Introduction

Research on anthropogenic impacts on population connectivity (Ricketts 2001; Baguette and Van Dyck 2007), and growing attention to connectivity’s effects on population persistence (Fahrig and Merriam 1985; Hanski and Gaggiotti 2004), have made the study of fragmented populations highly germane to biodiversity conservation (Opdam 1991; Hawkes 2009; Goldberg and Waits 2010; Murphy et al. 2010). Fragmented populations suffer increased extirpation risk because these populations are often relatively isolated, and smaller, making them more strongly influenced by stochastic processes, all of which may result in increased likelihood of inbreeding depression, and low evolutionary potential due to decreased genetic diversity (Keller and Weller 2002; Mills 2007; Frankham et al. 2009). As connectivity among habitat patches is reduced, genetic drift can promote divergence in allelic and haplotypic frequencies among populations, resulting in genetic structuring when there is little or no emigration or immigration to counteract it (Selkoe and Toonen 2006; Broquet and Petit 2009). Conversely, habitat connectivity (Taylor et al. 1993) can ameliorate many risk factors and allow physically disjunct populations to persist as part of a network (Macdonald and Johnson 2001; Crooks and Sanjayan 2006) or, under particular circumstances, as a metapopulation of interconnected habitats (Hanski 1999; Smith and Green 2005; Doleman 2012). Thus, genetic structure can be a useful indicator of the interactions between spatially isolated populations, that provides potentially valuable information for their conservation (Gibbs and Reed 2007; Mills 2007).

Though many studies have investigated population connectivity in stable populations, the development of genetic structure in growing populations (e.g., recovering or following colonization) is less well-understood. Specifically, it is unclear at which temporal and spatial scales fragmentation (and consequently population differentiation) will occur in dynamic, human-altered landscapes. Research on the population structure of fragmented, recovering populations at small spatial scales may yield valuable insights into the demographic and genetic effects of recovery and landscape change of populations at larger scales (Greenbaum and Fefferman 2017; Moreno-Mateos et al. 2017).

Islands are particularly vulnerable to landscape changes because anthropogenic pressures can more readily affect a significant portion of the available habitat (Fordham and Brook 2010). Here we investigate the population structure of an endangered bird, the Hawaiian gallinule (`Alae `ula, Gallinula galeata sandvicensis), whose populations have been recovering from near-extinction in a highly modified landscape. The island of O`ahu, a moderately sized (1500 km2) volcanic island in the central Pacific Ocean once supported a diversity of terrestrial habitats, including extensive lowland forests and coastal freshwater wetlands (Ziegler 2002). Throughout the twentieth century, urban land cover has increased markedly with growing tourist visitation and resident population size while agricultural land use on the island decreased (Giambelluca 1986; Klasner and Mikami 2003; NOAA Coastal Services Center 2014). O`ahu is now heavily urbanized, supporting > 60% of the population of the state of Hawai`i (953,000 of 1.4 million people) in ~10% of its land area. More than 65% of O`ahu’s freshwater lowland wetlands have been lost to land conversion, primarily because they occupied ideal locations for agricultural residential areas (van Rees and Reed 2014). Due to the island’s topography, wetland habitats have always been spatially separated on the landscape, but habitat loss and anthropogenic landscape change may have introduced new barriers to dispersal among animal populations persisting in remaining wetland patches on the island. Accordingly, information on the population structure of wetland-specialist taxa on the island would be valuable for their conservation and management.

