Directed evolution of phage Romulus in biofilm-embedded Staphylococcus aureus: mutations in baseplate proteins enhanced its antibiofilm activity

Abstract

The ability of Staphylococcus aureus to form biofilm and the emergence of multidrug-resistant strains make staphylococcal infections often chronic and difficult to treat. To face these challenges, alternative or adjunct strategies to antibiotics are urgently required. In this context, phage therapy gained renewed interest as promising approach to target multidrug-resistant bacteria. To enhance their efficacy as natural phages, they can undergo directed evolution via serial host passages. To date, most protocols focus on planktonic cultures, while the effects towards biofilm-targeted evolution remain poorly explored.

Our study aims at investigating the potential of a new directed evolution protocol designed to specifically enhance the efficacy of phage Romulus to target staphylococcal sessile communities and to identify whether specific phage proteins are involved in this process.

The method involved 31 serial passages with a two-step incubation: 1 h for phage adsorption and infection, followed by 8 h for its amplification. Mutant phages were isolated, sequenced, and phenotypically characterised.

Mutations emerged in two baseplate proteins (gp54 and gp58), involved in host adsorption. Three mutants (R31, R31p2, R31p5) showed enhanced bactericidal activity against planktonic cells and improved efficacy against biofilm, achieving up to a 4-log₁₀ reduction. While their host range remained consistent with the wildtype, phage Romulus mutants exhibited higher efficiency of plating against the nine out of 21 sensitive S. aureus strains.

Overall, our results underscore the potential of biofilm-adapted phages to improve phage efficacy towards both planktonic and sessile cells, without impacting on the phage host range. The analysis of mutations suggested that the baseplate plays a crucial role in targeting biofilm-embedded cells, even if further investigation is necessary to explain the molecular basis responsible for the enhanced lytic efficacy.

Highlights

• •Directed evolution enhanced phage Romulus antibiofilm efficacy.
• •Mutations in baseplate proteins improved biofilm-targeting ability.
• •Phage mutants determined up to 4-log₁₀ CFU/ml reduction in biofilms.
• •Directed evolution improved lytic activity without altering host range.

1. Introduction

A hallmark trait of Staphylococcus aureus, particularly in chronic infections, is its ability to form biofilms, communities of sessile cells embedded within a self-produced extracellular polymeric substance (EPS) matrix, with the ability to adhere to both biotic and abiotic surfaces [1]. The EPS matrix comprises a complex mixture of exopolysaccharides, extracellular proteins, extracellular DNA (eDNA), lipids, surfactants, and water, conferring viscoelasticity to the biofilm and contributing up to 90 % of biofilm dry weight [2]. Biofilms are often associated with medical devices, such as prosthetic joints, catheters and cardiac valves [3], posing significant challenges to clinical treatment. Biofilm-associated infections are more likely to persist as chronic conditions, mainly due to their recalcitrance towards antibiotic treatments. On the one hand, they exhibit increased resistance to antibiotics, because of the presence of the additional EPS physical barrier [4]. On the other hand, this tolerance generally results from altered physiological states, such as nutrient limitation-induced dormancy and the existence of persister cells, metabolically inactive subpopulations that can survive otherwise lethal antibiotic therapy [5]. The treatment of biofilm-associated infections is even more challenging because of the increasing global problem of antimicrobial resistance (AMR). According to current estimations, the impact of AMR will lead to 1.91 annual deaths attributable to AMR globally and 8.22 million annual deaths associated with AMR, including more than 100,000 deaths attributable to methicillin-resistant S. aureus (MRSA) [6].

In this context, the need to design and optimise alternative strategies to target bacterial infections has utmost priority. In the last twenty years, the threat of AMR together with the advances in research techniques (e.g. high-throughput sequencing, metagenomics, genetic engineering, and synthetic biology), has driven the reconsideration of phage therapy as suitable strategy in multidrug-resistant bacterial infection treatment, especially those associated to biofilms [7]. A critical appraisal of this approach is demonstrated by the increasing number of successful treatments and clinical trials worldwide [[8], [9], [10], [11], [12], [13]].

Despite the potential of phages in therapeutic application, phage efficacy might be impaired by the bacterial ability to develop resistance, similar to what happens with antibiotics. Nevertheless, the advantage of phages over antibiotics lies in their ability to naturally co-evolve with their host, potentially overcoming bacterial resistance in a continuous and dynamic equilibrium. This distinct feature is the rationale behind phage (pre)adaptation or training protocols. Viral populations are trained to anticipate how the target bacterial will evolve phage resistance, by allowing phages to accumulate random mutations. One of the most common approaches is based on the coevolution of phages and bacteria during serial passages. For example, this strategy was applied to evolve a strictly lytic variant of phage lambda against E. coli, determining a reduced bacterial expression of the receptor LamB which led the phage to infect targeting a secondary receptor (OmpF) [14]. Moreover, it has been demonstrated that phage training protocols for obtaining Pseudomonas aeruginosa phage variants led to a tenfold greater lytic efficiency than the ancestral naturally isolated phage, towards not only the natural host, but also a panel of a further 20 clinical isolates [15]. Hence, one of the most desired features of phages for therapeutic purposes is a sufficiently wide host range, with the aim to use the same phage preparation to target several strains. One of the most common and successful approaches, currently applied to evolve phages targeting different strains, is the Appelmans protocol. This strategy relies on combining multiple phages and challenging them against a panel of both sensitive and not-sensitive bacterial isolates, for multiple evolution rounds. The progressive accumulation of mutations, together with the recombination occurring between phages, increases the chance to widen the host range making the insensitive strains susceptible to the emerged viral variants [16]. The different strategies to drive phage evolution suggest that this approach is a powerful tool to increase phage fitness according to different aims. Nevertheless, it is crucial to be aware that the outcomes of (co)evolution assays are often unpredictable because of possible evolutionary scenarios. For instance, there are evidence of trade-offs and fitness costs emerging while applying training protocols, such a contraction of the host range as a drawback of an increased growth rate [17,18]. For this reason, phage training outcomes should be evaluated according to the specific phage-bacteria combinations, without assuming predictable evolutionary trajectories.

In this study, we developed a directed evolution protocol to enhance the infectivity of the staphylococcal phage Romulus, a member of the Silviavirus genus, specifically targeting biofilm-embedded S. aureus cells. Our protocol was tailored to shape the phage population by increasing its fitness and infective capacity specifically towards sessile cells. Interestingly, while most of the directed evolution methods have relied on planktonic cultures of bacteria [14,19,20] such environments do not mimic the complex selective conditions characteristic of biofilms. Biofilm populations impose different constraints on phage infection, e.g., reduced diffusion, altered bacterial metabolism, and the presence of an extracellular polymeric matrix that can impede adsorption and replication. We hypothesised that S. aureus phage Romulus propagating in biofilm-associated cells would be more likely to favour the selection of variants with specific adaptations to bypass these biofilm-mediated limitations. This pressure may also impact phage-encoded depolymerases, enzymes capable of hydrolysing extracellular polysaccharides and providing phage access to embedded bacterial cells, enhancing antibiofilm activity further [21]. This approach is predicted to grant phage populations with increased efficiency against sessile cells without compromising infectivity against planktonic populations, a core deficiency in current phage therapeutic practice.

Hence, based on our observations, phage infection dynamics in biofilm communities differ notably from planktonic infections, particularly in terms of timing and replication efficiency. These parameters are critical factors to consider when evaluating phage infectivity in the context of biofilm-associated infections.

