Bombardiers and assassins: mimetic interactions between unequally defended insects

In defensive mimicry, resemblance between unequally defended species can be parasitic; this phenomenon has been termed quasi-Batesian mimicry. Few studies have used real co-mimics and their predators to test whether the mimetic interactions were parasitic. Here, we investigated the mimetic interaction between two well-defended insect species, the bombardier beetle Pheropsophus occipitalis jessoensis (Coleoptera: Carabidae) and the assassin bug Sirthenea flavipes (Hemiptera: Reduviidae), using their potential predator, the pond frog Pelophylax nigromaculatus (Anura: Ranidae), which coexists with these insect species in the same habitat in Japan. We observed behavioural responses of this frog species (adults and juveniles) to adult Ph. occipitalis jessoensis and adult S. flavipes under laboratory conditions. Among the frogs, 100% and 75% rejected Ph. occipitalis jessoensis and S. flavipes, respectively, suggesting that, compared with the assassin bug S. flavipes, the bombardier beetle Ph. occipitalis jessoensis is more well-defended against frogs. An assassin bug or a bombardier beetle was provided to a frog that had encountered the other insect species. Frogs with a history of assassin bug encounter demonstrated a lower rate of attack toward bombardier beetles. Similarly, frogs with a history of bombardier beetle encounter demonstrated a lower rate of attack toward assassin bugs. Therefore, both the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes benefit from the mimetic interaction.

To elucidate the nature of the mimetic interaction between bombardier beetles and assassin bugs, we observed behavioural responses of the frog Pe. nigromaculatus to Ph. occipitalis jessoensis and S. flavipes under laboratory conditions. Specifically, we compared rates of rejection by Pe. nigromaculatus between Ph. occipitalis jessoensis and S. flavipes. To determine whether Ph. occipitalis jessoensis and/or S. flavipes benefits from the mimetic interaction, we experimentally investigated whether a frog would attack an insect species after it had encountered the other insect species. The rate of attack by frogs that had encountered bombardier beetles (or assassin bugs) was compared with the rate of attack by frogs that had not encountered the indicated species. Finally, we discuss the adaptive significance of mimetic interactions between bombardier beetles and assassin bugs.

Sampling
Forty-three adults of the bombardier beetle Ph. occipitalis jessoensis were collected from Honshu (Hyogo, Shiga, and Shimane Prefectures) in July-September 2020 and May-October 2021. Adult beetles were housed separately in plastic cases (diameter: 85 mm; height: 25 mm) under laboratory conditions (25 • C; cf. Sugiura, 2018;Sugiura & Sato, 2018;Sugiura, 2021;Sugiura & Date, 2022) and fed dead larvae of Spodoptera litura (Fabricius) (Lepidoptera: Noctuidae) (cf. Sugiura, 2018;Sugiura & Sato, 2018;Sugiura, 2021;Sugiura & Date, 2022). Prior to experiments, body length and weight were measured to the nearest 0.01 mm and 0.1 mg using electronic slide callipers (CD-15AX, Mitsutoyo, Kawasaki, Japan) and an electronic balance (CPA64, Sartorius Japan K.K., Tokyo, Japan), respectively ( Table 1). The same beetles were not used for multiple experiments. Twenty adults of the assassin bug S. flavipes were collected from Honshu (Hyogo and Shimane Prefectures) in August-October 2021. Adult bugs were housed separately in plastic cases (diameter: 85 mm; height: 25 mm) under laboratory conditions (25 • C). Prior to experiments, body length and weight were measured to the nearest 0.01 mm and 0.1 mg using electronic slide callipers and an electronic balance, respectively (Table 1). Adults and nymphs of the mole cricket G. orientalis were provided as prey (cf. Hayashi, 2023). Some assassin bugs were used repeatedly in different experiments.

