Diversity of functionally permissive sequences in the receptor-binding site of influenza hemagglutinin

SUMMARY

Influenza A virus hemagglutinin (HA) initiates viral entry by engaging host receptor sialylated glycans via its receptor-binding site (RBS). The amino-acid sequence of the RBS naturally varies across avian and human influenza virus subtypes and is also evolvable. However, functional sequence diversity in the RBS has not been fully explored. Here, we performed a large-scale mutational analysis of the RBS of A/WSN/33 (H1N1) and A/Hong Kong/1/1968 (H3N2) HAs. Many replication-competent mutants not yet observed in nature were identified, including some that could escape from an RBS-targeted broadly neutralizing antibody. This functional sequence diversity is made possible by pervasive epistasis in the RBS 220-loop and can be buffered by avidity in viral receptor binding. Overall, our study reveals that the HA RBS can accommodate a much greater range of sequence diversity that previously thought, which has significant implications for the complex evolutionary interrelationships between receptor specificity and immune escape.

eToc Blurb

Wu et al. performed a large-scale mutational analysis of the influenza hemagglutinin receptor-binding site (RBS). A large number of replication-competent RBS mutants were observed. Such functional sequence diversity is made possible by pervasive epistasis in the RBS 220-loop and can be buffered by avidity in viral receptor binding.

INTRODUCTION

Influenza A viruses are moving targets for vaccine and drug development due to their rapidly evolving nature. Emerging drug resistance and immune escape after vaccination or natural infection demand a continuous effort in improving existing antivirals or developing new ones, and in annual re-formulation of influenza vaccines. Among the proteins encoded by the 8-segment influenza A virus genome, the hemagglutinin glycoprotein (HA) is the major surface antigen, and evolves at an exceptionally high rate (Bhatt et al., 2011) as an RNA virus (Burton et al., 2012; Duffy et al., 2008) and due to selection pressure by the humoral immune system. Influenza A virus HA has been classified into 18 subtypes (H1 to H18) based on its antigenic properties. Despite the high genetic divergence among subtypes, only three subtypes (H1, H2, and H3) have been associated with human pandemics, but five other subtypes (H5, H6, H7, H9, and H10) have sporadically emerged in human population. The co-circulation of multiple subtypes adds another layer of complexity to the study of influenza A virus evolution and increases the challenge in the development of antivirals and vaccines.

HA plays a critical role in initiating the influenza A virus replication cycle by binding to its sialic acid receptor. The hemagglutinin receptor-binding site (RBS) is composed of the 130-loop, 150-loop, 190-helix, and 220-loop (Wilson et al., 1981) with Tyr98, Trp153, His183, and Tyr195 (H3 numbering), being highly conserved and important for sialic acid interaction (Ha et al., 2001; Skehel and Wiley, 2000). While the 130-loop, 150-loop, 190-helix are relatively conserved among HA subtypes, a higher genetic diversity has been detected in the 220-loop. Variation in the RBS among subtypes results in different binding modes to the host receptors and different mechanisms for switching tropism (Matrosovich et al., 2000; Shi et al., 2014). Such a difference has been well characterized for human adaptation of avian influenza viruses, where E190D/G225D have been employed in the H1 subtype (Glaser et al., 2005; Matrosovich et al., 1997; Stevens et al., 2006a; Tumpey et al., 2007), and Q226L/G228S in H2 and H3 subtype (Connor et al., 1994; Pappas et al., 2010; Rogers et al., 1983; Xu et al., 2010b). In addition, the HA RBS continues to evolve as influenza A virus circulates in the human population (Lin et al., 2012). These observations suggest that the functional constraints on the HA RBS do not completely prohibit genetic diversification, and that the HA RBS can tolerate some mutational changes without abolishing its receptor-binding function. Comprehending the structural flexibility and potential constraints in HA RBS will facilitate the understanding of the possible diversification of the receptor-binding mode among influenza species and subtypes, especially from evolution under antibody pressure for human viruses.

A number of mutational analyses have been performed to characterize the functional constraints of the HA RBS (Ayora-Talavera et al., 2009; Bradley et al., 2011; Martin et al., 1998; Xu et al., 2010a; Yamada et al., 2006). However, previous studies have usually focused on a small panel of mutations from natural isolates or from adaptations in the laboratory. Mutations that were not associated with replication-competent viruses, as well as combinations of mutations, were rarely investigated. In this study, we aimed to systematically investigate the in vitro evolutionary properties and functional sequence space of the HA RBS. Specifically, we probed the fitness consequences of up to three substitutions across 11 residues in influenza A/WSN/33 (H1N1) virus HA RBS using deep mutational scanning. Our results revealed that many single substitutions were deleterious when presented alone, but were viable in combination with other substitutions. Such a phenomenon has been noted in natural adaption of avian influenza viruses to human receptors (Stevens et al., 2006a; Vines et al., 1998; Xu et al., 2010b). The epistatic effect observed in this study is particularly pervasive in the HA RBS 220-loop. Interestingly, some of the newly discovered, replication-competent variants were able to escape from bnAb neutralization. A second deep mutational scanning experiment on HA RBS 220-loop of influenza A/Hong Kong/1/1968 (H3N2) virus demonstrated that, despite the existence of some strain-specific evolutionary properties, many of our findings could be generalized across genetically distinct strains and subtypes. Biophysical and structural characterization further suggested that minor structural changes that diminished the monovalent receptor binding affinity could be buffered by avidity in viral receptor binding without imposing a cost in viral replication fitness. Overall, this systematic study uncovers several intrinsic evolutionary properties and perhaps fewer constraints on the HA RBS than previously appreciated, and highlights that, despite decades of research, much is still unknown regarding the HA RBS.

RESULTS

Deep mutational scanning of A/WSN/33 hemagglutinin receptor-binding site

By coupling saturation mutagenesis with deep sequencing, the fitness effects of a large number of mutations can be accessed in a parallel manner (Fowler and Fields, 2014). This approach is commonly known as deep mutational scanning and has been successfully applied to influenza virus to examine the phenotype of all single-residue mutations across HA (Thyagarajan and Bloom, 2014; Wu et al., 2014). Here, we aimed to investigate not only the fitness effects of individual mutations in HA RBS, but also the fitness effects of combinations of mutations.

Three plasmid mutant libraries were created based on the reverse genetics system of influenza A/WSN/33 (WSN, H1N1) virus (Neumann et al., 1999) (see Methods). The first library contained all possible single amino-acid substitutions across 11 residues in the HA RBS, namely residues 134, 136, 153, 155, 183, 190, 194, 195, 225, 226, and 228 (Figure 1A). The second library contained all pairwise combinations of amino-acid substitutions across the same 11 residues. The third library contained all triple combinations of amino-acid substitutions across residues 225, 226 and 228, which have been shown to be responsible for changes in receptor specificity of avian and human viruses (Matrosovich et al., 2000; Shi et al., 2014), and hence naturally more diverse compared to the other 8 residues of interest, some of which are almost absolutely conserved (e.g. Trp153, His183, Tyr195) for binding the sialic acid moiety. Our deep mutational scanning dataset allowed us to determine the fitness effect of all single substitutions (209 out of 209), 50% of all double substitutions (9,834 out of 19,855), and 82% of all triple substitutions (5,636 out of 6,859). The relative fitness index (RF index) was used as a proxy to estimate the replication fitness effect for each mutant as previously described (Wu et al., 2015) (see Methods). Briefly, the more deleterious the mutant was the smaller the RF index. The RF index of wild type (WT) was set as 1. The RF index of silent mutations (mean = 4.3) was significantly higher than that of nonsense mutations (mean = 3e-4) (p < 2.2e-16, Wilcoxon rank-sum test, Figure S1A), validating the fitness selection.

Figure 1. Profiling of fitness effects of mutations in WSN HA RBS.

(A) The locations of mutated residues are indicated on the structure. PDB 1RVX is used (Gamblin et al., 2004), since it is the only hemagglutinin structure that matches WSN for the amino-acid identities at the residues of interest. (B) The maximum RF index among all mutants that carry the indicated substitution is shown. If all mutants that carry a given substitution are lethal in the deep mutational scanning experiment, the cell in the heatmap that corresponds to that particular substitution is colored in white. If at least one mutant that carries a given substitution is as fit as the wild-type (WT) in the deep mutational scanning experiment, the cell in the heatmap that corresponds to that particular substitution is colored in red. “Single mutants” means the RF index of the indicated single substitution. “Double mutants” means the maximum RF index of all single and double mutants that contained the indicated substitution. “Triple mutants” means the maximum RF index of all single, double and triple mutants that contained the indicated substitution. The RF index of the amino acid representing the WT sequence is set as 1. For visualization purpose, the maximum RF index is capped at 1.

To understand the functional sequence diversity of HA RBS, we first aimed to identify viable amino-acid substitutions. A preliminary analysis was performed to examine the RF index of amino-acid substitutions at each of the 11 residues of interest and the maximum RF index that a specified single-substitution could achieve when combined with other substitutions. Our data indicated that most amino-acid substitutions were deleterious to the virus when presented alone (Figure 1B). For the substitutions at residues 225, 226, and 228, the maximum RF index that a specified single-substitution could achieve when combined with another substitution is significantly higher than when present alone (P = 1.8e-6, Wilcoxon signed-rank test, Figure S1B; n.b. the addition of a second additional substitution did not significantly alter the RF index (P = 0.69)). This suggests the presence of a non-additive mutational fitness effect, also termed epistasis. The prevalence of epistasis can be further highlighted by the fact that while 96% (55 out of 57) of the substitutions in the 220-loop are highly deleterious (RF index < 0.1) in the wild-type genetic background (Figure 1B), 20% of them (11 out of 55) are beneficial (RF index increases by > 0.5) in at least one of the genetic backgrounds examined in the deep mutational scanning experiment (Figure S1C). This positive epistatic effect expands the possible functional sequence diversity in the 220-loop of the HA RBS. It should be noted that such observations are consistent with the general notion that epistasis can increase the number of acceptable mutations (Povolotskaya and Kondrashov, 2010; Weinreich et al., 2005).

