Filtration through a 0.22 µm membrane is one of the most routine operations at a peptide bench and one of the least examined. It is quick, it looks decisive, and the filtrate is visibly clearer than what went in. That combination makes the step feel free. It is not. A membrane is a large area of engineered polymer with an extensive internal surface, and a peptide in dilute solution is a molecule that finds surfaces. This guide covers what the rating claims, where peptide is lost, what filtration does not remove, what it can add, and how the surviving particles are counted.
What a 0.22 µm rating claims
The rating is a retention performance claim tied to a defined bacterial challenge, not a geometric statement that every pore measures 0.22 µm. The distinction matters because the challenge conditions are specific.
Sundaram and colleagues challenged twenty-five integral 0.2 and 0.22 µm rated filter cartridges — eight filter types from four manufacturers, all claimed to retain Brevundimonas diminuta at 10⁷ CFU/cm² — with bacteria taken from a natural water source. They observed consistent penetration across every cartridge over the 18–24 hour challenge period, at challenge levels of about 10¹–10⁴ CFU/cm², and reported that the penetration was qualitatively independent of membrane type and manufacturer [1]. In a companion study whose penetrating challenge organism was a laboratory strain rather than natural water bacteria, thirty-two 0.2/0.22 µm rated cartridges spanning nine filter types were all penetrated, with log titer reductions between 3.5 and 7.7, while B. diminuta was fully retained by nineteen of the twenty filters challenged with both organisms [2].
That second result carries a correction worth knowing. The strain used, ATCC 700892, was later shown to be phylogenetically linked to the genus Curvibacter rather than to Hydrogenophaga pseudoflava, the name under which it had been reported. When genuine H. pseudoflava ATCC 33668 was challenged in cartridge format, it was completely retained by both 0.2 µm and 0.1 µm rated filters; retention of Curvibacter sp. ATCC 700892 varied with filter material, and only 0.1 µm rated filters retained it consistently [3]. The organism attribution changed; the underlying lesson did not. A rating is a claim about a named organism under named conditions.
How often does a filter validation test fail? Folmsbee analyzed roughly 1,162 individual filtrations to identify drivers of failure in low-surface-tension fluids, examining bacterial load, load rate, volume throughput and maximum flux. That analysis reports that approximately 97% of filter validation tests demonstrate absolute retention, making validation failure rare; the 27% failure rate within the assembled data set reflects a set deliberately weighted with penetration events and is not a general failure rate [4]. Sterilizing filtration usually works, the conditions of the fluid matter, and "usually" is not "always."
None of this describes a syringe filter pushed by hand at a bench: not integrity-tested per unit, not a validated sterilizing assembly. Treating a 0.22 µm syringe filter as a bioburden-reduction step is defensible; treating it as a sterility guarantee is not.
Where the peptide goes
The most direct loss mechanism is adsorption, and the relevant surface is not just the face of the membrane. Habibi and colleagues quantified fouling of a microfiltration membrane by two biomolecules — bovine serum albumin and glutathione, a tripeptide — using ATR-FTIR to measure accumulated mass directly rather than inferring it from flux. The two behaved differently. Glutathione adsorbed throughout the porous medium, penetrating the pores, altering the membrane's hydrophilicity at low concentration and constricting pores at high concentration. BSA instead built a dense, irreversible layer within the first minutes of filtration, containing material equivalent to more than 45 monolayers and resistant to both water rinsing and NaOH cleaning [5]. That was one tripeptide on one membrane, but it suggests the surface available to a small peptide may extend into the pore network rather than stopping at the membrane's apparent area, by an amount the study did not quantify.
Adsorption is also charge-dependent. Martínez and colleagues measured γ-globulin fouling of a 0.1 µm alumina membrane across concentrations and pH values, fitted the concentration dependence to a Freundlich heterogeneous isotherm, and observed a sharp maximum in adsorbed mass at the point of zero electrostatic force between solute and membrane [6]. That maximum was observed in their particular solute–membrane system; the shape it suggests — adsorption peaking where solute and surface charge cancel, not at any fixed pH — is a reason to record the pH of a filtration alongside its result.
