Elevated N-Linked Glycosylation of IgG V Regions in Myasthenia Gravis Disease Subtypes

The vast diversity of IgG V regions (IgG-Vs) is critical for host immunity. This diversity arises through VDJ recombination and somatic hypermutation (SHM). Historically, IgG-V diversity has been represented by amino acid sequence alone with little focus on posttranslational modifications. Recently, the presence of N-linked glycosylation of IgG-Vs (IgG-V^N-Glyc) has been shown to contribute to diversity (1, 2). IgG-V^N-Glyc is contingent on the presence of the predictive amino acid motif N-X-S/T, where X can be any amino acid except for proline. This motif is most often introduced as a consequence of SHM (3). Less often it can be provided by the few germline gene segments (IGHV1-8, IGHV4-34, IGHV5-10-1, IGLV3-12, and IGLV5-37) in which it is encoded (4).

Higher frequencies of IgG-V^N-Glyc than that found in healthy individuals have been observed in B cell malignancies (5–9) and in autoimmune diseases (10). Specifically, increased frequencies have been reported for ANCA-associated vasculitis (AAV) (11–13), rheumatoid arthritis (RA) (14–17), and primary Sjogren’s syndrome (pSS) (18, 19). The in vivo function of glycosylation in the IgG-V, a critical region of Ag contact, is not thoroughly understood. Follicular lymphomas may leverage N-linked glycosylation of their BCR V regions to activate Ag-independent signaling pathways that support survival (20). Ag binding can also be influenced by IgG-V^N-Glyc; this includes both increases and decreases in affinity and modulated functional activity. This is well highlighted by anti-citrullinated protein autoantibodies (ACPAs) found in patients with RA, where 80–100% of the autoantibodies include IgG-V^N-Glyc, and Ag binding properties are consequently altered (15, 16, 21).

Myasthenia gravis (MG) is an autoimmune disorder that affects neuromuscular transmission. Patients with MG experience severe muscle weakness and increased fatigability (22, 23). The molecular immunopathology of MG is directly attributed to the presence of circulating IgG isotype autoantibodies specifically targeting extracellular domains of postsynaptic membrane proteins at the neuromuscular junction (23, 24). The most common subtype of autoantibody-mediated MG (∼85% of patients) is characterized by autoantibodies against the nicotinic acetylcholine receptor (AChR) (23). In many of the remaining patients, autoantibodies targeting the muscle-specific tyrosine kinase (MuSK) or lipoprotein receptor–related protein 4 (LRP4) are present (25–28). Although MG autoantibodies cause disease, the underlying immune pathophysiology of the MG subtypes is distinct (29). AChR MG is governed primarily by IgG1 and IgG3 subclass autoantibodies (30, 31), which facilitate pathology through blocking acetylcholine, activating complement-mediated damage, or initiating internalization of AChRs (32–35). LRP4 autoantibodies putatively disrupt Agrin-LRP4 signaling and primarily belong to the IgG1 and IgG2 subclasses (27, 36). Conversely, the MuSK MG subtype is most often associated with IgG4 subclass autoantibodies, which are incapable of activating complement and other effector functions, but rather mediate pathology through blocking MuSK binding partners and its kinase activity (37–39).

Given that IgG isotype autoantibodies directly facilitate MG pathology, and that their autoimmune mechanisms are divergent, we hypothesized that N-linked glycosylation might be differentially represented in the distinct AChR and MuSK MG subtypes. To that end, we applied complementary sequencing and proteomic-based approaches to investigate V region N-linked glycosylation patterns of the bulk BCR repertoire and total circulating IgG in AChR and MuSK MG. Nucleotide-level sequencing was used to test for elevated IgG-V^N-Glyc frequency in AChR and MuSK MG BCR repertoires. Total IgG from sera was then evaluated with proteomic approaches to determine whether elevated IgG-V^N-Glyc could be observed in the circulation. Finally, we explored whether N-linked glycans impact binding to pathogenic targets by using four patient-derived monoclonal autoantibodies with N-linked glycan occupancy validated by mass spectrometry. We show that IgG-V^N-Glyc sequence motifs and total IgG-V^N-Glyc are more frequent in both AChR and MuSK MG in comparison with healthy control subjects, and that the patterns differ between the two MG subtypes. However, the presence of IgG-V^N-Glyc was not required for Ag binding by the four patient-derived pathogenic MG autoantibodies we tested.

