A plasma or serum stability assay puts a known amount of peptide into a biological matrix at a controlled temperature, samples it at set times, stops the reaction, and quantifies what survives. The output is a depletion curve and from it a half-life. It is one of the cheapest experiments in peptide work and one of the easiest to over-read.
The measurement is narrower than the number suggests. It reports how quickly the peptidases in one particular tube consumed one particular substrate under one particular set of conditions. Change the matrix, the anticoagulant, the donor, the dilution or the readout, and the half-life changes with them — sometimes by more than the difference you set out to detect. This guide covers what each choice does, and where the arithmetic tends to go wrong.
What is actually removing the peptide
It is worth separating this assay from the chemistry that runs on a peptide sitting in a vial or a buffer stock. Backbone hydrolysis, deamidation and oxidation are non-enzymatic, and on the one- to four-hour timescale of a typical matrix incubation they are rarely the dominant loss route. In plasma and serum, proteases are.
Those proteases are not a uniform pool. Yi and colleagues spiked an isotopically labeled peptide into human plasma and serum and followed it by time-course MALDI-TOF, using fibrinogen peptide A as a model substrate. What they saw was not a single cut but a sequence: intrinsic peptidases produced an observed first-order sequential multiple-step reaction, in which the step reactions differed in their kinetic rate constants. They took that difference to suggest a significant contribution of the truncated end residue to the substrate specificity of the peptidase or peptidases involved [1]. In practice this is why a stability run often shows a ladder of products differing by one or two residues rather than a clean two-fragment split.
Named activities behind that pattern include dipeptidyl peptidase IV (DPP IV/CD26; EC 3.4.14.5), a serine-type peptidase that removes N-terminal dipeptides [2]. Brandt and colleagues showed that purified DPP IV cleaves human BNP (1-32) with a specificity constant of 0.37 × 10⁶ L·mol⁻¹·s⁻¹, and that DPP IV activity in EDTA-plasma was able to truncate BNP (1-32) ex vivo — truncation that a specific DPP IV inhibitor prevented in a concentration-dependent manner. Under in vitro conditions where ANP was hydrolyzed extensively, both BNP (1-32) and BNP (3-32) were very resistant to cleavage by neutral endopeptidase [2].
The mix is also matrix-specific in ways that do not transfer from tissue work. Weinberger and Martinez, reasoning from metabolite accumulation with and without selective peptidase inhibitors, attributed approximately 85-90% of [Leu]enkephalin hydrolysis in rat plasma to the combined action of aminopeptidase M and angiotensin converting enzyme; "enkephalinase" and aminopeptidase MII activity against that substrate were not detectable. For the D-Ala² analog, 80-90% of hydrolysis was likewise attributed to aminopeptidase M plus ACE — in contrast to the high susceptibility to enkephalinase, and low susceptibility to aminopeptidase, reported for that analog in brain tissue. The authors offered several possible explanations, among them differences in enzyme conformation [3].
Matrix choice: blood, plasma, or serum
Serum is the supernatant left after blood clots; plasma is the supernatant after blood collected with an anticoagulant is centrifuged. Böttger and colleagues tested whether that difference matters for peptide degradation. They drew blood by cardiac puncture from CD-1 mice into a syringe already containing peptide solution; with coagulation activators such as glass surfaces and damaged cells absent, visible clotting was prevented, letting them follow degradation in fresh blood for one hour. Half-lives were then compared against serum prepared fresh from the same animal and against commercial serum and plasma.
Peptides from three different families showed considerably different stabilities across these matrices. Peptides were generally degraded faster in serum than in plasma, but all of them were more stable in fresh blood, and the order of degradation rates among the peptides varied across the six different incubation experiments. The authors' conclusion was pointed: degradation measured this way may often be misleading, prompting efforts to stabilize peptides at sites relevant only in vitro but of lower importance in vivo [4].
The rank-order instability is the part to take seriously. If the assay exists to rank analogs, and that ranking depends on which matrix you used, it is not answering the question you asked.
The HUPO Plasma Proteome Project reached a compatible position from the biomarker side, finding platelet-depleted plasma preferable to serum for certain peptidomic studies, and recommending samples be aliquoted and stored preferably in liquid nitrogen [5].
The anticoagulant is a protease inhibitor cocktail you did not design
Choosing an anticoagulant chooses an inhibition profile. EDTA chelates divalent cations and so bears on metallo- and calcium-dependent proteases; heparin acts on the coagulation cascade, including thrombin and factor Xa [4]. Neither touches the serine exopeptidases — which is precisely why DPP IV remained active enough in EDTA-plasma to truncate BNP [2].