2. Materials and methods

2.1. Bacterial strains and growth conditions

S. aureus PS47 strain (NCTC 8325) was used as phage host strain, while twenty other S. aureus strains were used for the efficiency of plating (EOP) test (Table S1). Three E. coli strains were used in this study, TOP10 (Thermo Fisher Scientific, Waltham, MA, USA), BL21 (DE3) (Thermo Fisher Scientific, Waltham, MA, USA) and C41 (DE3) OverExpress™ (Sigma-Aldrich®, Merck KgaA, Darmstadt, Germany) strains. Strains were cultured either in Brain Heart Infusion (BHI) or Lysogeny Broth (LB) medium or on 1.5 % agar plates. LB was supplemented with ampicillin (100 μg/ml) or kanamycin (50 μg/ml) for plasmid selection.

2.2. Staphylococcal biofilm formation and cell dislodging

S. aureus biofilms were formed on porous glass beads in 50-ml tubes by inoculating 10⁷ colony-forming units per millilitre (CFU/ml) of overnight grown bacterial suspensions (1 ml/bead), followed by incubation at 37 °C for 24 h [22]. After incubation, beads were washed three times with Phosphate-Buffered Saline (PBS: 137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, pH 7.4) to remove non-adherent planktonic cells, leaving only the surface-attached bacteria. Following treatment, each bead was again washed three times with PBS and transferred to a 2-ml centrifuge tube containing 1 ml of PBS. To dislodge biofilm-associated bacteria, the tubes were vortexed for 30 s, sonicated for 60 s, and vortexed again for 30 s. The resulting sonication fluids were subjected to tenfold serial dilutions, and 20 μl of each dilution was plated onto BHI agar. Plates were incubated overnight at 37 °C, and cell concentration was determined.

Biofilm biomass was quantified using a crystal violet staining assay, as previously described by Stepanović et al. [23]. Briefly, S. aureus overnight cultures were diluted 1:100 in fresh medium and 200 μl per well were dispensed into 96-well microtiter plates (EuroClone®, Italy). After incubation at 37 °C for 24 h, planktonic cells were removed, and wells were gently washed three times with PBS. The attached biofilm was stained with 0.1 % (w/v) crystal violet for 15 min at room temperature. After removal of the crystal violet solution, each well was washed twice with PBS and allowed to air-dry. The bound crystal violet was solubilized with 30 % (v/v) acetic acid. The absorbance of the resulting solution was measured at 570 nm using a microplate reader. The experiment was performed as three biological replicates with four technical replicates each.

2.3. Phage amplification and titration

The bacteriophages used in this study included phage Romulus (NCBI accession number PV821423) and its evolved variants. High-titer phage stocks were obtained by propagation on S. aureus PS47 strain, using the soft agar overlay method [24]. Briefly, 4 ml of 0.7 % BHI soft agar, supplemented with 300 μl of an overnight S. aureus PS47 culture and an equal volume (300 μl) of phage solution (10⁶ PFU/ml), was poured onto 1.5 % BHI agar plates and incubated at 37 °C for 24 h. After visible lysis, the soft agar layer was scraped off using a disposable plastic loop, collected in a 50-ml centrifuge tube, and resuspended in 10 ml of 1 × phage buffer (10 mM Tris-HCl, 10 mM MgSO₄, 150 mM NaCl, pH 7.5). The phage suspension was incubated at 4 °C for 3 h with continuous shaking (120 rpm) and then centrifuged at 4600×g for 20 min at 4 °C. The supernatant was filtered through a 0.22-μm PVDF syringe filter (Merck, Darmstadt, Germany) to remove bacterial contaminants and stored at 4 °C. To determine phage titers (PFU/ml), 5 μl of tenfold serially diluted phage suspensions in phage buffer were spotted onto a BHI soft agar lawn of S. aureus PS47, incubated overnight at 37 °C.

2.4. Phage directed evolution protocol against staphylococcal biofilm

To identify the phage life cycle in biofilm, a one-step growth curve versus sessile bacteria was carried out. Briefly, S. aureus PS47 biofilm was formed on porous glass beads for 24 h, as previously described. Each bead, after three consecutive washes in PBS, was transferred to a 2-ml microcentrifuge tube containing 1 ml of BHI of medium and inoculated with 10⁷ PFU/ml of phage Romulus. Samples were statically incubated at 37 °C for varying time periods (0, 5, 10, 30, 60, 120, 180, 240, 300, and 360 min). Beads were recovered after incubation and treated as mentioned above. Then, the sonication fluid of bacterial suspensions was plated to determine CFU/ml values. Two technical replicate samples of each timepoint were combined, centrifuged at 8000×g for 5 min to remove detached bacterial cells, and filtered using a 0.22-μm PVDF syringe filter before determining phage titers. Two independent experiments were carried out.

Directed evolution protocol was implemented as follows (Fig. 1). Biofilms were allowed to grow for 24 h on six glass beads in a 50-ml centrifuge tube. Following three PBS washes, one bead was removed to quantify biofilm formation, and the remaining five beads were added to 5 ml of BHI broth supplemented with 10⁷ PFU/ml of phage Romulus and incubated statically at 37 °C for 1 h. Subsequently, the phage-containing medium was discarded and replaced with 5 ml of fresh pre-warmed BHI medium and incubated statically at 37 °C for 8 h.

Fig. 1.

Two-step directed evolution protocol on S. aureus biofilm. A 24-h pre-formed S. aureus PS47 biofilm (on glass beads) (1) is treated for 1 h with 10⁷ PFU/ml phage Romulus (2); medium is discarded and replaced with fresh BHI medium (3) and the biofilm incubated for further 8 h (4); beads and phages are collected (the latter stored for the following evolution round) and CFU/ml and PFU/ml are assessed (5). (Created using Microsoft® PowerPoint v16.89.1).

After the second incubation, the culture medium was harvested, centrifuged at 4000×g for 20 duplicate, while bacterial cell concentrations were assayed from three beads, at the end of each round of evolution. Phages of each round end were used to infect a fresh S. aureus biofilm, following the aforementioned steps. The directed evolution experiment was performed as a single evolutionary line, initiated from a single ancestral phage population. This design aimed to establish a proof-of-concept protocol to explore the feasibility and phenotypic outcomes of phage adaptation under biofilm-associated conditions. Directed evolution protocol was performed for 31 rounds. The number of passages was determined empirically: phage lytic activity was monitored every ten rounds by CFU counting of sessile cells, and after round 31 a progressive reduction in CFU/ml was observed compared to earlier time points (R11, R21), which guided the decision to proceed with clones from this round. At the end of round 31, the lysate was diluted to 10³ PFU/ml and plated to obtain well-separated plaques; ten plaques, representing approximately one-third of the total formed, were selected based on size (R31p1–R31p10, from smallest to largest).

2.5. Genome DNA extraction, sequencing and analysis

Phage DNA was extracted using the Zymo Research DNA Clean & Concentrator™-5 kit after treating the phage lysate with DNAse I (Thermo Fisher Scientific, USA) and RNAse A (Thermo Fisher Scientific, USA) to remove bacterial genetic material and with proteinase K (Thermo Fisher Scientific, USA) to digest the capsid proteins and release phage genomic DNA [25]. Whole genome sequencing of both wildtype and mutant phages was performed twice at KU Leuven (Laboratory of Gene Technology) in Belgium. Illumina sequencing libraries were created using the Nextera Flex DNA Library Kit sequenced using an Illumina Miniseq device (2 x 150 bp reads). The raw data were submitted to BV-BRC online platform v3.6.12 [26] and their contig assembly was performed using Unicycler (v0.4.8) [27], following default settings. The genome of phage Romulus wildtype was re-annotated by using RASTtk [28], followed by manual functional annotation comparing BV-BRC predicted CDSs against the non-redundant GenBank protein database through BLASTp [29]. The mutations in the evolved phages were assessed by using the tool Snippy v4.6.0 [30], by mapping the reads obtained with the Illumina sequencing against the reference GenBank file of the wildtype phage. Phylogenetic relationships and genomic similarity among Staphylococcus aureus strains were assessed through a multi-step bioinformatics pipeline. Genomic annotation of bacterial DNA was performed using Prokka (v1.14.6) [31] with default parameters. Core genome alignment was conducted using Roary (v3.13.0) [32], applying default settings, including a minimum percentage identity of 95 % for BLASTp and a 99 % threshold for gene presence across isolates to be considered part of the core genome. Intergenomic distances were then calculated using RAxML (v8.2.12) [33], and an unrooted phylogenetic tree was constructed based on these distances. The resulting tree was visualized using the Interactive Tree of Life (iTOL) platform [34].