Experiment 1: initial response
We used the predator Pe. nigromaculatus to test whether the bombardier beetle Ph. occipitalis jessoensis or the assassin bug S. flavipes is better defended under laboratory conditions. In accordance with the method established by Sugiura (2018), we experimentally investigated behavioural responses of Pe. nigromaculatus to Ph. occipitalis jessoensis and S. flavipes in the laboratory (Graduate School of Agricultural Science, Kobe University) between September 2021 and October 2021. We used frogs that had fed on mealworms >24 h before experiments. First, a frog was placed in a plastic cage (length: 120 mm; width: 85 mm; height: 130 mm). Next, a bombardier beetle (or an assassin bug) was transferred to the cage containing the frog. The behaviours of the frog and the bombardier beetle (or assassin bug) were recorded using a digital camera (iPhone 12 Pro Max; Apple Inc., Cupertino, CA, USA) and a digital video camera (Handycam HDR-PJ790V, Sony, Tokyo, Japan). The footage of recorded behaviour was reviewed to investigate how each insect could defend. The bombing sounds of bombardier beetles were checked to investigate whether bombing forced the frogs to reject the beetles. Stabbing by assassin bugs was investigated to determine whether stabbing forced the frogs to reject the bugs. When a frog did not attack a bombardier beetle or an assassin bug within 2 min, we considered it to have ignored the insect. When a frog swallowed a bombardier beetle or an assassin bug, we observed whether the frog subsequently vomited the insect (cf. Sugiura & Sato, 2018;Sugiura, 2018;Sugiura & Date, 2022). When the frog did not vomit the insect, we inferred that the frog had digested the insect. In Experiment 1, we used 20 bombardier beetles, 20 assassin bugs, and 40 frogs. The body size (snout-vent length and body weight) of frogs that attacked bombardier beetles did not significantly differ from the body size of frogs that attacked assassin bugs (t -test; snout-vent length, t = −0.77455, df = 37.923, P = 0.4434; body weight, t = 0.98556, df = 35.179, P = 0.3311). The same individuals of bombardier beetles, assassin bugs, and frogs were not used repeatedly in this experiment. The sample size was determined based on the number of assassin bugs collected in this study. Experiment 1 was part of the following experiment; specifically, the initial responses of 35 frogs observed in Experiment 2 were used as the data for Experiment 1.

Experiment 2: generalisation tests
We experimentally investigated the interaction between the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes via the potential predator Pe. nigromaculatus under laboratory conditions. Specifically, we investigated how a frog responded to a bombardier beetle or an assassin bug after the frog had encountered the other insect species (i.e., generalisation test; Fig. 3). The same plastic cages and video cameras (see Experiment 1 for details) were used in this experiment. A bombardier beetle was provided to a frog that had encountered (attacked or ignored) an assassin bug (n = 23; Fig. 3A). We tested whether the frog attacked or ignored the bombardier beetle approximately 6 min (median: 6 min; range: 5-14 min) after the frog had encountered an assassin bug. The rate of attack on bombardier beetles by frogs that had encountered assassin bugs was compared to the rate of attack by frogs that had not encountered assassin bugs. Similarly, an assassin bug was provided to a frog that had encountered a bombardier beetle (n = 20; Fig. 3B). We tested whether the frog attacked or ignored the assassin bug approximately 6 min (median: 6 min; range: 5-7 min) after the frog had encountered a bombardier beetle. The rate of attack on assassin bugs by frogs that had encountered bombardier beetles was compared to the rate of attack by frogs that had not encountered bombardier beetles. Although the duration used in our generalisation tests (5-14 min) was shorter than the generalisation time of a spider (1 h; Raška et al., 2020) and the memory retention time of a bird (35 days; Kojima & Yamamoto, 2020), our field observations (Fig. 1) suggest that pond frogs frequently encounter bombardier beetles and assassin bugs under field conditions. Survivals of bombardier beetles, assassin bugs, and frogs used in this study were checked within 24 h after experiments. When a frog died within 24 h after the experiment, we dissected the frog to investigate the cause of death. In Experiment 2, we used 43 bombardier beetles, 17 assassin bugs, and 43 frogs. The same individuals of bombardier beetles and frogs were not used repeatedly in this experiment. The initial behavioural responses of 35 frogs observed in this experiment were also used as the data for Experiment 1. In addition, eight frogs were exclusively used in Experiment 2.
All experiments were performed in accordance with Kobe University Animal Experimentation Regulations (Kobe University's Animal Care and Use Committee, No. 30-01). Only one pair of insect species was provided to each frog to minimise the negative  impacts of well-defended insects on frogs. Healthy frogs were released after the experiments had been completed. No frogs were euthanised in this study.

