Phage–antibiotic synergy restores β-lactam efficacy in MDR Klebsiella quasipneumoniae biofilms and suppresses resistance

Abstract

Background

Biofilms formed by multidrug-resistant (MDR) Klebsiella spp. present a significant clinical challenge due to elevated antibiotic tolerance. Bacteriophages (phages) represent a promising alternative, particularly in combination with antibiotics, where phage–antibiotic synergy (PAS) can increase antibiofilm activity. Evaluating treatment efficacy in these complex structures requires real-time, noninvasive viability analysis.

Methods

To address this, we used light-sheet fluorescence microscopy (LSFM), a high-resolution, minimally invasive approach, for dynamic tracking of PAS in intact biofilms. To our knowledge, this is the first in vitro application of LSFM for investigating PAS. We studied the combined activity of a virulent phage (vB_KpUKJ_2) and ceftazidime (CAZ) against an extended-spectrum β-lactamase-producing Klebsiella quasipneumoniae.

Results

In planktonic cultures, PAS was strongly affected in a dose-dependent manner. In mature biofilms, LSFM imaging revealed that high-dose phages (10⁸ PFU/mL) combined with CAZ at a 0.25 × minimum inhibitory concentration (MIC) induced a rapid and sustained reduction in viability over 24 h. This regimen significantly outperformed mono-treatments (p < 0.01), demonstrating that phage coadministration can reduce the required antibiotic dose. Mechanistically, treatment resulted in phage-mediated degradation of α- and β-polysaccharides within the extracellular polymeric substance (EPS). Crucially, while phage mono-treatment led to the emergence of resistant mutants, the combination treatment fully suppressed resistance. Whole-genome sequencing revealed mutations in genes such as fhuA, purA, and rpoC, suggesting diverse resistance mechanisms linked to fitness trade-offs such as impaired biofilm formation.

Conclusions

Our findings highlight a precision-guided strategy with translational potential for device-associated infections, providing a mechanistic and methodological foundation for optimizing PAS-based therapies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12929-026-01218-1.

Introduction

The rapid rise of antimicrobial resistance (AMR) represents one of the most urgent threats to global health, undermining antibiotic efficacy, complicating infection management, and increasing healthcare costs worldwide [1–3]. Among Gram-negative pathogens, Klebsiella spp. are of particular concern because of their prevalence in both hospital- and community-acquired infections [4]. These opportunistic bacteria are implicated in a broad spectrum of diseases, including pneumonia, sepsis, urinary tract infections, and device-associated infections [5].

Originally classified as Klebsiella pneumoniae, the species Klebsiella quasipneumoniae was redefined following comprehensive phylogenomic and whole-genome sequence analyses that revealed significant genetic divergence within the K. pneumoniae complex [6, 7]. This reclassification distinguished K. quasipneumoniae from K. pneumoniae sensu stricto, with an average nucleotide identity (ANI) of approximately 93–94%, confirming its status as a separate species. While some K. quasipneumoniae lineages remain susceptible to standard antibiotics, others harbor resistance determinants such as extended-spectrum β-lactamase (ESBL) and/or carbapenemases, which severely limits therapeutic options and contributes to increased morbidity and mortality [8, 9]. Although considered primarily environmental, K. quasipneumoniae is now increasingly recognized as an emerging nosocomial pathogen capable of causing invasive infections and exhibiting multidrug-resistant (MDR) phenotypes [10]. Notably, the widely used quality-control strain ATCC 700603, originally labeled K. pneumoniae, has since been reclassified as K. quasipneumoniae subsp. similipneumoniae (hereafter referred to simply as K. quasipneumoniae) [6]. This strain is a well-characterized MDR isolate that produces the ESBL SHV-18, which confers resistance to penicillins and most cephalosporins while remaining susceptible to carbapenems.

Compounding these challenges, Klebsiella species readily form biofilms, which are structured bacterial communities encased in an extracellular polymeric substance (EPS) that increases persistence and tolerance to antibiotics and host defenses [11, 12]. Biofilms on indwelling medical devices such as catheters and ventilators are a major cause of chronic and relapsing infections that are particularly difficult to eradicate [13, 14]. EPS, which is composed primarily of polysaccharides, functions as a physicochemical diffusion barrier that limits the penetration of antimicrobial agents and facilitates bacterial survival under stress conditions. Moreover, bacteria embedded within biofilms, particularly those located in deeper layers, are metabolically dormant and therefore poorly accessible to antibiotics [11, 15]. This reduced metabolic activity, combined with restricted drug penetration, results in pronounced antibiotic tolerance, rendering the minimum inhibitory concentration (MIC) determined for planktonic cells ineffective against biofilm-associated bacteria [16]. This dual burden of antimicrobial resistance and biofilm-associated tolerance underscores the urgent need for innovative therapeutic strategies to combat MDR Klebsiella infections.

Bacteriophages (phages), viruses that selectively infect and lyse bacteria, have emerged as promising antimicrobial agents, particularly when used in combination with antibiotics [17, 18]. In such combinations, a phenomenon known as phage–antibiotic synergy (PAS) can occur, in which the concurrent action of phages and antibiotics enhances bacterial clearance and suppresses the evolution of resistance across multiple pathogens [19, 20]. Interestingly, emerging evidence suggests that phages may help restore the efficacy of antibiotics rendered clinically ineffective by resistance, offering a route to repurpose existing drugs for renewed therapeutic use [21, 22]. This precision-guided strategy is especially attractive in MDR contexts, where few effective therapies remain. PAS has been studied in both planktonic and biofilm systems [23, 24]; however, studies capturing the real-time, spatially resolved dynamics and kinetics that provide mechanistic hints of these interactions in intact biofilms remain scarce and technically challenging. Such studies require noninvasive imaging approaches capable of resolving the spatial organization and temporal evolution of biofilm responses under antimicrobial pressure.

Among the imaging methods used in microbiology, confocal laser scanning microscopy (CLSM) is frequently employed because of its ability to capture detailed 3D structures [25]. However, its application in longitudinal studies is limited by limitations such as phototoxicity, photobleaching, and reduced efficiency for rapid volumetric imaging [26]. These drawbacks are especially restrictive for investigating live biofilms under antimicrobial treatment, where prolonged and minimally invasive imaging is critical [27]. To overcome these limitations, we used light-sheet fluorescence microscopy (LSFM), which enables high-resolution, real-time imaging over extended durations with minimal photodamage [28–30]. Despite its advantages, LSFM remains underutilized in microbiology and, to our knowledge, has not yet been applied to investigate phage–antibiotic interactions within biofilms. To support reliable viability monitoring during LSFM imaging, we used calcein acetoxymethyl ester (CAM). CAM is a non-fluorescent precursor that is hydrolyzed by intracellular esterases into green-fluorescent calcein, which accumulates in metabolically active cells and exhibits high resistance to photobleaching [31, 32]. While CAM specifically measures enzymatic activity rather than direct viability per se, it has been previously shown to correlate strongly with CFU counts across various biofilm-forming species, serving as an accurate indicator of biofilm vitality [32].

With this imaging platform, we investigated the effects of the PAS between the virulent phage vB_KpUKJ_2 and the β-lactam antibiotic ceftazidime (CAZ) on ESBL-producing K. quasipneumoniae ATCC 700603 under planktonic and biofilm conditions. Phage vB_KpUKJ_2, previously characterized by our group [33], was selected for its rapid replication and potential biofilm-disrupting activity. CAZ was chosen because its mode of action promotes bacterial filamentation, a key factor implicated in facilitating phage replication under PAS conditions [34]. Through the integration of real-time imaging and functional assays, this study establishes a mechanistic and methodological basis for optimizing PAS-based interventions targeting MDR biofilms. These findings reinforce the potential of PAS as a precision-guided therapeutic strategy against MDR Klebsiella infections.

