Transient anti-interferon autoantibodies in the airways are associated with recovery from COVID-19

Abstract

Pre-existing anti-interferon alpha (anti-IFN-α) autoantibodies in blood are associated with susceptibility to life-threatening coronavirus disease 2019 (COVID-19). However, it is unclear whether anti-IFN-α autoantibodies in the airways, the initial site of infection, can also determine disease outcomes. In this study, we developed a multiparameter technology, FlowBEAT, to quantify and profile the isotypes of anti-severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and anti-IFN-α antibodies in longitudinal samples collected over 20 months from the airways and blood of 125 donors spanning mild to severe COVID-19. We found that nasal IgA1 anti-IFN-α autoantibodies were induced post-infection onset in more than 70% of mild and moderate COVID-19 cases and were associated with robust anti-SARS-CoV-2 immunity, fewer symptoms, and efficient recovery. Nasal anti-IFN-α autoantibodies followed the peak of host IFN-α production and waned with disease recovery, revealing a regulated balance between IFN-α and anti-IFN-α response. In contrast, systemic IgG1 anti-IFN-α autoantibodies appeared later and were detected only in a subset of patients with elevated systemic inflammation and worsening symptoms. These data reveal a protective role for nasal anti-IFN-α in the immunopathology of COVID-19 and suggest that anti-IFN-α autoantibodies may serve a homeostatic function to regulate host IFN-α following viral infection in the respiratory mucosa.

One sentence summary:

Nasal IgA1 autoantibodies against IFN-α associate with improved COVID-19 prognosis, including fewer symptoms and increased anti-SARS-CoV-2 immunity.

INTRODUCTION

Although most individuals are susceptible to respiratory severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, only a small fraction develop severe, life-threatening coronavirus disease 2019 (COVID-19). We and others have shown that life-threatening COVID-19 is associated with an uncontrolled hyper-inflammatory response in the airways (1–4) rather than an uncontrolled viral load (1, 5). Although these previous studies have revealed the hyper-inflammatory immune phenotypes (1–4) associated with life-threatening COVID-19, the immune-mediated mechanisms that restrict viral replication while protecting from hyper-inflammation and cytokine release syndrome in the airways during natural recovery remain unclear.

Of the many inflammatory cytokines produced in the airways during SARS-CoV-2 infection, type I interferons (IFNs) have been extensively studied due to their protective antiviral properties, particularly when produced early at the disease onset (6–9). However, type I IFNs can also worsen symptoms during viral infections. Delayed or exacerbated IFN-α production in the airways of COVID-19 patients (8) or animal models of coronavirus infection (9–11) is pathologic and was found to be associated with increased disease severity. Thus, there is a contradiction wherein early IFN-α elicits antiviral protection, whereas delayed or persistent IFN-α can trigger hyper-inflammation and worsening symptoms during viral infections.

Patients with pre-existing defects in IFN-α responses, including inborn genetic errors, are predisposed to life-threatening COVID-19 (12–14), supporting a protective role for IFN-α. Consistent with this, autoantibodies against IFN-α (anti-IFN-α) in the blood are associated with an increased risk of life-threatening disease (15–19). Paradoxically, increased and prolonged production of IFN-α in the airways is detrimental and associated with progression to severe COVID-19 (3, 8, 20–23), suggesting that a strictly regulated local IFN-α response in the airway mucosa is instrumental in determining disease outcome. Indeed, controlled IFN-α production strictly limited to the early stages of infection appears beneficial and associated with efficient recovery from COVID-19 (8, 11). Therefore, an important unanswered question is whether a regulated balance between host IFN-α, anti-IFN-α autoantibodies, and anti-SARS-CoV-2 antibodies in the airway mucosa is necessary for efficient recovery during infection and whether dysregulation of this balance in the airways is detrimental, leading to life-threatening COVID-19. In this study, we developed FlowBEAT (flow cytometry-based Bead assay to detect Antigen-specific antibody isoTypes) to determine the longitudinal dynamics of anti-SARS-CoV-2 and anti-IFN-α autoantibodies in the airway, the site of infection, and in matching blood, revealing their contribution to the progression of COVID-19 spanning from disease onset to full recovery.

RESULTS

FlowBEAT reveals distinct anti-SARS-CoV-2 and anti-type I IFN autoantibody responses across tissues and disease states

To quantify the full breadth of antibody responses, including isotype usage and antigen specificity across COVID-19 severity states, we developed a multiparameter assay called FlowBEAT. FlowBEAT is a modular technology that measures up to 176 antibody parameters per sample, including eight human antibody isotypes (IgG1, IgG2, IgG3, IgG4, IgA1, IgA2, IgE, IgM) against a panel of up to 22 host and viral antigens (Fig. 1A, fig. S1A, tables S1 and S2), including host type I IFNs (anti-IFN-α and anti-IFN-ω) and SARS-CoV-2 proteins including anti-spike protein receptor binding domain (RBD), anti-spike protein subunits S1 and S2, and structural proteins (anti-nucleocapsid, -membrane, and -envelope) and non-structural proteins (anti-NSP), including the open reading frame (anti-ORF) (24). We showed the high sensitivity (> 200-fold linear range), specificity, and reproducibility of the FlowBEAT assay in replicate serial dilutions of mouse and human monoclonal antibodies, an NIH COVID-19 human serology standard (25), and pre-pandemic plasma as negative controls (Fig. 1B, fig. S1B to D). We used bovine serum albumin (BSA)-coated control beads to measure non-specific background signal (noise) and establish the lower limits of detection of the assay (fig. S1E). Thus, FlowBEAT reproducibly measured antibodies at a high dynamic range, effectively discriminating low to high antibody titers.

Figure 1. FlowBEAT reveals the breadth of antibody response to COVID-19.

(A) Graphical overview of FlowBEAT sampling from different tissues to reveal antibody isotype and subclasses and antigen specificity. asp., aspirate. (B) Dot plot showing the linear range of the anti-RBD median fluorescence intensity (MFI) signal by FlowBEAT serial dilution of seropositive NIH standard (n = 3 independent replicates) and monoclonal human IgG1 anti-RBD. (C) Graphical summary of study cohort spanning mild, moderate, and severe COVID-19 with multi-tissue collection and longitudinal follow-up. See cohort details in tables S3-S4 and data files S1-S2. (D) Sunburst plots of FlowBEAT showing the breadth of antibody response against SARS-CoV-2 antigens and autoantigens (8 antibody isotypes against 18 antigens) in the airway and systemic samples, grouped by disease severity over the first 90 days post-onset. Individual points represent the per-donor maximum FlowBEAT MFI signal on the Y-axis moving outward from the center. Each outward circle corresponds to the MFI signal noted on the scale. The radial X-axis corresponds to the antigens described on the outermost circle. The color of each dot represents the individual antibody isotypes. The pie chart inserts in the center of the plot indicate the antigen type, as described in panel (A). The mean value for the most prevalent isotype in each tissue (IgA1 in airways and IgG1 in blood) was plotted as a shaded line on the top of each sunburst to provide contrast. Mild donors: n = 17 nasal and 28 systemic; Moderate donors: n = 19 nasal and 24 systemic; Severe donors: n = 22 endotracheal aspirates and 23 systemic.

