Content uploaded by Ron van der Oost
Author content
All content in this area was uploaded by Ron van der Oost on Sep 04, 2018
Content may be subject to copyright.
Aquatic Toxicology 39 (1997) 45-75
Biomonitoring aquatic pollution with feral eel
(Anguilla anguilla) .
III. Statistical analyses of relationships between
contaminant exposure and biomarkers
Ron van der Oost”a*, Eric Vindimianb, Paul J. van den Brink”,
Karel Satumalay”, Henk Heida”, Nice P.E. Vermeulend
“Department of Environmental Toxicology, OMEGAM Environmental Research Institute,
City of Amsterdam, P. 0. Box 94685, 1090 GR Amsterdam, The Netherlands
‘Direction scien@que et de la q&it& INERIS, Pare Technologique ALATA, Verneuil en Halatte, France
‘DLO Winand Staring Centre for Integrated Land, Soil and Water Research, Wageningen, The Netherlands
‘Department of Pharmacochemistry, Molecular Toxicology Division, Free University, Amsterdam,
The Netheriandr
Accepted 25 November 1996
Abstract
In a large-scale field study, sediments and eel (Anguillu anguillu) samples were collected from
six Amsterdam freshwater sites with varying degrees of pollution. All sediment and eel
samples were analyzed for organic trace pollutants, such as polychlorinated biphenyls
(PCBs), organochlorine pesticides (OCPs) and polycyclic aromatic hydrocarbons (PAHs).
In addition, the pollution-induced responses of a suite of 21 biochemical parameters in eel
(notably phase I and phase II biotransformation enzymes, antioxidant enzymes, PAH me-
tabohtes, DNA adducts and serum transaminases) were measured. The resulting comprehen-
sive database was subjected to statistical analyses in order to determine the biomarkers which
were most suitable to assess inland water pollution and to classify the environmental quality
of the sites.
Bivariate correlation analysis, principal component analysis (PCA) and residual maximum
likelihood analysis (REML) all revealed that the eel tissue levels of most PCB and OCP
analyte groups were suitable to assess exposure to these contaminants, whereas PAH tissue
levels were not. The phase I biotransformation enzymes in eel were found to be the most
responsive to organic pollutants in the environment. Phase II enzymes and cofactors, as well
as DNA adducts, were found to be less sensitive biomarkers, whereas the antioxidant en-
zymes and the serum transaminases did not show statistically significant correlations with
pollutant levels. Similar results were obtained by means of the postulated bivariate correla-
*Corresponding author.
0166-445X/97/$17.00 0 1997 Elsevier Science B.V. All rights reserved.
PIISO166-445X(96)00851-X
46 R. vun der Oost et aLlAquatic Toxicology 39 (1997) 45-75
tion-significance index (CSI) and the multivariate PCA analysis.
Discriminant analysis (DA) was used to classify the pollution status of the various sites. It
appeared that the best discrimination between reference sites, moderately polluted sites and
heavily polluted sites was obtained using DA on data of the nine most responsive biochem-
ical markers. The importance of monitoring biota for the classification of the pollution status
or environmental quality of freshwater sites was demonstrated in the present study, since no
clear discrimination between moderately and heavily polluted sites could be made using
sediment pollutant levels only. The results indicate that biological effect monitoring is the
only appropriate method providing a reliable environmental risk assessment.
0 1997 Elsevier Science B.V.
Keywords: Feral eel (Anguillu anguilla); Organic trace pollutants; Bioaccumulation; Bio-
markers; Uni- and bivariate correlation analyses; REML analysis; Multivariate analyses
(PCA, DA); Environmental risk assessment
1. Introduction
Many Dutch freshwater sites are polluted with organic trace contaminants, such
as polychlorinated biphenyls (PCBs), organochlorine pesticides (OCPs), polycyclic
aromatic hydrocarbons (PAHs) and, to a lesser extent, polychlorinated dioxins
(PCDDs) and dibenzofurans (PCDFs). It is virtually impossible to monitor all
contaminants which form a potential threat to the environment, i.e., both of
anthropogenic origin (predominantly halogenated hydrocarbons) and of natural
origin (heavy metals and most polyaromatic hydrocarbons). To assess the quality
of the aquatic environment, it is, therefore, more appropriate to evaluate the impact
of chemical substances in terms of molecular responses reflecting the potential for
impairment of physiological and biological processes in exposed organisms (Ever-
aarts et al., 1994). Various monitoring techniques may be performed in order to
assess risks of contaminants for organisms and to classify the environmental quality
of ecosystems (Henderson et al., 1989; Everaarts et al., 1994; Van Welie et al.,
1992).
l Environmental monitoring (EM): measuring levels of a selected set of well-
known contaminants in abiotic environmental compartments.
