Seyoum et al.: Genetically determined population structure of Lachnolaimus maximus in the southeastern United States 
447 
A 
08 
09 
07 06 
05 
02 03 qi 
04 
PC 1 
B 
O Cluster 1 I 0 Cluster 2 Cluster 3 
Figure 2 
Principal coordinate analysis based on a nonstandardized distance 
method of a matrix of pairwise (A) uncorrected and (B) corrected 
genetic distances (Fgj) between sampling areas 1-9 for specimens 
of hogfish (Lachnolaimus maximus) collected from November 2005 
through August 2013 in the southeastern United States (from north- 
west Florida to the Carolinas). Analysis separated the 9 sampling 
areas into 3 main genetic clusters, with principal coordinate 1 (PCI) 
and PC2 explaining 58.9% and 38.8% of the variability for uncor- 
rected and 68% and 32% of the variability for corrected, respectively. 
Sampling areas 1-9 are labeled numerically by geographic region as 
defined in Figure 1 and are identified as cluster 1 (circles), cluster 2 
(stars), or cluster 3 (diamond). 
Pairwise Fqt values between uncor- 
rected (and corrected) Fgx values were 
0.015 (0.016) between clusters 1 and 2, 
0.027 (0.023) between clusters 1 and 3, 
and 0.030 (0.039) between clusters 2 and 
3 (Table 2). All pairwise Fct comparisons 
among the clusters were highly significant 
(P<0.001). Overall, the Fgj statistic and 
the AMOVA confirmed the presence of 3 
geographically based clusters, but these 
clusters appeared to be hierarchically ar- 
ranged; the least genetic differentiation 
was seen between clusters 1 and 2 and 
the differentiation between clusters 2 and 
3 exceeded that observed between the geo- 
graphically disjunctive clusters 1 and 3. 
Interestingly, the observed pattern of ge- 
netic differentiation did not conform with 
the expected greatest differentiation be- 
tween clusters separated by the greatest 
geographic distances. 
Bayesian population assignment 
Lacking a method for determining wheth- 
er values of L (K) statistically differed, 
we derived inferences herein by evaluat- 
ing replicate likelihoods and resultant 
A K statistics for different values of K. 
L (K) increased quickly from K= 1 to K= 2 
and somewhat less quickly from K= 2 to 
K= 3, reaching a plateau at successive 
values (L(F); Fig. 3). The largest value 
of A K occurred for K=2 (AF=281) and, 
secondarily, for K - 3 (AF=118) (A K; Fig. 
3). At the base hierarchical level of K= 2, 
sampling areas of clusters 1 and 3 were 
predominantly conjoined within a single 
Bayesian cluster, whereas sampling areas 
of cluster 2 were predominantly assigned to a dis- 
crete Bayesian cluster (Fig. 4A). Virtually the same 
result was obtained under the model of no admix- 
ture analysis for K= 2. At the next hierarchical level 
of K= 3, sampling areas of clusters 1 and 3 were as- 
signed to different Bayesian clusters (Fig. 4B). The 
CLUMPP analysis indicated that sampling areas 5 
and 6, were admixtures of Bayesian clusters 1 and 
2, respectively, exhibiting graduated mean genomic 
proportions (Fig. 4C). 
Mantel test 
The number of paired comparisons within cluster 1 
(from Big Bend to the Everglades) was sufficient to al- 
low only a within-cluster Mantel test. No significant 
correlation was observed between genetic and geo- 
graphic distances (P=0.125, r=0.334). The Fgx value be- 
tween sampling areas 7 and 8 was not significant (Fgx= 
-0.0007 P>0.9). However, when sampling areas 7 and 
8 were included in the Mantel test with the sampling 
areas from cluster 1, there was a significant correla- 
tion between genetic and geographic distance (P=0.004, 
r=0.543). This correlation was attributed to the genetic 
break between cluster 1 and cluster 2 rather than to 
isolation by distance. 
Effective population size 
The 1% allele-frequency criterion was used to de- 
termine the following estimates of N e for each of 
the 3 clusters: cluster 1=1368.2 (95% confidence in- 
terval [CI] = 1022. 6-2033. 4; n=324); cluster 2=1035.7 
(95% 01=833.5-1750.2; n= 223); and cluster 3=285.6 
(95% 01 = 216.2-411.2; n = 102). The 2% allele-fre- 
quency criterion yielded the following: cluster 
1=1478.4 (95% CI= 1022.6-2581.1); cluster 2=1075.5 
(95% 01=748.3-1853.8); and cluster 3=327.5 (95% 
01=231.9-537.23). The effective population size of 
cluster 3 (sampling area 9, from the Carolinas) was 
approximately 3 times smaller than that of the oth- 
er 2 clusters. 
