A peptide map is a proteolysis experiment with a mass spectrometer bolted on the end, and most of what goes wrong with one goes wrong before the digest reaches the column. The protease decides which bonds exist to be observed. The buffer decides which side chains change while the digestion runs. The search parameters decide which of the resulting masses you are prepared to call a peptide. Sequence coverage — the single number most maps get summarised by — sits downstream of all three, which makes it a statement about your experiment at least as much as about the molecule.
What the protease decides before you inject anything
Trypsin is the default because its specificity is narrow enough to serve as a search filter, and that filter does real work. Using sub-parts-per-million mass accuracy to raise confidence in peptide assignment by more than a hundred-fold over typical ion trap experiments, trypsin was shown to cleave solely C-terminal to arginine and lysine; the apparently non-tryptic peptides in the data occurred only as the C-terminal peptides of proteins and as breakup products of fully tryptic peptides N-terminal to an internal proline. Simulating lower mass accuracy on the same data produced a large number of proteins erroneously identified through non-tryptic hits [1].
The proline question is not closed. The long-standing rule that trypsin cleaves after Arg and Lys but not before proline was re-examined against 14.5 million tandem spectra — two orders of magnitude more observed cleavages than the rule was originally built on — and that analysis revealed a surprisingly large number of cleavages before proline, arguing that specificity rules in identification algorithms should be modified to legitimatise them [2]. The two studies read the same neighbourhood differently: one treats sequence around internal proline as a source of breakup products from fully tryptic peptides, the other as genuine proteolysis. For your map the practical consequence is identical. If the search rule is stricter than the enzyme, peptides spanning Lys-Pro and Arg-Pro bonds are absent from the output, and the hole reads as a coverage problem when it is a parameter problem. Re-searching a subset with the restriction relaxed costs an afternoon and settles which it is.
Choosing a protease for coverage rather than for convenience
Trypsin's convenience is also its ceiling. In yeast, digesting with five proteases in parallel — trypsin, LysC, ArgC, AspN and GluC — and tailoring the MS2 method to the precursor by a decision tree yielded 92,095 unique peptides mapping to 3,908 proteins at 1% false discovery rate, against 27,822 unique peptides and 3,313 proteins from trypsin alone. The additional 595 proteins were mainly low-abundance species below roughly 1,000 copies per cell, and sequence coverage for those proteins improved nearly three-fold. The authors' conclusion is the operationally useful one: large portions of a proteome are simply inaccessible after digestion with a single protease, and multiple proteases rather than technical replicates are the direct route to more coverage [3].
A systematic version of the same experiment ran seven commercially available enzymes singly and in double and triple combinations across the HeLa proteome, 48 digests in total. Analysing the unfractionated lysate digests directly detected 5,223 proteins at 42% mean sequence coverage. The reasons tryptic coverage runs short are worth memorising because they are the reasons your gaps exist: peptide length, ionisation efficiency, and co-localisation of post-translational modifications. From the strong-anion-exchange fractionated data, three or more proteotypic SRM candidates were found within a single digest for 6,056 genes, and in 25% of those cases the best digest was neither trypsin nor Lys-C [4].
Single-protein maps sharpen the point. A multiple-parallel-protease strategy using five proteases with high-resolution MS consistently achieved 100% sequence coverage in maps of two monoclonal antibodies alongside detection of their modifications [5]. Coverage can also be pushed by pairing enzymes that cut the same residues from opposite sides: Tryp-N, which cleaves N-terminal to lysine and arginine, gave coverages comparable to trypsin across four monoclonal antibodies, with the combination slightly improving overall coverage, and the two enzymes generated identical peptide sequences apart from the terminal ones [6]. Choose the second enzyme against the specific region you cannot see, not from habit.
Missed cleavage and non-specific cleavage are competing errors
Missed cleavages are not stochastic noise. A predictor built on information theory identified experimentally defined missed cleavages with up to 90% accuracy from amino acid sequence alone, across a peptide mass fingerprinting dataset and a tandem MS dataset; masking the confidently predicted sites in the searched database improved identification with two different search engines [7]. If missed cleavage is that predictable, an unexpected missed-cleavage peptide is information about your digest, not a nuisance to be absorbed by widening the search.