We applied FlowBEAT to paired longitudinal samples from the airway (nasal swabs and endotracheal aspirates, ETA) and blood (serum or plasma) of COVID-19 patients before and after vaccination (Fig. 1C). We recruited 125 donors spanning from pre-pandemic infection-naïve (n = 36), mild (n = 34), moderate (n = 32), and severe life-threatening (n = 23) COVID-19. We collected outpatient samples longitudinally from 1 to 97 days post-onset with additional post-recovery samples collected for more than 2 years. We collected samples from severe COVID-19 inpatients as available, which spanned days 7 to 48 post-onset, a window included within the timing of the outpatient cohort. We analyzed the nasal swabs and blood from COVID-19 outpatients, whose disease severity (mild versus moderate) we defined based on the total number of assessed symptoms and the time needed to recover from those symptoms (table S3, data file S1). We analyzed the ETA and blood from severe COVID-19 inpatients who were hospitalized in the intensive care unit (ICU) under mechanical ventilation. In total, we analyzed 505 unique FlowBEAT samples (205 airway samples and 300 serum or plasma samples, Fig. 1C, tables S3 and S4, data files S1 and S2). FlowBEAT revealed that the amount and diversity of antibody responses increased with disease severity, represented as peak antibody signal during acute infection (Fig. 1D). As the disease severity increased from mild to severe COVID-19, additional antibody isotypes and specificities against SARS-CoV-2 non-structural proteins (NSP and ORF) and host IFNs were observed (Fig. 1D). Thus, FlowBEAT distinguished the full breadth of antibody response against SARS-CoV-2 and host type I IFNs across tissues and disease states.

Airway-specific antibody isotype and specificity against SARS-CoV-2 proteins distinguish mild, moderate, and severe disease

To better understand the full breadth of antibody responses associated with disease progression and recovery, we assessed the isotype and specificity of antibodies against SARS-CoV-2 proteins longitudinally across disease states. We sampled the nasal mucosa and blood starting at the time of disease onset, determined as the date of polymerase-chain reaction (PCR)-confirmed infection, and through recovery and subsequent mRNA vaccination. In mild and moderate COVID-19 outpatients who recovered from the disease, IgG1, IgG3, and IgA1 were the predominant anti-spike protein antibody isotypes in both nasal mucosa and blood, reaching their respective peaks before 30 days post-onset (Fig. 2A). Although most antibody isotypes were maintained for greater than three months post-onset in blood, they quickly waned in the nasal mucosa after disease recovery, except for IgA1 which was still detectable in more than 60% of mild and moderate-recovered donors three months post-onset (Fig. 2A). Nasal IgA1, IgE, and IgM isotypes of anti-nucleocapsid protein pre-dated the detection of anti-spike protein by at least one week (fig. S2A), likely representing pre-existing immunity to nucleocapsid epitopes conserved between SARS-CoV-2 and endemic common cold coronaviruses (26, 27).

Figure 2. The type and specificity of antibodies against SARS-CoV-2 structural and non-structural proteins change with disease severity, vaccination, and source.

(A) Longitudinal plots showing isotype-specific anti-spike protein MFI signal (Y-axis) across infection and up to four doses of mRNA vaccination (X-axis) in the nasal mucosa and blood of patients with mild and moderate COVID-19 (n = 36 nasal, 107 longitudinal samples; n = 52 systemic, 116 longitudinal samples). Solid lines connect longitudinal samples from individual patients and an overlaid Loess regression smooth curve. The shaded area represents the confidence interval around the smoothed curve. (B) Bar plots show the maximum anti-spike protein signal by individual patients (one point per patient) and are grouped by disease severity (fill texture) and antibody isotype (color). Statistics by Wilcoxon rank-sum test (mild donors: n = 17 nasal and 28 systemic; moderate donors: n = 19 nasal and 24 systemic; severe donors: n = 22 ETA and 23 systemic). (C) Tissue-specific antibody signature against SARS-CoV-2 non-structural proteins (NSP), including open reading frame (ORF) proteins (NSP signature). The gray shaded area highlights the 0-14 days post-onset. Pie charts show the percentage of donors with detectable anti-SARS-CoV-2 NSP signal, defined by MFI > 10². The gradient shading in the pie chart corresponds to peak antibody MFI signal during infection. Colored shaded boxes drawn on the systemic IgG1 anti-NSP plots represent gates to include only donors who generate anti-NSP IgG1 > 10² MFI. These donors were analyzed for maximum (Max.) C-reactive protein (CRP) concentration (mg/L; healthy < 8.1 mg/L) in blood (n = 37) and peak viral load (copies/mL) in the nasal swabs (n = 35). Statistics by Wilcoxon rank-sum test, p = 2x10⁻⁴.

In contrast to the relative stability of isotype usage, particularly IgG1, IgG3, and IgA1, throughout SARS-CoV-2 infection and recovery, mRNA vaccination induced de novo systemic and mucosal IgG4 and systemic IgG2 after two or more vaccine doses (Fig. 2A). Moreover, we found that intramuscular mRNA vaccination boosted a distal nasal IgA1 and IgG1 anti-spike protein antibodies (Fig. 2A), but not anti-nucleocapsid protein antibodies (fig. S2B) supporting a role for vaccination in promoting mucosal sterilizing immunity.

We also found airway- and blood-exclusive antibody isotypes and specificities (per-patient maximum signals during the peak of infection) that were associated with COVID-19 severity. In the airway, the production of IgA2, often related to hyperinflammatory responses in the mucosa (28, 29), increased with disease severity and was detected the highest within endotracheal aspirates (ETA) of patients with severe COVID-19 (Fig. 2B). In blood, whereas IgM was comparable across disease states, an increase in the production of IgG1 and IgG4 anti-spike protein was associated with an increase in COVID-19 severity (Fig. 2B). High IgG3 in the blood, which can suppress type I IFN (30), distinguished patients with moderate to severe disease (Fig. 2B). For example, whereas 61% of moderate cases developed systemic IgG3 anti-spike protein, only 17% of mild cases had detectable systemic IgG3 anti-spike protein during acute infection (< 90 days post-onset) (Fig. 2B, fig. S2C).

Finally, we identified a nasal IgA1 antibody signature against SARS-CoV-2 NSP and ORF (anti-NSP signature), which emerged within two weeks of infection (Fig. 2C). Subsequently, a fraction of patients (25% of mild and 46% of moderate) went on to develop a matching IgG1 anti-NSP signature in the blood (Fig. 2C). This sequential progression from nasal IgA1 to systemic IgG1 anti-NSP was linked to an increased concentration of systemic C-reactive protein (CRP, a clinical marker of inflammation) at the peak of infection, independent of viral load (Fig. 2C, Wilcoxon rank-sum test, p = 4.0x10⁻⁴). Thus, FlowBEAT identified IgA2, IgG3, and IgG4 as features associated with COVID-19 severity, revealed an IgG2 and IgG4 signature induced by repeated vaccination, and identified an initial IgA1 anti-NSP response restricted to the airway site of infection that later progressed to a systemic IgG1 response primarily in patients with increased disease severity.

Anti-IFN-α autoantibodies are induced post-SARS-CoV-2 infection of the airways and peak after local production of host IFN-α

To reveal the longitudinal dynamics between anti-SARS-CoV-2 antibodies and autoantibodies against type I IFN (anti-IFN-α and anti-IFN-ω), we longitudinally assessed 36 mild and moderate COVID-19 patients for whom we collected matched nasal and blood samples throughout infection, recovery, and subsequent mRNA vaccination. To quantify anti-IFN autoantibodies, we coated our assay beads with the same IFN-α (IFN-α2a) and IFN-ω proteins used in previous studies (31). In addition, we included other IFN-α subtypes, IFN-α5 and IFN-α14, shown to restrict SARS-CoV-2 replication (32). To distinguish the background from the antigen-specific signal, we included BSA-coated control beads in every multiplexed assay to calculate non-specific antibody binding. We only considered positive signals above the BSA background and confirmed the specificity of low anti-IFN signals (signals above but near the BSA background) by serial dilution of samples with positive signals (fig. S3A and B).

