Populations that are connected by immigrants play an important role in evolutionary and conservation biology, yet we have little direct evidence of how such metapopulations change genetically over evolutionary time. We compared historic (1894–1906) to modern (1988–2006) genetic variation in 11 populations of warbler finches at 14 microsatellite loci. Although several lines of evidence suggest that Darwin's finches may be in decline, we found that the genetic diversity of warbler finches has not generally declined, and broad-scale patterns of variation remained similar over time. Contrary to expectations, inferred population sizes have generally increased over time (6–8%) as have immigration rates (8–16%), which may reflect a recent increase in the frequency and intensity of El Niño events. Individual island populations showed significant declines (18–19%) and also substantial gains (18–20%) in allelic richness over time. Changes in genetic diversity were correlated with changes in immigration rates, but did not correspond to population size or human disturbance. These results reflect the expected stabilizing properties of whole metapopulations over time. However, the dramatic and unpredictable changes observed in individual populations during this short time interval suggests that care should be taken when monitoring individual population fragments with snapshots of genetic variation.

A principal goal of evolutionary genetics is to understand how genetic changes occur over time, yet empirical studies are largely confined to snapshots of genetic variation at a single time point. Studies that directly measure genetic change over evolutionary time are rare, and tend to focus on single populations undergoing recent decline (Bouzat et al. 1998; Nichols et al. 2001). The need to understand genetic change over time is especially important for fragmented populations. Populations connected by immigrants are a main concern for population genetics and speciation theory (Mayr 1942; Wright 1969; Butlin 1987). Although classical studies have viewed gene flow largely as a constraining influence on speciation, it is now clear that speciation often proceeds without complete isolation from gene flow (Nosil 2008). Recent studies have shown that under some circumstances, low levels of gene flow may promote, rather than constrain, local adaptation (Whitlock et al. 2000; Church and Taylor 2002).

Fragmented populations receive an enormous amount of attention in ecology and conservation biology (Fahrig 2003). A key concern of landscape management is the degree to which population fragments are connected by corridors that promote the exchange of immigrants (Harrison and Bruna 1999). Metapopulation theory has traditionally focused on the dynamics of dispersal, and extinction and recolonization of population fragments (Hanski and Gaggiotti 2004). Metapopulation structure can buffer a species from small-scale environmental or ecological fluctuations (Dey and Joshi 2006; Su et al. 2009), and similar effects can occur with genetic variation. Immigrants carry genes that can help a population avoid local extinction if it has undergone genetic erosion. This notion of “genetic rescue” is a mainstay of conservation genetics and captive breeding programs (Brown and Kodric-Brown 1977; Whitlock et al. 2000), but genetic rescue has only rarely been documented in nature (e.g., Keller et al. 2001). A final reason to investigate the complexities of genetic change in fragmented populations is because genetic monitoring is increasing dramatically (Moritz 1994; Schwartz et al. 2007; DeBarba et al. 2010; Luikart et al. 2010; Antao et al. 2011). A modest sample of genetic information can provide more accurate estimates of effective population size and growth trajectory than other, much more labor intensive, methods. However, we know little about the robustness of genetic monitoring in a fragmented landscape (Waples 2010).

Natural history collections offer valuable opportunities to directly study genetic change through time (Austin and Melville 2006; Wandeler et al. 2007; Leonard 2008). Direct cross-temporal genetic comparisons can reveal more complex population histories than single time point estimates (Ramakrishnan and Hadly 2009). For example, in populations with low genetic diversity, it is often difficult to determine if recent population decline or an extended history of small population size has created the observed patterns when only a single time point is available for analysis (Matocq and Villablanca 2001). Genetic data from historically preserved specimens can serve as a reference point for past genetic diversity (Bouzat 2001). The reconstruction of population history using historic specimens is becoming increasingly common, especially in threatened and endangered taxa, including fish (Hansen et al. 2002, 2009; Guinand et al. 2003), mammals (Pertoldi et al. 2001; Miller and Waits 2003), and birds (Nichols et al. 2001; Johnson and Dunn 2006; Taylor et al. 2007). However, most studies of historic collections have been limited to one or a small number of isolated populations.

The adaptive radiation of Darwin's finches offers an ideal natural system in which to directly evaluate the history of metapopulations. Most Darwin's finch species have attributes commonly associated with metapopulations, including regular exchange of immigrants among island populations (Petren et al. 2005), several population extinctions, and one documented founding event (Grant and Grant 1995; Dvorak et al. 2004; Grant et al. 2005). Because even the oldest species still regularly exchange immigrants among populations, it is likely that the adaptive radiation occurred with ongoing gene flow (Petren et al. 2005). During 1895–1905, several natural history expeditions visited the Galápagos Islands and collected large numbers of finches, at a time when human settlers numbered only in the low hundreds. These specimens can provide a historical point of reference, and allow us to measure changes in genetic diversity over time across a well-defined, fragmented landscape. A 100-year time span represents roughly 25-30 generations in Darwin's finches (average generation time of 3–5 years; Grant 1999).

Based on recent studies, substantial demographic, morphological, and genetic changes have occurred in Galápagos finch populations within the last 100 years. El Niño fluctuations produce periods of dramatic expansion and contraction in finch population sizes (Grant et al. 2000). One goal of our study is to determine whether these dramatic fluctuations observed in a few species on a single island are a general characteristic of other species of Darwin's finches and of other islands across the archipelago. Climatic impacts vary according to island characteristics, but the frequency and severity of El Niño episodes appears to be on the rise (Guilderson and Schrag 1998). Changes in natural selection pressures on small islands have been demonstrated to cause significant morphological changes in heritable beak traits in just a few years time (Grant et al. 2004; Grant and Grant 2006). Resident human populations and tourism have increased exponentially over the last century (Watkins and Cruz 2007), along with the introduction of nonnative species. Habitat disturbance has increased (Watson et al. 2009), diseases such as avian pox and plasmodium parasites have been introduced (Wikelski et al. 2004; Kleindorfer and Dudaniec 2006; Levin et al. 2009; Parker et al. 2011), and introduced dipteran nest parasites are negatively affecting reproduction and recruitment in endemic bird communities (Dudaniec and Kleindorfer 2006; Dudaniec et al. 2007).

Other island systems demonstrate how human activities can have devastating effects on endemic avian communities. For example, the Hawaiian endemic finch radiation has lost most of its species to extinction (James and Price 2008). In contrast, the endemic finches of Galápagos have not yet experienced a known species extinction, yet there are signs of decline, and their future trajectory is uncertain. Several local extinctions have been documented in the finch radiation over the past century (Grant 1999; Grant et al. 2005), and two species, the mangrove finch and medium tree finch, are now critically endangered (IUCN Redlist 2010). The increase in environmental pressures in the Galápagos over the past century allows us a unique opportunity to directly investigate how individual island populations are responding to this changing environment, and what role metapopulation dynamics may play in this system.

We compared historic and modern genetic diversity in 11 different populations of warbler finches to test the hypothesis that populations are in general decline. We predicted that populations on smaller islands, or those most directly disturbed by human settlements, would show the greatest signs of decline. Small islands are more prone to extended droughts associated with El Niño/La Niña climatic cycles (Grant 1999), and genetic changes should be more easily detected in smaller populations. Even if whole metapopulations are stable, individual subpopulations may show dramatic changes (Hanski and Gaggiotti 2004). We therefore assessed changes in individual populations, again expecting small populations to show the most change. We also tested whether the magnitude and direction of interisland migration was consistent over time. Finally, we expect that any gains or losses in genetic diversity will be reflected by changes in inferred immigration rates over time.



Warbler finches are the most widespread of all Darwin's finches. All 14 known populations are morphologically and ecologically similar (Grant and Grant 2002), but they actually comprise two genetically distinct lineages based on mtDNA and microsatellite data (Freeland and Boag 1999; Petren et al. 1999a). Certhidea olivacea inhabits the large, central islands, whereas C. fusca occupies the smaller, peripheral islands of the archipelago (Tonnis et al. 2005). These species are strictly allopatric and represent the greatest genetic divergence found in the entire Darwin's finch radiation (Petren et al. 2005); we therefore treated them separately in this study. Warbler finches are morphologically distinct from, and distantly related to, finch species within the radiation that are known to hybridize, thus we assume that introgression is negligible.


A total of 219 modern and 192 historic tissue samples were collected for cross-temporal comparisons of 11 warbler finch populations (Table 1 and Fig. 1). For modern specimens, whole blood samples were collected in the field by venipuncture and dried on EDTA-treated filter paper. These samples were collected on various field expeditions to the Galápagos Islands during the years 1988–2006 (Petren et al. 2005). Museum specimen tissue for DNA extraction was obtained from toe pad shavings (approximately 3 × 2 mm) of Darwin's finches from the California Academy of Sciences, the British Natural History Museum and the American Museum of Natural History. The majority of specimens (∼80%) were obtained from Rollo Beck's collection from an 1899 expedition, and the California Academy of Sciences Galápagos expedition (1905–1906), both housed at the California Academy of Sciences, San Francisco, CA. Collection dates for all museum specimens were from 1894 to 1906 (Table S1).

Table 1.  Certhidea populations used for cross-temporal analysis. Time periods are historic (H) and modern (M); Sources for historic specimens are California Academy of Sciences (CAS), British Natural History Museum (BNHM), and American Museum of Natural History (AMNH).
IslandTime periodSourceDate n
  1. 1Samples for Española come from the main island (10) and the satellite island of Gardner (8). Excluding Gardner samples did not change the overall results for genetic diversity measures.

  2. n = number of samples analyzed.

  M field 1988, 1997 29
FernandinaHBNHM, AMNH, CAS1894, 1897, 189913
  M field 1999 19
GenovesaHBNHM, CAS1897, 190625
  M field 1988, 1997 23
  M field 1999 25
MarchenaHBNHM, CAS1897, 1899, 190622
  M field 1988  8
PintaHCAS1899, 190612
  M field 1997, 2001 19
PinzónHBNHM, CAS1899, 190619
  M field 2004 19
San CristóbalHBNHM, CAS1897, 190620
  M field 1999 19
Santa CruzHCAS190610
  M field 1988–1999 15
Santa FeHBNHM, CAS1897, 1899, 190618
  M field 1998–1999, 2004 12
  M field 1996 31
Figure 1.

A map of the Galápagos indicating islands sampled. A dashed line separates central C. olivacea occupied islands from more peripheral C. fusca occupied islands. Island abbreviations shown are used in subsequent figures.


DNA was extracted from modern blood samples using previously published methods (Petren 1998). Museum samples were stored and processed in a room dedicated to ancient DNA work, and separated from any modern specimens to avoid contamination (Petren et al. 2010). All equipment and work area surfaces were UV irradiated prior to and after each use, work areas were frequently bleached, disposable protective clothing was worn in the room, and access was restricted. DNA was extracted from museum specimens using GeneClean Ancient DNA kits (QBiogene, Carlsbad, CA) following the manufacturer protocol. Extracted DNA was eluted to a total volume of approximately 50 μL. Blank extractions (prepared with no tissue) were periodically processed to check reagents for contamination.

Fourteen autosomal microsatellite loci previously developed for Darwin's finches (Petren 1998; Petren et al. 1999b) were used to obtain genotype information from both modern and historic specimens. Amplification success declines rapidly with fragment size in degraded genetic samples (Sefc et al. 2003), thus PCR primers were redesigned to generate shorter PCR products (Petren et al. 2010). Total DNA was subjected to PCR in multiplex reactions (four loci per reaction with differing color fluorescent dyes) to increase genotyping efficiency and to conserve extracted template DNA. Negative control PCR reactions were also run with each batch of reactions prepared. PCR amplifications were performed in a total volume of 15 μL containing 7.5 μL QIAGEN (Valencia, CA) multiplex PCR master mix, 0.30 μM of primers, and 1 μL of extracted DNA under the following conditions: an initial denaturation step at 95°C for 15 min, followed by 33 cycles (40 cycles for historic samples) of 30 sec at 94°C, 1 min 30 sec at 52°, and 1 min 30 sec at 72°, and a final extension step of 72°C for 10 min. PCR reactions for historic samples were replicated three times each to recover alleles that may have failed to amplify (dropouts), and thereby reduce genotyping error. PCR products were analyzed with a LIZ-labeled size standard, on an Applied Biosystems 3730xl DNA Analyzer at the Cornell University Life Sciences Core Laboratories Center. Raw traces were genotyped by hand with the aid of GeneMapper software (Applied Biosystems, Carlsbad, CA). Modern and historic genotypes were scored independently, and historic specimens were scored without knowledge of population origin. Singleton alleles were identified by population and reevaluated for accuracy. Individual museum specimens with less than 50% genotype recovery across loci were excluded from analyses.

To assess the quality and repeatability of historic genotypes, a subset of 10 randomly chosen individuals were subjected to sixfold replicated genotyping across all loci. Replicates were scored independently, then compared to quantify frequencies of allelic dropout and spurious alleles. A principal coordinates analysis (PCA) was conducted using all available microsatellite markers previously developed for this group (16 total loci including two sex-linked; Petren 1998; Petren et al. 1999b) with both modern and historic genotypes to examine the congruence between past and present datasets.


Fourteen autosomal loci were subjected to genetic analysis using GenAlex (Peakall and Smouse 2006) and FSTAT (Goudet 1995) to calculate basic genetic summary statistics of the populations at each time point, including allelic richness (AE), and expected (HE) and observed (HO) heterozygosities. To account for differences in sample size, allelic richness was calculated using a rarefaction method for a minimum sample size of three individuals (FSTAT; Goudet 1995). GDA (Genetic Data Analysis, version 1.0; Lewis and Zaykin 2001) was used to perform an exact test (Guo and Thompson 1992) with sequential Bonferroni correction (Rice 1989) to determine which loci deviated significantly from Hardy–Weinberg proportions. Historic and modern mean values for AE, HE, and HO were compared to determine if there was a significant change over time within a single population using Wilcoxon tests paired by locus. Genetic parameters were also calculated and compared for pooled populations at both time points to detect any changes in genetic diversity across the archipelago as a whole over time. In addition, theta (θ) values (Weir and Cockerham 1984), the FST analogue that accounts for variation among populations, were calculated between time points using the GDA program, and a 95% confidence interval was used to determine if significant divergence in allele frequencies occurred within each population over time. Theta was also calculated for all population pairs within species at each time point. Weir and Cockerham's θ will be referred to as FST to avoid confusion with θ used by MIGRATE for population size. Mantel matrix correlations (rm) were conducted for pairwise FST between time points to determine if genetic structuring among populations remained stable over time, and tests of significance were based on 1000 permutations (Smouse, Long and Sokal 1986). We also used a resampling approach with a minimum sample size of 10 individuals and 100 replicates to determine if changes in genetic diversity could be attributable to sampling effects in historic samples. This approach was previously used for modern samples, and indicated that variation in measures of heterozygosity did not differ substantially between sample sizes of six and 16 individuals (Petren et al. 2005).

To test the hypothesis of general population decline, genetic diversity measures were evaluated across time with a repeated measures ANOVA. Changes in genetic diversity measures over time (calculated as percent change from past to present) were also analyzed for correlation with island elevation and log10 transformed island area to test whether smaller islands had the largest changes in diversity through time.

We searched for evidence of recent interisland immigration by evaluating genetic structure using the Bayesian clustering method of Pritchard et al. (2000) implemented in STRUCTURE (version 2.3.2). We evaluated support for populations (K) using 10 replicate runs of the admixture model under default conditions with alpha inferred. Island of origin was used to initialize simulations and infer recent immigrants up to two generations prior to sampling using the admixture model. We used a burn in of 30,000 prior to an equal umber of Markov chain Monte Carlo (MCMC) replicates.


We used three methods to reconstruct population history based on a single recent snapshot of genetic variation. MSVAR (Beaumont 1999) was used to evaluate historic population size trends. Specific conditions and details are provided in Table S2. The BOTTLENECK (Cornuet and Luikart 1996) program was used to test each population at each time point for evidence of a recent population bottleneck. A two-phase mutation model (80% stepwise) was used because this is more appropriate for microsatellite data than a strict stepwise model (Luikart and Cornuet 1998). A coalescent-based method was used to estimate population size while also providing estimates of migration rates (MIGRATE 3.0.3; Beerli and Felsenstein 1999, 2001), for both past and present time points. MIGRATE is widely used, highly cited (ISI listed 885 citations in February 2011), and it is relatively robust with regard to missing populations and violations of assumptions, including changes in population size, migration and mutation rates over time if used properly (Beerli 2004, 2007). MIGRATE was run under the maximum likelihood framework using the Brownian motion model of microsatellite evolution, with randomly generated starting trees. The Brownian motion model typically provides results that are very similar to the more time-consuming stepwise model with our data. Unknown alleles were excluded, and searches included 10 short chains and three long chains run with an adaptive heating scheme to increase the parameter space explored. Samples were taken every 20 steps, with a burn-in of 10,000. Simulations were used to evaluate all population sizes and bidirectional migration parameters (θ and M, respectively). Profile likelihoods were calculated but not reported because the 90% confidence intervals for any single simulation were consistently narrower than the variation between simulations using identical starting conditions. Following the manual's recommendation, 10 replicate runs were performed for each dataset and results were averaged and compared using a Wilcoxon signed rank test. Mantel tests were conducted to determine if overall migration rates (mean number of immigrants exchanged for each pair of populations) and directionality of migration (difference between bidirectional estimates for pairs of islands) were significantly correlated between past and present time points. We also tested for changes in population size and migration rate over time with a repeated measures statistical model using separate estimates of migration and population size for each locus.

We expected that genetic diversity estimates and population size (θ= 4Neμ) would be higher overall for C. olivacea populations than for C. fusca, because the latter inhabit smaller, peripheral islands. An ANOVA was used to test the hypothesis that population sizes were generally larger for C. olivacea than for C. fusca, with factors including species, time period, the interaction of time and species, and island included as a random nested factor within species. A similar ANOVA was used to test whether migration rate estimates differed between C. olivacea and C. fusca populations.



Of the 2688 historic individual × locus genotypes possible, 84% were recovered for this study. Genotyping success varied greatly by locus (66–99%), with four loci falling below 80% recovery. Exclusion of these four loci increased the genotyping success rate to 90%. Based on the subset of 10 individuals with sixfold replication, allelic dropout was estimated to impact approximately 26% of PCR replicates, whereas spurious alleles affected about 4% of PCR replicates. Using this calculated allele dropout rate, the probability of missing a second allele in all three PCR replicates for a given sample was less than 2%. Twenty-five of the 154 total locus/population combinations were out of Hardy–Weinberg equilibrium after Bonferroni correction. A single locus (Gf13), accounted for 28% of these deviations, whereas the remaining locus/population combinations were nearly equally distributed between historic and modern samples (45% and 55%, respectively), suggesting allelic dropout in historic specimens was not the primary cause of deviations from equilibrium. Modern and historic population genotypes clearly clustered together according to the PCA analysis, which supports the conclusion that historic specimens provide reliable genetic information about these populations (Fig. 2).

Figure 2.

PCA plot of historic (hollow symbols, dashed lines) and modern (solid symbols and lines) genetic data. Circles are 50% centroids. The first two axes account for 51.1% and 14.5% of the variation in the data.


Pairwise FST distances among islands were significantly correlated between time points (C. olivacea P < 0.02; rm = 0.88 and C. fusca P = 0.02; rm = 0.85), indicating no substantial changes in overall population structure at the landscape level over the past century (Table S3). Four individual populations, Fernandina, Genovesa, Isabela and Santiago had significant cross-temporal FST values (Fig. 3 and Table S4). Thus, substantial changes in allele frequencies occurred between sampling time points for at least one population within both the C. olivacea and C. fusca groups.

Figure 3.

Summary genetic data: (A) allelic richness (AE), (B) expected heterozygosity (HE), and (C) observed heterozygosity (HO) calculated as means across 14 loci. White represents historic populations and shaded modern. Bars indicate standard deviation. Islands are presented largest to smallest within species, with C. olivacea to the left of dashed line, and C. fusca to the right. Islands with significant cross-temporal FST values are indicated by squares around the island abbreviations. P values are indicated by * (<0.05) or ** (<0.01).

Population structure analyses confirmed the six past and present C. fusca populations, but the five C. olivacea populations had equivocal support because Fernandina and Isabela birds were not partitioned in either time frame (Fig. S1). Only three recent immigrants were detected, on Pinzon, from Santa Cruz. Low levels of structure were detected among past and present populations within some islands, especially when sampling time was used to help define groups (Table S5).


Direct cross-temporal comparisons of genetic diversity did not reveal consistent declines over time across the archipelago (F1,143= 0.22, P = 0.64 for allele richness, AE; F1,143= 0.02, P = 0.89 for expected heterozygosity, HE). Genetic diversity was significantly higher in C. olivacea populations from larger central islands when compared to C. fusca from smaller peripheral islands at both time points (F1,9= 22.7, P < 0.01 for AE; F1,9= 21.7, P < 0.01 for HE), whereas the interaction of time and species was not significant (F1,295= 0.78, P = 0.38 for AE; F1,295= 0.67, P = 0.41 for HE). Changes in allele richness were not correlated with island size (R2 = 0.02; P = 0.70) or elevation (R2 = 0.09; P = 0.36). When data were pooled among populations, allele richness for the metapopulation as a whole did not change over time (P = 0.22). Significant declines in genetic diversity over time were not apparent for islands with permanent human settlements; Santa Cruz, Isabela and San Cristóbal (P = 0.19, P = 0.72, and P = 0.43 for allelic richness, respectively).

Changes in genetic diversity over time for individual populations did not parallel changes in allelic composition (FST) over time (Fig. 3). Four island populations (Pinta, Pinzon, San Cristóbal, and Santa Fé) showed no evidence of change over time in any of the genetic diversity measures evaluated, whereas the remaining seven islands showed significant changes in at least one genetic measurement over time (Fig. 3 and Table S4).

Genovesa and Marchena showed significant declines (P < 0.05) in allelic richness (20% and 18%, respectively), with accompanying decreases in expected heterozygosity of >25% over the time interval examined (Fig. 3). Resampling of genotypes for Genovesa indicated a standard deviation of only ±0.12, or 6% of the mean value, for allelic richness, and 3.3% for expected heterozygosity. Therefore, significant differences observed across time are not likely due to sampling artifacts. Tests for genetic bottlenecks (Cornuet and Luikart 1996) did not reveal any evidence of recent population decline in these populations (Table S6). Of the 14 loci examined, five and four previously variable loci have become fixed for a single allele in these two respective modern populations. These fixed loci are uninformative for bottleneck inferences based on the method used, but contributed to the significant change in allelic composition (FST) across time that was noted for Genovesa (Fig. 3). Fernandina was the only population with a statistically significant test for a recent population bottleneck, which was indicated by modern samples (Table S6), but there was no supporting evidence of a loss of genetic diversity over time.

Populations of C. olivacea on Isabela and Santa Cruz showed a significant increase in observed heterozygosity over time (P < 0.01; Fig. 3). Genetic diversity increased substantially (∼20%) on Santiago, but not evenly across loci, thus the change was not statistically significant. The population of C. fusca on the small peripheral island of Española showed a statistically significant increase in allelic richness (18%; P = 0.002) and expected heterozygosity (35%; P = 0.001) over time.


Inferred scaled population sizes (θ= 4Neμ) generated by the program MIGRATE generally increased over time across all populations (F1, 207= 6.22; P = 0.013). Scaled population sizes were generally larger for C. olivacea populations from larger central islands than for C. fusca populations from smaller peripheral islands (F1,9= 12.8, P = 0.006), whereas the species by time interaction revealed that C. olivacea populations increased to a greater extent over time (F1, 207= 5.78; P = 0.017). These results are in stark contrast to results obtained from a different method (MSVAR; Beaumont 1999) that indicated dramatic declines in population size (90–100%) in all warbler finch populations based only on genotypes from recently collected specimens (Table S2). The discrepancy is likely due to the fact that MIGRATE accounts for immigration and MSVAR does not.

Simulations with migration indicated statistically significant increases in C. olivacea population sizes (θ= 4Neμ) on two of the five islands, Santiago and Pinzon, between time points (P < 0.01). However, these changes were small in magnitude, less than 7% (Fig. 4). Significant changes in population size were noted in two of the six C. fusca populations; Marchena decreased over time, whereas the Española population increased. However, as with C. olivacea, these changes were relatively small in magnitude, 6–8% (Fig. 4).

Figure 4.

Average (A) scaled population sizes (θ= 4Neμ) and (B) total number of immigrants per generation coming into a population from all source populations, generated from 10 replicate MIGRATE runs. White represents historic populations and shaded modern. Bars indicate standard deviation. C. olivacea are to the left of dashed line, C. fusca to the right. P values are indicated by * (<0.05) or ** (<0.01).


Overall levels of migration among islands were similar between C. olivacea and C. fusca (Fig. 4; F1,9= 0.016, P = 0.90), thus there was no clear difference between smaller and larger populations. However, the proportional genetic effect of these migrants would likely be greater in the smaller C. fusca populations. Levels of migration for both species increased significantly over time (F1,207= 0.016, P < 0.001), but these increases appeared to be largely driven by increases in just a few populations. The species by time interaction was not significant (P = 0.186). Mantel tests for similarity of migration rates for individual island pairs across time were highly correlated for both C. olivacea (P = 0.01; Rm= 0.92) and C. fusca (P < 0.01; Rm = 0.72) populations. Mantel tests for patterns of directional migration revealed significant similarity for C. fusca populations over time (P = 0.01; Rm = 0.69), but not for C. olivacea (P = 0.13; Rm = 0.48), indicating substantial changes in the directionality of gene flow over time among larger central islands. Figure 5 shows directional migration rates by receiving island, each with several source islands, for the two time points. Linear relationships are apparent, but slopes are shallower than 1:1. Islands that received fewer immigrants historically tended to receive more immigrants recently, and vice versa.

Figure 5.

Historic versus modern migration rates (number of migrants per generation) by receiving island for (A) C. olivacea and (B) C. fusca populations. Each point represents migration from a single source population. Solid line indicates best fit line, dotted line indicates equality of past and present values. Points below the dotted line have higher migration in the past, above the line higher migration at present.

Significant changes in migration rates over time were noted for two of the five C. olivacea populations. Total immigration into the Santa Cruz population increased by 33%, from 2.7 to 3.6 immigrants per generation. Estimated immigration rates for Santiago more than doubled over time, from 1.1 to 2.8 immigrants per generation (Fig. 4).

For the C. fusca populations, immigration into Pinta and Española increased significantly over time (40% and 132%, respectively, Fig. 4). Immigration rates into Genovesa were lower (by 15%) in the modern dataset (3.4 to 2.9 immigrants per generation), although the change was not statistically significant (P = 0.14). However, the changes in migration rates for both Genovesa and Española were statistically significant (P < 0.01) when migration data were analyzed by locus across time.

Changes in migration rate generally corresponded to changes in genetic diversity over time (Fig. 6). Significant correlations were noted between percent change in migration and change in allelic richness (P < 0.01; R2 = 0.66), and change in expected heterozygosity (P < 0.01; R2 = 0.58). Santa Cruz showed consistent but small magnitude increases in all three categories. Santiago and Española both showed consistent increases in diversity, population size, and immigration. Unusually low historic migration rates in these populations rebounded to higher levels over time. Marchena and Genovesa lost genetic diversity and decreased in size, whereas inferred levels of immigration remained consistently low over time.

Figure 6.

Percent change in mean values over time for (A) allelic richness, (B) total migration, and (C) estimated population size (θ= 4Nμ) from past to present. Islands are presented largest to smallest within species, with C. olivacea to the left of dashed line, and C. fusca to the right.


Genotypes from museum specimens allowed us to examine genetic changes in several interconnected populations of Darwin's finches over the last century. Our results suggest that the dramatic population fluctuations observed on Daphne Major (Grant et al. 2000, Grant & Grant 2006) are a general feature of other populations across the archipelago. We rejected the hypothesis that warbler finches are in general decline as reflected by genetic diversity. Decline was not associated with direct human disturbance and habitat loss, nor was there evidence of consistent decline in smaller populations on smaller islands that are more prone to drought. Declines in diversity occurred in two smaller populations. Surprisingly, several populations showed an increase in genetic diversity over time, accompanied by increased migration rates compared to historical levels. Although the broadest patterns of immigration were consistent over time, there were substantial changes in the direction and rate of gene exchange among specific islands over time. Tests for genetic bottlenecks, recent immigration and recent population growth did not reveal this history of change. This study highlights both the limitation of inferences based on single time points, and the utility of direct cross-temporal comparisons using historical collections.


The quality assessment of historic genotypes conducted for this study will be informative for future cross-temporal comparisons using natural history collections. The reliability of genotyping varied by locus, with a significant negative correlation between amplification success and fragment size (P = 0.003). Reliability also varied by specimen, which may be related to environmental and storage conditions during or shortly after study skin preparation (i.e., temperature and humidity). Ten individuals (∼5% of total museum specimens evaluated) were excluded from the study as a result of failure to recover >50% of their genotype information. Overall success rates for specimens in this study (1899–1906) were much greater, and therefore were more reliable, than those from the Beagle voyage in 1835 (Petren et al. 2010).

Reducing the size of targeted loci, replicate genotyping, extensive use of controls, and an internal quality analysis were the most effective approaches to ensure high-quality genetic data from these historic specimens (Gilbert et al. 2005). The underestimate of heterozygosity due to allelic dropout was less than 2%. The recovery of heterozygotes with sixfold replication suggests that misidentification of heterozygotes is most likely to occur when PCR replicates fail, rather than through allelic dropout in several successful PCR reactions, which is informative for quality control in future studies. In this study, the majority (92%) of the historic genotypes were based on multiple successful amplifications. We conclude that the magnitude of cross-temporal changes found in this study greatly exceeds what could be attributable to genotyping error.


Declines in genetic diversity were confined to two small, peripheral islands of the archipelago and were not associated with human inhabited islands where anthropogenic disturbance is most extensive. The erosion of genetic diversity in these populations is most likely due to genetic drift caused by periods of population decline associated with natural climate cycles. Genovesa and Marchena are low elevation, peripheral islands that are highly impacted by periodic El Niño events, which bring abundant rain to otherwise arid islands, generating a boom of primary productivity that continues up the food chain. This period of abundant food resources leads to a spike in reproductive output of finch populations (Grant et al. 2000). However, dry conditions typically follow El Niño events, resulting in food scarcity and massive mortality (Grant and Grant 1992). These periodic “boom and bust” cycles may differentially impact a subset of smaller islands and erode genetic diversity over time without evidence of a single, recent bottleneck event (Vucetich and Waite 1999).


We detected widespread evidence of genetic movement among warbler finch populations. Islands received an average of three to four migrants per generation, which lies at the lower end of the spectrum of inferred migration rates for Darwin's finches (Petren et al. 2005). Overall patterns of immigration were relatively constant, whereas directional patterns were more variable over time, as indicated by matrix correlations. Substantial changes in gene flow occurred for particular islands over time and largely corresponded to cross-temporal trends in genetic diversity (Fig. 6). Surprisingly, three large island populations had substantial changes in migration rate and/or inferred population size.

Española showed significant increases in all measures of genetic diversity over time. This increase is attributable to immigration, because other causes are much less likely. Mutation is unlikely to introduce substantial new genetic variation over the 100-year time frame examined here. Introgression is not likely for the morphologically and genetically distinct warbler finches, which have a lower propensity to hybridize compared to other Darwin's finches (Grant 1999). Locations where specimens were collected on each island may have differed at the two time points and contributed to observed differences, but within-island geographical population structure tends to be subtle (e.g., de León et al. 2010).

Migration rate estimates increased 132% for Española between the historic and modern datasets (Fig. 4), which complements the genetic diversity increases of 18% for AE and 35% for HE (Figs. 3 and 6). No specific source population for migrants could be identified, as substantial increases were inferred from several other islands (90–273%). Española had the lowest historical genetic diversity, so the increase may represent a recovery from a historical population crash and return to equilibrium. An influx of immigrants is expected to have a more substantial impact on genetic diversity in a genetically depauperate population than on a more genetically diverse one. This pattern is consistent with a natural genetic rescue (Brown and Kodric-Brown 1977), but any affect of the genetic diversity increase on fitness remains unknown.


The Galápagos warbler finches show many of the predicted dynamics of metapopulations over century-long time scales. Metapopulation dynamics can buffer species from extinction by recolonization of empty habitat patches, but also through the maintenance of genetic variation (Reed 2004). Although genetic diversity may be lost in individual populations through isolation, drift, and inbreeding (Wright 1969; Keller 1998), alleles may be retained elsewhere in the metapopulation and be infused back into a declining population through migration at a later time (Brown and Kodric-Brown 1977; Reed 2004). The gain in diversity on Española shows that genetic infusion through immigration does occur, and may therefore play a role in species persistence, evolution, or both. Low levels of gene flow into small populations can be a recipe for adaptive divergence under some conditions (Whitlock et al. 2000; Church and Taylor 2002). Across all islands, past migration rates that were either high or low tended to change more and in the opposite direction (Fig. 5). This pattern of compensation is consistent with expectations based on metapopulation dynamics, where local stochastic fluctuations tend to be evened out over time. Local stochastic fluctuation amid global stability is a hallmark feature of metapopulations (Young and Clarke 2000; Hanski and Gaggiotti 2004).

The lack of evidence for a general loss of genetic diversity, and the apparent increase in population sizes over time, suggests warbler finches have a high probability of long-term persistence. An increase in population size is consistent with the recent increase in the frequency and severity of El Niño events (Guilderson and Schrag 1998). El Niño events bring rain, an increase in reproduction, and an apparent increase in finch movements between islands (Grant 1999, Grant et al. 2000). Thus, increases in population size and immigration are expected to be coupled if El Niño events are responsible. However, these results should be viewed with caution. The apparent extinction of the Floreana warbler finch (Grant et al. 2005) may indicate other processes are at work. An overall decline in habitat quality may promote greater movement among islands as birds search for suitable habitat. It is conceivable that increased immigration could temporarily mask an overall decline, but the conditions over which this could happen are not clear. It also remains to be seen whether other species of Darwin's finches will show a similar pattern of population growth over time. There is reason to suspect that other species may be on a different trajectory. For instance, diet differs dramatically among the seed-eating ground finches, the tree finches, and the vegetarian and warbler finches (Grant 1999). Larger bodied species may also be more heavily impacted by a recently introduced dipteran nest parasite that causes very high nestling mortality (Dudaniec et al. 2007; Kleindorfer and Dudaniec 2009).

Three species of Darwin's finches are not part of metapopulations, and it is worth noting that their lack of adaptive divergence in one instance, and their apparent decline in the other two, may be affected by the absence of population substructure. The Cocos finch (Pinaroloxias inornata) remains undifferentiated within the remote, isolated Cocos Island off the coast of Costa Rica (Grant 1999). The mangrove finch (Cactospiza heliobates) and the medium tree finch (Camarhynchus pauper) are currently confined to single islands (Grant 1999), and they are the only two species currently listed as endangered (IUCN Redlist 2010).


In our analyses, methods that take migration into account performed better than those that did not, particularly those that attempted to reconstruct ancestral population size and bottlenecks. Migration may yield misleading results if not accounted for or ruled out in other systems. The use of natural history collections enabled us to reveal substantial changes in genetic diversity and gene flow over time. This suggests that inferences based on genetic variation from a single time point should be regarded with caution. The stochastic nature of environmental change in the Galápagos is a likely cause of the observed rapid evolutionary change in Darwin's finches. It appears that only historical samples allow one to recover this recent history, yet their availability is regrettably limited in most systems. It remains to be seen whether the volatile history of populations revealed here is a common feature of other natural populations, or whether the Galápagos finches are unusual in this regard, just as they are unusual in their rate of adaptive radiation.

Associate Editor: D. Posada


We thank T. Chesser and J. Cracraft of the American Museum of Natural History, J. Dumbacher, M. Flannery, D. Long, and L. Baptista of the California Academy of Sciences and R. Prys-Jones from the British Natural History Museum for access to valuable historical specimens. We thank the Galápagos National Parks and Charles Darwin Research Station for field support. We thank K. Short, J. Niedzwiecki, and E. Ristagno for laboratory and field assistance and H. Lisle Gibbs, T. Culley, S. Matter, R. DeBry, L. Kubatko and two anonymous reviewers for constructive comments. This work was partially supported by the National Science Foundation (DEB-0317687 to KP), Sigma Xi, The American Ornithologists’ Union and the University of Cincinnati University Research Council and Wieman-Wendell grant funds.