HDX-MS and MD Simulations Provide Evidence for Stabilization of the IgG1â€”FcÎ3RIa (CD64a) Immune Complex Through Intermolecular Glycoprotein Bonds

Previous reports present different models for the stabilization of the Fc—FccRI immune complex. Although accord exists on the importance of L235 in IgG1 and some hydrophobic contacts for complex stabilization, discord exists regarding the existence of stabilizing glycoprotein contacts between glycans of IgG1 and a conserved FG-loop (MGKHRY) of FccRIa. Complexes formed from the FccRIa receptor and IgG1s containing biantennary glycans with N-acetylglucosamine, galactose, and a2,6-Nacetylneuraminic terminations were measured by hydrogen–deuterium exchange mass spectrometry (HDX-MS), classified for dissimilarity with Welch’s ANOVA and Games-Howell post hoc procedures, and modeled with molecular dynamics (MD) simulations. For each glycoform of the IgG1—FccRIa complex peptic peptides of Fab, Fc and FccRIa report distinct H/D exchange rates. MD simulations corroborate the differences in the peptide deuterium content through calculation of the percent of time that transient glycan-peptide bonds exist. These results indicate that stability of IgG1—FccRIa complexes correlate with the presence of intermolecular glycoprotein interactions between the IgG1 glycans and the KHR motif within the FG-loop of FccRIa. The results also indicate that intramolecular glycanprotein bonds stabilize the Fc region in isolated and complexed IgG1. Moreover, HDX-MS data evince that the Fab domain has glycan-protein binding contacts within the IgG1—FccRI complex. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/ 4.0/). ier Ltd.This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Journal of 167391 K.W. Anderson, C. Bergonzo, K. Scott, et al. Journal of Molecular Biology 434 (2022) 167391


Introduction
Monoclonal antibody (mAb) pharmaceuticals comprise potent, targeted therapeutics for addressing cancer, 1 autoimmune conditions,2-4 osteoporosis, 5 macular degeneration, 6 migraine, 7 and infectious diseases including SARS-CoV-2. 8,9 Most therapeutic mAbs are IgG1s formed by a dimer of two longer heavy chains, where each is bound to a shorter light chain via disulfide bonds. The dimer is depicted in a canonical Y-shape, where the arms represent Fab regions that recognize a specific antigen, and the vertical stroke represents the dimeric Fc region. Each Fc monomer has a distribution of complex biantennary glycan structures bound at a conserved N-glycosylation site, asparagine 297 (N297). 10 Glycan structure affects effector functions, [11][12][13] pharmacokinetics, 13,14 stability, [15][16][17][18] and aggregation propensity. 19,20 The subject of this report, the strongly bound IgG1-FccRI complex, is formed when the Fc region binds to integral membrane glycoprotein FccRI (CD64) high-affinity receptors, residing on the surface of monocytes, macrophages, dendritic cells, and on activated neutrophils and eosinophils. 21 This complex promotes killing of mAb-coated targets by phagocytosis or cellmediated cytotoxic processes. 22 Novel antibodyinduced immunotherapies for treating metastatic melanoma through FccRI have been proposed. 23 Adversely, these interactions can also result in thrombocytopenia, anaphylaxis, and arthritis induction. 24 The role of glycans in the stabilization of IgG1-FccRI complexes is unclear. Three x-ray crystal studies have provided detailed structures of glycosylated IgG1-FccRI complexes. One structure is of the P1 space group 25 and two structures, determined at higher resolution, are of the C2 space group. 26,27 Even though most backbone atoms in these structures are nearly superimposable, the nature of binding interactions remain unresolved, as the P1 and C2 structures differ in critical interface contacts. The P1 space group structure assigns strong bonding to the FG-loop in the D2 domain of the receptor, which enables a charged 173 KHR 175 motif to interact with proximal carbohydrate units of the Fc glycans. Additional stabilization comes from a hydrophobic cluster comprising K130, L131, and Y133 of FccRI and Fc residues L234/L235 and a hydrophobic sandwich comprising W104/W127 of FccRI and Fc residue P239. 25 Contrariwise, the C2 space group structures report that the placement of Fc L235 into a hydrophobic pocket at the surface of FccRI and hydrogen bonds involving Fc K173/L131 and other interactions stabilize the Fc-FccRI complex. One C2 structure assigns a solitary intermolecular van der Waals bond to exist between the proximal IgG1 GlcNAc and receptor. 26 The other C2 structure study finds no evi-dence for intermolecular glycan-protein interactions. 27 Discord over bonding contacts may exist because X-ray crystallography captures the lowest-energy forms of biomolecules and does not sample less populated, higher energy states that can manifest transient bonds in dynamic equilibrium under physiological conditions. Solution phase studies do sample these less populated, higher-energy forms. A summary conclusion of the solution studies is that the essential elements of strongly bound IgG1-FccRI complexes require Fc dimers containing N297 glycosylation and L235 residues [25][26][27][28] and a receptor containing a conserved FG-loop ( 171 MGKHRY 176 ). 25,29,30 The dissociation constant of wild type, glycosylated IgG1 ligands with FccRI is K d = (8.8 ± 7) nmol, which is increased by a factor of 350 for deglycosylated IgG1 ligands. 25 Even so, the increased K d permits substantial ligand occupancy of FccRI. [31][32][33] Presence or absence of fucosylation has no measurable effect on relative binding activity. 33 Glycan structures decorating FccRI have minimal effect on complex formation. 34 Questions remain on the roles of antennary glycans, core GlcNAc, and core mannose in the stabilization of IgG1 and of the IgG1-FccRI complex. X-ray crystallography structures evince the presence of intramolecular contacts between glycans and Fc residues. [25][26][27] Better resolved xray structures appear to show galactoseterminated glycans buried between the C H 2 domains, seemingly unavailable for interaction. 35,36 On the other hand, solution phase studies report that the a1-3 and a1-6 glycan arms are available to enzymatic remodeling. [37][38][39][40] In accord with these observations, nuclear magnetic resonance (NMR) spin-relaxation studies have determined that galactose-terminated a1-6 and a1-3 arms undergo large excursions with 2 ns correlation times. 41 Thus, both arms appear available to interact with FccRI receptors.
Molecular dynamics simulations are a powerful way to gain understanding about conformational dynamics by projecting a fourth dimension (time) to the static 3D coordinates from x-ray crystallography or other structural biology methods. Quantitative comparison to HDX-MS data is difficult due to the timescales, solution conditions, temperatures, and complex kinetics involved. 42 Many groups have had success in either post hoc ensemble pruning or restraining simulations using HDX-MS data for simpler systems. 43 Here, MD simulations provide insight into dynamics in the potential minima described by x-ray crystallography structures and allow easy testing of different hypotheses to decouple direct from indirect effects on binding. 44 This study examines the solution-phase dynamics of IgG1 glycoforms and IgG1-FccRI complexes with special attention to the roles of glycans in stabilization. For this investigation we conducted hydrogen-deuterium exchange mass spectrometry (HDX-MS) measurements on aIL8hFc (IgG1) glycoforms in isolation and in complexation with soluble FccRIa. We analyze the HDX-MS data by qualitative application of a statistical thermodynamics (type 1) scenario that can account for changes of H/D exchange rates between chemophysical states though glycanprotein and protein-protein contacts. 45,46 The HDX-MS experiments are accompanied by molecular dynamics simulations of Fc-FccRIa glycoform complexes, which can predict and quantify contribution changes to binding of transient glycan-protein bonds correlated with glycan dynamics. The results indicate that intermolecular bonds between Fc glycans and the 173 KHR 175 motif of the receptor FGloop provide some stabilization energy to the IgG1-FccRIa complex.

