Spatial population structure in an obligate plant pathogen colonizing oak Quercus robur



    Corresponding author
      *Author to whom correspondence should be addressed. E-mail:
    Search for more papers by this author

    1. Metapopulation Research Group, Department of Biological and Environmental Sciences, PO Box 65 (Viikinkaari 1), FI-00014 University of Helsinki, Finland
    Search for more papers by this author

    1. Metapopulation Research Group, Department of Biological and Environmental Sciences, PO Box 65 (Viikinkaari 1), FI-00014 University of Helsinki, Finland
    Search for more papers by this author

*Author to whom correspondence should be addressed. E-mail:


  • 1The spatial population structure of parasites may have profound effects on both ecological and evolutionary dynamics. Strong gene flow among local populations has been hypothesized to prevent differential performance on local and foreign hosts.
  • 2We studied the spatial population structure of an obligate pathogen, the oak mildew (Microsphaera alphitoides) on the pedunculate oak (Quercus robur). Patterns of distribution and colonization–extinction dynamics were recorded in a set of 113 trees over 3 years, and in an additional set of 77 trees over 2 years. Colonization of uninfected hosts was examined in a set of 100 experimentally transplanted hosts, and differential performance on local and foreign hosts tested by reciprocal inoculations conducted in the laboratory.
  • 3Across years, the mildew occupied a high proportion of host trees ( ≥ 2/3). Patterns of mildew infection showed either no spatial aggregation (years 2003, 2005), or aggregation at a large spatial scale (up to 400 m; year 2004). High and even incidence of infection reflects efficient dispersal: during a single year, transplanted uninfected host plants were rapidly colonized in all parts of the landscape.
  • 4Local mildew populations still performed best on their original hosts: in reciprocal inoculations conducted on mature foliage, local mildew populations infected a significantly higher proportion of leaves from their original host tree than from other trees. Yet, even on the local host less than half of the leaves were infected.
  • 5To account for the observed patterns, two selection processes are invoked. The colonization experiment suggests that mildew spores are spread widely, with selection filtering out the genotypes that cannot infect the focal host. Local mildew populations are subject to further selection events later in the summer, favouring the strains best adapted to phenological changes in the local environment.
  • 6In conclusion, the patterns observed and processes inferred in M. alphitoides suggest that complex selection pressures may affect local parasite populations, blurring any clear-cut relation between gene flow and performance on local and foreign hosts.


The spatial structure of natural populations has important consequences both for population dynamics and for evolutionary change. Dynamically, the rate of migration among different sites will determine to what extent local populations may fluctuate as independent units. Evolutionarily, the level of gene flow will determine how local populations are bound together as cohesive evolutionary units, and at what scale evolutionary trajectories may diverge.

A satisfactory understanding of a spatially structured population can only be reached through quantification of key features such as local population sizes and turnover rates, as well as of the amount of migration between local populations (Harrison 1991; Harrison & Taylor 1997). So far, the well-studied systems are relatively few (for reviews see Hanski & Gilpin 1997; Hanski 1999; Ehrlich & Hanski 2004; Hanski & Gaggiotti 2004), and much more empirical work is needed before we can say something about the prevalence of particular spatial structures in nature, or their potential affinities to particular taxa (but see Hanski & Kuussaari 1995).

Among parasites, spatial population structures may have profound effects on the scale over which the organism may adapt to its host. Metapopulation dynamics of both hosts and pathogens may strongly affect the stability of local interactions, and hence the spatial scale and timespan over which the pathogen can adapt to its host (Comins, Hassel & May 1992; Gandon et al. 1996; Boots & Sasaki 1999; Roy & Kirchner 2000; Thrall, Burdon & Bever 2002; Laine 2005; Laine & Hanski 2006). In particular, local adaptation may be affected by the scale and strength of migration. If gene flow between a given site and others is extensive, foreign genes may swamp the local gene pool and keep local populations in a maladapted state (e.g. Camin & Ehrlich 1958; McNeilly 1968; Antonovics 1976; Stearns & Sage 1980; Dhondt et al. 1990; Holt & Gaines 1992; Holt & Gomulkiewicz 1997). Recent work reveals more versatile effects: If we assume that migration rates are not so high as to homogenize populations completely, parasites are predicted to be locally adapted if they migrate more than their hosts (Gandon et al. 1996; Gandon & Michalakis 2002). Furthermore, in the complete absence of migration, parasites tend to be locally maladapted (Morgan, Gandon & Buckling 2005).

