Hide and seek: interplay between influenza viruses and B cells

Abstract

Influenza virus constantly acquires genetic mutations/reassortment in the major surface protein, hemagglutinin (HA), resulting in the generation of strains with antigenic variations. There are, however, HA epitopes that are conserved across influenza viruses and are targeted by broadly protective antibodies. A goal for the next-generation influenza vaccines is to stimulate B-cell responses against such conserved epitopes in order to provide broad protection against divergent influenza viruses. Broadly protective B cells, however, are not easily activated by HA antigens with native structure, because the virus has multiple strategies to escape from the humoral immune responses directed to the conserved epitopes. One such strategy is to hide the conserved epitopes from the B-cell surveillance by steric hindrance. Technical advancement in the analysis of the human B-cell antigen receptor (BCR) repertoire has dissected the BCRs to HA epitopes that are hidden in the native structure but are targeted by broadly protective antibodies. We describe here the characterization and function of broadly protective antibodies and strategies that enable B cells to seek these hidden epitopes, with potential implications for the development of universal influenza vaccines.

Introduction

Seasonal influenza virus infection causes severe morbidity and mortality annually. It is estimated that between 291 000 and 646 000 people worldwide die from flu-related illnesses each year (1). The economic losses caused by the influenza illness (deaths and disabilities) are estimated at $5.8 billion in the USA alone (2).

The most effective way to reduce the influenza burden is to elicit protective antibodies by vaccination to the major surface protein, hemagglutinin (HA) (3). Influenza HAs are highly diverse and are composed of at least 18 subtypes (H1–H18) for influenza A virus. On the basis of sequence similarity, the 18 HA subtypes are further categorized into two groups (group 1 and group 2). Owing to the antigenic variations, current influenza vaccines require annual updates to reduce the risk of antigenic mismatch between the vaccine strains and the circulating strains, which can result in low vaccine effectiveness (4, 5). Seasonal influenza vaccines generally elicit strain-specific antibody responses against the highly variable globular head domain of HA but induce poorly antibody responses that target the conserved domain (e.g. the stem domain) (6, 7). As mutations are constantly introduced in the HA head domain by antigenic drift, strain-specific responses acquired by vaccination and infection quickly lose the efficacy (8). In addition, strain-specific responses do not provide protection against new reassortant strains that are generated by antigenic shift (9, 10). Therefore, a major challenge for the development of influenza vaccines is to counteract the antigenic variation and evolution of the HA antigens.

Although influenza HA antigens are highly diverse, there are conserved epitopes across multiple HA subtypes. In general, such conserved epitopes are structurally or functionally important for the virus; therefore, mutations in these regions often reduce viral fitness. Thus, a rational strategy for new influenza vaccines is to elicit antibody responses that target the ‘Achilles’ heel’ of the virus. This strategy has been encouraged by an isolation of large numbers of broadly neutralizing antibodies (bnAbs) against the conserved epitopes on the stem domain and the receptor-binding site (RBS) of the head domain (11, 12).

More recently, we and others have identified HA epitopes that are conserved across influenza viruses and are targeted by broadly protective antibodies in humans (13–15). One striking feature of these epitopes is that they are structurally hidden from antibody recognition in the HA antigens that form a homotrimer in the native structure. Antibodies targeting these epitopes do not neutralize the virus in conventional neutralization assays but provide protection through Fc-mediated mechanisms. Here, we summarize from the immunological point of view the structural and functional properties of broadly protective antibodies to these hidden HA epitopes. We also describe our findings on B-cell selection in germinal centers (GCs) in response to HA immunization or virus infection and extend the discussion to strategies for increasing the frequency of otherwise rare antibody responses against hidden epitopes.

