{"paper_id":"46372d0b-34fd-4b1e-8e8b-cb5f351cab7d","body_text":"Discovery of anti-enterococcal phage lysins from environmental metagenomes using protein embedding-based classification | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Discovery of anti-enterococcal phage lysins from environmental metagenomes using protein embedding-based classification Iris Pottie, Alexandre Boulay, Lander De Coninck, Jelle Matthijnssens, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8328502/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 1 You are reading this latest preprint version Abstract Phage lysins, enzymes encoded by bacteriophages that degrade bacterial cell walls, are emerging as a promising class of antimicrobial agents. This study aimed to discover novel lysins with activity against Enterococcus species using a sequence-based metagenomic discovery pipeline. Viral metagenomic DNA was extracted and sequenced from five environmental samples originating from pig feces or sewage. Putative lysins were first predicted with SUBLYME, a protein embedding-based classifier. Subsequently, a specific protein embedding-based classifier was developed to predict lysins with potential activity against Enterococcus . A total of 8 825 candidate lysins were predicted, including 129 with potential anti-enterococcal activity. Comparative analysis revealed differences in domain architectures and physicochemical properties between lysins derived from fecal and sewage samples, suggesting distinct phage host origins. A subset of the predicted lysins was expressed in Escherichia coli , partially purified and tested for muralytic activity against three enterococcal species ( Enterococcus faecium , Enterococcus faecalis , and Enterococcus hirae ). Among the 21 expressed lysins with variable expression yields, four exhibited lytic activity against all three Enterococcus species, two were active against Ent. faecalis and Ent. hirae , and seven showed activity exclusively against Ent. hirae . Six of these active proteins contained previously unreported domain architectures, indicating that this approach can uncover structurally novel functional lysins. While this pipeline was applied to Enterococcus , it is broadly adaptable for the discovery of lysins targeting other bacterial pathogens, offering a scalable approach to expand the antimicrobial arsenal. sequence-based metagenomics bacteriophage lysin Enterococcus Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Introduction The rise of bacteria acquiring resistance against antibiotics renders many currently used antibiotic treatments ineffective, thereby causing a considerable threat to global health. Misuse and overuse of antibiotics further exacerbate antibiotic resistance. This growing resistance crisis coincides with a stagnating antibiotic development pipeline, creating what has been described as a “perfect storm” [ 1 ]. Moreover, broad-spectrum antibiotics impact the delicate balance of the human microbiota, a diverse community of microorganisms essential for various physiological functions [ 2 , 3 ]. This microbiota disruption (“dysbiosis”) has been associated with health complications, including gastrointestinal disorders [ 4 ], obesity [ 5 , 6 ], and autoimmune diseases [ 7 ]. To address these challenges, there is an urgent need for novel antimicrobial strategies that are both effective and specific. Among the most promising candidates are phage lysins, enzymes derived from bacteriophages that selectively degrade bacterial cell walls. These enzymes selectively target and degrade the bacterial cell wall of specific bacteria, resulting in rapid lysis, while rendering other non-targeted bacteria of the microbiome unaffected. Their unique mechanism of action, together with their vast diversity and engineering possibilities to further finetune their specificity and efficacy differentiate them from traditional antibiotic classes. However, a key bottleneck in advancing lysins from promising research tools to effective therapeutics is the early-stage identification and collection of candidate lysins and their building blocks [ 8 ]. Metagenomics, the study of genetic material recovered directly from environmental samples, offers a powerful solution to this bottleneck [ 9 ]. Classical sequence homology-based metagenomics typically relies on sequence homology searches to mine environmental datasets and has already enabled the discovery of diverse enzymes [ 10 ]. For example, Fernández-Ruiz et al. [ 11 ] identified 2 628 endolysins from 183 289 uncultured viral genomes using a sequence homology-based pipeline. Hits meeting thresholds for identity, coverage, alignment length, and E-value were considered as putative lysins. Many of these hits exhibited novel domain architectures, highlighting the potential of metagenomics to feed the lysin hit-to-lead pipeline with candidates possessing unexpected traits which are of value for combinatorial engineering. However, this approach inherently overlooks novel lysins that lack detectable similarity to known lysins. Advances in computational and artificial intelligence methods have leveraged the discovery of novel enzymes through sequence-based approaches. Recently, the SUBLYME (Software for Uncovering Bacteriophage LYsins in MEtagenomic datasets) tool has been developed to identify lysins from metagenomic datasets [ 12 , 13 ]. Protein embeddings generated by language models such as ProtT5 were used to create vector representations of lysins [ 14 ]. These representations served as input for support vector machines (SVMs), which were trained on a curated dataset from PhaLP [ 15 ], a comprehensive lysin database ( https://phalp.ugent.be ), to sequentially identify and classify lysins as virion-associated lysins (VALs) or endolysins. When applied to the same database as the previously mentioned study of Fernández-Ruiz et al. [ 11 ], 50 856 putative lysins were predicted, including 98% of the previously identified lysins. In this study, we applied a sequence-based metagenomic pipeline to discover novel lysins with activity against Enterococcus species, opportunistic pathogens that exemplify the need for targeted antimicrobials. Enterococcus faecalis and Enterococcus faecium , typically commensals of the human gastrointestinal tract, can become pathogenic under conditions such as antibiotic treatment, immunosuppression, or gut barrier disruption. They are now recognized as major contributors to hospital-acquired infections, including urinary tract infections, bacteremia, endocarditis, and wound infections. Their intrinsic resistance to many antibiotics and capacity to acquire additional resistance genes makes treatment increasingly difficult [ 16 ]. To address this, we extracted and sequenced metagenomic DNA from five environmental samples (originating from pig feces and sewage). Protein-coding regions were predicted from assembled contigs, and candidate lysins were identified using SUBLYME. A second SVM-based classifier was trained to predict lysins with potential activity against Enterococcus species (hereafter referred to as anti-enterococcal lysins). A subset of 21 predicted lysins was expressed and tested for muralytic activity against Enterococcus hirae , Ent. faecium , and Ent. faecalis , revealing promising candidates for further development. 2 Materials and methods Sample collection and virion isolation Five different samples were collected, all originating from pools of pig feces or hospital sewage (Table 1 ). Samples were either processed immediately (hospital sewage) or stored at -20°C until further use (pig feces). Table 1 Overview of the characteristics of the samples for which environmental DNA was obtained. For each sample, the collection date, source, number of stool samples and subsequent library is depicted Sample Name Collection date Source Number of stool samples/ volume of sewage Library 1 2024/02/23 Pig feces (Merelbeke 1 ) 14 samples Library 2 2024/02/26 Pig feces (ILVO 2 ) 10 samples Library 3 2024/03/08 Pig feces (ILVO 2 ) 20 samples Library 4 2024/04/13 Hospital sewage UZ Gent 3 1 L Library 5 2024/04/16 Hospital sewage UZ Gent 3 1 L Virus purification and concentration were performed according to the methods used at the Quadram institute and the NetoVIR protocol [ 18 , 19 ]. For fecal samples, approximately 4 g aliquots were prepared. Sterile phosphate buffered saline [PBS, pH 7.4; 137 mM NaCl (Thermo Scientific), 2.7 mM KCl (Carl Roth), 10 mM NaH 2 PO 4 (Thermo Scientific), 1.8 mM KH 2 PO 4 (Carl Roth)] was added to each aliquot in a 1:10 (w/v) ratio, followed by vortexing for at least 1 min until samples were homogeneous. Samples were then incubated on ice for 1 h to promote viral particle release and centrifuged (3 000 × g, 5 min at 4°C) to remove debris. The supernatant was transferred to a new tube and subjected to consecutive centrifugation steps (20 min at 7 196 × g or 15 min at 10 000 × g, both at 4°C) until full clarity was achieved, as assessed by naked-eye inspection. The clear supernatant was filtered using a 0.45 µm polyethersulfone bottle top filter (Thermo Scientific). Next, viruses were precipitated using polyethylene glycol (PEG). A solution containing 1 M NaCl and 25% (w/v) PEG 8000 was added to the filtered phage lysate to reach a final PEG concentration of 7.5% (e.g., 15 mL solution per 35 mL of lysate), followed by incubation on ice for 16 h and centrifugation (7 196 × g, 40 min at 4°C). The resulting pellets were dissolved in sterile PBS. For sewage samples, initial centrifugation was performed at 3 000 × g for 5 min, followed by a second centrifugation step of the supernatant at 3 220 × g for 1 h (library 4) or at 17 000 g for 3 min (library 5). All centrifugations were performed at 4°C. The clear supernatant was filtered using a 0.45 µm polyethersulfone filter. PEG precipitation was achieved by adding NaCl to a final concentration of 1 M, dissolving it with a magnetic stirrer, and centrifuging at 3 200 × g. PEG 8000 was added to the supernatant to a final concentration of 10% (w/v), and samples were incubated on ice for 16 h, followed by centrifugation at 7 196 × g for 40 min or 12 000–17 000 × g for 30 min, both at 4°C. Pellets were dissolved in PBS (library 4) or immediately in 1 × DNase buffer (library 5). DNA extraction Two preparatory steps were necessary before extracting viral DNA. First, host and environmental DNA and RNA were removed to minimize background interference; second, viral capsids were digested to access the viral genomic material. To remove contaminant RNA and DNA, phage suspensions were treated for 90 min at 37°C with 100 U/mL RNaseA (Thermo Scientific) and 2 U/mL DNaseI (Thermo Scientific) in DNaseI buffer (Thermo Scientific). DNaseI and RNaseA were inactivated by adding 20 mM ethylenediaminetetraacetic acid (EDTA; Acros Organics). Subsequently, viral capsids were disrupted by adding 0.5% (w/v) sodium dodecyl sulfate (SDS; Merck) and 50 µg/mL proteinase K (Thermo Scientific), followed by 3 h incubation at 56°C. DNA was then extracted by adding an equal volume of phenol:chloroform (1:1 ratio, phenol at pH 8.0 from Carl Roth, chloroform from VWR), gently shaking for 1 min, incubating 1 min, followed by centrifugation (3 750 × g, 2 min). The upper aqueous phase was recovered and subjected to consecutive phenol:chloroform extraction steps until the middle, white layer disappeared. An equal volume of chloroform was then added, followed by gentle shaking for 1 min and centrifugation (3 750 × g, 2 min). The upper aqueous phase was recovered and 0.8 volume 100% (v/v) isopropanol (Carl Roth), 0.1 volume 3 M sodium acetate (VWR), and 40 µg glycogen (Thermo Scientific) were added to the aqueous phase, followed by incubation at 4°C for 16 h. DNA was collected by centrifugation (10 000 × g, 15 min at 4°C), washed twice with 70% (v/v) ethanol (VWR), and air-dried. Finally, the DNA was resuspended in TE buffer [10 mM Tris (Carl Roth), 1 mM EDTA, pH 8.0]. DNA quality and concentration were assessed using a microvolume spectrophotometer (Denovix) and a Quantus Fluorometer. DNA sequencing Metagenomic DNA was sequenced by Eurofins Genomics, using the INVIEW METAGENOME service. This included standard genomic library preparation, followed by paired-end Illumina sequencing (2 × 150 bp), with guaranteed 10 M read pairs. Raw reads and contigs were deposited in the European Nucleotide Archive with study number PRJEB102709. Analysis of sequencing results Raw sequencing reads were processed with the ViPER v2.2 pipeline (Virome Paired-End Reads) to generate viral contigs [ 20 ]. This pipeline applies Trimmomatic v0.39 [ 21 ] to trim low-quality bases and adapters from raw reads, followed by assembling the paired-end reads of 1%, 10% and 100% subsets into contigs using metaSPAdes v3.15.5 [ 22 ]. These contigs were clustered using a combination of BLAST + v2.14.0 [ 23 ], and anicalc.py and aniclust.py scripts from the CheckV v1.0.1 packages [ 24 ]. Contigs larger than 500 bp were retained for further analyses. Contig completeness was assessed using CheckV v1.0.1. Abundance tables were then generated in the ViPER pipeline by mapping sequencing reads to each contig with bwa-mem2 v2.2.1 [ 25 ], thereby quantifying the number of reads aligned per contig. Taxonomic classification was performed in the ViPER pipeline with DIAMOND blastx v2.1.8 [ 26 ]. The geNomad pipeline v1.8.0 (database version v1.7) was subsequently employed [ 27 ]. GeNomad uses pyrodigal-gv, a Cython binding of Prodigal [ 28 ], to predict coding sequences and generate protein annotations. These predicted proteins were used in subsequent analyses, including the identification of potential phage lysins. Both the geNomad and the ViPER pipeline classify contigs. The geNomad pipeline was applied to assign contigs as viral, proviral, plasmid, or unknown [ 27 ]. This classification served as the basis for subsequent steps. Contigs labeled as unknown and those labeled as proviral were taxonomically classified using DIAMOND blastx within ViPER. Relative abundance across taxonomic kingdoms was inferred from read-mapping data generated by ViPER. Prediction of phage lytic enzymes The proteins obtained by pyrodigal-gv were used as input sequences for SUBLYME v1.1, a protein-embedding-based classifier that predicts lysins and classifies them as endolysins or VALs [ 12 , 13 ]. Predicted lysins were clustered using CD-HIT at 98% sequence identity to remove redundancy [ 29 ]. This reduction was performed separately for each library to preserve intra-library diversity, and subsequently on the combined dataset to retain only non-redundant sequences. For clarity, this final pooled and re-clustered set of lysins is hereafter referred to as “all lysins”. Characterization of lysins Relevant physicochemical properties of the predicted lysins were calculated using the Peptides package in R [ 30 ]. These included the pI, molecular weight (MW), grand average of hydropathy (GRAVY) index, and aliphatic index, all calculated with default parameters. Domain annotation was performed with InterProScan on the Galaxy server (Galaxy Version 5.59-91.0 + galaxy3). InterProScan integrates eighteen member databases, but only four were retained for analysis: Pfam, CDD, SMART and PROSITE. The remaining databases were excluded based on prior findings indicating that they often predicted overly broad, overly specific, or irrelevant motifs for phage lytic proteins [ 15 ]. To address overlapping domain predictions, a custom R function (code in Supplementary_information_01 ) was used. Domains predicted by the same database were all retained without filtering. For proteins with multiple domains predicted by different databases, pairwise overlaps were assessed by calculating the maximum percentage overlap between domain intervals (start and end positions). Overlaps greater than 80% were considered significant. To resolve these overlaps, a filtering strategy was applied: Pfam predictions were given priority, and any non-Pfam domain overlapping with a Pfam domain was removed. In the absence of Pfam domains, overlapping domains were filtered to retain only one representative (the first in the list, sorted by start position), while preserving the non-overlapping domains. The final output consisted of filtered domain data. Functional domains were annotated and classified following the framework proposed by Criel et al. [ 15 ], who identified lysin domain clusters in the PhaLP database and developed a classification system based on the predicted domain accession numbers ( Supplementary Table 1 ). Specifically, domains were classified according to their type into cell wall-binding domains (CBD) and enzymatically active domains (EADs). These were subdivided into 13 distinct CBD clusters and 26 EAD types. Additionally, EAD clusters were further grouped based on enzymatic activity, following the classification proposed earlier [ 15 ]. These seven groups included: (1) N -acetyl-β- d -muramidase, (2) transglycosylase, (3) combined N -acetyl-β- d -muramidase and transglycosylase, (4) N -acetyl-β- d -glucosaminidase, (5) N -acetylmuramoyl- l -alanine amidase, (6) peptidase, and (7) combined N -acetylmuramoyl- l -alanine amidase and peptidase activity. Domains not included in the classification of Criel et al. [ 15 ] were omitted. Predicting and selection of anti-enterococcal phage lysins To functionally characterize a subset of the predicted lysins against enterococci, a second tool was developed to identify lysins potentially active against enterococci. To this end, SVMs with radial basis function kernels and balanced class weights (inversely proportional to class frequencies) were trained on protein embeddings calculated using ProtT5 [ 14 ]. The implementation is available on GitHub at https://github.com/Rousseau-Team/lysin-target-pred . The PhaLP database ( https://phalp.ugent.be ) was used to train the models, containing 237 anti-enterococcal lysins (from phages infecting Enterococcus ) and 15 150 lysins associated with other hosts [ 15 ]. Rather than training one model on the whole dataset, which possesses almost 64 times more negative (associated to other hosts) than positive (anti-enterococcal) examples, 200 smaller models were trained using all 237 anti-enterococcal lysins combined with random subsets of 500 non-enterococcal lysins (Fig. 1 ). This approach ensured that each model was less biased toward the negative class while collectively covering the entire negative dataset. For evaluation, a stratified test set comprising 20% of the positive (~ 45/237) and negative (~ 100/500) examples was set aside, with the remaining 80% used for training. These training and testing sets were created based on sequence-level clusters at 30% sequence identity and 70% coverage using MMseqs2 to ensure that similar proteins were kept in either the training or the testing set [ 31 ]. This strategy allowed assessment of model generalization to unseen proteins. Models achieving a precision greater than 80% on the test set were retained, resulting in 79 of 200 models. All retained models were then used on the newly identified lysins to predict their potential anti-enterococcal activity. Lysins classified as anti-enterococcal by at least 80% of the retained models were selected for further analysis. Proteins exceeding 550 amino acids were excluded during manual curation to increase the likelihood of successful heterologous expression and purification [ 32 , 33 ]. Subsequently, and an uninformed random selection of 21 proteins was performed. The relevant metrics were calculated as indicated before. The three-dimensional structures of the selected proteins were predicted with AlphaFold3 [ 34 ]. The Segmentation of PhAge Endolysin Domains using Alphafold’s PAE matrix (SPAED; https://spaed.ca ) algorithm was used to delineate protein domains [ 35 ]. Bacterial strains and growth medium Escherichia coli BL21(DE3)pLysS was used as a strain for protein expression. E. coli was grown for 18 h at 37°C, shaking at 180 rpm in lysogeny broth (LB), or static on LB containing 1.5% agar (VWR). For cryopreservation, 50% and 20% glycerol stocks were made for respective storage at -20°C and − 80°C. Kanamycin (Carl Roth) at 50 µg/mL was used as selection marker to select for colonies with the inserted plasmids, 25 µg/mL chloramphenicol (Acros) was used for E. coli BL21(DE3)pLysS. A panel of three Enterococcus species was selected for this study: Ent. faecalis NJ-3 (ATCC 51299), obtained from the LGC collection [ 36 ]; Ent. faecium AW3576, kindly provided by Maia Merabishvili (Queen Astrid Military Hospital) [ 37 ]; and Ent. hirae ECC1, provided by Juan M. Rodríguez (Universidad Complutense de Madrid) [ 38 ]. All strains were cultured at 37°C in brain heart infusion (BHI) broth (Merck) or on BHI agar under static conditions. DNA manipulation of anti-enterococcal lysins The selected protein sequences were converted to coding sequences, adapted for E. coli codon usage, by Twist Bioscience. These insert sequences were then provided by flanking position markers and BsaI recognition sites and synthesized by Twist Bioscience ( Supplementary Table 2 ). Synthetic genes were diluted to a final concentration of 46 nM in ultrapure water. For the assembly reaction, 1 µL of the diluted gene was combined with 1 µL of 25 nM pVTD3 vector [ 39 ], 1 µL of 46 nM C-terminal hexahistidine tag (His-tag) containing position marker sequences for directional cloning (5′–GGA AGC–3′ overlap with the synthetic gene and 5′–ATA CTT–3′ overlap with the vector), 2 µL of 10 × T4 DNA ligase buffer (Thermo Scientific), 3.5 U T4 DNA ligase (Thermo Scientific), and 10 U of Eco31I (Thermo Scientific). Ultrapure water was added to bring the final reaction volume to 20 µL. The reaction was subjected to 80 thermal cycles alternating between 5 min restriction at 37°C and 5 min ligation at 22°C, followed by enzyme inactivation at 80°C for 5 min. Reaction products were purified using the GeneJET PCR Purification Kit (Thermo Scientific), according to the manufacturer’s instructions. Subsequently, 5 µL of the purified product was used to transform chemically competent E. coli TOP10 or E. coli BL21(DE3) pLysS cells, prepared using the rubidium chloride method [ 40 ]. Transformation mixtures were plated on LB plates with the required antibiotics, supplemented with 5% (w/v) sucrose (Thermo Scientific) as a negative selection marker. To verify the presence of the correct constructs, plasmids were isolated from a single bacterial colony using the GeneJET Plasmid Miniprep Kit (Thermo Scientific), following the manufacturer’s protocol. Sanger sequencing was performed by LGC Genomics using the T7 promoter primer (5′-TAA TAC GAC TCA CTA TAG GG-3′) and the T7 terminator primer (5′-GCT AGT TAT TGC TCA GCG G-3′). Sequencing results were analyzed using Benchling [ 41 ]. Protein expression and purification Protein expression was performed in terrific broth [0.4% glycerol (Acros Organics), 1.2% tryptone, 2.4% yeast extract, 1.7 mM KH 2 PO 4 (Carl Roth) and 7.2 mM K 2 HPO 4 (Carl Roth)] for 72 h at 16°C, shaking at 180 rpm. All previously selected proteins, together with PlyEF501 (UKM17465; an endolysin of Enterococcus phage AQEF5 as positive control protein), were expressed in 100 mL. Overnight precultures from single colonies were diluted 1:25 in terrific broth, grown to mid-exponential phase [optical density (OD 600 ) between 0.45–0.55], and induced by adding isopropyl β- d -1-thiogalactopyranoside (IPTG; Thermo Scientific) to a final concentration of 0.5 mM. Cultures were incubated at 16°C and 180 rpm for 72 h, cells were harvested by centrifugation (10 000 × g, 10 min, 4°C) and resuspended in 1 mL lysis buffer [20 mM imidazole (Carl Roth), 50 mM NaH 2 PO 4 -NaOH (Thermo Scientific), 300 mM NaCl, pH 7.4, supplemented with 10 µg/mL DNaseI from bovine pancreas (Sigma)]. Cell suspensions were disrupted with three freeze-thaw cycles, followed by sonication at 60% (3 min with 15 s on/off cycles). The resulting cell lysates were centrifuged at 10 000 × g for 40 min at 4°C, 300 µL HisPur Ni-NTA Superflow Agarose beads (Thermo Scientific) were added to the supernatant and incubated on an end-over-end shaker at 4°C for 16 h. Mixtures were centrifuged (700 × g, 2 min, 4°C), the supernatant was removed, and the pellets were washed twice with 1.5 mL wash buffer (50 mM imidazole, 50 mM NaH 2 PO 4 -NaOH, 300 mM NaCl, pH 7.4). Elution was performed twice with 250 µL elution buffer (50 mM NaH 2 PO 4 -NaOH, 300 mM NaCl, 500 mM imidazole, pH 7.4). Both fractions were pooled and a buffer exchange to PBS was performed with Zeba Spin Desalting Plates, 7K MWCO (Thermo Scientific), according to the guidelines provided by the manufacturer. Protein concentration was estimated using the microtiter plate protocol of the micro BCA protein assay kit (Thermo Scientific), following the manufacturer's guidelines with bovine serum albumin as a standard. Protein purity was assessed with SDS-PAGE, using 12% acrylamide gels. PageRuler Broad Range unstained ladder (Thermo Scientific) was used as a protein marker. Turbidity reduction assay Lytic activity of lysins against enterococci was evaluated using a turbidity reduction assay (TRA). An 18 h preculture of the target strain was diluted 1:25 in BHI and grown to mid-exponential phase (OD 600 = 0.45–0.55). Cells were harvested by centrifugation (12 000 × g, 10 min, 4°C), washed once with PBS, and resuspended in PBS. The bacterial suspension was adjusted so that, when 55 µL was combined with 45 µL PBS in the wells of a 96-well half-area microtiter plate (microplate, 96 well, polystyrene, half area, clear, Greiner Bio-One), the final OD 600 was approximately 1, accounting for dilution upon enzyme addition. For the assay, 55 µL of the bacterial suspension was added to 45 µL of semi-purified lysins in the half area microtiter plate. OD 600 was measured every minute using a Tecan Infinite plate reader for 3 h at 37°C. PlyEF501 was used as positive control and just PBS served as a negative control. Each measurement was performed in biological triplicates. The OD 600 decrease with time was plotted as a general kinetic view. To quantitatively assess the lytic activity, the area between the curve of the negative control and each treatment (ABC) was calculated as: $$\\:ABC=\\:\\:{AUC}_{control}-{AUC}_{protein}$$ With AUC control and AUC protein being the area under the curve of the negative control and the putative lysin, respectively. Statistical analysis Statistical analyses were performed in the RStudio interface (2023.03.01 + 446), with the R programming language v4.3.1 [ 42 , 43 ]. The packages that were employed include ggplot2 v3.5.1 [ 44 , 45 ], readxl v1.4.3 [ 46 ], DescTools v0.99.58 [ 47 ], dplyr v1.1.4 [ 48 ], tibble v3.2.1[ 49 ], WRS2 v1.1-7 [ 50 ], and multcomp v1.4-26 [ 51 ]. To compare putative lysins between libraries, robust statistical methods were applied to account for non-normal and heteroskedastic data distributions. Continuous protein metrics (length, MW, pI, aliphatic index, and GRAVY index) were analyzed using