New-Onset IgG Autoantibodies in Hospitalized Patients with COVID-19

Coronavirus Disease 2019 (COVID-19), caused by Severe Acute Respiratory Syndrome Coronavirus-2 (SARS-CoV-2), is associated with a wide range of clinical manifestations, including autoimmune features and autoantibody production. We developed three different protein arrays to measure hallmark IgG autoantibodies associated with Connective Tissue Diseases (CTDs), Anti-Cytokine Antibodies (ACA), and anti-viral antibody responses in 147 hospitalized COVID-19 patients in three different centers. Autoantibodies were identified in approximately 50% of patients, but in <15% of healthy controls. When present, autoantibodies largely targeted autoantigens associated with rare disorders such as myositis, systemic sclerosis and CTD overlap syndromes. Anti-nuclear antibodies (ANA) were observed in ~25% of patients. Patients with autoantibodies tended to demonstrate one or a few specificities whereas ACA were even more prevalent, and patients often had antibodies to multiple cytokines. Rare patients were identified with IgG antibodies against angiotensin converting enzyme-2 (ACE-2). A subset of autoantibodies and ACA developed de novo following SARS-CoV-2 infection while others were transient. Autoantibodies tracked with longitudinal development of IgG antibodies that recognized SARS-CoV-2 structural proteins such as S1, S2, M, N and a subset of non-structural proteins, but not proteins from influenza, seasonal coronaviruses or other pathogenic viruses. COVID-19 patients with one or more autoantibodies tended to have higher levels of antibodies against SARS-CoV-2 Nonstructural Protein 1 (NSP1) and Methyltransferase (ME). We conclude that SARS-CoV-2 causes development of new-onset IgG autoantibodies in a significant proportion of hospitalized COVID-19 patients and are positively correlated with immune responses to SARS-CoV-2 proteins.


Introduction
Coronavirus Disease 2019 , caused by Severe Acute Respiratory Syndrome Coronavirus-2 (SARS-CoV-2) infection, is associated with many different clinical features that are commonly found in autoimmune diseases, including arthralgias, myalgias, fatigue, sicca, and rashes 1- 3 . Less common manifestations of autoimmunity have also been observed in COVID-19 patients, including thrombosis, myositis, myocarditis, arthritis, encephalitis, and vasculitis 3 . These clinical observations, and the increasing proportion of "recovered" patients with persistent post-COVID-19 symptoms (so-called "long haulers", or "long COVID") suggest that inflammation in response to SARS-CoV-2 infection promotes tissue damage in the acute phase and potentially some of the longterm sequelae [4][5][6] .
Autoantibodies, a hallmark of most but not all autoimmune disorders, have been described in COVID-19 patients. In the earliest report, approximately half of hospitalized patients at an academic hospital in Greece had high levels of serum autoantibodies, often associated with clinical findings such as rashes, thrombosis, and vasculitis 7 . Serum anti-nuclear antibodies (ANA) were detectable in approximately one third of patients 7 . Woodruff et al. reported that 23 of 48 (44%) of critically-ill COVID-19 patients have positive ANA tests 8,9 . Zuo described an even higher prevalence of thrombogenic autoantibodies, reporting that up to 52% of hospitalized COVID-19 patients have antiphospholipid antibodies. They further showed that autoantibodies have the capacity to cause clots in mouse models 10 . In a large autoantibody screen, Gruber et al. demonstrated that Multisystem Inflammatory Syndrome in Children (MIS-C) patients develop autoantibodies, including autoantibodies against the lupus antigen SSB/La 11 . SSB/Ro autoantibodies have also been described 12 . The apparent link between clinical manifestations resembling those seen in patients with classifiable autoimmune diseases, and those observed in COVID-19 patients, has prompted searches for candidate target autoantigens that may be useful for diagnosis and for improving understanding of COVID-19 pathogenesis. The molecular targets of autoantibodies in individual patients with COVID-All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 1 1 19 are largely unknown, as are their associations with anti-viral immune responses, and the timing of their appearance in regard to infection with SARS-CoV-2.
