Heritable variation in prey defence provides refuge for subdominant predators

Abstract

Generalist predators with broadly overlapping niches commonly coexist on seemingly identical sets of prey. Here, we provide empirical demonstration that predators can differentially exploit fine-grained niches generated by variable, heritable and selective defences within a single prey species. Some, but not all, clones of the aphid Aphis craccivora are toxic towards the dominant invasive predatory ladybeetle, Harmonia axyridis. However, other less competitive ladybeetle species are not affected by the aphid's toxic trait. In laboratory and open field experiments, we show: (i) that subdominant ladybeetle species were able to exploit the toxic aphids, benefitting from the suppression of the dominant predator; and (ii) that this narrow-spectrum toxicity can function as an anti-predator defence for the aphid, but depends on enemy community context. Our results demonstrate that niche differentiation among generalist predators may hinge upon previously underappreciated heritable variation in prey defence, which, in turn, may promote diversity and stability of enemy communities invaded by a dominant predator.

1. Introduction

In a world full of predators and natural enemies, prey have evolved a myriad of morphological, physiological and/or behavioural defences to promote their own survival [1–3]. However, the community of potential enemies is diverse, and prey defences are often enemy-specific or ‘narrow-spectrum’, affecting only a subset of the enemy community [4–6]. Further, defensive capability can vary considerably within a prey species, owing to phenotypic plasticity [7], maternal effects [8], symbiotic microbial defences [9], or other inherited traits [10–12]. For predators that attack such prey, the consequences are twofold. First, only a subset of individuals within a prey species may be functionally consumable. Second, the consumable subset of prey may differ among predator taxa, depending on each predator's susceptibility to various prey defences. This brings up an intriguing possibility: can differential defence within a single prey species promote coexistence of competing enemy taxa?

Conceptually, prey defence is capable of modulating competition among multiple enemies, but the outcome depends on the competitive hierarchy of the enemies and their relative susceptibility to the defence (electronic supplementary material, figure S1) [13–15]. If prey are narrowly defended against only less competitive enemies, then the superior competitor will dominate, just as it would if the prey were entirely undefended. If the prey are broadly defended against all natural enemies, then the prey itself will benefit. The interesting scenario arises when the prey are narrowly defended against a competitively superior natural enemy, but not against less competitive enemies. In this instance, less competitive enemy species may be able to exploit a consumptive niche that is unavailable to the superior competitor, bypass any competitive exclusion and persist in the community [16]. The narrowly defended prey, in turn, may receive little benefit from its own defensive trait, if the less competitive enemies effectively replace the eliminated competitor [17,18]. Consequently, the selective benefit of a narrow-spectrum defence for the prey will likewise depend on the composition and relative susceptibility of the enemy community. For defensive traits that are heritable, one would then expect eco-evolutionary feedbacks between the natural enemy assemblage and the defensive topology of the prey population in a particular ecological community [10].

Intraspecific trait variation can be substantial and is receiving increasing attention in community ecology [19]. For heritable prey defences, variation owing to maternally inherited symbiotic microbes has been a particular focus [16,20]. For example, many aphid species are infected with inherited bacteria that confer resistance to some (but not all) parasitoid wasps [21,22]. These defended aphids are typically found at intermediate frequencies in natural populations [9]. In the laboratory, separate studies have demonstrated that defensive symbiont infection frequency in prey populations can vary as a function of parasitoid attack [23], and that the success of different parasitoids can vary as a function of aphid symbiotic defence [16,20]. Feedbacks between these prey defensive traits and enemy communities under field conditions are only beginning to be investigated [24,25], and only within the relatively narrow context of parasitoids and defensive microbes. Other enemy guilds, such as generalist predators with broadly overlapping niches, may also respond to heritable intraspecific variation in prey defence, but have not been tested.

