Song divergence between closely related taxa may play a critical role in the evolutionary processes of speciation and hybridization. We explored song variation between two Ecuadorian subspecies of the gray-breasted wood-wren (Henicorhina leucophrys) and tested the impact of song divergence on response behaviors. Songs were significantly different between the two subspecies, even between two parapatric populations 10 km apart. Playback experiments revealed an asymmetric response pattern to these divergent subspecies specific songs; one subspecies responded more to songs of its own subspecies than to the other subspecies’ songs, whereas the second responded equally strongly to songs of both subspecies. While song parameters revealed a mixed pattern of divergence between allopatric and parapatric populations, the majority of spectral characteristics showed increased divergence in parapatry, suggestive of character displacement. This increased song divergence in parapatry appeared to affect behavioral responses to playback as discriminating responses were most prominent in parapatry and against parapatric songs. The clear behavioral impact of subspecies-specific song differences supports a potential role for song as an acoustic barrier to gene flow. The asymmetric nature of the responses suggests that song divergence could affect the direction of gene flow and the position of the subspecies-specific transition.

Bird song plays an important role in species recognition in the context of territoriality and mate attraction (Catchpole and Slater 1995; Collins 2004). Emerging differences in subspecies-specific songs have the potential to promote assortative mating that may ultimately lead to reproductive isolation (Grant and Grant 1997a,b; Price 1998; Panhuis et al. 2001; Servedio 2004). When two subspecies that have diverged to some extent in allopatry come into secondary contact, songs may converge again, remain distinct, or become even more divergent. Song convergence would degrade acoustic boundaries and could promote rather than restrict gene flow. Maintenance or further divergence of song differences is likely to yield a more solid acoustic boundary and could prevent hybridization and complete the process of speciation.

Oscine passerines, which include half of the close to ten thousand known avian species, develop adult song by social learning (see Kroodsma 1982; Hultsch and Todt 2004). Theoretical models have shown that plasticity through social learning can lead to rapid generation of geographic diversification of acoustic signals, which under certain conditions could promote assortative mating (Ellers and Slabbekoorn 2003; Lachlan and Servedio 2004; Servedio et al. 2009) and accelerate the rate of speciation (Grant and Grant 1997a; Lachlan and Servedio 2004). However, field observations suggest that learning makes song among populations more likely to converge instead of diverge with auditory contact (Grant and Grant 1996; Secondi et al. 2003). Despite the well-studied role of avian song in sexual selection, and the extensive literature on geographic song variation (Podos and Warren 2007), we still have very limited insight into whether there are conditions in which differences in learned songs can remain distinct or diverge when potentially hybridizing taxa meet.

In order for song divergence to have an effect on gene flow and thus contribute to the formation of reproductive barriers, geographic song variation must be perceived by the birds themselves. A thorough understanding of the impact of song variation on reproductive exchange between populations therefore requires investigation of the behavioral impact of song differences among potentially hybridizing taxa. Relatively few studies of the role of sexual selection in avian speciation have specifically tested for geographic variation in responses to divergent sexual signals (Pearson and Rohwer 2000; Patten et al. 2004; Colbeck et al. 2010; Ripmeester et al. 2010) and so the behavioral impacts of divergence in learned songs are still not well understood.

Divergence in song and associated responses could affect gene flow through male–male competition or via female choice (see e.g., Slabbekoorn and Smith 2002). If females select mates based on divergent (sub)species-specific cues, then assortative mating is likely to lead to reproductive isolation (Patten et al. 2004). Even if females use different cues for mate choice than males use to recognize competitors, subspecies-specific song variation can have an impact on gene flow solely as a result of male–male competition. If a male sings a song that is not recognized by conspecific males, then the singer will not likely be able to establish a territory nor subsequently attract a mate. The potential impact of male–male interactions on population divergence and reduced gene flow has been supported by a theoretical model (Ellers and Slabbekoorn 2003) and an aviary study on two hybridizing chickadee species (Poecile atricapillus and P. carolinensis; Bronson et al. 2003).

An ability to differentiate between divergent songs can be inferred from a decreased response to songs that differ from an individual's own song (Slabbekoorn and Smith 2002). Female responses are notoriously difficult to assess in the field, so many studies have measured responses of territorial males to playback of geographic variants of male songs, which are assumed to reflect female choice (e.g., Irwin et al. 2001; Seddon and Tobias 2007; Uy et al. 2009). Field studies provide support for this assumption by demonstrating that both sexes respond more strongly to songs of their own subspecies than to songs of a closely related subspecies (Baker and Baker 1990; Patten et al. 2004; Derryberry 2007).