Materials and methods

Bacterial strain, phage and antibiotic

The Klebsiella quasipneumoniae subsp. similipneumoniae strain ATCC 700603 (GenBank accession number CP014696.2) used in this study was originally isolated from the urine of a hospitalized patient and was formerly classified as Klebsiella pneumoniae. This strain is well characterized as an ESBL-producing strain harboring plasmid-encoded (GenBank accession number CP014698.2) resistance genes and is widely used as a reference strain for antimicrobial susceptibility testing [35, 36].

The virulent phage vB_KpUKJ_2, previously isolated from hospital sewage water (GenBank accession number PQ505130.1) [33], was employed for phage experiments.

The third-generation cephalosporin CAZ (Thermo Fisher Scientific Inc., Waltham, US) was used for phage‒antibiotic combination studies. The minimum inhibitory concentration (MIC) of CAZ against ATCC 700603 was determined to be 32 mg/L via the broth microdilution method in accordance with the European Committee on Antimicrobial Susceptibility Testing (EUCAST) guidelines (ISO 20776-2:2021).

Growth kinetics of planktonic bacteria

Growth curve experiments were conducted via 96-well microtiter plates (Greiner Bio-One International GmbH, Frickenhausen, Germany). Overnight bacterial cultures were diluted in Mueller‒Hinton (MH) broth (Merck KGaA, Darmstadt, Germany) to an initial optical density (OD₆₀₀) of 0.08 corresponding to approximately 10⁸ CFU/mL, based on prior calibration for K. quasipneumoniae ATCC 700603. Treatments included CAZ alone (0.25 × to 2 × MIC; MIC = 32 mg/L), phage alone (10⁴–10⁸ PFU/mL), or combinations of both (10⁴, 10⁶, or 10⁸ PFU/mL in combination with each CAZ concentration). Accordingly, the estimated multiplicity of infection (MOI) ranged from 0.0001 to 1.0 across these phage doses. Bacterial growth (OD₆₀₀) was recorded every 15 min for 25 h at 37 °C via an Infinite 200 PRO microplate reader (Tecan Group AG, Männedorf, Switzerland). Before each absorbance measurement, the microplate was subjected to orbital shaking at an amplitude of 5 mm for 480 s. To prevent cross-contamination between wells, particularly under phage-treated conditions, the plates were sealed with sterile, breathable membrane films (Greiner Bio-One GmbH, Frickenhausen, Germany) during incubation. The treatment groups were distributed in a randomized layout across the plate to minimize positional effects.

Biofilm establishment

For biofilm formation, an overnight bacterial culture was adjusted to an OD₆₀₀ of 0.08 and 250 μL, which were applied to TruLive3D dishes (Luxendo GmbH, Heidelberg, Germany) or 200 µL into 96-well plates (glass or plastic bottom depending on the experiment; both Greiner Bio-One, Kremsmünster, Austria) and incubated at 37 °C and 5% CO₂ for 48 h under static and humidified conditions, respectively. To prevent evaporation, the TruLive3D dishes were carefully sealed with parafilm. After 24 h, approximately 150 μL of medium was gently removed from each well and replaced with an equal volume of fresh Luria–Bertani (LB) broth (Carl Roth GmbH, Karsruhe, Germany) to maintain nutrient levels.

Biofilm treatment with antibiotics and phages

After staining (see below) and prior to treatment, the supernatants containing planktonic bacteria were carefully aspirated. CAZ was freshly prepared as a 10 mg/mL stock solution in distilled water and diluted to the required concentrations in MH broth (64–8 mg/L). Phage stock solutions were diluted tenfold to achieve the desired titers (10¹⁰–10⁸ PFU/mL). Mature biofilms were treated with 250 μL or 200 µL per well (according to the previous medium volume used in the respective dish or plate) of antibiotic, phage, their combination, or medium only (controls). The biofilms were treated for varying durations depending on the experiment (see relevant results), and treatment effects were evaluated via LSFM, CLSM, or quantification of colony-forming units (CFU/mL) and plaque-forming units (PFU/mL) (see below).

LSFM experiments

Prior to staining, the supernatants were carefully aspirated, and the biofilms were gently washed twice with 250 µL of sterile phosphate-buffered saline (1 × PBS) (Carl Roth GmbH) in TruLive3D dishes. The biofilms were then stained with the CAM fluorophore (Thermo Fisher Scientific Inc., Waltham, MA, USA) following the protocol described previously [32]. Briefly, 150 µL of the staining solution was added to each well, yielding a final concentration of 100 µM. The samples were incubated for up to 60 min at room temperature in the dark to protect the fluorophore from light exposure. After incubation, the staining solution was carefully removed, and 250 µL of MH broth was added to the control wells to prevent biofilm desiccation during microscopy. In the treatment wells, 250 µL of the corresponding treatment solutions—phage, antibiotic, or their combination at the indicated concentrations—were added (see above).

The LSFM of the biofilms was performed via a TruLive3D Imager (Luxendo GmbH, Heidelberg, Germany) equipped with a 25 × /1.1 NA (numerical aperture) water-immersion objective and a 10 × ocular lens. Images were acquired with a 2048 × 2048-pixel sCMOS camera (Hamamatsu Orca Flash 4.0 V3; 6.5 µm pixel size), yielding an XY field of view of approximately 532 × 532 µm (effective sample pixel size ~ 0.26 µm at × 25). The light-sheet beam diameter was set to 1.6 µm to optimize the axial resolution. Z-stacks were acquired at 1 µm steps starting from the first plane containing bacteria, with one Z-stack per well. The number of Z layers varied according to the sample thickness from 50 to 100. Calcein was excited at 488 nm, and emission was measured at 522 nm (green channel). Biofilm stacks were recorded every 2 h for 24 h. Images were processed and analyzed in Fiji (ImageJ, NIH, Bethesda, MD, USA).

CLSM experiments

After treatment, planktonic cells were removed, and biofilms were washed twice with sterile PBS in glass bottom microtiter plates. The biofilm matrix was stained with calcofluor white (CFW) (Merck KGaA, Darmstadt, Germany) or concanavalin A (Con-A) (Thermo Fisher Scientific), which target β- or α-polysaccharides, respectively [37–39]. CFW was diluted in PBS to a final working concentration of 200 µg/mL, while Con-A was prepared at approximately 180 µg/mL in PBS. Staining (50 µL/well) was performed for 20–25 min at room temperature in the dark, followed by gentle removal of the dye and replacement with LB to maintain hydration during microscopy.

CLSM was performed on an LSM 980 system (Carl Zeiss AG, Oberkochen, Germany) with a 40 × /0.65 air objective. To minimize photobleaching and laser-induced biofilm disruption, CLSM was used only at discrete time points on independent samples to assess matrix polysaccharide degradation. Con-A fluorescence was excited at 488 nm and detected at 522 nm (green channel), whereas CFW was excited at 405 nm and detected at 430 nm (blue channel). Z-stacks were acquired over 200 µm × 200 µm areas at 1 µm intervals, starting from the first plane containing visible bacteria. The pinhole diameter was set to 0.95 Airy units, corresponding to an optical section thickness of approximately 1 µm. Image acquisition and processing were carried out via ZEN Blue software (Carl Zeiss AG).