Results
aIL8hFc is a murine-human chimeric IgG1, which inhibits IL-8 binding to human neutrophils. 47,48 For brevity, we refer to each IgG1 glycoform by the predominate glycan structure bound to N297 in each Fc, e.g., G0F refers to aIL8hFc-G0F. For the convenience of direct comparisons, we align all IgG1 heavy chains and their peptides to the EU numbering system. 10 For aIL8hFc the slightly longer V H sequence extends its sequence numbering to HC -4. In tables "HC" prefaces sequence start and stop indices of peptides from the heavy chain sequence, "LC" prefaces sequence indices of peptides from the Fab light chain, and "Rc" prefaces sequence numbers of peptides from FccRIa. In tables we abbreviate the terms, apo-and holo-to, a-and h-, respectively. In plots and discussions, the term, holo-FccRIa X , references peptides from FccRIa in complexation with aIL8hFc-X (X = G0F, G2F, SAF). Figure 1 diagrams the glycoform structures discussed in this manuscript. Table 1 lists the relative abundances of aIL8hFc glycoforms found in each sample by hydrophilic interaction liquid chromatography (HILIC). HILIC analyses of native aIL8hFc, expressed from unmodified CHO cells reveal traces of M5 and G1F glycoforms. 49 These glycoforms are also present in G0F, G2F, and SAF samples.

Composition of aIL8hFc and FccRIa
Tandem mass spectrometry (MS/MS) measurements of the peptic peptides from each aIL8hFc glycoform detected oxidation only in 241 FLFPPKPKDTLM 252 . Integrated MS peak areas of this peptide revealed the amount of oxidation: G0F (0.8% ± 0.1%), G2F (1.6% ± 0.1%), and SAF (2.2% ± 0.1%), where the uncertainty denotes one sample standard deviation (1s). MS/MS data did not detect oxidation in the FccRIa material. Peptic peptides of FccRIa and aIL8hFc glycoforms did not exhibit modifications indicative of deamination or phosphorylation.

