Sodium dodecylsulfate (SDS), bovine serum albumin, dimethylsulfoxide (DMSO), TGF, and galunisertib were sourced from Sigma (Oakville, ON, Canada). Electrophoresis reagents came from Bio-Rad (Mississauga, ON, Canada). The enhanced chemiluminescence reagent was from Amersham Pharmacia Biotech (Baie d’Urfé, QC, Canada). Micro bicinchoninic acid protein assay reagents were from Pierce (Rockford, IL, USA). The rabbit polyclonal antibody against GRP78 (SC-13968) and fibronectin (SC-8422) was from Santa Cruz Biotechnology (Dallas, TX, USA). Polyclonal antibodies against TGF-β (56E4, 3709 S), AKT (11521, 9272 S), phospho-AKT Ser473 (12604, 4060 S), Snail (C15D3, 3879 S), Smad2 (D43B4, 5339 S), Smad3 (C67H9, 9523 S), phospho-Smad2 Ser465/467 (138D4, 3108 S), and phospho-Smad3 Ser423/425 (C25A9, 9520 S) were from Cell Signaling Technology Inc. (Danvers, MA, USA). The monoclonal antibody against GAPDH (D4C6R, 97166 S) was from EMD Millipore (Oakville, ON, Canada). Horseradish peroxidase-conjugated secondary antibodies were from Jackson ImmunoResearch Laboratories (West Grove, PA, USA).

Human U87 GBM-derived cells and culture media were from the American Type Culture Collection (ATCC; Manassas, VA, USA). According to the manufacturer, mycoplasma contamination was eliminated in September 1975 from these cells. This was confirmed bi-annually in our laboratory, and we found no contamination. The STR profiling, Y-chromosome paint, and Q-band assay, provided by the manufacturer, confirm that the cell line was male in origin, and likely a GBM of CNS origin (https://www.atcc.org/products/htb-14). Cells were cultured as 2D monolayers without antibiotics in Dulbecco’s modified Eagle’s medium (DMEM) containing 10% fetal bovine serum (FBS). Cells were kept subconfluent and used over a maximum of eight passages, with a 1:5 bi-weekly split, in a humidified incubator at 37°C with 5% CO₂.

Intracellular 2DG-mediated ATP depletion was measured as described elsewhere using HPLC separation of purine nucleotides carried out with an Agilent 1260 Infinity system (Agilent Technologies) with the multiple-wavelength detector set to measure absorbance at 248, 254, and 262 nm simultaneously. Quantification of nucleotides was performed by peak integration of the area under the curve and verified using external standards of known concentration as determined by the molar extinction coefficient [23].

The formation of 3D spheroids was obtained using the hanging drop technique, which involves suspended inverted cell droplets and culturing cells in 100 mm non-adherent Petri dishes (low-attachment plates) to prevent substrate adhesion as previously described [21, 24]. The lids were removed, and 10 mL of sterile water was added to the bottom of each dish to create a hydration chamber. Using a multichannel pipette, 38 μL drops of cell suspension (at densities of 38,000 cells per drop) were carefully dispensed onto the inner surface of the inverted lid, ensuring sufficient spacing to prevent contact between drops. Up to 40 drops were placed per lid. The lid was then inverted over the water-filled bottom chamber and incubated at 37°C in a humidified atmosphere with 5% CO₂ and 95% humidity for 5 days. The droplets were monitored daily under a microscope to assess cellular aggregate formation. Spheroids were either analyzed directly or transferred to 96-well round-bottom plates (FALCON #351177) coated with poly(2-hydroxyethyl methacrylate) [poly-HEMA] (P3932, Sigma-Aldrich, Oakville, ON, Canada) to prevent cell adhesion to the growth surface. Each well contained 250 μL of fresh complete medium.

Spheroid morphology was examined under a Nikon Eclipse TE2000-U inverted microscope (Nikon Instruments Inc., Melville, NY, USA). Phase-contrast micrographs were taken with a Q Imaging QICAM-IR Fast 1394 Digital CCD camera using the NIH μManager software (National Institutes of Health, Bethesda, MD). For 96-well plates, an Agilent BioTek Cytation 5 cell imaging multimode reader with a 4× objective was used to image the entire well using the bright field imaging channel using Gen5 software. Quantification of spheroid size was performed using QuPath 0.5.1 along with the Segment Anything Model (SAM) extension, a validated tool that employs deep learning techniques to segment objects within images by identifying their contours and features [25].

