Research ArticleCORONAVIRUS

Longitudinal antibody repertoire in “mild” versus “severe” COVID-19 patients reveals immune markers associated with disease severity and resolution

See allHide authors and affiliations

Science Advances  05 Mar 2021:
Vol. 7, no. 10, eabf2467
DOI: 10.1126/sciadv.abf2467

Abstract

Limited knowledge exists on immune markers associated with disease severity or recovery in patients with coronavirus disease 2019 (COVID-19). Here, we elucidated longitudinal evolution of SARS-CoV-2 antibody repertoire in patients with acute COVID-19. Differential kinetics was observed for immunoglobulin M (IgM)/IgG/IgA epitope diversity, antibody binding, and affinity maturation in “severe” versus “mild” COVID-19 patients. IgG profile demonstrated immunodominant antigenic sequences encompassing fusion peptide and receptor binding domain (RBD) in patients with mild COVID-19 who recovered early compared with “fatal” COVID-19 patients. In patients with severe COVID-19, high-titer IgA were observed, primarily against RBD, especially in patients who succumbed to SARS-CoV-2 infection. The patients with mild COVID-19 showed marked increase in antibody affinity maturation to prefusion SARS-CoV-2 spike that associated with faster recovery from COVID-19. This study revealed antibody markers associated with disease severity and resolution of clinical disease that could inform development and evaluation of effective immune-based countermeasures against COVID-19.

INTRODUCTION

SARS-CoV-2 causes coronavirus disease 2019 (COVID-19), a severe disease in humans, and remains a global public health challenge. The ongoing SARS-CoV-2 pandemic has resulted in 1 million deaths and >35 million cases as of September 2020 (1). Therefore, development of effective vaccines and therapeutics for SARS-CoV-2 infection is a high global priority. Multiple vaccine candidates based on SARS-CoV-2 spike are being evaluated in preclinical and clinical studies (2). It is postulated that immune responses generated during SARS-CoV-2 infection likely provide protection among survivors, and therefore, convalescent plasma is being evaluated as therapeutics in several clinical studies (3, 4). Recent studies on SARS-CoV-2 infection in humans used monoclonal antibody (mAb) as surrogate readouts (57). However, there is no mapping of post–SARS-CoV-2 infection polyclonal antibody response in the ongoing COVID-19 pandemic. Moreover, there is limited knowledge about antibody repertoire following acute SARS-CoV-2 infection in patients with COVID-19 and its evolution over time to disease resolution or death due to COVID-19. Therefore, deeply characterizing antibody responses in “mild” versus “severe” COVID-19 patients may help identify immune markers involved in viral disease or recovery that may facilitate development and evaluation of vaccine and therapeutics against SARS-CoV-2.

Enzyme-linked immunosorbent assays (ELISAs) and neutralization tests have primarily been used to characterize antibody responses in humans following SARS-CoV-2 virus infection but provide limited insight into the diversity and quality of polyclonal antibody responses across the SARS-CoV-2 spike and its evolution over time following acute SARS-CoV-2 infection in patients with COVID-19 (811). These studies did not determine epitope specificity against the SARS-CoV-2 spike, nor did they closely characterize antibody kinetics during disease progression or resolution to delineate disease immune markers associated with diseases severity or symptom resolution.

The goal of our study was to perform a comprehensive longitudinal analysis of the humoral immune response following acute SARS-CoV-2 infection in critically ill patients with severe COVID-19 (hospitalized for >30 days with ventilator use and tracheal intubation) who either died or survived compared with patients with mild COVID-19 who recovered early (no ventilator use and <30 days in hospital) so as to identify antibody response associated with COVID-19 disease severity or symptoms resolution. Quantitative and qualitative analyses of immunoglobulin M (IgM), IgG, and IgA antibodies were performed on 115 longitudinal human sera collected frequently from 11 acutely SARS-CoV-2–infected patients with mild or severe COVID-19 disease during acute illness, before peak in symptoms and before the viral decline, disease resolution, or death, during the entire duration of their hospital stay (2 to 10 weeks). In addition to viral load, cytokine levels, and plaque reduction neutralization test (PRNT)–based neutralization titers, the evolution of antibody epitope repertoires following SARS-CoV-2 infection was elucidated using whole SARS-CoV-2 spike gene-fragment phage display libraries (GFPDLs). Phage display technique is suitable for display of properly folded and conformationally active proteins as it has been widely used for display of large functionally active antibodies, enzymes, hormones, and viral and mammalian proteins. We have adapted this technology for unbiased, comprehensive gene-fragment phage display library (GFPDL) approach for multiple viral pathogens including SARS-CoV-2, ebola virus, highly pathogenic avian influenza virus, respiratory syncytial virus, and zika virus (ZIKV), to define antibody epitope repertoire of postvaccination/infection samples to define both linear and conformation epitopes (1219).

In addition, we used surface plasmon resonance (SPR) technology to measure real-time antibody binding kinetics, immunoglobulin isotypes, and affinity maturation against the various SARS-CoV-2 spike proteins and domains. This was done to delineate the immune markers that associate with disease severity or disease resolution in these patients with COVID-19. The findings identified a framework of immune signatures that could inform development and evaluation of effective SARS-CoV-2 vaccines and therapeutics against COVID-19.

RESULTS

Evolution of antigenic fingerprint generated following SARS-CoV-2 infection in patients with COVID-19

We evaluated 115 human serum or plasma collected every third day from 11 acutely hospitalized patients with COVID-19 either until death or discharge from the hospital following recovery from disease, with all days being relative to the day (D) of symptom onset (tables S1 and S2 and fig. S1). We began the study very early during the COVID-19 pandemic and focused on samples collected frequently without any experimental antibody or immunomodulatory treatment. All patients were symptomatic males between the ages of 25 and 81 years. During the clinical study, any individual positive for SARS-CoV-2 by reverse transcription polymerase chain reaction (RT-PCR) was admitted to hospital in Japan, even with minimal clinical symptoms (either low fever, discomfort, or others). We defined patients as mild versus severe on the basis of the criteria defined by Grein et al. (20). “Mild” disease was defined as patients who do not require supplemental oxygen and have oxygen-supportdata of <3 and a combined symptom score of ≤5 (table S2). Six hospitalized patients (S-01 to S-05 and S-C02) were hospitalized for >30 days (four fatal cases; S-01 died on D43, S-04 died on D35, S-05 died on D57, and S-C02 died on D86; S-02 was still in hospital on D48; and S-03 was discharged on D34), required ventilatory support and tracheal intubation with extracorporeal membrane oxygenation (ECMO) support (S-01, S-02, S-04, S-05, and S-C02) or without (S-03) were defined as severe COVID-19 cases (table S1). Five patients (M-10, M-14, M-15, M-16, and M-18) were defined as mild COVID-19 cases as they did not require oxygen supplementation, had a combined symptom score of ≤5, and recovered early with discharge in less than 25 days from hospital (tables S1 and S2).

The viral load determined by RT–quantitative PCR (qPCR) on the first day of hospitalization for these 11 patients ranged from 19.85 to 30.99 Cq values, which declined over time in all patients, but became repeatedly negative for most of the mild cases by 14 days after symptom onset, apart from M-15 (Fig. 1, A to K, and table S2). The SARS-CoV-2 neutralization titers during acute illness as measured by PRNT reached >100 PRNT50 (plaque reduction neutralization titer 50) for only the patients with severe COVID-19 (apart from S-01) but not for the early-recovered mild (M-series) patients (Fig. 1, A to K, and table S2) similar to PRNT titers observed in acute infection during hospitalization (11). High serum PRNT50 titers (>500) were observed for three (S-04, S-05, and S-C02) fatal cases (Fig. 1, D to F, and table S2). Quantitative analysis of 17 cytokine/chemokines was performed in these patients with COVID-19, but all were not elevated in the severe patients (table S3). Elevated interleukin-6 (IL-6) and IL-8, as well as chemokines monocyte chemoattractant protein-1 (MCP-1) and macrophage inflammatory protein-1beta (MIP-1β) were observed in the patients with severe COVID-19 compared with the patients with mild COVID-19 (Fig. 1, L to O) during their hospital stay. Area under the curve (AUC) analysis revealed significantly higher levels of IL-6, IL-8, and MCP-1 in the severe patients compared with the patients with mild COVID-19 (Fig. 1P).

