Ward et al.: Population structure of Thunnus albacares 
567 
allozyme loci in Pacific Ocean collections have shown 
significant spatial heterogeneity at one locus (GPI- 
A*)\ the common allele in western and central regions 
differed from that in the east (Sharp, 1978; Ward et al., 
1994). This finding either indicates the existence of two 
reproductively isolated groups of yellowfin tuna in the 
Pacific Ocean or suggests that selection pressures are 
different in the two regions. There is no evidence of 
mitochondrial DNA (mtDNA) differentiation between 
eastern Pacific and western Pacific yellowfin tuna 
(Scoles and Graves, 1993; Ward et al., 1994). 
In the Atlantic Ocean, where it was once assumed 
that there were separate eastern and western stocks, 
recent taggings of large yellowfin tuna have resulted 
in 15 trans-Atlantic recoveries (ICCAT, 1992b); a 
single stock is now assumed (ICCAT, 1995). There 
have been no genetic comparisons of eastern and 
western Atlantic yellowfin tuna. 
The extent of genetic differentiation of yellowfin 
tuna from different oceans has been little studied. 
Suzuki (1962) found no differences in the incidence 
of the Tg2 blood group antigen in fish from the equa- 
torial Pacific and Indian Oceans. Scoles and Graves 
(1993) found no significant differentiation in mtDNA 
from one west Atlantic collection and five Pacific col- 
lections (each of 20 fish). Here we compare genetic 
variation in collections from the Pacific, Indian, and 
Atlantic oceans. We used larger sample sizes than 
those used in the study by Scoles and Graves (1993) 
and examined both allozyme and mtDNA variation 
to see if the increased statistical power would en- 
able us to reject the null hypothesis of no interoce- 
anic genetic differentiation. 
Materials and methods 
Samples were collected from one region of the Atlan- 
tic Ocean, two regions of the Indian Ocean, and six 
regions of the Pacific Ocean. Details of most of the 
Pacific collections (Philippines, Coral Sea, Kiribati, 
Hawaii, California, and Mexico) are given in Ward 
et al. (1994). For the present paper, the 1991 and 
1992 Hawaii collections were pooled. A second Phil- 
ippines collection was collected in October-Decem- 
ber 1994. This showed no significant genetic differ- 
entiation from the earlier collection; therefore the 
two collections were pooled for our study. The Atlan- 
tic collection was taken from the Caribbean Sea (Gulf 
of Mexico, approx. 28°N, 88°W) in September 1993. 
The Indian Ocean collections were taken from off Sri 
Lanka (approx. 6°N, 80°E) and off the Seychelles 
(approx. 7°S, 52°E) in December 1994. White muscle 
samples were flown (airfreight) frozen to Hobart and 
stored at -80°C. 
Specimens were studied by allozyme and mtDNA 
analysis. The experimental methods are given in 
Ward et al. (1994). Four allozyme loci known to be 
polymorphic in white muscle were examined: ADA* 
(adenosine deaminase, EC 3. 5. 4. 4), FH* (fumarate 
hydratase, EC 4.2. 1.2), GPI-A*, and GPI-B* (glucose- 
6-phosphate isomerase, EC 5.3. 1.9). MtDNA varia- 
tion was examined by using two restriction enzymes 
( Bel I and Eco RI) known to discriminate most of the 
mtDNA haplotypes revealed by eight restriction en- 
zymes in an earlier survey (Bam HI, Ban I, Bel I, 
Eco RI, Hind III, Pvu II, Sal I, and Xho I — see Ward 
et al., 1994). 
The homogeneity of allele and haplotype frequen- 
cies of the collections was tested by the randomized 
Monte Carlo chi-square procedure of Roff and 
Bentzen (1989). For each test, 2,000 randomizations 
of the data were carried out, each giving a randomized 
chi-square value ( X 2 nu ii )• The probability that the null 
hypothesis of genetic homogeneity was correct was 
given by P = n/2,000, where n is the number of ran- 
domizations that generate X 2 null - X 2 an d where % 2 is 
the chi-square value given by the actual observations. 
The extent of genetic differentiation among collec- 
tions was quantified by Nei’s gene diversity statistic 
G st (Nei, 1987), which estimates the proportion of 
total genetic variation attributable to differentiation 
between populations. For each allozyme locus, G sr 
was estimated as (H T - Hg) / H T , where H T represents 
total heterozygosity and H s is average (Hardy- 
Weinberg expected) population heterozygosity. The 
mtDNA data were analyzed in a similar way, treat- 
ing haplotypes as alleles and H r and H s as diversity 
estimates. The proportion or magnitude of G ST gen- 
erated by sampling error, which we have termed 
G sT.nuii > was estimated with a bootstrapping program, 
with the observed allele or haplotype frequencies and 
collection sizes. Simulations were run 1,000 times to 
provide a mean value of G STnull and a standard de- 
viation. The probability of obtaining a value of G STnull 
as large or larger than that obtained from the actual 
observations of G gT was given by P = n/1,000, where 
n is the number of randomizations that generate 
G sr.nuii - G S t- Values of P less than 0.05 indicated 
significant differentiation between areas that could 
not be explained by sampling error alone. 
Bonferroni adjustments of significance levels, to 
correct for multiple tests, were carried out with the 
sequential procedure advocated by Hochberg ( 1988). 
Tests are ordered according to their probability value. 
The highest probability value, P , is compared with 
the significance value a. Here we initially set a = 
0.05. If P ^ oc, that test is judged to be nonsignifi- 
cant, and comparisons continue with subsequent 
probabilities, each compared with a modified signifi- 