We first investigated whether anti-IFN autoantibodies (anti-IFN-α and anti-IFN-ω) could be detected locally in the airways throughout COVID-19 progression. We discovered a robust IgA1 anti-IFN-α autoantibody response induced shortly after infection, frequently emerging within two weeks post-onset in the nasal mucosa (anti-IFN-α were detected in 72% or 26/36 of mild and moderate cases of COVID-19, Fig. 3A). Consistent with the anti-SARS-CoV-2 antibody isotype response, nasal anti-IFN was dominated by IgA1 in mild and moderate COVID-19, whereas IgA2, often associated with mucosal hyperinflammation (28, 29), was detected at lower signals measured as median fluorescence intensity (MFI) (Fig. 3A). These new-onset nasal IgA1 anti-IFN-α autoantibodies were transient and waned as patients recovered from symptomatic disease (Fig. 3A). The transient IgA1 anti-IFN-α recognized all three IFN-α subtypes (−α2a, −α5, −α14) with strong positive correlation (anti-IFN-α5: r=0.61, p < 1x10⁻⁸, anti-IFN-α14: r=0.84, p<1x10⁻¹⁵, fig. S3C to E), whereas systemic IgG1 poorly recognized anti-IFN-α5 (fig. S3E), suggesting a systemic response unlinked from the nasal mucosa. We observed a similar pattern of transient production of anti-IFN-ω in the same donors but at a lower FlowBEAT MFI signal (fig. S3F). In contrast to anti-IFN-α, other autoantibodies, including the canonically autoreactive antibodies of the VH4-34 immunoglobulin gene family, which we previously showed to be increased in the blood of hospitalized COVID-19 patients (33), remained stable in the airways throughout the disease course (fig. S4A), revealing a transient property of the autoantibodies against type I IFN in COVID-19.

Figure 3. Anti-IFN-α autoantibodies are transiently induced in the nasal mucosa after infection and are associated with viral load and local IFN-α secretion.

(A) Longitudinal plot showing the post-onset induction of nasal IgA1, IgA2, and IgG1 anti-IFN-α2a autoantibodies in nasal swabs and blood samples from individuals with mild and moderate COVID-19. Days post-onset on the X-axis, and longitudinal samples from individual donors are linked by solid lines. Loess regression smooth curves of the same data are overlaid with the confidence interval shown as a shaded area, and color corresponding to the antibody isotype. The gray shaded area highlights the 0-14 days post-onset. Pie charts show the percentage of donors with detectable anti-IFN-α2a in nasal swabs (orange) and blood (green) samples. The gradient shading in the pie chart corresponds to the peak antibody MFI signal during acute infection. Mild + Moderate donors: n = 36 with matching nasal and systemic samples; 102 nasal and 112 systemic longitudinal samples. Naïve-vaccinated donors (n = 7) were included as a control. (B) Longitudinal viral load (viral copies per mL, qPCR) in nasal swabs of patients were grouped by the presence (IgA producers, orange) or absence (IgA non-producers, gray) of anti-IFN-α2a autoantibodies as in (A). Statistic by Wilcoxon rank-sum test, p = 0.001. (C) Longitudinal nasal IFN-α2a cytokine concentrations were plotted as pg/mL of nasal swab supernatant (n = 41 swabs from 31 donors) and grouped by the presence or absence of anti-IFN-α2a as in (B). Statistic by Wilcoxon rank-sum test, p = 3.0x10⁻⁴. (D) Longitudinal plots binned by week (7-day increments) show peak viral load, IFN-α2a secretion, and anti-IFN-α2a autoantibodies in the nasal mucosa. (E) In vitro IFN-α2a neutralization assay using a type I IFN-reporter cell line. The IFN-induced response signal following exogenous IFN-α2a (60 units/mL) stimulation is plotted as arbitrary units (AU). Lower AU values indicate neutralization of exogenous IFN-α2a in the presence of systemic serum or plasma (n = 22 donors, 27 samples) or nasal samples (n = 29 donors, 51 samples). Samples were grouped by antibody detection based on the FlowBEAT MFI signals. Samples with antibody MFI < 10² were considered as not detected (N.D.). Horizontal lines represent group means and standard error. A thin black line connects groups tested by one-way ANOVA, p = 0.007. (F) Pearson correlation between FlowBEAT MFI signal (X-axis) and cell response to exogenous IFN-α stimulation (Y-axis) as in (E) for systemic (n = 16 donors, 21 samples) and nasal samples (n = 16 donors, 18 samples). (G) CXCL10 measurements in systemic (n = 10 donors, 17 samples) and nasal samples (n = 13 donors, 14 samples) of IFN-α2a cytokine-positive samples. Statistics by Wilcoxon rank-sum test.

The initial IgA1 anti-IFN-α response was limited to the nasal mucosa, and fewer donors progressed to a systemic response in the blood. Only 36% of the patients (13/36) developed blood anti-IFN-α, which appeared more than 2 weeks post-onset (Fig. 3A). This systemic autoantibody response was dominated by the IgG1 isotype (Fig. 3A), which was rarely detected in the airways (fig. S4B). Unlike nasal IgA1, systemic IgG1 persisted for more than two months in some patients (Fig. 3A). They were more frequently detected in donors with moderate (42%) as compared with mild (14%) disease (fig. S4C). Finally, in contrast to SARS-CoV-2 infection, mRNA vaccination did not induce anti-IFN-α autoantibodies (Fig. 3A), indicating that anti-IFN-α autoantibodies are a specific response to SARS-CoV-2 infection.

Next, we investigated whether the anti-IFN-α response was associated with viral load by assessing the same nasal swabs for SARS-CoV-2 copy numbers. We found that patients who produced detectable anti-IFN-α autoantibodies experienced significantly greater peak viral load than those who never produced nasal anti-IFN-α (Fig. 3B, Wilcoxon rank-sum test, p = 0.001). The temporal separation between peak viral load and anti-IFN-α induction (Fig. 3A, 3B), confirms that nasal anti-IFN-α was induced post-viral exposure and not pre-existing and the association of anti-IFN-α with elevated viral load (Fig. 3B) further suggests that this is a viral-induced autoantibody response.

We then assessed whether viral-induced anti-IFN-α autoantibodies were associated with local secretion of IFN-α. We longitudinally measured host IFN-α in the same matched nasal swab and serum samples. We detected peak nasal IFN-α cytokine within the first week of infection, waning after two weeks (Fig. 3C). Only the patients who produced high nasal IFN-α early after infection developed nasal anti-IFN-α autoantibodies (Wilcoxon rank-sum test, p = 3x10⁻⁴) (Fig. 3C). Longitudinally, the peak production of nasal IFN-α preceded the anti-IFN-α autoantibody response, which peaked two weeks post-onset, when IFN-α waned (Fig. 3D). We observed a similar pattern of IFN-α and anti-IFN-α dynamics in the blood (fig. S4D).

Finally, we tested whether the viral-induced anti-IFN-α autoantibodies could neutralize the activity of IFN-α signaling using a type I IFN-reporter HEK293 cell line optimized using exogenous IFN− α2a (fig. S5A) and monoclonal anti-IFN-α2a (fig. S5B) in healthy control samples. We showed that patient serum or plasma containing anti-IFN-α autoantibodies neutralized exogenous IFN-α signaling in vitro (Fig. 3E, one-way ANOVA, p = 0.007), corroborating previous studies (16). Unlike blood samples, nasal mucosal samples neutralized exogenous IFN-α even without detectable anti-IFN-α (Fig. 3E), suggesting a robust mechanism to control IFN-α signaling in the nasal mucosa. This neutralization capacity of nasal samples was not due to in vitro technical artifacts such as cellular toxicity because the viability of the reporter cell line was comparable between blood and nasal samples (fig. S5C), and the neutralization signals were similar in buffer-exchanged samples (fig. S5D).

