Vibrational communication between a myrmecophilous butterfly Spindasis lohita (Lepidoptera: Lycaenidae) and its host ant Crematogaster rogenhoferi (Hymenoptera: Formicidae)

Ants are a dominant insect group in terrestrial ecosystems and many myrmecophilous species evolve to associate with ants to gain benefits. One iconic example is myrmecophilous butterflies that often produce ant-mimicking vibrational calls to modulate ant behaviors. Despite its popularity, empirical exploration of how butterflies utilize vibrational signals to communicate with ants is scarce. In this study, we reported that the myrmecophilous butterfly Spindasis lohita produce three types of larval calls and one type of pupal call, while its tending ant, Crematogaster rogenhoferi emit a single type of call. The results of discriminant analysis revealed that calls of the two species are quantitatively similar in their signal attributes; the potential role of butterfly calls are further confirmed by the playback experiments in which certain ant behaviors including antennation, aggregation, and guarding were induced when one of the butterfly calls was played to C. rogenhoferi workers. The findings in the current study represent the very first evidence on vibrational communication between Spindasis and Crematogaster and also imply that S. lohita may have been benefited from ant attendance due to the ability to produce similar calls of the ant C. rogenhoferi.


Results
Vibrational behavior and signals. The larvae of S. lohita produced three types of calls (Type A, B, and C) and pupae produced one type of call (Table 1, Figs. 1 and 2, also see Supplementary Audio S1-4). The larvae produced calls during moving, resting and even eating. The larvae emitted Type A calls constantly, and sometimes produced Type B calls. The type A call (N = 6, Σ = 82 pulses) consisted of a long pulse train, which resembles to rapid drumming, while Type B calls consisted of a single pulse (N = 4, Σ = 45 pulses) that resembles to a grunt. Type C calls also consisted of a single pulse (N = 6, Σ = 61 pulses) and occurred randomly. Type C calls were recorded between Type A and B calls intermittently. Despite no obvious abdominal segment contraction, pupae (N = 4, Σ = 88 pulses) emitted calls (Fig. 2) immediately once they were properly settled for the recording. Pupae calls resemble to finger snaps and occurred constantly.  Table 1. Call characteristics of Spindasis lohita and Crematogaster rogenhoferi (mean ± SD). Characters were compared using the Kruskal-Wallis test. Means followed by different lowercase letters in the same column indicate significant differences among different calls.
The ant workers produced calls by lifting their gaster and rubbing it against the postpetiole. During the recording sessions, the ants only made calls when being disturbed (tapping on the leaves) or when discovering food. Since tapping the leaves would incur extra noise and lead to stylus instability, mealworms were instead provided as food resource on the leaves to stimulate ant calls. Only a single type of call was observed in the workers (N = 3, Σ = 62 pulses), and consisted of two parts: part 1 occurred constantly and is sometimes followed by part 2 (Fig. 3, see Supplementary Audio S5). Both parts resemble to rubbing plastic material by human hands (Table 1, Fig. 3). No significant difference was found in dominant frequency (Mann-Whitney U test, U = 467.5, P = 0.949) and pulse period (Mann-Whitney U test, U = 412.5, P = 0.398) between part 1 and part 1 + 2 of ant calls. The duration of part 1 + 2 was significantly longer than that of part 1 (Mann-Whitney U test, U = 6, P < 0.001). It is worth noting that we failed to record any call signals emitted by queens despite the same recording effort made to queens and the presence of a stridulation apparatus also in queens (see further).
Type A and Type C calls possessed the highest and lowest dominant frequency, respectively, among the five recorded call types ( The result of discriminant analysis on the three call characteristics (dominant frequency, pulse duration and pulse period) of S. lohita and C. rogenhoferi is shown in Fig. 4. The call (part 1 + 2) of C. rogenhoferi overlapped with the Type A call of S. lohita, whereas Type B calls and pupal calls of S. lohita were grouped together with the call (part 1 alone) of C. rogenhoferi. Correct grouping in the discriminant analysis of these signals was 68.32%. All these data suggest that the vibrational characteristics of the lycaenid calls and the ant calls are similar.
Playback experiments. Among all the experiments, attack behavior was not observed. The occurrence of ant antennation was positively correlated with playback of larval Type A calls and pupal calls (Fig. 5, see Supplementary Table S1). The positive correlation of ant aggregation occurred with all six playback signals, especially for the butterfly's calls (three larval calls and a pupal call) (Fig. 5, see Supplementary Table S1). The occurrence of ant guarding was positively correlated with all playback signals except ant signals (Fig. 5, see Supplementary Table S1). Type A calls and pupal calls of S. lohita positively induced three ant behaviors (aggregation, antennation and guarding behavior), while Type B calls and Type C calls of S. lohita were positively correlated with ant aggregation and guarding behavior. Ant calls had only one positive correlation with ant aggregation.
