Role of sexual imprinting in assortative mating and premating isolation in Darwin’s finches

Significance

It is important to preserve the potential for future speciation because global biodiversity is being rapidly degraded. A key question is how incipient species become reproductively isolated from each other. Here we provide evidence that two species of Darwin’s finches choose mates on the basis of learning morphological features of their parents and possibly from inherited preferences. The evidence for imprinting is stronger in sons than in daughters and for imprinting by both sons and daughters is stronger on fathers, which sing, than on mothers, which do not. Imprinting establishes a barrier to interbreeding between morphologically different species. The barrier is leaky, species occasionally hybridize, and the hybrids can give rise to a new species based on learned mate preferences.

Abstract

Global biodiversity is being degraded at an unprecedented rate, so it is important to preserve the potential for future speciation. Providing for the future requires understanding speciation as a contemporary ecological process. Phylogenetically young adaptive radiations are a good choice for detailed study because diversification is ongoing. A key question is how incipient species become reproductively isolated from each other. Barriers to gene exchange have been investigated experimentally in the laboratory and in the field, but little information exists from the quantitative study of mating patterns in nature. Although the degree to which genetic variation underlying mate-preference learning is unknown, we provide evidence that two species of Darwin’s finches imprint on morphological cues of their parents and mate assortatively. Statistical evidence of presumed imprinting is stronger for sons than for daughters and is stronger for imprinting on fathers than on mothers. In combination, morphology and species-specific song learned from the father constitute a barrier to interbreeding. The barrier becomes stronger the more the species diverge morphologically and ecologically. It occasionally breaks down, and the species hybridize. Hybridization is most likely to happen when species are similar to each other in adaptive morphological traits, e.g., body size and beak size and shape. Hybridization can lead to the formation of a new species reproductively isolated from the parental species as a result of sexual imprinting. Conservation of sufficiently diverse natural habitat is needed to sustain a large sample of extant biota and preserve the potential for future speciation.

The enormously rich biodiversity of the planet poses two questions to biologists: How to explain it and how to maintain it. Understanding the origin of biodiversity involves the evolutionary search for mechanisms of speciation. The maintenance of biodiversity in a world that is experiencing environmental degradation on a global scale (1) appears to be an entirely separate endeavor. In the short term, this is largely true, except that the future for some species may hinge upon their evolutionary ability to respond to environmental stress, stress that is both continuous and increasing as well as episodic and recurring (2). Termed “evolutionary rescue from the threat of extinction” (3), the responsiveness of species depends largely upon the amount of standing genetic variation together with mutation (4–7). In the medium to long term, we should be thinking about how to maintain communities of organisms in a fit state for future speciation: not just the prevention of extinction but the generation of new species. Unless environmental provision is made to allow for future speciation, the current reduction in the number of species will be permanent. Therefore, whatever can be learned about how species arise is useful information for addressing the neglected problem of preserving the potential for future speciation.

Speciation involves the formation of a barrier to the exchange of genes between two divergent lineages derived from one (8). The barrier may be intrinsic or extrinsic and may function prezygotically or postzygotically (9). In the early stages of divergence the barrier is likely to be solely prezygotic and may be permeable, as shown by the numerous examples of introgressive hybridization between recently formed species (reviewed in refs. 10–13). Important questions that need to be addressed are how the barrier is constructed, how it evolved, why it occasionally leaks, and how gene exchange affects the evolution of the respective populations (14). Paradoxically, the breakdown of reproductive isolation may be especially informative in answering questions about how it was established by exposing candidate traits that constitute the normally effective barrier to interbreeding (15–17). Further insights can then be gained by conducting experiments on the genetic and experiential basis of mate choice and preferences of closely related species (16, 18–22).

Here we describe the results of a long-term field investigation into mating patterns of Darwin’s finches in the Galápagos archipelago, discuss their significance in the context of speciation, and comment on their relevance to the maintenance of biodiversity. The finches are a young adaptive radiation comprising 18 ecologically diverse species (23). According to mitochondrial dating, they diverged from a common ancestor 2.3 (24) or 2.5 (25) Mya and diversified in approximately the last 1 My (23). Previous work has illuminated the basis of reproductive isolation of closely related species of the group (26–30), but unanswered are questions of how mates are chosen from members of the same population and whether the rules of intraspecific mate choice apply to the avoidance of heterospecific mating (31). This is the subject of the present report.

