Epitope and Paratope Mapping of PD-1/Nivolumab by Mass Spectrometry-based Hydrogen/Deuterium Exchange, Cross-linking, and Molecular Docking

Abstract

Programmed cell death-1 (PD-1), an antigen co-receptor on cell surfaces, is one of the conspicuous immune checkpoints. Nivolumab, a monoclonal antibody therapeutic approved by FDA, binds to PD-1 and efficiently blocks its pathways. In this study, an integrated approach was developed to map the epitope/paratope of PD-1/Nivolumb. The approach includes hydrogen deuterium exchange mass spectrometry (HDX-MS) followed by electron-transfer dissociation (ETD), chemical cross-linking and molecular docking. HDX-ETD offers some binding-site characterization with amino acid resolution. Chemical cross-linking provides complementary information on one additional epitope (i.e., the BC-loop) and a potential paratope at the N-terminus of the heavy chain. Furthermore, cross-linking identifies another loop region (i.e., the C’D-loop) that undergoes remote conformational change. The distance restraints derived from the cross-links enable building high-confidence models of PD-1/Nivolumab, evaluated with respect to a resolved crystal structure. This integrated strategy is an opportunity to characterize comprehensively other antigen/antibody interactions and to enable understanding binding mechanisms and design future antibody therapeutics.

Introduction

Antibodies are key biosensors in the immune system that can neutralize antigens and evoke other biomolecules that fight pathogens.¹ The binding between epitopes and paratopes is exquisitely specific and of high affinity, contributing to numerous applications in biological research, diagnostics, and therapy.² Comprehensive description of the epitopes/paratopes, ideally to the residue level, is crucial to understand the binding mechanism and to design future therapeutic agents. Hydrogen deuterium exchange (HDX) coupled with mass spectrometry (MS), an approach that reflects the local solvent accessible surface area (SASA) and H-bond network of the protein backbone, is a valuable tool for probing protein interfaces.^3-9 Its advantages are the near-native conditions of the experiment, low sample amount, and high throughput compared to X-ray crystallography. A major limitation, however, can be the coarse spatial resolution limited by the length of proteolytic peptides generated in the HDX experiment.¹⁰ Besides proteolyzing the protein to smaller fragment peptides, another approach to increase spatial resolution is electron transfer dissociation (ETD), a fragmentation technique that can locate deuterium on one or a few residues. It utilizes electron transfer from a radical anion to fragment peptides or protein with minimal scrambling of the amide H and D, in contrast to collision-induced fragmentation that uses many low-energy collisions to induce fragmentation.^11-14 Another limitation of HDX-MS is the inability to distinguish between the direct binding interaction and remote conformational or allosteric effects. The use of a combination of other complementary methods may overcome this limitation.

Mass spectrometry-based chemical cross-linking (XL-MS) has developed rapidly owing to the increased availability of diverse cross-linkers, advanced analysis software, and improvements in sample handling.^15-17 Observed cross-links deliver information about not only the connectivity of adjacent protein subunits but also the distance ranges between specific amino acid residues as defined by the spacing between the reactive functional groups in cross-linking reagents. These features contribute to a wide range of successful applications, including structural elucidation of single proteins¹⁸, topological portrayal of large macromolecular assemblies^19-20, and interaction maps of an entire proteome.^20-21 Mapping epitopes/paratopes by using XL-MS, however, is an underutilized opportunity.²² In this study, we used a combination of XL-MS, HDX, and HDX-ETD to illustrate an analytical approach for epitope/paratope mapping of an important antigen/antibody system.

Programmed cell death-1 (PD-1)²³, an immune checkpoint, is an antigen-independent co-receptor, located on cell surfaces and expressed predominantly by T-cells.²⁴ The critical role of PD-1 is to bind with specific ligands, PD-1 ligand 1 (PD-L1)²⁵ and PD-1 ligand 2 (PD-L2)²⁶, to maintain immune tolerance by suppressing self-reactive T-cells²⁷ and preventing pathogenic autoimmunity. The signaling, however, can be utilized by tumor cells to escape immune surveillance.^28-29 Therefore, blockage of the PD-1 pathway has been an appealing target in recent development of immuno-therapeutics.^30-32 Nivolumab , one of two monoclonal antibodies (mAbs) on the market³³, is designed to bind with PD-1, demonstrating immune restoration in multiple tumor conditions^34-36 with impressive clinical efficacy.

