SEARCH

SEARCH BY CITATION

Keywords:

  • asynchrony;
  • epidemiology;
  • fungal pathogen;
  • host–pathogen interactions;
  • metapopulation

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

1 Analysis of the dynamics of rust disease caused by Uromyces valerianae in 30 discrete populations of Valeriana salina occurring on an archipelago of small islands in the Gulf of Bothnia, central Sweden, showed strong temporal and spatial effects.

2 Over a 13-year period, the population dynamics of the pathogen varied across the metapopulation, with disease incidence (presence/absence), prevalence and severity all showing strong population and year effects, indicative of heterogeneity among years and host populations in the suitability of conditions for the pathogen. While some individual populations were infected for the entire study period, others were infected for only 1 or 2 years. Local pathogen population extinction and recolonization events were relatively common, with annual recolonization rates of previously healthy populations ranging from 0% to 13.3%.

3 The incidence of disease within individual host populations was significantly affected by host population size, the prevalence of disease in the previous year and the proximity of neighbouring populations that were infected in the current year. The prevalence of disease in infected populations only depended on the prevalence of disease in the previous year. There was little to choose between disease prevalence and severity in predictive power.

4 Overall, the dynamical behaviour of this set of pathogen demes best fit that predicted for a metapopulation with considerable asynchrony in epidemiological patterns between different demes, despite evidence of among-population migration.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Our understanding of the evolutionary dynamics of plant–pathogen associations has advanced rapidly with the realization that the spatial context in which such interactions occur is of paramount importance in determining the evolutionary trajectory of co-evolving associations. Individual host and pathogen populations are no longer regarded as closed systems endlessly responding to each other. Rather, each such pair is seen as a small component in a network of populations between which gene flow, extinction and colonization events occur with varying frequency and intensity ( Burdon 1997). The importance of the ecological setting has been further stressed in recent heuristic arguments that detail a variety of ways in which life-history attributes of hosts and pathogens, interacting over space and time, may shape the spatial scale of interaction metapopulations ( Thrall & Burdon 1997) and have significant implications for the relative effectiveness of different disease-resistance mechanisms ( Burdon et al. 1996 ). A similar picture is emerging from an increasing number of computer simulation models that have all emphasized the importance of including spatial and temporal variability in developing an understanding of host–pathogen co-evolutionary dynamics ( Frank 1991; Thrall & Antonovics 1995; Gandon et al. 1996 ).

Despite this upsurge in interest in the theoretical consequences of spatial and temporal variability for the co-evolutionary dynamics of host–pathogen interactions, to date little empirical information has been gathered that assesses the behaviour of real world systems. However, those studies that do exist indicate considerable temporal and spatial stochasticity, with marked differences in disease incidence and severity occurring between adjacent host populations within a single season, as well as within populations between seasons (Filipendula ulmariaTriphragmium ulmariae: Burdon et al. 1995 ; Silene spp.–Ustilago violacea: Carlsson & Elmqvist 1992; Antonovics et al. 1994 ).

Here we report on the long-term epidemiology of the rust pathogen Uromyces valerianae in an interaction metapopulation involving 30 populations of its host, Valeriana salina. This pathogen shows an annual demographic cycle characterized by a very sharp drop in numbers during autumn and winter, when survival is restricted to off-season resting spores (teliospores). By monitoring changes in the incidence (presence/absence), prevalence and severity of disease caused by this pathogen, and fluctuations in host population sizes for each of 13 consecutive years, we were able to determine the effect of the size of host populations, the proximity of neighbouring host populations, and the incidence and prevalence of disease in previous years on the dynamics of disease in the metapopulation.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

The host, valeriana salina pleijel (valerianaceae)

Valeriana salina is a tall perennial herb considered to be endemic to Fennoscandia. It is an octoploid taxon, the taxonomy of which is not fully understood, but it is close to V. sambucifolia Mikan fil. ( Tutin et al. 1976 ). In the study area on the Gulf of Bothnia, V. salina is confined exclusively to sandy/gravely sea-shores. New growth from existing over-wintering rootstocks starts in late May, with flowering in mid-July. In mid-autumn (October) all above-ground parts of these plants die, leaving only the rootstock. Seedling recruitment into the population occurs in the first half of June. Population sizes vary greatly from year-to-year, being influenced by a range of factors including summer droughts, autumn and winter storms, and seed production in the previous year.

