by RNA-seq we further investigated the role of the FGF pathway in doxorubicin-enhanced adipogenesis. In addition to validating increased levels of FGF2 mRNA by RT-qPCR in doxorubicin-treated adipogenic differentiating MSC (Fig. 2D), we also confirmed increased levels of FGF2 at the protein level (Fig. 3A). Western blot analysis showed significant increases in the high molecular weight (HMW) isoform and trends for increases of the low molecular weight isoform (LMW) of FGF2 protein in doxorubicin-treated cells at day 10 of adipogenic differentiation (24 h post-treatment). Secreted FGF2 protein levels were also elevated in cell conditioned supernatants collected at day 10 from doxorubicin-treated cells (24 h post-treatment) as detected by ELISA (Fig. 3B). As not all MSC undergo adipogenesis in vitro and thus could contribute to FGF2 levels observed, we measured FGF2 levels in conditioned supernatant from undifferentiated MSC treated for 24 h with either vehicle control or 25 nM doxorubicin and observed that FGF2 secretion was below the range of detection by ELISA (data not shown). This suggests that cells undergoing adipogenesis are likely responsible for the observed increased secretion of FGF2 in response to doxorubicin in our experiments.

Fig. 3.

Doxorubicin increases both intracellular and secreted protein levels of FGF2, and depletion of FGF2 impedes doxorubicin-enhanced adipogenesis. A) Protein expression of FGF2 at day 10 following treatment with vehicle control or 25 nM doxorubicin at day 9. Densitometry of FGF2 bands normalized to GAPDH as an endogenous loading control is shown for the high molecular weight (HMW) and low molecular weight (LMW) isoforms from 3 independent biological replicate blots. B) Secreted FGF2 (pg/mL) as detected by ELISA in conditioned supernatant at day 10 following vehicle control or 25 nM doxorubicin treatment at day 9 of adipogenic differentiation. C) Schematic for adipogenic differentiation of human MSC treated on day 9 of differentiation with vehicle control or 25 nM doxorubicin and subsequently transfected with 100 nM non-targeting siRNA or siRNA targeting FGF2 and assessed at day 12 by RT-qPCR and BODIPY staining. D) RNA isolated as per (C), was used to confirm siRNA-mediated depletion of FGF2 gene expression by RT-qPCR. E) Representative images of BODIPY stained cells treated as per (C) (left), with fold change in the percentage of BODIPY+ cells with doxorubicin treatment as compared to the vehicle control treatment for each siRNA condition enumerated and graphically represented (right). F) Gene expression of PPARG was measured using RT-qPCR at day 12 following treatment as in (C). Graphs show the mean ± SEM. Statistical tests are unpaired t-test in Fig A) and B) and one-way ANOVA in Fig D), E) and F) (*p < 0.05; **p < 0.01; ***p < 0.001), n = 3 biological replicates each with 3 technical replicates.

To establish whether FGF2 plays a direct role in doxorubicin-mediated adipogenesis, we depleted FGF2 using siRNA-interference at day 9 of differentiation, immediately following doxorubicin treatment, and subsequently stained with BODIPY to enumerate the lipid containing cells at day 12 (Fig. 3C). Following effective siRNA-mediated depletion of FGF2 (Fig. 3D), the change in the percentage of lipid droplet-containing cells with doxorubicin treatment was significantly decreased in FGF2 depleted cells as compared to controls (Fig. 3E). Although normalization to vehicle controls in each sample were necessitated in this experiment due to variable efficiency of adipogenesis observed across MSC isolated from individual patients, we observed the expected doxorubicin induced increase in percentage of BODIPY+ adipocytes in vehicle treated non-targeting siRNA transfected cells (∼20% increase on average). If we considered the differences only in doxorubicin treated cells transfected with FGF2-targeting vs non-targeting siRNA (∼30% decrease), the data suggests that FGF2 depletion ameliorated a proportion of the doxorubicin-enhanced adipogenic effects observed. The change in gene expression of PPARG with doxorubicin treatment was also decreased when FGF2 was depleted with siRNA, although these comparisons did not quite reach statistical significance (Fig. 3F). Together, these results suggest that FGF2 contributes to the observed doxorubicin-enhanced adipogenesis.

