Tangential flow filtration for protein purification

Tangential flow filtration (TFF) is the workhorse of protein purification. Almost every purified protein, from a few micrograms of an enzyme on a bench to kilograms of a therapeutic antibody in a plant, passes through a TFF step at some point, usually to concentrate it and to exchange it into the buffer it will be stored or used in. This article looks at how TFF is used for proteins: the range of proteins it handles, how much diafiltration a given job needs, whether to run that diafiltration continuously or discontinuously, and how the membrane itself interacts with the protein. Antibodies get most of the attention, because they dominate both the biopharmaceutical pipeline and the hardest problems in protein TFF, but the same principles apply across the board.

Concentration and diafiltration

TFF does two things to a protein solution. In concentration, permeate is drawn off through the membrane and the protein, held back by the membrane, is left in a smaller volume. In diafiltration, fresh buffer is added to the retentate at the same rate as permeate leaves, so the volume stays constant while small molecules are washed out and the buffer is replaced. Nearly every protein process combines the two, which is why TFF in this setting is usually written as UF/DF, ultrafiltration and diafiltration.

The membrane is chosen to retain the protein completely and to pass the things you want to remove: salts, the old buffer, small-molecule additives, and sometimes smaller contaminating proteins. The driving force is the transmembrane pressure, and flow across the membrane surface sweeps away the layer of concentrated protein that would otherwise build up and choke the flux. We have covered the underlying principle and its history in an earlier article on the origins and applications of tangential flow filtration, and the mechanics of buffer exchange in more depth in our article on TFF diafiltration.

Which proteins, and which membrane cut-off

Proteins span a wide range of size, shape and behaviour, and TFF is used across all of it. At the small end are peptides and small recombinant proteins, insulin, cytokines and growth factors of a few kDa. In the middle sit most enzymes, serum albumin and antibody fragments. At the large end are intact antibodies at around 150 kDa, fusion proteins, and large multi-subunit assemblies. Each is retained by choosing a membrane whose nominal molecular weight cut-off (MWCO) is comfortably below the mass of the protein.

The usual guidance is to pick a cut-off three to five times smaller than the protein you want to keep. A tighter membrane gives more complete retention and less product loss, at the cost of lower flux and longer processing. A cut-off too close to the protein risks losing some of it through the membrane, and the risk grows with the volume of permeate generated, so it matters most in long diafiltrations. Shape and charge complicate the picture: an elongated or heavily glycosylated protein behaves as though it were larger than its mass suggests, and the apparent size of a protein shifts with pH and ionic strength. It is worth characterising a new feedstock rather than trusting the nominal mass alone.

These are starting points, not rules. A 30 kDa membrane is the default for an antibody, but some antibody processes run on a 50 kDa membrane for higher flux where a little passage can be tolerated, and others drop to 10 kDa to be safe with a fragment. The only reliable way to settle it is a small retention trial on the actual protein.

How much diafiltration is needed

Diafiltration is measured in diafiltration volumes, or diavolumes. One diavolume is a volume of permeate equal to the volume of retentate being held constant. The question of how much to apply has a clean answer for the simplest case and a messier one for everything else.

For a small molecule that passes freely through the membrane, run in constant-volume mode, the fraction remaining in the retentate falls off exponentially with the number of diavolumes. The concentration after N diavolumes is the starting concentration multiplied by e to the power of minus N. That gives the familiar figures below: roughly 95% removal after three diavolumes, better than 99% after five, and around 99.9% after seven. Most protein buffer exchanges are run to somewhere between five and ten diavolumes, which trades off completeness against the time, buffer and protein losses that each extra volume costs.

The exponential rule holds only when the solute passes the membrane completely and is not held up by anything. Two things break it. The first is partial retention: if the membrane holds back a quarter of the small molecule, the wash-out slows, and reaching 99.9% removal takes around nine diavolumes instead of seven. The larger the species you are trying to remove, the lower its passage and the more buffer it takes. The second is interaction with the protein, which matters most for antibodies and is dealt with below.

A practical point that saves a lot of buffer: concentrate before you diafilter. Because a diavolume is set by the retentate volume, halving that volume first halves the buffer each diavolume consumes. The usual sequence is to concentrate the protein, diafilter at the reduced volume, then make any final concentration adjustment in the new buffer. The exception is a protein that becomes too viscous or unstable when concentrated, where a gentler route is to diafilter at a higher volume and accept the extra buffer.

Continuous or discontinuous diafiltration

There are two ways to wash buffer through the retentate. Continuous diafiltration, also called constant-volume diafiltration, adds fresh buffer at the same rate as permeate is removed, so the retentate volume and the protein concentration stay fixed throughout. Discontinuous diafiltration, sometimes called sequential or batch diafiltration, works in cycles: dilute the retentate with a volume of buffer, reconcentrate back to the starting volume, and repeat.