Validation of the deep mutational scanning data and epistasis in the 220-loop

To validate the high-throughput genetics data, a panel of 31 WSN mutants was individually constructed and analyzed by a virus rescue experiment. The virus titer in the rescue experiment, which was measured by TCID₅₀, and the RF index from deep mutational scanning has a Pearson correlation of 0.81 (Figure 2A). The virus titer was also measured by a hemagglutination assay (Figure 2B–C and Figure S2A). The HA titer and RF index has a Pearson correlation of 0.68 (Figure 2B). Both of these experimental results confirm the validity of the deep mutational scanning data and analysis.

Figure 2. Experimental validation and epistasis in the 220-loop of WSN HA RBS.

(A) The relationship between log₁₀ RF index and log₁₀ TCID₅₀ from virus rescue experiment of WSN mutants is shown as a scatterplot. The green dashed line represents the lower detection limit of the TCID₅₀ assay. (B) The relationship between log₁₀ RF index and log₂ HA titer measured by agglutination of turkey red blood cell is shown. (C) Representative images of turkey red blood cell agglutination in the presence of WSN virus with increasing dilution factor. With a high dilution factor, red blood cell formed a palette. (D) The log₁₀ RF index (upper panels) and the log₁₀ TCID₅₀ (lower panels) of the indicated double-mutant cycle is shown. A double-mutant cycle includes the wild-type (WT), two single mutants, and the corresponding double mutant. The blue and red lines represent the each of the two possible evolutionary pathways from WT to the double mutant (from left to right). The green dashed line represents the lower detection limit of the TCID₅₀ assay. (E) The left panels represent the double-mutant cycle on wild-type (WT). The right panels represent the double-mutant cycle in the presence of D225E. The TCID₅₀, HA titer, and RF index for each validated WSN variants are listed in Dataset S1.

Our validation experiment was also designed to corroborate some of the epistatic interactions in the 220-loop by including several relevant mutants in the rescue experiment (Figure 2A–B). Here, we focused on the most extreme form of epistasis, namely reciprocal sign epistasis, in which a double substitution is less deleterious compared to each single substitution alone. The existence of reciprocal sign epistasis in the 220 loop of the HA RBS was indeed validated. For example, D225L or Q226S alone were deleterious, but in combination were neutral (Figure 2D). Reciprocal sign epistasis was also observed and validated for D225Q/Q226A and Q226N/G228A. Many more examples could be found in the deep mutational scanning data (Figure S2B). Interestingly, we also observed and validated an example of pairwise epistasis in the 220-loop that could be affected by the presence of a third mutation (higher-order epistasis). Q226A and G228A were deleterious when presented alone or together (Figure 2E). When a further substitution D225E was present, Q226A and G228A individually were still deleterious, but there was no fitness cost when introduced together (reciprocal sign epistasis) (Figure 2E). These results confirm the prevalence of epistasis in the 220-loop of HA.

Mutations in the 220-loop allow escape from broadly neutralizing antibodies

We hypothesized that some of the replication-competent mutants identified in this study would be able to escape from RBS-targeted broadly neutralizing antibodies (bnAb). The neutralization activities of two RBS-targeted broadly neutralizing antibodies, namely C05 IgG (Ekiert et al., 2012) and S139/1 IgG (Lee et al., 2012; Yoshida et al., 2009), were analyzed. While none of the tested WSN mutants reduced the neutralization activity of C05 (Figure 3A), many mutants were able to escape neutralization by S139/1 (Figure 3B). We also purified the HA ectodomain of WT and three genetically diverse mutants (D225Q/Q226A, D225L/Q226S/G228S, and D225M/Q226T/G228A) to test their binding activity against C05 IgG and S139/1 IgG using biolayer interferometry (Figure 3C–D and Table S1). Consistent with the neutralization assay, C05 IgG bound readily to all tested mutants, whereas S139/1 had poor binding activity against the three mutants as compared to WT. These results suggest that some replication-competent mutants identified in this study could impact the antigenic properties of the virus.

Figure 3. WSN escape mutants of a broadly neutralizing antibody.

(A) The neutralizing activity of C05 IgG against different mutants was measured by a cell viability assay. (B) The neutralizing activity of S139/1 IgG against different mutants was measured by a cell viability assay. (C–D) The binding of (C) 50 nM C05 IgG and (D) 50 nM S139/1 IgG to immobilized biotinylated purified HA mutants was measured by Bio-Layer Interferometry (BLI). The raw data (semitransparent) are overlaid with solid lines that represents the best fit model (1:2 bivalent analyte model). The dissociation constants (K_d) are listed in Table S1.

Deep mutational scanning of A/Hong Kong/1/1968 HA 220-loop

The observed functional sequence diversity in the 220-loop of WSN HA motivated us to ask whether the same phenomenon could be observed in other virus strains or other cell lines for viral passaging. Thus, we performed a deep mutational scanning on HA residues 225, 226 and 228 of influenza A/Hong Kong/1/1968 (HK68, H3N2) virus. HAs from HK68 and WSN are from distinct genetic and structural classes, in which WSN HA was from influenza A group 1 and HK68 HA from group 2. Here, a 6:2 reassortant (6 internal genes from WSN with chimeric HA and NA from HK68) was used (see Methods).

Our deep mutational scanning dataset on HK68 allowed us to determine the fitness effect of 95% of all single substitutions (54 out of 57), 60% of all double substitutions (649 out of 1,083), and 33% of all triple substitutions (2,289 out of 6,859) at residues 225, 226 and 228. The number of tolerated substitutions at these residues in HK68 increased when in combination with other substitutions (Figure 4A). For the substitutions at residues 225, 226, and 228 in HK68, the maximum RF index that a specified single-substitution could achieve when combined with one or two substitutions is significantly higher than when present alone (P = 7.2e-7 and P=7.4e-10, respectively, Wilcoxon signed-rank test, Figure S3A). In this case, there is a modest but significant difference between one and two additional substitutions (P = 0.02). In fact, 26% (15 out of 57) of the substitutions in the 220-loop of HK68 HA RBS are highly deleterious (RF index < 0.1) in the wild-type genetic background (Figure 4A), but 40% of them (6 out of 15) are beneficial (RF index increases by > 0.5) in at least one of the genetic backgrounds that were examined in the deep mutational scanning experiment (Figure S3B). This observation suggests that the 220-loop of HK68 HA RBS also has a high permissiveness of functional sequences and its residues are epistatic.

Figure 4. Profiling the fitness effects of mutations in HK68 HA RBS.

(A) The maximum RF index of all HK68 mutants that contained the indicated substitution is shown as heatmaps. This is the same analysis as in Figure 1B, but on HK68. The RF index of the amino acid representing the WT sequence is set as 1. For visualization purpose, the maximum RF index is capped at 1. (B) The relationship between log₁₀ RF index and log₁₀ TCID₅₀ from virus rescue experiment is shown as a scatterplot. The green dashed line represents the lower detection limit of the TCID₅₀ assay. (C) The relationship between log₁₀ RF index and log₂ HA titer measured by agglutination of turkey red blood cell is shown. The TCID₅₀, HA titer, and RF index for each validated HK68 variants are listed in Dataset S1. (D) The log₁₀ RF index (upper panels) and the log₁₀ TCID₅₀ (lower panels) of the indicated double-mutant cycle is shown. The green dashed line represents the lower detection limit of the TCID₅₀ assay. (E) The left panels represent the double-mutant cycle on wild-type (WT). The right panels represent the double-mutant cycle in the presence of G225E. (F) The neutralizing activity of S139/1 IgG against different mutants were measured by a cell viability assay. Data are represented as mean ± SD from three replicates.

For H3N2 strains, a single substitution D151G on neuraminidase (NA) is known to support HA-independent receptor binding (Lin et al., 2010). As a control experiment, next-generation sequencing was performed on the corresponding region on NA and showed that >99.9% of virus retained an Asp at residue 151 on NA in the post-passaged mutant library. This control experiment shows that the result from the deep mutational experiment of HK68 was not confounded by any de novo mutation at residue 151. We also compared the RF index of silent mutations and nonsense mutations for further validation of our deep mutational scanning data on HK68. The RF index of silent mutations (mean = 1.19) was significantly higher than that of nonsense mutations (mean = 0.04) (P = 1.9e-8, Wilcoxon rank-sum test, Figure S3C). In addition, a panel of 19 mutants were individually constructed and analyzed by virus rescue experiment. The virus titer measured by TCID₅₀ (Figure 4B) and the hemagglutination assay (Figure 4C) correlated well with the RF index (Pearson correlation = 0.89 and 0.91, respectively). These results confirm the validity of the deep mutational scanning data of 220-loop of HK68 HA RBS and imply generality of functional sequence diversity in the 220-loop of the HA RBS.

