The peak area a mass spectrometer returns is proportional to concentration only while ionisation efficiency holds constant, and in electrospray it does not hold. Anything co-eluting with the analyte competes for charge and for room on the surface of the evaporating droplet, and the response moves while the concentration stays put. Matuszewski, Constanzer and Chavez-Eng stated the consequence plainly: the early perception that HPLC-MS/MS practically guarantees selectivity is being challenged by reported examples of lack of selectivity due to ion suppression or enhancement caused by the sample matrix and interferences from metabolites, and although regulatory requirements include the need to assess and eliminate the matrix effect, the experimental procedures necessary to assess it are not detailed [1]. What follows is those procedures — what each one isolates, and what it leaves unmeasured.
Three quantities routinely collapsed into one
Matrix effect, recovery and process efficiency come from different sample sets and answer different questions. Matrix effect compares analyte spiked into an already-extracted blank against the same amount in neat solution, isolating what ionisation did. Recovery compares analyte spiked into matrix before extraction against that post-extraction spike, isolating what the extraction lost. Process efficiency compares the pre-extraction spike against neat solution and folds the two together, which is exactly why it cannot tell you which of them moved. A recent systematic assessment of all three — developed on a cerebrospinal-fluid lipid assay rather than a peptide one — argues for calculating absolute and relative values alongside their internal-standard-normalised factors, in order to determine the extent to which the internal standard compensates for the variability introduced by the matrix and the recovery fraction; the same authors record that guidelines on bioanalytical method evaluation are not harmonised and can occasionally be ambiguous [2].
Absolute and relative are not interchangeable, and the distinction between them is Matuszewski's [1]. In the usage that followed, the absolute figure is the ratio measured in one lot of matrix and the relative figure is how that ratio varies between lots, which is the quantity that decides whether a method survives donor-to-donor variation. The same paper describes experiments that determine the true recovery of analytes by HPLC-MS/MS, eliminating the uncertainty about the effect of matrix on this commonly measured method parameter [1]. That is not bookkeeping: a recovery figure computed from responses that are themselves suppressed is not a recovery figure.
The relative effect is usually reported as the coefficient of variation of the internal-standard-normalised matrix factor across lots. Rudzki and co-workers compared the internal-standard-normalised matrix factors recommended by the European Medicines Agency against the Matuszewski calculation, both computed from post-extraction spiked samples, across nine analytes of diverse chemical structure; practical differences were minimal, the EMA approach appeared slightly more conservative on their datasets, and they add that further studies are encouraged to confirm their observations [3]. Either calculation is defensible. Choosing one and reporting the lot-to-lot spread is the part that matters.
Post-column infusion tells you where; the spike tells you how much
Bonfiglio, King, Olah and Merkle built the experiment that maps suppression onto the chromatogram: a constant flow of analyte is infused into the eluent downstream of the column while a blank matrix extract is injected, so any depression or rise in the otherwise flat infused trace marks a retention window in which matrix is altering ionisation [4]. The trace localises the problem along the gradient. It does not hand you a correction factor, because the infused compound arrives continuously at a concentration you chose rather than as a peak at the concentration in your sample; the number still has to come from the post-extraction spike.
What that infusion system revealed about sample preparation is worth carrying forward. Comparing methyl-t-butyl ether liquid–liquid extraction, two solid-phase extraction formats and acetonitrile protein precipitation, the acetonitrile precipitation samples showed the greatest amount of electrospray response suppression while the liquid–liquid extracts demonstrated the least, and the suppression was compound-dependent [4].
Phospholipids are a large part of the reason. Ismaiel and colleagues tracked glycerophosphocholines through their characteristic fragment ions and used post-column infusion to confirm that the suppression and enhancement regions coincided with the phospholipid elution profiles; in their assay for chlorpheniramine — a small molecule, so read the magnitude as an illustration and not as a peptide result — the mean matrix effect under atmospheric-pressure chemical ionisation was 75% less than the mean matrix effect in electrospray, and they concluded that a matrix effect study in one source of biological fluid may not be sufficient [5].
