Author Correction: Human DC-SIGN and CD23 do not interact with human IgG

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Human immunoglobulin G (IgG) molecules contain a conserved N-linked glycan in their crystallizable fragment (Fc) at the asparagine residue at position 297 (N297). In addition, approximately 15% of total IgGs have N-glycans in the antigen binding fragment (Fab) in the antigen binding part (VH and VL domains) 1,2 . These glycans have a biantennary core structure compose d of mannose and N-acetylglucosamine (GlcNAc) residues which can be extended with other sugar moieties such as fucose, galactose and sialic acid. Fc-glycans are required for the open conformation of the CH2 region of the IgG Fc domain to facilitate binding of Fc gamma receptors (FcγRs) 3 . Subtle changes in Fc-glycan composition influence the affinity of IgG for FcγRs and effector functions 4 . For example, it is well established that afucosylated IgG has an enhanced binding affinity to FcγRIIIa and FcγRIIIb [4][5][6][7][8] .
Recently, it has become apparent that galactosylation, which has less effect on binding affinity of fucosylated IgG to FcγR, further enhances binding affinity of afucosylated IgG to FcγRIIIa/b 7,9 . Notably, the elevated binding of these IgG glycoforms to members of the FcγRIII family (FcγRIV in mice) seems to be related to the fact that only these receptors have an N-linked glycan at position 162 that also directly interacts with the IgG Fc-glycan, causing favorable binding to afucosylated IgG [10][11][12] . Irrespective of the fucosylation status, IgG Fc-galactosylation and -sialylation, increases the binding of IgG to C1q and thereby affects classical complement activation 7,13 . Therefore, IgG glycosylation encodes for an important immunomodulatory ability.
Because of its anti-inflammatory effect, intravenous immunoglobulin (IVIg) therapy is used to treat a wide range of patients with autoimmune diseases and other inflammatory conditions. The precise mechanisms underlying this treatment still remain to be elucidated. The current ideas are that several mutually nonexclusive mechanisms exist, IgG Fab-and/or Fc-mediated, that are responsible for its clinical actions 14 . Because of the immunomodulatory potential, IgG glycosylation has been suggested to be an additional factor responsible for the effectiveness of IVIg treatment [15][16][17][18][19][20] . During the last decade, a considerable amount of prominent work has been put forward hinting at an alternative mechanism 17,18,[21][22][23] . One suggested mechanism is that the IVIg fraction with sialylated Fc (approximately 16% of total IVIg 24 ) is responsible for the anti-inflammatory effect via the C-type www.nature.com/scientificreports www.nature.com/scientificreports/ lectin DC-SIGN (dendritic cell-specific ICAM-3 grabbing non-integrin), which has been designated by some as a type II FcR 25 . Another member of this Type II FcR family is the C-type lectin CD23 (low affinity receptor for IgE Fc). In the proposed mechanism, these homologous receptors (both located on chromosome 19 26 ) can both bind  the IgG CH2-CH3 interphase, in a similar fashion as CD23 binds to IgE 23 . IgG binding to DC-SIGN, found on  regulatory macrophages and dendritic cells, has been reported to result in the release of regulatory cytokines and  subsequently upregulation of the inhibitory FcγRIIb on effector macrophages, causing an increased threshold for  inflammation 17,19,25 . IgG binding to CD23, found on mature B-cells, has also been reported to cause upregulation of FcγRIIb, which increases the threshold for B-cell receptor signaling and affects Ig-affinity maturation 27,28 . According to this model, CD23 and DC-SIGN only bind to IgG, in a calcium-dependent manner, when the IgG Fc-glycan is α2,6-sialylated. Upon sialylation, the Fc domain was hypothesized to undergo a conformational change from an "open" to a "closed" state 23 . However, follow-up work has suggested that these structural changes are minimal and confined to the CE loop in the upper CH2 region 29,30 . In addition, direct evidence for the interaction between IgG and CD23/DC-SIGN consists of two cell-based ELISA assays demonstrating that CD23/DC-SIGN-expressing Chinese hamster ovary (CHO) cells preferentially bind to sialylated human IgG Fc derived from IVIg or recombinant human IgG1 18,23 . Predicted interactions between crystal structures and the effects that can be observed in in vitro and in vivo tests provide further indirect evidence 18,21,25,27,28,31,32 . One of these in vivo studies used a transgenic mouse model in which a SIGN-R1 (specific ICAM-grabbing non-integrin R1) knock-out resulted in an abolishment of the anti-inflammatory response after administration of sialylated IgG 21 . Transgenic expression of human DC-SIGN in these SIGN-R1-deficient mice resulted in restoration of this immunomodulatory response 27,28 .
