The 1.4E7 human β cell line (ECACC) was cultured either in ML or as spheroids in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and 2 mM L-glutamine (all from Biosera), referred to as complete medium. Cultures were maintained in a humidified incubator at 37°C with 5% CO₂. For ML culture, 1.4E7 cells were seeded at 2 × 10⁴ cells/cm² in T-flasks with 0.2 mL/cm² of complete medium and routinely passaged upon reaching 80%–90% confluence. For spheroid culture, 1.4E7 ML cultures were detached using trypsin and seeded at 0.3 × 10⁶ cells/mL in 0.125 L or 0.5 L SpF (Corning) in 80% of working volume. The SpF were maintained under clockwise agitation on a magnetic stir plate (2mag bioMIX) set at 90 rpm (0.125 L SpF) or 75 rpm (0.5 L SpF), resulting in a constant P/V of ∼8.5 W/m³. 1.4E7 spheroids were cultured either in complete medium or in a chemically defined medium (RPMI-1640 Advanced, Capricorn Scientific # RPMI-ADV-500ML). This medium which contains insulin, transferrin and bovine serum albumin was supplemented with 4 mM L-glutamine, 100 IU/mL penicillin, and 100 µg/mL streptomycin (Eurobio), and referred to as production medium. Cultures were regularly monitored for mycoplasma contamination using the MycoAlert™ detection kit (Lonza). The maximal growth rate µ _max (h⁻¹) was determined as the slope of the linear regression of the natural logarithm transformation of the exponential growth phase. Population doubling time (PDT, h) was inferred as PDT = ln(2)/µ _max .

1.4E7 spheroids were cultured in the Ambr^®250 modular STBr using dual 30° three-segment pitched-blade impeller mammalian culture vessels (Sartorius Stedim). The cells were inoculated at 0.3 × 10⁶ cells/mL in 245 mL of production medium. The temperature was set to 37°C, with an agitation rate of 300 rpm (to achieve ∼8.5 W/m³), pH maintained at 7, and dissolved oxygen (DO) saturation set to 50%. pH was regulated by the addition of 1M NaOH and sparged CO₂, while DO was maintained through air sparging at a maximum rate of 2.45 mL/min.

Samples from SpF or STBr cultures were collected, and 30–200 µL was distributed in triplicate into a flat-bottomed 96-well plate for spheroid counting. Images were captured from each well using an Axiovert A1 FL LED Inverted Microscope coupled with an AxioCam MRc (Zeiss), and the largest diameter of each spheroid was measured using Fiji software. The interfacial area a (mm²/mL) of either the spheroids or the free cells (i.e., cells not incorporated into spheroids) in 3D cultures was calculated using Equation 1 where r is the mean radius of the free cells or spheroids (in mm), measured using Fiji software, and [X] is the concentration of spheroids or free cells (per mL). a = 4 × π × r 2 × X (1)

Mixing and hydrodynamics in SpF and STBr cultures were characterized by various parameters. The Reynolds number (Re, dimensionless) was calculated using Equation 2, while the laminar-turbulent transition (Re _T ) was estimated using Equation 3 (Grenville and Nienow, 2003). The volumetric power input P/V (W/m³) was calculated using Equation 4, the mixing time Θ _m (s) was estimated using Equation 5, the mean shear rate ϒ ˙ (s⁻¹) using Equation 6, the mean Kolmogorov length scale λ K (µm) using Equation 7 and the impeller tip speed υ _tip (m/s) using Equation 8. R e =   ρ × N × D 2   µ (2) R e T = 6370 × N P ‐ 1 / 3 (3) P / V = N p × ρ × N 3 × D 5   V (4) Θ m   = 5.9 × ε − 1 / 3 × D T − 1 / 3 × T   2 / 3   with  ε = P / ρ × V (5) γ ˙ = P µ × V   1 / 2 (6) λ K =   μ 3 ρ 3 × ε 1 / 4 (7) v t i p = π × N × D (8)

