(AGENPARL) - Roma, 25 Febbraio 2026 - (AGENPARL) – Wed 25 February 2026 For consideration in Annals of the New York Academy of Sciences
Page 1 of 84
Original Article
Rhythmic signaling of ants and butterflies with varying degrees
of myrmecophily
Submission ID
d9310f04-5eb5-412f-a5e0-01e712950bf7
Manuscript ID
annals-ORr-1124
Submission Version
Revision 2
PDF Generation
26 Jan 2026 04:25:50 EST by Atypon ReX
Authors
Dr. Chiara De Gregorio
Corresponding Author
Submitting Author
ORCiD
https://orcid.org/0000-0001-7017-6181
Mrs. Izabela Sondej
ORCiD
https://orcid.org/0000-0001-8548-9738
Mr. Stefano Previdi
Prof. Francesca Barbero
ORCiD
https://orcid.org/0000-0003-2667-0435
Prof. Luca Pietro Casacci
ORCiD
https://orcid.org/0000-0003-3083-7357
Affiliations
• Department of Psychology, University of Warwick,
Coventry, United Kingdom
Affiliations
• Department of Natural Forests, Forest Research
Institute, Białowieża, Poland
Affiliations
• Department of Life Sciences and Systems Biology,
University of Turin, Torino, Italy
Affiliations
• Department of Life Sciences and Systems Biology,
University of Turin, Torino, Italy
Affiliations
• Department of Life Sciences and Systems Biology,
University of Turin, Torino, Italy
Page 3 of 84
Rhythmic signaling of ants and butterflies with varying degrees of myrmecophily
Chiara De Gregorio1, Izabela Sondej2, Stefano Previdi3, Francesca Barbero3, Luca Pietro Casacci3
1 Department of Psychology, University of Warwick, Coventry, United Kingdom
2Department of Natural Forests, Forest Research Institute, Białowieża, Poland
3Department of Life Sciences and Systems Biology, University of Turin, Torino, Italy
Correspondence
Chiara De Gregorio,
Francesca Barbero,
Graphical Abstract
Myrmecophilous butterflies evolved diverse adaptations to communicate with ants, including
acoustic ones. We analyzed the rhythmic properties of vibroacoustic signals in two ant and nine
butterfly species differing in myrmecophily. Highly myrmecophilous species and ants shared
complex rhythmic structures (isochrony, double-meter), suggesting convergent evolution for
Page 4 of 84
mutualistic coordination. Species with weaker associations showed simpler or irregular rhythms,
highlighting rhythm’s key role in signal efficiency and recognition.
Abstract
Myrmecophilous organisms have evolved in several arthropod lineages, developing specialized
traits to communicate with ants. In butterflies, these include morphological, visual, chemical,
behavioral, and acoustic adaptations. While acoustic communication has long been overlooked,
recent studies show that vibrational signals mediate key aspects of caterpillar–ant interactions.
Yet, no study has specifically investigated the rhythmic structure of such signals in
myrmecophilous contexts, despite growing evidence that rhythm is a fundamental component of
signal architecture across taxa. We examine the rhythmic properties of vibroacoustic signals from
two ant and nine butterfly species differing in myrmecophily degree. We tested whether rhythmic
features such as pulse train tempo, inter-train interval, and rhythmic patterns vary across taxa and
reflect the strength of their ecological association with ants. Our results reveal that ants and
highly myrmecophilous species share a complex rhythmic organization (isochrony and doublemeter), likely reflecting convergent adaptation to tight mutualistic interactions. Species with
intermediate or no myrmecophily associations showed more variable or simplified rhythms.
Temporal regularity and precision could possibly balance the need for signal recognizability with
energetic constraints and avoidance of detection by unintended receivers. These findings highlight
the role of temporal patterning in vibroacoustic communication, influencing signal efficiency and
recognition in ant–butterfly interactions.
Keywords: Interspecific signaling, Rhythmic patterning, Isochrony, Double meter
Page 5 of 84
1. Introduction
Communication plays a central role in animal behavior, enabling the transfer of information
between individuals through a variety of sensory channels. Among these, vibroacoustic
communication is a specific form of signaling that involves the production and detection of
mechanical vibrations, typically transmitted through air or various substrates such as soil, leaves,
or plant stems [1]. Although across invertebrates chemical communication represents the
dominant modality of communication, vibroacoustic signaling is also widespread, especially
among several arthropod taxa, including web-building spiders, crickets and katydids, leafhoppers,
and planthoppers, where it serves key functions such as prey detection, mate attraction, and
territorial interactions [2,3].
