Human adipose-derived MSCs from three different donors (hMSC lines 1-3) were isolated, characterized and expanded as previously described (24) and their identity was confirmed by flow cytometry using a panel of MSC markers CD73, CD90, CD44, CD36, CD45 and CD105 (Supplementary Figure 1). Approval was obtained from the Research Ethics Committee of the Faculty of Health Sciences, University of Pretoria (protocol number 218/2010) and written informed consent was obtained prior to lipoaspirate harvesting from healthy donors undergoing routine plastic or reconstructive surgery procedures. SW1353 chondrosarcoma cells (HTB-94) and SW872 liposarcoma cells (HTB-92) were obtained from ATCC. SaOS-2 (ATCC HTB-85) osteosarcoma cells were a gift from Abe Kasonga (University of Pretoria). Embryonic kidney HEK293FT cells were obtained from Thermo Fisher Scientific, USA. Cell cultures were maintained under standard culture conditions (37°C, 5% CO₂, 65% humidity) in Dulbecco’s Modified Eagle Medium GlutaMax™ culture medium (DMEM, Gibco, Life Technologies/Thermo Fisher Scientific, USA) supplemented with 10-20% fetal bovine serum (FBS, Gibco, USA) and 1% penicillin and streptomycin (pen/strep, Gibco, USA). HEK293FT cells stably express the SV40 large T antigen and the neomycin resistance gene and were cultured with 500 μg/ml geneticin (G-418). At 80% confluence, hMSC cultures were passaged using 0.25% Trypsin/EDTA (Gibco, USA) and re-seeded at a density of 5000 cells/cm². All experiments were performed with hMSCs at passages lower than 20.

Immunophenotype analysis of hMSC lines 1-3 was performed as described previously (25). The hMSCs were positive for CD73, CD90, and CD105 and negative for CD45.

MSCs were plated at a density of 5000 cells/cm² in a 24-well plate and transfected at 60% confluency with either 50 nM siTBX3 #1 (SI03100426, Qiagen, USA), 100 nM siTBX3 #2 (SI00083503, Qiagen, USA), 50 nM sic-Myc #1 (SI00300902, Qiagen, USA), 50 nM sic-Myc #2 (SI02662611, Qiagen, USA) or 50-100 nM control (non-silencing, siCtrl) (1027310, Qiagen, USA). Cells were transfected using HiPerFect^® transfection reagent (Qiagen, USA) according to manufacturer’s instructions.

Lentiviral plasmids pCDH-Empty, pCDH-FLAG-TBX3 and envelope (p.VSV.G) and packaging (psPAX8) expression plasmids were kind gifts from Li Zhao (26). pCDH-FLAG-c-Myc was purchased from Addgene (plasmid # 102626; http://n2t.net/addgene:102626; RRID : Addgene_102626, a gift from Hening Lin) (27) and the FLAG-c-Myc construct was kindly provided by Professor Lüscher from Institut für Biochemie, Germany (28). The human TBX3 promoter luciferase reporter constructs were as previously described (16).

Lentivirus was generated by co-transfecting pCDH-FLAG-TBX3 or pCDH-FLAG-c-Myc with p.VSV.G and psPAX8 plasmids into HEK293FT cells using Lipofectamine LTX transfection reagent (Invitrogen Life Technology, USA) according to manufacturer’s instructions. The empty pCDH-Empty vector (EV) with no insert was used as a control. Viral supernatant was harvested at 48 and 72 h post-transfection and concentrated by ultracentrifugation. Viral titer was determined using limiting-dilution colony formation titering assay. Using polybrene infection reagent (8 μg/mL, Merck, Germany), hMSC line 1 was transduced with viral supernatant [multiplicity of infection (MOI) of 0.3] to achieve TBX3/c-Myc overexpression and subjected to 0.5 μg/ml puromycin (Sigma, USA) for 16 - 18 days until colonies formed. Single cells were grown into colonies and sub-cultured into 24-well plates. The efficacy of stable TBX3 overexpression was confirmed by qRT-PCR, western blotting, and immunofluorescence.

