The highly conserved influenza virus hemagglutinin (HA) stem domain is a major target for broadly neutralizing antibodies (bnAbs). However, despite being discovered more than a decade ago, the IGHV1-69-encoded CR9114 remains the only HA stem bnAb that cross-reacts with both influenza A and B viruses. To investigate the constraints on the breadth evolution of CR9114, this study performs four deep mutational scanning experiments to compare the binding affinity landscapes of the germline and somatic CR9114 against H1 HA, H3 HA, and influenza B virus HA. Many mutations that minimally affect or even improve the H1 HA binding are detrimental for binding to H3 HA and influenza B virus HA. We further reveal the prevalence of epistasis in IGHV1-69 HA stem bnAbs. Overall, our findings provide a mechanistic explanation for the scarcity of HA stem bnAbs with cross-reactivity against both influenza A and B viruses, and have important implications for developing broadly protective influenza vaccines.

Deep mutational scanning of the germline and somatic CR9114

To probe the evolutionary landscape of CR9114, we aimed to determine the binding affinity of all possible single amino acid mutations of the germline and somatic CR9114 against different HAs (Fig 1). Briefly, site-saturation mutagenesis was performed on the heavy chain variable domain of the germline and somatic CR9114. We did not include any mutations in the light chain, since it is not involved in binding [15]. Next-generation sequencing indicated that the frequencies of the mutations were fairly even (S1 Fig). The mutant libraries were then displayed on the yeast surface and selected for binding against different HAs. Since germline CR9114 does not cross-react with group 2 HAs and BHA [16], the mutant library of the germline CR9114 was selected against a recombinant HA stem construct that was designed based on H1N1 A/Brisbane/59/2007 HA (hereinafter referred to as H1 stem) [18]. On the other hand, the mutant library of somatic CR9114 was selected against H1 stem, H3N2 A/Singapore/INFIMH-16-0019/2016 (H3/Sing16) HA, and influenza B/Lee/1940 (B/Lee40) HA. The apparent binding affinity (K_D,app) of each mutation was then quantified by Tite-Seq [17], which combined fluorescence-activated cell sorting (FACS) and next-generation sequencing (Fig 2 and S1 Data). For each mutation, we used −Δlog₁₀ K_D,app, to quantify its effect on binding affinity. The −Δlog₁₀ K_D,app value represented the difference between the Δlog₁₀ K_D,app of a given mutation and that of the corresponding wild type (i.e., germline or somatic CR9114). A lower −Δlog₁₀ K_D,app value indicated a more detrimental effect on binding affinity. Pearson correlation coefficients of 0.94 to 0.99 were observed between the −Δlog₁₀ K_D,app values from two replicates of Tite-Seq experiments, demonstrating the high data quality (S2A and S2B Fig). As a control, we also measured the expression level of each mutation by FACS and next-generation sequencing. Our result showed that most mutations have high expression level, which did not correlate with binding affinity (S2C Fig). Moreover, the binding affinity and expression level of nonsense mutations were distinct from those of silent mutations (S2C Fig), further validating our data.

A previous study by Phillips and colleagues [16] has performed similar Tite-Seq experiments using a mutant library of CR9114 containing all possible combinations of 16 somatic hypermutations against HAs from H1N1 A/New Caledonia/20/1999, H3N2 A/Wisconsin/67/2005, and B/Malaysia/2506/2004 HA. Because the combinatorial mutant library from Phillips and colleagues [16] contained single amino acid mutants of the germline and somatic CR9114, we were able to compare a small subset of their data to ours. A Pearson correlation coefficient of 0.66 was observed between their −Δlog₁₀ K_D,app values and ours (S3A Fig). This correlation coefficient was lower than that between our Tite-Seq experimental replicates, likely due to the different HAs being used for H1 subtype, H3 subtype and influenza B virus. The major difference is that H1 stem was used in our present study, whereas the entire H1 HA ectodomain was used by Phillips and colleagues [16]. Consistently, when we compared only their −Δlog₁₀ K_D,app values against H1 HA and ours against H1 stem, the Pearson correlation coefficient decreased to 0.32. Additionally, CR9114 in fragment antigen-binding (Fab) format was used in our present study, whereas single-chain fragment variable (scFv) format was used by Phillips and colleagues [16]. This difference may also affect the impact of a given mutation on binding affinity.

Mutational tolerance of CR9114 correlates with binding affinity

A previous study demonstrated that the binding affinity of somatic CR9114 to H1 HA was stronger than its affinity to H3 HA, which in turn was stronger than its affinity to BHA [16]. Consistently, our biolayer interferometry (BLI) experiments showed that somatic CR9114 Fab bound to H1 stem, H3/Sing16 HA, and B/Lee40 HA with K_D values of <0.1 nM, 36 nM, and 110 nM, respectively (S4 and S5 Figs). This order of binding affinity of somatic CR9114 against different HAs appeared to correlate with the mutational tolerance, which decreased from H1 stem (median −Δlog₁₀ K_D,app = −0.22) to H3/Sing16 HA (median −Δlog₁₀ K_D,app = −0.40) to B/Lee40 HA (median −Δlog₁₀ K_D,app = −0.82, Figs 2 and S6). Similarly, germline CR9114 had a weaker binding affinity (K_D = 185 nM) and lower mutational tolerance (median −Δlog₁₀ K_D,app = −0.85) than somatic CR9114 against H1 stem (Figs 2, S4, and S6). It is well established that proteins with higher thermostability have higher evolvability [19]. Nevertheless, despite the tendency of affinity maturation to decrease antibody thermostability [20], the thermostabilities of germline CR9114 (T_m = 77.5 °C) and somatic CR9114 (T_m = 78.5 °C) were similar (S7 Fig). This observation showed that the difference in mutational tolerance between germline and somatic CR9114 was not due to their thermostability. While it is possible that this difference was due to the stronger binding affinity of somatic CR9114, it is also possible that the mutational stability effect, which we did not measure, was generally less deleterious in somatic CR9114.

