Why do flux and pressure change during tangential flow filtration?

Run a TFF concentration and the same thing happens every time. Flux starts high, drops away over the first few minutes, then levels off somewhere well below where it began. Feed pressure drifts upward as the retentate concentrates.

Most people accept this as the way the technique behaves and leave it at that. It repays a closer look, because several different processes are running at once and they do not have the same consequences. Some of the lost flux comes back the moment you flush the system. Some comes back when you clean the cartridge. Some does not come back at all, and working out which is which is what tells you whether a cartridge is still worth using.

This article covers the mechanisms behind flux decline, what a rising pressure reading is actually telling you, the recurring question of vertical versus horizontal mounting, and permeability data from three consecutive concentration and cleaning cycles on our 94 cm² cartridges. That last part shows a pattern most users will recognise but few have bothered to quantify.

The Resistance Picture: What Flux Actually Depends On

Flux through a TFF membrane is usefully described by a resistance-in-series model:

J = TMP / [ η (R_m + R_p + R_a + R_c) ]

Where J is flux, TMP is transmembrane pressure, η is permeate viscosity, R_m is the intrinsic resistance of the clean membrane, R_p is the resistance of the concentration polarisation layer, R_a is the resistance from adsorbed material, and R_c is the resistance of any deposited cake or gel layer.

Written out this way the diagnosis becomes tractable. R_m is fixed for a given cartridge. R_p only exists while cross-flow and TMP are applied, and vanishes on flushing. R_c is deposited material, usually removable by cleaning. R_a is adsorbed material bound to the polymer surface, and it is the term that decides whether a cartridge is genuinely reusable over many runs. Anything that changes flux is changing one of these four, or changing viscosity.

Why Flux Falls During a Run

Concentration polarisation

Polarisation is the fastest effect and the largest one in most runs. Solvent passes through the membrane; retained material does not. Retained material therefore accumulates in a thin boundary layer at the membrane surface, reaching a local concentration that can be one to two orders of magnitude higher than in the bulk retentate. That layer adds hydraulic resistance and generates an osmotic back-pressure that offsets part of the applied TMP.

Polarisation establishes itself within seconds of starting a run and reaches a steady state within a few minutes. It is reversible: if you stop the run and flush the cartridge, it is gone. Its magnitude is set by the balance between convective transport toward the membrane (driven by flux) and back-transport away from it (driven by cross-flow shear and diffusion). This is why increasing cross-flow velocity raises flux, and why above a certain TMP further pressure delivers no additional flux at all. In that pressure-independent regime, extra TMP simply compresses the polarisation layer, and the resistance rises in step with the driving force.

Adsorption

Proteins adsorb onto polyethersulphone and modified PES surfaces spontaneously, driven by hydrophobic interaction and by the entropy gained when ordered hydration water is displaced. It starts on first contact with the feed, well before there is any measurable change in flux, and it happens on the internal pore walls as much as on the lumen surface. Pore constriction affects flow as the fourth power of radius, so even a molecular monolayer inside a 20 nm pore costs a surprising amount of permeability.

Adsorption is also self-limiting, and that turns out to matter a great deal for how cartridges behave over repeated use. The high-energy binding sites go first. Once the accessible polymer surface is covered, protein arriving at the membrane meets an already-conditioned surface rather than bare PES, and it binds far more weakly. Weakly bound material comes off in an alkaline clean. The first layer largely does not. We come back to this when we look at the data.

Pore blocking and cake formation

Larger species block pores outright, partially occlude them, or accumulate as a cake layer on the membrane surface. The classical treatment separates complete blocking, standard (internal) blocking, intermediate blocking and cake filtration, each with a characteristic flux-decline profile. A real feed produces a mixture of all four: aggregates and debris block, proteins adsorb internally, and a compressible cake builds on top of everything else.

Rising retentate concentration

During a concentration run the retentate becomes progressively more concentrated, and both viscosity and polarisation severity increase with it. A 20-fold concentration of a 1 mg/ml protein solution ends at 20 mg/ml, at which point viscosity is measurably above that of buffer and the layer at the membrane wall is substantially more resistive than it was at the start.

