Spatial variation in resistance and virulence in the host–pathogen system Salix triandra–Melampsora amygdalinae

Authors

  • LENA NIEMI,

    1. Department of Ecology and Environmental Science, Umeå University, 901 87 Umeå, Sweden, *Department of Animal Ecology, and †Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences (SLU), 901 83 Umeå, Sweden
    Search for more papers by this author
  • ANDERS WENNSTRÖM,

    1. Department of Ecology and Environmental Science, Umeå University, 901 87 Umeå, Sweden, *Department of Animal Ecology, and †Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences (SLU), 901 83 Umeå, Sweden
    Search for more papers by this author
  • JOAKIM HJÄLTÉN,

    1. Department of Ecology and Environmental Science, Umeå University, 901 87 Umeå, Sweden, *Department of Animal Ecology, and †Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences (SLU), 901 83 Umeå, Sweden
    Search for more papers by this author
  • * PATRIK WALDMANN,

    1. Department of Ecology and Environmental Science, Umeå University, 901 87 Umeå, Sweden, *Department of Animal Ecology, and †Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences (SLU), 901 83 Umeå, Sweden
    Search for more papers by this author
  • LARS ERICSON

    1. Department of Ecology and Environmental Science, Umeå University, 901 87 Umeå, Sweden, *Department of Animal Ecology, and †Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences (SLU), 901 83 Umeå, Sweden
    Search for more papers by this author

Lena Niemi (tel. +4690 7869777, fax +4690 7866705, e-mail lena.niemi@emg.umu.se).

Summary

  • 1Host–pathogen interactions in isolated populations may result in the adaptation of pathogens to local hosts. However, results of earlier studies of local adaptation in plant–pathogen systems have been contradictory and it has been suggested that specific, species characteristics, for example distribution, dispersal and the degree of pathogen dependence on the host, are important for the outcome of host–pathogen interactions. In addition, the scale of the study may influence whether or not local adaptation is found.
  • 2We argue that local adaptation of the pathogen to the host can be expected in a system where: (i) the pathogen is host-specific with a short generation time compared with the host; (ii) populations are isolated, allowing localized evolution to occur; and (iii) the study is performed on a geographical scale exceeding the maximum dispersal range of the interacting species.
  • 3To test these predictions we examined the within- and among-population variation in resistance and virulence of the plant–pathogen system Salix triandraMelampsora amygdalinae. The pathogen occurs throughout the whole distribution range of the host, and the area of interest consists of highly isolated, small natural populations.
  • 4Resistance and virulence differed both within and between populations and all clones showed a unique resistance pattern. We conclude that M. amygdalinae is locally adapted to the host S. triandra as all pathogen populations have a higher probability of infecting sympatric than allopatric hosts. Furthermore, high resistance in the host population was accompanied by a high virulence in the pathogen population, suggesting that high resistance levels in a host population may select for more virulent pathogen populations, or vice versa.
  • 5The nature of host–pathogen interactions differs among systems, and the dynamics of the interaction between S. triandra and M. amygdalinae is governed by characteristics that have resulted in the evolution of local adaptation.
  • 6Thus, when studying local adaptation of pathogens to their hosts it is important to consider the biology, as well as the scale of dispersal and spatial distribution of both host and pathogen.

Introduction

In host–pathogen systems, where organisms intimately interact with one other and the pathogen is dependent on its host for survival, it would be advantageous for the pathogen to adapt to the local hosts. Although several studies have reported that local adaptation has evolved in different host–parasite systems (e.g. Parker 1985; Lively 1989; Ebert 1994; Thrall et al. 2002), other studies have failed to reveal patterns of adaptation to the local hosts (e.g. Davelos et al. 1996; Roy 1998; Kaltz et al. 1999). These contradictory results indicate that local adaptation of parasites to their hosts is not universal (Burdon & Thrall 1999; Gandon 2002; Dybdahl & Storfer 2003).

Different systems of hosts and associated pathogens display large variability in species-specific characteristics, e.g. mode of reproduction, host specificity, distribution and dispersal, that influence the potential for coevolution and the development of local adaptation (Thrall & Burdon 1997; Burdon & Thrall 1999; Thompson 1999). The evolutionary potential of the interacting species is also influenced by generation time, and because pathogens often have shorter generation times than their hosts, this may lead to an evolutioanary advantage (Gandon & Michalakis 2002).