Extraction: cleanliness bought with loss
Chambers, Wagrowski-Diehl, Lu and Mazzeo compared protein precipitation, liquid–liquid extraction, pure cation-exchange, reversed-phase and mixed-mode solid-phase extraction, and report that protein precipitation is the least effective sample preparation technique, often resulting in significant matrix effects, while the combination of polymeric mixed-mode SPE, appropriate mobile phase pH and UPLC provides significant advantages for reducing matrix effects [6].
For peptides the same trade has been measured directly. Esposito and co-workers digested four model peptides — somatostatin, GLP-2, human insulin and liraglutide — to generate catabolite mixtures spanning a wide range of hydrophobicity and isoelectric point, spiked them into human plasma, and ran several precipitation and SPE protocols. Precipitation with three volumes of acetonitrile or ethanol yielded the highest overall recoveries, more than 50% for the four parent peptides and all their catabolites; among five SPE sorbents, mixed-mode anion exchange was the only one that extracted all the peptides with recoveries above 20%, and matrix effects were generally reduced with the SPE approaches [7]. That is the whole trade inside one experiment: the protocol that keeps the most analyte is not the protocol that delivers the cleanest extract, and peptides at either extreme of hydrophilicity or hydrophobicity had a much narrower choice of precipitation solvent or SPE protocol.
Staging the two can move both numbers at once. Wang, Kong and co-workers placed a hydrophilic-interaction solid-phase extraction directly downstream of protein precipitation for triptorelin in plasma; against the alternatives they compared, this format gave the minimum matrix effect, 96.02–103.41%, together with the maximum recovery, 80.68–90.54% [8].
Immunocapture buys selectivity in a different currency. In the SISCAPA design, anti-peptide antibodies immobilised on 100 nanolitre nanoaffinity columns enrich specific peptides along with spiked stable-isotope-labelled internal standards; working with tryptic peptides from four human plasma proteins, Anderson and colleagues reported an average 120-fold enrichment of the antigen peptide relative to others, cycle-to-cycle coefficients of variation near 5%, and columns that could be recycled with little loss of binding capacity [9]. Multiplexed and moved between laboratories, an 8-plex immuno-MRM assay reached limits of detection at or below 1 ng/mL for the assayed proteins in 30 µL of plasma, with median intra- and interlaboratory coefficients of variation above the limit of quantification of 11% and below 14% — but trypsin digestion and its requisite sample handling contributed the most to assay variability and reduced the recovery of target peptides from digested proteins [10]. Enrichment does not abolish variability. It relocates it to whichever upstream step you have not controlled.
Adsorption is a loss you cannot correct away
Peptides disappear onto surfaces, and they do it twice. Maes and co-workers separate the two faces: carryover results from the aspecific adsorption of analytes to parts of the analytical system and introduces bias in both identification and quantification assays, while nonspecific binding at the surfaces of pipette tips, sample tubes and LC vials means linearity, sensitivity and repeatability of the analyses are negatively affected. Because of the great diversity in physicochemical properties of biomolecules there is no general approach available to minimise adsorption phenomena, which is why proper solubilisation is treated as one of the most important factors for a successful analysis [11].
The two controllable variables are the solvent the sample sits in and the vessel it sits in. Feickert and Burckhardt attacked both with a design-of-experiments approach for substance P and hemokinin-1: adding formic acid, dimethyl sulfoxide and organic solvents to the injection solvent produced a substantial gain of intensity for both peptides compared with using the start gradient as injection solvent, and the systematic investigation demonstrated polypropylene as the most favourable container material [12]. In the broader survey of quantitative peptide bioanalysis by van den Broek and colleagues, peptide adsorption and degradation sit alongside sample pretreatment, internal standard selection and assay validation as standing problems for this analyte class, in the same list as the chromatographic variables — mobile phase, stationary phase, flow rate, temperature — and the mass spectrometric ones, ionisation mode, collision-induced dissociation and response factors [13]. Analyte lost to a tube wall before the pre-extraction spike is even measured is invisible to every ratio described above, which is why adsorption is a quantification problem and not laboratory housekeeping.