0 Biological monitoring (BM): measuring contaminant levels in biota (bioac-
cumulation).
0 Biological effect monitoring (BEM): determining the critical dose at a target
site or the early adverse alterations that are partly or fully reversible (bio-
markers).
0 Health surveillance (HS) : examining the occurrence of irreversible diseases
or tissue damage in organisms.
Comprehensive data on environmental, biological and biological effect monitor-
ing of inland water pollution in recent years, conducted by our research institutes,
R van der Oost et al.lAquatic Toxicology 39 (1997) 45-75 47
have been presented in earlier papers (Heida et al., 1986, 1987; Van der Oost et al.,
1988, 1991a,b, 1994a,b Van der Oost et al., 1996a,b). The common European eel,
Anguilfa anguilla, has been used for most of these studies. Both the levels of organic
trace pollutants and several molecular and biochemical responses were recently
examined in specimens of sediment and eel from six Amsterdam freshwater sites
with varying degrees of contamination. In part I of the present paper, it was
demonstrated by means of EM and BM studies that persistent hydrophobic chem-
icals, notably PCBs and OCPs, tended to accumulate in feral eel through different
mechanisms: bioconcentration via the aqueous phase and biomagnification via
contaminated food (Van der Oost et al., 1996a). A large variation in biota sediment
accumulation factors (BSAFs) was observed between sites. This indicated that bio-
accumulation not only depends upon type of organism and congener (Van der Oost
et al., 1988), but that site-specific factors should also be taken into account. The
more biodegradable chemicals, like PAHs, were proven to be taken up by eel, but
not to accumulate to high levels in muscle tissues (Van der Oost et al., 1994b). In
part II of the present paper, pollution-induced biochemical responses (biomarkers)
were demonstrated in a BM/BEM study with the same eel as used for the bioaccu-
mulation study (Van der Oost et al., 1996a). A suite of 21 biochemical parameters
in eel (phase I-related enzymes, phase II enzymes and cofactors, antioxidant en-
zymes, PAH metabolites, DNA adducts and serum transaminases) was measured in
order to determine the potential of these parameters as biomarkers for the assess-
ment of inland water pollution. The most important criteria for suitable biomarkers
are an elevated bioaccumulation or a clear biochemical response which can be
related to the exposure to or the effects of toxic xenobiotic compounds in the
environment (Mercier and Robinson, 1993). The study revealed that six hepatic
parameters in the eel were most feasible for this purpose: cytochrome bs (cyt bs)
level, cytochrome P450 1A protein level (CYPlA), ethoxyresorufin-0-deethylase
activity (EROD) and turnover (EROD/P450), UDP glucuronyl transferase activity
(UDPGT) and DNA adduct levels.
Comprehensive relationships between environmental (EM), biological (BM) and
biological effect monitoring (BEM) methods, however, have not yet been studied
extensively. Since it seems most likely that for a reliable risk assessment of aquatic
pollution more than one monitoring method will have to be used, it is important to
determine mutual relationships and deviations between the different monitoring
methods. For the present investigations, 66 eels were caught at six freshwater sites
in and around Amsterdam. Both the levels of organic trace pollutants (Van der
Oost et al., 1996a) and their biochemical responses (Van der Oost et al., 1996b)
were measured in each eel, and the relationships between data from these EM, BM
and BEM studies were investigated, using various statistical analyses.
