The predominance of outcrossing despite the substantial transmission advantage of self-fertilization remains a paradox. Theory suggests that selection can favor outcrossing if it enables the production of offspring that are less susceptible to pathogen attack than offspring produced via self-fertilization. Thus, if pathogen pressure is contributing to the maintenance of outcrossing in plants, there may be a positive correlation between the number of pathogen species attacking plant species and the outcrossing rate of the plant species. We tested this hypothesis by examining the association between outcrossing rate and the number of fungal pathogen species that attack a large, taxonomically diverse set of seed plants. We show that plant species attacked by more fungal pathogen species have higher outcrossing rates than plants with fewer enemies. This relationship persists after correcting for study bias among natural and agricultural species of plants. We also accounted for the nested hierarchy of relationships among plant lineages by conducting phylogenetically independent contrasts (PICs) within genera and families that were adequately represented in our dataset. A meta-analysis of the correlation between pathogen and outcrossing PICs shows that there is a positive correlation between pathogen species number and outcrossing rates. This pattern is consistent with the hypothesis that pathogen-mediated selection may contribute to the maintenance of outcrossing in species of seed plants.
In hermaphroditic plants, self-fertilization enjoys a strong transmission advantage over outcrossing that should lead, all else being equal, to the selective elimination of biparental reproduction (Fisher 1941; Williams 1971; Maynard Smith 1978). However, outcrossing is the predominant mode of reproduction in the plant kingdom (Yampolsky and Yampolsky 1922; Schemske and Lande 1985; Vogler and Kalisz 2001), suggesting that there are substantial benefits of cross-fertilization. Although there are many theories explaining the benefits of producing genetically variable progeny and thus the maintenance of outcrossing, few have found consistent empirical support (West et al. 1999).
In an attempt to identify the factors maintaining outcrossing in plant species, biologists have long examined its ecological correlates. One recurring finding in these studies is the predominance of outcrossing in undisturbed, biologically complex habitats where disease and other “natural enemies” might be expected to be prevalent (Levin 1975; Lloyd 1980; Bell 1982). Levin (1975) proposed that this ecological pattern might result from pathogen-mediated selection for outcrossing in natural plant populations. In theory, an outcrossing strategy may be selectively favored if the genetically novel (Levin 1975) and/or rare (Clarke 1976; Jaenike 1978) offspring produced via outcrossing have higher fitness than common genotypes. This situation may occur if common genotypes are disproportionately attacked by coevolved pathogens (Hamilton et al. 1990). Similarly, selection for genetically variable progeny by pathogens may also explain variation in outcrossing rates among taxa that reproduce through varying amounts of cross-fertilization (Lively and Howard 1994; Agrawal and Lively 2001).
Infection by multiple pathogen species can favor recombination when (1) competition between pathogen species or genotypes favors more virulent species/genotypes than when hosts are singly infected (Nowak and May 1994; May and Nowak 1995; Davies et al. 2002) or (2) the cumulative virulence effects of multiple pathogen species are greater than the virulence of any one pathogen species (Hamilton et al. 1990). Even if simultaneous infection by multiple pathogen species per se does not favor outcrossing, selection for outcrossing may be more common in host species with many natural enemies because the probability of infection by a single, highly virulent pathogen is likely to increase with pathogen species number (sensu Huston 1997; Tilman et al. 1997). Regardless of the underlying mechanism, there should be a positive relationship between outcrossing rate and the number of pathogen species infecting each plant species whenever pressure from pathogens favors recombination. We examined this possibility by compiling data on outcrossing rates and pathogen species abundance across a wide array of seed plant species and analyzing the correlation between these two variables.
Methods
Compilation of Outcrossing Rate Data
The majority of species in the dataset were originally compiled by Barrett and Eckert (1990). These data were analyzed previously to estimate the distribution of natural outcrossing rates in seed plants (Vogler and Kalisz 2001). This dataset is current up to and including 1998. Consequently, we added all reported studies up to and including May 2003. For each species in the dataset, the outcrossing rate is an average across all separate studies reported in the scientific literature. Outcrossing rates are measured in natural populations by using statistical models to infer the parentage of offspring through the use of polymorphic genetic markers. Whenever possible, we used multilocus estimates of outcrossing rate in this study because these estimates have lower error variance and greater adherence to model assumptions than estimates taken from single loci (Ritland and Jain 1981). We did not include any apomictic plant species in our analyses because the lack of recombination per se in asexual lineages cannot be adequately translated into an outcrossing rate.
