Desalting and Buffer Exchange: What Each Method Costs the Peptide

A peptide is rarely already in the buffer the next measurement wants. It comes off preparative chromatography in acetonitrile and trifluoroacetic acid, out of a digest in ammonium bicarbonate, or out of a binding experiment in 150 mM sodium chloride, and the instrument downstream needs something else. Every route between those two states is a partition, and in each one the peptide is the species most likely to end up where you did not intend it — adsorbed on a membrane, retarded inside the pore volume of a resin, or sitting in the flow-through of a sorbent bed. This guide covers what the salt costs at the ion source, how each desalting format behaves for a small and often hydrophobic or charged analyte, and the two situations in which the step alters the molecule rather than only its surroundings.

What the salt costs at the ion source

Salt is not removed because it is dirty. King and colleagues designed experiments to separate gas-phase from solution-phase contributions to ionisation suppression in biological extracts, and reported that gas-phase reactions leading to loss of net charge on the analyte are not likely to be the most important process involved; their results pointed instead to changes in the droplet solution properties caused by the presence of nonvolatile solutes as the main cause ^[1]. A review describes it in the same terms: suppression arising from less volatile compounds that can change the efficiency of droplet formation or droplet evaporation, which in turn affects how much charged ion reaches the detector; salts and ion-pairing agents head its list of materials shown to cause it ^[2].

The magnitude is not subtle. In infusion experiments on insulin, myoglobin and bovine serum albumin, sodium phosphate buffer at 5 and 10 mM, pH 7.5, almost completely diminished the MS response, while 50 mM formic acid at pH 2.4 gave the most intense protein signals; ammonium formate and ammonium borate at up to 100 mM, pH 8.5, also caused protein ion suppression, although with the borate buffer significant MS intensity remained ^[3]. Two things follow. Single-digit millimolar nonvolatile buffer was enough to all but erase the protein signal there, and a volatile buffer is not free — it is cheaper, at a concentration-dependent price. Those figures were measured on intact proteins under direct infusion, not on a small peptide eluting from a gradient, and do not transfer as numbers.

On a MALDI target the failure mode looks different. Alkali adduction splits one analyte across several m/z channels, so the cost is precision as much as sensitivity. In an analysis of cyclic cyanobacterial peptides by MALDI time-of-flight, adding 200 µM zinc sulfate heptahydrate to samples before spotting enhanced detection of the protonated molecule while suppressing competing adducts, and the coefficient of variation for microcystin-LR fell from 31.1% to 5.2%; 18-crown-6 ether, lithium, sodium and potassium chloride and several transition metal salts were also tested and were unable to achieve comparable results ^[4]. Adduct heterogeneity is quantitatively expensive; the remedy, though, was specific to one additive in one sample class.

Size-exclusion spin columns

A desalting resin separates on hydrodynamic size: the analyte is excluded from the pore volume and leaves in the void, while salt ions enter the pores and are retarded. In a spin format the whole operation is one centrifugation, which is why it is the default whenever a sample must not sit around. Automated size-exclusion tips have been characterised in this role. Miniaturised size-exclusion columns in pipette tips on a liquid handler achieved buffer exchange of the denatured, reduced and alkylated intact antibody in under 15 minutes, ahead of digestion, with no significant change observed in product attributes and their quantities compared with manual, low-artifact sample preparation, and with sample recovery of 86–94% against 29–62% for the manual spin cartridges the same group used, across a wide variety of starting sample concentrations ^[5].

Two cautions attach to that. The recovery comparison is between formats, and what was recovered was intact protein measured before digestion; it says nothing about the absolute recovery of any particular peptide sequence on a size-exclusion tip. And the separation only exists if the peptide is genuinely excluded from the pore volume, which is a property of the resin's stated fractionation range, not something to assume because salt is small; a peptide falling inside that range is retarded along with the salt, in a proportion nobody has measured for you.

Dialysis, and the cutoff you are trusting

Dialysis is the slowest of these formats and the least mechanically aggressive, and its mechanics are worth stating plainly: conventional dialysis separates small molecules from large by allowing diffusion of only the small molecules through a selectively permeable membrane, with the sample sealed in tubing and immersed in a chosen buffer until small solutes equilibrate between sample and dialysate; concomitant with the movement of small solutes across the membrane, however, is the movement of solvent in the opposite direction ^[6]. The second half of that sentence is the one people forget: the concentration going in is not the concentration coming out, so a mass balance after dialysis must account for a volume that changed while nobody watched. And because each buffer change only brings the sample to a fresh equilibrium with a new reservoir, residual salt is governed by the volume ratio and the number of changes rather than driven to zero by any one of them.

