Spatial scale of local adaptation in a plant-pathogen metapopulation


Anna-Liisa Laine, Metapopulation Research Group, Department of Biological and Environmental Sciences, PO Box 65 (Viikinkaari 1), FI-00014 University of Helsinki, Finland.
Tel.: +358 9 191 57743; fax: +358 9 191 57694;


The rate and scale of gene flow can strongly affect patterns of local adaptation in host–parasite interactions. I used data on regional pathogen occurrence to infer the scale of pathogen dispersal and to identify pathogen metapopulations in the interaction between Plantago lanceolata and its specialist phytopathogen, Podosphaera plantaginis. Frequent extinctions and colonizations were recorded in the metapopulations, suggesting substantial gene flow at this spatial scale. The level of pathogen local adaptation was assessed in a laboratory inoculation experiment at three different scales: in sympatric host populations, in sympatric host metapopulations and in allopatric host metapopulations. I found evidence for adaptation to sympatric host populations, as well as evidence indicating that local adaptation may extend to the scale of the sympatric host metapopulation. There was also variation among the metapopulations in the degree of pathogen local adaptation. This may be explained by regional differences in the rate of migration.


Coevolution is considered to be one of the major forces in organizing Earth's biodiversity by linking the genomes of interacting species (Thompson et al., 2002). In the coevolutionary arms race between hosts and parasites, conventional wisdom holds that parasites with relatively short generation times and larger population sizes evolve faster than their hosts, quickly overcoming new host resistance adaptations. Such asymmetry should lead to local adaptation of the parasite, indicated by a higher fitness in sympatric than in allopatric host populations [Kaltz & Shykoff, 1998; see Gandon et al. (1998) for other definitions of local adaptation].

Local parasite adaptation is considered as indisputable evidence of coevolution and it has stimulated theoretical and empirical work on a wide range of biological systems. Indeed, several studies – including parasites of both plants and animals – have reported evidence for local adaptation (Parker, 1985; Lively, 1989; Ebert, 1994; Lively & Dybdahl, 2000). However, other studies have failed to find any evidence for parasite local adaptation (Davelos et al., 1996; Dufva, 1996; Strauss, 1997; Mutikainen et al., 2000) or even found parasites to be maladapted (Parker, 1989; Kaltz et al., 1999; Oppliger et al., 1999). In yet other systems evidence for variation among parasite populations in the level of local adaptation to the same host species has been found (Morand et al., 1996; Imhoof & Schmid-Hempel, 1998; Koskela et al., 2000).

Parasite local adaptation is not always the expected outcome of host–parasite interactions. Highly specific parasites with severe fitness effects on their hosts are more likely to become locally adapted than generalist parasites or those with little impact on their hosts fitness (Lively, 1999; Gandon, 2002). Similarly, parasites that migrate more than their hosts show greater potential of being locally adapted (Gandon et al., 1996; Gandon, 2002), although this advantage may break down for very large scales of parasite dispersal (Thrall & Burdon, 2002). Even in a situation where the host and the parasite are reciprocally coevolving, lack of evidence for local adaptation or maladaptation is not surprising given the dynamic nature of coevolution. By definition, in antagonistic coevolutionary interactions one species evolves in specific ways to mitigate the adaptation of the other species. Therefore, if multiple populations are studied at one point in time, different parasite populations may be in different phases of local adaptation (Kaltz & Shykoff, 1998; Thompson et al., 2002).

Negative results concerning parasite local adaptation may also reflect local coevolutionary processes operating at either larger (Hanks & Denno, 1994; Imhoof & Schmid-Hempel, 1998; Thrall et al., 2002) or smaller spatial scales (Parker, 1985; Karban, 1989; Lively & Jokela, 1996; Ebert et al., 1998) than at the scale of local populations. The spatial scale of dispersal, and hence the scale of differentiation of parasite populations, is likely to determine the scale of local adaptation (Thrall et al., 2002). In metapopulations, where the local populations are connected by gene flow, local parasite populations are composed of genotypes that have been locally coevolving with the host populations as well as recently arrived genotypes that have been coevolving with other host populations. Parasite populations linked by frequent migration might adapt to the overall frequency of particular host genotypes in the metapopulation (Dybdahl & Lively, 1996).