Fig. 1 Viral load, serum neutralizing titers, and cytokine analyses of patients with COVID-19 during hospitalization.

(A to K) Viral load in upper respiratory tract of the 11 patients with COVID-19 at various time points as measured by RT–quantitative PCR (qPCR) (blue symbols). SARS-CoV-2–neutralizing antibody titers in serum/plasma for six severe (A to F; red symbols) and five mild (G to K; green symbols) COVID-19 patients at various time points as determined by PRNT50. (L to O) Cytokine levels (L, IL-6; M, IL-8; N, MCP-1; and O, MIP-1β) of fourfold diluted plasma/serum samples at various time points in the 11 patients with COVID-19 as analyzed via a Bio-Plex Pro Human Cytokine Panel 17-Plex assay. (P) AUC for the four cytokine/chemokines of the 11 patients with COVID-19. Limit of detection (LOD) for the assay is shown as dashed line. The statistical significances between the groups were determined using Kruskal-Wallis multiple comparisons test for AUC (AUC values) of severe patients (red) and mild patients (green). The differences were considered statistically significant with a 95% confidence interval when the P value was less than 0.05. *P < 0.05, **P < 0.001.

The polyclonal IgM, IgG, and IgA antibody epitope repertoire of post–SARS-CoV-2 infection antibodies was analyzed in individual longitudinal serum/plasma obtained at several time points after symptom onset for the deceased patient with severe COVID-19 (S-01) and the two early-recovered mild patients (M-10 and M-14) by GFPDL containing sequences ranging from 50 to 1500 base pairs (bp) long from the SARS-CoV-2 spike with >107.1 unique phage clones (Figs. 2 to 4). In addition, GFPDL-based IgM, IgG, and IgA antibody epitope repertoire analysis was performed on samples pooled from remaining three deceased patients (S-04, S-05, and S-C02) and compared with pooled samples from three remaining mild patients (M-15, M-16, and M-18) to identify the overall antibody epitope signature following SARS-CoV-2 infection (fig. S3). The SARS-CoV-2 GFPDL displayed linear and conformational epitopes with random distribution of size and sequence of inserts that spanned across SARS-CoV-2 spike sequence (fig. S2A). Recently, we showed that SARS-CoV-2 GFPDL can recognize linear, conformational, and neutralizing epitopes in the postvaccination sera (21). Moreover, in the current study, the SARS-CoV-2 GFPDL adsorbed majority (>88%) of SARS-CoV-2 prefusion spike–specific antibodies in the postinfection polyclonal human serum/plasma samples, providing proof of concept for use of the SARS-CoV-2 GFPDL for epitope repertoire analyses of human sera (fig. S2B). An uninfected control human serum (collected in 2008) bound very few (<20) phages in two experiments with the SARS-CoV-2 GFPDL (fig. S3). The uninfected control serum antibodies recognized sites in the N-terminal domain (NTD) of S2 before fusion peptide and the S1 NTD.

Fig. 2 IgM, IgG, and IgA antibody repertoires elicited in patient with severe COVID-19 who succumbed to SARS-CoV-2 infection.

(A) Distribution of phage clones after affinity selection on post–SARS-CoV-2 infection samples. Number of IgM-, IgG-, and IgA-bound phage clones selected using SARS-CoV-2 spike GFPDL on polyclonal samples from days 6 (D6), 21 (D21), and 42 (D42) following symptom onset in patient with fatal COVID-19 (S-01). (B to D) IgM, IgG, and IgA antibody epitope repertoire recognized in the SARS-CoV-2–infected serum/plasma of deceased patient with COVID-19 (S-01) at different days after onset of symptoms and their alignment to the spike protein of SARS-CoV-2. CD, connector domain; CH, central helix; CT, cytoplasmic tail; FP, fusion peptide; HR, heptad repeat; SP, signal peptide. Graphical distribution of representative clones with a frequency of ≥2, obtained after affinity selection, is shown. The horizontal position and the length of the bars indicate the peptide sequence displayed on the selected phage clone to its homologous sequence in the SARS-CoV-2 spike on alignment. The thickness of each bar represents the frequency of repetitively isolated phage. Scale value for IgM, IgG, and IgA is shown enclosed in a red box beneath the respective alignments. The GFPDL affinity selection data were performed in duplicate (two independent experiments by researcher in the laboratory, who was blinded to sample identity), and similar number of phage clones and epitope repertoire was observed in both phage display analysis.

Fig. 3 Evolution of IgM, IgG, and IgA antibody repertoire in patient with mild COVID-19.

(A) Distribution of phage clones after affinity selection in serum following SARS-CoV-2 infection. Number of IgM, IgG, and IgA bound phage clones selected using SARS-CoV-2 spike GFPDL on polyclonal samples from days 5 (D5), 12 (D12), 15 (D15), and 25 (D25) following symptom onset in patient with mild COVID-19 (M-10). (B to D) IgM, IgG, and IgA antibody epitope repertoire recognized in the SARS-CoV-2–infected serum/plasma at different days after onset of symptoms and their alignment to the spike protein of SARS-CoV-2. Graphical distribution of representative clones with a frequency of ≥2, obtained after affinity selection, is shown. The horizontal position and the length of the bars indicate the peptide sequence displayed on the selected phage clone to its homologous sequence in the SARS-CoV-2 spike on alignment. The thickness of each bar represents the frequency of repetitively isolated phage. Scale value for IgM, IgG, and IgA is shown enclosed in a red box beneath the respective alignments. The GFPDL affinity selection data were performed in duplicate (two independent experiments by researcher in the laboratory, who was blinded to sample identity), and similar number of phage clones and epitope repertoire was observed in both phage display analysis.

Fig. 4 Elucidation of IgM, IgG, and IgA antibody repertoire following SARS-CoV-2 infection in patient with mild COVID-19.

(A) Distribution of phage clones after affinity selection with COVID-19 serum. Number of IgM, IgG, and IgA bound phage clones with days 1 (D1) and 14 (D14) serum following symptom onset of patient with mild COVID-19 (M-14). (B to D) SARS-CoV-2 IgM, IgG, and IgA epitope repertoire of the patient with mild COVID-19 (M-14). Graphical distribution of affinity-selected clones with frequency of ≥2 is shown. The horizontal position and length of bars indicate the peptide sequence displayed on selected phage clone to its homologous sequence in SARS-CoV-2 spike. Thickness of each bar represents frequency of repetitively isolated phage. Scale value for IgM, IgG, and IgA is shown in a red box beneath the respective alignments. (E to G) Antibody epitope profile following SARS-CoV-2 infection. Antigenic regions/sites within SARS-CoV-2 spike sequence (GenBank: MN908947.3) recognized by serum/plasma antibodies (based on data presented in Figs. 2 to 4). Antigenic sites shown in cyan were uniquely recognized by post–SARS-CoV-2 infection IgG (E), IgA (F), or IgM (G) only in patients with mild but not severe COVID-19. Antigenic sites shown in red were uniquely recognized by post–SARS-CoV-2 infection IgG (E), IgA (F), or IgM (G) antibodies only in patients with severe but not mild COVID-19. Epitopes of each protein are numbered in black.

