The Ecology and Evolution of Microbial Defense Systems in Escherichia coli

Margaret A. Riley*, John E. Wertz, and Carla Goldstone

[Section Editor: Monica Riley]

Posted February 27, 2004

Department of Ecology and Evolutionary Biology, Yale University, New Haven, CT 06511

*Corresponding author: Phone: (203) 432-3875, Fax (203) 432-2374, E-mail: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

WHAT ARE MICROBIAL DEFENSE SYSTEMS?

Microbes produce an extraordinary array of microbial defense systems. These include the broad-spectrum classical antibiotics so critical to human health concerns; metabolic by-products, such as the lactic acids produced by lactobacilli; lytic agents, such as lysozymes found in many foods; and numerous types of protein exotoxins and bacteriocins. The abundance and diversity of this biological arsenal is clear. Lactic acid production is a defining trait of lactic acid bacteria (19). Bacteriocins are found in almost every bacterial species examined to date, and within a species, tens or even hundreds of different kinds of bacteriocins are produced (23, 41). Halobacteria universally produce their own version of bacteriocins, the halocins (55). Streptomycetes commonly produce broad-spectrum antibiotics (46). It is clear that microbes invest considerable energy in the production and elaboration of antimicrobial mechanisms. What is less clear is how such diversity arose and what roles these biological weapons play in microbial communities.

One family of microbial defense systems, the bacteriocins, has served as a model for exploring evolutionary and ecological questions. In this review, current knowledge of how the extraordinary range of bacteriocin diversity arose and is maintained in one species of bacteria, Escherichia coli, will be assessed, and the role these toxins play in mediating microbial dynamics will be discussed.

BACTERIOCINS: THE MICROBIAL WEAPON OF CHOICE

Bacteriocins are loosely defined as biologically active protein moieties with a bacteriocidal mode of action (23, 51). They differ from traditional antibiotics in one critical way: they have a relatively narrow killing spectrum and are toxic only to bacteria closely related to the producing strain. These toxins have been found in all major lineages of Bacteria and, more recently, have been described as universally produced by some members of the Archaea (55). According to Klaenhammer and Raloff, 99% of all bacteria may make at least one bacteriocin, and the only reason we have not isolated more is that very few researchers have looked for them (26, 35).

The bacteriocin family includes a diversity of proteins in terms of size, microbial targets, modes of action, and immunity mechanisms. The most extensively studied, the colicins produced by E. coli, share certain key characteristics (1, 2, 4, 16, 22, 27, 33). Colicin gene clusters are carried on plasmids and are composed of a colicin gene, which encodes the toxin; an immunity gene, which encodes a protein conferring specific immunity on the producer cell; and a lysis gene, which encodes a protein involved in colicin release from the cell. Colicin production is mediated by the SOS regulon and is therefore principally produced in times of stress. Toxin production is lethal for the producing cell and any neighboring cells recognized by that colicin. A receptor domain in the colicin protein that binds a specific cell surface receptor determines target recognition. This mode of targeting results in the relatively narrow phylogenetic killing range often cited for bacteriocins. The killing functions range from pore formation in the cell membrane to nuclease activity against DNA, rRNA, and tRNA targets. Colicins—indeed, all bacteriocins produced by gram-negative bacteria—are large proteins. Pore-forming colicins range in size from 358 to 626 amino acids. Nuclease bacteriocins also have a broad size range and vary from 551 to 777 amino acids.

Although colicins are representative of gram-negative bacteriocins, there are intriguing differences within this subgroup of the bacteriocin family. E. coli carries its colicins exclusively on plasmid replicons (34). The nuclease pyocins of Pseudomonas aeruginosa,which share recent ancestry with colicins, and other as yet uncharacterized bacteriocins are found exclusively on the chromosome (48). Other close relatives of the colicin family, the bacteriocins of Serratia marcesens, are found on both plasmids and chromosomes (9, 12, 17).

Many bacteriocins isolated from gram-negative bacteria appear to have been created by recombination between existing bacteriocins (2, 28, 36, 45). Such frequent recombination is facilitated by the domain structures of bacteriocin proteins. In colicins, the central domain comprises ~50% of the protein and is involved in the recognition of specific cell surface receptors. The N-terminal domain (~25% of the protein) is responsible for translocation of the protein into the target cell. The remainder of the protein houses the killing domain and the immunity region, which is a short sequence involved in immunity protein binding. As we shall explore further below, the conserved domain configuration of these toxins is responsible for much of the bacteriocin diversity we find in nature.

EVOLUTION OF BACTERIOCIN DIVERSITY

Colicins as a Model for Evolutionary Studies

The colicins and other enteric bacteriocins, such as klebicins and alveicins, remain the only bacteriocins for which detailed evolutionary investigations have been undertaken. Among the colicins there are two main evolutionary lineages, which also distinguish the two primary modes of killing: pore formation and nuclease activity (39). The driving force behind colicin evolution appears to be positive selection for an altered killing spectrum. This can be realized either by altering the translocation or binding domain so that new target cells are recognized or by alterations in the killing domain that render formerly immune cells susceptible. Several studies have examined colicin evolution in an effort to elucidate the relative roles of recombination, positive selection, and genetic drift as mechanisms for generating diversity in colicin-like bacteriocins. These studies include DNA and protein sequence comparisons (2, 36), surveys of DNA sequence polymorphism in natural isolates (32, 43, 52), experimental evolution (14, 53), and mathematical modeling (14), and they have revealed two primary modes of colicin evolution (54).

The Role of Diversifying Recombination in Colicin Evolution

Recombination has been shown to be an important force in colicin evolution both through domain swapping within the colicin gene and by moving the entire colicin operon to novel backgrounds. The more abundant pore former colicins are generated by domain shuffling within colicin genes and by moving entire colicin operons (2, 54). All characterized pore former colicin proteins share one or more regions with high levels of sequence similarity to other pore former colicins (Fig. 1). This patchwork of shared and divergent sequences suggests frequent recombination. The locations of the different patches frequently correspond to the different functional domains of the proteins.

Fig. 1Pairwise comparisons of pore-forming colicin protein sequences. The values below each comparison indicate the sequence identity for the region indicated. The colicin proteins are not drawn to scale.

Evidence of domain swapping is seen frequently in colicin genes; for example, colicins Ia and Ib share translocation and binding domains but contain different killing domains (52, 54). Klebicin B, the first colicin-like plasmid isolated from Klebsiella pneumoniae, appears to have undergone recombination both in the plasmid backbone and in the bacteriocin genes themselves (42) (Fig. 2). Klebicin B is a nuclease klebicin that shows sequence similarity to both colicin A-like pore former sequences and pyocin S1-like nuclease sequences (42). The similarity of DNA sequences flanking the operon to sequences from pColA and pColE9, isolated from E. coli, and pyocin S1, found on the chromosome of P. aeruginosa, suggests that the klebicin B operon is a chimera whose history involves movement from a colicin A-like plasmid into a Klebsiella-specific plasmid backbone and at some point acquiring a colicin E9- or pyocin S1-like killing domain and immunity gene. Perhaps the most striking illustration of this form of functional diversification is the case of colicins 5, 10, and K: the binding and translocation domains of colicin 5 are nearly identical to those of colicin 10, whereas its killing domain is virtually identical to that of colicin K (31). The first bacteriocins isolated from Hafnia alvei, alveicins A and B, appear to be chimeras consisting of translocation domains from Tol-dependent colicins, such as E2 and DF13; unique binding domains; and killing and immunity domains similar to those of colicin Ia (J. E. Wertz and M. A. Riley, in press).

Fig. 2Patterns of sequence similarity in klebicins suggest recombination. The chimeric nature of the pKlebB plasmid sequence is indicated by different shading patterns. The key shows regions of sequence similarity to other bacteriocin gene clusters and plasmids. pKlebB illustrates a pattern typical of other bacteriocin-encoding plasmids in which sequences encoding plasmid functions are similar to sequences found in other plasmids segregating in the host species, whereas those sequences composing and flanking the bacteriocin gene cluster show similarity to bacteriocin sequences from other species.

Several E2 colicins isolated in Australia suggest that diversifying recombination is not restricted to pore former colicins (52). Half of the E2 producers carry the characterized E2 plasmid. The other half carry a recombinant plasmid with sequences derived from colicin E7 and the characterized E2 plasmid. The influence of diversifying recombination is also not limited to the closely related bacteriocins of enteric bacteria. As mentioned above, the S pyocins of P. aeruginosa are the result of recombination between several pore former and nuclease colicins and other, as yet uncharacterized, bacteriocins (47, 48). Even altering the domain structure of the protein, as is seen for pyocins, which have switched the receptor recognition and translocation domains relative to the order found in colicins, has not limited the influence of diversifying recombination.

The Role of Diversifying Selection in Colicin Evolution

An alternative mode of evolution is responsible for the current diversity of nuclease colicins (44, 54). These colicins, which include both RNase and DNase killing functions, share a recent common ancestry. Their DNA sequences are quite similar, ranging from 50 to 97% sequence identity. However, many pairs of nuclease colicins have elevated levels of divergence in the immunity region (Fig. 3). To explain this pattern of divergence, Riley and Tan have proposed a two-step process of mutation and selection (36, 40, 54). First, a point mutation in the immunity gene confers broadened immunity on the host cell, so that it is immune to the colicin itself, its immediate ancestor, and potentially other colicins in the environment. This broadened immunity provides a fitness advantage to the cell in environments where multiple colicins are present and thus allows time for the accumulation of a second mutation, this time in the colicin’s killing domain so that it no longer binds the ancestral immunity protein.

Fig. 3Graph indicating the average number of total nucleotide substitutions between pairs of nuclease-type colicin gene clusters (colicin pairs E2-E9 and E3-E6). Most of the divergence between colicins occurs in the immunity region of the gene cluster (composed of the immunity gene and the immunity binding region of the colicin gene).

This combination of mutations results in a "super-killer" phenotype, which enjoys nonreciprocal immunity to its immediate ancestral colicin. Strong positive selection for cells encoding the evolved super-killer colicin will allow them to displace cells carrying the ancestral colicin. This evolved colicin will ultimately be replaced by yet another novel colicin as the cycle repeats itself. Repeated rounds of this form of diversifying selection will result in the accumulation of a disproportionately high ratio of nonsynonymous to synonymous nucleotide substitutions in the immunity gene and the killing domain of the colicin gene. The nucleotide substitution patterns seen in pairwise comparisons of colicins E3-E6 and E2-E9 best fit this model of colicin evolution (39, 40, 54). The level of sequence divergence, as well as the ratio of nonsynonymous to synonymous substitutions per site, increases dramatically in the killing–immunity-binding domain of the colicin genes experiencing diversifying selection. This is demonstrated graphically in a sliding-window analysis of colicins E2 and E9 (Fig. 4).

Fig. 4Sliding-window pairwise comparison of colicins E2 and E9. Calculations were performed on pairs of aligned nucleotide sequences using a sliding window of 40 bp and a step size of 1 base. Gaps in aligned sequences were deleted from both sequences; there were no ambiguous bases. % Identity, percentage of nucleotide matches per total number of sites compared; ds, percentage of synonymous substitutions per synonymous sites; dn, percentage of nonsynonymous substitutions per nonsynonymous sites. The values are not corrected for multiple substitutions per site. The x axis is the number of the first nucleotide in the comparison window.

Recently, the DNA sequence of a new pore former colicin, Y, was determined (37). Colicin Y is a close relative of colicin U, another pore former colicin isolated from a different continent and over 20 years earlier (49). This pair of colicins has a pattern of DNA substitution identical to that observed among the nuclease colicins, with an elevated level of substitution in the immunity region. This observation suggests that the process of diversifying selection is not restricted to nuclease colicins.

A Two-Step Process of Colicin Evolution

Riley has developed a model of colicin diversification that involves two phases (39). When colicins are rare, as is the case for most known nuclease colicins, the occurrence of point mutations that alter immunity function may be the primary mode for generating novel bacteriocin phenotypes. Novel immunity and killing functions are rapidly selected for, since they allow a cell to avoid being killed by other bacteriocins or to displace its ancestors. These novel bacteriocins are then maintained until a new immunity or killing function emerges. When colicins are abundant, as is the case for many of the known pore former colicins, domain swapping may become a more frequent mode of diversification. This "switch" in evolutionary mechanism is due simply to the requirement for a set of bacteriocins to be abundant enough to serve as templates for recombination. Once they are abundant, recombination can more rapidly generate additional diversity.

We have just begun to tap the diversity of enteric bacteriocins. However, recent work suggests that similar evolutionary mechanisms may play a role in the diversification of other enteric bacteriocins. Sequence comparisons reveal that in several cases enteric bacteriocins are chimeras of known gram-negative bacteriocins (42; M. A. Riley, M. Chaven, C. Goldstone, unpublished data). Finally, some new enteric bacteriocins have no similarity to those characterized previously.A particularly interesting example of this observation is the recently described colicin Js (50). This plasmid-borne bacteriocin has a typical colicin gene cluster composition, with toxin, immunity, and lysis genes. However, the organization of the gene cluster is unique in that the lysis gene is transcribed 5' to the toxin gene. The genes themselves show no similarity to any known bacteriocin genes, and the encoded toxin is 94 amino acids long, which is smaller than any other described colicin.

Bacteriocin-encoding plasmids, like pColJs (which encodes colicin Js) and pKlebB (which encodes klebicin B), demonstrate another aspect of bacteriocin evolution (42, 50). These bacteriocin plasmids are chimeras with a plasmid "backbone" comprising replication and maintenance sequences typical of plasmids found in the bacteriocins’ host species. In the case of pKlebB, isolated from K. pneumoniae, the plasmid contains sequences similar to pNBL63 (57) and pJHC-MW1 (7), isolated from Klebsiella oxytoca and K. pneumoniae, respectively, and encoding plasmid maintenance functions. The sequence surrounding and comprising part of the klebicin B gene cluster shows similarity with colicins A and E9, originally isolated from E. coli (42). In the case of pColJs, the plasmid backbone is virtually identical to that of pColE1, whereas the DNA flanking the colicin Js gene cluster shows high similarity to that of pPCP1 from Yersinia pestis (20). The colicin Js gene cluster itself has a significantly lower G+C content (33.6%) than the rest of the plasmid (52.9%), indicating that it originated from yet a third source (50), perhaps even outside of the Enterobacteriaceae. This type of recombination, while it does not alter the bacteriocin genes proper, results in an increased host range.

As more bacteriocins from diverse enteric species are characterized, more pieces of this evolutionary puzzle fall into place. The picture that is starting to emerge is one in which colicin-like bacteriocins are composed of functional modules drawn from natural plasmid libraries, where one of the functional modules is the plasmid backbone itself. If recombination between functional modules occurs via homologous recombination, as is suspected, then there would have to be either a conserved universal interdomain sequence where recombination occurs or a "linkage network" connecting subsets of modules that can recombine with each other. There is no evidence for the existence of a universal conserved sequence between domains. If such a linkage network exists, then the central binding domain would govern which killing domain could be associated with a given translocation domain. Additional research in this area is necessary to determine the sizes and compositions of these module libraries, as well as the linkage relationships between specific families of modules, if any exist.

Ecological Role of Colicins

Without question, bacteriocins serve some function in microbial communities. This statement follows from the detection of bacteriocin production in all surveyed lineages of prokaryotes. Equally compelling is the inference of strong positive selection acting on enteric bacteriocins. Such observations argue that these toxins play a critical role in mediating microbial population or community interactions. What remains in question is what, precisely, that role is.

Bacteriocins may serve as anticompetitors, enabling the invasion of an established microbial community by a strain. They may also play a defensive role and act to prohibit invasion of an occupied niche by other strains or species or to limit the advance of neighboring cells. An additional role has recently been proposed for gram-positive bacteriocins: mediating quorum sensing (30).It is likely that whatever roles bacteriocins play, those roles change as components of the environment, both biotic and abiotic, change.

Theoretical and Experimental Studies of Colicin Ecology

Early experimental studies of the ecological roles of bacteriocins were inconclusive and contradictory (5, 10, 13, 18, 21, 24, 56). More recently, a theoretical and empirical base has been established that has defined the conditions that favor the maintenance of toxin-producing bacteria in both population and community settings. Almost exclusively, these studies have modeled the actions of colicins. Chao and Levin showed that the conditions for invasion of a colicin producer strain were much broader in a spatially structured environment than in an unstructured one (3). In an unstructured environment with mass action, a small population of producers cannot invade an established population of sensitive cells. This failure occurs because the producers pay a price for toxin production—the energy costs of plasmid carriage and lethality of production—but the benefits—the resources made available by killing sensitive organisms—are distributed at random. Moreover, when producers are rare, the reduction in growth rate experienced by the sensitive strain (owing to extra deaths) is smaller than the reduction felt by the producer (owing to its costs); therefore, the producer population goes extinct. In a physically structured environment, such as on the surface of an agar plate, the strains grow as separate colonies. Toxin diffuses out from a colony of producers, killing sensitive neighbors. The resources made available accrue disproportionately to the producing colony, owing to its proximity, and therefore killers can increase in frequency even when they are initially rare.

The RPS Model

Recent modeling efforts have incorporated additional biological realities. Two such efforts introduced a third strain, one that is resistant to the toxin but cannot itself produce it (6, 25). The authors of both studies reasonably assume that there is a cost of resistance and that this cost is less than the cost of toxin production borne by the killer strain (11). Owing to this third member, pairwise interactions among the strains have the nontransitive structure of the childhood game of rock-paper-scissors (RPS) (Table 1) (29). The producer strain (P) beats the sensitive strain (S), owing to the toxin’s effects on the latter. The sensitive strain beats the resistant strain (R), because only the latter suffers the cost of resistance. Finally, the resistant strain wins against the producer, because the latter bears the higher cost of toxin production and release while the former pays only the cost of resistance. In an unstructured environment, this game allows periodic cycles in which all three types coexist indefinitely, but each with fluctuating abundance. In a structured environment, this game permits a quasistable global equilibrium, one in which all three strains can persist with nearly constant global abundance (6).

Table 1Chemical warfare among microbes as a nontransitive three-way game similar to the RPS game

Further effects of evolution were incorporated into the model of Czárán et al. by allowing as many as 14 distinct systems of toxin production, sensitivity, and resistance, along with the genetic processes of mutation and recombination that can alter these traits and their associations (6). The permutations of these systems permit the existence of several million different strains. From this additional complexity emerge two distinct quasiequilibrium conditions, the "frozen" and "hyperimmunity" states. In the frozen state, all the toxins are maintained globally, but most colonies are single-toxin producers. That is, each colony produces one toxin, to which it is also immune. By contrast, in the hyperimmunity state, many colonies produce no toxin, many others make one, and still others produce several toxins, but only a few produce most of the available toxins. Resistance shows a very different distribution, with all of the colonies being resistant to most or all of the toxins. Which of these two outcomes is obtained depends upon the initial conditions. If the evolving system begins with the entire population sensitive to all of the toxins, then the frozen state results. The hyperimmunity state is reached if the system starts with enough diversity that most colonies already have multiple killer and resistance traits. Theoretical studies reveal that a combination of all three phenotypes (P, S, and R) provides a stabilizing factor in a mixed, spatially structured assemblage (6, 8, 25). This complex of three strains has the same formal structure as the RPS game (25), with P beating S (by killing), S beating R (by growth rate advantage), and R beating P (by growth rate advantage).

Numerous surveys of colicin production in natural populations suggest that populations of E. coli may closely match predictions of the model of Czárán et al. (15, 41). In E. coli, producer strains are found at frequencies ranging from 10 to 50%. Resistant strains are even more abundant and are found at frequencies from 50 to 98%. In fact, most strains are resistant to all cosegregating colicins. Finally, there is a small population of sensitive cells. Figure 5 provides a summary of phenotype distributions in a population of E. coli isolated from wild field mice in Australia (15). The model of Czárán et al. predicts that this distribution of phenotypes will result from frequent horizontal transfer of resistance and the significant cost to colicin production (6). In other words, if a strain can gain resistance and lose production, it will over time—just as was observed in the E. coli isolates from the field mouse population over the course of a summer (15).

Fig. 5Survey of colicin production and resistance in E. coli. Over 400 strains were isolated from two populations of feral mice in Australia over a period of 7 months. The isolates were scored for colicin production and resistance. (a) Colicin production is abundant, with just under 50% of the strains producing eight distinct colicin types. Col ⁻, nonproducer strains. (b) The majority of isolates are resistant to most co-occurring colicins. (c) A small proportion of the population is sensitive to co-occurring colicins.

The Killing Breadth of Colicins

We have assumed that colicins play a role in mediating within-species (or population level) dynamics. This assumption is based upon the narrow killing range exhibited by most colicins. However, recent work calls this assumption into question. Bacteriocins from natural isolates of several species of enteric bacteria were assayed for their killing effect against a large set of nonproducers isolated from the same sources (38). Figure 6 reveals that, contrary to expectations, the killing breadth varies significantly for different bacteriocins. Some are clearly most effective at killing within the producer strains’ own species. Others kill more broadly or kill quite specifically isolates of a different species. This diversity of killing breadth argues that bacteriocins, contrary to prior suggestions, play equally compelling roles in mediating both population level and community level interactions. A more thorough understanding of how bacteriocins function awaits the development of a more biologically realistic experimental approach. Prior studies have considered how producer, sensitive, and resistant strains within the same species interact. If the goal is to understand the roles these toxins play in nature, our experiments must incorporate more complex microbial communities and environments.

Fig. 6Phylogenetic breadth of bacteriocin killing. The killing spectrum of each class of bacteriocins was cross-referenced with a phylogenetic tree of the enteric species they were screened against. The heights of the solid boxes are proportional to the percentage of strains sensitive to each class of bacteriocin. Bacteriocins were screened against 40 natural isolates from each enteric species. The molecular phylogeny of a subset of enteric bacteria is based on a composite of five housekeeping genes ( gapA, groEL, gyrA, ompA, and pgi) and 16S ribosomal sequences. The tree is rooted using Vibrio cholerae as an outgroup. Abbreviations : KO, K. oxytoca; KP, K. pneumoniae; EB, Enterobactercloacae; CF, Citrobacter freundii; EC, E. coli; SM, S. marcescens; HA, H. alvei, VC, Vibrio cholerae.