Prevalence of epistasis and pleiotropy in the breadth evolution of CR9114

The binding affinity landscapes of the germline and somatic CR9114 against different HAs allowed us to analyze the involvement of epistasis and pleiotropy in the breadth evolution of CR9114. Here, epistasis refers to how a given mutation has different effects on binding affinity between germline and somatic CR9114, whereas pleiotropy describes how a given mutation affects the binding affinity against different HAs. We first observed that mutations with minimal effects on the binding affinity of germline CR9114 to the H1 stem rarely affected the binding affinity of somatic CR9114 to H1 stem (Figs 3A and S8A), but were frequently detrimental to the binding affinity of somatic CR9114 to H3/Sing16 HA (Figs 3B and S8B), and even more often deleterious to that of B/Lee40 HA (Figs 3C and S8C). These increasingly detrimental mutations throughout the screens mostly resided within the complementarity-determining regions (CDRs), while the effects of mutations in the framework regions on binding affinity correlated well between germline and somatic CR9114 (Figs 3A–3C and S8A–S8C). Our findings indicated that affinity maturation of germline CR9114 against H1 stem could impose a genetic barrier for evolving cross-reactivity to H3 HA and BHA due to epistasis and pleiotropy.

By comparing the binding affinity landscapes of somatic CR9114 against different HAs, we observed a pattern of nested constraint, similar to that described by a previous study on 16 somatic hypermutations of CR9114 [16]. For example, while somatic CR9114 mutations that were detrimental for H1 stem binding were also detrimental for H3/Sing16 HA and B/Lee40 HA binding, many somatic CR9114 mutations that impaired H3/Sing16 HA and B/Lee40 HA binding had minimal effects on H1 stem binding (Figs 3D, 3E, S8D, and S8E). Additionally, while somatic CR9114 mutations that were detrimental for H3/Sing16 HA binding were also detrimental for binding to B/Lee40 HA, many somatic CR9114 mutations that impaired B/Lee40 HA binding had minimal effects on H3/Sing16 HA binding (Figs 3F and S8F). Notably, most mutations that exhibited differential effects on binding against different HAs were in the CDRs (Figs 3D–3F and S8D–S8F). Together, our results suggested that further improving the affinity of somatic CR9114 against H1 HA could result in a loss of cross-reactive breadth due to pleiotropy.

Epistasis and pleiotropy involve non-paratope mutations

To experimentally validate mutations that exhibited differential effects on the binding affinity of the germline and somatic CR9114 against different HAs, we recombinantly expressed and purified mutants of interest, then measured their binding affinity using BLI (S4 and S5 Figs). The −Δlog₁₀ K_D,app values from Tite-Seq had a Pearson correlation coefficient of 0.87 with the −Δlog₁₀ K_D values from BLI (Fig 4A). Mutations in this validation experiment included V_H S24F and V_H S24L, which abolished the binding of somatic CR9114 to B/Lee40 HA, but not to H1 stem and H3/Sing16 HA (Figs 4C, S4, and S5). Our BLI results also showed that V_H I52G, V_H I52V, and V_H K73W had minimal effects on the binding of germline CR9114 to the H1 stem, whereas V_H S52G, V_H S52V, and V_H I73W slightly weakened the binding of somatic CR9114 to H1 stem while completely abolishing its binding to H3/Sing16 HA and B/Lee40 HA (Figs 4C, S4, and S5). The binding affinity of germline and somatic CR9114 mutants against H1 stem also showed strong correlation (Pearson correlation coefficient of 0.85) with their neutralization activity against H1N1 A/Brisbane/59/2007 virus (Fig 4B). Our validation experiment substantiated that epistasis and pleiotropy could constrain the breadth evolution of CR9114.

While V_H residue 73 is part of the CR9114 paratope (i.e., antibody region that contacts the epitope), V_H residues 24 and 52 are not (Fig 5A). In the somatic CR9114, V_H S24 H-bonds with the backbone carbonyl of V_H G27 and the side chain of V_H S29 to stabilize the conformation of CDR H1 (Fig 5B). Mutations V_H S24F and V_H S24L would abolish these H-bonds and increase the conformational flexibility of CDR H1. Together with our BLI results (Figs 4C, S4, and S5), this observation suggested that the conformational stability of CDR H1 may be required for somatic CR9114 to bind to BHA, but not to H1 HA and H3 HA. As for V_H residue 52, the germline CR9114 had an Ile, whereas the somatic CR9114 had a Ser. In the somatic CR9114, V_H S52 H-bonds with the backbone carbonyl of V_H Y98 to stabilize the conformation of both CDR H2 and CDR H3 (Fig 5C). However, this H-bond between CDR H2 and CDR H3 would be absent in the germline CR9114 due to its hydrophobic V_H I52. As a result, while mutating V_H residue 52 in the somatic CR9114 to Gly or Val would abolish an intramolecular H-bond, the same would not occur in the germline CR9114. This observation could explain the differential effects of V_H I/S52G and V_H I/S52V on the binding affinity of somatic and germline CR9114 (Figs 4C, S4, and S5).

A paratope mutation constrains the breadth evolution of CR9114

Our BLI results showed that mutating V_H residue 73 to Trp improved binding of germline CR9114 H1 stem, but abolished the binding of somatic CR9114 to H3/Sing16 HA and B/Lee40 HA (Figs 4C, S4, and S5). V_H residue 73 is part of the CR9114 paratope. In somatic CR9114, V_H I73 is adjacent to V_H F74 (Fig 5D). Our structural modeling indicated that mutation V_H I73W would introduce a steric clash with V_H F74 (Fig 5E), which could explain its detrimental effect on the binding of somatic CR9114 to H3/Sing16 HA and B/Lee40 HA. By contrast, this detrimental steric clash was not observed in the germline CR9114, which has the small amino acid Ser at V_H residue 74 (Fig 2). However, somatic CR9114 with double mutations V_H I73W/F74S was still unable to bind to H3/Sing16 HA or B/Lee40 HA (Figs 4C and S5). Therefore, V_H W73 could impose strong constraints for CR9114 to evolve breadth, which may not be easily overcome by a secondary mutation.