In the deceased patient with severe COVID-19 (S-01), the number of bound GFPDL phages was similar for IgM and IgA antibodies at early time point (D6) (Fig. 2A). By D21, the IgM binding phage titer was 10-fold higher than either IgA or IgG antibodies. However, the spike-specific phage titers peaked by D21 but then declined about 2- to 10-fold by D42 in this patient, just before his death (Fig. 2A). In the deceased COVID-19 (S-01) patient, postinfection IgM antibodies showed a diverse epitope repertoire distribution, displaying small and large sequences spanning the entire spike in this patient on D6 (Fig. 2B). At days 21 and 42, IgM epitope profile evolved with preferential recognition of sites in S2 domain, followed by NTD in S1, and limited binding to epitopes in receptor binding domain (RBD).

The IgG response in this deceased COVID-19 (S-01) patient was minimal at D6 with few recognized epitopes at the N and C termini of S2 domain, which evolved to predominant S2 binding IgG on D21, with immunodominance to β-rich connector domain (CD) and the heptad repeat 2 (HR2) domain at the C terminus of S2, followed by the N terminus of S2 and the C terminus of S1. Minimal recognition of short epitope sequences in RBD and no binding to receptor binding motif (RBM) were observed (Fig. 2C). The total number of IgG-bound phages declined by D42 (Fig. 2A), and the IgG profile showed antibody binding primarily to immunodominant region in CD and the HR2 at the C terminus of S2.

IgA response to spike following acute SARS-CoV-2 infection (D6) in this patient with severe COVID-19 (S-01) recognized the RBD, before the RBM and S2 site encompassing central helix and the CD (Fig. 2D). By D21, IgA epitope repertoire evolved to focus on five immunodominant sites; two sites in the NTD of S1, strong recognition of RBD with some overlap of RBM, N-terminal site before the fusion peptide in S2 domain, and the HR2 site in S2 domain (Fig. 2D). By D42, the IgA response recognized the S2 domain site in HR2, a previously unknown site in the fusion peptide, as well as epitopes in the RBD and S1-NTD (Fig. 2D).

For the patient with mild COVID-19 (M-10), at D5 after onset, very few IgM binding phages were selected that continuously increased until D25 before his discharge from hospital (Fig. 3A). The IgM antibodies recognized a diverse epitope repertoire across the entire spike protein with increased IgM binding to sites in N and C termini of S1 and the N-terminal half of S2 (Fig. 3B).

Negligible IgG phage binding was observed on D5 and minimal on D12 that increased substantially by D15 and continued to increase until D25 for this patient with mild COVID-19 using SARS-CoV-2 GFPDL (Fig. 3A). The IgG epitope repertoire on D15 and D25 after onset of symptom in this individual (M-10) showed strong recognition to C-terminal CD-HR2 sites in S2 domain (Fig. 3C). However, in contrast to the deceased patient with severe COVID-19 (S-01) or pooled severe patients, this mild recovered patient (M-10) generated a strong IgG recognition to sites containing the fusion peptide in S2 domain observed on both D15 and D25. Moreover, the IgG antibody profile showed specificity for large sequences in the RBD encompassing RBM (probably conformational epitopes) at D15 and D25 (Fig. 3C) in contrast to IgG binding to preferentially short RBD sequences for the deceased patients with COVID-19 (S-01) (Fig. 2C).

In contrast to the deceased patient with severe COVID-19 (S-01), this mild recovered patient (M-10) recognized 100-fold lower IgA phage titers compared with IgG phage titers on both D15 and D25 (Fig. 3A). IgA antibodies primarily recognized the CD-HR2 sites in the C terminus of S2 domain followed by large sequences within the RBD encompassing RBM, on both D15 and D25 in this patient with mild COVID-19 (Fig. 3D).

The second patient with mild COVID-19 (M-14), IgM/IgG/IgA antibodies bound very few clones in the SARS-CoV-2 spike on D1 after onset of symptoms that increased remarkably (100- to 1000-fold) by D14 just before his discharge from hospital (Fig. 4A). IgM epitope repertoire induced by SARS-CoV-2 infection in this individual was diverse, similar to other patients with COVID-19 (Fig. 4B). In contrast to the deceased patient with severe COVID-19 (S-01) and like the other mild patient (M-10), the IgG antibodies in this mild patient (M-14) on D14 preferentially recognized immunodominant site containing the fusion peptide in S2 domain, followed by long sequences encompassing the complete RBD, in addition to the CD-HR2 domain of S2 (Fig. 4C). As observed for the other mild patient (M-10), the IgA phage titers in this patient (M-14) were ~80-fold lower on D14 compared with IgG phage titers (Fig. 4A) that primarily recognized large antigenic sites in the N terminus of S2 domain containing the fusion peptide sequence followed by S1-NTD (Fig. 4D).

In addition to individual samples, GFPDL analysis was also performed for samples pooled from the three severe fatal cases (S-04, S-05, and S-C02) versus three mild patients (M-15, M-16, and M-18) at around 2 weeks after onset of symptoms (fig. S4). The analysis of IgM, IgG, and IgA epitope repertoire demonstrated that the differential antibody signature profile for the pooled samples was similar to that observed for the individual deceased patient with severe COVID-19 (S-01) or mild patients (M-10 and M-14). While the patients with “fatal” COVID-19 recognized 100-fold (IgM) and 1000-fold (IgA) higher phages compared with pooled serum from mild cases, the differences in IgG-bound phage clones was only 2-fold. Similar to the individual sample analyses, the pooled mild patient IgG showed specificity for large sequences encompassing RBM and the S2 fusion peptide in contrast to IgG binding from patients with fatal COVID-19 that preferentially recognized smaller RBD epitopes (fig. S4).

SARS-CoV-2 infection in these COVID-19 longitudinal patient samples generated a diverse antibody response across the spike protein that was defined by 12 antigenic regions and 13 antigenic sites contained within these regions (Fig. 4, E and F, and table S4). Antigenic regions of 35 to 250 amino acid residues were defined on the basis of antibody recognition by at least 4% of phage clones obtained after affinity selection on IgM/IgG/IgA antibodies with at least one serum/plasma sample at any time point. Most regions/sites were recognized by antibodies across samples from both patients with severe and mild COVID-19 (shown in black in Fig. 4, E and F). Site S5.2 in RBD and antigenic region S7 in S1 domain were only recognized by the IgA antibodies (Fig. 4F), while two sites (S10 and S10.2) were only recognized by IgM antibodies in these patients with COVID-19 (Fig. 4, E and F). Comparing IgG epitope repertoire between the patients with severe versus mild COVID-19 revealed that site 5.2 (representing short epitopes; shown in red in Fig. 4E) in RBD before RBM was preferentially recognized by the severe patient antibodies, whereas, the mild patients contained IgG that bound large RBD sites S5, S5.1, and S6 encompassing the RBM as well as sites S9.3 and S9.5 containing a fusion peptide in S2 domain (shown in blue in Fig. 4E). For IgA, the deceased patients with severe COVID-19 preferentially recognized site S7 at the C terminus of S1 domain (in red), while the patients with mild COVID-19 recognized antigenic region S9 (containing the fusion peptide) and S11 in S2 domain (in blue) at higher frequency (Fig. 4F). Most of these antigenic regions/sites in the S1 domain were not conserved among other CoV strains; however, some sites in S2 showed >50% sequence conservation across multiple human and bat CoV strains (fig. S5 and table S4). Structural depiction of these antigenic sites on the SARS-CoV-2 spike (fig. S6; on Protein Data Bank number 6VSB) demonstrated that several of these antigenic sites identified here are surface-exposed on the native prefusion spike (22).

