A stable trimeric influenza hemagglutinin stem as a broadly protective immunogen

Starting with the HA sequence of H1N1 A/Brisbane/59/2007, changes were introduced using a combination of rational design and library approaches (27). In the first design phase (stages I to III; Fig. 1A), we aimed to generate a soluble and stable HA stem monomer containing bnAb epitopes in the correct conformation, whereas in the final stages (stages IV and V), we aimed to create a mini-HA in its natural trimeric form. Details on the design, sequences, and screening are outlined in (27) and figs. S1 and S2. As a readout for a native-like stem conformation of the mini-HA constructs, we used the binding of bnAbs CR9114 and CR6261 (both of which recognize epitopes in the HA stem) (28). From each design stage, the best-performing candidates were selected for further optimization. Key mutations in these lead candidates are schematically depicted in fig. S2 and on a model structure of mini-HA in Fig. 1B.

Selected mini-HAs #4157 (stage II), #4454 (stage III), #4650 (stage IV), and #4900 (stage V) were expressed in human embryonic kidney (HEK) 293F cells and purified. On the basis of size, mini-HAs #4157 and #4454 appeared predominantly monomeric in solution; #4650 was larger, suggesting at least partial formation of dimers, whereas #4900 appeared trimeric (fig. S3A). Formation of an intermonomer cysteine bridge in trimeric mini-HA #4900 was confirmed using SDS–polyacrylamide gel electrophoresis under nonreducing and reducing conditions (fig. S3B). All selected mini-HA constructs have four putative N-linked glycosylation sites (Fig. 1B). Under reducing conditions, the mini-HA bands were diffuse and ran at higher molecular weight (35 to 45 kD) than expected (29 kD), indicating that the mini-HAs were glycosylated (fig. S3B). Accordingly, deglycosylation with peptide-N-glycosidase F resulted in sharper bands (~30 kD, as for mini-HA #4900 in fig. S3B). The glycans appear relevant for proper folding, as attempts to efficiently produce mini-HA with retention of bnAb epitopes in bacteria were not successful (29).

Size exclusion chromatography with inline multi-angle light scattering (SEC-MALS) analysis revealed a single peak (fig. S4A), with average molecular weights of 40 kD for #4157 and #4454, 60 kD for #4650, and 108 kD for #4900 (Table 1), which also indicated that #4157 and #4454 are monomeric in solution, #4650 forms dimers (at least in part), and #4900 forms trimers. In the presence of CR9114 or CR6261 Fabs, a peak shift to shorter retention time was observed for all mini-HAs, indicating formation of complexes (fig. S4A), which was confirmed by MALS (Table 1), with masses in agreement with binding of one Fab (monomeric mini-HAs #4157 and #4454), one to two Fabs (dimeric mini-HA #4650), or three Fabs (trimeric mini-HA #4900) (30).

Binding of CR9114 and CR6261 to all four candidate immunogens was in the nanomolar range, with affinity improving as the design process progressed (Table 1 and fig. S4B). The strongest binding was exhibited by mini-HA #4900, with dissociation constant (K_D) values comparable to those of trimeric FL HA, indicating that this mini-HA recapitulates the bnAb epitopes and thus accurately mimics the native conformation of the HA stem.

Modifications introduced throughout the design gradually increased the conformational stability as determined by hydrogen-deuterium exchange mass spectrometry (HDX-MS) (31, 32). In particular, modifications introduced in stage III (from #4157 to #4454) and stage V (from #4650 to #4900) stabilized the overall structure and the shorter A helix (Fig. 2). Furthermore, we observed only a low level of deuterium exchange on the long CD helix of mini-HA #4900, likely due to the shielding effect caused by trimerization. Most important, the HDX pattern of mini-HA #4900 is very similar to that of trimeric FL HA, further demonstrating that the final design mimics the structure and dynamics of native HA. In line with these findings, the overall stability of the mini-HAs also evolved, as is apparent from measurements of the thermal stability (fig. S5).

Complexes of mini-HAs with CR9114 Fab were first studied by negative-stain electron microscopy (EM) (Fig. 3A). The EM class averages illustrate the progression from monomeric (#4454) to a more heterogeneous mixture of monomers and dimers (#4650) and finally to a trimeric form (#4900). To provide further structural insights, we determined the crystal structures of #4454-CR9114 and #4900-CR9114 Fab complexes to 4.3 Å and 3.6 Å resolution, respectively (Fig. 3, B and C, and table S1) (27). In the #4454-CR9114 complex (Fig. 3B), CR9114 binds monomeric mini-HA and recognizes essentially the same epitope as in FL HA through contacts only with its heavy chain (table S2) (11). The solvent-accessible surfaces of the CR9114 epitope residues in #4900 are similar to those in A/California/04/2009 FL H1 HA (table S3). The #4900 trimer differs only slightly from the stem of trimeric A/California/04/2009 FL HA (and perhaps other HAs), as it adopts a more open, splayed conformation at the trimer base with the HA2 H helix, shifted by about 14.5 Å (Fig. 3, C and D), perhaps as a consequence of insertion of the GCN4 trimerization motif at the top of the long helix in #4900 (fig. S6). The primary amino acid interactions are maintained between CR9114 and #4454 as compared to those between CR9114 and FL H5 HA (3), and the CR9114 complex is trimeric (Fig. 3C). The #4900 protomer is superimposable with an HA protomer from the H1 A/California/04/2009 trimer (PDB ID 4M4Y) with a C_α RMSD of 1.2 Å, and only 1.0 Å for the CR9114 epitope residues. As a consequence of the removal of the HA1 head in #4900 and a more open trimer base (table S2), some residues in the HA2 A helix, B loop, C and D helix, and E strand become solvent-accessible. These results convincingly demonstrate that the iterative design process led to antigens with an increasingly improved HA stem conformation and ultimately yielded HA-like stem antigens with a preserved CR9114 epitope.

