Stable Isotope-Labelled Internal Standards for Peptide Quantitation

Between the moment a sample enters the tube and the moment an ion arrives at the detector, the amount of any given peptide changes. It adsorbs to surfaces, it is incompletely released from its parent protein by proteolysis, it is partly lost on a clean-up cartridge, and in the electrospray plume it competes for charge against whatever happens to be co-eluting with it at that second. None of those losses is constant from sample to sample. An external calibration curve cannot see any of them, because the calibrator never travelled the sample's route.

Isotope dilution addresses this by putting a second copy of the analyte into the tube early enough to share that route. The copy is the same molecule, differing only in nuclide composition, so it partitions, adsorbs, digests and ionises alongside the native peptide while remaining resolved from it in the mass domain. What is reported is the ratio of two signals acquired in the same scan under the same instantaneous conditions; the known amount of the spiked copy converts that ratio into an absolute figure.

The AQUA principle

Gerber and colleagues set out the version of this now standard in targeted proteomics ^[1]. Peptides are synthesised with stable isotopes incorporated so that they mimic native peptides formed by proteolysis, and the synthetic standards can additionally be prepared with covalent modifications — phosphorylation, methylation, acetylation — that are chemically identical to naturally occurring post-translational modifications. The labelled peptides are then used to measure the absolute levels of proteins and post-translationally modified proteins after proteolysis, using a selected reaction monitoring analysis in a tandem mass spectrometer ^[1]. The original work applied the strategy to low-abundance yeast proteins involved in gene silencing, to the cell-cycle-dependent phosphorylation of Ser-1126 of human separase, and to identifying kinases capable of phosphorylating Ser-1501 of separase in an in vitro kinase assay ^[1].

The procedural detail carrying most of the weight is where the standard enters. In the general procedure described by Kirkpatrick and colleagues, the AQUA peptide is introduced at a known concentration into cell lysates during digestion, and one SRM experiment then detects and quantifies both the native peptide and the labelled standard ^[2]. Everything downstream of that spike is normalised by the ratio. Everything upstream of it is not — it is simply assumed to have behaved.

Choosing the nuclide

The design premise is that the standard is chemically indistinguishable from the analyte apart from mass ^{[1, 2]}. A label that measurably separates the two forms in time weakens that premise, because the heavy and light signals are then no longer being acquired from the same instant of the gradient, and in complex matrices the co-eluting background is not benign: MRM measurements of peptides with stable isotope-labelled internal standards (SIS) can be highly precise and quantitative, but interferences occur frequently enough that the data have conventionally required manual review ^[5].

Deuterium is the label most likely to produce that separation, and the effect has been measured on peptides. Zhang and Regnier ran a BSA tryptic digest differentially labelled with the light and ²H₈ forms of the ICAT reagent, which derivatises cysteine-containing peptides: chromatographic resolution between the two isoforms exceeded 0.5 for 20% of the peptides in the digest, and three-quarters of that group contained two cysteine residues and were doubly labelled. Among peptides carrying a single ICAT residue, only 23% showed resolution greater than 0.4 ^[3]. In the same study, peptides differentially labelled with ¹³C- and ¹²C-succinate never exceeded a resolution of ±0.01 — within the limits of the method used to measure resolution — from which the authors concluded that the ¹³C- and ¹²C-coded isoforms did not resolve. With ¹³C coding, the isotope ratio could therefore be determined from a single mass spectrum taken at any point in the elution profile ^[3].

The structural basis was mapped in a companion study of six deuterated coding agents, which were found to vary widely in both their absolute and their relative contribution to the chromatographic isotope effect. The single most dominant factor was the placement of the deuterium atoms relative to hydrophilic functional groups in the coding agent: the probability of a deuterium atom interacting with the reversed-phase stationary phase, and so affecting resolution, is greatly diminished when it sits adjacent to a hydrophilic group, which the authors explained by solvophobic theory. Peptide size and coding agent size both correlated inversely with the magnitude of the effect ^[4].

Both of those studies used deuterated chemical derivatisation reagents rather than synthetic peptides carrying heavy amino acids, so the numbers do not transfer directly to an AQUA-style standard. What transfers is the mechanism and its direction: deuterium substitution can alter reversed-phase retention, the size of the effect depends on where the deuterium sits, and ¹³C substitution under the same conditions did not produce a measurable separation ^{[3, 4]}. That is the mechanistic case for preferring ¹³C labelling — and, by the same reasoning, other non-exchangeable heavy-atom labels such as ¹⁵N — in a peptide internal standard, and for treating co-elution of a deuterated standard as something to demonstrate on your own gradient rather than assume.