The Hawaiian gallinule is an endangered waterbird subspecies endemic to the Hawaiian Islands that lives only in shallow, coastal, freshwater wetlands. Hawaiian gallinules are hypothesized to be recent colonizers of the Hawaiian Islands (Fleischer and McIntosh 2001), with archaeological evidence of gallinules only found in deposits post-dating human colonization of the islands (~1500 years before present, James 1987). Hawaiian gallinules were formerly distributed among most of the main Hawaiian Islands (Hawai`i, Maui, Moloka`i, O`ahu, and Kaua`i), but were extirpated from all islands except Kaua`i and O`ahu during a period of severe population decline throughout the 19th and early twentieth centuries (Schwartz and Schwartz 1949; Munro 1960; Banko 1987). This decline was driven by the introduction of invasive mammalian predators (esp. small Indian mongoose Herpestes javanicus, domestic dogs Canis lupis familiaris, and cats Felis catus), habitat loss and degradation by agriculture and later urbanization, and habitat modification by exotic invasive plants (e.g., Urochloa mutica) (Griffin et al. 1989; Reed et al. 2012). The population reached a minimum estimated total population size of 57 individuals in the 1950s (Engilis and Pratt 1993). The establishment of National Wildlife Refuges on O`ahu and Kaua`i, and associated predator control and habitat restoration have helped Hawaiian gallinules make a steady recovery over the last 50 years, with current statewide population estimates from 400 to 1000 total individuals (Reed et al. 2011; Underwood et al. 2013; USFWS pers. comm.). Currently, between 200 and 300 Hawaiian gallinules persist on O`ahu (Reed et al. 2011; USFWS 2011) in a number of spatially isolated, relict or artificially maintained wetland habitats. These habitat patches occur within state, county, and federal protected areas, private farms, golf courses, and waterways. Wetland habitats are separated by a highly heterogeneous landscape of highways, agriculture, dense urban and suburban areas, and active military bases. To date, no formal studies have been undertaken to understand the genetic structure of gallinule populations among these physically isolated patches.

Hawaiian gallinules are a member of the family Rallidae (rails), which are known for their remarkable propensity to colonize islands and disperse long distances over open water despite many species being reluctant fliers (Ripley 1977). Rails are often freshwater wetland specialists, and can exhibit limited dispersal and subsequent genetic structuring under conditions of habitat loss and landscape change (e.g. California Ridgway’s rail, Rallus obsoletus obsoletus, Wood et al. 2017). Several rail species in North America are migratory (e.g. yellow rail Coturnicops noveboracensis, Sora Porzana carolina, common gallinule) with vagrant sightings well outside of their breeding range, indicating substantial dispersal abilities (Ripley 1977; Taylor and Van; Perlo 1998). In contrast to the migratory behaviors of other G. galeata subspecies, Hawaiian gallinules do not migrate, and are thought to be highly sedentary. For example, they have not colonized restored or unoccupied habitats on other islands in their historical range despite gradual population recovery (USFWS 2011). Hawaiian gallinules are weakly seasonal breeders, and can have multiple broods per year (average clutch size is three to six eggs; Byrd and Zeillemaker 1981) with an estimated generation time of 5.9 years (BirdLife 2016), though individuals are observed breeding at age 2 (van Rees, pers. obs.). For species with known limited dispersal propensities and relatively rapid generation times, understanding the level of connectivity (effective dispersal) among local wetlands is important when assessing the spatial ecology and likely persistence of small endemic populations (Van Strien et al. 2014). O`ahu’s small size, rapid landscape change, and the discrete distribution of wetland habitats provide an excellent study system for examining the spatial scales at which population structure may be influenced by changing landscapes, including habitat loss and matrix alteration, as well as by population recovery.

The structure and connectivity of O`ahu’s Hawaiian gallinule population is also pertinent to the long-term conservation prognosis of the subspecies. The majority of habitat patches on O`ahu support fewer than 30 individuals (CVR pers. obs.), and therefore each population may have a high extinction risk without genetic and demographic rescue. As has been demonstrated in other habitat specialist birds (e.g., Florida scrub-jay, Aphelocoma coerulescens, Chen et al. 2016), habitat fragmentation and reduced dispersal lead to increased inbreeding coefficients, affecting vital rates important to population persistence (e.g., increased hatch failure). Consequently, we investigated the spatial patterns of genetic diversity within Hawaiian gallinules on O`ahu to evaluate the effects of population decline and recovery in an urbanized, fragmented landscape.

