Novel and unique rheumatoid factors cross-react with viral epitopes in COVID-19

Abstract

Rheumatoid factors (RFs), polyreactive antibodies canonically known to bind two conformational epitopes of IgG Fc, are a hallmark of rheumatoid arthritis but also can arise in other inflammatory conditions and infections. Also, infections may contribute to the development of rheumatoid arthritis and other autoimmune diseases. Recently, RFs only in rheumatoid arthritis were found to bind novel linear IgG epitopes as well as thousands of other rheumatoid arthritis autoantigens. Specific epitopes recognized by infection-induced polyreactive RFs remain undefined but could provide insights into loss of immune tolerance. Here, we identified novel linear IgG epitopes bound by RFs in COVID-19 but not rheumatoid arthritis or other conditions. The main COVID-19 RF was polyreactive, binding two IgG and multiple viral peptides with a tripeptide motif, as well as IgG Fc and SARS-CoV-2 spike proteins. In contrast, a rheumatoid arthritis-specific RF recognized IgG Fc, but not tripeptide motif-containing peptides or spike. Thus, RFs have disease-specific IgG reactivity and distinct polyreactivities that reflect the broader immune response. Moreover, the polyreactivity of a virus-induced RF appears to be attributable to a very short peptide motif. These findings refine our understanding of RFs and provide new insights into how viral infections may contribute to autoimmunity.

1. Introduction

Rheumatoid factors (RFs) are antibodies of any isotype that bind the Fc region of IgG. Initially discovered in 1939 [1], RFs are a diagnostic marker for rheumatoid arthritis with 60–90% sensitivity [2, 3]. However, RFs also are found in people with other inflammatory conditions, including autoimmune diseases like Sjögren’s disease and lupus, smokers, and both acute and chronic infections [4–7]. In total, RFs are detectable in ~4% of “normal” individuals [8] despite being considered a hallmark of rheumatoid arthritis.

Canonically, RFs bind two conformational epitopes in the Fc region of IgG: the Ga determinant (an epitope comprised of loops from the CH2 and CH3 domains) [9] and an epitope in the hinge (a flexible region that connects the CH1 and CH2 domains) [10]. Of note, RFs do not bind native, circulating IgG; rather IgG must be enzymatically cleaved, be antigen-bound, or otherwise be modified to allow RF binding [11]. Recently, citrullinated and homocitrullinated linear IgG epitopes were identified as bound by IgG in rheumatoid arthritis and not in other autoimmune diseases, while a linear native IgG epitope in the hinge region was recognized in Sjögren’s disease [12, 13], suggesting that a distinct profile of IgG epitopes may be recognized by RFs in different autoimmune diseases. More recently, IgG Fc epitopes were demonstrated to be differentially targeted in rheumatoid arthritis, Sjögren’s disease, and healthy donors [14]. However, which, if any, IgG epitopes are uniquely bound by RFs elicited by infection is unknown.

In addition to binding IgG Fc, RFs are commonly polyreactive, binding a variety of self and non-self-antigens [15, 16]. For example, IgM-RFs (RFs of the IgM isotype) from rheumatoid arthritis and periodontitis patients can bind IgG and some oral bacteria [16]. Specific epitopes bound by polyreactive RFs in infection are unknown. However, a variety of infections, including respiratory infections, correlate with rheumatoid arthritis development [17, 18]. Thus, defining infection-induced RF polyreactivity could provide insights into how immune tolerance is lost after an infection, ultimately leading to rheumatoid arthritis.

Unfortunately, studying infection-induced RFs in humans is challenging due to the difficulty of generating a uniform study cohort (i.e., adults infected by a known pathogen at a similar time with the same number of previous exposures). However, in 2020, severe acute respiratory syndrome coronavirus two (SARS-CoV-2) emerged. In addition to causing the devastating coronavirus disease 2019 (COVID-19) pandemic, SARS-CoV-2 created a large cohort of individuals who generated a primary immune response to the same virus at a similar time. Also, RFs develop in 5–20% of COVID-19 patients [19–21]. Thus, COVID-19 presents a unique opportunity to study infection-induced RFs. Finally, since much of the world’s population was infected with SARS-CoV-2, millions of people experienced a rheumatoid arthritis risk factor, i.e., a viral infection, and developed RFs, adding importance to the study of RFs in COVID-19.

In this study, we evaluated antibody binding to IgG and viral epitopes in COVID-19, rheumatoid arthritis, and other conditions to reveal novel and unique features of SARS-CoV-2-induced RF reactivity that have important implications for our understanding of RFs and potentially virus-induced autoimmunity.

2. Methods

2.1. Human Subjects

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board at the University of Wisconsin (UW). Informed consent was obtained for experimentation with human subjects. Serum and clinical information from COVID-19 convalescent subjects (positive SARS-CoV-2 PCR test in the spring of 2020 and ~5 weeks post-symptom resolution) were obtained from the UW COVID-19 Convalescent Biorepository [22], and serum and plasma from subjects with acute COVID-19 (hospitalized at UW Health with a positive SARS-CoV-2 test and COVID-19 symptom onset <3 weeks prior to sample collection from May 2020 until March 2021) were obtained from the UW Carbone Cancer Center Translational Science BioCore BioBank. Serum and clinical information for subjects with rheumatologist-diagnosed rheumatoid arthritis, systemic lupus erythematosus, Sjögren’s disease, and age- and sex-matched controls with no autoimmune or inflammatory disease were obtained from the UW Rheumatology Biorepository [23]. Control and rheumatoid arthritis sera were collected prior to 2019. Lupus and Sjögren’s disease sera were either collected prior to 2020 or a review of the electronic medical record detected no history of COVID-19. Rheumatoid arthritis subjects either had negative clinical testing for anti-cyclic citrullinated peptide (CCP) and RF or had anti-CCP and RF levels greater than 2x the upper limit of normal. Lupus and Sjögren’s disease subjects were limited to those with a positive clinical test for RF.

