82 



Fishery Bulletin 93[1), 1995 



lensten and Erlich, 1988) carried out in a 100-uL 

 volume containing the same reactants as the initial 

 PCR but using 10 uL of the dissolved gel band and 

 reducing one primer concentration 100-fold. The 

 product was washed by centrifugal dialysis with ster- 

 ile water in Centricon microconcentrators (Amicon) 

 to remove excess dNTP's. Sequencing was performed 

 with the Sequenase kit (United States Biochemical, 

 Cleveland, Ohio) by using the limiting primer from 

 the asymmetric PCR reaction. Data from eight spe- 

 cies were obtained by directly sequencing double- 

 stranded PCR products. The template was purified 

 prior to sequencing (either directly from the PCR 

 reaction mix or following excision of the appropriate 

 band from low-melt agarose) with Magic PCR Preps 

 (Promega). Sequencing was performed with the 

 Sequenase kit according to the specifications of 

 Casanova et al. ( 1991 ). Sequences from Mycteroperca 

 and Morone was obtained after first cloning the PCR 

 products in pGEM t- vector (Promega) according to 

 the manufacturer's instructions. Transformation was 

 carried out by using XL-1 blue cells. Two positive 

 clones were selected for each PCR product. Double- 

 stranded sequencing (Sequenase 2.0) was performed 

 following alkaline denaturation as recommended by 

 the manufacturer. Sequence was obtained from both 

 strands of the amplified fragment for all individuals. 



Analysis 



Sequences were aligned by using the Mac Vector pro- 

 gram (IBI Biotechnologies). Maximum parsimony 

 analysis was performed with PAUP 3.1. (Swofford, 

 1991). Neighbor-joining (Saitou and Nei, 1987) and 

 UPGMA dendograms were constructed with Phylip 

 3.5 (Felsenstein, 1993). The strength of support for 

 various nodes was assessed by using the bootstrap 

 analysis (Felsenstein, 1985). Specific conditions for 

 each analysis are contained in the figure legends. 



Competing phylogenetic hypotheses were com- 

 pared by using the "enforce topological constraints" 

 option of PAUP 3.1. This option allowed us to deter- 

 mine the length difference between the most parsi- 

 monious trees that support each hypothesis. The cla- 

 distic permutation test for monophyly and 

 nonmonophyly (Faith, 1991) was then used to ascer- 

 tain whether the more parsimonious hypothesis is 

 significantly better than the competing hypothesis 

 according to the criterion of parsimony. The test was 

 performed as follows. The actual length difference 

 between trees supporting the two opposing hypoth- 

 eses was obtained. Then 99 permuted data sets were 

 constructed from the original data set by randomly 

 shuffling the character states for each character. We 

 then obtained the length difference between trees 



supporting the two opposing hypotheses for each 

 permuted data set. If the actual length difference was 

 matched or exceeded fewer than 5 times in all 100 

 data sets (the original data set plus 99 permuted data 

 sets), then the more parsimonious hypothesis was 

 considered to be significantly better than the less 

 parsimonious hypothesis. This corresponds to a to- 

 pology-dependent permutation tail probability, or T- 

 PTP, of less than or equal to 0.05. 



The effects of character weighting on parsimony 

 analysis were assessed by EOR weighting (Thomas 

 and Beckenbach, 1989; Knight and Mindell, 1993): 

 each type of nucleotide substitution was weighted 

 according to the ratio of its expected number of oc- 

 currences divided by its observed number of occur- 

 rences, or EOR. There are six types of nucleotide 

 substitutions if we disregard the direction of change: 

 A«G, CoT, G<=>T, G<=>C, A»T, and A<=>C. The ob- 

 served number of each substitution type was obtained 

 through pairwise sequence comparisons. Pairwise 

 comparisons were performed between sets of sister 

 species (sister species were identified through an 

 initial unweighted phylogenetic analysis; see Fig. 2). 

 Sister-species comparisons were used for two reasons. 

 First, within a clade, sister species will tend to rep- 

 resent relatively recent speciation events. This 

 recency lessens the chance that multiple substitu- 

 tions have occurred at the same site and that more 

 recent substitutions obscure older ones. Second, all 

 comparisons between pairs of sister species are mu- 

 tually independent. Therefore, if we restrict our com- 

 parisons to sister species, we cannot count the same 

 base substitution twice. 



We modified the method of Knight and Mindell 

 ( 1993) to derive the expected number of substitutions 

 in each class. This method accounts for differences 

 in the frequencies of the four nucleotides that greatly 

 influence the expected frequency of each substitu- 

 tion type. For instance, if guanine residues are very 

 rare, then substitutions of other nucleotides for gua- 

 nine will also be rare. The L-strand base composi- 

 tion of cytochrome b in scombroid fishes is strongly 

 skewed (Table 2), as it is in other groups examined 

 (for example, Irwin et al., 1991). Cytosines and thy- 

 midines each compose nearly 30% of the total nucle- 

 otide population whereas guanines compose less than 

 16%. In order to incorporate knowledge of the base 

 composition into our derivation of the expected num- 

 ber of each substitution type, we proceeded as fol- 

 lows. First, the average frequency of each nucleotide 

 (/) was obtained for all species used in the pairwise 

 sequence comparisons. Second, the observed num- 

 ber of each substitution type (S 0(i -j), where i and./ 

 represent two different nucleotides, was obtained by 

 summing the results from all pairwise comparisons of 



