Alizon and Michalakis (In press) raise issues about the interpretation of results in my recent study of virulence evolution in an experimental bacteria/plasmid system (smith 2011). They argue that the results do show that superinfection is necessary to explain virulence evolution but do not exclude the possibility that a trade-off between virulence and infectious transmission to uninfected hosts also played a role. I did not intend to claim that such a trade-off had no influence on the results, although in retrospect there are places where the text of smith (2011) could have more accurately described the results as inconsistent with the predictions of the trade-off hypothesis if superinfection is assumed to be unimportant. Nevertheless, the issues these authors raise are interesting and worth addressing in detail. What can we say about the role of a transmission/virulence trade-off in the experiments of smith (2011)?
Alizon and Michalakis develop theoretical models to predict the direction of virulence evolution with and without immigration of uninfected hosts. Because these predictions are the same when virulence is correlated with superinfection rate as when correlated with both superinfection and transmission to uninfected hosts, they argue that it is impossible to exclude the relevance of a virulence-transmission trade-off by measuring how immigration qualitatively influences virulence evolution. This is consistent with the theoretical predictions in smith (2011) that plasmids could evolve increased virulence in the immigration treatment because of either infectious transmission or superinfection. I drew no conclusions from the effect on virulence evolution of adding uninfected hosts, nor did I claim evidence for such an effect.
Alizon and Mickalakis also show how plasmid effects on host mortality and host reproduction respond differently to the addition of uninfected hosts. They identify a term in their expression for virulence evolution (Urσx,s/K in eq. 1) that has a net positive effect on virulence when plasmids reduce host fecundity. They claim the term shows that increasing the availability of uninfected hosts increases virulence with or without a trade-off. This needs some clarification. In my experiments, total population size was always less than or equal to carrying capacity, so the larger expression of which this term is a part (labeled “vert. transmission” in eq. 1 of Alizon and Michalakis In press) would always be negative. The expression describes selection against plasmid virulence caused by vertical transmission. Immigration of uninfected hosts reduces the strength of this selection because they consume resources and reduce population growth rate, thereby reducing the reproductive value of vertical transmission. In the experiments, total population size was much less than carrying capacity for much of each growth period (Fig. S1 in smith 2011), causing this term to be small. It should also be noted that the immigration treatment increased the density of uninfected hosts but also decreased the density of infected hosts, which reduces selection for virulence (“superinfection” in eq. 1 of Alizon and Michalakis In press). The net effect would be determined by the quantitative balance of these two manipulations. In any case, I presented no evidence for an effect of immigration on plasmid virulence.
The data in smith (2011) most useful for assessing the role of a trade-off are the associations between virulence, infectious transmission, and superinfection among plasmid genotypes. In both experimental treatments, plasmids evolved greatly increased virulence, greatly increased superinfection rates, and modest to no increase in transmission to uninfected hosts. These findings seem to suggest that superinfection was the major determinant of virulence evolution, although infectious transmission could have also played a minor role in the immigration treatment. Alizon and Michalakis point out that, as acknowledged in smith (2011), the method used to sample plasmids from experimental populations is biased toward plasmids with high infectious transmission (admittedly not the best experimental design, in retrospect). The bias could alter the observed correlations. How would it alter our interpretation of the results? Despite the bias, evolved plasmids showed at best a small increase in infectious transmission above the ancestral plasmid genotype. If highly infectious, highly virulent plasmids existed in these populations, they would have been sampled. One effect of the bias could have been undersampling of plasmids with decreased transmission and decreased virulence. This means that a trade-off between transmission and virulence may have played a role in these experiments, but it would have been to select for plasmids with reduced virulence, not the large increase decribed in smith (2011). Another effect of the bias could have been undersampling of plasmids with increased superinfection rates but decreased transmission to uninfected hosts, so it is possible that the bias actually overestimated the role of infectious transmission in the evolution of virulent plasmids.
Alizon and Michalakis also suggest that even if infectious transmission is not directly correlated with virulence, it could still play an indirect role by increasing the contact rate between infected hosts and thus the frequency with which plasmids compete within hosts. How would the observed modest increase in infectious transmission, acting through its effect on superinfection, contribute to plasmid evolution? By itself, the <10-fold increase in infectious transmission, applied proportionally to superinfection, would not be enough to select for virulent plasmids (Fig. 1C in smith 2001). Dividing the superinfection transmission constant of evolved plasmids by their transmission constant to uninfected hosts shows that most of the increase in superinfection rates among evolved plasmids is independent of infectious transmission to uninfected hosts (Fig. 1). Even if differences in superinfection were entirely mediated by differences in total infectiousness, transmission to uninfected and infected hosts are distinct sources of selection with different effects on virulence evolution. In the absence of within-host competition, for example, the trade-off hypothesis predicts stronger selection for virulence in epidemics than in endemic pathogens (Lenski and May 1994). Selection for virulence because of superinfection, however, can be strongest when endemic because of higher contact rates between infected individuals. Mechanistic links between traits do not mean their effects are inseparable.
All told, the available evidence suggests that within-host competition caused by superinfection was the main factor driving increased virulence evolution in the experiments of smith (2011). The data cannot exclude the possibility that infectious transmission to uninfected hosts also played a minor role in the immigration treatment, in selecting for decreased virulence, or indirectly increasing superinfection rates—but these possibilities are not necessary to account for the results, either. The goal of these experiments was to determine, of the many known processes that could influence virulence evolution, which best predicts how virulence actually evolves. The results were not consistent with expectations based on models that focus exclusively on selection among hosts, as many (but not all) discussions of the trade-off hypothesis do. Instead, the results showed that within-host interactions can be primary determinants of virulence evolution, especially in chronic, vertically transmitted infections like those caused by plasmids.