Methods

Sample Collection

We captured Hawaiian gallinules using walk-in Tomahawk live-traps baited with cracked corn, cat food, or other attractive food items (e.g. mango fruit Mangifera sp.). We captured Hawaiian gallinules at 13 coastal wetlands on O`ahu in summer (May–July) from 2014 to 2016 (Fig. 1). Wetland names and their corresponding four-letter codes are given by region in Table 1: North Shore (includes TURT, JCNR, SHRI, WAIM, LOTU), Windward (KLIP, HAMA, ENCH, KAWA, OLOM), Maunalua (KEAW), and Pearl Harbor (POUH, HONO). Our sampled wetland sites represent all known major habitats for the species on O`ahu. James Campbell National Wildlife Refuge was sampled only in 2014. Marine Corps Base Hawai`i, Kahuku, Kawainui marsh, and Pouhala marsh were sampled only in 2016. All other wetlands were sampled in 2015 and 2016. Inter-site Euclidean distances ranged from 1.5 to 54 km, with a mean of around 30 km. Notably, sample sizes for some wetlands (POUH, HONO, KEAW, KAWA, and SHRI) are small; however, sample sizes represent between 20 and 100% of birds occupying individual wetlands. Due to small sample size and close geographic proximity, samples from Pouhala and Hono`uli`uli wetlands were pooled into a regional population, Pearl Harbor (PEHA). We collected 3–6 body feathers from post-fledging Hawaiian gallinules and stored feather samples in plastic bags at room temperature. We extracted DNA using DNeasy Blood and Tissue Kits (QIAGEN; Valencia, CA).

Fig. 1
figure 1

Locations of sampled wetlands on O`ahu; four-letter codes are defined in Table 1. Note that PEHA (Pearl Harbor) represents the combination of two adjacent wetlands with smaller sample size, POUH (Pouhala Marsh) and HONO (Pearl Harbor National Wildlife Refuge, Honouliuli Unit). Light gray regions indicate mountain ranges (elevation > 300 m) and darker gray represents developed and urban areas. Parenthetic values indicate the number of birds sampled at each site

Full size image
Table 1 Code names and regions of sampled wetlands on O`ahu that are depicted in Fig. 1, listed in clockwise order starting at the western side of the island
Full size table

Laboratory techniques

We genotyped individuals at 12 autosomal microsatellite loci: Fal08, Fal10, Fal12, Fal14, Fal 17, Fal19, Gch06, Gch12, Gch13, Gch17, Gch19 (Sonsthagen et al. 2014), and KiRa9 (Brackett et al. 2013). The forward primer for locus KiRa9 was modified for this study (KiRa9.1F: 5′-GCGAGACTTGAAGTAGTGG-3′). We amplified microsatellites using polymerase chain reaction (PCR), and electrophoresed with fluorescent-tailed primers following the protocols described in Sonsthagen et al. (2004). We also sequenced 520 base pairs (bp) of the NADH dehydrogenase 2 (ND2) region of mitochondrial DNA (mtDNA) using primer pair ND2_224L (5′-CTMCTACTATTCTCCAGCAC-3′) and ND2_720H (5′-GCCTGCTAGGGAKAG-3′), following Sonsthagen et al. (2007) for PCR amplifications, cycle-sequencing protocols, and post-sequence processing. Lastly, 10% of samples were randomly selected, reamplified, and genotyped in duplicate for quality control. No inconsistencies in genotype scores were observed between replicates. Sequences are accessioned in GenBank (MF673902–MF673904).

Analysis of genetic diversity

We calculated allelic richness (AR), inbreeding coefficient (FIS), observed and expected heterozygosity (HO and HE, respectively) and tested for Hardy–Weinberg equilibrium and linkage disequilibrium among microsatellite loci in FSTAT version 2.9.3.2 (Goudet 1995). We estimated haplotype (h) and nucleotide (π) diversity of the ND2 region of mtDNA using Arlequin 3.1 (Schneider et al. 2000). We tested for selective neutrality for sequence data from the mtDNA ND2 region using Fu’s FS (Fu 1997) and Tajima’s D (Tajima 1989) in Arlequin. We constructed an unrooted haplotype network for mtDNA ND2 using the reduced median method (Bandelt et al. 1995) in NETWORK 5.0.0.0 (Fluxus Technology Ltd.).

Because we collected samples from both adult and hatch-year individuals, we used Queller and Goodnight’s (1989) index of relatedness (rxy) to calculate levels of familial relationship between pairs of individuals within wetlands using IDENTIX 1.1 (Belkhir et al. 2002). Relatedness values of < 0.5 represent first order (i.e. full-sibling, parent-offspring) relationships, 0.25 for second order (i.e. half-sibling) relationships, and 0 for unrelated individuals.