2.6. Lytic activity against planktonic and biofilm-embedded cells

The phage lytic activity of Romulus mutants derived from phage mix of round 31 was firstly assessed against planktonic cells. Exponential growing S. aureus PS47 cells (10⁶ CFU/ml) were treated with the wildtype and mutant phages with a multiplicity of infection (MOI) of 0.1, including an untreated control. Cell growth was assessed by CFU/ml counting, by plating cells 3, 6 and 24 h after phage treatment. The assay was performed with six independent biological replicates. To assess the phage lytic activity against sessile cells, pre-formed biofilms were treated with different phage titers (ranging from 10⁷ to 10⁹ PFU/ml) for 24 h at 37 °C, including an untreated control. Cell concentration was finally assessed as described above. The experiment was performed with five independent biological replicates.

2.7. Phage host range and efficiency of plating

The lytic activity of the wildtype and the mutated phages was determined by a spot test [35]. Briefly, 10 μl of eight tenfold serial dilutions of bacteriophage lysate (including the not diluted phage suspension) were spotted on the 0.7 % BHI soft agar bacterial lawn. For all phages, the efficiency of plating (EOP) was determined by dividing the phage titer calculated by spotting on the selected strains by the titer obtained with the host S. aureus PS47 strain [36]. The assay was performed in biological triplicate.

2.8. Phage adsorption assay

To determine the time required for the phage Romulus, wildtype (wt) and mutant variants, to attach to its host, adsorption assay was performed according to Kropinski with minor modifications [37]. Briefly, exponentially growing S. aureus PS47 cells were infected with the phage at an MOI of 0.001 in BHI (added with 5 mM of MgCl₂ and 5 mM CaCl₂) and statically incubated at 37 °C. At 0, 10, 15, 20 and 25 min post-infection, 200 μl of samples were collected from each experiment, combined with 10 μl of pre-cooled chloroform to kill the bacteria, and kept on ice. The titer of the non-adsorbed free phages in the supernatants collected at different time intervals was determined by double-layer plaque assay.

To assess phage adsorption on biofilm-associated cells, biofilm of S. aureus PS47 was formed as previously described. Single beads were transferred into 24-well plates and infected with 10⁴ PFU/ml phages, in a final volume of 1 ml per well. The plate was statically incubated at 37 °C, and 200 μl samples were collected at different time points (0, 5, 15, 30, 45, and 60 min), combined with 10 μl of pre-cooled chloroform to kill the bacteria, and kept on ice. The titer of non-adsorbed phages in the supernatants at each time interval was determined by the double-layer plaque assay, as described for the planktonic assay. The adsorption rate constant (k) was calculated using the equation $k=\frac{2.3}{B\times t}\mathit{log}(\frac{P{0}_0}{P}$) [38].

2.9. Gene cloning and protein expression

The gp58 baseplate protein was cloned into plasmid pET21b (EMD Biosciences, Inc. USA), by Gibson assembly method [39], and the plasmid was amplified in E. coli TOP10 strain (Thermo Fisher Scientific, USA). For the construction of pEJK144, the gp58 gene was amplified and introduced into a pET24d (Novagen) derivative modified for modular cloning via BsaI restriction sites. The plasmid was amplified in E. coli DH5a prior to its transformation into the expression strains E. coli BL21 (DE3) ClearColi™ and C41 (DE3) OverExpress™. The former one was used for the halo assay (see paragraph 2.10), while the latter one was used for protein purification and mass photometry analysis (MP).

Protein expression was carried out by growing E. coli C41 (DE3) in LB medium and supplementing with 1 % lactose to induce the expression at 30 °C for 18 h. Cells were harvested by centrifugation at 4000×g for 15 min at 4 °C and resuspended in buffer A (20 mM HEPES, 20 mM KCl, 200 mM NaCl, 40 mM imidazole, pH 8.0) and lysed using a microfluidizer (M110-L, Microfluidics). The lysate was transferred into and centrifuge at 20,000×g for 20 min at 4 °C. The protein was purified from the supernatant by nickel ion affinity chromatography and eluted with buffer B (20 mM HEPES, 20 mM KCl, 200 mM NaCl, 250 mM imidazole, pH 8.0). The eluted proteins were concentrated by centrifugation (10 kDa MWCO) and further polished by size-exclusion chromatography on an S6 XK16 column (Cytiva, Marlborough, USA) with a size-exclusion chromatography (SEC) buffer consisting of 20 mM HEPES-Na (pH 7.5), 200 mM NaCl, and 20 mM KCl. Visual analysis of the purified protein was performed by SDS-PAGE.

2.10. Halo assay

E. coli BL21 (DE3) transformants were screened for depolymerase activity on a 2 % LB agar plate containing 0.05 mM IPTG, the necessary selectable markers, and 1 % S. aureus cell wall substrate. This substrate was prepared by autoclaving a 300 ml overnight culture of the S. aureus PS47 strain followed by centrifugation (4000×g, 30’, 4 °C). Next, the pellet was washed in 20 ml sterile PBS and centrifuged under the same conditions. Finally, the pellet was resuspended in the amount of PBS needed to obtain a suspension of 0.2 g of substrate per ml [40].

2.11. Mass photometry analysis

The oligomeric state of gp58 was analyzed using mass photometry (MP). Measurements were conducted with a TwoMP mass photometer (Refeyn Ltd., UK) operated via AcquireMP v2023 R¹.1 software (Refeyn Ltd.). Microscope coverslips (No. 1.5H, 24 × 60 mm; Carl Roth) and CultureWell Reusable Gaskets (CW-50R-1.0, 3 × 1 mm; Grace Bio-Labs, USA) were cleaned by alternating rinses with isopropanol and ultrapure water (Milli-Q, Merck, Germany) three times, followed by air-drying under a stream of compressed air. Cleaned coverslips and gaskets were assembled and mounted on the instrument using immersion oil (Immersol™ 519F; Carl Zeiss, Germany). For each measurement, 18.5 μl of SEC buffer (20 mM HEPES, 20 mM KCl, 150 mM NaCl, pH 7.5) was applied to a single gasket well and brought into focus using the “Droplet Dilution” mode. Next, 1.5 μl of pre-diluted protein solution (750 nM) was added to the drop, yielding a final concentration of 56 nM. The mass photometer was calibrated with a custom molecular weight standard consisting of proteins ranging from 86 kDa to 344 kDa. Measurements were recorded over 60 s at 100 frames per second. Data were analyzed using DiscoverMP v2023 R¹.2 software (Refeyn Ltd.). Peaks corresponding to distinct molecular species were manually selected and fitted with Gaussian distributions. The resulting peak areas and their relative proportions were used to quantify the oligomeric states of gp58.

