Reprogramming bacteriophage host range: Design principles and strategies for engineering receptor binding proteins

Abstract

Bacteriophages (phages) use specialized tail machinery to deliver proteins and genetic material into a bacterial cell during infection. Attached at the distal ends of their tails are receptor binding proteins (RBPs) that recognize specific molecules exposed on host bacteria surfaces. Since the therapeutic capacity of naturally occurring phages is often limited by narrow host ranges, there is significant interest in expanding their host range via directed evolution or structure-guided engineering of their RBPs. Here, we describe the design principles of different RBP engineering platforms and draw attention to the mechanisms linking RBP binding and the correct spatial and temporal attachment of the phage to the bacterial surface. A deeper understanding of these mechanisms will directly benefit future engineering of more effective phage-based therapeutics.

Graphical Abstract

Introduction

As the most abundant biological entities within Earth’s biosphere [1], bacteriophages (phages) present a morphologically diverse repertoire of infection machinery to ensure effective recognition and attachment to their bacterial hosts. The inherent ability of phages to kill specific species or individual strains has led to a resurgence of interest in the development of phages as therapeutics to tackle the growing antibiotic resistance crisis [2]. Most phages currently being explored for therapeutic and diagnostic applications are tailed dsDNA viruses belonging to the order Caudovirales that utilize a tail organelle (cauda is tail in Latin) for translocation of their genomic DNA and proteins into the host cytoplasm during infection. These phages interact with specific ligands displayed on their bacterial host cells using receptor binding proteins (RBPs) that emanate from the tail. Physical proximity to the tail allows RBPs to spatially and temporally coordinate host recognition, irreversible attachment, and genome release.

RBPs can be categorized into two classes – tail fibers (TFs) and tailspike proteins (TSPs) – depending on their morphology. TFs are long and slender fibrous proteins lacking enzymatic activity. TSPs are shorter and stockier and usually have enzymatic activity towards a particular surface structure (commonly, a sugar moiety). RBPs interact with a variety of structures displayed on the bacterial surface, such as outer membrane proteins, lipopolysaccharides, teichoic acids, capsular polysaccharides and even organelles (e.g., flagella or pili) [3]. As the first point of contact with a bacterial host, the binding range of a phage’s RBP is the primary determinant of its host range. Thus, RBPs serve as the first and most important checkpoint in the infection process.

A bacterium can quickly become resistant to phage infection by altering its surface molecules through spontaneous mutation or phenotypic variation [4,5]. Akin to the spread of antibiotic resistance, the emergence of phage-resistant subpopulations of bacteria can be a major bottleneck when using phages as therapeutics or antibacterial agents [6,7]. Despite the impressive array of counterstrategies employed by phages for altering their host range [8], such adaptation does not occur within a suitable time frame to prolong their antibacterial activity. As a result, phages are typically applied as a cocktail of multiple unrelated phages that have different host ranges and are known (or sometimes assumed) to target different surface structures, which draws selective mutation pressure away from a single receptor. Unfortunately, while phage cocktails have proven successful in various clinical cases [9,10], their formulation and production can be time and labor intensive, especially if isolating and characterizing new phages is required to target a given pathogen or if phage resistance develops. For this reason, there is significant interest in engineering the genomes of individual phages with adaptable host ranges to bypass the need for continuous modulation of phage cocktail compositions and isolation of new phages.

Over the last decade, a combination of X-ray crystallography [11-15], cryo-electron microscopy (cryo-EM) [16-25] and biochemical studies [26-31] have provided high-resolution models of various tail architectures (e.g., myo-, sipho-, or podo-viral) and revealed the atomic structures of different RBPs [32] (Figure 1). These “blueprints” are being used to guide structure-based engineering of RBPs to modulate the host range, with different strategies generally falling into two categories:

Figure 1. Minimal and evolved baseplates of two contractile tail nanomachines.