Some of the flux decline seen in any concentration run is therefore not fouling at all, but a straightforward consequence of concentrating the sample. It is worth allowing for before concluding that a cartridge is deteriorating.

Why Pressure Rises During a Run

Pressure is where most of the diagnostic confusion arises, because two quite different pressures are in play and they mean different things.

Feed to retentate pressure drop (ΔP) is the axial pressure loss along the fibre lumen. It is set by flow rate, fibre internal diameter, fibre length and fluid viscosity, following Hagen-Poiseuille behaviour. It has almost nothing to do with the membrane itself.

Transmembrane pressure (TMP) is the net pressure across the membrane wall, conventionally the average of feed and retentate pressures minus permeate pressure. It is the driving force for flux.

A rising ΔP at constant pump speed means the lumen is becoming harder to flow through. Three causes account for nearly all cases. Viscosity is the first: as retentate concentrates, ΔP rises in proportion, which is expected and is not a fault. The second is partial occlusion of individual fibres by aggregates, cell debris or precipitated material, forcing the same volumetric flow through fewer parallel channels. The third is air. Gas trapped in the cartridge, whether from incomplete priming or from foam generated at a leaking fitting or a vortexing reservoir, increases apparent resistance dramatically and produces erratic readings.

A rising TMP is a different signal. Under constant-flux operation, TMP must rise to compensate for increasing membrane resistance, and TMP creep is then a direct measure of fouling. Under constant pressure, which is more usual at research scale, TMP will not rise on its own and flux falls instead. In practice many users see both at once, because a peristaltic pump at fixed speed sits somewhere between the two regimes.

One rule of thumb is worth holding onto. Fouling is gradual. A sharp step change in pressure is nearly always mechanical, so look for a kinked line, a clamp that was never fully opened, a blocked fibre bundle or a restricted permeate line before concluding the membrane is at fault.

Cartridge Orientation: Vertical or Horizontal?

This one comes up regularly. The assumption behind it is usually that vertical mounting must be better, because in a horizontal fibre retained material would settle along the lower wall and air bubbles would collect along the upper one. We have not seen any difference in performance between the two. The physics supports that, though the two halves of the assumption fail for quite different reasons, and the air half deserves more attention than it usually gets.

Take the hydrostatic argument first. For a hollow fibre cartridge with a housing diameter of roughly 1 to 2 cm, the hydrostatic head between the uppermost and lowermost fibre in a horizontal orientation is around 0.001 to 0.002 bar. Typical operating TMP is 0.5 to 1.5 bar. The gravitational contribution is three orders of magnitude below the driving force, so no measurable difference in flux distribution should be expected.

Sedimentation within the lumen is the second concern, and it fails on timescale. Stokes settling velocity for a 100 nm extracellular vesicle at a density of approximately 1.15 g/ml in aqueous buffer is on the order of 1 nm per second. Over a two hour run that amounts to less than 10 µm of vertical displacement, against a lumen internal diameter of several hundred micrometres. For proteins such as BSA the settling velocity is smaller still, and Brownian motion swamps it entirely. Meanwhile cross-flow at 10 to 30 cm per second is moving the fluid along the fibre roughly a hundred million times faster than gravity moves anything across it. Sedimentation does not get the chance to happen.

Air is the one thing gravity acts on strongly

Gas is a different case, and it is worth taking seriously rather than waving away. Buoyancy scales with the density difference between the dispersed phase and the fluid, and with the square of diameter. For a vesicle that density difference is about 150 kg/m3. For an air bubble it is close to 1000 kg/m3, and bubbles are far larger. A 100 µm bubble rises at roughly 5 mm per second. Gravity really does move gas around inside a TFF system on the timescale of a run, in a way it does not move anything else in the feed.

What stops this becoming an orientation-dependent performance problem is confinement. Inside a fibre lumen of a few hundred micrometres, surface tension dominates buoyancy by around two orders of magnitude, so a bubble in the lumen does not float to the top of the channel. It is held by capillary forces and moved by whatever the flow does to it. Cross-flow is twenty to fifty times faster than the bubble would rise even in open fluid, so gas that enters a fibre is swept through and out to the retentate whichever way the cartridge is mounted.

The balance reverses in the header volumes and the shell space, which are centimetre-scale rather than micrometre-scale. Buoyancy wins comfortably there, and gas does collect at the highest point.