The most commonly used measure to quantify an association between two var-
iables is the Pearson correlation coefficient, denoted by r (Norusis, 1993). The
absolute value of r, which ranges from 0 (no correlation) to 1 (perfect correlation),
indicates the strength of a linear relationship. Generally, it is useful to examine
correlation coefficients together with scatter plots, since one and the same correla-
tion coefficient can result from very different underlying relationships (Norusis,
48 R van der Oost et al.lAquatic Toxicology 39 (1997) 45-75
1993). For data that do not satisfy the normality assumption, another measure of a
linear relationship between two variables, the Spearman rank correlation coefficient,
is often used (Norusis, 1993). The rank correiation coefficient is the Pearson corre-
lation coefficient based on ranks of data. When interpreting correlation coefficients,
it should not be assumed that correlation automatically implies causation. Since eels
caught at the same site may be intercorrelated with one another, the conventional
Pearson and Spearman statistical tests may be too liberal. The method of residual
maximum likelihood (REML) is a univariate statistical analysis which also accounts
for an unbalanced, so-called “nested”, sampling design. The REML analysis is a
more appropriate statistical method when other site-specific factors may influence
the variables which are tested for mutual relationships (Searle et al., 1992). The
relationships between pollutant levels and biochemical responses in eel of the
present study were also tested with multivariate analyses. A commonly used multi-
variate analysis is the principal component analysis (PCA). PCA can be performed
to characterize the vicinity of different variables used for the characterization of a
set of individuals (Caillez and Pages, 1976). The PCA method has, for example,
been successfully used to identify analyte patterns of PCBs (Storr-Hansen and
Spliid, 1993), PAHs (Zitco, 1993) and PCDD/Fs (Fiedler et al., 1996) in environ-
mental samples. Discriminant analysis (DA) is a PCA on classes of individuals,
which in the present study are represented by eel from different sites. DA is, there-
fore, performed in order to maximize the inter-site variance, thus helping to char-
acterize the differences between the various sites. Recent examples of the use of DA
on environmental quality data concerned the characterization of different stations
along French rivers (Persat et al., 1985) and in a Norwegian fjord (Beyer et al.,
1996), based on the results of biological (effect) monitoring.
The primary aim of the present study was to examine the possible relationships
between environmental (EM), biological (BM) and biological effect monitoring
(BEM) methods, using data on the bioaccumulation of PCB, OCP, and PAH
analyte groups and their biochemical responses in eel. The results of the study
are used to make a selection of the most promising biomarkers for the assessment
of inland water pollution. It is discussed whether or not this selection tends to
support recently reported trends (Van der Oost et al., 1996a,b). Moreover, it is
investigated whether multivariate analyses on the comprehensive dataset of chem-
ical and biochemical parameters may be used to classify the pollution status of the
freshwater environment. A comparison will be made between classifications ob-
tained with different biomarkers and monitoring strategies.
2. Materials and methods
2.1. Sample collection and handling
Sediment and eel samples were collected from six Amsterdam freshwater sites
with different pollution levels in July, 1991. Sampling sites consisted of two rela-
tively clean reference sites, Diemerzeedijk (DZ) and Gaasperplas (GP), three mod-
R. van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 49
erately polluted sites, Lake Nieuwe Meer (NM), Amerika Harbour (AH) and En-
closed IJ (EY) and a heavily polluted site, the Volgermeerpolder (VM). The sam-
pling strategies and handling of the specimens are described in detail in parts I and
II of the present paper (Van der Oost et al., 1996a,b). A total of 60 sediment
samples and 66 eel were used for the different monitoring studies.
2.2. Chemical analyses
All samples of sediments and eel muscle tissues were analyzed for organic trace
pollutants. Since not enough liver tissue was available for both biochemical and
chemical analyses, these tissues were not chemically analyzed. Sediment and eel
muscle tissues were isolated, homogenized, dried and Soxtec@ extracted with hex-
ane-acetone (1 : 1). After column-chromatographic clean-up the PCB, OCP and
PAH levels were determined with GC-ECD and HPLC techniques, as described
in previous papers (Van der Oost et al., 1994b, 1996a). Levels of individual com-
pounds were summed in order to determine total pollutant levels: XPCB.i, COCP
and XPAH. The CPCB.i levels were determined by summing the levels of the
individual PCB congeners 28, 52, 101, 118, 138, 153 and 180. Total PCB levels
(XPCB) were determined by screening the GC profiles and adding up the estimated
levels of all congeners with retention times corresponding with those of a combined
Aroclor mixture (1221, 1242, 1254 and 1260). The PCBs were divided into five
congener groups according to the number of chlorine substituents (PCB 2/3: di-
and trichlorobiphenyls; PCB 4: tetrachlorobiphenyls; PCB 5 : pentachlorobiphen-
yls; PCB 6: hexachlorobiphenyls; PCB 7/S: hepta- and octachlorobiphenyls). The
OCPs were divided into four analyte groups of related compounds (XHCH: a-
HCH, B-HCH and y-HCH (lindane); Zdrins: aldrin, dieldrin, endrin, telodrin
and isodrin ; Zheptaclor : heptachlor, c-heptachlor epoxide, t-heptachlor epoxide;
XDDT: o,p-DDE, p,p-DDE, o,p-DDD, p,p-DDD, o,p-DDT and p,p-DDT) and one
single congener : hexachlorobenzene (HCB). Other OCPs (e.g. endosulfanes) were
virtually undetectable in fish tissues. The PAHs were divided into five analyte
groups according to the number of aromatic rings in the separate congeners
(PAH 2: naphthalene and acenaphthene; PAH 3: acenaphthylene, fluorene, phen-
anthrene and anthracene; PAH 4: fluoranthene, pyrene, benzo[a]anthracene and
chrysene; PAH 5 : benzo[b]fluoranthene, benzo[k]fluoranthene, benzo[a]pyrene
and dibenzo[ah]anthracene ; PAH 6 : indeno[ 1,2,3-cdlpyrene and benzo[ghi]peryl-
ene).