Quantification of viable bacteria and phages in treated biofilms

The biofilms used for bacterial viability assessment and phage quantification were grown, treated (with CAZ or phages at the respective concentrations) and washed in a similar manner as described for the LSFM. The CFU/mL and PFU/mL concentrations were determined at three time points: directly after treatment, as well as at 4 and 24 h post-treatment. Following treatment, biofilms were washed twice with 250 μL of sterile PBS to remove nonadherent cells, subsequently carefully scraped using a sterile loop, and resuspended in 250 µl of PBS. These steps took a few minutes; thus, the starting point corresponds to approximately 15 min and was set in the graphics, tables and analyses to zero. To determine the number of viable bacteria, 100 µl of the disrupted biofilm suspensions was serially diluted in 10 mM ferrous ammonium sulfate (FAS), a virucidal agent used to inactivate free phages prior to plating [40]. Next, 10 μL aliquots from each dilution were added in triplicate to MH agar plates and incubated overnight at 37 °C. Spots with 3 to approximately 30 colonies were counted, and CFU/mL values were calculated.

For phage count, 100 μL from the rest of the disrupted biofilm suspension was serially diluted in sterile PBS, and phage titers were determined via the double agar overlay assay as previously described [41].

Resistance development under ceftazidime and phage treatment

Phage and antibiotic resistance frequencies, as well as killing efficiencies, were evaluated in both planktonic and biofilm-associated bacterial populations under various treatment conditions. Planktonic cultures at a starting density of 1 × 10⁸ cells/mL and preformed 48-h biofilms were treated with phages at a concentration of 10⁸ PFU/mL, CAZ at 0.25 × , 0.5 × , or 1 × MIC (MIC = 32 mg/L), or a combination of both agents. Untreated controls were included for comparison. Planktonic samples (1 mL per condition) were incubated at 37 °C with shaking at 250 rpm for 3 h, while biofilm samples were incubated statically at the same temperature for the same duration. After the treatment period, the planktonic samples were centrifuged at 8,000 rpm for 10 min at room temperature. The biofilm samples were first washed and resuspended in PBS as described above and then centrifuged under the same conditions. The supernatant was discarded, and the resulting bacterial pellets were resuspended in 1 mL of sterile 10 mM FAS to halt further phage replication from any remaining phages. The samples were gently vortexed to ensure a uniform suspension and incubated at room temperature for approximately 10 min. From each sample, 100 µL of appropriate serial dilutions were plated onto selective and nonselective agar media to quantify the total number of viable cells and resistant subpopulations. The plates were incubated at 37 °C for 24 h (nonselective) or 48 h (selective, due to slower growth under pressure). The total viable counts were determined via nonselective LB agar. To quantify CAZ-resistant mutants, samples were plated on MH agar supplemented with 32 mg/L CAZ. To assess phage-resistant mutants, samples were plated on LB agar containing 10⁸ PFU/mL phage.

The resistance frequency was calculated as the ratio of resistant colonies (selective plates) to total viable bacteria (nonselective plates) under each treatment condition. Killing efficiency was calculated as the ratio of the total viable counts (on nonselective plates) of the treatment conditions to those of the nontreated controls.

Isolation of phage-resistant mutants

Following phage treatment of preformed biofilms, six randomly selected surviving bacterial colonies (hereafter referred to as mutants) were serially passaged on blood agar (Becton Dickinson GmbH, Heidelberg, Germany) across at least five consecutive rounds (incubation time 24 h). This procedure was conducted to evaluate colony morphology (e.g., emergence of small colony variants) and to assess the stability of phenotypic traits over time. One representative colony from the final passage of each plate, representing a clonal isolate of a putative phage-resistant mutant, was selected for subsequent characterization. The following phenotypic and genotypic traits of these mutants were assessed in comparison with those of the ancestral strain ATCC 700603: (i) phage resistance—phenotypically confirmed via a spot test (see below); (ii) the MIC of CAZ—determined via the broth microdilution method (ISO 20776-2:2021); (iii) growth kinetics—evaluated through growth curve assays (see above); (iv) biofilm-forming capacity—quantified via crystal violet staining (see below); (v) virulence—determined via a serum resistance assay (see below); and (vi) whole-genome sequencing—comparative genomic analysis (see below).

Spot test to assess phage resistance

Phage resistance was assessed via spot assays [42], in which 10 µL of serially diluted phage suspensions (10⁶–10⁹ PFU/mL) was applied to a bacterial lawn of putative phage-resistant mutants and monitored for lysis zones to confirm the resistant phenotype.

Assessment of biofilm formation ability

Biofilm mass was assessed via the crystal violet staining method in 96-well microtiter plates [43]. The log-phase bacterial cultures were adjusted to an OD₆₀₀ of 0.08 in LB broth, and 200 µL of the suspension was added to each well and incubated statically at 37 °C for 48 h under 5% CO₂. After incubation, the planktonic cells were removed, and the wells were washed twice with 200 µL of sterile PBS. The plates were air-dried before 200 µL of 95% ethanol was added to fix the biofilm for 15 min, after which they were dried again. Each well was stained with 200 µL of 0.1% crystal violet (prepared in distilled water) for 10–15 min at room temperature. Excess stain was removed, and the wells were rinsed 3–4 times with PBS and dried for 2 h. The bound dye was solubilized with 200 µL of 30% acetic acid for 10–15 min with gentle shaking. The absorbance was measured at 570 nm, and blank wells (broth only) were used for background correction.

Serum bactericidal assay

We used commercially available sterile-filtered human serum derived from pooled AB blood group male plasma (Sigma‒Aldrich, St. Louis, USA). In accordance with the manufacturer’s instructions, the serum is not heat inactivated and retains complement activity. The bactericidal activity of human serum against phage-resistant mutants was assessed via a previously established protocol [44]. The log-phase resistant isolates were resuspended in PBS and mixed with human serum at a 1:3 volume ratio, yielding a final volume of 400 μL containing 75% serum. We included an untreated bacterial suspension in PBS (negative control) to control for baseline viability and complement-specific effects. Heat-inactivated serum (56 °C for 30 min) was used to assess complement-independent effects. Active serum served as a positive control for complement-mediated killing. All the mixtures were incubated at 37 °C for 3 h. At 0, 1, and 3 h, 100 μL samples were collected, serially diluted, and plated on LB agar. The CFU/mL ratio was determined after incubation for 24 h at 37 °C.

Assessment of synergistic interactions

Synergistic interactions were calculated for the time-dependent kinetic assays (planktonic growth curves and LSFM-based biofilm analyses) via the area under the curve (AUC)-based highest single agent (HSA) synergy model [45] (Eq. 1):

$$S_{\text{HSA}}=min\left(\frac{{\text{AUC}}_P}{{\text{AUC}}_0},\frac{{\text{AUC}}_A}{{\text{AUC}}_0}\right)-\frac{{\text{AUC}}_{\mathit{AP}}}{{\text{AUC}}_0},$$ 1

where AUC_P and AUC_A correspond to monotherapies with phages or antibiotics, AUC_AP corresponds to the combination treatment, and AUC₀ corresponds to the untreated control. The function min identifies the most effective mono-treatment, i.e., the treatment producing the lowest normalized AUC value. A positive S_HSA value (> 0) indicates synergy, where the combination inhibits bacterial growth more effectively than either mono-treatment does. A value close to zero suggests an additive or indifferent effect, whereas a negative S_HSA (< 0) denotes antagonism, where the combination is less effective than at least one of the individual treatments. This approach integrates the entire growth trajectory, providing a comprehensive and endpoint-independent assessment of antimicrobial interactions.

