HDX-MS: A Comprehensive Overview of Mechanisms and Applications

Hydrogen-deuterium exchange mass spectrometry, or HDX-MS, is a technique that reveals how proteins move and breathe in solution by tracking which parts of a protein’s backbone swap their hydrogen atoms for deuterium, a heavier form of hydrogen. The method works because exposed or flexible regions of a protein exchange quickly, while buried or rigid regions exchange slowly, and a mass spectrometer can detect the mass increase that comes with each swap. Over roughly five decades of development, HDX-MS has grown from a niche biophysics tool into a workhorse for drug development, structural biology, and quality control of biologic medicines.

How the Exchange Happens

Every protein backbone contains amide hydrogens, one at each peptide bond linking amino acids together. When you place a protein in heavy water (D₂O), those hydrogens can swap out for deuterium atoms from the solvent. The swap is not random. It depends on whether a given amide is exposed to solvent and on the local chemical environment. Neighboring amino acid side chains influence the rate through a combination of electronic effects and physical blocking, and these neighbor effects are surprisingly additive along the chain.¹ The rate at which an unprotected amide exchanges, called the intrinsic exchange rate, can be estimated from the protein’s sequence, the temperature, and the pH of the solution.²

The practical insight here is that a protein in its folded state protects most of its amide hydrogens from solvent. To exchange, an amide must become temporarily exposed through local structural fluctuations or larger-scale unfolding events. The measured exchange rate at any position therefore reports on how dynamic that region is, making HDX-MS a readout of protein motion rather than a static snapshot.

Two Kinetic Regimes That Shape the Data

Protein dynamics in HDX-MS fall into two dominant patterns, known as EX1 and EX2. In the EX2 regime, which is by far the more common situation under normal conditions, a region of the protein flickers open and closed many times before a given amide finally exchanges. Each opening gives the amide a chance, and over many openings, individual amides gradually pick up deuterium at different rates. The mass spectrum shows a single isotope envelope that shifts smoothly to higher mass over time.

In the EX1 regime, the story looks different. Here, once a region opens, exchange happens so fast that every amide in that region swaps before the region closes again. The result in the mass spectrum is a distinctive bimodal pattern: you see one population that has not yet exchanged and another that has fully exchanged, with the relative sizes of the two peaks changing over time.³ EX1 behavior is relatively rare under physiological conditions, but when it does appear, it signals cooperative unfolding of an entire structural element, which can be biologically meaningful. Some researchers deliberately push conditions toward EX1 to study large-scale conformational transitions.

The Back-Exchange Problem

After labeling a protein with deuterium, you need to chop it into peptides and separate them before the mass spectrometer can measure anything. The trouble is that all of those steps happen in regular water, and the deuterium you just incorporated starts swapping back to hydrogen. This “back-exchange” erodes the signal you are trying to measure. Standard HDX-MS setups, which digest the protein and run chromatography at around 0 °C and pH 2.7, can lose more than 30% of the deuterium label within 15 minutes of sample injection.⁴

Several strategies help. One approach adjusts the salt concentration and pH during different stages of sample preparation: higher salt during proteolysis and trapping, then lower salt and pH before the sample enters the mass spectrometer. Optimizing these factors together, along with improvements in peptide detection, has achieved deuterium recovery around 90%.⁵ Another increasingly popular approach is running the chromatographic separation at subzero temperatures. Using antifreeze-like buffer additives such as methanol or ethylene glycol, separations can be performed at temperatures as low as −30 °C with almost no loss of the deuterium label, even during extended chromatographic runs.⁶ When applied to complex samples such as bacterial cell lysates, optimized subzero separations at −10 °C identified roughly three times as many deuterated peptides as a shorter standard-temperature run.⁷

Proteases Beyond Pepsin

Because digestion must happen at low pH and low temperature, the choice of protease is limited. Pepsin has been the default enzyme for decades. It works well at acidic pH, but it has blind spots: it strongly avoids cutting next to histidine, lysine, arginine, and proline residues, which can leave gaps in sequence coverage.⁸