However, a considerable amount of contradictive studies emerged, mostly showing that DC-SIGN is not involved and/or sialylated IgG is not a stronger anti-inflammatory substance in several in vitro functional assays 15,20,33,34 and/or in mouse models [35][36][37] . Some studies have suggested that it is not sialylation in the Fc but rather sialylation in the Fab domain that may exert certain anti-inflammatory effects 20 , while others found no additional effect of Fab-sialylated IVIg 36 . One group demonstrated that binding of recombinant DC-SIGN tetramers to hexameric IgG Fc-based scaffolds does not depend on sialylation, but rather on high-mannose levels on the N-linked glycan 38 , which was also found in the glycan binding assays performed by Guo et al. 39 . Interestingly, Yu et al. demonstrated in an ELISA-based study that well-characterized sialylated IgG Fc fragments do not bind directly to human DC-SIGN 40 .
For most of these binding studies, either a well-defined set of glyco-engineered IgG was lacking 18,23 , or the human CD23-IgG interaction was not studied 40 . The aforementioned controversies, extensively reviewed elsewhere 41,42 , need to be settled.
To study this in more detail, we extensively investigated the direct binding between native human CD23/ DC-SIGN and a well-characterized set of human IgG glycoforms. In addition to flow cytometry analysis, we also used a biosensor platform allowing us to study low affinity and avidity interactions for both receptors CD23 and DC-SIGN expressed in its native configuration on cells. Furthermore, the context of the IgG α2,6-sialylation was studied, by including both Fab-sialylated IgG and altering the naturally-occurring variable glycan adducts in the sialylated Fc-glycan, in any possible combination.

Results
Generation of human IgG1 glycoforms. In order to test whether IgG glycosylation determines the binding capacity to the proposed IgG Fc receptors DC-SIGN and CD23, we first generated a defined set of glyco-engineered recombinant human IgG1 variants ( Fig. 1) 7,43 . Our choice for IgG1 was based on the fact that it comprises the largest fraction in IVIg (approximately 60%) 44 and because studies demonstrating direct binding of sialylated IgG to CD23/DC-SIGN all used Fc fragments which originated either from IVIg or recombinant human IgG1 18,23 . Also the immunomodulatory effects in mouse models that have been shown for sialylated IgG from IVIg 18,22,45 could be fully recapitulated by sialylated recombinant human IgG1 27,28,46,47 . The produced IgG Fc-glycoforms (asialylated/unmodified; galactosylated; sialylated with (+) or without (−) fucose (F) or bisecting GlcNAc (B)) were characterized by tandem liquid chromatography and mass spectrometry ( Fig. 1b and Supplemental Fig. S1), and show the expected glycan features. Levels of di-galactosylation and di-sialylation of those IgG glycoforms enriched in galactose and sialic acid, were highly elevated above that of mono-galactosylated and mono-sialylated IgG species (Fig. 1b and Supplemental Fig. S1). In addition, IgG1 with Fab glycosylation sites (known to have high level of α2,6-sialic acid) 36,48 were produced and sialylation levels were verified by performing a Sambucus nigra agglutinin (SNA)-ELISA assay (Fig. 1c).