Plants and their parasites provide important model systems for studies on parasite population dynamics and their patterns of local adaptation in a spatial context, since plants or plant stands occupy fixed positions in space, and many of their parasites can be mapped with relative ease (Antonovics et al. 1994; Burdon & Thompson 1995; Mopper & Simberloff 1995; Mopper & Strauss 1998). The few available studies of natural plant–pathogen interactions have highlighted the importance of dispersal for the regional persistence of plant pathogens (Smith, Ericson & Burdon 2003; Antonovics 2004; Laine & Hanski 2006). These studies have also revealed striking variation in infection prevalence among interactions, as well as within interactions through high turnover rates of local pathogen populations between years (Burdon, Ericson & Müller 1995; Ericson, Burdon & Müller 1999; Smith et al. 2003; Antonovics 2004). At one extreme is the highly fragmented meadow network of Plantago lanceolata L. in the Åland islands, where its powdery mildew (Podosphaera plantaginis Castagne; U. Braun & S. Takamatsu 2001) has colonized only c. 5% of the host populations (Laine & Hanski 2006). In other interactions infection prevalence ranges from 10% to 73% (Burdon et al. 1995; Ericson et al. 1999; Smith et al. 2003; Antonovics 2004). While for many interactions tests of local adaptation have yet to be carried out, patterns of local adaptation appear to be closely coupled with the scale of pathogen dispersal (Thrall et al. 2002; Laine 2005).

Here, we use the oak mildew (Microsphaera alphitoides Griff. & Maubl.) as a test case. This species is a widespread fungal pathogen of oak (Quercus robur L. and Q. petraea (Matt.) Liebl.) foliage throughout North America and Europe (Nef & Perrin 1999). The oak and the mildew closely interact with each other: the mildew can only survive on host tissue (Bushnell 2002), and may reduce the net photosynthesis of infected host leaves up to 50% (Hewitt & Ayres 1976; Brüggemann & Schnitzler 2001). The oak–mildew system exhibits a well-defined spatial structure, since the generation time of the host is very long (hundreds of years) compared to that of the pathogen. Compared to small, herbaceous host plants, the oak system exhibits a distinctly different structure. For mildew on herbs, local populations of the host offer patches of suitable habitat embedded in a matrix of other, unsuitable habitats (Laine & Hanski 2006). In comparison, for mildew on oaks, each large host tree could be envisaged as a local host population, with leaves offering ‘individuals’ of identical genotype (J. Burdon, pers. comm.) but potentially strikingly different phenotypes (Gripenberg & Roslin 2005; Roslin et al. 2006; Gripenberg, Salminen & Roslin 2007). From a temporal perspective, the long-lived host offers a relatively stable landscape in which pathogen generations come and go.

Earlier studies on local adaptation in parasites (Lively 1999; Gandon 2002) suggest that several biological features of the oak–mildew system may promote the evolution of local adaptation. First, the association of M. alphitoides with its host is obligate, as the mildew requires host tissue throughout its life cycle (Bushnell 2002). Second, the interaction is highly specific, as there are no other known hosts for M. alphitoides in the study region; third, the generation time of the mildew is short compared to that of its host; and fourth, the mildew population sizes are considerably larger than population sizes of oak (Hamilton, Axelrod & Tanese 1990; Ebert & Hamilton 1996; Kaltz & Shykoff 1998).

The south west of Finland offers an ideal setting for spatial studies on oak–mildew interactions, since oaks are relatively uncommon in the landscape, and occur both as dense stands and as highly isolated individuals. Hence, it is easy to find oaks growing under variable spatial settings, and to relate local host–pathogen interactions to the structure of the surrounding landscape (cf. Gripenberg & Roslin 2005); that is the key objective of this paper. Here, we describe metapopulation structure, dynamics and patterns of local adaptation in the oak mildew. Specifically, we first describe the spatial population structure of the pathogen by monitoring its distribution across a large number of hosts over three consecutive years, and by assessing the rate with which it colonizes new hosts experimentally transplanted to different parts of the landscape. We then use a laboratory experiment to probe for differential performance of the mildew on native and novel hosts, and to relate the spatial population structure of the pathogen to the patterns hence observed.

