A perspective on the structural and functional constraints for immune evasion: insights from the influenza virus

Abstract

Influenza virus evolves rapidly to constantly escape from natural immunity. Most humoral immune responses to influenza virus target the hemagglutinin glycoprotein (HA), which is the major antigen on the surface of the virus. The HA is comprised of a globular head domain for receptor binding and a stem domain for membrane fusion. The major antigenic sites of HA are located in the globular head subdomain, which is highly tolerant of amino-acid substitutions and continual addition of glycosylation sites. Nonetheless, the evolution of the receptor-binding site (RBS) and the stem region on HA is severely constrained by their functional roles in engaging the host receptor and in mediating membrane fusion, respectively. Here, we review how broadly neutralizing antibodies (bnAbs) exploit these evolutionary constraints to protect against diverse influenza strains. We also discuss the emerging role of other epitopes that are conserved only in subsets of viruses. This rapidly increasing knowledge of the evolutionary biology, immunology, structural biology, and virology of influenza virus is invaluable for development and design of more universal influenza vaccines as well as novel therapeutics.

Graphical abstract

Introduction

Influenza virus imposes a substantial health and socioeconomic burden globally [1]. There are four known types of influenza virus, named as A-D. Type A and B viruses circulate in human population and are responsible for pandemics (type A), epidemics and seasonal outbreaks (types A and B), while type C and the recently discovered type D [2] viruses do not cause significant disease or epidemics. Influenza A viruses are further classified into subtypes according to the antigenic properties of their two viral surface glycoproteins, namely the hemagglutinin (HA) and neuraminidase (NA). There are 18 known HA subtypes (H1 to H18) and 11 known NA subtypes (N1 to N11) [3]. The 18 HA subtypes can further be classified into group 1 (H1, H2, H5, H6, H7, H8, H9, H11, H12, H13, H16, H17, and H18) or group 2 (H3, H4, H7, H10, H14, and H15) [4]. Out of the 198 (11 × 18) possible combinations, only three (H1N1, H2N2, H3N2) are known to have caused human pandemics. The main natural reservoir for influenza A viruses are wild aquatic birds, but domestic poultry also become infected and hence harbor influenza A viruses [5, 6]. Pigs and other mammals, such as horses, dogs, seals, minks, and bats [7], can also be infected by influenza A viruses and contribute to possible sources of viruses that infect humans [5]. Certain subtypes found in natural reservoirs occasionally emerge in the human population, as exemplified by H5N1, H5N6, H6N1, H7N7, H7N9, H9N2, and H10N8 viruses. Some of these zoonotic subtypes can be highly pathogenic and have a high mortality rate (>50% of hospitalized individuals) when infecting humans [8, 9].

Influenza has been a long-term threat to humans and the first major pandemic that was documented was that of the 1918 H1N1 Spanish flu that was responsible for more than 50 million deaths worldwide [10, 11]. Since then, there were three pandemics, namely the Asian flu (H2N2) pandemic in 1957, the Hong Kong flu (H3N2) pandemic in 1968, and the most recent swine flu (H1N1) pandemic in 2009. Over the past five decades, annual (seasonal) outbreaks have been caused by influenza A H1N1 and H3N2 subtypes as well as the two lineages of influenza B virus (B/Victoria/2/87 and B/Yamagata/16/88). As compared to influenza B, influenza A generally results in higher morbidity and mortality [12, 13]. In addition, influenza A viruses evolve three times faster than influenza B viruses [14, 15]. Therefore, influenza A viruses have often received more attention and concern as a global threat compared to influenza B viruses [16]. Of note, the quadrivalent influenza vaccine that is recently licensed in many countries [17] now offers protection against both lineages of influenza B virus as well as the two influenza A subtypes [18].

Among all influenza virus proteins, HA evolves at the highest rate [19, 20] due to it being the major target of the immune response. Phylogenetic analysis suggests that different HA subtypes of influenza A virus diverged around 2,000 years ago [21]. Although the protein sequences of their HAs share as low as 40% sequence identity, they adopt the same protein fold [22]. As a class I viral fusion protein, HA plays an important role for viral entry by binding to the host receptor, sialylated glycans on endothelial cells in the respiratory tract, and facilitating membrane fusion in the low pH environment of the endosomal compartments after cell entry via endocytosis. During virus replication, the uncleaved precursor of the HA, namely HA0, is synthesized and is then cleaved by cellular proteases into two subunits HA1 and HA2, to produce the fully functional form of the protein [23]. Although this cleavage is usually catalyzed by trypsin-like serine endoproteases [24, 25], HAs from highly pathogenic H5 and H7 subtypes that contain a polybasic cleavage site can also be cleaved by the ubiquitous protease furin [26–28]. This maturation process is a prerequisite to attain the fusion-competent, metastable form of the HA that undergoes the large conformational rearrangements required for the membrane fusion process.