In these equations, N (s⁻¹) is the agitation rate, D (m) is the diameter of the impeller, ρ is the density of the culture medium (assumed to be that of water at 37°C = 993 kg/m³) and μ is the dynamic viscosity (assumed to be that of water at 37°C = 7 × 10⁻⁴ Pa·s). For N _P (dimensionless), we used the experimentally determined values reported by Rotondi et al.: 2.51 for a single flat paddle impeller, mimicking SpF, and 1.37 for the dual 30° three-segment pitched-blade impeller of the Ambr^®250 STBr (Rotondi et al., 2021). V (m³) is the volume of the medium; T (m) is the vessel diameter, and ε (W/kg) is the mean specific energy dissipation rate. A summary of the geometric characteristics of the SpF and the STBr, along with the calculated and fixed mixing and hydrodynamic values for this study, are presented in Table 1.

Cell viability was assessed using two distinct methods: trypan blue exclusion and estimation of dead and lysed cells through lactate dehydrogenase (LDH) activity in the cell culture supernatant. Specifically for spheroid cultures in SpF or STBr, samples were collected and centrifuged at 300 g for 5 min; then the pellets were dissociated with trypsin to allow for viable and dead cell count via trypan blue exclusion. In parallel, LDH activity in cell culture supernatants was measured as an indicator of cell damage and death (Legrand et al., 1992) with the Cytotoxicity Detection Kit (Roche) with rabbit muscle L-LDH as a standard (#10127230001, Roche). Absorbance was measured using the FLUOstar OPTIMA plate reader (BMG Labtech). In parallel, 1.4E7 cells grown as ML or spheroids in complete medium were lysed with 2% Triton X-100 to assess intracellular LDH levels, this being used to determine the number of dead and lysed cells in the culture supernatant. Lastly, cell viability was calculated using Equation 9 where [X] _alive is the viable cell concentration determined using trypan blue counting and [X] _dead is the estimated lysed cell concentration in cell culture supernatant. V i a b i l i t y   % = X a l i v e X a l i v e + X d e a d × 100 (9)

Cell culture medium was collected, centrifuged at 300 g for 10 min, and stored at −80°C until metabolites were analyzed. The concentrations of glucose, glutamine, lactate and ammonium were quantified using the Gallery™ analyzer (ThermoFisher Scientific). Substrate (S) specific consumption rates (q _S ) for glucose and glutamine, and product (Pr) specific production rates (q _P ) for lactate and ammonium, between i-1 and i samplings (spaced by Δt interval), were determined respectively by Equations 10, 11 q S = − 1 X i × S i −   S i ‐ 1 Δ t (10) q P = 1 X i × ⁡ Pr ⁡ i −   ⁡ Pr ⁡ i ‐ 1 Δ t (11)

Lactate-to-glucose and ammonium-to-glutamine metabolic yields (Y _Lac/Glc and Y _NH4/Gln , respectively), as well as cell-to-glucose and cell-to-glutamine growth yields (Y _cells/Glc and Y _cells/Gln , respectively), were determined as the slope of the linear regressions of respectively f([S]_i0-[S]_i) = [Pr]_i-[Pr]_i0; and f([X]_i-[X]_i0) = [S]_i0-[S]_i, between 0 and 72 h.

All relevant data from our experiments were submitted to the EV-TRACK knowledgebase (EV-TRACK ID: EV240160) (Van Deun et al., 2017). Cells were expanded through 12 to 18 passages to produce sEV, reaching a cumulative population doubling level (PDL, Equation 12) of 18.8–33.4. X _harvested corresponds to the total number of viable cells harvested at each passage, and X _seeded corresponds to the total number of seeded cells. P D L = ln X h a r v e s t e d   ‐   ln X s e e d e d   ln 2 (12)

For ML cultures, after 72 h of culture (∼80% confluence), cells were washed twice with PBS, and the complete medium was replaced with 0.12 mL/cm² of production medium for 24 h. For 3D cultures in SpF and STBr, both spheroid formation and sEV production were performed in the production medium. After 48 h of culture, 1.4E7 spheroids were allowed to settle, and the culture medium was replaced with a fresh production medium (∼85% renewal).

