High‐Yield Production of Extracellular Vesicle Subpopulations with Constant Quality Using Batch‐Refeed Cultures

Abstract

The conventional manufacturing of extracellular vesicles (EVs) is characterized by low yields and batch‐to‐batch variability, hampering fundamental research on EVs and their practical applications. Perfusion operations have huge potential to address these limitations and increase the productivity and quality of EVs. In this study, perfusion cultures are simulated with batch‐refeed systems and their productivity is compared with that achieved using batch cultures. It is shown that a shift from batch to batch‐refeed system can increase the space‐time yields of a target EV subpopulation characterized by CD81 and CD63 biomarkers by threefold. Moreover, it is demonstrated that the method facilitates the consistent production of the target EVs from cells maintained under constant conditions for 13 days. These results indicate that the use of perfusion cultures is a promising strategy to increase the manufacturing yield of EVs and control the production of specific EV subpopulations with constant quality attributes, thereby improving reproducibility.

Low yields, batch‐to‐batch variability, and scarce control on sample composition are major issues in the production of extracellular vesicles (EVs). Here the potential of batch‐refeed and perfusion cultures to produce high amounts of EVs with constant quality attributes and enriched in specific subpopulations are demonstrated.

1. Introduction

Extracellular vesicles (EVs) have emerged as promising therapeutic agents and drug delivery systems over the last few years owing to favorable properties, including immunogenicity, half‐life, specific targeting, and ability to cross the blood–brain barrier.^{[ 1 ]} However, clinical studies involving EVs are still scarce, also because of the challenge in the large‐scale manufacturing of consistent samples required in these studies.^{[ 2} , ³ , ⁴ , ⁵ , ^{6 ]}

As of date, only a few strategies have been developed to increase upstream yields.^{[ 2} , ⁷ , ^{8 ]} In some cases, EV release is stimulated by applying chemical stress, such as increased calcium ion concentrations,^{[ 9} , ^{10 ]} serum deprivation,^{[ 11} , ^{12 ]} or hypoxia.^{[ 10} , ^{13 ]} For instance, a study reported that hypoxia enhanced the secretion of EVs from breast cancer cells by 1.8 times.^{[ 13 ]} In other approaches, mechanical stress, such as acoustic irradiation, has been exploited and reported to increase the yield by up to tenfold.^{[ 4 ]} However, care should be taken when applying stress conditions because they can modify the EV subpopulations produced, eventually altering the activity of the final product.^{[ 14 ]}

In some contexts, the choice of the EV source can drastically simplify the scale‐up of EV production, as in the case of EVs derived from milk^{[ 15 ]} or algae,^{[ 16} , ¹⁷ , ^{18 ]} where large quantities of EV‐containing medium can be obtained economically and sustainably.

In addition to the EV source and environmental factors, an appropriate bioreactor design can significantly optimize EV yield. For instance, cultures of human cell lines have been scaled up from 2D to 3D formats using stirred tank bioreactors.^{[ 19} , ^{20 ]} This approach increased yields by up to 20‐fold.^{[ 19 ]}

In the past decade, perfusion bioreactors have attracted increasing attention as alternatives to batch operation for commercial manufacturing of biologics, such as monoclonal antibodies.^{[ 21} , ²² , ²³ , ²⁴ , ²⁵ , ^{26 ]} The continuous supply of nutrients in these bioreactors allows higher cell densities, resulting in higher productivities than those achieved using batch cultures, without affecting cell viability. The constant cell culture conditions during the entire process are also reflected in consistent product quality attributes. Moreover, risks of product degradation are minimized by constant harvesting, which reduces the residence time in the bioreactor.^{[ 23 ]}

All these advantages are very attractive in the context of EV manufacturing. However, currently, only a few studies based on hollow fiber bioreactors have employed perfusion conditions,^{[ 27} , ²⁸ , ²⁹ , ^{30 ]} and characterization studies aimed at understanding the effect of perfusion conditions on the production of EVs are lacking. In most cases, EVs are quantified only using techniques, such as nanoparticle tracking analysis (NTA) and total protein assays, which cannot distinguish between different EV populations. Therefore, how these cultures affect the production of small EVs (sEVs), a subpopulation of vesicles with various therapeutic applications, remains unclear. A better understanding of the productivity of sEVs and their specific subpopulations in different cell cultures is urgently required to optimize their production.^{[ 31 ]}

In this work, we studied the production of EVs, characterized by tetraspanins, namely CD63 and CD81, which are well‐known surface biomarkers of sEVs^{[ 32} , ^{33 ]} and compared their yields in batch and perfusion cultures. Following an approach previously developed for monoclonal antibodies,^{[ 34 ]} we modeled perfusion at a small scale with a batch‐refeed culture and then used bead‐based flow cytometry to track the number of different EV subpopulations (namely CD81+, CD81+/CD63+, and CD81+/CD9+ EVs) directly in the medium. We performed the study using HEK‐293F cells, an established source of EVs, considering their scalability prospects, ease of transformation, well‐known properties, and safety.^{[ 35} , ³⁶ , ³⁷ , ³⁸ , ^{39 ]} We cultured cells in suspension in a chemically defined serum‐free medium. We demonstrated that the production of EVs is constant in batch‐refeed cultures, whereas it changes over time in batch cultures. Moreover, space‐time yields of the batch‐refeed systems are threefold higher than those of batch cultures. Overall, we demonstrate the potential of perfusion and batch‐refeed cultures in producing large amounts of specific EVs in a controlled and reproducible manner. This is a key step toward the identification of functional EV subpopulations for various therapeutic applications.