The cowpea aphid, Aphis craccivora, and its natural enemies in North America comprise an excellent system to investigate the intersection of intraspecific variation among prey, narrow-spectrum defence and competition among multiple enemy species. Aphids employ a varied arsenal of behavioural and physiological defences against their natural enemies [26,27]. Among these, Ap. craccivora demonstrates dramatic intraspecific variation in defence towards ladybeetles. Some clonal lineages exhibit toxicity to the multicoloured Asian ladybeetle, Harmonia axyridis, whereas other lineages support normal H. axyridis development [12,28,29]. The mechanism of this toxicity remains unknown, but is endogenous to certain aphid genotypes and is not a function of host plant or bacterial symbiont infection [12]. Furthermore, this defence is narrow-spectrum: toxic aphid clones have only minimal effect on other coccinellid species including the naturalized seven-spotted ladybeetle Coccinella septempunctata and native species such as Coleomegilla maculata and Cycloneda munda [5]. Harmonia axyridis is a dominant invasive species and intraguild predator implicated in the decline of other ladybeetles during invasions of new continents [30]. Consequently, it is particularly important to understand factors that might promote the continued persistence of native species within invaded communities [31]. Many niche-related mechanisms have been implicated in coexistence, or lack thereof, among this predator guild [32–34], but differential suitability within a single prey species has not been explored.

Here, we used intraspecific variation in Ap. craccivora toxicity as an experimental tool to determine: (i) whether other natural enemies benefit from the prey's narrow-spectrum defence against a dominant competitor, H. axyridis; and (ii) whether the prey themselves benefit from a trait that defends them from only a portion of their natural enemy community. We first tested the patch tenure and oviposition patterns of adult H. axyridis in the field, to verify that reduced usage of toxic aphids creates a potentially exploitable resource for other enemies. We next studied interactions among differentially defended prey and beetle larvae in a controlled laboratory environment (electronic supplementary material, figure S1), using Coc. septempunctata as a less toxin-susceptible subdominant competitor. We then expanded to an open field experiment where naturally occurring enemies were allowed to attack differentially defended aphid colonies, and we could evaluate whether a variety of different arthropod functional groups responded to our treatments. Unexpectedly, this experiment indicated that toxic aphid clones might be resistant to a common parasitoid wasp. Therefore, we executed a controlled laboratory experiment to confirm the effect on these wasps and inform the interpretation of our field experiment. Collectively, these experiments demonstrate, in the open field, that a generalist predator community can and does respond to heritable intraspecific variation in prey defences. Certain predators exploit the fine-grained niches generated by defended prey, suggesting that predator community composition and prey defensive arsenals dynamically shape one another.

2. Material and methods

(a). Study organisms

Aphids exhibit parthenogenetic reproduction, allowing maintenance of genetically identical clonal colonies over time. All experiments were conducted with six Ap. craccivora clones; three toxic to H. axyridis (LE, LW, LS) and three non-toxic clones (AC, AV, AL) [5,12]. Toxic clones were originally collected from black locust (Robinia pseudoacacia) and non-toxic clones were originally collected from alfalfa (Medicago sativa; see [35]). It is noteworthy, however, that toxic and non-toxic aphid clones can be found intermixed on the same host plant in the field (J. White 2016, personal observation), and the toxic trait is not a function of host plant [12]. All clones perform well on fava bean (Vicia faba) with comparable population growth rates [35], and aphids retain their respective toxic/non-toxic properties on fava [12] so this plant is used for all experiments herein. All aphid clones have been maintained in the laboratory for 5–7 years on fava, and were previously cured or naturally uninfected with facultative bacterial symbionts [12,35].

Ladybeetle larvae (both H. axyridis and Coc. septempunctata) used in our experiments were the offspring of field-collected adults that were maintained in the laboratory on a diet of pea aphids, Acyrthosiphon pisum (see methods in [12]). We initiated a laboratory population of Lysiphlebus testaceipes wasps from aphids retrieved in the field experiment, and maintained a colony on non-toxic Ap. craccivora clones (methods in [36]).

(b). Adult Harmonia axyridis behaviour

We first evaluated the behaviour of adult H. axyridis in an open field experiment to determine whether beetles discriminated against toxic Ap. craccivora. We compared: (i) the tenure of beetles in patches of toxic versus non-toxic Ap. craccivora, and (ii) total egg deposition in the two patch types. Potted fava plants, infested with approximately 3000 aphids from either a toxic or non-toxic clone, each were enclosed in a mesh bag along with two mated adult female H. axyridis that were marked with treatment-specific paint. We placed the plants into the ground 5 m apart in an old field on the University of Kentucky's Spindletop Research Farm (38°7′29″ N, 84°30′39″ W), and removed the bags. We counted marked beetles and total number of eggs hourly for 6 h, and once at 24 h. Experimental blocks were conducted in June, August and September 2015, with n = 8, 16 and 15 plants, respectively. For statistical analysis, experimental date was initially included as a random blocking factor, but was removed when we found no significant effect in any analysis. We compared H. axyridis adult counts between toxic and non-toxic treatments for the first 6 h using repeated measures MANOVA, final 24 h beetle counts with a t-test, and total egg counts with a generalized linear model (GLIM) using a Poisson distribution. All analyses were conducted in JMP^® (version 11 Pro. SAS Institute Inc., Cary, NC, 1989–2007).