2.3. Biochemical assays
Subcellular fractions of homogenized eel liver tissues were prepared by differ-
ential centrifugation. Microsomal and cytosolic fractions were stored at -80°C
until analyses of the biochemical parameters. The methods used to determine the
levels and activities of phase-I-related enzymes, phase II enzymes and cofactors,
antioxidant enzymes, l-OH pyrene, DNA adducts and serum transaminases have
been described in previous papers (Van der Oost et al., 1994b, 1996b).
50 R. van der Oost el al.lAquatic Toxicology 39 (1997) 45-75
Table I
Relationships (correlation coefficients and significance of the correlation) between mean OM-based lev-
els of PCBs, OCPs and PAHs in sediments (n=6) and mean levels and activities of biomarker en-
zymes and cofactors in eel from six Amsterdam sites (n = 6) and significance of REML correlations be-
tween individual samples of sediments (n = 60) and eel (n = 66)
Phase I-related enzymes
cyt P450 0.83 a I’ 0.66
cyt bs 0.90 a ” 0.89 B
CYPlA 0.66 0.83 B
CYP3A- 0.73 ‘l 0.65
like
EROD 0.64 0.66
PROD 0.70 0.54
EROD/ 0.61 0.49
P450
P450 0.67 ” 0.49
RED
Phase II enzymes and cofactors
GSH -0.35 -0.49
GSSG 0.43 0.20
GSH+ -0.19 -0.20
GSSG
GSH/ -0.23 -0.25
GSSG
GST 0.53 0.25
UDPGT 0.76 ” 0.54
Antioxidant enzymes
GPOX 0.01 -0.09
SOD 0.06 0.42
CAT -0.47 -0.14
PAH parameters
l-OH 0.77 E’ 0.66
pyrene
DNA 0.77 ii 0.89 *’
adducts
Serum transaminases
ALT 0.32 0.49
AST -0.03 0.37
a Significant correlation (p < 0.05).
” Significant correlation @ < 0.01).
’ Analysis not possible.