We hypothesized that SARS-CoV-2 induces the production of antibodies against autoantigens and cytokines/chemokines de novo, and these correlate with anti-viral responses. We assembled three different custom bead-based protein arrays to measure IgG antibodies found in CTDs, ACA, and anti-viral responses in 197 COVID-19 samples. Samples were obtained from 147 hospitalized patients infected by SARS-CoV-2, some of which were collected longitudinally, in three geographically distinct locations. Our results demonstrate that a large cadre of autoantigens are targeted by circulating antibodies in a substantial proportion of hospitalized patients with COVID-19, but less commonly in uninfected healthy controls (HC). Our studies confirm emerging reports of IgG autoantibodies in hospitalized COVID-19 patients and demonstrate that a significant subset of patients develop new-onset autoantibodies that could place them at risk for progression to symptomatic, classifiable autoimmunity in the future.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Anti-nuclear antibodies (ANA) are produced by one in four hospitalized COVID-19 patients
To determine if hospitalized patients with COVID-19 produce autoantibodies against prototypical autoantigens associated with systemic autoimmunity, we measured ANA using an indirect immunofluorescence assay in one of our cohorts (University of Pennsylvania). We found that seven out of 73 patients (10%) were positive at a dilution of 1:160 using a clinical-grade assay and that another 13 were weakly positive (Supplementary Fig. 1a). A variety of ANA patterns were observed including diffuse, speckled and nucleolar ( Supplementary Fig. 1b and 1c). One patient exhibited cytoplasmic staining but was negative for nuclear staining. Given the finding of positive and weakly positive ANAs, we measured dsDNA antibodies. Only one individual out of 73 tested was positive for dsDNA antibodies at a dilution of 1:270, and this individual also was ANA positive with a speckled pattern (Supplementary Fig. 2). Since several patients who were severely or critically ill had thromboembolic and vascular events, we also analyzed the same 73 patients for Myeloperoxidase (MPO) and Proteinase 3 (PR3) antibodies, as these antibodies are associated with autoimmune vasculitis. Only one individual tested positive for PR3 antibodies (Supplementary Fig.   2). The levels of positivity in these clinical-grade assays are in line with those of one of the authors (J.J.) who reported that 17 of 113 (15.8%) patients with positive SARS-CoV-2 serology had serum autoantibodies and/or antiphospholipid antibodies 13 . These findings prompted us to "cast a wider net" for autoantibodies using additional patients and assays that detected larger numbers of not only common, but also unusual autoantigens.

Protein microarrays identify autoantibody targets in hospitalized COVID-19 patients
To systematically and simultaneously measure a large number of different autoantibodies in serum or plasma derived from patients acutely infected with SARS-CoV-2, we constructed a 53-plex All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 1 3 COVID-19 Autoantigen Array (Fig. 1, left half of panel We characterized 51 cross-sectional COVID-19 serum or plasma samples from patients who provided samples within seven days of hospitalization (Fig. 1). As expected, prototype reference samples from patients with classifiable autoimmune diseases were strongly positive for autoantibodies, recognizing 25 of the 53 arrayed proteins (Fig. 1, bottom  HC samples. Both samples were therefore considered "positive" in our analysis, but we included them in calculating the 5 SD cutoff on the COVID samples. In striking contrast, 25 of 51 (49%) hospitalized patients with COVID-19 had autoantibodies recognizing at least one autoantigen (Fig. 1, top panel).
Using a stringent 5 SD cut-off, serum antibodies from eleven COVID-19 patients identified a single All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. antigen, thirteen recognized 2-3 antigens, and one subject (Subject UP40) recognized nine different autoantigens. Ribosomal P proteins (P0, P1, and P2) were most prominently targeted in patients (10 of 50 patients, 20%), but were not found in any of the HC. Similar results were observed in 48 Kaiser subjects analyzed using an earlier-generation 26-plex autoantigen microarray, identifying overlapping RNA-containing autoantigen complexes including RPP14 Th/To, the Ro/La particle, the U1-small nuclear ribonuclear protein (U1-snRNP), thyroid antigens, and chromatin proteins as targets in hospitalized COVID-19 patients, but in none of the HC (Supplementary Fig. 4).