To determine whether nasal anti-IFN-α autoantibodies can contribute to the neutralization potential of nasal samples, we repeated our cell-based assay using low-throughput optimized conditions to obtain a higher dynamic range near the assay’s lower limit of detection. We tested a subset of nasal swabs containing a range of IgA anti-IFN-α titers in the presence of exogenous IFN-α (60 activity units/mL). While we did not observe a significant decrease in IFN-α signal correlated with anti-IFN-α (Fig. 3F), we confirmed that nasal anti-IFN-α does not enhance endogenous or exogenous IFN-α signaling (fig. S4E). In blood, high titers of IgG1 anti-IFN-α showed a negative correlation with IFN-α signaling (Fig. 3F, r = −0.46, p = 0.03), further supporting that anti-IFN-α can neutralize IFN-α. To further validate whether the production of anti-IFN-α can neutralize IFN signaling, we measured C-X-C motif chemokine ligand 10 (CXCL10, also known as interferon gamma-induced protein 10, IP-10) longitudinally in the same samples, as CXCL10 was shown to be a reliable marker for IFN signaling in the nasal mucosa of patients with COVID-19 (34). We showed that serum or plasma anti-IFN-α was significantly associated with reduced CXCL10 (Fig. 3G, Wilcoxon Rank-sum test, p = 0.022). In the nose, this reduction did not reach statistical significance (Fig. 3G), even though, longitudinally, the peak nasal anti-IFN-α is associated with reduced CXCL10 (fig. S5F). Thus, the reduction in IFN-α signaling and CXCL10 in nasal samples is likely due to the combined effects of IgA1 anti-IFN-α autoantibodies and other unknown mediators in the nose of acutely infected individuals.

Transient IgA1 anti-IFN-α autoantibodies in the nasal mucosa are associated with fewer symptoms, more anti-SARS-CoV-2 antibodies, and full recovery

Next, we explored whether anti-IFN-α autoantibodies could be associated with symptomatology or other clinical features in patients with mild and moderate disease (table S3). First, we found that nasal autoantibody response was unlinked from the blood as there was no direct correlation between nasal and blood anti-IFN-α (Fig. 4A). We then grouped patients into three categories based on the presence or absence of nasal and blood anti-IFN-α (nasal , blood-only producers, and non-producers) (Fig. 4A). Between these groups, we found no statistical differences in sex, age, or body mass index (BMI) (Fig. 4B, fig. S6A-C), even though males produced the highest titers of blood IgG1 anti-IFN-α (fig. S6A-B). However, we identified a significant (p<0.05) association between anti-IFN-α autoantibodies and disease features, including shortness of breath, sleep disturbance, balance issues, and chills (Fig. 4B).

Figure 4. Nasal IgA1 anti-IFN-α autoantibodies are associated with fewer symptoms, less systemic inflammation, and increased anti-SARS-CoV-2 antibody responses.

(A) Biaxial plot of nasal IgA1 anti-IFN-α2a (Y-axis) and systemic anti-IFN-α2a IgG1 (X-axis) autoantibody signal. Points represent the patient’s peak antibody FlowBEAT signal (MFI) within the first six weeks post-onset. Patient populations (n = 36 donors with matched nasal swabs and systemic samples) are grouped (dashed lines) as nasal producers (pink), blood-only producers (green), and non-producers (black). (B) Heatmap of nasal and blood anti-IFN-α autoantibodies, demographics, airway cytokines, and frequency of symptoms (Chi-square test, with p-value reported) of patients grouped as in (A). Rows correspond to individual patients; the columns describe patient features as labeled. The pie chart shows the frequency of “shortness of breath” (SOB) symptoms between pink and green groups as in (A). (C) The scatterplot shows the maximum systemic C-reactive protein (CRP) concentration (healthy CRP < 8.1 mg/L) for each patient (n = 30) grouped as in (A), with non-producers indicated as double negative (DN). The thin black line connects group means tested by one-way ANOVA (p = 9.3x10⁻⁴). (D) Pearson correlation between anti-IFN-α2a (X-axis) and composite anti-SARS-CoV-2 NSP signal (Y-axis). All samples plotted (n = 57) are from donors in the pink group (nasal producers, n = 21 donors) as in (A). For IgA1 anti-NSP (NSP1, NSP7, ORF3a, ORF8) against IgA1 anti-IFN-α: Pearson’s r = 0.75, p = 9.0x10^-12. For IgA2 anti-NSP (NSP1, NSP2, ORF3a, ORF8) against IgA2 anti-IFN-α: Pearson’s r = 0.97, p = 2.2x10⁻¹⁶. (E) Maximum IgA1 (nasal) or IgG1 (blood) anti-SARS-CoV-2 NSP signal (as in D, composite signal of anti-NSP1, −NSP7, −ORF3a, −ORF8) grouped by gates displayed in (A): nasal IgA producers (nasal producers, pink gate), systemic IgG producers (blood-only producers, green gate), and double-negative (DN) (non-producers, gray gate). Statistics by Wilcoxon rank-sum test.

Patients who developed systemic IgG1 anti-IFN-α (blood-only producers) reported persistent shortness of breath and other relevant symptoms (Fig. 4B, Chi-square test, multiple p < 0.05). In contrast to the systemic IgG1 producers, nasal IgA1 anti-IFN-α autoantibody producers showed significantly fewer symptoms, including less shortness of breath (Fig. 4B, p < 0.01). COVID-19 treatment guidelines consider shortness of breath a symptom associated with disease severity and worse prognosis (35), likely reflecting viral spread to the lower respiratory tract. Consistent with this, patients who developed systemic IgG1 anti-IFN-α and shortness of breath had elevated systemic CRP (Fig. 4C, one-way ANOVA, p = 9.3x10⁻⁴).

Finally, we tested whether nasal anti-IFN-α autoantibodies were associated with an increased humoral immune response against SARS-CoV-2 in the mucosal site of infection. Indeed, we found a positive linear correlation between nasal IgA1 and IgA2 anti-IFN-α autoantibodies and antibodies against SARS-CoV-2 proteins, including an anti-NSP and ORF signature, showing that nasal anti-IFN-α production is linked to enhanced humoral immunity against SARS-CoV-2 in the upper airway (Fig. 4D). In contrast, patients who did not develop anti-IFN-α (non-producers) had fewer anti-SARS-CoV-2 antibodies in the nasal mucosa (Fig. 4E, Wilcoxon rank-sum test, p = 0.039). When compared to nasal IgA1 producers, patients who developed systemic IgG1 anti-IFN-α (blood-only producers) had higher anti-SARS-CoV-2 antibodies in the blood but not in the upper airway (Fig. 4E, Wilcoxon rank-sum test, p = 1.3x10⁻³). Thus, nasal IgA1 anti-IFN-α is associated with robust anti-SARS-CoV-2 immunity in the nose.

Hospitalized patients show elevated anti-IFN-α in the airways associated with uncontrolled IFN-α, higher viral load, and lower anti-SARS-CoV-2 antibody titers

To determine whether the coordinated balance between host IFN-α and anti-IFN-α we identified in mild- and moderate-recovered patients is disrupted in severe COVID-19, we performed FlowBEAT on endotracheal aspirates (ETA) and peripheral blood from 23 hospitalized and unvaccinated patients admitted to the ICU with life-threatening severe COVID-19. We detected IgA1 and IgA2 anti-IFN-α autoantibodies in the airway mucosa (ETA) of 86% of severe patients (Fig. 5A), and the remaining 14% also had detectable anti-IFN-α but of a different antibody isotype (fig. S6D). In contrast, only 43% of severe patients developed detectable systemic IgG1 anti-IFN-α (Fig. 5A), in agreement with a recent study that found anti-IFN autoantibodies in the blood of 45% of hospitalized patients using a similar bead-based assay (36).

Figure 5. Patients with severe COVID-19 show persistent IFN-α and anti-IFN-α responses associated with viral load, hyperinflammation, and lower anti-SARS-CoV-2 humoral immunity.

