Atypical and non-classical CD45RB^lo memory B cells are the majority of circulating SARS-CoV-2 specific B cells following mRNA vaccination or COVID-19

Abstract

Resting memory B cells can be divided into classical or atypical groups, but the heterogenous marker expression on activated memory B cells makes similar classification difficult. Here, by longitudinal analysis of mass cytometry and CITE-seq data from cohorts with COVID-19, bacterial sepsis, or BNT162b2 mRNA vaccine, we observe that resting B cell memory consist of classical CD45RB⁺ memory and CD45RB^lo memory, of which the latter contains of two distinct groups of CD11c⁺ atypical and CD23⁺ non-classical memory cells. CD45RB levels remain stable in these cells after activation, thereby enabling the tracking of activated B cells and plasmablasts derived from either CD45RB⁺ or CD45RB^lo memory B cells. Moreover, in both COVID-19 patients and mRNA vaccination, CD45RB^lo B cells formed the majority of SARS-CoV2 specific memory B cells and correlated with serum antibodies, while CD45RB⁺ memory are activated by bacterial sepsis. Our results thus identify that stably expressed CD45RB levels can be exploited to trace resting memory B cells and their activated progeny, and suggest that atypical and non-classical CD45RB^lo memory B cells contribute to SARS-CoV-2 infection and vaccination.

Introduction

Humoral immune memory against SARS-CoV2 forms a critical barrier preventing or limiting the severity of later infections¹. Humoral immunity relies on a complex interplay of CD4 T cells such as T-follicular helper cells (Tfh) and antibody-producing B cells and their memory populations².

Circulating memory B cells in human blood can be broadly divided into two main fractions. The first, classical memory, expresses CD27 and CD21 but lacks expression of CD11c while atypical memory cells, also referred to as DN2 or age-associated B cells (ABC), are typified by the opposite pattern of CD11c positivity but lacking CD27 and CD21^3,4,5,6. During acute infection or following recent vaccination, a population of activated memory B cells has been recognized by their expression of a variety of markers, most notably CD71^7,8 and lack of CD21 expression^9,10. Since these cells express markers associated with both classical (CD27) and atypical (CD11c, Tbet and FCRL5)^9,11 B cells, determining if they are derived from resting classical or atypical memory B cells has been difficult.

There is a lack of consensus on the best markers and nomenclature to define B cell memory^3,12. Several recent projects using large antibody panels to screen for markers of memory B cells identified a sialylated epitope of CD45RB, recognized by the MEM55 clone, as a key marker of circulating human classical memory B cells^13,14. CD45RB^MEM55 positivity was also described to precede CD27 expression in the classical branch of memory B cells, characterizing an early form of CD45RB⁺CD27^- memory^13,15. CD45RB is not expressed by either naïve or germinal center B cells but was found to be a marker of the transition from a germinal center (GC) to early memory and plasmablast phenotypes and thus has been considered a marker of GC derived memory^16,17.

Since the GC is the primary site of efficient affinity maturation, GC-derived memory cells play a critical role in humoral immunity to COVID-19 and related vaccination^18,19. However, there is also an increasing appreciation of the role of CD11c and Tbet expressing atypical B cells in a range of settings such as viral infections, including COVID-19, and autoimmunity^{8,20,21,22,23,24,25,26}.

Both the classical and atypical routes of antibody production are largely reliant on T cell help and T cell memory is another critical component of protection from viral infections such as COVID-19^1,27. Circulating Tfh (cTfh), have been identified to be strongly predictive of ongoing germinal center responses²⁸. Recently PD1^hiCXCR5^- T-peripheral helper (Tph) cells, have been identified as a major helper of B cell responses in extrafollicular regions^29,30. Tph were originally described in the context of autoimmune diseases such as RA and SLE^30,31 but have also recently been shown by us and others to be present in high numbers during COVID-19 and correlate with atypical B cells^31,32,33

In this study, we find that blood circulating memory B cells can be divided into branches of CD45RB⁺ positive or CD45RB^lo memory cells and that stable differences in CD45RB sialyation allows tracking of activated B cells and plasmablasts derived from these two branches. The majority of blood circulating memory B cells recognizing SARS-CoV2 following infection or mRNA vaccination are contained within distinct groups of CD23⁺ CD45RB^lo and atypical CD11c⁺ CD45RB^lo B cells. These results suggest that several groups of CD45RB^lo non-classical lineage B cells have a key role in the response to both COVID-19 and mRNA vaccination.

Results

Study design and mass cytometry analysis

To perform an in-depth comparison of different disease state and vaccine responses we examined a large longitudinal cohort of severe COVID-19 and bacterial sepsis patients alongside a healthcare worker cohort of Pfizer mRNA vaccine recipients (Fig. 1A, Supplementary Table 1). Samples were processed in large batches using sample barcoding and a split panel design for analysis of T cells, non-T cells and total lineage proportions in CD3+ magnetic bead enriched, CD3- depleted, or unenriched cells respectively (Fig. 1A, Supplementary Table 2). After quality control, removal of batch controls and debarcoding (Supplementary Fig. 1A, B) this gave a total of 632 unique samples from 218 donors split into 26.8, 34.7, and 7.3 million cells in the T cell, non-T cell and lineage datasets respectively (Fig. 1A). All samples were analyzed in a single combined workflow allowing in-depth comparison of cellular states across our separate cohorts^32,34,35.

Fig. 1: Mass cytometry analysis of B cells reveals distinct groups of activated and resting memory B cell populations.

A Overview of CyTOF study design. B ForceAtlas2 visualization of B cell FlowSOM clusters from merged analysis of dataset. Pre-gating shown in Supplementary Fig. 1A. C Expression of indicated protein markers on ForceAtlas2 map. D Heatmap of median scaled expression of type markers across merged B cell FlowSOM clusters and donut plots of proportions of isotypes for each B cell cluster. Derived from a total of 9.81 million B cells, with 632 samples from 218 individuals. Boxplots of selected markers shown in Supplementary Fig. 2A. A Created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

Analysis of total lineage populations in our ICU cohort showed that B cells (including plasmablasts) were notably increased at the earliest timepoints in COVID-19 but not Sepsis (Supplementary Fig. 1C–E). We then performed detailed analysis of B cells and annotated the clusters based on our previous work and recent studies using large screening approaches for B cell population analysis^13,14,32. IgM, IgG, and IgA were measured but excluded from clustering to avoid excessive fractionation of memory populations.

For visualization we used a force directed approach (ForceAtlas2) that preserves global structure (relationships between clusters) and was able to lay out cells in a biologically interpretable order (Fig. 1B)^36,37. Transitional B cells were recognized as CD38⁺CD24⁺IgD⁺CD27^- and were adjacent to CD38^loCD24^loIgD⁺CD27^- Naïve B cells (Fig. 1B–D). The CD45RB⁺CD27^- early memory population^13,15 was visible as a branch emerging from naïve B cells and terminating in CD45RB⁺CD27⁺ classical memory cells. A second branch of CD45RB^-CD11c^-CD21⁺IgD^- class-switched cells that we identified as having a similar progression from CD27 negative to positive was annotated as CD45RB^-CD27^- and CD45RB^-CD27⁺ Mem respectively. CD11c⁺CD21^loCXCR5^-CD27^-IgD⁺ Atypical (Atypical) naïve B cells³⁸ were visible and adjacent to class-switched CD11c⁺CD21^loCXCR5^-CD27^-IgD^- atypical (Atypical) memory cells.

Previous reports have demonstrated that activated memory B cells can be defined by expression of CD71, CD27, CD11c and lack of CD21⁷. Here we found that CD71⁺CD27⁺CD11c⁺Ki67⁺CD21^lo activated B cells could be clearly divided into two groups using CD45RB, whose expression has been reported as restricted to classical memory B cells^13,14,15. Accordingly, we annotated these cells as CD45RB⁺ activated memory (CD45RB⁺ Activated) and CD45RB^- atypical Activated memory (CD45RB^- Atypical Activated). Comparison of these two populations revealed that CD45RB^- Atypical Activated has significantly higher expression of CD11c and Ki67 while CD45RB⁺ Activated has higher expression of CD21, CXCR5, CD27, CD95 (Fig. 1C, Supplementary Fig. 2A). However, while these differences were statistically significant in this large dataset, they were also subtle, leaving CD45RB as the only non-overlapping marker allowing reliable separation of these cells. The activated phenotype of CD45RB⁺ Activated was confirmed by direct comparison with CD45RB⁺CD27⁺ memory by its significantly lower expression of CD21, CXCR5, CXCR4 and CD24 alongside significantly increased CD11c, CD39, CD71, CD38, CD95, Ki67 and CD19 (Supplementary Fig. 2A). Analysis of Isotypes also showed that Atypical memory cells had a high level of IgG while CD45RB⁺ classical memory groups had a mixture of IgA and IgG (Fig. 1D).

