Turnover in local parasite populations temporarily favors host outcrossing over self‐fertilization during experimental evolution

Abstract The ubiquity of outcrossing in plants and animals is difficult to explain given its costs relative to self‐fertilization. Despite these costs, exposure to changing environmental conditions can temporarily favor outcrossing over selfing. Therefore, recurring episodes of environmental change are predicted to favor the maintenance of outcrossing. Studies of host–parasite coevolution have provided strong support for this hypothesis. However, it is unclear whether multiple exposures to novel parasite genotypes in the absence of coevolution are sufficient to favor outcrossing. Using the nematode Caenorhabditis elegans and the bacterial parasite Serratia marcescens, we studied host responses to parasite turnover. We passaged several replicates of a host population that was well‐adapted to the S. marcescens strain Sm2170 with either Sm2170 or one of three novel S. marcescens strains, each derived from Sm2170, for 18 generations. We found that hosts exposed to novel parasites maintained higher outcrossing rates than hosts exposed to Sm2170. Nonetheless, host outcrossing rates declined over time against all but the most virulent novel parasite strain. Hosts exposed to the most virulent novel strain exhibited increased outcrossing rates for approximately 12 generations, but did not maintain elevated levels of outcrossing throughout the experiment. Thus, parasite turnover can transiently increase host outcrossing. These results suggest that recurring episodes of parasite turnover have the potential to favor the maintenance of host outcrossing. However, such maintenance may require frequent exposure to novel virulent parasites, rapid rates of parasite turnover, and substantial host gene flow.

