Rapid change in host specificity in a field population of the biological control organism Pasteuria penetrans

Abstract In biological control, populations of both the biological control agent and the pest have the potential to evolve and even to coevolve. This feature marks the most powerful and unpredictable aspect of biological control strategies. In particular, evolutionary change in host specificity of the biological control agent could increase or decrease its efficacy. Here, we tested for change in host specificity in a field population of the biological control organism Pasteuria penetrans. Pasteuria penetrans is an obligate parasite of the plant parasitic nematodes Meloidogyne spp., which are major agricultural pests. From 2013 through 2016, we collected yearly samples of P. penetrans from eight plots in a field infested with M. arenaria. Plots were planted either with peanut (Arachis hypogaea) or with a rotation of peanut and soybean (Glycine max). To detect temporal change in host specificity, we tested P. penetrans samples annually for their ability to attach to (and thereby infect) four clonal lines of M. arenaria. After controlling for temporal variation in parasite abundance, we found that P. penetrans from each of the eight plots showed temporal variation in their attachment specificity to the clonal host lines. The trajectories of change in host specificity were largely unique to each plot. This result suggests that local forces, at the level of individual plots, drive change in specificity. We hypothesize that coevolution with local M. arenaria hosts may be one such force. Lastly, we observed an overall reduction in attachment rate with samples from rotation plots relative to samples from peanut plots. This result may reflect lower abundance of P. penetrans under crop rotation, potentially due to suppressed density of host nematodes. As a whole, the results show local change in specificity on a yearly basis, consistent with evolution of a biological control organism in its ability to infect and suppress its target pest.


