assessed the approximate number of eluted viruses). We screened one-third of the library (that is, 2.1 X 106viruses) in three separate selection experiments (that is, 0.7 X 106 viruses per experiment) and isolated 370 viruses that bound to Siaα2,6-TRBCs (Supplementary Fig. 2). Individual viruses were then grown in Madin-Darby canine kidney (MDCK) cells modified to overexpress Siaα2,6Gal (AX4 cells17), and screened again for their ability to agglutinate Siaα2,6-TRBCs (Supplementary Fig. 2). The parental control virus (designated VN1203/PR8) with avian-type receptor-binding specificity agglutinated untreated TRBCs (which possess both human-and avian-type receptors on their surface), but not TRBCs possessing predominantly human-type receptors (Siaα2,6-TRBCs; Supplementary Table 1). By contrast, of the 370 viruses originally isolated, nine agglutinated Siaα2,6-TRBCs, albeit with different efficiencies (Supplementary Table 1). All nine viruses possessed mutations in the region targeted for random mutagenesis; one mutant also possessed an additional mutation (E119G) in an area that was not targeted for mutation. Most of the mutations clustered around the receptor-binding pocket (Fig. 1a). Several of the selected viruses possessed mutations known to increase binding to human-type receptors, including N186K (ref. 9), S227N (ref. 5) and Q226L (which confers human-type receptor binding together with G228S)15 (all shown in blue in Fig. 1a). The identification of known determinants of human-type receptor-binding specificity from a library of random mutants validates our approach. Notably, our screen also identified mutations not previously associated with receptor-binding specificity.
Although viruses were diluted to ~0.5 viruses per well for amplification in AX4 cells, we cannot exclude the possibility that some wells were infected with more than one virus, resulting in mixed populations. To confirm the significance of the identified mutations in HA for human-type receptor binding, the mutations were engineered into a VN1203/PR8 virus (possessing an avirulent HA cleavage site sequence, as described earlier). All nine mutants were generated; however, after two passages in MDCK cells, the S136N mutation reverted to the wild-type sequence. This mutant was excluded from further evaluation.
First, we confirmed the binding of the remaining eight variants to Siaα2,6-TRBCs (Supplementary Table 1). For comparison, we included a VN1203/PR8 virus with two changes in its HA (Q226L and G228S) previously shown to have increased binding to Siaα2,6Gal6,15. Indeed, compared to the wild-type VN1203/PR8 virus, the Q226L/G228S mutant displayed an increased ability to bind to human-type receptors. For the recreated variants, haemagglutination titres were higher and slightly different from the initial characterization, which we attribute to biological differences (the initial characterization was carried out with non-concentrated cell culture supernatant and potentially mixed virus populations, whereas the recreated viruses were concentrated and purified) and to experimental differences (that is, differences between the TRBC batches or the efficiency of α2,3-sialidase treatment, or both). Collectively, however, these experiments demonstrate that this random mutagenesis approach allows the identification of hitherto unrecognized amino acid substitutions that permit avian virus HAs to bind to human-type receptors.
To characterize further the receptor-binding properties of the selected variants, we used solid-phase binding assays in which sialylglycopolymers were absorbed to plates, which were then incubated with virus (Fig. 2a). A virus possessing the HA and NA genes of the seasonal human A/Kawasaki/173/2001 (H1N1; K173) virus and the remaining genes from PR8 (K173/PR8) served as a control virus with typical human-type receptor specificity. Indeed, K173/PR8 preferentially bound to Siaα2,6Gal. In contrast, VN1203/PR8 bound to only Siaα2,3Gal. As reported elsewhere6,15, the Q226L/G228S mutations led to increased binding to Siaα2,6Gal. Variants I202T/R220S, W153R/T160I, N169I/H184L/I217M and H130Q/K157E resembled VN1203/PR8 in their binding to glycans, despite the fact that these mutants weakly agglutinated Siaα2,6-TRBCs (see Supplementary Table 1). These viruses may have bound to glycans on TRBCs that were different from Siaα2,6Galβ1,4GlcNAc used in this study. However, variants N186K/M230I, S227N/G228A and Q226L/E231G showed an appreciable increase in binding to Siaα2,6Gal but also retained binding capacity for Siaα2,3Gal. Of all of the variants tested, only E119G/V152I/N224K/Q226L exhibited specificity for only Siaα2,6Gal. Thus, only one H5 HA variant with receptor-binding capability akin to that of seasonal influenza viruses was isolated from the library screen of 2.1 X 106 viruses. To identify the amino acid change(s) responsible for the conversion from Siaα2,3Gal to Siaα2,6Gal recognition in the E119G/V152I/N224K/Q226L virus HA, we tested the amino acid changes at positions 119, 152, 224 and 226 individually and in various combinations. Solid-phase binding assays demonstrated that the N224K/Q226L combination is critical for the shift from Siaα2,3Gal to Siaα2,6Gal recognition (Fig. 2b); Q226L in combination with V152I also conferred weak binding to α2,6-glycans.