This study was approved by Yale University′s Institutional Review Board (Clinicaltrials.gov: NCT03792659). Informed written consent was received from all participating patients before inclusion in this study. PBLs (PBMCs) and serum were collected from subjects with MG at the Yale Myasthenia Gravis Clinic (New Haven, CT) (40). BCR repertoire sequencing for the MuSK MG cohort (n = 3) was derived from our previous study (41). AChR MG patients (n = 11) and heathy control subjects (n = 9) were selected for BCR-based adaptive immune receptor repertoire sequencing (AIRR-seq) using PBMC-derived RNA. Serum samples from all three MuSK MG subjects and nine AChR MG subjects (eight overlapping with paired AIRR-seq) were also investigated for the presence of VH gene glycosylation-specific signatures in the serum. For all patients with MuSK MG diagnoses and several patients with AChR MG diagnoses, longitudinal serum samples were collected. Detailed demographics and clinical data for the study cohort are presented in Supplemental Table I.

A patient with RA (n = 1) was enrolled in an Institutional Review Board–approved research study at the University of California San Francisco for pathogen and autoantibody detection.

For the initial screening processes, patient serum samples were diluted 1:1 with 2× storage buffer (2× PBS, 20 mM HEPES, 0.04% sodium azide, 20% glycerol). The IgG from human sera or MuSK-specific mAb 4A3 and AChR-specific mAb 637 were captured with AG beads (Thermo Fisher) and then eluted by boiling at 95°C in 2× Laemmli buffer (with 10% 2-ME). For immunoblotting, the gel was transferred to 0.45-μm nitrocellulose membrane and blotted with secondary anti-human IgG conjugated to IR800 dye (LI-COR Biosciences). Nitrocellulose blots were imaged with LI-COR scanner and analyzed qualitatively by eye for the presence of altered migration patterns in IgG. Potential IgG H chain migration phenotypes were qualitatively called by an experimenter blind to experimental conditions. For the three MuSK-specific mAbs (MuSK1A, MuSK1B, and MuSK3-28) and the AChR-specific mAb 637, Mini-PROTEAN TGX Stain-Free Precast Gels (Bio-Rad) and Laemmli Sample Buffer (Bio-Rad) were used for SDS-PAGE. Before electrophoresis, the proteins were reduced with 0.1 M DTT (Thermo Fisher Scientific) and heat denatured at 95°C for 5 min. After electrophoresis, the gel was stained with Coomassie blue solution. Bands were visualized with the ChemiDoc Touch Imaging System (Bio-Rad). Enzymatic assays for PNGase F and Endo S were performed according to the manufacturer’s instructions (NEB). The effect of the enzymatic assays was analyzed by either Coomassie staining or immunoblotting.

First, RNA was isolated from frozen PBMCs using the RNeasy Mini kit (Qiagen) per the manufacturer’s instructions. Bulk libraries were prepared from RNA using reagents from New England Biolabs as part of the NEBNext Immune Sequencing Kit as described previously (42, 43). In brief, cDNA was reverse transcribed by a template-switch reaction to add a 17-nt unique molecular identifier to the 5′ end with streptavidin magnetic bead purification. This was then followed by two rounds of PCR; the first round enriched for Ig sequences using IGHA-, IGHD-, IGHE-, IGHG-, and IGHM-specific 3′ primers and added a 5′ index primer. Libraries were purified with AMPure XP beads (Beckman), after which another round of PCR added Illumina P5 Adaptor sequences to each amplicon. The number of PCR cycles was selected based on real-time quantitative PCR to avoid the plateau phase. Libraries were then purified again with AMPure XP beads. Libraries were pooled in equimolar libraries and sequenced by 325 cycles for read 1 and 275 cycles for read 2 using paired-end sequencing with a 20% PhiX spike on the Illumina MiSeq platform according to the manufacturer’s recommendations.

Processing and analysis of bulk BCR sequences were carried out using tools from the Immcantation framework, as done previously (44). Preprocessing was performed using pRESTO. In brief, sequences with a phred score less than 20 were removed, and only those that contained C region and template switch sequences were preserved. Unique molecular identifier sequences were then grouped, and consensus sequences were constructed for each group and assembled into V(D)J sequences in a two-step process involving an analysis of overlapping sequences (<8 nt) or alignment against the IMGT (international ImMunoGeneTics information system) IGHV reference (IMGT/GENE-DB v.3.1.19; retrieved December 1, 2019) if no significant overlap was found. Isotypes were assigned by local alignment of the 3′ end of the V(D)J sequence to C region sequences. Duplicate sequences were removed, and only V(D)J sequences reconstructed from more than one amplicon were preserved. Primer sequences used for this analysis are available online at: https://bitbucket.org/kleinstein/immcantation.