While a large portion of HA1 amino-acid sequence is highly variable [22] and is intrinsically tolerable to mutations, the receptor-binding site (RBS) is an exception [29, 30]. The HA RBS is composed of the 130-loop, 150-loop, 190-helix, and 220-loop, named after their relative positions on the HA amino-acid sequence. The stem region, which is composed primarily of HA2 with some residues from the N- and C- termini of HA1, is even more conserved [22]. During the membrane fusion process that is triggered by acidic pH [31–33], the α-helices in stem region [34] undergo large conformational rearrangements to form a 100 Å triple-helical coiled-coil [35]. This molecular machine consists of many moving parts that impose strong evolutionary constraints on many residues in the stem region.

Influenza H1N1 and H3N2 viruses entered the human population in 1918 and 1968, respectively [36]. While the seasonal H3N2 virus continues to circulate until the present day, the history of human H1N1 virus is more complex [36, 37]. In the last century, the human H1N1 virus first appeared in 1918, but discontinued circulating in human population in 1957 for around 20 years. It remerged in 1977 as a relatively benign epidemic and continued to circulate as a seasonal virus until 2009, when the pandemic swine flu (A(H1N1)pdm09) emerged and displaced the seasonal H1N1 virus. Of note, HAs from both seasonal H1N1 virus and A(H1N1)pdm09 are derived from the HA of 1918 Spanish flu, while the other genes have different origins [38]. Nonetheless, the seasonal H1N1 virus had mutated so much that was antigenically very distant from both the 1918 Spanish flu and the A(H1N1)pdm09. Thus, when the A(H1N1)pdm09 emerged, immunity was lacking in the younger to middle-aged population [39]. Despite the difference in circulation history between H1N1 and H3N2 subtypes in human population, both subtypes are subjected to continual pressure to escape from the human immune system.

In this review, we will discuss the interplay between immune evasion associated with influenza virus, the countermeasures offered by the humoral immune system to combat ongoing variation in influenza viruses, and the requirement for the virus to maintain function throughout this complex evolutionary process. We will mostly focus on the HA protein due to its role as the main antigen of influenza virus and the major seasonal vaccine target, but will also touch on the emerging role of anti-neuraminidase antibodies. Throughout this review, all residues on HA are named according to H3 numbering.

Antigenic drift: point mutations and glycosylation

Most antibodies elicited by influenza virus by natural infection and vaccination target the globular head domain of HA1, which is distal from the virus surface and readily accessible for immune recognition. Influenza virus HA, in turn, mutates to escape from pre-existing immunity. This mutation-based immune evasion process is known as antigenic drift. Early studies proposed five major antigenic sites in the HA1 globular head domain for both H1 and H3 HAs [40–44], namely Sa, Sb, Ca1, Ca2, and Cb for H1 HA [40, 41] (Fig. 1a), and sites A-E for H3 HA [42–44] (Fig. 1b). The locations of Sa, Sb and Ca2 of H1 HA, and antigenic sites A and B of H3 HA partially overlap with the RBS, whereas Ca1 and Cb of H1 HA, and antigenic sites C-E of H3 HA are more distant from the RBS. These antigenic sites provide a structural framework to understand the evolutionary dynamics and constraints of influenza virus in response to humoral immunity.

Figure 1. Major antigenic sites on HA.

(a) The locations of the five major antigenic sites on H1 HA [40, 41] are shown on the HA trimer structure of A/California/04/2009 (PDB 3UBE) [84]. (b) The locations of the five major antigenic sites on H3 HA [42–44] shown on the HA trimer structure of A/Hong Kong/1/1968 (PDB 2YPG) [85]. (a–b) The human receptor analog pentasaccharide lactoseries tetrasaccharide c (LSTc) is shown in sticks representation (yellow).