HDX-MS data sets
HDX-MS data for proteins were collected under physiologic conditions using the same stock reagents and chromatography solution lots, pH, salt concentrations, and temperature profile. The working dataset comprises %17,750 measurements across 7 states, defined by the biomolecules (IgG1 glycoforms and receptor) and complexation (apo-and holo-). The Supplementary Material contains spreadsheets listing mass uptake vs t HDX for peptides used for this study. Unabridged HDX-MS datasets for uncomplexed and complexed aIL8hFc glycoforms and FccRIa receptor are archived in a public database. 50 Table S1 and Figures S1-S3 in the Supplementary Material list protein sequences and graph the peptic peptides of aIL8hFc and FccRIa. In recommended formats Tables S2-S4  summarize HDX-MS experiments. Tables S5-S7 display D-uptake vs log 10 t HDX traces for selected peptides from aIL8hFc and FccRIa. Tables S8-S10 list D peptide X t HDX ð Þ for peptides common to all states. 51 Evaluation of deviations from similarity among states of IgG1 glycoforms and the receptor HDX-MS data for each isolated glycoform and for each binary mixture are organized into two collections: six peptide datasets for the light chain and six peptide datasets for the heavy chain. Peptides of the same sequence in the apo-and holo-G0F, G2F, and SAF states were tested for equal deuterium content, u X [" D peptide X t HDX ð Þ] using Welch's one-way ANOVA (a = 0.025, k = 6) and Games-Howell post hoc procedures (WAGH). Here, a is the probability of rejecting the null hypothesis when it is true and k is the number of datasets. These procedures identified means; l X , residing outside confidence limits of the null hypothesis (l 1 ¼ l 2 Á Á Á ¼ l k Þ. Peptides failing the null hypotheses for a majority of nonzero t HDX are designated as dynamically dissimilar. As described in the Methods section, the dissimilarity, or Jaccard distance d J , of a protein sequence between states A and B is computed using the binary results of the WAGH analyses. Values of d J can range between one, signifying no similarity between states A and B, and zero, indicating that the states are identical within measurement uncertainties. Y %E A;B , the absolute value of mean differences in %E for a peptide between states A and B and the difference polarity are computed by summing differences of dissonant %E peptide i ;X t HDX ð Þ. Eq. (1) converts raw HDX-MS measurements, Þ. Each peptide exhibiting dissimilar H/D exchange rates between states A and B is marked on a Deviation from Similarity vs Sequence # plot with a colored bar of location and width defined by the sequence start and stop indices. If the mean H/D exchange rate of a peptide in state A is greater than in state B, the bar projects to +y-value; and if mean H/D exchange in state B is greater than in state A, the bar projects to Ày-value. Because specific bar color and y-elevation have no statistical significance, these attributes are chosen arbitrarily to enable the reader to easily distinguish overlapping peptide clusters. However, once xcoordinates, y j j, and color of each bar are defined, and these attributes are conserved across all figures of this report. Only y-polarity may change. For example, in all panels of Figures 2, 3, and S4 peptide HC 242-251 is always colored orange, resides between x = 242 and x = 251, and has an elevation of y j j = 5 units. Peptide Rc 99-105 provides an example of a polarity change between similarity plots of holo-FccRIa G0F vs holo-FccRIa G2F ( Figure S5(d)) and holo-FccRIa G2F vs holo-FccRIa SAF ( Figure S5 (f)).
Each WAGH(a = 0.025, k = 6) calculation produces a set of Deviation from Similarity vs Sequence # plots composed of 15 comparisons among the protein states. For each chain three colored bar plots infer the effects of glycan structure on the dynamics of isolated apo-aIL8hFc ( Figure 2) and three colored bar plots infer effects of glycan structure on the dynamics of aIL8hFc-FccRIa complexes ( Figure 3). The magnitude of peptide D-uptake dissimilarity, Y %E A;B (red line), is shown above each colored bar plot. Figure S4 in the Supplementary Material provides additional similarity plots that compare the combined effects of glycan structure and complexation. For peptides exhibiting differences greater than confidence limits, Tables S8 and S9 in the Supplementary Material list Y %E A;B and Y D t ð Þ À A;B , the absolute value of differences in Daltons, for the light and heavy chains of aIL8hFc, respectively. Table S11 lists the Jaccard distances between the 15 states.
Fab of the IgG1-FccRIa complex exhibits effects of glycan structure Proteolytic peptides from the light chain in the G0F-FccRIa complex exhibit depressed H/D as compared to like peptides of apo-G0F (Figure 3 (a)). Four peptides report depressed exchange rates (Y %E A;B % (3.0% ± 0.3%) across residues spanning 52-76 in the V L region. Two peptides report depressed exchange rates (Y %E A;B % (1.9 and 4.6)% ± 0.3%) across residues 149-177 of the C L region. Figure 3(g and h) shows representative Duptake vs log 10 t HDX traces for the apo-and holostates. For light chains of the G2F and SAF glycoforms the differences of D peptide X t HDX ð Þ between apo-and holo-states fall inside the confidence limits of the WAGH(a = 0.025, k = 6) null hypothesis; thus, bars representing these peptides are absent from Figure 3(b and c).
Auxiliary WAGH(a = 0.025, k = 2) analyses, containing variances of one glycoform usually have narrower confidence limits than WAGH(a = 0 .025, k = 6) analyses. Such analyses find significant differences in D peptide X t HDX ð Þbetween apo-and holostates. For the four V L peptides of G2F these differences in D peptide X t HDX ð Þ are % 50% smaller,

Y %E
A;B % 1.6% ± 0.2%, and no differences are found in G2F peptides from the C L region. Similar analyses of SAF datasets find no differences of D peptide X t HDX ð Þ between apo-and holo-states in the V L and C L regions. Table S8 in the Supplementary Material lists the D-uptake relationships found for the light chain.
The WAGH(a = 0.025, k = 6) analyses find six G0F heavy chain peptides within sequence indices 0-75 of the V H region that have differences of D peptide X t HDX ð Þ between apo-and holo-states. For G0F the average integrated Duptake difference between peptides in the apoand holo-states is Y %E A;B % 3.0% ± 0.3%. For the G2F and SAF glycoforms no significant differences are found.
WAGH(a = 0.025, k = 2) pairwise analyses of heavy chain G2F peptides find D-uptake differences in four of the six peptides, but the differences are smaller, Y %E A;B % 1.6% ± 0.2%. Analyses of SAF data find that all peptides exhibit essentially equal D peptide X t HDX ð Þ. In summary, as the monosaccharide chains of N297 glycans in IgG1 increase in size within the IgG1-FccRIa complex, differences of D peptide X t HDX ð Þ in the V L and V H domains diminish in magnitude.

Glycan structure affects H/D exchange rates in apo-IgG1
Comparisons of the HDX-MS measurements for peptides observed from isolated glycoforms characterize the effects of glycan structure on H/D exchange rates. The light chains of apo-G0F, G2F, and SAF glycoforms ( Figure 2(a-c)) exhibit similar H/D exchange rates. Figure 2(b) displays a solitary violet bar representing LC 52-67, which indicates significantly faster H/D exchange rates in apo-G0F than in apo-SAF with an average differential effect of Y %E A;B = (1.6% ± 0.1%) ( Table  S8). Absences of this violet bar in Figure 2(a and c) indicate that the comparisons of apo-G0F vs apo-G2F and apo-G2F vs apo-SAF detect no differences in D-uptake vs t HDX , resulting in featureless similarity plots and Jaccard distances of d J = 0.0 (Table S11).
Peptides from the heavy chains of aIL8hFc glycoforms display distinct H/D exchange rate patterns ( Figure 2(d-e)). Overlapping peptides from the C H 2 domain report H/D exchange rates of residues within 241-252 with average differences of Y %E A;B = (7.9% ± 0.4%) (Table S9). Figure 2(i-j) show representative traces of Duptake vs log 10 t HDX . The order of H/D exchange kinetics rates is apo-G0F > (apo-G2F % apo-SAF).
Overall, H/D exchange rates in apo-G0F are more rapid than observed in apo-G2F and in apo-SAF. However, apo-G2F and apo-SAF report equal H/D exchange rates, as indicated by the featureless Deviation from Similarity vs Sequence # plot (Figure 2(f)) and the derived Jaccard distance of d J = 0.0.