Strategies to isolate broadly protective HA antibodies

Recent advances in characterizing the antigen-specific B-cell antigen receptor (BCR) repertoire at a single-cell level have led to the isolation of multiple classes of broadly protective influenza antibodies in humans and mice. For example, the analysis of recombinant antibodies cloned from single B cells (16–20) has made it possible to characterize the BCR repertoire from any kind of B cell. Indeed, several influenza bnAbs have been isolated from circulating human memory B cells or plasmablasts by this approach (6, 21, 22). A number of influenza bnAbs have also been isolated by other robust approaches: phage-display libraries constructed from various cell sources (23–26), Epstein–Barr virus (EBV) transformation of B cells (27, 28) and subsequent generation of hybridomas (14), single-cell culture of circulating plasma cells (29) and a combination of proteomic spectrotyping of antigen-specific serum antibodies coupled with high-throughput BCR sequencing (30).

We have recently developed a single-cell culture approach denoted as Nojima culture, which allows us to characterize the specificity, avidity and somatic genetics of thousands of BCR repertoires in mice and humans (13, 15, 31–37). B cells of interest (e.g. HA-binding human memory B cells) were sorted directly into each well of 96-well plates at no more than one cell per well and cultured in the presence of feeder cells and cytokines. Feeder cells that express retrovirally transduced mouse or human CD154 (and cytokines) support the expansion, class-switch recombination (to IgG) and plasmacytic differentiation of single B cells. Consequently, culture supernatants contain 1–5 μg ml⁻¹ for mouse B-cell cultures (31, 36) and 5–30 μg ml⁻¹ for human B-cell cultures (13, 35) of clonal antibodies produced by the progeny of single B cells.

Importantly, this culture does not support somatic hypermutation of B cells; therefore, clonal antibodies in culture supernatants retain both the specificity and avidity of the original BCRs expressed by B cells placed in the cultures. The cloning efficiency (frequency of IgG⁺ wells among total wells screened) ranges from 20 to 80%, depending on B-cell types (31, 33, 35, 36). As B cells produce clonal antibodies sufficient for small-scale assays, we can characterize the specificity, avidity and function (e.g. virus-neutralizing activity) of antibodies without the time-consuming and labor-intensive steps of cloning and making recombinant antibodies.

Importantly, Nojima cultures can be used to obtain the V(D)J sequences of the clonal antibodies from each well (>90% efficiency) that contains thousands of clonally expanded B cells. From any clone, typically from selected subsets of clones (e.g. broadly cross-reactive clones), we can make recombinant antibodies for further characterization including biochemical and structural analysis and functional assays. Successful isolation of ‘rare’ influenza antibodies that cross-react with group 1 and group 2 HAs, and the subsequent determination of previously uncharacterized HA epitopes recognized by the cross-reactive antibodies, have demonstrated the utility of this approach (13, 15, 33).

HA bnAbs

The majority of influenza bnAbs isolated thus far recognize either the RBS or the stem of HAs (11, 12). RBS antibodies are able to block initial virus attachment to host receptors and exhibit better neutralizing activity than stem antibodies that block the downstream fusion process (Fig. 1). RBS antibodies are diverse in V, D and J usage (38), in contrast to the relatively limited V_H gene use by stereotypical stem antibodies (Fig. 2). Despite strong neutralization and hemagglutination inhibition (HAI) activities, the breadth of RBS antibodies is normally limited to HA subtype or group (25, 38–44), and only a handful of RBS antibodies isolated to date exhibit cross-group specificity (24, 33, 45). The conserved RBS core is generally smaller than the footprint of typical RBS antibodies. Therefore, RBS antibodies need to contact with peripheral residues that are not highly conserved across multiple subtypes (38, 46).

Fig. 1.

Proposed mechanisms of protection provided by HA bnAbs and by HA broadly protective antibodies to hidden epitopes. Illustration of the protective pathways that confer broad protection by HA bnAbs and by HA hidden epitope antibodies. HA bnAbs (black) inhibit viral replication processes by blocking ① viral attachment (RBS antibodies), ② viral fusion (stem antibodies) and ③ budding of the virus (RBS and stem antibodies) in the absence of effector cells/functions. In contrast, broadly protective antibodies to hidden epitopes (head interface antibodies and stem helix antibodies; red) do not prevent viral attachment or fusion due to the relatively low availability of these hidden epitopes on the virus compared with the bnAb epitopes. Hidden epitope antibodies, instead, bind HA antigens on infected cells that have thousands of HA trimers on their surface and elicit ④ ADCC and antibody-dependent cellular phagocytosis (ADCP) in the presence of effector cells or activate ⑤ CDC pathways to kill infected cells. Both the effector functions and the protection by hidden epitope antibodies are dependent on Fc regions of IgGs as reported for stem antibodies. A subset of head interface epitope antibodies also blocks budding of the virus in the absence of effector cells/functions.

