Study species and sampling locations
A. polyacanthus were collected from 15 back-reefs from 3 regions on the Great Barrier Reef during 2003 and 2004 (Table 1, Figure 1). This species is very common and show not apparent variation in abundance on reefs either along or across the GBR [84,85]. The geographic distribution of this species extends from 15°N to 26°S http://www.fishbase.org webcite and the southern region sampled here was therefore near the southern limit of this species. A. polyacanthus is
polymorphic with a southern black morph, and a black and white morph in
the central and northern regions of the reef. Fish were captured using
small hand spears or baited fence nets and hand nets and were
transported either alive or on ice to the nearest shore where genetic
samples (fin clips) were taken and preserved in 80% EtOH. Genetic
effects of continental shelf position, inner, middle and outer, were
examined among replicate reefs in two regions (i.e. north and central).
Because the southern region contains no true inner- and mid-shelf
zones, the genetic structure in this region was explored using pairwise
DNA extraction and amplification
Genomic DNA was extracted from approx 0.25 cm2 of fin tissue (re-hydrated by several TE washes) using a modified CTAB extraction procedure [, excluding the phenol-chloroform step] and re-suspended in 50 μl of TE. Concentrated DNA stock was diluted 1:50 yielding a final DNA concentration of approximately 50 ng/μL.
A 400 base-pair region of the mitochondrial control region (hyper
variable region I, HVR I) was amplified using the specific forward
primer (dLoopF 5'-CATATATGTRTTATCAACATTA-3') and the universal primers
CR-E H16498 (5'-CCTGAAGTAGGAACCAGATG-3') . PCR reactions were carried out on a PE Applied Biosystems 9700 in 25 μl containing 1× PCR Buffer (Promega), 3.5 mM MgCl2, 200 μM each dNTP, 0.4 μM each primer, 10 ng template DNA and 0.1 unit of Taq Polymerase (Promega).
Amplification using the polymerase chain reaction (PCR) was conducted
with a cycling profile of 30 s at 94°C, 45 s at 48°C and 60 s at 72°C
for 30 cycles. The cycling profile was flanked by an initial 2 min
denaturing step (94°C) and a 10 min terminal extension phase (72°C).
PCR products were cleaned up using PCR clean up columns (Qiagen) and
re-suspended in 20 μL of ddH2O. Two μL
of the purified product was sequenced in the forward and reverse
direction using a dyenamic ET dye terminator kit (Megabase) chemistry
(Amersham Biosciences). Sequence products were purified using Sephadex
G-50 columns. Labelled extension products were sequenced on a Megabase
1000 (Amersham Biosciences). Representative sequences have been
deposited in a public database [GenBank: DQ199666 – DQ199947].
The control region sequences were aligned and edited using Sequencher 4.2 (GeneCodes Corp. Michigan USA) and ESEE . The best model of nucleotide substitution was determined using Modeltest 3.5  and PAUP* 4.0b10 . The hierarchical likelihood tests and Akaike Information Criteria agreed that the Tamura and Nei model  with γ
= 0.3012 fitted the data best (-LogLikelihood = 1220.65; AIC =
2453.30). This model and rate heterogeneity estimate was used in all
following analyses of population genetic structure. Base frequencies
and the ts/tv ratio from all sampled fish combined were calculated
using Modeltest. The role of saturation was explored by comparing the
topology of neighbour joining trees (implemented in PAUP*) including
and excluding transitions. All individuals retained membership in the
same major clades and transitions were included in all further analyses.
Population Genetic Structure
Estimates of mitochondrial haplotype and nucleotide diversity [91-93] and their associated standard deviations were calculated using Arlequin 3.11 
for each reef and region. Statistical differences in genetic diversity
among regions were determined using Kruskal-Wallis tests implemented in
Hierarchical population genetic structure of A. polyacanthus among regions and reefs was explored using AMOVA with 1000 permutations [95,96]
implemented in Arlequin. Pairwise genetic distances among reefs were
calculated using Arlequin and a false discovery rate to correct for
multiple tests (Benjamini-Hochberg) was applied to all pairwise
Differences in levels of migration among reefs were investigated further using Migrate 18.104.22.168 [33,34]. This program calculates reciprocal migration rates (i.e., 4Nem
from a to b, and vice versa) using a coalescence maximum likelihood
approach (Markov Chain Monte Carlo with Hastings Metropolis importance
sampling) and assumes constant mutation rates and equal effective
population sizes. Because of the molecular divergences detected by
phylogenetic and AMOVA analyses, Migrate was run on each geographical
region separately. Reciprocal migration rates were interpreted as
different when their 95% confidence intervals did not overlap.
Extensive sampling regimes including 10 short chains sampled 10,000
times each and 5 long chains sampled 100,000 each were averaged over 5
replicates. Migrate was implemented on a SGI Origin 3800 computer in
the James Cook University High Performance Computing Facility using a
ts/tv ratio of 1.53 (estimated by Modeltest). Earlier versions of
Migrate had problems with convergence of estimates, migration estimates
and their associated profile likelihoods in low signal data .
Here we used a newer version of Migrate with high signal data and found
no evidence of lack of convergence as repeated runs were highly
consistent using the implemented sampling strategy. We also found
congruence of migration estimates with conventional estimates of
Genetic distances were estimated using the conventional genetic distance estimator, ΦST, and geographical distances among locations were calculated using Vincenty's inverse method http://www.ga.gov.au/nmd/geodesy/datums/distance.jsp webcite.
Correlations between genetic and geographical distances were tested
using a Mantel test (1,000 permutations) of both log-transformed and
non-transformed data following  and implemented in GenAlEx 6 . A false discovery rate to correct for multiple tests (Benjamini-Hochberg) was applied to all pairwise comparisons . Log transformation did not affect the overall results and therefore, only non-transformed kms versus ΦST are presented here.
Demographic histories were explored using mismatch analysis in Arlequin and DnaSP 
using 1000 bootstrap replicates. These analyses computed the
distribution of pairwise nucleotide differences to that expected under
population models of constant and sudden expansion and assume that
sub-populations are panmictic. The best fit of models was determined
using log-likelihood ratio tests. The sums of square deviations (SSD)
from the observed mismatch distributions were calculated for each of
the models and log-likelihoods calculated following the methodology
outlined in .
The statistical significance of log-likelihood ratios was adjusted
using FDR as above and when different, the model with the lowest SSD
was accepted. The age of population expansion was estimated by τ = 2 μt, where μ = the mutation rate and t = generation time. τ values were not converted to absolute years because of uncertainty associated with estimating mutation rates in fishes  and because comparisons were relative among regions and reefs. Differences in τ
values were compared among regions using a Kruskal-Wallis test.
Predictions from the meta-population re-colonisation models were
examined by comparing estimates of genetic differentiation among older
and younger reefs. Relative ages of reef samples were defined using the
time of population expansion estimates (τ). Younger reefs were defined by having τ confidence intervals that could not be distinguished from the present (i.e. 95% CI of τ included 0). Older reefs were defined by having τ confidence intervals that did not encompass the present (i.e. 95% CI of τ
did not include 0). Following this methodology, replicate younger and
older reefs could only be compared in the central and southern regions
because the 95% CI of τ did not include 0 in any northern reefs. Because of the uncertainty associated with τ estimates and the relatively low number of reefs compared here, these results should be interpreted with caution.
The exponential population growth parameter (g) was calculated among reefs and regions using a maximum likelihood coalescence approach implemented in Fluctuate 1.4 .
This approach assumes that subpopulations are panmictic, that
population structure, growth, immigration and recombination rates have
remained constant throughout the lifespan of the underlying coalescent
tree . A search strategy, each 10000 steps long using ten short chains, sampling every 20th step, followed by ten long chains each of 20000 steps sampled every 20th step, gave consistent results among runs and was used in all analyses. Estimates of g were compared among regions using a Kruskal-Wallis test. We calculated Fu's Fs  and R2  neutrality indices using DnaSP because they are the most sensitive to population growth . Significance level was corrected for multiple testing using FDR as above.