This is where the real orientation difference lies, and it is geometric. In a vertically mounted cartridge the highest point is a single small end header, so trapped gas gathers in one compact pocket that is easy to purge and blanks very little membrane area. Mounted horizontally, the highest point is a line running the full length of the housing, and the same volume of gas spreads as a thin layer over a substantial fraction of the fibre bundle. Vertical mounting is more forgiving of a rushed prime, which is a handling advantage rather than a difference in what the membrane can do.

What trapped air actually does, and how to recognise it

A gas blanket does not transmit water. Any membrane area sitting under a bubble is out of service, so a cartridge holding 10% of its area under trapped gas reports approximately 10% lower permeability. On a water flux measurement that is indistinguishable from fouling, which is the trap. Cleaning will not shift trapped air and re-priming will not shift fouling, so if a permeability figure disappoints, re-prime and re-measure before reaching for the NaOH.

The other signatures are behavioural. Trapped or entrained air produces erratic, oscillating pressure readings rather than the smooth traces of a properly filled system, and it gives pulsatile permeate flow. It also degrades reproducibility between technical repeats of a water flux measurement, because the gas volume shifts slightly between readings. Where a permeability log shows unusually wide triplicate spread on an otherwise healthy cartridge, air is a more likely explanation than heterogeneous fouling.

There is a product quality dimension as well, and it is easy to overlook. Gas-liquid interfaces are strongly denaturing. Proteins adsorb and unfold at an air-water interface, and vesicles exposed to the same interface can aggregate or lose integrity. Persistent bubbles recirculating through a cartridge are an active source of aggregate formation, and those aggregates then go on to block pores. Very few problems in TFF damage both the membrane and the material being processed; air is one of them.

Where the air comes from

In practice most air enters by one of a handful of routes. Incomplete priming, particularly where the permeate line was left clamped and the shell side never filled. A reservoir allowed to run low, so the feed line draws gas once the level falls below the inlet. Vortexing in a vigorously stirred reservoir, which entrains gas continuously without any obvious fault. A loose fitting on the suction side of the pump, which is the one most often missed, since a suction-side connection draws air in without letting any liquid out and so gives no drip to alert you. And cold buffer taken straight from 4°C storage and warmed on the bench, which releases dissolved gas as it equilibrates.

Priming with the permeate line open, running buffer through at low flow before applying pressure, and letting buffer reach room temperature first will deal with nearly all of these. End-of-run recovery from the fibre dead volume is also a little more predictable when the cartridge drains in a consistent direction, which is a further, smaller argument for fixing an orientation and sticking with it.

So: either orientation is acceptable, and the choice should come down to what makes priming, tubing routing and cartridge removal easiest on your bench. Orientation only matters to the extent that it changes how easily air is purged, and air is the variable actually worth managing. Whichever you pick, use it for every run, so that any performance change you see can be attributed to the membrane rather than to a change in setup.

Real Data: Three Concentration and Cleaning Cycles on a 94 cm² Cartridge

The dataset below was generated on our 94 cm² hollow fibre cartridges. Each run was a 20-fold concentration of 500 ml of 1 mg/ml bovine serum albumin (BSA) in PBS down to approximately 25 ml. Between runs the cartridge was cleaned with 0.5 M NaOH at 40°C and 250 rpm for 40 minutes, followed by a further 20 minutes under the same conditions with 0.3 bar of back pressure applied to drive cleaning solution through the membrane wall.

Clean water permeability, in LMH per bar, was measured in triplicate at each stage. This is the single most informative measurement in TFF. It is made on buffer alone, so polarisation has been flushed away and what remains reflects only the intrinsic membrane resistance plus whatever material is genuinely bound to or deposited in it.

Reading the Data: Why Flux Drops, Then Holds

The first run does most of the damage

The biggest single change in the whole series happens on first use. Permeability falls from 124.8 to 66.9 LMH/bar, a loss of 46%. Alkaline cleaning recovers 20 percentage points of that and takes the cartridge back to 74%. The remaining 26 points are not recovered by cleaning, and they never are. They are the conditioning layer of adsorbed protein on the highest-affinity binding sites of the membrane polymer, together with whatever has lodged inside the pore structure where a surface clean cannot reach it.