Shown are the structure of the P. aeruginosa R2 pyocin baseplate in the pre-attachment state (A) and the E. coli phage T4 baseplate in pre- and post-attachment states (B). Orthologues are shown in the same colors. A mutant used for the determination of the T4 baseplate structure contained no sheath. In the contracted state of the baseplate, the tube does not interact with it and dissociates away from it. The short tail fibers are disordered, and their electron density averages out in the cryoEM reconstruction. Figures generated using UCSF Chimera [84].

1. Domain swapping. The modular architecture of RBPs can be exploited to produce chimeric RBPs with alternative receptor binding domains to recognize different hosts [33-38]. Given the increasing number of sequenced phage genomes and high-resolution structures of diverse RBPs, it is becoming relatively easy to identify boundaries of receptor binding domains within RBPs in order to design chimeric RBPs. This modular strategy has also been applied to engineer pyocins into strain-specific antibacterial agents.

2. Structure-guided mutagenesis. Analogous to antibody engineering [39] and reverse transcriptase-mediated tropism switching [40], sequence variability can be introduced into the receptor binding sites of RBPs via targeted [41] or random [33,35] mutagenesis. This creates a library of phage mutants that feature highly diverse RBPs with distinct ligand specificities. Subsequent screening of these libraries can identify phages capable of infecting a broader range of bacteria, including strains refractory to the parental phage.

Both approaches have their own advantages and disadvantages, for instance, swapping domains (or even whole tail components) can be done with minimal understanding of the receptor binding or potential enzymatic capabilities of an RBP; however, a major bottleneck lies in identifying an alternative RBP featuring the desired binding range. On the other hand, targeted randomization does require knowledge of the binding site. However, the payoff with this approach is great, as it is possible to produce a synthetic assortment of binding residues with novel binding properties. Random mutagenesis, e.g., by using mutagenizing agents such as ethyl methanesulfonate [35], can also achieve the same; however, there can be a lack of control over the locations and frequency of mutation.

Structure and function of receptor binding proteins

A significant amount of information regarding the organization and function of RBPs can be derived solely from the analysis of their amino acid sequences. For example, the modular architecture of RBPs and the size of the modules that are exchanged between different phages are apparent [36,42]. Furthermore, it is possible to predict the host range of the phage by finding all bacterial hosts carrying prophages with sufficiently similar RBPs [26]. However, the location of the ligands on the surface of the RBP – the information required for precise engineering of the RBP’s ligand specificity – cannot be derived from the sequence and a detailed knowledge of the RBP structure (desirably, with ligand bound) is needed.

Tail fibers

The “spines” of many TFs feature a variation of a homotrimeric β-helix. Their distal tips that confer specificity, however, come in different shapes and sizes [11-15] (Figure 2). For example, the tip of the T4 long tail fiber (LTF) is a thin rod (Figure 2A) [11], whereas the T7 fiber carries a large globular head domain (Figure 2D) [12]. The T7 TF binds LPS on the E. coli surface, while the T4 LTF can interact with LPS or the OmpC porin [11,31] Some TFs (and TSPs) feature C-terminal intramolecular chaperone domains that assist with the folding of the trimer prior to specific cleavage and dissociation from the mature protein [12,14,43]. Many other TFs require a chaperone for assembly that is often encoded by a small gene immediately downstream from it [15,44].

Figure 2. A selection of structurally distinct tail fiber binding tips.

The distal ends of tail fibers are shown as ribbon representations (top) and with sequence diversity mapped onto surface representations (bottom) for E. coli phage T4 gp37 (A) [11], P. aeruginosa pyocin R2 (B) [13], Salmonella phage S16 (C) [14] and E. coli phage T7 (D) [12]. Sequence variable regions are colored cyan as indicated in the color bars provided for each panel. In panel (C), AD stands for attachment domain. E) The variable region of a human antibody (light chain, purple; heavy chain, green) is shown. Highlighted are the sequence-variable complementarity determining regions (CDRs; red) that closely resemble the distal binding loops of phage tail fibers. For panels (A)-(D), the first 250 most similar sequences found by a BLAST [85] search were aligned by COBALT [86] and mapped on the molecular surface of the protein using UCSF Chimera [84]. Panel E was generated using PyMOL Molecular Graphics System, Version 1.4 Schrödinger, LLC.

Only a handful of TF structures are known, with the majority of structural information limited to their most distal fragments. The largest segment of a TF structure is known for R-type pyocins [13] (Figure 2B).

Interestingly, not only the C-terminal lectin-like domain of the pyocin fiber, but also its middle domains display patches of high sequence diversity. This pattern is characteristic to other TFs as well. Senseless mutations accumulate in proteins over time, but such mutations should be randomly scattered on the surface of the protein. However, as seen in Fig. 2, diverse residues in LPS-binding TFs appear to form patches and the size of these surface patches is similar to that of a typical sugar binding site. For this reason, this sequence diversity is unlikely to be senseless but instead is driven by natural selection. We suspect that these mutations are selected because they are beneficial for phage function and hence can participate in LPS binding. Thus, a fiber is likely to bind the sugar moiety of the LPS in several places along its length, which both orients the fiber with respect to the cell surface and fixes the fiber to it (i.e., results in ‘irreversible’ attachment). This line of thought is supported by images of phages T4 and P1 attached to the cell surface in which the distal part of the TF is oriented perpendicular to the cell surface [23,45]. We further speculate that such a restriction of the TF’s conformational space decreases the entropy of the fiber-baseplate system, which supplies the energy needed for triggering tail contraction.

The same activation principle could likely be at work in systems featuring TSPs and can explain how a TSP could first actively digest a surface sugar and then, upon reaching the core part of the LPS, trigger a conformational change in the tail that accompanies irreversible attachment. In this case, the restriction of conformational space must be even more severe as TSPs are less mobile and less rigid than TFs, meaning their initial entropy is lower.

In some TFs, distal tips are formed by separate proteins – adhesins and tail assembly proteins – that can play critical roles in or assist with receptor binding [14,15,46,47]. Among known T4-like phages most are, in fact, T2-like and their LTFs are equipped with an adhesin. The structure of one such LTF tip – that of Salmonella phage S16 – was recently solved [14] (Figure 2C). The C-terminus of the S16 adhesin is composed of ten polyglycine type II (PG_II) helices that form a PG_II sandwich with exposed residues and distal loops that confer host binding [48]. Organizational similarities can be drawn between the terminal receptor binding domains of tail fibers and the antigen-binding sites formed by the heavy and light variable chains of antibodies (Figure 2E). The conserved framework of the variable chains of antibodies is interspersed by three complementarity-determining regions (CDRs) that closely resemble the distal loops formed by the S16 LTF adhesin and the loops present at the tips of all known TFs (Figure 2A-D).

Tailspikes

Compared to their fibrous TF counterparts, TSPs appear more rigid in their architecture, which makes them relatively easier to crystallize. A much greater number of high-resolution structures of TSPs are currently known, including those interacting with fragments of their cell surface ligands [26,27,30,49-55] (Figure 3).

Figure 3. Representative RBPs used by phages to target different cell wall ligands.

Ribbon diagrams are shown for: (A) TSP3 (orf212/gp164) of E. coli phage CBA120 that features a middle β-helical glycosidase domain that digests the O77 O-antigen of E. coli [26,59], (B) the capsule-degrading TSP (gp38) of Klebsiella phage KP32 [52], (C) the RBP (gp15) of Listeria phage PSA that recognizes serovar 4b wall teichoic acids via the head binding domain [33], and (D) the large multidomain TSP of Staphylococcus phage Φ11 that binds α- or β-N-acetylglucosamine moieties of S. aureus wall teichoic acids via the central propeller domain [87]. CBM, carbohydrate binding module. Figures produced using PyMOL Molecular Graphics System, Version 1.4 Schrödinger, LLC.