Here, we applied HDX-MS to the PD-1/Nivolumab complex to obtain regional binding information, which was further refined by HDX-ETD to specify more closely the critical binding residues. The suggested epitope and paratope regions were subsequently evaluated by XL-MS, revealing complementary binding interfaces and differentiating remote conformational changes. Utilizing the distance restraints derived from various cross-linkers, we conducted molecular docking to generate high-confidence 3D models and evaluated the strengths and limitations of this approach. A previously resolved X-ray crystal structure of PD-1/Nivolumab Fab^37-38 was employed for final comparison purposes. The integration of several MS-based approaches enables more precise and detailed characterization of epitopes/paratopes, increasing our understanding of binding mechanisms and providing support of protein therapeutic discovery.

Experimental Section:

Hydrogen Deuterium Exchange Mass Spectrometry

HDX of PD-1 and Nivolumab was conducted under several conditions including PD-1 alone, Nivo Fab alone, and bound PD-1 and Nivo Fab at a molar ratio of 1:2 on an HDX PAL robot (LEAP Technologies, Carrboro, NC). More details are in SI.

Chemical Cross-linking

PD-1 (Bristol-Myers Squibb, NY, NY) and Nivolumab Fab (Bristol-Myers Squibb, NY, NY) were crosslinked by NHS-ester cross-linkers and EDC in individual trials. The extent of crosslinking was monitored by using gel electrophoresis followed by in-solution digestion. More details are in SI.

Molecular Docking with Cross-Link Derived Restraints:

Protein-protein docking for PD-1 and Nivolumab Fab was conducted by the Rosetta (v. 3.8) ^39-41 docking_protocol (RosettaDock) program ^42-43 with the X-ray structure of the PD-1/Nivolumab Fab complex (PDB ID: 5WT9) ³⁷. The distance restraints were derived from the cross-links list in Table 1, specifically, 6 – 16 Å⁴⁴ for the Cα-Cα distance(s) of crosslinked residues using EDC and 9 – 30 Å⁴⁵ using BS²G or BS³ crosslinks. More details are in SI.

Table 1.

Summary of observed cross-links

	PD-1	Nivo Fab	Cross-linker	Epitope	Paratope
1	S27 – K57(H)		BS2G /BS3	N-Loop	CDR-H2
2	D26 – K57(H)		EDC
3	S27 – Y35(L)		BS3		CDR-L1
4	S62– N-term (H)		BS3	BC-Loop	N-terminus (H)
5	E61 – N-term (H)		EDC
6	K135 – K57(H)		BS3	FG-Loop	CDR-H2
7	K135 – Y35(L)		BS3		CDR-L1
8	K135 – N-term (H)		BS3		N-terminal (H)

Results and Discussion

Epitope and Paratope Mapping by HDX

To map the epitope on PD-1, we performed HDX experiments with unbound PD-1 and PD-1 bound to the antigen-binding fragment of Nivolumab (Nivo Fab) at a molar ratio of 1:2. Deuterium uptake was monitored at 0.33, 1.0, 10, and 240 min on 19 unique peptides, covering 85% of the PD-1 sequence. Information on 15% of the sequence was lost likely because the antigen contains complex N-linked glycans, hampering peptide chromatography and identification at those sites. Accumulative deuterium uptake differences across the four time points were calculated (Figure 1A), revealing three regions that become more protected upon binding: ²⁵LDSPDRPWNPPTFSPALL⁴², ⁸⁰AAFPEDRSQPGQDCRF⁹⁵ and ¹²⁵AISLAPKAQIKESL¹³⁸. Specifically, regions ¹²⁵AISLAPKAQIKESL¹³⁸, located on the FG-loop of the PD-1 structure (Figure 1B), and ²⁵LDSPDRPWNPPTFSPALL⁴², part of the N-loop (Figure 1B), exhibit the most significant decrease in HDX upon binding to Nivo Fab with protection corresponding to decreases of 7 and 4 Da, respectively, in comparison with the unbound state. Region ⁸⁰AAFPEDRSQPGQDCRF⁹⁵ containing the C’D-loop (Figure 1B) showed protection corresponding to a decrease of approximately 2 Da. The deuterium uptake differences were similar across all time periods for the three regions (Figure 1C), suggesting stable protection, strong bonding, and associated small off rates in the binding equilibrium.