Longitudinal analysis of antibody binding kinetics of post–SARS-CoV-2 infection samples to spike proteins and domains

To understand the evolution of antibody response following SARS-CoV-2 infection in these 11 patients with COVID-19 during their hospital stay, quantitative and qualitative SPR analyses were performed for several dilutions (10-, 50-, and 250-fold) of all polyclonal serum/plasma using recombinant purified prefusion spike protein and its domains, S1, S2, and RBD (21). The conformational integrity of the spike proteins used in our SPR assay was assessed by human ACE2 (hACE2) receptor-binding assay (fig. S7). The purified proteins used in SPR antibody analysis—prefusion spike, the S1 domain, and the RBD proteins—demonstrated high-affinity interaction receptor-binding activity to hACE2 protein, the SARS-CoV-2 receptor, with affinity constants ranging from 6.9 to 28.4 nM (fig. S7). The S2 domain protein lacking the RBD did not bind to hACE2. Furthermore, these proteins were immunized into rabbits that generated neutralizing antibodies as expected from native folded functionally active proteins, wherein S2 domain generated weak neutralizing antibodies compared with other spike domains (21). Representative sensorgrams for different serial dilutions (10-, 50-, and 250-fold) of sample antibody binding to prefusion spike are shown in fig. S8A. The optimized SPR does not show nonspecific background reactivity with unimmunized or irrelevant immunized sera to SARS-CoV-2 spike proteins (21). Control human serum samples collected from 50 healthy adults in 2008 showed <4 resonance units (RU) binding to the SARS-CoV-2 prefusion spike protein in SPR (fig. 8B).

Antibody binding titers to most spike proteins were observed at earlier time points after onset of symptoms and at higher titers for the patients with severe COVID-19 compared with patients with mild COVID-19 (Fig. 5, A to D). The antibody titers gradually increased and peaked around days 18 to 21 for most patients against all the four spike proteins and domains (fig. S8). The antibody titers then declined remarkably against the prefusion spike, S1 domain, and RBD for the patients with severe fatal COVID-19 (S-01, S-05, and S-C02), while the anti-S2 domain antibody levels declined marginally for most patients (Fig. 5, A to C, versus Fig. 5D). For the other patient with severe COVID-19 (S-02), similarly, the anti-S1 and anti-RBD antibodies decline rapidly after peak on D18; however, the anti-S2 domain antibodies remained at appreciable binding levels until D48 (19 RU for anti-RBD versus 93 RU for anti-S2) (Fig. 5, C and D, and fig. S9). For this patient, the prefusion spike antibodies dropped substantially from days 26 to 29 (466 to 198 RU) but then increased on D32 and remained at high levels until D48 (Fig. 5A). The non-ECMO patient with severe COVID-19 (S-03) demonstrated minimal decline of anti-S2 antibodies and moderate decline for prefusion spike, S1, and RBD-binding antibodies. SARS-CoV-2 infection–induced antibodies in the patients with mild COVID-19 appeared around days 10 to 12 after symptom onset, a few days later than ECMO patients with severe COVID-19 (Fig. 5, A to D). The antibody titers for mild patients were lower for prefusion spike, S1, and RBD but not for S2 domain compared with those for patients with severe COVID-19. The anti-spike antibody titers kept increasing in the mild patients until the day of discharge. Increase in antibody binding titers to SARS-CoV-2 prefusion spike protein for the non-ECMO patient with COVID-19 (S-03) and all five mild patients (delayed for M-16) were associated with concomitant decline in clinical symptom scores in patients with COVID-19 resulting in disease resolution (fig. S10).

Fig. 5 SPR-based analysis of human antibody response following SARS-CoV-2 infection.

Serial dilutions of each serum/plasma sample collected at different time points from patients with COVID-19 were analyzed for antibody binding to SARS-CoV-2 spike and subdomains (figs. S7 and S8). (A to D) Total antibody binding is represented in SPR RU for six patients with severe COVID-19 (in shades of red) and five mild patients (in shades of green) for binding to prefusion spike (A), S1 domain (B), RBD (C), and S2 domain (D). Total antibody binding shown is observed maximum RU for 10-fold diluted serum/plasma sample. (E) Antibody isotype of SARS-CoV-2 prefusion spike–binding antibodies. The isotype composition of serum/plasma antibodies bound to prefusion spike. The RU for each anti-spike protein antibody isotype was divided by total RU for all antibody isotypes combined to calculate percentage of each antibody isotype for individual serum/plasma sample. (F) Mean percentage of IgM, IgG, and IgA antibody isotype bound to SARS-CoV-2 prefusion spike is shown for patients with severe (red) versus mild (green) COVID-19. The statistical significances between groups were determined using Kruskal-Wallis multiple comparisons test for patients with severe (red) versus mild (green) COVID-19. The differences were considered statistically significant with 95% confidence interval when P value was less than 0.05. **P < 0.01.

Class-switching of spike-binding antibodies following SARS-CoV-2 infection

Isotype analysis of spike-binding antibodies performed by SPR demonstrated the presence of IgM, IgA, and IgG following SARS-CoV-2 infection in these patients with COVID-19 (Fig. 5E and fig. S11). The absence of antibody binding to prefusion spike for 50 uninfected control human samples (collected in 2008) in SPR that measures binding of all antibody isotypes excluded the possibility that IgM binding may be due to polyreactive natural antibodies (i.e., sticky antibodies not induced by SARS-CoV-2) in SPR. On D6, in the deceased patient with severe COVID-19 (S-01), most of the prefusion spike–binding antibodies were of IgM isotype, which class-switched gradually with increasing contribution from IgA until D21 followed by IgG isotypes over time (Fig. 5E and fig. S11). D9 onward, 30 to 50% of prefusion spike antibodies were contributed by the IgA isotype, while D27 onward, IgG1 and IgG3 subclasses contributed >10% of prefusion spike–binding antibodies with minimal IgG2 and IgG4 binding in this deceased patient with COVID-19 (fig. S11). While S1- and S2-binding antibodies were initially IgM and IgA, the response class-switched over time with equal contribution from IgM, IgA, and IgG1 isotypes (fig. S11). Unexpectedly, 40% of RBD-binding antibodies were of IgA isotype on D6, which increased substantially to >70% by D15 and reached over 90% IgA on D42, before demise of this patient with severe COVID-19. A similar RBD-binding antibody isotype profile was observed in the second patient with severe COVID-19 (S-02) with 30% IgA (at D11) to 60% IgA by D47 (fig. S11). This patient had high anti-spike binding antibodies at the first time point (D11 after onset of symptoms) that were contributed mostly by IgG1 followed by IgA across all four proteins during the duration of hospital stay. In contrast, the spike-binding antibodies of the non-ECMO patient with severe COVID-19 (S-03) were primarily IgM followed by IgG1 and low IgA. The prefusion spike–binding antibodies for the two fatal cases (S-04 and S-05) demonstrated 40 to 60% IgA followed by IgM and only 7 to 26% IgG isotype (Fig. 5E). The longest hospitalized patient with COVID-19 (S-C02) showed a high IgG isotype binding to prefusion spike with equal contribution from all IgG subclasses, followed by IgM and then IgA. However, in the two fatal cases (S-05 and S-C02), the IgA antibodies to RBD (but not anti-S2) increased gradually overtime reaching ~50%, before the demise of these two patients with COVID-19 (fig. S11).