Yi and colleagues ranked the practical consequence. Across the peptides they tested, stability followed this general order, best first [1]:
- EDTA plasma with added protease inhibitors
- EDTA plasma
- Citrate plasma
- Heparin plasma
- Serum
Oliver and colleagues found a compatible split for oxytocin: it was stable in blood collected into EDTA tubes with or without protease inhibitors, but not in lithium heparin tubes [6].
That study also isolates a variable that is easy to miss — when the inhibitor arrives. Blood collected into any EDTA-containing tube gave near-complete inhibition of oxytocinase within 100 minutes or less. Adding EDTA to lithium heparin plasma after the fact produced only a 35% reduction in oxytocinase activity, leaving roughly two thirds of that activity still running [6]. An inhibitor added at the bench is not the same as one waiting in the tube.
Adding inhibitors is not free either. The HUPO group recommends protease inhibitors but advises they be incorporated early and used judiciously, since some form non-specific protein adducts and others interfere with peptide studies [5]. Rasmussen and colleagues compared five collection containers in 17 participants and found that container choice changed measured concentrations of glucagon, GLP-1 and GIP — though not uniformly. Fasting glucagon did not depend on container by either ELISA or radioimmunoassay, while GIP measured by ELISA was higher in the control tubes and was negatively affected by P800 tubes and by added aprotinin. On that basis they recommended EDTA-coated tubes without protease inhibitors, or P800 tubes, for those analytes [7]. Note the scope: that recommendation was drawn for three specific peptides read by immunoassay, and does not transfer to an arbitrary synthetic peptide read by LC-MS/MS.
Sampling and quench chemistry
Between pulling an aliquot and injecting it, proteolysis has to stop completely, the peptide has to stay in solution, and it has to remain recoverable. Those are three separate requirements, and a quench can meet one while failing another.
An organic solvent crash — typically acetonitrile or methanol, often acidified — does double duty by denaturing peptidases and precipitating bulk plasma protein. Acidification alone can be decisive. Developing a UPLC-MS/MS method for the pentapeptide MLIF, Liu and colleagues identified instability in matrix as the primary challenge and resolved it by adding acidification reagents such as sulfuric acid, after which the analyte proved stable across various handling processes and storage conditions. The same method reported a mean extraction recovery of 43.3% [8]. A quench that halts the enzymes but leaves more than half the peptide in the pellet is still workable, because depletion is measured against a zero-time aliquot processed identically — but recovery and stability are distinct failure modes, and the internal standard has to control for both.
The internal standard is not exempt from the problem it is meant to solve. Yi and colleagues observed that synthetic peptides introduced into plasma as internal controls can themselves be degraded, and argued that their instability should not be overlooked as a preanalytical variable [1]. A stable-isotope-labeled internal standard is chemically near-identical to the analyte, so it is cleaved at a near-identical rate: harmless if added at quench, a source of silent cancellation if added at time zero.
Stabilization sometimes has to begin before the assay does. Onorato and colleagues, working with (Pyr)¹apelin-13 in rodents, found that even with a highly selective LC/MS/MS method in hand, the peptide was not detectable in rodent study samples until a detailed stabilization protocol was implemented during blood collection [9].
Several common choices are conventions rather than validated standards: incubation at 37 °C, a peptide concentration low enough to sit below the Km of the relevant peptidases, and a sampling series along the lines of 0, 5, 15, 30, 60 and 120 minutes. The right window depends entirely on how fast the peptide turns over.
Fitting the half-life, and the arithmetic that goes wrong
The standard treatment is substrate depletion: incubate, sample at several time points, quantify parent compound remaining, fit. McNaney and colleagues describe an automated LC-MS implementation using liver microsomes and up to six user-selected time points, in which the half-life is, in general, the time necessary for metabolism of 50% of the initial compound following first-order kinetics, and intrinsic clearance is the proportionality constant between the rate of metabolism of a compound and its concentration at the enzyme site [10]. The same depletion arithmetic is carried over to plasma incubations by convention.
First-order kinetics means a fixed fraction is lost per unit time, so t½ = ln2 / k, with ln2 approximately 0.693. Two half-lives leave 25% of the starting material and three leave 12.5%. Regress the natural log of fraction remaining against time; the slope is negative k.