Characterization of HK68 HA 220-loop mutants

Similar to WSN, reciprocal sign epistasis was also identified and validated in the 220-loop of HK68 HA RBS. Among the panel of 19 validated mutants, L226N and S228A exhibited reciprocal sign epistasis (Figure 4D). L226N and S228A were deleterious individually (particularly L226N), but were WT-like in combination. Additional examples of reciprocal sign epistasis could be identified in the HK68 deep mutational scanning data (Figure S4A). Moreover, we also validated an example of higher-order epistasis, where reciprocal sign epistasis existed between L226R and S228E only in the presence of G225E (Figure 4E), analogous to the D225E-dependent epistatic interaction between Q226A and G228A in WSN (Figure 2E).

Several HK68 mutants were also tested for the ability to escape from bnAb S139/1 (Lee et al., 2012; Yoshida et al., 2009). Among the six mutants tested, a triple-substitution mutant G225M/L226T/S228A was able to escape from S139/1. This escape may attribute to the ~200-fold reduction in affinity of S139/1 IgG against G225M/L226T/S228A as compared to HK68 WT HA (Figure S4B–C). In fact, the same triple-substitution mutant in WSN also allowed escape from S139/1 (Figure 3B and 3D). Overall, these results suggest that the high diversity of functional sequences resulting from epistasis, along with its influence in bnAb escape, can be observed in the HA RBS 220-loop of genetically distinct strains.

Comparing the evolutionary properties of HK68 and WSN HA 220-loops

We next aimed to further compare and contrast the evolutionary properties of HK68 and WSN HA 220-loops. Based on the deep mutational scanning data, the distributions of reciprocal sign epistasis in HK68 and WSN HA 220-loops were compared (Figure S5). For WSN, the occurrence of reciprocal sign epistasis between residues 225 and 226 was significantly greater than between residues 225 and 228 (P = 3e-12, Fisher’s exact test) and between residues 226 and 228 (P = 2e-10, Fisher’s exact test, Figure 5A). In contrast, for HK68, the occurrence of reciprocal sign epistasis between residues 226 and 228 was significantly greater than between residues 225 and 228 (P = 0.002, Fisher’s exact test) and between residues 225 and 226 (P = 0.015, Fisher’s exact test, Figure 5B). In fact, residues 226 and 228 are the determinants for tropism switching in natural H3 strains (Connor et al., 1994; Matrosovich et al., 2000; Rogers et al., 1983; Shi et al., 2014). These results indicated that the hot spot for reciprocal sign epistasis in HA RBS might vary in different strains. Despite this difference, reciprocal sign epistasis that was conserved between strains could still be identified, as exemplified by substitutions resulting in 226N and 228A, which were validated in both WSN and HK68 (Figure 2D and 4D).

Figure 5. Comparison of the evolutionary properties of WSN and HK68 RBS 220-loops.

(A–B) The fraction of pairwise substitutions at the indicated residue pairs classified as reciprocal sign epistasis is shown for (A) WSN, and (B) HK68. A pair of substitution is classified as interacting with reciprocal sign epistasis when both single substitutions have a lower RF index than WT and the double substitution. Fisher’s exact test was performed to compute the p-value. (C) A total of 3,269 variants were included in the fitness profiling dataset for WSN and HK68. Each variant is denoted by the single-letter codes of amino acids at residues 225, 226, and 228. For each variant, the relationship between the log₁₀ RF index for WSN and for HK68 is shown. Variants carrying a stop codon (nonsense variants) are colored in red. Variants that are included in the validation experiment of both WSN and HK68 are colored in green. Variants that are included in the validation experiment of WSN but not HK68 are colored in blue. Variants that are included in the validation experiment of HK68 but not WSN are colored in yellow. Variants corresponding to WT sequence are colored in purple. DQG is the WT sequence for WSN and GLS for HK68. (D) Virus rescue experiments were performed for nine HK68 HA variants including WT. However, instead of pairing with HK68 neuraminidase (N2) as in Figure 4, WSN neuraminidase (N1) was used. This experiment was designed to examine the influence of neuraminidase subtype on the fitness of the tested HK68 HA variants.

We further compared the fitness effect of each amino-acid combination at residues 225, 226 and 228 in WSN and HK68 HA RBS 220 loops. For clarity, each variant is denoted by the single-letter codes of amino acids at residues 225, 226, and 228. For instance, variant DQG represents the WT sequence at 225, 226, and 228 for WSN and variant GLS represents the WT sequence for HK68. The relationships between the fitness effect in WSN and in HK68 for individual variants were visualized by a scatterplot (Figure 5C). An interesting observation was that many variants could be found in the upper left quadrant, but not the lower right quadrant. The lack of variants in this lower right quadrant indicated that HK68 could tolerate more variants as compared to WSN. The high tolerance of variants in HK68 was unlikely due to the difference in NA between WSN and HK68, since pairing with the NA from WSN did not affect the rescue efficiency of all tested HK68 HA variants (Figure 5D). Perhaps surprisingly, we also observed many variants that could be tolerated by both WSN and HK68 (upper right quadrant of Figure 5C). Of note, two of those variants, namely MTA and LSS could be rescued to a WT-phenotype in both WSN and HK68 as measured by TCID₅₀ (Dataset S1).

Together, these results demonstrate that while HA RBS 220-loops from different strains possess different evolutionary properties, some variants can be commonly tolerated. Thus, the difference in the functional constraints of RBS 220-loop among different HA subtypes may not be as pronounced as previously suggested based on natural evolution patterns (Shi et al., 2014).

Biophysical and structural characterization of HA RBS 220-loop variants

To obtain structural insight into how mutations can be tolerated by the HA RBS 220-loop, we determined crystal structures of three variants of HK68 in apo form and in complex with 3′-sialyl-N-acetyllactosamine (3′-SLN) or with 6′-SLN (Table S2). The three variants were MTA, LSS, and QAS. MTA and LSS were WT-like in both WSN and HK68 as described above (Figure 5C). QAS represented a double substitution G225Q/L226A for HK68, in which the same double substitution was WT-like in WSN. Although these variants had WT-like virus replication fitness and capacity for aggregating red blood cells (Dataset S1), their affinities as recombinant proteins towards 3′-sialylated di-N-acetyllactosamine (3′-SLNLN) and 6′-SLNLN were generally weaker (Figure S6A–B and Table S3). This observation suggests that the polyvalent or avidity effect of the HA-receptor interaction on viral surface can buffer the decrease in monovalent binding (Peng et al., 2017), which is analogous to the disproportionally high potency of RBS-targeted bnAbs, when compared to their individual Fab affinities, due to the bivalency of IgG (Ekiert et al., 2012; Lee et al., 2012). This is further supported by the results from binding assays that involved polyvalent interaction. For example, the off-rate (k_off) in the binding between 3′-SLNLN and WSN virus (Figure S6C) is several fold to a magnitude less (i.e. slower) than between 3′-SLNLN and WSN HA recombinant protein (Figure S6A and Table S4). This implies that the polyvalency of HA on virus surface facilitates tighter binding to the receptor. In addition, binding of polymerized sialylated di-N-acetyllactosamine (3′-SLNLN-PAA or 6′-SLNLN-PAA) to immobilized HK68 HA (WT or variants) exhibits a dissociation constant (K_d) in the pM range (Figure S6D and Table S5). The avidity of this polyvalent binding is much stronger than that for monovalent binding (Figure S6B), where the K_d in the high nM to μM range (Table S3). Overall, these results imply the weak monovalent HA-receptor binding can be greatly enhanced by polyvalency, which in turn is important for tolerating substitutions that enable immune escape, but that also decrease HA-receptor binding affinity. This is consistent with the previous observations on Y98F substitution (Martin et al., 1998), where a large reduction in HA-receptor binding affinity did not impair virus growth in vitro.

As compared to the WT (Ekiert et al., 2012), the three HK68 variants (MTA, LSS, and QAS) have a very similar overall backbone conformation (Figure 6A). Nonetheless, a minor shift of 220-loop backbone of around 0.8 to 1 Å is observed in each of these three variants (Figure 6B). The 220-loop conformations in these three variants are stabilized by intramolecular hydrogen bonds that are not observed in WT. In MTA, T226 forms a side chain to side chain hydrogen bond with Y98 (Figure 6C). In LSS, S226 also forms a a similar hydrogen bond with Y98 (Figure 6D). In QAS, the Q225 side chain hydrogen bonds with the main-chain carbonyl of N137 (Figure 6E). Consistently, there are no major backbone conformational differences when comparing the apo structure with complexes with 3′-SLN or 6′-SLN for each of the three variants (Figure S7).

Figure 6. Structural characterization of HK68 HA RBS 220-loop variants.

(A) Comparison of HA RBS backbone conformation between WT (gray) and three variants, namely MTA (lime), LSS (pink), and QAS (cyan). For the WT structure, PDB 4FNK was used (Ekiert et al., 2012). (B) Comparison of 220-loop backbone conformation between WT and three variants. Distance between the Cα of residue 225 (sphere representation) in WT and in each variant is indicated. (C–E) Side chains of residues 225, 226 and 228 and interacting residues (hydrogen bonding) are shown in sticks representation. Hydrogen bonds that are present in the HK68 HA RBS 220-loop variants but not in the wild type (WT) are shown in black dashed lines for (C) MTA, (D) LSS, and (E) QAS.

