Sustained coevolution of phage Lambda and Escherichia coli involves inner- as well as outer-membrane defences and counter-defences

Bacteria often evolve resistance to phage through the loss or modification of cell surface receptors. In Escherichia coli and phage λ, such resistance can catalyze a coevolutionary arms race focused on host and phage structures that interact at the outer membrane. Here, we analyse another facet of this arms race involving interactions at the inner membrane, whereby E. coli evolves mutations in mannose permease-encoding genes manY and manZ that impair λ’s ability to eject its DNA into the cytoplasm. We show that these man mutants arose concurrently with the arms race at the outer membrane. We tested the hypothesis that λ evolved an additional counter-defence that allowed them to infect bacteria with deleted man genes. The deletions severely impaired the ancestral λ, but some evolved phage grew well on the deletion mutants, indicating that they regained infectivity by evolving the ability to infect hosts independently of the mannose permease. This coevolutionary arms race fulfils the model of an inverse gene-for-gene infection network. Taken together, the interactions at both the outer and inner membranes reveal that coevolutionary arms races can be richer and more complex than is often appreciated.


INTRODUCTION
An issue of longstanding interest is whether the coevolution of bacteria and virulent (lytic) phages involves endless rounds of bacterial defences and phage counter-defences. Based on experiments in chemostats, Lenski and Levin [1] suggested that bacteria typically had the upper hand, as Escherichia coli often eventually evolved resistance by deleting or inactivating the phage's specific receptor, which the phage could not readily overcome. This resistance did not imply the extinction of the phage, however, because it often reduced the bacteria's competitiveness for resources. Instead, the typical outcome was coexistence of resistant and sensitive bacteria, with the latter more efficient at exploiting resources and thus able to sustain the phage's persistence [2,3]. A study of cyanobacteria and their phages in the marine environment also supported this pattern [4].
On the other hand, Lenski and Levin also pointed out that bacteria would lose the upper hand if the phage targeted a receptor that was essential for the bacteria to survive in their current environment. They cited then-recent work by Williams Smith and Huggins [5,6], who showed that they could successfully treat mice with otherwise lethal bacterial infections using a phage that specifically targeted a receptor required for the bacteria to colonize the mice. As the problem of bacterial resistance to antibiotics has grown, similar strategies are now being tested in which phage that specifically target drug efflux pumps are deployed as therapeutic agents [7][8][9]. In the meantime, yet other forms of bacteria-phage coevolution have been discovered, including CRISPR systems in bacteria and countermeasures to avoid these defences in phage [10][11][12][13].
Another part of the argument that bacteria had the upper hand in the coevolutionary arms race depended on the idea that, while phages could often counter minor mutations in OPEN ACCESS receptors, it was much more difficult for them to evolve the ability to use another receptor if the bacteria simply stopped producing the usual receptor [1]. However, more recent work has shown that some host-phage pairs can undergo longer coevolutionary cycles involving defences and counter-defences at the outer membrane [14][15][16], and some phages can evolve to use new receptors even on a short time scale [17]. This coevolutionary dynamic -in which hosts lose structures exploited by specific pathogens, and those pathogens evolve to exploit alternative structures -is called inverse gene-for-gene (IGFG) coevolution [18][19][20][21]. This IGFG framework is useful for representing changes in coevolving communities of bacteria and phage (Fig. 1). For example, if phage cannot evolve to exploit new features after bacteria have evolved resistance, then phage populations may be evolutionarily static [22,23]. Conversely, if phage exploit essential features of the bacteria that cannot be eliminated, then the host's evolution is constrained and phage infectivity may remain elevated [6,8]. Our study builds on one such example of IGFG coevolution, in which it was discovered that populations of a virulent strain of phage λ often evolved the ability to use another outer-membrane receptor after coevolving E. coli reduced their expression of the receptor that the phage had initially exploited [17,24].
Phage λ requires a two-step infection process to cross the outer and inner bacterial membranes (Fig. 1). The λ tail initiates infection at the outer membrane of the cell, where its J protein fibres adsorb to the bacterial protein LamB [25,26]. The tail proteins V and H allow λ to enter the periplasm and thereby interact with the mannose permease proteins (encoded by manY and manZ) in the inner membrane, which λ uses to eject its genome into the cytoplasm [27][28][29][30]. Resistance to λ can occur by blocking λ's entry at either the outer or inner membrane, with resistance mutations typically mapping to lamB, lamB's positive regulator malT [25,26], or the mannose permease genes [27,28,30] (Fig. 1). It has been shown that sensitive E. coli and lytic λ can coexist, along with resistant E. coli mutants, in both continuous [31] and batch culture regimes [17]. Previous analysis of this coevolving system has revealed IGFG dynamics focused on outer-membrane defences and counter-defences. That is, E. coli often first evolves malT mutations that reduce LamB expression, resulting in increased resistance to λ [17,31,32], and λ then regains infectivity through mutations in the J gene that increase its adsorption rate and fitness [31,33]. In some, but not all, experiments, specific sets of J mutations allow the novel exploitation of a second outer-membrane protein, OmpF, catalyzing further evolution including mutations in the ompF gene [17,34].
Despite extensive knowledge about the evolution of the initial (adsorption) and final (lysis) steps of λ infection of E. coli, much less is known about the evolution of the genetic networks during other stages of infection, including λ's passage through the periplasmic space and the ejection of its DNA into the host cytoplasm. Meyer et al. [17] found that E. coli coevolving with λ often acquired mutations that affected their ability to grow on mannose, which presumably were favoured because they disrupt entry of the phage genome via the mannose permease. In this study, we examine how this coevolutionary arms race -previously focused on the cell's outer membrane -also set off an arms race involving the host's inner membrane, including the mechanism λ uses to eject its DNA through that membrane and into the bacteria's cytoplasm.