Where the standard is introduced

A peptide-level spike is added after the protein has already been handled. It normalises solid-phase extraction, chromatography and ionisation, and, because it is present during proteolysis, it shares the digestion buffer and incubation ^[2]. What it cannot report on is the liberation of the native peptide from its protein, because the standard was never inside a protein. Brun and colleagues reported that use of chemically synthesised isotope-labelled marker peptides can lead to severe biases, and presented an alternative in which in vitro-synthesised isotope-labelled full-length proteins are used as standards. Because those protein standards match the biochemical properties of the target proteins, they can be added directly into the samples to be analysed, which the authors reported allowed accurate quantification even in prefractionated complex samples; the demonstration used water and urine samples containing Staphylococcus aureus superantigenic toxins ^[6].

A middle position is the concatenated standard. Beynon and colleagues identified the principal limitation on widespread adoption of peptide-level absolute quantification as the availability of standard signature peptides in accurately known amounts, and answered it by designing an artificial gene encoding a concatenation of tryptic peptides — a QconCAT — drawn from several chick skeletal muscle proteins, with features built in for quantification and purification ^[7]. In the full characterisation of that approach, a QconCAT containing one tryptic peptide from each of 20 soluble chicken skeletal muscle proteins was expressed in Escherichia coli in minimal medium containing selectively stable isotope-labelled amino acids, giving an equimolar series of uniformly labelled proteotypic peptides. Purified, quantified and added in known amount to a homogenised muscle preparation before trypsin digestion, the QconCAT was completely digested at a rate far higher than the analyte proteins, which the authors took as confirming the applicability of such artificial proteins for multiplexed quantification ^[8].

A head-to-head evaluation is more informative than any single platform's own report. Scott and colleagues compared synthetic peptides, QconCAT constructs and recombinant proteins as internal standards for the absolute quantification of three human cytokine proteins by MRM, assessing them by calibration curve construction and regression analysis. Full-length protein standards showed the broadest quantitative reliability, while accuracy was peptide-dependent for both QconCATs and synthetic peptides. Adding natural flanking sequences — the residues that in the real protein sit on either side of the signature peptide — greatly improved the quantitative performance of both the QconCAT and the synthetic peptide standards ^[9]. Since the flanking residues are what the protease has to cut through, that result suggests a substantial part of the peptide-level standard's disadvantage sits at the cleavage step rather than in the peptide sequence itself.

Where an affinity capture step precedes the digestion, the spike point can be tested directly. Faria and colleagues used an extended stable isotope-labelled signature peptide, matching the antibody epitope, as an internal standard while deliberately inducing immunocapture variability by varying the antibody amount per well from 150 to 4500 ng. With the standard added after immunocapture, the observed variability ranged from −80.9% to 77.0%; with it added before, from −37.5% to 20.3% ^[10]. Being present for the capture step narrowed the induced variability substantially without abolishing it, which is what one would expect when a standard is moved one step earlier, though this was one analyte and one deliberately induced source of variability.

At scale the same logic holds. In an eleven-laboratory study using 14 LC-MS systems and MRM assays for 125 peptides derived from 34 proteins in human plasma, digestion recovery of peptides was assessed and quantitative accuracy improved by using heavy-isotope-labelled versions of the proteins as internal standards in addition to the labelled peptides ^[11].

The standard's own assigned amount

Every absolute figure the method produces is proportional to the amount assumed for the spike. An error in that assumed amount is systematic: it multiplies every sample by the same factor, it does not enlarge the scatter between replicates, and no amount of replication will reveal it. This is where the argument joins the general problem of reference standards and metrological traceability — the number attached to the standard is a claim inherited from someone else's measurement, and it is worth knowing which measurement.

The usual measurement is amino acid analysis. As Stoppacher and colleagues describe it, common practice is to quantify the mass concentration of a peptide solution through quantitative determination of selected chemically stable amino acids produced after complete hydrolysis of the parent peptide, because there is generally insufficient material to perform a direct purity characterisation of the intact peptide and constitute a calibration solution from it ^[12]. Assignment by the alternative mass balance route means estimating the mass fraction of every impurity present and taking the difference from the theoretical limit. Working on the model peptide angiotensin I, the authors used data-dependent LC-high-resolution MS/MS to detect and structurally characterise peptide fragments generated on storage at elevated temperatures, and quantified the major impurities in a candidate reference material at a mass fraction of 10.4 mg/g. Failure to correct for those impurities, they calculated, would produce a 1% error in the peptide concentration determined by amino acid analysis ^[12].