For the endpoint CFU/mL data, synergistic effects were evaluated via a mixed-effects model [46]. The log₁₀-transformed CFU reductions were analyzed to quantify the interaction between treatments via the log-reduction interaction coefficient (LRIC) (Eq. 2):

$$\text{LRIC}=\text{log}\left(\text{APR}\right)-\left[log\left(\text{AR}\right)+log\left(\text{PR}\right)\right],$$ 2

where AR and PR represent the reduction in bacterial counts following antibiotic or phage treatment, respectively, and APR represents the reduction following the combined (phage + antibiotic) treatment.

Consistent with a positive $S_{\text{HSA}}$ value in the kinetic analysis, positive LRIC values indicated synergy, and negative LRIC values indicated antagonism. This CFU-based model complements kinetic AUC analysis by providing an endpoint-based assessment of treatment interactions within static biofilm assays.

Computed image analysis

To facilitate efficient processing of large volumetric LSFM datasets and enable robust quantitative analysis, we developed custom macros suitable for Fiji. These tools support semiautomated preprocessing and viability quantification. All macros used in this study are publicly available at https://github.com/Tiktcha/Supporting_macros_LSFM_data/. Biofilm viability analysis was performed via a custom Python script, which was originally adapted from Mountcastle et al. [47] and further modified in our previous work [32]. In the present study, the script was updated to support LSFM-specific file formats (e.g., h5) and accommodate single-channel imaging with the CAM fluorophore. The biofilm surface area coverage (SAC) was calculated for each Z-slice according to Eq. 3:

$$\text{SAC}\left(\%\right)=\left(\frac{A_{\text{biofilm}}}{A_{\text{total}}}\right)\times 100,$$ 3

where A_biofilm represents the biofilm-covered area and A_total represents the total imaged area of the Z-slice. The total area was defined as 2048 × 2048 pixels, corresponding to the dimensions of the acquired micrographs. This analysis was applied consistently across all slices in the 3D image stack.

Whole-genome analysis of the mutants

Genomic DNA was extracted via the ZymoBIOMICS DNA Miniprep Kit (Zymo Research) and quantified via the Qubit dsDNA High Sensitivity Assay Kit (Thermo Fisher Scientific). For size selection, DNA was purified via the use of a 0.45 × volume of AMPure XP beads (Beckman Coulter, Brea, USA) and eluted in 50 μL of EB buffer (Qiagen, Hilden, Germany). Library preparation was carried out with 1 µg of input DNA via the SQK-LSK114 kit (Oxford Nanopore Technologies, Oxford, UK) following the manufacturer's instructions.

Basecalling was performed with Dorado (v7.3.11; https://github.com/nanoporetech/dorado) via the superaccuracy model. The resulting reads were assembled and polished via Flye (v2.9.3) [48] for assembly, followed by polishing via Medaka (v1.11.3; https://github.com/nanoporetech/medaka) and Racon (v1.4.20; https://github.com/lbcb-sci/racon).

Pairwise variant calling was conducted via Snippy (v4.6.0; github.com/tseemann/snippy) aligning raw sequencing reads (FASTQ) against a previously assembled and annotated reference genome of the K. quasipneumoniae strain ATCC 700603. Annotation of the reference genome was performed via Bakta v1.9.3 [49]. As a preliminary quality control step, potential methylation-induced basecalling errors were assessed with MPOA (v1.5.0) [50] and confirmed to not impact the final variant calls.

Statistical analysis

Visualization and statistical analyses were performed via R Statistical Software v.4.5.1 (R Core Team 2021, Vienna, Austria). The following R packages were used: ggplot2 (v3.4.0) for data visualization; purrr (v1.1.0) for functional programming workflows; rstatix (v0.7.2) for statistical tests and summary statistics; and tidyverse (v2.0.0) for data processing and plotting. A p value of < 0.05 was considered statistically significant across all tests. The experiments were performed in three independent biological replicates (n = 3), each consisting of three technical replicates. For statistical analysis, the mean of the technical replicates was used for each biological replicate. Statistical testing included one-way and two-way ANOVA, followed by appropriate post hoc multiple comparisons tests, depending on the experimental design (indicated in the figure legends).

Results

Phage‒antibiotic synergy in planktonic cultures

To quantify PAS under planktonic conditions, we monitored the growth kinetics of K. quasipneumoniae ATCC 700603 over 24 h in the presence of CAZ (0.25 × –2 × MIC), vB_KpUKJ_2 phage (10⁴–10¹⁰ PFU/mL), or their combination. The combination treatments included all tested concentrations of antibiotic pairs with three representative phage titers of 10⁴, 10⁶, and 10⁸ PFU/mL.

CAZ mono-treatment had the expected concentration-dependent effect (Fig. 1A): at 2 × and 1 × MIC (64 and 32 mg/L), growth was fully suppressed, whereas at 0.5 × MIC, growth inhibition was transient, with regrowth observed after approximately 14 h. The lowest dose (0.25 × MIC) produced only a brief delay before the resumption of exponential growth, similar to the untreated control. Phage mono-treatment induced a uniform but temporary delay in bacterial growth across all the tested titers, with no clear dose dependency (Fig. 1B). Regardless of the input titer, regrowth occurred after 8–10 h, likely reflecting the emergence of phage-resistant subpopulations. Combination treatments showed phage titer-dependent PAS effects: the combination of phage titers of 10⁴ PFU/ml and 10⁶ PFU/ml with CAZ had no visible effects on growth (Fig. 1C and D). Only a high phage titer of 10⁸ PFU/ml showed visible effects when combined with CAZ, which was characterized by a lack of growth over an entire period of 25 h, even at a sublethal CAZ concentration of 0.25 × MIC (Fig. 1E).

Fig. 1.

Assessment of PAS effects on planktonic K. quasipneumoniae ATCC 700603. A–E Bacterial growth curves under different treatment conditions (as indicated in the diagrams). The concentrations of CAZ as the fold change in the MIC (MIC_CAZ = 32 mg/L) and phage titers are indicated in the legends. The data are presented as the means ± standard deviations (SDs, n = 3)

To quantify the interaction effects in planktonic cultures, we calculated HSA synergy scores (_SHSAs) on the basis of differences in the AUCs of the growth curves (more details are provided in Sect. 2.13). An additive effect was observed for the combination of CAZ 0.5 × MIC and phage 10⁸ PFU/mL (S_HSA = 0.05). In contrast, a strong synergistic effect emerged at CAZ 0.25 × MIC combined with phage 10⁸ PFU/mL, yielding a higher synergy score (S_HSA = 0.29) (see Supplementary Material, Table S1). These findings confirm that PAS is strongly dose dependent in planktonic cultures, with high phage input enabling subinhibitory antibiotic concentrations to achieve full suppression of bacterial growth.

Synergistic biofilm clearance of combined phage–antibiotic treatment

To evaluate the efficacy of PAS in reducing K. quasipneumoniae biofilms, we performed time-lapse LSFM imaging over 24 h and quantified the viable biofilm surface area covered (SAC). In the untreated controls, biofilm viability remained stable throughout the experiment (Supplementary Material, Fig. S1, A–D), maintaining 50–55% SAC (Fig. 2A, gray line), and yielding the highest AUC values (1271.0 ± 68.7 relative fluorescence units (RFU) × h) (Fig. 2B).

Fig. 2.

