A prominent hypothesis proposes that pathogen virulence evolves in large part due to a trade-off between infectiousness and damage to hosts. Other explanations emphasize how virulence evolves in response to competition among pathogens within hosts. Given the proliferation of theoretical possibilities, what best predicts how virulence evolves in real biological systems? Here, I show that virulence evolution in experimental populations of bacteria and self-transmissible plasmids is best explained by within-host competition. Plasmids evolved to severely reduce the fitness of their hosts even in the absence of uninfected cells. This result is inconsistent with the trade-off hypothesis, which predicts that under these conditions vertically transmitted pathogens would evolve to be less virulent. Plasmid virulence was strongly correlated with the ability to superinfect cells containing competing plasmid genotypes, suggesting a key role for within-host competition. When virulent genotypes became common, hosts evolved resistance to plasmid infection. These results show that the trade-off hypothesis can incorrectly predict virulence evolution when within-host interactions are neglected. They also show that symbioses between bacteria and plasmids can evolve to be surprisingly antagonistic.

Pathogens play an important role in biological evolution. They are first of all an important component of biological diversity; parasitic species probably outnumber nonparasites several times over (Windsor 1998). Pathogens can also have a large impact on the evolution of their hosts, having been implicated in shaping traits such as the vertebrate immune system (Apanius et al. 1997), sexual ornamentation (Hamilton and Zuk 1982), and even sex itself (Hamilton 1980). The pathogen trait that imposes selection on hosts—and the trait that makes pathogens pathogens—is their virulence, which I use here to mean the amount that an infection reduces a host's fitness. Virulence varies to a puzzlingly large degree among pathogens, even among closely related species or different strains of the same species. The bacterium Escherichia coli, for example, is known to many as a cause of diarrhea, but the majority of E. coli strains are harmless. RNA viruses can be as deadly as HIV or as innocuous as the common cold. When do pathogens evolve to be more virulent or less virulent? Some have argued that understanding this issue would allow us to manage human pathogens so that they evolve to be less severe or even benign (Dieckmann et al. 2001; but see Ebert and Bull 2003). Virulence theory is also important for understanding how interactions between species evolve to be parasitic or mutualistic (Herre et al. 1999; Sachs et al. 2004).

Intensive theoretical investigation has identified a large number of factors predicted to influence virulence evolution (Frank 1996; Dieckmann et al. 2001). The most prominent hypothesis holds that virulence is largely determined by a genetic trade-off with infectious transmission (Anderson and May 1982; Alizon et al. 2009). In this view, damage to hosts is an unavoidable side effect of traits that increase pathogen infectiousness, such as reproduction within hosts, coughing, or diarrhea. At the same time, increased virulence reduces the duration of infectiousness by causing host death, changes in host behavior, or stimulating immune clearance. When infections can be inherited by hosts’ offspring, increased virulence also reduces the contribution that vertical transmission makes to pathogen fitness (Fine 1975; Lipsitch et al. 1996). Selection favors pathogen genotypes that balance these conflicting effects. A central result is that evolution maximizes R0—the total number of new infections caused by an infected individual in a wholly susceptible population (Anderson and May 1982). The optimum virulence can be affected by many factors, including the availability of uninfected hosts and background host mortality.

A major alternative to the trade-off hypothesis is that virulence is determined by competition among pathogens within hosts. Within-host competition can occur if a host's initial infection contains multiple pathogen genotypes or if a host gets secondarily infected (superinfected) with another genotype. Competition for host resources can favor pathogens that reproduce more quickly and are thus more virulent (Levin and Pimentel 1981; Nowak and May 1994). On the other hand, interference or exploitative competition among pathogens can favor pathogens of lower virulence (Chao et al. 2000; Smith 2001; Brown et al. 2002; West and Buckling 2003). The net effect of within-host competition depends on the biological details of the infection process. Either way, pathogens are only predicted to maximize R0 if within-host competition is an unimportant component of pathogen fitness. Most applications of the trade-off hypothesis thus assume that within-host competition is an unimportant determinant of virulence evolution.

