Plasticity and Heterogeneity in the Generation of Memory B Cells and Long-Lived Plasma Cells: The Influence of Germinal Center Interactions and Dynamics

The major function of the primary immune response is to generate a robust and appropriate effector response to any foreign stimulus. In principle, the nature of this response has evolved to successfully eliminate foreign pathogens while sparing the host excessive collateral damage. To accomplish this, the immune system must balance producing a large number of effector cells targeting foreign Ags while safeguarding host cells by appropriately regulating and ultimately turning off the response. In addition to effector cells, the primary response produces Ag-specific memory cells that remain after the infection has been cleared. These cells can be reactivated upon re-exposure to the same Ag, rapidly differentiating into effector cells. In the humoral response, memory B (B_mem) cells, together with long-lived plasma cells (PCs), enhance host protection for long periods of time.

Although we understand the basic features of a humoral response (1), we still do not know the signals that invoke B_mem cell formation or fully understand what regulates the quality and quantity of the memory and PC populations. Why do different infections or types of vaccines generate different lengths and qualities of immune protection? How is the B_mem cell population maintained in the face of recurring infections? Are there different types of memory cells geared to be functionally different? Addressing these questions and gaining a deeper understanding of B cell memory will likely lead to a better understanding of the success of vaccines to different pathogens and, therefore, will inform vaccine design in the future.

The most basic definition of a B_mem cell population is that it is Ag experienced and persists in the absence of the immunizing agent. Two other qualities often attributed to B_mem cells are that, compared with naive cells, they respond more rapidly during a secondary response and they have a higher affinity for Ag. However, the latter attribute is associated with cells that have participated in a germinal center (GC) reaction, suggesting that GC B cells are precursors to B_mem cells. Although it seems clear that GC cells can be precursors to B_mem cells, it is uncertain whether all B_mem cells have derived from GC precursors.

There are two broad types of humoral responses: T independent (TI) and T dependent (TD). Although TI responses can generate abortive GCs (2, 3), TD responses typically are associated with fully developed GCs. In the TD GC, Ag-specific B cells undergo numerous rounds of division and can undergo somatic hypermutation (SHM), class-switch recombination, and affinity selection (4–7). Therefore, the GC response endows this population of Ag-activated B cells with qualities, such as higher affinity for the Ag and a relevant Ig isotype, which should result in more efficient clearance of Ag. However, the GC itself is transient, and GC cells themselves are not known to mediate effector functions. Instead, some GC B cells differentiate into PCs or B_mem cells, both of which persist after the primary response has subsided and directly or indirectly provide effector function.

In contrast to the more open-ended definition given above, the classical description of B_mem cells requires that they have undergone affinity maturation and selection. This led researchers to conclude that the GC environment is necessary for B_mem cell formation. Unfortunately, the arguments supporting GCs as the sole location in which B_mem cells are formed can be circular (reviewed in Refs. 8 and 9). Nevertheless, numerous studies demonstrated that in certain immunodeficiencies in which a functional GC response is abrogated, the B_mem cell pool is altered in number, affinity maturation, and/or ability to generate secondary responses. These immunodeficiencies are also marked by a decrease in circulating Ab and long-lived PC numbers (Table I). Therefore, although it is likely that long-lived cells can be formed in the absence of functional GCs, the GC microenvironment plays an important role in producing a classically defined, fully formed B_mem cell population, as well as long-lived PCs.

Under normal circumstances, fully developed GCs require T cell function. B cell–T cell interactions occur throughout the humoral response, either at the border between the T and B cell zones or within the follicle itself, and the nature of these interactions changes over time. Several such interactions and subsequent signaling are vital in the early stages of the response. In the absence of signaling downstream of CD40 or ICOS, for example, GC initiation is disrupted. Conversely, in the absence of B cells, T cells in the follicle (T follicular helper cells [T_FHs]) are also absent (38, 46). It remains unclear whether the absence of T_FHs in B cell-deficient animals is due to a complete inhibition of differentiation into T_FHs or whether differentiation can occur, but the phenotype of the T_FH subset requires B cells for maintenance (47). In this vein, the absence of signaling lymphocytic activation molecule-associated protein (SAP), which is expressed on T cells, directly impacted the duration of B cell–T cell interactions (48), leading to diminished T_FH development and expansion. Therefore, when the quality of signaling downstream of early B–T interactions is impaired, formation of GCs is also disrupted.