V(D)J germline genes were assigned to reconstructed V(D)J sequences using IgBLAST v.1.14.0 and also using the December 1, 2019 version of the IMGT gene database (45). V(D)J sequences with IGH-associated V and J genes were then selected for further analysis, and nonfunctional sequences were removed. Germline sequences were reconstructed for each V(D)J sequence with D segment and N/P regions masked (with Ns) using the CreateGermlines.py function within Change-O v.1.0.0 (46). VH gene nucleotides up to IMGT position 312 were translated from both the aligned sequence and germline reconstructed V(D)J sequence using BioPython v.1.75. To quantify the frequency of N-X-S/T glycosylation motifs, matches to the regular expression pattern “N[^P][S,T]” were quantified for each translated sequence, including for translated CDR and framework region (FWR) fragments of the VH gene sequence (defined by IMGT coordinates) (4). N-X-S/T glycosylation motifs in the CDR3 and FWR4 regions were similarly quantified separately and included for CDR and FWR distribution analyses.

To build the lineage tree in Supplemental Fig. 1A, we first clustered B cells into clones by partitioning based on common IGHV gene annotations, IGHJ gene annotations, and junction lengths. Within these groups, sequences differing from one another by a length-normalized Hamming distance of 0.2 within the junction region were defined as clones by single-linkage clustering using Change-O v.1.0.1 (46). The Hamming distance threshold was determined by manual inspection of the distance to the nearest sequence neighbor plot using SHazaM v.1.0.2 (47). Phylogenetic tree topology and branch lengths of an illustrative clonal lineage were estimated using the HLP19 model in IgPhyML v.1.1.3 and visualized using ggtree v.2.0.4 and custom R scripts (48, 49).

Sequencing data have been deposited in the Sequence Read Archive (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA738368/) under accession number PRJNA738368.

High-resolution mass spectrometry was used to confirm the change in glycosylation status between wild-type (WT) and mutated variants. To reduce complexity at the intact mass level, we used a “middle-down” approach (50, 51). Intact Abs were incubated with IdeS protease, followed by reduction of disulfide bonds. This workflow is well-known to break down Abs into three ∼25-kDa subunits, LC, Fc, and Fd, and thereby separate disease-associated glycosylation within the V region from standard glycosylation in the C region. The H chain V region (VH) is located within the Fd subunit. Purified mAb (2 mg/ml in PBS) was treated with 1 U of IdeS protease (Promega) per 1 µg of mAb, and the sample was incubated at 37°C in a shaking incubator for 1.5 h. The digested sample was then diluted into 6 M guanidinium chloride to a final IgG concentration of 1 mg/ml, and Tris(2-carboxyethyl) phosphine hydrochloride was added for a final concentration of 30 mM. The sample was incubated at 37°C in a shaking incubator for 1.5 h; then the reaction was quenched by the addition of trifluoroacetic acid (final trifluoroacetic acid concentration of 0.1% v/v). The sample was desalted by buffer exchange into liquid chromatography–mass spectrometry buffer (five rounds of buffer exchange with an Amicon Ultra-0.5 ml centrifugal filter unit, 10 kDa molecular mass cut off).

The digested and reduced Ab species were further desalted and separated on a monolithic C4 column (RP-5H, 100 mm, 0.5 mm i.d.; Thermo Scientific) with an Ultimate 3000 RSLCnano system (Thermo Scientific) using a binary gradient. The gradient used solvent A (95% water, 5% acetonitrile, and 0.2% formic acid) and solvent B (5% water, 95% acetonitrile, and 0.2% formic acid).

Data were acquired on a Q Exactive HF instrument with an attached HESI source (sheath gas = 10, auxiliary gas = 2, spare gas = 2, spray voltage = 3500 V, S-lens RF level = 65). MS1 acquisition used a scan range window of 400–2000 m/z with 1 microscan and an AGC target of 1e6, at a resolution of 15,000.

Glycosylation sites (N-X-S/T) present in the V regions of the mAbs were removed by mutating the asparagine (N) either to a glutamine (Q) or a serine (S). This was performed with Q5 Site-Directed Mutagenesis Kit (NEB) according to the manufacturer’s instructions. The primers were designed with NEBaseChanger. Sequences of all expression plasmids were verified by Sanger sequencing.

The mAbs were produced as previously described (38). In brief, HEK293A cells were transfected with equal amounts of the heavy and the corresponding L chain plasmid using linear PEI (catalog no. 23966; Polysciences). The media were changed after 24 h to basal media (50% DMEM 12430, 50% RPMI 1640, 1% antibiotic/antimycotic, 1% Na-pyruvate, 1% Nutridoma). After 6 d, the supernatant was harvested, and Protein G Sepharose 4 Fast Flow beads (GE Healthcare) were used for Ab purification.

