Results from the simulation demonstrate a significant negative relationship between the mean number of resistance genotypes per host population and disease prevalence (adj. r2 = 0·50, P < 0·0001; Fig. 1a), while a strong positive relationship (adj. r2 = 0·61, P < 0·0001; Fig. 1b) was predicted between the number of resistance genotypes per population and disease incidence (presence/absence of disease in host populations). This may seem counter-intuitive, but makes sense when considered in the context of the coevolutionary dynamic taking place. Conditions that lead to increased numbers of host resistance genotypes also select for an increase in pathogen virulence types (but not necessarily the average virulence of those pathogens). Results from the model show that the total number of host and pathogen genotypes present in the metapopulation are indeed highly positively correlated (r2 = 0·83, P < 0·0001) over a wide range of dispersal conditions. As the number of pathogen types rises in the system, the fraction of host populations in which infection can successfully establish also increases, even though average prevalence may decrease.
Figure 1. The relationship between (a) disease prevalence (average fraction of infected individuals/population), (b) disease incidence (fraction of populations with disease present) and the mean number of resistance genotypes per population as predicted by a simulation model of evolutionary dynamics in a gene-for-gene metapopulation. Plotted points are the means of 10 random runs, and show the equilibrium values reached after 1000 host generations.
Download figure to PowerPoint
Similar to the patterns seen for the number of host and pathogen genotypes per population, disease prevalence was negatively related to the total number of host resistance genotypes across the metapopulation (adj. r2 = 0·63, P < 0·0001), as well as the number of resistance genes per individual (adj. r2 = 0·26, P < 0·001). These variables were also both positively related to disease incidence (adj. r2 = 0·40, P < 0·0001; adj. r2 = 0·33, P < 0·0001, respectively).
Relationships between resistance measures and disease prevalence in Linum
The resistance and virulence structures of the studied host and pathogen populations were determined for the populations extant in the 1995–96 growing season. Association between those data and disease prevalence in individual populations was assessed for the following three growing seasons. Regressions based on arcsin-square root-transformed values showed a significant negative relationship between mean resistance and disease prevalence in 1998–99 (r2 = 0·44, P = 0·02; Fig. 2e), and a nearly significant relationship in 1996–97 (r2 = 0·26, P = 0·11; Fig. 2a). No such relationship was evident in 1997–98 (Fig. 2c) when disease levels across the metapopulation as a whole were very low (mean peak incidence levels: 1996–97 = 84·6%; 1997–98 = 9·3%; 1998–99 = 51·4%).
Figure 2. Regression analyses showing the relationship between the average resistance of 10 populations of L. marginale and the prevalence of disease caused by M. lini (left-hand column of panels), and the relationship between the diversity of resistance phenotypes and disease prevalence (right-hand column of panels) in the same populations at the peak of the epidemic in three successive growing seasons (a, b: 1996–97; c, d: 1997–98; e, f: 1998–99). Both average resistance and disease prevalence values were arcsin-square root-transformed prior to analysis. Note that prevalence data were only available for eight populations in the 1996–97 growing season.
Download figure to PowerPoint
Measures of diversity in disease resistance phenotypes were calculated using the corrected Shannon–Weaver index (Hutcheson, 1970) and these were then regressed against disease prevalence at the peak of the epidemic for the same three growing seasons as the previous analyses. Although this relationship was not significant at the P = 0·05 level for any year, the pattern across census periods was similar to that seen for mean resistance in that in years of high disease incidence (Figs 2b and f), the relationship closely approached significance (1996–97: r2 = 0·30, P = 0·09; 1998–99: r2 = 0·23, P = 0·09). It should be noted that even in the simulation models, where stochastic effects are, of necessity, much smaller than would occur in the real world, considerable unexplained variation is apparent in the relationship between population levels of resistance diversity and disease prevalence. In this light, it is encouraging that the patterns based on observations from natural populations (where chance effects play a far greater role) are as strong as they are, and in the predicted directions.
