Structural insights into the design of novel anti-influenza therapies

Abstract

A limited arsenal of therapies is available to tackle the emergence of a future influenza pandemic or even to effectively deal with the continual outbreaks of seasonal influenza. However, recent findings hold great promise for design of novel vaccines and therapeutics, including the possibility of more universal treatments. A major contribution to those advances comes from structural biology, in particular from the many studies on influenza hemagglutinin (HA), the major surface antigen. The HA primary function is to enable the virus to enter host cells, and structural work has revealed the various HA conformational forms generated during the entry process. Other studies have explored how human broadly neutralizing antibodies (bnAbs), designed proteins, peptides and small molecules, can inhibit and neutralize the virus. Here we review milestones in HA structural biology and how the recent insights from broadly neutralizing antibodies are leading to design of novel vaccines and therapeutics.

Introduction

Influenza virus is a negative-strand RNA virus that contains eight RNA segments, encoding at least 12 proteins (PB2, PB1, PB1-F2, PA, PA-X, HA, NA, NP, M1, M2, NS1, and NS2). Two of these proteins, hemagglutinin (HA) and neuraminidase (NA), are cell surface glycoproteins that enable the virus to enter and to escape from host cells, respectively. HA is a homotrimer that is synthesized as a single polypeptide chain (HA0), which is subsequently cleaved by host cell proteases to attain its fusion-competent state. The mature HA trimer is therefore composed of HA1 and HA2 subunits that remain cross-linked after cleavage through a single disulfide bond. The HA trimer can be divided into two structural as well as functional domains, the head and the stem, that comprise the receptor-binding site (RBS) and the fusion machinery, respectively. NA is an enzyme that cleaves the sialoside receptor off from the cell surface and enables progeny virus to escape from the infected cell to subsequently infect new cells. Both HA and NA activities are essential for viral infection. However, HA greatly outnumbers NA on the virus surface and consequently is the main target of the humoral immune response. Nevertheless, NA is the primary target for the anti-influenza drugs oseltamivir and zanamivir¹ due to the ability to more readily target the NA active site compared to the much shallower HA RBS.

Here we review progress on the structural and functional characterization of HA in particular with human broadly neutralizing antibodies (bnAbs), which have provided exciting new insights and stimulated structure-based design of novel vaccines and new classes of therapeutics to target influenza virus.

Hemagglutinin structure and function

Influenza A viruses have 18 different HA subtypes (H1–18), whereas influenza B viruses have two different lineages (Yamagata and Victoria lineages). The natural reservoir for influenza viruses are wild aquatic birds, and 16 of these 18 HA subtypes (H1–H16) are resident in the bird population. Genomic RNAs of the other two influenza A subtypes (H17 and H18) have recently been found in bats^{2, 3}, although live virus of these two subtypes has yet to be isolated. The influenza virus HA structure (H3 subtype from the 1968 influenza pandemic) was first determined in 1981, and was also the first viral antigen from an enveloped virus to be described^{4, 5} (Fig. 1). The identification of substitutions in HA that account for the differential recognition of avian-type versus human-type receptors (α2–3 versus α2–6 linked sialosides) in 1983⁶ facilitated structural determination of HA-bound receptor complexes with sialic acid analogues in 1988⁷. Another unknown was whether the precursor HA0 undergoes substantial conformational changes when converting to its fusion-competent form, as HA1 and HA2, and that was answered in 1988: the HA0 structure revealed surprisingly few differences between the cleaved and uncleaved forms⁸, other than at the cleavage site, which seem to differ from the larger changes suggested recently for some other viral envelope proteins, such as HIV-1 Env⁹. The next burning question was what conformational changes in HA lead to its membrane fusion activity in the low pH of endosomal compartments. The structure of a fragment of the HA stem in 1994 showed the massive rearrangements that HA undergoes to acquire its post-fusion form¹⁰.

Figure 1. Milestones of influenza HA structural biology.

