A backbone amide hydrogen in a folded protein exchanges with solvent far more slowly than the same amide would in a free peptide, and the size of that slowdown is the measurement. Detailed analysis of most of the backbone amide hydrogens of staphylococcal nuclease, read against the structure, found that hydrogens engaged in hydrogen bonding resist exchange even when they sit adjacent to solvent, and that dynamic structural rearrangement is necessary to disrupt the protecting bonds and permit the solvent exposure that makes a site exchange-competent [1]. The mechanisms found there ranged from complete unfolding of the protein, through rearrangement of regional secondary structure, down to localized fluctuations affecting minimal structural units, and the predominant motion determining any given hydrogen's rate reflected the surrounding structure rather than solvent proximity or bond characteristics [1].
Everything downstream in an HDX-MS experiment is bookkeeping on that fact, plus a deliberate effort to stop the clock before the label leaks back out. This guide covers the exchange chemistry that sets the timescale, the two kinetic regimes and how to tell them apart from the raw isotope envelope, what quench and digestion cost you, and the interpretive step where HDX work most often overreaches.
The chemistry sets the clock
Amide exchange is acid- and base-catalysed, so the rate is a strong function of pH. Chemical exchange rates go through a minimum between pH 2 and 3 and increase by a factor of 10 per pH unit at higher and lower pH [2]. That minimum is what makes the experiment possible at all: the rate for unprotected amide hydrogens at the pH minimum is fairly slow, over 1 h at pH ~2.5 and 0 °C [2]. Label at near-neutral pH, drop the sample to the minimum, and you have on the order of an hour of working time before the information bleeds away.
The unprotected rate is not a single number. Side chains next to an amide change its intrinsic rate through inductive and steric blocking effects, both of which are apparent when all 20 naturally occurring amino acids are studied in dipeptide models [3]. The additivity of nearest-neighbour blocking and inductive effects was tested in oligo- and polypeptides and, in the authors' own phrasing, surprisingly confirmed [3]. The practical consequence is that the intrinsic rate for a given amide is computed from its flanking residues, and any protection factor you report is a ratio against that computed value rather than against something you measured.
Which makes the calibration a live problem rather than settled background. Millisecond HDX-MS on three unstructured peptides — angiotensin, bradykinin and rat atrial natriuretic peptide amide — was compared against theoretical uptake, and using trialanine as the reference compound instead of the commonly used poly-DL-alanine gave experimental rates that agreed much better with predictions and never exceeded the theoretical maximum [4]. Molecular dynamics in the same work indicated significant helical propensity for poly-DL-alanine, with as few as 15 residues sufficient to form stable helices, which makes it unsuitable as an unstructured reference [4]. Reanalysis of previously published data attributed further discrepancies to unquantified salt effects on exchange rates, and the authors state that a comprehensive universal calibration remains elusive [4].
EX1, EX2 and the shape of the isotope envelope
Because exchange requires a transient opening [1], the observed rate depends on how the closing motion competes with the chemical step. Two limiting regimes are conventionally distinguished. In EX2, closing is fast relative to chemistry, many opening and closing cycles pass per successful exchange, and amides in the population exchange independently of one another. In EX1, the chemical step outruns closing, so a single opening event exchanges everything it exposes and the population splits into molecules that have experienced the event and molecules that have not.
That distinction is visible in raw spectra, not only in fits. The hallmark bimodal isotope pattern observed for EX1 kinetics is distinct from the binomial isotope pattern for uncorrelated exchange under EX2 [5]. Peak width is the practical discriminator: modelling of EX2 widening alongside empirical data from proteins and peptides containing purely EX2 kinetics showed that the widening attributable to stochastic forward and back exchange in a typical experiment is small and can be quantified [5]. Proteins and peptides with both obvious and less obvious EX1 behaviour were then analysed by peak width, which returns the half-life for the cooperative unfolding event and the relative number of residues involved [5].
Two things follow at the bench. A bimodal envelope is a result, and centroiding it produces a number corresponding to no physical state in the sample. And the converse does not hold: a unimodal envelope is not proof of EX2, only that any correlated event present is not resolvable against the widening you already expect from stochastic exchange [5].
