Broad Auto-Reactive IgM Responses Are Common In Critically Ill COVID-19 Patients.

The pathogenesis of severe COVID-19 remains poorly understood. While several studies suggest that immune dysregulation plays a central role, the key mediators of this process are yet to be defined. Here, we demonstrate that plasma from a high proportion (77%) of critically ill COVID-19 patients, but not healthy controls, contains broadly auto-reactive immunoglobulin M (IgM), and only infrequently auto-reactive IgG or IgA. Importantly, these auto-IgM preferentially recognize primary human lung cells in vitro, including pulmonary endothelial and epithelial cells. By using a combination of flow cytometry, LDH-release assays, and analytical proteome microarray technology, we identified high-affinity, complement-fixing, auto-reactive IgM directed against 263 candidate auto-antigens, including numerous molecules preferentially expressed on cellular membranes in pulmonary, vascular, gastrointestinal, and renal tissues. These findings suggest that broad IgM-mediated autoimmune reactivity may be involved in the pathogenesis of severe COVID-19, thereby identifying a potential target for novel therapeutic interventions.

these cells have never been exposed to SARS-CoV-2 naïve, antibodies detected in this assay re ect the targeting of self-antigens and are not the consequence of reactivity against SARS-CoV-2 antigens.
Analysis of cells using conventional ( Figure 1a) and imaging ow cytometry (Figure 1b-d) revealed the presence of antibodies binding to the plasma membrane. Scored against healthy and non-COVID controls, auto-reactive IgA, auto-reactive IgG and auto-reactive IgM were detected in 28 (51%), 23 (42%), and 51 (93%) out of 55 COVID ICU patients, respectively (Figure 1e). In each reaction, the percentage of cells that stained positively for IgM antibodies was far greater than IgA or IgG, suggesting higher circulating auto-reactive IgM titers. Although COVID ICU patients were associated with higher circulating interleukin-6 (IL-6) and C-reactive protein (CRP) (Supplementary Figure 1a-b), only auto-IgM levels were modestly associated with increased plasma interleukin-6 (IL-6) ( =0.29, p=0.0056; Supplementary Figure   2a-b). Of note, most COVID ICU patient plasma showed IgA, IgG, IgM, or a combination of, reactivity with cells of pulmonary origin (Figure 1f). Although a signi cant percentage of COVID ICU patients had detectable levels of auto-reactive IgA and IgG, we focused on auto-reactive IgM given its substantially higher titers and frequency. Overall, this rst set of data revealed that high-titer auto-reactive IgM are frequently detected in patients with severe COVID-19 and that the reactivity is most pronounced against cells of pulmonary epithelial and endothelial origin.
We next sought to understand which auto-antigens are targeted by these circulating auto-reactive IgM in COVID-19 patients. Plasma samples from COVID ICU patients with strong auto-reactive IgM titers (n=5), non-COVID ICU patients (n=3) and healthy controls (n=4) were surveyed in analytical human proteome microarrays (HuProt v4 array). The array epxresses over 21,000 intact proteins, therefore allowing for a thorough and comprehensive investigation of potential binding targets for auto-reactive IgM antibodies.
For stringency, a potential binding target was considered for any protein that had a uorescence signal at least 4 standard deviations (Z-score>4) above the array mean. Additionally, the target had to possess a uorescence signal at least 2 Z-scores above the same target across all healthy controls. This strict approach resulted in the identi cation of 260 candidate autoantigens that were uniquely linked to COVID ICU patients (Figure 2a and Supplementary Table 2). Of note, the auto-reactive IgM repertoire in COVID ICU patients is broad, and the candidate targets infrequently overlapped among different patients included in this cohort (Figure 2b). It is very likely, and anticipated, that interrogation of additional plasma samples from patients with severe COVID-19 by proteome microarray would identify further auto-antigen targets, and that the individual antigenic targets are likely less relevant to the disease pathogenesis than the overall abundance, breadth, and tissue speci city of the observed auto-antibodies.
