The spitting image of plant defenses: Effects of plant secondary chemistry on the efficacy of caterpillar regurgitant as an anti‐predator defense

Abstract In the arms race between plants, herbivores, and their natural enemies, specialized herbivores may use plant defenses for their own benefit, and variation in plant traits may affect the benefits that herbivores derive from these defenses. Pieris brassicae is a specialist herbivore of plants containing glucosinolates, a specific class of defensive secondary metabolites. Caterpillars of P. brassicae are known to actively spit on attacking natural enemies, including their main parasitoid, the braconid wasp Cotesia glomerata. Here, we tested the hypothesis that variation in the secondary metabolites of host plants affects the efficacy of caterpillar regurgitant as an anti‐predator defense. Using a total of 10 host plants with different glucosinolate profiles, we first studied natural regurgitation events of caterpillars on parasitoids. We then studied manual applications of water or regurgitant on parasitoids during parasitization events. Results from natural regurgitation events revealed that parasitoids spent more time grooming after attack when foraging on radish and nasturtium than on Brassica spp., and when the regurgitant came in contact with the wings rather than any other body part. Results from manual applications of regurgitant showed that all parameters of parasitoid behavior (initial attack duration, attack interruption, grooming time, and likelihood of a second attack) were more affected when regurgitant was applied rather than water. The proportion of parasitoids re‐attacking a caterpillar within 15 min was the lowest when regurgitant originated from radish‐fed caterpillars. However, we found no correlation between glucosinolate content and regurgitant effects, and parasitoid behavior was equally affected when regurgitant originated from a glucosinolate‐deficient Arabidopsis thaliana mutant line. In conclusion, host plant affects to a certain extent the efficacy of spit from P. brassicae caterpillars as a defense against parasitoids, but this is not due to glucosinolate content. The nature of the defensive compounds present in the spit remains to be determined, and the ecological relevance of this anti‐predator defense needs to be further evaluated in the field.