With the notable exception of Thrall et al. (2002), very little effort has been put into incorporating parasite spatial dynamics into experimental designs while investigating local adaptation by parasites. Here, I investigated pathogen local adaptation in the Plantago lanceolataPodosphaera plantaginis metapopulation in the Åland archipelago in southwest Finland. Infection by P. plantaginis is not lethal to the host plant, but it constitutes a stress factor that reduces growth and reproduction of the host. When infection coincides with other stressful environmental conditions, it may even induce host mortality (Laine, 2004). The interaction between P. lanceolata and P. plantaginis is characterized by strain-specific resistance in the host (Laine, 2004), suggesting a gene-for-gene type of interaction (Burdon et al., 1996). The host populations are highly variable and differentiated in their resistance to the pathogen (Laine, 2004). In this study system, there are approximately 3300 host populations, which have all been mapped within an area of 50 × 70 km, and they are surveyed annually for the occurrence of the pathogen (Fig. 1; Laine, 2004). The host populations occur in a highly fragmented landscape and only a small fraction (c. 5%) of them is infected at a given time. The local pathogen populations go typically through a decline at the end of the growing season as the perennial host dies back to rootstock, resulting in frequent local extinctions of the pathogen. The extinctions are compensated for by colonizations of wind-dispersed spores (Laine, 2004).

Figure 1.

The distribution of Podosphaera plantaginis in the networks of Plantago lanceolata populations in the Åland Islands in 2001. Black symbols depict infected host populations and white symbols uninfected host populations. The circles show locations of the four metapopulations, which are depicted in the right hand panel, where the pathogen population (circled) was sampled for the local adaptation study.

The frequent local extinctions and colonizations of the pathogen suggest that the pathogen may be more likely to adapt at the metapopulation scale rather than at the scale of individual host populations. To test this hypothesis I used empirical data on the occurrence of the pathogen to identify the spatial scale of metapopulations, and performed cross-inoculation experiments to compare the performance of pathogen populations on their sympatric host populations, sympatric host metapopulations and allopatric host metapopulations. The performance of the pathogen on a given host line was measured both as the ability to cause infection (referred to as infectivity throughout the paper) and by the severity of infection (referred to as infection stage). In brief, the aim of this study was to determine the spatial scale of local parasite adaptation.

Materials and methods

The study system

Plantago lanceolata L. (Plantaginaceae) is a wind-pollinated monoecious perennial herb that is an obligate outcrosser due to self-incompatibility induced by protogyny (Cavers et al., 1980). It reproduces via seeds that germinate in the following summer or remain in the soil seedbank, and clonally by producing side-rosettes. In Finland P. lanceolata occurs in the south-western archipelago and on the south coast, where it is rare. The fungal pathogen, P. plantaginis (Erysiphales), is a powdery mildew that grows on the surface of host tissue with only feeding organs inside the plant cells. It is an obligate pathogen requiring living host tissue throughout its life-cycle. Powdery mildews are not considered lethal pathogens, but infection may reduce the growth of the host plant (Bushnell, 2002). In Åland, infected host individuals had high mortality during drought periods (Laine, 2004). During the growing period the fungus propagates as clonally produced conidia that are dispersed by wind. Sexually produced resting spores (cleistothecia) appear towards the end of the growing season in August–September.

The Plantago–Podosphaera study system in the Åland Islands comprises approximately 3300 P. lanceolata populations (Fig. 1). The host populations are typically small, occurring on dry meadows separated by unsuitable habitat consisting of arable land, human settlement, roads and forest (Nieminen et al., 2004). Metapopulation studies of the Glanville fritillary butterfly (Melitaea cinxia) are conducted in this same meadow network, and every year in early September all known host populations were surveyed for the presence of the butterfly and the powdery mildew (Nieminen et al., 2004). At this time of the year the fungus was clearly visible on host leaves as white-greyish mycelial growth with conspicuous black resting spores. A leaf sample was collected from each infected site for subsequent microscopic identification.