Origin of electrophoresis reagents, total cell lysis procedure, SDS-polyacrylamide gel electrophoresis, electrotransfer to low-fluorescence polyvinylidene difluoride membranes, and immunodetection were conducted as described previously in detail [26]. Immunoreactive material was visualized by enhanced chemiluminescence.

U87 cells were transiently transfected with small interfering RNA (siRNA) sequences using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific). Gene silencing was performed using 20 nM siRNA against TGF-β (Hs_TGFB1_2 FlexiTube siRNA GeneGlobe ID: SI00013601), or scrambled sequences (AllStar Negative Control siRNA, 1027281) synthesized by QIAGEN (Valencia, CA, USA) and annealed to form duplexes. Gene silencing efficacy was assessed by real-time quantitative PCR (RT-qPCR) as described below.

Total RNA was extracted from 2D cell monolayers and 3D spheroids using TRIzol, following the manufacturer’s recommendations (Life Technologies, Gaithersburg, MD, USA). Total RNA concentration was measured using a NanoPhotometer P330 (Implen), and 2 μg was reverse-transcribed into cDNA using a high-capacity cDNA reverse transcription kit (Applied Biosystems; Foster City, CA, USA). Samples were prepared using primer sets and SsoFast EvaGreen Supermix (Bio-Rad; 1725204). The following QuantiTect primer sets were provided by QIAGEN: FN1 (Hs_FN1_1_SG QT00038024), SNAI1 (Hs_SNAI1_1_SG QT00010010), TGF-β (Hs_TGFB1_1_SG QT00000728), VEGF (Hs_VEGFA_1_SG QT01010184), GLUT1 (Hs_SLC2A1_1_SG QT00068957), HIF-1α (Hs_HIF1A_1_SG QT00083664), GAPDH (Hs_GAPDH_2_SG QT01192646) and PPIA (Hs_PPIA_4_SG QT01866137). Gene expression was quantified by RT-qPCR on a CFX Connect (Bio-Rad) with the Bio-Rad CFX Manager software version 3.0. The relative RNA quantities of each target gene were normalized against two housekeeping genes, GAPDH and PPIA, using the standard 2^–ΔΔCq method.

Premade RT² Profiler PCR arrays for Human Cell Death (PAHS-212Z) and for Human Epithelial-Mesenchymal Transition (PAHS-090Z) were purchased from QIAGEN and were used following the manufacturer’s instructions. Briefly, the genomic DNA was eliminated before reverse-transcribing 4 μg of total RNA using the RT² First Strand Kit (QIAGEN, 330404). Each plate was used to assess one cDNA sample prepared with the RT² SYBR Green qPCR Master Mix (QIAGEN, 330502). The relative expression analysis of 84 genes, as well as controls, was done through the GeneGlobe Analysis Center, a website provided by QIAGEN (https://geneglobe.qiagen.com/us/analyze), using the standard fold change 2^–ΔΔCq method. Fold regulation was used in the figures; for upregulated genes, fold regulation is the same as fold change, but for downregulated genes, fold regulation is calculated as 1/(fold change).

The STRING database v11 (https://www.string-db.org/) was used to identify and build protein-to-protein interaction networks [27], with a confidence score of 0.4. The maximum number of interactions shown was set to 10 to help with the readability of the predictions.

All statistical analyses were conducted using the GraphPad Prism 7 software (Dotmatics). Data and error bars are expressed as the mean ± standard error of the mean from three independent experiments, unless otherwise specified. Hypothesis testing was performed using the Kruskal‑Wallis test followed by a Dunn’s post-hoc test (> 2 groups). P < 0.05 (*) values are considered significant and are indicated in the figures.

