One of the earliest detailed structural studies about antigenicity of proteins (and one of the first protein-protein interaction studies) was carried out with Sperm whale myoglobin because its structure had become available (Atassi, 1964). The 3D structure of an antigen (myoglobin) allows structural rationalization of epitopes. Furthermore, an antibody raised against an antigen is not effective against homologous antigens with slightly different epitopes, suggesting that antibodies bind conformational epitopes (Benjamin et al., 1984). Owing to lack of high-resolution techniques to study large proteins, the antigen was proteolyzed, and those fragments of the antigen that bind were identified (Landsteiner, 1942; Atassi, 1964) (Figure 3).

As peptide-library synthesis became more feasible, scanning of many peptides having overlapping sequences (PEPSCAN) became an important tool for identifying epitopes (Geysen et al., 1984). Often, an epitope identified by this approach is validated by site-directed mutagenesis. During 1990s, the phage display library method for epitope mapping was introduced (Cwirla et al., 1990; Devlin et al., 1990; Scott and Smith, 1990). With this approach, each bacteriophage displays unique/overlapping sequences of an antigen, and by affinity selecting the phages and identifying the sequence of “displayed peptide,” an investigator could locate the epitope for a given antibody. Theoretically, the phage-display library approach has circumvented, by clever genetics, the enormous task of generating peptide libraries. An advantage of a library-based approach, however, is that once the library is constructed, one can use it for screening epitopes for many antibodies. Because small peptides are synthesized or displayed on the surface, only linear epitope mapping is possible. Recently, the high throughput phage display (Phage-based Representation OF ImmunoLigand Epitope Repertoire (PROFILER) method was introduced where large segments of antigens can be displayed to reveal potentially conformational epitopes (Cariccio et al., 2016).

In the 1980s, chemical modification of a protein was also applied to map an epitope on an antigen (Burnens et al., 1987) (Figure 3). An antigen was modified by selective labeling in the presence and absence of an antibody. Since binding will sterically hinder the amino acid labeling involved at epitope interface, comparing the antigen labeling in the absence and presence of the antibody can reveal the epitope (this approach is a predecessor of modern MS-based covalent footprinting). The differential labeling was initially analyzed by both High Performance Liquid Chromatography (HPLC) and sequencing of the labeled peptide to identify the epitope. This simple and elegant approach of differential labeling was later combined with MS-based detection (more below) of a differentially labeled peptide by Przybylski and coworkers (Fiedler et al., 1998) and in a modern day version by Gross and coworkers (Jones et al., 2011). Because the turnaround is rapid, the population-level information is accurate, and the assignment of peptide sequences can be nearly unambiguous (by MS/MS), such differential labeling (footprinting) is now a pillar of MS-based protein/protein interaction studies, especially epitope mapping (Liu et al., 2020).

Limited proteolysis or proteolytic footprinting is widely used for epitope mapping. One of the seminal observations that opened the door for limited proteolysis is that antibody-bound regions under native conditions are resistant to proteolytic cleavage. This property is exploited in proteolytic footprinting of antibody-antigen complexes (immune complexes). Upon immune complex formation, that part of the antigen interacting with the antibody will bury the cleavage site or stabilize it, making it inaccessible or less reactive to the protease. By examining the differential nature of the resultant peptides, an investigator can identify the potential epitope (Jemmerson and Paterson, 1986). Very early on, such differential protease digestions were analyzed by HPLC-based separation followed by peptide-sequence analysis. The introduction of limited proteolysis to epitope mapping was a major milestone in the field, and the approach also unravels discontinuous epitopes.

The next breakthrough was the coupling of limited proteolysis with MS, called “epitope excision” by Przybylski and coworkers (Suckau et al., 1990). In this method, proteolysis of the free antigen or antibody-bound antigen, was followed first by plasma desorption MS and later by MALDI MS (Papac et al., 1994; Zhao and Chalt, 1994). This was significant because MS provided rapid turnaround along with nearly unambiguous mass and sequence confirmation of peptides. Although this approach was initially used for linear epitope mapping, as MS instrumentation became more sophisticated, this approach was quickly adapted to proteins ranging from small globular proteins to large glycosylated viral antigens (Parker et al., 1996; Jeyarajah et al., 1998). Around the same time, a similar approach was introduced called “epitope extraction” by Chait and coworkers (Zhao and Chalt, 1994) where the full antigen is digested by a protease, and those peptides that bind are allowed to form an immune complex. Non-interacting peptides are washed off, and the interacting peptides released from the complex are analyzed quickly by MALDI MS. This approach was later extended to affinity proteomics where the antigen was either separated by using the antibody of interest and then digested, or a complex cellular matrix containing the antigen was digested and bound peptides were extracted and analyzed (Yu et al., 1998; Peter and Tomer, 2001).

Contemporaneously, chemical crosslinking coupled with MS was introduced by Zenobi and coworkers (Pimenova et al., 2008) to show that the combination has potential for epitope mapping. The epitope extraction method was also used in conjunction with native MS whereby the mass of the peptide bound in the antibody complex was analyzed under native conditions, and from the increase in mass, one can infer the peptide location and sequence (Lu et al., 2009). Very recently, another version of peptide extraction under native conditions was proposed by Glocker and coworkers (Yefremova et al., 2017) where bound peptides were ejected by collision energy ramping to distinguish the relative binding affinities of closely related peptides. The current status of excision and extraction methods for epitope mapping was recently summarized by Glocker and coworkers (Opuni et al., 2018).