Materials and methods

biology of microsphaera alphitoides

In the life cycle of the mildew, cycles of clonal reproduction are followed by an off-season single cycle of sexual recombination. Mildew hyphae will typically hibernate in only small parts of the canopy (Woodward, Waldie & Steven 1929; Kerling 1966). Of the buds on shoots heavily infected by mildew in the previous year, only some 7·6% are infected with mildew (Woodward et al. 1929). As colonization from these sources is passive (based on air currents), a given mildew strain will likely end up on different branches in different years. During the growing season, the oak mildew passes through several clonal generations (c. six to eight generations based on laboratory measured sporulation times). The clonal dispersal spores are produced in chains on the surface of the leaves at high densities, hence yielding massive potential for the production of clonal dispersal propagules. As the spores are detached from their chains, they are carried by wind to new hosts. Following dry and warm summers sexually produced fruitbodies appear. From a quantitative perspective, these fruitbodies are considered relatively unimportant in the disease cycle of the oak mildew (Nef & Perrin 1999).

spatial population structure of microsphaera alphitoides

The material used in this study was collected on Wattkast, south west of the city of Turku, south western Finland (60°11′N, 21°37′E). On this 5 km2 island, all oak trees with a height exceeding 50 cm have been mapped with an accuracy of a few metres (Gripenberg & Roslin 2005), providing a well-defined spatial setting for studies on plant–parasite interactions (Fig. 1).

Figure 1.

 Distribution of oaks and mildew on the island of Wattkast. (a) Naturally established trees monitored for mildew infection. Here, circles correspond to 113 trees monitored over 3 years (2003–2005), stars correspond to 77 additional trees monitored over 2 years (2004–2005). The relative shading of the map symbol reveals during how many years the tree was occupied by the mildew, with a black circle indicating a tree infested by mildew in all 3 years, a dark grey circle indicates tree infected in 2 years, a light grey circle indicates tree infected only during a single year, and a white circle indicates tree never infected by the mildew. The letter-labelled arrows indicate the location of the four trees included in the laboratory experiment on local adaptation. (b) All 1868 oak natural oak trees on Wattkast (small black dots). Here, bigger circles show 100 oak trees planted in May 2005. During the summer, trees shown in grey were colonized by mildew whereas trees shown in white were not.

To describe the regional distribution and colonization–extinction dynamics of the oak mildew, we surveyed a set of 113 small-sized oaks over 3 years (2003–2005), and an additional set of 77 trees over 2 years (2004–2005). The selected oaks were small enough for the entire tree to be inspected (1–3 m in height) and spread out across all parts of the landscape occupied by oaks (Fig. 1a). Between 8 and 11 September 2003, 5 and 20 September 2004, and 31 August and 18 September 2005, we visited each of the selected oaks, examined the full foliage and hence confirmed the local presence or absence of M. alphitoides. On each tree, the general infection rate was also scored on a 4-grade scale, where 0, no mildew; 1, small amounts of mildew; 2, parts of the tree have substantial amounts of mildew, other parts uninfected; and 3, the whole tree is infected. Scorings were done by different persons in different years, and a tree's score in the previous year was unknown to the observer. Since the trees were scored in the autumn, when mildew infection has reached its maximum level, we expect few false absences in our data set.

colonization of fresh hosts and spatial aggregation in natural mildew infection

To examine direct dispersal and potential gene flow of M. alphitoides among oak trees, we transplanted a set of 100 uninfected oak trees to Wattkast during 10–13 May 2005 (Fig. 1b). The trees were c. 80–120 cm in height, and reared for 4 years from acorns collected in three natural oak populations in southern Finland (not including Wattkast). Each collection included the offspring of tens of trees. All transplanted trees were judged free of mildew in the previous autumn. To avoid any secondary mildew contamination, the trees were transplanted well before leaf flush, with only a minimal amount of soil on the roots and no dead leaf material. The planting sites were selected a priori to cover the full range of geographic locations and oak densities present within the island (Fig. 1b), but randomized with respect to habitat type. Only locations next to inhabited buildings or in the middle of cultivated fields were systematically rejected. Trees of different origin were spread evenly across sites. All trees were watered once at plantation, but then left under natural conditions until inspected for mildew presence on 20–26 September 2005. During the summer, regular visits to randomly chosen trees revealed that the phenology and development of planted trees closely matched that of naturally established individuals.

Patterns of colonization among the experimental trees were compared to patterns of spatial aggregation in the natural occurrence of mildew infection. Here, metapopulation theory predicts that the local incidence of a species will reflect patch-specific probabilities of colonization and extinction, and that the scale of dispersal may therefore have a key impact on the scale of spatial aggregation in abundance and patch occupancy patterns (Lande, Engen & Saether 1999; Ovaskainen & Cornell 2006). In this context, sampling at different scales may reveal different patterns of aggregation. Since the criteria used in selecting naturally established trees to be monitored for mildew infection (Fig. 1a) differed from those used in selecting sites for the transplanted trees (Fig. 1b), the two sampling designs were partly complementary.