-
0.90 a ‘I
0.95 rl ‘3
0.97 ‘I ‘I
0.36
0.77
1.00 1’
0.94 h
0.60
0.92” ‘)
0.82 a ‘1
0.90 a ‘1
0.22
0.77 0.95 t’ c
0.83 a 0.98 ” c
0.94 h 0.79 b
0.43 0.71 S’
1 .oo ‘1 0.78 a
0.94 h 0.83 a h
0.94 ‘j 0.74 a
0.54 0.63 ;’
0.83 ’
0.66
0.71
0.37
-0.49
0.89 * ‘I
-0.06
-0.49 -0.33
0.83 ri 0.67
-0.03 -0.06
-0.60 -0.41
0.54 0.75 *
0.94 ” 0.92 a ”
-0.26 -0.08
0.09 -0.06
-0.20 -0.26
0.43 0.77
0.83 a 0.82 a ”
-0.54
0.54
-0.20
-0.77 ‘I -0.60
0.76 *
0.90 a ”
0.54
0.77
-0.09
0.10
-0.03
-0.31
0.14
0.14
0.43
0.72 “
0.60
0.89 a
0.06 -0.31 0.06 0.14
-0.31 -0.43 -0.26 0.09
Sediment PCB Sediment OCP
(organic matter) (organic matter)
____ _
Pearson REML Spearman Pearson REML
Sediment PAH
(organic matter)
Spearman Pearson REML Spearman
The following biochemical parameters were measured : phase I-related enzymes
(phase I): cytochrome P450 (cyt P450), cytochrome bs (cyt bs), cytochrome P450
lA- and P450 3A-like isoenzymes (CYPlA and CYP3A), ethoxyresorufin-O-de-
ethylase (EROD), pentoxyresorufm-0-dealkylase (PROD), EROD turnover
(EROD/P450) and NADPH cytochrome P450 (c) reductase (P450 RED); phase
Table 2
R van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 51
Relationships (correlation coefficients and significance of the correlation) between individual muscle
tissue levels of PCBs, OCPs and PAHs on a lipid weight basis, and levels and activities of biomarker
enzymes and cofactors in eel from six Amsterdam sites (n = 66)
Eel PCB (lipid weight) Eel OCP (lipid weight) Eel PAH (lipid weight)
Pearson REML Spearman Pearson REML Spearman Pearson REML Spearman
Phase I-related enzymes
cyt P450 0.32 a b 0.36 b
cyt bs 0.38 b 0.43 b
CYPlA 0.48 b b 0.42 ’
CYP3A- 0.38 b 0.37 b
like
EROD 0.49 b b 0.47 b
PROD 0.19 0.27 a
ERODl 0.45 h b 0.43 b
P450
P450 0.39 b 0.41 h
RED
Phase II enzymes and cofactors
GSH 0.11 0.11
GSSG 0.02 0.09
GSH+ 0.16 0.19
GSSG
GSHI 0.02 -0.02
GSSG
GST 0.21 0.22
UDPGT 0.60 b b 0.54 b
Antioxidant enzymes
GPOX 0.12 0.02
SOD 0.02 -0.02
CAT -0.16 -0.13
PAH parameters
l-OH 0.25 0.18
pyrene
DNA 0.11 0.10
adducts
Serum transaminases
ALT 0.13 0.05
AST 0.39 b b 0.16
0.32 b b 0.37 b -0.19 -0.12
0.61 b a 0.67 b -0.16 -0.11
0.70 b b 0.64 b -0.09 -0.08
0.36 b 0.41 b -0.11 -0.09
0.67 b b 0.65 b -0.25 H -0.22
0.35 b 0.39 h 0.01 0.02
0.65 b a 0.60 ’ -0.22 -0.23
0.26 a 0.28 a -0.17 -0.12
-0.12 -0.05 0.16 0.13
0.28 a 0.22 -0.13 -0.09
0.04 0.09 0.13 0.10
-0.27 a -0.21 0.16 0.13
0.39 8 0.44 b 0.15 0.16
0.53 b b 0.52 b -0.19 0.15
0.11
-0.01
-0.15
0.07 -0.07
-0.04 -0.21
-0.20 0.01
-0.05
-0.20
-0.01
0.16
0.28 a
0.19 -0.02 0.01
0.29 a -0.27 a -0.23
0.23 0.23 0.10 0.15
0.23 a 0.15 0.16 0.32 a
a Significant correlation (p < 0.05).
” Significant correlation 0, < 0.01).
II enzymes and cofactors (phase II): reduced glutathione (GSH), oxidized gluta-
thione (GSSG), total glutathione (GSH+GSSG), thiol : disulfide ratio (GSH/
GSSG), glutathione-S-transferase (GST) and UDP glucuronyl transferase
(UDPGT); antioxidant enzymes (antiox): glutathione peroxidase (GPOX), super-
oxide dismutase (SOD) and catalase (CAT); PAH parameters: OH pyrene in bile
and DNA adducts in liver tissue; serum transaminases (transamin): alanine trans-
aminase (ALT) and aspartate transaminase (AST) in blood plasma.
