Main

SARS-CoV-2 has resulted in the global pandemic of COVID-19 that causes severe disease most often in adults and elderly individuals with comorbidities. In children, SARS-CoV-2 infection is often asymptomatic or causes mild disease; however, children are susceptible to develop severe disease manifestations of COVID-19 and the associated postinfectious multisystem inflammatory syndrome in children (MIS-C). Although the adult COVID-19 immune response has been studied1,2,3,4, limited knowledge exists regarding the SARS-CoV-2 immune response in children with COVID-19 or MIS-C, especially against NSPs and accessory proteins of SARS-CoV-2 and their association with disease severity. Previous studies in children evaluated immune responses primarily focused on spike or nucleocapsid (N) antibodies but have provided limited insight into the diversity and quality of polyclonal antibody responses across the complete SARS-CoV-2 proteome3,5,6,7,8. These studies did not identify epitope specificity against the complete viral proteome, nor did they identify antibody kinetic parameters that correlate with disease severity or potential serodiagnostic candidates in these children. Moreover, the contribution of different antigenic sites in SARS-CoV-2 spike protein as an immunogen or vaccine in neutralization against SARS-CoV-2 is unknown.

Therefore, an in-depth understanding of the nature of the immune response to SARS-CoV-2 in hospitalized children is critical to deciphering host–pathogen interaction as well as identification and development of serologic and effective therapeutic/neutralizing targets for reducing SARS-CoV-2-induced disease in this understudied population.

Results

Immunological signatures of children with COVID-19 or MIS-C

A total of 74 hospitalized children (mean age, 9.4 + 6.1 years) were enrolled (Fig. 1a and Supplementary Table 1) at Children’s Healthcare of Atlanta (CHOA), which included 25 children meeting criteria for MIS-C, 35 with acute COVID-19, according to the Centers for Disease Control and Prevention (CDC) case definition, and 14 hospitalized controls (Fig. 1b and Supplementary Table 1). Viral RNA levels were determined by quantitative PCR with reverse transcription (RT–qPCR) using nasopharyngeal swabs (Supplementary Table 2). Demographic, clinical, laboratory, imaging, treatment and outcome data are summarized in Supplementary Table 1 and detailed clinical data for each individual is provided in the Supplementary Table 3. Patients were stratified into severe (S) or mild/moderate (M) disease groups based on admission to an intensive care unit (ICU) or not, respectively. The disease-severity classification of the patients correlated with the maximum pediatric sequential organ failure assessment (pSOFA) scores (Fig. 1a and Supplementary Table 3).

Fig. 1: Demographics and neutralization titers of children with COVID-19 or MIS-C.
figure 1

a, Overview of pediatric cohort, including hospitalized mild/moderate or severe control (SARS-CoV-2 negative) and children with mild/moderate (M) or severe (S) COVID-19 or MIS-C. b, Spearman correlations of clinical parameters among the groups assigned according to clinical disease severity was performed using R on log-transformed data. Color intensity and size of the circles are proportional to the correlation coefficients. Heat scale in the correlogram ranges from −0.7 (blue; negative correlation) to +0.7 (red; positive correlation). Statistical analyses among groups were performed using R and differences were considered statistically significant with a 95% confidence interval when the P value was less than 0.05 (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001). Original P values that show significance are marked on the circles. A black box around the circle indicates the adjusted P value of false discovery rate (FDR) < 0.05. c, Violin plots of SARS-CoV-2 serum PsVNA50 across clinical groups, as determined by spike PsVNA. Each dot represents a distinct patient color coded by groups: gray, mild/moderate control (n = 9); black, severe control (n = 5); orange, mild/moderate COVID-19 (n = 20); red, severe COVID-19 (n = 15); cyan, mild/moderate MIS-C (n = 8); blue, severe MIS-C (n = 17). Statistical analyses among groups were performed using R. The tests were two-sided using Tukey multiple-group comparisons to reduce type 1 error. The differences were considered statistically significant with a 95% confidence interval when the P value was less than 0.05 (*P ≤ 0.05, ** P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001). Detailed statistical data are provided in Supplementary Table 9. d, Correlation and linear regression of serum neutralization titers with DPoS in children with MIS-C and COVID-19. The associated Pearson’s correlation coefficients (r) and linear regression significance (P) are colored accordingly. The trend line fits are shown as center solid lines with the error bands representing the 95% confidence intervals shown as shaded colored areas. ALC, absolute lymphocyte count; ALT, alanine aminotransferase test; ANC, absolute neutrophil count; BNP, brain natriuretic peptide; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; WBC, white blood cell.

Source data

Full size image

Sera were collected upon study enrollment, with all days being relative to days post onset of symptoms (DPoS) that prompted hospitalization. Cytokine and chemokine analyses as well as in-depth antibody profiling were performed on the sera collected at study entry from these 60 children with COVID-19 or MIS-C. The controls were 14 hospitalized children (5 patients with severe illness in ICU and 9 non-ICU patients with mild illness) who were SARS-CoV-2 negative by RT–qPCR and by SARS-CoV-2 nucleocapsid antibody test and were found to have alternative diagnoses. Thus, these children with phenotypically similar disease states served as disease-matched controls to identify differences in immune responses unique to SARS-CoV-2 infection. This is a more relevant clinical comparator group than healthy children, as diagnostic and immunophenotyping analyses are needed to differentiate children with COVID-19 and MIS-C from hospitalized children with similar clinical phenotypes, not from healthy children in the community. None of the patients with MIS-C reported having a preceding COVID-19-like illness before hospitalization; thus, DPoS refers to the days after onset of the symptoms that prompted hospitalization. Differential immune profiles at the time of hospitalization could aid in establishing diagnosis and understanding immunopathogenesis of these distinct clinical phenotypes. Six of the 14 control children were suffering from Kawasaki disease, which serve as comparator control for children with MIS-C, as described in the study by Consiglio et al.7.

COVID-19 symptoms typically occur on average 1 week (range 2–12 d) after SARS-CoV-2 infection, whereas MIS-C is a hyperinflammatory syndrome that occurs approximately 4 weeks after infection. Because it is difficult to accurately determine the day of SARS-CoV-2 infection in these children, we defined these patients from the DPoS. Therefore, to control for differences in age, gender and DPoS among patients with COVID-19 and MIS-C, all statistical calculations across disease groups, were performed on log2-transformed data in R package using age, gender and DPoS as covariates.

Pseudovirus neutralization assay (PsVNA) showed higher 50% pseudovirion neutralizing antibody (PsVNA50) titers in the severe COVID-19 or MIS-C groups compared with the mild disease group, but this difference did not reach statistical significance in comparison with mild/moderate disease (Fig. 1c) from these patients. Hospitalized controls did not show measurable PsVNA titers (PsVNA50 < 60). No gender differences were observed for neutralization titers (Extended Data Fig. 1). Over time, a trend in decline of neutralization titers was observed in the severe COVID-19 group (Fig. 1d).

Cytokine and chemokine profiling identified hyperinflammatory response and revealed elevation of multiple cytokines (interleukin (IL)-2, IL-6, IL-10, IL-13 and IL-17) in children with severe MIS-C or COVID-19, in comparison to children with mild/moderate disease (Fig. 2, Extended Data Fig. 2 and Supplementary Table 4). Tumor necrosis factor (TNF) was significantly higher in children with mild MIS-C, in comparison to children with mild/moderate or severe COVID-19 (Fig. 2k). Temporal analyses of cytokines/chemokines in serum, either by linear regression or grouped intervals, support a distinct disease course in SARS-CoV-2-infected children. Although IL-10 levels were sustained in both patients with severe MIS-C and those with COVID-19; IL-6, IL-8 and macrophage inflammatory protein (MIP)-1β levels declined in severe COVID-19/MIS-C over time (Fig. 2 and Extended Data Fig. 2). Because all eight children with mild MIS-C had samples collected within 9 d of symptom onset, the temporal analysis was performed to compare severe COVID-19 versus mild COVID-19 versus severe MIS-C only. Statistical analysis for the regression lines to test ‘elevation’ or the differences in the intercepts between the trend lines revealed statistically significant differences for IL-2, IL-6, IL-10, IL-17 and monocyte chemoattractant protein-1 (MCP-1) in mild COVID-19 versus severe COVID-19, or severe COVID-19 versus severe MIS-C (Supplementary Table 5).

