224 
Fishery Bulletin 117(3) 
and 54°C (7, 7, and 20 cycles, respectively) for 1 min, and 
extension at 64°C for 2 min, followed by a final extension 
step at 64°C for 60 min (as in Darden et al., 2014). Both 
size standards (GenomeLab DNA Size Standard Kit 400, 
Beckman Coulter, Inc.) and reaction products were sepa¬ 
rated with a Beckman Coulter CEQ 8000 Genetic Analysis 
System, with fragment size analysis performed with CEQ 
8000 software. All chromatograms were scored manually 
by 2 independent readers. Discrepancies between read¬ 
ers were resolved in conference, or samples were rerun to 
obtain an unambiguous genotype for all individuals. 
In 2004, researchers at SCDNR began a cobia stock 
enhancement research program, releasing discrete 
numbers of cobia juveniles into the wild. Therefore, all 
hatchery-produced fish were removed from the data 
set prior to further analysis. We used a maximum likeli¬ 
hood parentage approach as implemented in the software 
Cervus, vers. 3.0.3 (Kalinowski et al., 2007), to provide 
a statistical evaluation of parentage taking into account 
mutation rates, population allele frequencies, and lab 
error rates. The power of the loci suite to correctly iden¬ 
tify hatchery fish as well as individual fish is high, with 
average parent-pair and identity non-exclusion probabil¬ 
ities of 1.7x10“' and 7.8xl0“ 12 , respectively, indicating 
very low probabilities of incorrectly identifying hatchery 
fish or individuals. Parentage simulations (number of 
simulations [n]=20) were run with known sex parentage 
analysis by using allele frequencies from individuals col¬ 
lected from 2007 through 2009 (n=1407). All simulations 
were conducted with 10,000 offspring, 8 candidate parent 
pairs (with all parents sampled), 95% genotyping, and low 
mistyping error (0.01) and mutation (0.001) rates. Critical 
delta scores were determined by using 99.0% and 99.9% 
confidence levels for the relaxed and strict criteria, respec¬ 
tively. Parentage analyses for the juvenile samples were 
conducted with the modal simulation file from the simu¬ 
lation runs. All parental assignments were designated at 
the strict confidence level (99.9%). 
All remaining individuals that were successfully geno- 
typed at 8 or more loci were subjected to sibship analyses 
as implemented in the software COLONY, vers. 2.0.6.4 
(Jones and Wang, 2010), to identify any potential large 
family groups within the data set that could confound 
further genetic structure analyses. Two simulations were 
run by using settings of polygamous breeding, weak prior, 
updating allele frequencies, no genotyping error, and full 
likelihood and pairwise likelihood combined method for 
a medium run length. Any identified duplicate samples 
were removed from the data set prior to further analy¬ 
ses. Results were evaluated for consistency among runs 
for individual fullsib relationships as well as family sizes 
present. 
For the initial analyses, the data set was partitioned into 
18 geographic sections based on natural latitudinal breaks 
in the collection data (Table 2, Fig. 2). Standard population 
genetic statistical analyses were applied to the resulting 
sample data set. Population genetic structure throughout 
the collection range was assessed through evaluations of 
Hardy-Weinburg equilibrium (HWE) in GenAlEx, vers. 
6.5 (Peakall and Smouse, 2006, 2012), analysis of molecu¬ 
lar variance (AMOVA) in Arlequin, vers. 3.5.1.2 (Excoffier 
and Lischer, 2010), pairwise F ST -style statistics calculated 
in GenAlEx and Arlequin, and the clustering algorithms 
implemented in STRUCTURE, vers. 2.3.4 (Pritchard et al., 
2000). Iterative AMOVAs (i? ST based) were conducted to 
evaluate areas of genetic discontinuity in the data set with 
potential location groupings under 2- and 3-population 
scenarios. Pairwise comparisons of sample locations and 
HWE were conducted initially at the smallest geographic 
scale, and locations were combined sequentially to repre¬ 
sent the smallest number of homogenous groupings. Esti¬ 
mates of Rgfp, F ST , Cg'p, G g^ (Nei, 1973), G g-p, and Dggp 
were initially calculated to verify consistency across met¬ 
rics. Because patterns of all estimates were consistent, only 
i? ST metrics are reported. The clustering model assignment 
employed in the program STRUCTURE by using a hierar¬ 
chical approach with the assistance of the web-based soft¬ 
ware STRUCTURE HARVESTER, vers. 0.6.94 (Earl and 
vonHoldt, 2012), was used to identify the most appropriate 
number of distinct populations ( K) of each run. Simulations 
were run by using the locprior parameter, with 5 replicates 
for each K, the length of the burn-in period was 20,000 
runs, and the number of Markov chain Monte-Carlo reps 
after burn-in was 20,000. Sites that were strongly assigned 
to one population were removed from the data set, and 
STRUCTURE was run iteratively until K= 1 was the most 
appropriate assignment for each cluster. The effective num¬ 
bers of migrants per generation and year (based on gener¬ 
ation time of 5-7 years for cobia) were calculated for each 
resulting homogenous cluster in Arlequin. 
Results 
Tag-recapture data 
The tagging data analyzed covers a 29-year period, with 
the first fish tagged in 1988 and the last recapture occur¬ 
ring in 2017. During that period, 25,867 cobia were tagged 
cumulatively by all 7 tagging programs (Table 1), and 
2124 cobia were subsequently recaptured and reported 
(8%) with the highest recapture rates occurring in South 
Carolina (18%) and Virginia (11%) and the lowest in the 
GOM (7%). After removing recaptures that were missing 
location or date information (n=110) as well as those that 
occurred less than 30 d after tagging («=264), the com¬ 
bined data set consisted of 1750 recaptures. Mean FL 
at tagging was largest in Virginia, South Carolina, and 
North Carolina (Table 3) and was smaller along the coast 
of Florida and in the GOM. Overall, mean FL at tagging 
was 786 mm, indicating that most tagging efforts were 
focused on sublegal cobia because the minimum legal size 
was 838 mm FL until September 2017 in federal waters in 
the western North Atlantic Ocean and until 2016 in North 
Carolina and Virginia state waters. Minimum legal size 
in the GOM remains 838 mm FL. Mean FL at recapture 
(Table 4) was largest from cobia recaptured in North Caro¬ 
lina, South Carolina, and Virginia but did not vary greatly 
