Extreme philopatry and genetic diversification at unprecedented scales in a seabird

Effective conservation requires maintenance of the processes underlying species divergence, as well as understanding species’ responses to episodic disturbances and long-term change. We explored genetic population structure at a previously unrecognized spatial scale in seabirds, focusing on fine-scale isolation between colonies, and identified two distinct genetic clusters of Barau’s Petrels (Pterodroma baraui) on Réunion Island (Indian Ocean) corresponding to the sampled breeding colonies separated by 5 km. This unexpected result was supported by long-term banding and was clearly linked to the species’ extreme philopatric tendencies, emphasizing the importance of philopatry as an intrinsic barrier to gene flow. This implies that loss of a single colony could result in the loss of genetic variation, impairing the species’ ability to adapt to threats in the long term. We anticipate that these findings will have a pivotal influence on seabird research and population management, focusing attention below the species level of taxonomic organization.

A topic of critical importance in conservation biology concerns the mechanisms through which evolutionary changes occur, including the influence of population connectivity, and the implications of variation in connectivity on a species' long-term survival and conservation needs 1,2 . Seabirds present several challenges to the generally accepted mechanisms of population differentiation in other vagile groups 3,4 and, despite recent advances in genetic theory and approaches, the scales at which genetic differentiation exists among their populations remains extremely difficult to predict 5 . Seabirds have an almost unsurpassed flight potential and have been known to cross physical features that limit dispersal in less motile groups [6][7][8] . Consequently, one could expect seabirds to disperse freely among their breeding sites yet growing evidence suggests that the potential for long-distance movement is not always the best indicator of gene flow 9,10 . This implies that philopatry and other intrinsic barriers to dispersal (e.g. differences in breeding phenology) may have important roles in the evolution of seabird diversity and endemism 1,3 , and that predicting genetic differentiation among populations for management or systematic purposes requires the assessment of these factors 4 . Despite this, little is known about the scales at which intrinsic barriers to genetic dispersal operate with few studies focusing on fine-scale inter-colony observations.
Understanding population genetic differentiation is particularly important in the case of rare, threatened, and highly localized seabirds 1,11 as the life-history attributes of most seabirds render their populations robust in the face of fluctuations in breeding success, but highly sensitive to changes in adult mortality 12 . An estimated 95% of all seabird species breed in limited numbers of highly synchronous colonies 12,13 , rendering them susceptible to extreme breeding failure in the event of adverse environmental conditions or prey shortages. Nevertheless, poor reproductive output must be long-term and extensive to result in decreases in populations, whereas even slight changes in adult mortality can have lasting population or even species-level consequences 12 . Delayed sexual maturity and high philopatric tendencies also imply that seabird populations are typically slow to recover and that recovery after local extirpation is unlikely without interventions such as translocation or social attraction 12 . Thus, assuming local populations differ genetically, the loss of even a single population may result in the loss of important genetic variation that might ultimately affect the species' ability to recolonize breeding sites, adapt to changing conditions, and possibly even to speciate 14,15 . Consequently, an understanding of the physical scales at which genetic differentiation occurs among populations is critical to the management of seabirds. This includes knowledge of their biology, ecology, and population trends including historical bottlenecks to identify population-specific management priorities.
Within the past few decades, seabirds have become exceptionally well-studied and knowledge of their overall conservations status and populations trends is considered more comprehensive than for any other group of marine organisms. However, Ref. 16 assessed the population trends for the world's seabirds and noted that as little as 19% of the global seabird population had been monitored more than five times between 1950 and 2010. Moreover, significant knowledge gaps exist with most colonies, and species in the tropics having been almost completely neglected. Among the most notable of the neglected species are the tropical petrels of the genera Pseudobulweria and Pterodroma (Procellariidae 11 ). Most of these species breed on one or a few isolated islands and almost all remain little-known despite their poor conservation status. In this regard, Réunion Island (western Indian Ocean) is unique among tropical islands in that it supports two endemic species of petrel, the Mascarene (Pseudobulweria aterrima) and Barau's (Pterodroma baraui) Petrels 17 . Both have unfavorable conservation status, suffering from threats imposed after the island was first colonized by humans in 1665 18 . These include light-induced mortality of fledglings and the predation of adults, eggs and chicks by invasive cats and rats 17 . The introduction and subsequent effects of these threats have increased dramatically following the rapid expansion of human activities during the last few decades. The Mascarene Petrel remains little-known, but over the last decade, considerable scientific attention has focused on the biology and conservation needs of the Barau's Petrel. Despite this, a critical knowledge gap concerns contemporary and historic genetic diversity, population size, and population structure.
