Host genotype by parasite genotype interactions
Who controls the epidemiological traits of a host-parasite association is a key factor in predicting the evolution of this association [16,26]. Until recently, most models of host-parasite coevolution assumed that traits like symptom severity, transmission or virulence were the characteristics of the parasite only. Now an increasing body of evidence on plant-pathogen systems [20,27], microorganisms , invertebrates [17,28], and vertebrates  as hosts shows that both host and parasite genotypes may interact in the determination of the level of these quantitative traits. In our experiment we found that the intensity of infection (number of infected leaves) and the associated transmission of H. arabidopsis were influenced by strong interaction effects, with nonetheless significant main effects of parasite type on the one hand and host line on the other. The host line always explained a major part of the variance in transmission or infection intensity, with more differences among lines for the transmission (four non overlapping groups) than for infection intensity (three overlapping groups). This impact of host genotype on parasite fitness traits had already been demonstrated in the same association  but also in malaria models [29,30]. Contrary to theses precedent results  however, we demonstrate here, with a larger number of parasite strains, that the parasite also has a significant effect on transmission and infection intensity. Indeed, parasite success differed according to their origin, with isolates from Orsay succeeding on average significantly less well than Laboratory isolates. The two strains from Orsay, though they had been isolated from infected plants growing only a few meters apart, also differed significantly from one to another.
Some host lines suffered more infection than others and some parasite strains had undeniably better average infection success than others. These average results, however, hide large differences among specific combinations, as even "successful" parasite genotypes failed in particular combinations. Both host and parasite identities thus determined parasite fitness traits. Because increased performance on, or adaptation to, a particular host or parasite type will not necessarily imply increased performance in interaction with another host or parasite type, the selective landscape experienced by the two protagonists is unstable. Each new association of genotypes could "erase" any adaptation achieved with a previous partner preventing the appearance of a universally high-performance generalist. Clearly these genotype by genotype interactions will permit the maintenance of genetic variation for characters under selection as can genotype by environment interactions. This is particularly relevant in our pathosystem, as we found such variation in quantitative fitness traits within one natural population. Indeed, the two parasite strains Ors3 and Ors5, collected in the same host population in Orsay, showed significant interactions for infection phenotypes over the range of host lines tested. Though such significant genotype by genotype interactions within populations have been demonstrated for qualitative traits such as infectivity/susceptibility [23,24] there are very few demonstrations for quantitative traits [19,31]. Here we generated new, probably not previously encountered combinations of host and parasite because the six hosts originated from different geographic areas from each other and from the parasite strains we used. Despite the large geographical scale, however, these novel combinations may not be different in kind to those that occur naturally. A. thaliana, though a selfing plant, is highly variable for neutral markers, with much intra- as well as inter-population variation, the latter showing little geographical pattern [32-35]. Therefore a given parasite isolate might be regularly confronted with novel host genotypes, from a nearby population or even from the same population, that differ in qualitative and quantitative resistance.
Trade-off between host and parasite fitness
The perfect organism should produce an infinite number of descendants immediately after its own birth. Indeed, following natural selection's rules, the best rate of reproduction should be strongly selected. So why are we not surrounded by such ideal organisms? A classical response is that reproduction is traded off against other traits that are also necessary for fitness. As an example, the number of descendants can be negatively correlated with their size or quality. In general, two traits are traded off if an increase in one leads to a decrease in the other because both require a common limited resource . Though such trade-offs are logical and compelling, it has proven difficult to find evidence for them in natural systems. Indeed, the comparison of different allocation strategies should be made for individuals possessing the same amount of available resources , which is not often the case in natural systems and even hard to achieve in controlled experiments.
By analogy, parasites are considered to harm their hosts because they divert and consume a common resource that the host also requires for its maintenance and reproduction. Virulence is then the by-product of the parasite using its host's resources for parasite reproduction . As a consequence, we expect that host resources are traded off between host and parasite fitness. Because of the implications for virulence evolution and management, the relationship between virulence (or traits linked to host fitness) and transmission (or parasite propagule production) has been investigated in many theoretical and experimental studies [see [6,38] for a review]. Several experimental studies have indeed shown a positive correlation between virulence and parasite fitness (or a negative correlation between host and parasite fitness) but the question of the relevance of such measures, and their applicability in the real world, remains . Different kinds of approaches have been used to demonstrate this relationship. In general, they compare parasite isolates of varying degree of virulence on a single host line. These isolates could be either different parasite species of the same clade , different experimental treatments [40,41], different genotypes of the same parasite species [42-44] or different lines evolved under experimental selection [9,14,15,45-47].