To examine spatial aggregation in mildew infection, we used a K-function statistic (reduced second moment measure; Ripley 1976; Diggle 2003; Ovaskainen & Roslin unpublished). Specifically, we tested: (i) whether the spatial distribution of infected host individuals among all hosts individuals is more aggregated than expected by chance alone; and (ii) whether the spatial distribution of host individuals in a given infection category (1–3; cf. above) is more aggregated than expected by chance. Hence, we ask: for a randomly chosen oak in a given infection category, what is the probability that a randomly chosen neighbour at distance d falls in the same infection category? (To examine spatial aggregation in general infection, we simply pool infected individuals in all classes 1–3.) The observed probability of this event was calculated by splitting distance into classes of 50 m, and establishing the fraction of individuals representing the relevant infection category i within respective distance from each individual in i. The pattern observed was compared to that expected under the null hypothesis of random association between spatial location and infection status. To derive the expected pattern, we used the programme toast (Ovaskainen & Roslin 2007) to swap observed infection levels among host individuals 1000 times. When the empirically observed probability fell outside of the 2·5% or 97·5% quantiles of the resulting distribution, infection status was considered significantly negatively or positively aggregated in space, respectively.

mildew performance on local vs. foreign hosts

To experimentally assess the performance of local mildew populations on different oak individuals, oak trees were selected in different parts of Wattkast (Fig. 1a), 285–1840 m from each other. Since a pilot experiment led us to expect relatively low infection rates and substantial heterogeneity in leaf quality within trees (Roslin et al. 2006), we wanted to replicate inoculations of each combination of oak tree and mildew population as intensively as possible. Hence, only four oak trees could be included, resulting in a rather limited material on oak individuals per se. All the selected trees were relatively large, with a trunk diameter of 58–139 cm and a height of c. 9–12 m. To preserve part of the foliage in an uninfected state, 25 branches on each tree were enclosed in spore-proof bags (One Window Polyester Pollination Bags, 450 × 230 mm, PBS International, Scarborough, UK) between 14 and 15 May 2005. Since leaf quality may vary significantly among different parts of the tree crown (Gripenberg & Roslin 2005; Roslin et al. 2006; Gripenberg et al. 2007), the bags were deliberately spread out across all parts of the foliage reachable by climbing or by lassoing the branches down to ground level.

At the time of branch bagging, the buds were still fully closed on all trees in the area; leaf-flush began 1–2 weeks later. The bags were monitored repeatedly throughout the summer through the small plastic window on the bags, and if necessary opened to verify infection status. All bags containing a single infected leaf were instantly discarded from the experiment (n = 7 bags during full experiment).

On 21 July and 9–10 August 2005, after the leaves had reached physico-chemical maturity (Salminen et al. 2004), uninfected leaves were collected from 15 to 21 bags on each tree and stored in closed polyethylene bags with a piece of moist paper. At the same time, a sample of leaves heavily infected with mildew was collected from each tree. Since we had no prior knowledge of the genetic diversity of local mildew populations, we strived for comprehensive samples: some 100 leaves were collected from several discrete sampling points spread across the full tree crown. The leaves were stored in hard plastic vials with moist paper until used in experimental inoculations.

All leaf material was transported to the laboratory in a cooler, and stored in closed plastic vials with wet tissue paper. On 22 July 2005 (inoculation 1) and 11 August (inoculation 2), the uninfected leaves were moved to large Petri dishes (diameter 14 cm) with a moist filter paper on the bottom. Two leaves from different trees were placed on each dish, and before closing the lid, the leaves were inoculated by dusting them with mildew spores brushed off from infected leaves. To avoid cross-contamination of mildew populations, each Petri dish was inoculated with only one of the four mildew sources. During each of the two inoculation rounds, 10 leaves from each tree were inoculated with mildew from each of the three other trees, and 20 leaves from each tree were inoculated with mildew collected on the tree itself. Hence, the total number of experimental inoculations was 200 (4 × 3 × 10 + 4 × 20) per inoculation event, 400 in total.

Inoculated leaves were incubated at 20 °C under a photoperiod of 16 L : 8D, and the filter paper was wetted every second day. The leaves were carefully examined under a stereomicroscope 11 or 12 (inoculation 2 and 1, respectively) days after inoculation, and the presence or absence of infection scored on a binary scale (0/1).

