Counting how many disulfides a peptide carries and saying which cysteines are joined are different measurements, and the second one has a problem the first does not: nearly every step that produces the answer can also rearrange the bonds it is meant to report. Thiol-disulfide exchange proceeds through the thiolate anion, so any stage held near or above neutral pH gives a native linkage a route to become a non-native one, and the product is indistinguishable by mass from a non-native linkage that was already in the vial. That ambiguity is stated directly in the primary literature: it is not easy to differentiate whether scrambled disulfide linkages are initiated by the sample itself or induced during the protease digestion process [1]. An assignment is therefore never one experiment. It is an experiment plus the controls that bound its own artefact rate.
The arithmetic that flags a linked pair
Non-reducing peptide mapping is a comparison of two digests of the same material. The reduced-and-alkylated digest gives the inventory of linear peptides the enzyme produces. The non-reduced digest, run on material that has never seen a reducing agent, contains those same peptides plus composite species in which two chains are still held together.
The mass bookkeeping is stoichiometric. Forming a disulfide from two thiols removes two hydrogen atoms, so a species built from peptides P1 and P2 and one interchain bond has monoisotopic mass M1 + M2 - 2.01565 Da. A single peptide closed by an intrachain bond sits 2.01565 Da below its fully reduced form. Three peptides held by two bonds sit at M1 + M2 + M3 - 4.03130 Da. Searching a non-reduced map means enumerating those sums over every cysteine-containing peptide the enzyme can produce and matching the observed accurate masses against them.
Two consequences follow immediately. First, the oxidised and reduced forms of the same peptide differ by 2.01565 Da, while the second isotope peak of the oxidised form sits 2.00671 Da above its monoisotopic peak. The gap is 0.00894 Da, which at m/z 1000 for a singly charged ion is roughly nine parts per million — comfortable on a high-resolution instrument, invisible on a low-resolution one, where a partly reduced species hides inside the isotope envelope of the intact one. Second, a mass match is a shortlist, not an assignment. Peptides of similar composition generate composite masses that coincide, and the number of candidate sums grows steeply with the number of cysteine-containing peptides. The linkage has to be localised, not merely inferred from a precursor mass.
Digestion is where the answer is usually lost
Two failure modes compete, and optimising against one worsens the other. Under non-reducing conditions the substrate is harder to denature and digestion is often incomplete. A workflow developed against that problem denatured the substrate with 8 M urea plus 0 to 1.25 M guanidine hydrochloride at 50 degrees C and then ran a two-step trypsin/Lys-C digestion as a one-pot reaction, completing sample preparation within three hours and, on comparison with a commercial kit using a low-pH digestion condition, showing higher digestion efficiency and shorter preparation time; the work was carried out on monoclonal and bispecific antibodies, and the authors present the method as particularly suited to those antibodies that carry digestion-resistant domains [2]. The opposing failure mode is that the denaturation and digestion conditions which make cleavage efficient are also the conditions under which exchange runs.
The pH is not the only variable, and treating it as such is a common error. In a systematic comparison run on bevacizumab with three enzyme sets followed by dimethyl labelling and mass spectrometry, no disulfide linkage scrambling was detected at pH 6 when Lys-C or trypsin plus Glu-C were used, whereas thermolysin produced identifiable scrambled bonds at pH 5, 6 and 7, with less scrambling at lower pH; all correct linkages were recovered [1]. The enzyme, not just the buffer, sets whether the map is clean.
Pushing the digestion into frank acid removes the thiolate. A workflow built on pepsin, which is active at acidic pH, was validated on a mixture of six proteins rich in natural disulfides and identified 31 unique Cys-Cys bonds out of the 43 present, with no disulfide reshuffling products detected; pepsin outperformed trypsin both in the number of authentic bonds detected and, more importantly to the authors, in preventing the artificially reshuffled bridges formed when digestion runs at neutral pH [3].
That trade is the basis of a dissenting reading worth taking seriously. One group examining non-reduced mapping of monoclonal antibodies describes the blanket avoidance of basic-pH digestion as a common misconception, arguing that the alternative reagents and conditions adopted to prevent scrambling can alleviate it but yield problematic digestion profiles, and offering instead a set of practical measures intended to keep basic-pH digests both accurate and high in quality [4]. The transferable point is not that acid is wrong. It is that the trade between suppression and digest quality has to be measured on the material in hand rather than assumed from the pH alone.
When no protease cuts between the cysteines
Digestion can only distinguish pairings it can physically separate. If two cysteines sit close enough in sequence that no cleavage site falls between them — adjacent cysteines being the limiting case — then every enzyme in the panel returns them on the same peptide, and the map cannot say whether they are joined to each other or to partners elsewhere. Complementary enzyme specificities widen the reach but do not solve this case; a chemical step is needed.
