IgM N-glycosylation correlates with COVID-19 severity and rate of complement deposition

The glycosylation of IgG plays a critical role during human SARS-CoV-2, activating immune cells and inducing cytokine production. However, the role of IgM N-glycosylation has not been studied during acute viral infection in humans. In vitro evidence suggests that the glycosylation of IgM inhibits T cell proliferation and alters complement activation rates. The analysis of IgM N-glycosylation from healthy controls and hospitalized COVID-19 patients reveals that mannosylation and sialyation levels associate with COVID-19 severity. Specifically, we find increased di- and tri-sialylated glycans and altered mannose glycans in total serum IgM in severe COVID-19 patients when compared to moderate COVID-19 patients. This is in direct contrast with the decrease of sialic acid found on the serum IgG from the same cohorts. Moreover, the degree of mannosylation and sialylation correlated significantly with markers of disease severity: D-dimer, BUN, creatinine, potassium, and early anti-COVID-19 amounts of IgG, IgA, and IgM. Further, IL-16 and IL-18 cytokines showed similar trends with the amount of mannose and sialic acid present on IgM, implicating these cytokines’ potential to impact glycosyltransferase expression during IgM production. When examining PBMC mRNA transcripts, we observe a decrease in the expression of Golgi mannosidases that correlates with the overall reduction in mannose processing we detect in the IgM N-glycosylation profile. Importantly, we found that IgM contains alpha-2,3 linked sialic acids in addition to the previously reported alpha-2,6 linkage. We also report that antigen-specific IgM antibody-dependent complement deposition is elevated in severe COVID-19 patients. Taken together, this work links the immunoglobulin M N-glycosylation with COVID-19 severity and highlights the need to understand the connection between IgM glycosylation and downstream immune function during human disease.


Introduction
SARS-CoV-2 (COVID-19) has impacted the world signi cantly since its outbreak in late 2019, killing more than 14 million between 2020-21 [1]. Once viral particles are inhaled and enter the human airway, the spike (S) protein trimer expressed on the surface of SARS-CoV-2 membranes binds and infects cells via the angiotensin-converting enzyme 2 (ACE2) abundant in airway epithelial and endothelial cells [2]. The resulting infection consists of two overlapping phases. The rst mainly consists of viral replication associated with mild constitutional symptoms. During the second phase, a combination of the host's adaptive and innate immune response can result in either the e cient clearance of virus-infected cells or the induction of multi-organ system damage requiring intensive care [3]. Patients in this second phase with severe COVID-19 often present with elevated D-dimer [4], C-reactive protein (CRP) [5], IL-6 [6], acute kidney injury [7], and heightened complement deposition [8,9].
Immunophenotyping assessment in a COVID-19 cohort (IMPACC) was designed at the beginning of the pandemic with the intent to enroll hospitalized patients with COVID-19 to collect detailed clinical, laboratory and radiography data with the intent of turning this into a prospective longitudinal study [10].
Biological samples including blood, nasal swabs, and endotracheal aspirates were collected at multiple time points during hospitalization. Five trajectory time points were identi ed previously based on clinical data from the entire IMPACC cohort. Patient trajectories were divided into 5 groups based on longitudinal While the N-glycosylation of IgM has been characterized previously in healthy pooled human serum, during cancer [35,[43][44][45], and in recombinant IgM [34,46], this is the rst characterization of the IgM Nglycosylation pro le isolated from humans infected with an acute viral disease. Here, we report signi cant differences in the IgM N-glycan content from cohorts of hospitalized COVID-19 separated by severity trajectory. Total mannosylation decreased while di-sialylation (S2) increased on IgM -opposing the trend detected in the same cohorts of reduced IgG sialylation. Moreover, glycosylation of IgM correlates with circulating immune cell glycosyltransferase expression of ST3GAL4 and MAN1A2, previously reported clinical markers of COVID-19 severity, and elevations in cytokines IL-16 and IL-18.
Lastly, we report an increased antibody-dependent complement deposition induced by IgM from the severe COVID-19 cohort.

Human samples Patient enrollment and consent
The IMPACC is a collaborative project developed by the NIAID and investigators from the Human Immunology Project Consortium (HIPC), the Asthma and Allergic Diseases, and the Cooperative Disease Research Centers (AADCRC). Drexel University collected 106 patient samples to be included in the IMPACC through the Tower Health Hospital network. Participants are enrolled within 48 hours of hospitalization where demographics, detailed medical history, and clinical data were taken. Consenting participants are enrolled within 48 hours of hospitalization under the IRB Protocols 2004007753 and 2102008337. Upon enrollment, demographics, COVID-19 symptom onset, detailed medical history (including comorbidities), and medical records were all recorded. Patients were con rmed positive with a SARS-CoV-2 polymerase chain reaction (PCR). Extensive clinical labs are taken during intake-and biological samples including blood, nasal swab, and endotracheal aspirates are collected. Clinical data and samples from days 4 and 7, representing patient admission to the hospital, were examined.