To explore the links between intrinsic barriers to dispersal and genetic population structure, we employed diversity and structure analyses based on polymorphic microsatellites to investigate the genetic relationships between two proximate breeding colonies of the Barau's Petrel separated by roughly 5 km on Réunion Island (Indian Ocean; Fig. 1). The overall objective was to explore the influence of the tendency for a high level of philopatry on population genetic divergence at a previously unexplored spatial scale and to understand the implications of this for the species' conservation needs. We additionally tested for any recent changes to effective population size, phenotypic differences in morphology and, to quantify philopatry, we performed an analysis of band recoveries based on the long-term monitoring efforts on the Barau's Petrel.

Results
Tests of assumption and indicators of genetic diversity. Successful microsatellite amplification was achieved across 14 loci in more than 94% of the sampled individuals. The fifteenth locus, PB_1030, was excluded due to a high number of false amplifications (see Primer Screening). Concerning indicators of genetic diversity across the 14 successfully amplified loci-the mean allelic diversity was roughly six alleles per locus. Both colonies contained private alleles amounting to 5% and 11% of total allelic diversity, carried by 7 and 25 individuals for the Grand Benare and Piton des Neiges colonies, respectively ( Table 1). Means of N A and A E were significantly different in the global dataset (Wilcox. test: n = 14, w = 406, p value = 4.0 × 10 -6 ) indicating a high proportion of rare alleles (allele frequency < 0.05) within each of the two colonies (39% and 47% of total allelic diversity for the Grand Benare and Piton des Neiges colonies, respectively). Most estimates of genetic diversity were higher for Piton des Neiges (Table 1), though the estimates of A R , based on a minimum sample size of 114 individuals, were not significantly different between the two colonies (p value = 0.59). Finally, the overall dataset showed no devia- Unexpectedly, following multiple approaches (Fig. 2), the best-supported model for the Bayesian clustering analysis was that of two genetic clusters using sampling location as a priori information (refer to "Methods" section). This was reinforced by the median value of L(K). The two genetic clusters correspond perfectly to the two breeding colonies, with individual membership coefficients averaging 0.85 ± 0.04 and 0.83 ± 0.04 for birds sampled from Grand Benare and Piton des Neiges, respectively (Fig. 3). The DAPC clustering procedure produced similar results to the STRU CTU RE analysis, assuming K = 2 and using the first 80 principal components. Assignment probabilities of individuals sampled at Piton des Neiges averaged 0.79 ± 0.26 for Cluster 1, and assignment probabilities of individuals sampled at Grand Benare averaged 0.82 ± 0.23 for Cluster 2 (Fig. 4).  Burn-in period was set to 50,000 simulations followed by 100,000 repetitions. Each value was obtained by averaging the posterior probabilities of 10 independent runs. Shaded region highlights the optimal value of K based on both approaches. Estimation of the optimal number of genetic clusters based on the Bayesian clustering analysis performed in STRU CTU RE and following two approaches 61,62 . The best-supported model was that of two genetic clusters as indicated by the maximum value of ΔK and the lowest standard error for lnP(X|K) at K = 2. www.nature.com/scientificreports/ tively. These were more than double the approximations for the Grand Benare colony where Ne was estimated as 514 (95% CI 269-2237) and 664 (95% CI 343-3541) individuals at the cut-offs of 0.05 and 0.02, respectively. The classic Bottleneck analysis based on the IAM, TPM and SMM models proved inconclusive in determining recent changes in Ne for the two breeding colonies or for the overall sampled population ( Table 2). Nevertheless, the L-shaped allele distribution approach indicated that the two breeding colonies were in mutation-drift equilibrium suggesting that Ne has remained stable for at least a few dozen generations (Fig. 5). This is consistent with the M-ratio, which was high for Piton des Neiges (M = 0.94 ± 0.03) and Grand Benare (M = 0.96 ± 0.02) colonies as well as the overall dataset (M = 0.95 ± 0.03), suggesting that no severe population declines had occurred on recent timescales. Finally, using the coalescent likelihood MCMC approach, the estimates of the posterior distribution of past Ne revealed a relatively constant population size in the overall population for at least the last 5,000 generations (Current Theta index = 2.71, Intermediate Theta index = 2.68; Past Theta index = 2.49).   www.nature.com/scientificreports/

Morphometric comparisons between the pair of colonies. Morphometric measurements for adult
Barau's Petrels sampled in each of the two breeding colonies are presented in Table 3. One-way ANOSIM highlighted strong morphometric similarity among adult Barau's Petrels sampled at the Grand Benare and Piton des Neiges breeding colonies (one-way ANOSIM: global R = 0.01, p value = 0.03). The significant p value linked to the very low R-value indicated that any difference between the pair of breeding colonies was the result of only a very small proportion of the measurements. The overlap among birds from the two breeding colonies is displayed in Fig. 6.