The evidence supporting these hypotheses (or any other, for that matter) is mixed at best. Transmission trade-offs seem sufficient to explain virulence in some important pathogens (Mackinnon and Read 2003; Alizon et al. 2009), but it is far from clear whether this is true for most species (Ebert and Bull 2003). In addition, much of the evidence fails to distinguish between competing alternative explanations. An increase in contact rate, for example, can increase both infectious transmission and the frequency of multiple-strain infections (Herre 1993, 1995; Frank 1996). Evidence is most useful when it is not only consistent with a hypothesis, but also inconsistent with one or more of its alternatives (Platt 1964). At present, it is unclear how general are trade-offs between transmission and virulence, to what degree they predict virulence evolution, and whether their effects are direct or are mediated by within-host competition.

Here, I test the trade-off hypothesis with experimental populations of bacteria and plasmids. Plasmids are molecules of DNA that inhabit bacterial cells, replicate independently of chromosomes, and are inherited by both daughter cells during cell division (Summers 1996). Genes for antibiotic resistance and bacterial pathogenicity are often carried by plasmids, but most plasmids seem to not benefit their hosts. Plasmids often reduce the reproductive rate of bacteria (Lenski and Bouma 1987; Modi and Adams 1991; Dahlberg and Chao 2003). This reduction in fitness can be considered the plasmid's virulence. Plasmid virulence may potentially evolve according to the trade-off hypothesis, because many plasmids transmit themselves infectiously among bacteria using organelles called conjugative pili, and expressing these pili increases the fitness cost of plasmid infection (Haft et al. 2009). Competition among plasmids within hosts may also influence virulence evolution. Plasmids have genes that reduce superinfection by related plasmids (Harrison et al. 1992), but this reduction is not absolute (two or three orders of magnitude). Increased replication within hosts could make plasmids more competitive but also increase virulence (Paulsson 2002). Toxin–antitoxin systems also contribute to within-host competition while increasing plasmid costs (Cooper and Heinemann 2005).

To test the extent to which the trade-off hypothesis predicts virulence evolution, I allowed plasmids and bacteria to evolve with or without immigration of uninfected hosts (Fig. 1A). With high levels of immigration, plasmids must continually infect new hosts to persist in the population. The trade-off hypothesis predicts that under these conditions—an abundance of uninfected hosts and high host mortality—plasmids will evolve increased virulence as a correlated effect of increased infectious transmission. With no immigration, the trade-off hypothesis predicts plasmids will evolve decreased virulence because infectious transmission does not increase plasmid fitness when all hosts are already infected (Fig. 1B; Lipsitch et al. 1996). If virulence evolution is determined by within-host competition, however, plasmids can evolve increased virulence in both experimental treatments as a correlated effect of increased superinfection (Fig. 1C).

Figure 1.

(A) A design of the evolution experiment. Populations of plasmids and bacteria were allowed to evolve under a daily serial transfer regime. In the no-immigration treatment, all cells began infected with plasmids and were propagated with no additional input of cells or plasmids. In the immigration treatment, all cells began infected but 90% of the population used to inoculate the next day's flask came from a separate paired uninfected population. Three separate lines evolved under each treatment. (B, C) Theoretical predictions when plasmid evolution is determined by trade-offs between virulence and (B) infectious transmission or (C) superinfection. Dotted lines show ancestral plasmid traits. Shaded regions show mutant plasmid genotypes that can increase in frequency in the no-immigration treatment (light gray), the immigration treatment (medium gray), both treatments (dark gray), or neither treatment (white). For model details, see Supporting information.

The experimental design is a modified version of that used by Turner et al. (1998). These authors sought to test the trade-off hypothesis, but their experimental design did not lead to the desired differences in infectious transmission; realized plasmid transmission was very low in all treatments. To better reflect the epidemiology of successful pathogens, I chose a plasmid, host strain, and immigration rate such that plasmids could invade and persist in bacterial populations with no external selection for plasmid-borne genes. That is, infectious transmission of the plasmid was sufficient to overcome any reduction in the fitness of its host (Supporting information). In epidemiological terms, this means that the plasmid's R0 was greater than one.