For large-scale protein production (>10L), continuous diafiltration is widely used in biopharmaceutical manufacturing as it is more efficient, reaching a given level of small-molecule removal in fewer diavolumes than the discontinuous approach, so it uses less buffer and less time.  A constant hydrodynamic environment at the membrane also makes the run easier to scale.

Discontinuous diafiltration is more frequently used at small scale (up to a few litres). It needs no reservoir feeding buffer into the retentate at a matched rate, so it can be done with a single vessel, a membrane and a source of fresh buffer, which is why it is common in manual bench work and in stirred cells. If you are desalting a stable, well-behaved protein in a few millilitres, the difference in buffer consumption is small and the simplicity can win. For anything sensitive, large or destined for scale-up, continuous is the safer default.

Membrane type and protein interaction

The membrane is not an inert sieve. Proteins adsorb to membrane surfaces, and that adsorption drives both product loss and fouling, the gradual fall in flux as material accumulates on and in the membrane. The choice of membrane material is largely a choice about how much the protein will stick.

Two materials dominate ultrafiltration for proteins. Regenerated cellulose is naturally hydrophilic and binds less protein than any other common ultrafiltration membrane. It fouls less, cleans easily and recovers dilute proteins well, which makes it the default for low protein loadings and for feedstocks prone to fouling. Its limits are a narrower working pH range, around pH 2 to 13, and lower mechanical strength than the alternatives.

Polyethersulfone (PES) is more hydrophobic. It is mechanically stronger, gives high flux, and tolerates a wider pH range for processing and cleaning, up to the harsh caustic regimes used in validated manufacturing. The trade-off is that it adsorbs more protein than regenerated cellulose and tends to foul more, so it usually needs more aggressive cleaning to restore flux. But at higher protein loadings which are commonly used, even at lab scale, above roughly 20 g per square metre of membrane, PES is often preferred precisely because the adsorptive loss becomes small relative to the amount processed, and its robustness pays off.

Two further interactions are worth naming. Charge matters: a protein and the membrane, or a protein and a charged contaminant, can attract or repel each other depending on pH and ionic strength, which shifts both retention and the sieving of the species you are trying to remove. And some proteins foul far more than their size predicts. PEGylated proteins are a good example, where the attached polymer chains increase apparent size, hydrophobicity and the tendency to accumulate at the membrane, so flux is lower and fouling heavier than for the native protein.

Antibodies: the hard case

Antibodies are the most purified class of therapeutic protein, and they expose the limits of protein TFF. The UF/DF step at the end of an antibody process has to bring the drug substance to its final concentration and exchange it into the formulation buffer, and the targets have climbed steeply as subcutaneous dosing has replaced the drip. Intravenous products might sit at 20 g/L; a subcutaneous antibody delivered in a small injection volume can need 100 to 150 g/L, and some reach higher still.

At these concentrations the membrane stops being the bottleneck. Viscosity does. A concentrated antibody solution can become viscous enough that the flux collapses and the pumps struggle, and the practical ceiling on how far you can concentrate is set by the solution, not the filter. Formulators push back against this with viscosity-lowering excipients and by choosing a pH away from the antibody's isoelectric point, while process developers reach for cassettes with wider feed channels and for single-pass configurations that avoid recirculating an already-viscous retentate.

The second antibody-specific problem is that the final buffer does not come out matching the diafiltration buffer. At high protein concentration and low ionic strength, charged buffer species and the charged protein interact, so the retentate ends up with excipient levels and a pH offset from the buffer you were washing in. This is the Donnan effect, together with volume exclusion by the crowded protein, and it is strong enough that the pH can shift by a few tenths of a unit and excipient concentrations by a meaningful margin. It is the reason you cannot assume that running enough diavolumes of formulation buffer leaves the protein in formulation buffer. Developers now model these shifts, or adjust the diafiltration buffer composition to land on the right final target, rather than discovering the offset after the fact.

Two TFF variants were developed largely with antibodies in mind. High performance TFF, introduced by a Genentech team in 1997, operates in a carefully controlled region of the flux curve to separate proteins of similar size, something conventional TFF cannot do, and was demonstrated by separating an antibody from albumin. Single-pass TFF (SPTFF) strings membrane modules in series so the target concentration is reached in one pass with no recirculation loop; an early study concentrated bovine IgG from about 45 g/L to over 190 g/L in a single pass at more than 95% recovery, and the approach has since been used to debottleneck antibody manufacturing and to feed continuous processes.

