Ten samples run as ten separate injections give you ten chromatograms acquired at ten different times, on a column whose retention behaviour drifted between them, sampled by a data-dependent acquisition that chose a slightly different subset of precursors each time. Isobaric labelling moves the comparison inside a single scan: each sample is derivatised with a tag chemically identical to the others and of the same total mass, the samples are pooled and injected once, and the abundance difference is read from fragment ions of a single precursor. What you gain is a comparison free of run-to-run variance. What you inherit is a quantitative measurement made on whatever else happened to be inside the isolation window at that moment.
Reporter, balancer, reactive group
Rauniyar and Yates describe the generic tandem mass tag as three functional groups: an amine-reactive group that labels the N-terminus and ε-amine group of lysine in peptides, a mass normalisation (balance) group that balances mass differences from individual reporter ions to ensure the same overall mass of the reagents, and a reporter group that provides the abundance of a peptide upon MS/MS in individual samples being mixed [1]. The isobaric condition is engineered by shuffling heavy nuclei between the two halves: the overall mass of the reporter and balance components is kept constant using differential isotopic enrichment with 13C, 15N and 18O atoms [1]. In the iTRAQ 4-plex set the reporter ranges from m/z 114 to 117 and the balance group from 28 to 31 Da, so the combined mass stays constant at 145 Da across all four reagents [1].
Two consequences follow, and between them they are the method. First, because the tags are isobaric, peptides labelled with isotopic variants of the tag appear as a single composite peak at the same m/z value in an MS1 scan with identical liquid chromatography retention time [1]. Thompson and colleagues stated the design intent directly: the tags are designed to ensure that identical peptides labelled with different TMTs exactly comigrate in all separations, permitting simultaneous determination of identity and relative abundance from a CID-based MS/MS analysis [2]. Ross and colleagues describe the same architecture from the iTRAQ side — a multiplexed set of isobaric reagents yields amine-derivatised peptides that are indistinguishable in MS but exhibit intense low-mass MS/MS signature ions that support quantitation [3]. Second, the reporter is only liberated by fragmentation: in the iTRAQ case, following fragmentation of the tag amide bond the balance carbonyl moiety is lost as a neutral loss whereas charge is retained by the reporter group [1].
Channel count is set by how finely reporter masses can be resolved, not by the reactive chemistry. The proline-based TMTpro structure described by Thompson and colleagues provides a set of 16 tags for use with resolution of 6.3 mDa mass differences in high-resolution mass spectrometers, and a set of 9 reagents with 1 Da spacing between reporter ions for single-dalton analysis using 9 heavy nuclei per tag [4]. The millidalton sets are a demand on the acquisition method rather than on the labelling bench.
Labelling at lysine and the N-terminus
The reaction is N-hydroxysuccinimide ester acylation: the tags employ NHS chemistry, and the amine-reactive, NHS-ester-activated group reacts with N-terminal amine groups and ε-amine groups of lysine residues to attach the tags to the peptides [1]. Rauniyar and Yates note that it does not occur where the primary amino group is already modified — when an N-terminal glutamine or glutamic acid has cyclised to pyroglutamic acid, or when the group is acetylated [1]. Zecha and colleagues formalise the target state as a peptide modified with TMT on all lysine side chains and on free, that is non-acetylated, peptide N-termini [5].
Getting there is a concentration problem rather than a stoichiometry problem. Zecha and colleagues evaluated under- and overlabelling of peptides from complex digests across peptide quantities of 12.5 to 800 µg and TMT-to-peptide ratios (wt/wt) from 8:1 to 1:2, and found that when reaction volumes were reduced to maintain TMT and peptide concentrations of at least 10 mM and 2 g/l, ratios as low as 1:1 (wt/wt) gave labelling efficiencies above 99% with excellent intra- and interlaboratory reproducibility [5]. Across their interlaboratory replicates an average of 99.7% of peptide-spectrum matches identified fully labelled peptides and underlabelling stayed below 0.5% of PSMs; dropping the ratio to 1:2 produced underlabelling of between 4.7 and 28.4% of PSMs depending on the laboratory [5].
Incomplete labelling costs you twice over. An unlabelled or partially labelled form sits at a different precursor mass, so unless the search looks for it, it is not identified at all — Zecha and colleagues ran separate searches with the tag as a variable rather than fixed modification on lysine and peptide N-termini in order to see underlabelling [5]. And what is identified is depleted: underlabelled peptides showed intensities always considerably lower than those of correctly labelled peptides [5]. Because the labelled fraction need not be identical in every channel, a reporter ratio computed over an incompletely labelled sample set is a ratio of labelled fractions rather than of the peptide.