Figure 1.

Epitope regions on PD-1 indicated by HDX. (A) Differential HDX kinetics plots of PD-1 and PD-1/Nivolumab Fab complex. (B) Epitope regions mapped onto the PD-1 crystal structure (PDB: 3RRQ). (C) HDX kinetics of three peptides corresponding to the epitope regions in PD-1. Unbound PD-1 is colored in black and bound PD-1/Nivolumab Fab is colored in burgundy. The green-shaded region represents the propagated error across all time points. HDX kinetics results for the other peptides are shown in Figure S1.

We also performed HDX experiments with PD-1 and full-length Nivolumab (Nivo mAb) for comparison (Figure S2). One of the peptides, ¹³²KAQIKESLRAELRVTE¹⁴⁷, was not found in PD-1-Nivo mAb complexes, possibly owing to the larger number of peptides when using the full Nivo mAb. The larger number of peptides from full-length mAb hampers the acquisition and identification of PD-1 peptides in the fast chromatography used in HDX. The other peptide regions, on the other hand, showed consistent and similar trends regarding the deuterium uptake differences for bound and unbound states, indicating comparable binding behaviors between the full Nivolumab and its Fab.

To map the paratope on Nivolumab, we conducted a series of similar HDX experiments. Non-bound and bound Nivo Fab (PD-1: Nivo Fab at a molar ratio of 2:1) were exchanged with deuterated buffer from 0.33 min to 4 h. The accumulative deuterium uptake differences between the two states confirm the involvement of the CDR regions in PD-1 binding (Figure 2). Peptide regions covering CDR-H2 on the heavy chain (Figure 2A) and CDR-L3 on the light chain (Figure 2B) showed the greatest protection, corresponding to more than 8 Da, and a constant difference of HDX as a function of time (Figure 2C). Peptides covering CDR-H1, CDR-H3 and CDR-L2 showed smaller decreases in HDX upon binding (~ 6 Da for the first and ~ 3 Da for the latter two). The HDX of the peptide covering ³³SSYLAWYQQKPGQA⁴⁶ exhibited only 1 Da deuterium uptake difference across all HDX time points, and the difference mainly occurred at the longer HDX time (Figure 2C). Given that this region only partially overlaps with CDR-L1, ²⁷RASQSVSSYLA,³⁷ few amino acids in this region are involved in PD-1 binding, resulting in small differences in HDX. The HDX paratope of Nivolumab, in general, is in accord with what is commonly viewed as CDR regions.

Figure 2.

Paratope regions on Nivo Fab as determined by HDX. Differential HDX kinetics plots of Nivo Fab vs. PD-1/Nivo Fab complex for (A) heavy chain and (B) light chain, respectively. (C) HDX kinetics of the peptides in the corresponding paratope regions in Nivo Fab (unbound PD-1 is colored in black and bound PD-1/Nivo Fab is in burgundy). The green-shaded region represents the propagated error across all time points. HDX kinetics plots of the other peptides are shown in Figure S3.