Cell-based assays for detection of AChR or MuSK Ab binding were performed as we have previously described (52). In brief, the cDNA encoding human AChR α, β, δ, and ε subunits and rapsyn-GFP were each cloned into pcDNA3.1-hygro plasmid vectors (Invitrogen, CA), and cDNA encoding human full-length MuSK was cloned into pIRES2-EGFP plasmid vector (Clontech). AChR and MuSK vectors were kindly provided by Drs. D. Beeson and A. Vincent of the University of Oxford. HEK293T (ATCC CRL3216) cells were transfected with either MuSK-GFP or the AChR domains together with rapsyn-GFP. On the day of the CBA, the mAbs were added to the transfected cells in a dilution series (ranging from 10 to 0.02 μg/ml). The binding of each mAb was detected with Alexa Fluor–conjugated AffiniPure Rabbit Anti-Human IgG, Fcγ (309-605-008; Jackson Immunoresearch) on a BD LSRFortessa (BD Biosciences). FlowJo software (FlowJo) was used for analysis.

R v.4.0.3 was used for all statistical analyses. Data frame handling and plotting were performed using functions from the tidyverse v.1.3.0 in R and pandas v.0.24.2 in python v.3.7.5. A significance threshold of p < 0.05 was used and shown on plots with a single asterisk; double asterisks correspond to p < 0.01. Unpaired one-tailed Wilcoxon tests were used for comparisons with healthy control subjects in repertoire analysis; the alternative hypothesis was that the average count of glycosylation motifs for each V(D)J sequence in MG BCR repertoires would be higher.

N-linked glycosylation sites occur only at amino acid sequence positions with the motif (N-X-S/T, where X = not proline). Elevated IgG-V^N-Glyc in MG could arise from the introduction of these sites by SHM or the use of germline sites found in a small subset of VH gene segments (IGHV1-8, IGHV4-34, and IGHV5-10-1). To quantify global differences in the glycosylation frequency of the B cell repertoire, we examined the encoded BCR repertoire generated by AIRR-seq from the mRNA of circulating PBMCs from healthy donors (HDs) and patients with MG. For the AIRR-seq, the MuSK MG patient cohort (n = 3) included 12 unique time point samples, the AChR (n = 10) included 10 unique time point samples, and each HD (n = 9) included a single time point (Supplemental Table I). The AIRR-seq library included a total of 10,565,778 (H chain only) raw reads; after quality control and processing, a high-fidelity dataset was generated that consisted of 764,644 unique error-corrected sequences, which was further filtered to include only IgG subclass sequences that consisted of 232,094 sequences.

We observed a statistically significant elevation in median IgG-V^N-Glyc site frequency for AChR MG (13.0%; p = 0.039, one-tailed Wilcoxon test) and MuSK MG (17.4%, p = 0.018, one-tailed Wilcoxon test) in comparison with healthy control subjects (10.3%) (Fig. 1A). To investigate whether the increased frequency of N-linked glycosylation sites was generated through preferred use of he three VH gene segments that encode an N-X-S/T motif or through SHM, we assessed the frequency of the motif in germline reversions of the VH gene segments (Fig. 1B). We observed no increases in the germline frequency of IgG-V^N-Glyc sites when comparing healthy control subjects and AChR MG patients (p = 0.55, one-tailed Wilcoxon test), whereas the MuSK MG cohort exhibited a significant elevation (p = 0.05, one-tailed Wilcoxon test), thus reflecting increases in the usage of select V gene segments (IGHV4-34, IGHV1-8, and IGHV5-10-1, in descending order of usage frequency). An illustrative example of N-X-S/T motif acquisition and conservation through the SHM process is shown for a B cell clonal family present in a MuSK MG repertoire, which includes acquisition of two motifs (Supplemental Fig. 1A).

We then examined whether the increases in glycosylation frequency were specific to CDRs, which are primarily responsible for Ag contact, or also included distribution in the FWRs, which maintain structural integrity of the variable domains. The motif could be found in all CDRs and FWRs in sequences from the HDs and patients with MG with the exception of FWR2, in which the motif was absent in all sequences (Supplemental Fig. 1B). The motif was most often observed in the FWR3 sequences from the HDs and patients with MG. Sequences from the AChR MG patients revealed a significant increase in motif frequency only in the CDR2 region in comparison with HD (p = 0.047, one-tailed Wilcoxon test). Elevated frequencies were observed in comparisons of MuSK MG and HDs at all regions (Supplemental Fig. 1B), and statistically significant increases were observed in the FWR1, CDR2, and FWR4 regions (p = 0.009, p = 0.0045, p = 0.042, respectively, one-tailed Wilcoxon tests). Examining the location of the motifs within each region (Supplemental Fig. 1C) showed that they were present throughout but were not uniformly distributed, because some areas showed enriched accumulation. Those present in FWR1, although rare, were found close to the CDR1 (Supplemental Fig. 1C).