An interesting alternative comes from an unexpected source: carnivorous pitcher plants. Proteases from Nepenthes plants, particularly nepenthesin I and nepenthesin II, thrive at low pH and low temperature, just like pepsin, but they cut at different sites. Nepenthesin II, for instance, cleaves next to basic residues that pepsin cannot touch, and it also cuts next to tryptophan. It is roughly as stable as pepsin once immobilized on a solid support and even more resistant to chemical denaturants and reducing agents.⁹ Nepenthesin I, meanwhile, has different cleavage preferences and also maintains stability when immobilized, surviving exposure to high concentrations of denaturing agents.¹⁰ Using these enzymes alongside or instead of pepsin generates shorter, more complementary peptides that can fill coverage gaps and improve the spatial resolution of HDX-MS data.

Gas-Phase Scrambling and How to Avoid It

Once peptides reach the mass spectrometer, an entirely different problem arises. To figure out exactly where deuterium sits along a peptide, you might try fragmenting the peptide inside the instrument and reading out individual fragment ions. But the energy used for fragmentation can cause hydrogen and deuterium atoms to shuffle around within the peptide before it breaks apart, erasing the site-specific labeling pattern you want to measure. This intramolecular migration is called hydrogen scrambling.

Collisional activation, the most common fragmentation method, causes extensive scrambling in most situations. Studies using selectively labeled peptides have shown that collisional activation essentially randomizes the deuterium distribution, whether the peptide carries a positive or negative charge.¹¹ There is one exception: fragmenting intact protein ions that carry a high charge density produces fragments with little or no scrambling. But when those same proteins carry fewer charges, or when multi-stage fragmentation is used, scrambling becomes severe.¹² Earlier work confirmed a similar trend: high-energy collisional activation on intact protein ions left labeling patterns largely intact, while low-energy, multi-collision fragmentation led to nearly random redistribution.¹³

This is where electron-based fragmentation methods changed the game. Electron transfer dissociation (ETD) breaks peptide bonds through a radical-driven mechanism that is fast enough to outrun scrambling. Using model peptides designed to detect even small amounts of scrambling, researchers showed that ETD in an ion trap preserves the solution-phase deuterium pattern and allows deuterium to be localized to individual residues.¹⁴ The combination of soft ionization conditions with ETD fragmentation has been implemented in automated workflows, achieving what the field calls single-amide resolution.¹⁵ A landmark validation applied this approach to beta-2-microglobulin, comparing deuterium levels at 60 individual backbone amides measured by HDX-ETD-MS against values from NMR spectroscopy. The correlation was excellent, confirming that mass spectrometry can match NMR’s resolution for hydrogen exchange measurements.¹⁶

Software and Statistical Analysis

A typical HDX-MS experiment produces thousands of isotope envelopes across multiple time points and conditions. Extracting meaningful information requires matching each deuterated peptide’s mass envelope to its undeuterated reference, then calculating the mass shift that represents deuterium uptake.¹⁷ Beyond simple centroid-based approaches, methods that analyze the full shape of isotope envelopes combined with kinetic information across a wide time window can extract exchange rates for individual amides, even from peptide-level data.¹⁸

The computational ecosystem has expanded considerably. Open-source tools like DECA handle post-processing tasks including back-exchange correction and statistical testing of differences between conditions.¹⁹ MEMHDX applies a mixed-effects statistical model to each peptide at each time point, accounting for the time dependence of the exchange reaction and the number of replicates, and classifies results using two adjusted p-values: one for whether exchange dynamics changed and one for how large the difference is.²⁰ A recent review catalogued the broader landscape of computational tools, noting that different algorithms now address automated peptide identification, quantification of differences across conditions, deconvolution of bimodal envelopes, and computational strategies to push resolution toward single residues.²¹