Time line plotting crystal structures that represent important contributions to our understanding of HA structure and function, and key discoveries of human heterosubtypic bnAbs that have led to development of HA-targeted antivirals and vaccines. HA stalk is shown in cyan; HA head in dark gray; broadly neutralizing antibodies (bnAbs) in pink. Sialic acid receptors, TBHQ, and Arbidol are shown in sphere representation (carton: yellow, oxygen: red, nitrogen: blue, sulfur: orange).

Further questions arose as to whether there are substantial structural differences in the HAs from subtypes in influenza viruses responsible for the other pandemics, such as the 1918 H1N1 Spanish influenza^{11, 12}, 1957 H2N2 Asian influenza^{13, 14}, and 2009 H1N1 swine-origin pandemic^{15, 16, 17}, or in the more recent emerging viruses of concern to human health, such as H5N1^{18, 19}, H7N9^{20, 21, 22}, H10N8^{23, 24, 25}, and H6N1^{24, 26, 27}, as well as influenza B HA and other influenza A subtypes^{3, 19, 28, 29, 30, 31, 32}. This compendium of structures indicated that the HA architecture is highly conserved, but the surface properties and glycosylation patterns differ extensively among influenza subtypes and types. Furthermore, as the virus evolves after entering the human population, the glycans cover more of the HA surface^{33, 34}. Influenza A viruses can also be classified phylogenetically into group 1 (H1, H2, H5, H6, H7, H8, H9, H11, H12, H13, H16, H17, and H18) and group 2 (H3, H4, H7, H10, H14, and H15). The main functional differences between these groups arise in the different ways that they interact with their receptor³⁵ and in the definition of their epitopes^{35, 36}, especially in the head region.

HA stem binders

Despite these advances, it was not known until relatively recently whether human antibodies could be found that are able to broadly neutralize influenza virus. The big breakthrough came in 2008–2009 with the isolation and characterization of bnAbs that could neutralize most influenza A group 1 viruses^{37, 38, 39} (Fig. 1). Surprisingly, those antibodies, such as CR6261 and F10, did not bind to the hypervariable head, as do the strain-specific antibodies that are normally elicited by infection and vaccination; instead, they were found to target the much more highly conserved and less accessible stem region using only the variable heavy chain for interaction^{39, 40}. Thus, the HA stem suddenly became of major interest as a bnAb target and suggested that a more universal vaccine might be possible. Indeed, back in 1993, a mouse antibody with some heterosubtypic breadth (C179) had been reported to target the HA stem region⁴¹, but this antibody was largely overlooked by researchers in the influenza field. A crystal structure of C179 some 20 years later confirmed that it bound to the same stem region as these human bnAbs⁴².

Further extensive searches then ensued for other human antibodies that could target the HA stem with binding profiles that differed from the first set of bnAbs, which were all derived from the same V_H1–69 germline family and exclusively targeted group 1 HAs. The Crucell group discovered another antibody (CR8020), whose heavy chain is encoded by V_H1–18 germline gene⁴³. CR8020 binds slightly lower down the stem than the original bnAbs and targets group 2 viruses (Fig. 2). The next major development was the isolation by the Lanzavecchia group of antibody FI6v3, which can target all group 1 and group 2 HAs⁴⁴. FI6v3 arose from a different germline gene family, V_H3–30, and targets the same stem region as the V_H1–69 antibodies, but with a completely different binding mode and in more conventional way with interactions from both the light and heavy chain variable regions (Fig. 2). In another major step forward, Crucell then identified an antibody (CR9114) that can bind to both influenza A and influenza B viruses⁴⁵ (Fig. 2). Such an antibody represents the epitome of what one would want to elicit in a universal vaccine. All of these broadly neutralizing antibodies inhibit the membrane fusion activity of HA by stabilizing the pre-fusion form of the HA trimer and preventing the conformational changes induced by low pH that release the fusion peptide^{39, 41, 43, 44, 45}.