Total protein lysates were prepared as described previously (29). Primary antibodies used were: rabbit polyclonal anti-TBX3 (1:1000; ab99302; Abcam, Cambridge, UK), mouse monoclonal anti-β-actin (1:3000; sc-47778), mouse monoclonal anti-NANOG (1:500; sc-374001), rabbit polyclonal anti-cyclin A (1:1000; sc-751), mouse monoclonal anti-CDK2 (1:500; sc-6248), mouse monoclonal anti-MDM2 (1:500; sc-965), rabbit polyclonal anti-p16 (1:250; sc-759), rabbit polyclonal anti-p19/p14 (1:250), mouse monoclonal anti-p53 (1:500; sc-126), mouse monoclonal anti-SNAI (1:500; sc271977), mouse monoclonal anti-SLUG (1:500; sc-166476), mouse monoclonal anti-TWIST1 (1:500; sc-81417) from Santa Cruz Biotechnology Inc (Texas, USA); mouse monoclonal anti-MMP9 (1:1000; 4A3 NBP2-13173), mouse monoclonal anti-MMP2 (1:500; 8B4 NB200-114) from Novus Biotechnologicals (Colorado, USA); mouse monoclonal anti-cyclin B1 (1:1000; V152), rabbit monoclonal c-Myc (1:1000; D84C12) and rabbit polyclonal anti-vimentin (1:1000, R28) from Cell Signaling Technology (Massachusetts, USA); rabbit polyclonal anti-p38 (1:5000; M0800), from Sigma-Aldrich (Missouri, USA). Secondary antibodies were used at 1:5000 dilution and included horseradish peroxidase (HRP)-conjugated goat anti-rabbit (Biorad, USA), goat anti-mouse (Biorad, USA). Densitometry was performed using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e) (30) and normalised to the appropriate loading control. All blots are representative of at least two independent repeats.

Total RNA was extracted using the High Pure RNA Isolation Kit (Roche, Germany) according to the manufacturer’s instructions, and reverse transcription was performed using 500 ng RNA and the InProm-II™ reverse transcription system (Promega A3800, USA) according to the manufacturer’s instructions. The reactions were carried out using 1 μL cDNA, 2x SYBR green master mix (Applied Biosystems, USA) and primers to amplify the human TBX3 (QT00022484, Qiagen, USA), GUSB (QT00046046, Qiagen, USA), c-Myc (F: CTGAGACAGATCAGCAACAACC; R: TTGTGTGTTCGCCTCTTGAC, Integrated DNA Technologies, USA), p16^INK4a (F: GTGGACCTGGCTGAGGAG; R: CTTTCAATCGGGGATGTCTG, Integrated DNA Technologies, USA). qRT-PCR was performed using the Light Cycler^® 2.0 system (Roche, Switzerland) and the following parameters: denaturation for 15 min at 95°C and combined annealing and extension for 40 cycles at 60°C for 1 min. Samples were prepared in triplicate and non-template controls were run to detect contamination or nonspecific amplification. The 2^−ΔΔCt method was employed to analyze results, and relative mRNA expression levels of c-Myc, TBX3, and p16^INK4a were normalized to mRNA levels of glucuronidase-beta (GUSB).

Immunofluorescence was performed as previously described (16). Briefly, cells were incubated with rabbit polyclonal anti-TBX3 antibody (1:100 dilution; ab99302, Abcam, USA) at 4°C overnight. Subsequently, cells were incubated with fluor-conjugated secondary Cy3 donkey anti-rabbit IgG (1:1000 dilution; Jackson ImmunoResearch Laboratories, Inc., USA). To visualize nuclei, cells were incubated with 1 mg/mL Hoechst 33258 (Thermofisher Scientific, South Africa). Fluorescent cells were viewed using an Axiovert confocal microscope (Zeiss, USA).