Fig. 2.

Four classes of broadly protective influenza HA antibodies isolated from humans. There are at least four epitopes for broadly protective influenza HA antibodies in humans: RBS (purple), stem (green), head interface (red) and stem helix (blue). Solid areas with purple and green colors represent epitopes that are exposed on HA surface and are recognized by RBS and stem antibodies, respectively. Dotted areas with red and blue colors represent epitopes that are not exposed on the HA surface in the native, trimeric structures and are recognized by head interface antibodies and stem helix antibodies, respectively. These antibodies show distinct structural and functional properties that are summarized in the table.

Another class of bnAbs recognizes the HA stem domain in which several conserved and conformational epitopes have been identified (Fig. 2) (11). Stem antibodies neutralize the virus, albeit poorly, by blocking the conformational change of HAs that is required for the fusion process (25, 29, 47). Instead, stem antibodies provide protection primarily through cell-mediated effector functions, such as antibody-dependent cellular cytotoxicity (ADCC) (29, 48). They frequently use the stereotypical V_H1–69 gene segment and exhibit heterosubtypic breadth within group 1 (27), although minor stem antibodies bind a wider spectrum of HA subtypes (25, 28, 29).

Broadly protective HA antibodies against hidden epitopes

High-throughput screenings of influenza HA antibodies have revealed two additional classes of broadly reactive antibodies. Both classes recognize the epitopes that are hidden in native HA trimers on the virus and do not exhibit neutralizating activity in conventional neutralization assays. Nonetheless, they confer cross-group protection in mice challenged with otherwise lethal doses of the viruses. Herein, we describe the structural and functional properties of broadly protective HA antibodies against hidden epitopes, in comparison to those against neutralizing epitopes (Fig. 2), all of which can be potential targets for ‘universal’ influenza vaccines.

Head interface antibodies

Kelsoe and Harrison’s group (13) and the group of Crowe (14) have independently reported a novel class of human antibodies that provide substantial protection against group 1 and group 2 viruses. This class of antibodies binds a wide range of HAs from group 1 and group 2 viruses but does not compete for binding to HAs with other known HA bnAbs. Instead, these antibodies recognize a head interface epitope that is occluded at the contact surface between HA head domains in the trimeric structure (13, 14, 49).

The occluded head interface epitope appears to be conserved among a wide spectrum of influenza HA subtypes, which likely explains the breadth of head interface antibodies. Members of the S5-C1 lineage antibodies (S5V2-29, S5V1-15 and S5V2-52) were isolated from a single donor, S5 who received trivalent influenza vaccine 2015–16. All three members of the S5-C1 lineage bind HAs from group 1 (H1, H2 and H9) and group 2 (H3, H4, H7 and H14) viruses.

In addition to these three antibodies, we isolated eight antibodies that compete with S5V2-29 from multiple donors. While all 11 head interface antibodies examined in our study exhibit cross-group reactivity to group 1 and group 2 HAs, some of them exhibit skewed binding to group 1 HAs over group 2 HAs or vice versa. A group of the head interface antibodies collectively bind all HA subtypes examined with the exception of HAs from influenza B viruses. Similarly, another head interface monoclonal antibody (mAb), FluA-20, which was isolated from a donor who had an extensive history of receiving influenza vaccinations, binds HAs from most influenza A subtypes with the exception of H16, H7 and influenza B viruses (14).