Data analysis
All analyses were performed using R version 3.5.2 (R Core Team, 2018).
Fisher's exact test was used to compare the rates of rejection by frogs between bombardier beetles and assassin bugs. Welch's t -test was used to compare the body size (body length and weight) of bombardier beetles and assassin bugs; it was also used to compare the body size (snout-vent length and body weight) of pond frogs that attacked bombardier beetles and assassin bugs. A generalised linear mixed model (GLMM) with a binomial error distribution and a logit link was used to investigate the effects of insect species and frog encounter history on the rate of attack by frogs. The frog response (attack, 1; or ignore, 0) was used as a response variable. The insect species (the bombardier beetle Ph. occipitalis jessoensis or the assassin bug S. flavipes), frog encounter history (an initial response or a response after encountering the other species), and the interaction between insect species and frog encounter history were used as fixed factors. Individual assassin bugs and frogs were used as random effects. The GLMM was conducted using the lme4 package version 1.1.19 in R (Bates et al., 2015). A significance threshold of 0.05 was used for all statistical tests.
The rate of bombardier beetle rejection by frogs (100.0%) significantly differed from the rate of assassin bug rejection by frogs (75.0%; Fisher's exact test, P = 0.0471). The body size (body length and weight) significantly differed between bombardier beetles and assassin bugs; bombardier beetles were shorter but heavier than assassin bugs (t -test; body

Notes.
Values: numbers of frogs. a Frog behaviour: Swallow: frogs successfully swallowed beetles (or bugs). Spit out: frogs spat out beetles (or bugs) immediately after capturing the insects in their mouths (frogs were bombed or stabbed). Stop attack: frogs stopped their attacks immediately after their tongues had contacted beetles (or bugs). Ignore: frogs did not attack beetles (or bugs).
A history of assassin bug encounter reduced the rate of attack on bombardier beetles by frogs from 75.0% to 21.7% (Fig. 7), although the rate of rejection by frogs did not change (100%; Fig. 8A). A history of bombardier beetle encounter reduced the rate of attack on assassin bugs by frogs from 91.3% to 40.0% (Fig. 7); the rate of rejection increased from 78.3% to 100.0% (Fig. 8B). The GLMM analysis showed that frog encounter history had a significant influence on the rate of attack by frogs, although insect species and the  interaction between insect species and frog encounter history did not have significant effects on the rate of attack (Table 4).

Survival
None of the bombardier beetles or assassin bugs that successfully defended against frogs died within 24 h after the experiments. Of the frogs (n = 48), one (2.1%) died within 24 h (4 h) after swallowing an assassin bug; a dead assassin bug was found in the stomach of the dead frog. Other frogs (97.9%) were not harmed by our experiments.

DISCUSSION
There has been controversy regarding whether mimetic interactions between unequally defended species are parasitic (Speed et al., 2000;Rowland et al., 2007;Rowland et al., 2010;  Swallow: frogs successfully swallowed beetles (or bugs). Spit out: frogs spat out beetles (or bugs) immediately after capturing the insects in their mouths (frogs were bombed or stabbed). Stop attack: frogs stopped their attacks immediately after their tongues had contacted beetles (or bugs). Ignore: frogs did not attack beetles (or bugs). Photo credit: Shinji Sugiura.
Full-size DOI: 10.7717/peerj.15380/ fig-8 Aubier, Joron & Sherratt, 2017). In the present study, we showed that both the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes were well-defended against the frog Pe. nigromaculatus (Fig. 4). In generalisation tests (Experiment 2), frogs with a history of assassin bug encounter attacked bombardier beetles less frequently compared with frogs that had no such encounter history (Figs. 3A, 7 and 8A). Similarly, frogs with a history of bombardier beetle encounter attacked assassin bugs less frequently compared with frogs that had no such encounter history (Figs. 3B, 7 and 8B). These results suggest that both the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes benefit from the mimetic interaction in terms of defence against the potential predator Pe. nigromaculatus.
Although Ph. occipitalis jessoensis demonstrated superior defensive abilities compared to S. flavipes (Figs. 4 and 8), the mimetic interaction between these two unequally defended species may be mutualistic, rather than parasitic.