Estimation of genetic structure

We calculated estimates of inter-population variance in allelic and haplotypic frequencies (FST) in Arlequin; P-values were adjusted for multiple comparisons using the B-Y method (Benjamini and Yekutieli 2001; Narum 2006) with a starting α of 0.05. We used RecodeData version 1.0 (Meirmans 2006) to calculate the maximum possible FST values for the sampled suite of microsatellite loci. Hierarchical analysis of molecular variance (AMOVA) was performed using Arlequin to determine the magnitude of spatial variance in haplotypic and allelic frequencies among wetland regions on the island. Because the Pearl Harbor and Maunalua regions were both represented by only a single wetland (Pearl Harbor made up of the combined wetland PEHA and Maunalua by KEAW), and groups must be represented by more than one population to avoid violating assumptions of the AMOVA analysis, we made comparisons only between the Windward side and North Shore regions. The Pearl Harbor region could not be considered a group based on the original populations HONO and POUH because of the small sample sizes of these sites. We tested for isolation by distance by comparing a matrix of Euclidean geographic distances and genetic distances (FST) between all wetland pairs using a Mantel test implemented in the isolation By distance web service (Jensen et al. 2005).

Finally we used STRUCTURE 2.3.2.1 (Pritchard et al. 2000; Hubisz et al. 2009) to infer the occurrence of population structure among sampled individuals without using prior knowledge of sample collection locations based on our genotypic microsatellite data. STRUCTURE was conducted using an admixture model assuming correlated frequencies and with sample group information as a prior, with the possible number of populations (K) ranging from 1 to 11 (search strategy; burn-in of 50,000 iterations followed by 500,000 Markov chain Monte Carlo iterations); the analysis was repeated ten times. We followed the method of Evanno et al. (2005) and maximum likelihood (Pritchard et al. 2000) to determine the most likely number of clusters (K) given the data. The results were summarized in STRUCTURE HARVESTER Web v0.6.94 (Earl and von Holdt 2012).

Estimation of gene flow

We estimated gene flow between wetlands in MIGRATE v3.6.11 (Beerli and Felsenstein 1999, 2001) based on the 12 microsatellite loci. We ran MIGRATE with a full gene flow model, θ (4 Neμ or Nfμ, composite measure of effective population size and mutation rate), and all pairwise gene flow parameters (M) were estimated individually from the data and were compared to a restricted island model for which θ was averaged and pairwise gene flow parameters were symmetrical between populations. We estimated gene flow using a maximum likelihood search parameter; ten short chains (5000 trees used out of 1,000,000 sampled), five long chains (15,000 trees used out 3,000,000 sampled), five static chains (start temperatures: 1, 1.5, 3, 6, and 12; swapping interval = 1) with a 5M burn-in per chain. Models were run three times to ensure the convergence of parameter estimates. The alternative model was evaluated for goodness-of-fit given the data using a log-likelihood ratio test. The resulting statistic from the log-likelihood ratio test is equivalent to a χ2 distribution with the degrees of freedom equal to the difference in the number of parameters estimated in the two models (Beerli and Felsenstein 2001).

We aggregated several wetlands that were adjacent and shared low genetic structure to reduce the number of estimated model parameters. JCNR and SHRI were pooled into JCSF (James Campbell-Shrimp Farms), and HAMA, OLOM, KAWA, and ENCH were combined into WIND (Windward side), resulting in a total of eight unique wetland patches in our gene flow analysis.

Results

Genetic diversity

We obtained multi-locus genotypes for 152 Hawaiian gallinules at 13 wetland sites on O`ahu (sample distribution shown in Fig. 1). Each individual had a unique multi-locus genotype. Four hatch-year birds had pairwise rxy values > 0.5 with adults sampled from the same wetland (HAMA n = 2, KEAW n = 1, WAIM n = 1), suggestive of a parent-offspring relationship. The four hatch-year birds were omitted from subsequent analysis; our final data set comprised 148 individuals (Table 2). Number of alleles per microsatellite locus ranged from 1 to 4, with a mean of 2.2 and standard deviation of 0.5 alleles per locus. The average number of alleles per locus was similar across sampled wetlands with values ranging from 2.1 to 2.3 (Table 2). Within wetlands, observed heterozygosity ranged from 33.9 to 52.0% (Table 2). All loci and populations conformed to the assumptions of Hardy–Weinberg equilibrium, and all loci were in linkage equilibrium.

Table 2 Estimates of genetic diversity of Hawaiian gallinule on O`ahu, Hawai`i, including; average number of alleles, allelic richness (AR), observed and expected heterozygosity (Ho, He, respectively), inbreeding coefficient (F IS ), and sample size calculated from 12 microsatellite loci, as well as, number of haplotypes (no. hapl.), haplotype diversity (h), and nucleotide diversity (π), Fu’s Fs, and Tajima’s D calculated from 295 bp of mtDNA ND2
Full size table

We observed three unique haplotypes at the ND2 region of the mtDNA (n = 148) characterized by two variable sites (Fig. 2). Number of ND2 haplotypes observed per wetland ranged from 1 to 3. Individuals from KLIP and KEAW were represented by a single haplotype (1), while all other wetlands were represented by at least two haplotypes. In wetlands with more than one ND2 haplotype observed, moderate levels of haplotype (h = 0.516–0.726) and nucleotide (π = 0.0026–0.0043) diversity were detected (Table 2).

Fig. 2
figure 2

Network diagram illustrating relationships of mtDNA haplotypes among sampled wetlands of Hawaiian gallinules on O`ahu. The size of the circle corresponds to the frequency of each haplotype, and connecting lines represent a change in a single nucleotide. Each sampled wetland has a unique shading pattern

Full size image

Genetic structure

Moderate to high genetic structure was detected based on the 12 microsatellite loci (FST = 0.098, p < 0.001, Table 3) and mtDNA (FST = 0.248, p < 0.001, Table 3). The upper limit of FST for our microsatellite data set is 0.584; therefore the FST of 0.098 accounts for 16.8% of the maximum possible level of genetic structure. KLIP and KEAW were differiented from most wetlands with the highest number of significant comparisons, followed by LOTU, HAMA, and TURT based on microsatellite data and WAIM and LOTU based on mtDNA variance. The regional analysis uncovered within-population structure (microsatellites FST = 0.096, p < 0.001; mtDNA FST = 0.248, p < 0.001); however, region (North Shore vs. Windward side) did not explain a significant portion the variance at either marker type (microsatellites FCT = 0.007, p = 0.301; mtDNA FCT = 0.112, p = 0.086). We observed no evidence of isolation by distance among wetlands based on microsatellite data or mtDNA (Mantel test, microsatellites: r = 0.107, p = 0.131; mtDNA: r = − 0.134, p = 0.926).

Table 3 Pairwise FST values for all sampled wetlands, separated by region
Full size table

STRUCTURE uncovered genetic partitioning of Hawaiian gallinules on O`ahu; two clusters were identified based on Evanno et al. (2005) method (ΔK = 53.3; LnPr|K = − 2417.6, Fig. 3a) and four clusters were identified based on the maximum likelihood given the data (ΔK = 2.0, LnPr|K = − 2374.3, Fig. 3b). Based on Evanno’s method, individuals from KLIP had high membership coefficient to group 1 (white), and individuals from WAIM had moderate coefficients (Fig. 3a). The individuals representing the remaining wetlands clustered into group 2 (black). Based on Pritchard et al. (2000) method, individuals from KEAW (group 2; light gray) and KLIP (group 1; white) formed wetland specific clusters (Fig. 3b). Individuals from KAWA, HAMA, ENCH, OLOM, and PEHA clustered together into group 3 (dark gray). Individuals from TURT and half of the LOTU individuals clustered into group 4 (black). Individuals from WAIM, SHRI, JCNR, and KAWA, had intermediate membership coefficients. The four-population model detected one known dispersal event (ENCH87856); that individual hatched in KEAW and dispersed to ENCH; this individual’s assignment profile more closely resembles those of birds from KEAW than ENCH. The sample group information was informative (r < 1.0) for both models. The average proportion of cluster membership across all individuals within each wetland are shown over their respective sampling location in Fig. 4.

Fig. 3
figure 3

Average membership coefficient of genotyped Hawaiian gallinules that were sampled at 12 wetlands on O`ahu, Hawaii; a shows assigned membership among two clusters (the most likely number according to the Evanno method), and b shows membership among four clusters (the most likely number according to the maximum-likelihood method) inferred with data from 12 microsatellite loci in STRUCTURE. Dashed lines separate individuals from different wetlands. Asterisk denotes an individual banded at KEAW (Keawawa wetland, Maunalua region) that was later resighted and captured at ENCH (Enchanted Lakes, Windward side)

Full size image
Fig. 4
figure 4

Within-wetland average membership coefficients of genotyped Hawaiian gallinules sampled at 12 sites on O`ahu, Hawai`i, USA. Image a shows assigned membership among two clusters (the most likely number according to the Evanno method), and b shows membership among four clusters (the most likely number according to the maximum-likelihood method) based on data from 12 microsatellite loci analyzed in STRUCTURE

Full size image

Gene flow

Asymmetrical gene flow was detected based on the microsatellite data; the full model (all parameters allowed to vary independently) had significantly higher likelihoods than did the restricted model (symmetric interpopulation M and θ) indicating asymmetric gene flow among wetlands [LnL(full) = − 4436, LnL(test) = − 4789, p < 0.001]. Gene flow (Nem) maximum-likelihood estimates and 95% confidence intervals are provided in Table 4, and gene flow between populations is displayed schematically in Fig. 5. We observed asymmetrical gene flow (as indicated by non-overlapping 95% confidence intervals) between 17 of 66 pairwise wetland combinations. Overall, most gene flow occurred from other wetlands into JCSF and WIND, with JCSF receiving the highest magnitude of asymmetrical gene flow from the greatest number of wetlands. Highly asymmetrical gene flow was detected from KLIP outward to almost every other wetland, and limited dispersal into KLIP. PEHA and TURT both had moderate levels of dispersal outward, with gene flow into both wetlands only from KLIP. It is important to note that gene flow estimates for KEAW are based on few individuals (n < 10); though sampling represents all birds occupying the wetland.

Table 4 Gene flow estimated among Hawaiian gallinules sampled on eight wetlands in O`ahu, based on 12 microsatelite loci
Full size table
Fig. 5
figure 5

Diagram showing direction of prevailing gene flow between wetland sites where Hawaiian gallinules were sampled on O`ahu, Hawai`i, USA. Gene flow estimates with non-overlapping 95% confidence intervals between wetland pairs are shown. Arrow thickness represents the magnitude of gene flow (Nem): small arrows 1.0–2.4, medium arrows, 2.5–5.0, and large arrows > 5.0

Full size image

Discussion

Hawaiian gallinules on O`ahu exhibit moderate to high genetic structure (sensu Wright 1978) among small and geographically close wetlands surrounded by a diverse and increasingly urbanized landscape matrix. These results are particularly noteworthy considering the small distances by which wetland habitats on the island are separated (1.5–55 km) coupled with moderate to high levels of genetic structure observed at two marker types (pairwise msat FST ≤ 0.261; pairwise mtDNA FST ≤ 0.919; Table 3). Genetic drift had limited variation upon which to act in this population, (AR ≤ 2.2; Table 2), so dispersal among remnant patches was likely markedly reduced in order to promote the levels of genetic partitioning observed. Hawaiian gallinules experienced a severe and recent population bottleneck (Engilis and Pratt 1993) and are relatively recent colonizers of Hawai`i (James 1987; Fleischer and McIntosh 2001). We thus hypothesize that the population bottleneck and subsequent isolation of some habitat patches throughout population recovery promoted the rapid development of genetic structure among wetlands through the process of genetic drift.

The magnitude of genetic differentiation observed among wetlands in this study is higher than that observed for highly vagile bird species (e.g., Swainson’s warbler Limnothlypis swainsonii, microsatellite FST = 0.023–0.049, Winker et al. 2000), and more closely resembles values for other wetland specialist birds in habitat fragmented by urbanization (e.g. white-fronted chats Epthianura albifrons, microsatellite FST = 0.035–0.183, Major et al. 2014), highly sedentary tropical birds (e.g. bicolored antbirds Gymnopythis bicolor, microsatellite FST = 0.012–0.278, Brown et al. 2004), and other rail species (Girard et al. 2010; Miller et al. 2012, 2015). Yellow rails, a migratory species, exhibit moderate genetic structure (microsatellite FST = 0.083–0.113) between an isolated population in Oregon, with the rest of the breeding population distributed over much of Eastern North America; no genetic structure was detected within the contiguous distribution (Miller et al. 2012). High genetic structuring (microsatellite FST = 0.19–0.63; distance 100–850 km) was found among nonmigratory Californian populations of black rail (Laterallus jamaicensis coturniculus; Girard et al. 2010). Similarly, the closely related Marianas common moorhen (Gallinula chloropus guami) is also highly structured, with interisland estimates (FST) of 0.152 for microsatellites and 0.390 for the ND2 region of mitochondrial DNA (distance ~200 km), between Saipan and Guam (Miller et al. 2015). Unfortunately no studies of within-island structure for Marianas moorhen or between-island structure for Hawaiian gallinules have been conducted for comparison.

Although inter-population estimates of genetic structure observed here are similar across a range of avian taxa, assessments described above were conducted at a much larger geographic scale relative to our study (distances 100–10,000 km vs 1.5–55 km, respectively). Indeed, the spatial scale of our study is smaller than most reports of notable avian population differentiation (e.g., Nicholls et al. 2006; Roberts et al. 2011; Nogueira et al. 2014; Vidal et al. 2015; Greenberg et al. 2016; Levy et al. 2016), even for sedentary tropical species (Brown et al. 2004; Abalaka et al. 2015). Genetic structure observed in our study is similar in magnitude and spatial scale to that observed by Delaney et al. (2010) in what was described as the highest amount of genetic structure observed over small distances in avian taxa. The small spatial scale at which genetic structure is observed within Hawaiian gallinule of O`ahu highlights two important drivers of genetic structure: low dispersal propensity and the influence of genetic drift as a result of a severe bottleneck event. We hypothesize that the alteration of O`ahu’s landscape and the reduction of habitat for waterbirds (Griffin et al. 1989; van Rees and Reed 2014) has likely played a role in the creation of a mosaic of spatially isolated wetlands enabling genetic drift to quickly act and promote genetic structure among populations in close geographic proximity.

Landscape and behavioral ecologists have challenged the notion that inter-patch Euclidean distances alone are an effective measure of potential population isolation or connectivity, suggesting instead that the type of landscape cover making up that distance (i.e., the landscape matrix) plays a larger role in many terrestrial systems (Ricketts 2001; Rodewald 2003; Revilla and Wiegand 2008; Aben et al. 2012). We observed a small number of populations (in particular KLIP, KEAW, WAIM, and TURT) that show strong and consistent structure that is not correlated with physical distance. The patterns of genetic structure we observed may be explained, at least in part, by the influence of the physical structure of the landscape matrix on Hawaiian gallinule dispersal among wetlands (i.e., functional connectivity; Taylor et al. 1993; Baguette and Van Dyck 2007), the history of O`ahu’s Hawaiian gallinule population, and source-sink dynamics among persisting populations (Pulliam 1988).

The severity of the population decline experienced by Hawaiian gallinule populations in the late 1800s to mid-1900s (gallinules reduced to ~60 individuals across the state of Hawai`i) and subsequent recovery (Shallenberger 1977; Reed et al. 2011), likely erased signatures of genetic structure prior to the decline. Genetic drift acting on remnant populations during the bottleneck and subsequent recolonization likely played a large role in current patterns of genetic structure observed. Hawaiian gallinules on O`ahu likely survived the population bottleneck in a few (2–3) isolated strongholds (as indicated by high pairwise FST estimates) and subsequently colonized unoccupied or newly available habitats (created through restoration or active management) as numbers increased. Wetland size and quality differ widely between contemporary habitats, and may have influenced dispersal rates and population connectivity via habitat effects and source-sink dynamics (Pulliam 1988; Pfluger and Balkenhol 2014). JCNR and HAMA, two wetlands that currently support larger populations of Hawaiian gallinules, are weakly differentiated from most other populations on the island, and are known to have supported few if any birds prior to recovery (Shallenberger 1977). Limited genetic structure coupled with larger census sizes suggests that these now presumably high-quality habitats received a high number of immigrants from other wetland populations during the recovery period (Fig. 5). Spatially isolated wetlands (e.g. KEAW, a wetland located in an urban development surrounded by mountains, and KLIP, located on a fenced military base) exhibited asymmetrical gene flow to most other wetlands on the island, notably those in the WIND and JCSF regions. Prior to the decline, these wetlands (KEAW, KLIP, and TURT) may have been strongholds for Hawaiian gallinules as they are located on restricted military and private properties, and potentially experienced reduced persecution from hunting, which was legal until 1939 (Shallenberger 1977). Wetland habitats in WIND and JCSF supported few Hawaiian gallinules prior to the 1970s and 1980s (Shallenberger 1977; Banko 1987; Silbernagle, USFWS, pers. comm.), after which significant habitat management and predator control efforts began. Once management was enacted and protected habitat became available, WIND and JCSF wetlands apparently received immigrants from throughout the island (see Fig. 5), as large populations developed quickly where birds had not been observed for some time (Shallenberger 1977; Banko 1987).

Hawaiian gallinules suffer mortality near roads (K. Doyle, Hawai`i State Department of Fisheries and Wildlife, pers. comm.), are typically shy of humans, and are rarely seen flying > 2 m off the ground. These factors suggest mechanisms by which anthropogenic landscape change could limit dispersal success through psychological, mortality-based, and physical barriers (Zeller et al. 2012). Population structure in other wetland birds (e.g. white-fronted chat, Major et al. 2014) is driven in part by urbanization around remaining habitats, regardless of the level of protection within these habitats; it may be that observed structure in Hawaiian gallinules follows similar patterns. Though the low connectivity of urban wetlands like KEAW and KLIP to nearby wetlands supports this hypothesis, observed structure between TURT and two adjacent wetlands (WAIM and JCNR) might be a notable exception, given that O`ahu’s North shore is among the island’s least-developed regions and that these wetlands are embedded in a largely rural matrix. However, there are other possible barriers to dispersal that might apply within O`ahu’s North shore, such as fence lines and dense scrub-forest. Linear features like roads can increase mortality and act as barriers to movement in many taxa (Andrews 1990; Vanak et al. 2010), including cursorial birds occupying open habitats (e.g. Tympanuchus pallidicinctus, Wolfe et al. 2007). In contrast, connectivity between spatially distant populations can be increased by corridors, or other landscape features that encourage or facilitate movement (Baum et al. 2004; Tewksbury et al. 2002). In the case of Hawaiian gallinules, birds use streams as breeding and foraging habitat (Banko 1987), and it is suspected that they may disperse more readily along stream margins and drainage canals (M. Silbernagle, USFWS, pers. comm). Finally, steep topography and mountain ranges may physically or psychologically (e.g. behavioral inhibition; sensu Harris and Reed 2002) impede dispersal. The Ko`olau mountain range, for example, separates the windward wetlands from PEHA, KEAW, and wetlands on the leeward side of the North shore (Fig. 1). These additional possible matrix effects on dispersal rates may also play a role in the observed structure among wetlands that are not located in urbanized areas. Though perhaps not of direct conservation concern, the influence of natural barriers on gene flow may be exaggerated by more recent modifications to the landscape. Given the diversity of potential landscape-relevant drivers of genetic structure in Hawaiian gallinules, formal analysis of patterns observed in this study using a landscape genetic framework (Manel et al. 2003; Storfer et al. 2010) should be considered an important next step in research on this threatened population.

Conservation implications

Our results illustrate the development of genetic structure in a taxon that is expanding its range after severe population decline, showing that biologically significant structuring can occur over small geographic distances. This raises an important, management relevant possibility—that populations recovering in altered landscapes may experience isolation or poor connectivity after recruitment. Given the relationship between fragmentation and extinction risk (Saunders et al. 1991; Fahrig 2003), this genetic structuring could be of conservation concern. For example, the recovery plan for the Great Lakes population of the Piping plover Charadrius melodus in the U.S. has as one of its delisting criteria the maintenance of genetic diversity sufficient for population persistence (USFWS 2003). If genetic structuring of this kind can be attributed to landscape characteristics, methods that increase connectivity in the context of altered landscapes (e.g. translocations, Wright et al. 2014; corridors, Tewksbury et al. 2002; stepping stone habitats; Saura et al. 2014) may be useful in ameliorating the potential demographic and genetic effects of fragmentation. The influx of gene flow we observed for James Campbell National Wildlife Refuge and Hamakua Marsh wildlife sanctuary suggests that restored and managed wetlands can be rapidly recolonized by Hawaiian gallinules. This suggests that wetland restoration and management may be effective tools for the management of this subspecies.