Rare antigens seen in patients with autoimmune myositis (MDA5, Mi-2, and tRNA synthetases such as PL-7 and Jo-1), and candidate autoantigens in autoimmune myocarditis (troponin and MYH6, fibrillarin, and the U11/U12 snRNP, Fig. 2b). A subset of autoantibodies (e.g., antibodies that bind the complement inhibitor C1q, thrombosis-associated antibodies that target beta 2 glycoprotein 1 (β2-GP1), and vasculitis-associated antigens such as bactericidal permeability inducing protein (BPI)) that have been implicated in pathogenic inflammation in target organs, were also found in individual patients ( Fig. 2c) 4-6 14-16 . Relatively common autoantigens such as Scl-70, CENP A/B, and Sm/RNP were infrequent. Thyroid autoantibodies were also commonly observed (12/147 subjects across our entire study, 8.2%, using cutoffs of 3,000 MFI and 5 SD above HC). Thyroid dysfunction, which is relatively common in the general population, has been reported in COVID-19 patients 4,5 . In all cases where samples from more than one time point were available, anti-TPO and anti-thyroglobulin (TG) were already present at high MFI levels in the baseline sample. Taken together, these findings reveal that hospitalized patients with COVID-19 produce an increased frequency of autoantibodies, but that there is substantial inter-individual variation in which autoantigens are targeted.

Secreted proteins are common autoantigens in hospitalized COVID-19 patients
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Table 2). We observed even more striking results with the secretome array, which revealed that serum antibodies in 41/51 (80%) of hospitalized COVID-19 patients recognized at least one secreted or cell surface autoantigen (Fig. 1, upper right half of panel), while only 2/31 (6%) HC subjects recognized a single antigen (interferon-gamma, IFNγ in one and CD74 in the other, Fig. 1, middle right half of panel). Interestingly, the IFN-γ+ HC subject (HC27) also had serum antibodies specific for Sm (a subunit of the U1-snRNP, using 5 SD cutoff) and for both Ro60 and La (using a 3 SD cutoff), suggesting this "healthy" subject is in preclinical evolution toward developing SLE, a disease in which we have previously described multiple different ACA including anti-IFN-α and anti-B cell activating factor (BAFF) 18 .
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. reactivities were observed in individual COVID-19 patients, including IL-12p70 (Subject UP47); the SARS-CoV-2 receptor angiotensin converting enzyme-2 (ACE-2, Subject UMR19); granulocyte macrophage colony stimulating factor which is the causative autoantibody target in PAP (GM-CSF, Subject UP25); oncostatin-M (OSM, Subject UP40); and soluble receptor activator of nuclear factor kappa B (sRANK-ligand, Subject UP19). Subject UP17 was being treated with a tumor necrosis factor-alpha (TNF-α) inhibitor at the time of SARS-CoV-2 infection, explaining the high MFI reactivity to TNF (Fig. 1). MFI for all antigens except IL-12p70 were very high (>10,000) in individual patients.
Autoantibodies against all interleukins, cytokines and ACE-2 identified in the initial screen were also observed using a 5 SD cutoff in a second COVID-19 cohort (n=98 longitudinal samples from 48 different patients, see although IL-1β was targeted using a 3 SD cutoff; IL-31, which met a 3 SD cutoff; and GM-CSF).

A subset of autoantibodies is triggered by SARS-CoV-2 infection
To determine if autoantibodies and anti-cytokine antibodies were generated de novo (versus As with the unpaired samples described in Fig. 1, autoantibodies from patients with paired samples had high MFIs in individual patients. Some patients were again identified whose serum All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. recognized a large number of autoantigens (Supplementary Figs. 5 and 6). Twenty-five (52%) of hospitalized COVID-19 patients had autoantibodies against at least one autoantigen. Serum autoantibodies recognized two or more antigens (range 2-7 antigens) in seven patients (15%) (Fig.   3a, and Supplementary Fig. 6).
