Plasmodium-specific atypical memory B cells are short-lived activated B cells

A subset of atypical memory B cells accumulates in malaria and several infections, autoimmune disorders and aging in both humans and mice. It has been suggested these cells are exhausted long-lived memory B cells, and their accumulation may contribute to poor acquisition of long-lasting immunity to certain chronic infections, such as malaria and HIV. Here, we generated an immunoglobulin heavy chain knock-in mouse with a BCR that recognizes MSP1 of the rodent malaria parasite, Plasmodium chabaudi. In combination with a mosquito-initiated P. chabaudi infection, we show that Plasmodium-specific atypical memory B cells are short-lived and disappear upon natural resolution of chronic infection. These cells show features of activation, proliferation, DNA replication, and plasmablasts. Our data demonstrate that Plasmodium-specific atypical memory B cells are not a subset of long-lived memory B cells, but rather short-lived activated cells, and part of a physiologic ongoing B-cell response.


Introduction
Atypical memory B cells (AMB) are an unusual B-cell subset detected in both mouse models and humans in the context of certain infections and autoimmune disorders, including HIV, HCV, tuberculosis, malaria, rheumatoid arthritis and systemic lupus erythematosus, and accumulated with age (Knox et al., 2017b;Naradikian et al., 2016a;Portugal et al., 2017;Rubtsov et al., 2017). In the context of infections, AMB were first described in HIV-viremic subjects, and termed tissue-like memory B cells, due to their similarity to an FCRL4-expressing memory B-cell subset found in human tonsillar tissues (Ehrhardt et al., 2005;Moir et al., 2008). In addition to FCRL4, these cells express relatively high levels of other potentially inhibitory receptors including CD22, CD85j, CD85k, LAIR-1, CD72, and PD-1, and show a profile of trafficking receptors including expression of CD11b, CD11c and CXCR3, consistent with migration to inflamed tissues. They are antigen-experienced classswitched B cells, which lack the expression of CD21 and the hallmark human memory B-cell marker CD27. Further studies demonstrated the expression of the transcription factor T-bet and the cytokine IFNg by these cells, also characteristic of Th1 CD4 + T cells (Knox et al., 2017b;Obeng-Adjei et al., 2017;Portugal et al., 2017). Due to their poor functional capacity upon in vitro re-stimulation with BCR ligands, AMB were characterized as dysfunctional B cells, and increased frequencies of these cells was proposed to be a consequence of B-cell exhaustion driven by chronic inflammation and stimulation, drawing parallels with T-cell exhaustion during chronic viral infections (Moir et al., 2008;Portugal et al., 2015;Sullivan et al., 2015). It has been hypothesized that expansion of AMB might contribute to the mechanisms driving autoimmune disorders and deficiencies in acquisition of immunity to chronic infections. However, due to lack of good tools and animal models to analyze antigen-specific atypical B cells in greater depth, many of these concepts remain speculative.
Several studies suggest that AMB might contribute to poor acquisition of long-term immunity to Plasmodium infection Portugal et al., 2015;Sullivan et al., 2015;Sullivan et al., 2016;Weiss et al., 2011;Weiss et al., 2009;Weiss et al., 2010). Indeed, some studies demonstrated that in the absence of constant re-exposure, Plasmodium-specific serum antibody levels rapidly wane, and full protection from clinical symptoms is lost, suggesting that B-cell memory is functionally impaired (Portugal et al., 2013). However, others have reported long-lasting maintenance of Plasmodium-specific antibodies and/or memory B cells in settings of differing malaria endemicity, and similar responses are also observed in mouse malaria models (Dorfman et al., 2005;Ndungu et al., 2009;Ndungu et al., 2013;Ndungu et al., 2012;Wipasa et al., 2010). Moreover, it has been shown that BCRs cloned from P. falciparum-specific AMB from malaria-exposed adults encode P. falciparum-specific IgG antibodies, which could contribute to P. falciparum-specific IgG antibodies in serum (Muellenbeck et al., 2013). These authors proposed that P. falciparum-specific AMB do not prevent, but rather contribute to the control of Plasmodium infection. These apparently contradictory results may reflect the fact that some studies were performed on the general peripheral blood B-cell pool and others focused on Plasmodium-specific B cells. In determining a role for these cells in a chronic infection it would be important to follow antigen-specific responses and to distinguish these from non-specific polyclonal B cell activation.
The study of the development of AMB is challenging and requires suitable mouse models, which allow for identification and isolation of antigen-specific B cells that exist often at very low frequency. Here, we generated a knock-in transgenic mouse with a high frequency of B cells specific to the 21 kDa C-terminal fragment of Plasmodium chabaudi Merozoite Surface Protein 1 (MSP1 21 ), to investigate memory B cells generated following mosquito-transmission of the rodent malaria, P. chabaudi. We identified a CD11b + CD11c + FCRL5 hi subset of MSP1 21 -specific B cells during the chronic infection with phenotypical and transcriptional features strikingly similar to those of human AMB. These AMB disappeared as the infection progressed, leaving a CD11b -CD11c -FCRL5 hi MSP1 21 -specific B-cell compartment with characteristics of long-lived classical memory B cells (B mem ) after the resolution of the infection. These short-lived MSP1 21 -specific AMB were also generated in response to immunization, suggesting they may be a normal but transient component of a B-cell response to antigen. In this chronic P. chabaudi infection, it appears that AMB require ongoing antigenic stimulation driven by the sub-patent infection to persist, and do not represent a true long-lived 'memory' B cell subset. Moreover, we show that generation of Plasmodium-specific AMB does not prevent the generation of Plasmodium-specific B mem , and does not prevent resolution of the infection.

Results
Generation of an immunoglobulin heavy chain knock-in transgenic mouse model to study Plasmodium-specific B cell responses To study Plasmodium-specific B cell responses in a rodent malaria model, we generated an Igh NIMP23/+ mouse strain on the C57BL/6J background (Materials and methods and Figure 1-figure supplement 1).
The Igh NIMP23/+ mice were healthy, with no unusual behavioral or physical characteristics. There were no alterations in total cellularity, pro-B, pre-B, immature B, mature B, total B220 + CD19 + B cells, and plasma cells in the bone marrow ( Figure 1A-B), and no alterations in number of T1, T2, T3, follicular, marginal zone, germinal center B cells, plasmablasts, plasma cells, and total cellularity in the spleen of Igh NIMP23/+ mice ( Figure 1C-D) (Sen et al., 1990;Young et al., 1994). Importantly, The Igh NIMP23/+ mice had a greatly increased frequency of B cells specific for MSP1 21 (approximately 60% of the total B-cell compartment), as demonstrated by flow cytometry analysis of splenocytes with a MSP1 21 fluorescent probe consisting of biotinylated recombinant MSP1 21 loaded on streptavidin-PE ( Figure 1E-F). Thus, in this model, a recombinant light chain is not required to bring about specificity. This suggest that most endogenous light chains will pair with the NIMP23 heavy chain to generate a BCR with detectable binding to MSP1 21 .

Increase in Plasmodium-specific B cells after mosquito transmission of P. chabaudi
To investigate B cells in P. chabaudi infections, which last several weeks, and to avoid potential problems with activation arising from very high frequencies of MSP1-specific B cells, we reduced the precursor frequency of MSP1 21 -specific B cells to match the natural level expected for antigen-specific B cells more closely, yet still readily detectable by flow cytometry. We generated mixed bone marrow (BM) chimeras by adoptively transferring a mixture of 10% bone marrow from either Igh NIMP23/+ or Igh +/+ mice (CD45.2 + ) together with 90% bone marrow from C57BL/6.SJL-Ptprc a mice (CD45.1 + ) into sub-lethally irradiated Rag2 -/-.C57BL/6.SJL-Ptprc a mice (CD45.1 + ) to generate NIMP23fiRag2 -/and WTfiRag2 -/bone marrow chimeric mice respectively ( Infection of C57BL/6J wt mice with P. chabaudi by mosquito bite gives rise to a short (48 hr) preerythrocytic infection, followed by an acute blood parasitemia peaking approximately 10d posttransmission. Thereafter, the infection is rapidly controlled, reaching very low parasitemias by 15d post-transmission, with a subsequent prolonged (~90 d), but low-level chronic infection before parasite elimination (Brugat et al., 2017;Spence et al., 2013). NIMP23fiRag2 -/mice infected with P. chabaudi by mosquito bite, showed a similar course of parasitemia to that of control WTfiRag2 -/mice (Figure 1-figure supplement 2F), and C57BL/6J wt mice (Brugat et al., 2017;Spence et al., 2013;Spence et al., 2012). Importantly, the MSP1 21 -specific Igh NIMP23/+ B cells (CD45.2 + MSP1 21 + ) in NIMP23fiRag2 -/chimeras showed a robust response to the infection, as demonstrated by a dramatic increase in the proportions and numbers of GL-7 + CD38 lo germinal centers (GC) and IgG2b + -IgDclass-switched B cells in the spleen at 35 days post-infection (dpi) (Figure 1-figure supplement 2G-H). Thus, we have generated a mouse model with detectable numbers of functional MSP1 21 -specific B cells capable of responding to P. chabaudi infection.
We detected an increased number of cells in a distinct CD11b + CD11c + MSP1 21 -specific B-cell subset at 28-35dpi, in the chronic phase of P. chabaudi infection (Figure 2A-B). This subset showed several AMB characteristics, including high expression of FCRL5 and low expression of CD21 and IgD ( Figure 2C-E). In addition, the CD11b + CD11c + MSP1 21 -specific B-cell subset was enriched with cells expressing CD80 and CD273 ( Figure 2C-E).
We then explored whether this CD11b + CD11c + MSP1 21 -specific B cell subset was detected during the memory phase,that is after resolution of the infection. As it takes up to 90 days for a bloodstage P. chabaudi infection to be eliminated from C57BL/6J mice (Achtman et al., 2007;Spence et al., 2013), we measured these responses from 155dpi onwards. Unexpectedly, the numbers of CD11b + CD11c + MSP1 21 -specific B cells were not significantly higher than background level (Figure 2A-B).
These data demonstrate that a mosquito-borne infection with P. chabaudi generates Plasmodium-specific B cells resembling human AMB. However, these cells do not persist and are not detected above background level after parasite clearance.
Transcriptome analysis confirms the AMB nature of CD11b + CD11c + MSP1 21 -specific B cells, and reveals a plasmablast-like signature for this subset To gain a better understanding of the identity of the CD11b + CD11c + Plasmodium-specific B cell subset, we isolated both CD11b + CD11c + and CD11b -CD11c -MSP1 21 -specific B cells from spleens of P. chabaudi-infected NIMP23fiRag2 -/mice (35dpi) (Figure 3-figure supplement 1), and MSP1 21specific B cells from the spleen of naïve NIMP23fiRag2 -/mice (Figure 2A), by flow cytometric sorting, and performed an mRNAseq transcriptional analysis on the three populations.
We ran a Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) with a gene list ranked according to their differential expression between MSP1 21 -specific CD11b + CD11c + AMB sorted from infected mice and MSP1 21 -specific B cells sorted from naïve mice, using gene sets a priori obtained from Reactome (Fabregat et al., 2018). Among the gene sets yielding the top 50 significant (fdr <0.001) highest normalized enrichment score (NES) we obtained gene sets corresponding to cell cycle, DNA replication, generation/consumption of energy, regulation of apoptosis, activation of NF-kB on B cells, and downstream signaling events of the BCR (Supplementary file 2 and Figure 3-figure supplement 2). These data further corroborate the activated and proliferative nature of MSP1 21 -specific CD11b + CD11c + AMB.
Taken together, these data demonstrate that CD11b + CD11c + MSP1 21 -specific mouse AMB present during the chronic phase of P. chabaudi infection are very similar to human AMB described in several chronic infections. In addition, this B-cell subset shows features of activation, proliferation,   DNA replication and plasmablasts, resembling previous observations in human AMB (Muellenbeck et al., 2013).

Generation of Plasmodium-specific AMB in response to immunization
The occurrence of CD11b + CD11c + AMB might be a consequence of aberrant B-cell activation driven exclusively by certain pathogens. Alternatively, they might be part of a normal B-cell response, which is exacerbated by the persistent nature of certain infections. To test whether CD11b + CD11c + FCRL5 + AMB could be generated in the absence of persistent infection, we immunized mice with MSP1 21 . A previous report had demonstrated the presence of CD11b + CD11c + Tbet + B-cells 24 hr post-immunization with R848, a TLR7/8 ligand (Rubtsova et al., 2013). Therefore, we immunized Igh NIMP23/+ mice with R848 together with the antigen MSP1 21 and looked for the appearance of MSP1 21 -specific CD11b + CD11c + FCRL5 + atypical B cells. We observed substantial numbers of MSP1 21 -specific CD11b + CD11c + B cells in the spleens of Igh NIMP23/+ mice 24 hr post-immunization ( Figure 4A). These cells expressed increased levels of both FCRL5 and CD80 ( Figure 4B-C) and did not display GC characteristics ( Figure 4D), similar to the MSP1 21 -specific CD11b + CD11c + FCRL5 + atypical B cells generated following Plasmodium infection. The MSP1 21 -specific CD11b + CD11c + B cells observed Figure 4. Generation of splenic MSP1 21 -specific CD11b + CD11c + AMB in response to immunization. (A) Flow cytometry showing differential expression of CD11b and CD11c on splenic MSP1 21 -specific B cells from Igh NIMP23/+ mice before immunization (day 0) and at days 1 and 3 post-immunization with R848 and MSP1 21 . (B) Flow cytometry showing expression of FCRL5 and CD80 on different subsets of splenic MSP1 21 -specific B cells from Igh NIMP23/+ defined based on CD11b and CD11c expression at day one post-immunization and naïve mice. (C) Geometric MFI of FCRL5 and CD80 expression on different subsets of splenic MSP1 21 -specific B cells from Igh NIMP23/+ defined based on CD11b and CD11c expression at day one post-immunization. Two-way ANOVA vs CD11b -CD11csubset. ****p<0.0001. (D) Flow cytometry of CD38 vs GL-7 (GC markers) on CD11b + CD11c + MSP1 21 -specific B cells from Igh NIMP23/+ at day one post-immunization. (E) Numbers of splenic CD11b + CD11c + MSP1 21 -specific B cells from Igh NIMP23/+ during the course of immunization. Kruskal-Wallis test compared to day 0. **p<0.01. Error bars are SEM. Data pooled from three independent experiments with 3-5 mice per group. DOI: https://doi.org/10.7554/eLife.39800.009 after immunization appeared only transiently, as they could no longer be detected at 3 and 7d postimmunization ( Figure 4A and E).
These data demonstrate that MSP1 21 -specific CD11b + CD11c + AMB with no functional characteristics of memory B cells can be generated independently of the infection and the presence of the pathogen, and that they are short-lived cells.
Plasmodium-specific CD80 + CD273 + B mem are generated and persist after resolution of P. chabaudi infection Identification of mouse B mem by flow cytometry originally relied on detecting B cells that had undergone Ig class-switching from IgM to IgG, and that did not express GC markers (i.e. IgG + CD38 hi GL-7 lo ) (Lalor et al., 1992;Ridderstad and Tarlinton, 1998). More recently, this set of markers has been extended to include CD80, CD273 (PD-L2) and CD73, with CD273 and CD80 being the most useful to discriminate memory from naïve B cells (Tomayko et al., 2010;Zuccarino-Catania et al., 2014). In combination, these markers allow the identification of different subsets of switched as well as non-class switched (i.e. IgM/D + ) B mem . Therefore, we used cell surface expression of CD80 and CD273 on MSP1 21 -specific B cells to identify B mem during and after resolution of P. chabaudi infection.
These data show that, in contrast to the transient AMB, splenic CD80 + CD273 + and CD80 + CD273class-switched and non-class-switched MSP1 21 -specific B mem persist after resolution of P. chabaudi infection.

Plasmodium-specific B mem express high levels of FCRL5
As discussed above, no single marker has been described so far that can identify all mouse B mem subsets. Surprisingly, we observed that after resolution of infection (155-170dpi), MSP1 21 -specific B cells expressing different combinations of CD80 and CD273 (CD80 + CD273 + , CD80 -CD273 + or CD80 + CD273 -) all expressed very high levels of FCRL5, in contrast to CD80 -CD273 -MSP1 21 -specific B cells that express no memory markers at this stage ( Figure 6A). This suggests that FCRL5 might be a marker for all B mem . In order to confirm this, we used unsupervised methods to analyze our multiparameter flow cytometry data. We used PhenoGraph and t-SNE within the Cytofkit package (Materials and methods, Chen et al 2016) to analyze MSP1 21 -specific B cells based on the expression of FCRL5, CD38, IgD, CD80 and CD273 on these cells, as determined by flow cytometry (Figure 6A and B). The analysis identified six clusters of cells with memory characteristics displaying high expression of CD38, CD80 and/or CD273, and variable expression of IgD, all of which expressed high levels of FCRL5 ( Figure 6C: clusters identified with purple arrows). We then used Isomap, (Cytofkit package) to infer the relatedness between those cell subsets identified by PhenoGraph. This confirmed high similarities between the cell clusters expressing high levels of FCRL5 with the clusters expressing high levels of the memory markers CD80, CD273 and CD38 ( Figure 6D).
To confirm the memory identity of MSP1 21 -specific FCRL5 hi B-cells detected after resolution of the infection, we isolated MSP1 21 -specific B cells expressing either high levels of FCRL5 or not expressing FCRL5 (i.e. FCRL5 hi and FCRL5 -MSP1 21 -specific B cells) from the spleen of P. chabaudiinfected NIMP23fiRag2 -/mice (155dpi) (Figure 6-figure supplement 1), and MSP1 21 -specific B cells from the spleen of naïve NIMP23fiRag2 -/mice, by flow cytometric sorting, and performed mRNAseq analysis on these three sorted cell populations ( Figure 6E). As expected, the MSP1 21 -specific FCRL5 hi B cell subset showed high expression of genes encoding the hallmark memory B cell markers Cd38, Cd80, Cd86, Nt5e (CD73) and Pdcd1lg2 (CD273), when compared with either MSP1 21 -specific FCRL5 -B cells sorted at the same time or MSP1 21 -specific B cells sorted from naïve mice ( Figure 6E). Moreover, the MSP1 21 -specific FCRL5 hi B cells sorted after resolution of the infection upregulated the anti-apoptotic Bcl2 gene, which is an additional hallmark characteristic of memory B cells ( Figure 6E). Importantly, FCRL5 also identified CD80 + and CD273 + MSP1 21 -specific B mem subsets generated following immunization with a model antigen ( Figure 6-figure supplement 2). Thus, after resolution of the infection, high expression of FCRL5 identifies P. chabaudi-specific B mem .

Plasmodium-specific AMB are a distinct short-lived activated B cell subset
After identifying and sorting MSP1 21 -specific AMB during chronic P. chabaudi infection, and MSP1 21 -specific B mem after resolution of the infection, we then compared the transcriptome of these two B-cell subsets. Principal component analysis (PCA) demonstrated a strikingly distinct transcriptome of MSP1 21 -specific AMB from that of MSP1 21 -specific B mem , as well as all other MSP1 21 -specific B-cell subsets sorted in this study ( Figure 7A). The MSP1 21 -specific AMB sorted at 35dpi formed a separated cluster at the extreme right of the PC1 axis of the PCA plot, which accounts for the majority of the variance ( Figure 7A). All the other subsets [including MSP1 21 -specific CD11b -CD11c -B-cells sorted from the same mice and at the same day post-infection as the MSP1 21 -specific AMB (i.e. 35dpi)] clustered on the left of the PC1 axis, and showed differences mostly along the PC2 axis of the PCA plot, which accounts for only 10% of the variance ( Figure 7A). Interestingly, MSP1 21 -specific CD11b -CD11cand MSP1 21 -specific B mem clustered on opposite sides of the MSP1 21 -specific naïve B-cell subset along the PC2 axis ( Figure 7A), which suggests that the MSP1 21 -specific B mem more closely resemble MSP1 21 -specific naïve B cells than MSP1 21 -specific CD11b -CD11c -B cells sorted at 35dpi. MSP1 21 -specific AMB sorted during chronic P. chabaudi infection, and MSP1 21 -specific B mem sorted after resolution of the infection shared the expression of a series of mouse memory markers, including Cd80, Fcrl5, Nt5e (CD73), and Cd86 ( Figure 7B). However, these two subsets showed differences in the expression pattern of anti-and pro-apoptotic genes ( Figure 7C). While MSP1 21 -specific B mem from after infection resolution showed the highest levels of expression of the antiapoptotic Bcl2 gene, MSP1 21 -specific AMB sorted during chronic P. chabaudi infection showed the lowest levels of expression of this hallmark anti-apoptotic gene ( Figure 7C). In contrast to MSP1 21specific B mem , MSP1 21 -specific AMB expressed high levels of the pro-apoptotic genes Bad, Bax, Fas and Fasl ( Figure 7C). Interestingly, MSP1 21 -specific AMB expressed very high levels of class-switched immunoglobulins, including Igha, Ighg1, Ighg2b, Ighg2c and Ighg3 ( Figure 7D). Finally, MSP1 21 -specific AMB highly expressed Cd11b, Cd11c, Tbx21, Ifng and Pdcd1 ( Figure 7E), all hallmarks of human AMB, as well as Mki67, indicative of active cell division, as previously shown in human AMB.
These data demonstrate that AMB and B mem share the expression of memory markers. However, they show striking differences in the expression of pro-and anti-apoptotic genes, immunoglobulins genes, and cell proliferation genes. The increased expression of Mki67, pro-apoptotic genes and class-switched immunoglobulins in AMB suggests that they resemble activated B cells. By contrast, B mem express much less Mki67 (similar to naïve cells) and present an anti-apoptotic gene expression pattern, consistent with being long-lived quiescent B cells.
Mouse FCRL5 has been shown to be expressed on marginal zone (MZ) and B1 B cells (Won et al., 2006). Therefore, it is possible that the AMB identified in this study might represent a specific subset of either MZ or B1 B cells. As discussed elsewhere (Baumgarth, 2011;Garraud et al., 2012;Pillai et al., 2005;Zouali, 2011), B1 and MZ B cells are CD1d mid/hi , CD9 + , IgM hi and CD23 -. B1 are further CD43 + , B220 lo , and may (B1a) or may not (B1b) express CD5. MZ are further characterised by CD22 hi , CD21/CR2 hi , and the expression of the lysophospholipid sphingosine-1 phosphate receptor S1P 1 , and the lineage master regulator Notch2. Indeed, MSP1 21 -specific CD11b + CD11c + AMB showed high expression of some markers associated with B1 and/or MZ, including CD5, CD9 and CD43. However, these markers have also been shown to be highly expressed by activated B2 B cells and plasma cells (Baumgarth, 2011). Studying all these canonical markers, we found no other similarities between B1/MZ and MSP1 21 -specific CD11b + CD11c + AMB to suggest a relationship between these subsets ( Figure 8A). This was further corroborated by flow cytometry analysis ( Figure 8B). Moreover, unlike MZ and B1 B cells, MSP1 21 -specific AMB were class-switched and expressed very high levels of IgG ( Figure 7D). To further characterize the identity of MSP1 21 -specific CD11b + CD11c + AMB, we explored if these cells resemble GC B cells. Flow cytometry analysis demonstrated that MSP1 21 -specific AMB did not display a GC phenotype (i.e. CD38 hi GL-7 lo ) ( Figure 8C) while the MSP1 21 -specific CD38 lo GL-7 hi GC B cells showed low expression of both CD11b and CD11c. Thus, MSP1 21 -specific AMB are not GC B cells. All together, these data further show that P. chabaudi-specific AMB represent a distinct subset of short-lived activated B cells.

Discussion
Similar to other chronic infections [e.g. HIV, HCV and Mycobacterium tuberculosis (Knox et al., 2017b;Portugal et al., 2017)], Plasmodium infection, the cause of malaria, leads to an increase in the frequency of AMB [originally termed tissue-like memory B cells (Moir et al., 2008)] in peripheral blood from P. falciparum-exposed subjects Portugal et al., 2015;Sullivan et al., 2016;Sullivan et al., 2015;Weiss et al., 2011;Weiss et al., 2010;Weiss et al., 2009). However, mouse models to investigate Plasmodium-specific AMB are lacking. Here we have generated an IgH knock-in transgenic mouse strain to study the generation of Plasmodium-specific AMB in a Plasmodium chabaudi infection. We demonstrate the generation of P. chabaudi-specific AMB in response to blood-stage mosquito-transmitted chronic P. chabaudi infection, and show the short-lived nature of these cells. P. chabaudi infections in mice present some outstanding characteristics for the study of AMB; a chronic phase which allows investigation of the impact of constant immune activation driven by persistent subpatent parasitemia, followed by a clearance phase which allows the study of immune responses after the infection is naturally resolved. Thus, it is possible to study both exhaustion driven by chronic immune activation, and memory immune responses which remain after P. chabaudi elimination.
AMB resemble age-associated B cells (ABC) which accumulate with age as well as in autoimmunity, and were also proposed to be a subset of long-lived memory B cells (Naradikian et al., 2016a;Portugal et al., 2017;Rubtsov et al., 2017). Expression of T-bet, CD11c and CXCR3 are shared by AMB, tissue-like memory B cells, and ABC (Knox et al., 2017b;Naradikian et al., 2016a). Moreover, similar to ABC, expansion of human AMB associated with malaria is driven by IFNg . IL-21, which is highly expressed by follicular helper T cells in response to Plasmodium infection (Carpio et al., 2015;Obeng-Adjei et al., 2015;Pérez-Mazliah et al., 2015), also directly promotes T-bet expression in B cells in the context of TLR engagement (Naradikian et al., 2016b). Taken together, these data strongly suggest that this is the same T-bet + B-cell subset, which accumulates with time due to repetitive antigenic exposure. In agreement with previous data (Rubtsova et al., 2013), we show here that immunization with MSP1 21 and R848, a TLR7/8 ligand, promotes a robust but short-lived CD11b + CD11c + P. chabaudi-specific AMB response. T-bet + atypical B cells are critical to eradicate various murine viral infections (Barnett et al., 2016;Rubtsova et al., 2013), and a recent study showed that yellow fever and vaccinia vaccinations of humans stimulated an acute T-bet + B-cell response and suggested that these T-bet + B-cell population may function as an early responder during acute viral infections (Knox et al., 2017a). Thus, T-bet + B cells, even in the context of malaria, are likely to be a normal component of the immune compartment that becomes activated and expands, most probably in response to BCR, endosomal TLR, and IFNg or IL-21 stimulation. Moreover, a recent study shows a TLR9/IFNg-dependent activation of autoreactive T-bet + CD11c + atypical B cells in response to P. yoelii 17XNL infection in mice genes representing memory B-cell markers, anti and pro-apoptotic genes, immunoglobulins and atypical memory B-cell markers, respectively, for all five groups described in (A). Each bar represents an individual mouse. Data generated with five mice per group. DOI: https://doi.org/10.7554/eLife.39800.014 Figure 8. Analysis of MZ, B1 and GC B cell characteristics on MSP1 21 -specific AMB. (A) MSP1 21 -specific CD11b + CD11c + (AMB) and CD11b -CD11c -B cells were flow cytometry sorted from the spleen of NIMP23fiRag2 -/chimeric mice at 35dpi; MSP1 21 -specific B cells were flow cytometry sorted from the spleen of naïve NIMP23fiRag2 -/-, and these three B cell populations were submitted to mRNAseq analysis. The heat map displays level of expression of selected individual genes known to be up or downregulated on either MZ, B1 B cells or both. Each column represents an individual mouse. (B) Flow cytometry analysis of surface markers of either MZ, B1 B cells or both, on MSP1 21 -specific CD11b + CD11c + (AMB) (blue) and MSP1 21specific IgM hi (red) B cells from the spleen of Igh NIMP23/+ mice at 20dpi. (C) Flow cytometry analysis of GC markers on MSP1 21 -specific CD11b -CD11c -(non-AMB, left) and CD11b + CD11c + (AMB, right) B cells from the spleen of Igh NIMP23/+ mice at 20dpi. (D) Flow cytometry analysis showing the expression of CD11b and CD11c on MSP1 21 -specific CD11b + CD11c + (AMB, blue) compared to GC (CD38 lo GL-7 hi , red) B cells from the spleen of Igh NIMP23/+ mice. Data generated with 5-6 mice per group. DOI: https://doi.org/10.7554/eLife.39800.015 (Rivera-Correa et al., 2017). However, whether these cells remained part of the long-lived memory B-cell pool after resolution of the infection was not explored.
Here, we show that P. chabaudi-specific AMB are short-lived activated B cells. These cells were absent after resolution of the infection, and immunization with purified antigen and TLR agonists resulted in a transient, yet robust, activation of P. chabaudi-specific AMB which lasted no more than 48 hr. Moreover, the Igh NIMP23/+ mouse model allowed us to obtain a deep insight of the transcriptome profile of MSP1 21 -specific AMB during natural infection, and compare it side-by-side with the transcriptome of MSP1 21 -specific naïve B cells. This allowed us to demonstrate their heavily pro-apoptotic and activated transcription profile, further explaining their short-lived nature. P. chabaudi-specific AMB showed very low expression of Bcl2, and high levels of expression of several pro-apoptotic genes including Bad, Bax, Fas and Fasl. In addition, these cells expressed very high levels of classswitched immunoglobulins and genes associated with DNA replication and proliferation.
The association of P. chabaudi-specific AMB with ongoing infection explains several observations in human studies: reduction of HIV plasma viremia by ART resulted in a significant reduction of HIVspecific AMB without altering the frequency of HIV-specific B mem (de Bree et al., 2017;Kardava et al., 2014); individuals living in high malaria endemicity present higher frequencies of AMB than individuals living in areas with moderate transmission Sullivan et al., 2015); repetitive Plasmodium episodes result in higher frequencies of AMB ; the percentage of AMB is larger in children with persistent asymptomatic Plasmodium falciparum parasitemia as compared with parasite-free children (Weiss et al., 2009); previously exposed subjects significantly reduce the frequency of AMB following a year of continuous absence of exposure to Plasmodium falciparum infection (Ayieko et al., 2013). These observations all support the view that constant immune activation rather than impaired memory function leads to the accumulation of AMB in malaria.
After resolution of infection, P. chabaudi-specific AMB did not persist, but instead subsets of P. chabaudi-specific B mem were readily detected. These cells expressed different combinations of previously described mouse B-cell memory markers [i.e. CD80, CD273 and CD73 (Anderson et al., 2007;Tomayko et al., 2010;Zuccarino-Catania et al., 2014)]. P. chabaudi-specific B mem included both class-switched and non-class-switched cells, which show different responses to a secondary challenge infection (Krishnamurty et al., 2016). Independently of the combination of previously described memory markers expressed on P. chabaudi-specific B mem , all of these cells displayed very high expression of FCRL5. Previous data showed most prominent expression of FCRL5 in marginal zone B cells, while much less evident in the newly-formed and follicular splenic B cell subpopulations (Davis et al., 2004;Won et al., 2006). In agreement with this, we observed expression of FCRL5 on a subset of splenic P. chabaudi-specific B cells obtained from naïve mice. However, the level of FCRL5 expression on P. chabaudi-specific B mem detected after resolution of the infection was noticeably higher than that of naïve B cells both at the protein and mRNA levels. In contrast to MSP1 21specific AMB, these CD11b -CD11c -FCRL5 hi MSP1 21 -specific B mem showed high expression of the hallmark memory and anti-apoptotic gene Bcl2 (Bhattacharya et al., 2007). Thus, after resolution of the infection, high expression of FCRL5 acted as a universal B mem marker. Due to its complex dual ITIM/ITAM signaling capacity (Zhu et al., 2013), it is tempting to speculate that FCRL5 might serve as an important signal in the differentiation/maintenance of B mem .
Tracking the fate of the different MSP1 21 -specific B-cell subsets identified in this work will allow detailing the interplay between them. A model can be proposed in which antigen-specific AMB serve as an intermediate stage of differentiation between naïve and B mem . Alternatively, antigen-specific AMB might represent early plasmablasts or recent GC emigrates, based on class-switching and the high expression of IgG by these cells. These scenarios are not necessarily antagonist, and might even occur in parallel. Moreover, B1 B cells have been shown to class-switch and contribute to serum IgG1, IgG2a and IgA to influenza (Baumgarth et al., 2005), and IgG-producing B1a B cells have been shown to accumulate in the spleen of a mouse model of systemic lupus erythematosus (Enghard et al., 2010). Therefore, we can't rule out the possibility that AMB might represent a particular B1 B cell subset that expands in the spleen and blood in response to Plasmodium infection.
Our data suggest that the expansion of AMB in malaria is not a consequence of B-cell exhaustion, but rather a physiologic stage of B-cell activation, and that these cells are sustained in high frequencies by ongoing chronic infections. Thus, Plasmodium-specific AMB are neither 'memory', nor 'atypical'. Importantly, our data demonstrate that robust expansion of Plasmodium-specific AMB does not hinder clearance of the infection, activation of germinal centers, or generation of Plasmodium-specific long-lived quiescent B mem upon resolution of the infection.

Continued on next page
Two weeks after mosquito feeding/infection, each experimental mouse was exposed to 20 infected mosquitos for 30 min. Blood parasitemia in infected experimental mice was routinely monitored by thin blood smears.

Immunizations
Mice were immunized i.p. with a combination of 100 mg of MSP1 21 (Quin and Langhorne, 2001) and 50 ml of Titermax Gold emulsion (Sigma), or a combination of 50 mg of MSP1 21 and 50 mg of R848 (Invivogen).

Flow cytometry and cell sorting
Spleens, lymph nodes and bone marrows were dissected and single cell suspensions were obtained by mashing the organs through a 70 mm filter mesh in HBSS, 6 mM Hepes buffer (Gibco, Invitrogen). After removal of red blood cells from spleens and bone marrows by treatment with lysing buffer (Sigma), the remaining cells were resuspended in complete Iscove's Modified Dulbecco's Medium [IMDM supplemented with 10% FBS Serum Gold (PAA Laboratories, GE Healthcare), 2 mM L-glutamine, 0.5 mM sodium pyruvate, 100U penicillin, 100 mg streptomycin, 6 mM Hepes buffer, and 50 mM 2-ME (all from Gibco, Invitrogen)] and viable cells were counted using trypan blue (Sigma) exclusion and a hemocytometer. Cells were then resuspended in PBS and incubated with APC-or PElabelled MSP1 21 fluorescent probes and/or different combinations of fluorochrome-conjugated antibodies (key resources table), and either acquired after two washes with PBS, or fixed with 2% paraformaldehyde and stored in staining buffer at 4˚C until acquisition. The APC and PE MSP1 21 fluorescent probes were produced as previously described (Krishnamurty et al., 2016;Taylor et al., 2012). Briefly, purified MSP1 21 (Quin and Langhorne, 2001) was biotinylated using an EZ-link Sulfo-NHS-LC-Biotinylation kit (Thermo Fisher Scientic) using a 1:1 ratio of biotin to protein, and loaded onto Streptavidin-APC conjugated or Phycolink Streptavidin-R-PE conjugated (ProZyme) in a 6:1 ratio of MSP1 21 :Streptavidin-fluorochrome.
Cell sorting was performed on a MoFlo XDP (Beckman Coulter) or a BD FACSAria Fusion (BD Biosciences) and the target cell populations were directly dispensed into TRIreagent (Ambion) and stored at À80˚C until RNA isolation. Purity checks were routinely performed for all assays by sorting aliquots of cells into PBS containing 2% FCS and reacquiring them on the cell sorter.
Dead cells were routinely excluded from the analysis by staining with LIVE/DEAD Fixable Aqua or Blue stain (Invitrogen). Singlets were selected based on FCS-A vs FCS-H and further based on SSC-A vs SSC-H. 'Fluorescence minus one' (FMO) controls were routinely used to verify correct compensation and to set the thresholds for positive/negative events. Analysis was performed with FlowJo software version 9.6 or higher (Tree Star).
PhenoGraph and t-distributed stochastic neighbor embedding (t-SNE) were combined to analyze multiparameter flow cytometry data using the Cytofkit package . t-SNE renders high-dimensional single-cell data based on similarities into only two dimensions, and thus helps visualize multiparameter data (van der Maaten, 2008). PhenoGraph (Levine et al., 2015) allows partitioning of high-dimensional single-cell data into phenotypically coherent subpopulations (i.e. clusters). The relatedness of the cell clusters identified by PhenoGraph was inferred using Isomap (Cytofkit package), in which related clusters/subsets can be visualised close to each other.

RNA isolation, sequencing and data analysis
Total RNA from 1À5 Â 10 4 cells sorted into TRIreagent (Ambion) was isolated using the Ribopure kit (Ambion). Concentration of purified RNA was determined by Qubit fluorometric quantitation using the HS assay kit (ThermoFisher Scientific), and the quality analyzed with a 2100 Bioanalyzer (Agilent). Samples with a RIN score above 8.50 were used for the next steps. cDNA was generated from total RNA with the SMART-Seq v4 Ultra Low Input RNA Kit (Takara Bio USA). Next-generation sequencing libraries were produced with the Ovation Ultralow System V2 (Nugen), and run as PE100 on a HiSeq 4000 sequencer (Illumina). GEO accession: GSE115155.
For bioinformatics analysis, paired-end sequence reads were adapter and quality trimmed using cutadapt v1.9.1 (Martin, 2011) with the following non-default settings: '-a AGATCGGAAGAGC -A AGATCGGAAGAGC -minimum-length 30 -q 20,20'. Gene-level abundance estimates were generated from the trimmed reads using RSEM v1.2.31 (Li and Dewey, 2011) running STAR v2.5.1b (Dobin et al., 2013) with default settings, aligned against the Mus musculus Ensembl release 89 transcriptome (mm10). All further analysis was conducted using the DESeq2 (Love et al., 2014) package from Bioconductor v3.5 run in R v3.4.0. The expected counts were imported and rounded to integers to generate a counts matrix. Differential expression between phenotype groups was assessed using the DESeq function with default settings. In the case of comparisons of different MSP1 21 -specific B cell subsets obtained from the same experimental mouse, an additional mouse factor was added to the design formula to accommodate the paired nature of the data. Significance was thresholded using an FDR 0.01. PCA analysis was conducted using DESeq's plot PCA function with the regularized log (rlog) transformed count data. Heat maps were generated using the regularized log (rlog) transformed count data, scaled per gene using a z-score. Mouse homologues to genes previously associated with human AMB were selected (Supplementary file 1), and those showing significant differential expression on MSP1 21 -specific AMB were used to produce separate heat maps split by functional annotation (Figure 3). The GSEA pre-ranked function from the Broad's Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) suite was used to assess significant enrichment of MSigDB's C2 Reactome gene sets associated with differential expression between cell types. The function was run using a list of genes ranked for differential expression using DESeq2's Wald test statistic with default settings except for: collapse dataset to gene = false enrichment statistic = classic

Statistical analysis
Statistical analysis was performed using Mann Whitney U test, Kruskal-Wallis test followed by Dunn's multiple comparisons test, or Two-Way ANOVA followed by Dunnett's multiple comparisons test on Prism software version 6 (GraphPad). p<0.05 was accepted as a statistically significant difference.