592 



Fishery Bulletin 103(4) 



Data analyses 



Canonical variates analysis (CVA) was used to visualize 

 the separation between species and the relative impor- 

 tance of all variables (morphometric characters, pigment 

 patterns, and month of capture) in that separation. 

 Results from the CVA were used to help drive charac- 

 ter selection for subsequent analyses. The significance 

 of the canonical axes was obtained with a Monte Carlo 

 permutation test (499 iterations). The canonical analyses 

 were performed with the software CANOCO (version 4.5, 

 Microcomputer Power, Ithaca, NY), and plotted with the 

 associated software CANODRAW. 



In the CVA, all the molecularly identified white mar- 

 lin (21) and blue marlin (68) with full measurement sets 

 (i.e., no missing values) and a subset of sailfish (135) 

 with full measurement sets were compared. Every at- 

 tempt was made to include fish from different locations, 

 different years and months of collection, and across 

 the full available size range of each species, in order to 

 capture as much intra- and inter-species variation as 

 possible. Forward selection was used as a guide for the 

 creation of a reduced set of variables by retaining those 

 that were significant for discrimination at oc=0.05 in a 

 Monte Carlo permutation test (499 iterations). Months 

 that were excluded by selection were restored to the 

 variable set to insure that the entire spawning season 

 was represented. It was assumed that pigment on the 

 right lower jaw ramus was of equal importance as pig- 

 ment in the corresponding location on the left lower jaw 

 ramus; thus if a pigment grid from only one side of the 

 jaw was selected, the corresponding grid from the other 

 side of the jaw was added back to the reduced set. 



In addition to its function as an exploratory tool for 

 character selection, CVA with the reduced set of vari- 

 ables was used to identify unknown larvae to species. 

 Ordination coordinates of an unknown larva were ob- 

 tained by summing the products of the canonical coef- 

 ficients and the character values for the unknown (stan- 

 dardized to mean 0, standard deviation 1). The identity 

 of an unknown larva was determined by its placement 

 in the ordination with respect to the reference larvae. 



The CVA provided clues as to which individual pig- 

 ment grid cells were important for species discrimina- 

 tion, but cluster analysis was employed to examine 

 overall lower jaw pigment patterns. Simple average 

 link cluster analysis of Jaccard similarity indices was 

 executed on pigment grid cell presence (binary coding) 

 in the suite of lower jaw grid cells with BioDiversity 

 Pro 1 software for the 26 white marlin with undamaged 

 lower jaws and for equal numbers of randomly chosen 

 blue marlin and sailfish. Analyses were conducted on 

 all larvae together, and separately by flexion stage. Pig- 

 ment drawings of the individual larvae within single- 



1 McAleece, N., P. J. D. Lambshead, G. L. J. Paterson, and J. 

 D. Gage. 1997. The National History Museum and The 

 Scottish Association for Marine Science. Website: http:// 

 www.sams.ac.uk/. [Accessed 5 February 2003.] 



species clusters were examined visually for commonali- 

 ties. If a pattern was detected, the entire database of 

 pigment position, number, and shape of all molecularly 

 identified larvae was searched for that pattern. Lower 

 jaw pigment patterns that were confined to one species 

 only were deemed diagnostic characters. 



Lower jaw pigment patterns alone did not resolve the 

 differences among the species sufficiently for identifica- 

 tion of all larvae. Therefore, for each species, continuous 

 variables related linearly to SL were regressed against 

 SL by using SAS (version 8.02, SAS Institute, Cary, 

 NO software. Two ratios were also examined in this 

 way — snout length divided by eye orbit diameter, and 

 snout length divided by eye diameter. Both ratios were 

 suggested by the results of the full-model CVA because 

 the influence of snout length was large and opposite in 

 sign to the large and similar vectors of orbit diameter 

 and eye diameter. The former ratio was also considered 

 by Ueyanagi (1963, 1964, 1974b) to be an important 

 distinguishing character for istiophorid larvae. The 

 same larvae that were used in the CVA analyses were 

 used for the regressions, plus three white marlin, two 

 sailfish, and two blue marlin that were excluded from 

 CVA because of a missing measurement. Suitability 

 of the characters for linear regression was assessed 

 visually. Confidence intervals of 95%, 99%, and 99.9% 

 were constructed around the regressions. Intersections 

 of the three levels of confidence intervals for the three 

 species were examined for maximum discrimination at 

 the smallest standard length. The relationships that 

 provided the best separation were included in the iden- 

 tification key. 



The identification key was constructed from the vari- 

 ous characters that showed differences among the three 

 species. All of the larvae used in developing the key 

 were tested with it, as well as 12 blue marlin and 61 

 sailfish that were previously excluded from the analy- 

 ses. A set of 50 larvae were independently identified by 

 two observers unfamiliar with the key (naive observ- 

 ers). The only information about the fish provided to 

 them was month of capture, so that each made his own 

 measurements and pigment evaluations. The percent 

 accuracy of their identifications was taken as a measure 

 of the utility of the key. 



Results 



Molecular identification 



The molecular identification technique was applied to 

 1044 larvae. Amplification success rates appear to have 

 been negatively affected by the addition of BHT to etha- 

 nol and by the use of the Ruzzante et al. (1996) DNA 

 extraction protocol. Overall, 714 (68.4%) istiophorids 

 were successfully identified to the species level. Sailfish 

 represented 82.8%- of this group (591 larvae), whereas 

 96 blue marlin (13.4%) and 27 white marlin (3.8%) were 

 identified. No longbill spearfish were identified. Sailfish 

 larvae (2.9 mm-18.3 mm SL) were collected from April 