Materials and Methods


Unless otherwise indicated, all cultures were grown in 50-mL Erlenmyer flasks containing 10.0-mL Davis minimal media supplemented with 1000 μg/mL glucose and 20 μg/mL uracil. All cultures were incubated at 37°C while shaking at 200 rpm. Serial transfer cultures were passaged with daily 100-fold dilution. Cell densities were estimated from colony counts on tetrazolium arabinose (TA) plates (Levin et al. 1977). Colonies able to ferment the sugar arabinose (Ara+ strains) appear white on TA plates, while colonies that cannot ferment arabinose (Ara strains) appear red. Where indicated, media was supplemented with antibiotics at the following concentrations: nalidixic acid (Nal) at 2.0 μg/mL, rifampicin (Rif) at 10 μg/mL, kanamycin (Km) at 25 μg/mL, and chloramphenicol (Cm) at 25 μg/mL.


All bacteria were derivatives of E. coli K12 strain MG1655 (Blattner et al. 1997). A spontaneous Ara mutant of MG1655 was isolated by visually screening colonies on TA plates. Spontaneous NalR and RifR mutants were isolated by plating on TA Nal and TA Rif, respectively. Plasmid R1–19 (also known as R1drd19) was obtained from B. Levin (Emory University). R1–19 is a finO mutant of the IncFII plasmid R1 (Meynell and Datta 1967). The FinOP antisense RNA system represses plasmid transfer, so R1–19 is more infectious than R1. R1–19 carries genes for resistance to the antibiotics ampicillin, streptomycin, spectinomycin, kanamycin, and chloramphenicol. Plasmid R1a, a KmS mutant of R1 (Meynell and Datta 1967), was obtained from Kurt Nordström (Uppsala University, Sweden). All plasmids were introduced into host strains by conjugation. Table 1 lists the strains used in this article.

Table 1. Escherichia coli strains.
StrainChromosomal markers1Plasmid2Reference or source3
  1. 1Ara, arabinose; Nal, nalidixic acid; Rif, rifampin.

  2. 2ev, evolved derivative.

  3. 3NCTC=National Collection of Type Cultures, London.

js10Ara+ NalR This study
js40AraThis study
js41Ara NalR This study
js42AraR1–19This study
js99Ara RifRR1aThis study
js171Ara+ RifRThis study
js177AraR1–19evThis study
js178Ara NalRR1aThis study


The design of the evolution experiment is shown in Figure 1. Populations in the no-immigration (N) treatment were founded by Ara cells infected with plasmid R1–19 (strain js42) and propagated without any additional input of uninfected cells. Populations in the uninfected (U) treatment were founded by plasmid-free Ara cells (js40) and propagated without additional input. Populations in the immigration (I) treatment were founded with Ara+ cells infected with plasmid R1–19 (js10). For each transfer thereafter, 0.1 mL of the previous day's I culture was mixed with 0.9 mL of a paired U culture and 0.1 mL of this mix used to inoculate the next day's I flask. All treatments were performed with threefold replication, for a total of nine evolving populations. Populations were propagated for 60 transfers (about 400 generations). Every 50 or so generations, a 2.0-mL sample from each population was mixed with 1.0 mL 60% glycerol (v/v) and stored at −80°C. The immigration treatment increases background mortality and opportunities for infectious transmission to uninfected hosts, for both of which the trade-off hypothesis predicts increased virulence (Anderson and May 1982; Lipsitch et al. 1996). The experimental treatments do not alter vertical transmission. Infected cells can lose their plasmids, but this is extremely rare—around 10−7 per cell division (Nordström and Aagaard-Hansen 1984). Spontaneous plasmid loss is thus unlikely to create a measurable abundance of uninfected hosts.


To assay ecological dynamics during the evolution experiment, I measured the stationary phase cell densities of revived samples from experimental populations. Samples were revived by looping frozen material into 10-mL media and incubating for 24 h. Total cell densities were then measured from colony counts on TA plates. Densities of KmR and CmR cells were estimated from colony counts on TA Km and TA Cm plates, respectively.