The rust, uromyces valerianae (dc.) lév.

Uromyces valerianae is an autoecious, macrocyclic rust pathogen that has only been recorded on species of Valeriana ( Wilson & Henderson 1966; Gjaerum 1974). During the summer growing period the pathogen occurs as small, dark brown uredial infections on the leaves and fleshy stems of its host, V. salina. This stage of the pathogen’s life cycle is asexual and, under favourable environmental conditions, many generations may follow one another in quick succession, leading to local epidemics. The above-ground parts of host-plants die back during the autumn and local over-winter survival of the pathogen is entirely dependent on teliospores. Given suitable conditions, these germinate in the following spring to initiate a process of sexual recombination, followed by the development of aecial infections. Aeciospores subsequently initiate the uredial stage of the life cycle.

The study sites

The study was focused on all known populations of V. salina on 30 of the 31 islands (island 31 was never colonized by V. salina) occurring in an approximately 16-km2 area of the Harkskärsfjärden archipelago 20 km north of Gävle, in central Sweden (60°46–48′N, 17°20–24′E; Fig. 1). The islands differed in their suitability as habitats for V. salina, with population sizes ranging over the study period from 0 to 8800 (average = 1340). The relative position of each population was determined on large-scale maps of the region (Topografiska Kartan, scale 1 : 50 000) using position 6739/1588 on map 13H Gävle as the N/W 0/0 position. These coordinates were used in subsequent spatial data analysis.

image

Figure 1. Spatial distribution of Valeriana salina populations in the Harkskärsfjärden archipelago near Gävle, central Sweden. The asterisk on the inset map shows the position of the archipelago in Fennoscandia.

Download figure to PowerPoint

Experimental assessments

Disease incidence, prevalence and severity

In 1984, the islands of the archipelago were surveyed for the presence of populations of V. salina. At that time distinct, physically separated populations were identified on 27 islands to which were added, in subsequent years, a further three very small populations detected in annual surveys of previously unoccupied islands. Over the period 1984–96 inclusive, the size of each V. salina population and the incidence, prevalence and severity of disease caused by U. valerianae, were determined in a yearly early August census. At each site, plant population size was estimated either by counting all plants in small populations or, for larger populations, by counting all plants occurring in a series of 20 randomly distributed 1-m2 quadrats and correcting this number to allow for the total area covered by the population. (Because populations varied greatly in shape, areas occupied were estimated by subdivision into manageable pieces that were then measured by tape or careful pacing). At the same time, disease incidence (presence/absence) and, where present, the prevalence (percentage of plants infected) and severity (percentage leaf area covered by lesions) of disease caused by U. valerianae on individual plants were recorded. The severity of disease in the population as a whole was determined as a mean of the percentages of leaf area infected for each individual. In populations of less than 100 plants, all individuals were examined and the areas occupied by uredia estimated visually using a modified James scale ( James 1971). In larger populations a complete assessment was not possible and a random sample of 100 plants was examined. To ensure that all parts of the population were covered, plants were randomly chosen along a series of rough transects. When no disease symptoms were recorded on the 100 sample plants, the population was further searched to determine if the pathogen was present at all.

Data analysis

The relationships between disease incidence, disease prevalence, disease severity and host population size, and log-transformed versions of the last three variables, were initially examined graphically using all year × population combinations (n = 351). Correlations between severity and prevalence were also calculated.

The relationships of each of disease incidence, disease prevalence and disease severity to five predictor variables, population size in the current year, population size in the previous year, prevalence of disease in the previous year, severity of disease in the previous year and incidence of disease in the previous year (binary variable, presence/absence), were investigated using stepwise regression. In this analysis, the predictor variable that most improves the relationship is added at each step, until further additions do not significantly improve the relationship. Log transformations were used on disease prevalence, disease severity and host population size to overcome heteroscedasticity in the data. Disease incidence was analysed using generalized linear models with binomial errors and logit link ( Dobson 1990). In such analyses, an analysis of deviance is produced that is analogous to an analysis of variance for normally distributed data. Individual deviances, analogous to sums of squares, are distributed as chi-squares. Disease prevalence and severity were analysed using stepwise regression with normal errors, and were performed only on data from populations where disease was present. This resulted in a two-stage inference procedure, (i) determining factors that affected presence of disease, and (ii) determining factors that influenced the amount of disease, given that disease was present.