In patients with mild COVID-19 (M-10, M-14, M-16, and M-18), SARS-CoV-2 infection–induced anti-spike antibodies were primarily IgM on D12 after onset of symptoms that class-switched to 20 to 50% contribution from IgG1, with lower IgA and minimal IgG3 at later time points in these patients with mild COVID-19 (Fig. 5E and fig. S11). Prefusion spike–binding antibodies showed higher class switching to IgG (>50% by D14) with contribution from all IgG subclasses in mild patient M-15 who had the longest stay in the hospital (discharged on 23 days after onset of symptoms). Comparison of percent isotype contribution to prefusion spike–binding antibodies demonstrated significantly lower IgA and higher IgM isotype in the mild patients compared to patients with severe COVID-19 (Fig. 5F).

Affinity maturation of spike-binding antibodies following acute SARS-CoV-2 infection in patients with COVID-19

To determine the antibody affinity maturation over time against different spike proteins/subdomains following virus infection, the dissociation kinetics (off-rate constants) of antigen-antibody complexes that are independent of antibody concentration were used as a surrogate for overall average affinity of polyclonal antibody against spike proteins using SPR (14, 18, 19, 21, 23). Technically, because antibodies are bivalent/multivalent, the proper term for their binding to multivalent antigens such as viruses is avidity, but here, we use the term affinity throughout because we measured primarily monovalent interactions (18, 21, 23).

The off-rates of polyclonal antibodies bound to spike proteins were fast (between 0.1 and 0.01 per second) at initial time points after SARS-CoV-2 infection, indicating weak antibody affinity early during illness that matured thereafter although differentially against various spike proteins and domains in these patients with COVID-19 (Fig. 6, A to D, and fig. S11). Anti-spike antibodies in patient with severe COVID-19 (S-01) demonstrated minimal affinity maturation from D6 until demise to either prefusion spike or the three spike domains (fig. S11). In the second patient with severe COVID-19 (S-02), minimal anti-prefusion spike affinity maturation was observed through D26 (0.025 per second); however, on D29, a twofold increase in antibody affinity (0.0137 per second) and a gradual increase in antibody affinity through D47 (0.00834 per second) was observed (Fig. 6A). Similar observations were made for this patient’s antibodies against S1, RBD, and S2 domain (Fig. 6, B to D). For non-ECMO patient with COVID-19 (S-03) a minimal antibody affinity maturation was observed for first 4 weeks after onset of symptoms; however, an inflection point with fourfold increase in antibody affinity from D27 (0.0267 per second) to D31 (0.0079 per second) was observed against prefusion spike (Fig. 6A). This jump in affinity maturation from D27 to D31 was not evident for the antibodies against the spike S1, RBD, or S2 domains (Fig. 6, B to D). The antibody kinetics for other fatal cases (S-04, S-05, and S-C02) demonstrated weak affinity maturation (<10-fold) from initial sample until their demise against all the four proteins, with off-rate reaching only 0.01 per second, just before their demise (Fig. 6, A to E).

Fig. 6 Antibody affinity maturation of human antibody response following SARS-CoV-2 infection in patients with COVID-19 and association with disease severity.

(A to D) Polyclonal antibody affinity maturation to SARS-CoV-2 spike proteins was determined by SPR. Binding affinity of serially diluted after infection serum/plasma of each of the six patients with severe COVID-19 (in shades of red) and five mild patients (in shades of green) to prefusion spike (A), S1 domain (B), RBD (C), and S2 domain (D). All SPR experiments were performed twice, and data shown are average value of two experimental runs. Off-rate was calculated and shown only for samples that demonstrated antibody binding of 10 to 100 RU in SPR. (E) Antibody affinity (as measured by dissociation off-rate per second) against SARS-CoV-2 prefusion spike, S1, S2, and RBD for the final day samples from each of the severe (red) versus mild (green) COVID-19 patients. The statistical significances between the groups were determined using Kruskal-Wallis multiple comparisons test for severe patients (red) and mild patients (green). The differences were considered statistically significant with 95% confidence interval when P value was less than 0.05. *P < 0.05. (F to P) Relationship of serum antibody affinity against prefusion spike (blue symbols; off-rate per second) and clinical symptom scores for six severe (F to K; red symbols) and five mild (L to P; green symbols) at various days after onset of symptoms in patients with COVID-19.

For the five patients with mild COVID-19, a substantial affinity maturation was observed for antibodies binding to the prefusion spike from D12 after onset onward (approximately 2 to 4 weeks after SARS-CoV-2 exposure) reaching an antigen-antibody complex dissociation rate of low 10−3 per second, which was 5 to 10 times higher antibody affinity compared with patients with severe COVID-19 (Fig. 6, A to E). The affinity against the RBD was ~2-fold higher in “mild patients” compared with severe patients. Antibody affinity to S1 and S2 domain was not substantially different between the mild versus severe COVID-19 patients (fig. S12). Comparison of antibody affinity for the final time-point serum/plasma sample from these patients demonstrated significantly higher antibody affinity (slow antigen-antibody dissociation rate) only against the prefusion spike (but not S1, S2, or RBD) in the mild patients compared with patients with severe COVID-19 (Fig. 6E).

To identify the relevance of antibody affinity with clinical disease, prefusion spike–specific antibody affinity was plotted against the clinical scores overtime in these patients with COVID-19 (Fig. 6, F to O). Of the severe patients, only S-03 patient resolved his symptoms and was discharged from hospital. The marked increase in anti-prefusion spike antibody affinity from D27 to D31 (for patient S-03) coincided with decline in clinical score from D27 onward and eventual discharge by D34 (Fig. 6H and table S2). Similarly, predominant increase in antibody affinity to prefusion spike in each of the five patients with mild COVID-19 from D10 onward was associated with rapid decline in clinical symptom severity and disease resolution in these patients with COVID-19 (Fig. 6, K to O). In comparison, the antibody affinity maturation to prefusion spike was minimal or absent in any of the four patients who succumbed to the disease (Fig. 6, F and I to K). The clinical scores decreased before an increase in neutralizing titers was measured in the PRNT50 assay, indicating that neutralizing antibodies may not be the sole factor to control SARS-CoV-2 infection or that sensitivity of the PRNT50 assay is low in determining the functional immune response that curtails COVID-19 disease (table S2).

DISCUSSION

Differences in antibody kinetics between patients with COVID-19 with different disease severity have been reported, including virus neutralization titers and spike-binding antibodies following acute infection or during convalescence (24). However, the predictive value of neutralization titers with disease outcome has not been established, because patients with severe COVID-19 show much faster and stronger neutralizing antibodies compared with mild cases. Therefore, an immune marker that correlates with protection is not well defined and is critically needed to inform COVID-19 medical countermeasures during ongoing pandemic.

To our knowledge, this study represents a high-fidelity comprehensive longitudinal characterization of IgM, IgG, and IgA repertoire following acute SARS-CoV-2 infection frequently during acute illness until disease resolution or death to identify the immune markers that correlate with disease severity or resolution in severe versus mild COVID-19 patients. The primary findings were that while the severely infected individuals had higher overall binding and neutralizing antibody titers, it did not correlate with either positive clinical outcome or apparent recovery from COVID-19. However, there was a distinct increase in antibody affinity to the prefusion SARS-CoV-2 spike that associated with disease recovery. Patients with mild disease demonstrated an immunodominant IgG binding profile to epitopes encompassing the fusion peptide. The severely SARS-CoV-2–infected patients also had high plasma levels of IL-6, IL-8, and MCP-1 and sustained increased IgA compared with patients with mild COVID-19. Correlates of protection against SARS-CoV-2 infection may be different from correlates of protection from COVID-19 disease. Antibodies may not be a correlate of protection for primary SARS-CoV-2 infection, but they can be a good correlate of protection for COVID-19 disease and against secondary viral infection. The recent observation that mAb, when administered early during the COVID-19 disease, provided beneficial effect on clinical disease outcome (25) suggests that rapid generation of antibodies targeting protective sites and faster antibody affinity maturation following infection might be good correlates of protection against COVID-19.