Proteins possess dynamic personalities, and even under native conditions, they sample a wide variety of open confirmations (Henzler-Wildman and Kern, 2007). Conformational opening involves breaking and forming (polar) hydrogen bonds. A proton is mobile, and during structural opening of H-bonding, a “new” proton from the solvent usually replaces it (Eigen, 1964; Hvidt and Nielsen, 1966). This phenomenon is utilized in HDX. By diluting the protein (ideally >10 fold) with D₂O, the reformation of an H-bond will occur mostly with deuterium (forward exchange). As HDX occurs, the probability of finding protons in the protein diminishes rapidly. In addition to amide hydrogens on the backbone, those attached to electronegative oxygen (O), sulfur (S), and nitrogen (N) atoms in the side chains are also involved in reversible hydrogen bonding and undergo exchange. Because peptide-bond hydrogens undergo H-bonding and in enol-keto isomerization, the backbone amide hydrogens are more stable and slower to exchange than the side-chain H-bonds.

The HDX reaction is pH- and temperature-dependent. At physiological pH, exchange is OD⁻ catalyzed, and at low pH, it is D₃O⁺ catalyzed. At low temperature (≤0°C) and pH −2.6 HDX, rates are minimum, and, hence, these conditions are utilized to quench HDX at specific times and minimize deuterium back exchange (exchange out those deuteriums that had exchanged in the protein with solvent protons) during LC-MS (Bai et al., 1993). In the back exchange occurring after quench and before detection, the side-chain Hs back exchange quickly, and hence, HDX predominantly measures the backbone dynamics (Skinner et al., 2012).

The ability of HDX to “footprint” backbone amide bonds of different open conformations makes HDX one of the most valuable techniques to study both protein dynamics (James et al., 2022) (Jethva and Udgaonkar, 2017; 2018) and changes accompanying protein-protein interactions (e.g., epitope mapping). Apart from structural aspects of protein dynamics that affect the exchange reaction, there are other physicochemical factors (e.g., pH, temperature, and ionic strength) that influence HDX. Thus, it is important to control carefully the physicochemical parameters for comparative analysis of protein dynamics as in epitope mapping (Hamuro, 2021).

Hydrogen and deuterium differ in several physicochemical properties including density (Hvidt and LinderstrØM-Lang, 1954), infrared absorbance (Haggis, 1957; Blout et al., 1961; Goormaghtigh, 2013), NMR absorbance (Wishnia and Saunders, 1962; Otting and Wüthrich, 1990; Paterson et al., 1990; Englander and Mayne, 1992), radioactivity (for hydrogen tritium exchange, HTX) (Leach and Springell, 1962; Englander, 1963; Ashman et al., 1971), UV absorbance (Englander et al., 1979), neutron diffraction (Kossiakoff, 1982), and molecular weight. All these properties can be exploited to follow the extent of HDX. The first use of MS was to measure the changes in exchanged water (Nabedryk-Viala et al., 1976), but subsequently MS was utilized for monitoring the exchange in peptides in a bottom-up analysis (reconstructing the protein level HDX behavior by looking at the peptide level exchange behavior) (Sethi et al., 1983; Verma et al., 1986; Katta and Chait, 1991; Mandell et al., 1998). Currently, MS analysis of exchanged peptides is a mainstream methodology because it affords medium structural resolution, high sensitivity, and practically no size limit for the target protein (antigen) (Engen and Komives, 2020; James et al., 2022). Thus, locating an epitope map is most often met by using a bottom-up HDX-MS approach (Figure 4).

Like HDX, the sister technique of HTX, pioneered by Englander and coworkers (Englander, 1963), was among the first to be deployed for antigen-antibody interface analysis (Ashman et al., 1971; Liberti et al., 1972; Liberti et al., 1981). As other detection methods evolved, HDX was coupled with infrared spectroscopy (Závodszky et al., 1981), subsequently with NMR for epitope mapping (Paterson et al., 1990). Such studies brilliantly elucidated the role HDX can play for epitope and paratope mapping to reveal binding-induced conformational changes (Benjamin et al., 1992; Mayne et al., 1992). Although NMR provides atomic level details of the epitope map, the relatively high concentrations of protein required (often mM concentration) and the size of antigen that can be studied (<30 kDa) are shortcomings. Nevertheless, NMR is still used (Bardelli et al., 2015; Simonelli et al., 2018; Di Muzio et al., 2020; Valente and Manzano-Rendeiro, 2021). To address the shortcomings of NMR, investigators turned to MS in the early 1980s with fast atom bombardment (FAB) (Barber et al., 1981) and in the late 1980s with MALDI (Karas and Hillenkamp, 1988; Tanaka et al., 1988) and finally with ESI (Fenn et al., 1989). The first MS application of HDX to intact peptide/proteins was by Katta and Chait (Katta and Chait, 1991) followed by design of the bottom up approach by Zhang and Smith using FAB (Zhang and Smith, 1993), but the complexities of FAB were an obstacle to widespread adoption. In the 2000s, the first HDX in combination with ESI-FTICR-MS (Yamada et al., 2002) and around same time the coupling with MALDI (Baerga-Ortiz et al., 2002) were utilized for the epitope mapping. The compatibility of ESI with online LC systems has made it the most widely used soft ionization approach for HDX. Early epitope mapping studies, using both HDX-NMR and -MS, clearly indicate not only that antigen-antibody bonding primarily stabilizes side-chain interactions but also slows the backbone dynamics, affording a map of an epitope.

Epitope mapping by HDX-MS is comparative in nature, meaning that the HDX kinetics of the unbound antigen is compared with that of the antibody-bound antigen (Figure 4). The first step is to consider the feasibility of HDX for a given epitope mapping problem (Hamuro and Coales, 2018). HDX can be monitored at the intact protein level, but that gives no spatial resolution, motivating protein digestion post HDX and bottom-up MS analysis. As mentioned earlier, HDX, being a reversible footprinting technique, requires monitoring the isotopic exchange via fast gradient LC-MS runs. Quenching the HDX reaction at pH 2.6, approximately 0°C can preserve the labeling pattern for a longer time than at pH 7 and room temperature, hence making possible reliable detection by MS.