Epistasis is prevalent in IGHV1-69 HA stem antibodies

CR9114 is encoded by IGHV1-69, which is a recurring sequencing feature of HA stem antibodies [21–23]. Our previous study has curated over 5,000 influenza virus HA antibodies from the literature, of which 160 are IGHV1-69 HA stem antibodies [24]. Here, we analyzed the somatic hypermutations present in these 160 IGHV1-69 HA stem antibodies in terms of their effects on the binding affinity of germline CR9114. Most of the IGHV1-69 HA stem antibodies contained at least one somatic hypermutation with a −Δlog₁₀ K_D,app value of less than −1 (Fig 6), which was a stringent cutoff for deleterious mutations according to our validation experiment (Fig 4A). These observations suggested that epistasis is prevalent among the affinity maturation trajectories of IGHV1-69 HA stem antibodies. This epistasis could be due to a given mutation having different effects on the binding affinity of the germline precursors of different IGHV1-69 HA stem antibodies. Consistently, different binding modes have been reported for IGHV1-69 HA stem antibodies [22,25]. Another possibility is that a given mutation may be detrimental to the germline precursors of all IGHV1-69 HA stem antibodies, but may become neutral or beneficial in the presence of other somatic hypermutations, as observed in CR9114 [16].

To further examine the presence of epistasis in another IGHV1-69 HA stem antibody, we focused on the somatic hypermutation V_H S35N in 310-18G10, which was previously isolated from an H5N1 vaccinee [26]. V_H S35N is highly deleterious for both germline and somatic CR9114 (Fig 2), yet it is present in 8% of IGHV1-69 HA stem antibodies (S9 Fig), including 310-18G10. The binding affinity of the inferred germline of 310-18G10 (hereinafter referred to as germline 310-18G10) to H1 stem was weakened by 15% upon the addition of V_H S35N (S10 Fig). At the same time, reverting V_H N35 to V_H S35 weakened the binding affinity of somatic 310-18G10 to H1 stem by almost 4-fold. Therefore, V_H S35N contributed to the affinity maturation of 310-18G10, despite being deleterious in the germline 310-18G10. These results further substantiated the present of epistasis among the binding affinity landscapes of IGHV1-69 HA stem antibodies.

As a somatic hypermutation in CR9114, V_H F29S has previously been shown to have a negative first-order effect on CR9114 binding to H1N1 A/New Caledonia/1999 HA, but it has positive epistatic interactions with V_H I52S, V_H K73I, V_H S74F, V_H T75S, and V_H S76N [16]. Similarly, we observed that V_H F29S was highly deleterious for the binding of germline CR9114 to H1 stem (Fig 2). We found that V_H F29S was present in two other IGHV1-69 HA stem antibodies, namely 1009-3B05 and 1009-3E06 (Figs 6 and S11). Notably, both 1009-3B05 and 1009-3E06 were isolated from the same influenza patient and belong to the same clonotype [24,27]. While 1009-3E06 contained one of the known compensatory mutations, V_H S76N, 1009-3B05 did not contain any (S11 Fig). This observation suggested that an unknown epistatic interaction involving V_H F29S was present in 1009-3B05.

Yeast display plasmids

Yeast display plasmids for germline and somatic CR9114 Fabs were constructed based on the pCTcon2 vector, with an open reading frame encoding (from N-terminal to C-terminal): Aga2 secretion signal, somatic CR9114 Fab light chain, V5 tag, equine rhinitis B virus (ERBV-1) 2A peptide, Aga2 secretion signal, germline or somatic CR9114 Fab heavy chain, HA-tag, and Aga2p.

Construction of yeast Fab mutant libraries

Two separate mutant libraries were generated based on different versions of the CR9114 heavy chain variable domain sequence. The fully developed or somatic CR9114 sequence, and the pre-developmental, or germline CR9114 sequence, were each used as bases for the respective libraries. The heavy and light chain sequences of somatic CR9114 were obtained from Genbank JX213639 and JX213640, respectively [15]. The CR9114 germline sequence was obtained from Phillips and colleagues [16], who themselves had reconstructed the sequence through the international ImMunoGeneTics information system (IMGT) [50] and IgBLAST [51]. We chose to mimic their treatment of site S109N, and not acknowledge a mutation at that site [16]. For both libraries, the somatic CR9114 light chain sequence was used, to focus solely on heavy chain variant effects. The mutant libraries of variant Fabs were generated that differed from the somatic or germline CR9114 sequences by only one amino acid per mutant, such that every possible amino acid, as well as stop codons, was represented in every possible position of the heavy chain variable domain, excluding the final amino acid in the heavy chain variable domain, which is highly conserved.

The pCTcon2 plasmid containing either the somatic or germline CR9114 sequence in Fab form was used as a template to generate the respective mutant library inserts and the linearized vector through polymerase chain reaction (PCR), separately. The inserts for each library were then generated as follows, per library. To generate the linearized vectors, we used 5′-TGG TGT CCA CAC TTT CCC TGC TGT T-3′ and 5′-AGT GGA TTG GGG ATT GGC TTT CCG C-3′ as primers. To generate the full-length inserts, we first needed to manufacture two sets of 15 split inserts per library in parallel before merging them in paired PCRs using overlap extension PCR. The first set of split inserts included 15 separate reactions; each performed with a cassette of pooled forward primers and a universal reverse primer 5′-AGG AGT ACA AAC CGG AAG ATT GC-3′. Each forward cassette of primers was composed of eight forward primers in equal molar ratios with an identical 21 nt at the 5′ end and 15 nt at the 3′ end. NNK (N: A, C, G, T; K: G, T) sequences were present within each cassette, placed in such a way that saturation mutagenesis was achieved. As described previously, silent (or synonymous) mutations were introduced into each primer to discern between sequencing errors and target mutations through data analysis [52]. In S1 Table, the forward primers are listed for each mutant library, identified in the table by cassette number and the position of the amino acid mutated. The second set of split inserts also included 15 separate reactions, designed to match up with the first set of split inserts. This final set of 15 insert fragments was generated through PCR, with each reaction using the universal forward primer 5′-ACT GTC GCC CCA ACT GAG TGC TC-3′ and a unique reverse primer (S2 Table). The first and second sets of split inserts were joined to generate full-length cassettes through 15 overlapping PCRs, using universal forward and reverse primers to perform the amplifications with 10 ng of the respective inserts per PCR. For each library, these 15 full-length cassettes were gel purified, then mixed using equimolar amounts of each cassette to generate the final insert pool for use in the yeast transformation. PrimeSTAR Max polymerase (Takara Bio, cat no. R045B) was used for all PCRs. Gel purification was performed using the Monarch Gel Extraction Kit (New England Biolabs, cat no. T1020L). The manufacturer’s instructions were used for all samples in each case.

