Tracking B cell responses to the SARS-CoV-2 mRNA-1273 vaccine

SUMMARY

Protective immunity following vaccination is sustained by long-lived antibody-secreting cells and resting memory B cells (MBCs). Responses to two-dose SARS-CoV-2 mRNA-1273 vaccination are evaluated longitudinally by multimodal single-cell analysis in three infection-naıïve individuals. Integrated surface protein, transcriptomics, and B cell receptor (BCR) repertoire analysis of sorted plasmablasts and spike⁺ (S-2P⁺) and S-2P⁻ B cells reveal clonal expansion and accumulating mutations among S-2P⁺ cells. These cells are enriched in a cluster of immunoglobulin G-expressing MBCs and evolve along a bifurcated trajectory rooted in CXCR3⁺ MBCs. One branch leads to CD11c⁺ atypical MBCs while the other develops from CD71⁺ activated precursors to resting MBCs, the dominant population at month 6. Among 12 evolving S-2P⁺ clones, several are populated with plasmablasts at early timepoints as well as CD71⁺ activated and resting MBCs at later timepoints, and display intra- and/or inter-cohort BCR convergence. These relationships suggest a coordinated and predictable evolution of SARS-CoV-2 vaccine-generated MBCs.

Graphical Abstract

In brief

Using multiomic single-cell analyses, Assis et al. show a coordinated trajectory involving plasmablasts and activated and resting memory B cells in response to primary SARS-CoV-2 mRNA vaccination. Spike-specific BCR repertoire analysis shows incremental affinity maturation across the 6-month study period and reveals evidence of convergence among study participants and other cohorts.

INTRODUCTION

The emergence and rapid spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has led to an equally rapid worldwide effort to develop COVID-19 vaccines across several different platforms.¹ Among the most successful and widely approved are mRNA-based vaccines, including Moderna mRNA-1273 and BNT162b2 Pfizer/BioNTech, both of which encode a stabilized ectodomain of the spike protein trimer (S-2P) that elicits strong neutralizing antibody responses after two doses.^2,3 Despite the strong protective effects of mRNA and other SARS-CoV-2 vaccines, numerous studies have shown that there is a wide range of immune responses to the vaccines, driven in part by host factors such as age, immune deficiencies, and other underlying conditions.⁴ Furthermore, recent epidemiological studies have demonstrated that protection wanes over time and that the increase in susceptibility to SARS-CoV-2 infection, albeit with decreased severity, is at least in part associated with diminishing effectiveness of vaccines.^5,6 Several immunologic studies have demonstrated a decay of humoral immunity to SARS-CoV-2 vaccines over time,⁷ underscoring the importance of vaccine-induced neutralizing antibodies in protection from severe illness.^8,9 Despite these advances, immune correlates of protection remain poorly understood yet are critical for the development of effective vaccine strategies against rapidly emerging variants.¹⁰

B cell responses to vaccination contribute to durable immunity through the antibodies they secrete, mainly from long-lived plasma cells that reside in the bone marrow, and memory B cells (MBCs) that can rapidly respond upon re-exposure to viral antigens.¹¹ Several studies have shown that while antibodies against SARS-CoV-2 decline over several months after vaccination and/or infection, MBC responses either rise or stabilize over the same period.^12–20 However, MBCs are highly heterogeneous and display a range of phenotypes that vary over time, anatomical location, and source of exposure.^11,21,22 After SARS-CoV-2 infection, the early response is dominated by activated populations, including atypical MBCs (CD21^loCD27^+/− CD11c⁺), plasmablasts (PBs), CD71⁺-activated B cells, and some with intermediate phenotypes.^16,23,24 Similar populations have been reported early after primary vaccination,^12,13 albeit at lower frequencies compared with infection.²⁴ After infection and/or vaccination, activated spike-specific MBCs are gradually replaced by MBCs with a classical resting phenotype (CD21⁺CD27⁺CD11c⁻ CD71⁻) that have been associated with a durable memory response.^{12,15,16,18,19,23,25}

In addition to their longevity, we and others have shown that spike-specific MBC responses are strong correlates of spike-specific antibody titers following the initial two mRNA vaccine doses in uninfected individuals,^12,13,20 suggesting coordination and/or relationships between antibody-secreting cells and MBCs. Furthermore, responses among several distinct PBs and MBCs at baseline of dose two and in the weeks thereafter were found to predict antibody titers as far out as 6 months after vaccination,¹³ indicating a role for these early B cell populations in driving antibody responses. Despite these correlates, trajectories and relationships between B cell populations following vaccination remain poorly understood. To address these relationships and to gain insight into the evolution of B cell responses to SARS-CoV-2 mRNA-based vaccination, we performed in depth longitudinal single-cell transcriptomic and epitope sequencing (CITE-seq), and B cell receptor (BCR) repertoire profiling on B cells of three study participants from our primary vaccine study.¹³ We found a coordinated trajectory involving PBs, and activated and resting MBCs that provide insight into the B cell populations involved in the development of immunologic memory following SARS-CoV-2 vaccination.

RESULTS

Transcriptomic and surface protein analyses of spike-specific B cell response after primary vaccination

To track the development of B cell responses to the two-dose mRNA-1273 vaccine, we performed longitudinal droplet-based single-cell transcriptome and epitope sequencing (CITE-seq) and BCR sequencing on PBs and non-PB B cells of three SARS-CoV-2 uninfected healthy adults who had participated in a previous study on primary immune responses to mRNA vaccination.¹³ Sorting of spike-specific (S-2P⁺) and non-spike-specific (S-2P⁻) B cells was performed at six timepoints post-vaccination: vaccine dose 1 (v1) day 14 (D14), v2D6 and v2D9, v2D14, and v2D28 and month (M) 6 (Figure 1A). Total PBs were also sorted at v1D14, v2D6, and v2D9, the first three time-points where PBs were enough to be processed separately. The three study participants were chosen based on frequencies of S-2P⁺ B cells that would yield enough cells at all timepoints. These frequencies among total CD19⁺ B cells ranged from 0.2%–0.4% at v1D14 to peak at v2D7/9 of 0.9%–2.3% and 0.6%–1.4% at M6 (Table S1). Purified B cells from the three study participants were combined by time point. Total PBs, S-2P⁺, and S-2P⁻ B cell populations were then sorted (Figures 1B and 1C) and subjected to single-cell sequencing using appropriate barcoding (hashtag [HTO] antibodies) and individual single nucleotide polymorphisms (SNPs) to demultiplex the data (see methods and Figure 1C). Of note, two oligo-tagged spike antigens (RBD and S1) were included among the antibody-derived tags (ADT; Table S2) at the last two timepoints where there were enough cells to perform the ADT step after sorting. To minimize cell loss, the ADT step was otherwise combined with sorting but excluded RBD and S1 because of interference with the S-2P tetramer used to sort.

Figure 1. Study design.

(A) Graphical depiction of blood draws, cell sorting, and ADT staining for SARS-CoV-2-uninfected vaccinees (n = 3) after receiving the two-dose mRNA-1273 vaccine. See also Table S1.

(B) B cell gating strategy for sorting PBs (CD19⁺CD20⁻), CD20⁺ S-2P⁻ (−), and S-2P⁺ (+) B cells.

(C) Workflow for single-cell analyses showing steps from sample processing through data analysis.

