370 



Fishery Bulletin 105(3) 



100°W 



90°W 



80°W 



70°W 



40°N 



30°N 



20°N 



40°N 



30°N 



20°N 



100°W 



90°W 



80°W 



70°W 



Figure 1 



Geographic range of the four menhaden species: Gulf menhaden (Brevoortia patronus) — 

 smooth gray line. Gulf; Atlantic menhaden (B. tyrannus) — dotted gray line, Atlantic; fine- 

 scale menhaden (B. gunteri) — dotted black line, western Gulf; and yellowfin menhaden (B. 

 smithi) — dashed black line, eastern Gulf). Sample sites are indicated by black boxes. 



species, and among all populations, was determined 

 by using Fstat software. A neighbor-joining (N-J) den- 

 drogram was constructed with genetic divergence as 

 a distance metric, and populations were revealed as 

 terminal nodes by using the freeware program MEGA, 

 vers. 2.1 (Kumar et al., 2001). 



With a small subset of the data (to accommodate 

 differences in sample size among species), we used a 

 Bayesian procedure, as implemented in the freeware 

 program Structure 2.1 (Pritchard et al., 2000), to ex- 

 amine historical admixture among species. Admixture 

 may be defined as any measure of combination and 

 introgression of genotypes (through hybridization) be- 

 tween two distinctive genetic sources, such as definitive 

 populations, subspecies, or species. Structure 2.1 uses 

 a Markovian chain Monte Carlo method (MCMC) to 

 assign individuals to population clusters such that sat- 

 isfaction of Hardy-Weinberg and linkage equilibrium ex- 

 pectations are maximized within populations. Because 

 of the mechanics of the MCMC method, each iteration 

 in a simulation is necessarily connected to the previous 

 iteration, and therefore initial iterations are subject to 

 the stochastic nature of random sampling. In order to 

 account for this, data from simulations should not be 

 collected until summary statistics begin to converge (to 

 "burn-in"). A burn-in length of 10,000 repetitions was 



chosen by examination of summary statistics for con- 

 vergence in preliminary runs. A run length of 100,000 

 repetitions was chosen after several preliminary trials, 

 after which parameter values were evaluated for consis- 

 tency. For all analyses, the admixture model was used, 

 and allele frequencies were assumed to be independent 

 among species. Although this procedure allows estima- 

 tion of the number of populations present {K), while 

 simultaneously assigning individuals, we constrained 

 the population number to K = A, to reflect the prior 

 assumption that only four species were present. Indi- 

 viduals were then assigned profiles that included the 

 proportion of their genotype contributed by each species 

 cluster. These profiles were then aligned by species in 

 order to visually examine the influence of genetic ad- 

 mixture upon species groups. 



Multilocus expected heterozygosity, estimated from 

 microsatellite allele frequencies, was used to generate 

 relative estimates of the value N^f.i for each of the four 

 species, where N^ is the effective population size and ;< 

 represents the overall mutation rate, assuming a step- 

 wise mutation model (Valdes et al., 1993). The estimator 

 dp (Xu and Fu, 2004) was used to approximate Af.^ and 

 was calculated as follows: 



0^= 0.5 (i^2_ I) 