Biological Sample Processing
Blood samples and nasal swabs were collected at each timepoint and processed at Drexel University within 6 hours of collection according to the IMPACC standardized operating procedure [10]. Whole blood, nasal swabs, peripheral blood mononuclear cells (PMBCs), and plasma collected from each patient was processed at Drexel University and sent to IMPACC core facility sites for further analysis as previously reported [10,11]. PBMCs were used to identify immune cell populations and changes in cell populations, gene expression, and activation markers. Plasma was used to characterize antibody titers, anti-RBD titers, antibody isotype, proteomics, and metabolomics. At Drexel, plasma was additionally used for ELISA antibody abundance analysis, Luminex cytokine and chemokine assays, and glycomic analysis. Whole blood was used in genome-wide association study (GWAS) and cytometry by the time-of-ight (CyTOF) and bulk RNA transcriptomics. Nasal Swabs were used for bulk RNAseq and viral load quantitaion. PBMC Isolation: Patient blood samples were spun down at 1000 x g for 10 minutes at room temperature, and plasma was aliquoted. The remaining blood was diluted 1:2 with DPBS (Ca + 2 Mg + 2 free) and slowly pipetted into a 50mL SepMate-50 tube (with 15mL Lymphoprep below the insert). Samples were spun at 800 x g for 20 minutes at 20ºC with brakes off. The top layer with PBMCs was transferred to a new tube and cells were washed at 400 x g for 5 minutes. Cells were resuspended in 20mL EasySep Buffer, then spun again at 300 x g for 10 minutes at room temperature. For RNASeq, cells were resuspended at 5 million per mL, and 50uL was aliquoted into CRYSTAL Gen tubes. Cells were spun at 500 x g for 5 minutes at room temperature and the excess media was removed. 200uL QIAGEN RLT Buffer with (BME) was added and vortex until the pellet was fully dissolved. Samples were stored at -80ºC for shipment. The remaining PBMCs were frozen down in FBS + DMSO for storage at Drexel University.
Anti-SARS-CoV-2 nucleocapsid IgA, IgG, and IgM quantitation Monobind AccuBind® ELISA Anti-SARS-CoV-2 kits were used as a qualitative determination of Anti-SARS-CoV-2 speci c IgA, IgG and IgM antibodies at Drexel's IMPACC site. These kits utilize a sequential sandwich ELISA method. This test utilizes recombinant nucleocapsid protein (rNCP) from SARS-CoV-2 coated on microwells to capture antibodies in human plasma. Patient plasma was diluted 1:100 and added directly to the ELISA plate. Following incubation and washing, IgA, IgG or IgM labeled antibodies were added. After a second incubation and wash, reagent substrate is added to produce a measurable color through the reaction with enzyme and hydrogen peroxide. After the addition of a stop substrate, absorbance was read in each well at 450nm within 15 minutes of adding the stop solution.
RealTime Quantitative Polymerase Chain Reaction: Master mixes containing nuclease-free water, combined primer/probe mixes, and One-Step RT675 qPCR ToughMix (Quantabio) were prepared on ice, and 15 µL was dispensed in each well of a 384-reaction plate (Thermo sher) CoV2 was quantitated using the CDC qRT-PCR assay (primers and probes from IDT). Brie y, this comprises two reactions targeting the CoV2 nucleocapsid gene (N1 and N2) and one reaction targeting RPP30 (RP). Each batch included positive controls of plasmids containing N1/N2 and RP target sequence Plates were centrifuged for 30 seconds at 500 x g, 4C. The quantitative polymerase chain reaction was performed using a Quantstudio5 (Thermo Fisher) with cycling conditions: 1 cycle 10 min at 50°C, followed by 689 3 min at 95°C, 45 cycles 3 sec at 95°C, followed by 30 sec at 55.0°C.
RNA-sequencing cDNA Library Production: From each nasal RNA sample, 10ul was aliquoted to a library construction plate using the Perkin 692 Elmer Janus Workstation (Perkin Elmer, Janus II). Ribosomal depletion, cDNA synthesis, and library construction steps were performed using the Total Stranded RNA Prep with Ribo-Zero Plus kit, following the manufacturer's instructions (Illumina). All steps were automated on the Perkin Elmer Sciclone NGSx Workstation to reduce batch-to-batch variability and increase sample throughput. Final cDNA libraries were quanti ed using the Quant-it dsDNA High Sensitivity assay, and library insert size distribution was checked using a fragment analyzer (Advanced Analytical; kit ID DNF474). Samples, where adapter dimers constituted more than 4% of the electropherogram area, were failed before sequencing. Technical controls (K562, Thermo Fisher Scienti c, cat# AM7832) were compared to expected results to ensure that batch-to-batch variability was minimized. Successful libraries were normalized to 10nM for sequencing.

RNA-sequencing Clustering and Sequencing
Barcoded libraries were pooled using liquid handling robotics prior to loading. Massively parallel sequencing-by-synthesis with uorescently labeled, reversibly terminating nucleotides was carried out on the NovaSeq 6000 sequencer using S4 owcells with a target depth of 50 million 100 base-pair pairedend reads per sample (25 million read pairs).

Total IgG isolation
Total IgG was isolated from 20µL of plasma using a Protein G spin plate as described by the manufacturer (ThermoFisher, MA). Four 200µL 1X PBS washes removed unbound plasma protein using a vacuum manifold apparatus. Next, IgG was eluted by incubating 150µL of 0.1M glycine HCl pH 2-3 for 5 minutes at room temperature. The eluate was collected into a 96-well 2mL collection plate pre-loaded with 15µL of 1.5M Tris pH 8 to neutralize the glycine elution buffer. The wash process was repeated a second time to ensure a high yield of IgG. The resulting 315µL of the neutralized eluate was concentrated and buffer-exchanged to 20µL of 1X PBS using Amicron Ultra-0.5 centrifugal Filter 10 kDa MWCO (Millipore) following the manufacturer's instructions. NanoDrop 1000 spectrophotometer readings monitored protein yield through the isolation process.