Frogs as predators
Frogs have been frequently used as predators to investigate the effectiveness of anti-predator defences in insects (Taniguchi et al., 2005;Ito, Taniguchi & Billen, 2016;Matsubara & Sugiura, 2017;Sugiura, 2018;Shinohara & Takami, 2020;Sugiura, 2020a;Sugiura, 2020b;Sugiura & Date, 2022;Sugiura & Tsujii, 2022). In this study, we used pond frogs as predators of bombardier beetles and assassin bugs to show that the frog species Pe. nigromaculatus was unable to distinguish between the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes. This finding could be explained by the adaptive generalisation in predators, where they learn to recognise dangerous prey by generalising the appearance of previously encountered prey to that of subsequently encountered prey (Ruxton et al., 2008).
In the Experiment 2, the time used in the generalisation tests ranged from 5-14 min. The time used in previous memory and generalisation tests (1 h-35 days;Ito, Taniguchi & Billen, 2016;Kojima & Yamamoto, 2020;Raška et al., 2020) was longer than the time of our generalisation tests. For example, the tree frog Dryophytes japonica (Günther) reportedly retains the memory of unpalatable prey for at least 1 day (Ito, Taniguchi & Billen, 2016). In addition, the Pe. nigromaculatus individuals used in this study were collected from the study sites where both Ph. occipitalis jessoensis and S. flavipes were found, suggesting that some individuals of Pe. nigromaculatus may have already experienced Ph. occipitalis jessoensis and/or S. flavipes at the sites prior to our experiments. Therefore, the use of short experimental durations and wild-collected individuals may have influenced the results of our experiments in Pe. nigromaculatus. Further studies are needed to investigate detailed memory retention in Pe. nigromaculatus.
Some predators have evolved counter defences, such as specific skills to avoid welldefended prey by detecting toxic chemicals or recognising warning signals (Edmunds, 1974;Endler, 1991;Ruxton, Sherratt & Speed, 2004;Skelhorn & Rowe, 2006;Williams et al., 2010). In the present study, 50% and 40% of frogs stopped attacking Ph. occipitalis jessoensis and S. flavipes before they had been bombed and stabbed, respectively (Fig. 4). Because Pe. nigromaculatus individuals stopped attacking immediately after their tongues had contacted these insects, this frog species may quickly detect deterrent characteristics on the body surfaces of Ph. occipitalis jessoensis and S. flavipes with its tongue; this enables avoidance of damage (Sugiura, 2018). Such reactions to well-defended prey have been reported in other predators such as tree frogs (Ito, Taniguchi & Billen, 2016) and quails (Kojima & Yamamoto, 2020).

Bombardier beetles as models and mimics
Bombardier beetles can chemically defend themselves against various types of predators (Eisner, 1958;Eisner & Meinwald, 1966;Eisner & Dean, 1976;Dean, 1980;Eisner et al., 2006;Bonacci et al., 2008;Sugiura & Sato, 2018;Sugiura, 2018;Kojima & Yamamoto, 2020;Sugiura, 2021;Sugiura & Date, 2022). Many bombardier beetle species have aposematic body colour patterns that advertise their toxicity to predators (Schaller et al., 2018;Anichtchenko et al., 2022). Therefore, bombardier beetles are visually mimicked by distantly related insects that coexist with them in the same habitats (Shelford, 1902;Bonacci et al., 2008;Kojima & Yamamoto, 2020). However, very few studies have elucidated the nature of mimetic interactions that include bombardier beetles. In the present study, we used the frog Pe. nigromaculatus as a potential predator to investigate that the mimetic interaction between the bombardier beetle Ph. occipitalis jessoensis and the assassin bug S. flavipes. We found that a history of encounter with Ph. occipitalis jessoensis reduced the rate of attack on S. flavipes (Fig. 8B), suggesting that the coexistence with Ph. occipitalis jessoensis is beneficial for S. flavipes. However, Ph. occipitalis jessoensis consistently repelled the frog Pe. nigromaculatus in our study (Figs. 4 and 8A). Therefore, a history of encounters with the assassin bug S. flavipes may not benefit Ph. occipitalis jessoensis. Nevertheless, the mortality risk from frog attacks is not zero, as a previous study reported that 3.6% of the frog Pe. nigromaculatus successfully ate Ph. occipitalis jessoensis (Sugiura, 2018). Thus, the coexistence with the assassin bug S. flavipes is beneficial for the bombardier beetle Ph. occipitalis jessoensis, although the mutualistic interaction between these two insect species may be asymmetric.
In this study, we showed that the frog species Pe. nigromaculatus frequently rejected the assassin bug S. flavipes (Figs. 4 and 8). Although some frogs successfully swallowed S. flavipes individuals (Figs. 4 and 8B), one frog died 4 h after a successful swallowing event. These results suggest that S. flavipes venom is sufficiently strong to repel predators. In addition, some frogs stopped attacking assassin bugs immediately after their tongues had contacted the bugs (Fig. 4), suggesting that chemicals on the body surfaces of S. flavipes act as a deterrent to Pe. nigromaculatus.
The distribution of the assassin bug S. flavipes overlaps with the distribution of the bombardier beetle Ph. occipitalis jessoensis in East and Southeast Asia (Chłond, 2018;Fedorenko, 2021). However, S. flavipes is also found in South and West Asia where Ph. occipitalis jessoensis is not distributed (Chłond, 2018;Fedorenko, 2021). In the assassin bug S. flavipes, the body colour pattern of the South and West Asian populations partially differs from the body colour pattern of the East and Southeast Asian populations; the pronotum of the South and West Asian populations is redder than the pronotum of the East and Southeast Asian populations, although both types share the black and yellow pattern on other body parts (Chond, Bugaj-Nawrocka & Sawka-Gadek, 2019). Notably, the body colour pattern of South and West Asian S. flavipes is very similar to the body colour pattern of another bombardier beetle, Pheropsophus (Stenaptinus) catoirei (Dejean); adult Ph. catoirei individuals have a reddish head and pronotum (Fedorenko, 2021). Pheropsophus catoirei, which is closely related to Ph. occipitalis jessoensis, shares its distribution area (South and West Asia) with S. flavipes in East and Southeast Asia (Chond, Bugaj-Nawrocka & Sawka-Gadek, 2019;Fedorenko, 2021). Therefore, the mimetic partner of S. flavipes could differ between East-Southeast Asia and West-South Asia.

CONCLUSIONS
Some aposematic species form 'mimicry rings' (Kunte, Kizhakke & Nawge, 2021;Chatelain et al., 2023). Mimicry rings are composed of at least two Müllerian co-mimics or one aposematic species plus one Batesian mimic (Kunte, Kizhakke & Nawge, 2021); the smallest mimicry rings include only two species (Kunte, Kizhakke & Nawge, 2021), while the largest mimicry rings include >100 species (Pekár et al., 2017). Although many studies have investigated mimicry rings that are composed of closely related taxa (Kunte, Kizhakke & Nawge, 2021), fewer studies have focused on mimicry complexes that involve distantly related taxa (Linsley, Esiner & Klots, 1961;Pekár et al., 2017). Our results suggest that the bombardier beetle Ph. occipitalis jessoensis and the distantly related S. flavipes form a multi-order mimetic complex. Other insects such as the rove beetle species Ocypus weisei Harold (Coleoptera: Staphylinidae) may be included in the mimetic complex; O. weisei adults share a similar microhabitat (on the ground in grassland), body colour pattern (black and yellow pattern), and movement pattern with Ph. occipitalis jessoensis and S. flavipes adults in Japan. Furthermore, the black and yellow body colour pattern of these insects is similar to the typical aposematic colour of stinging hymenopteran insects, such as paper wasps and bees (Chatelain et al., 2023). The presence of a colour pattern similar to a typical aposematic pattern likely provides S. flavipes and Ph. occipitalis jessoensis with more robust protection from predators, compared with other colour patterns.