2. Results

2.1. Cultivation of 293F Cells in Batch‐Refeed Mode

We modeled the perfusion process with a batch‐refeed culture of HEK‐293F cells following a method previously developed using CHO cells.^{[ 34 ]} The strategy employed by us is illustrated in the schematic in Figure  1 . In a perfusion culture, the medium is continuously harvested and replaced with a fresh feed, and cells are simultaneously removed in the bleed. These operations occur at specific rates to maintain a constant cell density and nutrient concentration in the solution. Batch‐refeed cultures follow a similar principle; however, cell bleeding and medium exchange are not performed continuously but at regular intervals of time. In the present study, 20 mL cultures were maintained for up to 21 days in vented shake flasks and the medium was exchanged daily. To monitor the product concentrations with minimal purification steps, we characterized the harvests using bead‐based flow cytometry after clarification of the medium by centrifugation and analysis of particle sizes using nanoparticle tracking analysis and microfluidic diffusion sizing.^{[ 40 ]}

Figure 1.

Schematic of the experimental setup. For the production of EVs, perfusion cultures were simulated with batch‐refeed cultures. A perfusion culture involves a bioreactor with a cell retention device. Under normal operations, the medium is constantly harvested, fresh medium is fed, and cells are bled to keep the cell culture conditions in the bioreactors stable. In contrast to batch cultures, high cell densities can be obtained in perfusion bioreactors, which can be maintained for several days. In a batch‐refeed culture, cell bleed and medium exchange are not performed continuously but at regular intervals of time. In this study, cells were partially removed from the culture every 24 h, pelleted, resuspended in a fresh medium, and further incubated in shake flasks. This daily medium replacement allowed maintenance of a high density of viable cells over several days in batch‐refeed operations. For the characterization of EVs, the samples were analyzed using nanoparticle tracking analysis, microfluidic diffusion sizing,^{[ 40 ]} and bead‐based flow cytometry. The quantification of EVs in the harvested medium was performed using bead‐based flow cytometry. The surface marker CD81 was used to capture EVs on the beads, and CD63 was used to stain the EVs. With this approach, we measured the production of CD81+/CD63+ EVs in the culture. Owing to the stable culture conditions and constant harvest of the product, the EV concentration in perfusion bioreactors remains constant over time. In contrast, in batch systems, EVs reach a maximum concentration in the culture and are eventually taken up by cells.

The first step in establishing batch‐refeed cultures involved the definition of the maximum cell densities achievable in our system. We fixed the perfusion rate at 1 RV day⁻¹, where RV is the reactor volume. This value was a convenient starting point because lower rates are known to limit cell densities, whereas, according to the literature, greater values increase densities only for some HEK293 cell strains.^{[ 26} , ^{41 ]} With this setup, we observed an increase in cell densities up to 2 × 10⁷ cells mL⁻¹ in 14 days (Figure S1, Supporting Information). The available glucose progressively decreased over time and was almost completely consumed after 9 days, with a concomitant reduction in the production of lactate. This indicated a shift in the metabolism due to a lack of nutrients, which was confirmed by the decrease in the viability of cells.^{[ 34 ]} The decrease in viability is not desirable during EV production owing to the possibility of contamination of EV samples with cell debris and apoptotic bodies.

Next, we related the growth rate of cells to cell‐specific perfusion rate (CSPR), the rate at which the medium is exchanged for a given cell, to select suitable conditions for steady‐state experiments (Figure S2, Supporting Information). We observed a drop in the cell‐specific growth rate at CSPR lower than 100 pL cell⁻¹ day⁻¹. Because perfusion requires operation at a CSPR that keeps the cells stably growing, we selected a CSPR of 120 pL cell⁻¹ day⁻¹ for steady‐state experiments. This corresponded to a viable cell density (VCD) of 8.5 × 10⁶ cells mL⁻¹ (high VCD reactor in Figure  2 ). In parallel, to evaluate the effect of the metabolic state of cells and the nutrient amount on the production of EVs, we also selected a higher CSPR (480 pL cell⁻¹ day⁻¹) for a second steady state experiment, where the viable cell density was maintained at 2.1 × 10⁶ cells mL⁻¹ (low VCD reactor in Figure 2). Cells were then cultivated under the two selected steady‐state conditions for 13 days. In this case, the cells were bled daily to maintain the batch‐refeed culture at constant cell densities.