Stridulatory structure of the ants. Scanning microscopy showed the existence of a prominent stridulation apparatus on the mesodorsal part of the first gastral segment on worker and queen of C. rogenhoferi (Fig. 6). This consists of a stridulation file with parallel transverse ridges on the anterior part of the first gastral tergite and a scraper that is formed by the posterior and slightly downward-bent edge of the postpetiole (Fig. 6B,C). The stridulation signal is produced by moving the scraper over the file, which is achieved by the action of two antagonistic muscles. Both in the worker and the queen, these consist of two paired laterally located dorsoventral muscles and an unpaired centrally located longitudinal muscle (Fig. 6D,F).  www.nature.com/scientificreports www.nature.com/scientificreports/ The full size of the stridulation file, however, is usually not visible on scanning micrographs as its anterior portion is hidden underneath the postpetiole. We therefore performed histological examination to view the entire stridulation file. This revealed that queens had a larger stridulation file with more ridges and wider spacing of the ridges than workers (Fig. 6E,G; Table 2).

Discussion
We found that the lycaenid S. lohita produce a total of four types of call including three larval calls and one pupal call, while the ant C. rogenhoferi produces one type of call (Table 1; Figs. 1-3). With playback experiments and behavioral assays, this study demonstrated the vibrational communication between Spindasis and Crematogaster for the first time (Fig. 5). Furthermore, the discriminate analysis also revealed the similarity between the calls of S. lohita and C. rogenhoferi in their signal attributes (Fig. 4), echoing the results of playback experiments that ant benevolent behaviors were induced significantly by different types of butterfly calls.
To assess the potential role of butterfly calls, we compared the calls of butterflies and ants, and performed the playback experiments. The discriminate analysis (Fig. 4) indicates that the pupal call and larval Type B calls of S. lohita adjoining with the ant call (part 1) form a group, while Type A and the ant call (part 1 + 2) overlap to some extent, implying that butterfly calls are similar to ant calls in call attributes. Playback of ant calls and all types of butterfly calls induce ant aggregation, further supporting that calls of both lycaenids and ants share a similar  www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ function and that the lycaenids are capable of producing similar calls of the ant possibly for the purpose of inducing the ant's benevolent behavior 18,28,29 . Type A call of larvae and pupal call of S. lohita frequently occurred during our observation and invariably induced one of three ant's benevolent behavioral reactions (antennation, aggregation, and guarding). This suggests that the two calls of S. lohita may serve as a primary call for C. rogenhoferi. Type B and Type C calls of S. lohita, however, occurred occasionally and only induce two types of ant responses (e.g., aggregation and guarding behavior), leading to a possibility of these calls likely functioning as communication cues with non-host recipients such as conspecific larvae or other co-inhabiting organisms 30 in addition to communicating with the host ant. Despite variations in ant responses to the calls of the lycaenid, our results strongly indicate that vibrational cues play a significant role in the interactions between S. lohita and its attending ants.
It is unexpected that the playback of ant calls failed to induce strong behavioral reactions in ant workers. This may be explained by the fact that communication in ants typically involves both chemical and physical cues, and that vibrational calls alone may not be sufficient or intimate enough to produce a strong behavioral response. Equally unexpected is that no queen calls were recorded during the entire experiment, which prevents us from playing back queen's calls. While it remains unclear whether a C. rogenhoferi queen can make calls, we did observe a functional stridulatory apparatus in queens that is structurally similar to that of workers (the only difference is the spacing between ridges, Table 2), implying that the C. rogenhoferi queen is certainly capable of producing vibrational signals. One explanation is that the environmental conditions while this study was being conducted may not be favorable in inducing calling behavior of the C. rogenhoferi queen. To our surprise, in playback experiments white noise was able to induce ant's benevolent behavior. One of the possible explanations is that ants are generally rather sensitive to vibration, especially to unfamiliar vibrational signals 1,31,32 .
How S. lohita larvae make calls is still ambiguous thus far. Species in Riodinidae are known to produce signals using vibratory papillae, while species in Lycaenidae lack this structure 21 . It is believed that vibrational signals of lycaenid caterpillars may emanate from muscular contraction and air compression through the tracheae 33 . Our behavioral observation revealed that the larvae of S. lohita produced calls under any circumstances, and sometimes produced two types of call simultaneously. No apparent movements were observed in outer larval appearance while signals were being produced, thus leading us to speculate muscular contraction as a more likely signal-producing mechanism for S. lohita. Pupae of S. lohita produced calls from time to time in any condition while being recorded. Pupal calls are thought to have been generated from tooth-and-comb stridulatory organs between the fifth and sixth segments of abdomen in members of Riodinidae and Lycaenidae 34,35 . As no stridulatory movements of pupae while producing signals, future study should focus on the identification of the pupal vibrational organ and how pupal calls sustain high ant maintenance.
The stridulation apparatus of the ants represents the typical structural features as in other stridulating ants 36,37 . This consists of a file with transverse cuticular ridges on the anterior portion of the first gastral tergite, and a scraper that is formed by the posterior downward-bend posterior edge of the postpetiole 38 . Signal production is realized by moving the gaster up and down, which is achieved by the alternating action of two antagonistic muscle groups in the postpetiole: the paired dorsoventral muscles upon contraction pull the sternal part of the gaster forward and hence the tergal part backward, which results in one chirp. Contraction of the longitudinal muscle, on the other hand, pulls the tergal part of the gaster forward again and thus acts as the antagonist of the dorsoventral muscles [36][37][38][39] . Repeated stridulation will be the result of the alternating action of both muscles. These muscles represent the common muscular outfit that is known in the postpetiole of other ants, with the paired dorsoventral muscles corresponding with muscles 9 and 10, and the unpaired longitudinal muscle corresponding with muscle 8 in the study by Hashimoto 39 .
The finding of larvae of S. lohita producing three types of call is similar to what has been reported for J. evagoras 22 , although J. evagoras produce two types of pupal calls. However, the call attributes (both frequency and durations) differ in the two species. In contrast, Maculinea rebeli, as reported in Barbero,et al. 18 , produce only one type of larval call and one pupal call. Differences in call diversity and characteristics between S. lohita and other butterflies are in parallel with Riva, et al. 24 that compared vibrational signals of 12 species from eight lycaenid genera (Cacyreus, Lycaena, Cupido, Lycaeides, Scolitantides, Plebejus, Maculinea, and Polyommatus) and concluded that the calls of Lycaenidae are species-specific, even for those of the same genus. Such differences may also hint that vibrational signals of caterpillars would serve as an applicable tool in species delimitation. Similarly, the temporal and spectral properties of vibrational signals of C. rogenhoferi differ from those of other ant species 18,40 . Ferreira, et al. 40 reported that vibrational signals can be used in discovering cryptic species of ants as ants recognize conspecific signals and refuse heterospecific signals 41 . Combined with all empirical evidence, one may utilize vibrational pattern as a systematic character for taxa such as butterfly and ant. The practice of vibrational signals in α-taxonomy can also be seen in other insect taxa, such as treehoppers 42 , lacewings 43 , and psyllids 44 . www.nature.com/scientificreports www.nature.com/scientificreports/ In Taiwan, there are two other myrmecophilous butterfly species that are closely related to S. lohita, namely S. syama and S. kuyanianus. Both species are believed to be obligate mutualists with their attendant ants, C. popohana and C. amia for the former, and C. laboriosa for the latter. This system provides a great opportunity to test interspecific competition for the vibrational niche between ants and butterflies. For example, whether these species make distinct and species-specific vibrational calls to attract their respective host ants remains unclear. Furthermore, one also can test if these call characters can be used for species identification.

Methods
Collection and lab rearing of butterfly and host ants. Ten to fifteen S. lohita females were collected from the montane areas in northern Taiwan during September 2014 and August 2015. All adult female butterflies were caged with the ant C. rogenhoferi and a branch of the host plant Mallotus paniculatus (Euphorbiaceae) in an 18.5 × 10.5 × 4.5 cm box for egg laying. Newly hatched larvae were separated individually to a branch of the host plant M. paniculatus in a 5.5 × 8 × 3 cm cage with daily provision of fresh M. paniculatus leaves and removal of frass.
The ant C. rogenhoferi builds a ball-shaped nest on various tree species in the montane areas of Taiwan. Twelve ant colonies were collected from the same areas where S. lohita were collected between July 2013 and October 2015. Every collected nest was searched in a detailed manner for the presence of the queen. Once found, the queen and the remainder of the colony were transferred and maintained in an artificial harborage made of plaster which is placed in a box (34 × 24 × 15 cm). Fluon was carefully applied onto the inner walls of the box to prevent the ants from escaping. All ant colonies were maintained at 25 °C with a schedule of 12-hour light and 12-hour dark with daily supplement of water, honey solution and mealworms. All ant colonies were allowed at least one-month acclimatization period.
Signal recording experiment. The 22 6 th instar larvae and 16 pupae were taken to a noiseless recording studio for signal recording. The larvae and pupae were placed on a branch of their host plant. The 3 M Scotch mounting putty was used to secure the branch (ca. 15 cm height) to reduce its movement. After 5-min acclimatization, the recording started and lasted 20 mins. The signal recording method followed Liao and Yang 45 and Liao, et al. 46 . Vibrational signals were recorded using a gramophone stylus through an amplifier (Lzban, DRA-455, China) and the stylus slightly touched the surface of the host plant leaf. The signals were saved in a dictaphone (Laxon, USB-F20, Taiwan). The number of signal-producing individuals was denoted as N and the number of signal used in sequential analysis was denoted as Σ in the Results section. For the recording of ant workers, the six entire artificial ant harborages were taken to the noiseless recording studio for the signal recording. A branch of M. paniculatus (ca. 15 cm) was put in the nest and one mealworm was positioned on the leaf to stimulate ant calls. After 30-min acclimatization, the recording started and lasted 30 mins. The gramophone stylus was directly put on the plant. For the case of ants, N was representative of the number of artificial ant harborages with signal production. Three ant queens were tested with the recording procedure of caterpillar as described earlier. All the resulting calls were processed through a two-stage noise reduction before analysis. Sampling rate for signal recording was 48,000 Hz and bit depth was 32-bit resolution. Recordings can be accessed in the Supplementary Audios. Playback experiments. The behavioral experiment setup followed Barbero, et al. 18 , and was carried out in a 10 × 10 × 8 cm acrylic arena. A loudspeaker from which the speaker cone has been removed was attached through a hole in the side wall, and sealed on the outside with the scotch removable mounting putty. Ten workers from the same colony were placed in the arena and allowed to settle for 10 minutes before the test calls were played from a MP3 player. The calls were composed of loops of the original recordings, with the volume adjusted to the level reached during the recording.
Three types of larval call, pupal call and ant call were played to the ants. A white noise which was the background sound generated by the recording machine was used as control, a procedure to test if ants react to a potentially meaningless signal. A silence treatment was added to the experiment as a second control and defined as intercept in regression analysis. Ants' behavioral reactions were recorded and categorized as attack, antennation (antenna drumming and vibrating), aggregation (moving toward the speaker) and guarding (standing by the speaker and lifting the gaster). The behavioral experiment lasted for 10 mins, during which different ants' behavior reactions toward the speaker were recorded every minute. The number of each reaction of every minute was summed up to represent the level of each behavior during the 10-minute period (N = 10 colonies, each colony was tested for 3 times, with a total of 30 trails for each treatment).
Statistical analysis and plotting. Dominant frequency, pulse duration and pulse period of all vibrational signals from pupae, larvae and ants were measured by Audacity 2.1.0 to obtain call characteristics. The dominant frequency is the strongest frequency in the spectrogram. The pulse duration is the time length of pulse or pulse trains specific for Type A call of caterpillar, while the pulse period is the time length between the starting points of two consecutive pulses. Call characteristics of both caterpillars and ants were analyzed by using a Mann-Whitney U test and Kruskal-Wallis tests by Past 3.14 47 . A P value < 0.05 was considered significant. Discriminant analysis (DA) was performed using Past 3.14 to reveal signal characteristics of different types of call. A multivariate normality test was conducted, and the raw data did not match the normal distribution (Mardia's test, statistic = 484.1, df = 10, P < 0.001). Hence, all data were transformed (log(n + 1)) prior to the DA 48 . To determine whether an ant was attracted by caterpillar signals, the association between ant behavior and playback signal was analyzed using multiple linear regression analysis using Past 3.14. Figures were produced using Sigma Plot 10.0 (Systat Software, San Jose, CA). The oscillogram and spectrogram of call signals were generated through the Matlab 8.0. (R2012b, The MathWorks, Natick, MA, USA). For audio file reading we used scripts by Ellis 49 . The script for noise reduction and plotting was modified from Vincent 50 and Zhivomirov 51 . Spectrograms were computed using Fast-Fourier transformation (FFT) with a 1024-point window size (Hann window) and a 97% overlap.