Enhancing phage therapy through synthetic biology and genome engineering

Abstract

The antimicrobial and therapeutic efficacy of bacteriophages is currently limited, mostly due to rapid emergence of phage-resistance and the inability of most phage isolates to bind and infect a broad range of clinical strains. Here, we discuss how phage therapy can be improved through recent advances in genetic engineering. First, we outline how receptor-binding proteins and their relevant structural domains are engineered to redirect phage specificity and to avoid resistance. Next, we summarize how phages are reprogrammed as prokaryotic gene therapy vectors that deliver antimicrobial ‘payload’ proteins, such as sequence-specific nucleases, to target defined cells within complex microbiomes. Finally, we delineate big data- and novel artificial intelligence-driven approaches that may guide the design of improved synthetic phage in the future.

Introduction

In recent decades, the widespread and often poorly regulated use of antibiotics in medicine and agriculture has escalated the emergence of antibiotic-resistant pathogens [1]. The increasing burden placed on health care systems by these multidrug-resistant (MDR) microorganisms, alongside the simultaneous decline in pharmaceutical companies developing and stockpiling novel antibiotics [2], presents a dire threat to continued human prosperity [3]. As a result, there is rekindled interest in the use of phages as potential antimicrobial therapeutics [4], underlined by a series of recent successful compassionate use cases and upcoming randomized clinical trials (e.g. NCT03808103 and NCT04191148) [5,6]. However, in vitro and clinical data suggest that the efficacy of phage therapy using natural phage may be limited due to the rapid selection of phage resistant bacteria [6,7], due to immunogenicity as a consequence of prolonged treatment [8], and also because the host ranges of natural phage isolates rarely cover all clinically relevant pathogenic strains.

Advances in synthetic biology, where engineering principles are utilized to design biological systems from interchangeable parts, and the increasing pace of phage discovery may enable the augmentation of natural phage properties to overcome many of these shortcomings. Recent studies describing naturally evolved phage systems to broaden host range [9] and evade host defenses such as CRISPR-Cas [10,11], suggest that phage genomes contain a treasure trove of natural antimicrobial mechanisms and products to add to the synthetic biology toolbox. Here, we review the phage engineering techniques used to integrate such components into designer phages that possess enhanced properties over their naturally occurring counterparts, such as altered/broadened host ranges and the ability to transduce therapeutic payload genes to defined targets within the microbiota. We also give a speculative outlook on the future use of artificial intelligence and machine learning approaches to design the next generation of phage-based therapeutics.

Phage genome engineering methods

The stable integration of temperate phage genomes into host chromosomes results in a resident prophage, and generally enables virus manipulation via the same methodologies as those for bacterial genomes. In contrast, strictly lytic (virulent) phages require specialized genome engineering methods, which can all be classified into two broad conceptual groups: i) homologous recombination (HR), and ii) genome rebooting, that is, the activation of exogenously assembled, synthetic phage DNA.

Homologous recombination

Recombination-based approaches (Figure 1a), where the phage genome undergoes HR-driven allelic exchange with a cytoplasmic editing template during infection, are the most commonly used methods of phage genome engineering. However, with respect to strictly lytic phage, these methods are limited by low natural recombination frequencies and require extensive screening to obtain progeny phages with the desired mutations [12–14].

Figure 1.

Phage genome engineering methods. (a) Homologous recombination is often used to engineer phage genomes. In recombineering, heterologous recombination proteins are expressed within the cell, thereby increasing recombination frequency and protecting the recombination template from intracellular degradation. Both positive selection (e.g. fluorescence markers) and negative selection (e.g. CRISPR-Cas9 targeted against the wild-type phage) can be used for downstream selection of engineered phage particles. (b) Phage genomes can also be assembled from synthetic DNA fragments to introduce desired genetic alterations. Both in vivo (e.g. TAR-cloning) and in vitro (e.g. Gibson) assembly techniques are used. The fully assembled phage genome is rebooted to produce fully functional phage particles either in Gram-negative bacteria following electrotransformation, or using cell-free transcription-translation (TXTL) systems. Phage genomes of Gram-positive bacteria are rebooted in cell-wall deficient L-form bacteria after PEG-assisted transformation.

Recombination efficiency has been significantly improved by co-opting the natural recombination systems of temperate phages in a method termed recombineering. Expression of phage recombination proteins (e.g. lambda Red and Rac RecE/RecT) within the recombination host protects the editing template from degradation and facilitates annealing with the injected phage genome (in vivo recombineering), thereby increasing recombination frequency and reducing the homology arm length requirement [15]. The identification of RecE/RecT homologs in mycobacteriophage Che9c expanded these techniques to Gram-positive bacteria and enabled the bacteriophage recombineering of electroporated DNA (BRED) technique, in which the phage genome and editing templates are co-electroporated into recombineering hosts [16].