| INTRODUC TI ON
Host specificity is the foremost challenge for the safe and effective use of biological control agents (reviewed in Brodeur, 2012;Fagan, Lewis, Neubert, & Driessche, 2002;McEvoy, 1996). Upon release into a novel range, generalist predators or parasites may attack unintended prey or host species, leading to a decline in their population size and an increase in extinction risk (e.g., Boettner, Elkinton, & Boettner, 2000;Louda, Kendall, Connor, & Simberloff, 1997;reviewed in Louda, Pemberton, Johnson, & Follett, 2003;Simberloff & Stiling, 1996). Such nontarget effects motivated a shift toward biological control agents with narrow host ranges. Efforts to identify these more specialized candidates include extensive prescreening of potential host ranges and a focus on native enemies, for which the host range within the local community is known and likely limited by a prior evolutionary history (Brodeur, 2012;Greathead, 1995;Roderick, Hufbauer, & Navajas, 2012;Secord & Kareiva, 1996;Waage, 2001). Narrow host ranges bring their own set of limitations, however. If a population of a biological control agent is specialized to attack only a subset of genotypes within a single host species, we would predict limited efficacy against the majority of host populations (Parker, 1985) (reviewed in Brodeur, 2012).
In theory, rapid evolution can ameliorate the problem of high specificity in biological control. Indeed, a major proposed advantage of biotic over abiotic control is that biological antagonists can rapidly adapt as host populations evolve (e.g., Dwyer, Levin, & Buttel, 1990;Hajek, Humber, & Elkinton, 1995;reviewed in Hopper, Roush, & Powell, 1993;Roderick & Navajas, 2003). The field of host-parasite coevolution provides a valuable theoretical framework for predicting evolutionary trajectories. Coevolutionary theory predicts that parasites adapt to infect locally common host genotypes (Haldane, 1949), and several studies support this prediction (Chaboudez & Burdon, 1995;Koskella & Lively, 2009;Lively & Dybdahl, 2000;Wolinska & Spaak, 2009). Thus, high host specificity may not prevent a parasite from serving as an effective biological control. With sufficient selection and genetic variation, a population of parasites applied as biological control could adapt to infect and suppress a diversity of host genotypes across populations (Gandon, 2002;Gandon & Michalakis, 2002). Clearly, the potential for rapid evolution of host specificity is a significant factor to consider in evaluation of biological control agents. Yet there exists little evidence of the evolution of specificity in biological control systems (though see Le Masurier & Waage, 1993;Salt & van den Bosch, 1967) (reviewed in Hufbauer & Roderick, 2005;van Klinken & Edwards, 2002).
Pasteuria penetrans offers a biological alternative to chemical controls, which are expensive, toxic, and increasingly difficult to obtain (Onkendi et al., 2014). Endospores of P. penetrans attach to the cuticles of juvenile nematodes as they migrate through the soil in search of roots. Once a nematode establishes within a root, the attached bacterium penetrates the cuticle and develops vegetatively within the host body (Sayre & Wergin, 1977). Females infected with P. penetrans produce few to no offspring (Bird, 1986;Bird & Brisbane, 1988).
Application of P. penetrans to plots can dramatically reduce nematode densities and increase crop yield (Brown, Kepner, & Smart, 1985;Mankau, 1975Mankau, , 1980Mankau & Prasad, 1972). As a result, P. penetrans is under consideration as a commercial product for control of Meloidogyne spp. A related species, P. nishizawae, is already being marketed to control the soybean cyst nematode Heterodera glycines.
Prior studies suggest that genotypes of P. penetrans are highly specialized, infecting a narrow subset of host genotypes. A single population of P. penetrans varies dramatically in its potential to infect different host populations (Davies, Kerry, & Flynn, 1988;Duponnois, Fargette, Fould, Thioulouse, & Davies, 2000;Spaull, 1984;Stirling, 1985). For example, when populations of M. incognita in Ecuador were challenged with a single population of P. penetrans, the frequency at which endospores attached to the cuticles of hosts from different populations ranged from <10% to >90% (Trudgill et al., 2000).
In addition, a single population of P. penetrans contains a diversity of genotypes that vary in their host specificity (Davies, Redden, & Pearson, 1994;Joseph, Schmidt, Danquah, Timper, & Mekete, 2017;Timper, 2009). Extensive study of P. ramosa, a related parasite of the cladoceran Daphnia magna, indicates that high host specificity may be a feature of this genus. Infection success of P. ramosa varies with its ability to attach to the cuticle in the host esophagus Duneau, Luijckx, Ben-Ami, Laforsch, & Ebert, 2011;Ebert et al., 2016). Switching the identity of a single host allele can render a resistant host genotype susceptible, likely by enabling parasite attachment (Luijckx, Fienberg, Duneau, & Ebert, 2012, 2013Metzger, Luijckx, Bento, Mariadassou, & Ebert, 2016). Similarly, the interaction of proteins on the spore surface and the nematode cuticle may mediate attachment and, thereby, specificity of P. penetrans (Davies, 1994(Davies, , 2009Davies et al., 1994). There is some evidence that specificity can evolve in this system (Oostendorp, Dickson, & Mitchell, 1990;Timper, 2009;Tzortzakakis & Gowen, 1994). Channer and Gowen (1992) reared a population of P. penetrans on three novel host populations. In one case, they found that the parasite population increased in its ability to attach to the novel host population with which it was reared. It simultaneously lost its ability to attach to its original host population. Hosts can also evolve resistance: glasshouse experiments showed a decrease in attachment rate as a host population was continually challenged with a static parasite population (Tzortzakakis, Gowen, & Goumas, 1996), with the evolved resistance specific to the parasite population in the experiment (Tzortzakakis & Gowen, 1994). A fundamental question remains as follows: Does host specificity rapidly change in field populations of this parasite?
We addressed this question by testing for spatial and temporal variation in the host specificity of P. penetrans sampled from plots in an experimental agricultural field. Long-term sampling of this study site provided preliminary evidence for change in host specificity: Starting in 1998, we used a single laboratory population of M. arenaria to assay the soil for abundance of P. penetrans endospores. Large numbers of endospores attached to assayed nematodes in 1998 and 1999. In subsequent years, the numbers of attached endospores declined. Repeating the assay with different host lines revealed that this decline in attachment did not occur because endospores had disappeared from the soil, but because the P. penetrans population had lost the ability to attach to the standard laboratory population of M. arenaria (Timper, 2009). We hypothesized that P. penetrans populations evolve rapidly in their host specificity. To test this hypothesis, we sampled endospores from eight plots of the experimental field from 2013 to 2016. For each plot, we measured variation in host specificity by quantifying the attachment rate of each endospore sample to four clonal lines of M. arenaria. After controlling for variation in endospore abundance, we observed substantial yearly change in attachment rates of field-sampled P. penetrans to the tested host lines, consistent with the hypothesis of rapid evolution of host specificity.
Though a single spore suffices for an infection to establish, a J2 can acquire many endospores. Attachment of multiple endospores can impair J2 mobility, preventing it from finding and establishing in a plant root (Stirling, 1984). Germination of attached endospores depends upon cues associated with establishment of the J2 host in the root. The endospore produces a germ tube that penetrates the host cuticle, entering the pseudocoelom. The bacterium then develops and proliferates, reducing or eliminating host reproduction in the process (Mankau & Imbriani, 1975;Phani & Rao, 2018).
This study focuses on host-parasite specificity at the attachment stage. Variation in infection could arise at multiple stages in the life cycle of Meloidogyne and P. penetrans (for full life cycle diagram, see Figure 1 of Preston et al., 2003). Successful infection requires that a Pasteuria bacterium persist in soil, make contact with a J2 host, attach to the host's cuticle, penetrate the cuticle to enter the host's body cavity, overcome within-host defenses (Tarr, 2012), and reproduce. Each of these steps may be influenced by host genetics, parasite genetics, environmental factors, and their interactions Kruitwagen, Beukeboom, & Wertheim, 2018).
For example, endospore attachment varies with temperature, pH (Chen & Dickson, 1998), and plant root exudates (Liu, Timper, Ji, Mekete, & Joseph, 2017). In addition, replication rate of P. penetrans increases with temperature (Chen & Dickson, 1998;Lopes, Orr, & Blok, 2018). In spite of all this variation, it is possible to identify steps at which host and parasite are most likely to respond to selection: Experiments spanning the infection process for P. ramosa on Daphnia hosts highlight attachment as the infection step most strongly influenced by the genetic interaction of host and parasite. Variation in the attachment step is thus most likely to drive coevolutionary interactions for P. ramosa (Duneau et al., 2011;Ebert et al., 2016;Luijckx, Ben-Ami, Mouton, Pasquier, & Ebert, 2011). It remains to be determined whether these findings also apply to P. penetrans-nematode interactions.