Local adaptation should be more likely to evolve in an interaction where the pathogen is highly dependent on the host, the degree of dependence increasing with host specificity. A host-specific pathogen is totally dependent on a single host species for both survival and reproduction, whereas a non-specific pathogen is able to attack two or more different host species (Burdon 1987). Thus, non-specific pathogens may have a higher probability of finding a host than host-specific pathogens, and therefore there may be less selection for adaptations to specific populations of hosts.

The interaction between the host and associated pathogen is also dependent on their geographical distribution and dispersal abilities. Most, if not all, natural species distributions consist of uneven sized populations with different degrees of connectedness (Thompson 1994, 1999). In each population, the interacting species are subject to local selection processes that may result in local adaptation (Thompson & Burdon 1992; Burdon & Thrall 1999; Thompson 1999). However, in cases where gene flow among populations is large, and/or local extinctions are frequent, the development of local patterns of resistance or virulence (‘virulence’ is defined as the ability of the pathogen to cause disease symptoms, i.e. sporulate, on the host) is prevented (Thompson & Burdon 1992; Gandon et al. 1996; Burdon & Thrall 1999; Lively 1999; Thompson 1999). Furthermore, the relative dispersal of host vs. pathogen is important in determining which organism has the greatest potential to adapt to the other; if host dispersal is limited relative to pathogen dispersal, the pathogen will be able to track local host resistance and adapt accordingly (Gandon et al. 1996).

However, geographical distribution and dispersal may also have implications for detecting coevolution. If the geographical pattern of population distribution includes isolated populations, then different dynamics of, for example, migration, frequency-dependent selection and extinction events in each population will result in populations that differ in genotypic composition (Burdon 1987; May & Andersson 1990; Jarosz & Burdon 1991; Whitlock 1992; Clay & Kover 1996; Kaltz & Shykoff 1998; Burdon & Thrall 1999; Carlsson-Granér & Thrall 2002; Thrall et al. 2002, 2003). These dynamics occur independently in different populations and at different times, making it unlikely that we will discover local adaptations in all populations (Thompson 1999). However, if there is substantial gene flow among populations, the distinctiveness of particular populations can diminish and gene flow may mask local evolutionary events. As a result, local adaptations may not be found within a single population but only between more or less isolated populations (Burdon & Thrall 1999; Thrall et al. 2002).

Thus, to understand local adaptation it is important to have a knowledge of the basic characteristics that govern the interaction and also to control for genetic variations in both the host and the pathogen populations (Thrall et al. 2002). It may also be necessary to broaden the scale from local populations to larger geographical patterns when studying the dynamics of host–pathogen interactions. So far, few studies of coevolution, in terms of local adaptation of pathogens to their hosts, have covered large geographical scales and even fewer have combined comparisons of parasite performance on allopatric and sympatric hosts with comparisons of allopatric and sympatric pathogens on a given host (but see Thrall et al. 2002).

We examined the interaction between the host plant Salix triandra and the pathogen Melampsora amygdalinae, a system for which it is possible to work on a large geographical scale. It is also a system where both the pathogen and the host have the potential to reproduce sexually, thus allowing continuous genetic recombination; moreover, the pathogen is host-specific and host populations are isolated. Our aim was to investigate whether: (i) there is spatial variation in resistance and virulence, and (ii) the pathogen has evolved adaptations to local host resistance. We hypothesize that the interaction between S. triandra and M. amygdalinae is governed by characteristics and processes that promote coevolution, and that local adaptation of the pathogen to the host will be found in this system.

Materials and methods

the host plant

Salix triandra L. (Salicaceae) is a dioecious plant that reproduces both sexually and asexually (broken-off twigs or branches are easily rooted). It is widely distributed from Japan to Western Europe. The main distribution range is in Russia, where S. triandra forms large continuous stands along rivers, over vast areas (Skvortsov 1999). The western limit of its continuous distribution is the Karelian Isthmus (Russia), the Baltic states and eastern Poland. Further west, S. triandra is rare, and occurs in more or less isolated populations. Historically, S. triandra has probably been part of the Scandinavian flora since the Weichselian glaciation, which started to withdraw approximately 10 000 years ago (Lundqvist & Robertsson 1994). Salix triandra probably colonized Sweden from the east and established itself on river banks near the coastline. Since then, isostatic uplift has raised this former coastal region to higher altitudes, and S. triandra has spread downstream. Salix triandra now inhabits sandy inundated riverbeds and is native to six river valleys, where it reaches high densities. These river systems are well separated, and the three Swedish rivers included in the study (the Dalälven, Öreälven and Torneälven rivers) are located approximately 450 km from each other. The distance from the closest Swedish population (Dalälven river) to the Russian population studied, situated at Oredezh river valley south of St Petersburg, is approximately 750 km. Salix triandra produce numerous seeds in these populations. However, the rate of seedling establishment is not known but seems to be rare. Vegetative dispersal is more common and is mainly directed downstream (our personal observations). In addition, because S. triandra is insect pollinated (e.g. by bumblebees and syrphids), opportunities for pollen transfer between river systems are limited.