HR-based editing techniques have also been improved through the use of positive and negative selection of progeny phages. The insertion of reporter genes (e.g. luciferase or fluorescent proteins) or phage-specific marker genes (e.g. trxA or cmk in coliphages) into phage genomes facilitates rapid positive selection of recombinant phages, despite low recombination frequencies [13,17,18]. Recently, bacterial CRISPR-Cas systems have been adapted as a mechanism for the negative selection of unmodified phage progeny, effectively enriching rare mutants from recombinant phage lysates. To date, type I-E, II-A, and III-A CRISPR systems have been used to select for recombinant phages targeting Gram-negative and Gram-positive hosts [19].

Genome rebooting

To address the problem that phage gene products may be toxic to their bacterial host, synthetic methods for genome assembly outside of the natural bacterial hosts have been developed. These techniques rely on the assembly of small to medium-size DNA fragments into full-length phage genomes through transformation-associated recombination (TAR) or in vitro enzymatic assembly (Gibson assembly) followed by transformation into competent bacterial hosts for rebooting and assembly into mutant phage particles (Figure 1b). Phage genome assembly from synthetic DNA fragments enables flexible engineering to introduce mutations, deletions, or insertions at any genomic locus, scales easily for genetic library construction, and eliminates the need to select against wild-type sequences.

Jaschke et al. [20] first demonstrated the efficacy of such approaches to reconstruct and archive the ΦX174 phage genome in yeast. The small 5.4 kb ΦX174 genome could be assembled directly from PCR products and synthetic fragments into a yeast artificial chromosome through TAR, released through restriction enzyme digestion, and rebooted in Escherichia coli cells. Ando et al. [21] utilized similar approaches to modulate the host range of several T7-like phages to target various Gram-negative pathogens (see below). Because of the requirement for high transformation efficiency and access to the E. coli molecular machinery, these approaches were initially limited to phages infecting Gram-negative hosts. Recently, cell wall-deficient L-form bacteria have been developed as effective Gram-positive phage rebooting hosts following genome transformation using polyethylene glycol. Kilcher et al. [22^●] demonstrated cross-genus rebooting of various Gibson-assembled and wild-type phage genomes in L-form Listeria monocytogenes, suggesting that this approach may be broadly applicable for phages infecting Gram-positive bacteria. This technique was subsequently used to construct reporter phages [23] and to modify the host-range of a Listeria phage (see below) [24^●●].

New synthetic biology techniques to reboot phage genomes outside host cells eliminate the need for DNA transformation and enable the synthesis of phages infecting unknown or undomesticated hosts. High yields of self-assembling MS2, T4, T7, and ΦX174 phage particles have been generated in a test tube from a small quantity of phage DNA and optimized E. coli extracts, via cell-free transcription-translation (TXTL) systems [25–27]. Although genetic engineering has in the past been limited to phages infecting well-studied laboratory hosts, recent high-throughput screens for recombination protein homologs [28] and the development of TXTL systems [29] from additional bacteria is expected to expand future phage engineering to new bacterial hosts (e.g. Vibrio, Streptomyces, Bacillus, and Pseudomonas species).

Designer phage applications

The techniques outlined above have been used to create genetically modified phages for use in bacterial diagnostics, therapeutics, and drug delivery [30,31]. Below, we will highlight efforts to engineer designer phages with tunable host range and enhanced payload delivery to develop the next generation of phage therapy.

Programming host specificity

Phage host specificity is a double-edged sword: although phages can selectively infect and kill host bacteria within complex microbial communities, individual phages often lack sufficient host range to target all strains responsible for clinical infections [9]. Multi-phage cocktails can target various pathogenic strains and prevent the proliferation of phage-resistant mutants; however, the isolation and characterization of the constituent phages is time-consuming and requires laborious regulatory approval [32]. To ameliorate these limitations, synthetic biologists are now implementing design principles from the natural host range expansion mechanisms of broad-host-range phages for the scalable synthesis of modular phages, creating a tunable host range based upon previously validated phage scaffolds (Figure 2a).

Figure 2.

Designer phage applications. (a) Host specificity can be (re-)programmed by altering the phage receptor binding modules. Genetic alterations of these modules may be introduced at different scales, ranging from full gene replacements (e.g. tail or RBP swapping) to individual point mutations (e.g. amino acid exchanges in distal loops at the host-RBP interface). (b) Payload engineering refers to ‘arming’ the phage with genes and properties aimed at increasing antimicrobial activity. For example, including CRISPR-Cas9 as a genetic payload can introduce intracellular sequence-specific toxicity whereas phage-encoded biofilm depolymerases target extracellular bacterial polymers in the vicinity of the lysed host cell.

The primary host-range determinants are receptor-binding proteins (RBPs) located either at the distal end of the phage adsorption apparatus (the baseplate), or at the tip of the tail fibers, which mediate the interaction with carbohydrate or protein receptors at the host cell surface. Broad-host-range phages have evolved strategies to expand or switch host range, including: i) RBP allele and domain exchange, ii) targeted RBP diversification in hypervariable regions, and iii) polyvalent RBP virions [9].

Rational engineering of phage host range has so far been accomplished by tail fiber exchange via HR between closely related phages in the T2, T4, and T7 families [12,14,33–37]. Ando et al. expanded these techniques to a broad range of T7-like phages using a yeast-based genome rebooting strategy in order to create phage scaffolds with modular exchange of those tail components dictating the host range [21]. Whereas exchanging complete or partial T7 phage scaffold tail fibers (gp17) with those of closely related phages effectively modulated host range, the creation of viable phage hybrids between Escherichia phage T7 and Klebsiella phage K11 required the exchange of the complete tail apparatus (gp11, 12, and 17), thus also demonstrating certain limits of RBP modularity.

Most mature RBPs are involved in multiple protein-carbohydrate and protein-protein interactions with the host cell receptor as well as with structural phage tail proteins. These interactions are required to maintain structural integrity, to mediate receptor recognition, and to initiate and orchestrate receptor-mediated structural rearrangements. It is therefore not an easy task to modify RBPs without losing infectivity. Domain shuffling techniques have been developed to create chimeric RBPs by exchanging the globular receptor-binding domains located in the C-terminal domain (CTD) of phage tail fibers, while leaving the N-terminal domain required for interactions with the phage tail intact [21,24^●●,33,34].

Recently, high-throughput RBP diversification strategies have been implemented to screen for phage RBP mutants with shifted or expanded host ranges. Yehl et al. [38^●●] used a targeted mutagenesis approach inspired by antibody engineering to generate functional diversity within unstructured loops located at the RBP-host cell interface of Enterobacteria phage T3. The resulting synthetic ‘phagebody’ libraries (10⁷ variants) contained individual phages with expanded host range and suppressed the evolution of phage resistance. Yosef and co-workers integrated tail/RBP allelic exchange and iterative cycles of targeted mutagenesis to engineer the host ranges of phage particles for optimized DNA transduction and payload delivery (see below) into new bacterial targets [39^●●]. These techniques rapidly increase RBP diversity at a pace unachievable through natural evolution and help to create synthetic phage populations one step ahead of their bacterial hosts in the co-evolutionary arms race.

Dunne et al. combined RBP and lysogeny engineering to reprogram the temperate, narrow-range Listeria siphovirus PSA into a strictly lytic phage with broadened host specificity [48]. Akin to the T7 phagebody approach, the PSA host range was shifted via RBP diversification using targeted mutagenic PCR and subsequent rebooting of genome libraries in Listeria L-form cells. Additionally, a second rbp allele targeting a different Listeria serovar was inserted into the PSAgenome to generate a synthetic phage featuring polyvalent host cell binding. Finally, the elucidation of the RBP CTD crystal structure enabled the structure-guided design of chimeric RBPs. Globular, C-terminal affinity domains identified in sequenced Listeria genomes were swapped at conserved helical bundle motifs connecting the CTD to the baseplate. This approach generated viable chimeric phages, whose modified and expanded host-ranges were predicted through globular domain phylogeny and the glycotype of the donor lysogen.

It is evident that all these novel rational design principles depend on a detailed understanding of the molecular interactions dictating individual phage-host interactions. Thus, high-throughput tools to identify phage RBPs and their conserved structural domains are required to rapidly engineer modular phages at sufficient scale for therapy.