| Experimental site and design
The study site was a 0.77 hectare agricultural field at the University of Georgia Gibbs Farm, Tifton, Georgia, USA, that was naturally infested with M. arenaria race 1 and P. penetrans. The soil was a Tifton loamy sand (fine-loamy, siliceous, thermic Plinthic Kandiudult; pH 6.1). The site contained a crop rotation experiment where the primary crop was peanut. There were 17 rotation sequences (treatments) arranged in a randomized complete block design with four replications. Samples for P. penetrans were collected from replicate plots of two treatments in the experiment: peanut (cv. GA-06G) rotated with soybean (cv. Pioneer 95Y20) (P-S-P) and continuous peanut (P-P-P). Both peanut and soybean are good hosts for M. arenaria.
The plot dimensions and locations of sampled plots are shown in Supporting information Figure S1.
In the spring, the soil was plowed to a depth of 20 to 25 cm before shaping into planting beds 1.8 m wide and 10 to 15 cm high. Each plot included four beds. The field was planted with a new seed lot each year. Both peanut and soybean were planted in May with seeds spaced in two rows, 0.9 m apart on the bed, with 20 seed/m for peanut and 24 seed/m for soybean. For the P-S-P rotation, soybean was planted in 2012, 2014, and 2016. Crop management, including fertilization and pesticides, were conducted based on recommendations for the area (Guillebeau, 2009). Crops were harvested at optimum maturity in mid-September to late September.

