Author response: Single-cell analysis reveals dynamics of human B cell differentiation and identifies novel B and antibody-secreting cell intermediates

Abstract

Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Differentiation of B cells into antibody-secreting cells (ASCs) is a key process to generate protective humoral immunity. A detailed understanding of the cues controlling ASC differentiation is important to devise strategies to modulate antibody formation. Here, we dissected differentiation trajectories of human naive B cells into ASCs using single-cell RNA sequencing. By comparing transcriptomes of B cells at different stages of differentiation from an in vitro model with ex vivo B cells and ASCs, we uncovered a novel pre-ASC population present ex vivo in lymphoid tissues. For the first time, a germinal-center-like population is identified in vitro from human naive B cells and possibly progresses into a memory B cell population through an alternative route of differentiation, thus recapitulating in vivo human GC reactions. Our work allows further detailed characterization of human B cell differentiation into ASCs or memory B cells in both healthy and diseased conditions. Editor's evaluation To recapitulate B cell differentiation process in vitro, the authors established an in vitro system to identify a cluster and performed extensive analyses to demonstrate that the cluster mimics human germinal center antibody-secreting cells (ASCs). They provide stepwise trajectories of plasma cell differentiation from human naive B cells upon stimulation with CD40 ligands, IL-4, and IL-21. Since intermediate clusters of cells show features of germinal center B cells, the authors propose novel intermediate stages of B cells before a complete differentiation into plasma cells. This study is valuable in the differentiation of primary naive B cells into ASC ex vivo and may interest immunologists with an emphasis in B cell biology as it provides an in-depth description of the B cell differentiation pathways. https://doi.org/10.7554/eLife.83578.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Protection from invading pathogens is, in part, provided by antibody-secreting cells (ASCs) that secrete high-affinity class-switched antibodies. Antibody formation, however, can also be highly undesired. This is the case for detrimental autoantibodies in autoimmune disorders, for alloantibodies that react with donor blood cells after a blood transfusion, or for antibodies that are formed against therapeutic biologics and that interfere with their therapeutic efficacy. Plasma cells secreting high-affinity antibodies are generated from naive B cells that, upon foreign antigen recognition by the antigen-specific B cell receptor (BCR), get activated and after CD4 T cell help can start a germinal center (GC) reaction. Here, B cells differentiate into GC B cells and begin to cycle between the GC dark zone (DZ) and the GC light zone (LZ). In the former, GC B cells undergo several rounds of cell cycle and receptor editing processes, such as somatic hypermutation and class-switching. In the latter, B cells need to reacquire antigen and present antigenic peptides to cognate antigen-specific CD4 T follicular helper (Tfh) cells to receive the required co-stimulatory, survival, and differentiation signals (Bryant et al., 2007; King and Mohrs, 2009; King et al., 2008; Reinhardt et al., 2009; Yusuf et al., 2010; Glatman Zaretsky et al., 2009). After several rounds of DZ to LZ cycling, this process results in the selection of high-affinity, class-switched B cells (reviewed in Victora and Nussenzweig, 2022). While early memory B cell and early plasmablast formation occur before GC establishment (Glaros et al., 2021), the GC reactions give rise to a later stage, higher affinity memory B cell formation followed by the eventual formation of high-affinity, ASCs (plasmablasts and plasma cells) (Weisel et al., 2016). Migration of plasma cells to the bone marrow and incorporation into bone marrow survival niches establishes long-lived plasma cells that may secrete antibodies for decades (Belnoue et al., 2008; Cassese et al., 2003; Radbruch et al., 2006). The transition of B cell into ASC is tightly regulated on the