2.12. Functional and structural analysis of mutated proteins

Functional and structural protein predictions were used to determine whether a protein sequence or part of it shared similarities with known structures in the Protein Data Bank (PDB) [41]. First, the Basic Local Alignment Search Tool for Protein (BLASTp) v2.15.0 algorithm in the NCBI database was used to infer biological functions of novel protein sequences from annotation [42]. Results were interpreted using metrics including E-value and identity percentage to predict functionally similar proteins. Next, HmmerWeb v2.43 software was used to identify conserved domains within protein sequences [43,44]. Profile Hidden Markov Models (HMMs) were built from Pfam alignments to identify existing protein domains. Query sequences were retrieved in FASTA and searched against the Pfam HMMs, with an appropriate default E-value cut-off of 0.01 [45]. The HMMER search provided information on the type, number and location of Pfam domains within the query sequences, together with annotations. In addition, the HHPRED v2.08 software tool was used to identify conserved domains and biological functions of the proteins of interest. HHPRED enables remote homology detection and structure prediction by pairwise comparison of HMMs [46]. It iteratively builds a multiple sequence alignment (MSA) for the query protein, similar to PSI-BLAST [45,47], to capture evolutionary relationships of homologous proteins. Search options have been set to default. Moreover, a prediction of putative depolymerases was carried out through the tool PhageDPO [48]. Finally, three-dimensional structure predictions were performed through AlphaFold 3, running the analysis on of AlphaFold Server (beta) with default settings [49].

2.13. Data analysis

Data were analyzed using appropriate statistical tests and plotted using GraphPad Prism v9.0.0. Data normality was evaluated using the Shapiro–Wilk test (α = 0.05). For the bactericidal assay on biofilm, the activity of each mutant phage was compared with that of the ancestral wildtype phage using a one-way ANOVA followed by Dunnett's post-hoc test. Similarly, in the adsorption assay, the adsorption efficiency of mutant phages at each time point was compared with that of the wildtype phage using one-way ANOVA followed by Dunnett's post-hoc test. Significance thresholds for each analysis are reported in the figure legends. Circos plot was created using circlize v0.4.11 [50].

3. Results

3.1. Elucidating bacteriophage life cycle in staphylococcal biofilms

Biofilm formation, a key virulence determinant of Staphylococcus aureus, plays a crucial role in its ability to colonize prosthetic material and resist both antibiotic treatment and host immune clearance [51].

Before investigating the infection dynamics, the ability of S. aureus PS47 to form biofilm was evaluated to ensure that the experimental setup reflected biofilm-associated conditions. Quantification of biofilm biomass by crystal violet staining revealed that PS47 is a strong biofilm producer, even exceeding the biomass formed by the reference strain S. aureus ATCC 43300. This strain is widely recognized as a potent biofilm producer (Fig. S1). This confirmed that PS47 is a suitable model for studying phage–biofilm interactions.

To establish a protocol for the directed evolution of phages targeting bacterial biofilms and to select phage mutants with enhanced antibiofilm activity, key kinetics parameters of phage Romulus infection of biofilm-associated S. aureus PS47 cells were determined. Specifically, we assessed the timing of the adsorption phase of the phage to sessile bacterial cells and the timing of novel phage particle release following infection. These data were crucial for defining the timing parameters of the subsequent evolution protocol. A one-step growth curve (OSGC)-like experiment was performed to characterise the eclipse phase duration in biofilm-associated infections. The results revealed a prolonged eclipse phase of approximately 120 min in biofilm embedded cells, in stark contrast to the 20-min eclipse phase observed during infection of planktonic cells [52] (Fig. 2). Following this initial eclipse phase, novel phage particles began to be released, with the exponential phase of replication extending beyond 6 h.

Fig. 2.

OSGC-like of phage Romulus on S. aureus PS47 biofilm. Pre-formed 24-h biofilm was treated with 10⁷ PFU/ml of phage Romulus, incubated at 37 °C, and sampled at the timepoints plotted on x-axis. On y-axis the phage titer (PFU/ml) evaluated at different timepoints post-infection; n = 2 independent biological replicates.

3.2. Genome analysis of phage mutants

To improve phage activity against sessile bacteria, a novel phage-directed evolution protocol in biofilm was set up. Following the 31-round evolution protocol, evolved phages were collected. Monitoring of CFU/ml during the evolution process showed that the average CFU decreased from 8.23 × 10⁵ in the first round to 3.27 × 10⁵ in round 31, representing an approximate 60 % reduction in sessile cell survival (Fig. S2). Ten individual plaques (numbered from R31p1 to R31p10) were then isolated based on plaque size. Genomic analysis was performed by aligning the sequences of the ten evolved Romulus phage mutants against the wildtype reference genome (Fig. 3). This analysis revealed a total of eighteen point mutations emerged during the directed evolution experiment. The nucleotide profiles of these mutations across all isolates, including the wildtype control, are presented in Table S2.

Fig. 3.

Genome wide distribution of mutations. The circos plot shows the ten phage mutants aligned to Romulus wildtype sequence. The reference genome is 131,314 bp long and displayed in the outermost circle. Strain and mutant specific variations are represented by black ticks, and their details are listed in Table 1 (The plot was made with circlize v0.4.16).

Among the ten isolates, only R31p7 maintained the wildtype sequence, whereas all other mutants exhibited at least one nucleotide substitution, indicating the accumulation of genetic diversity during the evolution protocol. The presence of a wildtype-like clone (R31p7) after 31 evolutionary cycles may reflect several non-exclusive possibilities. It could indicate residual ancestral genotypes persisting under moderate selection pressure, stochastic effects associated with limited sampling, or local microenvironmental variation within biofilms allowing coexistence of both wildtype and mutant variants.

The genomic locations and nature of these mutations are detailed in Table 1. Notably, a single missense mutation (G312D) in gp54, a structural protein, was detected in all isolated clones (except the R31p7), indicating a high-frequency variant that may have been selectively favoured during evolution. Additionally, eight distinct missense mutations were identified in gp58. Some of these mutations were shared among multiple clones, while others were unique to specific isolates.

Table 1.

Summary table of mutations occurred in evolved phage Romulus variants, including SNP position, mutated protein, amino acid substitution and mutants carrying the listed mutation.

SNP position	Gene product	Function	aa mutation	Mutant
39,855	gp 54	structural protein	G312D	R31p1, R31p2, R31p3, R31p4, R31p5, R31p6, R31p8, R31p9, R31p10
46,137	gp 58	structural protein	P67L	R31p6
46,268	gp 58	structural protein	W111G	R31p1, R31p10
46,842	gp 58	structural protein	D302G	R31p1, R31p2, R31p3, R31p4, R31p5, R31p6, R31p8, R31p9, R31p10
47,592	gp 58	structural protein	S552F	R31p3
47,597	gp 58	structural protein	H554Y	R31p8
47,604	gp 58	structural protein	A556E	R31p4, R31p9
47,654	gp 58	structural protein	N573D	R31p2
47,666	gp 58	structural protein	D577 N	R31p1, R31p2, R31p3, R31p4, R31p5, R31p6, R31p8, R31p9, R31p10
73,123	gp91	DNA Polymerase	SM in D codon	R31p4
100,028	intergenic region	–		R31p10
100,035	intergenic region	–		R31p10
100,056	intergenic region	–		R31p10
100,057	intergenic region	–		R31p10
100,084	intergenic region	–		R31p10
100,085	intergenic region	–		R31p10
100,234	intergenic region	–		R31p10
130,831	gp 188	unknown function	SM in C codon	R31p1

In addition to these missense mutations, two silent mutations were detected: one at position 73,123 in gp91 (DNA polymerase) and another at position 130,831 in gp188 (a gene of unknown function). An additional nine mutations were identified in an intergenic region, exclusively in the R31p10 isolate.

