Autoreactivity profiles of influenza hemagglutinin broadly neutralizing antibodies

Epitope-focused approaches for selective clonal induction of broadly neutralizing antibodies (bnAbs) inform most current vaccine strategies for influenza virus and other rapidly evolving pathogens. The two conserved epitopes on the influenza hemagglutinin (HA) - the “stem” and the receptor-binding site (RBS) on the “head” - are the focus of the current “universal” influenza vaccine development efforts. Because stem-directed serum bnAbs are much less abundant than head-directed ones, we hypothesized that the HA stem bnAbs may be autoreactive and thus eliminated through the mechanisms of self-tolerance. We compared autoreactivity profiles of a set of stem and head-directed bnAbs. Most of the stem bnAbs we examined bound autoantigens; several showed staining of HEp-2 cells. A smaller proportion of the head-directed bnAbs were polyreactive. Gene usage did not correlate with autoreactivity. We suggest that complex foreign antigens may often have surface patches resembling some host epitope; our results indicate that HA stem epitopes resemble a host epitope more frequently than does the RBS.

www.nature.com/scientificreports www.nature.com/scientificreports/ ready-made, hydrophobic contact surface, including an important contact from HCDR2 residue Phe54 seen in various crystal structures 9,19,20 . A previous study 21 showed that stem-directed antibodies bound more tightly to dsDNA, LPS and insulin than did head binders. V H 1~69 encoded, anti-HA antibodies that did not bind the stem had much lower affinity for these potential autoantigens, suggesting a correlation between properties that give rise to polyreactivity (e.g., hydrophobicity of the CDR surface and positive charge on HCDR3) and adequate access to stem surfaces occluded on virions. Specific epitopes within the stem and head were not determined. Few, if any, of the antibodies in that panel gave nuclear staining of HEp-2 cells, leading the authors to conclude that the extent of polyreactivity exhibited by stem-directed antibodies might not be pathogenic.
These observations led us to pose two sets of questions. First, for stem-directed antibodies, what are the polyreactivity characteristics of those not encoded by V H 1~69? Second, like many non-MPER directed HIV bnAbs, and that of stem-directed CR6261 (red). The model was derived from the crystal structure of full-length HA (gray and light blue) in complex with mAb CH65 (PDB ID 5UGY), onto which were superposed the HA head complex with mAb CH67 (purple; PDB ID 4HKX) and an HA bound with mAb CR6261 (red; PDB ID 3GBM). HA residues in contact with each antibody are shown in their respective color. Fabs were removed for clarity. (b) Representative confocal fluorescence microscopy images of HEp-2 cell staining. Antibody names are indicated for each image. No primary antibody control -no prim and anti-HIV-1 MPER mAb 2F5 were used as controls. All panels are a single plane taken with 20x objective N.A. = 0.7. The scale bar is 50 μm. All HEp-2 cell slides were co-stained with DAPI (blue) to localize the cell nucleus. Channel intensity was adjusted to facilitate visualization of the pattern. Antibodies were grouped and boxed according to their HA epitope -head (purple) and stem (red). (c) Mean Fluorescence Intensity (MFI) quantifying nuclear and cytoplasmic signal for each antibody tested. The values are shown as mean ± SD. www.nature.com/scientificreports www.nature.com/scientificreports/ do RBS-directed influenza HA bnAbs also tend to show polyreactivity? The broader question is, with what probability does any foreign antigen have a B-cell epitope (somewhere on that antigen) similar to an epitope on some host protein.
To examine the polyreactivity of HA-directed antibodies, we chose a 12-antibody panel, six specific for the HA stem and six for the head. Of the latter, five bound the RBS and one a non-RBS head epitope. Ten of the antibodies came directly from donor BCR sequences; one of the stem antibodies was generated through in vitro affinity maturation, and one of the head antibodies came from screening a human-antibody phage display library.
We carried out three more specific assays on subsets of the twelve mAbs.