For each sEV batch, around 400 mL of conditioned medium was used (Supplementary Table 1), obtained from 10 to 19 flasks for ML, two vessels for STBr, or single 0.5L SpF.

After a 24 h sEV production period, 1.4E7-derived sEV were isolated using a method combining differential centrifugation, tangential flow filtration (TFF), ultrafiltration (UF), and size exclusion chromatography (SEC). Briefly, the production medium was collected and sequentially centrifuged to remove cell debris and large extracellular vesicles: 300 g for 10 min, 2,000 g for 20 min, and 16,000 g for 20 min, followed by filtration through a 0.22 µm PES filter. sEV were isolated from the conditioned medium by five cycles of diafiltration in PBS using a 300 kDa MidiKros column (Repligen) on a KrosFlo Research IIi TFF instrument (Spectrum Europe). The input flow rate was set to 50 mL/min, with transmembrane pressure regulated below 300 mbar. Twenty mL of retentat were recovered in PBS and further concentrated down to 150 µL using a 100 kDa AMICON (Merck Millipore) filter by centrifugation at 2,000 g for 20 min. sEV were then passed through a qEV Gen2 35 nm single SEC column (Izon). sEV were collected in flow-through fractions two, three, and four in PBS filtered 0.1 µm. sEV were stored at −80°C in PBS containing 25 mM trehalose (Sigma-Aldrich) (Bosch et al., 2016).

sEV size and concentration were determined by nanoparticle tracking analysis of the tetraspanin-labeled particles using the ZetaView^® PMX 220 TWIN (Particle Metrix) and the associated software (ZetaView 8.05.16). Briefly, sEV were incubated with AF-488-conjugated anti-human CD81 (454720, Bio-Techne), anti-human CD9 (209306, Bio-Techne), and anti-human CD63 (H5C6, BioLegend) antibodies at a 1:200 dilution overnight at 4°C. After incubation, samples were diluted in 0.1 µm-filtered PBS to achieve a minimum 1:50,000 final dilution of the antibodies. Before analysis, the ZetaView^® was calibrated using 100 nm polystyrene standard particles. For acquisition, samples were excited with a 488 nm laser coupled with a 500 nm filter. Videos were recorded at 25°C across 11 positions, with a sensitivity of 95, a shutter speed of 100, and a frame rate of 30 frames per second.

For total protein analysis, cells were lysed in RIPA lysis buffer (ThermoFisher Scientific) containing 1% proteases/phosphatases inhibitors (Ozyme). sEV were lysed in 0.1% Triton X-100 (Sigma-Aldrich). Protein concentration was determined using Coomassie plus assay reagent (ThermoFisher Scientific) following the manufacturer’s recommendations. Protein absorbance was read on the FLUOstar OPTIMA plate reader.

For morphological analysis, sEV were prepared for cryo-electron microscopy as described previously (Giri et al., 2020) and imaged on a Tecnai G2T20 Sphera electron microscope (FEI Company) equipped with a CMOS camera (XF416, TVIPS) at 200 kV. Micrographs were acquired under low electron doses, using the camera in binning mode 1, at a nominal magnification of ×50,000.