In summary, the frequency of V region N-linked glycosylation sites among IgG-switched B cells differ when comparing healthy control subjects and AChR or MuSK MG patients. These differences result from SHM in AChR MG, while differences found in MuSK MG result from both SHM and elevated usage of V genes with germline encoded N-linked glycosylation sites.

The heavy chains of total IgG isolated from serum of patients with autoimmune disease, such as RA and ANCA-associated vasculitis, migrate at a higher m.w. than those of healthy control subjects because of the presence of a large fraction of IgG-V^N-Glyc (13, 15). Having demonstrated that the B cell repertoire of both AChR and MuSK MG include elevated IgG-V^N-Glyc site frequency, we next sought to investigate whether total circulating IgG from patients with MG reflected this m.w. increase. To that end, we analyzed total IgG purified from serum samples from the MG cohort (MuSK MG, n = 3; AChR MG, n = 9) for the presence of IgG-V^N-Glyc (Supplemental Table I). Longitudinal samples were also included to evaluate the temporal stability of IgG-V^N-Glyc patterns (Supplemental Table I). The IgG-V^N-Glyc presence was tested through the assessment of IgH migration patterns by SDS-PAGE. Serum-derived IgG from a patient with RA was included as a positive control (Fig. 2A). Total IgG migration patterns between healthy individuals and patients with MG were compared (Fig. 2B); differences were noted for one AChR patient (AChR MG-1) and one MuSK patient (MuSK MG-1). Longitudinal samples were assessed spanning a period of 4 years of clinical disease; the altered migration patterns remained consistent through all of the time points collected from these two subjects (Fig. 2C). These two subjects also demonstrated an elevated frequency of IgG-V^N-glyc in their B cell repertoire (Fig. 1, arrows).

To assess whether the altered migration patterns observed in MuSK MG-1 and AChR MG-1 reflect elevated IgG glycosylation as opposed to other possible modifications, such as phosphorylation or ubiquitination, we subjected IgG to digestion with PNGase F or Endo S. Enzymatic digestion of MuSK MG-1 IgG with PNGase F, which nonspecifically cleaves N-linked glycosylation units, resulted in a loss of the atypical IgG migration pattern (Fig. 2D). By comparison, Endo S, which cleaves N-linked glycosylation at N297 of IgG C region, caused a shift in gel mobility in both samples but no change in the atypical pattern (Supplemental Fig. 2). Removal of phosphates with calf intestinal phosphatase, a phosphatase enzyme, had no effect on migration (Supplemental Fig. 2). These results suggest that a subset of patients with MG possess atypical IgG glycosylation specifically in the Fab region, likely as a result of IgG-V^N-Glyc, which appears to be a stable feature over long periods (3–4 y).

Having demonstrated that the bulk BCR repertoire and total circulating IgG included elevated IgG-V^N-Glyc frequencies, we next turned our attention to MG-specific autoantibodies. We had previously generated three human recombinant MuSK-specific mAbs that demonstrated in vitro pathogenic capacity (38, 53, 54). We found glycosylation motifs (N-X-S/T) in the V region of all three MuSK mAbs, in either the heavy (MuSK1A and 3-28) or L chain (MuSK1B) (Fig. 3A–C). Specifically, the motif was present in the H chain FWR3 of MuSK1A due to the use of IGHV1-8, where it is encoded in the germline. MuSK1B acquired the motif in the L chain (FWR1) through SHM, and the H chain lost the motif in the CDR2, which was present in the germline VH (IGHV4-34). MuSK3-28 acquired the motif in the H chain (CDR2) through SHM. We sought to test whether these sites were occupied. Digestion with PNGase F reduced the m.w. of the heavy (MuSK1A and MuSK3-28) and L chain (MuSK2A) of the mAbs, suggesting the presence of N-linked glycosylation on the Abs (Supplemental Fig. 3A–C). We then removed these putative glycosylation sites by mutagenesis and screened all constructs for variations in migratory pattern as a result of m.w. changes. Removal of glycosylation sites led to a change in gel mobility as expected in all three MuSK mAbs, which was also consistent with site-specific occupancy (Supplemental Fig. 3D–F). Next, we performed intact mass spectrometry analysis to more precisely detect these glycosylation sites (Fig. 3A–C). Differences in mass and mass spectra can be used to confirm the presence of IgG-V^N-Glyc. All three MuSK autoantibodies were found to be glycosylated, and the mutated variants were significantly less heterogeneous and lighter in mass by ∼2 kDa (Fig. 3A–C). Because N-glycans are extremely heterogenous molecular moieties, proteins containing IgG-V^N-Glyc have elevated mass spectra heterogeneity; these findings confirm the presence of glycosylation and that mutations were successful in disrupting the introduction of glycosylation in all three MuSK mAbs.