a Kruskal–Wallis test, followed by Dunn’s post hoc test with Holm adjustment for multiple comparisons. For comparisons between lysins predicted to be anti-enterococcal and all remaining lysins, a Wilcoxon rank-sum test was performed. Effect sizes (ES) were estimated as described by Wilcox and Tian [ 52 ] with a heteroskedastic analysis of variance (ANOVA) on trimmed means (default trimming level γ = 0.20) for comparison between the libraries, using the ζ statistic. Interpretation of ζ followed the guidelines of that reference: approximately 0.1 indicated a small effect, 0.3 a medium effect, and higher than 0.5 a high effect [ 52 ]. Distribution of predicted protein domains were compared by a χ 2 -test, followed by evaluation of Pearson’s standardized residuals if significant differences (α = 0.05) were observed. Differences in lytic activity during TRA were evaluated with an ANOVA, followed by Dunnett’s post-hoc test to check if the treatment with a specific endolysin resulted in an ABC significantly different from zero. 3 Results From raw DNA reads to predicted lysins In this study, viral DNA was extracted from five distinct samples derived from pig feces or sewage (Table 1 ), followed by Illumina sequencing. Raw reads (18.9–23.47 million per sample) were processed through the ViPER pipeline for trimming and assembly, yielding 41 978–76 946 contigs per sample (Table 2 ). Table 2 Illumina sequencing metrics and lysin identification for five libraries. ‘Raw reads’ are expressed in millions (M). ‘No. contigs’ indicates the total number of assembled contigs, while ‘Viral contigs’ shows the percentage of contigs classified as viral. ‘Viral abundance’ represents the proportion of reads mapping to viral contigs. Columns ‘All proteins’ and ‘Lysins’ report the total predicted proteins and lysins, respectively; numbers in round and square brackets denote predicted endolysins and virion-associated lysins (VAL), respectively. Some lysins could not be assigned to either category. The percentage of lysins among all proteins is shown under ‘% Lysin’. ‘Representative lysins’ indicates the number of non-redundant lysins after clustering at 98% sequence identity Library Source Raw reads (M) Contigs (No.) Viral contigs (%) Viral abundance (%) All proteins (No.) Lysins (No.) (endolysin) [VAL] % Lysins Representative lysins (No.) 1 Pig feces 23.47 41 978 46.28 88.1 117 593 1 348 (1 241) [83] 1.1 1 327 2 Pig feces 19.71 63 031 50.27 81.3 158 086 1 904 (1 821) [50] 1.2 1 881 3 Pig feces 18.91 76 946 49.07 75.9 175 203 2 079 (1 994) [64] 1.2 2 059 4 Sewage 20.1 56 841 31.11 81.3 144 410 1 707 (1 481) [211] 1.2 1 695 5 Sewage 18.88 71 885 40.51 69.8 161 639 2 101 (1 823) [260] 1.3 2 087 Contigs were classified using a combination of the geNomad and ViPER pipelines. GeNomad initially categorized contigs as viral, proviral, or plasmid. Contigs classified as proviral, as well as those not assigned by geNomad, were subsequently classified based on their ViPER annotation. Of all contigs, for each library, between 17 688 and 37 754 (31.11–50.00%) were identified as viral. Relative abundance was estimated by summing the reads mapped to the contigs for each kingdom (Supplementary Fig. 1) , with viral reads comprising 69.8–88.1% of the total reads (Table 2 ). No significant differences in abundance of phage kingdom-level classifications were observed across libraries (χ² test, P = 0.81). Contig completeness was subsequently assessed with CheckV for contigs classified as viral by ViPER. CheckV compares contigs to a reference database of complete viral genomes and assigns quality categories: complete, high (> 90%), medium (50–90%), low (0–50%), and undetermined. This analysis indicated that approximately 80–85% of viral contigs were of low quality and thus incomplete (< 50% completeness; Supplementary Fig. 2 ). Next, protein-coding sequences were predicted using pyrodigal-gv within geNomad, resulting in 117 593 − 175 203 predicted proteins per library. These proteins were used as input for SUBLYME. This tool is a protein embedding-based classifier to identify phage lysins among viral proteins, using ProtT5 to generate vector-based representations (embeddings) of protein sequences, which are then classified using SVMs [ 12 , 13 ]. Across libraries, 1 348 to 2 101 lysins were predicted, representing approximately 1.1% (range: 1.1–1.3%) of all predicted proteins. The number and proportion of lysins per library are summarized in Table 2 . In addition to lysin prediction, SUBLYME predicts the type of lysin, either endolysin or VAL, indicated between round and square brackets in Table 2 . Remarkably, most lysins are predicted to be endolysins, with VALs contributing to approximately 7.4% (653 lysins) of the predicted lysins. The other predicted lysins are either endolysins (8 158 in total) or could be either endolysins or VALs (14 in total). After clustering predicted lysins at 98% sequence identity using CD-HIT, each library contained between 1 327 to 2 087 unique lysins. When all libraries were combined and re-clustered, the dataset was reduced to 8 825 non-redundant lysins. This final set is referred to as ‘all lysins’ throughout the manuscript. Exploring the metagenomic lysin landscape 3.1.1 Physicochemical properties To characterize the lysin landscape, several physicochemical properties were calculated for lysins from each individual library as well as for the combined dataset. These properties included protein length, MW, pI, aliphatic index, and hydrophobicity. Previous studies have shown that full-length lysins from Gram-negative phages are typically shorter than those from Gram-positive phages [ 53 ]. The pI, which indicates the pH at which a protein carries no net charge, and the GRAVY score, which reflects overall hydrophobicity [ 54 ], tend to be higher in lysins from phages infecting Gram-negative bacteria [ 53 ]. Similarly, the aliphatic index, which represents the relative volume of aliphatic side chains and correlates with protein thermostability [ 55 ], has also been reported to be elevated in Gram-negative phage lysins [ 53 ]. Distributions of these properties across the 5 samples are visualized in violin plots (Fig. 2 ). All the differences in physicochemical features across libraries found were low-to-moderate (ES = 0.2), with a lower ES for the pI (0.07), which strongly signals that there are no differences in pI distributions between the libraries. Libraries from pig feces (L1-L3) contained more lysins that were longer and with a higher MW, with a broader distribution (Mean = 22 kDa, Q1 = 17 kDa, Q3 = 28 kDa) than in sewage samples L4-L5 (mean = 19 kDa, Q1 = 15 kDa, Q3 = 25 kDa). In contrast, fecal samples contained more lysins with a lower aliphatic index and GRAVY than sewage samples. 3.1.2 Architecture and functional domain families Domain architectures of lysins were predicted using four InterProScan member databases. Based on these predictions, protein domains were classified using the framework described by Criel et al. [ 15 ], which organizes domains into clusters, comprising thirteen CBD clusters and twenty-six EAD types, with EADs further categorized into seven enzymatic activity groups (see 2.6 Characterization of lysins). Out of the 8 825 proteins analyzed, 6 700 (75.9%) contained at least one predicted domain, with library-specific proportions ranging from 71.6% to 80.0%. When restricting the analysis to proteins with domains annotated within the used framework, the proportion decreased to 72.9% (range: 69.7–76.6%; Supplementary Table 3 ). Complete domain architectures are listed in Supplementary Tables 4 and 5 . Unless otherwise stated, percentages refer only to proteins with at least one domain included in the framework. Most proteins contained a single domain (68%), predominantly an EAD (63%) rather than a CBD (5.3%). Among proteins with two annotated domains, the most common configuration was an N-terminal EAD followed by a C-terminal CBD (17%), while the reverse orientation (CBD–EAD) was rare (2%). Architectures with two EADs or two CBDs were observed in 2.5% and 1.9% of cases, respectively. Although most other configurations occurred at frequencies below 0.57%, an N-terminal EAD followed by two (4.6%) or three CBDs (1.1%) were relatively more common. These results should be interpreted cautiously, as repeated domains may form a single functional module [ 53 ]. For example, in this study, 267 of the 645 proteins with three or more domains were found to have two or more adjacent CW_7 CBDs. Additionally, the reliance on sequence-based domain prediction may underestimate the number of functional domains, particularly in proteins with divergent sequences or domains not yet represented in reference databases. No significant differences were found between libraries in the number of predicted domains per protein (χ² test), although some tendencies were noted. Fecal libraries (L1–L3) contained more proteins with two predicted domains (17–19%) and fewer with one domain (77–79%) compared to sewage libraries (L4–L5), which had 11–13% two-domain proteins and 85–86% single-domain proteins). This difference reflects the shorter length of lysins in sewage libraries and may indicate variation in the relative abundance of phages infecting Gram-negative versus Gram-positive bacteria, as the former typically encode single-domain lysins while the latter often exhibit multi-domain architectures related to the shorter length of these lysins [ 53 ]. Among all predicted domains, 73.3% were EADs, 20.8% were CBDs, and 6% were other domains not included in further analyses (Fig. 3 a ) . CBDs were classified into 13 domain clusters, PG_1 (37%), CW_7 (18%), SH3 (14%), LysM (11%), and PG_3 (10%) being the most frequent. These percentages represent the distribution of each cluster among all predicted CBDs. A significant difference in CBD cluster distribution was observed between libraries according to a χ 2 test ( P = 9.653 × 10 − 5 ). Standardized Pearson residuals (Fig. 3 b) identified PG_3 as being more prevalent in sewage libraries L4–L5 (25–26%; residuals 3.1–3.3 ) and less common in fecal libraries L1–L3 (ranging 2.5–10%; standardized Pearson’s residual − 3.0 to -0.88). The SPOR domain was more frequent in L2 and L3 (13% and 9% with Pearson’s standardized residual of 2.3 and 1, respectively), while CW_7 was enriched in fecal libraries ( 19–21%; standardized Pearson’s residual 0.65 and 1 ) compared to sewage libraries ( 8–14%; standardized Pearson’s residual − 2.1 to -0.61 ). The ratio of EADs with specific activities (or belonging to specific domain clusters) to the total number of EADs was calculated and expressed as a percentage for each library (Fig. 3 c). The most abundant EAD clusters were Ami_2 (14%), PET_M15 (12%), and SLT_related (12%), with all others below 10% each. The most frequently predicted enzymatic activities were N -acetylmuramoyl- l -alanine amidase (24%), N- acetyl-β- d -muramidase (22%) and peptidase activity (19%). No significant differences were identified between the libraries for activities (χ² test; P = 0.9117) or domain clusters (χ² test; P = 0.9998). However, libraries obtained from similar samples (feces or sewage) appeared to have more similar enzymatic domains. Fecal libraries (L1- L3) contained more EADs from the Ami_2, Ami_3, CHAP, and GH25 clusters, whereas sewage libraries (L4 and L5) had more EADs in the GH_24 and PET_M15 domain clusters. Similarly, N -acetylmuramoyl- l -alanine amidase activity was more frequent in fecal libraries (L1–L3), where N -acetyl-β- d -muramidase activity was more common in sewage libraries (L4 and L5). Together, these findings highlight distinct physicochemical and architectural profiles between lysins derived from fecal and sewage environments, reflecting underlying differences in phage host specificity and ecological context (Table 3 ). Table 3 Summary of differences between lysins predicted in fecal libraries and sewage libraries. GRAVY: grand average of hydropathy, ES: effect size, MW: molecular weight, ns: not significant, pI: isoelectric point Feature Fecal libraries (L1–L3) Sewage libraries (L4–L5) Statistical/interpretive note Protein length and MW Longer, higher MW Shorter, lower MW Overall differences small (ES ≈ 0.2) pI ~ no difference ~ no difference Negligible difference (ES = 0.07) Aliphatic index lower higher Small effect (ES ≈ 0.2) Hydrophobicity (GRAVY) lower higher Small effect (ES ≈ 0.2) No. predicted domains per protein More 2-domain proteins (17–19%); Fewer singledomain (77–79%) Fewer 2-domain proteins (11–13%) More singledomain (85–86%) χ² ns (Tendencies only) CBD composition (cluster-level) Higher CW_7 Lower PG_3 Higher SPOR Higher PG_3 Lower CW_7 χ² significant EAD composition (cluster-level) Higher Ami_2, Ami_3, CHAP, GH25 Higher GH_24, PET_M15 χ² for EAD clusters ns (Tendencies only) Enzymatic activities Higher N- acetylmuramoyl- l -alanine amidase Higher N -acetyl-β- d -muramidase χ² for activities ns (Tendencies only) Exploring the anti-enterococcal lysin landscape Next, lysins from phages targeting the genus Enterococcus (anti-enterococcal lysins) were predicted. To this end, a new protein embedding-based SVM classifier was trained, using 237 lysins from an enterococcal background and random subsets of 500 non-enterococcal lysins retrieved from the PhaLP database (https;//phalp.ugent.be) [ 15 ]. Models achieving a precision of at least 80% were retained, as this threshold provides confidence in prediction accuracy, despite potentially excluding some candidates. Of the 200 trained models, 79 met this criterion and were subsequently used to predict anti-enterococcal lysins among the newly identified sequences (Fig. 1 ). Lysins predicted as anti-enterococcal by at least 80% of these retained models were considered positive, yielding 129 lysins (1.5% of all lysins) predicted to be anti-enterococcal. Among these, 102 (79%) were predicted endolysins and 23 (18%) were predicted VALs. The distribution of the calculated MW, length in amino acids, aliphatic index, GRAVY index (hydrophobicity), and pI of the anti-enterococcal lysins versus other lysins can be found in Fig. 4 . Anti-enterococcal lysins exhibited a broad, bimodal distribution in both protein length and MW, with primary peaks around 275 and 380 amino acids (~ 30 and ~ 43 kDa), alongside several outliers. Compared to non-anti-enterococcal lysins, they were longer and had a higher MW (ES = 0.81 for both). Anti-enterococcal lysins showed a bimodal pI distribution, similar to the overall lysin dataset, with few proteins near the physiological pH of 7.4, indicating that most are charged under physiological conditions. Compared to non-anti-enterococcal lysins, their pI profile differed moderately (ES = 0.25), with an approximately even split between peaks at pI 5.3 and 9.3 (median = 6.8). In contrast, non-anti-enterococcal lysins showed two modes at pI 4.8 and 9.3 (median of 8.5). Moderate differences were observed for the GRAVY (ES = 0.26) and aliphatic index (ES = 0.48), both lower in anti-enterococcal lysins. These trends are consistent with typical features of Gram-positive lysins, which tend to be larger and more hydrophilic. The domain composition of anti-enterococcal lysins was examined using InterProScan to infer potential functional characteristics. Among the 129 proteins analyzed, 49 (38%) lacked any predicted domains, suggesting the presence of novel or highly divergent sequences not captured by the four domain databases used. Only 19 proteins (15%) contained at least one CBD, distributed across four domain clusters: SH3 (10 proteins), LysM (4), CW_7 (3), and PG_1 (3). In contrast, 79 proteins (61%) carried at least one EAD, with four proteins containing two EADs. Of the 83 predicted EADs, 22 were predicted as N -acetylmuramoyl- l -alanine amidases, six as peptidases, and twenty-six as having both activities (CHAP and NLPC_P60 domains). Additionally, nineteen proteins were predicted to have N -acetyl-β- d -muramidase activity, nine combined muramidase and lytic transglycosylase activity, and one was annotated with N -acetyl-β- d -glucosaminidase activity. The four proteins with dual EADs included combinations such as CHAP with GH25, CHAP with SLT-related, and GH25 with PET_M23 domains. Eleven proteins contained both a CBD and an EAD, with five showing an N-terminal EAD and C-terminal CBD, five having the reverse orientation, and one with an N-terminal EAD followed by two CBDs. In several cases, terminal regions lacked predicted domains, potentially indicating the presence of functional modules that are either novel or too divergent to be detected by current domain models. 3.1.3 Selection of anti-enterococcal lysins In the next step, a subset of 21 lysins was selected for expression and purification. For practical reasons, a maximum protein length of 550 amino acids was preferred. This threshold was set to enhance the likelihood of successful heterologous expression and purification [ 32 , 33 ]. After excluding twenty longer than 550 amino acids, 109 lysins remained. From the remaining pool, 21 lysins were selected through an uninformed, random selection. The physicochemical properties (length, MW, pI, aliphatic index, GRAVY index) of the 21 selected proteins were plotted as dots on the violin plots depicting the properties of all anti-enterococcal phage lysins (Fig. 4 ). This revealed that the random selection achieved to retain a similar distribution of the properties as observed for the whole population. Protein structures for the 21 selected putative lysins were predicted using AlphaFold3 [ 34 ]. Domain boundaries were identified with SPAED [ 35 ], and annotations were assigned via InterProScan [ 56 ] ( Supplementary Table 6 ). Of these 21 lysins, only ten contained at least one predicted domain as classified according to Criel et al. [ 15 ], while the remaining eleven comprised domains not included in that classification or lacked InterProScan predictions. Functional characterization of enterococcal phage lysins Twenty-one constructs were successfully cloned into the expression vector, and all were expressed and purified based on metal affinity. After semi-purification, the total protein concentration as estimated with the micro BCA method (bovine serum albumin as standard) ranged between 0.14 and 2.5 mg protein per 100 mL expression medium ( Supplementary Table 7 ), with highly heterogenous purities ( Supplementary Fig. 3 ). Due to the considerable variability in protein purity, equimolar comparisons between the tested proteins were not possible, and the concentrations given refer only to the total protein amount –not to the concentration of the protein of interest. The muralytic activity of the lysins was evaluated using a TRA, which measured the turbidity (as OD 600 ) of a bacterial suspension over time. A reduction in OD 600 was interpreted as an indication of cell lysis. Because the lysins were predicted on the genus level, a selection of three enterococcal species was used: Ent. hirae ECC1, Ent. faecium AW3576, and Ent. faecalis NJ-3. These strains represent two distinct peptidoglycan chemotypes found in enterococci (A3α for Ent. faecium and Ent. hirae , and A4α for Ent. faecalis [ 57 ]). Additionally, they include two clinically relevant species associated with human infections ( Ent. faecium and Ent. faecalis ). The progress of the OD over time ( Supplementary Fig. 4 ) was quantified by calculating the ABC, which represents the integrated difference in OD 600 between lysin-treated samples and the negative control across all time points (Fig. 5 ). The positive control exhibited significant lytic activity against all three Enterococcus species tested. Among the tested strains, Ent. hirae appeared most susceptible to lysis. Notably, some degree of autolysis was observed in the negative control of this strain, which may have weakened the peptidoglycan layer and sensitized cells, thereby enhancing the sensitivity to detect lysins with only minor to moderate activity. In total, four lysins showed lytic activity against Ent. faecium (E_24, E_25, E_31, E_65), six against Ent. faecalis (including the aforementioned plus E_82 and E_101), and thirteen against Ent. hirae (all previously mentioned plus E_28, E_46, E_60, E_61, E_63, E_76, and E_119), respectively. Thus, thirteen out of the twenty-one lysins tested (62%) displayed lytic activity against at least one Enterococcus species, confirming their anti-enterococcal potential. Notably, six of these active lysins (E_24, E_28, E_31, E_65, E_82, E_86) lacked any domains classified by Criel et al. [ 15 ]. 4 Discussion Phage lysins are enzymes encoded by bacteriophages that break down bacterial cell walls, representing a promising new class of antibiotics. However, discovering novel lysins is essential to fully realize their potential, as it expands the range of available lysins and provides insights into their sequences and corresponding activities. Sequence-based metagenomics reveals the genomic content of various environments, and, when paired with sequence mining, it becomes a valuable tool for discovering lysins in these diverse settings. In this study, a sequence-based metagenomic pipeline was implemented, which included the following steps: (i) extracting and sequencing viral metagenomic material from five samples originating from pig feces and sewage, (ii) assembling the sequences into contigs and predicting open reading frames along with related proteins, (iii) identifying lysins across the proteins, (iv) in silico characterization of the predicted lysins, (v) prediction of enterococcal phage origin, and (vi) functional screening of a selection of putative anti-enterococcal phage lysins. Sampling, DNA extraction and sequencing Five different samples were taken from two types of material: pig feces [from two locations: the faculty of veterinary Medicine of Ghent University (Merelbeke) and Flanders research institute for agriculture, fisheries, and food (ILVO)] and sewage from the Ghent University Hospital (UZ Gent). Both types of material are expected to contain enterococci and thus also their phages and lysins. In addition, the distinct types of starting materials (semi-solid and liquid) highlight the diversity in source material. Yet, this diversity in material type, location and date remains a biased snippet of the wide environmental diversity that exists. Moreover, this study focused on DNA extraction, and the Illumina sequencing with its associated library preparation primarily targets dsDNA, excluding RNA and ssDNA phages. Nevertheless, most characterized RNA and ssDNA phages are known to employ alternative lysis mechanisms that do not primarily rely on enzymatic degradation of peptidoglycan [ 58 , 59 ], but instead use other lysin systems, such as inhibiting new cell wall synthesis [ 60 ]. Such lysis systems would not be detected with the computational and experimental methods used in this work and thus fall out of scope. However, recent transcriptomic studies have uncovered greater diversity among RNA phages and their lytic strategies [ 61 ], suggesting potential for novel lysin discovery. The proposed pipeline could be adapted to include RNA and ssDNA phages by incorporating RNA-to-cDNA conversion [ 62 ] or converting ssDNA to dsDNA [ 63 ] during sample preparation. After assembling sequence reads, 310 681 contigs were obtained, of which 135 675 contigs were classified as viral. The abundance of reads mapping to the viral contigs amounted to 69.8–88.1%, (Table 2 ). Although classification and abundance estimates depend on the bioinformatic tools used for virus identification, viruses typically represent less than 5% of reads in non-enriched samples, suggesting the effectiveness of the enrichment strategy applied here [ 64 , 65 ]. Analysis with CheckV indicated that most viral contigs predicted using DIAMOND blastx were incomplete, meaning they represented only partial viral genomes with an estimated completeness of 0–50%. This is expected in metagenomic datasets due to the low sequencing coverage of rare viruses [ 64 ]. In total, 280 viral contigs were classified as complete genomes. Lysin prediction was performed on all contigs, regardless of completeness, which is an advantage of our approach since a full viral genome is not required to identify a lysin. However, when a viral genome is incomplete, it may lack the lysin gene entirely, which explains why a lysin could not be predicted for every contig. Prediction of candidate lysins This study utilized SUBLYME to predict candidate phage lysins, a protein embedding-based classifier offering an alternative to traditional sequence homology-based methods. While the latter approaches rely on sequence alignment tools to detect similarity between candidate proteins and known lysins using specific thresholds (e.g., sequence identity, alignment coverage, alignment length, and E-value), such approaches may fail to detect highly divergent or novel lysins with limited sequence similarity to known references. SUBLYME partly overcomes this limitation by leveraging protein embeddings to capture features beyond sequence similarity. In a previous benchmark study using the dataset from Fernández-Ruiz et al. [ 11 ], which included 3.8 million proteins from 183 298 uncultured phages, SUBLYME predicted 41 007 thousand endolysins (1.1%), compared to just 2 628 (0.069%) detected by the sequence homology-based methods. Additionally, 9 849 VALs were identified within this dataset (0.26%) [ 13 ]. Consistent with these findings, the present study predicted 8 825 representative lysins among 756 931 proteins (1.1%), including 8 158 endolysins (1.1%). In contrast, only 653 predicted VALs were detected (0.08%). While strong conclusions should be avoided given the uncertainty surrounding the prevalence of VALs across phage genomes, this low proportion suggests that the classifier may have difficulty accurately detecting VALs. This could be due to limitations inherent to the model, particularly the smaller training set for VALs compared to endolysins (respectively 4 429 versus 10 970 proteins), which may not sufficiently capture the expected diversity. Additionally, the presence of structural protein domains lacking muralytic activity within VALs further complicates their accurate identification. To improve VAL detection, a dedicated classifier trained specifically on VALs, using a broader and more diverse dataset, may be required, combined with further computational and experimental insight into the true VAL diversity. Moreover, the current model may possibly favor lysins with features resembling those in the PhaLP database, potentially overlooking novel variants. To identify completely novel lysins, alternative approaches may be required. One such approach is functional metagenomics, which bypasses the need for sequence similarity or prior knowledge by cloning environmental DNA into a heterologous host and screening the resulting expression libraries for lytic activity. This strategy enables the discovery of enzymes based on function rather than