(A) Pie charts of IgA1+IgA2 (orange) and IgG1 (green) anti-IFN-α2a autoantibody frequency and peak FlowBEAT signal (shaded gradient) in the endotracheal aspirate (ETA) and blood of hospitalized patients (n = 23 donors) with severe COVID-19. (B) Pearson correlation between ETA IgA1 and IgA2 anti-IFN-α autoantibodies and IFN-α2a cytokine. Pearson’s r = 0.59, p = 0.0063 (IgA1), and r = 0.61, p = 0.0035 (IgA2). (C) Bar chart comparing the relative IgA1 and IgA2 subclass usage in the airways of mild/moderate (nasal swabs) and severe (ETA) patients. Statistics by Wilcoxon rank-sum test, p = 0.02 (anti-IFN-α2a IgA1), p = 9.1x10⁻⁴ (anti-IFN-α2a IgA2), and p = 5.9x10⁻⁵ (anti-spike protein IgA2). (D) Biaxial plot of individual patients showing antibody signals corresponding to systemic IgG1 and airway IgA1 + IgA2 anti-IFN-α2a. Patients are grouped into airway IgA producers (pink) and non-producers (gray). The bar chart shows the viral load (viral copies/mL by qPCR) in the ETA of each individual in the biaxial plot. Wilcoxon rank-sum test, p = 0.028. (E) Heatmap of the airway (ETA IgA1+IgA2) and blood (IgG1) anti-IFN-α2a and airway (ETA) cytokine concentrations (n = 20 donors). Displayed cytokines showed a significant Pearson correlation with anti-IFN-α2a (IgA1+IgA2) in the ETA (*p < 0.05; **p < 0.01; ***p < 0.001). Rows correspond to individual patients as in (D), and columns indicate features as labeled. (F) Correlation plots between anti-IFN-α2a (X-axis) and the composite anti-SARS-CoV-2 NSP signal (Y-axis) as described in Fig. 4D. All samples plotted are from donors in the pink group (nasal producers) as in (D). Pearson’s product-moment correlation coefficient tests did not reach significance for either IgA1 or IgA2 (correlation not statistically significant, p > 0.05).

Of note, earlier studies have reported a lower frequency (approximately 5% to 10%) of anti-IFN-α in the blood of patients hospitalized with COVID-19 (17, 31). However, subsequent studies using different assays, including bead-based antigen arrays and IFN-α neutralization, detected a greater frequency of anti-IFN-α producers among hospitalized patients, ranging from approximately 20% (particularly among older adults) (16) to 45% (36), which corroborates the frequency we observed using the FlowBEAT assay (Fig. 5A, fig. S6E). Perhaps due to differences in assay sensitivity, earlier studies may have underestimated the frequency of patients who produce lower titers of anti-IFN.

In contrast to mild and moderate outpatients, we detected sustained IFN-α production in the ETA of hospitalized patients that positively correlated with IgA1 and IgA2 anti-IFN-α autoantibodies (Fig. 5B), revealing a dysregulated IFN-α/anti-IFN-α response at the airway site of infection. Moreover, patients with severe COVID-19 produced more hyperinflammation-associated IgA2 subclass anti-IFN-α, consistent with sustained hyperinflammation in the airways (28, 29) (Fig. 5C, Wilcoxon rank-sum test, p = 9.1x10⁻⁴). Similarly, the anti-SARS-CoV-2 response also switched from the IgA1 to the hyperinflammation-associated IgA2 in the airway (Fig. 5C, Wilcoxon rank-sum test, p = 5.9x10⁻⁵). Given that nasal swabs and ETA are sampled from different anatomic locations (upper and lower airways), these differences in the IgA subclass may also reflect differences in humoral responses at each mucosal site.

Consistent with our findings in mild and moderate disease, airway IgA anti-IFN-α autoantibodies in severe disease were associated with increased viral load (Fig. 5D, Wilcoxon rank-sum test, p = 0.028), suggesting an ongoing antiviral immune response in the hospitalized patients. Indeed, airway anti-IFN-α was positively associated with local (ETA) pro-inflammatory cytokines, including those that promote myeloid recruitment and activation, such as IFN gamma (IFN-γ), IFN lambda 1 (IFN-λ1), CXCL10, CXCL11, C-C motif chemokine ligand 19 (CCL19), Tumor necrosis factor (TNF) alpha (TNF-α), and TNF-related apoptosis-inducing ligand (TRAIL) (Fig. 5E, Pearson correlation, all p < 0.05). However, whereas airway IgA anti-IFN-α correlated with enhanced anti-SARS-CoV-2 humoral immunity in mild and moderate patients (Figs. 4D and E), this correlation no longer persisted in the airways of severe patients (Fig. 5F, and fig. S6F, suggesting weaker airway humoral immunity as a feature of severe disease.

DISCUSSION

The nasal epithelium is the initial site of SARS-CoV-2 infection. As such, antiviral immune responses locally in the nasal mucosa can determine disease recovery or progression to severe COVID-19 (8, 21, 37, 38). Here, we show that transient autoantibodies against type I IFNs (anti-IFN-α) in the nasal mucosa are a common feature of the host immune response to SARS-CoV-2 infection and are associated with efficient recovery from mild and moderate disease. Mild and moderate-recovered patients who produced nasal IgA1 but not systemic IgG1 anti-IFN-α showed fewer symptoms, less systemic inflammation, and increased anti-SARS-CoV-2 humoral immunity in the airways. In contrast, patients who progressed directly to a systemic IgG1 anti-IFN-α without detectable nasal IgA1 showed worsened symptoms (particularly shortness of breath), increased systemic inflammation, and a switch in anti-SARS-CoV-2 antibody isotypes from IgG1 to the inflammation-associated IgG3. Thus, the high prevalence (> 70%) of nasal IgA anti-IFN-α autoantibodies in mild- and moderate-recovered patients suggests that autoreactivity to type I IFN at the nasal site of infection is an intrinsic aspect of the host protective antiviral immunity and not a pre-condition to severe COVID-19, as has previously been suggested for blood anti-IFN-α (16). Most importantly, our findings showing a direct association between nasal anti-type I IFN and disease resolution should warn against depleting anti-IFN-α autoantibodies or administering nasal type I IFN in ongoing clinical trials to treat patients with COVID-19 (39, 40).

Type I IFN is an essential antiviral cytokine that protects against most viral infections (6). However, its role in protection and recovery from COVID-19 and other coronavirus infections is uncertain. Animal models of coronavirus infection in mice (9, 10) and non-human primates (11) suggest that type I IFN is dispensable or detrimental to resolving the disease. Mice that lack the IFN-αβ receptor (IFNAR^−/−) are protected from lethal infection with SARS-CoV without affecting viral load (9). Similarly, a mouse model of SARS-CoV-2 infection shows that type I IFN is not required for viral clearance but is responsible for airway infiltration of inflammatory cells and drives the lung pathology associated with severe disease (10). In non-human primates, persistent type I IFN can lead to lung pathology but could be protective if produced in small amounts and only at the early stages of SARS-CoV-2 infection (11). In humans, persistent or dysregulated type I IFN in the airways also leads to a worse prognosis (8). Our data corroborate previous studies and support a mechanism in which SARS-CoV-2 infection induces early and transient anti-IFN-α autoantibodies to counteract the detrimental effects of excessive IFN-α secretion in the airways.

Previous studies showed that blood anti-IFN-α and inborn errors of type I IFN immunity are associated with increased susceptibility to severe, life-threatening COVID-19 (12, 16). Conclusions drawn from these studies infer that anti-IFN-α autoantibodies are pre-existing rather than infection-induced, based on their findings showing no detectable anti-IFN-α in the blood of mild outpatients measured at disease onset, whereas systemic anti-IFN-α was detected in over 10% of patients hospitalized with COVID-19. To reconcile the apparent discrepancy with previous findings, we observed that mild patients develop systemic anti-IFN-α only later, over 2 weeks after infection. Thus, it is possible that previous studies missed the infection-induced anti-IFN-α in mild patients as autoantibodies were measured in blood and too early (when anti-IFN-α is only detectable in the airways). Moreover, hospitalized patients with systemic anti-IFN-α may have developed their infection-induced autoantibodies before being admitted to the ICU. Indeed, subsequent studies corroborated our findings and showed new-onset anti-IFN-α in the blood of patients hospitalized with COVID-19 (36). Therefore, by considering tissue-specific immune responses (transient nasal IgA1 versus persistent blood IgG1), our findings reconcile previous studies and provide a framework to understand the immunopathology of COVID-19 in different tissues, challenging the notion that anti-IFN-α autoantibodies are universally pathologic.

