Does hollow fiber bioreactor culture make placental MSC extracellular vesicles more potent?

A donor-matched comparison of flasks and bioreactor

The paper asks whether the vessel a cell grows in changes the therapeutic quality of the particles it releases. PMSCs were isolated from chorionic villus tissue of discarded second-trimester human placenta and expanded first in conventional 2D flasks. Portions of the same cell populations were then seeded either into Falcon 875 cm2 five-layer flasks or into a FiberCell Duet hollow fiber bioreactor, so that cells from the same donor could be compared at matched passage.

The bioreactor is the intervention. It holds a cartridge of parallel porous polysulfone fibers with a 20 kDa cutoff, through which medium is pumped at 30 ml per minute; cells attach in a 3D arrangement that the authors describe as mimicking capillary beds. Four PMSC donor lines were used across the study, with two lines cultured simultaneously in the bioreactor alongside donor-matched 2D cultures. EVs from both conditions were isolated by the same tangential flow filtration sequence, first through a 0.65 µm filter to remove cells and debris, then through a 500 kDa cutoff filter to remove medium and small soluble factors.

That shared isolation step matters for reading the results: the comparison is between culture systems, not between purification methods. The 2D comparator was a six-day culture in five-layer flasks harvested at roughly 95% confluence, described by the authors as the maximum EV output of one 2D batch, while bioreactor medium was collected weekly over 30 days of continuous culture.

Bioreactor cells keep their MSC identity

Before comparing EVs, the study checked whether the bioreactor changed the cells themselves. After 30 days of continuous culture, PMSCs were trypsinized from the cartridge and counted; final counts were consistent among donor lines and all cultures reported at least 90% viability. Average doubling times stayed around 45 hours across all donor lines in both systems.

Phenotyping by flow cytometry followed International Society for Cell Therapy guidelines: bioreactor-grown PMSCs expressed CD29, CD90, CD44, CD105 and CD73, and were negative for CD31, CD45 and CD34. Karyotyping at the 30-day timepoint found no significant differences in chromosome structure between 2D and bioreactor PMSCs, indicating no large chromosomal changes from the bioreactor system. Trilineage differentiation testing showed strong chondrogenic, modest osteogenic and low adipogenic potential, with 2D-cultured PMSCs showing similar capabilities under the same protocol.

This is the control that makes the potency result interpretable. If bioreactor EVs were more active while the producer cells still looked like MSCs and carried a normal karyotype, the difference is better attributed to culture-induced changes in what the cells package and release than to a shift in cell type.

Yield rises far more than particle size changes

The most concrete production result is yield. Comparing total PMSC-EVs isolated per donor line per culture condition, bioreactor culture increased total EV yield by 30–40x, up to 8E+11, and EV yield per cell by roughly 5x, up to 8000 particles per cell. These are the paper's own culture-and-isolation outputs, not projections for a manufacturing run.

Physical characterization was largely stable. Nanoparticle tracking analysis showed consistent size distributions between 2D and 3D EVs across all four collection timepoints, and zeta potential stayed around −30 mV across conditions and timepoints, indicating no significant change in net surface charge on a bulk scale. Cryogenic transmission electron microscopy showed the canonical lipid bilayer and circular shape with little morphological difference between conditions.

Two details complicate a simple "smaller is better" reading. PMSC-EVs collected at the 30-day timepoint showed a left shift in size distribution with fewer large-diameter particles than 2D counterparts, and average EV size decreased with statistical significance in later bioreactor timepoints and from 2D to 3D culture. Separately, single-molecule dSTORM imaging of CD9, CD63 and CD81 found that the share of triple-positive EV clusters rose from 70–75% in 2D EVs to 75–85% in 3D timepoints. Yield, size and tetraspanin positivity are different measurements and should not be treated as one potency score.

The EV protein cargo shifts with culture system

To ask why potency might change, the authors ran quantitative LC-MS/MS on five EV samples: one from 2D culture and four collected across successive weeks of bioreactor culture. Across these five conditions they identified 3449 proteins, 827 of which were shared by all EV samples. Principal component analysis separated the EV groups, and the paper reports significantly enriched proteins in week 4 bioreactor EVs relative to 2D EVs, with the top five most significantly expressed proteins highlighted.

The abstract names examples of bioreactor-enriched proteins: neuraminidase 1, dynamin-related protein 1, and matrix metalloproteinases involved in tissue remodeling, apoptosis regulation and immune modulation. Gene ontology analysis of week 4 EVs highlighted terms reflecting increased interaction with extracellular entities such as the surrounding extracellular matrix, foreign antigens, and free or cell surface-bound proteins. The discussion adds that several differentially expressed proteins have known roles in neuronal development and neurological disorders, listing neurocan core protein, neuropilins, neurofibromin, brain-derived neurotrophic factor, neurotensin receptors and neurolysin.

These are compositional associations, not demonstrated causal mechanisms. The paper's own framing is that the proteomic differences suggest possible mechanisms of action for the neuroprotection results; it does not show that any single enriched protein produces the functional gain.

More neuroprotection and faster wound closure

Functional potency was tested in two assays. In the neuroprotection assay, SH-SY5Y neuroblastoma cells were treated with staurosporine to induce apoptosis and then given equal EV doses of 2000 EVs per cell, or 2E+08 EVs per well, from 2D or bioreactor culture. Untreated cultures showed low confluence and reduced neurite branching. Both EV sources increased cellular networks and branching points, and quantification showed greater neuroprotective potential for bioreactor-produced EVs from donor line 802. A trend of increasing neuroprotection appeared with later bioreactor collection timepoints, with week 4 EVs giving the greatest effect.

In the endothelial scratch assay, human umbilical vein endothelial cells were wounded and treated with matched doses of 2000 EVs per cell, or 1.8E+08 EVs per well. Uncovered-area fold change measured in ImageJ showed increasing potency of bioreactor-produced EVs, with a slight trend toward greater potency at later harvest timepoints. The PBS negative control showed little to no wound closure at six hours, while the 10% FBS medium positive control showed the greatest healing.

The consistency claim rests on donor replication: the remaining three donor lines showed a similar increase in neuroprotective efficacy and a similar wound-healing effect. Because the assays use equal EV numbers rather than equal protein or equal volume, the potency comparison is per particle, and it is separate from the 30–40x yield gain.

What the comparison does and does not establish

Several boundaries sit close to these results. The 2D comparator was a single six-day batch harvested near confluence, while bioreactor EVs accumulated over weekly collections across 30 days; the paper attributes the potency difference to the 3D culture environment, but the two production schedules are not identical in duration. Functional testing used one neuroblastoma line and one endothelial cell type in vitro, so the neuroprotective and wound-healing labels refer to those assays, not to outcomes in animals or patients.

The authors also note that much published EV therapeutic literature comes from 2D culture while clinical translation requires bioreactor systems, and they describe this as, to their knowledge, the first comparative profiling of PMSC-EV cargo and function from hollow fiber bioreactor versus traditional 2D flask culture. Their stated next step is investigation of PMSC and PMSC-EV production in a cGMP setting for clinical trial testing, which the paper does not perform.

The protein-level explanation remains correlative: enriched proteins and gene ontology terms are reported alongside functional gains, and the discussion presents them as potential mechanisms rather than demonstrated causes.

Later bioreactor harvests carry the strongest signal

The paper's most decision-relevant pattern is internal to the bioreactor run. Neuroprotection trended upward with later collection timepoints and peaked at week 4; wound healing showed a slight trend in the same direction; week 4 EVs carried the significantly enriched protein set and the gene ontology terms pointing to extracellular matrix and protein interaction. If a reader is weighing hollow fiber production for PMSC-EVs, the comparison to prioritize is not simply 2D versus 3D but which bioreactor harvest week is used, because the potency and cargo evidence is strongest at the late timepoint.

The second priority is to keep the two gains separate. The 30–40x total yield increase and roughly 5x per-cell increase describe how many particles a culture produces; the neuroprotection and wound-closure results describe what equal particle doses do in two in vitro assays. A production decision that needs volume and a potency decision that needs per-particle activity draw on different numbers from this study, and only the donor-matched, in vitro evidence reported here supports either one.

Culture system is a potency variable, not just a scale variable

The study's central claim is that switching PMSCs from 2D flasks to a hollow fiber bioreactor is not a neutral scale-up step. Donor-matched cells retained MSC markers, differentiation potential and a normal karyotype, yet their EVs shifted in protein cargo, became more uniformly tetraspanin-positive, and performed better per particle in neuroprotection and endothelial wound-closure assays, with the strongest effects at the latest bioreactor timepoints.

For readers comparing production options, the practical threshold implied by this paper is the harvest timepoint: week 4 bioreactor EVs, not an averaged bioreactor output, carry the enriched protein profile and the largest functional effect. The yield numbers and the potency numbers answer different questions, and the paper supports using both, while leaving in vivo efficacy and cGMP manufacturing as work still to be done.

Sources used in this answer