Here in addition to using different parasite types we also used a number of host lines. This allowed us not only to test the relationship between host and parasite fitness but also to examine whether the nature of this relationship varied with host as well as with parasite identity. Globally, confounding all combinations, we found no correlation between host and parasite fitness [but see [8,20,29]]. However, we found significant heterogeneity for the relationship between host and parasite fitness among host lines, with a significant negative correlation for one of six, the line Gb. Interestingly, the Gb line also happened to be the most susceptible to H. arabidopsis, i.e. had the highest average transmission. In corollary we assessed how the relationship varied for each parasite origin exposed to a range of host lines. The slopes of the relationship between host and parasite fitness did not differ among the diverse parasite origins, although we had previously found, using a slightly different measure of host fitness, variation between two laboratory parasite strains . This and our previous experiment  give inconsistent patterns for some combinations of host and parasite that were used in both experiments. For example the Noco strain transmitted on the Pyr ecotype in this experiment but had failed to do so in our previous experiment. However, these inconsistencies may be due to environmental differences between the two experimental conditions, as similar influences of environmental variation on infection phenotype are known .
Why did we observe a negative relationship between host and parasite reproductive success so rarely? Clearly, if hosts and parasites use the same resource base for reproduction there must be a trade-off between their respective fitness. However, we succeeded in revealing this in only one case. One possibility is that the asexual transmission success of the parasite that we measured in this experiment is not a good estimator of the global fitness via both asexual and sexual stages. Of course, the production of sexual oospores also consumes some host resources, and an additional trade-off between these two modes of transmissions could explain why, in some cases, infections with poor asexual transmission greatly reduced host fitness (and vice versa). However different parasite strategies for allocating host-derived resources to sexual versus asexual reproduction does not explain our observation that the relationship between host fitness and parasite asexual transmission differed among host lines, unless this potential trade-off also depended on host genotype by parasite genotype interactions. We thus propose a more general hypothesis consistent with our results. Superimposed upon underlying variation in host size and resource availability, that were standardized as much as possible in our experiment, shared host and parasite control of the infection phenotype may directly modify this resource pool. As we have discussed above, each combination represents a particular specificity between host and parasite. This specificity could, in addition to influencing the infection phenotype, control the amount of resources available that can be converted into both host and parasite (asexual or sexual) reproduction, thus blurring any trade off for the allocation of these resources. We imagine that particularly compatible interactions reduce the shared resource pool little while conflictual ones leave little resource for either host or parasite. By "compatibility" we mean, here, the adequacy of the host-parasite association from a quantitative point of view, rather than the qualitative ability of the parasite to infect which is how this term is employed in the gene for gene literature. As an example, activating defense systems in plants , insects , and vertebrates  is costly so parasites that induce or hosts that mount a strong defensive reaction will reduce the total resource pool available for reproduction of both parties. In addition incompatible parasites may be less efficient in resources conversion or may consume "dead end" tissues that do not permit their dissemination . We propose that different interactions vary both for their compatibility, hence total resource pool available for host and parasite reproduction, and for the proportion of resources appropriated for parasite reproduction alone (Figure 3 gives a representation of hypothetical partitioning of resources toward host and parasite reproduction). When only this proportion varies, negative correlations are expected between host and parasite reproduction. When only compatibility varies, positive relationships should be found because increasing compatibility would then increase the resource pool available for the parasite but also the residual resources available for the host (hypothetical relationships are represented in Figure 4). Variation in both could generate the large number of possible relationships between host and parasite reproductive success that we observe (Figure 2). Under this hypothesis we propose that negative relationships would be found more often in systems with highly susceptible hosts, which are uniformly compatible to all parasite genotypes, such as the host line Gb in our experiment. The systematic use of highly susceptible hosts in host-parasite studies would then hide the diversity of possible relationship between traits such as parasite and host fitness.