Given the high Z-scores of each candidate target, the auto-reactive IgM antibodies are circulating at robust titers and/or bind with high avidity to the respective targets. We next sought to determine whether the candidate autoantigens were expressed in key tissue types. Using arterial tissues as surrogates for endothelial sites, small intestinal and colonic tissues as surrogates for gastrointestinal sites, as well as renal, neural, and pulmonary sites, we found 226 candidate autoantigens expressed at above-background levels in these cell types ( Figure 2c). Importantly, we identi ed 16 autoantigens associated with the human plasma membrane proteome 15 and therefore considered these molecules as important prospective candidate targets for circulating pathogenic auto-reactive IgM (Figure 2d). We next investigated whether these proteins shared similar motifs. Although N-linked glycosylation was predicted in 11 candidate autoantigens, heterogeneity in amino acid sequences anking predicted N-linked glycosylated residues indicated minimal in uence of N-linked glycosylation on potential IgM binding motifs (Supplementary Figure 3a). However, an arti cial neural network prediction model 16  non-survivors, compared to three COVID-19 negative control patients for whom lung tissue was available from cancer-related resection ( Figure 3a). It should be noted that some modest IgM deposition in the COVID-19 negative patient controls was expected, as auto-reactive IgMs can develop during lung cancer progression 17 and/or following radiation therapy 18 . While we cannot formally rule out that the IgM detected in COVID-19 + lung tissue are reactive against SARS-CoV-2 surface antigens, the observed staining patterns are not consistent with the distribution patterns observed for SARS-CoV-2 antigens such as the Spike protein 19,20 . Importantly, the extensive IgM staining patterns are at levels at least three times higher than COVID-19 negative controls (Figure 3b), and are not described for other causes of acute respiratory distress 21 . Further histological analysis revealed, in the lung of severe COVID-19 patients, signi cant alveolar damage and patchy hemorrhage, alongside extensive in ammatory in ltrate breaching the alveolar lumen. Previous studies have linked alveolar damage to dysregulated cytokine release and neutrophil extracellular traps seeded by resident macrophages [22][23][24][25] . Yet, these observations could also be linked to auto-reactive IgM, through the capacity of these immunoglobulins to x complement and induce cytotoxicity. Indeed, staining for complement component 4 (C4d), a marker of complement activation, showed a two-fold increase in COVID-19 patients compared to negative controls (Figure 3c), indicating frequent in vivo complement xation.
Complement-dependent cytotoxicity (CDC) and complement deregulation have been proposed to play a roles in the pathogenesis of ARDS 26 . Additionally, as there is considerable pulmonary microangiopathy observed in severe COVID-19 patients 27,28 , it is conceivable that CDC can precede or even cause the damage to the pulmonary endothelium. Given the observed IgM and C4d binding to pulmonary targets and to con rm that the auto-reactive IgM can mediate CDC, we next tested plasma samples from severe COVID-19 patients for their capability of xing complement and inducing cytotoxicity in vitro. To this end, we investigated patient plasma samples that showed greater than 10% binding to the respective cell type in the screening assay. Interestingly, we consistently observed higher rates of CDC in cells of pulmonary origin (Figure 3d-h). In addition, while non-COVID-19 ICU patient plasma samples induced limited or no cell death, most COVID-19 ICU patients plasma samples induced cell death at frequencies proportional to their measured level of cell binding (Figure 3i). Collectively, these data indicate that auto-reactive IgM present in plasma from severe COVID-19 patients can x complement and induce cytotoxicity.
The identi cation of auto-reactive IgM as a potential contributing factor to the pathogenesis of severe COVID-19 has two immediate implications. First, this observation may explain how COVID-19 is disproportionately more serious in the elderly 29 , who typically manifest higher plasma levels of circulating auto-reactive antibodies 30 . This phenomenon would be exacerbated by decreases in functional T follicular helper cells that promote antibody class switching 31 , a process associated with better disease outcomes 32 . Given that IgM levels peak within a week of the clinical onset of COVID-19 and persist at similar levels for weeks thereafter 34 , the elderly face a protracted period where there is steadfast secretion of auto-reactive IgM that maintain relatively low a nity for the same epitope without either switching to alternate antibody class types or undergoing somatic hypermutation and a nity maturation. In this perspective, the elderly may be more prone to severe COVID-19 due to a more protracted exposure to the cytopathic effects of auto-reactive IgM.
Secondly, it is conceivable that this type of immunopathology can be limited by therapeutic interventions that inhibit the IgM-complement axis. In the immediate term, this approach could mitigate the SARS-CoV-2 associated alveolar damage and ARDS [35][36][37] , and consequently protect against mortality 38 and/or reduce the need for invasive mechanical ventilation 39 . In the long term, preservation of lung integrity may prevent pathogenic sequelae such as pulmonary brosis 40,41 , which diminishes lung function postrecovery 42 . These therapeutic goals could be implemented through the use of immunosuppressants, such as dexamethasone, that can attenuate the production of auto-reactive IgM 43 , plasma exchange to remove auto-reactive IgM once formed 44 , or to synergize and supplement proposed anti-brotic therapies 45 . Alternatively, the complement cascade can be directly inhibited through conestat alfa 46 or eculizumab 46 , and indeed, both drugs are presently undergoing evaluation through clinical trials to determine e cacy 47 . Optimistically, our ndings cast support for interventions that can be readily and swiftly implemented in the clinic to alleviate or prevent serious COVID-19 complications.