The set of infected leaves collected from each tree was assumed to provide a representative sample of the mildew population present in the host individual. While using purified strains would allow for a more detailed assessment of genotype–genotype interactions (cf. Thrall et al. 2002; Laine 2005), the purification procedure (Nicot, Bardin & Dik 2002) may result in unintentional elimination of the less infective and/or sporulating strains. The current approach is also virtually identical to techniques used to demonstrate local adaptation in insects, where transplants among trees are usually based on tree-wide samples from local insect populations (Mopper & Strauss 1998). Importantly, the experiments were conducted at a phase of leaf development when local mildew strains had already gone though several phases of clonal reproduction. Given the mature stage of the foliage, colonization of novel foliage should then offer substantial resistance to mildew infection (Nef & Perrin 1999).

statistical models

Our observations of natural trees followed over time allowed us to assess whether the general intensity of mildew infection differed between years, and whether a tree heavily infected in 1 year remained so in the following. To test for general differences in the intensity of mildew infection among years, we conducted Kruskal–Wallis non-parametric AOV on ranks of four-level infection scorings. To assess whether the same tree maintained its level of infection across years, we calculated Spearman rank correlations of infection rates across years. Here, we focused on the subset of trees monitored over the longest period of time: 103 of the original set of 113 trees survived over the full 3-year period. (During the survey period, 10 trees were destroyed by mammalian browsing and human interference.) In both cases, we omitted trees unoccupied in any given year, to separate between regional extinction–colonization dynamics (infection rate ‘0’) and local interactions (infection rates ‘1’, ‘2’ and ‘3’).

The difference in performance of local mildew populations on local vs. foreign hosts was tested with data from the two reciprocal inoculation experiments using a generalized linear mixed model (Breslow & Clayton 1993). The probability with which a single leaf was infected by a mildew inoculum was modelled as a function of fixed effects Mildew origin (the identity of the tree on which the inoculum was collected), Leaf origin (the identity of the tree on which the leaf was collected) and Sympatry (i.e. whether the mildew inoculum originated from the same oak individual as the leaves). To test whether the infectiveness of a mildew population differed between its own and other host trees (a key test of local adaptation), Sympatry was nested under Mildew origin. By this nesting structure, we obtained clear-cut estimates of the infectiveness of each strain on the local vs. foreign hosts. To account for variation in leaf quality among different parts of the oak canopy, we modelled Bag (i.e. the identity of the exact bag nested under Leaf origin) as a random effect, and to account for slight variation in experimental conditions among the two inoculation events, we also included the identity of the Experiment as a random effect. The model was fitted by the Glimmix macro (Littell et al. 1996) as implemented in sas 8·0 (SAS Institute 1999). Since we were modelling a binary response variable (whether or not the leaf was infected) to estimate a proportion (the probability with which the leaf was infected), we assumed a logit link-function and binomially distributed errors (Littell et al. 1996). All mouldy leaves (n = 5) were excluded from the analysis.


spatial distribution and dynamics of mildew–host interactions

The oak mildew, M. alphitoides, is widely distributed within the island of Wattkast (Fig. 1a, Table 1). During each year covered by this study, over two-thirds of host trees examined were occupied by the species (Table 1). Of the set of 113 trees surveyed over 3 years, 56 trees remained occupied all 3 years, 34 were occupied in 2 years, 18 in a single year and only 5 never. Of the 77 trees followed over 2 years, 55 were occupied in both years, 16 in a single year and 6 never. Thus, most trees in the landscape appeared intrinsically suitable for mildew growth.

Table 1.   Metapopulation dynamics of Microsphaera alphitoides across two sets of oak trees: 113 trees monitored over 3 years (2003–2005) and 77 additional trees monitored over 2 years only (2004–2005)
MaterialtOldNewExtinctTotalTotal treesInfected (%)
  1. Column ‘Old’ refers to the number of trees infected by mildew in both years t and t – 1, ‘New’ to trees uninfected in t – 1 but infected in t, and ‘Extinct’ to trees infected in t – 1 but uninfected in t. ‘Total trees’ refers to the count of trees examined year t and ‘Infected’ to the ratio Total/Total trees. Trees disappearing from the original set of 113 trees were destroyed by mammalian browsing or human interference.

113 trees2003   8511375·2
20056727 39410391·3
77 trees2004   58 7775·3
20055513 368 7788·3

The pattern of mildew infection did not exhibit any consistent spatial aggregation. Only in 1 out of 3 years (2004) did we detect significant spatial aggregation, then at a scale up to 400 m (Fig. 2a; non-significant results for other years not shown). The level of infection seemed to vary randomly in space: neighbouring infected trees were not significantly more likely to share the same infection category than expected by chance (Fig. 2c,d). This general pattern was only broken by trees of category 1 in 2004, which showed some spatial aggregation up to 400 m (Fig. 2b).

Figure 2.