(c). Controlled community microcosm experiment

We next used a laboratory microcosm study to test: (i) whether an alternate natural enemy, Coc. septempunctata, could indirectly benefit from the Ap. craccivora toxic defence via a negative effect on H. axyridis and (ii) whether toxic Ap. craccivora benefited from their narrow-spectrum defence in different enemy community contexts. To isolate the effects of enemy species identity, we used a substitutive design, which changed community composition but held ladybeetle density constant. We evaluated aphid density and ladybeetle survival across two aphid toxicity treatments (toxic/non-toxic) in factorial combination with two ladybeetle community treatments, one that included both Coc. septempunctata and H. axyridis (H. axyridis+), and one that was only Coc. septempunctata (H. axyridis−). We ran a total of eight microcosms per treatment. Two microcosms per treatment were run on each of four blocked experimental dates. Microcosms consisted of 60 × 60 × 60 cm fine mesh and plastic enclosures stocked with four greenhouse-grown fava seedlings in 10 cm pots, each hosting 300 aphids at the start of the experiment. Each microcosm received one type of aphid (toxic or non-toxic) that was a mixture of two different clones (LE/LW or AC/AV). All cages received eight neonate ladybeetle larvae: H. axyridis+ enclosures received four H. axyridis and four Coc. septempunctata larvae while H. axyridis− enclosures received eight Coc. septempunctata larvae. The cages were maintained indoors under fluorescent grow lamps on a 16 L : 8 D cycle at 25°C and given approximately 100 ml of water per plant every 48 h. Within each cage, we counted surviving ladybeetle larvae and total aphid density after 7 days.

We compared H. axyridis survival between the two H. axyridis+ treatments using GLIM with binomial distribution and logit link function, including aphid toxicity as the fixed factor. We analysed Coc. septempunctata survival and overall ladybeetle survival in the same fashion but included the predator community treatment as a fixed factor. We compared final aphid densities among treatments using ANOVA with log-transformed data. Experimental date was initially included as a random blocking factor in all analyses, but was subsequently removed when we found no significant effect.

(d). Effects of aphid toxicity in the field

We also conducted an open field experiment in which we manipulated aphid toxicity and H. axyridis presence, but allowed plants to be colonized by naturally occurring arthropods in an agroecosystem. We were therefore able to evaluate: (i) whether the enemy community recruited differentially to toxic versus non-toxic aphids in the presence or absence of H. axyridis, and (ii) whether such recruitment undercut the defensive virtue of the toxic trait for the aphid. We stocked each of 90 greenhouse-grown, three-week-old fava plants with 500 aphids of mixed developmental stages from either a toxic or non-toxic aphid clone. We enclosed the plants in mesh bags and transferred them to the Spindletop Research Farm in April 2016. We buried empty pots flush with the soil on a 5 × 5 m grid within a 30 × 95 m field plot that had been tilled and was bare of vegetation. We randomly assigned each experimental plant to an empty pot on the grid. Plants were watered daily unless there was rainfall the previous day. After 3 days, we unbagged 72 plants that were split equally across toxic/non-toxic and H. axyridis+/− treatments in a 2 × 2 factorial design (n = 18 plants/treatment). Harmonia axyridis+ plants received 20 neonate H. axyridis larvae, which corresponds to an average sized egg clutch [37]. Harmonia axyridis− plants were unbagged but not augmented with H. axyridis larvae. To control for the possibility of differential population growth rates among aphid clones, we kept an additional 18 plants (nine with toxic aphids and nine with non-toxic) enclosed in their bags. At the conclusion of the experiment, we counted aphids in these no-predator controls and compared population size between toxic and non-toxic colonies using a two-sample t-test of log-transformed aphid numbers.

The experiment lasted for 10 days. Each day, we visually inspected all unbagged plants, counting aphids, H. axyridis larvae, and any other arthropods found occupying the plants. We sight-identified arthropods to the lowest taxonomic unit possible (electronic supplementary material, table S1) without removing or collecting any individuals. We also recorded developmental stage for H. axyridis larvae. After the final visual sample, we individually bagged plants and returned them to the laboratory for destructive sampling and a more detailed count of aphids and visibly parasitized aphids (mummies).