Figure 2.

Comparison of batch‐refeed and batch cultures of HEK293 F cells. Viable cell density and viability a–c), glucose d–f), and lactate g–i) concentrations in the cultures. In the batch‐refeed cultures, cell bleeds were performed daily to maintain the cell densities constant for 13 days. The experiment was performed in triplicates, and data are represented as means ± SD. Dotted arrows mark the days at which cell bleeds were started to maintain the cultures under steady‐state conditions.

As expected, the daily exchange of medium enabled us to achieve significantly higher cell densities than those achieved in the batch culture (Figure 2a–c). Moreover, in both the low and high‐cell density cultures, the cell viability remained above 94% during the entire culture period, whereas it decreased below 94% after only 5 days in the batch culture.

The concentrations of glucose and lactate in batch‐refeed cultures changed similarly every day after starting the cell bleed (Figure 2d,e,g,h). The low‐density culture had excess glucose, whereas glucose was mostly consumed in the high‐density one (Figure 2d,e). In contrast to the steady‐state values in the batch‐refeed, the glucose concentration progressively decreased over time in the batch culture (Figure 2f). The lactate concentrations were similar in the two batch‐refeed cultures (Figure 2g,h). A higher concentration of lactate, which is a by‐product of metabolism, was expected in the high‐density culture, given the higher number of producing cells. As previously reported,^{[ 34} , ^{42 ]} the lower amount of lactate might indicate that the cells consumed a part of the produced lactate due to the lack of glucose, and were, thus, in a different metabolic state than the cells in the low‐density cultures. In the batch culture, the lactate concentration changed drastically over time in a non‐monotonous manner, initially increasing and then decreasing after the depletion of glucose in the medium (Figure 2i).

2.2. Quantification of EVs in Conditioned Medium

We next quantified the EVs produced during batch and batch‐refeed cultures by analyzing the medium harvested every day from each of the cultures. We minimized the processing steps between collection and analysis to avoid biases in the measurement due to the loss or damage of EVs. To this end, after two clarification steps using centrifugation, we captured EVs employing immuno‐affinity of magnetic particles coated with CD81 antibodies and directly quantified the amount of bound EVs using bead‐based flow cytometry. The EVs on the beads were stained with an anti‐CD63 IgG coupled with the fluorescent dye, phycoerythrin (PE).

The subpopulation of EVs characterized using the surface markers CD81 and CD63, consists of small EVs with diameters below 150 nm, derived from multivesicular endosomes.^{[ 32} , ³³ , ^{43 ]} This subpopulation of EVs is commonly applied for therapeutic purposes. We, therefore, selected the CD81+/CD63+ population of EVs as our target for yield optimization. However, our approach is generic and can be applied to optimize the yield of other EV subpopulations of interest.

We first checked whether the CD81+ particles in our clarified medium had the expected size. Light scattering‐based NTA showed that the clarified medium contained particles with an average size of 146 nm (Figure  3a), and transmission electron microscopy (TEM) images showed that these particles had the typical cup‐shape morphology of EVs (Figure 3b). We subsequently applied fluorescence‐based microfluidic diffusion sizing,^{[ 40 ]} a technique that we recently developed to size EVs with multiple nonspecific and specific staining. Staining with anti‐tetraspanin CD81 confirmed that CD81+ particles exhibited an average diameter of 88 nm (Figure 3c; Figure S5, Supporting Information). This result was in good agreement with the known association of CD81 with small EVs.^{[ 32} , ^{33 ]}

Figure 3.

Characterization of CD81+ EVs directly from clarified conditioned medium. a) Light scattering‐based nanoparticle‐tracking analysis of particles in clarified conditioned medium (CCM). b) Transmission electron microscopy image of the EVs in the CCM after isolation by size exclusion chromatography. c) Fluorescence‐based microfluidic diffusion sizing of EVs stained with an APC‐labeled anti‐CD81 IgG. d) Bead‐based flow cytometry of CD81+/CD63+ EVs (left) and CD81+/CD9+ EVs (middle). Controls were used to prove the absence of nonspecific binding of the antibodies to the beads (left and middle panels) and to ensure that antibodies were specifically binding to the EVs (right panel).

Next, the presence of other EV surface markers on the CD81+ particles in the medium was verified. We captured the particles on magnetic beads coated with anti‐CD81 IgGs and stained the beads with antibodies against CD63 and CD9. The flow cytometry analysis showed that both proteins were present on CD81+ particles, but CD63 was more abundant than CD9 (Figure 3d). Indeed, the shift in the median fluorescence intensity of the samples compared with that in the controls was significantly larger for the sample stained for CD63.

Based on these analyses, we concluded that using magnetic beads with anti‐CD81 antibodies and anti‐CD63 staining, we could mostly capture and quantify sEVs. We optimized the flow cytometry conditions to avoid saturation of the beads with EVs and to work over a linear range (Figure S6, Supporting Information).