Fig. 2: Cytokine profile of children with COVID-19 or MIS-C.
figure 2

a, Heat map of 17 cytokines and chemokines in serum samples of children with mild/moderate or severe COVID-19 versus MIS-C and hospitalized SARS-CoV-2-negative controls. The color scale represents the concentration of each protein. b–k, Cytokine/chemokine levels reported over DPoS with linear regression analyses. IL-2, IL-6, IL-7, IL-8, IL-10, IL-13, IL-17, MCP-1, IFN-γ and TNF concentrations in serum are shown in log10 of pg ml−1. Statistical differences among groups were performed using R and considered statistically significant with a 95% confidence interval with a P value less than 0.05 (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001). The detailed statistical data are provided in Supplementary Table 9. Correlation of cytokine and chemokine measurements were reported over DPoS and continuously as linear regressions. The trend line fits were performed for mild/moderate COVID-19 (orange) versus severe COVID-19 (red) versus severe MIS-C (blue), using the linear regression model. Associated Pearson’s correlation coefficients and linear regression significance are colored accordingly. The trend line fits are shown as central, colored solid lines with the error bands representing 95% confidence intervals shown as shaded colored areas. l, Spearman correlations and significance of cytokines/chemokines with disease-severity groups. Multivariate linear regression analysis on log10-transformed data was performed using R. DOH, days of hospitalization; DOI, days of ICU admission. The Spearman’s rank correlation coefficient (ρ) is indicated by circle size and heat scale in the correlogram. Heat scale is indicated as gradient color from ρ = −0.7 (blue; strong negative correlation) to ρ = 0 (white, no correlation) to ρ = 0.7 (red; strong positive correlation). The P values that show significance are marked on the circles; *P < 0.05; **P < 0.01; ***P < 0.001. A black box around the circle indicates the adjusted P value of FDR < 0.05.

Source data

Full size image

To gain insights into key differences between COVID-19 and MIS-C with mild/moderate versus severe disease, we correlated the measurements of these soluble cytokine/chemokines using multivariable analysis (Fig. 2l). Positive associations of the cytokines linked to cytokine-release syndrome, such as IL-6 and IL-10, as well as IL-2, IL-7 and IL-13, were observed in children with severe disease.

SARS-CoV-2 antibody epitope repertoire in children

To elucidate the IgM, IgG and IgA antibody epitope repertoire across the entire SARS-CoV-2 proteome in children, epitope mapping was performed using the SARS-CoV-2 genome-fragment phage display library (GFPDL) containing sequences ranging from 50–1,500-bp long from the complete SARS-CoV-2 genome with >107.9 unique phage clones (Extended Data Fig. 3). Previously, GFPDL analyses of highly pathogenic avian influenza virus, respiratory syncytial virus (RSV), Zika virus and Ebola virus were used to map the antibody repertoires of convalescent sera and in individuals after vaccinations. These studies revealed several diagnostic and protective targets9,10,11,12.

The SARS-CoV-2 GFPDL displayed linear and conformational epitopes with random distribution of size and sequence of inserts that spanned the entire SARS-CoV-2 genome (Extended Data Fig. 3). In preliminary studies to characterize the SARS-CoV-2 spike GFPDL, epitope mapping of monoclonal antibodies targeting the SARS-CoV-2 spike protein or receptor-binding domain (RBD) identified the expected linear or conformation-dependent epitopes recognized by these monoclonal antibodies. Recently, we showed that SARS-CoV-2 spike GFPDL can recognize both linear, conformational and neutralizing epitopes in the post-vaccination sera of rabbits13 and after SARS-CoV-2 infection in adults and the elderly14,15. Moreover, the SARS-CoV-2 full-genome GFPDL adsorbed >92% of SARS-CoV-2 virion-specific serum antibodies in patients with COVID-19 and in those with MIS-C (Extended Data Fig. 3), providing proof of concept for use of the SARS-CoV-2 GFPDL for epitope repertoire analyses of pediatric samples.

First-drawn samples collected from each child at the same time point to match DPoS from intravenous immunoglobulin (IVIG)-untreated, either mild/moderate or severe COVID-19 or MIS-C patients were pooled and evaluated to delineate the epitopes of IgM, IgG and IgA separately across the SARS-CoV-2 proteome compared with hospitalized controls (Fig. 3). Altogether, we analyzed targets of IgM, IgG, and IgA antibodies in five serum pools in duplicate, in total assessing >108 potential antibody epitope specificities. SARS-CoV-2 infection induced IgM/IgG/IgA antibodies with diverse overlapping epitope repertoire distribution, displaying small and large sequences spanning the entire SARS-CoV-2 proteome (Fig. 3 and Extended Data Figs. 3–5). Overall, IgM, IgG and IgA recognized similar protein antigenic sites across the SARS-CoV-2 proteome, suggesting that SARS-CoV-2 infection induced a shared epitope profile in these children, but IgM antibodies from severe COVID-19 and severe MIS-C serum recognized roughly a ten-fold higher number of phage clones compared with IgG or IgA phage titers in the same cohort (Extended Data Fig. 3). However, when fine epitope specificities of SARS-CoV-2-induced IgM, IgG and IgA were compared, we observed considerable divergence among some epitopes that were recognized especially by only IgG or only IgA in these children with COVID-19 and MIS-C (Extended Data Figs. 4 and 5).

Fig. 3: SARS-CoV-2 IgM, IgG and IgA epitope repertoires in children.
figure 3

IgM, IgG and IgA antibody epitope repertoire recognized in the SARS-CoV-2-infected sera from children with mild/moderate (M) or severe (S) COVID-19 or MIS-C across the whole proteome of SARS-CoV-2 encoding four structural proteins (S, E, M and N) as well as 16 NSPs and 9 accessory proteins. The schematic on top shows various proteins in the full proteome of SARS-CoV-2. The ORF1ab polypeptide encoding various NSPs (numbered sequentially based on occurrence) is shown. Other proteins are labeled as spike (S), ORF3a (3a), ORF4 (E), ORF5 (M), ORF7a (7a), ORF8 (8), ORF9a (N), ORF9b (9b), ORF9c (9c) and ORF10 (10). Epitope coverage map shows alignment position (amino acid position) on the x axis and coverage (phage clone frequency) of each position on the y axis defined by epitopes recognized by IgG, IgM and IgA in mild/moderate and severe COVID-19 (orange and red, respectively) and in mild/moderate and severe MIS-C (cyan and blue, respectively). The y axis shows phage clone frequency in the range of 0–200 for IgM, IgG or IgA selections with these samples, 0–1,000 for IgM clones with severe COVID-19 and 0–500 for severe MIS-C. Owing to higher frequency of selected phage clones, the original clone frequencies (y axis) of spike and ORF3a antigenic sites were scaled (a scaling ratio of 5.25 was applied for spike sites 7,738–7,799 and 3.05 for ORF3a sites 8,407–8,438) to aid in data visualization. Samples were fresh and not frozen and thawed. All samples were run in one batch at the same time for each assay. The GFPDL affinity selection was performed in duplicate (two independent experiments were performed by a researcher in the laboratory, who was blinded to sample identity). Similar numbers of bound phage clones and epitope repertoire were observed in the GFPDL panning.