Yeast transformation

Transformation of yeast was performed for each mutant library by electroporation as previously described [53]. Strain EBY100 of Saccharomyces cerevisiae (ATCC) were grown overnight at 30 °C with 250 rpm agitation in YPD medium (1% w/v yeast nitrogen base, 2% w/v peptone, 2% w/v D(+)-glucose). Once the culture reached OD₆₀₀ of 3, a new culture was seeded at OD₆₀₀ of 0.3 in 100 mL of YPD media and continued to grow at 30 °C with 250 rpm agitation. Following OD₆₀₀ reaching 1.6, the culture of yeast cells was centrifuged for 3 min at 1,700g at room temperature, and the media was removed. Fifty milliliters of ice-cold water was used to wash the cell pellet twice; the pellet was then washed with 50 mL of ice-cold electroporation buffer (1 M sorbitol, 1 mM calcium chloride) before resuspension in 20 mL conditioning media (0.1 M lithium acetate, 10 mM dithiothreitol). The new mixture was then shaken at 250 rpm at 30 °C for 30 min, then centrifuged at 1,700g for 3 min at room temperature. Following wash with 50 mL ice-cold electroporation buffer, the pellet was resuspended in 1 mL of electroporation buffer and placed on ice. In parallel, 4 µg of the linearized vector was mixed with 5 µg of the full-length cassette mutant library insert, and pipetted into 400 µL of the prepared 1 mL conditioned yeast in a pre-chilled BioRad GenePulser 2 mm electrode gap cuvette. Following a 5-min incubation on ice, electroporation was performed at 2.5 kV, 25 µF, such that the time constant was within 3.7 and 4.1 ms. The electroporated cell mixture was immediately transferred into 8 mL of a 1 to 1 mixture of YPD and 1 M sorbitol, then shaken at 30 °C, 250 rpm for 1 h.

Cells were collected via centrifugation at 1,700g for 3 min at room temperature, resuspended in 0.6 mL synthetic dextrose casamino acid medium (SD-CAA: 2% w/v D-glucose, 0.67% w/v yeast nitrogen base with ammonium sulfate, 0.5% w/v casamino acids, 0.54% w/v Na₂HPO₄, 0.86% w/v NaH₂PO₄·H₂O, all dissolved in deionized water), plated onto SD-CAA plates (2% w/v D-glucose, 0.67% w/v yeast nitrogen base with ammonium sulfate, 0.5% w/v casamino acids, 0.54% w/v Na₂HPO₄, 0.86% w/v NaH₂PO₄·H₂O, 18.2% w/v sorbitol, 1.5% w/v agar, all dissolved in deionized water) and incubated at 30 °C for 48 h. Colonies were then collected in SD-CAA medium, centrifuged at 1,700g for 5 min at room temperature, and resuspended in SD-CAA medium with 15% v/v glycerol such that OD600 was 50. Glycerol stocks were stored at −80 °C until used.

Cell culture

Sf9 cells (Spodoptera frugiperda ovarian cells, female, ATCC, cat no. CRL-1711) were maintained in Sf-900 II SFM medium (Thermo Fisher Scientific, cat no. 10902088) at 37 °C. Expi293F cells (Thermo Fisher Scientific, cat no. A14527) were grown and maintained in Expi293 Expression Medium (Thermo Fisher Scientific, cat no. A1435101) at 37 °C, 8% CO₂, and 95% humidity with shaking at 125 rpm according to the manufacturer’s instructions. Madin-Darby canine kidney cells stably transfected with the cDNA of human 2,6-sialyltransferase (MDCK-SIAT1) (ATCC, cat no. CRL-3743) were grown and maintained in Dulbecco’s Modified Eagle’s Medium (Thermo Fisher Scientific, cat no. 11995073) containing 10% fetal bovine serum (Fisher Scientific, cat no. A5669701), 1× GlutaMAX supplement (Thermo Fisher Scientific, cat no. 35050079), and 100 U mL⁻¹ penicillin/streptomycin (Thermo Fisher Scientific, cat no. 15140122) at 37 °C, 5% CO₂, and 95% humidity, according to the manufacturer’s instructions.

Influenza virus

Recombinant 6:2 reassortant virus with hemagglutinin (HA) and neuraminidase (NA) segments from H1N1 A/Brisbane/59/2007 in H1N1 A/Puerto Rico/8/1934 (PR8) backbone were rescued using the eight-plasmid influenza virus reverse genetics system [54]. Briefly, DNA plasmids were transfected into a co-culture of HEK293T cells and MDCK-SIAT1 cells (ratio of 6:1) and incubated at 37 °C for 48 h. Supernatants were then injected into 8- to 10-day-old embryonated chicken eggs for viral rescue at 37 °C for 48 h. Virus was plaque-purified on MDCK-SIAT1 cells. Individual plaques were picked and injected into embryonated eggs, and viral RNAs were extracted from allantoic fluids. HA and NA segments were confirmed by Sanger sequencing.