Excess reagent produces the opposite defect. NHS-based isobaric tags may acylate the side-chain hydroxyl groups of serine, threonine and tyrosine under the conditions normally used for acylation of primary amines [1]. Zecha and colleagues measured the scale: TMT-labelled serine, threonine and tyrosine residues were present in on average 10.8% of identified spectra at a TMT-to-peptide ratio of 8:1, falling to 6.3% at 1:1 and 1.5% at 1:2, with serine accounting for about two-thirds of the overlabelled residues [5]. Overlabelling was strongly sequence-contingent — between 55.5 and 78.1% of overlabelled PSMs contained a histidine residue [5]. Unlike underlabelled peptides, overlabelled peptides showed comparable intensities to correctly labelled ones [5]. Reversal of peptide O-acylation can be achieved by treatment with hydroxylamine, which has no disruptive effect on acyl modifications of primary amines [1].
Co-isolation, and why ratios compress
The quantitative claim rests on an assumption that is usually false: that the ions co-isolated with your precursor are your precursor. Michalski, Cox and Mann filtered isotope clusters from standard LC runs of a cell lysate and found more than 100,000 likely peptide features, of which only about 16% had been targeted for MS/MS; the median precursor ion fraction — the proportion of the target peptide ion intensity in the MS/MS selection window — was only 0.14, which they identify as the greatest limitation on coverage [6]. That figure was measured on unlabelled peptides with a conventional isolation width, but it sets the scale: the isolation window routinely holds more foreign signal than target signal.
Every co-isolated tagged peptide deposits its own reporter ions into the same low-mass region of the same spectrum. Since the interfering population is drawn from the whole pooled sample, its reporter distribution tends toward the pool average rather than toward the target peptide's ratio, and the measured ratio is pulled toward 1:1. Rauniyar and Yates state the effect: the presence of co-eluting peptides within the isolation window used for the selection and subsequent fragmentation of individual peptide ions typically results in an underestimation or compression of actual protein abundance differences [1].
Ting and colleagues quantified it in a two-proteome model, a mixture of human and yeast proteins in a sixplex isobaric labelling system [7]. For yeast peptide ions the median compression of expected 10:1 ratios was 3.2-fold, 2.2-fold for 4:1 ratios and 1.3-fold for 2.5:1 ratios [7]. Precision degraded alongside accuracy: the 10:1 log2 ratio distribution standard deviation was 0.5 for channels without interference and 0.8 for channels with human peptide interference, an increase of 60% [7]. Note the shape of that bias: it is largest where the true difference is largest, so the fold changes you most want to trust are the ones most understated.
Ow and colleagues, assessing iTRAQ across low- and high-complexity mixtures, put ceiling and cause together: iTRAQ has the potential to provide accurate quantification spanning two orders of magnitude, but they demonstrate the interference of mixed MS/MS contribution occurring during precursor selection, an issue they described as currently very difficult to minimise [8]. They also report that the interferences were largely scenario-specific and dependent on sample complexity [8] — which is why a compression factor measured on one sample does not carry over to another.
Savitski and colleagues showed both halves of the trade this forces. Applying stringent interference filters enabled substantially more accurate TMT quantification but came at the expense of 30–60% fewer proteins quantified; an algorithm that instead corrects experimental TMT ratios on the basis of determined peptide interference levels achieved comparable accuracy while limiting the loss in coverage to under 10% [9].
Brenes and colleagues supplied a control that makes interference visible without any spike-in. In a large multibatch TMT experiment on induced pluripotent stem cell lines derived from both male and female donors, peptides mapping to the Y chromosome should be absent from the female lines; they were nonetheless consistently detected in the female channels of all TMT batches, and were then used to quantify the level of ion co-isolation and the effect of primary and secondary reporter ion interference [10].
MS3, synchronous precursor selection, and the sensitivity bill
The fix is to insert a further gas-phase selection between isolation and quantification. Ting and colleagues reported that applying triple-stage mass spectrometry almost completely eliminates interference [7]: in their MS3 analysis, median ratios for yeast peptides with and without human peptide interference were 10.5 and 11.7 against an expected ratio of 10, and 4.4 and 4.7 for the four-fold comparison [7].
The bill is paid in ions. Because the method required acquisition of two spectra per peptide ion, Ting and colleagues saw a 13% reduction in identified non-redundant peptide sequences and 8% fewer proteins assigned, with a 12% reduction in the number of proteins quantified overall — a decrease they attribute to both an expected sensitivity loss through quantification using MS3 and a reduction in acquisition speed [7]. McAlister and colleagues put the general case the same way: gas-phase manipulations such as MS3 and proton-transfer reactions negate the interfering signals but carry a significant sensitivity penalty [11].
Synchronous precursor selection is the standard answer to that deficit. Using isolation waveforms with multiple frequency notches, McAlister and colleagues co-isolated and co-fragmented multiple MS2 fragment ions, increasing the number of reporter ions in the MS3 spectrum 10-fold over the standard MS3 method; the added signal improved the dynamic range of reporter ion quantitation, reduced reporter ion signal variance and produced more high-quality quantitative measurements [11].
What SPS does not recover is duty cycle. The answer there is to stop acquiring quantitative scans for spectra that will not be identified. Erickson and colleagues implemented a real-time search that scores fragment spectra within milliseconds as they are acquired and triggers the MS3 scan only upon confident peptide identification; benchmarked against standard workflows on a complex two-proteome model of interference and on a whole-proteome 10-plex, where twelve fractions acquired over 36 hours of standard MS3 against 18 hours of real-time-search MS3 gave comprehensive characterisation of the 10-plex proteome in 50% less acquisition time [12].
A different route avoids the third stage entirely. Wühr and colleagues quantify from the complement reporter ion cluster — the fragment ions that carry most of the TMT mass balance — rather than from the low-m/z reporters [13]. Because the location of these ions in the spectrum is precursor-specific, co-eluting peptides do not generally affect the measurement of the cluster of interest, and the method can be implemented on a wide range of high-resolution instruments without an additional ion-purification step [13]. The stated limitation is that the efficiency of complement-ion formation is affected by both peptide sequence and peptide ion charge state [13].
Isotopic impurity, and an artefact of the detector
Reporter ions carry a second error that has nothing to do with co-isolation. The labelling reagents are not isotopically pure, so each reporter ion peak has some contribution from adjacent reporter ions and must be corrected for isotopic overlap using the values reported in the manufacturer's instruction sheet; uncorrected data will appear distorted and confound the observed change in expression levels [1]. Ow and colleagues identified this as one of two factors limiting attainable dynamic range, correctable only given accurate isotopic factors [8]. The correction is a routine feature of quantitation software — Census 2 implements reporter ion impurity correction alongside a reporter ion intensity threshold filter and an option for weighted normalisation to correct mixing errors [14]. Impurity correction unmixes reagent cross-talk between adjacent channels using a supplied matrix; it leaves interference from co-isolated peptides untouched, because that is a different error with a different cause.
Where reporter masses are separated by millidaltons rather than daltons, the measurement can also fail inside the detector. Werner and colleagues found that on Q Exactive Orbitrap instruments the proximate TMT10 reporter ion pairs become prone to coalescence — the fusion of the different reporter ion signals into a single measurable entity — with a detrimental effect on peptide and protein quantification [15]. They established that the main reason was the commonly accepted maximum ion target for MS2 spectra of 1e6, and the artefact was completely removed by lowering the maximum MS2 ion target from 1e6 to 2e5 without any loss in identification depth or quantification quality of proteins [15]. This one is set by an acquisition parameter, not by the chemistry, so no amount of care at the labelling bench prevents it.
Normalisation, and what “relative” means
Reporter intensities are not comparable across channels until the channels are equalised for loading, and not comparable across plexes without a shared anchor. Brenes and colleagues integrated twenty-four 10-plex TMT batches in a single analysis and identified a significant inflation rate of protein missing values as batches were added, a pattern aggravated at the peptide level, and showed that without normalisation strategies to address the batch effects, the high precision of quantitation within a single multiplexed batch is not reproduced when data from multiple batches are integrated [10]. Their structural recommendation is to include a common reference line in every batch, which increases precision by facilitating normalisation across batches, and to design experiments so as to minimise the effect of cross-population reporter ion interference [10].
Accuracy and precision come apart here, and conflating them is the commonest misreading of an isobaric data set. Hogrebe and colleagues, comparing quantification strategies for global phosphoproteomics in a mixed-species comparison with fixed phosphopeptide ratios, found label-free quantification and SILAC to be the most accurate, with MS2-based TMT yielding the highest precision but the lowest accuracy due to ratio compression, which MS3-based TMT can partly rescue [16]. In their biological comparison, however, MS2-based TMT outperformed MS3-based TMT, because its higher precision and larger identification numbers allowed detection of a greater number of significantly regulated phosphopeptides [16]. Which acquisition mode is right therefore depends on whether the experiment needs the size of a change or the fact of one.
Finally, a reporter ratio is relative and confined to its plex: it says a peptide's reporter signal is higher in one channel than in another by some factor, not how much peptide was in either. Ross and colleagues built the bridge to an absolute figure in the original iTRAQ work, using four-fold multiplexing to enable relative protein measurements simultaneously with determination of absolute levels of a target protein using synthetic isobaric peptide standards [3]. That is isotope dilution, and it carries isotope dilution's requirements: the absolute figure is only as good as the assigned amount of the standard, and it is a claim about the peptides you actually spiked, not about every peptide in the plex. The library's guide to stable isotope-labelled internal standards covers where such a standard must enter the workflow and how its assigned amount propagates into systematic error.
Stated plainly, an uncorrected MS2 reporter ratio licenses a directional claim about a peptide within one plex, at a magnitude that is a lower bound on the true difference [1, 7, 9]. It becomes a magnitude claim only after interference has been either excluded by a further gas-phase selection [7, 11, 13] or measured and corrected [9], and after the reagent impurity matrix has been applied [1, 8, 14]. It becomes a between-plex claim only through a common reference channel or an equivalent normalisation [10]. And it becomes an absolute claim only through a separately quantified standard [3].