In ant colonies, vibroacoustic signals are involved in a wide range of vital social functions, such as
coordinating alarm and defense responses, marking territory, attracting mates, recruiting
nestmates, initiating rescue behaviors, and conveying information related to caste or social
hierarchy [4–7]. Despite their vital role in the organization of ant colony life, vibroacoustic signals
can also become a point of vulnerability. Several myrmecophilous organisms, i.e., species that live
in close association with ants, have evolved, besides the ability to imitate chemical signals, the
capacity to mimic ant vibrational signals to manipulate their behavior, gaining access to food,
protection, or integration within the colony [8–12].
This phenomenon has been particularly well studied in lycaenid butterflies (Lepidoptera:
Lycaenidae), a family in which over 75% of species establish facultative or obligate associations
with ants [13–16]. Since the pioneering work of DeVries[17], it has become clear that both caterpillars
and pupae of several lycaenid species are capable of emitting substrate-borne vibrational signals,
often functioning as acoustic mimicry to manipulate the behavior of their host ants [8]. By
Page 6 of 84
producing vibrational signals, lycaenid caterpillars could recruit, retain, or modulate the tending
workers, often eliciting trophallaxis or protective behaviors[7]. On the other hand, ants respond to
these signals with increased antennation, exploration, or context-dependent actions such as
digging [9,10]. In particular, studies on obligate myrmecophilous Phengaris (=Maculinea) butterflies
have shown that their caterpillars and pupae produce signals closely resembling those of ant
queens, which elicit strong guarding and rescue responses by Myrmica workers [9,10]. The acoustic
properties of these signals have been shown to vary across species with different degrees of
myrmecophily, suggesting an evolutionary link between signal structure and the level of host
integration [7,18].
However, the role of the temporal and rhythmic structure of vibroacoustic signals in effectively
mediating communication between ants and their symbionts remains largely unexplored. In
particular, the extent to which rhythm influences signal transmission and recognition in these
systems is still unknown. However, studies on rhythmic structure may provide hints about its
potential functional roles. Rhythm may play a key role not only in enhancing signal detectability
and discrimination [19,20] but also in conveying information through the temporal regularity of the
signal itself [21]. Investigating rhythmic structure can thus provide important insights into how
vibratory signals are produced, perceived, and function in interspecific and intraspecific
communication.
This study aimed to investigate the rhythmic properties of insects’ vibroacoustic signals to uncover
the role of temporal patterning in within and between-species communication. In particular, our
first aim was to test whether the rhythm of vibroacoustic signals differs between the species
examined, to understand the degree of species-specificity, as this may reflect the extent to which
signal properties are evolutionarily tuned to particular ecological or social interactions, such as
Page 7 of 84
species recognition. A high degree of species-specificity suggests finely adapted communication
strategies, while more conserved patterns indicate general constraints or shared evolutionary
pressures.
We also investigated the degree of rhythmic similarity between the vibroacoustic signals produced
by ants and those emitted by butterfly caterpillars, with a specific focus on whether the degree of
myrmecophily could predict variation in temporal and rhythmic features. Given the importance of
rhythmic patterning in animal communication, we tested the hypothesis that higher levels of
myrmecophily would be associated with greater rhythmic similarity to ants. To address this, we
examined four main aspects: 1) Are caterpillar trains faster or slower than ant signals? We
predicted that species with higher myrmecophily would match ant tempo more closely. 2) Do
caterpillars differ from ants in the timing of pauses between trains, and is this linked to
myrmecophily? Higher myrmecophily species were expected to show intervals more similar to
ants. 3) Do caterpillar signals exhibit the same rhythmic structures as ants (e.g., isochrony or
double meter)? We hypothesized that species closely associated with ants would display more
ant-like rhythmic patterns. 4) How consistent are the trains in maintaining rhythmic structure? We
predicted that species with stronger ant associations would show higher regularity. Together,
these analyses enable us to investigate whether temporal and rhythmic convergence in
vibroacoustic signaling may reflect adaptive adjustments linked to the strength of ant–butterfly
associations, potentially shedding light on the evolutionary pressures shaping interspecific
communication.