Dialysis and centrifugal ultrafiltration both rest on a molecular-weight cutoff, and that rating is a calibration rather than a sieve dimension. Its limits show most clearly where the calibrant and the analyte differ in shape. Work on ultrafiltration of polysaccharides makes the point explicitly: the commonly applied MWCO is based on globular marker proteins with a narrow relation between molecular weight and hydrodynamic volume, and ultrafiltration of randomly coiled pullulan and rigid rod-like scleroglucan through the same 10 kDa polyethersulfone membrane yielded significantly different recoveries of 1% and 71% respectively, with deviations greater than 77-fold between nominal MWCO and apparent molecular weight ^[7]. Those numbers were measured on polysaccharides and do not transfer to peptides. What transfers is the mechanism: the number on the box is a calibration against one solute class, and an analyte not shaped like the calibrant will not honour it.

Nominal identity between devices does not guarantee behavioural identity either. Four commercially available filter membranes of similar nominal molecular weight cutoff, but differing membrane chemistries and filter orientations, were evaluated for isolating the low-molecular-weight fraction of human plasma; of those, only one — a tangential membrane device rated at 20K — was effective in the non-retention of species above 50K and in recovery and enrichment of low-molecular-weight components ^[8]. The same study then optimised buffer composition, centrifugal force, duration and temperature around that device: 100 µL of plasma in 10% acetonitrile, 4000 × g, 35 min, 20 °C. A cutoff rating is the beginning of a method, not the method.

The problem sharpens when the peptide sits near or below the rating. Ultrafiltration of bovine haemoglobin hydrolysates through 1 and 3 kg·mol^-1 regenerated cellulose membranes was used deliberately to drive a small peptide into the permeate, reaching a maximal enrichment factor of about 29 and a permeate purity of about 70% for the target sequence ^[9]. That is a separation working as designed, and it is the same physical situation as trying to diafilter salt away from a peptide of comparable size on a membrane of that rating. A peptide that permeates is not retained: the wash that carries salt to waste carries the peptide with it, and the only open question is the ratio.

Tangential flow and hollow fibres

Recirculating the sample across the membrane instead of pushing it through relieves the polarisation that builds on a dead-end filter, and substitutes a surface problem. Hollow fibre membranes are often hydrophobic, and this can result in adsorption of protein and/or peptides ^[10]. The effect is analyte-specific in the way that matters most: adsorption was tested with one hydrophobic and one hydrophilic peptide against two biocompatible hollow fibres made from different materials, and the fibre made from polyethylene coated with ethylene-vinyl alcohol gave near 100% recovery for both peptides, in contrast to the polysulfone fibre when sampling the hydrophobic peptide ^[10]. Loss in a membrane process is therefore not a scalar you can correct with a factor. It is a composition change, and the sequences it removes are the ones a hydrophobic surface prefers.

Reversed-phase solid-phase extraction

C18 in a tip or a spin cartridge is among the most common desalting steps in peptide work, and the one with the most behaviour. It is a chromatographic step run without a gradient, so its selectivity is set by wash strength and by how much material the bed is asked to hold. A systematic characterisation of a C18 microcolumn found that peptide losses can be reduced with low acetonitrile concentration in the washing buffer, giving a recovery rate of approximately 82%; that the dynamic binding capacity of the column was approximately 26 µg; and that with increased loading amounts, more hydrophilic peptides were replaced by hydrophobic peptides ^[11]. The last is the important one: past capacity, a desalting step quietly becomes an enrichment step, and it enriches against precisely the polar sequences that were already hardest to see. The wash strength is a dial with a cost at both ends, and only the loss-reducing end was quantified in that work.

The complementary failure is retention rather than displacement. Graphite powder packed in constricted tips was introduced as an alternative to reversed-phase microcolumns because such columns are able to retain small and/or hydrophilic peptides that can be lost when using RP microcolumns ^[12]. Note the hedge in the original — can be lost, in the mixtures examined — and note the practical consequence. If your analyte is short and polar, a C18 tip is a plausible place for it to vanish, and the flow-through is worth collecting and checking rather than discarding on faith.

Sorbent choice has been compared systematically at low load. Sixteen stationary phases — eight reversed-phase, five graphite, two ion-exchange and one hydrophilic-lipophilic balance, configured as thirteen self-packed and three commercial centrifugal cartridges or spin tips — were evaluated on 1 µg tissue and cell line digests. Two reversed-phase, one graphite and the HLB self-packed tips gave excellent results, with sample loss significantly reduced compared with one of the commercial methods, 22–58% more unique phosphopeptides identified, and recovery higher by 132–155% ^[13]. Those percentages are relative to that one commercial comparator, on phosphopeptides, at 1 µg input. They are not a general ranking of formats.