Forty years ago, a bottom-up protocol using porcine pepsin (an acid endo-protease) was introduced for improving structural resolution (Rosa and Richards, 1979). Although the field has grown rapidly since, most studies still utilize pepsin (Section 7, meta-analysis of literature) because it is reproducible, active at low pH and at quench conditions involving denaturant and reducing agents (Blumenfeld et al., 1960) and has broad specificity for producing many overlapping peptides (Hamuro et al., 2008; Ahn et al., 2013). The non-specific nature of pepsin also improves spatial resolution and increases confidence because results are recapitulated in several different peptides. Currently, the bottom-up protocol has essentially the same features as 40 years ago; that is, the modern protocol seamlessly interfaces with online digestion, LC, and MS (Zhang and Smith, 1993). The peptides generated by online acid protease digestion are trapped, eluted, separated by using C-18 RP-HPLC, and, after separation, submitted to the mass spectrometer usually operating with ESI.

The bottom up HDX-MS experiment starts with the generation of a peptide map for a given antigen (Figure 4, step 1) in the absence of HDX (unexchanged) where CID-MS/MS helps identify the peptides and determine their retention times. Given the labile nature of H and D positions in peptides, however, CID cannot be used for locating deuterium in the HDX experiment (Section 5.3).

The unexchanged antigen is diluted in H₂O followed by a quench whereby the pH is decreased to approximately 2.6, and the protein injected onto the online protease column where it is digested. The generated peptides are captured on a trap column while salts, unfriendly to the mass spectrometer, are washed off. When an LC gradient is applied, the peptides are shifted to an analytical column where they are separated and transferred online to the mass spectrometer. One way to improve spatial resolution is to tune the conditions of proteolysis (e.g., the time for passage of the protein through protease columns and/or concentration of reducing and denaturing agents) to create many overlapping peptides (see below).

For mapping, the spectrometer is usually run in a Data Dependent Analysis (DDA) mode where ions representing predefined precursors are selected and fragmented individually. The saved data are analyzed using LC-MS/MS data search programs containing the antigen sequence. Typically, peptides that have a precursor mass within a predefined tolerance (set for the MS instrumentation) of an expected mass of a given sequence are accepted but refined in a second pass to include those peptides undergoing reliable fragmentations that are consistent with the sequence (typically 3 to 4 fragments per 10 amino acids) are considered. In a third pass, the above procedure is repeated at least three times to assess the reproducibility of the quench conditions, digestion, gradient, and LC-MS parameters.

Setting up the HDX protocol takes advantage of the unfolding of most proteins at pH 2.6 used for the quench. Often parts of proteins are disulfide-bonded and acid stable, adding difficulty for acid proteases to digest and provide full coverage. In such scenarios, addition of denaturants and reducing agents (e.g., urea, a mild denaturant, guanidinium hydrochloride, a strong denaturant, tris-carboxyethyl-phosphine hydrochloride (TCEP-HCl), an acid compatible reducing agent) or an organic solvent (Guo et al., 2020; Fang et al., 2022) can be explored to achieve good coverage and overlapping peptides. Additionally, the use of “fast” (non-scrambling fragmentation methods) in lieu of CID for the peptides is gaining some traction, giving improvement in spatial resolution of the HDX data (see below).

Once digestion is optimized and good coverage is obtained, the next step is to monitor the HDX kinetics for each peptide (Figure 4, step 2). Typically, a protonated protein is diluted 5 to 10-fold in deuterated buffer of the same pH (ideally of the same composition as well), and HDX is initiated, in the presence and absence of the antibody. At predetermined times, the HDX is quenched by decreasing the pH to 2.6, approximately 0°C followed by online protease digestion (most commonly used) or offline digestion (historically used) (Figure 4, step3). The generated peptides are trapped by using a small C-8/C-18 trap column and then eluted as resolved by the C-18 analytical column and submitted to the mass spectrometer. During HDX-MS analysis, the spectrometer is in the MS-only mode (not MS/MS) to measure the isotope distribution of the peptides. This workflow is repeated as a function of time (from seconds to hours), to cover the time scales of protein (antigen) breathing motions in the presence and absence of the antibody.

The collected HDX kinetic data are analyzed manually by first obtaining extracted ion chromatograms (EIC) for integration or by using sophisticated, automated software for the data analysis (Figure 4, step 4). Such software packages use the LC-MS/MS mapping runs to locate the correct peptide from the unexchanged controls based on retention time, ppm mass error, and expected isotopic distribution. A non-exhaustive list of software includes HXExpress (Weis et al., 2006a), DynamX (Waters, Milford, MA), HDX Workbench (Pascal et al., 2012), HDExaminer (Hamuro et al., 2003), Mass Spec Studio (Raval et al., 2021), Deuteros (Lau et al., 2019), HDX suite from PMI, and others (Wales et al., 2013). Some are open source.

Finally, the differential HDX pattern must be visualized, and this is done by either HDX kinetics plots showing patterns in the absence and presence of antibody (Figure 5A) or in a more condensed differential format such as Woods’ plots (Yamada et al., 2002) (Figure 5B), bar graphs (Figure 5C), heat maps (Zhang et al., 2010), and butterfly plots (Houde et al., 2011). Although each visualization has its own appeal, it is beyond the scope of this review to comment on each. It is the authors’ opinion that kinetic plots are simplest and most informative. As an example, the kinetic plot for the POV4 bound state of peptide 12–21 shows the most protection (decrease in rate of deuterium incorporation upon complex formation) at the earliest time point of HDX, but at the longest time point, the kinetic curve shows HDX similar to that of the unbound state (Figure 5A). Furthermore, peptide 12–24 does not show any merging of the HDX kinetic curves. Both peptides indicate that the 12–21 region remains dynamic, and that the epitope contacts are near residues 22–24. Similarly, for POWV-63 peptides, residues 73–90 and 243–253 show protection. The strong protection and non-converging HDX kinetics indicate that the main epitope contact lies in the region 73–90, and that region 243–253 remains dynamic as compared to the 73–90 region in the immune complex. Figures 5B,C show Woods’ plots and bar graph presentations where cumulative differences across all HDX timepoints (bound-unbound) are plotted along with the propagated error for each peptide (for Woods’ plot global error), respectively. The Woods’ plot is more informative as multiple dimensions of HDX data can be presented including secondary structure, sequence coverage, peptide lengths, and the extent of HDX as a function of sequence. Furthermore, the confidence interval can be color-coded in a fashion that is similar to that in volcano plots.