molecular level by a complex network of transcription factors that control the switch from the B cell stage to the ASC stage (reviewed in Verstegen et al., 2021). The ASC-specific transcription factors IRF4, XBP1, and BLIMP1 are upregulated only once the B cell-specific transcription factors PAX5, IRF8, and BACH2 are downregulated. These transcriptional changes are then followed by surface expression of stage-specific cellular markers, such as CD27, CD38, and CD138 on ASCs, with simultaneous downregulation of the B cell markers CD19 and CD20 (Loken et al., 1987; Sanderson et al., 1989; Tangye et al., 1998; Terstappen et al., 1990; Victora et al., 2012). Although these transcription factors associated with plasma cell differentiation and plasma cell markers have been well-studied, the differentiation trajectories of naive B cells into short-lived and long-lived ASCs and the regulators that control transition at the different stages of human B cell differentiation remain largely unknown. Bulk RNA sequencing on sorted B cells ex vivo has identified differences between naive B cells, GC B cells, memory B cells, plasmablasts, and plasma cells (Halliley et al., 2015; Minnich et al., 2016; Tarte et al., 2003), but does not allow delineation of the actual differentiation process and the various stages through which B cells transition to becoming ASCs. The development of single-cell sequencing techniques circumvents this problem and allows for the discovery of new cellular subsets and cells that are transitioning from one cellular differentiation state to another. To date, single-cell sequencing has been used, among others, to identify B cell subsets in the immune landscape of tumor-infiltrating lymphocytes (Chung et al., 2017), delineate malignant B cells in follicular lymphomas (Andor et al., 2019), and identify transiting bone marrow progenitor cell populations that commit to the B cell lineage (Alberti-Servera et al., 2017; Miyai et al., 2018). In addition, software to reconstruct B cell receptor sequences from single-cell RNA sequencing data has been developed in recent years (Canzar et al., 2016; Lindeman et al., 2018; Rizzetto et al., 2018; Upadhyay et al., 2018). In-depth characterization of the differentiation process of mature naive B cells into ASCs has been investigated by focusing on parts of this differentiation process, for example, GC LZ to DZ migration (Dominguez-Sola et al., 2015; Milpied et al., 2018; Sander et al., 2015). Although this contributed to new insights into GC reactions, it remains unclear how differentiation into plasma cells occurs and is regulated. More insight into this differentiation process might uncover targets to regulate B cell differentiation into ASC at an early stage and to intervene in the generation of undesired antibodies in disease. We previously established a minimalistic in vitro culture system that uniquely supports efficient in vitro differentiation of human naive B cells into CD27++CD38++ ASCs while also maintaining high cell numbers along the differentiation stages (Unger et al., 2021). Here, we applied in-depth single-cell transcriptomics, B cell receptor reconstruction and trajectory inference on the in vitro-generated ASCs and combined analyses with data from publicly available single-cell datasets of ex vivo obtained B cells and ASCs from different human tissues. Our data demonstrate that in vitro-generated ASCs are highly comparable to their ex vivo counterparts. The differentiation trajectories uncover a hitherto unknown pre-ASC cellular stage that is detected also ex vivo in lymphoid tissue with active ASC differentiation processes. For the first time, GC-like B cells are identified in vitro from human naive B cells and possibly progress through an alternative route of differentiation into memory B cells. In addition to known regulatory transcription factors and cell surface markers, potential novel transcriptional regulators and plasma membrane markers are identified that may control or, respectively, mark the process of human plasma cell differentiation. Results Single-cell transcriptomic analysis of in vitro differentiated antibody-secreting cells We previously described a minimalistic in vitro system that effectively differentiates human naive B cells into ASCs while also maintaining high cell numbers at the various stages of B cell differentiation (Unger et al., 2021). Briefly, human naive B cells (CD19+CD27-IgG-IgD+) are sorted from healthy donor peripheral blood mononucleated cells (PBMCs) and cultured on a feeder layer of human CD40L-expressing mouse fibroblasts and restimulated after 6 days. Cytokines typically expressed by follicular T cells (Tfh) (IL-21, IL-4) are added to mimic Tfh help for B cell differentiation. Restimulation with CD40 costimulation and Tfh cytokines in vitro was demonstrated to be needed to drive efficient ASC differentiation, in line with in vivo GC reactions (Figure 1a). Triggering of the IgM BCR with soluble anti-IgM antibodies or IgM-coated beads did not contribute to enhanced ASC differentiation in the minimalistic cultures (Unger et al., 2021). To assess the transition of naive B cells (CD27-CD38-) into ASCs (CD27++CD38++), cells from the four CD27/CD38 quadrants were sorted on day 11 and the expression of master regulators of B cells to ASC differentiation was analyzed (Figure 1b; Verstegen et al., 2021). CD27-CD38- cells gradually lose expression of B cell signature genes PAX5 and BACH2 upon acquisition of CD27 and/or CD38 expression with CD27++CD38++ cells showing the most downregulated expression of PAX5 and BACH2 and upregulation of ASC signature genes IRF4, XBP1, and PRDM1 (Figure 1c–g). In line with this, CRISPR-Cas9-mediated knockout of PRDM1 on day 3 strongly suppressed the formation of CD27+CD38+ cells on day 11 compared to control and CD19 knockout (Figure 1h). Figure 1 with 2 supplements see all Download asset Open asset Unbiased analysis of in vitro-induced human naive B cell differentiation by scRNA-seq reveals five transcriptionally distinct B cell clusters. (a) Overview of the scRNA-seq experiment in short human naive B cells was isolated and cultured with a human CD40L-expressing mouse fibroblast (3T3) cell line and recombinant IL-4. Secondary cultures were initiated on day 6 using 9:1 wild type (WT) 3T3 to CD40L-expressing 3T3 cells and IL-4/IL-21 as this was necessary to induce substantial differentiation into the CD27++CD38++ ASC subset. On day 11, 384 cells were single-cell sorted based on the expression of CD27 and CD38 and sequenced using the Smart-seq2 method. (b) Overview of the cellular stages and important transcription factors involved in B cells differentiation from naive to antibody-secreting cell (GC, germinal center; ASC, antibody-secreting cell). (c–g) Expression of PAX5 (c), BACH2 (d), IRF4 (e), XBP1 (f), and PRMD1 (g) mRNA in sorted populations was analyzed by qPCR and related to levels present in CD27-CD38- cells. Each data point represents the mean of an individual experiment (n = 3) with triplicate measurements. Mean values are represented as bars. (h) FACS analysis of surface CD19 levels (top left) and ASC differentiation (top right) in cultured primary human B cells expressing the indicated control or CD19-targeting RNP. FACS analysis of ASC differentiation (bottom left) and quantification (bottom right) in cultured primary human B cells expressing the indicated control or three different PRDM1-targeting RNPs. Representative of three independent experiments with triplicate measurements. (i) Uniform Manifold Approximation and Projection (UMAP) projection of single-cell transcriptomes of in vitro differentiated human naive B cells (276 high-quality cells). Each point represents one cell, and colors indicate graph-based cluster assignments. (j–o) UMAP projection as in (i) colored by the transcriptional regulators IRF4 (j), XBP1 (k) and PRMD1 (l), PAX5 (m), BACH2 (n), SPIB (o), which are important in B cell differentiation (left graph of each panel), along with corresponding distribution of average expression levels (ln(TPM+1)) across the B cell clusters (1, 2, and 3) and the ASC clusters (4 and 5) (right graph of each panel). Since the minimalistic in vitro system efficiently drives differentiation of primary human naive B cells into ASCs and follows the in vivo observed transitions of the master regulators of transcription, the dynamics of human naive B cell differentiation were investigated in detail by single-cell RNA sequencing. Cultured cells were single-cell sorted on day 11 based on the expression of CD27 and CD38 to obtain a faithful representation of cells at varying stages of B cell to ASC differentiation. Next, cells were processed for single-cell RNA sequencing using the SMARTseq2 method. After raw data processing and quality control, a total of 275 (out of 382 sorted cells) high-quality cells that together express 17,716 genes were included in the final dataset (Figure 1—figure supplement 1a–e). Cells were assigned to one of three phases of the cell cycle (G1, G2/M, S) as determined by scores calculated based on the expression of G2/M and S phase genes (Figure 1—figure supplement 1f). As the cell cycle is an essential component of GC reactions, data were analyzed with and without regressing cell-cycle heterogeneity. When the cell cycle was not regressed out, cells were clustered based on cell-cycle scores, and most differentially expressed genes were related to the cell cycle (Figure 1—figure supplement 1g). Cell-cycle heterogeneity was regressed out in the final dataset to primarily focus our analysis on factors influencing B cells differentiation apart from the cell cycle (Figure 1—figure supplement 1h and i). Unsupervised hierarchical clustering and visualization with Uniform Manifold Approximation and Projection (UMAP) identified five clusters of differentiating cells (Figure 1i, Figure 1—figure supplement 1j). Clusters 4 and 5 showed a clear ASC gene signature with a prominent expression of known ASC-specific genes IRF4, XBP1, and PRDM1 (Figure 1j–l). Of note, 99% of CD27+CD38+ double-positive sorted cells were represented in clusters 4 and 5 (Figure 1—figure supplement 2a and b). The other three clusters (clusters 1–3) still showed a pronounced B cell signature with higher expression of genes known to repress the ASCs state such as PAX5, BACH2, and SPIB (Figure 1m–o). Thus, upon 11 days of culture with CD40 costimulation and IL-21/IL-4, human naive B cells differentiate into distinct B cells and ASC clusters based on overall gene signatures. In vitro-generated ASCs are highly comparable with ex vivo obtained ASCs from different human tissues To further characterize in vitro-generated ASCs, transcribed immunoglobulin genes were reconstructed from the single-cell transcriptomic data using BraCeR, and transcription levels and properties of immunoglobulins genes were compared between B cell and ASC clusters. A functional B cell receptor heavy chain rearrangement was successfully reconstructed for 255 out of 276 cells analyzed. Expression of the reconstructed B cell receptor (BCR) genes in clusters 4 and 5 was up to 10 times higher compared to clusters 1–3, demonstrating a very high transcriptional activity of immunoglobulin genes in the ASC clusters (Figure 2a). Isotype analysis of the reconstructed BCR heavy chains revealed a homogeneous distribution of isotypes in the different clusters (Figure 2b, Figure 2—figure supplement 1a and b). When combining data for B cell and ASC clusters, the percentage of cells expressing immunoglobulin of the IgM isotype was lower in clusters 4 and 5 compared to clusters 1–3 (clusters 4 and 5: 36.6% vs. clusters 1–3: 50.3%; p-value<0.01), while the opposite was true for expression of immunoglobulins of the IgG1 isotype (cluster 4 and 5: 43.7% vs. cluster 1–3: 24.6%; p-value <0.001). In addition, 87% (226/255) of all the reconstructed BCR heavy genes were identical to the germline sequence, that is, not somatically hypermutated, independently of their cluster origin (Figure 2c). Thus, in vitro-generated ASCs express a high load of, mostly, IgG1 class-switched but no significant mutated immunoglobulin genes. Figure 2 with 4 supplements see all Download asset Open asset Identification of antibody-secreting cells with substantial gene expression overlap as compared to ex vivo-derived plasmablast/plasma cells among the in vitro-differentiated human naive B cells. (a) Violin plot of the reconstructed B cell receptor immunoglobulins gene expression as determined by BraCeR within each B cell cluster. (b) Stacked bars denotes the frequency of isotype analysis of the reconstructed immunoglobulin heavy chain within each cluster. (c) Stacked bars showing the percentage of cells with a given B cell receptor mutation count within each B cell cluster. (d–g) Volcano plot depicting significantly (adjusted p-value <0.05) down- and upregulated genes. (d) In vitro-generated antibody-secreting cells (clusters 4 and 5), ex vivo bone marrow-derived plasma cells (e), ex vivo peripheral blood-derived plasmablast (f), and ex vivo tonsil-derived plasmablast/plasma cells (g) were compared to each specific B cell counterparts. (h, i) UpSetR plot (h) and Venn diagram (i) depict the intersection among all upregulated genes identified in in vitro and ex vivo antibody-secreting cells. To assess how the transcriptional profile of the in vitro-generated ASCs was representative of the in vivo situation, the specific gene signature of the in vitro-generated ASCs was first identified by performing differential expression analysis in the ASC clusters 4 and 5 compared to the B cell clusters 1–3 (Figure 2d). A total of 3858 genes were differentially expressed in the ASC clusters, with 1406 overexpressed among which PRDM1 and XBP1 were top hits. Next, publicly available scRNA-seq datasets from bone marrow from the Human Cell Atlas Data Coordination Portal ‘Census of Immune Cells’ project (https://data.humancellatlas.org/explore/projects/cc95ff89-2e68-4a08-a234-480eca21ce79; Figure 2—figure supplement 2), human peripheral blood (Rizzetto et al., 2018; Figure 2—figure supplement 3), and tonsils (Attaf et al., 2020; Figure 2—figure supplement 4) were used to perform differential expression analysis in which each tissue-specific ASC cluster was compared to its tissue-specific B cell counterpart. A total of 7653, 1079, and 1860 genes were upregulated in ex vivo obtained ASCs from bone marrow, peripheral blood, and tonsils, respectively (Figure 2e–g). Analysis of the overlap between the identified upregulated genes selected in in vitro and ex vivo data revealed that out of the 1406 upregulated genes found in in vitro-generated ASCs, 1113 (79%) were shared with at least one ex vivo ASC population, of which 349 (25%) were shared with all populations (Figure 2h and i). The remaining 293 (21%) genes were uniquely upregulated in in vitro-generated ASCs, indicating that these genes might be linked to in vitro culture rather than be general for ASCs. Additionally, ex vivo ASC populations also possess unique differentially expressed genes that are not shared with other ex vivo or in vitro-generated ASC populations (Figure 2i). These findings indicate that in vivo ASCs carry specific, possibly tissue-related, gene signatures in addition to the ‘core ASC signature’ shared among ASCs. Taken together these data show that the in vitro-generated ASCs have highly similar transcriptional profiles compared to ex vivo-obtained ASCs. A novel B cell to ASC intermediate cellular stage is identified as a precursor of terminally differentiated ASCs In our dataset, cells with a prominent ASC gene signature were separated into two distinct transcriptional clusters. The cells in these clusters also show a slightly different expression of the ASCs phenotypic markers CD27 and CD38, with cells in cluster 5 being predominantly CD27++CD38++, while cluster 4 also includes CD27-CD38+ cells (Figure 1—figure supplement 2a). When performing clustering analysis on the dataset without cell-cycle regression, again cells from clusters 4 and 5 separated from the B cells clusters. Interestingly, the two former ASC clusters now separated into two new clusters, whereby the cell cycle stage became the major common denominator between the two clusters (Figure 1—figure supplement 1j). This shows that original clusters 4 and 5 contain both cells in the cell cycle and cells that are out of the cell cycle. To further investigate the differences between cells in clusters 4 and 5, the similarity and the enrichment score for ASC gene signature were analyzed for both clusters (Figure 3a and b). Cells in cluster 4 scored lower than cells in cluster 5 for both ASC similarity and enrichment, indicating a less dominant ASC gene signature in this cluster. Differential expression analysis between clusters 4 and 5 identified a total of 1076 differentially expressed genes, with 114 and 962 genes up- and downregulated in cluster 5, respectively (Figure 3c). Network analysis of Gene Set Enrichment Analysis (GSEA) results for the differentially expressed genes revealed a predominant enrichment for terms concerning cytoskeleton organization, including supramolecular fiber organization, immune cell receptor signaling, and metabolism of steroids (Figure 3d, Figure 3—figure supplement 1). Of note, among the enriched terms concerning cytoskeleton organization, we found ‘Arp2/3 complex-mediated actin nucleation,’ recently shown to be important in immune synapses formation (Bolger-Munro et al., 2019; Roper et al., 2019), BCR signaling and B cell activation, together with other terms involved in antigen uptake and processing (Figure 3—figure supplement 1). The majority of the processes indicated by the predominantly enriched terms were suppressed in cluster 5 compared to cluster 4, while specifically activated processes in cluster 5 were ‘protein N-linked glycosylation’ and ‘retrograde protein transport’ (Figure 3e). Thus, cells in cluster 4 are still active in antigen presentation and processing, BCR signaling, and fatty acid metabolism, while such processes are being shut down in cells from cluster 5 that conversely upregulate genes involved in protein glycosylation and protein transport. Figure 3 with 1 supplement see all Download asset Open asset Pre- and terminally differentiated antibody-secreting cells (ASCs) separated in different clusters. (a) Uniform Manifold Approximation and Projection (UMAP) projection colored by ASC similarity score (left), along with corresponding distribution of the similar score within each cluster (right). (b) Gene Set Enrichment Analysis (GSEA) enrichment plots for ASC gene signature in cells from clusters 4 (left) and 5 (right) compared to the other clusters (NES, normalized enrichment score; FDR, false discovery rate). (c) UMAP projection as in Figure 1i colored by clusters 4 and 5 membership (left) and volcano plot depicting significantly (adjusted p-value<0.05) down- and upregulated genes in cluster 5 compared to cluster 4. (d) Functional grouping network diagram of GSEA comparing clusters 4 and 5. The dot size and the dot color represent the number of genes in the pathway and the NES, respectively. (e) The top 3 most significantly differentially activated pathways as determined by GO enrichment analysis comparing clusters 4 and 5. The dot size and the dot color represent the number of genes in the pathway and the adjusted p-value, respectively. (f) Velocyto force field showing the average differentiation trajectories (velocity) for cells located in different parts of the UMAP plot. Arrow size conveys the strength of predicted directionality. (g) PAGA graph showing the connectivity between the clusters. Each node corresponds to each of the clusters identified using Seurat. The most probable path connecting the clusters is plotted with thicker edges. (h) UMAP projection of ex vivo B cells and ASC derived from bone marrow, peripheral blood, and tonsil colored by cluster 4 and 5 similarity score. To assess whether cluster 4 is a precursor population of cluster 5, we applied two algorithms for analyzing respectively the connectivity and the dynamics between clusters in scRNA-seq datasets, namely, partition-based graph abstraction (PAGA) (Wolf et al., 2019) and RNA velocity (Bergen et al., 2020). Projection of the RNA velocity vectors on the UMAP representation revealed an area of increased velocity in cluster 4 in the direction of cluster 5, indicating that cells in cluster 4 are rapidly transitioning toward the transcriptomic signature of cells in cluster 5 (Figure 3f). At the top-right extremity of cluster 5, where most of the non-cycling cells are located, no velocity vectors are projected. This implies that no subsequent cellular stage could be appointed, suggesting an endpoint of differentiation here. In line with this, results of PAGA showed increased connectivity among clusters 4 and 5 (Figure 3g). Interestingly, only cluster 4 and not cluster 5 was shown to be connected with all three B cell clusters, with cluster 3 being the most pronounced connection followed by cluster 1. This indicates that cluster 4 is more similar to the B cell clusters than cluster 5. To confirm that cluster 4 represents a previously overseen intermediate B cell to ASCs population, we obtain a set of hallmark genes for cluster 4 and tested them on the publicly available scRNA-seq dataset from bone marrow, tonsil, and peripheral blood (Figure 3h). In the bone marrow dataset, the highest similarity score with the cluster 4 hallmark genes was observed in the cells connecting naive B cells to plasma cells. Interestingly, the GC B cells in the tonsil dataset, both non-cycling LZ and cycling DZ GC B cells, showed the highest similarity score with cluster 4 cells, while in the peripheral blood dataset, few cells in the naive/memory cluster showed increased similarity. Of note, when performing the same analysis with the hallmark genes from cluster 5, the highest similarity was observed in each tissue-specific ASC population, as expected. Thus, taken together these results show that cluster 5 represents a terminally differentiated ASC population and that cluster 4 represents a novel B cell to ASC intermediate cellular stage, a precursor of terminally differentiated ASCs. This population can also be found ex vivo, mainly in lymphoid tissues with active ASC differentiation. Differentiating B cells transit through a GC-like and early memory phenotype To analyze the earlier stage of naive B cell to ASC differentiation, differential expression analysis among the three B cell clusters was performed to identify cluster-specific gene signatures (Figure 4a). Interestingly, genes that mark a GC B cell phenotype, such as IRF8, BCL6, CD22, and CD83, were among the top 30 differentially expressed genes in cluster 2 (Basso and Dalla-Favera, 2010; Meyer et al., 2021; Victora et al., 2012; Wang et al., 2019). When performing GSEA using hallmark gene sets on the differentially expressed genes in cluster 2, MYC (Calado et al., 2012; Dominguez-Sola et al., 2012) and E2F (Béguelin et al., 2017) targets, known to promote GC B cell differentiation, were the top 3 significantly enriched gene sets and, in fact, MYC was expressed at its highest in cluster 2 compared to the other clusters (Figure 4b and c). Finally, the similarity score with ex vivo-sorted human GC B cells was the highest in cluster 2, followed by cluster 3 (Figure 4d). Thus, we concluded that cells in cluster 2 represent GC-like B cells. Figure 4 Download asset Open asset Terminal differentiation into antibody-secreting cells (ASCs) starts from germinal center (GC)-like cells. (a) Top 30 differentially expressed genes identified in clusters 1–3. The size of the dot encodes the percentage of cells within a cluster, while the color encodes the average gene expression across all cells within a cluster. (b) Significantly enriched hallmark gene sets overlapping with the differentially expressed genes identified in cluster 2. (c) Uniform Manifold Approximation and Projection (UMAP) projection colored by expression of the transcription factors MYC (left), along with corresponding distribution of expression levels (ln(TPM+1)) within each cluster (right). (d) Corresponding distribution of the GC similarly score within each cluster (left) and Gene Set Enrichment Analysis (GSEA) enrichment plot (right) for GC B cell gene signature in cells from cluster 2 (NES, normalized enrichment score; FDR, false discovery rate). (e–k) UMAP projection colored by expression of the BANK1 (e), CCR6 (f), MZB1 (g), ITGAX (h), SIGLEC6 (i), FCRL5 (j), HHEX (k; left), along with corresponding distribution of expression levels (ln(TPM+1)) within each cluster (right). (l) The similarity score of ex vivo germinal center B cell (GC), memory B cell (MBC), and ASC gene sets ordered by UMAP1. For cells in cluster 3, no definite phenotypic signature could be appointed based on the cluster-specific gene signature. However, given the intermediate similarity with both GC B cells and ASC gene signatures (Figures 3a and 4b) and the higher connectivity indicated by PAGA with the pre-ASCs cluster 4 compared to the other B cell cluster, we hypothesize that cluster 3 might represent a post-GC population primed to become ASCs. Among the top differentially expressed genes of cluster 1, we found BANK1, recently identified as a marker of a pre-memory B cell stage (Holmes et al., 2020) t