With this approach, we measured the median fluorescence intensities (MFI) of the beads introduced in the clarified medium, which was harvested during the batch‐refeed and batch cultures. The MFI values are proportional to the amounts of CD81+/CD63+ EVs in the clarified conditioned medium. In particular, a twofold increase of MFI (reported as arbitrary unit AU) indicates a 1.7‐fold increase in CD81+/CD63+ EVs (Figure S6, Supporting Information). MFI increased until the cell bleed was started in both the high‐ and low‐density cultures (Figure  4a,b). When the cell density was kept constant, the concentration of CD81+/CD63+ EVs reached a stationary value similar to all other cell culture parameters (i.e., glucose and lactate concentrations, cell viability, and cell density, see Figure 2). Because no loss in the viability of cells was recorded at the end of the batch‐refeed cultures, EV production could have been extended to longer periods.

Figure 4.

The titer of CD81+/CD63+ EVs and cell‐specific productivities in batch‐refeed and batch cultures. a–c) Bead‐based flow cytometry analysis of the CD81+/CD63+ EVs at each culture day. The amount of CD81+/CD63+ EVs in the medium corresponds to the MFI (AU) of the beads. A twofold increase in AU corresponds to a 1.7‐fold increase in EV particles. d–f) Cell‐specific productivities of CD81+/CD63+ EVs.

A different trend was observed for the batch culture (Figure 4c). In this case, the concentration of EVs initially increased and reached the maximum after 4 days. Following glucose depletion and the corresponding loss of cell viability, the concentration of EVs in the solution decreased. These results indicate that the analyzed EV subpopulation was mainly released when cells were under conditions that favored their growth. The arrest of cell growth probably led to a stop in EV production, and the uptake of EVs by the cells led to a decrease in EV concentration over time.

To achieve a better understanding of the conditions that favored EV production, we computed the cell‐specific productivities, that is the amount of EVs produced by a given cell each day (Figure 4d–f). The overall productivity was lower at higher cell densities, possibly due to higher uptake rates by the viable cells or due to slower secretion, for example, because of glucose depletion and slow growth rate. Importantly, we observed that the productivity dropped below zero in batch cultures, indicating that the produced EVs can be taken up by cells or can be damaged by the cell culture conditions quicker than they can be secreted.

In addition to the concentration of CD81+/CD63+ EVs, we verified whether the expression of other EVs remained constant under long cultivation times by bead‐based flow cytometry. We observed that both the overall amounts of CD81+ and CD81+/CD9+ EVs were constant in both batch‐refeed cultures (Figure  5 ). Therefore, the two‐weeks culture time did not impact several key biochemical properties of the produced CD81+ EVs.

Figure 5.

The titer of CD81+ and CD9+/CD81+ EVs. Bead‐based flow cytometry analysis of a,b) the CD81+ EVs and c,d) the CD81+/CD9+ EVs at different culture days. The amount of EVs in the medium corresponds to the MFI (AU) of the beads. The fluorescence intensity of beads in the absence of samples was 2.4 × 10³ AU in the APC channel, and 5.0 × 10⁴ AU in the FITC channel. The experiment was performed in triplicates and data represent means ± SD. Dotted lines are guides to the eyes.

After analyzing the production and the properties of CD81+ EVs, we measured the number of total particles in the medium using light scattering‐based NTA to get an understanding of the conditions that maximized the amount of CD81+/CD63+ EVs in the medium, that is, the target EV subpopulation in this study (Figure  6a–c). The measured concentration of particles in each sample was two orders of magnitude larger than the value measured in fresh CD293 medium (1.2 × 10⁸ particles mL⁻¹).

Figure 6.

The titer of particles and fraction of target CD81+/ CD63+ particles in batch‐refeed and batch cultures. a–c) Light‐scattering‐based NTA of the concentration of particles in the medium over time. d–f) Fraction of the target CD81+/CD63+ EV subpopulation over time, measured as the ratio between the MFI and the total particles determined using the NTA. The experiment was performed in triplicates and data represent means ± SD. Dotted lines are guides to the eyes.

The concentration of total particles in both batch‐refeed cultures was constant over 13 days and was 3.4‐fold higher in the high‐density culture than in the low‐density one. From the ratio between the MFI (reporting on CD81+/CD63+ EVs) and the total particle number, we observed that in the low‐density culture we obtained a higher fraction of the target CD81+/CD63+ subpopulation (twofold compared with that in the high‐density culture) (Figure 6d,e). As a result, the operating conditions of the batch‐refeed cultures can be adjusted to produce samples enriched in different EV subpopulations.

In the batch culture, the particle concentration is not constant and increases over time, in particular when the cell viability drops as a result of apoptosis. This induces a progressive decrease in the fraction of target EVs in the medium with the culture days and complicates the production of reproducible samples (Figure 6f).

