Combinatorial antibody libraries from survivors of the Turkish H5N1 avian influenza outbreak reveal virus neutralization strategies

The Outbreak and Source of Material.

Between December 2005 and January 2006, an outbreak of avian influenza H5N1 occurred in Turkey (9). In total, 12 individuals were infected and only 8 survived. Because bone marrow RNA contains the archived record of all antibodies made by an individual, we selected it as our source material. We obtained bone marrow and serum from six of the Turkish survivors after their recovery and successfully prepared antibody libraries from five of the six bone marrow samples. In the sixth sample, the RNA was degraded.

Serological Analysis.

The hemagglutinin (HA) protein is essential for binding the influenza virus to the cell that is being infected and is generally considered to be the main target of neutralizing antibodies (1). Therefore, we tested by ELISA each of the individual serum samples at high serum dilutions to detect antibodies against H5 HA proteins [see supporting information (SI) Fig. S1] and intact viruses (data not shown). This analysis showed that the patients had readily detectable serum antibodies, even when the serum was diluted 10,000-fold. We selected the Vietnam/1203/04 HA as a target because it was readily available and is thought to be related to the influenza virus strain that caused the disease outbreak in Turkey.

Library Construction.

Our primary objectives were to understand the nature of the immunological response to infection and to identify specific antibodies that might be used passively for the prevention of spread and/or for treatment of H5N1 influenza virus infections. We wished to recover every possible solution to H5N1 infections, with minimal or no bias. Because gene expression for the individual Ig families is not equal—making them prone to bias and over-representation—we decided against using the standard pooled approach to Ig recovery. Instead, we individually rescued 20 of the 23 distinctly amplifiable gene families during construction of the libraries. The remaining three gene families (V_H 2, 5, and 6) were recovered as a pool because they are infrequently used. We further normalized gene content by creating equimolar pools of each Ig family DNA for cloning into a phagemid display vector.

A unique DNA barcode was embedded into a nondisruptive portion of the phagemid vector to allow each clone to be tracked back to the original patient source (Fig. S2). This barcoding enables assignment of clones to individual patients, even when phage libraries from multiple survivors are screened simultaneously. As a result of this tagging, every clone isolated from any library can be confidently attributed to the cognate survivor.

Using this vector with its coding system, we successfully cloned repertoires from the bone marrow of five of the six survivors in both single-chain (scFv) and Fab phagemid formats. Each collection from an individual survivor has a diversity >1.0 × 10⁸ members. Furthermore, we created additional barcoded libraries comprising mixed survivor light and heavy chains with a final diversity of 1.1 × 10⁹. Collectively, the five donor-specific collections and the pooled libraries from all donors have a total diversity of 1.0 × 10⁹ as an scFv collection and 4.2 × 10⁹ as a Fab-displayed collection (Table S1).

Selecting Binding Antibodies.

As indicated above, one interesting feature of these studies was that we initially selected antibodies for binding against a virus strain and antigen that were related to, and different from, the one that caused the infection. This was done because viral isolates from the patients were not available. The necessity of using a related strain to select antibodies could have proven fortuitous because it may have led to the isolation of more broadly neutralizing antibodies (see below).

The libraries were panned against inactivated virus containing the Vietnam/1203/04 virus HA and neuraminidase proteins and recombinant purified HA (10). Typically, after three to four rounds of phage panning, individual clones from enriched phage pools were analyzed by ELISA against H5N1 virus or purified HA, and the positive clones were sequenced to determine their heavy and light chain sequences and to read their survivor barcode (11). From these studies, we isolated specific H5 HA binding clones from all five of the individual libraries from survivors. We have thus far recovered >300 hundred different antiviral antibodies, of which 146 specifically bind the H5 HA protein.

General Features of the Selected Clones.