52
Table 3
R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75
Relationships (correlation coefficients and significance of the correlation) between individual muscle
tissue levels of PCBs, OCPs and PAHs on a fresh weight basis, and levels and activities of biomarker
enzymes and cofactors in eel from six Amsterdam sites (n = 66)
Eel PCB (fresh weight) Eel OCP (fresh weight) Eel PAH (fresh weight)
Pearson REML Spearman Pearson REML Spearman Pearson REML Spearman
Phase I-related enzyme.r
cyt P450 0.24 * 0.22
cyt bs 0.49 ” 0.50 h
CYPlA 0.53 ” 0.50 Ii
CYP3A- 0.16 0.22
like
EROD 0.62 ” 0.56 ”
PROD 0.23 0.21
EROD/ 0.61 ” 0.58 ”
P450
P450 0.25 0.28 ”
RED
Phase II enzymes and cojktors
GSH 0.05 0.07
GSSG 0.31 it 0.26 R
GSH+ 0.19 0.19
GSSG
GSHI -0.23 -0.20
GSSG
GST 0.10 0.12
UDPGT 0.40 h iL 0.37 ‘I
Antioxidant enzymes
GPOX 0.12 0.13
SOD 0.15 0.14
CAT -0.07 -0.13
PAH purumerers
1 -OH 0.08 0.09
pyrene
DNA 0.31 ;’ 0.37 1’
adducts
Serum rrunsaminases
ALT 0.03 0.02
AST 0.01 -0.15
a Significant correlation (p < 0.05).
h Significant correlation @ < 0.01).
0.25” ”
0.63” ”
0.68 ”
0.20
0.70 ”
0.33 ”
0.70 ‘)
0.18 0.19 -0.09 -0.14
-0.11
0.4 1 ” ”
0.08
-0.39 ‘a ii
0.28 a
0.39 r) ;’
0.10
0.09
-0.09
0.06
0.37 ”
0.14
-0.01
0.21 -0.04
0.64 k’ 0.14
0.61 ” 0.19
0.32 * -0.26 *
0.60 b 0.16
0.34 h 0.20
0.61 ” 0.18
-0.06 0.08
0.32 a 0.14
0.08 0.16
-0.31 a -0.08
0.32 a 0.03
0.37 h -0.27 a h
0.11 0.13
0.06 0.04
-0.16 -0.02
0.17 -0.22
0.46 ” 0.01
0.19 -0.09
-0.06 -0.19
-0.11
0.15
0.20
-0.23
0.11
0.12
0.16
0.04
0.15
0.10
-0.14
0.09
-0.13
0.09
0.01
0.03
-0.06
0.08
-0.04
-0.09
2.4. Vni- and bivariate statistical analyses
2.4.1. Test of’ normal distribution
The Lilliefors test, based on a modification of the Kolmogorov-Smirnov test,
was used to determine whether data were normally distributed.
R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75 53
2.4.2. Correlations and correlation-signljicance index (CSI)
Correlations between the chemical and biochemical parameters were determined
with both the Pearson correlation coefficient on log-transformed data and the
Spearman rank correlation coefficient on the original data. Significance of the
correlations was determined with a two-tailed Student’s t test for significance. As-
says were performed with SPSS@ 6.0 statistical software (Norusis, 1993). In order
to visualize the individual relationships between tissue pollutant levels and bio-
chemical parameters, a correlation-significance index (CSI) was postulated. The
CSI value is defined as:
CSI=CC+SF (1)
Here, CC refers to the absolute value of the Spearman rank correlation coeffi-
cient (0 < CC < 1) and SF to the significance factor of the correlation. The SF
value depends upon the two-tailed significance of the correlation:
0 SF = 0.0, no significance (p > 0.05);
0 SF = 0.5, significance (0.05 >p > 0.01);
l SF = 1 .O, strong significance (p < 0.01).
Consequently, the CSI value ranges from 0 (no correlation) to 2 (perfect corre-
lation).
2.4.3. Residual maximum likelihood (REML) analysis
Because the experimental design is hierarchical, which in this case means that
two distinct levels (location and eel-within-location) have to be analyzed, a multi-
level model (Goldstein, 1995) is called for. Such models have several variance
components (Searle et al., 1992). Biomarkers were, therefore, regressed against
pollutant concentrations using a regression model with two variance components.
Calculations were carried out on log-transformed data. The model is fitted using
the method of residual maximum likelihood (REML). The REML analyses were
carried out using GENSTAT 5 version 3.1. (GENSTAT 5 Committee, 1993). For
testing the statistical significance of the relation between biomarkers and pollu-
tion, the Wald test was used. Asymptotically, the Wald statistic has a x2 distribu-
tion (GENSTAT 5 Committee, 1993; Fowler and Cohen, 1990). More informa-
tion on the use of REML is presented in GENSTAT 5 Committee (1993) and in
Goldstein (1995) and Baar and ter Braak (1996).