2.3. Comparison of EV Yields in Batch and Batch‐Refeed Cultures

To quantify the benefits of batch‐refeed cultures, we computed the space‐time yield (STY), which represents the overall EV yield divided by the process duration and the bioreactor volume. This parameter is recommended for the comparison of the overall productivity of perfusion and batch processes.^{[ 44 ]} To compute the STY of the batch cultures, we considered that 1 day of downtime is usually required for reactor cleaning, vessel setup, and sterilization. On the contrary, for the STY of the batch‐refeed systems, we considered only the EVs produced during steady‐state conditions.

The trends of the STYs for the batch and batch‐refeed cultures were first plotted (Figure  7a). The STY of the batch culture increased up to day 3, whereas the STY of the batch‐refeed cultures constantly increased during the entire 13 days of culture. At the end of the culture, the STY of the low‐ and high‐density cultures were 2.8‐ and 3.2‐times higher, respectively, than the maximum STY of the batch system (Figure 7b). This result indicates that, at a fixed bioreactor volume, the EVs produced per day in a perfusion culture are approximately three times more than those produced in batch culture.

Figure 7.

Comparison between the STY of batch‐refeed and batch cultures. a) STY of the batch culture and low‐ and high‐density batch‐refeed cultures. b) Fold change in the overall STY of the low‐ and high‐density batch‐refeed cultures relative to the maximum STY of the batch.

This result also showed that the concentration of CD81+/CD63+ EVs is only minimally affected by the concentration of cells in the culture. Indeed, the culture with a fourfold higher concentration produced only 14% more EVs. This behavior can be explained with the following mass balance equation:

$$\frac{d{c}_p}{dt}=q_r{c}_{VC}-q_u{c}_{VC}{c}_p-P{c}_p$$ (1)

where c_p is the EV concentration, q_r and q_u are the cell‐specific release and uptake rates of the EVs, c_VC is the concentration of viable cells, and P is the perfusion rate, which is null in the batch reactor. At the steady‐state conditions, the accumulation of product is null and the concentration of EVs is described using the following equation:

$$c_p=\frac{q_r}{q_u+P/c_{VC}}$$ (2)

In agreement with our results, Equation (2) shows that the concentration of the product is not linearly proportional to the concentration of cells. Moreover, when $P/c_{VC}$ is negligible compared with q_u , the product amount is mostly determined by the ratio between the release and uptake rates of individual cells.

3. Discussion

In this study, we demonstrated the benefits of perfusion cultures for EV production. We modeled perfusion cultures on a smaller scale with batch‐refeed operations, also called semi‐perfusion operations, which are one of the most used methods for perfusion‐like cultures.^{[ 34} , ⁴⁵ , ⁴⁶ , ^{47 ]} This method has been used to optimize process conditions, and study product quality and productivity, while avoiding the high costs and risks of actual perfusion runs. Although this study does not involve a real perfusion culture, the batch‐refeed model demonstrates for the first time the impact of steady‐state perfusion operations on the yield and quality of EV.

In this work, we tested HEK293 F cells adapted for suspension culture in chemically‐defined serum‐free media which is one of the preferred culture modes in industrial settings.^{[ 23 ]} Cultures can be scaled up in stirred tank bioreactors (STRs) operated in perfusion mode, a setup that allows cell densities up to 80 × 10⁶ cells mL⁻¹,^{[ 41 ]} and operations at scales up to 2000 L.^{[ 46 ]} Compared to conventional methods applied in the EV field, such as hollow fiber bioreactors (HFBRs) with adherent cells, this setup significantly facilitates culture scale‐up and control.^{[ 48 ]} Indeed, it allows perfusion and continuous operations by constant cell removal and medium replacement, as well as simple cell sampling and culture monitoring. Moreover, it avoids the formation of gas and nutrient gradients within the bioreactor, a common issue of HFBRs that contributes to decrease in the reproducibility and control on EV batches.^{[ 48} , ^{49 ]} STRs are often used to grow adherent cells on microbeads or as aggregates, but in such setup constant and controlled removal of cells for continuous operations is still challenging.^{[ 49 ]}

We analyzed the amounts of CD81+, CD81+/CD63+, and CD81+/CD9+ EVs in the cell culture medium of the batch‐refeed cultures.^{[ 34 ]} Importantly, we observed that when cells were grown under constant conditions either with an excess or with a defect of nutrients, the productivities could be kept constant over time. This approach can be easily adapted to other EV subpopulations by simply changing the antibodies used for staining.

For the CD81+/CD63+ EV subpopulation, we observed that EVs are mostly produced during the growth phase of cells and that cell‐specific productivity decreases as the cell number increases. This trend is in agreement with previous observations reported in the literature, wherein EV productivity was shown to decrease with time as the number of cells in the culture increased.^{[ 31 ]} These results can be explained by the fact that EVs are both released and taken up by producing cells and show the importance of maintaining the culture under steady‐state conditions at which the release of EVs is faster than their uptake.