Over the past half-century, the HAs from human H3N2 viruses have accumulated at least 75 substitutions (13% of the entire protein) (Fig. 2a–b). Interestingly, most mutations that account for major antigenic drift in human H3N2 viruses reside within or immediately proximate to the RBS [45–48] (Fig. 2c). Similar observations are found in the recent antigenic drift of A(H1N1)pdm09 and type B viruses [48–50]. Thus, there seems to be a preference for neutralizing antibodies to target the RBS-proximal region. One explanation is that targeting the RBS-proximal region can sterically hinder receptor binding to achieve neutralization. Another plausible explanation is the location of the glycosylation sites on the HA. In both H1N1 and H3N2 viruses, the number of N-glycosylation sites on the HA1 globular head domain has substantially increased since they were introduced into the human population [51–58]. For example, the number of N-glycosylation sites in the HA1 globular head domain of the human H1N1 virus increased from one to at least three sites per protomer (3 to 9 per trimer) from 1918 until its temporary eradication from the human population in 1957 [56, 57]. Similarly, the number of N-glycosylation sites in the HA1 globular head domain of the human H3N2 virus has constantly increased with up to 10 new sites being added over the past half-century [57–59], but only at least six additional sites per protomer (18 per trimer) being present in any circulating strain. Including the 5 highly conserved glycosylation sites per protomer (15 per trimer), the end result is a total of at least 33 glycosylation sites per trimer in current H3N2 viruses (Fig. 2d–e). Of note, some glycosylation sites only appear temporarily during the evolution of human H3N2 virus, such as Asn81, Asn276, and potentially Asn144. Accretion of conformationally flexible oligosaccharides on the HA further shields the antigenic sites to facilitate immune evasion [43, 59–62]. Nonetheless, the added N-glycosylation sites can also affect binding of the natural receptor when they are proximal to the RBS [54, 63–67]. The functional constraints of the RBS render it difficult, if not impossible, to be completely shielded by oligosaccharides. As an increasing proportion of the HA1 globular domain becomes masked by oligosaccharides, the RBS and its immediate proximal region becomes relatively more vulnerable to be targeted by antibodies which have specific features, such as long CDR loops (see below). This may explain why the more exposed RBS-proximal antigenic site B (Fig. 1b) has become immunodominant in recent years [47, 68].

Figure 2. Natural substitutions and emerging N-glycosylation sites in the HA of human H3N2 viruses.

(a) Stacked graph showing the cumulative average number of amino-acid substitutions in the HA in strains isolated from different years as compared to the ancestral strain, A/Hong Kong/1/1968. (b–c) HA residues that have different amino-acid identities between A/Hong Kong/1/1968 and the consensus sequence of the 2015 human H3N2 strains are shown as spheres on one protomer of A/Hong Kong/1/1968 HA trimer (PDB 4FNK) [69]. The other two protomers are shown in surface representation. Color scheme follows that of panel a. (c) Zoom-in view for the receptor binding site (RBS) of A/Hong Kong/1/1968 in complex with a human receptor analog pentasaccharide lactoseries tetrasaccharide c (LSTc) (PDB 2YPG) [85]. LSTc is shown as yellow sticks (carbons) with nitrogens in blue and oxygens in red. Residues within the RBS are labeled. Residues responsible for major antigenic changes are labeled with asterisks [48]. (d) Stacked graph showing the normalized percentage of strains that contain the indicated N-glycosylation sites. While the N-glycosylation site Asn81 disappeared in year 1977, nine other N-glycosylation sites have emerged in human H3N2 viruses since 1968. Five absolutely conserved N-glycosylation sites in human H3 strains (HA1 Asn22, Asn38, Asn165, Asn285; HA2 Asn154) are not included in the stacked graph. This analysis includes a total of 4625 sequences were downloaded from Influenza Research Database [190]. This plot is an updated version of the plot in Lee et al., 2014 [58]. (e) Oligomannoses (cyan) on A/Hong Kong/1/1968 (PDB 4FNK) [69] and A/Victoria/361/2011 (PDB 4O5N) [58] were modeled by Glyprot [191]. HA1 is colored in grey and HA2 is colored in black. The receptor-binding site (RBS) is colored in lime. The absolutely conserved N-glycosylation sites in human H3 strains (HA1 Asn22, Asn38, Asn165, Asn285; HA2 Asn154) are included in the display.