Cyanylation supplies one. Cyanylating a nascent thiol enables chemical cleavage at that cysteine, so the cut is placed by the chemistry at the residue of interest rather than by whatever the protease happens to recognise; the strategy identifies free sulfhydryls and disulfide linkages in the same experiment and has been described as particularly powerful for assigning disulfide bonds in proteins containing adjacent or closely spaced cysteines [5].
Partial reduction with differential alkylation is the other route, and it has been worked out on peptides rather than borrowed from protein practice. For alpha-conotoxin SII, roughly 30 percent of whose mass is cysteine, partial reduction with tris(2-carboxyethyl)phosphine was monitored in real time by direct-infusion electrospray mass spectrometry so that the differentially reduced intermediates could be caught at high yield; those intermediates were alkylated with either N-phenylmaleimide or 4-vinylpyridine, assigned by mass, and their alkylation sites placed by conventional Edman degradation, giving a connectivity of 3-8, 2-18 and 4-14 in residue numbering [6].
The alkylation step can sometimes be dropped. In a method established on ziconotide, a synthetic conotoxin whose six cysteines form three disulfides, partial reduction with tris(2-carboxyethyl)phosphine was followed directly by liquid chromatography tandem mass spectrometry of the products still carrying one disulfide; tandem MS reports the sequence lying outside the surviving ring, which identifies the pair that closes it, and the authors obtained the full connectivity from two partially reduced products [7]. What partial reduction buys in resolution it spends in risk: it is a deliberate, controlled version of exactly the reaction the rest of the workflow is designed to suppress, and the intermediates have to be captured and separated before they equilibrate.
Localising the bond in the tandem spectrum
Fragmentation behaviour differs enough between activation methods that the choice determines what can be assigned. In a study using a Glu-C duplet peptide from insulin as the model — peptide p1 from the A chain and p2 from the B chain, carrying one interchain bond between them and one intrachain bond within p1 — collision-induced dissociation did not cleave disulfide bonds at all, and electron transfer dissociation cleaved only the interchain bond, at low efficiency; in-source dissociation cleaved the intrachain bond in addition to the backbone, producing the fragment ions needed to assign both linkages, and the same approach was applied to double disulfide linkages in endothelin 3 and relaxin 2 [8].
Electron transfer chemistry on peptides carrying an intrachain bond has been characterised in more detail. Fragments arising inside the cyclic region were observed, including c and z ions along with c-33, z+33, c+32 and z-32 types, indicating cleavage at both the disulfide and the N-C-alpha backbone from a single electron transfer event; direct transfer at the linkage correlated with a signature loss of 33 Da from the charge-reduced precursor. Cleavage at the disulfide was promoted among ions of lower charge state while backbone fragmentation dominated at higher charge states, and across the 13 peptides examined the method returned 75 to 100 percent of possible backbone fragmentation [9]. Charge state is thus a tunable parameter, not an incidental property of the precursor you happened to isolate.
Hybrid activation exploits both. In electron-transfer/higher-energy collision dissociation of disulfide-bridged peptides, the electron transfer step preferentially cleaves the S-S bond to give two intense disulfide-cleaved peptides as primary fragments, while the collisional step fragments unreacted and charge-reduced precursors along the backbone; a dedicated search engine then reports the accurate monoisotopic masses of the two released peptides and the sequence of each by matching the remaining product ions against a linear peptide database [3]. That is the ideal case: the composite precursor is resolved into its two halves and each half is sequenced in one acquisition.
The mechanistic premise underneath it is not settled, and an assignment should not lean on it alone. The proposal that disulfide cleavage is preferred over backbone cleavage in electron capture or transfer dissociation has been debated, and experimental work on the question reports that the mechanism of protein disulfide bond cleavage is more intricate than previously anticipated [10]. Read a missing S-S cleavage as uninformative rather than as evidence that no disulfide was present.
Distinguishing intrachain from interchain
Before it is a fragmentation problem this is a mass and charge problem. Both an intrachain loop and an interchain link cost the same 2.01565 Da set out above, so the mass shift alone cannot say which of the two it is. The awkward case is an interchain bond joining two copies of one sequence, at 2M - 2.01565 Da: that dimer observed at charge 2z falls at an m/z only 1.00783/z from the monomer at charge z — below the fully reduced monomer, and the same distance above the intrachain-looped one — so at z = 2 the separation is about half an m/z unit, and a deconvolution working from an assumed charge state can swap the species. Isotope spacing on the raw spectrum settles the charge, and therefore the species.
