Bacteriophages as Trojan Horses for Antimicrobial Peptides Delivery

Abstract

The spread of multidrug-resistant (MDR) bacteria has renewed interest in combining the targeted killing of bacteriophages with immunomodulatory antimicrobial peptides (AMPs). AMPs offer broad antimicrobial, antibiofilm, and immunomodulatory effects, although their efficacy is limited by stability, delivery, and toxicity. Phage-based systems may help address some of these limitations by localizing antimicrobial activity and improving bacterial targeting. In this perspective, we treat engineered phages as programmable “Trojan horses” that deliver AMPs into their bacterial targets, framing this concept alongside the rapid growth of AI-guided AMP design as well as phage–host matching. The evidence thus far is largely preclinical. AMP-armed phages have shown activity in vitro and in animal models, while engineered phages have only recently entered early-phase clinical trials. Reasons why phage-delivered AMPs remain largely in the preclinical and early translational stages are delineated. We argue that the primary hurdle lies in the gap between the separate advancement of AMP design on one end and phage–host matching on the other. The alignment of these interests, along with manufacturing and regulatory efforts, will likely be what allows this therapy to reach the bedside.

1. The Rise in MDR Bacteria and Phage Therapy Limitations

The rapid rise in multidrug-resistant (MDR) organisms has created a critical need for alternative treatment options [1]. The discovery of newer antibiotics is laborious and complex; thus, new alternatives such as phage therapy are attracting renewed interest [2].

Bacteriophages are viruses that specifically infect bacteria. During the lytic cycle, they replicate within and destroy susceptible bacterial cells while not infecting human cells, and they generally have a much smaller impact on the normal microbiota than broad-spectrum antibiotics due to their host specificity [3]. In contrast to traditional chemically static antibiotics, phages can co-evolve with their bacterial hosts through mutations, recombination, or gene acquisition [4].

Clinical reports and compassionate-use cases have shown promising results of bacteriophages in treating difficult MDR infections, with clinical improvement and bacterial eradication exceeding 70% in treated cases in several recent series [5]. Phages have also been shown to disrupt bacterial biofilms, increase bacterial sensitivity to antibiotics, and synergize with antibiotics [6].

However, the therapeutic application of bacteriophages faces several distinct challenges. The effectiveness of phages is largely determined by their host range. Efforts to expand host range include phage cocktail development, phage engineering, and combination therapies [3]. Furthermore, several animal studies have shown evidence of reduced phage efficiency over time due to acquired phage resistance [7]. Other hurdles in the use of bacteriophages for MDR bacteria include the lack of regulation and the need for standardized protocols [1].

Arming phages with antimicrobial peptides (AMPs) provides a way to address several of these limitations. In this article, we treat AMPs and phages as a unified tool. We make the case that what most limits progress in these areas is not any single component but the separate pursuit of two computational efforts, i.e., AMP design and phage–host matching. The question is what the current evidence supports and where the concept still runs ahead of the data.

2. The Advantages of AMPs

AMPs are part of the innate immune system and have broad antimicrobial, antibiofilm, antiviral, and immunomodulatory properties [8]. They commonly work by disrupting bacterial cell membranes, although some AMPs can also interfere with cell wall formation, nucleic acid synthesis, and protein synthesis [9]. AMPs have shown activity against MDR organisms and generally have a lower tendency for resistance development compared to traditional antibiotics [10]. AMPs exert their bactericidal effects through electrostatic interactions that target microbial membranes and disrupt intracellular pathways, ultimately leading to cell death. This broad-spectrum efficacy against both Gram-positive and Gram-negative species minimizes the likelihood of pathogen resistance [11]. They may also contribute to wound healing and regulation of inflammation [12].

Several limitations, however, have restricted their widespread clinical use, including cytotoxicity, susceptibility to proteolytic degradation, poor pharmacokinetics, high manufacturing costs, and regulatory challenges [13,14].

Advanced delivery systems, including nanoparticles, hydrogels, and other carrier platforms, can improve the stability, bioavailability, and targeted release of AMPs, helping to reduce the effects of AMPs intrinsic toxicity and protect them from degradation [15,16].