Short-term growth was measured as described previously (29). hMSCs were seeded in triplicate at a density of 5000 cells/cm² in 24-well plates. Cells were collected by trypsinization and counted on a haemocytometer at 2–3 days intervals. As an alternative assay for proliferation, cell viability was determined using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) Cell Proliferation Kit (Roche, Switzerland). Cells were seeded in quadruplicate in 96-well plates (5000 cells/cm²) and cell viability determined over 5 days according to the manufacturer’s instructions. Absorbance (595 nm) was measured using GloMax^® plate reader (Promega, USA) and the absorbance of the medium only control was subtracted from the samples.

Cell migration was measured using a two-dimensional (2D) in vitro scratch motility assay as previously described (31). The wound areas were measured over time and calculated using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e).

Cells were seeded at 100 cells/35 mm dish in triplicate and the culture medium was changed every 3 days. After 18 days, the colonies were washed twice with 1x PBS, fixed with ice-cold methanol for 15 min, stained with 1% crystal violet dye (C3886, Sigma, USA) for 1 h, and washed once with 1x PBS and twice with tap water to remove excess crystal violet. Photographic images were captured to count the colonies.

Cells were seeded at 5000 cells/cm² in 6-well plates and 48 h later they were stained using a Senescence β-Galactosidase Staining Kit (#9860, Cell Signaling Technology, USA) following the manufacturer’s protocol. Images were taken using an EVOS M5000 Imaging System microscope (Thermo Fisher Scientific, USA).

To establish spheroids that finally consist of 5000 cells/spheroid, cells were suspended in DMEM at 50,000 cells/mL and 100 μL of this suspension was plated per well in a 96-well plate coated with 1.2% agarose (SeaKem LE Agarose 50004, Lonza, USA) to prevent cell adhesion. Cells were incubated for 4 days to allow for compact spheroid formation before proceeding with the experiments described below.

Images (EVOS XL AMEX Core Imaging System) were taken (t=Day 1) and cell growth was monitored over five days (t=Day 5). The area of spheroid core and periphery was measured using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e) and the ratio of spheroid core to periphery was calculated.

The following three fluorescent stains were used to determine cell viability: calcein AM (1 mg/mL, C1430, Invitrogen, Massachusetts, USA), propidium iodide (PI) (1 mg/mL), and 4′,6-diamidino-2-phenylindole (DAPI) (500 µg/mL, Thermo Fisher Scientific, USA). A 2x staining solution in medium was prepared by mixing calcein AM, PI, and DAPI to working concentrations of 2 µM, 8 µg/mL, and 20 µg/mL, respectively. Spheroids were stained by removing 50 µL of medium and replaced with 50 µL of the 2x staining solution, for final well concentrations of 1 µM, 4 µg/mL, and 10 µg/mL, respectively. Plates were then incubated at 37°C and 5% CO₂ for 60 min and imaged using an EVOS M5000 Imaging System microscope (Thermo Fisher Scientific, USA). Calcein AM staining intensity was measured using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e) and calculated relative to the spheroid area.

Spheroids were formed as described earlier and each spheroid was then transferred to individual wells of a 96-well plate where they were embedded in 70 μL of 1.5 mg/mL collagen type I rat tail matrix (Gibco, A1048301, Thermo Fisher Scientific, USA) and covered with 100 μL of complete media. Cell invasion was monitored using an EVOS M5000 Imaging System microscope (Thermo Fisher Scientific, USA) and images were taken after 0, 24, and 48 h. The invasive area was determined by calculating the difference between the final area (t=24, 48 h) and the initial area (t=0 h) using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e), and data were normalized to control cells. Three independent experiments including four replicates for each condition were performed.

Dishes (35 mm) were layered first with 0.6% agar prepared in cell culture medium followed by 0.3% agar prepared in cell culture medium containing 5000 cells/dish. The dishes were incubated for 12 weeks, and colonies were imaged using an EVOS M5000 Imaging System microscope (Thermo Fisher Scientific, USA). Relative colony area was measured using the image analysis software Fiji (Version 2.0.0-rc-68/1.52e), and data were normalized to control cells. Two independent experiments including three replicates for each condition were performed.

