Modulation of immunoglobulin G oligomerization by variable domain glycans: A mechanism to regulate complement activation

Expression of IgG carrying VDGs

To investigate the effect of VDGs on complement activation, we produced three monoclonal antibodies (mAbs), all of the IgG1 isotype, from ACPA-BCR sequences derived from patients with RA (30) carrying natural N-linked glycosylation sites in their variable domains (VDG+). The N-glycosylation sites are located either in the frame-work regions (FR1 and FR3) or in complementary-determining region 1 (CDR1; Fig. 1A) (13). To generate IgG1 counterparts lacking VDGs (VDG−) (Fig. 1B), the amino acids encoding for the N-linked glycosylation motifs were mutated to the corresponding germline sequences (Table S1) as previously described (13). The integrity of the VDG+ and VDG− mAbs was verified by gel electrophoresis (Fig. 1C). The shift in the apparent molecular weight between both variants indicated the presence of glycans (∼2,500 to 2,600 Da per disialylated, diantennary N-glycan with bisecting GlcNAc and core fucose) and was in line with an increasing number of glycosylation sites expressed by the three different mAbs (7E4 < 3F3 < 2G9). As depicted in Fig. 1C, all glycosylation sites in the heavy chain (HC) and LC of the mAbs 7E4 and 3F3 were occupied by a glycan. Similarly, the majority of 2G9 VDG+ mAbs showed a size shift, although a fraction had the same apparent molecular weight as the VDG-variant, suggesting that the glycosylation sites of this mAb were not fully occupied. To generate mAbs with similar VDG compositions as observed for ACPA IgG isolated from patients (14), we glycoengineered the mAbs using glycosyltransferase-transfected producer cell lines (see Materials and methods (31)). To exclude possible differences in Fc glycan composition introduced by glycoengineering, we similarly produced the VDG− mAbs. Subsequently, we assessed Fc glycosylation using liquid chromatography coupled with mass spectrometry (LC–MS). The observed Fc glycans showed high core fucosylation (93 to 97%), >61% galactosylation, ≥25% sialylation, and ≥45% bisection. The Fc glycan composition was similar between the VDG+ and VDG− variant of each set of mAbs (7E4, 3F3, and 2G9) (Fig. 1D). To confirm the effect of Fc glycan changes on complement activation reported in the literature (1), we produced glycoengineered and nonengineered 7E4 mAb variants that expressed either high or low levels of Fc galactosylation, sialylation, and bisection (Figs. 1E and S1A).

To characterize the glycan composition of the VDGs, we determined the total released N-glycan profile using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS). The results obtained showed low and high levels of galactosylation, sialylation, and bisection for the nonengineered and glycoengineered mAb variants, respectively (Fig. 1F). Importantly, the glycoengineered VDG+ mAb variants showed a dominant glycan peak at mass-to-charge (m/z) ratios ∼2,651 (Figs. 1F and S1B) corresponding to the most prominent VDG identified on secreted ACPA IgG from patients with RA (14).

Thus, these results indicate that ACPA IgG1 mAbs produced in glycosyltransferase-transfected producer cell lines harbor a similar VDG profile as present on ACPA obtained from patients with RA.

Glycosylation of the IgG variable domain inhibits classical complement activation

To delineate the effect of IgG VDGs on the activation of complement, we determined the ability of the generated VDG+ and VDG− mAbs to activate the classical complement pathway using an enzyme-linked immunosorbent assay (ELISA) (32). Firstly, similar binding of the VDG+ and VDG− antibody variants to a citrullinated antigen was controlled by ELISA. For mAb 7E4, a slight difference in antigen binding was detected (22). To account for this, we adjusted the antigen-coating concentration between the VDG+ and VDG− variants (Materials and methods) to achieve comparable antigen binding (Fig. 2A), thereby eliminating potential confounding effects on complement activation. No significant effect of VDGs on antigen binding was observed for mAb pairs 3F3 and 2G9 (Fig. 2A).

Next, we analyzed CP activation by the addition of exogenous complement to the antigen-bound mAbs and assessed C1q binding and deposition of C4 and C3. Our data showed increased CP activation for the 7E4 VDG− mAb with high levels of glycan galactosylation, sialylation, and bisection (Fig. 2B and C), confirming data from previous studies investigating the influence of Fc glycan compositions on complement activation (1, 33). In the same assays, we included VDG+ mAbs, and our data showed a profound reduction of CP activation in the presence of these additional glycans on the variable domains of the antibodies (Fig. 2B and C). This effect was 4× larger than the effect of Fc glycan composition changes (Fig. 2C). Similar results for C1q binding and C4 and C3 deposition were observed for all three VDG+ vs. VDG− mAb pairs (Figs. 2D and E and S2A) that express in total either four, six, or eight N-linked VDGs. No complement activation was detected in the absence of normal human serum (NHS) or in the presence of ethylenediaminetetraacetic acid (EDTA) that leads to a loss of C1 activity (Fig. 2D).

