Completing the BASEL phage collection to unlock hidden diversity for systematic exploration of phage–host interactions

Overview: Including O-antigen-dependent phages in the BASEL collection

The cell surface of Gram-negative bacteria is dominated by LPS and proteins embedded in the outer membrane on which additional layers of protective exopolysaccharides such as a capsule can be present [32]. To recognize their host and pass through these layers, phages usually directly bind exposed glycans as an initial host receptor before moving down to the cell surface [10,31,33,34]. There, the virus typically binds a second so-called terminal receptor, which results in irreversible adsorption and DNA ejection [34]. The receptor-binding proteins (RBPs) that target primary and terminal receptors are typically displayed on different structures with longer, lateral tail fibers or stubby tailspikes commonly targeting primary receptors followed by central tail fibers or tail tip complexes binding terminal receptors [reviewed in reference 34].

Bacterial exopolysaccharides and the dense O-antigen layer of the LPS effectively shield the cell surface from the access of tailed phages unless the phage carries enzymatic tailspikes or glycan-targeting tail fibers to specifically bind them as part of primary receptor recognition [28,29,31,33,34] (Fig 1A). Consequently, the variability of the ca. 180 different types of O-antigen and other surface structures greatly contribute to the observed narrow host range of most enterobacterial phages [10,30,35,36]. This diversity is further increased by accessory O-antigen modification systems that can alter the polysaccharide chains, e.g., by acetylation or glycosylation [36]. However, laboratory strains of the E. coli K-12 lineage do not express protective exopolysaccharides such as capsules and lack even the O-antigen layer due to disruption of the wbbL gene by a transposon [27] (Fig 1A). Phages can therefore readily access the bacterial cell surface, which enabled broad sampling of E. coli phages for the original BASEL collection [10,28]. However, most of these phages are unable to infect E. coli K-12 with restored O16-type O-antigen with few exceptions that likely carry RBPs targeting this glycan such as Hanrivervirus JakobBernoulli (Bas07) or Tequintavirus IrisvonRoten (Bas32; Fig 1A and 1B). In addition, some phages like N4 or the Vequintavirinae can bypass the O-antigen layer by targeting the “N4 glycan receptor” (NGR), a recently described glycan that is broadly conserved among enterobacteria [10,17] (Fig 1A and 1B). The shielding of surface receptors by the O-antigen layer seems to be primarily due to its density because it is not more than 30 nm deep, while the tail fibers of model phages T4 (ca. 145 nm [37]), T5 (80 nm [38]), or T7 (30 nm [39]) could otherwise easily reach through it to the cell surface. However, these phages are completely blocked by restored O16-type O-antigen expression [10] (Fig 1B).

Fig 1. O-antigen glycans as a barrier or host receptor for phage infections.

(A) The illustration shows cell surface glycans of E. coli K-12 laboratory strains, including rough lipopolysaccharide (LPS) without O-antigen and smooth LPS (with genetically restored O16-type O-antigen; see details in Materials and methods). The O-antigen can be a barrier to phage infection (green), used as a host receptor (blue), or be bypassed by phages that target the N4 glycan receptor (NGR) chains (brown). (B) Serial dilutions of different phages characterized previously [10] were spotted on the E. coli K-12 ΔRM strain without (left) or with (right) genetically restored O-antigen expression to illustrate the three different modes of interaction with these surface glycans (same color code as in panel A).

https://doi.org/10.1371/journal.pbio.3003063.g001

While the absence of an O-antigen barrier in the K-12 strain used for sampling the BASEL phages made it much easier to isolate many diverse phages, it also meant that phages that depend on this glycan as part of host receptor recognition could not be sampled. Using different sewage samples, we had previously shown that these are quantitatively rare, though they would of course be qualitatively interesting to expand the assorted diversity of the BASEL collection [10]. We have therefore now systematically isolated and characterized diverse phages that infect E. coli K-12 with O16-type O-antigen but not the isogenic strain without this glycan (Fig 2A and 2B). Fortuitously, we observed that phages depending on O16-type O-antigen as host receptor are much more common in environmental samples than in sewage (S1 Table). Most, but not all of these viruses belonged to phage groups that had already been known for their dependence on surface glycans, including host O-antigen such as podoviruses of Gamaleyavirus genus or siphoviruses of the Kagunavirus genus [33] (see Fig 2B).

Fig 2. Expanding the BASEL collection with phages dependent on O-antigen or inhibited by K-12 cryptic prophages.

(A) Phages dependent on O16-type O-antigen were isolated and characterized as shown in the illustration (analogous to our previous work [10]). (B) O-antigen dependence of diverse newly isolated phages characterized in this work compared to JakobBernoulli and IrisVonRoten from our previous study that can infect E. coli K-12 without (left) and with (right) O-antigen [10]. (C) Phages inhibited by the cryptic prophages in the K-12 genome were isolated analogously to those dependent on O-antigen (see panel A) with help of the E. coli K-12 Δ9CP strain (details outlined in Materials and Methods).

https://doi.org/10.1371/journal.pbio.3003063.g002

Overview: Including phages inhibited by resident immunity of E. coli K-12 in the BASEL collection

Bacteriophages encounter multiple layers of antiviral defenses outside and inside their host cells that must be overcome for successful viral replication [40]. Some of these defenses prevent phage adsorption or DNA ejection, while others directly attack the invading virus such as restriction-modification (RM) systems, which cleave foreign DNA based on self/ non-self recognition through certain DNA modifications. Beyond direct attack, a deeper layer of defense systems protects the bacterial population through abortive infection, i.e., the altruistic suicide of infected cells prior to successful phage replication [41]. Recent work on anti-phage defense systems has vastly expanded our knowledge about their abundance and diversity through the discovery of many new systems that target various aspects of phage biology and are countered by phages in diverse ways [40–45].

The phages of the original BASEL collection had been sampled with a derivative of the commonly used E. coli K-12 MG1,655 laboratory strain in which we had knocked out all known RM systems, i.e., the EcoKI type I RM system and three type IV restriction systems [10]. This enabled broad sampling of E. coli phages, including several taxonomic groups of phages that displayed considerable sensitivity to EcoKI and would otherwise have been excluded [10]. However, the literature and commonly used prediction tools such as DefenseFinder and PADLOC indicate that E. coli K-12 encodes several additional bona fide defense systems that might limit the range of phages accessible with this genomic background [46,47] (see details at Bacterial strains in Materials and methods). We had not considered this to be a widely relevant factor when composing the original BASEL collection because phage defenses beyond RM systems were thought to be rather specific. However, our own results showing broad antiviral defense by the Fun/Z system and the work of others, e.g., on the Zorya system revealed that this is not necessarily the case [10,24].

Notably, all these bona fide defense systems are encoded on the cryptic prophages of E. coli K-12 that are genomic islands descended from once active prophages, i.e., the facultatively host-integrated form of so-called temperate phages [46,47]. This is in line with the common observation that anti-phage defenses are frequently encoded on mobile elements and genomic islands [48–50]. To overcome possible sampling biases caused by these bona fide phage-defense systems (and possibly others that are yet unknown), we thus performed phage isolation experiments with a variant of E. coli K-12 in which a previous study had removed all cryptic prophages, E. coli K-12 BW25113 Δ9CP (both with and without restored O-antigen) [51] (Fig 2C). Similar to previous work in Vibrio that had used an analogous approach [50], these experiments indeed resulted in the isolation of an additional diversity of phages both from groups already covered in the BASEL collection and from a genus that had never been isolated before using any E. coli laboratory strain (Wifcevirus myoviruses).

Composition of the expanded BASEL phage collection

The original BASEL phage collection had been composed of 68 new handpicked phage isolates that we characterized alongside a panel of 10 well-studied model phages such as the classic T-phages [10]. These phages represent many major groups of E. coli phages, which have been previously described in phage isolation studies with E. coli K-12 [52,53]. The (mostly genus-level) groups comprising these phages showed characteristic and specific patterns of features regarding receptor usage, host range/ recognition, and sensitivity or resistance to bacterial immunity in our data and those from other studies [10,20–25]. These observations make it possible to learn about the properties of newly isolated phages merely from their genome sequences or taxonomic position, e.g., to inform a targeted selection or engineering of effective phages for individual cases of phage therapy [54–56].

However, a close look at published work describing phages isolated with E. coli hosts beyond the K-12 laboratory strain reveals important gaps in the assorted diversity of the original BASEL collection. These include, e.g., several groups of podoviruses such as the Gordonclarkvirinae with characteristic prolate virions [57], diverse small siphoviruses such as the genus Kagunavirus, and a few myoviruses such as the genus Wifcevirus [35,58,59]. Using E. coli K-12 with restored O-antigen expression and/ or without cryptic prophages, we have now isolated and characterized more than a hundred new phages that are unable to grow on the regular E. coli K-12 host used for the original BASEL collection. These new isolates cover all major groups of phages (and some rarely isolated ones) that have been reported to infect E. coli, and we selected 37 isolates for an expansion of the BASEL collection based on their genomic and phenotypic diversity (Fig 3 and S2 Table). We systematically characterized these phages analogous to the phages in our previous study (Fig 4A–4D) and used several original BASEL phages that could overcome the O16 O-antigen barrier as controls for the phenotypic experiments. For these, no relevant differences to their phenotypes in the original paper were observed when the host strain expresses O16 O-antigen (S1B Fig). These results suggest that the O-antigen barrier and other defenses are largely orthogonal and that our results from this study are comparable to the original publication on the BASEL collection.

Fig 3. Overview of new phages included in the BASEL collection.

The illustration shows new phage isolates included in the BASEL collection with their Bas## identifiers. Isolates are sorted by morphotype and their current ICTV classification [68]. The color code distinguishes phages based on morphotype and genome size. For siphoviruses, small phages (ca. 30–50 kb) are shown in red and medium-sized phages (ca. 50–100 kb) are shown in orange. Small myoviruses (ca. 30–100 kb) are shown in green. For podoviruses, small phages (ca. 30–50 kb) are shown in violet and medium-sized phages (ca. 50–100kb) are shown in pink.

https://doi.org/10.1371/journal.pbio.3003063.g003

Fig 4. Overview of E. coli surface glycan variants and the immunity systems used in this study.

(A) Surface glycans of different E. coli K-12 variants are shown schematically (compare Fig 1A and see further details in Materials and Methods). None of the newly isolated phages in this work showed any dependence on the NGR glycan (see S1A Fig) which has thus been omitted from the other illustrations. (B, C) Restriction-modification (B) and other defense systems (C) used to characterize phage isolates are summarized schematically (same as used in our previous work [10]). (D) The genome of E. coli K-12 contains nine cryptic prophages that encode all remaining antiviral defense systems predicted by DefenseFinder or PADLOC after deletion of all restriction systems in our previous work (including McrA; see details on these defense systems in Materials and Methods under Bacterial strains) [10,46,47]. Due to the known accumulation of antiviral defenses in prophages [48,49], it seems likely that also many of the possibly still unknown defense systems of E. coli K-12 would be encoded in these loci (indicated by question marks). The E. coli K-12 Δ9CP variant of Wang and colleagues [51], which we use in the current study has been cured of all nine cryptic prophages.

https://doi.org/10.1371/journal.pbio.3003063.g004

Properties of siphoviral genus Kagunavirus

Small siphoviruses like the Drexlerviridae family (relatives of phage T1) or genus Dhillonvirus are frequently isolated using E. coli as a host and typically have genomes of 40−50kb size. The diverse representatives of these groups in the original BASEL collection mostly showed considerable sensitivity to RM systems and—as expected from previous work—targeted different porins as terminal host receptors [10,60,61]. Only one of these phages, JakobBernoulli (Bas07), is able to infect E. coli K-12 with restored O-antigen expression, suggesting that its lateral tail fibers can target O16-type O-glycans [10] (see also S1A Fig and below). Our targeted search for phages that depend on O16-type O-antigen resulted in the isolation of various small siphoviruses of genus Kagunavirus within the subfamily of Guernseyvirinae (Fig 5A–5D). These phages have small genomes of ca. 41−44kb, which makes them attractive models for synthetic biology applications, e.g., in the case of phage CM001 [62]. While all our isolates fully depend on restored O-antigen expression, only some of them (Bas70−73) infect a regular E. coli K-12 host while others (Bas74−76) could only be isolated using a K-12 host devoid of cryptic prophages (Fig 5D). Phenotypically, both groups of Kagunavirus isolates behave very similarly and display a high sensitivity to RM systems and a variable sensitivity to the Old antiphage defense system that is typical for small siphoviruses [10] (Fig 5D). The dependence on glycans as host receptor seems to be a general property of Guernseyvirinae phages, including, e.g., the prototypic Kagunavirus K1G targeting the K1 capsule of E. coli [15] or the Jerseyvirus SETP3 and several relatives targeting O-antigen of Salmonella enterica [63].