Expression and purification of H1 stem, H3 HA, and BHA

Oligonucleotide encoding H1 stem #4900 protein [18], H3 HA (A/Singapore/INFIMH-16-0019/2016; GISAID accession num. EPI1858150), or BHA (B/Lee/1940; GISAID accession num. EPI243230) was fused with N-terminal gp67 signal peptide and a C-terminal BirA biotinylation site, thrombin cleavage site, T4-foldon trimerization domain, and a 6× His tag, and then cloned into a customized baculovirus transfer vector [55]. Recombinant bacmid DNA that carried the protein construct was generated using the Bac-to-Bac system (Thermo Fisher Scientific, cat no. 10359016) according to the manufacturer’s instructions. Baculovirus was generated by transfecting the purified bacmid DNA into adherent Sf9 cells using Cellfectin reagent (Thermo Fisher Scientific, cat no. 10362100) according to the manufacturer’s instructions. The baculovirus was further amplified by passaging in adherent Sf9 cells at a multiplicity of infection (MOI) of 1. Recombinant protein was expressed by infecting 1 L of suspension Sf9 cells at an MOI of 1. Infected Sf9 cells were incubated at 27 °C with shaking at 170 rpm for 72 h. On day 3 post-infection, Sf9 cells were pelleted by centrifugation at 4,000g for 25 min, and soluble recombinant protein was purified from the supernatant by affinity chromatography using Ni Sepharose excel resin (Cytiva, cat no. 17371202) and then size exclusion chromatography using a HiLoad 16/100 Superdex 200 prep grade column (Cytiva, cat no. 90100137) in 20 mM Tris-HCl pH 8.0, 100 mM NaCl. The purified protein was concentrated by Amicon spin filter (Millipore Sigma, cat no. UFC901008) and filtered by 0.22 mm centrifuge Tube Filters (Costar, cat no. 8160). Concentration of the protein was determined by a NanoDrop One^C spectrophotometer (Fisher Scientific). Protein was subsequently aliquoted, flash frozen in dry-ice ethanol mixture, and stored at −80 °C until use.

Biotinylation and PE-conjugation of HAs

Purified H1 stem, H3 HA, and BHA were biotinylated using the Biotin-Protein Ligase-BIRA kit according to the manufacturer’s instructions (Avidity, cat no. BirA500). After buffer exchange into phosphate-buffered saline (PBS), the biotinylated HA were conjugated to streptavidin-PE (Fisher Scientific, cat no. S866A) by incubating at room temperature for 15 min.

Fluorescence-activated cell sorting (FACS) of yeast display libraries

Hundred microliters glycerol stock of the yeast display library was recovered in 50 mL SD-CAA medium by incubating at 27 °C with shaking at 250 rpm until OD₆₀₀ reached between 1.5 and 2.0. Then 15 mL of the yeast culture was harvested and pelleted via centrifugation at 4,000g at 4 °C for 5 min. The supernatant was discarded, and sucrose/galactose/raffinose/casamino acids medium (SGR-CAA: 2% w/v galactose, 2% w/v raffinose, 0.1% w/v D-glucose, 0.67% w/v yeast nitrogen base with ammonium sulfate, 0.5% w/v casamino acids, 0.54% w/v Na₂HPO₄, 0.86% w/v NaH₂PO₄·H₂O, all dissolved in deionized water) was added to make up the volume to 50 mL. The yeast culture was then transferred to a baffled flask and incubated at 18 °C with shaking at 250 rpm. Once OD₆₀₀ reached between 1.3 and 1.6, 1 mL of yeast culture was harvested and pelleted via centrifugation at 4,000g at 4 °C for 5 min. The pellet was subsequently washed with 1 mL of 1× PBS twice. After the final wash, cells were resuspended in 1 mL of 1× PBS. For each expression sort, PE anti-HA.11 (epitope 16B12, BioLegend, cat. no. 901517) that had been buffer-exchanged into 1× PBS was added to the cells at a final concentration of 30 nM.

For the expression sorts, PE anti-HA.11 (BioLegend, cat no. 901518) was added to a subset of the washed cells at a final concentration of 30 nM. For Tite-Seq, cells were labeled with PE-conjugated H1 stem, H3 HA, or BHA at each of eight antigen concentrations. For the somatic CR9114 mutant library sorts, one-log increments of PE-conjugated probe spanning 0.01 nM–10 nM and 0.003 nM–3 nM (H1 stem); 0.1 nM–100 nM and 0.03 nM–33 nM (H3 HA); or 0.2 nM–200 nM and 0.067 nM–67 nM (BHA) were used for FACS. For the germline CR9114 mutant library sorts, one-log increments of PE-conjugated H1 stem spanning 16 nM–0.016 nM and 5 nM–0.005 nM, along with an unstained sample, were used for FACS. A negative control was set up with no probe added to the resuspended cells. For the somatic CR9114 mutant library sorts, all cells were stained in solution containing 1× PBS with 1% bovine serum albumin (BSA), while for the germline CR9114 mutant library sorts, all cells were stained in solution of 1× PBS lacking BSA. The reason for this discrepancy is that the method was improved during this project: the addition of 1% BSA was found to provide a cleaner sort with lower background signal. Samples were incubated with their respective probe concentrations for 1 h at 4 °C with rotation. Then the yeast pellet was washed twice in 1× PBS and resuspended in FACS tubes containing 2 mL 1× PBS. Cells in the selected gates were collected in 1 mL of SD-CAA containing 100 U mL⁻¹ penicillin/streptomycin (ThermoFisher Scientific, cat no. 15140122). The somatic CR9114 mutant library was sorted using a BD FACSMelody Cell Sorter and BD FACSChorus Software (BD Biosciences), while the germline CR9114 mutant library was sorted using a BD FACSAria and BD FACSDiva Software (BD Biosciences).