Digestion architecture moves the number. A quantitative comparison of in-solution protocols on yeast lysate — double tryptic, surfactant-assisted, and tandem Lys-C then trypsin — found the tandem Lys-C/trypsin scheme most efficient at yielding fully cleaved peptides while reducing the abundance of miscleaved peptides, with the improvement concentrated in miscleaved stretches flanked by charged basic and particularly acidic residues [8]. That sequence-context detail is the useful part: if your missed cleavages cluster around acidic neighbours, the fix is the protocol, not longer incubation.
The reflex response to missed cleavage — digest longer, digest hotter — buys coverage in one currency and spends it in another. Removing guanidine, a known trypsin inhibitor, from the digestion buffer allowed near-complete digestion of reduced and alkylated immunoglobulin gamma in 30 minutes, and the short digest gave cleaner tryptic maps precisely because there was less trypsin self-digestion and fewer non-specific cleavages [9]. Non-specific cleavage costs twice. It puts peptides in the map at termini you did not predict, and the usual remedy — opening the search to semi-specific or unspecific termini — expands the search space so that marginal spectra acquire plausible matches. That is the same failure mode the mass-accuracy simulation reproduced deliberately [1]. Treat semi-specific searching as a diagnostic run on a subset, then go back and fix the digest.
Sequence coverage is a figure about the experiment
A gap in a map licenses exactly one statement: no peptide from that stretch was identified under those conditions. It is not evidence that the region is absent, that it is intact, or that it carries no modification. The mechanisms behind the gap are mundane and enumerable — the tryptic peptide from that stretch is too short or too long to be selected and sequenced, it ionises poorly, or a modification on it moves its mass away from the unmodified form the search expected and splits its signal across forms [4]. None of those mechanisms is informative about the molecule; all of them are informative about the method.
Two consequences follow. First, coverage percentages are only comparable between analyses that used the same enzyme set, the same acquisition method and the same identification criteria, since each of those changes the denominator of what was observable at all [3, 5]. A map reported at 95% with one protease and one at 95% from a five-protease composite are not the same measurement. Second, a coverage percentage says nothing about depth at any given residue. A residue covered by a single peptide identified from a single spectrum and a residue covered by overlapping peptides from two enzymes producing complementary ion series [6] both count once toward the percentage. If a specific residue matters — a suspected modification site, an engineered substitution — the number to report is the evidence at that residue, not the coverage of the whole sequence.
The modifications your own preparation makes
Trypsin digestion can induce artificial asparagine deamidation and N-terminal glutamine cyclisation because of the temperature and the alkaline buffers used, and the amount of these artefacts is directly proportional to the incubation time in the reduction and alkylation buffer and, more importantly, in the digestion buffer where the peptides are completely solvent exposed [9]. That last clause is the one to act on: exposure time after the protein has been unfolded and cut is where the artefact accumulates fastest, so a protocol that reaches complete digestion sooner is chemically safer, not just quicker.
The stronger intervention is to move the chemistry rather than shorten it. A platform peptide mapping method combining a low-pH-resistant Lys-C with modified trypsin was developed to maintain succinimide stability and eliminate the deamidation assay artefact while achieving digestion efficiency equivalent to a conventional tryptic map run under alkaline conditions [10]. If your assay question is deamidation, running the digestion at the pH that drives deamidation is a design flaw, however well controlled.
Oxidation is the artefact that behaves least predictably. In a tryptic-digest, high-resolution LC-MS workflow that carried a frozen reference standard in every run for system suitability and longitudinal trending, a seven-year record showed apparent total methionine oxidation in an Fc-fusion protein reference standard rising abruptly and staying elevated, up to roughly five-fold; the shift was traced to one specific bag of microcentrifuge tubes used during buffer exchange and resolved when those tubes were discontinued. In other antibodies, variability in observed methionine oxidation was attributed to on-column oxidation, which produced a broad, noisy peak that interfered with automated integration and was reduced by EDTA flushing, implicating exposure to metal ions [11]. Neither cause is visible in the chromatogram as a labelled error; both are visible in a reference standard trended over time.