Consistent with their B-cell origin (i.e. circulating memory B cells or plasmablasts with the exception of FluA-20, whose origin is unknown), all head interface antibodies carry substantial frequency of V(D)J nucleotide substitutions compared with their computationally inferred, unmutated common ancestors (UCAs). Notably, head interface antibodies are encoded by diverse combinations of V, D and J gene segments (13), suggesting the possibility that diverse and polyclonal germline precursors that recognize this head interface epitope can be activated by vaccinations or infections, engage in humoral responses and become mature through the process of antigen-driven clonal selection and affinity maturation in GCs (50–57).

That different head interface antibodies use different modes of contact to bind HAs creates clonal diversity of BCRs/antibodies against this epitope (13, 49). Some of the head interface antibodies appear to be dependent on structures peripheral to the core epitope more than others. Although this dependency on the peripheral structures likely limits the binding breadth of antibodies, such clonal diversity may be important to limit viral escape. For example, single mutations at the core epitope (e.g. amino acid position 220 or 223 or 229) abolish binding of most, but not all of the head interface antibodies (13, 14). Importantly, some of head interface antibodies that rely on peripheral structures for HA binding and do not exhibit the greatest binding breadth still bind to such viral mutants (13), implying the importance of the peripheral binders to limit viral escape. In other words, a virus would have to mutate several positions at the core epitope/peripheral structures to escape completely from survey by a group of head interface antibodies when several lineages are present, making the occurrence of escape mutants rare.

Stem helix antibodies

The long alpha helix (LAH) in the HA2 stem region includes epitopes that are relatively conserved among divergent HA subtypes (58). The presence of the highly immunogenic, globular head domain on the top of trimeric HA conceals the conserved LAH epitopes from antibody access, thereby hiding the epitopes within the native form of HA antigens. This explains why LAH antibodies are poorly recalled by seasonal vaccination in humans although we cannot formally exclude contributions of other unidentified factors (58). The discovery of a mouse mAb, 12D1 (59), which confers protection in both prophylactic and therapeutic usage, initiated studies into LAH-binding antibodies. A subsequent study demonstrated that immunization of mice with an H3-derived LAH peptide elicited broadly protective antibodies not only against drifted H3N2 viruses but also against heterosubtypic H1N1 and H5N1 viruses (58).

More recently, using a single-cell culture approach, we have cloned some LAH-binding antibodies from human memory B cells (15). A substantial fraction of human LAH-binding antibodies exhibit breadth, covering all major influenza A subtypes that infect in humans, including HAs from the H1, H3, H5 and H7 subtypes (15). The LAH antibodies do not neutralize the viruses in conventional virus neutralization assays but, nonetheless, confer heterosubtypic cross-group protection against otherwise lethal influenza infection (15). Thus, LAH antibodies are cross-group protective antibodies targeting hidden epitopes. It is also important to stress that LAH antibodies are present in multiple seronegative donors (15), increasing the feasibility of strategies to consider LAH antibodies as targets for broadly protective vaccines.

B-cell strategies to seek the hidden epitopes for broad protection

Recovery of memory B cells specific for the head interface epitopes or the LAH epitopes from multiple human donors (13–15) strongly suggests that these hidden epitopes are exposed at some point, resulting in B-cell stimulation, and that BCRs against these epitopes are not vanishingly rare but are rather common among human populations. Consistent with their hidden nature, however, neither type of epitope appears to be an immunodominant epitope on HAs (13, 15, 49, 58). Herein, we describe B-cell strategies to seek these hidden epitopes and discuss the local environment and/or potential strategies to activate selectively B cells specific to these hidden epitopes in animal models.

Head interface antibody responses

Using the Nojima culture approach, we characterized the BCR repertoire (specificity, binding avidity and somatic genetics of antibodies) of GC B cells in response to immunization with influenza HAs in mice (31, 49). We showed that during the GC reaction, affinity maturation of B cells was accompanied by the accumulation of V(D)J point mutations and, in fact, by an increased clonal diversity. As a result, the dominant B-cell clones among naive and early GC B cells were generally replaced by initially rarer, high-affinity clones in the late GCs. This mode of B-cell selection and affinity maturation contrasts the purifying selection of a single, high-affinity clone in GCs in response to haptenated proteins (60–62).