Most data analysis software packages include data visualization as well, and many dedicated software packages are available (mostly for free). These packages help in data visualization, and they include MSTools (Kavan and Man, 2011), MEMHDX (Hourdel et al., 2016), HDfleX (Seetaloo et al., 2022), HD-eXplosion (Zhang et al., 2021), and others. As with most comparative analysis, HDX differential plots also require statistical analysis, described by (Houde et al., 2011; Masson et al., 2019; Weis et al., 2019).

In the next sections, we organize our review around the use of several improvements to complement/validate observations by HDX-MS and, in that way, provide a current view of the field. Although we center our discussions around epitope mapping, we will show how improvements in the HDX-MS technology overcome past shortcomings of the general HDX-MS methodology.

Very early on, it was realized that digestion of a protein would improve the spatial resolution afforded by HDX. As mentioned earlier, porcine pepsin was introduced 40 years ago for digesting a protein post HDX. Its utility is due to its compatibility with the HDX quench conditions (low pH, presence of denaturants) (Blumenfeld et al., 1960). Subsequently, other proteases including acid proteases from type XIII fungus (Aspergillus saitoi) (Cravello et al., 2003; Englander et al., 2003), type XVIII fungus (Rhizopus species) (Cravello et al., 2003), nepenthesin I and II, acid proteases from a carnivorous plant (Rey et al., 2013), from rice field eel (Monopterus albus Zuiew) (Ahn et al., 2013), Rhizopuspepsin (Rey et al., 2009), and Pepsin A1 and A2 (Brier et al., 2007). Because each protease has broad but different specificities, tandem (Cravello et al., 2003; Mayne et al., 2011; VanBlargan et al., 2021; Doyle et al., 2022), parallel (Nirudodhi et al., 2017), or concurrent digestions (Hamuro and Zhang, 2019; Mullahoo et al., 2020) using multiple enzymes can greatly improve the sequence coverage of hard to digest proteins and increase the number of overlapping peptides.

Marshall and coworkers (Zhang et al., 2011) used for the first time for epitope mapping tandem protease digestion for the large nut allergen protein Ana o 2 (95 kDa) to improve coverage and give many overlapping peptides. Using a combination of an automated HDX platform, tandem protease digestion, and high-resolving power of FT-ICR-MS (Zhang et al., 2008; Kazazic et al., 2011), Marshall and coworkers (Zhang et al., 2011) investigated more than 100 peptides from the antibody-bound complex to find both linear (for 1F5) and conformational (for 2B5) epitopes. It is important to note here that Ana o 2 is a homotrimeric protein (monomer 95 kDa, trimer −285 kDa) in solution. Considering that each homotrimer will bind one antibody (150 kDa), then the solution complex is nearly half a mega Dalton (0.435 MDa).

Any digestion that gives many peptides leads to other problems of chromatographic crowding, spectral overlap, and demanding data analysis. Furthermore, ESI forms multiple charge states for many peptides, adding complexity to the mixture of ions. Especially for epitope mapping, the presence of the large antibody (150 kDa MW) compounds the problem because its digestion affords many more peptides than does the antigen. Hamuro and coworkers (Coales et al., 2009) tried to minimize the chromatographic crowding by doing on-column exchange using an immobilized antibody. The antigen is deuterated in solution and then in the on-column, antibody-bound state, and the antigen is eluted under quench conditions. This approach increases the number of analyzable peptides detected post HDX. Jørgensen and coworkers (Jensen et al., 2013) introduced a modified protocol for affinity separation of biotinylated antibodies post HDX. Such affinity-based antibody separation, however, requires preparation and characterization of an immobilized antibody for its binding. Another difficulty arising from the large numbers of peptides is that deuteration causes each peptide to shift on the m/z scale, adding more signals and complicating interpretation. Although managing those shifts manually is an enormous task, this issue can be addressed in several ways. The easiest is to use automated HDX data analysis software packages (see HDX data analysis Section 4.2). This allows the analysis of hundreds of peptides in a relatively short time. Additionally, software packages also provide statistical analysis and different ways of HDX data visualization.

Englander and coworkers (Casina et al., 2015) used a multiprong strategy for high resolution mapping of epitopes for autobodies targeting MDTCS (Metalloprotease, Disintegrin, TSP1, Cys-rich, and Spacer) regions of ADAMTS13 (a disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13). Inhibition of ADAMTS13 by autobodies leads to a severe pathological condition called acquired thrombotic thrombocytopenic purpura (TTP). Thus, it is important to identify those epitopes that are targeted by autobodies and render ADAMTS13 ineffective. To do this, the authors first used pepsin and fungal type XIII protease digestion to obtain many overlapping fragments (Figure 6A). Following that, the relevant domain (MDTCS) of the larger antigen (ADAMTS13) was used as an antigen. Third, HDX was carried out on-column with immobilized single-chain fragments of the variable regions (scFvs) so that post quench, only the antigen would elute. All these steps reduced chromatographic complexity. Furthermore, to record the isotopic distribution post exchange, the authors used ExMS (Kan et al., 2011) software followed by HDsite (Kan et al., 2013).

For improving the spatial resolution when overlapping peptides are available, most investigators rely on simple subtraction of centroid mass of smaller peptides from that of the larger peptide. This strategy works best when peptides have a common N-terminus. It is often not possible that there is a perfect series of overlapping peptides to obtain a deuterium location with high certainty. As another metric, the isotopic distribution envelope shape is often overlooked in HDX-MS analysis although its broadness can provide a clear signature of nature of exchanging deuteriums (Weis et al., 2006b; Zhang J. et al., 2013).