Fig 5. Overview of Guernseyvirinae genus Kagunavirus phages.

(A) Schematic model of host recognition by Kagunavirus phages (O-antigen chains in blue). In analogy to other siphoviruses infecting E. coli like lambda or T5, we anticipate that these phages also use a phage-derived conduit for DNA injection across the outer membrane after the tail fibers recognizing the terminal host receptor have bent off [110,141]. (B) Representative TEM micrograph of phage SonjaBuckley (Bas74). (C) Maximum-likelihood phylogeny of Kagunavirus phages and relatives based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. Newly isolated phages of the BASEL collection are highlighted by red phage icons. The phylogeny was rooted between the genera Kagunavirus and Jerseyvirus. (D) The results of quantitative phenotyping experiments with Kagunavirus phages regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and methods). Data points and error bars represent the average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g005

Properties of other O16-dependent small siphoviruses and Dhillonvirus

Besides phages of the genus Kagunavirus, we also obtained different other small siphoviruses that depend on restored O16-type O-antigen to infect E. coli K-12 and new isolates of the Dhillonvirus genus (Fig 6A–6G). Some of the new O16-dependent isolates belong to a cluster of siphoviruses formed primarily by the genus Skatevirus (Bas77–81, Fig 6C and 6D). These viruses have genomes of 46–48 kb, which is well within the typical range for small siphoviruses. Additionally, we present phage KarinMuster (Bas82), belonging to the rarely reported Nonanavirus genus comprising relatives of the more commonly known Salmonella phage 9NA (Fig 6E) [64]. The genome of this phage is ca. 58.5 kb and thus around 10 kb larger than that of the other small siphoviruses presented here, reminiscent of the Queuovirinae characterized in our previous work [10]. This latter group encodes a 7-deazaguanosine modification pathway in their additional genome content which provides considerable resistance to RM systems [10,65]. However, there is no recognizable additional locus in KarinMuster and its relatives that could encode such supplemental biological functions [64], though their distant relationship to other known phages and limited knowledge about their biology makes it difficult to draw firm conclusions. Very similar to small siphoviruses characterized previously and the Kagunavirus phages shown above, all these phages are markedly sensitive to RM systems and to the Old defense system, which in the latter case results in quantifiable decrease of plaque counts only for some phages (Fig 6G). The additional genome size of KarinMuster therefore does not seem to make this phage more robust than other small siphoviruses when facing the tested examples of bacterial immunity.

Fig 6. Overview of other O16-dependent siphoviruses of genera Skatevirus and Nonanavirus as well as new isolates of siphoviral genus Dhillonvirus.

(A) Schematic model of host recognition by O16-dependent small siphoviruses just as shown for Kagunavirus phages in Fig 5A. (B) Schematic model of host recognition by small siphoviruses of the Dhillonvirus genus that use first a lateral tail fiber to overcome the O-antigen barrier before binding a porin receptor with the central tail fiber (instead of only glycan-targeting tailspikes for the LPS-dependent small siphoviruses; see also our previous work [10]). (C) Representative TEM micrograph of Skatevirus ClaudiaMuster (Bas77). (D) Maximum-likelihood phylogeny of the Skatevirus genus and relatives based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. The phylogeny was rooted between genus Roufvirus and the others. Note that most reference phages presented here as Skatevirus have no formal ICTV classification while others are scattered over several seemingly paraphyletic genera, prompting us to show all phages as Skatevirus [68]. (E) Maximum-likelihood phylogeny of the Nonanavirus genus and relatives based on a curated whole-genome alignment with bootstrap support of branches shown if >70/100. The phylogeny was rooted between phage 9NA and the other relatives based on their much closer relationship. Note that Sasha and Sergei are formally classified as Sashavirus while 9NA and AVIO78A are classified as Nonanavirus [68], which—according to this phylogeny—would be paraphyletic. (F) Maximum-likelihood phylogeny of the Dhillonvirus genus based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. The phylogeny was rooted as previously [10] between the group of isolates, including phage WFI and the others. In (D-F), newly isolated phages of the BASEL collection are highlighted by red or orange phage icons. The color code of phage names in (F) refers to porin host receptors with blue indicating FhuA, red LptD, green BtuB, and orange LamB [10] (see also S2B Fig). (G) The results of quantitative phenotyping experiments with Skatevirus, Nonanavirus, and new Dhillonvirus isolates regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and METHODS). Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g006

Finally, the phage isolation using a host without prophages identified additional members of the Dhillonvirus genus (Bas83−85), which—unlike their relatives of the original BASEL collection (Bas14−18)—fail to form plaques on regular E. coli K-12 (Fig 6B, 6F, and 6G). These new isolates do not cluster separately within the Dhillonvirus genus but are rather split widely across this group, suggesting that no vertically inherited feature but rather highly variable aspects of phage biology are responsible for this difference (Fig 6F). Phenotypically, the three additional phages selected for this work behave mostly similar to the original Dhillonvirus phages, e.g., by showing only moderate sensitivity to tested RM systems [10] (Fig 6G). Like the other siphoviruses characterized for the original BASEL collection, these Dhillonvirus phages display the typical arrangement of two separate types of tail fibers with distinct receptor-binding proteins [10,34] (see below and S2A Fig). Their central tail fiber has a bona fide target recognition module that is predicted to encode LptD specificity (S2B Fig) [10].

Properties of podoviruses of Gordonclarkvirinae genus Kuravirus and relatives

Besides small siphoviruses, our phage isolation with E. coli K-12 expressing its O16-type O-antigen resulted in multiple phages belonging to podoviruses in the Gordonclarkvirinae subfamily (Fig 7A–7D). These phages form characteristic prolate virions of the rare C3 morphology [66] (Fig 7B) and have been frequently isolated using diverse strains of E. coli. However, so far, no representatives infecting the E. coli K-12 or B laboratory strains have been obtained, and the molecular biology of these phages has been only poorly studied. The genome sizes of this group differ widely from 67−69kb for one clade of the Kuravirus genus (at the bottom of Fig 7C) to 76-77kb for other Kuravirus phages and genera Suseptimavirus as well as Nieuwekanaalvirus (see details for all in S2 Table). Our phylogeny largely reproduces the previously identified subgroups of Gordonclarkvirinae, which have in the meantime been recognized as different genera in the ICTV taxonomy [67,68] (Fig 7C). The isolates infecting E. coli K-12 are widely scattered over the different genera of Gordonclarkvirinae, suggesting that no vertically inherited factor plays major roles in the ability of these phages to infect that host (Fig 7C).

Fig 7. Overview of Gordonclarkvirinae phages.

(A) Schematic model of host recognition by Gordonclarkvirinae phages as an interpretation of different molecular studies [57,117]. (B) Representative TEM micrograph of phage LorandFenyves (Bas91). (C) Maximum-likelihood phylogeny of the Gordonclarkvirinae subfamily based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. Phages of the BASEL collection are highlighted by little violet phage icons. The phylogeny was rooted between the genus Suseptimavirus and the genera Nieuwekanaalvirus + Kuravirus based on closer relationship of the latter two. (D) The results of quantitative phenotyping experiments with Gordonclarkvirinae phages regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and Methods). Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g007

All our isolates completely depend on the restored expression of O16-type O-antigen for infection, suggesting that the O-antigen and possibly other surface glycans are the conserved host receptors of these phages (Fig 7D). This is in line with previous work reporting that spontaneous E. coli mutants resistant to Kuravirus phage ES17 had always lost O-antigen expression [69]. In one of the few dedicated studies on Gordonclarkvirinae, the Plevka group reported the virion structure of Kuravirus SU10, which—in the absence of decades-long research available for standard model phages—was critical for the functional understanding and annotation of structural genes in Kuravirus genomes [57]. Unlike O-antigen-dependent siphoviruses (see above) and other well-studied podoviruses such as T7 [70,71], the Gordonclarkvirinae have two separate bona fide RBPs, identifiable as long-tail fibers and short-tail fibers, that likely recognize different host receptors. These two sets of tail fibers appear to be used successively analogous to classical T-even myoviruses where the long-tail fibers mediate initial host recognition by targeting porins and O-antigen while the short-tail fibers subsequently bind conserved motifs in the LPS core as the terminal receptor [10,37,57,72–74] (see also below). The functionality of long and short-tail fibers of T-even phages is linked in a way that host recognition by long-tail fibers is required for the deployment of short-tail fibers and, consequently, viral genome injection [10,73,74]. Similarly, direct observation of Kuravirus SU10 infection using cryogenic electron tomography shows how the lateral tail fibers initially target an exposed receptor—likely O-antigen glycans—before the short-tail fibers target the cell surface and contribute to the formation of a nozzle which extends the virion tail for genome delivery [57] (see Fig 7A).

The Gordonclarkvirinae are remarkably resistant to RM systems despite the lack of known pathways of covalent genome modification or restriction inhibitors (Fig 7D and S2 Table). However, they encode a bona fide dCTP deaminase and a nucleotidyltransferase with their DNA replication genes (S3 Fig), suggesting that these phages might feature a covalent DNA modification—most simply deoxyuracil instead of thymidine like Bacillus phage PBS1—which may protect their genome from restriction [75]. The only detectable sensitivity to tested anti-phage defense systems was observed for Fun/Z which targets phages of the Kuravirus genus (Fig 7D).

Properties of other groups of podoviruses and myoviruses

Xuquatrovirus relative HelliStehle (Bas95).

Besides small siphoviruses and Gordonclarkvirinae podoviruses, all other phage groups isolated for this work were individually much rarer, and only one or a few isolates of each could be obtained. This includes two podoviruses and one myovirus that are each the only isolates of their respective groups (Fig 8A–8G). One of these is a podovirus related to the Xuquatrovirus genus for which there are only two closely related representatives, the original PTXU04 of the study by Korf and colleagues [58] and phage Pondi of Sidi Mabrouk and colleagues [76] (Fig 8D). These phages have an intermediate genome size for a podovirus of around 60 kb, which is ca. 20 kb larger than Autographiviridae (T7 relatives; see below) but ca. 15–20 kb smaller than Gordonclarkvirinae (see above). Our new isolate HelliStehle (Bas95) is clearly related to the Xuqutatovirus phages with a largely colinear genome of similar size (S4 Fig) but is phylogenetically distinct (Fig 8D). The biology of these phages has largely remained enigmatic besides a genomic and proteomic analysis of PTXU04 when it had originally been described, which was crucial for a functional annotation of at least the structural genes due to low sequence similarity to well-characterized phages [58]. Consequently, a recent transcriptome-wide CRISPR interference study determined the essential genes of PTXU04 but could in the absence of additional phenotypic data not use this information for deeper insights as has been possible, e.g., with model phages T4 and T5 [77]. Our experiments show that phage HelliStehle is completely dependent on restored O-antigen expression, indicating that it is using this glycan as an essential receptor (Fig 8G). Furthermore, this phage is very sensitive to RM systems, suggesting that—unlike other larger podoviruses such as N4 or Gordonclarkvirinae—it lacks potent mechanisms to protect its genomic DNA (Fig 8G).