50 µg of cellular protein lysate or 5 × 10⁹ sEV were denatured in Bolt™ sample buffer (Thermo Fisher Scientific) and separated on a 4%–12% Bis-Tris Plus gradient SDS-PAGE gel under non-reducing conditions, followed by transfer to a nitrocellulose membrane (Thermo Fisher Scientific). Membranes were blocked with TBS containing 0.05% Tween-20% and 5% milk, then incubated with the following primary antibodies from BioLegend: mouse anti-CD63 (H5C6, 1:1,000), mouse anti-CD81 (5A6, 1:500), mouse anti-CD9 (HI9a, 1:1,000), mouse anti-LAMP1 (H4A3, 1:500), HRP-rat anti-Flotillin-1 (W16108A, 1:2,500), HRP-rat anti-β-actin (W16197A, 1:10,000), mouse anti-Hsp70 (W27, 1:1,000), or rat anti-Calnexin (W17077C, 1:1,000). Membranes were then incubated with the appropriate horseradish peroxidase-conjugated secondary antibodies: polyclonal rabbit anti-mouse (1:2,000, Agilent Technologies) or polyclonal goat anti-rat (1:20,000, Thermo Fisher Scientific), except for β-actin and Flotillin-1 antibodies. Signals were detected using enhanced chemiluminescence substrate Pico or Femto (Thermo Fisher Scientific), depending on the target, and visualized on a Fusion FX6 instrument (Thermo Fisher Scientific).

sEV were analyzed using the ExoView™ Human Tetraspanin Kit (Unchained Labs) according to the manufacturer’s instructions. Briefly, sEV were diluted to a final concentration of 2 × 10⁸ sEV/mL in the provided incubation solution. A volume of 50 µL of the sEV solution (1 × 10⁷ sEV) was added to silicon dioxide chips coated with antibodies against human CD9 (HI9a), CD63 (H5C6), and CD81 (JS-81). The chips were incubated at room temperature for 16 h and then washed with the supplied washing solution. Subsequently, the chips were incubated for 1 h at room temperature with fluorochrome-conjugated antibodies against CD9 (1:500, HI9a, CF^®488A), CD63 (1:500, H5C6, CF^®647), and CD81 (1:500, JS-81, CF^®-555), prepared in the supplier’s blocking solution. After incubation, the chips were washed, dried, and imaged using the ExoView™ R100 platform and the ExoView Scanner 3.2.2 acquisition system. Data were analyzed using ExoView Analyzer 3.2 software. The data from each tetraspanin capture spot (CD9, CD63, and CD81) were calculated by subtracting the IgG control spot values.

For immunofluorescence staining, 3 × 10⁴ 1.4E7 cells/cm² were seeded onto 8-chamber Nunc LabTek slides. The following day, cells were washed in PBS and fixed with PBS 4% paraformaldehyde for 15 min at room temperature. Cells were rehydrated in PBS and permeabilized in PBS containing 0.2% Tween-20% and 5% BSA for 30 min at room temperature. Cells were washed in PBS and incubated in PBS 5% BSA blocking buffer for 30 min at room temperature, followed by overnight incubation at 4°C with primary antibodies or isotypic controls diluted in blocking buffer [anti-human CD63, H5C6, 1:50 (BioLegend); anti-human CD81, 454720, 1:25 (Bio-Techne); anti-human CD9, 209306, 1:25 (Bio-Techne)]. The next day, cells were washed and incubated with secondary AF488-anti-mouse IgG H+L antibody (Thermo Fisher Scientific, #A32723) for 1 h at room temperature. Nuclei were counterstained with Hoechst 33342 (Sigma-Aldrich) at 1 μg/mL for 10 min at 37°C before imaging on an LSM780 confocal microscope (Zeiss) with a Plan-Apochromat objective lens 63x/1,4 Oil DIC M27, and analysis with Fiji software (Schindelin et al., 2012).

Statistical tests were performed using GraphPad Prism 8. Data are presented as means ± SD. Differences between two groups only were analyzed using the Mann-Whitney test. The Kruskal–Wallis test, followed by Dunn’s multiple comparisons test, was used for comparing data sets of three or more groups. Correlation analyzes were performed using Spearman’s rank correlation test. Significant differences are indicated in the figures as: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

The human β cell line 1.4E7 was cultured either in T-flasks as 2D static ML with a complete medium containing 10% FBS, or in suspension spheroids in 0.125 L SpF at an agitation rate of 90 rpm and a seeding density of 0.3 × 10⁶ cells/mL. The 3D cultures were maintained in either a complete medium; or in a chemically defined medium, without FBS (production medium), to implement a fully serum-free 3D culture process and prevent contamination from serum-derived EV.