Next, we analyzed in more detail whether the effect on CP activation depended on the abundance or location of N-linked VDGs. Therefore, we generated four mAb variants of 7E4 and 3F3 including mAbs with VDGs (VDG+), with VDGs only expressed in the HC (HC-VDG+), with VDGs only expressed in the LC (LC-VDG+), and without VDGs (VDG−) (Fig. 3A). As shown in Fig. 3B, the apparent molecular weight of the antibodies produced corresponded to the number and location of N-linked glycosylation sites expressed by the respective Ab chains. These glycoengineered mAb variants had similar Fc glycan profiles as assessed by LC–MS, with the exception of the 7E4 HC-VDG+ mAb, which had slightly higher Fc sialylation (Fig. S1C). When assessing the variants in complement activation assays, we observed for both ACPA a significant stepwise decrease in C1q binding with increasing numbers of VDGs (Fig. 3C–E) despite identical antigen binding (Fig. S2B). This effect was less pronounced for C3 and C4 deposition; however, for both ACPAs the mAbs expressing the highest number of VDG consistently showed the lowest CP activation followed by mAbs expressing a lower number of VDGs. The highest C3 and C4 deposition was detected for the VDG− mAbs (Figs. 3C–E and S2C and D). In the case of 7E4, the mAb variant expressing the glycan in the LC (LC-VDG+) was associated with a stronger inhibitory effect, in particular C1q binding, compared with the variant expressing the VDG in the antibody HC (HC-VDG+) (Figs. 3C and D and S2C and D). The mAb 3F3 expressing two VDGs in the antibody HC and one in the LC showed lower binding of C1q when expressed as a HC-VDG+ variant (Fig. 3C and E).

These data indicate that the degree of N-linked glycosylation on the variable domain of IgG correlates with its ability to reduce initiation of the classical complement pathway, possibly by interfering with binding to C1q and subsequent deposition of C4 and C3.

Effects of the glycan position in relation to the C1q-binding platform

To elucidate the underlying mechanism of the inhibitory effect of IgG VDGs on C1q binding and subsequent CP activation, we hypothesized that the glycan position in relation to the C1q-binding interface is important. To investigate this further, we produced four “bispecific” antibody (bsAb) variants of the ACPA clone 3F3 expressing VDGs on either the antigen-binding or “innocuous” Fab arm (Fig. 4A). The “innocuous” Fab arm does not bind the antigen and is probably positioned differently to the C1q-binding platform (34) as it will display a higher degree of freedom compared with the Fab bound to the antigen. Based on the crystal structure of 3F3 cocrystallized with a citrullinated antigen (Protein Data Bank [PDB]: 6YXK (22)), we identified that the amino acid threonine at position 100 (T100) in the HC is involved in citrullinated antigen binding. We subsequently produced 3F3 VDG+ and VDG− mAbs with a threonine to valine point mutation (T100V) (Fig. 4A) to obtain 3F3 variants not able to bind their antigen (Fig. S3A and C). As depicted in Figs. 4C and S3D, these mAbs that show almost no detectable binding to citrullinated antigens cannot trigger CP activation. To generate monovalent bsAb molecules with one Fab arm recognizing the antigen, we introduced additional mutations in the Fc region allowing a controlled Fab arm exchange (35, 36). More specifically, we expressed the antigen-binding VDG+/VDG− 3F3 mAbs and the VDG+/VDG− (T100V) mutants as parental IgG including two matching point mutations, K409R and F405L, respectively, at the CH3:CH3 interface. Controlled reduction and reoxidation of the interchain disulfide bonds allowed the mutation-based recombination of HC–LC pairs resulting in four different monovalent bsAb variants (VDG+/VDG+ [T100V], VDG−/VDG+ [T100V], VDG+/VDG− [T100V], and VDG−/VDG− [T100V]) (Fig. 4A). The bsAb variants carrying only one glycosylated Fab arm, (VDG−/VDG+ [T100V] and VDG+/VDG− [T100V]), both correspond to a molecular weight between the fully glycosylated (VDG+/VDG+ [T100V]) and nonglycosylated (VDG−/VDG− [T100V]) variants (Fig. 4D, left panel). A similar finding was made when the antibodies were analyzed under reducing conditions, indicating successful Fab arm exchange of the parental molecules (Fig. 4D, right panel). Using these monovalent bsAbs, we addressed the question whether C1q binding is more restricted when VDGs are located on the “innocuous” Fab arm directed towards the C1q-binding platform. Despite similar antigen binding (Fig. S3B and E), we observed a decreased C1q deposition on the VDG+/VDG+ (T100V) bsAb carrying VDGs on both Fab arms compared with the level of C1q binding facilitated by the bsAb expressing no VDGs (VDG−/VDG− [T100V]) (Fig. 4E). Both bsAbs carrying VDGs on only one Fab arm showed an identical, intermediate C1q deposition, irrespective of the location of the VDGs (Figs. 4E and F and S3F). Thus, no additional inhibitory effect was detected when the glycans were presented on the Fab arm pointing towards the C1q-binding platform or vice versa (Fig. 4F).