ATP serves as the primary energy currency supporting cytoskeletal remodeling, cell–cell adhesion, and vesicular trafficking necessary for 3D spheroid compaction [22]. Therefore, investigating ATP requirements during the 3D in vitro GBM spheroid formation provides mechanistic insight into how energy metabolism may govern 3D tumor organization and acquisition of an invasive phenotype. Here, we assessed the impact of glycolytic inhibition on EMT in 3D spheroids treated with increasing concentrations of 2DG. Morphological analysis revealed that 2DG led to dose-dependent disruption of spheroid integrity, indicating reduced cell–cell adhesion under metabolic stress (Figure 1A). RT-qPCR confirmed higher VEGF, GLUT1, and HIF-1α transcript levels upon spheroid formation, indicating activation of hypoxia and angiogenesis adaptive responses (Figure 1B). Importantly, 2DG treatment was found to prevent these increases, supporting the spheroid-disrupting events as a consequence of ATP depletion (Figure 1B). Western blotting showed elevated TGF-β, fibronectin, and phospho-Smad2 (P-Smad2) expression in 3D spheroids compared to 2D monolayers, consistent with enhanced EMT (Figure 1C), and this was confirmed by increased other EMT biomarkers using gene array screens, including TGF-β signaling intermediates, ECM organisation proteins, and cell surface integrins (Table S1). While GRP78 expression increased upon spheroid formation, 2DG treatment further potentiated that effect and led to a decreased EMT phenotype, suggesting 2DG-mediated glycolytic inhibition altered TGF-β signaling pathways. Finally, along with its increased expression at the protein level in spheroids, TGF-β transcript levels were also reduced upon 2DG treatment (Figure 1D). These findings demonstrate that 3D culture promotes EMT and that glycolytic inhibition potentially alters TGF-β autocrine signaling. In addition, reduced AKT phosphorylation also appears indicative of a cell death phenotype (Figure 1C). Decreased total Akt and P-Akt in 3D spheroids can rationally be a reflection of biologically meaningful shifts driven by the 3D microenvironment. Spheroid architecture alters growth factor access, cell–matrix signaling, metabolic state, phosphatase activity, and subpopulation composition; any of these can reduce Akt protein abundance or phosphorylation while stemness phenotypes are maintained or regulated through alternative pathways [28]. Importantly, this suggests that glioma stemness in our model can be maintained through Akt‑independent or context‑dependent pathways.

To determine the role of TGF-β/TGF-βR1 signaling in EMT within 3D spheroid cultures, cells were subjected to siRNA-mediated TGF-β knockdown (siTGFβ) or treated with galunisertib, a TGF-βR1 inhibitor [29, 30]. Morphological analysis showed that control spheroids (siScrambled) remained compact, whereas galunisertib treatment disrupted spheroid integrity and siTGFβ also reduced spheroid size (Figure 2A). Western blotting revealed elevated TGF-β and fibronectin expression in 3D spheroids compared to 2D monolayers, which were markedly reduced by siTGFβ and dose-dependent galunisertib treatment (Figure 2B). Both strategies targeting TGF-β downstream activity further led to a strong decrease in P-Smad2 levels, confirming effective inhibition of a potential TGF-β/TGF-βR1 signaling axis (Figure 2B). Consistent with these findings, RT-qPCR analysis demonstrated significant downregulation of TGF-β and fibronectin transcripts following siTGFβ and galunisertib exposure (Figure 2C, middle and right panels). Interestingly, spheroid-induced expression of HIF-1α was also altered (Figure 2C, left panel), as well as VEGF and GLUT1 in accordance with spheroid disruption (not shown). These results suggest that TGF-β signaling is critical for EMT induction and maintenance in 3D spheroids, and its inhibition attenuates EMT-associated markers and signaling activity.

Given that autocrine TGF-β-mediated signaling was already involved in 3D spheroids, we decided to characterize the dynamics of TGF-β signaling through phosphorylation of Smad2 and Smad3 over time, in 2D GBM cell monolayers. Western blot analysis revealed rapid phosphorylation of both Smad2 and Smad3 within 2–5 min (Figure 3A), reaching near-maximal levels by 30 min and remaining sustained for up to 60 min, while total Smad2 and Smad3 levels remained unchanged. Quantitative analysis confirmed a sharp increase in P-Smad2/Smad2 and P-Smad3/Smad3 ratios during these first 30 min, then followed by a plateau, indicating rapid and stable activation of canonical TGF-β signaling (Figure 3B). These results suggest that Smad activation, a key mediator of cellular migration, proliferation, and differentiation processes, is also required for 3D spheroid regulation during an autocrine TGF-β process.