The mechanism is not always monomer sticking to polymer. Maa and Hsu investigated severe membrane fouling during 0.22 µm sterile filtration of recombinant human growth hormone and tested four candidate explanations. Pore constriction and tortuosity played only a limited role. Nonspecific adsorption appeared insignificant, judged by comparing filters of differing protein-binding tendency. Shear stress during filtration was small, and material deliberately sheared in a high-speed rotating device did not foul worse. What dynamic light scattering showed instead was a trace amount of aggregate always present in equilibrium with the hydrophobic air–water and membrane–water interfaces; those aggregates adsorbed preferentially and, the authors proposed, generated further aggregate through the monomer–aggregate equilibrium until the membrane fouled [7]. In that system the filter was not removing a fixed impurity so much as interacting with a population that kept re-forming.
Adsorption is strongly sequence-dependent, which is why no general recovery figure exists. Siebels and colleagues built an assay that quantifies surface adsorption using a reference mixture of thousands of tryptic peptides and differential LC-MS/MS. A significant number of hydrophobic peptides adsorbed to conventional polypropylene vials; only a few adsorbed to polypropylene sold as low-protein-binding; and the largest number adsorbed to glass, driven by electrostatic interactions [8]. Those measurements were made on containers rather than membranes, but they establish two variables that govern loss anywhere in a workflow: the chemistry of the surface and the chemistry of the individual peptide.
Membrane chemistry and what it brings with it
The membranes in common use — polyethersulfone (PES), polyvinylidene fluoride (PVDF), cellulose acetate and regenerated cellulose, polyamides, PTFE, mixed cellulose esters, polycarbonate — differ in more than binding. Kushwah and colleagues profiled extractables and leachables from five polymeric filter membranes by LC-MS and GC-MS, then spiked the recovered leachable stocks into formulations of a recombinant protein at two pH values. Polyamide, polycarbonate and PES membranes yielded higher extractables and leachables content than PTFE and regenerated cellulose. Extractables and leachables from regenerated cellulose and PES were surface-active; those from polyamide and regenerated cellulose significantly altered particle size and structure, producing aggregation of the test protein. Bisphenol A was identified among the extractables and leachables from polycarbonate filters [9]. The authors present this as a qualitative rank ordering for membrane selection, not a per-unit specification.
"Low protein binding" is a supplier's descriptive label, not a standardized specification, and it has not been tested against your peptide. It is a reasonable default; it is not a measurement.
A worked example: why recovery depends on volume
Adsorption onto a finite surface can saturate. If it does, the mass lost to a filter is roughly fixed, and the fraction lost falls as more solution passes through. Percentages quoted without volumes are therefore uninterpretable.
Take assumed numbers. You filter 1.00 mL of peptide at 1.00 mg/mL and the filtrate assays at 0.86 mg/mL. Mass loaded is 1.00 mg, mass recovered is 0.86 mg, so 0.14 mg was retained. The loss is 0.14 mg divided by the 1.00 mg loaded, which is 14.0%. Now assume that 0.14 mg is a saturating property of that filter and does not change. Filter 5.00 mL of the same 1.00 mg/mL solution: 5.00 mg loaded, 5.00 − 0.14 = 4.86 mg recovered. The loss is 0.14 divided by 5.00, which is 2.8%, and recovery is 97.2%. The absolute loss was identical in both runs. The percentage fell fivefold only because the denominator grew fivefold.
Real recovery data has this shape. Zheng and colleagues measured recovery of three perfluorinated acids across six syringe filters and found recovery rose with filtration volume — 80 ± 28% at 1 mL, 85 ± 21% at 5 mL, 90 ± 18% at 10 mL — and also differed sharply by membrane, with mixed cellulose ester (0.20 µm, 25 mm) giving 97 ± 11% at 1 µg/L against 61 ± 37% for PTFE (0.45 µm, 13 mm), and by analyte, with PFOS recovering at 55 ± 25% where PFOA reached 96 ± 12% and PFBA 95 ± 8% [10]. Those analytes are not peptides, and the authors did not model the effect as fixed-mass saturation; the study is cited for the shape of the dependence on membrane, volume and analyte identity.
Two consequences follow. A recovery figure measured on a 1 mL trial does not transfer to a 10 mL run. And within a single run, the earliest filtrate is the most depleted — the reasoning behind discarding the first fraction.
What filtration does not remove
Anything dissolved passes. Salts, the trifluoroacetate or acetate counterion, residual synthesis solvents, scavengers and small-molecule impurities are all far below the pore rating and leave at the concentration they entered. Endotoxin is likewise not removed by a 0.22 µm membrane; it is a separate question.