Sequence ions do not break that tie either, because the two halves of a homodimeric link generate identical fragment masses; the discriminating evidence is the precursor mass together with its charge. And where both a loop and a bridge sit inside a single digestion product, an activation method that leaves the intrachain bond intact reports only the interchain connection, which is the outcome described above for the insulin duplet [8]. If the digest cannot be pushed to separate the two cysteines of a loop onto different peptides, the loop has to be localised by fragmentation or by chemistry, not by mass alone.
Controls that expose the artefact
The control that matters most is generating the artefact on purpose. One reported strategy exploits intentional disulfide scrambling as a reference tool, used to demonstrate that a proposed preparation is robust [4]. Deliberately scrambled material tells you where the non-native species elute, what they weigh and how your search engine reports them, so a subsequent clean run is a positive statement rather than an absence of evidence.
Detecting low-level scrambling needs its own sensitivity work. Adding tris(2-carboxyethyl)phosphine post-column to the eluent of a non-reduced digest partially reduces the disulfide-linked peptides after separation, so the parent composite and its reduced daughter peptides co-elute and are detected together; combined with glycine addition to enhance signal, the approach was developed to detect, characterise and quantify low-abundance scrambling in monoclonal antibodies [11]. Co-elution of parent and daughters converts a mass coincidence into a linked observation.
The search side needs discipline too, because the candidate space is combinatorial. A dedicated engine for disulfide-linked peptides mapped all native disulfide bonds of a monoclonal antibody and ten standard proteins and, applied at proteome scale, identified 199 disulfide bonds in Escherichia coli and 568 in proteins secreted by human endothelial cells [12]. A single synthetic peptide presents a far smaller search space than that, which is a reason to score it formally rather than to eyeball the masses.
Free thiol as the orthogonal check
A connectivity assignment says nothing about what fraction of the material carries no bond at all, and a free thiol is the substrate for every scrambling reaction downstream. Quantifying it is harder than it looks, and the difficulty is the same exchange chemistry: dissecting the extent of oxidation and reduction of disulfides remains an ongoing challenge in part because of the facility of thiol-disulfide exchange, and several established reagents for the job remain underutilised [13].
The default colorimetric assay has a documented blind spot. Evidence has accumulated that many protein sulfhydryls react incompletely with Ellman's reagent even during prolonged assay times; the kinetic problem was addressed either by including cystamine as a mediator between the protein sulfhydryl and the reagent, or by substituting 4,4'-dithiodipyridine, which reacts quickly with poorly accessible sulfhydryls by virtue of its small size, amphiphilic nature and lack of charge. Both optimised methods reached a detection limit of 0.2 nmol per mL, a determination limit of 0.6 nmol per mL and a 5 min assay time, and agreed with each other in validation [14]. An unexpectedly low free-thiol figure may report a buried thiol rather than a fully oxidised sample.
Mass spectrometry answers the same question with location attached. Labelling intact protein with maleimide-PEG2-biotin detected as little as 2 percent of cysteine residues bearing free sulfhydryls, that is 0.02 mol SH per mol protein; differentially labelling free sulfhydryls with N-ethylmaleimide and disulfide-bonded cysteines with d5-N-ethylmaleimide, then digesting and analysing by nanoLC-MS, quantified free-sulfhydryl abundance at individual cysteine residues down to 2 percent. The workflow had to be optimised specifically to avoid non-specific labelling, disulfide scrambling, and maleimide exchange and hydrolysis, and it was demonstrated on three monoclonal antibody standards and one small protein standard [15]. Residue-level free-thiol data and a connectivity map constrain each other: a cysteine reported as bonded in the map should not also be carrying appreciable free thiol.
What an assignment licenses
An independent route is worth having when the stakes are high, and it does not have to be mass spectrometry. Phenylthiohydantoin derivatives of S-methyl cysteine and S-carbamidomethyl cysteine have been characterised and validated as bioanalytical standards for cysteine detection and quantification by Edman degradation, and used to determine disulfide connectivity in the peptides CCAP-vil and mu-conotoxin KIIIA and in human insulin completely independently of mass spectrometry. The same authors set out the general position plainly: it is still not possible to detect disulfide bonds and the connectivity of multiply bridged peptides directly by a simple and sufficiently validated sequencing or peptide mapping method, because partial or complete reduction and chemical cysteine modification are required as initial steps [16].
That is the honest ceiling on the whole exercise. Every assignment route rests on a chemical perturbation of the bond being measured, so what the result licenses depends entirely on how well that perturbation was controlled. A defensible record therefore states the enzyme and its specificity, the pH, temperature and duration of denaturation and digestion, the alkylating agent and when it was added, the activation method and charge states used for localisation, the free-thiol figure and the method that produced it, and the outcome of a deliberate-scrambling control. Where those are absent, a connectivity written as 1-3, 2-4 is a hypothesis carried over from a homologue, and it should be read as one.