The research was conducted on the Oriental tit and the Varied tit breeding in nest boxes (different number each year) in mixed forests of Seoul (three locations: 37°26′50′′N 126°56′44′′E, 37°35′46.1′′N 127°02′42.9′′E, 37°35′10.5′′N 127°06′03.9′′E), Republic of Korea. Based on monitoring of 100–200 nest boxes each year since 2011, we established that snakes are the main nest-predators for both species25. The field experiments were performed from 2013 to 2020 and comprised predator presentations and playback experiments.
Predator presentation experiments
From 2013 to 2020, we conducted predator presentation experiments on 23 breeding pairs of the Varied tit and 35 breeding pairs of the Oriental tit. In each experiment we recorded alarm calls of the pair (without differentiating between male and female due to practical reasons) in response to a presentation of a predator model (jay, cat, and snake). Next, we compared the alarm calls toward snake models with the alarm calls toward non-snake predator groups in order to determine if the snake-induced vocalizations are clearly different from the others.
In preliminary trials, we presented an empty cage on top of the nest box, and it did not induce any alarm calls: after a short period of initial cautiousness, the parents resumed their regular visits. We used four types of predator models to imitate the presence of a predator at a nest box: a live Steppe rat snake (Elaphe dione), a flexible plastic snake that resembled a live snake, a furry toy cat that resembled a domestic cat, and a stuffed Eurasian jay (Garrulus glandarius). The toy cat and the live and model snakes were placed on top of the nest box, while the stuffed jay was placed just next (within 20 cm) to the nest box (Table 1). All these predators (snake, jay, and cat) occur in the study area and are able to depredate nestlings and/or fledglings of small passerines. By using an attached transparent string to a body part of the predator model (neck of the plastic snake, tail of the furry toy cat, and leg of the Eurasian jay) we were able (hidden behind a tree or bush at a distance of 10–20 m) to imitate small movements of the snake body (rising up and lowering down frontal part of the body), cat’s tail (imitating a sitting cat that moves its tail) and jay’s head and upper body (imitating a bird looking around). Each breeding pair was tested with only one of the three predator treatments for the Varied tit (11 pairs in the plastic snake treatment, 5 pairs in the jay treatment, and 7 pairs in the cat treatment) and one of the four predator treatments for the Oriental tit (9 pairs in the live snake treatment, 10 pairs in the plastic snake treatment, 8 pairs in the jay treatment, and 8 pairs in the cat treatment). We used only one model per predator type, but the presentations varied randomly due to variability in the movements of a live snake and the movements of predator models induced manually. We aimed to test as many nests of each species as possible in each season while balancing the sample size and the distribution of testing dates among the treatment groups. Varying field conditions from year to year, such as the number of nests available and the rate of predation, ultimately determined the final sample sizes. For the same reason, we could not plan to randomly distribute the nests into treatment groups ahead of time, as it was practically impossible to know which nest will be available at the end of the breeding cycle. Finally, as the experiments were conducted in multiple years, we cannot formally exclude the possibility of pseudoreplication by conducting experiments on the same individuals in different years. However, considering a small sample size of experimental pairs among the possible number of pairs in the study area (roughly 50 pairs per year; Ha et al., unpublished data), we believe that the experiments can yield a valid dataset pertaining to the original purpose of the study.
Recording equipment and recording conditions
We used two different microphones (Primo EM172 and Sennheiser me67) which were connected to one of the three audio recorders (Tascam DR-05, Tascam DR-100, and Marantz pmd661 mk2) (see Table 1). We mounted the recording equipment on a 0.5 m high tripod and placed it 1 m away from the nest box. After the setup, the researcher waited 30 m away from the nest box until parent(s) entered the nest to became assured that the birds were habituated to the presence of the recording equipment. When the parents left their nest, we approached the nest box to set up a predator model on top of the box or within 20 cm from the box on one side of the nest box (jay). At their return, the alarm calls of the parents were recorded for 4 min from their first vocal response at 48 kHz sampling rate and 24-bit depth for jay presentations to the Oriental tit, and 96 kHz sampling rate and 24-bit depth for all the other presentations. At each presentation, we moved the predator model from a distance by using the string attached to the model. The movements of the model were repeated by pulling the string twice within a 10-s interval, followed by a release after an additional 5-s pause (for a visual demonstration, please refer to Supplementary Video V4). Based on our field experiences, we believe that these repeatable movements of the predator model are essential to induce clear alarming response of birds similar to natural responses to real predators. It was impossible to perform the experiments blind with respect to the predator type, study species, or nest ID. However, fixed procedures were implemented for moving the predator models by the experimenter, and standard methods were used for recording the bird responses, in order to avoid biases in the results. The presentations were conducted between 9:00 am and 5:00 pm, and rainy days were excluded from the experimental schedule.