Human CD23 and DC-SIGN expressed on human embryonic kidney cells recognize natural ligands.
Human CD23 and DC-SIGN are normally expressed on cell surfaces as multimers (reported as trimers and tetramers, respectively) with intertwined C-type lectin domains held together by an α-helical coiled-coil stalk 26 . In order to mimic this native configuration, we generated human embryonic kidney (HEK) 293 Freestyle cells stably expressing either human CD23 or DC-SIGN. FcγRIIa (CD32a)-expressing HEK cells were generated and used as positive control for low affinity IgG binding. FACS results showed that human FcγRIIa, CD23 and DC-SIGN were specifically recognized by antibodies directed against the corresponding receptor (

Soluble human IgG1 glycoforms do not bind to CD23-or DC-SIGN-expressing cells. In order
to investigate binding to human IgG1, we then tested binding of the IgG1 glycoforms to the cells by FACS, in the presence of calcium. No binding was observed at 37 °C for cells expressing either CD23 or DC-SIGN, not even www.nature.com/scientificreports www.nature.com/scientificreports/ for the Fab-sialylated or the most prevailing naturally-occurring sialylated Fc-glycoform (+F−B) ( Fig. 2d and Supplemental Fig. S2 for representative histograms). Also, a lower binding temperature (4 °C) or lower IgG1 concentrations did not result in binding (Supplemental Fig. S2). In contrast, substantial binding was observed for all Fc-and Fab-glycosylated IgG1 to the low affinity FcγRIIa.  Fig. S2) was detected by FACS, we could not rule out that even lower affinity interactions were not being detected in these assays. Therefore, and to study potential avidity-dependent interactions between human IgG1 and CD23/DC-SIGN, cellular Surface Plasmon Resonance imaging (cSPRi) tests were performed. Here we used trinitrophenyl-conjugated human serum albumin (HSA-TNP) as a model antigen which was spotted onto the sensor. Anti-TNP IgG1 glycoforms were injected and binding and formation of antigen-antibody complexes was observed in real-time ( Fig. 3a; 1-2). This was followed by the injection of the generated cell lines ( Fig. 3a; 3). First, without flow, the cells were allowed to sediment and interact with the antigen-bound antibodies in the presence of calcium ( Fig. 3a; 4). Subsequently, the immune-complex-bound cells were washed with increasing flow speed ( Fig. 3a; 5). Importantly, none of the cell lines showed binding to the negative control spots (Fig. 3b). Compared to Mock-transfected cells, FcγRIIa-expressing HEK cells bound asialylated IgG1 and, as expected, no binding was observed for Mock-, CD23-, or DC-SIGN-expressing HEK cells (Fig. 3b). Both CD23-and DC-SIGN-expressing cells showed an increased binding response on spotted anti-CD23 and anti-DC-SIGN, respectively (Fig. 3c). Similarly, the FcγRIIa-expressing HEK cells showed binding to anti-CD32 on the SPRi-sensor.
Importantly, no significant binding of CD23-, DC-SIGN-, or Mock-expressing HEK cells was observed to human IgG1 with varying degrees of Fc-glycan sialylation (Fig. 4a-c), irrespective of the presence or absence of fucose and/or bisecting GlcNAc moieties (Fig. 4c-f). Moreover, the presence of sialylated Fab-glycans did not trigger the binding of any of these cells (Fig. 4g). In addition, varying the HSA-TNP antigen density level either 10 times higher or lower (100 and 1 nM) did not show specific binding of CD23/DC-SIGN-expressing cells to Fc-or Fab-sialylated IgG1 (Supplemental Fig. S3).