Statistically, we first compared H. axyridis counts over time in the two H. axyridis+ treatments, to confirm that aphid toxicity negatively affected the predator. We tested for differences in H. axyridis larvae counts between toxicity treatments (toxic, non-toxic) over time using repeated measures MANOVA and compared treatments at each time point using associated univariate ANOVA.

Next, we evaluated treatment effects on aphid density to determine whether toxic aphid colonies benefitted from their narrow-spectrum defence under field conditions, either in the presence or the absence of H. axyridis. First, we log-transformed and compared aphid numbers per plant on day 10 using a fully factorial ANOVA, with aphid toxicity (toxic and non-toxic) and H. axyridis presence (H. axyridis−, H. axyridis+) as fixed factors. We used planned contrasts between toxic/non-toxic aphids for each H. axyridis treatment. However, because we detected a difference between the intrinsic growth rate of non-toxic and toxic aphids in bagged control plants (mean ± s.e. non-toxic aphids = 4918 ± 467, toxic aphids = 3546 ± 270; t = −2.46, d.f. = 16, p = 0.026) we also considered final aphid density as a function of potential population size. As a correction factor, we divided final aphid density of each unbagged plant by the mean density of the corresponding aphid strain (toxic and non-toxic) in bagged plants. We then compared these values across treatments using a fully factorial ANOVA, with log-transformation of the data to satisfy homogeneity of variance assumptions.

For other observed arthropods, taxa were assigned to functional ecological groups (electronic supplementary material, table S1). Total observed counts for each functional group were summed per plant across the 10 day experiment with the exception of aphid mummies, where only the final count was used. To assess whether raw abundance of these groups differed by treatment, we first tested for differences in the observed counts of each functional group across treatments using fully factorial GLIM with either a normal or Poisson distribution [38], depending on the distribution of count data (electronic supplementary material, tables S2–S4). All GLIMs included an overdispersion parameter and used the Firth-adjusted maximum likelihood estimation method. In cases with significant interactions, we used planned contrasts between toxic/non-toxic aphids for each H. axyridis treatment. To evaluate whether abundance differences in these groups might be a function of aphid density, we next regressed counts for each functional group against log-transformed aphid density, measured as the sum of counts from across the 10 day experiment (final aphid density for aphid mummy analysis). For functional groups that showed significant evidence of aphid density dependence (non-Harmonia ladybeetles, adult parasitoid wasps, mummies, ants; electronic supplementary material, table S3) we incorporated log(aphid density) into the model as an offset [39] and re-evaluated treatment effects with aphid toxicity (toxic, non-toxic) and H. axyridis presence (H. axyridis−, H. axyridis+) as fixed factors. In this way we were able to assess whether there were treatment effects beyond what could be accounted for by differences in aphid density among treatments.

(e). Effects of toxicity on a parasitoid wasp

Based on parasitoid mummy distribution in the field, we tested L. testaceipes parasitism of toxic versus non-toxic aphids in a controlled laboratory experiment. We infested each of 42 fava plants with 100 mixed-age Ap. craccivora from either a toxic or non-toxic clone. Aphid age structure was consistent between treatments, to avoid biases owing to parasitoid preference. Replicates were split across two experimental dates with 20 plants on the first date and 22 plants on the second. We caged the plants in plastic jars with mesh panels, and after 24 h added 1 female L. testaceipes to each cage. We removed wasps after 24 h; seven replicates were discarded because the wasp was not recovered or was found dead. After 10 days we counted the number of mummified aphids and compared between aphid toxicity treatments using a GLIM with Poisson distribution and log link function. Experimental date was initially included as a random factor in this analysis but subsequently removed when we found no significant effect. This model included an overdispersion parameter and used the Firth-adjusted maximum likelihood estimation method.