While we used a 1-h incubation time for probe staining, longer incubation times were also tested as part of our pilot experiments. Briefly, each yeast display library was incubated with different concentrations of PE-conjugated probes. After incubation for either 1 h, 2 h, or 24 h, samples were measured by a NovoCyte Opteon (Agilent). We saw no obvious difference in binding trends after extending the incubation time for more than 1 h (S12A Fig). Moreover, the MFI of the 1-h probe incubation was similar to that of the 24-h incubation, indicating that binding was close to equilibrium, if not already reached, after 1 h. To measure probe depletion effects, a pre-depletion sample set was prepared by incubating yeast mutant library samples with 0.1–200 nM of PE-conjugated probe for 1 h, then pelleting the yeast by centrifuging at 4,000g at 4 °C for 5 min. To prepare the post-depletion samples, the probe-containing supernatants were removed from the pre-depletion sample yeast pellets. This probe-containing ‘depleted’ supernatant was applied to an equivalent number of yeast cells for 1 h to generate the post-depletion sample set. Both pre- and post-depletion samples were analyzed by a NovoCyte Opteon (Agilent). The result showed that there was no probe depletion at all the tested concentrations (S12B Fig).

For expression sort, single cells were gated into four equally populated bins, with bin 1–4 having the increasing PE signal. For Tite-Seq, single cells were gated into four bins along the PE-A axis based on unstained and CR9114 controls, with bin 1 comprising all PE-negative cells, bin 4 comprising PE-positive cells with comparable expression or binding affinity to the germline or somatic CR9114 positive population, and bins 2 and 3 splitting the intermediate population between bin 1 and bin 4. Following sort, the sort tubes were directly placed at 30 °C with 250 rpm agitation. After 20–40 h, the cells in each tube were collected by centrifugation at 3,800g for 15 min once a large pellet was visible or the culture was cloudy. Frozen stocks were made per sorted sample by reconstituting the pellet in 15% v/v glycerol (in SD-CAA medium) and then stored at −80 °C until use.

Next-generation sequencing of FACS samples

Plasmids from the yeast cells before and after sorting were extracted using a Zymoprep Yeast Plasmid Miniprep II Kit (Zymo Research, cat no. D2004) following the manufacturer’s protocol. The region of interest in the mutant library was subsequently amplified by PCR using primers 5′-C ACT CTT TCC CTA CAC GAC GCT CTT CCG ATC TGT GTA CTT GCC GCC GCT CAA CCA −3′ and 5′-G ACT GGA GTT CAG ACG TGT GCT CTT CCG ATC TAC GGA AGG TCC CTT AGT AGA AGC-3′ for both the germline and somatic mutant libraries. Subsequently, adapters containing sequencing barcodes were appended to the amplicon using primers 5′-AAT GAT ACG GCG ACC ACC GAG ATC TAC ACX XXX XXX XAC ACT CTT TCC CTA CAC GAC GCT-3′, and 5′-CAA GCA GAA GAC GGC ATA CGA GAT XXX XXX XXG TGA CTG GAG TTC AGA CGT GTG CT-3′. Positions annotated by an ‘‘X’’ represented the nucleotides for the index sequence. All PCRs post-FACS were performed using Q5 High-Fidelity DNA polymerase (NEB, cat no. M0492S) according to the manufacturer’s instructions. PCR products were purified using PureLink PCR Purification Kit (Thermo Fisher Scientific, cat no. K310002). The final PCR products were submitted for next-generation sequencing using NovaSeq SP PE250 (Illumina).

Computing the binding score, expression score, and apparent K_D values

The sequencing data were initially obtained in FASTQ format, then analyzed using a custom Python Snakemake pipeline [56]. Briefly, Paired-End reAd mergeR (PEAR) was used to merge the forward and reverse reads [57]. The number of reads corresponding to each variant in each sample was counted. To avoid division by zero in downstream analysis, a pseudocount of 1 was added to the final count.

For each expression sort, the enrichment value of each mutation were computed as follows:

where the is the occurrence frequency of mutation in bin and is the mean fluorescence intensity of bin . The expression score for each variant was further computed from the enrichment value as follows:

where is the average enrichment value of the silent mutations and is the average enrichment value of the nonsense mutations. The final score for each variant var is the average of two biological replicates.

To compute apparent value of each variant from the Tite-Seq data, we adopted the analysis approach as previously described [58]. Briefly, to determine the mean bin of PE fluorescence for each variant at each HA concentration, a simple weighted mean calculation was applied:

where is the number of cells with mutation that fall into bin at HA concentration . This calculation computes a weighted average by assigning integer weight to the bin .

For each variant var, we estimated its sorted cell count that corresponds to bin at HA concentration as follows:

where variant read count is the read counts for variant var in bin at HA concentration , is the total read counts for bin at HA concentration , is the total number of cells in bin at HA concentration . We then determined the apparent binding affinity value for each mutation via a nonlinear least-squares regression using a standard non-cooperative Hill equation:

where free parameters a is titration response range and b is titration curve baseline. Mutations with an occurrence frequency below 0.01% in our pre-sort samples were discarded from our downstream analysis. To quantify the effects of each mutation on binding affinity, we calculate the difference between the values of the given mutation and the corresponding wild type:

For this analysis, ligand depletion effects were not considered because probe depletion was not observed in a pilot experiment (S12B Fig).

Of note, while the log₁₀ K_D,app values (not Δlog₁₀ K_D,app values) from Tite-Seq correlated well with the K_D values from the BLI experiment (Pearson correlation coefficients of 0.79), the log₁₀ K_D,app values were systematically lower than the log₁₀ K_D values for germline CR9114 mutants against H1 stem (below the diagonal line in S3B Fig). Similarly, while germline CR9114 is known to have a much weaker binding affinity than somatic CR9114 to H1 stem (S4 Fig) [16,59], the K_D,app value for germline CR9114 against H1 stem was 0.3 nM, which was similar to that for somatic CR9114 against H1 stem 0.7 nM. These observations indicated that the Tite-Seq experiment for germline CR9114 mutant library against H1 stem overestimated the binding affinity. One possible explanation for this artifact is that the PE-conjugated probe for the Tite-Seq experiment of the germline CR9114 mutant library against the H1 stem was prepared much earlier than the ones for other Tite-Seq experiments in this study, including that for the somatic CR9114 mutant library against the H1 stem. There may have been a batch effect in the preparation of the PE-conjugated probe, which introduced systematic noise into the K_D,app value measurement.