In another approach, Englander and coworkers (Kan et al., 2013) developed the HDsite algorithm that takes the isotopic distribution post HDX as input and quantifies deuterium retention solely based on envelope shape. By looking at many overlapping peptides (although not perfectly overlapped), one can validate deuterium location in employing an iterative protocol and obtain higher spatial resolution and even residue-resolved deuteration. For example, the effort by England and coworkers (Kan et al., 2013) identified five contiguous loops on ADAMTS13 that were targeted by autobodies indicating a conformational epitope (Figure 6D). Interestingly, for some of the loop regions, residue-level resolution could be obtained to define precise boundaries for the epitope (Figures 6B,C) by using the HDsite algorithm. Epitope validation studies revealed that all five loops are essential for the autobody binding as deletion of any one loop leads to complete abrogation of binding.

The most common way of improving spatial resolution is to use of multiple proteases in tandem. This approach, however, can lead to noisy, overlapped isotopic distributions. At the same time, overlapping peptides may not back exchange similarly owing to sequence, near-neighbor effects, and different responses to the reversed-phase column (Sheff et al., 2013). Such compounding effects can lead to faulty sub-localization of deuteriums based solely on the degree of overlaps. To address this issue, rapid gas-phase fragmentation may be successful.

As mentioned earlier, CID cannot be used because it raises stepwise the vibrational energy of a peptide ion, leading to deuterium scrambling prior to fragmentation (Jørgensen et al., 2005). Electron Capture Dissociation (ECD, using electrons from heated filaments) and Electron Transfer Dissociation (ETD, transferring an electron from a radical anion) give similar outcomes where electron transfer to a protonated peptide leads to rapid (allegedly picosec) fragmentation, thereby preempting scrambling and preserving the sensitive pattern of deuterium labeling (Zubarev et al., 1998; Syka et al., 2004). Although the ETD/ECD approaches have exciting potential, an investigator needs to ensure that scrambling is minimal because “naked” gaseous ions are labile, and even the acceleration and consequent collisions while transferring ions from one instrument compartment to another can lead to redistribution of labels (scrambling) (Demmers et al., 2002; Hoerner et al., 2004; Rand and Jørgensen, 2007; Wang et al., 2019; Wollenberg et al., 2020). Fortunately, there is a well-established protocol available for checking and minimizing scrambling for HDX by using a properly designed standard peptide (Rand and Jørgensen, 2007). ETD/ECD fragmentations can be used in two ways: top-down (where the intact protein is fragmented and its regional behavior studied) and bottom up (Pan et al., 2009; Catherman et al., 2014; Karch et al., 2018). Currently, a bottom-up approach for epitope mapping (see case studies described below) is most often used. Recently, a promising approach, ultraviolet photodissociation (UVPD), was optimized for gas-phase fragmentation that gives minimal H/D scrambling (Mistarz et al., 2018; Modzel et al., 2021).

In another example, Chen and coworkers (Huang et al., 2018) identified mAb1, a strong binder with TL1A, a member of the tumor necrosis factors superfamily. In the presence of a soluble decoy protein, the death domain receptor 3 (Dr3) protein inhibits TL1A-mediated activity. By inhibiting TL1A and Dr3 interactions, the necrosis activity of TL1A can be restored. Competitive binding studies reveal that mAb1 and Dr3 have overlapping binding sites. Because TL1A is a functional trimer (90 kDa MW for monomer) and binds with mAb1 with 3:3 stoichiometry, the functional complex has a MW of ∼0.55 MDa, challenging traditional structural methods. When the authors looked at differential deuterium labeling on TL1A in the presence of mAb1, two regions, encompassing two solvent-exposed loops, and one structured region became most protected upon antibody binding (Huang et al., 2018). Protection in the structured region was hypothesized to be due to binding-induced remote conformational changes. Based on solvent accessibility docking studies, the epitope was localized within two HDX protected loop regions. Because both loop-containing peptides are long (residues 102–116 and 166–180), the investigators used ETD to fragment peptides post HDX in the presence and absence of mAb1 and located the protected sites to be 113–116 and 169–172, indicating that both loops are involved in antibody binding and that only a few residues in the large peptides contact mAb1.

Recently, the same group deployed HDX-MS for identifying epitopes for four antibodies targeting major histocompatibility complex class I chain-related A (MICA) (Huang et al., 2020a). MICA’s interaction with natural killer cells leads to cytotoxicity and antitumor activity. Proteolysis of MICA by tumor-specific proteolysis leads to inhibition of its antitumor activity. Thus, targeting a proteolytic site on MICA by monoclonal antibodies may effectively restore its antitumor activity. Three mAbs, mAb2, mAb39, mAb40, with similar binding constants, target similar epitope regions whereas mAb39, which binds 10 times stronger than the other mAbs, chooses a similar region but a different conformational epitope. HDX footprints indicate that mAb36 binding is stabilized by multiple epitope contacts, which may be responsible for its stronger binding. When two strongly protected peptides in the presence of Fab36 were subjected to ETD post HDX, only one small (8 amino acids) peptide showed complete fragmentation whereas a longer peptide (15 amino acids) did not fragment well under scrambling-minimized ETD. ETD indicates that only two residues (234 and 237) are responsible for protection, pinpointing specific contacts by the antibody.

These studies demonstrate that the spatial resolution can be improved for epitope boundaries by gas-phase, electron-based fragmentation. Further, understanding the extent of scrambling is important for successful application of ETD/ECD methods for epitope mapping. Unfortunately, few peptide fragments are seldom sufficiently long and multiply charged to be amenable to ETD/ECD (Good et al., 2007).