Next, to extend these findings to AChR MG, we evaluated the patient-derived AChR-specific mAb 637 (55) because its sequence shows two predicted N-linked glycosylation sites (N66 and N84) within the V region of the FR3 of H chain, which were acquired through SHM (Fig. 3D). The H chain of mAb 637 migrated at a lower m.w. when treated with PNGase F in comparison with untreated mAb (Supplemental Fig. 3G). We subsequently performed mutagenesis and produced several different constructs disrupting the two predicted glycosylation sites—at N66 and N84. Mutation of N66 alone or N66 and N84 together resulted in a construct that migrated at lower m.w., while mutation of only the N84 did not affect migration (Supplemental Fig. 3H, 3I). We further explored this result using intact mass spectrometry analysis of the Fd from the WT and three mutants (Fig. 3D). The WT construct and the construct containing a mutation at position N84 had a complex mixture of proteoforms clustered between 28.3 and 28.8 kDa, whereas constructs containing mutations at N66 (including a mutant at both N66 and N84) were less heterogeneous, with proteoforms clustering closer to 26.4 kDa (Fig. 3D). The lighter mass and simplified proteoform signature of variants containing a mutation at N66 suggest that N66 is the main site of glycosylation in mAb 637 and not N84. In addition, removal of the glycosylation site N66 did not shift glycosylation to the second predicted site at N84. In summary, autoantigen-specific mAbs in MG can contain occupied N-glycosylation of their IgG-Vs.

We then sought to test the contribution of IgG-V^N-Glyc to MG mAb binding. Given that these mAbs were previously validated for their capacity to bind AChR or MuSK in live cell-based assays (38, 53, 54), we tested the contribution of IgG-V^N-Glyc sites to binding using the same approach (Fig. 4). When tested over a wide range of concentrations (ranging from 10 to 0.02 µg/ml), we found that loss of IgG-V^N-Glyc did not affect binding of the three anti-MuSK mAbs or the anti-AChR mAb to their respective targets.

The percentage of IgG in healthy individuals that includes V region glycosylation is estimated to be between 10% and 25% (1). The wide range reflects different approaches of measurement, particularly whether the measurement was derived from genomic sequencing or analysis of Igs (1). Nucleotide-level sequencing demonstrates that the frequency of V region glycosylation motifs in healthy individuals is 9–12% (8, 19, 21, 56). Our BCR sequencing data of the IgG H chain VH gene segment from healthy control subjects identified ∼0.1 glycosylation motif per sequence. Because the motif is rarely present more than once in each sequence, this indicates that ∼10% of VH gene segments carry a glycosylation motif, which is consistent with these previous studies. We used this baseline value to establish that elevated numbers of the IgG-V^N-Glyc motif are found in the IgG VH gene segment sequences derived from MG disease subtypes. Two mechanisms are thought to contribute to increased IgG-V^N-Glyc motif frequency. The first is enriched usage, at the naive B cell stage, of the five germline VH or VL gene segments that contain N-linked glycosylation motifs (IGHV1-8, IGHV4-34, IGHV5-10-1, IGVL3-12, and IGVL5-37). The second mechanism is selection, during affinity maturation, of B cell clones that acquire N-linked glycosylation motifs through the SHM process. The germline encoded motifs in the three heavy chains are found in CDR2 of IGHV4-34 and in the FWR3 of IGHV1-8 and IGHV5-10-1. Our sequence analysis showed that these motifs, when they are acquired through SHM, are distributed throughout the V region with the exception of FWR2. The distribution mirrors SHM patterns in that mutations accumulate preferentially in CDRs and FWR3. Although replacement mutations can be observed in FWR2, our data [and that of others (21)] suggest that a glycosylation motif is not tolerated in this region, suggesting that such alterations are constrained by the role of the FWRs in conserving the overall structure of the Ab. Similarly, motifs found in FWR1 were restricted to regions near the flexible CDR loops that it flanks. These collective findings indicate that the acquisition of the motif may be driven by positive selection. It is also possible that the motifs could be selectively neutral but arise as a consequence of SHM. If so, MG repertoires could have more motifs than the HD repertoires simply by having more SHMs, and the motifs could be concentrated around the CDRs because of the presence of known hotspot motifs in those regions.