To assess polyreactivity, the binding patterns of human anti-HA bnAbs were compared to an isotype-matched human myeloma protein 151 K in lot-matched arrays. Essentially, the binding is assessed as a pairwise comparison of signal (anti-HA bnAb) to background (151 K) ratios. As 151 K binds one human protein, betaine-homocysteine methyltransferase 2 (BHMT2) 16 , it is an appropriate negative control for polyreactivity, and serves as a positive control to identify autoreactivity. Previous studies with this assay led us to define the autoreactivity cutoff for HIV bnAbs as a signal/background ratio greater than 500 16 , shown as dashed lines in Fig. 4. Because one of the protein antigens is an influenza A HA, however, we could also define for these tests a more specific cutoff limit. 41-5E04, CR6261 and CR9114 had a highly skewed binding distribution, with some degree of association with most of proteins on the array ( Fig. 4b-d). FI6 interacted with interleukin-15 (IL-15) ( Fig. 4a; Table 1) and, to a lesser extent, with human erythropoietin (EPO) and progesterone-induced-blocking factor 1 (PIBF1). We also plotted the distance of each ProtoArray data point from the diagonal as a frequency histogram and defined a "polyreactivity index" (PI) as the Gaussian mean of the distances from the reference line. The PIs of CR6261, 41-5E04 and CR9114 were 0.34, 0.48 and 0.39, respectively, indicating a significant displacement of the frequency histogram (Fig. S2). Three HA head-directed bnAbs K03.12, 6649 and CR8033 showed no reactivity in the protein arrays (Figs 4h,k,l and S2), but C05, H2526 (which derives from the V H 1~69 gene segment) and CH67 bound most of the proteins on the array (Fig. 4k,l,g); Table 1).
In summary, CR6261 and FI6 stem-directed bnAbs scored positively for autoreactivity in the HEp-2 staining, Luminex-based AtheNA, and ProtoArray assays. CR6261 also bound phospholipids. 41-5E04 anti-HA stem mAb was positive in HEp-2, lipid ELISA and ProtoArray assays. FY1, the precursor of an in vitro affinity-matured antibody (MEDI8852), bound RNP when tested against the AtheNA panel. The heavy chains of these four antibodies derive from V H 1~69, 3~30, 3~53 and 6~1, respectively. Another V H 1~69-derived stem bnAb, CR9114, was polyreactive in the ProtoArray assay. Although none of the six head-directed bnAbs produced HEp-2 nuclear staining, three had polyreactive profiles. C05 had modest autoantigen reactivity in the AtheNA assay, and C05, H2526 and CH67 showed a significant degree of polyreactivity and recognized autoantigens on protein arrays. C05 was isolated by phage display; it is therefore unlikely to have its naturally paired light chain. H2526 has a V H 1~69 derived heavy-chain variable domain. CH67 does not have any evident characteristics predisposing it to polyreactivity. Although previous work suggested that absence of HEp-2 nuclear staining among the stem antibodies in that study diminished the likelihood of pathogenic polyreactivity 21 , nuclear staining by CR6261, 41-5E04 and FI6 (the latter two with heavy chains not derived from V H 1~69) and interaction of FI6 and CR6261 with UBE3A as well as CR6261 binding to Jo-1 and of CR6261 and 41-5E04 binding to phospholipids suggest that at least some stem antibodies have pathogenic signatures.
www.nature.com/scientificreports www.nature.com/scientificreports/ From the autoreactivity profiles of the bnAbs we tested, there appeared to be a correlation between epitope and autoreactivity. Five of the six stem-directed antibodies, CR6261, CR9114, FI6, 41-5E04, and MEDI FY1, were autoreactive to varying degrees. Of these, only CR6261 and CR9114 are encoded by V H 1~69; the others derive www.nature.com/scientificreports www.nature.com/scientificreports/ from the V H 3~30, V H 3~53 and V H 6~1 genes, respectively. MEDI8852 showed no detectable autoreactivity, but as this antibody was evolved in vitro from MEDI FY1 27 , the introduced mutations may have diminished its autoreactivity profile. Of the six HA head-directed bnAbs, three were polyreactive when assayed on protein arrays, although none were positive in the HEp-2 test. A much larger screen would be needed, however, to conclude that head-directed bnAbs are less prone to autoreactivity than are stem-directed ones.
Influenza HA head epitopes have been under a strong selective immune pressure and have thus been evolving rapidly. Unlike the HA head, the stem has remained largely invariant 28 , probably because its occlusion on virions makes it poorly immunogenic 29 . Anti-HA stem specific antibodies are present in human sera -either natural antibodies or derived from memory -although at low level 30 . Upon (re)exposure to influenza HA in a form that does not occlude the stem (e.g. a split vaccine), one might expect these responses to be rapidly amplified. Yet, in homologous vaccination boost regimens, the immune responses upon re-vaccination are disproportionately skewed towards the globular head epitopes -the head responses increase after re-vaccination, while the stem  www.nature.com/scientificreports www.nature.com/scientificreports/ responses remain the same pre-and post-vaccination 30 . These data are consistent with purging of stem responses through mechanisms of self-tolerance, but a firm conclusion would require a direct test of purging in vivo.