Treat the virgin figure of 124.8 LMH/bar as a manufacturing specification. The number that matters day to day is the conditioned baseline you get after the first run and clean.

Cycles two and three show no further decline

If fouling were cumulative, each successive cycle would be worse than the last. It is not. Post-run values across the three cycles are 54%, 61% and 53%, a spread of eight points with no direction to it, which is about what you would expect from ordinary run-to-run variation in feed handling and measurement. Post-clean values are 74%, 75% and 83%.

The self-limiting adsorption described earlier is the reason. Once the accessible polymer surface is covered, the membrane presents a protein-conditioned face to every subsequent feed. Protein arriving on that face binds to other protein rather than to PES, and protein-protein association is much weaker than the original hydrophobic protein-polymer interaction. Alkaline cleaning hydrolyses and solubilises that weakly bound layer efficiently, which is why every clean returns the cartridge to a similar place. What the run takes away, the clean puts back.

The third clean, and what we cannot say about it

Post-clean 3 came in at 83%, six to eight points above the first two cleans. There are at least two candidate explanations.

The first is a cumulative benefit from the 0.3 bar back pressure step. Forcing cleaning solution through the membrane wall in the reverse direction reaches material that cross-flow leaves undisturbed, and repeated application might be progressively clearing pore-internal deposits. The second is simply temperature. Water viscosity falls by roughly 2 to 3% per degree Celsius, so buffer a few degrees warmer than in the previous session reads high, and a difference of this size can be accounted for by temperature alone.

We cannot separate the two from three cycles of data. A longer series with temperature-corrected measurements would be needed to say whether post-clean permeability genuinely improves with repeated back-pressure cleaning or whether this is scatter. We have left it open rather than pick the more interesting answer.

Measurement noise is informative too

There is a pattern in the triplicate spread worth noticing. On the virgin cartridge the three readings span 0.9% of the mean. On post-run measurements the spread is 4.5 to 10%. On post-clean measurements it sits between 1 and 6%.

A pristine membrane is a uniform hydraulic structure and gives highly reproducible readings. A fouled membrane is heterogeneous, with some fibres and some pore populations more obstructed than others, and small variations in how the measurement is made produce correspondingly larger variation in the result. Rising variability between technical repeats is worth treating as an early fouling indicator in its own right, and it is a reason not to lean too hard on any single permeability reading.

One caveat on our own numbers. The two widest post-run spreads, 8.4% on run 1 and 10.0% on run 3, are put down to heterogeneous fouling above, but residual air in the cartridge would give the same signature. Re-priming between technical repeats would separate the two. We did not do that here.

Note on the challenge used

A 20-fold concentration of BSA to a final 20 mg/ml is a deliberately aggressive protein challenge. BSA is small, abundant and strongly surface-active, and it is close to a worst case for adsorptive fouling of PES. Most EV workflows put a lower absolute protein load on the membrane, but a more heterogeneous one, with lipoproteins, aggregates and cell debris in the mix. The qualitative behaviour is the same. The balance shifts, though: with EV feeds a larger share of the total loss tends to be cake-type and therefore recoverable, provided pre-clarification is adequate.

Practical Monitoring: Building a Permeability Log

The dataset above doubles as a template for a monitoring protocol, and the discipline it needs is not onerous.

Measure clean water permeability on every new cartridge before its first use, in triplicate, and record the buffer temperature. Measure again after the first run and after the first clean. Those two numbers define the conditioned baseline for that specific cartridge. From then on, measure post-clean permeability after every cleaning cycle and track it against the conditioned baseline rather than against the virgin value. Correct for temperature if the measurements are not being made in a thermally controlled environment.

Define the retirement criterion in advance and apply it to the post-clean value. A common bioprocess approach is to retire the device when post-clean permeability falls below 60% of the value achieved at the first clean, or when it declines by more than about 10% over consecutive cycles. A cartridge behaving like the one in our data, returning to 74 to 83% after every clean with no downward trend, has plenty of working life left.