To distinguish between loss of plasmid infection and loss of antibiotic resistance genes from evolved plasmids, I directly assayed plasmid DNA in ancestral and evolved host populations. To isolate plasmid DNA, frozen samples were first looped into 10-mL LB broth and cultured overnight. Plasmid DNA was then isolated using QIAfilter Midiprep kits (QIAGEN, Valencia, CA) as per the instructions, except that DNA was eluted at 65°C. DNA was resuspended in 200-μL deionized water. A total of 20-μL sample was mixed with 4-μL 6× loading dye (Promega, Madison WI) and run on an 1.0% agarose gel at 75 V for 6 h. The gel was stained for 45 min in 1 μg/mL ethidium bromide and destained for 60 min in deionized water. An image was acquired using a Bio-Rad Gel Doc gel documentation system (Bio-Rad Laboratories, Hercules CA). The image was uniformly adjusted for the brightness and contrast in Adobe Photoshop 7.0.


To control for host evolution, the plasmid traits were assayed in a common ancestral host background. To sample plasmids from experimental populations, flasks containing 10.0-mL culture media were inoculated with frozen samples of both an experimental population and an uninfected ancestral Ara+ RifR host (strain js171). These coinoculated flasks were then cultured for 1 day to allow plasmids to infect the ancestral hosts. Cultures were then plated onto TA or TA Rif plates for single colonies. A random Ara+ colony was picked from 10.0-mL fresh media and cultured another day. These strains were then screened for a plasmid carriage by assaying their sensitivity to antibiotics and phage MS2 (obtained from B. Levin, Emory University). Phage lysates were spotted onto soft agar lawns of the strains to be tested and cultured overnight at 37°C. MS2-sensitive strains show plaques whereas MS2-resistant strains remain turbid. Antibiotic sensitivity was assayed using BBL Sensi-Discs (Becton Dickinson, Sparks, MD). Strains were considered infected if they were resistant to at least one antibiotic or were sensitive to MS2. At least five genotypes were sampled at each time point, except for population N2400, for which sampling yielded only one plasmid out of eight attempts. It should be noted that this sampling method is likely biased toward plasmids with high infectious transmission rates. Sampled plasmids may therefore not reflect the full distribution of traits in experimental populations.


I assayed the virulence of sampled plasmids by competing infected strains against a standard plasmid-resistant competitor (js177, an Ara strain isolated from population N2400). For these competition experiments, 0.5 mL of overnight cultures of both js177 and the tested strain were mixed to create an initial population. A total of 0.1 mL of this mix was then used to inoculate 9.9-mL fresh media. After 1-day incubation, 0.1 mL of this culture was used to inoculate a fresh flask and incubated for another day. Stationary phase densities of infected (I) and resistant (R) cells at days 0, 1, and 2 were estimated from colony counts on TA and TA Rif plates. Fitness of infected strains was measured as selection coefficients relative to the resistant competitor, calculated as s=b[ln(I/R)]/ln(100) where b is the regression coefficient over time. Lower values of s indicate plasmids of higher virulence. Single-fitness measurements were made for each plasmid genotype sampled. Five measurements were made for the uninfected control strain.


Infectious transmission was measured as the ability of plasmids to infect ancestral Ara NalR hosts (strain js41). Superinfection was measured as the ability to infect the ancestral Ara NalR hosts already infected with plasmid R1a (strain js178). For these assays, overnight cultures of donor strains were first diluted 10,000-fold in saline and 0.5 mL mixed with 0.5 mL of overnight recipient culture. A total of 0.1 mL of this mix was then added to 9.9-mL fresh media and cultured for 1 day to allow the plasmid transmission and superinfection. From this culture, stationary phase densities of recipients were estimated from colony counts on TA plates. Densities of donors were estimated from colony counts on TA Km and/or TA Rif plates. Densities of transconjugants were estimated from colony counts on TA Km or TA Nal Km plates.

Plasmid transfer rates were calculated using a modified version of the end-point method (Simonsen et al. 1990), which measures the transmission constant for plasmids with mass-action infection dynamics. Because the end-point method assumes that infection rates are directly proportional to cell growth rates, only β (the proportionality between growth and infection) was measured. The end-point method also assumes plasmids have no cost, but is insensitive to even very large costs under the conditions used here, where plasmids remain rare (Simonsen et al. 1990). The infection constant was calculated as β= ln(1 +TN/RD)/(N − N0), where D is the density of the donor strain, R is the density of recipients, T is the density of transconjugants, N is total population density, and N0 is the initial density of the inoculated population.