The deviation of the 220-loop backbone from WT can be contributed to the left-handed helical conformation of residue 225 (positive ϕ), which is where the largest displacement of the 220-loop backbone occurs (Figure 6B–E). The ϕ angle of residue 225 is also likely constrained by the proximity of the side chain of W222 (Figure 7A), where the backbone amide between residues 224 and 225 forms an aromatic hydrogen bond with the six-membered aromatic ring of W222, which can translate to around 3 kcal/mol in stabilizing enthalpy (Levitt and Perutz, 1988). This may explain why residue 225 still adopts a left-handed helical conformation despite the non-Gly substitutions in the three variants. Flipping of the peptide bond between residues 224 and 225 into a right-handed conformation would result in unfavorable electrostatic interactions between the carbonyl group and the Π electrons of the Trp aromatic ring (Figure 7B). Nevertheless, Gly at residue 225 of HK68 WT (Ekiert et al., 2012; Lee et al., 2015; Yang et al., 2015) often has a larger ϕ angle and smaller ψ angle compared to the three variants (MTA, LSS, and QAS) (Figure 7C). This positions G225 away from the more limited, but still favored left-handed region for non-Gly amino acids in the Ramachandran plot (Lovell et al., 2003). Subsequently, in those three variants, the ϕ and ψ constraints of non-Gly amino acids (Met, Leu, and Gln) at residue 225 cause a shift in the 220-loop towards the more favored region of the Ramachandran plot for non-Gly residues. Along with the results from the binding kinetic assay (Figure S6 and Table S3), our data suggest that non-Gly substitutions at residue 225 decrease the affinity against sialic acid receptor by causing minor changes in the 220-loop backbone conformation. This agrees with previous structural analyses on H2 and H3 subtype showing that minor changes in the 220-loop backbone conformation impact the affinity against the sialic acid receptor (Weis et al., 1988; Xu et al., 2010b).

Figure 7. Structural comparison of HK68 HA RBS 220-loop variants with WT.

(A) Residue 225 in the crystal structures of wild type (WT) (Ekiert et al., 2012) and all three variants (MTA, LSS, and QAS) is in left-handed helix conformation. The side chain of W222, and main chain of residues 224 and 225 are shown in sticks representation. In the left-handed helix conformation, the distance between the backbone amide nitrogen atom of residue 225 to the center of the aromatic ring is around 3.4 to 3.6 Å. (B) When residue 225 is modeled in right-handed helix conformation, the main-chain carbonyl of residue 224 points towards the plane of the indole ring in W222. These two nucleophilic groups would result in a repulsive force. In this conformation, the distance between the backbone carbonyl oxygen atom of residue 224 to the center of the aromatic ring would be around 2.4 to 2.6 Å. (C) The Ramachandran plot of residue 225 in the apo form of WT and the three indicated variants are shown. Here, three different crystal structures of WT were analyzed, namely PDB 4FNK (Ekiert et al., 2012), PDB 4ZCJ (Lee et al., 2015), and PDB 4WE4 (Yang et al., 2015). Residue 225 from each HA monomer within the crystal asymmetric unit was analyzed. The orange shading area was created based on data points for the general case of non-Gly (Lovell et al., 2003). Region in white, lighter, and darker orange represents disallowed, allowed and favored region, respectively. (D) The side-chain position and conformation of residue 226 in the crystal structures of WT (PDB 2YPG) (Lin et al., 2012) and the three indicated variants were compared in the presence of a bound human receptor analog.

Besides the backbone conformation, substitutions at residue 226 also contribute to the decrease in affinity against the sialic acid receptor. Among the side chains of the three mutated residues (225, 226, and 228) in the H3 subtype, residue 226 is known to make the largest number of contacts with the sialic acid receptor (Weis et al., 1988). Specifically, L226 in H3 HA makes van der Waals contact with C-6 of galactose-2 in α2,6-linked sialic acid (Lin et al., 2012). This van der Waals contact is lost in the three variants (MTA, LSS, and QAS) due to the shorter side chains of Thr, Ser, and Ala (Figure 7D). Overall, our results suggest that viral receptor binding avidity is important for tolerating minor structure changes that would decrease the monovalent affinity against sialic acid receptor. This buffering effect permits sequence diversification of HA RBS 220-loop without diminishing replication fitness.

DISCUSSION

Influenza A virus hemagglutinin (HA) has been of substantial interest in many biomedical fields such as virology, immunology, structural biology, and evolutionary biology. This diverse interest in HA is not only due to its critical role in receptor-binding and membrane fusion for entry of influenza virus into the host cells, but also because it is the major antigen of influenza A virus, an important determinant in host tropism, and the prototype for type 1 membrane fusion proteins (Skehel and Wiley, 2000). The receptor-binding site (RBS) is particularly intriguing region in HA. RBS binds to its host receptor, sialic acid, as the first step in viral replication cycle. At the same time, it can also be targeted by antibodies, including some that are broadly neutralizing, although not to the same extent as to the HA stem (Lee and Wilson, 2015). The HA RBS from different influenza A virus subtypes possesses slight, but important variations in their amino-acid sequences, as well as in glycosylation. The limit of HA RBS sequence diversity and the underlying functional constraints are thus not entirely clear. This study provides insight in the evolutionary limitations and the range and diversity of constraints on HA RBS that allow it to retain function while escaping from immune pressure.

We chose to focus on only 11 RBS residues due to the technical limitations on the genetic bottleneck for virus rescue (Bloom, 2014; Thyagarajan and Bloom, 2014), sequencing read length, and sequencing error rate (Zhang et al., 2016). Additional residues, such as 98 (Weis et al., 1988), 145 (Xu et al., 2012a), 186 (Shi et al., 2013; Wang et al., 2015), 222 (Xu et al., 2012a), and 227 (Gambaryan et al., 2006), also influence the receptor-binding function of HA. Given the findings in this study, sequencing diversity and evolutionary properties in those other residues deserve further investigation. With the continuous improvement in the deep mutational scanning platform for influenza virus (Doud and Bloom, 2016) and next-generation sequencing technology (Goodwin et al., 2016), we anticipate that future studies can be performed on a larger scale.

One interesting and unexpected observation in our study is the pervasive epistasis in HA RBS 220-loop. In fact, pairwise epistasis in the HA RBS has been shown to be important for tropism switching (Stevens et al., 2006b; Xu et al., 2010b). The double substitution Q226L/G228S changes the receptor specificity of H2 and H3 subtype from α2,3-linked sialic acid (avian receptor) to α2,6-linked sialic acid (human receptor) (Connor et al., 1994; Matrosovich et al., 2000; Pappas et al., 2010; Rogers et al., 1983), while G228S alone reduces the affinity towards avian receptors without increasing affinity towards human receptors (Matrosovich et al., 2000) and Q226L alone abolishes binding to most sialosides (Xu et al., 2010b). For the H1 subtype, the double substitution at G225D/E190D changes specificity from avian to human receptors, whereas each single substitution alone significantly diminishes or abolishes binding towards both receptors (Stevens et al., 2006b). Our results suggest epistasis to be an intrinsic evolutionary property of the HA RBS 220-loop. For the tropism switching in H2 subtypes, the epistatic effect between Q226L and G228S could be attributed to the fine-tuning of the 220-loop backbone conformation (around <1 Å) in addition to the side-chain effects (Xu et al., 2010b). We speculate that this constraint underpins the pairwise epistasis observed at particular residues in this study, where one substitution may cause a backbone conformation disruption that cannot be buffered by viral receptor avidity, but can be compensated by another substitution. Nonetheless, the exact structural mechanisms for such individual cases of epistasis could be different and need further investigation.

After more than three decades since the first HA structure was determined (Wilson et al., 1981), the properties and evolution of the RBS remains an active area of research (Connor et al., 1994; Glaser et al., 2005; Lin et al., 2012; Matrosovich et al., 2000; Matrosovich et al., 1997; Pappas et al., 2010; Rogers et al., 1983; Shi et al., 2014; Stevens et al., 2006a; Tumpey et al., 2007; Xu et al., 2010b; Yang et al., 2010; Zhu et al., 2015). To the best of our knowledge, this study represents the largest scale mutational analysis of the HA RBS to date. While our study demonstrates the functional sequence diversity of HA RBS and reveals several of its intrinsic evolutionary properties, our understanding is far from complete. We acknowledge that constraints underlying the evolution of RBS are also likely influenced by many factors that are not investigated in this study. For example, the density and types of glycans present on the cell surface vary in different cell types (Matrosovich et al., 2004) and in different locations on the respiratory tract in humans (Shinya et al., 2006), which would have the most direct impact on the functional requirement for HA receptor-binding to initiate viral entry and subsequent replication. In addition, HA receptor-binding affinity needs to balance with neuraminidase (NA) activity for efficient replication (Mitnaul et al., 2000; Xu et al., 2012b). Studies have also shown that the HA spike density on the virus surface can vary in different strains (Moules et al., 2011), which potentially could impact the avidity effect and, hence, the mutational tolerance. Some of these factors were described and analyzed in our study but not in detail. Future in-depth studies are needed to examine how these factors constrain or promote the evolution of the HA RBS. A comprehensive understanding of the evolutionary constraints and capacity of the RBS in different HA subtypes may lead to a unifying theme to predict how the HA RBS can continue to evolve, which will in turn be valuable for antiviral and vaccine development.