Source data

Full size image

The severe hospitalized control (SARS-CoV-2-negative) serum IgM, IgG or IgA bound ~1,000-fold fewer phages that primarily recognized epitopes in NSP3, NSP12–16, spike and N protein (Extended Data Fig. 3). The cross-reactivity observed in control serum potentially could be due to antibodies generated by endemic human coronaviruses (hCoVs) that cross-react due to conservation in homologous protein sequences (for example, NSP3, NSP12, NSP13, NSP14 and spike; Supplementary Table 6). This may also result in an anamnestic response boosting B cell memory against hCoVs following SARS-CoV-2 infection in children. Within NSPs, the COVID-19 IgM recognized several antigenic sites in NSP3, NSP12, NSP13 and NSP14, followed by few sites within NSP1, NSP2, NSP5, NSP15 and NSP16 (Fig. 3 and Extended Data Fig. 5). In many cases, we observed recognition of overlapping epitope sequences, indicating an epitope in the common region. IgM in children with mild/moderate COVID-19 recognized NSP2 (249–331), few sites in NSP3 (1,329–1,483, 1,678–1,815 and 2,475–2,660) and NSP12 (4,724–4,791) at higher frequency compared with IgM from severe COVID-19 (Extended Data Fig. 5). In ORFs, IgM from children with COVID-19 preferentially recognized N, followed by sites in ORF5 (M), ORF8 and ORF9b (9,786–9,828). In children with severe COVID-19, a strong response was observed to S1 immunodominant antigenic epitopes (7,729–7,761 and 7,738–7,799) at the C terminus of the RBD as well as a sequence containing a fusion peptide (7,972–8,013), compared to that in patients with mild/moderate COVID-19. The IgM repertoire in MIS-C was diverse across the entire SARS-CoV-2 proteome; however, higher antibody binding to epitopes in NSP3 (1,071–1,158 and 2,437–2,526), NSP5/6 and NSP7/8, ORF7a (9,110–9,173) and N (9,586–9,600) was observed compared with that in patients with COVID-19 (Extended Data Figs. 4 and 5). IgM from mild/moderate MIS-C recognized sites in NSP2 (249–331), NSP3 (1,265–1,465), NSP7/8 (3,905–3,994), ORF7a (9,110–9,173), ORF8 (9,301–9,314) and N (9,413–9,485) at a higher frequency than that from severe MIS-C, whereas severe MIS-C IgM showed a higher frequency of epitopes in NSP3 (1,071–1,158 and 2,437–2,526) compared with mild/moderate MIS-C IgM (Extended Data Fig. 5).

The IgG epitope repertoire mimicked the diverse IgM epitope profile, with primary recognition of sites in NSP3, NSP12, NSP14, spike, ORF3a, ORF4, ORF5, ORF8 and N (Fig. 3 and Extended Data Figs. 4 and 5). IgG from children with severe COVID-19 showed a higher frequency of sites in NSP3 (1,265–1,465 and 1,447–1,552), NSP12 (5,164–5,218), NSP14 (5,942–6,040), ORF3a (8,407–8,438), ORF8 (9,301–9,314), and few sites in N (9,655–9,663) and spike (7,323–7,430 and 8,066–8,257) compared with IgG from children with mild/moderate COVID-19 (Extended Data Fig. 5). The IgG repertoire in MIS-C was also diverse, with additional recognition of ORF4 (8,760–8,782) and ORF5 (8,844–8,912) by mild/moderate MIS-C in comparison with severe MIS-C, and stronger IgG binding to NSP3 (1,732–1,775) and ORF3a (8,407–8,438) in severe MIS-C compared with mild MIS-C (Fig. 3 and Extended Data Figs. 4 and 5). Compared with severe MIS-C, IgG from severe COVID-19 selected a significantly higher frequency of clones displaying NSP3 (1,447–1,552 and 1,573–1,658), spike (8,066–8,257) and N (9,647–9,663), whereas severe MIS-C recognized NSP3 (1,732–1,775) and ORF3a (8,407–8,438) significantly higher than severe COVID-19 (Extended Data Fig. 5).

The IgA repertoire was equally diverse, wherein all sera recognized immunodominant epitopes in NSP3 (1,678–1,815), NSP12 (5,164–5,218) and IgM/IgG/IgA epitopes in spike (7,738–7,799) and N (9,655–9,663; Fig. 3 and Extended Data Figs. 4 and 5). Interestingly, the severe MIS-C IgA uniquely recognized several epitopes located in NSP5/6 (3,506–3,598), NSP14 (6,296–6,503) and ORF4 (E; 8,760–8,782) that were absent or minimally observed by IgA/IgM/IgG from patients with mild/moderate or severe COVID-19 or patients with mild/moderate MIS-C (Extended Data Fig. 5).

To test the probability that host autoantibodies may contribute to the identified SARS-CoV-2 antibody epitope repertoire, we performed homology searches for the GFDPL-identified antigenic sites with a human protein database. This analysis identified some sequences in nucleocapsid, ORF9b and ORF10 proteins that demonstrated >70% sequence conservation hits with human proteins (Supplementary Table 7).

Antibody binding and avidity profile in MIS-C versus COVID-19

To further investigate the individual antibody profiles to different proteins encoded by SARS-CoV-2, quantitative and qualitative surface plasmon resonance (SPR) analyses were performed with 24 recombinant SARS-CoV-2 proteins, including 4 structural proteins (spike, S; nucleocapsid, N; envelope, E; and membrane, M), 8 accessory proteins and 13 NSPs using serial dilutions of polyclonal serum. These proteins were produced in mammalian cells, insect cells or bacterial cells, and were either obtained from commercial sources or produced in-house (Supplementary Table 8). Owing to lack of protein availability, NSP4, NSP6, NSP10, NSP15 and ORF6 were not included in this SPR analysis. Moreover, weak or no binding was observed to NSP4, NSP6, NSP10 or ORF6 in GFPDL selection (Extended Data Fig. 5), in agreement with earlier studies in convalescent adults16. In addition, antibody-binding kinetics were analyzed using three spike subdomains (S1, RBD and S2), prefusion spike proteins from six different hCoVs (SARS-1, Middle East respiratory syndrome (MERS), 229E, HKU, OC43 and NL63) as well as influenza H1N1 hemagglutinin (HA) and RSV prefusion F protein, as controls. Most humans are exposed to RSV by 2 years of age. RSV fusion protein (F) is widely used for antibody evaluation following RSV infection and for development of RSV vaccines17.

Control pediatric serum (n = 14) did not display antibody binding against most SARS-CoV-2 proteins. However, weak binding was observed against a few NSPs (NSP3, NSP7/8, NSP12, NSP13 and NSP16) in serum from seven older controls (ages 8–14 years), but not in serum from the seven younger hospitalized controls (ages 0.3–4 years; Fig. 4a and Extended Data Fig. 6). Serum antibody-binding titers to most SARS-CoV-2 proteins were higher in children with severe MIS-C or COVID-19 than in children with mild/moderate disease; however, interestingly, NSP2 and NSP13 showed significantly higher reactivity in children with mild/moderate disease (Fig. 4a). Severe MIS-C serum showed higher antibody reactivity to ORF3a and ORF9c than COVID-19 serum, whereas severe COVID-19 serum showed higher antibody binding to ORF5 (M) than did MIS-C serum (Fig. 4a and Extended Data Fig. 6). Analysis over time revealed that differential antibody levels to ORF3a, ORF3b and ORF9c were sustained in patients with severe MIS-C, whereas children with severe COVID-19 showed sustained titers against ORF5 (M; Fig. 4b). Because all eight children with mild MIS-C had samples collected at early time point of hospitalization (within 9 d of symptom onset), the temporal analysis in Fig. 4b was performed to compare mild COVID-19 versus severe-COVID-19 versus severe MIS-C only.