3. Results

(a). Adult Harmonia axyridis behaviour

Gravid H. axyridis females reduced usage of toxic aphid colonies, but did not completely eliminate exposure for their offspring. Adult H. axyridis beetles dispersed from plants with toxic or non-toxic aphids at a statistically indistinguishable rate over the first 6 h (electronic supplementary material, figure S2a; rep. measures MANOVA toxicity F_1,37 = 3.52, p = 0.068, time F_6,32 = 10.28, p < 0.001, time × toxicity F_6,32 = 1.10, p = 0.39). However, by 24 h toxic aphid patches retained 0.2 ± 0.1 beetles whereas non-toxic aphid patches averaged 1.2 ± 0.2 beetles (electronic supplementary material, figure S2a; t₃₇ = 4.02, p < 0.001). Beetles laid 2.4 times more eggs on non-toxic aphid patches than toxic, but still left an average of 8.5 ± 2.3 eggs in toxic patches (electronic supplementary material, figure S2b; GLIM log link: toxicity p = 0.017). Subsequent experiments using larvae, therefore, represent a realistic but conservative scenario; adult dispreference for toxic aphids would amplify effects demonstrated by the larval ladybeetle experiments.

(b). Coccinellid community effects in microcosms

The overall number of surviving coccinellids was higher in H. axyridis– treatments compared to H. axyridis+ treatments but was unaffected by aphid toxicity (figure 1a; GLIM logit link: toxicity p = 0.792, H. axyridis p < 0.001, toxicity × H. axyridis p = 0.414). However, when each species was analysed separately, aphid toxicity had strong effects on community composition.

Figure 1.

Mean ± s.e. (a) ladybeetle larval survival and (b) final aphid density from a caged competition experiment using toxic or non-toxic aphid clones and a substitutive design for ladybeetle community composition: either eight Coccinella septempunctata larvae (Harmonia axyridis−), or four Coc. septempunctata and four H. axyridis larvae (H. axyridis+). (a) Harmonia axyridis survival is lower on toxic aphid clones. Coccinella septempunctata survival is not affected by aphid toxicity in the absence of H. axyridis but drops in the presence of H. axyridis and non-toxic aphids. Asterisks indicate significant differences between treatments, n.s. indicates no significant difference at α = 0.05. (b) Toxic aphids have significantly higher density in cages with H. axyridis comprising part of the ladybeetle community. Different letters indicate significant differences at α = 0.05. (Online version in colour.)

As expected, H. axyridis larvae experienced significantly lower survival in microcosms with toxic aphids than non-toxic aphids (figure 1a; GLIM logit link: toxicity p < 0.001). Larvae that survived in toxic aphid cages never developed beyond first instar larvae. By contrast, Coc. septempunctata survival was not affected directly by aphid toxicity, with larvae showing normal development and high survival rates on toxic (89.1 ± 7.0%) and non-toxic (84.4 ± 4.4%) aphids (figure 1a) in the H. axyridis− treatments. In the presence of H. axyridis, however, aphid toxicity had a significant effect on Coc. septempunctata survival (GLIM logit link: toxicity p < 0.001, H. axyridis p = 0.067, toxicity × H. axyridis p < 0.001); survival of Coc. septempunctata larvae remained high in microcosms with toxic aphids (93.8 ± 4.1%), but was reduced to 34.4 ± 1.3% in the microcosms with non-toxic aphids. Healthy H. axyridis larvae were regularly observed feeding on Coc. septempunctata larvae within this treatment. Although both H. axyridis+ treatments initially had equal numbers of both species, non-toxic aphid microcosms favoured H. axyridis survival while toxic aphid microcosms favoured Coc. septempunctata.

This shift in coccinellid community interacted with aphid toxicity to affect final aphid densities (figure 1b; ANOVA: date F_3,25 = 40.09, p < 0.001, toxicity F_1,25 = 16.52, p < 0.001, H. axyridis F_1,25 = 3.57, p = 0.071, toxicity × H. axyridis F_1,25 = 6.34, p = 0.019). For the microcosms that had only Coc. septempunctata, there was no difference in final aphid density between toxic and non-toxic aphids. For the mixed predator microcosms (H. axyridis+), the outcome depended on aphid toxicity. Toxic aphids prevented H. axyridis development, and because of the substitutive design of the experiment, these aphids enjoyed lower ladybeetle predation pressure and had double the population at the end of the experiment compared to other treatments (figure 1b). By contrast, the non-toxic aphids primarily had H. axyridis instead of Coc. septempunctata by the end of the experiment, which served to depress aphid densities to the same levels as the Coc. septempunctata-only microcosms even though there were fewer surviving ladybeetles in the mixed predator microcosms.

(c). Effects of aphid toxicity in the field

Aphid defensive phenotype had significant effects on H. axyridis larvae, the population density of the aphids and other natural enemies. Beginning with H. axyridis, we observed significantly more larvae on the non-toxic clones than the toxic clones starting on day 3 of the experiment (figure 2a; rep. measures MANOVA toxicity F_1,33 = 8.0, p = 0.008, time F_10,24 = 8125.7, p < 0.001, time × toxicity F_10,24 = 11.2, p < 0.001). On days 3, 4 and 5, non-toxic aphid clones supported approximately 40% more larvae than toxic aphids (figure 2a). Most importantly, the developmental stages of these larvae differed (figure 2a). On plants infested with non-toxic aphids, average H. axyridis larval instar had progressed to the second instar by day 3, and the third instar by day 6, after which larvae rapidly dispersed from their plants. By contrast, H. axyridis larvae on plants infested with toxic aphids rarely developed beyond the first stadium; we observed only 13 second and eight third instar larvae on the toxic aphids across the entire experiment. Naturally occurring H. axyridis did occasionally colonize H. axyridis– plants (four larvae and 15 adults) but were removed to maintain treatments.

Figure 2.

Mean ± s.e. counts of (a) Harmonia axyridis larvae and (b) Aphis craccivora over time from an open field experiment, in which plants infested with either toxic or non-toxic aphids were initially infested with 20 first instar H. axyridis larvae (HA+) or not (HA−). Colour-coded numbers in panel (a) indicate the average H. axyridis larval instar at each time point on plants infested with toxic aphids or non-toxic aphids. Asterisks indicate significant density differences at each time point at α = 0.05. (Online version in colour.)

Aphid population densities diverged over the course of the experiment (figure 2b), such that by day 10, both H. axyridis and aphid toxicity treatments influenced aphid density in an interactive fashion (figure 3a; ANOVA: toxicity F_1,68 = 12.28, p < 0.001, H. axyridis F_1,68 = 29.62 p < 0.001, toxicity × H. axyridis F_1,68 = 4.41, p = 0.039). Overall, aphid densities were lower in the presence of H. axyridis, as might be expected, and by the end of the experiment were substantially (5.5 times) higher for toxic than non-toxic aphids in the presence of H. axyridis (figure 3a). However, in the absence of H. axyridis, toxic and non-toxic aphids were statistically indistinguishable from each other (figure 3a). The same pattern was evident when aphid densities were corrected for population size in bagged controls (ANOVA: toxicity F_1,68 = 6.77, p = 0.012, H. axyridis F_1,68 = 29.74, p < 0.001, toxicity × H. axyridis F_1,68 = 4.74, p = 0.033).

Figure 3.

Mean ± s.e. observational counts for (a) Aphis craccivora and different functional ecological groups of invertebrates (b–e) between fava plants stocked with toxic or non-toxic aphid clones and with or without Harmonia axyridis larvae. Functional ecological groupings include (b) coccinellids other than H. axyridis, (c) ants, (d) adult aphid parasitoid wasps and (e) developing parasitoids (aphid mummies). Observational counts for (b–d) represent total observations summed across the 10 day experiment. Aphid counts (a) and aphid mummies (e) represent the final count on day 10. Specific taxa classified into each grouping can be found in the electronic supplementary material, table S1. (Online version in colour.)

A diverse community of arthropods colonized the fava plants and their associated aphid colonies. We observed, not including Ap. craccivora, a total of 8645 invertebrates over the course of the experiment, 97% of which could be identified to at least the family level (electronic supplementary material, table S1). When classified into functional groups, we found significant differences among treatments for other ladybeetles, ants and parasitoids, each of which we will address in turn. We also compared counts of potential aphid predators (other than coccinellids, parasitoid wasps and ants) and herbivorous insects but found no effect of experimental treatments (electronic supplementary material, tables S2–S3).

For ladybeetles other than H. axyridis, we found an interactive effect between H. axyridis and toxicity treatments (figure 3b; toxicity , p = 0.081, H. axyridis p = 0.005, toxicity × H. axyridis p = 0.005). If H. axyridis was absent, there was no difference in abundance of other ladybeetles between aphid toxicity treatments ( p = 0.395), but if H. axyridis was present, more ladybeetles (of other species) were observed on the toxic than non-toxic aphids ( p = 0.004). This reduction in (other) ladybeetles on the non-toxic aphids was probably mediated in part by aphid density, which was lower in the H. axyridis+ non-toxic aphid treatment: other ladybeetles were significantly and positively associated with aphid density (electronic supplementary material, figure S3 and table S3; p = 0.003). However, when aphid density was incorporated into the model as an offset, we still found an interactive effect between H. axyridis and aphid toxicity treatments (electronic supplementary material, table S4; toxicity p = 0.167, H. axyridis p = 0.209, toxicity × H. axyridis p = 0.033), with lowered abundance of other ladybeetles on non-toxic than toxic aphids in the presence of H. axyridis ( p = 0.027), but not in the absence of H. axyridis ( p = 0.537). In communities where H. axyridis is present, toxic aphids were associated with higher abundance of other ladybeetle species, above and beyond what can be accounted for by aphid density.

Several genera of ant (electronic supplementary material, table S1) were observed tending aphids. Ants did not respond to differences in aphid toxicity but were significantly less abundant on plants that were stocked with H. axyridis larvae (figure 3c; electronic supplementary material, table S2; toxicity p = 0.840, H. axyridis p < 0.001, toxicity × H. axyridis p = 0.647). While ant density had a strong positive relationship with aphid density (electronic supplementary material, table S3; p < 0.001), incorporation of an aphid density offset in the model had no effect on the outcome (electronic supplementary material, figure S3; toxicity p = 0.840, H. axyridis p = 0.002, toxicity × H. axyridis p = 0.929).

Parasitoids were the most commonly observed aphid natural enemy (electronic supplementary material, table S1), 99% of which were a single species, L. testaceipes. Similar to (non-Harmonia) ladybeetles, there were significant differences in total observed counts of adult parasitoids among treatments (figure 3d; electronic supplementary material, table S2; toxicity p = 0.147, H. axyridis p = 0.002, toxicity × H. axyridis p = 0.044). This result was entirely attributable to the strong positive association between wasp abundance and aphid density (electronic supplementary material, figure S3 and table S3b; p < 0.001). When aphid density was included in the model, there were no significant treatment effects on adult parasitoid abundance (electronic supplementary material, table S4; toxicity p = 0.494, H. axyridis p = 0.965, toxicity × H. axyridis p = 0.165). Thus, based on adult parasitoid observations, it would appear that L. testaceipes should benefit from aphid toxicity as an escape from H. axyridis competition, but only insofar as mediated by aphid density.

The effect of the experimental treatments on parasitoids was very different when evaluating successful parasitism as measured by aphid mummies. Total L. testaceipes mummies produced on non-toxic/H. axyridis– plants were 8–16× higher than other treatments (figure 3e; electronic supplementary material, table S2; toxicity p = 0.147, H. axyridis p = 0.002, toxicity × H. axyridis p = 0.044). Few mummies were produced from the toxic aphids, regardless of whether H. axyridis was present or not. While mummies, unsurprisingly, were strongly positively correlated with final aphid density (electronic supplementary material, table S3; p < 0.001), inclusion of aphid density in the model had a relatively modest effect on the comparison among treatments (electronic supplementary material, figure S3 and table S4; toxicity p = 0.061, H. axyridis p = 0.003, toxicity × H. axyridis p = 0.033). In the absence of H. axyridis, we observed more mummies [mummies/log(aphids)] in the non-toxic than toxic treatment (χ²₁ = 28.42, p < 0.001), suggesting a direct effect of the toxic trait on parasitoid success.

(d). Effects of toxicity on a parasitoid wasp

Using a controlled laboratory bioassay, we confirmed our findings from the field that L. testaceipes has lowered success when parasitizing toxic Ap. craccivora clones. At 10 days following exposure to a gravid female wasp, we observed 70.2 ± 5.3 mummies in the non-toxic treatment, versus only 19.6 ± 1.8 in the toxic treatment (electronic supplementary material, figure S4; GLIM log link: toxicity , p < 0.001).

4. Discussion

Our results unequivocally demonstrate that subdominant (but less susceptible) predators can benefit from narrow-spectrum prey defence. Harmonia axyridis is a more efficient predator of aphids than many other natural enemies, and is also a notorious intraguild predator [30]. Our results verify that adult H. axyridis reduce usage of and deposit fewer eggs in association with toxic aphid patches [12], similar to findings for other ladybeetles and unsuitable prey [40], but do not avoid oviposition entirely. When toxic aphids killed or halted development of H. axyridis larvae, we then showed that less competitive ladybeetle species were able to exploit a niche that was unavailable to the dominant competitor, in both the laboratory and the field. This effect was probably mediated largely through aphid numbers, because H. axyridis larvae drew down non-toxic aphid populations faster than toxic colonies, thus toxic aphids were a more abundant resource for other coccinellids [41,42]. However, density dependence was not the entire explanation for the toxicity effect: coccinellid numbers remained significantly higher on toxic than non-toxic aphids even when accounting for aphid population size in the statistical model. Speculatively, it is possible that indirect effects mediated by other members of the community, such as ants, may have played a role [43,44]. Regardless of mechanism, toxic aphids were an exploitable resource for subdominant coccinellids, upon which competition and intraguild predation from a dominant species was reduced. In natural populations, the magnitude of this effect is probably amplified even further by adult H. axyridis dispreference for toxic aphids. Globally, H. axyridis is a major invasive species [30], and the distribution of ‘refuge’ prey resources that are unavailable to H. axyridis, such as toxic Ap. craccivora, may be key to understanding the persistence of other coccinellids in face of this invasive threat.

Our results also demonstrate that for prey, the benefit of a narrow-spectrum defence is contingent on community context. When H. axyridis was present in the community, the toxic trait mattered for aphid populations, and toxic aphids retained higher population densities than non-toxic aphids in both laboratory and field studies. This result provides a clear demonstration of the efficacy of a putative defence, which is often difficult to evaluate in other predator–prey systems, particularly if they have had long coevolutionary histories with one another [1,2]. When H. axyridis was absent from the community, however, there was no difference in the performance of toxic versus non-toxic aphid populations, and the narrowly defended toxic aphids did not enjoy a fitness benefit from their trait. Consequently, narrow-spectrum defences will only provide a selective advantage to their possessors when susceptible enemies constitute an important component of the community.

In contrast with subdominant coccinellids, an important parasitoid did not benefit from the aphid's narrow-spectrum defence. Adult L. testaceipes wasps were abundantly and equivalently observed in association with both toxic and non-toxic aphid colonies in our field experiment, but produced only a fraction as many mummies on toxic as non-toxic aphids, a pattern validated in our subsequent laboratory experiment. This suggests a type of anti-parasitoid physiological defence similar to the defensive symbioses some aphids have with bacteria [45], but facultative bacterial endosymbionts were not causative in our experiments because we used aphids that lacked these symbionts. A similar innate (non-symbiont) defence to parasitoids has been described in pea aphids [11]. In Ap. craccivora, the most parsimonious explanation is that the same aphid defensive mechanism affects both H. axyridis and the parasitoid; however, unrelated but correlated mechanisms are also possible [12]. These same toxic strains of Ap. craccivora sometimes receive high levels of parasitism by L. testaceipes in the field [46], suggesting either that some toxic strains are not defended against parasitoids, or that the parasitoid also exhibits substantial intraspecific variation in susceptibility, perhaps as a coevolutionary response to variation in prey defence [47].

Heritable variation in a defensive trait provides the opportunity for eco-evolutionary feedbacks between prey and their enemies. Intraspecific variation is common, and can have substantial community consequences [48,49]. When such variation is heritable, it provides the raw material for natural selection. The defensive composition of prey populations will evolve as a function of enemy community context, leading to spatial and temporal structuring of prey defensive arsenals [10,50,51]. The enemy community, in turn, may respond to the distribution of prey defensive traits [47], potentially subdividing and exploiting differentially defended individuals. Thus, intraspecific variation in defence could contribute to maintenance of biodiversity in higher trophic levels.

In conclusion, our results demonstrate: (i) that certain predators can and do exploit niches generated by narrow-spectrum heritable defences of prey, (ii) that these niches can be fine-grained, corresponding to only a subset of the prey population, and (iii) that the defensive efficacy of narrow-spectrum traits is contingent on enemy community composition. Our findings suggest that dynamic eco-evolutionary feedbacks between generalist predator communities and prey defensive traits may be ongoing, which has the potential to profoundly influence the structuring of predator communities and their resilience to disruption by invaders.

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material (electronic supplementary material, files S2–S5).

Authors' contributions

P.L., K.J. and J.W. collaborated to design the experiments; P.L. and K.J. collected and analysed data, all authors drafted and revised the manuscript.

Competing interests

We have no competing interests.

Funding

This work was supported by the National Institute of Food and Agriculture, U.S. Department of Agriculture, under Awards no. 2014-67013-21576 and Hatch no. 0224651, as well as the University of Kentucky.