sequence and has already been successfully applied to discover lysins from (meta-)genomic sources, as reviewed elsewhere [ 66 ]. By incorporating newly discovered lysins from functional screens into the training set, machine learning models can be refined to improve detection of highly divergent lysins. Thus, sequence-based and functional metagenomics represent complementary strategies for fully exploring lysin diversity. Lysins across different environments Lysins from the different libraries were characterized by physicochemical properties (MW, length, pI, aliphatic index, and GRAVY index) and predicted domain composition. This characterization revealed that libraries from pig feces samples appeared to have characteristics mostly observed in lysins targeting Gram-positives, where sewage samples also contained lysins with characteristics typically found in lysins targeting Gram-negatives (Table 3 ). That is, the length and MW of lysins from fecal samples were significantly higher, a characteristic frequently observed in the (generally multimodular) lysins targeting Gram-positives [ 53 ]. Moreover, the domain cluster PG_3, exclusively observed in Gram-negative-targeting lysins [ 15 ], was significantly more present in lysins of sewage libraries (Fig. 3 ), together with a more frequent occurrence of GH24 enzymatic domains, an EAD cluster mostly observed in Gram-negative targeting bacteria [ 15 ]. Although these disparities may still be attributed to differences during the phage precipitation protocols, these findings suggest that (pig) feces were a rich source of lysins with a typical Gram-positive architecture, whereas sewage contained both phage lysins with typical Gram-positive and Gram-negative architecture. Similarly, Fernández-Ruiz et al. [ 11 ] found phages from aquatic samples to contain endolysins predicted to be active against both Gram-positive and Gram-negative bacterial hosts, where in the human microbiome, lysins with a typical architecture for targeting Gram-positives were preferentially found. Whether these architectural differences truly reflect variations in phage host range remains an open question and could be further investigated, for instance through refined host prediction approaches based on phage genomic data [ 67 ]. Anti-enterococcal lysins This study employed a new machine learning tool to predict putative phage lysins against a given bacterial genus, using an ensemble of classifiers trained on balanced datasets selected to have a precision higher than 80%. In this case, dealing with the prediction of anti-enterococcal lysins, the employed dataset (derived from PhaLP database) allowed to build 79 models which individually evaluated every lysin candidate. Lysins predicted to be anti-enterococcal lysins by at least 80% of the models were retained, providing high-confidence associations. The latter criterium is, however, tunable, and while in this work we chose to keep a conservative cut-off value, lowering the proportion of models with a positive outcome could still be valuable, especially if the aim is finding additional diversity –though at the cost of potentially introducing more false positives. This tool thus offers a pipeline for identifying anti-enterococcal lysins, with potential for adaptation to other bacterial genera. However, the approach may be overly permissive, potentially predicting non-active lysins as functional or misclassifying lysins with different host specificities as anti-enterococcal. To improve specificity, an alternative strategy could involve first predicting phage host taxonomy at the genus level using tools such as iPHoP [ 68 ], followed by lysin identification. Nonetheless, this method may overlook lysins with broad or cross-species lytic activity, as (endo)lysins often exhibit host ranges that extend beyond those of their associated phages [ 69 ], this approach may miss lysins with promiscuous or cross-species lytic activity. Selection of anti-enterococcal lysins To assess the predictive model and address concerns regarding potential false positives, a subset of candidate proteins was selected for experimental validation. To facilitate heterologous expression and purification, proteins longer than 550 amino acids were excluded. This criterion inherently removed most predicted VALs (sixteen out of twenty-three), retaining primarily endolysins. Although excluded from this initial screen, these larger VALs may still possess lytic potential. A feasible strategy for characterizing VALs while maintaining compatibility with expression systems could involve isolating and expressing only their lytic domains, omitting structural regions. An uninformed random selection was then performed to select 21 proteins for further characterization. Their physicochemical properties were broadly distributed across the spectrum observed in all predicted anti-enterococcal lysins. Their physicochemical properties were broadly distributed across the spectrum observed in all anti-enterococcal lysins (Fig. 4 ), suggesting that the selected subset encompassed a diverse range of characteristics within the full dataset. These proteins were heterogeneously expressed in E. coli , and semi-purified using metal affinity-based chromatography. Protein yields varied considerably, from 0.14 to 2.5 mg per 100 mL expression culture, with differing purity levels. Such variability is to be expected when expressing proteins identified in metagenomic samples. For example, Fu et al. [ 70 ] reported yields of 0.7–24.3 mg per 100 mL for lysins identified from metagenomic samples. Moreover, Cremelie et al. [ 71 ] highlighted that recombinant production of phage lysins often faces solubility and host-dependent functional limitations. Balaban et al. [ 72 ] similarly reported that only 30–40% of two staphylococcal endolysins could be recovered in the soluble fraction after E. coli expression. These observations point out that lysins are not naturally optimized for high-level expression in heterologous systems, making consistent production difficult. While strategies such as codon optimization, alternative hosts, expression of individual domains, or fusion partners may improve yield, success is not guaranteed. Muralytic activity of the selected lysins Partially purified lysins, exhibiting a wide range of yields and purity levels, were qualitatively screened for muralytic activity using a single dose against three enterococcal species ( Ent. faecium, Ent. faecalis , and Ent. hirae ). Four lysins were active against all three species, two additional lysins showed activity against two species ( Ent. faecalis and Ent. hirae ) and seven were active only against Ent. hirae . The latter strain was likely the most sensitive due to observed partial autolysis activity under experimental conditions, a phenomenon also reported previously [ 73 ]. Despite suboptimal expression and purification yields, 62% of the tested proteins exhibited activity against at least one enterococcal species, supporting the effectiveness of the pipeline in identifying functional anti-enterococcal lysins. Notably, six active lysins (E_24, E_28, E_31, E_65, E_82, E_86) lacked any predicted domain using InterProScan, suggesting that the approach may also uncover lysins with novel domain architectures. In conclusion, this study introduced a sequence-based metagenomic pipeline integrated with embedding-based tools to identify and characterize phage lysins from diverse environmental samples. The approach successfully retrieved a large and diverse set of candidate lysins, including several with confirmed muralytic activity against enterococci. Notably, some identified lysins combined novel domain architectures with demonstrated activity, underscoring the potential of this pipeline to discover lysins with both structural novelty and functional potential. Further, comparative analysis revealed differences in lysin properties between fecal and sewage samples, suggesting distinct ecological sources connected to different lysins. These findings underscore the potential of metagenomics to expand the lysin repertoire and provide insights into their diversity and functionality. However, key challenges remain. Protein yields varied widely, and several lysins exhibited low purity. Additionally, functional assays were limited to three species, single strains, and one condition, restricting conclusions about host range and clinical relevance. Future work should focus on refining predictive algorithms, improving expression systems for scalable production, and expanding functional characterization under diverse conditions. Addressing these challenges will be critical for translating lysin discovery into clinical applications. Declarations Conflict of Interest YB is co-founder and scientific advisor of Obulytix. Competing Interests YB is co-founder and scientific advisor of Obulytix. Funding IP was supported by the Research Foundation – Flanders (FWO) under Research project FWO SB 1SC9424N. RV was supported by the ‘Bijzonder Onderzoeksfonds’ (BOF) Ghent University with a postdoctoral fellowship [01P10022]. LDC was supported by the Research Foundation – Flanders (FWO) with doctoral fellowship [11L1325N]. Author Contribution Conceptualization: IP, RV, YB; Funding acquisition: IP, TVdW, YB; LDC, JM Software: AB, LDC, IP; Formal analysis: IP, LDC, AB, RV; Investigation: IP; Methodology: IP, RV, TVdW, YB; Supervision: RV, TVdW, YB; Visualization: IP; Writing - original draft preparation: IP; Writing - review and editing: all authors. Acknowledgement We thank Maya Merabishvili and Jean-Paul Pirnay (Queen Astrid Military Hospital, QAMH) for providing Enterococcus phage AQEF5 and its host, Enterococcus faecium AW3576. We would like to acknowledge Juan M. Rodríguez from UCM for providing the Ent. hirae ECC1. We are also grateful to Thomas Martens (ILVO), Fëllanza Halimi (Ghent University), Tony Nimmegeers (UZ Gent), and Nils and Dirk Mouton (Bioboerderij De Zwaluw) for sharing the environmental samples that were crucial for this study. Data Availability The datasets for this study can be found in the European Nucleotide Archive: [http://www.ebi.ac.uk/ena/browser/view/PRJEB102709](http:/www.ebi.ac.uk/ena/browser/view/PRJEB102709) . The implementation of the classifier to predict anti-enterococcal lysins can be found on GitHub at: [https://github.com/Rousseau-Team/lysin-target-pred](https:/github.com/Rousseau-Team/lysin-target-pred) . Protein datasets are provided in the supplementary information (Supplementary_files_03). References Piddock LJV, Alimi Y, Anderson J, de Felice D, Moore CE, Rottingen JA et al (2024) Advancing global antibiotic research, development and access. Nat Med 30(9):2432–2443. https://doi.org/10.1038/s41591-024-03218-w Fishbein SRS, Mahmud B, Dantas G (2023) Antibiotic perturbations to the gut microbiome. Nat Rev Microbiol 21(12):772–788. https://doi.org/10.1038/s41579-023-00933-y Sung J, Rajendraprasad SS, Philbrick KL, Bauer BA, Gajic O, Shah A et al (2024) The human gut microbiome in critical illness: disruptions, consequences, and therapeutic frontiers. J Crit Care 79:154436. https://doi.org/10.1016/j.jcrc.2023.154436 Bosnjak M, Karpe AV, Van TTH, Kotsanas D, Jenkin GA, Costello SP et al (2023) Multi-omics analysis of hospital-acquired diarrhoeal patients reveals biomarkers of enterococcal proliferation and Clostridioides difficile infection. Nat Commun 14(1). https://doi.org/10.1038/s41467-023-43671-8 Li P, Chang XL, Chen XY, Wang CA, Shang Y, Zheng DY et al (2022) Early-life antibiotic exposure increases the risk of childhood overweight and obesity in relation to dysbiosis of gut microbiota: a birth cohort study. Ann Clin Microbiol Antimicrob 21(1). https://doi.org/10.1186/s12941-022-00535-1 Xu ZL, Jiang W, Huang WL, Lin Y, Chan FKL, Ng SC (2022) Gut microbiota in patients with obesity and metabolic disorders - a systematic review. Genes Nutr 17(1). https://doi.org/10.1186/s12263-021-00703-6 Dahal RH, Kim S, Kim YK, Kim ES, Kim J (2023) Insight into gut dysbiosis of patients with inflammatory bowel disease and ischemic colitis. Front Microbiol 14. https://doi.org/10.3389/fmicb.2023.1174832 McCallin S, Drulis-Kawa Z, Ferry T, Pirnay JP, Nir-Paz R, Antibacterials E-E S. G. N. (2025). Phages and phage-borne enzymes as new antibacterial agents. Clin Microbiol Infect 31(6):910–921. https://doi.org/10.1016/j.cmi.2023.10.018 Ariaeenejad S, Gharechahi J, Shahraki MF, Atanaki FF, Han JL, Ding XZ et al (2024) Precision enzyme discovery through targeted mining of metagenomic data. Nat Prod Bioprospecting 14(1). https://doi.org/10.1007/s13659-023-00426-8 Santos CA, Morais MAB, Mandelli F, Lima EA, Miyamoto RY, Higasi PMR et al (2025) A metagenomic 'dark matter' enzyme catalyses oxidative cellulose conversion. Nature 639(8056). https://doi.org/10.1038/s41586-024-08553-z Fernández-Ruiz I, Coutinho FH, Rodriguez-Valera F (2018) Thousands of novel endolysins discovered in uncultured phage genomes. Front Microbiol 9. https://doi.org/10.3389/fmicb.2018.01033 Boulay A (2024) sublyme Boulay A (2025) Development of computational tools for the study of bacteriophages and their proteins . master's thesis, Université Laval Elnaggar A, Heinzinger M, Dallago C, Rehawi G, Wang Y, Jones L et al (2022) ProtTrans: Toward understanding the language of life through self-supervised learning. IEEE Trans Pattern Anal Mach Intell 44(10):7112–7127. https://doi.org/10.1109/TPAMI.2021.3095381 Criel B, Taelman S, Van Criekinge W, Stock M, Briers Y (2021) PhaLP: A Database for the Study of Phage Lytic Proteins and Their Evolution. Viruses-Basel 13(7). https://doi.org/10.3390/v13071240 Wagner TM, Poentinen AK, Fenzel CK, Engi D, Janice J, Almeida-Santos AC et al (2024) Interactions between commensal Enterococcus faecium and Enterococcus lactis and clinical isolates of Enterococcus faecium. 5. Fems Microbes https://doi.org/10.1093/femsmc/xtae009 Halimi F, Vanderhaeghen T, Timmermans S, Croubels S, Libert C, Vandewalle J (2024) Comparative analysis of hepatic gene expression profiles in murine and porcine sepsis models. Int J Mol Sci 25(20). https://doi.org/10.3390/ijms252011079 Conceiçao-Neto N, Zeller M, Lefrère H, De Bruyn P, Beller L, Deboutte W et al (2015) Modular approach to customise sample preparation procedures for viral metagenomics: a reproducible protocol for virome analysis. Sci Rep 5. https://doi.org/10.1038/srep16532 Brown T, Tariq MA, Haagmans R, Ansorge R, Telatin A, Adriaenssens E (n.d.). Viromics of faecal samples. https://quadram.ac.uk/viromics-specific-protocols/ De Coninck L, Faye L, Basler N, Jansen D, Van Espen L (2024) Viper. (version 2.2 ed.) Bolger AM, Lohse M, Usadel B (2014) Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30(15):2114–2120. https://doi.org/10.1093/bioinformatics/btu170 Nurk S, Meleshko D, Korobeynikov A, Pevzner PA (2017) metaSPAdes: a new versatile metagenomic assembler. Genome Res 27(5):824–834. https://doi.org/10.1101/gr.213959.116 Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K et al (2009) BLAST plus: architecture and applications. BMC Bioinformatics 10. https://doi.org/10.1186/1471-2105-10-421 Nayfach S, Camargo AP, Schulz F, Eloe-Fadrosh E, Roux S, Kyrpides NC (2021) CheckV assesses the quality and completeness of metagenome-assembled viral genomes. Nat Biotechnol 39(5):578–. https://doi.org/10.1038/s41587-020-00774-7 Vasimuddin M, Misra S, Li H, Aluru S, Ieee (2019) Efficient architecture-aware acceleration of BWA-MEM for multicore systems. In 33rd IEEE International Parallel and Distributed Processing Symposium (IPDPS), Rio de Janeiro, BRAZIL, May 20–24 2019 2019 (pp. 314–324, International Parallel and Distributed Processing Symposium IPDPS). https://doi.org/10.1109/ipdps.2019.00041 Buchfink B, Reuter K, Drost HG (2021) Sensitive protein alignments at tree-of-life scale using DIAMOND. Nat Methods 18(4):366–. https://doi.org/10.1038/s41592-021-01101-x Camargo AP, Roux S, Schulz F, Babinski M, Xu Y, Hu B et al (2024) Identification of mobile genetic elements with geNomad. Nat Biotechnol 42(8). https://doi.org/10.1038/s41587-023-01953-y Hyatt D, Chen GL, LoCascio PF, Land ML, Larimer FW, Hauser LJ (2010) Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC Bioinformatics 11. https://doi.org/10.1186/1471-2105-11-119 Fu LM, Niu BF, Zhu ZW, Wu ST, Li WZ (2012) CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics 28(23):3150–3152. https://doi.org/10.1093/bioinformatics/bts565 Osorio D, Rondón-Villarreal P, Torres R (2015) Peptides: a package for data mining of antimicrobial peptides. R J 7(1):4–14. https://doi.org/10.32614/RJ-2015-001 Steinegger M, Söding J (2017) MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nat Biotechnol 35(11):1026–1028. https://doi.org/10.1038/nbt.3988 Demain AL, Vaishnav P (2009) Production of recombinant proteins by microbes and higher organisms. Biotechnol Adv 27(3):297–306. https://doi.org/10.1016/j.biotechadv.2009.01.008 Francis DM, Page R (2010) Strategies to optimize protein expression in E. coli. Current protocols in protein science. 61(1):5 21-25.24. 29. https://doi.org/10.1002/0471140864.ps0524s61 Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A et al (2024) Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630(8016). https://doi.org/10.1038/s41586-024-07487-w Boulay A, Cremelie E, Galiez C, Briers Y, Rousseau E, Vázquez R (2025) SPAED: harnessing AlphaFold output for accurate segmentation of phage endolysin domains. Bioinformatics 41(10). https://doi.org/10.1093/bioinformatics/btaf531 Swenson JM, Clark NC, Sahm DF, Ferraro MJ, Doern G, Hindler J et al (1995) Molecular characterization and multilaboratory evaluation of Enterococcus faecalis ATCC 51299 for quality control of screening tests for vancomycin and high-level aminoglycoside resistance in enterococci. J Clin Microbiol 33(11):3019–3021. https://doi.org/10.1128/jcm.33.11.3019-3021.1995 Bhatta J (2022) Isolation and characterization of lytic bacteriophages with therapeutic potential for the treatment of bacterial infections. University of Antwerp, University of Antwerp Jiménez E, Ladero V, Chico I, Maldonado-Barragán A, López M, Martín V et al (2013) Antibiotic resistance, virulence determinants and production of biogenic amines among enterococci from ovine, feline, canine, porcine and human milk. BMC Microbiol 13. https://doi.org/10.1186/1471-2180-13-288 Gerstmans H, Grimon D, Gutiérrez D, Lood C, Rodríguez A, van Noort V et al (2020) A VersaTile-driven platform for rapid hit-to-lead development of engineered lysins. Sci Adv 6(23). https://doi.org/10.1126/sciadv.aaz1136 Green R, Rogers EJ (2013) Transformation of chemically competent E. coli. In: Lorsch J (ed) Methods in Enzymology, vol 529. Academic, pp 329–336 Benchling (n.d.). Benchling [Biology software]. https://benchling.com . Accessed 2023–2025 Posit Team (2023) RStudio: Integrated Development Environment for R. (R version 4.3.3 ed). Posit Software, PBC, Boston, MA R Core Team (2023) R: A language and environment for statistical computing. (R version 4.3.3. R Foundation for Statistical Computing, Vienna, Austria, R Foundation for Statistical Computing Wickham H (2016) Data analysis. ggplot2: elegant graphics for data analysis. Springer, pp 189–201 Wickham H (2016) ggplot2: elegant graphics for data analysis. ggplot2: elegant graphics for data analysis. Springer-, New York, pp 189–201 Wickham H, Bryan J, Kalicinski M, Valery K, Leitienne C, Colbert B et al (2019) Readxl: read excel files. (R package version 1.4.5 ed., Vol. 13) Signorell A (2025) (R package version 0.99.59 ed.) Wickham H, François R, Henry L, Müller K, Vaughan D (2023) (R package version 1.1.4 ed.) Müller K, Wickham H (2023) Tibble: Simple data frames. (R package version 3.2.1 ed.) Mair P, Wilcox R (2020) Robust statistical methods in R using the WRS2 package. Behav Researcb Methods 52(2):464–488. https://doi.org/10.3758/s13428-019-01246-w Hothorn T, Bretz F, Westfall P (2008) Simultaneous inference in general parametric models. Biom J 50(3):346–363. https://doi.org/10.1002/bimj.200810425 Wilcox RR, Tian TS (2011) Measuring effect size: a robust heteroscedastic approach for two or more groups. J Applied Statistics 38(7):1359–1368. https://doi.org/10.1080/02664763.2010.498507 Vázquez R, García E, García P (2021) Sequence-function relationships in phage-encoded bacterial cell wall lytic enzymes and their implications for phage-derived product design. J Virol 95(14). https://doi.org/10.1128/JVI.00321-21 Kyte J, Doolittle RF (1982) A simple method for displaying the hydropathic character of a protein. J Mol Biol 157(1):105–132. https://doi.org/10.1016/0022-2836(82)90515-0 Ikai A (1980) Thermostability and aliphatic index of globular proteins. J Biochem 88(6):1895–1898. https://doi.org/10.1093/oxfordjournals.jbchem.a133168 Jones P, Binns D, Chang HY, Fraser M, Li WZ, McAnulla C et al (2014) InterProScan 5: genome-scale protein function classification. Bioinformatics 30(9):1236–1240. https://doi.org/10.1093/bioinformatics/btu031 Schleifer KH, Kandler O (1972) Peptidoglycan types of bacterial cell walls and their taxonomic implications. Bacteriological reviews 36(4):407–477. https://doi.org/10.1128/br.36.4.407-477.1972 Chamakura KR, Tran JS, O'Leary C, Lisciandro HG, Antillon SF, Garza KD et al (2020) Rapid de novo evolution of lysis genes in single-stranded RNA phages. Nat Commun 11(1). https://doi.org/10.1038/s41467-020-19860-0 Chamakura K, Young R (2019) Phage single-gene lysis: Finding the weak spot in the bacterial cell wall. J Biol Chem 294(10):3350–3358. https://doi.org/10.1074/jbc.TM118.001773 Reed CA, Langlais C, Wang IN, Young R (2013) A2 expression and assembly regulates lysis in Qβ infections. Microbiology-Sgm 159:507–514. https://doi.org/10.1099/mic.0.064790-0 Hör J (2025) Advancing RNA phage biology through meta-omics. Nucleic Acids Res 53(8). https://doi.org/10.1093/nar/gkaf314 Urayama SI, Fukudome A, Hirai M, Okumura T, Nishimura Y, Takaki Y et al (2024) Double-stranded RNA sequencing reveals distinct riboviruses associated with thermoacidophilic bacteria from hot springs in Japan. Nat Microbiol 9(2):514–523. https://doi.org/10.1038/s41564-023-01579-5 Székely AJ, Breitbart M (2016) Single-stranded DNA phages: from early molecular biology tools to recent revolutions in environmental microbiology. FEMS Microbiol Lett 363(6). https://doi.org/10.1093/femsle/fnw027 Kosmopoulos JC, Klier KM, Langwig MV, Tran PQ, Anantharaman K (2024) Viromes vs. mixed community metagenomes: choice of method dictates interpretation of viral community ecology. Microbiome 12(1). https://doi.org/10.1186/s40168-024-01905-x Wu LY, Wijesekara Y, Piedade GJ, Pappas N, Brussaard CPD, Dutilh BE (2024) Benchmarking bioinformatic virus identification tools using real-world metagenomic data across biomes. Genome Biol 25(1). https://doi.org/10.1186/s13059-024-03236-4 Georgakis N, Premetis GE, Pantiora P, Varotsou C, Bodourian CS, Labrou NE (2025) The impact of metagenomic analysis on the discovery of novel endolysins. Appl Microbiol Biotechnol 109(1). https://doi.org/10.1007/s00253-025-13513-2 Nie WC, Qiu TY, Wei YW, Ding H, Guo ZX, Qiu JX (2024) Advances in phage-host interaction prediction: in silico method enhances the development of phage therapies. Brief Bioinform 25(3). https://doi.org/10.1093/bib/bbae117 Roux S, Camargo AP, Coutinho FH, Dabdoub SM, Dutilh BE, Nayfach S et al (2023) iPHoP: An integrated machine learning framework to maximize host prediction for metagenome-derived viruses of archaea and bacteria. PLoS Biol 21(4). https://doi.org/10.1371/journal.pbio.3002083 Liu H, Hu Z, Li MY, Yang Y, Lu SG, Rao XC (2023) Therapeutic potential of bacteriophage endolysins for infections caused by Gram-positive bacteria. J Biomed Sci 30(1). https://doi.org/10.1186/s12929-023-00919-1 Fu YR, Yu ST, Li JF, Lao ZS, Yang XF, Lin ZL (2024) DeepMineLys: Deep mining of phage lysins from human microbiome. Cell Rep 43(8). https://doi.org/10.1016/j.celrep.2024.114583 Cremelie E, Vázquez R, Briers Y (2024) A comparative guide to expression systems for phage lysin production. Essays Biochem 68(5):645–659. https://doi.org/10.1042/EBC20240019 Balaban CL, Suárez CA, Boncompain CA, Peressutti-Bacci N, Ceccarelli EA, Morbidoni HR (2022) Evaluation of factors influencing expression and extraction of recombinant bacteriophage endolysins in Escherichia coli. Microb Cell Fact 21(1):40. https://doi.org/10.1186/s12934-022-01766-9 Shockman GD (1992) The autolytic (‘suicidase’) system of Enterococcus hirae: From lysine depletion autolysis to biochemical and molecular studies of the two muramidases of Enterococcus hirae ATCC 9790. FEMS Microbiol Lett 100(1–3):261–267. https://doi.org/10.1111/j.1574-6968.1992.tb14050.x Footnotes 1 Merelbeke: Department of Pathobiology, Pharmacology and Zoological Medicine, Faculty of Veterinary Medicine, Ghent University, 9820 Merelbeke, Belgium. Pigs were part of a pig sepsis model, for which the study design was approved by the Ethical Committee of the Faculty of Veterinary Medicine and the Faculty of Bioscience Engineering of Ghent University (EC 2017/24) [ 17 ] ILVO: Instituut voor Landbouw-, Visserij- en Voedingsonderzoek, Varkenscampus, Van Gansberghelaan 92/1, 9820 Merelbeke-Melle, Belgium UZ Gent: Universitair Ziekenhuis Gent, Corneel Heymanslaan 10, 9000 Ghent, Belgium. Additional Declarations Competing interest reported. YB is co-founder and scientific advisor of Obulytix. Supplementary Files Supplementaryfiles03.7z Supplementaryfile01CodeDomainreductioninterpro.rmd supplementaryfiles02figuresandtables.docx Cite Share Download PDF Status: Under Review Version 1 posted First submitted to journal 10 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-8328502\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":558295453,\"identity\":\"bd6c7729-c389-46cf-a700-31d266914a71\",\"order_by\":0,\"name\":\"Iris Pottie\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Ghent University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Iris\",\"middleName\":\"\",\"lastName\":\"Pottie\",\"suffix\":\"\"},{\"id\":558295454,\"identity\":\"f2420cb9-397c-470f-8193-4fa4f19b6cf8\",\"order_by\":1,\"name\":\"Alexandre Boulay\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Université Laval\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Alexandre\",\"middleName\":\"\",\"lastName\":\"Boulay\",\"suffix\":\"\"},{\"id\":558295455,\"identity\":\"ac46f498-0b98-4fa5-8561-76a08aa8aed6\",\"order_by\":2,\"name\":\"Lander De Coninck\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"KU Leuven\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Lander\",\"middleName\":\"\",\"lastName\":\"De Coninck\",\"suffix\":\"\"},{\"id\":558295456,\"identity\":\"f0473715-98ea-462e-90f9-03fbf30d4c6a\",\"order_by\":3,\"name\":\"Jelle Matthijnssens\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"KU Leuven\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Jelle\",\"middleName\":\"\",\"lastName\":\"Matthijnssens\",\"suffix\":\"\"},{\"id\":558295457,\"identity\":\"bb16cbda-b77f-4630-bbd3-4ae27f727fd2\",\"order_by\":4,\"name\":\"Tom Van de Wiele\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Ghent University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Tom\",\"middleName\":\"Van\",\"lastName\":\"de Wiele\",\"suffix\":\"\"},{\"id\":558295458,\"identity\":\"6bd385f7-011d-4f4c-98a0-7e1660e65fc7\",\"order_by\":5,\"name\":\"Roberto Vázquez\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Ghent University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Roberto\",\"middleName\":\"\",\"lastName\":\"Vázquez\",\"suffix\":\"\"},{\"id\":558295459,\"identity\":\"44d32e38-80fa-47eb-8794-b692e3dc4d53\",\"order_by\":6,\"name\":\"Yves Briers\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIie3OsQrCMBCA4QsFXa51TVHQRygIcRH7Ki2FTuLS0aVTu0TfwPdwVDp0CbgKgiiCk0PERURQqzi4xI6C+YeEg3zkAHS6X43EgMW1eY1YnhjOC5QkRRVaijjLIJNkum7U6vl+iJfuwAVzvlGTMKRERGiP+2xljsMIwQocJRGCAUk8dARWVibPfA7IqJosTrIgrqjuI5PfCtI5K0nOgT5/QWAGnmfPX1QC7Dxh1BceUoFtexIHPs+sQLmYlRk7eZx6bo1Xt/Jw7flpOppLFWnNHof3nkjyOAzlWgDN+GO8fnmu0+l0f9kdCqRGYuirJtYAAAAASUVORK5CYII=\",\"orcid\":\"\",\"institution\":\"Ghent University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Yves\",\"middleName\":\"\",\"lastName\":\"Briers\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-12-10 14:23:19\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-8328502/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-8328502/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":98422426,\"identity\":\"78037aa9-320f-4201-9910-e0417ac75c99\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:02\",\"extension\":\"docx\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":1276840,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Finalmanuscript.