Discussion
In this study, we found no evidence that human IgG1 binds either native CD23 or DC-SIGN regardless of the sialylation status of IgG, whereas binding of these receptors to their native ligands was easily detected. These data are, however, in full agreement with those of Crispin et al. who also found no binding of sialylated IgG Fc to DC-SIGN 30,40 . Here, we investigated the binding in more detail by using a biosensor platform, which allowed us to study avidity interactions, and the study was extended by including CD23. In addition, the context of the IgG www.nature.com/scientificreports www.nature.com/scientificreports/ sialylation was studied extensively, by including both Fab-sialylated IgG and altering the naturally-occurring variable glycan adducts (fucose and bisecting GlcNAc) in the biantennary sialylated Fc-glycan, in any possible combination. Still no binding could be detected, while binding was detected for the low affinity FcγRIIa which acted as a positive control. Taken together, these data do not support CD23 and DC-SIGN to be classified as true FcγRs. www.nature.com/scientificreports www.nature.com/scientificreports/ Part of the explanations for the opposing findings here and in literature concerning the binding of IgG to either CD23 and/or DC-SIGN 14,[17][18][19]21,23,25,30,31,33,40,42 may be due to differences in methodology and materials. For instance, most previous studies lack a characterization of the glycoforms used and relied on SNA-purified IgG from serum or IVIg, or ex vivo treatment of IgG or fragments thereof. Differences in used disease models may also be a reason for the apparent contradictions in literature. Many of these studies used human IgG in a murine context, which also introduces slight differences in structural composition of sialic acid between human and mice [49][50][51][52] . Other studies demonstrating binding of sialylated human IgG to DC-SIGN-or CD23-expressing CHO cells 18,23 , possibly introducing an incomplete human context e.g. the lack of α2,6-linked sialic acid, addition of alpha-Gal, and perhaps elevated level of mannose-containing structures 52,53 . In addition, in most in vivo studies human IgG was administered to mice which express SIGN-R1, potentially resulting in an incorrect translation from murine to human setting 18,21,33,41 .
Considerable part of the evidence for the binding between CD23/DC-SIGN to sialylated human IgG is based on predicted interactions between crystal structures. The structure of the sialylated human IgG1 Fc domain was determined by Ahmed et al. and showed that upon sialylation the Fc domain indeed undergoes a conformational change in the CE loop of the CH2 domain resulting in an increased interdomain flexibility 29 . However, no considerable conformational differences can be observed in the IgG1 CH2-CH3 interphase where CD23 and DC-SIGN have been proposed to bind 23 . The group of Crispin et al. also determined the structure of the human sialylated Fc domain and was not able to observe conformational changes upon sialylation, even not in the CH2 domains 30 . Within the same group, Yu et al. demonstrated that there is no specific binding of IgG to DC-SIGN tetramers, regardless of sialylation status, and that serum IgG was not able to inhibit the binding of DC-SIGN to natural ligand HIV envelope glycoprotein gp120 40 . Therefore, these authors concluded that the direct binding between DC-SIGN and human IgG is uncertain. Another aspect of the proposed sialylation-induced "closed" IgG Fc state, is the major reduction (up to 10-fold) in binding affinity of IgG1 to conventional FcγRs 18,22 . Currently, evidence is growing that IgG Fc-sialylation does not or only subtly affect binding affinities to FcγRs 54,55 . While Subedi and Barb found a modest increase in affinity (1.16-fold) for afucosylated IgG1 to both FcγRIIa and FcγRIIIa/b 55 , our www.nature.com/scientificreports www.nature.com/scientificreports/ group previously found a slight decrease (approximately 2-fold) in affinity of whole IgG1 for FcγRIIIa/b due to sialylation, but only when the Fc-glycan was also afucosylated and bisected 7 .
Our results do not imply that the function of IgG1 (or other subclasses) Fc-/Fab-sialylation or the involvement of C-type lectins (or other lectins, e.g. CD22) in IVIg therapy can be completely excluded. The mode of action of IVIg is complex, involving many underlying mechanisms that are both Fab-and Fc-mediated 14,22 . The most important mode of action is likely depending on its context. However, we conclude that a direct binding of either CD23 or DC-SIGN to sialylated IgG is very unlikely to occur. Therefore, these molecules should not be considered as bona fide receptors for IgG in humans.