Total IgM isolation
Total IgM was isolated from plasma by incubating 80µL of goat anti-IgM agarose-conjugated agarose beads (A9935, Millipore Sigma, MA) with 80µL plasma and 100µL 1X PBS for 2 hours at room temperature. Following the incubation, the solution was transferred to a 1.2um MultiScreen HTS 96-well lter plate. Four 200µL 1X PBS washes removed unbound plasma protein using a vacuum manifold apparatus. Next, IgM was eluted by incubating 150µL of 0.1M glycine HCl pH 2-3 for 5 minutes at room temperature. The eluate was collected into a 96-wel 2mL collection plate pre-loaded with 15µL of 1.5M Tris pH 8 to neutralize the glycine elution buffer. The wash process was repeated a second time to ensure a high yield of IgM. The resulting 315µL of the neutralized eluate was concentrated and buffer-exchanged to 20µL of 1X PBS using Amicron Ultra-0.5 centrifugal Filter 10 kDa MWCO (Millipore) following the manufacturer's instructions. NanoDrop 1000 spectrophotometer readings monitored protein yield through the isolation process Immunoglobulin N-glycan analysis N-glycans from IgG and IgM were released, labeled, and analyzed as described previously using the Waters GlycoWorks RapiFluor MS kit, adapted for PCR tubes [47]. Brie y, samples were denatured using the RapiGest reagent for 5 minutes at 95°C using a PCR thermocycler. Next, glycoprotein samples were deglycosylated using PNGase F for 6 minutes at 60°C using a PCR thermocycler. Afterward, samples were labeled with RapiFluor label (RFMS) for 5 minutes at room temperature. A solid-phase extraction (SPE) clean-up module isolated RFMS labeled N-glycans which were then eluted into a 96-well 2mL Waters ANSI plate capped with a PFTE 96-well membrane top for high-throughput N-glycan analysis. An ACQUITY Premier UPLC System was used following the setting and protocol described previously [47]. Brie y, a ACQUITY UPLC BEH Amide Column, 130Å, 1.7 µm, 2.1 mm X 50 mm column (Waters, MA) was used to chromatographically separate N-glycans during the 18.3 min run employing a gradient of 50mM Ammonium Formate pH 4.4 (Waters) made with LC-MS Water (Millipore), LC-MS ACN (VWR, Honeywell) 25%-75% gradient transitioning over 12 min to 60%-40%. N-glycans separated by charge and stereochemistry were quantitated using Waters AQUITY Fluorescent detector set to 265/425 em/ex, 10Hz using Empower 3 software. Lastly, N-glycan identity was con rmed using a Waters AQUITY QDa Mass spectrometer. The resulting UPLC uorescent trace was analyzed with Empower v3.3.1 software, UPLC trace percent-area was combined with collected MS-spectra to identify eluted peaks as described previously [47]. Pooled N-glycans labeled with the RapiFluor tag were digested with Neuraminidase S (New England BioLabs, MA, P0743L) or Neuraminidase (New England BioLabs, MA, P0720S) for 12 hours at 32°C following the manufacturer's instructions. Digested N-glycans were cleaned up using Water's SPE kit and analyzed using the UPLC detailed above.
Antigen-speci c complement deposition assay Antibody-speci c complement deposition against the RBD and Spike S1 antigens were assayed following the previously developed protocol [48]. Brie y, 20µL FluoSpheres™ NeutrAvidin™-Labeled Microspheres (ThermoFisher) were incubated with 20µg RBD (aa319-541, Invitrogen) (biotinylated in-house using the EZ-Link™ Sulfo-NHS-LC-Biotinylation Kit) or 20µg biotinylated SARS-CoV-2 (2019-nCoV) Spike S1-His Recombinant Protein, Biotinylated (SinoBiological) antigen for 4 hours at 37°C. After washing twice with 200µL 1X PBS, the antigen-bound beads were blocked with 200µL 5% BSA in 1X PBS for 1 hour at 37°C. Next, the beads were washed twice with 500µL of 0.1% BSA in 1X PBS and diluted 1:100 in 1X PBS. A subset of plasma and puri ed IgM samples were treated with either a Mannosidase (New England BioLabs, MA, P0768S) or Neuraminidase (New England BioLabs, MA, P0720S) for 12 hours at 32°C prior to antigen-speci c complement deposition analysis following the manufacturer's instructions. Next, 15µL of the 1:100 bead solution was transferred to low-binding 1.5mL tubes (Corning) and incubated with 20µL of 1:10 1X PBS diluted pooled severe or nonsevere plasma or 5µg of IgM isolated from pooled severe or nonsevere plasma for 2 hours at 37°C. Next, the immune-complexed beads were incubated for 15 minutes with freshly resuspended Guinea pig complement (Cedarlane, CL4051) and diluted 1:50 in Gelatin Veronal Buffer with Mg2+ & Ca2+ (GVB++) at 37°C. The complement deposition was halted with two washes of 200µL 15mM EDTA. Next, 50µL of a 1:100 diluted FITC labeled Goat anti-Guinea pig Complement C3 antibody (MP Biomedicals, 085538) was incubated for 30 minutes with the immunecomplexed beads. Lastly, two 200µL 1X PBS washes removed unbound FITC labeled anti-C3 antibody.
Washed samples were re-suspended in 100uL and analyzed using a Fortessa Flow Cytometer (BD). Beads were gated for the presence or absence of the FITC antibody, and the MFI of the bead content was divided by the total number of beads to determine the rate of complement deposition in each sample. The gating strategy is displayed in Fig. 4B. Flow Minus One (FMO) control samples were run with the same protocol to con rm a low background signal and inform the gating cut-off strategy.

