Blaylock et al : Use of parasites In discriminating stocks of Hippoglossus stenolepis 



Pacific halibut (Blaylock et al., 1998a), eight taxa met 

 these criteria: the juvenile acanthocephalans Corynosoma 

 strumosum (body cavity) and C. villosum (body cavity), the 

 metacestode A^v6e/;>;(a surmenicola (stomach wall), the di- 

 genean metacercaria Otodistomum sp. (stomach wall), and 

 the larval nematodes Anisakis simplex (body cavity, or- 

 gans, musculature), Pseudoterranova decipiens (body cav- 

 ity, organs, musculature), Contracaecum sp.(body cavity), 

 and Spirurid gen. sp. (stomach wall). A ninth taxon, the 

 larval nematode Hysterothylaciun7 adiincum (body cavity 

 and organs) was included for the analysis of juveniles. 



Because individual fish varied extensively in size (fork 

 length), and the number of a parasite individuals was strong- 

 ly correlated with fish size (Blaylock et al., 1998a), parasite 

 numbers were corrected for differences in host size. Counts of 

 individual parasites were first log-transformed (ln(A:-i-l)). To 

 adjust for the effect offish length, a regression of the trans- 

 formed parasite numbers on fish length for each species in 

 each locality (and haul) was calculated. This relationship was 

 then used to adjust the number of parasite individuals within 

 each fish in each locality (and haul) to that expected for the 

 average-size fish in the overall sample (80.9 cm for adults, 

 39.2 cm for juveniles). These data were then used in discrimi- 

 nant function analyses. We applied the most widely used (and 

 available) method of discriminant function analysis, in which 

 the data were divided into training and test sets, and a dis- 

 criminant function calculated on the training set was used to 

 classify the test set. Interpretations were based on patterns 

 in the test sets. To insure that any identified patterns were 

 due to differences among localities rather than simply differ- 

 ences among individual hauls, we performed the same analy- 

 sis on both the locality and the individual haul data. 



Our training set consisted of six fish randomly selected 

 from each haul ("haul" training set) or these fish plus four 

 from the northern Queen Charlotte Islands and six from 

 Unimak Pass ("locality" training set). Discriminant func- 

 tions calculated from data on these "training" fish were 

 used to classify each of the remaining fish from each haul 

 ("haul" test set) or those fish plus all remaining fish ("local- 

 ity" test set). The test set fish were first classified into one 

 of the 13 hauls or 15 localities. Classification matrices were 

 examined for the degree of misclassification. Hauls or local- 

 ities were then grouped and regrouped into four and three 

 groups based on patterns in the 13 or 15 category analyses 

 and the zoogeographic zones from Blaylock et al. (1998b). 

 Analyses were then repeated. Classifications were exam- 

 ined for misclassification, and boundaries adjusted for re- 

 testing. Results presented are those from the best fit "test" 

 classifications. Statistical analyses were performed in SYS- 

 TAT for Windows version 5.05 (Wilkinson et al., 19921. The 

 entire data set from which the data for this analysis came is 

 available for purchase from the Depository of Unpublished 

 Data, Document Delivery, CISTI, National Research Coun- 

 cil of Canada, Ottawa, ON KIA 0S2, Canada. 



Results 



Of the taxa that met the Arthur and Albert ( 1993) criteria, 

 A^. surmenicola was most common and abundant in north- 



ern localities and fairly common and abundant in central 

 localities. Corynosoma strumosum, although variable in 

 prevalence and abundance, was much more common in the 

 northernmost localities. Corynosoma villosum, although 

 prevalent everywhere, was more abundant in northern 

 fish. Otodistomum sp. and Spirurid gen. sp. were more 

 common and abundant in southern localities. Anisakis 

 simplex, although present in virtually every fish from 

 every locality, was more abundant in southern fish. Pseu- 

 doterranova decipiens and Contracaecum sp. appeared to 

 be more common in central areas (Table 1). In the juve- 

 niles, A. simplex and P. decipiens were more common in 

 central localities, whereas C. villosum, C. strumosum, and 

 Hysterothylacium aduncum were more common in north- 

 ern localities (Table 2). 



The haul analyses indicated that the majority of fish 

 from some hauls (12/14 Vancouver Island |VI] fish, 3/4 

 Southeast Alaska 1 |SA1| fish, 3/5 from the Pribilof Islands 

 |PI|, and all 4 from St. Matthew's Island ISMI]) could be 

 correctly classified but that fish from surrounding areas 

 also were incorrectly classified to these hauls. Moreover, 

 the percentage offish correctly classified by the haul func- 

 tions was, in all cases, within only a few percentage points 

 of that correctly classified by the equivalent locality func- 

 tion. Thus, patterns do not appear to be associated with 

 independent hauls. Therefore, we present only the results 

 of the locality analyses. 



Fifteen category discriminant analyses revealed severe 

 misclassification in most areas. Only 39% were correctly 

 classified to locality (Table 3). The functions did assign 

 correctly the majority of test fish from two localities ( 19/26 

 from Vancouver Island [VI] and 14/22 from the southern 

 Bering Sea [SBl ). However, misclassification of fish from 

 surrounding areas to these localities indicated less than 

 accurate discrimination. The clearest indications from 

 these analyses were that localities from the vicinity of the 

 Queen Charlotte Islands south should be grouped together 

 and that there is a suggestion that the two northern Ber- 

 ing Sea locations (PI and SMI) should be grouped. 



Regrouping the localities into four categories by using 

 boundaries from zoogeographic analyses (Blaylock et al., 

 1998b) plus the apparent northern Bering Sea grouping 

 (PI-SMI), considerably improved the predictive ability of 

 the functions. The "best fit" four-category grouping gave ap- 

 proximately 62% correct classification at the locality level 

 (Table 4). The four-category functions were good predictors 

 for the California-Oregon (OR) to southern Queen Char- 

 lotte Islands (SQC) fish; over 80% of these southern fish 

 were correctly classified, and only about 6% of the other fish 

 were misclassified to this group. Over 70% of the Pribilof-St. 

 Matthew Island (PI-SMI) fish were correctly classified, and 

 only 7% of the other fish were incorrectly classified to this 

 group. There was much misclassification in the two central 

 groups, and adjustment of the boundary between these two 

 groups did not produce marked improvement (not shown). 



Grouping into three categories by combining the two 

 central groups resulted in substantial improvement in 

 discrimination (83% correct) (Table 5). Shifting of the 

 boundary between the northern and central group re- 

 vealed that discrimination broke down when the southern 