Banding recoveries within and between colonies. The long-term banding information is consistent with our estimates of genetic differentiation, indicating extremely high colony fidelity among 2842 individuals banded as both adults (1984) and pre-fledglings (858). More precisely, among 2690 re-sightings of banded birds, all were recovered from the same colony in which they were first banded. This is indicative of extremely high natal philopatry (no birds banded as fledglings in a given colony have been re-sighted as either prospecting or breeding adults at the other colony) and breeding site-fidelity (no birds banded as a breeding adults in a given colony have been re-sighted as a breeding or prospecting adult at the other colony).

Discussion
The 'seabird paradox, ' or the contradiction between the high vagility of seabirds and their apparent reluctance to disperse among breeding sites 19 , has drawn attention to the scale of population genetic differentiation, the mechanisms driving differentiation, and the implications these factors have for seabird conservation 4 . The scale of population genetic structure varies extensively among studied seabird species, down to extremely small spatial and temporal distances [20][21][22] , although the scale to which we observed population structure in the Barau's Petrel is finer and more robust than has previously been recognized or explored ( Table 4). The three indices of genetic  In terms of the mechanisms of genetic population differentiation, our findings are clearly linked to the species' extremely high philopatric tendencies, the apparent avoidance of cross-breeding between colonies and the absence of any other obvious barriers to genetic connectivity. A suite of individual, social, ecological, and historical factors are frequently invoked to explain patterns of genetic population differentiation in seabirds 3,4 . Characteristics such as biometrics and coloration are often of limited use given their similarity among different populations of seabirds 25 and, in the case of the Barau's Petrels, birds from the two breeding colonies are phenotypically indistinguishable. In addition-birds from our two colonies do not differ in either their breeding phenologies or coloration 25,26 , though each of these factors has been used to explain genetic population divergence in at least a few other species 21,27,28 . It has been hypothesized that non-visual reproductive signals including olfactory and acoustic cues may serve as better indicators of reproductive isolation in largely nocturnal seabirds 1 , although these are logistically and technically difficult to explore under field conditions and remain untested for most species 29 . Behaviorally, Barau's Petrels are unique among small to medium sized petrels in that they arrive back at their colonies during daylight hours 17 . This is believed to be an energy-saving tactic, with birds using thermals to gain in altitude. However, recent radar surveys and thermal camera observations have shown that a large proportion of birds arrive at the colonies after sunset (authors' unpublished data), and the peak activity around the colonies happens after dark as in other related species. Thus, daytime arrival at colonies is unlikely to play a role in cross-colony interactions and colony partitioning. Moreover, several molecular studies have suggested that contemporary gene flow may not immediately override the influence of historical population genetic structure 3 . In the case of the Barau's Petrel, however, there is no evidence of historical range contractions www.nature.com/scientificreports/ or recent bottlenecks and the major threats the species faces (e.g. light-induced mortality, predation by invasive mammals 17 ;) are all relatively recent and linked to man's colonization and expansion on Réunion Island 18 . This implies that the main threats to Barau's Petrel populations may not yet have affected genetic diversity in this long-lived seabird. Nevertheless, the influence of small population size on genetic diversity cannot be neglected and, although our results highlight good genetic diversity with no evidence of inbreeding, both colonies contain rare and private alleles which are likely to disappear if these threats are maintained. Some seabirds exhibit population genetic structure in the absence of recognizable barriers to gene flow, suggesting that selective or social processes including philopatry can limit gene flow. Indeed, philopatry has been recognized as the second most obvious potential barrier to gene flow in seabirds after the physical isolation of breeding populations 3 . Given sufficient time, philopatry has the potential to restrict gene flow sufficiently to lead to reproductive isolation 5 , and has contributed to total speciation among populations of seabirds in a handful of examples 9,30,31 . However, philopatry alone may be insufficient to result in complete reproductive isolation as it usually acts in combination with other barriers to gene flow. In the case of the Barau's Petrel, the genetic implications of the species' philopatric tendencies appear to be exacerbated by high mate fidelity 32 . Barau's Petrels nest in high altitude elfin forest above 2400 m a.s.l. on the two isolated peaks of Réunion Island 23 . The locations of the colonies are determined primarily by habitat conditions, particularly where soil conditions are suitable for burrow construction. The colony on Piton des Neiges is the largest known colony, situated in an area of optimal Table 4. Estimates of population genetic structure based on microsatellite variation in nine diverse seabird species, by order of geographic distance between pairs of breeding colonies. Asterisks denote significant values as follows: *p < 0.05, **p < 0.01, ***p < 0.001. www.nature.com/scientificreports/ habitat conditions around 2400 m a.s.l., whereas the habitat on Grand Benare ranges from optimal to sub-optimal between the altitudes of 2600 m a.s.l. and 2800 m a.s.l. Thus, local environmental factors may additionally inhibit effective dispersal between the two breeding colonies, through the isolation of the nesting habitats and the habitat selectivity of the birds themselves. Importantly, however, it is not yet known whether Barau's Petrels will engage in subadult prospecting at non-natal colonies though the limited number of band recoveries of fledglings at colonies has yet to detect such behavior and colony exchanges of breeding adults appears to be genetically negligible. Thus, although philopatry may not be universal among all seabird lineages 33 , the extremely fine-scale at which it obviously contributes to genetic population divergence in the Barau's Petrel further emphasizes its role as an intrinsic barrier in the evolution of seabird diversity and endemism. Our results highlight that genetic studies below the taxonomic level of species are critical to the practice of applied conservation. Seabirds are disproportionately represented among the birds that are most at risk of extinction 11 . Their highly adapted biology reduces the capacity of populations to absorb additive mortality, particularly of adults, and slows their potential for recovery 1,11 . Moreover, the breeding distribution of many species is restricted to a limited number of sites, often on single islands or archipelagos, where populations can rapidly diverge 11,25 . The loss of genetic diversity within sub-divided populations can therefore have lasting impacts on the ability of the species to adapt, and possibly also to speciate. In practice, this means that seabird populations should be maintained across all conservation units to retain genetic variation at the species level and to protect the potential for future speciation. In addition to the genetic implications, the philopatric tendencies of most seabirds further imply that, without intervention (e.g. translocation or social attraction), a species would be slow to recover or repopulate areas following local extirpation 34 . This reinforces the fact that modern molecular approaches and population-level studies provide an invaluable perspective from which to view and delimit the appropriate units for management, offering powerful approaches for the assessment of the effects of disturbance on populations to the guidance of long-term recovery programs.

Materials and methods
Sample collection and microsatellite amplification. This study was carried out in compliance with the ARRIVE guidelines. Fieldwork was performed between the austral summers of 2008/2009 and 2012/2013 at two large breeding colonies of the Barau's Petrel on Réunion Island, Indian Ocean (Fig. 1). Breeding is highly synchronous between the two colonies, with no differences in foraging behavior during either the breeding or non-breeding periods 26,35 . A total of 259 adult birds were captured at their nests in the Grand Benare (n = 117) and Piton des Neiges (n = 142) colonies. A blood sample of approximately 0.2 ml was collected from each bird through medial metatarsal venipuncture and stored in 70% ethanol for further processing. All procedures were performed in accordance with relevant guidelines and regulations. We also received ethical approval from the Animal Ethics Committee (ZOOL-01-2013) at Rhodes University (South Africa), the Université de La Réunion (Réunion Island), Parc National de La Réunion, the Ethical Committee of Réunion Island, and the Centre de Recherche sur la Biologie des Populations d'Oiseaux (CRBPO, personal program of Matthieu Le Corre, PP6109). DNA was subsequently extracted from whole blood subsamples using the DNeasy Blood and Tissue extraction Kit (Qiagen).
A classic 3-primer Polymerase Chain Reaction (PCR) approach was used to amplify DNA at fifteen microsatellite loci following ref. 36 , alongside four fluorescently labelled dyes (6-FAM, PET, VIC, and NED) for the universal M13 forward primer enabling fragment analysis multiplexing 37 . PCR product sizes were determined using a 3730XL DNA analyzer (Applied Biosystems) by Gentyane platform (Clermont-Ferrand, France) and were estimated with the LIZ(500) standard using GeneMapper version 4.0 (Applied Biosystems).
Primer screening. Successful microsatellite amplification was achieved across the 15 loci in more than 55% of all samples from the two breeding colonies of the Barau's Petrel. Around 3.09% of the global dataset consisted of null values. Locus PB_1030 contained an extremely high percentage of missing data with 31.62% and 47.18% missing data for Grand Benare and Piton des Neiges, respectively. Removal of this locus for all further analyses reduced the proportion of missing data to within acceptable limits (a maximum of 2.56% at any one locus in either colony). The resulting genotype data were checked for amplification errors and the presence of null alleles using MicroChecker 2.2.3 38 . The only examples of stuttering were at loci PB_2742 and PB_3916, indicated by the highly significant shortage of heterozygote genotypes with alleles of one repeat unit difference. Null allele frequency was low and within acceptable limits, though null alleles were present for loci PB_1890, PB_2742, and PB_4708 in both colonies, and PB_3916 only in the Grand Benare colony. All loci for which null alleles were detected were associated with homozygote excesses. Each sample-locus combination was tested for linkage disequilibrium using GenePop 4.0.10 39 , employing the exact probability test (Markov chain parameters: 10,000 dememorizations, 100 batches, 1000 iterations per batch), and with False Discovery Rate correction 40 . Presence/ absence of linkage disequilibrium was confirmed with the method of Index of Association 41 , in the R package poppr 42 . No linkage disequilibrium was observed among any of the loci (all p values not significant) using either approach.
Estimates of genetic diversity. Estimators of genetic variability were calculated across the 14 successfully amplified loci including the allele frequencies at every locus, the average number of alleles per locus (N A ), the average number of effective alleles (A E ), private allele richness (PA), and the observed (H O ) and unbiased expected (H E ) heterozygosities according to Ref. 43 using GenALEx 6.5 44 . The statistical significance of differences between N A and A E were assessed using non-parametric Wilcoxon Signed Rank tests implemented in PAST 3.1.5 45 . Allelic richness (A R ; 46 ), adjusted for discrepancies in sample size using rarefaction, was additionally calculated for each breeding colony using FSTAT 2.9.3 47 , and compared between the colonies using 10,000 ) according to Ref. 51 using GenAlEx version 6.5 44 .
Genetic differentiation and structure. Wright's overall multi-locus fixation index (F ST 49 ), and its associated p value, were computed over all 14 loci to assess genetic differentiation following ref. 45 , using the R package adegenet 52 . The statistical significance of differences between the pair of breeding colonies in this index, was tested using 10,000 random permutations. Mutation processes occur at relatively high rates and with stepwise changes in allele sizes at microsatellite loci, which introduces bias into classical measures of population differentiation such as F ST 53 . Thus, to corroborate these results, we calculated two additional measures of genetic differentiation: R ST using SPAGeDi 53 , and G′ ST using GenoDive 3.04 54 . R ST uses a stepwise mutation model to assess variances in allele sizes rather than allele frequencies (as in F ST ) and better reflects population differentiation among microsatellite loci in instances when stuttering is observed 55 . G′ ST is equivalent to F ST but with different statistical properties 56 , and with correction for bias stemming from sampling a limited number of populations 57 . The simple test of permutated R ST (pR ST ), using 10,000 permutations of the genotypes, was additionally used to determine whether F ST equalled R ST and to assess whether stepwise-like mutations or genetic drift contributed to genetic differentiation 53 .
The multilocus genotype data were additionally used to perform a Bayesian clustering analysis implemented in STRU CTU RE 2.3.4 58 . Multiple independent simulations, using a burn-in period of 50,000 simulations followed by 100,000 repetitions, were first run using different settings (e.g. admixture versus no-admixture model, LOCP-RIOR verses no-LOCPRIOR) to assess convergence. The Dirichlet parameter (ɑ) was used to determine whether the standard admixture model was appropriate following Refs. 58,59 . Similarly, the r-index was used to determine the informativeness of the sampling location (LOCPRIOR), with low values of r indicating that sampling locations are informative to the overall model 60 . An admixture model (ɑ = 3.26 ± 0.57 58,59 ), assuming sampling location (LOCPRIOR; r = 0.74 ± 0.12 60 ), was most appropriate and provided the best levels of convergence. Correlated allele frequencies were assumed 58,60 . The number of genetic clusters (termed K) was subsequently determined following three criteria: (1) the log likelihood given K (lnP[X|K] 61 ), (2) the second-order rate of change of mean log-likelihood (ΔK 62 ) and (3) the median value of L(K). The first two were calculated using STRU CTU RE HAR-VESTER online Web server 63 , while the third was calculated using CLUMPAK 64 . Ten independent simulations, at the optimal value of K, were then run using 1,000,000 iterations each (after a burn-in of 500,000 steps). Finally, CLUMPAK 64 was used to find the optimal individual alignments of replicated cluster analyses and to plot the estimates of individual membership for all genetic clusters.
Lastly, population structure was explored using Discriminant Analysis of Principal Components (DAPC 65 ). This approach does not make any assumptions about HWE or linkage disequilibrium but is sensitive to fine genetic differences among populations. We used K-means clustering of principal components for K = 1 to K = 5 and Bayesian Information Criteria (BICs) to assess the optimal number of genetic clusters. The value of K with the lowest BIC value was considered optimal 66 . DAPC was applied using the Adegenet package 2.1.1 in R (52).
Effective population size and test of bottleneck. Contemporary estimates of effective population size (Ne) based on the 14 microsatellite markers were calculated for the two Barau's Petrel breeding colonies using the molecular co-ancestry method of Ref. 67 , as implemented in NeEstimator 2.0 68 . Ne was calculated assuming a monogamous mating pattern 32 . Two allele frequencies were adopted as conservative cut-offs to minimize the effect of the presence of rare alleles 68 : 0.02 and 0.05. Estimates of 95% confidence interval (CI) were calculated by jackknifing over loci for each estimate.
Finally, we also used Bottleneck 1.2.02 69 to test for recent changes in population size using the heterozygosity excess approach, assuming mutation-drift equilibrium as the null hypothesis. We used the Infinite Allele (IAM), two-phase (TPM; non-stepwise = 0.22%, variance = 12; typical values for many microsatellite markers 70 , and stepwise mutation models (SMM) based on 10,000 replications and the Wilcoxon sign-rank test of significance (two tailed for heterozygosity excess or deficiency). These results were confirmed with the L-shaped method of allele mutation-drift equilibrium 71 , the M-ratio calculated (mean ± S.E.) across all 14 loci using the R package StrataG 2.4.905 72,73 , and the approximate likelihood MCMC approach using modelled Theta values (Θ = 4Ne × Mu; where Ne is the effective population size and Mu is the microsatellite mutation ratio) using the VarEff 1.2 R package 74 .
Banding information and morphometric measurements. Annual monitoring at the Grand Benare and Piton des Neiges breeding colonies began over the austral summers of 2007/2008 and 2002/2003, respectively. Subsequently, the two colonies have been visited at least twice annually up to the present for banding and control of breeding adults (November/December) and for banding of pre-fledging chicks (March). All re-sightings to the present day (13 and 18 years of data for the Grand Benare and Piton des Neiges colonies, respectively), were collated to quantify philopatry and levels of individual exchange between colonies. Re-sightings of birds banded at two age groups (adults, pre-fledglings), within and between each breeding colony, were determined using dplyr 1.0.2 75 in R.
Morphometric measurements of adult Barau's Petrels were taken from a sample of individuals and compared between the two breeding colonies using One-way Analysis of Similarities (ANOSIM) based on an Euclidean distance measure and 9999 permutations in PAST 3.1.5 45 . Barau's Petrels are morphologically indistinguishable between the sexes 32 , and thus all individuals were pooled for the purposes of this analysis so as not to limit our ability to detect differences between the two colonies. Morphometric measurements were taken in the field with 1 mm precision and included the following: wing chord (AP), culmen length (LC), bill depth at the maximum gonydeal expansion (HC), maxillary unguis length (CR), and tarsus lengths (TA). www.nature.com/scientificreports/

Data availability
All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Information. Raw microsatellite genotypes of the 259 individual Barau's Petrels, for all 15 loci, are available in the supporting information (Table S1).