K-function analysis of the distribution of mildew infection categories in year 2004. Each examined oak tree (n = 188) was classified into one of four infection categories, where 0, no mildew; 1, small amounts of mildew; 2, parts of the tree with substantial amounts of mildew, other parts uninfected; 3, the whole tree infected. The solid line shows the probability that an oak at distance d from a randomly chosen oak individual shares the same infection category as the focal individual. The dotted line gives the average infection level and the dashed lines the one-sided 95% confidence intervals. Panel (a) shows the aggregation of infected trees (categories 1–3 pooled), whereas panels (b)–(d) shows the spatial aggregation of trees in category 1, 2 and 3, respectively.

While the total fraction of occupied hosts remained high over time, there was still substantial change in the exact set of trees occupied in different years. From one year to the next, 4%–25% of earlier mildew populations went extinct, and 46%–75% of previously uninfected trees were colonized. Local interactions between the host and the mildew are then relatively dynamic in terms of population turnover. Yet, trees that did remain occupied over time showed a remarkable consistency of infection levels: the general level of infection did not vary significantly between years (Kruskal–Wallis non-parametric AOV, F2,177 = 0·31, P = 0·74), and for individual trees, transitions between years within the same infection category were much more likely than transitions among categories (Table 2, Spearman rank correlations for infection levels 2003 vs. 2004 rs = 0·40, P = 0·002; 2004 vs. 2005 rs = 0·46, P = 0·0005; 2003 vs. 2005 rs = 0·51, P = 0·0001).

Table 2.   Transitions between infection levels 2004 and 2005
Infection level 2004→
Infection level 2005↓
  1. Cell entries show the number of trees for which the infection level of 2004 was that of the top row, and the infection level of 2005 that of the left-hand column (total n = 122 for both years). To separate between regional extinction–colonization dynamics and local interactions, trees unoccupied in one of the years (i.e. trees reaching infection level 0) were omitted.

38 8 3 19
211915 35
102345 68

colonization dynamics by the mildew

Experimentally introduced host individuals in all parts of the landscape were rapidly colonized by the mildew (Fig. 1b). During the single growth season of 2005, the mildew colonized 73 of 100 planted oak trees. By the end of the summer, the incidence of mildew on planted trees was still somewhat lower on planted trees (73 occupied: 27 unoccupied) than on 180 naturally established host individuals screened in that year (162 occupied: 18 unoccupied; Fisher's exact test P = 0·0003). Yet, on the trees actually colonized, the intensity of the infection did not differ significantly from that of naturally established trees (Kruskal–Wallis non-parametric AOV, F1,233 = 2·53, P = 0·11). If anything, the planted trees showed slightly higher infection levels (mean rank 127·4, n = 73) than naturally established trees (mean rank 113·8, n = 162). There were no signs of spatial aggregation in patterns of colonization: a tree close to a tree that got colonized was no more likely to be colonized itself than was a distant tree (observed probabilities well within 2·5%–97·5% quantiles of the distribution generated by random permutations).

performance of local mildew populations

Reciprocal inoculations conducted in the laboratory revealed that local mildew populations infected leaves from local and foreign hosts with different probability (Fig. 3, Table 3). The general probability of infection differed significantly between mildew populations from different source trees: Mildew from tree C had the lowest infectivity and mildew from tree A the overall highest infectivity (Table 3; main effect of Mildew origin). However, the specific probability of infection depended on whether the mildew originated from the tree itself, or from another tree (Table 3; main effect of Sympatry [Mildew origin]): the likelihood of successful infection was, on average, significantly higher on the original host tree than on other host individuals (Fig. 3). This result was evident in spite of the limited number of oak trees and mildew populations examined. Overall, infection success was relatively low on leaves from all origins. The mildew populations were usually able to infect less than half of the leaves from their source original tree: the total fraction of sympatric leaves infected was 37% (59 out of 158 inoculated leaves).

Figure 3.

Infection rates observed in the laboratory experiment. The height of each bar shows the total proportion of leaves from a given tree infected by respective mildew population; error bars shows the variance of these binomially distributed proportions. Black bars correspond to sympatric mildew–host combinations (a mildew population grown on leaves from its original host tree), white bars to allopatric mildew–host combinations (a mildew population grown on leaves from a foreign host individual).

Table 3.   Generalized linear mixed model of the probability of infection in reciprocal inoculations
SourceVarianceF-rationdf, ddfP
  1. For fixed effects, we provide type 3 F-statistics with associated degrees of freedoms and P-values, for random effects, we show variance estimates and associated standard errors.