CD45RB^lo activated B cells respond to COVID-19 and vaccination while CD45RB^hi activated B cells expand during bacterial Sepsis

We then examined changes to the proportions of B cells in our longitudinal ICU cohort (Fig. 2A, Supplementary Table 1). Most B cells in age matched healthy donors were found in naïve and classical memory (CD27⁺CD45RB⁺) groups with atypical cells, activated B cells and plasmablasts comparatively rare (Supplementary Fig. 2B). During COVID-19 and Sepsis, CD45RB^- Atypical Activated B cells were significantly expanded at all measured time points in comparison to healthy controls (Fig. 2B). In contrast CD45RB⁺ Activated B cells were significantly expanded in Sepsis in comparison to either COVID-19 or healthy controls suggesting that this population is responsive to bacterial, but not viral infection (Fig. 2B). Both CD45RB^lo and CD45RB^hi plasmablasts were expanded in both COVID-19 and Sepsis with an initial skew to COVID-19 at days 1–4 after ICU admission but greater proportions of these cells in Sepsis from day 5 onwards (Fig. 2B). A broad shift in loss of CD45RB^hi memory cells but gain of CD45RB^lo cells was observed in non-activated memory groups with both CD45RB⁺CD27⁺ and CD45RB⁺CD27^- cells reduced in COVID-19 and Sepsis while the CD45RB^-CD27^- group being expanded most strongly in COVID-19 (Supplementary Fig. 2B). Naïve cells were unaffected while transitional B cells suffered an increasing loss over time (Supplementary Fig. 2B). We also examined a cohort of moderate COVID-19 patients and found that these trends were broadly similar with CD45RB^- Atypical Activated B cells increased while CD45RB⁺ Activated B cells were not (Supplementary Fig. 2C).

Fig. 2: CD45RB^lo activated B cells respond to COVID-19 and vaccination but CD45RB^hi activated B cells expand during bacterial sepsis.

A Schematic of COVID-19 and Sepsis cohort. B Proportions of B cell clusters in COVID-19 and Sepsis ICU patients across time compared to age-matched healthy controls (n = indicated in 2 A). Other B cell populations shown in Supplementary Fig. 2B. C Schematic of SARS-CoV2 mRNA vaccination cohort. Created with Biorender.com. D Proportions of B cell clusters in SARS-CoV2 mRNA vaccination cohort (n = indicated in C). Significance testing in comparison to with pre-first except when indicated. Multiplicity adjusted p-values from EdgeR (two sided). Single symbols P ≤ 0.05, double P ≤ 0.01, triple P ≤ 0.001. Box plots are median +/- interquartile range (IQR), whiskers are 1.5*IQR. A–C Created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

We then examined a longitudinal cohort of healthcare worker BNT16b2 mRNA vaccine recipients (Fig. 2C). Analysis of B cells during mRNA vaccination revealed that by day 12–21 post-primary vaccination a significant expansion of CD45RB^lo but not CD45RB^hi plasmablasts was observed (Fig. 2D). Following secondary vaccination this increase of CD45RB^lo plasmablasts was sustained. At this time point a significant increase in CD45RB^- Atypical Activated was also observed while CD45RB⁺ Activated had no clear response. By 3 months after the secondary vaccination, these increases in activated cells had resolved. However, following tertiary vaccination CD45RB^- Atypical Activated B cells showed a significant expansion while no other B cell subset responded (Fig. 2D) suggesting that although this population is rare, it is highly responsive to mRNA vaccination.

Overall, we saw significant increases in CD45RB^- activated atypical B cells following both SARS-CoV2 infection and vaccination and that CD45RB^lo subsets of memory, atypical memory and plasmablasts seemed to have a closer relationship to COVID-19 and vaccination while CD45RB⁺ Activated memory cells were only expanded during bacterial Sepsis.

Tph cells are predominant during early COVID-19, but Tph and Tfh responses are more balanced during vaccination

Using the same analysis approach as for B cells we examined the CD4 T cell compartment in the same samples (Fig. 3A). We defined Tph as CD45RA^-CD45RO⁺ PD1⁺ICOS⁺HLA-DR⁺CD38⁺Tbet^loCXCR5^- (Supplementary Fig. 3A) and further divided them based on expression of the proliferation marker Ki67 into Tph and Ki67⁺Tph. Both Tph and Ki67⁺Tph were expanded in the blood of COVID-19 and Sepsis patients over time (Fig. 3B). The Ki67⁺Tph subgroup was most associated with COVID-19 while bacterial Sepsis had a significantly higher proportion of the less activated Tph. Proliferating effector memory (Ki67⁺ EM) Th1 cells were also increased in both settings, again to a significantly higher degree in COVID-19 than Sepsis (Fig. 3B).

Fig. 3: Tph cells are predominant during COVID-19, but Tph and Tfh1 responses are balanced during vaccination.

A UMAP and Heatmap of markers of CD4 T cell FlowSOM clusters. Boxplots of selected markers shown in Supplementary Fig. 3A. B, C Proportions of CD4 (B) and Tfh (C) cell clusters in COVID-19 and Sepsis ICU patients across time compared to age-matched healthy controls (n = indicated in Fig. 2A). Other CD4 populations seen in Supplementary Fig. 3B. Further analysis of Tfh is shown in Supplementary Fig. 3C, D. * COVID-19 vs healthy, †Sepsis vs healthy, ‡COVID-19 vs Sepsis. D, E Proportions of CD4 (D) and Tfh (E) clusters in SARS-CoV2 mRNA vaccination cohort (n = indicated in Fig. 2C). Significance indicate comparison with the pre-first timepoint. Multiplicity Adjusted p-values from EdgeR (two-sided). Single symbols P ≤ 0.05, double P ≤ 0.01, triple P ≤ 0.001. Box plots are median +/- interquartile range (IQR), whiskers are 1.5*IQR.

We saw that overall circulating Tfh (cTfh) were slightly but significantly reduced in the blood of COVID-19 patients (Fig. 3B). Since cTfh themselves are known to harbor a range of subpopulations we clustered these cells to fully dissect Tfh subgroups. We split cTfh into the large central memory (CM)-like group of PD1 negative cells and subgroups based on expression of CCR4, CCR6, CD161, CXCR3 and Tbet^39,40. PD1, ICOS, and Tox were then used to define more activated cells and Ki67 used to define the most highly activated proliferating groups (Supplementary Fig. 3C).

PD1⁺CXCR3⁺Tbet⁺ Tfh1 were reduced as a proportion of Tfh while their rarer proliferating subgroup Tfh1 Ki67⁺PD1⁺ was increased, with the overall result that Tfh1 were reduced but those that remained were highly activated (Fig. 3C). CCR6⁺CD161⁺ Tfh17 were significantly increased in Sepsis but not COVID-19 while CCR6^-CXCR3^- Tfh2 were increased over time in both COVID-19 and Sepsis (Supplementary Fig. 3D). Taken together, there was a slight decrease in cTfh, but fine analysis of subgroups revealed a more complex picture with shifts within Tfh towards Tfh17 in Sepsis and Tfh1 being proportionally reduced but highly activated (Fig. 3C, Supplementary Fig. 3D).

During vaccination, analysis of CD4 T cells revealed a significant expansion of Ki67⁺ Tph at the same time points (post-primary and post-secondary) where we observed expansion of CD45RB^lo B cells (Fig. 3D). The post-secondary time point also saw expansion of proliferating Th1 similar to that seen in the initial trials of BNT162b1⁴¹. In contrast to COVID-19, overall cTfh were slightly increased during vaccination (Fig. 3D). Close inspection of Tfh sub-phenotypes revealed a significant expansion of proliferating Tfh1 (Fig. 3E).

Correlation analysis of the CD4 and B cell subsets from the time points with maximum activation of COVID-19 and Sepsis (both ICU Day 1–2) or mRNA vaccination (Post 2^nd) indicated that in COVID-19 and mRNA vaccination CD45RB^- Atypical Activated B cells had a significant correlation with Tph, Th1 and Tfh1. In contrast CD45RB⁺ Activated B cells were only correlated with these T cells in Sepsis (Fig. 4). Correlations derived from all times points also highlighted the strong relationship between CD45RB^- Atypical Activated B cells, Tph and Tfh (Supplementary Fig. 4A). We also noted that there was not a clear relationship between the identified B cell subsets and later mortality, pneumonia or non-pneumonia sepsis or the Gram classification of causative bacteria in Sepsis (Supplementary Fig. 4B–E).

Fig. 4: Correlations between B and T cell subsets.

Spearman rank correlation matrix of indicated CD4 and B cell subsets during day 1–2 of COVID-19 or Sepsis or Post 2^nd Vaccination timepoint. (n = 85, 34 or 28 respectively). Significance *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001 by two-sided Spearman’s rank correlation.

Overall, expansion or activation of Tph and Tfh1 was not an exclusive property of either infection or vaccination, but vaccination shows a balanced response of Tph and Tfh1 while COVID-19 is more strongly skewed to a Tph response. Tph and Tfh1 are also expanded in Sepsis but with greater Th17 and Tfh17 responses than that seen in either COVID-19 or vaccination.