Functionally conserved region 1: Receptor-binding site

As a functionally conserved region on the HA1 globular head domain, the RBS is a target for broadly neutralizing antibodies (bnAbs) [58, 69–81] (Fig. 3a–c). RBS-targeted bnAbs inhibit viral entry by preventing HA binding to its host receptor, and may also prevent HA conformational changes during membrane fusion by cross-linking neighboring subunits of the HA trimer [79]. Common features of RBS-targeted bnAbs have been gleaned through crystal structure determination of multiple antibody HA complexes [74] (Fig. 4a). These characteristics involve mimicking the binding mode of sialic acid to some extent using a single complementarity determining region (CDR) loop [78]. The most prevalent mimicry strategy is to insert a hydrophobic amino acid into the hydrophobic pocket that interacts with the acetamide group of sialic acid [78]. Many RBS-targeted bnAbs also interact with a polar pocket that is occupied by the sialic acid carboxylate group that forms numerous hydrogen bonds to the HA. Some antibodies mimic this interaction precisely by inserting an aspartic acid at the tip of one of its CDRs (H2 or H3) into this binding pocket [75, 76, 78–82].

Figure 3. Binding footprints of RBS-targeted bnAbs.

(a) Amino-acid sequence conservation is projected onto the HA structure of A/Hong Kong/1/1968 (PDB 4FNK) [69]. One protomer is shown in Cα ribbon representation and the other two protomers are shown in a surface representation. Human H1, H2, and H3 strains were used for computing the amino-acid sequence conservation. For each subtype, we randomly sampled at most 5 strains from each year for this analysis. (b) The sequence conservation of the HA RBS region of A/Hong Kong/1/1968 (PDB 4FNK) [69] is shown. Several highly conserved residues that interact with the sialic acid receptor are labeled. (c) The binding footprints of RBS-targeted bnAbs are shown on the structure of the HA RBS of A/Hong Kong/1/1968 (PDB 4FNK) [69]. Epitope usage represents the frequency of a specified residue being targeted by RBS-targeted bnAbs. Epitopes from nine RBS-targeted bnAbs, namely C05 [69], S139/1 [71], F045–092 [58], 2G1 [74], 8M2 [74], 5J8 [76], 1F1 [192], HC63 [79], and CH65 [75], were analyzed here. Locations of naturally occurring indels are indicated. (d–e) Same as panel b and c, except that the stem region is now shown. N38, which is a highly conserved N-glycosylation site in group 2 HAs, but not in group 1 HAs, is labeled. (e) Epitopes from nine stem-binding bnAbs, namely CR6261 [99], CR9114 [101], CR8043 [109], F10 [100], CR8020 [107], C179 [102], MEDI8852 [105], 39.29 [113], and FI6v3 [111], were analyzed here. Amino-acid substitutions associated with stem-binding bnAbs escape were annotated. HA1: T318K (C179 escape) [103]. HA2: D19N (CR8020 escape) [107], R25M (CR8043 escape) [109], G33E (CR8020 escape) [107], Q34R (CR8020 and CR8043 escape) [109], Q42K (39.29 escape) [118], D46Y/G (39.29 escape) [118], V52E (C179 escape) [103], and H111L (CR6261 escape) [108]. Of note, HA1 H111L is a buried behind the surface. “Site 1” indicates the epitope region of CR9114, CR6261, and other V_H1–69 encoded stem-binding bnAbs. “Site 2” indicates the epitope region of the two group 2-specific stem-binding bnAbs, namely CR8020 and CR8043. Site 2 is lower down the stem as compared to site 1. “Both” indicates the overlapping region of sites 1 and 2.

Figure 4. Binding of broadly neutralizing antibody (bnAb) to influenza hemagglutinin (HA).

(a) Interaction between RBS-targeted bnAbs and HA is shown. Binding of C05 [69] (cyan; PDB 4FP8), S139/1 [71] (pink; PDB 4GMS), F045–092 [58] (lime; PDB 4O58), 2G1 [74] (purple; PDB 4HG4), 8M2 [74] (orange; PDB 4HFU), 5J8 [76] (blue; PDB 4M5Z), 1F1 [192] (green; PDB 4GXU), HC63 [79] (gray; PDB 1KEN), and CH65 [75] (red; PDB 5UGY) Fabs to HA trimer (white). (b) Interaction between stem-binding bnAbs and HA is shown. Binding of CR6261 [99] (yellow; PDB 3GBN), CR9114 [101] (blue; PDB 4FQI), CR8043 [109] (orange; PDB 4NM8), CR8020 [107] (cyan; PDB 3SDY), C179 [102] (red; PDB 4HLZ), MEDI8852 [105] (pink; PDB 5JW4), 39.29 [113] (green; PDB 4KVN), and FI6v3 [111] (gray; PDB 3ZTJ) Fabs to HA trimer (white). The curved arrow indicates that the approach angle has to be from no more than perpendicular to at an upward disposition to avoid a steric clash with the membrane. CR6261 Fab is obscured by CR9114 Fab, as they both have very similar angles of approach to the HA and, therefore, is not visible in these views. (c) The approach angle of RBS-targeted bnAbs and stem-binding bnAbs was quantified by adapting the method described in [115]. Briefly, the trimer axis of HA is on the z-axis such that the x-y plane (horizontal plane) represents the viral membrane. Vertical plane represents the y-z plane. The long axis of the Fab is defined as the line connecting the averaged coordinate of Cα-atoms of the conserved cysteines in the constant domain to that in the variable domain. The angle between the long axis of the Fab and the horizontal plane or the vertical plane is shown. (d) Binding of H5M9 [128] (wheat; PDB 4MHJ) and CR8071 [101] (teal; PDB 4FQJ) Fabs to the same protomer of the HA trimer (white). To visually compare their epitope locations with other bnAb epitopes, the RBS and the epitope of a stem-binding bnAb CR6261 are colored in lime and pink, respectively.