Taken together, these data indicate that the reduced complement activation due to the presence of VDGs is independent of the position of the glycans with respect to the C1q-binding platform.

VDGs affect hexamerization of IgG for efficient complement-dependent cytotoxicity

To initiate efficient CP activation, monomeric IgG needs to hexamerize to form an optimal binding platform for the hexavalent C1q molecules (27, 28, 34, 37). To further investigate the mechanism by which VDGs affect CP activation, we therefore determined whether VDGs influence the propensity of IgG to form hexamers. To this end, we generated RGY triple mutants (E345R, E430G, and S440Y) of the VDG+ and VDG− ACPA IgG 3F3. These mutations enhance Fc/Fc interactions and thereby promote hexamerization in solution, allowing the assessment of VDG effects on oligomerization and C1q binding in an antigen-independent manner (34). The presence of RGY monomers and hexamers in a concentration-dependent manner was demonstrated by size exclusion chromatography (SEC) (Fig. 5). The size shift between the VDG+ and VDG− IgG RGY reflects the presence or absence of VDGs. As depicted in Fig. 5A, hexameric (IgG)⁶ species (eluting at 12 to 13 mL) were predominantly present in both VDG+ and VDG− IgG RGY mutants at a concentration of 0.1 mg/mL and higher. Interestingly, reduced hexamer formation, as indicated by increased monomer peaks eluting at 16 to 17 mL, was observed for the VDG+ IgG RGY mutant at all concentrations analyzed, indicating that the presence of VDGs reduces hexamer formation (Fig. 5A and B). We subsequently investigated the potential of the IgG RGY mutants to activate the CP, as determined by C1q binding and C3 deposition. Despite a similar IgG deposition in ELISA (Fig. 5D), we observed higher C1q binding and C3 deposition induced by the IgG RGY mutants compared with their Fc-nonmutated monomeric IgG counterparts (Figs. 5C and D and S4A). These observations were in agreement with the enhanced hexamerization of IgG RGY mutants as obligate hexamers can directly interact with C1q. Interestingly, when comparing VDG+ with VDG− IgG RGY mutants, we observed significantly reduced binding of C1q and deposition of C3 mediated by the variable domain glycosylated variant (Fig. 5C and E). This difference in CP activation concurs with the inhibitory effect of VDGs on the propensity of IgG to form hexamers. An effect of Fc glycans could be excluded as both RGY mutants showed similar Fc glycan composition (Fig. S4B).

To determine whether the effect of IgG VDGs on CP activation is confined to autoantibodies from patients with RA (ACPA IgG) or also holds true for other specificities, we generated VDG+ and VDG− IgG RGY mutants of an antidrug antibody. As previously described, antidrug antibodies also exhibit increased glycosylation of the variable domain, which is why we selected antiadalimumab (anti-Ada) clone 2.10 (38), which naturally harbors an N-glycosylation site in the HC and LC (Table S1). A size shift between the VDG+ and VDG− IgG RGY mutant confirmed the expression of VDGs on the 2.10 mAb (Fig. S5A). Similar to ACPA IgG, anti-Ada IgG hexamers were preferentially formed at higher concentrations, whereas the equilibrium shifted towards the monomeric anti-Ada IgG at lower concentrations (Fig. S5A and B). VDGs also affected the ability of IgG RGY mutants of anti-Ada clone 2.10 to form hexamers, as we observed reduced hexamer formation, albeit only significant at lower concentrations (Fig. S5B). In complement activation assays, more C1q binding and C3 deposition were observed for the 2.10 VDG− IgG RGY mutant compared with the VDG+ IgG RGY variant, despite similar IgG deposition in ELISA (Figs. S5C and D). To further validate that VDGs affect IgG hexamer formation and thereby subsequent C1q binding, we coated the ELISA plate with purified C1q. We assessed C1q binding of the VDG+ and VDG− IgG RGY mutants (ACPA 3F3 and anti-Ada 2.10) in relation to binding of monomeric Fc-nonmutated IgG (ACPA 3F3). No direct C1q binding of monomeric 3F3 was observed (Fig. 5F), consistent with the notion that IgG only interacts with C1q when oligomerized. In contrast, the IgG RGY mutants, which readily formed hexamers in solution, showed dose-dependent binding to purified C1q (Figs. 5F and S5E). Importantly, a significantly higher interaction with C1q was observed for the VDG− IgG RGY mutant compared with the VDG+ IgG RGY mutant for both ACPA and anti-Ada IgG (Fig. 5G).

