Rapid evolution of Medicago polymorpha during invasion shifts interactions with the soybean looper

Abstract The Enemy Release Hypothesis posits that invasion of novel habitats can be facilitated by the absence of coevolved herbivores. However, a new environment and interactions with unfamiliar herbivores may impose selection on invading plants for traits that reduce their attractiveness to herbivores or for enhanced defenses compared to native host plants, leading to a pattern similar to enemy release but driven by evolutionary change rather than ecological differences. The Shifting Defense Hypothesis posits that plants in novel habitats will shift from specialized defense mechanisms to defense mechanisms effective against generalist herbivores in the new range. We tested these ideas by comparing herbivore preference and performance of native (Eurasia)‐ and invasive (New World)‐range Medicago polymorpha, using a generalist herbivore, the soybean looper, that co‐occurs with M. polymorpha in its New World invaded range. We found that soybean loopers varied in preference and performance depending on host genotype and that overall the herbivore preferred to consume plant genotypes from naïve populations from Eurasia. This potentially suggests that range expansion of M. polymorpha into the New World has led to rapid evolution of a variety of traits that have helped multiple populations become established, including those that may allow invasive populations to resist herbivory. Thus, enemy release in a novel range can occur through rapid evolution by the plant during invasion, as predicted by the Shifting Defense Hypothesis, rather than via historical divergence.