A total of 109,225 cells from all study participants and timepoints combined were analyzed, among which were 66,155 S-2P⁻ and 35,667 S-2P⁺ cells and 7,403 PBs (Table S3). The proportional contribution to each sorted population varied by time point sampled, ranging from 9% to 23% for S-2P⁻, 6%–35% for S-2P⁺, and 11%–64% for PBs (Figure 2A and Table S3). However, the number of cells contributed to each time point was nonetheless substantial with the lowest being 2,217 S-2P⁺ cells at v2D6 and 842 PB at v2D9 (Table S3), the latter reflecting a rapid waning of PBs in circulation after a peak at v2D6.¹³ Unsupervised clustering was performed on gene expression using the Seurat v4.1.0 pipeline.^26,27 Seven distinct cell clusters (C) in the integrated dataset were identified and visualized using uniform manifold approximation and projection (UMAP) dimensionality reduction (Figures 2B and S1A). The clusters were annotated based on gene and protein expression of nine canonical surface markers (Figure 2C). Among the seven clusters, the three identified as naive (C1, C2, and C7) collectively contained the most cells, as expected for peripheral blood B cells. Three other clusters were identified as MBCs (C3, C4, and C5) and one distinct cluster comprised the PBs (C6). There was concordance between gene and protein expression, although lack of an available ADT for immunoglobulin (Ig)A precluded its detection at the protein level, and mRNA levels of CD21 and to a lesser extent CD38 were low, consistent with other single-cell-based analyses on human peripheral blood B cells.²⁸ Clusters C1, C2, and C7 expressed markers consistent with naive B cells (IgD^hiIgM⁺ CD38⁺CD21⁺CD27⁻) and differential gene expression (DGE) analysis confirmed the graph-based clustering and delineated both commonalities and differences within the three naive B cell clusters (Figure 2D). Similarly, three predominantly MBC clusters were identified by their common CD38^loCD21⁺CD27⁺ profile and distinguished by their predominant immunoglobulin isotype: unswitched IgM⁺IgD^+/− C3 and IgG/A switched C4 and C5 (Figure 2C). Of note, while IgM⁺ and IgA⁺ MBCs were present in both C4 and C5, IgG was clearly the dominant isotype in C5 but not C4 (Figure 2C). The PB cluster C6 had a characteristic CD20^loCD38^hiCD21^loCD27^hi profile and equally strong distribution of IgA and IgG, consistent with previous observations.¹³

Figure 2. Single-cell analyses of sorted populations show distinct clusters and enrichment of S-2P⁺ cells.

(A) Bar graphs showing each time point contribution to the three sorted populations. See also Table S3.

(B) UMAP showing unsupervised clustering of 109,225 B cells (integrated dataset) using gene expression data. See also Figure S1A.

(C) Violin plots showing mRNA and surface protein (ADT) expression of canonical B cell markers and Ig isotypes for each cluster. The x axis shows the globalscaled and log-normalized gene expression value obtained using the function “NormalizedData” from the Seurat R package.

(D) Dot plots showing the top-10 differentially expressed genes ordered by average log2 fold change for each cell cluster. Color intensity indicates average expression, whereas dot size denotes the percentage of cells expressing the gene.

(E) Bar graphs showing contribution of non-naïve B cell clusters to each time point for the three sorted populations. The cell populations are indicated at the top of the bars. The p values shown reflect enrichment of MBC-C5 among S-2P⁺ versus S-2P⁻ cells at each time point. Differences were evaluated by chi-square test. ***p ≤ 0.001. See also Figure S1B.

(F) UMAP showing kinetics of S-2P⁺ cell distribution among all clusters. Red dots represent S-2P⁺ cells at each time point and gray areas represent all other cells.

The DGE analyses revealed expected differences among the three major populations: naive B cell clusters expressed higher levels of IL4R, FCER2 (CD23), BACH2, and TCL1A than MBC and PB clusters, while the PB cluster expressed high levels of genes associated with antibody secretion (Figure 2D). Peripheral blood MBCs tend to be diverse with fewer distinct transcriptomes than other B cell populations.²⁹ Nonetheless, GPR183 (EBI2) and TNFRSF13B (TACI), two MBC-defining genes,^28,30 were expressed in all three MBC clusters (Figure 2D). In addition to the clear contrasts among naive, MBC, and PB clusters, we noted several features that were distinct to MBC-C5 (Figures 2C and 2D). Notably, MBC-C5 contained cells with a CD19^hiCD20^hiCD38^‒ CD21^loCD27^lo profile, distinct from the other MBC clusters (Figure 2C), and consistent with atypical MBCs (also known as tissue-like memory, age-associated [ABCs], or double-negative [DN2] B cells; reviewed in Courey-Ghaouzi et al.³¹). DGE analyses also revealed a unique set of genes for MBC-C5 (Figure 2D), including HOPX, PLEK, FGR, and CIB1 that have been shown to be upregulated in atypical MBCs.^32,33 In addition, S-2P⁺ cells were significantly enriched among MBC-C5 (p < 0.0001, chi-square test), especially at v2D9 and v2D14 timepoints (Figures 2E and S1B), consistent with previous observations showing increased frequencies of atypical MBCs at these timepoints.¹³ Temporal changes in frequencies of S-2P⁺ cells among clusters were also observed (Figure 2F), with the majority of S-2P⁺ cells detected among naive clusters at v1D14 when overall frequencies of spike-specific B cells were low and a shift to MBC clusters as these frequencies increased after dose two.¹³ In contrast to S-2P⁺ cells, the frequency of MBC-C5 remained more stable over time among sorted S-2P⁻ cells (Figures 2E and S1B). These observations are consistent with large expansions and contractions over time after exposure to SARS-CoV-2 infection or vaccination being concentrated among PBs as well as activated antigen-specific B cell populations similar to those found within MBCC5.^{16,23,24,34,35}

Collectively, several distinct populations of B cells were identified in response to two-dose mRNA-1273 vaccination and distinct fluctuations were revealed by combined transcriptomic and surface protein single-cell longitudinal profiling of PBs and S-2P-sorted cells.

Kinetics and characterization of spike-specific B cells within MBC-C5

Given the enrichment and temporal changes of S-2P⁺ cells within MBC-C5 and its distinct albeit heterogeneous features, we further explored this cluster by re-clustering its 6,524 cells. This yielded five subclusters, named MBC-SC5.1 through MBC-SC5.5 that fluctuated over time (Figures 3A and S1C). We identified distinct gene expression profiles for four of the five subclusters (MBC-SC5.2–5.5), as shown by the heatmap of the top 10 differentially expressed genes within each subcluster (Figure 3B). MBC-SC5.1 was least distinctive, with a bridging pattern that suggested a mixed population (Figure 3A). Its signature was dominated by a strong expression of C-X-C motif chemokine receptor 3 (CXCR3), which has been associated with homing of B cells to sites of inflammation.³⁶ The remaining sub-clusters were more distinct and further annotated by analysis of combined gene and protein expression of the canonical cell-surface markers based on those shown in Figure 2C plus three additional gene/protein pairs: CD11c, which has become the main defining marker for atypical MBCs,³¹ and the activation/proliferation markers CD71 and CD95^23,37 (Figure 3C). MBC-SC5.2 was a resting MBC, based on low expression of activation markers and CD11c. The remaining three subclusters expressed two sets of distinct markers: activated CD71⁺CD95⁺CD38⁺CD11c⁻ MBC-SC5.3 and atypical CD71⁻CD95^+/−CD38⁻CD11c⁺ MBCSC5.4 and MBC-SC5.5. These latter two clusters also contrasted with the other MBC-C5 subclusters by their CD19^hiCD20^hiCD21^loCD27^lo profile, as well as several genes shown by the asterisks in Figure 3B consistent with atypical MBCs.^32,38,39 Of note, while MBC-SC5.4 and MBC-SC5.5 clustered together by UMAP analysis and displayed similar gene expression profiles, several genes highly expressed among activated MBC-SC5.3 were also expressed by MBC-SC5.4 (Figures 3A and 3B), and MBC-SC5.5 contained IgD/M-expressing cells while the other MBC-C5 subclusters expressed predominantly IgG (MBC-SC5.2) or IgG and IgA (MBC-SC5.1/3/4). MBC-SC5.1 also contained both IgG and IgA-expressing cells, as well as intermediate profiles for canonical markers associated with atypical and activated MBCs (Figure 3C), consistent with a mixed cluster. Notwithstanding mixed MBC-SC5.1, the subclustering of MBC-C5 revealed that it contained several distinct MBC populations that could be categorized as activated, atypical, or resting.

Figure 3. Single-cell analyses within MBC-C5 show distinct subclusters and enrichment of S-2P⁺ cells.

(A) UMAP (left) showing unsupervised re-clustering of 6,524 cells from cluster MBC-C5 in Figure 2B using gene expression data and bar graphs showing contribution of each subcluster at each time point. See also Figure S1C.

(B) Heatmap showing the top-10 differentially expressed genes for each subcluster. Red asterisks show genes associated with atypical MBCs.