Parasites are predicted to incur selection to infect the most common host genotypes, which can impose negative frequency-dependent selection on host populations (Koskella & Lively, 2009). Outcrossing hosts can produce genetically diverse offspring by exchanging genetic material across lineages and recombining beneficial mutations with different origins into novel or rare genotypes. Conversely, selffertilizing hosts are more likely to produce offspring with common genotypes that are predicted to suffer disproportionately from parasite infection. The fitness advantage that outcrossed offspring gain from reduced parasitism can fluctuate over time; if this advantage periodically outweighs the costs of sex, then outcrossing will be maintained in the long term (Vergara, Jokela, & Lively, 2014). Many studies of natural and experimental host-parasite coevolution have provided support for the Red Queen hypothesis (Dybdahl & Lively, 1998;Jokela, Dybdahl, & Lively, 2009;King, Delph, Jokela, & Lively, 2009;Koskella & Lively, 2009;Lively & Dybdahl, 2000;Masri et al., 2013;Morran, Schmidt, Gelarden, Parrish, & Lively, 2011;Slowinski et al., 2016;Vergara et al., 2014).
Although the Red Queen hypothesis specifically invokes negative frequency-dependent selection driven by host-parasite coevolution as a mechanism that can favor the long-term maintenance of outcrossing, other sources of recurring environmental change could have similar effects. If adaptive landscapes are frequently shifting over time, outcrossing lineages may gain substantial fitness advantages by producing genetically diverse offspring and assembling beneficial mutations from multiple lineages into novel genotypes. Theory predicts that natural selection will favor increased outcrossing and recombination if the sign of epistasis for fitness changes every two to five generations, such that genotypes with high fitness in any given generation become unfit a few generations later (Barton, 1995;Gandon & Otto, 2007;Peters & Lively, 1999). Conversely, populations evolving under relatively consistent environmental conditions will reach an adaptive peak, after which outcrossing is likely to break up adaptive gene complexes and epistatic relationships. This may lead to outbreeding depression and the re-emergence of self-fertilization or asexual reproduction (Lynch, 1991;Lynch & Deng, 1994). Therefore, it is critical to evaluate whether changing environmental conditions can produce sufficient shifts in adaptive landscapes to favor outcrossing and ultimately whether such environmental changes occur often enough to explain the maintenance of outcrossing in nature.
When spatial heterogeneity in selective pressures is coupled with frequent migration between environments, interactions between epistasis and selection will determine whether sex and recombination are favored. Sex and recombination can break down linkage between beneficial and deleterious alleles (Hill-Robertson interference) and promote the incorporation of incoming beneficial mutations into genotypes that confer high levels of fitness, but they may also break apart locally adapted genotypes (Hill & Robertson, 1966). These opposing effects may combine to favor sex and recombination if offspring from local × migrant crosses have higher average fitness than offspring from local × local and migrant × migrant crosses (Agrawal, 2009;Otto, 2009 (Anderson, Morran, & Phillips, 2010). Caenorhabditis elegans populations consist of males and hermaphrodites. Hermaphrodites cannot mate with each other but can self-fertilize or outcross with males (Brenner, 1974). Mutations at the mating system loci xol-1 and fog-2 can be exploited to generate obligately selfing and obligately outcrossing populations (Miller, Plenefisch, Casson, & Meyer, 1988;Schedl & Kimble, 1988). Morran, Parmenter, and Phillips (2009) exposed nematode populations to two different selection environments during 40-to 50-generation evolution experiments: (1) a chemical mutagen coupled with a migration barrier and (2) the pathogenic bacterium Serratia marcescens. Obligately outcrossing populations showed stronger adaptation to the challenging environmental conditions than wild-type and obligately selfing populations. Wild-type populations evolved higher levels of outcrossing and exhibited stronger adaptation than obligately selfing populations. Morran et al. (2011) conducted a 30-generation evolution experiment in which nematode populations were exposed to either a fixed strain of S. marcescens or a potentially coevolving S. marcescens population that was isolated from nematode carcasses every generation. Both parasite treatments led to significant increases in host outcrossing rates over the first eight generations, but only coevolving parasites selected for the maintenance of elevated outcrossing rates throughout the experiment. Slowinski et al. (2016) expanded on these results by showing that wild-type C. elegans hermaphrodites (which are capable of self-fertilization) rapidly invaded mutant obligate-outcrossing populations when the mixed populations were exposed to avirulent or fixed parasite genotypes over 33 generations, but self-fertilization did not invade C. elegans populations that were exposed to potentially coevolving parasites. Masri et al. (2013) did not find elevated host outcrossing rates during 48 generations of coevolution between C. elegans and the pathogenic bacterium Penley & Morran, 2017), we established an initial outcrossing rate of ~0.5 by manipulating the ratio of hermaphrodites to males. We made five replicates of this ancestral host population and passaged each of them with the S. marcescens strains CoSm, ES1, Rec320, and Sm2170 for 18 generations. The novel strains (CoSm, ES1, and Rec320) were derived from Sm2170 and experienced different selective pressures during previous evolution experiments: ES1 was selected to cause higher host mortality, Rec320 was selected to cause nonlethal infections, and CoSm was passaged apart from the host (Gibson et al., 2015;Morran et al., 2011). Therefore, we expected our parasite strains to have different initial levels of virulence, resulting in different host evolutionary trajectories and perhaps different outcrossing rates. We measured outcrossing rates every six generations to test whether higher levels of outcrossing would be favored as host populations adapted to novel parasite strains. At the end of our 18-generation evolution experiment, we compared competitive fitness in the presence of parasites between ancestral hosts and evolved (generation 18) hosts to test whether novel parasite strains would induce faster rates of host adaptation. We also tested whether host resistance to the original parasite strain (Sm2170) would decrease as host populations adapted to the novel parasite strains, which would indicate crossresistance trade-offs.

| Study system
Caenorhabditis elegans is a free-living nematode that colonizes ephemeral bacterial blooms in rotting fruits and herbaceous stems.
Conditions such as extreme temperatures, scarce food, and high population density cause C. elegans to enter a nonfeeding life stage known as dauer, in which larvae are resistant to environmental stresses and starvation. Dauer larvae actively seek invertebrate vectors for long-distance dispersal to new bacterial blooms through nictation behavior, in which they stand on their tails and wave their heads (Cutter, 2015;Felix & Braendle, 2010;Frezal & Felix, 2015).
Caenorhabditis elegans has an androdioecious mating system with males and self-fertilizing hermaphrodites. The hermaphrodites cannot outcross with each other but may outcross with males (Brenner, 1974). All known natural strains predominantly self-fertilize, although natural outcrossing rates are variable (Teotonio, Manoel, & Phillips, 2006). Given the characteristics of its mating system and its need for frequent migrations that may result in exposure to new parasites, C. elegans seems to be an appropriate model for studying how parasite turnover affects host outcrossing rates and adaptive potential.
For our parasite, we used S. marcescens, a virulent bacterium that infects many plant and animal species (Grimont & Grimont, 1978).