The distribution of outcrossing rates in the 182 plant species that we surveyed closely matches the bimodal pattern observed in earlier studies of plant outcrossing (Fig. 1; Vogler and Kalisz 2001). A Kolmorov-Smirnov test shows that the distribution of outcrossing rates in our study does not differ in location, dispersion, or skew from the most recently published distribution of outcrossing rates in plants (Dobs = 0.049 ≪ 0.138 = Dcrit(α=0.05); Sokal and Rohlf 1995). This means that the distribution of outcrossing rates in the present study is likely to be an unbiased sample of the reproductive systems found in species of seed plants (Schemske and Lande 1985; Vogler and Kalisz 2001).
Estimation of the Number of Pathogen Species Attacking Plants
We cross-referenced plant species against a database listing all fungal pathogens documented to infect many species of natural and agricultural plants (Farr et al. 1989, 2003). This database lists all of the fungal species known to infect both living plant species and their agriculturally important products. To avoid counting single fungal species more than once, we only tallied pathogen species for which a species epithet was listed. Before conducting our analyses, the number of pathogen species was natural log-transformed so that this variable was normally distributed. There were significant differences between the number of fungal pathogen species known to attack wild and agricultural species (mean agricultural species = 43.1, wild species = 6.3; independent samples t-test, t = 10.249, df = 212, P < 0.001). These reported means are back-transformed marginal means taken from a one-way ANOVA.
Removal of Study Biases
It is important to account for the likelihood that the significant difference in mean fungal pathogen species number between agricultural and wild species is driven by the inherent biases in study effort toward species that are important to humans (e.g. Gregory 1990; Nunn et al. 2003). We dealt with this issue by using a statistical approach, partial correlations, that accounts for the positive correlation expected between study effort and pathogen number. By removing the influence of study effort on the number of pathogen species recorded for each plant species, we were able to compute an unbiased correlation between outcrossing rate and the number of pathogens known to attack a diverse group of plants. We removed the study bias from our data by counting the number of Web of Science citations found for each plant's Latin binomial species name from 1987 to 2003 and removing correlations between this variable and the two variables of interest, pathogen species number and outcrossing rate (Zar 1996). More specifically, we performed separate linear regressions of pathogen species number and outcrossing rate on citation number and saved the unstandardized residuals, which represent the variation in pathogen species number and outcrossing rate not explained by citation biases. There was a significant negative relationship between outcrossing rate and study effort (F = 8.297; df = 181; P < 0.004; R2 = 0.044; β = −0.21), and a significant positive relationship between pathogen species number and study effort (F = 11.199; df = 181; P < 0.001; R2 = 0.059; β = 0.242). We used the residuals from these regressions to compute a nonparametric rank correlation between these unbiased measures of pathogen species number and outcrossing rate. We used a nonparametric test because the bimodal distribution of outcrossing rates (Fig. 1) violated assumptions of bivariate normality necessary for interpretation of hypothesis tests (Zar 1996).
Accounting for Phylogenetic Nonindependence
Another potentially confounding factor is phylogenetic nonindependence. More specifically, because species exhibit varying amounts of relatedness, one must control for the possibility that relatedness may contribute to similarity between closely related species, thereby violating the assumption that outcrossing values or pathogen species numbers estimated for different species are independent datapoints (Felsenstein 1985). We attempted to control for phylogenetic nonindependence in several ways. First, we computed the mean residual outcrossing rate and mean residual pathogen number within the five plant families with an adequate sample of species (N > 10; Pinaceae, Poaceae, Asteraceae, Myrtaceae [all species of Eucalyptus], and Fabaceae) and examined the correlation between these variables. This analysis between families corrects for the nonindependence of datapoints within these families.