Overall, the individual patients use different germ lines for both heavy and light chains, demonstrating that these individuals have found different solutions to the same potentially lethal immunological challenge. The major features of combinatorial antibody libraries that can be used both to give confidence as to the quality of the obtained repertoire and to provide information about the chemistry of antibody binding and/or neutralization are seen in these clones. The clones contain all of the hallmarks of the previously described repeated clones (“jackpot solution”) to antigen binding that is found in the natural progression of affinity maturation, as well as in selected synthetic antibody libraries (8, 12). The presence of jackpots in these large collections validates the screening procedure because, unless the phage was selected on the basis of activity, the chance of obtaining the same clone multiple times is highly improbable. Moreover, when one analyzes the heavy chain differences within groups, it is observed that many of the amino acid substitutions were chemically and structurally conservative (Table 1). As with repeated clones, the appearance of multiple amino acid substitutions that are chemically reasonable is unlikely to be a random event.

Binding Specificity of Recovered Antibodies.

Initial testing of a set of Fabs by using biolayer interferometry binding to the H5 Vietnam HA protein indicated that we had identified at least four distinct epitopes (data not shown). We selected six clones, from three survivors, that recognized two different epitopes for conversion into full IgG1 proteins. The binding of three of these antibodies was mapped to the HA1 subunit of the HA protein by Western blot analysis (data not shown).

One goal of these studies was to recover those rare antibodies that broadly neutralize divergent viral strains. It was suggested that some of our antibodies might be broadly reactive because the serum from the donors had high-titer antibodies against a divergent subfamily of H5N1 viruses that extended beyond the virus with which they were infected. To determine the degree of cross-reactivity at the level of individual antibodies, we analyzed binding of our clones to various influenza HA antigens (Fig. 1 and Table 2). Not surprisingly, these antibodies recognize HA from the corresponding infecting Turkey/65596/06 strain and, in addition, recognize the heterologous HA from the Vietnam/1203/04 strain used for selection. Furthermore, they recognize the antigenically divergent Indonesian/5/05 H5 HA. We performed kinetic binding analyses on the four prototype antibodies and found that the antibodies from survivor 5 bound Vietnam/1203/04 HA with single-digit nanomolar affinities, whereas the survivor 2 antibody bound more strongly, with a measured affinity of 13 pM (data not shown).

To determine whether our antibodies are even more broadly reactive, we studied their binding to a larger collection of HAs from different influenza A subtypes (Fig. 1 and Table 2). We found that the four prototype antibodies bound HA from the closely related subtype H1N1 contemporary reference strain New Caledonia/20/99. Notably, the three neutralizing antibodies belonging to survivor 5 also bound HA from the H1N1 South Carolina/1/18 isolate that emerged during the 1918 Spanish flu pandemic. Conversely, none of these four antibodies bound HA from the contemporary H3N2 Wisconsin/67/05 reference strain, indicating that even though the antibodies display broad-spectrum binding among and between influenza subtypes, the reactivity did not extend to all influenza subtypes.

To further explore the immunochemical basis of the H1/H5 cross-reactivity, we rescreened the libraries against the H1N1 New Caledonia/20/99 HA protein. From this selection, we found clones (Table S2) that bore significant similarity to the sequences obtained from survivor 5 when the H5 HA protein was used in the panning (Table 1).

Neutralization Studies.

Initially, the antibodies were assayed for their ability to neutralize an H5 HA (Vietnam/1203/04) containing influenza virus. One antibody derived from survivor 2 and three from survivor 5 that recognized a common epitope (epitope “A”) were all neutralizing, whereas the two antibodies derived from survivor 1 that recognized a second epitope (epitope “B”) were not.