This behavior is different from the situation observed, for instance, with monoclonal antibodies produced from CHO cells, wherein the cell‐specific productivity increases linearly with cell density.^{[ 34} , ^{50 ]} In contrast with other biologics, for EV manufacturing, the effect of cell density could be smaller for yield optimization, although increasing cell density is generally beneficial for overall EV production.

With our batch‐refeed cultures, we achieved STYs threefold higher than in batch cultures. We note that the STYs of the batch‐refeed culture can be further optimized by implementing several improvements. The effect of the initial transient period, in which steady‐state and the desired cell density have not been reached, can be decreased by increasing the overall duration of the batch‐refeed cultures. Moreover, this transient time can be decreased by selecting a faster‐growing cell line. Last, the perfusion rates could be further optimized to test their effect on the growth of cells and nutrient availability, and thereby, on the release of CD81+/CD63+ EVs.

We note that higher yields are not the only advantage of using batch‐refeed and perfusion cultures over batch cultures. Most importantly, with batch‐refeed culture, we achieved a consistent production of the target EV subpopulation over 13 days. Because cell growth conditions affect the properties of EVs, such as their metabolic signature,^{[ 51 ]} the constant and controlled conditions in perfusion cultures increase the reproducibility, homogeneity, and quality of EV samples, as demonstrated by the combination of microfluidic diffusion sizing, bead‐based flow cytometry, and NTA. Despite these advantages, we highlight that perfusion processes and product qualities can be affected by the genomic instabilities of the source cells. Although in this work we have shown that key properties of our target EV subpopulation are preserved over two weeks, this result cannot be extended to all subpopulations and EVs from other cell sources. Constant product analysis is necessary to prevent drifts in key quality attributes of different EV samples during perfusion processes.

In this context, our recently developed microfluidic diffusion sizing method with fluorescence detection^{[ 40 ]} is important to complement the current analytical methods and go toward innovative process analytical technologies (PAT). These methods require a limited amount of material and allow rapid analysis, for both quality control and process design, and in the future can be integrated, for instance, with machine learning methods.^{[ 52 ]}

Finally, we remark that batch‐to‐batch variability is a major issue in the field, whose solution is challenged by the complex and scarcely understood heterogeneity of EVs.^{[ 7 ]} We have shown that in batch cultures the composition of EV samples is extremely sensitive to the culture state and is constantly changing over time. In this context, we believe that perfusion cultures allow the production of more controlled and reproducible EV samples, eventually enriched in specific subpopulations, and are hence a step toward the identification of functional EV subpopulations, an important milestone in the EV field.^{[ 53 ]}

4. Conclusion

In this study, we investigated the use of perfusion cultures for the upstream manufacturing of EVs. We simulated perfusion cultures with batch‐refeed ones and measured the rates of production and yields for a target subpopulation of EVs, characterized by CD81 and CD63 biomarkers. The batch‐refeed cultures allowed constant production of our target EVs for 13 days and space‐time yields up to threefold higher than those of the batch process. Moreover, the concentration of the total particles in the medium and, therefore, the fraction of our target EV subpopulation in the medium were also steady during the entire culture period. Overall, we demonstrate the potential of perfusion cultures for controlling the production of specific EV subpopulations as well as for improving the yield in EV manufacturing.

5. Experimental Section

Cell Culture

293F cells (ThermoFisher Scientific) were cultured in CD293 medium (ThermoFisher Scientific) in vented shake flasks without baffles. Flasks were shaken in an incubator with 8% CO₂ and 90% humidity at 125 rpm and 37 °C.

In the batch cultures, the cells were grown in 100 mL medium in 500 mL flasks. Three milliliters of conditioned medium were collected every day and clarified in two centrifugation steps, at 200 × g for 5 min and then at 3000 × g for 15 min. The clarified medium was then stored at −20 °C until analysis.

In the batch‐refeed cultures, the cells were grown in 20 mL of medium in 125 mL flasks, and a perfusion rate of 1 RV day⁻¹ was used. The perfusion rate P (day⁻¹) is defined as:

$$P=\frac{V_{\mathrm{Exchange}}}{V_{\mathrm{Culture}}\mathrm{\Delta }t}$$ (3)

where V _Exchange,i is the volume of medium fed and V _Culture is the total volume of the cell culture. The maximum density of viable cells was determined by allowing the cells to grow for 3 days in the batch mode and 11 days in the batch‐refeed mode. Every 24 h, the cell density, cell viability, and glucose and lactate concentrations were measured. The cells were then pelleted at 200 × g for 5 min; the conditioned medium was removed, and the cells were resuspended in 20 mL of fresh medium.