In order to address the potential role for an autocrine TGF-β/TGF-βR1 signaling axis activated in 3D spheroids, we next questioned whether glycolytic inhibition affected TGF-β signaling and cell chemotaxis. Cells were treated with increasing concentrations of 2DG in the presence or absence of TGF-β. Intracellular ATP depletion was confirmed (Figure S1), and Western blot analysis showed that TGF-β strongly induced phosphorylation of Smad2 and Smad3, whereas 2DG dose-dependently reduced P-Smad2 and P-Smad3 levels without altering total Smad protein abundance (Figure 4A). Functional assays revealed that TGF-β significantly enhanced cell migration over 2 hours, and that this effect was dose-dependently suppressed by 2DG, with near-complete inhibition at concentrations ≥ 300 μM (Figure 4B). These findings indicate that glycolytic activity is required for optimal TGF-β signaling and migration, linking metabolic regulation to invasive behavior. Because 2DG can induce endoplasmic reticulum (ER) stress and impair N-linked glycosylation in addition to inhibiting glycolysis, we interpret ATP depletion as evidence of energetic stress rather than a specific blockade of a single metabolic pathway; ATP measurements in 2D monolayers were used to establish effective dosing and timing for subsequent 3D spheroid experiments, and follow-up assays to resolve the contributing mechanisms are planned.

To determine whether glycolytic inhibition affects TGF-β-induced EMT, cells were treated with increasing concentrations of 2DG in the presence or absence of TGF-β. Western blot analysis showed that TGF-β markedly increased fibronectin and Snail protein expression, confirming EMT activation, whereas 2DG dose-dependently suppressed both markers (Figure 5A, middle panels). In contrast, GRP78 (BiP), an ER stress marker, was strongly upregulated at higher 2DG concentrations, indicating metabolic stress (Figure 5A, upper panel). Quantitative analysis revealed near-complete inhibition of fibronectin and Snail protein levels at ≥ 300 μM 2DG (Figure 5B), and RT-qPCR confirmed a similar dose-dependent reduction in their mRNA transcripts (Figure 5C). These findings demonstrate that glycolytic activity is required for TGF-β-driven EMT, and its inhibition antagonizes EMT signaling while inducing stress responses.

We next explored whether a common TGF-β-mediated molecular signature linked cell death pathways between 2D TGF-β-treated cell monolayers and 3D spheroids. Total RNA was isolated from each condition, and gene arrays were performed as described in the Methods section. Overall, the waterfall plots showed differential cell death-related gene expression profiles after TGF-β treatment (Figure 6A) and 3D spheroid formation (Figure 6B) compared to the 2D monolayer control, with upregulated genes highlighted in green and downregulated genes in red. Accordingly, Venn diagrams further illustrate the overlap of apoptosis-related genes (Figure 6C, left) and autophagy-related genes (Figure 6C, right) between TGF-β treatment and 3D spheroid conditions. Among the upregulated apoptotic genes, BCL2, CASP7, FAS, FASLG, and GADD45A were triggered by both TGF-β and in 3D spheroids. Similarly, commonly induced autophagic genes included ATG7, ATG16L1, IRGM, ULK1, and PIK3C3. Protein–protein interaction network, generated using the STRING database (confidence score ≥ 0.7) for the commonly selected genes involved in macroautophagy (Figure 6D, red) and apoptosis (Figure 6D, green), showed functional associations and clustering that revealed strong connectivity between autophagy regulators and apoptosis mediators, suggesting pathway crosstalk.

The relative expression of apoptosis- and autophagy-related genes was next assessed in 2D cultures treated with TGF-β (Figure 7A, white bars), 2D cultures treated with TGF-β + 2DG (Figure 7A, black bars), and 3D spheroids treated with 2DG (Figure 7A, grey bars), and compared to the untreated 2D monolayer control. All were found to be sensitive to 2DG-mediated depletion of ATP, and their expression was inhibited. Particularly, the expression of the autophagic marker ATG7 and the key apoptosis-related genes BCL2 and FAS were found to be differentially regulated. Gene expression changes were similarly assessed in 3D spheroids from scrambled siRNA-transfected cells (Figure 7B, white bars), siTGFβ-transfected cells (Figure 7B, black bars), and upon galunisertib treatment (Figure 7B, grey bars). We found that altering the TGF-β/TGF-βR1 signaling axis significantly prevented these increases, supporting the interplay between TGF-β signaling, glycolytic inhibition, and TGF-βR1 blockade in regulating cell death and survival pathways in 3D spheroids.