Sequence-Function Relationships in Phage-Encoded Bacterial Cell Wall Lytic Enzymes and Their Implications for Phage-Derived Product Design

Abstract

Phage (endo)lysins are thought to be a viable alternative to usual antibiotic chemotherapy to fight resistant bacterial infections. However, a comprehensive view of lysins' structure and properties regarding their function, with an applied focus, is somewhat lacking. Current literature suggests that specific features typical of lysins from phages infecting Gram-negative bacteria (G-) (higher net charge and amphipathic helices) are responsible for improved interaction with the G- envelope. Such antimicrobial peptide (AMP)-like elements are also of interest for antimicrobial molecule design. Thus, this study aims to provide an updated view on the primary structural landscape of phage lysins to clarify the evolutionary importance of several sequence-predicted properties, particularly for the interaction with the G- surface. A database of 2,182 lysin sequences was compiled, containing relevant information such as domain architectures, data on the phages' host bacteria, and sequence-predicted physicochemical properties. Based on such classifiers, an investigation of the differential appearance of certain features was conducted. This analysis revealed different lysin architectural variants that are preferably found in phages infecting certain bacterial hosts. In particular, some physicochemical properties (higher net charge, hydrophobicity, hydrophobic moment, and aliphatic index) were associated with G- phage lysins, appearing specifically at their C-terminal end. Information on the remarkable genetic specialization of lysins regarding the features of the bacterial hosts is provided, specifically supporting the nowadays-common hypothesis that lysins from G- usually contain AMP-like regions. IMPORTANCE Phage-encoded lytic enzymes, also called lysins, are one of the most promising alternatives to common antibiotics. The potential of lysins as novel antimicrobials to tackle antibiotic-resistant bacteria not only arises from features such as a lower chance to provoke resistance but also from their versatility as synthetic biology parts. Functional modules derived from lysins are currently being used for the design of novel antimicrobials with desired properties. This study provides a view of the lysin diversity landscape by examining a set of phage lysin genes. We have uncovered the fundamental differences between the lysins from phages that infect bacteria with different superficial architectures and, thus, the reach of their specialization regarding cell wall structures. These results provide clarity and evidence to sustain some of the common hypotheses in current literature, as well as making available an updated and characterized database of lysins sequences for further developments.

Keywords: antimicrobial agents; bacteriophage therapy; bacteriophages; bioinformatics; endolysins; genomics.

FIG 1

Schematic examples of the architecture of peptidoglycan hydrolases. The top row in each panel corresponds to the domain structure based on the three-dimensional folding according to the corresponding PDB entry (red boxes are EADs; blue boxes are CWBD repeats). In the bottom rows, the domain structure predicted using HMMSCAN is shown. Numbers indicate the amino acid coordinates. (A) LytA autolysin from S. pneumoniae. Note that only five choline-binding repeats (CW_binding_1) of the six present in the three-dimensional structure are predicted. (B) Lysozyme from lambda coliphage.

FIG 2

General properties of lysins from phages that infect G+ or G−. (A) Distribution of the number of PF hits predicted per protein. (B) Distribution of domain types. (C) Distribution of protein lengths. (D and E) Distributions of the number of amino acids before (D) or after (E) predicted EADs. (F) Percentage of lysins with a predicted N-terminal signal peptide according to Phobius. (G) PF domain variability (different colors stand for different PF domain families, corresponding to those in Table 2). In distribution charts (C, D, and E), the y axis shows an estimation of the distribution density.

FIG 3

Differential distribution of PF hits among G− and G+ bacterial hosts. The y axis shows the proportion of PF hits found in G+ within a given domain family. Gray bars and numbers above the graph represent the total number of hits of each PF domain.

FIG 4

Heat map of PF hits distribution across host bacterium genera. Numbers within each tile indicate the number of hits predicted for the corresponding taxon and PF family. The color scale represents the number of hits from low (red) to high (yellow). Gray bars on the right represent the total number of PF hits predicted within each genus.

FIG 5

Relevant architectures observed in lysins from phages infecting different taxonomic groups of bacteria. Different colors indicate different domains; brackets denote domains that appear in only some representatives of the depicted architecture.

FIG 6

Heat map depicting PF hits distribution among different streptococci. Numbers within each tile indicate the number of hits for the corresponding taxon, whereas colors express a scale from lower (red) to higher (yellow) number of hits. Gray bars show the total number of PF hits for each streptococcal species.

FIG 7

Differential distribution of CWBDs and catalytic activities across peptidoglycan chemotypes and taxonomic groups of bacterial hosts. (A) Schematic representation of the relevant peptidoglycan chemotypes present for the bacterial hosts in our data set. (B) Distribution of CWBD PF hits among chemotypes. (C) Distribution of catalytic activities of EAD PF hits among chemotypes and taxonomic groups.

FIG 8

SSNs of the PF hits in our data set corresponding to different domain families. (A) Classification of sequences according to the taxonomic group of the corresponding bacterial host. (B) Classification by peptidoglycan chemotype of the host. Each node represents a single sequence. Dashed lines separate recognizable similarity clusters. The edge similarity cutoff was ≈ 40% for CHAP, LysM, and SH3_5 and ≈ 30% for Amidase_2.

FIG 9

Random forest prediction and classification of Gram groups of bacterial hosts based on physicochemical properties of lysins (A, B, and C) or on those properties plus others relative to lysin architecture (D, E, and F). (A and D) ROC curves of the random forest predictive models (TPR, true-positive rate; FPR, false-positive rate). ROC best points of positive-group (G+) probability for outcome maximization are presented, as well as the AUCs. (B and E) Random forest castings of bacterial host Gram group on the testing subset of lysin sequences. The dashed lines represent the G+ probability threshold for classification based on the respective ROC best points. (C and F) Importance (i.e., mean Gini index decrease for each variable) of each of the four descriptors used for classification within each model. HM, hydrophobic moment.

FIG 10

Differential physicochemical properties distribution among G+ and G− phage lysins. (A) Distribution of net properties calculated along the whole protein sequences of lysins from phages infecting G− or G+. (B) Local computation of physicochemical properties. Each dot represents the particular value calculated for an 11-aa window in a given lysin. Continuous lines are average tendencies based on either all G− or all G+ data points. (C) Distribution of different properties at quartiles of lysin sequences. Asterisks indicate P values (**, P ≤ 0.01; ***, P ≤ 0.001) obtained from the Yuen-Welch test for trimmed means with a trimming level (γ) of 0.2. ES indicates the Wilcox and Tian ζ measurement of effect size.

FIG 11

Net charge distribution of lysins from G−-infecting phages classified according to the predicted EAD. The rightmost gray bars depict the number of lysins classified into each EAD group (lysins within the NA group are those for which an EAD was not assigned). All groups were compared with the distribution of the Amidase_2 domain, as a highly represented, near-neutral control using Welch’s test on γ = 0.2 trimmed means with post hoc Bonferroni correction (*, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001).

FIG 12

Local computation of physicochemical properties in lysins from G− infecting phages classified according to EAD predictions. Each dot represents the particular value predicted for an 11-aa window from a given lysin. Continuous lines are average tendency lines.