Monocyte interferon signature seen in COVID-19 and vaccination but not bacterial sepsis

Previously, atypical B cell activation has been associated with both Tph, Th1 and Tfh1 which are themselves induced by early type 1 interferon and its effects on monocytes and dendritic cells (DC)⁴². Analysis of classical monocytes in our dataset (Supplementary Fig. 5A) demonstrated that at day 1 following ICU admission over 90% of monocytes express the interferon signature protein CD169 (Siglec-1) which rapidly drops to baseline levels within 5 days (Supplementary Fig. 5B). No clear increase in CD169 expressing monocytes was seen in Sepsis despite these cells also being driven into a characteristic HLA-DR^lo state shared by COVID-19 and Sepsis^43,44, supporting previous results that CD169 is an effective separator of viral and bacterial inflammation⁴⁵.

During vaccination, both CD169 expressing HLA-DR^hi and HLA-DR^lo monocytes showed a similar increase 2 days after vaccination, preceding changes to CD4 and B cells (Supplementary Fig. 5C). This response, although much lower in magnitude than COVID-19, suggests an early role for interferon signaling in monocytes in driving the similarity in CD4 and B cell responses.

CITE-seq analysis of B cell memory confirms phenotype of CD45RB⁺ and CD45RB^– Activated memory groups

Having established the broad changes to the immune system in these large cohorts, and the importance of CD45RB as a marker, which is not recoverable by mRNA, we then performed CITE-seq⁴⁶ to further explore B cell populations of interest (Fig. 5A). We first hashtagged PBMCs from 3 sepsis, 3 COVID-19, 3 post-secondary and 3 post-tertiary vaccine timepoints. Cells were stained with fluorescent and TotalSeqC antibodies and double stained with SARS-CoV2 spike tetramer and then IgD^- B cells were sorted and sequenced via the chromium 10x platform to obtain information on antigen specificity, surface protein, mRNA and BCR sequences.

Fig. 5: CITE-seq analysis of memory B cells and plasmablasts in COVID-19, Sepsis, and vaccination.

A Schematic of CITE-seq experiment. FACS sorting strategy in Supplementary Fig. 6. B Weighted nearest neighbor (wnn) UMAP of IgD^- memory B cells in CITE-seq colored according to SLM clusters. Dot plot of expression of selected differentially expressed mRNA. C Single cell Violin plots of Antibody Derived Tag (adt) protein markers. Selected other mRNA and protein markers shown in Supplementary Fig. 7A and (E). D Expression of indicated markers displayed on wnnUMAP. Other markers are shown in Supplementary Fig. 7B. E Volcano plots of differentially expressed genes between indicated populations. Other comparisons shown in Supplementary Fig. 7D. Figure 5A Created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

Clustering yielded a similar population structure that we had seen in the IgD^- compartment of the CyTOF dataset consisting of CD45RB^- Atypical memory, Atypical Activated B cells (that we further divided into Atypical early Activated and Atypical Activated), a CD45RB^lo memory group, and CD45RB⁺ Activated and CD45RB⁺ memory (Fig. 5B). We could not clearly separate CD27^+/- compartments of CD45RB⁺ and CD45RB^lo memory on either mRNA or CD27 staining suggesting these are transcriptionally similar groups. As before, the Activated memory groups co-express CD27, CD71 (TFRC) and CD11c (ITGAX) at both protein and RNA levels alongside activation markers such as FAS, S100A10 and ANXA4 (Fig. 5B, C, D, Supplementary Fig. 7A, B, D). Separate groups of CD45RB^lo and CD45RB^hi plasmablasts were also resolved. We also confirmed the identification of the Atypical (Atypical) groups through their expression of characteristic markers such as TBX21 (Tbet), ZEB2 and FCRL5 (Fig. 5B, E).

CD45RB^lo resting memory cells are characterized by expression of CD23 and IL4R

In addition to further analysis of atypical cells we found that the CD11c^-CD45RB^lo memory cells had a broadly resting phenotype (lack of activation markers) shared with CD45RB⁺ Memory cells, but that differential gene analysis showed enriched expression of FCER2 (CD23), IL4R, STAG3, IL13RA1 and IGHE (Fig. 5D, E, Supplementary Fig. 7B, 7C, 7D). Several groups have recently described a similar signature in both CD27^-IgD^-47 or CD27⁺IgD^- memory B cells in the context of allergy^48,49. All three studies and our own data display a striking similarity of phenotype between these cells.

CD45RB^lo phenotypes form the majority of SARS-CoV2 reactive B cells during early COVID-19 or post-secondary and tertiary vaccination

We then analyzed antigen specificity in the CITE-seq data by identification of memory B cells that dual bound to the spike tetramers used (Fig. 6A, B). While CD45RB⁺ B cells made up the majority of SARS-CoV2 spike negative memory B cells, we saw that in both COVID-19 and vaccination, SARS-CoV2 spike positive cells were strongly concentrated into the CD45RB^lo clusters, particularly the Atypical early Activated and Atypical Activated groups (Fig. 6C, D, Supplementary Fig. 8A), supporting our finding of expanded CD45RB^- Atypical Activated in the CyTOF dataset (Fig. 2B, C).

Fig. 6: CD45RB^lo B cells form the majority of SARS-CoV2 reactive B cells during early COVID-19 and post-secondary and tertiary vaccination.

A wnnUMAP of memory B cells highlighted by SARS-CoV2 Spike tetramer binding cells. B Scatterplot of staining intensity for both tetramers with spike positive cells highlighted (>0.2 in each channel). C Proportions of (B) memory clusters for Spike-negative and Spike-positive cells for COVID-19 samples (n = 3). Median proportions of all memory and plasmablasts shown in Supplementary Fig. 8A. D Median proportions of (B) memory clusters for Spike-negative and Spike-positive B cells from SARS-CoV2 mRNA vaccination samples (6 samples from 3 donors). E Variable region mutation frequency across (B) memory clusters for Spike-negative, and Spike-positive cells following the 2nd and 3rd vaccine doses (n = 3). Spike negatives are pooled from both time points (6 samples from 3 donors). F Top two spike binding clonal lineage trees. G Heatmaps of Jaccard index of BCRs between SARS-CoV2 spike positive B cell clusters from all conditions. Numbers displayed on map are clone numbers.

B cell receptor analysis shows clonal overlap between atypical subsets and CD45RB^lo plasmablasts

We then examined levels of mutation in BCR sequences in vaccine samples and found that all groups harbored mutations in their Spike non-binding populations, with Activated Memory cells having the highest levels (Fig. 6E). Examination of spike binding B cells showed that both the Atypical early Activated and the Atypical Activated groups had low levels of mutation after the second vaccination but by post-3^rd vaccine there was a clear increase (Fig. 6E). Examination of individual clone trees also demonstrated that single clones could be found to gain mutations over time and that the same clone could be detected in the Atypical like groups, plasmablasts and CD45RB^lo memory (Fig. 6F). A broader look at clonal overlaps between all spike positive clones revealed strong clonal sharing between the Atypical early Activated, Atypical Activated and plasmablasts (Fig. 6G). Examining all B cells showed that while Atypical Activated had a strong overlap with plasmablasts in COVID-19 and vaccination, the CD45RB⁺ Activated population had the strongest overlap with plasmablasts in Sepsis (Supplementary Fig. 8B), supporting the Sepsis-specific expansion of CD45RB⁺ Activated seen in the CyTOF dataset (Fig. 2B). Together these results suggest that CD45RB^lo atypical-like cells are the primary source of plasmablasts in COVID-19 but in Sepsis CD45RB^hi memory cells were the primary source of plasmablasts and that it is possible to track the plasmablast progeny of different memory B cell subgroups.

RNA velocity analysis⁵⁰ of vaccine and COVID-19 samples also showed that the less activated CD45RB⁺ memory and Atypical memory groups appeared to flow through the corresponding activated memory groups and terminate in plasmablasts as expected (Supplementary Fig. 8C).

CD45RB^lo Atypical and CD23⁺ memory cells are a large proportion of SARS-CoV2 specific cells during acute or resting time points

To further confirm the accuracy of these results, we used mRNA-based anchor transfer⁵¹ of previously published CITE-seq datasets onto our cluster identities (Supplementary Fig. 8D–G). These datasets included peripheral blood BNT162b2 mRNA vaccine recipients following secondary, pre-, and post-tertiary vaccination and a mixed dataset of peripheral blood and tonsil cells from mRNA vaccine recipients⁵². We excluded naïve or GC-like cells since they were not in our dataset. Comparison of the expression patterns of the differentially expressed genes from our dataset (Fig. 5B) and the PBMC dataset (Supplementary Fig. 8E) confirmed that the transfer was effective although analysis of plasmablasts was limited due to a low frequency of plasmablasts in the transferred data.