Longitudinal analysis identified prominent increases in autoantibodies at the second available time point (Fig. 3b, red boxes). In 9 individual patients (19%), autoantibody measurements were above the average for HC at the earliest available time point and MFI increased by at least 50%, exceeding the 5 SD and 3,000 MFI cutoff at the later time point (Fig.   3b, e.g., MDA5, subject UP50; BPI, subject UP52; Supplementary Fig. 6). Some autoantibodies were at or below the average for HC at the first time point and increased over time (e.g., histones and histone H3, subject UP65; and β 2GP1, subjects UP65 and UP52), suggesting these autoantibodies were directly triggered by SARS-CoV-2 infection. Others were already elevated at the first time point and did not have large increases in MFI over time (n=22, 45%, blue boxes) ( Fig. 3c and Supplementary Fig. 6). In a small number of cases, autoantibody MFI levels decreased below the SD and MFI cutoffs over time (n=5, 10%, green boxes), suggesting that their development might be transient (e.g., PL-7, subject UP70, Fig. 3b). Anti-TPO and anti-Scl-70 (Fig. 3c, blue boxes) remained elevated at high levels in all seropositive subjects regardless of the time of measurement, suggesting that these autoantibodies were already present at hospitalization and likely represent preclinical (asymptomatic), unreported, or undiagnosed autoimmunity.
To further evaluate the potential evolution of autoantibodies, we performed ANA testing on 21 individuals with paired samples. Eight individuals (38%) had positive or weak positive ANA reactivity.
Among these 8 individuals, ANAs were present at both time points in three, changed in intensity of staining in two, and were positive at only one of the two time points in the final three ( Supplementary   Fig. 1b). Taken together, these data indicate that autoantibody levels change over time in individual COVID-19 patients, consistent with their production and, in some cases, transience during acute illness.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 1 8 We next examined whether IgG ACA are triggered by SARS-CoV-2 infection. Paired samples from the same 48 subjects described above were used to probe the 41-plex cytokine array, again in a single, batched run. As observed with unpaired samples (Fig. 1), 28 of 48 (58%) of COVID-19 patients had at least one ACA (Supplementary Fig. 6). Of these twenty-eight, sera from fifteen patients recognized one cytokine, five recognized two cytokines, and eight recognized three or more cytokines (range 3-12 antigens). Interferons, IL-17, and RANK-L were the most common targets, and interferons, IL-17, and IL-22 were new targets in some patients (Fig. 3d). In addition to Subject UMR19 (Fig. 1), a second patient with high MFI ACE-2 autoantibodies was also identified (Subject To further evaluate the change in autoantibodies and ACA over time, we performed a targeted analysis of 21 of the 48 patients who had paired autoantibody and ACA data specifically at D0 and D7 of hospitalization (Supplementary Fig. 7). Almost all patients (18/21, 86%) had demonstrable changes in the number of antibodies, defined at varying thresholds of sensitivity (>3 SD vs. 3-5 SD vs. >5 SD) between D0 and D7 (Supplementary Fig. 7a). When combining the number of autoantibodies or ACA (Supplementary Fig. 7b), there is a trend towards increased numbers both of autoantibodies and ACAs per subject over time. Higher numbers of individuals with more All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 1 9 autoantibodies and ACAs at D7 compared to D0 at the 3-5 SD threshold were observed (Supplementary Fig. 7b and 7c), but the difference in medians between D0 and D7 was not statistically significant. Nevertheless, these data clearly show that there is ongoing evolution in both the numbers and levels of autoantibodies and ACAs with time in hospitalized COVID-19 patients.

patients
We have used protein arrays for epitope mapping and to measure antibody responses in influenza vaccines 22 and in a nonhuman primate human immunodeficiency virus (HIV) vaccine study 23 . We used a similar approach here to characterize anti-viral responses following SARS-CoV-2 infection. We created a 28-plex COVID-19 viral array that included structural and surface proteins from SARS-CoV-2 as well as eight commercially available recombinant nonstructural proteins localized to the interior of the virus (Supplementary Table 3). As an initial validation, we compared array-based detection and measurement using a clinical-grade ELISA (R=0.81, Spearman's, p<0.0001 for anti-SARS-CoV-2 nucleocapsid; R=0.60, Spearman's, p<0.0001 for anti-SARS-CoV-2 RBD, Supplementary Fig. 8a and 8b, respectively). By studying the anti-viral antibody (AVA) response, we hoped to understand if certain viral antigens might correlate with the development of autoimmune responses. We hypothesized that poorly controlled SARS-CoV-2 infection leads to the development of serum antibodies that recognize not just structural proteins such as the SARS-CoV-2 spike protein, but also nonstructural proteins, and that a subset of these viral proteins might correlate with the development of autoimmunity. Proteins from related coronaviruses were also included to explore whether pre-existing antibodies to seasonal coronaviruses might correlate negatively or positively with disease severity, and with autoimmunity. Fig. 4 depicts a heatmap representation of IgG reactivity based on MFI (Fig. 4a, left panel) and calculation of SD above average MFI for HC (Fig. 4b, right panel). As expected, nearly all All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. patients had broad immune responses to viral structural proteins (first seven antigens on left, Fig. 4a and b). Twelve patients had low MFI levels at the earliest time point (almost all were day 0, defined as collection within the first 24-72 hours of hospitalization but developed high MFI IgG SARS-CoV-2 antibodies when tested at later time points, consistent with previously published findings in the setting of acute illness 24 (Fig. 4a). Other subjects (e.g., subject UP50) already had broad AVA responses at day 0, suggesting they had been infected for a significant period of time prior to hospital admission.
IgG antibody levels against non-structural SARS-CoV-2 proteins were significantly elevated in antibody responses in which multiple non-structural antigens were targeted in the same subject. In rare patients (e.g., subject UP65, see SARS-CoV-2 protein PLpro, Fig. 4a and 4b), the initial immune response was focused on an internal protein (or was pre-existing) and later evolved to target spike and other SARS-CoV-2 surface or structural proteins. We conclude that antibody responses in hospitalized COVID-19 patients are not limited to structural proteins, that linked responses to multiple non-structural proteins are observed over time, and that NSP9 is the most commonly recognized internal SARS-CoV-2 protein of those tested on the array.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

New-onset IgG autoantibodies are temporally associated with anti-SARS-CoV-2 IgG responses
We next identified a subgroup of patients (n=12) whose anti-SARS-CoV-2 antibody responses suggested that they had been infected at a time point that was proximate to hospitalization and capture of the first sample. Selection criteria for patients who were early in their anti-viral responses included (i.) the first available sample was within three days of hospitalization; (ii.) anti-spike S1 IgG levels were <5,000 MFI at baseline; (iii.) anti-RBD IgG levels were <20,000 MFI at baseline; and (iv.) at least a 2-fold increase in MFI for IgG against both S1 and RBD was observed at the next available timepoint. We then studied these patients to further determine if new IgG autoantibodies appeared at the second time point, providing evidence that SARS-CoV-2 directly triggers development of autoantibodies.
We compared IgG reactivities at both time points for all twelve subjects who met the above criteria on COVID-19 autoantigen arrays (Fig. 5a, left panel) and cytokine arrays (Fig. 5b, middle panel) with anti-viral responses using the virus array (Fig. 5c, right panel). Four of twelve patients were found to have at least one newly induced autoantibody at the later time point (white boxes). Two of these four patients had two or more new autoantibodies (Subjects UP52, n=5 antigens; and subject UP65, n=10 antigens). β 2GP1, histones, and the 54 kD component of the myositis autoantigen signal recognition particle (SRP 54) were the most common antigens identified (n=2 subjects each). Given the small sample size, no clear correlations were identified between individual autoantibodies and an IgG response to a specific viral protein ( Fig. 5d and Supplementary Fig. 9).
Finally, we correlated autoantibodies and ACA with anti-viral IgG responses using array data from the cohort described in Fig. 3, focusing on Penn and Marburg samples which had been (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021.

Discussion
We have used a multiplexed, bead-based platform to identify circulating antibodies in hospitalized patients with COVID-19 and have generated integrated results from three different protein microarrays to discover COVID-19 associated autoantigens and link them to anti-viral responses. Our studies have led to several important findings that provide further insights into COVID-19 pathogenesis. First, we found that approximately half of hospitalized COVID-19 patients develop serum autoantibodies against one or more antigens on our array even though only a quarter of all patients are ANA+. Increased levels of autoantibodies are not simply a reflection of hypergammaglobulinemia because they are produced out of proportion to total IgG serum concentration. In most individuals, only a small number of autoantigens are targeted, which is more consistent with a sporadic loss of self-tolerance than a global increase in autoantibody production.
Second, the autoantibodies we discovered are found in relatively rare connective tissue diseases that are not typically measured in clinical labs, and some are predicted to be pathogenic. Third, a surprisingly large number of ACA were identified, far more than just the interferon autoantibodies described recently 21 . Fourth, antibodies recognizing nonstructural SARS-CoV-2 proteins were identified that correlate positively with autoantibodies. Finally, and perhaps most importantly, some autoantibodies are newly triggered by SARS-CoV-2 infection, suggesting that severe COVID-19 can break tolerance to self.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 2 3 Approximately 60-80% of all hospitalized COVID-19 patients in our study had at least one ACA, with a greater number of different ACA specificities generated in individual patients than In addition to ACAs modulating the immune response and potentially causing more destructive inflammation, autoantibodies have the potential to contribute in a number of other ways to COVID-19 pathogenesis. Several autoantigens we discovered are naturally complexed with a structural RNA All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. lung tissue and normal-appearing skin 37,38 . Magro and colleagues hypothesize that spike protein on the surface of circulating pseudovirions binds to endothelial ACE-2 (whose gene is interferon-All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Severe infection may also result in an "all-hands-on-deck" immune response that results in loss of tolerance due to the presence of pro-inflammatory mediators that may lessen the requirement for T cell help. Some patients with severe acute COVID-19 appear to mount extrafollicular B cell One of the most important unanswered questions raised by our studies is why specific molecules are targeted in hospitalized COVID-19 patients. For newly triggered ACA, the most likely explanation is that they arise as a consequence of severe disease along with high levels of viremia, tissue injury, and elevated local levels of pro-inflammatory cytokines and chemokines. However, it is also possible that the presence of ACA could affect the regulation of self-reactive lymphocytes by altering the half-lives of the receptor interactions of the target molecules. For traditional autoantigens, one possibility is that viral proteins or the SARS-CoV-2 RNA genome and self-molecules physically interact, and that the initial immune response to the viral protein in a highly inflammatory microenvironment expands to include self-proteins through linked recognition and intermolecular All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 2 6 epitope spreading. Another possibility is molecular mimicry in which one or more viral proteins or epitopes cross reacts with self-proteins leading to loss of tolerance and development of autoimmunity 51,52 . Experiments to explore these mechanisms are ongoing. Many studies of hospitalized COVID-19 patients, including our study, suffer from important limitations. First, confounding variables exist including heterogeneous demographics, medications at All rights reserved. No reuse allowed without permission.

The vast majority of studies on SARS-CoV
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 2 7 hospitalization, individualized treatment approaches, and, in some cases, unknown history of preexisting medical or autoimmune conditions. Second, "Day 0" is not day 0 of infection but instead refers to a time point most proximate to hospitalization. Our viral array results (Figs. 4 and 5) confirm that the time between initial infection and sample acquisition was heterogeneous, potentially confounding interpretation of autoantibody and ACA results. Third, not all antigens (e.g., lipids, hydrophobic proteins and carbohydrates) are compatible with our screening methodology, and as a result we have certainly missed some reactivities. Fourth, we did not include patients who were asymptomatic, had mild COVID-19, were vaccinated for SARS-CoV-2, had other severe viral illnesses, or were children. Finally, our analysis was limited to hospitalized patients during acute illness, with follow up times of days rather than months or years.
Although beyond the scope of these studies, our data generate many more questions that While the COVID-19 pandemic is leaving a wake of destruction as it progresses, it also provides an All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 2 8 unprecedented opportunity to understand how exposure to a new virus could potentially break tolerance to self, potentially giving rise to autoimmunity and other chronic, immune-mediated, diseases.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Bead-based antigen array content
We created three different custom, bead-based antigen arrays modelled on similar arrays that we previously used to study autoimmune and immunodeficiency disorders, and for characterizing vaccine responses 18,20,22,[53][54][55][56][57] . Antigens were selected based on our published datasets; literature searches that have implicated specific antigens in COVID-19; potential for mechanistic contribution to COVID-All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 19 pathogenesis; and compatibility with bead-based platforms. A complete list of all antigens, vendors, and catalogue numbers can be found in Supplementary Tables 1-3.

Array construction
Antigens were coupled to carboxylated magnetic beads (MagPlex-C, Luminex Corp.) such that each antigen was linked to beads with unique barcodes, as previously described 53,58 . In brief, unless stated otherwise, 8 μg of each antigen or control antibody was diluted in phosphate buffered saline (PBS) and transferred to 96-well plates. Diluted antigens and control antibodies were conjugated to 1×10 6 carboxylated magnetic beads per ID. Beads were distributed into 96-well plates (Greiner BioOne), washed and re-suspended in phosphate buffer (0.1M NaH 2 PO 4, pH 6.2) using a 96-well plate washer (Biotek). The bead surface was activated by adding 100 μ l of phosphate buffer containing 0.5 mg 1-All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 3 2 washed with 3 × 60 µl PBS-Tween and re-suspended in 60 µl PBS-Tween prior to analysis using a (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 3 3 patients known to have clinically elevated PR3 and MPO antibody levels. Results were scored as positive or negative based upon the kit instructions. All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Statistical Analyses
All data analysis and statistics were performed using R and various R packages 61 . For normalization, average MFI values for "bare bead" IDs were subtracted from average MFI values for antigen conjugated bead IDs. The average MFI for each antigen was calculated using samples from healthy subjects known to be uninfected with SARS-CoV-2 (all obtained before December 2019). Antibodies were considered "positive" if MFI was > 5 SD above the average MFI for HC for that antigen, and MFI All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
A less stringent 3 SD cutoff used in a Luminex assay to measure SARS-CoV-2 immunoglobulins in blood and saliva 62 was also employed for comparison in some experiments. An example can be found in Supplementary Figs. 3 and 4. ELISA and antibody number data were visualized in GraphPad Prism v.9.0.0 (86). Upon publication of this study in a peer-reviewed journal, deidentified array data will be uploaded to the Gene Expression Omnibus (GEO) database.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. All remaining authors report no conflicts of interest with the research reported in this manuscript.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Jennifer Okwara for assistance with ANA indirect immunofluorescence image analysis; and Scott Hensley for providing the RBD antigen for the ELISA.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 4 8 indicate autoantibodies whose MFI measurements are >5 SD (red) or < 5 SD (black) above the average MFI for HC. MFIs <3,000 were excluded. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Figure Legends
The copyright holder for this preprint this version posted January 29, 2021. ; https://doi.org/10.1101/2021.01.27.21250559 doi: medRxiv preprint 4 9 axis. Subjects are shown on x-axis (COVID-19 patients, HC, and Prototype Autoimmune). Average MFI for HC, 3 SD above the average MFI for HC, and 5 SD above the average MFI for HC are shown with orange lines. Error bars represent one standard deviation of the MFI for sample replicates.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted January 29, 2021. All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Supplementary
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.