Disrupted fine-scale population processes in fragmented landscapes despite large-scale genetic connectivity for a widespread and common cooperative breeder: the superb fairy-wren (Malurus cyaneus)

Authors


Correspondence author. E-mail: katherine.harrisson@gmail.com

Summary

  1. Understanding how habitat fragmentation affects population processes (e.g. dispersal) at different spatial scales is of critical importance to conservation. We assessed the effects of habitat fragmentation on dispersal and regional and fine-scale population structure in a currently widespread and common cooperatively breeding bird species found across south-eastern Australia, the superb fairy-wren Malurus cyaneus.
  2. Despite its relative abundance and classification as an urban tolerant species, the superb fairy-wren has declined disproportionately from low tree-cover agricultural landscapes across the Box-Ironbark region of north-central Victoria, Australia. Loss of genetic connectivity and disruption to its complex social system may be associated with the decline of this species from apparently suitable habitat in landscapes with low levels of tree cover.
  3. To assess whether reduced structural connectivity has had negative consequences for genetic connectivity in the superb fairy-wren, we used a landscape-scale approach to compare patterns of genetic diversity and gene flow at large (landscape/regional) and fine (site-level) spatial scales. In addition, using genetic distances, for each sex, we tested landscape models of decreased dispersal through treeless areas (isolation-by-resistance) while controlling for the effect of isolation-by-distance.
  4. Landscape models indicated that larger-scale gene flow across the Box-Ironbark region was constrained by distance rather than by lack of structural connectivity. Nonetheless, a pattern of isolation-by-resistance for males (the less-dispersive sex) and lower genetic diversity and higher genetic similarity within sites in low-cover fragmented landscapes indicated disruption to fine-scale gene flow mechanisms and/or mating systems.
  5. Although loss of structural connectivity did not appear to impede gene flow at larger spatial scales, fragmentation appeared to affect fine-scale population processes (e.g. local gene flow mechanisms and/or mating systems) adversely and may contribute to the decline of superb fairy-wrens in fragmented landscapes in the Box-Ironbark region.

Introduction

Human alteration of landscapes has had profound effects on processes operating within populations. Potentially disrupted processes may include dispersal, mechanisms of communication (e.g. song-sharing; Pavlova et al. 2012), kin interactions and mating and social systems (Banks et al. 2007). Dispersal–the movement of individuals from one location to another–is a fundamental process of particular relevance in fragmented landscapes, where populations may exist as spatially isolated subunits. In these landscapes, dispersal acts to link subpopulations together in a metapopulation (Hansson 1991). Dispersal influences overall persistence of metapopulations through both demographic connectivity [colonization and extinction rates (Hanski & Gilpin 1997) and source-sink dynamics (Pulliam 1988)] and genetic connectivity (e.g. gene flow; Lowe & Allendorf 2010). Where subpopulations are small, immigration (dispersal into a patch) can improve chances of local persistence by mitigating the interacting effects of demographic, environmental and genetic stochasticity (Caughley 1994; Spielman, Brook & Frankham 2004). Understanding how large-scale structural changes to landscapes affect within- and among-population processes (including dispersal) is therefore of great importance to conservation management (Banks et al. 2007; Sunnucks 2011).

Although deterministic factors (e.g. habitat loss) and demographic and environmental stochasticity can contribute to population decline, these rarely operate in the absence of genetic effects (Spielman, Brook & Frankham 2004; Sunnucks 2011). Small, genetically isolated populations are likely to be negatively affected by inbreeding depression, loss of genetic variation through genetic drift and loss of evolutionary potential (Frankham 1996; Spielman, Brook & Frankham 2004). These effects are potentially reversible if genetic connectivity can be re-established via dispersal into isolated patches. Here, we distinguish genetic connectivity from dispersal, as the latter does not always equate to gene flow, as migrants are often characterized by lower reproductive success (e.g. because of poor body condition; Hansson, Bensch & Hasselquist 2004; Coulon et al. 2010). Genetically effective dispersal and the maintenance of natural levels of genetic variation are therefore critical for the persistence of wildlife populations in fragmented landscapes (Palstra & Ruzzante 2008).

Many fragmentation studies involve small populations of threatened taxa, but studies of nonendangered species often have advantages in terms of sample sizes, replication and breadth of factors and scenarios that can be examined (Honnay & Jacquemyn 2007; Lancaster et al. 2011). Here, we investigate genetic connectivity of a well-studied, common and widespread south-eastern Australian cooperatively breeding passerine, the superb fairy-wren Malurus cyaneus, across fragmented agricultural landscapes in the Box-Ironbark region of north-central Victoria, Australia.

Superb fairy-wrens are small (8–11 g), sedentary birds that live in cooperatively breeding groups consisting of a dominant breeding pair and typically 1–4 male subordinate helpers (Mulder 1997; Double et al. 2005). Family groups usually occupy small (c. 1·3–2·4 ha) territories and typically occur in various types of woodland and human-altered habitats (e.g. urban gardens, edge habitats), wherever there is a dense, shrubby understorey (Tidemann 1990; Higgins, Peter & Steele 2001). Dispersal is strongly female biased: during the first year of life, the majority of immature females disperse permanently over four or more territories (> 500 m), whereas the majority of males remain on natal territory or move to adjacent ones to fill vacancies in reproductive queues (Mulder 1995; Cockburn et al. 2003, 2008, 2009; Double et al. 2005). Superb fairy-wrens are renowned for extremely high levels of female-mediated extra-pair (and extra-group) fertilization (Mulder et al. 1994; Mulder 1997; Double & Cockburn 2000; Cockburn et al. 2003). Breeding females make pre-dawn forays across up to six territories and obtain extra-group fertilizations (Cockburn et al. 2008). Therefore, females mediate gene flow at two spatial scales: (i) relatively large-scale (landscape/regional, > 1 km) via permanent dispersal and (ii) relatively fine-scale (site-level, < 1 km) via extra-pair and extra-group fertilization.

Persistence of the superb fairy-wren in urban areas implies some inherent resilience to habitat alteration (Trollope, White & Cooke 2009), although effects of habitat fragmentation on population processes can be complex and difficult to predict (Van Houtan et al. 2007; Callens et al. 2011). The superb fairy-wren has declined disproportionately with decreasing landscape-level extent of tree cover in agricultural landscapes in north-central Victoria, disappearing from apparently suitable habitat in landscapes with low levels (< 18%) of native vegetation cover (Radford, Bennett & Cheers 2005; Amos et al. 2012). As a sedentary species and relatively weak flier, the superb fairy-wren may be reluctant to cross large gaps (e.g. in the order of 100 m) between habitat patches embedded in a matrix of cleared farmland, as observed for several other Australian woodland bird species (Brooker, Brooker & Cale 1999; Robertson & Radford 2009; Doerr, Doerr & Davies 2011). The complex social systems of cooperatively breeding species such as the superb fairy-wren may leave them particularly vulnerable to changes in landscape structure, if key population processes (e.g. dispersal, gene flow and mating systems) at large and/or fine spatial scales are disrupted (Blackmore, Peakall & Heinsohn 2011).

Integrating a landscape-scale study design and a suite of molecular approaches, our study examines the effects of habitat fragmentation on population processes. Given disruption of gene flow mechanisms can occur at both large (landscape/regional) and small (site-level) spatial scales, we used a combination of allele frequency-based and genotype-based genetic analyses (including landscape models of gene flow) to understand population processes better on different spatial and temporal scales (Sunnucks 2011). Our aims were as follows: (i) to compare patterns of genetic diversity and genetic connectivity among twelve 100-km2 landscapes of different total tree cover; (ii) to use landscape genetic models to examine whether reduced dispersal across a treeless agricultural matrix could contribute to the absence of the superb fairy-wren from apparently suitable habitat patches in low-cover landscapes; (iii) to test for genetic consequences of reduced genetic connectivity in fragmented landscapes and (iv) to explore population subdivision at the regional scale.

Materials and methods

Study Region and Sampling

The Box-Ironbark region in north-central Victoria, Australia, extends over c. 20 500 km2 of the inland slopes of the Great Dividing Range (Radford, Bennett & Cheers 2005; Radford & Bennett 2007). Since colonization of Australia by non-Indigenous people, the Box-Ironbark forests and woodlands have been heavily cleared and fragmented, primarily for timber, mining and agriculture (ECC 2001). Remnant native vegetation occurs typically on low-fertility soils and is embedded within an agricultural matrix (ECC 2001).

Superb fairy-wrens were sampled from twelve 100-km2 (10 by 10 km) landscapes within the study region (Fig. 1). Given earlier work suggesting the extent of tree cover in landscapes in the Box-Ironbark region was a good predictor of the likelihood of occurrence of superb fairy-wrens (Radford, Bennett & Cheers 2005; Radford & Bennett 2007; Amos et al. 2012), we used tree cover as an approximate measure of suitable superb fairy-wren habitat/structural connectivity. While superb fairy-wrens generally require the presence of a low shrub layer, spatially explicit models of shrub-cover do not exist over the study region. Nevertheless, the presence of tree cover can be sufficiently correlated with that of shrubby vegetation, because treeless areas in the Box-Ironbark region are mainly cleared farmland, lacking a shrub layer. Fragmented landscapes varied in extent of tree cover: we denote landscapes with 11–20% total tree cover as low-cover fragmented, and landscapes with 25–45% total tree cover as high-cover fragmented (Fig. 1; Table 1). Three landscapes with the highest extent of continuous tree cover remaining in the region were used as reference landscapes (> 70% tree cover; Fig. 1). This design has been used to address conservation management of a suite of bird species in the study region (Amos et al. 2012; Harrisson et al. 2012). While superb fairy-wrens are not likely to be distributed evenly across all available tree cover within the study landscapes (i.e. there is likely to be a bias towards occurence along watercourses and patch edges), this is unlikely to affect major inferences of our study, as we did not observe a sampling bias in low-cover landscapes (where there are more edges), and riparian vegetation was distributed more-or-less evenly across landscapes.

Figure 1.

Location of study landscapes in north-central Victoria, Australia. Location of the study region within Victoria is shown in inset. Grey shading represents tree cover. Boxes define 10 km by 10 km landscapes. Sites within landscapes are marked. Site codes refer to the landscape and site number. Landscape codes (with extent of total tree cover in brackets) are: St – Stuart Mill (19%), Tu – Tunstalls (20%), We – Wehla (11%), Gl – Glenalbyn (18%), Du – Dunolly (78%), Ha – Havelock (45%), Sh – Shelbourne (12%), Ax – Axe Creek (35%), Cr – Crosbie (26%), Re – Redcastle (72%), Ru – Rushworth (78%) and Mu – Murchison (27%).

Table 1. Landscape-based statistics (for landscapes where N ≥ 8) for superb fairy-wrens: landscape extent of tree cover (%Tree), number of male and female individuals (#M-F), mean allelic richness (AR), observed (Ho) and expected (UHe) heterozygosity, mean number of unique alleles (#UA), FIS and number of first-generation migrants detected (#1M)
LandscapeTree (%)#M–FARHoUHe#UAFIS#1M
  1. a

    Indicates one individual of unknown sex.

Wehla115–35·70·680·740·090·090
Shelbourne121–0
Tunstalls207–54·90·740·740−0·0011
Glenalbyn1822–135·60·700·7400·053
Stuart Mill195–3a4·90·810·760−0·070
Crosbie266–76·00·710·7400·050
Murchison2712–146·40·700·750·360·071
Axe Creek3518–23a6·50·730·760·730·044
Havelock4510–115·60·680·720·360·063
Redcastle7212–116·20·720·730·360·023
Rushworth7811–86·30·750·740·09−0·011
Dunolly781–2

Variance in male reproductive success as a result of female-mediated extra-group copulation has been shown to contribute to local genetic structure of superb fairy-wrens (Double et al. 2005) and has the potential to confound conclusions about the effects of habitat fragmentation on spatial genetic patterns. To address this, sampling sites within landscapes were spread over intervals longer than female extra-territorial forays (which are usually < 1 km, Double & Cockburn 2000) to ensure that the genetic influence of any individual male did not extend across multiple sample sites. Four to seven sampling sites per landscape were visited twice, c. 6 months apart, between November 2007 and February 2010; from one to five sites in each of 12 landscapes yielded superb fairy-wren samples (Fig. 1).

Superb fairy-wrens were caught in mist-nets, measured, banded, aged as juvenile, immature or adult (according to Rogers, Rogers & Rogers 1986), sampled and released. Blood samples for genetic analysis were taken from the brachial vein and stored in ethanol (at −20 °C upon return from field).

Genotyping and Molecular Sexing

Individuals were screened by polymerase chain reaction (PCR) for 12 putatively selectively neutral nuclear markers [11 microsatellites and 1 Exon-Primed-Intron-Crossing (EPIC) region; Tables S1–S2, Supporting information) and sexed using a standard molecular protocol. Extraction protocol, primer sequences, PCR conditions and protocols are provided in Supporting information. PCR products were screened using Li-Cor 4200 and 4300 Global IR2 two-dye DNA sequencers.

Establishment of Appropriate Genetic Marker Behaviour

A total of 252 superb fairy-wrens were genotyped for a minimum of 9 loci (< 0·001% data were missing). Tests for deviations from Hardy–Weinberg and linkage equilibria were performed using GENEPOP 4.0 (Rousset 2008) for all locus/site combinations. The presence of null alleles was checked by testing for consistent departure from Hardy–Weinberg equilibrium (in the direction of homozygous excess) across multiple sites and landscapes and by looking for individuals that failed to amplify at a locus in multiplex reactions that were otherwise successful (i.e. putative null homozygotes).

Parentage Analysis to Remove Potential Offspring of Sampled Parents

Including offspring of sampled parents in analyses can bias analyses that assume a random sample of unrelated individuals. We performed per-landscape parentage analyses in CERVUS 3.0 (Kalinowski, Taper & Marshall 2007) to identify and remove potential progeny from all analyses (details in Supporting information).

Regional- and Landscape-Scale Genetic Structure

To assess the extent of population subdivision at regional and landscape scales, an individual-based Bayesian spatial algorithm implemented in TESS 2.3.1 (Chen et al. 2007) was used with all individuals across the study region (excluding progeny) and individual landscapes (with more than ten individuals), respectively. TESS was run using the CAR admixture model (100 iterations of 106 sweeps, discarding the first 30 000) with spatial interaction parameter set to 0·6, and the number of genetic clusters (K) set from 2 to 12 for the study region and 2 to 5 for individual landscapes. The point of greatest change (or elbow) in the plot of the mean DIC value across 100 runs against K determined the most likely value of K. Cluster probabilities were averaged for the ten runs with the lowest DIC values for the most likely K value using the Greedy algorithm option with 1000 random input orders in clumpp 1.1.2 (Jakobsson & Rosenberg 2007). Results were visualized using DISTRUCT 1.1 (Rosenberg 2004).

Testing for Qualitative Time-Scale of Population Divergence

To determine whether observed genetic structure across the study region was shaped over ancient or more recent timescales, pairwise allele size permutation tests were performed among four observed geographical clusters (based on TESS analysis) using SPAGeDi 1.3 (Hardy & Vekemans 2002). Allele sizes were randomly reassigned among alleles, with 10 000 permutations. If allele evolutionary relationships as inferred from allele sizes are informative with respect to population structure (i.e. the mutation process and allele-frequency changes lead to consistent patterns of allele size difference between identified TESS clusters and thus contribute to observed differentiation), observed pairwise RST values should be significantly higher than values permuted (ρ RST). If observed population structure is the result of drift alone (i.e. characterized by differences in allele state rather than size), RST values should not differ significantly from values permuted (ρ RST).

Hierarchical amova was performed in Arlequin 3.0 (Excoffier, Laval & Schneider 2005) to partition genetic variation into amounts explained by four large geographic regions identified by TESS analysis, variance among landscapes within regions, and variance within landscapes (that includes variance within and among sites). A large amount of genetic variance explained by regions would suggest strong genetic differentiation between regions, possibly reflecting long-term isolation.

Migration and Gene Flow among Regions and Sites

The amount of recent migration (one to three generations) among four geographic regions detected by TESS was tested using BayesAss 1.3 (Wilson & Rannala 2003). This method relaxes the assumption of genetic equilibrium and symmetric migration rates necessary for estimation of historic gene flow (FST). BayesAss was run for 10 000 000 iterations with a 2 000 000 burn-in and sampling frequency of 2000. For the final analysis, the delta value (which defines the maximum amount a parameter can be changed each iteration) for allele frequency, migration rate and inbreeding rate were set to 0·1, 0·15 and 0·1, resulting in accepted changes of 47, 46 and 42%, respectively; the increase in log probabilities reached a plateau after c. 12 000 iterations.

Potential first-generation migrants at sites with ≥ 5 individuals were identified using GENECLASS2 (Piry et al. 2004). We used the Bayesian method of Rannala & Mountain (1997) with Monte-Carlo resampling algorithm (Paetkau et al. 2004) to determine the likelihood that the sampling site of an individual was its population of origin (Lhome). The test used 10 000 simulated individuals with a type I error set to 0·01.

Reduced connectivity and subsequent genetic drift in small, isolated populations are expected to result in increased genetic differentiation between isolated sites. Two allele-frequency-based measures of genetic differentiation (reflecting patterns of gene flow on longer temporal scales than individual, genotype-based analyses) with different strengths were computed for all pairs of sites with ≥ 5 individuals across the study region: FST in GENEPOP 4.0 (Rousset 2008) and Jost's D, adjusted for small sample sizes (Dest), in the DEMEtics package in R (Gerlach et al. 2010). Jost's D is thought to accumulate faster than FST following disruption of gene flow (Landguth et al. 2010) and is more comparable across data sets with different heterozygosity than is FST (Jost 2008).

Testing Landscape Models of Gene Flow

Modelling of gene flow across the study region using ‘resistance’ surfaces (isolation-by-resistance; McRae 2006), which assume differential permeability of different landscape context to a moving individual, is a powerful approach for investigating the relationship between structural and functional connectivity (McRae & Beier 2007). Pairwise genetic distances among individuals can be used to test such models while controlling for the effect of isolation-by-distance, expected for organisms with restricted dispersal. Amos et al. (2012) classified every 25 × 25 m cell in our study region as either ‘treed’ or ‘treeless’ and built the models of isolation-by-distance (IBD), which assumed equal resistance of treed and treeless cells, and isolation-by-resistance (IBR), which assumed that treeless cells had twice the resistance (harder for an individual bird to cross) value of treed cells using CIRCUITSCAPE 3.5.1 (McRae & Shah 2008; McRae et al. 2008). These two models yielded resistance surfaces, from which pairwise resistance distances for each pair of sites were calculated (methods in Amos et al. 2012). Results of TESS analysis suggested a potentially historical genetic break between the east and west of the study region (Fig. 2). To account for any biological differences between fairy-wrens in the east and west that might have arisen as a result of longer-term isolation, we tested models for east and west separately. Because male superb fairy-wrens have much shorter dispersal distances than do females (above), we tested the models using genetic distances calculated separately for each sex. Mantel tests were used to test for correlations between pairwise site-based resistance (IBD and IBR) and pairwise individual-based genetic distance matrices (Smouse & Peakall 1999). Support for IBR above isolation-by-distance was determined using partial Mantel tests: an IBR model was considered to fit better than IBD if it explained significant amount of genetic variance when IBD was partialled out.

Figure 2.

Genetic structure of superb fairy-wrens displayed at the scale of (a) the study region and (b) individual landscapes. Each column on the x-axis is representative of an individual fairy-wren and the y-axis represents the proportional assignment (Q) of each individual to each identified genetic cluster based on TESS analysis. For the study region plot (a) cluster labels correspond to the Stuart Mill/Tunstalls block (StTu block), the Glenalybn/Wehla block (GlWe block), the Dunolly/Havelock block (DuHa block) and the eastern block (East block). The dotted line indicates the east–west division applied in habitat resistance modelling. Landscape abbreviations are as for Fig 1. Sites are arranged from west to east.

Fine-Scale Population Structure Within Landscapes Inferred from Genetic Data

Habitat fragmentation can alter social structure and patterns of genetic similarity among individuals within isolated sites (Banks et al. 2007). We used individual-based spatial autocorrelation analysis in GENALEX 6.41 (Peakall & Smouse 2006) to compare fine-scale spatial genetic patterns among low-cover, high-cover and reference landscapes. The sizes of distance class bins were 0–1 km (reflecting within-site similarity), > 1–5 km, > 5–10 and > 10–15 km (reflecting maximum distance class between two sites within a landscape). Tests were performed separately for males and females, because male superb fairy-wrens are more sedentary and thus more genotypically structured than are females (Double et al. 2005). We used 999 random permutations to estimate the expected range of genetic similarity under random spatial association of individuals; 95% confidence intervals around r were estimated using 999 bootstraps (Peakall, Ruibal & Lindenmayer 2003). The rare cases of individuals with data missing for a locus (< 0·001% data were missing) were assigned the most common landscape genotype for that locus.

Population Consequences of Landscape-Level Habitat Fragmentation Inferred from Genetic Data

If gene flow is reduced due to habitat alteration, the more impacted (low tree cover) landscapes should display fewer immigrants, fewer unique alleles, loss of genetic diversity (allelic diversity and heterozygosity), and elevated relatedness. For each landscape with N ≥ 8, mean allelic richness (AR) across loci was calculated in FSTAT 2.9.3 (Goudet 2001), and observed heterozygosity (Ho), unbiased expected heterozygosity (UHe) and mean number of unique alleles were calculated in GENALEX 6.4.1 (Peakall & Smouse 2006). To explore patterns of genetic diversity across the study region, single factor analysis of variance (anovas) were performed to test for an effect of landscape-level tree cover (low-cover, high-cover and reference) on landscape-based statistics: AR, Ho, UHe, proportion of migrants and number of unique alleles in R 2.10.1 (R Development Core Team 2009). In addition, linear regressions of landscape-based statistics against landscape-level tree cover were performed. Longitudinal biogeographical gradients (potentially corresponding to gradients in environmental factors including rainfall, climate and soil fertility) across the Box-Ironbark region have been observed in a number of woodland bird species (Radford & Bennett 2007), so we also performed regressions of landscape-based statistics against longitude.

Results

Establishment of Appropriate Genetic Marker Behaviour

Of 12 loci screened, only Smm3 showed strong evidence of null alleles, with 13 individuals failing to amplify in multiplex reactions. High likelihood of null alleles for locus Smm3 was reflected in consistent significant homozygote excess across site-samples, when most other loci were not affected, and it was therefore excluded from all analyses. Ten of the 209 tests for Hardy–Weinberg equilibrium showed significant homozygous excess, but there was no consistent pattern in relation to any locus or sampling location. There was no evidence of consistent linkage disequilibrium between any loci across site samples and so independence of loci was assumed.

Regional Population Structure Inferred from Genetic Data

Strong geographic structure across the study region was evident from genotypic data (TESS analysis), with genotypic clusters generally corresponding to four spatial regions: three in the west (Stuart Mill/Tunstalls – StTu block), north-west (Glenalbyn/Wehla – GlWe block), south-west (Dunolly/Havelock – DuHa block) and one in the east (East block, Fig 2a). The most likely number of clusters was five, but as no individuals were most strongly assigned to the fifth cluster, for simplicity four clusters are presented (Fig. 2; Figs S1 and S2, Supporting information). The four geographical groups were supported by significant allele frequency differentiation among them (FST = 0·035–0·05; Table S5, Supporting information). Population differentiation as estimated by FST was relatively low among the three western spatial regions compared to between these and the eastern block (Table S5).

Testing for Time-Scale of Population Divergence

Observed pairwise RST values were not significantly higher than permuted RST values among identified genotypic clusters of landscapes (Table S5), suggesting that the predominant force shaping observed genetic differentiation was genetic drift on recent timescales rather than mutational processes over the timescales on which microsatellites evolve suites of alleles. Recent divergence as a result of drift was supported by amova results, with only 2·6% (FST = 0·05, < 0·001) of total genetic variance explained by membership of the four regional blocks and 2·4% (FSC = 0·03, < 0·001) by membership to landscapes within regions, whereas 94·9% (FCT = 0·03, < 0·001) was explained by genetic variation within landscapes (which also encapsulates within-site variance).

Recent Migration and Gene Flow Among Sites and Regions

BayesAss non-equilibrium estimates of recent migration among the four geographic regions (based on TESS analysis) ranged from very low (0·3%, confidence interval 0–1·7% from DuHa to GlWe) to moderate (9·9%, CI 0·1–31% from East block to StTu block; Table S6, Supporting information), suggesting that some gene flow occurs among regions each generation.

GENECLASS analysis identified 16 likely first-generation migrants: 11 were adult females (the dispersing sex), one was a female of unknown age, one was an immature female, and three were adult males (Table S7, Supporting information). This strong sex-bias in the direction expected from knowledge of the species ('Introduction') lends credence to the identification of dispersers in this analysis. Detection of first-generation migrants was not limited to sites within contiguous habitat blocks. Three migrants were identified in isolated sites (Gl5, Gl6 and Tu3, with < 6% vegetation within a 500 m radius of site).

Pairwise FST and Dest values between sites ranged from 0 to 0·28 and from 0 to 0·53, respectively, across all comparisons and were generally higher for sites from different genotypically defined regions (based on TESS analysis) than for sites within the same region (pairwise FST and Dest matrices provided in Table S8, Supporting information). Mean pairwise FST and Dest values between sites were significantly higher in fragmented (mean FST = 0·08, mean Dest = 0·21) than reference (mean FST = 0·02, mean Dest = 0·05) landscapes (FST: = 3·35, d.f. = 8·6, = 0·009, Dest: = 3·17, d.f. = 9·85, = 0·01). All seven within-landscape pairwise comparisons of FST values for fragmented landscapes were significant. In contrast, only a single pair of sites in seven comparisons in the reference landscapes showed significant differentiation. Dest showed similar patterns: six of seven comparisons within fragmented landscapes showed significant genetic differentiation, compared to two within reference landscapes.

Testing Isolation-by-Distance and Isolation-by-Resistance Landscape Models of Gene Flow

Genetic distances for adult males were consistent with isolation-by-distance (IBD; Mantel r = 0·323, P = 0·001) and isolation-by-resistance (IBR; Mantel r = 0·352, P = 0·001), and IBR explained variance above that explained by IBD (Partial Mantel r = 0·153, P = 0·003; Table 2). For adult female fairy-wrens (the more dispersive sex), genetic distances were best explained by IBD (Mantel r = 0·212, P = 0·001); IBR was significant on marginal test (Mantel r = 0·193, P = 0·002) but not on partial test (Partial Mantel r = −0·04, P = 0·64; Table 2), indicating that the isolation-by-resistance model did not explain genetic variance beyond that explained by the null model of isolation-by-distance. These results provided evidence of reduced dispersal of males through treeless patches, whereas for females, isolation-by-distance could not be rejected in favour of isolation-by-resistance. Mantel results were consistent across the east and west of the study region (Table 2; supported by genetic distances in Table S5), suggesting that fairy-wrens in both areas respond similarly to habitat fragmentation. Similar results are obtained if eastern and western areas are analysed as one (data not shown).

Table 2. Summary of IBD (isolation-by-distance) and IBR (isolation-by-resistance) landscape models of gene flow tested for adult male (N = 54, 41) and adult female (N = 55, 26) superb fairy-wrens in the east and west of the study region, respectively. Mantel tests are correlations between pairwise site-based resistance (based on CIRCUITSCAPE models) and pairwise individual-based genetic distance matrices. Partial Mantel tests show partial correlations when effects of the alternative model are partialled out. Significant Mantel r values are in bold font
RegionSexLandscape modelMantel testPartial Mantel tests
IBD partialled outIBR partialled out
r P r P r P
WestFemalesIBD 0·212 0·0010·0980·15
IBR 0·193 0·002−0·040·64
MalesIBD 0·323 0·001−0·040·70
IBR 0·352 0·001 0·153 0·03
EastFemalesIBD 0·105 0·001 0·0420·297
IBR 0·097 0·002−0·0120·55
MalesIBD 0·111 0·001 0·2110·995
IBR 0·182 0·001 0·255 0·002

Fine-Scale Population Structure within Landscapes Inferred from Genetic Data

Within individual landscapes, TESS identified more than one genetic cluster for four out of five of the fragmented landscapes analysed (Gl, Mu, Cr, Ha), compared to only a single cluster within each of the two continuous-cover reference landscapes (not shown) and the combined reference block (Re-Ru, Fig. 2b). The exception among fragmented landscapes was Axe Creek (35% tree cover), where all but a single individual were assigned to one cluster.

Both sexes showed evidence of altered fine-scale genetic structure in low-cover fragmented landscapes: genetic similarity among males and among females at the site-scale (0–1 km) was consistently highest in the low-cover fragmented landscapes, lower in the high-cover fragmented landscapes and lowest in the reference landscapes (Fig. 3, details in Tables S9–S11, Supporting information). This difference was more pronounced in males, where genetic similarity was significantly higher among individuals in low-cover compared to high-cover fragmented (T= 7·0, = 0·007) and to reference landscapes (T2 = 6·5, = 0·01; Fig. 3). Even for females, which are more mobile and less genetically structured (Double et al. 2005), genetic similarity was significantly higher in low-cover compared to reference landscapes (T= 5·05, = 0·01; Fig. 3).

Figure 3.

Correlograms showing differences among landscapes with different levels of tree cover in the genetic correlation coefficient (r) between pairs of individuals as a function of distance class (0–1 km, 1–5 km, 5–10 km, 10–15 km) for male and female superb fairy-wrens separately. Symbols represent low-cover fragmented (black squares), high-cover fragmented (dark grey triangles) and reference (light grey circles) landscapes. Sample sizes were: low = 24, 39, high = 55, 47, reference = 21, 23, for females and males respectively. The 95% confidence error bars for each r value were estimated with 999 bootstraps. *Significant difference (< 0·05) between landscape-types.

Lower Allelic Richness with Less Tree Cover and in a Westerly Direction

There was a significant relationship between landscape tree cover classification (e.g. low-cover, high-cover or reference) and allelic richness within landscapes (= 6·6, d.f. = 2, = 0·02). This was driven by significantly lower allelic richness in the lowest cover landscapes (Tukeys test: = −3·174, = 0·037). However, allelic richness also showed a strong positive correlation with longitude (R2 = 0·76, = 0·001), decreasing east-to-west across the study region. Neither longitude nor landscape-level tree cover was a good predictor of observed or expected heterozygosity, number of migrants or number of private alleles (all tests > 0·1), although low-cover landscapes typically lacked unique alleles (unlike high-cover and reference landscapes).

Discussion

Altered Fine-Scale Population Processes in Fragmented Landscapes Despite Large-Scale Genetic Connectivity

Lack of support for isolation-by-resistance above isolation-by-distance for adult female superb fairy-wrens (the dispersive sex) indicated that genetic connectivity across the fragmented Box-Ironbark region was constrained mainly by distance rather than structural connectivity (physical continuity of habitat). High levels of gene flow across fragmented habitat have been reported in earlier studies of some bird species (Veit et al. 2005; Mylecraine et al. 2008), including in our study system (Harrisson et al. 2012; Pavlova et al. 2012). This is consistent with findings that species persisting in habitat fragments may be those who actually show increased dispersal distances as a consequence of fragmentation, or are tolerant of the habitat matrix (Van Houtan et al. 2007; Coulon et al. 2010). Although there was little apparent association between gene flow and structural connectivity at large spatial scales, with female fairy-wrens still apparently able to undertake obligate long-distance dispersal, our results suggest that habitat fragmentation may have adverse consequences for population processes at finer spatial scales. Disruption to local processes in another woodland bird species (grey shrike thrush Colluricincla harmonica) was shown by Pavlova et al. (2012), who found reduced acoustic connectivity through treeless habitat across the Box-Ironbark region. In our study, disruption of fine-scale processes (e.g. local gene flow mechanisms and/or mating systems) in landscapes with low levels of tree cover was indicated by reduced dispersal of philopatric males through treeless habitat (isolation-by-resistance) and by higher genetic similarity and lower genetic diversity within sites.

Models of decreased mobility through treeless compared with treed areas best described genetic patterns for male superb fairy-wrens, indicating that dispersal patterns of males may be altered by biotic and/or abiotic changes associated with habitat loss and fragmentation. Long-distance dispersal events by philopatric male fairy-wrens are thought to be rare, and usually males will disperse only to fill breeding vacancies in neighbouring territories (Pruett-Jones & Lewis 1990; Double et al. 2005; Cockburn et al. 2008, 2009). In fragmented habitat, movement of males may become even rarer, with decreased habitat translating into reduced availability of neighbouring territories and reduced ability to exploit gaps in reproductive queues in territories that may occur across the agricultural matrix. Similar sex-limited patterns may apply in disrupted habitat for the habitat specialist yellow-throated scrubwren Sericornis citreogularis, where models including habitat resistance were significant for only males (Shanahan, Possingham & Riginos 2011), although in that study the full analysis was not conducted on each sex owing to small sample size.

Higher genetic similarity among individuals within sites in low-cover fragmented landscapes, relative to high-cover fragmented and reference landscapes may reflect disruption of local gene flow mechanisms and/or mating systems as populations contract in isolated habitat patches. Increased genetic similarity in sites in low-cover landscapes was seen in both sexes, but most pronounced among male superb fairy-wrens, the more philopatric sex. Because dispersal by male superb fairy-wrens is infrequent (particularly when mate and territory availability are limiting), female superb fairy-wrens are thought to mediate gene flow at small spatial scales via extra-group copulation, introducing new alleles into social groups away from their natal area (< 1 km; Pruett-Jones & Lewis 1990; Mulder 1995; Cockburn et al. 2003; Double et al. 2005). During breeding, frequent, short-distance movements of female superb fairy-wrens to pursue extra-group matings should reduce the extent of local genetic similarity among males below that expected to result from strong male philopatry (Double & Cockburn 2000). Substantially higher genetic similarity in fragmented than continuous-cover landscapes may signal disruption of complex social and mating systems, and reflect a reduction in both infrequent male dispersal and in the amount of novel genetic variation being introduced into groups via short-distance female-mediated extra-group copulation.

The absence of spatial autocorrelation within continuous-cover reference landscapes for female superb fairy-wrens was consistent with the findings of Double et al. (2005), who detected positive genetic structure only among male superb fairy-wrens (separated by small geographic distances, < 1 km) in continuous habitat within the Australian National Botanic Gardens, Canberra. Our detection of elevated genetic similarity among female superb fairy-wrens in sites in low-cover landscapes compared to our reference landscapes (like the continuous-cover area of Double et al. 2005) may be indicative of disrupted fine-scale population processes.

While patterns of decreasing allelic richness with tree cover were unavoidably confounded by longitude, fairy-wrens in the lowest cover landscapes had significantly lower genetic diversity compared with high-cover and reference landscapes, and typically lacked unique alleles. Reduced allelic richness and stronger genetic differentiation (higher FST and Dest, stronger genetic subdivision detected by TESS) in low-cover fragmented landscapes may reflect lower effective population sizes resulting from reduced availability of habitat, rather than fragmentation per se. Nonetheless, enhanced effects of genetic drift in small populations would mean that gene flow across fragmented habitat may still be occurring below the rate that would be required to counteract the loss of genetic variation. Delaney, Riley & Fisher (2010) observed a similar effect of fragmentation on genetic diversity in urban populations of the wrentit Chamaea fasciata, a similarly sized territorial bird, with typically short dispersal distances (c. 400 m; Baker, Nur & Geupel 1995). Erosion of genetic variation in fragmented habitat, as a result of smaller population sizes, the enhanced effects of genetic drift and the disruption of local gene flow mechanisms and/or mating systems, ultimately leaves populations more susceptible to inbreeding depression and loss of evolutionary potential (Frankham 1996).

Regional-Scale Genetic Subdivision across the Study Region

In the study area, we identified an east–west division in genetic structure corresponding to a large discontinuity in structural connectivity west of Shelbourne, between the Bendigo and Dunolly forest blocks (c. 20 km; Fig. 1), and three smaller geographic clusters in the west of the study region. Restricted gene flow between these eastern and western areas may reflect a response to European habitat clearance over the past 150 years and/or a longer-term effect such as naturally occurring differences in vegetation types. Absolute timescales of population separation are challenging to estimate (Landguth et al. 2010) and will be addressed in a separate publication. The available evidence, including low genetic distinctiveness of the geographic regions in their allele identity (amova), and the ρ RST test for evolutionary timescale, suggest that subdivision detected across the Box-Ironbark region is probably quite recent (potentially either before or after European habitat clearance), as opposed to long evolutionary time-scales. Low-to-moderate migration among regions in recent generations and an apparent lack of strong evolutionary divergence would suggest that despite some large-scale spatial genetic structuring, superb fairy-wrens across the Box-Ironbark region have been operating as a single population or metapopulation.

Conclusions and Management Implications

We found evidence of altered fine-scale genetic structure in heavily fragmented agricultural landscapes for a currently common and widespread bird species with a complex social system. While long-distance dispersal by female fairy-wrens does not appear to be strongly inhibited by the habitat alteration experienced in the study area, we did observe effects that are likely to signal disruption of local gene flow mechanisms and/or mating systems (increased genetic similarity and reduced allelic diversity, potentially flagging population size contractions in isolated areas). While superb fairy-wrens may be able to successfully utilize the structural modifications and resources of some urbanized environments (Trollope, White & Cooke 2009), our study suggests that the type of land-use change (e.g. large tracts of land cleared for agriculture) experienced in the Box-Ironbark region has had negative consequences for fine-scale population processes.

Effective management of fairy-wren populations may require revegetation or restoration of habitat adjacent to areas of existing habitat in low-cover landscapes, which will provide additional habitat and improve the capacity for male short-distance dispersal and female-mediated extra-group copulation (i.e. local gene flow mechanisms). The study region contains spatial groups identified by cluster analysis. Dating the separations of these geographic groups would be a valuable goal for future work, but on the current evidence these seem recent in origin (although still possibly before European habitat clearance), in which case management should seek to reconnect rather than keep them distinct (Frankham et al. 2011; Weeks et al. 2011).

Acknowledgements

Funding was provided by the Australian Research Council Linkage Grant (LP0776322), the Victorian Department of Sustainability and Environment (DSE), Museum of Victoria, Victorian Department of Primary Industries, Parks Victoria, North Central Catchment Management Authority, and Goulburn Broken Catchment Management Authority. Birds Australia contributed towards NA's PhD stipend, and Monash University Science Faculty funded a Dean's Scholarship. We thank Holsworth Wildlife Research Endowment for valuable support to NA. Samples were collected under DSE permit number 10004294 under the Wildlife Act 1975 and the National Parks Act 1975, DSE permit number NWF10455 under section 52 of the forest Act 1958 and the Australian Bird and Bat Banding Scheme permit under approval and monitoring of Monash University ethics processes (BSCI/2007/07). We thank volunteers for assistance with fieldwork, and other Birds Linkage team members for diverse inputs. Jian Yen and two anonymous reviewers provided helpful comments on earlier drafts. Computationally intensive analyses (TESS) were performed on Monash Sun Grid.

Ancillary

Advertisement