Regardless of FBS supplementation, 3D spheroid cultures maintained high cell viability (>93%), as assessed using trypan blue exclusion, throughout the process, with no significant differences between the ML and 3D cultures (Figure 1A). However, 3D spheroid cultures significantly reduced 1.4E7 cell growth compared to 2D static ML culture (Figure 1B). The average growth rate (µ _max ) in ML culture was 0.028 h⁻¹, compared to 0.01 h⁻¹ in 3D cultures with 10% FBS, and 0.006 h⁻¹ in 3D cultures without FBS, corresponding to PDT of 25.1 h ± 4.6, 100.2 h ± 76.2, and 145.1 h ± 178, respectively. Although the µ _max of cells grown as spheroids with or without FBS did not significantly differ, fewer cells were obtained after 72 h of culture in the absence of FBS (Figure 1C).

LDH activity was assessed in cell culture supernatants as an indicator of cell death. 1.4E7 cells, grown as ML or spheroids, were first lysed in 2% Triton X-100 to evaluate their LDH content. Intracellular LDH levels were notably higher in the lysates of 1.4E7 spheroids compared to ML (Figure 1D). This allowed the estimation of the number of dead and lysed cells based on LDH activity in the supernatant depending on culture conditions (Figure 1E). Since LDH activity in ML culture supernatants was minimal or undetectable, higher cell viability was inferred using this method, compared to 3D cultures (Figure 1F, ML vs. 3D with FBS: p < 0.05; ML vs. 3D without FBS: p < 0.0001). At 72 h, 3D cultures exhibited a cell viability of 92.7% ± 1.1 in the presence of FBS compared to 83.6% ± 5.8 in its absence, which was not significantly different. Thus, LDH measurement is a sensitive method for 3D cultures, allowing the detection of lysed cells that are not identified using trypan blue exclusion.

Monitoring the metabolites in the culture medium confirmed the absence of substrate limitation (glucose and glutamine, Figure 2A, C) and excess metabolic waste (lactate <20 mM and ammonium ions <5 mM, Figure 2B, D) (Hassell et al., 1991; Cruz et al., 2000). The glucose q _S (Figure 2E) and lactate q _P (Figure 2F) decreased over the culture duration in all conditions. However, both parameters were higher in ML cultures compared to 3D cultures, regardless of FBS supplementation, which is consistent with higher cell proliferation in ML cultures. Interestingly, the glutamine q _S (Figure 2G) and ammonium q _P (Figure 2H) gradually decreased with time in ML cultures, in line with exponential growth, which was not the case in 3D cultures. A higher lactate-to-glucose yield (Y _Lac/Glc ) was observed in ML cultures (1.4 mol/mol ±0.4) compared to 3D cultures (0.9 mol/mol ±0.1 in the presence of FBS and 1 mol/mol ±0.1 in the absence of FBS) (Figure 2I). This relatively high ratio in ML cultures suggests that cells are primarily using glycolysis, a phenomenon commonly seen under hypoxia, but also in proliferating cells regardless of oxygen concentration (known as Warburg effect) (Vander Heiden et al., 2009). In contrast, the ratio of 1 mol/mol in 3D cultures indicates a lower lactate production relative to glucose consumption, suggesting that cell metabolism shifted to oxidative phosphorylation which allows for more efficient ATP production from glucose (∼36 ATP instead of 2). Concerning the ammonium-to-glutamine metabolic yield (Y _NH4/Gln ) (Figure 2J), a similar ratio of <1 mol/mol was observed in all conditions, suggesting a balanced metabolic state where ammonium production is proportional to glutamine availability, indicating efficient glutamine utilization. However, the growth yield, expressed as cells generated per mole of glucose (Y _cells/Glc , Figure 2K) or glutamine (Y _cells/Gln , Figure 2L) consumed, was higher in ML cultures, indicating a more efficient utilization of these metabolites for cell growth compared to 3D cultures.

