Recombinant IgG antibodies specific for biotin [bt; (34, 35)] and trinitrophenyl [TNP; (36)] were produced using FreeStyle™ HEK293-F cells (ThermoFisher), which were cultured in FreeStyle™ 293 Expression Medium (ThermoFisher). Using Polyethylenimine-MAX, pcDNA3.1 expression vectors encoding human κ, IgG1 and IgG4 were co-transfected under serum-free conditions into HEK293-F cells according to the manufacturer’s protocol (Invitrogen). In the case of the anti-TNP antibodies, an supplemental vector was used with a mutation of D86N in the light chain to introduce an additional glycosylation site (37). 0.9% w/v Peptone Primatone^® RL (Sigma-Aldrich) was added at least 4 h post transfection. The galactose-glycovariants were produced as described previously (19). To produce low galactose-glycovariants, 0.5 mM of the galactose analogue 2-deoxy-2-fluoro-d-galactose (2FG) (Carbosynth) was added 4 h post transfection. To produce high galactose-glycovariants, 5 mM D-galactose (Sigma-Aldrich) was added to the cell culture 1 h before transfection, and cells were co-transfected with 1% DNA (of total DNA transfected) encoding the β-1,4-galactosyltransferase 1 (B4GALT1).

High mannose-glycovariants were produced using HEK293-6E cells with a MGAT1 knock-out (KO), which lack N-acetylglucosaminyl transferase I (GlcNAc-T I) (38, 39). These cells were kindly provided by Prof. H. Clausen from the Copenhagen Center for Glycomics, University of Copenhagen, Denmark. The MGAT1-KO cells were cultured in FreeStyle™ F17 Expression Medium (ThermoFisher) supplemented with 4 mM L-glutamine (Sigma-Aldrich) and 0.1% w/v Kolliphor^® P 188 (Sigma-Aldrich), at 5% CO₂, 37°C while shaking at 125 rpm.

Culture supernatant was harvested five or six days after transfection and filtered using a 0.2 nm Puradisc™ syringe filter (Whatman, GE Healthcare). Antibodies were purified on a HiTrap^® protein G HP 1 mL column (GE Healthcare) and eluted with 0.1 M glycine pH 2.5 (Merck). The eluate was then dialyzed against 5 mM NaAc (pH 4.5). Antibody concentrations were determined by measuring A280 on a NanoDrop One spectrophotometer (ThermoFisher) and samples were stored at -20°C.

To be able to produce bispecific IgG1 antibodies, a mutation of K409R was introduced into the heavy chain of the IgG1 anti-bt clone, which allows for Fab arm exchange of IgG1 in a manner similar to IgG4. IgG1 K409R anti-bt was reacted with an in-house generated IgG1 K409R variant of adalimumab, and IgG4 anti-bt or anti-TNP with IgG4 natalizumab (TYSABRI^®), both in a ratio of 1:4 (anti-bt or anti-TNP to irrelevant clone). Fab arm exchange was induced by mild reduction with 10 and 0.5 mM L-glutathione (GSH; Sigma-Aldrich) for IgG1 and IgG4 respectively, in presence of 0.1% w/v Kolliphor^® P407 (BASF). Samples were incubated at 37°C for over 20 h, after which the reaction was stopped with 12 mM and 0.6 mM 2-iodoacetamide (IAM; Merck) for IgG1 and IgG4 respectively. IgG1 and IgG4 bispecific antibodies were made fresh for each experiment due to aggregation of the antibodies during storage. Bispecificity was confirmed for all clones.

The purified glycoengineered IgG1 and IgG4 were dried by vacuum centrifugation and subjected to tryptic cleavage followed by LC-MS as described previously (40, 41). The raw LC-MS spectra were converted to mzXML files and LaCyTools, an in-house developed software, was used for the alignment and targeted extraction of raw data. The analyte list for targeted extraction of the 1+, 2+, 3+ and 4+ charge states was based on manual annotation as well as on literature reports (40). The inclusion of an analyte for the final data analysis was based on quality criteria including signal-to-noise (higher than 9), isotopic pattern quality (less than 25% deviation from the theoretical isotopic pattern), and mass error (within a ±20 parts per million range) leading to a final analyte list ( Table S1 and S2 ). Relative intensity of each glycopeptide in the final analyte list was calculated by normalization the glycopetide value to the sum of the total area of all glycopeptides included. Normalized intensities were used to calculate fucosylation, bisection, galactosylation and sialylation levels ( Table S3 ).