A similar procedure was followed for the steady‐state experiments. Cells were grown for 3 days in the batch mode and then for a few days in the batch‐refeed mode, without a cell bleed, as described above. When the cell density reached the target cell density, which was 2.1 × 10⁶ cells mL⁻¹ in the low‐density culture (reached on day 5) and 8.5 × 10⁶ cells mL⁻¹ in the high‐density culture (reached on day 8), every day the medium was exchanged and a fraction of the cultures (the cell bleed) was removed to bring the number of cells to the desired value. This volume was on average 40% and 30% for the low‐ and high‐density cultures, respectively. After the cells were bled from the culture, the remaining cells were pelleted at 200 × g for 5 min and resuspended in 20 mL of fresh medium. The harvested medium was then centrifuged at 3000 × g for 15 min and stored at −20 °C. This process was performed every 24 h for 13 days.

Cell Culture Analysis

Cell density and viability were measured using the trypan blue exclusion method (Cedex HiRes, Roche Diagnostics). Cell culture (300 µL) was measured undiluted or three times diluted in PBS and pictures of the cells were taken at every measurement (Figure S3, Supporting Information). Glucose and lactate concentrations were measured using the enzymatic‐amperometric measuring principle with the Super GL Compact (Hitado) according to the manufacturer's instructions. The instrument performance was checked daily with a calibration solution. Cell culture (5 or 10 µL) was spiked in the measurement cartridges and analyzed immediately.

Analysis of the Cell‐Specific Perfusion and Growth Rates

The cell‐specific perfusion rate CSPR (pL cell⁻¹ day⁻¹) was calculated as follows:

$$CSP{R}_i=\frac{P}{X_i}$$ (4)

where, X_i is the cell density on day i. The cell‐specific growth rate µ _i (day⁻¹) was calculated as follows:

$${\mu }_i=\frac{\ln X_i-\ln X_{i-1}}{\mathrm{\Delta }t}$$ (5)

where, X_i is the cell density on day i and t is the process time.

Transmission Electron Microscopy (TEM)

For TEM analysis, EVs were first isolated by size exclusion chromatography as previously reported.^{[ 54 ]} In brief, clarified conditioned media (50 mL) was filtered through a 0.22 µm membrane and incubated with 25 U of Pierce Universal Nuclease (ThermoFisher Scientific) for 2.5 h at room temperature. The sample was then concentrated 100 times using an Amicon‐15 centrifugal filter (50 kDa MWCO, RC membrane, Merck Millipore) and the concentrated EVs were loaded on a gravity flow chromatography column packed with 10 mL Sepharose CL4B resin (Cytiva) using PBS as running buffer. Fractions (20, 500 µL each) were collected and analyzed by ELISA (as previously described^{[ 54 ]}), NTA, Micro BCA Protein Assay (ThermoFisher Scientific), and Quant‐IT dsDNA Assay (High sensitivity, ThermoFisher Scientific) (Figure S4, Supporting Information). Fractions 6–8 were pooled together and analyzed by TEM. Five µL of the sample were placed on glow discharged (negatively at 25 mA for 30 s in an Emitech K100X glow discharge system, Quorum Technologies Ltd.) carbon‐coated grids (Plano GmbH) and left to adsorb at room temperature for 60 s. The excess liquid was drained with a filter paper and the samples were exposed to negative staining with 2% Uranyl acetate (w/v) by two successive incubations of 1 and 15 s, respectively. The grids were air‐dried and imaged in a JEM‐1400Flash Electron Microscope (Jeol) in bright field mode operated at 100 kV.

Nanoparticle Tracking Analysis (NTA)

NTA of EVs was performed on a ZetaView instrument equipped with a CMOS camera and a 405 nm laser (Particle Metrix). The chamber was calibrated daily with polystyrene nanoparticle standards according to the manufacturer's recommendation. Samples were diluted in phosphate‐buffered saline (PBS) to a particle concentration between 10⁷ and 10⁹ particles mL⁻¹ and injected into the sample chamber using a 1 mL syringe until it was filled. Video acquisition was performed for all samples at 11 positions, applying 80% scattering intensity, 100 shutters in light scattering mode, with a trace length of 15 frames and a frame rate of 30 s⁻¹. Data were analyzed using the ZetaView analysis software (ZetaView 8.04.02 SP2).

Fluorescence‐Based Microfluidic Diffusion Sizing (fluoMDS)

Fluorescence‐based microfluidic diffusion sizing analysis was performed as described previously.^{[ 40 ]} In brief, 500 µL of conditioned medium was concentrated 12 times in a Vivaspin500 (50 kDa MWCO, RC membrane, Sartorius). Ten microliters of the conditioned medium were blocked for 1 h at room temperature with 0.1% BSA in PBS and then incubated for 2 h at room temperature with either APC‐conjugated anti‐CD81 antibody (1D6, 1:500 dilution in blocking buffer, Invitrogen) or APC‐conjugated mouse IgG1 kappa isotype control (1:500 dilution in blocking buffer, Invitrogen). Five microliters of sample were then loaded in the fluoMDS device that was run at 60 µL h⁻¹ using the blocking buffer as a running buffer. After image acquisition with a Ti2‐U inverted microscope (Nikon), the diffusion profiles were fitted and average particle sizes were computed as previously reported.^{[ 40 ]}