Statistical analysis
A biomarker was removed from analysis if its overall number of missing values was greater than 3 (13.6% of 22 patients) to reduce potential bias [49][50][51]. Data analysis was performed using R and GraphPad Prism 8. COVID-19 trajectory groups were categorized as "1-3" and "4-5" for the averages of measured transcriptomic, proteomic, Luminex, and clinical data. Gender and COVID-19 trajectory group categories were summarized as counts and percentages, continuous variables were summarized as the median and interquartile range (IQR) overall and by trajectory group category. For transcriptomic data, raw counts were normalized to counts per million (CPM), then values were log2 transformed for statistical analysis. A pseudo-count of 2 was added to all count data prior to log transformation because zero cannot be 'logged' [52][53][54]. Mann-Whitney U test was used to test the signi cance of continuous variables between trajectory group categories. A chi-square test was used to test the association between gender and trajectory group category. Associations between IgM Mannosylated or total S2 and other variables were tested using simple linear regression. Raw trajectory group values were used in simple linear regression. Coe cient of determination R2 was obtained from linear regression. p < 0.05 was considered statistically signi cant for all tests.

Results
IgM di-sialylation and mannosylation associate with COVID-19 severity Plasma from patients admitted to the hospital after testing positive for COVID-19 was analyzed 4-and 7days post-admission. Clinical characteristics of the patients are presented in Table 1 strati ed by trajectory 1-5, with 1 being a mild COVID-19 infection and 5 being death from complications of COVID-19 infection. N-glycan pro les isolated from puri ed total IgM were analyzed (Fig. 1A), with N-glycan identities listed in Supplemental Table 1. N-glycans ranging from mono-antennary to tri-antennary as well as hybrid and mannosylated moieties were observed in all IgM samples. The 36 individual IgM N-glycan peaks with identities con rmed by mass-spectrometry from day 4 and day 7 are included in Supplemental Figs. 1 and 2. To analyze general trends in the IgM N-glycan pro le across disease severity, glycans were grouped by size, charge, and type into classes (G0, G1, G2, S1, ect.) as denoted below the IgM N-glycan pro le in Fig. 1A.
Protein glycosylation is impacted by factors including sex, age, and BMI [55][56][57][58][59][60][61][62][63][64][65][66]. Therefore, COVID-19 patient cohorts from the IMPACC study were analyzed to determine if there were statistically signi cant differences between mild (trajectories 1 and 2), moderate (trajectory 3), and severe (trajectories 4 and 5) (Fig. 1B). There was no statistically signi cant difference between cohorts based on sex, age, BMI, the number of days of COVID-19 symptoms prior to hospitalization, or viral load. Furthermore, we determined there was no statistically signi cant difference in the concentration of total IgM isolated between each patient cohort (Supplemental Fig. 3). After con rming that cohort characteristics were comparable, we analyzed the IgM N-glycosylation pro les from day 4 and 7 hospitalized COVID-19 IMPACC patients across mild, moderate, and severe cohorts (Fig. 1C). Di-sialylated (S2) N-glycans on IgM increased signi cantly in the severe COVID-19 cohort on day 4 of hospitalization compared to the mild and moderate cohorts. In addition, total mannose, including hybrid N-glycans, decreased signi cantly in the severe COVID-19 cohort on day 4 IgM. On day 7, the severe cohort's IgM N-glycosylation maintained the trends observed on day 4, but lost signi cance likely due to the death of four of the COVID-19 patients in the severe trajectories reducing the power of the analysis. Taken together, the changes in IgM Nglycosylation correlate with the severity of COVID-19 infection in humans.
IgG and IgM N-glycans responses differ during COVID-19 We next compared the glycosylation of bulk IgM and IgG isolated from COVID-19 patients to characterize the plasma blast glycosylation response to viral infection. Patients were sorted into nonsevere (trajectories 1-3) and severe (trajectories 4 and 5) cohorts to compare the change in immunoglobulin Nglycosylation by glycan class. First, IgG N-glycans from healthy control, nonsevere, and severe COVID-19 cohorts were analyzed as grouped classes (G0, G1, G2 ect.) as described in Supplemental Fig. 4. IgG in both severe and nonsevere COVID-19 exhibited reduced di-galactosylation (G2) and mono-sialylation (S1) while agalactosylation (G0) signi cantly increased compared to healthy controls in the severe COVID-19 cohort (Fig. 2A). Interestingly, the IgG N-glycosylation of the severe and nonsevere cohorts did not exhibit statistically signi cant differences between one another. In contrast, the IgM glycosylation from the same patients revealed statistically signi cant changes between severe and nonsevere cohorts (Fig. 2B). Agalactosylated (G0) and mono-galactosylated (G1) N-glycans signi cantly decreased in severe patients compared to the nonsevere cohort. Further, the increase in S2 remained signi cant while tri-sialylated (S3) content also increased signi cantly in the severe COVID-19 cohort. In comparison, the sialyation of severe patient IgG N-glycans remained lowered or unchanged on day 4 compared to healthy controls ( Fig. 2A), aligning with previous studies of IgG N-glycosylation in hospitalized COVID-19 patients [24,27,28]. Lastly, the decrease in mannose remained signi cant in severe trajectory patients compared to nonsevere patients on day 4 of hospitalization.
The decrease in total mannose content required further interrogation because 11 hybrid and mannosylated N-glycans contribute to the overall decrease observed in the IgM during severe COVID-19 (Fig. 2C). The decrease in total mannose was predominantly due to lowered levels of the smaller hybrid moieties: M4G1, FM4A1, and M5A1 in combination with the mannosylated moieties: M5 and the two isoforms of M6. Mannosylated structures or co-eluting peaks larger than M6 did not signi cantly decrease, while M9 signi cantly increased in the severe COVID-19 cohort. Next, mannose and hybrid structures ranging from M4-M6 were compared to mannose structures M7-M10, revealing a potential reduction in the degree of mannose processing by Golgi-bound mannosidases during IgM production. Taken together, the glycosylation pattern of IgM was consistently altered in the severe COVID-19 cohort, with major classes of IgM N-glycans trending in opposite directions compared to the IgG N-glycan classes.