Delivering Cas nucleases as antimicrobial payloads

Besides engineering host range, phage antimicrobial activity can be enhanced or modulated through the in-situ production of heterologous proteins, often described as genetic ‘payloads’ (Figure 2b) [30]. For example, phages have been engineered to deliver biofilm-depolymerases and capsule-depolymerases [40,41], quorum-quenching enzymes [42], and cell wall hydrolases with cross-genus lytic activity [22^●]. Infected cells produce and release these proteins upon host cell lysis where they act on target cells or substrates in their vicinity. Another group of payload proteins that have recently gained much attention are CRISPR-Cas nucleases. Here, the phage serves as a target cell-specific vector to deliver a programmable Cas nuclease toxin, effectively creating a nucleotide sequence-specific antimicrobial. Two initial landmark studies describe the use of non-replicative phagemids that target antimicrobial resistance- or virulence genes in E. coli [43] and Staphylococcus aureus [44], using sequence-specific crRNA and the type II-A effector nuclease Cas9. Phagemid delivery enabled selective removal of resistant or virulent bacterial sub-populations and can be employed to immunize against plasmids containing antibiotic-resistance genes. The same pathogens were later also targeted by engineered temperate phages delivering Cas9 or Cas3 (type I-E) effectors [45,46]. Very recently, Selle et al. presented a slightly modified approach that makes use of conserved type I-B CRISPR-Cas systems present in most Clostridium difficile genomes [47^●●]. Based on the temperate phage ΦCD24–2, the authors engineered a set of phage derivatives that delivered host-targeting crRNA as a toxin and/or feature genomic deletions that convert the phage lifestyle from temperate to virulent. Both life-style conversion and crRNA-mediated genome targeting improved killing in vitro and in a C. difficile mouse infection model. Similarly, Lam et. al used engineered, Cas9-delivering M13 phagemids to selectively deplete one of two E. coli strains competitively colonizing the gut of mice [48]. These two studies highlight the potential applicability of engineered phage therapy for gastrointestinal targeting and in vivo microbiome engineering.

Outlook and conclusions

The advent of big data and the pioneering of novel machine learning (ML) methodologies constitute a major technological advancement in biological and medical research. Whereas the rational engineering efforts described above have been modeled on natural phage systems, synthetic biology approaches integrating recent advancements in ML may further boost phage therapy as a viable clinical treatment.

Research combining ‘omics with ML-powered bioinformatics is already being widely used in cancer and other biomedical fields [49]. These methods have thus far been limited in phage therapeutics, barring a handful of publications [50,51,52^●,53], an inattention most likely due to the scarcity of data available from online resources. As it stands, the most abundant dataset for any class of phage is an Actinobacteriophage collection comprising more than 3400 complete genome sequences (The Actinobacteriophage Database; URL: https://phagesdb.org). One trending approach is to utilize phage and bacteria whole-genome sequences from such databases, along with corresponding experimentally determined host-range information, to predict virus-host interactions (Figure 3a,b). While the Actinobacteriophage collection covers a sequence space of sufficient size for ML applications, the vast majority of these phages were isolated on a single bacterial species (Mycobacterium smegmatis), which drastically reduces its applicability. Generally, comprehensive and reliable host-range data is difficult to obtain as it requires labor-intensive experimentation as well as standardized protocols and scoring systems that allow comparing host range data from different laboratories. Negative interaction data (i.e. bacterial hosts not infected by a given phage) is particularly underrepresented, a limitation that could be solved only by implementing complex data augmentation methods [54] or by making unvalidated assumptions regarding phage specificity.

Figure 3.

Artificial intelligence-driven phage research. The figure outlines different applications of machine learning (ML) and artificial intelligence (AI) to phage engineering. (a) Given a novel phage sequence, a ML model can by trained on phage whole-genome sequences as features and corresponding host range information as classification labels. This approach could enable the prediction of phage-host interactions and provide insights into potential genetic elements dictating host range. (b) Similarly, by combining bacterial whole-genome sequences with information on phage infectivity, an AI algorithm could be trained to identify a selection of phages likely to infect a novel pathogen. (c) By associating bacterial surface glycans, that is, using (partial) surface glycan structures as input data, with information on phage infectivity, the model could predict phage binding to a host cell surface.

Just as sequence-based, ML-driven phage research is on the rise, so are the inferences and predictions of biological processes being made on the basis of structural data. In addition to utilizing host-range information to infer phage sensitivity based on predicted protein-protein interactions [52^●], recent advances in glycan research are similarly promising. Glycans make up a large portion of the bacterial cell surface and often function as phage attachment ligands and receptors. Two recent studies by Bojar et al. [55,56] use natural language processing to create a comprehensive database of glycan structures with associated phage-binding information readily available and infer the evolutionary relationships of the bacterial species [57,58]. This opens an interesting angle for AI-driven phage research. Analogous to using host-range information as a classification label, these types of experimental data could very well serve as structural input data to predict phage-host interaction (Figure 3c).

To fully develop the untapped potential of phage therapy, the authors firmly believe that an interdisciplinary approach is required. Besides genome engineering, synthetic biology, structure-guided design, and machine learning, this includes a plethora of other fields such as drug formulation and administration, pharmacokinetics/- dynamics, and immunology. When coupled with a non-prohibitive regulatory framework for the manufacturing and application of phage-based biologics, this integrative approach paints a carefully optimistic picture of a future with targeted phage therapeutics effective against multi-drug-resistant pathogens.