Although mutations in non-coding DNA and silent mutations have the potential to influence phage phenotypes, either by modifying codon usage and protein synthesis [53] or gene regulation [54], the missense mutations within gp54 and gp58 were the primary interest of this research. The high frequency of the G312D mutation within gp54 suggests that it conferred an evolutionary advantage to the phage. Moreover, the multiple mutations observed in gp58 indicate stringent selective pressure on this protein. Given that gp54 and gp58 are baseplate-associated proteins (see Section 3.3), the evidence supports to the hypothesis that these proteins play an important role in phage-host interaction, possibly in host recognition as well as adsorption efficiency.

3.3. Functional annotation and structural prediction of mutated proteins

To understand how the observed mutations might alter phage behaviour, we functionally annotated the affected proteins and predicted their structural features. We first focused on gp54, which carried the high-frequency G312D mutation.

Based on bioinformatic analysis, gp54 matched to nine highly significant sequences (E-value ≈ 0), indicating a likely biological functional closeness (Table S3). Specifically, four matches out of the nine matches were structural proteins of the staphylococcal phages ΦRNIID, qdsa001, SA11, LJLAME001 with a percentage identity greater than 99 %. The other five matches of gp54 include hypothetical proteins of the staphylococcal phages stAP1, vB_Sau-RP15, SAPYZU_15, vB_Sau-SPJ2 and stAP1 with (percentage identities between 99 % and 99.56 %). Most of the staphylococcal phages from the annotated proteins, including ΦRNIID, qdsa001, vB_Sau-RP15 and SAPYZU_15, are classified within “unclassified Silviavirus”. This means that they share features with known Silviavirus phages but lack definitive species classification [55].

The gp54 HMMER analysis reported a carbohydrate binding domain (CBD) (residues 267–406) matching sequences from Bacillus phage Taffo16 (E-value of 1.1e-42) and Lactobacillus phage Dionysus (E-value of 7.9e-38) (Table S4). While CBDs are typically associated with receptor binding proteins (RBPs) of phage baseplates, they are also found in other phage parts of unrelated Lactobacillus and Bacillus phages. For example, Lactobacillus phages harbour CBDs in their distal tail (Dit) proteins for host binding [56]. In addition, the putative tail fiber of Bacillus phage SPP1 contains tail spike binding domains similar to those of Salmonella phage P22 [57]. These results suggest that CBDs may be present in other baseplate structures despite the RBP itself.

The HHPRED analysis was performed for gp54, and the top hits were proteins associated with the phage baseplate (Table S5), including baseplate wedge protein (gp23 of 7YFZ_Mpam3 and gp17 of 7KH1_B2) with high probability (99.29 % and 99.19 %, respectively) and low E-values (4.5e-11 and 2.3e-10, respectively). These results suggest a very close evolutionary relationship with known phage baseplate components.

The AlphaFold 3 prediction for gp54 revealed a complex structural organisation with a unique arrangement of distinct domains (Fig. S3); however, the prediction's reliability was limited by a low predicted template modelling (pTM) with a score of 0.33. Additionally, attempts to model multimeric assemblies resulted in similarly low confidence scores, suggesting limitations in the reliability of both monomeric and multimeric predictions. The N-terminus starts with a packed triple helical bundle, followed by some additional short α-helices. Following the N-terminal domain is a region of extended loops without a defined secondary structure. The following region has some features consistent with a CBD (residues 267–406) as predicted by HHMER. This domain displays a β-sheet architecture, with several α-helices exposed on the outer surface. Notably, the β-sheets fold inwards to form a distinctive bowl-like structure which suggests a potential role for the CBD in encapsulating the carbohydrate ligands during binding [58].

HMMER identified matches from unrelated Bacillus phage Taffo16 and Lactobacillus phage Dionysus, and the presence of an additional short α-helix in the predicted CBD of gp54 has not been previously reported.

We next analyzed gp58, which accumulated multiple independent mutations across clones. The best matches for gp58 were either capsid and scaffold proteins or structural proteins of staphylococcal phages StAP1 (x2), SAP6, MR003 and staphylococcal phages PM56, SAC, PM93, SA11, SSP49 respectively (Table S6). These nine BLASTp hits showed 99–100 % sequence identity and near-zero E-values, suggesting functional similarity with the query. For example, phages StAP1 and Romulus both belong to the subfamily Twortvirinae (genus Silviavirus), suggesting also an evolutionary relationship between the two proteins [59]. In addition, the best hits of the closely related phages PM56 and PM93, derived from phages Remus and Romulus, respectively, suggest functional similarity with the predicted protein gp58 [60]. Phage SA11 is also close to phage Romulus and Remus, sharing 82.1 % and 84.6 % nucleotide identity, respectively [52]. This suggests shared biological functions between these structural proteins due to the close genetic relationship. In short, the retrieved hits are potentially strong candidates for functional or evolutionary relatedness to gp58.

HMMER analysis results of the gp58 sequence identified an N-terminal baseplate upper protein (BppU_N) domain and a C-terminal CBD. The identified BppU_N domain (residues 6–155) shared similarities with 14 staphylococcal phages such as 37, LH1, Φsa2wa_st80, Φ7401PVL, among others (Table S7). The identified CBD (residues 208–345) shared homology with seven staphylococcal phages (Table S8). Nevertheless, the E-values associated with these BppU_N domain matches were significantly higher compared to the CBD domain predictions, suggesting a lower level of confidence in the BppU_N domain homology.

Analysis of the HHPRED results for gp58 revealed strong candidates for its role in phage adsorption (Table S9). The top hit, 5M9F_C, aligned with the C-terminal domain (residues 401–637) of gp144 in staphylococcal myophage phage K [61]. Notably, gp144 is a well-characterised RBP located within the phage tail baseplate and is responsible for adsorption to S. aureus via GlcNAc binding in the WTA. Shared functional annotations (“receptor-binding protein” and “phage tail tip”) strongly suggest that gp58 plays a similar role in phage adsorption.

Receptor-binding proteins are typically assembled as homotrimers [62,63]. Mass photometry of purified gp58 revealed a single peak at 236 kDa (Fig. 4), consistent with the theoretical molecular weight of a trimer, given that each gp58 monomer is approximately 74 kDa. These results indicate that gp58 adopts a trimeric conformation. Based on this observation, a trimeric structural model of gp58 was successfully predicted using AlphaFold3 (Fig. 5). The N-terminus starts with an α-helix interwoven with a β-sheet with some surface-exposed loops and a ‘hinge’ region between them. This N-terminal region covers the BppU_N domain predicted by HMMER. The central part has the shape of a β-propeller and covers the region of the predicted CBD. The propeller folds inwards, creating a cavity in the middle where gp58 is likely to bind its target. A small α-helix and β-sheet act as a bridge, connecting the central propeller to the C-terminus of the protein. The C-terminus consists of two very conserved structures, each consisting of an antiparallel β-sheet and some loops exposed to the outside.

Fig. 4.

Mass photometry measurement of purified gp58 protein from Romulus phage (monomer mass: 74 kDa).

Fig. 5.

Three-dimensional structure of the gp58 in its homotrimeric structure, with a side view (on the left) and bottom view (on the right). The CBD (in green) and BppU (in blue) regions are shown; in red, the mutated amino acids are highlighted. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Finally, phageDPO analysis predicted with a probability of 92 % a depolymerase function of gp58. To investigate this hypothesis, a halo assay was carried out, but the analysis did not allow to confirm an enzymatic activity of the protein. No signs of substrate degradation were observed around the E. coli colonies expressing gp58, likely due to the absence of protein secretion, which might be hindered by the protein size (Fig. S4).