3.4. Doxorubicin-enhanced adipogenesis leads to enhanced tumour cell proliferation

Given the potential role of BMA in promoting tumour growth in vitro, and our findings that doxorubicin promotes secretion of FGF2 from cells undergoing adipogenesis – a factor which has been shown to be released by visceral adipose tissue to directly promote tumour cell growth [47] – we sought to investigate the effects of doxorubicin-enhanced adipogenesis on tumour cell proliferation. Two co-culture models were employed: direct co-culture and indirect co-culture by conditioned media transfer (Fig. 4A). With the direct co-culture model in which tumour cells were added directly to differentiated BMA in adipogenic media, we observed a significant increase in the number of PC3 tumour cells when co-cultured with adipogenic differentiated MSC previously treated with doxorubicin compared to with vehicle control (Fig. 4B). We also performed a conditioned media transfer model to assess whether PC3 proliferation is enhanced when exposed to secreted factors from doxorubicin-treated adipogenic differentiated MSC. Conditioned media isolated from doxorubicin or vehicle control treated adipogenic differentiating MSC cells at day 10 (Fig. 4C) or day 12 (Fig. 4D) were added at a 30:70 volumetric ratio (conditioned media:PC3 growth media, the ratio at which non-cell conditioned BMA media alone was found to not adversely affect PC3 cell growth) to PC3 cells. We observed that the number of viable PC3 cells was significantly increased following incubation with both day 10 or day 12 conditioned media from doxorubicin-treated compared to vehicle control-treated adipogenic differentiating MSC beginning as early as 48 h following conditioned media treatment. Non-cell conditioned media controls that were incubated for 24 h or 72 h with vehicle control or 25 nM doxorubicin added showed no difference in PC3 growth at 24 h and 48 h post-media addition (Fig. S4).

Fig. 4.

Tumour cell proliferation is increased by culture with or media transfer from doxorubicin-treated adipogenic differentiated MSC and is dependent in part on FGF2. A) Proliferation of PC3 cells was assessed at 24 h, 48 h and 72 h following I) treatment with conditioned media from doxorubicin-treated BMA or II) direct seeding on top of doxorubicin-treated BMA. B) Proliferation of PC3 following direct co-culture with day 12 adipogenic differentiated MSC that were treated at day 9 of adipogenic differentiation with vehicle control or 25 nM or 50 nM doxorubicin. PC3 cells were enumerated at 24 h, 48 h and 72 h following addition to adipogenic differentiated MSC. Proliferation of PC3 treated with 30% conditioned media isolated at C) day 10 or D) day 12 from differentiated MSC treated at day 9 of adipogenic differentiation with vehicle control or 25 nM doxorubicin. E) Proliferation of PC3 in response to treatment with 30% conditioned media isolated at day 10 from adipogenic differentiating MSC following treatment at day 9 with 25 nM doxorubicin and siRNA targeting FGF2 or non-targeting control siRNA. F) Conditioned media was collected at day 10 from differentiating MSC/BMA cells treated with vehicle or 25 nM doxorubicin at day 9 of the differentiation process. K562 or 4T1 tumour cells were treated with 30% adipocyte conditioned media and were imaged and counted 48 h after addition of conditioned media. Mean cell count normalized to vehicle control is presented. In all graphs, mean cell count is normalized to cell count at 0 h prior to addition of conditioned media. Graphs show the mean ± SEM, using the statistical test two-way ANOVA in Figure B) and one-way ANOVA in Figure C), D) and E), (*p < 0.05, ***p < 0.001; ****p < 0.0001), n = 3 biological replicates each with 3 technical replicates.

We confirmed the ability of doxorubicin-treated BMA to promote tumour growth in two additional cell lines known to grow in the bone, 4T1 breast cancer cells and K562 leukemia cells. Using day 10 conditioned media isolated from 3 different MSC lots, we consistently saw a statistically significant ∼15–25% increase in tumour cell growth in the presence of doxorubicin treated BMA-conditioned media compared to vehicle treated BMA-conditioned media (Fig. 4F).

To determine the role of increased FGF2 levels observed in doxorubicin-enhanced adipogenesis on proliferation of PC3 cells, we performed similar experiments as described above following transfer of conditioned media isolated at day 10 from differentiating MSC treated at day 9 of adipogenic differentiation with vehicle control or 25 nM doxorubicin in combination with non-targeting siRNA or siRNA targeting FGF2 (Fig. 4E). At 48 h following addition of conditioned media, the fold-change in cell number was decreased when conditioned media from FGF2-depleted cells was transferred compared to that from non-targeting siRNA. These findings support that FGF2 is at least in part promoting the observed increase in tumour cell proliferation when treated with conditioned media from doxorubicin-treated adipogenic differentiating MSC.