Although autoantibodies are often seen as pathologic, the prevalence of anti-IFN-α in over 70% of infected individuals and its association with efficient recovery suggests a common mechanism of host antiviral immunity instead of a break in B cell tolerance. Indeed, recent studies show that new-onset autoantibody responses against a broad array of inflammatory cytokines and chemokines is a common feature of host immune responses during COVID-19 (17, 36, 41), and their neutralizing capacity likely plays a role in attenuating inflammation during infection (42). As such, these autoantibodies were not associated with disease severity but with efficient recovery, similar to our findings for nasal anti-IFN-α, which we show may partially neutralize IFN-α signaling.

Other studies have suggested that certain anti-cytokine autoantibodies could prolong the half-life of the target cytokine and enhance, instead of neutralizing, their activity in vivo (43, 44). However, in our studies, nasal and blood anti-IFN-α did not enhance exogenous IFN-α signaling or CXCL10, a readout for IFN signaling (34). Moreover, anti-IFN-α correlated with a significant decrease in IFN signaling and CXCL10 in blood but not nasal samples. The complex composition of the nasal samples may partly explain this difference. Unlike blood, nasal samples naturally neutralized IFN-α even without anti-IFN-α, revealing a separate mechanism to control IFN-α in mucosal sites. Even though the nasal mucosa of uninfected donors may constitutively neutralize IFN-α, anti-IFN-α autoantibodies are only induced post-SARS-CoV-2 infection and only in patients with measurable IFN-α production in the nose, suggesting additional mechanisms regulate IFN-α abundance beyond autoantibodies. IFN-α production must be tightly regulated to prevent adverse effects, including hyperinflammatory diseases, autoimmunity, cancer, and even failed pregnancy due to placental remodeling (45, 46). Thus, it is conceivable we have evolved a robust mechanism to prevent excessive IFN-α in mucosal sites that continuously survey the environment. Collectively, our data suggest that nasal anti-IFN-α autoantibodies may contribute to the neutralizing activity of the nasal mucosa and may help control hyperinflammatory responses by removing excess IFN-α during viral infections.

The direct progression to systemic IgG anti-IFN-α without an initial nasal IgA response characterizes a small group of patients with worse prognoses. If measured after disease recovery, the persistent IgG anti-IFN-α could be interpreted as “pre-existing”, as previously suggested (16, 31). Moreover, donors who experienced higher systemic inflammation and are therefore at elevated risk for post-acute sequelae of COVID-19 (PASC) (42, 47) are more likely to develop systemic IgG1 anti-IFN-α during acute infection. Systemic IgG1 anti-IFN-α may contribute to the low IFN-α cytokine detected in the blood of individuals with PASC (48) or render them susceptible to subsequent viral infections that might contribute to PASC symptoms. Our studies identified only one donor with pre-existing systemic anti-IFN-α autoantibodies measured at diagnosis (donor ER029). This individual resolved the disease without hospitalization and experienced moderate symptoms. Similarly, patients with known pre-existing systemic anti-IFN-α autoantibodies, such as autoimmune polyendocrine syndrome type-1 (APS-1), showed only mild symptoms of COVID-19 in a prospective study (49) but susceptibility to more severe disease in another (50). Thus, future studies with larger patient cohorts are needed to resolve these discrepancies and determine the functional differences between nasal (IgA1) and systemic (IgG1) anti-IFN-α that are viral-induced or pre-existing.

The mechanisms that lead to the activation of autoreactive B cell clones against cytokines, particularly IFN-α, are unknown. Our data showing that only patients who produced IFN-α in the airways developed anti-IFN-α suggests that B cell activation is antigen-specific. The fast and transient nature of airway autoantibodies against local IFN-α indicates that B cells might be activated locally independent of germinal centers or T cell help, leading to short-lived (transient) IgA antibody-secreting cells (51, 52). In contrast, systemic IgG1 anti-IFN-α persisted for months, suggesting a systemic B cell response separate from the nasal mucosa. Indeed, whereas nasal IgA1 bound all IFN-α subtypes we tested, systemic IgG1 poorly bound IFN-α5 anti-git, further supporting the notion that systemic IgG1 represents a separate humoral immunity comprised of different antigen-specific B cells against new IFN-α epitopes, likely affecting their effector functions. Given the interest in developing a nasal vaccine (27), it will be critical to determine whether it induces mucosal anti-IFN-α responses. Our results suggest that such a response would occur only if the vaccine (or its adjuvant) induced local IFN-α cytokine. Extrapolating from our findings, we do not expect a detrimental effect of IgA1 anti-IFN-α if produced in the nasal mucosa following local IFN-α production.

What determines the time and amount of nasal IFN-α production during infection is unclear. Recent studies show that airway epithelial cells are heterogeneous and respond differently to SARS-CoV-2 infection, including their ability to produce IFN-α (37, 38). Thus, it is possible that the type of infected epithelial cell in the airways could influence IFN-α production, determining disease outcome to mild or severe COVID-19. Our findings of persistent anti-IFN-α in the lower airways of ICU patients are consistent with a model in which persistent or dysregulated IFN-α production by certain epithelial cells in the lower airways induces local hyperinflammation and a continuous IgA anti-IFN-α response in a feedforward loop to counteract IFN-α. This inflammatory milieu in the airways of hospitalized patients favors class-switching to IgA2, which forms immune complexes that can further exacerbate inflammation (28), as confirmed by other studies on severe COVID-19 (14, 53). It is also possible that the neutralization capacity of IgA2 in the airways of ICU patients may not be as effective as nasal IgA1 or blood IgG1. Indeed, we showed that nasal IgA1 equally binds IFN-α2a and IFN-α5 whereas systemic IgG1 effectively binds IFN-α2a but not IFN-α5. This suggests that different autoantibody isotypes may originate from different B cell clones or antibody class-switching may induce additional B-cell receptor mutations that affect their binding and neutralization capacity. Thus, the type of epithelial cell infected, the amount and duration of IFN-α production, and the clone of autoantibodies against IFN-α subtypes may all contribute to determining disease outcome and together may explain the different immunopathology in mild or moderate versus severe COVID-19.

Our study also has limitations. Although our neutralization assays suggest that anti-IFN-α autoantibodies may neutralize IFN-α signaling, they can only report the sample effects on a reporter cell line in vitro. These assays generally cannot reveal the effects of anti-IFN-α in the airway mucosal in vivo, which comprise a heterogeneous population of epithelial cells surrounded by cytokines, antibodies, and other unknown effector molecules absent from in vitro systems. In addition, purifying IgA from airway samples was not feasible due to limited sample volume and because anti-IFN-α is associated with local IFN-α cytokine; the purified IgA could include varying amounts of antibody-bound IFN-α. This limitation makes functional assays that rely on IFN-α as readout challenging to interpret. Although many types of IFN can be protective against SARS-CoV-2 (8, 34), our studies are limited to autoantibodies against type I IFN, particularly anti-IFN-α. Although our longitudinal samples revealed the viral-induced and transient nature of nasal anti-IFN-α associated with efficient recovery in mild and moderate patients, our sample collection for severe patients in ICU was limited. The lack of longitudinal samples from this cohort before hospital admission prevents us from determining whether their anti-IFN-α was pre-existing or viral-induced. Comparisons of the airway immunity between mild and moderate versus severe patients were based on different sample types, i.e., nasal swabs versus ETA. We also could not isolate the IgA2 from the airways of patients in the ICU to determine whether ETA IgA2, unlike nasal IgA1, represents a distinct clone of B cells with different avidity and neutralization capacity. Although we found associations between systemic IgG1 and disease severity, future studies should determine the functional and clonal differences between nasal IgA1 and systemic IgG1 autoantibodies in the immunopathology of COVID-19. Finally, our results do not exclude the possibility of another functional role for anti-IFN-α in the nose, or that factors beyond anti-IFN-α may contribute to patient outcomes.