(C) Violin plots showing mRNA and surface protein (ADT) expression of B cell markers and Ig isotypes for each cluster. The x axis shows the global-scaled and lognormalized gene expression value obtained using the function “NormalizedData” from the Seurat R package.

(D) UMAP showing kinetics of S-2P⁺ cell distribution among MBC-C5 subclusters. Red dots represent S-2P⁺ cells at each time point and gray areas represent all other cells. See also Table S3.

(E) UMAP showing the distribution of S1⁺ (top) and RBD⁺ (bottom) cells among MBC-C5 subclusters at v2D28 (red) and M6 (blue) timepoints.

(F) UMAP showing the single-cell trajectories inferred by Monocle3 within MBC-C5. Mixed MBC-SC5.1 was used as the root node based on kinetics depicted in Figure 3D. The encircled number denotes the positioning of the root node. The lines represent the principal graph generated by Monocle3 and the cells are colored according to relative pseudotime.

Among the few S-2P⁺ cells detected in MBC-C5 at early time-points v1D14 and v2D6, most were concentrated in the CXCR3⁺ subcluster MBC-SC5.1 (Figures 3D, and S1B). From days 9 through 14 after dose 2 (v2D9 and v2D14), the distribution of S-2P⁺ cells expanded to include activated MBC-SC-5.3 and atypical MBC-SC5.4/5 but not resting MBC-SC5.2 (Figure 3D). By v2D28, S-2P⁺ cells had decreased in MBC-SC5.1 while they were increasing among resting MBC-SC5.2. By M6, the majority (84%) of C5 S-2P⁺ cells were in resting MBC-SC5.2 (Figure 3D and Table S3). RBD and S1 tetramer ADTs were also included at the last two timepoints (Figure 1A), providing an opportunity to confirm specificities among S-2P⁺ cells and display their binding among clusters. The distribution of RBD⁺ and S1⁺ cells was similar to that of S-2P⁺ cells at v2D28 and M6 within MBC-C5 (Figure 3E), where enrichment among resting MBC-SC5.2 increased from averages of 28% and 32%–89% and 90%, respectively, between the two timepoints (Figure S1D).

We further investigated the development of the S-2P response within MBC-C5 and whether temporal associations between subclusters were also identifiable transcriptionally. To test this, we used Monocle3, a program designed for pseudotime analyses that reconstructs single-cell differentiation trajectories based on the transcriptional reconfiguration involved in cell-state transitions.⁴⁰ Given the enrichment of S-2P⁺ cells in MBC-SC5.1 at early timepoints (Figures 3A and 3D), this subcluster was used as the root node. We identified two major trajectories: one was directed toward MBC-SC 5.2/3, while the second was directed toward subclusters MBC-SC 5.4/5 (Figure 3F). Overall, the trajectory analysis was consistent with the kinetics of S-2P⁺ B cells (Figure 3D), with one of two major trajectories ending with activated MBC-SC5.3 and resting MBC-SC5.2.

Clonal expansion, somatic hypermutation, and diversification of S-2P⁺ B cells

To further elucidate and track the B cell response following the two-dose SARS-CoV-2 mRNA-1273 vaccine, we analyzed the BCR repertoire from single cells collected at each of the six timepoints post-vaccination using the Immcantation framework (Figure 1A, immcantation.org). A clone of B cells was defined as a set of B cells that descended from a common naive B cell and potentially differ by somatic hypermutations (SHMs). We identified clones by clustering sequences from the same participant with common V/J genes, CDR3 lengths, and an individually determined CDR3 nucleotide distance threshold (see methods). From the full dataset, we identified a total of 120,734 clones, including 4,899 expanded clones (two or more cells). Among expanded clones, 23.7% contained members from two or more cell populations (Figures 4A and S2). The frequency of highly expanded clones (≥ 10 cells) varied across cell populations (PBs = 1.03%, S-2P⁺ cells = 0.055%, and S-2P⁻ cells = 0.005%).

Figure 4. BCR repertoire and clonal overlap are distinct for S-2P⁺ cells.

(A) Venn diagram showing clonal distribution of expanded clones among cell types. The number of unique and shared clones is indicated. See also Figure S2.

(B) Clonal diversity of S-2P⁺ and S-2P⁻ B cells within each participant over time. Clonal diversity is quantified as diversity order q = 2 (inverse Simpson’s index). Points show the mean of 500 resampling repetitions, error bars show 95% confidence intervals. Clones were classified as S-2P⁺ if they contained at least one S-2P⁺ cell. Only non-naı¨ve S-2P⁺ and S-2P⁻ sorted B cells were included in this analysis.

(C) Heatmap showing overlap of S-2P⁺ (lower triangle) and S-2P⁻ (upper triangle) clones across timepoints. Color intensity indicates overlap strength as measured by the Jaccard index. Only non-naïve S-2P⁺ and S-2P⁻ sorted B cells were included in this analysis.

(D) Mean SHM frequency among clones sampled at M6 versus earlier timepoints. Each point represents the mean SHM frequency of all heavy chain sequences comprising a clone sampled within the specified time period. Lines connect clones across time periods. p values were calculated using a paired Wilcoxon test, and numbers in parentheses indicate number of clones tested. See also Figure S3.

To characterize the dynamics of clonal expansion over time, we quantified clonal diversity and overlap among timepoints. We classified clones as “S-2P⁺” if they contained at least one sorted S-2P⁺ B cell, and as “S-2P⁻” if they contained no S-2P⁺ B cells. Consistent with clonal expansion that occurs during an adaptive immune response, the overall inter-clonal diversity of clones from S-2P⁺ B cells was lower than that of their S-2P⁻ counterparts, and this difference was maintained through 6 months after the first vaccine dose (Figure 4B). To ensure these results were not biased by changes in cell type composition, these analyses were restricted to non-naıïve B cell clusters MBC-C3-C5 (Figure 2B). S-2P⁺ clones were more likely than S-2P⁻ clones to contain cells from multiple timepoints (Figure 4C), further indicative of persistent clonal expansion among S-2P-binding B cells, and consistent with prolonged germinal center reactions observed with SARS-CoV-2 mRNA vaccines.⁴¹ For all three participants, S-2P⁺ clones were most likely to overlap between adjacent timepoints, particularly v2D9 and v2D14 in VAC003 and VAC716, and v2D28 and M6 in VAC611 (Figure 4C).

To determine whether affinity maturation through the acquisition of SHMs was occurring after vaccination, we compared the mean mutation frequency of each B cell clone sampled at early (v1D14–v2D28) and late (M6) timepoints. In all three participants, S-2P⁺ clones contained significantly higher levels of SHM at M6 than at earlier timepoints (Figure 4D). In contrast, there were no significant differences in SHMs among S-2P⁻ clones between M6 and earlier timepoints (Figure 4D). These results indicate that S-2P⁺ B cell clones as a group accumulated new mutations following vaccination.

To explore whether distinct clusters accumulated new mutations over time, we compared SHM levels across timepoints within each cell cluster and MBC-C5 subcluster. More specifically, within each clone we calculated the mean SHM frequency of cells from each cluster and MBC-C5 subclusters at M6. We then compared this to the mean SHM frequency of all sequences from the same clone, regardless of cluster, at earlier timepoints (Figure S3). In agreement with the comparison of all clones at early and late timepoints (Figure 4D), S-2P⁻ clones did not show a significant accumulation of SHMs over time when analyzed by cluster (Figure S3). In contrast, S-2P⁺ cells within switched MBC clusters MBC-C4, resting MBC-SC5.2, and activated MBC-SC5.3 showed significantly higher SHM levels at M6 compared with sequences from the same clone at earlier time points. These results indicate ongoing SHM among S-2P⁺ B cells within cluster MBC-C4, and subclusters MBC-SC5.2 and MBC-SC5.3.

Having shown that S-2P⁺ clones increased in SHMs over time as a group, we next searched for individual clones that accumulated mutations post-vaccination. To this end, we built B cell lineage trees for clones containing S-2P⁺ cells using the R package dowser⁴² and IgPhyML,⁴³ and applied a previously developed test for measurable evolution to determine whether each lineage had a statistically significant increase in SHMs over time.⁴⁴ We identified 12 measurably evolving clones (Figure S4). Because these clones contained at least one S-2P⁺ B cell, they were likely responding to vaccine-expressed antigens. At early timepoints, these evolving clones were primarily from IgG⁺ PBs, followed by an increasing contribution from MBC-C5 subclusters after v2D9 (Figure S5A). Among non-evolving S-2P⁻ clones, their source varied over time from PBs, to unswitched MBC-C3 and switched MBC-C4, while several clusters contributed to non-evolving S-2P⁺ clones (Figure S5A). These results indicate that the composition of B cell clones evolving in response to vaccination shifted over 6 months from PBs to MBC clusters–primarily resting MBC-SC5.2 and activated MBC-SC5.3.

Clonal association between early activated CD38⁺ MBCs and month 6 resting MBCs

In previous work, S-2P⁺ resting MBCs were found to be strong predictors of the antibody response to the two-dose mRNA-1273 vaccine and the spike-specific response at M6 was dominated by these resting MBCs.¹³ Having confirmed here that the spike-specific B cell response was indeed enriched among resting MBCs (MBC-SC5.2) at M6 (Figure 3D), we next investigated how these MBCs were clonally connected to other clusters at different timepoints. Given that B cell clones and lineage trees can link cells of different subtypes to a common ancestor, we used BCR clonal analyses to detect clonally related cell types and potential ancestor/ descendant relationships. In the participants VAC003 and VAC611, the resting MBC-SC5.2 cells had the highest clonal overlap with activated MBC-SC5.3, followed by atypical MBC-SC5.4 in VAC003 and mixed MBC-SC5.1 in VAC611 (Figure 5A). In the participant VAC716, the resting MBC-SC5.2 had the highest clonal overlap with mixed MBC-SC5.1, followed by activated MBC-SC5.3 (Figure 5A). Next, we evaluated whether resting MBC-SC5.2 at M6 was associated with cell types at earlier timepoints. Consistent with the clonal overlap analyses, clones containing resting MBC-SC5.2 cells at M6 had precursors at v2D9 and v2D28 within activated MBC-SC5.3, mixed MBC-SC5.1, and atypical MBC-SC5.4 (Figure S5B). These clones were also found among early PBs (v1D14, Figure 5B), indicating that these clones may be selected primarily following the first vaccine dose in both PB and MBC lineages. Importantly, these patterns of clonal overlap were only observed among S-2P⁺ clones, indicating that they were likely driven by the B cell response to the vaccine.

Figure 5. Tracking of B cell responses shows relationships between MBCs and PBs.

(A) Heatmap showing overlap of S-2P⁺ (lower triangle) and S-2P⁻ (upper triangle) clones across cell type clusters. Color intensity indicates overlap strength as measured by the Jaccard index.

(B) Three largest lineage trees from S-2P⁺ clones containing resting MBC-SC5.2 clones at M6. SP tests showing enrichment of (C) switches from cell types to resting MBC-SC5.2 at all timepoints, and (D) switches from earlier timepoints to resting MBC-SC5.2 at M6 within B cell trees. Cells are shaded per SP test p value. Comparisons with p < 0.05 are marked with an X.

Lineage trees were generated to explore the history of shared mutations and trace contributing populations within clonal families. Among the three largest S-2P⁺ clones that contained resting MBC-SC5.2 at M6, all were populated with activated MBC-SC5.3, atypical MBC-SC5.4, and PBs (Figure 5B). To quantify the associations between these cell types, we used the switch proportion (SP) test in dowser.⁴² Briefly, this test compares the proportion of cell type switches in each tree to the same trees with the cell types randomized among all S-2P⁺ trees with multiple cell types within a participant. Using this test, we found a significant enrichment of switches from activated MBC-SC5.3 to resting MBC-SC5.2 in participants VAC003 and VAC611, whereas these switches were from IgG/A switched MBC-C4 in participant VAC716 (Figure 5C). Given the importance of MBC-SC5.2 at M6 in driving the B cell and antibody responses to vaccination,¹³ we repeated this test on switches to resting MBC-SC5.2 at M6 from earlier timepoints. We found similar trends, with significant enrichment of switches from mixed MBC-SC5.1 in participant VAC716, and MBC-SC5.3 in participant VAC003 (Figure 5D). Taken together, our results from both clonal overlap and phylogenetic analyses indicate a close association between resting and activated MBCs, consistent with the pseudotime trajectory analysis (Figure 3F). These analyses indicate that activated MBCs may be a precursor population of resting MBCs at M6.

Convergent BCRs among S-2P⁺ MBCs

Both V(D)J recombination and affinity maturation occur through highly stochastic processes, and as such, similar BCR sequences among individuals are expected to be rare.⁴⁵ However, highly similar (convergent) SARS-CoV-2 spike-specific antibodies have been documented from several independent studies,^8,28,46–49 suggesting similarities in responses to the spike protein. To identify B cells with convergent BCRs in the current study, we clustered B cells across the three participants that contained the same heavy chain V and J gene, CDR3 length, and at least 80% CDR3 amino acid similarity, as well as the same light chain V gene, J gene, and CDR3 length. Sequence clusters formed by similar BCRs across multiple participants were defined as convergent sequence clusters. We identified 202 convergent clusters in total, many of which contained multiple sorted populations: 158 of these clusters contained S-2P⁺ cells, 106 contained PBs, and only 53 contained S-2P⁻ cells. In all three participants, S-2P⁺ convergent sequence clusters (Tables S5 and S6) were enriched for subclusters within MBCC5 (p < 2.2e-16, chi-square test), especially MBC-SC5.1–4 (Figure 6A). Compared with sequence clusters found in only one participant, convergent sequence clusters among PB and S-2P⁺ cells had higher frequency of IGHV1–69 and IGHV3–30, which were also found at high frequency in the public SARSCoV-2 antibody database CoV-AbDab⁵⁰ (Figure 6B). Similarly, light chains among convergent clusters had higher frequencies of IGLV1–40 and IGLV3–1 among PBs and S-2P⁺ but not S-2P⁻ cells. These light chain genes were also found at high frequency among publicly reported SARS-CoV-2 antibodies (Figure S6).

Figure 6. Intra- and inter-cohort convergence of S-2P⁺ clones mainly enriched in MBC-C5.

(A) Proportion of cell types among S-2P⁺ B cells within BCR sequence clusters convergent across multiple participants, compared with S-2P⁺ B cells in clusters found in only one participant. Cluster 5 subclusters are arranged at the bottom of each barplot.

(B) IGHV gene usage among convergent clusters. Colored bars indicate whether the sequence cluster was found in multiple participants (convergent) or only one (not convergent). The frequency of IGHV genes in the SARS-CoV-2 antibody database, CoV-AbDab, is also shown. Only IGHV occupying at least 5% of any comparison group are shown. See also Figure S6.

(C) Proportion of cell types among S-2P⁺ B cells with heavy chain BCR sequences similar to previously published SARS-CoV-2 antibodies (“public”), compared with S-2P⁺ B cells that did not match previously published antibodies.

(D) Phylogenetic tree of a measurably evolving clone expressing similar BCR sequences in all three participants. Amino acid alignment of convergent IGHV sequences from participants, as well as sequences listed in the Cov-AbDab database. See also Table S5.

Next, we determined whether our dataset contained BCR heavy chain sequences that were similar to those of vaccine and infection-derived coronavirus antibodies deposited in the CoV-AbDab database.⁵⁰ In this analysis, we directly compared each heavy chain to previously published antibody sequences. We classified heavy chains as “public” if they used the same V gene, J gene, CDR3 length, and at least 80% CDR3 AA similarity to a previously published anti-SARS-CoV-2 heavy chain. Consistent with our analysis of sequence convergence among participants, S-2P⁺ cells with public heavy chains similar to the CoV-AbDab database were highly enriched within MBC-C5 sub-clusters (Figure 6C, p < 2.2e-16, chi-square test), further suggesting a key role for MBC-C5 in the B cell response to SARSCoV-2 vaccination and infection.

Finally, we found one S-2P⁺ B cell clonal lineage from participant VAC003 that displayed all salient features described thus far (Figure 6D and Table S5): the lineage was detected at v1D14 in PBs, followed by atypical MBC-SC5.4 and activated MBC-SC5.3 after dose two, and resting MBC-SC5.2 at M6. This lineage was also measurably evolving, accumulating a statistically significant number of SHMs over the 6 months of study (p = 0.039, Figure S4). Finally, this lineage contained B cell sequences that were part of a convergent cluster found across all three participants, as well as public heavy chains similar to multiple previously published anti-SARS-CoV-2 antibodies (Figure 6D). Overall, these findings demonstrate a consistent trajectory across several distinct B cell populations, with evidence from convergence that this trajectory may be widespread.

DISCUSSION

B cell responses to SARS-CoV-2 vaccination are thought to contribute to protection from infection and severe COVID-19 through the production of neutralizing antibodies and long-lasting MBCs.^11,22 However, the trajectory and interconnectivity of these responses remain poorly understood, yet are essential for identifying robust correlates of immunity, which in turn are essential for advancing vaccine strategies aimed at improving the longevity and breadth of protective immune responses. To this end, we examined the trajectory of the B cell response to the primary two-dose SARS-CoV-2 mRNA-1273 vaccine by performing longitudinal analyses of the transcriptome, surface proteins, and BCR repertoire of single cells of three cellular populations (PBs at early timepoints and S-2P⁺ and S-2P⁻ B cells at all timepoints) that were sorted from the peripheral blood of three vaccinees who remained uninfected throughout the 6-month study period. We found similar trajectories, incremental accumulation of mutations, and evidence of convergence in BCR sequences among S-2P⁺ B cells of the three vaccinees studied, as well as convergence with BCR sequences reported from other SARS-CoV-2 vaccine and infection studies.^8,28,46–49 While we did not directly address whether the evolving clonal lineages translated into increased potency, this would be expected based on the above studies where antibodies were reconstituted from the convergent BCR sequences.^25,28,47 Among the B cell clones we identified, several contained combinations of PBs, and activated and resting subclusters within MBC-C5, suggesting a key role for these cell types in the B cell response to vaccination. Furthermore, intra- and inter-cohort BCR convergent sequences were also enriched within MBC-C5, further suggesting a key role for this cluster in the B cell response to vaccination.

Human MBCs are highly heterogeneous,^51,52 indicative of the complexity and distinct functionalities of each cell type, as well as underscoring the importance of identifying those responsible for the development of durable immunity. The combined tracing of BCR repertoires and phenotypic/transcriptomic properties using single-cell analysis, even if limited by the number of cells examined, has the power to delineate relationships between cell types involved in the generation of B cell memory.^53–56 In a longitudinal study on B cell responses in people with mild and severe COVID-19, several distinct cell types dominated the early phase of infection, including PBs and several activated MBCs, while these were largely replaced by resting MBCs after 6 months.²³ Several studies have since demonstrated that resting MBCs are responsible for a stable long-lasting memory B cell response following infection and/or vaccination, even as antibody titers wane.^{12–14,17,39}

In the current study, we initially identified seven distinct clusters based on transcriptomics. Among these were three MBC clusters, one of which, MBC-C5, was enriched with S-2P⁺ cells, consistent with its predominance of IgG that drive responses to SARS-CoV-2 mRNA vaccines, while IgA and IgM, which dominated MBC-C3 and MBC-C4, are not strongly elicited by these vaccines.^13,20,57 We further delineated the cell types involved in the development of the B cell response to vaccination by subclustering MBC-C5. We identified four clearly delineated subclusters, namely atypical CD11c-expressing (MBC-SC5.4 and MBC-SC5.5), activated CD71-expressing (MBC-SC5.3) and resting (MBC-SC5.2) MBCs, along with a mixed bridging MBC (MBC-SC5.1). Within these five subclusters, S-2P⁺ cells were mostly found among mixed MBC-SC5.1 at early timepoints, followed by increases among atypical and activated MBC-C5 subclusters and finally the resting MBC-SC5.2, of which a large majority were IgG. Several studies involving infection and/or vaccination cohorts have confirmed these temporal dynamics,^{16,20,24,35,39,58,59} although few have recapitulated the complexity of responding cell types described in the earlier study.²³ We also identified two distinct vaccine-induced trajectories from the mixed MBC-SC5.1, one in the direction of atypical CD11c-expressing MBC-SC5.4 and MBC-SC5.5 and the other in the direction of activated CD71-expressing MBC-SC5.3 and resting MBC-SC5.2. BCR lineage analysis confirmed the strong link between the latter two MBCs, suggesting that activated MBCs are more dominant than atypical MBCs as key intermediates in the generation of resting MBCs. It should be noted that RNAseq-based trajectory and BCR lineage analyses were based on separate sources of information (mRNA transcription and BCR sequence similarity, respectively). Thus, while our BCR lineage analyses were based on a relatively small number of clones, the fact that they were consistent with transcriptional trajectory analysis validates these findings.

In designing the current study, our goal was to capture the full extent of vaccine-induced B cell responses. To this end, we used the vaccine-encoded protein S-2P to sort all spike-binding B cells without restriction on epitope or immunoglobulin isotype. However, by not including two S-2P probes, a strategy that can improve specificity,⁶⁰ we likely captured B cells that might not be specific for S-2P, especially among naive B cell clusters, which others have shown to be rare events.^61,62 Nonetheless, with the inclusion of oligo-tagged RBD and S1 within ADTs at the last two timepoints where there were enough cells to perform the ADT step post sort, we observed similar binding and temporal dynamics for the three spike proteins within MBC-C5. In addition, differences between S-2P⁺ and S-2P⁻ cells in BCR diversity and gene usage, clonal overlap, accumulation of SHMs, as well as intra- and inter-cohort convergence provided further evidence that the majority of S-2P⁺ cells within MBC-C5 expressed a BCR that was spike-specific.

PBs are rapidly and strongly induced following exposure to antigen and can be readily detected in circulation, albeit for a brief period.⁵¹ Several studies have shown increased frequencies of PBs early after SARS-CoV-2 infection or vaccination,^{13,24,28,37,41,46,48,53,63} although evidence of relationships between PBs and other B cell types remain limited. We previously demonstrated a strong correlation between early PBs and late antibody titers following primary two-dose mRNA-1273 vaccination,¹³ perhaps indicative of a link between circulating PBs and plasma cells in the bone marrow, which are the main source of antibodies present in blood.⁵² Such a link between PBs and plasma cells has been demonstrated for influenza-specific responses to vaccination, both through correlations in frequencies and clonal lineages,⁶⁴ the latter also demonstrated for SARS-CoV-2 mRNA vaccine responses.⁴¹ Clonal relationships between PBs and MBCs to recall antigens have also been shown,⁶⁴ with evidence from multiyear serial blood draws for PBs being derived from MBCs.⁵³ Similarly, in the current study, we found evidence of extensive relationships between SARS-CoV-2 vaccine-induced PBs and S-2P⁺ cells based on BCR repertoire analyses. Among the cells with BCRs in PB-C6 (5,928 cells) analyzed from the first three post-vaccination timepoints, 967 cells (16.3%) showed clonal overlap with S-2P⁺ cells, despite the PBs not having been sorted for antigen specificity. Several lineage trees contained mixtures of PBs and MBCs, with PBs being closest to germline, consistent with their presence at early timepoints in an evolving response. However, in contrast to recall responses where intermingling of MBCs and PBs are thought to reflect the latter being derived from the former,⁵³ this is unlikely to occur with primary vaccination. The reverse, MBCs deriving from PBs, is also unlikely based on established developmental pathways.^51,52 The PB-MBC relationships are more likely to be the result of parallel yet common germline engagement, consistent with the evidence of extensive convergence associated with B cell responses to SARS-CoV-2.^8,28,46–49

Our longitudinal analyses identified a common developmental trajectory for S-2P⁺ MBCs, which produced BCRs that were convergent both among the three vaccinees and with published SARS-CoV-2 binding antibodies. Several studies have identified convergent BCRs in response to SAR-CoV-2 exposures,^8,28,46–49 possibly due to common engagement of unmutated SARSCoV-2 binding BCRs,⁶⁵ and suggesting that these convergent BCRs may be effective targets for vaccine development.⁴⁹ Convergent BCRs among our three vaccinees were highly enriched within memory B cell clusters and followed a common developmental trajectory that led to resting MBCs enriched in IgG. Public BCRs that matched previously published antibodies in the CoV-AbDab were also highly enriched within memory B cell clusters. The source publications of matching antibodies in the CoV-AbDab revealed that some of these public BCRs were derived from IgG⁺CD27⁺ B cells,^66,67 consistent with resting MBCs. However, whether anti-SARS-CoV-2 public BCRs arise from a similar trajectory will need to be confirmed with larger cohorts and more extensive characterization.

In summary, we have shown that while the B cell response to the primary two-dose SARS-CoV-2 mRNA-1273 vaccine involves several cell types and trajectories, strong clonal relationships between PBs and MBCs and among MBCs were identified that may help provide insight into the development of durable immunity against the virus. We identified a distinct population of CD71-expressing activated MBCs as the predominant precursor of resting MBCs while CD11c-expressing atypical MBCs were not frequent sources of resting MBCs, despite being a major source of the S-2P response early after the second dose. Finally, the extensive convergence observed among S-2P⁺ cells suggests that the trajectory described in our study may represent engagement of a common differentiation pathway in the generation of B cell memory to SARS-CoV-2.

Limitations of the study

The main limitation of our study was that it was restricted to three participants and these participants had higher than average spike-specific B cell responses to the vaccine than the rest of the cohort,¹³ which could have skewed our findings. However, these were not outliers and combined with the evidence of convergence, we believe our findings are reflective of common differentiation pathways. While convergence across the three vaccinees was observed and consistent with other studies, we did not address the impact it might have on the evolution of the virus.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Susan Moir (smoir@niaid.nih.gov).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Single-cell RNA-seq data were deposited at GEO and are publicly available as of the date of publication. The accession number is listed in the key resources table. The original code generated in this study is available on https://doi.org/10.5281/zenodo.7925074. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-human CD19 PE-Cy7	ThermoFisher	Cat# 25-0198-42; RRID: AB_10671548
Anti-human CD20 APC-H7	BD Biosciences	Cat# 560734; RRID: AB_1727449
Anti-human CD20 TotalSeq-C	BioLegend	Cat# 302363; RRID:AB_2800743
Anti-human CD21 TotalSeq-C	BioLegend	Cat# 354923; RRID:AB_2800953
Anti-human CD10 TotalSeq-C	BioLegend	Cat# 312233; RRID:AB_2800817
Anti-human CD19 TotalSeq-C	BioLegend	Cat# 302265; RRID:AB_2800741
Anti-human CD38 TotalSeq-C	BioLegend	Cat# 303543; RRID:AB_2800758
Anti-human CD138 TotalSeq-C	BioLegend	Cat# 356539; RRID:AB_2810567
Anti-human CD27 TotalSeq-C	BioLegend	Cat# 302853; RRID:AB_2800747
Anti-human IgD TotalSeq-C	BioLegend	Cat# 348245; RRID:AB_2810553
Anti-human IgM TotalSeq-C	BioLegend	Cat# 314547; RRID:AB_2800835
Anti-human CD71 TotalSeq-C	BioLegend	Cat# 334125; RRID:AB_2800885
Anti-human IgG Fc TotalSeq-C	BioLegend	Cat# 410727; RRID:AB_2801087
Anti-human CD11c TotalSeq-C	BioLegend	Cat# 371521; RRID:AB_2801018
Anti-human CD95 TotalSeq-C	BioLegend	Cat# 305651; RRID:AB_2800787
Anti-human CD307e TotalSeq-C	BioLegend	Cat# 340309; RRID:AB_2819969
Anti-human Hashtag 2 TotalSeq-C	BioLegend	Cat# 394663; RRID:AB_2801032
Anti-human Hashtag 3 TotalSeq-C	BioLegend	Cat# 394665; RRID:AB_2801033
Anti-human Hashtag 4 TotalSeq-C	BioLegend	Cat# 394667; RRID:AB_2801034
Anti-human Hashtag 5 TotalSeq-C	BioLegend	Cat# 394669; RRID:AB_2801035
Anti-human Hashtag 6 TotalSeq-C	BioLegend	Cat# 394671; RRID:AB_2820042
Anti-human Hashtag 7 TotalSeq-C	BioLegend	Cat# 394673; RRID:AB_2820043
Anti-human Hashtag 8 TotalSeq-C	BioLegend	Cat# 394675; RRID:AB_2820044
Anti-human Hashtag 9 TotalSeq-C	BioLegend	Cat# 394677; RRID:AB_2820045
Anti-human Hashtag 10 TotalSeq-C	BioLegend	Cat# 394679; RRID:AB_2820046
Mouse IgG1 κ isotype TotalSeq-C	BioLegend	Cat# 400187; RRID:AB_2888921
Mouse IgG2a κ isotype TotalSeq-C	BioLegend	Cat# 400293; RRID:AB_2888922
Mouse IgG2b κ isotype TotalSeq-C	BioLegend	Cat# 400381; RRID:AB_2888923
Rat IgG2a κ isotype TotalSeq-C	BioLegend	Cat# 400577; RRID:AB_2894971
Rat IgG2b κ isotype TotalSeq-C	BioLegend	Cat# 400677; RRID:AB_2894967
Biological samples
Human peripheral blood samples from SARS-CoV-2 mRNA vaccine recipients	Collected at Clinical Center, NIH	N/A
Chemicals, peptides, and recombinant proteins
SARS-CoV-2 S-2P spike S-2P biotin This study		N/A
Streptavidin PE-Cy5.5 ThermoFisher		Cat# SA1018
SARS-CoV-2 spike RBD APC tetramer TotalSeq-C0956/C0971	BioLegend and custom conjugation	Cat# 793906
SARS-CoV-2 spike S1 PE tetramer TotalSeq-C0961/0972	BioLegend and custom conjugation	Cat# 793806
Critical commercial assays
Chromium Next GEM Single Cell 5′ Kit v2	10X Genomics	Cat# 1000263
Chromium Next GEM Chip K Single Cell Kit	10X Genomics	Cat# 1000286
5′ Feature Barcode Kit	10X Genomics	Cat# 1000256
Chromium Single Cell Human BCR Amplification Kit	10X Genomics	Cat# 1000253
Dual Index Kit TN Set A	10X Genomics	Cat# 1000250
Dual Index Kit TT Set A	10X Genomics	Cat# 1000215
Library Construction Kit	10X Genomics	Cat# 1000190
TapeStation High Sensitivity D5000 Reagents	Agilent Technologies	5067–5593
TapeStation High Sensitivity D1000 Reagents	Agilent Technologies	5067–5585
KAPA Lib Quant Kit	Roche	KK4824
NovaSeq 6000 S4 Reagent Kit (200 Cycles)	Illumina	20027466
NovaSeq 6000 SP Reagent Kit (200 Cycles)	Illumina	20040719
RNeasy micro Kit	QIAGEN	74104
NEBNext® Ultra™II DNA Library Prep Kit for Illumina	New England Biolabs	E7645
Deposited data
Raw sequencing data files for single-cell RNA sequencing	This paper	GEO accession: GSE219098
Software and algorithms
10X Cell Ranger package	10X Genomics	https://support.10xgenomics.com
R (Version 4.2.1, 4.1.0 for VDJ analysis)	The Comprehensive R Archive Network	https://cran.r-project.org/
RStudio (Version 2022.07.1 Build 554)	RStudio, Inc.	https://www.rstudio.com/
Seurat (Version 4.1.0)	Stuart etal., 201927	https://satijalab.org/seurat/index.html
Scripts for scRNA seq processing	This paper	https://doi.org/10.5281/zenodo.7925074
DittoSeq (Version 1.10.0)	Bunisetal., 202168	https://rdrr.io/github/dtm2451/dittoSeq/
Change-O (Version 1.1.0)	Gupta etal.,201569	https://changeo.readthedocs.io
Immcantation (Version 4.2.0)	Kleinstein lab	https://immcantation.readthedocs.io
Alakazam (Version 1.2.1)	Gupta etal.,201569	https://alakazam.readthedocs.io
IgBLAST (Version 1.17.1)	Ye etal., 201370	https://www.ncbi.nlm.nih.gov/igblast/
Scoper (Version 1.2.0)	Nouri etal., 201871	https://scoper.readthedocs.io
SHazaM (Version 1.1.2)	Gupta etal.,201569	https://shazam.readthedocs.io
IgPhyML (Version 1.1.4)	Hoehnetal., 201943	https://igphyml.readthedocs.io
Dowser (Version 1.1.0)	Hoehn et al., 202242	https://dowser.readthedocs.io
Ggtree (Version 3.0.4)	Yu etal., 201772	https://github.com/YuLab-SMU/ggtree
Phangorn (Version 2.10.0)	Schliep, 201173	https://github.com/KlausVigo/phangorn
Monocle3 (Version 1.0.0)	Cacchiarelli, 202240	https://cole-trapnell-lab.github.io/monocle3
Demuxlet	Kang et al., 201874	https://github.com/statgen/demuxlet
GraphPad Prism (Version 9.3.1)	GraphPad	RRID:SCR_002798
FlowJo (Version 10.8.1)	FlowJo	RRID:SCR_008520

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Ethics statement

Phlebotomy was performed at the NIH Clinical Research Center in Bethesda, MD under protocol NCT00001281. The protocol was approved by the NIH Institutional Review Board and written informed consent was obtained from all study participants.

Human participants and study design

Longitudinal single-cell analyses were performed on peripheral blood mononuclear cells (PBMCs) collected and cryopreserved from three SARS-CoV-2-uninfected adults who previously participated in a study evaluating B cell and antibody responses to the twodose mRNA-1273 vaccine.¹³ Timepoints and participants in the current study were selected based on availability of cells and frequencies of target cells calculated from data collected in the previous study.¹³ Details of demographics, cell frequencies and exact visit timepoints are provided in Figure 1A and Table S1.

METHOD DETAILS

Recombinant COVID S-2P protein

An uncleaved version of the ectodomain (residues 1–1208) of SARS-CoV-2 spike protein, namely SARS-CoV-2 S-2P was constructed as sorting probe that contains the following modifications - D614G mutaiotn, ₆₈₂RRAR₆₈₅ →SGAG substitution at the furin cleavage site, and two proline substitutions ₉₈₆ KV₉₈₇→PP. Additionally in the S-2P construct, the C terminus of the S ectodomain is appended with a GSG peptide linker, a foldon trimerization motif, an HRV3C protease cleavage site, an 8 × His Tag, and a TwinStrep Tag (SAWSHPQFEKGGGSGGGSGGSAWSHPQFEK) as previously described.^75,76 The S-2P protein encoding sequence were codon optimized for human cell expression (GenScript, NJ), cloned into the mammalian cell expression vector pcDNA3.1(−), and confirmed by sequencing prior to transient transfection in FreeStyle 293-F cells with 293fectin transfection reagent (Thermo Fisher Scientific). Culture supernatants were harvested at 5 days post transfection, filtered, and purified by StrepTactin resin (IBA Lifesciences). Elutes were subjected to size exclusion chromatography using a Superpose 6 10/300 GL column (Sigma-Aldrich, MO) to collect trimer fraction. Elutes from both purification procedures were concentrated and buffer exchanged with phosphate-buffered saline (PBS) with an Amicon Ultra 10 kDa molecular weight cutoff concentrator (Millipore).

Sample processing, cell sorting, and CITE-seq

B cells were enriched from thawed cryopreserved PBMCs by negative magnetic bead-based selection (StemCell Technologies). An approximately equal number of S-2P⁺ B cells from each of the three participants was pooled and stained with PE-Cy7-conjugated anti-human CD19 (ThermoFisher, cat# 25–0198-42) and APC-H7-conjugated anti-human CD20 (BD Biosciences, cat# 560734). SARS-CoV-2-binding B cells were identified using biotinylated spike protein (S-2P) tetramerized with PE-Cy5.5-labeled streptavidin (SA) (ThermoFisher, cat# SA1018) at a molar ratio of 4:1, as previously described.¹³ CD20⁻ PBs and CD20⁺ S-2P⁺ and S-2P⁻ cells were sorted using a BD FACSAria Fusion instrument (BD Biosciences) (Figure 1B). CITE-seq and cell hashing were performed as previously described,^76–78 with the following details. Sorted cells from v2D28 and M6 were Fc blocked (Human TruStain FcX), mixed with a timepoint-specific oligo-tagged hashtag antibodies (HTOs), in addition to TotalSeq-C ADTs, and RBD and S1 oligo-tagged tetramers (all from BioLegend, Table S2). For v1D14 through v2D14 where the number of target cells was lower than at later timepoints, cell loss was minimized by adding HTOs and ADTs before cell sorting. RBD and S1 tetramers were excluded at these timepoints due to interference with S-2P during the sorting step.

Single cell RNA sequencing

Sorted cells were resuspended in PBS with 0.04% BSA, counted, mixed with a reverse transcription (RT) mix, and partitioned into single cell Gel-Bead in Emulsion (GEM) using the 10x Chromium Single Cell Immune Profiling Next GEM v2 Chemistry Kit (10x Genomics, Pleasanton, CA). The RT step was conducted using a C1000 touch Thermo Cycler (BioRad, Hercules, CA). Single-cell gene expression (GEX), cell surface protein (ADT) and B cell receptor (BCR) libraries were prepared following the manufacturer’s guidelines.⁷⁹ A total of 84 libraries were quality control tested using a TapeStation 4200 (Agilent, Santa Clara, CA) and quantified using the KAPA Library Quantification kit (Roche, Wilmington, MA). The libraries were pooled and sequenced on the Illumina NovaSeq platform (Illumina, San Diego, CA) using the sequencing parameters recommended by the 10X Genomics Single Cell Immune Profiling kit v.2 user guide. Minimum sequencing depths were as follows: 50,000 read pairs/cell for the GEX library, 5,000 read pairs/cell for the BCR library, and 5,000 read pairs/cell for the ADT library. Sequencing saturation of the libraries was 80% for the GEX and 10% for the BCR and ADT libraries.

Bulk RNA sequencing and sample demultiplexing

Cells from each participant were demultiplexed in silico using SNPs, which were determined by bulk sequencing 25,000–50,000 PBMCs from each participant. Total RNA was extracted from samples stored in Trizol using the miRNAeasy micro kit (QIAGEN, Germantown, MD). Standard RNA sequencing libraries were generated using NEBNextII DNA Library prep kit (#E745). Libraries were then sequenced on Illumina NovaSeq6000 with 100bp paired-end reads, yielding 420M total reads. Sequencing data were converted to a FASTQ format using the Illumina bcl2fastq software. The sequence reads were adaptor and quality trimmed and then aligned to the human genome using the splice-aware STAR aligner, and SNP calls were generated using a previously published protocol.⁸⁰ We used the software package demuxlet⁷⁴ to then match single cells from the 10x RNA-seq data to each participant and identify doublets, of which were removed. Because multiple samples from different time points for each participant may be collected and could not be demultiplexed by this method alone, we also used ‘hashtag’ antibodies (HTOs) (BioLegend) to uniquely label the different time points.⁷⁸

Single cell data processing

The GEX and ADT libraries were mapped to the GRCh38 human reference genome using CellRanger (10X Genomics) version 6.0.0. Data were further processed using Seurat version 4.1.0²⁶ running in R v4.1.0. After filtering the single cells based on the demuxlet output, we further demultiplexed the samples from different timepoints using Seurat’s HTOdemux function. Cells were discarded if they met one of the following criteria: 1) there were fewer than 200 or greater than 4,000 detected genes; 2) the percentage of mitochondrial gene counts was >10%; or, 3) the read counts for ADT and HTO were >30,000 and 15,000, respectively. Additionally, genes expressed in fewer than 5 cells were discarded. Gene expression was normalized (scale factor = 10000) and log transformed (log1p) before analysis.

All pre-processed data were integrated into a Seurat object using the regressed scaled counts of the top 2000 highly variable features across the datasets. Ig genes were removed from the list of variable genes given their abundance in certain cell types and influence on clustering. Further dimensional reduction was carried out via Principal Component Analysis (PCA) and UMAP. For clustering, the k.param nearest neighbors was computed and seven clusters were identified (resolution = 0.5) using a shared nearest neighbor (SNN) modularity optimization based clustering algorithm implemented in Seurat.²⁶ Additionally, CITE-seq protein data were normalized using centered log ratio (CLR) transformation across cells, followed by scaling and dimensional reduction (PCA) using Seurat.

Differential gene expression analysis

Differentially expressed genes between different clusters were identified using the FindAllMarkers function from Seurat (Wilcoxon test and Bonferroni p value correction). Significant genes with average log fold changes >0.5 and expression in at least 50% of cells in a cluster were ranked according to fold change and represented using a Feature Plot (Seurat) or heatmap (dittoSeq R package).⁶⁸

Trajectory analysis

Trajectory inference and pseudotime calculations were performed with Monocle3⁸¹ on the Seurat object. First, we converted the Seurat integrated object to a Monocle3 object using the SeuratWrappers pipeline. Next, we fitted a principal graph within each partition, and then the cells were ordered according to their progress through the developmental program, measured here as pseudotime. For the analysis shown in Figure 3F, MBC-SC5.1 was selected as the root node of the trajectory based on temporal distribution of S-2P⁺ cells (Figure 3D).

B cell receptor processing and analysis

B cell receptor (BCR) V(D)J data processing and analysis were carried out using tools in the Immcantation framework (www.immcantation.org) in R 4.1.0.⁸² V(D)J genes from the CellRanger output were re-assigned by alignment to the IMGT GENE-DB⁸³ reference database obtained 2/22/21 using IgBLAST v1.17.1⁷⁰ and Change-O v1.1.0.⁸⁴ To maximize the number of available sequences, BCRs from cells that were assignable to a single sample but were filtered out during gene expression analyses were included, but with a missing cell type annotation. Following V(D)J gene annotation, cells with non-functional sequences or multiple heavy chain sequences were removed.

Functional V(D)J sequences were grouped into clones using scoper v.1.2.0.⁷¹ Sequences within each participant were first partitioned based on common IGHV gene annotations, IGHJ gene annotations, and CDR3 lengths. Within these groups, sequences differing from one another by a participant-specific normalized Hamming distance threshold within the CDR3 were defined as clones by single-linkage clustering.⁸⁴ This threshold was determined by fitting a gamma/Gaussian mixture model to the distance to nearest sequence neighbor distribution using SHazaM v.1.1.2.⁶⁹ This performs a maximum-likelihood fitting procedure for learning the parameters of a gamma and Gaussian distributions which fit the bimodal distributions representing within and between clone comparisons, respectively. A threshold was selected to achieve an estimated specificity of 0.995. These thresholds were 0.085, 0.156, and 0.07 for VAC003, VAC611, and VAC716, respectively. Between-participant sequence comparisons were used to initialize gamma/ Gaussian parameters. These heavy chain-defined clonal clusters were further split if their constituent cells contained light chains that differed by V gene, J gene, or CDR3 length.⁸⁵ Consensus clonal germline V and J sequences were then reconstructed for each clone with D segment and N/P regions replaced with “N” nucleotides using dowser v1.1.0.⁴² SHM was calculated as the frequency of non-ambiguous mismatches from each cell to the V gene (IMGT positions 1–312) of its reconstructed germline sequence.

Clonal diversity is an important metric of B cell repertoires, and low B cell clonal diversity is consistent with clonal expansion. To quantify B cell clonal diversity, we calculated the inverse Simpson’s diversity (diversity order q = 2) for S-2P⁺ and S-2P⁻ clones within each sample using alakazam v1.2.1.^69,86 Lower values of clonal diversity indicate a greater probability of two random sequences belonging to the same clone. To account for differences in sequence depth, samples within each comparison were downsampled to the same number of sequences, and the mean of 500 such re-sampling repetitions was reported. Only participant/sort/time point samples with at least 50 B cells were included. Clonal overlap among tissues can also be used as a measure of immunological connectivity. Clonal overlap was calculated using the Jaccard index, which for each pair of either timepoints or cell annotations is the number of unique clones found in two groups (intersect) divided by the total number of unique clones among the two groups (union). To account for differences in naive B cell proportions between S-2P⁺ and S-2P⁻ sorts, only non-naïve B cells were included in clonal diversity and overlap analyses. Further, to make diversity and overlap values more comparable between S-2P⁺ and S-2P⁻ clones, only S-2P⁺ and S-2P⁻ sorted (not plasmablast-sorted) cells were included in these analyses.

Convergent antibody identification

To identify putative SARS-CoV-2-specific antibodies shared by multiple participants, we first grouped together cells using single linkage hierarchical clustering with the heavy chain V/J genes, CDR3 lengths, and at least 80% CDR3 amino acid similarly. These heavy chain-defined sequence clusters were further split if their constituent cells contained light chains that differed by V and J genes. Sequence groups containing cells obtained from multiple participants were considered convergent sequence clusters. To identify BCR sequences similar to previously published antibodies, we compared each BCR heavy chain sequences to all human heavy chains deposited in the CoV-AbDab database (obtained 10/11/22).⁵⁰ We classified a heavy chain as “public” if it has the same V gene, J gene, CDR3 length, and at least 80% CDR3 AA similarity to a heavy chain in the CoV-AbDab.

B cell lineage tree analysis

To infer lineage trees, we estimated tree topologies, branch lengths, and participant-wide substitution model parameters using maximum likelihood under the GY94 model.⁸⁷ Using fixed tree topologies estimated from the GY94 model, we then estimated branch lengths and participant-wide parameter values under the HLP19 model in IgPhyML v1.1.4.⁴³ Trees were visualized using dowser v1.1.0⁴² and ggtree v3.0.4.⁷² Six BCRs contained premature stop codons due to insertions relative to their V gene germline sequence and were removed. We tested for measurable evolution in each lineage tree using the date randomization test implemented in dowser.⁴⁴ This test calculates the observed correlation between divergent (sum of branch lengths to the most recent common ancestor) and sample time for each tip and compares it to a null distribution from randomized time point at the tips.⁴¹ Due to the small number of available sequences, timepoints were randomized uniformly among tips rather than among single-timepoint monophyletic clades.

The “switch proportion” (SP) permutation test⁴² was used to understand the phylogenetic relationship among cell type annotations. Briefly, given the set of annotations in a tree, a maximum parsimony algorithm was used to identify the set of internal node tissue labels resulting in the fewest number of tissue changes along the tree. For all S-2P⁺ trees containing cells from at least two distinct cluster/subcluster within a participant, the number and direction of tissue changes along all trees was recorded and normalized by the total number of switches to give the switch proportion statistic (SP). Annotations were then randomized among trees within a participant, and the resulting SP statistic was calculated for these permuted trees. Annotations were permuted among trees to account for the small number of cells in individual trees. The difference between observed and permuted SP statistics (δ) was recorded, and this process was repeated for 1000 replicates. The p value for enrichment of changes between annotations (i.e., δ > 0) is the proportion of replicates in which δ ≤ 0. If p < 0.05 and δ > 0 for a given pair of annotations, this indicates significantly more connection from one annotation to the other than expected by chance in the lineages surveyed. To account for uncertainty in tree topology, sequence alignment columns were re-sampled with replacement, and tree branch length and topologies were re-estimated for each replicate. For computational expediency, the pratchet function in the phangorn v2.10.0⁷³ was used to estimate tree topologies and branch lengths. To control the false positive rate of the SP test, all lineages were down-sampled to a maximum tip to statechange ratio of 20 for each repetition. The SP test was performed twice: once considering all switches leading to resting MBCs (MBC-SC5.2), and again considering only switches leading to resting memory cells from v2M6 from earlier timepoints.

QUANTIFICATION AND STATISTICAL ANALYSIS

For single-cell and B cell repertoire analysis, statistics are described in the corresponding section of method details and figure legends. The chi-squared test, performed with GraphPad Prism 9 software, was used to evaluate the enrichment of S-2P⁺ versus S-2P⁻ cells within C5 at different timepoints. A p value of <0.05 was considered statistically significant.

Highlights.

• Clonally related spike-specific plasmablasts and memory B cells

• Durable resting memory B cells arise from CD71⁺ activated B cells

• Incremental accumulation of BCR somatic mutations

• Evidence of convergence and common differentiation pathways

INCLUSION AND DIVERSITY

One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. One or more of the authors of this paper self-identifies as a member of the LGBTQ+ community. One or more of the authors of this paper self-identifies as a gender minority in their field of research.