Songs of Darwin's finches diverge when a new species enters the community
Contributed by B. Rosemary Grant, October 12, 2010 (sent for review August 7, 2010)
Profile
June 20, 2011
Abstract
Bird species sing different songs and as a result rarely breed with each other. Species are not static but can shift in acoustic and morphological space, yet maintain their distinctiveness. Investigating such a situation in a community of Darwin's finches sheds light on the origin and maintenance of premating barriers between species. Explanations for songs divergence generally invoke morphological changes to the sound-producing apparatus, environmental changes influencing transmitting properties of song, avoidance of acoustical interference with other species, and random processes including copying errors. We investigated changes in songs of Geospiza fortis (medium ground finch) and Geospiza scandens (cactus ground finch) from 1978 to 2010 on Daphne Major Island, Galápagos. The habitat did not change significantly; however, the finch community changed. The socially aggressive congener Geospiza magnirostris (large ground finch), singing in the same frequency band (2–4 kHz), colonized Daphne in 1983 and increased in numbers. Temporal features of the songs of G. fortis and G. scandens, especially trill rate and song duration, diverged from G. magnirostris songs as it became increasingly common. Changes in song were not a passive consequence of a change in beak morphology. Instead they arose as a bias during song imprinting and production. Sons of both G. fortis and G. scandens sang faster songs than their respective fathers and thereby differed more from G. magnirostris in their songs than did their fathers. Divergence from an aversive or confusing stimulus during learning illustrates a “peak shift” that may be a common feature of song evolution and speciation.
Sign up for PNAS alerts.
Get alerts for new articles, or get an alert when an article is cited.
Accounting for the origins, persistence, and ultimate extinction of species is one of the most challenging problems for evolutionary biologists. In most cases the divergence of species from a common ancestor happened a long time ago; therefore the process must be inferred from genetic differences between extant species, and the causes are rarely understood. The smaller are the differences between species, and the more recent is their divergence from a common ancestor, the greater is the potential for discovering how and why one species gives rise to two. There are several model systems for studying speciation across a broad taxonomic range of closely related species, from Drosophila flies (1) to Helianthus flowers (2), and each has its advantages and disadvantages. Darwin's finches on the Galápagos islands of Ecuador are one such group. They are suitable for two reasons (3). First, the 14 species have been derived from a common ancestor comparatively recently, estimated to be in the last 2–3 My (3), and the small differences between them allow us to reconstruct the evolutionary transitions. Second, they occupy a diversity of islands, some of which are close to a pristine state, never having been inhabited, ensuring that any changes we observe are the result of natural conditions and not influenced directly by humans. The finches are easy to capture, measure, genotype, observe, and follow in successive generations.
Species coexist sympatrically as a result of barriers to the exchange of genes. The reproductive barrier between closely related species of Darwin's finches is behavioral: They occasionally hybridize with little if any intrinsic fitness loss (4). Experiments in the field have demonstrated that closely related species in the genus Geospiza living sympatrically can discriminate between each other on the basis of song in the absence of visual cues (5, 6) and on the basis of morphology in the absence of acoustic cues or movement (7). Furthermore, discrimination between allopatric populations of the same species is possible when they differ in song (5, 8, 9) or morphology (7).
Thus differences in morphology in association with song constitute the barrier to interbreeding (10, 11). What is the origin of this premating barrier to gene exchange? Divergence takes place initially in populations on different islands with different ecologies through natural selection and random genetic drift. Islands differ in their food supplies, and the role of natural selection when feeding conditions change has been well documented (6, 7). Phenotypic differences arising in allopatry may then be enhanced by selection when populations establish sympatry on secondary contact (12, 13). Premating barriers arise when the mating signals (song and morphology) diverge in parallel.
The question of how songs diverge is not so easy to answer, although it has stimulated much research (14). Explanations generally invoke random song changes such as those caused by an accumulation of copying errors, morphological changes to the sound-producing apparatus, environmental changes influencing the transmitting properties of song, or acoustical interference with other species (13, 15, 16).
Songs may change as a result of stochastic processes such as an accumulation of copying errors between generations in a manner analogous to random genetic drift (17, 18), and differences between populations could build up over time. Random change constitutes a null hypothesis for the alternative explanations that invoke systematic factors.
Body size affects syrinx volume and hence fundamental frequencies, in Darwin's finches (10) as well as in other species (19, 20). Beak size influences the rate of note (syllable) production and frequency range. Darwin's finches with large beaks are constrained to have songs of lower note repetition rate and narrower frequency bandwidth than birds with smaller beaks by their inability to open and close their beaks rapidly (21–23). Changes in song can therefore occur as a passive consequence of changes in beak morphology (21, 24) due to natural selection from an altered food environment (3).
Song features are adapted to the sound transmission properties of different habitats (25, 26). Owing to differential attenuation and degradation, songs with lower frequencies are transmitted better through dense vegetation than through open environments (10, 16, 20, 25). Rate of note repetition, or trill rate (21), also varies in relation to habitat features. Trills transmit better in open habitats than in forests where interference from the vegetation and reverberation blurs the distinction between notes (25). Two longitudinal studies of white-crowned sparrows (Zonotrichia leucophrys) showed that song changes were associated with more efficient transmission in altered environments over periods of 30 (27) and 35 y (28).
Competition for signal space among species using the same acoustic environment may also be a factor influencing change in song structure (14, 19). The other species may comprise closely related finch species but also distantly related species such as seabirds. Finches arriving at an island may encounter both a different habitat and a different constellation of species. The presence of congeneric species could select for divergence in song if there is a fitness penalty to interbreeding (16, 29) or more generally if there is a communication benefit from minimizing acoustical interference (30–32). Peak shift, a learning phenomenon whereby an individual responds more to an exaggerated form of its own signal in a direction away from a congeneric signal, is a mechanism that would lead to divergence between songs of related species as a consequence of direct social interactions (33–35). Therefore songs of the residents and colonists may gradually change over the next few generations under pressures for unambiguous transmission to other members of the same breeding population, both from the habitat and from other species.