After a GC has formed, T_FHs critically affect the outcome. T_FHs influence B cells via two mechanisms: direct cell contact and cytokine secretion. T_FHs can control the initiation and termination of humoral responses through the expression of positive and inhibitory ligands and receptors, such as CD40L, ICOS, programmed death-1 (PD-1), CD28, and CTLA-4, thereby regulating the extent of the response and possibly the quantity of effector cells (10, 11, 13–15, 17, 23, 49). T_FHs also affect the quality and types of B cells produced (e.g., via the well-document effects of T cell-derived cytokines in influencing B cell survival, proliferation, isotype switching, and differentiation) (27, 50–53). In addition, the availability of T cell help and the expression of FasL (CD178) regulate selection within the GC, affecting the affinity of resulting long-lived PCs (28, 33, 52). Although it is evident from these studies that T cells can affect GC dynamics, are they involved in the decision of GC B cells to differentiate into B_mem cells or PCs?

How is GC B cell differentiation regulated? Is there is an actual fate decision per se, or is differentiation the result of an accumulation of factors that overrides a death or GC-retention signal? One hypothesis is that a master regulator of transcription directs the fate of a B_mem cell. B cell lymphoma (Bcl)-6 is required to induce the GC transcriptional program (54), and B lymphocyte-induced maturation protein-1 plays a similar role in PCs (55). However, despite extensive investigation of gene-expression profiles of human and murine B_mem cells (56–59), no single deterministic transcription factor for B_mem cells has come to light. In fact, unlike the comparison between naive cells and GC B cells or PCs, the transcriptomes of naive and B_mem cells are very similar (56–59).

There have been a number of proposals for factors that may determine B_mem cell differentiation. Researchers suggested that Bcl-6 (56) or CD40 (57) can regulate B_mem cell fate decisions, although these proposals have not been borne out (58). Alternatively, affinity selection, in some cases by follicular dendritic cell (FDC)-bound immune complexes, has also been suggested to determine the fate of B cells, with high-affinity cells directed to differentiate into PCs (63, 64). However, subsequent studies demonstrated that higher-affinity cells were undergoing more expansion early in the response, rather than being actively directed into the PC-differentiation pathway (65). Furthermore, in the absence of immune complexes, GC responses and B_mem cells still form robustly (66); therefore, signals from immune complexes on FDCs are likely not involved in the fate decision per se.

Given the failure to find deterministic signals that direct memory cell formation from a GC precursor, it is possible that such a singular directive does not exist. Hodgkin and colleagues (67–69) hypothesized, based on a series of in vitro experiments and computer models, that a uniform population of B cells is intrinsically able to commit to any one of several different fates postactivation (such as division, differentiation, or death) based on stochastic factors, influenced by integrated external and intrinsic signals. For example, B cells stimulated with CD40 will not all behave in the same manner. Some remain undivided, some divide, and a proportion differentiate. Hodgkin and colleagues proposed that at each division, some cells commit to each of these fates, but changing intrinsic or extrinsic factors in the response only affect the probability of one event occurring over the other. In this model, the ability of a B cell to differentiate is part of its cellular make-up and can be influenced in a probabilistic sense, but not determined, by extrinsic stimuli.

In the case of the GC, we suggest that T cell-derived signals regulate the quantity and quality of the B_mem cell population by influencing B cell behavior. The role of signals in producing a fully formed B_mem cell population can be subdivided into those that are important for GC homeostasis and those that may influence B cell differentiation itself. Indeed, a number of recent publications investigated the effects of specific genetic and/or cellular defects on GC B cell dynamics (progression, cell death, and size), as well as differentiative outputs of GCs (i.e., B_mem cells and PCs).

In this regard, it is important to recognize that the output from the GC seems to change over time. At early stages, the reaction seems to generate a substantial population of memory phenotype B cells, which have been identified as early as day 7 postimmunization (70); this observation led to the plausible proposal that some early memory cells derive independently of the GC (19). Similarly, short durations of BrdU administration conducted early in the GC response, in which cells that stop dividing shortly after incorporating BrdU will remain BrdU⁺, lead to very efficient labeling of many long-lived B_mem cells, as detected at week 8 and beyond postimmunization (66) (S. Anderson and M. Shlomchik, unpublished observations). Takahashi et al. (33) found that the memory compartment was much less mutated than the PC compartment, and its level of mutation matched that of the early, but not late, GC population, again indicating that memory cells emerge relatively early from the GC. Conversely, the frequency of mutations among long-lived PCs in the bone marrow (BM) matches that of the late, but not the early, GC (33). Together with evidence that the size of the long-lived PC compartment in the BM continues to increase (7) even beyond 16 wk postimmunization, a time when the memory compartment has long since stabilized (66), it seems that the late output of the GC is more skewed to the generation of PCs, some of which migrate to and reside in the BM as part of the long-lived PC pool. Commensurate with this interpretation, the average affinity of the BM PC population increases for 8–12 wk after immunization with a nonreplicating Ag (7, 71).