Soluble aggregates and oligomers smaller than the rating also pass. Worse, a filter cannot fix the composition of a population that is at equilibrium. Maa and Hsu reported trace aggregate in continual equilibrium with monomer and with the hydrophobic interface in their rhGH system [7]; they did not test post-filtration re-equilibration, so the inference that removing part of an equilibrium population does not permanently change it is reasoning from their observation, not their finding.
Organisms other than the validation strain can pass: naturally occurring waterborne bacteria penetrated every cartridge tested [1], and retention of a given strain varied with filter material [2, 3]. Those studies do not establish why. And filtration reverses nothing chemical. Hydrolysis, oxidation and deamidation that have already occurred produce soluble species that a membrane cannot distinguish from intact peptide.
What filtration can add
The membrane itself is a source of extractables and leachables [9]; housings and connectors are further polymer contact surfaces, not covered by that study. Filters may also shed particles: it has been reported that protein–filter interaction and particle shedding from filter materials can contribute to aggregation under routine handling stresses, and Winter and colleagues designed a study around that observation, measuring particle formation at the visible, subvisible and submicron scales, along with structural change, in a glycoprotein formulation after both constant and interrupted "stop and go" filtration through 0.22 µm PVDF, PES and cellulose acetate membranes [11].
Filtration also creates interface. Fresh membrane–water contact area is generated as the sample passes, and pushing the last of a sample through with air introduces an air–water interface — the same class of hydrophobic interface at which Maa and Hsu found aggregate concentrating [7].
Measuring what is left: subvisible particles
Subvisible particle counts are reported by two main method families — light obscuration (LO) and flow imaging microscopy (FIM) — and what either one reports depends on more than the sample.
The two families also disagree with each other. Shibata and colleagues measured insoluble particles in several commercial protein products by both. Flow imaging counts were much higher, with the size of the gap depending on the product and on particle features; some products contained transparent, elongated particles that could escape LO detection entirely. In one product the FI count of particles ≥10 µm exceeded the 6,000-per-container criterion of the compendial particulate matter test — a criterion written against LO — which is why the authors urge caution in setting acceptance criteria using FI [13].
The dependence runs deeper than method family. Matter and colleagues showed that LO sizing accuracy depends strongly on the reference material used for calibration: substituting lower-refractive-index silica beads in sucrose for the usual polystyrene beads in water changed the result, and flow imaging microscopy suffered the same artifact to a smaller degree. Sizing and counting accuracy varied between sensors and between instrument suppliers, and the authors concluded that LO and FIM are not truly orthogonal analytical methods [12]. A subvisible particle count is a number produced by an instrument under a calibration, not a bare property of the sample. Record the instrument and calibration, or the number cannot be compared with anything.
The units also invite arithmetic errors. Compendial criteria are stated per container while instruments report per milliliter. A sample counted at 2,800 particles/mL at ≥10 µm in a 2.0 mL container gives 5,600 per container, under the 6,000 threshold. The same material counted at 3,200 particles/mL gives 6,400 per container, over it. That is a 14.3% increase — 400 divided by the lower count of 2,800 — and it straddles the limit. Small method-to-method differences in per-milliliter counts become pass/fail differences after multiplication.
What is not established
Almost none of the literature above concerns small synthetic peptides handled at the 1–10 mg scale. It concerns proteins and glycoproteins, industrial-scale dead-end filtration, and validated sterilizing assemblies. Synthetic peptides differ in ways that plausibly matter — no tertiary structure to unfold at an interface, much smaller hydrodynamic size, and solutions that often contain organic modifier or a preservative. Extrapolating protein filtration behavior to a 10-residue peptide is an assumption, not an inference.
There is no published, peptide-by-peptide table of recovery across membrane chemistries, and the variable list — sequence, concentration, buffer, pH, membrane, area, volume — suggests there never will be. The only recovery number that means anything for a given material is the one measured on that material: split a reconstituted stock, filter one portion, assay both by the same method, and compute recovery against the unfiltered portion as the denominator. That experiment is standard practice rather than a published protocol, as are pre-wetting a membrane, discarding the first fraction, choosing the smallest filter diameter that will pass the volume, using steady rather than forceful plunger pressure, and using each filter once.
One final point of consistency. Filtration is not a remedy for a cloudy reconstituted solution. Persistent cloudiness after reconstitution points to incomplete solubilization or an unsuitable solvent; filtering it removes the visible evidence along with some of the peptide and leaves the cause in place. Decide the solvent question first, and let the filter do only the job it can do.