Mildew origin 2·863, 3140·04
Leaf origin 0·543, 690·66
Sympatry (Mildew origin) 2·424, 3140·05
Experiment2·40 ± 3·46   
Bag (Leaf origin)0·54 ± 0·28   
Residual0·86 ± 0·07   


This study reveals the spatial population structure of M. alphitoides on Wattkast as a tightly connected metapopulation. New hosts are quickly colonized almost anywhere on the island, and a large proportion of old hosts is infected each year despite frequent extinction of local populations. Still, in the face of what appears to be extensive gene flow among local mildew populations, we found that such populations were more efficient in infecting leaves from their own host tree than from other trees – even though overall, a relatively low proportion of mildew strains present on a tree managed to re-infect mature foliage.

mildew and hosts in space

Within the study area of 5 km2, local mildew populations occupy oaks in most parts of the landscape, and spatial aggregation is weak or lacking. A high and even incidence may reflect either low extinction rates or high colonization rates (Hanski 1999): our direct observations of extinction–colonization dynamics and our experiment with planted oaks both support the latter interpretation. With a total count of 1868 oaks in the island, distances between neighbouring hosts for the mildew will rarely exceed some hundred metres; a scale over which the dispersal of M. alphitoides seems highly efficient. Our experiment with the planted oaks shows that the mildew is able to reach nearly any part of the landscape in a single growing season – perhaps with some reservations. After the summer, the incidence of the mildew was still lower on planted trees than on naturally established oaks, an observation which we may attribute either to slight dispersal limitation at the scale of the study area, or to the fact that none of the planted trees were infected from before (i.e. that they sustained no pre-existing mildew populations surviving across years). In themselves, the planted trees seem just as suitable for the mildew as any wild trees in the area, as revealed by the lack of a difference in infection levels between those planted trees which were actually colonized, and that of naturally established trees.

While patterns of colonization should not uncritically be interpreted as patterns of gene flow (Slatkin 1994; Whitlock & McCauley 1999; Saccheri & Brakefield 2002), the scale and extent of colonization events observed by us do suggest that in the landscapes of Wattkast, mildew spores will travel efficiently from one oak to the next. Naturally, the introduced trees may differ from resident ones in their sensitivity to local pathogens – or they may be suffering from transplant-induced stress. Both factors may affect their susceptibility to mildew infection. Nevertheless, neither consideration will affect our main conclusion that mildew spores travel frequently over the distances between oaks of Wattkast. The same notion is also supported by the lack of spatial aggregation in the incidence of mildew infection on naturally established trees. While the spatial configuration of these trees differed from that of the planted trees, spatial aggregation of mildew infection was only evident in 1 out of 3 years, and then over large distances.

Once a mildew population is established on a host, the length of association is uncertain. On the small trees observed by ourselves, half of the local mildew populations (56/113) which were present by the start of the observation period persisted for the full length of the study, over 3 years. We still lack direct data on population persistence in larger trees, but expect it to be higher – potentially much higher – due to the larger size of the resource and the mildew population that it can sustain. Hence, we expect local pathogen–host interactions to persist over several years at a time, but local adaptation to be continuously challenged by the inflow of genes from nearby mildew populations.

The spatial population structure observed in M. alphitoides then contrasts substantially with that described in another well-studied mildew –Podosphaera plantaginis on the ribworth plantain (Plantago lanceolata, cf. Laine & Hanski 2006). Both the incidence of the pathogen and the stability of the local interaction differ widely among the systems (cf. Laine & Hanski 2006): in the Pl. lanceolata–Po. plantaginis interaction infection prevalence has remained below 5% over several years, while in the oak mildew over 90% of hosts may be infected in any given year. For Po. plantaginis, the small, herbaceous host also has a short life cycle and strong fluctuations in density among years (Laine 2004), thereby offering a much less stable resource than the long-lived oaks. Overall, the infection prevalence observed in M. alphitoides falls at the high end of values reported from other plant–pathogen interactions (10%–73%; Burdon et al. 1995; Ericson et al. 1999; Smith et al. 2003; Antonovics 2004).

local adaptation in microsphaera alphitoides

Despite substantial gene flow observed in the oak mildew, there were indications of ‘local adaptation’ (sensu differential performance on local and foreign hosts). Even in our small material, local mildew populations were able to infect a higher proportion of leaves from their original host tree than leaves from any of the other trees. The failure of mildew strains to infect 100% of leaves from their tree of origin is understandable, as young leaves are more susceptible to mildew infection than leaves hardened during the growing season (Nef & Perrin 1999). Hence, these mildew strains have successfully colonized their tree earlier in the growing season, and in late July or early August when the experiments were carried out, they were able to infect only 37% of their local hosts’ leaves. However, the ability to infect a higher proportion of leaves of their local vs. foreign host trees suggests that these strains have, to some extent, adapted to their local trees and the changes the leaves pass through during the growing season. The lower infection success of the mildew strains originating from other parts if the island may be attributed to both maladaptation to the seasonal changes in leaf chemistry and structure, and to any attributes of the host – be they genetic or phenotypic – that will prevent establishment throughout the growing season.