after it has evolved in an unfamiliar environment, compared to its native habitat. These traits can influence the outcome of interactions with local enemies and may be key to explaining why we see such biogeographic variation in plant performance (Parker & Gilbert, 2007).
Classical theories relating species invasion to host-enemy interactions assume historical coevolutionary processes that drive strictly ecological processes. According to Ehrlich and Raven (1964), coevolutionary studies provide a starting point for understanding community evolution and ecology, especially as these interactions are likely to have played a key role in species diversity in plants and insects.
Several competing ecological theories have been developed to predict the successful establishment of introduced plants based on the response of native herbivores to these exotics. The most prevalent theory is known as the Enemy Release Hypothesis (ERH), which predicts that native herbivores will prefer to feed on plants from their native range and that they will have better growth rates on these plants. Under this theory, exotic plants have greater success in their new environment because they have left behind coevolved natural enemies and because native generalist herbivores have a greater impact on native plants than on these exotics (Keane & Crawley, 2002;Schaffner et al., 2011). Conversely, the Biotic Resistance Hypothesis (BRH) posits that native herbivores would limit the range expansion of exotic plants due to their preference and higher growth rates on evolutionarily naïve plants (Parker & Hay, 2005). In contrast, the Novel Weapons Hypothesis (NWH) posits that invasive plant defense systems will be more effective in novel interactions with herbivores (Callaway et al., 2008;Callaway & Ridenour, 2004;Schaffner et al., 2011;Zheng et al., 2015).
However, predictions about invasive plant success based solely on these ecological theories fail to consider rapid evolution as a mechanism leading to their establishment. Several studies have found that herbivory is a strong selective force on plant defensive traits. Plants can rapidly adapt to their environment as herbivores drive ecological and evolutionary changes in plant populations quickly leading to increased survival in their new habitat (Agrawal, Hastings, Johnson, Maron, & Salminen, 2012;Zangerl & Berenbaum, 2005;Zangerl, Berenbaum, & Mallet, 2003;Züst et al., 2012). If exotic plants do leave behind enemies, this may lead to changes in plant antiherbivore defenses. Resistance traits in particular are costly to make. A lack of herbivorous insects could lead to the evolution of reduced phytochemical production and the channeling of those resources toward growth, known as the Evolution of Increased Competitive Ability (EICA; Agrawal et al., 2012;Ali & Agrawal, 2012;Blossey & Notzold, 1995;Uesugi & Kessler, 2013) Müller-Schärer, Schaffner, and Steinger (2004) argue that moving to a new location releases plants from some natural enemies, mostly specialists; but changing environments does not lead to a decrease in selective pressure on defense traits but rather changes them from focusing on both specialists and generalists to mainly focusing on generalist herbivores. Under this theory, known as the Shifting Defense Hypothesis (SDH), there should be selection for an increase in defensive toxins, which are more likely to affect nonadapted generalists (Doorduin & Vrieling, 2011). Lankau found that removing the dominant specialist of the invasive black mustard, Brassica nigra increased the concentration of the defensive compound sinigrin, which significantly deterred generalist herbivores, but removing the dominant generalist led to intermediate sinigrin concentrations (2007). Evidence for the SDH has also been shown in the California poppy (Leger & Forister, 2005), in Arabidopsis (Züst et al., 2012), and in Senecio pterophorus (Castells, Mulder, & Pérez-Trujillo, 2014). These studies not only provide evidence of contemporary selection and the rapid evolution of plant defensive traits in response to herbivore interactions but also show that evolutionary processes can have an effect on ecological process.
Plants have developed secondary metabolites that have a crucial role in both direct and indirect defenses against herbivores (Orians & Ward, 2010;War et al., 2012). While these defenses include both constitutive and inducible defenses, inducible defenses that are only activated after insect attack are especially interesting because it requires the plant to recognize the attacking insect and produce effective defensive compounds (Fürstenberg-Hägg, Zagrobelny, & Bak, 2013;Woodard, Ervin, & Marsico, 2012).
Both the Enemy Release and Biotic Resistance Hypotheses rely on evolutionary unfamiliarity between interacting species, a concept known as evolutionary mismatching, which can occur through two different mechanisms (Verhoeven, Biere, Harvey, & Putten, 2009).
The medics are a group of legume species that are indigenous to the Mediterranean Basin (Bena, Prosperi, Lejeune, & Olivieri, 1998).
Medicago polymorpha is one of the most common medic species outside cultivated alfalfa (Medicago sativa) and is now considered invasive (Paredes et al., 2002;Small & Jomphe, 1989). It was purposefully introduced to North and South America and Australia in the 1800s and has spread worldwide (de Haan & Barnes, 1998;Lesins & Lesins, 1979;Small & Jomphe, 1989;Spira & Wagner, 1983). Medicago polymorpha may have been accidentally introduced much earlier with records of its presence in S. America dating back 450 years (Del Pozo, Ovalle, & Avendaño, 1989). Previous studies of plant-herbivore interactions using medics have mainly focused on the direct and indirect effects the presence of an invasive species has on native host plants or, have been field studies so only ambient herbivory was measured (Lau, 2012;Lau & Strauss, 2005;Leakey & Lau, 2012). Those studies have in some instances shown evidence of genotypic differences between the ranges but are not direct assessments of how herbivores respond to Eurasian genotypes versus those from the New World.
In this study, we measure the feeding preference and growth of a common generalist herbivore, the soybean looper (Chrysodeixis includens), when allowed to feed on M. polymorpha genotypes originating from contrasting ranges: M. polymorpha's native range (Eurasia) and its invasive range (the New World, which overlaps with the herbivore; Figures 1 and 2). If Medicago populations underwent rapid evolutionary changes due to novel biotic interactions, then some of those changes may have allowed it to successfully establish in their new environment and avoid herbivory from insects present in that range. Specifically, we predict that Medicago populations in the New World will have evolved during the course of invading their novel habitat in ways that result in them interacting less strongly with generalist herbivores than naïve populations from their native range around the Mediterranean. This result would demonstrate evidence of rapid evolution in the introduced populations of M. polymorpha, which could be due to interactions with diverse herbivores in the invaded range, selection imposed by other novel biotic or abiotic stresses, or nonadaptive evolutionary processes such as genetic drift during founder events.

| Medicago polymorpha germplasm, germination, and growth
Medicago polymorpha is an annual legume that forms a symbiotic relationship with the rhizobium Ensifer medicae and is highly selfing in nature. We used 19 genotypes of M. polymorpha from their invasive range (North America, South America, and Australia,) and 19 genotypes from their native range (Eurasia) obtained from both the USDA NGRP and personal field collection ( Table 1). The field collection was created by collecting individual pods from the ground at least 1 m apart. These pods were assumed to represent unique maternal lineages because M. polymorpha is highly selfing in nature and the pods rarely disperse large distances (Lesins & Lesins, 1979). All parental plants of the genotypes we used were started from a single pod, selfed from a single progeny for three generations and grown in a tested showed low levels (0%-4%) of intra-accession crossing when they attempted to simulate a natural environment in a greenhouse (Vitale, Pupilli, Labombarda, & Arcioni, 1998

| Soybean looper larvae (C. includens)
The soybean looper is a generalist herbivore whose hosts include field crops such as soybeans, tomatoes, and peanuts, and wild plants such as the common cocklebur and sow thistle. Third instar soybean loopers, obtained from Benzon Research, were used in all assays.
The loopers were fed a multispecies artificial diet from Southland Products Inc. until they were used in the experiment.

| Food preference assay
We used 38 M. polymorpha genotypes in 84 pairings between plant ranges (one native-range genotype with one invasive-range genotype; see Appendix 1). The original plan was to compare preference between all possible combinations of native-and invasive-range genotypes, but we became tissue-limited during experimental setup. We then randomly chose pairings, which is why some genotypes occur more than others, but all pairs are only represented five times in the dataset to prevent overinfluence of a specific pairing. By setting up the experiment in this manner, we are able to examine range effects on preference but not genotypic variation. We placed an equal amount of native-range or invasive-range leaf material (~30 mg) on each side of a 100-mm petri dish. One caterpillar was placed in the middle of the petri dish and left to feed for 24 hr. Each pairing had 5 replicate petri dishes.
The dishes were placed in large plastic bins with moist paper towels to prevent the leaves from drying out. Petri dishes containing only leaves (~25 mg on average) of the genotypes served as controls to account for changes in leaf weight due to water loss. After 24 hr, we reweighed the remaining leaf material. The amount of leaf tissue consumed of each genotype was divided by the total amount of food consumed within the petri dish. Values over 0.5 indicate more of that genotype was preferred over the other available genotype and were given a score of 1, and the nonpreferred genotype was given a score of 0. Larvae Weight) + Factorial(# Days of experiment) * Average weight change per day))/ # Days of experiment] and following the procedure described by Waldbauer (1968). We also measured the mortality rate of the soybean loopers on each plant. It is unlikely that RGR or mortality rate was influenced by cannibalism because the experiments were not conducted under conditions that encouraged it (e.g., overcrowded containers, limited food source, and late stage larvae).

| No-choice assay
We compared the amount of leaf material consumed by the soybean loopers when they were not given a choice for 10 of the genotypes used in the full experiment. This experiment used the same set of plants but was set up one week after the choice experiment. Soybean loopers, approximately the same size as the ones that were used in the preference experiment, were individually placed in six-well tissue culture plates (Corning-Cat #3516) and given ~30 mg of leaf material from only one genotype. Leaf tissue was reweighed after 72 hr.

| Statistical analyses
Data analysis was done using R version 3.5.2 (R Core Team, 2018).
Range was treated as a fixed effect in all models. We ran exploratory analyses to determine whether the absolute value of latitude had any explanatory power, but model selection using the likelihood ratio test found the models to be a better fit when it was not included.
Genotype was included as a random variable nested within range to account for differences between genotypes not represented by the range term, and population was included as a random variable to account for the spatial aggregation of some genotypes within our sampling design. The population term thus accounts for any nonindependence due to genetic similarity of host genotypes originating from the same population. Pair was included to control for nonindependence of the genotypes within petri dishes for the preference assay. Normality, overdispersion, and heteroscedasticity of residuals were checked on all models where appropriate using DHARMa (Hartig, 2019). The preference data were analyzed using a generalized linear mixed model with a binomial distribution with the lme4 package (Bates, Mächler, Bolker, & Walker, 2014). To analyze the data from the insect growth assay, we removed three data points that represented larvae that lost body weight greater than two standard deviations of the mean over the course of the experiment, as these caterpillars were almost certainly not healthy prior to the start of the experiment. Then, we analyzed RGR using a linear mixed model with the lme4 package. Caterpillar mortality based on a count of the number of dead or missing caterpillars was analyzed using a generalized linear mixed model with a Poisson distribution. Not collapsing missing caterpillars into the dead column may have given a better model to analyze mortality but we did not record when we found dead caterpillars versus when they were just missing in this experiment. This decision was made based on a small pilot experiment where we did record alive caterpillars versus intact dead bodies versus missing caterpillars. Before declaring a caterpillar missing, all areas around the top layer of the soil were inspected, and then, plants were removed from their containers for a closer inspection of the soil. Several times, we found remnants of what were either decomposed bodies or molted skin, which could not be categorized with certainty. To analyze the no-choice consumption data, the proportion of leaf tissue eaten by the 10 genotypes when alone versus when paired was analyzed using a generalized linear mixed model with a beta distribution using glmmTMB (Brooks et al., 2017). An initial check of the residuals revealed heteroscedasticity, which was then built into a new model.
Comparison of the two models did not show a significant difference (χ 2 = 0.911, p = .3398), but the second model was used for analysis.
Fixed effects were analyzed using likelihood ratio tests for all models.

| Native-and invasive-range genotype herbivore choice experiments
Feeding preference of the soybean looper was significantly influenced by the range from which the M. polymorpha host genotypes originated (p = 0.047; Table 2 and Figure 3a). The native-range host genotypes (Eurasian) were preferred in 55.1% (±2.4%) of the trials over the co-occurring New World genotypes.

| Caterpillar growth and mortality on intact plants
The growth rate of the caterpillars is a performance metric that indicates the suitability of the plant. Although the soybean loopers tended to have a higher growth rate on Eurasian M. polymorpha, it was not significantly different from their growth rate on invasive New World tissue (Eurasian MP: 1.24 × 10 −2 mg/day ± 5.29 × 10 −3 mg/day; New World MP: 9.77 × 10 −4 mg/day ± 4.32 × 10 −3 mg/day, p = .183;  Figure 3c).

| No-choice leaf consumption
We compared the proportion of leaf material eaten by the soybean loopers when they were only given either Eurasian or New World  (Table 3 and Figure 3d).

| D ISCUSS I ON
The ability of exotic plants to become invasive upon establishment in a novel environment is puzzling, and current studies do not provide a clear consensus over the role of herbivorous insects in their establishment (Bezemer, Harvey, & Cronin, 2014;Chun, Kleunen, & Dawson, 2010;Harvey et al., 2010). While evolution is often thought a slow process that takes thousands of years, novel interactions between species can lead to rapid evolution that affect ecological dynamics (Johnson, Vellend, & Stinchcombe, 2009;Ohgushi, 2016;Slobodkin, 1980;Thompson, 1998Thompson, , 1999Turcotte, Reznick, & Hare, 2011). This is especially true in the case of antagonistic plant-insect interactions where there may be a continual defense-counter response arms race.
In our study, we compared the preference of a generalist herbivore, the soybean looper (C. includens), for native-range (Eurasian) M. polymorpha populations versus New World host populations that have evolved in the same range as the looper, but for less than 500 years (invasive-range M. polymorpha). We found that the soybean loopers showed a significant preference for the Eurasian M. polymorpha over the New World M. polymorpha when they were given a choice of leaf tissue (Figure 3a). We also compared the amount of leaf tissue that the soybean loopers ate when they were not given a choice versus when they were given a choice for a subset of the plants. In both cases, the soybean loopers showed a higher rate of consumption for Eurasian M. polymorpha genotypes, that is, the genotypes that did not have range overlap with the herbivore.
One of the problems with attempting to apply ecological theo- adapt to their environment as herbivores drive ecological and evolutionary changes in plant populations in their new habitat Züst et al., 2012). Both Hawkes (2007) and Schultheis, Berardi, and Lau (2015) found that enemy release declined over time as native and introduced species coevolved, to where, approximately 150 years after introduction, invasive species have similar responses to herbivory as native community members. It is possible that something comparable happens with biotic resistance. It highlights the importance of knowing the range of the focal plants used in experiments as well as the distribution of the herbivores used in the assays.
M. polymorpha has been a very successful invader, especially compared to other medic species (Bena, Lyet, Huguet, & Olivieri, 2005), which is suggestive of enemy release as a means for their expansion.
However, introduced M. polymorpha genotypes have been present in their invaded range (Del Pozo et al., 1989;de Haan & Barnes, 1998;Spira & Wagner, 1983) between 200 and 500 years, allowing sufficient time to adapt to native herbivores such as the soybean looper, and other environmental factors Lankau, 2007;Züst et al., 2012). When exotic M. polymorpha first arrived in the introduced range, it may have been susceptible to herbivore attack but over subsequent generations, it evolved to be less susceptible either through genetic drift or natural selection. Alternatively, it is possible that only better defended M. polymorpha genotypes were able to establish in the New World invaded range, resulting in evolution through lineage sorting. Any of these processes would yield a pattern similar to enemy release but achieved through evolutionary processes. Our results suggest that novel genotypes, particularly those from regions above southern Portugal (Helliwell et al., 2018), The difference in host preference and performance by the soybean looper could be an indication of the evolution of herbivore defense in M. polymorpha where changes in defense strategies are based on risk of herbivore attack (Orians & Ward, 2010). If herbivores are present but rarely attack in the invaded range, possibly because of differences in food cues (Stutz, Croak, Proschogo, Banks, & McArthur, 2017)  (native range) M. polymorpha genotypes, which in conjunction with the preference and no-choice consumption data suggest a shift toward inducible defenses. If escape is only from generalist herbivores (Müller-Schärer et al., 2004), then overall defensive compounds against generalists would be either maintained or increased (Doorduin & Vrieling, 2011;Orians & Ward, 2010) in New World (invaded range) plant genotypes. The slightly increased mortality on Eurasian native-range M. polymorpha is puzzling but herbivore preference does not always fall in line with performance metrics (Ikonen, Tahvanainen, & Roininen, 2002;Orians et al., 1997;Tomas, Box, & Terrados, 2011), which may be due to differences in selection on traits related to performance versus preference (Orians et al., 1997). Another explanation is that there has been a change in the type or quantity of defensive compound produced. In this study, we only measured RGR, but did not analyze nutritional indices, which may have further illuminated shifts in secondary metabolites or differences in nutritional quality. If we had included efficiency of conversion of ingested food into biomass (ECI) and efficiency of conversion of digested food into biomass (ECD), it would allow us to determine whether there was a shift from compounds that promote toxicity to those that are more antinutritional or antidigestive (Chen, 2008 there is some evidence of variation in the average fitness of genotypes in response to herbivory (terHorst & Lau, 2015), indicating the potential for rapid evolution in this species. We sampled as many populations as possible and included replicate genotypes for as many populations as were available. This hierarchical sampling accounts for range, between-population, and within-population genetic variation and is captured in the mixed models we used for analysis. The study by Helliwell et al. (2018) based on a novel, expanded collection that was not available to us showed that M. polymorpha variation is highest at the population level and lowest at the range level and found evidence supporting rapid evolution for flowering time in the invaded region. Future work leveraging genetic and genomic tools could partition how much of the change in plant-herbivore interactions is due to adaptation through natural selection versus genetic drift (Agrawal et al., 2015;Keller & Taylor, 2008;Schrieber et al., 2017). Thus, although there are alternative evolutionary hypotheses for the shift in plant-herbivore interactions that we document, this does not change our main assertion that there has been an evolutionary change, be it via adaptation through natural selection or evolution through genetic drift, resulting in New World genotypes rapidly evolving to have traits that impact their interactions with herbivores.

| CON CLUS ION
Our study found that the generalist herbivore, C. includens (soybean looper), preferred to feed on leaves from Eurasian native-range M. polymorpha rather than from M. polymorpha that have recently invaded the New World, the range where the herbivore occurs.
This preference for evolutionarily naïve genotypes shows that New World (invasive range) M. polymorpha populations underwent an evolutionary change in less than 500 years, which may have made it more similar to native, nonintroduced plants. This would result in a pattern similar to enemy release but occurring through contemporary evolution by the plant during invasion rather than historical divergence at the species or genus level.

ACK N OWLED G M ENTS
We thank members of the Friesen lab for helpful discussions, two anonymous reviewers, and the associate editors who provided helpful suggestions and comments on previous versions of this manuscript. We acknowledge support from the National Science

CO N FLI C T O F I NTE R E S T S
All authors declare they have no conflict of interest.

AUTH O R CO NTR I B UTI O N S
CNJ and MLF conceived and designed the experiment and wrote the manuscript. CNJ performed the experiment and analyzed the data.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data and R scripts used to analyze data are available at Dryad Digital Repository (https ://doi.org/10.5061/dryad.84h07s1).