Figure 2. Anti-HA broadly neutralizing antibodies (bnAbs).

(a) Structures of representative RBS-targeted (C05 (PDB 4FP8)⁵⁰ and F045–092 (PDB 4O58)⁵¹) and stem-binding (CR9114 (PDB 4FQI)⁴⁵ and FI6v3 (PDB 3ZTJ)⁴⁴) bnAbs in complex with HA are shown. (b) The epitopes of four stem-binding bnAbs, namely CR8020 (PDB 3SDY)⁴³, CR6261 (PDB 3GBM)⁴⁰, FI6v3 (PDB 3ZTJ)⁴⁴, and CR9114 (PDB 4FQI)⁴⁵ are compared. The epitope of the indicated bnAb is colored lime. For comparison, regions that correspond to the epitope of the other three bnAbs are colored white. (c) The binding and neutralization breadths of CR8020⁴³, CR6261^{38, 40}, FI6v3⁴⁴, CR9114⁴⁵, F045–092^{49, 51}, and C05⁵⁰ are shown. The phylogenetic relationship of HAs from influenza B virus and from different influenza A virus subtypes is shown on the left. A check mark indicates that binding, neutralization activity, or in vivo protection have been demonstrated; otherwise, a cross is shown.

HA head binders

We will return later to stem binders but now focus on exciting new developments on bnAbs to the head region. It was not clear initially whether bnAbs could be found against the head, as that region is so hypervariable. However, a number of groups have now identified antibodies to the head that have greater breadth than the isolate-specific antibodies normally elicited by natural infection and vaccination. In 2009, the Takada group reported an antibody (S139/1) that cross-reacts with different influenza virus subtypes (H1, H2, H3, H5, H9, and H13) close to antigenic site B⁴⁶. A flurry of papers in 2011 also reported on head antibodies that have considerable breadth within the H1 subtype (e.g. 5J8⁴⁷ and CH65⁴⁸), and an exceptional pan-H3 antibody F045–092 that also had some activity against H1, H2, H5, and H13 subtypes⁴⁹ (Fig. 2). These head antibodies interact with the conserved RBS by inserting their H3 or H2 complementarity determining loops (CDRs) into the same site as the sialic acid receptor, explaining their breadth. Perhaps the archetype antibody against the RBS is C05, which essentially uses a single, long CDR H3 to engage the RBS, thereby minimizing contact with the hypervariable residues that surround the RBS⁵⁰ (Fig. 2). Indeed, the most effective of these bnAbs to the RBS have smaller binding footprints that emulate to some extent the limited surface engaged by the sialic acid receptor⁵¹. Despite the quite different angles of approach of these antibodies to the HA surface, analysis of CH65⁴⁸, CH67⁵², 5J8⁵³, F045–092⁵¹, 641 I–9⁵⁴, and H2526⁵⁴ revealed that a dipeptide corresponding to an aspartic acid-hydrophobic motif was the primary antibody motif for recognition of the receptor binding site^{54, 55}. Specifically, this dipeptide is located at the tip of the CDR loop that is inserted into the RBS. This observation further underscores that it is feasible to target the RBS for vaccine and therapeutic design.

Design of a more universal vaccine

A number of different families of antibodies have now been found to target the stem. Originally, most of these antibodies were from the V_H1–69 germline family, including CR6261 and CR9114, with a very hydrophobic CDR H2 that inserts two hydrophobic residues (Ile, Phe) at the tip into hydrophobic stem pockets. A conserved Tyr residue in CDR H3 provides the basis for the interaction with the HA stem. Together, these recognition elements, arising from CDR H2 (IF) and H3 (Y), form the so-called IFY motif^{56, 57}. Two other stem-binding bnAbs, FI6v3⁴⁴ and 39.29⁵⁸ of the V_H3–30 germline family, which bind the HA at a very different angle and use both light and heavy chains for interaction, compared to V_H1–69, where the heavy chain makes the majority, if not all, of the stem interactions. Nonetheless, V_H1–69 and V_H3–30 antibodies have very similar binding footprints and fill the same set of hydrophobic pockets in the HA stem⁵⁹. However, the V_H3–30 family uses an extended CDR H3 to occupy the same pockets as the IFY motif of the V_H1–69 antibodies. More recently, a number of vaccine-induced antibodies that arise from other heavy chain germline families have been shown to target this same site⁶⁰. Thus, these finding have led to the concept of a headless HA as an immunogen that would focus the immune response on the HA stem and away from the more immunogenic head region⁶¹.