Glycosyltransferase expression correlates with IgM N-glycosylation
The observed changes in IgM N-glycosylation likely result from glycosyltransferase expression within the Golgi of plasmablasts. The IMPACC study collaborators at Emory University provided 61 glycosyltransferase and glycosidase transcript expression data isolated from peripheral blood mononuclear cells (PBMCs) collected on day 0 of patient hospitalization. After normalizing the data by total read count and transforming by log2 for comparability, expression pro les were compared between the severe and nonsevere COVID-19 cohorts.
The expression of the mannosidases MAN1A2 and MAN2A1 decreased signi cantly in the severe cohort compared to the nonsevere cohort (Fig. 3A). These mannosidases are responsible for processing high mannose structures into smaller mannose moieties [67]. The decrease in mannosidase expression aligns with data in Fig. 2C where we observe less mannosidase-processed M5 and M6 content in the severe COVID-19 cohort IgM. In addition, IgM total mannose correlated with MAN1A2, the o-mannosyltransferase TMTC2, and the α-2,3 sialyltransferase ST3GAL4 (Fig. 3B).
The expression of the α-2,3 sialyltransferase ST3GAL4 and the O-glycan α-2,6 sialyltransferase ST6GALNAC2 were signi cantly elevated in the severe COVID-19 cohort (Fig. 3A). Interestingly, the ST6GAL1 did not signi cantly differ between COVID-19 severity suggesting that a portion of the increased sialylation on IgM is due to the α-2,3 sialyltransferase ST3GAL4 (Supplemental Table 4). When IgM N-glycans were digested with the exoglycosidase Neuraminidase S, speci cally cleaving α-2,3-linked sialic acids, we detect a signi cant reduction in the A3G3S3 glycan species and a concomitant increase in the A3G3S2 abundance (Supplemental Fig. 5). Because ST6GALNAC2 adds an α-2,3 linked sialic acid to the O-glycans expressed on leukocyte cell surfaces, it is unlikely to add sialic acid to IgM [68]. However, the increased ST6GALNAC2 expression in the severe COVID-19 cohort PBMCs may re ect a reduced propensity for leukocytes to migrate into tissues due to sialic acid blocking P-/L-selectin ligand a nity [69]. Lastly, we report that a summation of all the sialic acids (S1, S2, and S3) from IgM positively correlated with the expression of ST3GAL4 (Fig. 3B). This nding suggests a potential role for ST3GAL4 adding sialic acid to IgM, but future studies will need to con rm this phenomenon speci cally in plasma blast transcriptomic studies. All in all, the PBMC transcriptomic data aligned with our observations of IgM glycosylation alterations within the severe COVID-19 cohort.
Clinical markers of disease severity correlate with IgM glycosylation Next, we sought to determine if the changes in IgM N-glycosylation were associated with clinical laboratory data and additional cytokine panels collected by Drexel's IMPACC study [10,11]. After omitting clinical parameters with less than 90% complete datasets [70], the remaining data were analyzed for correlations to IgM total mannose and S2 content using a linear regression model (Supplemental Tables 2 and 3). The reduction of IgM mannose in severe COVID-19 patients negatively correlated with increased D-dimer, blood urea nitrogen (BUN), creatinine, and potassium (K+) (Fig. 4A). In addition, the increased IgM S2 content positively correlated with the same clinical measurements -except for a nonsigni cant correlation with potassium, p = 0.186 (Fig. 4B).
The severity of COVID-19 has also been associated with higher anti-SARS-CoV-2 IgG and IgA antibody abundance at the time of hospital admission [71]. Therefore, we sought to correlate IgM mannose and S2 glycosylation with the relative abundance of anti-SARS-CoV-2 nucleocapsid (anti-N) IgA, IgM, and IgG.
Anti-N IgA relative abundance negatively correlated with IgM mannose content, while the increase in IgM S2 content positively correlated with anti-N titers of IgA, IgM, and IgG relative abundance (Fig. 4C).
Lastly, we examined Luminex data from a 32-plex cytokine panel to determine if circulating cytokines were associated with the glycosylation changes observed on IgM. Interestingly, cytokines previously demonstrated to alter glycosyltransferase activity such as IFN-γ, TNF-α, IL-6, IL-17A, or IL-10 [72, 73] did not signi cantly correlate with either IgM mannose or S2 content (Supplemental Tables 2 and 3).
Moreover, only the cytokine IL-18 was signi cantly higher between the nonsevere and severe hospitalized COVID-19 cohorts (Fig. 4D). While not statistically signi cant, IL-18 and IL-16 correlated positively with IgM S2 (p = 0.099 and p = 0.0538 respectively). IgM mannose content correlated negatively with IL-18 and IL-16 (p = 0.057 and p = 0.059 respectively). Taken together, the IgM glycosylation pro le closely correlates with COVID-19 severity on clinical and serological parameters and the cytokines IL-16 and IL-18 may play a role in controlling downstream glycosyltransferase expression in plasma blasts during COVID-