Resistance was measured as the relative ability of cells to be infected by the ancestral plasmid. To obtain genetically marked recipients, single RifR clones from populations of interest were isolated by plating overnight population cultures onto TA Rif plates. Typical-looking colonies were streaked onto LB plates and cultured overnight. Typical-looking colonies were chosen because some RifR mutations are known to decrease fitness and colony size (Reynolds 2000). Single well-isolated typical-looking colonies were picked from the LB plates into 10.0-mL media and cultured overnight. These RifR mutants were screened for plating efficiency on TA Rif relative to TA. For mutants with plating efficiency >50%, 2.0 mL of culture was added to 1.0-mL 60% glycerol and stored at −80°C. Resistance was also measured for the ancestral uninfected strain (js40) and an Ara RifR ancestral strain (js99) carrying plasmid R1a.

Resistance was measured against a common plasmid donor: an Ara+ NalR ancestral host infected with the ancestral plasmid (strain js10). For these assays, 0.5 mL of overnight donor culture was first mixed with either 0.5 mL of overnight recipient culture (for plasmid-infected recipients) or 0.5 mL of overnight recipient culture diluted 10,000 fold in saline (for uninfected recipients). A total of 0.1 mL of this mix was then added to 9.9-mL fresh media and incubated 1 day to allow the plasmid transfer. From this culture, stationary phase densities of recipients were estimated from colony counts on TA plates. Densities of donors were estimated from colony counts on TA or TA Nal plates. Densities of Ara RifR KanR transconjugants were estimated from colony counts on TA Km or TA Rif Km plates. Transmission constants were calculated using the modified end-point method described above.


Statistical analyses were performed using R 2.8.1 (The R Foundation for Statistical Computing). Because the distribution of plasmids sampled from experimental populations was likely biased toward those with high infectious transmission rates, an analysis of virulence trends over experimental time may be misleading. Instead, I tested whether plasmids of altered virulence appeared during the evolution experiment by comparing the virulence of the ancestral plasmids (sampled from populations after only one transfer, when mutants are expected to still be rare) against those sampled later in the experiment (generation 50 and onwards). Virulence data were fit to linear mixed effects models (lme procedure) fit by maximum likelihood with population as a random effect (Crawley 2002). P-values were calculated using likelihood ratio tests on models fit with or without experimental evolution (ancestral vs. evolved) as a fixed effect.

Relationships between infectious transmission, superinfection, and virulence were tested using Spearman rank correlations (cor.test procedure). Partial rank correlations were calculated following Sokal and Rohlf (1995). P-values for the significance of partial rank correlations were obtained from bootstrapped distributions based on 100,000 resamplings of the experimental data. Differences among immigration treatments in the relationship between virulence and infectious transmission or superinfection were tested using generalized additive models. Nonparametric fits for plasmid virulence as a function of log10(infectious transmission) or log10(superinfection), plus interaction of these terms with immigration treatment, were calculated using the gam procedure.

Data for host resistance showed significant heterogeneity of variances among groups even after log10 transformation (Bartlett's K2= 21.96, df = 7, P= 0.0026) and so was transformed as ln(−log10β− 7) (K2= 6.19, df = 7, P= 0.518). Differences among means were tested using the T-method (TukeyHSD procedure) for multiple unplanned comparisons at a 99% confidence level (Sokal and Rohlf 1995).



Between generations 50 and 150 of the evolution experiment, all plasmid-infected populations showed a marked drop in stationary phase cell density (Fig. 2). Population densities in the immigration treatment stayed low or decreased over the course of the experiment, whereas in the no-immigration treatment the decreases were transitory. All replicate lines of the no-immigration treatment were at some point visibly late becoming turbid (dashed vertical lines in Fig. 2) but regained normal growth later in the experiment. By chance, samples of population N2 were frozen and stored the day before it was late. When these samples were later revived and cultured in the laboratory, they were also late to become turbid and reached a much reduced population density after 24 h of growth.

Figure 2.

Reduced population densities in evolving plasmid-infected populations. Reductions were sustained in the immigration treatment (I) but transitory in the no-immigration treatment (N). Each panel is a replicate evolving population. Solid lines show total cell densities after 1-day growth. Dashed lines show densities of cells resistant to kanamycin. Dotted lines show densities of cells resistant to chloramphenicol. Points are geometric mean of three independent samplings. Error bars show range of data. Dashed vertical lines indicate when slow culture growth was first observed.