Three failures recur. The first is extrapolating past the data. If the last time point still shows 90% of the peptide present, the run has covered only about 0.15 of one half-life, and the half-life it reports is roughly 6.6 times longer than the entire experiment. That figure is an extrapolation, not a measurement, and belongs in the record as a lower bound — half-life greater than the incubation length — rather than as a point estimate.
The second is treating a clean-looking exponential as evidence of a single reaction. The observed decay of the parent can be first-order even when the underlying process is a sequential multi-step digestion whose individual steps carry different rate constants [1]. The fitted half-life is a useful summary; it is not a mechanism.
The third is dilution. Assays are often run in diluted matrix, and the dilution factor propagates straight into the answer. A 25% plasma incubation — one part plasma to three parts buffer — contains one quarter of the peptidase content of neat plasma. If depletion is first-order in enzyme concentration and the peptide sits well below Km, the observed rate constant should fall about fourfold and the apparent half-life should be about four times longer. Half-lives measured at different dilutions are therefore not comparable, and a half-life reported without its matrix percentage is close to uninterpretable.
Why species and donor change the answer
Species differences are real but not predictable in direction, and the honest position is that they must be measured, not assumed. Murza and colleagues mapped cleavage of Pyr¹-apelin-13 in mouse, rat and human plasma and in rat cerebrospinal fluid, and found the Leu⁵-Ser⁶ bond to be the first cleavage observed in plasma, independently of species — a conserved result, not a divergent one. The same study found in vitro and in vivo maps diverging in a different way: replacing Phe¹³ with unnatural amino acids gave a 10-fold increase in plasma stability, although hydrolysis of the Pro¹²-Phe¹³ bond, previously described as a site of cleavage by ACE-2, was not observed. In vivo that bond was cleaved, yet appeared as a minor product compared with hydrolysis of the Pro¹⁰-Met¹¹ bond [11].
Donor state can matter as much as species. Oliver and colleagues measured approximately 11-fold higher oxytocinase activity in plasma from donors in late-stage pregnancy than in plasma from non-pregnant participants [6]. Commercial pooled matrix averages over donors, hiding that variance rather than removing it.
Storage behavior is analyte-specific and worth checking, not assuming. Yoon and colleagues, validating a fluorogenic Gly-Pro-AMC assay, reported that plasma DPP4 was stable under various storage temperatures, even after three freeze-thaw cycles [12] — a result for one enzyme activity in human plasma, not a general license to freeze-thaw matrix.
The readout decides part of the answer
Two methods pointed at the same tube can disagree substantially. Oliver and colleagues reported plasma oxytocin concentrations up to 33 times higher when analyzed by ELISA than by LC-MS/MS [6]. Onorato and colleagues noted that commercial ELISA kits for apelin suffer from a lack of selectivity, recognizing a range of active and inactive isoforms of the peptide [9]. An immunoassay that binds a truncated product about as well as the parent will report a stability the parent does not have — a plausible mechanism for such gaps, though not something either study set out to demonstrate.
Labeling introduces its own artifacts. De Spiegeleer and colleagues characterized iodinated mouse obestatin peptides in plasma, liver and kidney, and found cleavage sites depended on both the biological matrix and the position of the iodine-bearing residue, with the degrading proteases targeting bonds further away from the iodine while cleavage of nearby bonds was more limited. Loss by deiodination ran alongside proteolysis: in plasma, the percentage of peptide degradation attributed solely to deiodinase activity after 20 minutes of incubation reached up to 25% for 2,5-diiodo-H¹⁹-obestatin, compared with 20% for 3,5-diiodo-Y¹⁶- and only 3% for 3-iodo-Y¹⁶-obestatin [13].
What the assay does not tell you
A matrix half-life is not a circulating half-life. It contains no distribution, no renal or hepatic clearance, no tissue binding — and, as the fresh-blood comparison shows, not even the full protease environment of blood [4]. Nor does it establish which cleavage will dominate in an intact system; the apelin work is a direct example of an in vitro map that did not match the in vivo one [11].
What the assay does support is a ranking, and only within a fixed protocol: same matrix, same anticoagulant, same lot, same dilution, same quench, same readout, analogs run side by side on the same day. Reported that way — with the matrix percentage, the incubation length and the fraction remaining at the final time point all stated alongside the number — a plasma stability half-life is genuinely informative. Reported as a bare figure in minutes, it is a number whose denominator has gone missing.