To study the influence of glycosylation on the molecular weight of IgG1 and IgG4, a reduced SDS-PAGE was performed on a NuPAGE 4-12% Bis-Tris gel according to manufacturer’s protocol (Invitrogen). The gel was stained with InstantBlue^® Coomassie Protein Stain (Abcam) for 1 h and the bands were visualized using a ChemiDoc™ Universal Hood III (Bio-Rad).

To assess complement activation by IgG1 and IgG4 variants, C3b deposition was measured in ELISA as described previously (42). In short, human serum albumin (HSA; Albuman, Sanquin) was either biotinylated with different concentrations (30 to 240 µM) of biotin (EZ-Link™ Sulfo-NHS-LC-Biotin No-Weigh; ThermoFisher) in PBS or TNPlated with 1 mM of 2,4,6-trinitrobenzenesulfonic acid (TNBS, Sigma-Aldrich) in 0.2 M Na₂HPO₄ (Merck, F173578238). Nunc MaxiSorp™ flat-bottom plates (ThermoFisher) were coated with 5 µg/mL of biotinylated HSA (HSA-bt) or 10 µg/mL of TNPlated HSA (HSA-TNP) in PBS and incubated at 4°C overnight. IgG antibodies were serially diluted in PBS supplemented with 0.1% v/v Tween^®-20 and 0.2% w/v gelatin (Merck) and incubated shaking at room temperature for at least 1 h. After washing, C3b deposition was allowed for 1 hour at RT by addition of 2.5% normal human serum (NHS; pool of minimum healthy donors) in veronal buffer (3 mM barbital (Fagron), 1.8 mM sodium-barbital (Fagron), 0.146 M sodium chloride (Merck)) supplemented with 1 mM calcium chloride (CaCl₂, Merck), 0.5 mM magnesium chloride (MgCl₂, Merck) and 0.2% Tween-20 (VB+/+). For inhibition conditions, VB++ was supplemented with additional anti-C1q-85 (6.3 µg/mL; Sanquin (43, 44)), anti-MBL1 (0.2 µg/mL; Sanquin (45, 46)) and/or anti-complement factor D (a-FD; 0.13 µg/mL; Lampalizumab; Genentech (47, 48);) or inhouse anti- complement factor B-1 (a-FB; 24,3 µg/mL; Sanquin) in order to inhibit C1q, MBL and FD/FB respectively. All inhibitors were shown to be effective blockers of their respective route of activation (49).

Alternatively, instead of the addition of serum dilution, IgG binding to the coat was measured using mouse-anti-IgG-horseradish peroxidase (0,33 µg/mL; MH-16-1-HRP; Sanquin). Deposited C3b was detected with mouse anti-human C3-19-HRP (0.5-1 µg/mL; Sanquin (50)) and bound C1q was detected with mouse anti-human C1q-2-HRP (3 µg/mL; Sanquin (43)) and visualized using tetramethylbenzidine (TMB; Merck) mix. The reaction was stopped with 0.2 M sulfuric acid (Merck) and the optical density (OD) was measured at Δ450–540 nm using an ELISA plate reader (SynergyTM 2 Multi-Detection Microplate Reader, Biotek).

The CDC assay to assess terminal pathway activation by the anti-bt IgG antibodies was performed as described earlier, with minor alterations (42). To biotinylate red blood cells (RBCs), one part packed cells was reacted with seven parts of a biotinylation reagent at a final concentration of 2.5 mM Sulfo-NHS-LC-Biotin (ThermoFisher) in PBS. RBCs were incubated at 37°C for at least 1 h with serial dilutions of anti-bt IgG antibodies, 10% of NHS pool and inhibitors (indicated for each respective experiment) in veronal buffer supplemented with 0.05% w/v gelatin, 1 mM CaCl₂ and 0.5 mM MgCl₂. Conditions with complement inhibitors were included as described above, but with four times higher concentrations. A positive control (100% lysis) was generated by adding 0.5% w/v saponin (Merck/Millipore) to RBCs and used to normalize the data. After incubation, 40 µL of buffer was added, plates were centrifuged at 1800 rpm for 2 min and 100 µL supernatant was transferred to MaxiSorp™ plates. The OD was measured as ΔOD(412 nm) – OD(690 nm). The percentage of cell lysis for each sample was calculated as follows: % Lysis = OD_sample/mean OD_100% * 100.