the pathogen

Salix triandra is attacked by the rust fungus, Melampsora amygdalinae Kleb. (Wilson & Henderson 1966; Gjaerum 1974), a non-systemic annual rust that is autoecious and host specific to S. triandra. It occurs throughout the distribution range of the host. The life cycle of M. amygdalinae is macrocyclic and includes a sexual stage in spring, when overwintering teliospores germinate and give rise to basidiospores that infect the host. The haploid basidiospores germinate to produce mycelia that form spermogonia. After dikaryotization by transfer of spermatia to the receiving structures of compatible mating types, aecia are formed. From these, aecidiospores are dispersed, which, after germination, form uredia on infected host tissue. During summer, the disease symptoms, identified as pustules with orange uredospores on leaves and shoots, are seen. The uredospores are produced asexually and several generations can appear during one season. The spores are wind-dispersed and can infect the same or different individuals. In years when conditions for infection are favourable, the current, annual shoots can be covered with pustules, which may result in premature leaf shedding (our personal observations). In autumn, teliospores are produced. M. amygdalinae is also able to overwinter as mycelium in S. triandra buds (Gäumann 1959; Wilson & Henderson 1966).

inoculation experiments

Cuttings of S. triandra were collected in spring, prior to leaf burst, from three river populations in Sweden, namely the Dalälven (60°34′ N, 17°27′ E), Öreälven (63°43′ N, 19°31′ E) and Torneälven (66°23′ N, 23°40′ N) populations, and from one population from the Oredezh River, south of St. Petersburg in Russia (58°49′ N, 30°20′ E). The cuttings were transported to a glasshouse at the university campus and rooted in water. When leaves started to develop, the cuttings were planted individually in pots. We selected viable plants, comprising 29 clones (11 males and 18 females) from Öreälven, 19 clones (10 males and 9 females) from Dalälven, 13 clones (10 males and 3 females) from Torneälven, and 17 clones (8 males and 9 females) from the Russian population. These plants were kept in 3-L plastic pots in the experimental garden. The potted clones were tested for resistance against rust originating from Öreälven, Dalälven, Torneälven and Russia.

The rust spores (uredospores) were collected early in the season when individual pustules could be easily separated. Leaves with single pustules of approximately 3 mm in diameter were picked, returned to the laboratory and stored separately. We used eight individual rust pustules from Öreälven, six from Dalälven, six from Torneälven and five from Russia. The pustules were allowed to dry for 24 h at room temperature. Separate pustules are hereafter referred to as ‘rust isolates’.

To test the resistance/virulence of the host clones/rust isolates, different rust isolates were inoculated onto cuttings of each of the potted host clones. Each rust isolate was inoculated onto three replicates of all host clones. For these experiments, we cut three small pieces, each with 1–2 attached leaves, of the current annual shoots from each of the 78 clones. The cuttings were randomly inserted into a 25 × 50 × 0.5 cm piece of Styrofoam perforated with holes. The Styrofoam with the 234 cuttings was floated on water in a larger container. Each rust isolate was put into a glass vial and water was added (10 mL). The water with the suspended pustules was thereafter sprayed directly onto all leaves in the container. This procedure was repeated for all rust isolates, which were kept in different containers. A spore count of pustules with a diameter of 3 mm resulted in between 3 × 104 and 4 × 104 spores per mL water. This is sufficient to obtain susceptibility responses and the amount of spores is also within a range that makes the different inoculations comparable (Pei et al. 2003). The containers with the cuttings were covered with transparent plastic sheets to ensure high humidity, and kept in a glasshouse. Two weeks after application of the spores, each cutting was checked for disease symptoms. At this stage, plants were considered to be susceptible if pustules occurred on leaves of any of the three replicates; if pustules were absent from all three replicates the plant was scored as resistant. No new pustules developed after 2 weeks.