While the portion of the RBS that interacts with the sialic acid of sialoside receptors is very highly conserved, some amino-acid variation can be observed in natural circulating strains in the other regions of the RBS that interact with the other sugar moieties of the receptor. Structural and functional characterization has also revealed how HAs from human H1N1 and H3N2 viruses interact differently with human receptor analogs [83–91]. These differences are mostly attributed to amino-acid substitutions at residue 190 in the helix atop the RBS and to the 220-loop that interacts with Gal-2 of the receptor analog. Furthermore, evolution of residues in the 150-loop, 190-helix, and 220-loop in recent human H3N2 viruses are observed [85], which has resulted in bias of the receptor preference towards long, branched sialylated glycans [92]. However, despite those structural variations among strains and subtypes, certain RBS-targeted bnAbs are able to neutralize virus strains from both human H1 and H3 subtypes [58, 69–72]. In addition, one pan-H3 bnAb F045–92 is able to neutralize all tested H3 strains isolated over five decades [58, 72]. These observations suggest that the neutralization activity of the RBS-targeted bnAbs can be robust to many structural variations within or proximal to the RBS.

Besides the small binding footprint that is preferred for bnAbs to specifically target the relatively small RBS [58], which thereby restricts the interaction to the most conserved region of this site, the neutralizing activity of RBS-targeted bnAbs is promoted by an avidity effect [69, 71, 76]. The bivalency of IgG allows RBS-targeted bnAbs to enhance neutralizing activity and to increase breadth against strains that interact weakly with its Fab (fragment antigen binding) only. This avidity effect relies on the high density of HAs on influenza virus [93, 94]. While the high density of receptor-binding proteins is not a universal feature for all viruses [95], it promotes high receptor-binding avidity during virus entry and is critical for influenza virus due to the extremely weak monovalent binding between sialylated glycan and HA (K_d in low mM range) [96]. As a consequence, this functional constraint allows RBS-targeted bnAbs to accommodate the changing landscape of the RBS and increase neutralization breadth by avidity.

One common obstacle that restricts the neutralization breadth of RBS-targeted bnAbs is the insertions and deletions (indels) near the RBS (Fig. 3c). The 133a insertion (between residues 133 and 134), which is common in H1 and H5 strains, is perhaps the most well-known example. This insertion produces a bulge in the 130-loop. Binding of RBS-targeted bnAbs CH65 [75], C05 [69], S139/1 [71], and F045–092 [58] are abolished by the 133a insertion. In addition to 133a insertion, many isolates from subtypes H4, H6, H7, H10, H14, and H15 have single (158a) or double (156a/158b) amino-acid insertions in the 150-loop that can negatively influence the binding of C05 [69], 139/1[71], and F045–092 [58]. Some H7 viruses also have an 8-residue deletion in the 220-loop [97, 98], which may decrease the contact surface and hence affinity of RBS-targeted bnAbs. Although it is much more difficult for the virus to acquire indels than point mutations, it is necessary for the development of an RBS-targeted universal vaccine or therapeutic to account for potential indels.