Epitope Refinement by HDX-ETD

The spatial resolution of epitope identification by HDX is limited by the size of peptic peptides and by overlapping peptides, providing only regional information. To increase the spatial resolution, we coupled ETD fragmentation with HDX-MS to refine further the protected regions. To set up the ETD measurements, we used a previously published procedure¹¹ to measure the extent of H/D scrambling of a synthetic peptide, HHHHHHIIKIIK, under several conditions and selected those that show minimal scrambling. In addition, the bound-state PD-1 was achieved with Nivo Fab to produce fewer peptides, lower complexity in the separation step of HDX than with full-length Nivo mAb, and the choice of the Fab led to increased the signal-to-noise ratios for peptide peaks. The incubation time for deuterium labeling was controlled as 1 min, a time point that gives distinct differences in the peptide-level HDX.

We submitted to ETD the three peptides (Figure 1C) that cover the epitope regions identified by the HDX kinetics. Not all the peptides could be successfully resolved owing to their nature (e.g., residue composition) and to incomplete fragmentation of the peptide at their available charge states. For example, the doubly charged peptide ²⁵LDSPDRPWNPPTFSPALL⁴², part of the N-loop, suffers from multiple proline residues in the sequence, showing a limited number of product ions. The peptide ⁸⁰AAFPEDRSQPGQDCRF⁹⁵ containing the C’D loop has only a moderate difference in HDX between bound and unbound (~ 0.5 Da at 1 min), and that makes it difficult to measure differences in HDX given the large size of the peptide (16 residues) and the experimental error associated with HDX-ETD. On the other hand, a triply charged peptide ¹²⁵AISLAPKAQIKESL¹³⁸, located on the FG-loop, successfully produced a series of C fragment ions upon ETD, allowing further epitope refinement for this region.

We plotted the cumulative deuterium uptake for all C-ions and calculated the deuterium uptake difference between the bound and unbound states of PD-1 (Figure 3: note that the deuterium uptake of each C_n-ion represents that occurring on the n + 1 residue). The absence of C₁ and C₂ ions is likely an undesirable consequence of the ionization optimization in which we balanced the intensity of the fragment ions and the extent of deuterium scrambling. The deuterium uptake measured on the C₃ ion was reduced (~ 0.8 Da) upon binding. Given that deuteriums on the first two residues are almost always lost owing to back exchange⁴⁶ and the C₂ ion was not resolved in the experiment, protection of residue ¹²⁷S or ¹²⁸L or both could account for the observed decrease in deuterium uptake on C₃. Further increases in HDX protection were observed at C₆, showing 1.6 Da reduction in the PD-1/Nivo Fab complex compared to the unbound PD-1; the considerable drop in deuterium uptake pinpoints the protected residue ¹³¹K, given the HDX-silent ¹³⁰P and the similar deuterium uptake difference of C₄ compared to that of C₃. In addition, the C₇ ion exhibits additional HDX protection, increasing from 1.6 at C₆ to ~ 2.3 Da. The HDX for C₉ shows an additional uptake in protection as at C₇, suggesting protection on residues ¹³⁴I and ¹³²A, respectively, upon binding to the Nivo Fab. Other C_n fragments showed insignificant differences between the two states compared to the adjoining C_n-1 ions. In summary, HDX-ETD allows epitope refinement to ¹²⁷S and/or ¹²⁸L, ¹³¹K, ¹³²A and ¹³⁴I in the region ¹²⁵AISLAPKAQIKESL.¹³⁸

Figure 3.

Cumulative deuterium uptake plot for C-type ions of ¹²⁵AISLAPKAQIKESL¹³⁸ in PD-1 and PD-1/Nivo Fab complex by HDX-ETD. Deuterium uptake differences are calculated and labeled for each C-ion. The potential binding residues are indicated with arrows and colored in red.

Although HDX-ETD successfully reveals several epitope binding residues, it can be challenging to differentiate HDX protection induced by direct binding or by a remote conformational change induced by binding. Additional information on the interacting domains is desirable, and that prompts our subsequent investigation to complement the HDX results with chemical cross-linking.