Quantitative evaluation of treatment effects on K. quasipneumoniae ATCC 700603 biofilms following exposure to CAZ (0.25 × MIC), phage vB_KpUKJ_2 (10⁸ PFU/mL), or their combination. A Surface area covered (SAC) by viable biofilm (in %) over time, as determined by the calcein fluorescence signal after staining with CAM. A total pixel area of 2048 × 2048 µm was analyzed from LSFM micrographs; B AUC representation of the kinetic data from (A) over a treatment period of 0–24 h, based on relative fluorescence units (RFU) × h; C Viable bacterial counts (CFU/mL) after biofilm treatments at three distinct time points. Statistical analysis was performed via two-way ANOVA followed by Tukey’s multiple comparisons test. D Phage particles (PFU/mL) after combined biofilm treatment at three time points. Statistical analysis was conducted via one-way ANOVA followed by Dunnett’s post hoc test. The data are presented as the means ± SDs (n = 3). Significance was assumed at p < 0.05 and is indicated by asterisks: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001

CAZ mono-treatment (8 mg/L) produced a modest reduction in the viable area during the first 4–6 h (Supplementary Material, Fig. S1, E–G), reaching its lowest extent at 6 h (38.2% ± 6.76%; Fig. 2A, blue line), followed by regrowth from 8 h onward, with a high AUC (1166.0 ± 79.2 (RFU) × h) (Fig. 2B), highlighting the limited antibiofilm activity of CAZ. Phage mono-treatment (10⁸ PFU/mL) induced a pronounced early reduction in viable biofilm coverage (Supplementary Material, Fig. S1, I–J), reaching a minimum of 16.9% ± 7.8% at 8 h, followed by partial regrowth to approximately 30% by 24 h (Fig. 2A, green line). Despite this rebound, the overall biofilm mass remained significantly lower than that of the untreated control, as reflected by a reduced AUC (715.6 ± 90.2 RFU × h; p < 0.0001) (Fig. 2B).

Combining the same high-dose phage (10⁸ PFU/mL) and subinhibitory concentration of CAZ (8 mg/L) led to a rapid and sustained reduction in biofilm coverage (Supplementary Material, Fig. S1, M–P), decreasing below 10% by 6 h and remaining suppressed through 24 h (Fig. 2A, red line). This yielded the lowest AUC of all the groups (386.6 ± 46.0 RFU × h), which was significantly lower than that of both mono-treatments (p < 0.0001) (Fig. 2B). AUC-based synergy analysis revealed an S_HSA of 0.21, confirming a strong synergistic effect. These results demonstrate that PAS facilitates biofilm eradication at antibiotic concentrations two dilution steps below the MIC (i.e., 8 mg/mL compared with 32 mg/mL), highlighting its therapeutic potential.

To complement the LSFM analysis, we quantified viable bacteria (CFU/mL) and phage titers (PFU/mL) in mature biofilms after treatment with CAZ, phages, or their combination. Directly after exposure to the treatment and processing of the biofilms (these steps took a few minutes, the starting point was approximately 15 min; however, it was set in the graphics, tables and analyses to zero), the bacterial counts were comparable across the control (1.10 × 10⁹ ± 1.73 × 10⁸ CFU/mL), CAZ (6.59 × 10⁸ ± 1.39 × 10⁸ CFU/mL), and phage groups (1.20 × 10⁹ ± 3.46 × 10⁸ CFU/mL), with no significant differences observed (Fig. 2C). In contrast, the combination treatment group presented a significantly reduced count of 2.21 × 10⁷ ± 4.78 × 10⁶CFU/mL (~ 2 log₁₀ reduction, p < 0.0001) for this short exposure time, indicating high killing activity. After four hours, the treatments began to diverge: phage mono-treatment reduced the number of viable bacteria to 3.41 × 10⁵ ± 1.74 × 10⁵ CFU/mL (~ 4 log₁₀ reduction vs. control, p < 0.01), whereas compared with the control, CAZ alone slightly but significantly reduced the number of viable bacteria (5.42 × 10⁸ ± 1.53 × 10⁸ CFU/mL) (p < 0.01). In contrast, the combination treatment resulted in a strong ~ 6 log₁₀ reduction in the bacterial count (to 3.00 × 10³ ± 1.01 × 10³ CFU/mL, p < 0.0001 vs. all other groups) (Fig. 2C). After 24 h, only the effect of CAZ on the biofilms increased, resulting in a further reduction in the number of viable bacteria to 3.73 × 10⁷ ± 2.81 × 10⁷ CFU/mL. Neither the phages alone nor in combination with CAZ improved antibiofilm activity (Fig. 2C).

For the endpoint CFU/mL data, synergistic effects were assessed via a mixed-effects model [46] to calculate the LRIC coefficient. Positive interaction coefficients were observed at all time points: 1.51 at 0 h, 1.48 at 4 h, and 0.80 at 24 h (p < 0.0001 for all). These results confirm the strong and sustained synergistic effect of the phage–CAZ combination in reducing biofilm viability.

Phage dynamics after biofilm treatment

Phage replication dynamics provided further mechanistic insight (Fig. 2D). Despite an initial inoculum of 1 × 10⁸ PFU/mL, the measured titer decreased to 2.48 × 10⁵ ± 1.86 × 10⁵ PFU/mL after 15 min (the initial time point, t ≈15 min, on the graph indicated as 0 h), indicating rapid adsorption of phages to bacterial cells. The titers subsequently increased significantly after 4 h (to 8.52 × 10⁷ ± 6.73 × 10⁷) and 24 h (1.66 × 10⁸ ± 9.31 × 10⁷ PFU/mL) (p < 0.05), indicating active phage replication within the biofilm.

Rapid and progressive degradation of EPS polysaccharides after phage treatment

The polysaccharide fraction of EPS forms the core structural scaffold of bacterial biofilms, acting as both a physical barrier and a protective shield against antimicrobials. Understanding how phages disrupt the matrix is therefore crucial for elucidating the mechanisms of biofilm destruction. To assess these effects, we repurposed established staining methods [37] to visualize α- and β-polysaccharides within the EPS. Using CLSM, we tracked polysaccharide dynamics at 0, 4, and 24 h after phage exposure. The α-polysaccharides were labeled with Con-A (green), and the β-polysaccharides were labeled with calcofluor white (magenta), yielding complementary signals that revealed progressive changes in EPS architecture (Fig. 3). The control biofilms exhibited dense, well-structured matrices with strong Con-A and moderate CFW fluorescence, indicating abundant α- and β-polysaccharide components with no immediate structural disruption evident (Fig. 3A); similar structures were observed directly after phage treatment (Fig. 3B). While the α- and β-polysaccharide compositions remained stable in the control biofilms over time, confirming further biofilm maturation, the phage-treated biofilms presented visibly reduced Con-A and CFW fluorescence after 4 h (Fig. 3B), suggesting that EPS degradation, particularly that of β-polysaccharides, occurred. After 24 h, phage-treated biofilms presented nearly complete loss of α- and β-polysaccharide signals and a collapsed EPS architecture (Fig. 3B).

Fig. 3.

Representative CLSM images showing the effects of phage vB_KpUKJ_2 (10⁸ PFU/mL) on mature K. quasipneumoniae ATCC 700603 biofilms at 0, 4, and 24 h post-treatment. Biofilms were stained with Con-A (green) for α-polysaccharides or with CFW (magenta) for β-polysaccharides to visualize extracellular matrix integrity. A Untreated control, B phage-treated samples. To minimize photobleaching and laser-induced biofilm disruption, independent samples were used for each time point. Each image represents a representative Z-stack projection from three independent experiments. Scale bars: 20 μm

Frequencies of phage and antibiotic resistance in planktonic bacteria and biofilms

Determining resistance frequencies is critical for predicting the risk of phage therapy failure and developing control strategies. To address this, phage and antibiotic treatments were assessed individually and in combination for their efficacy against planktonic and biofilm bacteria. Phage mono-treatment achieved high killing efficiency in both planktonic bacteria and biofilms but was associated with the emergence of resistant mutants, particularly in planktonic cultures (Table 1). In contrast, CAZ mono-treatment resulted in a dose-dependent reduction in killing efficiency, with substantially lower efficacy in biofilms. The resistance frequency under CAZ treatment remained low (Table 1). Remarkably, all phage–CAZ combinations resulted in complete bacterial eradication (100% killing efficiency) under both planktonic and biofilm conditions, with no detectable resistant colonies.