Expression and purification of antibodies

The heavy and light chains of CR9114 or 310-18G10 fragment antigen-binding (Fab) or immunoglobulin G (IgG) germline or somatic heavy chain sequence mutants were cloned into phCMV3 vector. The mutants were generated using mismatch PCR to modify the bases desired using PrimeSTAR Max polymerase (Takara Bio, catalog no. R045B). The heavy and light chain sequences of somatic 310-18G10 were obtained from GenBank KJ149431.1 and KJ149356.1, respectively [26]. The germline heavy chain sequence was inferred using IMGT [50] and IgBLAST [51]. For both somatic and germline 310-18G10, the somatic 310-18G10 light chain sequence was used. Gibson cloning was then performed using the NEBuilder HiFi DNA Assembly Master Mix (NEB, cat no. E2621), using the manufacturer’s instructions following mismatch PCR to modify the bases desired. The plasmids were co-transfected into Expi293F cells at a 2:1 (HC:LC) mass ratio using ExpiFectamine 293 Reagent (Thermo Fisher Scientific, cat no. A14525). At 5 days post-transfection, the supernatant was collected, filtered using 100.0 µm Millipore Steriflip Vacuum Tube Top Filters (Millipore Sigma, cat no. SCNY00100), and the Fab was purified using a CaptureSelect CH1-XL Affinity Matrix (Thermo Fisher Scientific, cat no. 194371201L).

Biolayer interferometry binding assay

Binding assays were performed by BLI using an Octet Red96e instrument (Sartorius) at room temperature. Briefly, biotinylated H1 stem, H3 HA, or BHA proteins at 20 µg mL⁻¹ in 1× kinetics buffer (PBS at pH 7.4, 0.01% w/v BSA, and 0.002% v/v Tween 20) were loaded onto streptavidin (SA) biosensors and incubated with the indicated concentrations of Fabs. The assay consisted of five steps, each run for at least 60 seconds: (1) baseline: 1 × kinetics buffer; (2) loading: HA proteins; (3) baseline: 1× kinetics buffer; (4) association: Fab samples; and (5) dissociation: 1× kinetics buffer. Two to three Fab concentrations at 3-fold serial dilutions were used for each assay. For estimating the K_D, a 1:1 binding model was used. In cases where 1:1 binding model did not fit well due to the contribution of non-specific binding to the response curve, a 2:1 heterogeneous ligand model was used to improve the fitting.

Differential scanning fluorimetry assay

Thermal shift assays were performed using differential scanning fluorimetry (DSF) on the CFX Opus 96 Real-Time PCR System (Bio-Rad, cat no. 12011319). SYPRO Orange Protein Gel Stain (5,000× concentrate in dimethyl sulfoxide) (Thermo Fisher Scientific, cat no. S6650). Five micrograms of each Fab was used per assay in a total volume of 25 µL with SYPRO Orange Protein Gel Stain at 1× in PBS in a Hard-Shell 96-Well PCR Plate (Bio-Rad, cat no. HSP9601). The assay was performed from 10 to 95 °C in increments of 0.5 °C for 10 s plus plate reading using the Förster Resonance Energy Transfer channel of the instrument. Output files were analyzed with R.

Microneutralization assay

MDCK-SIAT1 were seeded overnight in a 96-well plate. After reaching 100% confluency, MDCK-SIAT1 cells were washed once with 1× PBS. Minimal essential media (MEM, Thermo Fisher Scientific, cat. no 12492013) containing 25 mM 4-(2-hydroxyethyl)−1-piperazineethanesulfonic acid (HEPES, Thermo Fisher Scientific, cat. no 15630080) was then added to the cells. IgGs were serially diluted 2-fold starting from 100 μg mL⁻¹ and mixed with 100 TCID₅₀ (median tissue culture infectious dose) of viruses at equal volume and incubated at 37 °C for 1 h. Subsequently, the mixture was inoculated into cells and incubated at 37°C for another hour. Supernatants were replaced with MEM containing 25 mM HEPES and 1 g mL⁻¹ TPCK-trypsin (Thermo Fisher Scientific, cat. no 20233). After incubation, Antibodies were added back to the supernatants at the same concentrations as before the medium replacement. Each antibody condition was performed in duplicate within the same plate. Plates were incubated at 37 °C for 48 h, and virus presence was assessed by cytopathic effect to determine the minimum inhibitory concentration.

Sequence alignment

The heavy chain sequences of CR9114, 1000-3B05, and 1009-3E06 were aligned to the IGHV1-69 germline sequence by MAFFT (v7.526) [60].

S1 Fig. Occurrence frequency of each mutation before sorting.

Heatmaps showing the pre-sorting frequency of each mutation in each deep mutational scanning experiment. “H1”, “H3”, and “BHA” indicate H1 stem, H3/Sing16 HA, and B/Lee40 HA, respectively. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3004021.s001

(TIFF)

S2 Fig. Correlations between replicates and expression score comparisons during sort.

(A) Correlation of log₁₀ K_D,app values between replicates. (B) Correlation of −Δlog₁₀ K_D,app values between replicates. (C) Correlation of expression scores with −Δlog₁₀ K_D,app values. Each teal, blue, or orange dot placed at the coordinate values represents a missense, nonsense, or silent mutation, respectively. “H1”, “H3”, and “BHA” indicate H1 stem, H3/Sing16 HA, and B/Lee40 HA, respectively. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3004021.s002

(TIFF)

S3 Fig. Correlation of Tite-Seq data in this study with that of a published study and with biolayer interferometry (BLI) data.