SHM appears to be a major contributor to the increased frequency of the IgG-V^N-Glyc sites in the AChR MG patients we studied. Positive selection leading to enriched N-linked glycosylation DNA motifs has been observed in the parotid gland of patients with pSS, a structure known to contain ectopic lymphoid follicles in these patients (19). Similarly, the thymus in a subset of patients with MG includes germinal centers, which are thought to contribute to the generation and maturation of AChR autoantibody-producing B cells (57–59). Thus, positive selection of N-linked glycosylation motifs may occur in this compartment, and support for this possibility is provided by our previous study, where we found that the IgG-switched BCR sequences in MG thymus were enriched in N-linked glycosylation motifs (43). This study was not designed to investigate whether thymic lymphofollicular hyperplasia or the surgical removal of the thymus is associated with the frequency of IgG-V^N-Glyc in the periphery. However, it is reasonable to speculate that the heterogeneity of IgG-V^N-Glyc frequency found in the AChR MG patients studied here may relate to MG disease subtypes, some of which are defined by thymus-related pathology.

Both V gene usage and the SHM process contributed to the elevated frequencies we observed in the MuSK MG patients. Defects in B cell tolerance checkpoints can skew the developing repertoire (42). Such defects are known to exist in both AChR and MuSK MG (60), and thus may contribute to enrichment of the V genes containing N-linked glycosylation motifs, which we observed in some patients with MuSK MG. However, the accumulation of additional motifs through SHM suggests that, in the MuSK disease subtype, Ag-driven positive selection also plays a role in the conspicuously elevated IgG-V^N-Glyc frequency. It remains possible that this selection is an Ag-independent process. Examples of this mechanism include interactions between glycosylated BCRs and lectins (20), which are thought to drive proliferation in B cell malignancies and some autoimmune diseases (19).

The Ag binding of the four human mAbs that we studied was not disturbed by N-linked glycosylation. Although a limited number of mAbs were available for study, these results align well with data derived from other human MuSK-binding mAbs, some of which include glycosylation motifs and others that do not, indicating that the glycan is not essential for binding (39). These collective results indicate that selection of the IgG-V^N-Glyc in human MuSK autoantibodies may not have been driven by MG-specific self-antigen positive selection. This is somewhat unexpected given that the IgG-V^N-Glyc sites could be found in regions responsible for Ag contact (CDRs). However, it remains possible that the positions of the IgG-V^N-Glyc in the human MuSK-binding mAbs we studied were not essential for Ag–autoantibody contact. Structural studies of MuSK–mAb complexes would be required to test this possibility. Similarly, the V regions of ACPAs from patients with RA are consistently glycosylated, but their binding is not influenced by the modification (17). However, other investigations suggest that binding can be modulated as a consequence of their presence (15, 16, 21). These findings suggest that an autoantigen-independent selection mechanism may influence the IgG-V^N-Glyc motif frequency in the autoimmune repertoire in some, but not all, autoimmune diseases.

Our proteomic analysis of the serum-derived total IgG from two of the study subjects (one from each of the AChR and MuSK MG cohorts) showed a higher m.w. band in the electrophoresis studies. These results indicate that a large fraction of circulating IgG included V region N-linked glycosylation. The higher m.w. bands are not attributable to only the AChR- or MuSK-specific IgG, given that these autoantibodies are present at very low concentrations in the circulation. Although a number of the patients with MG were observed to have elevated IgG-V^N-Glyc motif frequency in the B cell repertoire sequencing data, the serum-derived IgG from some of the same patients did not include a higher m.w. band in the proteomic analysis. Moreover, the intensity of the higher m.w. bands in some samples suggests that the fraction of IgG-V^N-Glyc is higher than the B cell sequencing data indicate. These results align well with empirically derived models that describe a lack of clonal overlap between the serum IgG and circulating B cell repertoire, highlighting the discordance between serum Ab repertoire (humoral immunity) and the VH gene repertoire (61–63). Rather, these findings may reflect that much of the circulating IgG is derived from long-lived plasma cells residing in the bone marrow.

