The likelihood of hybridization after secondary contact is reduced when assortative mating is strong and fitness of hybrids is low; thus, maintaining reproductive isolation. To this end, signals that convey population identity may enhance recognition and reduce hybridization by promoting assortative mating. Here, we studied visual and acoustic signals, plumage colour and song, that commonly mediate species recognition in birds to evaluate their role in reproductive isolation in a recently diverged clade of Neotropical birds. Subspecies of the Variable Seedeater (Sporophila corvina) that differ dramatically in plumage colour have established secondary contact at two independent contact zones in Costa Rica and Panama. We first evaluated divergence in song in allopatric populations close to the most recent contact zone (Costa Rica), where the two subspecies do not hybridize, and found significant differences in song structure and composition of song elements between subspecies. Then, we used an experimental approach to measure a territorial male's aggressive response to visual and acoustic signals, as a proxy of the role of divergent secondary sexual traits in reproductive isolation. These playback experiments did not show clear effects of divergent colour and song on aggressive response -- we found no effect of plumage coloration and song in one subspecies, and males of the other subspecies were more aggressive towards homotypic songs but, unexpectedly, to heterotypic plumages. Separately, we also characterized the song structure of hybrid individuals from the Panamanian contact zone to evaluate how songs covary given evidence for gene flow. These hybrid individuals sang songs of different element composition than allopatric subspecies but did not differ in most acoustic characteristics. Overall, our experiments suggest that reproductive isolation based on secondary sexual signals is weak in this group, and after secondary contact, the breeding phenology may synchronize, and divergent acoustic signals may homogenize, facilitating further hybridization.

Song Recording and measurement

We evaluated variation in song structure and element composition between two S. corvina subspecies, S. c. corvina (here after named as corvina) and S. c. hoffmanni (here after named as hoffmanni). Male songs were recorded in the field at sites in the Caribbean (corvina males) and Pacific (hoffmanni males) slopes of central and eastern Costa Rica. We also recorded a few hybrid individuals from Gamboa, Panama to evaluate how acoustic signals from this black-pied hybrid population compares to the allopatric populations. Songs were recorded in two consecutive seasons between April and May 2019, and between July and August 2020, using a Tascam DR-100MKIII Linear PCM Recorder (Tascam Corporation, Japan), coupled with a Sennheiser ME66/K6 directional microphone system (Sennheiser Electronic Corporation, Germany) housed in a Rycote 055208 standard Windshield, at 16 bits and a 48-kHz sampling rate in a WAV file. To supplement our field recordings, we included recordings available from the Macaulay Library (Cornell Lab of Ornithology, Ithaca, NY) and the Laboratorio de Bioacústica (Universidad de Costa Rica, San José, Costa Rica), increasing our sample size to a total of 89 males (Supplementary Material Table 1).

We measured three to five songs per male selected randomly (using a random number  generator in Excel). In each song, we measured the following eight characteristics: 1) duration (s), 2) element richness (number of different elements in each song), 3) total number of elements in each song, 4) song pace (elements/duration), 5) lowest frequency (Hz), 6) highest frequency (Hz), 7) average frequency of maximum amplitude (Hz), and 8) average energy entropy (bits). All measurements were obtained using Raven Pro 1.6 sound analysis software (Cornell Lab of Ornithology, Ithaca, NY), using a combination of wave form, sonogram, and power spectrum windows. Our spectrograms were built using a frequency resolution of 188 Hz and temporal resolution of 5.8 ms, with a Hann window with 50% overlap and 256 Hz transform size.

Song playback and mount presentation experiments

We performed playback experiments to evaluate the aggressive response of local males to different playbacks and mounts, simulating intrusions on the focal territory. Playback experiments were carried out from July to August 2020, in the Central Pacific and Central Caribbean slopes of Costa Rica (Fig.1A, Supplementary Material Table 2), along secondary growth habitats. We found territories by finding singing males and placed the speaker and mount after a few minutes of observation to ensure that males were actively singing and that  the target male was the territory owner (i.e., no other males nearby). We placed the speaker and mount at 3m height because territorial males usually display from high exposed perch (personal observation), and ~10m away from the target male close to vegetation. Songs were broadcasted from an Anker AK-A3121011 loudspeaker placed just below the taxidermic mount at a sound pressure level of 80 dB by a Sper Scientific mini sound level meter (model 840014), with a fast response setup, and located 1m away. We then moved away 10 m from the speaker and the target male and waited 3 minutes before starting the acoustic stimuli. Each target individual was 
exposed to just one treatment, which was selected randomly from the pool of all possible combinations of stimuli. Treatments consisted of eight playback recordings of both subspecies from allopatric populations (4 per subspecies), in all possible combinations with four mounts of corvina or hoffmanni (2 per subspecies), for a total of 32 possible combinations.

Each experiment consisted of two minutes of stimuli, followed by three minutes of silence. For each experiment we estimated the initial distance (m), as the distance when acoustic stimulus started, and measured the aggressive response as: 1) vocal latency (s) or time to first vocal response, 2) song latency (s) or time to first song response, 3) number of songs produced in response to the first playback, 4) number of calls, 5) flight latency (s) or the time to first flight, 6) number of flights, 7) time to approach the mount at < 3m (s), 8) total time interacting with the mount at < 3m (s), and 9) minimum distance (m) from the mount and speaker. We recorded the response behaviours for five minutes, starting when the acoustic stimuli were first broadcasted.