The gray-breasted wood-wren Henicorhina leucophrys provides an excellent model system to test for a possible role of song divergence in reproductive isolation between potentially hybridizing populations. Wood-wrens are oscine passerines and are presumed to learn their songs, as has been shown for several other wren species (Kroodsma 1974; Morton 1987; Kroodsma et al. 2002). Henicorhina leucophrys occurs as two genetically distinct subspecies that occupy adjacent yet distinct ecological niches in montane forest areas of Ecuador (Dingle et al. 2006, 2008). One subspecies, H. l. hilaris, occurs at relatively low elevations (∼500–1800 m) and is restricted to the west slope of the Andes and an isolated mountain range near the coast. The other subspecies, H. l. leucophrys, occurs at relatively high elevations (∼1800–2500 m) and is found on both outer slopes of the Andes, but not on the Andean peaks nor in the inter-Andean valley (Fig. 1). A previous study showed that the two subspecies form genetically monophyletic subgroups with consistent habitat-dependent differences in several song characteristics (Dingle et al. 2008). For example, leucophrys has a slower song than hilaris, with a lower delivery rate of relatively long notes. Leucophrys also sings notes with a broader frequency range than hilaris, and this is associated with a relatively less-noisy environment in the high frequency range for leucophrys habitat due to divergent invertebrate communities (Dingle et al. 2008). These genetic and acoustic differences suggest that the two subspecies have diverged significantly in allopatry. However, they still hybridize to some extent in zones of secondary contact, indicating that prezygotic isolation is not yet complete (Halfwerk et al., unpubl. data).

Figure 1.

Divergence in bird song and behavioral response. (A) The sampling transect from coastal Ecuador across the Andes. (B) Song examples from each of the four sampling sites. (C) Mean and standard deviation for mean frequency bandwidth (bandwidth of individual notes averaged across each song). Sample sizes inside bars show number of individuals; arrows with asterisks indicate significant differences between pairs of populations. (D) Results of the first four playback experiments (PC1 shown as the combined response measure). Within each pair of bars, bars on the left represent responses to hilaris songs, bars on the right represent responses to leucophrys songs; “OWN” means that the bar represents responses to playback of own subspecies songs, whereas “FOR” means that the bar represents responses to playback of songs from the other (foreign) subspecies. Asterisks over individual bars indicate that the response to playback differed significantly from preplayback activity. Asterisks over brackets between pairs of bars indicate that the response to own and other subspecies’ songs differed significantly in that population.

We studied four populations of the gray-breasted wood-wren (one allopatric and one parapatric population of each subspecies) to examine song variation and the behavioral impact of acoustic differences. For the acoustic analysis, we tested whether differences in subspecies-specific songs were as great in parapatry as was found in allopatry, and examined patterns of change in song characteristics between allopatric and parapatric populations within each subspecies. To determine whether song variation has biological relevance, we conducted playback experiments to test experimentally whether both subspecies discriminated between songs of their own and the other subspecies. If subspecies-specific acoustic differences are important for recognition of relevant competitors or mates, then individuals should respond more strongly to song of their own subspecies than to song of a different subspecies. Upon finding that one of the subspecies (leucophrys) consistently discriminated between own and the other subspecies’ songs, we conducted additional playback experiments in the parapatric leucophrys population to further explore the responses of males in this population. We compared relative response levels to intraspecific variation in hilaris songs as well as in songs of their own subspecies, and to a familiar control species without any biological relevance. If response to the other subspecies is affected by the presence of a nearby population, then we predicted that parapatric leucophrys would show the weakest response to parapatric hilaris songs.



We collected song recordings from male wood-wrens at four sites: (1) allopatric hilaris: coastal Cordillera de Colonche at Loma Alta (900 m; 1°52′ S, 80°34′ W); (2) parapatric hilaris: west slope of the Andes at Mindo (1500 m; 0°03′ S, 78°45′ W); (3) parapatric leucophrys: west slope of the Andes at Bellavista (2200 m; 0°02′ S, 78°41′ W); and (4) allopatric leucophrys: east slope of the Andes at Yanayacu (2000 m; 0°36′ S, 77°53′ W). Both allopatric populations were separated from the parapatric populations by large areas of unsuitable habitat (the high Andean peaks and inter-Andean valley in the case of leucophrys and lowland rainforest in the case of hilaris). The two parapatric populations used in this study were adjacent to each other, separated by less than 10 km along a steep elevational gradient on the west slope of the Andes (Fig. 1A), with a small zone of overlap between the two subspecies. On the west slope, the wood-wrens of both subspecies are distributed continuously along an elevational gradient and occur at very high densities, so dispersing individuals from each of the two parapatric populations are likely to at least occasionally move into the breeding range of the other subspecies.

Both male and female gray-breasted wood-wrens sing, either as solos or as duets. Male solos are the most common song type sung, whereas female solos are rare (Dingle 2009). Males sing two different types of solos: one where songs are repeated at a fast rate (“fast” song) and the other where songs are repeated at a slower rate, with more time between each song repeat (“slow” song) (Dingle 2009). In this study, we focused on male fast song characteristics as this singing mode was most frequently recorded and song characteristics did not appear to differ qualitatively from male song used in duets. In addition, overlapping songs render sonographic measurements on duet recordings less reliable compared to solo songs.