2.5. Multivariate statistical analyses
2.5.1. Principal component analysis (PCA) and discriminant analysis (DA)
PCA was used to characterize the correlations between almost all variables
used in the present study. Except for the l-OH pyrene bile levels and eight of the
66 eel, which were discarded due to missing values, the whole biomonitoring da-
R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75
A. Phase i-related enzymes
.,’
2.0 I ”
& 2:
f3. Phase II enzymes and cofactors
C. Antioxidant enzymes
7
2.0 1’ .A ‘1
,.5!’ _:
-;I:-( i
csI1.ol'
; PAHS
n PCf3s /
~ n OCPS I
R van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 55
Fig. 1. Correlation-significance indices (CSIs) to visualize relationships between individual biomarkers
((A) Phase I-related enzymes, (B) phase II enzymes and cofactors and (C) antioxidant enzymes) and
groups of fresh-weight-based tissue levels of organic trace pollutants in eel from six Amsterdam sites.
Abbreviations are explained in Section 2.
+
taset was subjected to principal component analysis (PCA). Consequently, an ar-
ray of 58 rows, representing the individual eel, and 46 variables, being 20 bio-
chemical parameters, 19 tissue analyte residue groups and 7 physiological and
morphological parameters of eel, was considered for the multivariate analyses.
All variables were log-transformed in order to minimize the impact of outliers. In
order to prevent certain variables from having more influence on the PCA than
others, due to scaling effects, normalization of the data was necessary. Therefore,
variables were centered to obtain a zero average and normalized to obtain a var-
iance of 1.
DA was carried out in order to classify the different sampling sites according
to the monitoring results. DA allows the separation of the different sites using
factors, or discriminating functions, as linear combinations of the original varia-
bles. The discriminated stations were checked for significance at the 0.01 level us-
ing a Student’s t test on the individual scores for the two first discriminating
functions. The correlation of the variables with the factors was used to character-
ize the importance of different variables for classification. DA uses PCA as a first
step in order to use orthogonal variables before maximizing the inter-class var-
iance.
In both PCA and DA analyses, factors are calculated as the eigenvectors of a
matrix using individuals as rows and variables as columns. The factors are or-
thogonal and ordered using decreasing eigenvalues or inertia. The inertia associ-
ated with each factor is a fraction of the total variance and can be expressed as
a percentage. The results are presented using graphs of the first two factors Fl
and F2, thus covering the largest fraction of variance that can be shown within
two dimensions. PCA, DA calculations and graphical representations were per-
formed using ADE software running on an Apple Macintosh@ computer. ADE
is provided by the laboratory “Ecologic des eaux deuces et des grandes fleuves”
from the French national research organization CNRS (Chessel and Doledec,
1992).
3. Results
All levels of organic trace pollutants (i.e., PCBs, OCPs, PAHs, PCDFs and
PCDDs) in both sediments and eel muscle tissues, as well as the levels and activities
of a suite of 21 biochemical markers (notably phase I and phase II biotransforma-
tion enzymes, antioxidant enzymes, PAH metabolites, DNA adducts and serum
transaminases; see Tables l-3), determined in samples from six Amsterdam fresh-
water sites, have been presented in previous papers (Van der Oost et al., 1994b,
R. van der Oost et ul.lAquatic Toxicology 39 (1997) 45-75
A. PAHs ,1 ~,
2.00 ,’ : *. /
1.50 1 ! i
6 PAH
tot
c. OCPS
I
_y
2.00 ; ; !
&,’ i
1_ 1 anti-oxidant
tot OCP
tot
R. van der OOSI et aLlAquatic Toxicology 39 (1997) 45-75 57
Fig. 2. Correlation-significance indices (CSIs) to visualize relationships between individual fresh-
weight-based tissue levels of organic trace pollutants ((A) PAHs, (B) PCBs and (C) OCPs) and groups
of biomarkers in eel from six Amsterdam sites. Abbreviations are explained in Section 2.
t
1996a,b). This comprehensive dataset was subjected to a series of statistical analyses
in order to determine relationships between the data obtained by various monitor-
ing methods. Data on polychlorinated dioxins (PCDDs) and dibenzofurans
(PCDFs) were discarded from the present statistical analyses since they were only
analyzed in eel from two sites, the Diemerzeedijk reference site and most polluted
Volgermeerpolder site (Van der Oost et al., 1996a). Neither the liver somatic index
(LSI) nor the condition factor (CF) differed significantly between eel originating
from the different sites.