Fig 8. Overview of rare O-antigen-dependent phage groups (Xuquatrovirus relatives, Gamaleyavirus, Jilinvirus).

(A) Schematic model of host recognition by podoviruses with enzymatic tailspikes targeting O-antigen (e.g., Gamaleyavirus) based on previous work on phage T7 and P22 [71,101,203]. (B) Schematic model of host recognition by myoviruses targeting O-antigen with a tail fiber (e.g., Jilinvirus). Note that Jilinvirus HeidiAbel encodes two sets of tail fibers (see below) whose relationship is unknown, so only a simple schematic with one set is shown. (C) Representative TEM micrograph of Jilinvirus phage HeidiAbel (Bas97). This phage forms myovirus virions with similar appearance to other Jilinvirus phages and relatives that have been studied previously [79,80,82]. (D) Maximum-likelihood phylogeny of the genus Xuquatrovirus and relatives based on a curated concatenated alignment of major capsid protein, portal protein, and terminase large subunit protein sequences with bootstrap support of branches shown if >70/100. The phylogeny was rooted between two marine Vibrio phages and the others based on a large phylogenetic distance between these groups. (E) Maximum-likelihood phylogeny of the Enquatrovirinae subfamily of Schitoviridae and relatives based on a curated whole-genome alignment with bootstrap support of branches shown if >70/100. The phylogeny was rooted between subfamilies Rothmandenesvirinae and subfamily Enquatrovirinae. (F) Maximum-likelihood phylogeny of the Iiscvirinae subfamily based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. The phylogeny was rooted between the genus Jilinvirus and the other genera. The exemplary prophages of E. coli and Aeromonas media were identified in strains USECESBL382 (NCBI GenBank AATBGZ010000031) and T0.1-19 (NCBI GenBank CP038441), respectively, using PHASTEST [204]. In (D-F), phages of the BASEL collection are highlighted by pink and green (new) or gray (old) phage icons. (G) The results of quantitative phenotyping experiments with the three newly described phage isolates regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and methods). Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g008

Properties of myoviruses in genus Wifcevirus

Phages of the genus Wifcevirus within Caudoviricetes have been described comparably rarely with currently less than 20 sequenced isolates in NCBI GenBank that all infect E. coli or Shigella (Fig 9A–9E). They have genomes of ca. 66–72kb and were described as a genus when two new isolates, WFC and WFH, had been linked to several phages that had been sequenced earlier [58]. Interestingly, distant relatives of these phages in the Pbunavirus genus are among the P. aeruginosa phages most used for personalized therapy [84,85]. Based on clear homology, the cryogenic electron microscopy structure of virions of a Pbunavirus has greatly helped us to annotate the structural genes of Wifcevirus phages [86]. No Wifcevirus infecting E. coli laboratory models K-12 or B has ever been reported, limiting our knowledge about the biology of these phages and their interaction with the host. However, one previous study on two Wifcevirus isolates reported remarkably fast replication with only 3–5 min between adsorption and release of progeny [87]. This is much faster than even mutants of phage T7—another phage considered to be fast—which had been evolved for rapid replication but did not get faster than seven minutes in comparable assays [88]. This property of Wifcevirus phages could be of great interest for biotechnology or clinical applications, but the underlying molecular mechanisms have not been explored.

Fig 9. Overview of Wifcevirus phages.

(A) Schematic model of host recognition by Wifcevirus myoviruses targeting host O-antigen (blue) using tail fibers. Note that these phages carry two sets of tail fibers (see below) whose relationship is unclear, so only a simple schematic with one set is shown. (B) Maximum-likelihood phylogeny of the Wifcevirus genus and relatives based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. The phylogeny was midpoint-rooted between the distantly related Wifcevirus and Pbunavirus genera. Phages of the BASEL collection are highlighted by little green phage icons. (C) Representative TEM micrograph of phage HermannAdler (Bas98) with a virion morphology very similar to other representatives that have been visualized previously [58]. (D) Wifcevirus isolate ManiWeber (Bas99) was streaked for single plaques on E. coli K-12 strains with and without cryptic prophages. Kuravirus SkaterBoy (Bas90; see Fig 7) is shown as a control without sensitivity to the presence of cryptic prophages. (E) The results of quantitative phenotyping experiments with Wifcevirus phages regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and methods). Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g009

Our four isolates (Bas98−101) belong to both major subgroups of Wifcevirus phages (Fig 9B) and all depend totally on restored O16-type O-antigen as host receptor (Fig 9E), suggesting that their tail fibers target this glycan as host receptor (see also the dedicated section below). This is reminiscent of their Pbunavirus relatives which are known to target the LPS of P. aeruginosa as host receptor [86,89,90]. All our isolates could only be obtained using E. coli K-12 without cryptic prophages as host and their plaque formation is severely impaired in their presence (Fig 9D), though a quantitative reduction of plaque counts was only observed for VerenaGrendelmeier (Bas101; Fig 9E). All isolates are not sensitive to any tested bacterial defense system including RM systems despite the lack of any known genome protection mechanism (Fig 9E and S2 Table). One notable exception is the EcoP1_I type III RM system, to which they are very sensitive even though they have similar numbers of recognition sites for this system as for the EcoCFT_II type III RM system to which they are highly resistant (Fig 9E and S2 Table).

Properties of Autographiviridae: Studiervirinae genera Teseptimavirus and Kayfunavirus

Podoviruses of Autographiviridae subfamily Studiervirinae are well-known from classical representatives T3 (Teetrevirus genus) and T7 (Teseptimavirus genus) and have been frequently isolated against diverse enterobacterial hosts. In our previous work, we characterized T3 and T7 together with several new isolates from different genera of this subfamily [10]. As also suggested by the literature, we found that all these phages depended on the LPS core of E. coli K-12 as a host receptor when infecting this host, though they were invariably unable to infect this host when O-antigen expression was restored [10,91]. These phages exhibited considerable flexibility in their glycan recognition in a way that truncations of the LPS core reduced infectivity but that usually no individual truncation fully abolished it [10,91]. Given the ubiquity of O-antigen expression among wild E. coli strains and its barrier function, it seems likely that these phages have evolved to target these or other surface glycans as host receptor and more accidentally recognize also the LPS core of E. coli K-12. Previous work has already well established that diverse representatives of Autographiviridae target specific types of O-antigen as host receptor which determines the host range of these phages [29,33,92–94].

In the current study, we indeed isolated several different Teseptimavirus phages that depend on restored O16-type O-antigen expression of E. coli K-12 and included three of these in the expanded BASEL collection (Bas102–104; Fig 10). Furthermore, we isolated ElsieAttenhofer (Bas105), an O16-dependent phage of the distantly related Kayfunavirus genus that also contains various representatives known to target host glycans such as phage K1F targeting the K1 capsule [95] or CLB_P1 targeting the O104-type O-antigen of E. coli strain 55,989 [96] (Fig 10B). Our four new isolates encode a single tailspike or tail fiber in their genomes analogous to other Autographiviridae phages and likely use it to target O16-type O-antigen (see in the dedicated section below). Like their relatives, they have genomes of ca. 40kb size and are sensitive particularly to type II and type III RM systems if they have a considerable number of recognition sites (Fig 10E and S2 Table) [10]. Beyond restriction, studies on model phages T3 and T7 as well as our previous work suggested that Autographiviridae phages are specifically targeted by the PifA defense protein dependent on the allelic variant of the gp1.2 dGTPase inhibitor gene [10,97,98]. Of the three new Teseptimavirus phages, only MeretOppenheim (Bas102) encodes a gp1.2 allele that is similar to those of sensitive phages T7 and JacobBurckhardt (Bas65), and it is indeed significantly inhibited by PifA (Figs 10E and S6A). However, also the Kayfunavirus ElsieAttenhofer (Bas105) is sensitive to PifA (Fig 10E), but it carries a distinct variant of the gp1.2 gene (S6A Fig). These results suggest that Autographiviridae sensitivity to PifA is more complex than two allelic variants of one locus and/ or that additional factors such as the major capsid protein may play a role (as indicated by previous work [98]).

Fig 10. Overview of Autographiviridae: Studiervirinae and Lederbergvirus phages.

(A) Schematic model of host recognition by podoviruses with enzymatic tailspikes targeting O-antigen (e.g., Kayfunavirus and Lederbergvirus) based on previous work on phage T7 and P22 [71,101,203]. (B) Maximum-likelihood phylogeny of the Studiervirinae subfamily of Autographiviridae based on a curated whole-genome alignment with bootstrap support of branches shown if > 70/100. The phylogeny was midpoint-rooted between distantly related genus Kayfunavirus and the other genera. Phages of the BASEL collection are highlighted by violet (new) or gray (old) phage icons. (C) Representative TEM micrograph of Kayfunavirus ElsieAttenhofer (Bas105). (D) Maximum-Likelihood phylogeny of the Lederbergvirus genus based on an alignment of their structural genes with bootstrap support of branches shown if >70/100. This phylogeny was midpoint-rooted between two major clades of this genus. Phage CurroJimenez (Bas106) is highlighted by a violet phage icon while phages Huey, Dewey, and Louie are highlighted by gray phage icons. (E) The results of quantitative phenotyping experiments with Autographiviridae phages regarding sensitivity to altered surface glycans and bacterial immunity systems are presented as efficiency of plating (EOP). Small notes of “ΔRM” or “Δ9CP” indicate on which host strain the respective phage has been characterized (see Materials and Methods). Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3003063.g010

Properties of podoviruses in genus Lederbergvirus

Phages in the genus Lederbergvirus are among the most well-known phages that use tailspikes to target enterobacterial surface glycans and particularly the O-antigen as host receptors (Fig 10A) [33,99,100]. These phages are podoviruses with genomes of around 40 kb size and a temperate lifestyle typically governed by lambda-like lysogeny control elements comprising a cI-like repressor, cII, and Cro [99]. The prototypic Lederbergvirus is phage P22 that has become a key model for viral host interaction. This phage targets the O:4 O-antigen of its Salmonella Typhimurium host with its tailspike before a tail needle penetrates the outer membrane and eventually a trans-envelope channel for DNA injection is formed (Fig 10A) [101,102]. Similarly, well-known Lederbergvirus phages like E. coli phage HK620 and Shigella phage Sf6 bind specific O-antigen variants of their respective hosts, while others target different glycans like the K1 capsule in the case of E. coli phage CUS-3 [103,104]. Its potency in the transduction of genes between Salmonella Typhimurium strains has made P22 a crucial tool for research on this organism [105], and the propensity of this phage for lateral transduction due to the initiation of viral replication and packaging before excision from the host genome can result in extremely high rates of gene transfer [106].

Beyond these classical models, Lederbergvirus phages infecting enterobacterial hosts are known to be common and diverse [107]. It was thus unsurprising that we isolated three different phages of this genus—Huey, Dewey, and Louie—which showed homogeneous and highly turbid plaques indicative of lysogen formation and self-immunity (S6B Fig; NCBI GenBank accessions PQ850600, PQ850609, and PQ850616) [108]. However, we also isolated one representative called CurroJimenez (Bas106) that did not show temperate behavior when infecting E. coli K-12 with restored O-antigen and was included in the BASEL collection that focuses on virulent phages (S6B Fig). Since it does not have obvious defects in its lysogeny control genes, we speculate that E. coli K-12 may lack the specific attB site which would be targeted by the integrase of this phage. The phylogeny of Lederbergvirus phages shows different distinct clades among which CurroJimenez, Dewey, and Louie are closely related in a sister group of P22 itself, while Huey is found in a more distant clade closer to CUS-3 and also HK620 (Fig 10D). However, the high genomic mosaicism of this viral genus makes it difficult to draw conclusions about the relationships of individual phages beyond the specific genes that have been used to calculate the phylogeny [33,109]. Phage CurroJimenez totally depends on the expression of restored O16-type O-antigen for the infection of E. coli K-12, which identifies this glycan as the host receptor targeted by its tailspike (Fig 10E). The phage is also highly sensitive to restriction-modification systems, which is not unexpected given the absence of any predictable anti-restriction mechanisms (Fig 10E and S2 Table) and seems to be common among temperate phages [10].

Tailspikes and tail fibers to overcome the O16-type O-antigen barrier

Since the O-antigen barrier is ubiquitous among enterobacteria, it must be overcome or bypassed by any phage infecting these hosts (Fig 1). For classical model phages, the O-antigen has been seen as a facultative first receptor for initial host recognition followed by the binding to a second, terminal receptor for irreversible adsorption with a distinct receptor-binding protein [29,34]. As an example, phage T5 efficiently infects the E. coli K-12 model strain—not expressing any O-antigen—by directly targeting the FhuA porin as a terminal receptor with its central tail fiber [110]. While its lateral tail fibers are not required to infect a host without O-antigen, increased expression of the polymannose O8-type O-antigen that they target results in concomitantly faster adsorption of T5 to these hosts [111]. Conversely, increased expression of O8-type O-antigen results in decreasing adsorption of a T5 mutant specifically lacking these lateral tail fibers [111]. Restoration of the O16-type O-antigen of E. coli K-12 (which is not bound by the lateral tail fibers of T5) results in complete abrogation of T5 infections (Fig 1B and our previous work [10]). Similar results have been obtained for other phages of different morphotypes that express two distinct RBPs for successive recognition of O-antigen and a terminal cell surface receptor ([recently reviewed in reference 29]), and the concept also applies to analogous recognition of other initial receptors like capsules [112]. However, previous work had already indicated that some phages—primarily podoviruses like classical models T7 and N4—can also merely depend on a single receptor such as O-antigen or other glycans for host recognition and irreversible adsorption [17,29,31,70]. Similar to these, the O16-dependent phages characterized in this study typically feature a single RBP to target this glycan, seemingly for both initial host recognition and subsequent irreversible adsorption (Figs 11 and 12).

Fig 11. Phylogenetic analyses of tailspikes and tail fibers.

(A) Maximum-likelihood phylogeny of pectin lyase-like receptor-binding domains (InterPro accession IPR011050 [114]) of tailspikes encoded by our new BASEL phages and relevant controls. The phylogeny was midpoint-rooted between previously characterized tailspikes targeting the O:4 O-antigen of Salmonella Typhimurium (blue) and those of our new isolates targeting E. coli O16-type O-antigen (orange; including a few additional previously published phages with very similar tailspikes). Bootstrap support of branches is shown if >70/100. Phages of the BASEL collection are highlighted by colorful phage icons. (B) Maximum-likelihood phylogeny of Gp12-like receptor-binding domains (InterPro accession SSF88874 [114]) of tail fibers encoded by our new BASEL phages and relevant controls. The phylogeny was midpoint-rooted between a clade of distant Gordonclarkvirinae tail fibers at the top (blue) and those of Wifcevirus, Tequintavirus, and the Gordonclarkvirinae presented in this study (orange). Note that a clade with Wifcevirus receptor-binding domains (bottom) includes diverse phages from the database as well as Tequintavirus IrisVonRoten (Bas32 [10]) scattered between the Wifcevirus isolates characterized in the current study.

https://doi.org/10.1371/journal.pbio.3003063.g011

Fig 12. Predicted structures of representative tailspikes and tail fibers.

(A) AlphaFold-Multimer [198] predicted structures of homotrimeric tailspikes from selected BASEL phages (see also Fig 11A). (B) Predicted structures of the representative T-shaped tail fiber complexes of IrisVonRoten (Bas32, Tequintavirus), AnnaHegner (Bas87, Gordonclarkvirinae), and ManiWeber (Bas99, Wifcevirus) which all feature the dual Gp12-like RBD1 and T5-like RBD2 (predicted with the C-terminal intramolecular chaperone domain (IMC) attached; see also Figs 11B and S10). While ManiWeber features an N-terminal anchor domain for baseplate attachment, the T-shaped tail fibers of IrisVonRoten and AnnaHegner attach to proximal fibers (Gp22 and Gp9, respectively), which connect the complete fiber structure to the distal baseplate complex. The proximal fiber protein Gp22 of IrisVonRoten contains an additional galectin-like domain (inset box; colored N- (blue) to C- (red) terminus) that protrudes from the C-terminus of the fiber and could contribute to host receptor interaction as an additional RBD. While the overall setup of the T-shaped tail fibers is highly conserved, the RBDs are highly variable and seem to be frequently transferred horizontally (Figs 11B, S9C, and S10D). (C) Predicted structures of other tail fibers with single RBDs (more shown in S7 Fig). The short-tail fiber (Gp13) of AnnaHegner (Bas87, Gordonclarkvirinae) and the second tail fiber (Gp32) of ManiWeber (Bas99, Wifcevirus) have an elongated tip similar to the long-tail fiber of phage T4 including conserved HxH motifs for ion coordination [118]. Tail fibers of JakobBernoulli (Bas07, Drexlerviridae; Gp36) and RuthEpting (Bas85, Dhillonvirus; Gp29) both contain elongated N-terminal coiled coils connected to receptor-binding T5-like fiber tips with C-terminal IMC domains. In (A)-(C), N-terminal anchor domains and fiber shafts are colored blue, while the RBDs are colored red. A ~5-nm vertical scale bar is provided for reference. In all panels, viral morphotypes are annotated by phage icons analogous to Fig 11A and 11B.

https://doi.org/10.1371/journal.pbio.3003063.g012

Our new isolates of small siphoviruses in genera Kagunavirus, Skatevirus, and others encode tailspikes for recognition and enzymatic processing of the O16-type O-antigen that are encoded directly downstream of a J-like central tail fiber protein (S2 Fig). Notably, this is the same locus where the bona fide RBPs described in our previous work are encoded [10] as well as the capsule-targeting tailspikes of small siphoviruses targeting Klebsiella spp. (S2 Fig) [113]. We therefore conclude that these different RBPs are used analogously for terminal receptor recognition. Interestingly, the tailspikes of these distinct siphoviruses as well as Kayfunavirus ElsieAttenhofer (Bas105), Gamaleyavirus RudolfBernhard (Bas96), and Lederbergvirus CurroJimenez (Bas106) are all homologous and feature an enzymatic pectin lyase-like receptor-binding domain (InterPro accession IPR011050 [114]) connected to variable anchor domains (Figs 11A and 12A). A phylogeny including distant relatives known to target the O:4 O-antigen of Salmonella Typhimurium such as P22 shows that the tailspikes of our O16-targeting phages form a distinct, though diverse, group alongside tailspikes from a few phages isolated in previous studies (Fig 11A) [64]. This close relationship between tailspikes from very different phages targeting O16-type O-antigen highlights the frequent horizontal exchange of loci encoding these RBPs [33,115].

Based on prominent examples of glycan-targeting tailspikes, we had initially expected that phages depending on O-antigen as host receptor would mostly use enzymatic tailspikes to penetrate the glycan barrier [33,78,94]. However, many of our O16-dependent phage isolates had tail fibers instead. While these varied in length and the types of receptor-binding domains (RBDs), many contained an RBD homologous to that of Gp12 short-tail fibers of T-even phages targeting the E. coli LPS core [116] (InterPro accession SSF88874 [114]). These Gp12-like RBDs are located in the distal parts of lateral tail fibers across various phages, including myoviruses of the Wifcevirus genus, siphovirus IrisVonRoten (Bas32), and podoviruses of the Gordonclarkvirinae subfamily (Fig 11B). However, such a broad distribution of closely related RBDs is not unexpected, given the frequent horizontal transfer of tail fiber components [115].

Interestingly, a phylogeny shows that the Gp12-like RBDs of Wifcevirus isolates targeting O16-type O-antigen are scattered between homologs from diverse other Wifcevirus phages, suggesting that they either mediate very broad host recognition or that these other Wifcevirus isolates could also target this specific glycan (Fig 11B). Conversely Gp12-like RBDs of Gordonclarkvirinae targeting O16-type O-antigen are only very distantly related to homologous domains of other phages from this group (Fig 11B). For one of these others, EP335, it had been shown previously that its lateral tail fiber directly binds O157-type O-antigen, suggesting that this may be a conserved feature of these tail fibers with Gp12-like RBDs [117]. Other phages from the expanded BASEL collection encode different types of tail fibers with other RBDs that all likely target this glycan, highlighting the versatility of these non-enzymatic RBPs (see predicted structures in Figs 12B, 12C, and S7).

Phages with several receptor-binding proteins or dual receptor-binding domains

Notably, some of our new isolates that depend on O16-tpe O-antigen of E. coli K-12 encode two and not only one bona fide RBP. Of these, Jilinvirus HeidiAbel (Bas97) encodes separate long and short-tail fibers that are likely used successively during host recognition analogous to the long and short-tail fibers of much larger T-even myoviruses [10,73,74] (S7 Fig). A very different situation is apparent for Gamaleyavirus RudolfBernhard (Bas96), which encodes two tailspikes of which one carries a pectin lyase-like domain closely related to others from our collection (Figs 11A and 12A) and the other one seems unrelated. Such a setup of two tailspikes is common for Gamaleyavirus phages and other Schitoviridae (see also S8 Fig) [33,78]. It seems likely that the two tailspikes are used for alternative target recognition similar to Autographiviridae phage SP6, which carries two separate tailspikes that it can use to target different types of O-antigen to infect different strains of Salmonella [94].

While the presentation of two independent RBPs is thus not uncommon for podoviruses, the Gordonclarkvirinae are the only phages of this morphotype in the BASEL collection that encode separate long- and short-tail fibers, which are likely used successively and not alternatively [57]. The short-tail fibers of these phages are highly conserved and appear to be vertically inherited (Figs 12C and S9A) [57]. Similarly, the Wifcevirus phages also feature two different tail fibers, of which one—predicted to have an elongated RBD tip similar to the T4 Gp37 long-tail fiber [37] (Fig 12C)—is highly conserved across this genus (S9B Fig). We therefore suggest that these conserved RBPs have a more mechanical role during the infection process rather than recognizing divergent host receptors. Conversely, both the Gordonclarkvirinae and Wifcevirus phages feature a second tail fiber with a predicted distinctive T-shaped structure presenting dual RBDs, the Gp12-like RBD1 (see above) and a second, T5-like RBD2 that could not be identified from primary sequence alone (Figs 12B and S10A–S10C). Both the Gp12-like RBD1 and the T5-like RBD2 differ widely between homologs of the T-shaped tail fibers (S9C Fig) and have a phylogenetic distribution that is scattered across viral groups and morphotypes (Figs 11B and S10D). These observations are in line with the recognition of highly variable host receptors like O-antigen, suggesting that these T-shaped tail fibers may generally target these glycans and possibly others like capsules.

This remarkable T-shaped tail fiber structure is predicted to have a very similar arrangement in Gordonclarkvirinae (exemplified by Gp10 of AnnaHegner, Bas87), Wifcevirus (exemplified by Gp114 of ManiWeber, Bas99) and also the lateral tail fibers of some Tequintavirus phages like IrisVonRoten (Gp24, Bas32; see Fig 12B). Using Gp24 of IrisvonRoten as a representative, the DALI server identified structural similarities and shared structural elements between the Gp12-like RBD1 with X-ray crystal structures of the distal knob domain (D10) of the T4 long-tail fiber Gp37 [118] (Z-score 8.8, RMSD 3.2, 100 residues) and—as expected from primary sequence—with the Gp12 short-tail fiber distal tip [118] (Z-score 6.2, RMSD 4.1, 127 residues; S10A–S10C Fig). Likewise, the T5-like RBD2 of Gp24 of IrisVonRoten also shares structural similarity with the distal binding tip and intramolecular chaperone (IMC) domain of the phage T5 lateral tail fiber [119] (Z-score 6.3, RMSD 4.8, 179 residues; see S10A–S10C Fig). Finally, while the T-shaped RBPs of IrisVonRoten (Gp24) and AnnaHegner (Gp10) connect to a long proximal fiber (Gp23 and Gp9, respectively), the T-shaped RBP of ManiWeber (Gp114) instead features a prominent attachment domain at its N-terminus that likely facilitates direct baseplate attachment (Fig 12B). This diversity highlights the modular versatility of bacteriophage RBPs for host recognition and the initiation of viral infections.

Host range across pathogenic E. coli and Salmonella typhimurium

The host range of bacteriophages is a crucial property both with regard to their ecology and for applications in clinics and biotechnology [120–122]. It can be defined in different ways, but researchers generally distinguish a phage’s lytic host range—estimated by the ability to lyse a host strain in a spot test—and its plating host range, i.e., the ability to form plaques [120,121]. While plaque formation requires the phage to bind the host cell and to overcome or bypass all layers of bacterial immunity at least partially, the lytic host range is primarily informative about viral host recognition due to “lysis from without” at high phage-to-host ratios even in the absence of viral replication [123]. In our previous work, we tested the host range of BASEL phages and classical model phages against commonly used model strains of E. coli and S. typhimurium [10]. For the current study, we anticipated that the diverse O-antigen variants of these strains would largely exclude the mostly O16-dependent phages presented here. This would make it impossible to estimate their interaction with bacterial immunity. We therefore included three additional E. coli strains encoding O16-type O-antigen that belong to sequence types (ST)-131, −144, and −8,281 (see section Bacterial strains and strain construction in Materials and methods).

Our experiments to determine the lysis host range indeed revealed that most O16-dependent phages lysed all or some of the O16-expressing strains but barely any of the strains with rough LPS or another type of O-antigen (Fig 13A). This is in line with the expected dominant role of O-antigen for the host recognition of these phages, especially since most of them do not encode additional receptor-binding proteins besides the O16-targeting tailspikes or tail fibers (Figs 11 and 12) [29–31,35]. However, there were stark differences between the different groups of phages that we tested. All small siphoviruses (genera Kagunavirus, Skatevirus, and relatives; Bas70–82) and many podoviruses (Bas95/96 Bas102–106) had a very narrow host range beyond strains encoding O16-type O-antigen and lysed almost none of the strains without it (Fig 13A). These phages have either an O16-targeting enzymatic tailspike or simple tail fibers (Figs 11, 12 and S7), suggesting that their ability to target additional host receptors is limited. Conversely, isolates of Gordonclarkvirinae and Wifcevirus groups showed a very broad host recognition and could in some cases even lyse strains with rough LPS (Fig 13A). This indicates that the T-shaped tail fibers of these phages are more versatile and might be able to bind different glycan structures with their two RBDs (Figs 11, 12, S7, S9B, S9C, and S10). Previous work had indeed already observed a comparably broad recognition of some isolates of large myoviruses (e.g., Tevenvirinae, Vequintavirinae and relatives, etc.) or large siphoviruses (e.g., Markadamsvirinae) that commonly carry complex long-tail fibers [10,35,59,124,125]. Our data do not show unambiguously which of the two RBDs of the T-shaped tail fibers—or both—may be critical for O-antigen recognition and bacterial infection in our experiments. However, the phylogeny of the T5-like RBD2 does not correlate with any host recognition data (compare Figs 13A and S10D). Conversely, the phylogeny of the Gp12-like RBD1 shows different clusters for Gordonclarkvirinae with broadest host recognition (Bas87 and Bas91) and others for those with narrower host range (Bas86/89/94 and Bas88/89/82/83; compare Figs 11B and 13A). This may suggest a bigger role for the Gp12-like RBD in host recognition of Gordonclarkvirinae across our set of host strains.

Fig 13. Host range of new phage isolates in the BASEL collection.

(A) Lysis host range of all new BASEL phages and controls (same as in S1B Fig) as determined by qualitative spot test assays with high-titer phage stocks (>10⁹ pfu/ml if possible; see Materials and methods). The O-antigen type of different E. coli and Salmonella Typhimurium strains and possibly rough LPS (as O(-)) are highlighted at the top, while phages with O16-targeting tail spikes or tail fibers are distinguished on the left side. Lysis of a host strain is shown by a red cell lysis icon while the inability to lyse a strain is shown as a green icon in the shape of a cell with full integrity. (B) Plating host range of all new BASEL phages and controls (like in (A)) as determined by quantitative efficiency-of-plating (EOP) assays using serial dilutions of phage stocks. Data points and error bars represent average and standard deviation of at least three independent experiments. Raw data and calculations are available in S1 Data. In (A) and (B), K-12 strains have been selected differently for phages inhibited by its cryptic prophages (Δ9CP) and for those who are not (ΔRM).

https://doi.org/10.1371/journal.pbio.3003063.g013

The siphoviruses with O16-targeting lateral tail fibers and additional central fibers to bind a porin as terminal receptor—new Dhillonvirus isolates Bas83–85 as well as Bas07 (JakobBernoulli) and Bas32 (IrisVonRoten)—lysed all rough strains but differed in their ability to infect others (Fig 13A). This confirms the barrier function of mismatched O-antigen types for these phages and the expected facultative nature of O-antigen binding when the terminal receptor on the cell surface is accessible [10,28]. As shown already previously [10,12,17], phages that can bypass O-antigen by targeting the NGR glycan (Bas60–62 and N4 as well as Bas69) have a very broad host recognition with the Justusliebigvirus phages Bas60–62 featuring the largest lysis host range which even includes Salmonella Typhimurium (Fig 13A). This is likely due to the array of different tailspikes and tail fibers expressed by these phages, which enables them to overcome diverse glycan barriers [126]. Conversely, we also observed that many apparently O16-dependent phages failed to lyse one or more of the strains with this O-antigen type that we tested, especially when the phages use enzymatic tailspikes for host recognition (Fig 13A). Since these phages do not encode any other receptor-binding proteins, we find it unlikely that bacterial variation in other potential receptors such as the LPS core may be responsible for this behavior. Instead, it seems likely to us that strain-specific O-antigen modifications—which are abundant and diverse ([recently reviewed in reference 36])—could prevent the binding of some highly specific tailspikes to certain variants of O16-type O-antigen.

The plating host range of all phages was usually much narrower than their lysis host range (Fig 13A and 13B), which shows the impact of bacterial immunity in the diverse strains that we used. This effect was particularly strong for all phages with considerable sensitivity to RM systems such as small siphoviruses Bas70−82 (Fig 13B; compare Figs 5D and 6G). Conversely, phages with pronounced resistance to RM systems of the Gordonclarkvirinae (Bas86-93) and Gamaleyavirus (Bas96) groups could form plaques on many of the strains that they can lyse, indicating that this abundant and broadly targeting form of defense systems has a major impact on phage host range (Fig 13B; compare Figs 7D and 8G) [40,46,47]. This is in line with previous work showing a broad plating host range for individual isolates of Gordonclarkvirinae [117,121,127] and a narrower one for individual Wifcevirus phages or small siphoviruses like Kagunavirus isolates [35,59]. Intriguingly, Autographiviridae phages (Bas102−105) and in particular the new Dhillonvirus isolates (Bas83−85) formed plaques on nearly all strains that they could lyse despite their only moderate resistance to RM systems (Fig 13A and 13B; compare Figs 6G and 10D). It would be interesting to study the underlying molecular mechanisms that apparently enable these phages to overcome the native RM systems of these strains while the Wifcevirus phages—displaying considerable resistance to restriction in the K-12 host (Fig 9E)—fail to form plaques on any of the many strains which they can lyse (Fig 13A and 13B).

Cryptic prophages inhibit some BASEL phages via O-antigen glycosylation

One of the most curious biological results of this study has been the observation that phages belonging to very different viral groups—Bas74–76 (Kagunavirus), Bas83–85 (Dhillonvirus), Bas87 + Bas91 (Gordonclarkvirinae), and Bas98-Bas101 (Wifcevirus)—are inhibited by the presence of the cryptic prophages of E. coli K-12 (see Figs 5D, 6G, 7D, and 9E). Based on the diverse bona fide antiphage defense systems encoded in these genomic islands (Fig 4D), we had initially assumed that one or more of these would target the phages and impair their replication. The phylogenies of these phages show that sensitivity to inhibition by the cryptic prophages does not correlate with certain clades of their respective viral groups (Figs 5C, 6F, 7C and 9B). This observation suggested that sensitivity to this inhibition must be due to variable genes of the susceptible phage groups that could, e.g., be sensed/ targeted by bacterial immunity or may serve as antidefense factors, as shown in other cases [40,45,128]. However, we also realized that the loci encoding the receptor-binding proteins of the phages sensitive to inhibition by the cryptic prophages were distinct from their insensitive relatives. This can be seen, e.g., for the Kagunavirus tailspikes in Fig 11A where those of the isolates inhibited by cryptic prophages (Bas74–76) cluster in a completely different clade than the insensitive ones (Bas70–73). Similarly, the only features specifically shared by Gordonclarkvirinae AnnaHegner (Bas87) and LorandFenyves (Bas91), which are sensitive to the cryptic prophages, is the variant of their Gp12-like tail fiber RBD which is different from those of all other isolates (Fig 11B). We thus suspected that viral adsorption and not classical anti-phage defense systems could be responsible at least for some of the observed differences in sensitivity to the presence of cryptic prophages.

Interestingly, the cryptic prophage CPS-53 (also known as KplE1) of E. coli K-12 encodes a GtrABS O-antigen glycosylation system that adds the glucose side chain to the O16-type O-antigen of this strain for which its O-antigen biosynthesis locus itself lacks a dedicated glycosyltransferase (Fig 14A) [129,130]. Consistently, this side chain could not be found when the O-antigen structure of another E. coli strain with O16-type O-antigen was solved [131]. Given that O-antigen modifications can impair the adsorption of bacteriophages [36], we tested the role of the CPS-53 GtrABS system for the host range of our phage isolates that were inhibited by the K-12 cryptic prophages. Remarkably, genetic inactivation of this system resulted in susceptibility of E. coli K-12 to all these phages, suggesting that O-antigen glycosylation and no other aspect of bacterial immunity had inhibited their growth on E. coli K-12 with cryptic prophages (Fig 14B). These results were reminiscent of previous work showing that homologous tailspikes of phages EP75 and HK620 target differently glucosylated variants of E. coli O18 O-antigen, possibly due to differences in glycan repeat recognition [117]. Unlike the other phages that depend on O-antigen as host receptor, the Dhillonvirus isolates Bas83−85 were only affected by the GtrABS system in presence of restored O-antigen expression while they infected E. coli K-12 with and without cryptic prophages equally well in the absence of this glycan barrier (Figs 14B and S11). The Dhillonvirus phages have two RBPs and do not need recognition of the O-antigen as primary receptor to target their terminal porin receptor, but they depend on O-antigen recognition—which is apparently impaired by GtrABS glycosylation—to cross this layer when it is present [10,100].

Fig 14. GtrABS-mediated O-antigen glycosylation inhibits phage infection.

(A) O16-type O-antigen of E. coli K-12 with (left) or without (right) the GtrABS system of CPS-53 differs in the presence or absence of the glucose side chain, respectively [129,130]. (B) Serial dilutions of different phages were spotted on variants of E. coli K-12 BW25113 with restored O-antigen expression as well as deletion of the GtrS glucosyltransferase (ΔgtrS) or the nine cryptic prophages including CPS-53 (Δ9CP). Three Dhillonvirus phages (Bas83−85), two Gordonclarkvirinae (Bas87 + Bas91), and the Wifcevirus phages (Bas98−101) are inhibited by the K-12 cryptic prophages if the GtrABS system in CPS-53 is intact (see S11 Fig for the same experiment without O-antigen expression). Gordonclarkvirinae phage SkaterBoy (Bas90) and Tequintavirus IrisVonRoten (Bas32) were used as GtrABS-insensitive controls that do or do not depend on O16-type O-antigen, respectively. A small growth defect of the three Kagunavirus phages in the gtrS mutant compared to the Δ9CP strain (top right) is indicative of a slight inhibition of these phages by other factor(s) encoded in the cryptic prophages (C) Serial dilutions of lambda PaPa and Ur-lambda were spotted on variants of E. coli K-12 BW25113 either in its wildtype form (left) or with restored O-antigen expression and either active (middle) or inactivated (right) GtrABS system. Phages SkaterBoy (Bas90, Gordonclarkvirinae) and IrisVonRoten (Bas32, Tequintavirus) were added as controls that do or do not depend on the O16-type O-antigen and are both insensitive to GtrABS.

https://doi.org/10.1371/journal.pbio.3003063.g014

These observations with the Dhillonvirus phages were reminiscent of the conundrum that phage lambda—initially found as a prophage in E. coli K-12—was unable to infect this strain after restoration of its O-antigen [10]. No plaque formation in the presence of this glycan barrier could be seen both with the commonly used laboratory variant lambda PaPa or the original Ur-lambda (Fig 14C) [10,132]. Unlike the domesticated lambda PaPa, Ur-lambda does not only feature a central tail fiber targeting the LamB porin as a terminal receptor but also side tail fibers suggested to bind OmpC as the primary receptor [132]. Our results raised the question how lambda would initially have lysogenized the original E. coli K-12 strain that expressed O16-type O-antigen [27]. Interestingly, we find that Ur-lambda with its side tail fibers—but not lambda PaPa without them—can overcome the O16-type O-antigen of E. coli K-12 in the absence of GtrABS-mediated glycosylation (Fig 14C). These results suggest that the side tail fibers of Ur-lambda can bind unmodified but not glucosylated O16-type O-antigen as host receptor similar to the recognition of O-antigen as a primary receptor by other siphoviruses like JakobBernoulli (Bas07) or IrisVonRoten (Bas32; see Fig 1) [10]. Phage lambda might thus have lysogenized E. coli K-12 either before acquisition of CPS-53 or while the GtrABS system was transiently not expressed since related systems, e.g., in Salmonella, are controlled by phase variation [133].

Phages depending on O-antigen glycans are diverse and abundant in nature

While it is curious that we find O16-targeting phages to be much more abundant in environmental samples compared to sewage treatment plant inflow (S1 Table), previous work has also observed such differences in other contexts. As an example, phages targeting the conjugative pili of different enterobacterial plasmids were much more common in wastewater compared to river water [134]. While the underlying ecological phenomena remain elusive, it seems certain that wastewater does not merely contain a large abundance of E. coli phages compared to most environmental samples but also has its own biases.

One interesting observation for many O16-dependent phages characterized in this work is that they seem to use these glycans as terminal host receptors, while their genomes lack any recognizable strategy to inactivate this receptor during infections. Conversely, virulent phages targeting porins characteristically encode proteins such as Cor of T1 or Llp of T5 that disable their porin receptors to prevent adsorption of viral offspring to the debris of the lysed virocell [135,136]. Similar mechanisms are also common among temperate phages, including those targeting O-antigen glycans, which they can chemically modify, e.g., by glycosylation or acetylation ([recently reviewed in reference 36]), for “superinfection exclusion” of related viruses [14,137]. These examples show that such exclusion mechanisms would in principle also be possible for virulent phages, but it seems likely that the abundance of O-antigen molecules on the cell surface (e.g., compared to porins) makes a complete modification during lytic infections futile. Instead, recent work suggested an alternative strategy of phages that transiently block their offspring’s host recognition by “caps” made of chaperones on receptor-binding proteins instead of disabling the host receptors [138]. We expect that this strategy and similar concepts may be more widespread among glycan-targeting phages, and we look forward to future research uncovering their molecular mechanisms and evolutionary context.

Traditionally, siphoviruses infecting E. coli were considered to exclusively target porins as terminal host receptors like well-studied model phages T1 or T5 [60,61]. However, more recent studies had already described siphoviruses infecting E. coli or other enterobacteria that target flagella, conjugative pili, or surface glycans as host receptors [33,93,139,140]. Using E. coli K-12 with restored O16-type O-antigen expression as host, we indeed isolated diverse small siphoviruses from genera Kagunavirus, Skatevirus, and others that use this surface glycan as an essential host receptor. It seems likely that the abundance and diversity of these groups had previously been overlooked due to the traditional focus on E. coli laboratory strains with rough LPS (see Fig 1A). Our results strongly suggest that O-antigen is the only and terminal host receptor of these isolates because they only carry a single receptor-binding protein—in all cases enzymatic tailspikes (Figs 11A and 12A)—which is encoded directly next to the J-like central tail fiber hub (S2A Fig). This is the same locus where other small siphoviruses infecting E. coli encode bona fide receptor-binding proteins as well as superinfection exclusion factors targeting different porins (S2B Fig [10,135]), and host recognition by these central tail fibers is known to trigger irreversible adsorption and genome injection [110,141]. We look forward to future studies characterizing the mechanistic link of O-antigen recognition and genome injection in more detail. One specific aspect deserving further investigation is the way how these processes are linked on the molecular level compared to phages with two distinct tail fibers carrying receptor-binding domains. Electron microscopy of small siphoviruses depending on O-antigen clearly shows three distal tail appendages that are likely the tailspikes presented on the J-like fibers (see Figs 5B and 6C). Conversely, siphoviruses targeting porins generally have only one central tail fiber targeting this terminal receptor besides three lateral tail fibers targeting O-antigen or other initial receptors [100]. This indicates that the single RBPs of siphoviruses targeting only O-antigen may be presented on the virion like lateral tail fibers while being functionally equivalent to central tail fibers for irreversible adsorption and triggering DNA genome ejection.

Our study highlighted once more that—besides a few models like P22, G7C, and T4 [37,72,78,142]—we lack a molecular understanding how many phages use their diverse tail fibers and tailspikes to recognize host receptors and initiate infections. While we had initially expected to isolate many phages with tailspikes targeting O16-type O-antigen that would move the virion to the cell surface by enzymatic activity, many instead carried tail fibers lacking enzymatic properties (Figs 12B–12C and S7). Beyond a mere binding to their LPS receptor (which has been well studied for several examples [37,70,119]), we speculate that successive transient interactions of tail fibers with O-antigen receptors may enable the fibers to “glide” through this barrier towards the cell surface (see Figs 7A and 9A). Of particular interest are the novel T-shaped tail fibers with separate Gp12-like and T5-like RBDs (Figs 12B and S10) that—to the best of our knowledge—are the first example of a tail fiber protein with two distinct RBDs in the same polypeptide. Conversely, branched structures combining distinct RBPs have been well described [143], and the long-tail fiber tip Gp37 of phage T4 features different patches in the same RBD to target distinct receptors [37].

Phage host range, glycan receptors, and bacterial immunity

Understanding the molecular mechanisms underlying the host range of different bacteriophages is important not only for ecological insights but also to enable effective applications in clinics [35,54–56,122]. While diverse aspects of phage–host interactions like bacterial metabolism and antiviral immunity all influence phage host range, the access to and recognition of viral host receptors generally play major roles [144,145]. As expected from previous work [10,28,30,35], we observed a dominant role of the O-antigen barrier for phage host range in a way that O-antigen types mismatched to the phages’ receptor-binding proteins abolished host recognition while phages with no other host receptor also fully depended on its presence. Interestingly, our data show that phages with tail fibers tend to have a broader host recognition than phages with tailspikes among the tested viruses (Fig 13A). Understanding these differences on the molecular level by studying distinct functionalities of these different types of RBPs more systematically would not only expand our knowledge of phage–host interactions but could also help applications in clinics and biotechnology, e.g., by phage host range engineering [146,147]. The expanded BASEL collection with diverse phages that target the same receptor—O16-type O-antigen—could be a useful tool to investigate these interactions.

The dual role of O-antigen as a barrier and as a host receptor is reminiscent, e.g., of the well-studied capsules of Klebsiella spp., which form a robust barrier to phage infection but also serve as a dominant factor for phage host range by specific recognition of capsule types by certain variants of RBPs [113,148,149]. The specificity and dominance of these interactions even enable their use for quite reliable phage host range predictions across Klebsiella strains [150]. Conversely, for other hosts like P. aeruginosa, the availability of long retractable type IV pili reduces the importance of glycans as host receptors, and for many Gram-positives like Staphylococcus aureus at least the host recognition is usually broader due to a lower inter-strain diversity of bacterial surface structures [14,151–153].

For E. coli, the picture is certainly more nuanced because, e.g., most of the O16-targeting phages that can lyse our different tested strains with this O-antigen type fail to form plaques on them (compare Fig 13A and 13B). While this shows a clear influence of bacterial immunity, a recent study that experimentally tested the host range of many E. coli phages across many strains did not observe a simple correlation between phage host range and the number of predicted antiviral defense systems [35], unlike what had been observed for P. aeruginosa and Vibrio spp. [16,154,155]. In agreement with recent work by David, Plantady, and colleagues on a different strain of E. coli [145], our results primarily suggest a link between the ability to form plaques on diverse strains and viral resistance to RM systems as the most common and most broadly targeting form of bacterial immunity (see above) [46,47,156]. However, the relative importance of different antiviral defense systems for observed phage host range would certainly be a promising field for future study beyond mere predictions, e.g., by contrasting the phage susceptibility of wildtype hosts with mutants specifically lacking one or more defense systems. Our results indicate that—after removal of restriction systems in our previous work [10]—the E. coli K-12 genomic background may not encode anti-phage defense systems in narrow sense that play a major role in phage host range. Even when all predictable defense systems were absent in the Δ9CP strain lacking all cryptic prophages (Fig 4D; see Bacterial strains in Materials and methods), the new viruses that could be isolated were merely susceptible to an O-antigen glycosylation system encoded in one of these genomic islands (Fig 14).

Expanding the BASEL collection to cover all major groups of E. coli phages

The value of the BASEL collection as a tool for molecular biology research lies in its assorted diversity across and within phage groups and on their ability to infect the model organism E. coli K-12 [6]. In this study we therefore aimed at enhancing the assorted diversity to cover all known major groups of E. coli phages without a major compromise on domesticated E. coli K-12 as host strain (Fig 15). After restoring its O-antigen and inactivating all predicted antiviral defenses—of which only O-antigen glycosylation played a detectable role –we indeed readily isolated representatives of diverse previously inaccessible viral groups, most of which strictly depend on O-antigen as host receptor. These include, e.g., diverse groups of small siphoviruses like the Kagunavirus genus, myoviruses of the Wifcevirus genus, and podoviruses of Gordonclarkvirinae or genus Kayfunavirus. Since phages from most of these diverse groups had never been isolated with any E. coli lab strain, there were only scarce—if any—previous studies on their molecular biology. This limitation restricted our analysis of the genomic and phenotypic data for some of them largely to data from the current study, while in our previous work on more well-studied groups of phages we could draw heavily from the generations of previous research in the field [10]. To improve the genome annotation of our new phage isolates, we therefore created a new tool—GAPS (Gene Annotation Pipeline based on sequence and structure Similarity; see Materials and methods) to infer functional information from remote homologies and structure predictions.

Fig 15. Overview of the complete BASEL collection.

The illustration shows all isolates of the complete BASEL collection with their Bas## identifiers. Isolates are sorted by morphotype and their current ICTV classification [68]. The color code distinguishes phages based on morphotype and genome size. For siphoviruses, small phages (ca. 30–50 kb) are shown in red, medium-sized phages (ca. 50–100 kb) are shown in orange, and large phages (>100 kb) are shown in yellow. For myoviruses, small phages (ca. 30–100 kb) are shown in green, medium-sized phages (ca. 100–150kb) are shown in cyan, and large phages (>100kb) are shown in blue. For podoviruses, small phages (ca. 30–50 kb) are shown in violet, and medium-sized phages (ca. 50–100kb) are shown in pink.

https://doi.org/10.1371/journal.pbio.3003063.g015

We anticipate that the completed BASEL collection will support various applications, from fundamental research on viral ecology and evolution over the work on new antiviral defense systems to studies on phage therapy. With an expanded biological diversity in still manageable numbers, these phages can all be grown on a regular E. coli K-12 by using a strain with inactivated GtrABS system in two variants, once with and once without restored O-antigen expression. Given the increasing interest in phage therapy but also its notable incidence of treatment failure [85,157,158], it is clear that not only fundamental research but also clinical applications could profit from a better understanding of links between phage genomes and taxonomy on one side and relevant biology/ phenotypes + clinical properties on the other side [35,54–56]. This is becoming increasingly evident as more and more engineered phages for therapeutic applications are developed [159]. For these applications, phages are mostly used as a vehicle to deliver proteins from nonviral origin such as CRISPR-Cas nucleases or antibacterial toxins [62,159–162]. This limitation highlights our lack of knowledge about the diverse molecular biology of phage–host interactions beyond a few traditional models. Unraveling the molecular mechanisms underlying new aspects of phage–host interactions—e.g., with tools like the BASEL collection or modern biochemistry and structural biology—could thus open up the untapped potential of engineered phages for applications in clinics or biotechnology.

Preparation of culture media and solutions

Lysogeny Broth (LB) was prepared by dissolving 10 g/l tryptone, 5 g/l yeast extract, and 10 g/l sodium chloride in Milli-Q H₂O and then sterilized by autoclaving. LB agar plates were prepared by supplementing LB medium with agar at 1.5% w/v before autoclaving. Phosphate-buffered saline (PBS) was prepared as a solution containing 8 g/l NaCl, 0.2 g/l KCl, 1.44 g/l NA₂HPO₄·2H₂O, and 0.24 g/l KH₂PO₄ with the pH adjusted to 7.4 using 10 M NaOH and sterilized by autoclaving. SM buffer was prepared as 0.1 M NaCl, 10 mM MgSO₄, and 0.05 M Tris (pH 7.5) and sterilized by autoclaving.

Bacterial handling and culturing

Escherichia coli and Salmonella enterica subsp. enterica serovar Typhimurium strains were routinely cultured in LB medium at 37 °C in glass tubes, plastic tubes, or Erlenmeyer flasks with agitation at 170 rpm. Selection for genetic modifications or plasmid maintenance was performed with ampicillin at 50 µg/ml, kanamycin at 25 µg/ml, and zeocin at 50 µg/ml.

Bacteriophage handling and culturing

Bacteriophages were generally cultured using the double-agar overlay method [163] with improvements of classical techniques described by Kauffman and Polz [164]. Top agar was prepared as LB agar with only 0.5% w/v agar supplemented with 20 mM MgSO₄ and 5 mM CaCl₂. Top agar plates were incubated at 37°C and plaques were counted as soon as they were identified by visual inspection. However, the plates were always incubated for at least 24 hours to record also slow-growing plaques.

High-titer stocks of bacteriophages were generated using plate-grown viruses. Briefly, phages were spread with a loop throughout the top agar before solidification to achieve confluent lysis after incubation for 16−24 hours at 37°C. Subsequently, top agars were scraped off from the bottom agar into each 15 ml of SM buffer and centrifuged at 8000g for 10 min. Supernatants were used as high-titer stocks and stored in the dark at 4°C. Phage Ur-lambda was obtained from the supernatant of an exponentially growing culture of the E. coli K-12 EMG2 strain (CGSC #4,401)—carrying an Ur-lambda prophage—after 2 hours of treatment with 0.5 µg/ml of mitomycin C to trigger prophage induction [132,165].

Phages of the original BASEL collection from our previous work [10] were generally grown on E. coli K-12 MG1,655 ΔRM. New isolates from this study were propagated on E. coli K-12 MG1,655 ΔRM with restored O-antigen (carrying pAS001 or the wbbL(+) genotype) except those with sensitivity to GtrABS-mediated O-antigen glycosylation. These phages were cultured on E. coli K-12 BW25113 Δ9CP or E. coli K-12 BW25113 ΔgtrS (both with restored O-antigen, i.e., carrying pAS001 or the wbbL(+) genotype). All bacteriophages used in this study are listed in S2 Table.

Bacterial strains and strain construction

All bacterial strains used in this work are listed in S3 Table, all oligonucleotide primers in S4 Table, and all plasmids in S5 Table.

Variants of Escherichia coli K-12 MG1,655 and BW25113

The primary host of our experimentation with E. coli K-12 was the ΔRM mutant of the MG1,655 laboratory strain, which we had reported in our previous work [10]. Briefly, we had engineered this strain to be a more permissive host for bacteriophages by knocking out all known restriction systems of E. coli K-12, i.e., the EcoKI type I RM system (encoded by hsdRMS) and the McrA, Mrr, and McrBC type IV restriction systems. For phage isolation and phenotyping, the strain carried a variant of the F-plasmid in which the pifA immunity gene had been deleted [10]. Phenotyping with plasmid-encoded antiviral defense systems was done with a strain carrying pBR322_ΔPtet as an empty vector control [10,166] (see S3 Table). Whenever indicated, we used the E. coli K-12 ΔRM variant with a repaired chromosomal wbbL gene that has restored O16-type O-antigen expression from our previous work [10] (see S3 Table). To rapidly restore O-antigen expression in diverse strains (e.g., porin mutants of the Keio single-gene knockout collection [167]), we created plasmid pAS001 that constitutively expresses wbbL (see below in section Plasmid construction). Fortuitously, the plasmid-encoded wbbL does not fully restore the O-antigen barrier (because most phages inhibited by it still show infectivity; compare Figs 1 and S11A) but enough of it to enable robust infections by all O16-dependent phages (compare Figs 2 and S11B).

To identify possible limitations of phage infectivity beyond host recognition, we focused on possible antiviral defenses in the K-12 genomic background beyond the restriction systems knocked out in E. coli K-12 MG1,655 ΔRM. One of these is the enigmatic enterobacterial type I-E CRISPR-Cas system known to be not functional in phage defense without dedicated genetic engineering [168,169]. In addition, E. coli K-12 strains encode Lit (reported to cleave bacterial EF-Tu when sensing the major capsid protein of some Tevenvirinae [170]), RnlAB (toxin-antitoxin system that aborts T4 infections in absence of the viral dmd “master key” antitoxin [171]), DicB (inhibitor of the bacterial ManYZ inner membrane conduit of lambda phage DNA translocation [172]), and a Hachiman type I system (targeting DNA replication of diverse phages; also known as AbpBA [44,47,173]). With the exception of the seemingly inactive type I-E CRISPR-Cas system, all these bona fide antiviral defenses are encoded in the cryptic prophages of E. coli K-12. We therefore used the E. coli K-12 BW25113 Δ9CP strain devoid of all cryptic prophages as an additional host for phage isolation and characterization [51]. This strain is derived from the BW25113 laboratory wildtype of E. coli K-12 which has a few minor differences compared to the MG1,655 laboratory wildtype [174,175] (see also Fig 4D). Most important for our work, E. coli K-12 BW25113 carries active variants of the three type IV restriction systems knocked out in our MG1,655 ΔRM strain as well as the hsdRMS locus encoding the EcoKI type I RM system. Unfortunately, due to the high sensitivity of E. coli K-12 BW25113 Δ9CP to stress and possibly also as a consequence of its already extensive genomic engineering, our attempted knockouts of these systems consistently failed. However, the EcoKI locus of BW25113 is restriction-deficient, and we are not aware of any wildtype phages that would be sensitive to the three type IV restriction systems of E. coli K-12, suggesting that the presence of these loci might have only a negligible effect. Analogous to our work with E. coli K-12 MG1,655 ΔRM [10], we also generated a variant of E. coli K-12 BW25113 Δ9CP with chromosomally restored wbbL(+) to isolate and study phages that would depend on this glycan (see S3 Table). The different variants of E. coli K-12 BW25113 Δ9CP were used for phage isolation and characterization just like the ΔRM variants with plasmid-encoded antiviral defenses and a pifA-deficient F-plasmid (see S3 Table).

Later during our study, we found that only the GtrABS O-antigen modification system of CPS-53 was responsible for the observed inhibition of phage growth by the cryptic prophages of E. coli K-12 (Fig 14). We therefore additionally used an E. coli K-12 BW25113 ΔgtrS mutant as well as a derivative of this knockout with chromosomally repaired wbbL—restoring O-antigen expression—for different experiments shown in this manuscript (see S3 Table). Briefly, a clean deletion of gtrS in the E. coli K-12 BW25113 strain was generated by lambda red recombineering in two steps analogous to the knockouts in our previous work [10]. In the first step, gtrS was replaced with a double-selectable cassette encoding kanamycin resistance and sucrose sensitivity (amplified from pUA139_cat-sacB_v3 using prDP0079/80). Subsequently, the double-selectable cassette was removed by recombineering with annealed 80 nt oligonucleotides spanning the desired deletion with 40 nt on each side (prDP0081/82). Oligonucleotide primers prDP0083/84 were used to probe the success of each step of recombineering by overspanning colony PCR over the gtrS gene (all listed in S4 Table). Finally, the wbbL(+) gene was restored by precisely removing the IS5 element disrupting this gene, as described in our previous work [10].

Other enterobacterial strains used for bacteriophage phenotyping

To determine the host range of our new phage isolates and relevant controls, we initially used the same strains as in our previous work (see there for details [10]). Briefly, besides the E. coli K-12 variants with and without restored O-antigen (see above; phylogroup A, ST10, O16(+/-):H48, K-12 LPS core) these were the rough E. coli B REL606 laboratory strain (phylogroup A, O7(-):K5:H(-), truncated R1 LPS core), uropathogenic E. coli model strains UTI89 (phylogroup B2, ST95, O18:K1:H7, R1 LPS core) and CFT073 (phylogroup B2, ST73, O6:K2:H1, R1 LPS core), and enteroaggregative E. coli model strain 55,989 (phylogroup B1, ST678, O104:H4, R3 LPS core). Similarly, we again used the common Salmonella enterica subsp. enterica serovar Typhimurium laboratory strains 12023s (also known as ATCC 14,028) and SL1344 that share the Salmonella Typhimurium O4-type O-antigen. Because the results with these strains were indistinguishable, only those with S. Typhimurium 12023s are included in this manuscript. Beyond these strains already included for host range analyses in our previous work, we included three additional E. coli strains with O16-type O-antigen (i.e., the same as E. coli K-12) belonging to ST131 (Egli lab collection isolate 720,834−18; phylogroup B2, O16:H5), ST144 (Egli lab collection isolate 714,478−19; phylogroup B2, O16:H6), and ST8281 (Egli lab collection isolate 711,043−19; phylogroup B2, O16:H5) [176].

Plasmids and plasmid construction

For the phenotyping of phage resistance or sensitivity to diverse bacterial immunity systems, we used plasmids encoding different antiviral defenses constructed in our previous work (see Fig 4B and 4C as well as S5 Table). Plasmid pAS001 encoding wbbL was cloned by inserting the wbbL gene downstream of a constitutively expressed chloramphenicol resistance gene on a plasmid with SC101 origin of replication (see details of plasmid construction in S6 Table).