Carbamylation is the artefact you introduce with the denaturant. In aqueous solution urea is in equilibrium with ammonium and isocyanate, and under alkaline conditions the isocyanate reacts with primary amines at free N-termini and with the ε-amino groups of lysines. The process is relatively slow and accelerated at elevated temperature, so prolonged incubation in urea buffers induces carbamylation that hampers tryptic digestion and peptide identification and interferes with isobaric labelling. Measured under common preparation conditions, carbamidomethylation in the presence of 8.0 M urea induced carbamylation of 17% of N-termini and 4% of lysine residues [12]. A carbamylated lysine is also a blocked cleavage site, so the same reaction that adds a mass shift adds missed cleavages, and the two artefacts appear together.
Controls that separate a real modification from one you made
The general shape of the control is to make the artefact isotopically distinguishable from the finding. Preparing the sample in H218O does this for deamidation: formation of isoAsp or Asp during sample preparation incorporates 18O into the side chain for a 3 Da increase, whereas deamidation that was already present gives 1 Da. The complication is that up to two 18O atoms can also be incorporated into the peptide C-terminal carboxyl group during enzymatic digestion, so the intact-peptide mass difference is only interpretable when a fixed number of 18O atoms is incorporated; the reported procedure therefore calculates the split from b ions, which do not carry the C-terminus, and is presented as applicable under general peptide mapping conditions [13].
Where an isotopic control is not available, the substitutes are procedural: an incubation-matched blank that goes through every buffer step, a digest of the same material prepared by the artefact-suppressing chemistry for comparison [10], and a reference standard carried in every run and trended, which is what exposed the tube-borne oxidation above [11]. A modification reported without at least one of these is a modification whose provenance is unknown, and the direction of the bias is always the same — toward overestimating what is on the molecule.
Three claims, not one
Peptide mapping is routinely used for three distinct purposes, and a map that supports one does not automatically support the others. Identity confirmation asks whether the observed peptide masses and fragment spectra are consistent with an expected sequence; it tolerates gaps, because a confident subset can be diagnostic. Coverage mapping asks what fraction of the sequence was observed at all, and as above it is conditional on the enzyme set and acquisition method. Modification site assignment is the strictest of the three and the one most often overclaimed.
Search engines are relatively effective at identifying modified peptides with a defined measure of reliability, but their localisation of the site or sites of modification is often arbitrary and unreliable, and the field has lacked a widely accepted false localisation rate metric that describes the certainty of site assignment in published datasets [14]. In practice a modified peptide identification and a localised modification site are two different results with two different error rates, and reporting the first as the second is the commonest overreach in a peptide map. Localisation improves when the spectrum actually contains ions that discriminate between candidate sites, which is one reason pairing enzymes that place a basic residue at opposite ends of the same sequence is useful: the complementary ion series obtained that way have been used for precise localisation of deamidation and oxidation sites and of single amino acid variants [6].
Reproducibility across analysts
Two layers of variability sit on top of the chemistry. The digestion itself is labour-intensive and challenging to reproduce accurately; a workflow using immobilised trypsin on magnetic beads was built specifically to address that and was run at four independent sites, with data showing robust digestion efficiency and low levels of protocol-induced modifications [15]. Automating the step that generates the artefact is a more durable answer than training analysts to hurry.
The acquisition layer is stochastic in a way no amount of care removes. Across 144 LC-MS/MS experiments on eight instruments, peptide lists from pairs of technical replicates overlapped by only 35–60%, and reproducibility between different instruments of the same model lagged single-instrument repeatability by several percent; sample complexity did not appear to affect peptide identification repeatability. The most repeatable peptides were those at conventional tryptic cleavage sites, those producing intense signal, and those from proteins yielding many distinct peptides [16]. That overlap figure is the reason a single injection is a weak basis for declaring a region uncovered, and the reason a map intended to be compared across analysts needs replicate injections and a fixed inclusion strategy rather than data-dependent sampling alone.
The discipline that follows is unglamorous. State the enzyme, buffer, pH, temperature and digestion time alongside every coverage figure; report modification levels with the control that bounds the artefact; distinguish in writing between a peptide identified, a residue covered and a site localised. A peptide map that carries those three qualifications is reproducible. One that reports a coverage percentage and a list of modification percentages is a number that another analyst cannot reconstruct.