To our surprise, about 75 and 50% of clonal IgG antibodies recovered from early and late GC B cells, respectively, did not bind the native form of immunogens in standard binding assays (31). We concluded that complex protein antigens, such as the influenza HA antigen, drive permissive clonal selection in GCs.

Among B-cell clones recovered from late GC B cells in response to immunization with either H3 HA or the glycan-modified counterpart, we identified BCRs/antibodies specific for the HA head interface epitope (49). Clones 8H10, FL-1056 and FL-1066 belong to different clonal lineages but all use the same V_H5-9-1 gene segment paired with the lambda light chain. These antibodies bind a broad range of H3 HAs that had previously circulated as well as H4 HA. In addition, they bind hyper-glycosylated H3 HAs carrying a number of additional putative N-linked glycosylation sites that cover all surface-exposed HA epitopes (designated as gHA^shield) (49).

In response to immunization with wild-type H3 HA, V_H5-9-1⁺ B cells appeared to be infrequent in primary GCs (~0.7% of antigen-binding B cells). This frequency increased to ~10% in GCs in response to hyper-glycosylated HAs. As the overall magnitude of both GC and plasmacytic responses was not impaired by glycosylated HK-68 HAs, the simplest explanation for the over-representation of the V_H5-9-1⁺ B cells is that B-cell responses to normally sub-dominant HA epitopes become dominant in response to HAs that have a reduced number of surface-exposed, likely immunodominant, epitopes. Of note, virtually all antigen-binding GC B cells elicited by the glycosylated HA also bind their wild-type counterpart (49). These results indicate that no glycan-specific responses are elicited by the glycosylated HA and that the elicited antibodies target epitope(s) common to wild-type and glycosylated HAs. Such epitopes include the head interface epitope and might also include other occluded epitopes that have not yet been characterized.

Stem helix antibody responses

Immunization with native HA trimers barely elicits LAH antibodies in humans and mice (15, 58). By contrast, LAH-binding B cells are preferentially selected within local GCs that develop after live virus infection. Following infection, inducible bronchus-associated lymphoid tissues form in the infected lungs and support GC B cells and plasma cells locally (63, 64). GC responses in the lungs are functionally competent for supplying lung-resident memory B cells and presumably plasma cells as well (65–67). One striking feature of the lung GC responses is to select preferentially cross-reactive, LAH-binding B cells that are usually less competitive than strain-specific B cells (66). Therefore, local GC responses at the site of infection, unlike those systemically elicited by HA immunization, select LAH-binding B cells and support plasmacytic differentiation of these B cells.

How are LAH-binding B cells selected in the lung GCs following infection? As antibodies cloned from LAH-binding GC B cells bound poorly to native HA trimers presented on the virus surface (15), LAH-binding B cells would be outcompeted by other B-cell clones that avidly bind HAs when native HA trimers are exclusively presented to B cells in GCs. By contrast, the HA structure is not fixed all the time during the life cycle of the virus, as conformational plasticity exists on the structure (68). It is possible that non-native forms of HAs (e.g. resulting from denaturation, protease digestion and protein modifications) are presented to B cells in the lung GCs following infection.

It is intriguing to speculate that the non-native HA antigens drive the selection of the LAH-binding B cells in local GCs. Indeed, dissociation of HA structures by acid treatment enhanced the binding avidity of LAH antibodies, and immunization with these acid-treated HAs enriched cross-reactive, LAH-binding GC B cells in secondary lymphoid organs. These results recapitulated the cross-reactivity and LAH dominance observed in lung GCs in response to influenza infection (15) and supported the hypothesis that non-native forms of HA created by acid treatment act as selecting agents in local GCs following influenza virus infection.

