Your server is running PHP version 5.3.29 but this plugin requires at least PHP 5.4. Please run an upgrade.

Toxins in Fresh Water Fish | Michael Spratt

Toxins in Fresh Water Fish

Toxins in Fresh Water Fish  (AERM)

Data analysis and statistical study of pollution and contamination concentrations in fresh water fish in Northern France.


In summary there were 8 different species for a total of 176 fish, caught from 28/mai/91 to 26/oct/95 at 32 different dates, from 5 different rivers (12 different sites).  The rivers were : Chiers (Tetaigne), Meuse (Chooz, Chooz amont, Chooz aval), Moselle (Arancy, Berg, Koenig Amo, Mereville, Pont a Mouson, Uckange,), Rhin (Gambsheim), Sarre (Willerwald).   The fish captured were :

Species No.Indiv Grouped Total







The toxins studied are listed below.  The study included the concentration in grams and as a percentage of fat.












These toxins were analyzied by species in relation to : Sex, size, ……



Interest in using fish as bio-indicators of pollution has caused an increase in work and publication concerning methods and results. A variety of opinions exist concerning most aspects of this research. The following comments are a resume of other author’s observations and are presented to illustrate difference approaches.

I have 3 major observations concerning the experimental fishing data. They are :
1) the collection,
2) measurement methods and
3) the results.
1) Collection – many advise a collection method which is designed to prove (or disprove) a hypothesis. A plan is usually designed with the aid of a statistician who can advise on the number of individuals to capture, there characteristics (species, size, sex, etc.), when and where and what statistical analysis and tests which should be performed.

2) Methods – This is the most controversial subject as it determines the validity of the results. Two techniques which cause much disagreement are A) grouping individuals and B) normalizing for lipid concentrations.

A) grouping individual in order to measure contaminates is a commonly employed practice. Most often used when funds are inadequate for individual measurements. Many authors warn against this practice as they contend it can bias results.

They argue that :

1) it is impossible to completely and evenly mix individuals together.

2) Different sized individuals may not have the same concentration levels or have the contamination distributed evenly throughout the body, as it maybe concentrated in certain organs.

3) This method can eliminate extremes and exceptions.

Certain authors suggest that equal wet weight samples be homogenized individually and then equal size samples are mixed together in order to obtain a mean concentration for the population. They also suggest that sample size should be based upon the individuals and the contaminate studied and that each case requires evaluation before proceeding. In resume they agree that if just a mean is required this method maybe acceptable but if information concerning the variability and precision are required then sampling every individual is necessary.


 B) normalizing for lipid concentrations.

Normalize contaminants based on lipid content is most often used for comparison of different individuals, groups, species for bio accumulation studies. Lipid normalization is usually accomplished by dividing the contaminate concentration by lipid concentration which provides a ratio.

It is assumed that contaminates accumulate in muscles in proportion to tissue lipid content and that this procedure eliminates the influence of lipid covariation.

Several authors indicate that lipid ratios (covariate) are correct for only isometric relations between the toxin and lipid concentration which risk to be the exception not the rule and that individuals who departure from an isometric relation will produce unpredictable and confusing results.

Examples of this problem can be found in C. Hebert et aL article “To Normalize or not to Normalize? Fat is the Question”. In this article several cases were presented which illustrate how normalized lipids can cause interpretation errors.

In the first case they illustrate that an analysis of the lipid and PCB concentration indicate that no relationship exists, thus the use of lipid normalization is unnecessary and that normalizing the data causes an interpretation error.

The second case is more subtle in that a relation exists between the lipid and contaminate concentration and that normalizing the data may eliminate the confusing effects of lipid concentrations which indicated that no interspecific differences existed in mean contaminate concentration. However an evaluation of wet-weight indicated that individuals from both groups with similar wet-weights had different levels of contamination and that normalizing the data did not remove the effect of the lipid concentration. Instead they performed an analysis of the covariance (ANCOVA) which combined the features of regression and ANOVA. Linear regression was used to examine the relationship, which if linear then normalization is necessary if the relation is not linear then it must be transformed to a linear relationship. This approach entails a test for significant differences between the slopes of the regression lines for each species (ANOCOVA interaction term). If there are no significant differences in the slope then a common slope is adjusted to contaminate and lipid concentrations. If there is significant differences then individual regression lines are used to normalize the data but with caution concerning the interpretation of the intercepts. Finally, the test concerning the interspecific differences between adjusted means is performed.

In the example they provide an ANCOVA test indicated that no interspecific differences existed in the slope of their regression lines. The interaction term was removed and the ANOCOVA recalculated which reveled significant relation between species and lipids. They then used the common slope to adjust the contaminate concentration for percentage lipid. The common regression model is fitted to the data and residuals are calculated from the regression which represents the variation in contamination concentrations after the effects of lipids are removed. Any positive or negative residual are re-scaled by adding the grand mean contaminant concentration to each residual. When using this method in the example they provided species A had a higher concentration than species B while the coefficients of variation are lower than the normalized method or wet-weight methods.

A third example was provided an example where there is a positive relation between contaminant and lipid concentration and after normalization no interspecies differences exist. But after normalization the coefficients of variation are greater than in the case of wet-weight and that after normalization the precision of the data has been reduced, thereby reducing the power of statistical tests to detect interspecific differences. However, when using the ANCOVA approach a positive relation exists between the contaminate and the lipid concentration and that there is no interspecific differences in the slope of the regression lines. The interaction term is removed and the ANCOVA recalculated which reveals significant lipid and species effects. A common regression line is used to lipid-adjust the contaminate concentrations which indicates that species A has significantly greater lipid-normalized contaminate concentration and the coefficients of variation are lower than wet-weight or lipid-normalized data.

In conclusion normalizing is necessary when the relation between toxin (i.e. PCB) and lipid concentration has been reviewed and the results justify the procedure. i.e. – percentage (concentration of contaminant / lipid concentration – Testing techniques using Analysis of Covariance (ANCOVA) can remove confusing results by manipulating lipid percentages before lipid normalization. Contaminate and lipid relations exist and vary among, individuals, groups, and animal species. Certain tests (ANOCOVA, ANCOVA, etc.) can add precision to concentration calculations while reducing unknown variations.



3) Results – Due to a lack of scientific rigor concerning data collection and sampling any results and their interpretation must be performed with extreme prudence.

Even the most global results, such as mean contamination concentrations, risk to provide misleading conclusions. Also our evaluation did not include a study of lipid concentration which if performed, could eliminate other sources of confusions.


Finally ….

It is important to perform an exhaustive analysis of data in order to evaluate the data’s full potential. A critically review of the entire collection, sampling and evaluation procedures might reveal that no concrete or valid results can be obtained from the data.

Last updated 08/01/08 05:58:48