Infected populations began losing resistance to antibiotics at roughly the same time as the first decreases in stationary phase density (Fig. 2). In most populations, antibiotic-resistant cells eventually decreased to an undetectable frequency. In population N1, cells with wild-type antibiotic resistance profiles decreased in frequency but eventually rebounded and persisted for ∼100 generations at a frequency of ∼10%. Loss of antibiotic resistance was not caused by loss of plasmids. At generation 400, all infected populations retained plasmid DNA, but of a smaller size than the ancestral plasmid (Fig. 3). All three populations in the immigration treatment contained multiple plasmid bands.

Figure 3.

Experimental populations lost resistance to antibiotics but retained evolved plasmids. Shown are plasmid DNA preps from the ancestral infected strain and from experimental populations after 400 generations of evolution. R1–19 = ancestral plasmid, N= no-immigration populations, I = immigration populations, U = uninfected populations.


Plasmid virulence was measured as the fitness of infected hosts relative to a common plasmid-resistant competitor. To control for the effects of host evolution, plasmids were sampled from evolving populations and measured in a common ancestral host background. As expected, the ancestral plasmid reduced host fitness (uninfected s=−0.171 ± SE 0.005, n= 5; infected s=−0.202 ± SE 0.008, n= 32; t=−3.271, P= 0.0014, one-sided). Plasmids sampled from evolved populations (generation 50 and later) were more virulent than the ancestor in both the immigration treatment (Fig. 4; LR= 19.16, df = 1, P < 0.0001) and the no-immigration treatment (LR= 9.67, df = 1, P= 0.0019). Evolutionary change did not occur as a monomorphic shift in population mean virulence. Instead, it occurred primarily through the addition of virulent genotypes. Some of these plasmids were extremely virulent, with selective coefficients less than negative one—less than what would be observed if infected cells did not reproduce at all. Many evolved plasmids reduced colony size and prevented cultures from becoming turbid (data not shown). At the same time, plasmids of low virulence persisted for the duration of the experiment. These less-virulent plasmids no longer conferred antibiotic resistance, so they were not simply a persisting ancestral genotype.

Figure 4.

Highly virulent plasmids appeared in all populations of both immigration treatments. Virulence was measured as the fitness of infected hosts relative to a common plasmid-resistant competitor. Lower values indicate plasmids of higher virulence. Circles show single measurements of independently isolated plasmid genotypes. Solid lines are arithmetic mean of genotypes at each time point. Each panel is an independently evolving population. Shaded circles show plasmid genotypes for which infectious transmission and superinfection were measured (Fig. 5).


When sampled plasmids retained resistance to at least one antibiotic, I measured their ability to infect plasmid-free cells (infectious transmission) and to infect cells infected with an incompatible competitor plasmid (superinfection). These were calculated as transmission constants, which control for the densities of plasmid donors and recipients and are equivalent to the transmission constants in epidemiological models of infectious diseases (Simonsen et al. 1990). Plasmid virulence was weakly correlated with infectious transmission (Fig. 5; Spearman ρ=−0.31, P < 0.05, two-sided, n= 53) and strongly correlated with superinfection (ρ=−0.60, P < 0.0001, two-sided, n= 54). Infectious transmission increased less than 10-fold, whereas superinfection increased over 100-fold, approaching the ancestral infectious transmission rate. All sampled plasmids more virulent than the ancestral genotype had increased superinfection, but not all had increased infectious transmission.

Figure 5.

Plasmid virulence was weakly associated with the ability to infect plasmid-free hosts (A) but strongly associated with the ability to superinfect hosts carrying an incompatible competitor plasmid (B). Dotted lines show ancestral plasmid traits (mean of plasmids isolated from the first experimental transfer, when mutants are expected to be rare). Circles show plasmids isolated from evolved populations (generation 50 and later). Gray circles: plasmids from the no-immigration treatment. Black circles: plasmids from the immigration treatment.