Fig. 4: Serum antibody kinetics of COVID-19 or MIS-C with SARS-CoV-2 proteins.
figure 4

Serum from children with mild/moderate (M) or severe (S) COVID-19 or MIS-C or hospitalized controls were analyzed for antibody binding to purified proteins by SPR. a,b, Total antibody binding of tenfold-diluted serum is represented in SPR resonance units (RUs; a) and as linear regressions with DPoS (b). Each dot represents a distinct patient and DPoS, color coded by groups: gray, mild/moderate control (n = 9); black, severe control (n = 5); orange, mild/moderate COVID-19 (n = 20); red, severe COVID-19 (n = 15); cyan, mild/moderate MIS-C (n = 8); blue, severe MIS-C (n = 17). a, The mean values for each group are shown. b, Correlations of binding antibodies reported over DPoS and continuously as linear regressions were performed for mild/moderate COVID-19 (orange) versus severe COVID-19 (red) versus severe MIS-C (blue), using linear regression analysis in GraphPad Prism. Associated Pearson’s correlation coefficients and linear regression significance values are colored accordingly. The trend line fits are shown as the center solid lines, with error bands representing 95% confidence intervals (shaded colored areas). c, Antibody–antigen off-rate constants as a surrogate of antibody avidity of post-SARS-CoV-2 infection serum to SARS-CoV-2 proteins were measured using SPR and plotted in the respective color coding for each patient group. Statistical differences among groups were calculated using R. The differences were considered statistically significant with a 95% confidence interval when the P value was less than 0.05 (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001). Antibody binding of all SARS-CoV-2-infected children (COVID-19 or MIS-C) were significantly higher than control samples for different SARS-CoV-2 proteins. Therefore, for clarity of representation, only statistically significant differences between SARS-CoV-2-infected patient groups are shown (COVID-19 versus MIS-C with mild versus severe disease). Detailed statistical data are provided in Supplementary Table 9. 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%.

Source data

Full size image

Serum antibody binding to the SARS-CoV-2 prefusion form of spike and S-RBD was significantly higher in patients with severe disease, compared with that in controls (Fig. 4a). However, binding antibodies against the S2 domain were observed in six of the seven older uninfected controls (ages 8–14 years) but were absent in the seven younger children (ages 0.3–4 years; Fig. 4a). SARS-CoV-2-induced S2-binding antibodies were not significantly different between children with MIS-C and those with COVID-19. Regression analysis did not identify differences among groups for spike-, RBD- or S2-specific antibody responses (Fig. 4b).

To determine the antibody avidity maturation against different SARS-CoV-2 proteins following 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 avidity of polyclonal antibody against SARS-CoV-2 proteins using SPR10,13,18. Previously, we showed that antibody kinetics measured under optimized SPR conditions primarily represent the monovalent interactions between the antibody–antigen complex, as antigen–antibody-binding off-rates of the IgG and Fab interaction with protein antigens were similar12,13,14,15. Because sera from pediatric controls have minimal or no binding to the SARS-CoV-2 proteins, antibody avidity was determined for children with COVID-19 or MIS-C. SARS-CoV-2 infection of children induced weak antibody avidity against most NSPs (apart from NSP2 and NSP12) with off-rates ranging from 0.8 to ~0.1 per second (Extended Data Fig. 7). In addition to antibodies against spike and N proteins, SARS-CoV-2-induced antibodies showed higher avidity (off-rates of ~0.1–0.01 per second) to several accessory proteins including ORF3a, ORF3b, ORF4 (E), ORF5 (M), ORF7a, ORF7b, ORF8 and ORF9b (Fig. 4c and Extended Data Fig. 7). Children with milder disease showed significantly higher antibody avidity (slower off-rates) against NSP2 and NSP13, whereas children with severe disease demonstrated higher antibody avidity to NSP11 and NSP12 (Fig. 4c and Extended Data Fig. 7). Severe MIS-C serum contained higher avidity antibodies against NSP7/8 complex and ORF9c compared with either mild or severe COVID-19 serum (Fig. 4c and Extended Data Fig. 7). Although children with mild MIS-C demonstrated significantly higher antibody avidity against NSP9 and ORF3a compared to patients with either mild/severe COVID-19 or severe MIS-C (Extended Data Fig. 7).

Significantly higher avidity against SARS-CoV-2 prefusion spike protein was observed in children with mild/moderate COVID-19 and MIS-C than in children with severe disease (Fig. 4c). But this difference in antibody avidity did not translate to spike subdomains S1, RBD or S2, suggesting that antibodies recognize epitopes present on the native trimeric conformation of prefusion spike, but not those retained in individual spike subdomains.

SPR analysis with prefusion spike protein from different hCoVs including pathogenic SARS-CoV-1 and MERS-CoV, as well as seasonal hCoVs (229E, HKU1, OC43 and NL63), revealed SARS-CoV-2 infection in children induced antibodies that cross-reacted with the prefusion spike protein of SARS-CoV-1, MERS-CoV and seasonal hCoVs (Fig. 4a and Extended Data Fig. 6). Patients with severe disease demonstrated higher antibody binding to the prefusion spike protein of other hCoVs compared with that of children with mild/moderate disease. High levels of binding antibodies against RSV-F protein were observed in children with severe MIS-C (Extended Data Fig. 6); however, there was no significant correlation between SARS-CoV-2 spike protein-binding antibodies with RSV-F or influenza HA-binding antibodies. RSV-F protein lacks sequence conservation with SARS-CoV-2 spike protein, and most children get infected with RSV by the age of 2 years17. The 14 age-matched hospitalized SARS-CoV-2-negative controls showed no or minimal antibody binding to SARS-CoV-2 proteins in SPR (Fig. 4a) and no SARS-CoV-2 neutralization (Fig. 1a), demonstrating no nonspecific cross-reactivity of anti-RSV antibodies with SARS-CoV-2. This observation was further confirmed by SPR analysis of an additional set of sera from hospitalized seronegative controls and SARS-CoV-2-seropositive COVID-19 and MIS-C patients obtained from later time points from some of the children in our study (Extended Data Fig. 6). Therefore, the increase in anti-RSV-F-binding antibodies following SARS-CoV-2 infection in children with severe COVID-19 and MIS-C seems to be a true biological observation, which agrees with inflamed antibody profiles specific for multiple pathogens, including RSV, generated following SARS-CoV-2 infection in children19.

Immune correlates of disease severity in COVID-19 and MIS-C

To identify the immune markers that associate with clinical disease severity in pediatric MIS-C and COVID-19, linear regression analysis was performed for maximum pSOFA score for each patient and corresponding serum cytokine/chemokine, total antibody-binding titers and antibody avidity against each of the SARS-CoV-2 proteins (Fig. 5 and Extended Data Fig. 8–10). Higher neutralization titers did not associate with the decline of pSOFA scores in these children (Fig. 5a). A significant positive correlation was observed between pSOFA scores and IL-2, IL-7 and IL-10, which were upregulated in most patients with severe disease (Fig. 5b and Extended Data Fig. 8).