When tested in mice, all mini-HAs were highly immunogenic, eliciting high titers of antibodies binding to the FL HA of the A/Brisbane/59/2007 (H1N1) strain used as the basis for mini-HA generation (fig. S7A), as well as to FL HAs from a number of other group 1 (H1, H5, H9) and group 2 (H3, H7) influenza A strains (fig. S7B). These results demonstrate that antibodies induced with mini-HAs efficiently recognize epitopes that are present in native sequences of FL HA and are conserved among different group 1 and even some group 2 influenza A strains (33).

Next, we evaluated the ability of mini-HA immunogens to provide protection against stringent lethal challenges and characterized the elicited immune response (Fig. 4 and fig. S7). Protection against influenza challenge with H1N1 A/PR/8/34 (heterologous to the parental strain used for design of the mini-HAs) was assessed after one, two, or three immunizations with mini-HAs #2759 (monomer, long variant of #4157; fig. S2), #4650 (mix of monomers and dimers), and #4900 (trimer). Mice immunized three times with monomeric mini-HA #2759 were only partially protected against mortality and showed severe weight loss and clinical symptoms, whereas #4650 and #4900 provided complete protection, with #4900-immunized mice showing neither weight loss nor clinical symptoms (Fig. 4A and fig. S7C). Only mini-HA #4900 showed the same full protective ability when administered twice. Even after one immunization, #4900 exhibited 90% protection against mortality, although animals exhibited some (recoverable) weight loss and clinical symptoms, whereas #4650 and #2759 were not able to protect animals. Antibodies elicited with mini-HA #4900 after two and three immunizations competed with bnAb CR9114 for binding to the FL HA stem (Fig. 4A), which shows that this mini-HA could readily elicit antibodies specific for bnAb stem epitopes.

Mice immunized three times with an expanded set of mini-HAs were challenged with heterosubtypic H5N1 A/Hong Kong/156/97 virus to evaluate the breadth of protection. Mini-HAs #2759 and #4157 (monomers) failed to elicit protection, #4454 (improved library monomer) and #4650 exhibited partial protection, whereas #4900 exhibited complete protection with neither weight loss nor clinical symptoms (Fig. 4B and fig. S7D). To further investigate antibody-mediated effector mechanisms, we tested prechallenge sera in a neutralization assay using pseudoparticles derived from H5N1 A/Vietnam/1194/04 (34). Sera from mice immunized with mini-HA #4900, and to some extent #4650, showed detectable neutralization, demonstrating the ability of these mini-HAs to elicit bnAbs (Fig. 4C). Besides direct virus neutralization, Fc-mediated effector mechanisms, such as antibody-dependent cellular cytotoxicity (ADCC), contribute substantially to protection against influenza, with stem-directed bnAbs being particularly effective in these mechanisms (35). We therefore used a mouse-adapted ADCC surrogate assay to test prechallenge sera (36–38). The evolving mini-HAs demonstrated an increasing ability to elicit antibodies with potential to induce ADCC (Fig. 4D), with ADCC potency of serum antibodies corresponding to survival proportions observed in the H5N1 challenge model. In accordance with the binding data (fig. S7, A and B), mini-HA #4900 elicited ADCC responses to a wide range of group 1 influenza A HAs (fig. S7E). To assess whether serum antibodies are indeed responsible for the observed in vivo protection, we passively transferred serum from mice immunized with #4650 and #4900 to naïve mice and subsequently challenged the recipient mice with A/HK/156/97 (H5N1) virus (Fig. 4E and fig. S7F). In contrast to mice that received mock serum, mice that received serum from mini-HA–immunized mice were significantly protected, proving that antibodies play a key role in the protection elicited by mini-HAs.

These combined results demonstrate that the progressive structural refinement and stabilization of mini-HAs directly translates to increasing ability to elicit stem-targeting, neutralizing, and ADCC-mediating antibodies that can protect mice against heterologous and heterosubtypic group 1 influenza strains, with trimeric #4900 exhibiting the most favorable features. In a recent study (26), a bacterially produced trimeric HA stem construct also seemed more effective than monomeric versions in eliciting protection in an animal influenza challenge model.