We used two playback treatments representing snake alarm calls (snake-induced alarm calls) of the Oriental tit and snake alarm calls of the Varied tit. We played the calls at the nests of the two species at the time when nestlings are sufficiently old to be able to jump out of the nest. This allowed us to test if nestlings of each species fledge upon hearing the conspecific as well as heterospecific snake alarm calls. The experiments were performed from 2018 to 2020.
When the nestlings in a nest reached the age at which fledging can be induced (more than 14 days old), we played back the recordings of conspecific or heterospecific alarm calls to them. Thus, only nests with nestlings that survived until 15 days after hatching were included in the playback experiment. Our previous research on the Oriental tit in this study area has shown that nestlings older than 14 days old are capable of fledging when they hear conspecific “snake call”17. We carried out playback tests on 30 nests of the Oriental tit and 20 nests of the Varied tit; 15 Oriental tit broods and 10 Varied tit broods were tested with the Oriental tit playback, and the same number of broods for each species were tested with the Varied tit playback. We used 5 different playback stimuli for each species. Hence, each playback stimulus was used at 3 broods of the Oriental tit and at 2 broods of the Varied tit. Even though multiple uses of the same playback cannot provide full independence between tests, we followed this method to assure that we use only the best quality recordings without excessive noises or alarm calls of the other individuals, or species, except the parents of the nest. We believed that using clear signals (with relatively low noise level) to induce nestlings’ response expected in a natural situation of vocalizing parents is more important than keeping formal independence of all data. This design cannot be formally used to statistically generalize the results to the responses of nestlings to a whole population of alarm calls of each species in the study area. Despite this drawback, the main difference between the two sets of these good-quality recordings is the species from which they were recorded (the Oriental tit or the Varied tit). Therefore, we think that we can use the results as indicators to assess if the alarm calls from the two species trigger similar or different responses in nestlings of the two species. For each species, we applied the ‘minimization’ randomization approach, as outlined by the guidance for Animal Research: Reporting of In Vivo Experiments (ARRIVE)42. In this method, the subsequent allocation of playback stimuli to nests was influenced by prior allocations, with the objective of mitigating potential biases associated with testing dates and sites in the field experiment. Specifically, we employed a rotation of treatments to ensure a balanced distribution of sample sizes across various dates to achieve robustness in the study’s outcomes.
The playback stimuli
The playback samples were created from the recordings of predator presentations in the current study (see section “predator presentation experiment” above). The preparation of acoustic stimuli generally followed the protocol used in the previous study17. We created 3-min long playback stimuli by using alarm call recordings from plastic snake presentations for the Varied tit and from live snake presentations for the Oriental tit (see Supplementary Audios S1, S2). Our preliminary observations of the Varied tit confirmed that parental vocalizations in response to the plastic snake trigger fledging of the Varied tit nestlings, the same as Oriental tit vocalizations to live and artificial snakes trigger fledging of Oriental tit nestlings. We selected the five best recordings (i.e. those with the highest signal-to-noise ratio) for each species. From the 4 min-long recordings, we removed sections with following issues; excessive background noise, vocalizations of other species, overlapped calls by multiple individuals. We then removed sections without any vocalizations and fixed the maximum inter-phrase duration to one second to control the intensity of the alarm calls in the playback stimuli. This filtering procedure produced an audio file where inter-phrase durations do not exceed one second, and that only includes alarm call phrases from the parents of a specific nest. From this file, we chose the first minute and applied a 1 kHz high-pass filter. Finally, to create a 3-min playback stimulus, we converted the 1-min recording to 48 kHz sampling rate, 24-bit depth, and repeated it three consecutive times in the playback stimulus (Supplementary Fig. S4). We produced 5 playback stimuli from each species in this way. Our method allowed the playback stimuli to retain within and between species variation in acoustic properties of phrases among the recordings, while maintaining similar temporary dynamics of phrase delivery among the playback stimuli. Based on previous studies that revealed only alarm calls induced by snakes can trigger the fledging response, we did not include control playback groups in the experimental design. This allowed us to minimize the total number of nests required for the study.