K-function analysis and pathogen metapopulations

I tested whether the spatial distribution of infected host populations among all host populations was more aggregated than would be expected by chance. An aggregated pattern of occurrence would be indicative of metapopulation structure in the fungus and the degree of aggregation would also help define the spatial scale of these metapopulations. This was done using data for year 2001 with a spatial K-function statistic (reduced second moment measure; Bailey & Gatrell, 1995). The result is expressed as the probability of finding an infected host population at distance d from a randomly chosen infected population. One-sided 95% confidence limits were based on a random distribution generated by 100 simulations. When the observed probability is outside the one-sided 95% confidence limits the degree of pathogen aggregation is considered to be statistically significant.

Four metapopulations were chosen in 2002 to study the dynamics of P. plantaginis in greater detail. In order to objectively classify host populations into metapopulations a hierarchic cluster analysis as implemented in the program SPOMSIM was used (Moilanen, 2004). Based on the results of the K-function analysis I chose four habitat patch networks with a radius of approximately 500 m, each of which consisted of five P. lanceolata populations. Furthermore, in each metapopulation at least one of the host populations had to be infected in 2001 (see Fig. 1). Six of the host population networks fulfilled these criteria, and four of them were chosen to represent different parts of the Åland Islands. In the period 2002–2004, all host populations in these four metapopulations were surveyed in August, and the coverage of host plants and the proportion of infected host plants were estimated in 10 × 10 m grid cells.

Laboratory inoculation experiment

I tested local adaptation of four P. plantaginis populations in a laboratory inoculation experiment. One host population that had been infected throughout 2001–2004 was chosen from each metapopulation (population IDs 114, 382, 2224 and 3270; Table 1; Fig. 1). In September 2003 seeds were collected from all 20 host populations in the four metapopulations. Twenty-five seeds were collected from the four populations from which the fungus had been sampled (114, 382, 2224 and 3270), whereas 10 seeds were collected from each of the four other host populations in each metapopulation. Additional seeds were collected from populations 114, 382, 2224 and 3270 for the purification and propagation of the pathogen strains. Seeds were collected into paper envelopes and stored at room temperature. On 2–4 June 2004 seeds were germinated by placing them in 0.8 L pots in a 30% vermiculate-70% potting soil mixture in greenhouse conditions of 16 h of light and at +22 °C. Of the maternal lines producing seedlings, 20 lines were chosen to represent the sympatric host population in each network and 20 lines were chosen to represent the other populations in the metapopulation, each population represented by five maternal lines. Altogether, the experiment consisted of 160 maternal lines.

Table 1.  Prevalence (%) of Podosphaera plantaginis in Plantago lanceolata populations in the four metapopulations in years 2002–2004. In 2001 only the pathogen incidence (presence) in the host populations was recorded. The pathogen populations sampled for the local adaptation study are depicted in bold.
 Geta metapopulationLumparland metapopulationEckerö metapopulationFlaka metapopulation
Colonizations 34 1 2
Extinctions 13 0 0

The pathogen populations 114, 382, 2224 and 3270 were sampled for the inoculation experiment on 29 July 2004. Because disease prevalence was quite low (Table 1), only five to seven infected plants were sampled as detached leaves not to affect too much the local dynamics. The detached leaves were stored in plastic bags with moistened tissue paper and transported in a cool box to the laboratory in Helsinki. Within 48 h of sampling, spores from the detached leaves were brushed over healthy leaves taken from individuals grown in the greenhouse. The leaves were inoculated in Ø 9 cm Petri dishes and placed in a growth chamber at 20 ± 2 °C with 16L/8D photoperiod. After 12 days single, discrete fungal colonies growing on the inoculated leaves were isolated and again spores were brushed over healthy leaves. This procedure was repeated three more times to assure that isolates used in the experiment were genetically homogeneous (Nicot et al., 2002).