Similar to our initial findings, CD45RB⁺ memory dominated wild-type (wt)Spike-negative memory B cells but wtSpike, wtRBD and B.1.1.529-Spike positive B cells had a larger proportion of CD45RB^lo memory cells (Fig. 7A). The Atypical like groups could be observed shortly after post-secondary and post tertiary vaccinations but, interestingly, at the resting time point 6 months post-secondary vaccination, the resting CD45RB^lo mem population made up a significant proportion of tetramer positive memory cells. This high proportion remained after the third vaccination alongside expansion of the CD45RB^lo Atypical Activated groups (Fig. 7A) as also seen in our CITE-seq and CyTOF datasets (Fig. 2D, Fig. 5C, D). While our own CITE-seq data had a higher proportion of activated cells, possibly due to differences in time of blood collection after vaccination, the pattern of the result was similar, and it was notable that the majority of SARS-CoV2 binding B cells lacked CD45RB expression following the third vaccination.

Fig. 7: Confirmation of results by anchor transfer.

A Median proportions of memory B cells from Zurbuchen et al.⁵² PBMC vaccine dataset anchor transferred onto our cluster identities. UMAP and marker dotplots in Supplementary Fig. 8D and E. B Single cell Violin plots of indicated markers derived from memory B cells from Zurbuchen et al.⁵² PBMC and Tonsil vaccine dataset anchor transferred onto our cluster identities. C Median proportions of memory B cells from Zurbuchen et al.⁵² PBMC and tonsil vaccine dataset anchor transferred onto our cluster identities. UMAP and marker dotplots in Supplementary Fig. 8F, G.

While the first transferred PBMC dataset did not include CD45RB staining, the second matched tonsil and PBMC dataset included it, allowing clearer confirmation that mRNA-based anchor transfer was able to accurately identify the same CD45RB^lo atypical and CD23⁺ groups cells that we had previously observed (Fig. 7B, Supplementary Fig. 8D-G). Since this dataset was derived from resting timepoints long after vaccination and Atypical cells are expected to be low in lymphoid organs outside the spleen⁵³, only the CD45RB⁺ mem and CD45RB^lo mem were observed in significant numbers. Again, comparison of SARS-CoV2 binding and non-binding cells demonstrated that CD45RB^lo cells were strongly enriched in spike or RBD binding fractions and were often the majority of antigen specific cells in both the blood and tonsils (Fig. 7C). Overall, the analysis of our and other CITE-seq datasets supported the findings of the CyTOF analysis and suggested that further close analysis of the pre- and post-3^rd vaccination time point was warranted.

CyTOF analysis of tetramer specific B cells

We then adapted our B cell CyTOF panel by incorporating tetramers for Spike, wtRBD and Omicron RBD (Fig. 8A) alongside additional markers such as Tbet and CD23 (Supplementary Fig. 9A, Supplementary Table 3). We concentrated on a smaller confirmatory subset of COVID-19 samples and a larger group of pre- and post-tertiary vaccination samples to examine the distribution of virus specific B cells at both resting and acute time points. Initially we used the expanded marker panel to confirm that, as predicted Atypical memory and Atypical Activated B cells express high levels of Tbet, FCRL5 and CD20 (Supplementary Fig. 9A). We also observed that CD45RB⁺ Activated expressed intermediate levels of Tbet, FCRL5 and CD20 both lower than Atypical cells but significantly higher than CD45RB⁺CD27⁺ memory cells.

Fig. 8: Analysis of tetramer reactive B cells by CyTOF confirms that CD45RB^lo groups are the main SARS-CoV2 reactive memory B cells pre and post tertiary vaccination.

A Gating of SARS-CoV2 tetramer-bound B cells. Cells past the indicated thresholds were considered positive. B Percent wtSpike positive cells per total B cell of the indicated cell type (Naive, Memory or Plasmablasts) per total B cells. (n = 14). C Percent wtSpike positive, wtRBD positive and B1.1.529-RBD positive cells of the indicated cell type per total (B) memory cells (excluding Naive and Plasmablasts). (n = 14). D Median proportions of (B) memory clusters of wtSpike negative and indicated tetramer positive cells pre and post third vaccination (n = 14). E Median proportions of B memory clusters of wtSpike negative and indicated tetramer positive cells for COVID-19 (n = 3) Significance *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001 by two-sided Paired Wilcoxon (B, C, D). Box plots are median +/-  interquartile range (IQR), whiskers are 1.5 * IQR.

CD45RB^-CD27⁺ memory cells had significantly higher (although still low) expression of Tbet and CXCR3 than CD45RB⁺CD27⁺ memory cells. CD23 expression was seen on CD45RB^lo memory B cells confirming that these are the same cells seen in the CITE-seq datasets (Supplementary Fig. 9A). CD23 levels decreased as these cells became CD27 positive which, alongside a decrease in CXCR4 and increased CD95 (Supplementary Fig. 2A) suggests that CD27 may act as a maturation marker in a similar manner to CD45RB⁺ memory cells described previously¹³. BLIMP1 was restricted to plasmablasts with CD45RB^lo plasmablasts having lower BLIMP1 but higher Tbet and HLA-DR expression than their CD45RB^hi counterparts suggesting a less mature phenotype with a recent switch from Tbet expressing cells (Supplementary Figs. 2A, 9A).

CyTOF analysis confirms that CD45RB^lo groups are the main SARS-CoV2 reactive memory B cells pre and post tertiary vaccination

To examine the distribution of the SARS-CoV2 antigen specific memory populations, we first gated tetramer positive B cells (Fig. 8A) and confirmed that most spike positive B cells pre and post vaccination were class-switched memory (IgD negative non-plasmablasts) (Fig. 8B).

Proportional analysis of memory subsets demonstrated that CD45RB⁺ memory cells were dominant among spike negative cells while most spike positive cells fell into the CD45RB^lo memory groups (Fig. 8C, D). As expected, the distribution of spike positive subsets changed across time with resting cells making up the majority of antigen specific memory prior to the third vaccination with both CD45RB⁺ and CD45RB^lo CD27⁺ and CD27^- subgroups being enriched (Fig. 8C, D). Most SARS-CoV2 specific cells expanding post vaccination were found in the Atypical Mem, CD45RB^- Atypical Activated and CD45RB⁺ Activated groups while CD45RB^-CD27^- mem were reduced as a proportion of total memory groups (Fig. 8C). wtSpike, wtRBD and B1.1.529-RBD all showed a similar distribution although both RBDs had a slightly higher proportion of CD45RB⁺ cells, particularly pre-third vaccination (Fig. 8D). This data, along with our CITE-seq analysis (Figs. 6D, 7A) suggests resting and recently vaccinated time points are distinguished by the shift from CD45RB⁺ and CD45RB^lo resting populations to activated and atypical phenotypes seen shortly after vaccination.

Examination of our COVID-19 cohort also showed a strong shift in phenotypes with CD45RB^- Atypical Activated being the dominant antigen specific populations at day 1 post ICU admission (Fig. 8E). By day 7 the same donors had begun have a more balanced distribution with non-Atypical cells beginning to recover as a proportion of antigen specific cells, mirroring the shifts in non-antigen specific B cells seen over time in the large cohort (Fig. 2B).

Serum antibody levels are significantly correlated with SARS-CoV2 reactive CD45RB^lo Atypical B cells but not classical memory B cells

We then examined levels of SARS-CoV2 specific antibodies in the blood of the same vaccinated cohort. As expected, levels of IgG recognizing the RBD, S1 and S2 subunits of spike protein strongly increased following the third vaccination while a lack of anti-nucleocapsid antibodies confirmed an absence of asymptomatic infections (Fig. 9A). We then examined the correlation between serum antibodies and proportions of antigen specific B cells in the same patients. Positive correlations were observed between antibody levels and antigen specific B cells but only the Atypical mem and CD45RB- At Activated populations reached significance (Fig. 9B, C, Supplementary Fig. 7B, C). Since the CD45RB^loCD23⁺ subset of B cells has previously been associated with sequential switching to IgE production^48,49, we also assessed levels of SARS-CoV2 spike reactive IgE but were unable to detect spike reactive IgE by ELISA (limit of detection 2.5 ng/ml) either pre or post vaccination despite substantial increases in spike reactive IgG within the same donors (Fig. 9A). This suggests that while these cells may be primed to switch to IgE production in some settings, they do not appear to do so during mRNA vaccination. Indeed, while IgE transcripts were enriched in this subset, the majority (84.7%) of the cells lacked its expression (Supplementary Fig. 7C).

Fig. 9: Correlation of anti-SARS-CoV2 antibodies and tetramer reactive B cells.

A MFI of plasma levels of indicated antibodies in vaccine samples Pre and post 3rd vaccination. (n = 14). Antibody correlations in Supplementary Fig. 9B. B Heatmaps of Spearman rank correlation coefficients of % of tetramer positive cells of B memory cells correlated with plasma MFI of indicated SARS-CoV2 antibodies. Other correlations in Supplementary Fig. 9C. C Scatterplots of selected data from (G). Significance *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001 by two-sided Paired Wilcoxon (A) or two-sided Spearman’s rank correlation (B, C). Box plots are median +/- interquartile range (IQR), whiskers are 1.5 * IQR.

In vitro tracking of B cell subsets confirms that CD45RB is a stable marker of cellular origin and that all B cells can upregulate atypical makers in a T cell dependent manner

Together our results suggested that heterologous CD45RB^lo memory B cells may have a key role in the antibody responses to both vaccination and SARS-CoV2 infection. While CD45RB has previously been considered to be a stable marker of classical memory B cells^13,15,16,17 recent results have also demonstrated that classical memory B cells can switch to a Tbet expressing phenotype, making these boundaries less clear^52,54. While relationships between resting and activated populations can be implied by their similar phenotypes, RNA velocity and clonal analysis, a more direct way to show this is through direct tracking of the fate of sorted CD45RB^lo or high memory groups upon further stimulation.

Therefore, we sorted IgD⁺ Naïve, CD45RB^-CD27⁺ (CD45RB^loCD27⁺CD11c^-IgD^-), CD45RB⁺CD27⁺ (CD45RB^hiCD27⁺CD11c^-IgD^-), CD45RB⁺ Activated (CD45RB^hiCD27⁺CD11c⁺IgD^-) and CD45RB^- Atypical Activated (CD45RB^loCD27⁺CD11c⁺IgD^-) CD19⁺ B cells with the aim of tracing their phenotypes after in vitro stimulation (Fig. 10A, Supplementary Fig. 10A). To provide relatively physiological conditions, we added the sorted B cell subsets into B cell depleted PBMCs and then stimulated them indirectly via anti-CD3 driven T cell activation and cytokine supplementation and used CyTOF to readout the resulting phenotypes.

Fig. 10: In vitro tracking of B cell subsets confirms that CD45RB is a stable marker of cellular origin and that all B cells can upregulate atypical makers.

A Schematic diagram of in vitro assay. Gating strategy for sorting in Supplementary Fig. 10A. B UMAP distribution of indicated input subsets after culture. C UMAP with scaled expression of indicated markers. D Heatmap of median scaled expression of all markers used. E UMAP of B cells from in vitro assay colored by cell cluster. F Proportion across non-Naive B cells (Activated or Plasmablast) for assay input B cell type. (n = 4). Bars are Medians. G Median expression of markers expressed by Activated B cells and Plasmablasts derived from indicated input B cell. P-values from, comparison with Atypical Mem shown (n = 4). Bars are Medians. Other markers in Supplementary Fig. 10B. Significance in E & F in comparison to Atypical Mem *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001 ****P ≤ 0.001 from one way ANOVA with Holm adjustment. A Created with BioRender.com released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license (https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en).

Examination of the resulting phenotypes and placement in the UMAP demonstrated that all cell types were capable of producing both activated and plasmablast phenotypes (Fig. 10B, C). We broadly divided the resulting cells into Naïve, activated B cells and plasmablasts using markers such as IgD, CD20 and BLIMP1 (Fig. 10B–E) and determined that all sorted B cell subsets could differentiate into plasmablasts (Fig. 10F). On examination of activated memory B cells derived from any of these initial populations we found that all sorted subsets were capable of upregulating CD11c and Tbet to a similar degree (Fig. 10G). However, CD45RB was neither lost by initially CD45RB⁺ cells nor gained by initially CD45RB^lo cells prior to plasmablast differentiation, while CD27 was also slightly lower (Fig. 10G). This difference could not be explained by lower proliferation or activation since levels of Ki67, CD71 and active RNA and protein production measured by incorporation of BrU and puromycin⁵⁵ were either not different or higher for Atypical Mem derived cells (Supplementary Fig. 10B) as also observed previously¹³. When examining CD45RB expression in plasmablasts we again saw that Atypical mem derived cells had significantly lower CD45RB expression (Fig. 10G) although it was upregulated in comparison to activated B cells. Atypical mem derived cells could also be seen to be concentrated in the CD27 and CD45RB low areas of the UMAP (Fig. 10B, C, Supplementary Fig. 10C). Together with the different levels of HLA-DR, BLIMP-1 and Tbet between these populations, we interpreted these findings as indicating that CD45RB^lo plasmablasts are a less mature intermediate derived from CD45RB^lo memory cells but may later switch to a more mature CD45RB⁺ phenotype.

Discussion

Here we used multimodal analysis of longitudinal COVID-19, Sepsis, and mRNA vaccination cohorts to dissect the cellular immune response in these settings, with a focus on a deep analysis of B cells. Several reports have already described clear increases in classical, atypical, and activated memory following mRNA vaccination and during COVID-19^{8,20,21,22,23,24,26}. However, there is considerable diversity in the strategies used to define atypical cells in these reports with CD27 + CD21- B cells being defined as either activated or atypical cells in different studies^56,57,58. This confusion may be due to current difficulties in separation of the atypical and activation signatures which overlap in many cases.

In this report we have demonstrated that the stable absence of CD45RB in non-classical B cells can be used to better demarcate activated and plasmablast B cells derived from initially CD45RB^lo non-classical B cells. We found that while they are a rarer subset, CD45RB^lo activated memory cells with an atypical phenotype had a stronger response to COVID-19 and mRNA vaccination than CD45RB⁺ activated memory cells which were primarily increased in Sepsis, regardless of Gram status. The frequencies of these cells did not have a clear relationship with mortality in either Sepsis or COVID-19. Both phenotypic and clonal relationships suggest that these activated cells were at an intermediate stage between CD45RB^lo or CD45RB^hi resting memory B cells and terminated in matching groups of CD45RB^lo or CD45RB^hi plasmablasts.

While atypical memory cells are detectable in blood, it is a diverse group and many of the markers used to identify it may also catch B cells recently activated in the presence of interferon or IL-21⁵⁹. Several recent reports identified some degree of identity switching, as classical memory B cells were seen to gain atypical-like activation signatures defined by Tbet and FCRL5 upon reactivation^52,54. Here we also support this conclusion demonstrating that essentially all B cells can gain Tbet expression upon re-stimulation. However, we also suggest that while both CD45RB^lo and CD45RB^hi activated B cells express CD11c and Tbet, the former may be the product of atypical memory cells while the latter may derive from interferon-driven activation of classical memory B cells.

Strikingly, the majority of tetramer-positive cells following both COVID-19 and mRNA vaccination were in the CD45RB^lo compartment of B cell memory, and the question arises as to what is the source of these cells? At the resting time-point before the 3^rd dose of mRNA vaccines, most CD45RB^lo SARS-CoV2 reactive B cells in both the blood and tonsils did not express atypical markers, but rather a signature characterized by CD23 and IL4R that has recently been described in both CD27^-47 and CD27⁺ memory cells in the context of allergy^48,49.The observed CD23⁺IL4R⁺CD45RB^lo B cells had little in common with the interferon driven signature seen in atypical B cells, lacking markers such as ZEB2, CD11c and Tbet⁶⁰. However, we demonstrated that during in vitro culture that both these and classical memory B cells were capable of upregulating Tbet and CD11c. While initially CD45RB^lo and CD45RB^hi cells could still be separated on the basis of CD45RB, the CD11c⁺ (Atypical mem) and CD11c^- CD45RB^lo memory cells were essentially indistinguishable after culture. Since both in our own and other datasets we saw that CD45RB^lo non-atypical cells were a significant reservoir of resting SARS-CoV2 memory cells prior to 3^rd vaccination, this raises the possibility that they may be one source of the activated CD45RB^lo Tbet positive cells seen following vaccination. Another non-mutually exclusive possibility is that the cells are also derived from rare but still detectable Atypical memory cells that may expand rapidly upon revaccination. Thirdly, naïve B cells are also a potential source of CD45RB^lo activated cells, however in this case we also saw clonal lineages across vaccination time points and a clear increase of variable gene mutations in the later vaccination suggesting that memory cells are a likely source. While increased mutations over time could be an indicator of reentry into secondary GCs, SHM is not completely exclusive to GCs^61,62 and Tbet expressing B cells have previously been reported to have lower mutations than GC derived B cells but higher than naïve cells⁶³. Here we also observed that the CD45RB^lo activated subset of B cells had a lower level of mutation than CD45RB⁺ activated B cells, another potential indicator of different origins.

Due to the recent discovery of CD45RB^loCD23⁺IL4R⁺ memory B cells very little is known about their origin^47,48,49. However, a recent preprint suggests that a murine equivalent to these cells is observable and may be partially independent of germinal centers⁴⁹. The recent descriptions of these cells have focused on their presence in allergy and ability to switch to IgE production^48,49, but here we saw no evidence of SARS-CoV2 reactive IgE after vaccination.

While it is generally understood that both classical and non-classical memory B cells are broadly T-dependent⁶⁴ the balance between Tfh and Tph may also have a key role in determining B cell fates. A recent study demonstrated that during severe, but not mild, COVID-19 there is an overall decrease in cTfh and a delay in the development of antigen specific Tfh⁶⁵. Together with previous findings demonstrating suppressed GC in deceased COVID-19 patients, this indicates that some suppression of Tfh may occur during severe acute COVID-19^66,67. However, studies of convalescent patients also show a clear increase in antigen specific Tfh and functional GCs underlining that this may be a temporary phenomenon specific to severe disease^18,68,69,70. More specifically the Tfh1 subset has also been observed to be functional in COVID-19 and associated with atypical phenotypes of B cells^71,72,73. Tph have also been shown to increase in COVID-19^32,33. Previous reports have suggested that atypical cells can be induced by either IFNγ or IL-21, both of which may be sourced from either Tph or Tfh1^74,75,76. As a result, both Tph and Tfh1 are credible sources of B cell help in this context and here we saw that all these previously reported phenomena were observable in a time and context dependent manner. While severe COVID-19 was characterized by an expansion in Tph but relatively weak response from Tfh, mRNA vaccination showed a more balanced response with both Tph and Tfh1 expanded at the same time points suggesting that multiple branches of T cell help can be active simultaneously.

The primary limitation of this study is that since the data was exclusively derived from blood, we could not directly examine the tissue origin or homing of the indicated cells as regards either extra-follicular or GC origins. Additionally, further work is needed to dissect the signals that drive naïve and memory B cells into the different forms we have seen here.

In conclusion, using multimodal analysis of large longitudinal cohorts from several settings we have been able to provide a comprehensive structure of circulating B cells in human blood. The stable difference in the expression of CD45RB made it possible to track B cells derived from three branches of memory CD27⁺CD45RB^hi (classical memory), CD23⁺CD45RB^lo (non-classical memory) and Tbet⁺CD45RB^lo (atypical memory) through matching stages of activated and plasmablast differentiation. Importantly most SARS-CoV2 binding B cells following mRNA vaccination COVID-19 fell into the CD45RB^lo branch of memory/plasmablast differentiation and were correlated with serum antibody levels. This suggest that the non-classical route of B cell activation may have a key role in the response to viral infections. Understanding this balance in both infection and vaccination will be critical to the development of new treatments and increasing the efficacy of vaccines to emerging and existing threats from infectious disease.

Methods

Human participants recruitment and sampling ICU Cohort

PBMC and plasma samples were collected longitudinally from cohorts of moderate COVID-19 (non-Intensive care) and severe (Intensive care) COVID-19 and severe (Intensive care) Sepsis patients, and healthy age-matched healthy controls (Supplementary Table 1). Hospitalized cases diagnosed as COVID-19 were enrolled by physicians using clinical manifestation and PCR test results at Osaka University Hospital and Osaka General Medical Center. Sepsis patients were enrolled by physicians using clinical manifestation based on sepsis-3 criteria at Osaka University Hospital and Osaka General Medical Center. Age-matched healthy control volunteers were collected at Osaka University Hospital and Osaka General Medical Center. Patients with COVID-19 were further categorized by the WHO ten-point clinical progression scale, (Supplementary Table 1)⁷⁷.

Additionally, PBMC and plasma samples were collected from a cohort of healthy donors during a vaccination time-course with the Pfizer BioNTech BNT162b2 SARS-CoV-2 vaccine (BNT162b2) (Supplementary Table 1). Samples were collected from August 2020 to May 2021 at Osaka University Hospital and Osaka General Medical Center. This study followed the principles of the Declaration of Helsinki and was approved by the institutional review board of Osaka University Hospital (permit nos. 907 and 885 [Osaka University Critical Care Consortium Novel Omix Project; Occonomix Project]). Informed consent was obtained from the patients or their relatives and the healthy volunteers for the collection of all blood samples.

Peripheral blood sample processing

Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood samples using Ficoll-Paque density gradient centrifugation, resuspended at 1-2E6 cells/ml Cellbanker 1 (Takara Bio) before being subject to controlled freezing at -80 °C in a CoolCell device (Corning) and later stored in gas-phase liquid nitrogen.

Mass cytometry antibody production

We obtained Indium-113 and -115 and gadolinium-157 (Trace Sciences) for conjugation to antibodies. X8 polymer MaxPar kits (Standard Biotools) were used to conjugate Indium and lanthanide isotopes, while MCP9 polymer kits were used to conjugate Cadmium isotopes according to the manufacturer’s instructions. Platinum-labeled antibodies were conjugated with cisplatin by treating 100 μg purified antibody in 4 mM tris(2-carboxyethyl)phosphine (TCEP) solution (Thermo) for 30 min at 37 °C before conjugation with 20 nmol of mono-isotopic cisplatin (Standard Biotools) as previously described⁷⁸. Conjugated antibodies were stored in a PBS-based antibody stabilizer or HRP-protector stabilizer for cadmium labeling (Candor Biosciences). All antibodies were titrated using control PBMCs to obtain optimal staining concentrations. Antibody staining panels were either prepared fresh, or prepared in bulk and stored as aliquots at -80 °C.

CD45 barcoding and cell staining for mass cytometry

Samples were measured across 21 experimental runs. The 21st run used an updated panel for B cells that included tetramers for SARS-CoV-2 spike protein.

Frozen PBMC samples were defrosted in a 37 °C water bath for 2 min and poured into a 15 ml tube and 5 ml of prewarmed RPMI containing 10% FCS and 20 IU/ml Pierce Universal Nuclease for Cell Lysis was added. Samples were then washed with the same buffer without resuspending, then resuspended in 2 mL of CyFACS buffer, and live cells counted by Acridine Orange/Propidium Iodide staining on a LUNA-FL fluorescence cell counter (Logos Biosystems). In each experiment, up to 4E6 viable cells per sample were labeled with a seven choose-three pattern of anti-CD45 barcodes (89Y, 113In, 115In, 194Pt, 195Pt, 196Pt, and 198Pt) to give a combination of up to 35 barcoded samples per experiment, including a spike in control used to monitor batch effect (Supplementary Fig. 1B)⁷⁹. Barcoding antibody aliquots in CyFACS were prepared in bulk and stored at -80 °C. Samples were incubated with CD45 barcodes together with FC-block and anti-CXCR5 biotin (in run 21 a direct anti-CXCR5 antibody was used to avoid interference with streptavidin-biotin B cell tetramer conjugates) in CyFACS buffer (PBS with 0.1% BSA and 2 mM EDTA) for 30 min at room temperature (RT), and then washed once by adding 5 mL CyFACS buffer, centrifugation at 400 g and the supernatant was poured off. The barcoded cells were resuspended in 700 μl CyFACS buffer, pooled and filtered through a 70 μm filter (Miltenyi) to remove dead cell debris. At this stage 5-10% cells were transferred to a 15 mL tube (Lineage stain tube, total PBMCs used for lineage proportions across CD45+ PBMCs). The remainder of the cells were separated into a CD3+ fraction (T stain tube) and CD3-depleted fraction (non-T stain tube) using a magnetic bead-based CD3 positive selection kit (StemCell). Cells in each tube were stained with separate surface and intracellular panels for T cells (CD4, Tfh, Treg, CD8 and γδT), non-T cells (Monocytes, DCs, B and NK cells), and total lineage proportions (panel information Supplementary Table 2). Premixed and frozen master mixes were used to reduce batch effects. For the 21^st experiment, only the non-T stain tube was stained using panel for B cells.

Cells were stained with the respective surface stain antibody cocktail for 45 min at RT. The cells were then washed twice in CyFACS buffer (400 g, 5 min, RT), stained using 500 nM Zirconium based viability stain⁸⁰ in PBS for 5 min at RT, washed, and then fixed and permeabilized using the Foxp3 Transcription Factor Staining Buffer Set according to the manufacturer’s protocol (eBioscience). The cells were subsequently stained with the respective intracellular antibody cocktail for 45 min at 4 °C and then washed twice (centrifugation at 800 g, 5 min, 4 °C) in CyFACS buffer and once in PBS. The cells were then fixed overnight in 2% formaldehyde (Thermo) in PBS containing 1 μM DNA Cell-ID Intercalator-103Rh (Standard Biotools).

Preparation of B cell tetramers for mass cytometry staining

B cell tetramers were prepared in a similar way to previous reports^81,82. Briefly, biotin-labeled SARS-CoV-2 wild-type full-length Spike (Miltenyi), wild-type RBD domain or B.1.1.529/OmicronRBD (Acro Biosystems) that had been reconstituted at 100–200 μg/mL in PBS10 were mixed with Streptavidin (SA)-FITC, SA-PE or SA-APC (Biolegend) respectively in three steps separated by 15 min to give a final 4:1 molar ratio of protein:SA and total incubation time of RT for 45 min. Tetramer staining mix was prepared by adding each tetramer to CyFACS with 2.5 μM Biotin to saturate unbound streptavidin and minimize probe cross-reactivity. Following barcoding and mixing of cells, they were stained with the three tetramers (wtSpike-FITC, wtRBD-PE and B.1.1.529/OmicronRBD-APC) at RT for 45 min in a total volume of 550 μl. Then cells were washed three times in CyFACS buffer (400 g, 5 min, RT) following which the staining protocol was identical to that for experiments 1–20 but using the B cell staining panel (Supplementary Table 3).

Mass cytometry data acquisition

Cells were washed (800 g, 5 min, 4 °C) once in 5 mL CyFACS buffer and twice in 2 mL Cell Acquisition Buffer (CAS, Standard Biotools). The cells were diluted to 1.1E6 cells/mL in CAS with 15% EQ Four Element Calibration Beads (Standard Biotools) and passed through a 35 μm filter immediately before collection. Cells were run at 200 to 500 cells/s on a Helios mass cytometer (Standard Biotools), or CyTOF XT. Flow Cytometry Standard (FCS) files were normalized to EQ beads and concatenated using Standard Biotools software before export.

B cell in vitro plasmablast differentiation assay

The B cell in vitro plasmablast differentiation assay was performed using frozen PBMC from healthy Leukopak donors (StemCell). For each assay, ~1E8 cells in liquid nitrogen were thawed as above except 50 mL tubes were used for the initial thawing and washing step. 10% of PBMC was set aside for sorting of non-B cells while 1% of PBMC was set aside on ice for the whole PBMC control. The remainder was enriched for B cells using the EasySep™ Human B Cell Isolation Kit (StemCell). Enriched B cells were stained with a panel of fluorescently labeled antibodies (Supplementary Table 5) and total PBMC were stained with anti-CD19 and anti-IgD to allow sorting of non-B cells (CD19-IgD-). Cells were stained for 30 min at RT and washed twice in 5 mL PBS + 10% FCS (PBS10), then resuspended in 1–2 mL of PBS10 and sorted at 4 °C on a BD FACSymphony S6 cell sorter (BD). B cells were sorted into six different populations (Supplementary Fig. 7D) after which non-B cells were sorted. Sorted cells were transferred from 5 mL FACS tubes into 15 mL tubes and spun down. Up to 18 conditions were included in the B cell assay. The base media for the B cell assay (RPMI + 10% FBS + Pen/strep (cRPMI) with 1 μg/mL anti-CD3 antibody (Biolegend)), 10 U/mL IL2 (Immunace35, Shinogi) and 10 ng/mL IL-1β (Peprotech). For further stimulation of B cells, some conditions were supplemented with a cocktail of 10 ng/mL IL-10, 10 ng/mL IL-4 and 50 ng/mL IL-21 (Peprotech) added on day 4 of the assay. To prepare the assay, in 96-well U bottom plates (Corning #163320), 100 μl of media (containing 2x concentration of stimulants) was added to wells followed by 100 μl of cells in cRPMI. For the whole PBMC condition, 1E5 of the whole PBMC (100 μl at 1E6 cells/mL) was added to respective wells. For B cell assay conditions, sorted non-B cells were added to each well (9E4 cells in 50 μl) followed by sorted B cells (1E4 cells in 50 μl) yielding an approximately physiological ratio of 10% B cells to 90% non-B cells in 200 μl per well. Plates were incubated at 37 °C 5% CO₂ for 6 days. On the 4th day, 50 μl of media was aspirated from each well and replaced with 50 μl cRPMI or cRPMI with 4x cocktail for the day 4 cocktail condition.

On day 6 samples were stained and measured by CyTOF with the following modifications. Cells were pipetted out of wells into 15 mL tubes, spun down and first labeled with a six choose-three pattern of anti-CD45 barcodes (89Y, 113In, 115In, 194Pt, 195Pt, and 196Pt) to give a combination of up to 18 barcoded samples per experiment with 198Pt to be used for dead cell staining. To allow incorporation of puromycin, IdU and BrU into cellular macromolecules⁵⁵, the staining protocol included a 30 min pre-stain at 37 °C in cRPMI at which time cells were also stained for chemokine receptors (CCR7, CXCR3, CXCR5) (Supplementary Table 4). Cells were then stained with surface and intracellular panels. B cell assay samples were run on a CyTOF XT using suspension mode in CAS+ buffer with 10% 6-element EQ beads (Standard Biotools). Flow Cytometry Standard (FCS) files were normalized to EQ beads and concatenated using Standard Biotools software before export.

Mass cytometry data analysis

FCS files were uploaded to Cytobank software (Beckman Coulter) and gated as positive for DNA, negative for EQ beads, below a threshold on the Zr live/dead stain (or 198Pt negative in the B cell assay), with normal ion cloud Gaussian parameters (Supplementary Fig. 1A), before being manually debarcoded based on the CD45 barcodes (Supplementary Fig. 1A, B) and exported as FCS files.

Firstly, FCS files were imported into R (v.4.3.0) via flowCore (v.2.12.2) and batch corrected (SOM size 10 by 10, arcsinh 5 transformed) with cyCombine (v.0.2.15)⁸³. Spike in samples in every experiment were used to confirm the accuracy of batch correction. In the T stain tube, CCR2, which was not in the panel for experiments 1 and 2, was imputed onto these two batches using the salvage_problematic() function of cyCombine. In the Lineage stain tube, CD33, CD95, CXCR4, CD27, IgD, CD56, Helios, Granzyme B, TCRγδ, CD38, HLA-DR were imputed from experiments 2-20 onto experiment 1 using the CyTOF 2 panels vignette of cyCombine.

Following batch correction samples were exported as FCS files and then re-imported to R and converted to single cell experiment format (SCE) and analysed with CATALYST (v.1.24.0)^34,35 as previously described³². Briefly, data was then compensated using a compensation matrix derived from a compensation panel of single-metal-labeled compensation beads. Data was clustered using flowSOM (using a 10 by 10 SOM and subsequent meta-clustering) using a multi-step process. The first step employed a limited set of markers optimized to separate into the main immune lineages. Secondly, each lineage population was sub-clustered to obtain fine cell sub-types. Detailed cell type annotation was based on prior knowledge combined with analysis of heatmaps and UMAPs (downsampling to 1000 cells per sample and 40 nearest neighbors). In the case of CD4 cells, CXCR5⁺Foxp3^- Tfh were also handled as a subcluster to allow their in-depth analysis.

Non-T stain tube data for experiments 1-20 was clustered to obtain B cells, NK cells and Myeloid & DCs. For B cells, the panel for experiments 1-20, which was designed for clustering all CD3- cells, was different from that of experiment 21, which was focussed specifically on B cells and tetramers. However, we sub-clustered the B cell data from experiments 1-21 together using only shared markers after batch correction with cyCombine using the CyTOF 2-panels vignette. For simplicity we merged the CD73+ or CD73- Naive and CD73+ or CD73- Classical B cell types previously identified by Glass et al. ¹³. To calculate isotype proportions, the B cell SCE was reclustered using only IgG, IgA, IgD and IgM and isotype proportion per phenotypic cluster was calculated.

Differential abundance testing among clusters was performed with diffcyt (v.1.20.0) in DA mode using edgeR (v.3.42.4) and default settings. Prior to plotting, clusters with 0% abundance were set to 0.01% for display on Log scaled plots. Expression levels of markers were tested for statistical significance using Kruskal-Wallis followed by Dunn’s with Holm adjustment with rstatix (v. 0.7.2).

ForceAtlas2 visualization

For visualization of B cells by forceatlas2, 5000 cells per cluster were used to generate a KNN edge matrix in vortex software⁸⁴, the resulting exported XML file containing edge and node lists was imported into Gephi (v.0.10.1) network analysis software for layout by the Forcealtas2 algorithm using, Tolerance = 1, Approximate repulsion = 1.2, scaling = 1, gravity = 4 and dissuade hubs ON³⁶.

CITE-seq

CITE-seq was performed on 12 samples: three COVID-19 day 1 samples, three Sepsis day 1 samples, and samples from three vaccine cohort donors after their second and third vaccine dose. PBMC for each sample were thawed as above and cells were barcoded with TotalSeq-C Hashtag antibodies. Cell pellets were resuspended in a total of 100 μl barcode mix (0.5 μl hashtag antibody in PBS10) and stained for 30 min at 4 °C. Cells were then washed twice in 5 mL PBS10 (400 g, 5 min, 4 °C), resuspended in 1 mL PBS10 and then mixed into a single tube and centrifuged (400 g, 10 min, 4 °C).

The cell pellet was resuspended in a total of 200 μl staining mastermix containing fluorescently labeled sorting antibodies, TotalSeqC antibodies and two freshly prepared SARS-CoV-2 wild-type full-length Spike tetramers (Supplementary Table 4 and 6) and stained for 30 min at 4 °C. Cells were then washed twice in 5 mL PBS10 (400 g, 5 min, 4 °C) and sorted on a FACSaria III (BD) into 1.5 mL tubes. IgD- B cells (CD19+IgD-CD4-CD14-) (post sort viability >95%) were resuspended at 700 cells per μl and submitted to the Osaka University Biken NGS Core for library preparation and sequencing.

Single-cell RNA library construction and sequencing

Single-cell suspensions were processed with a 10 x Genomics Chromium Controller using the Chromium Next GEM Single Cell 5’ Kit v2, Chromium Next GEM Chip K Single Cell Kit and Dual Index Kit TT Set A according to the manufacturer’s instructions. Library preparation and sequencing was done by loading ~16,500 live cells per well on the Chromium controller to generate 10,000 single-cell gel-bead emulsions (5 wells total). Oil-encapsulated single cells and barcoded beads (GEMs) were reverse-transcribed on a Veriti Thermal Cycler (Thermo). The resulting cDNA tagged with a cell barcode and unique molecular index (UMI) was amplified to generate single-cell libraries, which were quantified with an Agilent Bioanalyzer High Sensitivity DNA assay (Agilent). The amplified cDNA was enzymatically fragmented, end-repaired, and polyA tagged followed by cleanup and size selection using SPRIselect magnetic beads (Beckman-Coulter). Next, Illumina sequencing adapters were ligated to the size-selected fragments followed by another round of cleanup. Finally, sample indices were selected and amplified, followed by a double-sided size selection using SPRIselect magnetic beads.

For VDJ repertoire profiling, full-length VDJ regions were enriched from amplified cDNA by PCR amplification with primers specific to the Ig constant regions using the Chromium Single Cell Human BCR Amplification Kit and enriched VDJ segments were used for VDJ-library construction. After final library quality assessment using an Agilent Bioanalyzer High Sensitivity DNA assay, samples were sequenced on an Illumina NovaSeq 6000 as paired-end mode (read1: 28 bp; read2: 91 bp).

Preprocessing of single-cell RNA sequencing data

Data was pre-processed using the CellRanger multi pipeline (version 7.0.0). Analysis was done in R (v.4.3.0) using the Seurat package (v.4.3.0). Since samples were split across multiple GEM wells, a well ID was first added to the barcodes followed by sample assignment based on Hashtags using the HTODemux() function. Data was then sent through the following QC pipeline. The data was filtered for hashtag singlets and the following QC metrics (500 > sample > 8000 mRNA features per cell, <10% mitochondrial reads, >5% ribosomal reads and <3000 ADT counts per cell). Ribosomal and mitochondrial genes were then removed along with the MALAT1 gene. Data was then normalized according to recommended settings with RNA (method = LogNormalize, scale.factor = 10,000) and ADT (method = CLR, margin = 2). Cell cycle was then assigned using CellCycleScoring and data was scaled regressing out the %mitochondrial and cell cycle difference (S score minus G2M score) from the RNA data. This resulted in a Seurat object with 28491 QC-passed cells (10341 for COVID-19, 2296 for Sepsis and 15854 for Vaccine samples).

Analysis of single-cell RNA sequencing data

FindVariableFeatures() was used to identify the list of top 3000 variable genes. IGHV, IGKV and IGLV immunoglobulin genes were removed from this list. Seurat (v.4.3.0) multimodal clustering of the GEX (mRNA) and ADT data was performed. Principal component analysis (PCA) was run on the GEX data using the variable features and another PCA was run on all ADT features. FindMultiModalNeigbors() was then run using PCs 1–6 of GEX and PCs 1–4 of ADT at k = 50 nearest neighbors. The wnnUMAP was made and data was clustered using FindClusters with SLM algorithm and resolution = 2. The 24 resulting clusters were then manually merged and annotated to give clusters shown (Fig. 5B). SARS-CoV-2 wtSpike tetramer positive cells (total: 1042) were annotated as ≥ 0.2 in both the PE and APC ADT channels (Fig. 5B). Differential expression between populations was performed with FindMarkers() at default settings.

Anchor transfer

Anchor transfer of the datasets⁵² was done with Seurat (v.4.3.0) in R (v.4.3.0) using a supervised PCA made from the wnn graph, and k.filter set to NA. IgD+ cells were removed from the query data, as the reference was sorted for IgD- cells (RNA cutoff: 0.1, protein cutoff: 0.5).

RNA velocity analysis

Count matrices of unspliced and spliced RNA was generated using velocyto (v.0.17.17) alignment to hg38 in zsh (v.5.9) and merged with loompy (v.3.0.6) in python (3.9.16). RNA velocity was calculated using the dynamic model in scVelo (v. 0.2.5) with variable features pre-defined in the seurat analysis, 50 principal components, 50 neighbors, and 10 shared counts. scVelo was run in Jupyter notebook (v.7.0.0) using the following dependencies: anndata (v.0.9.1), scanpy (v.1.9.3), numpy (v.1.21.1), scipy (v.1.10.1), pandas (v.1.5.3), scikit-learn (v.1.2.2), matplotlib (v.3.7.1), python-igraph (v.0.10.6), louvain (v.0.8.0), pybind11 (v.2.11.1), hnswlib (v.0.6.2), ipython (v.8.14.0), tqdm (v.4.65.0), and ipywidgets (v.8.0.7). For use in python, the seurat object was converted to h5ad using SeuratDisk (v.0.0.0.9020) in R (v.4.3.0).

Analysis of BCR data

The Immcantation⁸⁵ and scRepertoire (v. 1.11.0)⁸⁶ workflows were used for analysis of BCR data. In scRepertoire the ‘strict’ definition of clonotype was used for all analyses. The Immcantation workflow starts by running igblast (version 1.21.0) using Python scripts provided by Immcantation to give the “filtered_contig_igblast_db-pass.tsv files”. Given that the data spanned 5 GEM wells, a well ID was appended to the cell_id and sequence_id columns, after which the data was concatenated, therefore allowing cells found in both the GEX and BCR datasets to be matched by their cell barcode.

Briefly, data was filtered for productive BCRs, and cells with none or multiple heavy chains were removed (yielding 21300 cells with a single heavy chain). Then clonotyping was done with Scoper using an automatically-determined threshold. For calculating clonal overlap between Seurat clusters, clonotyping results were filtered for IgH and then metadata columns from the seurat object were attached giving 15103 cells found in both datasets (6692 for COVID-19, 1538 for Sepsis and 6873 for Vaccine), of which 460 were spike positive.

Somatic Hypermutation (SHM) on heavy chain data was then calculated using observedMutations (regionDefinition = IMGT_V) (Shazam) (IMGT database updated May 2023), yielding 20687 cells mapped to germline of which 14659 were also found in the GEX data (453 spike positive). Dowser (v.2.0.0) was then used to format and plot the germline-mapped cells as phylogenetic clone trees (build = “pml”) with metadata attached from the Seurat object.

Clonal overlap between Seurat clusters was calculated as the Jaccard index, which is the ratio of the number of intersecting clonotypes to the union count of clonotypes between two clusters. The raw count of intersecting clonotypes between pairs of clusters is also written on the heatmaps. IgH gene usage heatmaps by Seurat cluster were calculated using vizGenes() function in the scRepertoire pipeline.

Analysis of plasma antibody levels by Luminex and ELISA

During preparation of PBMC, plasma samples were also collected and stored at -80 °C. Serum titers of SARS-CoV-2 antibodies were assessed using the Bio-Plex Pro Human IgG SARS-CoV-2 N/RBD/S1/S2 4-Plex Panel kit (BioRad) according to the manufacturer’s directions on a BioPlex-200 machine (BioRad). Freshly thawed plasma samples were centrifuged (1000 g, 10 min, 4 °C), diluted 1:1000 (4 μl into 196 μl then 5 μl in 95 μl), and run in duplicate. Background subtracted MFI duplicates were averaged and spearman correlations with cellular proportions from the corresponding PBMC samples were calculated with Corrplot v.0.92.

An Anti-SARS-Cov-2 SPIKE IgE ELISA kit (Matriks Biotech) was run according to the manufacturer’s directions using 1:100 and 1:500 dilution of each plasma sample. The same samples were run as for measuring SARS-CoV-2 IgG antibodies.

Statistics and reproducibility

No statistical method was used to predetermine sample size. We estimated required sample sizes from previous studies. Three samples with <150 total live cells were not used for immune subset analysis (detailed T and and non-T cell analysis). This criteria was not pre-established. The experiments were not randomized. The CyTOF experiments were performed in 20 batches with donors assigned in a random manner. Sepsis and COVID-19 were in mixed batches but mRNA vaccination samples were specifically in batches 11,12,13,14 and 15. Batch 21 was only for the pre and post third mRNA vaccination samples with a modified antibody panel. CITE-seq was performed in a single experiment. Patients were stratified based on clinical manifestation based on sepsis-3 criteria or categorized by the WHO ten-point clinical progression scale for CIOVID-19. Further randomization was not performed due to the exploratory nature of the study. The Investigators were not blinded to allocation during experiments and outcome assessment. Technical reproducibility of the main mass cytometry dataset was confirmed by monitoring of batch control spike in samples across each of the 21 separate experiments. Intraexperimental variation was controlled by barcoding/hashtagging so that all staining, washing and data collection was performed with mixed samples. Statistic tests are described in figure legends.

Figure production

Overview diagrams in Figs. 1A, 2A, 2C, 5A and 10A were created with BioRender.com. Figures were assembled in Adobe Illustrator (v.27.3-28.0).

Reporting summary

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