19.
Antibody-dependent complement deposition is increased in severe COVID-19 patients After examining clinical factors and cytokines associated with IgM N-glycosylation changes, we sought to interrogate the differences in complement deposition rates initiated by SARS-CoV-2 circulating plasma antibodies in general, and IgM speci cally. We adapted an antibody-dependent complement deposition (ADCD) assay employing uorescent beads conjugated to a biotinylated antigen to compare complement deposition rates with SARS-CoV-2 antigens: receptor binding domain (RBD), and Spike S1 (Fig. 5A) [48].
After incubating either diluted plasma or puri ed IgM with antigen-coated beads, deposition of guinea pig complement was detected using ow cytometry (Fig. 5B). RBD induced low ADCD in diluted nonsevere and severe plasma, aligning with previously reported ADCD trends [74,75] (Fig. 5C). Further, puri ed IgM did not induce complement deposition above the PBS background control (dotted line). However, the spike S1 antigen induced signi cantly higher ADCD in both plasma and puri ed IgM assay cohorts (Fig. 5D). Plasma from severe patients deposited higher levels of complement compared to nonsevere COVID-19 plasma, but not to a signi cant degree. However, IgM from the severe COVID-19 cohort induced signi cantly higher levels of complement deposition compared to the nonsevere cohort IgM. Next, plasma and IgM samples were digested with a mannosidase (M) or sialidase (S) and assayed for ADCD (Fig. 5E). Mannosidase treatment signi cantly reduced the deposition of complement on Spike S1 antigen in both plasma and IgM samples. However, the treatment with a sialidase only reduced the deposition of complement in the severe cohort IgM. Taken together, we report that severe COVID-19 cohort IgM induces higher levels of antigen-speci c complement -which could be related to the alteration in the glycosylation of its mannose or sialic acid content.

Discussion
IgG N-glycosylation and effector function have been well characterized during acute COVID-19 infection [17][18][19] [24][25][26][27][28]. However, IgM antibodies also play vital roles during immune responses, promote a nity maturation, maintain hemostasis at mucosal sites including the gut and lung, and induce signi cantly higher levels of complement deposition compared to IgG [76]. We suggest that IgM has been overlooked as a key player during the acute COVID-19 immune response. Within the IMPACC cohort enrolled at Drexel University, we nd host IgM N-glycosylation correlates with disease severity.