The focus on spike was primarily because, currently, most SARS-CoV-2 vaccines or antibody-based therapeutics in clinical development target SARS-CoV-2 spike. Over the course of illness in these patients with COVID-19, we observed a differentially evolving diverse antibody response, in terms of antibody epitope repertoire, isotype class switch, and affinity maturation in mild versus severe disease patients during hospitalization. In the deceased patients, the IgG and IgA antibody epitope repertoire was primarily focused to sites in the N and C termini of S1 and S2 domains, with minimal IgG recognition of RBM in RBD or fusion peptide in the S2 domain. The enrichment of IgG binding to epitopes at the C terminus of S2 domain in these patients was peculiar, and studies are ongoing to understand the role of these antibodies in virus-host interaction. A bioinformatics approach identified potential B cell epitopes in the spike glycoproteins of SARS-CoV, based on human antibody responses to SARS-CoV-1 infection and the corresponding epitopes in SARS-CoV-2 spike (26). Several of these predicted B cell epitopes overlapped with the sequences that we identified in our GFPDL analysis that were conserved between SARS-CoV-1 and SARS-CoV-2 (table S4). Majority (18 of 25) of antigenic sites uniquely identified by the COVID-19 serum antibodies in the current study were missed by the prediction algorithms (26, 27), underscoring the limitations of the predictive algorithms in the in silico approach. One of the possible limitations of GFPDL-based assessments is that while the phage display is likely to detect both conformational and linear epitopes on SARS-CoV-2 spike, they are unlikely to detect paratopic interactions that require posttranslational modifications and rare quaternary epitopes that are cross-protomers. However, in the current study (fig. S2) and prior studies with SARS-CoV-2 vaccination (21) and Ebola virus/ZIKV/Respiratory syncytial virus/influenza virus postvaccination or postinfection serum polyclonal antibodies, 86 to 91% of antivirus antibodies were removed by adsorption with the GFPDL, supporting the use of the GFPDL for analyses of postinfection human samples (1315, 1719). As previously reported (28, 29), the severe patients had sustained high levels of proinflammatory cytokines compared with the patients with mild disease who reached statistical significance for IL-6, IL-8, and MCP-1.

The patients with severe COVID-19 demonstrated moderate to high SARS-CoV-2 neutralizing antibody response for 5 to 9 weeks since symptom onset compared with the minimal neutralization titers observed for mild patients and only after resolution in disease symptoms before discharge. Increased titers in severe patients likely reflect the duration of immune response and antigen exposure. Five patients with severe COVID-19 (receiving ECMO treatment) and four fatal cases generated a high spike-binding antibody titer with moderate to high PRNT50 titers, similar to previous observations in hospitalized patients with COVID-19 (8, 11). However, we observed none or minimal anti-spike antibody affinity maturation over the 5 to 10 weeks’ time period in these patients with severe COVID-19. This suggests that other immune parameters including antibody isotype, specificity, and affinity in addition to neutralizing antibodies are contributing to protection from COVID-19 disease and therefore should be carefully evaluated in vaccine/therapeutic studies against COVID-19.

The higher levels of IgG3 isotype in prefusion spike–binding antibodies in the deceased patients with COVID-19 (S-01, S-04, and S-C02) during the later stage but not in the other patients may be related to antigen drive but equally likely to be driven by the severe cytokine storm during acute SARS-CoV-2 infection, as postulated before (30). In the patients with severe COVID-19, a significantly higher predominance of IgA antibodies were observed especially against the prefusion spike and RBD compared with patients with mild disease. The IgA repertoire emerged very early and continued to increase as disease progressed in the patients with severe COVID-19 who received ECMO, and four of them died of COVID-19. In a separate study of SARS-CoV-2–infected individuals with subclinical COVID-19 disease (who did not require hospitalization), majority of spike-binding antibody were IgM (>65%), while 20 to 30% were IgG and minimal IgA isotypes (clinical study in progress). Other studies observed both rapid waning of IgA (31, 32) and sustained high IgA (33) in patients with COVID-19. While we did observe declining IgA in mild patients, however, sustained high levels of IgA was a hallmark of all patients with severe COVID-19. These apparent contradictions between studies may be due to distinct patient populations as well as dependent on various factors in different type of assays and antigens used to quantify IgA in these studies and therefore difficult to draw meaningful inferences. Our assays used spike proteins that are captured using His tag in a fixed orientation on the chip surface to provide native spatial conformation for antibody binding in contrast to directly antigen-coated plate ELISA used in earlier studies that may result in denaturation of some (if not all) of the protein molecules bound to the polystyrene surface and present non-native epitope for binding antibodies resulting in loss of discrimination between antibodies specific to native versus non-native conformations. The association of IgA titers with disease severity is unclear but may be important at mucosal sites, because increased IgA results from the increased exposure of mucosal sites (lungs) to viral antigens. However, it is possible that dimeric IgA detected in serum may not correlate with secretory IgA (sIgA) on mucosal surfaces and suggests that this observation requires further investigation. This differential antibody kinetics suggests disparate expression and/or antigen exposure/recognition by the human immune system following SARS-CoV-2 infection in these patients with COVID-19 and therefore should be investigated further in follow-up studies. Our findings of an IgA antibody signature in patients with severe COVID-19 and not in the mild patients raise potential for prefusion spike–specific IgA as an immune marker of disease severity that should be tested in patients with COVID-19.

The antibody repertoire induced by SARS-CoV-2 infection in the patients with mild COVID-19 showed greater diversity, antibody class switching, and affinity maturation compared with severe COVID-19–induced antibody responses. In the severe patients who expired, there was a “hole” in the epitopes spanning the RBM and S2 fusion peptide. The rapid resolution of oxygen support, fever, and C-reactive protein (CRP) levels, causing an overall reduction in the clinical symptom score, was immediately preceded by a surge in high-titer, high-affinity prefusion spike antibodies with immunodominance of epitopes at the large RBD sequences encompassing RBM, and the fusion peptide in the early recovery patients with mild COVID-19. Antibodies generated against the moderately conserved site in the fusion peptide (>50% protein sequence across SARS-CoV-1, Middle East respiratory syndrome (MERS), and bat CoVs but not human CoVs) suggests a potential for fusion peptide as a component for design of cross-protective vaccine/therapeutic across multiple pathogenic CoV strains (34).

An inflection point with rapid rise in antibody affinity against prefusion spike protein was observed that was followed by decrease in clinical symptoms, associated with a rapid decline in clinical symptoms score thereafter in all the five mild patients and the non-ECMO patient with less severe COVID-19 (S-03) who recovered from the COVID-19 disease. The anti-prefusion spike antibody inflection point (but not S1 or S2 domain) was observed for both titers, and affinity preceding rapid decline of clinical symptoms onward suggests the presence of some relevant protective epitopes on the prefusion native form of spike protein that are not present on individual S1 and S2 spike domains (7, 22). In earlier studies, with vaccines against H7N9 avian influenza virus, we found important correlation between antibody affinity against the hemagglutinin HA1 globular domain and control of virus loads after challenge of ferrets with H7N9 (35). In the study of patients recovering from ZIKV infections, their antibody affinity against ZIKV E-DIII correlated with lower clinical scores (17). In a large randomized clinical trial of human intravenous immunoglobulin (hIVIG) hyper-enriched for influenza virus antibodies, in adults hospitalized with confirmed influenza A or B virus infections, a statistically significant virological benefit and clinical benefit for patients infected with B strains, directly correlated with stronger antibody affinities of the hIVIG for circulating B strains (36). In a recent longitudinal study of Ebola virus disease survivor, affinity maturation to Ebola virus glycoprotein (GP) was associated with a rapid decline in viral replication and illness severity in this patient (18). On the other hand, complex immune dysregulation in patients with severe COVID-19 includes CD4 cytopenia and macrophage activation syndrome (37, 38). Recently, it was shown that in patients with fatal COVID-19, there is notable loss of germinal centers (GC) in lymph nodes and spleens as well as depletion of Bcl-6 + T follicular helper cell differentiation, suggesting an underlying basis for the lower quality of humoral immune responses (39). Therefore, it seems that in patients with more severe COVID-19, although they can generate high binding and neutralizing antibody titers, there is a block to antibody affinity maturation that may be linked to deficiency in CD4 cells, and especially T follicular helper cells subsets, which are required for entry into GC. Thus, the ability of virus-specific B cells to enter GC in lymph nodes and undergo affinity maturation may play an important determinant in the ultimate effectiveness of viral control and disease resolution. Thus, vaccines that can elicit high-affinity antibodies may have a substantial advantage for in vivo clinical outcome of SARS-CoV-2 infection and contribute to amelioration of disease in infected individuals. The potential role of prefusion form of spike protein antibodies in protection from disease provides support for the role of the prefusion spike protein as a potential vaccine or therapeutic target (22, 40, 41).

We began the exploratory study very early during the COVID-19 pandemic and focused on sample collected frequently without any experimental antibody or immunomodulatory treatment. Although our study sample size is small, the comprehensive immune profiling reveals differences between the mild versus severe groups that these differences reach statistical significance for cytokines, antibody isotype, and antibody affinity for prefusion spike. A larger number of longitudinal samples from patients with COVID-19 is desirable, but high-quality well-defined carefully curated longitudinal samples with frequent collection during acute phase of SARS-CoV-2 infection before disease resolution or death, which did not receive any experimental treatment, are very limited. However, these findings need to be confirmed by expansion to larger cohort of acutely infected longitudinal samples throughout the hospitalization in patients with COVID-19. Together, in our study, sustained high levels of proinflammatory cytokines/chemokines (IL-6, IL-8, and MCP-1), high serum IgA, and blunted affinity maturation against the prefusion spike protein were predictive of a worse outcome for the hospitalized patients, while disease resolution in survivors was associated with antibody affinity maturation to native prefusion spike. These findings demonstrate the importance of prefusion-specific antibody as a correlate with COVID-19 disease outcome.

In summary, this longitudinal analysis study demonstrated a differential evolving antigenic fingerprint following SARS-CoV-2 infection in terms of antibody epitope repertoire diversity, antibody isotype class switching, and antibody affinity maturation in mild versus severe COVID-19 patients. Our study indicates that binding and neutralizing antibody titers do not correlate with disease outcome, while antibodies to the fusion peptide, and most importantly, higher antibody affinity specifically to prefusion spike protein, were associated with disease resolution and clinical outcome in the patients with COVID-19. Our findings indicate that close immunological characterization of the host-viral interaction can identify immune markers of disease severity and resolution of clinical disease that may contribute to the disease outcome in patients with COVID-19. The evaluation of antibody quality, specificity, and affinity maturation following vaccination/infection may facilitate development and evaluation of more targeted immune-based countermeasures against COVID-19.

MATERIALS AND METHODS

Proteins and serum samples

Prefusion spike ectodomain (amino acids 1 to 1208) lacks the cytoplasmic and transmembrane domains (CT-TM), S1 domain (amino acids 16 to 685), RBD (amino acids 319 to 541), and S2 domain (amino acids 686 to 1213), all containing His tag at C terminus, were produced in either human embryonic kidney–293 mammalian cells (prefusion spike, S1, and RBD) or insect cells (S2 domain). The SARS-CoV-2 spike plasmid expressing genetically stabilized prefusion 2019-nCoV_S-2P spike ectodomain, a gene encoding residues 1 to 1208 of 2019-nCoV S fused to HisTag was a gift from B. Graham (Vaccine Research Center, National Institutes of Health). This expression vector was used to transiently transfect FreeStyle293-F cells (Thermo Fisher Scientific) using polyethylenimine. Protein was purified from filtered cell supernatants using Strep-Tactin resin and subjected to additional purification by size exclusion chromatography in phosphate-buffered saline (PBS). Recombinant SARS-CoV-2 protein domains were purchased from Sino Biologicals (S1, 40591-V08H; RBD, 40592-V08H; or S2, 40590-V08B). The native receptor-binding activity of the spike proteins was determined by binding to the hACE2 protein (5 μg/ml) (fig. S6).

The heat-inactivated clinical serum/plasma samples were obtained from University of Tokyo and Fujisawa City Hospital (table S1). We collected serum/plasma samples from admission until discharge (one severe patient, S-02, still in the hospital at 48 days after onset of symptoms) or death of the patient (all days are after symptom onset) under a research protocol approved by the Research Ethics Review Committees of the Institute of Medical Science, University of Tokyo (approval number 2019-71-0201) and Fujisawa City Hospital (approval number F2019049). We began the exploratory study very early during the COVID-19 pandemic and focused on samples collected frequently without any experimental antibody or immunomodulatory treatment. Samples were tested in different antibody assays with approval from the U.S. Food and Drug Administration’s Research Involving Human Subjects Committee (FDA-RIHSC). This study was fully approved by University of Tokyo Institutional Review Board (IRB) under protocol number 2019-71-0201. The FDA IRB reviewed the research approved by an independent IRB from the ethics committee of the Institute of Medical Science, University of Tokyo. The study at the Center for Biologics Evaluation and Research, FDA was conducted with deidentified samples; all assays performed fell within the permissible usages in the original consent. This study complied with all relevant ethical regulations for work with human participants, and informed consent was obtained. All adults hospitalized with COVID-19 disease were eligible without any specific selection criteria. Samples were collected from patients who provided informed consent to participate in the study. All assays performed fell within the permissible usages in the original informed consent.

Samples and assays were run in duplicate or triplicate. Research fellows running the antibody assays were blinded to the identity of the groups for assessments of outcomes.

Viral load determination

Viral load in the respiratory samples was determined by one-step RT-qPCR using LightCycler 96 System (Roche) according to the protocol of National Institute of Infectious Disease, Japan (42).

SARS-CoV-2 PRNT

After initial 10-fold dilution of serum or plasma samples in PBS, they were serially diluted 3-fold in duplicate. The diluted serum or plasma samples were incubated with 100 plaque-forming units of SARS-CoV-2 at room temperature for 1 hour. The virus-sample mixtures were inoculated into Vero-TMPRSS2 cells and incubated for 1 hour at 37°C. After the mixture was removed, the cells were incubated with Dulbecco’s modified Eagle’s medium containing 5% fetal calf serum, 10 mM Hepes, gentamicin sulfate (100 μg/ml), amphotericin B (2.5 μg/ml), and 1% agar for 2 days at 37°C. The maximum dilution to reduce the plaque number by more than 50% compared to that without serum and plasma samples was used as the PRNT50 value.

Clinical illness severity scoring

The oxygen-support status score consists of the following categories: 1, not hospitalized; 2, hospitalized, not requiring supplemental oxygen; 3, hospitalized, requiring supplemental oxygen; 4, hospitalized, requiring high-flow oxygen therapy, noninvasive mechanical ventilation, invasive mechanical ventilation, or both; and 5, hospitalized, requiring invasive mechanical ventilation, ECMO, or both (20). Combined clinical symptom score quantified with following status: body temperature: 0, under 37.5°C; 1, 37.5°C or more; and CRP used as a maker of inflammation: 0, under 1; 1, 1 or more and less than 5; 2, 5 or more and less than 10; and 3, 10 or more.

Measurement of cytokine levels in sera

All plasma/serum samples were diluted fourfold in Bio-Plex Sample Diluent HB buffer. The plasma/serum samples were analyzed via a Bio-Plex Pro Human Cytokine Panel 17-Plex assay per the manufacturer’s instructions. Plates were read using the Bio-Plex 200 system (Bio-Rad, Hercules, CA).

Antibody binding kinetics of post–SARS-CoV-2 infection human plasma/serum to recombinant SARS-CoV-2 proteins by SPR

Steady-state equilibrium binding of post–SARS-CoV-2–infected human polyclonal plasma/serum was monitored at 25°C using a ProteOn SPR (Bio-Rad). The purified recombinant SARS-CoV-2 proteins were captured to a Ni–nitrilotriacetic acid sensor chip with 200 RU in the test flow channels. The protein density on the chip was optimized such as to measure monovalent interactions independent of the antibody isotype (18). Serial dilutions (10-, 50-, and 250-fold) of freshly prepared plasma/serum in bovine serum albumin (BSA)–PBST buffer [PBS (pH 7.4) buffer with Tween 20 and BSA] were injected at a flow rate of 50 μl/min (120-s contact duration) for association, and disassociation was performed over a 600-s interval. Responses from the protein surface were corrected for the response from a mock surface and for responses from a buffer-only injection. SPR was performed with serially diluted plasma/serum of each time point in this study. Antibody isotype analysis for the SARS-CoV-2 spike protein bound antibodies in the polyclonal plasma/serum was performed using SPR. Total antibody binding and antibody isotype analysis were calculated with Bio-Rad ProteOn manager software (version 3.1). All SPR experiments were performed twice, and the researchers performing the assay were blinded to sample identity. In these optimized SPR conditions, the variation for each sample in duplicate SPR runs was <5%. The maximum RU (Max RU) data shown in the figures were the calculated RU signal for the 10-fold diluted serum sample.

Antibody off-rate constants, which describe the stability of the antigen-antibody complex, i.e., the fraction of complexes that decays per second in the dissociation phase, were determined directly from the human polyclonal plasma/serum sample interaction with recombinant purified SARS CoV-2 prefusion spike ectodomain, S1, S2, and RBD using SPR in the dissociation phase only for the sensorgrams with Max RU in the range of 10 to 100 RU (fig. S8) and calculated using the Bio-Rad ProteOn manager software for the heterogeneous sample model as described before (14, 18, 21). Off-rate constants were determined from two independent SPR runs.

SARS-CoV-2 GFPDL construction

DNA encoding the spike gene of SARS-CoV-2 isolate Wuhan-Hu-1 strain (GenBank: MN908947.3) was chemically synthesized and used for cloning. A gIII display-based phage vector, fSK-9-3, was used where the desired polypeptide can be displayed on the surface of the phage as a gIII-fusion protein. Purified DNA containing spike gene was digested with deoxyribonuclease I to obtain gene fragments of 50- to 1500-bp size range and used for GFPDL construction as described previously (19, 21). The phage libraries were constructed from the SARS-CoV-2 spike gene potentially display viral protein segments ranging in size from 18 to 500 amino acids, as fusion protein on the surface of bacteriophage (fig. S2).

Affinity selection of SARS-CoV-2 GFPDL phages with polyclonal human plasma/serum

Before panning of GFPDL with polyclonal plasma/serum antibodies, plasma/serum components that could nonspecifically interact with phage proteins were removed by incubation with ultraviolet-killed M13K07 phage-coated petri dishes (19, 21). Equal volumes of each human plasma/serum were used for GFPDL panning. GFPDL affinity selection was carried out in-solution with anti-IgM, or protein A/G (IgG), or anti-IgA–specific affinity resin as previously described (13, 14, 18, 19, 21). Briefly, the individual serum or the pooled serum/plasma was incubated with the GFPDL and the specific resin, and the unbound phages were removed by PBST (PBS containing 0.1% Tween 20) wash followed by washes with PBS. Bound phages were eluted by addition of 0.1 N Gly-HCl (pH 2.2) and neutralized by adding 8 μl of 2 M tris solution per 100-μl eluate. After panning, antibody-bound phage clones were amplified, the inserts were sequenced, and the sequences were aligned to the SARS-CoV-2 spike gene, to define the fine epitope specificity in these patients with COVID-19.

The GFPDL affinity selection data were performed in duplicate (two independent experiments by research fellow in the laboratory, who was blinded to sample identity). Similar numbers of bound phage clones and epitope repertoire were observed in the two GFPDL panning.

Adsorption of polyclonal human postinfection plasma on SARS-CoV-2 GFPDL phages and residual reactivity to SARS-CoV-2 prefusion spike

Before panning of GFPDL, 500 μl of 10-fold diluted plasma antibodies from postinfection sample was adsorbed by incubation with SARS-CoV-2 GFPDL phage-coated petri dishes. To ascertain the residual antibodies specificity, an ELISA was performed with wells coated with 100 ng/100 μl of purified recombinant SARS-CoV-2 prefusion spike. After blocking with PBST containing 2% BSA, serial dilutions of human plasma (with or without adsorption) in blocking solution were added to each well, incubated for 1 hour at room temperature, followed by addition of 5000-fold diluted horseradish peroxidase–conjugated goat anti-human IgA + IgG + IgM–specific antibody, and developed by 100 μl of O-phenylenediamine dihydrochloride (OPD) substrate solution. Absorbance was measured at 490 nm.

Statistical analysis

All experimental data were analyzed in GraphPad Prism version 8 (GraphPad Software Inc., San Diego, CA). Kruskal-Wallis nonparametric test was used to compare the differences among multiple data group. The differences were considered statistically significant with a 95% confidence interval when the two-tailed P value was less than 0.05 (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001).

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/7/10/eabf2467/DC1

https://creativecommons.org/licenses/by-nc/4.0/

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: We thank H. Golding and K. Peden for insightful review of the manuscript. We thank M. Nozaki of Fujisawa City Hospital and S. Hagiwara of Saiseikai Utsunomiya Hospital for support of the clinical study. Funding: The antibody characterization work described in this manuscript was supported by FDA intramural funds. Clinical study was supported by a Research Program on Emerging and Re-emerging Infectious Diseases (JP19fk0108113 and JP19fk0108166), a Project Promoting Support for Drug Discovery (JP20nk0101612, JP20nk0101614, and JP20nk0101603), the Japan Initiative for Global Research Network on Infectious Diseases (J-GRID) (JP19fm0108006), Japan Program for Infectious Diseases Research and Infrastructure (JP20wm0125002) from the Japan Agency for Medical Research and Development (AMED), and the National Institutes of Allergy and Infectious Diseases funded Center for Research on Influenza Pathogenesis (CRIP; HHSN272201400008C). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. Author contributions: Designed research: S.K. and Y.K. Performed research: S.R., Y.L., E.M.C., L.K., G.G., S.Y., M.I., M.K., A.Y., S.Yamay., Y.S.-T., M.Im., and S.K. Collected clinical samples and provided clinical data: O.A., M.Ko., E.A., M.S., H.Y., I.N., T.O., and R.B. Arranged clinical sample collection: K.I.-H. and Y.K. Contributed to writing: S.K. and Y.K. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. The proteins and patient serum samples can be provided by S.K. and Y.K. pending scientific review and a completed material transfer agreement. Requests for the clinical samples should be submitted to Y.K.

Stay Connected to Science Advances

Navigate This Article