Lane et al.: Ontogenetic and temporal variability in the fat content and fatty acid composition of Clupea harengus 
115 
Statistical analysis of total lipid content 
Statistical analysis was conducted by using SPSS, vers. 
16.0 (SPSS, Inc., Chicago, IL) and Plymouth Routines in 
Multivariate Ecological Research (PRIMER 6, Primer-E, 
Ltd., Ivybridge, UK) statistical software and a signifi- 
cance level of 0.05 for each test performed. Error was 
reported as standard deviation. Analysis of covariance 
(ANCOVA) tests were evaluated for equality of vari- 
ance by using Levenes test. Kolmogorov-Smirnov and 
Shapiro-Wilk normality tests were conducted on raw 
total percent lipid data. For comparisons of age classes, 
the age of each fish was estimated from length-at-age 
curves published in the literature (Penttila et al., 1989). 
Corrected Akaike information criteria (AIC c ) scores were 
calculated for linear, quadratic, and cubic distributions 
to determine the best fit for the relationship between 
total lipid content and fish size (Burnham and Ander- 
son, 2002), and the remaining analyses were conducted 
by using the model of best fit. The normalized relative 
likelihood of any model to best represent the data is its 
Akaike weight, w p and lower w ( values indicate better 
model fits. The AIC model w t statistics heavily favored 
the linear model (linear: w t = 0; quadratic: zc ( =0.46; cubic: 
tt/ ( = 0.54), and therefore the linear model was used in 
subsequent regression analyses. 
The relationship between total percent lipid (wet 
weight) and fish fork length over the entire size range of 
fish collected was examined with linear regression. Be- 
cause size had a significant effect on total percent lipid 
(see Results section), the remaining analyses were con- 
ducted by using ANCOVA to account for this covariation. 
The annual, seasonal, and monthly variation in total 
percent lipid content of Atlantic herring was examined 
by ANCOVA. Only samples from 2006-07 were included 
in seasonal analyses because the winter of 2006-07 was 
the only winter in which samples were collected. 
Statistical analysis of fatty add composition 
Fatty acid signatures of individual fish were compared 
by using PRIMER, vers. 6 software (Clarke, 1993; 
Clarke and Warwick, 2001; Clarke and Gorley, 2006). Of 
the suite of 67 fatty acids present in herring, a subset 
of 23 were selected and analyzed to determine whether 
patterns existed in fatty acid signatures in herring of 
different sizes, years, and seasons. These fatty acids 
were 14:0, 16:0, 16:1/7-11, 16:1/7-9, 16:1/7-7, 16:l/z-5, 
18:l/z-ll, 18:1/2-9, 18:2/7-6, 18:3/7-3, 18:4/7-3, 20:1/7-11, 
20:1/7-9, 20:1/7-7, 20:4/7-6, 20:4/7-3, 20:5/7-3, 22:1/7-11, 
22:1/7-9, 22:1/7-7, 22:5/7-3, 22:6/7-3, and 24:1/7-9. These 
fatty acids were chosen if they were present in at least 
95% of the individual fish analyzed. If a particular 
fatty acid was not detected in an individual, the con- 
centration of that fatty acid was changed from zero to 
0.005% because it was below the minimum detectable 
level (0.01%), but it was not so small that it would result 
in extreme outliers (Iverson et al., 2002). Individual 
fatty acids were standardized before analysis by divid- 
ing the value of each fatty acid in each sample by the 
5 10 15 20 25 30 
Fork length (cm) 
Figure 1 
Bivariate scatterplot of total percent lipid by fork length 
for Atlantic herring ( Clupea harengus , n= 889). Although 
the linear r 2 value is low, the relationship is highly 
significant. The low r 2 value may be representative of 
the highly variable nature of the nutritional quality of 
herring in the Bay of Fundy ecosystem. 
standard deviation of that fatty acid in all samples 
and resemblance matrices were created on the basis of 
Bray-Curtis similarity. 
Nonmetric multidimensional scaling (MDS, 25 re- 
starts, Kruskal scheme 1) analyses were conducted on 
the fatty acid profiles of all samples. MDS stress val- 
ues range from 0 to 1. Low stress values indicate high 
confidence in the model, and stress values less than 
0.2 were assumed to adequately represent the relation- 
ships of the samples in the model (Clark and Warwick, 
2001). Analyses of similarities (ANOSIM, one-way, max. 
permutations = 999) were conducted on all samples to 
evaluate the effect of fish age, year, and season on fatty 
acid signatures. Because similar patterns were observed 
when fish were separated first by age and then by year 
or season, all fish were pooled to examine annual and 
seasonal variability in fatty acid signatures to allow for 
higher power in the analyses. ANOSIM global r values 
range from 0 to 1, and higher global r values are more 
significant. One-way similarity percentages analysis 
(SIMPER, one-way, based on Bray-Curtis similarity, 
cut-off percentage=90) was conducted on all samples if 
the analysis of similarities was significant, to determine 
the fatty acids that contributed the most to the differ- 
ences observed between groups. 
Results 
Total lipid content 
A total of 889 individual fish collected between 2005 and 
2008 were analyzed for trends in total lipid content (per- 
centage of wet weight; Table 1). The linear regression of 