Bacteriophage isolation

Efficiency of plating assays

We quantified the infectivity of a phage on a host strain as the efficiency of plating (EOP) [177], i.e., as the ratio of its plaque formation on that host over the plaque formation on susceptible reference strains. As references, we used E. coli K-12 MG1,655 ΔRM wbbL(+) or E. coli K-12 BW251139 Δ9CP wbbL(+), depending on the isolates’ sensitivity to the presence of cryptic prophages, since all tested phages grew very well with restored O16-type O-antigen expression. All variants of E. coli K-12 used for phenotyping carried an F-plasmid without pifA defense gene (unless that was to be tested) and pBR322_ΔPtet as empty vector control or plasmids expressing different antiviral defenses (see Bacterial strains above and S3 Table for a list). Note that the phenotyping of phages sensitive to the presence of cryptic prophages enabled us to use the isogenic host with prophages as a test condition for efficiency of plating but prevented us from testing phage sensitivity to the EcoKI type I RM system because the E. coli K-12 BW251139 Δ9CP strain has active EcoKI methylation (see Bacterial strains above and the genotype in S3 Table). It is therefore highlighted at all graphs whether ΔRM (EcoKI tested, cryptic prophage sensitivity not tested) or Δ9CP (EcoKI not tested, cryptic prophage sensitivity tested) have been used as hosts. To test for the effect of surface glycans on phage infectivity, we used host variants expressing no O-antigen but—unlike in our previous work [10]—no mutants with deeper truncations in the LPS core because almost all phages completely depended on O-antigen expression. The results with an E. coli K-12 mutant lacking NGR expression due to the inactivation of nfrA are merely shown in S1A Fig since evidently no new phage isolate used this glycan for host recognition.

For these quantitative experiments, we prepared top agars on regular LB agar plates by overlaying them with a soft agar composed of LB agar containing only 0.5% agar and additionally 20 mM MgSO₄ as well as 5 mM CaCl₂ supplemented with a suitable bacterial inoculum. We used 3 ml of this soft agar with 100 µl of bacterial overnight culture as inoculum for regular round Petri dishes (ca. 9.4 cm diameter) and 10 ml of top agar supplemented with 200 µl of bacterial overnight culture for larger square Petri dishes (ca. 12 cm × 12 cm). EOP experiments were performed with serial dilutions of phage stocks grown on the respective reference hosts prepared in sterile phosphate-buffered saline (PBS). For each experiment, 2.5 µl of all serial dilutions were spotted on the top agar plates and dried into the top agar before incubation at 37 °C for at least 24 hours. We then recorded plaque formation repeatedly throughout this time to accurately quantify the infectivity of fast- and slow-growing phages. Finally, we determined the EOP of each phage on each host as the ratio of plaques obtained on the experimental host over the number of plaques obtained on the respective reference strain [177]. Plaque counts and the calculation of EOP of all replicates of all experiments reported in this study (Figs 5D, 6G, 7D, 8G, 9E, 10E, 13B, S1A and S1B) as well as their summary statistics are compiled in S1 Data.

In cases when visual inspection could not unambiguously record plaque formation, we determined the EOP to be below the detection limit of 10⁻⁶ even if significant lysis from without was apparent (i.e., lysis zones caused by bacterial cell death without significant phage replication simply from the death of infected cells [123]). In cases where a phage/ host pair was on the edge between strong lysis from without and very poor plaque formation, we cautiously recorded the result as an EOP below detection limit.

Qualitative top agar assays

We also determined the lysis host range of all tested phages on diverse E. coli strains (including K-12 ΔRM or Δ9CP with and without restored O-antigen) as well as Salmonella Typhimurium, i.e., the range of host strains on which a certain phage caused lysis zones irrespective of whether these were accompanied by plaque formation or not (Fig 13A). These experiments were performed as qualitative top agar assays, as described in our previous work [10]. Briefly, top agar plates of each host strain were prepared as described for efficiency-of-plating assays. After solidification, we spotted each 2.5 µl of undiluted high-titer stocks of all tested phage isolates (typically >10⁹ pfu/ml) onto the top agar plates. When the spots had dried in, we incubated the plates at 37 °C for at least 24 hours before visual inspection. The appearance of a lysis zone in more than half of all replicates for a phage–host pair meant that the host was counted as part of the lysis host range for this phage. In all such cases, we subsequently quantified phage infectivity in efficiency-of-plating assays (see above).

Bacteriophage genome sequencing and assembly

Genomic DNA of phage isolates was prepared from culture lysates using the Norgen Biotek Phage DNA Isolation Kit and sequenced at SeqCenter (https://www.seqcenter.com/) or SeqCoast Genomics (https://seqcoast.com/) using Illumina technology. The short reads (151 bp) were trimmed and assembled using the Geneious Assembler implemented in Geneious Prime 2024.0.2 with a coverage of typically 50–100× (see S2 Table for each phage). Circular contigs (indicating complete assembly of a viral chromosome due to the fusion of characteristic repeats at the genome ends [178]) were typically obtained easily using the “Low Sensitivity/ Fast” setting with ca. 100′000 reads as input for the assembly. If necessary, ambiguities or incomplete assemblies were resolved by PCR amplification of the high-fidelity polymerase Phusion (NEB) followed by Sanger Sequencing. Sequences were linearized at the 5′ end either to the first nucleotide of the small or large terminase subunit gene or to the first position of the operon containing the small or large terminase subunit gene.

Initial bacteriophage genome annotation

The genomes of new phage isolates were annotated successively in several steps. To start, we used Pharokka to generate an automated, preliminary annotation of the genes in all phage genomes, which included tRNA prediction using tRNAscan-SE [179,180]. Subsequently, the annotations were refined manually based on identified orthologs in whole-genome alignments to different related phages generated using MAFFT v7.490 [181] implemented in Geneious Prime 2024.0.2. In parallel, we used the InterPro 101.0 protein domain signature database to maximize the functional information accessible from the primary sequence of phage proteins [114].

Curation of genome annotations using remote homologies and structure prediction

For the curation of phage genome annotations, we developed a computational pipeline called GAPS (Gene Annotation Pipeline based on sequence and structure Similarity). Briefly, this pipeline employs a combination of sequence and structure analyses, database searches, interpretation, annotation, and visualization routines to infer functional information for a given set of genes, of any organism, in an automated way. For the current study, phage gene sequences were fed to GAPS as features in a.gb file and translated to protein. For each protein sequence: (i) the protein and domain families were determined by searching the Pfam, NCBI_CD, CATH_S40 and PHROGS databases with HHblits [182–187]; (ii) identical sequences were found by searching the PDB, SwissProt and RefSeq databases [188–190]; (iii) similar sequences were found by searching the UniRef30 and PDB70 databases with HHblits [186–189]; (iv) the protein structures were predicted with AlphaFold2 [191]; (v) similar structures were found by searching the PDB, AlphaFold-Proteome and AlphaFold-UniProt50 databases with Foldseek [188,191,192]. The results were parsed and integrated programmatically for presentation in an interactive report file that summarizes the available information for all genes of the input genome in one combined overview page. Detailed results of all individual genes are presented in subpages. An exemplary report for Jilinvirus HeidiAbel (Bas97) is included in this publication as S2 Data and the added value of GAPS for the functional annotation of its genes is highlighted in S13 Fig. GAPS is written in Python 3.8 and is open-source. The code used to run GAPS for generating the data presented in this manuscript is available on GitHub (https://github.com/hiller-lab/gaps) and on Zenodo (https://www.doi.org/10.5281/zenodo.14277981) in the form of a Jupyter notebook.

Bacteriophage naming and taxonomy

Phage isolates were classified according to applicable rules of the International Committee on the Taxonomy of Viruses (ICTV) [68,193] largely as described in our previous work BASEL [10] (see S2 Table). Briefly, phages were first roughly assigned to viral family and, if possible, genus based on whole-genome BLASTN searches against the nonredundant nucleotide collection database [https://blast.ncbi.nlm.nih.gov/Blast.cgi; see also reference 194]. Subsequently, we unraveled the evolutionary relationships between our new isolates and reference sequences from NCBI GenBank across the diversity of relevant viral groups by calculating phylogenies based on either whole-genome alignments or alignments of conserved marker genes (see details in the dedicated section below). We then compared the clusters of viral sequences observed in these phylogenies to phage taxonomy as established by the ICTV and the relevant literature [68,195]. For classification on the species level, whole-genome alignments of new phage isolates with close relatives were generated as described below, and the nucleotide sequence identity was determined as the fraction of aligned genome sequence multiplied by the nucleotide identity of aligned segments. Phage isolates were classified as the same species as previously described relatives if their genomes showed >95% nucleotide sequence identity [196].

We named most phages in honor of scientists or historically significant persons linked to the city of Basel, Switzerland, where most of the phages had been isolated. To overcome the inherent gender imbalance due to historical biases, we included diverse female personalities from across the whole of Switzerland. Phages of Skatevirus and Nonanavirus genera were given names composed of common female first names in Switzerland in 1971 (when Swiss women obtained voting rights in federal elections) and “Muster” as surname (the German word for paradigm or example).

Sequence alignments and phylogenetic analyses

Sequence alignments of bacteriophage genomes, specific loci, or certain genes/ proteins were generated using MAFFT v7.490 implemented in Geneious Prime 2024.0.2 [181]. Occasionally, we supplemented poor or missing annotations in these sequences using the ORF finder tool of Geneious Prime 2024.0.2 guided by orthologous loci. Alignments were typically constructed using default settings of MAFFT with the fast FFT-NS-2 algorithm and 200PAM/k=2 or BLOSUM62 scoring matrices for nucleotide and amino acid sequences, respectively. Subsequently, we curated the alignments manually and masked nonhomologous parts for exclusion during phylogenetic analyses.

Maximum-likelihood phylogenies were calculated using PhyML 3.3.20180621 implemented in Geneious Prime 2024.0.2 from sequence alignments of orthologous genes/ proteins or orthologous sequence stretches of different bacteriophage genomes [197]. We used the HYK85 and LG substitution models for nucleotide and amino acid sequences, respectively, and always calculated 100 bootstraps. Phylogenetic relationships between phage genomes were inferred from whole-genome alignments (whenever possible) or from concatenated sequence alignments of conserved core genes on nucleotide or amino acid level as indicated for each phylogeny. Relevant reference sequences and controls were selected to represent the diversity of taxonomic units or protein families and typically identified by BLAST searches against the non-redundant nucleotide collection database [https://blast.ncbi.nlm.nih.gov/Blast.cgi; see also reference 194]. When available, well-studied model phages such as P22 and their sequences were always included to provide intuitive reference points for data interpretation. The tree files and underlying sequence alignments in NEXUS format are included in this manuscript as S3 Data file.

Structure predictions and analyses of receptor-binding proteins

Structure predictions were performed using AlphaFold-Multimer on the Cosmic² platform or on the AlphaFold3 webserver [198–200]. All predictions presented high internal confidence scores, which are presented in S7 Table. Illustrations were generated using the PyMOL Molecular Graphics System (Version 2.4.1, Schrodinger LLC). The Dali server was used to identify structural homologs in the Protein Data Bank (PDB) [201].

Morphological analyses by transmission electron microscopy

We analyzed the virion morphology of representative isolates from the different taxonomic groups covered in this manuscript by negative-stain transmission electron microscopy [202]. Briefly, we applied 5 μl drops of high-titer phage lysates to 400 mesh carbon-coated grids that had been rendered hydrophilic using a glow-discharger at low vacuum conditions. Subsequently, virions were stained on 5 μl drops of 2% (w/v) uranyl acetate and examined using an FEI Tecnai G2 Spirit transmission electron microscope (FEI Company) at 80-kV accelerating voltage, a Hitachi HT7700 transmission electron microscope (Hitachi) at 100 kV, or an FEI Talos F200C transmission electron microscope (FEI Company) at 200 kV. Images were recorded with a side-mounted Olympus Veleta CCD camera 4k (Tecnai G2 Spirit), an AMT XR81B Peltier cooled CCD camera (8 M pixel; Hitachi HT7700), or a Thermo Fisher Scientific Ceta 16M CCD camera (Talos F200C) using EMSIS RADIUS software at a nominal magnification of typically 150′000×.

Quantification and statistical analysis

Data sets were analyzed quantitatively by calculating mean and standard deviation of at least three independent biological replicates per experiment. Detailed information about replicates and statistical analyses is provided in the figure legends. In addition, we determined the probability for all efficiency-of-plating values that they are significantly different from the E. coli K-12 vector control using Student t test (two-sample unequal variance/ heteroscedastic) with a result considered significant if pS1 Data).