Kruskal-Wallis test with Dunn’s multiple comparisons test was performed for statistical comparison of the IgG4 glycovariants.

To study complement activation by IgG4, C3b deposition and CDC were determined in a recombinant antibody system with anti-hapten antigens which allowed for control of antigen density. Using recombinant anti-biotin antibodies with biotinylated HSA, we found that IgG1 activated complement efficiently in all conditions tested, in contrast to IgG4 which could only activate complement at high antigen densities (HSA-bt 120 and 240 µM) and high antibody concentrations ( Figure 2A ). IgG4 required a roughly 100 times higher antibody concentration to reach the same OD as IgG1. Comparable results were obtained utilizing another antigen, TNP, where IgG4 was also only able to activate complement at high antigen densities and antibody concentrations ( Supplementary Figure 1A ). In addition, two monoclonal antibodies against protein antigens (rather than anti-hapten) were tested, anti-Betv1a (Birch pollen) and anti-idiotype Fabs (against adalimumab). No complement activation with either of these antibodies was observed, probably because sufficient densities of antibody binding were not reached ( Supplementary Figure 1B, C ).

Next, the ability of IgG4 antibodies to induce complement-dependent lysis of biotinylated RBCs was determined. In line with the results for the C3b deposition, IgG4 induced lysis at high antigen densities and antibody concentrations almost to the same extent as IgG1 antibodies ( Figure 2B ). However, IgG1 induced cell lysis at all conditions tested at much lower antibody concentrations. Taken together, this shows that IgG4 can activate complement to some extent, but only in conditions where both sufficient antigen and antibodies present.

It has been suggested that agalactosylated IgG4 autoantibodies can induce complement activation through the lectin pathway via MBL (33). To study this effect of galactosylation on complement activation by IgG4 in more detail, glycoengineered variants with relatively low and high levels of galactosylation were produced. These anti-biotin IgG1 and IgG4 antibodies were classified as ‘low galactose’ (LG; 10% galactose), ‘normal galactose’ (NG; 26-40%) and ‘high galactose’ (HG; 75-80%) based on their galactosylation levels determined by mass spectrometry ( Supplementary Figure 2A ). The antibodies were also analyzed by gel electrophoresis under reducing conditions and subtle shifts in size can be observed among the heavy chains due to varying levels of galactosylation ( Supplementary Figure 2B ). As a glycan control, cells deficient for N-acetylglucosaminyl transferase 1 were used to produce ‘high mannose’ (HM) variants (38, 39). An IgG1 antibody variant (referred to as IgG1 PG-LALA) with a combination of mutations that eliminate Fc-mediated effector functions, including complement was included as a “complement-dead” control (51).

The capacity of these glycovariants to activate complement was determined in C3b deposition ELISAs and CDC assays at high antigen densities (biotinylated HSA with 240 μM biotin and RBCs biotinylated at 2.5 mM respectively). At the level of C3b deposition, differences between the IgG4 glycovariants were observed, with the HG variant steadily outperforming the NG and LG variants ( Figure 3A ). The differences across glycoengineered IgG1 antibodies were less pronounced, but nevertheless consistent with previous results (23), as complement activation reached saturation at even low antibody concentrations. Both IgG1 and IgG4 HM variants induced the lowest levels of C3b deposition of all variants within each subclass. IgG1 PG-LALA was not able to mediate any detectable C3b deposition when tested in similar conditions as IgG4. No deposition was found with heat-inactivated serum and the binding of all glycovariants was found to be identical ( Supplementary Figure 3A, C ). In contrast, the differences in IgG4 galactosylation did not result in similarly clear differences at the level of cell lysis ( Figure 3B ). Furthermore, IgG1 PG-LALA, which is incapable of inducing complement activation, showed similar levels of cell lysis as the IgG4 variants. For both the IgG4 antibodies as well as the IgG1 PG-LALA variant, this effect appears to be predominantly complement-mediated, as heat-inactivated serum does not cause any cell lysis ( Figure 3C and Supplementary Figure 3C ). Red blood cells are sensitive to lysis as the formation of a single complement membrane attack complex can already cause cell lysis, especially at very high antibody concentrations and even when the antibody is considered ‘complement-dead’. Taken together, no enhanced complement activation for hypogalactosylated IgG4 was observed, on the contrary, high levels of Fc-galactosylation resulted in increased complement activation by IgG4, which is in line with observations using differently galactosylated IgG1 (20, 21, 23, 52).