Fig. 5: Immune markers of disease severity in children with COVID-19 or MIS-C.
figure 5

a–d, Correlation and linear regression of serum neutralization titers (a), cytokines (b), total antibody binding (c) and antibody avidity (d) for the first-drawn time-point sample for each patient, with corresponding maximum pSOFA clinical disease-severity scores across all pediatric MIS-C and COVID-19 groups, regardless of disease-severity classification across all SARS-CoV-2-infected children. The associated Pearson’s correlation coefficients and linear regression significance values are shown. Each dot represents a unique patient antibody binding and associated pSOFA score: orange, mild/moderate COVID-19 (n = 20); red, severe COVID-19 (n = 15); cyan, mild/moderate MIS-C (n = 8); blue, severe MIS-C (n = 17). The black center lines indicate the trend lines for all patients with the error bands representing 95% confidence intervals, shown as shaded areas. e, PCA for antibody binding and avidity to selected SARS-CoV-2 proteins (prefusion spike, NSP5, ORF3a, ORF5 (M) and ORF9c) of the first-drawn serum samples collected from 74 children. The percentage of variables explained by each principal component is indicated in the parenthesis. Points are colored by sample groups: gray, mild/moderate control; black, severe control; orange, mild/moderate COVID-19; red, severe COVID-19; cyan, mild/moderate MIS-C; blue, severe MIS-C. Ellipses show the 95% confidence interval around the colored centroid for each group. f, The top five antibody variables describing component loadings (PC1 and PC2) of PCA analysis are shown in the loading plots.

Source data

Full size image

Antibody-binding titers (maximum RUs) to ORF3a, ORF5 (M) and ORF9c demonstrated a significant positive correlation with pSOFA scores, whereas antibody binding to NSP2 correlated with decrease in pSOFA scores (Fig. 5c and Extended Data Fig. 9). A significant positive association of antibody avidity against ORF3a, NSP11, NSP12 and NSP16 with disease severity was observed (Fig. 5d and Extended Data Fig. 10). However, the increase in antibody avidity (decreasing off-rates) against NSP2, NSP13 and SARS-CoV-2 prefusion spike protein (but not S1, RBD or S2) significantly correlated with the decrease in clinical pSOFA score in these MIS-C and COVID-19 pediatric patients (Fig. 5d and Extended Data Fig. 10). These data suggest that both the levels of cytokines as well as antibody specificity and avidity may be associated with SARS-CoV-2-mediated disease severity or its resolution in children.

To characterize the SARS-CoV-2 immune profile difference more finely between patients with MIS-C and pediatric patients with COVID-19, we conducted principal-component analysis (PCA) for cytokines/chemokines, antibody titers and avidity, or both antibody parameters and cytokines/chemokines. PCA identified unique antibody parameters that separated hospitalized controls from patients with SARS-CoV-2 and appear to distinguish between children with mild/moderate versus severe COVID-19 or MIS-C (Fig. 5e,f). This antibody signature includes antibody-binding titers and antibody avidity to SARS-CoV-2 prefusion spike protein, ORF9b and ORF3a, which could potentially serve as markers for disease-severity classification of children with SARS-CoV-2 infection (Fig. 5e,f).

Serodiagnosis and neutralization targets of SARS-CoV-2

To evaluate the relevance of immunodominant epitopes identified in SARS-CoV-2-infected children, we sought to determine if synthetic peptides, reflective of these antigenic sites, would be targets for SARS-CoV-2 serodiagnosis or, upon immunization, generate SARS-CoV-2-neutralizing responses. Currently, in addition to spike-based vaccines, several whole-virus-inactivated vaccines (for example, Sinovac and Covaxin) containing spike and nucleocapsid proteins are being used for mass vaccination in several countries including China, India and others20,21,22,23. These inactivated vaccines will induce antibodies that will react in SARS-CoV-2 S-based or N-based serodiagnostic tests leading to the phenomenon of vaccine-induced seropositivity (VISP), which would confound the interpretation of current S-based and N-based SARS-CoV-2 serodiagnostic assays. Therefore, to address the phenomenon of VISP, we focused on differential antigenic sites in non-spike and non-nucleocapsid proteins using SARS-CoV-2 GFPDL analysis to determine potential serodiagnosis candidates that identifies true SARS-CoV-2 infection but does not react with vaccine-induced antibodies (Fig. 3 and Extended Data Fig. 5). Four peptides representing immunodominant antigenic sites from ORF8, ORF3a, NSP3 and NPS12 were chemically synthesized and evaluated by IgG ELISA (Fig. 6). In addition, SARS-CoV-2 RBD and nucleocapsid proteins were used as controls. Although both peptides and proteins showed high specificity, only ORF3a (8,407–8,438) and ORF8 (9,301–9,314) peptides matched >90% sensitivity observed for the RBD and N protein. The absolute total binding antibodies against ORF8 and ORF3a antigenic site peptides varied; patients with severe disease showed higher antibody binding compared to those with mild/moderate disease versus controls. NSP3 (1,447–1,483) and NSP12 (5,164–5,218) showed high specificity with low reactivity. The combination of four peptides demonstrated 95% sensitivity for the SARS-CoV-2 sera (Fig. 6g). The RBD and N proteins demonstrated high reactivity to serum from children with COVID-19/MIS-C representing an aggregate sum of different antibodies recognizing multiple epitopes in these two large proteins compared with fewer overlapping epitopes in the shorter peptide sequence.

Fig. 6: Peptide-based SARS-CoV-2 serodiagnostic ELISA for SARS-CoV-2.
figure 6

IgG detection in post-SARS-CoV-2-infection sera from children (black, control (n = 20; 10 hospitalized controls and 10 additional nonhospitalized healthy pediatric samples (ages 12–35 months, collected in 2009)); orange, mild/moderate COVID-19 (n = 16); red, severe COVID-19 (n = 10); cyan, mild/moderate MIS-C (n = 2); blue, severe MIS-C (n = 10)) in comparison with 72 pre-COVID influenza-infected adult controls (flu; collected in 2010; gray) are shown on the left, against each protein (a and b), or individual peptides (c–f) or a mixture of four peptides (g), by ELISA using HRP-conjugated anti-human IgG-Fc-specific antibody. Longitudinal analysis of IgG responses to the SARS-CoV-2 proteins and peptides in serum (n = 100) collected at different DPoS from 27 COVID-19 adult patients (shades of red) are shown on the right of each protein/peptide. Dots represent individual serum samples collected at the indicated DPoS, and the samples from the same COVID-19 patients are connected by the lines. Absorbance (y axis) is for 100-fold dilution serum. Cutoff values for each antigen in ELISA were determined by area under the receiver operating characteristic (ROC) curve analysis of the negative controls (using ten control hospitalized children serum samples at 100-fold serum dilution) in GraphPad Prism. The trend line fits were performed for longitudinal adult COVID-19 patients using a nonlinear regression model with polynomial distribution through the origin in GraphPad Prism. The trend line is depicted as the central, colored line with the error bands representing 95% confidence intervals (shaded colored area). Area under the ROC curve analysis using GraphPad Prism was performed to calculate sensitivity (%) and specificity (%) values with 95% confidence intervals. Data are the average values of two experimental ELISA runs. The variation for each sample in duplicate runs was <7%.

Source data

Full size image

To substantiate the findings with children’s sera, we tested serum from 72 pre-COVID influenza-infected adult controls (collected in 2010) and 100 sequential serum samples collected from 27 adults with COVID-19 (ref. 14) for longitudinal antibody analysis against RBD or N, or individual peptides, or a mixture of four peptides in the serodiagnostic ELISA (Fig. 6). The antibody binding of post-COVID-19 adult samples to most peptides evolved with similar trends. In addition to RBD and N proteins, all individual peptides and peptide mixtures demonstrated >90% sensitivity and 100% specificity with SARS-CoV-2-infected adults.

Because SARS-CoV-2 spike is the major target for vaccine and therapeutic development, we evaluated the functional importance of GFPDL-identified sites within the spike protein as an immunogen to generate SARS-CoV-2-neutralizing antibodies. Two rabbits each were immunized intramuscularly with either of five individual keyhole limpet hemocyanin (KLH)-conjugated synthetic peptides representing GFPDL-identified immunodominant antigenic sites (Extended Data Fig. 5), a control RSV-p27 peptide, SARS-CoV-2 prefusion spike protein, RBD or heat-killed SARS-CoV-2 particles. Sera from three of the five SARS-CoV-2 peptide immunized rabbits (but not p27 peptide) following the second immunization generated SARS-CoV-2-neutralizing antibodies, representing S7,363–7,430 in the RBD, S7,972–8,013 encompassing the fusion peptide, and S8,213–8,279 containing the heptad repeat 2 (HR2) sequence, in both the 293-ACE2 cell-based and Calu-3-based neutralization assays (Fig. 7). The fusion peptides of S2 (S7,972–8,013) and HR2 domain (S8,213–8,279) generated stronger neutralizing titers in Calu-3 cells, but were lower than those generated by the prefusion spike protein, RBD protein and virus particles (Fig. 7b). These rabbit studies confirmed that GFPDL-identified antigenic spike peptides are immunogenic, and that three of these peptides, including the RBD, conserved antigenic sites in fusion peptides and HR2 in the S2 domain, eliciting SARS-CoV-2-neutralizing antibodies.

Fig. 7: SARS-CoV-2 neutralization following GFPDL peptide immunization.
figure 7

a,b, SARS-CoV-2 neutralization titers of serum from rabbits immunized with GFPDL-identified SARS-CoV-2 spike peptides or control RSV p27 peptide by PsVN assay using either 293T-hACE2 (a) or Calu-3 (b) cells. Sera from rabbits immunized with SARS-CoV-2 prefusion spike protein, RBD protein or heat-killed SARS-CoV-2 virions were used as comparators. The data (50% neutralization titers; PsVNA50) represent the median value of two rabbits immunized with each immunogen from assays run in duplicate. Pre-vaccination rabbit sera did not neutralize SARS-CoV-2 tested in PsVN assay. Samples were fresh and not thawed. All samples were run in one batch at the same time for each assay. Differences among groups were analyzed using multiple-group comparisons by non-parametric (Kruskal–Wallis) statistical testing using Dunn’s post hoc analysis in GraphPad Prism. The differences were considered statistically significant with a 95% confidence interval when the P value was less than 0.05 (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001).

Source data

Full size image

Discussion

Deciphering the humoral response to SARS-CoV-2 infection in children is critical for understanding host–pathogen interaction and for developing diagnostics, therapeutics and vaccines in this important but understudied population. In this study, we provide a comprehensive SARS-CoV-2 proteome-wide in-depth antibody profiling in children with mild COVID-19 versus severe COVID-19 versus mild MIS-C versus severe MIS-C that revealed antibody markers associated with clinical disease severity. We identified serodiagnosis peptides and neutralizing targets that could guide rational design of diagnostics, therapeutics and vaccines in this critically underrepresented and understudied SARS-CoV-2 infection in children.

To summarize our key findings, SARS-CoV-2 proteome-wide antibody profiling revealed that higher antibody avidity to NSP2 and the SARS-CoV-2 prefusion spike protein was associated with decreased clinical disease severity and could potentially be used to develop a prognostic test of clinical disease outcomes in children. Children with mild COVID-19 and mild MIS-C showed higher antibody binding and antibody avidity to NSP2 and NSP13, compared to children with severe COVID-19 or severe MIS-C. Moreover, we identified ORF8 and ORF3a peptides for differential serodiagnosis of SARS-CoV-2 infection, as well as the spike fusion peptide and HR2 in the S2 domain as potential candidates for development of better therapeutics and vaccines against SARS-CoV-2 (Fig. 8).

Fig. 8: Implications of SARS-CoV-2 deep immunoprofiling of COVID-19 and MIS-C.
figure 8

Schematic highlighting the importance of the key messages for SARS-CoV-2-induced human antibody signature against the viral proteome in children with COVID-19 or MIS-C with different levels of disease severity for various diagnostic, prognostic, therapeutics and vaccine countermeasures against SARS-CoV-2.

Full size image

Although MIS-C is temporally associated with SARS-CoV-2 infection in children, its immunopathogenesis is still poorly understood. Consiglio et al. found that children with MIS-C lacked the IL-17-driven cytokine storm characteristic of Kawasaki disease7. In our study, we found that the distinguishing pro-inflammatory cytokines associated with disease severity in children with COVID-19 included IL-2, IL-6, IL-10, IL-13 and IL-17, complementing and extending the findings of Pierce et al.24. Children with MIS-C, however, had higher TNF, which may be indicative of a resolved viral infection or waning antiviral cytokinemia. The non-COVID-19 IVIG used for treatment early during the pandemic was unlikely to have any neutralizing antibodies to SARS-CoV-2 (ref. 25). Sub-analyses of antibody fingerprinting between children treated without or with non-COVID IVIG did not reveal any differences in antibody response. Robust serologic and hyperinflammatory immune responses were observed in children with MIS-C despite administration of any immunomodulatory therapies. In our study, children with MIS-C did not have elevated levels of IL-1β, which brings into question the utility of these targeted biologic treatment modalities.

We observed a differentially evolving diverse antibody epitope repertoire and avidity maturation to various SARS-CoV-2 proteins in children with mild/moderate versus severe COVID-19 or MIS-C. Higher antibody avidity against NSP8 and prefusion spike protein (but not neutralization titers) was associated with lower clinical disease severity in these children. Antibodies against both NSPs and spike protein could help in disease resolution possibly by Fc-mediated functions. Thus, vaccines or antibody-based therapeutics that can elicit high-affinity antibodies may have a substantial advantage for in vivo clinical outcomes in SARS-CoV-2 and contribute to amelioration of disease in infected individuals. Some of the current SARS-CoV-2 vaccines are based on a ‘stabilized’ prefusion spike protein20. The potential role of antibodies against the prefusion spike protein in protection from disease, provides support for the prefusion spike protein as a vaccine or therapeutic target23,26,27.

Current vaccines have been shown to reduce disease severity, but they do not provide sterilizing immunity; therefore, vaccinated individuals get infected with SARS-CoV-2 and remain asymptomatic28,29. These breakthrough infections will be missed or under-reported if they are not tested in the acute phase of infection by PCR, because the SARS-CoV-2 PCR-based tests are typically negative after 2 weeks of SARS-CoV-2 infection. This means that the true number of breakthrough infections will not be captured and may result in transmission to susceptible unvaccinated individuals like children <12 years of age, who are not yet eligible for vaccination. To address VISP, we identified differential immunodominant epitopes covering non-spike and non-N antigenic sites that showed sensitive and specific antibody detection using ORF3a (8,407–8,438), ORF8 (9,301–9,314) and NSP12 peptide or the peptide combination. ORF3a and ORF8 were recently identified as serologic markers for SARS-CoV-2 infection in adults4. These peptide antigens are not part of current advanced vaccine candidates, are not present in virion particles, and are only expressed following SARS-CoV-2 infection. As SARS-CoV-2-specific antibodies are expected to persist for longer compared with SARS-CoV-2 RNA following viral infection, this serodiagnostic test may allow greater sensitivity to detect breakthrough infections long after the acute phase (1–4 weeks) of SARS-CoV-2 infection.

Although GFPDL-based assessments are likely to detect both conformational and linear epitopes on SARS-CoV-2 proteins, they are unlikely to detect paratopic interactions that require posttranslational modifications and rare quaternary epitopes that cross protomers. However, ~90% of anti-SARS-CoV-2 antibodies either from post-vaccination or postinfection plasma were removed by adsorption with the SARS-CoV-2 GFPDL, supporting the use of the SARS-CoV-2 GFPDL for analyses of human sera13,14,15. Another limitation is our sample size for COVID-19 and MIS-C needs to be expanded to a larger cohort. Even with these limitations, we identified unique pediatric serum antibody-binding and avidity signatures against few SARS-CoV-2 proteins (prefusion spike, ORF3a and ORF9b) that could be used for development of diagnostic and/or prognostic tests for clinical outcomes and disease-severity classification of children with SARS-CoV-2 infection. In a recent study of elderly patients with COVID-19 where samples were collected every 1–3 d, an increase in prefusion spike-specific antibody affinity maturation was immediately followed by decrease in symptom severity15. Identification of fusion peptide and HR2 domain neutralization targets that are conserved among various SARS-CoV-2 variants may facilitate design of targeted therapeutics and more effective SARS-CoV-2 vaccines using reverse vaccinology approaches.

