The diversity of the plasmablast signature across species and experimental conditions: A meta‐analysis

Abstract

Antibody‐secreting cells (ASC) are divided into two principal subsets, including the long‐lived plasma cell (PC) subset residing in the bone marrow and the short‐lived subset, also called plasmablast (PB). PB are described as a proliferating subset circulating through the blood and ending its differentiation in tissues. Due to their inherent heterogeneity, the molecular signature of PB is not fully established. The purpose of this study was to decipher a specific PB signature in humans and mice through a comprehensive meta‐analysis of different data sets exploring the PB differentiation in both species and across different experimental conditions. The present study used recent analyses using whole RNA sequencing in prdm1‐GFP transgenic mice to define a reliable and accurate PB signature. Next, we performed similar analysis using current data sets obtained from human PB and PC. The PB‐specific signature is composed of 155 and 113 genes in mouse and human being, respectively. Although only nine genes are shared between the human and mice PB signature, the loss of B‐cell identity such as the down‐regulation of PAX5, MS4A1, (CD20) CD22 and IL‐4R is a conserved feature across species and across the different experimental conditions. Additionally, we observed that the IRF8 and IRF4 transcription factors have a specific dynamic range of expression in human PB. We thus demonstrated that IRF4/IRF8 intranuclear staining was useful to define PB in vivo and in vitro and able to discriminate between atypical PB populations and transient states.

The loss of the B‐cell identity during the late differentiation of B cells into plasmablast and plasma cells is an evolutionary conserved mechanism. There is a high heterogeneity within up‐regulated markers, in particular regarding the chemokine receptors and markers involved in the immune regulation. The differential expression of IRF4 and IRF8 is efficient to track the plasmablast in mice and in humans.

Abbreviations

ASC

antibody‐secreting cells

Bhlha15

basic helix–loop–helix protein 15

DEG

differentially expressed gene

ePB

ex vivo plasmablast

FC

fold change

FCS

fetal calf serum

FDR

false discovery rate

GEO

Gene Expression Omnibus

GO

gene ontology

GO‐BP

gene ontology biological process

iPB

in vitro‐generated plasmablast

LLPC

long‐lived plasma cells

PC

plasma cells

PB

plasmablast

PBMCs

peripheral blood mononuclear cells

RMA

robust multi‐array average

RE

reticulum endoplasmic

SC

signature score

ssGSEA

single sample gene set enrichment analysis

TF

transcription factor

UPR

unfolded response

INTRODUCTION

The production of antibodies protecting against viruses and other pathogens is the unique feature of antibody‐secreting cells (ASC), the ultimate end‐point of the B‐cell differentiation. ASC have been for decades considered as a homogeneous stage. However, experimental data accumulated in recent years from both rodents and humans demonstrated that ASC are a complex and heterogeneous subset [1]. This heterogeneity stands on the localization of the ASC compartment throughout the body, their ability to proliferate or not, their life span and also their ultimate function. Usually, ASC are divided into two principal subsets including the long‐lived plasma cell (LLPC) subset residing in the bone marrow and the short‐lived PC subset or plasmablast (PB) that may be proliferating and could be released in the blood and towards tissues after an immune reaction [2]. The differentiation process of B cells towards ASC is controlled by a network of transcriptional factors displaying a strict time sequence expression [3]. A core triad of transcription factors (TFs) guides this developmental programme involving IRF4 (encoded by IRF4), BLIMP1 (encoded by PRDM1) and XBP1 (encoded by XBP1) [4]. Among them, BLIMP1 is the critical regulator of the loss of the B‐cell identity by repressing BSAP (PAX5), PU.1 (SPIB), CD22 (CD22) and ID3 (ID3) but appears dispensable to maintain the global transcriptional identity of mature PC [5]. Moreover, the conditional deletion of PRDM1 in PC was not essential to their long‐term survival in bone marrow but impaired the function of immunoglobulin secretion.

However, less is known about the phenotype and the functional signature of the PB subset. The PB compartment has been described not only in peripheral blood of individuals after infection and vaccination [6, 7] but also in a numerous autoimmune and inflammatory diseases [8, 9, 10]. PB may encompass cells with different origins from either naive or recall of cross‐reactive memory B cells. The authors agree to define CD19⁺ CD38⁺⁺ CD27⁺⁺ as the most accurate representative subset of circulating human PB cells but with a high degree of heterogeneity [11, 12, 13, 14, 15]. In mice, CD138 (syndecan‐1), CD22, CD19 and TNFRSF13B (TACI) are broadly used to identify PC and PB but with more limited analogy with human PB [13, 16, 17, 18]. Although the mRNA signature of normal and malignant human PC in bone marrow [19, 20, 21] is well established, the molecular signature of PB appears to be highly dependent on the nature of the experimental set‐up. Different human PB mRNA profiles were described from 1‐ASC generated after in vitro B‐cell activation [22, 23, 24]; 2‐ASC sorted at steady state [11, 25, 26]; or 3‐ASC sorted from ex vivo subsets after infection and vaccination [6, 14, 20]. Although these accumulating data provided a high quality of information about human ASC, they challenged an overall view of the molecular identity of PB not only across experimental conditions but also across species. Furthermore, beyond the Ig production, the functional identity of PB appears more complex as recent studies have underlined that this subset could exert regulatory function on the immune response [27, 28].

To gain insight into the understanding of the molecular identity of the PB population, we performed a meta‐analysis of different data sets exploring the differentiation from mature B cell to PC in mice and humans. The present study attempted to use new recent analyses using whole RNA sequencing in Prdm1‐GFP transgenic mice to define an accurate PB signature based on the reliability and the stability of the differential gene variation. We therefore drew a molecular identity of PB in mice revealing new molecular aspects on the in vivo progression of B cells towards terminal differentiation. We next performed a similar analysis using six recent available data sets from human PB and PC including RNA sequencing and microarray data. We uncovered unique and conserved molecular features for PB and pre‐PB across experiments and species. Finally, we explored ex vivo and in vitro the protein‐related expression of some conserved and atypical markers obtained from the human PB signature. We thus underlined the robustness of using intracellular markers targeting IRF4 and IRF8 to identify the heterogeneity of human PB subsets.

MATERIALS AND METHODS

Public data

RNA sequencing or microarray data were obtained from the public functional genomic data repository: Gene Expression Omnibus (GEO) (Table 1). All data were published during or after 2015. The different cell subsets used (ex vivo and in vitro) are detailed in Figure [Link], [Link] for mouse and in Figure [Link], [Link] for human being. Raw data in ‘cel’ format from the GSE68878, GSE69033, GSE75534 and E‐MEXP‐3034/E‐MEXP‐2360 experiments were exported and analysed with the ‘oligo’ R package v1.44.0, and a robust multi‐array average (RMA) correction was used for the normalization. Probes were associated with their gene target using pd. huex.1.0.st.v2v3.14.1, pd. hugene.2.1. Stv3.14.1 and pd.hg. u133.plus.2v3.12.0 R packages. Normalized data and raw count (GSE60927) were analysed with iDEP platform [29], using DESeq2 or Limma R packages, respectively. Differentially expressed genes (DEGs) were obtained with a false discovery rate (FDR) of 5% and a fold change |FC|>2 for RNA‐seq and |FC|>1·5 for microarrays.

TABLE 1.

Description of data sets used in the meta‐analysis

Specie	ID	Sequencing platform	Reference genome	Aligner	Counting tool	Publishing date
RNA‐seq
Mouse	GSE60927	Illumina HiSeq 2500	GRCm37	Subread	FeatureCounts	2015
Mouse	GSE71698	Illumina HiSeq 2000	GRCm37	TopHat v.1.4.1	HTseq v.0.5.3	2015
Mouse	GSE77744	Illumina HiSeq 2500	GRCm37	TopHat v.1.4.1	HTseq v.0.5.3	2016
Mouse	GSE91040	AB 5500xl Genetic Analyzer	GRCm38	TopHat v.1.4.1	Cufflinks v.2.2.0	2017
Human	GSE81443	Illumina HiSeq 2500	Ensembl R72	TopHat	Partek	2016
Human	GSE107011	Illumina HiSeq 2000	GENCODE v.26	Kallisto	Tximport	2019

Specie	ID	Microarray	Scanner	Analysis package	Annotation source	Publishing date
Microarray
Human	GSE41208	Illumina HumanHT‐12 V4.0 expression beadchip	Illumina BeadScanner	Lumi	HUGO Gene Nomenclature Committee annotations	2012
Human	GSE75534	Affymetrix Human Gene 2.1 ST Array	–	Oligo v.1.44.0	pd. hugene.2.1. stv3.14.1	2016
Human	GSE68878/GSE69033	Affymetrix Human Exon 1.0 ST Array	–	Oligo v.1.44.0	pd. huex.1.0.st.v2v3.14.1	2016
Human	E‐MEXP‐3034/E‐MEXP‐2360	Affymetrix GeneChip Human Genome U133 Plus 2.0	–	Oligo v.1.44.0	pd.hg. u133.plus.2 v3.12.0	2011

RNA sequencing and sc‐RNA sequencing on in vitro‐differentiated B cells

Bulk RNA‐sequencing data from in vitro differentiation of human naive B cells obtained from three independent healthy blood donors (Etablissement Français du Sang, Rennes, France) were analysed. Cell culture conditions to differentiate naïve B cells into PB were as described in Hipp et al. [30]. In vitro‐generated PB (iPB) were sorted as CFSE^low CD20^low CD38⁺ at day 7 of culture to be compared with unstimulated naive B cells. Qualities of RNA‐sequencing libraries were evaluated by FastQC‐0.11.7‐MultiQC‐1.4.dev. FASTq was trimmed with Trimmomatic‐0.33/BBMap_38.05 and aligned with STAR 2.5.3a over GRCh38.91. Raw and normalized counts were calculated with FeatureCounts (subread‐1.6.0). The DEG analysis was performed on iDEP, and genes with FPKM ≤1 for 70% of samples were removed.

For single‐cell RNA sequencing, bone marrow PC (CD38^hi CD138⁺) and iPB were sorted by flow cytometry and captured in C1 IFC for mRNA seq (10–17 µm) (Fluidigm). The Seurat package was used for data filtering and DEG analysis. Cells with less than 1000 expressed genes and with more than 20% of mitochondrial genes were removed. Genes expressed in less than 3 cells and with a mean expression less than 20 reads were removed. RNA‐seq counts and sc‐RNA‐seq comparison data for the 113 genes analysed are in Table S5.

Meta‐analysis pipeline

Gene names in mouse and human data were harmonized before comparison, as detailed in Figure [Link], [Link]A,B. Briefly, gene name synonyms were obtained from R packages ‘org.Mm.eg.db’ v3.6.0 for mouse data sets or from ‘org.Hs.eg.db’ v3.6.0 for human data sets. ‘org.Mm.eg.db’ and ‘org.Hs.eg.db’ contain toorg.Mm.eg.dbdatabase and toorg.Hs.eg.dbdatabase, respectively. Alias2Symbol function from ‘Limma’ v3.36.5 converts identified gene synonyms into a unique gene name. If gene names are not described in the first used database, a second research was performed in the database from ‘HGNChelper’ v0.5.2 package thanks to CheckGeneSymbols function. Gene names not identified in all tested databases keep their original names.

Normalized data and raw count (GSE60927) were analysed with the iDEP platform [30], using DESeq2 or Limma R packages, respectively (Figure [Link], [Link]A,C). Raw counts data were normalized by DESeq2, and then, normalized counts were filtered removing genes with FPKM ≤1. Genes from microarray experiments were also filtered depending on the signal background. Finally, the DEGs were obtained with FDR of 5%, and a fold change |FC| > 2 for RNA‐seq and |FC| > 1·5 for microarrays.

To construct the signature, an original approach was designed to compare DEG from several technologies and analysis methods (Figure [Link], [Link]A,D). Briefly, for each DEG, the frequency of up‐regulation and down‐regulation was calculated across all comparisons. Then, if the up‐regulation frequency (FreqUp)> down‐regulation frequency (FreqDown), a corrected frequency (Corr Freq) was calculated in the following manner: FreqUp–FreqDown=Corr Freq. If FreqDown>FreqUp, Corr Freq=FreqDown –FreqUp. The median of fold change (medFC) was calculated using all comparisons. Finally, a signature score (SC) was obtained as follows: MedFC*Corr Freq= SC.

To define the murine signatures, genes were filtered using |medFC| ≥ 2, Corr Freq ≥0·5 and |SC| ≥1·5. To define the human signatures, genes were filtered using |medFC| ≥2, Corr Freq ≥0·5 and |SC| ≥0·7 due to the low resolution of the microarrays (Figure [Link], [Link]D). Genes related to light or heavy antibody chains were removed to gene lists.

Single sample gene set enrichment analysis (ssGSEA)

Single sample gene set enrichment analysis (ssGSEA) was performed using ssGSEAProjection module v9.0.10 on the GenePattern platform with default parameter. For each data set, results were normalized through z‐score. Mean and standard deviation were calculated using the correlation scores obtained across all samples from one data set. In Figure 2, we regrouped in ‘Activated B cell’ all samples including in vitro‐activated B cells with different stimuli (LPS, CD40, IL4); in ‘Mature B cell’ samples including follicular B cells, spleen IgD⁺ and spleen B220⁺ B cells and in ‘PC’ Plasma cells samples obtained from the spleen and bone marrow.

FIGURE 2.

The core of the plasmablast signature is shared with plasma cells. (a) Single sample gene set enrichment analysis (ssGSEA) correlation value for each sample or groups of samples with the plasmablast (PB) mouse signature (882 genes). Down‐regulated genes are in blue, and up‐regulated genes are in red. GCB, germinal centre B cell; MZB, marginal zone B cell; PB, plasmablast; PC, plasma cell; pre‐PB, pre‐plasmablast. Each group of samples is associated with its RNA‐sequencing ID. (b) Ratio of the ssGSEA correlation for down‐regulated genes and up‐regulated genes in the pre‐PB (N = 4), PB (N = 4) and PC (N = 2) murine molecular signatures. The Kruskal–Wallis test, ns: P‐value > 0·05, *P‐value ≤ 0·05, **P‐value ≤ 0·01, ***P‐value ≤ 0·001

Gene ontology analysis

Gene ontology biological process (GO‐BP) annotation of each gene was performed in DAVID database and GO‐BP, and the genes were grouped in global biological process based on MGI GO‐BP classification.

Cell isolation and Flow cytometry

Peripheral blood mononuclear cells (PBMCs) were isolated by density gradient centrifugation on lymphocyte separating medium, Pancoll human (PAN Biotech). CD19⁺ B cells were purified from human PBMCs using the REAlease^® CD19 MicroBead Kit (Miltenyi Biotec) with purity greater than 98%. Cells were cultured in 24‐well plates (Falcon) in complete medium (RPMI 1640 medium, Sigma‐Aldrich) supplemented with 10% of heat‐inactivated fetal calf serum (FCS) (BD, Biosciences), 4 mM l‐glutamine (Gibco) and penicillin (200 U/mL) and streptomycin (100 µg/mL). B cells were stimulated for 84 h (3, 5 days), in the presence of the AffiniPure Goat anti‐Human IgG+IgM (H + L) (2 µg/mL), the AffiniPure F(ab′)2 Fragment Goat Anti‐Human Serum IgA, α Chain Specific (2 µg/mL; Jackson ImmunoResearch Laboratories), CpG oligodeoxynucleotide (ODN‐2006) (0, 25 µM; InvivoGen), Recombinant Human Interleukin‐2 (rh IL‐2) (20 ng/mL, ImmunoTools) and Recombinant Human Interferon alpha 2a (rh IFN‐α2a) (10 µg/mL; ImmunoTools). Fresh PBMCs or stimulated B cells were stained with different combinations of fluorochrome‐conjugated antibodies. All antibodies are listed in Table S4. Intracellular staining for the transcription factors was carried out after cell permeabilization using cytofix/cytoperm permeabilization kit (BD Biosciences) according to the manufacturer's instructions. Stained cells were analysed using a Navios cytometer (Beckman Coulter). The phenotype of the cells was analysed using Kaluza Flow Cytometry Analysis and FlowJo software.

RESULTS

A blimp‐1‐dependent PB signature in mice across different experimental data sets

In the first part of the analysis, we used four sources of data representing so far, in mice, the most exhausting publicity available data sets using RNA sequencing for pre‐PB, PB and PC (Table 1, Table S3 and Figure [Link], [Link]). All the different studies were performed using the well‐established mouse model expressing a GFP‐reporter allele within the Prdm1 locus [31]. These data sets encompassed not only ex vivo‐sorted PB and PC based on GFP (prdm1) and CD138 expression but also in vitro‐generated pre‐PB and PB following 3–4 days of LPS or CD40 and IL‐4 stimulation. The kinetic progression of the ASC differentiation is representing by the progressive acquisition of BLIMP1 following by the up‐regulation of CD138 [32]. Therefore, we can distinguish in the spleen and lymph nodes, following LPS stimulation, between pre‐PB GFP⁺ CD138⁻, PB GFP⁺ CD138⁺ and LLPC in bone marrow CD28⁺ CD138⁺ GFP⁺. Our primary objective was to define a mouse PB signature reliable across the different experimental conditions. We used a pipeline analysis (see Material and methods) that allowed us to compose with different data sets (Figure [Link], [Link]A). Briefly, after gene name normalization between the different experiments (Figure [Link], [Link]B), we first performed comparisons in each data set of the PB transcriptome vs the other B‐cell subsets (referred to as early B cells) (Figure [Link], [Link]C). We chose to gather all other B‐cell subsets to exclude potent differential expressed genes that would have appeared due to the intrinsic nature of the different B‐cell subsets (marginal zone vs germinal centre cells as an example) and the different experimental conditions. We excluded PC from this analysis to keep the progressive dynamic of the gene network involved in the PB initiation programme. Next, from these 29 comparisons (Table S3), we realized the meta‐analysis generating a 12·793 gene signature (Figure [Link], [Link]). We filtered those genes using FC < −2 or >2, a corrected frequency ≥0·5 and a |signature score| (|SC|) ≥ 1·5 revealing a PB signature with 1068 genes (Table S1 and Figure [Link], [Link]D). As expected, the up‐regulation of immunoglobulin‐related gene transcription occurring at the PB stage represented up to 25% of the 704 up‐regulated genes, which were thus filtered out. Therefore, we proposed a mouse PB signature of 882 genes with 520 up‐regulated genes and 362 down‐regulated genes highly conserved between the different experimental conditions (Figure 1a, Table S1 and Figure [Link], [Link]E).

FIGURE 1.

The plasmablast signature includes 880 differential expressed genes compared with early B cells. (a) Pie chart of the number of up‐regulated and down‐regulated genes in murine plasmablasts (PB) compared with early B cells (including all B‐cell samples excepting plasma cells) based on the 882 gene signature. (b) Heatmaps of the signature score of murine genes according to their ontology and molecular function: cell surface receptor, reticulum endoplasmic (RE) stress and survival, transcription factor or immune signalling function. The genes mentioned in text are coloured in red for up‐regulated and in blue for down‐regulated genes. Row min and row max are minimal and maximal values for each row. (c) REVIGO representation of semantic association of gene ontology (GO) process significantly enriched in the murine molecular signature (FDR ≤ 0·05) (GO process enrichment test was performed with Gorilla)

The 882 gene PB signature delineated not only different groups of gene expression (Figure 1b), including a high number of genes involved in protein production, in transport and in the cellular cycle, but also a significant amount of surface cell receptors, transcription factors and genes involved in immune processes underlined in the gene ontology (GO) analysis (Figure 1c). As expected, the Prdm1 gene (SC = 4·7) stands among the most up‐regulated genes with several other transcription factors less described in the B‐cell differentiation process. Two prominent families of transcription factors emerged (SC > 4): the basic helix–loop–helix family bHLH and the zinc finger proteins. The class A basic helix–loop–helix protein 15 (Bhlha15) known as Mist1 is one of the most differentially expressed genes in PB (SC = 6·2), this protein is well established as a critical factor regulating secretory vesicle trafficking not only in zymogenic cells in the gut [33] but also in salivary secretory acinar cells [34]. Recently, Mist1 was reported as a ‘scaling’ factor programming the secretory architecture in cells independently of the cell lineage [35]. This transcription factor is linked to a network of different regulators (Xbp1, Ern1, Pycr1 and Derl3) involved in the resolution of secretory stress and the regulation of the unfolded response (UPR) [36]. This observation revealed that the ability to secrete protein is an early and stable feature acquired during the differentiation process that is not restricted to fully differentiated plasma cells. Only 55/362 down‐regulated genes have a SC < −3 and 7 have a SC < −4. The vast majority of these genes are primary B‐cell lineage regulators such as Pax5, Bach2, Spib, Ciita and genes involved in cell activation and signalling, including Cd22, Fcer2a, Tlr9, Id3 and Icosl. Interestingly, the down‐modulation of the B‐cell identity occurs along with the up‐regulation of atypical B‐cell markers mostly involved in the negative regulation of immune processes (Ctla4 (SC = 3·2), Tigit (SC = 4), Lag3 (SC = 2·6), Il10 (SC = 3·1), C4bp (SC = 3·8) and Cd9 (SC = 1·8)) [28] (Table S1).

The mouse PB signature is highly similar to PC

We next compared this PB signature with the different transcriptome data sets gathering activated, mature B‐cell subsets, pre‐PB and PC using single sample gene set enrichment analysis (ssGSEA) (Figure 2a,b). Mature and activated B cells are strongly segregated from the PB signature (r < 0·1), and we observed a slightly but significant difference with the Pre‐PB subset (GFP‐Prdm1 ⁺ CD138 ⁻), suggesting that additional cofactors to the Prdm1 gene are mandatory for the commitment of the PB differentiation. However, we did not observe a significant difference between the PB signature and individual data set coming from the fully differentiated PC (from spleen or bone marrow).

To gain insight into putative factors involved in the early PB commitment, we examined the genes differentially expressed between PB and pre‐PB (Table S1), we found a group of 98 genes with a FC > |2| and a frequency >|0·5| (Figure [Link], [Link]A). Using EnrichR data analysis tools [37], we thus identified Klf4, Suz12, Irf8, Rela and the chromatin‐modifying enzyme Ezh2 as putative coregulatory genes of Blimp1 conducting the first events towards the B‐cell differentiation (Figure [Link], [Link]B). In this regard, Ezh2 and Suz12 were recently described as a key controller of the B‐cell fate regulating epigenetic programmes during B‐cell division and differentiation [38, 39, 40].

Towards a mouse PB‐specific signature

Our precedent analysis revealed that the PB gene signature is shared with PC. To define a unique PB signature distinct from PC and other B‐cell subsets, we next conducted a similar meta‐analysis between the PB and PC transcriptomes applying FC > |2|, frequency > |0·5| and SC > |1·5| as filters. According to these standards, the Venn diagram of the 155 DEGs between both subsets defined four different clusters (Figure 3a, Table S1). Clusters A and B (121 genes) included DEGs with a progressive up‐ or down‐regulation between mature B cells and PC. This kinetic occurs with successive gaps of gene induction and gene repression, suggesting that progressive and distinct environmental inputs might conduct transcriptional dynamics towards the PC differentiation (Figure 3b). We observed the evolution of B‐cell trafficking receptor expression underlined by the progressive loss of Ccr6, Ccr7, Siglecg, Cd22, S1rp3 and Cxcr5 together with the up‐regulation of Ccr10, Cdh2, Ly6c1 and Ly6c2 but also the syndecan family protein Sdc1 that characterize profound modification in cell homing. Another interesting cluster of genes is involved in cytokine signalling and cell survival. These observations confirm previous experiments demonstrated that during the differentiation process PB and then PC acquire a specific localization that reprogrammes their survival‐dependent signals. The loss of Il4ra, Il21r, Cxcr5 and Ccr6 is consistent with the exit of PB from secondary lymphoid organs together with the acquisition of their independency of follicular helper T cells. The subsequent expression of Ccr10, kit and Tnfrsf17 suggests that the final differentiation process occurs in different targeted tissues such as the mucosal tissues or the bone marrow, allowing PC to reside in survival niches highly dependent on the Tnfrsf17 ligands Tnfsf13b (BAFF) and Tnfsf13 (APRIL) [41]. The two last clusters represent genes that might be considered as the specific functional gene identity of PB. The clusters C and D (34 genes) present a pattern of expression including up‐regulated or down‐regulated genes from early B cells to PB and then inversely expressed in PC (Figure 3b, Table S1). Using functional gene annotation, we observed that the unique PB identity is related to cell adhesion (Palpn, Hapln4, Nid2) and cell proliferation (Mki67, Amigo2) (Figure 3c). However, except Bcam (Cd239), no specific membrane receptor defined the murine PB population. Finally, we underlined for the first time that the bHLH family protein atonal homolog 8 (Atoh8) might represent a TF for the PB subset. Atoh8 is a TF implicated in the development of many tissues, including kidney, nervous system, retina and muscles (42), and performs lineage determination function by reprogramming cell specification [43]. However, its role in the B‐cell terminal differentiation is unknown.

FIGURE 3.

Definition of the murine plasmablast‐specific signature. (a) Workflow defining the different clusters according to their gene expression patterns. The 155 genes dispatched in the 4 clusters characterize the plasmablast (PB)‐specific gene expression compared with early B cells and plasma cells (PC), respectively. The clusters A and B described genes with a continuous up‐ or down‐regulation during the B‐cell differentiation (transitional expression level in PB). The clusters C and D described genes with a peak of up‐regulation or down‐regulation at the PB stage (specific expression level in PB). (b) Heatmaps of the median fold change (FC) of the differential expressed genes (DEGs) in PB versus early B cells and in PC versus PB. Each gene is classed in the four different clusters. The genes mentioned in text are coloured in red when up‐regulated and blue when down‐regulated genes in PB. Row min and row max are minimal and maximal values for each row. (c) Gene ontology analysis of the genes with transitional expression level in PB (clusters A and B) or with a specific expression level in PB (clusters C and D)

Altogether, these data suggest that a unique transcriptional gene network programme directs the identity of the PB population as a proliferative subset with tissue‐dependent migration and survival abilities.

The comparison of the different PB signatures across species reveals an evolutionary conserved differentiation process

In the second part of this work, we realized the same analysis using 6 human data sets (Table 1, Table S3 and Figure [Link], [Link]). The definition of human PB populations in healthy situations is highly variable and depends on the nature of the experiment, the physiological conditions (vaccination or not) and also the B‐cell sorting strategy. We thus gathered six data sets representing the diversity of these experimental conditions studying the generation of PB and PC in humans that included both ex vivo and in vitro studies. Because of the scarcity of studies using RNA sequencing, we used both microarray and NGS‐based experiments. Using the same approach described in Figures 1, 2, 3, we first generated a human PB signature that regroups the most stable gene expression across all the experimental conditions (Figure [Link], [Link]A and Table S2). This PB signature included 411 genes, with 129 genes that are shared with the mouse PB signature (Figure [Link], [Link]B). The core regulatory TF network driving the terminal B‐cell differentiation in mouse is conserved in human and included BHLHA15, PRDM1, MZB1, IRF4, ELL2 and XBP1 as up‐regulated genes and BACH2, PAX5, IRF8, BANK1 and SPIB as down‐regulated genes associated with a loss of mature B‐cell identity markers (CD22, MS4A1, CD40, CXCR5 and CD72). Interestingly, some other conserved genes have not been fully explored in human B cells and may have a decisive role in controlling PB and PC generation. In this regard, the involvement of the FKBP11 gene (encoded the FKBP19 protein) is intriguing. Although its role in the B‐cell differentiation process is not well understood, FKBP11 was not only found as a putative target of IRF4 initiating the pre‐PB stage [44] but also increased in lupus B cells inducing B‐cell tolerance breakdown [45]. Additionally, we underlined proteins involving in cell survival that may be valuable targets for controlling lymphoid neoplastic disorders such as PIM2, TRIB1, SEL1L and ELL2 [46, 47]. On the other hand, the human PB signature encompassed 283 genes mostly involved in cell activation (Figure [Link], [Link]C,D). The up‐regulation of CD27 and CD38 and the down‐regulation of CD24 and CD1c are consistent with the use of these markers for distinguishing differentiated B cells in humans. The absence of these proteins in mouse suggested that their expression may be uncoupled for the core transcriptional programme conducting the terminal B‐cell differentiation but linked to the experimental activation conditions. Finally, we noticed that unlike what has been observed in mouse, the negative checkpoint inhibitors Il10, Tigit, Lag3 and Ctla4 were not found in the human PB‐specific signature.

The human PB signature underlined the importance of specific‐metabolic pathways

We next generated the specific human PB signature by comparing this signature network with the different PC transcriptome (Figure 4a). The 113 stable DEGs clustered into four groups, as described in Figure 3. Both clusters A and D are under‐represented with only three genes AMPD1, RGCC and SEL (in Cluster A) involved mostly in metabolic and activation pathways and one gene in the cluster D: AKT3. Cluster B is the more consistent group of genes and is related to the regulation of the transcription, cell proliferation and survival (Figure 4b). Surprisingly, none of these genes are shared with the cluster B identified in mouse but are present in the murine PB signature, revealing important differences in the quantitative variation of these genes in both species. As an example, IRF8 has more important successive down‐regulation gaps during human PB differentiation than in mice. Cluster C involved the metabolic enzymes ADA, SGK1, GOT1, GGH, EDEM2 GALK2 (Figure 4a,b) underlining the importance of metabolic processes in the human terminal B‐cell differentiation pathway and may represent future and ongoing exciting tracks into the comprehension of the ASC generation [48, 49]. In cluster C, we found that PRDM15 and IRF4 have a unique expression peak at the human PB stage not observed in the murine PB signature.

FIGURE 4.

Definition of a human plasmablast‐specific signature conserved across experimental conditions. (a) Heatmaps of median fold change (FC) of the differential expressed genes (DEGs) of the 113 common genes between human being and mouse. All genes are dispatched in the 4 clusters characterize the plasmablast (PB)‐specific gene expression compared with early B cells and also with plasma cells (PC). The genes mentioned in text are coloured in red for up‐regulated and in blue for down‐regulated genes. Row min and row max are minimal and maximal values for each row. (b) Gene ontology analysis of the genes with a transitional expression level in PB (clusters A and B) or with a specific expression level in PB (clusters C and D). (c) Venn diagram between human and mouse PB‐specific signature. Transcription factor and membrane receptor genes are shown for each group of genes with their associated cluster

To confirm these results, we analysed bulk RNA‐seq (iPB versus naive B cells; Figure [Link], [Link]A and Table S5) and single‐cell RNA‐seq data (bone marrow PC versus iPB; Figure [Link], [Link]B and Table S5) generated from a model of in vitro differentiation of human naive B cells [30] (C. Delaloy, unpublished data). We showed that PB‐specific genes (observed in cluster C) such as PRDM15, IRF4, ADA, GGH, SKG1 and EDEM2 were up‐regulated in iPB, on the one hand in comparison with naive B cells and on the other hand versus bone marrow PC highlighting the soundness of our PB signature.

The loss of the B‐cell lineage‐related mRNA is the only feature that we observed conserved between humans and mice across all experimental variations (Figure 5c). We noticed that down‐regulation of the gene REL is a conserved event during the PB generation between mice and humans. This observation underlined that the NF‐kB pathway could represent an important crossroad between the different activation processes leading to PB differentiation. REL is a crucial regulator of the expression of BACH2 and may be implicated in B‐cell survival and B‐cell migration. PB seems to acquire an independency of the classical B‐cell transcriptional programme very early in the differentiation process underlined by the progressive loss of PAX5, IRF8 and REL. Our analyses have not revealed specific surface markers (belonging to cluster C or D) in the human PB signature confirming that typical and robust identity PB‐related proteins are highly variable and are challenged by the diversity of the experimental conditions. However, we observed that IRF4 and IRF8 progressive differential expression might represent a robust molecular identity to characterize the human terminal B‐cell differentiation.

FIGURE 5.

Experimental exploration of the human plasmablast gene signature in ex vivo and in vitro‐induced plasmablast. (a) Dot plots depicted ex vivo plasmablast (PB) from peripheral blood (ePB: dotted red line), or in vitro‐induced PB (iPB; red line) and ex vivo mature B cells (dotted blue line) and in vitro‐activated B cells (blue line) according to CD27 and CD38 expression. Below panels depicted expression of B‐cell‐related canonical surface markers. (b) Expression of up‐regulated markers identified in the mouse and human PB signature. (c) Representative dot plots of identification of B‐cell subsets according to IRF4 and IRF8 expression (N = 5 for ex vivo B cells and N = 13 for in vitro‐stimulated B cells). Differentiated B cells (IRF8^low IRF4⁺B cells are in red) and activated B cells (IRF8⁺ IRF4^low B cells are in blue) (left). For each subset, the expression of surface markers CD27 and CD38 was analysed (middle panel), as well as the expression of the transcription factor BLIMP1 (right panel)

Experimental exploration of the human PB gene signature

Finally, to assess the protein relevance of the PB human signature, we conducted a phenotypic analysis of human ex vivo PB (ePB) and in vitro‐generated PB (iPB) after 84 h of stimulation. Human PB are frequently characterized by the high expression of CD27 and CD38 compared with mature or activated B cells (Figure 5a and Figure [Link], [Link]A). The iPB and ePB shared the loss of the B‐cell identity underlined by the down‐regulation of CD19, CD20, CD22, CXCR5, CCR7 and CD40 (Figure 5a). Examining other up‐regulating markers described in the mouse PB signature, we observed high heterogeneity of protein expression with significant difference between ePB and iPB (Figure 5b). Surprisingly, CXCR4 expression was observed to be down‐regulated in ePB but stayed unchanged in iPB, which is in contrast with the requirement of this receptor for long‐term survival in bone marrow. Inversely, CD43 was highly expressed in ePB, whereas we observed a dual expression in culture with positive and negative subsets. CXCR3 has an intermediate variation level, although its expression is slightly increased in ePB, we showed a decrease in CXCR3 expression between activated and differentiated B cells (Figure 5b). CD166 is only expressed in vitro, and CD138, robustly used to discriminate ASC in mice, is expressed in a subset of ePB and induced at low level in vitro in our experimental conditions. We noticed that CD239 (BCAM), the unique membrane receptor identified in the mouse PB signature, is expressed on ePB but not in vitro. We did not observe the up‐regulation of CTLA‐4, TIGIT, CD83 and SIGLEC10 in none of the human PB subsets (Figure [Link], [Link]B). Interestingly, we found a low expression of LAG‐3 in ePB that could be induced in vitro but not restricted to PB. Overall, this first analysis revealed the complexity of defining a specific pattern of membrane receptor expression in PB. To overtake this limitation, we proposed to adopt a different population identification strategy using the molecular identity. Our precedent analysis has suggested that the progressive differential expression of IRF4 and IRF8 may recapitulate the true continuum related to the human B‐cell differentiation. We thus performed intracellular analysis of these two markers on ePB and in vitro‐stimulated B cells (Figure 5c). We thus confirmed that the progressive acquisition of IRF4 coincided with the down‐regulation of IRF8. Using supervised gating, we demonstrated that IRF8⁺ IRF4^low cells encompassed not only a majority of CD38^high CD27⁻‐activated B cells but also 20% of CD27⁺ CD38^low cells. The IRF8^low IRF4⁺ population showed a great enrichment in CD27^high CD38^high cells but also in activated B cells suggesting the accurate ability to track the progressive evolution of B cells towards PB differentiation. We confirmed that the IRF8^low IRF4⁺ population expresses BLIMP1 and represents differentiated B cells (Figure 5c). In the in vitro experiments, we observed a third population expressing IRF8 with an intermediate level of IRF4 expression, this population has two distinct phenotypes (CD27⁺ and CD27⁻) that may represent two distinct progenitors of PB previously defined as pre‐PB (Figure [Link], [Link]C). The subset of PB identified with IRF8^low IRF4⁺ phenotype was closed to the PB subset characterized in vitro by CD38^high CD20^low expression, as previously described [24] (Figure [Link], [Link]A), and includes the three populations (pop) of circulating ASC: (pop) 1 (CD19⁺ CD38⁺ CD138^–), (pop) 2 (CD19⁺ CD38^high CD138^–) and (pop) 3 (CD19⁺ CD38^high CD138⁺) recently described by Garimalla et al. [14] (Figure [Link], [Link]).

Altogether, this analysis emphasized that consistent phenotypic characterization of human PB may be challenged by the nature of the experiments. However, we underlined the existence of different PB populations and PB progenitors expressing different densities of CD27, CD38 or CD138 but sharing a unique molecular identity highlighted by the differential expression of IRF8 and IRF4.

DISCUSSION

The approach adopted in the present report was a horizontal genomic meta‐analysis based on the variation of differentially expressed genes across several comparisons. This qualitative approach allows performing meta‐analysis using data processed with different methods (e.g. RNA‐seq or microarray), instruments (e.g. Illumina 2500 or Illumina 2000), analysis pipeline (e.g. reference genome or aligner) and experimental protocols (Table 1). This strategy opens up a large possibility of usable datasets and increases the robustness of results. The ssGSEA has demonstrated that the PB signature effectively discriminates PB against early B cells and pre‐PB supporting the soundness of such approach. However, such analyses rely on the quality of the data sets and are expected to be more robust using RNA sequencing. At the time of the analysis, human PB and human PC data available were coming essentially from microarray experiments (4/6 data sets) leading to the inevitably loss of information [50, 51], but the pipeline presented here could be easily adapted to future data sets.

This meta‐analysis revealed the complexity of defining a PB signature. Our work underlined that the loss of the B‐cell identity during the late differentiation of B cells into PB and PC is an evolutionary conserved mechanism. The down‐regulation of canonical B‐cell markers CD19, CD20, CD22, CXCR5, CCR7 and CD40 mRNA is well conserved between mouse and human being, and is also found in human ex vivo PB and in vitro‐induced PB at the protein level. However, we observed a high heterogeneity within up‐regulated markers, in particular regarding the chemokine receptors and markers involved in the immune regulation (CTLA‐4, TIGIT or SIGLEC10) that we could not detect in human PB at steady state. These observations may underline that such protein expression is dependent on a particular physiological or pathological context. Indeed, PB harbouring immune‐regulatory properties have been related to controlling experimental autoimmune encephalomyelitis, the response to chemotherapy, but also increasing bacterial infection susceptibility [52, 53, 54, 55]. Similar observation was made regarding the expression of syndecan‐1 (CD138). Although this marker appears greatly associated with the murine PB signature (SC = 3·5), we did not find its expression significantly enriched in the human PB signature. Moreover, only a few parts of IRF8^low IRF4⁺ ePB expressed CD138. Recent studies underlined that CD138 expression is dependent on cell‐to‐cell contact with stroma and specific soluble derived factors [41] and that its ability to bind APRIL in bone marrow promotes PC survival [56]. However, CD138 expression seems not required to the human PB generation (at least in the early events).

Based on our results, we explored the differential expression of IRF4 and IRF8 as molecular markers to track the human PB. This double‐negative feedback loop was demonstrated in mice regulating the initial developmental bifurcation of LPS‐activated B cells [57]. IRF8 is a key actor controlling GC B‐cell selection while antagonizing the PB fate [58], whereas IRF4 is an obligate controller of the formation of the germinal centre and the PC differentiation [59]. Both of these TFs have revealed a similar pattern of expression in the human PB signature belonging to the two‐opposite clusters (B and C). We thus used and validated this approach for the first time in human identifying in vitro‐generated PB but also circulating PB based on their differential expression of IRF8 and IRF4. Moreover, this strategy revealed the existence of an intermediate phenotype during the in vitro differentiation with an increase in IRF4 expression while conservation of IRF8 that might represent the pre‐PB stage. Tracking more carefully the different trajectories leading to the generation of PB could be a crucial advantage in the understanding of infection and vaccination response [60, 61]. This identification strategy has the advantage to be well conserved between species, which facilitates the translation of findings to human system; unfortunately, it remains incompatible with cell sorting and functional experiments.

Altogether, our analysis attempts to draw a comprehensive overview of the PB and PC transcriptomes across species and across experimental conditions. This analysis provides exploratory resources and new insights not only into the tissue‐dependent expression of PB‐specific membrane markers but also into the molecular programme of human and mouse PB, offering reliable tools to follow those subsets in animal models and diseases.

CONFLICT OF INTEREST

The authors declare no competing interests.

AUTHOR CONTRIBUTIONS

A.G, M.B, L.L.P and S.H designed the study, performed experiments, analysed data and wrote the paper. M.M.C and C.D performed experiments and analysed data. O.M, D.C, C.J and J‐O. P discussed data and edited the manuscript.

Supporting information

Figure S1‐S8

Table S1

Table S2

Table S3

Table S4

Table S5

Supplementary Material