Conservation of stem epitopes over decades of influenza virus variation is consistent with the low abundance of stem-directed antibodies and the consequent absence of immune pressure 29 . Imprinting from virus encountered in childhood, which biases subsequent responses to vaccination or infection with a later strain 31 , will then further contribute to the immunodominance of the head. The stem of HA in split vaccines, although largely clustered into rosettes, may be more accessible than on virions, perhaps accounting for the transient appearance of stem-directed antibodies in early recipients of the 2009 vaccine. Suppression by tolerance mechanisms could then explain their failure to persist following subsequent exposure to the same vaccine. www.nature.com/scientificreports www.nature.com/scientificreports/ Current research on vaccine strategies for rapidly evolving viral pathogens like HIV and influenza emphasizes epitope-focused, selective clonal induction 32 . For example, recent discussion of a "universal influenza vaccine" has concentrated particularly on HA stem epitopes [33][34][35][36] . The data presented here indicate that strategies focused exclusively on the HA stem-or, indeed, on any single epitope-might fail to induce adequate antibody titers owing to negative selection of autoreactive B cell clones. Including multiple conserved epitopes in an epitope-focused vaccine will probably increase the likelihood of a robust response while also decreasing the likelihood that a single mutation will escape an individual's protective immunity.

Materials and Methods
IgG expression and purification. For IgG production the genes for the heavy-and light-chain (kappa or lambda) variable domains were synthesized by Integrated DNA Technologies and cloned into pVRC protein expression vectors containing mouse and human heavy-and light-chain constant domains. Briefly, the bnAb V(D)J sequences were cloned into vectors containing mouse or human IgG1 framework using conventional endonuclease restriction digestion, as previously reported 37 . The mouse isotype was used for HEp-2, lipid ELISA and Luminex screening and the human isotype for ProtoArray microchips. All subsequent secondary reagents were species-specific to limit background signal. Constructs were confirmed by sequencing at the DNA Sequencing Core Facility at Dana Farber Cancer Institute. IgGs were produced by transient transfection of suspension HEK293F cells using polyethylenimine (PEI; Polysciences). Supernatants were harvested 7 days later, clarified by centrifugation and IgGs were purified using Protein G Plus Agarose (Thermo Fisher Scientific): the IgG supernatants were incubated overnight with agarose slurry, eluted with 0.1 M glycine, pH 2.5, neutralized with 1 M Tris-HCl, pH 8.0 and dialyzed against PBS buffer.
HEp-2 cell staining. Ready to use HEp-20-10 slides (Euroimmun; product No. FA 1512-20) were stained according to manufacturer's instructions with minor modifications. Recombinant mAbs (mouse IgG1) were diluted to 5 µg/ml or 25 µg/ml in block buffer (0.1% tween, 0.5% BSA in PBS). After one hour of incubation with the mAbs, slides were washed in block buffer and stained with goat anti-mouse IgG-Alexa488 (Life Technologies) and 2 µg/ml DAPI (Sigma) for one hour. After washing, coverslips were mounted with Fluoro-gel (Electron Microscopy Sciences). Images were acquired using an Olympus Fluoview FV1000 confocal system.
Quantification. Images were analyzed using Cell Profiler 38 . Nuclei were identified by the DAPI signal.
Cytoplasmic masks were generated by expanding the nuclei mask by 10 pixels and subtracting the original nucleus mask. Mean fluorescent intensities in the Alexa488 channel were quantified for each cell (defined by nucleus).
ProtoArrays. Recombinant human IgG1 mAbs were screened for binding on protein microarrays (ProtoArray) (catalog no. PAH0525101; Invitrogen) precoated with >9,400 human proteins in duplicate. The binding patterns of human anti-HA bnAbs were compared to the human myeloma protein 151 K in lot-matched arrays. Array-bound anti-human IgG served as the loading control for the detection Ab, and array-bound human IgG served as the loading control for the secondary reagent. Abs were screened for reactive antigens on protein microarrays following the manufacturer's instructions and as described previously 16 . The ProtoArray microarray (Invitrogen) was blocked and incubated on ice with 2 μg/ml of HA mAb or isotype control 151 K for 90 min. Ab binding to array protein was detected with 1 μg/ml of Alexa Fluor 647-labeled anti-human IgG secondary Ab (Invitrogen). The ProtoArray microarrays were scanned using a GenePix 4000B scanner (Molecular Devices) at 635 nm, with 10-μm resolution. Fluorescence intensities were quantified with GenePix Pro 5.0 program (Molecular Devices) using lot-specific protein location information provided by the microarray manufacturer.