GFPDL are a powerful tool to decipher all the primary antigenic sites in influenza viruses following natural exposure or vaccination. Previously, this approach led to the development of HIV-SELECTEST for differential diagnosis of HIV infections in vaccine recipients 
. In the current study, we used H5N1 GFPDL to identify recognition sites of antibodies in convalescent sera obtained from five Vietnamese individuals with a history of H5N1 infection and two H5-specific neutralizing MAbs derived from two of these survivors. The study identified antibodies targeting the HA, NA, and the M2e among other proteins. Furthermore we provide evidence that the PB1-F2, a putative viral virulence factor, was recognized by antibodies in all five convalescent sera of H5N1-infected individuals (mean end-point titers of ≥1
Previously, murine MAbs and escape mutants were used to map binding sites on the structures of human influenza HA 
, and more recently, of H5 HA 
. In contrast, until recently, very limited information was available on human antibody epitopes in type A viruses, and even less for avian H5N1 viruses 
Elucidating the repertoire of influenza-specific human antibodies against all the viral proteins is desired for understanding virus-host interactions and identifying new targets for protection. To that end, we used influenza GFPDL to unravel the antibody specificities of five H5N1-recovered individuals and to map the epitopes of two neutralizing MAbs derived from their memory B cells. Our main findings were: (1) H5N1 convalescent sera contained highly diverse antibody specificities against HA1, HA2, NA, and internal proteins of the virus for at least 6 mo; (2) two human MAbs, which were protective in mice, recognized conformation-dependent nonlinear epitopes in large HA1 fragments that encompass the RBS; (3) the HA1[(-10)-223)] protein showed high avidity binding to the two human MAbs and adsorbed a significant proportion of the neutralizing activity of polyclonal anti-H5N1 sheep and ferret sera; (4) strong antibody reactivity against the NA catalytic site and M2e were identified; (5) convalescent sera bound PB1-F2 peptides providing evidence that this protein is expressed during acute H5N1 infection; (6) sera from 20 Vietnamese adults with no history of H5N1 infection, and from ten US human influenza-confirmed infections, revealed very low reactivity to most of the H5N1 epitopes.
Deciphering the epitopes of the H5-specific human MAbs explained the restricted neutralization pattern of FLA5.10 compared with FLD21.140. MAb FLA5.10 binding requires L129, a critical amino acid within the RBS. A L129
S change was reported in clade 2 H5N1 viruses from China and Southeast Asia in 2002–2005 
, and could explain the clade 1-restricted protection provided by FLA5.10. Preliminary data from escape mutants (ALS and KS) support the FLA5.10 epitope identified by phage display libraries (unpublished data). On the other hand, the predicted contact residues for FLD21.140 are highly conserved among clade 1 and clade 2 H5N1 viruses and the high binding affinity of this MAb for the HA [(-10)-223)] fragment may explain its broad cross-protection in vivo. Importantly, for both MAbs the combined GFPDL/RPL approach identified noncontinuous conformation dependent epitopes.
Panning of GFPDL (HA+NA) with sera from individuals who had recovered from H5N1 virus infection revealed broad antibody reactivity against both HA1 and HA2 domains, including the large HA1 fragments bound by the protective human MAbs. Epitope profiling of HA led to identification of six antigenic clusters (I–VI). HA antigenic sites “a–e” were defined primarily using mouse monoclonal antibodies that are encompassed in clusters-I and -II, described in this study.
A recent paper by Throsby et al., describes heterosubtypic neutralizing MAbs that cross react against H5N1 and H1N1 
. The epitopes of these monoclonal antibodies were mapped to amino acids 43–58 in HA2. This sequence corresponds to H5-HA-2723-2378 in our study, and was part of several peptide sequences displayed on the affinity selected phage clones (, peptides 10–12; Table S1
). While all five H5N1 recovered individuals had titers against these HA2 peptides (reciprocal titers 500–2,500), binding of control plasma (from Vietnam or the US) to these peptides was observed at low frequency, and the titers did not exceed 1
100 (; Table S2
Among the H5N1 convalescent plasma, most ELISA reciprocal titers were lower in the 6-mo postinfection sample. However, HA2 peptide (H5-HA-2838-2866) located in the membrane proximal domain was strongly recognized by all convalescent sera (), but did not bind control sera from Vietnam or from US individuals with recent culture-confirmed seasonal influenza infections. This highly conserved HA2 region contains aa changes between H5, H1, and H3 viruses 
and could be useful for surveillance of H5N1 infections in endemic areas.
This study also identified a strong binding by the H5N1 convalescent sera to a 178-aa fragment containing the NA catalytic site that has not been reported previously. This NA segment was not recognized by US control sera with high HI titers against H1N1. Therefore, we did not find evidence that repeated exposure to human H1N1 influenza viruses elicits high-titer antibodies against the heterologous avian N1 NA, as was previously predicted 
Studies in mice have shown that serum anti-M2e antibodies can reduce virus replication and death. Currently, M2 is being evaluated as a component of several “universal vaccines”
, and possibly as target for monoclonal antibody therapies 
. However, the immunogenicity of M2e during natural influenza infections is still debated 
. Surprisingly, the GFPDL analyses revealed high titer anti-M2e antibodies in post-H5N1 infection sera (but not in control sera). Although the N-terminal (1–9 aa) sequence of M2e is highly conserved across different influenza-A subtypes, there is significant diversity in the C-terminal (10–21 aa) sequence of M2e (http://www.flu.lanl.gov
). Also, similar GFPDL analyses using the pooled sera from 20 control H5N1-uninfected Vietnamese females did not select any phage-displaying peptide sequence from the M2 region. This result suggests that strong antibody response is generated against M2e following primary infection with HP H5N1 strain. Thus, the contribution of M2e antibodies to viral clearance after H5N1 infection should be further evaluated.
PB1-F2, a recently discovered proapoptotic influenza-A viral protein, contributes to viral pathogenesis in mice 
. However, this protein is not incorporated into viral particles, and evidence for its expression during influenza infection of animals or humans is lacking. In our study, 35 PB1-F2–expressing GFPDL clones were isolated, and sera from all five H5N1 infection survivors, but not control sera, reacted with synthetic PB1-F2 peptides in ELISA. To our knowledge, this is the first report indicating that PB1-F2 is expressed during H5N1 virus infection in humans. It is not clear how this short-lived protein activates B cells. However, high viral replication and cell lysis during H5N1 infection could present the newly expressed viral proteins to the immune system.
The use of GFPDL to decipher the complete B cell repertoires in AIV-exposed individuals is limited by the fact that all protein segments are expressed in a bacterial system. Therefore, epitopes that are strictly dependent on post-translational modifications or on the multimeric forms of influenza proteins in the native structure might have been missed in our analyses. In spite of these limitations, the use of GFPDL led to new insights into the repertoire of anti-influenza antibodies following H5N1 virus infection and of epitopes that may contribute to resolution of avian influenza infections and could be incorporated into future vaccines. Finally, conserved epitopes recognized by sera from convalescent individuals may be useful for monitoring outbreaks of avian influenza.