Finally, variables that represented spatial spread of disease were introduced into the models. For each population in each year, indices of the total influence of disease from other populations in the current year were calculated by summing over all other populations an indicator calculated as the level of disease multiplied by the population size and divided by the distance to the population. Four such indices were calculated, where disease prevalence, disease severity, log disease prevalence and log disease severity were each used as the measure of ‘level of disease’. A further four indices were also calculated, one for each measure of ‘level of disease’, by dividing by the squares of the distances rather than the distances.

Stepwise regression was used to add these indices to the best non-spatial models for incidence, prevalence and severity to determine whether significant improvement occurred, using the same transformations and error structures as in the non-spatial analyses.

For both the analyses of deviance and variance, the tests of significance of the terms included in the models depend on the assumption that the residuals are independent and identically distributed. To test these assumptions we examined the Pearson residuals from each of the models for incidence and prevalence that included all significant terms. Sample variograms were calculated individually for incidence and prevalence to investigate both possible temporal and spatial dependence, in which half the squared differences between residuals for all pairs of observations were graphed against the temporal difference (years) and spatial difference (distance between populations; km) ( Diggle et al. 1994 ). For the spatial variograms a smoothing spline on six degrees of freedom was fitted to illustrate the underlying trend. Due to the large number of points in the spatial variogram for incidence, separate variograms were calculated for each year.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

The incidence of disease and its prevalence and severity in different populations were highly variable from year-to-year and population-to-population (the prevalence of infected individuals in four closely neighbouring populations is shown in Fig. 2). Over the entire 13-year period of the study, the number of populations in which disease was present in any given year ranged between 13 and 22. Between years, disease status showed considerable fluctuations, with many previously disease-free populations being colonized by the pathogen, while in others pathogen extinction occurred ( Table 1). To some extent these patterns were also reflected in the prevalence of disease occurrence summed across the metapopulation as a whole, where the percentage of infected individuals ranged from a high of 33.5% in 1988 to lows of 4.9% and 5.4% in 1987 and 1994, respectively ( Fig. 3).

image

Figure 2. Year-to-year fluctuations in the prevalence of individuals infected with Uromyces valerianae in four representative populations occurring on four islands (12, 13,14 and 26) within 150 m of each other.

Download figure to PowerPoint

Table 1.  Changes in the overall percentage of populations of Valeriana salina infected with Uromyces valerianae and in the frequency of local pathogen extinction and recolonization events (expressed as a percentage of all populations)
 Year
1984198519861987198819891990199119921993199419951996
% of populations infected57636068736868575757475043
% extinctions07731313131032077
% recolonization731310713013310103
image

Figure 3. Year-to-year fluctuations in the prevalence of individuals infected with Uromyces valerianae summed across the metapopulation as a whole.

Download figure to PowerPoint

These year-to-year fluctuations undoubtedly reflect variations in the suitability of the gross physical environment for disease development. In addition, however, some populations consistently sustained higher or lower prevalences of infected individuals than others. Thus, several very small populations (5, 9, 21) were free of disease for the entire study period while other populations showed infection in most years. Moreover, although the large populations 24 and 7 were similar in size, their mean prevalence of infected individuals was markedly different [7.5% (range: trace–40%) and 42.2% (range: 1–100%), respectively ( Fig. 4)].

image

Figure 4. Variation between different Valeriana salina populations in the 13-year mean prevalence of individuals infected with Uromyces valerianae.

Download figure to PowerPoint

The length of time for which disease is present in a population is another measure of the stochasticity of the interaction. Moreover, in host–pathogen systems like that between V. salina and U. valerianae, where epidemic development is reliant on the pathogen being present early in the season, this measure of association time may be an important reflection of the pathogen’s potential to act as a selective agent. Such occupancy time can be measured in two ways. The first measure, at the level of the individual population, is the number of years disease is present. Out of 13 populations, this ranged from continuous in almost a quarter of all populations, to total absence in a few populations, and presence in only 1 or 2 years ( Fig. 5a). Secondly, the rate of conversion of populations from diseased to healthy and back again is an equally important measure of the way in which the host and pathogen interact. The high rate of local pathogen extinction is well illustrated by the dominance of short time periods (1–3 years) during which the pathogen was either present ( Fig. 5b) or absent.

image

Figure 5. (a) Differences between Valeriana salina populations in the total number of years (out of 13) in which the pathogen Uromyces valerianae was present. (b) Variation in the relative frequency of different periods of time for which the pathogen Uromyces valerianae was present in Valeriana salina populations.