Distinct and presumably genetically differentiated lines were chosen for the experiment based on their ability to infect the plants used for the purification procedure. As could be expected from the small sizes of the pathogen populations (see Table 1), the samples did not seem to contain a high diversity of pathogen isolates. In pathogen populations 114, 382 and 3270, three pathogen strains that were not identical in their infectivity on the same set of host genotypes were distinguished. From pathogen population 2224, I was able to identify only two distinct pathogen strains. Following the purification procedure, repeated cycles of inoculations were performed to obtain adequate stocks of sporulating fungal material for the inoculation trials. The experiment consisted of 160 host lines, which were all reciprocally infected with the 11 pathogen isolates resulting in 1760 inoculations. The experiment had to be done in two parts, on 27 September and 11 October, because scoring of the infection would not have been possible to accomplish within a time frame that would have produced comparable results. Plants were divided randomly among the two inoculation dates.

In the experiment, plants were exposed as detached leaves to a single pathogen strain at a time. The leaf was placed in a Ø 9 cm Petri dish on moist filter paper, and spores from an infected leaf were gently brushed with a fine paintbrush over the entire leaf surface (Nicot et al., 2002). Fungal colonies of approximately 1 cm in diameter and similar age (12–15 days) were used for the experiment to obtain as similar spore densities as possible. Inoculated dishes were incubated in the growth chamber at 20 ± 2 °C with 16L/8D photoperiod for 10 days, after which the infection status was scored using a microscope. The pathogen was scored as infective on the host plant line when there was mycelial growth and conidia on the detached leaf and no severe chlorosis. When there was a chlorotic response and no mycelial growth could be observed under a microscope, or it had died off at an early stage, the pathogen was scored as noninfective on that host. The infection stage was assessed 10 days after inoculation, ranging from mycelial growth to heavy spore production [1 = sparse mycelium but no conidia, 1.5 = mycelium producing very few conidia and colonies visible only under a dissecting microscope, 2.5 = colonies visible with the naked eye but exhibiting sparse sporulation, 3 = profuse sporulation on colonies of moderate size (<5 mm diameter) and 4 = profuse sporulation on large colonies (>5 mm diameter): key adapted from Bevan et al., 1993].

Statistical analyses

Infectivity of the pathogen populations was analysed with a generalized linear mixed model with a binomial error distribution and a logit link function (GLIMMIX macro of SAS version 8.02; SAS Institute, 1999). In the model, the host individuals and pathogen strains were defined as random factors. They were nested within the source of host population and pathogen population, respectively. Pathogen population, host population and host metapopulation were fixed factors in the model, with host population nested within the respective metapopulation. The level of sympatry–allopatry was a fixed factor, nested within pathogen population (1 = sympatric host population, 2 = sympatric host metapopulation and 3 = allopatric metapopulation).

In the analysis of infection stage of P. plantaginis the response variable consists of the five infection categories described above. As the infection categories were ordered (from low to high), I used an ordinal logistic regression as implemented in JMP 5.0.1/SAS (JMP, 2002). In the model the explanatory fixed variables were host population nested within the metapopulation, the pathogen population and level of sympatry–allopatry (see above) nested within pathogen population. Contrast tests were used to find out whether the infectivity and infection stage of P. plantaginis populations differed among their respective sympatric populations, sympatric metapopulations and allopatric metapopulations. Finally, the correlation between infectivity and infection stage of the four pathogen populations on the 20 host populations was tested.


Spatial distribution of infected populations

The occurrence of P. plantaginis in the P. lanceolata meadow network was more aggregated than would be expected from a random distribution. The infected populations were particularly aggregated at distances shorter than 500 m, and aggregation was statistically significant up to 1 km (Fig. 2). The probability of finding an infected population at 500 m distance from a randomly chosen infected population is 2.4 times higher than the average infection prevalence in the Åland Islands (Fig. 2).

Figure 2.

K-function analysis of the distribution of infected populations in the entire Plantago lanceolata population network in the Åland Islands in 2001. The solid line shows the probability that a population is infected at distance d from a randomly chosen infected population. The dotted line gives the average infection level in Åland and the dashed lines represent one-sided 95% confidence intervals.