We next investigate the possible role of the Fab glycans on complement activation as they are relatively abundant on IgG4 (25). Therefore, WT IgG4 antibodies were produced against TNP (containing only an Fc glycan), and variants with a mutation in the variable region (D86N) of the light chain that introduces an additional glycosylation site within the Fab domain. These antibodies were generated as ‘low’ (10%) galactose and ‘high’ (70%) galactose variants and had Fc glycan composition levels like the anti-biotin variant as described above ( Supplementary Figure 1C, D ). Again, both HG variants induced more C3b deposition than the LG variants ( Figure 3D ), suggesting this is Fc-glycan mediated. The introduction of the additional glycan in the Fab interfered with the ability of the antibody to activate complement, as both antibodies with the additional Fab glycan performed worse than the WT variants. Antigen binding of all variants appears to be similar, indicating that this is not the cause of the decreased ability to activate complement. ( Supplementary Figure 3A ).

Next, we investigated which pathway of the complement system was activated by these IgG4 glycovariants. C3b deposition and RBC lysis were performed in the presence of inhibitors of C1q, MBL, and FD, to determine the contribution of the classical pathway, lectin pathway and alternative pathway, respectively. Despite the suggestion that IgG4 might activate complement via MBL (33), we found that C3b deposition by all IgG1 and IgG4 glycovariants was fully inhibited when blocking C1q, and not, or only slightly, in case of blocking MBL or FD ( Figure 4A, C ). This indicated that the main mode of activation of IgG4 is via the classical pathway and moreover, that this activation is irrespective of glycosylation as all variants rely on C1q for activation. Moreover, we found that C1q could also bind to IgG4, although to a lesser extent than to IgG1 ( Figure 4E ). Inhibition of FD decreased C3b deposition slightly compared to the condition without any inhibitor, suggesting that the alternative pathway may amplify activation via the classical pathway. As the alternative pathway requires higher serum concentrations to be studied, we also performed the experiment at 10% serum and allowed deposition at 37°C and we found similar results ( Supplementary Figure 3D ). No clear inhibition was found when MBL was blocked. The blocking antibody was able to completely block mannan-induced C3b deposition at equal serum concentrations, showing that the antibody is effective ( Supplementary Figure 4A ). Furthermore, when the assay was repeated with serum of an MBL deficient donor, IgG4 was still able to activate complement ( Supplementary Figures 4A, B, D ) The glycovariants were not affected differently by any of the inhibitions. Similar results at the level of cell lysis were observed: C1q inhibition significantly reduced IgG4-mediated lysis ( Figures 4B, D ), whereas inhibition of either MBL or FD had no impact on the induced lysis by IgG4. Furthermore, IgG4 could still induce cell lysis with serum from an MBL deficient donor ( Supplementary Figure 4C ).

Taken together, no role for the lectin pathway in IgG4 complement activation was found, which instead appears to mainly employ the classical pathway that was slightly enhanced by galactosylation on the level of C3b deposition.

In our previous assays, only bivalent-monospecific antibodies were used, but IgG4 antibodies can recombine half-molecules with different antigen specificities in vivo. Through this Fab arm exchange process, monovalent-bispecific IgG4 antibodies with decreased avidity for their targets are generated (15). We therefore exchanged IgG4 monoclonal anti-biotin and anti-TNP half-molecules with an irrelevant clone (natalizumab with no target in our assays) to create a bispecific antibody that would bind to biotin and TNP respectively with only one arm. In addition, a K409R mutation in the IgG1-Fc was introduced, which allowed for the IgG1 anti-biotin clone to be exchanged with an adalimumab clone (also no target in our assays) with the same mutation. For the fab arm exchange process mild reduction conditions are necessary. These antibodies were then tested for C3b deposition via ELISA, which was found to be reduced for both anti-biotin and anti-TNP bispecific IgG4 clones, but also for the bispecific anti-biotin IgG1 clone in comparison to their monospecific counterparts. The binding of the monospecific and bispecific IgG antibodies to either biotin ( Figure 5A ) or TNP ( Figure 5B ) was comparable for parental (mock condition) and Fab arm exchange clones. To conclude, this shows that Fab arm exchange greatly reduces the capacity of IgG4 to induce complement activation.