In conclusion, our comprehensive immunoprofiling across the SARS-CoV-2 proteome demonstrated differential epitope repertoire diversity, antibody specificity and antibody avidity maturation in children with COVID-19 or MIS-C with different levels of disease severity. Antibody avidity measurement against the prefusion spike protein can be used to develop a prognostic test for clinical disease outcomes in children. We identified differential serodiagnosis peptides and demonstrated an important role for the fusion peptide and HR2 in the S2 domain for the generation of SARS-CoV-2-neutralizing antibodies. These observations provide a more in-depth understanding of quantitative and qualitative aspects of immune responses that could aid in the development and evaluation of targeted, more effective SARS-CoV-2 serodiagnostics, therapeutics and next-generation vaccines for these understudied pediatric populations.

Methods

Patient enrollment

Children and adolescents, from 0 to 21 years of age, hospitalized at CHOA with confirmed or suspected COVID-19 or MIS-C were enrolled into an institutional review board (IRB)-approved specimen collection protocol (Emory University IRB protocol nos. STUDY00022371 and STUDY00000723) following informed consent and assent as appropriate for age6. Children were classified as having MIS-C if they met the CDC case definition (https://emergency.cdc.gov/han/2020/han00432.asp; https://www.cdc.gov/mis-c/hcp/). They were classified as having acute COVID-19, per the CDC case definition (https://www.cdc.gov/coronavirus/2019-ncov/symptoms-testing/symptoms.html), if they were hospitalized with symptomatic disease and had SARS-CoV-2 detected by real-time PCR (Supplementary Table 2). Clinical disease severity was classified as either severe or mild/moderate. Patients with severe disease were those who required ICU admission. Patients with mild/moderate disease consisted of children who were hospitalized with symptomatic COVID-19 or MIS-C, but who were never admitted to ICU (Supplementary Table 4). These criteria for disease classification and severity scales were previously described for SARS-CoV-2-infected children6. All children with COVID-19 or MIS-C were confirmed as having SARS-CoV-2 infection by the US Food and Drug Administration (FDA) Emergency Use Authorization-approved SARS-CoV-2 RT–PCR and/or SARS-CoV-2 antibody ELISA. The molecular tests performed at CHOA for inpatients during this time period included the DiaSorin Molecular Simplexa COVID-19 Direct RT–PCR assay30, the Cepheid GeneXpert Xpress SARS-CoV-2 test31 and the Hologic Panther Fusion SARS-CoV-2 test32. The qualitative SARS-CoV-2 nucleocapsid antibody test (Abbott Laboratories) was used for serologic testing. Prospective blood samples were collected from patients at the time of enrollment, and residual leftover samples from clinical laboratories were also obtained from the clinical laboratory. SARS-CoV-2-seropositive samples were collected 1 month after initial sample collection on a subset of enrolled patients. Specimens from children who were evaluated for COVID-19 or MIS-C, but tested negative, were utilized as hospitalized contemporaneous controls. The controls were alike-disease hospitalized controls (not infected with SARS-CoV-2), which served as true disease-matched controls. Of these 14 control children, 6 had Kawasaki disease, 5 were admitted to ICU and were categorized as ’severe’, and 9 were never admitted to ICU and served as the ‘mild’ disease comparator group. Because all the samples used in the study were collected from children during their stay in hospitalization, we have not considered any samples as convalescent. Convalescence refers to the later stage of an infectious disease or illness when the patient recovers and returns to previous health. Demographic, clinical, laboratory, imaging, treatment and outcome data were abstracted from the medical record and summarized in Supplementary Table 1, and individual data are provided in Supplementary Table 3. As our study samples were collected early during the pandemic, the effects of non-COVID IVIG administration on SARS-CoV-2 serology were likely negligible. The maximum pSOFA score within a 24-h period during hospitalization was calculated for each patient33. Healthy pediatric control samples from 1- to 3-year-old children were collected in 2009 (ref. 34). The longitudinal plasma samples from patients hospitalized with COVID-19 were described previously14.

Samples were tested in different antibody assays with approval from the FDA’s Research Involving Human Subjects Committee (FDA-RIHSC) under exemption protocol ‘252-Determination-CBER-2020-08-19’. Sample were frozen on the day of collection and kept frozen below −70 °C. The samples were thawed only a single time on the day when different assays were run simultaneously. All samples were run in one batch at the same time for each assay. All the data are presented in an unbiased way, and there was no normalization unless explicitly mentioned in the text or figure legends.

Ethics statement for human studies

The study at the Center for Biologics Evaluation and Research (CBER), FDA was conducted with de-identified samples (Emory University IRB protocol nos. STUDY00022371 and STUDY0000723) under RIHSC protocol no. ‘252-Determination-CBER-2020-04-02’. Samples were collected under IRB-approved specimen collection protocol (Emory University IRB protocol nos. STUDY00022371 and STUDY00000723) after informed consent was obtained from at least one parent or legal guardian. All assays performed fell within the permissible usages in the original consent.

Proteins

Purified recombinant SARS-CoV-2 proteins that were produced in mammalian, insect or Escherichia coli cells or chemically synthesized were developed in-house or obtained from commercial sources (Supplementary Table 8).

The plasmids expressing genetically stabilized SARS-CoV-2 prefusion 2019‐nCoV_S‐2P spike ectodomain, and prefusion spike protein of different hCoVs, SARS-CoV-1, MERS, HKU, OC43, NL63 and 229E were a kind gift from B. Graham (VRC, National Institutes of Health (NIH))26. This expression vector was used to transiently transfect FreeStyle293F cells (Thermo Fisher, R79007) and protein was purified from filtered cell supernatants using StrepTactin resin (Cytiva, 29401326) and by size‐exclusion chromatography and tested for binding to human ACE2 protein13.

Measurement of cytokine levels in serum

All samples were diluted fourfold in Bio-Plex Sample Diluent HB buffer and analyzed via a Bio-Plex Pro Human Cytokine Panel 17-Plex assay (Bio-Rad, M5000031YV) per the manufacturer’s instructions.

SARS-CoV-2 pseudovirus production and neutralization assay

Human codon-optimized cDNA encoding SARS-CoV-2 S glycoprotein (NC_045512) was cloned into eukaryotic cell expression vector pcDNA 3.1 between the BamHI and XhoI sites. Pseudovirions were produced by co-transfection of Lenti‐X 293T cells (Takara, 632180) with psPAX2(gag/pol), pTrip-luc lentiviral vector and pcDNA 3.1 SARS-CoV-2-spike-deltaC19, using Lipofectamine 3000 (Thermo Fisher, L3000015). The supernatants were collected at 48 h after transfection and filtered through 0.45-µm membranes and titrated using 293T-ACE2 cells (HEK293T cells expressing ACE2 protein). The following reagent was obtained through BEI Resources, National Institute of Allergy and Infectious Diseases (NIAID), NIH: Human Embryonic Kidney Cells (HEK-293T) Expressing Human Angiotensin-Converting Enzyme 2, HEK-293T-hACE2 cell line, NR-52511.

For the neutralization assay, 50 µl of SARS-CoV-2 S pseudovirions were pre-incubated with an equal volume of varying dilutions of serum at RT for 1 h, then virus–antibody mixtures were added to 293T-ACE2 cells in a 96-well plate. After 3 h of incubation, the inoculum was replaced with fresh medium. After 24 h, cells were lysed and luciferase activity was measured. Controls included cell-only control, virus without any antibody control and positive control sera.