At adequate load the picture is less dramatic. Four single-step desalting devices — two ZipTip formats, a C18 TopTip and an HLB µElution plate — were compared on tryptic peptides from whole saliva. Each facilitated the identification of about 340 proteins, inter-approach Pearson correlation coefficients exceeded 0.95, and the authors concluded that the devices performed equally well in removing low-molecular-weight contaminants; purification-method-dependent variations in protein composition were nonetheless observed ^[14]. Aggregate counts can agree while the identity of what survived does not. A device swap in the middle of a study is a change to the sample, not only to the workflow.

Recovery is a function of how little you have

Adsorptive loss tracks wetted surface and is largely indifferent to how much peptide is dissolved, so the fraction lost climbs as concentration falls — which is why a step that is invisible on a stock is decisive on a standard. A step-wise assessment of a conventional in-solution proteomic workflow starting from about 1000 mammalian cells, roughly 100 ng of material, found that surface adhesion, along with desalting and speed-vacuum drying steps, all contribute heavily to sample loss, in particular for low-abundance proteins. Adapting slippery liquid-infused porous surface treatment, commercial low-binding tubes and in-line desalting together allowed identification of on average 829 proteins from 1000 U2OS cells with 75-minute LC/MS/MS runs, an 11-fold increase in protein identification ^[15]. The 11-fold figure belongs to the combination of all three changes and cannot be assigned to the desalting change alone. What does transfer is that desalting was one of the steps carrying heavy loss at the 100 ng scale, and that reducing transfers and dry-downs was part of what recovered it.

When the step changes the peptide

Two changes are chemical rather than positional. The first is the counterion. A synthetic cationic peptide arrives paired with an anion, and any method that passes it over an ion exchanger, through a reversed-phase bed with a different acid modifier, or through a deprotonation step hands it back paired with something else. In a critical comparison on the dicationic octapeptide lanreotide, reversed-phase HPLC and an ion-exchange resin allowed partial to almost complete exchange of the tightly bound trifluoroacetate counterion for an ion from an acid weaker than TFA, such as acetic acid, while a deprotonation/reprotonation cycle of the amino groups required a basic solution that permitted complete removal of the trifluoroacetate ^[16]. Partial to almost complete is the operative phrase: one pass is not an exchange guarantee. The same work notes that the classical alternative, repeated freeze-drying in the presence of an excess of hydrochloric acid, means working at pH below 1, which can induce peptide degradation.

The second is pH. Desalting buffers are chosen for volatility and instrument compatibility, rarely for what they do to labile residues, and one common volatile choice, ammonium bicarbonate near pH 8, sits in an unhelpful range. Deamidation of peptidyl asparagine generates aspartic acid or isoaspartic acid; it is a spontaneous, non-enzymatic modification that can easily be introduced during sample preparation, and the newly generated residues may block or introduce new proteolytic sites and convert one asparagine-containing peptide into multiple species that affect quantification. Neutral to mildly basic conditions favour the reaction, while mildly acidic conditions markedly slow it — which is why a Glu-C digestion was moved to pH 4.5 in ammonium acetate, a volatile buffer compatible with mass spectrometry, in place of pH 8.0 ammonium bicarbonate, with nearly identical sequence specificity observed at both pH values ^[17]. That result concerns a proteolysis step, not a desalting step, and is not a measured deamidation rate for your peptide. The chemistry, though, does not care which step holds the residue at pH 8: hours spent equilibrating in a mildly basic volatile buffer are hours in a deamidation-favouring condition.

Not doing the step at all

The cheapest desalting is the buffer you never used. Where the assay tolerates it, running the peptide in a volatile system from the outset — a formate or an acetate rather than a phosphate — removes the requirement instead of managing it, and the infusion data above put a number on the difference: phosphate at 5–10 mM almost completely diminished the protein signal in a system where 50 mM formic acid maximised it ^[3]. Ammonium acetate is used as a volatile, mass-spectrometry-compatible buffer at mildly acidic pH for exactly this reason ^[17]. The tradeoffs are real. Volatile buffers still suppressed protein ion signal at up to 100 mM in the same infusion study ^[3], their useful buffering ranges are narrow, and whether a given peptide's solubility and aggregation state survive the drop in ionic strength is a question to answer experimentally before the switch, not after.

Measuring what the step did

Recovery through a desalting step is not observable from the fact that liquid came out the other end. The minimum standard proposed for characterising suppression is a signal recovery study using extracts with added analyte, with post-column infusion of the analyte offered as the more comprehensive approach for detecting protracted ionisation effects across a run ^[2]. Ported to desalting, that means putting a known quantity of your peptide through the entire procedure and quantifying it by the assay you actually intend to use, in the buffer it will actually be in. Both failure modes described above — adsorptive loss on the device and selective retention by the sorbent — yield an eluate that looks perfectly clean. And because those losses are sequence-selective rather than uniform ^{[10, 11, 12]}, a single recovery percentage for a mixture is a summary that can hide the thing you most need to know: which peptides are no longer in the tube.