Chemical Cross-linking of PD-1 and Nivolumab Fab

To achieve better coverage and more comprehensive information of the epitope and paratope regions provided by HDX, we utilized several cross-linkers, including BS³-H₁₂/D₁₂ and BS²G-H₄/D₄, different in spacer lengths, and EDC/NHS targeting glutamic (D) and aspartic acids (E), complementary to the usual NHS-ester reactive residues (e.g., lysine (K), serine (S), tyrosine (Y) and the N-terminus). We tested several concentrations of cross-linkers with respect to those of the proteins and could monitor the success of cross-linking by the band of PD-1/Nivo Fab on an SDS PAGE gel (Figure S4). Individual cross-linked samples were digested in solution followed by LC-MS/MS analysis and cross-linking identification with pLink.^47-48 In total, we identified eight distinct inter-molecular cross-links (Table 1; representative mass and product-ion (MS/MS) spectra are shown in Figure 4) located on different regions of PD-1 and the Nivo Fab, consistent with the HDX results that the binding interface is discontinuous.

Figure 4.

Representative mass spectra and product-ion spectra (XL-8)

For PD-1, multiple cross-links were formed on the N-loop (cross-link 1-3) and the FG-loop (cross-link 6-8), two regions that also showed upon binding significant protection in the HDX kinetics, consistent with the assignment as epitope regions. XL-MS results not only complement those of HDX but also reveal an additional binding region on PD-1, the BC-loop, identified by cross-links 4-5 by both BS³ and EDC chemistry. It is worth mentioning that, for EDC cross-linking, the Euclidean distance between the cross-linked atoms is only ~ 3 Å, the length of one amide bond. Thus, this XL reagent locates and defines the binding interfaces with higher spatial resolution than do other reagents. Cross-link 5 emphasizes the vicinity of the BC-loop to the N-terminus of the heavy chain on Nivo Fab, indicating physical contacts between the epitope and paratope.

On the other hand, we observed no cross-links on the C’D-loop, and this could result from the lack of reactive residues or may indicate that it is not a binding interface. Although the sequence of C’D-loop contains several eligible residues for cross-linking (e.g., D and S), their side chains may orient in an unfavorable way to preclude cross-linking. Because the C’D-loop region showed only a low extent in the accumulative deuterium uptake in HDX upon binding with Nivo Fab, we suggest this is not a region that involves strong interactions between the proteins. Binding-induced remote conformational changes more likely account for the reduced HDX upon binding.

On the heavy chain of the Nivo Fab, we observed three cross-links involving ⁵⁷K (i.e., cross-links 1, 2 and 6) located in the CDR-H2 region identified from the HDX kinetics. This region, based on the supporting evidence from both HDX and XL-MS, is confidently assigned to be a paratope. A newly revealed paratope region is the N-terminus of the heavy chain, which affords multiple cross-links not only with the BC-loop of PD-1 but also with the FG-loop. On the light chain of Nivo Fab, we observed only one cross-linked residue, ³⁵Y, on the CDR-L1 peptide. The identified cross-links (i.e., 3 and 7) support the CDR-L1 region as a binding interface with PD-1. For the four other CDR regions, we identified no cross-links, showing the limitations of using stand-alone XL-MS for mapping. Restricted numbers of reactive residues, considering both the intrinsic and low reactivity, side-chain orientation, and complexity of the cross-linked species diminish the possibility of using XL alone to characterize epitope/paratope interfaces. A more confident assignment than either approach alone is integrating HDX and XL-MS, an approach that is validated by a comparison of our MS results and the resolved X-ray crystal structure in the next section.

Epitope/Paratope Assignments by MS and Comparison with X-ray Crystallography

The HDX results mapping the epitope and paratope on the peptide-level of PD-1 with both Nivo Fab and full-length Nivolumab show good agreement and indicate comparable binding behavior. Thus, we conducted the residue-level analysis with PD-1/Nivo Fab by using HDX-ETD and cross-linking MS.