Functionally conserved region 2: Membrane fusion

The stem region is the other functionally conserved region on HA. As compared to the RBS, there is much less sequence variation in the stem region across strains and subtypes. In fact, the stem region contains the most conserved epitopes across HA for antibody recognition (Fig. 3a, d and e). As a result, stem-binding bnAbs can acquire higher neutralization breadth than RBS-targeted bnAbs [99–113]. These stem-binding bnAbs inhibit virus replication by preventing the pH-induced conformational changes that are essential for membrane fusion. Unlike the RBS, the stem region does not have a natural interacting partner. Nevertheless, similar to RBS-targeted bnAbs, a number of stem-binding bnAbs also share recurring motifs [114]. Due to the proximity to the viral membrane, stem-binding bnAbs have less conformational freedom for approach angles as compared to RBS-targeted bnAbs (Fig. 4a–b). They have to approach the stem perpendicularly or at an upward disposition as otherwise they would clash with the membrane. The difference between the approach angles of RBS-targeted bnAbs and stem-binding bnAbs can be quantified using a method that was previously employed to compare approach angles in HIV bnAbs [115] (Fig. 4c). The approach angle of RBS-targeted bnAbs has a range (maximum – minimum) of 37° in the horizontal plane and 60° in the vertical plane, whereas stem-binding bnAbs have a range of 29° in the horizontal plane and 39° in the vertical plane. This analysis supports the notion that the approach angle of stem-binding bnAbs.is more restricted due to proximity to the membrane.

Stem-binding bnAbs are often encoded by the V_H1–69 germline gene, which is characterized by encoding hydrophobic residues at positions 53 (Ile) and 54 (Phe) in CDR H2, as well as an aromatic Tyr98 in CDR H3 [99–101, 114]. These three residues insert into the hydrophobic groove in the stem region and are critical for high affinity binding. In fact, the V_H1–69 germline gene has been proposed as a critical “SOS component” of the antibody repertoire that allows rapid and robust response against influenza and other infections [116, 117]. However, several stem-binding bnAbs that are encoded by V_H1–69 germline gene neutralize group 1 but not group 2 influenza subtypes [78, 99, 100, 108]. The problem of neutralizing group 2 influenza subtypes has been attributed to an N-glycosylation site at HA1 Asn38, which is immediately adjacent to the conserved epitope [99, 100] (Fig. 3d). This N-glycosylation site at Asn38 is conserved among group 2 HAs but is not present in group 1 HAs. Binding of these bnAbs to HA can be sterically hindered by the oligosaccharide at Asn38. In fact, two group 2-specific stem-binding bnAbs, namely CR8020 [107] and CR8043 [109], find a solution by recognizing an epitope lower down the stem, which may avoid a direct clash with the oligosaccharide at Asn38. Another stem-binding bnAb MEDI8852 binds to an epitope slightly higher than that of CR8020 and CR8043 and can neutralize both group 1 and 2 subtypes [105]. These bnAbs do not use V_H1–69 germline gene. CR8020, CR8043, and MEDI8852 are derived from V_H1–18, V_H1–3, and V_H6-1 germlines, respectively.

In comparison, the V_H1–69 encoded stem-binding bnAbs analyzed to date are more group 1-specific. Nonetheless, a V_H1–69 encoded stem-binding bnAb CR9114 has been identified that binds both group 1 and 2 influenza subtypes by interacting with a similar epitope as the group 1-specific V_H1–69 encoded stem-binding bnAbs but accesses this site through reorientation of the group-2 specific glycan at HA1 Asn38 [101]. This binding feature is also observed in a V_H3–30 encoded stem-binding cross-group bnAbs FI6v3 [111]. Overall, these observations suggest that Asn38 is a major obstacle for stem-binding bnAbs to overcome against influenza A group 2 and also influenza B viruses. At the same time, it demonstrates that N-glycosylation is critical not only for shielding the HA1 globular head domain, but also for interfering with recognition of the HA surface by stem-binding bnAbs.

While the shielding effect from oligosaccharides in the stem region can be overcome by some stem-binding bnAbs, HA mutants that are able to escape from stem-binding bnAbs have been identified by extensive passaging, although they are much fewer than for head binding antibodies [103, 106–109, 118] (Fig. 3e). For example, passaging an H3N2 virus in the presence of 39.29, a stem-binding bnAb that can neutralize both group 1 and 2 influenza viruses [113], has led to identification of three escape mutants [118]. One mutant completely abolished binding of antibody 39.29 to the HA stem, whereas the other two mutants escape by increasing fusion efficiency [118]. Other examples include escape mutants against stem-binding bnAbs C179 [103], CR8020 [107, 119], CR6261 [108], and CR8043 [109]. Although some escape mutants against stem-binding bnAbs have been shown to reduce viral replication fitness [106, 118], compensatory mutations may exist. Although stem-binding bnAbs have provided valuable insights into vaccine design [120–122] and computational design of antiviral proteins [123, 124], we should always be prepared for the possibility of emerging escape mutants in circulating influenza viruses.