Given the complex T–B interplay required to generate and then sustain a GC response, it is not surprising that a large number of mutations affecting molecules and cells critical to the immune system disrupt the GC reaction, to at least some degree. Although each of these mutations has its own distinct phenotype, it is interesting to consider them with respect to the timing and nature of the defects with which they are associated. One category of such mutations blocks signals and/or cellular interactions critical for the formation of GCs or the very early events involved in GC development. Mice deficient in Bcl-6, SAP, CD40, or CD40L do not form GCs (10–12, 20, 39) (Table I). B_mem cell numbers, as well as long-lived PCs and Ab titers, are also markedly diminished in these animals, consistent with the lack of GC formation. Similarly, human patients with deficiencies in CD40, CD40L, ICOS, and SAP have dramatically decreased numbers of switched B_mem cells. In contrast, they often have normal or slightly reduced numbers of mutated IgM⁺CD27⁺ B cells, with CD27 used as a marker for human memory cells (8). In mice, low-affinity B_mem cells can be formed in the absence of Bcl-6 (39). Taken together, these data suggest that early signaling events, which occur before the formation of a functional GC, are sufficient to induce a degree of B_mem cell formation (Table I). However, in the absence of these signals, the numbers of B_mem cells are nonetheless markedly reduced, presumably because a GC environment in which B_mem cell precursor expansion takes place is absent. In all of the mutants in this category, isotype-switched long-lived Ab responses were severely impaired, as expected from the lack of a GC response.

Another category of mutations is at least somewhat permissive for GC formation but seems to block GC progression or maintenance. These mutants that affect late GC responses tend to show differential regulation of B_mem cells and PCs, with generally normal or even increased numbers of memory cells but decreased numbers of PCs (Table I). For example, phospholipase Cγ2 (41) and CD21/35 (42, 72) are vital for GC B cell survival but not formation. Similarly, mice expressing mutant CD19 alleles form GCs, but these GCs fail to mature (45), and there are significant defects in the long-term Ab response. Although it was originally thought that ICOS deficiency completely blocked the GC response (13–15, 17, 18), ICOS seems to be in the middle of these two categories. More detailed research showed that GCs do form in the absence of this molecule, albeit to a much lesser degree; ICOS blockade induced during the GC response leads to a reduction in V_H mutations in B_mem cells but not the total numbers of memory cells themselves (16, 19). Importantly, total Ab responses and particularly high-affinity Ab were reduced by such treatment. ICOS may act, in part, to sustain CD40L expression by T cells, which, in turn, is required to maintain CD40 expression on B cells (16). In the absence of signaling through receptors of T cell-derived cytokines, GCs can form, but their frequency is decreased in the late response. In the case of IL-4, there was a 3-fold reduction in the frequency of GC B cells in mesenteric lymph nodes 2 wk postinfection with an intestinal parasite (25), whereas in lymph nodes responding to (4-hydroxy-3-nitrophenyl)acetyl (NP)-OVA infection, there was a decrease in the frequency of GC B cells in IL-4–deficient animals only at days 14 and 21 (26). IL-21 acts directly on GC B cells to maintain Bcl-6 expression; without it, the GCs dissipate later in the response (28, 29). Still other molecules are important in the maintenance of even later stages of the GC response, suggesting that constant B–T interactions are important for maintenance of GC, although the proteins responsible change as the GC evolves. Our recent research demonstrated that the absence of PD-1 resulted in increased apoptosis in the late GC, as well as smaller GCs at these time points (24). Although the numbers of cells of a T_FH phenotype increased during the late GC response in the absence of PD-1 signals, they produced less IL-4 and IL-21 directly ex vivo. This correlated with a decrease in survival of GC B cells, similar to the IL-4Rα and IL-21 gene-targeted mice (24, 25, 28, 29). Therefore, B and T cells interact to direct the formation of GC responses, as well as the maintenance of the T_FH phenotype (28, 29) and, as a result, the maintenance of the late GC response.