Song Acquisition and Transmission
All Darwin's finch males sing a single short song. Song and parental morphology are learned from day 10 to 40 after hatching through an imprinting process coinciding with the time when fathers are singing and feeding their young (10, 36). This effect of early experience was shown experimentally many years ago with finches in captivity (10). Once learned, a song is retained for life. As in other imprinting processes, learning species-specific song early in life influences mate-choice behavior later in life (37–39), resulting in individuals mating according to their conspecific song (11, 36, 37). Such vertical cultural transmission of song from father to son (6, 36) is expected to be culturally stable (16, 40), with small changes resulting from an accumulation of copying errors (18). Females do not sing, although the pattern of mating reveals that they learn their father's song (36, 37, 41), as shown experimentally in other oscines (42). Hybrid males sing the same species-specific song as their fathers’ (36, 41) and not a mixture of songs of the parental species as occurs in a few other species (43).
Darwin's Finches on Daphne Major Island
We addressed the question of what drives song divergence by studying the songs of two species of finches on Daphne Major Island (34 ha) from 1978 to 2010. The species are Geospiza fortis, the medium ground finch, and Geospiza scandens, the cactus ground finch (Fig. 1), and they differ in beak morphology and in song (Fig. 2). Even though both species vary from island to island in morphology and song features (10), song remains discretely different between species in sympatry (Fig. S1).
Fig. 1.
Fig. 2.
The study is uniquely suitable for studying two of the factors hypothesized to be important in song divergence: acoustic avoidance of other species’ songs and morphological determinants of song. A third species, Geospiza magnirostris, the large ground finch, colonized the island in 1983 and established a breeding population and numbers built up gradually over the next two decades (Fig. 3). Acoustic interference with G. fortis and G. scandens is to be expected because all three species concentrate sound energy within the same frequency band of 2–4 kHz and songs of G. magnirostris are especially loud.
Fig. 3.
Song Changes Across Years
All Darwin's finch males sing one single short song, retained unaltered throughout life (36). Song variation within a population is continuously distributed. Examples in Fig. 2 reflect the range of song variation seen within G. fortis and G. scandens on Isla Daphne. We used measurements of individual songs recorded from 1978 to 2010, grouped according to the decade in which the recorded birds hatched (Table 1), to identify the song variables that changed (Methods). Song features were heterogeneous among years, in both G. fortis (MANOVA; Wilk's Λ = 0.836, F = 6.855, df = 9, 808, P = 0.0001) and G. scandens (MANOVA; Wilk's Λ = 0.573, F = 13.022, df = 9, 455, P = 0.0001). Component variables changed in one or both species. Changes were most pronounced in the last two decades in trill rate (number of notes per second), especially in G. scandens, and song duration (Table 1 and Table S1).
Table 1.
| Species year | No. of notes  ±SEM |
Internote interval, s  ±SEM |
Trill rate, no. of notes/s  ±SEM |
Bandwidth, Hz  ±SEM |
Duration of song, s  ±SEM |
Center frequency, Hz  ±SEM |
N |
|---|---|---|---|---|---|---|---|
| G. fortis | |||||||
| 1970 | 2.4 ± 0.12 | 0.05 ± 0.01 | 4.08 ± 0.25 | 3.1 ± 0.01 | 0.62 ± 0.02 | 3.6 ± 0.01 | 27 |
| 1980 | 2.9 ± 0.09 | 0.04 ± 0.03 | 4.59 ± 0.13 | 3.1 ± 0.01 | 0.65 ± 0.02 | 3.6 ± 0.01 | 120 |
| 1990 | 2.9 ± 0.09 | 0.03 ± 0.00 | 5.08 ± 0.15 | 3.1 ± 0.01 | 0.58 ± 0.02 | 3.5 ± 0.00 | 117 |
| 2000 | 3.3 ± 0.14 | 0.01 ± 0.00 | 6.05 ± 0.20 | 2.8 ± 0.01 | 0.53 ± 0.01 | 3.1 ± 0.00 | 72 |
| G. scandens | |||||||
| 1970 | 8.9 ± 0.64 | 0.05 ± 0.0 | 8.20 ± 0.28 | 1.6 ± 0.01 | 1.10 ± 0.08 | 2.9 ± 0.00 | 23 |
| 1980 | 9.4 ± 0.40 | 0.05 ± 0.00 | 8.32 ± 0.23 | 1.8 ± 0.01 | 1.16 ± 0.05 | 2.9 ± 0.00 | 61 |
| 1990 | 8.6 ± 0.60 | 0.04 ± 0.01 | 9.73 ± 0.71 | 1.7 ± 0.01 | 0.96 ± 0.07 | 2.9 ± 0.00 | 34 |
| 2000 | 6.4 ± 0.15 | 0.004 ± 00 | 10.55 ± 0.18 | 1.8 ± 0.01 | 0.61 ± 0.01 | 2.8 ± 0.00 | 70 |
| G. magnirostris | |||||||
| 1990 | 2.0 ± 0.00 | 0.20 ± 0.02 | 1.94 ± 0.06 | 2.7 ± 0.01 | 1.00 ± 0.03 | 3.5 ± 0.01 | 7 |
| 2000 | 2.0 ± 0.00 | 0.17 ± 0.01 | 1.80 ± 0.02 | 2.6 ± 0.01 | 1.10 ± 0.01 | 2.7 ± 0.00 | 57 |
Fluctuating changes in amount of vegetation also occurred as a result of pronounced annual variation in rainfall (3). The habitat is a mixture of cactus, shrubs, trees, and open areas, with no closed-canopy forest. The changes were mainly restricted to low-growing vegetation close to the ground where finches do not sing, whereas song perches are 1–5 m above ground on cactus (Opuntia echios) and trees (Bursera graveolens and Croton scouleri). Such changes in vegetation are ephemeral when considered against a finch life span. The average generation length of G. fortis is 4.5 y and of G. scandens is 5.5 y, the maximum age for both species being 16 y (3). Thus for singing finches the habitat remained the same, so changes in songs must have been caused by other factors.