IgG1-FccRIa complexes exhibit depressed H/ D exchange rates in the Fc region
Proteolytic peptides from the Fc region of aIL8hFc show depressed H/D exchange rates in the C H 2 and C H 3 domains of the aIL8hFc-FccRIa complex as compared to like peptides of apo-aIL8hFc ( Figure 3(d-f)). Figure 3(i and j) exhibit traces of D-uptake vs log 10 t HDX for representative peptides. Figure S4 presents traces of other overlapping peptides that violate the null hypothesis. Suppression of H/D exchange between apo-G0F and holo-G0F ranges Y %E A;B % (1.9-9.6)% ± 0.5% (Table S9). At the peptide level the Jaccard distance of d J = 0.38 between the heavy chains of apo-and holo-G0F shows that these species are dynamically dissimilar (Table S11). Table 2 lists the peptides showing distinct H/D exchange rates as a function of glycan structure and lists residue and saccharide interactions, drawn from prior literature, 15,[25][26][27][52][53][54][55] which can account for differences of H/D exchange rates.
Five overlapping peptides from the 241-252 sequence of G0F, G2F, and SAF report substantial suppression of H/D exchange rates, Y %E A;B % 9% (Table S9), between the apo-and holo-states. D-uptake behaviors of these peptides are best represented by traces for peptide HC 242-251 (Figures 3(i) and S6(j)), showing that HDX rates follow the order of apo-G0F > (apo-G2 Peptide HC 334-348 reports on fractions of C H 2 and C H 3. Data for apo-and holo-G0F and G2F show H/D exchange differences for the HC 334-348 peptide, averaging Y %E A;B = (4.2% ± 0.3%). HC 334-348 also shows significant suppression of Y %E A;B = (2.5% ± 0.3%) between apo-and holo-SAF, as determined by a WAGH(a = 0.025, k = 2) analysis. The relative H/D exchange rates for peptide HC 334-348 follow the order of (apo-G0F % apo-G2F % apo-SAF) > (holo-G0F % holo-G2F % holo-SAF), indicating that glycan structure has no detectable impact on local dynamics in this portion of the C H 2 domain.
H/D exchange rates in FccRIa are modulated by IgG1 glycan structure Mixtures containing FccRIa and each aIL8hFc glycoform enable HDX-MS measurements of holo-FccRIa X (X = G0F, G2F, SAF) peptides. WAGH(a = 0.025, k = 4) analysis of the apo-FccRIa and holo-FccRIa X (X = G0F, G2F, SAF) datasets yields six Deviation from Similarity vs Sequence # plots for peptic peptides of FccRIa ( Figure S5). Figure 4 displays comparisons of apo-FccRIa vs holo-FccRIa X glycoforms, and Table S12 lists the Jaccard distances. Table 2 lists the peptides showing distinct H/D exchange rates as a function of glycan structure and offers primary interactions affecting D-uptake rates, drawn from prior literature. [25][26][27] Table S10 lists the D peptide X t HDX ð Þ. Table S7 shows D-uptake vs log 10 t HDX traces of peptides from FccRIa that exhibit exchange rates discordant with the null hypothesis H-bond and van der Waals contacts between Fc glycan-Fc K334 a,b  Table S10 reports kinetic relationships of the discordant peptides. The preponderance of positive bars in Figure 4 (a-c) shows that apo-FccRIa is more dynamic than holo-FccRIa X (X = G0F, G2F, SAF) complexes. The seven peptides presented in Sequence #s 83-113, highlighted by a greenish stripe, exhibit H/D exchange rates in the order: apo-Rc > (holo-Rc G0F % holo-Rc G2F % holo-Rc SAF ), indicating an insensitivity to glycan structure.