Altogether, our findings challenge the notion that anti-IFN-α autoantibodies are solely a pathologic feature of COVID-19. Instead, these data support a model in which nasal IgA1 anti-IFN-α may serve an important regulatory function to counteract the detrimental effects of excessive IFN-α and restore homeostasis following SARS-CoV-2 invasion of the respiratory mucosa. Our data also challenge the concept that systemic anti-IFN-α is a risk factor for life-threatening COVID-19. Instead, we show that systemic IgG1 anti-IFN-α is associated with disease severity in all patient groups, including patients with moderate symptoms that efficiently recover from COVID-19 without hospitalization. Future studies should resolve the opposing roles of airway versus blood anti-IFN-α in COVID-19 and determine whether progression to systemic IgG1 anti-IFN-α can predispose recovered patients to subsequent viral infections or PASC. Moreover, subsequent studies should determine whether nasal anti-IFN-α autoantibody response is a common mechanism of host protective antiviral immunity in other respiratory viral infections, including influenza and other viruses that infect mucosal sites.

METHODS

Study design:

At the onset of the COVID-19 pandemic, we developed a prospective longitudinal cohort study to evaluate clinical, demographic, behavioral, immunologic, and genetic risk factors associated with COVID-19 acquisition, recovery, and severity. This study included 126 participants enrolled through the University of California San Francisco (UCSF)’s COVID-19 Host Immune Response Pathogenesis (CHIRP) cohort, Emory University’s PROATECT (Profiling of Antibody-secreting cells and Tissue-resident myeloid cells: Emory COVID-19 Task Force) cohort previously described (1), and the pre-pandemic and healthy control samples were provided by the Clinical and Translational Discovery Core at Children’s Healthcare of Atlanta (CHOA) and Emory University under IRB 00089506. All participants were assigned a new, sequential ID in the format “ERxxx.” IDs are known only to the study investigators as per Emory IRB 00003368.

For the CHIRP cohort, the inclusion criteria were SARS-CoV-2 RT-PCR nasal swab positive results regardless of symptoms. We set up an off-campus location and mobile study van to enroll RT-PCR+ individuals. Participants had matched nasal swabs and serum biospecimens collected between 4/2020 and 11/2022 (table S3, data file S1). All participants provided written informed consent. Longitudinal biospecimens were collected at weeks 0, 1, 3, 10, and 24 from baseline visits. A subset of participants also provided additional samples post-vaccination (2 weeks post-dose #1, 2 weeks post-dose #2, 8 weeks post-dose #1 and 2 weeks post-dose #4). Participants completed a detailed questionnaire, an adapted Somatic Symptom Scale-8 (SSS-8), at each study visit, including demographic, social, and behavioral history, medical history, concomitant medication use, and symptom severity reporting as previously described (54) (table S3). The date of symptom onset (or lack of symptoms if asymptomatic disease) was recorded at the baseline visit. COVID-19 symptom severity (mild versus moderate) was defined based on symptom burden (total number of assessed symptom categories) and duration (weeks of symptoms) (table S3, data file S1). Samples for clinical laboratory tests, including SARS-CoV-2 viral load (nasal swab), complete blood count with differential, comprehensive metabolic panel, erythrocyte sedimentation rate (ESR), and high sensitivity C-reactive protein (hs-CRP) were collected at each visit, as well as whole blood for processing for plasma and serum. This study was approved by the UCSF and Emory University Committee on Human Subjects Research (UCSF IRB 20-30588; Emory IRB 00003368). All participants provided written informed consent, and all biospecimens used for research were de-identified.

For the Emory PROATECT cohort, participants were identified by confirmed SARS-CoV-2 RT-PCR test results and enrolled as previously described (1) (tables S3 and S4, data files S1 and S2). Briefly, patients with severe, life-threatening COVID-19 were recruited from the ICU of Emory University, Emory St. Joseph’s, Emory Decatur, and Emory Midtown Hospitals between 6/2020 and 2/2021 (table S4, data file S2). One patient (ERE242) donated blood after ICU discharge and was not included in this study (data file S2). We also recruited outpatients with mild and moderate COVID-19 in the Emory Acute Respiratory Clinic. Outpatients were classified by disease severity (mild or moderate) following NIH guidelines (35) using symptom data collected by a healthcare provider (tables S3, data file S2). Infection-naïve patients were recruited for pre- and post-vaccination antibody assessments in blood and nasal swabs (table S3, data file S1). Informed consent was obtained for all study participants. Emory’s IRB approved all studies under protocol numbers 00003368, 00058507, 00057983, and 00058271.

Plasma and serum processing:

Samples were collected for all study participants, regardless of COVID-19 diagnosis or severity. To obtain serum, whole peripheral blood was collected in BD vacutainer (gold top) tubes and allowed to clot, and supernatant serum was isolated and cryopreserved. For plasma collection, whole blood was collected in EDTA tubes (BD vacutainer, purple top), and plasma was isolated by centrifugation at 400 × g for 10 minutes at 4°C followed by removal of the supernatant. To further precipitate platelets, contaminating cells, or remaining debris, the isolated plasma and serum samples were centrifuged at 4000 × g for 10 minutes at 4°C. Samples were distributed in aliquots and cryopreserved at −80C. The Human SARS-CoV-2 Serology Standard was obtained by request from the Frederick National Laboratory.

Nasal swab processing:

Samples were collected with flocked nasal swabs from both nostrils and transferred to 3 mL of viral preservation media (BD Universal Viral Transport media). Swabs (in media) were vortexed to liberate collected antibodies into media. Samples were further mechanically dissociated using a syringe and 25 Ga needle to disrupt aggregates of mucins and other respiratory secretions. Samples were centrifuged at 2000 × g to precipitate non-soluble aggregates. Processing of samples from SARS-CoV-2 positive donors was performed within a biosafety cabinet following BSL2+ protocols approved by the Emory University Institutional Biosafety Committee and Biosafety Officer.

Endotracheal aspirate (ETA) processing:

Samples were collected by healthcare providers in the ICU from either endotracheal intubation or tracheostomy (table S4). ETA samples were transferred to a USDA-approved BSL3 containment facility at Emory for further processing. To disrupt mucin aggregates, ETA was mixed 1:1 with a 50 mM EDTA solution (VWR International) (final concentration 25 mM EDTA) in custom RPMI-1640 media deficient in biotin, L-glutamine, phenol red, riboflavin, and sodium bicarbonate (defRPMI-1640) and dissociated using a syringe as previously described (1). Samples were centrifuged at 250 × g, and supernatants were cryopreserved at −80°C. Before removal from BSL3 containment, ETA supernatants were UV-inactivated following our published protocols (55).