The main problem was how to design and engineer such a construct, given that the headless HA region comprises three segments: the N and C- terminal regions of the HA1 chain, and the entire HA2 chain. In 2015, two such designs, namely mini-HA and headless HA, were reported by Janssen⁶² and the NIH Vaccine Research Institute (VRC)⁶³, respectively (Fig. 3a). These engineered immunogens were tested in mice, monkeys and ferrets. The immunized animals survived challenge with a lethal dose of virus, thus providing the all-important proof-of-principle that a protective response could indeed be generated against the HA stem by vaccination. Furthermore, the response was relatively broad: for example, in the immunization studies with the mini-HA #4900 design, a heterologous response was produced in mice against an H1 subtype (A/Puerto Rico/8/1934) that was quite distant from the template (A/Brisbane/59/2007) used for design of the immunogen⁶². Heterosubtypic responses were also elicited against viruses from another subtype (H5, A/Hong Kong/156/1997). Similar results were observed in cynomolgus monkeys: the mini-HA was able to protect against both heterologous and heterosubtypic viruses, in contrast to a trivalent seasonal vaccine that could generate only strain-specific responses⁶². Passive protection against heterosubtypic challenge (H5), by introducing serum from the immunized mice into naïve mice, was also observed. The VRC headless HA construct was also displayed on ferritin nanoparticles particles to enhance its immunogenicity⁶³ (Fig. 3b).

Figure 3. HA stem immunogen design.

(a) Structures of stem-binding bnAbs (CR6261 or CR9114) in complex with HA (PDB 3GBM)⁴⁰, Gen4 HA-SS construct (PDB 5C0S)⁶³, and mini-HA #4900 construct (PDB 5CJQ)⁶² are shown. (b) EM reconstruction of a nanoparticle that displays the Gen6 HA-SS construct (EMDB EMD-6332)⁶³. (c–d) The structural conformations of HA, Gen4 HA-SS (without the trimerizing gp41 peptide region), and mini-HA #4900 (without the trimerizing GCN4 region) are aligned by minimizing their RMSD. (c) Alignment was performed on a single protomer. (d) Alignment was also performed on the trimeric form. Although the protomers from both Gen4 HA-SS and mini-HA #4900 align well with HA, their trimeric forms adopt a more open, splayed conformation at the base as compared to native full-length HA.

Crystal structures of both mini-HA and headless HA designs indicated some splaying of the headless HA trimers at the base compared to the full-length HA (Fig. 3c–d), which is likely attributable to the engineering that involved additional trimerization domains (GCN4⁶² or a trimerizing peptide from HIV-1 gp41⁶³) to help stabilize the triple helices in the center of the stem. Notwithstanding, the neutralizing epitopes were faithfully preserved in these designs, as indicated by their strong binding to known bnAbs, such as CR6261 and CR9114.

Besides the mini-HA and VRC headless HA constructs, other stem-based immunogens have been designed and reported^{64, 65, 66, 67, 68, 69}. Another proposed approach to elicit heterologous and heterosubtypic immunity to the HA stem was the use of HA chimeras, where head domains from different subtypes are fused with a single stem domain, to direct the antibody response to the conserved stem region by sequential immunization with different chimeras^{70, 71}.