Overall, these results indicate that IgG VDGs impair hexamer formation, which possibly explains the reduced binding to the hexavalent C1q molecule.

Antibody variable domain glycosylation reduces complement-mediated membrane lysis

We investigated the effect of VDGs on CDC using native lipid bilayers as cell membrane mimetics. To this end, we synthesized liposomes displaying citrullinated antigens, enabling VDG+ and VDG− ACPA IgG to bind and move freely to form oligomers on the surface. Complement activation was then assessed by measuring sulforhodamine B (SRB) release from the liposomes, which becomes fluorescent upon release due to MAC pore formation (39). No liposome lysis was detected in the absence of NHS or when using liposomes coated with the arginine control antigen (Figs. 6A and S6). In contrast, in the presence of 10% NHS, both VDG+ and VDG− ACPA IgG activated the CP and induced MAC pore formation and liposome lysis, resulting in an increase in fluorescent signal (Fig. 6A). At mAb concentrations of 10 nM and higher, the VDG− 3F3 induced significantly greater complement-mediated membrane lysis compared with its VDG+ counterpart (Fig. 6B).

In summary, these results link the presence of VDGs to a reduced ability of IgG to induce complement-mediated membrane lysis. Thus, these results provide further evidence that VDGs impair IgG's propensity to form ordered oligomers, even in the membrane-bound state, thereby reducing efficient C1q recruitment and hence leading to diminished CP activation.

Visualization of reduced ability of VDG IgG to form ordered hexameric structures and recruit complement

To gain insights into the impact of VDGs on the structure of IgG–C1 complexes and thus to visualize any structural differences between the VDG+ and the VDG− 3F3 IgG–C1 complexes on the surface of liposomes, we used cryoelectron tomography (cryoET) (40). Either VDG+ or VDG− 3F3 antibodies were bound to the surface of citrullinated antigen-coated liposomes and NHS was added as a source of complement components, as previously described (27). Using cryoET, we could observe C1 complexes on top of both antigen-bound 3F3 VDG+ and 3F3 VDG− IgG (Fig. 7A and B). This yielded n = 249 and n = 486 C1 complexes on VDG+ and VDG− 3F3 IgG, respectively, from the same number of tomograms (Fig. 7C), corresponding to significantly fewer C1 complexes on the antibodies containing VDGs. Next, we performed subtomogram averaging on the C1 complexes bound to VDG+ and VDG− 3F3 antibodies, which clearly showed density corresponding to a C1 complex on top of an antibody platform (Fig. 7D and E). However, slices through the subtomogram averages at the layer of the Fc platform only showed hexameric density for the VDG− condition (Fig. 7E), while the VDG+ antibodies appeared to adopt a much less-ordered structure (Fig. 7D). This less-ordered IgG oligomerization was also apparent in the 3D volumes (Fig. 7F and G). In spite of low resolution (45 Å), the resulting map of C1 bound to the VDG− 3F3 IgG revealed a similar structure to previously published C1–IgG1 maps (28) (Fig. 7G), with six clearly resolved C1q head domains bound to the Fc platform, as well as density corresponding to the C1r2s2 protease platform and the C1q stalk domain on top. In contrast, the map of C1 bound to the VDG+ antibodies was of lower resolution (57 Å), and the map indicated a much more flexible and less-ordered structure (Fig. 7F).

Taken together, antibodies show less C1 complex recruitment in the presence of glycans linked to the variable domain, likely caused by a hindering effect of VDGs during hexamer formation after antigen binding. Thus, VDGs represent another way in which IgG can be functionally and structurally diversified.