ChIP assays were performed as previously described (32). Briefly, hMSCs were fixed in 1% formaldehyde, the chromatin extracted and sonicated to obtain DNA fragments of between 300 and 500 bp. Protein-bound DNA was immunoprecipitated using antibodies against c-Myc (8 μg; E5Q6W, Cell Signaling, USA) or normal rabbit IgG (8 μg; sc-2729, Santa Cruz Biotechnology Inc, USA). The DNA to which c-Myc bound was purified using the phenol-chloroform extraction method and precipitated DNA was analysed by qRT-PCR using primers spanning TBX3 E-box -1936; -1789 (forward, 5′- GGGAATTCTCAACGCTGG -3′; reverse, 5′ CAGCACAGGCCTCTCTCG 3′); E-box -1210 (forward, 5′- GAG AAG ATA CCA GGC TGG C-3′; reverse, 5′ CAT ATT CCA CCT GGA TGT GGG3′); E-box -701 (forward, 5′- GAG ACC AGC ACC GAG ACA C-3′; reverse, 5′ GGC CAC TCG GTT CTA CAA AAG-3′) and GAPDH (forward, 5′-GAAGGCTGGGGCTCATTT-3′; reverse, 5′-CAGGAGGCATTGCTGATGAT-3’). Crossing values (Ct) of c-Myc and IgG precipitated DNA were adjusted by normalizing against the Ct value of 1% of input DNA and the ΔΔCt method was used to determine fold enrichment using the following equation: 2^{-(ΔCt1−ΔCt2)} (ΔCt1 = c-Myc, ΔCt2 = IgG).

For luciferase assays, hMSCs were plated in 12-well plates at a density of 5000 cells/cm². The next day, 500 ng of the TBX3 luciferase reporter plasmid and increasing amounts (50, 100 and 200 ng) of the FLAG-c-Myc expression vector or empty vector were transfected using Lipofectamine LTX (Invitrogen Life Technology, USA) according to manufacturer’s instructions. The vector pRL-TK, which contains the thymidine kinase promoter driving expression of a renilla reporter, was included as an internal control for transfection efficiency (20 ng per transfection). Cells were lysed 48 h after transfection, and firefly and renilla luciferase activities were determined using the Dual-Luciferase Reporter Assay System (Promega, USA) according to the manufacturer’s instructions and measured using a Luminoskan Ascent luminometer (Thermo Labsystems, USA). Extracts were subjected to western blot analysis to confirm expression of transfected c-Myc protein.

Total RNA was extracted from EV (EV #1, EV #2) and TBX3 (TBX3 #1, TBX3 #2) hMSCs (three biological replicates per clone) using the High Pure RNA Isolation Kit (Roche, Germany) according to the manufacturer’s instructions and was used for microarray gene expression analysis on the human Affymetrix Clariom S array. Briefly, a total of 5 ng RNA from each sample was used to synthesize first strand cDNA, followed by 3’ adaptor and double stranded cDNA (ds-cDNA) using the Clariom S array (IVT) Pico kit according to the manufacturer’s protocol. The ds-cDNA was used to synthesize cRNA by in vitro transcription for 14 h at 40°C and the cRNA was purified using purification beads on a magnetic stand. A total of 20 µg purified cRNA was used for 2nd-cycle ds-cDNA synthesis followed by RNA hydrolysis and ds-cDNA purification as described above. Thereafter, 5.5 µg of each sample was used for fragmentation and labelling, followed by hybridization on the Clariom S array in an Affymetrix GeneChip^® Hybridisation Oven-645 rotating at 60 rpm at 45°C for 16 h. Hybridised chips were washed and stained using GeneChip™ Hybridization, Wash and Stain Kit in an Affymetrix GeneChip^® Fluidics Station-450Dx before scanning using an Affymetrix GeneChip^® Scanner-7G. From each scanned chip, the Affymetrix system generates a CEL file which has intensity values for all probes present on it. The CEL files were imported into the Affymetrix Transcriptome Analysis Console (TAC) v4.0.2 software for differential gene expression analysis. Genes considered to be differentially expressed (DEGs) between EV and TBX3 hMSCs were those with fold-change (FC) values ≥ 2 or ≤ −2 and ANOVA adjusted-p value < 0.05. The DEGs between EV and TBX3 hMSCs were further visualized on a volcano plot and hierarchical clustering using TAC 4.0.2 software. The microarray data files of this study have been deposited in NCBI GEO (Gene Expression Omnibus) with assigned GEO accession number GSE183848.

