Abstract
Virulent bacteriophages infecting Klebsiella pneumoniae often show capsule-driven host tropism due to the presence of capsule-specific depolymerases. Yet for temperate phages the genetic and functional basis of such capsular specificity remains less well understood. Depolymerases appear unexpectedly rare in prophage genomes, raising unresolved questions about which prophage genes mediate capsular specificity, whether this apparent scarcity reflects biological or ecological differences versus annotation limitation, and whether prophage-encoded receptor-binding proteins (RBPs) are functionally active. To address these questions, we analysed 3,900 Klebsiella genomes from diverse ecological niches to identify prophage-encoded proteins mediating capsular specificity. We conducted a genome-wide association study (GWAS) correlating prophage protein clusters (from 8,105 prophages) with confidently assigned bacterial K-loci. GWAS revealed statistically supported predictors of capsular specificity for 16 of the 35 most diverse K-loci analysed. These predictors were dominated by diverse RBPs, including classical [Formula: see text]-helix depolymerases (6 predictors), SGNH-domain hydrolases predicted to deacetylate polysaccharides (6 predictors), and structurally novel RBPs lacking known depolymerase folds (2 predictors). Nearly one-third of K-loci yielded no statistically significant predictors. A targeted experimental screen of 50 candidate prophage depolymerases showed that 34 failed to yield detectable recombinant expression, and neither sequence similarity, structural prediction, nor prophage genomic context reliably predicted activity. Of the 14 active enzymes, 5 targeted a K-type different from that predicted of their bacterial host, and enzyme specificity was not consistently explained by sequence or structural homology. Comparison with GWAS predictions revealed that 10 of the 12 strongest GWAS predictors were experimentally validated, while 2 remained inconclusive. Together, these results highlight the intrinsic difficulty of predicting activity and capsular specificity of prophage-encoded RBPs from genomic information alone. Finally, analysis of 4,598 high-completeness prophages revealed that SGNH-domain hydrolases are among the most prevalent enzymatic domains in prophage RBPs. Two SGNH-domain RBPs identified by GWAS were experimentally confirmed as active esterases, supporting capsule deacetylation as a widespread alternative to polysaccharide depolymerisation in temperate phages. Our findings reveal that Klebsiella prophages encode structurally diverse RBPs, suggesting temperate phages may rely not only on depolymerisation but also on capsule modification-such as deacetylation-for infection. This also suggests that capsule modification may contribute to phage-host interactions in ways not fully captured by current K-locus assignments, with potential implications for phage specificity, competition and vaccine design.