Song Changes in Response to G. magnirostris: Tests of the Acoustic Interference Hypothesis
G. magnirostris is distinctly larger than G. scandens and G. fortis and persistently harasses both of them for nest sites in cactus bushes and when feeding. Song characteristics of the three species overlap, and therefore song divergence of the socially subordinate G. fortis and G. scandens from G. magnirostris is expected from the hypothesis of acoustic interference. The number of singing male G. magnirostris, paired and unpaired, increased from 3 in 1983 to a maximum of 51 in 2010 (Fig. 3). If G. magnirostris singing influenced the songs of G. fortis and G. scandens, we should expect their songs to change in temporal and spectral features in a direction away from G. magnirostris song characteristics. The expectation was not met with spectral features. The intercorrelated frequency variables shifted downward in all three species from the 1990s to the 2000s (Table 1 and SI Methods).
Songs of G. fortis and G. scandens are expected to differ from the songs of G. magnirostris increasingly in time, and more in the 2000s, when there were many of them, than in the 1980s and 1990s when there were fewer. We tested these expectations with the five temporal variables: trill rate (number of notes per second), song duration, average interval between notes, and duration of each of the first two notes. Because these variables are positively correlated with each other, we performed a principle components analysis (PCA) with the first three variables (SI Methods).
Both univariate and multivariate analyses showed that song traits in G. fortis changed significantly in the expected direction away from G. magnirostris (one-way ANOVAs). Duration of the song as well as duration of the first two notes and the average internote interval became shorter across the four decades, whereas trill rate, illustrated in Fig. 4, increased (Table 1). The average number of notes in a song increased. Most of these changes took place from the 1990s to the 2000s (Table 1 and Tables S1 and S2). Mean principal components (PC 1) scores varied strongly among the decades (F3,332 = 27.50, P = 0.0001). A post hoc test identified the decades that differed significantly at P = 0.01 as follows: 2000s > 1990s > 1980s = 1970s. Remaining variation captured by the second principal component was more homogeneous among decades (F3,332 = 2.64, P = 0.0495). The 1970s differed from the 2000s, but only at P = 0.05.
Fig. 4.
G. scandens showed the same trends (Table 1), except that the average number of notes in the song decreased (Table 1), and duration of the first two notes did not change (Table S1). As in G. fortis, mean PC 1 scores varied strongly among the decades (F3,182 = 27.50, P = 0.0001) and in the same way: 2000s > 1990s > 1980s = 1970s. PC 2 scores did not vary (F3,182 = 2.61, P = 0.0529). Thus in both species changes were in the direction away from G. magnirostris in acoustical space, consistent with the hypothesis of acoustic avoidance of G. magnirostris songs (Fig. 5).
Fig. 5.
Song Change Related to Morphological Change Caused by Character Displacement from G. magnirostris: Tests of the Biomechanics Hypothesis
A severe drought began 20 y after G. magnirostris arrived on the island, causing competition with G. fortis for the large and hard Tribulus seeds that constitute the diet of both species. As a result, large-beaked G. fortis died to a disproportionate extent, and average beak size diverged away from G. magnirostris (Fig. 6). This character displacement event was restricted to the influence of G. magnirostris on the G. fortis population (44); morphological features of G. scandens did not change.
Fig. 6.
Changes in G. fortis song might have been a consequence of changes in beak size (Fig. 6). A biomechanics hypothesis (21) predicts a decrease in beak size will cause an increase in frequency bandwidth and trill rate. Given the trade-off between these two song features, we tested the hypothesis with a new metric: frequency bandwidth of the first note divided by its duration as a measure of time-adjusted frequency sweep. The first note was chosen because it has an up-sweep and a down-sweep in the same note (Fig. 2). Two tests were performed with 251 G. fortis individuals. In the first test we compared the frequency sweep of songs grouped in the four decades, expecting no differences except between the 1990s and the 2000s, that is to say before and after the character displacement; the frequency sweep should be greater in the 2000s than in the 1990s. Frequency sweep was heterogeneous among decades (Kruskal–Wallis ANOVA, χ23 = 12.15, P = 0.0069), as a result of a difference between the 1980s and the 1990s [Tukey–Kramer honestly significant difference (HSD) test, significant at 0.05 but not at 0.01]. However, there was no difference between the 1990s and the 2000s (P > 0.1). In the second test we used the dataset in a single analysis. Frequency sweeps were predicted to be inversely correlated with beak size (PC 1), but in fact the two variables were not correlated (r = 0.08, P = 0.287). We obtained the same results when frequency bandwidth was used without the note duration adjustment, as was done in the original development of the biomechanics hypothesis (21).
The maximum frequency in G. fortis songs decreased in the 2000s. At this time beak sizes decreased whereas, given a decrease in maximum frequency (Table S2), the hypothesis predicts an increase in beak size. Note the average frequency bandwidth is approximately the same in G. fortis as it is in the much larger G. magnirostris (Table 1). Trill rate is hypothesized to be negatively associated with beak size. As predicted, the increase in trill rate of G. fortis songs from the 1990s to 2000s was associated with a decrease in beak size (F1,160 = 13.681, P = 0.0003). However, this is not likely to be a functional association as the trill rate in G. fortis is correlated so weakly with beak size (r = −0.14, df = 1, 235, P = 0.031): Only 2% of the variation in trill rate is explained statistically by variation in beak size. A further complication is that beak size is positively correlated with body size (r = 0.81, P < 0.0001) and body size also decreased across the two decades (F1,117 = 16.949, P < 0.0001). To test the hypothesis that variation in beak size independent of body size variation differed between the decades in association with trill rate differences we compared residuals from a regression of beak size on body size (23) in the two periods. The residuals did not differ (F1,117 = 1.7359, P = 0.1903), which is not surprising given the strong correlation between beak and body size. The power of the test is extremely weak because there is little beak size variation left after body size variation has been statistically controlled.