In conclusion, the combination of HMMER and HHPRED analyses provided strong evidence for the involvement of gp54 and gp58 in phage-host interactions. Gp58 plays a role in initial attachment to the host cell wall through its CBD, while gp54 may be a baseplate component involved in baseplate assembly and interaction with phage attachment structures. However, further structural analysis, in particular the arrangement and interactions of the identified domains (CBD, and BppU_N), is required to clarify the molecular mechanisms by which gp54 and gp58 contribute to phage-host interaction.

3.4. Host range and efficiency of plating of mutant phages

To determine whether the evolved mutations broadened or altered host specificity, we evaluated the host range and the EOP of the mutants on a panel of S. aureus biofilm-producer clinical isolates. Specifically, phage Romulus and its evolved mutants were tested against a panel of 21 S. aureus strains (Fig. 6). The analysis revealed that the wildtype Romulus and its mutants successfully infected nine out of the 21 tested S. aureus strains. Notably, all Belgian strains, except MSSA V 190809–88, and three out of five Swiss strains (MRSA AOC19, MRSA AOC42, MRSA AOMu81) exhibited increased susceptibility to the evolved mutants compared to the wildtype Romulus. Among the German strains, only MRSA 9 showed increased sensitivity to the mutants. As expected, phage R31p7, which lacked genomic mutations, displayed an infection pattern similar to that of wildtype phage Romulus. Additionally, the spot test revealed unique interactions between the phages and strain AOMu05 (Switzerland). While the wildtype phage Romulus and the clone R31p7 were not active on this strain, the mutants produced a lysis of the bacterial lawn at the highest tested titer, even though no single plaques were countable.

Fig. 6.

Efficiency of plating (EOP) of phage Romulus, both wildtype and mutants, against a panel of 21 S. aureus strains. A higher EOP value, indicated by a darker colour, represents a mutant phage that infects the target strain more efficiently than the wildtype Romulus; n = 3 independent biological replicates. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Finally, to investigate whether strain sensitivity correlates with genetic closeness, we constructed a phylogenetic tree of the tested S. aureus strains based on their intergenomic distances. The strains grouped into three distinct clusters, highlighted in red, blue, and yellow (Fig. S5). Overall, no clear correlation between phage sensitivity and genetic proximity was observed. However, none of the strains in the yellow cluster were susceptible to the phages, except for strain AOMu05, which exhibited partial sensitivity to the evolved mutants in the spot test.

3.5. Lytic activity of mutated phages against both planktonic and sessile cells

To assess whether genomic mutations translated into enhanced killing ability, we compared the lytic activity of all evolved phages against planktonic and biofilm-associated S. aureus cells. To select mutant phages with higher lytic activity compared to the wildtype phage, a preliminary descriptive screening of evolved phages, clones from R31p1 to R31p10 and their mixture (R31), was conducted against S. aureus planktonic cells at 3, 6, and 24 h post-incubation (Fig. 7). The results displayed extensive phage lytic activity variation among mutant phages. Eradication of bacterial culture or a strong reduction in CFU/ml compared to the wildtype phage treatment was observed with the incubation of phages R31, R31p1, R31p2, R31p4, R31p5, R31p6 and R31p8, particularly 6- and 24-h post-treatment. In most of the cases, however, regrowth of bacteria was observed, suggesting the potential development of tolerance or resistance mechanisms, with final CFU/ml values comparable to the untreated control (10⁸–10⁹ CFU/ml). Notably, three treatment conditions consistently resulted in a strong bactericidal effect across all six replicates. Specifically, phage R31 and R31p2 achieved complete eradication of S. aureus cultures, while R31p5 induced at least a 3-log₁₀ reduction in bacterial counts at both 6- and 24-h post-infection. These findings suggest that one or more stages of the infection process, such as host recognition, receptor binding, genome injection, or phage assembly, may differ among these mutants, conferring a selective advantage and enhanced ability to elude bacterial defence mechanisms.

Fig. 7.

CFU/ml of S. aureus PS47 cells treated with phage Romulus wt and mutants (MOI = 0.1) for 3, 6 and 24 h. In brown, the untreated control (GC); in red, the wt phage; in ochre, the three phages with a bactericidal activity in all replicates (R31, R31p2 and R31p2). The other phages are indicated in dark and light blue for an easier reading. Phage R31p7 was excluded since it matched with the wt phage; n = 6 independent biological replicates. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Based on the promising initial findings, three phages, R31, R31p2 and R31p5, were selected as candidates for further bactericidal assays targeting staphylococcal biofilm. Their activity was assessed against a pre-formed S. aureus PS47 biofilm at three titers (ranging from 10⁷ to 10⁹ PFU/ml) for 24 h (Fig. 8). When tested at 10⁹ PFU/ml, the wildtype phage determined a 1.5-log₁₀ reduction of viable cells in biofilm-associated bacteria over the original bacterial loading (t₀). However, no significant reduction in CFU/ml was achieved with titers of 10⁷ and 10⁸ PFU/ml. In contrast, mutant phages displayed a higher efficacy in targeting sessile cells, determining a CFU/ml reduction ranging between 2- and 4-log₁₀, when tested at 10⁸ and 10⁹ PFU/ml. Nevertheless, the mutant phages at the lowest concentration tested (10⁷ PFU/ml) had a similar activity to the wildtype phage, suggesting that a minimal phage titer may be required to achieve effective biofilm degradation, as bacterial reduction was only observed above a certain threshold of phage load. These results support the hypothesis that the application of directed evolution protocols resulted in beneficial mutation within the phage genome, determining an improved lytic efficiency against both planktonic and sessile cultures of S. aureus.

Fig. 8.

CFU/ml of S. aureus PS47 cells collected from biofilms formed of porous glass beads and treated for 24 h with the wildtype phage, the R31, the R31p2 and the R31p5 (10⁷, 10⁸ and 10⁹ PFU/ml). One-way ANOVA Dunnett's multiple comparisons test, ns p-value ≥0.05, ∗p < 0.05, ∗∗∗p < 0.001; df = 19; n = 5 independent biological replicates.

3.6. Phage adsorption

Since all missense mutations emerged through the evolution protocol involved baseplate proteins, it was hypothesised that phage mutants differed from wildtype phage Romulus in their adsorption to the bacterial host. Phage adsorption, which is critical for phage infectivity and host specificity, relies on viral RBPs located on the phage baseplate. RBPs recognize and bind specific receptors on the bacterial cell wall, enabling phage-host interaction. By comparing the behaviour of the selected mutants during the adsorption phase with the wildtype phage Romulus, it was possible to determine how the occurred mutations affect this initial attachment step. Phage adsorption results showed that 97 % of the wildtype Romulus particles were irreversibly adsorbed within 20 min, with an adsorption rate constant of 2.5 × 10⁻¹⁰ ml/min (Fig. 9). Like the wildtype, R31p2 achieved a slightly more rapid initial adsorption (1.0 × 10⁻⁹ ml/min) but adsorbed slightly less at its maximum (20 min), with 92 % of the phage particles bound. Conversely, R31p5 reached its maximal absorption to bacterial cells more rapidly than the wildtype, with only 59 % of the particles irreversibly bound within 15 min. The adsorption rate constant was 6.0 × 10⁻¹⁰ ml/min. R31p5 showed a faster adsorption than the wildtype, starting its burst 20 min after infection, and leading to a 77 % increment of free phages with respect to the 15-min post-infection timepoint. Conversely, phage R31p2 and Romulus burst occurred in more than 25 min, since the increment of free phages was still not clearly detectable in the timeframe adopted in the assay.

Fig. 9.