To more adequately test the idea that pathogens may favor outcrossing, it is necessary to examine changes in both pathogen susceptibility and outcrossing between plant lineages that are not a product of shared recent ancestry (Harvey and Pagel 1991). To remove potentially confounding variation caused by shared inheritance among groups, we conducted formal phylogenetically independent contrasts (PICs) using the method described by Felsenstein (1985). This method is based upon the assumption that traits evolve by a process of Brownian motion along the branches of phylogenetic trees. Under such a model, trait evolution occurs independently within each lineage, such that the expected difference between two taxonomic units has a mean of zero and a variance proportional to time, or the sum of the branch lengths separating the two groups (Felsenstein 1985). As a consequence, the observed differences between taxa can be standardized by their divergence times. Once all of the possible independent comparisons are conducted along the topology of a tree, it is possible to test hypotheses using standard parametric statistics because the changes observed along more terminal branches are independent of those occurring on more basal branches (Felsenstein 1985).
Phylogenetic Analyses and Meta-Analysis
We constructed phylogenetic trees describing relationships at both the family and genus level (Pinaceae, Pinus, Poaceae, Asteraceae, Eucalyptus, Fabaceae), and then generated a supertree connecting the five plant families (Pinaceae, Poaceae, Asteraceae, Myrtaceae, Fabaceae). In light of the fact that different sequences evolve at different rates and are informative only at certain levels of genetic divergence, not all phylogenetic trees were based upon the same loci. Sequence data were obtained from the NCBI website ( www. ncbi.nlm.nih.gov) for the species or genera represented in the dataset. The following genes were used to estimate relationships: matK (Asteraceae, Fabaceae, and Pinus); NdhF (Poaceae); the 18S rRNA subunit (Pinaceae); and the 18S, 5.8S, and 26S rRNA subunits along with two internal transcribed spacer regions (Eucalyptus), and matR (supertree). The relationships among taxa in these groups were estimated using the neighbor-joining algorithm in PAUP (Swofford 1998).
Once the topology of the tree and the branch lengths were estimated, the residual number of pathogens and outcrossing rate for a plant species or genus were then computed and mapped onto the phylogenetic tree. Phylogenetically independent contrasts were conducted in Mesquite (Maddison and Maddison 2003). In particular, pathogen and outcrossing contrasts were computed using the PDAP:PDTree module (Midford et al. 2003). The correlation between outcrossing and pathogen species number contrasts was then computed in SPSS (Rel. 11.5.2.1, SPSS Inc. 2003). For the PIC analysis conducted on the supertree, the estimates of pathogen and outcrossing for each family were obtained by inferring the ancestral state of both variables within each plant family using a model of squared-change parsimony (Maddison 1991).
A meta-analysis was conducted on the correlation observed between the PICs of residual pathogen species number and outcrossing rate taken from each level of analysis. Meta-analyses were conducted using MetaWin (Rosenberg et al. 2000). The within-family correlations (ρi) were z-transformed to normalize the distribution of coefficients and to make the variance independent of the common coefficient: zi = 1/2 ln(1 + ρi/1 − ρi). These normalized coefficients were then weighted by their relative sample size and summed together (Hedges and Olkin 1985). We compared the test statistic obtained from the Fisher's z-test to the two-sided critical value of the standard normal distribution. In light of the fact that the z-transformed correlation coefficients were not normally distributed (Shapiro-Wilk statistic = 0.735; df = 7; P = 0.009), the results of this statistical test must be interpreted with caution. As an alternative, we used resampling methods to directly evaluate the significance of the correlation given the data. We performed 1000 bootstrap replicates by randomly sampling with replacement from the seven correlations to generate a distribution of possible values. If the 95% confidence interval of the bootstrapped correlation failed to overlap zero, then the correlation was judged to be significantly different than zero (Sokal and Rohlf 1995).