| INTRODUCTION
The idea that plants affect not only their consumers, but also the interactions between the second and third trophic levels has received considerable attention since Price et al.'s seminal article in 1980. Plant traits that confer direct or indirect resistance against herbivores may have effects cascading up the food chain (Desurmont, Harvey et al., 2014;Hare, 2011;Harvey, Van Dam, & Gols, 2003). For example, the performance of parasitoids, which develop inside or on the outside of an herbivore host, has often been found to be correlated with the performance of the host: if plant defenses affect the host, parasitoid's performance is likely to be impacted the same way (Gols & Harvey, 2009;Gols et al., 2008). On the other hand, specialized herbivores may have adapted to their host plant's defenses and use them for their own benefit through the sequestration of plant toxins making them less palatable or better defended against natural enemies (Agrawal, Petschenka, Bingham, Weber, & Rasmann, 2012;Nishida, 2002;Opitz & Müller, 2009). In such cases, variation in plant defenses among host plants may directly impact the herbivore's level of defenses or unpalatability. For example, a host-specialist aphid species known to sequester plant toxins was found to be less well defended against a generalist predator when feeding on plants with low levels of noxious chemicals (Francis, Lognay, Wathelet, & Haubruge, 2001). At an evolutionary level, the benefits that herbivores derive from plant defenses have the potential to drive host plant preferences, dietary specialization, and speciation processes for both herbivores and their host specialized natural enemies (Abrahamson & Blair, 2008;Ali & Agrawal, 2012;Quicke, 1997).
Studies on the effects of plant secondary chemistry on antipredator defenses have almost exclusively focused on sequestered toxins (Hartmann, 2004;Heckel, 2014), largely missing out on other strategies deployed by herbivores to fend off attackers (Karban & Agrawal, 2002). Moreover, most of these studies concern just a few host plants with a limited range of chemical profiles (Calcagno, Avila, Rudman, Otero, & Alonso-Amelot, 2004;Francis et al., 2001;Sword, 2001). Here, we tested the hypothesis that plant chemistry affects the efficacy of another type of herbivore defense: the active regurgitation of oral secretions on natural enemies.
The secretion or regurgitation of defensive chemicals to fend off attackers has been documented in nature for a wide range of animals, including lizards (Middendorf & Sherbrooke, 1992) and birds (Canestrari et al., 2014;Warham, 1977). Among insects, examples of this defense mechanism include autohemorrhaging behavior (i.e., reflex bleeding) in several species of beetles and grasshoppers (Bateman & Fleming, 2009), secretion of deterrent chemicals from specialized glands (Eisner & Aneshansley, 1999;Pasteels, Rowell-Rahier, & Raupp, 1988), and regurgitation of oral secretions and/or semi-digested plant material (Hunter, 2000;Tullberg & Hunter, 1996). Defensive regurgitation is common among lepidopterans at the larval stage (Gross, 1993), and it is typically associated with an enlarged foregut in caterpillars using this weaponry (Grant, 2006). In the Pieridae family, the cabbage white Pieris brassicae is known to actively regurgitate on attacking natural enemies (Müller, Agerbirk, & Olsen, 2003), including its main parasitoid Cotesia glomerata (Hymenoptera: Braconidae) ( Figure S1). Interestingly, closely related species such as Pieris rapae or Pieris virginiensis do not display such behavior (Hunter, 2000). Moreover, regurgitation was found to have an intrinsic cost for P. brassicae caterpillars, reducing pupal weight in caterpillars that regurgitated more frequently or more abundantly (Higginson, Delf, Ruxton, & Speed, 2011). These findings suggest that defensive regurgitation must have an adaptive value to be maintained in P. brassicae in nature. Müller et al. (2003) found that P. brassicae regurgitant is deterrent to ants, suggesting the presence of compounds with anti-predator properties. Being a specialist feeder on glucosinolate-containing plants, it is possible that P. brassicae derives its chemical weaponry from its host plants' secondary metabolites.
Glucosinolates are a class of plant compounds known for their defensive function (Ahuja, Rohloff, & Bones, 2010;Hopkins, van Dam, & van Loon, 2009). There are more than 120 described glucosinolates, present in 16 plant families (Fahey, Zalcmann, & Talalay, 2001). Although they are known to be toxic or deterrent to many species of herbivores, several host-specialized insect herbivores have adapted to deal with these compounds (Winde & Wittstock 2011). This is the case for several butterflies belonging to the Pieris genus, which use them as oviposition and/or feeding stimulants (Hopkins et al., 2009). Our study species, P. brassicae, develops mainly on wild and cultivated crucifers (family Brassicaceae), but is also known to infest other glucosinolate-containing plants in nature, for example the ornamental nasturtium Tropaeolum majus (family Tropaeolaceae) (Geervliet et al., 2000).  (Schweizer et al., 2013). To investigate the link between plant chemical profile and regurgitant efficacy, we extracted and measured the glucosinolate content in leaves of all host plants. Our working hypothesis was that C. glomerata behavior should be more strongly impacted by regurgitant originating from plants with higher glucosinolate concentrations, and we tested this prediction at three different levels: across all host plants, within the three B. rapa populations, and between the two A. thaliana lines.

| Plant and insect material
We used a total of 10 host plant species and varieties for the dif- Pieris brassicae caterpillars used for the experiments came from a laboratory rearing originally started with individuals collected in the field in the Zürich area (Switzerland). The braconid parasitoid C. glomerata is the main natural enemy of P. brassicae in temperate Western Europe and attacks early larval instars of its host. Parasitoids used in this study came from a laboratory rearing originally started with individuals collected in the field in the Neuchâtel area (Switzerland). Parasitoids were reared on P. brassicae caterpillars from the laboratory rearing.
Newly emerged parasitoids of both sexes were placed in Bugdorm-1 cages (30 × 30 × 30 cm, Mega View Science Education Services Co. Ltd, Taiwan) at ambient temperature (ca. 25°) with water and honey for 48 h, a period that is typically sufficient to ensure successful mating.
Then, the rearing cages were transferred in a growth chamber at 13°C (16/8 L:D light regime) with water and honey until parasitoids were needed for the experiments (1-3 weeks after emergence).