| Field sampling and bioassay of specificity
Four clonal lines of M. arenaria (C3, C6, C8, and C40) previously found to differ in patterns of endospore attachment were used to assay P. penetrans endospores from subsamples of soil from each plot. The clonal lines were obtained in 2008 by picking single egg masses from peanut growing in field soil infested with the nematode (Timper, 2009). They were maintained on eggplant (Solanum melongena, cv. Black Beauty) in a glasshouse at 22-30°C. We also assayed P. penetrans endospores using hosts from a glasshouse line (GH) of M. arenaria that had been originally isolated from Gibbs Farm in the early 1990 s. This line was maintained on tomato (Solanum lycopersicum L. cv. Rutgers) and eggplant in the glasshouse. In this study, we focus on data collected from the four clonal host lines, except in the case of one analysis. Hosts for the attachment assay were obtained by placing roots with egg masses separately in a mist chamber. The hatched J2 were collected 3-4 days later.
Soil for assaying P. penetrans specificity was collected from the center of two beds of each plot in early October from 2013 to 2016. For each plot, 10 root-zone soil cores (2.5 cm diam; 15 cm deep) were collected in each row and mixed thoroughly to obtain a large, representative sample of endospores from the plot. All soil was heated at 60°C for 2 hr before use to kill the native M. arenaria.
Specificity was measured using a bioassay previously described by Timper et al. (2001). In this assay, the four clonal host lines served as probes to detect shifts in specificity of P. penetrans in the field.
A subsample of 100 cm 3 soil was added along with tap water to a flask, and the flask was shaken vigorously to make a slurry before decanting the soil-water suspension into another 250-ml flask.
Second-stage juveniles (1,500 J2) of one clonal line of M. arenaria were added to the soil-water suspension and shaken on a rotary shaker at 150 rpm. After 24 hr, the J2s were extracted by centrifugal floatation (Jenkins, 1964), and the number of endospores adhering to 25 randomly selected J2s was determined at 400× magnification with an inverted microscope. For each soil sample, we repeated this bioassay for all four clonal host lines. The relative differences in attachment rate to J2s of the four clonal host lines provided a quantitative estimate of the host specificity of the tested parasite sample.

| Statistical analyses
For the following analyses, we converted spore counts per host to a binomial variable: 0 for zero endospores attached and 1 for one or more endospores attached. This conversion enabled us to broadly compare hosts that could potentially be parasitized (endospores attached) to those that could not (no endospores attached). All models were fit to this binomial response variable using a logit link function.
Both the abundance and the specificity of P. penetrans endospores in tested soil may contribute to variation in the rates of endospore attachment to tested hosts. For example, attachment rates could be elevated in a particular assay if the parasite population size is large (i.e., abundance or dose) or if alleles conferring ability to attach to the tested host line are at high frequency in the parasite population. We used a statistical approach that enabled us to test for changes in specificity by controlling for the contribution of endospore abundance to variation in attachment rates. Specifically, we included main effects (e.g., year, plot) and, where relevant, a two-way interaction (i.e., plot × year) to control for intrinsic variation between parasite sources. Abundance of parasite endospores is an intrinsic difference between parasite sources; for example, attachment rates will be relatively high from soil drawn from a plot with high parasite abundance, regardless of the host line used. In this case, including plot as a main effect in the model controls for this difference in endospore abundance. After controlling for differences in endospore abundance with these statistical terms, we could use the remaining interactions with host line to test for changes in specificity, which we describe further in the subsection below entitled "Change in host specificity."

| Variation in endospore abundance
We first evaluated variation in endospore abundance. Specifically, we compared endospore abundance in soils collected from the two crop treatments. We fit a generalized estimating equation ( A first-order autoregressive correlation structure between batches from the same plot was selected due to the longitudinal nature of the sampling (Wang & Carey, 2003;Ziegler & Vens, 2010). The results of Wald tests dictated the exclusion of insignificant interaction effects (Zuur, Ieno, Walker, Saveliev, & Smith, 2009).
The host clonal lines used in the bioassays were collected in 2008 from the peanut plots that we surveyed from 2013 to 2016.
This shared origin raised the possibility that endospores sampled from peanut plots had greater ability to attach to the host clonal lines than did endospores sampled from rotation plots. It is unlikely that parasite samples from interspersed peanut and rotation plots would strongly differ in adaptation to this small sample of host genotypes. Nonetheless, with the above analysis, we cannot conclusively attribute differences in attachment rates between treatments to differences in endospore abundance between treatment plots. To address this problem, we repeated the analysis using hosts from an independent lineage (GH), which was established in the early 1990s.
This host line does not share a coevolutionary history with parasites from either the peanut or rotation plots. Hence, we can attribute variation in attachment rate between treatments to variation in endospore abundance between treatments. We fit a GEE as described above with year (2013-2016), treatment (peanut, rotation), and their interaction as predictors of the probability of endospores attaching to a host.