4. Discussion

This study investigated the effects of chemotherapy on bone marrow adipogenesis and the resultant effects on tumour cell growth. While a few patient studies have shown that chemotherapy can increase adiposity of the bone marrow [16], [17], [18], it was unknown whether this was due to increased adipogenesis or hypertrophy. Furthermore, the underlying mechanisms of increased bone marrow adiposity following chemotherapy were unknown. As it has been shown in vitro that bone marrow adipocytes can support and confer chemotherapeutic resistance to tumour cells [19], [20], [21], [22], [23], [24], chemotherapy intended for treatment of the primary tumour may inadvertently prime the bone marrow with high levels of bone marrow adipocytes which could create a supportive microenvironment for maintenance of tumour cell dormancy or metastatic growth.

We have shown that bone marrow mesenchymal stem cell adipogenesis is enhanced in vitro with low-dose doxorubicin treatment at mid- and late- stages of adipogenic differentiation based on evidence of increased percentage of lipid droplet-containing cells and increased gene expression of the adipocyte marker gene PPARG. Although we had expected to also find an increase in CEBPA gene expression in the doxorubicin-treated adipogenic differentiated MSC, CEBPA gene expression was not significantly altered. This finding may perhaps be explained by findings that cold stress on brown adipose tissue causes disassociation of PPARG from CEBPA enhancers [48] and as doxorubicin treatment also causes cellular stress, this could rationalize why CEBPA gene expression is not significantly elevated in the doxorubicin-treated cells as is PPARG. Although we can not entirely rule out a role for adipocyte hypertrophy in our model, our results suggesting increases in adipocyte numbers with no significant changes in adipocyte cell diameter with doxorubicin treatment supports a role for doxorubicin in de novo bone marrow adipocyte differentiation. Together, these results suggest that the increased adiposity of the bone marrow following chemotherapy in clinical and in vivo rodent studies could be due in part to enhanced adipogenesis.

Our findings of the positive effects of doxorubicin on bone marrow adipogenesis are in line with reports in patient and in vivo bone marrow adipose tissue studies [13], [16], [17] but are in contrast to studies on white adipose tissue. This may highlight the differential effects of doxorubicin and other chemotherapies on bone marrow MSC versus pre-adipocytes of white, brown or beige adipose tissue. Alternatively, this may also indicate the effects of differential concentrations of chemotherapy in different organs of the body, with low concentration in bone compared to higher concentration in areas such as breast tissue [49]. The effects of chemotherapy on these other types of adipose tissue have been studied in multiple clinical trials and overall have shown that patients receiving chemotherapy experience reduced total adipose tissue and visceral adipose tissue [50]. In an in vivo study by Biondo et al. [51], doxorubicin administration at 15 mg/kg in rats resulted in a reduction in adipocyte size and glucose uptake following insulin stimulation of adipocytes from the retroperitoneal adipose tissue. They also observed increased cell death and decreased mRNA levels of PPARG and CEBPA in 3T3-L1 cells treated with 1μM doxorubicin [51]. In a separate study, C57BL/6 mice were treated with 2.5 mg/kg doxorubicin and adipose tissue fat pads were found to be lower in mass and extracted adipocytes lower in size and adiponectin secretion [52]. In contrast, brown adipose tissue activity has been shown to be increased in patients treated with doxorubicin [53]. The varying phenotypes of adipose tissue in response to doxorubicin further illustrate the unique properties of adipocytes from white, brown, beige and bone marrow adipose tissue.

Doxorubicin has also been shown in vitro to cause apoptosis, cellular shrinkage and increased ROS production in MSC at doses > 1 µM [54]. Although we did see death of some MSC cells at doses of doxorubicin > 100 nM, we did not observe significant cell death at the concentrations used throughout our study. Moreover, lower doses of chemotherapy are relevant to tumours in the bone as although it is challenging to determine the concentration of drug present in bone following chemotherapy, it is likely that concentrations within the bone are less than the doses previously investigated in vitro. This is supported by recent modelling data which showed doxorubicin accumulation within the bone is at levels less than 250 nM in the hours following treatment which then declines further over time to < 100 nM by approximately 100 h post-treatment; these concentrations are considered to be relatively low compared to accumulation in other organs [49], and are in line with those tested in our experiments. Using low-dose carboplatin treatment is also relevant as it has been shown that only ∼0.003–0.01% of administered doses are found to accumulate in bone in patients [55]. Thus, it remains possible that in some cases, the lower accumulated amount of chemotherapy drug in bone could lead to enhanced adipogenesis.