Although numbers of B_mem cells are not greatly reduced in the face of mutations that block GC progression, the affinity of the resultant memory and/or PC populations can be affected, depending on the mutant (Table I). The process that controls selective enrichment of high-affinity cells in the GC population and their subsequent recruitment into the long-lived populations could be influenced by several factors. In the absence of competition between cells of varying affinities, low-affinity GC B cells are intrinsically more prone to undergo apoptosis compared with high-affinity B cells (73). The presence of competing high-affinity cells also suppresses the response of low-affinity cells, although the mechanisms underlying this are unknown (74). When the balance between pro- and antiapoptotic molecules downstream of the BCR is artificially tipped toward survival, such as in B cell leukemia-xL transgenic (Tg) and Bim-deficient mice, there is an accumulation of low-affinity B_mem cells (6, 35). In contrast, in Fas-deficient mice, B cells continue dividing within the GC, undergoing more rounds of SHM than usually occur in normal mice, resulting in a more mutated B_mem cell population (33, 34). Administering CD40L-blocking Ab in the late GC response suppresses this accumulation, suggesting that extension of the GC response itself was responsible for the alteration of the memory population, rather than a direct effect of the absence of Fas signaling in B_mem cells (33).

The availability of T cell help provides further extrinsic modulation of selection. In the absence of signals through IL-21 (28, 29), there is an accumulation of low-affinity B_mem cells. In contrast, in the absence of PD-1 (24) or complement receptor 2 (43) signaling, the affinity of the B_mem cell pool seems to be relatively unaffected, but the remaining long-lived PCs are of higher affinity than the wild-type (WT) controls. In the case of PD-1 deficiency, this could be related to the partial reduction of IL-4 and IL-21 found at late GC time points (24). Although the results from IL-21– and PD-1–deficient animals may seem contradictory, we hypothesize that the difference can be explained by the differing degrees of T cell help available in the two systems (Fig. 1). In the complete absence of IL-21, low-affinity cells are produced in the initial phase of the response. As a result of the lack of IL-21 signals, Bcl-6 expression is not maintained in GC B cells, and they exit from the GC too early to produce substantial numbers of high-affinity cells. As a result, low-affinity post-GC (memory phenotype) cells accumulate; in fact, there is an excess of such cells (28). This could reflect more that the process was terminated prematurely rather than a direct effect on selection per se. In contrast, in the absence of PD-1 signaling, IL-4 and IL-21 production was reduced, but not absent; this was most evident at later time points (24). When low- and high-affinity cells are formed, but there is reduced T cell help, competition for T cell help could result in high-affinity cells outcompeting the low-affinity cells. This idea is supported by data showing that when competition with high-affinity cells is reduced, even very low-affinity cells are capable of forming GCs and entering the B_mem cell pool (75). Because PCs are the main output of the late GC reaction, the net result is seen within the PC compartment, which is smaller and particularly lacking in lower-affinity PCs when PD-1 signals do not occur (24). Therefore, it is possible that the levels of cytokines, and possibly other undefined T cell-derived signals, can affect the quality of the long-lived populations by affecting selection of high-affinity GC precursors.

Integrating these findings, we propose a model for how a population of B_mem cells is formed. After each division, an activated B cell can continue dividing, or it can exit the cell cycle and differentiate into a memory cell or PC. This can occur within or outside of a GC. However, this fate decision is regulated in a stochastic fashion by the microenvironment, with some signals [e.g., those that downregulate Pax5 (76)] favoring PC over memory cell commitment, or vice versa. Within the GC, cells expand and mutate; although low- and high-affinity cells may differentiate into B_mem cells, competition for T cell help favors survival of the fittest, presumably the high-affinity cells (Fig. 1). Therefore, T cells and their products influence the magnitude and quality of the response, as well as the ratio of B_mem cells/PCs, in part depending on when during the GC response individual mediators and mutations exert their effects (Fig. 1).

Although, in general, B cell memory has been referred to in monolithic terms, the B_mem cell population is actually phenotypically and functionally heterogeneous. The identification of such diversity within the B_mem cell pool raises additional, interesting questions about the signals that influence B_mem cell formation. How would variations within the memory compartment arise? In fact, there are almost no data that speak to this question. However, we speculate that in much the same way that stochastic interactions with environmental cues could dictate GC cell fates in general, diverse experiences and timing could result in variation of B_mem cell properties. Specifically, it is likely that the specific nature of the components of the primary response alter the types of B_mem cells produced. These alterations may include the type and amount of the infectious agent, route of infection, number of precursor cells, and types of T_H cells (and thus, cytokines).

Defining subsets of B_mem cells requires the phenotypic, rather than functional, identification of B cell memory, something that until recently has mainly been pursued using cells from human tissues or blood. Studies of human B_mem cells have long used CD27 as a marker of B_mem cells, although it subsequently was found that there is a small subset of CD27⁻, isotype-switched B cells (77–79). The notion that CD27 marked human B_mem cells derived initially from the finding that this subset contained the vast majority of isotype-switched and somatically mutated B cells, which, strictly speaking, only indicates that the cells have undergone prior activation.

Heterogeneity of BCR isotype is the best-studied aspect of diversity within the memory compartment. In humans, CD27⁺ B cells can be subdivided into isotype-switched and nonswitched (IgM⁺) B_mem cells. However, these IgM⁺CD27⁺ cells could be found in patients with CD40L deficiency, suggesting to some that they were not real memory cells, because they could form in the presumed absence of a functional GC (8). Yet, B_mem cells can be generated in mice in which a fully functional GC does not form (Table I). Therefore, these B cells are formed independently of a GC, or signaling that takes place during early GC formation is enough to produce some B_mem cells before abrogation of the reaction occurs. There are other indications from murine studies that GC-independent B_mem cells can be formed. Responses to TI Ags can form memory (80), but whether these cells begin to undergo a TI GC pathway is unclear. Unmutated, low-affinity memory cells are found in immune mice with B cells lacking Bcl-6 and, hence, lacking GCs (39). Similarly, B1b cells can generate TI IgM memory cells that protect from Borellia infection (81). It is quite likely that, in addition to isotype, TI and TD B_mem cells differ qualitatively with respect to function and maintenance; indeed, NP-specific B_mem cells generated with TI immunization have a half-life similar to naive B cells (75), in contrast to the very extended half-life of TD NP-specific B_mem cells (66).

To extend the debate about the origin of IgM⁺CD27⁺ B cells in humans, researchers recently transferred human B cell subsets into immunodeficient mice and immunized them with TI or TD Ag (82). After immunization with the TD Ag, IgM⁺CD27⁺ and isotype-switched CD27⁺ B cells responded by making IgG Ab, indicating a memory-like function. Other studies of human IgM⁺CD27⁺ B cells showed somatic mutations in bcl6, an event that occurs in the GC (83), and the presence of V-region mutations in many IgM⁺CD27⁺ B cells suggests that they are bona fide memory cells. The presence of mutated IgM memory cells in mice has long been known (84), but these cells are understudied. We showed that such cells are common in immune mice that received small numbers of NP-specific B cells with the B1-8 knockin H chain; such IgM memory cells retained some dependence on B lymphocyte stimulator (BLyS) for survival, unlike their IgG1-expressing counterparts, indicating functional heterogeneity (85). Recently, the existence of murine IgM memory cells was confirmed by use of an AID-Cre YFP-reporter mouse, which demonstrated that IgM and switched Ag-experienced B cells are formed in response to immunization with sheep RBCs or NP-chicken gamma globulin (71). Transfer of IgM⁺YFP⁺ (putative memory) cells induced GC formation in a secondary response, whereas transfer of isotype-switched YFP⁺ cells induced a predominantly plasmablast response, leading the investigators to conclude that the heterogeneity seen in Ig isotypes directly relates to different functions during a secondary response. Whether this was purely a correlative finding or whether the isotype differences were mechanistically responsible for the different responses of IgM and IgG B_mem cells was not determined; however, differences in IgM and IgG1 cytoplasmic tails were hypothesized to account for the rapid differentiation into Ab-forming cells by B_mem cells (86).

In addition to differences between switched and unswitched isotypes, further heterogeneity lies in functional differences of memory cells with different (IgG, IgA, and IgE) isotypes. The importance of producing different isotypes lies in the ability of the immune system to effectively respond appropriately to a multitude of foreign Ags and at different anatomical sites, such as by producing IgA in mucosal tissues. Interestingly, studies of human cells show that multiple subsets of T_FHs, as defined by differential production of cytokines, can be induced in the presence of the same cytokine milieu (87). Furthermore, isolation and study of B cell–T cell conjugates demonstrated that the same response could induce T_FH subsets, as defined by distinct patterns of cytokine production, which, in turn induced B cells with diverse, corresponding isotypes (26). Although more work is required to explore the presence, timing, and location of T_FH subsets during an immune response, it is possible that they play a role in producing diversity of the B_mem cell pool, inducing switching of most B cells to one isotype, as well as producing some B cells of other isotypes, allowing for plasticity during the secondary response.

Comparisons of human naive, IgM, and switched B_mem cells demonstrated that IgM and switched B_mem cells are genotypically and phenotypically similar and behave similarly in culture to various stimuli (56). Although they express many of the same proteins, the abundance of some of these proteins does differ, suggesting slightly different experiences during the primary response and potentially underlying subtle differences in in vivo responses (56). Unfortunately, studying variations in human B_mem cells is limited because comparisons between cells generated by the same Ag cannot be performed. However, differences have also been found in murine B_mem cells generated in response to the same Ag. Recent work from our laboratory showed that certain surface markers are differentially expressed on B_mem cells. Subsets of B_mem cells differentially upregulate CD80, PD-L2, CD73, and CD21/35 (59, 88 and M. Tomayko, N. Steinel, S. Anderson, and M. Shlomchik, manuscript in preparation). These markers are induced independently of each other, again consistent with a model that stochastic events, rather than deterministic signals, influence their expression. Nonetheless, there is a hierarchy, such that some memory cells express multiple memory-specific markers and others express few or none. These subsets have differential rates of mutation, as well as BLyS dependency, suggesting a “maturation gradient,” with the cells lacking expression of these markers having fewer mutations and greater BLyS dependency, more like naive B cells, whereas the cells expressing multiple markers (particularly PD-L2 and CD80) seem to have more classical, fully-developed memory-like features (M. Tomayko et al., manuscript in preparation). In addition to IgM B_mem cells, Dogan et al. (71) divided B_mem cells into peanut agglutinin (PNA)⁺ and PNA⁻ fractions; however, these subsets were only present in a chronic response (71). Therefore, it is possible that the PNA⁺ fraction had only recently exited the GC or may have even contained persistent GC cells; thus, it may not form part of the memory response in the absence of Ag. Nevertheless, this article clearly confirmed the existence of IgM memory cells that were PNA⁻. Lastly, identification of a subset of FCRL4⁺ tissue-resident memory cells in humans (9, 89) suggests that location may also play a role in producing specialized subsets of B_mem cells. Overall, there is an emerging literature that is uncovering substantial diversity in murine and human B_mem cells, not just at the level of the BCR isotype, but also with respect to expression of key surface markers, although much more needs to be learned about the functional significance of these newly defined subsets.

We propose that the quality of the B_mem cell population depends on the time spent within the GC, as well as the quality of interactions with T cells while there. This results in a gradient of B_mem cell maturation, with the more classically defined B_mem cells emerging later in the response, although before most of the PCs. Within this gradient, one could broadly characterize some B_mem cells as more naive-like, and others as more (classically defined) memory-like. Hence, we would predict that TI and early B_mem cells, as well as B_mem cells from mutants that block early GC progression, are more naive-like and less mutated, with reduced expression of memory-specific surface markers. Such cells may be more likely to re-enter GCs upon Ag re-exposure. More memory-like cells would have reciprocal properties and would tend to form only in TD responses, perhaps as a consequence of more receipt of T cell signaling and/or numbers of division or other signals received in the GC. Therefore, the heterogeneity of the B_mem cell population is a result of fine-tuning mechanisms during the primary response.

In addition to parameters that affect GC dynamics, other fine-tuning mechanisms, such as selection of affinity variants and control of B_mem cell numbers postexit of the cell cycle, exist to shape the B_mem cell repertoire. These processes all converge to form an optimized, yet phenotypically and functionally diverse, B_mem cell population.

To survive, an organism must be able to respond to multiple pathogens and appropriately adapt to variations in Ag stimulation and availability and type of T cell help. Therefore, heterogeneity in the B_mem cell pool would be adaptive with respect to the variety of different infectious agents, recurring infections, and/or evolution of viruses over time. Furthermore, to provide protection over the lifetime of the host, the B_mem cell pool must also persist in the face of what could be many antigenic challenges. Diversity that includes some cells with more self-renewal capacity or ability to re-enter the GC reaction and others with a greater tendency to quickly differentiate into plasmablasts would be one way to ensure this. The next phase of investigation into B_mem cell development may include testing of whether different infections, through different dynamics of the response and/or different innate immune signals, affect the make-up of the B_mem cell pool. If so, it may help to explain why some vaccinations and infections induce memory over the lifetime of the host, whereas others do not.

We thank Mary Tomayko, Steven Kerfoot, Saheli Sadanand, and Griselda Zuccarino-Catania for helpful discussion and/or critical review of this manuscript.

Disclosures The authors have no financial conflicts of interest.