In 3D cultures, 1.4E7 cells rapidly self-agglomerated until forming well-defined and spherical spheroids regardless of FBS concentration (Figure 3A). The concentration of spheroids decreased with time (Figure 3B), while their size increased (Figure 3C) suggesting that spheroids grew larger as they fused together. While after 48 h, we observed similar spheroids’ concentration and size with or without FBS, at 72 h larger but fewer spheroids were obtained in the absence of FBS. Indeed, we counted 333 spheroids ±116 per mL, with a size of 248 µm ± 20 in the absence of FBS against 562 spheroids ±146 per mL (p < 0.01), with a size of 198 µm ± 10 (p < 0.001) in the presence of FBS. Six to 10% of the cells remained as alive or dead free cells (i.e. not incorporated within spheroids), and their viability decreased over time in culture (Figure 3D), suggesting that these cells likely underwent anoïkis due to the lack of adhesion. Interestingly, free cell viability appeared lower in the absence of FBS, although the differences were not statistically significant at any time point. Lastly, similar numbers of cells per spheroid were obtained regardless of FBS supplementation (Figure 3E). In the absence of FBS, spheroids contained 590 cells ±232 at 48 h, increasing to 1244 cells ±883 at 72 h.

In conclusion, while 2D ML cultures achieved higher growth yields, 3D spheroid cultures allowed to culture 1.4E7 cells in suspension while maintaining high cell viability. Although the proliferative capacity of cells was lower in 3D cultures, the cells metabolism shift towards oxidative pathway respiration improving glucose utilization for ATP production. Additionally, 3D cultures of 1.4E7 cells in a serum-free and chemically defined medium preserved equivalent cell viability, metabolic profiles, and spheroid morphology while avoiding contamination of the medium by exogenous EV.

For further investigation of the upstream culture mode on 1.4E7 spheroids and sEV derived thereof, we used an sEV production process comprising an expansion phase in 2D ML cultures containing FBS, followed by a transition to a 3D culture system using a serum-free, chemically defined medium, where spheroids were formed for 48 h. Subsequently, the medium was renewed for an additional 24 h culture phase to produce sEV.

The impact of hydrodynamics on spheroid formation was assessed by measuring spheroid size and concentration after 72 h of culture at different agitation rates (60, 90, and 120 rpm) in 0.125 L SpF with a constant volume of production medium (100 mL) and, therefore, a constant liquid height (H) and liquid height-to-vessel diameter ratio (H/T). This resulted in P/V values of 2.6, 8.6, and 20.4 W/m³, respectively. Visually, 1.4E7 cells cultured at 60 rpm tended to form larger and more heterogeneous aggregates, while cultures at 120 rpm produced smaller aggregates with less-defined morphology. In contrast, cultures at 90 rpm consistently generated homogenous and well-defined spheroids (Figure 4A).

These agitation rates in 0.125 L SpF correspond to the operational window that enables spheroid formation. Below 60 rpm and above 120 rpm, large (several mm) heterogeneous clusters, consisting mainly of dead cells, along with a significant number of free cells, were obtained (data not shown).

Thus, within this operational window, we observed that P/V was linearly and negatively correlated to spheroid size (Spearman, R² = 0.85, p < 0.0001) and positively to spheroid concentration (R² = 0.5, p < 0.001) (Figure 4B). As spheroid formation was dependent on the system’s P/V, we applied a constant P/V-based strategy to standardize spheroid formation during process scale-up. As expected, in a larger SpF (0.5 L), applying a constant P/V of ∼8.6 W/m³ generated spheroids of similar size to those obtained in the reference 0.125 L SpF at 90 rpm (Figure 4C). Conversely, maintaining a constant agitation rate of 90 rpm in a 0.5 L SpF increased the P/V to 15 W/m³, resulting in smaller spheroids (p < 0.05). An alternative scale-up strategy, which uses a constant υ _tip, required setting the agitation rate to 61 rpm. This condition, with a lower P/V value of 4.7 W/m³, generated larger spheroids (p < 0.05). These data demonstrate that spheroid formation strongly depends on the system’s P/V, and maintaining this parameter constant during process scale-up in geometrically similar vessels is crucial to standardizing spheroid size and morphology.