What the internal standard fixes, and the residue it leaves
A stable isotope-labelled version of the analyte is the correction that comes closest to following the analyte through both suppression and extraction loss, because it co-elutes and ionises in the same environment. The qualifier is co-elutes. Wang, Cyronak and Yang showed on carvedilol enantiomers — small molecules, though the mechanism is not specific to them — that a slight difference in retention time between analyte and SIL internal standard, caused by the deuterium isotope effect, resulted in a different degree of ion suppression in certain lots of plasma [14]. A standard eluting fractionally early is corrected against a different ionisation environment from the one the analyte experienced, and where the local suppression gradient is steep that offset propagates straight into the ratio. This is one of the more useful things the post-column infusion trace can be used for: it shows how steep that gradient is in the retention window you actually occupy, and therefore what a small offset would cost.
The label also has to enter the workflow at the right point. Where a peptide is quantified after enzymatic digestion, a labelled peptide spiked into the digest cannot report anything that happened during digestion. In the interlaboratory immuno-MRM work, using a stable isotope-labelled protein as an internal standard instead of stable isotope-labelled peptides, so that the standard passes through digestion as well, nearly doubled assay accuracy while improving assay precision 5% [10]. Arnold, Stevison and Isoherranen observed considerable differences in digestion efficiency, sensitivity and matrix effects between peptides, and concluded that different sample matrices have peptide, time and matrix specific effects on protein digestion and absolute quantification [15]. A surrogate peptide characterised in one matrix is not characterised in another.
Selectivity against endogenous forms of the same sequence
An assay for a peptide that also occurs endogenously has to separate the two, and a transition shared between them cannot do it — selectivity has to come from a mass difference the instrument can see or from a separation that resolves the forms. Chambers, Fountain and co-workers quantified intact human insulin and five recombinant analogues directly in plasma using mixed-mode SPE and a multidimensional LC method built around a solid-core anion-exchange column. Matrix factors for all analogues were calculated in six sources of human plasma with coefficients of variation below 15% in all cases, lower limits of quantification were 50–200 pg/mL from 250 µL of plasma, and the presence of high levels of human insulin and bovine insulin did not interfere with quantification of any of the analysed analogues [16]. These are intact polypeptides measured without digestion; the selectivity argument rests on sequence differences carried through to the intact mass and its fragments.
Where two endogenous peptides are closely related, the antibody-based alternative is the one that fails first. Substance P and hemokinin-1 were predominantly examined by immunoassays with their limitation to differentiate appropriately between both peptides, and the LC-MS/MS method built on the adsorption work above enabled the simultaneous and selective determination of endogenous substance P and hemokinin-1 plasma levels [12].
Calibrating in a matrix that already contains the analyte
When the peptide is endogenous there is no blank matrix, and the calibration line has to be constructed some other way. Thakare and colleagues set out four routes — standard addition, background subtraction, surrogate matrix and surrogate analyte — and are explicit about the price of the last: for the surrogate analyte approach it is required to demonstrate similarity in matrix effect and recovery between surrogate and authentic endogenous analytes [17]. That demonstration is not a formality. It is the assumption every reported concentration rests on, and it is testable with the same post-extraction and pre-extraction spike sets used to characterise the method in the first place. The surrogate matrix route shifts the burden rather than removing it, because a stripped or artificial matrix then has to be shown to suppress like the real one.
Calibrator identity matters as much as calibrator matrix once digestion is in the workflow. The validated peptide method discussed above used recombinant protein calibrators together with stable isotope-labelled peptide internal standards [15], which places the calibrator upstream of the digestion step rather than downstream of it.
What the numbers license
An internal-standard-normalised matrix factor near unity with a low coefficient of variation across six donor lots licenses exactly one conclusion: in those lots, at those concentrations, the internal standard tracked the analyte. It says nothing about a lot you did not test, nor about a co-eluting species present in only some samples — the argument for multiple matrix sources made earlier. Reporting matrix effect and recovery separately is what tells you which term moved when an assay drifts; process efficiency, being their product, cannot. And no ratio recovers analyte that never reached the vial.