Generation and glycoengineering of recombinant mAbs

Human mAbs were generated based on full-length IgG BCR sequences obtained from antigen (citrulline or adalimumab)-specific B cells as previously descried (38, 46, 47). To generate mAbs without VDGs (VDG−), the N-linked glycosylation motifs (N–X–S/T, X ≠ P) in the variable domains were replaced by the germline amino acid sequence (based on the ImMunoGeneTics information system database (48)) at the respective position (Table S1). VDG+, including N-linked glycosylation sites, and VDG−, including only the conserved Fc N-linked glycosylation site, sequences were codon-optimized, and each VH and VL was ordered from GeneArt (Life Technologies) flanked by the IGHV1-18*01 leader sequence, the Kozak sequence, and 5′-BamHI and 3′-XhoI restriction sites. The constructs were ligated into the pcDNA3.1 (+) expression vector (Invitrogen) carrying the IGHG1 or the IGLC3/IGKC constant regions (UniProt), respectively, and the flanking 3′-XhoI site. Sequences were verified by Sanger sequencing performed on Applied Biosystems 96-capillary (ABI3730) systems (LGTC Facility; Macrogen). The recombinant mAbs were produced in Freestyle 293-F cells (Gibco) as previously stated (30). To modulate the antibodies’ glycan composition, recombinant mAbs were glycoengineered by adding 5 mM D-galactose (Sigma; G0750-5G) and by cotransfecting 1% β-1,4-N-acetylglucosaminyltransferase III (GnTIII), 2.5% α-2,6-sialyltransferase 1 (ST6galT), and 1% β-1,4-galactosyltransferase 1 (B4GalT1). The supernatants were harvested 5 to 6 days posttransfection. IgG1 antibodies were purified using a 1-mL HiTrap Protein G HP affinity column (GE Healthcare) followed by a direct buffer exchange (53 mL HiPrep 26/10 Desalting column; GE Healthcare).

In total, three glycoengineered VDG+ and VDG− ACPA IgG1 (7E4, 3F3, and 2G9) mAb pairs were produced. Additionally, both VDG+ and VDG− 7E4 IgG1 mAbs were produced without glycoengineering to yield low levels of galactosylation, sialylation, and bisection. Next to the VDG+ and VDG− mAb variants, we generated the 7E4 and 3F3 mAbs with only the HC (HC-VDG+) or the LC (LC-VDG+) N-linked glycans, respectively. To produce 3F3 (ACPA) and 2.10 (antiadalimumab) VDG+ and VDG− IgG1 RGY mutants, three point mutations (E345R, E430G and S440Y) were introduced into the IGHG1*01 constant domain (34). Mutations were introduced using site-directed mutagenesis PCR (NEB) and verified by Sanger sequencing on Applied Biosystems 96-capillary (ABI3730) systems (LGTC facility; Macrogen). IgG1 RGY mutants were recombinantly expressed with glycoengineering.

Generation of bsAbs

Human bsAbs of the 3F3 clone, only able to bind with one arm to their cognate (citrullinated) antigen, were generated by introducing the T100V point mutations into the VH CDR3 sequence. The mutation abrogating antigen binding was identified based on 3F3 crystal structure data (PDB: 6YXK). In addition, two matching point mutations (K409R and F405L) were introduced into the CH3 interface of VDG+, VDG−, VDG+ (T100V), and VDG− (T100V), respectively. All point mutations were generated using the Q5 Site-Directed Mutagenesis Kit (NEB; E0554S). The parental mAbs (VDG+ [K409R], VDG− [K409R], VDG+ [T100V, F405L], and VDG− [T100V, F405L]) were produced in Freestyle 293-F cells (Gibco) and purified as described above. BsAb molecules were generated by controlled Fab arm exchange as previously described (35). In brief, 50 µg of the parental 3F3 mAbs was mixed in equimolar amounts and incubated with 750 mM cysteamine hydrochloride (Sigma-Aldrich) for 5 h at 31° C for reduction. Reoxidation and buffer exchange were performed using Ultra Centrifugal filters (50,000 molecular weight cut-off; Merck Millipore) and phosphate buffered saline (PBS). This method allows mutation-based recombination of HC-LC pairs resulting in four 3F3 bsAb molecules (VDG+/VDG+ [T100V], VDG−/VDG+ [T100V], VDG+/VDG− [T100V], and VDG−/VDG− [T100V]) with one “innocuous” Fab arm and one Fab arm that can bind citrullinated antigens.