Quench, digest, and the deuterium you lose regardless
The peptide-level experiment has had the same shape since it was introduced. Label in D2O, transfer into slow exchange conditions at pH 2–3 and 0 °C, fragment with pepsin, and read the fragments by directly coupled microbore HPLC–mass spectrometry [6]. Applied to horse heart cytochrome c, that first implementation resolved amide exchange rates spanning very rapid, k > 140 h-1, to very slow, k < 0.002 h-1 [6]. In the classic protocol one takes samples in time from the exchanging protein and plunges them into 0 °C solution at pH 2.5 [2].
From the moment of quench, label is leaving. Calculation from the chemical factors leads one to expect a loss of only about 1% of the carried amide D-label per minute, so a final recovery of D-label in the range of 70 to 80% can be expected [2]. Measured workflows sit near that on average but hide a wide spread across peptides. Preparing a maximally deuterated control — the same protein deuterated to completion and carried through the identical buffer, quench, chromatography and MS conditions — is how the loss is quantified for each peptide separately. A protocol using strong denaturation first, then deuteration, and avoiding high temperature and low pH during the deuteration step to limit aggregation, produces such a control in roughly 40 minutes for non-membrane proteins and gave relative fractional uptake and back-exchange values equivalent to slower alternatives [7]. Across the proteins tested, average back-exchange spanned 25 to 35%; within myoglobin, per-peptide values ranged from around 20% in the peptide showing the lowest loss of deuterons to around 55% in a peptide containing three histidines, and myoglobin showed intermediate levels of 35.4 ± 0.89% [7]. A single global correction factor therefore misstates individual peptides in both directions.
The other lever is the separation itself. A UPLC system built for HDX achieved near-baseline resolution in 6-minute separations at 0 °C for proteins exceeding 95 kDa individually and 250 kDa in mixtures [8]. Against a loss of roughly one percent of label per minute [2], minutes saved on the gradient are label retained.
Digestion and the resolution you actually have
Pepsin does the digestion because aspartic proteases are active at low pH and therefore work under hydrogen exchange quench conditions [9]. Pepsin is less specific than other endoproteinases [9], and that low specificity is useful rather than merely tolerated, because different peptides can enhance protein sequence coverage and improve the spatial resolution of HDX-MS data [9]. It is not random, though. Across more than 30 replicate digestions of the same protein, the number of reproducible peptides produced during pepsin digestion becomes constant above 5–6 replicate digestions [9]. A comparison of porcine pepsin with aspergillopepsin and a new aspartic protease from the stomach of the rice field eel found unique cleavage specificity for the eel enzyme at arginine, asparagine and glycine, and the different peptides produced by the various proteases can enhance protein sequence coverage and improve the spatial resolution of HDX-MS data [9]. If a region of interest is bare, changing protease is a real option; adding replicates past six is not.
Coverage still is not guaranteed. An HDX-MS protocol for apolipoprotein-D had to solve high protein rigidity and low pepsin cleavage using rigorous quenching conditions and longer deuteration times, and reached 85% sequence coverage with 50% deuterium exchange [10]. The uncovered fraction contributes nothing to the result, and its silence is not evidence about that region's dynamics.
Within a peptide there is no resolution at all: the measured mass is a sum over that peptide's exchangeable amides. Localization by enzymatic cleavage typically achieves single-residue resolution for only a few residues [11]. Gas-phase fragmentation is the route to finer localization and carries its own failure mode. A set of peptides with unique regioselective labelling, designed for high sensitivity to scrambling, showed that the selective labelling is completely erased upon low-energy collisional activation in a tandem mass spectrometry experiment as a result of extensive hydrogen scrambling [11]. Any residue-level assignment from a fragmentation method needs accompanying evidence that scrambling was suppressed under the exact conditions used.