Download figure to PowerPoint

The relationship between disease prevalence (percentage plants infected) and host population size based on all populations × years combinations pooled is presented in Fig. 6. Most small populations were free of disease ( Fig. 6), but when disease-free populations were omitted there was no apparent relationship between prevalence and population size (r = 0.11; n = 148). Similar patterns were obtained for the relationship between disease severity and population size. There were very high correlations between the two disease parameters both when all populations were included and when disease-free populations were excluded (for log severity–log prevalence, r = 0.92 in both cases). Consequently only one of these parameters would be expected to appear in the best models obtained by stepwise regression.

image

Figure 6. The relationship between prevalence of Uromyces valerianae and size of Valeriana salina populations for all year × population combinations (n = 351).

Download figure to PowerPoint

When applied without spatial variables, the generalized linear model indicated that log population size in the current year and log disease prevalence in the previous year were the only two significant variables in predicting disease incidence. Both of these were highly significant (P < 0.001; Table 2). The eight variables representing spatial spread of disease were very highly correlated, with correlation coefficients ranging from 0.83 to 0.98. When added to the model, the effect of each in explaining further variation in incidence was therefore similar, and only one variable significantly increased the amount of variation explained by the model when the stepwise procedure was used. The measure of disease severity in surrounding populations, based on summing untransformed severity × population size divided by distance, was the best of these variables (deviance = 13.26; P < 0.001). Thus the final model for incidence ( Table 2) involves three predictive terms, each with positive coefficients. The variogram for temporal dependence constructed from the Pearson residuals for this final model is given in Fig. 7a, and the variogram for spatial dependence in 1989 is given in Fig. 7b. In both these cases, and in the other 11 spatial variograms, there was no consistent directional trend that would have been indicative of any remaining correlation structure in the residuals. The probabilities of disease resulting from the final model are shown in Fig. 8 for the full range of population sizes and for low (0), medium (10) and high (100) values of prevalence of disease in the previous year. In all three cases, the disease severity index for other populations is set at its median value. This demonstrates the wide range of possible probabilities of disease as prevalence and population size vary.

Table 2.  Model for incidence of disease caused by Uromyces valerianae in populations of Valeriana salina
Term
(a) Analysis of deviance
 d.f.Change in devianceMean deviance
Log population size131.6631.66
Log prevalence in previous year138.5738.57
Disease severity index for other populations113.2613.26
Residual320363.251.14
All changes in deviance are significant at P < 0.001
(b) Estimates of regression coefficients
  EstimateStandard error
Constant –4.350.90
Log population size 0.4800.123
Log prevalence in previous year 0.5590.099
Disease severity index for other populations 0.022580.00781
image

Figure 7. Variograms of Pearson residuals from the incidence model for (a) temporal and (b) spatial dependence (1989 only), and from the prevalence model for (c) temporal and (d) spatial dependence. The line in (b) and (d) is the fitted spline on six degrees of freedom.

Download figure to PowerPoint

image

Figure 8. The relationship between the probability of occurrence of disease caused by Uromyces valerianae and the size of host populations. Relationship based on the accumulated data from all populations across all years generating a mean value for spatial influence of disease; the three curves show the relationship assuming the prevalence of disease in the previous year was 100% (solid line), 10% (dotted line) and 0% (dashed line).

Download figure to PowerPoint

Due to high correlations between the prevalence and severity variables, the results of model fitting for these two variables were very similar, and only the results for prevalence are presented. The stepwise regression for log prevalence, performed on diseased populations only, picked out only one significant variable, prevalence in the previous year (P < 0.001), again with a positive coefficient ( Table 3); there was no evidence of influence of spatial spread of disease in the current year. The variograms constructed from the Pearson residuals for this final model are given in Fig. 7c,d. As for disease incidence, there was no consistent directional trend in the prevalence residuals that would have been indicative of any remaining correlation structure. The estimated regression coefficient (0.217) together with the positive intercept (1.993) ( Table 3) lead to the equation:

Table 3.  Model for log prevalence of disease caused by Uromyces valerianae in populations of Valeriana salina with disease present
Term
(a) Analysis of variance
 d.f.SSMSF
Log prevalence in previous year118.3118.3111.45
Residual146233.521.60 
F ratio is significant at P < 0.001; percentage variation explained = 6.6
(b) Estimates of regression coefficients
  EstimateStandard error 
Constant 1.9930.154 
Log prevalence in previous year 0.2170.064 

log prevN+ 1 = 1.993 + 0.217 log prevN

suggesting that low prevalence will give rise to an increase in the following year, whereas moderate to high prevalence will predict a decrease. For example, a prevalence of 10 gives a predicted prevalence of 12.1, whereas 25 gives 14.8, and 100 gives 19.9. This reflects the transient nature of severe levels of disease, which are seldom present in the same population for consecutive years.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Studies of individual populations may provide a detailed picture of the fitness consequences of given levels of pathogen damage ( Augspurger & Kelly 1984; Paul & Ayres 1986; Jarosz & Burdon 1992). However, from an evolutionary point of view, restriction of studies to such narrow spatial scales often carries implicit assumptions of permanence of associations, and a lack of extinction and/or migration. In reality, understanding the long-term dynamics of associations is likely to be greatly affected by the occurrence of multiple populations, by a range of life-history attributes of both host and pathogen, and by the interplay of these in a spatial setting ( Thrall & Burdon 1997, in press).

The current association between Uromyces and Valeriana is one involving a mixture of life-history attributes that are likely to result in considerable fluctuations in the size of individual pathogen populations. The pathogen’s multiple within-season cycles of urediospore production may lead to substantial within-season disease epidemics, thereby enhancing the possibility of migration to currently disease-free host populations. However, in the study area U. valerianae is restricted to V. salina and at the end of the growing season the uredial population crashes to virtually zero as plants die back to underground tubers. On-site over-winter survival for the pathogen is strongly affected by its ability to produce sufficient telia, as re-establishment of the pathogen the following spring occurs via germination of such survivors or through migration from other populations. While local pathogen population extinction events were fairly frequent over the 13 years of this study ( Table 1), and within the range found for other fungi ( Burdon et al. 1995 ), the contribution of disease prevalence in one year to incidence and prevalence in the next ( Tables 2 and 3) clearly indicates significant levels of local survival. Equally, the significant relationship between disease incidence and the disease severity distance index reflecting distance ( Table 2) confirms that local migration events are an important part of the dynamics of the association.

Asynchrony of pathogen population dynamics

Asynchrony of development in different populations is one feature that has been predicted to occur between different demes within a metapopulation ( Pacala et al. 1990 ; Gilpin & Hanski 1991; Hanski & Gilpin 1997). In the interaction between U. valerianae and V. salina, such asynchrony was apparent in the pathogen populations in two distinct ways. First, as shown in Fig. 2, disease epidemics in neighbouring populations were often at quite different stages. Secondly, a further manifestation of asynchrony was apparent in the fluctuations that occurred in the length of time over which pathogen and host were associated within individual populations ( Fig. 5a). For a few populations the disease was present continuously; however, for the majority, disease was present for 1–12 of the 13 years of the study. There was considerable variability in the length of the continuous association of pathogen and host ( Fig. 5b): a measure of the rate of conversion of populations between diseased and healthy states. In many cases, these periods ran for 1 or 2 years only before extinction occurred. However, in some cases the pathogen was continuously present for 5 or more years.

What are the consequences of such asynchrony of pathogen development across the metapopulation? At this stage we cannot provide a definitive answer to this question. However, it is not unreasonable to suggest that such asynchrony may have implications for the generation, maintenance and spatial structuring of genetic variation within pathogen metapopulations ( McCauley 1991). Pathogen populations at sites with high turnover rates are likely to show lower within-population variation (due to founder effects) than populations that persist for longer periods. Differences in the presence and temporal persistence of U. valerianae between demes of the metapopulation may also lead to differences in the intensity of selective pressures experienced by local populations of V. salina. In turn, this may have an effect on local pathogen population structure.