| Observations of regurgitation events on different host plants
The aim of this experiment was to evaluate the efficacy of caterpillar regurgitation as an anti-predator defense by observing the behavior of . We categorized the body parts of C. glomerata wasps as follows: head, thorax, abdomen, legs, and wings. Grooming behavior was defined as parasitoid leg movements aimed at cleaning different body parts (Zhukovskaya, Yanagawa, & Forschler, 2013). We recorded the total duration of grooming until the parasitoid attacked a second caterpillar, or until the observation period was over (15 min after the initial attack). If a parasitoid interrupted grooming for more than a few seconds and resumed grooming afterwards, the durations of all grooming sequences were added to obtain the total grooming time. The behavior of an individual parasitoid was only recorded once (i.e., one parasitoid was used per replicate). Caterpillars used for this experiment were feeding in groups of various sizes (10-25 caterpillars per group) and had been developing on their host plant since egg hatch. The number of regurgitation events that were recorded per group of caterpillars varied depending on the host plant and availability of caterpillars.

| Manual application of regurgitant on Cotesia glomerata parasitoids
We collected regurgitant from P. brassicae caterpillars feeding on leaves of 10 host plant species (listed above). Regurgitant was obtained by holding the caterpillar between two fingers and gently applying pressure below the head, while collecting the regurgitant directly from the mouthparts using a micropipette. The regurgitant was immediately stored on ice, pooled for each plant species, and frozen at −20°C until use. Regurgitant was collected from third to fifth instar larvae, except for the two A. thaliana lines for which second and third instar larvae were used because these plants did not produce enough foliage to sustain bigger caterpillars. For the bioassay, we placed a leaf circle of B. rapa var. pekinensis with groups of newly hatched first instar P. brassicae on moist filter paper into a Petri dish. Newly hatched larvae were chosen for this experiment because they are too small to produce significant amounts of regurgitant and thus should not interfere with the manual application of regurgitant. Using an aspirator, we transferred individual C. glomerata females, to the Petri dish and placed them in close proximity to the feeding larvae. We applied 2.5 μl regurgitant with a micropipette onto the thorax once the wasp had inserted its ovipositor into a larva. An application of 2.5 μl water was used as control treatment. We then recorded the behavior of the parasitoid until it attacked a second larva (re-attack), or for a maximum observation time of 15 min. The following parameters were recorded: attack interruption upon regurgitant application (yes/no), initial attack duration (s), grooming time (s), re-attack within 15 min after the initial attack (yes/no), and time until re-attack after regurgitant application (s). As before, all grooming sequences were added to obtain the total grooming time. Because all plant species could not be grown at the same time, we conducted three separate series of this experiment with different parasitoids, larval batches, and host plants tested. A total of 332 manual applications of P. brassicae regurgitant from 10 host plants  Table 1. Glucosinolate profiles were analyzed following a modified protocol from Glauser, Schweizer, Turlings, and Reymond (2012). In brief, 11.5 (±1.5) mg of frozen leaf powder were suspended in 1 ml of cold methanol/water (70:29.5 v/v; 0.5% formic acid) and five glass beads were added. In contrast to Glauser et al. (2012), we used acidified extraction solvent to quench residual enzyme activity without any detectable influence on the analysis' outcome compared to heat inactivation.

| Glucosinolate extraction and analysis
Following a brief vortex and agitation in a bead mill at 30 s −1 for 3 min, all samples were centrifuged at 14 000 rpm for 5 min (20 800× g).
Gluconapin was used at 0.2, 1, 5, and 20 μg/ml to establish a calibration curve for glucosinolate quantitation except for the two A. thaliana accessions where glucobrassicin and glucoraphanin were used at a single concentration (2.5 μg/ml) to estimate glucosinolate contents.

| Statistical analysis
For the first experiment, parasitoid grooming time was analyzed by running a two-way ANOVA with host plant, body part reached, and the interaction between these two terms as effects in the model.
Data were square root transformed to meet the assumptions of the model. Means were compared using a Tukey-Kramer post-hoc test.
An additional analysis was conducted to determine whether grooming time differed when more than one body part was reached. For this analysis, data from all observations where one body part was reached were pooled and were compared to observations where more than one body part was reached using a one-way ANOVA. For the second experiment, data for the treatments that were common among the three series of the experiment (BR1, BR3, R, and water) were pooled together and the effects of host plant, series, and the interaction between the two terms on the parameters recorded (attack interruption upon regurgitant application (yes/no), initial attack duration (s), grooming time (s), re-attack within 15 min after the initial attack (yes/no), and time until re-attack (s)) were analyzed. In addition, data were also analyzed separately for each series, testing the effect of host plant on the parameters recorded. For these analyses, general ANOVAs for continuous variables and Chi-square tests for proportion variables were used. Means were compared using a Tukey-Kramer post-hoc test, and data were square root transformed to meet the assumptions of the model if necessary. Total glucosinolate levels were calculated for each plant by summing the concentrations of individual compounds. Genotypic effects on total glucosinolates were then analyzed using a one-way ANOVA followed by a Holm-Sidak post-hoc test. Because the glucosinolate extraction and quantification protocol were not done at the same time for the two A. thaliana accessions and the other host plant tested, they were excluded from this analysis. All data were square root transformed prior to the analysis. Associations between glucosinolate content (total levels and individual compounds) and parameters of parasitoid behavior (attack duration, grooming time, and time to re-attack) were investigated using linear regression procedures with a Bonferroni correction (JMP9).

| Manual application of regurgitant on Cotesia glomerata parasitoids
The analysis of the pooled data of the four treatments common to the three experimental series (BR1, BR3, R, and water) overall revealed that the behavior of parasitoids was less affected when water was applied rather than regurgitant and that regurgitant originating for radish-fed caterpillars (R) had a stronger impact on parasitoid behavior (significantly lower proportion of re-attacks within 15 min and highest grooming time) than regurgitant originating from other host plants (Figure 3). In detail, the variables included in the model (host plant, series, and interaction between both terms) did not explain variation in parasitoid attack duration (F 7,125 = 1.6, p = .14), but they had a significant effect on the proportion of attacks interrupted (df = 7, χ 2 = 14.8, p = .04), on grooming time (F 11,182 = 7.1, p < .0001), on the proportion of observing a re-attack within 15 min (df = 11, χ 2 = 68.3, p < .0001), and on the timing to re-attack (F 11,182 = 8.6, p < .0001). Host plant (df = 3, χ 2 = 13.5, p < .01) had a significant effect on the proportion of attacks interrupted, but not series (df = 1, χ 2 = 2.9, p = .08), nor the interaction between both terms (df = 3, χ 2 = 4.5, p = .2). More parasitoid attacks were interrupted with regurgitant from BR1, BR3, and R, than with water ( Figure 3a). Host plant (F 3,182 = 11.0, p < .0001) and series (F 2,182 = 15.1, p < .0001) had a significant effect on grooming time, but not the interaction between both terms (F 6,182 = 1.8, p = .1): grooming time was longer with BR1, BR3, and R than with water ( Figure 3b). Host plant (df = 3, χ 2 = 15.7, p = .001) and series (df = 2, χ 2 = 22.1, p < .0001) had a significant effect on the proportion of observing a re-attack within 15 min, but not the interaction between both terms (df = 6, χ 2 = 10.7, p = .09). Proportion of second attacks was the lowest with R, higher with BR1 and BR3, and the highest with water ( Figure 3c) proportion of re-attacking a caterpillar, and shorter time to re-attack) during series 3 (Fig. S1).
The additional analyses of the three experimental series separately further supported that applying water to the parasitoids had much less impact on their behavior than the application of regurgitant (Table 1).
However, we found very few significant differences among host plants within each series. Specifically, in series 1, proportion of re-attacks within 15 min was lower on R than on all other host plants; in series 3, grooming duration was higher on BR3 than on BR1 (Table 1). Importantly, we did not find any difference in parasitoid behavior when exposed to regurgitant originating from an Arabidopsis line containing glucosinolates or an Arabidopsis line deficient in glucosinolates (Table 1, series 3).

| Glucosinolate analysis
Following infestation by P. brassicae, we detected 20 different glucosinolates across the nine host plants that were tested, and the different host plants differed remarkably in their glucosinolate profiles both in quantity and quality ( Figure 4, Table S1). As expected, F I G U R E 4 Leaf glucosinolate profiles following 24 h of Pieris brassicae caterpillar feeding (μg/g fresh weight; means ± SE; N = 5, except Col-0: N = 4). BR1, Brassica rapa rapa wild population 1; BR2, Brassica rapa rapa wild population 2; BR3, Chinese cabbage Brassica rapa var. pekinensis; CAP, nasturtium Tropaeolum majus; R, radish Raphanus sativus; O, green cabbage Brassica oleracea; S, spider flower Cleome hassleriana; Col-0, Arabidopsis thaliana wild-type; myc234, A. thaliana glucosinolate-free mutant. Means followed by a different letter indicate that the total glucosinolate quantity is statistically different between two host plants (one-way ANOVA, α = 0.05, JMP9). Glucosinolate quantification for A. thaliana accessions was done at a later time than the other host plants and they were excluded from the analysis; the dashed line indicates this separation glucotropaeolin and glucoraphenin, respectively. The three B. rapa accessions had slightly more complex glucosinolate blends (Table S1).
The quantitative variation among the three accessions, however, was similarly high ranging from 1583 μg/g FW for BR1 to 117 μg/g FW for BR3. BR2 showed intermediate glucosinolate levels. Overall, host plant explained 87% of the observed quantitative variation in glucosinolate content (F 6,28 = 32.462, p < .001). We found no significant associations between total glucosinolate levels or the concentrations of individual glucosinolates and parasitoid behavior parameters (attack duration, grooming time, time to re-attack) (Ps > .05, JMP9).

| Regurgitation as a defense against natural enemies
Overall, we found good evidence of the efficacy of regurgitant as a defense, with parasitoids more likely to interrupt their attack, spending more time grooming and less likely to re-attack a caterpillar within 15 min after exposure to regurgitant than after exposure to water ( Figure 3). Regurgitant was observed to stick to the hydrophobic cuticle of the parasitoid, whereas droplets of water were simply deflected.
This confirms that P. brassicae regurgitant is highly amphiphilic and acts as a surfactant that can wet the cuticle of arthropod attackers, as it is the case for regurgitant of other caterpillars (Rostás & Blassmann, 2009). Moreover, grooming time was dependent on the body part that was covered with the regurgitant, with wings being the most time-consuming body part to groom (Figure 2). We also found direct evidence that host plant influences to a certain extent the efficacy of regurgitation. Indeed, parasitoid behavior was more affected when regurgitant originated from radish than from wild or cultivated B. rapa

| Host plant glucosinolate profile and regurgitant efficacy
Overall, we found very little support for the hypothesis that plant glucosinolate content is associated with efficacy of P. brassicae regurgitant. There were no significant correlations between total glucosinolate levels or quantities of single glucosinolate compounds and parasitoid behavior for any of the response parameters tested, and parasitoid behavior was not differently affected when parasitoids were exposed to regurgitant originating from B. rapa populations with high, medium, and low glucosinolate content (BR1, BR2, and BR3, respectively). Moreover, the fact that regurgitant was not less effective when originating from a glucosinolate-deficient A. thaliana accession rules out an important role for glucosinolates in anti-parasitoid defense. Still, it should be noted that the host plant that had the strongest effect on the parasitoids was R. sativus (R), a host plant with a unique glucosinolate profile, albeit in low concentrations. Indeed, the major glucosinolate present in R. sativus, glucoraphenin, is absent in the other host plants tested (Table S1).
It appears that the anti-predator properties of P. brassicae regurgitant may be based on self-synthetized compounds or on plant-derived compounds other than glucosinolates. Most of the research on plant defenses in the Brassicaceae family has focused on the role of glucosinolates as major mediators of insect-plant interactions (Hopkins et al., 2009), but non-glucosinolate key compounds have also been documented (Roessingh, Städler, Baur, Hurter, & Ramp, 1997). Testing the spit of caterpillars reared on artificial diet would be one possible way to further elucidate the effects of host plant chemistry on regurgitant efficacy, but is challenging to do with P. brassicae.

| Significance of this research and future directions
Plant-mediated effects of defensive regurgitation in insects have been studied in orthopterans (Calcagno et al., 2004;Sword, 2001), but have rarely been examined for lepidopteran larvae (but see Higginson et al., 2011). Our study is a first attempt to unravel the link between plant chemistry and anti-predator defenses in P. brassicae, a nonsequestering herbivore. The two experimental designs that we used In conclusion, our study supports the idea that the value of P. brassicae's regurgitant as an anti-predator defense is mainly due to its amphiphilic proprieties (Rostás & Blassmann, 2009), with little effect of plant secondary metabolites on the deterrence of the regurgitant. Ecologically, possessing an anti-predator defense whose efficacy is independent of plant chemistry may be beneficial for P. brassicae, by alleviating some possible costs of host plant switching and diet mixing in nature. Given that amphiphilic properties have been reported for a wide range of caterpillar regurgitants (Rostás & Blassmann, 2009), it is therefore puzzling that close relatives of P. brassicae, P. rapae, and P. virginiensis do not actively regurgitate on natural enemies (Hunter, 2000). This may have something to do with the gregarious feeding behavior of P. brassicae, which is also distinctive among pierids. Young P. brassicae larvae feed in close groups up to the third instar, whereas its relatives are typically solitary. The benefits of defensive regurgitation for P. brassicae may only be present or enhanced in the context of gregarious feeding. In our study, regurgitation rarely prevented caterpillars from getting stung by parasitoids, as only a few attacks were interrupted after exposure to regurgitant (Table 1). Yet, successful regurgitation on a parasitoid by a caterpillar within a feeding group may help protect its kin within the group, and thus regurgitation may be more effective for gregarious species than solitary species. Indeed, each exposure to regurgitant made parasitoids waste time and energy grooming, which may benefit members of a group by reducing the overall attack rate (Turchin & Kareiva, 1989).
Moreover, repeated exposures to deterrent chemicals may have more serious effects on parasitoids than one-time exposures. Consequently, the adaptive value of defensive regurgitation for P. brassicae should be investigated in the field in the context of its gregarious behavior, without underestimating the possible effects of host plant traits on the benefits of group living (Desurmont, Weston & Agrawal 2014).

ACKNOWLEDGMENTS
The authors thank Tom de Jong (Leiden University) for providing seeds of wild B. rapa accessions from the Netherlands, and Phillipe Reymond (University of Lausanne) for providing Arabidopsis seeds. We are also thankful to Marcel Dicke, who suggested including a glucosinolatedeficient Arabidopsis accession in the study, and Emilie Demarsy for her help with growing the Arabidopsis plants in Neuchâtel. Yves Borcard and Thomas Degen, from University of Neuchâtel, kindly provided the illustration of a C. glomerata wasp. We are also thankful to Marie-Jeanne Tschudi for her help with growing the plants needed in the study, and to the APP (Apprentissage par problème) program of the University of Neuchâtel without which this study would not have existed.