3. Phage–Antibiotic Synergy

Naturally occurring bacteriophages can present synergistic bactericidal activity to routine antibiotics [17]. AMPs have also demonstrated synergistic effects when combined with conventional antibiotics. This combination may lower the required antibiotic dose [18,19] and improve activity against drug-resistant pathogens [20]. When the three elements are combined, i.e., bacteriophage-AMP-ATB, bacterial survival of E. faecium is significantly decreased compared to using either of the elements alone [21]. Their use is especially effective in biofilm-producing bacteria like Pseudomonas and Staphylococcus aureus [22].

AMPs can be encoded in the phage for specific delivery. This has shown promising against A. baumannii, Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus, and Bacillus subtilis [23].

A most important aspect is that phage–antibiotic combinations can take advantage of bacterial efflux pump trade-offs and help re-sensitize ESKAPEE pathogens [24] (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter species, and Escherichia coli) to antibiotic therapy [4,5].

4. Engineering Bacteriophages as Trojan Horses for AMP Delivery

Several reviews have discussed engineered or programmable bacteriophages as potential future therapies. Beyond direct bacterial lysis, these phages could serve as vehicles for molecular reprogramming or combination therapies with small molecules and nanoparticles [4,25].

Bacteriophages and bacteriophage-based systems are being used to deliver AMPs directly into bacteria (hence the phrase “bacteriophages as Trojan horses for AMP delivery”). These include phage gene delivery, the use of phage capsids as carriers, and synthetic phage-like nanoparticles.

Table 1. Studies Demonstrating Phage-Mediated Delivery or Expression of Antimicrobial Peptides.

Study	Phage System	AMP	Delivery Strategy	Target Organism	Key Findings
[26]	Engineered T7Select phage, 1018-T7	1018	AMP gene inserted into the T7Select phage genome and expressed during bacterial infection	E. coli	Enhanced killing of planktonic bacteria and improved disruption of established biofilms compared with unmodified T7Select phage.
[27]	Engineered T7Select phages	OmpA-Api805, CRAMP, melittin	AMP-encoding sequences inserted into the T7Select phage genome for expression during infection	E. coli	Engineered phages showed similar lysis to unmodified T7Select but partially reduced regrowth of potentially phage-resistant E. coli; melittin expression was directly confirmed.
[28]	CRISPR/Cas9-engineered lytic Salmonella phage, pST_PMR_LL37	LL-37	LL-37 displayed on the phage capsid surface	Salmonella Typhimurium	Enhanced adsorption, reduced bacterial regrowth and phage-resistant mutant emergence, decreased epithelial cell invasion/intracellular survival, and improved survival in a Galleria mellonella infection model.
[29]	Engineered M13-based phagemids	Cecropin PR-39 and apidaecin Ia, with CcdB toxin in optimized constructs	Phagemid particles delivered synthetic antimicrobial gene networks for intracellular expression without bacterial lysis	E. coli	Reduced bacterial viability in vitro and improved survival in a murine E. coli peritonitis model.
[30]	Engineered nonlytic M13 phage targeting LPS, PMB-M13αLPS	Polymyxin B	Polymyxin B chemically conjugated to the surface of engineered M13 phage for targeted delivery to Gram-negative bacteria	Multidrug-resistant Pseudomonas aeruginosa	Improved antibacterial potency in vitro and effectively treated P. aeruginosa pneumonia and corneal infection in mice with reduced toxicity compared with free polymyxin B.
[31]	Engineered Salmonella phage selzHA-TAT	HA-TAT cell-penetrating peptide	CPP displayed on phage GP94 to enhance mammalian cell uptake	Intracellular Salmonella	Increased intracellular phage uptake and improved killing of intracellular Salmonella in epithelial cell models without detectable cytotoxicity.
[32]	Engineered T7Ag-XII phage armed with silver nanoparticles	Silver nanoparticles	AgNP-binding peptide displayed on T7 phage capsid to bind silver nanoparticles	E. coli biofilms	T7 phages armed with AgNPs showed stronger biofilm eradication than phage or nanoparticles alone and were not toxic to eukaryotic cells at effective concentrations.
[22]	Phage-derived endolysins T7L and T4L	Colistin, polymyxin B, nisin	Combination of purified bacteriophage endolysins with AMPs	Pseudomonas aeruginosa and Staphylococcus aureus biofilms	T7L with polymyxin B or colistin synergistically eradicated P. aeruginosa biofilms, while T4L with nisin showed synergy against S. aureus biofilms.
[33]	Heterologous effector phage therapeutics, HEPTs	Colicin-like bacteriocins and cell wall hydrolases	Effector genes integrated into phage genomes for in situ production and release during host lysis	Uropathogens including E. coli, Klebsiella pneumoniae, and Enterococcus faecalis	HEPTs improved uropathogen killing, suppressed regrowth/resistance, controlled polymicrobial communities, and a colicin E7-producing HEPT improved control of patient E. coli bacteriuria ex vivo.
[34]	Phage-inspired targeted polymeric micelles using ϕ11 Gp45 or Gp45-derived peptides	Vancomycin and oxacillin	Phage receptor-binding protein or peptide-conjugated micelles used to target antibiotic delivery	Staphylococcus aureus sepsis model	Targeted micelles reduced MIC values and MiGp45 improved survival, reduced bacterial load, inflammation, lung injury, and oxidative stress in a mouse S. aureus sepsis model.
[35]	Bacteriophage-mimicking nanomedicines using RBPsb1	Nisin	Phage receptor-binding protein conjugated to nisin-loaded modules for infection-responsive release	MRSA/Staphylococcus aureus pneumonia	RBP-targeted nisin nanomedicines localized to infected lungs, reduced nisin toxicity, and improved therapeutic efficacy in a mouse MRSA pneumonia model.