The framework for our investigation is the hypothesis of sexual imprinting (reviewed in refs. 32 and 33). According to this hypothesis, offspring learn morphological features of their parents in early life—they imprint on them—and the mental images are used by the fully grown offspring as cues when choosing a mate (34, 35). An alternative view to learning-based mate preferences is that preferences are genetically inherited. Distinguishing between learned and inherited predispositions will require experimental studies in the future (Discussion). We have chosen imprinting as a framework because of the abundant literature on early learning and mate choice consequences (32–36) in contrast to a lack of evidence for genetic variation in mate preferences for metric traits in passerine birds with biparental care.

The imprinting hypothesis predicts a similarity in the morphology of a chooser’s mate and the chooser’s parents. To the extent that morphological cues are heritable (37), a second prediction is that mates should resemble each other (assortative mating). This follows from the fact that the imprinted morphology of mother and father is both a genetic predictor of the offspring and a cultural predictor of the offspring’s mate. These predictions apply to members of the same species. However, closely related species may be similar morphologically and therefore, as in other organisms, prone to hybridize (13, 38, 39). A third prediction of the imprinting hypothesis is that the individuals that hybridize should be more similar to the other species than are nonhybridizing individuals, and a fourth prediction is that their parents should also be more similar to the other species than are nonhybridizing individuals.

We have taken advantage of small island populations whose pedigrees can be documented to answer questions of mate choice and social relationships. A preliminary study of Geospiza fortis (medium ground finch) on the small (0.34 km²) and undisturbed Galápagos island of Daphne Major gave mixed support to the imprinting hypothesis. A weak tendency was found for mates to be similar in body size and for extreme individuals to mate with similar extreme individuals (40). However, the study was restricted to one species and one trait (body size), and hybrids were ignored. Moreover, the analysis was confounded by inclusion of the hybrid Big Bird lineage that was isolated reproductively from G. fortis after two generations (41, 42).

Here we test for nonrandom mating and the influence of parental morphology on the mate choices of sons, daughters, and hybrids, with data from the period 1976–1998 when comprehensive pedigree information was available. Our analyses were repeated with a second species, the cactus finch (Geospiza scandens). Beak traits were included because these were identified as salient cues in the species discrimination experiments (26, 27): The ground finch species (Geospiza) do not differ in plumage or courtship behavior. Males are territorial, sing, and build nests, and pairs are formed after females have visited the territories of several males. Both males and females feed the nestlings and fledglings (30).

Sons learn their father’s song and mate according to species-specific song. Learning another species’ song early in life can lead to hybridization (30). To test the third and fourth predictions concerning heterospecific mating, we analyzed mating patterns of G. fortis that hybridized. Male and female G. fortis rarely hybridize with G. scandens, a common resident species, and with Geospiza fuliginosa (small ground finch), an occasional immigrant species (43). Additionally, six G. fortis females bred sequentially with an immigrant large cactus finch (Geospiza conirostris) from Española Island, and one of the pairings gave rise to the Big Bird lineage (42, 43). Mating patterns of these females were analyzed, but the Big Bird lineage itself was excluded. In all these cases of conspecific and heterospecific mating, members of a pair are both social and reproductive mates. Despite social monogamy, a low incidence of extrapair mating occurs (44). Pairs affected by extrapair paternity have been excluded from the analyses. Nevertheless, they have a value in providing a natural cross-fostering experiment (SI Appendix, Section 1).

Results

The Influence of Imprinting on Mothers and Fathers in Mate Choice.

The first prediction of the imprinting hypothesis—that a chosen mate resembles the chooser’s parents—is supported in the two species but in different ways (Table 1). For G. fortis mates, the body size of the female is strongly predicted by the body size of the male’s parents, and the body size of the male mate is predicted by the size of the female’s parents. In both cases the father’s size (Fig. 1) is a statistically stronger predictor than the mother’s size.

Table 1.

The imprinting influence of parents on mate choice

Trait	n	F	P	Adjusted R2	Father t test	P	Mother t test	P
G. fortis: male choice of female
PC1 body	417	13.01	<0.0001	0.059	4.48	<0.0001	1.68	0.1052
PC1 beak	439	4.93	0.0076	0.018	3.11	0.0020	−0.02	0.9854
PC2 beak	439	1.08	0.2778	0.005	1.56	0.1186	−1.16	0.2455
Beak length	450	0.94	0.6642	0.008	1.11	0.2665	1.31	0.1942
Beak depth	450	6.99	0.0010	0.026	3.64	0.0003	0.21	0.8308
Beak width	450	3.67	0.0262	0.012	2.70	0.0071	−0.37	0.7109
G. fortis: female choice of male
PC1 body	534	14.82	<0.0001	0.053	4.27	<0.0001	2.23	0.0265
PC1 beak	574	4.18	0.0158	0.011	1.09	0.2766	2.37	0.0180
PC2 beak	574	0.39	0.6794	0.000	0.82	0.4099	0.33	0.7405
Beak length	586	1.82	0.1636	0.003	1.24	0.2140	1.27	0.2046
Beak depth	586	5.70	0.0035	0.019	1.39	0.1639	2.59	0.0098
Beak width	586	3.43	0.0331	0.008	0.73	0.4632	2.37	0.0180
G. scandens: male choice of female
PC1 body	121	2.20	0.1156	0.036	1.01	0.3182	1.49	0.1376
PC1 beak	149	0.57	0.5655	0.000	0.96	0.3402	0.35	0.7239
PC2 beak	149	3.27	0.0407	0.030	−1.49	0.1372	2.39	0.0180
Beak length	150	1.05	0.3517	0.001	0.88	0.3799	0.82	0.4152
Beak depth	150	1.25	0.1900	0.003	1.57	0.1177	0.13	0.8947
Beak width	150	0.13	0.8794	0.000	−0.49	0.6225	−0.01	0.9935
G. scandens: female choice of male
PC1 body	178	3.74	0.0257	0.030	1.82	0.0710	0.85	0.3962
PC1 beak	225	1.09	0.3368	0.001	1.21	0.2267	0.80	0.4248
PC2 beak	225	6.72	0.0015	0.057	2.39	0.0179	1.59	0.1127
Beak length	228	8.57	0.0003	0.063	2.65	0.0087	1.92	0.0556
Beak depth	228	0.70	0.4968	0.000	1.18	0.2409	0.23	0.8155
Beak width	228	1.07	0.3441	0.001	0.97	0.3314	1.03	0.3058

In each group, the chosen mate’s morphology is predicted by the combined parents of the chooser in multiple regression analyses and then by the individual parents by partial regression. Two analyses were performed for each group, one with the three PC scores and one with the three univariate beak variables. Statistical significance is highlighted in italics (P < 0.05) or boldface (P < 0.01). All morphological associations between the mates and parents of choosers are positive except where indicated by a negative t test of the partial regression coefficients.

Fig. 1.

Prediction of a G. fortis mate’s body size by the father of the female or male. (Upper) Body size (PC1) of a male mate predicted by the father of the female (F_1,431 = 24.50, P < 0.0001, adjusted R² = 0.04). (Lower) Body size of a female mate predicted by the father of the male (F_1,431 = 30.90, P < 0.0001, adjusted R² = 0.06).

In contrast to the G. fortis results, the G. scandens male’s choice of a mate is not predicted by the size of either of his parents. However, the beak shape of a female’s parents is a statistically strong predictor of her mate’s beak shape. The prediction is significant for the father’s but not for the mother’s beak shape. A separate analysis of beak variables shows that the beak shape association stems from beak length and not beak depth or beak width, whereas for G. fortis beak depth is the primary predictor.

Nonrandom Mating.

The second prediction of assortative mating on the basis of body size is supported by statistically strong positive correlations between mates in both G. fortis and G. scandens (Fig. 2 and Table 2). Further, there is strong support for assortative mating by beak size [principal component (PC) scores] in G. fortis but not in G. scandens. The difference between the species is expected because beak size is more strongly correlated with body size in G. fortis than in G. scandens. The adjusted R² values for the correlations between the two traits (PC scores) are 0.67 for male and 0.49 for female G. fortis. These are larger than the corresponding values of 0.33 and 0.27, respectively, in G. scandens (P < 0.0001 in each case).

Fig. 2.

Assortative mating by bill length in G. fortis (r = 0.16, n = 787, P < 0.0001) (Upper) and G. scandens (r = 0.20, n = 378, P < 0.0001) (Lower).

Table 2.

Assortative mating by morphological traits

Species	Trait	n pairs	r	P
G. fortis	PC1 body	712	0.198	<0.0001
	PC1 beak	746	0.138	0.0002
	PC2 beak	746	0.053	0.08
	Beak length	764	0.104	0.0023
	Beak depth	763	0.151	<0.0001
	Beak width	763	0.085	0.0104
G. scandens	PC1 body	306	0.189	0.0005
	PC1 beak	350	0.000	0.4044
	PC2 beak	350	0.091	0.0487
	Beak length	357	0.193	0.0002
	Beak depth	357	0.000	0.8547
	Beak width	357	0.026	0.6049

Statistical significance is highlighted in italics (P < 0.05) or boldface (P < 0.01).

In contrast to these positive correlations between mates, there is only statistically weak evidence of assortative mating by beak shape [principal component 2 (PC2) beak] in G. scandens and no evidence of assortative mating by beak shape in G. fortis. Partitioning the composite beak shape measure into its components shows that G. scandens mates resemble each other in beak length, and G. fortis mates resemble each other in all three traits but more strongly in beak depth than in the other two traits (Table 2). Length is the primary trait in which the two species differ (43).

For those individuals that changed mates, the positive associations are expressed in the G. fortis male sample (n = 164) at first mating but not at the second. Associations are weaker or nonexistent at second mating of the G. fortis female sample (n = 168) and the two (smaller) G. scandens samples (SI Appendix, Section 2 and Table S1). However, sample sizes are much smaller than in the analysis of parental influence on mate choice (Table 1). A possible source of bias (SI Appendix, Section 3) in the G. fortis results arising from natural selection in 1977 (43) was checked by deleting the 1976 pairs and analyzing the remainder. This had little effect on the results (SI Appendix, Table S2 compared with Table 1). Exclusion of rare hybrids also had little effect on the results (SI Appendix, Table S2).

Hybridization of G. fortis with G. fuliginosa and with G. scandens.

G. fortis is larger than G. fuliginosa in body size and size-related traits (45).The third, interspecific, prediction of the imprinting hypothesis is those G. fortis individuals that hybridize are unusually similar to their heterospecific mates. The prediction is supported by hybridization of G. fortis and G. fuliginosa (or F₁ and backcrosses). G. fortis females that hybridize with G. fuliginosa are significantly smaller in body size than matched nonhybridizing G. fortis females (F_1,62 = 4.94, P = 0.03). Hybridizing G. fortis males exhibit a nonsignificant trend in the same direction (F_1,66 = 1.77, P = 0.19). G. fortis that pair with G. scandens, a larger species, do not differ from matched controls in body size (or beak traits). This applies to both males (F_1,64 = 0.48, P = 0.49) and females (F_1,45 = 2.56, P = 0.12).

The fourth prediction of the imprinting hypothesis is parents of G. fortis that hybridized with G. fuliginosa are smaller than nonhybridizing individuals of the same sex. There is no evidence in support of the hypothesis. Neither fathers (F_1,94 = 0.63, P = 0.43) nor mothers (F_1,81 = 0.38, P = 0.54) of hybridizing G. fortis differ in size from nonhybridizing G. fortis of the same sex. Furthermore, neither fathers (F_1,88 = 0.01, P = 0.94) nor mothers (F_1,88 = 0.01, P = 0.94) of G. fortis that hybridized with G. scandens differ in size from nonhybridizing G. fortis of the same sex. Although the G. fortis × G. fuliginosa hybrids and the G. fortis × G. scandens are rare, they do show a tendency to mate with each other (SI Appendix, Section 4 and Table S3).

The samples of G. scandens are too small for comparable analyses. Nine males and two females were measured, as were the parents of only five of the males.

Hybridization of G. fortis with G. conirostris.

An immigrant G. conirostris male bred sequentially with six G. fortis females during his 14-y life. All six females had mated with another male before mating with the immigrant, and one had mated with two males. Since second mates within the G. fortis population do not reflect a signature of parental imprinting, there is no clear expectation of nonrandom mate choice. Nonetheless, hybridization is different from intrapopulation mating, and the effects of imprinting in these few cases may have been unusually enduring; for that reason the imprinting hypothesis was tested.

If the six females that mated with the immigrant imprinted on their parents, they should be unusually large, like their G. conirostris mate. This is the third prediction of the hypothesis. In agreement with predictions, the first five mates were unusual in one or more traits (Table 3).

Table 3.

Hybridizing G. fortis females compared with other breeding females at the time they first bred with the immigrant G. conirostris

Female ID	Comparisons	Year	Body size	Beak size	Beak shape	Beak length
4734	114,118	1983	0.44	0.43	−2.19	1.10
5628	90,100	1987	0.89	−0.27	−2.60	1.77
5626	123	1990	1.44	−0.01	−1.47	0.59
5821	111	1991	1.11	2.10	−0.55	2.13
15210	115	1992	1.43	0.55	−1.15	0.90
16594	130	1993	0.70	1.27	0.70	0.95

Deviations of the hybridizing females from each comparison group are shown as SDs, in boldface when in the most extreme 5% in the tail of a distribution. Negative signs indicate the hybridizing females are smaller in size or have a more pointed beak. Where two sample sizes are given, the first refers to body size only.

The fourth prediction is the parents of the six females, especially the fathers, should also be unusually large. The data provide mixed support. The fathers of two of the mates are unusually large in body size, and the father of another of the mates has an exceptionally large beak (Table 4), in agreement with prediction. In three cases beak shapes are statistically significant in the direction opposite the one predicted. However, beak shape is a composite trait, and in one dimension, length, the data support the hypothesis. Beak lengths of all six fathers are above the means of the samples with which they are compared (Fig. 3). Four of them are more similar to the immigrant male than to the sample means (SI Appendix, Section 5), which is consistent with the third prediction. Fathers are also unusual compared with other males breeding in the year the mates hatched (SI Appendix, Table S4).

Table 4.

Unusual fathers of hybridizing G. fortis females

Female ID	Father ID	Comparisons	Year	Body size	Beak size	Beak shape	Beak length
4734	2666	139,151	1983	1.24	0.07	−3.70	1.83
5628	4053	104	1987	1.79	0.73	−2.61	1.76
5626	4053	98	1990	2.37	0.94	−2.77	2.01
5821	1738	125	1991	1.23	2.24	0.03	2.10
15210	14533	130	1992	−0.30	−0.37	−1.49	0.24
16594	14856	160	1993	0.96	0.44	0.21	0.30

Fathers are compared with males breeding in the years the females first bred with an immigrant male G. conirostris. Deviations of the fathers from each comparison group are shown as SDs, in boldface when in the most extreme 5% in the tail of a distribution. Negative signs indicate the fathers are smaller in size or have a more pointed beak. Where two samples are given, the first refers to body size only.

Fig. 3.

Unusual morphology of the fathers of the first four G. fortis females that mated with an immigrant G. conirostris male. The females’ fathers’ beak lengths (*) are compared with the distribution of beak lengths of G. fortis males breeding at the time of hybridization. Boxes enclose 50% of the sample. SDs are indicated by blue horizontal lines, and 95% of the samples are indicated by red horizontal lines. n = 98–160 samples (Table 4).

The mothers of the immigrant’s mates are not unusual (SI Appendix, Table S5), except for the mother (3367) of the first mate (4734) that has an unusually pointed beak, like her mate (the father, 2666) (Table 4). This is consistent with the general pattern of assortative mating in the G. fortis population (Table 2). In fact, both members of this pair have the most extreme beak shape in their respective samples.

Discussion

Mate Choice Within Populations.

We have adopted the framework of sexual imprinting to investigate mating patterns of identified individuals in a natural environment and their relevance to reproductive isolation from closely related species. An alternative view to learning-based mate preferences is that they are genetically inherited. In contrast to the abundant literature on early learning and mate choice consequences (36), there is no evidence of genetic variation in mate preferences for varying metric traits in passerine birds with biparental care (46). It is possible that inherent predispositions exist and interact with learning-based preferences to determine mate choices. Mate preferences are functions of trait distributions (47–49), and preference functions may be governed by learning but limited probabilistically at the extremes by inherent predispositions. Cross-fostering experiments would be needed to investigate these possibilities (SI Appendix, Section 1) (18, 19, 33, 50). Experimental manipulation of finches in nature is not an option for studies in the protected Galápagos National Parks environment. Therefore, we have taken the approach of hypothesizing an imprinting influence of parents’ morphology on the mate choice of their offspring and testing expected consequences with correlation analyses.

Both predictions of the hypothesis were supported, to varying degrees, when applied to mate choice within populations. The first prediction is morphological similarity of the offspring’s mate to his or her own father and mother. The prediction was supported in both G. fortis and G. scandens and involved body size and beak traits. Interestingly, the influence of fathers on mate choice was statistically stronger than the influence of mothers. A caveat is necessary, however. As mentioned above, without experiments we cannot rule out the possibility that some or all of the associations were produced by inherent predispositions or genetically based preferences, as found, for example, in fish (51, 52). Natural cross-fostering that occurs with extrapair paternity offers an opportunity to distinguish between cultural and genetic inheritance (SI Appendix, Section 1). However, our analysis is limited by small sample sizes of extrapair paternity. We assume that if inherent preferences exist they are of minor importance and are masked by the effects of imprinting. This assumption needs to be investigated experimentally with other bird species. It should also be noted that much of the variation in the regression relationships is left unexplained statistically by the independent variables. Mate preference functions based on experience may be broad and shallow (53, 54). Factors other than parental imprinting contribute to the variation, including mate availability, song, environmental suitability, behavioral interactions in courtship, and chance.

The second prediction of assortative mating by size traits was demonstrated in both G. fortis and G. scandens. Assortative mating, defined as “the pattern that results when like mates with like” (31), has been shown to occur in a variety of organisms including flies (55), butterflies (56), fish (52, 57), frogs (58), lizards (59), and birds (60, 61). The pattern of assortative mating by class (e.g., hybrids vs. species) or by traits (e.g., plumage morphs, size) can arise directly as a result of choice by one or both members of a pair or indirectly through a choice of a habitat by individuals that are alike and in which random mating occurs (62–64). These alternatives cannot be distinguished in two other situations in which assortative mating in Darwin’s finches has been studied: two species of tree finches (Camarhynchus spp.) and their hybrids on Floreana Island (65) and small and large forms of G. fortis on Santa Cruz Island (66). On the small island of Daphne Major there is little habitat heterogeneity that would allow similar individuals to segregate into different patches, and no association has been found between natal and breeding microhabitats of G. fortis phenotypes (43). Therefore, the pattern of assortative mating in each of the two species, G. fortis and G. scandens, arises directly from selective mating. Since the species have biparental care, the choice of mates is likely to be mutual, although the cues used by males and females need not be exactly the same. Assortative mating is weaker or nonexistent when individuals pair for the second time (SI Appendix, Table S1). This suggests that although early learning of parental traits influences initial mate choice, the influence wanes with additional experience of the environment, and additional factors such as nest site and territory quality become more important (67, 68).

Our findings on imprinting on metric traits extend the research into sexual imprinting carried out with plumage traits in captive birds by providing different insights into mechanisms and consequences. The most extensive studies have been conducted with sexually dichromatic zebra finches (Taeniopygia guttata) (33). By experimentally manipulating the rearing environment of offspring, Vos (69) showed that males choose mother-like females in plumage color in preference to father-like females. Females, in contrast, did not prefer partners of the opposite-sex parent’s color morph; instead they preferred males of the mother’s plumage type (70), potentially discriminating against other species. Darwin’s finches are also sexually dichromatic, and the degree of blackness in the plumage of males is a factor in mating success on Daphne Major (67). A study of Geospiza propinqua (conirostris) on Genovesa Island also showed that males in fully black plumage had the highest mating success, but black plumage and age (and hence experience) are positively correlated, and age was found to be the more significant factor in mating success (68).

Song is another cue used by birds in choosing a mate. Cross-fostering experiments with zebra finches have demonstrated early learning of song in an imprinting-like process and subsequent mating based on song features (71). Darwin’s finches are similar, in that they learn their father’s single song during a sensitive period that lasts for the 2 wk they are in the nest and the additional 3 wk or more when they are being fed as fledglings (both parents feed the offspring throughout this period). Learning of song has been demonstrated by experiments with playback of tape-recorded song to naive nestlings in captivity (29) and inferred from field observations of parent–offspring associations and multigenerational pedigree analyses (30). Sons sing the song they learn from their fathers, whereas females, although they do not sing, mate according to the species song of the father (72). Female learning of song is known from the few cases in which males of one species have acquired the song of another species through accidental perturbation of the normal song-learning process, for example through death of the father and learning from a heterospecific neighbor (30). Daughters breed with males that sing the species song they learned early in life; hence in such cases they hybridize. Within a species there are small variations among males in song pattern and structure, but daughters do not systematically choose males that sing the same song variant as their father’s (30); if anything, they tend to avoid such males when they pair for the second time. Nor is there evidence from Daphne Major (72, 73) that the choice of a male with a particular beak morphology is an indirect effect of choice of song arising from a correlation between beak size and song features, as has been found with other populations and species (74–76). Song and beak morphological variation of G. fortis do not covary on this island (30).

Mate Choice of Hybridizing Species.

The third, interspecific, prediction of the imprinting hypothesis is those individuals that hybridize are unusually similar to their heterospecific mates. In agreement with the prediction, G. fortis females that bred with G. fuliginosa, F₁ hybrids, and backcrosses are significantly smaller than nonhybridizing G. fortis. They are more similar to G. fuliginosa than they are to some other members of their own population that did not breed with G. fuliginosa. However, other tests did not support the hypothesis. G. fortis males are not significantly smaller than their matched nonhybridizing counterparts. The parents of hybridizing G. fortis individuals are not unusually small, and this finding is inconsistent with sexual imprinting being the cause of choice of another species as a mate (prediction 5). Likewise, G. fortis that hybridized with G. scandens were not more similar to them than to nonhybridizing G. fortis, and their parents did not differ in size from controls.

Hybridization of G. fortis and G. conirostris is more consistent with imprinting. Numbers are small; nevertheless, some of the mating patterns conform to the third and fourth predictions. Six G. fortis females bred sequentially with an immigrant G. conirostris male. In agreement with the third prediction, the first five were unusually large in comparison with other females breeding in the same year, like their G. conirostris mate. In agreement with the fourth prediction, fathers of the first four mates were unusually large in body size and beak length in comparison with other breeding males in (i) the year the mates hatched and (ii) the year the mates bred with the immigrant. The mothers were not unusually large. Collectively, these nonrandom patterns constitute evidence of imprinting on paternal morphology. An additional unusual fact is that the first three females were first-generation backcrosses of F₁ G. fortis × G. scandens hybrids to G. fortis (FFS). The low probability of FFS females mating with the G. conirostris male is shown by the extremely low frequencies of FFS in those years, i.e., 1983, 1987, and 1990. The frequencies were respectively 0.03 (n = 178), 0.06 (n = 119), and 0.03 (n = 62). The combined probability of random mating for all six females is 5.24E-05.

Song and morphology may function synergistically as cues that are used by finches in choosing a mate. The evidence is stronger for imprinting on fathers than on mothers (Table 1), perhaps indicating that two cues of species identity provide a stronger basis for learning than morphology alone. Song was a factor in the interbreeding of G. fortis with G. fuliginosa and with G. scandens: All males except for one G. scandens sang G. fortis songs (30, 43). In contrast, song was not a factor in the hybridization of G. fortis and G. conirostris. The G. conirostris male sang a distinctive trilled song, recognizably different from the songs of all residents on the island (illustrated in ref. 41), whereas the fathers of all six G. fortis females sang a type 1 G. fortis song. Song has also been implicated in the assortative mating of small and large forms of G. fortis on Santa Cruz Island (66, 76).

In summary, hybridization with G. fuliginosa was apparently caused by song similarity, but in G. conirostris it occurred despite song dissimilarity. A common denominator was size. Small G. fortis females hybridized with a smaller species (G. fuliginosa), and large G. fortis females hybridized with a larger species (G. conirostris).

Mate Avoidance Between Populations.

Hybridization is rare (SI Appendix, Table S3) (30, 43). Learning of parental morphology and song early in life binds individuals of similar culture and genetics into breeding collectives (populations), so the questions arise: What are the barriers to interbreeding between different collectives? Do the same rules of intraspecific mate choice apply to the avoidance of heterospecific mating? There is debate about whether species recognition and avoidance of heterospecific mating is a categorically different phenomenon from mate choice within species or whether the two phenomena are the same except that the cues differ in magnitude along an axis of continuous variation (31, 56, 57, 77–79). Our data show the answer depends on the particular cue: Song variation is categorical, whereas morphological variation is continuous. In combination, song and morphological differences between species constitute a barrier to interbreeding.

The components of the barrier are liable to break down in different ways. The barrier is breached if individuals copy and produce the song of another species. The probability of interbreeding is either 0 or close to 1. This differs from the probability of interbreeding through morphological resemblance, which varies more continuously. Morphological discrimination between species is an exaggerated manifestation of discrimination between individuals of the same species. Thus, song and morphology are complementary aspects of the barrier.

As species diverge through time and morphological differences between them increase, the differences become more important as barriers to interbreeding. This has been argued for cichlid fish (16) and is indicated in the present study by two facts. First, G. scandens imprint on beak length, the trait in which it differs most from G. fortis, and G. fortis imprint on beak depth and width, the traits in which it differs most from a smaller (G. fuliginosa) and a larger (Geospiza magnirostris) congener. Second G. magnirostris, the fourth ground finch on Daphne, is much larger than both G. fortis and G. scandens, does not overlap them in body size or beak depth and width, and does not hybridize with them. Even though eight G. fortis males and one G. scandens male learned and produced with high fidelity the unique G. magnirostris song and not their respective species-specific songs, they did not breed with G. magnirostris females (43). Large morphological differences acted as a barrier to interbreeding. Together these results suggest that sexual imprinting narrows the range of possible mates, reducing the chance of mating heterospecifically under normal circumstances to a degree that depends on the magnitude of morphological differences between species. Therefore, sexual imprinting has both intraspecific (mating) and interspecific (avoidance) consequences (30).

Two Routes to Speciation.

It is well recognized that speciation can proceed in different ways (9, 80). Darwin’s finches exemplify two of them. The first is the widely accepted classic model of allopatric speciation that begins with divergence caused by ecologically based natural selection and/or genetic drift in allopatry and culminates in possibly further divergence and reproductive isolation in sympatry (81, 82): This is speciation by fission. Gene exchange may occur throughout the process, either retarding or accelerating speciation (50, 83, 84) and eventually declining to zero. In an archipelago such as Galápagos, with numerous ecologically different islands that have fluctuated in size and isolation through glacial cycles (85), there have been many opportunities for rapid divergence in both allopatry and sympatry (86) and many opportunities for the enhancement of divergence through gene exchange, resulting in adaptive radiation. Although these conditions and opportunities are especially prevalent in archipelagos, they are not restricted to them (87–89).

Reproductive isolation has been hypothesized to arise as a nonselected consequence or by-product of ecological divergence (90). The hypothesis has been applied to Darwin’s finches (26, 82, 91), recognizing that beak and body size differences between species simultaneously create both feeding niche differences and a behavioral barrier to interbreeding, as pointed out by Lack (26) many years ago. Mate preferences that develop from sexual imprinting on parental body size and beak traits and from learning paternal song (30, 72) act conservatively and tend to maintain the reproductive isolation. Further directional change in the signal–response systems can occur by the evolutionary process of sexual selection on traits and preferences (31, 67, 92) in which learning may play a role (36), by hybridization (89, 93), or by the behavioral process of a shift in song characteristics by adult males in response to songs of other species that are learned by offspring (72).

The second, and apparently much rarer, mode is homoploid hybrid speciation (93–97): speciation by fusion. This form of speciation is exemplified by the hybridization of an immigrant G. conirostris and resident G. fortis on Daphne Major (42, 43). Interbreeding initiated a new lineage. From generation 2 onwards members of the lineage bred within their lineage and were reproductively isolated from G. fortis by a premating barrier. In light of the foregoing discussion, endogamous breeding from generation 2 onwards is explained by sexual imprinting on the shared features of a unique song and unique morphology: beaks and body size were larger than those of G. fortis, and, perhaps more significantly, beaks were large relative to body size. The absence of interbreeding, especially from generation 4 onwards, can be attributed to the availability of several potential mates with those shared features: Like bred with like. In contrast, there were no available conspecific mates for the immigrant. He had difficulty in attracting (or choosing) a mate and succeeded only after an exceptional delay of 3 mo in the prolonged 8-mo breeding season of 1983 (an el Niño year). Unique beak sizes and associated diets explain successful coexistence with the potentially competitive sympatric species (43).

In both types of speciation, the standard allopatric and the hybridization model, the key features were environmental heterogeneity on both landscape (archipelago) and local (island) scales, the opportunity for gene exchange, and mate choice based on learning possibly combined with inherited preferences. Environmental conditions that promoted diversification in the past are those that are needed to allow diversification to continue in the future. They provide guidelines on how to maximally foster future speciation. To borrow a famous metaphor from Hutchinson (98), the ecological theater needs to be preserved on such a scale as to permit the evolutionary play to continue in ways that are already known and in other ways yet to be discovered.

Methods

Sampling Design.

Daphne Major is 0.34 km² in area, 0.75 km long, and 120 m high (43). Adult finches were captured in mist nets placed in all habitats throughout the island, banded with a combination of one numbered metal band and three colored leg bands, measured, and released. They were weighed, and the wings, tarsi, and all three dimensions of the beaks were measured (99). At maximum (in 1992), all finches on the island were banded. An attempt was made to find every nest and identify the parents in every year of breeding from 1976 to 1998, inclusive. From 1999–2012 the coverage of breeding was less complete. Chicks were banded in the nest at age 8 d. They were captured in mist nets when fully grown at age 60 d or older and were measured (99). Parentage was determined by observation and microsatellite analysis, and species and hybrids were identified by a combination of measurements, pedigrees, and microsatellite DNA (100). Princeton University Animal Care Committee approved the research procedures.

Statistical Analyses.

Principal components analysis was used to reduce the six measured variables to three traits: body size determined by weight and wing and tarsus length (PC1 body), beak size (PC1 beak), and beak shape (PC2 beak) determined by beak length, depth, and width, as described in detail in refs. 40, 82, and 98. We used all data available for each test but were selective for the analysis of G. fortis × G. fuliginosa and G. fortis × G. scandens mating patterns. To compare hybridizing G. fortis with nonhybridizing G. fortis banded at the same time, we used individuals that were listed in the data file immediately before and after or as close as possible to each hybridizing individual. All tests are two-tailed. Analyses were performed in JMP (SAS Institute).