Infectious transmission was also correlated with superinfection (ρ= 0.47, P < 0.001, two-sided, n= 52). Controlling for the interrelationships among all three traits, the partial correlation between infectious transmission and virulence was nonsignificant (Spearman partial ρ=−0.04, P= 0.78, two-sided, n= 53) whereas the other two remained significant (superinfection vs. virulence: ρ=−0.54, P < 0.0001, two-sided, n= 54; superinfection vs. infectious transmission: ρ= 0.38, P < 0.01, two-sided, n= 52). The apparent association between infectious transmission and virulence was thus entirely mediated by the correlation between infectious transmission and superinfection. There was a shallower relationship between superinfection and virulence among plasmids from the no-immigration treatment (F= 5.91, df = 2.71, P= 0.0013, n= 54). For a given superinfection rate, plasmids from the immigration treatment were less virulent. There was no difference between treatments in the relationship between infectious transmission and virulence (F= 0.06, df = 1.0, P= 0.90, n= 53).


Coinciding with or shortly following the no-immigration treatment's drops in a population density, strains appeared that produced mucoid colonies and were resistant to phage MS2 (which infects cells expressing plasmid pili). In population N1, these strains were first visible at generation 120 and remained abundant thereafter. They were first visible in population N3 at generation 100, were absent at generation 200, reappeared at generation 250, and remained abundant thereafter. Mucoid cells were never observed in the N2 population. In the immigration treatment, drops in population density coincided with the appearance of strains with small colonies, but neither mucoidy nor MS2 resistance was observed except when populations were propagated several additional days without immigration (data not shown). Mucoidy was never observed in uninfected populations.

Because mucoidy is known to confer resistance to bacteriophage, I tested MS2-resistant evolved strains for resistance to infection by the ancestral plasmid. At generation 400, strains isolated from the no-immigration treatment were more resistant to infection than the ancestral uninfected, evolved uninfected, and ancestral infected strains (Fig. 6). The nonmucoid strains from population N2 were more resistant than the mucoid strains of N1 and N3. Preparations of plasmid DNA from resistant strains contained distinct bands (data not shown), so it is possible that evolved plasmids contributed to resistance.

Figure 6.

Strains from the no-immigration treatment evolved resistance to plasmid infection. Resistance was measured from the ability of tested recipients to be infected by the ancestral plasmid genotype. Smaller transmission constants indicate greater resistance. Circles show individual measurements of four independently isolated strains from each population. Bars show geometric mean. Evolved populations were sampled at generation 400. Groups with different letters differ significantly at a 99% level of confidence (Tukey's HSD).


These results show that trade-offs between infectious transmission and damage to hosts are not always the primary determinants of virulence evolution. Highly virulent plasmid genotypes evolved in all populations of the no-immigration treatment and became abundant enough to be sampled (Fig. 4), to reduce population growth (Fig. 2), and to select for resistant hosts (Fig. 6). These results are inconsistent with the trade-off hypothesis, which predicts that virulent plasmid mutants should have been selected against in these populations because infectious transmission only increases plasmid fitness when there are uninfected hosts (Fig. 1B; Lipsitch et al. 1996). In addition, the often-assumed virulence/transmission trade-off was weakly supported, at best (Fig. 5).

Instead, the results are most consistent with models that emphasize within-host competition (Supporting Information; Levin and Pimentel 1981; Nowak and May 1994). The strong correlation between plasmid virulence and superinfection ability (Fig. 5) can account for these genotypes’ rise to abundance. When all hosts are infected, plasmids that can better transmit themselves to other infected cells and displace the resident plasmids are able to increase in frequency despite being more virulent (Fig. 1C). The experimental immigration treatment does not strongly reduce opportunities for within-host competition because almost all immigrants are infected by resident plasmids. For a substantial fraction of the growth cycle, uninfected cells are rare (Fig. S1).

The trade-off hypothesis has received a lot of attention (reviewed in Alizon et al. 2009), but some authors have questioned its usefulness and generality (Ebert and Bull 2003). Much of its empirical support in vertically transmitted infections has come from evolution experiments in which researchers propagate either infected hosts or infectious particles, effectively enforcing either complete vertical or complete horizontal transmission (Bull et al. 1991; Messenger et al. 1999; Stewart et al. 2005; Sachs and Wilcox 2006). Here, rather than forcing the system to fit the assumptions of any particular model, I manipulated ecological conditions (immigration and background host mortality) and asked which of the available theoretical models, if any, predicted how virulence evolves. The results give further reason to question the generality of the trade-off hypothesis and suggest that attempts to predict virulence evolution should explicitly take within-host interactions into account.