Molecular dynamics simulations of Fc-FccRIa complexes report on glycan-mediated binding
Simulations of FccRIa in complexation with Fc regions containing the same glycan structures as G0F, G2F, and SAF were performed to Dt simulation ¼ 1 ls per structure, with four copies of each system. Each model was built using coordinates of a crystal structure Protein Data Bank (PDB entry: 4X4M), 25 using glycoforms built from sequence and energy minimized using the Glycam web server 56 and subsequently modeled into the Fc structure based on existing atomic coordinates. Though the simulations are likely unconverged in terms of sampling excited states present during HDX experiments, they can capture dynamic details unavailable from crystal structures, especially for the glycans, which are highly dynamic on the sampled timescales.
Simulations show that glycan interactions with Fc are mainly localized to the C H 2 domain. Additional interactions involve the C H 3 domain and the D1 and D2 domains of the receptor ( Figure S6). As the length of the glycan arms increase, the glycans can interact with residues residing further away from the covalently bound N297 site in the C H 2 domain of the Fc region.
Interestingly, glycan size is anti-correlated in the simulations with interaction time with the receptor. Figure 5 shows the percentage of the total simulated time each glycan spends interacting with any heavy atom in the FccRIa receptor, where interactions are defined by a distance cutoff of 0.45 nm. For the core monosaccharides, which are either covalently attached to the Fc region located adjacent to the receptor binding site or within 1 or 2 monosaccharides from this covalent attachment point, these interactions are more heavily occupied. Fuc (2) interacts with the receptor a significant amount of the simulated time. Interactions between the glycan arms and the Fc region ( Figure 5(a)) mirror known interactions, with the a1-6 arm showing increased interaction with Fc, overall, as compared to the a1-3 arm. [25][26][27]35,36,55 Glycan interactions with the receptor occur for a lower percentage of time than interactions with domains in the Fc. The highest interactions appear in the GlcNac(1) and Fuc (2). The glycans on the a1-6 and a1-3 arms show a decreasing percentage of the trajectory interacting with the receptor as the glycan increases in size (from G0F > G2F > SAF, Figure 5(c)). The total interaction decreases by %50% as the glycan increases to the G2F form ( Figure 5(c), GlcNAc, %t sim = 29.1%), and then again by a few percent of the simulation time as the glycan arms become larger in the SAF glycoform ( Figure 5(c), %t sim = 11.7%, a2,6-Neu5Ac).
Further analysis of simulated trajectories quantifies the residual contributions to binding energy of the Fc region to the receptor. Figure 6 presents these data colored as non-zero perresidue contribution to binding energy on the surface area of the binding site. The bottom-up view of the receptor and top-down view of the Fc region show the asymmetry of fit during binding. These views also highlight the receptor 'FG loop,' centered on K173 (Figure 6(a)), which has the highest contribution to binding, and fits into the C H 2 domain alongside L235/P329, located in the hinge B region of the C H 2 domain (Figure 6(b)).
Quantitative results of the pairwise energy decomposition between the Fc region and FccRIa receptor is given in Table S13 in the supplementary material.
Interaction energies between the G0F, G2F, and SAF glycoforms and the receptor, presented in Table 3, show that most interactions between glycans and receptor occur with the FG-loop, specifically, the 173 KHR 175 motif. Chain A glycans interact with the receptor FG-loop via the core glycans (1-3-4), but notably, not the fucose. FGloop interactions with chain B glycans occur via Fuc(2) in the a1-6 arm. The sum of binding energy contributions between the receptor and each glycoform is similar between the three glycoform systems. Additionally, the measured and observed volume of binding site residues is similar for each glycoform ( Figures S7 and S8).
Chain A glycans interact with receptor residues, N134, L136, and F146, via the core monosaccharides (1-3-4). These residues exhibit glycan-dependent differences in D-uptake by receptor peptides, as described in Table 2, with dynamics measured as apo-FccRIa > holo-FccRIa G2F > holo-FccRIa G0F > holo-FccRIa SAF (decreasing exchange). Though the energies are similar overall, it is interesting to note that these residues are the only other glycan-mediated contacts in the receptor that report a contribution to binding. Thus, the reach of different glycans can help clarify the distinct Duptake rates observed among glycoforms by HDX-MS measurements (Figures 3 and 4).
As glycans increase in size, only minor changes to the binding surface of the C H 2 domains of the Fc are observed ( Figure S8) and quantified using volume calculations ( Figure S7). Table 3 lists these glycan-receptor interaction energies and Table S13 in the Supplementary Material quantifies the pairwise per-residue interaction energy between Fc and FccRIa residues, showing good correlation to the specific interactions outlined in Table 2. While a few differences in binding enthalpy, if not free energy, of the ground state can be identified, kinetic modulation of Fc-FccRIa binding may account for many of the differences in the HDX-MS datasets. Predicting the kinetic equilibrium between exchanging conformations is beyond the scope of these simulations.
Finally, umbrella sampling simulations were run to gauge the effect of glycans on the Fc-FccRIa complex, and the free energy analyzed using the Weighted Histogram Analysis Method (WHAM). 57,58 Chain A and B C H 2 domains were pulled apart and pushed closer together from a minimum distance of 4.0 nm to 5.5 nm and 3.0 nm, respectively ( Figure 7). Simulations were performed in duplicate for each glycan condition (G0F glycans present (red) or absent (black)). Figure 7 shows resulting free energy curves averaged over the last 500 ps per window from the two independent runs per condition. As the Fc-FccRIa complex is disrupted through displacement of the C H 2 domains, the system free energy increases. However, there is a significant difference between systems with glycans present vs. those with glycans absent as longer distances are reached. It is significantly easier by % 20 kJ/mol to disrupt the complex when glycans are absent.

Fab fragment binding dynamics
Formation of the aIL8hFc-FccR1a complex causes the light and heavy chains of the Fab domain in the G0F glycoform to exhibit decreased D-uptake rates in specific regions of the VL, CL, and VH domains, indicating that bonding Portions of the structure, colored orange, denote sequences that change H/D exchange rates between the apo-and holo-states. Although the IgG1 in Figure 8 is depicted as a canonical Y-structure, three-dimensional cryo-electron tomography shows that the hinge region linking the Fc and Fab domains is remarkably flexible and capable of positioning a Fab in contact with the Fc in multiple orientations. 63 The net suppression of H/D exchange depends upon the ratio of bonding conformation space to nonbonding conformation space. The available bonding conformation space is relatively small and expands slowly with increasing glycan size. The size of the nonbonding conformation space is affected, separately, by the number of conformations available to the glycan and to the Fab. Table 3 Interaction energy between Fc monosaccharides and receptor residues. Residues K173, H174, R175 indicate the FG-loop in FccRIa. The numeric suffix on each monosaccharide label refers to the position shown in Figure 1 for the glycoforms: G0F, G2F, and SAF.   The effects of H/D exchange suppression in Fab are prominently observed in the deuterium content of peptides from holo-IgG1 than from apo-IgG1 glycoforms. This difference originates from the difference of conformation space available to the Fab domains of holo-and apo-IgG1 glycoforms. In IgG1-FccRIa complexes steric constraints close off much of the nonbonding conformational space from exploration by the Fab. This restriction of Fab conformation space results a more favorable ratio of bonding conformation space to nonbonding conformation space, producing peptides from Fab that contain distinctly different amounts of H/D suppression. In apo-IgG1 the nonbonding conformation space is at its maximum, and the Fab domains explore a larger configuration space. Consequently, for all glycoforms the ratios of bonding to nonbonding conformation spaces are at their minima. Thus, the differences in mean deuterium content in peptides from the Fab domains of apo-G0F, apo-G2F, and apo-SAF reside within the confidence limits of the null hypothesis (i.e., l G0F ¼ l G2F ¼ l SAF Þ, in accord with the present observations ( Figure 2). As glycan size increases, the conformation space explored by the monosaccharides of glycans rapidly expands, and the ratio of bonding to nonbonding conformations becomes diminished, resulting in less suppression of H/D exchange at Fab amides. Thus, an expansion of nonbonding conformation space that accompanies increased glycan size accounts for the rapid decline of D peptide apo t HDX ð ÞÀD peptide holo t HDX ð Þ h i X in IgG1-FccRIa complex glycoforms from X = G0F to G2F to SAF, as displayed in Figure 3 and listed in Tables S9 and S10.
An alternate accounting of the present HDX-MS data would posit that the Fab sequences acquire protection from H/D exchange through contacts either with the Fc region, with the D1 domain (e.g., Rc 44-56 and Rc 83-98), or with portions of the protein not detected by HDX-MS measurements. Protein-protein contacts would be expected to exhibit D-uptake suppression of similar magnitude for all glycoforms. For example, due to protein-protein bonding (Table 2), peptides from sequence indices 98-107 of holo-FccRIa X (X = G0F, G2F, SAF) exhibit nearly equal H/D exchange rates. However, because peptides from the Fab exhibit D-uptake rates that vary strongly with IgG1 glycoform, as expected for glycanprotein contacts, we discard explanations involving protein-protein contacts.

Intramolecular glycoprotein bonds affect H/D exchange in the Fc region
The present HDX-MS dataset for apo-aIL8hFc contains five overlapping peptides from the Fc sequence 241-251 (A-strand) exhibiting H/D exchange rates that are sensitive to glycan structure ( Figure 2). Previous HDX-MS studies have reported sensitivities of D-uptake rates to glycan structures by peptides from Fc sequence 241-251 in high mannose and complex type IgG1 glycoforms. 20,32,54,55,60,66,67 These present results are in accord with previous crystal structures, NMR, proteolysis, and HDX-MS studies have reported stabilizing hydrogen bonds and van der Waals contacts between the a1-6 arm and the Fc, specifically, GlcNAc (6) and Gal (7) with Fc F241 and Fc F243. 15,26,[52][53][54] A previous HDX-MS study found that D-uptake rates of G1aF and G0F were greater than G1bF and G2F (Figure 1), which is interpreted to indicate that a1-6 arm Gal(11) interacts with K246 and that Man(9) stabilizes the C H 2 domain through CH-p interactions of Man(9) with Fc F241 and Fc F243. The a1-3 arm Man(5) remains more solvent exposed. 55 These stabilizing interactions by a1-6 arm monosaccharides correlate with a reduction in the conformational flexibility of the Fc and diminish H/ D exchange rates between apo-G0F and apo-G2F in Fc sequence 241-252 (Table 2).
Peptides from the mAb fraction of IgG1-FccRIa complexes report H/D exchange rates in the order apo-G0F > (apo-G2F % apo-SAF) > holo-G0F > (ho lo-G2F % holo-SAF). These data indicate that addition of a2,6-Neu5Ac to G2F affects the stability of apo-IgG1 and IgG1-FccRIa complexes in amounts less than the detection limits of the experiment, resulting in the Jaccard distance, d J = 0.0. This result is in accord with previous HDX-MS and limited proteolysis experiments demonstrating that terminal a2,6sialylation (Neu5Ac(12)) is not destabilizing to the C H 2 domain, whereas a2,3-sialylation, as produced in CHO cell cultures, is destabilizing. 17,53 The a1-6 arm extends from N297 toward the C H 3 domain. 26 Crystal structure data can account for suppressed H/D exchange in HC 334-348 by the presence of van der Waals contacts between the Fc glycan and FccRIa and possibly through secondary effects of a hydrogen bond between the glycan and K334. 30

Transient intermolecular glycoprotein bonds affect dynamics of IgG1-FccRIa complexes
Peptide Fc 83-98 of the FccRIa reports uniform H/D rate depression for all glycoforms, suggesting a possible protein-protein interaction between the D1 domain and the Fc region. Protein-protein contacts in holo-FccRIa between the Fc and the D2 domain can account for the depressed Duptake rates reported by six peptides from FccRIa sequence 98-113. Regardless of the IgG1 glycoform, these peptides of FccRIa exhibit equal depressions of D-uptake rates. This result is in accord with x-ray crystal structures that assign protein-protein contacts between Fc-A and Fc-B to specific residues in the FccRIa D2 domain ( Table 2). [25][26][27] Glycan structure distinctly affects H/D exchange rates in holo-FccRIa X (X = G0F, G2F, SAF) glycoforms. An x-ray structure of the IgG1-FccRIa complex (PDB: 4W4O) assigns intermolecular protein-protein bonds involving resides of C H 2-A and C H 2-B to bind specific residues of the receptor (Table 2) including solvent mediated interactions. On the other hand, peptides from sequence 110-146 of FccRIa exhibit H/D exchange rates in the order apo-FccRI a > holo-FccRIa G2F > holo-FccRIa G0F > holo-FccR Ia SAF . This pattern of increasing conformational stability suggests that all sugar groups within glycans contact the FccRIa receptor, favoring increased conformational stability as the glycan chain lengthens. Specific local effects can disorder the magnitudes of glycan effects.
Peptides from the FccRIa D3 domain report essentially no change in H/D rates between apoand holo-forms, suggesting that the D3 domain plays no direct role in the stabilization of IgG1-FccRIa complexes. This conclusion is consistent with that by Asaoka et al., who report that the absence of D3 domain in a FccRIa minimally reduces stability of FccRI-IgG1 complexes. 68 Extending x-ray structure to a dynamic model To further report on the intermolecular interactions in the Fc-FccRia complex, and to quantify the potential interacting residues between Fc glycans and the receptor, molecular dynamics simulations were run in quadruplicate for each Fc-FccRIa glycoform. MD simulations indicate specific interactions between the glycans and FccRIa receptor form and persist enough to contribute significant energy above thermal fluctuations to the Fc-FccRIa binding (Table 3  and Table S13). Indeed, a reasonable correspondence exists between the interactions denoted in Table 2 and the interaction energies between Fc and FccRIa residues calculated from the simulated data (Table S13). The similarity of calculated interaction energies between receptor residues and glycans qualitatively matches the known similarity in binding with each of these glycoforms present. 33 Additionally, while it is known that the loss of glycans will not prevent binding, the dissociation constant increases significantly. 25 This observation is directly corroborated by the PMF calculations (Figure 7), showing a decrease in free energy required to disrupt the unglycosylated Fc-FccRIa complex.
The present results support the role of the FG loop, originally proposed in Lu et al., 25 and show with energy decomposition the direct interactions of this loop with all glycans (Table 3 and  Table S13), as well as the "lock-and-key" fit of this loop with L235 in C H 2 domain. This bonding configuration is described in previous reports ( Figure 6). 26,27 The MD simulations, though not quantitatively predicting the HDX-MS data, helps resolve seemingly disparate results from crystal structures of the complex by virtue of modeling the glycans as a dynamic entity.
aIL8hFc mAbs were expressed from CHO DP-12 clone#1934 cell line (American Type Culture Collection, Catalogue # CRL-12445). Briefly, cells were inoculated at 2.5x10 5 cells/ml into 250 mL shake flasks each containing 80 ml Biogro CHO media (Biogro Technologies Inc, Winnipeg, Canada) with 25 mmol/L glucose and 0.5 g/L yeast extract (BD, Sparks, USA). Cells were cultured by incubating the shake flasks in a humidified incubator (Nuaire, Minnesota, USA) at 120 rpm, 10% CO 2 and 37°C. After 4 days growth, the cultures were centrifuged at 1500g n for 5 min to collect the culture supernatant that was filtered through a 0.2 lm Steritop filter (EMD Millipore, Etobicoke, ON).
IgG1 glycoforms of aIL8hFc were prepared by solid-phase enzymatic remodeling. 40 Since S1F and S2F glycoforms were prepared using human sialyltransferase, both sialylated structures have a2,6-linkages. Each IgG1 glycoform sample comprised % 100 mg material. Sequence alignment shows that the Fc shares 99.4% identity with Uni-Prot accession number P01857. Table S1 in the Supplementary Material lists the sequence of light and heavy chains of aIL8hFc.

Measurement of glycan distributions
Relative abundances of aIL8hFc glycoforms in samples were determined by releasing N-linked glycans from IgG1 with peptide-N-glycosidase F; tagging the filtered, released glycans with fluorescent 2-aminobenzamide (2AB) label; separating tagged glycans by hydrophilic interaction liquid chromatography (HILIC), and evaluating glycan abundance from peak areas of observed fluorescent signals. 49 Peptide identifications from mass spectrometry data
Samples were equilibrated at 1°C. HDX was conducted on a HDX PAL robot (LEAP Technologies, Carrboro, NC). Protein solutions (5 mL) were diluted into 31 mL D 2 O buffer (10 mmol/L sodium phosphate, 137 mmol/L sodium chloride, 2.7 mmol/L potassium chloride at pD 7.4) at 25°C. After immersion in D 2 O solution for selected times (t HDX = (0, 30, 300, 900, 3600, and 14,400) s) the HDX sample was quenched by mixing with 30 mL quench buffer (4 mol/L GdmHCl, 0.2 mol/L sodium phosphate, 0.5 mol/L TCEP at pH 2.5) at 1°C. This solution was injected into a liquid chromatography apparatus that housed its LC connection lines and valves in a refrigerated compartment at %1°C. The quenched solution flowed through the immobilized pepsin column for 3 min at 15°C.
Peptic peptides in the solution digest were trapped on a C18 guard column (%1°C; 1.0 mm dia. Â 5 mm length, 5 mm particles; Grace Discovery Sciences, Deerfield, IL) and separated with a C18 analytical column (%1°C, 1.0 mm dia. Â 50 mm length, 1.9 mm particles, Hypersil GOLD; Thermo Scientific) via a Thermo Scientific Ultimate NCS-3600RS binary pump with a 9.5 min gradient operated with a binary mixture of solvents A and B at 50 mL/min flow rate. The gradient settings used were: 5-35% solvent B for 3 min, 35-60% solvent B for 5 min, 60-100% solvent B for 0.5 min, isocratic flow at 100% solvent B for 0.5 min, and a return in 5% solvent B for 0.5 min. Solvent A was water containing 0.1% formic acid and solvent B was 80% acetonitrile and 20% water containing 0.1% formic acid.
Peptides were measured on a Thermo Orbitrap Elite. The instrument settings were spray voltage, 3.7 kV; sheath gas flow rate, 25 (arbitrary units); capillary temperature, 275°C. In the Orbitrap stage MS spectra were acquired with the resolution set at 25,000. 72

HDX-MS analyses
The program, HDX Workbench, was used to compute, %E peptide i ;X t HDX ð Þ, which is the percent of peptide undergoing deuterium exchange obtained for the ith measurement of a peptide in state X (e.g., apo-and holo-glycoform). 73 The recovery parameter in the software was set to 100%. Deuterium mass (Da) of a peptide from state X is computed using: where F D 2 O is the molar fraction of solution D 2 O, n is the number of amino acids and p is the number of prolines in the peptide excluding the first two N-terminal residues, and m H þ and m D þ are proton and deuteron masses. We computed the mean for n measurements, D peptide X t HDX ð Þ, and its associated sample standard deviation, s k .
Since this study concurrently evaluates four to six protein states, we employed Welch's one-way ANOVA and Games-Howell post hoc calculations (WAGH) to test the hypothesis that means; u X ("D peptide X t HDX ð Þ), for k ! 2 protein states fall within the chosen 1 À a ð ÞÂ100% confidence level, affirming the null hypothesis robust for treating datasets of unequal variance and sample size, [74][75][76][77][78][79][80][81] and it provides moderate control against a-inflation. 76,78 The Games-Howell post hoc procedure applies a distinct confidence criterium, computed from pooled measurement variances, for each pairwise test of the null hypothesis (Tables S3-S4). Software for comparison procedures was written in Labview 7.1 (National Instruments Co., Austin, TX). HDX-MS measures the aggregate rates of Duptake by backbone amides of the peptide. Although the overall temporal D-uptake trace for each peptide sequence is distinct, like peptides from two or more protein states sometimes exhibit equal D peptide Þat early and late t HDX . 82 We designate a peptide sequence of State W to be dynamically dissimilar to States X, Y, Z, . . . when its D peptide W t HDX ð Þ trace falls outside confidence levels at most measurement t HDX . The average absolute value and the sign of the average magnitude of the perturbations affecting the amide deuterium uptake rates between states A and B, D D peptide A;B , are computed: where "m" is the number of t HDX for which the difference of

Computation of dynamical dissimilarity
A census of the residues within peptides observed for states A and B enables direct computation of the Jaccard distance d J (A,B): where S ! i is the subject protein vector composed of elements representing residues indexed from 0 = 1 to 0 = m; m is the protein sequence length; p is the number of prolines in the sequence between 0 = 3 and 0 = m; P ! obs iþ1 j; k ð Þ vectors represent members of the set of n peptides common to datasets of states A and B; j; k ð Þ vectors represent members of the subset of n' peptides containing deuterium mass falling within the critical threshold for significance of the null hypothesis (l A ¼ l B ); j and k are the peptide start and end indices referenced to subject protein sequence index 0; and "" is the Hadamard product operator. Vector notation recognizes that the sequence index 0 runs from N-to C-termini. All vectors contain m elements. Elements of S ! i¼0 with indices 0 > 2 that represent exchangeable residues are set equal to one, and elements representing proline and indices of 0 2 are set equal to zero. Elements of P ! obs j j; k ð Þ and P ! l A ¼l B j ðj; k Þ with indices 0 = 1, . . . j + 1 and 0 > k that represent exchangeable residues are set equal to one, and elements between 0 = j + 2 and 0 = k are set equal to zero. (This indexing scheme acknowledges that rapid back-exchange generally erases information from the first two residues.) The summed vector elements of S ! n and S ! n 0 , each resulting from recursive operations involving S ! i , are used in the computation of Jaccard distance, which characterizes the dynamical dissimilarity of states A and B.

Molecular dynamics simulations
All atom structures of Fc bound to FccR1a were created using coordinates from a crystal structure (PDB: 4X4M). 25 Missing loop density for receptor residues 219-223 were adopted from a second structure (PDB: 4W4O). 26 Point mutations to match the sequence used in this work were introduced using SWISS-MODEL rotamer libraries. 83 Glycans (G0F, G2F, SAF) were built from sequence using GLYCAM-Web, 56 minimized by the web server, and grafted onto the Fc region.
Structures were built using the FF14SB protein forcefield, 84 GLYCAM_06j-1 carbohydrate force field, 56 and SPC/E water model, 85 with Joung-Cheatham monovalent ion parameters tuned to that water model. 86 Three chlorine ions were added to neutralize the charge in the system. About 48,500 water molecules and 367Na + and Clions were added, resulting in a 350 mmol/L concentration. Ions were randomly swapped with water positions so that ions were at least 0.60 nm from any solute atom and 0.40 nm from each other, yielding four separate starting solvent orientations per glycoform.
Minimization and relaxation were performed according to Roe and Brooks 87 on CPUs and GPUs, respectively, using AMBER18, 88 with decreasing positional restraint weights. Final equili-bration of each system with no restraints was performed for 1 ns using a Monte Carlo (MC) barostat 89 in an isothermal-isobaric (NPT) ensemble, with pressure = 101.325 kPa. Production simulations were run using AMBER18 GPU code 90 with SHAKE 91 to constrain bonds to hydrogen, allowing a 2 fs timestep. Simulations were run in NTP, with a MC barostat set to 101.325 kPa and temperature regulated by a Langevin thermostat. 92 Four simulations per glycoform (G0F, G2F, SAF) were run to 1 ls each, allowing accumulation of statistics on this timescale. Analysis was performed using CPPTRAJ. 93 A distance cutoff of 0.45 nm was chosen for the 'mask' command since this can reasonably be considered the longest distance for significant inter-atomic interactions. 42 MM-GBSA analysis was performed using the Onufriev-Bashford-Case model (OBC, igb = 2 in Amber), with mbondi2 Born radii parameters and a salt concentration of 200 mmol/L. 94 Pairwise per-residue energy decomposition was performed on 1000 frames per simulation with 1-4 force field terms added to the internal potential terms (idecomp = 3 in Amber).

Umbrella sampling simulations
The first two minimized structures for the G0F were stripped of their glycoforms and reminimized as described above to generate "no glycosylation" starting structures for umbrella sampling. To disrupt the structure, a reaction coordinate was selected to pull apart the Fc domain by restraining the distance between the centers of mass of each chain's C H 2 domain Ca atoms (104 atoms per com restraint point). Each of these structures was restrained to a distance 0.1 nm apart from 3.0 to 5.5 nm, totaling 26 windows. The structures were equilibrated at these distances using 41.84 kJ/mol (10 kcal/mol) restraints over a 500 ps simulation in an NPT ensemble with simulation parameters described above. The resulting structures were used for production dynamics of 1 ns per window at 0.05 nm distances from 3.0 to 5.5 nm, totaling 52 windows. Restraints were kept at 41.84 kJ/mol (10 kcal/mol). The last 500 ps of these windows were used in WHAM analysis to calculate free energies. 57,58 The procedure was performed independently for each set of starting structures (two per glycosylation condition, four total), and averages and standard deviations are reported for these independent simulations.

Disclaimer
Certain commercial equipment, instruments, or materials are identified in this paper to foster understanding. Such identification does not imply recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equipment identified are necessarily the best available for the purpose.