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/9d56edfd30962bfa13402102.docx\"},{\"id\":97950933,\"identity\":\"f49dbf95-5f64-4b63-b419-0f066f6d17dc\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:58:59\",\"extension\":\"json\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":9303,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"da0f73793c1a4669968ace29b399fb7e.json\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/ed444d2b6a31c944803b355f.json\"},{\"id\":97950938,\"identity\":\"a08454fd-ad12-4d6c-80ca-6a4becfdd759\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:58:59\",\"extension\":\"rmd\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":5359,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementaryfile01CodeDomainreductioninterpro.rmd\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/a73751c1c0e8f0b702dfd6a4.rmd\"},{\"id\":97950957,\"identity\":\"f115b791-8bd1-446e-89c0-feed70eb443d\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:01\",\"extension\":\"7z\",\"order_by\":3,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":99069933,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementaryfiles03.7z\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/874dc295f3a60651800ded66.7z\"},{\"id\":97950955,\"identity\":\"8f02d8b0-9241-43a1-b4b7-51fe886e50ac\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"docx\",\"order_by\":4,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":20799986,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"supplementaryfiles02figuresandtables.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/08ba508c8f8878eb12a9a664.docx\"},{\"id\":97950951,\"identity\":\"78530a90-cb52-41ed-92b1-c4259a401f13\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"xml\",\"order_by\":5,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":206273,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"da0f73793c1a4669968ace29b399fb7e1enriched.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/d9dfa151303d08ba1cfbca99.xml\"},{\"id\":98422977,\"identity\":\"ecb9b3f8-968e-4040-a081-0cbb155def27\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:42\",\"extension\":\"png\",\"order_by\":6,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":119179,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/26517803533b75cecb25ae2c.png\"},{\"id\":97950941,\"identity\":\"222f1a86-d524-471d-9851-86c8fca4ebd9\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"png\",\"order_by\":7,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":246252,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/1a7d068f5b47fc1cc1548df6.png\"},{\"id\":98422327,\"identity\":\"390af813-a948-4772-8fa3-923ecc50f86d\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:30:50\",\"extension\":\"png\",\"order_by\":8,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":228391,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/57ad8ab7d19b264da3311833.png\"},{\"id\":98423212,\"identity\":\"3db6edf2-0a39-4b75-b6a1-e661b62195da\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:58\",\"extension\":\"png\",\"order_by\":9,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":177598,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/a0cb730a9d58bc2d288e5d22.png\"},{\"id\":98423796,\"identity\":\"e08a4363-d328-4d03-b668-395127f99701\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:32:38\",\"extension\":\"png\",\"order_by\":10,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":115769,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"floatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/51081a6b973c72d9dc9d723b.png\"},{\"id\":98422531,\"identity\":\"5488b068-ae40-4758-9f76-6812de34eeed\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:10\",\"extension\":\"png\",\"order_by\":11,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":38251,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/6c060f6c2013b4d966edd8b1.png\"},{\"id\":97950945,\"identity\":\"72a19df9-36b4-481e-9bf7-a7f99daab916\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"png\",\"order_by\":12,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":98285,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/10b41fa13f1b39a88d41d21d.png\"},{\"id\":97950949,\"identity\":\"78297862-09af-43ef-b701-f315e43ff163\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"png\",\"order_by\":13,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":78761,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/96653c08bd21e17096f868e2.png\"},{\"id\":98422389,\"identity\":\"ba720eaf-2430-4405-a283-0c94dc2b4089\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:30:57\",\"extension\":\"png\",\"order_by\":14,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":59490,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/8ca187e950014778f8a1c77d.png\"},{\"id\":98422485,\"identity\":\"eaf77d2d-962a-449b-b07e-abda6ccb525c\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:06\",\"extension\":\"png\",\"order_by\":15,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":35600,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Onlinefloatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/b3fd2ff24a79b388d8413fff.png\"},{\"id\":97950953,\"identity\":\"4d83504c-9bd0-46ee-af8e-a7ef4106a743\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"xml\",\"order_by\":16,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":204554,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"da0f73793c1a4669968ace29b399fb7e1structuring.xml\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/2e79430ccf8ead626cee5bb9.xml\"},{\"id\":98422555,\"identity\":\"51a0fd19-1ee4-4626-a4ec-9bff5e7689a2\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:12\",\"extension\":\"html\",\"order_by\":17,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"acdc-reference\",\"size\":221776,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"earlyproof.html\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/b30d1e0f847ec7a90091c6c7.html\"},{\"id\":98422455,\"identity\":\"6f6a14e7-a22f-4b4c-9c5b-afa5644c7b3a\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:04\",\"extension\":\"jpg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":31767,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eWorkflow for training a model to predict potential anti-enterococcal lysins Two hundred models were trained using all 237 enterococcal-targeting lysins from the PhaLP database, and 500 randomly sampled non-enterococcal lysins. Models with precision \\u0026gt; 80 % were retained for prediction, while those below this threshold were discarded. Predictions were aggregated based on prediction support (number of models classifying a lysin as anti-enterococcal): lysins with high prediction support (\\u0026gt; 80 %) were likely anti-enterococcal, those with moderate support (30-80 %) were less likely, and those with little or no support (\\u0026lt; 50 %) were considered unlikely anti-enterococcal. In this paper, lysins were considered to be likely enterococcal when 80 % of the models predicted it to be likely anti-enterococcal\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture1.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/6c9a804f938b67ffae6fced0.jpg\"},{\"id\":98422976,\"identity\":\"0855e12a-8ad1-4051-8028-5fafd739116e\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:42\",\"extension\":\"jpg\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":62084,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePhysicochemical metrics of predicted lysins across libraries. Libraries L1-L3 originated from pig feces, L4 and L5 from wastewater, “all” represents a non-redundant set of lysins from all libraries combined. Metrics include: (A) aliphatic index (aindex). (B) grand average of hydropathy (GRAVY). (C) length in number of amino acids (No. AA). (D) molecular weight (MW) in kilodalton (kDa), and (E) isoelectric point (pI). Differences in properties were compared across libraries (L1-L5; the pooled lysins not included). Significance was determined with the Kruskal-Wallis test (\\u003cem\\u003eP\\u003c/em\\u003e \\u0026lt; 0.001: ***), and significant differences between libraries as determined by a post-hoc Dunn’s test (with Holm correction) are displayed as letters. Effect sizes (ES, corresponding to Wilcox and Tian’s ζ ) were estimated using a heteroskedastic one-way ANOVA on trimmed means.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture2.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/7fde9eb25b775788ef3372b8.jpg\"},{\"id\":98421966,\"identity\":\"7a2d7653-6c84-41d0-aac8-03b387cf01f6\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:30:03\",\"extension\":\"jpg\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":84042,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eDistribution of enzymatically active domains (EADs) and cell wall-binding domains (CBDs) across lysin libraries. (a) Relative proportions of proteins (in %) assigned to specific domain clusters, based on proteins with at least one predicted domain; proteins lacking domain annotations are excluded. Unclassified: proteins containing a domain that was not included in the classification of Criel et al. [15]. Predicted catalytic activities are indicated in brackets: [Ami] Amidase: \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-l-alanine amidase. [Ami + Pet] amidase and peptidase: \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-l-alanine amidase and peptidase. [Pet] Peptidase. [GA], glucosaminidase: \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-d-glucosaminidase [MUR], muramidase: \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-d-muramidase, [Mur + TG], muramidase, transglycosylase: \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-d-muramidase, lytic transglycosylase, [TG] transglycosylase: lytic transglycosylase. (b) Standardized Pearson residuals indicating variation in CBD cluster representation across libraries, expressed as the percentage of each CBD cluster relative to the total number of CBDs. (c) Relative distribution of predicted lytic activities among EADs per library, shown as the percentage of proteins assigned to each activity out of all proteins with a predicted enzymatic function\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture3.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/4ce2e24905b7e63a7d9480d9.jpg\"},{\"id\":98423185,\"identity\":\"b930dea9-0fba-4e7d-b43f-d3f6f31559bb\",\"added_by\":\"auto\",\"created_at\":\"2025-12-17 16:31:55\",\"extension\":\"jpg\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":52967,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePhysicochemical properties of lysins predicted to be anti-enterococcal (n = 129), compared to the remaining lysins (n = 8 696). These metrics include (a) aliphatic index (aindex), (b) grand average of hydropathy (GRAVY), (c) length in number of amino acids (No. AA), (d) molecular weight (MW, in kDa), and (e) isoelectric point (pI). Horizontal lines represent the median of the distributions. Asterisks represent the \\u003cem\\u003eP\\u003c/em\\u003e-value as determined by a Wilcoxon test (\\u003cem\\u003eP \\u003c/em\\u003e\\u0026lt; 0.001: ***). The effect size (ES) is Wilcox and Tian’s ζ\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture4.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/452f670a6fcea6283b537547.jpg\"},{\"id\":97950935,\"identity\":\"7fee1cec-b0ab-43b3-898e-7e45dfd41185\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:58:59\",\"extension\":\"jpg\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":48944,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eMuralytic activity of 21 putative enterococcal phage lysins. Muralytic activity was tested in a turbidity reduction assay (TRA) against three distinct species: \\u003cem\\u003eEnt. faecium\\u003c/em\\u003eAW3576, \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e ECC1, and \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e NJ-3. The area between the curve (ABC) of each treatment and the negative control was calculated and used as metric for muralytic activity. Asterisks represent the significance as determined with a one-way ANOVA (analysis of variance) and \\u003cem\\u003epost-hoc\\u003c/em\\u003eDunnett’s test (\\u003cem\\u003eP \\u003c/em\\u003e\\u0026lt; 0.001: ***, P \\u0026lt; 0.01: **, P \\u0026lt; 0.05: *)\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture5.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/f16c9093ee9e9379e96f66e3.jpg\"},{\"id\":98774617,\"identity\":\"d496ac25-2ad6-457f-aabb-ecc1794398e1\",\"added_by\":\"auto\",\"created_at\":\"2025-12-22 12:05:11\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1359722,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/fbcc8002-bfbc-4429-be07-ba27748acbac.pdf\"},{\"id\":97950958,\"identity\":\"8e5faaea-4f06-4dba-ad5e-38aee6b2e68d\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:03\",\"extension\":\"7z\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":99069933,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementaryfiles03.7z\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/cd8a08f1bc52c7691e4ee92b.7z\"},{\"id\":97950942,\"identity\":\"37ebbc25-4f43-4727-becb-6abec47dac40\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"rmd\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":5359,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementaryfile01CodeDomainreductioninterpro.rmd\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/c6c577815f2f12114262a55c.rmd\"},{\"id\":97950956,\"identity\":\"97f32bd6-2f6a-4404-a079-3bc3aa80a50c\",\"added_by\":\"auto\",\"created_at\":\"2025-12-11 06:59:00\",\"extension\":\"docx\",\"order_by\":2,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":20799986,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"supplementaryfiles02figuresandtables.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8328502/v1/c72e7c2891d404f07636f216.docx\"}],\"financialInterests\":\"Competing interest reported. YB is co-founder and scientific advisor of Obulytix.\",\"formattedTitle\":\"Discovery of anti-enterococcal phage lysins from environmental metagenomes using protein embedding-based classification\",\"fulltext\":[{\"header\":\"1 Introduction\",\"content\":\"\\u003cp\\u003eThe rise of bacteria acquiring resistance against antibiotics renders many currently used antibiotic treatments ineffective, thereby causing a considerable threat to global health. Misuse and overuse of antibiotics further exacerbate antibiotic resistance. This growing resistance crisis coincides with a stagnating antibiotic development pipeline, creating what has been described as a \\u0026ldquo;perfect storm\\u0026rdquo; [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e]. Moreover, broad-spectrum antibiotics impact the delicate balance of the human microbiota, a diverse community of microorganisms essential for various physiological functions [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e]. This microbiota disruption (\\u0026ldquo;dysbiosis\\u0026rdquo;) has been associated with health complications, including gastrointestinal disorders [\\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e], obesity [\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e], and autoimmune diseases [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eTo address these challenges, there is an urgent need for novel antimicrobial strategies that are both effective and specific. Among the most promising candidates are phage lysins, enzymes derived from bacteriophages that selectively degrade bacterial cell walls. These enzymes selectively target and degrade the bacterial cell wall of specific bacteria, resulting in rapid lysis, while rendering other non-targeted bacteria of the microbiome unaffected. Their unique mechanism of action, together with their vast diversity and engineering possibilities to further finetune their specificity and efficacy differentiate them from traditional antibiotic classes. However, a key bottleneck in advancing lysins from promising research tools to effective therapeutics is the early-stage identification and collection of candidate lysins and their building blocks [\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eMetagenomics, the study of genetic material recovered directly from environmental samples, offers a powerful solution to this bottleneck [\\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e9\\u003c/span\\u003e]. Classical sequence homology-based metagenomics typically relies on sequence homology searches to mine environmental datasets and has already enabled the discovery of diverse enzymes [\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e]. For example, Fern\\u0026aacute;ndez-Ruiz et al. [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e] identified 2 628 endolysins from 183 289 uncultured viral genomes using a sequence homology-based pipeline. Hits meeting thresholds for identity, coverage, alignment length, and E-value were considered as putative lysins. Many of these hits exhibited novel domain architectures, highlighting the potential of metagenomics to feed the lysin hit-to-lead pipeline with candidates possessing unexpected traits which are of value for combinatorial engineering. However, this approach inherently overlooks novel lysins that lack detectable similarity to known lysins.\\u003c/p\\u003e \\u003cp\\u003eAdvances in computational and artificial intelligence methods have leveraged the discovery of novel enzymes through sequence-based approaches. Recently, the SUBLYME (Software for Uncovering Bacteriophage LYsins in MEtagenomic datasets) tool has been developed to identify lysins from metagenomic datasets [\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e]. Protein embeddings generated by language models such as ProtT5 were used to create vector representations of lysins [\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e]. These representations served as input for support vector machines (SVMs), which were trained on a curated dataset from PhaLP [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e], a comprehensive lysin database (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://phalp.ugent.be\\u003c/span\\u003e\\u003cspan address=\\\"https://phalp.ugent.be\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e), to sequentially identify and classify lysins as virion-associated lysins (VALs) or endolysins. When applied to the same database as the previously mentioned study of Fern\\u0026aacute;ndez-Ruiz et al. [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e], 50 856 putative lysins were predicted, including 98% of the previously identified lysins.\\u003c/p\\u003e \\u003cp\\u003eIn this study, we applied a sequence-based metagenomic pipeline to discover novel lysins with activity against \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species, opportunistic pathogens that exemplify the need for targeted antimicrobials. \\u003cem\\u003eEnterococcus faecalis\\u003c/em\\u003e and \\u003cem\\u003eEnterococcus faecium\\u003c/em\\u003e, typically commensals of the human gastrointestinal tract, can become pathogenic under conditions such as antibiotic treatment, immunosuppression, or gut barrier disruption. They are now recognized as major contributors to hospital-acquired infections, including urinary tract infections, bacteremia, endocarditis, and wound infections. Their intrinsic resistance to many antibiotics and capacity to acquire additional resistance genes makes treatment increasingly difficult [\\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e16\\u003c/span\\u003e]. To address this, we extracted and sequenced metagenomic DNA from five environmental samples (originating from pig feces and sewage). Protein-coding regions were predicted from assembled contigs, and candidate lysins were identified using SUBLYME. A second SVM-based classifier was trained to predict lysins with potential activity against \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species (hereafter referred to as anti-enterococcal lysins). A subset of 21 predicted lysins was expressed and tested for muralytic activity against \\u003cem\\u003eEnterococcus hirae\\u003c/em\\u003e, \\u003cem\\u003eEnt. faecium\\u003c/em\\u003e, and \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e, revealing promising candidates for further development.\\u003c/p\\u003e\"},{\"header\":\"2 Materials and methods\",\"content\":\"\\u003cp\\u003eSample collection and virion isolation\\u003c/p\\u003e \\u003cp\\u003eFive different samples were collected, all originating from pools of pig feces or hospital sewage (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e). Samples were either processed immediately (hospital sewage) or stored at -20\\u0026deg;C until further use (pig feces).\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eOverview of the characteristics of the samples for which environmental DNA was obtained.\\u003c/b\\u003e For each sample, the collection date, source, number of stool samples and subsequent library is depicted\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"4\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e Sample Name\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eCollection date\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eSource\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eNumber of stool samples/ volume of sewage\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary 1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e2024/02/23\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003ePig feces (Merelbeke\\u003csup\\u003e1\\u003c/sup\\u003e)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e14 samples\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary 2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e2024/02/26\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003ePig feces (ILVO\\u003csup\\u003e2\\u003c/sup\\u003e)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e10 samples\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary 3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e2024/03/08\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003ePig feces (ILVO\\u003csup\\u003e2\\u003c/sup\\u003e)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e20 samples\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary 4\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e2024/04/13\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eHospital sewage UZ Gent\\u003csup\\u003e3\\u003c/sup\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e1 L\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary 5\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e2024/04/16\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eHospital sewage UZ Gent\\u003csup\\u003e3\\u003c/sup\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e1 L\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003eVirus purification and concentration were performed according to the methods used at the Quadram institute and the NetoVIR protocol [\\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e18\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e19\\u003c/span\\u003e]. For fecal samples, approximately 4 g aliquots were prepared. Sterile phosphate buffered saline [PBS, pH 7.4; 137 mM NaCl (Thermo Scientific), 2.7 mM KCl (Carl Roth), 10 mM NaH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e (Thermo Scientific), 1.8 mM KH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e (Carl Roth)] was added to each aliquot in a 1:10 (w/v) ratio, followed by vortexing for at least 1 min until samples were homogeneous. Samples were then incubated on ice for 1 h to promote viral particle release and centrifuged (3 000 \\u0026times; g, 5 min at 4\\u0026deg;C) to remove debris. The supernatant was transferred to a new tube and subjected to consecutive centrifugation steps (20 min at 7 196 \\u0026times; g or 15 min at 10 000 \\u0026times; g, both at 4\\u0026deg;C) until full clarity was achieved, as assessed by naked-eye inspection. The clear supernatant was filtered using a 0.45 \\u0026micro;m polyethersulfone bottle top filter (Thermo Scientific). Next, viruses were precipitated using polyethylene glycol (PEG). A solution containing 1 M NaCl and 25% (w/v) PEG 8000 was added to the filtered phage lysate to reach a final PEG concentration of 7.5% (e.g., 15 mL solution per 35 mL of lysate), followed by incubation on ice for 16 h and centrifugation (7 196 \\u0026times; g, 40 min at 4\\u0026deg;C). The resulting pellets were dissolved in sterile PBS.\\u003c/p\\u003e \\u003cp\\u003eFor sewage samples, initial centrifugation was performed at 3 000 \\u0026times; g for 5 min, followed by a second centrifugation step of the supernatant at 3 220 \\u0026times; g for 1 h (library 4) or at 17 000 g for 3 min (library 5). All centrifugations were performed at 4\\u0026deg;C. The clear supernatant was filtered using a 0.45 \\u0026micro;m polyethersulfone filter. PEG precipitation was achieved by adding NaCl to a final concentration of 1 M, dissolving it with a magnetic stirrer, and centrifuging at 3 200 \\u0026times; g. PEG 8000 was added to the supernatant to a final concentration of 10% (w/v), and samples were incubated on ice for 16 h, followed by centrifugation at 7 196 \\u0026times; g for 40 min or 12 000\\u0026ndash;17 000 \\u0026times; g for 30 min, both at 4\\u0026deg;C. Pellets were dissolved in PBS (library 4) or immediately in 1 \\u0026times; DNase buffer (library 5).