Pathogen dynamics in the four metapopulations

The dynamics of the pathogen in the four metapopulations is summarized in Table 1. In all four networks there was at least one host population that remained infected throughout the period 2001–2004 (Table 1; Fig. 1). In the Geta metapopulation, there were two host populations that remained infected throughout the survey period and the population with higher pathogen prevalence (114) was chosen for the local adaptation experiment. In the Geta and Lumparland metapopulations, extinction and colonization events were recorded every year. In both cases most extinctions were recorded in 2003 (Table 1), a summer characterized by an unusually severe drought. This drought caused a widespread decline in host density (Laine, 2004), and as a result, the prevalence of the pathogen declined in all the four metapopulations in this study (Table 1). Pathogen prevalences remained low in 2004, although there were altogether seven colonization events recorded in the four metapopulations (Table 1). A more detailed analysis of large-scale regional dynamics of P. plantaginis will be reported elsewhere (A.-L. Laine & I. Hanski, unpublished data).

Pathogen local adaptation

The infectivity of the four pathogen populations was significantly affected by the degree of sympatry of the host populations (Table 2; Fig. 3). Pathogen populations 114 and 382 did not differ in their ability to infect hosts from sympatric populations and sympatric metapopulations, but both pathogen populations were significantly less effective in infecting hosts from the allopatric metapopulations (Fig. 3). This result suggests that the pathogen is locally adapted at the metapopulation level (Fig. 3). Pathogen population 2224 was locally adapted to its sympatric host population (Fig. 3), but in contrast to the other pathogen populations, its infectivity was higher on hosts from the allopatric than from the sympatric metapopulation (Fig. 3). Finally, no significant pattern of local adaptation could be detected in pathogen population 3270 when adaptation was measured as infectivity. However, note that in all four pathogen populations infectivity was highest in the sympatric host population (Fig. 3). The other fixed explanatory variables had no significant effect on the infectivity of pathogen populations (Table 2).

Table 2.  Results of the generalized linear mixed model of infectivity and the ordinal regression model of infection stage of Podosphaera plantaginis.
SourceInfectivityInfection stage
Host population160.880.5911.870.0001
Pathogen population30.310.81559.910.008
Figure 3.

Average infectivity of the four Podosphaera plantaginis populations on their sympatric host populations, sympatric metapopulations and allopatric metapopulations.

The stage of infection development of the pathogen strains was significantly affected by the degree of sympatry of the host populations (Table 2; Fig. 4). However, only for population 382 was infection stage significantly higher on hosts from its sympatric population than on hosts from allopatric metapopulations (Fig. 4). Infection stage differed significantly among host populations within metapopulations and among pathogen populations (Table 2). The stage of infection was highest in population 3270 (2.51 ± 0.06 SE) and lowest in population 2224 (1.8 ± 0.06 SE). The average infection stage of pathogen populations 114 and 382 were 1.88 ± 0.05 and 2.16 ± 0.06, respectively. There was no correlation between infectivity and infection stage of the four pathogen populations (‘Population 114’r18 = 0.039; P = n.s.; ‘Population 382’r18 = 0.352; P = n.s.; ‘Population 2224’r18 = 0.373; P = n.s.; ‘Population 3270’r18 = 0.185; P = n.s.).

Figure 4.

Average infection stage of the four Podosphaera plantaginis populations on their sympatric host populations, sympatric metapopulations, and allopatric metapopulations.


These results demonstrate spatial variation in the pattern of local adaptation across the study landscape in the interaction between P. lanceolata and P. plantaginis, as could be expected from the dynamic and complex nature of the coevolutionary process. The geographical mosaic theory of coevolution proposes that spatial variation in natural selection and gene flow across the landscape may create among-population variation in the strength of local adaptation (Thompson, 1994, 1999). When pathogen performance was measured as infectivity, all four populations of P. plantaginis were locally adapted to their sympatric host population, and in three of four cases, pathogen infectivity was significantly higher on sympatric hosts than on hosts from the allopatric metapopulations. In two metapopulations, infectivity of the pathogen populations did not differ between the sympatric host population and host populations in the respective metapopulations, which suggests that these pathogen populations are adapted at the metapopulation level. Thrall et al. (2002) have also reported evidence for pathogen local adaptation at the metapopulation scale in the interaction between Linum marginale and Melampsora lini. In their pathosystem, the dispersal scale of the pathogen is assumed to exceed that of the host (Thrall et al., 2001).

