Wildes et al.: Genetic variation in outer-coastal and fjord populations of Clupea pallasii in the eastern Gulf of Alaska 
385 
Homogeneity tests of allelic frequencies were con- 
ducted for all pairs of collections by using x 2 prob- 
abilities (Markov chain algorithm, GENEPOP, vers. 
4.0). P-values were corrected for multiple testing with 
the false discovery rate test (Benjamini and Yekutieli, 
2001). Pairwise F ST values, a measure of genetic diver- 
gence, were calculated with the Weir and Cockerham 
(1984) algorithm in FSTAT and evaluated with a per- 
mutation test. 
The amount of molecular diversity was measured 
within sites at locations for which multiple years of 
data from the same location were available (Berners 
Bay and Sitka Sound). The diversity among years was 
compared with the diversity between locations, by using 
a hierarchical AMOVA (analysis of molecular variance) 
with 1000 permutations in ARLEQUIN, vers. 3.5 (Ex- 
coffier and Schneider, 2005). The amount of variation 
was partitioned in the following categories: within the 
individual sample (Fsc), among collections from dif- 
ferent years at the same location (Fct), and between 
locations (Fst). 
Results 
Seven of the 22 loci were dropped from further analy- 
ses because of either large allele drop out ( Cpal02 ), 
suspected null alleles ( Cpa8 ), stutter bands ( Cpal04 , 
Chal23), one-bp shifts ( CpalOO ), or because of our 
inability to resolve the loci for all data sets {CpalOl, 
Cpal07a ), or because of one-bp shifts (CpalOO)', however, 
further optimization in the laboratory may prove these 
loci useful for future studies. The remaining 15 loci 
were Cpal03, Cpal08, Cpalll, Cpall2, Cpall3, Cpall4, 
(Olsen et ah, 2002) Cpa4, Cpa6, Cpa27, Cpal07, Cpal25, 
Cpal34, (Miller et ah, 2001) Chal017 and Chal020, 
(McPherson et al., 2001) and Cha63 (O’Connell et al., 
1998b) (Appendix). These loci were highly polymorphic 
in general; the average number of alleles was 30.8, 
ranging from 7 at Chal017 to 64 at Cpall2 (Appendix). 
Several loci with a large number of unique alleles, such 
as Cpal34 (58 alleles), and Cpall2 (64 alleles), exceeded 
the number of individuals (n = 52) in the Berners07 
collection. This finding illustrates the need for ample 
sample sizes for use with highly polymorphic markers. 
Average observed heterozygosity across all populations 
for each locus varied from 0.45 to 0.99. The number of 
alleles (n a ) found in each collection was similar overall, 
with the exception of those from the Berners07 collec- 
tion, which were typically lower. The allelic richness (a) 
and effective number of alleles (n eff ) were also similar 
overall (Appendix). 
Observed and expected heterozygosities were overall 
in close agreement for all loci. Nine of the 195 tests 
for Hardy Weinberg equilibrium had an excess of ho- 
mozygotes — an amount expected by chance alone — and 
importantly, no excesses of homozygotes at any one lo- 
cus were found in more than two collections — evidence 
that null alleles did not contribute significant bias to 
estimates derived from these 15 microsatellites. Low 
F IS values for most loci indicate random mating within 
each collection. Chal017 and Cpa27 had a slightly high 
overall F ls value of 0.167 and 0.114, respectively. Nine 
out of 105 linkage disequilibrium tests in GENEPOP 
were significant. No locus had a significant value at 
more than two collections, indicating that all loci were 
likely inherited independently. 
Global AMOVA results for the three Berners Bay 
collections and two Sitka Sound collections revealed 
that nearly all the variation was within the individuals 
(99.6%) and within the collection (0.35%). The remain- 
der of the genetic variation was attributable to regional 
differences between the two locations (0.05%; P=0.058). 
No variation was attributable to temporal differences 
among collection years within a region. 
Pairwise homogeneity tests of allelic frequency re- 
vealed statistically significant genetic differentiation 
at 10 (7 after correction for multiple testing) of the 21 
sample pairs of spawning fish collections (below diago- 
nal of boxed area in Table 2). The most notable result 
was that Berners08 and 09 spawners were significant- 
ly different from the Sitka Sound and Hoonah Sound 
spawners in all pairwise homogeneity tests of allele 
frequencies. The Berners07 collection of prespawning 
fish was generally homogeneous with all other collec- 
tions, possibly because of small sample size. Berners 
Bay spawning herring and those collected in the win- 
ter from nearby Lynn Canal, were homogeneous, with 
the exception of the early winter collection (Lynn07). 
The two Lynn Canal collections in the early spring 
2008 (Lynn08, a and b) were nearly identical to each 
other and were also highly similar to those collections 
of spawning herring in the nearby Berners08 sample 
taken several months later. 
Among the collections of spawning fish, seven of the 
21 F st estimates were significant before correction. 
Only one pairwise F sr value was significant after cor- 
rection: namely the F ST value for Hoonah Sound and 
Berners08. Berners Bay and Lynn Canal collections, 
with the exception of Lynn07, exhibited an F ST of 0, 
indicating a high level of genetic homogeneity in this 
region over several years of collection. 
Significant P ST values and differences in allele fre- 
quencies among regional groups were more evident after 
spatially homogeneous and temporal collections were 
pooled (Table 3). Collections of herring from Prince Wil- 
liam Sound were significantly different (P= 0.001) from 
both outer-coastal and interior collections in South- 
east Alaska (except for the single collection of Hobart 
Bay). Herring samples from the two outer-coastal loca- 
tions, Sitka Sound and Nichols Bay, were homogeneous 
(P=0.28), but were collectively divergent from interior 
Berners Bay and Lynn Canal collections. 
Genetic distances and the resulting neighbor-joining 
tree generally mirrored the results in Tables 2 and 3, 
however, bootstrap support for the tree was weak. PWS 
grouped together in 60% of the resamplings, Bern- 
ers08, Lynn08a, and Lynn08b grouped together 50% 
of the time, and all other branches grouped less than 
50%. 