Pairing HDX-MS with Molecular Dynamics Simulations

One of the more exciting recent developments is the integration of HDX-MS data with molecular dynamics (MD) simulations. Since HDX-MS measures the average behavior of proteins in solution, it provides a natural experimental benchmark for computer-generated structural models. A method called HDX ensemble reweighting (HDXer) predicts what the HDX data should look like for a given set of simulated structures, then reweights the simulation to better match the experimental measurements. This approach has been extended to model protein-ligand interactions: by combining computational docking, MD simulations, HDXer reweighting, and clustering, researchers can extract high-resolution drug binding poses that agree with the HDX-MS data.²²

A complementary strategy uses HDX-MS data to steer the simulation itself. In one proof of concept with the protein calmodulin, researchers ran adaptive sampling where simulations were scored by how well they matched experimental HDX rates, and the best-scoring trajectories were used to seed the next round of simulations. Over a cumulative three-microsecond trajectory, the guided simulations consistently improved their agreement with experiment and captured domain motions more accurately than unguided simulations.²³

Drug Development and Epitope Mapping

HDX-MS has become a staple in the biopharmaceutical industry, and one of its headline applications is mapping where an antibody grabs onto its target, known as epitope mapping. In a study of a broadly neutralizing influenza antibody, HDX-MS identified two stretches of the hemagglutinin protein that the antibody contacts. Comparison with a high-resolution crystal structure confirmed the result: of 24 amino acids flagged by HDX-MS, 16 were directly involved in contacts visible in the crystal structure. The regions that HDX-MS missed were obscured by sugar modifications on the protein surface, a known limitation with heavily glycosylated targets.²⁴

A refinement of the approach maps not only the epitope (where the antibody binds the antigen) but also the paratope (which parts of the antibody do the binding). In a study of two antibodies targeting the meningococcal antigen NadA, standard epitope mapping showed a clear binding footprint for only one of the two antibodies. But paratope mapping revealed that both antibodies engaged NadA through multiple complementarity-determining regions across both heavy and light chains. The combination of both mapping directions gave a much fuller picture of the interactions than either alone.²⁵

Quality Control of Biosimilars

Biosimilar manufacturers need to demonstrate that their product has the same higher-order structure as the reference biologic. Standard statistical tests designed to find differences are not appropriate here; you need to show equivalence. A method adapted for HDX-MS data uses a two one-sided test (TOST), which asks whether the differences fall within a predefined equivalence margin. Applied to three infliximab biosimilars (Remsima, Renflexis, and Inflectra), all three were found statistically equivalent to the reference product Remicade. In a deliberate negative control, a partially deglycosylated monoclonal antibody was correctly flagged as not equivalent.²⁶ This kind of formal equivalence testing positions HDX-MS as a potential quality-control tool in regulatory filings.

Capturing Receptor Dynamics and Allostery

One of HDX-MS’s greatest strengths is its ability to detect conformational changes that static structural methods miss. A study of the β1-adrenergic receptor, a drug target for heart conditions, found that agonists and antagonists produce clearly different dynamic signatures. All tested antagonists stabilized a specific intracellular loop, while both full and partial agonists did the opposite, increasing dynamics in that same loop. This distinction between drug classes was invisible in crystal structures and cryo-electron microscopy images of the receptor.²⁷

HDX-MS also excels at revealing allosteric communication, where binding at one site changes the behavior at a distant site. Work on the lac repressor, a classic model for gene regulation, showed that the DNA-bound and inducer-bound forms of the protein adopt mutually incompatible low-energy conformational ensembles. Regions far from the inducer binding site showed altered exchange patterns, mapping the path of allosteric signaling through the protein.²⁸

Intrinsically Disordered Proteins

Proteins that lack a fixed three-dimensional structure, called intrinsically disordered proteins, present a particular challenge. Their amide hydrogens exchange so quickly under normal conditions that standard HDX-MS experiments, which typically start at around ten seconds of labeling, cannot capture the action. By lowering the pH of the labeling step (which slows the intrinsic exchange rate), researchers can push the accessible time window into the millisecond range. This strategy was used to map the regions of a disordered coactivator protein (PGC-1α) that become structured upon binding its nuclear receptor partner.²⁹