\\u003c/p\\u003e \\u003cp\\u003eDNA extraction\\u003c/p\\u003e \\u003cp\\u003eTwo preparatory steps were necessary before extracting viral DNA. First, host and environmental DNA and RNA were removed to minimize background interference; second, viral capsids were digested to access the viral genomic material. To remove contaminant RNA and DNA, phage suspensions were treated for 90 min at 37\\u0026deg;C with 100 U/mL RNaseA (Thermo Scientific) and 2 U/mL DNaseI (Thermo Scientific) in DNaseI buffer (Thermo Scientific). DNaseI and RNaseA were inactivated by adding 20 mM ethylenediaminetetraacetic acid (EDTA; Acros Organics). Subsequently, viral capsids were disrupted by adding 0.5% (w/v) sodium dodecyl sulfate (SDS; Merck) and 50 \\u0026micro;g/mL proteinase K (Thermo Scientific), followed by 3 h incubation at 56\\u0026deg;C. DNA was then extracted by adding an equal volume of phenol:chloroform (1:1 ratio, phenol at pH 8.0 from Carl Roth, chloroform from VWR), gently shaking for 1 min, incubating 1 min, followed by centrifugation (3 750 \\u0026times; g, 2 min). The upper aqueous phase was recovered and subjected to consecutive phenol:chloroform extraction steps until the middle, white layer disappeared. An equal volume of chloroform was then added, followed by gentle shaking for 1 min and centrifugation (3 750 \\u0026times; g, 2 min). The upper aqueous phase was recovered and 0.8 volume 100% (v/v) isopropanol (Carl Roth), 0.1 volume 3 M sodium acetate (VWR), and 40 \\u0026micro;g glycogen (Thermo Scientific) were added to the aqueous phase, followed by incubation at 4\\u0026deg;C for 16 h. DNA was collected by centrifugation (10 000 \\u0026times; g, 15 min at 4\\u0026deg;C), washed twice with 70% (v/v) ethanol (VWR), and air-dried. Finally, the DNA was resuspended in TE buffer [10 mM Tris (Carl Roth), 1 mM EDTA, pH 8.0]. DNA quality and concentration were assessed using a microvolume spectrophotometer (Denovix) and a Quantus Fluorometer.\\u003c/p\\u003e \\u003cp\\u003eDNA sequencing\\u003c/p\\u003e \\u003cp\\u003eMetagenomic DNA was sequenced by Eurofins Genomics, using the INVIEW METAGENOME service. This included standard genomic library preparation, followed by paired-end Illumina sequencing (2 \\u0026times; 150 bp), with guaranteed 10 M read pairs. Raw reads and contigs were deposited in the European Nucleotide Archive with study number PRJEB102709.\\u003c/p\\u003e \\u003cp\\u003eAnalysis of sequencing results\\u003c/p\\u003e \\u003cp\\u003eRaw sequencing reads were processed with the ViPER v2.2 pipeline (Virome Paired-End Reads) to generate viral contigs [\\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e20\\u003c/span\\u003e]. This pipeline applies Trimmomatic v0.39 [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e] to trim low-quality bases and adapters from raw reads, followed by assembling the paired-end reads of 1%, 10% and 100% subsets into contigs using metaSPAdes v3.15.5 [\\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e]. These contigs were clustered using a combination of BLAST\\u0026thinsp;+\\u0026thinsp;v2.14.0 [\\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e23\\u003c/span\\u003e], and anicalc.py and aniclust.py scripts from the CheckV v1.0.1 packages [\\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e]. Contigs larger than 500 bp were retained for further analyses. Contig completeness was assessed using CheckV v1.0.1. Abundance tables were then generated in the ViPER pipeline by mapping sequencing reads to each contig with bwa-mem2 v2.2.1 [\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e], thereby quantifying the number of reads aligned per contig. Taxonomic classification was performed in the ViPER pipeline with DIAMOND blastx v2.1.8 [\\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e26\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eThe geNomad pipeline v1.8.0 (database version v1.7) was subsequently employed [\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e]. GeNomad uses pyrodigal-gv, a Cython binding of Prodigal [\\u003cspan citationid=\\\"CR28\\\" class=\\\"CitationRef\\\"\\u003e28\\u003c/span\\u003e], to predict coding sequences and generate protein annotations. These predicted proteins were used in subsequent analyses, including the identification of potential phage lysins.\\u003c/p\\u003e \\u003cp\\u003eBoth the geNomad and the ViPER pipeline classify contigs. The geNomad pipeline was applied to assign contigs as viral, proviral, plasmid, or unknown [\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e]. This classification served as the basis for subsequent steps. Contigs labeled as unknown and those labeled as proviral were taxonomically classified using DIAMOND blastx within ViPER. Relative abundance across taxonomic kingdoms was inferred from read-mapping data generated by ViPER.\\u003c/p\\u003e \\u003cp\\u003ePrediction of phage lytic enzymes\\u003c/p\\u003e \\u003cp\\u003eThe proteins obtained by pyrodigal-gv were used as input sequences for SUBLYME v1.1, a protein-embedding-based classifier that predicts lysins and classifies them as endolysins or VALs [\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003ePredicted lysins were clustered using CD-HIT at 98% sequence identity to remove redundancy [\\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e29\\u003c/span\\u003e]. This reduction was performed separately for each library to preserve intra-library diversity, and subsequently on the combined dataset to retain only non-redundant sequences. For clarity, this final pooled and re-clustered set of lysins is hereafter referred to as \\u0026ldquo;all lysins\\u0026rdquo;.\\u003c/p\\u003e \\u003cp\\u003eCharacterization of lysins\\u003c/p\\u003e \\u003cp\\u003eRelevant physicochemical properties of the predicted lysins were calculated using the Peptides package in R [\\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e30\\u003c/span\\u003e]. These included the pI, molecular weight (MW), grand average of hydropathy (GRAVY) index, and aliphatic index, all calculated with default parameters. Domain annotation was performed with InterProScan on the Galaxy server (Galaxy Version 5.59-91.0\\u0026thinsp;+\\u0026thinsp;galaxy3). InterProScan integrates eighteen member databases, but only four were retained for analysis: Pfam, CDD, SMART and PROSITE. The remaining databases were excluded based on prior findings indicating that they often predicted overly broad, overly specific, or irrelevant motifs for phage lytic proteins [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e]. To address overlapping domain predictions, a custom R function (code in \\u003cb\\u003eSupplementary_information_01\\u003c/b\\u003e) was used. Domains predicted by the same database were all retained without filtering. For proteins with multiple domains predicted by different databases, pairwise overlaps were assessed by calculating the maximum percentage overlap between domain intervals (start and end positions). Overlaps greater than 80% were considered significant. To resolve these overlaps, a filtering strategy was applied: Pfam predictions were given priority, and any non-Pfam domain overlapping with a Pfam domain was removed. In the absence of Pfam domains, overlapping domains were filtered to retain only one representative (the first in the list, sorted by start position), while preserving the non-overlapping domains. The final output consisted of filtered domain data.\\u003c/p\\u003e \\u003cp\\u003eFunctional domains were annotated and classified following the framework proposed by Criel et al. [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e], who identified lysin domain clusters in the PhaLP database and developed a classification system based on the predicted domain accession numbers (\\u003cb\\u003eSupplementary Table\\u0026nbsp;1\\u003c/b\\u003e). Specifically, domains were classified according to their type into cell wall-binding domains (CBD) and enzymatically active domains (EADs). These were subdivided into 13 distinct CBD clusters and 26 EAD types. Additionally, EAD clusters were further grouped based on enzymatic activity, following the classification proposed earlier [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e]. These seven groups included: (1) \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase, (2) transglycosylase, (3) combined \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase and transglycosylase, (4) \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-glucosaminidase, (5) \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidase, (6) peptidase, and (7) combined \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidase and peptidase activity. Domains not included in the classification of Criel et al. [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e] were omitted.\\u003c/p\\u003e \\u003cp\\u003ePredicting and selection of anti-enterococcal phage lysins\\u003c/p\\u003e \\u003cp\\u003eTo functionally characterize a subset of the predicted lysins against enterococci, a second tool was developed to identify lysins potentially active against enterococci. To this end, SVMs with radial basis function kernels and balanced class weights (inversely proportional to class frequencies) were trained on protein embeddings calculated using ProtT5 [\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e]. The implementation is available on GitHub at \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://github.com/Rousseau-Team/lysin-target-pred\\u003c/span\\u003e\\u003cspan address=\\\"https://github.com/Rousseau-Team/lysin-target-pred\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/p\\u003e \\u003cp\\u003eThe PhaLP database (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://phalp.ugent.be\\u003c/span\\u003e\\u003cspan address=\\\"https://phalp.ugent.be\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e) was used to train the models, containing 237 anti-enterococcal lysins (from phages infecting \\u003cem\\u003eEnterococcus\\u003c/em\\u003e) and 15 150 lysins associated with other hosts [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e]. Rather than training one model on the whole dataset, which possesses almost 64 times more negative (associated to other hosts) than positive (anti-enterococcal) examples, 200 smaller models were trained using all 237 anti-enterococcal lysins combined with random subsets of 500 non-enterococcal lysins (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e). This approach ensured that each model was less biased toward the negative class while collectively covering the entire negative dataset. For evaluation, a stratified test set comprising 20% of the positive (~\\u0026thinsp;45/237) and negative (~\\u0026thinsp;100/500) examples was set aside, with the remaining 80% used for training. These training and testing sets were created based on sequence-level clusters at 30% sequence identity and 70% coverage using MMseqs2 to ensure that similar proteins were kept in either the training or the testing set [\\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e31\\u003c/span\\u003e]. This strategy allowed assessment of model generalization to unseen proteins. Models achieving a precision greater than 80% on the test set were retained, resulting in 79 of 200 models. All retained models were then used on the newly identified lysins to predict their potential anti-enterococcal activity. Lysins classified as anti-enterococcal by at least 80% of the retained models were selected for further analysis.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eProteins exceeding 550 amino acids were excluded during manual curation to increase the likelihood of successful heterologous expression and purification [\\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e32\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e]. Subsequently, and an uninformed random selection of 21 proteins was performed. The relevant metrics were calculated as indicated before. The three-dimensional structures of the selected proteins were predicted with AlphaFold3 [\\u003cspan citationid=\\\"CR34\\\" class=\\\"CitationRef\\\"\\u003e34\\u003c/span\\u003e]. The Segmentation of PhAge Endolysin Domains using Alphafold\\u0026rsquo;s PAE matrix (SPAED; \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://spaed.ca\\u003c/span\\u003e\\u003cspan address=\\\"https://spaed.ca\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e) algorithm was used to delineate protein domains [\\u003cspan citationid=\\\"CR35\\\" class=\\\"CitationRef\\\"\\u003e35\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eBacterial strains and growth medium\\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eEscherichia coli\\u003c/em\\u003e BL21(DE3)pLysS was used as a strain for protein expression. \\u003cem\\u003eE. coli\\u003c/em\\u003e was grown for 18 h at 37\\u0026deg;C, shaking at 180 rpm in lysogeny broth (LB), or static on LB containing 1.5% agar (VWR). For cryopreservation, 50% and 20% glycerol stocks were made for respective storage at -20\\u0026deg;C and \\u0026minus;\\u0026thinsp;80\\u0026deg;C. Kanamycin (Carl Roth) at 50 \\u0026micro;g/mL was used as selection marker to select for colonies with the inserted plasmids, 25 \\u0026micro;g/mL chloramphenicol (Acros) was used for \\u003cem\\u003eE. coli\\u003c/em\\u003e BL21(DE3)pLysS.\\u003c/p\\u003e \\u003cp\\u003eA panel of three \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species was selected for this study: \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e NJ-3 (ATCC 51299), obtained from the LGC collection [\\u003cspan citationid=\\\"CR36\\\" class=\\\"CitationRef\\\"\\u003e36\\u003c/span\\u003e]; \\u003cem\\u003eEnt. faecium\\u003c/em\\u003e AW3576, kindly provided by Maia Merabishvili (Queen Astrid Military Hospital) [\\u003cspan citationid=\\\"CR37\\\" class=\\\"CitationRef\\\"\\u003e37\\u003c/span\\u003e]; and \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e ECC1, provided by Juan M. Rodr\\u0026iacute;guez (Universidad Complutense de Madrid) [\\u003cspan citationid=\\\"CR38\\\" class=\\\"CitationRef\\\"\\u003e38\\u003c/span\\u003e]. All strains were cultured at 37\\u0026deg;C in brain heart infusion (BHI) broth (Merck) or on BHI agar under static conditions.\\u003c/p\\u003e \\u003cp\\u003eDNA manipulation of anti-enterococcal lysins\\u003c/p\\u003e \\u003cp\\u003eThe selected protein sequences were converted to coding sequences, adapted for \\u003cem\\u003eE. coli\\u003c/em\\u003e codon usage, by Twist Bioscience. These insert sequences were then provided by flanking position markers and BsaI recognition sites and synthesized by Twist Bioscience (\\u003cb\\u003eSupplementary Table\\u0026nbsp;2\\u003c/b\\u003e). Synthetic genes were diluted to a final concentration of 46 nM in ultrapure water. For the assembly reaction, 1 \\u0026micro;L of the diluted gene was combined with 1 \\u0026micro;L of 25 nM pVTD3 vector [\\u003cspan citationid=\\\"CR39\\\" class=\\\"CitationRef\\\"\\u003e39\\u003c/span\\u003e], 1 \\u0026micro;L of 46 nM C-terminal hexahistidine tag (His-tag) containing position marker sequences for directional cloning (5\\u0026prime;\\u0026ndash;GGA AGC\\u0026ndash;3\\u0026prime; overlap with the synthetic gene and 5\\u0026prime;\\u0026ndash;ATA CTT\\u0026ndash;3\\u0026prime; overlap with the vector), 2 \\u0026micro;L of 10 \\u0026times; T4 DNA ligase buffer (Thermo Scientific), 3.5 U T4 DNA ligase (Thermo Scientific), and 10 U of Eco31I (Thermo Scientific). Ultrapure water was added to bring the final reaction volume to 20 \\u0026micro;L. The reaction was subjected to 80 thermal cycles alternating between 5 min restriction at 37\\u0026deg;C and 5 min ligation at 22\\u0026deg;C, followed by enzyme inactivation at 80\\u0026deg;C for 5 min. Reaction products were purified using the GeneJET PCR Purification Kit (Thermo Scientific), according to the manufacturer\\u0026rsquo;s instructions. Subsequently, 5 \\u0026micro;L of the purified product was used to transform chemically competent \\u003cem\\u003eE. coli\\u003c/em\\u003e TOP10 or \\u003cem\\u003eE. coli\\u003c/em\\u003e BL21(DE3) pLysS cells, prepared using the rubidium chloride method [\\u003cspan citationid=\\\"CR40\\\" class=\\\"CitationRef\\\"\\u003e40\\u003c/span\\u003e]. Transformation mixtures were plated on LB plates with the required antibiotics, supplemented with 5% (w/v) sucrose (Thermo Scientific) as a negative selection marker.\\u003c/p\\u003e \\u003cp\\u003eTo verify the presence of the correct constructs, plasmids were isolated from a single bacterial colony using the GeneJET Plasmid Miniprep Kit (Thermo Scientific), following the manufacturer\\u0026rsquo;s protocol. Sanger sequencing was performed by LGC Genomics using the T7 promoter primer (5\\u0026prime;-TAA TAC GAC TCA CTA TAG GG-3\\u0026prime;) and the T7 terminator primer (5\\u0026prime;-GCT AGT TAT TGC TCA GCG G-3\\u0026prime;). Sequencing results were analyzed using Benchling [\\u003cspan citationid=\\\"CR41\\\" class=\\\"CitationRef\\\"\\u003e41\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eProtein expression and purification\\u003c/p\\u003e \\u003cp\\u003eProtein expression was performed in terrific broth [0.4% glycerol (Acros Organics), 1.2% tryptone, 2.4% yeast extract, 1.7 mM KH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e (Carl Roth) and 7.2 mM K\\u003csub\\u003e2\\u003c/sub\\u003eHPO\\u003csub\\u003e4\\u003c/sub\\u003e (Carl Roth)] for 72 h at 16\\u0026deg;C, shaking at 180 rpm.\\u003c/p\\u003e \\u003cp\\u003eAll previously selected proteins, together with PlyEF501 (UKM17465; an endolysin of \\u003cem\\u003eEnterococcus\\u003c/em\\u003e phage AQEF5 as positive control protein), were expressed in 100 mL. Overnight precultures from single colonies were diluted 1:25 in terrific broth, grown to mid-exponential phase [optical density (OD\\u003csub\\u003e600\\u003c/sub\\u003e) between 0.45\\u0026ndash;0.55], and induced by adding isopropyl β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-1-thiogalactopyranoside (IPTG; Thermo Scientific) to a final concentration of 0.5 mM. Cultures were incubated at 16\\u0026deg;C and 180 rpm for 72 h, cells were harvested by centrifugation (10 000 \\u0026times; g, 10 min, 4\\u0026deg;C) and resuspended in 1 mL lysis buffer [20 mM imidazole (Carl Roth), 50 mM NaH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e-NaOH (Thermo Scientific), 300 mM NaCl, pH 7.4, supplemented with 10 \\u0026micro;g/mL DNaseI from bovine pancreas (Sigma)]. Cell suspensions were disrupted with three freeze-thaw cycles, followed by sonication at 60% (3 min with 15 s on/off cycles). The resulting cell lysates were centrifuged at 10 000 \\u0026times; g for 40 min at 4\\u0026deg;C, 300 \\u0026micro;L HisPur Ni-NTA Superflow Agarose beads (Thermo Scientific) were added to the supernatant and incubated on an end-over-end shaker at 4\\u0026deg;C for 16 h.\\u003c/p\\u003e \\u003cp\\u003eMixtures were centrifuged (700 \\u0026times; g, 2 min, 4\\u0026deg;C), the supernatant was removed, and the pellets were washed twice with 1.5 mL wash buffer (50 mM imidazole, 50 mM NaH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e-NaOH, 300 mM NaCl, pH 7.4). Elution was performed twice with 250 \\u0026micro;L elution buffer (50 mM NaH\\u003csub\\u003e2\\u003c/sub\\u003ePO\\u003csub\\u003e4\\u003c/sub\\u003e-NaOH, 300 mM NaCl, 500 mM imidazole, pH 7.4). Both fractions were pooled and a buffer exchange to PBS was performed with Zeba Spin Desalting Plates, 7K MWCO (Thermo Scientific), according to the guidelines provided by the manufacturer.\\u003c/p\\u003e \\u003cp\\u003eProtein concentration was estimated using the microtiter plate protocol of the micro BCA protein assay kit (Thermo Scientific), following the manufacturer's guidelines with bovine serum albumin as a standard. Protein purity was assessed with SDS-PAGE, using 12% acrylamide gels. PageRuler Broad Range unstained ladder (Thermo Scientific) was used as a protein marker.\\u003c/p\\u003e \\u003cp\\u003eTurbidity reduction assay\\u003c/p\\u003e \\u003cp\\u003eLytic activity of lysins against enterococci was evaluated using a turbidity reduction assay (TRA). An 18 h preculture of the target strain was diluted 1:25 in BHI and grown to mid-exponential phase (OD\\u003csub\\u003e600\\u003c/sub\\u003e\\u0026thinsp;=\\u0026thinsp;0.45\\u0026ndash;0.55). Cells were harvested by centrifugation (12 000 \\u0026times; g, 10 min, 4\\u0026deg;C), washed once with PBS, and resuspended in PBS. The bacterial suspension was adjusted so that, when 55 \\u0026micro;L was combined with 45 \\u0026micro;L PBS in the wells of a 96-well half-area microtiter plate (microplate, 96 well, polystyrene, half area, clear, Greiner Bio-One), the final OD\\u003csub\\u003e600\\u003c/sub\\u003e was approximately 1, accounting for dilution upon enzyme addition.\\u003c/p\\u003e \\u003cp\\u003eFor the assay, 55 \\u0026micro;L of the bacterial suspension was added to 45 \\u0026micro;L of semi-purified lysins in the half area microtiter plate. OD\\u003csub\\u003e600\\u003c/sub\\u003e was measured every minute using a Tecan Infinite plate reader for 3 h at 37\\u0026deg;C. PlyEF501 was used as positive control and just PBS served as a negative control. Each measurement was performed in biological triplicates. The OD\\u003csub\\u003e600\\u003c/sub\\u003e decrease with time was plotted as a general kinetic view. To quantitatively assess the lytic activity, the area between the curve of the negative control and each treatment (ABC) was calculated as:\\u003cdiv id=\\\"Equa\\\" class=\\\"Equation\\\"\\u003e\\u003cdiv format=\\\"TEX\\\" class=\\\"mathdisplay\\\" id=\\\"FileID_Equa\\\" name=\\\"EquationSource\\\"\\u003e\\n$$\\\\:ABC=\\\\:\\\\:{AUC}_{control}-{AUC}_{protein}$$\\u003c/div\\u003e\\u003c/div\\u003e\\u003c/p\\u003e \\u003cp\\u003eWith AUC\\u003csub\\u003econtrol\\u003c/sub\\u003e and AUC\\u003csub\\u003eprotein\\u003c/sub\\u003e being the area under the curve of the negative control and the putative lysin, respectively.\\u003c/p\\u003e \\u003cp\\u003eStatistical analysis\\u003c/p\\u003e \\u003cp\\u003eStatistical analyses were performed in the RStudio interface (2023.03.01\\u0026thinsp;+\\u0026thinsp;446), with the R programming language v4.3.1 [\\u003cspan citationid=\\\"CR42\\\" class=\\\"CitationRef\\\"\\u003e42\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR43\\\" class=\\\"CitationRef\\\"\\u003e43\\u003c/span\\u003e]. The packages that were employed include ggplot2 v3.5.1 [\\u003cspan citationid=\\\"CR44\\\" class=\\\"CitationRef\\\"\\u003e44\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR45\\\" class=\\\"CitationRef\\\"\\u003e45\\u003c/span\\u003e], readxl v1.4.3 [\\u003cspan citationid=\\\"CR46\\\" class=\\\"CitationRef\\\"\\u003e46\\u003c/span\\u003e], DescTools v0.99.58 [\\u003cspan citationid=\\\"CR47\\\" class=\\\"CitationRef\\\"\\u003e47\\u003c/span\\u003e], dplyr v1.1.4 [\\u003cspan citationid=\\\"CR48\\\" class=\\\"CitationRef\\\"\\u003e48\\u003c/span\\u003e], tibble v3.2.1[\\u003cspan citationid=\\\"CR49\\\" class=\\\"CitationRef\\\"\\u003e49\\u003c/span\\u003e], WRS2 v1.1-7 [\\u003cspan citationid=\\\"CR50\\\" class=\\\"CitationRef\\\"\\u003e50\\u003c/span\\u003e], and multcomp v1.4-26 [\\u003cspan citationid=\\\"CR51\\\" class=\\\"CitationRef\\\"\\u003e51\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eTo compare putative lysins between libraries, robust statistical methods were applied to account for non-normal and heteroskedastic data distributions. Continuous protein metrics (length, MW, pI, aliphatic index, and GRAVY index) were analyzed using a Kruskal\\u0026ndash;Wallis test, followed by Dunn\\u0026rsquo;s \\u003cem\\u003epost hoc\\u003c/em\\u003e test with Holm adjustment for multiple comparisons. For comparisons between lysins predicted to be anti-enterococcal and all remaining lysins, a Wilcoxon rank-sum test was performed. Effect sizes (ES) were estimated as described by Wilcox and Tian [\\u003cspan citationid=\\\"CR52\\\" class=\\\"CitationRef\\\"\\u003e52\\u003c/span\\u003e] with a heteroskedastic analysis of variance (ANOVA) on trimmed means (default trimming level γ\\u0026thinsp;=\\u0026thinsp;0.20) for comparison between the libraries, using the ζ statistic. Interpretation of ζ followed the guidelines of that reference: approximately 0.1 indicated a small effect, 0.3 a medium effect, and higher than 0.5 a high effect [\\u003cspan citationid=\\\"CR52\\\" class=\\\"CitationRef\\\"\\u003e52\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eDistribution of predicted protein domains were compared by a χ\\u003csup\\u003e2\\u003c/sup\\u003e-test, followed by evaluation of Pearson\\u0026rsquo;s standardized residuals if significant differences (α\\u0026thinsp;=\\u0026thinsp;0.05) were observed.\\u003c/p\\u003e \\u003cp\\u003eDifferences in lytic activity during TRA were evaluated with an ANOVA, followed by Dunnett\\u0026rsquo;s \\u003cem\\u003epost-hoc\\u003c/em\\u003e test to check if the treatment with a specific endolysin resulted in an ABC significantly different from zero.\\u003c/p\\u003e\"},{\"header\":\"3 Results\",\"content\":\"\\u003cp\\u003eFrom raw DNA reads to predicted lysins\\u003c/p\\u003e \\u003cp\\u003eIn this study, viral DNA was extracted from five distinct samples derived from pig feces or sewage (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e), followed by Illumina sequencing. Raw reads (18.9\\u0026ndash;23.47\\u0026nbsp;million per sample) were processed through the ViPER pipeline for trimming and assembly, yielding 41 978\\u0026ndash;76 946 contigs per sample (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e).\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eIllumina sequencing metrics and lysin identification for five libraries.\\u003c/b\\u003e \\u0026lsquo;Raw reads\\u0026rsquo; are expressed in millions (M). \\u0026lsquo;No. contigs\\u0026rsquo; indicates the total number of assembled contigs, while \\u0026lsquo;Viral contigs\\u0026rsquo; shows the percentage of contigs classified as viral. \\u0026lsquo;Viral abundance\\u0026rsquo; represents the proportion of reads mapping to viral contigs. Columns \\u0026lsquo;All proteins\\u0026rsquo; and \\u0026lsquo;Lysins\\u0026rsquo; report the total predicted proteins and lysins, respectively; numbers in round and square brackets denote predicted endolysins and virion-associated lysins (VAL), respectively. Some lysins could not be assigned to either category. The percentage of lysins among all proteins is shown under \\u0026lsquo;% Lysin\\u0026rsquo;. \\u0026lsquo;Representative lysins\\u0026rsquo; indicates the number of non-redundant lysins after clustering at 98% sequence identity\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"10\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c8\\\" colnum=\\\"8\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c9\\\" colnum=\\\"9\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c10\\\" colnum=\\\"10\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLibrary\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eSource\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRaw reads (M)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eContigs\\u003c/p\\u003e \\u003cp\\u003e(No.)