(A) The correlation of −Δlog₁₀ K_D,app values between this study and a previously study by Phillips and colleagues [16]. Each data point represents a mutation for which data is available in both datasets. The Pearson correlation coefficient (R) is indicated. (B) The log₁₀ K_D values determined by biolayer interferometry (BLI) are plotted against the log₁₀ K_D,app values determined by Tite-Seq in the deep mutational scanning experiments. Each data point represents a mutation. Mutations with no detectable binding in BLI are assigned a log₁₀ K_D value of −5. The genetic background of each mutation (i.e., germline or somatic CR9114) and the binding target (i.e., H1 stem, H3/Sing16 HA, or B/Lee40 HA) are color-coded. Pearson correlation coefficient (R) is indicated. The data underlying this Figure can be found in S3 Data.

https://doi.org/10.1371/journal.pbio.3004021.s003

(TIFF)

S4 Fig. Binding kinetics of germline and somatic CR9114 mutants against H1 stem.

The binding kinetics of the indicated (A) germline and (B) somatic CR9114 mutants in Fab format against H1 stem were measured by biolayer interferometry. The Y-axis indicates the signal response. Blue lines represent the response curve, and red lines represent the best fit model (1:1 binding model or 2:1 heterogeneous ligand model, see Methods). Binding kinetics were measured for two to three concentrations of Fab at 3-fold dilution. The dissociation constants (K_D) and the goodness-of-fit values (R²) are shown. The data underlying this Figure can be found in S4 Data.

https://doi.org/10.1371/journal.pbio.3004021.s004

(TIFF)

S5 Fig. Binding kinetics of somatic CR9114 mutants against H3 HA and BHA.

The binding kinetics of the indicated somatic CR9114 mutants in Fab format against (A) H3 HA and (B) BHA were measured by biolayer interferometry. The Y-axis indicates the signal response. Blue lines represent the response curve, and red lines represent the best-fit model (1:1 binding model or 2:1 heterogeneous ligand model, see Methods). Binding kinetics were measured for two to three concentrations of Fab at 3-fold dilution. The dissociation constants (K_D) and the goodness-of-fit values (R²) are shown. The data underlying this Figure can be found in S4 Data.

https://doi.org/10.1371/journal.pbio.3004021.s005

(TIFF)

S6 Fig. Distribution of missense mutations effects.

The distribution of −Δlog₁₀ K_D,app values for each deep mutational scanning experiment is shown as a violin plot. “H1”, “H3”, and “BHA” indicate H1 stem, H3/Sing16 HA, and B/Lee40 HA, respectively. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3004021.s006

(TIFF)

S7 Fig. Measuring the thermal stability of germline and somatic CR9114 through a thermal shift assay.

The first differential curves for the relative fluorescence unit (RFU) with respect to temperature are shown for germline CR9114 (green) and somatic CR9114 (purple). The reported T_m is an average of three independent biological replicates. The data underlying this Figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3004021.s007

(TIFF)

S8 Fig. Correlation of mutational effects on binding affinity across different deep mutational scanning experiments for amino acid mutations that are one nucleotide different from the wild type.

(A–F) The −Δlog₁₀ K_D,app value of each mutation was compared between the deep mutational scanning experiments of (A) the germline CR9114 against H1 stem and the somatic CR9114 against H1 stem, (B) the germline CR9114 against H1 stem and the somatic CR9114 against H3 HA, (C) the germline CR9114 against H1 stem and the somatic CR9114 against BHA, (D) the somatic CR9114 against H1 stem and the somatic CR9114 against H3 HA, (E) the somatic CR9114 against H1 stem and the somatic CR9114 against BHA, (F) the somatic CR9114 against H3 HA and the somatic CR9114 against BHA. Each data point represents a mutation. Pearson correlation coefficients for the mutations in the CDRs and the DE loop (blue) as well as for those in the framework regions (FWRs, red) are indicated. Only those amino acid mutations that are one-nucleotide different from the wild type (i.e., germline or somatic CR9114) are included here. “H1”, “H3”, and “BHA” indicate H1 stem, H3/Sing16 HA, and B/Lee40 HA, respectively. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3004021.s008

(TIFF)

S9 Fig. Correlation between the frequency of somatic hypermutations observed among known IGHV1-69 HA stem antibodies and their effects on germline CR9114 binding to H1 stem.

The occurrence frequency of each somatic hypermutation among the 160 known IGHV1-69 HA stem antibodies [24] is plotted against their corresponding −Δlog₁₀ K_D,app value from the deep mutational scanning experiment of CR9114 against H1 stem. Points in red represent somatic mutations that are present in CR9114. The data underlying this Figure can be found in S3 Data.

https://doi.org/10.1371/journal.pbio.3004021.s009

(TIFF)

S10 Fig. Binding kinetics of IGHV1-69 antibody somatic and germline 310-18G10 mutants against H1 stem.

The binding kinetics of the indicated somatic and germline mutants of 310-18G10 in Fab format against H1 stem were measured by biolayer interferometry. The Y-axis indicates the signal response. Blue lines represent the response curve, and red lines represent the best fit model (1:1 binding model or 2:1 heterogeneous ligand model, see Methods). Binding kinetics were measured for three concentrations of Fab at 3-fold dilution. The dissociation constants (K_D) and the goodness-of-fit values (R²) are shown. The data underlying this Figure can be found in S4 Data.

https://doi.org/10.1371/journal.pbio.3004021.s010

(TIFF)

S12 Fig. Impacts of incubation time and probe depletion on median fluorescence intensity (MFI).

(A) MFI is plotted for each sample according to concentration of HA probe and mutant library identity (germline or somatic CR9114). The color of each point denotes the incubation time: 1 h, 2 h, or 24 h. (B) The MFI is shown for pre-depleted (pink) and depleted (black) conditions for 1 h incubations. MFI is normalized to each respective unstained mutant library background signal for each time point. The data underlying this Figure can be found in S6 Data.

https://doi.org/10.1371/journal.pbio.3004021.s012

(TIFF)