| Host and parasite populations
Caenorhabditis elegans stock populations were maintained at 20°C in 10-cm-diameter Petri dishes filled with 30 ml of autoclaved nematode growth medium lite (NGM) (US Biological, Swampscott, MA, USA). These dishes were seeded with 200 μl of Escherichia coli strain OP50 culture that was grown overnight at 28°C in Luria-Bertani broth (LB). After the E. coli lawns grew overnight at 28°C, the dishes were stored at 4°C for future use. Caenorhabditis elegans stock strains were derived from single wild-caught individuals. We obtained the wild-type strain CB4856 (from Hawaii, USA) and the GFP-marked strain JK2735 from the Caenorhabditis Genetics Center (University of Minnesota, Minneapolis, MN, USA).
Our ancestral C. elegans host population, EW2-30, was derived from PX382, a systematically inbred variant of CB4856 (Morran et al., 2009). EW2-30 resulted from a previous evolution experiment in which hosts were passaged with a nonevolving S. marcescens population (strain Sm2170) for 30 generations; the complete protocol is published in Morran et al. (2011). Briefly, a population of PX382 (the ancestral population of EW2-30) was mutagenized with ethyl methanesulfonate and passaged on Serratia selection plates (SSPs), which required nematodes to migrate through live S. marcescens and ampicillin to reach their food source, E. coli strain OP50. Under these conditions, naïve host populations can suffer mortality rates up to 80% (Morran et al., 2009;Schulenburg & Ewbank, 2004). Only the offspring of nematodes that reached the food source proceeded to a new SSP to begin the next generation. At the end of the 30-generation evolution experiment, EW2-30 was frozen at −80°C for future use.
Our novel S. marcescens strains (ES1, Rec320, and CoSm) were derived from Sm2170 during previous evolution experiments. ES1 underwent selection for increased infectivity and virulence as it was passaged with a static CB4856 host population for 30 generations.
Bacteria that killed nematodes after 24 hr of exposure were harvested every generation and used to infect the next generation of hosts; see Morran et al. (2011) for further details. Host and parasite populations can be copassaged to allow for coevolution; in this case, parasites are harvested from dead nematodes every generation and used to infect the offspring of surviving nematodes. Selection for reduced antagonism is also possible; a previous study identified hosts carrying mild upper intestine infections that were not cleared but did not cause death or prevent reproduction. During a 20-generation evolution experiment, Gibson et al. (2015) copassaged offspring from infected parents with bacteria that caused those mild infections, resulting in the Rec320 parasite strain. As an alternative means of generating a less-virulent parasite strain, CoSm was passaged for 20 generations on SSPs without nematodes; see Gibson et al. (2015) for further details regarding CoSm and Rec320.

| Experimental evolution
Before starting our evolution experiment, we manipulated our ancestral host population (EW2-30) to establish initial male frequencies of ~0.25 in each of the experimental populations. Groups of 20 L4 nematodes were transferred to NGM dishes seeded with OP50 and allowed to produce offspring; five dishes had a 1:1 ratio of hermaphrodites to males and five dishes had all hermaphrodites.
Matings between hermaphrodites and males result in 50% male offspring, whereas selfing hermaphrodites produce ~0.02% male offspring, a frequency that approximates the rate of spontaneous X chromosome nondisjunction (Anderson et al., 2010). Therefore, mixing equal quantities of offspring from those 10 dishes resulted in a population with ~25% males. We created five replicates from this mix, transferring ~1,000 offspring to each replicate population, and passaged each of them for 18 generations on SSPs with four parasite strain treatments: CoSm, ES1, Rec320, and Sm2170. SSP construction and nematode transfers were performed using published protocols (Morran et al., 2011). Briefly, groups of ~1,000 L3-L4 nematodes were washed into M9 buffer and transferred to live S. marcescens lawns. Only the offspring of nematodes that successfully migrated through the parasite and a streak of ampicillin to reach their food source (E. coli strain OP50) were transferred to a new SSP to begin the next generation. Each of the five replicate ancestral populations was separately passaged with CoSm, ES1, Rec320, and Sm2170 for 18 generations, resulting in 20 evolved host populations. The replicate ancestral populations were frozen at −80°C at the beginning of the experiment and the experimentally evolved host populations were frozen at −80°C every six generations so they could be tested simultaneously in later assays.