Results and Discussion
Once the potentially confounding effects of study effort were removed, we found a strong positive partial correlation between the number of fungal pathogen species attacking each plant species and its outcrossing rate (Fig. 2; Spearman ρ = 0.234, P < 0.001, N = 182). We also found that there was a significant positive correlation between pathogen species number and outcrossing rate across the Asteraceae, Fabaceae, Myrtaceae, Pinaceae, and Poaceae (Pearson r = 0.975; P = 0.005; Fig. 3). However, both of these correlations suffer from a large amount of phylogenetic nonindependence. For example, there is a large degree of recent shared ancestry between angiosperm families (e.g. Asteraceae, Fabaceae, Myrtaceae, and Poaceae) that separates them from gymnosperms (Pinaceae), such that the datapoints contributed by angiosperm families are not statistically independent of each other.
From a comparative standpoint, is also important to consider the timescale over which pathogen-mediated selection may favor higher outcrossing rates (Felsenstein 1985). Since taxonomic treatments of plant groups suggest that outcrossing rates are often highly variable within plant genera and families (Stebbins 1950; Barrett et al. 1996), it was imperative to account for shared inheritance over these relatively short timescales. In all of the PIC analyses, the correlation between outcrossing rate and susceptibility to pathogens was positive, though not significantly greater than zero (Fig. 4A– G). To conduct a more general test of the idea that pathogens may selectively favor outcrossing in plants, we performed a meta-analysis combining the correlations observed at each of the seven taxonomic levels. Results of the meta-analysis suggest that there is a significant and positive correlation between outcrossing rate and fungal pathogen number in seed plants (Table 1). This finding is consistent with the idea that there may be a link between pathogen attack and the maintenance of outcrossing.
Both in the genus Pinus and within the family Pinaceae, the correlation coefficient between outcrossing rate and pathogen number exceeds 0.85 with a sample size of only four datapoints. In contrast, the correlation coefficients for other plant groups were of a much lower magnitude (Fig. 4). Interestingly, Pinus and Pinaceae represent two of the three groups characterized primarily by woody and relatively long-lived organisms. This trend for a tighter correlation between outcrossing rates and pathogen attack in longer-lived organisms is consistent with the pattern expected if pathogen pressure favors outcrossing in natural populations. In particular, whenever pathogens have much shorter generation times than their hosts, pathogen populations may adapt to individual host genotypes as they age (Rice 1983). Closer adaptation of pathogen populations to individual host genotypes also increases the chances that one or more pathogen genotypes will be highly virulent on any given host (Mutikainen et al. 2000; Thrall et al. 2002). In such a situation, the production of rare or novel genotypes by outcrossing may generate offspring that are capable of evading locally adapted pathogens. The trend for tighter correlations between outcrossing rates and pathogen attack in long-lived organisms is similar to the observation that longer-lived mammals have higher rates of recombination (Burt and Bell 1987). Continued combat with antagonists over evolutionary time should favor high levels of recombination whenever hosts have much longer generation times than their pathogens (Hamilton 1982), and may help to explain the well-documented correlation between outcrossing and the perennial habit in seed plants (Stebbins 1950; Barrett et al. 1996).
An alternative explanation for the positive correlation between outcrossing and the number of pathogen species attacking plants is that self-fertilizing taxa may have smaller range sizes and thus encounter fewer pathogen species. In this case, a positive correlation between outcrossing rate and pathogen species number would not clearly implicate pathogen-mediated selection for outcrossing. Although self-fertilizing populations may often be effectively small (Schoen and Brown 1991), the demographic flexibility associated with self-fertilization allows such species to successfully found new populations following long-distance dispersal, thereby allowing them to spread over greater geographical distances (Baker 1955, 1967). If inbreeding taxa have larger ranges, then we would have observed a negative relationship between outcrossing rate and pathogen species number. Currently there is no consensus on the differences in range size between inbreeding and outcrossing plant species because this pattern has been studied in relatively few taxonomic groups (Stebbins 1957; Lloyd 1980).
It is possible that other factors unrelated to parasite-mediated selection may also contribute to the detected correlation between outcrossing rate and attack by fungal pathogen species. Although we are unable to eliminate all potentially confounding factors, our results support the idea that high rates of outcrossing in seed plants may be favored by pathogen-mediated selection in natural populations. The conclusion that outcrossing is favored by pathogen pressure is strengthened in light of the fact that short-term field experiments have shown that outcrossing is favored when pathogen or parasite pressure is intense (Schmitt and Antonovics 1986; Kelley 1994), and by field surveys showing that asexuality may be more common in regions where parasites are rare or absent (Lively 1987, 1992; Schrag et al. 1994; Kumpulainen et al. 2004).
Conclusions
In general, the results of these analyses demonstrate that there is a positive correlation between outcrossing rate and the number of fungal pathogens known to attack species of seed plants. We showed that this correlation is not driven by biases resulting from the disproportionate study of species important to humans. Moreover, the fact that the correlation between outcrossing and pathogen number remains after controlling for phylogenetic nonindependence suggests that this pattern is most likely not an artifact of shared inheritance. We also discussed the possibility that differences in range size between species using different reproductive strategies may confound this correlation, and tentatively conclude that outcrossing and self-fertilizing plant species are not likely to encounter natural enemies at consistently different rates. Taken together, these results provide evidence in support of the hypothesis that pathogen pressure may contribute to the maintenance of outcrossing in natural plant populations.
Acknowledgments
The authors thank S. Barrett and C. Eckert for generously providing access to their data on outcrossing rates in seed plants. A copy of our dataset is available upon request to C. Eckert via eckertc@biology.queensu.ca. The authors also thank G. Burleigh, S. Mathews, J. Palmer, and A. Richardson for providing relevant sequence data and advice on phylogenetic analyses. We thank A. Agrawal, J. Antonovics, K. Clay, L. Delph, F. Frey, S. Hamm, B. Koskella, C. Lively, E. Osnas, J. Rudgers, J. Webster, and members of the Clay and Delph-Lively labs for comments on drafts of the manuscript. We also thank A. Peters for his suggestions and help throughout the review process, and an anonymous reviewer for helpful criticisms of earlier versions of our paper. This research was funded by the National Science Foundation (JWB), the Graduate School at Indiana University (MN), and the Indiana University Department of Biology (JMK).
Literature Cited
Fig. 1.
Distribution of outcrossing rates among sexually reproducing plant species for which the number of fungal pathogen species was estimated. This histogram describes variation in the amount of outcrossing within the 182 sexual plant species included in the study
Fig. 2.
Bivariate plot of the residual outcrossing rate versus a plant species' residual log transformed number of fungal pathogen species infecting it. The residuals represent the variation not explained by possible study bias. Residuals were computed from univariate regressions of each variable against the number of Web of Science citations for each plant species from 1987 to 2003. The line segment represents the relationship predicted by linear regression. The outlying point represents Arabidopsis thaliana, which has the largest number of citations in the database
Fig. 3.
Bivariate plot of the average residual outcrossing rate and residual log transformed pathogen species number in five well-studied plant families. Residuals were computed from the univariate regressions of each variable against the number of Web of Science citations for each plant species from 1987 to 2003 and averaged within families. The line segment represents the relationship predicted by linear regression
Fig. 4.
Bivariate plots of residual outcrossing rate and residual log-transformed pathogen species numbers obtained from phylogenetically independent contrasts. Contrasts for both variables are standardized by the sum of the branch lengths separating taxa (A, Pinaceae; B, Pinus; C, Poaceae; D, Asteraceae; E, Eucalyptus; F, Fabaceae; G, supertree). The correlation for each group is shown along with the P-value associated with testing the null hypothesis that the correlation equals zero. Although all of the correlations are positive, none of them are significantly different from zero. The line segments represent the relationships predicted by linear regression
Table 1.
Results of the meta-analysis conducted on results of the seven groups of phylogenetically independent contrasts. The cor relation coefficient was estimated through the use of z-transfor mations weighted by the sample size of each correlation. The com bined z-statistic was compared against the standard normal distri bution. The P-value of this statistical test is shown next to the parametric confidence interval (CI) of the correlation. A nonpara metric resampling test was conducted because the sample corre lations were not normally distributed. The bootstrapped 95% con fidence interval was computed by sampling with replacement 1000 times from the seven correlations. This test showed that the 95% confidence interval did not include zero, suggesting that the positive correlation between plant outcrossing rate and fungal pathogen number is significantly different than zero