lists studies showing bacteriophage-based AMP delivery systems against bacterial infections. Collectively, these studies suggest that phages can function not only as self-replicating antibacterials, but also as programmable delivery platforms capable of enhancing bacterial killing through a wide range of mechanisms. At present, however, AMP-delivery phages remain a proposed strategy rather than a well-established clinical approach.

In addition to targeting antimicrobial resistance, CRISPR-Cas9-engineered phages may be able to confer enhanced adsorption, as well as thermal and pH stability, compared to wild-type phages. Furthermore, wild-type phages were shown to permit bacterial regrowth compared to the engineered phages [28]. Most strikingly, in an epithelial cell interaction assay, the engineered phage significantly reduced bacterial attachment and internalization compared with its wild-type counterpart.

CRISPR-Cas9 can be used to characterize which AMPs are crucial for certain microorganisms [36]. The most important approach is to integrate DNA fragments encoding derivatives of an AMP into the phage genome to induce the production and release of such AMP within the phage infection cycle, targeting phage-resistant bacteria [27]. Through this genetic modification, bacteriophages with AMP-encoded peptides can partially inhibit the growth of phage-resistant bacteria, offering a potential new strategy to combat multi- and pan-resistant bacteria.

The bacteriophage Trojan horse concept has been validated in vitro and, in limited cases, in vivo. Multiple preclinical strategies have demonstrated the feasibility of phage-mediated antimicrobial payload delivery. These include AMP-expressing T7 phages [26], polymyxin-conjugated filamentous phages [30], effector-expressing phages producing colicin-like bacteriocins and hydrolases [33], and synergistic endolysin-AMP combinations [22] (Figure 1). Other approaches include non-lytic phagemids encoding antibacterial toxins [29], cell-penetrating peptide-displaying phages [31] and nanoparticle-armed phages targeting biofilms [32].

The antibacterial effect of bacteriophages can be enhanced with AMPs, conferring broad-spectrum antimicrobial or antibiofilm effects. Inserting the coding sequence for a particular AMP into the bacteriophage genome can lead to genetically modified bacteriophages that are effective against bacterial planktonic and biofilm cultures [26]. Colicins are a class of naturally occurring, high-molecular-weight antimicrobial proteins (bacteriocins) secreted by Escherichia coli and related bacteria. Bacteriophages engineered for target-specific effector gene delivery and host-dependent production of colicin-like bacteriocins demonstrated better therapeutic control of E. coli bacteriuria in urine samples from patients with UTI [33].

Phagemids, hybrid cloning vectors that combine the advantageous properties of both a plasmid and a bacteriophage, can be engineered to express AMPs and protein toxins, demonstrating successful bacterial death in an in vivo murine peritonitis infection model [29].

Most importantly, bacteria that were previously resistant to a particular bacteriophage can be overcome through bacteriophage genome manipulation. Lytic bacteriophages, along with highly active AMPs, are sometimes used in these processes. Such genetically engineered AMP-secreting lytic bacteriophages induce synergistic killing and prevent the rapid regrowth of phage-resistant bacterial subpopulations that often survive standard phage therapy [27].