Cross-group protection by antibodies against the hidden epitopes

Despite the lack of neutralization activity in vitro, both the head interface antibodies and the LAH antibodies provide substantial protection against lethal influenza infection when passively transferred into mice (13–15, 49). This protection depends largely on the Fc region of IgG, similar to the protection provided by stem antibodies (48) (Fig. 1). S5V2-29, H2214 and 8H10 (head interface antibodies), and mLAH1 and mLAH2 (LAH antibodies) recombinantly expressed as mouse IgG2c (or IgG2a for LAH antibodies) protect mice from H3N2 A/Aichi/2/1968 (all antibodies), H1N1 A/Narita/1/2009 (S5V2-29) and H7N9 A/Anhui/1/2013 (LAH antibodies) viruses. In contrast, the mouse IgG1 and IgG2a N297A [asparagine (N) to alanine (A) substitution at position 297] mutant of the same antibodies provide only partial protection (13–15, 49).

Given that mouse IgG2c and IgG2a fix complement efficiently and strongly interact with activating FcγRs, whereas mouse IgG1 does so poorly, and that the mouse IgG2a N297A mutant cannot interact with FcγRs (69), these results strongly suggest an involvement of an IgG effector function in the protection provided by these antibodies. Consistently, S5V2-29 (as mouse IgG2c) mediates both complement-dependent cytotoxicity (CDC) and ADCC in vitro, as measured by the killing of an HA-expressing cell line and by the luciferase activity of the FcγRIV-expressing reporter cells, respectively (13).

In contrast, although another head interface antibody (FluA-20) can activate natural killer (NK) cells in vitro, this effector function (i.e. ADCC) appears to be dispensable for protection provided by this antibody. Passive transfer of the N297A mutant or the LALA mutant [leucine (L) to alanine (A) substitutions at position 234 and 235 of IgG Fc] of FluA-20 protects mice from death caused by influenza challenge as efficiently as their wild-type human IgG1 counterpart does (14).

As these mutant IgGs cannot engage FcγR or complement receptors (69) and therefore cannot trigger Fc-mediated pathways, the authors conclude that the Fc-mediated ADCC activity is dispensable for FluA-20’s protective role in vivo. Although the experimental data support their conclusion, it is important to note that mice that received the N297A mutant exhibited significantly more severe weight loss in response to influenza infection compared with those that received the wild-type human IgG1 counterpart.

This observation suggests a role of the effector function of FluA-20 in optimal protection against influenza challenge. It may be that head interface antibodies, like stem antibodies (48), rely more on Fc-mediated effector functions when the effective antibody concentration at local sites is low, a scenario most likely recapitulated in humans, as the majority of anti-HA antibodies are strain-specific and polyclonal. When the effective concentration reaches certain levels, head interface antibodies may inhibit the shedding of viruses via functions independent of Fc, such as the dissociation of HA trimers into HA monomers shown for the FluA-20 (14). Clearly, further investigation is required to fully understand the mechanisms of protection by these broadly protective antibodies that target hidden epitopes.

Concluding remarks

Studies have isolated multiple classes of broadly protective antibodies that target different conserved HA epitopes from humans and mice. These findings not only provide knowledge on how our immune systems cope with the viral variants, but also raise the possibility that new vaccines and vaccination strategies to reliably elicit broadly protective antibodies could be developed. Substantial progress has been made to improve vaccine efficacy to drive broadly protective antibody responses against the target influenza antigens (e.g. stem) (70–73) and a clinical trial has just begun for one of the stem vaccines (ClinicalTrials.gov Identifier: NCT03814720) although the antigenicity and immunogenicity of stem antigens could be improved (7, 72).

There are four key immunological steps that involve the elicitation of all classes of broadly protective antibodies: antigen recognition by naive and/or memory B cells. B-cell selection in the GCs, establishment of memory B cells and long-lived plasma cells, and recall of memory B cells. Understanding the biology of humoral responses to broadly protective HA epitopes would provide substantial merit for developing efficacious vaccines and vaccination strategies against rapidly mutating influenza viruses.

Funding

The work related to this review was supported in part by Japan Agency of Medical Research and Development under JP19fk0108051 and JP20fk0108141 (to Y.T.) and by National Institutes of Health grants P01 AI089618 (to Stephen C. Harrison), U19 AI117892 (to Garnett Kelsoe and Thomas B. Kepler) and R01 AI128832 (to Garnett Kelsoe).

Conflicts of interest statement: the authors declared no conflicts of interest.