How much the choice of characterisation method itself contributes was tested in a multi-laboratory comparison on the nonapeptide oxytocin, setting an HPLC assay, quantitative NMR and amino acid analysis against one another. The HPLC assay, run against the same peptide bulk material as the standard, gave the lowest inter-laboratory variability — but that coefficient of variation was calculated without counting the uncertainty attached to the mass balance assignment of the standard itself, so the comparison is not a like-for-like ranking of total uncertainty. The authors noted that qNMR is a direct measurement of the peptide against an internal standard, straightforward to perform under common laboratory conditions, and concluded that it deserves further exploration as a primary method for peptide reference standard value assignment ^[13].

Transition selection

In an SRM experiment the two mass filters of a triple quadrupole select a predefined precursor ion and one of its fragments and monitor that pair over time; a series of such transitions combined with the retention time of the targeted peptide can constitute a definitive assay ^[14]. The internal standard doubles the transition list, because each light transition needs its heavy counterpart.

Selecting proteotypic peptides is the conventional first step, but Sherman and colleagues argue that the usual justification for it is the wrong one. Avoiding redundancy at the sequence level is incidental; the reason that matters is that homologous peptides produce redundancy in the mass-to-charge domain — which is the domain the instrument actually filters in ^[15]. Two peptides that are not identical in sequence can still share a precursor and a product mass, and the quadrupoles cannot tell them apart.

The heavy standard is also the tool for finding interference after the fact. The AuDIT algorithm compares the relative product ion intensities of the analyte peptide with those of the SIS peptide and applies a t-test to decide whether they differ significantly, alongside a coefficient of variation computed from the analyte-to-SIS peak area ratio across replicate samples. It identified problematic transitions with accuracies of 94–100% and with sensitivity and specificity of 83–100% for correct identification of errant transitions ^[5]. The logic is worth internalising even if the algorithm is not used: the heavy peptide supplies the expected fragmentation pattern, so a light transition whose relative intensity departs from its heavy twin is carrying something that is not the analyte.

Isotope overlap at low mass shift

The assumption behind every ratio above is that there is no cross-talk between the standard's channel and the analyte's. Rule and colleagues point out that this has been an implicit assumption rather than a demonstrated fact, and that in some cases naturally occurring isotopes of the analyte do contribute to the signal of the internal standard. The contribution becomes more pronounced for isotopically rich compounds — those containing sulfur, chlorine or bromine — for higher molecular weight compounds, and at high analyte-to-internal-standard concentration ratios, and it can create non-linear calibration behaviour that may bias results. The same correction handles the reverse case, in which unlabelled analyte is present in the internal standard as an impurity ^[16].

That work is general quantitative mass spectrometry rather than a peptide study, and its worked examples are not peptide measurements. What carries across is the shape of the problem: the cross-contribution is governed by the mass shift the label provides relative to the analyte's own natural isotope envelope, it grows with molecular weight, and it grows as the analyte comes to outweigh the standard ^[16]. The practical consequences for a peptide standard are to prefer a mass shift generous enough to clear that envelope over the smallest shift that separates monoisotopic peaks, to avoid running the standard far below the analyte, and, where a small shift is unavoidable, to treat the overlap as a quantity to be measured and built into the calibration model rather than an unknown to be hoped away ^[16].

Linearity, and what the ratio licenses

Linear dynamic range, limits of detection and limits of quantification are properties of the whole assay rather than of the isotope dilution principle, and they are established the same way as for any other method — by the experiments a validation protocol prescribes, in the matrix the assay will actually run in. The NCI-CPTAC multi-site study assessed exactly these figures for multiplexed MRM-based assays and reported that, using common materials and standardised protocols, the assays could be highly reproducible within and across laboratories and instrument platforms and were sensitive to low µg/ml protein concentrations in unfractionated plasma ^[17]. The later eleven-laboratory study reported median intra- and inter-laboratory reproducibility below 20% for its 125-peptide panel, with low to sub-ng/ml sensitivity achieved by one-step immunoaffinity depletion of 14 abundant plasma proteins before analysis ^[11]. Both sets of figures are specific to those matrices and those panels; neither is a performance level you inherit by adopting a labelled standard.

What the ratio licenses is narrow, and worth stating plainly. It gives the amount of that peptide, in that digest, traceable to the assigned amount of the standard and to the transitions monitored. It becomes a statement about the parent protein only to the extent that proteolysis was complete or that the standard corrected for its being incomplete — which, in the platform comparison, was peptide-dependent for peptide-level standards ^[9]. It becomes a specific statement only to the extent that the monitored transitions are free of co-isolating interference, which is a property of the sample and has to be checked in the sample ^{[5, 15]}. And it is only as accurate as the number attached to the standard, which is a separate measurement made elsewhere and carrying its own uncertainty ^{[12, 13]}. Those are four independent claims, and a validation protocol has to earn each of them separately.