Methods
Cloning. DNA encoding human DC-SIGN (AAK20997.1) was synthesized and codon-optimized by Geneart (Invitrogen, USA) and DNA encoding human CD23 cDNA (BC062591) and FcγRIIa 131 H (XM_017000664.1) was synthesized by Integrated DNA Technologies, USA. All constructs were generated with flanking HindIII and EcoRI restriction sites for cloning into pcDNA3.1 (+) expression vector (Invitrogen, USA) with ampicillin and neomycin resistance genes using Rapid Ligation Kit (Thermo Scientific, USA). Ligation products were transformed into DH5α heat-shock competent cells (Thermo Scientific, USA). DNA sequences isolated from Ampicillin resistant clones were verified by Sanger sequencing in a 3730 DNA Analyzer. Antibodies and other reagents. Human anti-TNP IgG1 production and glyco-engineering was performed as described by Dekkers et al. for Fc-glycoforms 8,43 . In brief, vectors encoding p21, p27 and pSVLT and a single-gene construct encoding human anti-TNP IgG1 heavy chain and kappa light chain were transiently transfected into HEK 293 FreeStyle cells using 293fectin (Invitrogen, USA). To add bisecting GlcNAc to the N297 glycan, cells were co-transfected with a vector encoding GlcNAc transferase (GNTIII). To decrease fucosylation, decoy substrate 2-fluorofucose (2FF) was added to the medium. For IgG1 galactosylation, D-galactose (Sigma-Aldrich Chemie N.V., The Netherlands) was added to the medium and cells were co-transfected with a vector encoding beta-1,4-galactosyltransferase (B4GALT1). For α2,6-sialylation, the latter treatment was combined with co-transfection with a vector encoding beta-galactoside alpha-2,6-sialyltransferase 1 (ST6GALT) and after purification these antibodies were further sialylated in vitro by incubation with recombinant ST6GALT enzyme and the substrate cytidine-5′-monophospho-N-acetylneuraminic acid. Unmodified anti-TNP IgG1 was used as a control for asialylation. For the Fab-glycoform, glycosylation sites in the VH and VL domain were introduced by substituting respectively the serine residue on position 29 and aspartic acid residue on position 86 into asparagine. Constructs encoding the anti-TNP IgG1 heavy chain-(S29N) and kappa light chain-(D86N) Fab glycosylation were transfected into HEK cells in the same manner as the anti-TNP Fc-glycoforms. For all glycoforms, cell supernatant was harvested five days after transfection and purified using a HiTrap Protein A HP column (GE Healthcare, USA) on an ÄKTAprime plus system (GE Healthcare, USA). Commercially available antibodies and other reagents used in ELISA, FACS and cSPRi tests are depicted in Table 1.
Trypsin digestion and mass spectrometric glycosylation analysis. The IgG samples were dried in a centrifugal vacuum concentrator (Eppendorf, Germany) for 2 hours at 60 °C. Subsequently, the IgG was digested overnight at 37 °C in 40 µl (25 mM) ammonium bicarbonate (Sigma-Aldrich Chemie N.V., The Netherlands) containing 200 ng trypsin (sequencing grade, Promega, Madison, WI). The resulting glycopeptides were analyzed by nanoLC-ESI-QTOF-MS on an RSLCnano Ultimate 3000 system (Thermo Fisher, The Netherlands) coupled to an Impact quadrupole-TOF-MS (micrOTOF-Q, Bruker Daltonics, Germany). The system and glycopeptide analysis are described in detail by Dekkers et al. 43 .
