568 



Fishery Bulletin 94(3). 1996 



a principal components analysis of the correlation 

 matrix with NTSYS software (Rohlf, 1994). 



A noteworthy characteristic of coho salmon popu- 

 lations south of Alaska is their unique propensity to 

 return to spawn at three years of age (Sandercock, 

 1991). We classified samples from the same locale as 

 "on cycle" if the brood years were three years apart 

 and as "off cycle" when otherwise. We tested for a 

 significant effect of brood-year cycle within a locale 

 on allele frequencies by nesting the brood-year cycle 

 within-locale in the above mentioned ANOVA. 



Results 



Four alleles were observed with relative mobilities 

 of 106, 103, 100, and 97. The *106 allele was observed 

 at only three locations (Cowlitz, Lewis, and Trinity 

 rivers) at a low frequency (P<0.03). Frequencies of 

 the other alleles varied considerably among samples: 

 0.116<P<1.000 for the *103 allele, 0.000<P<0.535 for 

 the *100 allele, and 0.000<P<0.764 for the *97 al- 

 lele. Allele frequencies for the 48 new samples ana- 

 lyzed in our laboratory are included in Table 1 (iden- 

 tified as "NMFS"). 



Of the 48 samples, 3 (6.3%) deviated from expected 

 Hardy-Weinberg proportions: Nehalem Hatchery, 

 brood year 1982 (P=0.022); Bonneville Hatchery, 

 brood year 1989 (P=0.008 ); and Minter Creek Hatch- 

 ery, brood year 1990 (P=0.039). Although the total 

 number of significant tests (n=3) was only slightly 

 greater than the number expected by chance alone 

 (n=2.4), we have no reason to reject the assumptions 

 that the samples represented panmictic populations 

 and the genetic model used to interpret the observed 

 allozyme patterns was correct. Data from an inher- 

 itance study have confirmed that the transferrin al- 

 leles segregate according to expected Mendelian seg- 

 regation proportions (Van Doornik, unpubl. data). 



Transferrin data compiled from nine sources in 

 addition to those collected for this study are shown 

 in Table 1. The total data set included 135 samples 

 from sites ranging from California to southern Brit- 

 ish Columbia ( Fig. 1 ). Sample sites were grouped into 

 the following eight geographic regions: Vancouver Is- 

 land, mainland British Columbia, Puget Sound, coastal 

 Washington, Columbia River, northern Oregon (north 

 of Cape Perpetual, southern Oregon, and California. 



Sixty-three pairwise, G-tests were made by com- 

 paring allele frequencies of samples from the same 

 location but from different brood years. A total of 4 1% 

 (n=26) of the temporal comparisons were statistically 

 significant (P<0.05). Despite this substantial tem- 

 poral variability, the ANOVA indicated significant 

 between-sample variation (F=2 1.1, df=89, P=0). ( The 



rare *106 allele was pooled with the '103 allele fre- 

 quencies for this analysis. ) The contribution of brood- 

 year cycle within sample locations to the overall ge- 

 netic variation was not statistically significant for 

 the *100 allele (P=2.1, df=24, P=0.06) and the +103 

 allele (F=0.9, df=24, P=0.60). 



The gene diversity analysis indicated that most of 

 the genetic diversity was found within locations 

 (72.5%). The other components of relative gene di- 

 versity were 22.3% between geographic regions, 2.1% 

 between stocks within a geographic region, and 3.1% 

 between temporal comparisons. Relative gene diver- 

 sity (G sl ) equaled 27.5%. Total genetic diversity (H T ) 

 was 0.586, and the mean genetic diversity within 

 populations (H ) equaled 0.425. 



A dendrogram of Nei's genetic identity (I) for the 

 compiled data for 135 samples revealed three major 

 clusters at 1=0.900 that followed regional geographic 

 patterns (Fig. 2). All samples from Vancouver Island 

 (n=2), the Washington coast («=8), Puget Sound 

 (rt=32), and northern Oregon (n=22) formed cluster 

 1. Cluster 1 is distinguished by samples with moder- 

 ate frequencies of the *103 (0.080<P<0.501) and *97 

 (0.250<P<0.77) alleles. The 21 samples from south- 

 ern Oregon appeared in all three clusters. Samples 

 from Marlow Creek, Five Mile Creek, Elk Creek, Coos 

 River, North Fork Siuslaw River, and two Umpqua 

 River stock samples (brood year 1981, 1992) also 

 grouped in cluster 1; cluster 3 contained samples from 

 the Rogue River (n=2) and Morton Creek; and the 

 remaining 11 southern Oregon samples formed the sole 

 members of cluster 2. Cluster 2 contained samples with 

 relatively high frequencies of the *100 allele (0.356 

 <P<0.540). All of the California (n=10), Columbia River 

 (ra=36), and mainland British Columbia (n=4) samples 

 grouped together in cluster 3. Cluster 3 consisted of 

 samples with high frequencies of the '103 allele 

 (0.640<P<1.000). 



The principal components analysis yielded results 

 identical to those from Nei's genetic identity values, 

 and therefore, are not included here. 



Discussion 



The geographical pattern of genetic variability shown 

 by these data supports several conclusions regard- 

 ing the genetic population structure of some of these 

 stocks. Results from cluster analysis of the total data 

 set (Fig. 2) indicated that samples from large river 

 systems (Fraser River, Columbia River, Rogue River, 

 and Klamath River) are genetically similar to each 

 other despite the geographical distances separating 

 them. Several studies have shown that coho salmon 

 from small coastal rivers differ from those from large 