Overall, the longevity of established trees and the dispersal propensity of the mildew suggest the following scenario: the spores of individual mildew strains spread relatively widely, but only genotypes able to colonize a given host will establish. Hence, the spore rain is subject to selection filtering out genotypes that cannot infect the host. While this may appear as ‘local adaptation’ in reciprocal transplant experiments, it will hardly fulfil a strict definition of the concept.

Interestingly, our results are akin to those observed in leaf-mining moths (Mopper et al. 1995, 2000), where indirect data suggest efficient dispersal between trees, but reciprocal transplants still reveal superior performance on the native tree. Even though in principle, a local population may then extend well beyond an individual tree, local selection pressures will confine it to this scale – or to sets of similar trees. High levels of gene flow over neighbouring trees may consequently result in hierarchic genetic structuring of the oak mildew, and in adaptation to the overall frequency of particular tree genotypes in the metapopulation rather than to individual trees (cf. Dybdahl & Lively 1996).

While our results suggest strong and differing selection pressures acting in different oak trees, it is clearly based on a small material. The current experimental design will also preclude firm conclusions with respect to the underlying agents. Oaks as individuals do show differential susceptibility to M. alphitoides are clearly revealed by the consistency of infection scores among years: a tree particularly infected by mildew in one year was also highly infected in another (see also Ayres 1976). The factor responsible will have to be a general attribute of the tree. The laboratory experiment included foliage from all parts of the canopy, and still the signal was clear. This argues against mildew susceptibility being a mere function of the physicochemical characteristics of the foliage, such as phenolic profile, water content and leaf size, since all of these measures vary drastically within a single oak crown (Gripenberg & Roslin 2005; Roslin et al. 2006; Gripenberg et al. 2007).

What we are looking for is rather an attribute of the full tree crown, consistent from year to year. It might be based on fine-tuned recognition systems, but detailed investigations into the resistance response of several oak species to M. alphitoides reveal a rather limited response of Q. robur in terms of cell necrosis and structural cytological changes (Edwards & Ayres 1981). This, together with the quantitative, not qualitative susceptibility of oaks observed in our reciprocal transplants, seems to argue against the existence of any fine-tuned resistance response. Unfortunately, our experiment does not suffice to fully separate between genetic and environmental impacts on a tree's susceptibility, and calls for continued work.


Our study reveals multiple selection processes in the oak mildew. While the species seems an efficient colonizer, reciprocal inoculations suggest that local mildew populations may perform best on mature foliage of their original hosts. Based on our description of the spatial population structure of M. alphitoides, two selection processes appear likely: the colonization experiment suggests that mildew spores spread efficiently among trees, with selection filtering out the genotypes that cannot infect the focal host. Results from the reciprocal inoculations reveal that local mildew populations are subject to further selection events later in the summer, favouring strains best adapted to phenological changes in the local environment. In conclusion, the patterns observed and processes inferred in M. alphitoides indicate that complex selection pressures may affect local parasite populations, blurring any clear-cut relation between gene flow and performance on local and foreign hosts. Local evolutionary processes will then be a result both of the surrounding landscape and of species-specific life-history, emphasizing the role of species-specific ecology (Roslin 2000, 2001; Gutierrez et al. 2001; Roslin & Koivunen 2001; Biedermann 2004; McGeoch & Price 2004).


Pertti Pulkkinen at the Finnish Forest Research Institute provided the oak saplings used to record mildew spore dispersal. Otso Ovaskainen helped with the K-function analyses. Ainhoa Apesteguia, Katja Bonnevier, Riikka Kaartinen, Maiju Lanki, Elly Morriën, Alberto Oviedo, Tommi Salmela, Markku Salo, Ayco Tack and Riina Uvanto assisted with crucial tasks in the field and in the laboratory. We thank them all for their kind contributions. This project was funded by the Academy of Finland through Academy Researcher Fellowship no. 00677 and accompanying grant number 111704, and the Finnish Centre of Excellence Programmes 2000–2005 (grant number 44887) and 2006–2011 (grant number 213457). Financial contribution by the Ella and Georg Ehrnrooth Foundation is also gratefully acknowledged.