An alternative approach uses the exchange midpoint of each peptide, relative to a fully deuterated control, as a measure of how fast exchange is occurring. By computing a weighted residue-by-residue average across overlapping peptides, researchers can map residual structure at close to single-residue resolution. This was validated on the p300-binding domain of ACTR, a well-characterized disordered protein, and the secondary structure elements it identified agreed well with prior NMR measurements.³⁰

Membrane Proteins in Lipid Nanodiscs

Membrane proteins are notoriously difficult to study because they normally sit inside a lipid bilayer, and the large amounts of lipid needed to keep them happy overwhelm mass spectrometry. For years, HDX-MS of membrane proteins was limited to detergent-solubilized samples, which do not perfectly mimic the native membrane environment. Lipid nanodiscs solved this problem. These are small, disc-shaped patches of lipid bilayer held together by scaffold proteins, and they hold a single membrane protein in a near-native environment with a well-defined and low ratio of lipid to protein. This makes the samples compatible with HDX-MS and has opened up membrane protein dynamics to routine analysis.³¹

Large Assemblies and Viral Systems

HDX-MS scales surprisingly well to very large molecular machines. Over the past two decades it has been applied to a range of viral proteins, glycoproteins, and complexes, often capturing multiple conformational states of the same system.³² It has also been used to study chromatin remodelers like the ISWI complex, where HDX-MS revealed that the enzyme samples an ensemble of relatively open conformations in its resting state and undergoes major conformational changes in its ATPase domain during active nucleosome sliding.³³

How HDX-MS Compares to Other Footprinting Methods

HDX-MS is not the only mass spectrometry-based method for probing protein structure. Covalent labeling (CL) MS modifies amino acid side chains rather than the backbone, and the two techniques naturally complement each other. The backbone exchange that HDX-MS detects is sensitive to both solvent accessibility and the structural fluctuations of the protein. Covalent labeling, because it operates on a slower timescale, responds only to changes in solvent accessibility and is blind to dynamics. When the two methods are combined on the same protein-ligand system, the paired data more clearly reveal binding sites and distinguish them from allosteric changes distant from the binding site.³⁴

Another complementary technique, hydroxyl radical footprinting (HRF-MS), uses reactive oxygen species to label solvent-exposed side chains. A comparison of structural resolution across many applications found that HRF-MS achieved coverage of about 40% or more of residues in many cases, whereas HDX-MS often reached full structural coverage.³⁵ HDX-MS’s high coverage is one of its defining advantages, though the other methods bring distinct information about side-chain environments that backbone exchange cannot provide.

Microfluidics and Millisecond Time Scales

Conventional HDX-MS setups have a practical lower limit for labeling times. Mixing a protein with D₂O, waiting, and then quenching the reaction by hand takes at least several seconds. That is too slow for studying the fastest exchange events, particularly in loops and other weakly structured regions. Microfluidic devices address this by shrinking the mixing and incubation steps into tiny channels where diffusion distances are short. One early integrated microfluidic platform demonstrated millisecond-to-low-second HDX measurements, successfully characterizing dynamics in weakly structured regions that were invisible in conventional experiments.³⁶

More recent work has pushed the concept further with droplet microfluidics, where the protein sample and D₂O are brought together inside individual picoliter droplets. The high diffusive flux of both D₂O and protons within these tiny droplets enables microsecond mixing, and a surfactant-free droplet merging approach handles the acid quench. Forward exchange measurements on a model peptide established that D₂O labeling itself is the rate-limiting step, setting a practical floor of about ten milliseconds for the shortest useful incubation time.³⁷ Access to these fast timescales matters not only for disordered proteins but for capturing folding intermediates and early binding events that are invisible on the conventional timescale.