2.2. High Density Peptide Array

IgA, IgM, and IgG binding to an array that contained the constant regions of the heavy chains of IgG1-4 (Uniprot P01857, P01859, P01860, P01861), tiled as 16 amino acid peptides overlapping by 15 amino acids, was quantified as previously described [24]. Additionally, the array contained peptides from the following proteins tiled in an identical manner: the proteomes of 7 human coronaviruses including SARS-CoV-2, proteomes and spike proteins from other coronaviruses, and proteins from poliovirus, 7 strains of rhinovirus, and human cytomegalovirus [24]. Antibody binding data to viral peptides on this array as well as IgA, IgM, and IgG binding data from an array containing the constant regions of the heavy chains of IgG1-4 tiled as 12 amino acid peptides overlapping by 11 amino acids [12] were used in secondary analysis.

2.3. Enzyme Linked Immunosorbent Assay (ELISA)

For peptide ELISA, Costar 96-well high binding ELISA plates (Corning, Corning, NY) were coated with 5μg/ml streptavidin (Thermo Scientific Pierce, Waltham, MA) in phosphate buffered saline (PBS) overnight at 4°C, washed twice with PBS, and coated with either 0.1μM peptide (Supplemental Table 1) conjugated to biotin at the C terminus (Biomatik, Cambridge, Ontario, Canada or Peptide 2.0, Chantilly, VA) in PBS or PBS alone (uncoated wells) for one hour at room temperature. Plates were washed three times with wash buffer (PBS with 0.2% Tween 20), blocked with blocking solution (5% nonfat dehydrated milk in wash buffer) for at least 2.5 hours at room temperature, and then incubated with serum (diluted 1:100 in blocking solution) or purified antibodies (20ng/ml or as indicated) overnight at 4°C. Plates were then washed four times with wash buffer, incubated for one hour at room temperature with mouse anti-human IgG, goat anti-human IgM, or goat F(ab’)2 anti-human kappa and goat F(ab’)2 anti-human lambda all conjugated to horseradish peroxidase (Southern Biotechnology, Birmingham, AL) diluted 1:5000 in blocking solution. Plates were washed four times with wash buffer, developed with tetramethylbenzidine substrate solution (Thermo Scientific Pierce) and the reaction was stopped using 0.18M sulfuric acid. Endpoint absorbance (450–562nm) was read on a FilterMax F3 spectrophotometer (Molecular Devices, San Jose, CA) and absorbance values from uncoated wells were subtracted from coated wells for each sample as nonspecific binding.

For protein ELISA, human IgG Fc fragment (MilliporeSigma, Burlington, MA) was first depleted of any contaminating intact IgG, light chain, and IgM using streptavidin magnetic beads coated with biotinylated goat F(ab’)2 anti-human kappa, goat F(ab’)2 anti-human lambda, and goat F(ab’)2 anti-human IgM (Southern Biotechnology). Plates (Corning) were coated with 2.5 μg/ml human Fc fragment or SARS-CoV-2 spike protein (GenScript USA, Piscataway, NJ) in PBS or with PBS alone. Plates were washed as described above and blocked overnight at 4°C. Serum or plasma (diluted 1:5000) or purified antibodies in blocking solution were added to the wells and incubated overnight at 4°C. Plates were washed as above, and goat F(ab’)2 anti-human kappa and lambda conjugated to horse radish peroxidase (Southern Biotechnology) at 1:5000 in blocking solution were added to each well for one hour at room temperature. Plates were washed, developed, and absorbance was recorded as described above.

A standard curve was used for some ELISAs. After the above streptavidin and wash steps, wells were either incubated with biotinylated goat F(ab’)2 anti-human kappa and lambda (Southern Biotechnology) diluted 1:5000 in PBS for one hour at room temperature followed by washing, blocking, and incubation with serially diluted human IgG (Bethyl Laboratories, Montgomery, TX) or IgM (Fitzgerald Industries, Acton, MA) overnight at 4°C or incubated with serially diluted biotinylated human IgM and human IgG (Novus Biologicals, Englewood, CO) in PBS for one hour at room temperature. Plates were washed, incubated with detection antibody, developed and absorbance determined as above. Absorbance values for the standards were fit to a four parameter logistic curve to determine antibody concentrations for samples (myassays.com).

2.4. Antibody Purification

Streptavidin sepharose beads (0.25ml of a 50% bead slurry; BioVision, San Francisco, CA or Abcam, Waltham, MA) were washed three times with tris buffered saline (TBS; Fisher Bioreagent, Pittsburgh, PA) and then incubated with 0.25ml of 0.05mM biotinylated peptide in TBS and tumbled at 4°C for two hours. Peptide-labeled beads were resuspended in blocking solution and loaded onto a frit membrane in a 1ml spin column (G-Biosciences, St. Louis, MO) and washed with blocking solution. Columns were stoppered, and sera diluted 1:5 in blocking solution were added. Columns were capped and stored overnight at 4°C. The stopper and cap were removed and serum was allowed to flow from the column. The beads were washed twice with 0.25 ml blocking solution, then rinsed with 0.25ml conditioning buffer (Classic IP Kit, Thermo Scientific Pierce). Adsorbed antibodies were eluted from the beads with 0.1ml elution buffer (100mM glycine pH 2.6) and collected directly into a microcentrifuge tube containing 5 μl of neutralizing buffer (1.5M Tris HCl pH 8.7) by brief centrifugation at 8000xg. The elution step was repeated. The concentration of purified antibodies was approximated by ELISA as described above.

2.5. Statistical analysis

Except for the calculation of the local false discovery rate using MixTwice [25] and R statistical software in the array analysis, statistical analyses were performed using Prism (Graphpad, San Diego, CA). To avoid distributional assumptions, we used nonparametric tests. A Kruskal Wallis test with Dunn’s multiple comparisons test was used to compare >2 groups and a Mann-Whitney test was used to compare 2 groups. P<0.05 was considered significant for both. A Wilcoxon matched-pairs signed rank test was used when two sets of data were paired, and p=0.0625, the lowest possible p value for n=5, was considered significant. Motif analysis was performed with Meme Suite [26].

3. Results

To evaluate RFs in COVID-19, first we quantified RFs by a traditional anti-IgG Fc ELISA. As shown in Supplemental Figure 1, when selecting a cut-off that gives the expected 4% RF positivity for controls [8], ~11% of subjects either acutely ill with COVID-19 or COVID-19 convalescent would be considered RF positive. We then used high density peptide array to quantify antibody binding to all possible linear native peptides along the length of the constant region of the IgG heavy chain in COVID-19 convalescent and control sera. As shown in Figure 1 and Supplemental Figure 2, COVID-19 convalescent sera had several sites of high IgG and IgM binding to the peptides located within the constant region of IgG, whereas IgA binding was overall low. To determine if these regions of IgG bound by antibodies post-COVID-19 were also recognized in rheumatoid arthritis, we performed a secondary analysis of previously obtained IgA, IgM, and IgG binding data to the native peptides in the same region of IgG for subjects with the four serotypes of rheumatoid arthritis: those with a positive anti- CCP test (a marker for anti-citrullinated protein antibodies, the second hallmark autoantibody of rheumatoid arthritis) only, a positive RF test only, neither test positive, and both tests positive [12]. As expected given the binding of rheumatoid arthritis RFs primarily to citrulline- and homocitrulline-containing IgG peptides, for all rheumatoid arthritis serotypes, there was similar antibody binding to the native IgG peptides as controls (Figure 1 and Supplemental Figure 2).

Figure 1. COVID-19 convalescent, but not rheumatoid arthritis, serum IgM and IgG bind to linear native IgG1-derived peptides.

Serum IgA, IgM, and IgG binding to linear native peptides from the constant region of the heavy chain of IgG1 were quantified for subjects 5 weeks post-COVID-19 (n=40) and matched controls (n=20) as well as for rheumatoid arthritis (anti-CCP+RF+, anti-CCP+RF-, anti-CCP-RF+, and anti-CCP-RF- RA; n=12) and matched controls (n=12) by high density peptide array and reported as fold change.

Next, we selected peptides derived from IgG1, the most abundant IgG in humans [27], that were bound at high levels after COVID-19 to confirm by ELISA, specifically peptides starting at amino acid positions 104 (hinge region), 131 (CH2 region), 238 (CH3 region), and 293 (CH3 region). As shown in Figure 2, IgG1-104 was bound at a similar level in the larger set of COVID-19 convalescent subjects as control subjects. However, the other three peptides showed significantly higher binding by COVID-19 convalescent compared to control sera, with IgG1-131 and IgG1-238 highly bound by IgM and IgG1-293 moderately bound by both IgM and IgG. These results identify novel linear epitopes of IgG bound in COVID-19.

Figure 2. IgG1 peptides bound by COVID-19 convalescent sera.

COVID-19 convalescent sera (5 weeks post-COVID-19, n=120) and sera from matched controls (n=54) were assayed by ELISA to determine IgM and IgG binding to four IgG1-derived peptides starting at amino acid positions 104, 131, 238 and 293. Absorbance (abs) values were compared by Mann-Whitney test. Dots indicate individual serum samples, lines indicate medians, and ***p<0.001, ****p<0.0001.

Given the presence of RFs in other conditions, we next evaluated antibody binding to the four IgG1 peptides in rheumatoid arthritis (anti-CCP+RF+ and anti-CCP-RF-), RF+ Sjögren’s disease, RF+ lupus, and smokers. We found that in rheumatoid arthritis, IgM and IgG binding to IgG1-131, IgG1-238 and IgG1-293 was not different from controls, but IgG1-104 had significantly higher IgG binding in anti-CCP+RF+, but not anti-CCP-RF-, rheumatoid arthritis compared with controls (Figure 3). A similar pattern was seen in Sjögren’s disease with significantly elevated IgG binding only to IgG1-104 (Supplemental Figure 3). In lupus and smokers, there was no increased antibody binding to any of the four IgG1 peptides (Supplemental Figures 4 and 5). Taken together, these data suggest that IgG1-131, IgG1-238 and IgG1-293 are unique linear epitopes of IgG that are bound primarily by IgM after COVID-19 but are not bound in other RF+ conditions. In contrast, the peptide at position 104 of IgG1 (hinge region) is bound prominently only in Sjögren’s disease and rheumatoid arthritis.

Figure 3. Only the hinge region peptide is bound by rheumatoid arthritis sera.

Sera from 32 anti-CCP+RF+ rheumatoid arthritis (RA+), 24 anti-CCP-RF- rheumatoid arthritis (RA-), and 20 matched controls were tested by ELISA for IgM or IgG binding to IgG1-derived peptides starting at amino acid positions 104, 131, 238, and 293. Absorbance (abs) values were compared using a Kruskal Wallis test with Dunn’s multiple comparisons test. Dots indicate individual serum samples, lines indicate medians, and **p<0.01.

We then evaluated the timing of COVID-19 RF development. We tested sera from patients with acute COVID-19 by ELISA for IgM against all four IgG1 peptides and IgG against IgG1-104 and IgG1-293. We found that patients with acute COVID-19 had higher IgM binding to IgG1-131, IgG1-238 and IgG1-293 compared to controls that rose significantly by 2–3 weeks post-symptom onset (Supplemental Figure 6). Thus, unique RFs develop soon after SARS-CoV-2 infection.

Next, we evaluated clinical features associated with COVID-19 RFs. We found that COVID-19 convalescent subjects with age >60 years, male sex, and severe disease had higher IgG levels against IgG1-293 (Supplemental Figure 7). In our cohort, males were older than females (mean 49 versus 43 years, p = 0.045 by Mann-Whitney test) and older males were more likely to have severe or critical disease [22]. However, within the moderate severity group, which had an approximately equal distribution of sex and age, IgG anti-IgG1-293 was not different between sexes (p = 0.161, Mann-Whitney test) but was significantly higher in adults >60 years old compared to <40 years old when comparing three groups as in Supplemental Figure 7 (p = 0.007, Kruskal-Wallis test with Dunn’s multiple comparisons test). Thus, age may be the underlying risk factor for anti-IgG1-293. Consistent with this finding, higher levels of IgM anti-IgG1-293 were seen for age >60 years, but not for severe disease or male sex. IgM anti-IgG1-131 and IgG1-238 levels did not vary with age, sex, or disease severity. Taken together, these data suggest that some COVID-19 RFs may be driven by older age.

Having characterized the development and unique reactivity of these IgG1-binding antibodies in COVID-19, we then evaluated potential cross-reactivity. First, we noticed that typically the same subjects had elevated IgM anti-IgG1-131 and anti-IgG1-238 with limited overlap with IgM anti-IgG1-293 (Figure 4A). Further, both IgG1-131 and IgG1-238 had a positively charged amino acid in position one and an aspartic acid in position two (Figure 4B). Hypothesizing that a single antibody might bind to both IgG1-131 and IgG1-238, but not IgG1-293, we purified antibodies that bound to IgG1-131 or IgG1-238 from COVID-19 convalescent sera and evaluated their binding to IgG1-131 and IgG1-238, as well as to IgG1-293, homocitrullinated (J) IgG1-219J and IgG1-289J (two peptides highly bound by IgG in rheumatoid arthritis [13] with abundant antibody levels in serum), and a SARS-CoV-2 membrane peptide that is highly bound in COVID-19 [24] as a negative control. As shown in Figure 4C, IgM purified by binding to either IgG1-131 or IgG1-238 recognized both IgG1-131 and IgG1-238, but not any other peptide tested. We then performed similar experiments with antibodies purified from COVID-19 convalescent sera that bound to IgG1-293 and antibodies purified from rheumatoid arthritis sera that bound to IgG1-219J and IgG1-289J (Figure 4C). IgM anti-IgG1-293 did not bind to any of the IgG1 peptides apart from IgG1-293, but did bind the SARS-CoV-2 membrane peptide, potentially due to sequence similarities (Figure 4B). Rheumatoid arthritis IgG anti-219J and anti-289J cross-reacted to some extent with each other, as expected [13, 28], but not with any other peptide. Together, these data suggest that two cross-reactive antibodies develop in COVID-19 that bind native linear IgG1 peptides, but not the homocitrullinated IgG1 peptides bound by RFs in rheumatoid arthritis.

Figure 4. Cross-reactivity among antibodies that bind IgG1 epitopes.

A. Binding of IgM to IgG1-131, IgG1-238, and IgG1-293 that was quantified by ELISA for 120 COVID-19 convalescent subjects was plotted to compare binding for each peptide for each subject. B. Comparison of peptide sequences (orange: same amino acid; green: related amino acid; blue: homocitrulline; gray: neighboring serine or glycine). C. Antibodies purified based on binding to IgG1-131, IgG1-238, IgG1-293, IgG1-219J, and IgG1-289J were evaluated by ELISA for binding to IgG1-131, IgG1-238, IgG1-293, IgG1-219J, IgG1-289J, and a SARS-CoV-2 membrane (mem) peptide (binding level of indicated purified antibody on y axis, antigen on X axis). Groups were compared to the first column in each graph by Kruskal-Wallis test with Dunn’s multiple comparisons test. Dots represent individual experiments using 5 unique subjects, boxes indicate mean +/− SEM, and *p<0.05, **p<0.01.

Next, we evaluated potential cross-recognition of the purified anti-IgG1-131/238 antibodies with viral peptides. Since the peptide array also included peptides derived from several viruses including SARS-CoV-2 [24], we conducted a secondary analysis to identify viral peptides bound by IgM in those subjects with high levels of IgM recognizing IgG1-131 or IgG1-238. We divided the COVID-convalescent subjects into two groups: subjects with high IgM binding to IgG1-131 or IgG1-238 (>1 standard deviation above mean IgM binding to all IgG1-derived peptides on the array) and subjects with low binding (<1 standard deviation above the mean). We then evaluated which array peptides were bound >5x more by IgM in the high binding group (n=10) as compared to the low binding group (n=30). We excluded array peptides with increased binding by IgG (>1.5 fold) to reduce the possibility of secondary detection of IgM-RFs bound to IgG bound to viral peptides on the array. We found that COVID-19 convalescent sera with high levels of IgM recognizing IgG1-131 or IgG1-238 also had IgM that bound 18 peptides derived from 15 proteins from 11 viruses, including 2 peptides in SARS-CoV-2 spike, at a threshold of >5x greater than the low binding group (Supplemental Table 2, locFDR<0.1). Moreover, these 18 peptides had a strong motif with an aspartic acid (D) at position 2 in every peptide (Figure 5A). Notably, despite the uniform presence of aspartic acid in position 2, not all array peptides with this “second D” are highly bound by COVID-19 convalescent sera with high levels of IgM anti-IgG1-131 or IgG1-238, but second D motif peptides are bound more overall (Supplemental Figure 8).

Figure 5. Cross-reactive binding of COVID-19 antibodies to IgG1 and viral peptides.

A. Motif for 18 peptides bound at a high level by IgM from COVID-19 convalescent sera that also had high levels of IgM that bound IgG1-131 or IgG1-238. B. Antibodies that bound to IgG1-131 or IgG1-238 (B) or IgG1-293, IgG1-219J, or IgG1-289J (C) were purified and evaluated by ELISA for binding to the IgG1 peptide (striped column), viral peptides with the second D motif (SARS-CoV-2 orf1ab-32, SARS-CoV-2 spike-570, poliovirus 1 polyprotein-112), and viral peptides without the motif (SARS-CoV-2 nucleocapsid (nucl)-390 and spike-1253). The binding level of the indicated purified antibody is on the y axis and the antigens are on X axis. Groups were compared to the first column in each graph by Kruskal-Wallis test with Dunn’s multiple comparisons test. For (B) and (C): orange shading indicates peptides with a second D motif; dots represent individual experiments with individual subjects (n=5); boxes indicate mean +/− SEM; and *p<0.05, **p<0.01 and ***p<0.001.

Given the binding of multiple peptides with a similar motif by subjects with IgM anti-IgG1-131/238, we evaluated cross-reactivity with three virus-derived peptides with the second D motif. We purified antibodies that bound IgG1-131 and IgG1-238 from COVID-19 convalescent sera and evaluated binding to SARS-CoV-2 ORF1ab-32, SARS-CoV-2 spike-570, and polyprotein from poliovirus 1 (polio)-112 (all with the second D motif) as well as two SARS-CoV-2 peptides previously found to be highly bound by antibodies in COVID-19 without the motif (nucleocapsid-390 and spike-1253) [24]. We found that antibodies that bound IgG1-131 and IgG1-238 also bound the three second D motif viral peptides but not the viral peptides without the motif (Figure 5B). This cross-reactivity was limited to anti-IgG1-131/238, since antibodies purified based on binding to IgG1-293, IgG1-219J, and IgG1-289J (all peptides without the second D motif) generally did not bind any of the viral peptides (Figure 5C).

Next, we evaluated if anti-IgG1-131 is a true RF, i.e., able to bind IgG Fc protein. As shown in Figure 6A, with increasing concentration of anti-IgG1-131 purified from five COVID-19 convalescent subjects, increased binding to IgG Fc was detected. Further, anti-IgG1-131 purified from each subject bound IgG Fc more than a negative control peptide, although binding was variable with an average of only 13% of the level of binding to the IgG1-131 peptide used to purify the antibody (Figure 6B). To compare with an antibody that binds a linear IgG1 epitope in rheumatoid arthritis, we purified anti-IgG1-219J from five rheumatoid arthritis subjects. As expected, with increasing concentration of purified anti-IgG1-219J, increased binding to IgG Fc was detected, and anti-IgG1-219J bound IgG Fc more than the negative control peptide (Figure 6A and 6B). However, at the same concentration, anti-IgG1-219J bound IgG Fc more consistently, on average at 51% of the level of binding to IgG1-219J peptide (Figure 6B).

Figure 6. Cross-reactive antibodies in COVID-19 bind both IgG Fc and spike protein.

A. Increasing concentrations of anti-IgG1-131 purified from 5 COVID-19 convalescent subjects or anti-219J purified from 5 rheumatoid arthritis subjects were used to assess binding to the peptide against which the antibody was purified, IgG Fc, or a negative control peptide (spike-1253, based on Figure 5) by ELISA. Mean +/− SEM are depicted. B. For the 333 ng/ml concentration of the purified antibodies, binding to the negative control peptide and IgG Fc were divided by binding to the peptide used to purify the antibody with percent binding compared by Wilcoxon matched-pairs signed rank test. Each symbol represents a different subject and #p=0.0625. C and D. Antibodies were purified based on binding to spike-570, IgG1-131, IgG1-219J, and mem-8 peptides (n=5). Binding of each purified antibody (250 ng/ml) to SARS-CoV-2 spike protein and IgG Fc was determined by ELISA. C. Results are displayed in a scatter plot. D. Binding to spike versus IgG Fc was compared by Wilcoxon matched-pairs signed rank test (#p=0.0625).

Given the polyreactivity of the anti-IgG1-131 antibodies, including with the spike-570 peptide (Figure 5), this antibody may have been generated in a polyreactive response to SARS-CoV-2 spike, whereas a rheumatoid arthritis RF that was generated prior to the emergence of COVID-19, would not have formed in this manner. To evaluate this possibility further, we determined the comparative abilities of antibodies purified based on binding to spike-570, IgG1-131, IgG1-219J, and mem-8 to bind SARS-CoV-2 spike protein and IgG Fc. As expected, anti-mem-8 had minimal binding to either protein (Figure 6C). Anti-spike-570 and anti-IgG1-131 generally clustered together, as expected given their cross-reactivity, with all ten antibodies binding spike protein more than IgG Fc protein (Figures 6C and 6D). However, antibodies purified based on binding to IgG1-131 bound spike and IgG Fc relatively similarly, whereas antibodies purified based on binding to spike-570 had more distinctly increased binding to spike than IgG Fc (Figure 6D). In contrast, all five rheumatoid arthritis anti-IgG1-219J antibodies bound IgG Fc more than spike (Figures 6C and 6D). Taken together, the data in Figure 6 suggest that COVID-19 induces polyreactive antibodies that are both RFs and anti-SARS-CoV-2 spike protein antibodies, whereas rheumatoid arthritis antibodies that bind linear IgG epitopes have relatively high RF activity and minimal binding to spike protein.

Finally, given the binding of RFs to citrulline- and homocitrulline-containing IgG epitopes and a dominant antibody response overall to citrullinated antigens in rheumatoid arthritis [12, 13], we evaluated IgA, IgM, and IgG binding to the IgG1-131 peptide in the native form and in a form in which arginines were replaced by citrullines (B) using five week COVID-19 convalescent, rheumatoid arthritis, and age- and sex-matched control sera. As shown in Figure 7A, we detected high IgM binding in COVID-19 convalescent sera, but not in rheumatoid arthritis sera, to both IgG1-131 and IgG1-131B with little difference between the peptides. In contrast, there was high IgG binding in rheumatoid arthritis sera, and not COVID-19 convalescent sera, to IgG1-131B only (Figure 7B). There was no increased IgA binding to either peptide in COVID-19 or rheumatoid arthritis. Thus, the presence of citrulline does not affect COVID-19 IgM-RF factor binding, but as expected, drives high IgG-RF binding in rheumatoid arthritis.

Figure 7. Citrulline does not alter IgM binding to the IgG1-131 peptide post-COVID-19 but is essential for IgG binding in rheumatoid arthritis.

IgM (A) and IgG (B) binding to IgG1-131 in the native form and in a form in which the arginine was replaced by citrulline (represented as B) was evaluated by ELISA using sera from five week COVID-19 convalescent, rheumatoid arthritis (RA), and matched control subjects (n=20). All conditions were compared to control serum binding to the native IgG1-131 peptide by Kruskal-Wallis test with Dunn’s multiple comparisons test (***p<0.001, ****p<0.0001), and lines indicate medians. Also, IgM binding for post-COVID-19 sera and IgG binding for rheumatoid arthritis sera were plotted to compare the binding for each subject to the IgG1-131 and IgG-131B peptides.

4. Discussion

In this manuscript, we report that SARS-CoV-2 infection is associated with the presence of unique cross-reactive RFs that recognize both novel linear IgG epitopes and multiple viral antigens. While RFs are known to develop in SARS-CoV-2 and other infections [5–7, 20, 21], specific IgG epitopes have not been characterized. Here, we identified three linear epitopes of IgG that are bound post-COVID-19. IgG1-131 and IgG1-238 appear to be novel IgG epitopes for RFs, with purified IgM antibodies able to co-recognize both of these epitopes as well as IgG Fc at a relatively low level, consistent with polyreactive IgM-RFs in general [29, 30]. Part of the IgG1-293 peptide was identified in an array of seven amino acid linear peptides derived from the CH3 region of IgG bound by rheumatoid arthritis IgM-RFs [31]. Interestingly, the seven amino acids identified as bound by rheumatoid arthritis RFs, WQQGNVF, are not the amino acids in our IgG1-293 peptide that are similar to the SARS-CoV-2 membrane peptide. We did not detect binding to IgG1-293 in rheumatoid arthritis, perhaps due to differences in conformation of their seven amino acid peptide [32] versus our longer peptide. Indeed, we did not detect binding to any of the three COVID-19-related linear IgG epitopes in rheumatoid arthritis, Sjögren’s disease, lupus, or smokers, suggesting that these peptide epitopes are unique to COVID-19. In contrast, IgG-RFs in anti-CCP+RF+ rheumatoid arthritis uniquely bind linear citrullinated and homocitrullinated IgG epitopes [12, 13], including the citrulline-containing version of the IgG1-131 peptide (Figure 7). Also, rheumatoid arthritis and Sjögren’s disease IgG recognizes IgG1-104, consistent with binding by Sjögren’s disease IgG to an overlapping hinge peptide (IgG1-109) [13]. However, in that study, rheumatoid arthritis IgG had only a trend towards increased binding to IgG1-109 [13], a discrepancy likely due to the slightly different peptide and increased sample size in the current study. Together, these findings support the idea that RFs with distinct reactivities can develop in different conditions, a phenomenon that could be leveraged to design more informative diagnostic RF assays.

Of the COVID-19 RFs that we identified, anti-IgG1-131/238 is polyreactive, and anti-IgG1-293 is at least cross-reactive. RF polyreactivity is well-established [15, 16] and has been attributed generally to antigenic similarity [16] with specific epitopes undefined. Here, we identified specific peptides that share a tripeptide motif as bound by polyreactive RFs post-COVID-19. Polyreactivity driven by recognition of a motif (D/E-W-D/E-Y-S/G) also has been described for a subset of pathogenic anti-DNA antibodies in lupus [33] and for some anti-citrullinated protein antibodies in rheumatoid arthritis (citrulline-glycine, citrulline-serine, or homocitrulline-glycine pairs) [12, 28]. Polyreactivity due to motifs, at least partially independent of protein identity, is a substantial departure from how antigenic similarity, as well as molecular mimicry, typically refers to a small number of similar proteins [34]. For example, peptides from myelin basic protein and a hepatitis B protein that both contain YGSLPQ can drive encephalitis, leading to the conclusion of molecular mimicry between these two proteins [35]. However, a correlation between hepatitis B infection and multiple sclerosis has not been observed, and YGSLPQ is present in thousands of proteins in fish, birds, reptiles, insects, worms, plants, fungi, bacteria, viruses, and mammals. Thus, a polyreactive antibody that binds those two antigens due to the motif is more plausible than hepatitis B driving multiple sclerosis due to molecular mimicry between two proteins. Our finding of a motif-driven polyreactive RF post-COVID-19 not only adds to this growing body of evidence that polyreactivity due to epitope motifs may drive autoreactivity more than protein identity, but also appears to be the first study to link virus-induced motif-driven polyreactivity to an autoantigen, challenging the traditional view of molecular mimicry and providing important new insights into how viruses contribute to autoreactivity and potentially ultimately autoimmune disease.

Although the antibodies that bind peptides with the second D motif also bind full-length proteins in which the motif would be present in a variety of positions, the strongest binding was to peptides with the motif in positions 1–3, suggesting an antibody response driven by a protein fragment. While SARS-CoV-2 proteases could generate some of these peptides, neutrophil- and macrophage-derived cathepsin G generates peptides with a negatively charged amino acid in the P2’ position [36]. Thus, perhaps early recruitment of innate immune cells provides proteases that generate proteolytic fragments with the maximally stimulatory position of the motif to induce the observed antibody response.

The development of post-COVID-19 RFs is more likely due to a polyreactive IgM antibody response to tripeptide motif-containing peptides derived from SARS-CoV-2 than IgG. Consistent with this idea, anti-IgG1-131 binds SARS-CoV-2 spike protein more than IgG Fc, and the post-COVID-19 RFs have similar kinetics and demographic correlations as other anti-SARS-CoV-2 antibodies [22, 37]. Polyreactive antibodies are common in infection, both in the initial IgM response and the later IgG response, and broadly neutralizing antibodies, i.e., antibodies that neutralize similar strains of the same virus, can be polyreactive [38–40]. However, the cost of polyreactivity can be self-reactivity, and an inability for some people to generate broadly neutralizing antibodies against some viruses may be due to peripheral tolerance mechanisms [38–40]. Tolerance in the T cell compartment [41] likely explains the absence of IgG-RFs that bind IgG1-131/238 post-COVID-19, whereas the unique RFs in rheumatoid arthritis are IgG-RFs, consistent with known loss of T cell tolerance for IgG in rheumatoid arthritis [42]. Of note, IgG cross-reactive with SARS-CoV-2 and self-antigens has been reported in COVID-19, but primarily in critically ill patients, suggesting that severe inflammation can overcome additional tolerance mechanisms [43].

Like the distinct IgG epitopes bound by RFs in different conditions, their polyreactivity also differs. Whereas one subset of COVID-19 RFs binds second D-containing peptides and spike protein, rheumatoid arthritis RFs that bind IgG1-219J do not recognize those antigens, but do bind the thousands of citrullinated and homocitrullinated antigens recognized by anti-citrullinated protein antibodies [13, 28]. Similarly, antibodies in lupus, Sjögren’s disease, and rheumatoid arthritis do not bind the second D containing IgG1-131 or IgG1-238 peptides, despite extensive polyreactivity to other antigens [28, 44, 45]. Interestingly, similar to the distinct RFs in COVID-19 versus RA, anti-nuclear antibodies in COVID-19 are distinct from those in lupus and Sjögren’s disease, given their lack of reactivity against extractable nuclear antigens in COVID-19 [43, 46]. These findings, in addition to our findings of distinct IgG epitopes, support a model in which RFs are not a single polyreactive entity in multiple diseases, but rather disease-specific antibodies with polyreactivity that reflects the broader immune response.

There are several caveats to our study. Related to the technology, the array contained only short native peptides derived from viruses and human IgG heavy chain. We did not evaluate conformational, post-translationally modified, or human epitopes apart from IgG. Also, results for IgM, IgA, and IgG binding cannot be directly compared given likely differences in affinity. Related to the human subjects, this study was performed in a single cohort of patients. Moreover, we did not have pre-COVID-19 samples, so we compared COVID-19 subjects to age- and sex-matched controls. While this comparison is commonly used, we cannot exclude with 100% certainty the possibility that the COVID-19 RFs developed prior to COVID-19. However, we believe that COVID-19 RFs developing prior to COVID-19 is very unlikely given the absence of high levels of these RFs in age- and sex-matched controls. Also, since few COVID-19 subjects generated high levels of anti-IgG1-293, we were limited in further evaluation of this antibody’s cross-reactivity. Finally, the only infectious disease we evaluated was COVID-19, so it remains unknown if the RFs that we identified are unique to SARS-CoV-2 or are elicited by other pathogens in a primary immune response. In support of a lack of specificity to SARS-CoV-2, antibody autoreactivity observed in patients with acute respiratory distress syndrome from bacterial pneumonia was similar to critical COVID-19 [43]. In contrast, and perhaps more relevant for our study of primarily moderate COVID-19, four times more COVID-19 patients developed anti-nuclear antibodies compared to other causes of fever and/or pneumonia [46]. However, in this study, as well as in our study, whether self-reactivity and specific RFs are unique to SARS-CoV-2 versus polyreactivity in a primary immune response in general is not known. Thus, future studies are needed to define the full repertoire of self-antigens bound by COVID-19 RFs, to identify additional motifs such as what may underlie anti-IgG1-293 reactivity, and to use additional patient cohorts to explore SARS-CoV-2 versus general primary immune response phenomena.

Despite these limitations, we present novel, unique, polyreactive RFs in COVID-19. The implications of these RFs are unknown. They could be simply polyreactive antiviral antibodies with little functional impact like other promiscuous autoantibodies in COVID-19 [47]. However, RFs are typically polyreactive and have been shown to bind virus-antibody immune complexes to enhance neutralization by complement and uptake by macrophages [48, 49], suggesting a possible beneficial role. Further, given their reactivity with spike protein, COVID-19 RFs might directly bind and neutralize SARS-CoV-2. Thus, an important future direction for this work is to evaluate the capacity for COVID-19 RFs to neutralize live SARS-CoV-2 virus and improve outcomes in COVID-19.

On the other hand, COVID-19 RFs could be harmful. Infections, including SARS-CoV-2, not only induce RFs but are risk factors for rheumatoid arthritis, although data are mixed for SARS-CoV-2 [5, 6, 17, 18, 50–52]. The modest relationship between COVID-19 and new rheumatoid arthritis may be due to the relatively short amount of time since SARS-CoV-2 emergence as well as the likelihood that most individuals regain immune tolerance. However, in a small percentage of people, loss of tolerance for IgG could be a first step towards rheumatoid arthritis [53] with a transient rise in COVID-19 RFs, leading to increased and more pathogenic autoantibodies via changes in polyreactivity and/or epitope spreading over many years, a typical chronology for autoimmune disease [30]. For example, reactivity could shift from the second D motif at amino acids 131–133 of IgG1 early post-COVID-19 to the serine-citrulline motif at position 137–138 (should the IgG molecule become citrullinated) on a path towards rheumatoid arthritis. Alternatively, for individuals who previously developed anti-citrullinated protein antibodies, which typically arise before RFs in pre-clinical rheumatoid arthritis [54], SARS-CoV-2 could trigger the rise in RFs associated with more imminent rheumatoid arthritis, raising the possibility that some long COVID patients with arthralgias and signs of inflammation [19, 55] could have pre-clinical rheumatoid arthritis. Thus, it will be extremely important to follow COVID-19 subjects longitudinally to evaluate for resolution, persistence, and reactivity changes in their autoantibodies as well as the development of rheumatoid arthritis. Future studies are also needed to evaluate RFs and rheumatoid arthritis development in long COVID as well as after SARS-CoV-2 vaccination given the ability of vaccines to induce RFs [6], the potential for polyreactive anti-SARS-CoV-2 spike antibodies, and case reports of new rheumatoid arthritis after COVID-19 vaccination [56, 57]. Such studies will provide important insights into how immune tolerance is maintained or lost post-infection.

• COVID-19 rheumatoid factors (RFs) bind IgG epitopes not bound in rheumatoid arthritis or other conditions.

• The main COVID-19 RF was polyreactive, binding two IgG and multiple viral peptides with a tripeptide motif, as well as IgG Fc and SARS-CoV-2 spike proteins.

• A rheumatoid arthritis-specific RF recognizes IgG Fc, but not motif-containing peptides or spike.

• Thus, RFs have disease-specific IgG Fc reactivity and distinct polyreactivities that reflect the broader immune response.

• Moreover, a very short motif appears to underlie the polyreactivity of a virus-induced RF.