We recorded male wood-wrens from February through May 2005 and September through October 2006. In 2005, all recordings were collected using a Sennheiser ME67 directional microphone (Sennheiser Electronic Corporation, Old Lyme, CT) and a Sony TCM-5000EV tape recorder (Tokyo, Japan). In 2006, the same microphone was used with either a Sharp MD-MT190H(S) (Osaka, Japan) or an M-Audio Micro Track 24/96 digital recorder (Irwindale, CA), using a 16-bit, 44.1 kHz recording rate. The two digital recorders did not compress sounds when recording, and there were no obvious differences in recording quality between these devices and we believe that device-dependent variation is highly unlikely to affect any aspect of our analyses. We also tested and excluded possible seasonal effects on song characteristics or recording quality.


We recorded and analyzed 1074 songs of 65 individuals (mean 16.8 ± 9.6 songs per individual). Song recordings were digitized at a sampling rate of 44.1 kHz and sonograms were generated (FFT size = 1024, 80% frame overlap) using Luscinia software for acoustic analyses (Lachlan 2007). These settings led to an effective spectral resolution of 10.5 Hz for the individual means. We measured the following spectral and temporal song characteristics: mean and maximum frequency bandwidth of individual notes; peak, maximum, and minimum frequency within a song; song duration; number of notes per song; and note length. Song duration and number of notes per song were used to calculate the song delivery rate, defined as the number of notes sung per second. These measures were selected based on a previous study of subspecies-specific variation across Ecuador, which indicated that these measurements varied significantly between subspecies (Dingle et al. 2008). All variables were tested for normality using the Kolmogorov–Smirnoff test and a few variables were log-transformed to meet requirements for parametric testing (delivery rate, note length, maximum bandwidth, and maximum frequency). Statistical tests of patterns of divergence were conducted using a factorial analysis of variance (ANOVA), with subspecies and population (allopatry or parapatry) as independent categorical predictors. For the song characteristics that showed a significant interaction between subspecies and population, Tukey's multiple comparison tests were used post-hoc to test for within-subspecies differences between allopatric and parapatric populations. All analyses were conducted using Statistica 7 (StatSoft Inc.).


We conducted seven playback experiments to test receiver response patterns. The first set of four playback experiments compared the behavioral response of territorial males in each of the four study populations to playback of songs of their own and the other subspecies. The main aim of these experiments was to test whether the two subspecies showed signs of being able to discriminate between the divergent subspecies-specific songs. In the allopatric populations, males heard stimuli recorded from the two parapatric populations (Figs. 2A, C). In the parapatric populations, males heard own-subspecies’ songs recorded from the distant allopatric population and the other-subspecies’ songs recorded from the nearby parapatric population (Figs. 2B, D). Because there is also song variation between populations of the same subspecies, birds may respond stronger to own-subspecies’ songs from their own population than to own-subspecies songs from a more distant population. Our experiment was designed to exclude the potential for such familiarity with a local dialect to affect our tests of subspecies-specific discrimination. By playing back a distant and unfamiliar variant of own-subspecies song, we take a conservative approach and, as a consequence of our design, we can exclude the possible impact of particularly high response levels to a local dialect and attribute a reduced response strength to discrimination against other-subspecies songs.

Figure 2.

Schematic overview of six of the seven playback experiments. Arrows indicate from which two populations stimuli were taken for playback in the test population in each experiment. The top two rows represent experiments one through four; top row: playbacks to hilaris in (A) allopatry and (B) parapatry; middle row: playbacks to leucophrys in (C) allopatry and (D) parapatry. The bottom row represents experiments five and six: playback experiments to leucophrys in parapatry, with (E) songs of hilaris originating from allopatric and parapatric populations, and with (F) songs of own subspecies originating from the test population and the allopatric population.

After discovering that one subspecies (leucophrys) was discriminating between the subspecies-specific songs, we further explored the impact of more subtle song variation on the behavioral response for this subspecies in parapatry. In experiment five, we tested the response of leucophrys males to allopatric and parapatric hilaris, with the prediction that parapatric leucophrys would respond less to parapatric hilaris males than to the allopatric hilaris males (Fig. 2E). Experiment six tested the response of parapatric leucophrys to leucophrys songs from their own population and from the allopatric population (Fig. 2F). This was to test the impact of familiarity with a local dialect on response strength and allowed for an evaluation of discrimination and absolute response levels of the between-subspecies comparisons relative to this within-subspecies comparison. In experiment seven, we tested the response of parapatric leucophrys males again to parapatric hilaris songs but now compared to the response to a locally common heterospecific bird species that occurs syntopically with the wood-wrens: the russet-crowned warbler Basileuterus coronatus. This control experiment allowed us to compare responses to wood-wren songs to any behavioral changes related to setting up equipment and playing back a natural sound without biological relevance to the wood-wrens.