IgM N-glycosylation
We report a signi cant decrease in total IgM mannose in patients with severe COVID-19 (trajectory 4 and 5) compared to those with nonsevere COVID-19 (trajectories 1-3). By examining the mannose and hybrid structures contributing to this decrease, we conclude IgM contains fewer (M4-M6) mannose structures during severe COVID-19. Instead, IgM in severe COVID-19 contains larger mannose structures. These conclusions are supported by decreased mannosidase MAN2A1 and MAN1A2 expression within patient PBMC mRNA glycosyltransferase (GT) expression datasets. Previously, MAN1A2 genetic variability was identi ed as a potential correlate with susceptibility to COVID-19 infection [77]. During severe in uenza, MAN1A2 was also downregulated and predicted to be highly regulated by miRNA [78,79]. More work into the regulation of mannosidase expression is required to con rm if the changes observed in PBMC mRNA are maintained within plasma blast cell population mRNA expression, or if SARS-CoV-2 infection of PBMCs is the main factor inducing these changes in GT expression. Furthermore, the majority of IgM mannosylation is site-speci c, populating the C-terminus of IgM on the Asn-563 and Asn-402 amino acids [80]. These two glycosylation sites are positioned to potentially interact with the C1q component, and impact complement activation rates in mice and humans [81,82]. When IgM binds to an antigen target, it converts from a "planar" to a hexagonal "dome" or "staple" con guration [83]. Based on in situ cryostructures, the antigen-bound IgM con rmation binds to complement C1q close to where the IgM Cterminus mannose structures are present. More work is required to determine if human IgM mannosylation impacts the a nity of C1q binding due to steric hindrance or by interacting with the mannose-binding lectin or H-colin [84].
Because no studies of human IgM N-glycosylation during human viral infections have been completed, we sought to compare reports of IgM N-glycan pro les characterized during other human disease states.
In ovarian cancer patients, the IgM N-glycans M7 and M8 decreased on glycosite N439 (Asn563) with concomitant increases of mono-and di-sialylated N-glycans occupying N209 (Asn332) [43]. However, the group determined that the IgG and IgA N-glycan pro les predicted patients with ovarian cancer with higher accuracy than IgM N-glycans. In contrast, we observe signi cant differences in IgM Nglycosylation stratifying COVID-19 disease severity with decreases only in the processed M5 and M6 Nglycans. Another group reported higher levels of sialic acid detected in IgM protein fraction isolated from cancer patient sera compared to non-cancer patients [45] while another study reported no signi cant glycomic response associated with either IgG or IgM N-glycan pro les following tumor ablation therapy [44]. Our ndings expand upon the previous reports of a general increase in sialic acid content on IgM. These sialylated IgM N-glycans likely populate the Asn-395, Asn-332, and Asn-171 glycosylation sites and could play roles in immunomodulatory signaling. When we examined the PMBC sialyltransferase mRNA expression data, we did not observe signi cant changes in the ST6GAL1 mRNA levels. However, we did detect increased ST3GAL4 mRNA expression, which positively correlated with the summation of all sialic acid content on IgM.
A previous high-throughput glycomic analysis of COVID-19 patients identi ed increased α-2,6 and α-2,3 sialylation in the total plasma, lung, and liver tissues [85]. The group associated this change with the increase in α-2,6 sialylation of the complement proteins and heightened rates of complement deposition during severe COVID-19. However, α-2,3 sialylation also has been demonstrated to modulate immune responses. Increased ST3GAL4 expression responding to the NF-κB pathway resulted in sialylated CD44 expression, exacerbating osteoarthritis in a mouse model [86]. Further, human primary chondrocytes treated with IL-1β or TNF-α increased ST3GAL4 expression and increased cellular α-2,3 sialic acid content which resulted in cartilage homeostasis disruption [87]. ST3GAL4 activity has been associated with adding the α-2,3 sialic acid required for recognition by the Siglec-3, -8, and − 9 [88] and ST3GAL4 expression regulates the synthesis of E-, P-and L-selectin ligands vital for neutrophil adhesion by increasing the binding avidity of the surface antigen sialyl-LewisX [89]. Taken together, the increase in ST3GAL4 during COVID-19 may be exerting proin ammatory downstream effects during disease pathogenesis.
While the receptor FcµR for IgM was demonstrated to bind IgM in a glycan-independent manner [90], IgM has been separately demonstrated to impair T cell proliferation in a sialic acid-dependent manner [91].
For example, the inhibitory sialic acid-binding Ig-type lectin G (Siglec G or CD22) expressed on B-cells and the human Galetcin-9 receptor expressed on the surface of APCs have been reported to bind sialylated IgM [92,93]. Therefore, multiple receptors on immune cells may interact with IgM with increased sialic acid content, resulting in functional consequences of the humoral immune response. More work is required in this area to better understand how the changes in IgM N-glycosylation are associated with immune signaling, effector cell function, and the adaptive immune response during severe disease.
Comparing N-glycans from IgM to N-glycans from IgG Both IgG and IgM are glycosylated by a set of highly regulated glycosyltransferases and glycosidases (GTs) within the Golgi of plasma blast cells. GT expression is regulated by multiple cytokine and chemokine factors during an immune response, and the regulatory factors are not fully elucidated [94]. During COVID-19, we observed signi cant increases in the agalactosylated N-glycans on IgG with concomitant decreases of G2 and S1. In contrast, the IgM N-glycan pro le lost G0 and G1 content, instead gaining S2 and S3 sialic acid as well as acquiring larger, unprocessed mannose content. The differences in glycosylation observed between IgG and IgM from the same COVID-19 patients suggest that ST3GAL4 could add sialic acid to IgM. Because IgG contains nearly all α-2,6 sialic acid [95] and ST6GAL1 transcripts remain unchanged, the upregulation of ST3GAL4 could explain how only IgM gains sialic acid while IgG does not. This is supported by our observation that α-2,3 linked sialic acids are released from IgM A3G3S3 isomers when enzymatically digested with neuraminidase S.

IgM N-glycan correlation with markers of severity and cytokines
Predicting COVID-19 severity continues to be important to appropriately distribute healthcare resources.
Markers of severe COVID-19 infection include elevated D-dimer, blood urea nitrogen (BUN), creatinine, and circulating potassium. These markers of severity signi cantly correlated with the IgM N-glycosylation: mannose and S2 content. It is likely that the potassium, BUN, and creatinine re ect acute kidney injury often observed in severe COVID-19 patients [96]. In previous COVID-19 studies, mild hyperkalemia is associated with COVID-19 severity and acute kidney injury due to severe COVID-19 [97] and altered potassium levels are associated with a poorer prognosis for COVID-19 survival [98]. BUN levels obtained upon emergency room evaluation signi cantly correlated with COVID-19 disease severity [99]. D-dimer indicates recent coagulation cascade activation by providing a marker of clot brinolysis, thus indirectly re ecting circulatory thrombosis [100]. During COVID-19, D-dimer has commonly been found to be elevated in severe COVID-19 patients [4]. In addition, increased titers of IgA and IgG have been associated with a more severe COVID-19 [101]. IgM N-glycan's signi cant correlation with anti-nucleocapsid (N) antibodies suggests that the mechanism of severity in part correlates with IgM N-glycosylation alterations. Taken together, the signi cant correlations between IgM N-glycosylation pro le and markers of severity suggest a potential role or response to the pathogenesis of severe COVID-19.
Of the measured cytokines in the Luminex assays performed at Drexel's IMPACC site, only IL-16 and IL-18 were correlated with the changes in total mannose and S2 content of IgM. Most cytokines elevated in COVID-19 are proin ammatory [102] and IL-16 and IL-18 are no exception. IL-16 has previously been associated with promoting asthma severity by increasing the release of other proin ammatory cytokines [103] and promoting T cell activation by acting as a T-cell chemoattractant after being released by monocytes [104]. No studies of COVID-19 have reported IL-16 elevation associated with the severity of the disease. However, IL-18 is known to be elevated in patients with acute respiratory distress syndrome (ARDS) resulting from in uenza virus infections [105]. One study correlated the level of IL-18 determined from a genome-wide association study (GWAS) was protective against severe COVID-19, and the authors suggested that IL-18 could participate in producing IFN-γ [106]. We postulate that the glycosylation of IgM could be a downstream response to the increased levels of IL-18 and IL-16 because glycosyltransferase expression can respond to cellular stimuli such as cytokine signaling [72].

Antigen-speci c complement deposition
Overactivation of complement has been associated with mortality and morbidity from COVID-19 in severe cases [74,[107][108][109][110]. Because IgM is highly effective at inducing complement, and the N-glycans on IgM were signi cantly altered in severe vs nonsevere COVID-19 patients, we sought to determine if we could con rm the previously reported increases of SARS-CoV-2 antigen-speci c complement deposition. The RBD antigen complement deposition was low, likely due to the lower levels of anti-RBD antibodies present during the rst 10 days of COVID-19 naïve patients lacking previous vaccinations. Thus, we assayed ADCD with the spike S1 antigen and observed higher complement deposition in the severe COVID-19 cohort. Next, we sought to determine if puri ed IgM could activate complement for these SARS-CoV-2 antigens. Puri ed IgM has been previously measured for complement deposition using Guinea pig complement [111,112], but not using COVID-19 antigens. Similar to total plasma, low levels of complement deposition were detected with puri ed IgM from COVID-19 patients in the RBD antigen ADCD assay. However, the Spike S1 antigen complexed with IgM from the severe cohort induced signi cantly higher levels of complement deposited compared to IgM from the nonsevere COVID-19 cohort. IgM interacts with complement C1q via conformational shift on the IgM antigen binding face [83,113], thus we hypothesize that the mannose or sialic acid N-glycans could impact the rate of complement deposition. We digested plasma and IgM with a mannosidase or a sialidase and determined that the ADCD for the Spike S1 antigen was signi cantly reduced when IgM from severe COVID-19 was treated with either a mannosidase or a sialidase. It is intriguing to see that the severe patient IgM glycosylation could be in part responsible for promoting complement deposition during COVID-19 pathogenesis. We hypothesize that complement deposition by IgM could, in turn, promote acute respiratory distress syndrome (ARDS) or acute kidney injury (AKI) observed in severe COVID-19 patients.

Limitations
This report analyzed relatively small cohorts from Drexel's IMPACC study. Larger studies are required to con rm these ndings. Furthermore, this cohort was collected early in 2020 when COVID-19 was predominantly driven by the Wuhan strain. Patients at this time lacked access to life-saving vaccines, antiviral medications, and rapid testing. Therefore, newer variants of the virus, more effective treatments, and vaccination may alter the characteristics of severe COVID-19 patient IgM N-glycosylation. One extraneous source of N-glycans is the IgM pentamer J-chain, however, only one out of the ~ 60 N-glycans per IgM pentamer is associated with the J-chain and thus this potential N-glycan contribution was ignored during data analysis. We have also not determined the ratio of hexamers and pentamers of IgM, however, this ratio could confound the reported increased complement activity. We also analyzed total IgM, limited by our detection method requiring at least 4µg of IgM to obtain adequate uorescent and mass spectrometry signal-to-noise ratios for N-glycan identi cation. Antigen-speci c glycopeptide mapping of IgM N-glycans using an LC-MS/MS platform would provide more accurate information about the speci c immune response to severe COVID-19 infections.

Conclusion
IgM N-glycosylation changes in interesting and unexpected ways compared to IgG N-glycans in severe COVID-19 patients. The identi cation, quanti cation, and correlation of the IgM N-glycan pro le within a well-characterized cohort provided opportunities to learn more about how the human immune system responds to acute viral infections. We align glycosyltransferase expression to the increased mannose complexity and sialic acid content on IgM and contrast these ndings to what is canonically observed in IgG N-glycan pro les from patients with severe COVID-19. We correlate the IgM N-glycan pro le to markers of disease severity and report that spike S1 speci c complement deposition driven by IgM may contribute to severe COVID-19 pathophysiology. A better understanding of IgM N-glycosylation could one day result in novel therapeutics to reduce the severity of acute infectious diseases in humans. Taken together, this data opens the eld for immunoglobulin M to be characterized during other infectious disease states. Table   Table 1 is available in the Supplementary Files section. Figure 1 IgM N-glycosylation analysis reveals differences in COVID-19 patients strati ed by trajectory A) IgM Nglycans labeled with the RapiFluor (RFMS) were pro led with UPLC-FLR-ESI-MS. The resulting N-glycans were identi ed using mass spectrometry and retention time data. Please see Supplementary Table 1 for a complete list of N-glycans. Dashed lines represent N-glycans without con rmed mass identities due to the limitation of the RFMS label in the QDa mass spectrometer. IgM monomer is displayed with the 5 explanation of the N-glycan classes. C)IgM mannosylated N-glycans from non-severe compared to severe COVID-19. A summation of the indicated mannose/hybrid N-glycan sub-groups are graphed to the right.

Figures
IgM N-glycan classes graphed as mean +/-S.D. with signi cance determined using multiple unpaired Ttests *p < 0.05, **p < 0.01, ***p < 0.001   Antigen-speci c complement deposition (ADCD) induced by plasma and IgM from severe and nonsevere COVID-19 cohorts A) Spike S1 and RBD antigen location on SARS-CoV-2 Spike glycoprotein (left), an example of how ADCD assay was quantitated using ow cytometry of plasma compared to PBS-blank sample (center), and the glycosylation of IgM pentamer displaying the c-terminus of IgM containing mannose in orange color while the purple portions of the heavy chain on IgM are complex-type N-glycans