STAR METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
DMEM medium	Thermo Fisher Scientific	11995065
RPMI medium	Thermo Fisher Scientific	11875093
Expi293 expression medium	Thermo Fisher Scientific	A1435101
OPTI-MEM	Thermo Fisher Scientific	31985070
HyClone insect cell culture medium	GE Healthcare	SH30280.03
Phosphate-buffered saline (PBS)	Thermo Fisher Scientific	14040133
Fetal Bovine Serum (FBS)	Thermo Fisher Scientific	16000044
MEM non-essential amino acids	Thermo Fisher Scientific	11140050
Trypsin-EDTA	Thermo Fisher Scientific	25200056
Penicillin-Streptomycin	Thermo Fisher Scientific	15140122
Lipofactamine 2000	Thermo Fisher Scientific	11668019
Turkey red blood cells	Lampire Biological Laboratories	7249409
Ni-NTA Superflow	Qiagen	30450
Antibodies
C05	(Ekiert et al., 2012)	N/A
S139/1	(Lee et al., 2012)	N/A
Bacterial and Virus Strains
A/WSN/33 (H1N1)	(Neumann et al., 1999)	N/A
Chimeric A/Hong Kong/1/1968 (H3N2)	This study	N/A
DH10Bac competent cells	Thermo Fisher Scientific	10361012
MegaX DH10B T1R Electrocomp cells	Thermo Fisher Scientific	C640003
Chemicals, Peptides, and Recombinant Proteins
WSN wild-type	This study	N/A
WSN D225M/Q226T/G228A	This study	N/A
WSN D225L/Q226S/G228S	This study	N/A
WSN D225Q/Q226A/G228S	This study	N/A
HK68 wild-type	This study	N/A
HK68 G225M/L226T/S228A	This study	N/A
HK68 G225L/L226S	This study	N/A
HK68 G225Q/L226A	This study	N/A
BsmBI	New England Biolabs	R0580L
DpnI	New England Biolabs	R0176L
T4 DNA Ligase	New England Biolabs	M0202L
Trypsin	New England Biolabs	P8101S
TPCK-Trypsin	Thermo Fisher Scientific	20233
RNaseOUT	Thermo Fisher Scientific	10777019
Sodium chloride (NaCl)	Sigma-Aldrich	S9888
Tris Base	Sigma-Aldrich	11814273001
Concentrated hydrochloric acid (HCl)	Sigma-Aldrich	H1758
Sodium azide (NaN3)	Sigma-Aldrich	S2002
Bovine Serum Albumin (BSA)	Sigma-Aldrich	A9418
Tween 20	Fisher Scientific	BP337-500
3′-SLNLN	This study	N/A
6′-SLNLN	This study	N/A
3′-SLNLN-PAA	This study	N/A
6′-SLNLN-PAA	This study	N/A
Chemicals for protein crystallization	Hampton Research	N/A
Critical Commercial Assays
CellTiter-Glo Luminescent Cell Viability Assay	Promega	G7570
KOD Hot Start DNA Polymerase	EMD Millipore	71086-3
Superscript III reverse transcriptase	Thermo Fisher Scientific	18080044
QIAamp Viral RNA Mini Kit	Qiagen	52904
CellTracker Orange CMTMR Dye	Thermo Fisher Scientific	C2927
PCR Clean-Up and Gel Extraction Kit	Clontech Laboratories	740609.250
QuikChange XL Mutagenesis kit	Stratagene	200516
QIAprep Spin Miniprep Kit	Qiagen	27106
NucleoBond Xtra Maxi	Clontech Laboratories	740414.100
Deposited Data
Raw sequencing reads	This study	BioProject PRJNA353496
X-ray coordinates and structure factors	This study	PDB: 5VTX, 5VTU, 5VTZ, 5VTV, 5VTQ, 5VTY, 5VTW, 5VTR, 5VU4
Experimental Models: Cell Lines
HEK 293T cells	N/A	N/A
Expi293F cells	Thermo Fisher Scientific	A14527
A549 cells	ATCC	CCL-185
MDCK-SIAT1 cells	Sigma-Aldrich	05071502-1VL
Sf9 cells	ATCC	CRL-1711
High Five cells	Thermo Fisher Scientific	B85502
Oligonucleotides
See Methods	Integrated DNA Technologies	N/A
Recombinant DNA
pFUSE-CHIg-hG1	InvivoGen	pfuse-hchg1
pFUSE2-CLIg-hK	InvivoGen	pfuse2-hclk
WSN 8-plasmid reverse genetics	(Neumann et al., 1999)	N/A
pHW2000-chimeric HK68 HA	This study	N/A
pHW2000-chimeric HK68 NA	This study	N/A
pFast-WSN wild-type	This study	N/A
pFast-WSN D225M/Q226T/G228A	This study	N/A
pFast-WSN D225L/Q226S/G228S	This study	N/A
pFast-WSN D225Q/Q226A/G228S	This study	N/A
pFast-HK68 wild-type	(Ekiert et al., 2012)	N/A
pFast-HK68 G225M/L226T/S228A	This study	N/A
pFast-HK68 G225L/L226S	This study	N/A
pFast-HK68 G225Q/L226A	This study	N/A
Software and Algorithms
R	https://www.r-project.org	N/A
Python	https://www.python.org	N/A
GraphPad Prism	https://www.graphpad.com	N/A
HKL2000	(Otwinowski and Minor, 1997)	N/A
Phaser	(McCoy et al., 2007)	N/A
Coot	(Emsley et al., 2010)	N/A
Refmac5	(Murshudov et al., 2011)	N/A
MolProbity	(Chen et al., 2010)	N/A
Custom scripts	This study	N/A

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Ian A. Wilson (wilson@scripps.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell cultures

HEK 293T cells were maintained in DMEM medium supplemented with 10% FBS, 1x MEM non-essential amino acids, and 100 U mL⁻¹ of Penicillin-Streptomycin. A549 cells (human lung carcinoma cells) were maintained in RPMI medium supplemented with 10% FBS, 1x MEM non-essential amino acids, and 100 U mL⁻¹ of Penicillin-Streptomycin. Sf9 cells and High Five cells were maintained HyClone insect cell culture medium. Expi293F cells were maintained in Expi293 expression medium. MDCK-SIAT1 cells were maintained in DMEM medium supplemented with 10% FBS, 1x MEM non-essential amino acids, and 100 U mL⁻¹ of Penicillin-Streptomycin.

Influenza virus

WSN

Influenza viruses (WT, mutant, or mutant libraries) were constructed based on the eight-plasmid reverse genetics system (Neumann et al., 1999). For virus rescue experiments performed with WSN, HEK 293T cells in DMEM medium supplemented with 10% FBS was seeded at a 6-well plate at a density of 0.5 million cells per well, and incubated overnight at 37°C the day prior to transfection. Transfection was performed using lipofactamine 2000 according to manufacturer’s instructions. For passaging and tittering, A549 cells in RPMI medium supplemented with 10% FBS were used.

HK68

The chimeric HK68 HA segment was constructed by flanking the HK68 HA ectodomain (amino-acid residues 1 _HA1 to 175_HA2, H3 numbering) with the 32-nucleotide 3′ non-coding region and the first 19 amino acids of WSN HA at the N-terminus, and with the last 46 amino acids of WSN HA and the 48-nucleotide 5′ non-coding region at the C-terminus. The chimeric HK68 NA segment was constructed by flanking the HK68 NA coding region with the 19-nucleotide 3′ non-coding region at the N-terminus, and with the 5′ non-coding region at the C-terminus. Similar chimeric designs have been utilized for vaccine design (Gomila et al., 2013; Harvey et al., 2011; Harvey et al., 2010; Ping et al., 2015).

For virus rescue experiments performed with chimeric HK68, transfection was performed in HEK 293T/MDCK-SIAT1 cells co-culture (ratio of 6:1) using lipofactamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s instructions. Virus rescue experiments for chimeric HK68 (WT, mutant, or mutant libraries) were performed with the HA and NA segments from chimeric HK68 (Figure S5) and the other six WT segments from WSN. At 24 hours post-transfection, cells being washed twice with PBS and cell culture medium was replaced with OPTI-MEM medium supplemented with 0.8 μg/mL TPCK-trypsin. Virus was harvested at 72 hours post-transfection. MDCK-SIAT1 cells were used for tittering and infection. For titering and infection, cells were washed twice with PBS prior to the addition of virus, and OPTI-MEM medium supplemented with 0.8 μg/mL TPCK-trypsin was used.

METHOD DETAIL

Construction of mutant libraries

The HA plasmid mutant libraries were created by ligating a mutant library insert and a PCR-generated vector. The mutant libraries were built based on the pHW2000 plasmid (Neumann et al., 1999) that encoded HA influenza from A/WSN/33 (WSN) or from chimeric A/Hong Kong/1/1968 (HK68). For WSN mutant libraries, two silent mutations were introduced at the codons before and after the codon that was being randomized. These silent mutations acted as an internal barcode to indicate which codon position was being randomized and allowed distinguishing randomized codon positions from sequencing errors (Olson et al., 2014). In the chimeric HK68 HA and NA, the WSN vRNA packaging signals were preserved to increase rescue efficiency (Fujii et al., 2003; Watanabe et al., 2003) (Figure S5). To alter the HA segment, we replaced the coding region of the WSN HA ectodomain with that of HK68 HA (Figure S5A). For the NA segment, the entire coding region of full-length WSN was replaced with that of HK68 (Figure S5B). In both HA and NA segments, the non-coding regions were from WSN.