Integrity and size characterization of mAbs

The integrity of the mAbs and size shift due to the expression of VDGs was analyzed by sodium dodecyl sulfate (SDS)–polyacrylamide gel electrophoresis (PAGE). In brief, 1.5 µg of mAb was diluted in 4× Laemmli buffer (Bio-Rad) with (reduced) or without (nonreduced) 2% β-mercaptoethanol (Sigma-Aldrich) and incubated for 5 min at 95 °C. Samples and PageRuler Plus Prestained Protein Ladder (Thermo Fisher Scientific) were loaded on 4 to 15% SDS–PAGE gels (Bio-Rad). For protein detection, gels were stained with Coomassie brilliant blue G-250 Dye (Thermo Fisher Scientific). To analyze the presence of IgG RGY mutant monomers and hexamers, SEC was performed using a Superose6Increase, 10/300 GL column (GE Healthcare). The recombinant proteins were monitored by UV absorption at 280 nm. The percentage of oligomerization was calculated based on the area under the curve (AUC) of the hexamer peak eluting at 12 to 13 mL.

Fc peptide and VDG analysis

MALDI-TOF MS analysis of released N-glycans was performed as previously described (49). In brief, the IgG samples (10 µg) were reduced in 2% SDS and incubated with 0.5 U N-glycosidase F, PNGase F (Roche Diagnostics, Germany), in 1:1 5×PBS/4% nonidet (N)P-40, overnight at 37 °C for N-glycan release. N-glycans were stabilized with ethylation reagent (0.5 M N-(3-Dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride and 0.5 M 1-hydroxybenzotriazole hydrate) for 1 h at 37 °C and purified in 85% acetonitrile (ACN) (Biosolve, Valkenswaard, The Netherlands) with 3-mm cotton hydrophilic interaction liquid chromatography (HILIC) 15-µL tips using washing steps of 3 × 15 µL 85% ACN + 1% trifluoroacetic acid and 85% ACN. Released and purified N-glycans were eluted from the HILIC tips using 10 µL MQ and 1 µL spotted on a MALDI target (MTP AnchorChip 800/384 TF; Bruker Daltonics) together with 1 µL 5 mg/mL super-3,4-dihydroxybenzoic acid in 50% ACN and 1 mM NaOH. Spots were dried at room temperature (RT) and analyzed on an UltrafleXtreme (Bruker Daltonics) operated under flexControl 3.3 (Build 108; Bruker Daltonics). A mass spectrum from m/z 1,000 to 3,000 was recorded, combining 10.000 shots in a random walk pattern at 1,000 Hz. The instrument was calibrated with a peptide calibration standard (Bruker Daltonics). MS data were analyzed using flexControl 3.3, and glycan peaks above S/N of 9 were included into the analysis.

LC–MS glycan analysis was performed as previously described (50). Briefly, for Fc peptide glycan analysis, mAbs were trypsinized using 200 ng sequencing grade modified trypsin (Promega) in 50 mM ammonium bicarbonate overnight at 37 °C. For total N-glycan analysis, glycans were released using 2 U PNGase F as described above, labeled with 2-aminobenzoic acid (Sigma-Aldrich) and 2-picoline borane (Sigma-Aldrich), and purified using 15 µL cotton (3 mm) HILIC tips. Trypsinized IgG or released and purified N-glycans were separated and analyzed on an Ultimate 3000 UHPLC system (Thermo Fisher Scientific, Breda, The Netherlands) coupled to a MaXis Impact HD quadrupole–TOF mass spectrometer (MaXis; Bruker Daltonics) equipped with a NanoBooster source (Bruker Daltonics) as previously described (50–52). Mass spectra were acquired within a mass range of m/z 550 to 1,800. LC–MS data were first examined manually using DataAnalysis (Bruker Daltonics). Data processing, including peak integration, was performed using LaCyTools v1.1.0. Glycans with a signal/noise ratio above 9, a mass accuracy of ±20, and an isotopic peak quality of 0.2 were included into the analysis (53).