Gene ontology (GO) biological process and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using g:Profiler (version e104_eg51_p15_3922dba) with the Benjamini-Hochberg FDR method applying a significance threshold of 0.05 (33). To identify gene signatures or pathways enriched in TBX3 hMSCs, gene set enrichment analysis (GSEA) was performed with the GSEA program (v.4.1.0) (34, 35). The Broad Molecular Signatures Database (MSigDB v7.4) set H (hallmark gene sets, 50 gene sets, h.all.v7.4.symbols.gmt) was selected as the reference. The permutation number was set to 1,000 and cut-off values of nominal p  <  0.05 and FDR  <  0.25 were selected.

Statistical significance was determined using a student’s t-test (Excel, Microsoft, Redmond, WA, USA). Significance was accepted at *p < 0.05, **p < 0.01, ***p < 0.001. Unless otherwise stated, all data were obtained from at least three independent experimental repeats with error bars representing standard error of the mean (SEM). Graphs were generated using GraphPad Prism software 6.0 (GraphPad Prism software, USA).

To explore whether the c-Myc/TBX3 axis exists in hMSCs we first investigated whether, as is the case in sarcomas, c-Myc transcriptionally activates TBX3. To address this, we determined whether altering c-Myc levels resulted in a corresponding change in TBX3 levels in hMSCs. Indeed, we show that depleting c-Myc by siRNA (sic-Myc #1 or sic-Myc #2) resulted in decreased levels of TBX3 mRNA (Figure 1A) and protein (Figure 1B) and overexpressing c-Myc resulted in increased TBX3 levels (Figure 1C). These results show that c-Myc is indeed upstream of TBX3 in hMSCs. The TBX3 promoter has four highly conserved c-Myc consensus E-box sites at positions -1936, -1789, -1210 and -701 base pairs (bps) (Figure 1D). ChIP assays showed that while c-Myc did not occupy the regions of the TBX3 promoter containing E-box sites at -1936; -1789 and -1210 bps, its binding to the canonical E-box site at -701 (CACGTG) was enhanced by 3.88-fold (Figure 1E). Luciferase reporter assays confirmed that 200 ng c-Myc significantly activated the TBX3 promoter in hMSCs (Figure 1F), and importantly, mutation of the E-box at -701 bp abrogated this activation (Figure 1G). Western blot analysis of lysates from luciferase assays confirmed expression of transfected c-Myc in these experiments (Figures 1F, G). Together these results confirm that, in hMSCs, the TBX3 promoter is transcriptionally activated by c-Myc through the -701 bp E-box motif.

If the upregulation of TBX3 by c-Myc is a key molecular mechanism involved in transforming hMSCs to sarcomas, then one would expect that parental hMSCs will have lower levels of TBX3 than TBX3-driven sarcomas. Therefore, before we engineered hMSCs to overexpress TBX3, we compared the status of TBX3 mRNA and protein in three adipose-derived hMSC cell lines (hMSC line 1-3) as well as chondrosarcoma (SW1353) and liposarcoma (SW872) cells by qRT-PCR and western blotting respectively. Our results show that TBX3 mRNA and protein levels were lower in hMSCs compared to chondro- and liposarcoma cells (Figures 2A, B). We speculated that endogenous TBX3 contributes to hMSC proliferation and migration and that during the malignant transformation of hMSCs, TBX3 is upregulated and promotes these as well as other oncogenic processes. Indeed, transiently depleting TBX3 by siRNA (siTBX3) significantly inhibited hMSC proliferation (Figure 2C), decreased levels of the cell cycle progression markers cyclin A and CDK2, increased the negative cell cycle regulators p14^ARF and p53, and decreased levels of the negative p53 regulator, MDM2 (Figure 2D). Furthermore, depleting TBX3 retarded the migratory ability of hMSCs (Figure 2E). We next investigated whether TBX3 promotes hMSC proliferation downstream of c-Myc. To this end, TBX3 was depleted in hMSCs stably overexpressing c-Myc (c-Myc siTBX3) or hMSC empty control (EV siTBX3) cells and the impact on cell proliferation measured. Growth curve assays show that overexpressing c-Myc (c-Myc siCtrl) significantly enhanced hMSC proliferation and depleting TBX3 (EV siTBX3) inhibited their proliferation (Figures 2F, G). Importantly, depleting TBX3 abrogated the pro-proliferative activity of c-Myc (c-Myc siTBX3). Together, these data suggest that the upregulation of TBX3 by c-Myc is a key downstream event in mediating the ability of c-Myc to promote hMSC proliferation.