One of two possible glycosylation sites in an autoantibody known to be specific for AChR, mAb 637, was shown to be unoccupied. We speculate that this is indicative of context-specific N-glycosylation (local amino acid sequence containing the motif or cellular environment) or inherent selectivity for one site over the other possibly as a result of conformation or solvent accessibility. Nevertheless, this site did not appear to contribute to binding activity, similar to our observations obtained by testing the MuSK mAbs. We recognize, as a study limitation, that the in vitro expression of these mAbs may not emulate the glycosylation occupancy in vivo. We did use a mammalian expression system (human embryonic kidney cells) to achieve the best approximation of the in vivo status, and we experimentally confirmed occupancy for the Ag-binding studies. It remains to be investigated whether variables, such as the stage of B cell activation or tissue residence, could alter the occupancy.

Finally, a consensus on the function of IgG-V^N-Glyc in health or disease is unclear. Several possibilities have been described, including perturbation of Ab–antigen interactions (binding affinity, specificity), altered metabolism of B cells or IgG in vivo (half-life, clearance), mislocalization of IgG to host tissue, redemption of autoreactive B cells, and inappropriate selection/expansion of autoreactive B cells in germinal centers (1, 2, 64). Elimination of these motifs or removal of the glycosyl moiety itself have been observed to impair Ag binding, such as in anti-adalimumab/infliximab Abs derived from patients treated for RA (21). However, other studies have suggested a more nuanced picture; a study of anti–cyclic citrullinated protein autoantibodies showed no contribution of N-linked glycans to binding (17). In this article, we present unequivocal evidence that the presence of N-linked glycans is not required for binding in the case of four MG autoantibodies. This finding is in agreement with a number of studies focusing on human autoantibodies, which demonstrated that IgG-V^N-Glyc can alter binding or activity but is not required for these properties (15, 21, 65).

Our study is not without limitations. MG is a heterogeneous disease; a number of AChR MG subtypes are well documented. Accordingly, we recognize the limited number of patients included in this investigation and thus encourage caution in generalizing our findings. Furthermore, we acknowledge that the investigation of several human mAbs is unlikely to precisely represent the circulating MG autoantibody repertoire. Thus, we cannot conclude that AChR or MuSK autoantibodies are universally glycosylated. Other investigations (13, 15) of human autoimmune disease, which showed the conspicuous presence of IgG-V^N-Glyc by electrophoresis, specifically focused on Ag-enriched autoantibodies rather than the BCR repertoire and total circulating IgG, which was the focus of our study. Although we studied a small set of specific autoantibodies in the form of human-derived mAbs, we recognize that the mAbs serve only to represent a fraction of the total circulating autoantibody repertoire. Although challenging, it will be important to isolate serum-derived MuSK- and AChR-specific autoantibodies to investigate both the frequency and the properties of the V region glycans in this broad population of circulating MG autoantibodies.

In summary, IgG-V^N-Glyc of the total circulating IgG and bulk BCR repertoire is elevated in a subset of AChR and MuSK MG patients. These findings are consistent with those of a previous study (56) that showed elevated (albeit not statistically significant) V region N-glycosylation sites in the BCR repertoire of MG patients compared with healthy control subjects. Our findings extend this molecular phenotype beyond RA, pSS, SLE, and ANCA-associated vasculitis. IgG-V^N-Glyc did not affect AChR or MuSK autoantibody binding in the limited number of human mAbs we tested. We speculate an elevation in N-linked glycosylation motifs containing V gene sequences may be driven by the presence of dysregulated germinal centers that contribute to B cell selection defects observed in the disease. Our findings contribute to efforts to understand the basic biology of IgG-V^N-Glyc and its association with disease.

We thank Karen Boss for expert copyediting and proofreading, Dr. Bailey Munro-Sheldon for verifying autoantibody titers, and Charlotte A. Gurley for assisting with manuscript and reference formatting.

K.C.O. has received research support from Ra Pharma and is a consultant and equity shareholder of Cabaletta Bio. K.C.O. is the recipient of a sponsored research subaward from the University of Pennsylvania, the primary financial sponsor of which is Cabaletta Bio. K.C.O. has received speaking and advising fees from Alexion and Roche. M.L.F. has received research support from Grifols. R.J.N. has received research support from Genentech, Alexion Pharmaceuticals, argenx, Annexon Biosciences, Ra Pharmaceuticals, Momenta, Immunovant, and Grifols. R.J.N. has served as consultant/advisor for Alexion Pharmaceuticals, argenx, CSL Behring, Grifols, Ra Pharmaceuticals, Immunovant, Momenta Cabaletta Bio, and Viela Bio. S.H.K. receives consulting fees from Northrop Grumman. M.R.W. has received research support from Roche/Genentech. K.B.H. receives consulting fees from Prellis Biologics. The other authors have no financial conflicts of interest.