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eViral contigs (%)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003eViral abundance (%)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003eAll proteins (No.)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003eLysins (No.)\\u003c/p\\u003e \\u003cp\\u003e(endolysin)\\u003c/p\\u003e \\u003cp\\u003e[VAL]\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e% Lysins\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003eRepresentative lysins \\u003c/p\\u003e \\u003cp\\u003e(No.)\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003ePig feces\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e23.47\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e41\\u0026thinsp;978\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e46.28\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e88.1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e117\\u0026thinsp;593\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;348\\u003c/p\\u003e \\u003cp\\u003e(1\\u0026thinsp;241)\\u003c/p\\u003e \\u003cp\\u003e[83]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;327\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003ePig feces\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e19.71\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e63\\u0026thinsp;031\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e50.27\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e81.3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e158\\u0026thinsp;086\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;904\\u003c/p\\u003e \\u003cp\\u003e(1\\u0026thinsp;821)\\u003c/p\\u003e \\u003cp\\u003e[50]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;881\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003ePig feces\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e18.91\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e76\\u0026thinsp;946\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e49.07\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e75.9\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e175\\u0026thinsp;203\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e2\\u0026thinsp;079\\u003c/p\\u003e \\u003cp\\u003e(1\\u0026thinsp;994)\\u003c/p\\u003e \\u003cp\\u003e[64]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e2\\u0026thinsp;059\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e4\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eSewage\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e20.1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e56\\u0026thinsp;841\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e31.11\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e81.3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e144\\u0026thinsp;410\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;707\\u003c/p\\u003e \\u003cp\\u003e(1\\u0026thinsp;481)\\u003c/p\\u003e \\u003cp\\u003e[211]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e1\\u0026thinsp;695\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e5\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eSewage\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e18.88\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e71\\u0026thinsp;885\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e40.51\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e \\u003cp\\u003e69.8\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e \\u003cp\\u003e161\\u0026thinsp;639\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c8\\\"\\u003e \\u003cp\\u003e2\\u0026thinsp;101\\u003c/p\\u003e \\u003cp\\u003e(1\\u0026thinsp;823)\\u003c/p\\u003e \\u003cp\\u003e[260]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c9\\\"\\u003e \\u003cp\\u003e1.3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c10\\\"\\u003e \\u003cp\\u003e2\\u0026thinsp;087\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003eContigs were classified using a combination of the geNomad and ViPER pipelines. GeNomad initially categorized contigs as viral, proviral, or plasmid. Contigs classified as proviral, as well as those not assigned by geNomad, were subsequently classified based on their ViPER annotation. Of all contigs, for each library, between 17 688 and 37 754 (31.11\\u0026ndash;50.00%) were identified as viral. Relative abundance was estimated by summing the reads mapped to the contigs for each kingdom \\u003cb\\u003e(Supplementary Fig.\\u0026nbsp;1)\\u003c/b\\u003e, with viral reads comprising 69.8\\u0026ndash;88.1% of the total reads (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e). No significant differences in abundance of phage kingdom-level classifications were observed across libraries (χ\\u0026sup2; test, \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.81). Contig completeness was subsequently assessed with CheckV for contigs classified as viral by ViPER. CheckV compares contigs to a reference database of complete viral genomes and assigns quality categories: complete, high (\\u0026gt;\\u0026thinsp;90%), medium (50\\u0026ndash;90%), low (0\\u0026ndash;50%), and undetermined. This analysis indicated that approximately 80\\u0026ndash;85% of viral contigs were of low quality and thus incomplete (\\u0026lt;\\u0026thinsp;50% completeness; \\u003cb\\u003eSupplementary Fig.\\u0026nbsp;2\\u003c/b\\u003e).\\u003c/p\\u003e \\u003cp\\u003eNext, protein-coding sequences were predicted using pyrodigal-gv within geNomad, resulting in 117 593\\u0026thinsp;\\u0026minus;\\u0026thinsp;175 203 predicted proteins per library. These proteins were used as input for SUBLYME. This tool is a protein embedding-based classifier to identify phage lysins among viral proteins, using ProtT5 to generate vector-based representations (embeddings) of protein sequences, which are then classified using SVMs [\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e]. Across libraries, 1 348 to 2 101 lysins were predicted, representing approximately 1.1% (range: 1.1\\u0026ndash;1.3%) of all predicted proteins. The number and proportion of lysins per library are summarized in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e. In addition to lysin prediction, SUBLYME predicts the type of lysin, either endolysin or VAL, indicated between round and square brackets in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e. Remarkably, most lysins are predicted to be endolysins, with VALs contributing to approximately 7.4% (653 lysins) of the predicted lysins. The other predicted lysins are either endolysins (8 158 in total) or could be either endolysins or VALs (14 in total). After clustering predicted lysins at 98% sequence identity using CD-HIT, each library contained between 1 327 to 2 087 unique lysins. When all libraries were combined and re-clustered, the dataset was reduced to 8 825 non-redundant lysins. This final set is referred to as \\u0026lsquo;all lysins\\u0026rsquo; throughout the manuscript.\\u003c/p\\u003e \\u003cp\\u003eExploring the metagenomic lysin landscape\\u003c/p\\u003e \\u003cdiv id=\\\"Sec4\\\" class=\\\"Section3\\\"\\u003e \\u003cdiv class=\\\"Heading\\\"\\u003e3.1.1 Physicochemical properties\\u003c/div\\u003e \\u003cp\\u003eTo characterize the lysin landscape, several physicochemical properties were calculated for lysins from each individual library as well as for the combined dataset. These properties included protein length, MW, pI, aliphatic index, and hydrophobicity. Previous studies have shown that full-length lysins from Gram-negative phages are typically shorter than those from Gram-positive phages [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e]. The pI, which indicates the pH at which a protein carries no net charge, and the GRAVY score, which reflects overall hydrophobicity [\\u003cspan citationid=\\\"CR54\\\" class=\\\"CitationRef\\\"\\u003e54\\u003c/span\\u003e], tend to be higher in lysins from phages infecting Gram-negative bacteria [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e]. Similarly, the aliphatic index, which represents the relative volume of aliphatic side chains and correlates with protein thermostability [\\u003cspan citationid=\\\"CR55\\\" class=\\\"CitationRef\\\"\\u003e55\\u003c/span\\u003e], has also been reported to be elevated in Gram-negative phage lysins [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e]. Distributions of these properties across the 5 samples are visualized in violin plots (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e).\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eAll the differences in physicochemical features across libraries found were low-to-moderate (ES\\u0026thinsp;=\\u0026thinsp;0.2), with a lower ES for the pI (0.07), which strongly signals that there are no differences in pI distributions between the libraries. Libraries from pig feces (L1-L3) contained more lysins that were longer and with a higher MW, with a broader distribution (Mean\\u0026thinsp;=\\u0026thinsp;22 kDa, Q1\\u0026thinsp;=\\u0026thinsp;17 kDa, Q3\\u0026thinsp;=\\u0026thinsp;28 kDa) than in sewage samples L4-L5 (mean\\u0026thinsp;=\\u0026thinsp;19 kDa, Q1\\u0026thinsp;=\\u0026thinsp;15 kDa, Q3\\u0026thinsp;=\\u0026thinsp;25 kDa). In contrast, fecal samples contained more lysins with a lower aliphatic index and GRAVY than sewage samples.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec5\\\" class=\\\"Section3\\\"\\u003e \\u003cdiv class=\\\"Heading\\\"\\u003e3.1.2 Architecture and functional domain families\\u003c/div\\u003e \\u003cp\\u003eDomain architectures of lysins were predicted using four InterProScan member databases. Based on these predictions, protein domains were classified using the framework described by Criel et al. [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e], which organizes domains into clusters, comprising thirteen CBD clusters and twenty-six EAD types, with EADs further categorized into seven enzymatic activity groups (see 2.6 Characterization of lysins).\\u003c/p\\u003e \\u003cp\\u003eOut of the 8 825 proteins analyzed, 6 700 (75.9%) contained at least one predicted domain, with library-specific proportions ranging from 71.6% to 80.0%. When restricting the analysis to proteins with domains annotated within the used framework, the proportion decreased to 72.9% (range: 69.7\\u0026ndash;76.6%; \\u003cb\\u003eSupplementary Table\\u0026nbsp;3\\u003c/b\\u003e). Complete domain architectures are listed in \\u003cb\\u003eSupplementary Tables\\u0026nbsp;4 and 5\\u003c/b\\u003e. Unless otherwise stated, percentages refer only to proteins with at least one domain included in the framework.\\u003c/p\\u003e \\u003cp\\u003eMost proteins contained a single domain (68%), predominantly an EAD (63%) rather than a CBD (5.3%). Among proteins with two annotated domains, the most common configuration was an N-terminal EAD followed by a C-terminal CBD (17%), while the reverse orientation (CBD\\u0026ndash;EAD) was rare (2%). Architectures with two EADs or two CBDs were observed in 2.5% and 1.9% of cases, respectively. Although most other configurations occurred at frequencies below 0.57%, an N-terminal EAD followed by two (4.6%) or three CBDs (1.1%) were relatively more common. These results should be interpreted cautiously, as repeated domains may form a single functional module [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e]. For example, in this study, 267 of the 645 proteins with three or more domains were found to have two or more adjacent CW_7 CBDs. Additionally, the reliance on sequence-based domain prediction may underestimate the number of functional domains, particularly in proteins with divergent sequences or domains not yet represented in reference databases.\\u003c/p\\u003e \\u003cp\\u003eNo significant differences were found between libraries in the number of predicted domains per protein (χ\\u0026sup2; test), although some tendencies were noted. Fecal libraries (L1\\u0026ndash;L3) contained more proteins with two predicted domains (17\\u0026ndash;19%) and fewer with one domain (77\\u0026ndash;79%) compared to sewage libraries (L4\\u0026ndash;L5), which had 11\\u0026ndash;13% two-domain proteins and 85\\u0026ndash;86% single-domain proteins). This difference reflects the shorter length of lysins in sewage libraries and may indicate variation in the relative abundance of phages infecting Gram-negative versus Gram-positive bacteria, as the former typically encode single-domain lysins while the latter often exhibit multi-domain architectures related to the shorter length of these lysins [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eAmong all predicted domains, 73.3% were EADs, 20.8% were CBDs, and 6% were other domains not included in further analyses (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003ea\\u003cb\\u003e)\\u003c/b\\u003e. CBDs were classified into 13 domain clusters, PG_1 (37%), CW_7 (18%), SH3 (14%), LysM (11%), and PG_3 (10%) being the most frequent. These percentages represent the distribution of each cluster among all predicted CBDs. A significant difference in CBD cluster distribution was observed between libraries according to a χ\\u003csup\\u003e2\\u003c/sup\\u003e test (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;9.653 \\u0026times; 10\\u003csup\\u003e\\u0026minus;\\u0026thinsp;5\\u003c/sup\\u003e). Standardized Pearson residuals (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003eb) identified PG_3 as being more prevalent in sewage libraries L4\\u0026ndash;L5 (25\\u0026ndash;26%; residuals 3.1\\u0026ndash;3.3\\u003cem\\u003e)\\u003c/em\\u003e and less common in fecal libraries L1\\u0026ndash;L3 (ranging 2.5\\u0026ndash;10%; standardized Pearson\\u0026rsquo;s residual \\u0026minus;\\u0026thinsp;3.0 to -0.88). The SPOR domain was more frequent in L2 and L3 (13% and 9% with Pearson\\u0026rsquo;s standardized residual of 2.3 and 1, respectively), while CW_7 was enriched in fecal libraries \\u003cem\\u003e(\\u003c/em\\u003e19\\u0026ndash;21%; standardized Pearson\\u0026rsquo;s residual 0.65 and 1\\u003cem\\u003e)\\u003c/em\\u003e compared to sewage libraries \\u003cem\\u003e(\\u003c/em\\u003e8\\u0026ndash;14%; standardized Pearson\\u0026rsquo;s residual \\u0026minus;\\u0026thinsp;2.1 to -0.61\\u003cem\\u003e).\\u003c/em\\u003e\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe ratio of EADs with specific activities (or belonging to specific domain clusters) to the total number of EADs was calculated and expressed as a percentage for each library (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003ec). The most abundant EAD clusters were Ami_2 (14%), PET_M15 (12%), and SLT_related (12%), with all others below 10% each. The most frequently predicted enzymatic activities were \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidase (24%), \\u003cem\\u003eN-\\u003c/em\\u003eacetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase (22%) and peptidase activity (19%). No significant differences were identified between the libraries for activities (χ\\u0026sup2; test; \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.9117) or domain clusters (χ\\u0026sup2; test; \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.9998). However, libraries obtained from similar samples (feces or sewage) appeared to have more similar enzymatic domains. Fecal libraries (L1- L3) contained more EADs from the Ami_2, Ami_3, CHAP, and GH25 clusters, whereas sewage libraries (L4 and L5) had more EADs in the GH_24 and PET_M15 domain clusters. Similarly, \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidase activity was more frequent in fecal libraries (L1\\u0026ndash;L3), where \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase activity was more common in sewage libraries (L4 and L5).\\u003c/p\\u003e \\u003cp\\u003eTogether, these findings highlight distinct physicochemical and architectural profiles between lysins derived from fecal and sewage environments, reflecting underlying differences in phage host specificity and ecological context (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e).\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab3\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 3\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eSummary of differences between lysins predicted in fecal libraries and sewage libraries.\\u003c/b\\u003e GRAVY: grand average of hydropathy, ES: effect size, MW: molecular weight, ns: not significant, pI: isoelectric point\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"4\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eFeature\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eFecal libraries (L1\\u0026ndash;L3)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eSewage libraries (L4\\u0026ndash;L5)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eStatistical/interpretive note\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eProtein length and MW\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eLonger, higher MW\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eShorter, lower MW\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eOverall differences small (ES\\u0026thinsp;\\u0026asymp;\\u0026thinsp;0.2)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003epI\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e~ no difference\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e~ no difference\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eNegligible difference\\u003c/p\\u003e \\u003cp\\u003e (ES\\u0026thinsp;=\\u0026thinsp;0.07)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAliphatic index\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003elower\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003ehigher\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eSmall effect (ES\\u0026thinsp;\\u0026asymp;\\u0026thinsp;0.2)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eHydrophobicity (GRAVY)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003elower\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003ehigher\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eSmall effect (ES\\u0026thinsp;\\u0026asymp;\\u0026thinsp;0.2)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eNo. predicted domains per protein\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eMore 2-domain proteins (17\\u0026ndash;19%); \\u003c/p\\u003e \\u003cp\\u003eFewer singledomain (77\\u0026ndash;79%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eFewer 2-domain proteins (11\\u0026ndash;13%)\\u003c/p\\u003e \\u003cp\\u003eMore singledomain (85\\u0026ndash;86%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eχ\\u0026sup2; ns \\u003c/p\\u003e \\u003cp\\u003e(Tendencies only)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eCBD composition\\u003c/p\\u003e \\u003cp\\u003e(cluster-level)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eHigher CW_7 \\u003c/p\\u003e \\u003cp\\u003eLower PG_3 \\u003c/p\\u003e \\u003cp\\u003eHigher SPOR\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eHigher PG_3 \\u003c/p\\u003e \\u003cp\\u003eLower CW_7\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eχ\\u0026sup2; significant\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEAD composition\\u003c/p\\u003e \\u003cp\\u003e(cluster-level)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eHigher Ami_2, Ami_3, CHAP, GH25\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eHigher GH_24, PET_M15\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eχ\\u0026sup2; for EAD clusters ns (Tendencies only)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEnzymatic activities\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eHigher \\u003cem\\u003eN-\\u003c/em\\u003eacetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidase\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eHigher \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eχ\\u0026sup2; for activities ns (Tendencies only)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003eExploring the anti-enterococcal lysin landscape\\u003c/p\\u003e \\u003cp\\u003eNext, lysins from phages targeting the genus \\u003cem\\u003eEnterococcus\\u003c/em\\u003e (anti-enterococcal lysins) were predicted. To this end, a new protein embedding-based SVM classifier was trained, using 237 lysins from an enterococcal background and random subsets of 500 non-enterococcal lysins retrieved from the PhaLP database (https;//phalp.ugent.be) [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e]. Models achieving a precision of at least 80% were retained, as this threshold provides confidence in prediction accuracy, despite potentially excluding some candidates. Of the 200 trained models, 79 met this criterion and were subsequently used to predict anti-enterococcal lysins among the newly identified sequences (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e). Lysins predicted as anti-enterococcal by at least 80% of these retained models were considered positive, yielding 129 lysins (1.5% of all lysins) predicted to be anti-enterococcal. Among these, 102 (79%) were predicted endolysins and 23 (18%) were predicted VALs. The distribution of the calculated MW, length in amino acids, aliphatic index, GRAVY index (hydrophobicity), and pI of the anti-enterococcal lysins versus other lysins can be found in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eAnti-enterococcal lysins exhibited a broad, bimodal distribution in both protein length and MW, with primary peaks around 275 and 380 amino acids (~\\u0026thinsp;30 and ~\\u0026thinsp;43 kDa), alongside several outliers. Compared to non-anti-enterococcal lysins, they were longer and had a higher MW (ES\\u0026thinsp;=\\u0026thinsp;0.81 for both). Anti-enterococcal lysins showed a bimodal pI distribution, similar to the overall lysin dataset, with few proteins near the physiological pH of 7.4, indicating that most are charged under physiological conditions. Compared to non-anti-enterococcal lysins, their pI profile differed moderately (ES\\u0026thinsp;=\\u0026thinsp;0.25), with an approximately even split between peaks at pI 5.3 and 9.3 (median\\u0026thinsp;=\\u0026thinsp;6.8). In contrast, non-anti-enterococcal lysins showed two modes at pI 4.8 and 9.3 (median of 8.5). Moderate differences were observed for the GRAVY (ES\\u0026thinsp;=\\u0026thinsp;0.26) and aliphatic index (ES\\u0026thinsp;=\\u0026thinsp;0.48), both lower in anti-enterococcal lysins. These trends are consistent with typical features of Gram-positive lysins, which tend to be larger and more hydrophilic.\\u003c/p\\u003e \\u003cp\\u003eThe domain composition of anti-enterococcal lysins was examined using InterProScan to infer potential functional characteristics. Among the 129 proteins analyzed, 49 (38%) lacked any predicted domains, suggesting the presence of novel or highly divergent sequences not captured by the four domain databases used. Only 19 proteins (15%) contained at least one CBD, distributed across four domain clusters: SH3 (10 proteins), LysM (4), CW_7 (3), and PG_1 (3). In contrast, 79 proteins (61%) carried at least one EAD, with four proteins containing two EADs. Of the 83 predicted EADs, 22 were predicted as \\u003cem\\u003eN\\u003c/em\\u003e-acetylmuramoyl-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003el\\u003c/span\\u003e-alanine amidases, six as peptidases, and twenty-six as having both activities (CHAP and NLPC_P60 domains). Additionally, nineteen proteins were predicted to have \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-muramidase activity, nine combined muramidase and lytic transglycosylase activity, and one was annotated with \\u003cem\\u003eN\\u003c/em\\u003e-acetyl-β-\\u003cspan type=\\\"SmallCaps\\\" class=\\\"SmallCaps\\\" name=\\\"Emphasis\\\"\\u003ed\\u003c/span\\u003e-glucosaminidase activity. The four proteins with dual EADs included combinations such as CHAP with GH25, CHAP with SLT-related, and GH25 with PET_M23 domains. Eleven proteins contained both a CBD and an EAD, with five showing an N-terminal EAD and C-terminal CBD, five having the reverse orientation, and one with an N-terminal EAD followed by two CBDs. In several cases, terminal regions lacked predicted domains, potentially indicating the presence of functional modules that are either novel or too divergent to be detected by current domain models.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec6\\\" class=\\\"Section3\\\"\\u003e \\u003cdiv class=\\\"Heading\\\"\\u003e3.1.3 Selection of anti-enterococcal lysins\\u003c/div\\u003e \\u003cp\\u003eIn the next step, a subset of 21 lysins was selected for expression and purification. For practical reasons, a maximum protein length of 550 amino acids was preferred. This threshold was set to enhance the likelihood of successful heterologous expression and purification [\\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e32\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e]. After excluding twenty longer than 550 amino acids, 109 lysins remained. From the remaining pool, 21 lysins were selected through an uninformed, random selection.\\u003c/p\\u003e \\u003cp\\u003eThe physicochemical properties (length, MW, pI, aliphatic index, GRAVY index) of the 21 selected proteins were plotted as dots on the violin plots depicting the properties of all anti-enterococcal phage lysins (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e). This revealed that the random selection achieved to retain a similar distribution of the properties as observed for the whole population. Protein structures for the 21 selected putative lysins were predicted using AlphaFold3 [\\u003cspan citationid=\\\"CR34\\\" class=\\\"CitationRef\\\"\\u003e34\\u003c/span\\u003e]. Domain boundaries were identified with SPAED [\\u003cspan citationid=\\\"CR35\\\" class=\\\"CitationRef\\\"\\u003e35\\u003c/span\\u003e], and annotations were assigned via InterProScan [\\u003cspan citationid=\\\"CR56\\\" class=\\\"CitationRef\\\"\\u003e56\\u003c/span\\u003e] (\\u003cb\\u003eSupplementary Table\\u0026nbsp;6\\u003c/b\\u003e). Of these 21 lysins, only ten contained at least one predicted domain as classified according to Criel et al. [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e], while the remaining eleven comprised domains not included in that classification or lacked InterProScan predictions.\\u003c/p\\u003e \\u003cp\\u003eFunctional characterization of enterococcal phage lysins\\u003c/p\\u003e \\u003cp\\u003eTwenty-one constructs were successfully cloned into the expression vector, and all were expressed and purified based on metal affinity. After semi-purification, the total protein concentration as estimated with the micro BCA method (bovine serum albumin as standard) ranged between 0.14 and 2.5 mg protein per 100 mL expression medium (\\u003cb\\u003eSupplementary Table\\u0026nbsp;7\\u003c/b\\u003e), with highly heterogenous purities (\\u003cb\\u003eSupplementary Fig.