The improvement in LC (UPLC and nanoLC) (Wu et al., 2006) systems and the availability of refrigerated chromatography for HDX (Wales et al., 2008) are pushing HDX-MS into the mainstream structural proteomics. The introduction of a complete automated system enables HDX-MS (Chalmers et al., 2006) to become more routine and be deployed as an early structural biology approach to interrogate solution dynamics of protein complexes. A recent apt example is the use of HDX-MS to characterize epitopes for nine neutralizing antibodies against the receptor-binding domain of SARS CoV2 spike protein to assist the design of antibody cocktails for efficient neutralization of viral infections (Hansen et al., 2020). A study by Weis and collaborators (Toth IV et al., 2017) determined the epitopes of Ricin binding protein A (RiVax) for nine neutralizing and one non-neutralizing antibodies. Based on regional protection found by HDX-MS upon antibody binding, the nine neutralizing antibodies were binned to four clusters.

Similarly, Gross and collaborators (Fernandez et al., 2018; Adhikari et al., 2021) characterized epitopes for five antibodies, and Simmons et al. (Simmons et al., 2021) determined epitopes for six nanobodies. In a recent study, Weis and collaborators (Haque et al., 2022; Haque et al., 2023) characterized epitopes for eleven mAbs against the recombinant outer-surface protein A (OspA) from Borrelia burgdorferi sensu stricto (ss). These studies clearly indicate that HDX-MS can characterize immune responses for vaccine candidates during different stages of development. Automation for HDX-MS (including data analysis) enables these large studies of epitope mapping for several antibodies, bringing HDX-MS more mainstream.

In contrast to running multiple comparative HDX studies, Marshall and collaborators (Zhang Q. et al., 2013) used polyclonal antibodies (pool of antibodies extracted from serum) to characterize the immune response of a nut allergen. Subsequently, others described polyclonal antibodies for study of immune response against tumor necrosis factor α (TNF-α) (Abbott et al., 2013) and factor H-binding protein (fHbp) (Ständer et al., 2021). Further, other investigators compared the immune response for a given antigen in different serum samples (Yang et al., 2016) to evaluate different diagnostic polyclonal antibodies against C-reactive protein, an important biomarker for inflammation (Sun et al., 2022) and recently against Cardiac troponin I (cTnI), a biomarker for acute myocardial infarction (Song et al., 2023). Such studies showcase the exquisite tolerance of HDX-MS for handling molecular heterogeneity, which is difficult with other structural methods.

An obvious way to decrease chromatographic crowding and spectral overlap is to increase the gradient length. Traditionally, the gradient used for chromatographic separation is <15 min to minimize back exchange. The gradient can be lengthened while minimizing back exchange by running the chromatography at low temperatures (e.g., −10°C (Fang et al., 2021), −20°C (Wales et al., 2017), and −30°C (Venable et al., 2012; Anderson and Hudgens, 2022). Fang and others (Fang et al., 2021) showed that a chromatography gradient at −10°C can be completed in 90 min, providing a 3-fold increase in number of peptides identified. Additionally, subzero temperature HDX can be coupled with gas-phase fragmentation (Amon et al., 2012; Pan et al., 2014). Recently, a low-temperature chromatographic set up was characterized for its ability to retain deuterium; at −30°C, back exchange slowed by 40-fold as compared to 0°C. At −20°C, back exchange was ∼10%, and nearly 90% of exchanged deuteriums were retained.

Venable et al. (2012) demonstrated that a −30°C chromatographic set up can be employed to map a linear epitope (3XFLAG-epitope) recombinantly attached to C-terminus of GFP protein. The 3X-FLAG peptide sequence retained ∼2.5 fold more D (close to 90%) at −20°C as compared to chromatography at 0°C (Figure 7A). Additionally, the difference in D retention in the presence and absence of the M2 antibody was approximately 2.5 fold, nearly completely preserving the epitope footprint (Figure 7B).

In another study, Zhang et al. (Zhang et al., 2018) used chromatography at −9°C and increased the gradient length, reduced spectral overlap, and improved overall sequence coverage for the Birch Pollen Allergen. Fang et al. (Fang et al., 2022) used a −10°C chromatographic set up with a 30 min gradient for epitope mapping of Anthrax Protective Antigen and four monoclonal antibodies.

The above studies indicate that lowering the chromatography temperature can significantly expand the time window for the gradient and allow more and larger protein assemblies/antigens to be interrogated for epitope mapping. There is also growing interest in use of Ion mobility (Iacob et al., 2008; Cryar et al., 2017) and capillary electrophoresis (CE) (Black et al., 2015) separation for tackling chromatographic crowding. Wider adoption of automation (Section 7) could further integrate low temperature (∼30°C) chromatography, and with many studies utilizing the Waters Synapt G2 system, it may be possible to see routine incorporation of ion mobility and ETD (for obtaining residue-level resolution) for not only epitope mapping but also for other HDX-MS studies. Ultra-low temperature, however, has its own challenges in terms of back pressure and the need to redesign the chromatographic system. Nevertheless, it remains a promising approach.

Many surface-exposed viral antigens (e.g., SARS CoV2 spike protein) and soluble serum antibodies are rich in disulfide bonds. As described earlier, successful HDX-MS requires good sequence coverage of the antigen. Given that proteases cannot cleave the tightly packed regions of a protein, reduction of disulfide bonds becomes imperative. TCEP-HCl is a widely used reducing agent as it is odorless and works under acidic conditions. For each new protein, the concentration of TCEP-HCl, time, and pH of the quench must be optimized. An important consideration is the reduction potential of TCEP because its reactivity is affected at low pH and temperature. Most investigators add denaturants to facilitate protein unfolding so that efficient reduction can be achieved.