All PCR reactions were performed using KOD DNA polymerase (EMD Millipore) with 1.5 mM MgSO₄, 0.2 mM of each dNTP (dATP, dCTP, dGTP, and dTTP), and 0.6 uM of forward and reverse primer according to the manufacturer’s instructions. All PCR products were purified by gel extraction using PCR Clean-Up and Gel Extraction Kit (Clontech Laboratories).

Insert for WSN single-substitution library

The WSN single-substitution mutant library was generated by two independent series of PCRs and a subsequent overlapping PCR. A total of 11 reactions in the first series of PCR represented the codon randomization at each of the 11 sites. These 11 PCRs used 5 ng of wild type (WT) plasmid as the template and shared a reverse primer: 5′-CCC TGG CGT CTC TGA CAC TTC GTG TTA CAC TCA TGC ATT GAC GCG-3′ (“universal reverse primer”). The forward primers carried three nucleotides “NNK” at the site of codon randomization and a silent mutation at each of the codon before and after the site being randomized. The sequences of the forward primers were as follows (product names are in parenthesis):

• 134F: 5′-AAG TTC ATG GCC CAA CCA CAC ATT CAA TNN KGT TAC AGT ATC ATG CTC CCA TAG GG-3′ (134-Frag1)

• 136F: 5′-ATG GCC CAA CCA CAC ATT CAA CGG AGT CNN KGT CTC ATG CTC CCA TAG GGG AAA AA-3′ (136-Frag1)

• 153F: 5′-AAA AAG CAG TTT TTA CAG AAA TTT GCT TNN KCT CAC GAA GAA GGG GGA TTC ATA CC-3′ (153-Frag1)

• 155F: 5′-CAG TTT TTA CAG AAA TTT GCT ATG GCT ANN KAA AAA GGG GGA TTC ATA CCC AAA GC-3′ (155-Frag1)

• 183F: 5′-GAA AGA AGT CCT TGT ACT ATG GGG TGT ANN KCA TCC GTC TAG CAG TGA TGA GCA AC-3′ (183-Frag1)

• 190F: 5′-GGG TGT TCA TCA CCC GTC TAG CAG TGA CNN KCA GCA GAG TCT CTA TAG TAA TGG AA-3′ (190-Frag1)

• 194F: 5′-CCC GTC TAG CAG TGA TGA GCA ACA GAG CNN KTA CAG TAA TGG AAA TGC TTA TGT CT-3′ (194-Frag1)

• 195F: 5′-GTC TAG CAG TGA TGA GCA ACA GAG TCT ANN KAG CAA TGG AAA TGC TTA TGT CTC TG-3′ (195-Frag1)

• 225F: 5′-GGA AAT AGC TGC AAG GCC CAA AGT AAG GNN KCA GCA TGG GAG GAT GAA CTA TTA CT-3′ (225-Frag1)

• 226F: 5′-AAT AGC TGC AAG GCC CAA AGT AAG AGA CNN KCA CGG GAG GAT GAA CTA TTA CTG GA-3′ (226-Frag1)

• 228F: 5′-TGC AAG GCC CAA AGT AAG AGA TCA ACA CNN KAG AAT GAA CTA TTA CTG GAC CTT GC-3′ (228-Frag1)

A total of 11 reactions were performed in the second series of PCR. These 11 PCRs used 5 ng of WT plasmid as the template and shared a forward primer: 5′-AAT GCG CGT CTC TGC TTC CAG CGA GAT CAT GGT CCT ACA TTG TAG-3′ (“universal forward primer”). The sequences for reverse primers were shown as follows (product names are in parenthesis):

• 134R: 5′-GAA TGT GTG GTT GGG CCA TGA ACT TTC CTT-3′ (134-Frag2)

• 136R: 5′-TCC GTT GAA TGT GTG GTT GGG CCA TGA ACT-3′ (136-Frag2)

• 153R: 5′-CAA ATT TCT GTA AAA ACT GCT TTT TCC CCT-3′ (153-Frag2)

• 155R: 5′-CCA TAG CAA ATT TCT GTA AAA ACT GCT TTT-3′ (155-Frag2)

• 183R: 5′-ACC CCA TAG TAC AAG GAC TTC TTT CCC TTT-3′ (183-Frag2)

• 190R: 5′-ACT GCT AGA CGG GTG ATG AAC ACC CCA TAG-3′ (190-Frag2)

• 194R: 5′-CTG TTG CTC ATC ACT GCT AGA CGG GTG ATG-3′ (194-Frag2)

• 195R: 5′-ACT CTG TTG CTC ATC ACT GCT AGA CGG GTG-3′ (195-Frag2)

• 225R: 5′-TAC TTT GGG CCT TGC AGC TAT TTC CGG GGT-3′ (225-Frag2)

• 226R: 5′-TCT TAC TTT GGG CCT TGC AGC TAT TTC CGG-3′ (226-Frag2)

• 228R: 5′-TTG ATC TCT TAC TTT GGG CCT TGC AGC TAT-3′ (228-Frag2)

The products from the second series of PCR paired with the products from the first series of PCR as the templates for a subsequent overlapping PCR. For example, 134-Frag1 was mixed with 134-Frag2 at a 1:1 molar ratio as a template. These 11 overlapping PCRs used the “universal forward primer” and the “universal reverse primer”. Each of the 11 products from overlapping PCRs represented a site-specific codon randomization insert. Here, we denoted these 11 site-specific codon randomization inserts as “134 insert”, “136 insert”, “153 insert”, “155 insert”, “183 insert”, “190 insert”, “194 insert”, “195 insert”, “225 insert”, “226 insert”, and “228 insert”, where the number represents the residue that was being randomized. The insert for the single-substitution library was generated by mixing all 11 site-specific codon randomization inserts at equal molar ratio.

Insert for WSN double-substitution library

Similar to the generation of the insert for the WSN single-substitution library, an overlapping PCR strategy was employed for the generation of the insert for the WSN double-substitution library. Briefly, 5 ng insert of the single-substitution library was used as a template for a series of eight PCRs using the “universal reverse primer” as the reverse primer and 134F, 136F, 153F, 155F, 183F, 190F, 194F, and 195F as the forward primers. These eight PCRs allowed paired-codon randomization among most sites, but not the neighboring sites. Therefore, another independent 9 PCRs were performed using 5 ng WT plasmid as template. Again, the “universal reverse primer” as the reverse primer and the forward primers are shown as follows:

• 134/136F: 5′-TTC ATG GCC CAA CCA CAC ATT CAA TNN KGT TNN KGT CTC ATG CTC CCA TAG GGG AAA AA-3′

• 153/155F: 5′-AAG CAG TTT TTA CAG AAA TTT GCT TNN KCT CNN KAA AAA GGG GGA TTC ATA CCC AAA GC-3′

• 183/190F: 5′-AGA AGT CCT TGT ACT ATG GGG TGT ANN KCA TCC GTC TAG CAG TGA CNN KCA GCA GAG TCT CTA TAG TAA TGG AA-3′

• 190/194F: 5′-GGG TGT TCA TCA CCC GTC TAG CAG TGA CNN KCA GCA GAG CNN KTA CAG TAA TGG AAA TGC TTA TGT CT-3′

• 190/195F: 5′-GGG TGT TCA TCA CCC GTC TAG CAG TGA CNN KCA GCA GAG TCT ANN KAG CAA TGG AAA TGC TTA TGT CTC TG-3′

• 194/195F: 5′-CCC GTC TAG CAG TGA TGA GCA ACA GAG CNN KNN KAG CAA TGG AAA TGC TTA TGT CTC TG-3′

• 225/226F: 5′-GGA AAT AGC TGC AAG GCC CAA AGT AAG GNN KNN KCA CGG GAG GAT GAA CTA TTA CTG GA-3′

• 225/228F: 5′-GGA AAT AGC TGC AAG GCC CAA AGT AAG GNN KCA ACA CNN KAG AAT GAA CTA TTA CTG GAC CTT GC-3′

• 226/228F: 5′-AAT AGC TGC AAG GCC CAA AGT AAG AGA CNN KCA CNN KAG AAT GAA CTA TTA CTG GAC CTT GC-3′

Similar to the production of the insert from the single-substitution library, a subsequent overlapping PCR was performed by mixing with at an equal molar ratio with the corresponding Frag2; the “universal forward primer” and the “universal reverse primer” were used as primers. Mixing all the resultant products from the overlapping PCR at an appropriate ratio such that individual pairs of randomized codon were at equal molar ratio produced the insert for the double-substitution library.

Insert for WSN triple-substitution library

The insert for the WSN triple-substitution library was also generated through an overlapping PCR strategy. The first PCR used 5 ng WT plasmid as the template and the “universal reverse primer” as the reverse primer. The sequence of the forward primer was as follow:

• 225/226/228F: 5′-GGA AAT AGC TGC AAG GCC CAA AGT AAG GNN KNN KCA CNN KAG AAT GAA CTA TTA CTG GAC CTT GC-3′

To perform the overlapping PCR, the product of the first PCR was mixed at an equal molar ratio with 225-Frag2, and the “universal forward primer” and the “universal reverse primer” were used as primers. The resultant product from the overlapping PCR was the insert for the triple-substitution library.

Insert for HK68 triple-substitution library

The insert for HK68 triple-substitution library was created by PCR using 5 ng of chimeric HK68 WT plasmid as template and primers: 5′-CGT ACG TCT CAG GGT AAG GNN KNN KTC TNN KAG AAT AAG CAT CTA TTG GAC AAT-3′ and 5′-CGT ACG TCT CAT ACG TTT TGA AAG GGC TTG TCA-3′.