| Change in host specificity
We then evaluated changes in the specificity of P. penetrans through time. Peanut and rotation plots were analyzed in two separate models. For each treatment, we fit a logistic model (generalized linear model-GLM) with year, clonal line (C3, C6, C8, C40), plot, and all possible interactions as predictors of the probability of endospores attaching to a host. The response variable was the attachment status (0, 1) of an individual host. For these two models, the predictors' year, plot, and their interaction controlled for variation in endospore abundance. After controlling for differences in endospore abundance, we tested for differences in specificity using two-way and three-way interactions of year, line, and plot. The interaction "line × plot" addressed spatial variation in attachment rate to host lines, across all years. A substantial line × plot interaction would indicate static differences between plots in specificity. The interaction "year × line" addressed temporal variation in attachment rate to host lines, shared across plots. A substantial year × line interaction would indicate change in specificity, with the temporal trajectory similar across plots. The interaction "year × line × plot" simultaneously addressed spatial and temporal variation in specificity. A significant three-way interaction would indicate change in specificity, with the temporal trajectories differing between plots.
In an effort to weigh the relative importance of model terms, we The models were otherwise as described for the prior analysis. In these eight models, the predictor year controlled for yearly variation in the abundance of endospores in soil sampled from the focal plot.
Controlling for differences in endospore abundance, we could then test for temporal change in specificity using the interaction term. A substantial interaction effect would indicate change in specificity.

| Variation in endospore abundance
Four of the eight study plots were continuously planted with peanut from 2013 to 2016. For the other four study plots, the crop alternated annually between peanut and soybean. We refer to these as the peanut and rotation treatments, respectively.
We first compared rates of endospore attachment between soils derived from plots subjected to these two crop treatments. A significantly higher fraction of tested hosts acquired endospores when exposed to soils from peanut plots (46.1% ± 3.0%) relative to rotation plots (26.6% ± 2.7%) (Figure 2, Table 1a). Repeating this analysis using a distinct host line (GH) gave the same result: A significantly higher fraction of GH hosts acquired endospores when exposed to soils from peanut plots (42.2% ± 4.2%) relative to rotation plots (21.3% ± 4.1%) (Table 1b). These results suggest the maintenance of a higher abundance (i.e. dose) of endospores in soils of peanut plots: Attachment rates were higher with soils from peanut plots, regardless of the host lines tested.
Mean attachment rates did not vary with year, indicating no overall variation in endospore abundance over time (Table 1, Figure 3).
Attachment rates also did not vary with host clonal line, indicating that, averaging over all samples collected from the field plots, P. penetrans did not attach better to one host line than another (Table 1a, Figure 3). There was, however, a significant interaction of year and line, indicating that sampled P. penetrans varied among years in attachment rate to specific host lines (Table 1a). This effect appeared to be driven by an overall decline in the attachment rate to C6 from 76.0% ± 8.0% in 2013 to 30.0% ± 4.8% in 2014 (GEE coefficient estimate = −1.60 ± 0.67, p = 0.016), 27.5% ± 5.5% in 2015 (−1.72 ± 0.77, p = 0.025), and 32.0% ± 6.2% in 2016 (−1.88 ± 0.72, p = 0.009).