The marked year-to-year variation in the strength of these relationships can be best understood by an examination of the explanatory power of measures of resistance in relation to peak mean disease prevalence (Fig. 3). For both average resistance and resistance diversity, there was a broadly asymptotic relationship, such that adjusted r2 values for these two parameters approached a maximum for disease prevalence levels greater than 60–80%. This reflects the dynamics of the interplay between the pathogen and its host. In years where disease increase is limited, the opportunities for selective interaction between pathotypes and different host resistance phenotypes are also restricted. As a consequence, the amount of variation that can be explained is also limited. In epidemic years, on the other hand, there is far greater opportunity for this interaction to occur and have an immediate impact on disease.
Figure 3. Relationship between the prevalence of disease caused by M. lini at peak epidemic (averaged across the 10 host populations) and the explanatory power of mean population level resistance (●), and resistance diversity (○).
Download figure to PowerPoint
In sharp contrast to the patterns observed for mean host resistance, there was no relationship between disease prevalence at any of the epidemic peaks and mean virulence (the number of resistance phenotypes in the differential set that could be attacked) of the 1995–96 pathogen populations. However, there were highly significant relationships between these variables when prevalence data derived from earlier phases of each epidemic was used. This strongly suggests that early in the development of an epidemic, pathogen virulence and its relationship to host resistance are important factors determining establishment in host populations. This relationship was highly positive in the 1996–97 epidemic (r2 = 0·63, P = 0·01, Fig. 4a), the growing season immediately following that in which host resistance and pathogen virulence was originally assessed. In contrast, it was significantly negative in all subsequent years examined (1997–98: r2 = 0·48, P = 0·02, Fig. 4b; 1998–99: r2 = 0·39, P = 0·03, Fig. 4c). Given the severe epidemic during the 1996–97 growing season, these results may indicate selection against L. marginale genotypes susceptible to the pathogen populations present when virulence was assessed during the 1995–96 season. This possibility is supported by evidence from the same system demonstrating that population resistance structure can change substantially from one growing season to the next as a consequence of high disease levels (Burdon & Thompson, 1995). If true, and such selection has occurred, then one would expect the slope of the relationship to change in the observed direction as most host populations would show resistance against the pathogen populations present in 1995–96.
Figure 4. Relationship between mean virulence of populations of M. lini (as assessed during the 1995–96 epidemic) and disease prevalence in those same populations during the early phase of the epidemic in each of three successive field seasons. Note that mean virulence was calculated based on the number of host differential lines that could be attacked (see Methods). Both mean virulence and disease prevalence values were arcsin-square root-transformed prior to analysis.
Download figure to PowerPoint
Disease prevalence and pathogen diversity (calculated using the corrected Shannon–Weaver index) were not correlated at any point in the epidemics for the years investigated. There are two likely explanations for this result. First, M. lini undergoes severe bottlenecks in population size at the end of every growing season, and consequently is subject to the vagaries of local colonization-extinction dynamics (Burdon & Jarosz, 1992). Secondly, pathogen virulence was assessed on a host differential set containing 10 different resistance alleles, only half of which are known to occur in the immediate vicinity of the Kiandra metapopulation.
At first glance, the finding that average resistance is negatively correlated with disease prevalence may not appear surprising. However, within this metapopulation, there were always pathotypes capable of overcoming all of the resistance phenotypes identified in this study (J. J. Burdon, unpublished data). Thus, the observation that disease declines as average resistance increases is not just a simple reflection of an increase in host lines resistant to all locally occurring pathotypes. Earlier work on this system has shown that pathogens isolated from diverse host populations are more virulent than those derived from host populations with low variability for resistance (Burdon & Jarosz, 1992; PH Thrall, JJ Burdon & AG Young, unpublished data). Given that these pathogen populations are generally highly variable, and relatively mobile, this suggests that the trade-offs between aggressiveness and virulence observed in agricultural studies (Chin & Wolfe, 1984) may also be an important factor in the dynamics of the interaction between L. marginale and M. lini.
It has been noted by several workers (May & Anderson, 1990; Frank, 1997) that very few empirical studies of natural host–pathogen interactions have incorporated both spatial and temporal aspects of disease dynamics and genetic structure. The present study addresses these issues and provides the first evidence from a natural host–pathogen system that increasing diversity of resistance reduces disease. This extends observations made in the highly controlled conditions of agricultural fields to native systems, and reinforces the value of long-term studies of multiple populations for understanding coevolutionary systems.