WSN vector generation

The vector for WSN mutant libraries was created by PCR using the WT plasmid as template and primers: 5′-ATG ATC CGT CTC GAA GCA GTG AGT CGC ATT CTG GAT TTC CCA AGA GCC ATC C-3′ and 5′-GTG TAA CGT CTC GTG TCA AAC ACC CCA GGG AGC TAT AAA CAG CAA TCT CCC T-3′.

Chimeric HK68 vector generation

The vector for chimeric HK68 triple-substitution library was created by PCR using chimeric HK68 WT plasmid as template and primers: 5′-CGT ACG TCT CAC GTA AAC AAG ATC ACA TAT GGA GCA-3′ and 5′-CGT ACG TCT CAA CCC AGG GTC TGG ACC CGA TAT TCG-3′.

Restriction digestion, ligation and transformation

Both the vector and inserts were digested with BsmBI (New England Biolabs). Ligation was performed for each mutant library using T4 DNA ligase (New England Biolabs). The ligated products were transformed into MegaX DH10B T1R Electrocomp cells (Thermo Fisher Scientific). At least one million colonies were collected for each mutant library. Plasmid mutant libraries were purified from the bacteria colonies using Maxiprep Plasmid Purification (Clontech Laboratories).

Construction of individual mutants

Individual mutants for validation experiments were constructed using the QuikChange XL Mutagenesis kit (Stratagene) according to the manufacturer’s instructions.

Transfections, infections and titering

Virus rescue experiments for WSN (WT, mutant, or mutant libraries) were based on the eight-plasmid reverse genetic system (Neumann et al., 1999), with the HA segment from WT, mutant or mutant libraries, along with the other seven WT segments. For virus rescue in the deep mutational scanning experiments, one T225 flask was used for the single-substitution library, five T225 flanks for the double-substitution library, and three T225 flasks for the triple-substitution library. For validation experiments, virus rescue was performed in 6-well plates. For passaging of each virus mutant library, a monolayer A549 cells (for WSN) or MDCK-SIAT1 cells (for HK68) in a T225 cm² plate were infected with an MOI (multiplicity of infection) of 0.05. At 2 hours post-infection, infected cells were washed three times with PBS followed by the addition of fresh medium. Virus was harvested at 24 hours post-infection.

Sequencing library preparation

Viral RNA was extracted using QIAamp Viral RNA Mini Kit (Qiagen). The extracted RNA was then reverse transcribed to cDNA using Superscript III reverse transcriptase (Thermo Fisher Scientific). The plasmid mutant libraries or the cDNA from the post-infection viral mutant libraries were amplified by PCR. For the WSN single-substitution library and double-substitution library, the primer sequences were: 5′-CAC TCT TTC CCT ACA CGA CGC TCT TCC GAT CTC CCA AGG AAA GTT CAT GGC CCA-3′ and 5′-GAC TGG AGT TCA GAC GTG TGC TCT TCC GAT CTG TTC TAG CAA GGT CCA GTA ATA-3′. For WSN triple-substitution library, the primer sequences were: 5′-CAC TCT TTC CCT ACA CGA CGC TCT TCC GAT CTA GGG AAA GAA GTC CTT GTA CTA-3′ and 5′-GAC TGG AGT TCA GAC GTG TGC TCT TCC GAT CTG CCG GAC TCA AAC CCT CTA CTC-3′. For the HK68 triple-substitution library, the primer sequences were: 5′-CAC TCT TTC CCT ACA CGA CGC TCT TCC GAT CTA GCC AGC AAA CTA TAA TCC CGA-3′ and 5′-GAC TGG AGT TCA GAC GTG TGC TCT TCC GAT CTT ACC AGT ACG TCA CCC GGC TTA-3′. For the HK68 NA segment, the primer sequences were: 5′-CAC TCT TTC CCT ACA CGA CGC TCT TCC GAT CTA CTC GGG CAG GGG ACC ACA CTA GAC-3′ and 5′-GAC TGG AGT TCA GAC GTG TGC TCT TCC GAT CTA ATG AAA TGG AAC ACC CAA CTC ATT-3′. The primer sequences of the first PCR contained part of the adapter sequence required for Illumina sequencing. A second PCR was performed to add the rest of the adapter sequence and index to the amplicon using primers: 5′-AAT GAT ACG GCG ACC ACC GAG ATC TAC ACT CTT TCC CTA CAC GAC GCT-3′ and 5′-CAA GCA GAA GAC GGC ATA CGA GAT XXX XXX GTG ACT GGA GTT CAG ACG TGT GCT-3′. Positions annotated by an “X” represented the nucleotides for the index sequence:

• WSN single-substitution plasmid mutant library: 5′-CGT GAT-3′

• WSN single-substitution post-infection mutant library: 5′-TGG TCA-3′

• WSN double-substitution plasmid mutant library: 5′-CAC TGT-3′

• WSN double-substitution post-infection mutant library: 5′-ACA TCG-3′

• WSN triple-substitution plasmid mutant library: 5′-GCC TAA-3′

• WSN triple-substitution post-infection mutant library: 5′-ATT GCC-3′

• HK68 triple-substitution plasmid mutant library: 5′-GAT CTG-3′

• HK68 triple-substitution post-infection mutant library: 5′-TCA AGT-3′

• HK68 NA plasmid: 5′-AAG CTA-3′

• HK68 NA post-infection: 5′-GTA GCC-3′

The first PCR that amplified the cDNA was performed using KOD DNA polymerase (EMD Millipore) with 1.5 mM MgSO₄, 0.2 mM of each dNTP (dATP, dCTP, dGTP, and dTTP), and 0.6 uM of forward and reverse primer. The thermocycler was set as follows: 2 minutes at 95°C, then 25 three-step cycles of 20 seconds at 95°C for denaturation, 15 seconds at 56°C for primer annealing, and 20 second s at 68°C for extension, and a 1 minutes final extension at 68°C. In the second PCR, the conditions were the same except the cycle number was 9. All PCR products were purified by gel extraction using PCR Clean-Up and Gel Extraction Kit (Clontech Laboratories).

The final PCR products were mixed and submitted for next-generation sequencing. One lane of Illumina MiSeq PE300 was used for WSN libraries and one lane of Illumina MiSeq PE75 was used for HK68 libraries.

Sequencing data analysis

For the WSN single-substitution mutant library and double-substitution mutant library, the positions of randomized codon were first identified by the internal barcode. A paired-end read was filtered and removed if the corresponding forward and reverse reads did not match at the randomized codon. Each mutation was called by comparing individual paired-end reads to the WT reference sequence. Sequencing data for each library were processed independently. A relative fitness index (RF index) was computed for individual mutants as previously described (Wu et al., 2015). Briefly, for a mutant i in mutant library n of sample t (in this study, n could be WSN single-substitution library, WSN double-substitution library, WSN triple-substitution library, or HK68 triple-substitution library, and t could be input plasmid library or post-selection library):

Occurrence frequency_i,n,t = Read count_i,n,t/Coverage_n,t, where Read count_i,n,t represented the number of read in mutant library n of sample t that carried mutation i and coverage_n represented the sequencing coverage of the mutant library n of sample t.

Similarly, Occurrence frequency_WT,n,t = Read count_WT,n,t/Coverage_n,t, where Read count_WT,n represented the number of read that matches with the WT sequence in mutant library n of sample t and coverage_n,t represented the sequencing coverage of the mutant library n of sample t.

Subsequently, Relative frequency_i,n,t = (Occurrence frequency_i,n,t)/(Occurrence frequency_WT,n,t), and RF index = (Relative frequency_{i,n,post-selection})/(Relative frequency_i,n,plasmid)

For each WSN mutant library, a given mutant would be discarded if the number of reads in the WSN plasmid mutant library were <5 (i.e. <0.001% to <0.006% input frequency, depending on mutant library). For mutants that were present in more than two WSN mutant libraries (e.g. a single-substitution mutant should be present in at least the single-substitution library and the double-substitution library), the reported RF index was computed by averaging the RF index across different WSN mutant libraries. For deep mutational scanning of HK68, a given mutant would be discarded if the number of reads in the HK68 plasmid mutant library were <180 (i.e. <0.007% input frequency). Of note, variant ENA in HK68 mutant library did not satisfy the cutoff, but was included in our validation experiment as it was one of the variants that could commonly be tolerated by both WSN and HK68. The RF index for each WSN variant and HK68 variant is listed in Dataset S2.

Hemagglutination assay

Turkey red blood cells (Lampire Biological Laboratories) were washed twice with PBS, and incubated in PBS containing 5 uM CellTracker Orange CMTMR Dye (Thermo Fisher Scientific) at 37 °C for 30 min. Turkey red blood cells were then washed again and resuspended in PBS. To each well of a round-bottomed 96-well plate, 100 μL of 0.5% turkey red blood cells (Lampire Biological Laboratories) were added and the virus sample were serially two-fold diluted. Turkey red blood cells were allowed to settle to the bottom of the well at room temperature for 2 hours, and plates were imaged using IN Cell Analyzer 6000 (GE Healthcare). The fluorescence of dots in the center of round-bottomed plates was quantified. Sigmoidal curve fitting was performed in GraphPad Prism (GraphPad Software). Dilution factor at which the signal was half of the saturation value was taken as the HA titer.