Viral replication fitness effect of thousands of influenza virus mutants can be measured in parallel using deep mutational scanning [29, 30, 125]. A proof-of-concept study has shown that deep mutational scanning can be applied to rapidly identify antibody escape mutants [126]. This technique will be useful in mapping bnAb escape mutants, which will facilitate a better mechanistic understanding of potential escape strategy, if any. Recently, we demonstrated that many single amino-acid substitutions in the RBS alone impose a detrimental effect on viral replication fitness, but are neutral when combined together [127]. This phenomenon is known as epistasis. Interestingly, a number of the combinatorial mutants can escape RBS-targeted bnAb S139/1 [71, 127]. Whether such epistatic effects can be found in the stem region and whether it will promote stem-binding bnAb escape remain to be explored.

Other conserved antibody epitopes

Isolation and characterization of broadly neutralizing or cross-reactive antibodies have facilitated the discovery of other conserved region as antibody epitopes. H5M9 [128] and CR8071 [101] are bnAbs that can neutralize diverse H5N1 strains and influenza B strains, respectively. H5M9 is able to block both receptor binding and membrane fusion steps [128], whereas CR8071 inhibits the viral budding process [101]. Interestingly, they both bind to the vestigial esterase subdomain of HA1 that is conserved within a given subtype or in type B [101, 128, 129], but not across all subtypes (Fig. 3a and Fig. 4d). A fully functional esterase domain is present in the influenza C virus haemagglutinin-esterase-fusion glycoprotein (HEF) [130–132], which also contains the RBS and membrane fusion machinery. The esterase domain in influenza C virus HEF is responsible for cleaving the host receptor to facilitate viral budding [133], which is analogous to the function of NA in influenza A and B viruses. When influenza A and B viruses diverged from influenza C viruses, which is estimated to be around 4,000 years ago [21], a separate receptor-destroying enzyme, the neuraminidase (NA) emerged and the esterase subdomain became obsolete.

Notwithstanding, while the evolutionary constraints associated with epitopes of RBS-targeted bnAbs and stem-binding bnAbs are intuitive, they are not as clear for the HA vestigial esterase subdomain in influenza A and B viruses. Similarly, sera before and after immunization with a trivalent seasonal influenza vaccine in young adults were analyzed and a large fraction provided in vivo protection against H1N1 and H3N2 viruses by binding to another conserved epitope in the head domain but without any in vitro neutralizing activity [134]. The in vivo protection effect for these types of antibodies are likely to rely on the effector functions of the Fc domain, such as antibody-dependent cell-mediated cytotoxicity (ADCC), as for HA anti-stem antibodies [101, 111, 119, 120, 135–137]. The evolutionary constraints of these types of epitopes require further investigation and will be important for evaluating the genetic barrier for the emergence of escape mutants.

Protective immunity against zoonotic influenza viruses

Pandemic threats from highly pathogenic zoonotic influenza viruses have been a global concern. In recent years, avian influenza virus H5N1 and H7N9 subtypes have caused hundreds of human infections with a mortality rate of 20% to 60% [9]. Studies have demonstrated that “original antigenic sin” [138] is a key determinant for the susceptibility of zoonotic influenza viruses [138–140]. “Original antigenic sin” arises when our immune system employs immunological memory based on an earlier infection by a virus with an slightly different antigenicity, which then skews the current immune response towards antibodies that were previously produced and to epitopes recognized during previous infections [138]. During 1918 Spanish flu H1N1 pandemic, people aged between 15 and 35 had an atypically high fatality rate [141, 142]. Based on phylogenetic, seroarcheological, and epidemiological evidence, this atypically high fatality rate was proposed to be associated with childhood exposure of H3N8 virus, which was antigenically distinct from Spanish flu [139, 143, 144]. On the other hand, it is suggested that those that were born before or after this age group were exposed to H1 or H1-like influenza viruses during childhood [139]. These studies implied higher protection against strains that are antigenically similar to those that were encountered during childhood.

In fact, a serological study on seasonal H3N2 viruses has indeed shown that individuals produce high antibody titers towards strains that were encountered during childhood and that the antibody titer was less against strains that subsequently emerged [145]. Based on statistical modeling, such childhood imprinting is also important for determining the susceptibility against emerging avian influenza viruses H5N1 and H7N9 subtypes [140]. Briefly, those who were infected with seasonal H1 or H2 subtype (group 1) during their first encounter of influenza virus are less susceptible to H5 viruses (group 1), whereas those who were infected with seasonal H3 subtype (group 2) during their first encounter of influenza virus are less susceptible to H7 viruses (group 2) [140]. Overall, these observations highlight the importance of “original antigenic sin” in protection against zoonotic influenza viruses. They also reaffirm that seasonal influenza virus infection or vaccination can elicit heterosubtypic antibodies against antigenically similar zoonotic influenza viruses [108, 146–149].