Adsorption kinetics of phage Romulus wildtype and mutants R31p2 and R31p5 to the host S. aureus PS47 strain, infected with MOI of 0.001. Percentage of free phages (y axis) is indicated at 15, 20 and 25 min after infection (x axis). One-way ANOVA Dunnett's multiple comparisons test ns = not statistically significant; ∗p < 0.05; df = 6; n = 3 independent biological replicates.

According to our observations, missense mutations in baseplate proteins altered the adsorption efficiency and timing. While phage R31p2 exhibited adsorption kinetics similar to the wildtype Romulus, R31p5 showed a higher rate constant of adsorption and an earlier onset of the burst phase, albeit with a reduced maximum adsorption efficiency. These findings highlight the impact of baseplate mutations on the early phases of phage infectivity.

To find out whether the phage adsorption kinetics were also different when infecting biofilm, a further adsorption assay was performed with mature S. aureus biofilms (Fig. S6). In general, the adsorption kinetics for the wildtype Romulus and mutants R31p2 and R31p5 were similar to each other, and statistically significant differences in the rates of adsorption or in the percentage of free phage particles were not detected within 60 min of incubation. These findings demonstrate that, while baseplate protein mutation influences the adsorption performance under planktonic conditions, factors other than initial adhesion may be accountable for enhanced activity of some mutants against biofilm-adherent bacteria.

To further assess potential differences in infection dynamics, one-step growth curve assays were performed on the evolved mutants and the wildtype under biofilm conditions (Fig. S7). The resulting curves displayed comparable profiles, with no significant differences observed in key parameters such as latent period, burst size, or overall timing of the replication cycle.

4. Discussion

The challenge in biofilm-associated S. aureus infection treatment, particularly those involving multidrug-resistant strains, underlines the priority for new therapy strategies. Responding to this clinical demand, our study adopted a novel directed evolution strategy to improve the efficacy of bacteriophages in targeting mature S. aureus biofilms.

Phage-directed evolution protocols have been used to improve phage characteristics related to their host range, infectivity, or resistance to bacterial defences. To date, most of the protocols have been directed at optimizing phage performance against planktonic bacterial cultures [14,19,20]. This means that the challenges provided specifically by biofilms remain only lightly explored. The extracellular polymeric matrix of biofilms provides a complex environment that protects bacteria against phage penetration and antibiotic treatments, serving as a persistent cause of chronic infections [64]. Although phages are already described to be active against biofilm-embedded bacteria of S. aureus, degrading matrix and targeting persister cells [65], only a modest in vitro reduction in terms of bacterial load is observed when they are tested alone [66]. To the best of our knowledge, only a few studies applying a directed evolution protocol with an emphasis on enhancing phage activity against biofilms, particularly versus biofilm-embedded P. aeruginosa, are reported [67].

To specifically target S. aureus biofilms, we developed a directed evolution protocol involving the coincubation of the staphylococcal phage Romulus with mature biofilm. The phage population evolved over 31 sequential rounds, whereby, in each round, phages recovered from the previous cycle were challenged against a fresh biofilm. At the completion of the evolution protocol, phage variants showing an enhanced biofilm-targeting profile were isolated and characterised.

Whole-genome sequencing of the evolved mutants revealed missense mutations within gp54 and gp58. The prediction of their structural domains and the identification of similar phage proteins allowed us to identify them as baseplate proteins involved in phage attachment to the bacterial surface. For instance, gp54 showed high similarity to gp23 of the myophage Pam3, which carries two docking sites for tail fiber attachment located at the periphery of its minimal baseplate [68]. Although Pam3 infects a different host (cyanobacteria), the presence of a similar wedge protein with docking sites further supported the putative role of gp54 in baseplate assembly and interaction with phage attachment structures. Moreover, it displayed similarity to the wedge protein (gp17) within the baseplate of Vibrio myophage XM1, which forms the outer layer structure and binds to tail spikes via its flexible parts [69]. Taken together, the evidence suggested that gp54 is probably involved in baseplate assembly, possibly interacting with tail attachment structures.

A CBD in gp54 was also identified and it is consistent with established knowledge of staphylococcal phages. The β-sheet architecture and the presence of a cavity that could accommodate GlcNAc ligands reflect the potential role of the CBD within staphylococcal RBPs [70]. The closest related CBDs were identified in Bacillus phage Taffo16 and Lactobacillus phage Dionysus. A previous study suggested the presence of CBDs in the distal tail (Dit) proteins of Lactobacillus phages to facilitate the binding. The Dit protein of Lactobacillus phage J-1 consists of 18 β-strands and extended loops, showing some similarities to the predicted CBD structure of gp54 [71]. Interestingly, Bacillus phage SPP1 has a putative tail fiber protein (gp21) with a right-handed β-helix fold sharing structural similarity with the tail spike domain of Salmonella phage P22 [57].

The analysis of gp58 revealed the presence of a BppU_N domain, which features a β-sandwich fold, critical for its function in bacteriophages [72]. Studies of the staphylococcal phage Φ11, have shown that a significant portion of the BppU domain within the N-terminus of gp54 of phage Φ11 is located within the baseplate [70]. This is consistent with the findings for the lactococcal phages Tuc2009 and TP901-1, which previously showed similarity to gp54 of Φ11 [73]. In these phages, the N-terminal region of the RBP forms a bowl-like structure interacting with the BppU loop, which allows a stable association between the RBP and the baseplate core. Interestingly, the N-terminal region of gp62 of S. aureus phage 80α also corresponds to the N-terminal part of gp54 in Φ11 (crystal structure 97 % identical), further suggesting a potential functional conservation of this domain in related phages [74]. These findings highlighted the potential role of the BppU_N domain in baseplate assembly and stability in different phages.

Gp58 also carries a CBD, crucial for the initial recognition and reversible attachment to specific carbohydrate receptors on the bacterial host [73]. Staphylococcal phages are known to display multiple CBDs within their RBPs, each with a specific function during the infection process. For example, the putative receptor binding site within Φ11 allows for the positioning to encapsulate the GlcNac of the cell WTA [58]. This specific interaction allows the efficient attachment and infection of the host. Similarly, multiple CBDs in the putative RBPs encoded by orf103 and orf105 of the Twort-like phage SA012 interact with the WTA backbone of the host cell [75]. Moreover, multiple RBPs allow the phage to expand its potential host range by targeting multiple carbohydrate receptors present on different bacterial strains [76].

In support of the predicted CBD structure, comparisons with RBPs from closely related phages reveal high similarity in both structure and function. For example, the trimeric RBP of Φ11 (gp45) contains a centrally located five-bladed β-propeller domain (residues 142–439) with a cavity that binds the GlcNAc moieties of WTA on the S. aureus cell surface [58]. This region is consistent with the predicted CBD, and both structures share β-strands folded in the center of the structure. Further, phage 80α shares a high similarity (97 %) with the closely related phage Φ11. Its RBP (gp61) also has a centrally located five-bladed β-propeller platform domain (residues 143–442) within its trimeric structure [74], previously demonstrated to bind GlcNAc moieties of WTA. In addition, the sequence homology of the described RBPs of phage Φ11 (gp45) and Twort-like phage Φ812 (gp123) with phage K indicates a similar structure for its RBP (gp144), known to bind with high affinity to GlcNAc residues of the WTA [70,77].