Table 1.

Killing efficiency and resistance frequency in planktonic and biofilm bacteria following mono- and phage‒antibiotic combination treatments [mean ± standard deviation (SD), n = 3]

Treatment	Planktonic		Biofilm
	Killing efficiency (%)	Resistance frequency (mean ± SD)	Killing efficiency (%)	Resistance frequency (mean ± SD)
Phage	99.995 (± 0.001)	4.76 × 10–3 (± 5.6 × 10–3) *	99.781 (± 0.283)	2.76 × 10–4 (± 3.26 × 10–4) *
1 × CAZ	98.652 (± 0.407)	8.66 × 10–5 (± 1.09 × 10–4) #	38.210 (± 7.388)	1.33 × 10–5 (± 1.12 × 10–6) #
0.5 × CAZ	84.779 (± 3.868)	nd	25.452 (± 6.252)	nd
0.25 × CAZ	62.652 (± 13.215)	nd	24.964 (± 7.207)	nd
Phage + 1 × CAZ	100	0 #	100	0 #
Phage + 0.5 × CAZ	100	nd	99.997 (± 0.001)	nd
Phage + 0.25 × CAZ	100	nd	99.999 (± 0.001)	nd

^*refers to resistance against phage vB_KpUKJ_2 (selection at 10⁸ PFU/mL), ^# refers to resistance against CAZ (selection at 1 × MIC). CAZ ceftazidime, nd not determined

Phenotypic characterization and fitness assessment of phage-resistant mutants

In contrast to combination treatment, phage mono-treatment readily facilitated the evolution of resistance. To assess potential fitness trade-offs, factors critical for treatment efficacy and long-term therapeutic success, we characterized the phage-resistant isolates microbiologically and at the molecular level.

To begin this characterization, we first confirmed a stable phage-resistant phenotype in all the mutants via spot assays, which revealed no detectable lysis zones. The resistance was maintained after five serial passages without selective pressure, indicating genetic stability. Whole-genome sequencing further verified that all the isolates were K. quasipneumoniae ATCC 700603, thereby excluding the possibility of contamination.

To evaluate whether phage resistance affects susceptibility to the antibiotic CAZ, we determined the CAZ MICs for all the mutant strains. This analysis aimed to identify possible cross-resistance (increased antibiotic resistance following phage resistance) or collateral sensitivity (increased antibiotic susceptibility as a cost of phage resistance). Two mutants, Mut_2 and Mut_4, retained the same MIC as the parental strain (32 mg/L), whereas the remaining mutants presented an increased MIC of 64 mg/L. According to the EUCAST breakpoints, however, all strains remained within the resistant category (Supplementary Material, Table S2).

In addition, three mutants (Mut_2, Mut_3, Mut_4) presented a stable small-colony phenotype with visibly reduced colony size compared with the parental strain (Supplementary Material, Fig. S2). This morphological change persisted after repeated passages, suggesting underlying metabolic or regulatory alterations.

Functional fitness was assessed through growth kinetics and biofilm formation (Fig. 4) and the ability to survive in human serum, reflecting resistance to complement-mediated killing (Supplementary Material Fig. S3). Growth performance, as assessed by the AUC over 24 h, revealed that Mut_1, Mut_2, and Mut_4 presented significantly reduced growth compared with the parental strain (p < 0.001, p < 0.01, and p < 0.0001, respectively), with Mut_4 showing the greatest reduction (19.71 ± 0.32 vs. 23.4 ± 0.50 for ATCC 700603) (Fig. 4A). Mut_3, Mut_5, and Mut_6 presented AUC values comparable to those of the wild type.

Fig. 4.

Characteristics of six phage-resistant mutants in comparison with the parental K. quasipneumoniae ATCC 700603 strain. A Differences in the AUCs based on growth kinetics. B Biofilm formation ability at 48 h quantified by crystal violet staining and absorbance measured at 570 nm; the data are presented as the means ± SDs (n = 3). Statistical analyses were performed via one-way ANOVA followed by Tukey’s multiple comparisons test. Significance was assumed at p < 0.05 and is indicated by asterisks: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0. 0001

Biofilm mass quantification revealed that Mut_2, Mut_4, and Mut_5 resulted in significantly less biomass than did the parental strain (0.74 ± 0.06, 0.71 ± 0.07, and 0.50 ± 0.04 vs. 0.98 ± 0.12; p < 0.05, p < 0.001, p < 0.05, respectively; Fig. 4B). In contrast, Mut_1, Mut_3, and Mut_6 produced biofilm levels comparable to those of ATCC 700603, indicating a preserved capacity for biofilm formation.

Serum survival assays revealed that all strains displayed an initial decrease in CFU/mL after 1 h in 75% human serum, followed by recovery for 3 h (Supplementary Material Fig. S3). Most of the mutants displayed survival kinetics similar to those of the parental strain, except Mut_4, which maintained slightly lower counts at all time points, with a significant difference at 3 h (p < 0.05), indicating a minor but detectable fitness cost in this strain.

Genomic features responsible for phage resistance

To identify genetic determinants of phage resistance, we performed whole-genome sequencing on resistant isolates and mapped the reads to the parental K. quasipneumoniae ATCC 700603 reference genome. Variant calling revealed a range of mutations, including single nucleotide polymorphisms (SNPs), small insertions and deletions (indels), and frameshifts affecting coding regions across multiple functional categories, such as outer membrane proteins, metabolic enzymes, and transcriptional regulators (Table 2).

Table 2.

Genetic changes identified in phage-resistant K. quasipneumoniae mutants

Mutant	Gene	Nucleotide position	Product	Variation	Amino acid position	Mutation type	Protein effect	Gene function
Mut 1	rpoC	2162234	RNA polymerase β′ subunit	T → A	430/1407	Missense (SNP)	p.His430Leu	RNA polymerase β′ subunit
	55 bp upstream of mtnC*	5048720	Acireductone synthase	C → T	–	Missense (SNP)	Putative promoter region disruption	Methionine salvage pathway
Mut 2	purA	1942670	Adenylosuccinate synthase	A → T	127/432	Missense (SNP)	p.Ile127Asn	Purine biosynthesis
Mut 3	dadA	1025853	D-amino acid dehydrogenase	A → C	374/433	Missense (SNP)	p.Thr374Pro	Cell wall remodeling
	fhuA	1459656	Ferrichrome porin	G → GGC	95/735	Frameshift insertion	Premature truncation of protein	Loss/alteration of receptor
	piuB	3638660	PepSY-domain protein	A → C	264/452	Missense (SNP)	p.Thr264Pro	Iron uptake
Mut 4	dadA	1025853	D-amino acid dehydrogenase	A → C	374/433	Missense (SNP)	p.Thr374Pro	Cell wall remodeling
	piuB	3638660	PepSY-domain protein	A → C	264/452	Missense (SNP)	p.Thr264Pro	Iron uptake
Mut 5	rlhA	4789077	Protease YdcP	C → A	535/653	Missense (SNP)	p.His535Asn	RNA hydroxylation
	40 bp upstream of digH#	875348	Glycosyl hydrolase	G → A	–	Missense (SNP)	Putative transcriptional impact	Cell wall remodeling
Mut 6	fhuA	1458597	Ferrichrome porin	37 bps → G	437/735	In-frame deletion	p.Val437_Ala448del	Loss/alteration of receptor

bp base pairs, SNP single-nucleotide polymorphism

^*The mtnC gene is located at positions 875403–876092;

^#The digH gene is located at positions 5048760–5050052

Two isolates (Mut_3 and Mut_6) carried disruptive mutations in fhuA, which encodes the ferrichrome porin, a known outer membrane receptor for phages in Enterobacteriaceae [51, 52]. Mut_3 harbored a frameshift insertion that introduced a premature stop codon, whereas Mut_6 exhibited an in-frame deletion. Given the established role of FhuA in phage adsorption [53], these mutations likely confer resistance by preventing receptor-mediated entry.