The SpF process was subsequently transferred to the fully controlled Ambr^®250 STBr by applying a constant P/V of approximately 8.5 W/m³, despite significant geometric and impeller design differences between the SpF and the STBr. Additionally, in contrast to SpF, volumetric aeration was introduced through the sparger in STBr, allowing for oxygen level control. However, its impact on the overall P/V under our conditions was minimal and negligible. 1.4E7-derived sEV were produced in both stirred systems, following a process in two steps: (i) a spheroid formation phase of 48 h, followed by (ii) an ∼85% renewal of the production medium for an additional 24 h, before collection of the supernatants to isolate sEV.

Representative images of the 1.4E7 spheroids obtained in the 0.5 L SpF and the STBr at the end of the sEV production phase are shown in Figure 5A. Throughout the process, significantly greater numbers of spheroids were obtained in the STBr at any time point (Figure 5B), with these spheroids being smaller in size (Figure 5C). At the end of the process, in STBr, 3572 spheroids ±943, of 92 µm ± 7 were obtained per mL, with each spheroid containing 84 cells ±17 (Figure 5D), compared to 327 spheroids ±90 (p < 0.01) of 230 µm ± 23 (p < 0.01) per mL, each containing 1137 cells ±361 (p < 0.01) in SpF. As illustrated in Figure 5E, a tangential flow regime was achieved in the SpF equipped with a flat paddle impeller, while the dual 30° three-segment pitched-blade impellers of the STBr generated an axial flow regime. These striking differences in the vessels and, notably, the impellers geometry, affected the flow regime, thereby impacting spheroid formation.

The hydrodynamic characteristics of both systems are summarized in Table 1. No major difference was observed concerning the Re (6173 in 0.5 L SpF vs. 4795 in STBr) and similar Θm were determined (7.4 s in the SpF vs. 6 s in the STBr). As P/V remained constant, both systems achieved comparable mean shear stress. The calculated shear rate γ ˙ was approximately 110 s⁻¹ while the λ K was equal to ∼80 µm in both systems. Since all spheroids in the SpF were larger than the λ K , they likely experienced shear stress (Cherry, 1993), whereas in the STBr, ∼25% of spheroids remained unaffected by shear stress (Figure 5F). However, these global measures neglect the heterogeneity of energy dissipation. Specifically, the energy dissipation rate and associated shear stress, are higher and maximal near the impellers. As the υ _tip was nearly 2-fold higher in the STBr (0.41 m/s vs. 0.23 m/s), and since there were two impellers, shear stress was likely locally higher in STBr compared to SpF.

Regarding cell performance and viability, slight cell proliferation was observed in the 0.5 L SpF despite the absence of FBS, whereas no proliferation was noted in the STBr (Figure 5G, H). We observed similar and high cell viability (>95%) using trypan blue exclusion in both stirred systems throughout the process (Figure 5G). A slightly higher LDH release was detected in the STBr, resulting in a calculated cell viability of 93.6% ± 1.5 in the SpF and 90.1% ± 3.1 in the STBr at 72 h (p < 0.05) (Figure 5I). A slightly higher percentage of free cells was found in the STBr at 48 h: 19.8% ± 4.6 in the STBr vs. 13.8% ± 3.5 in the SpF (p < 0.05) and this difference remained after medium renewal (13.4% ± 3.9 in the STBr vs. 5.0% ± 1.4 in the SpF, p < 0.05). However, at the end of the sEV production phase, the percentage of free cells was not significantly different anymore (Figure 5J). Lastly, we estimated that the more numerous, smaller spheroids in the STBr increased the cell exchange surface, referred as interfacial area. The spheroids generated in the STBr represented 92 mm²/mL ± 16 compared to 54 mm²/mL ± 12 in the SpF at 72 h, while no significant difference was observed in the interfacial area of the free cells (Figure 5K). Regarding cell metabolism, no major difference was observed between both systems (Supplementary Figure 1).

