We obtained tissue samples from 238 Lampropholis delicata, representing 120 different populations, from across the entire Australian range of the species (Figure 2, 3; Additional files 1, 2). Samples were obtained from the frozen-tissue collections of several Australian Museums (Australian Museum, South Australian Museum, CSIRO Australian National Wildlife Collection), along with our own field collections (Additional files 1, 2). We included the closely related L. guichenoti (Australian Museum NR2639) and an Australian Eugongylus-lineage skink Niveoscincus pretiosus (Australian Museum NR391) as outgroups in our study.

Total genomic DNA was extracted from liver, muscle, toe or tail-tip samples using a Qiagen DNeasy Blood and Tissue Extraction Kit (Qiagen, Hilden, Germany). For each sample we sequenced portions of four mitochondrial genes: ND2 (~600 bp), ND4 (~700 bp), 12SrRNA (~700 bp), and 16SrRNA (~500 bp). These regions were targeted because work across several taxonomic levels in squamate reptiles has indicated useful levels of variability [e.g. 26 31 32 33 34 ]. The primers used to amplify and sequence these regions are provided in Additional file 3. PCR was conducted as outlined in Greaves et al. 35 , except on a Corbett Research GC1-960 thermal cycler. PCR products were purified using ExoSAP-IT (USB Corporation, Cleveland, Ohio USA). The purified product was sequenced directly using a BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) and then analysed on an ABI 3730XL capillary sequencer.

Sequence data were edited using CONTIGEXPRESS in VECTOR NTI ADVANCE v9.1.0 (Invitrogen), and aligned using the default parameters of CLUSTAL X v1.83 36 . We translated all coding region sequences to confirm that none contained premature stop codons. Sequence data were submitted to GenBank [GenBank:JF438009-JF438959, EF567304, EU567726, EU567768, EU567769, EU567927, EU567928, EU568019, EU568020] (Additional file 1).

Maximum Likelihood (ML) and Bayesian tree building methods were used. We used MODELTEST 3.7 37 to identify the most appropriate model of sequence evolution based on the AIC criterion. MODELTEST, conducted in PAUP* 4.0b10 38 , was also used to estimate base frequencies, substitution rates, the proportion of invariable sites (I) and the among-site substitution rate variation (G). These values were then used as settings in PhyML 3.0 39 to generate a ML tree with 500 bootstraps.

MRBAYES 3.1.2 40 was then used to complete Bayesian analyses. Preliminary analysis of each mtDNA region revealed congruent tree topologies. In order to evaluate partitioning strategies, we used MODELTEST to determine the most appropriate model for each partition. We then conducted a Bayesian analysis for each partitioning strategy, applying the appropriate model of evolution to each partition, and allowing among-partition rate variation. We ran each Bayesian analysis for five million generations, sampling every 100 generations (i.e. 50,000 sampled trees). We ran each analysis twice, using four heated chains per run. We discarded the first 25% of samples as burn-in and the last 37,500 trees were used to estimate the Bayesian posterior probabilities. In order to calculate the AIC and BIC scores for different partitions strategies, we calculated the number of parameters for each. Following McGuire et al. 41 , for each parameter we added the number of substitution rates for the model suggested by MODELTEST for that partition (6 for GTR, or 2 for HKY), the number of free equilibrium base frequencies (3 for GTR and HKY), plus one parameter per partition where appropriate for each of I and/or G. For multi-partition strategies, we also added one parameter per partition, corresponding to the among-partition rate multiplier. To calculate the AIC and BIC scores, we used the equations: AIC = -2L_i + 2k_i and BIC = -2L_i + (k_i ).(ln n) (where L _i is the harmonic mean log likelihood for partitioning strategy i, k_i is the total number of parameters for partitioning strategy i, and n is the total number of nucleotides). The program TRACER 1.5 42 was used to check for chain convergence and mixing. Specifically, raw traces of sampled values versus MCMC step numbers were examined to confirm that there was no trend away from the mean and that there were no large fluctuations in the likelihood values.

Bootstrap values (500 ML bootstraps) and Bayesian posterior probabilities were used to assess branch support. We considered branches supported by bootstrap values of 70% or greater 43 , and/or posterior probability values greater than or equal to 95% 44 to be supported by our data.

Estimates of genetic diversity within L. delicata clades (number of haplotypes, h; haplotypic diversity, Hd; number of polymorphic sites, S; nucleotide diversity, π) were calculated in DNASP v4.50 45 . Tamura-Nei (TrN)-corrected genetic distances within and among clades were calculated in MEGA 4 46 . Genetic differentiation among clades within L. delicata was estimated in ARLEQUIN v3.5 47 . Pairwise Φ_ST values (an analogue of Wright's fixation index F _ST) were calculated to estimate among clade differentiation. We conducted hierarchical Analysis of Molecular Variance (AMOVA; 48 ) to investigate the impact of the a priori (Tables 1 and 2) and a posteriori biogeographic barriers on the partitioning of genetic variation within L. delicata. Both tests used TrN genetic distances with gamma correction (using the value calculated from MODELTEST). Significance levels of all the estimated values were calculated by 10,000 permutations, and adjusted according to the Bonferroni correction procedure 49 for multiple pairwise comparisons as described by Holm 50 .

We used Tajima's D 50 , Fu's F statistic 51 (calculated in ARLEQUIN) and mismatch distributions to test for signatures of population expansion within L. delicata clades. Significant and negative Tajima's D and Fu's F statistic values are indicative of possible population expansion. Mismatch frequency histograms were plotted in DNASP to determine whether the clades exhibited evidence of spatial range expansion or a stationary population history 52 . A smooth bell shape signifies either population expansion or spatial range expansion, whereas a multimodal distribution represents a long history in situ 53 54 55 56 . To distinguish between these two types of distribution, a raggedness index (RI, sum of the squared difference between neighbouring peaks) and the sum of squared deviations (SSD) between the observed and expected mismatch were calculated using the methods of Schneider & Excoffier 57 in ARLEQUIN. The spatial expansion hypothesis (both RI and SSD) was tested using a parametric bootstrap approach (200 replicates).

As there are no suitable fossil calibration points available for Lampropholis skinks, we estimated the divergence time of L. delicata clades using an evolutionary rate of 1.3-1.63% sequence divergence per million years, based on mitochondrial DNA calibrations from other squamate reptile groups (1.3%, 58 ; 1.42-1.63%, 59 ; 1.55%, 60 ; 1.62%, 61 ; 1.63%, 62 ). A strict molecular clock (0.0065-0.00815 per lineage substitution rate), implemented in BEAST v1.6.1 63 , was used to estimate the divergence times within L. delicata. The Australian Eugongylus lineage is estimated to have originated ~20 mya 64 , and this information was used as the maximum age of the tree root. A GTR+I+G model of evolution was employed with a coalescent (Bayesian skyline) tree prior. The analysis was run twice, with 20 million generations per run (total 40 million generations). The output was viewed in TRACER to check that stationarity had been reached, and ensure that the effective sample size (ESS) exceeded 200 63 . The two separate runs were then combined using LOGCOMBINER v1.6.1, with a maximum clade credibility tree generated in TREEANNOTATOR v1.6.1 and visualised in FIGTREE v1.3.1. A Bayesian skyline plot 65 was also generated in TRACER to examine the magnitude and timing of population size changes in L. delicata.

The edited alignment comprised 2426 characters (550 bp ND2, 671 bp ND4, 708 bp 12SrRNA, 497 bp 16SrRNA; Additional file 4), of which 813 (33.5%) were variable and 587 (24.2%) were parsimony-informative. For the ingroup only, the alignment contained 638 (26.3%) variable characters, of which 543 (22.4%) were parsimony-informative. Base frequencies were unequal (A = 0.3694, T = 0.2172, C = 0.2889, G = 0.1245), but a χ² test confirmed the homogeneity of base frequencies among sequences (df = 498, P = 1.0). The phylogenetic analyses were conducted on a dataset comprising the 165 unique haplotypes that were present within L. delicata (Table 3).

The AIC from MODELTEST supported the GTR + I + G substitution model as the most appropriate for our unpartitioned dataset. Parameters estimated under this model were: relative substitution rates (A↔C = 2.1053, A↔G = 44.1898, A↔T = 2.5350, C↔G = 1.1020, C↔T = 29.2135, relative to G↔T = 1.00), proportion of invariable sites (0.5241), and gamma distribution shape parameter (0.7015). We evaluated three partitioning strategies for our dataset (Table 4). The unpartitioned and by gene partitioning strategy were analysed using the GTR + I + G model for all nucletotides. When the data was partitioned by codon, we used a mixture of GTR and HKY models for each partition (Table 4). Both the AIC and BIC scores ranked the most highly parameterised strategy (by gene and codon) as the most appropriate. However, the topologies of the ML, unpartitioned Bayesian and partitioned Bayesian trees were congruent, therefore we present the optimal ML tree (-ln L = 14501.43296) with ML bootstrap (BS) values and unpartitioned Bayesian posterior probabilities (PP) indicating branch support (Figure 4, 5). Nine well-supported, non-overlapping clades (labelled from the top to bottom of the tree, and roughly related to their north to south geographic distribution) are present within L. delicata (Figure 2, 3, 4, 5), with high levels of haplotypic and nucleotide diversity within each clade (Table 3). The PP's of the main clades and subclades in the partitioned Bayesian analysis were identical to those from the unpartitioned Bayesian analysis presented in Figure 4 and 5, except that the support value for Clade 9 was lower (0.85 rather than 0.95).

Clade 1 encompasses populations from coastal northern and eastern QLD, and is comprised of three subclades (Figure 2, 3, 4, 5). Subclade 1a includes populations north of Townsville, subclade 1b contains populations in the Mackay region, and subclade 1c stretches from the Rockhampton region to Bania State Forest (inland from Bundaberg) in south-east QLD (Figure 2, 3, 4, 5). Clade 2 is restricted to Kroombit Tops (Figure 2, 3, 4, 5). A complex mosaic of geographically non-overlapping clades occurs throughout south-eastern QLD and northern NSW (Figure 2, 3, 4, 5). Clade 3 includes populations from the Sunshine Coast and Brisbane region of south-eastern QLD (Figure 2, 3, 4, 5). There is strong support for a close affinity between Clade 2 and Clade 3 (Figure 4, 5). Subclade 3a occurs within Cooloola National Park, while subclade 3b is restricted to the Bunya Mountains (Figure 2, 3, 4, 5). Subclade 3c extends from the Maryborough region to the northern suburbs of Brisbane, with subclade 3d containing populations from the southern suburbs of Brisbane to Barney View on the western side of Lamington National Park (Figure 2, 3, 4, 5).

Clade 4 (96 BS) occurs along the Main Range and throughout the QLD/NSW border region, and in inland northern NSW (Figure 2, 3, 4, 5). Subclade 4a (93 BS) extends from Deongwar State Forest and other areas in south-eastern QLD (Lamington NP, Main Range) through the elevated regions of inland northern NSW to Riamukka State Forest, inland from Port Macquarie (Figure 2, 3, 4, 5). Subclade 4b is restricted to the more coastal regions of northern NSW (Mt Warning NP, Border Ranges NP, Nightcap NP, Whian Whian State Forest, Alstonville), while subclade 4c occurs further inland at Bolivia Hill, Torrington State Forest and near Armidale (Figure 2, 3, 4, 5). Clade 5 extends along the northern NSW coastal region from Yamba to Cairncross State Forest near Port Macquarie, and is comprised of two subclades: subclade 5a (Yamba to Coffs Harbour) and subclade 5b (Styx River State Forest to Cairncross State Forest) (Figure 2, 3, 4, 5). Clade 6 represents a disjunct inland population on the Blackdown Tableland in southern QLD (Figure 2, 3, 4, 5).

Clade 7 is geographically widespread, occurring from the Australian Capital Territory (ACT) and inland southern NSW (subclade 7a) across to western VIC (Little Desert NP) and south-eastern SA (subclade 7b) (Figure 2, 3, 4, 5). Clade 8 is a disjunct population that occurs in Coolah Tops National Park in inland northern NSW (Figure 2, 3, 4, 5). Clade 9 is distributed from the central coast of NSW and throughout eastern VIC and TAS (Figure 2, 3, 4, 5). Subclade 9a encompasses the central coast of NSW and the Sydney region, subclade 9b occurs at Brayton, while subclade 9c comprises populations from coastal southern NSW (Figure 2, 3, 4, 5). Subclade 9d represents a shallow clade that is distributed throughout eastern VIC and TAS (Figure 2, 3, 4, 5).

Considerable genetic differentiation was evident amongst the nine L. delicata clades, with extremely high and statistically significant pairwise Φ_ST values among clades (Table 5). The only comparisons that were not significant were those involving clades with low sample sizes (e.g. Clade 8). Substantial genetic distances are evident among the clades (4.3-7.4%; Table 5), indicating that the divergences within L. delicata occurred during the late Miocene-Pliocene (Figure 6, Additional file 5). The intra-clade genetic divergences in L. delicata were 0.0-2.6% (Table 3). The vast majority (97.7%) of genetic variation in L. delicata was partitioned among populations (Table 6). The nine a priori biogeographic barriers (Tables 1 and 2) accounted for 64.1% of the genetic variation in L. delicata (Table 6). This value increased to 66.5% when the two barriers identified a posteriori (Blackdown Tableland, Coolah Tops) were included in the analysis (Table 6).

The Bayesian skyline plot indicated recent (i.e. last 0.2 myr) contraction then expansion of L. delicata populations (Figure 7), although there was no consistent support for the model of spatial expansion in L. delicata clades or subclades. Three main clades (2, 4 and 7) and four subclades (3a, 4b, 7b and 9d) deviated significantly from the expectations of neutrality (Tajima's D, Fu's F statistic) (Table 3), suggesting recent population expansion. However, the RI and SSD values indicated that a model of population expansion could only be conclusively rejected for two main clades (1 and 4) and two subclades (1c and 4c) (Table 3).

Despite its widespread distribution along the east coast of Australia, there is substantial phylogeographic structure across the native range of the delicate skink (Figure 4, 5). In many instances these breaks are concordant with dry habitat corridors, indicating that regions of drier vegetation represent effective barriers to dispersal for the mesic-adapted delicate skink. For instance, the delicate skink exhibits a moderate genetic break (4.5%, subclade 1a vs 1b; Figure 6) across the Burdekin Gap in North QLD. Equivalent Pliocene divergences between populations either side of the Burdekin Gap have been reported in open forest frogs 66 67 , rainforest lizards 22 , woodland lizards 68 , rainforest birds (Pleistocene; 19 ) and freshwater fish (Miocene; 69 ) (Table 2). In contrast, the St Lawrence Gap north of Rockhampton on the central QLD coast has only been identified as a significant barrier for one lizard species (late Pleistocene-early Pliocene, 68 ; Table 2). Although divergence is also evident across the St Lawrence Gap in the delicate skink (4.8%, mid-late Pliocene, subclade 1b vs 1c), a more substantial break is evident a little to the south, between clades 1 and 3 (7.1%, late Miocene) in the Gladstone region (Table 5, Figure 2, 3, 4, 5).

Two high elevation areas (Kroombit Tops, Blackdown Tableland) inland from the main range of L. delicata in southern QLD were found to harbour genetically divergent lineages. Both areas are remnant patches of moist forest that are surrounded by drier lowland eucalypt woodland, and are disjunct from the main distribution of the delicate skink along the east coast ( 70 ; Figure 2, 3). Kroombit Tops (~730 m) was identified a priori as a potential habitat isolate for the delicate skink, as the region provides a cooler and wetter refuge for mesic-adapted species ( 70 71 ; Tables 1 and 2). The Kroombit Tops population of the delicate skink diverged from the surrounding coastal populations during the mid Pliocene (4.9%, Clade 2 vs 3; Table 5, Figure 6), a pattern that has also been observed in a rainforest bird 71 , two open forest frogs 66 67 , and several open forest reptiles 72 (Table 2). The Blackdown Tablelands are a moderate elevation plateau (~600 m) that provides an isolated refugium for numerous mesic-adapted species. Our results for the delicate skink (4.3-6.7%, Pliocene; Figure 6) provide evidence that the fauna of this region may also be genetically divergent from the coastal populations, a pattern also seen in other open forest reptiles 72 .

The Coolah Tops are a high elevation (1200 m) plateau located in inland northern NSW, just to the north of the Hunter Valley (Figure 2, 3). The refugial population of the delicate skink that occurs on the Coolah Tops was found to have diverged from the nearby populations in northern NSW during the early-mid Pliocene (4.6-4.8%; Table 5, Figure 6). The dry habitat in the Hunter River Valley has been demonstrated to represent a major barrier to dispersal in both woodland and wet forest species ( 73 74 75 76 ; Table 2). While the divergence across the Hunter Valley was estimated to have occurred in the Miocene in the congeneric L. guichenoti, which is frequently sympatric with L. delicata 26 , an early-mid Pliocene break was observed across this barrier in the delicate skink (subclade 9a vs Clades 4-5; Figure 2, 3, 4, 5). The divergence estimate for the delicate skink is consistent with those reported for most other species across the Hunter Valley (Table 2).

Our analyses revealed a complex mosaic of geographically structured, non-overlapping clades and subclades (Clades 3-5) in the delicate skink in south-eastern QLD and northern NSW (Figure 2, 3, 4, 5). The McPherson Range that occurs along the border region (Figure 1) is concordant with the phylogeographic break between Clades 3 and 4 (Figure 2, 3). The distribution of Clade 4 extends northwards into south-east QLD to the Main Range, which runs perpendicular to the western edge of the McPherson Range (Figure 1). Similar biogeographic patterns involving the Main and McPherson Ranges occur in wet forest (Litoria pearsoniana; 77 ) and open forest frogs (Litoria fallax; 66 ) (Table 2). The early-mid Pliocene split found across the McPherson/Main Ranges in the delicate skink (4.8%; Figure 6) is concordant with that observed in L. guichenoti, which also inhabits open woodlands and dry sclerophyll forest 28 , but intermediate between that reported for frogs (Miocene; 66 77 ) and a wet forest snake (Pleistocene; 78 ) (Table 2).

Some relatively minor phylogeographic structure is evident among the populations in the Maryborough, Sunshine Coast and Brisbane regions of south-east QLD (subclades 3a-d; Figure 2, 3, 4, 5). A more substantial break occurs between inland (Clade 4) and coastal (Clade 5) delicate skink populations north of the Hunter Valley in NSW (4.5%, early-mid Pliocene; Table 5, Figure 6). A similar coastal vs inland divergence in northern NSW is shared with White's skink (Liopholis whitii, 74 ), and reflects an equivalent pattern that is regularly observed in southern NSW (Table 2). Indeed, an analogous pattern is evident within Clade 4 in northern NSW, with subclade 4b occurring near the coastal margin, subclade 4a present in intermediate areas (with secondary contact between 4a and 4b occurring in the Border Ranges NP), and subclade 4c occurring further inland (Figure 2, 3). These biogeographic patterns, combined with the break observed in southern NSW (Figure 2, 3), indicate that high elevation areas may represent barriers to dispersal in the delicate skink.

The five phylogeographic studies that have had sufficient sampling to examine the impact of the elevational and habitat barriers in southern NSW have reported a genetic break between the inland (including the ACT) and coastal regions ( 26 71 73 74 79 ; Table 2). The impact of this barrier is pronounced in the delicate skink, as populations from the ACT and inland southern NSW are more closely related to the SA populations (3.0%, early Pleistocene-late Pliocene) than the adjacent populations along the NSW coast (5.3%, mid-Pliocene; Figure 2, 3, 4, 5). This indicates that the delicate skink most likely reached SA from the southern NSW region via an inland route, rather than along a coastal dispersal pathway through VIC. The delicate skink may have dispersed through the mesic vegetation that is located along the Murray River, which forms the border between NSW and VIC for the majority of its length (Figure 2, 3). Indeed, the eastern water skink (Eulamprus quoyii) has a continuous distribution through the Murray-Darling River system that connects populations along the east coast (North QLD to southern NSW) to an isolated population in south-eastern SA 28 . This biogeographic pattern was not previously suspected for the delicate skink and explains the large distributional gap across western VIC between the eastern suburbs of Melbourne and Little Desert NP in north-western VIC (Figure 2, 3). Given this pattern, it was not possible to examine the impact of the Murray Basin on the delicate skink (Table 2).

Several frog and lizard species exhibit deep phylogeographic breaks in the East Gippsland region 26 67 74 79 , a pattern that is believed to be the result of repeated marine inundation of the area since the Miocene (Tables 1 and 2). However, the East Gippsland region does not appear to constitute a barrier to dispersal in the delicate skink, with an extremely shallow clade (intraclade divergence 0.2%) distributed across eastern VIC and across Bass Strait into TAS (Figure 2, 3). The coastal area in East Gippsland has been relatively stable since the late Pleistocene 15 80 , enabling the delicate to disperse across eastern VIC from southern NSW.

The delicate skink has colonised TAS during the late Pleistocene, with the presence of shared haplotypes between populations in eastern VIC and TAS indicating a connection between these two regions until relatively recently (~12-15 kya; Table 5, Figure 6, Additional file 2). The timing coincides with the inundation of the Bassian Isthmus, which connected eastern VIC (Wilsons Promotory) to north-eastern TAS between 43-14 kya ( 16 ; Table 1, Figure 1). Although the delicate skink does not currently occur on Wilsons Promontory, it is present on Flinders Island (which formed part of the Bassian Isthmus) and is common throughout north-eastern TAS ([ 28 81 , DGC, personal observation]). While a second land bridge was located from western VIC, through the King Island region to western TAS from ~43-17.5 kya 16 , the absence of the delicate skink from western VIC would have precluded dispersal of the species via this western route. Fossil evidence for a Nothofagus tree species on King Island 38 kya suggests that moist forest vegetation occurred along the Bass Strait land bridges 82 , enabling dispersal of the delicate skink into TAS. Although some other species appear not to have used these recent land bridges (frog: Crinia signigera, 79 ; reptiles: Acritoscincus duperreyi, 76 ; Lerista bougainvilli, 83 ; mammals: Dasyurus maculatus, 84 ), others appear to have dispersed across these Bass Strait land bridges (frogs: Limnodynastes peronii and tasmaniensis, 67 ; reptiles: Liopholis whitii, 73 ; Notechis scutatus, 85 ) (Table 2). The repeated presence of the land bridges has also restricted east-west gene flow across Bass Strait in several marine invertebrate species ( 86 87 88 89 ; Table 2).

There is no evidence to support the hypothesis of Rawlinson 29 that the delicate skink is a 'glacial relic' with a relatively long presence in southern Australia. In contrast, our analyses indicate that the delicate skink only colonised VIC and TAS during the late Pleistocene from coastal southern NSW. Although the delicate skink (also known as the rainbow or plague skink in its introduced range) is a successful invasive species in the Hawaiian Islands, New Zealand, and Lord Howe Island 90 , there is no strong evidence to suggest that it represents an introduced species in TAS. However, given the relatively shallow genetic divergences within subclade 9d, we are unable to completely exclude the possibility that the delicate skink reached TAS via human-associated colonisation.