Gating EVs by Size with Sequential Tangential Flow Filtration
Most tangential flow filtration (TFF) protocols for extracellular vesicles (EVs) use the membrane as a net. Everything bigger than the cut-off stays in the retentate, everything smaller goes to waste, and the EVs come out as a single concentrated population. TFF can also be used to gate EVs based on their size.
The idea is straightforward. Run the sample over a large-pore cartridge first: larger EVs stay in the retentate and smaller EVs pass into the permeate. Then run that permeate over a smaller-pore cartridge, which retains the smaller EVs and lets free protein through. If only the large EVs are wanted, the first round is all you need.
In this article we look at whether this idea actually works in practice, what has been published, and why the boundaries between fractions are softer than the numbers on the cartridge labels suggest.
Why Separate EVs By Size
EV preparations are mixtures. The MISEV2018 guidelines recommend describing EVs by operational terms rather than by presumed biogenesis, with "small EVs" for particles below 200 nm and "medium/large EVs" for those above, precisely because isolation methods cannot yet reliably tell exosomes from microvesicles.1 Size is the property most methods can actually measure.
And size seems to matter. Asymmetric flow field-flow fractionation of cancer cell secretions resolved two exosome subsets, large (90 to 120 nm) and small (60 to 80 nm), plus a population of non-membranous particles of around 35 nm that the authors named exomeres, each with distinct protein, lipid and nucleic acid content.2 Proteomic comparison of EVs separated by centrifugation speed and density also found that markers often treated as exosome-specific are shared across subtypes, so a size band is not the same thing as a biogenesis pathway.3 More recently, mesenchymal stem cell small EVs split by a multi-module TFF system showed size-dependent differences in cellular uptake and in their effects on proliferation and migration.4
For anyone developing an EV therapeutic or a diagnostic, that raises an awkward question: which part of the bulk preparation is doing the work?
How Sequential TFF Divides A Sample
The two-round scheme maps onto a small number of designs, depending on which fraction is wanted.
| Gating EVs by size with one or two TFF stages | ||
| Target fraction | Configuration | Where the target ends up |
| Larger EVs only | Single stage on a large-pore cartridge (e.g. 1,000 kDa), followed by diafiltration | Stage 1 retentate |
| Smaller EVs, larger EVs excluded | Stage 1 permeate fed to a smaller-pore cartridge (e.g. 300 or 500 kDa) | Stage 2 retentate |
| Both, as separate fractions | Two stages in series, collecting both retentates | Stage 1 retentate and stage 2 retentate |
| Three or more bands | Cascade of cartridges with decreasing pore size, each fed by the permeate of the one before | The retentate of each stage |
One detail is worth emphasising: Concentrating on the first cartridge does not by itself give a clean large-EV fraction, because the retentate liquid still contains small EVs at roughly their starting concentration. Smaller particles only leave the retentate as liquid passes through the membrane. To deplete them, the stage 1 retentate needs diafiltration: buffer added to the feed while permeate is drawn off, washing the smaller particles through over several volumes. Each of those volumes ends up in the stage 1 permeate, which is then larger and more dilute when it reaches the second cartridge.
Sequential filtration as a concept is not new. A three-step protocol published in 2014 combined dead-end prefiltration, TFF to remove free protein, and a final low-pressure pass through a track-etched membrane to select exosome-sized vesicles from large volumes of conditioned medium and biofluid.5 What has changed since is that the size gating itself has moved onto tangential flow cartridges.
What Has Been Published
Several groups have now put two or more TFF membranes of different pore size in series.
| Published multi-membrane TFF approaches to EV size separation | ||
| Study | Membranes | Outcome |
| Kim et al. 20216 | 200 nm and 30 nm membranes in a dual cyclic microfluidic TFF device | Isolated the 30 to 200 nm range in one step, with better size selection than direct filtration or a single cyclic stage |
| Bae et al. 20247 | 200 nm and 50 nm hollow fibre filters run simultaneously in tandem | Isolated 50 to 200 nm small EVs from sample volumes of 15 µL up to 2 L |
| Su et al. 20254 | Cassette-based TFF with multiple membrane modules | Processed 2 L of conditioned medium in about 25 minutes and split MSC small EVs into functionally different size subpopulations |
| Scarborough et al. 20268 | Five hollow fibre modules in series: 0.2 µm clarification, then 0.05 µm, 500 kDa, 300 kDa and 100 kDa | Separated larger EVs, smaller EVs and exomeres from HEK293T medium, routinely handling 360 mL in about three hours |
The Scarborough study is the most detailed test of the cascade idea, and it reaches below EVs altogether. The exomere fraction collected on the 100 kDa membrane was negative for CD9, CD63 and flotillin-1 and enriched in the RNA interference proteins AGO2 and HSP90AB1, and it appeared purer than the equivalent fraction from differential ultracentrifugation.8 The cartridges were cleaned with sodium hydroxide and reused for five to seven isolations.
So the short answer to "can it be done" is yes. The longer answer depends on how sharp you need the boundaries to be.
Why The Cut-Off Is Not Definitive
A cartridge label gives a single number, but a membrane has a distribution of pore sizes. Rejection of a particle is a probability that rises with particle size, not a switch.
Scarborough and colleagues used a log-normal pore distribution model to choose their membranes, and the predictions are instructive.8 For a 300 kDa membrane with a mean pore diameter of about 15 nm, the model predicted rejection of 72% of 60 nm particles and 88% of 100 nm particles. Both are mostly retained. The practical consequence is that a cartridge only discriminates between EVs when its pores are on the same scale as the vesicles. The 300 and 500 kDa cartridges routinely used for EV concentration hold back almost all EVs, which is exactly why they are used: they separate EVs from free protein, not EVs from each other. For the upper gate, the published studies above used microfiltration membranes of 0.05 to 0.2 µm.
The most open ultrafiltration ratings reach into the EV range too. Approximate pore diameters for the EVlution™ hollow fibre cartridges are shown below.
| EVlution™ cartridge ratings and their role in sequential TFF | ||
| Rating (kDa) | Approx. pore diameter (nm) | Likely role |
| 300 | ~25 | Stage 2: retains small EVs, passes free protein |
| 500 | ~45 | Stage 2: retains small EVs with better protein clearance, at some risk to the smallest vesicles |
| 750 | ~65 | Stage 1 for a low boundary: retains most EVs, passes a share of the smallest |
| 1,000 | ~100 | Stage 1: enriches larger EVs in the retentate, passes a substantial share of smaller EVs |
A 1,000 kDa then 300 kDa sequence therefore gives a split at around 100 nm, inside the small EV range, rather than at the 200 nm line used in the MISEV convention. Because rejection is probabilistic, many vesicles below 100 nm will still be held in the first retentate. Conversions between kDa ratings and pore diameter also vary between manufacturers and membrane chemistries: the 300 and 500 kDa membranes modelled by Scarborough were given mean pore diameters of about 15 and 25 nm.8 The figures above are a starting point for design, and the fractions they produce should be measured rather than assumed.
Even then, fractions overlap. Exomere-like particles turned up at moderate abundance in the Scarborough EV fraction,8 which is what partial rejection would predict.
Three further effects shift the effective cut-off during a run:
- The fouling layer. Retained particles and protein build up at the membrane surface and act as a second, tighter filter. As a run progresses, particles that passed freely at the start may be held back. This is the same polarisation layer that drives rising pressure during concentration, which we discussed in an earlier article on flux and pressure in TFF.
- Transmembrane pressure. Vesicles are deformable. Liposome extrusion studies show that lipid vesicles are forced through pores smaller than themselves once pressure exceeds a threshold, emerging smaller as they do so.9 Whether EVs behave the same way at TFF operating pressures has not been properly measured, but it argues for running the gating stage at low transmembrane pressure and high crossflow.
- The yield and purity trade-off. In a two-stage process for EVs from dairy whey, a 750 kDa ultrafiltration membrane kept EV loss close to zero but removed impurities less effectively than a 0.15 µm microfiltration membrane.10 A more open membrane cleans better and loses more.
None of this makes size gating unworkable. It means the fractions should be treated as enriched rather than pure, and characterised rather than assumed.
Setting Up A Two-Stage Run
| Practical choices for sequential TFF of EVs | ||
| Step | Suggested approach | Reason |
| Clarification | Low-speed centrifugation to remove cells, debris and apoptotic bodies before any TFF | Large debris fouls the first cartridge and contaminates the large-EV fraction |
| Stage 1 cartridge | The largest pore that sits near the boundary you want: 1,000 kDa (~100 nm) for a split within the small EV range, or 750 kDa (~65 nm) for a lower one | Only pores on the scale of EVs discriminate between EV sizes |
| Stage 1 operation | Low transmembrane pressure, high crossflow, then diafiltration of the retentate if the large-EV fraction is to be kept | Limits fouling and deformation; washes smaller EVs out of the large-EV fraction |
| Stage 2 cartridge | 300 or 500 kDa, with enough membrane area for the larger, more dilute stage 1 permeate | Retains small EVs while passing most free protein |
| Polishing | Size exclusion chromatography of each retentate | Removes residual soluble protein that diafiltration leaves behind |
| Verification | Particle size distribution and marker profile for every fraction, including the permeates | Shows how much overlap there actually is between bands |
Recovery deserves attention too. Every stage loses some material to the membrane and the hold-up volume, and those losses multiply across a cascade. Measuring particle number in the feed, each retentate and each permeate gives a mass balance that tells you where the EVs went, which matters more than usual when a fraction is expected to be small.
What Is Still Open
There is no standard protocol. The studies above used different membrane materials, formats and pore ratings, and nominal ratings are not directly comparable between manufacturers, so a "0.05 µm" fraction from one system may not match another.
It is also unclear how closely size fractions correspond to biologically meaningful subtypes. The functional differences reported between size bands are encouraging,4 but a band defined by a membrane will always contain a mixture of origins.3 Whether size gating by TFF can deliver fractions consistent enough for a therapeutic specification, rather than a research comparison, has not yet been shown.
For now it is probably best seen as a scalable enrichment step: a practical way to ask whether larger and smaller EVs from the same cells behave differently, with the answer checked by characterisation rather than read off the cartridge label.
References
- Théry C, Witwer KW, Aikawa E, et al. Minimal information for studies of extracellular vesicles 2018 (MISEV2018): a position statement of the International Society for Extracellular Vesicles and update of the MISEV2014 guidelines. J Extracell Vesicles. 2018;7(1):1535750.
- Zhang H, Freitas D, Kim HS, et al. Identification of distinct nanoparticles and subsets of extracellular vesicles by asymmetric flow field-flow fractionation. Nat Cell Biol. 2018;20(3):332-343.
- Kowal J, Arras G, Colombo M, et al. Proteomic comparison defines novel markers to characterize heterogeneous populations of extracellular vesicle subtypes. Proc Natl Acad Sci USA. 2016;113(8):E968-E977.
- Su X, et al. Multi-module cassette tangential flow filtration enables high-throughput isolation and functional analysis of small extracellular vesicle subpopulations. Anal Chem. 2025. doi:10.1021/acs.analchem.5c02800.
- Heinemann ML, Ilmer M, Silva LP, et al. Benchtop isolation and characterization of functional exosomes by sequential filtration. J Chromatogr A. 2014;1371:125-135.
- Kim K, Park J, Jung JH, et al. Cyclic tangential flow filtration system for isolation of extracellular vesicles. APL Bioeng. 2021;5(1):016103.
- Bae JH, Lee CH, Jung D, et al. Extracellular vesicle isolation and counting system (EVics) based on simultaneous tandem tangential flow filtration and large field-of-view light scattering. J Extracell Vesicles. 2024;13(7):e12479.
- Scarborough T, Kawai-Harada Y, Brennan O, Chan C, Harada M, Walton SP. Tangential flow filtration for isolating exomeres and other nanoscale extracellular particles. Nanoscale. 2026. doi:10.1039/d6nr01765g.
- Hunter DG, Frisken BJ. Effect of extrusion pressure and lipid properties on the size and polydispersity of lipid vesicles. Biophys J. 1998;74(6):2996-3002.
- Onesti R, Giancaterino S, Masciolini L, Piunti C, Cimetta E, Boi C. Continuous tangential flow filtration to isolate extracellular vesicles from dairy by-products. Ind Eng Chem Res. 2025.
The EVlution™ TFF system runs polyethersulfone hollow fibre cartridges from 5 kDa to 1,000 kDa, so both stages of a size-gated run can be carried out on one system: a 1,000 kDa or 750 kDa cartridge for the first gate, then a 300 or 500 kDa cartridge to concentrate the smaller EVs that pass it. Our Exo-spin™ columns provide the size exclusion polishing step, and the NTA size profiling service can confirm the size distribution of each fraction. For background on the method itself, see our introduction to tangential flow filtration for EV research.
Learn more about cytogenetics and cell characterisation from Cell Guidance Systems:
