The study is in accordance with the declaration of Helsinki and according to Dutch regulations. Data and samples were collected only from voluntary, non-remunerated, adult donors who provided written informed consent as part of routine blood collection procedures of the Sanquin Blood Supply Foundation (Blood Bank). The study was approved by the Ethics Advisory Council of Sanquin Blood Supply Foundation. Donors having recovered from mild convalescent COVID-19 disease were identified at Sanquin Blood Bank, Amsterdam, the Netherlands (410 donors). The participants were further selected on availability of more than two frozen PBMC vials at the early time point, two or more serological measurements, and more than 49 days between the first and last time points of serology (132 donors). For these selected donors, all available plasma samples were measured at least twice to determine RBD-IgG titers as described below and donors with signs of reinfection were excluded. Antibody levels show exponential decay over time; therefore, we assumed that the decay in log-transformed antibody levels was linear to estimate the half-life. For each donor, a linear model was fitted on the log-transformed antibody levels, with time expressed as days post symptom onset (PSO) as described before (34). The estimated slope of the model was used to calculate the donor’s antibodies’ half-life in days: t 1 / 2 = log ( 1 2 ) / s l o p e . The top 10% with shortest half-life (DAb, n=13) and top 10% with longest half-life, including increasing titers (SAb, n=14), were selected for this study.

IgM, IgG, and IgA to RBD and nucleocapsid (NC) were measured as described previously (7). RBD and NC proteins were produced as described before (7). Pooled convalescent plasma or serum was included on each plate as a calibrator (set to a value of 100 AU/mL) to quantify the signals. Results were expressed as arbitrary units (AU) per mL (AU/mL) and represent a semiquantitative measure of the titer of IgG, IgA, and IgM antibodies to RBD and NC.

Serum samples from the time point of PBMC sampling (T1) and the last time point of sampling (Tfinal) were tested for the neutralizing capacity of their SARS‐CoV‐2 antibodies using a competitive assay as described previously (7). The IC50 value was calculated by using a non-linear fitting model with a restrained slope of −1 and weighted by 1/Y² on the output of dilution factor versus the percentage of non-inhibition. By multiplying the calculated IC50 value with the RBD titer of the same serum sample, we excluded the influence of possible variations in the RBD titer between individuals.

Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll Paque Plus gradient separation (GE Healthcare, Chicago, IL, USA) and stored in 10% dimethyl sulfoxide in fetal bovine serum (Thermo Fisher Scientific, Waltham, MA, USA) for future use. All procedures were in accordance with guidelines established by the Sanquin Medical Ethical Committee and in line with the Declaration of Helsinki.

SARS-CoV-2 S-2P, RBD, influenza A hemagglutinin (H1N1pdm2009, A/Netherlands/602/2009, GenBank: CY039527.2), RSV prefusion stabilized F (DS-Cav1), and constructs with avi-tag and/or hexahistidine (his)-tag and/or strep-tag were expressed and purified as previously described (2, 28). Avi-tagged proteins were biotinylated with a BirA500 biotin-ligase reaction kit according to the manufacturer’s instruction (Avidity). Tetanus toxoid was purchased from Creative Biolabs (Vcar-Lsx003). Nucleocapsid and tetanus toxoid were aspecifically biotinylated using EZ-Link Sulfo-NHS-LC-Biotinylation Kit (Thermo Fisher) according to the manufacturer’s instructions. Biotinylated protein antigens were labeled individually with fluorescently labeled streptavidin and multimerized (BB515, BD Biosciences; BUV615, BD Biosciences; AF647, BioLegend; BV421, BioLegend), as described previously (28). Biotinylated proteins and fluorescently labeled streptavidin were mixed at a 2:1 protein to fluorochrome molar ratio and incubated at 4°C for 1 h. Unbound streptavidin conjugates were blocked with 10 µM biotin (GeneCopoiea) for at least 10 min. A combinatorial probe staining strategy was used for simultaneous identification of multiple B cell specificities in a single sample. This strategy uses multiple combinations of two fluorophores to increase the number of specificities that can be detected and decrease aspecific binding. In our study, we were able to detect six different antigen-specificities using five distinct fluorophores. Probes were labeled in the following manner: SARS-CoV-2 S (AF647, BV421), H1N1 HA (BUV615, BV421), RSV F (AF647, BUV615), NC (AF647, BB515), tetanus toxoid (BB515, BV421), and RBD (PE-Cy7). Individually labeled proteins were then equimolarly mixed and kept at 4°C before usage. A final concentration of 45.5 nM of each probe is used to label B cells.

PBMCs were thawed in Iscove’s Modified Dulbecco’s Medium (Lonza, Verviers, Belgium) with 10% fetal calf serum (Bodinco, Alkmaar, The Netherlands), 10 million PBMCs were depleted of CD3+ cells using EasySep™ Human CD3 Positive Selection Kit II and EasyPlate™ EasySep™ Magnet (STEMCELL Technologies, Vancouver, Canada), and between 0.8 and 3.5 million B cells were obtained for staining. A 29-color spectral flow cytometry panel was designed to enable comprehensive and simultaneous immunophenotyping of total B cells and B cells with antigen specificity to six proteins ( Supplementary Tables S1 , S2 ). A total of 23 colors were assigned for deep characterization of the B cell compartment, 5 colors were assigned to capture antigen specificity using double discrimination, and 2 colors were assigned to exclude dead cells and for a dump channel to remove cells expressing CD3, CD4, CD16, or CD56. LIVE/DEAD™ Fixable Blue Dead Cell Stain Kit from Invitrogen™ (Thermo Fisher Scientific, Waltham, MA, USA) was used to exclude non-viable cells. BD Horizon™ Brilliant Stain Buffer Plus (BD Biosciences, Franklin Lakes, NJ, USA) was supplemented to the staining mixture according to the manufacturer’s instructions. Cells were stained at 4°C for 30 min with the mix of multimerized proteins and the mix of fluorochrome-conjugated antibodies simultaneously. Following staining, cells were washed twice with a washing buffer containing 1% bovine serum albumin (Sigma-Aldrich, Saint Louis, USA) and 1 mM ethylenediaminetetraacetic acid (EDTA) in phosphate-buffered saline (Fresenius Kabi, ‘s-Hertogenbosch, The Netherlands), and then fixed with 1% cold paraformaldehyde for 10 min at room temperature on a shaker and then washed twice with washing buffer. Data were acquired on Cytek Aurora 5L using SpectroFlo^® software (Cytek Biosciences, Fremont, California, United States).

PBMCs were depleted of CD3+ cells and labeled with a panel of fluorochrome-conjugated antibodies to surface proteins and fluorochrome-labeled probes and acquired using spectral flow cytometry for a deep profiling of antigen-specific B cells in convalescent COVID-19 patients and healthy controls in three batches. All FCS files were gated to remove debris, doubles, and dead cells, and cells positive for CD3-, CD4-, CD16-, or CD56-positive events were excluded in a dump channel. Subsequently CD19 was used to identify B cells (gating strategy is shown in Supplementary Figure S2A ) and a range from 101,469 to 708,000 CD19+ B (median: 239,799) cells were measured per donor. B cells specific for each of the six proteins were gated based on the combination of fluorochrome-conjugated streptavidin. RBD-specific cells were gated from the Spike-specific gate. Gated data were further processed with the R programming language (http://www.r-project.org) and Bioconductor (http://www.bioconductor.org) software. First, antigen specificity was integrated as a logical variable in the data. Anomalies in the data were removed using the flow_auto_qc function from the FlowAI package in R (35). Data were transformed with an inverse hyperbolic sine (arcsinh) transformation. Batch effects were modeled using reference samples stained and acquired with each batch to control for signal fluctuation that might occur over time due to changes in instrument performance. The model was then used to remove batch effects from the data using a normalization algorithm. Modeling of batch effect and data normalization was done using the CytoNorm package in R (36).

For initial clustering, FlowSOM was used, which is based on self‐organizing maps and hierarchical consensus meta-clustering (36). Total B cells and antigen-specific B cells were clustered into 400 nodes, using a 20 × 20 grid with the scaled expression of 23 cell surface proteins used as input. Nodes were then meta-clustered using the ConsensusClusterPlus function with k = 40 for hierarchical consensus clustering as implemented in the ConsensusClusterPlus package in R (37). A random subset of the cells in the data were then aggregated per meta-cluster and median of the unscaled expression of each cell surface protein was calculated to generate colored heatmap for meta-clusters versus cell surface proteins using the make.pheatmap function from Spectre package in R (38). Meta-clusters were then annotated, and some of the meta-clusters were merged based on visual inspection of heatmaps (where the differential expression of cell surface proteins was not biologically meaningful for some neighboring meta-clusters).

Frozen PBMCs with median 18.6 million cells per vial (min: 8.4 million, max: 30.2 million) were thawed. Two million PBMCs were stimulated overnight at 37°C with or without SARS-CoV-2 peptide pools (PepMix™, JPT, 1 μg/ml in DMSO) in the presence of 3 μg/ml brefeldin A (eBioscience™, Thermo Fisher). Peptide pools consisted of 15-mers with an 11-amino acid overlap spanning the entire protein sequence of Spike and nucleocapsid. Spike peptides were divided into two pools (S1 and S2) and separately used for PBMC stimulation. Cells were stained with FSV575V viability dye and the chemokine markers anti-CCR6, anti-CCR4, anti-CXCR5, and anti-CXCR3; the activation markers anti-PD1, anti-HLADR, anti-CD38, and anti-ICOS; and the differentiation markers anti-CD45RA, anti-CD27, and anti-CD14 antibodies for 30 min at room temperature. Cells were subsequently fixed and permeabilized using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience™, Thermo Fisher) and stained for the cytokines IL21, IL17A, IL4, IFNg, and TNF; the lineage markers CD4, CD8, and CD3; and the activation markers CD40L (BioLegend), CD137, and CD69 for 30 min at 4°C and analyzed on a BD Symphony flow cytometer. Data were analyzed by FlowJo (Tree Star). Median 287,340 CD4+ T cells (min: 188,305, max: 701,800) were measured per donor. Antibodies were purchased from BD Biosciences-Pharmingen unless otherwise stated ( Supplementary Table S3 ). Unfortunately, the surface expression of the chemokine CXCR5 on activated cells was highly affected by the assay setup, and thus antigen-specific T follicular helper cells (Tfh, CXCR5 positive) could not be identified and are likely mixed in the other T helper populations. The gating strategy is shown in Supplementary Figure S4A .

Statistics and data visualization was performed using GraphPad Prism version 9.1.1 and the programming language R, using RStudio. For visualization of marker expression, cell frequencies between groups, ggplot2 (V3.3.2), ggpubr (V0.2.5), rstatix (V0.7.0), and ggridges (V0.5.3) packages in R were used. As no normal distribution was assumed, the Wilcoxon signed-rank test was used to compare two or more groups, with unpaired and paired analysis as necessary. The results were adjusted for multiple comparisons using the Bonferroni correction as implemented in the rstatix package. The non-parametric Spearman’s rank-order correlation was used to test for correlation. We used the following convention for symbols indicating statistical significance; ns P > 0.05, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. As there were certain B cell populations identified in the FlowSom clustering that were not present in all 21 donors measured for B cell analysis, we implemented a cutoff of minimal events to be analyzed per population of 1% of the total antigen-specific number for Spike, RBD, and NC. The distribution of antigen-specific cells per cluster was also checked between all donors to make sure all contributed to each cluster. Antigen-specific populations that fell below the cutoff were not taken along in further statistical analysis.

In this study, 410 SARS-CoV-2 experienced individuals >18 years of age were sampled for peripheral blood at the beginning of the pandemic, between 08 April 2020 and 15 September 2020, as part of a larger convalescent blood donor observational study (34). There were 132 individuals who had experienced mild infection that were selected for evaluation of RBD-IgG antibody kinetics based on presence of >2 sample points for serology up to 150 days post-symptom onset (PSO), of which first and last samples were at least 50 days apart and availability of >1 vial of PBMCs ( Figure 1A , Supplementary Figure S1A ). Linear mixed-effects modeling integrating all measuring points of RBD-IgG was used to calculate the half-life of RBD-IgG (34). Based on the estimated half-life of RBD-IgG, 13 individuals were selected who showed the fastest declining antibody titers (Declining Antibody group) as well as 14 individuals who showed the most stable antibody titers over time (Sustaining Antibody group) ( Figures 1A, B ). The estimated half-life for RBD-IgG was clearly different between the two groups ( Figure 1C ). The same was observed for estimated half-life for Spike-IgG and to lesser extent for NC-IgG ( Supplementary Figure S1 ). The median age and sex of donors were comparable between both groups, as were the number of days PSO for the PBMC samples analyzed ( Figure 1D ). Evaluation of RBD-IgG, RBD-IgA, RBD-IgM, Spike-IgG, and NC-IgG levels at time point of PBMC sampling (T1, median: 44d [27-80]) showed no significant differences between the SAb and DAb groups ( Figure 1E ). In line with selection criteria for both groups, at the final time point, around 128 days PSO (Tfinal, median: 128d [57-148]), RBD-IgG and Spike-IgG titers were significantly higher in the SAb group, whereas RBD-IgA, RBD-IgM, and NC-IgG levels were similar between the SAb and DAb groups ( Figure 1F ).

Next, to investigate if differences in antibody RBD-IgG decay over time are associated with formation of specific SARS-CoV-2 B cell subsets, deep profiling of the SARS-CoV-2-specific B cell compartment in blood was performed relatively early after infection, around 45 days PSO (T1). In addition to the three SARS-CoV-2 antigen specificities (Spike, RBD, and NC), B cells reactive to hemagglutinin (HA) from H1N1/pdm2009 influenza virus, fusion glycoprotein (F) from respiratory syncytial virus (RSV), or tetanus toxoid vaccine (TT) were evaluated to compare the composition of the specific B cell compartments early after infection (SARS-CoV-2) to those specific B cell compartments (Flu/RSV/TT) that were established longer ago. Of note, during the time period of analysis, no influenza or RSV infections occurred and due to travel restrictions, almost no adult received a TT vaccination. No differences were observed between the SAb and DAb groups in frequency of Flu-HA, RSV-F, and TT specific total B cells ( Figure 2A ). In addition, the frequency of SARS-CoV-2-specific B cells (Spike, RBD, and NC) was similar between the DAb and SAb groups ( Figure 2A ). With respect to RBD, these results indicate that the size of the circulating RBD-specific B cell compartment at 45 days PSO is not indicative for the measured differences in anti-RBD titer decay over time.

In order to analyze the different differentiation and activation stadia of the antigen-specific B cell compartments, a total of 25 unique flow cytometry markers ( Supplementary Table S1 ) were analyzed with unsupervised high-dimensional analysis, using a computational pipeline including the FlowSOM algorithm (36). The antigen-specific B cells were gated using combinatorial probe gating within live CD19+ B cells ( Supplementary Figure S2A ). In total, 30 different B cell clusters were identified when B cells of all six different antigen specificities were combined with bulk CD19+ B cells ( Figures 2B, C ). Overarching populations were identified based on expression of core B cell markers and subclusters were annotated based on presence of activation markers ( Figures 2C, D ). Similar populations were found previously by us upon deep profiling of specific B cell compartments in individuals having experienced COVID-19 with different grades of disease severity (28). Naïve B cells (IgD+, NBC), classical MBCs (CD27+CD38−) and PB (CD27+CD38+) populations were identified in the antigen-specific compartments. In addition, clusters were identified that contain characteristics of the atypical B cells, including Double Negative (CD11c+IgD-CD27−, DN) B cells, activated Naïve (CD11c+IgD+, aNAV) cells, and age-associated B cells (CD11c+CD27+, ABC), which we recently investigated and characterized after SARS-CoV-2 vaccination (39) and appear during the extrafollicular response prior to GC formation (40). Finally, IgA+, IgM+, and IgG+ activated B cell populations (ActBC) were identified being CD27+, CD71+, CD95+, and CD43+ (clusters 3, 4, and 5). Our previous research on B cell immune profiles after infection identified CD71 and CD43 as defining markers for these ActBC populations (28). Furthermore, a population of antigen-specific CD71+CD27+ B cells was previously shown to arise in the blood at the same time of GC formation after infection and vaccination (41). In addition, a population described as CD27 high but CD21 low B cells (42) shares similarities with this ActBC population and is shown to be early precursors prior to PB differentiation. Taken together, ActBC populations may reflect recent GC experience and be indicative of ongoing GC reactions. These data show that the used multiparameter approach can successfully identify specific B cell populations at different stages of extrafollicular and GC-driven differentiation, hereby generating in depth insight in the composition of antigen-specific B cell compartments and phenotypic B cell subset relationships after infection.

UMAP representations of the individual SARS-CoV-2-specific B cells in the SAb and DAb groups indicate that Spike-, RBD-, or NC-specific B cells ( Figure 2E ) show a different distribution across B cell clusters compared with Flu-, RSV-, or TT-specific B cells, in line with the more recent encounter of SARS-CoV-2 ( Supplementary Figure S2B ). To further investigate this, overall frequencies of the annotated B cell populations in Spike-specific B cells ( Figure 3A ), exceeding a preset minimal threshold of number of events (see M&M), were compared with those in total B cells ( Supplementary Figure S2C ), revealing the Spike-specific level of enrichment of annotated B cell populations ( Figure 3A ). The frequency of the eight most frequent populations in the Spike-specific compartment were next depicted for all antigens investigated and for total B cells ( Figure 3B ). IgG+ ActBC, as well as IgM+ and IgA+ ActBC, are clearly enriched in the Spike-specific B cell compartment of convalescent donors for both groups ( Figure 3A ). These populations are not present when studying total B cell compartments specific for Flu, RSV, or TT that were encountered longer ago ( Figure 3B ). In addition, IgG+ ActBC form a substantial fraction of the Spike-specific response ( Figure 3B ), indicating active ongoing GC reactions 6–8 weeks after mild infection. Also, HLA-DR+CD95+CD86+ PB are strongly enriched in both groups in the Spike-specific compartment. IgG+(CD95+) MBC and IgM+CD95+CD21− MBC (clusters 10, 11, and 14) are enriched for Spike specificity, but these populations are also observed in the B cell compartments of the TT, Flu, and RSV antigens encountered longer ago ( Figure 3B ), consistent with the formation of a long-lived memory response after infection. IgG+(CD11c-CD24+) DN cells are present in Spike-, RBD-, and Flu-specific B cells. CD21− NBC and aNAV (clusters 26 and 27) are among the most frequent Spike-specific B populations, which might indicate recently primed naïve populations ( Figures 3A, B ). Interestingly, CD21− NBC clustered together with the atypical phenotype in the UMAP and heatmap ( Figures 2B, C ) rather than other NBC. The majority of the Spike-specific B cell populations induced after SARS-CoV-2 are comparable in both DAb and SAb groups ( Figure 3C ). Strikingly, within the Spike- and RBD-specific B cells, the IgG+ MBC ( Figure 3C , Supplementary Figure S3A ) are a significantly higher proportion in the DAb group compared with the SAb group, with also a higher overall frequency of RBD-specific IgG+ MBCs ( Supplementary Figure S3B ). This indicates that people who generate stable SARS-SoV-2 antibody levels over time generated a relatively lower GC output toward a B cell memory phenotype at 6–8 weeks after mild infection compared with persons who show a greater decay of SARS-CoV-2-specific antibodies.

Ongoing GC reactions are needed to generate antigen-specific antibodies with higher affinities due to the GC processes of somatic hypermutation and affinity maturation. Over time, this results in antibody outputs with enhanced neutralization capacities. Indeed, other studies have shown an increase in neutralization capacity of SARS-CoV-2 antibodies during late convalescence (16, 19). To study if the observed differences in B cell responses against SARS-CoV-2 between SAb and DAb groups are also reflected in the neutralization capacities of specific antibodies in both groups, the neutralizing potency of plasma samples was investigated at median 44 days (T1) and 128 days PSO (Tfinal) by a competitive ELISA assay. At both timepoints, there were no significant differences in neutralization potency between both groups. Interestingly, neutralizing potency increased significantly over a period of 5 months after infection in the SAb group, as demonstrated by a decrease in the IC50 value, whereas neutralization potency remained unchanged in the DAb group ( Figure 4A ). In vaccination studies, an antigenic shift toward less dominant and more conserved Spike epitopes after second and third vaccination has been described (43, 44). To examine if a broadening of the antibody repertoire to non-RBD epitopes occurred over time, the ratio of Spike to RBD antibody titers was studied and compared between the SAb and DAb groups at the two time points. The Spike to RBD ratio did not change significantly between T1 (median SAb: 1,01; DAb: 1,15) and Tfinal (median SAb: 1,46; DAb: 1,54) and was similar between both groups ( Figure 4B ). Together, these data show that individuals with stable antibody levels over time show an increase in neutralization capacity of the generated specific antibodies.

As the SARS-CoV-2-specific immune response showed signs of GC reactions and thus T cell dependency, the SARS-CoV-2-specific CD4+ T cell response in circulation was compared between the SAb and DAb groups using an activation-induced marker assay. Upon stimulation with Spike (S1 or S2) or NC peptide pools, convalescent donors upregulated activation markers CD40L and CD137 compared with unstimulated control, revealing the SARS-CoV-2-specific CD4+ T cell population ( Figure 4C , Supplementary Figure S4A ). The frequency of SARS-CoV-2-specific CD4+ T cells at 45 days PSO was similar between the SAb and DAb groups, with a high inter-donor variability ( Figure 4D ). Phenotypic analysis of the SARS-CoV-2-specific CD4+ T cells displayed an enrichment of the central memory phenotype (CD27+ CD45RA−) compared with the whole CD4+ T cell population ( Supplementary Figure S4B ). The distribution and frequency of the various memory populations did not differ between the SAb and DAb groups ( Supplementary Figure S4C ). Characterization of the T helper phenotype using chemokine markers to identify a mixed population of T follicular and T helper (T (follicular) helper) 17 cells (T(f)h17, CCR6+), T (follicular) helper 1 cells (T(f)h1, CXCR3+, CCR6−), and T (follicular) helper 2 cells (T(f)h2, CCR4+, CXCR3−, CCR6−) demonstrated a significant enrichment and dominance of T(f)h1 cells in the SARS-CoV-2-specific population compared with total non-naïve CD4+ T cells, and a lower frequency of T(f)h17 and T(f)h2 cells was observed in both the SAb and DAb groups ( Supplementary Figure S5A ). No significant differences in the percentage of the T helper cell populations in SARS-CoV-2-specific CD4+ T cells were found between the SAb and DAb groups ( Supplementary Figure S5B ), indicating that early after mild infection, the balance between T(f)h1, T(f)h2, and T(f)h17 cells is not associated with longevity of the SARS-CoV-2 antibody response. In support, frequencies of SARS-CoV-2-specific cytokine producing CD4+ T cells did not differ between groups, also not for IL-21 producing CD4+ T cells ( Supplementary Figure S6 ).

To assess if and how cellular and humoral parameters relate in both SAb and DAb groups, circulating SARS-CoV-2-specific B and T cell subset frequencies, antibody titers, and neutralization capacity were correlated in both SAb and DAb groups. Clear differences were observed in the overall pattern of correlations between Spike-specific B and T cell subpopulations and antibody titers between DAb and SAb groups ( Figure 5 ). For the DAb group, Spike-specific IgG+ ActBC cell populations positively correlate with the Spike-specific T(f)h2 cell populations ( Figure 5 , Supplementary Figure S7A ). In comparison, in the SAb group, not only did IgG+ ActBC and MBC B cell populations strongly correlate with Spike-specific T (f)h2 ( Supplementary Figure S7A ) but also additional correlations were observed between distinct atypical B cell and PB populations with T(f)h1 and T(h)2. In support, Spike-specific IgG+ MBC also strongly correlated with Spike-specific IFNγ+ or TNF+ CD4+ T cells, whereas some Spike-specific IgG+ DN cells correlated with Spike-specific IL-21 producing CD4+ T cells ( Supplementary Figure S7B ). None of the B cell populations correlated with antibody titers at T1 or Tfinal for either group. For the SAb group, the Spike-specific B cell subsets had a tendency toward positive correlation with the neutralization capacity at T1 in contrast to the DAb group; however, the DAb group displayed a strong negative correlation with the IC50 at the final timepoint for several B cell subsets. All Spike-specific CD4+ T cell helper subsets including cytokine-producing T cells were strongly correlated in the DAb groups with antibody titers at both timepoints, which was not the case for the SAb ( Figure 5 , Supplementary Figures S7A, B ). In conclusion, the DAb group showed strong correlations between Spike-specific CD4+ T cells and Spike-specific IgG antibodies, but overall no major correlations between B cells with CD4+ T cell or antibodies was found. In contrast for the SAb group, a larger array of correlations between Spike-specific B cell populations and CD4+ T cells were observed but no cellular correlations with Spike-IgG titers.