IgG expression and purification

The C05 (Ekiert et al., 2012) and S139/1 (Yoshida et al., 2009) heavy chains and light chains were cloned into pFUSE-CHIg-hG1 and pFUSE2-CLIg-hK respectively. The plasmids were co-transfected into Expi293F cells at 2:1 ratio (light to heavy) using lipofactamine 2000 (Thermo Fisher Scientific) according to manufacturer’s instructions. The supernatant was collected at 72 hours post-transfection. Full-length IgG proteins were purified from the supernatant using protein G column on AKTAexpress (GE Healthcare).

IgG neutralization assay

Around 3,000 WT or mutant viruses were incubated with C05 IgG or S139/1 IgG at the indicated concentration for 1 hour. The virus-IgG mixture was added to A549 or MDCK-SIAT1 cells in a 96-well plate. At 3-day post-infection, cell viability was measured using CellTiter-Glo Luminescent Cell Viability Assay (Promega) according to manufacturer’s instructions. Sigmoidal curve fitting was performed in GraphPad Prism (GraphPad Software).

HA expression and purification

WSN and HK68 HAs were prepared for binding experiments as previously described (Ekiert et al., 2011). Briefly, the ectodomain of HA, which corresponded to 11–329 (HA1) and 1–176 (HA2) based on H3 numbering was fused with an N-terminal gp67 signal peptide and a C-terminal tag consisting of (from N-terminal to C-terminal) BirA biotinylation site, thrombin cleavage site, foldon trimerization domain, and a His₆ tag and cloned into a customized baculovirus transfer vector (Ekiert et al., 2011). Recombinant bacmid DNA was generated using the Bac-to-Bac system (Thermo Fisher Scientific). Baculovirus was generated by transfecting purified bacmid DNA into Sf9 cells using FuGene HD (Promega). WSN HA was expressed by infecting suspension cultures of High Five cells with baculovirus at an MOI of 5 to 10 and incubating at 28 °C shaking at 110 rpm for 72 hours. The supernatant was concentrated. HA0 was purified by Ni-NTA and buffer exchanged into 20 mM Tris-HCl pH 8 and 150 mM NaCl. For binding experiments, HA0 was biotinylated as described (Ekiert et al., 2012) and purified by size exclusion chromatography on a Hiload 16/90 Superdex 200 column (GE Healthcare) in 20 mM Tris pH 8.0, 150 mM NaCl, and 0.02% NaN₃. For crystallization, the HA0 was treated with trypsin (New England Biolabs) to remove the C-terminal tag (BirA biotinylation site, thrombin cleavage site, trimerization domain, and the His₆ tag) and to produce the cleaved mature HA (HA1/HA2). The trypsin-digested HA was then purified by size exclusion chromatography on a Hiload 16/90 Superdex 200 column (GE Healthcare) in 20 mM Tris pH 8.0, 150 mM NaCl, and 0.02% NaN₃ and concentrated to ~9 mg/mL in 10 mM Tris pH 8.0, 50 mM NaCl, and 0.02% NaN₃.

Concentration of viruses

WSN viruses were grown in MDCK-SIAT1 cells in 50 mL of OPTI-MEM medium supplemented with 0.8 μg/mL TPCK-trypsin. Supernatants were collected after 3-day post-infection and centrifuged at 2000 rpm for 15 minutes at 4°C. They were then filtered using a 0.45 uM vacuum filter. Following filtration, the supernatants were aliquoted to polycarbonate ultracentrifuge tubes. The viruses were spun down using an ultracentrifuge at 18,000 rpm for 3 hours at 4°C. The supernatants were then aspirated and the pellets were air-dried for 10–20 minutes at room temperature. The pellets were then resuspended in 500 uL of 1x PBS buffer.

IgG and glycan binding assay

Binding assay was performed by biolayer interferometry (BLI) using an Octet Red instrument (ForteBio). HA0 with the C-terminal tag described above were used for these measurements. HA0 at ~10–50ug/mL (for IgG binding assay), or biotinylated 3′SLNLN or 6′-SLNLN at 3 ug/mL (for glycan binding assay) in 1x kinetics buffer (1x PBS, pH 7.4, 0.01% BSA and 0.002% Tween 20) was loaded onto Ni-NTA biosensors (for IgG binding assay) or streptavidin biosensors (for glycan binding assay) and incubated with 50 nM IgG (for IgG binding assay), indicated concentration of HA0 (for glycan binding assay), or with concentrated virus (for glycan binding assay). For binding assay of 3′-SLNLN-PAA or 6′-SLNLN-PAA against HK68 HA0, HA0 was immobilized onto Ni-NTA biosensors. Non-specific binding to the sensor was measured using a reference sensor and was subtracted from the sample signal. In the glycan binding assay, K_d was determined by the 2:1 heterogeneous ligand model as described previously (Xu et al., 2013).

Crystallization and structural determination of HA ectodomain for HK68 mutants

Initial crystal screening was carried out using our high-throughput, robotic CrystalMation system (Rigaku) at TSRI. The initial crystal screening was based on sitting drop vapor diffusion method with 35 uL reservoir solution and each drop consisting 0.1 uL protein + 0.1 uL precipitant. For HK68 variants G225M/L226T/S228A (MTA) and G225Q/L226A (QAS), after optimization based on the initial hit, crystals that routinely diffracted to <3 Å were obtained using the sitting drop vapor diffusion method with 500 μL reservoir solution containing 0.1 M sodium cacodylate pH 6.5, 5% PEG 8000, and 38% 2-methyl-2,4-pentanediol. Drops consisting 0.8 μL protein + 0.8 μL precipitant were set up at 20 °C and crystals appeared within a week. To generate HA-receptor complexes, crystals were soaked in reservoir solution supplemented with 20 mM of 3′-sialyl-N-acetyllactosamine (3′-SLN) or 6′-SLN for 30 mins to 1 hour. For HK68 variant G225L/L226S (LSS), the crystal for the apo structure was obtained from the initial screen with reservoir solution containing 0.2 M potassium nitrate and 20% PEG 3350. However, the condition was not reproducible. Subsequently, diffraction-quality crystals that were used for generating HA-receptor complexes were obtained using the same condition as that of MTA and QAS.

Diffraction data were collected at the APS GM/CA-CAT 23ID-D. The data were indexed, and integrated and scaled using HKL2000 (HKL Research) (Otwinowski and Minor, 1997). The structure was solved by molecular replacement using Phaser (McCoy et al., 2007) with PDB: 4FNK (Ekiert et al., 2012) as the molecular replacement model, modeled using Coot (Emsley et al., 2010), and refined using Refmac5 (Murshudov et al., 2011). Ramachandran statistics were calculated using MolProbity (Chen et al., 2010).

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analyses were performed using the R software package. Pearson correlation coefficients were computed for the data presented in Figure 2A–B, Figure 4B–C, and Figure 5D, and are reported in the figure. P-values reported in Figure 5A–B were computed by Fisher’s exact test. P-values reported in Figure S1B and Figure S3A were computed by Wilcoxon signed-rank test. P-values of < 0.05 were defined as statistical significant. The R² for the model fitting and the standard error of mean (SEM) for the K_d value that were reported in Figure S4B–C, Table S1, Table S3, and Table S5 were computed by Octet Data Analysis software version 9.0 (ForteBio). The neutralizing data of S139/1 IgG in Figure 4F are shown as mean ± SD from three replicates.

DATA AND SOFTWARE AVAILABILITY

Raw sequencing data have been submitted to the NIH Short Read Archive under accession number: BioProject PRJNA353496. Custom scripts for deep mutational scanning analysis have been deposited to https://github.com/wchnicholas/HARBS. The X-ray coordinates and structure factors have been deposited in the RCSB Protein Data Bank under accession codes 5VTX, 5VTU, 5VTZ, 5VTV, 5VTQ, 5VTY, 5VTW, 5VTR, and 5VU4.

Supplementary Material

1

Dataset S1. Experimental validation data.

2

Dataset S2. RF index.

3

Table S1, related to Figure 3. Dissociation constants of two bnAbs with WSN recombinant HA proteins.

Table S2, related to Figure 6. X-ray data collection and refinement statistics for HK68 variants.

Table S3, related to Figure 7. Dissociation constants of HAs with 3′-SLNLN and 6′-SLNLN.

Table S4, related to Figure 7. Off-rates of WSN recombinant HA or virus with 3′-SLNLN or 6′-SLNLN.

Table S5, related to Figure 7. Dissociation constants of HAs with 3′-SLNLN-PAA and 6′-SLNLN-PAA.

Figure S1, related to Figure 1. Analysis of WSN HA RBS deep mutational scanning data.

Figure S2, related to Figure 2. Experimental validation and examples of epistasis.

Figure S3, related to Figure 4. Analysis of HK68 HA RBS deep mutational scanning data.

Figure S4, related to Figure 4. Examples of epistatic interaction in the HK68 HA RBS and an S139/1 escape mutant.

Figure S5, related to Figure 5. Pairwise reciprocal sign epistatic interaction is enriched in specific position pairs.

Figure S6, related to Figure 6. Glycan binding properties of HA mutants.

Figure S7, related to Figure S7. Structural comparison of the apo, 3′-SLN-bound, or 6′-SLN-bound forms of HK68 variants.

Highlights.

• Large-scale mutational analysis of influenza hemagglutinin receptor-binding site (RBS)

• Many new replication-competent RBS mutants were identified

• Some previously unobserved RBS mutants escape from a broadly neutralizing antibody

• Epistasis is common in the 220-loop of the hemagglutinin receptor-binding site