Lectin enzyme-linked immuno sorbent assay. To confirm the occupancy of the Fab glycosylation sites, lectin ELISA was performed as described previously by van de Bovenkamp et al. 48 . In brief, 96-well flat bottom plates (Nunc MaxiSorp, Denmark) were coated overnight with neuraminidase (New England Biolabs, USA)-treated mouse anti-human IgG diluted in PBS. Plates were incubated for 1 hour with 100 µl sample (25 ng/ ml) diluted in PBS supplemented with 0.02% Tween-20. Subsequently, plates were incubated with biotinylated Sambucus nigra agglutinin (SNA) for 2 hours for sialylation measurement. Then, plates were incubated with HRP-conjugated streptavidin for 10 minutes. In order to correct sialylation optical densities for total IgG, in parallel, total IgG levels were determined for all samples by incubating for 1 hour with HRP-conjugated mouse anti-human IgG. Plates were washed three times in between each incubation step and detection was performed with tetramethylbenzidine and stopped with 2 M H 2 SO 4 . Absorbance was measured at 450 and 540 nm on a BioTek plate reader. www.nature.com/scientificreports www.nature.com/scientificreports/ IgG1 glycoforms or biotinylated mouse anti-CD32 and subsequently incubated with fluorescently labeled secondary antibodies/F(ab') 2 fragments or streptavidin. Incubation times were one hour and all steps were performed at 4 °C, except for the ligand and IgG glycoform 37 °C binding studies. Sample fluorescence was measured by a FacsCanto II cytometer (BD Biosciences, USA) and at least 10,000 events were recorded per sample. Reported data show geometric mean fluorescence intensities with the unstained cell signals (autofluorescence) subtracted. Cellular surface plasmon resonance imaging. An Easy2Spot pre-activated SensEye P-type sensor (Ssens, The Netherlands) was spotted, in triplicate, with HSA-TNP 7 (100, 10, 1 nM), HSA (100 nM) (Sanquin, The Netherlands) and antibodies against CD23/DC-SIGN/CD32 (100 nM) by a continuous flow microspotter (CFM) (Wasatch Microfluidics, USA). Spotting buffer consisted of 10 mM sodium acetate/acetic acid supplemented with 0.075% Tween-80 at pH 5.5, except for HSA and HSA-TNP, which was pH 4.0. Deactivation of the sensor was performed with 100 mM ethanolamine, pH 8.0 supplemented with 0.075% Tween-80 and 0.1% HSA in acetate buffer, pH 3.7 supplemented with 0.075% Tween-80. Analyte binding was performed in an IBIS MX96 (IBIS Technologies B.V., The Netherlands). After injection, antibodies were allowed to bind to HSA-TNP spots for 5 minutes. A short wash was performed to remove the unbound antibodies from the sensor. Subsequently, cells were injected and allowed to sediment/bind to immobilized glycoforms or control spots for 25 minutes 56 . During the next 11 minutes, the wash flow speed was increased stepwise from 1 to 120 µl/s. Regeneration was performed using glycine HCl pH 2.2 supplemented with 0.075% Tween-80.
In the first (antibody) flow either one of the human IgG1 glycoforms (100 nM) was injected into the flow chamber, in the second (cell) flow either HEK DC-SIGN/CD23/FcγRIIa/Mock (5.0 • 10 6 cells per ml). System buffer consisted of 20 mM HEPES buffer supplemented with 1 mM CaCl 2 , 1 mM MgCl 2 and 1% BSA, pH 7.4. Data analysis. FACS data was analyzed using FlowJo software (FlowJo, USA). cSPRi data was analyzed using SprintX software (IBIS Technologies B.V., The Netherlands) and Rstudio software (RStudio, USA) 57 . Graphpad Prism 8.0.2 (GraphPad Software, Inc., USA) was used to generate sensorgrams and to perform statistical analyses. To optimize visualization of the cell binding during sedimentation and flow phase in the cSPRi data, per HSA-TNP density, consecutively individual sensorgrams were normalized to the representative antibody binding plateau and buffer signals were spot-specifically subtracted for each time point.

Data Availability
We will make materials, data and associated protocols available to readers. www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/