Interestingly, both mutated proteins contained CBDs, which are important in the recognition of bacterial surface carbohydrates [78]. Biofilm matrices are composed of a high amount of polysaccharides, and mutations within these CBDs could alter the phage affinity for biofilm-specific receptors, hypothetically enhancing the ability of the phages to bind and penetrate biofilms [79]. This suggests that the mutations might enhance phage interactions with biofilm components, such as polysaccharides in the extracellular matrix, enabling deeper biofilm penetration and more effective bacterial cell lysis [80]. While these annotations and structural similarities strongly support a role in adsorption and baseplate assembly, the specific functional effects of the observed mutations have not been experimentally confirmed and therefore remain hypothetical.

The predominance of mutations in adsorption-related proteins may reflect the complex role of initial attachment and penetration as a key passage through the physical barriers imposed by the biofilm matrix. However, this does not imply that the protocol exclusively selects for adsorption-related changes. It is also possible that under different biofilm conditions, or after longer evolution, various beneficial mutations increasing replication efficiency, or the ability to take advantage of the microenvironments surrounding a biofilm may occur. These findings highlight the complexity of biofilm-directed phage evolution and might suggest that the biofilm itself, despite the experimental design, imposes a selective pressure favouring enhanced binding to biofilm-specific structures.

The impact of the occurred mutations on the phage behaviour was assessed further. After 31 rounds of directed evolution, the mixture of evolved phages, R31, and two single phages mutants, R31p2, and R31p5, consistently demonstrated bactericidal activity against planktonic cells over 3, 6, and 24 h. These mutants were later tested against biofilms, demonstrating higher lytic activity compared to wildtype phage after 24 h, with a CFU/ml reduction up to 4-log₁₀. The enhanced lytic activity of the evolved phages R31, R31p2, and R31p5 against S. aureus sessile cells underscores the efficacy of our directed evolution approach to target staphylococcal biofilms. This not only appears as a solution with potential to weaken biofilm protection but also brings in possibilities to create more efficient phage-based treatments.

The phage evolution training was focused on increasing phage activity against S. aureus PS47 biofilms and selected for mutations that improved the ability of the phage to target this specific trait. A potential drawback previously observed in directed evolution approaches is a narrowing host range [17]. Phages could become more efficient on the host strain but perform worse on others, as a result of a trade-off between activity and host range [81]. Surprisingly, all phage Romulus mutants displayed a host range comparable to the wildtype. Even more interesting, the efficiency of plating of the phage mutants generally increased, suggesting that the mutants were overall more efficient than wildtype phage Romulus in a subset of the tested strains. This suggests that the occurred mutations appear to be more critical for efficient infection rather than determining the overall host range. This finding is supported by the study of Burrowes et al. [16], who found that spontaneous mutations alone were insufficient to alter the host range. Therefore, it is hypothesised that mutations may have increased the ability of the phage to adhere to primary receptors on the bacterial surface, improving the initial attachment critical for successful infection. Similar observations have been made in Bacillus phage mutants, where point mutations in tail proteins restored infectivity without affecting host range [82]. Finally, mutant phages displayed a slight activity even towards the S. aureus AOMu05 strain, which was not sensitive to the wildtype Romulus (and the clone R31p7). Even the mutants failed to produce single plaques, suggesting insufficient phage progeny production within the host, they demonstrated some level of bacterial binding, since lysis of the bacterial lawn was detected. This phenomenon may be attributed to either abortive infection or lysis from without [36]. Abortive infection results in incomplete phage replication and the production of non-infectious particles [83]. In contrast, lysis from without occurs when an excess of phages binds to and damages the bacterial surface, leading to cell lysis without effective phage propagation [84].

To further explore the functional consequences of these mutations, we assessed the adsorption behaviour of both wildtype and mutant phages under planktonic conditions. This revealed a paradoxical trend that offers deeper insight into the specialisation of the evolved phages. Although R31p2 did not differ significantly from the wildtype phage, the mutant R31p5 exhibited a lower overall adsorption rate under planktonic conditions (59 % for R31p5 versus 97 % for the wildtype). However, R31p5 also showed faster adsorption kinetics, suggesting that although it is less efficient in total adsorption, its initial binding is more rapid. This may confer an advantage in biofilm environments, where phage attachment must occur quickly at transiently exposed binding sites. This apparent contradiction may be explained if the mutants became specialised for the biofilm conditions. Most likely, the mutation allowed phages to recognize and bind biofilm-specific molecules that were absent in planktonic cells; this may account for their reduced adsorption in planktonic environments, yet increased activity against biofilms are achieved. These findings underline the possible trade-offs involved in optimizing phages for specific conditions, such as biofilm eradication, at the cost of reduced performance in other environments.

Taken together, these results demonstrate that biofilm-adapted mutations, particularly in structural proteins involved in adsorption, can lead to significant phenotypic advantages under biofilm conditions, even when such adaptations come at a cost to planktonic performance. This work illustrates that directed evolution protocols can indeed be used to develop phages with superior biofilm-targeting properties. The evolved mutants, R31, R31p2, and R31p5, showed better staphylococcal biofilm eradication, although their adsorption rate in the planktonic assays was lower. Therefore, the development of phage through evolution can be targeted specifically for improved antibiofilm activity an important need for the treatment of biofilm-associated infections resistant to conventional therapies.

A limitation of this work is that the directed evolution was conducted as a single independent evolutionary line. While this design enabled us to establish a proof-of-concept for phage adaptation under biofilm-associated conditions and to characterise the resulting genomic and phenotypic changes in detail, it does not allow assessment of the reproducibility or generality of the observed adaptive events. Future studies involving multiple parallel evolutionary lines will be required to differentiate stochastic effects from consistent adaptive processes and to evaluate the robustness of evolutionary trajectories in biofilm environments.

In addition, it would be valuable to examine whether the adaptive mutations identified here are maintained or lost when selective pressure from biofilm-associated hosts is removed, for instance during propagation under planktonic conditions.

This could allow, from an applied research point of view, the development of more effective phage therapies against biofilm-associated infections and, consequently, improved patient outcomes when antibiotic therapy has failed. Furthermore, from a fundamental research perspective, the study will contribute to a deeper understanding of the interactions between phages and sessile cells and of the molecular adaptations that increase the efficacy of phages in the context of biofilms.

CRediT authorship contribution statement

Claudia Campobasso: Writing – original draft, Visualization, Supervision, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Pauline Henderix: Investigation, Formal analysis. Ekaterina Jalomo-Khayrova: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation. Sara Bolognini: Investigation. Gert Bange: Writing – review & editing, Supervision, Resources. Rob Lavigne: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Funding acquisition. Arianna Tavanti: Writing – review & editing, Supervision, Resources. Jeroen Wagemans: Writing – review & editing, Supervision, Project administration, Methodology. Mariagrazia Di Luca: Writing – review & editing, Supervision, Resources, Project administration, Conceptualization.

Funding

This research was funded by i) Hub multidisciplinare e interregionale di ricerca e sperimentazione clinica per il contrasto alle pandemie e all'antibioticoresistenza (PAN-HUB)” funded by the Italian Ministry of Health (POS 2014–2020, project ID: T4-AN-07, CUP: I53C22001300001); ii) PNRRTHE e Tuscany Health Ecosystem; Spoke 7—Innovating Translational Medicine-Sub-project 5—Innovative models for management of infections caused by antibiotic-resistant bacteria (Project code: ECS00000017; CUP I53C22000780001); iii) KU Leuven, Internal Funds KU Leuven, Interdisciplinary Networks (ID-N) grant (IDN/20/024); iv) the Deutsche Forschungsgemeinschaft (DFG; Projektnummer 464366151) in the framework of the priority program “SPP 2330 – New Concepts in Prokaryotic Virus-host Interactions” (to G.B.).

Declaration of interest statement

MDL is a co-founder of start-up company named Fagoterapia LAB S. r.l. All other authors declare no competing interests.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

Data will be made available on request.