Overall, these stem immunogens are currently the most promising for development of a more universal vaccine and current work is in progress to increase the breadth of the response by further design.

Design of small proteins and peptides against the HA stem domain

Shortly after the first structures of stem bnAbs were determined, Baker and colleagues designed small proteins, inspired by the interactions between the antibodies and the HA stem. The concept was to select side chains to fill some of same pockets in the HA stem targeted by antibodies, and then identify protein scaffolds onto which those amino-acid side chains could be attached in an appropriate conformation and configuration to optimize binding. Two such designs, HB36 and HB80, bound to H1 and H5 HAs with low nanomolar affinity. The crystal structure of HB36 in complex with 1918/H1 HA showed the remarkable fidelity of the design⁷² (Fig. 4). The designs were then further optimized to produce a 51-residue protein (HB80.4) that was able to bind to all influenza A group 1 HAs and neutralize H1N1 viruses with potencies akin to the best bnAbs⁷³. The most recent advance in these small protein designs came from using a massively parallel approach, whereby 22,600 mini-proteins with different backbone scaffolds of 37–43 residues were screened against influenza HA⁷⁴. Binders with very high affinity (K_d < 10 nM) and stability (T_m > 95 ºC) were identified and, importantly, those designed proteins were not found to be immunogenic even after repeated injections in mice. One mini-protein design, HB1.6928.2.3, is smaller than the previous designs (Fig. 4) and represents an excellent alternative to bnAbs for prophylaxis and therapy.

Figure 4. Protein and peptide design that target the HA stem region.

Four different designs that target the HA stem region, namely HB36.3 (PDB 3R2X)⁷², HB80.4 (PDB 4EEF)⁷³, HB1.6928.2.3 (PDB 5VLI)⁷⁴, and P7 (PDB 5W6T)⁵⁹, along with CR9114 Fab (PDB 3GBM)⁴⁰ are shown in burgundy. HA is colored white. Their binding surfaces are colored lime in the bottom panel. Figure inspired in part from Whitehead 2017⁷⁵.

In another recent breakthrough, small peptides were designed and characterized against this same site in the HA stem, also based on how bnAbs target the stem, in a collaboration between Janssen and our laboratory⁵⁹. Using the CDR interacting loops of two bnAbs FI6v3 and CR9114 as templates, cyclic peptides that also incorporated non-proteinogenic amino acids were designed to fill the same hydrophobic pockets in the HA stem as the stem-binding bnAbs⁵⁹. The crystal structure of one of these peptides, P7, confirmed that it indeed targets exactly the same highly conserved region in the stem as the bnAbs. These 11-residue peptides bind with low nanomolar affinity to group 1 HAs and also inhibit the low-pH induced conformational changes that lead to membrane fusion. This design of a peptide mimic of antibody loops is perhaps the most successful design of an antibody surrogate⁷⁵.

Design of a small protein against the HA head

The mode of binding of bnAbs to the HA head has also inspired design of small proteins that mimic the key interactions of antibody C05⁵⁰ with the RBS. The peptide binding loop from C05 H3 was the basis for the design, where a scaffold (cystatin) that could incorporate the sequence of this loop in a similar conformation was used as the template⁷⁶ (Fig. 5). The design was further optimized by fusing three of these small protein binders with a natural homotrimer. In this way, the design (HSB.2A) matched the geometry of the HA trimer and substantially increased the avidity and reduced the off-rate by more than three-orders of magnitude. HSB.2A also had similar breadth and potency as C05 IgG and protected mice in a challenge study. This study now also provides further proof-of-concept that the information derived from how bnAbs target the RBS can be used to design a small protein that inhibits influenza virus.

Figure 5. Protein and peptide design that targets the HA receptor-binding site.

(a) Crystal structures of HA-C05 Fab (PDB 4FP8)⁵⁰ and HA-HSB.2A⁷⁶ complexes are shown. C05 Fab and HSB.2A are colored burgundy. (b) C05 and HSB.2A are colored pink. The loop region that interacts with the HA receptor-binding site is colored red. HA is colored white. HA receptor-binding site is colored light blue.