Results from the inoculations with rust isolates from Russia were limited, as these isolates were attacked by an unidentified hyperparasite (neither taxonomy nor systemic position could be determined). This hyperparasite rapidly consumed the rust spores and made scoring possible in only two containers representing two isolates. Therefore, the results concerning the inoculation with Russian isolates are not included in the statistical analyses.

statistical methods

For the statistical analysis of our data we have used a generalized linear mixed model (GLMM) with a logit link function, logit(µ) = log(µ/(1 − µ)), that transforms the binary data interval (0, 1) into (–∞, ∞). In ordinary linear mixed models with unbalanced data sets, significance of fixed effects is evaluated using approximate Wald tests (Pinheiro & Bates 2000). However, because our data had variables with non-normal response, the asymptotic properties of the Wald test no longer hold in a GLMM (McCulloch & Searle 2001). Instead we adopted a Bayesian approach where Gibbs sampling is used for inference in the GLMM (Zeger & Karim 1991; Clayton 1996).

Inference in Bayesian statistics is based on the posterior distribution, which combines the information from prior distributions of the parameters in the model with the likelihood of the data (e.g. Beaumont & Rannala 2004; Ellison 2004; Gelman et al. 2004). In our analyses we use Gibbs sampling, which is a Markov Chain Monte Carlo method (MCMC), to obtain samples from the posterior distribution. Models with different degrees of complexity can be compared using model selection, and the deviance information criterion (DIC) has recently been proposed for comparing the goodness of fit for hierarchical models (Spiegelhalter et al. 2002). The DIC criterion is defined as

image( eqn 1)

where D(θ̄) is the deviance of the posterior mean of the parameters and pD is a measure of the effective number of parameters in the model. Hence, DIC can be thought of as a measure of model fit that adjusts (penalizes) for the fact that more complex models fit data better. Smaller values of DIC indicate better fit to the data.

Three separate GLMM model comparisons were performed because of missing results from the inoculation with isolates from the Russian population. Males and females of the host plant showed no significant difference in the response to the rust, so data were pooled in all comparisons. The first comparison was based on the three Swedish populations and their respective pathogens. To find the model with the best fit (smallest DIC), we sequentially added factors from the mean up to the full model,

image(        eqn 2)

where pop is host population, path is pathogen population and clone is host individual. The factors popi, pathj and popi × pathj are fixed factors, and clonek a random factor. Of primary interest is the interaction popi × pathj because this factor will tell us whether there is any local adaptation.

In the second comparison, the DIC of a model that includes the mean and clone factor was compared with the DIC of a model where the population factor was added. This comparison was based on the total data (including the Russian host population), but the Swedish populations were merged into one level. This comparison evaluated whether the Russian population responded differently from the Swedish populations to all pathogens. The third model comparison was based only on the Russian data, to evaluate whether different pathogen populations had different infection probabilities in this population. For the model with the best fit in each comparison, we estimated fixed effects (on the logit scale) and back-transformed means for each level of the fixed factor(s).

Gibbs sampling GLMMs were implemented using the program WinBUGS14 (Spiegelhalter et al. 2003). The first level of each of the fixed effects was constrained to be zero. The prior for the mean and the non-constrained fixed effects were chosen to be non-informative (Normal[0, 104]). The prior for the random effects was constructed by using a hyper prior for the variance (Z. hypUniform[0, 7.5]), and then sampling from Normal[0, Z. hyp]. This prior can also be considered to be non-informative. Based on preliminary runs of MCMC that were inspected and checked for convergence with the tools available in WinBUGS14, we decided to run one chain per model of length 110 000, discarding the first 10 000 iterations as burn-in and saving each 10th iteration (yielding a total of 10 000 iterations).

Results

All host clones, including one Russian clone that was resistant to all rust isolates, displayed unique patterns of resistance. It is notable that none of the clones from any of the four host populations was susceptible to all isolates. Of the 20 rust isolates used, 20 unique infection profiles were detected. No rust isolate was able to sporulate on all clones or on more than 24% of the Russian clones. The average virulence of the Swedish rust isolates and the resistance of the Swedish host populations in sympatric and allopatric inoculation combinations is shown in Figs 1 and 2, and the average virulence on Swedish compared with Russian hosts in Fig. 3.

Figure 1.

Virulence (mean ± SE) of the rust populations from Dalälven (n = 6), Öreälven (n = 8) and Torneälven (n = 6), measured as frequency of successful inoculations, on sympatric and allopatric host clones. White bars represent the Dalälven host population, light grey the Öreälven host population and dark grey the Torneälven host population.