\\u0026nbsp;3\\u003c/b\\u003e). Due to the considerable variability in protein purity, equimolar comparisons between the tested proteins were not possible, and the concentrations given refer only to the total protein amount \\u0026ndash;not to the concentration of the protein of interest.\\u003c/p\\u003e \\u003cp\\u003eThe muralytic activity of the lysins was evaluated using a TRA, which measured the turbidity (as OD\\u003csub\\u003e600\\u003c/sub\\u003e) of a bacterial suspension over time. A reduction in OD\\u003csub\\u003e600\\u003c/sub\\u003e was interpreted as an indication of cell lysis. Because the lysins were predicted on the genus level, a selection of three enterococcal species was used: \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e ECC1, \\u003cem\\u003eEnt. faecium\\u003c/em\\u003e AW3576, and \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e NJ-3. These strains represent two distinct peptidoglycan chemotypes found in enterococci (A3α for \\u003cem\\u003eEnt. faecium\\u003c/em\\u003e and \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e, and A4α for \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e [\\u003cspan citationid=\\\"CR57\\\" class=\\\"CitationRef\\\"\\u003e57\\u003c/span\\u003e]). Additionally, they include two clinically relevant species associated with human infections (\\u003cem\\u003eEnt. faecium\\u003c/em\\u003e and \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e).\\u003c/p\\u003e \\u003cp\\u003eThe progress of the OD over time (\\u003cb\\u003eSupplementary Fig.\\u0026nbsp;4\\u003c/b\\u003e) was quantified by calculating the ABC, which represents the integrated difference in OD\\u003csub\\u003e600\\u003c/sub\\u003e between lysin-treated samples and the negative control across all time points (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e). The positive control exhibited significant lytic activity against all three \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species tested. Among the tested strains, \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e appeared most susceptible to lysis. Notably, some degree of autolysis was observed in the negative control of this strain, which may have weakened the peptidoglycan layer and sensitized cells, thereby enhancing the sensitivity to detect lysins with only minor to moderate activity. In total, four lysins showed lytic activity against \\u003cem\\u003eEnt. faecium\\u003c/em\\u003e (E_24, E_25, E_31, E_65), six against \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e (including the aforementioned plus E_82 and E_101), and thirteen against \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e (all previously mentioned plus E_28, E_46, E_60, E_61, E_63, E_76, and E_119), respectively. Thus, thirteen out of the twenty-one lysins tested (62%) displayed lytic activity against at least one \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species, confirming their anti-enterococcal potential. Notably, six of these active lysins (E_24, E_28, E_31, E_65, E_82, E_86) lacked any domains classified by Criel et al. [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"4 Discussion\",\"content\":\"\\u003cp\\u003ePhage lysins are enzymes encoded by bacteriophages that break down bacterial cell walls, representing a promising new class of antibiotics. However, discovering novel lysins is essential to fully realize their potential, as it expands the range of available lysins and provides insights into their sequences and corresponding activities. Sequence-based metagenomics reveals the genomic content of various environments, and, when paired with sequence mining, it becomes a valuable tool for discovering lysins in these diverse settings.\\u003c/p\\u003e \\u003cp\\u003eIn this study, a sequence-based metagenomic pipeline was implemented, which included the following steps: (i) extracting and sequencing viral metagenomic material from five samples originating from pig feces and sewage, (ii) assembling the sequences into contigs and predicting open reading frames along with related proteins, (iii) identifying lysins across the proteins, (iv) \\u003cem\\u003ein silico\\u003c/em\\u003e characterization of the predicted lysins, (v) prediction of enterococcal phage origin, and (vi) functional screening of a selection of putative anti-enterococcal phage lysins.\\u003c/p\\u003e \\u003cp\\u003eSampling, DNA extraction and sequencing\\u003c/p\\u003e \\u003cp\\u003eFive different samples were taken from two types of material: pig feces [from two locations: the faculty of veterinary Medicine of Ghent University (Merelbeke) and Flanders research institute for agriculture, fisheries, and food (ILVO)] and sewage from the Ghent University Hospital (UZ Gent). Both types of material are expected to contain enterococci and thus also their phages and lysins. In addition, the distinct types of starting materials (semi-solid and liquid) highlight the diversity in source material. Yet, this diversity in material type, location and date remains a biased snippet of the wide environmental diversity that exists. Moreover, this study focused on DNA extraction, and the Illumina sequencing with its associated library preparation primarily targets dsDNA, excluding RNA and ssDNA phages. Nevertheless, most characterized RNA and ssDNA phages are known to employ alternative lysis mechanisms that do not primarily rely on enzymatic degradation of peptidoglycan [\\u003cspan citationid=\\\"CR58\\\" class=\\\"CitationRef\\\"\\u003e58\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR59\\\" class=\\\"CitationRef\\\"\\u003e59\\u003c/span\\u003e], but instead use other lysin systems, such as inhibiting new cell wall synthesis [\\u003cspan citationid=\\\"CR60\\\" class=\\\"CitationRef\\\"\\u003e60\\u003c/span\\u003e]. Such lysis systems would not be detected with the computational and experimental methods used in this work and thus fall out of scope. However, recent transcriptomic studies have uncovered greater diversity among RNA phages and their lytic strategies [\\u003cspan citationid=\\\"CR61\\\" class=\\\"CitationRef\\\"\\u003e61\\u003c/span\\u003e], suggesting potential for novel lysin discovery. The proposed pipeline could be adapted to include RNA and ssDNA phages by incorporating RNA-to-cDNA conversion [\\u003cspan citationid=\\\"CR62\\\" class=\\\"CitationRef\\\"\\u003e62\\u003c/span\\u003e] or converting ssDNA to dsDNA [\\u003cspan citationid=\\\"CR63\\\" class=\\\"CitationRef\\\"\\u003e63\\u003c/span\\u003e] during sample preparation.\\u003c/p\\u003e \\u003cp\\u003eAfter assembling sequence reads, 310 681 contigs were obtained, of which 135 675 contigs were classified as viral. The abundance of reads mapping to the viral contigs amounted to 69.8\\u0026ndash;88.1%, (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e). Although classification and abundance estimates depend on the bioinformatic tools used for virus identification, viruses typically represent less than 5% of reads in non-enriched samples, suggesting the effectiveness of the enrichment strategy applied here [\\u003cspan citationid=\\\"CR64\\\" class=\\\"CitationRef\\\"\\u003e64\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR65\\\" class=\\\"CitationRef\\\"\\u003e65\\u003c/span\\u003e]. Analysis with CheckV indicated that most viral contigs predicted using DIAMOND blastx were incomplete, meaning they represented only partial viral genomes with an estimated completeness of 0\\u0026ndash;50%. This is expected in metagenomic datasets due to the low sequencing coverage of rare viruses [\\u003cspan citationid=\\\"CR64\\\" class=\\\"CitationRef\\\"\\u003e64\\u003c/span\\u003e]. In total, 280 viral contigs were classified as complete genomes. Lysin prediction was performed on all contigs, regardless of completeness, which is an advantage of our approach since a full viral genome is not required to identify a lysin. However, when a viral genome is incomplete, it may lack the lysin gene entirely, which explains why a lysin could not be predicted for every contig.\\u003c/p\\u003e \\u003cp\\u003ePrediction of candidate lysins\\u003c/p\\u003e \\u003cp\\u003eThis study utilized SUBLYME to predict candidate phage lysins, a protein embedding-based classifier offering an alternative to traditional sequence homology-based methods. While the latter approaches rely on sequence alignment tools to detect similarity between candidate proteins and known lysins using specific thresholds (e.g., sequence identity, alignment coverage, alignment length, and E-value), such approaches may fail to detect highly divergent or novel lysins with limited sequence similarity to known references. SUBLYME partly overcomes this limitation by leveraging protein embeddings to capture features beyond sequence similarity. In a previous benchmark study using the dataset from Fern\\u0026aacute;ndez-Ruiz et al. [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e], which included 3.8\\u0026nbsp;million proteins from 183 298 uncultured phages, SUBLYME predicted 41 007 thousand endolysins (1.1%), compared to just 2 628 (0.069%) detected by the sequence homology-based methods. Additionally, 9 849 VALs were identified within this dataset (0.26%) [\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eConsistent with these findings, the present study predicted 8 825 representative lysins among 756 931 proteins (1.1%), including 8 158 endolysins (1.1%). In contrast, only 653 predicted VALs were detected (0.08%). While strong conclusions should be avoided given the uncertainty surrounding the prevalence of VALs across phage genomes, this low proportion suggests that the classifier may have difficulty accurately detecting VALs. This could be due to limitations inherent to the model, particularly the smaller training set for VALs compared to endolysins (respectively 4 429 versus 10 970 proteins), which may not sufficiently capture the expected diversity. Additionally, the presence of structural protein domains lacking muralytic activity within VALs further complicates their accurate identification. To improve VAL detection, a dedicated classifier trained specifically on VALs, using a broader and more diverse dataset, may be required, combined with further computational and experimental insight into the true VAL diversity.\\u003c/p\\u003e \\u003cp\\u003eMoreover, the current model may possibly favor lysins with features resembling those in the PhaLP database, potentially overlooking novel variants. To identify completely novel lysins, alternative approaches may be required. One such approach is functional metagenomics, which bypasses the need for sequence similarity or prior knowledge by cloning environmental DNA into a heterologous host and screening the resulting expression libraries for lytic activity. This strategy enables the discovery of enzymes based on function rather than sequence and has already been successfully applied to discover lysins from (meta-)genomic sources, as reviewed elsewhere [\\u003cspan citationid=\\\"CR66\\\" class=\\\"CitationRef\\\"\\u003e66\\u003c/span\\u003e]. By incorporating newly discovered lysins from functional screens into the training set, machine learning models can be refined to improve detection of highly divergent lysins. Thus, sequence-based and functional metagenomics represent complementary strategies for fully exploring lysin diversity.\\u003c/p\\u003e \\u003cp\\u003eLysins across different environments\\u003c/p\\u003e \\u003cp\\u003eLysins from the different libraries were characterized by physicochemical properties (MW, length, pI, aliphatic index, and GRAVY index) and predicted domain composition. This characterization revealed that libraries from pig feces samples appeared to have characteristics mostly observed in lysins targeting Gram-positives, where sewage samples also contained lysins with characteristics typically found in lysins targeting Gram-negatives (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e). That is, the length and MW of lysins from fecal samples were significantly higher, a characteristic frequently observed in the (generally multimodular) lysins targeting Gram-positives [\\u003cspan citationid=\\\"CR53\\\" class=\\\"CitationRef\\\"\\u003e53\\u003c/span\\u003e]. Moreover, the domain cluster PG_3, exclusively observed in Gram-negative-targeting lysins [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e], was significantly more present in lysins of sewage libraries (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e), together with a more frequent occurrence of GH24 enzymatic domains, an EAD cluster mostly observed in Gram-negative targeting bacteria [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e]. Although these disparities may still be attributed to differences during the phage precipitation protocols, these findings suggest that (pig) feces were a rich source of lysins with a typical Gram-positive architecture, whereas sewage contained both phage lysins with typical Gram-positive and Gram-negative architecture. Similarly, Fern\\u0026aacute;ndez-Ruiz et al. [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e] found phages from aquatic samples to contain endolysins predicted to be active against both Gram-positive and Gram-negative bacterial hosts, where in the human microbiome, lysins with a typical architecture for targeting Gram-positives were preferentially found. Whether these architectural differences truly reflect variations in phage host range remains an open question and could be further investigated, for instance through refined host prediction approaches based on phage genomic data [\\u003cspan citationid=\\\"CR67\\\" class=\\\"CitationRef\\\"\\u003e67\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eAnti-enterococcal lysins\\u003c/p\\u003e \\u003cp\\u003eThis study employed a new machine learning tool to predict putative phage lysins against a given bacterial genus, using an ensemble of classifiers trained on balanced datasets selected to have a precision higher than 80%. In this case, dealing with the prediction of anti-enterococcal lysins, the employed dataset (derived from PhaLP database) allowed to build 79 models which individually evaluated every lysin candidate. Lysins predicted to be anti-enterococcal lysins by at least 80% of the models were retained, providing high-confidence associations. The latter criterium is, however, tunable, and while in this work we chose to keep a conservative cut-off value, lowering the proportion of models with a positive outcome could still be valuable, especially if the aim is finding additional diversity \\u0026ndash;though at the cost of potentially introducing more false positives.\\u003c/p\\u003e \\u003cp\\u003eThis tool thus offers a pipeline for identifying anti-enterococcal lysins, with potential for adaptation to other bacterial genera. However, the approach may be overly permissive, potentially predicting non-active lysins as functional or misclassifying lysins with different host specificities as anti-enterococcal. To improve specificity, an alternative strategy could involve first predicting phage host taxonomy at the genus level using tools such as iPHoP [\\u003cspan citationid=\\\"CR68\\\" class=\\\"CitationRef\\\"\\u003e68\\u003c/span\\u003e], followed by lysin identification. Nonetheless, this method may overlook lysins with broad or cross-species lytic activity, as (endo)lysins often exhibit host ranges that extend beyond those of their associated phages [\\u003cspan citationid=\\\"CR69\\\" class=\\\"CitationRef\\\"\\u003e69\\u003c/span\\u003e], this approach may miss lysins with promiscuous or cross-species lytic activity.\\u003c/p\\u003e \\u003cp\\u003eSelection of anti-enterococcal lysins\\u003c/p\\u003e \\u003cp\\u003eTo assess the predictive model and address concerns regarding potential false positives, a subset of candidate proteins was selected for experimental validation. To facilitate heterologous expression and purification, proteins longer than 550 amino acids were excluded. This criterion inherently removed most predicted VALs (sixteen out of twenty-three), retaining primarily endolysins. Although excluded from this initial screen, these larger VALs may still possess lytic potential. A feasible strategy for characterizing VALs while maintaining compatibility with expression systems could involve isolating and expressing only their lytic domains, omitting structural regions.\\u003c/p\\u003e \\u003cp\\u003eAn uninformed random selection was then performed to select 21 proteins for further characterization. Their physicochemical properties were broadly distributed across the spectrum observed in all predicted anti-enterococcal lysins. Their physicochemical properties were broadly distributed across the spectrum observed in all anti-enterococcal lysins (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e), suggesting that the selected subset encompassed a diverse range of characteristics within the full dataset. These proteins were heterogeneously expressed in \\u003cem\\u003eE. coli\\u003c/em\\u003e, and semi-purified using metal affinity-based chromatography.\\u003c/p\\u003e \\u003cp\\u003eProtein yields varied considerably, from 0.14 to 2.5 mg per 100 mL expression culture, with differing purity levels. Such variability is to be expected when expressing proteins identified in metagenomic samples. For example, Fu et al. [\\u003cspan citationid=\\\"CR70\\\" class=\\\"CitationRef\\\"\\u003e70\\u003c/span\\u003e] reported yields of 0.7\\u0026ndash;24.3 mg per 100 mL for lysins identified from metagenomic samples. Moreover, Cremelie et al. [\\u003cspan citationid=\\\"CR71\\\" class=\\\"CitationRef\\\"\\u003e71\\u003c/span\\u003e] highlighted that recombinant production of phage lysins often faces solubility and host-dependent functional limitations. Balaban et al. [\\u003cspan citationid=\\\"CR72\\\" class=\\\"CitationRef\\\"\\u003e72\\u003c/span\\u003e] similarly reported that only 30\\u0026ndash;40% of two staphylococcal endolysins could be recovered in the soluble fraction after \\u003cem\\u003eE. coli\\u003c/em\\u003e expression. These observations point out that lysins are not naturally optimized for high-level expression in heterologous systems, making consistent production difficult. While strategies such as codon optimization, alternative hosts, expression of individual domains, or fusion partners may improve yield, success is not guaranteed.\\u003c/p\\u003e \\u003cp\\u003eMuralytic activity of the selected lysins\\u003c/p\\u003e \\u003cp\\u003ePartially purified lysins, exhibiting a wide range of yields and purity levels, were qualitatively screened for muralytic activity using a single dose against three enterococcal species (\\u003cem\\u003eEnt. faecium, Ent. faecalis\\u003c/em\\u003e, and \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e). Four lysins were active against all three species, two additional lysins showed activity against two species (\\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e and \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e) and seven were active only against \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e. The latter strain was likely the most sensitive due to observed partial autolysis activity under experimental conditions, a phenomenon also reported previously [\\u003cspan citationid=\\\"CR73\\\" class=\\\"CitationRef\\\"\\u003e73\\u003c/span\\u003e]. Despite suboptimal expression and purification yields, 62% of the tested proteins exhibited activity against at least one enterococcal species, supporting the effectiveness of the pipeline in identifying functional anti-enterococcal lysins. Notably, six active lysins (E_24, E_28, E_31, E_65, E_82, E_86) lacked any predicted domain using InterProScan, suggesting that the approach may also uncover lysins with novel domain architectures.\\u003c/p\\u003e \\u003cp\\u003eIn conclusion, this study introduced a sequence-based metagenomic pipeline integrated with embedding-based tools to identify and characterize phage lysins from diverse environmental samples. The approach successfully retrieved a large and diverse set of candidate lysins, including several with confirmed muralytic activity against enterococci. Notably, some identified lysins combined novel domain architectures with demonstrated activity, underscoring the potential of this pipeline to discover lysins with both structural novelty and functional potential. Further, comparative analysis revealed differences in lysin properties between fecal and sewage samples, suggesting distinct ecological sources connected to different lysins. These findings underscore the potential of metagenomics to expand the lysin repertoire and provide insights into their diversity and functionality. However, key challenges remain. Protein yields varied widely, and several lysins exhibited low purity. Additionally, functional assays were limited to three species, single strains, and one condition, restricting conclusions about host range and clinical relevance. Future work should focus on refining predictive algorithms, improving expression systems for scalable production, and expanding functional characterization under diverse conditions. Addressing these challenges will be critical for translating lysin discovery into clinical applications.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e \\u003ch2\\u003eConflict of Interest\\u003c/h2\\u003e \\u003cp\\u003eYB is co-founder and scientific advisor of Obulytix.\\u003c/p\\u003e \\u003c/p\\u003e\\u003cp\\u003e\\u003cstrong\\u003eCompeting Interests\\u003c/strong\\u003e\\u003cp\\u003eYB is co-founder and scientific advisor of Obulytix.\\u003c/p\\u003e\\u003c/p\\u003e\\u003ch2\\u003eFunding\\u003c/h2\\u003e \\u003cp\\u003eIP was supported by the Research Foundation \\u0026ndash; Flanders (FWO) under Research project FWO SB 1SC9424N. RV was supported by the \\u0026lsquo;Bijzonder Onderzoeksfonds\\u0026rsquo; (BOF) Ghent University with a postdoctoral fellowship [01P10022]. LDC was supported by the Research Foundation \\u0026ndash; Flanders (FWO) with doctoral fellowship [11L1325N].\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eConceptualization: IP, RV, YB; Funding acquisition: IP, TVdW, YB; LDC, JM Software: AB, LDC, IP; Formal analysis: IP, LDC, AB, RV; Investigation: IP; Methodology: IP, RV, TVdW, YB; Supervision: RV, TVdW, YB; Visualization: IP; Writing - original draft preparation: IP; Writing - review and editing: all authors.\\u003c/p\\u003e\\u003ch2\\u003eAcknowledgement\\u003c/h2\\u003e\\u003cp\\u003eWe thank Maya Merabishvili and Jean-Paul Pirnay (Queen Astrid Military Hospital, QAMH) for providing Enterococcus phage AQEF5 and its host, Enterococcus faecium AW3576. We would like to acknowledge Juan M. Rodr\\u0026iacute;guez from UCM for providing the Ent. hirae ECC1. We are also grateful to Thomas Martens (ILVO), F\\u0026euml;llanza Halimi (Ghent University), Tony Nimmegeers (UZ Gent), and Nils and Dirk Mouton (Bioboerderij De Zwaluw) for sharing the environmental samples that were crucial for this study.\\u003c/p\\u003e\\u003ch2\\u003eData Availability\\u003c/h2\\u003e\\u003cp\\u003eThe datasets for this study can be found in the European Nucleotide Archive: [http://www.ebi.ac.uk/ena/browser/view/PRJEB102709](http:/www.ebi.ac.uk/ena/browser/view/PRJEB102709) . The implementation of the classifier to predict anti-enterococcal lysins can be found on GitHub at: [https://github.com/Rousseau-Team/lysin-target-pred](https:/github.com/Rousseau-Team/lysin-target-pred) . Protein datasets are provided in the supplementary information (Supplementary_files_03).\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003ePiddock LJV, Alimi Y, Anderson J, de Felice D, Moore CE, Rottingen JA et al (2024) Advancing global antibiotic research, development and access. Nat Med 30(9):2432\\u0026ndash;2443. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41591-024-03218-w\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41591-024-03218-w\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFishbein SRS, Mahmud B, Dantas G (2023) Antibiotic perturbations to the gut microbiome. Nat Rev Microbiol 21(12):772\\u0026ndash;788. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41579-023-00933-y\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41579-023-00933-y\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSung J, Rajendraprasad SS, Philbrick KL, Bauer BA, Gajic O, Shah A et al (2024) The human gut microbiome in critical illness: disruptions, consequences, and therapeutic frontiers. J Crit Care 79:154436. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.jcrc.2023.154436\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.jcrc.2023.154436\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBosnjak M, Karpe AV, Van TTH, Kotsanas D, Jenkin GA, Costello SP et al (2023) Multi-omics analysis of hospital-acquired diarrhoeal patients reveals biomarkers of enterococcal proliferation and Clostridioides difficile infection. Nat Commun 14(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41467-023-43671-8\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41467-023-43671-8\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLi P, Chang XL, Chen XY, Wang CA, Shang Y, Zheng DY et al (2022) Early-life antibiotic exposure increases the risk of childhood overweight and obesity in relation to dysbiosis of gut microbiota: a birth cohort study. Ann Clin Microbiol Antimicrob 21(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s12941-022-00535-1\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s12941-022-00535-1\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eXu ZL, Jiang W, Huang WL, Lin Y, Chan FKL, Ng SC (2022) Gut microbiota in patients with obesity and metabolic disorders - a systematic review. Genes Nutr 17(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s12263-021-00703-6\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s12263-021-00703-6\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eDahal RH, Kim S, Kim YK, Kim ES, Kim J (2023) Insight into gut dysbiosis of patients with inflammatory bowel disease and ischemic colitis. Front Microbiol 14. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3389/fmicb.2023.1174832\\u003c/span\\u003e\\u003cspan address=\\\"10.3389/fmicb.2023.1174832\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMcCallin S, Drulis-Kawa Z, Ferry T, Pirnay JP, Nir-Paz R, Antibacterials E-E S. G. N. (2025). Phages and phage-borne enzymes as new antibacterial agents. Clin Microbiol Infect 31(6):910\\u0026ndash;921. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.cmi.2023.10.018\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.cmi.2023.10.018\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAriaeenejad S, Gharechahi J, Shahraki MF, Atanaki FF, Han JL, Ding XZ et al (2024) Precision enzyme discovery through targeted mining of metagenomic data. Nat Prod Bioprospecting 14(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1007/s13659-023-00426-8\\u003c/span\\u003e\\u003cspan address=\\\"10.1007/s13659-023-00426-8\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSantos CA, Morais MAB, Mandelli F, Lima EA, Miyamoto RY, Higasi PMR et al (2025) A metagenomic 'dark matter' enzyme catalyses oxidative cellulose conversion. Nature 639(8056). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41586-024-08553-z\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41586-024-08553-z\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFern\\u0026aacute;ndez-Ruiz I, Coutinho FH, Rodriguez-Valera F (2018) Thousands of novel endolysins discovered in uncultured phage genomes. Front Microbiol 9. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3389/fmicb.2018.01033\\u003c/span\\u003e\\u003cspan address=\\\"10.3389/fmicb.2018.01033\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBoulay A (2024) sublyme\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBoulay A (2025) \\u003cem\\u003eDevelopment of computational tools for the study of bacteriophages and their proteins\\u003c/em\\u003e. master's thesis, Universit\\u0026eacute; Laval\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eElnaggar A, Heinzinger M, Dallago C, Rehawi G, Wang Y, Jones L et al (2022) ProtTrans: Toward understanding the language of life through self-supervised learning. IEEE Trans Pattern Anal Mach Intell 44(10):7112\\u0026ndash;7127. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1109/TPAMI.2021.3095381\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/TPAMI.2021.3095381\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eCriel B, Taelman S, Van Criekinge W, Stock M, Briers Y (2021) PhaLP: A Database for the Study of Phage Lytic Proteins and Their Evolution. Viruses-Basel 13(7). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3390/v13071240\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/v13071240\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWagner TM, Poentinen AK, Fenzel CK, Engi D, Janice J, Almeida-Santos AC et al (2024) Interactions between commensal Enterococcus faecium and Enterococcus lactis and clinical isolates of Enterococcus faecium. 5. Fems Microbes\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/femsmc/xtae009\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/femsmc/xtae009\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHalimi F, Vanderhaeghen T, Timmermans S, Croubels S, Libert C, Vandewalle J (2024) Comparative analysis of hepatic gene expression profiles in murine and porcine sepsis models. Int J Mol Sci 25(20). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3390/ijms252011079\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/ijms252011079\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eConcei\\u0026ccedil;ao-Neto N, Zeller M, Lefr\\u0026egrave;re H, De Bruyn P, Beller L, Deboutte W et al (2015) Modular approach to customise sample preparation procedures for viral metagenomics: a reproducible protocol for virome analysis. Sci Rep 5. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/srep16532\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/srep16532\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBrown T, Tariq MA, Haagmans R, Ansorge R, Telatin A, Adriaenssens E (n.d.). Viromics of faecal samples. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://quadram.ac.uk/viromics-specific-protocols/\\u003c/span\\u003e\\u003cspan address=\\\"https://quadram.ac.uk/viromics-specific-protocols/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eDe Coninck L, Faye L, Basler N, Jansen D, Van Espen L (2024) Viper. (version 2.2 ed.)\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBolger AM, Lohse M, Usadel B (2014) Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30(15):2114\\u0026ndash;2120. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/bioinformatics/btu170\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/bioinformatics/btu170\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNurk S, Meleshko D, Korobeynikov A, Pevzner PA (2017) metaSPAdes: a new versatile metagenomic assembler. Genome Res 27(5):824\\u0026ndash;834. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1101/gr.213959.116\\u003c/span\\u003e\\u003cspan address=\\\"10.1101/gr.213959.116\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eCamacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K et al (2009) BLAST plus: architecture and applications. BMC Bioinformatics 10. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/1471-2105-10-421\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/1471-2105-10-421\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNayfach S, Camargo AP, Schulz F, Eloe-Fadrosh E, Roux S, Kyrpides NC (2021) CheckV assesses the quality and completeness of metagenome-assembled viral genomes. Nat Biotechnol 39(5):578\\u0026ndash;. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41587-020-00774-7\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41587-020-00774-7\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eVasimuddin M, Misra S, Li H, Aluru S, Ieee (2019) Efficient architecture-aware acceleration of BWA-MEM for multicore systems. In \\u003cem\\u003e33rd IEEE International Parallel and Distributed Processing Symposium (IPDPS), Rio de Janeiro, BRAZIL, May 20\\u0026ndash;24 2019 2019\\u003c/em\\u003e (pp. 314\\u0026ndash;324, International Parallel and Distributed Processing Symposium IPDPS). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1109/ipdps.2019.00041\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/ipdps.2019.00041\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBuchfink B, Reuter K, Drost HG (2021) Sensitive protein alignments at tree-of-life scale using DIAMOND. Nat Methods 18(4):366\\u0026ndash;. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41592-021-01101-x\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41592-021-01101-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eCamargo AP, Roux S, Schulz F, Babinski M, Xu Y, Hu B et al (2024) Identification of mobile genetic elements with geNomad. Nat Biotechnol 42(8). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41587-023-01953-y\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41587-023-01953-y\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHyatt D, Chen GL, LoCascio PF, Land ML, Larimer FW, Hauser LJ (2010) Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC Bioinformatics 11. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/1471-2105-11-119\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/1471-2105-11-119\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFu LM, Niu BF, Zhu ZW, Wu ST, Li WZ (2012) CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics 28(23):3150\\u0026ndash;3152. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/bioinformatics/bts565\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/bioinformatics/bts565\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eOsorio D, Rond\\u0026oacute;n-Villarreal P, Torres R (2015) Peptides: a package for data mining of antimicrobial peptides. R J 7(1):4\\u0026ndash;14. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.32614/RJ-2015-001\\u003c/span\\u003e\\u003cspan address=\\\"10.32614/RJ-2015-001\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSteinegger M, S\\u0026ouml;ding J (2017) MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nat Biotechnol 35(11):1026\\u0026ndash;1028. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/nbt.3988\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/nbt.3988\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eDemain AL, Vaishnav P (2009) Production of recombinant proteins by microbes and higher organisms. Biotechnol Adv 27(3):297\\u0026ndash;306. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.biotechadv.2009.01.008\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.biotechadv.2009.01.008\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFrancis DM, Page R (2010) Strategies to optimize protein expression in E. coli. Current protocols in protein science. 61(1):5 21-25.24. 29. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/0471140864.ps0524s61\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/0471140864.ps0524s61\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAbramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A et al (2024) Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630(8016). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41586-024-07487-w\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41586-024-07487-w\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBoulay A, Cremelie E, Galiez C, Briers Y, Rousseau E, V\\u0026aacute;zquez R (2025) SPAED: harnessing AlphaFold output for accurate segmentation of phage endolysin domains. Bioinformatics 41(10). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/bioinformatics/btaf531\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/bioinformatics/btaf531\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSwenson JM, Clark NC, Sahm DF, Ferraro MJ, Doern G, Hindler J et al (1995) Molecular characterization and multilaboratory evaluation of Enterococcus faecalis ATCC 51299 for quality control of screening tests for vancomycin and high-level aminoglycoside resistance in enterococci. J Clin Microbiol 33(11):3019\\u0026ndash;3021. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1128/jcm.33.11.3019-3021.1995\\u003c/span\\u003e\\u003cspan address=\\\"10.1128/jcm.33.11.3019-3021.1995\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBhatta J (2022) Isolation and characterization of lytic bacteriophages with therapeutic potential for the treatment of bacterial infections. University of Antwerp, University of Antwerp\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJim\\u0026eacute;nez E, Ladero V, Chico I, Maldonado-Barrag\\u0026aacute;n A, L\\u0026oacute;pez M, Mart\\u0026iacute;n V et al (2013) Antibiotic resistance, virulence determinants and production of biogenic amines among enterococci from ovine, feline, canine, porcine and human milk. BMC Microbiol 13. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/1471-2180-13-288\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/1471-2180-13-288\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGerstmans H, Grimon D, Guti\\u0026eacute;rrez D, Lood C, Rodr\\u0026iacute;guez A, van Noort V et al (2020) A VersaTile-driven platform for rapid hit-to-lead development of engineered lysins. Sci Adv 6(23). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1126/sciadv.aaz1136\\u003c/span\\u003e\\u003cspan address=\\\"10.1126/sciadv.aaz1136\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGreen R, Rogers EJ (2013) Transformation of chemically competent E. coli. In: Lorsch J (ed) Methods in Enzymology, vol 529. Academic, pp 329\\u0026ndash;336\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBenchling (n.d.). Benchling [Biology software]. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://benchling.com\\u003c/span\\u003e\\u003cspan address=\\\"https://benchling.com\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. Accessed 2023\\u0026ndash;2025\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003ePosit Team (2023) RStudio: Integrated Development Environment for R. (R version 4.3.3 ed). Posit Software, PBC, Boston, MA\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eR Core Team (2023) R: A language and environment for statistical computing. (R version 4.3.3. R Foundation for Statistical Computing, Vienna, Austria, R Foundation for Statistical Computing\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWickham H (2016) Data analysis. ggplot2: elegant graphics for data analysis. Springer, pp 189\\u0026ndash;201\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWickham H (2016) ggplot2: elegant graphics for data analysis. ggplot2: elegant graphics for data analysis. Springer-, New York, pp 189\\u0026ndash;201\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWickham H, Bryan J, Kalicinski M, Valery K, Leitienne C, Colbert B et al (2019) Readxl: read excel files. (R package version 1.4.5 ed., Vol. 13)\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSignorell A (2025) (R package version 0.99.59 ed.)\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWickham H, Fran\\u0026ccedil;ois R, Henry L, M\\u0026uuml;ller K, Vaughan D (2023) (R package version 1.1.4 ed.)\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eM\\u0026uuml;ller K, Wickham H (2023) Tibble: Simple data frames. (R package version 3.2.1 ed.)\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMair P, Wilcox R (2020) Robust statistical methods in R using the WRS2 package. Behav Researcb Methods 52(2):464\\u0026ndash;488. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3758/s13428-019-01246-w\\u003c/span\\u003e\\u003cspan address=\\\"10.3758/s13428-019-01246-w\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHothorn T, Bretz F, Westfall P (2008) Simultaneous inference in general parametric models. Biom J 50(3):346\\u0026ndash;363. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/bimj.200810425\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/bimj.200810425\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWilcox RR, Tian TS (2011) Measuring effect size: a robust heteroscedastic approach for two or more groups. J Applied Statistics 38(7):1359\\u0026ndash;1368. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1080/02664763.2010.498507\\u003c/span\\u003e\\u003cspan address=\\\"10.1080/02664763.2010.498507\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eV\\u0026aacute;zquez R, Garc\\u0026iacute;a E, Garc\\u0026iacute;a P (2021) Sequence-function relationships in phage-encoded bacterial cell wall lytic enzymes and their implications for phage-derived product design. J Virol 95(14). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1128/JVI.00321-21\\u003c/span\\u003e\\u003cspan address=\\\"10.1128/JVI.00321-21\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKyte J, Doolittle RF (1982) A simple method for displaying the hydropathic character of a protein. J Mol Biol 157(1):105\\u0026ndash;132. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/0022-2836(82)90515-0\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/0022-2836(82)90515-0\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eIkai A (1980) Thermostability and aliphatic index of globular proteins. J Biochem 88(6):1895\\u0026ndash;1898. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/oxfordjournals.jbchem.a133168\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/oxfordjournals.jbchem.a133168\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJones P, Binns D, Chang HY, Fraser M, Li WZ, McAnulla C et al (2014) InterProScan 5: genome-scale protein function classification. Bioinformatics 30(9):1236\\u0026ndash;1240. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/bioinformatics/btu031\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/bioinformatics/btu031\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSchleifer KH, Kandler O (1972) Peptidoglycan types of bacterial cell walls and their taxonomic implications. Bacteriological reviews 36(4):407\\u0026ndash;477. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1128/br.36.4.407-477.1972\\u003c/span\\u003e\\u003cspan address=\\\"10.1128/br.36.4.407-477.1972\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eChamakura KR, Tran JS, O'Leary C, Lisciandro HG, Antillon SF, Garza KD et al (2020) Rapid de novo evolution of lysis genes in single-stranded RNA phages. Nat Commun 11(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41467-020-19860-0\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41467-020-19860-0\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eChamakura K, Young R (2019) Phage single-gene lysis: Finding the weak spot in the bacterial cell wall. J Biol Chem 294(10):3350\\u0026ndash;3358. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1074/jbc.TM118.001773\\u003c/span\\u003e\\u003cspan address=\\\"10.1074/jbc.TM118.001773\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eReed CA, Langlais C, Wang IN, Young R (2013) A2 expression and assembly regulates lysis in Qβ infections. Microbiology-Sgm 159:507\\u0026ndash;514. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1099/mic.0.064790-0\\u003c/span\\u003e\\u003cspan address=\\\"10.1099/mic.0.064790-0\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eH\\u0026ouml;r J (2025) Advancing RNA phage biology through meta-omics. Nucleic Acids Res 53(8). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/nar/gkaf314\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/nar/gkaf314\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eUrayama SI, Fukudome A, Hirai M, Okumura T, Nishimura Y, Takaki Y et al (2024) Double-stranded RNA sequencing reveals distinct riboviruses associated with thermoacidophilic bacteria from hot springs in Japan. Nat Microbiol 9(2):514\\u0026ndash;523. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41564-023-01579-5\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41564-023-01579-5\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSz\\u0026eacute;kely AJ, Breitbart M (2016) Single-stranded DNA phages: from early molecular biology tools to recent revolutions in environmental microbiology. FEMS Microbiol Lett 363(6). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/femsle/fnw027\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/femsle/fnw027\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKosmopoulos JC, Klier KM, Langwig MV, Tran PQ, Anantharaman K (2024) Viromes vs. mixed community metagenomes: choice of method dictates interpretation of viral community ecology. Microbiome 12(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s40168-024-01905-x\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s40168-024-01905-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWu LY, Wijesekara Y, Piedade GJ, Pappas N, Brussaard CPD, Dutilh BE (2024) Benchmarking bioinformatic virus identification tools using real-world metagenomic data across biomes. Genome Biol 25(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s13059-024-03236-4\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s13059-024-03236-4\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGeorgakis N, Premetis GE, Pantiora P, Varotsou C, Bodourian CS, Labrou NE (2025) The impact of metagenomic analysis on the discovery of novel endolysins. Appl Microbiol Biotechnol 109(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1007/s00253-025-13513-2\\u003c/span\\u003e\\u003cspan address=\\\"10.1007/s00253-025-13513-2\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNie WC, Qiu TY, Wei YW, Ding H, Guo ZX, Qiu JX (2024) Advances in phage-host interaction prediction: in silico method enhances the development of phage therapies. Brief Bioinform 25(3). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/bib/bbae117\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/bib/bbae117\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eRoux S, Camargo AP, Coutinho FH, Dabdoub SM, Dutilh BE, Nayfach S et al (2023) iPHoP: An integrated machine learning framework to maximize host prediction for metagenome-derived viruses of archaea and bacteria. PLoS Biol 21(4). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1371/journal.pbio.3002083\\u003c/span\\u003e\\u003cspan address=\\\"10.1371/journal.pbio.3002083\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLiu H, Hu Z, Li MY, Yang Y, Lu SG, Rao XC (2023) Therapeutic potential of bacteriophage endolysins for infections caused by Gram-positive bacteria. J Biomed Sci 30(1). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s12929-023-00919-1\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s12929-023-00919-1\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFu YR, Yu ST, Li JF, Lao ZS, Yang XF, Lin ZL (2024) DeepMineLys: Deep mining of phage lysins from human microbiome. Cell Rep 43(8). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.celrep.2024.114583\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.celrep.2024.114583\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eCremelie E, V\\u0026aacute;zquez R, Briers Y (2024) A comparative guide to expression systems for phage lysin production. Essays Biochem 68(5):645\\u0026ndash;659. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1042/EBC20240019\\u003c/span\\u003e\\u003cspan address=\\\"10.1042/EBC20240019\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBalaban CL, Su\\u0026aacute;rez CA, Boncompain CA, Peressutti-Bacci N, Ceccarelli EA, Morbidoni HR (2022) Evaluation of factors influencing expression and extraction of recombinant bacteriophage endolysins in Escherichia coli. Microb Cell Fact 21(1):40. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s12934-022-01766-9\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s12934-022-01766-9\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eShockman GD (1992) The autolytic (\\u0026lsquo;suicidase\\u0026rsquo;) system of Enterococcus hirae: From lysine depletion autolysis to biochemical and molecular studies of the two muramidases of Enterococcus hirae ATCC 9790. FEMS Microbiol Lett 100(1\\u0026ndash;3):261\\u0026ndash;267. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1111/j.1574-6968.1992.tb14050.x\\u003c/span\\u003e\\u003cspan address=\\\"10.1111/j.1574-6968.1992.tb14050.x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"},{\"header\":\"Footnotes\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003e\\u003csup\\u003e1\\u003c/sup\\u003eMerelbeke: Department of Pathobiology, Pharmacology and Zoological Medicine, Faculty of Veterinary Medicine, Ghent University, 9820 Merelbeke, Belgium. Pigs were part of a pig sepsis model, for which the study design was approved by the Ethical Committee of the Faculty of Veterinary Medicine and the Faculty of Bioscience Engineering of Ghent University (EC 2017/24) [\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e]\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003e ILVO: Instituut voor Landbouw-, Visserij- en Voedingsonderzoek, Varkenscampus, Van Gansberghelaan 92/1, 9820 Merelbeke-Melle, Belgium\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003e UZ Gent: Universitair Ziekenhuis Gent, Corneel Heymanslaan 10, 9000 Ghent, Belgium.\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"probiotics-and-antimicrobial-proteins\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"paap\",\"sideBox\":\"Learn more about [Probiotics and Antimicrobial Proteins](http://link.springer.com/journal/12601)\",\"snPcode\":\"12602\",\"submissionUrl\":\"https://submission.nature.com/new-submission/12602/3\",\"title\":\"Probiotics and Antimicrobial Proteins\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"em\",\"reportingPortfolio\":\"Springer Hybrid\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false},\"keywords\":\"sequence-based metagenomics, bacteriophage, lysin, Enterococcus\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-8328502/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-8328502/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003ePhage lysins, enzymes encoded by bacteriophages that degrade bacterial cell walls, are emerging as a promising class of antimicrobial agents. This study aimed to discover novel lysins with activity against \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species using a sequence-based metagenomic discovery pipeline. Viral metagenomic DNA was extracted and sequenced from five environmental samples originating from pig feces or sewage. Putative lysins were first predicted with SUBLYME, a protein embedding-based classifier. Subsequently, a specific protein embedding-based classifier was developed to predict lysins with potential activity against \\u003cem\\u003eEnterococcus\\u003c/em\\u003e.\\u003c/p\\u003e \\u003cp\\u003eA total of 8 825 candidate lysins were predicted, including 129 with potential anti-enterococcal activity. Comparative analysis revealed differences in domain architectures and physicochemical properties between lysins derived from fecal and sewage samples, suggesting distinct phage host origins. A subset of the predicted lysins was expressed in \\u003cem\\u003eEscherichia coli\\u003c/em\\u003e, partially purified and tested for muralytic activity against three enterococcal species (\\u003cem\\u003eEnterococcus faecium\\u003c/em\\u003e, \\u003cem\\u003eEnterococcus faecalis\\u003c/em\\u003e, and \\u003cem\\u003eEnterococcus hirae\\u003c/em\\u003e). Among the 21 expressed lysins with variable expression yields, four exhibited lytic activity against all three \\u003cem\\u003eEnterococcus\\u003c/em\\u003e species, two were active against \\u003cem\\u003eEnt. faecalis\\u003c/em\\u003e and \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e, and seven showed activity exclusively against \\u003cem\\u003eEnt. hirae\\u003c/em\\u003e. Six of these active proteins contained previously unreported domain architectures, indicating that this approach can uncover structurally novel functional lysins.\\u003c/p\\u003e \\u003cp\\u003eWhile this pipeline was applied to \\u003cem\\u003eEnterococcus\\u003c/em\\u003e, it is broadly adaptable for the discovery of lysins targeting other bacterial pathogens, offering a scalable approach to expand the antimicrobial arsenal.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Discovery of anti-enterococcal phage lysins from environmental metagenomes using protein embedding-based classification\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-12-11 06:58:39\",\"doi\":\"10.21203/rs.3.rs-8328502/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"submitted\",\"content\":\"Probiotics and Antimicrobial Proteins\",\"date\":\"2025-12-10T14:03:05+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"probiotics-and-antimicrobial-proteins\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"paap\",\"sideBox\":\"Learn more about [Probiotics and Antimicrobial Proteins](http://link.springer.com/journal/12601)\",\"snPcode\":\"12602\",\"submissionUrl\":\"https://submission.nature.com/new-submission/12602/3\",\"title\":\"Probiotics and Antimicrobial Proteins\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"em\",\"reportingPortfolio\":\"Springer Hybrid\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false}}],\"origin\":\"\",\"ownerIdentity\":\"1cee210e-1075-4c58-a8b8-f3da9159ceed\",\"owner\":[],\"postedDate\":\"December 11th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"under-review\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2026-04-17T06:25:53+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-12-11 06:58:39\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-8328502\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-8328502\",\"identity\":\"rs-8328502\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}