| Host mortality rate assays
Host mortality rates were assayed in 10-cm-diameter Petri dishes filled with 30 ml of NGM and seeded with 200 μl of S. marcescens culture that was grown overnight at 28°C in LB. The S. marcescens lawns grew overnight at 28°C, then groups of 200 L4 nematodes were transferred into the dishes. Dead nematodes were counted after 24 hr of parasite exposure, and 24-hr host mortality rates were calculated as the number of dead nematodes divided by 200 transferred nematodes. Serratia marcescens strain Sm2170 is capable of establishing systemic infections and killing C. elegans within 24 hr (Morran et al., 2011), and host mortality is most accurately measured at this time point because the parasite disintegrates nematode carcasses after 24 hr. Before starting our evolution experiment, we used these assays to compare the virulence levels of our four S. marcescens strains (CoSm, ES1, Rec320, and Sm2170) toward the ancestral host population. For each of the five replicate ancestral host populations, we performed three technical replicates per treatment. The effects of parasite strain on 24-hr mortality rates were analyzed using a generalized linear model (GLM) with quasi-binomial error distribution and logit link function, and then, pairwise differences were assessed using Tukey's honest significant difference (HSD) tests. At the end of our evolution experiment, we used these assays to compare resistance to Sm2170 (to which our ancestral hosts were well-adapted) across replicate host populations that had been passaged with each of the four parasite strains for 18 generations. Each experimental evolution treatment had five replicate host populations, and we performed two technical replicates per population. The effects of host evolution treatment on 24-hr mortality rates were analyzed using a GLM with quasi-binomial error distribution and logit link function. These analyses were performed in R version 3.2.3 (R Development Core Team, 2014) using the multcomp package (Hothorn, Bretz, & Westfall, 2008). ANOVA tables are provided in Appendix S1.

| Measuring host outcrossing rates
Male frequencies were measured in each host population at the beginning of the experiment and after every six generations of experimental evolution. A transect of each experimental population was counted on the OP50 portion of the SSP every six generations. Approximately 200 L4 offspring were counted and sexed prior to passage to the next round of selection. The outcrossing rate for each host population was calculated by multiplying the male frequency by two after correcting for the number of males that typically result from spontaneous X chromosome nondisjunction (Stewart & Phillips, 2002). These data are presented in terms of outcrossing rates in Results but were analyzed in terms of male frequencies to enable the use of binomial GLMs. The effects of parasite strain, host generation, and their interaction on male frequencies in our host populations were analyzed using a GLM with quasi-binomial error distribution and logit link function.
We analyzed the effects of parasite strain on male frequencies at host generations 6, 12, and 18 using quasi-binomial GLMs and then performed pairwise comparisons using Tukey's HSD tests.
For each parasite strain treatment, we analyzed the effects of host generation on male frequencies using quasi-binomial GLMs.
We then performed pairwise comparisons using Tukey's HSD tests and tabulated the key differences: generation 6-ancestor, generation 12-ancestor, generation 18-ancestor, and generation 18-generation 12. These analyses were performed in R version 3.2.3 (R Development Core Team, 2014) using the multcomp package (Hothorn et al., 2008). ANOVA tables are provided in Appendix S1.

| Male versus hermaphrodite susceptibility to parasites
We compared the survival rates of males and hermaphrodites from the ancestral host population when they were exposed to parasites within our selection regime. SSPs were prepared using

| Competitive fitness assays
We measured the competitive fitness of each ancestral and generation 18 host population relative to a common tester strain in our se- Therefore, we used a nonparametric Kruskal-Wallis H test to assess the effects of host evolution treatment (the parasite strain each host population was passaged with) on percent changes in mean fitness.
Pairwise differences were assessed using Steel-Dwass tests. These analyses were performed in JMP 12.0 (SAS Institute, Cary, NC, USA).

| RE SULTS
Before starting our evolution experiment, we conducted host mortality rate assays to measure the virulence of our chosen S. marcescens strains toward the ancestral C. elegans population, which had previously adapted to Sm2170 during a 30-generation evolution experiment (Morran et al., 2011;. Our chosen parasite strains varied significantly in virulence toward the ancestral host population (Figure 1; F 3,16 = 17.6, p < .0001); ES1 caused the greatest host mortality, Sm2170 caused intermediate host mortality, and CoSm and Rec320 caused the lowest host mortality (Tukey's HSD tests, p < .03). These results match our predictions based on the parasite strains' evolutionary histories: Sm2170 was the ancestor, ES1 was selected to cause higher host mortality, Rec320 was selected to F I G U R E 1 Mortality rates suffered by the ancestral Caenorhabditis elegans population after 24 hr of exposure to the four Serratia marcescens strains we used for experimental evolution (±1 SEM). Different letters indicate significant differences between parasite strains (Tukey's honest significant difference tests, p < .03).  Table 1). In host populations passaged with ES1, outcrossing rates were significantly higher than ancestral levels at generations 6 and 12, and then decreased back to ancestral levels by generation 18 (Figure 2; Table 2). In the CoSm and Rec320 treatments, host outcrossing rates were not significantly different between generation 0 and generation 6, 12, or 18. In the Sm2170 treatment, host outcrossing rates showed a marginally significant decline from generation 0 to generation 6 (p = .054) and were significantly lower at generations 12 and 18. Outcrossing rates decreased significantly between generations 12 and 18 in the ES1 and Rec320 treatments ( TA B L E 1 Significance groups that resulted from calculating all pairwise differences in host outcrossing rates between parasite strain treatments ( Figure 2) at generations 6, 12, and 18 (Tukey's honest significant difference tests, p < .015)

TA B L E 2
The following contrasts in host outcrossing rates within each parasite strain treatment ( Figure 2) are presented: generation 6-ancestor, generation 12-ancestor, generation 18-ancestor, and generation 18generation 12. We calculated outcrossing rates in our host populations over time based on observed male frequencies. However, these measurements could be biased if males and hermaphrodites differ in susceptibility to any of the parasite strains used for experimental evolution.
To address this possible concern, we compared the survival rates of males and hermaphrodites in the ancestral population within our selection regime. Nematodes that migrated out of the S. marcescens side of SSPs and were alive 48 hr after exposure to the parasite were counted as survivors. There was no significant difference in survival rate between males and hermaphrodites across the four parasite strains (Table 3; F 1,38 = 1.41, p = .24). When the parasite strains were considered individually, male survival was significantly lower than hermaphrodite survival in the ES1 treatment (F 1,8 = 6.52, p = .034), but no other significant differences were found (F 1,8 < 2.2, p > .17).
Therefore, the elevated outcrossing rates (and male frequencies) observed at generations 6 and 12 in the ES1 experimental evolution treatment ( Figure 2; Table 2) occurred in spite of higher ancestral male susceptibility to this parasite strain ( Table 3), suggesting that the adaptive benefits of outcrossing were sufficient to overcome this initial barrier.
We evaluated the degree of host adaptation to each parasite strain over 18 generations of repeated exposure by assaying the competitive fitness of ancestral and generation 18 host populations.
In each assay, equal numbers of nematodes from the focal host population and a GFP-marked tester strain were mixed and exposed to the parasite strain the focal host was passaged with, using the same selective environment. After 4 days of exposure, we calculated the frequency of GFP-marked offspring; frequencies below the starting level of 50% indicated that the focal hosts had greater competitive fitness. The percent change in mean fitness for generation 18 hosts relative to their ancestors differed significantly across host evolution treatments ( Figure 3; χ 2 3 = 13.7, p = .003). Specifically, host populations that were passaged with ES1 exhibited the greatest rates of adaptation (Steel-Dwass tests, p < .03), but there were no significant pairwise differences between the CoSm, Rec320, and Sm2170 treatments (p > .8).
We conducted host mortality rate assays to compare susceptibility to Sm2170 across our generation 18 host populations. Because the ancestral host population had previously adapted to Sm2170, decreased resistance against Sm2170 in host populations that were passaged with novel strains would suggest that resistance to different S. marcescens strains is subject to trade-offs. However, there were no significant differences among generation 18 host populations in mortality rate following exposure to Sm2170 (Figure 4; F 3,16 = 0.83, p = .50), suggesting a lack of cross-resistance trade-offs.

| D ISCUSS I ON
To study the effects of parasite turnover on host adaptation and outcrossing rates, we passaged replicate C. elegans populations that had previously adapted to the S. marcescens strain Sm2170 with three novel parasite strains and Sm2170 for 18 generations. Our novel strains were derived from Sm2170 during previous evolution experiments: ES1 was selected to cause higher host mortality, Rec320 was selected to cause nonlethal infections, and CoSm was passaged outside of the host (Gibson et al., 2015;Morran et al., 2011). Their levels of virulence toward the ancestral host population aligned well with their evolutionary histories; ES1 caused the greatest host mortality, Sm2170 caused F I G U R E 3 Percent changes in mean fitness for generation 18 hosts relative to their ancestors during exposure to the parasite strain each host was passaged with, as determined by competitive fitness assays against a GFP-marked tester strain (±1 SEM). Different letters indicate significant differences between host evolution treatments (Steel-Dwass tests, p < .03). For ancestral hosts, N = 3-5 replicates per parasite strain; for generation 18 hosts, N = 5 replicate populations per treatment with two to four technical replicates per population  Table 1) and their outcrossing rates increased significantly over the first 12 generations, an effect that was not observed in any other treatment (Table 2).
ES1 was more virulent toward ancestral males than ancestral hermaphrodites within our selection regime (Table 3), suggesting that ancestral males surmounted an initial barrier before increasing in frequency in the ES1 evolution treatment. Conversely, in host populations passaged with Sm2170, outcrossing rates were lowest throughout the experiment (Table 1) and significantly below ancestral levels at generations 12 and 18 ( Host populations passaged with the most virulent novel parasite strain, ES1, exhibited not only the highest outcrossing rates throughout the experiment (Figure 2), but also the strongest adaptation to parasites ( Figure 3). However, links between parasite virulence, host outcrossing rates, and host adaptation were not clear across all treatments. Although host populations in the CoSm and Rec320 treatments generally maintained higher outcrossing rates than those in the Sm2170 treatment ( Figure 2; Table 1), rates of host adaptation were not significantly different between these treatments ( Figure 3). Furthermore, hosts did not lose resistance to Sm2170 as they adapted to the novel strains, suggesting that trade-offs in resistance to different parasite strains did not occur (Figure 4). There are at least two possible explanations for this lack of cross-resistance trade-offs: (1) evolved resistance to the novel strains is independent of ancestral resistance to Sm2170 and (2) evolved resistance to the novel strains builds off of ancestral resistance to Sm2170. However, our experiments were not designed to test for mechanisms of resistance or virulence. These results may also be explained by the connected evolutionary histories of our host and parasite strains. Our ancestral hosts adapted to Sm2170 during a previous 30-generation experiment (Morran et al., 2011;. Our novel parasite strains were each derived from Sm2170; ES1 was selected for increased virulence over 30 generations (Morran et al., 2011), whereas CoSm and Rec320 were selected for lower virulence over 20 generations (Gibson et al., 2015). Perhaps parasite turnover events are more likely to produce the expected patterns of elevated host outcrossing rates, rapid host adaptation, and cross-resistance trade-offs when the parasites involved are more diverse and more virulent than those used in our study.
The patterns we observed in host outcrossing rates over time broadly agree with theoretical predictions and previous empirical results. Although outcrossing can increase the efficacy of selection during adaptation to changing environmental conditions (Agrawal, 2009;Otto, 2009), it is likely to be disfavored after a population reaches an adaptive peak because recombination and segregation will disassemble adaptive genotypes (Lynch, 1991;Lynch & Deng, 1994). Previous studies in which C. elegans populations were passaged with nonevolving S. marcescens found that outcrossing rates increased initially but decreased back to control levels by the end of the experiment; peak outcrossing rates were observed at generation 20 of 40 (Morran et al., 2009) and generation 8 of 30 (Morran et al., 2011). Furthermore, Slowinski et al. (2016) found that a wildtype C. elegans lineage capable of self-fertilization rapidly invaded mutant obligate-outcrossing populations over 33 generations of exposure to both avirulent and fixed parasite genotypes, but could not invade host populations that were passaged with potentially coevolving parasites. Similar results have also been found in the absence of parasites; populations of facultatively sexual rotifers produced more sexual offspring during the initial stages of adaptation to new environments, but asexual offspring were favored after the populations reached new fitness plateaus (Becks & Agrawal, 2012). Therefore, it seems highly unlikely that single episodes of environmental change can favor the long-term maintenance of high levels of outcrossing in species that normally self-fertilize or reproduce asexually.
Host-parasite coevolution can generate the constantly shifting adaptive landscapes that are likely necessary for the longterm maintenance of outcrossing, as shown in natural populations Vergara et al., 2014) and experimental systems (Masri et al., 2013;Morran et al., 2011). While other mechanisms may be capable of producing similar dynamics, it remains to be determined whether selective pressures apart from host-parasite coevolution can drive the long-term maintenance of outcrossing.
Here, we found short-term adaptive increases in host outcrossing rates after the novel, and virulent parasite strain ES1 was introduced into replicate host populations. In the ancestral host population, males were more susceptible to ES1 than hermaphrodites (Table 3), yet males overcame this initial disadvantage to ultimately increase in frequency ( Figure 2) and help the evolved populations achieve higher competitive fitness (Figure 3). Our results contrast slightly with those of Masri et al. (2013) (Barton, 1995;Gandon & Otto, 2007;Peters & Lively, 1999).
One potential limitation to the long-term maintenance of host outcrossing by frequent parasite turnover is that persistent directional selection imposed by parasites will likely deplete additive genetic variation in the host population. In an experimental evolu-  Stewart and Phillips (2002) can both be described as "low-diversity" (Sivasundar & Hey, 2003), their origins and observed outcrossing rate dynamics were very different. Taken together, these results suggest that both the level and the nature of standing genetic variation in a population will affect outcrossing rate dynamics, depending on whether a population is experiencing inbreeding depression or outbreeding depression. The ancestral host population used in our study was chemically mutagenized before being passaged with the S. marcescens strain Sm2170 for 30 generations (Morran et al., 2011). Our results indicate that the ancestral host population possessed sufficient additive genetic variation to respond to selection pressure from a novel, virulent parasite over 18 generations and that high levels of outcrossing facilitated this response (Figures 2 and 3). However, for host outcrossing to be maintained in the long term, recurring parasite turnover may need to be coupled with periodic replenishment of additive genetic variation in the host population.
Host migration is a potential mechanism for mitigating the loss of additive genetic variation in host populations as they adapt to environmental changes, such as parasite turnover. Migration between different host populations can introduce novel genetic variation. Further, theory predicts that outcrossing may be favored in the context of frequent migration because sex can break down unfavorable genetic associations that are introduced from outside populations (Agrawal, 2009;Lenormand & Otto, 2000). An experimental evolution study in which sexual and asexual yeast populations were subjected to varying levels of migration between two different selective environments showed that only the sexual populations that experienced migration adapted to both environments (Gray & Goddard, 2012). Therefore, frequent host migration coupled with frequent environmental change may provide the most likely context for the maintenance of outcrossing, apart from host-parasite coevolution. Our results suggest that environmental change resulting from parasite turnover can favor host outcrossing in the short term, but further work is required to determine whether recurring episodes of host migration and parasite turnover can contribute to the long-term maintenance of outcrossing.
The natural context for interactions between these mechanisms also requires further investigation, because any explanation for the maintenance of outcrossing must ultimately account for the distribution of outcrossing in nature.
Host-parasite coevolution remains the most well-supported mechanism for generating constantly shifting adaptive landscapes that may favor outcrossing over selfing despite the inherent costs of sex . However, our results suggest that periodic parasite turnover events could have similar effects, if only temporarily, in the absence of host-parasite coevolution. A significant challenge for future studies is to determine whether selection imposed by coevolving parasites is uniquely able to maintain host outcrossing or whether coevolving parasites represent one of multiple forms of parasite turnover that can favor the long-term maintenance of host outcrossing.

ACK N OWLED G M ENTS
We thank C.W. Beck, B.J. Brosi, J.C. de Roode, N.M. Gerardo, and two anonymous reviewers for comments that improved the manuscript. We thank S. Hesse and G. Ha for logistical support and the Caenorhabditis Genetics Center for providing nematode strains.
Z.R.L. was supported by an NSF Graduate Research Fellowship and an Emory University Graduate Diversity Fellowship.

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