Based on the striking sequence similarity of clones separately isolated from survivor 5 against either H5N1 or H1N1 HA, we predicted that their cross-reactivity would extend beyond simple binding and they would also have the highly unusual property of neutralizing both H5N1 and H1N1 virus. To examine the cross-neutralizing activity of the IgGs, we tested representative antibodies from the H5N1 screen in a neutralization assay to see whether they would also neutralize H1N1 or H3N2 virus (Table 3). We studied the H1-bearing virus A/New Caledonia/20/99 and the H3-bearing virus A/Hong Kong/68. A collection of viruses bearing H5 subtype HA (A/Vietnam/1203/04; A/Indonesia/5/05; A/Turkey/65596/06; A/Egypt/06) was also tested. The antibodies showed no activity against H3 subtype influenza; however, three of the monoclonal antibodies that neutralized H5-containing viruses also strongly neutralized A/New Caledonia/20/99, which bears an H1 subtype HA (Table 3).

Immunochemical Basis of Neutralization.

One advantage of antibody libraries is that when large numbers of antibodies are obtained, they can be grouped as to their relatedness. Thus, when a function for a given antibody in the collection is observed, one can predict that other members of the group to which the antibody belongs will have similar activity.

All members of the group that contained the neutralizing antibody collection against epitope A from survivor 5 analyzed to date are shown in Table 1. The group consists of 61 unique members that most closely resemble the V_H1e germ-line heavy chain. Some heavy chains are paired with more than one light chain. In total, these heavy chains have 115 unique pairings to both κ and λ light chains. In comparing these heavy chains to the highly related V_H1e germ line, we observe three types of point substitutions. Some changes appear to be required, others are dominant, and some residues have only been changed sporadically. The required changes occur in every clone in the group within CDR2 at positions 52A (Pro → Gly), 53 (Ile → Met), and 57 (Ala → Thr), as well as in the framework 3 region at positions 73 (Lys → Glu) and 74 (Ser → Leu or Met), all of which vary from the germ-line side chain chemistries, suggesting that these mutations are critical to antigen binding and neutralization. The second set of mutations are dominant and are found in most clones. The first mutation, in framework 1 at position 24 (Ala → Thr), represents a significant chemical change. The next three are conservative changes in CDR1 at positions 34 (Ile → Val) and 35 (Ser → Thr) and also in CDR2 at position 50 (Gly → Ala). All four of these dominant substitutions, however, are dispensable, suggesting that, although beneficial, they are not essential. The sporadic changes found throughout framework regions 1, 3, and 4, as well as CDR3, are all conservative and likely represent minor optimization events. The positions of the required mutations in the structure of the antibody are shown in Fig. 2 superimposed on the crystal structure of a highly related anti-HIV Fab called 47e (Protein Data Bank ID code 1RZI) (13). The required mutations 52A (Pro → Gly), 53 (Ile → Met), 73 (Lys → Glu), and 74 (Ser → Leu or Met) create a remarkably tight cluster on the exposed surface of the heavy chain variable domain, where they form a ridge that protrudes prominently from the protein surface (Fig. 2). The remaining required mutation, 57 (Ala → Thr), is partially buried at the base of the CDR2 loop. The surface-exposed changes in CDR 2 and framework 3 are likely to have a direct role in antigen binding, whereas the less exposed required mutation and the nonessential dominant mutations may have indirect effects through stabilizing and/or positioning of the CDR2 loop.

The probability that a given mutation is important to the activity of an antibody increases as a function of the number of times the mutation was independently selected. To determine whether the required mutations were selected during somatic mutation from independent clones or were from the progeny of a single clone that mutated further during subsequent replications, the codon usage of the dominant mutations was analyzed (Dataset S1). The data reveal that, although different codons were used, they resulted in the same amino acid changes, demonstrating that these mutations arose independently in different clones and thus were selected multiple times. This convergent outcome for independently selected events is strong evidence that these dominant mutations play a critical role in the binding to the virus and/or in its neutralization.

The antibodies from survivor 2 consist of two unique heavy chains that most closely resemble the V_H4–4b germ-line heavy chain (Table 4). The first heavy chain has been found paired with five unique λ light chains, four of which are from the infrequently used λ 6 light chain family, and the other is paired with a single κ light chain. Antibody 4, whose neutralization profile was more restricted, came from this group.