| Change in host specificity
We tested the hypothesis that the specificity of P. penetrans changed in these plots. To do so, we evaluated variation in the attachment rate of P. penetrans against the four clonal M. arenaria lines. We used a statistical approach that enabled us to test for variation in host specificity by controlling for differences in the abundance of endospores in sampled plots (see Materials and Methods). Because of the substantial differences between treatments identified above, we analyzed peanut and rotation plots separately.
For both treatments, attachment rates varied with year, plot, and their interaction (Table 2). These effects likely reflect the expected spatial and temporal variation in the abundance of P. penetrans endospores in tested soils. Inclusion of these predictors in the models controlled for variation in endospore abundance, allowing us to then evaluate variation in host specificity of sampled parasites.
After controlling for variation in endospore abundance, we found evidence for change in host specificity in plots of both treatments (Table 2). For peanut plots, a three-way interaction of year, clonal line, and plot contributed strongly to variation in attachment rates (Table 2a). This interaction indicated that parasites varied through time F I G U R E 2 Endospore attachment rates by treatment. Endospore attachment rates were higher with soils collected from peanut plots vs. rotation plots. Attachment rate was calculated as the fraction of 25 tested hosts with endospores attached. Each boxplot represents 16 estimates of attachment rate (four clonal lines by four plots/treatment) in their ability to attach to the four host clonal lines and that these temporal trajectories varied significantly between the four peanut plots.  (Table 2a).
For rotation fields, the three-way interaction of year, clonal line, and plot also contributed to variation in attachment rates (Table 2b, Figure 4e-h). In these fields, however, the two-way interaction of year and line contributed slightly more (1.3-fold) to variation in attachment rates than did the three-way interaction.
This result suggests that the trajectories of host specificity were relatively similar across the four rotation plots. This finding suggests that temporal change in global, across-plot factors were important in driving variation between parasites sampled from rotation plots.
To complement the above analyses, we investigated change in host specificity for the parasites from each individual plot. For each of the eight plots, attachment rate varied substantially with the interaction of year and clonal line after controlling for yearly differences in endospore abundance (Supporting information Table S1).
This result further supports the above findings of change in specificity through time (Figure 4).

| D ISCUSS I ON
In this study, we tested for the evolution of host specificity of a parasite proposed as a biological control agent. After controlling for variation in parasite abundance, we detected yearly changes in the ability of Pasteuria penetrans to attach to a collection of clonal lines of its host, the nematode Meloidogyne arenaria (Figure 4, Table 2, Supporting information Table S1). This result is consistent with the original hypothesis: It demonstrates rapid change in specificity for host attachment, a phenotype that is fundamental to the ability of this parasite to infect its host. Temporal trajectories of host specificity differed substantially across individual plots sampled within a single field (Table 2, Figure 4). This result suggests that change in host specificity occurs at a local scale ( Figure 3 vs. 4). Below, we present hypotheses for the drivers of TA B L E 1 Variation in endospore attachment rate across crop rotation treatments Year × Line × Treatment 9 6.40 0.700 Notes. These tables present the results of generalized estimating equations with the number of hosts with and without attached endospores as a binomial response variable. The Wald chi-square statistics were obtained by sequentially adding each factor and comparing models with and without the factor of interest.
F I G U R E 3 Mean attachment rates through time. For each line, the graph shows the mean attachment rate across the eight plots surveyed each year. Attachment rates were estimated using 25 hosts for each line-by-plot-by-treatment combination. Error bars are 95% confidence intervals for the mean calculated using the function groupwiseMean in the R package rcompanion (Mangiafico, 2015) change in host specificity in P. penetrans and discuss the implications of changing specificity for the use of P. penetrans in biological control.
Before presenting hypotheses for the drivers of changing specificity, we would like to note a few limitations of this study. First, the assay of host specificity only used four clonal host lines. Assuming a genetic basis to attachment, the assay could only detect variation at those loci and alleles associated with attachment to the four tested host lines. The assay, therefore, was conservative, and variation in host specificity likely exceeds that shown here.
Second, the assay measured variation in attachment of P. penetrans to M. arenaria. Attachment does not ensure a successful infection (Davies et al., 1988;Oostendorp, Dickson, & Mitchell, 1991;Sayre & Wergin, 1977;Stirling, 1984). Infection cannot, however, proceed without attachment, so variation in specificity rests upon this first step in the infection process. Moreover, Duneau et al. (2011) primarily attributed variation in specificity of the parasite P. ramosa to variation in attachment to Daphnia hosts. Third, we cannot exclude the possibility that the observed changes in host specificity are not genetic and hence not evolutionary. We measured attachment rates of endospores directly following field collection.
This approach preserved the natural parasite diversity, but it did not allow for a period of propagation in the laboratory to remove any environmental effects on host specificity. Studies of P. ramosa demonstrate that variation in attachment rests largely upon the genetic interaction of host and parasite (Carius, Little, & Ebert, 2001;Duneau et al., 2011;Ebert et al., 2016;Luijckx, Fienberg, Duneau, & Ebert, 2013). The extent to which this is also true for P. penetrans remains to be determined, though prior studies are suggestive of a genetic basis (Davies et al., 1988Espanol et al., 1997;Stirling, 1985). Hence, we find evidence of change in specificity that is consistent with, but not equivalent to, the evolution of specificity.