The results from HDX point to three epitope regions on PD-1, namely the N-loop, the FG-loop and the C’D-loop, whereas XL-MS supported the former two and revealed an additional BC-loop (Figure 5). A lack of cross-links and a small HDX difference between bound and unbound make the C’D-loop less likely to be a binding interface but rather a region undergoing a remote conformational change induced by binding elsewhere. These conclusions agree well with the assigned epitope regions from the crystal structure of PD-1/Nivo Fab (PDB: 5WT9),^37-38 where the N-loop, BC-loop and FG-loop are identified as epitopes. The C’D loop, however, is not resolved, and there are no physical contacts between that region of the antigen and any paratope regions.

Figure 5.

Summary of binding regions identified by HDX (blue) and XL-MS (red) for (A) the PD-1 and the Nivo Fab complex including (B) heavy chain and (C) light chain. Critical binding residues indicated by HDX-ETD are pinpointed with triangles. Epitope/paratopes assigned from the crystal structure (PDB: 5WT9) are underlined in grey.

The integrated information from HDX and XL-MS affords additional information on the epitopes/paratopes, in accord with that from the crystal structure. For example, a resolved H-bond between ²⁵L on the N-loop of PD-1 and ⁵⁷K on the Fab heavy chain is consistent with two cross-links (cross-link 1 and 2, Table 1) to nearby reactive residues, ²⁶D and ²⁷S. In addition, the binding between the N-loop and CDR-L1 is established with cross-link 3 and with H-bonding between ²⁶D on PD-1 and ³⁵Y on the Fab light chain. HDX coupled with ETD fragmentation further identifies the binding residues in the FG-Loop, three of which, ¹²⁸L, ¹³¹K and ¹³²A, contact with the Nivo Fab through H-bonding and van der Waals interactions. Assignment of this epitope region is also supported by cross-link 7, resembling the H-bond between ¹³¹K on PD-1 and ³⁷A on the Fab light chain. Additionally, we observed that the FG-loop can cross-link with other domains through ¹³⁵K; those cross-links include the CDR-H2 (cross-link 6) and the N-terminus of the heavy Fab (cross-link 8). The N-terminus also cross-linked with ⁶¹E and ⁶²S on the BC-loop of PD-1 (cross-links 4 and 5), suggesting interprotein binding that is not seen in the solid-state structure.

In addition, the cross-linking network for the epitopes/paratopes delivers topological information of the PD-1/Nivo Fab complex by providing distance restraints. These restraints allow a description of the interaction regions and even of the overall architecture when the information is coupled with other approaches (e.g., protein-protein docking). The resulting 3D-information of the binding complex can provide a foundation for even more accurate determinations of the epitope/paratope.

Protein-protein Docking of PD-1 and Nivolumab Fab with Cross-link Derived Restraints

We conducted a protein-protein docking study with the RosettaDock program^42-43 by starting with the structures of PD-1 and the Nivo Fab extracted from the X-ray structure of the complex (PDB: 5WT9)³⁷. Docking of unbound Nivolumab and PD-1 was not feasible because many critical residues in the unbound PD-1 are not resolved. For each docking run, we separated the two proteins with the same intial configuration, followed by rotation to arbitrary extents (details in SI). Our earlier work⁴⁹ demonstrated that incorporation of multiple cross-link-derived restraints in the protein-protein docking computations can effectively yield high-quality models; thus, the restraints based on all eight of the identified cross-links (Table 1) were utilized here. We generated 250 RosettaDock docking runs, each of which gave up to 400 PD-1/Nivo Fab models. The 20 best-scoring models based on the RosettaDock “total_score” metric occupied a tight cluster, all of which closely recapitulated the X-ray structure of the PD-1/Nivo Fab complex (Figure 6A). For these 20 models, the root-mean-square deviation (r.m.s.d.) of all Cα atoms for PD-1 in the models relative to PD-1 in the X-ray structure ranged from 0.4 – 1.8 Å with a mean of 1.0 ± 0.4 Å, which is less than 2 Å, a threshold that often is used to define a high-confidence model. The successful generation of the complex architecture showing the protein/protein interface enables an in-depth view of the epitopes/paratopes.

Figure 6.