Bead‐Based Flow Cytometry

For the initial biomarker characterization, 5 µL beads (Exosome‐Human CD81 Flow Detection Kit, Invitrogen) were incubated for 22 h at 4 °C with a clarified conditioned medium (20 µL), diluted five times in the assay buffer (0.1% BSA in PBS). After two washes with the assay buffer, the beads (with and without EVs) were incubated for 1.5 h at room temperature in 100 µL antibody solutions. These contained either PE‐conjugated anti‐CD63 antibody (H5C6, 1:80 dilution in the assay buffer, Invitrogen), FITC‐conjugated anti‐CD9 antibody (eBioSN4, 1:80 dilution in the assay buffer, Invitrogen), or PE‐conjugated mouse IgG1 kappa isotype control (1:80 dilution in the assay buffer, Invitrogen). After two washing steps, the beads were resuspended in 150 µL assay buffer and analyzed on a CytoFLEX S flow cytometer (Beckman Coulter).

The same protocol was applied for the semi quantification of EVs in the harvested conditioned medium. However, 5 µL of clarified conditioned medium diluted 20 times in the assay buffer (0.1% BSA in PBS) was added to the beads, and either the PE‐conjugated anti‐CD63 antibody, APC‐conjugated anti‐CD81 antibody (1D6, Invitrogen) or FITC‐conjugated anti‐CD9 antibody were used. All antibodies were added to reach a 1:80 dilution in the assay buffer.

Analysis of Productivity and Space‐Time Yield

The harvest rate H_i (day⁻¹) was calculated as follows:

$$H_i=\frac{V_{\mathrm{Harvest},i}}{V_{\mathrm{Culture}}\mathrm{\Delta }t}$$ (6)

where, V _Harvest,i is the volume of medium harvested on day i, V _Culture is the total volume of the cell culture, and Δt is the time interval between each harvest, corresponding to 1 day in the present study.

The cell‐specific productivity q_p of batch (B) and batch‐refeed (BR) cultures (AU cell⁻¹ day⁻¹) were computed using the following equations:

$$q_{p,B,i}=\frac{\left(c_{p,i}-c_{p,i-1}\right)}{\mathrm{\Delta }t\ast \overline{X_i}}$$ (7)

$$q_{p,BR,i}=\frac{c_{p,i}\ast H_i}{\overline{X_i}}$$ (8)

where c _p,i is the concentration of EVs on day i (AU mL⁻¹), t is the processing time (day), and X_i is the viable cell density on day i (cells mL⁻¹). For these calculations, H_i was assumed equal to 1 to evaluate the overall amount of EVs produced by each cell and not only by the harvested ones.

The space‐time yield, STY, of the batch (B) cultures (particles mL⁻¹ day⁻¹) was calculated as follows:

$$ST{Y}_{B,i}=\frac{c_{p,i}}{(t_i-t_0)}$$ (9)

where, c _p,i is the concentration of EVs on day i (particles mL⁻¹), t_i is the processing time at which the EV concentration is maximum (day), and t ₀ is the processing time at which the culture was started. For batch‐refeed (BR) cultures, the STY was calculated as follows:

$$ST{Y}_{BR,i}=\frac{Y_{BR,i}}{V_{\mathrm{culture}}\left(t_i-t_0\right)}$$ (10)

where, V _culture is the volume of the culture, t ₀ and t_i are the process times at the beginning and on day i, and Y _BR,i is the overall product yield of the batch‐refeed culture on day i (particles), calculated using the following equation:

$$Y_{BR,i}=\sum _{i=1}^n{c}_{p,i}{H}_i\mathrm{\Delta }t{V}_{\mathrm{Culture}}$$ (11)

For these calculations, H_i was assumed equal to 1 and only the product produced after starting the cell bleed was considered. Moreover, for the STY calculations, the concentration of EVs in particles mL⁻¹ was obtained from the MFI values through a calibration curve generated with EVs purified using size exclusion chromatography (Figure S6, Supporting Information).

EV Track

The authors have submitted all relevant data of the experiments to the EV‐TRACK knowledgebase (EV‐TRACK ID: EV220316).^{[ 55 ]}

Statistical Analysis

Data analysis and graphs were performed using Matlab 2019a (Mathworks) and Python Spyder IDE (Anaconda). All data with error bars were presented as mean ± standard deviation (SD).

Conflict of Interest

The authors declare no conflict of interest.

Supporting information

Supporting information

Paganini C., Boyce H., Libort G., Arosio P., High‐Yield Production of Extracellular Vesicle Subpopulations with Constant Quality Using Batch‐Refeed Cultures. Adv. Healthcare Mater. 2023, 12, 2202232. 10.1002/adhm.202202232

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.