We did a meta-analysis of >80 papers for this review (see below) and found that approximately 60% of the epitope mapping studies use TCEP-HCl denaturant for disulfide bond reduction (Section 7). Some highly disulfide-bonded proteins are difficult to reduce in the short times needed for HDX-MS. Furthermore, the presence of TCEP also hampers the chromatographic and MS performance owing to formation of ion clusters (Zhang et al., 2010; Mysling et al., 2014). To address the shortcomings and “widen the net” for HDX-MS, online electrochemical reduction was implemented by Jørgensen and collaborators (Mysling et al., 2014). Subsequently, an online electrochemical reduction workflow was evaluated by the same group for highly disulfide-bonded proteins under different solvent conditions (Trabjerg et al., 2015), and for epitope mapping (Comamala et al., 2020b; Sheff et al., 2021). An ECD-based approach for improved gas-phase reduction of disulfides gave improved sequence coverage and localization of the deuterium label (Zubarev et al., 1999; Bobst and Kaltashov, 2014). ECD/ETD based approaches, however, still require optimization of instrument conditions such that H/D scrambling is minimal.

Glycosylation (N (Asp)- or O (Ser/Thr)-linked), with its usual high heterogeneity, leads to poor MS signal and inadequate fragmentation, causing incorrect or no identification in HDX of glycosylated proteins. Furthermore, bulky glycans sterically hinder protease digestion, often leading to poor sequence coverage around the glycosylation site. The glycan acetamido group can lead to errors in locating deuterium (Guttman et al., 2011), whereas pre-HDX removal of glycans raises concerns that the native state of protein is not being studied. To address these issues, HDX protocols incorporating N-deglycosylation enzymes such as PNGase A (Jensen et al., 2016; Wagner et al., 2021) and PNGase H+ (acid compatible) (Comamala et al., 2020a) were introduced. Recently, Rand and collaborators (Comamala et al., 2021) integrated online electrochemical reduction along with chip-based “Immobilized Enzyme Reactors (IMERs)” with pepsin and PNGase Dj (newly introduced acid-compatible N-deglycosylation enzyme) for disulfide-bond reduction, digestion, and deglycosylation of peptides for epitope mapping of the Sema domain of c-Met (SD c-Met), a proto-oncogene (refer to figure 2 in original reference Comamala et al., 2021). Using a conventional HDX-MS protocol with TCEP as a reducing agent, the investigators observed only −68% sequence coverage. Missing were cysteine-containing and glycan-containing regions. With the introduction of online EC, the use of pepsin and PNGase Dj chip columns gave >80% sequence coverage, and all the Cys-containing and most of glycan-containing peptides were detected. With better sequence coverage, the investigators assessed the epitope regions and identified the antibody-induced conformational changes on SD c-Met. This study highlights the value of a flexible HDX-MS platform that incorporates modular functionality in tandem for addressing improvements for epitope mapping. Currently, no enzymatic counterpart exists for the removal of O-glycans, and hence, similar strategies cannot be utilized for addressing poor sequence coverage around O-glycosylation sites (Wilkinson and Saldova, 2020). An ETD/ECD based approach was shown to be promising for fragmenting glycosylated peptides, and, hence, scrambling optimized conditions can also be utilized for improving sequence coverage (Mikesh et al., 2006).

One drawback of using HDX for epitope mapping and other protein-ligand interaction studies is the difficulty to distinguish protection induced by binding vs. that due to allosteric or remote conformational changes. Distinguishing these effects becomes more difficult for epitope mapping as epitopes can be conformational. During the equilibration stage, binding-induced structural changes (decreases in solvent accessibility) occur before allosteric changes (Ramirez-Sarmiento and Komives, 2018). Therefore, at short times, protection is seen at a binding site (orthosteric) as well as at distant sites (allosteric sites). Often allosteric protection in HDX can be seen from changes in protection and confirmed by site-directed mutagenesis (Hamuro et al., 2006). If structural details are not available, however, such assignments become tenuous (Sowole and Konermann, 2014; Konermann, 2016). Because most antibodies target solvent-exposed dynamic loop regions, protection achieved at the earliest times would be indicative of binding (Rob et al., 2013). Monitoring exchange in the millisecond time region can decouple protection induced by binding from that resulting from allostery. By adding the antibody (viewed as a ligand) in D₂O itself, one can initiate binding concurrent with HDX and then separate in time the two phenomena.

For monitoring millisecond exchange, a custom-built, time-resolved electrospray ionization MS source for HDX (TRESI-HDX) was introduced by Konermann and coworkers (Wilson and Konermann, 2003; Rob et al., 2012) (refer to figure 1 in original reference Deng et al., 2017). Subsequently, Wilson and coworkers (Deng et al., 2017) used TRESI-HDX to reveal the allosteric changes in myoglobin upon antibody binding. They compared exchange when the antibody and antigen are pre-equilibrated with that when the antibody was introduced in the D₂O. The results show, as predicted, that the epitope site developed protection earlier than the sites of allosteric changes (refer to figure 6 in original reference Deng et al., 2017, kinetic experiment). At approximately 4 s of HDX, the allosteric and binding changes are complete, motivating the need for millisecond time resolution. Recently, these investigators used the millisecond platform to evaluate the biosimilarity of a mAb under development and the original (Brown et al., 2020).

Millisecond time resolution for HDX is not new (Lento and Wilson, 2022). Traditionally, fast-pulse labeling was achieved in HDX with a quenched flow for studying fast protein folding (Roder et al., 1988; Udgaonkar and Baldwin, 1988; Miranker et al., 1993). For monitoring allosteric changes, a careful design of mixing sequences on the commercially available LEAP PAL robotic system enables sub-second HDX (Chalmers et al., 2006).