Additional studies in vitro and in vivo have shown that doxorubicin treatment can affect other bone resident cell populations as well. As an example, although the number of osteoblasts and osteoclasts remained unchanged with doxorubicin treatment in vivo, the activity of these cells can be suppressed [56]. Carboplatin has also been shown to significantly reduce trabecular bone volume via upregulated osteoclast activity in preclinical models [32]. Disseminated tumour cells within the bone are also largely unaffected by chemotherapy while they remain dormant [57], perhaps in part by chemoresistance conferred by bone marrow adipocytes [19], [20], [21]. Based on our findings, chemotherapy may also drive further adipogenesis in the bone, which may facilitate tumour cell reactivation post-treatment, in part via increased number of BMA and their ability to secrete growth factors that may drive tumour cell proliferation.

We performed RNA-seq to identify changes in gene expression in doxorubicin-treated adipogenic differentiated MSC and identified a number of GO-term associated pathways that would be expected to change following doxorubicin treatment, including ‘response to drug’, ‘response to toxic substances’ and ‘apoptosis’. We also validated changes in some of the genes included in these pathways, including changes in MDM2 and CDKN1A (which encodes p21). Although their upregulation is usually associated with cell cycle arrest and induction of apoptosis, it should be noted that they have both previously been implicated in adipogenesis [58], [59]. This may explain why the GO-term ‘apoptosis’ was upregulated and yet we did not observe enhanced apoptosis by TUNEL staining. MDM2 can promote adipogenesis in a fashion dependent on activation of the STAT pathway [58], and p21 can be induced in late adipogenic differentiation by high PPARG levels leading to terminal differentiation and cell cycle exit [59]. Although we validated changes in genes that would promote adipogenesis, these are cell intrinsic factors and would not directly drive the enhanced adipocyte-dependent tumour cell growth we observed (Fig. 4). Multiple other adipogenesis-related genes are also among the most significantly upregulated, including CLCA2 (a gene that encodes a chloride channel regulator and has been positively correlated with lipid droplet size [60]), TLCD4 (a gene that is predicted to encode a membrane-bound protein that may have a role in lipid homeostasis and lipid mobilization [61]) and TREM2 (previously shown to have enhanced gene expression in mature adipocytes [62]). Surprisingly, SDC1 is significantly upregulated in the doxorubicin-treated cells; others have shown that knockdown of SDC1 in adipogenic differentiating MSC resulted in enhanced adipogenesis [63]. Among the downregulated genes in the doxorubicin-treated cells are those related to adipogenesis: GLUL, MT1X, MT2A, PTK2B, CNTFR, CADM3 and TAGLN. GLUL encodes an enzyme that produces glutamine and it has been shown that inability to metabolize glutamine in mice is linked to increased marrow fat [64]. Interference in expression of the metallothionein genes, MT1X and MT2A, has been shown to result in increased lipid accumulation in 3T3-L1 adipocytes [65]. Expression of the protein tyrosine kinase, PTK2B, has been shown to inhibit adipogenesis and osteoblastogenesis of human MSC [66]. CNTFR gene expression has also been shown to be higher in undifferentiated human multipotent adipose-derived stem cells compared to adipocytes [67]. In contrast, a few genes are downregulated which seem to play a role in promoting adipogenesis: CADM3 gene expression has been shown to be driven by the transcription factor PPARG [68], TAGLN overexpression has been linked to enhanced adipogenesis via its regulation of cytoskeleton organization [69], and MXRA8 encodes a protein that initiates glucose transport and synthesis of triacylglycerol, associated with the late stage of adipogenesis [70]. In summary, there are significant alterations in gene expression related to adipogenesis in adipogenic differentiated MSC treated with doxorubicin.