Small molecule antivirals against the HA stem

On the small molecule front, a small molecule named tert-butyl hydroquinone (TBHQ) had been shown to bind influenza HAs (with K_d of 5 to 50 μM) and inhibit its fusion activity⁷⁷. The crystal structure of TBHQ bound to H3 and H14 HAs revealed that it interacted with a hydrophobic pocket further up the stem region than the regions targeted by the bnAbs, but could still stabilize protomer interactions in the trimer, accounting for its inhibition of membrane fusion⁷⁸.

In 2017, the crystal structure of a broad-spectrum antiviral that has been sold as an over-the-counter drug in some countries since 1990, in Russia as Arbidol and in China as Enerxin, was determined in complex with HA⁷⁹. This drug is currently manufactured by Pharmstandard in Russia. Our laboratory found that Arbidol indeed interacts with influenza HA and binds in the pocket that was only partially occupied by TBHQ⁷⁸ (Figs. 1 and 6). This compound has modest binding affinity (K_d ~40–100 μM) to group 1 and group 2 influenza A viruses, and also acts as a fusion inhibitor. Crystal structures of H3 and H7 HAs with Arbidol showed that a water molecule was also present in the binding pocket⁷⁹. Structure-based design of Arbidol analogs to incorporate a hydroxyl at the position corresponding to the water molecule improved binding to 90–500 nM, depending on the HA⁸⁰. Thus, Arbidol analogs represent promising lead compounds for further improvement of drug binding and stability.

Figure 6. Inhibition of HA-mediated membrane fusion by small molecules.

The structural basis of two HA-mediated membrane fusion inhibitors, namely TBHQ (left, PDB 3EYK)⁷⁸ and Arbidol (right, PDB 5T6S)⁷⁹, has been characterized. HA1 is colored grey. HA2 is colored teal. TBHQ and Arbidol are shown as spheres representation and colored by atom type (yellow: carbon, red: oxygen, blue: nitrogen). The receptor-binding site and CR9114 epitope are colored lime and pink, respectively. The Cα of each resistance mutation is shown as a purple sphere. The change in pH of fusion, as measured by hemolysis assay^{81, 82}, is indicated for each resistance mutation.

Some resistance mutations have been described that do not seem to directly interfere with either TBHQ or Arbidol binding^{81, 82} (Fig. 6). Instead, these mutations alter the pH of fusion and slightly raise the susceptible pH for triggering the conformational rearrangements that lead to fusion. Thus, increasing the affinity of Arbidol may make it more difficult for the virus to escape. Such inhibitors now provide yet another target for therapeutic development and further options for combating influenza virus.

Conclusions

In the last ten years, immense progress has been made on developing more effective strategies to combat influenza virus either by vaccination or by therapy. The discovery of human bnAbs to influenza virus proved that it was indeed possible to contemplate design of a more universal vaccine. The determination of a large number of bnAb structures against the HA head and stem has enabled common modes of recognition to be deciphered, even if the antibodies differ in their germline origin and angle of approach to the HA surface^{33, 55, 83}. These bnAbs therefore provide a blueprint for designing candidate immunogens that focus the immune response against the HA stem and the RBS. The headless HA designs and their characterization have convincingly demonstrated the feasibility of this approach.

These bnAbs have also inspired design of small proteins and peptides to specifically target the HA and emulate the neutralization properties and mechanisms of the bnAbs. The finding that small molecules can further target an additional conserved hydrophobic pocket in the HA stem provides yet another opportunity for further design of small molecules as therapeutics against influenza virus. The next few years therefore hold great promise both for design of new vaccines and therapies against influenza virus and hopefully these will be realized before any new pandemic emerges on the scale of the devastating H1N1 1918 influenza virus.