| Hypothetical drivers of specificity change
We now propose and discuss three hypotheses to explain the rapid change in host specificity of P. penetrans. First, temporal and spatial variation in specificity may simply reflect genetic drift. Given the relatively limited dispersal distances of both P. penetrans (Oostendorp et al., 1990) and Meloidogyne spp. (Prot & Netscher, 1979), patterns of random variation in allele frequencies could differ across, and even within, plots. Current data cannot reject a contribution of drift.
There is, however, good reason to suspect that P. penetrans populations experience strong selection on their ability to attach to local hosts: Development of P. penetrans cannot begin without attachment (Sayre & Wergin, 1977).
Second, abiotic factors may drive the evolution of host specificity in P. penetrans. We predict that some abiotic variables would have similar trajectories of temporal change across the whole field (e.g., seasonal variation in temperature and precipitation). These abiotic factors should drive parallel trajectories of specificity change across plots within the field. We, however, observed substantial variation in evolutionary trajectories among plots (Figure 4). To explain this localized variation, changes in specificity must be driven by abiotic factors that vary locally, between plots. Evaluation of this hypothesis requires knowledge of the abiotic variables that might impact specificity of P. penetrans.
Third, the change in host specificity of P. penetrans may reflect negative frequency-dependent selection. According to this hypothesis, parasite populations adapt to infect the most common genotypes in the local host population. Burdened by infection, these common host genotypes decline in frequency. As new host genotypes increase in frequency, selection on the parasite population shifts (Bell, 1982;Hutson & Law, 1981;Jaenike, 1978 Notes. These tables present the results of generalized linear models with the attachment status of an individual host (endospores attached or not) as a binomial response variable. The same model was separately fit to data from peanut plots and rotation plots. For each factor, we show the results of likelihood ratio tests of models with and without the factor. D is the deviance accounted for by each factor. R 2 L reflects the explanatory power of the model. adapt to one another (Engelstadter & Bonhoeffer, 2009;Hamilton, 1980;Hamilton, Axelrod, & Tanese, 1990;Parker, 1994). Prior studies suggest that the interaction of P. penetrans and Meloidogyne spp. meets these requirements. G × G interactions contribute to attachment (Davies et al., 1988Espanol et al., 1997;Stirling, 1985).
Host populations can rapidly evolve resistance (Channer & Gowen, 1992): In Tzortzakakis et al. (1996), M. javanica evolved resistance to a population of P. penetrans after four generations of selection. Data on adaptation of P. penetrans to new host genotypes are limited.
Nonetheless, Channer and Gowen (1992) suggest that P. penetrans populations can evolve increased attachment to a novel host genotype and reduced attachment to the ancestral host genotype within a single growing season.
Negative frequency-dependent selection should manifest as rapid and continual temporal change in host specificity of P. penetrans. Moreover, trajectories of change should be highly localized in space, such that sites diverge from one another at the scale of host dispersal. Finally, no genotype of P. penetrans should be universally more fit; rather, parasites from different plots should differ in the genotype that is most fit at any given point in time. The results of this study are consistent with these patterns: We detected rapid and continual change in host specificity, with trajectories that diverged across sampled plots (Figure 4, Table 2). The results also suggest that no genotype of P. penetrans was universally more fit: Averaged across fields, attachment rates were approximately equivalent across clonal host lines (Figure 3, Table 1).