SARS-CoV-2 gene fragment phage display library

DNA plasmid encoding the full-length SARS-CoV-2 isolate Wuhan-Hu-1 strain (GenBank: MN908947.3) was a kind gift from V. Thiel and used for cloning35. A gIII display-based in-house 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 the entire SARS-CoV-2 genome was digested with DNase I (NEB, M0303L) to obtain gene fragments of 50–1,500 bp in size (18 to 500 amino acids) and used for GFPDL construction12,13,18.

Adsorption of SARS-CoV-2 serum on SARS-CoV-2 GFPDL phages

Before panning of GFPDL, 500 μl of tenfold-diluted serum antibodies from SARS-CoV-2-infected samples 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 104 per 100 μl of heat-inactivated SARS-CoV-2 virion particles. After blocking with PBST (PBS containing 0.1 % Tween-20) containing 2% BSA, serial dilutions of human serum (with or without adsorption) in blocking solution were added to each well, incubated for 1 h at room temperature (RT), followed by addition of 5,000-fold-diluted HRP-conjugated goat anti-human IgA + IgG + IgM-specific antibody (Jackson ImmunoResearch, 109-035-064) and developed by 100 μl of OPD substrate (Thermo Fisher, 34006) solution. Absorbance was measured at 490 nm.

Affinity selection of SARS-CoV-2 GFPDL with human serum

Before panning of GFPDL with polyclonal serum antibodies, serum components, which could nonspecifically interact with phage proteins, were removed by incubation with ultraviolet-killed M13K07 (NEB, N0315S) phage-coated Petri dishes13,18.

Equal volumes of first-drawn time-point samples from the same time point (DPoS) to match the days after onset of symptoms from each patient in the corresponding cohort were pooled from the control, COVID-19 or MIS-C samples from mild/moderate (M) or severe (S) disease and were used for GFPDL panning. GFPDL in-solution affinity selection was carried out with anti-IgM, or protein A/G (IgG), or anti-IgA-specific affinity resin12,13,36. Briefly, the pooled serum was incubated with the full-length SARS-CoV-2 GFPDL and the specific resin, and the unbound phages were removed by PBST wash followed by PBS. Bound phages were eluted by addition of 0.1 N Gly-HCl (pH 2.2). After panning, antibody-bound phage clones were amplified, the inserts were sequenced, and the sequences were aligned to the SARS-CoV-2 genome.

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

Surface plasmon resonance-based antibody-binding kinetics of human serum

Steady-state equilibrium binding of pediatric serum was monitored at 25 °C using a ProteOn SPR (Bio-Rad). The purified recombinant SARS-CoV-2 or other proteins were captured to a Ni-NTA sensor chip (Bio-Rad, 176-5031) with 200 RUs in the test flow channels. The protein density on the chip was optimized so as to measure monovalent interactions independent of the antibody isotype12,13,14,15,37. Serial dilutions (10-, 30- and 90-fold) of freshly prepared sample in BSA-PBST buffer (PBS pH 7.4 buffer with Tween-20 and BSA) were injected at a flow rate of 50 µl min−1 (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 serum of each individual time point in this study. Total antibody binding was 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 data shown in the figures are the RU signal for the tenfold-diluted serum sample.

Antibody off-rate constants, which describe the stability of the antigen–antibody complex, that is, the fraction of complexes that decays per second in the dissociation phase, were determined directly from the human polyclonal sample interaction with recombinant purified SARS-CoV-2 proteins using SPR. To that end, serially diluted samples at 10-, 30- and 90-fold dilutions were analyzed to determine antibody off-rate constants in the dissociation phase only for the sensorgrams with maximum values in the range of 5–150 RUs and calculated using the Bio-Rad ProteOn manager software for the heterogeneous sample model as described before10,12,13,14,15,37. Off-rate constants were determined from two independent SPR runs. All SPR experiments were performed twice, and the researchers performing the assay were blinded to sample identity. The variation for each sample in duplicate SPR runs was <5%. The data shown are average values of two experimental runs.

Seroreactivity to SARS-CoV-2 antigenic peptides by ELISA

Ninety-six-well Immulon plates were coated with 100 ng/100 µl of recombinant spike, N protein or synthetic peptides in PBS overnight at 4 °C. Starting at a 1:100 dilution, serum samples were serially diluted at a 1:5 ratio and applied to the coated well for 1 h at ambient temperature. Serum samples were assayed in duplicate. Naïve serum samples were assayed along with the experimental samples. After three washes with PBS/0.05% Tween 20, bound antibodies were detected with a 1:5,000 dilution of horseradish peroxidase (HRP)-conjugated anti-human IgG Fc-specific antibody (Jackson ImmunoResearch, 109-035-098). After 1 h, plates were washed as before and o-phenylenediamine dihydrochloride (oPD) was added for 10 min. Absorbance was measured at 492 nm. The absorbance shown is reported for the 100-fold-diluted serum/plasma sample. Cutoff values for each antigen in ELISA were determined by the area under the ROC curve analysis of the negative controls (using samples from ten hospitalized control children at 100-fold serum dilution) in GraphPad Prism.

Rabbit immunization studies

GFPDL-selected immunodominant SARS-CoV-2 spike peptide conjugation to Maleimide Activated KLH was performed as described in the product manual of Imject Maleimide Activated KLH (Thermo Scientific, 77605). Female New Zealand white rabbits (4–6 weeks old) were immunized twice intra-muscularly at 21-d intervals with 25 μg of KLH-conjugated peptides or 50 μg of purified recombinant proteins or heat-inactivated 105 plaque-forming units of SARS-CoV-2 virion particles mixed with Emulsigen Adjuvant. Two rabbits were used per immunogen. Sera collected before vaccination and 8 d after the second vaccination were analyzed in a pseudovirion neutralization assay.

All animal experiments were approved by the FDA Institutional Animal Care and Use Committee under protocol no. 2008-10. This protocol meets NIH guidelines.

Statistical analysis

All experimental data to compare differences between groups were analyzed using lme4 and emmeans packages in R (RStudio version 1.1.463). Data from SPR (antibody-binding and antibody off-rates), cytokine and neutralization titers (used as independent variables) were analyzed for statistical significance among various disease groups to control for age, gender and DPoS, which served as covariates (predictor variables), using a linear regression model. To ensure robustness of the results, absolute measurements were log2 transformed before performing the analysis. Within each disease group, the respective measurement was regressed on age, gender and DPoS. For comparisons between various disease categories (factor variable; mild and severe controls, mild and severe COVID-19, mild and severe MIS-C), pairwise comparisons were extracted using emmeans, and Tukey-adjusted P values were used for denoting significance to reduce type 1 error due to multiple testing. The tests were two sided. The statistical analysis tested for significant differences in cytokine, antibody binding, antibody off-rates and neutralization titer measurements among various disease categories. For the temporal analysis and correlations, the DPoS values for COVID-19 and MIS-C were controlled statistically by multivariable analysis in R. Detailed statistical data for the figures and extended data figures are provided in Supplementary Table 9.

PCA analysis of 74 antibody-binding and avidity variables and 17 cytokines/chemokines from 74 samples was conducted using R packages factoMineR and factoextra. Antibody-binding variables (maximum RUs), avidity (off rate) and cytokines/chemokines (concentration) data were log10 transformed before analysis. Correlation and regression analyses were performed by computing Pearson’s coefficient in GraphPad Prism. Multivariate linear regression models for cytokine and clinical disease (Fig. 1b,p) were performed in R and used to calculate Spearman’s rank correlation coefficient and significance.

For rabbit data, differences among groups were analyzed using multiple-group comparisons by non-parametric (Kruskal–Wallis) statistical testing using Dunn’s post hoc analysis in GraphPad Prism.

Reporting Summary

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