Complement activation ELISA

To analyze the complement-activating potential of mAbs with different numbers of VDGs, complement activation ELISA was performed in a similar way as previously described (32). In short, mAbs were diluted in PBT (PBS/0.05% Tween/1% bovine serum albumin [BSA]) and added to streptavidin plates (Microcoat; 65001) precoated with 1 µg/mL biotinylated cyclic citrullinated peptide 2 (CCP2; patent protected sequence; EP2071335) overnight at 4 °C (30). The mAb concentrations used are mentioned in the respective figures. To obtain a similar antigen binding, CCP2 concentrations were adjusted for the 7E4 VDG+ (1.9 µg/mL) and VDG− (0.95 µg/mL) mAbs. IgG RGY mutants that hexamerize in solution without the need of oligomerization through binding to their cognate antigen were coated to Nunc Maxisorp plates (Thermo Fisher Scientific) in coating buffer (100 mM Na₂CO₃/NaHCO₃ [pH 9.6]) overnight at 4 °C. IgG deposition on the ELISA plate was identified using a horseradish peroxidase (HRP)-conjugated rabbit antihuman IgG secondary detection antibody (DAKO; P0214; 1:5,000). For complement activation, 0%, 0.5% (3F3 and 2G9), 1% (7E4), or 10% complement-active NHS diluted either in GVB++ (gelatin veronal buffer with calcium and magnesium; veronal buffered saline [VBS] containing 0.5 mM MgCl₂, 2 mM CaCl₂, 0.05% Tween, and 0.1% gelatin [pH 7.5]), in GVB++ in the presence of EDTA (loss of C1 activity) or Mg–ethylene glycol tetraacetic acid (EGTA; VBS containing 10 mM EGTA, 5 mM MgCl₂, 0.05% Tween, and 0.1% gelatin [pH 7.5]), buffer was added for 1 h at 37 °C. NHS was obtained from the in-house LUMC LuVDS (Vrijwillige Donoren Service) and consisted of a pooled sample from six healthy donors (male and female, varying ages). Complement deposition on the IgG mAbs was determined using rabbit anti-C1q (DAKO; A0136; 1:1,000), goat anti-C4 (QUIDEL; A305; 1:1,000), or rabbit anti-C3 (DAKO; A0062; 1:1,000) diluted in PBT for 1 h at 37 °C. Binding was detected by HRP-labelled goat antirabbit (DAKO; P0448; 1:3,000) or rabbit antigoat (DAKO; P0449; 1:3,000) detection antibodies, respectively, diluted in PBT for 1 h at 37 °C.

To identify mAb binding to C1q, in the absence of cognate antigen, 10 µg/mL purified C1q was coated on Nunc Maxisorp plates (Thermo Fisher Scientific) in coating buffer for 1 h at 37 °C. After blocking with PBS/1% BSA for 1 h at 37 °C, mAbs were added in PBT and incubated for 1 h at 37 °C. IgG binding to C1q was detected using HRP-labelled F(ab')₂ fragment rabbit antihuman IgG (Jackson ImmunoResearch; 309-036-003), diluted 1:5,000 in PBT, and incubated for 1 h at 37 °C. All ELISA plates were developed using the substrate azinobis-ethyl-benzothiazoline sulfonic acid (ABTS), and absorbance was read at 415 nm.

Complement-mediated liposome permeabilization assay

Liposomes presenting the cognate antigen (cyclic citrullinated peptide 4 [CCP4] (30)) or the control antigen (CargP4) were generated using dimyristoylphosphatidylcholine (Avanti Polar Lipids), dimyristoylphosphatidylglycerol (Avanti Polar Lipids), cholesterol (Avanti Polar Lipids), and CCP4/CArgP4-cholesterol in a ratio of 43:5:50:2 mol%, and classical complement activation after antibody binding was measured via MAC pore formation as previously described (29). Briefly, thin lipid films were produced under constant nitrogen flow. After at least 12 h, lipid films were rehydrated at 60 °C for 1 h with a self-quenching concentration of SRB (20 mM; Sigma-Aldrich; S1402) in PBS to a final lipid concentration of 0.8 mg/mL. Liposomes were generated by sonification for 10 min at 60 °C in a water bath. Purification of liposomes from free SRB was performed through SEC using a prepacked NAP-25 column (GE Healthcare), followed by an additional centrifugation step. Liposomes were centrifuged for 15 min at 20,000 g at 4 °C using a Sigma 2-16k centrifuge (Sigma Laborzentrifugen GmbH, Harz, Germany). Supernatant was removed, and pellets were dissolved in fresh PBS. Purified antigen-coated liposomes were mixed with ice-cold NHS (10% final concentration), and fluorescent intensity was measured for 100 s at 21 °C. Afterwards, VDG+ and VDG− 3F3 ACPA IgGs were added at a concentration range from 300 to 0.412 nM (final concentration). Complement activation was detected by MAC pore formation and resulting release of the fluorescent dye SRB. Fluorescence intensity was measured with an excitation wavelength of 565 nm and an emission wavelength of 585 nm using a CLARIOstar microplate reader (BMG Labtech) at 21 °C for 20 min.