Bacteriophages can be utilized as nanocarriers, since they naturally and specifically act as natural gene carriers to bacterial hosts. Bacteriophage-based systems using unique peptide libraries can provide bacteriophages with novel targeting properties [37].

Nanoparticle Strategies for AMPs Delivery

Mimicking bacteriophage-bacteria interactions through manipulation of the bacteriophage receptor proteins and coupling them to nano-drugs can enable precise delivery and targeting of antimicrobials [34]. Nanodelivery systems can mimic bacteriophages’ genome-delivery mechanism by utilizing bacteriophage receptor-binding proteins conjugated to loading modules. This approach leads to efficient targeting of bacterial pathogens [35, 38] and proves highly effective against bacterial biofilms [32].

Promising AMPs, such as nisin [39], have a strong cationic nature that causes membrane disruption. To mitigate any hemolytic and cytotoxic effects of nisin on mammalian cells, while preserving AMPs’ antimicrobial efficacy, researchers are actively exploring several promising strategies, including nanotechnology delivery systems. Such nanodelivery systems are not only remarkably specific in terms of the bacterial target but have also effectively delivered the antimicrobial in Methicillin-Resistant Staphylococcus aureus (MRSA)-infected tissue [35].

AMPs such as polymyxin B [39] have good antibacterial effects but are toxic, which limits their clinical use. Bridging the gap between in silico prediction and clinical translation remains a significant hurdle. Drug delivery to bacterial cells using a nanomaterial is a possible approach to ensure safe use of this drug. A non-lytic phage that recognized the lipopolysaccharide of Gram-negative bacteria conjugated to polymyxin filamentous cell-penetrating peptides was shown to be effective in vivo against P. aeruginosa pneumonia or corneal infection [30]. Combining polymyxin B and colistin with purified bacteriophage endolysin T7 can eradicate the biofilm formation of Pseudomonas aeruginosa [22]. In the same way, combining different AMPs such as colistin, nisin, and polymyxin B with phage-endolysin T4 is effective against Staphylococcus aureus biofilms [22].

5. What Is the Evidence Showing So Far

As MDR bacteria increasingly evade conventional antibiotics, precision antimicrobial strategies have re-emerged as a major area of therapeutic interest. Although AMP-armed phages remain largely experimental, advances in synthetic biology and artificial intelligence are rapidly accelerating their development.

At this time, no AMP-armed phage has entered clinical trials. However, two bioengineered phage products carrying CRISPR-Cas payloads indicate that engineered phages can be administered to humans with an acceptable safety profile while still retaining antibacterial activity [40]. SNIPR001, a cocktail of four CRISPR-Cas-armed bacteriophages, recently completed a Phase I randomized clinical trial in 36 healthy volunteers. The trial reported only mild/moderate adverse events and a 78% reduction in E. coli CFUs at the highest dose [40]. The second major clinical trial is the open-label ELIMINATE Phase 2 trial, LBP-EC01. Here, six CRISPR-Cas3-enhanced phages showed rapid E. coli reduction in urine by 4 h with complete UTI symptom resolution and no serious adverse events [41]. Together, these studies establish an important translational foundation for future AMP-armed systems as they demonstrate the feasibility of programmable antibacterial therapy.

6. Major Clinical Challenges

Although the results appear promising, at this time there are several barriers that must be considered. Restriction-modification, CRISPR-Cas immunity, and other bacterial defense systems limit phages, engineered or not. Furthermore, immune clearance, narrow host range, and evolving regulation for genetic modification complicate clinical translation.

Moreover, one of the greatest barriers to phage-AMP therapy is the enormous combinatorial complexity involved in identifying effective phage–host matches, designing stable antimicrobial payloads, and predicting resistance evolution. AMP mining, in which discriminative models review sequences to predict activity, toxicity, and stability, as well as AMP generation, in which generative models create novel peptide sequences that are optimized for efficacy, serve as strategies by which AI can help address these fundamental issues [38].

Despite their considerable promise, there are major challenges and reasons why phage-delivered AMPs remain largely in the preclinical and early translational stages. These clinical bottlenecks include:

-

Efficient delivery to the infection site. The engineered phage must reach the target bacteria in sufficient numbers and remain active in the host environment.