1.4E7-derived sEV were isolated from the conditioned medium of cells grown as ML (T-flask) or spheroids in SpF and STBr for 24 h in the absence of FBS, as described. Data regarding sEV production volumes, cell concentrations and sEV yields per condition are summarized in the Supplementary Table 1. Cryo-transmission electron microscopy revealed round-shaped membrane particles in all conditions (Figure 6A, see Supplementary Figure 2 for uncropped images). NTA analysis of tetraspanin-labeled particles also showed a homogenous sEV size distribution across conditions (Figure 6B). The downstream process, combining TFF and SEC, allowed for the recovery of highly pure sEV samples of up to 2 × 10⁹ sEV/µg proteins (Figure 6C). The highest sEV yield was obtained from STBr cultures, which was 2-fold higher than that from ML cultures, while the lowest yield was observed in SpF cultures (Figure 6D). We hypothesize that spheroid size and concentration may significantly contribute to these differences in sEV yields. Interestingly, when normalized to the interfacial area, the lowest sEV yield was observed in ML cultures, while production in stirred systems increased the yield (2-fold in SpF and 5-fold in STBr) (Figure 6E). To confirm the sEV identity of the isolated particles, and in accordance with the MISEV2023 recommendations (Welsh et al., 2024), western blots were performed to detect sEV-enriched proteins and to ensure the absence of proteins associated with irrelevant intracellular compartments (e.g., calnexin). As shown in Figure 6F, the sEV isolated from each culture mode displayed typical expression of membrane-bound proteins (Lamp1, CD63, CD81, and CD9) and cytosolic proteins (Hsp70 and flotillin-1). Notably, the expression of flotillin-1 was higher in stirred systems, particularly in SpF, as observed in four independent western blots. Finally, sEV from each condition did not express β-actin (cytoskeleton) or calnexin (endoplasmic reticulum), confirming the purity of the isolated sEV.

In addition, an ExoView™ array was performed to quantify the amount of 1.4E7-derived sEV expressing the CD9, CD63, and CD81 tetraspanins from each sEV production method. Representative images of sEV captured on CD63 chips are shown in Figure 7A. CD81 was the most highly expressed tetraspanin in each sEV sample (Supplementary Figure 3A). Interestingly, the percentage of sEV captured on CD63 chips was significantly higher in ML cultures than in the STBr (Figure 7B). In addition, the percentage of CD63⁺ sEV captured on CD9 chips was higher in ML cultures compared to both stirred systems (Figure 7C). A similar profile was observed with CD81 chips, although the differences were not significant (Figure 7D). In contrast to CD63, culture mode did not affect the expression of CD9 and CD81 (Supplementary Figures 3A–F).

Immunostaining of CD63, CD9, and CD81 was performed on 1.4E7 cells grown as ML to determine their cellular localization. CD63 was predominantly expressed at intracellular locations (Figure 7E), whereas CD9 (Figure 7F) and CD81 (Figure 7G) were almost exclusively found on the plasma membrane. These observations suggest that sEV expressing CD63 primarily originate from subcellular compartments as exosomes, while sEV lacking CD63 but expressing CD9 or CD81 derive from the plasma membrane as ectosomes.

Altogether, these data indicate that highly pure and intact sEV expressing typical EV markers were obtained from each culture system. Interestingly, a lower proportion of CD63⁺ sEV was isolated from spheroid cultures in stirred systems, suggesting an enrichment of plasma membrane-derived sEV.