We next tested the immunogenicity and protective efficacy of a three-dose regimen of mini-HA #4900 in six cynomolgus monkeys (Macaca fascicularis). As controls, groups of six monkeys received three injections of phosphate-buffered saline (PBS) or two human doses of a trivalent seasonal influenza vaccine (the standard of care for naïve children). As previously observed in mice, #4900 elicited high titers of antibodies in monkeys that were able to bind to a wide range of group 1 FL HAs (Fig. 5A and fig. S8A); compete with stem-binding bnAb CR9114 for binding to homologous, heterologous, and heterosubtypic HAs (Fig. 5B); and elicit ADCC responses to these divergent HAs (Fig. 5C). Furthermore, #4900 elicited H5N1 neutralizing antibodies that were detectable not only in a pseudoparticle-based neutralization assay (fig. S8B), as for the mouse sera (Fig. 4C), but also in a standard microneutralization assay (Fig. 5D). Relative to mock-immunized monkeys, animals vaccinated with mini-HA #4900 had significantly reduced fever after challenge with A/Mexico/InDRE4487/2009 (H1N1) virus (Fig. 5E and fig. S8E) (39), although no significant effect on the tracheal viral load was detected (fig. S8C). These results are comparable to those from monkeys vaccinated with the trivalent seasonal influenza vaccine (Fig. 5E and fig. S8C). However, the immune response elicited by mini-HA #4900 differed from the response elicited with the trivalent vaccine. As expected, vaccination with the seasonal vaccine, but not with “headless” #4900, elicited hemagglutination-inhibiting (HI) antibodies against A/California/07/09 (H1N1) virus, from which the H1 component of the trivalent vaccine is derived and which is closely related (two amino acid differences) to the A/Mexico/InDRE4487/2009 (H1N1) virus used to challenge the monkeys (fig. S8D). Furthermore, because the A/Texas/50/2012 HA antigen was present in the trivalent seasonal vaccine, the vaccine elicited higher antibody titers against this H3 strain than did mini-HA #4900 (fig. S8A). However, unlike the mini-HA–immunized monkeys, no antibodies able to compete with CR9114, induce ADCC, or neutralize H5N1 virus were detectable in the sera of monkeys immunized with the seasonal vaccine (Fig. 5, B to E, and fig. S8B); these results show that the mini-HA elicited a qualitatively different and much broader immune response than the seasonal vaccine.

In these in vivo studies, we used a single antigen dose and a potent adjuvant. Although this adjuvant has not yet been registered for human use, it has been tested in a number of clinical trials with favorable safety and immunogenicity outcomes (40). Future studies will need to address the minimal doses of mini-HA and adjuvant for a protective response. Because preexisting immunity may have a profound effect on the breadth of the response and no animal model recapitulates the complex preexisting immunity against influenza found in humans, such studies should be performed in humans (41).

We have described the design and characterization of a series of soluble HA immunogens solely composed of the HA stem. Although all selected mini-HAs elicited comparable levels of antibodies to FL HA, the breadth and protective ability of the elicited antibodies progressively increased with the structural evolution of mini-HA configuration. The final candidate—stabilized trimeric mini-HA #4900—demonstrated its unique ability to elicit broad and protective immune response in mice and nonhuman primates. It has been reported (42, 43) that stabilization of respiratory syncytial virus F antigen improves immune response and protection. Our results demonstrate that the same principle holds for influenza HA and provide further direction for the design of an epitope-based, universal influenza vaccine.

We thank R. Vogels, R. van der Vlugt, D. Zuijdgeest, N. Kroos, V. Klaren, S. Schmit-Tillemans, L. Kil, S. Barens, and O. Diefenbach for scientific input and technical support; J. Klap, M. Koldijk, and G. J. Weverling for statistical support; Algonomics for input in early designs; Novavax for supplying Matrix-M; Promega for early access to mouse surrogate ADCC assay; E. Montomoli, D. Perini, and S. Piccirella from VisMederi for pseudoparticle assay; Janssen Diagnostics for quantitative polymerase chain reaction of tracheal swabs; P. Mooij and G. Koopman of the Biomedical Primate Research Center (BPRC), TNO Triskelion, Central Veterinary Institute, and PreClinBiosystems AG for performing the animal studies; and L. Dolfing and A. Dingemans for critical reading of the manuscript. The data presented in this manuscript are tabulated in the main paper and in the supplementary materials. X-ray diffraction data sets were collected at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 12-2. Use of the SSRL, SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences under contract no. DE-AC02-76SF00515. The SSRL Structural Molecular Biology Program is supported by the DOE Office of Biological and Environmental Research and by National Institute of General Medical Sciences (NIGMS) grants including P41GM103393. The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of NIGMS or NIH. Coordinates and structure factors of the crystal structures are deposited in the Protein Data Bank as entries 5CJQ and 5CJS. Crucell Holland B.V., a Janssen company, has the following pending patent applications in this field: WO 2013/079473, WO 2014/191435, WO2008/028946, WO2013/007770, U.S. 62/062,746, and U.S. 62/062,754. Sharing of materials will be subject to standard material transfer agreements.