As many of the genes described above were associated with cell intrinsic effects on adipogenesis, we chose to further study another significantly upregulated pathway identified by our RNA-seq analysis that might have the ability to participate in doxorubicin-enhanced adipogenesis and tumour growth: the FGF pathway. We validated that gene expression, protein expression and secretion of FGF2 is increased in doxorubicin-treated compared to vehicle control treated adipogenic differentiating MSC (Fig. 2, Fig. 3). Furthermore, upon siRNA-mediated depletion of FGF2, we found that the enhanced adipogenesis associated with doxorubicin treatment was abrogated, indicating a critical role for FGF2 in doxorubicin-enhanced adipogenesis. While it appeared that FGF2 depletion could ameliorate a proportion of the doxorubicin-induced increased adipogenesis in these experiments, there may be additional factors induced following doxorubicin treatment contributing to adipogenesis that remain to be validated. Results in Fig. 3E also reflect alterations in the percentage of BODIPY+ adipocytes observed in the vehicle treated FGF2 targeting siRNA transfected cells which was used as a normalization control, which appeared to unexpectedly increase compared to non-targeting siRNA transfection. We suspect this apparent ‘increase’ is possibly due to a reduction in MSC cell number, as FGF2 depletion has been shown to affect the proliferation of MSC precursor cells [71]. It is thus possible that our measure of BODIPY+ cells is increased in vehicle control treated siFGF2 treated cells due to decreases in numbers of MSC leading to overall decreased total MSC cell numbers in culture and hence seemingly increased numbers of adipocytes under these conditions. Despite this, our findings are supported by others, who found that stimulation of human breast adipose stem cells (hbASCs) with exogenous FGF2 promoted adipogenesis in a PI3K/Akt dependent manner [72]. We also assessed protein levels related to the PI3K/Akt pathway in vehicle control versus doxorubicin-treated adipogenic differentiating cells but found that there was no general observable trend and high variation in phosphorylated Akt protein levels between lots of MSC preventing us from confirming a role for PI3K/Akt in our model system.

We have demonstrated that secreted factors in conditioned media from doxorubicin-treated adipogenic differentiating MSC enhanced proliferation of multiple tumour cell lines known to grow in bone in both direct co-culture and supernatant transfer models (Fig. 4). In the direct co-culture method, an ∼30–40% increase in PC3 tumour cell growth was observed when tumour cells were co-cultured with doxorubicin treated BMA versus vehicle control treated BMA. In the media transfer experiments we saw similar increases with an ∼40–50% increase in PC3, ∼20–25% increase in 4T1 and ∼30–35% increase in K562 tumour cell growth when 30% v/v conditioned media from doxorubicin treated BMA was added compared to when conditioned media from vehicle control treated BMA was added. Notably, the enhancement of tumour cell proliferation was reduced when conditioned media was harvested from adipogenic differentiating MSC depleted of FGF2 by siRNA, suggesting adipocyte derived FGF2 may be an important contributor to tumour cell growth. Indeed, in support of our findings there is evidence that stromal sources of FGF2 can drive prostate tumour growth, as osteoblasts engineered to secrete increased levels of FGF2 were shown to promote bone metastatic growth of prostate tumours in vivo [73]. The authors further showed that administration of neutralizing antibody to FGF2 inhibited PC3 bone metastases growth in vivo, demonstrating the importance of the growth factor in the bone metastatic process. Clinical relevance of stromal FGF2 has been demonstrated in prostate cancer patients, whereby increased stromal FGF2 expression was associated with higher clinical stage and worse biochemical recurrence-free patient survival [74]. Our findings add to this knowledge and support a role for adipocyte-derived FGF2 in the promotion of not only further adipogenic differentiation but promotion of tumour cell growth. Given that research shows chemotherapy can increase bone marrow adiposity in vitro, validation of our findings in vivo is warranted, as is study of the potential for chemotherapy to create a tumour-supportive niche within the bone marrow.