Antigen preparation and conjugation:

SARS-CoV-2 structural and non-structural proteins, type I IFN proteins, and the 9G4 antibody clone to detect VH4-34 autoantibodies were either produced in-house, in collaboration with ChemPartner using the expression constructs from Quantitative Biosciences Institute (QBI) – Coronavirus Research Group (24) or obtained from commercial sources (table S1). All antigens, including a purified BSA control antigen, were biotin-conjugated using EZ-link NHS-biotin reagents (Thermo Scientific) at a 20-fold molar excess. Antigens were loaded onto commercially sourced streptavidin-coated beads (Spherotech) at an empirically determined saturation point in 50 µL of PBS + 0.05% BSA + 0.2% Tween-20 (PBS-BSA) and allowed to bind for 20 minutes on ice before the excess (uncaptured) antigen was removed by washing with additional PBS-BSA. We experimentally determined the antigen-bead saturation point for the SARS-CoV-2 receptor binding domain (RBD), S1 subunit, and S2 subunit, as well as for the IFN-α2a and IFN-ω antigens. All were found to have a saturation point of 2 pmol per 50,000 beads in 50 µL of PBS-BSA. We then confirmed the capacity of saturated beads to capture antigen-specific antibodies by flow cytometry using serial dilution of mouse monoclonal antibodies (details in table S2). Antigen-specific binding was revealed by a secondary stain with goat polyclonal antibodies fluorescently conjugated with phycoerythrin (PE) or Alexa Fluor (AF) 488 (table S2). Each antigen-specific antibody had a maximum signal output at least 1000-fold greater than the unstained control, which decreases linearly with serial dilution.

Fluorescent detection antibodies:

Monoclonals anti-IgG1 and anti-IgA2 were purchased pre-conjugated from Southern Biotech (table S2) and anti-IgE and anti-IgM were purchased pre-conjugated from BD Biosciences (table S2). Monoclonals anti-IgG2, anti-IgG3, anti-IgG4 and anti-IgA1 were purchased unconjugated from Southern Biotech and conjugated using Lightning-Link fluorescent conjugation kits (Abcam) following the manufacturer’s protocol. Each lot of antibody was individually titrated to identify the dilutions used for our assay (table S2).

FlowBEAT assays:

After thawing, samples were centrifuged at 2,000 × g for 10 minutes at 4°C to precipitate any remaining aggregates. Serum or plasma samples were diluted to a final concentration of 1:125 with PBS-BSA in a reaction mix containing beads coated with antigen at an empirically determined saturation point. Airway samples were diluted to a final concentration of 1:2. Samples and beads were incubated on ice in the dark for 30 minutes, mixing at least every 15 minutes to prevent beads from settling. Beads were washed with PBS-BSA and resuspended before fluorescent antibody staining. Antibody staining was performed by mixing equal volumes of resuspended beads and a master mix containing the eight fluorescently conjugated monoclonal antibodies specific to IgG1, IgG2, IgG3, IgG4, IgA1, IgA2, IgE and IgM described above and in table S2. The optimal concentration of each antibody species was empirically determined by titration and is reported in table S2. The reaction was incubated on ice in the dark for 30 minutes, mixing at least every 15 minutes. Stained beads were washed with PBS-BSA and resuspended in 1X PBS. Samples were fixed in 1:6 final dilution of FACS Lysing solution (BD Biosciences), which we showed inactivates SARS-CoV-2 (55). During fixing, samples were kept in the dark at room temperature for 20 minutes, mixing at least every 10 minutes. A final wash with PBS-BSA was performed, and beads were resuspended in PBS-BSA. All samples were analyzed within 24 hours of staining using a 5-laser Cytek Aurora flow cytometer. All data collection occurred on the same instrument, and the cytometer was calibrated with a consistent lot of Cytek SpectroFlo quality control (QC) beads before all data collection. To standardize each FlowBEAT batch, we established a daily standard plasma sample with strong reactivity to SARS-CoV-2 spike protein subunits. The standard positive and a sample-free control were included in all FlowBEAT assays to confirm the assay’s performance and control for potential variations. To ensure the reproducibility and rigor of our assays, we ran our samples in duplicate or triplicate in each experiment, including our daily standard calibration sample of known anti-SARS-CoV-2 values and BSA-coated control beads in each sample to account for background noise signal. Signals can be further corrected by removing the expected component of signal derived from non-specific BSA-binding, as informed by linear regression between BSA and SARS-CoV-2 antigens in pre-pandemic-collected samples (fig. S1E). Most samples were analyzed twice on different days using freshly prepared antigen-coated beads. The patient sample repeats, and the daily standard calibration values showed high concordance (Fig. 1B, fig. S1C, D).

Interferon neutralization assays:

HEK-Blue IFNα/β cells (InvivoGen, catalog # hkb-ifnabv2), a commercially available human embryonic kidney cell line engineered to report on type I IFN activity by secreted embryonic alkaline phosphatase (SEAP), were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco) enriched with 10% fetal bovine serum (FBS, Gibco), 1% penicillin-streptomycin and selective antibiotics included with the HEK-Blue cells (blasticidin, zeocin, and normocin, InvivoGen) following manufacturer-recommended protocol. Cells were seeded at 50,000 cells per well in 96-well plates and allowed to adhere for up to 4 hours (80% confluency) at 37°C in a humidified atmosphere containing 5% CO₂ using media free from selection antibiotics. Media was gently aspirated and replaced by donor samples in DMEM containing 60 activity units per mL (U/mL) of IFN-α2a (Acro Biosystems, catalog # IFA-H5253) per well. Nasal swabs were incubated at a final concentration of 50%, and plasma or serum was incubated at a final concentration of 20% for one hour, after which samples were gently aspirated and replaced with antibiotic-free DMEM media.

To detect SEAP activity, conditioned media was collected after overnight incubation. 20 μL of media (from a 200 μL total volume/well) was added to 180 μL of QUANTI-Blue solution (InvivoGen, catalog # rep-qbs), prepared according to the manufacturer’s recommendations. This mixture was incubated for an additional 1.5 hours under the same conditions to allow for the development of the colorimetric reaction. The optical density (O.D.) of each well was measured at 620 nm using a spectrophotometer (BioTek Synergy HTX) to quantify the IFN-induced activity. Measurements of IFN-free (standard negative) and sample-free, IFN-containing (standard positive) conditions were used to define the range of each assay. Cell viability and density were assessed using a Nexcelom cell counter that reveals live and dead cells.

Cytokine quantification:

The concentration of human cytokines in serum, plasma, nasal swabs, and ETA was measured using a Meso Scale Discovery U-plex custom multiplex assay kit, as previously described (1). We measured an additional ten analytes (IFN-α2a, IFN-γ, IL-15, IL-29/IFN-λ1, CXCL10/IP-10, MCP-1, MDC, MIP-3β, SDF-1a, and TRAIL) following the manufacturer’s protocol (Meso Scale Discovery) in plasma, nasal swabs, and ETA. These samples were prepared following our established UVC-inactivation protocol (55).

Statistical analysis:

Individual-level data are presented in data file S3. Flow Cytometry Standard (FCS) files were gated and analyzed in FlowJo version 10 software (BD Biosciences) to identify median fluorescence intensity (MFI) values for each antigen-specific antibody isotype fluorescence (example gating strategy in fig. S1A). Exported MFI values were background-corrected using a linear regression correlating background antigen-specific signal (as determined in pre-pandemic donors) and control BSA signal. Anti-spike protein signals were calculated as the mean anti-S1, −S2, and -RBD signal (MFI) for each antibody isotype. Plots were generated using R version 4.2.2 using the packages “ggplot2”, “ggcyto,” and “flowcore.” Statistical tests were performed using R version 4.2.2, and p-values were generated using Student’s T test, ANOVA, or Wilcoxon rank-sum test, as noted in figure legends. Horizontal bars represent the mean and uncertainty measures when comparing groups are plotted as standard error bars. Reported correlation r-values and p-values were generated using Pearson’s product-moment correlation. The uncertainty in the linear regression is represented by a shaded area corresponding to the confidence interval, as drawn by the ggplot function “geom_smooth”. The code required to reproduce figures and statistical analyses is available as R notebook at the Ghosn Lab GitHub (github.com/Ghosn-Lab/FlowBEAT) and deposited in Zenodo (doi.org/10.5281/zenodo.1388696) (56).

Data and materials availability:

All data associated with this study are in the paper or supplementary materials.