Nonetheless, heterosubtypic antibodies are only infrequently elicited by seasonal influenza vaccination [147–150]. Even if they are elicited, such immunity may not be sustainable due to “original antigenic sin”. For example, while HA stem-binding bnAbs were commonly elicited during the 2009 pandemic H1N1 influenza season [151], this response was not sustained in subsequent years [152, 153]. Given the significance of “original antigenic sin”, sustainable protection against zoonotic influenza virus infection likely requires the induction of heterosubtypic antibody responses during childhood. Furthermore, as most of the epitopes targeted to influenza virus during natural infection or vaccination are against the highly variable HA1 head, heterosubtypic response may be more readily triggered by focusing on the highly conserved HA stem region [154–158]. Over the past few years, several immunogens have been designed based on the conserved HA stem region [120–122, 157, 159–162]. All of these immunogens are able to consistently elicit heterosubtypic antibody responses [120–122, 157, 159–162], which then provide a framework for making progress towards a universal influenza vaccine.

Emerging role for neuraminidase in the humoral immune response

While both HA and NA are targets for the humoral immune response, the antigenic properties of the NA are much less well characterized as compared to the HA. The number of NAs is around 5- to 10-fold less than HAs on the viral surface [163, 164]. However, NA is required for completing the viral replication cycle by cleaving host receptor during viral budding and, hence, is a key target for antivirals. Although somewhat slower than HA, the adaptation rate of NA is also very high [19, 20, 165]. As observed in HA, NA evolution can also result in antigenic drift [166]. Over the past 30 years, several epitopes on NA have been described based on structural characterization [167–172] and mutagenesis studies [173–176]. Some anti-N1 antibodies are able to provide cross-reactivity against both human H1N1 and avian H5N1 viruses [175, 177], or against both seasonal H1N1 and A(H1N1)pdm09 viruses [178]. Furthermore, one antibody that targets a highly conserved region on NA can protect against influenza viruses from N1-N9 subtypes [176]. Emerging evidence suggests that serum titer of anti-NA antibodies also correlates with protection against influenza infection and can be an independent predictor for disease outcome [179–181]. Yet, the role of NA as a vaccine immunogen remains to be explored [182].

There is a tight interplay between HA receptor-binding function and NA receptor-destroying activity. Experimental studies have shown that a functional balance between HA and NA is required for efficient viral replication [183–186]. When the functional activities of HA and NA are not in balance, mutations often rapidly arise, either in HA or NA or both, to restore the optimal functional balance [185–188]. Furthermore, certain mutations in the HA can compensate against the fitness cost of an oseltamivir-resistant mutation on NA [189]. Given that mutations that cause immune evasion can impact the protein functional activity (as seen in HA), antigenic drift in either HA or NA may also drive mutations in the other protein. However, further studies are needed to comprehend more on how the functional balance between HA and NA influences antigenic drift.

Concluding Remarks

It has been almost four decades since the crystal structure of influenza virus hemagglutinin (HA), which is the major antigen of influenza virus, was determined [34] that allowed an understanding of the effects of natural variation and structural identification of the major antigenic sites [42]. The high evolutionary capacity allows influenza to constantly escape natural immunity. However, functional constraints on receptor binding and membrane fusion that are required for viral entry result in sequence conservation of the receptor-binding site and the stem region. Recent identification and characterization of human influenza broadly neutralizing antibodies (bnAbs) have provided important insights into the development of antivirals and vaccines against these functional regions. Identification of bnAbs against the vestigial esterase subdomain and neuraminidase (NA) can further expand the anti-influenza arsenal. As the knowledge from evolutionary biology, immunology, structural biology, and virology of influenza virus continues to accumulate, we are hopefully moving closer towards better control of influenza with a universal vaccine that confers long-term immunity.

Highlights.

• Antigenic variation is a major challenge for influenza vaccine design

• Comprehending evolutionary constraints for influenza function is critical for vaccine development

• Broadly neutralizing antibodies reveal conserved epitopes

• Eliciting long-term heterosubtypic immunity is the ultimate goal for vaccine design