A strength of our study was the use of bone marrow MSC isolated from multiple donors, thereby increasing the chance that our findings are more broadly applicable across patients. Our study did not identify any differences in chemotherapy induced adipogenesis associated with race in our limited donor sample size. However, it should be noted that in vitro differentiation methods are not fully representative of all signaling pathways and cues that would initiate and maintain adipogenesis within the bone microenvironment [75]. Our study is also limited by the fact that during in vitro differentiation of MSC, only a proportion of treated MSC differentiate into adipocytes, and thus we cannot definitively say whether increases in FGF2 for example are due to increased FGF2 expression within adipocytes or overall increased levels resulting from the increased numbers of adipocytes observed. Irrespective of this, the elevated levels of FGF2 in conditioned media did confer growth advantages to tumour cells in vitro (Fig. 4). Future studies could evaluate the role of FGF2 and FGF signaling in bone marrow adipogenesis and doxorubicin-enhanced adipogenesis using in vivo models. As well, the tumour cell lines used within this study are immortalized cell lines and as such are highly proliferative. These findings suggest that use of chemotherapy such as doxorubicin that is used to treat existing bone metastases [76] could potentially lead to increases in the adipocyte compartment that in turn further support tumour growth in the bone, counteracting the therapeutic effects of these drugs. We did not assess effects of adipocytes on ‘dormant’ tumour cells, which would be more reflective of disseminated tumour cells found in the bone of patients with cancer, which later overcome microenvironmental barriers to form a bone metastasis in a process that can unfold over many years [77], [78]. Thus, it will be important to assess the role of specific anti-cancer drugs used in adjuvant therapies in order to determine whether these agents similarly expand the adipocyte compartment potentially supporting tumour growth in part by facilitating tumour cell ability to overcome microenvironment-enforced dormancy.

As we have established an important role for FGF2 in doxorubicin-enhanced adipogenesis, this opens the door for investigation of anti-FGF2 therapy to mitigate the effects of chemotherapy on bone marrow adiposity. One such therapy is umedaptanib pegol (formerly called RBM-007), an anti-FGF2 aptamer which has been investigated in clinical trials for treatment of wet age-related macular degeneration (NCT04200248, NCT04640272 and NCT04895293) and has been demonstrated to be safe, well tolerated and an effective treatment for this condition as an intravitreal injection [79]. There is ongoing research on the efficacy of FGFR inhibitors for treatment of cancer, including multiple with FDA approval, such as infigratinib, pemigatinib, futibatinib and erdafitinib, which have shown efficacy in treatment of some types of cholangiocarcinoma and bladder cancer [80]. Pemigatinib, an FGFR1/2/3 inhibitor, has also been shown to prevent growth of prostate cancer tumour cells in in vivo preclinical models [81]. We speculate that administration of FGFR inhibitors as an adjuvant to chemotherapy may enhance tumour cell killing by prevention of increased bone marrow adiposity and thus prevention of the protective effects which bone marrow adipocytes can confer to tumour cells. Future in vivo studies could assess the efficacy of co-delivery of doxorubicin and umedaptanib pegol or FGFR inhibitors on bone marrow adiposity and potentiation for tumour growth in bone.

In summary, our study shows that in vitro adipogenesis of bone marrow mesenchymal stem cells is enhanced by doxorubicin or carboplatin treatment in part via the upregulation of FGF2. These finding are relevant as anthracycline or platin-based therapies are frequently used for treatment or palliation of breast, prostate or lung cancer bone metastases, chronic myeloid leukemia, acute myeloid leukemia, osteosarcoma, or Ewing sarcoma [76], [82], [83]. We further demonstrated that secreted factors from these adipocytes promote growth of tumour cells in vitro with FGF2 contributing to the observed enhanced tumour cell growth. FGF2 thus plays a critical role in both doxorubicin-enhanced adipogenesis and release of secreted factors that promote tumour cell proliferation, suggesting its inhibition could be a viable intervention to prevent adipocyte-mediated tumour growth in the bone.

Funding Sources

This work was funded through grants awarded to CLA from the Cancer Research Society (award number 1281149) and from the Prostate Cancer Fight Foundation with money raised by the Motorcycle Ride for Dad Ottawa Chapter. LK was supported by a Doctoral Award from the Cancer Research Society and the Canadian Institutes of Health Research – Institute for Cancer Research (award number 1157762). E. Yakubovich was supported by a scholarship from the Canadian Nuclear Laboratories. SY, E. Yang and JC were supported by Undergraduate Summer Research Awards from the Natural Sciences and Engineering Research Council.

CRediT authorship contribution statement

Lauren M. Kreps: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Edward Yakubovich: Writing – review & editing, Writing – original draft, Software, Methodology, Investigation, Formal analysis, Data curation. Huijun Zhao: Writing – review & editing, Investigation. Selam Yimer: Writing – review & editing, Investigation. Edward Yang: Writing – review & editing, Investigation. Juliet Cruz: Writing – review & editing, Investigation. Barbara Vanderhyden: Writing – review & editing, Funding acquisition. Christina L. Addison: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: