Progress towards understanding the fate of plasmids in bacterial communities

Abstract

Plasmid-mediated horizontal gene transfer influences bacterial community structure and evolution. However, an understanding of the forces which dictate the fate of plasmids in bacterial populations remains elusive. This is in part due to the enormous diversity of plasmids, in terms of size, structure, transmission, evolutionary history and accessory phenotypes, coupled with the lack of a standard theoretical framework within which to investigate them. This review discusses how ecological factors, such as spatial structure and temporal fluctuations, shape both the population dynamics and the physical features of plasmids. Novel data indicate that larger plasmids are more likely to be harboured by hosts in complex environments. Plasmid size may therefore be determined by environmentally mediated fitness trade-offs. As the correlation between replicon size and complexity of environment is similar for plasmids and chromosomes, plasmids could be used as tractable tools to investigate the influence of ecological factors on chromosomes. Parallels are drawn between plasmids and bacterial facultative symbionts, including the evolution of some members of both groups to a more obligate relationship with their host. The similarity between the influences of ecological factors on plasmids and bacterial symbionts suggests that it may be appropriate to study plasmids within a classical ecological framework.

horizontal gene transfer, plasmid ecology, fitness trade-offs, facultative symbionts

Introduction

The horizontal gene pool

The horizontal gene pool (HGP) refers to genetic information that may be accessible to more than one bacterial species, potentially resulting in phenotypes of one being acquired by another. It includes the constituent genes of mobile genetic elements (MGEs) and also genes that, whilst not mobile themselves, may be mobilized by MGEs. Examples of MGEs include plasmids, bacteriophages, conjugative transposons and integrative conjugative elements. The HGP may have profound effects on both the ecology and the evolution of bacteria. Successful horizontal transfer of sequences that confer adaptive traits can significantly alter the ecology of the recipient bacterium, allowing colonization of otherwise hostile niches. An important example of this is the acquisition of antibiotic resistance as a result of plasmid transfer, which has been demonstrated for numerous bacterial genera in diverse habitats, such as wastewater (Mach & Grimes, 1982), animal products (Jayaratne et al., 1989), and the guts of mice (Mus musculus) (Doucet-Populaire et al., 1992) and house flies (Musca domestica L.) (Petridis et al., 2006). Horizontal transfer is also responsible for dissemination of other potentially ecologically significant traits, such as virulence. For example, benign strains of Xanthomonas citri may become pathogenic via acquisition of the plasmid pXcB, to cause citrus canker disease (El Yacoubi et al., 2007). Horizontally transferred sequences that persist also affect bacterial diversity and evolution (Doolittle, 1999; Ochman et al., 2000; Jain et al., 2002; Gogarten & Townsend, 2005). Evidence for widespread acquisition and persistence of horizontally transferred genes comes from regions of genomes that have a discontinuous distribution between closely related species and genera. This may be apparent from unusual G+C content or patterns of codon usage; these are reasonably uniform within species but show significant variation between species. Marked differences in isolated regions of genomes indicate recent horizontal acquisition from distantly related species (Lawrence & Ochman, 1997). Using these methods, Lawrence & Ochman (1998) estimated that up to 18% of the extant chromosome of Escherichia coli may have been acquired via horizontal transfer since its divergence from Salmonella 100 million years ago. That is not to say that 18% of the genetic content of E. coli is currently mobile. Horizontally transferred sequences that become fixed in a chromosome lose their ecological relevance as MGEs and are no longer considered components of the HGP.

Plasmids

Plasmids are amongst the best characterized and most widely recognized members of the HGP. They are a broad class of extrachromosomal elements, capable of horizontal transmission and also of regulating their copy number independently of the host chromosome. Replication is usually via one of two general strategies, rolling circle or theta and strand displacement (del Solar et al., 1998), and horizontal transmission proceeds either via conjugation, transformation or transduction (Thomas & Nielsen, 2005). Other characteristic survival functions of plasmids include loss prevention strategies, such as copy number control and multimer resolution (Summers, 1998), active partitioning systems (Ebersbach & Gerdes, 2005) and postsegregational killing (Hayes, 2003). Plasmids are ubiquitous entities, found in most bacterial species, as well as some archaeal and eukaryotic species, and in virtually all ecosystems, from hydrothermal vents (Prieur et al., 2004) to arctic soils (Fagerli & Svenning, 2005). They have been termed ‘individuality replicons’ (Chain et al., 2006) due to their propensity to encode accessory phenotypes that render their host better able to survive in atypical environments. Examples include virulence, resistance to antimicrobials and heavy metals and the ability to catabolize xenobiotics and other complex carbon sources. However, plasmid carriage is only beneficial to the host if environmental conditions are such that there is positive selection for these plasmid-encoded phenotypes. Also, any fitness advantage is countered by a potentially small but permanent fitness cost associated with plasmid carriage. The combination of these factors dictates the survival, spread or demise of a plasmid in a population. Stable maintenance of plasmid numbers can only be achieved if rates of plasmid loss, via vegetative segregation and/or fitness disadvantages due to the fitness costs of plasmid carriage, are at least equalled by rates of plasmid gain, via horizontal transfer and/or fitness advantages due to plasmid-encoded beneficial phenotypes (Fig. 1).

1

Plasmid survival, spread or demise is dictated by the combined rates of transmission processes (a, b) and selection processes (c, d). Plasmids will only be stably maintained in a population if rates of loss, via vegetative segregation (a) and negative effects on host growth rate due to the fitness costs associated with plasmid carriage (c), are at least equalled by rates of gain, via horizontal transfer (b) and positive effects on host growth rate due to plasmid-encoded beneficial phenotypes (d).

Plasmid population dynamics

Quantitative mathematical approaches to population biology, devised for higher organisms (e.g. Lotka, 1925; Volterra, 1926), were first applied to plasmid population dynamics in the 1970s. Stewart & Levin (1977) produced the first theoretical investigation of the ability of plasmids to establish and spread in natural populations and provided a basis for many subsequent studies (Levin et al., 1979; Levin & Stewart, 1980; Freter et al., 1983; Lundquist & Levin, 1986; Simonsen et al., 1990; Bergstrom et al., 2000). Many of the empirical estimates of process rates used in these models were determined experimentally in batch or chemostat culture using laboratory strains and plasmids constructed using recombinant DNA technology as well as novel host–plasmid combinations. These approaches fail to take into account some of the physiological and ecological complexities of plasmids and their hosts in natural environments. For example, horizontal transfer rates may vary according to the growth stage and physiological state of the bacterium. In log phase, rates may be reasonably approximated by a mass action model but in lag or stationary phase, they may not (Levin et al., 1979). Transitory derepression of conjugative pili synthesis may occur in newly formed transconjugants, increasing rates of horizontal transfer and the potential for plasmids to establish and spread through a population (Lundquist & Levin, 1986). To complicate matters further, the goodness of fit of the mass action model and the effect of transitory derepression is different for bacteria colonizing surfaces than for bacteria in liquids (Simonsen, 1990). Moreover, horizontal transfer rates of some plasmids are affected by external cues, such as cell density or stress. For example, specific opines, which are small carbon compounds produced by crown gall tumours, and a quorum-sensing signal, the acyl-homoserine lactone ligand Agrobacterium autoinducer (AAI), produced by the bacterium itself, regulate conjugal transfer of Ti plasmids in Agrobacterium spp. (Oger & Farrand, 2002). Similarly, enterococcal mating pheromones allow donor cells to regulate expression of conjugal transfer of the plasmid pCF10 in response to recipient cell density (Kozlowicz et al., 2006). R27 and other IncI1 plasmids have an unusual, thermosensitive mode of conjugation, with transfer occurring optimally between 22 and 28 °C but inhibited above 37 °C (Sherburne et al., 2000). Different estimates of context-dependent parameters, such as rates of horizontal transfer, will probably have dramatic effects on plasmid population model predictions.

Since Stewart & Levin's seminal paper, there has been a great deal of work to identify relevant parameters and incorporate more accurate, quantitative estimates of process rates into models of plasmid dynamics in natural environments. Here, we review recent observational, experimental and theoretical studies that have brought us closer to understanding how ecological parameters influence the fate of plasmids in bacterial populations. The focus is particularly on the effect of selection processes (Fig. 1c and d) as these are a frequently overlooked component of plasmid population dynamics (Eisen, 2000), and transmission processes (Fig. 1a and b) have been reviewed comprehensively by others (van Elsas & Bailey, 2002; Sorensen et al., 2005). We will instead take more of an overview of the survival and spread of a plasmid through a recipient population following transfer. We also discuss how the physical form of plasmids, including plasmid size and gene content, is influenced by ecological features. Parallels are drawn between plasmids and a similarly diverse group of biological entities, the facultative bacterial symbionts, and comparisons are made between sizes of plasmid genomes from different environments.

What makes a plasmid a plasmid?

There are three interlinked and defining features of plasmids: separateness (i.e. the potential to be physically distinct and to replicate autonomously) from the chromosome, transmissibility (i.e. the ability to transfer or be transferred as a discrete molecule) and dispensability (i.e. lack of essential genes). Plasmids tend to have discontinuous distributions in their host population (which may be composed of clones or different species and/or genera) due to two of these three general features: they can be horizontally transferred to novel hosts and, as dispensable entities, physically lost from individual hosts.

The enormous diversity of plasmids, in terms of size, structure, transmission, evolutionary history and accessory phenotypes, in many ways mirrors that of another group, the facultative bacterial symbionts. This group of bacteria is capable of either a free-living or a host-associated lifestyle. Well-known examples include Wolbachia spp. in arthropods and Rhizobium and Frankia spp. in plants. The facultative symbionts share all three of the general features of plasmids: horizontal transmission, dispensability and separateness. They similarly have a discontinuous distribution in host populations. For example, although rhizobia have a worldwide distribution, unique Rhizobium spp. occur in isolated geographical regions (Martinez-Romero & Caballero-Mellado, 1996). Symbionts also have complicated fitness impacts on their host; although they are generally considered to be beneficial, symbiont–host relationships may be mutualistic or pathogenic depending on host identity and environment (Dale & Moran, 2006). Rhizobium spp. and Frankia spp. benefit their leguminous and actinorhizal plant hosts by fixing nitrogen (Postgate, 1978) and Hamiltonella defensa defends its pea aphid host (Acyrthosiphon pisum) against parasitic wasps (Aphidius ervi) (Oliver et al., 2003). However, Wolbachia pipientis can cause reproductive abnormalities in a number of insect host species (Stouthamer et al., 1999) and Serratia symbiotica may reduce fecundity in pea aphids that are ‘superinfected’ with both S. symbiotica and H. defensa (Oliver et al., 2006). Also, just as for plasmid–host relationships, these symbioses have probably evolved multiple times, with examples from many different bacterial genera in many different host lineages (Thao et al., 2000; Moran et al., 2005).

There is considerable variety in the degree of dependency of a host on its symbiont; facultative symbionts, as noted previously, are dispensable whereas obligate symbionts are absolutely required for host survival. For example, female tsetse flies (Glossina spp.) undergo a loss of fertility when cured of their symbionts, Wigglesworthia glossinidia (Hill et al., 1973). Obligate symbionts are usually descended from ancient associations whereas facultative symbionts are more modern, implying that obligate relationships may be the evolutionary endpoints of facultative ones (Ochman & Moran, 2001; Moran, 2002). Further evidence for this comes from extant symbionts, such as the tsetse fly endosymbiont Sodalis glossinidius, which appear to be in transition from facultative to obligate status. The genome of S. glossinidius is characterized by a high proportion of pseudogenes for functions associated with a free-living lifestyle, such as defence and transport and metabolism of diverse carbohydrates (Toh et al., 2006). Similarly, plasmid hosts may be more or less dependent on their plasmids for survival. Some bacteria even have ‘obligate’ plasmid-like replicons which encode essential functions, the secondary chromosomes (for reviews see Mackenzie et al., 2004; Maclellan et al., 2004; Egan et al., 2005). There is some ambiguity in the definition of an essential function and there are a number of large replicons for which the designation of plasmid or secondary chromosome is not straightforward. For example, the legume symbiont Sinorhizobium meliloti 1021 has a composite genome comprising a chromosome (3.65 Mb) and two megaplasmids, pSymA (1.35 Mb) and pSymB (1.68 Mb), both with distinctive repABC plasmid replication genes (Galibert et al., 2001). However, as pSymB carries the only copy of an essential gene, the arginine tRNA, ^ArgtRNA_CCG, it is probably more appropriately termed a secondary chromosome. pSymA does not carry any essential genes and pSymA segregants have been demonstrated to grow under laboratory conditions (Oresnik et al., 2000). It is difficult to make definitive statements about the transferability of pSymA and pSymB because, although transfer has not been detected under laboratory conditions, this may not be relevant to transfer in the rhizosphere. For example, there is evidence that pSymA is self-transmissible but that transfer is repressed by expression of the pSymA-encoded rctA gene under laboratory conditions (Perez-Mendoza et al., 2005). One interpretation of the S. meliloti 1021 genome (Galibert et al., 2001) is that pSymB has lost all transfer genes, except for a paralogue of the pSymA traA and an oriT sequence, and gained an essential gene, the ^ArgtRNA_CCG, and is now ‘locked’ within its host, like W. glossinidia within tsetse flies. pSymA contains putative conjugative transfer genes (traACDG) and a putative oriT sequence but lacks the traIRMBF and trbDJKLFH genes found on other rhizobial plasmids and so, like S. glossinidius within tsetse flies, may be in transition from being facultative to being tightly associated with its host. Just as the evolutionary path from facultative to obligate symbioses seems to be a common one, it is also likely that the transition from plasmids to secondary chromosomes has occurred in multiple lineages on separate occasions. Mackenzie et al. (2004) list 44 species of bacteria, predominantly from the Alphaproteobacteria, but also from the Beta- and Gammaproteobacteria, spirochetes and deinococci, as having replicons that could be designated secondary chromosomes. Furthermore, phylogenetic relationships of the ParA ATPases of chromosomes, secondary chromosomes and plasmids from these species show that the secondary chromosome protein sequences cluster with plasmid-associated proteins rather than their chromosomal counterparts (Maclellan et al., 2004). Genomics has therefore highlighted an additional parallel between plasmids and the facultative bacterial symbionts: the potential to evolve from a facultative to an obligate symbiosis.

Does size matter?

Plasmids occupy an enormous size range, with an c. 2500-fold difference between the smallest and largest published bacterial plasmid sequences. These range from the 0.85 kb of pRKU1 from the obligate anaerobic heterotroph Thermotoga petrophila RKU1 (Nesbo et al., 2006) to the 2.09 Mb of megaplasmid pGMI100 from the soil-borne plant pathogen Ralstonia solanacearum GMI100 (Salanoubat et al., 2002). This size difference is far larger than the c. 80-fold difference between the smallest and largest published bacterial genome sequences, from the 160.00 kb of the psyllid endosymbiont Carsonella rudii PV (Carsonella-PV) (Nakabachi et al., 2006) to the 13.00 Mb of the myxobacterial strain Sorangium cellulosum So ce56, an important producer of secondary metabolites (Schneiker et al., 2007). There is substantial evidence to suggest that total genome size is related to ecological factors, with residents of stable environments, such as obligate parasites or symbionts, having smaller genomes and residents of complex, variable environments, such as soil-dwelling, facultative symbionts, having larger genomes (Bentley & Parkhill, 2004; Raes et al., 2007). We have investigated the relationship between ecological factors and plasmid genome size for 980 bacterial plasmids, including megaplasmids, from the Plasmid Genome Database (http://www.genomics.ceh.ac.uk/plasmiddb/) (Molbak et al., 2003). The plasmids were binned according to: (1) size (0–25, 25–50, 50–100, >100 kb) and (2) the environment from which the plasmid and/or host was isolated (animal, including animal products such as milk or faeces, or nonanimal with nonanimal further subdivided into aquatic and terrestrial for the two largest plasmid size fractions) (Fig. 2). Host environment was determined from metadata within the primary literature and/or project data from sequencing centres. In instances where the provenance of the plasmid and/or host was not apparent from these sources, isolation environment was recorded as unknown. Generally, there were less metadata available for plasmids belonging to the smaller size fractions, which prevented a meaningful subdivision of nonanimal into aquatic and terrestrial. Of all plasmids, over half were from animal- (55.20%) and approximately one-third from nonanimal- (32.86%) associated hosts with the remainder (11.94%) unknown. The higher proportion of plasmids from animal- than nonanimal-associated hosts in the database is presumably a reflection of sampling bias, with more studies focusing on microbial communities of human, animal and food products than of terrestrial and aquatic environments. This trend held for all but the largest size fraction, >100 kb, where the trend was significantly different, with far fewer plasmids from animal- (27.56%) than nonanimal- (71.15%) associated hosts (χ²=67.99, P<0.0005). Of the >100-kb plasmids from nonanimal-associated hosts, almost two-thirds were from terrestrial environments (65.77%) with fewer than one-third from aquatic environments (27.03%). This bias was less marked in the 50–100-kb size range of plasmids from nonanimal-associated hosts, where almost half (45.95%) were from terrestrial environments and approximately one-third (32.42%) from aquatic environments. With only 980 plasmid genomes available on the Plasmid Genome Database at the time of writing, these data are from what is clearly a limited and potentially biased sample of plasmids. However, these data do raise the possibility that plasmid size is under the same ecological constraints as chromosome size, with larger genomes associated with more complex, heterogeneous environments, such as soil. It may be that this similarity between plasmid and chromosome responses to ecological pressures extends to other genomic features, such as physical architecture and gene content and distribution. Therefore, insights into the ecological forces that shape plasmids may inform our understanding of the ecological forces that shape chromosomes.

2

In total, 980 bacterial plasmids from the Plasmid Genome Database (http://www.genomics.ceh.ac.uk/plasmiddb/) were binned according to size: 0–25, 25–50, 50–100 and >100 kb; and host environment: animal including animal products, e.g. milk and faeces (blue), nonanimal (green) and unknown (yellow). Of the 156 large plasmids (>100 kb), more were from nonanimal- (71.15%) than animal- (27.56%) associated hosts with the remainder unknown (1.28%). This is significantly different from the proportions in the total data set, where more were from animal- (55.20%) than nonanimal- (32.86%) associated hosts with the remainder unknown (11.94%) (χ²=67.99, P<0.0005). Of the 111 large plasmids from nonanimal-associated environments, 65.77% were from terrestrial (bright green) and 27.03% were from aquatic (mid green) environments with 7.21% unknown (light green). Of the 110 plasmids in the 50–100-kb size range, 45.95% were from terrestrial and 32.43% were from aquatic environments with 21.62% unknown.

In complex environments, there may be selection for large plasmids if they encode traits that render the host bacterium better able to survive in multiple niches. Is there also, however, a counter selection that limits the size of plasmids? The mechanisms by which the fitness costs of plasmid carriage operate are unclear. It has been variously suggested that costs are related to plasmid-encoded protein expression levels (Rozkov et al., 2004), replication and maintenance of plasmid DNA (Bjorkman & Andersson, 2000) or disruption of cellular regulatory status (Ricci & Hernandez, 2000). There are little data available on the influence of plasmid size and copy number on fitness costs for environmental plasmids and hosts. One study did investigate the effect of carriage of 101 antibiotic resistance plasmids isolated from medical material on the growth rate of E. coli K12 921 (Zund & Lebek, 1980). Although the authors reported no clear relationship between host population generation time and plasmid size, they did note that plasmids that extended host generation times by more than 15% tended to be large (>80 kb). Preliminary experiments conducted in our laboratory using the pQBR collection of plasmids (Lilley et al., 1996) have similarly shown an increase in fitness cost to the host with plasmid size (unpublished data). Further research is clearly needed to determine exactly which environmental, host and plasmid factors dictate the magnitude of the fitness costs associated with plasmid carriage. Initial indications are, however, that fitness trade-offs of plasmid carriage are environmentally mediated and are a determinant for plasmid size.

Are plasmids selfish or altruistic elements?

The perceptible fitness costs associated with carriage of many plasmids coupled with examples of plasmids that apparently confer no benefit on host cells has led to them being dubbed ‘selfish’ or ‘parasitic’ elements (Kado, 1998). Any benefit conferred on the host may also be viewed as a selfish strategy as survival of the host ensures survival of the plasmid. However, a number of common traits borne by plasmids are examples of cooperative social, or altruistic, behaviour, i.e. they benefit not only the host but also other members of the population. Examples include nodulation, catabolism of complex carbon sources and degradation of xenobiotic compounds. As with any cooperative social behaviour, there is the capacity for individuals to cheat and reap the benefits of the ‘public good’ while avoiding the metabolic cost of the behaviour. If these cheaters gain sufficient fitness advantages then they are liable to spread rapidly through a population, creating a tragedy of the commons (Hardin, 1968). If plasmids are indeed selfish, why do they carry public good genes? And how do public good plasmids persist when there is a net fitness benefit to cheating and plasmids may be physically lost via vegetative segregation?

Horizontal transfer has been proposed as a potential strategy for limiting the prevalence of cheating strains. Smith (2001) produced a mathematical model of the population dynamics of pathogenic bacteria that secrete virulence factors into the extracellular environment (cooperators) and their nonsecreting (cheating) counterparts. The model predicted that horizontal transfer of virulence factors would reduce the number of cheaters and result in a larger peak bacterial population size. It is difficult to comment on the adequacy of this model as the author did not compare model predictions with any experimental results. It is, however, an interesting hypothesis that warrants further investigation.

Another potential mechanism for limiting the prevalence of cheating strains is negative, or stabilizing, frequency-dependent selection, i.e. the fitness of a particular phenotype increases as it becomes less common in a population. Negative frequency-dependent selection is important in many eukaryotic systems for maintaining genetic polymorphisms (i.e. multiple alleles). Examples include foraging gene alleles in larvae of the common fruitfly (Drosophila melanogaster) (Fitzpatrick et al., 2007) and self-incompatibility alleles in plants (de Nettancourt, 1977). Negative frequency-dependent selection is also important for maintaining genetic variation in bacteria (Levin, 1988; Rainey & Travisano, 1998). The coexistence of plasmid-carrying and plasmid-free strains in a population may be thought of as a bacterial dimorphism. Negative frequency-dependent selection for plasmid-encoded genes has been experimentally demonstrated for a number of host–plasmid systems and therefore may be an important determinant for plasmid persistence. Dugatkin et al. (2005) performed competition experiments with cooperators (E. coli carrying an ampicillin resistance plasmid) and cheaters (plasmid-free E. coli protected from ampicillin in the vicinity of cooperators) in mixed liquid culture. The cheaters maintained frequencies of 5–12% in the presence of ampicillin. Also, when cheaters arose via vegetative segregation from pure cultures of cooperators in the presence of ampicillin, they were similarly maintained at frequencies of c. 10%. More recently, Ellis et al. (2007) demonstrated coexistence between environmental isolates of cooperators (Pseudomonas fluorescens SBW25 carrying the mercury resistance plasmid pQBR103) and cheaters (plasmid-free P. fluorescens protected from mercury in the vicinity of cooperators). It may be that these mechanisms, whereby selection for plasmid-encoded traits is influenced by community structure, are responsible for the stable persistence of some plasmids. The propensity for plasmids to be physically lost and horizontally transferred makes them suitably responsive agents for frequency-dependent processes.

Environmental variation

The vast majority of bacteria in natural environments colonize surfaces, such as rocks, soil particles, plant leaves and roots, gut linings, skin and liquid–air interfaces (Costerton et al., 1994). These environments are all spatially and temporally heterogeneous and some, such as soil, are extremely so (Young & Crawford, 2004). Almost 20 years ago, Eberhard (1990) proposed that sporadic selection for plasmid-encoded genes, typical of that in heterogeneous environments, was an important determinant for plasmid persistence. Bergstrom et al. (2000) subsequently produced a mathematical proof that demonstrated sporadic selection alone was not sufficient to ensure plasmid persistence but further conditions, such as selective sweeps (Turner et al., 1998), were also required. However, there have been few attempts to test Eberhard's hypothesis experimentally. This is due to the difficulties of both quantifying changes in physical, chemical and biological conditions in natural environments at the microscale and also predicting their effects on parameters such as horizontal transfer rates (van Elsas et al., 2003). Recently, however, there has been some progress, for example with the development and implementation of stochastic models (discussed below), towards identifying the role of environmental variation in plasmid persistence.

Spatial structure

The spatial structuring of bacterial populations in natural surface environments can affect plasmid persistence. For example, the dynamics of horizontal transfer differ between populations on surfaces and in liquids (for a review see Molin & Tolker-Nielsen, 2003). Also, the degree to which selection for plasmids is frequency-dependent may differ between surface and liquid environments. Chao & Levin (1981) very elegantly demonstrated this in competition experiments using E. coli B carrying the colicinogenic plasmid ColE3 and plasmid-free E. coli B. The colicinogenic phenotype is a particularly extreme form of cooperative behaviour as the metabolic cost of colicin production is death (lethal synthesis). At any one time, only a small proportion of the colicinogenic population produces colicin; both these producing cells and sensitive, plasmid-free cells are killed whereas the remaining colicinogenic, nonproducing cells are immune. Surviving cells are then able to utilize nutrients released from the dead, lysed cells. In mixed environments (liquid serial batch culture), colicinogenic populations were at an overall fitness advantage relative to sensitive populations, increasing in frequency over successive cultures, only when the initial frequency of colicinogenic cells was high. In spatially structured environments (soft agar matrix), however, colicinogenic populations were at an overall fitness advantage relative to sensitive populations even at far lower initial cell densities. As the colicinogenic cells were found to have a lower intrinsic growth rate than the sensitive cells, these differences must be understood in terms of the effects of spatial structure on the distribution of colicin and nutrients from dead cells. In mixed environments, colicin and nutrients will be distributed equally to colicinogenic, nonproducing cells and sensitive cells. If colicin concentrations are at sublethal levels, sensitive cells will have a fitness advantage due to their higher intrinsic growth rate. In spatially structured environments, however, colicin and nutrients will be at higher concentrations in areas around colicinogenic colonies. The colicinogenic, nonproducing cells that are able to survive in these areas will therefore benefit from the higher concentrations of nutrients. The influence of spatial structure on frequency-dependent selection has also been demonstrated for plasmids harbouring other cooperative traits. The negative frequency-dependent selection for P. fluorescens SBW25 carrying the mercury resistance plasmid pQBR103 demonstrated by Ellis et al. (2007) was found to differ between mixed and spatially structured environments. In mixed environments, coexistence of plasmid-carrying and plasmid-free strains occurred over a narrower range of higher mercury toxicity levels when the initial cell density was high. In spatially structured environments, the initial cell density had no effect. The findings of both these studies suggest that spatial structure increases the range of environmental conditions in which coexistence of plasmid-carrying and plasmid-free strains can occur. Spatial structure may therefore facilitate plasmid persistence and play a role in maintaining discontinuous distributions of plasmids in bacterial populations.

With the exception of a few recent studies, most mathematical models of plasmid population dynamics have not incorporated a spatial component. However, Lagido et al. (2003) did produce a model for horizontal transfer of plasmids on surfaces where donor and recipient cells were considered to initiate separate colonies which grew exponentially until nutrient exhaustion. Horizontal transfer through conjugation occurred instantaneously when donor and recipient colonies met. Experimental results were reasonably well described by the model, although it tended to overestimate instances of conjugation. When conjugation was set to not occur instantaneously, model predictions were improved. However, the authors conceded that a number of model assumptions, such as conjugation occurring every time a donor and recipient met and the exponential growth of all cells in a colony, were biologically unrealistic. More recently, Krone et al. (2007) produced a spatially explicit, stochastic individual-based model (IBM) of plasmid persistence on surfaces that incorporated plasmid loss as well as horizontal transfer. Spatially explicit IBMs simulate the behaviour of individuals occupying sites according to their interactions with other individuals in local neighbourhoods of proximal sites. In the model of Krone et al., local effects were captured by allowing the replication rate of an individual (bacterial cell) to vary according to: (1) cell type, i.e. donor, recipient or transconjugant, of the individual; and (2) nutrient level in a specified local neighbourhood. Similarly, horizontal transfer rates for donors were allowed to vary according to numbers of different cell types within another specified local neighbourhood of a donor. Stochastic effects were captured by updating sites asynchronously, i.e. at each time step, randomly chosen individuals were allowed to behave, i.e. replicate, transfer or lose plasmid etc., according to the constraints specified. The model was given a ‘hint’ of three-dimensional (3D) structure by allowing up to two individuals to occupy any one site. Experimental results were mostly well described by the model, except for dynamics of plasmid R1 in E. coli K12 from Simonsen (1990). This plasmid undergoes transitory derepression following transfer and so transfers at different rates according to residence time in the host cell. The authors predicted that not accounting for this variation in the model may have reduced the accuracy of its predictions. They also predicted that the lack of real 3D structure in the model would reduce its accuracy in capturing the behaviour of long-term experiments. Importantly, both models predicted features of plasmid persistence on surfaces consistent with experimental findings that were not predicted by the mass-action models after Stewart & Levin (1977). These models have provided a theoretical framework within which to investigate the dynamics of plasmids in spatially structured environments. The application of IBMs is in its infancy but there is the potential for making predictions as to the effects of realistic ecological variables, such as sporadic selection (Eberhard, 1990) and disturbance events (Buckling et al., 2000), on plasmid persistence.

Temporal fluctuations

Selection acting on plasmids changes over time according to variation in extrinsic (i.e. environmental) and intrinsic (i.e. plasmid and host) factors. There is often a correlation between plasmid prevalence and selection. For instance, greater numbers of mercury-resistant plasmids have been isolated from mercury-impacted soils than from pristine soils (Dronen et al., 1998; Smit et al., 1998). However, antibiotic resistance plasmids have been shown to persist in the absence of antibiotic selection (Chaslusdancla et al., 1987; Johnsen et al., 2005) so the cause and effect relationship is obviously not a simple one. Amelioration of the fitness costs associated with plasmid carriage may be partially responsible for this, either by adaptation of the host in the presence of selection (Bouma & Lenski, 1988) or by host–plasmid coadaptation in the absence of selection (Modi & Adams, 1991; Dahlberg & Chao, 2003). Some of the variations that influence plasmid persistence may be of a stochastic nature. Plasmid loss, for instance, has been noted to be a rather unpredictable event (Corchero & Villaverde, 1998).

Most models of plasmid persistence to date have ignored the potential influence of these temporal variations. However, Ponciano et al. (2007) recently devised a stochastic model for the dynamics of plasmid persistence that included the relative fitness of plasmid-free cells as a random variable influenced by environmental fluctuations (where ‘environment’ referred to the host itself as well as the environment surrounding the host). This model was then applied to seven plasmid–host combinations (of the broad host range IncP-1β plasmid pB10, and alpha-, beta- and gammaproteobacterial hosts) that showed plasmid loss over up to 600 generations in serial batch culture (De Gelder et al., 2007). Plasmid loss was better explained by this model than deterministic models in four of seven plasmid–host associations. Although the model was not mechanistic and could therefore not predict the cause of the stochasticity, we might speculate that it was due to environmental heterogeneity or genetic drift in the host and/or plasmid. Given these interesting initial results, greater exploration of the importance of stochasticity for plasmid persistence in natural environments seems pertinent. A comparison of the degree to which stochasticity explains plasmid persistence in heterogeneous (natural) environments and homogeneous (batch culture) environments would delineate the influence of extrinsic (environment) and intrinsic (host and plasmid) factors.

Concluding remarks

The focus of plasmid ecology research to date has been on plasmid population dynamics. In the future, there is likely to be increased emphasis on ecological factors that influence physical features of plasmids, such as size. This type of analysis is made vastly more powerful by the amount of genomics data now available; as of June 2007, the 1000th bacterial plasmid genome was logged on the Plasmid Genome Database. Genomics resources such as this will prove to be of enormous value for testing hypotheses about ecological influences on plasmid genomes. It may be that features such as physical architecture and gene composition will, like genome size, be similarly ecologically constrained for plasmids and chromosomes. If there are indeed universal truths that describe the influence of ecological factors on genomes, then plasmids are useful tools for investigating them. They are, by the simple virtue of being smaller, more tractable replicons than chromosomes.

As for most sequenced bacterial genomes, plasmid genomes often contain a significant proportion of genes of unknown function. It appears that our knowledge of plasmid-encoded gene function is more limited than that of chromosomally-encoded gene function as a number of composite genome-sequencing projects have found higher proportions of genes of unknown function on plasmids than chromosomes. In R. solanacearum GMI100, a greater proportion were found on the mega plasmid (18.7%) than on the chromosome (12.6%) (Salanoubat et al., 2002). Similarly, in S. meliloti 1021, a greater proportion were found on pSymA (11.5%) and pSymB (12.3%) than on the chromosome (5.0%) (Galibert et al., 2001). The mercury resistance plasmid pQBR103, the subject of over a decade of research and referred to earlier in this review, also has an exceptional degree of novelty; 80% of its 478 putative coding sequences could not be ascribed a function (Tett et al., 2007). Understanding of ecological factors that impact on plasmids will remain incomplete until knowledge of gene function is more comprehensive.

Although genome sequences may tell us a lot about plasmids, they are not a panacea for the problem of investigating plasmids in situ. As just mentioned, plasmid gene sequences are only meaningful in the context of gene function, plasmid gene functions are only meaningful in the context of gene expression, gene expression is only meaningful in the context of a host, and host bacteria are only meaningful in the context of their environment. Therefore, a variety of experimental, observational and theoretical approaches, at different scales, are needed. There has recently been a call to explore whether classical ecological theory might provide a framework within which to investigate microbial communities (Prosser et al., 2007). Given the parallels between plasmids and a group of microorganisms, the facultative bacterial symbionts, it seems appropriate to extend application of these tools to the ecology of plasmids.

Acknowledgements

We are grateful to Dawn Field for helpful discussions and to four anonymous reviewers and the editor for considered and constructive comments on early versions of the manuscript.

Author notes

Editor: Jan Dirk van Elsas