To determine if TBX3 overexpression in hMSCs can promote their transformation, we first established hMSCs that stably overexpress TBX3 using a lentiviral transduction system (Supplementary Figure 2). Briefly, hMSCs were transduced with lentiviruses delivering FLAG-TBX3 or Empty vector (EV) and resistance to the mammalian antibiotic puromycin enabled selection of successfully transduced cells. In total, 13 EV hMSC and 15 FLAG-TBX3 hMSC puromycin-resistant clones were obtained and they all tested positive for FLAG-TBX3 (data not shown). Figures 3A –C show EV hMSC (hereafter referred to as EV #1 and EV #2) and FLAG-TBX3 hMSC (hereafter referred to as TBX3 #1, TBX3 #2 and TBX3 #3) clones that were selected for further analyses. Compared to EV hMSCs, TBX3 hMSCs expressed significantly higher levels of TBX3 protein (Figure 3A), mRNA (Figure 3B) and nuclear localization (Figure 3C). Furthermore, TBX3 protein levels in the TBX3 clones were comparable to those in chondrosarcoma (SW1353), liposarcoma (SW872) and osteosarcoma (SaOS-2) cells (Figure 3A). The immunofluorescence results shown in Figure 3C confirm that TBX3 localizes to the nucleus and that the FLAG-tag therefore did not affect its localization and consequently its function as a transcription factor.

Stemness is a feature of cancer cells that drives tumor progression by enhancing self-renewal capacity, survival, proliferation, and metastasis (36). While expanding the EV and TBX3 hMSCs we noticed that with increasing ex vivo passages the EV hMSCs exhibited a flattened and senescent-like morphology but TBX3 hMSCs at the same passages maintained an elongated and spindle-shape morphology like early passage parental hMSCs (Passage 3) (Figure 4A). Furthermore, compared to EV hMSCs, TBX3 hMSCs had on average a 3.67-fold enhanced colony forming ability (Figures 4B, C) and expressed higher levels of the stem cell marker NANOG (Figure 4D). Together, these findings suggest that overexpressing TBX3 in hMSCs promotes their stemness characteristics and prevents them from differentiating and undergoing senescence.

We next investigated whether, when overexpressed in hMSCs, TBX3 indeed functions as an anti-senescence factor. Our results show that, compared to EV hMSCs, TBX3 hMSCs exhibited substantially less SA-β-gal activity (Figure 5A) and expressed significantly lower levels of the senescence marker p16^INK4a (Figures 5B, C). These results are important because for cancer cells to achieve a state of immortality, they need to escape cellular senescence.

The ability to sustain persistent proliferation is one of the most fundamental traits of cancer cells (37). MTT and growth curve assays showed that TBX3 significantly promoted hMSC viability and proliferation (Figure 6A) respectively, which was associated with an increase in cyclin A, CDK2 and cyclin B1 (Figure 6B). Furthermore, TBX3 hMSC had a decrease in p14^ARF, an increase in MDM2 (a negative regulator of p53), and a corresponding decrease in p53 levels (Figure 6C). These results are consistent with previous studies that showed that TBX3 promotes proliferation and immortalization by interfering with the p14^ARF-MDM2-p53 pathway (17, 22, 38, 39). To form tumors as well as to migrate and metastasize, cancer cells have to exhibit anchorage-independent cell growth (40). Results from soft agar assays show that TBX3 hMSCs proliferated and formed colonies (indicated by red arrows) in the absence of a substrate whereas EV hMSCs remained as single cells (Figures 6D, E).

