504 
Fishery Bulletin 107(4) 
Dive software (Meirmans, 2006). Finally, genetic chord 
distance, D CE (Cavalli-Sforza and Edwards, 1967), be- 
tween each population pair was calculated in GENETIX 
with permutation tests (1000 randomizations) used to 
estimate P-values. 
Individual-based analyses may be more suited to 
questions of dispersal and connectivity because in these 
analyses, information contained in each individual mul- 
tilocus genotype is used. By comparison, in population- 
based analyses, allele frequencies and heterozygosities 
are calculated for each population. Three different indi- 
vidual-based analyses were employed in this study. 
The frequency-based assignment method of Paetkau 
et al. (1995) was implemented in GENECLASS software 
(Piry et al., 2004). Populations were determined a priori. 
by sampling locations, and GENECLASS generated 
allele frequencies for each population, excluding the 
individual to be assigned in the given procedure (Waser 
and Strobeck, 1998). The expected frequency of each 
individual’s genotype at each locus across all popula- 
tions was calculated and each individual was assigned 
to the population from which its multilocus genotype 
most likely originated. Alleles that were absent from a 
population were designated a frequency of 0.001. 
Genotypes were also analyzed by a Bayesian pro- 
cedure implemented in the program STRUCTURE 
(Pritchard et al., 2000). STRUCTURE uses a Markov 
Chain Monte Carlo (MCMC) algorithm to cluster in- 
dividuals into populations that each exhibit Hardy- 
Weinberg and linkage equilibrium (HWLE), without 
prior definition of the number or geographic location of 
these populations. Five runs were performed at each 
value of K genetic clusters, with K varied from 1 to 5, to 
ensure proper mixing in the MCMC chain of iterations 
and consistent results. For all runs we used a burn-in 
period of 10 6 iterations and followed it by 10 6 MCMC 
iterations. We assumed an admixture model, in which 
individuals may have mixed ancestry and correlated 
allele frequencies, which could account for similarity 
between closely related populations. 
As a third method for inferring genetic structure in 
L. analis, we also used the landscape genetics program 
Geneland, available in the R statistical package (Guil- 
lot et al., 2005a, 2005b). This software operates like 
STRUCTURE in using Bayesian inference of Mendelian 
populations in HWLE. But unlike STRUCTURE, Gene- 
land incorporates geographic coordinates of the samples 
into the prior parameters of the estimation procedure. 
Recent applications (e.g., Galarza et al., 2009) show 
promise for inferring structure at low levels of genetic 
differentiation between marine populations. For spatial 
coordinates, we ran separate analyses with and without 
a variable “uncertainty” factor — roughly interpretable 
as encompassing the home range of an individual fish 
and appropriate in the case of highly mobile animals 
(Guillot et al., 2005a). Each run comprised 10 5 MCMC 
iterations with a thinning set at 100 and K genetic 
clusters varying from 1 to 10; Dirichlet (uncorrelated) 
allele frequency distributions were assumed and null 
allele frequencies were explicitly considered (Guillot et 
al., 2008a, 2008b). Ten independent runs under each 
set of conditions were launched to check for conver- 
gence on K populations. Once a reliable estimate of 
K was found, a run with this value fixed was used to 
estimate and map posterior probabilities of population 
membership. 
Results 
High levels of polymorphism were observed in all five 
populations of mutton snapper at the eight microsatel- 
lite loci. The number of alleles detected per locus ranged 
from nine to 32, and expected and observed heterozy- 
gosities ranged from 0.771 to 0.968 and 0.500 to 0.982, 
respectively (Table 2). Predictably, populations with 
larger sample sizes exhibited slightly increased levels 
of allelic diversity; there were 149 alleles present in the 
JP population and only 135 in the DT population. Also, 
only two private alleles (i.e. alleles present in only a 
single population) were present in the DT population, 
whereas all other populations contained seven or eight. 
However, there were no apparent trends towards reduced 
heterozygosity in populations with smaller sample sizes, 
and estimations of allelic richness indicated that no 
single population was particularly deficient in genetic 
diversity across loci. 
Seven out of 40 tests indicated significant departures 
from HWE (0<P<0.05). These significant tests were 
distributed evenly across populations, yet four of the 
significant tests involved locus La39 (Table 2). After 
the implementation of sequential Bonferroni correc- 
tions (Rice, 1989), only three tests, those involving 
La39, remained significant. Pairwise locus-population 
tests of linkage disequilibrium yielded six out of 140 
significant comparisons (0.01<P<0.05); however, no test 
remained significant after sequential Bonferroni correc- 
tions. Three out of 80 tests indicated significant hetero- 
geneity in allelic distribution between population pairs 
(0.01<P<0.05). Tests of genotypic distributions between 
population pairs indicated significant heterogeneity 
in two out of 80 tests (0.01<P<0.05). After sequential 
Bonferroni corrections, no test of genic or genotypic 
differentiation remained significant. 
Values of F ST can range from zero to one with zero in- 
dicating the absence of population substructure; values 
for estimators of this parameter can also be negative, 
indicating greater heterozygosity within than between 
populations. In this study estimates of F ST (6) for each 
locus ranged from -0.005 to 0 and estimates of R ST 
(p) ranged from -0.009 to 0.005. Pairwise estimates 
of F st (0) ranged from -0.0035 to 0.0022 and pairwise 
genetic distances, D CE , ranged from 0.012 to 0.018; none 
of these pairwise comparisons were significant. To ad- 
dress the effect of high variation on estimates of Fst> 
Meirmans (2006) has developed, <p ST , a standardized 
measure of genetic variation based on the analysis of 
molecular variance (AMOVA) framework. As with Fst 
( d ), all pairwise estimates of <p ST were negative, ranging 
from -0.033 to -0.059. 