There were altogether ten colonization events recorded in the four metapopulations in 2001–2004. This suggests substantial migration at the scale of 500 m, which corresponds to the scale at which pathogen populations are significantly aggregated in Åland. It has been proposed that pathogens are likely to adapt at greater spatial scales if the scale of pathogen dispersal exceeds the scale of host dispersal (Gandon et al., 1996; Thrall & Burdon, 1997; Burdon & Thrall, 2000; Gandon, 2002; Gandon & Michalakis, 2002; Thrall et al., 2002). This is likely to apply to the interaction between P. lanceolata and P. plantaginis. Throughout the growing season powdery mildews produce vast numbers of inoculum that are carried by wind to new host plants. Although the pollen of P. lanceolata is also wind-dispersed, pollen dispersal is restricted to a short time frame and pollen grains are fewer and larger than the dispersal spores of the pathogen. The mature seeds have no specialized dispersal structures and they travel very limited distances (van Damme, 1986).

Although two of the pathogen populations were locally adapted at the metapopulation level, the two other populations exhibited different patterns of local adaptation. The Eckerö pathogen population was strongly adapted to the local host population but even maladapted to the sympatric metapopulation in comparison with allopatric metapopulations. In Flaka, although there is some evidence of local adaptation, the differences in infectivity among the host populations were not statistically significant. Gandon (2002) has proposed that the evolutionary potential of parasites to adapt locally may be restricted by insufficient gene flow. The map on the regional occurrence of P. plantaginis in Åland (Fig. 1) suggests that differences in migration rate may explain the observed patterns in Eckerö in the western part of the Islands where infection incidence is rare. However, to validate this hypothesis replicated metapopulations at different degrees of connectivity should be tested for local adaptation.

Measuring pathogen performance as the stage of infection did not produce strong evidence for pathogen local adaptation. Infection stage unlike infectivity varied significantly among host populations within metapopulations and the average infection stage differed among the pathogen populations. This difference between the traits is likely to reflect the more complex mechanisms affecting the stage of infection. Although infectivity is mainly determined by a gene-for-gene interaction between respective resistance and virulence loci, infection stage may be determined by both partial resistance of the host as well as aggressiveness of the pathogen. Partial resistance is a polygenically controlled host trait that may affect the development of the pathogen once it has become infected. Partial resistance has pleiotropic effects that typically reduce the size of the feeding structures, haustoria, and limit the growth and sporulation of colonies (Bushnell, 2002).

No positive correlation between the stage of infection and infectivity could be found, in contrast to the results of Kaltz & Shykoff (2002). Hence, those pathogen populations that succeed in infecting more host genotypes do not develop faster than less infective pathogen populations, which may reduce some of the benefits of being locally adapted. A negative trade-off between virulence and spore production in the interaction between L. marginale and M. lini suggests that there is a cost to carrying extra virulence genes. Such a trade-off between infectivity and spore production will strongly influence the generation of local adaptation by impeding the emergence and evolution of highly virulent pathotypes capable of attacking all host genotypes (Thrall & Burdon, 2003).

In conclusion, the present results support the notion that the pathogen P. plantaginis often adapts to a host metapopulation in a fragmented landscape rather than to a local host population. I found evidence for geographical variation in the pattern of local adaptation of the pathogen, which may reflect spatially varying rates of gene flow in the pathosystem. These results are expected under complex spatially structured coevolutionary dynamics, and they point to the conclusion that understanding coevolution requires understanding of the spatial scales of parasite dispersal and the spatial dynamics of the host and the parasite.


Comments by Ilkka Hanski, Saskya van Nouhuys, Juha Merilä, Pia Mutikainen and one anonymous reviewer greatly helped improve an earlier draft of this paper. Otso Ovaskainen carried out the K-function analysis and Evgeniy Meyke produced the maps. Saila Kuokkanen, Riitta Ovaska and Mikko Putkonen assisted in the fieldwork. LUOVA Graduate School (Ministry of Education) and Academy of Finland (Grant no. 44887 to I. Hanski, Finnish Centre of Excellence Programme 2000–05) are acknowledged for financial support.