We conclude that changes in G. fortis song did not arise from a character displacement shift in beak size (Fig. 6) caused by G. magnirostris. Changes in the songs of G. fortis and G. scandens reflect a change in behavior.
Peak Shift as a Mechanism of Change
Males.
Changes in song might occur through a “peak shift” or learning-based bias, a common phenomenon well documented in a variety of species from insects to humans (45). During the process of discriminatory learning and production, an individual learns to respond to an exaggerated form of the desired signal in a direction away from a negative stimulus (33–35). The response can be expected to occur in two ways if there is a penalty to song transmission or in interactions with a heterospecific individual. A son, surrounded by singing heterospecifics during his sensitive period for learning or production of song, would be expected to learn and produce elements of conspecific song that were clearly distinct from sympatric heterotypic song. A female would be expected to respond to a male with a song that is an exaggerated form of a conspecific song and clearly distinct from heterospecific songs, resulting in a pattern of biased mating.
Sons learn songs early in life, mainly although not exclusively from their fathers. Consequently there is a strong relationship between the songs of sons and the songs of their fathers (36). The regression relationship (Fig. 7) is an estimate of cultural heritability, i.e., a cultural equivalent to the genetic heritability of quantitative genetics. A prediction of the peak shift hypothesis is that sons initially copy their fathers’ songs and then increase their trill rate in response to G. magnirostris songs. Following early learning trill rates did undergo small changes. For those individuals tested twice trill rate was faster at ages 4–14 y in the 1990s than at 2–6 y in the 1980s: in G. fortis (paired t = 2.79, P = 0.021) and G. scandens (paired t = 3.12, P = 0.0168; one outlier deleted). This trend occurred in all 10 G. fortis individuals (binomial P = 0.002) and in 8 of the 9 G. scandens individuals (binomial P = 0.020) that were recorded twice in different years. Because average song duration and interval between notes did not change (Methods), trill rate alone among the intercorrelated temporal variables increased. Thus sons (generation 1) may begin by singing a faithful copy of the song of their fathers (generation 0) but then later increase the trill rate, so that some of them will produce their own sons (generation 2) with a faster song than their father. In principle this process of song modification could continue repeatedly for several generations, resulting at each step in an increase in distance from the songs of G. magnirostris.
Fig. 7.
We tested the expectation that sons should sing faster songs on average than their fathers in successive generations by comparing the trill rate of 72 G. fortis fathers with their sons, having first averaged the trill rates of sons (one to six) of each father. In agreement with expectation, sons sang faster songs than their fathers (Wilcoxon matched pairs, T = −506, P = 0.0031). Sons sang faster songs in each of the 3 y of largest samples, 1983 (n = 14), 1987 (n = 13), and 1991 (n = 31): all 0.01 < P < 0.05. G. scandens sons also sang faster songs than their fathers in the 3 y combined (T = −131, P = 0.0377, n = 36) and in 1983 alone (T = −12, P = 0.0469, n = 6), although not to a significant extent in 1987 (T = −18, P = 0.1230, n = 11) or 1991 (T = −10, P = 0.5995, n = 13). In all cases the difference from G. magnirostris songs increased (Fig. 5) as a result of G. fortis and G. scandens sons singing faster songs than their fathers (Fig. 7).
A second test of the peak shift hypothesis is possible by regression analysis. A regression of song characteristics of sons [generation (gen) 2] on fathers (gen 1) should have a higher intercept than a regression of fathers (gen 1) songs on the songs of their own fathers (gen 0). Our samples of the songs of grandsons of recorded grandfathers are too small, <10 for both G. fortis and G. scandens, to perform a sufficiently sensitive test.
Females.
Females may amplify the peak shift by preferentially mating with males that sing an exaggerated form of a conspecific song that is clearly distinct from heterospecific (G. magnirostris) songs. A pattern of biased mating might give rise to relatively high reproductive success as a consequence. Although these ideas are plausible they are not supported by data. Seventy-one G. fortis males that hatched in 1991 and acquired mates did not differ in trill rate from 29 unsuccessful others from the same cohort that were tape recorded in the same year (1993) (F1,98 = 1.990, P = 0.1609). Lifetime reproductive success, measured as production of recruits to the breeding population after correcting for extrapair paternity, is known for members of the 1987 cohort. Nineteen G. fortis that produced recruits did not differ in trill rate from 16 that failed to produce a single recruit (F1,33 = 0.346, P = 0.5605). G. scandens show a similar pattern. Fifteen males of the 1991 cohort that bred did not differ from 10 that did not (F1,23 = 0.481, P = 0.4948), and six members of the 1987 cohort that produced one or more recruits did not differ from another 13 that failed to produce one (F1,17 = 2.770, P = 0.1141). All of the observed differences were in the expected direction away from G. magnirostris song; nevertheless there is no statistical evidence of a biased response of females to exaggerated male signals in either population.
Song and the Addition of New Species to the Community
G. magnirostris became incorporated into the finch community on Daphne as a result of a feeding advantage over G. fortis in the large-seed niche, social dominance, and a distinctive song. G. fortis and G. scandens, not G. magnirostris, made the adjustments in song. Another addition to the community occurred at about the same time that G. magnirostris established a breeding population, and the outcome was different.
A new lineage was established when a young immigrant male arrived on the island in 1981 and bred in 1983 (46, 47). It was identified genetically as a fortis–scandens backcross to G. fortis, originating from the nearby large island of Santa Cruz. In 1987 it bred with a resident female of similar hybrid constitution. For the next three generations members of this lineage bred with each other and with G. fortis. Only two individuals, a brother and a sister, survived the drought of 2003–2004, and they bred with each other when the drought ended in 2005. Their descendants bred with each other for the next three generations and not with G. fortis or G. scandens. The lineage is functioning as a new species, being distinct in song, in morphology, and genetically, and breeding endogamously. It has increased in numbers to nine breeding pairs in 2010.
The lineage has fitted into the community by being unique in song, as well as ecologically and morphologically distinct. Songs are short like some G. fortis songs, and they have an initial down-sweep like them, but they also resemble the songs of G. scandens in very short intervals between successive notes (Fig. 2). Numbers have been low, so far, and there have been no morphological, ecological, or song adjustments by the lineage to the other three species. They are larger in body and beak size than G. fortis and most G. scandens. However, they are not consistently socially dominant in interactions as G. magnirostris are, but ignore G. fortis and G. scandens and are ignored by them. The difference underscores the importance of social interactions in the behavioral adjustment of G. fortis and G. scandens to G. magnirostris.
Discussion
Sympatric assemblages of closely related species are notable for their diversity in ecological niches and sexual signaling systems (48, 49). From such observations, coupled with the generally smaller differences between allopatric populations of the same species (16, 19, 29; but see ref. 50), evolutionary biologists attempt to identify the processes that gave rise to the observed patterns. It is possible, although rare, to be able to observe processes directly through the study of organisms in natural environments. We have been fortunate to accomplish this in a long-term study of Darwin's finch populations, by witnessing the establishment of a third species on an island occupied by two others. The study has revealed an important role for behavior in the adjustment of species to sympatric living.
Songs of the medium ground finch (G. fortis) and cactus finch (G. scandens) changed following the arrival and increase in numbers of a third and socially dominant species, the large ground finch (G. magnirostris). The rate of note repetition, or trill rate, of the first two species became faster over the next three decades (Fig. 4). As a result, the songs of both species diverged from the songs of the large ground finch (Fig. 5). It is possible that the changes were the result of random cultural drift and that G. magnirostris was irrelevant to the song changes. We consider this to be unlikely given the coincidence of changes in G. fortis and G. scandens. Their songs differ but they changed in the same way.
We investigated two possible mechanisms of song divergence. The first one is a correlated effect of ecological character displacement (44). During a severe drought G. magnirostris outcompeted the larger members of the G. fortis population through superior ability at exploiting the supply of large and hard seeds. The G. fortis population underwent a selective shift toward smaller average beak size: divergent character displacement. In principle this might have led to an increased trill rate simply as a passive consequence of decreased beak size. Several facts argue against this possibility. For example, an expected inverse correlation between frequency sweep and beak size was not observed. The second explanation—behavioral modification of the songs through learning—was supported. Sons differed more from G. magnirostris than did their fathers, in both G. fortis and G. scandens (Fig. 7). These facts are consistent with a peak shift mechanism. According to this mechanism a shift in a song signal away from a negative heterospecific song signal occurs at the time of song learning and production. Songs are malleable to some degree. Elsewhere it has been shown that swamp sparrows (Melospiza georgiana) increase their trill rate in simulated encounters with an aggressive species (51).
The necessary caveat is that our study is purely observational, without experimental control of potentially confounding variables, and hence our identification of causes should be considered as an hypothesis rather than a demonstration. Members of the hybrid lineage could have contributed to the change, but only to a minor degree because the number of singing males before 2010 never exceeded four except in one year (1993), and their breeding territories were restricted to one small area of the island (47), so their influence could have been only highly localized. Moreover, unlike G. magnirostris they ignore rather than act aggressively toward G. fortis and G. scandens except when an intruder is close to their nests.
Experiments are needed to isolate a causal factor. In their absence one has to use natural observations of a quasi-experimental nature. In the present case there are two such observations. The first one is the absence of habitat change. Observations were made on changes in song in the presence of change in the finch community but in the absence of habitat change. In a loose sense the effect of the variable “habitat” on finch song was controlled (held constant). The second one involved fortuitous observations of G. fortis on the nearby island of Santa Cruz. Songs recorded in 1961 by Robert Bowman were compared with those made at the same locality in 1999 by Jeffrey Podos and found to be remarkably similar in note structure (52). Of particular relevance to our study, the mean trill rate, number of notes, and song duration all remained the same and there was no shift over time in acoustic space. These findings can be considered loosely as a control for the Daphne experiment because the finch community remained unaltered on Santa Cruz at the same time that it changed on Daphne.
In conclusion, species remain distinct even though components of the barrier to interbreeding are flexible because the behavioral imprinting mechanism on which the barrier is based is robust. Behavioral modification of mate-signal learning, through a peak shift mechanism, without genetic change, may be widespread as an ingredient in the response of organisms to the challenge of survival and selecting a mate in a species-rich environment.
Implications for Speciation
The changes in song features we have documented over several decades are minor in comparison with the differences among populations and species that arose in the course of the radiation of Darwin's finches (10) in the last 2–3 My (3). Nevertheless they provide some insight into long-term diversification by supporting the idea that behavioral interactions among previously separated populations cause songs to diverge. In this way they throw light on what happens at a crucial stage in speciation when formerly allopatric, partly differentiated, populations establish contact.
Much current research effort is devoted to understanding the genetic basis of reproductive isolation, a key component of speciation (1). How rapidly do genetic incompatibilities arise and what are the consequential effects on the evolution of mate preferences? If mutations responsible for genetic incompatibilities gradually accumulate when populations are spatially separated, hybrids formed during secondary contact will have relatively low fitness, in which case differences in mating signals and responses between the populations may be reinforced by selection, resulting in diminished interbreeding and gene exchange (12, 13, 53). Typically this stage is reached in birds after many millions of years (54), much longer than the entire period of the Darwin's finch radiation. Not surprisingly, genetic incompatibilities have not been detected in Darwin's finches, nor reinforcement found (3).
The premating barrier to interbreeding among the finches is an interesting composite of genetic and cultural factors learned through imprinting (1). When introgression occurs, the species exchange one component (genes) but not the other (song). In this respect they differ from some other species where hybrids sing elements of the songs of both of the parental species (43, 55). Thus in contrast to many other species of birds the defining feature of a population of Darwin's finches is a cultural learned signal.
Sexual imprinting may enhance the potential for speciation, in two ways. First, sibs with similar preferences that are learned from an atypical father may mate with each other and initiate a new lineage, as shown experimentally in viduine finches (56) and seen in the initiation of a new lineage of finches on Daphne (47). Second, paradoxically, sexual imprinting may enhance speciation when the barrier leaks through copying of heterospecific song. Hybridization with minimal fitness loss, followed by backcrossing, can lead to an increase in genetic variation. In a new environment genetically unusual individuals from such a population may begin a lineage that becomes reproductively isolated from the parental populations. The fortiscandens lineage on Daphne, a new incipient species, provides an example of reproductive isolation and shows that speciation involving hybridization can be very fast when driven by learned behavior (47).
Methods
A Sony TCM 5000 analog machine with a Sennheiser directional microphone was used to record songs of 625 birds in the years 1984–2010. Quality of songs varied owing to wind, noise of seabirds, etc. We chose for analysis the clearest recording with minimal background noise for each bird. A previous analysis showed very little variation among recordings of the same bird on different days (36). Recordings were made in 1984, 1986, 1987, 1990–1993, 2007, and 2010. In addition some G. magnirostris were recorded in 2000 and 2005. For the purpose of comparing song characteristics across years we grouped the data into four decade groups, 1970s, 1980s, 1990s, and 2000s, according to the year in which a bird hatched and not when it was recorded.
All measurements were made in Raven (57). The following spectral and temporal song features were measured: duration of the whole song as well as the first two notes (seconds), the interval between the first and second notes (internote interval, seconds), the maximum and minimum frequency (kHz) for calculating frequency bandwidth as the difference between them, and central frequency (the frequency that divides the song into two frequency regions of equal energy; kHz). The number of notes in each song was counted, and trill rate (seconds) was calculated as the number of notes divided by duration of the song. A note is defined as a continuous sound, appearing discretely on a spectrogram.
A possible stretching of stored tapes could introduce heterogeneous measurement error across years. To examine this possibility we took advantage of measurements made from the spectrograms in the 1980s and 1990s. We had previously used a Kay Elements DSP model 5500 Sona-Graph for producing spectrograms (36) and measured temporal and frequency features of songs by hand. We first had to calibrate the Kay and Raven systems. We obtained help from Dean Hawthorne, Tim Krein, Dounan Hu, and Russ Charif at the Cornell University (Ithaca, NY) Laboratory of Ornithology to match Kay and Raven machine parameters. To match the Kay setting of DC-8 kHz with a sampling rate of 20,480 Hz the Raven software was adjusted to record at 22,050 Hz, using a device setting of 44,100 Hz and a decimation factor of 2. The window was set to 6.25 ms or 138 samples, producing a 3-dB filter bandwidth of 208 Hz. A Hamming window was used with 50% overlap. We then compared song measurements of 88 G. fortis individuals made from Kay Sona-Graph spectrograms and from Raven software. They were almost identical. Mean duration of songs measured from the Kay spectrograms (mean ± SD, 0.5959 ± 0.082 s) is almost identical to the mean measured in Raven (0.5966 ± 0.081 s).
We performed a second test by comparing the songs of 10 G. fortis males and 9 G. scandens males recorded twice, 4–9 y apart. The first recordings were made in 1984–1987 and the second ones were made in 1990–1993. All of the birds except for one G. fortis individual were first recorded at a minimum age of 2 y, having had experience of at least one breeding season. There were no differences in song lengths between recordings, in either G. fortis (paired t = 1.16, P = 0.2725) or G. scandens (paired t = 1.42, P = 0.1926). There were also no differences in internote duration in either G. fortis (paired t = 0.15, P = 0.8830) or G. scandens (paired t = 0.63, P = 0.5447). However, in both species trill rate increased from the first to the second recordings, in G. fortis (paired t = 2.79, P = 0.021) and G. scandens (paired t = 3.12, P = 0.0168; one outlier deleted). This trend occurred in all 10 G. fortis individuals (binomial P = 0.002) and in 8 of the 9 G. scandens individuals (binomial P = 0.020). These results show there was no stretching of tapes in the intervening years, but a change in trill rate with age.
Two hundred nineteen G. fortis and 114 G. scandens were both tape recorded and measured morphologically when captured in mist nets (3). The measurement comprises body mass (weight, grams), wing length, tarsus length, beak length, beak depth, and beak width (all in millimeters). At the time finches were captured or when they were banded as nestlings, a single drop of blood was taken from the brachial vein and stored on EDTA-soaked filter paper in Drierite for later genetic (microsatellite) analysis (46). Hybrid backcrosses (n = 9) from G. fortis to G. scandens were identified from pedigrees and confirmed genetically with assignment tests (58) applied to variation at 14 autosomal microsatellite loci. Their song characteristics were tested against the characteristics of contemporary G. scandens. None of the differences were significant by t tests (all P > 0.20); therefore the backcrosses were included in the G. scandens samples. G. magnirostris has never hybridized with either species on Daphne.
Principal components analyses were performed on song, body size, and beak measurements separately. Song principal component scores were generated in analyses of six song temporal parameters and, separately, four frequency parameters. The percentage of variance explained by the first two components varied from 80 to 90% (for further details and the eigenvector loadings see SI Methods and Tables S3 and S4). Body size principal component scores were generated from weight, wing length, and tarsus length, and beak size principal component scores were generated from beak length, beak depth, and beak width. The percentage of variance explained by the first two components was 94% in the body analysis and 97% in the beak analysis. In the body size analysis loadings on PC 1 (PC 2) were 0.59 (−0.17) for weight, 0.58 (−0.6) for wing length, and 0.57 (0.78) for tarsus length. In the beak analysis loadings were 0.41 (0.9) for length, 0.64 (−0.31) for depth and 0.65 (−0.27) for width. Two G. scandens (numbers 19527 and 19518) were statistical outliers, up to 5 SDs from means, and were removed from analyses.
Statistical analyses were performed in JMP 7.0 (59). For ANOVA we checked for equal variances with Levene's tests and used a Welch ANOVA when variances were unequal. In multiple-group analyses we used a Kruskal–Wallis ANOVA when skewness was strong and Tukey–Kramer post hoc HSD tests at α = 0.05 or 0.01 to identify which groups differed significantly. All statistical tests were two-tailed.
Acknowledgments
We thank the many field assistants who have helped us on Daphne and the Galápagos National Parks Service and The Charles Darwin Foundation for logistical support. We thank Carel ten Cate, Steven Nowicki, and Nathalie Seddon for valuable and constructive comments and helpful review of the manuscript. We thank K. Thalia Grant for the photographs in Fig. 1. The National Science Foundation and Princeton University's Class of 1877 endowment provided funding.
Supporting Information
Supporting Information (PDF)
Supporting Information
- Download
- 174.59 KB
References
1
JA Coyne, HA Orr Speciation (Sinauer, Sunderland, MA, 2004).
2
LH Rieseberg, et al., Major ecological transitions in wild sunflowers facilitated by hybridization. Science 301, 1211–1216 (2003).
3
PR Grant, BR Grant How and Why Species Multiply (Princeton Univ Press, Princeton, NJ, 2008a).
4
PR Grant, BR Grant, Hybridization of bird species. Science 256, 193–197 (1992).
5
LM Ratcliffe, PR Grant, Species recognition in Darwin's finches (Geospiza, Gould). III. Male responses to playback of different song types, dialects, and heterospecific songs. Anim Behav 33, 290–307 (1985).
6
BR Grant, PR Grant Evolutionary Dynamics of a Natural Population (Univ of Chicago Press, Chicago, 1989).
7
LM Ratcliffe, PR Grant, Species recognition in Darwin's finches (Geospiza, Gould). I. Discrimination by morphological cues. Anim Behav 31, 1139–1153 (1983a).
8
J Podos, Acoustic discrimination of sympatric morphs in Darwin's finches: A behavioral mechanism for assortative mating? Philos Trans R Soc B 365, 1031–1039 (2010).
9
BR Grant, PR Grant, Simulating secondary contact in allopatric speciation: An empirical test of premating isolation. Biol J Linn Soc Lond 76, 545–556 (2002b).
10
RI Bowman Patterns of Evolution in Galápagos Organisms, eds RI Bowman, M Berson, AE Levinton (AAAS Pacific Division, San Francisco), pp. 237–537 (1983).
11
BR Grant, PR Grant Endless Forms: Species and Speciation, eds DJ Howard, SH Berlocher (Oxford Univ Press, New York), pp. 404–422 (1998).
12
G-P Sætre, et al., A sexually selected character displacement in flycatchers reinforces premating isolation. Nature 387, 589–592 (1997).
13
T Price Speciation in Birds (Ben Roberts, Greenwood, CO, 2008).
14
H Slabbekoorn, TB Smith, Bird song, ecology and speciation. Philos Trans R Soc B 357, 493–503 (2002).
15
J Podos, SK Huber, B Taft, Bird song: The interface of evolution and mechanism. Annu Rev Ecol Evol Syst 35, 55–87 (2004).
16
N Seddon, Ecological adaptation and species recognition drives vocal evolution in neotropical suboscine birds. Evolution 59, 200–215 (2005).
17
RB Payne, LL Payne, SM Doehlert, Biological and cultural success of song memes in Indigo buntings. Ecology 69, 104–117 (1988).
18
RB Payne Ecology and Evolution of Acoustic Communication in Birds, eds DE Kroodsma, EH Miller (Cornell Univ Press, Ithaca, NY), pp. 198–220 (1996).
19
ANG Kirschel, DT Blumstein, TB Smith, Character displacement of song and morphology in African tinkerbirds. Proc Natl Acad Sci USA 106, 8256–8261 (2009).
20
AV Badyaev, ES Leaf, Habitat associations of song characteristics in Phylloscopus and Hippolais warblers. Auk 114, 40–46 (1997).
21
J Podos, Correlated evolution of morphology and vocal signal structure in Darwin's finches. Nature 409, 185–188 (2001).
22
J Podos, JA Southall, MR Rossi-Santos, Vocal mechanics in Darwin's finches: Correlation of beak gape and song frequency. J Exp Biol 207, 607–619 (2004).
23
SK Huber, J Podos, Beak morphology and song features covary in a population of Darwin's finches (Geospiza fortis). Biol J Linn Soc Lond 88, 489–498 (2006).
24
AV Badyaev, RL Young, KP Oh, C Addison, Evolution on a local scale: Developmental, functional, and genetic bases of divergence in bill form and associated changes in song structure between adjacent habitats. Evolution 62, 1951–1964 (2008).
25
RH Wiley, Associations of song properties with habitats for territorial oscine birds in eastern North America. Am Nat 138, 973–993 (1991).
26
JA Tobias, et al., Song divergence by sensory drive in Amazonian birds. Evolution 64, in press, 2820–2839 (2010).
27
D Luther, L Baptista, Urban noise and the cultural evolution of bird songs. Proc R Soc Lond B 277, 469–473 (2010).
28
EP Derryberry, Ecology shapes birdsong evolution: Variation in morphology and habitat explains variation in white-crowned sparrow song. Am Nat 174, 24–33 (2009).
29
N Seddon, JA Tobias, Song divergence at the edge of Amazonia: An empirical test of the peripatric speciation model. Biol J Linn Soc Lond 90, 173–188 (2007).
30
DA Nelson, P Marler Comparative Perceptions. Complex Signals, eds WC Stebbins, MA Berkeley (John Wiley & Sons, New York) 2, 443–478 (1990).
31
D Luther, The influence of the acoustic community on songs of birds in a neotropical rain forest. Behav Ecol 20, 864–871 (2009).
32
DA Luther, RH Wiley, Production and perception of communicatory signals in a noisy environment. Biol Lett 5, 183–187 (2009).
33
C ten Cate, MN Verzijden, E Etman, Sexual imprinting can induce sexual preferences for exaggerated parental traits. Curr Biol 16, 1128–1132 (2006).
34
C ten Cate, C Rowe, Biases in signal evolution: Learning makes a difference. Trends Ecol Evol 22, 380–387 (2007).
35
MN Verzijden, E Etman, C van Heijningen, M van der Linden, C ten Cate, Song discrimination learning in zebra finches induces highly divergent responses to novel songs. Proc R Soc Lond B 274, 295–301 (2007).
36
BR Grant, PR Grant, Cultural inheritance of song and its role in the evolution of Darwin's finches. Evolution 50, 2471–2487 (1996).
37
PR Grant, BR Grant, Hybridization, sexual imprinting and mate choice. Am Nat 149, 1–28 (1997a).
38
DE Irwin, TD Price, Sexual imprinting, learning and speciation. Heredity 82, 347–354 (1999).
39
C ten Cate, DR Vos, Sexual imprinting and evolutionary processes in birds: A reassessment. Adv Study Behav 28, 1–31 (1999).
40
LL Cavalli-Sforza, MW Feldman, KH Chen, SM Dornbusch, Theory and observation in cultural transmission. Science 218, 19–27 (1982).
41
PR Grant, BR Grant, Mating patterns of Darwin's finch hybrids determined by song and morphology. Biol J Linn Soc Lond 60, 317–343 (1997b).
42
K Riebel, The ‘mute’ sex revisited: Vocal production and perception learning in female songbirds. Adv Study Behav 33, 49–86 (2003).
43
H-W Helb, F Dowsett-Lemaire, H-U Bergmann, K Conrads, Mixed singing in European songbirds: A review. Z Tierpsychol 69, 27–41 (1985).
44
PR Grant, BR Grant, Evolution of character displacement in Darwin's finches. Science 313, 224–226 (2006).
45
SK Lynn, J Cnaani, DR Papaj, Peak shift discrimination learning as a mechanism of signal evolution. Evolution 59, 1300–1305 (2005).
46
PR Grant, BR Grant, Pedigrees, assortative mating and speciation in Darwin's finches. Proc R Soc Lond B 275, 661–668 (2008b).
47
PR Grant, BR Grant, The secondary contact phase of allopatric speciation in Darwin's finches. Proc Natl Acad Sci USA 106, 20141–20148 (2009).
48
RB Stelkens, KA Young, O Seehausen, The accumulation of reproductive incompatibilities in African cichlid fish. Evolution 64, 617–633 (2010).
49
J Losos Lizards in an Evolutionary Tree (Univ of California Press, Berkeley, 2010).
50
N Seddon, JA Tobias, Character displacement from the receiver's perspective: Species and mate recognition despite convergent signals in suboscine birds. Proc R Soc Lond B 277, 2475–2483 (2010).
51
AL DuBois, S Nowicki, WA Searcy, Swamp sparrows modulate vocal performance in an aggressive context. Biol Lett 5, 163–165 (2009).
52
E Goodale, J Podos, Persistence of song types in Darwin's finches, Geospiza fortis, over four decades. Biol Lett 6, 589–592 (2010).
53
T Dobzhansky Genetics and the Origin of Species (Columbia Univ Press, New York, 1937).
54
BM Fitzpatrick, Rates of evolution of hybrid inviability in birds and mammals. Evolution 58, 1865–1870 (2004).
55
J Haavie, et al., Flycatcher song in allopatry and sympatry—convergence, divergence and reinforcement. J Evol Biol 17, 227–237 (2004).
56
RB Payne, LL Payne, JL Woods, MD Sorenson, Imprinting and the origin of parasite-host species associations in brood-parasitic indigobirds, Vidua chalybeata. Anim Behav 59, 69–81 (2000).
57
RA Charif, CW Clark, KM Fistrup Raven 1.3 User's Manual (Cornell Laboratory of Ornithology, Ithaca, NY, 2006).
58
JK Pritchard, X Wen, D Falush Documentation for Structure Software. Version 2.2, Available at http://pritch.bsd.uchicago.edu/software.html. Accessed May 23, 2008. (2007).
59
; SAS Institute JMP 7.0 (SAS Institute, Cary, NC, 2007).
Information & Authors
Information
Published in
Classifications
Submission history
Published online: November 3, 2010
Published in issue: November 23, 2010
Keywords
Acknowledgments
We thank the many field assistants who have helped us on Daphne and the Galápagos National Parks Service and The Charles Darwin Foundation for logistical support. We thank Carel ten Cate, Steven Nowicki, and Nathalie Seddon for valuable and constructive comments and helpful review of the manuscript. We thank K. Thalia Grant for the photographs in Fig. 1. The National Science Foundation and Princeton University's Class of 1877 endowment provided funding.
Notes
This contribution is part of the special series of Inaugural Articles by members of the National Academy of Sciences elected in 2008.
Authors
Competing Interests
The authors declare no conflict of interest.
Metrics & Citations
Metrics
Citation statements
Altmetrics
Citations
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.
Cited by
Loading...
View Options
View options
PDF format
Download this article as a PDF file
DOWNLOAD PDFGet Access
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Personal login Institutional LoginRecommend to a librarian
Recommend PNAS to a LibrarianPurchase options
Purchase this article to get full access to it.