Additional mutations implicated intracellular pathways potentially required for phage replication. Mut_2 harbors a missense mutation in purA, which encodes adenylosuccinate synthase, an essential enzyme in purine biosynthesis [54]. Mut_1 carried a variant in rpoC encoding the β′ subunit of RNA polymerase [55].

Several mutations have also been identified in genes associated with envelope physiology and transport. Mut_3 and Mut_4 both presented SNPs in piuB, which encodes a PepSY-domain protein potentially involved in iron uptake and transport regulation [56, 57], as well as in dadA, a gene linked to D-amino acid metabolism and peptidoglycan remodeling [58]. Additionally, Mut_5 carried a missense variant in rlhA encoding a predicted RNA hydroxylase or peptidase [59].

Finally, variant calling detected intergenic SNPs in Mut_1 and Mut_5. Both mutations are located within the proximal promoter regions, 55 and 40 bp upstream of mtnC and digH, respectively.

Discussion

In this study, we demonstrate that combining the virulent phage vB_KpUKJ_2 with the β-lactam antibiotic CAZ not only eradicates both planktonic and biofilm-associated populations of CAZ-resistant K. quasipneumoniae but also fully suppresses the emergence of phage-resistant mutants. Although the strain used was resistant to CAZ (MIC = 32 mg/L), we tested whether subinhibitory antibiotic concentrations could enhance phage activity and vice versa. PAS levels were first quantified in planktonic cultures across a range of phage–antibiotic ratios to define optimal dosing conditions. We then monitored treatment dynamics and spatiotemporal interactions within biofilms via LSFM in a time-resolved experiment, leveraging the low phototoxicity of LSFM to enable real-time imaging of live biofilms. The acquired data were processed through semiautomated ImageJ workflows custom-developed in-house for volumetric preprocessing and quantitative viability analysis. To our knowledge, this study presents the first application of LSFM to dynamically monitor PAS in biofilms, enabling minimally invasive, high-resolution visualization of treatment responses over a 24-h period. In parallel with live imaging, CFU and PFU assays were performed to provide complementary measures of bacterial killing and phage propagation. To further dissect the mechanisms underlying PAS, extracellular polysaccharide profiling was performed via CLSM, which, owing to its greater phototoxicity, was employed only at discrete time points on separate samples to assess polysaccharide degradation in the matrix. Finally, we quantified mutation frequencies under both planktonic and biofilm conditions, characterized phage-resistant isolates, and conducted whole-genome sequencing to identify underlying genetic alterations possibly associated with fitness trade-offs.

Phage–antibiotic combinations have been widely proposed as a strategy to increase antimicrobial efficacy and reduce the emergence of resistance [24]. Several prior studies have demonstrated that such combinations suppress phage-resistant populations across various bacterial pathogens [20, 60, 61]. Similarly, our findings highlight that a synergistic effect between phage and antibiotic treatments effectively enhanced bacterial killing while completely suppressing the emergence of resistance in both the planktonic and biofilm states.

Studying EPS polysaccharides provides a valuable proxy for assessing the structural integrity of the biofilm matrix, which plays a key role in protecting embedded bacteria from external agents [62]. Using established matrix-binding dyes, we observed the progressive depletion of both α- and β-polysaccharides during phage mono-treatment. These observations suggest that phage vB_KpUKJ_2 is associated with the disruption of the EPS structure and maintains a sustained effect over 24 h, leading to a reduction in key matrix components.

This matrix disruption is likely to facilitate phage amplification and enhance antibiotic penetration [63]. While such effects are commonly attributed to phage-encoded depolymerases, enzymes that selectively degrade exopolysaccharides to weaken the biofilm barrier [23, 64], this mechanism remains a likely, though not yet biochemically confirmed, driver of the depletion observed here. Supporting this hypothesis is the genomic and phenotypic profile of phage vB_KpUKJ_2, which encodes a putative tail-associated depolymerase and displays a halo phenotype, both hallmark indicators of enzymatic matrix digestion [33]. Similar activities have been reported in other systems; for example, Dpo71 enhanced colistin efficacy against Acinetobacter baumannii [65], whereas Dep37 improved kanamycin delivery in K. pneumoniae biofilms [66]. These findings underscore the potential of depolymerase-carrying phages to augment biofilm clearance by weakening matrix defenses [67].

Moreover, vB_KpUKJ_2 replicates rapidly and lacks a discernible latent period [33], which is consistent with our observations of fast phage adsorption and sustained propagation within biofilms, further supporting its suitability for combination treatment against pathogenic biofilms. Notably, in our study, CAZ combined with phage was effective against biofilm-associated K. quasipneumoniae even at sublethal concentrations, despite the strain’s ESBL genotype [36]. This observation aligns with existing studies showing that low-dose β-lactams can modulate bacterial surface receptors and stress responses in ways that increase phage adsorption and lysis [68]. For example, in E. coli, subinhibitory antibiotics were found to upregulate phage receptor expression and increase susceptibility to phage infection [34]. Moreover, plasmids encoding ESBLs impose a metabolic burden on the host cell [69], potentially diminishing the bacterium's ability to withstand additional selective pressures such as those imposed by phage predation.

In our planktonic experiments, we observed a clear dose-dependent effect of PAS. High phage titers (10⁸ PFU/mL; MOI 1.0), when paired with subinhibitory CAZ (8 mg/L), achieved complete clearance, whereas lower phage inputs (10⁴–10⁶ PFU/mL; MOI 0.0001–0.01) allowed for bacterial regrowth. This threshold likely reflects a requirement for sufficient MOI to reduce bacterial populations below the recovery threshold; at high MOIs, phages replicate exponentially and maintain pressure; at low MOIs, subpopulations of bacteria that persist despite antibiotic administration can escape and repopulate [70, 71]. These findings reinforce the notion that PAS depends not only on mechanistic synergy but also on optimized dosing ratios [19, 72]. With respect to the outcomes of CAZ mono-treatment at lower concentrations, the observed regrowth is likely not due to antibiotic degradation, since CAZ remains relatively stable over 24 h [73], but rather to bacterial adaptation driven by stochastic gene expression.