Uptake plots and the significance question
The standard presentation is deuterium uptake against labelling time, one curve per peptide per state, with a difference plot summarising state-to-state changes along the sequence. To extract a biophysically meaningful value for a perturbation from medium-resolution data there are two main routes: estimating the area between the two deuterium buildup curves using centroid values with and without the perturbation when plotted against log time scale, or dissecting the data into multiple single-exponential curves using the isotope envelopes [12]. The log time axis is not cosmetic. A single fixed timepoint samples one slice of a process whose informative features may lie well away from it in time, and both time window coverage and time window resolution are named as things to improve deliberately during the HDX stage rather than accept by default [12].
Deciding which differences are real is a multiplicity problem, since a typical experiment produces many differential values across many peptides monitored at several labelling times. Using null measurements on maltose binding protein and a monoclonal antibody, a hybrid statistical analysis based on volcano plots, combining individual significance testing with an estimated global significance limit, was shown to simultaneously decrease the risk of false positives and retain superior power [13]. What travels with the plot matters as much as the plot. Community recommendations for HDX-MS data collection, analysis, presentation and interpretation were assembled out of discussions emerging from the first International Conference on Hydrogen-Exchange Mass Spectrometry, and are meant to represent a consensus viewpoint on the best practices whose adoption would make the technique accessible to non-specialists [14].
What a protection difference licenses
HDX-MS data provide not structural information but dynamic information [15]. In an analysis correlating X-ray crystallographic parameters with protection factors in cytochrome c, hydrogen bonding was found to significantly influence HDX rates representing protein dynamics, while solvent accessibility has weaker effects [15]. So a peptide that takes up less deuterium in the presence of a ligand has become, over your labelling window, less able to open. It has not necessarily become less exposed, and it is not necessarily touching anything.
The consequence is stated directly in the same analysis: since HDX-MS identifies the change in dynamics upon binding, it can detect allosteric effects upon the binding in addition to the direct binding effects, and this character could be a disadvantage to identify the direct binding site but could be an advantage to understand the energy flow of the binding event [15]. A protection map is a map of where the energetics changed. Which part of that map is the interface is a different question, and one HDX does not answer on its own.
The apolipoprotein-D work shows both signals in a single dataset. Binding of a small hydrophobic ligand was detected as a significant reduction in deuterium exchange in eight peptides, and the stabilisation of dynamics was interpreted as a combined orthosteric effect in the ligand binding pocket and allosteric effect at the N-terminus and C-terminus, mapping effects relayed to distal parts of the protein [10]. Crystallography of the recombinant protein had demonstrated no major conformational changes upon binding of the same ligand [10]. That pairing is the honest version of the result: HDX registered something the static structure did not, and part of what it registered was not the pocket.
The failure modes run in both directions and have been enumerated: false negatives due to sequence coverage, false negatives due to time window coverage, false positives due to sequence resolution, and false positives due to allosteric effects [12]. A coverage gap or a badly chosen timepoint set hides a real difference. A long peptide spanning one changed and one unchanged element averages them into an ambiguous middle value, which is the sequence-resolution failure. And a reproducible, statistically significant protection can sit well away from the ligand. A further complication is that ligand binding changes may be complicated by the presence of unbound protein [15], so incomplete saturation can attenuate differences across the dataset by a factor you have not measured.
What is not settled
Absolute protection factors are still not routine output. Establishing HDX-MS as an absolute and quantitative rather than a relative and qualitative measurement is stated as an aim, and the recalibration work that brought unstructured-peptide rates into much better agreement with prediction also concluded that a comprehensive universal calibration remains elusive, with unquantified salt effects among the contributions [4]. Until that closes, most HDX-MS output is comparative by construction: this state against that state, same instrument, same session, same peptide list.
Nor is there a general rule for converting a protection difference into a contact. In the worked example above, the split between orthosteric and allosteric contributions was drawn using position relative to a binding pocket already known from crystallography [10] — a resource you do not have when the point of the experiment is to find the pocket. What the measurement licenses is a statement about where backbone dynamics changed, at peptide resolution, over a defined labelling window, under one set of solution conditions, corrected by a back-exchange control you should be able to show. Turning that into a binding site takes a different experiment: mutagenesis, a structure, or a competition measurement that tests contact rather than dynamics.