Feed pressure at fixed pump speed on clean buffer is a useful second indicator, because it catches a different failure mode. Permeability tracks the condition of the membrane. Buffer-only feed pressure tracks whether the lumens are still open. A cartridge with acceptable permeability but a creeping buffer-only feed pressure has partially occluded fibres, and its effective area is below the label value even though the remaining fibres are working normally.

What This Means for EV Workflows

Three things follow for extracellular vesicle work.

Plan around the conditioned baseline rather than virgin performance. Process development runs done on a brand new cartridge will produce flux figures that no subsequent run reproduces, and parameters set that way will not hold. Establishing the protocol on a conditioned cartridge gives numbers you can rely on.

Pre-clarification is the highest-leverage intervention available. Adsorptive fouling is largely unavoidable and self-limiting, but pore blocking by aggregates and debris is neither. Getting that material out before it reaches the membrane, by 0.22 or 0.45 µm filtration or low-speed centrifugation, protects the part of the cartridge's performance that would otherwise be lost permanently.

Finally, be careful about blaming the cartridge for a drop in EV recovery without measuring permeability. Lower particle yield can just as easily come from aggregation in the retentate, a change in the conditioned medium, or loss of smaller vesicles through the membrane. A permeability measurement separates the membrane variable from the biology and takes a few minutes.

Summary

Flux decline during a TFF run has several causes with different consequences. Concentration polarisation is immediate, large and completely reversible. Adsorption is rapid, self-limiting and partly irreversible. Pore blocking is largely preventable through pre-clarification. Cake formation is progressive and mostly removable by alkaline cleaning. Rising retentate viscosity accounts for a real share of late-run flux decline in any concentration process and is not a fault.

Rising pressure has to be read against which pressure is rising. A gradual increase in feed to retentate pressure drop tracking the concentration factor is expected. A sudden step change is mechanical. Erratic pressure indicates air. TMP creep under constant-flux operation is the cleanest available measure of fouling.

Orientation, vertical or horizontal, makes no measurable difference to performance in our hands, and the physics explains why: hydrostatic effects across a cartridge are three orders of magnitude below operating TMP, and sedimentation velocities are negligible against cross-flow. Air behaves differently, and vertical mounting is more forgiving of a rushed prime, but that is a handling consideration. Choose whichever orientation suits your bench and keep to it.

The permeability data from our 94 cm² cartridges shows the pattern to expect. A large first-use loss to around 54% of virgin permeability, recovery to roughly 74% after the first clean, then stable cycling between approximately 53 to 61% post-run and 74 to 83% post-clean across three consecutive cycles with no cumulative decline. Flux drops and then holds because the irreversible component of fouling saturates once the membrane surface is conditioned, after which each run deposits only the reversible material that the next clean takes off again.

Further Reading

• Van Reis R, Zydney A. Bioprocess membrane technology. J Membr Sci. 2007;297(1-2):16-50.
• Field RW, Wu D, Howell JA, Gupta BB. Critical flux concept for microfiltration fouling. J Membr Sci. 1995;100(3):259-272.
• Bacchin P, Aimar P, Field RW. Critical and sustainable fluxes: theory, experiments and applications. J Membr Sci. 2006;281(1-2):42-69.
• Hermia J. Constant pressure blocking filtration laws: application to power-law non-Newtonian fluids. Trans Inst Chem Eng. 1982;60:183-187.
• Marshall AD, Munro PA, Trägårdh G. The effect of protein fouling in microfiltration and ultrafiltration on permeate flux, protein retention and selectivity: a literature review. Desalination. 1993;91(1):65-108.
• Rabe M, Verdes D, Seeger S. Understanding protein adsorption phenomena at solid surfaces. Adv Colloid Interface Sci. 2011;162(1-2):87-106.
• Nordin JZ, et al. Ultrafiltration with size-exclusion liquid chromatography for high yield isolation of extracellular vesicles preserving intact biophysical and functional properties. Nanomedicine. 2015;11(4):879-883.

The EVlution TFF system is designed for reproducible concentration and diafiltration of extracellular vesicles at research scale, with the pressure monitoring needed to track cartridge condition run to run. For groups who prefer processed material without in-house instrumentation, a dedicated EV processing service is also available, and NTA size profiling can be used to verify particle recovery and size distribution after TFF.

Learn more about EVlution TFF and EV purification tools from Cell Guidance Systems: