Diverse Functional Autoantibodies in Patients with COVID-19

Abstract

COVID-19 manifests with a wide spectrum of clinical phenotypes that are characterized by exaggerated and misdirected host immune responses^1–6. While pathological innate immune activation is well documented in severe disease¹, the impact of autoantibodies on disease progression is less defined. Here, we used a high-throughput autoantibody (AAb) discovery technique called Rapid Extracellular Antigen Profiling (REAP)⁷ to screen a cohort of 194 SARS-CoV-2 infected COVID-19 patients and healthcare workers for autoantibodies against 2,770 extracellular and secreted proteins (the “exoproteome”). We found that COVID-19 patients exhibit dramatic increases in autoantibody reactivities compared to uninfected controls, with a high prevalence of autoantibodies against immunomodulatory proteins including cytokines, chemokines, complement components, and cell surface proteins. We established that these autoantibodies perturb immune function and impair virological control by inhibiting immunoreceptor signaling and by altering peripheral immune cell composition, and found that murine surrogates of these autoantibodies exacerbate disease severity in a mouse model of SARS-CoV-2 infection. Analysis of autoantibodies against tissue-associated antigens revealed associations with specific clinical characteristics and disease severity. In summary, these findings implicate a pathological role for exoproteome-directed autoantibodies in COVID-19 with diverse impacts on immune functionality and associations with clinical outcomes.

Humoral immunity plays dichotomous roles in COVID-19. Although neutralizing antibodies afford protection against SARS-CoV-2 infection^8,9, growing evidence suggests that dysregulated humoral immunity also contributes to the characteristic immunopathology of COVID-19^10–16. Recent reports have identified isolated autoantibody reactivities in COVID-19 patients, including those that are characteristic of systemic autoimmune diseases^11–15. Importantly, some autoantibodies, particularly neutralizing antibodies against IFN-I, appear to directly contribute to COVID-19 pathophysiology by antagonizing innate antiviral responses^11,12. While striking examples of disease-modifying autoantibody responses have been described, the full breadth of autoantibody reactivities in COVID-19 and their immunological and clinical impacts remain undetermined.

A particularly important class of autoantibodies are those that target extracellular and secreted proteins (the “exoproteome”). Exoproteome-targeting autoantibodies can exert a wide range of functional effects. such as perturbation of cell signaling (as with the case of anti-IFN-I autoantibodies^11,12) and targeted killing of specific cell populations via Fc receptors (FcR) and/or complement. We thus sought to identify functional autoantibody responses in COVID-19 patients by screening for autoantibody reactivities against the human exoproteome.

Widespread AAb elevation in COVID-19

To discover functional autoantibodies that could influence COVID-19 outcomes, we used a high-throughput autoantibody discovery method called Rapid Extracellular Antigen Profiling (REAP)⁷. REAP enables highly multiplexed detection of antibody reactivities through biopanning of patient IgG against a genetically-barcoded library of 2,770 human extracellular proteins displayed on the surface of yeast, converting an antibody:antigen binding event into a quantitative sequencing readout (“REAP Score”) based on the enrichment of each protein’s barcodes (Extended Data Fig. 1a). To allow for detection of antibodies against coronavirus proteins, we additionally included the receptor binding domain (RBD) of SARS-CoV-2 and other common coronaviruses in the library (full antigen list in Supplementary Table 1).

We used REAP to screen samples from SARS-CoV-2 infected patients who were prospectively followed as part of the Yale Implementing Medical and Public Health Action Against Coronavirus CT (IMPACT) study (Extended Data Fig. 1b). This cohort includes 172 patients seen at Yale-New Haven Hospital with a range of clinical severities (as reported previously¹) and 22 healthcare workers (HCWs) with mild illness or asymptomatic infection. Longitudinal samples were screened for a subset of the cohort. As uninfected controls, we screened 30 HCWs who tested negative for SARS-CoV-2 by RT-qPCR throughout their follow-up period in the IMPACT study. Patient demographics can be found in Extended Data Table 1. To validate the performance of REAP, we compared SARS-CoV-2 RBD REAP reactivity to that of ELISA (Extended Data Fig. 1c) and compared IL-6R REAP reactivity in patients who received anti-IL-6R therapeutic antibodies to those who did not (Extended Data Fig. 1d). We found strong concordance between REAP scores for SARS-CoV-2 RBD or IL-6R and SARS-CoV-2 RBD ELISA positivity or anti-IL-6R treatment respectively.

Next, we examined the total degree of autoreactivity in patients by quantifying the number of autoantibodies at different REAP score thresholds. Irrespective of the REAP score cutoff used, COVID-19 patients had a greater number of reactivities compared to controls, and the highest scoring reactivities were preferentially enriched in severe patients (Fig. 1a,b, Extended Data Fig. 2a). Of note, there was not a statistically significant difference in days from symptom onset (DFSO) between severe and moderate COVID-19 patients (Extended Data Fig. 2b), suggesting that the effects of temporal confounding were minimal. Given previously reported sex differences in the immune response to SARS-CoV-2¹⁷, we compared the number of autoantibody reactivities between male and female patients and found no significant differences in reactivity numbers at any score cutoff (Extended Data Fig. 2c). Finally, compared to REAP profiles of SLE and autoimmune polyglandular syndrome type 1 (APS-1; APECED) patients, COVID-19 patients had greater numbers of reactivities compared to SLE, but fewer numbers of reactivities compared to APECED (Extended Data Fig. 2d). SLE and APECED patient demographics and clinical characteristics have been previously described⁷. Altogether, these results indicate that autoantibodies that target the exoproteome are elevated in COVID-19.

Figure 1: Immune-targeting autoantibodies are elevated in COVID-19.

a, Heatmap of immune-related protein REAP scores across all patient samples stratified by disease severity. b,c, Number of positive (REAP score ≥ 6) total reactivities (b) and immune-targeting reactivities (c) in severe disease (n=66), moderate disease (n=160), mild or asymptomatic disease (mild/asym.; n=36), and uninfected (negative; n=54) samples. Dashed lines indicate medians and dotted lines indicate first and third quartiles. d, Score plot of principal component analysis performed on immune-targeting reactivities in hospitalized COVID-19 patient samples (n=226) colored by clinical score (CS). e, Loadings for the first and second principal components from d in descending order. In b-d, n values include longitudinal samples from the same patient. Significance in b,c was determined using a linear mixed model with correction for multiple comparisons (see Methods).

To investigate the temporal nature of these reactivities relative to COVID-19, we assessed longitudinal REAP score dynamics. Although definitive assignment was not possible due to lack of pre-infection samples, we inferred reactivities as “likely pre-existing”, “newly acquired”, or “waning” based on REAP score trajectories plotted against DFSO and anti-spike S1 IgG development. We found that some reactivities were present with high REAP scores within 10 DFSO and prior to the development of anti-spike S1 IgG responses, suggesting that they were likely pre-existing (Extended Data Fig. 3a,d). Others increased in score and tracked together with increasing anti-spike S1 IgG levels, indicating that they were newly acquired post-infection (Extended Data Fig. 3b,e). Finally, some reactivities decreased in REAP score over time while anti-spike S1 IgG increased or remained high, which suggests waning autoantibody titers (Extended Data Fig. 3c,f).

To further explore potential cellular sources of the elevated autoantibody reactivities in COVID-19 patients, we examined B cell phenotypes in peripheral blood mononuclear cells (PBMC) matching the REAP plasma samples. Similar to previous reports¹⁰, we found that extrafollicular DN B cells are expanded in moderate and severe COVID-19 patients compared to uninfected controls (Extended Data Fig. 2f).

AAbs target immune-related proteins

Analysis of specific reactivities detected by REAP indicated that autoantibodies targeting immune-related proteins were elevated in severe COVID-19 patients (Fig. 1a,c, Extended Data Fig. 2e). These proteins included those involved in lymphocyte function/activation, leukocyte trafficking, type I and type III interferon responses, type II immunity, and the acute phase response. Confirming a recent report from Bastard et al.¹¹, we identified anti-IFN-I autoantibodies in 5.2% of hospitalized COVID-19 patients. Using ELISA, we orthogonally validated a subset of 22 autoantibodies that target cytokines, chemokines, growth factors, complement factors, and cell surface proteins (Extended Data Fig. 4a-d). These results demonstrate that COVID-19 patients possess autoantibodies that may affect a wide range of immunological functions.

To uncover the low-dimensional features in immune-targeting reactivities, we performed principal component analysis (PCA; Fig. 1d,e, Extended Data Fig. 5a). We found that the first principal component (PC1) was primarily composed of autoantibody reactivities against type I interferons (Fig. 1e) and separated a distinct cluster of samples with severe disease phenotype (Fig. 1d), consistent with Bastard et al.¹¹. The next largest principal component (PC2) was composed of autoantibody reactivities against cytokines, chemokines, and the type III interferons, IFN-λ2 and IFN-λ3 (Fig. 1e). Notably, COVID-19 disease severity was a significant predictor of PC2 score (Extended Data Fig. 5b,c). Altogether, these findings suggest that autoantibodies against cytokines and chemokines may contribute to disease severity in COVID-19.

Virological/immunological AAb effects

Because autoantibodies in patients may influence circulating concentrations of their target proteins, we examined the plasma concentrations of cytokines and chemokines in patients with autoantibodies against these proteins. In some cases, autoantibodies were associated with apparent increases in their autoantigen targets (Extended Data Fig. 6b,f,j,m), whereas in other cases they correlated with apparent decreases (Extended Data Fig. 6k,l).

To more directly assess potential immunomodulatory effects of cytokine/chemokine targeting autoantibodies in COVID-19 patients, we assessed the in vitro activity of selected autoantibodies. We found that IgG from patients with anti-GM-CSF, anti-CXCL1, or anti-CXCL7 autoantibodies could neutralize signaling of GM-CSF, CXCL1, and CXCL7 respectively (Fig. 2a,b). In addition, we found that serum from patients with anti-CD38 or anti-CD3ε autoantibodies led to increased macrophage antibody dependent cellular phagocytosis (ADCP) of Raji B cells or Jurkat T cells respectively (Fig. 2c, Extended Data Fig. 4e,f). Altogether, these results demonstrate that immune-targeting autoantibodies in COVID-19 patients can directly inhibit the activity of cytokines/chemokines and engage FcR effector functions that could lead to immune cell depletions in affected patients.

Figure 2: Immune-targeting autoantibodies in COVID-19 patients have functional effects.

a, GM-CSF signaling assay performed with anti-GM-CSF autoantibody positive COVID-19 patient and uninfected healthcare worker (HCW) IgG. Results are averages of technical duplicates from one experiment. b, CXCL1 and CXCL7 signaling assays performed with anti-CXCL1 or anti-CXCL7 autoantibody positive COVID-19 patient and HCW IgG. Results are averages of 3 technical replicates or duplicates. c, Macrophage phagocytosis assay performed with Raji or Jurkat cells, anti-CD38 or anti-CD3ε autoantibody positive COVID-19 patient serum, anti-CD38 or anti-CD3ε monoclonal antibodies, and HCW serum (n=1 for all groups). Technical replicates are shown. Results in b,c are representative of two independent experiments. d, Longitudinal comparisons of SARS-CoV-2 viral load between patients with and without anti-IFN-I autoantibodies. Linear regressions (solid lines) and 95% confidence bands (shaded areas) for each group are displayed. n values include longitudinal samples from the same patient. e-g, Average percent B cells (e), relative proportions of classical, intermediate (int.), and nonclassical (n.c.) monocytes (f), and average percent CD4⁺ T cells (g) among peripheral leukocytes in HCWs and COVID-19 patients stratified by disease severity and positive REAP reactivity (AAb⁺; REAP score ≥ 2) against B cell displayed proteins (CD38, FcμR, FCRL3; e), proteins preferentially displayed on classical/intermediate monocyte (CCR2, CCRL2, FFAR4, SYND4, CPAMD8; f), and CD3ε (g) respectively. Data from e,g were presented as boxplots with the first quartile, median, third quartile, whiskers (minimum/maximum value within the first or third quartile ± 1.5 times the interquartile range), and individual data points indicated. Significance was determined using a generalized linear mixed model (d; see Methods) or a two-sided Wilcoxon rank-sum test (e). In b,c,e,f,g, n values indicate samples from unique patients. All error bars represent standard deviation.

To investigate the potential virological effects of cytokine/chemokine targeting autoantibodies, we examined a subset of COVID-19 patients with anti-IFN-I autoantibodies. Consistent with Bastard et al.¹¹, we found that these autoantibodies can neutralize IFN-I signaling activity in vitro (Extended Data Fig. 4g,h). To further assess their functional impacts, we compared longitudinal composite viral loads in patients who had anti-IFN-I autoantibodies to those who did not. After controlling for the contributions of age, sex, DFSO, and anti-SARS-CoV-2 IgG responses, patients with anti-IFN-I autoantibodies had significantly increased average viral loads relative to patients without anti-IFN-I autoantibodies (Fig. 2d, Extended Data Fig. 4i). We also found that patients with anti-IFN-I autoantibodies had extended hospital admission durations (Extended Data Fig. 4j). These results indicate that anti-IFN-I autoantibodies impair virological clearance in COVID-19 patients.

To investigate the in vivo effects of autoantibodies against immune cell surface proteins in COVID-19, we looked for associations between these autoantibodies and blood leukocyte composition. First, we focused on two groups of antigens: those expressed on B cells (CD38, FcμR, FcRL3) and those preferentially expressed on classical/intermediate monocytes (CCR2, CCRL2, FFAR4, SYND4, CPAMD8; identified in a public RNA-seq dataset¹⁸). We found that patients with autoantibodies against B cell or classical/intermediate monocyte antigens had lower frequencies of B cells (Fig. 2e) and anti-SARS-CoV-2 RBD IgM (Extended Data Fig. 7a) or classical/intermediate monocytes (Fig. 2f, Extended Data Fig. 7b) respectively compared to autoantibody negative severity-matched patients. Looking at individual reactivities, we found that a patient with anti-CD3ε autoantibodies had intact B and NK cell compartments but dramatically reduced levels of CD4⁺, CD8⁺, and natural killer T cells (Fig. 2g, Extended Data Fig. 7c,d). Similarly, a patient with anti-CD38 autoantibodies exhibited a lower frequency of NK cells and activated CD4⁺ or CD8⁺ T cells, all of which also express CD38 (Extended Data Fig. 7f). Of note, we found that serum from these patients could mediate ADCP against Jurkat or Raji cells respectively in vitro (Fig. 2c). In aggregate, these data show that autoantibodies targeting immune cell surface proteins may lead to depletion of specific immune cell populations in COVID-19 patients.

AAbs exacerbate disease severity in mice

To directly assess the impact of cytokine-targeting autoantibodies in COVID-19 pathogenesis in vivo, we used mice that transgenically express human ACE2 under the human keratin 18 promoter (K18-hACE2) and recapitulate aspects of human COVID-19 pathogenesis upon SARS-CoV-2 infection^19–21. Given the enrichment of anti-IFN-I autoantibodies in severe COVID-19 patients, we first examined the impact of antibody-mediated IFN-I blockade in vivo. We found that mice pre-treated with neutralizing antibodies against the interferon-α/β receptor were more susceptible to SARS-CoV-2 infection; they had increased weight loss (Fig. 3a) and reduced survival (Fig. 3b). Additionally, compared to PBS-treated SARS-COV-2 infected mice, anti-IFNAR treated infected mice exhibited impaired monocyte recruitment, maturation, and proinflammatory macrophage differentiation in the lungs (Extended Data Fig. 8b-d) as well as marked decreases in relative frequency and absolute number of activated (CD44⁺,CD69⁺) NK cells and CD4⁺, CD8⁺, and γδ T cells (Extended Data Fig. 8e,f). Collectively, these findings demonstrate that early blockade of IFN-I signaling by antibodies (which mimics the effects of pre-existing anti-IFN-I autoantibodies) results in exacerbated disease and interferes with myeloid/lymphoid activation in response to SARS-CoV-2 infection.

Figure 3: Immune-targeting autoantibodies exacerbate disease severity in a COVID-19 mouse model.

a, K18-hACE2 mice were intranasally infected with 10³ PFU (b,c) or 10⁴ (d-g) PFU SARS-CoV-2. b-c, Normalized body weight (b) and survival defined as 10% weight loss (c) of PBS or α-IFNAR-treated mice from day 1 to 14 post infection. d-g, Survival defined as 20% weight loss of α-IL-18 (d), α-IL-1β (e), α-IL-21R (f), α-GM-CSF (g), and PBS-treated K18 mice from day 1 to 14 post infection. Significance was determined using two-way ANOVA followed by Sidak correction (b) or log-rank (Mantel-Cox) test (c-g). All error bars in this figure represent standard error of the mean. All n values indicate biologically independent animals examined over 2 independent experiments.

In addition, we had identified autoantibodies against the interleukin-18 (IL-18) pathway (IL-18Rβ in particular; Extended Data Fig. 4d), which plays a critical role in antiviral NK and CD8⁺ T cell responses^22,23. To examine the impact of IL-18 pathway disruption in SARS-CoV-2 infection, we administered neutralizing anti-IL-18 antibodies to K18-hACE2 mice immediately prior to infection. We found that IL-18 blockade greatly enhanced susceptibility to SARS-CoV-2 infection (Fig. 3d, Extended Data Fig. 8k), resulted in significantly higher viral burden (Extended Data Fig. 8g,h), and led to decreased frequency and number of effector NK cells with enhanced cytotoxic properties (CD11b⁺ or KLRG1⁺; Extended Data Fig. 8i,j). These results highlight the disruptive role that autoantibody-mediated IL-18 blockade can have in the immune response to SARS-CoV-2.

Furthermore, we identified IL-1β, IL-21, and GM-CSF as cytokine autoantibody targets in COVID-19 patients. IL-1β and IL-21 both participate directly in host antiviral defense^24,25, and IL-21 in particular was a major contributor to PC2 in our PCA analysis of immune-targeting autoantibodies in COVID-19 (Fig. 1e). GM-CSF plays a critical role in augmenting alveolar macrophage-mediated innate antiviral defense²⁶. Consistent with the antiviral properties of these three cytokines, we found that mice receiving anti-IL-1β, anti-IL-21R, or anti-GM-CSF antibodies became substantially more susceptible to SARS-CoV-2 infection as they had significantly decreased survival (Fig. 3e-g) and lost more weight (Extended Data Fig. 8l-n) upon SARS-CoV-2 infection compared to PBS-treated control mice.

Tissue-targeting AAb correlations

In addition to immune-targeting autoantibodies, we also observed a high prevalence of tissue-associated autoantibodies in COVID-19 patients (Fig. 4a). These autoantibodies were directed against vascular cells, coagulation factors and platelets, connective tissue, extracellular matrix components, and various organ systems including lung, the CNS compartment, skin, gastrointestinal tract, and other tissues. To assess whether any of these putative autoantigens were associated with significant perturbations in clinical phenotypes, we performed exploratory data analysis using a common, generalized linear mixed effects model (Extended Data Fig. 10). We found that certain autoantigens (e.g., NXPH1, PCSK1, SLC2A10, and DCD) significantly correlated with clinical markers known to be associated with worsened COVID-19 disease severity (e.g., D-dimer, ferritin, CRP, lactate)^27,28. Given the extent of CNS-specific autoantigens identified in our REAP screen, and recent reports on the potential for SARS-CoV-2 neuroinvasion²⁹, we further examined whether any autoantibodies correlated with individual patient’s Glasgow Coma Scale (GCS) scores. Intriguingly, we found that ten unique COVID-19 patients developed autoantibodies against HCRTR2, an orexin receptor enriched in the hypothalamus. We noted a marked negative correlation between levels of HCRTR2 autoantibodies in these patients and their exceptionally low GCS scores encompassing the time of sample collection (Fig. 4b). Furthermore, we validated the presence of these autoantibodies in two patients using an ELISA assay (Fig. 4c) and, using an in vitro orexin signaling assay, found that HCRTR2 autoantibodies in one patient were neutralizing (Fig. 4d).

Figure 4: Autoantibodies against tissue-associated antigens are prevalent and functional in COVID-19 patients.

a, Heatmap of tissue-associated REAP scores stratified by disease severity. b, Correlation of HCRTR2 REAP scores with Glasgow Coma Scale scores in COVID-19 patients (n=89, Spearman’s ρ=−0.20, p=0.052). Linear regression (solid lines) and 95% confidence bands (shaded areas) are displayed. Patients are colored by clinical score (CS). Significance of Spearman’s correlation was determined via asymptotic t approximation with a two-sided test. c, Single point pan-IgG HCRTR2 autoantibody ELISA conducted with 1:50 plasma dilution. Dotted line represents the uninfected control (HCW) average plus 3 standard deviations. Results are averages of technical duplicates from one experiment. d, Orexin signaling assay performed with anti-HCRTR2 autoantibody positive COVID-19 patient and HCW IgG. Results are averages of 3 technical replicates from one experiment. All n values indicate samples from unique patients. All error bars represent standard deviation.

Discussion

The surprising extent of autoantibody reactivities seen in patients with COVID-19 suggests humoral immunopathology is an intrinsic aspect of COVID-19 disease pathogenesis. Screening patient samples with the REAP platform, we identified and validated numerous protein targets across a wide range of tissues and immunological and physiological functions. These autoantibodies had potent functional activities and could be directly correlated with various virological, immunological, and clinical parameters in vivo within COVID-19 patient samples. Analysis of REAP score trajectories and comparisons to SARS-CoV-2 humoral responses suggest that some of these autoantibodies likely predated infection while others were induced following infection. Furthermore, murine surrogates of these autoantibodies led to increased disease severity in a mouse model of SARS-CoV-2 infection. Altogether, these results provide evidence that autoantibodies are capable of altering the course of COVID-19 by perturbing the immune response to SARS-CoV-2 and tissue homeostasis.

The diversity of autoantibody responses in COVID-19 patients also underscores the importance of high-throughput and unbiased proteome-scale surveys for autoantibody targets. Beyond validating the biologically-compelling example of anti-IFN-I antibodies in COVID-19, our studies implicated numerous other immune pathways targeted by autoantibodies in COVID-19 that were not previously associated with the disease. We also detected antibodies against various tissue-associated antigens and identified correlations between these antibodies and inflammatory clinical markers like D-dimer, ferritin, CRP, and lactate in COVID-19 patients. Intriguingly, many tissue autoantibodies we identified were also present across diverse physiological compartments frequently implicated during post-COVID syndrome (PCS)³⁰. For example, we identified autoantibodies against the orexin receptor HCRTR2 that, ex vivo, could inhibit orexin signaling, which plays an important role in regulating wakefulness and appetite³¹. Ultimately, whether the specific autoantibodies identified here play a role in the establishment of PCS, and whether they persist beyond the acute phase of COVID-19, warrants further investigation.

In summary, our analyses revealed an expansive autoantibody landscape in COVID-19 patients and identified distinct autoantibodies that exerted striking immunological and clinical outcomes. These results implicate previously underappreciated immunological pathways in the etiology of COVID-19 and suggest novel therapeutic paradigms centered around modulating these pathways, as well as attenuating the autoantibodies themselves. Finally, our findings provide a strong rationale for a wider investigation of autoantibodies in infectious disease pathogenesis.

Methods

Ethics statement

This study was approved by Yale Human Research Protection Program Institutional Review Boards (FWA00002571, protocol ID 2000027690). Informed consent was obtained from all enrolled patients and healthcare workers.

Patients

As previously described¹ and reproduced here for accessibility, 197 patients admitted to YNHH with COVID-19 between 18 March 2020 and 5 May 2020 were included in this study. No statistical methods were used to predetermine sample size. Nasopharyngeal and saliva samples were collected as described³², approximately every four days, for SARS-CoV-2 RT–qPCR analysis where clinically feasible. Paired whole blood for flow cytometry analysis was collected simultaneously in sodium heparin-coated vacutainers and kept on gentle agitation until processing. All blood was processed on the day of collection. Patients were scored for COVID-19 disease severity through review of electronic medical records (EMR) at each longitudinal time point. Scores were assigned by a clinical infectious disease physician according to a custom-developed disease severity scale. Moderate disease status (clinical score 1–3) was defined as: SARS-CoV-2 infection requiring hospitalization without supplementary oxygen (1); infection requiring non-invasive supplementary oxygen (<3 l/min to maintain SpO₂ >92%) (2); and infection requiring non-invasive supplementary oxygen (>3 l/min to maintain SpO₂ >92%, or >2 l/min to maintain SpO₂ >92% and had a high-sensitivity C-reactive protein (CRP) >70) and received tocilizumab). Severe disease status (clinical score 4 or 5) was defined as infection meeting all criteria for clinical score 3 and also requiring admission to the ICU and >6 l/min supplementary oxygen to maintain SpO₂ >92% (4); or infection requiring invasive mechanical ventilation or extracorporeal membrane oxygenation (ECMO) in addition to glucocorticoid or vasopressor administration (5). Clinical score 6 was assigned for deceased patients. For all patients, days from symptom onset were estimated as follows: (1) highest priority was given to explicit onset dates provided by patients; (2) next highest priority was given to the earliest reported symptom by a patient; and (3) in the absence of direct information regarding symptom onset, we estimated a date through manual assessment of the electronic medical record (EMRs) by an independent clinician. Demographic information was aggregated through a systematic and retrospective review of patient EMRs and was used to construct Extended Data Table 1. The clinical data were collected using EPIC EHR and REDCap 9.3.6 software. At the time of sample acquisition and processing, investigators were unaware of the patients’ conditions. Blood acquisition was performed and recorded by a separate team. Information about patients’ conditions was not available until after processing and analysis of raw data by flow cytometry and ELISA. A clinical team, separate from the experimental team, performed chart reviews to determine relevant statistics. Cytokines and FACS analyses were performed blinded. Patients’ clinical information and clinical score coding were revealed only after data collection.

Clinical data acquisition

Clinical data for patients and healthcare workers were extracted from the Yale-New Haven Health computational health platform^33,34 in the Observational Medical Outcomes Partnership (OMOP) data model. For each research specimen, summary statistics including minimum, mean, median, and maximum values were obtained for relevant clinical measurements, including the Glasgow Coma Scale, within ±1 day from the time of biospecimen collection. Disease severity endpoints, including admission, supplemental oxygen use, and invasive ventilation were validated as previously described³⁵.

Yeast induction

All yeast were induced as previously described⁷. In short, one day prior to induction, yeast were expanded in synthetic dextrose medium lacking uracil (SDO -Ura) at 30 °C. The following day, yeast were induced by resuspension at an optical density of 1 in synthetic galactose medium lacking uracil (SGO -Ura) supplemented with 10% SDO - Ura and culturing at 30 °C for approximately 18 hours.

Rapid Extracellular Antigen Profiling (REAP)

IgG antibody isolation for REAP was performed as previously described⁷. In short, Triton X-100 and RNase A were added to plasma samples at final concentrations of 0.5% and 0.5 mg mL⁻¹ respectively and incubated at RT for 30 min before use to reduce risk from any potential virus in plasma. 20 μL protein G magnetic resin (Lytic Solutions) was washed with sterile PBS, resuspended in 75 μL sterile PBS, and added to 25 μL plasma. plasma-resin mixture was incubated overnight at 4 °C with shaking. Resin was washed with sterile PBS, resuspended in 90 µL 100 mM glycine pH 2.7, and incubated for five min at RT. Supernatant was extracted and added to 10 µL sterile 1M Tris pH 8.0. At this point, IgG concentration was measured using a NanoDrop 8000 Spectrophotometer (Thermo Fisher Scientific). To generate yeast-depleted IgG for use in REAP, purified IgG was added to 10⁸ induced empty vector (pDD003) yeast and incubated for 3 hours at 4 °C with shaking. Yeast-IgG mixtures were placed into 96 well 0.45 um filter plates (Thomas Scientific) and yeast-depleted IgG was eluted into sterile 96 well plates by centrifugation at 3000 g for 3 min.

Yeast library selection for REAP was performed as previously described⁷. In short, 400 µL of the induced yeast library was set aside to allow for comparison to post-selection libraries. 10⁸ induced yeast were added to wells of a sterile 96-well v-bottom microtiter plate, resuspended in 100 µL PBE (PBS with 0.5% BSA and 0.5 mM EDTA) containing 10 µg patient-derived antibody, and incubated with shaking for 1 hour at 4 °C. Yeast were washed twice with PBE, resuspended in 100 µL PBE with a 1:100 dilution of biotin anti-human IgG Fc antibody (clone HP6017, BioLegend), and incubated with shaking for 1 hour at 4 °C. Yeast were washed twice with PBE, resuspended in 100 µL PBE with a 1:20 dilution of Streptavidin MicroBeads (Miltenyi Biotec), and incubated with shaking for 30 min at 4 °C. All following steps were carried out at RT. Multi-96 Columns (Miltenyi Biotec) were placed into a MultiMACS M96 Separator (Miltenyi Biotec) in positive selection mode and the columns were equilibrated with 70% ethanol and degassed PBE. Yeast were resuspended in 200 µL degassed PBE and placed into the columns. The columns were washed three times with degassed PBE. To elute the selected yeast, columns were removed from the separator and placed over 96-well deep well plates. 700 µL degassed PBE was added to each well of the column and the column and deep well plate were centrifuged briefly. This process was repeated 3 times. Yeast were recovered in 1 mL SDO -Ura at 30 °C.

DNA was extracted from yeast libraries using Zymoprep-96 Yeast Plasmid Miniprep kits or Zymoprep Yeast Plasmid Miniprep II kits (Zymo Research) according to standard manufacturer protocols. A first round of PCR was used to amplify a DNA sequence containing the protein display barcode on the yeast plasmid. PCR reactions were conducted using 1 µL plasmid DNA, 159_DIF2 and 159_DIR2 primers, and the following PCR settings: 98 °C denaturation, 58 °C annealing, 72 °C extension, 25 rounds of amplification. A second round of PCR was conducted using 1 µL first round PCR product, Nextera i5 and i7 dual-index library primers (Illumina) along with dual-index primers containing custom indices, and the following PCR settings: 98 °C denaturation, 58 °C annealing, 72 °C extension, 25 rounds of amplification. PCR products were pooled and run on a 1% agarose gel. The band corresponding to 257 base pairs was cut out and DNA (NGS library) was extracted using a QIAquick Gel Extraction Kit (Qiagen) according to standard manufacturer protocols. NGS library was sequenced using an Illumina NextSeq 500 and NextSeq 500/550 75 cycle High Output Kit v2.5 with 75 base pair single-end sequencing according to standard manufacturer protocols. A minimum of 50,000 reads per sample was collected and the pre-selection library was sampled at ten times greater depth than other samples.

REAP data analysis

REAP score was calculated as previously described⁷ and reproduced here. First, barcode counts were extracted from raw NGS data using custom codes. Next, aggregate and clonal enrichment was calculated using edgeR³⁶ and custom codes. For aggregate enrichment, barcode counts across all unique barcodes associated with a given protein were summed, library sizes across samples were normalized using default edgeR parameters, common and tagwise dispersion were estimated using default edgeR parameters, and exact tests comparing each sample to the pre-selection library were performed using default edgeR parameters. Aggregate enrichment is thus the log₂ fold change values from these exact tests with zeroes in the place of negative fold changes. Log₂ fold change values for clonal enrichment were calculated in an identical manner, but barcode counts across all unique barcodes associated with a given protein were not summed. Clonal enrichment for a given reactivity was defined as the fraction of clones out of total clones that were enriched (log₂ fold change ≥ 2). Thus, the clonal enrichment metric progressively penalizes proteins with lower fractions of clones enriched. This metric was implemented because a true reactivity should theoretically enrich all yeast clones displaying a given protein.

Aggregate (E_a) and clonal enrichment (E_c) for a given protein, a scaling factor (β_u) based on the number of unique yeast clones (yeast that have a unique DNA barcode) displaying a given protein, and a scaling factor (β_f) based on the overall frequency of yeast in the library displaying a given protein were used as inputs to calculate the REAP score, which is defined as follows.

$$\mathrm{REAP\; score}={\mathrm{E}}_{\mathrm{a}}\times ({\mathrm{E}}_{\mathrm{c}}{)}^2\times {\mathrm{\beta }}_{\mathrm{u}}\times {\mathrm{\beta }}_{\mathrm{f}}$$

β_u and β_f are logarithmic scaling factors that progressively penalize the REAP score of proteins with low numbers of unique barcodes or low frequencies in the library. β_u is applied to proteins with ≤ 5 unique yeast clones in the library and β_f is applied to proteins with a frequency ≤ 0.0001 in the library. β_f was implemented to mitigate spurious enrichment signals from low frequency proteins, which could occur due to sequencing errors or stochasticity in the selection process. β_u was implemented because the clonal enrichment metric is less valid for proteins with low numbers of unique yeast clones, decreasing confidence in the validity of the reactivity. β_u and β_f are defined as follows where x_u is the number of unique yeast clones for a given protein and x_f is the log₁₀ transformed frequency of a given protein in the library.

$$\begin{gathered} {\mathrm{\beta }}_{\mathrm{u}}=\ln ({\mathrm{x}}_{\mathrm{u}}+0.5)/1.705 \\ {\mathrm{\beta }}_{\mathrm{f}}=\ln \left({\mathrm{x}}_{\mathrm{f}}+7.1\right)/1.16 \end{gathered}$$

Antigens (excluding coronavirus RBDs and IL6R) with an average REAP score greater than 0.5 across all samples were defined as “sticky” and excluded from further analysis. Antigens defined as immune-targeting and tissue-associated were manually identified.

Autoantibody enzyme-linked immunosorbent assay (ELISA) measurement

200 ng of purchased or independently produced recombinant protein in 100 µL of PBS pH 7.0 was added to 96-well flat-bottom Immulon 2HB plates (Thermo Fisher Scientific) and placed at 4 °C overnight. Plates were washed once with 225 µL ELISA wash buffer (PBS + 0.05% Tween 20) and 150 µL ELISA blocking buffer (PBS + 2% Human Serum Albumin) was added to the well. Plates were incubated for 2 hours at RT. ELISA blocking buffer was removed from the wells and appropriate dilutions of sample plasma in 100 µL ELISA blocking buffer were added to each well. Plates were incubated for 2 hours at RT. Plates were washed 6 times with 225 µL ELISA wash buffer and 1:5000 goat anti-human IgG HRP (Millipore Sigma) or anti-human IgG isotype-specific HRP (Southern Biotech; IgG1: clone HP6001, IgG2: clone 31–7-4, IgG3: clone HP6050, IgG4: clone HP6025) in 100 µL ELISA blocking buffer was added to the wells. Plates were incubated for 1 hour at RT. Plates were washed 6 times with 225 µL ELISA wash buffer. 50 µL TMB substrate (BD Biosciences) was added to the wells and plates were incubated for 20–30 min in the dark at RT. 50 µL 1 M sulfuric acid was added to the wells and absorbance at 450 nm was measured in a Synergy HTX Multi-Mode Microplate Reader (BioTek). Proteins used are as follows: ACKR1-mIgG2a-Fc (produced in-house), BAMBI (Sino Biological, 10890-H08H-20), C1qB (Sino Biological, 10941-H08B-20), CCL15 (PeproTech, 300–43), CCL16 (PeproTech, 300–44), CNPY3 (produced in-house), CNPY4 (produced in-house), CST5 (produced in-house), CD38 (R&D Systems, 2404-AC-010), GM-CSF (produced in-house), CXCL1 (PeproTech, 300–11), CXCL3 (PeproTech, 300–40), CXCL7 (PeproTech, 300–14), FcμR (R&D Systems, 9494-MU-050), HCRTR2-mIgG2a-Fc (produced in-house), IFN-ω (PeproTech, 300–02J), IL-13 (PeproTech, 200–13), IL-1α (RayBiotech, 228–10846-1), IL-6 (produced in-house), Leptin (R&D Systems, 398-LP-01M), SLC2A12-mIgG2a-Fc (produced in-house), TSLP (PeproTech, 300–62), IL-18Rβ (produced in-house).

Functional validation of anti-GM-CSF and anti-IFN-I autoantibodies

For validation of GM-CSF autoantibodies, TF-1 cells (ATCC, CRL-2003) were starved of recombinant GM-CSF (PeproTech, 300–03) eighteen hours prior to experiments. GM-CSF at 200 pg/mL was incubated with dilutions of purified IgG for 15 min at RT and then used to stimulate TF-1 cells in a 96-well plate (2×10⁵ cells per well) in a final volume of 100 µL (final concentration of 100 pg/mL). For validation of IFN autoantibodies, IFN⍺2 (R&D Systems, 11100–1) and IFN⍵ (Peprotech, 300–02J) at 1500 pg/mL and 2000 pg/mL, respectively, were incubated with dilutions of purified IgG for 15 min at RT and then used to stimulate THP-1 cells (ATCC, TIB-202) in a 96-well plate (3.5 × 105 cells per well) in a final volume of 100 µL (final concentrations of 750 and 1000 pg/mL, respectively). IgG was purified from plasma using protein G magnetic beads (Lytic Solutions) as previously described⁷. After 15 min of stimulation, cells were fixed in 4% paraformaldehyde for 30 mins, washed with PBS, and permeabilized in 100% methanol on ice for 45 mins. Cells were then washed twice with PBE and stained with PE conjugated anti-STAT5 pY694 (1:50) (BD Biosciences, 562077) or anti-STAT1 pY701 (1:50) (BD Biosciences, 612564) and human TruStain FcX (1:100) (Biolegend, 422302) for 1 hour at RT. Cells were washed with PBE and acquired on a SONY SA3800 flow cytometer. Data were analysed using FlowJo software version 10.6 software (Tree Star). pSTAT signal was measured as a function of mean fluorescence intensity (MFI). Percent max signal was calculated by subtracting background MFI and calculating values as a percentage of cytokine induced pSTAT MFI in the absence of IgG. Curves were fit using a sigmoidal 4 parameter logistic curve.

Functional validation of anti-CXCL1, anti-CXCL7, and anti-HCRTR2 autoantibodies

CXCL1, CXCL7, and orexin signaling was assayed using the PRESTO-TANGO system³⁷. HTLA cells, a HEK293-derived cell line that stably expresses β-arrestin-TEV and tTA-Luciferase, were seeded in wells of a sterile tissue culture grade flat bottom 96-well plate (35,000 cells/well) in 100 μL DMEM (+ 10% FBS, 1% Penicillin/Streptomycin). 18–24 hr after seeding (approximately 80–90% cell confluence), 200 ng CXCR2-Tango or HCTR2-Tango plasmid in 20 μL DMEM and 600 ng Polyethylenimine-Max (Polysciences, 24765–1) in 20 μL DMEM were mixed, incubated at RT for 20 min, and added to each well. 18–24 hr after transfection, medium was replaced with 100 μL DMEM (+ 1% Penicillin/Streptomycin, 10 mM HEPES) containing 10 ng CXCL7 (Peprotech, 300–14) or CXCL1 (PeproTech, 300–46), or 100 nM orexin A (Millipore Sigma, O6012) and 5 μg isolated IgG. IgG was purified from plasma using protein G magnetic beads (Lytic Solutions) as previously described⁷. 18–24 hr after stimulation, supernatant was replaced with 50 μL Bright-Glo solution (Promega) diluted 20-fold with PBS with 20 mM HEPES. The plate was incubated at RT for 20 min in the dark and luminescence was quantified using a Synergy HTX Multi-Mode Microplate Reader (BioTek). HTLA cells were a gift from Noah Palm. Tango plasmids were a gift from Bryan Roth (Addgene plasmid # 66260)

Functional validation of anti-CD38 and anti-CD3ε autoantibodies

Bone marrow stem cells were isolated from 8-week old C57BL/6 mouse femur and tibia. Cells were plated in RPMI (+ 10% heat inactivated FBS, 10 U/mL Penicillin, 100 mg/mL Streptomycin; cRPMI) with 30% (vol/vol) L929 fibroblast conditioned media as a source of M-CSF. On day 3 post-isolation, 10 mL L929 fibroblast conditioned media was added to plates. 200,000 BMDMs (7 days post isolation) were plated on non-TC treated 6-well plates in cRPMI + 10% L929 media. The following day, Raji (ATCC, CCL-86) or Jurkat (ATCC, TIB-152) cells were labelled using CellTrace Far Red (Thermo Fisher) according to standard manufacturer protocols. For the T cell ADCP assay, labelled Jurkat cells were incubated with 100 μg/mL healthy control IgG or CD3ε AAb⁺ patient IgG for 30 minutes on ice. Mouse IgG2a anti-human CD3ε (clone OKT3, Biolegend) was used at 5 μg/mL as a positive control. Jurkat cells were washed with 10 ml PBS. 10⁶ Jurkat cells were added to each well and incubated for 3 hours. For the B cell ADCP assay, Raji cells were incubated with complement inactivated patient serum at 1:50 dilution in PBS for 30 minutes on ice. Mouse IgG2a anti-human CD38 (clone MAB2404–100, R&D systems) was added at 5 μg/mL as a positive control. Raji cells were washed with 10 mL PBS once. 500,000 Raji cells were added to each well and incubated for 3 hours. BMDMs were detached from the plate after 3 hour incubation using 10 mM EDTA PBS and stained with anti-mouse CD45 – Pacific blue (clone 30-F11, Biolegend) for 30 minutes at 4 °C. Samples were acquired on a CytoFLEX flow cytometer (Beckman Coulter).

Mice

B6.Cg-Tg(K18-ACE2)2Prlmn/J (K18-hACE2) mice were purchased from the Jackson Laboratories and were subsequently bred and housed at Yale University. 5- to 10-week-old mixed sex mice were used throughout the study. All procedures used in this study (sex-matched, age-matched) complied with federal guidelines and the institutional policies of the Yale School of Medicine Animal Care and Use Committee.

SARS-CoV-2 mouse infections and antibody treatments

Before infection, mice were anesthetized using 30% (vol/vol) isoflurane diluted in propylene glycol. 50 µL of SARS-CoV-2 isolate USA-WA1/2020 (NR-52281; BEI Resources) at 2 × 10⁴ or 2 × 10⁵ PFU/mL was delivered intranasally to mice, equivalent of 10³ or 10⁴ PFU/mouse, respectively. Following infection, weight loss and survival were monitored daily. For IFNAR blockade, mice were treated once with 2 mg of blocking antibodies one day prior to infection (Clone MAR1–5A3). For IL-18 blockade, mice were treated three times each with 0.5 mg of blocking antibodies at day 0, 2, and 4 day post infection (Clone YIGIF74–1G7). For blockade of IL-1β, GM-CSF, or IL-21R, mice were treated three times each with 0.2 mg of blocking antibodies at day 0, 2, and 4 day post infection (Anti-IL-1β: Clone B122; anti-GM-CSF: Clone MP1–22E9; anti-IL-21R: clone 4A9). All blocking antibodies were purchased from BioXCell.

Statistical analysis

Details of linear models and principal component analysis can be found in Supplementary Information. Specific details of other statistical analysis are found in relevant figure legends. Data analysis was performed using MATLAB, GraphPad Prism, R, and the following R packages: ggplot2, edgeR, tidyverse, tidyr, dplyr, stringr, forcats, lme4, and ggpubr.

Reporting Summary

Further information on research design will be made available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Figure 1: REAP screen of COVID-19 patients.

a, A simplified schematic of REAP. Antibodies are incubated with a barcoded yeast library displaying members of the exoproteome. Antibody bound yeast are enriched by magnetic column-based sorting and enrichment is quantified by next-generation sequencing. Schematic was created with Biorender.com. b, Heatmap of all profiled reactivities across all patient samples stratified by disease severity and using the same color scale as in Figure 1a. Sticky antigens (as defined in Methods) were removed from the heatmap. c, SARS-CoV-2 RBD REAP scores for COVID-19 patient samples stratified by positive or negative ELISA RBD reactivity. d, IL-6-R REAP scores for COVID-19 patient samples stratified by treatment with an anti-IL-6R biologic therapy (tocilizumab or sarilumab). Samples collected at least one day after infusion were considered treated. Samples collected on the day of infusion were excluded from analysis due to uncertainty in the timing of sample collection. Significance in c,d was determined using a linear mixed model (see Methods). In c,d n values include longitudinal samples from the same patient.

Extended Data Figure 2: Comparisons of reactivities and clinical/ immunological parameters between patient groups.

a, Number of positive reactivities per sample at various REAP score cutoffs, stratified by disease severity. b, Days from symptom onset (DFSO) in severe and moderate COVID-19 samples. DFSO data was not available for a limited number of samples from each group and was not available for any mild or asymptomatic COVID-19 samples. The median (solid line) and first and third quartile (dashed lines) are shown. c, Number of positive reactivities per sample at various REAP score cutoffs, stratified by patient sex. d, Number of positive reactivities in COVID-19, SLE, and APECED patient samples at various score cutoffs. SLE and APECED patients were screened as previously described⁷. Due to the smaller size of the yeast exoproteome library used to screen the SLE and APECED samples, reactivities in COVID-19 samples against proteins that were not in the previously described yeast exoproteome library were removed from these analyses. e, Number of positive immune-targeting reactivities per sample at various REAP score cutoffs, stratified by disease severity. f, Average percentages of IgD-/CD27- B cells among peripheral leukocytes in COVID-19 patients stratified by disease severity and uninfected controls (neg.). In a-e n values include longitudinal samples from the same patient. In f n values indicate samples from unique patients. Significance was determined using linear mixed models (a-e; see Methods) or a Kruskal-Wallis test followed by a two-sided Dunn’s test (f). Medians are represented by a dashed line and first/third quartiles are represented by dotted lines for all plots in this figure.

Extended Data Figure 3: Autoantibodies exhibit varied developmental kinetics in COVID-19 patients.

a, Percentage of reactivities (REAP score greater than score cutoff) in COVID-19 patients present within 10 days from symptom onset at various score cutoffs. b, Percentage of reactivities in COVID-19 patients that had a REAP score less than the score cutoff (using various score cutoffs) at the first time point sampled and an increase in REAP score of at least 1 at the last time point. c, Percentage of reactivities in COVID-19 patients that had a REAP score greater than the score cutoff (using various score cutoffs) at the first time point sampled and a decrease in REAP score of at least 1 at the last time point. For a-c, all COVID-19 patients with longitudinal samples available (n=77 patients) were included in calculations. d, Plots of longitudinal changes in REAP score for “likely pre-existing”, “newly acquired”, and “waning” REAP reactivities in individual patients alongside scaled anti-spike S1 ELISA values in the same patients. Scaled ELISA values are defined as anti-spike S1 ELISA OD (450 nm - 570 nm) values multiplied by four. In each plot, unique patients are represented by uniquely colored lines. For a given patient, solid lines connect REAP scores of reactivities against respective antigens at various time points and dashed lines connect scaled anti-spike S1 ELISA values at those same time points. The dotted grey line indicates the scaled ELISA positivity cutoff value (see Methods).

Extended Data Figure 4: Biochemical and functional validation of autoantibodies in COVID-19 patients.

a, Single point pan-IgG autoantibody ELISAs conducted with 1:25 or 1:50 plasma dilution (indicated in graph titles). Dotted line represents the uninfected control (healthcare worker) average plus 3 standard deviations. For controls, results (averages of technical duplicates) from biologically independent samples are displayed in the same column (n indicated below each column). For COVID-19 patients, results from one patient are displayed in each column and technical duplicates are depicted as distinct points. b, GM-CSF, c, CD38, and d, IL-18Rβ pan-IgG autoantibody ELISAs conducted with serial dilutions of COVID-19 patient or uninfected control plasma. Results are averages of 2 technical replicates. Curves were fit using a sigmoidal 4 parameter logistic curve. Experiments in a-d were performed once. e,f Flow cytometry plot and gating for the Raji (e) and Jurkat (f) macrophage phagocytosis assay in Figure 2c. g,h IFN-α2 (g) and IFN-ω (h) signaling assay based on STAT1 phosphorylation performed with anti-IFN-α2 or anti-IFN-ω autoantibody positive COVID-19 patient IgG and uninfected control IgG. Results are averages of 2 technical replicates from one experiment. i, Fixed effects model fits for the generalized linear mixed effects model in Figure 2d. The model consisted of composite viral load as the dependent variable and covariates for anti-IFN-I autoantibody presence, days from symptom onset (DFSO), sex, age, and a random effects term per patient (see Methods). j, Hospital stay length in patients with and without anti-IFN-I autoantibodies, stratified by disease severity. Significance in j was determined using a two-sided Wilcoxon rank-sum test. All n values in this figure indicate samples from unique patients.

Extended Data Figure 5: Principal component analysis of immune-targeting autoantibody reactivities in hospitalized COVID-19 patients.

a, Percent of variance explained for principal components from the principal component analysis in Figure 1d. b, PC2 scores of COVID-19 samples stratified by disease severity (clinical score; CS). The mean for each group is represented by a solid black line. n values include longitudinal samples from the same patient. c, Fixed effects model fits from a generalized linear mixed effects model with PC2 score as the dependent variable and covariates for COVID-19 disease severity score, days from symptom onset (DFSO), sex, age, and a random effects term per patient (see Methods).

Extended Data Figure 6: Effects of anti-cytokine autoantibodies on corresponding cytokine plasma concentrations.

a-t, Average concentration of plasma CCL11 (xbo ), CCL15 (b), CCL2 (c), CCL26 (d), CCL8 (e), CXCL1 (f), CXCL12 (g), CXCL13 (h), FLT3LG (i), IFNA2 (j), IL1A (k), IL1B (l), IL13 (m), IL16 (n) and IL21 (o), IL22 (p), IL6 (q), PDGFA (r), TGFA (s), and TSLP (t) measured by a Luminex assay in patients stratified by COVID-19 disease severity and REAP reactivity (AAb⁺; REAP score ≥ 2 at any time point) against the corresponding cytokine. Data are presented as boxplots with the first quartile, median, third quartile, whiskers (minimum/maximum value within the first or third quartile ± 1.5 times the interquartile range), and individual data points indicated. Significance was determined using two-sided, Wilcoxon rank-sum tests. All n values in this figure indicate samples from unique patients.

Extended Data Figure 7: Additional immune cell surface directed autoantibody correlations.

a, Average anti-SARS-CoV-2 RBD IgM reactivity as measured by ELISA in patients stratified by COVID-19 disease severity and REAP reactivity against B cell displayed proteins (as defined in Figure 2e). b, Average percentage among total monocytes of classical monocytes, intermediate monocytes, and nonclassical monocytes in patients stratified by COVID-19 disease severity and REAP reactivity against proteins preferentially displayed on classical/intermediate monocyte (as defined in Figure 2f). c, Average percent CD8⁺ T cells, NKT cells, and NK cells among peripheral leukocytes in patients stratified by COVID-19 disease severity and REAP reactivity against CD3ε. d, Average CD4⁺ T cell to NK cell ratio, CD8⁺ T cell to NK cell ratio, and NKT cell to NK cell ratio among peripheral leukocytes in patients stratified by COVID-19 disease severity and REAP reactivity against CD3ε. e, Representative flow plot of T cells (CD3⁺), NK cells (CD56⁺), and NKT cells (CD3⁺CD56⁺) for Extended Data Fig. 7c,d. f, Average percent B cells, NK cells, activated CD4⁺ T cells, and activated CD8⁺ T cells among peripheral leukocytes in patients stratified by COVID-19 disease severity and REAP reactivity against CD38. g, Representative flow plot of B cells (CD19⁺HLA-DR⁺) for Extended Data Fig. 7f. h, Representative flow plot of activated CD8⁺ T cells (CD38⁺HLA-DR⁺) for Extended Data Fig. 7f. Significance in a was determined using a two-sided Wilcoxon rank-sum test. All n values in this figure indicate samples from unique patients. All boxplots are presented with the median, first and third quartile, whiskers (minimum/maximum value within the first or third quartile ± 1.5 times the interquartile range), and individual data points indicated.

Extended Data Figure 8: Additional immunological and clinical characterization of autoantibody-treated in a COVID-19 mouse model.

a, K18-hACE2 mice were intranasally infected with 10³ PFU (b-f) or 10⁴ (g-n) PFU SARS-CoV-2. b,c, Relative frequency (b) and absolute number (c) of lung Ly6C⁺CD11b⁺CD64⁺ macrophages from mock-infected, SARS-CoV-2-infected PBS-treated, and SARS-CoV-2-infected α-IFNAR-treated K18 mice. d, Expression of CD64 on lung-infiltrating CD11b⁺Ly6C^high monocytes. e,f, Relative frequency (e) and absolute number (f) of CD44⁺CD69⁺ lymphocytes (CD4⁺ T cells, CD8⁺ T cells, NK1.1⁺ cells, and γδ T cells). g,h, Viral RNA loads (g) and infectious titers (h) from lung tissue homogenates of mock-infected, SARS-CoV-2-infected PBS-treated, and SARS-CoV-2-infected α-IL-18-treated mice measured by RT-qPCR and plaque assay, respectively. i,j, Relative frequency (i) or absolute number (j) of CD11b⁺ and KLRG1⁺ NK1.1⁺ cells in lung tissues of PBS and α-IL-18-treated mice. k-n, Normalized body weight of α-IL-18 (k), α-IL-1β (l), α-IL-21R (m), α-GM-CSF (n), and PBS-treated K18 mice from day 1 to 14 post infection. Significance was determined using one-way ANOVA followed by Tukey correction (b-f, g), and unpaired two-tailed t tests (i,j). All n values in this figure represent biologically independent animals examined over 2 independent experiments.

Extended Data Figure 9: Flow cytometry gating strategies for immunophenotyping experiments.

a, Gating strategy to identify B cells described in Figure 2e, monocytes described in Figure 2f & Extended Data Figure 7b, and T cells/NKT cells/NK cells described in Extended Data Figure 7c,d,f, in human PBMCs. b, Gating strategy to identify CD11b⁺Ly6C^high monocytes and Ly6C⁺CD11b⁺CD64⁺ macrophages in mouse lung tissues described in Extended Data Figure 8b-d. c, Gating strategy to identify CD44⁺CD69⁺ lymphocytes in mouse lung tissues described in Extended Data Figure 8e,f. d, Gating strategy to identify KLRG1⁺ and CD11b⁺ NK cells described in Extended Data Figure 8i,j.

Extended Data Figure 10: Tissue-associated autoantibodies are correlated with clinical parameters in COVID-19 patients.

Heatmap of pairwise, -log₁₀ transformed REAP protein p-values from generalized linear model (GLM) fits accounting for age, sex, and maximum REAP score in estimating average or minimum clinical values from hospital admission. A total of 135 samples from unique patients were included in this analysis (n=135). Intensity represents p-values and color indicates directionality of gene-parameter pairs (red indicates positive change, blue indicates negative change. Grey columns indicate genes excluded from analysis due to insufficient patient number (n<2). Asterisks denote pairwise relationships that were significant after correction for multiple comparisons (FDR). Specific details of the GLMs can be found in the Methods. Proteins were classified as in Figure 4a.

Extended Data Table 1 |.

Demographics and clinical characteristics for patients in the IMPACT cohort

	Severe	Moderate	Mild	Asymptomatic	Negative	Total
n	55	103	7	29	54	248
Age (years)	63.15 ± 17.79	64.56 ± 15.77	35.14 ± 14.48	46.55 ± 19.82	39.17 ± 12.28	56.05 ± 19.43
Sex (M|F)	30 (54%) | 25 (46%); n=55	49 (48%) | 54 (52%); n=103	0 (54%) | 7 (100%); n=7	8 (36%%) | 14 (64%); n=22	25 (46%) | 29 (54%); n=54	112 (46%) | 129 (54%); n=241
BMI	32.25 ± 8.8; n=50	30.48 ± 8.28; n=97	--	28.88 ± 8.1; n=18	--	30.84 ± 8.44; n=176
COVID Risk Factors None	13 (24%)	25 (24%)	--	--	--	38 (24%); n=158
Cancer (<1 year)	5 (9%)	6 (6%)	--	--	--	11 (7%); n=158
Chronic Heart Disease	15 (27%)	30 (29%)	--	--	--	45 (28%); n=158
Hypertension	28 (51%)	55 (53%)	--	--	--	83 (53%); n=158
Chronic Lung Disese	9 (16%)	25 (24%)	--	--	--	34 (22%); n=158
Immunosuppresion	4 (7%)	9 (9%)	--	--	--	13 (8%); n=158

Consortia (Yale IMPACT Team)

Abeer Obaid, Adam Moore, Arnau Casanovas-Massana, Alice Lu-Culligan, Allison Nelson, Anderson Brito, Angela Nunez, Anjelica Martin, Annie Watkins, Bertie Geng, Camila D. Odio, Cate Muencker, Chaney Kalinich, Christina Harden, Codruta Todeasa, Cole Jensen, Daniel Kim, David McDonald, Denise Shepard, Edward Courchaine, Elizabeth B. White, Erin Silva, Eriko Kudo, Giuseppe DeIuliis, Harold Rahming, Hong-Jai Park, Irene Matos, Jessica Nouws, Jordan Valdez, Joseph Fauver, Joseph Lim, Kadi-Ann Rose, Kelly Anastasio, Kristina Brower, Laura Glick, Lokesh Sharma, Lorenzo Sewanan, Lynda Knaggs, Maksym Minasyan, Maria Batsu, Mary Petrone, Maxine Kuang, Maura Nakahata, Melissa Campbell, Melissa Linehan, Michael H. Askenase, Michael Simonov, Mikhail Smolgovsky, Nicole Sonnert, Nida Naushad, Pavithra Vijayakumar, Rick Martinello, Rupak Datta, Ryan Handoko, Santos Bermejo, Sarah Prophet, Sean Bickerton, Sofia Velazquez, Tara Alpert, Tyler Rice, William Khoury-Hanold, Xiaohua Peng, Yexin Yang, Yiyun Cao & Yvette Strong

Code Availability

Custom codes used for analysis in this study are available publicly at https://github.com/ring-lab/COVID-19_REAP_nature_2021.

Data Availability

Published RNA-seq dataset is publicly available at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE107011 (GEO: GSE107011). All data analysed in this study are available in the Article, the Source Data files, or the Supplementary Information. Data not available within the manuscript are available from the corresponding author upon reasonable request.