Pocillopora damicornis was sampled from 28 sites in the northern, central and southern regions of the Great Barrier Reef. In the northern region samples were collected from Martin Reef (1 site), Eyrie Reef (1 site), Lizard Island (9 sites), MacGillivray Reef (1 site) and Yonge Reef (1 site). In the central region samples were collected from Fantome Island (1 site), Orpheus Island (4 sites), Pelorus Island (3 sites), Trunk Reef (2 sites), Dip Reef (1 site) and Myrmidon Reef (1 site). Samples were collected in the southern region from Chinaman Reef (1 site), East Cay Reef (1 site) and Frigate Reef (1 site).
Fragments of Pocillopora damicornis, approximately two cm long, were collected from each of 1040 colonies. Each colony was photographed and the location recorded by towing a handheld Garmin Etrex GPS unit in a waterproof container on the surface. Between 6 and 99 individuals were sampled per site. Collections were made haphazardly on the upper reef slope along a zigzag transect approximately 200 m long, between 2 m and 10 m of depth. Unattached or asymmetrical colonies within one metre of a colony already sampled were not sampled due to the possibility that these were clones by fragmentation. Coral branches were fixed in absolute ethanol.
DNA was extracted using a modified protocol of the salt precipitation method. Samples were genotyped using nine microsatellite markers ( xxxxxx) according to the multiplex groups, primers and protocols described in Torda et al. (2013). Samples from four sites, Dip Reef, Chinaman Reef, East Cay Reef and Frigate Reef were not genotyped for marker Pd4.
To determine the lineage identity of each sample, a rapid genetic assay was used. The vast majority of samples (72%; i.e.745 samples) were
118 identified as Type alpha. The remaining samples were Type beta (22%; i.e. 228 samples), 'other Pocillopora' (5%; i.e. 56 samples) or did not give reliable results (1%; 11 samples) (Table 1).
All subsequent analyses were carried out on Type alpha and Type beta samples separately, omitting 'other Pocillopora' and unidentified samples, which potentially include the poorly resolved genetic lineage Type gamma and P. verrucosa (Schmidt-Roach et al. 2013).
Data analyses
To assess the discriminative power of sets of loci, Genotype Probability (GP) was calculated for each sample and each locus in GENALEX 6.4 (Peakall & Smouse 2006). Repeated multilocus genotypes (MLG) were considered to be clone mates if the product of GP for all 128 loci was < 0.001. Allopatric clone mates (for this study defined as originating from different 129 sampling sites) were kept, while all but one copy of sympatric clone mates were removed 130 before subsequent analyses. Deviations of populations from Hardy¿Weinberg Equilibrium (HWE) and genotypic linkage disequilibrium (LD) were tested in Genepop web version 4.0.10 (Raymond & Rousset 1995; Rousset 2008) using the log likelihood ratio statistic (G-test). Descriptive statistics were obtained in GENALEX. The Fis analogue Gis was calculated in GenoDive (Meirmans & Van 136 Tienderen 2004). Allelic richness was calculated in FSTAT v2.9.3.2 (Goudet 1995) for each locus and population. Populations that lacked data for a locus (DIP, CH, EC and FR for locus Pd4) were excluded from analyses based on all nine microsatellite loci, and instead analysed separately for the eight loci for which data were available. Allelic richness was compared between populations with Kruskal-Wallis tests.
To estimate genetic differentiation among populations, we used Dest (Jost 2008), calculated in SMOGD 1.2.5 (Crawford 2010), because this statistic is not sensitive to genetic diversity and because it accounts for both migration and mutation rates, being based on the finite island model. Significance levels of Dest values were determined by a permutation test, randomizing alleles over all compared populations, using R code from Alberto et al. (2011). For easier comparison with results of other studies, we explored other statistics as well, including (i) uncorrected pairwise Fst values by the ¿weighted" analysis of variance method (Weir & Cockerham 1984), as implemented in Genepop; (ii) the standardised pairwise F¿st, estimated using an AMOVA (Meirmans 2006) in GenoDive; (iii) pairwise Fst values corrected for null alleles (ENA correction), computed in FreeNA (Chapuis & Estoup 2007). To account for unbalanced sample sizes, the significance of uncorrected Fst values was assessed by a Fisher exact test (Goudet 1995) in Genepop with the default Markov chain parameters. To facilitate direct comparison of our results with those of previous allozyme studies, we also carried out a hierarchical analysis of standardised genetic variance as Weir & Cockerham¿s (1984) ¿ using the program TFPGA, following Ayre & Hughes (2000). To approximate the sampling design of Ayre & Hughes (2000), only samples collected from around Lizard Island and the Palm Islands were used for this analysis and samples from Type ¿ and s were pooled, as information on these distinct genetic lineages was not available in the earlier studies.
To detect putative first generation migrants, the probability that each individual belongs to each reference population was computed in GeneClass2 (Piry et al. 2004) using the criteria and probability computation algorithm of Rannala and Mountain (1997), with 10,000 simulated genotypes. Individuals were excluded from a reference population if the probability of exclusion was greater than 0.99 (¿<=0.01), and assigned to another reference population as a potential source if assignment probabilities were greater than 0.1.
Genetic structuring of samples without prior definition of populations was analysed using the Bayesian clustering method implemented in InStruct (Gao et al. 2007). As opposed to the more commonly used method implemented in Structure (Pritchard et al. 2000), InStruct accounts for potential selfing. Five independent chains were run for each K from K = 1 to K = 20, with a burn-in of 100,000 and another 100,000 MCMC replications after the burn-in, using the ¿infer population structure and population selfing rates¿ function with the default samplers.
To assist in the interpretation of the results of the genetic analyses, the potential dispersal capacity of brooded larvae was simulated using virtual Lagrangian particle transport modelling in the 0.025° x 0.025°-cell circulation model of the GBR in Connie 2.0 (CSIRO Connectivity Interface, http://www.csiro.au/connie2/). Collection sites were selected as both sources and sinks for dispersal of passive particles at a depth of 5 m over a dispersal period of 1, 15, 50 and 100 days.