We followed the protocol used in our previous study17, in which we proved that the jumping-out reactions of nestlings do not happen at all in response to ‘churr’-based alarm vocalizations of the Oriental tit to a mammal brood predator – chipmunk, while it does happen in response to snake alarm calls. This protocol considers that in the field situation there is no way of knowing for each nest at what age exactly the nestlings are old enough to be capable of jumping out of the nest box. Depending on the rate of growth and maturation of the brood, it may happen on day 15, 17, 18, or maybe even later. The protocol described below assures that the “no jumping out” outcome of the test procedure is not due to the nestlings being too young, but at the same time the protocol minimizes the chances that the nestlings fledge naturally before the researcher arrives at the nest to conduct the playback test. This protocol has proven effective in properly differentiating between calls that trigger nestlings to jump out of the nest and those that do not17.
Playback trials were performed between 9:00 am and 5:00 pm during non-rainy days, and the stimuli were played through a portable speaker (JBLAB HRS-20XB) when the parents were away from the nest. The speaker was positioned at the sidewall of a nest box. The researcher was standing 10-15 m away behind a bush or tree hiding from the view of the parents or from nestlings potentially jumping out of the box. Nestlings’ response to the playback was recorded as a binary variable (i.e. fledging response present or absent). We stopped a playback trial when the nestlings started to fledge and to minimize negative effects on nestlings we captured and put back the nestlings to the nest if they were too young to be able to properly fly. If parents arrived during the presentation of the playback, we stopped the playback and resumed it about five minutes after the parents left the area. We then repeated the same trial from the beginning of the playback until we successfully finished playing the 3-min playback stimulus without interruption by parental visits, or until we observed the first behavioral signs of fledging by nestlings (a nestling sitting in the nest box entrance). When this procedure did not result in fledging of the nestlings on one day, then we repeated the playback procedure on the next day in the same manner, and we repeated it again and again on the following days if no fledging was observed on a preceding day. This procedure was applied from the time when nestlings are 15 days old until the day when nestlings fledged in response to playback or until we found an empty nest indicating the likely natural fledging event.
A brood that showed signs of fledging during this procedure was assigned a binary value “response present”, and a brood that showed no signs of fledging until the nest became empty (presumably due to natural fledging) was assigned the value “response absent”. In the playback experiment, we defined the experimental units as the nests of the target species. The experiments could not have been conducted blind with respect to the experimental treatment or the nest ID/nest location. However, the potential for bias in the recorded variable values remains minimal due to the unequivocal binary nature of the response variable—yielding a ‘yes/no’ outcome, which has no chance of being subject to interpretation errors. Additionally, as described in “Experimental design”, an effort was made to ensure balanced sample size within each individual species, considering the distribution of testing dates and sites throughout the playback treatments.
Sound analysis and statistical analysis
Syllable types and their frequency in vocal responses
We aimed to determine whether vocal responses of the Varied tit and the Oriental tit to snakes differed from their vocal responses to other predators. First, for each species, we visually inspected a 2-min sonogram recording from the initial call to classify syllables into distinct types based on their unique appearance. We then assigned a different letter from the English alphabet to each syllable type (Fig. 1; Supplementary Table S5 contains our syllable type symbols matched with the syllable type names previously used in the literature). For each syllable type we summed the total duration of all syllables in the initial 2 min of response in each recording. Then, we calculated the proportion of time that each syllable type occupied in the total duration of the vocalization per recording (duration for all syllable types). Analyzing these proportions allowed us to identify syllable types (for each species separately) that were predominantly used in responses to specific predators, including those syllables that comprised majority of time in vocalizations triggered by a snake, thus constituting what we refer to as ‘snake alarm calls’. Our classification of snake alarm call syllables closely matched the sonograms of snake alarm calls described in previous studies15,17. We decided to base the proportions on time (duration) rather than number of syllables to more precisely determine the syllable types that stimulate the receivers for most of the alarm vocalization duration.
Statistical comparisons of acoustic properties of syllable types present in snake alarm calls
Oriental tits used the broadly defined ‘LF/churr/rattle/jar’ type of syllable17,19,20 (labeled as type ‘L’ in this study) in alarm calls to all three predators. In snake presentations this was the only type of syllable used by the Oriental tit parents. Visual inspection of sonograms combined with auditory inspection of recordings is often sufficient to further categorize this type as either the typical ‘jar’ call characteristic for snake alarm calls15,43 or the other ‘churr’ sub-type used toward other predators (e.g. chipmunks)17. However, in some cases the difference is more subtle, especially based only on visual inspection of sonograms. Therefore, we statistically compared acoustic properties of this broad syllable type (i.e. syllable type ‘L’) between different predator presentations.
We also compared the acoustic properties of snake alarm call phrases (syllable type ‘L’ in the Oriental tit and syllable type ‘f’ in the Varied tit) between species. Here, the unit of analysis was a phrase composed of only the syllable type used in responses to snakes: the ‘L’ syllable type (for the Oriental tit) or only the ‘f’ syllable type (for the Varied tit). For these comparisons, we measured 13 acoustic properties of each phrase in the initial 2 min of each recording, including: Number of Syllables, Phrase Duration, Average Syllable Duration, Standard Deviation of Syllable Duration, Average Calling Rate, Standard Deviation of Calling Rate, Average Inter-syllable Interval, Average Center Frequency of Syllables within a Phrase, Standard Deviation of Center Frequency of Syllables within a Phrase, Average 1st Quartile of Frequency of Syllables within a Phrase, Average 3rd Quartile of Frequency of Syllables within a Phrase, Average Inter-quartile Frequency, Standard Deviation of Inter-quartile Range Frequency of Syllables within a Phrase. We then grouped these variables into 7 “duration variables” (or “timing variables”) and 6 “frequency variables” (Supplementary Table S6). These variables represent the “outcome measures” in the ARRIVE guidelines42. The measurements of acoustic variables were performed in Raven Pro 1.644 with spectrogram configuration set at a Hann sample window size of 512, 256-sample hop size with 50% overlap, and frequency grid spacing and DFT size were set to 86.1 Hz and 512 samples. As these variables are often correlated with each other, we used parallel analysis (function ‘fa.parallel’)45 and principal components analysis (function ‘principal’)45 with varimax rotation to extract composite variables (principal components). Next, we conducted statistical analysis (Generalized Additive Model for Location, Scale and Shape, function ‘gamlss’)46 to test if the acoustic properties (the principal components) of the specific types of syllables (syllable type ‘L’ of the Oriental tit and syllable type ‘f’ of the Varied tit) are different among predators or between the two species. The goodness of fit of the models was assessed using a worm plot (function ‘wp’ in R package ‘gamlss’)46, which generated residual quantiles that aligned with a uniform distribution, indicating a reasonable fit of the model to the data. The analysis of ‘L’ syllables of the Oriental tit was based on n = 322 phrases from 10 nests in response to a plastic snake, n = 229 phrases from 8 nests in response to a jay, and n = 196 phrases from 8 nests in response to a cat. The analysis of ‘f’ syllables of the Varied tit was based on n = 288 phrases from 11 nests in response to a plastic snake. Additionally, we conducted variance component analysis (function ‘anovaVCA’ in R package ‘VCA’)47 for each principal component to explore the variability of acoustical properties by levels of explanatory or random variables (e.g., predator, species, and nest). All the statistical analyses were conducted in R48 (see Supplementary Table S7 for the basic information of acoustical dataset).
As we used two types of microphones that may differ in their frequency response curves, comparisons involving frequency variables can only be conducted within the set of data obtained using the same microphone model. Therefore, we decided to conduct all statistical comparisons within the same microphone model. Hence, we conducted three separate statistical analyses (Table 4): comparison of the Oriental tit’s ‘L’ syllables in snake calls with those in jay alarm calls (Sennheiser me67 microphone) and with those in cat alarm calls (Primo EM172 microphone), as well as comparison of snake alarm call syllables (‘L’ for Oriental tits and ‘f’ for Varied tits) between species (Primo EM172 microphone).
Statistical analysis of playback experiments
When analyzing the result of playback experiments, we used Fisher’s exact test (function ‘fisher test from the stats package in R’)48 and Mixed effects logistic regression (function ‘gamlss’)46 to compare the fledging responses between nestlings of two different species induced by playbacks of conspecific and heterospecifics snake alarm calls. In the logistic regression, we used the fledging response of nestlings (binary: 0 & 1) as the dependent variable and the nestling species (categorical with two levels: Varied tit nestlings & Oriental tit nestlings) as the explanatory variable. All the statistical analyses were conducted in R48.
Approval for animal experiments
The experimental procedures were approved by the Institutional Animal Care and Use Committee of Seoul National University: SNU-180205-1, SNU-190413-2, and SNU-200413-1-1. All methods were carried out in accordance with relevant guidelines and regulations. We also confirm that we followed the ARRIVE guidelines.