2. Methods
Page 8 of 84
2.1. Data collection
Sounds produced by caterpillars (Figure 1, S1) of nine lycaenid butterfly species, i.e., Lycaena
dispar, L. phlaeas, Cupido argiades, Polyommatus icarus, P. bellargus, P. coridon, Scolitantides
orion, Plebejus argus, and Phengaris (=Maculinea) alcon, as well as two ant species of the genera
Myrmica and Tetramorium, collected from various locations in Northern Italy, were recorded and
compared. These genera are also widespread in European grasslands and often co-occur with
lycaenid caterpillars, making them ecologically relevant reference models for examining broader
patterns of temporal convergence. All butterfly species exhibiting myrmecophily in this study
belong to the subfamily Polyommatinae, whereas L. dispar and L. phlaeas are members of the
subfamily Lycaeninae. Data on the butterfly and ant species used in this study were originally
collected by Riva and colleagues[18]. Caterpillars and ant colonies were collected from various
locations in Northern Italy between May 2012 and April 2014. Whenever possible, fully developed
butterfly caterpillars were collected; otherwise, plants on which eggs had been laid were gathered
to rear caterpillars in the laboratory. Specimens and their host plants were kept in transparent
boxes (30 × 20 × 20 cm³) covered with mesh at temperatures of 18 °C during the night and 25 °C
during the day, under a 14-hour light/10-hour dark cycle. After experiments, adults were released
back into their original collection sites. Concurrently, ant nests were also collected in the field and
subsequently maintained in plastic boxes (28 × 15 × 10 cm³) under controlled laboratory
conditions and reared on a diet of sugar and Drosophila larvae. Selection of lycaenid species was
based on their varying degrees of interaction with ants, classified following Fiedler[13] slightly
modified by Schönrogge and colleagues[7] 0: no association with ants (absence of myrmecophily);
1: unstable associations with ants, occurring only occasionally (weakly myrmecophilous); 2:
variable larval attendance by ants (moderately myrmecophilous); 2.5: between moderately and
constantly myrmecophilous; 3: most or all mature caterpillars consistently associated with ants
Page 9 of 84
(constantly myrmecophilous); 3.5 between constantly and obligate myrmecophilous; 4:
caterpillars entirely dependent on ants as parasites (obligate myrmecophilous). For the statistical
analyses, species of butterflies were grouped into four categories of myrmecophily based on
previously assigned scores:
No myrmecophily (“NO”) – L. dispar, L. phlaeas (score = 0)
Low myrmecophily (“Low”) – C. argiades (score = 2), P. icarus (score = 2.5)
Medium myrmecophily (“Medium”) – P. bellargus, P. coridon, S. orion (score = 3)
High myrmecophily (“High”) – P. argus (score = 3.5), P. alcon (score = 4)
Sound recordings of late-instar caterpillars and ants were obtained using a custom-made device
designed to minimize stress and disturbance. Samples were not stimulated/threatened with any
tools to obtain the recordings. The recording setup included a compact chamber (12.5 × 8 × 2 cm³)
equipped with a highly sensitive moving-coil miniature microphone (sensitivity: 2.5 mV/Pa at 1
kHz), with a second identical microphone recording ambient noise in anti-phase to reduce
interference. Signals from both microphones were combined through a mixer and amplifier
system covering a frequency range of 20 Hz to 20 kHz with a total gain of approximately 83 dB.
The entire setup was powered by a 12V gel cell battery and housed inside an anechoic chamber to
minimize background noise. Individual caterpillars and ants were placed directly on the
microphone surface and recorded at room temperature (23–25°C) for 20-minute intervals,
beginning 5 minutes after placement. For full methodological details, see Riva and colleagues[18].
2.2. Data preprocessing
Page 10 of 84
Acoustic recordings were carefully inspected for quality, saved in WAV format (16-bit resolution),
and analyzed for temporal and spectral properties using Praat software. We obtained a total of 56
recordings, each one from a different subject: 11 for Myrmica, 12 for Tetramorium, five for P.
alcon, five for C. argiades, three for P. argus, five for P. bellargus, one for P. coridon, two for L.
dispar, five for P. icarus, four for S. orion, three for L. phlaeas.
(Figure S1), we annotated the onsets and offsets of all pulses (i.e., call fundamental unit) emitted
in each recording and measured the starting point and duration of each interval[22]. Moreover,
since pulses occur in temporally structured sequences, we defined “pulse trains” as groups of
regularly spaced pulses forming the basic units of caterpillar calls. Each pulse generally exhibits a
harmonic structure composed of at least three frequency components, with the first
corresponding to the fundamental frequency[18]
Using the software RStudio to evaluate the rhythmic structure of the vibroacoustic signals, we
calculated the duration between the onsets of each couple of adjacent pulses (inter-onset interval
or IOI; hereafter tk). We obtained 5581 tk values for Tetramorium, 8754 for Myrmica, 5324 for P.
alcon, 1507 for P. argus, 6643 for P. bellargus, 2392 for P. coridon, 7239 for S. orion, 5817 for P.
icarus, 5897 for C. argiades, 417 for L. dispar, and 869 for L. phlaeas. Then, we calculated rhythmic
ratios (rk) by dividing each interval tk for itself plus the following one: rk = tk/(tk + tk+1) (see Refs. 23–
25). We obtained 5070 rk for Tetramorium, 8474 for Myrmica, 5263 for P. alcon, 141 for P. argus,
6500 for P. bellargus, 2384 for P. corindon, 6532 for S. orion, 5663 for P. icarus, 5480 for C.
argiades, 321 for L. dispar, 643 for L. phlaeas.
To evaluate the occurrence of small integer ratios, we tested whether the number of data points
falling within predefined ratio boundaries is significantly greater than expected under a null
Page 11 of 84
distribution. In particular, we were interested in investigating the presence of the two simplest
small-integer ratios: isochrony (1:1), when two consecutive temporal units have the same duration
(e.g., the ticking of a metronome), and double meter (1:2), which refers to rhythmic patterns in
which one event is followed by another that is approximately twice as long or short in duration. To
do so, we divided the ratio distribution into on-integer and off-integer ratio ranges, and we
centered the on-integer range around 0.5 (or 1:1), 0.33 (or 1:2), and 0.66 (or 2:1). In accordance
with Ref. 23, a 1:1 on-integer ratio is considered when ratio values fall within the values between
0.444 and 0.555, and an off-integer ratio when it is between 0.4−0.444 and 0.555−0.6. The range
of the 1:2 on-integer ratio is 0.308−0.364, while the off-integer ratio is within 0.286–0.308 and
0.364−0.4. Finally, for the 2:1 ratio, the on-integer ratio is within 0.636−0.692, while the respective
off-integer ratio is within 0.6−0.636 and 0.692−0.714. We then tallied all the ratios occurring
within the respective on-integer and off-integer ratio intervals for each group of pulses (trains)
contained in each file.
2.3. Statistical analyses
We performed all statistical analyses below using RStudio. We first investigated differences in
tempo and interval duration at the species level, and then at the myrmecophily degree level. We
ran a total of 12 models. To test for the significance of our full models, we compared them against
null models containing only the random factors[26] with a likelihood ratio test (Anova with
argument “Chisq”[27]).
Page 12 of 84
To test whether the tempo of trains and intervals between trains differed across species, we used
linear mixed models (LMM) (lme4 package [28]) with species identity as a fixed factor and file ID as
a random factor, followed by post-hoc pairwise comparisons among species (emmeans package
[29]). The response variables were log-transformed, and in both cases the full model differed
significantly from the null (tempo of trains: χ² = 34.42, df = 10, p < 0.001; intervals between trains:
χ² = 81.54, df = 10, p < 0.001).
We ran a total of two LMMs and eight generalized linear mixed models (GLMMs), using a null vs
full approach: to test for the significance of our full models, we compared each model with the
same one containing random factors only (Anova with chi-sq argument[27]). If null and full
significantly differed, then the fixed factors affected the distribution of the response variable, and
we therefore applied post hoc tests (emmeans package [29]) to perform all pairwise comparisons
for the levels of the fixed factors. Before fitting the models, we log-transformed the duration
values for the LMM and used Poisson as the distribution for the GLMMs (except for two cases –
the one focusing on butterflies with a low degree of myrmecophily was fitted as negative binomial
after checking the overdispersion parameter; the one testing regularity rate was fitted with a beta
distribution).
Page 13 of 84
To test whether the tempo of pulses (in trains) emitted by caterpillars differed from the ones
emitted by ants, we used a LMM (lme4 package, [28]). We used the log-transformed duration of tk
as the response variable, with the level of myrmecophily (No, Low, Medium, High) as a fixed
factor, and the ID of the file from which tk were extracted as a random factor. The full model
significantly differed from the null one (Null vs. Full—χ2 = 18.56499, df = 4, p < 0.001). We then
ran a post-hoc test to perform all pairwise comparisons for all levels of the variable
“myrmecophily degree” (emmeans package[29]).
To test whether the duration between different trains of pulses emitted by caterpillars differed
from the ones emitted by ants, we used a LMM (lme4 package [28]). We used as a response
variable the log-transformed duration of intervals, as a fixed factor the level of myrmecophily (No,
Low, Medium, High), and as a random factor the ID of the file from which intervals were extracted.
The full model significantly differed from the null one (Null vs. Full—χ2 = 12.91156, df = 4, p =
0.012). We then ran a post-hoc test to perform all pairwise comparisons for all levels of the
variable “myrmecophily degree” (emmeans package[29]).
We used five GLMM (glmmTMB package[30]), one for each level of category (Ants and the four
levels of myrmecophyly: No, Low, Medium, High) to investigate if caterpillar trains are temporally
organized as ants’ ones. In particular, we tested whether their rhythmic organization would match
Page 14 of 84
an isochronous peak (1:1 ratio) or double meter (1:2 and 2:1 rhythmic ratios). For each model, our
response variable was the number of rk for each level, as a fixed factor rk bin type (factor levels:
OFF1:1, ON1:1, OFF1:2, ON1:2, OFF2:1, ON2:1). For all models, we used the file ID as a random
factor. We accounted for the presence of zeros in the dataset by specifying ziformula = 1.
Additionally, we included an offset term to adjust the rk counts relative to the width of each bin in
the probability density function, thus controlling for differences in bin size. The comparison with
the null model was significant for all models (Ants: χ2 =10783.05, df = 5, p