Sample preparation for cryoelectron microscopy

Cryoelectron microscopy samples were prepared as previously described (27). Briefly, VDG− and VDG+ 3F3 IgGs (1,500 nM final concentration) were added to 2 mol% CCP4 displaying liposomes and incubated at RT for 20 min before cooling to 4 °C for 20 min. Afterward, ice-cold PBS or NHS (1.5% v/v final concentration) was added and samples were incubated at 4 °C for an additional 20 min before ice-cold BSA-coated 5-nm gold colloids were added. Lacey carbon-supported copper grids, 200 mesh (Electron Microscopy Sciences, LC200-CU-100), were plasma cleaned directly before they were loaded into a Leica EMGP (Leica Microsystems, Wetzlar, Germany). Samples (6 µL) were applied on the grids and incubated for 60 s at 4 °C at 85% humidity before 1.5-s blotting from the back and vitrifying in liquid ethane. Grids were clipped and stored in liquid nitrogen until usage.

CryoET data collection and reconstruction

Tilt series acquisition was done on either a Titan Krios (Thermo Fisher Scientific, Waltham, MA, USA; at Netherlands Centre for Electron Nanoscopy) operating at 300 kV with a K3 direct electron detector and a Bioquantum energy filter (Gatan) in counting mode or on a Talos Arctica (Thermo Fisher Scientific, Waltham, MA, USA; at EM facility in the LUMC) operating at 200 kV with a K3 direct electron detector and a Bioquantum energy filter (Gatan) in counting mode. The energy filter was set to a slit width of 20 eV for both microscopes. On the Titan Krios, tilt series were acquired using the Serial EM software (version 4.1.0beta) (54) at 42,000× magnification with a pixel size of 2.16 Å using a dose-symmetric scheme from 0 to ±57°, in 3° increments in a dose-symmetric tilt scheme (55). On the Talos Arctica, tilt series were acquired using the Tomography 5 (version 5.5.0; Thermo Fisher Scientific, Waltham, MA, USA) at 49,000× magnification with a pixel size of 1.74 Å using a dose-symmetric scheme from 0 to ±57°, in 3° increments in a dose-symmetric tilt scheme. Nominal defocus range was set between −2.5 and −6 µm, and tracking was performed before each tilt acquisition in both cases. A total dose of 60 e−/Ų was used for each tilt series. Raw frames were aligned using “alignframes” in IMOD (version 4.11) (56). Tomogram reconstruction including CTF estimation, correction, and gold beat erasing was done in IMOD after 2 × beginning of the tilt series using weighed backprojection.

Particle picking and subtomogram averaging

Tomograms were blinded by randomizing the names to remove any reference to the original dataset. All particles were then picked before names were reverted and the particles were extracted. Manual particles picking, extraction, and subtomogram averaging were performed using EMAN2 (version 2.91) (57). For picking and extraction, the command e2spt_boxer_old.py was used. For the Titan dataset, 651 VDG+ and 1,674 VDG− 3F3 IgG–C1 particles were picked, extracted, and normalized with inverted contrast using a box size of 256 pixels with a final pixel size of 4.34 Å/pixel. For the Talos dataset, 249 VDG+ and 486 VDG− 3F3 IgG–C1 particles were picked, extracted, and normalized with inverted contrast using a box size of 184 pixels with a final pixel size of 3.48 Å/pixel. For subtomogram averaging, only the dataset acquired on the Titan Krios system was used. An initial model for subtomogram averaging of the VDG− and VDG+ 3F3 IgG antibody during complement activation was generated using the command e2spt_binarytree.py. This model was filtered to 60 Å for further reconstruction. Subtomogram averaging of the 3F3 IgG–C1 complex was performed using e2spt_classaverage.py. All gold-standard procedures were followed, and particle sets were split into even and odd sets. Alignment of particles was achieved over 30 iterations with C1 symmetry and a loose spherical mask in center. For the VDG− 3F3 IgG–C1 and VDG+ 3F3 IgG–C1 complexes, resolutions of 45 and 57 Å were reached, respectively.

Statistical analyses

Statistical analyses were performed using the GraphPad Prism 8.0.1 software. Repeated measure (RM) one-way ANOVA including the Geisser–Greenhouse correction or paired t tests were performed for column analyses. Multiple (un)paired t tests were performed for grouped analyses. Specifications on independent or technical replicates are indicated in the respective figure legends. P-values <0.05 were considered significant: ns P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; or ****P ≤ 0.0001.