Host population size affecting pathogen persistence

As has been shown in a number of studies of other host–pathogen associations involving both systemic flower infecting smuts (Viscaria vulgaris–Ustilago violacea: Jennersten et al. 1983 ; Silene dioica–Ustilago violacea: Carlsson et al. 1990 ; S. alba–U. violacea: Antonovics et al. 1994 ) and annual rusts (Filipendula ulmariaTriphragmium ulmariae: Burdon et al. 1995 ), the size of host populations has a significant effect on disease. While the importance of the threshold size for pathogen persistence was less marked for U. valerianae than in the other interaction involving a rust pathogen (the Triphragmium study), the current data still clearly indicate the importance of larger populations in providing greater continuity of persistence for the pathogen. The importance of larger populations in providing this increased security of persistence has also been highlighted by recent simulation models ( Thrall & Burdon, in press).

Once disease was present, the size of the V. salina population had no significant effect on the prevalence or severity of disease. Given that population size is not necessarily a good predictor of population density, a lack of a positive relationship was not surprising in this wind-dispersed pathogen system. This was similar to the results from the interaction between T. ulmariae and its host where disease severity was only marginally positively affected by population size ( Burdon et al. 1995 ). In contrast, in the interaction between S. alba and U. violacea, small host populations showed consistently higher frequencies of disease than larger ones ( Antonovics et al. 1994 ). In that system, pathogen dispersal was almost exclusively achieved by insect vectors, whose foraging activity is frequency- rather than density-dependent. These differing outcomes highlight the way in which differences in life-history attributes may be reflected in patterns in the field.

Spatial associations in disease distribution

Typically, disease is unevenly distributed both within and among populations. However, in a metapopulation situation where migration is occurring between demes, we would expect to find evidence of spatial effects such that disease incidence in neighbouring populations may show greater association with each other than with more distantly placed ones. Such an effect was found in 2 of the 4 years of a study of the epidemiology of Triphragmium ulmariae in the Skeppsvik archipelago, about 500 km north of the present study area ( Burdon et al. 1995 ). In the current study, the probability of the incidence of rust caused by U. valerianae in a given population showed a weak but significant linear relationship to disease levels in surrounding populations divided by their separation distance. This implies a somewhat flatter dispersal gradient than for many air-borne pathogens where dispersal gradients fitting an inverse distance squared relationship are more typical ( Gregory 1968). However, as Gregory points out, data derived from area (as used here for U. valerianae) rather than point sources tend to flatten dispersal gradients. This flattening is further enhanced by the dispersal of telia on plant debris detached and washed about in drift during autumn and winter storms.

Metapopulation characteristics of uromyces valerianae

The population dynamics of U. valerianae across the study sites was very diverse. Within years, the prevalence of disease in different populations ranged from 0% to 100% ( Fig. 2), while across years, disease levels at individual sites showed similar fluctuations in amplitude. The analyses presented here show that these fluctuations are generated by a range of factors, with the epidemiological dynamics of the pathogen occurring at a given site being driven by a complex interplay of site quality, metapopulation-wide differences in the yearly suitability of the physical environment for pathogen development, host population size, disease prevalence in the previous year, and finally, the incidence of disease in other populations.

Viewed across the entire study area these patterns are not consistent with those expected from either (i) a single epidemiologically homogeneous population, or (ii) a series of entirely separate populations in each of which the interaction between host and pathogen is played out in isolation from all others. Rather, the patterns observed fit those expected of a metapopulation of multiple demes in which the incidence, prevalence and severity of disease may fluctuate violently from time to time and place to place; where pathogen extinction and subsequent recolonization is fairly frequent; where disease incidence is affected by population size and previous history; where neighbouring populations have an influence on disease levels; and where there is some evidence of spatial patterning in disease distribution.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

This study was supported by the Swedish Natural Science Research Council. The authors are grateful to Dr P.H. Thrall for his constructive comments and to Professor Alan Welsh for expert advice on techniques for examining correlation structure in residuals.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • Antonovics, J., Thrall, P.H., Jarosz, A.M., Stratton, D. (1994) Ecological genetics of metapopulations: the Silene–Ustilago plant–pathogen system . Ecological Genetics(ed. L.A.Real), pp. 146 170. Princeton University Press, Princeton, NJ.
  • Augspurger, C.K. & Kelly, C.K. (1984) Pathogen mortality of tropical tree seedlings: experimental studies of the effect of dispersal distance, seedling density and light conditions. Oecologia, 61, 211 217.
  • Burdon, J.J. (1997) The evolution of gene-for-gene interactions in natural pathosystems. The Gene-for-Gene Relationship in Plant–Parasite Interactions (eds I.R.Crute, E.B.Holub & J.J.Burdon), pp. 245 262. CAB International, Oxford, UK.
  • Burdon, J.J., Ericson, L., Müller, W.J. (1995) Temporal and spatial changes in a metapopulation of the rust pathogen Triphragmium ulmariae and its host, Filipendula ulmaria. Journal of Ecology, 83, 979 989.
  • Burdon, J.J., Wennström, A., Elmqvist, T., Kirby, G.C. (1996) The role of race specific resistance in natural plant populations. Oikos, 76, 411 416.
  • Carlsson, U. & Elmqvist, T. (1992) Epidemiology of anther-smut disease (Microbotryum violaceum) and numeric regulation of populations of Silene dioica. Oecologia, 90, 509 517.
  • Carlsson, U., Elmqvist, T., Wennström, A., Ericson, L. (1990) Infection by pathogens and population age of host plants. Journal of Ecology, 78, 1094 1105.
  • Diggle, P.J., Liang, K.-Y., Zeger, S.L. (1994) Analysis of Longitudinal Data. Clarendon Press, Oxford, UK.
  • Dobson, A.J. (1990) An Introduction to Generalized Linear Models. Chapman & Hall, London, UK.
  • Frank, S.A. (1991) Ecological and genetic models of host–pathogen coevolution. Heredity, 67, 73 83.
  • Gandon, S., Capowiez, Y., Dubois, Y., Michalakis, Y., Olivieri, I. (1996) Local adaptation and gene-for-gene coevolution in a metapopulation model. Proceedings of the Royal Society of London, B, 263, 1003 1009.
  • Gilpin, M. & Hanski, I. (1991) Metapopulation Dynamics: Empirical and Theoretical Investigations. Academic Press, London, UK.
  • Gjaerum, H.B. (1974) Nordens Rustsopper. Fungiflora, Oslo, Norway.
  • Gregory, P.H. (1968) Interpreting plant disease gradients. Annual Review of Phytopathology, 6, 189 212.
  • Hanski, I. & Gilpin, M. (1997) Metapopulation Biology: Ecology, Genetics and Evolution. Academic Press, San Diego, CA.
  • James, W.C. (1971) An illustrated series of assessment keys for plant diseases, their preparation and usage. Canadian Plant Disease Survey, 51, 39 65.
  • Jarosz, A.M. & Burdon, J.J. (1992) Host–pathogen interactions in natural populations of Linum marginale and Melampsora lini. III. Influence of pathogen epidemics on host survivorship and flower production. Oecologia, 89, 53 61.
  • Jennersten, O., Nilsson, S.G., Wästljung, U. (1983) Local plant populations as ecological islands: the infection of Viscaria vulgaris by the fungus Ustilago violacea. Oikos, 41, 391 395.
  • McCauley, D.E. (1991) Genetic consequences of local population extinction and recolonization. Trends in Ecology and Evolution, 6, 5 8.
  • Pacala, S.W., Hassell, M.P., May, R.M. (1990) Host–parasitoid associations in patchy environments. Nature, 344, 150 153.
  • Paul, N.D. & Ayres, P.G. (1986) The impact of a pathogen (Puccinia lagenophorae Cooke) on populations of groundsel (Senecio vulgaris L.) overwintering in the field. I. Mortality, vegetative growth and the development of size hierarchies. Journal of Ecology, 74, 1069 1084.
  • Thrall, P.H. & Antonovics, J. (1995) Theoretical and empirical studies of metapopulations: population and genetic dynamics of the Silene–Ustilago system. Canadian Journal of Botany, 73 (Supplement 1), S1249 S1258.
  • Thrall, P.H. & Burdon, J.J. (1997) Host–pathogen dynamics in a metapopulation context: the ecological and evolutionary consequences of being spatial. Journal of Ecology, 85, 743 753.
  • Thrall, P.H. & Burdon, J.J. (in press) The spatial scale of pathogen dispersal: consequences for disease dynamics and persistence. Evolutionary Ecological Research, in press.
  • Tutin, T.G., Heywood, V.H., Burges, N.A., Moore, D.M., Valentine, D.H., Walters, S.M., Webb, D.A. (1976) Flora Europaea, Vol. 4. Cambridge University Press, Cambridge, UK.
  • Wilson, M. & Henderson, D.M. (1966) British Rust Fungi. Cambridge University Press, London, UK.

Received 6 May 1998revision accepted 20 January 1999