Running protein TFF well

Much protein TFF at the bench is done by hand, with a peristaltic pump and a clamp on the retentate line, and no measurement of pressure or permeate flow. It works, and for a robust protein it is perfectly adequate. The cost is that when the flux falls you cannot tell whether the membrane is fouling or the retentate is simply getting more viscous, and a run done this way is hard to reproduce or to hand on to a larger system. Measuring transmembrane pressure and tracking permeate volume turns a diafiltration from a feel-based procedure into one you can specify in diavolumes and repeat.

The EVlution™ TFF system from Cell Guidance Systems uses hollow fibre cartridges in ten pore sizes, so a single set-up covers proteins, antibodies, viruses and extracellular vesicles, and we have seen users adopt it for protein concentration and buffer exchange well beyond its original vesicle work, often on the 10 kDa cartridge. The EVlution™ Pro instruments add monitoring of volume and pressure through the run on the same cartridges, which is what makes a diafiltration measured in diavolumes rather than in clamp turns. Choosing a cartridge for a given protein and volume is covered in our guide to choosing the right TFF filter.

Frequently asked questions

How many diafiltration volumes are needed to buffer-exchange a protein?
For a small molecule that passes freely through the membrane, five to seven diavolumes remove more than 99%, and seven remove about 99.9%, because in constant-volume diafiltration the fraction of a freely permeating solute remaining falls exponentially with the number of diavolumes. Species that are partly retained, or that bind the protein, need more: a species that is 25% retained takes around nine volumes to reach 99.9%. Most protein buffer exchanges run to between five and ten diavolumes.

Is continuous or discontinuous diafiltration better for proteins?
Continuous, constant-volume diafiltration is generally preferred. It reaches a given level of removal in fewer diavolumes than the discontinuous dilute-and-reconcentrate approach, and it holds the protein at a steady concentration, which is gentler and keeps viscosity and flux stable. Discontinuous diafiltration is simpler at small scale and needs no matched buffer feed, but it exposes the protein to repeated concentration swings.

Which ultrafiltration membrane is best for protein purification?
Regenerated cellulose binds very little protein and fouls less, which suits dilute proteins and fouling feedstocks; polyethersulfone is stronger and tolerates a wider pH range but adsorbs more protein and needs harsher cleaning. For the cut-off, choose a nominal molecular weight limit three to five times below the protein, so a 30 kDa membrane is typical for a 150 kDa antibody.

References

1. Pall Corporation. Diafiltration: a fast, efficient method for desalting or buffer exchange of biological samples. Scientific and Technical Report. 2008.
2. van Reis R, Zydney A. Bioprocess membrane technology. J Membr Sci. 2007;297(1-2):16-50.
3. van Reis R, Gadam S, Frautschy LN, et al. High performance tangential flow filtration. Biotechnol Bioeng. 1997;56(1):71-82.
4. Casey C, Gallos T, Alekseev Y, Ayturk E, Pearl S. Protein concentration with single-pass tangential flow filtration (SPTFF). J Membr Sci. 2011;384(1-2):82-88.
5. Dizon-Maspat J, Bourret J, D'Agostini A, Li F. Single pass tangential flow filtration to debottleneck downstream processing for therapeutic antibody production. Biotechnol Bioeng. 2012;109(4):962-970.
6. Molek JR, Zydney AL. Ultrafiltration of PEGylated proteins: fouling and concentration polarization effects. J Membr Sci. 2008;319(1-2):206-213.
7. Steele A, Arias J. Accounting for the Donnan effect in diafiltration optimization for high-concentration UFDF applications. BioProcess International. 2014;12(1):50-54.
8. Zhao Y, Moomaw JF, Thyagarajapuram NR, et al. A mechanistic model to account for the Donnan and volume exclusion effects in ultrafiltration/diafiltration process of protein formulations. Biotechnol Prog. 2021;37(2):e3106.
9. Jabra MG, Yehl CJ, Zydney AL. Strategies for high concentration drug substance manufacturing to facilitate subcutaneous administration: a review. Biotechnol Bioeng. 2020;117(9):2920-2932.
10. Yang Y, Velayudhan A, Thornhill NF, Farid SS. Multi-criteria manufacturability indices for ranking high-concentration monoclonal antibody formulations. Biotechnol Bioeng. 2017;114(9):2043-2056.

Cell Guidance Systems supplies EVlution™ TFF systems and hollow fibre cartridges for concentrating and buffer-exchanging proteins, antibodies, viruses and extracellular vesicles, alongside a range of EV and exosome services that can help accelerate your research with confidence.

Learn more about powerful technologies that are enabling research: