CH2 domain orientation of human immunoglobulin G in solution: Structural comparison of glycosylated and aglycosylated Fc regions using small-angle X-ray scattering.

The N-linked glycan in immunoglobulin G is critical for the stability and function of the crystallizable fragment (Fc) region. Alteration of these protein properties upon the removal of the N-linked glycan has often been explained by the alteration of the CH2 domain orientation in the Fc region. To confirm this hypothesis, we examined the small-angle X-ray scattering (SAXS) profile of the glycosylated Fc region (gFc) and aglycosylated Fc region (aFc) in solution. Conformational characteristics of the CH2 domain orientation were validated by comparison with SAXS profiles theoretically calculated from multiple crystal structures of the Fc region with different CH2 domain orientations. The reduced chi-square values from the fitting analyses of gFc and aFc associated with the degree of openness or closure of each crystal structure, as determined from the first principal component that partially governed the variation of the CH2 domain orientation extracted by a singular value decomposition analysis. For both gFc and aFc, the best-fitted SAXS profiles corresponded to ones calculated based on the crystal structure of gFc that formed a "semi-closed" CH2 domain orientation. Collectively, the data indicated that the removal of the N-linked glycan only negligibly affected the CH2 domain orientation in solution. These findings will guide the development of methodology for the production of highly refined functional Fc variants.

Introduction

Immunoglobulin G (IgG), the antibody class that is most widely produced for therapeutic purposes,[1,2] is a large multi-domain protein with an N-linked glycan at residue N297 of the CH2 domain in the crystallizable fragment (Fc) region. In recent years, therapeutic use of aglycosylated IgG molecules has gained popularity because the lack of glycan is thought to bypass issues with IgG glycan heterogeneity and suppress undesired effector function.[3-5] Further, aglycosylated IgG molecules can be produced in lower eukaryotes and bacteria, with considerable advantages in terms of short culture time and potential for excellent scale-up in comparison with the conventional IgG generation using mammalian cells.[6,7] However, it is generally accepted that removal of the N-linked glycan from the Fc region affects the properties of the protein, e.g., the stability and function of the Fc region.[8-14] Any aglycosylated IgG developed as a therapeutic should thus be evaluated to determine the effect of deglycosylation on these properties.

It has long been assumed that removal of the N-linked glycan might result in the alteration of the orientation of the CH2 domain in the aglycosylated Fc region (aFc).[15-17] For example, an aFc crystal structure (Protein Data Bank accession number, PDB: 3S7G) revealed a closed CH2 domain orientation, with the two CH2 domains more proximal to one another than in the crystal structure of the glycosylated Fc region (gFc) (PDB: 4W4N).[15,18] In contrast, a recent residual dipolar coupling (RDC) analysis by nuclear magnetic resonance (NMR)[19] examined the CH2-CH3 domain orientation in both gFc and aFc in solution, as judged by the relative orientation of each observable N-H bond vector of 15N-labeled gFc and aFc. The experiment revealed that the CH2-CH3 domain orientation in both gFc and aFc is similar to that in a crystal structure of gFc (PDB: 1L6X[20]). The authors proposed that “the effect of glycosylation on CH2 domain orientation is restricted to small amplitudes or small populations”. Their finding suggested the observed perturbation of the CH2 domain orientation in the crystal structure of aFc was mainly attributed to the crystal packing forces.[21]

In this study, using small-angle X-ray scattering (SAXS), a powerful tool for the evaluation of protein conformation in solution,[22] we proved that the removal of the N-glycan only minimally perturbed the CH2 domain orientation in aFc in solution. In SAXS, the SAXS profile [I(q)] is expressed by a Fourier transformation of the pair-distance distribution function [P(r)], and can also be theoretically calculated if the three-dimensional coordinates of a crystal or modeled protein structures are available.[23] Comparison of the experimentally determined and theoretical SAXS profiles validates the protein conformational characteristics in solution. Accordingly, we determined the SAXS profiles of gFc and aFc in a solution at pH 7, and characterized the domain orientation by comparing the experimentally derived profiles with the theoretical SAXS profiles which were calculated using multiple crystal structures and homology modeling.

Results

Measurements of the SAXS profiles

To investigate the conformational characteristics of gFc and aFc in a solution at pH 7, we collected their SAXS profiles (Figure 1). The shapes of gFc and aFc SAXS profiles were similar, but clearly different. The radius of gyration (Rg) and the scattering intensity at zero-angle [I(0)] were calculated from the Guinier approximation in the low q region (Figure 1(a) and Table S1). The Rg of gFc and aFc showed slight concentration dependence within the range of 2 to 5 mg/mL of proteins, suggesting weak intermolecular interaction[24-26] (Figure S1). Therefore, we estimated an interference-free SAXS profiles of gFc and aFc by extrapolating the measured SAXS profiles to an infinite dilution condition (Figure S2).[27] Rg from the interference-free SAXS profiles of gFc and aFc was determined to be 26.44 ± 0.31 Å and 28.88 ± 0.31 Å, respectively. Molecular masses estimated from I(0) of gFc and aFc were 53.3 kDa and 50.7 kDa, respectively, which agreed with the molecular masses determined using mass spectrometry (Table S1). Although there was little difference between the raw measured and normalized interference-free SAXS profiles in the small q region (0.005 to 0.05 Å−1), both profiles were employed in further analyses. A Kratky plot [q2I(q) vs. q] is used to characterize the structural properties of a protein, such as its globularity and flexibility.[28] The Kratky plots of gFc and aFc displayed a bell-shaped peak pattern, indicating the formation of a globular structure (Figure 1(b) and Figure S2b). Both plots were obviously bimodal. By using the value of the scattering parameter when the intensity of the Kratky plot is at a maximum (qm) [i.e., dq2I(q)/dq = 0], Rg can be also estimated from the Guinier approximation, as/qm.[29] Further, I(0) may be estimated using the intensity at qm in the Kratky plot, i.e., I(0) = [qm2I(qm)]exp(1)/qm2. Rg and I(0) values of gFc and aFc determined using qm in the smaller q region of the two peaks (qm,small) were close to the Rg and I(0) values determined by the linear fitting of lnI(q) vs. q2 (Table S1). These results indicated that gFc had a smaller Rg than aFc despite having a larger molecular mass, which was atypical for globular proteins.

Figure 1.

The experimentally determined SAXS profiles of 3 mg/mL protein concentration. q, I(q), r, and P(r) are a scattering parameter, a scattering intensity, a distance between electrons in a particle, and a pair-distance distribution function, respectively (see SUPPLEMENTARY METHOD). The log-linear plot (a) and Kratky plot (b). Inset in (a), Guinier plot. The black straight line indicates the fitted region of the Guinier approximation. The pair-distance distribution function calculated based on the experimentally determined SAXS profiles (c). The dotted and solid lines indicate the direct and indirect Fourier transformation, respectively, of the determined SAXS profiles.

To reveal the origin of the difference in Rg between gFc and aFc, we analyzed a pair-distance distribution function [P(r)]. P(r) derived from a Fourier transformation of the SAXS profile provides insight into additional structural properties of a protein in terms of real space (Figure 1(c) and S2d). The P(r) of gFc [P(r)gFc] in the 10–60 Å r range was larger than that of aFc [P(r)aFc], while P(r)gFc and P(r)aFc agreed with each other in the r ranges of 0–10 Å and 60–100 Å. Rg, calculated as the center of gravity of P(r)gFc, was smaller than that of P(r)aFc (Table S1), which was consistent with the result of the Guinier approximation. We interpreted these results as follows. The increased integrated area of P(r)gFc in the 10–60 Å range was derived from the scattering from the N-linked glycan of gFc, and the shift of the center of gravity in P(r)gFc was affected by such increased distance distribution from the N-linked glycan. Moreover, the intra-domain structure and the inter-domain orientations, corresponding to the shortest (0–10 Å) and longest (60–100 Å) distances, respectively, in gFc and aFc were equivalent. This implied that the change of the CH2 domain orientation caused by deglycosylation was small. In other words, the removal of the N-linked glycan corresponded to the generation of a hollow sphere from a solid sphere, without altering the size of the outer shell.

To demonstrate the appropriateness of these interpretations, we further evaluated the structural characteristics of gFc and aFc in solution by comparing the experimentally determined SAXS profiles with the theoretical SAXS profiles calculated based on the available crystal structures of gFc and aFc.

Elucidation of attributes that affect the variation among the Fc crystal structures

Due to advances in X-ray crystal structure analysis, a number of Fc crystal structures are now available. These structures exhibit a well known and appreciable variation with respect to the spatial orientation of the CH2 domain, while the CH3 domain is well aligned.[17,30] As a typical example, one aFc crystal structure (PDB: 3S7G) shows a closed CH2 domain orientation, in which the CH2 domain distance, often determined as the distance between two P238 residues, is relatively small (10.7 Å, measured using Pymol software) compared with that of gFc (e.g., PDB: 4W4N,[18] 19.5 Å), while gFc in complex with Fc gamma receptor (FcγR; e.g., PDB: 4W4O[18]) shows an open CH2 domain orientation in which the CH2 domain distance is relatively large (25.2 Å).

In this study, we first elucidated the intrinsic or extrinsic attributes that affected the variation among the available Fc crystal structures using principal component analysis (PCA). PCA is often used to identify important protein conformations, motions, and noise factors from data with a large number of dependent variables, such as a molecular dynamics trajectory.[31] Singular value decomposition (SVD), often used in PCA, is a mathematical approach of calculating eigenvalues and eigenvectors from data matrix, and transforming a large number of dependent data variables into a smaller number of independent variables (i.e., principal components), which simplifies the phenomena of interest.[32] We performed SVD on a dataset comprising 42 available Fc crystal structures. We extracted the principal components as the attributes independent of each other that intrinsically or extrinsically contributed to the variation between the Fc crystal structures. Each dataset for an individual crystal structure without the N-linked glycan exhibited a deviation from the averaged three-dimensional coordinates of Cα atoms between S239 to S442 when the flexible hinge and C-terminal regions were removed.

The results of SVD are shown in Figure 2. The singular values from the first to seventh components were relatively larger than others. The sum of their contribution ratios was 94.8% (Figure 2(a)). These observations implied that the variation among Fc crystal structures could be mainly described by these seven principal components. The left singular vector from the employed dataset could be interpreted as the direction and magnitude of deviation from the averaged three-dimensional coordinates of each Cα atom, as visualized by a porcupine plot[33] in Figure 2(b) and Figure S3. The first left singular vector showed the direction of the CH2 domain orientation to open or closed orientation from the averaged Cα positions (Figure 2(b)). The left singular vector of the second to seventh principal components described the deviation of the Cα atoms of the CH2 domain from the averaged position (Figure S3). We noted that the fifth left singular vector showed a deviation of the C′E loop (Figure S3i). The intrinsic flexibility of this region agreed with the conclusions of previous hydrogen/deuterium exchange mass spectrometry and NMR analyses.[8,19,34] The magnitude of the right singular vector could be interpreted as weight factor of the left singular vector of each crystal structure (Figure 2(c) and Table S2). For the first principal components, the magnitude of the first right singular vector of each crystal structure correlated with the trends of the open or closed CH2 domain orientation compared with the averaged orientation among the 42 crystal structures. For example, the value of the first right singular vector of the crystal structure of aFc (PDB: 3S7G), which typically adopts the closed orientation, was large and negative (−0.399), while it was large and positive (e.g., 0.131 in 4W4O) in the crystal structures of the Fc region in complex with FcγR, which adopts the open orientation. The values of the first right singular vectors of most gFc structures (not in complex) were slightly negative, indicating that they adopted a somewhat closed CH2 domain orientation. In this study, we defined the CH2 domain orientation of gFc (not in complex) as a ”semi-closed” orientation. In gFc in complex with other binding proteins, the value of their first right singular vector varied from negative to positive.

Figure 2.

Results of the SVD analysis. The singular value of each component is shown in (a). Inset in (a), the contribution ratio (red bars) and the cumulative contribution ratio (blue dots). The porcupine plots of the first left singular vector are given in (b). The yellow line indicates the averaged three-dimensional coordinates of Cα atoms calculated based on the crystal structure dataset. The green spine indicates the direction and magnitude of the deviation of the first left singular vector from the averaged three-dimensional coordinate of Cα atoms. The first right singular vector of each crystal structure is shown in (c). BP denotes a binding protein such as protein A (See Table S2).

Theoretical SAXS profiles based on crystal structures

To demonstrate the structural similarity between gFc and aFc in solution, we calculated theoretical SAXS profiles based on the individual Fc crystal structures, and fitted these to the experimentally determined SAXS profiles. Evaluation of the characteristics of the best-fitted crystal structure was then confirmed based on the singular vectors deduced from the SVD analysis described above.

We first prepared an intact model structure of gFc and aFc inclusive of the hinge and C-terminal regions by homology modeling. We prepared 80 template structures of gFc and aFc covering the region between S239 to S442 by extracting from 40 crystal structures of the Fc region, and calculated 100 intact model structures from each set of the 80 template structures (Figure 3(a)). As for the N-linked glycan coordinates, the rigid coordinates from a specific gFc crystal structure were used as a template (See Materials and Methods section). This was considered reasonable in light of previous SAXS analysis of gFc, in which the N-linked glycan was successfully assumed to fill the internal space between CH2 domains.[26,35] Then, the SAXS profiles of each intact model structure of gFc and aFc were calculated and fitted to the experimentally determined SAXS profiles at 3 mg/mL protein concentration of gFc and aFc, respectively. The reduced chi-square values from the fitting analysis of 100 intact model structures showed a random-like distribution (Figure S4) because of the conformational difference of the added hinge and C-terminal regions in each intact model structure. Nevertheless, the differences between the reduced chi-square values among the 100 intact model structures were smaller than among the 80 template structures, indicating that the variation of the CH2 domain orientation affected the shape of SAXS profiles to a greater extent than the conformational variation in the hinge and C-terminal regions. We then extracted a median of the reduced chi-square values of the 100 intact model structures from each of the 80 template structures, and 80 intact median model structures were used as representative model structures.

Figure 3.

Homology modeling and fitting. (a) Ten example model structures of gFc (left) and aFc (right) calculated from the template structure of 1HZH. (b) A correlation plot of reduced chi-square values for 3 mg/mL SAXS profiles of gFc and aFc. BP denotes a binding protein. The black straight line represents the result of the regression analysis. The inset contains the plot of the lowest reduced chi-square value. (c–f) SAXS profiles with the lowest reduced chi-square values calculated based on five model structures. Log-linear plots (c, d) and Kratky plots (e, f) are shown. The gray dots indicate the experimentally determined SAXS profile of 3 mg/mL protein concentration.

As shown in Figure 3(b), the reduced chi-square values for 3 mg/mL protein concentration from each representative model structure of gFc correlated well with those of aFc. We note that the same correlation was also observed in the reduced chi-square values for the interference-free SAXS profiles (Figure S5). This indicated that the presence or absence of the N-linked glycan only minimally affected the variation of the goodness of fit, and the difference in the experimentally determined SAXS profiles between gFc and aFc mainly stemmed from the scattering of the N-linked glycan in gFc. We next statistically evaluated the goodness of fit and determined the best-fitted representative model structure, by calculating the probability value (P-value) using a pair-wise correlation map (CorMap) analysis.[36] It should be noted that the reduced chi-square value does not necessarily represent statistical validity because of the difficulty in determining the accurate experimental errors.[36] In fact, the propagation of the experimental errors in the processes of calculating the one-dimensional SAXS profiles and averaging the different data frames rendered the experimental errors relatively large, which resulted in quite low reduced chi-square values (<<1.0, Table S2). When the P-values were calculated using the CorMap analysis, most of the representative model structures with higher chi-square values exhibited P-values below 0.01, indicating that the null hypothesis (H0) of similarity of the experimental and theoretical SAXS profiles could be rejected (Table S2). This also indicated that the theoretical SAXS profiles using those representative model structures were insufficient to express the experimental SAXS profiles. In contrast, six representative model structures generated using the crystal structures of 1HZH, 1OQX, 2DTS, 3AVE, 5U4Y, and 5U52 had P-value above 0.01 for both gFc and aFc fitting, indicating that the null hypothesis (H0) could not be rejected. In statistical analysis, failure to reject the null hypothesis (H0) does not ascertain statistical similarity. Nevertheless, the theoretical SAXS profiles generated from those six representative model structures indeed all agreed very well with the experimental SAXS profiles (Figure 3(c–h)). Therefore, we concluded that those six representative model structures could indeed represent the experimental SAXS profiles, considering the best-fitted representative model structures.

To confirm the characteristics of the best-fitted representative model structures, we generated correlation plots of the reduced chi-square values and the right singular vectors of each principal component extracted by the SVD analysis described above (Figure 4). The right singular vector of the first principal component, which reflected the open or closed orientation characteristics, showed an association with the reduced chi-square value of each representative model structure, while other principal components did not (Figure S6). The first right singular vectors of the best-fitted representative model structures were in the range of −0.123 to 0.053 (Table S2), and most of their CH2 domain orientations were classed as semi-closed. This indicated that the CH2 domain orientations in both gFc and aFc were almost identical.

Figure 4.

Correlation plots between the chi-square value for 3 mg/mL SAXS profiles and the first right singular vector; gFc (a) and aFc (b). BP denotes a binding protein.

Discussion

In this study, we experimentally determined the SAXS profiles of gFc and aFc to characterize their conformations in solution. The shapes of the experimentally determined SAXS profiles of gFc and aFc were similar, but also clearly different (Figure 1(a,b) and S2a-c). Interestingly, Rg of gFc was smaller than that of aFc, despite the larger molecular mass of gFc than aFc (Table S1). The reason for this inconsistency became clear after comparing their P(r) (Figure 1(c) and S2d). We interpreted these observations to indicate that the difference in P(r) between gFc and aFc was derived from the scattering from the N-linked glycan of gFc; that the intra-domain conformation and the inter-domain orientation of polypeptide portions were equivalent; and that the change of the CH2 domain orientation in aFc caused by deglycosylation was small. To demonstrate that the presence of the N-linked glycan was sufficient to lead to the difference in the SAXS profiles of gFc and aFc, we further evaluated the structural characteristics in the experimentally determined SAXS profiles of gFc and aFc by performing a fitting analysis using theoretical SAXS profiles based on the existing Fc crystal structures (Figure 3 and S5). The data indicated that the measured SAXS profiles of both gFc and aFc were best-fitted using the theoretical SAXS profiles based on the common Fc crystal structure, which adopted a “semi-closed” CH2 domain orientation (Figure 4).

SAXS analysis of gFc and aFc was previously performed by Borrok et al.[15] The authors found that Rg is larger in aFc than in gFc, and concluded that aFc forms an open orientation in solution. The SAXS analysis presented here reproduced the observation that Rg of aFc was larger than that of gFc (Figure 1 and Table S1), indicating that the divergent conclusions did not result from the differences of experimental conditions, e.g., sample preparation methods and the measuring equipment. The N-linked glycan of gFc is located near the center of the three-dimensional Fc structure, filling the hollow between the two CH2 domains. When the gFc and aFc structures are modeled as solid and hollow spheres, respectively, Rg of a solid sphere is smaller than that of a hollow sphere whose outer shell has the same diameter (see also SUPPLEMENTARY DISCUSSION).[37] Therefore, it is reasonable to assume that the scattering from the N-glycan reduces the Rg of gFc compared with aFc, without any changes in the CH2 domain orientation.

The analysis presented here provided information about the characteristics of an averaged conformation of gFc and aFc in solution, and we did not analyze a conformational distribution. As for the distribution, the previous study using RDC technique suggested that gFc has little conformational deviation in solution.[19] In contrast, Remesh et al. recently reported an obvious distribution in the CH2 domain orientation of gFc by combining atomistic modeling with experimental SAXS data in a multi-conformational analysis.[26] Their SAXS profile at pH 7.5 (SASDBD accession code: SASDDT4) is in excellent agreement with the SAXS profile in our study. The scattering curves measured by Remesh et al. were higher (~0.5 Å−1) than the scattering vectors in our experiment and they used these data to refine the characteristics of the local conformations including the N-terminal hinge, the N-linked glycan, the CH2-CH3 linker, the C-terminal region, and the short flexible loop in the CH3 domain. They showed that the derived SAXS profiles can be best explained by the weighted sum of theoretical SAXS profiles composed of four dominant model structures with different CH2 domain orientations. To clarify the relevance between the findings of Remesh et al. and our analytical results, we overlaid their four model structures on the correlation plots of 40 crystal structure data (Figure S7). This showed that the four model structures fell in the range of the semi-closed CH2 domain orientation, and neither conformation was classified into either the closed or open conformations. This indicates that the distribution of gFc conformations is constrained within a relatively narrow conformational space, compared to the vast conformational space that includes the closed and open CH2 domain orientations. In addition, Yogo et al. detected the open orientation of gFc in complex with FcγRIIIb, in solution, using deuteration-assisted small-angle neutron scattering (SANS).[38] An important conclusion reported by the authors is that the experimental population of the semi-closed orientation and the open orientation, in complex with FcγRIIIb, was in agreement with the theoretical population of free and bound form, calculated based on the dissociation constant between gFc and FcγRIIIb. This suggests that the transition of the CH2 domain orientation in gFc from semi-closed to open would be governed by a classical induced-fit mechanism. Therefore, in solution and in the absence of FcγRIIIb, gFc would be expected to predominantly adopt the semi-closed orientation, within the narrow range of conformational distribution. As for the conformational distribution of aFc, recent studies employing molecular dynamics simulation and Förster resonance energy transfer (FRET) analyses proposed that deglycosylation increases the range of conformational distribution in the CH2 domain,[30,33,39-41] although the characterization of the conformational distribution of aFc by SAXS has not been performed.

Deglycosylation of the Fc region causes a reduction of its binding affinity for FcγR.[14,19,42-44] One hypothesis explaining this phenomenon involves alteration of the static CH2 domain orientation by deglycosylation,[15-17] but the results of SAXS analysis presented in this study reasonably disproved it. Subedi et al. proposed that the conformational change in the C′E loop in the CH2 domain that results from deglycosylation is responsible for the reduced affinity to FcγRIIIa.[19] The authors found that the magnitude of the chemical shifts of Y300 in the C′E loop in a range of Fc variants was strongly correlated with FcγRIIIa affinity. In addition, protein engineering approaches involving random amino acid replacement have succeed in generating aFc variants with recovered FcγR affinity.[3,5] Ju et al. performed single-molecular FRET analysis with one such aFc variant (Fc5), in which the FcγRI affinity was recovered by amino acid replacement in the CH3 domain,[45,46] and demonstrated that the restriction of the orientation and domain motion of the CH2 domain is critical for the recovery of FcγR affinity.[39] Combining the results of high-resolution analyses, such as SAXS and NMR, with the Fc variant repertoire that exhibits recovered affinity will contribute to further elucidation of the molecular mechanisms that underpin the reduced FcγR affinity associated with deglycosylation.

In conclusion, using SAXS, we proved that the removal of the N-linked-glycan from the Fc region only minimally perturbed the CH2 domain orientation. The pair-distance distribution function calculated from the experimentally determined SAXS profiles indicated that the conformational characteristics of the CH2 domain orientation between gFc and aFc, were equivalent. Further, comparison of the experimentally determined SAXS profiles and the theoretical SAXS profiles calculated based on the model structures revealed that both gFc and aFc adopted a “semi-closed” CH2 domain orientation in solution. These findings will guide the development of methodology for the production of highly refined functional Fc variants.

Materials and methods

Sample preparation

Glycosylated human IgG1 Fc region (gFc) was prepared by papain digestion, and the recombinant human aglycosylated IgG1 Fc region (aFc) was produced using the Escherichia coli expression system, as previously described.[47] The averaged molecular masses of the purified samples (> 95% monomeric) were determined and analyzed using Acquity UPLC I-Class and SYNAPT G2-Si systems (Waters, Millford, MA, USA) and the Expressionist Refiner MS software (GeneData, Basel, Switzerland), respectively (Figure S8 and Table S1). The mass spectrum of gFc revealed several peaks representing a repertoire of glycoforms (Figure S8a). The molecular mass of the main peak agreed with the theoretical molecular mass estimated based on the amino acid sequences of the two heavy chains (from T236 to G447) and the attached G0F/G1F glycan. The experimental information on the glycan was used for the construction of N-linked glycan template during homology modeling, as described later. Papain digestion occurred at the same position as previously reported.[48] The averaged molecular mass of aFc agreed with the theoretical molecular mass estimated for the D232–K448 fragment with an additional N-terminal alanine (Figure S8b). The formation of the interchain disulfide bridges in the hinge region of aFc has been confirmed in a previous study.[47] The original buffer in the purified protein samples (2–5 mg/mL) was exchanged to 20 mM citrate-phosphate buffer (pH 7.0) by dialysis.

SAXS

The SAXS analyses were performed using the beamline BL-10C at the Photon Factory (PF) of the High Energy Acceleration Research Organization (KEK; Tsukuba, Japan). The X-ray wavelength was 0.12 nm; the camera length (2028 mm) was determined by using a scattering pattern of silver behenate as a standard.[49] X-ray scattering intensities were recorded using PILATUS 2M detector (Dectris, Baden-Daettwil, Switzerland). The temperature was controlled at 25.0 ± 0.1°C; 15 images were collected for each sample; and each image was recorded during a 2-s exposure time. A circular one-dimensional average of the image was computed by the program Nika.[50] Additional SAXS measurement parameters[51] are summarized in Table S1. Data processing procedure was described in SUPPLEMENTARY METHOD.

The scattering data collected using the BL-10C beamline are deposited in the Small Angle Scattering Biological Data Bank (SASBDB),[52] under the following accession codes: SASDDG2 and SASDDH2.