Each experiment involved 20 trials in 20 different territories. Territories were not used twice across experiments, except for in the parapatric leucophrys site where we conducted four playback experiments. At that site, territories were not used twice within an experiment, and subsequent trials in the same territory were always separated by at least 50 days. Ten unique pairs of song stimuli were used in each playback experiment. Avisoft SASlab Pro (R. Specht, Berlin, Germany) was used to process recordings and to create playback stimulus files. We selected only high-quality recordings with high signal-to-noise ratios, which were filtered with a digital high-pass filter at 1.0 kHz and normalized afterwards at the peak amplitude of the songs. A Creative Zen Touch 20 GB player was used for playback through a Creative Travelsound 400 portable speaker (Singapore). Playback levels were standardized at 85 dB(A) at 1 m from the speaker, as measured with a Sphynx digital (Sphynx Electronics, Kenzingen, Germany) sound pressure level meter. Each of the 20 trials in an experiment had a unique responding male, and stimulus pairs were used in two trials, but each time in a different order. Our setup limits the potential impact of pseudo-replication (Kroodsma 1989; Slabbekoorn and Bouton 2008), while optimally exploiting the available song recordings. A stimulus pair consisted of 2 min of recordings of a single individual from one site and 2 min of a single individual from another site. Each stimulus consisted of four bouts of four to 10 songs (for a total of 28 songs) to mimic natural singing behavior (total singing time per stimulus = 28.6 sec ± 1.59 standard error).

Playback trials lasted 30 min and consisted of two test periods, one for each stimulus. Each test period consisted of 5 min of preplayback observation (preplayback period), 2 min of playback and 3 min of postplayback observations (combined together as response period). The second playback test period was conducted on the same territory after a 10-min break. The playback speaker was attached to vegetation 1 m above the ground, a height at which wood-wrens are commonly found. Colored tape was used as a distance marker at 2 and 5 m around the speaker. We recorded movements and vocal responses continuously throughout the 30-min trials. Response variables were scored separately for the preplayback and response period and included: approach distance, response duration, and vocal output. Approach distance was measured as the minimum distance between subject and speaker. Distance categories were: 0 = less than 16 m; 1 = 8–16 m; 2 = 4–8 m; 3 = 2–4 m; and 4 = less than 2 m (after Nelson and Soha 2004). Response duration was measured as the accumulated time between initial response and the time at which birds were not visible or audible anymore. Vocal output included the total number of song bouts (fast and slow solos and duets) sang by the responding male.

We combined the three response variables into a single variable using principal component analysis (PCA). PCA was conducted on each of the seven experiments separately, as each experiment was independent from the other. PC1 explained 63.6 to 77.2% of the variation in the seven datasets (Eigenvalues 1.91 to 2.32) and all three response measures always correlated highly with this first principal component (average factor loadings were: 0.87 for minimum approach distance; 0.90 for response duration; and 0.75 for vocal output; see Table S1), so PC1 was used for further analysis. Two statistical analyses were conducted on the playback data to test for variation in response patterns to the two stimuli presented in each experiment. First, we tested whether overall responsiveness differed between the two different stimuli presented in each trial (Wilcoxon signed rank test; n= 20 for each experiment). We used nonparametric tests as response data were not normally distributed, and transformations were not able to correct for this. Second, we compared behavior during the response period to behavior recorded during the preplayback period to assess whether the stimuli led to a detectable behavioral change over “normal” levels of activity. In the preplayback period, we measured distance to the speaker to determine whether birds moved into or out of the area after the playback period. “Response duration” measured after playback was compared to “activity time” during the preplayback period (the total time spent singing prior to playing the stimulus). Wilcoxon signed rank tests were also used for this analysis, on only those trials in which the target stimulus was first in order of playback (n= 10 for each stimulus set). The second periods in a trial were not used here to avoid carry-over effects of the first stimulus onto the second.

Order effects were tested for by comparing responses to a stimulus when it was broadcast first in a trial versus when it was broadcast second, with a Mann–Whitney U test. Such an order effect was found only once, and the significance level of the stimulus comparison did not change when we just compared responses to the stimuli broadcast first.



Acoustic analysis of hilaris and leucophrys songs revealed significant acoustic differences between songs of the two subspecies for all parameters (population means in Table 1, results of factorial ANOVA in Table 2). All of the song parameters remained distinct in parapatry, but there were different patterns of change between allopatric and parapatric populations within subspecies; most of the song parameters remained the same or diverged between allopatric and parapatric populations, while one song parameter (minimum frequency) revealed convergence in parapatry (Tables 1 and 2).

Table 1.  Mean values±standard deviations (SD) for the seven song variables measured from parapatric and allopatric populations of hilaris and leucophrys. Number in parentheses after allopatry or parapatry indicates the number of males analyzed from each site.
FactorHilaris allopatry (n=12)Hilaris parapatry (n=11)Leucophrys parapatry (n=11)Leucophrys allopatry (n=31)
Mean frequency bandwidth1241.46±142.98 829.99±110.942085.44±188.671784.21±241.98
Max frequency bandwidth2191.36±372.881784.74±301.693823.28±302.633295.63±513.55
Peak frequency2461.68±145.652360.02±82.413755.42±288.263376.77±362.64
Maximum frequency3902.75±571.773471.00±244.827662.63±422.016914.64±926.75
Minimum frequency1038.64±70.731070.16±159.041288.78±105.791464.80±174.40
Delivery rate  12.88±1.39  11.97±0.76   6.70±0.61   6.59±0.79
Note length  52.22±4.92  55.11±7.52 104.87±9.50 110.73±16.51
Table 2.  Results from a factorial ANOVA testing for differences in spectral and temporal song characteristics between allopatric and parapatric populations of both subspecies. N=65 (the number of individual males compared). Tukey's post-hoc tests were used to test for differences between allopatry and parapatry for hilaris and leucophrys separately. Numbers in bold represent significant results.
FactorSubspeciesPopulationSubsp×Population Allopatry vs Parapatry (Tukey's post-hoc test)
  1. *These variables were log-transformed to meet normality assumptions.

Mean frequency bandwidth277.681<0.0011.0440.31143.62<0.001<0.01<0.01
Max frequency bandwidth*187.9<0.0010.30.56617.90<0.0010.010.03
Peak frequency223.075<0.0013.2060.078 9.640.003 0.82<0.01
Maximum frequency*454.4<0.0010.01.00011.900.001 0.120.05
Minimum frequency66.232<0.0013.3260.073 6.860.011 0.950.01
Delivery rate*467.16<0.0010.740.392 2.41 0.126 0.39 0.95
Note length*396.49<0.0010.000.993 1.86 0.178 0.92 0.95

The most notable difference between the songs of the two subspecies related to the presence of wide bandwidth notes in leucophrys songs that were rarely used in hilaris songs (Fig. 1B). Wide bandwidth notes were especially wide in the parapatric leucophrys population, completely absent in the parapatric hilaris population, and distinct to only a moderate degree in both allopatric populations (Table 1). Post-hoc analyses confirmed that the songs of both subspecies differed significantly between parapatric and allopatric populations for bandwidth measures (Table 2), with both subspecies shifting their songs away from the other subspecies in parapatry (Table 1).

Maximum and peak frequency were also more divergent between parapatric populations than between allopatric populations, with both subspecies shifting their songs away from the other subspecies in parapatry (Table 1), although this was only significant for leucophrys (Table 2). In contrast, minimum frequency showed a pattern of convergence, with leucophrys converging toward hilaris in parapatry (Tables 1 and 2). Temporal measures differed significantly between the two subspecies, and these differences were maintained between allopatric and parapatric populations within each subspecies but did not show signs of divergence or convergence (no interaction between subspecies and population, Table 2).


Experiments 1–4

Territorial leucophrys males discriminated clearly between songs of their own and the other subspecies, responding less to hilaris songs than to leucophrys songs in both allopatric and parapatric populations (Fig. 1D, Table 3). In addition, allopatric and parapatric leucophrys males differed in their responses to parapatric hilaris songs. Individuals from the allopatric leucophrys population showed a low-level response to hilaris songs, which was significantly higher than activity during preplayback periods (Fig. 1D, Table 3). In contrast, the parapatric leucophrys population, where leucophrys males could potentially meet individuals singing hilaris songs, individual males showed no significant response to hilaris songs (no difference between activity during the preplayback and response periods, Table 3). This divergence in response activity was not found in hilaris; territorial males from both hilaris populations responded equally strongly to songs of both subspecies (Fig. 1D, Table 3). In all cases, the response levels of hilaris males were significantly greater than preplayback activity levels (Table 3).

Table 3.  Wilcoxon signed rank tests of responses to playback in Experiments 1–4. “Own vs Other” shows results of the “across conditions” test (response to own song vs. the other subspecies’ songs). “Own vs pre” and “Other vs pre” show the results of the “within conditions test,” comparing responses during the playback period to a baseline level of activity as measured during the preplayback period (“pre”) for each trial. Ha and Hp refer to the allopatric and parapatric hilaris populations, respectively, and similarly La and Lp refer to the two leucophrys populations.
ExperimentSubjects Own vs other Own vs pre Other vs pre
Exp1Ha−0.221 0.836−3.593<0.001−2.9470.002
Exp2Hp−0.370 0.734−3.1540.001−2.4850.011
Exp4Lp−3.724<0.001−3.724<0.001−1.581 0.125

Experiments 5–7

Males in the parapatric leucophrys population discriminated not only on the basis of subspecies-specific variation, but also discriminated between allopatric and parapatric hilaris songs. They showed a detectable response (a significant difference between activity during playback and pre-playback periods) only to allopatric hilaris songs and no response to parapatric hilaris songs (Experiment 5; Fig. 3, Table 4). The response to allopatric hilaris songs was also significantly stronger than the (lack of a) response to parapatric hilaris songs (Table 4).

Figure 3.

Response of leucophrys to near and distant populations of own and hilaris songs. Response strength as reflected by PC1, the combined response measure. Asterisks over individual bars indicate a significant difference between activity in response to the stimulus and preplayback activity. Asterisks over brackets between the pair of bars on the left indicate a significant difference in response to allopatric and parapatric hilaris populations (Ha and Hp, respectively).

Table 4.  Wilcoxon signed rank tests of results from Experiments 5–7, all performed in the parapatric leucophrys population. “A” and “B” refer to the first and second stimulus presented during each playback trial. RCWA stands for russet crowned warbler (B. coronatus), the heterospecific stimulus. For more details, see the legend for Table 3.
ExperimentStimuli A vs B A vs B A vs pre B vs pre
Exp 5Ha v Hp−2.3410.017−3.059<0.001−2.023 0.062
Exp 6La v Lp−0.9330.368−3.575<0.001−3.883<0.001
Exp 7Hp v RCWA−0.9440.438−0.507 0.688 – –

The response of parapatric leucophrys to its own subspecies’ songs was strong and independent of whether the stimulus came from an individual's own or a distant population (Experiment 6; Fig. 3, Table 4). The low response level of parapatric leucophrys males to playback of parapatric hilaris songs was not different from the low response level to a different species (Experiment 7; Table 4), nor was it again any different than activity recorded during the preplayback intervals.


We found consistent differences in both spectral and temporal characteristics between songs of two subspecies of the gray-breasted wood-wren, even between two adjacent and potentially hybridizing populations. Although individual song parameters showed different patterns of change between allopatric and parapatric populations, spectral song parameters showed a striking pattern of divergence in parapatry. This was especially true for the presence of wide-bandwidth notes, with song differences between the two parapatric populations being greater than between the two allopatric populations. Playback results revealed a strong asymmetric response pattern. The high-altitude leucophrys subspecies responded strongly to its own subspecies’ songs, but less or not at all to the other subspecies’ songs. The low-altitude hilaris subspecies responded strongly to all songs of both subspecies, independent of origin of recording and location of playback (allopatry or parapatry).

Interestingly, the increased spectral song divergence in parapatry relative to allopatry was reflected twice in playback response patterns of the discriminating subspecies (leucophrys). The responses of leucophrys males to hilaris songs recorded in parapatry were weak in allopatry (but significant) but totally absent in parapatry. Furthermore, although this absence of any response to parapatric hilaris songs was shown repeatedly for birds in this parapatric population, parapatric leucophrys did show a detectable response to allopatric hilaris songs.


Our first main finding concerns the consistent song differences between the two subspecies in line with our previous study on geographic variation in 10 populations across Ecuador (Dingle et al. 2008). Although the two parapatric populations of gray-breasted wood-wrens were less than 10 km apart, subspecies-specific songs remained clearly distinct. This is remarkable because these birds are presumed to learn their songs from neighboring conspecifics (Kroodsma 1974; Morton 1987; Kroodsma et al. 2002). Given the small distance between the sampling locations, and that territories of both subspecies occur at high densities up to the point of altitudinal transition, occasional encounters among individual birds from both subspecies are likely. Also, not far from where we sampled, birds will have more regular auditory experience with the other subspecies’ songs. Previous empirical studies in other songbird species suggest that this situation would be more likely to lead to song copying across subspecies-specific boundaries and to acoustic convergence (Secondi et al. 2003), which may eventually yield genetic leakage through hybridization (Grant and Grant 1996, 1997a).

However, convergence is not the only possible outcome in all circumstances and there are many examples in which intraspecific dialectal variation is maintained with distinct acoustic boundaries between potentially hybridizing taxa (Podos and Warren 2007). Dialectal variation is characterized by a disjunct transition in acoustic variation and could apply to neighboring territories; one neighbor singing one dialect, the other neighbor singing another dialect. Dialects and dialect boundaries may remain for long periods of time (Derryberry 2007; Wright et al. 2008), even when mixed singers occur in areas of auditory contact (Baptista and Morton 1982; Caro et al. 2009) and dispersal of individuals between dialects is not necessarily restricted in any way (Baker and Cunningham 1985; Slabbekoorn and Smith 2002). The potential role for distinct geographic variants of song in assortative mating may be enhanced by ecological factors, and if song provides acoustic cues for finding locally adapted mates (MacDougall-Shackleton et al. 2002; Slabbekoorn and Smith 2002).

We envision two processes by which the ontogeny of song and cultural transmission could counteract between-group convergence of culturally inherited traits. First, song learning may be biased to parental songs through an early sensory phase and predispersal learning (Ellers and Slabbekoorn 2003; Hultsch and Todt 2004) or through restricted dispersal with settlement next to, or within, parental territories (e.g., in the case of cooperative breeding), as is assumed typical for many tropical species (Greenberg and Gradwohl 1997). Second, birds can have genetic predispositions to learn preferentially from conspecific tutors in the presence of heterospecific alternatives (Marler and Peters 1977). Such genetic guiding of song copying has been shown to occur at the subspecies level (Nelson 2000). For a better understanding of the stability of acoustic boundaries in learned songs, within and between species or subspecies, there is clearly a need for more detailed data on ontogenetic pathways and the effect of song exposure under field conditions.


Our playback results showed a clear pattern of asymmetric response to mutually divergent sexual signals: male wood-wrens of only one subspecies discriminated between divergent songs, whereas the other subspecies responded equally to both song types. Although relatively few studies have considered variation in response to divergent sexual signals, the evidence to date suggests that such an asymmetric response may not be uncommon. One of the first documented examples of an asymmetrical response pattern was between two Hawaiian Drosophila species (Kaneshiro 1976). Since then, asymmetrical responses have been reported in other insects (Polynesian field crickets Teleogryllus oceanicus: Tinghitella and Zuk 2009), reptiles (Carlia skinks: Dolman 2008), anurans (green-eyed treefrogs Litoria genimaculata: Hoskin et al. 2005), and birds, both within species (black-throated blue warblers Dendroica caerulescens; Colbeck et al. 2010) and between species (Dendroica warblers Pearson and Rohwer 2000; Vermivora warblers: Martin and Martin 2001). Despite the fact that asymmetric responses have been documented in a wide variety of taxa, the mechanisms causing such a skew in response strength are not well understood. Kaneshiro hypothesized that asymmetric responses were due to relaxation of female choice in derived populations, driven by a full or partial loss of a male's sexual signal during a founding event (Kaneshiro 1976; Kaneshiro and Boake 1987). However, this explanation, which is based on intersexual interactions, has gained little empirical nor theoretical support (Arnold et al. 1996).

A second possibility is that asymmetric responses to heterospecific signals are a consequence of intrasexual interactions and due to asymmetric competitive ability or aggressiveness between two taxa. Such differences in aggressiveness are thought to explain the asymmetric response to heterospecific playback in Townsend's warblers and hermit warblers (Pearson and Rohwer 2000). Townsend's warblers respond strongly to mounts of both species, whereas hermit warblers respond more strongly to conspecific mounts. Townsend's warblers also respond more strongly to hermit warbler mounts than do hermit warblers themselves, suggesting that Townsend's warblers are more aggressive overall. Pearson (2000) argues that this asymmetry in aggression results in Townsend's warbler males more easily establishing territories and attracting mates, thereby outcompeting and replacing hermit warblers across a moving hybrid zone. In our study, there is no indication that hilaris males respond more aggressively in general to playback of song than leucophrys males. Nevertheless, if the response asymmetry alone could give hilaris a similar advantage over leucophrys in territory establishment and mate attraction, it could mean that hilaris is replacing leucophrys in an upward distribution shift. However, we have no data on the hybrid or contact zone moving up or down the Andean west slope. Furthermore, the sharp ecological gradient along which these two subspecies come into contact, together with potentially strong local adaptations of both subspecies, may restrict such dynamic changes in distribution in the wood-wrens.

A third, mechanistic, explanation for the asymmetric response pattern is that the subspecies have a skewed perceptual sensitivity due to distinct overlap differences with respect to the frequency ranges used by each subspecies. Leucophrys songs overlap almost completely in frequency range with hilaris songs, although hilaris songs only cover part of the frequency range of leucophrys and do not contain the particularly high-frequency end of the distinctive wide bandwidth notes typical of leucophrys songs. It is well known that spectral features can play a critical role in triggering behavioral responses (Nelson 1988; Slabbekoorn and ten Cate 1998) and in our study the amount of frequency overlap between the songs used for playback and the songs of the local population correlates well with response strength. The positive correlation between frequency overlap and response strength is not only true for comparisons of songs between the subspecies, but is also reflected in the more subtle variation in response strength to parapatric and allopatric songs of the same subspecies. Spectral similarity could translate directly and inherently into aggression level, but parapatric leucophrys males may also have learned, through experience with parapatric hilaris song, to use frequency overlap as a cue to avoid responding to foreign subspecies. Further studies are needed to test whether and how previous experience with the other subspecies play a role and to explore the fitness consequences of discrimination.

Independent of the underlying mechanism, the discrimination of subspecies-specific song differences in at least one subspecies provides evidence that vocal interactions among males have the potential to affect gene flow across acoustic boundaries. Although it remains to be tested whether male responses to playback reflect female song preferences in the wood-wrens, a similar asymmetric female response pattern could theoretically lead to biased introgression of male leucophrys into the distribution area of hilaris, and female hilaris introgression into the distribution area of leucophrys. Such patterns of introgression can be investigated by collecting integrative data on song and genetic make-up of individuals around the acoustic boundary. We believe the wood-wren model system provides excellent opportunities to investigate such rarely explored phenomena.


Character displacement refers to a pattern of increased trait divergence that is due to the presence of a closely related species (or subspecies) in sympatry or parapatry (Brown and Wilson 1956; Grant 1972). Character displacement is a concept typically applied to genetically inherited traits (acoustic examples: Gerhardt 1994; Seddon 2005; Jang and Gerhardt 2006; Kirschel et al. 2009), but it is interesting to consider whether a culturally transmitted trait could also yield such a pattern. Two previous empirical studies provide some evidence for character displacement in learned traits. Blue tits (Cyanistes caeruleus) add a trill to the end of their songs by which they avoid interspecific aggression in areas of sympatry with larger great tits (Parus major) (Doutrelant et al. 2000). In a study on collared and pied flycatchers (Ficedula albicollis and F. hypoleuca), Haavie et al. (2004) found a pattern of asymmetric character shift of learned songs in sympatry: pied flycatcher songs converge toward collared flycatcher songs through heterospecific song copying, whereas collared flycatcher songs diverge from pied flycatcher songs, thereby maintaining species-specific song differences.

In our study, acoustic analysis revealed that several song parameters show a distinct pattern of divergence consistent with character displacement. The mutual spectral divergence was perceptually salient to the human ear and playback experiments provided evidence for an impact of the song divergence on territorial responses by the birds themselves. However, a pattern of increased divergence in parapatry alone is not enough to provide convincing evidence for character displacement as such a pattern could arise for reasons other than the presence of another subspecies. For example, a pattern of divergence among four populations could emerge from a mosaic (random) pattern of dialectal variation. Alternatively, the sampled populations could be the endpoints of two independent geographic clines, giving the appearance of character displacement in parapatry, as was shown when a classic case of ecological character displacement was reanalyzed using additional allopatric populations (Sitta nuthatches, Grant 1975). These alternatives need to be ruled out in order for the current study to be considered a strong case of character displacement. Although we have selected the most suitable set of four populations available to conduct a well-balanced, reciprocal series of playback experiments to test for auditory recognition of subspecies-specific songs, we only provide an unreplicated case study in terms of the song and response patterns congruent with character displacement. However, there is not much opportunity to replicate the experiment with this set of subspecies as the allopatric hilaris population occurs in the uniquely isolated coastal Cordillera de Colonche and we are not aware of any other truly allopatric populations. Therefore, further studies are required to achieve a more complete insight into whether learned birdsong, and the responses they elicit, can provide clear cases of character displacement, but future replication will have to exploit opportunities in subspecies or species other than the wood-wren subspecies pair we studied here.


In conclusion, the two subspecies of the gray-breasted wood-wren provide a clear example of a distinct acoustic boundary in learned birdsong. We believe this is remarkable especially in view of the small distance between the parapatric populations. Our playback results reveal a consistent asymmetric response pattern, largely independent of location (parapatry and allopatry). The fact that one subspecies strongly discriminates between the subspecies-specific songs suggests a potential role for acoustic variation in promoting speciation and avoiding hybridization. The fact that the other subspecies does not discriminate between the same song differences suggests that restrictions on gene flow may reveal leakage patterns with a subspecies-specific bias. Ecology seems to play a key role at multiple levels in our wood-wren study system. We have shown that habitat-dependent ambient noise profiles shape subspecies-specific song divergence (Dingle et al. 2008), while strong local adaptation by both subspecies can be inferred from the ecological segregation on the steep Andean slopes with mutual exclusion in altitudinal zones (Dingle et al. 2006). Such parallel divergence in sexual traits and fitness-related traits may provide the ecological link required for a critical role for song in assortative mating (Slabbekoorn and Smith 2002).

Associate Editor: M. Webster


We thank the Ecuadorian government and the Museo Ecuatoriano de Ciencias Naturales for permission to work in the country. We thank the following for permission to work on their property: H. Greeney (Yanayacu Biological Station), The Bustamantes (San Isidro), T. Quesenberry and M. Tenorio (Mindo), and R. Parsons (Bellavista). Many thanks to I. Pen for advice on statistical analysis, and to J. Komdeur for advice and support to JP and DB. Thanks also to B. Duperron for permission to use her image in Figure 1. This manuscript was improved considerably after comments from P. den Hartog, C. Jiggins, K. Riebel, N. Seddon, T. B. Smith, C. ten Cate, M. Verzijden, J. Welbergen, and three anonymous reviewers. Funding for this study was provided by grants from the Gates Cambridge Trust, St. John's College, and the American Ornithologists’ Union to CD, a VSB grant to WH, and funds from Groningen University, the Marco Polo Fonds, and Dr. L.A. Bumastichting.