Deep mutational scanning maps the K1F tailspike β-helix as a host-range tuning knob

What the K1F tailspike deep mutational scanning actually changed

Before this study, the K1F endosialidase was known structurally: crystal structures of the catalytic domain in apo and substrate-bound states had identified three polysialic acid binding sites and catalytic residues H350, R549, and E581 [1][2]. That work established the architecture but relied on purified protein and short oligosaccharide substrates, leaving open how individual residues contribute to infection on intact bacteria [1]. The new paper closes that gap by generating 22,365 designed single-amino-acid variants across the full 3.2-kb TSP gene using an enhanced ORACLE phage engineering platform, recovering 18,866 variants (85%) after genome insertion [1]. Functional scores (Fn) were calculated from variant abundance before and after selection on E. coli K1 strain EV36, normalized to wild type, with strong replicate correlation (R² = 0.826) between independent assemblies [1]. The result is the largest functional library of a phage tailspike protein reported to date, and it maps function in the native phage context rather than in isolation [1].

The methodological advance matters because scaling ORACLE to a 3.2-kb enzymatic gene required two innovations: DNA barcoding to track variants after genome insertion, and relocating Cre recombinase under inducible genomic control to stabilize donor plasmids [1]. These changes improved recombination efficiency 10-fold, from roughly 1 in 10,000 to 1 in 1,000 phages [1]. The approach is conceptually distinct from DMS protocols developed for viral proteins such as HIV-1 Env, which rely on replication-competent virus libraries and long-read sequencing but are limited to viruses that package one genotype per virion and can be efficiently rescued in producer cells [4]. The K1F work instead engineers the phage genome directly, which is necessary when the target is a large, multifunctional enzyme rather than a single envelope protein [1][4].

Structural fragility coexists with adaptive hotspots

The functional map reveals a protein that is broadly intolerant of mutation yet contains discrete regions of flexibility. Catalytic residues H350, R549, and E581 tolerated very few substitutions, and stop codons and start-codon mutations were generally not tolerated, consistent with prior biochemical assignments [1]. Variant dropout during library construction was reproducible across replicates and occurred even when the library was built in E. coli 10G, which K1F cannot infect, arguing against host-driven selection and pointing instead to intrinsic structural constraints [1]. This pattern — a fragile global architecture with local adaptive potential — parallels observations in other DMS studies where stability-based assays show stronger predictive generalizability than activity-based assays, because activity depends on complex cellular contexts that stability screens do not capture [6]. The K1F dataset is explicitly an activity-based screen in native phage context, which may explain why some variants that appear tolerated in vitro nonetheless fail during infection [1][6].

The adaptive flexibility is not randomly distributed. Mutations at positions such as N565 and Q615 enhanced phage fitness despite lying outside canonical binding sites, suggesting interactions with other bacterial surface moieties beyond the capsule [1]. This echoes findings in other systems where mutations far from active sites can have profound functional effects through allosteric or dynamic coupling — for example, graph neural network models trained on protein dynamics outperform sequence-based approaches at predicting epistatic effects of distal mutations [9]. The K1F data provide an experimental counterpart: peripheral and stalk residues shape function even when they do not contact substrate directly [1].

The β-propeller active site and β-helix stalk as distinct functional modules

Within the β-propeller active site, residues Q330 and K585 exhibited limited mutational tolerance, which the authors interpret as accommodation of longer sialic acid chains than structural studies had captured [1]. This is a functional inference from mutational sensitivity rather than a direct structural observation, and it suggests that the active site has more substrate-binding capacity than the crystallographic snapshots revealed [1]. The data also overturn a prior biochemical assignment: K410, proposed as a gateway residue, tolerated broad substitutions during phage infection, implying that another residue fulfills that role in vivo [1]. This kind of discrepancy between purified-protein biochemistry and native-context genetics is exactly what DMS is suited to detect, and it reinforces the value of assaying function in the full infection cycle rather than in isolation [1][8].

The β-helix stalk emerged as a distinct adaptive module. Variants in this region were enriched on EV36 and appear to destabilize sialic acid interactions, which the authors propose modulates processivity — the rate at which the enzyme translocates along the polysialic acid chain [1]. The interpretation is that tighter binding is needed to overcome steric hindrance when O-antigen is present, while faster cleavage is advantageous when O-antigen is absent, as in EV36, which carries the rfb-50 mutation in wbbl [1]. This trade-off between binding strength and catalytic rate parallels behavior observed in other carbohydrate-active enzymes [1]. The stalk is therefore not a passive structural element but a tuning knob whose mutational tolerance varies with host surface chemistry [1].

Host range is shaped by capsule modifications and O-antigen, not capsule serotype alone

Comparative selections across multiple K1 strains identified discrimination hotspots in β-barrel loops and distal residues outside canonical binding sites [1]. Clonal validation confirmed strain-specific effects: G730D and G742Q showed growth defects only on E. coli O45:H10 and A204, S460R was defective exclusively on those two strains, and N386M produced a subtle defect only on A204 [1]. These results demonstrate that single point mutations can alter host range even among strains sharing the same K1 capsule serotype [1]. The only known K1 capsule modification, O-acetylation by the prophage-encoded NeuO acetylase, did not explain the discrimination patterns, as only one strain carried the gene and the associated mutational profile was not observed there [1]. LPS structure also did not appear to contribute directly, since the tested strains except one share highly similar LPS architecture [1].

The emerging model is that capsule and O-antigen presence, rather than LPS, are the dominant modulators of K1F host range [1]. This is consistent with the broader principle that phage host range is a multi-step phenotype — binding, genome entry, defense evasion, replication, and release — and that alterations to receptor-binding proteins can change host range without necessarily expanding it [3]. The K1F DMS data add residue-level resolution to that principle for a capsule-degrading enzyme, showing that discrimination is encoded across a distributed network of residues whose effects depend on host surface chemistry [1]. The limitation is that the library did not productively infect E. coli K5 or Klebsiella pneumoniae K1/K2 strains, indicating that single point mutations are insufficient to reprogram substrate specificity across distinct capsule types [1].

What the functional map does not yet establish

The functional scores come from in vitro DMS and K1 strain selections, not from in vivo infection models or therapeutic efficacy testing [1]. The paper explicitly frames the work as a roadmap for designing phages that overcome capsule-based defenses, but the engineered variants have not been validated in animal models or against clinical isolates beyond the tested K1 strains [1]. This boundary matters because phage therapy outcomes depend on population-level dynamics that single-strain selections cannot capture. In Klebsiella pneumoniae, for example, phage-resistant populations display substantial phenotypic heterogeneity, with reversible acapsular variants, stable acapsular mutants, and capsule-reverted isolates coexisting; cocktail efficacy depends on targeting the full spectrum of resistant variants, not just the dominant clone [5]. A tailspike variant optimized against a single K1 strain may face different selective pressures in a heterogeneous bacterial population [1][5].

The DMS approach also has inherent coverage limitations. The designed library encompassed 22,365 variants, but only 18,866 (85%) were recovered after ORACLE, and dropout was attributed to intrinsic structural constraints rather than technical failure [1]. Computational imputation methods such as VEFill can fill missing DMS scores with moderate accuracy (Pearson r = 0.80) for stability-based datasets, but performance drops markedly on activity-based assays [6]. Since the K1F dataset is activity-based, imputation would likely be less reliable for the missing variants [1][6]. The broader DMS literature shows that multi-phenotype screens can reveal pleiotropic effects that single-readout assays miss — for example, P2RY8 variants can affect surface expression, migration, and proliferation discordantly [7], and influenza hemagglutinin mutations can have opposing effects on cell entry, acid stability, and antibody escape [10]. The K1F study measured a single composite phenotype (phage fitness), so variants with opposing effects on binding versus catalysis could be masked [1]. Whether the β-helix stalk tuning knob generalizes to other capsule-degrading tailspikes, and whether the identified discrimination residues can be combined to produce phages with expanded host range, remain open questions [1].

About These Sources

This research page is built on 10 peer-reviewed studies — published from 2010 to 2026, 9 from 2024 or later, collectively cited 68 times — selected as the most relevant from 13 studies that passed quality screening, drawn from 57 papers retrieved from a database of over 500 million.

Sources used in this answer