In summary, we found that broadly auto-reactive IgM are common in the plasma of patients with severe COVID-19. These auto-reactive antibodies bind pulmonary epithelial and endothelial targets, at which point they can be potent mediators of cytopathicity through the recruitment of complement. Future studies will investigate the relationship between SARS-CoV-2 infection and the emergence of autoreactive antibodies, and determine whether immunosuppressive therapy can reduce the levels of autoreactive IgM in plasma and consequently attenuate the clinical severity of COVID-19. IL-6 ELISA. Plasma levels of IL-6 were quanti ed using a Human IL-6 ELISA kit (ab178013, Abcam) and following the manufacturer's instructions.
Histology. Five-micrometer sections from formalin-xed, para n-embedded lung tissue sections were tested for IgM expression using a rabbit anti-IgM polyclonal antibody (cat. no. F0203, Agilent, Santa Clara, CA) at 1:400 dilution and for C4d expression using a rabbit anti-C4d polyclonal antibody (cat. no. 04-BI-RC4D, ALPCO Diagnostics, Salem, NH) at 1:100 dilution. IgM staining was performed on a Dako Link48 Autostainer with the EnVision FLEX dual-link system (Dako, Carpinteria, California) after heatinduced epitope retrieval in citrate buffer for 30 minutes. C4d staining was performed on a Leica Bond III automated stainer with the Bond Polymer Re ne Detection Kit (Leica Microsystems, Bannockburn, IL) after on-board epitope retrieval using Bond epitope retrieval solution 1 (ER1) for 20 minutes. Images were analyzed in ImageJ using the IHC Image Analysis Toolbox for the enumeration of nuclei, and to identify stained regions. The Color Pixel Counter plugin was further used to quantify the extent of staining in each image.
Complement Fixing Assay. Target cells were dissociated from culture asks by TrypLE Express reagent (Gibco) and resuspended in PBS at a concentration of 1x10 6 cells/ml. 50µl of the cell suspension was transferred to wells of a 96-well V-bottom plate. 50µl of plasma was added to each well and plates were incubated at 4ºC for one hour. 2 non-COVID (ICU) and 2 healthy donor plasma samples without IgM reactivity were selected as controls. Cells were washed with cold DPBS twice and resuspended in 100µl DPBS. 11µl of reconstituted rabbit complement (Low-Tox-M rabbit complement, Cedarlane) was added to each well. To one well, 0.1% Triton X-100 was added to induce cell lysis. Plates were then transferred to a 37ºC incubator for two hours. Plates were then centrifuged at 500g for 5 minutes to pellet cells. 50µl of the supernatant was transferred to a at-bottom 96-well plate in duplicate. 50µl of reconstituted lactose dehydrogenase assay reagent (CyQUANT LDH Cytotoxicity Assay, Invitrogen) was then added to each well, and the plate was subsequently protected from light and left at ambient temperature for 30 minutes, after which 50µl of the included stop solution was added. Absorbance was read at 490nm and 680nm (Varioskan LUX multimode plate reader, Thermo Fisher). Absorbance values at 680nm were subtracted from absorbances at 490nm and duplicate values averaged. Percentage cytotoxicity was calculated by comparing the absorbance values against the lysed-cell and healthy-donor controls.
Protein Array. 5 COVID-19 (ICU) and 3 non-COVID-19 (ICU) samples characterized as enriched with auto-IgM by the ow cytometry assay described above were submitted alongside 4 randomly chosen healthy control samples to CDI laboratories (Baltimore, MD) for antigen-speci city screening across >21,000 full-  Analytical human proteome microarray reveals a broad auto-reactive IgM repertoire. a: Plasma samples from 4 healthy human donors, alongside demonstrated auto-reactive IgM+ plasma samples from 5 COVID ICU patients and 3 Non-COVID ICU patients, were analyzed in a proteome microarray to determine potential IgM targets. Z score ranges (the number of standard deviations above/below the microarray signal intensity median) of targets in healthy controls were ranked in ascending order and represented as the black trace. Candidate targets of auto-reactive IgM in patient samples were selected through possession of a 2.0 Z-score difference compared to maximum in healthy controls, and a nal Z-score greater than 4.0. Each patient is uniquely color-coded. Greater shape opacity and size indicate higher Zscores, and are linked to higher antibody titers and/or avidity. b: Z-scores of all candidate auto-reactive IgM targets (n=260) found in COVID ICU patients are represented in this heatmap alongside Z-scores of the corresponding target in non-COVID ICU patients and healthy human controls. The IgM repertoire of each COVID ICU patient uniquely target a broad variety of proteins, which in turn are generally not targeted by auto-reactive IgM found in non-COVID ICU patients and healthy controls. Key shows Z-scores. c-d: Gene expression pro le of candidate COVID ICU patient auto-reactive IgM targets in key organ sites shows most of the targets are normally expressed at meaningful levels (c). Heatmap representation of candidate genes empirically shown to localize at the plasma membrane, and their relative expression in the listed tissue sites (d). Key shows expression levels in natural log scale. Figure 3