-

Narrow host range. While advantageous for preserving the microbiota, most bacteriophages infect only specific bacterial species or strains. This requires accurate pathogen identification, limiting effectiveness against polymicrobial infections.

-

Bacterial resistance. Bacteria can evolve resistance to phages by altering surface receptors, using restriction-modification systems, or activating CRISPR-Cas defenses.

-

Stable AMP expression by engineered phages. Inserted AMP genes can impose a fitness cost on the phage or be lost over time, reducing therapeutic reliability [42].

-

Immune clearance. The host’s immune system may still recognize and neutralize bacteriophages, particularly after repeated dosing. Immune clearance may then decrease therapeutic effectiveness before sufficient bacterial killing occurs [42].

-

AMP toxicity and expression control. AMPs must be produced at levels high enough to kill bacteria but low enough to avoid unintended toxicity or inflammatory responses [14].

-

Scalable manufacturing and regulatory approval pathways. In addition to strict control of purity, potency, genetic stability, and absence of bacterial contaminants such as endotoxins [43]; regulatory pathways for approval are still evolving. This makes clinical development more complex than for conventional antibiotics [44].

7. Using AI for Optimization

The recently unveiled ProteoGPT is a pre-trained protein large language model. AMPs generated by this model showed reduced susceptibility to resistance development by Carbapenem-resistant Acinetobacter baumannii (CRAB), MRSA in vitro [42], and comparable or superior efficacy to clinical antibiotics in in vivo infection models without organ damage [45]. Another approach is an explainable deep learning model known as EvoGradient, which acts similarly to in silico directed evolution and resulted in the virtual evolution, i.e., the computationally guided iterative mutation and selection of 32 peptides. It represents the same process performed in laboratory-based optimization, but instead done entirely in a virtual space. The most potent AMP, pep-19-mod, demonstrated activity against Carbapenem-resistant Enterobacteriales and Vancomycin-resistant Enterococci (VRE), achieving over 95% reduction in bacterial load in mouse models [46]. Finally, a recently developed M3-CAD, a multimodal pipeline integrating 3D structural features, enables de novo design of multi-mechanism AMPs with validated in vivo efficacy [47].

Despite this progress, reviews continue to highlight important caveats: training datasets are biased toward well-studied organisms, negative data are scarce, activity labels are assay-dependent, and bridging the gap between in silico prediction and clinical translation remains a significant hurdle [48].

AI is also addressing the congestion within phage-pathogen matching, in which specific phages are identified as effective antibacterial agents. Trained Random Forest models on phage–bacteria interactions can validate the ability to predict phage cocktails against untested E. coli strains directly from genome sequences. This is particularly important because strain-level genomic variation strongly influences phage susceptibility [49]. More sophisticated architectures are emerging for strain-level prediction, though current models face limitations from sparse training data and limited generalization [50,51,52].

8. What Is in the Future

The most immediate shift currently underway is a compression of the design phase. Platforms that are able to integrate computational prediction with wet-lab validation are substantially shortening iterative loops that once took years to mere weeks. This alone may be the advancement that brings AMP-phage to the clinical setting. There is also growing interest in designing peptides in accordance with the resistance profiles actively circulating in a given area, which would provide direct clinical relevance. Additionally, on the phage side, near-infrared bioimaging may offer a way to track phage distribution in real time, helping clarify questions regarding phage behavior that remain poorly understood [53,54].

The role of genomics and transcriptomics has a practical application. They help find AMP payloads and phage candidates, place and tune AMP genes in phage genomes, track phage–host responses and resistance, and screen safety and ecological effects [55,56]. To effectively integrate these approaches in MDR infections, phages, antibiotics, host immunity, and the local microbiota should be seen as one ecological system. Ecological partitioning and microbiome modulation play a role in enabling phage–antibiotic cooperation in a specific niche [56,57].

Phage-delivered AMPs represent a promising but still largely preclinical strategy against antibiotic resistance. As progress continues to be made, sequencing a patient’s isolate, matching it to a phage, and then arming it with a designer AMP payload in a clinically useful timeframe may not be too far away. The most significant hurdle, however, remains the disconnect between the parallel progress of AMP design and phage–host matching. Both the computational tools for AMP design and phage–host matching need to be unified rather than advancing separately. Regulatory frameworks and manufacturing at scale are issues that need to be solved.