Increasingly the ongoing technological developments in HDX position it to tackle even larger protein complexes, as in viral particles, to understand their dynamics under physiologically relevant conditions. Flaviviruses are highly pathogenic, vector-borne, single-stranded RNA viruses (Pierson and Diamond, 2020). Transmission between cold-blooded mosquito (Aedes sp.) to warm-blooded humans occurs with structural changes even before the virus attaches to host cells (Zhang et al., 2015). During late endocytosis, the acidic pH of the endosome leads to structural reorganization of the protein, fusion of the virus to the membrane, and release (Mukhopadhyay et al., 2005). Although X-ray structures are available for different physicochemical perturbations and facilitate understanding of the structural changes in viral particles, at best they are static pictures, endpoints that miss the dynamics of the change. For HDX, the effects of pH (Matthew and Richards, 1983; Bai et al., 1993), temperature, and ionic strength (Kim and Baldwin, 1982; Christoffersen et al., 1996) can be deconvolved from the overall change, leaving behind the changes on the viral-particle and providing new insights on the nature of human immune response against deadly viruses. Flavivirus on its surface has 180 copies of an envelope protein, each with approximately 53 kDa MW (total approximately 10 MDa complex). Despite the size, HDX-MS of the intact virus-like particles of the proteolytic peptides could be measured (Wang et al., 2001; Wang and Smith, 2005).

Anand and coworkers (Lim et al., 2017b) first deployed HDX-MS to understand the temperature-dependent structural rearrangement of Dengue virus strains (DENV1 and DENV2) particles and revealed that only the DENV2 envelope is expanded at 37°C (human body temperature) as compared to 28°C (vector temperature). DENV1 viral particles exhibit structural expansion only at 40°C (conditions of high fever in humans). The difference can explain how changes in temperature facilitate early stages of DENV strain-specific infection. In a subsequent study, the same group (Lim et al., 2017a) measured the dynamics of the virus-like DENV2 particle by using HDX-MS and located temperature-specific changes in epitope and paratope dynamics. Importantly, the study revealed that an antibody (2D22) primarily targets an envelope protein (E) and the E protein dimer interface (refer to figure 6 in original reference Lim et al., 2017a). By targeting the interface, 2D22 can dampen the temperature-associated changes in the viral envelope during early stages of infection. Paratope mapping revealed that heavy-chain contacts are maintained irrespective of temperature whereas the light chain contacts are lost upon temperature-mediated expansion. Such epitope and paratope mapping of the intact infectious virus in a pathophysiologically relevant state represents a quantum leap in epitope and paratope mapping compared to the study of constituent viral proteins and offers clear insights into antibody action at the target viral surface.

Factor H binding protein (fHbp) is the main virulence factor for the Neisseria meningitides, a leading cause for meningitides and sepsis. Owing to high sequence variability of the fHbp protein, the goal of developing a broad-based universal vaccine remains a challenge. Epitope mapping on such a variable antigen via a quick but reliable method is a pressing need. Enrico and coworkers (Malito et al., 2013) used several epitope mapping methods to reveal finer details of the epitope targeted by the anti fHbp antibody. Scanning a synthetic library comprised of 64 dodecameric peptide sequences of fHbp revealed only one binding peptide, 238–249. Scanning a phage-display library containing on average 55 amino acid length segments reinforced the conclusion of the PEPSCAN results and pointed to three peptides encompassing region 238–249. SPR binding analysis using PEPSCAN, however, identified a peptide from the C-terminal region of fHbp that showed poor binding (K _D > 1 mM) as compared to the full-length protein (K _D < 0.05 nM). The poor binding of the identified peptides clearly shows that PEPSCAN and a phage-display library failed to reveal a complete epitope for 12C1. To resolve, the investigators deployed HDX-MS (using a similar protocol to that discussed in Section 4.2). They found protection not only at the C-terminal but also in the N-terminal domains. HDX-MS results clearly reveal the conformational epitope for 12C1. To corroborate, the investigators also conducted X-ray crystallographic analysis to verify that the interface is the same as that determined by HDX-MS. Furthermore, mutational analysis supports a broad binding interface also revealed by HDX-MS and X-ray crystallography. This study underscores not only the importance of using multiple techniques to determine the full epitope but also the ability of HDX-MS to provide peptide-level information, yielding results similar to X-ray crystallography but with more efficiency. Thus, deploying HDX-MS very early on during lead optimization is less resource-intensive, recommending that resource-intensive X-ray and Cryo-EM be deployed later when the problem has some definition.

There are other complementary footprinting techniques to HDX, and they often utilize an advantage of irreversible labeling (Limpikirati et al., 2018; Liu et al., 2020; Ralston and Sharp, 2022). Conceptually, the process remains the same; the differential labeling of an antigen in the presence and absence of binding partner is compared. Such chemical labeling can be specific for one or a few amino acids (glycine ethyl ester (GEE) or diethylpyrocarbonate (DEPC)) or it could be more broadly reactive (•OH or •CF₃ radicals as in FPOP). Irreversible labeling has an advantage compared to HDX-MS that CID-MS can be used for gas-phase fragmentation to locate sites of labeling. Such covalent labeling was applied during the late 1980s for epitope mapping but not with MS detection (Burnens et al., 1987). As MS capabilities grew, studies began to utilize MS for locating covalent modification sites (Fiedler et al., 1998). Investigators employed several reagents including GEE (labels “D” and “E” residues) (Wecksler et al., 2015; Lin et al., 2018), DEPC (Tremblay et al., 2022), •OH (Jones et al., 2011; Yan et al., 2014; Zhang et al., 2017; Lin et al., 2018) and others (Dhungana et al., 2009; Liu et al., 2020) to map the epitope. It is •OH, generated by Fast Photochemical Oxidation of Proteins (FPOP), that enjoys some acceptance owing to its ability to control labeling and take advantage of the broad and high reactivity of free radicals and other reactive species (e.g., carbenes and carbocations), of the near pH independence of labeling, and of the ability to provide complementary information with respect to HDX-MS. FPOP depends on side-chain dynamics whereas HDX-MS uncovers backbone dynamics. These advantages make FPOP a good complement to alleviate some of the shortcomings of HDX-MS for epitope mapping. The Gross lab and their collaborators at Bristol Myers Squibb (BMS) have taken a lead in pushing the use of several mass spectrometry-based approaches to fine map the epitopes. In the following section, we discuss some of them.