SVD

Forty-two crystal structures of the human IgG Fc region (Table S2)[17,18,30,53-75] were selected based on the information deposited in the UniProt Knowledgebase (accession number: P01857) and were obtained from PDB. All crystal structures of human IgG1 Fc region available in the PDB that satisfied the following criteria were selected: (a) the three-dimensional coordinates of the Cα atoms of each amino acid residue from S239 to S442 were completely assigned; (b) two heavy chains must be present in the crystal structure; (c) the structure represented a wild-type sequence without any amino acid substitutions. The dataset for SVD calculation was prepared as follows. First, because of the existence of unassigned amino acid residues in the N-terminal hinge and C-terminal regions of several crystal structures, the three-dimensional coordinates of Cα atoms of each amino acid residue between S239–S442 without the N-linked glycan were extracted from the original crystal structure of each Fc region. Next, the three-dimensional coordinates of Cα atoms in the CH3 domain (G341–S442) were aligned using the Pymol software. Then, the deviation of the three-dimensional coordinates of each Cα atom in the 42 crystal structures was calculated by subtracting the averaged three-dimensional coordinates of each Cα atom. Finally, SVD was performed on the deviation dataset using LAPACK package in IGOR Pro (Wavemetrics, Portland, OR, USA).[32,47,76,77]

Homology modeling and determination of theoretical SAXS profiles

The three-dimensional coordinates of amino acid residues in the N-terminal hinge region (gFc, T225–P238; aFc, the N-terminal alanine and D221–P238) and the C-terminal region (gFc, S441–G447; aFc, S441–K448) of the Fc region were predicted using the MODELLER program.[78,79] The three-dimensional coordinates of all atoms between S239–S442 from the 42 Fc crystal structures were used as fixed template structures for calculating the model structure of aFc and gFc (Table S2). In the fixed template structure for aFc, the N-linked glycan was removed if it existed within the original coordinates. For the template structure of gFc, the three-dimensional coordinates of the N-linked glycan were reconstructed because they had not been completely assigned in some of the original gFc crystal structures. In this study, we used the rigid N-linked glycan coordinates from a specific gFc crystal structure. Investigations using NMR revealed significant mobility of the N-linked glycan, such as α1-3 and α1-6Man branches, in solution.[80,81] Meanwhile, the fitting analysis of the measured SAXS profile using the theoretical SAXS profiles from the pool of different modeled glycan structures revealed that the glycan filled the internal space between CH2 domains.[26,35] Based on these reports, we expected that the glycan would fill the internal space of the CH2 domains in solution, and potential dynamics of glycans observed by NMR experiments would occur in the limited internal space. Thus, the rigid glycan coordinates used in this study was a valid option as a glycan model. In the template structure of gFc, the pair of G0F and G1F glycans (G0F/G1F) was selected as the N-linked glycan template, with reference to the three-dimensional coordinates of the N-linked glycan of a specific gFc crystal structure (PDB: 4BYH[58]), by aligning the CH2 domain (S239–K340) of 4BYH and that of each template structure. One hundred intact model structures were predicted for each of the 84 template structures of gFc and aFc. Two inter-chain disulfide bridges at the hinge region (C226 and C229) were introduced by using a manual restraint. Some of the intact model structures calculated based on the template structures for 1T89, 1H3V, 1H3Y, 3DO3, 3S7G, 4BYH, 4WI2, and 5U52, returned an unacceptable structure, in which the N-terminal hinge region penetrated the DE loop of the CH2 domain. For 1T89, 3DO3, and 5U52, unacceptable structures were not computed frequently. The unacceptable structures were manually replaced with the accepted structures with the non-penetrated hinge region. For 3S7G, 4BYH, and 4WI2, the unacceptable structure accounted for the majority of output. The population of the unacceptable structures was reduced by adding one or more N-terminal amino acid residues to the fixed template structures; the unacceptable structures were then replaced with accepted ones. For 1H3V and 1H3Y, the unacceptable structure also accounted for the majority of structures, but the population of the unacceptable structures could not be reduced using the described procedure. Therefore, the intact model structures for 1H3V and 1H3Y were excluded from further analysis. Consequently, 40 crystal structures and 80 template structures were included in analysis. SAXS profiles of each model structure and the fitting analysis in the q range of 0.015 to 0.250 Å−1 was performed using the Crysol software.[23] The fitting results were evaluated based on the reduced chi-square value and the P-value using the pair-wise correlation map (CorMap) analysis in the DATCMP program within the ATSAS software.[23,36,82]