F I G U R E 4 Temporal change in
Pasteuria penetrans specificity at the local level. The fraction of hosts with attached endospores is shown for each of the four peanut (a-d) and rotation (e-h) plots. The relative attachment rates to host lines, and even their rank order, varied between years in the same plot and between plots in the same year, consistent with spatial and temporal variation in host specificity. Each data point is estimated from 25 hosts. Error bars are 95% confidence intervals for the proportion calculated using the function binom.test in R

| Implications for biological control strategies
Regardless of the driving force, what are the implications of these results for the use of P. penetrans as a biological control agent? Clearly, host specificity of P. penetrans can change rapidly, as evidenced by substantial variation in specificity in space and time. If the genetic composition of Meloidogyne populations varies in space and time, and there is a genetic basis to infection, we expect significant variation in the efficacy of different P. penetrans genotypes in controlling a Meloidogyne population.
Applying a mixture of P. penetrans genotypes may overcome this problem: Channer and Gowen (1992) showed that single isolates of P. penetrans varied substantially in their ability to attach to different host lineages, while mixtures of four to five parasites isolates showed little variation between host lineages (see also Tzortzakakis & Gowen, 1994). If negative frequency-dependent selection drives change in specificity in this system, we expect that a population of P. penetrans could adapt to infect the most common host genotypes present in a field, given sufficient genetic diversity and a lag time. Indeed, the literature commonly refers to a lag time in the efficacy of P. penetrans, with substantial suppression of Meloidogyne evident a few years after first application to the field (Chen & Dickson, 1998;Chen, Dickson, McSorley, Mitchell, & Hewlett, 1996;Oostendorp et al., 1991;Timper, 2009). This lag time may reflect adaptation of the P. penetrans population to local hosts. An alternate, though not mutually exclusive, explanation is that the lag time reflects the build-up of endospores to suppressive densities (reviewed in Hufbauer & Roderick, 2005).
Comparison of the crop rotation treatments suggests a conflict between different pest management strategies. Crop rotation reduced attachment rates by nearly 50% relative to continuous planting ( Figure 2). The substantial treatment effect, independent of assayed host line, indicates differences in parasite abundance between treatments rather than differences in specificity (Table 1). Though both peanut and soybean are good hosts to M. arenaria, nematode reproduction is tenfold greater on peanut than soybean (Noe, 1991). Hence, rotation with soybean can suppress densities of M. arenaria (Rodríguez-Kábana, Robertson, Backman, & Ivey, 1988), in turn suppressing densities of P. penetrans. Temporal trajectories of P. penetrans specificity were also less divergent across rotation plots than across continuous peanut plots. One possible explanation for this is that, for rotation plots, there was simply less variation in attachment rates to explain because of the low attachment rates in the bioassay. These observations raise an interesting problem for pest management: Interventions, like crop rotation, that reduce the density of Meloidogyne hosts may also suppress P. penetrans densities, reducing its efficacy in biological control (Madulu, Trudgill, & Phillips, 1994;Timper, 2009;Timper et al., 2001). This finding demonstrates a potential cost of using specific biological control agents that cannot maintain their population sizes on alternative hosts (Fagan et al., 2002).

| Conclusion
We have demonstrated rapid change in host specificity in a biological control system. Above, we hypothesize that this change in host specificity reflects adaptation to locally common host genotypes, consistent with host-parasite coevolution. This hypothesis remains to be tested. It predicts that, with sufficient genetic variation, the biological control agent should improve following its initial application, becoming more effective at limiting the population density of its host.

ACK N OWLED G EM ENTS
We thank David Clements and Susan Drawdy for technical assistance. C. Liu was supported by Syngenta. A.K. Gibson was supported by the NIH IRACDA program Fellowships in Research and Science Teaching (FIRST) at Emory University (K12GM000680).

CO N FLI C T O F I NTE R E S T
None Declared.

DATA ACCE SS I B I LIT Y
Data and analysis scripts for this study are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.0pv6570