TSPs are attached to the tail or other TSPs (as part of a multicomponent network [26,56]) with help of their N-terminal domains that could be as short as a few amino acids. Such attachment domains comprise a smaller, more conserved set than ligand-binding (sometimes called catalytic) modules that form the rest of the TSP structure. These attachment domains are “reused” across diverse families of phages alongside their interacting partners, e.g., T4 gp10-like docking domains are found in different phage families that contain multiple TSPs [26,27,57]. The rest of the TSP structure (the ligand-binding “module”) is also reused as it is linked to recognition and enzymatic processing of a sugar displayed on the surface of a particular bacterial host.

The ligand-binding module usually consists of at least two domains, one of which typically exhibits either lyase, hydrolase, or esterase activity (e.g., G7C [27]) towards cell surface sugar moieties such as lipo- and capsular polysaccharides (O- and K-antigens, respectively) or teichoic acids (Gram-positive hosts) (Figure 3A&B). Most TSPs disassemble their polymeric sugar substrate into short fragments thus creating a path for the phage particle to reach the cell membrane [26,49,54,55,58,59]. However, TSPs with esterase activity remove the small O-acetyl group from specific sugar residues and leave the polymeric chain intact, which raises a question of how the phage particle equipped with such a TSP reaches the cell surface [27]. Unlike TFs, where their least conserved regions are likely responsible for ligand binding (Figure 2), the active site in catalytically active TSPs constitutes their most conserved part. The enzymatic domain is therefore the key determinant of TSP specificity and, as a consequence, of the phage’s host range. The other domain(s) of the ligand-binding module play(s) a role in substrate binding and, therefore, host recognition, but the function of these domains in the initial host recognition and irreversible attachment is unknown. Many TSPs feature a lectin-like domain at the C-terminus that is most proximal to the cell membrane during infection.

We can assume with sufficient confidence that TSPs represent bona fide TFs with enzymatic activity that emerged as a part of the arms race between phages and bacteria after the latter evolved protective extracellular layers that are difficult for fibers to penetrate [60,61]. Most likely, phages captured non-critical catabolic enzymes (e.g., sialidases and pectate lyases) from host bacteria and incorporated them into preexisting RBPs, which allowed phages to “drill” through different outer bacterial polysaccharidic layers [62]. Nevertheless, such acquisition places serious constrains on the host range of the phage as all known TSPs typically recognize one substrate or at best a few very closely related ones [26,27,58]. Clearly, efficient receptor recognition and processing require a precise combination of affinity and optimal receptor processing kinetics, which limits the number of possible substrates [63,64]. TSP-carrying phages with wider host ranges have complex adsorption devices that contain several types of TSPs, each responsible for recognizing a certain host [65,66].

An important concept for the function of TF-like RBPs (i.e., TSPs lacking enzymatic activity) (Figure 3 C&D) is avidity as exemplified by the composition of host cell adsorption organelles of certain phages. To overcome the weak binding of an individual RBP, Lactococcus phage TP901-1 [67], Listeria phage A511 [19], Bordetella phage BPP-1[25], and many others, feature a staggering number of identical RBPs emanating from their tails [68]. To further improve the probability of interacting with a host many phages carry carbohydrate binding modules (CBMs) on various components of the phage particle, for instance, protruding from “evolved” distal tail proteins (Dits), neck passage proteins (NPSs) or major tail proteins (MTPs) [69]. These CBMs appear to have similar ligand specificities as the phage RBPs and should be careful considered when attempting to engineer the host range of such a phage.

Current RBP engineering strategies

Non-targeted recombination of phage genomes

Random mutation and recombination between phages is a long-used but mostly unreported approach to adapt phage host ranges [70-72]. For example, the so-called “Appelmans protocol” [70] involves cycling a cocktail of phages with a group of susceptible and resistant bacteria until a recombinant phage appears that lyses the resistant strains. This approach was recently used to generate a phage capable of infecting ten P. aeruginosa strains starting with two parental phages infecting only one or two strains [70]. This procedure contained over 30 rounds of co-infection in which a minimum of 48 recombination events between the parental phages and one point mutation took place. A large fraction of recombination events occurred within structural and RBP genes.

Trading tail fibers by directed homologous recombination

A more direct approach to modify host range is to only recombine the RBP gene(s) of an infecting phage or electroporated phage genome with a cytosolic RBP template featuring the desired host range. The limitation of this approach is the two RBP genes must feature high sequence similarity to allow recombination. This approach has been used to expand the host range of T4- and T2-like phages [38], for instance by swapping T2 LTF genes 37 and 38 with counterparts from phages PP01 or IP008 to expand T2 infectivity towards different E. coli strains [37,73]. To increase the frequency of recombinant phage identification, a CRISPR/Cas counterselection step can be included to remove any wildtype phages remaining after recombination [74-78]. Such a two-step approach has allowed replacing both short (gp12) and long tail fibers of T2 with those of PP01. The resulting T2 phage had improved adsorption to E. coli O157:H7 hosts, especially when compared to wildtype T2 or when only the LTFs had been switched with PP01 [79].

Complete replacement of tail components

Ando et al. developed a yeast-based platform to produce synthetic phage genomes assembled from individual fragments [34]. This approach allows the exchange of RBP and tail components between different phage backbones. The utility of the platform was demonstrated by swapping whole TFs or bioengineering chimeric TFs of different T7/T3-like phages with modified host ranges. A synthetic T3 phage carrying Yersinia phage TFs was able to cross the genus barrier and infect Y. pseudotuberculosis and E. coli strains. The advantage of this approach is the ability to synthesize phages with structurally different host recognition machinery by exchanging multiple components of the phage tail apparatus all at once regardless of their genomic position. For example, the tail proteins gp11 and gp12 and the TF protein gp17 of T7 can be replaced with their counterparts from the Klebsiella phage K11, which is equipped with a capsule-degrading TSP instead of a TF. The resulting synthetic T7_{K11(gp11-12-17)} phage infected Klebsiella. Similar to recombination-based engineering, the major drawback of this approach is that the replacement of phage components still requires sufficient sequence similarity at chosen domain boundaries to ensure they can be attached to the engineered phage.

An alternative platform for producing synthetic Gram-positive targeting phages was developed by Kilcher et al. for rebooting in vitro-assembled synthetic genomes in Listeria L-forms [80]. This platform was recently used to re-engineer the temperate Listeria monocytogenes phage PSA with a broader host range [33]. Error-prone PCR was used to produce a library of randomized genomic fragments encoding the phage RBP (gp15), which were assembled into synthetic phage genomes. Rebooted phages carrying different RBP mutations presented a shift in host range from serovar 4b to 4d strains, which removed the need for galactose as an essential binding element of the teichoic acid receptor. Secondly, a polyvalent phage encoding both wildtype and mutated (S334R) RBPs demonstrated infectivity against both serovars. Finally, a structure-guided approach was used to generate chimeras, whereby heterologous RBPs of low sequence similarity identified using BLAST within genomes of Listeria lysogens were fused at conserved domain boundaries (e.g., “neck chimeras” were fused after the α3 coiled coil (Figure 3C) to produce phages capable of infecting a broader selection of species and serovars (4a, 4b, 4d, 5 and 6b).

Directed mutagenesis of tail fiber binding loops

As sequence composition of the distal loops of TFs is directly responsible for binding specificity [48] it is not surprising that randomization of these loops can generate phages with new host ranges. The crystal structure of the phage T7 TF (Figure 2D) was used as a blueprint by Yehl et al. to identify four exposed distal loops for targeted randomization of related phage T3 [41]. Recombination plasmids featuring the TF gene of T3 were generated using site-directed mutagenesis to replace codons within the loops with NNK codons, such that loop sequences were completely randomized at the DNA level. Upon infection, T3 recombined with the randomized TF to produce phages with unique loop compositions. The mutant phage libraries were capable of infecting T3-resistant E. coli and provided long-term suppression of E. coli resistance development when tested in vitro and in a murine model [41]. It is important to note that not all loop mutations are functional, or provide a benefit compared to the parental phage. However, as sequence variability can be concentrated to specific regions (i.e., TF binding loops), the probability of generating a mutant phage with the desired binding is considerably greater than relying on natural phage evolution. For example, the smallest loop randomized consisted of only four amino acids, yet this provided a possible 10⁶ unique sequences within the phage library. The current limitation of this approach is our lack of understanding of the mechanism of cell surface receptor binding by TFs because not a single ligand-bound high-resolution structure of a TF is available.

Yosef et al. [35] recently demonstrated the expansion of host range through similar randomization of tail and TF genes. In their GOTrap (general optimization of transducing particles) platform, a T7 mutant lacking tail (genes 11 and 12) and TF genes infected E. coli hosts that carried randomized tail-TF genes on a plasmid containing an antibiotic resistance marker and a T7 packaging sequence. After infection, only phage particles that carried functional tail and TF complexes and contained a plasmid encoding for these proteins could successfully transduce a novel host that could be selected using the antibiotic marker (e.g., Klebsiella, Enterobacter, Salmonella etc.). Successive rounds of plasmid purification and mutation using ethyl methanesulfonate were used to select for tail and TF mutants with improved binding ranges [35].

Pyocin engineering

Phage tail-like bacteriocins, such as the R-type pyocins produced by P. aeruginosa, employ a P2 tail-like contractile machinery to kill competing bacteria by dissipating their membrane potential [16,81]. However, unlike phages that propagate during infection, pyocins can be used only as “single-shot” precision antimicrobials. While the wildtype pyocins have very narrow killing spectra, many of the engineering strategies developed for phages can be used to expand their therapeutic potential. For example, the Pseudomonas aeruginosa pyocin pyR2 (i.e., R2-type pyocin) could be retargeted to kill uropathogenic E. coli or Y. pestis strains by replacing most of its cognate TFs with those from E. coli phage P2 or Yersinia phage L-413c, respectively [36] (Figure 2B). Using the same strategy, the pyR2 pyocin was further developed into an O157-specific antimicrobial by fusing an O157-specific TSP from podovirus φV10 to the native fiber [82]. This hybrid pyocin could prevent and ameliorate E. coli-induced diarrhea and intestinal inflammation when tested in a rabbit model [83]. It is interesting to note that in the absence of structural “blueprints” of the fiber and TSP components, libraries of chimeric fibers had to be constructed and tested, which was a time-consuming task [36,82]. In addition to the structure of R1- and R2-type pyocin fibers described above, structures of the entire R2 pyocin particle were recently reported in pre- and post-contracted states providing intricate details of its internal mechanics [16]. This high-resolution model has already been used to engineer certain interface residues within the baseplate to produce an acid-stable pyocin, demonstrating the significance of high-resolution structures for future engineering of pyocins and phages.

Conclusion

Our current understanding of how a phage recognizes its host and attaches to its surface is far from complete. For example, cell envelope components that are responsible for “uncorking” the phage particle are unknown for most podo- and myo-phages that have already been studied for decades. Nevertheless, the necessary step in the infection process – the interaction of the protein that defines the specificity of the phage, its RBP – is much better understood. The work with the R2 pyocin platform demonstrates that both TSPs (with enzymatic activity) and TFs (sans enzymatic activity), can place the pyocin particle onto the path of irreversible attachment. This shows that the functions of TSPs and TFs are fundamentally similar. Not only are they responsible for the specificity, but they are capable of triggering a conformational change in the particle that is required for irreversible attachment. Together with the enormous progress in structural studies of bacterial surface polysaccharides and genetics of their biosynthetic pathways, it will soon be possible to identify in the existing pool or generate synthetically a comprehensive set of RBPs that will target all known cell surface sugars. Such RBPs can then be used in a phage- or pyocin-like mono- or multi-valent platform to provide more effective phage-based therapeutics.

Highlights.

• Phages use structurally diverse receptor binding proteins (RBPs) to target different cell wall structures.

• RBPs play a critical role in spatial and temporal positioning of phages on the bacterial surface to ensure correct attachment of the tail apparatus.

• High-resolution structures of RBPs are essential for understanding the mechanism of receptor recognition and provide “blueprints” to guide phage engineering.