The severity of plasmid/bacteria antagonism in these experiments is surprising. In previous studies, plasmid virulence ranged from near-zero to ∼20% reduction in fitness, usually decreasing after experimental evolution (Lenski and Bouma 1987; Modi and Adams 1991; Turner et al. 1998; Dahlberg and Chao 2003; Dionisio et al. 2005). Here, however, virulence increases to a level that would be expected if plasmids prevented their hosts from reproducing and caused them to actually die. The different results seen in different plasmid evolution experiments may stem from the fact that in this experiment, unlike those previous, the ancestral plasmid can invade host populations by infectious transmission (Fig. S2). If ancestral levels of infectious transmission and superinfection were very low, mutations that increase these traits might still be insufficient to allow mutant genotypes to increase in frequency. Mutations may be more likely to be beneficial when these traits are already substantial components of plasmid fitness.

The mechanistic causes of high plasmid virulence are unclear from these data, but there are several possibilities. Competition among plasmids within cells, caused for example by increased superinfection, is predicted to select for plasmids that replicate faster but impose a larger metabolic burden on host cells due to DNA replication (Paulsson 2002). Postsegregational killing systems have also been shown to make plasmids more competitive at within-cell competition, again at a cost to host fitness (Cooper and Heinemann 2005). The ancestral plasmid carries multiple postsegregation killing systems (Nordström and Austin 1989), alterations in which might lead to increased competitiveness and virulence. Production of conjugative pili increases plasmid virulence (Haft et al. 2009), but the lack of a substantial increase in infectious transmission among evolved plasmids argues against an increased production of pili. Instead, virulence may stem from the infection process itself. If a cell is infected by many donors at the same time, membrane integrity can become compromised—an effect called lethal zygosis (Skurray and Reeves 1973). This effect could come into play if superinfection becomes very prevalent in a population.

Regardless of the mechanism, these results show that plasmids are capable of being very virulent. But how representative is this of plasmids in natural populations? It is possible that a spatially structured environment such as the mammalian gut (the usual environment of E. coli) would limit the ability of excessively virulent plasmids to invade bacterial populations (Kerr et al. 2006; Boots and Mealor 2007). On the other hand, there may be a bias against observing virulent plasmids in natural environments: bacterial isolates that grow poorly are unlikely to become a part of strain collections.

Hosts evolved resistance to plasmid infection in all three no-immigration lines. In the immigration treatment, the large daily influx of susceptible hosts likely prevented resistant mutants from increasing in frequency. There appear to be multiple mechanisms of resistance at work: one involving mucoid colony morphology and another higher level of resistance that does not (Fig. 6). Insertion mutants in lipopolysaccharide biosynthesis genes are known to confer both mucoidy and resistance to infection by IncF plasmids (Perez-Mendoza and de la Cruz 2009). At least some of the resistant strains harbor plasmid DNA, so it is also possible that evolved plasmids contribute to resistance.

The finding that bacteria can readily evolve resistance to plasmid infection complicates efforts to assess whether plasmids are parasitic or mutualistic in natural populations (Simonsen 1991; Gordon 1992; Levin 1993). Variation in the ability of natural E. coli isolates to act as plasmid recipients (Gordon 1992) could be driven at least in part by selection for resistance to plasmids. Under some types of host/pathogen coevolution, pathogens can be maintained in largely resistant host populations even when susceptible hosts are relatively rare (Frank 1994). One cannot expect, then, to tell whether plasmids are usually parasitic or mutualistic simply by measuring virulence and infectious transmission in the laboratory and then comparing these values with the cell densities of gut populations (Simonsen 1991; Gordon 1992; Levin 1993).

Associate Editor: P. Turner


I would like to thank T. Cooper for pointing out the ancestral plasmid's high superinfection rate. I would also like to thank R. Antia, J. Koella, and G. Saxer for helpful discussion and S. Altizer, B. Levin, J. Logsdon, and J. Strassmann for comments on the manuscript. This work was supported by a Howard Hughes Medical Institute Predoctoral Fellowship to the author and National Institutes of Health grant GM33782-17 to B. Levin (Emory University).