(A) Twenty best-scoring protein-protein docking models of PD-1/Nivo Fab complex (PD-1 in cyan, the heavy chain of the Nivo Fab in dark pink, the light chain of the Nivo Fab in light pink), superimposed on the X-ray structure (red). (B) A representative model of PD-1/Nivo Fab complex with an enlarged view of the epitopes/paratopes (BC-loop in magenta, C’D-loop in black, N-terminus of Nivo Fab heavy chain in marine, CDR-H1 in purple).

We chose a representative model and enlarged the binding interface between PD-1 and Nivo Fab (Figure 6B). The BC-loop of PD-1 locates at similar proximity with CDR-H1 and the N-terminal region of heavy Fab, indicating that the two regions may contribute simultaneously to the binding interaction. The conformation is consistent with the N-terminal region being a paratope. The X-ray structure suggested the BC-loop is in physical contacts with CDR-H1, whereas we identified cross-links on the N-terminus of the heavy chain, complementing the scheme and emphasizing the necessity of examing binding events using solution-based approaches. In addition, another questionable region, the C’D-loop on PD-1, is far from the binding interfaces, minimizing the likelihood of being an epitope. Moreover, unlike it is resolved in the unbound PD-1, the C’D-loop is missing in the bound PD-1 complex, giving supporting evidence of undergoing considerable remote conformational changes companied with different strucutal dynamics. The quaternary structure of the PD-1/Nivo Fab complex enables more confident assignment of the epitope/paratope regions.

There are cautions, however, in generalizing the integrated method that includes XL-MS and molecular docking to other protein-binding systems. One obstacle we encountered in the docking study of the PD-1/Nivo Fab complex is the dissimilar structures of PD-1 in its unbound state (PDB: 3RRQ) and bound state (PDB:5WT9). Epitopes on PD-1 are mainly loops, some of which cannot be resolved in the X-ray structure owing to their high flexibility in absence of bonding to the Nivo Fab (e.g., the N-loop). Consequently, distance constraints derived from the residues within this region are of little use for downstream docking. More importantly, for loops that are resolved in the unbound state may have different orientation in the bound state. The conformations of dynamic loop regions are susceptible to amino acid substitutions (mutations), truncations, and to crystallization conditions including the ionic strength of the medium; these changes can lead to incorrect conformations for the solid-state structure. Given that the docking protocol does not readily accommodate changes in protein tertiary structure, biased initial structures could lead to erroneous 3D models of the binding complex. Incorporation of other computational methods (e.g., discrete molecular simulation) can better accommodate the structural changes. In fact, binding interfaces that contain mainly helices and beta-sheets, which possess relatively fixed high order structures, are preferable inputs for docking studies.

Conclusion

This study provides convincing evidence that epitope and paratope mapping by HDX, cross-linking, and docking can be effective. Although HDX-MS, as a stand-alone method, has shown fruitful applications in mapping binding interfaces, it has limited spatial resolution and has trouble distinguishing binding from remote conformational change. Using PD-1/Nivo Fab as an example, we demonstrated that integrating HDX-ETD, XL-MS, and molecular docking gives a more comprehensive description of epitopes/paratopes. Some critical binding residues can be successfully identified from HDX-ETD and chemical cross-linking results, further delineating interactions along a protein/protein interface. In addition, XL-MS affirms epitopes/paratopes characterized by HDX-ETD and distinguishes binding from sites showing protection as remote conformational changes. The distance restraints afforded by XL-MS allow building high-confidence 3D models with molecular docking. The integrated platform amplifies the ability of each biophysical method, offering an approach for other antigen/antibody systems that are difficult to crystallize for X-ray diffraction. It is noteworthy that docking exercises require careful consideration even with the availability of high-quality protein structures obtained by high-resolution techniques or computational methods. Even without molecular docking, the combination of HDX-ETD and XL-MS, the latter which is not often used in epitope/paratope mapping experiments, gives insights that deepen our understanding of antigen-antibody binding and assist the design of future antibody therapeutics.