Figure 2.

Resistance (mean ± SE) of the host populations from Dalälven (n = 15 (S), 19(A)), Öreälven (n = 29(S), 28(A)) and Torneälven (n = 11(S), 13(A)), to sympatric (S) and allopatric (A) rust isolates.

Figure 3.

Virulence (mean ± SE) of the three Swedish rust populations (n = 20, pooled data) on Swedish (SWE) and Russian (RUS) host clones.

The results of the first GLMM model comparison of the Swedish data show that the full model had the lowest DIC (1351.0) and therefore also the best fit to the data (Table 1). In addition, the predicted means of the sympatric combinations are higher than the predicted means of the allopatric combinations (Table 2). Taken together, this is strong evidence of local adaptation.

Table 1.  Model comparison of the Swedish populations using DIC. Factor popi = host population, pathj = pathogen population and clonek = host individual. A lower DIC value indicates that the model has a better fit. The penalty factor (pD) adjusts (penalizes) for the increase of model complexity [DIC = D(θ̄) + 2pD]
ModelDICpD
µ1424.00.9966
µ+popi1424.02.999
µ+popi+pathj1405.05.003
µ+popi+pathj+popi × pathj1368.09.024
µ+popi+pathj+popi × pathj+clonek1351.036.13
Table 2.  Summary statistics regarding the posterior distributions of the parameters in the full model of the Swedish data. The different populations are indicated with letters: D = Dalälven, Ö = Öreälven, T = Torneälven. pop = host population, path = pathogen population. Fixed effects with credible intervals excluding zero are indicated with an asterisk. The predicted means indicate the probability of infection by the different rust populations. Predicted means for sympatric combinations of host and pathogen are indicated in bold type
 Fixed effects (logit scale)Predicted means (original scale)
MedianCI 2.5%CI 97.5%MedianCI 2.5%CI 97.5%
  • *

    Credible intervals excluding zero.

  • ×10−5.

  • a

    Predicted means of the probability of infection (virulence) by isolates from the rust population from Dalälven (a), Öreälven (aa) or Torneälven (aaa).

µ−0.219−0.7180.278   
Parameters
 popD0−1.941.94†0.34510.22120.4842
 pop−0.626−1.2740.02420.32480.20730.4629
 popT−1.442*−2.299*−0.617*0.28720. 17510.4092
 pathD0−1.951.91†0.31a0. 19710.4446
 path−1.516*−2.188*−0.881*0.2101aa0.12220.3247
 pathT−0.082−0.670.5250.4384aaa0.29610.5765
 popD × pathD0−1.951.960.44490.21020.7083
 popD × path0−1.941.930.15090.051610.3442
 popD × pathT0−1.991.970.42470. 19850.6871
 pop × pathD0−1.951.930.30180.1280.55
 pop × path1.655*0.897*2.484*0.32850.14370.5837
 pop × pathT0.187−0.5880.9370.32440.13850.5781
 popT × pathD0−1.981.960.16060.05320.3855
 popT × path1.238*0.1603*2.287*0.12640.041410.311
 popT × pathT1.952*0.9751*2.965*0.55420.28730.7962
 inline image0.2290.06940.520   

Comparison of the back-transformed predicted means revealed that the predicted means for isolates from Torneälven were higher than for isolates from the other populations, and that both the Dalälven and the Öreälven isolates had a lower probability of causing infection in the Torneälven host population than in the other allopatric population (Table 2). Together, this suggests higher virulence among the Torneälven isolates than among the other isolates and higher resistance among Torneälven clones than among the other Swedish clones (see Figs 1 and 2).

In the second model comparison we included the data from the Russian host population and found that the model that included the population factor [logit(p0) =µ + popi + clonek] had a smaller DIC than the model excluding it (DIC was 1638.0 and 1655.0, respectively). The median of the posterior distribution of the fixed effect of the Russian population (compared with the Swedish populations chosen as corner contrasts) was −1.393 and the 95% credible interval deviated considerably from zero (−1.875, −0.9474). The back-transformed predicted means were 0.3062 (95% CI: 0.1464, 0.5259) and 0.09818 (95% CI: 0.03712, 0.2242) for the Swedish and Russian populations, respectively. Together, this shows that the probability of the Swedish pathogens causing infection is considerasbly lower in the Russian population than in the Swedish populations (Fig. 3).

The ability to infect the Russian population did not differ between the Swedish pathogens, i.e. on the third data set the model without the pathogen factor had the best fit (DIC was 230.0 compared with 232.4).

Discussion

In the present system, there was considerable variation within and between populations in terms of both resistance and virulence. This is in agreement with results from other studies and can be explained by the spatial isolation of populations, which allows the dynamics of the host–pathogen interaction to work independently in different populations (Burdon 1987; Burdon & Jarosz 1992; Gandon et al. 1996; Harrison & Hastings 1996; Burdon & Thrall 1999; Carlsson-Granér & Thrall 2002). Our results suggest that the pathogens have evolved adaptations to local host populations. All three rust populations had higher probabilities of infecting sympatric rather than allopatric clones. Similarly, clones from Torneälven and Dalälven were more resistant to allopatric than sympatric isolates of the pathogen, although clones from Öreälven seemed to be equally resistant to both allopatric and sympatric pathogens.

The studied host–pathogen system has several important characteristics that allow coevolutionary processes and local adaptations. First, there was considerable variation in pathogen virulence and host resistance, both between individuals and between the sampled populations. It is notable that in no case did we find that different host genotypes or rust isolates displayed the same response pattern in the inoculation experiments. Second, the pathogen is host-specific and thus dependent on this host species for survival. Third, the Swedish populations are isolated by distances great enough to reduce the degree of gene flow following dispersal, at least in the case of the host.

In some systems, local adaptation may favour the host adapting to a local pathogen, i.e. ‘maladaptation’ (Kaltz et al. 1999). In our studied system the pathogen reproduces sexually, has a generation time shorter than that of the host, and has a larger population size and dispersal range than the host. Together this creates a large evolutionary potential and, consequently, it is the pathogen population that may respond to the selection pressure from the host by evolving adaptation strategies to host resistance.

If selection for resistance and virulence is not reciprocal, or is more or less intense in some populations than in others, it may contribute to a pattern where the levels of resistance and virulence may differ between populations (Thompson 1999). It has previously been suggested that high levels of host resistance may select for high levels of virulence in the associated pathogen population, or vice versa (Thrall et al. 2002; Thrall & Burdon 2003). This is supported by data relating to the Swedish rust populations. The rust population from Torneälven was the most virulent of the Swedish rust populations, and in addition, the host population from Torneälven showed high resistance to allopatric rust isolates. By contrast, lower levels of resistance found in the other two Swedish host populations were associated with lower levels of virulence.

The probability of Swedish rust isolates causing disease on Russian hosts was much lower than on Swedish hosts, and the ability to infect Russian clones was equally low for all Swedish pathogens. The great differences between Russian and Swedish host clones seen in our study may well be due to the fact that the Swedish populations have little, or no, connection with the main distribution range in Russia.

All three pathogen populations displayed differences in terms of the probability of infecting sympatric vs. allopatric host clones, but the response of the three host populations was not consistent. The host population from Öreälven differed from the other host populations in that the response to sympatric and allopatric rust isolates did not differ from each other. Such results could be a consequence of, for example, selection for resistance differing among populations or the fluctuating dynamics of host–pathogen interactions temporarily putting populations in a phase where results of coevolution cannot be detected (Kaltz & Shykoff 1998; Thompson & Cunningham 2002). This further emphasizes the importance of considering both pathogen virulence and host resistance in studies of host–pathogen interactions (Thrall et al. 2002). It also indicates the necessity not only of using a large spatial scale in host–pathogen studies, but also of including several populations, because studies on a smaller scale or of single populations do not always reflect the processes that take place on a regional or larger scale (Thompson 1999).

In conclusion, the interaction between S. triandra and M. amygdalinae has characteristics that have enabled local adaptation of the pathogen to the host in the three Swedish populations. Future studies of different host–pathogen systems on both smaller and larger scales are needed in order to gain an understanding of the dynamics and processes that create the different trajectories of evolution in populations of hosts and their associated pathogens.

Acknowledgements

We would like to thank J. Burdon for helpful comments on earlier versions of this paper and P. Ingvarsson for comments on the statistics. We are grateful to H. Roininen, University of Joensuu, for providing us with the Russian host plants and to A. Kovalenko from the Botanical Institute in St Petersburg, Russia, for valuable help when we visited St Petersburg. A.W. was financially supported by the Oscar and Lili Lamm foundation.

Ancillary