Cell migration and invasion of cancer cells into surrounding tissue and lymphatic and/or vascular systems is the initial step of tumor metastasis (41). To explore whether TBX3 promotes the migratory ability of hMSCs, we performed two-dimensional scratch motility assays and the results show that TBX3 hMSCs migrated significantly faster than EV hMSCs (Figure 6F). This was associated with an increase in the mesenchymal migration marker vimentin and the transcription factors SLUG, SNAIL and TWIST1 (Figure 6G).

To test whether overexpression of TBX3 also promotes hMSC viability, proliferation, and invasion under more physiologically relevant conditions, we generated 3D spheroids of EV hMSCs and TBX3 hMSCs. 3D spheroid culture models are superior to 2D cell cultures because they better mimic the structure of the natural in vivo cell environment such as cell-cell and cell-extracellular matrix interactions (42). The periphery of TBX3 hMSC spheroids was significantly larger than that of the EV hMSC spheroids which is indicative of viable and proliferating cells (Figures 7A, B). Calcein AM staining for live cells (green) and propidium iodide staining for dead cells (red) confirmed that the cells on the periphery of the TBX3 hMSC spheroids are indeed more viable and proliferative than those on the periphery of the EV hMSC spheroids (Figures 7C, D). These results were accompanied by a downregulation of p16^INK4a and an increase in cyclin A and cyclin B1 (Figure 7E).

We next investigated the ability of hMSC spheroids to invade collagen matrices which are organized 3D structures that mimic a tumor micro-region (43). Our results show that, compared to EV hMSC spheroids, TBX3 hMSC spheroids were significantly more invasive at 24 h and 48 h (Figures 8A, B) which correlated with higher levels of the invasion markers MMP2 and MMP9 (Figure 8C).

To elucidate the molecular pathways that TBX3 impacts to induce a transformed phenotype in hMSCs, we performed microarray analysis with RNA from TBX3 hMSCs (TBX3 #1, TBX3 #2) and EV hMSCs (EV #1, EV #2). The results identified a total of 1256 differentially expressed genes (DEGs) (903 upregulated and 353 downregulated in TBX3 hMSCs) (Figures 9A, B). Gene ontology analysis revealed that the significantly upregulated genes were enriched in biological processes related to different aspects of the cell cycle including mitosis and cell division, and the significantly downregulated genes were enriched in biological processes such as development, cell adhesion and differentiation (Figure 9C). Furthermore, gene set enrichment analysis (GSEA) of the DEGs showed significant differences (FDR < 0.25, NOM p-value < 0.05) in the enrichment of MSigDB Collection Hallmarks (h.all.v7.4.symbols.gmt) in TBX3 hMSCs including gene sets related to E2F targets, MYC targets, the G2/M checkpoint, mitotic spindle and DNA repair (Figure 9D). Analysis of the DEGs revealed that they correlated with the biological processes that TBX3 was shown to promote in this study such as cell cycle progression, migration, and invasion (Table 1) and KEGG analysis confirmed that the genes upregulated in TBX3 hMSCs were involved in several oncogenic signaling pathways (Table 2). Importantly, several of the top upregulated genes (FC > 5) were associated with sarcomagenesis (Supplementary Table 1) and 57 of the upregulated genes in TBX3 hMSCs were present in the CINSARC (Complexity INdex in SARComas) database which contains 67 genes associated with sarcoma aggressiveness and poor prognosis (44) (Table 3). Together, these results reveal the genes and pathways that TBX3 impact to promote cellular transformation in hMSCs and provide evidence of a key role for TBX3 in sarcomagenesis.