In line with previous findings, our experiments revealed that phage mono-treatment effectively reduced bacterial loads but also led to the emergence of phage-resistant mutants, a well-documented limitation of phage monotherapy, even when resistance is associated with fitness costs [20, 44]. To characterize these resistant variants, we performed whole-genome sequencing and identified mutations spanning multiple functional categories, including outer membrane receptors (fhuA), purine biosynthesis (purA), and transcriptional regulation (rpoC), suggesting the presence of diverse resistance mechanisms. Given the established role of FhuA in phage adsorption [53], these mutations likely confer resistance by preventing receptor-mediated entry. Perturbations in the purine biosynthesis pathway could reduce intracellular nucleotide pools, limiting the DNA replication capacity of the phage [74], whereas disruption of global transcriptional dynamics can reduce the availability of host factors required for efficient phage propagation [75]. While the precise roles of these genes in phage resistance are unclear, they may contribute to membrane remodeling or stress responses that indirectly alter phage adsorption or provide compensatory benefits under phage pressure. These mutations may be associated with distinct physiological trade-offs: while fhuA mutants appeared to retain biofilm-forming capacity, purA and rpoC variants presented reduced growth and impaired biofilm formation. We also frequently observed alterations in genes involved in surface structure and stress responses, indicating that resistance may compromise cooperative traits critical for biofilm persistence [76]. Such trade-offs highlight exploitable weaknesses in resistant subpopulations, offering a rationale for evolution-informed therapeutic strategies [77]. Two intergenic SNPs located 40 bp and 55 bp upstream of digH and mtnC, respectively, were identified within the proximal promoter regions. Although subtle, these mutations may influence transcriptional regulation and fine-tune gene expression, thereby potentially contributing to the observed phage-resistant phenotype [78, 79].

In addition to these genetic alterations, Klebsiella spp. are known to employ additional, often reversible, resistance strategies that may affect therapeutic outcomes. These include capsule modulation, which enables transient evasion of phage adsorption [80], as well as less common CRISPR‒Cas systems and abortive infection mechanisms that can restrict phage propagation [81, 82]. Together, these mechanisms complement the mutations observed here and underscore the multifaceted nature of phage–bacterium interactions relevant to phage therapy [83].

Our findings contribute to the growing body of evidence challenging the conventional assumption that antibiotics lacking standalone efficacy have no therapeutic value, an idea increasingly questioned in the context of PAS. Clinical observations support this evolving view: Bao et al. reported successful treatment of MDR K. pneumoniae infections using a phage–antibiotic combination that included a previously ineffective drug [21]. Similarly, in our study, CAZ, which is clinically inactive against an ESBL-producing strain (MIC = 32 mg/L), regained potent antibiofilm activity when combined with phages, achieving clearance at sublethal concentrations (8 mg/L). This may have important clinical implications, particularly given the challenges of achieving high antibiotic concentrations in vivo due to pharmacokinetic limitations and toxicity risks [84]. By enabling bacterial eradication at lower, pharmacologically attainable doses, PAS may offer a strategy to enhance the utility of otherwise ineffective antibiotics [60], particularly in biofilm-associated infections linked to implanted medical devices, where tolerance and relapse are common [85]. Moreover, the observed dose‒dependent synergy highlights the potential for rational optimization of PAS regimens, though further studies are needed to confirm their efficacy and role in resistance suppression.

In addition to therapeutic insights, our study introduces a scalable and reproducible image analysis pipeline tailored for LSFM datasets. By developing custom macros for semiautomated preprocessing and quantification, we aimed to address a persistent challenge in volumetric imaging of biofilms: the lack of standardized tools for high-throughput viability analysis. These tools help lower technical barriers and may serve as a useful foundation for integrating LSFM into microbiological workflows, particularly for researchers seeking to visualize and quantify treatment effects in complex, living communities. As LSFM continues to gain interest in microbial and antimicrobial research, such pipelines have the potential to support wider adoption and promote analytical consistency across studies.

While our findings offer valuable insights into PAS, the study’s limitations should also be noted. Our experiments were conducted using a static in vitro biofilm model on abiotic surfaces. While appropriate for mechanistic work, these conditions do not fully replicate the complexity of flow-associated device biofilms, as they lack the shear stress and nutrient gradients typically found in clinical settings. Furthermore, this study utilizes a single-strain–single-phage model centered on the reference strain ATCC 700603. While the use of a well-characterized reference strain enhances reproducibility and provides a clear baseline for this proof-of-concept study, it limits the immediate generalizability of the results to the vast diversity of clinical isolates. Host-specific factors and the high degree of genomic variation in K. pneumoniae mean that outcomes may vary across different phage-antibiotic combinations. Therefore, broader testing and in vivo validation remain essential for translating these findings into clinical applications.

In this context, Klebsiella species are increasingly recognized as important pathogens in implant-associated biofilm infections, including those involving artificial hearts and mechanical circulatory support (MCS) devices [86]. Because the removal of infected MCS systems is often not feasible due to high mortality risk, patients frequently rely on suppressive rather than curative antibiotic therapy [87]. In a large retrospective study, ESBL-producing Klebsiella species were identified as causative agents in approximately 10% of MCS device–associated infections [86], underscoring their clinical relevance. In such refractory cases, phage–antibiotic combinations such as the PAS approach explored here could represent a promising adjunctive strategy, pending further preclinical and clinical validation. The phage used in this study, previously characterized for its broad host range across Klebsiella species, including K. pneumoniae and K. quasipneumoniae [33], may support future translational exploration, though its activity remains strain-specific.

Conclusion

This study provides preclinical evidence that a virulent phage can enhance the efficacy of an otherwise ineffective β-lactam against MDR K. quasipneumoniae biofilms, likely by promoting matrix degradation and improving antibiotic penetration. This approach may help expand future treatment strategies by repurposing existing antibiotics through PAS. The rapid adsorption, robust amplification, and treatment ratio dependence of vB_KpUKJ_2 suggest that phage biology may play a key role in influencing PAS outcomes. From a translational perspective, our findings suggest a possible precision-guided avenue for managing recalcitrant, device-associated infections, in line with emerging evidence supporting phage–antibiotic combinations. However, further in vivo and clinical studies will be essential to assess this potential. By integrating microbiology, live imaging, and genomics, this work supports the rationale for PAS and contributes a methodological framework that may inform future translational research.

Abbreviations

AMR

Antimicrobial resistance

ANI

Average nucleotide identity

AUC

Area under the curve

Phages

Bacteriophages

bp

Base pairs

CAZ

Ceftazidime

CAM

Calcein acetoxymethyl ester

CFW

Calcofluor White

CFU

Colony-forming units

CLSM

Confocal laser scanning microscopy

Con-A

Concanavalin A

EPS

Extracellular polymeric substances

ESBL

Extended-spectrum β-lactamase

EUCAST

European Committee on Antimicrobial Susceptibility Testing

FAS

Ferrous ammonium sulfate

HSA

Highest single agent

LB

Lysogeny broth

LRIC

Log-reduction interaction coefficient

LSFM

Light-sheet fluorescence microscopy

MCS

Mechanical circulatory support

MDR

Multidrug-resistant

MH

Mueller–Hinton

MIC

Minimum inhibitory concentration

MOI

MOI Multiplicity of infection 

NA

Numerical aperture

OD

Optical density

PAS

Phage–antibiotic synergy

PBS

Phosphate-buffered saline

PFU

Plaque-forming units

RFU

Relative fluorescence units

SAC

Surface area coverage

SNPs

Single nucleotide polymorphisms

Author contributions

All the authors reviewed and approved the article. Conception, study design and methodology: TT; Supervision: OM, MP and RE; Analysis: TT, OM, ML, EG; Investigation and visualization: TT and MM; Interpretation of data: TT, OM, EG; Resources: KM, OM, MP; Funding acquisition: MP and OM; Project administration: OM; Writing – original draft: TT; Critical revision and final approval of the manuscript: OM.

Funding

The author(s) declare that financial support was received for the research, authorship, and/or publication of this article. This project has received funding from the Federal Ministry of Education and Research (BMBF, Germany) Grant Numbers 13N15467 and 13N15720.

Data availability

The sequencing datasets supporting the conclusions of this article are available in the NCBI public database under BioProject accession number PRJNA1333837 and hyperlink to datasets in https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1333837.

Declarations

Competing interests

The authors declare no competing interests.