Bacteria and phage strains
Meyer et al. [17] founded 96 replicate cultures with E. coli B strain REL606 and lytic phage λ cI26, serially passaged the communities for 20 days, and froze mixed-community samples daily. Some of the phage populations evolved the ability to use the outer-membrane protein OmpF as a receptor, some of the bacterial populations evolved mutations that affected mannose metabolism and some communities changed in both respects. We obtained phage isolates from two of the populations ( Table 1, Pop-A and Pop-B) that changed in both of these key respects; in each case, however, the isolates were taken 4 days before the phage had evolved the new ability to use the OmpF receptor (Table 1, Supplementary Results and Discussion). E. coli K12 strains BW25113, JW1807 and JW1808 are from the Keio collection [35]. REL606 ∆manZ was constructed using a two-step allelic exchange (Supplemenatry Methods, Tables S1 and S2).

Phage growth assays
We measured the population growth of the ancestral and evolved phages under the same culture conditions as those in which the communities evolved (Supplementary Methods) [17]. The initial densities were ~9×10 6 cells ml −1 and ~1×10 4 phage ml −1 . We calculated the phage's net population growth as the ratio of its final density after 1 day to its initial density; we show the resulting net growth on a log 10 -transformed scale. We enumerated the initial and final phage populations using dilution plating and soft agar

Impact Statement
Laboratory studies of coevolution help us understand how host defences and pathogen counter-defences change over time, which is often essential for predicting the future dynamics of host-pathogen interactions. One particular model, termed inverse gene-for-gene coevolution, predicts that coevolution proceeds through alternating steps, whereby hosts lose the features exploited by pathogens, and pathogens evolve to exploit alternative features. Using a classic model system in molecular biology, we describe the nature and timing of a previously overlooked step in the coevolution of E. coli and bacteriophage lambda. Our work demonstrates that this mode of coevolution can profoundly reshape the interactions between bacteria and phage. In both scenarios, host alleles affect selection on pathogen phenotypes, and pathogen alleles influence selection on host phenotypes. However, the two models have different implications for understanding historical coevolution and predicting future changes. During GFG coevolution, hosts evolve resistance by gaining resistance genes, and pathogens evolve by losing genes that elicit host defences. GFG coevolution is common among plants and their bacterial pathogens; it may also occur in bacteria-phage interactions that involve restriction-modification and CRISPR defences. During IGFG coevolution, pathogen infectivity requires the exploitation of specific host features, and resistance involves eliminating the exploited features. Unlike in the GFG model, host defences in the IGFG model do not require pathogen recognition, and the pathogen's evasion of host resistance does not require the loss of a defence elicitor.  overlays (Supplementary Methods). We performed five or six replicate assays for each phage-host combination shown in Fig. 2.

Frequency of mutants with altered mannose phenotypes
We estimated the frequency of bacteria with mutations affecting the mannose permease by plating from the time series of frozen samples taken from populations Pop-A and Pop-B on tetrazolium mannose agar, as done previously [17]. Mutants with reduced ability to metabolize mannose form deeply pigmented colonies that can be readily distinguished from those of the ancestral strain REL606, which forms light pink colonies on that medium.
The data are available as Supplementary Datasets S1 (net population growth of phage λ on wild-type and knockout bacteria) and S2 (temporal dynamics of man mutants in E. coli populations).

RESULTS AND DISCUSSION
Our experiments focus on two independently coevolved communities of mixed E. coli and λ populations, designated Pop-A and Pop-B [17]. Both λ populations evolved from a common ancestral phage (strain cI26). From each evolved population, we isolated a single phage clone: λ-A from Pop-A and λ-B from Pop-B (Table 1). Each clone was isolated 4 days before its population evolved the ability to use the OmpF receptor; hence, the phage clones were isolated on different days of the coevolution experiment performed by Meyer et al. [17].
To examine whether and how coevolution affected λ's dependence on the ManY and ManZ proteins, we measured the population growth of the ancestral (cI26) and the two coevolved phage isolates (λ-A and λ-B) on bacterial strains with and without the manY and manZ genes (Table 1). Both the ancestral and evolved phage isolates grew well on bacterial strains with intact manY and manZ genes, including both the ancestral E. coli B strain, REL606, used in the coevolution experiment, and the K12 genetic background in which the Keio collection was made (Fig. 2, Table 1). Deletion of either the manY or manZ gene in either background severely reduced the ancestral phage's population growth. In two cases (REL606 ∆manZ and Keio ∆manY), we saw no growth whatsoever in the ancestral phage (cI26) population after 24 h; in the other case (Keio ∆manZ), the ancestral phage population increased ~10-fold, but that was 5 orders of magnitude less than the increase on the same background with both mannose permease genes present. In striking contrast, both evolved phage isolates showed substantial growth on all three bacterial strains that lacked either the manY or manZ gene (Fig. 2). These results thus indicate an inverse gene-for-gene coevolutionary interaction at the inner membrane. That is, the bacteria modified or lost the mannose permease, which the ancestral phage used to eject its genome into the cytoplasm, and the phage countered by evolving independence of that function.
To determine when the mutant mannose permease mutants arose in the two E. coli populations studied here, we plated frozen samples from the coevolution experiments on tetrazolium mannose agar, on which man mutants form pigmented colonies distinguishable from the wild-type (Supplementary Results and Discussion) [36]. We are particularly interested in the timing of the appearance of the man mutants relative to two other steps in the coevolutionary arms race that were previously characterized: (i) the malT mutations that reduced the bacteria's expression of LamB and thus the adsorption of the ancestral phage [33]; and (ii) λ's new ability to adsorb to OmpF as an alternative receptor [17]. Our phage growth data demonstrate that manY and manZ deletions confer substantial resistance even to the ancestral phage, which can use only the LamB surface receptor (Fig. 2). That result suggests the possibility that the man mutants could have arisen early in the coevolution experiments, perhaps alongside or even before the malT mutations that provided resistance at the outer membrane. However, time course data show that the man alleles consistently reached high frequencies (above the detection limits, shown as grey dashed lines in Fig. 3) only after the fixation of the malT mutations, which occurred by day 8 in both populations studied here (Figs. 3 and 4, Table  S3) [17].
These temporal data also show that the man mutations had nearly fixed in both bacterial populations (frequencies >95 % on day 10 in Pop-A and on day 12 in Pop-B), but then the mutants sharply declined the next day. This reversal suggests that these mutants were killed by phages that evolved independence of the mannose permease, and it is consistent with previous data Columns indicate bacterial genotypes with four exploitable features, and rows indicate λ genotypes that exploit those features: mal, maltose transport across the outer membrane; man, mannose transport across the inner membrane; ompF, glucose and electrolyte transport across the outer membrane; imx, a hypothetical inner membrane feature that is exploited by λ that evolved independence of the mannose permease. The '+' symbol indicates that either the bacteria have the feature or the phage exploit the feature. The '-' symbol indicates that the bacteria lack the feature, express it to a reduced degree, or otherwise modify it to minimize phage infection. Asterisks (*) indicate infectivity for each host-phage pair, with more asterisks indicating greater infectivity. Adaptive changes through the network can proceed by two types of move: E. coli resistance (to the right along rows) and increased λ infectivity (downward along columns). The coevolving communities were founded by host genotype a and phage genotype vi (shown by the black circle). The communities analysed in this study appear to have moved through the shaded nodes in five steps, as indicated by the arrows.
showing that mutant man alleles rarely fixed in the bacterial populations [17]. Meyer et al. (Fig. S2 in [17]) reported that the bacterial population densities remained high (~2×10 9 cells ml −1 , near the carrying capacity of the medium) throughout this period of the evolution experiment. Therefore, the mutant frequencies that we observed (Fig. 3) correspond to ~4×10 7 cells ml −1 (about 2 % of the total population, the limit of detection in that assay) to almost 2×10 9 cells ml −1 (the carrying capacity).
With such large population sizes, any phage mutants that gained the ability to infect the man mutants would have access to a large number of hosts, and correspondingly, a large fitness benefit. The resulting growth of the man-independent phage population would drive the frequency of man mutants down, especially if the man-independent phages preferentially infected and killed the man mutants relative to other cells that retained the wildtype permease. Fitness costs associated with loss of the mannose permease may also have contributed to the reversal, although the costs of the resistance mutations are small compared to their benefit in the presence of phage [36].
In host Pop-A, variation in colony morphology further suggested that different man alleles were present before and after the sudden decline in the frequency of man mutants on day 11 (Fig. 4, Supplementary Results and Discussion). The initial boom and bust of the mutant man alleles in both populations also occurred before the phage had evolved to use OmpF (Fig. 3, dashed arrows). Whether λ gained independence from the mannose permease by exploiting another inner-membrane protein, and whether E. coli did (or could) respond by eliminating such a structure, are interesting questions for future work.
Our results are broadly consistent with genetic and molecular biology studies of λ host range mutations. Scandella and Arber [30] isolated E. coli mutants that allowed phage adsorption to the cell envelope but interfered with ejection of the phage genome, thereby reducing infection success to a small fraction of that observed on wild-type cells. The responsible mutations were mapped to the mannose permease operon [27,37], and λ mutants that could infect these mutant bacteria had mutations in phage genes V or H [38]. Mutations in V and H have also been observed in another population in this study system [39].
Williams et al. [37] found that, for E. coli strain K12, manZ is not strictly required for wild-type λ to eject its genome, and our results are in accord with that finding (Fig. 2, Keio background). However, our results suggest that λ cI26 does require manZ when infecting E. coli strain B, at least in the culture conditions that we used (Fig. 2, REL606 background). Alternatively, λ cI26 might occasionally infect and replicate in hosts without manZ, but at a rate that is offset by the decay or inactivation of free virus particles under these conditions [17,33]. In any case, the net population growth of the ancestral phage on either the ∆manY or ∆manZ bacteria is insufficient to offset the 100-fold daily dilutions (Fig. 2, dashed line) that took place during the coevolution experiment [17].
Taken together, our results imply that E. coli and λ coevolved in an inverse gene-for-gene manner [18] ( Fig. 1). This coevolution involved two infection steps -crossing first the outer and then the inner membraneand at least three, and probably four, distinct host features (Figs 1, 5). E. coli evolved resistance to phage λ through the loss or alteration of maltose transport across the outer membrane (via mutations in malT) and mannose transport across the inner membrane (via mutations in manY or manZ), while λ evolved to exploit other E. coli features, including another outer-membrane protein (OmpF) and, presumably, some as yet unidentified, alternative inner-membrane protein (shown as encoded by the hypothetical imx gene in Fig. 5). While our study addresses one particular bacteria-phage interaction in a simple laboratory setting, it illustrates the extent to which the resulting coevolutionary arms races can be richer and more complex than is often appreciated.
There are many alternative coevolutionary paths through an inverse gene-for-gene network that has four features subject to host defences and parasite counter-defences (Fig. 5). This multiplicity of potential paths suggests that mutation and selection could drive replicate communities to different regions of the coevolutionary landscape, raising other interesting questions.
How might different first-step resistance mutations affect the subsequent host range evolution of the phage and the further evolution of host resistance? To what extent can IGFG systems continuously evolve host defences and parasite counterdefences? What is the effect of such prolonged coevolution for community diversity? Do communities become increasingly divergent as the coevolving populations follow different paths through the network, or might they eventually converge on the same phenotypic states after a period of divergence? How important are evolutionary innovations in opening new paths, relative to pleiotropic tradeoffs that may close off certain paths? Future work should investigate these and other questions about the coevolution of bacteria and phage and the structure of their genetic interaction networks.

Funding information
This work was supported by the National Science Foundation Graduate Research Fellowship (DGE-1424871) to A. R. B., the BEACON Center for the Study of Evolution in Action (NSF Cooperative Agreement DBI-0939454), the John Hannah endowment from Michigan State University to R. E. L., a Marie Curie IIF Fellowship to J. G., and the Max Planck Society to J. G. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Any opinions, findings, and conclusions or recommendations expressed in this paper are those of the authors and do not necessarily reflect the views of the funders.