No significant differences were found in the water temperature of the sites (ap-
proximately 2O”C), while seasonal variations were excluded since all specimens were
collected in the summer, within a period of three weeks. All eel specimens inves-
tigated were immature (most likely female) fish of the same age range (seven to ten
years). Therefore, the investigated eel appeared to be suitable for comparative
research (Van der Oost et al., 1996a,b).
3. I. Correlations and correlation-sign$cance index (CSI)
The results of the univariate correlation analyses between PCB, OCP and PAH
pollutant levels in organic matter of sediments and biochemical parameters in eel
are presented in Table 1. Since both the Pearson and Spearman correlation coef-
ficients range between - 1 and +l, the interpretation is the same for these coeffi-
cients, except that the Spearman coefficient examines the relationships between
ranks and not values (Norusis, 1993). All datasets were tested for normal distribu-
tion with a Lilliefors assay. With the exception of cyt P450, P450 RED and GST,
none of the datasets were distributed normally. Therefore, the dataset was log-
transformed before examining it with Pearson correlation coefficients and REML
analysis. No correlation coefficients but only significance levels of the relationships
are determined with the quantitative REML analysis. The hypothesis of linearity
for the Pearson and Spear-man relationships was tested for significance with a two-
tailed Student’s t assay, while the statistical significance of the correlations exam-
ined by REML analysis was assessed by the Wald test. It is notable that the most
frequent significant correlations were found between total OCP sediment levels and
phase I-related enzymes (particularly cyt bg, CYPIA, EROD, PROD and EROD/
P450). It is remarkable that REML analyses revealed significant correlations be-
tween cyt P450 levels and PCB and OCP sediment levels, although cyt P450 levels
were only slightly elevated in eel from the polluted sites (Van der Oost et al.,
1996b). Significant correlations were also observed between sediment OCP levels
and phase II cofactors (GSSG) and enzymes (UDPGT). In addition, REML anal-
ysis showed significant correlations between OCP levels and GSH/GSSG ratio and
GST activity. The sediment PAH levels correlated significantly with phase I-related
58 R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75
enzymes (notably cyt b:, and CYPIA). REML analysis revealed more significant
correlations between PAH levels and (mainly phase I) biomarkers than both of the
conventional methods, which may be partly due to the fact that 66 individual
R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75 59
Fig. 3. Representation of the variables on the first factorial plane obtained from principal component
analysis (PCA) on a dataset of 58 eels from six Amsterdam sites, consisting of biomarkers (phase I,
phase II, antiox, DNA adducts and transamin), physiological and morphological data (physlmor) and
tissue pollutant levels (PCB, OCP and PAH) on either (A) a fresh weight or (B) a lipid weight basis.
Circles with bold names represent the centers of gravity of variable groups (white: biomarkers; grey:
physical/morphological parameters; black: pollutant tissue levels). Abbreviations are explained in Sec-
tion 2.
t
sediment samples could be used in the REML analysis instead of the 6 mean sedi-
ment and eel tissue levels that were used in the other correlation analyses. PCB
sediment levels correlated only weakly with biochemical responses.
The correlations between tissue pollutant levels and biochemical responses could
be determined with individual eel values, so that 66 observations could be tested for
each parameter. The relationships between the lipid-weight-based eel tissue levels of
organic trace pollutants and biochemical parameters are presented in Table 2. Most
of the statistically significant correlations were observed between total PCB and
OCP tissue levels on the one hand and phase-I-related enzymes on the other. Con-
taminant levels in eel tissues did not correlate with phase II cofactor levels, but
statistically significant correlations were observed with phase II enzyme activities:
UDPGT vs. PCB and OCP, and GST vs. OCP. Virtually no statistically significant
correlations were observed between the lipid-weight-based PAH tissue levels and
any of the biochemical parameters. The relationships between fresh-weight-based
eel tissue levels of organic contaminants and biochemical parameters are presented
in Table 3. Again, most of the significant correlations were observed between total
PCB and OCP tissue levels on the one hand and phase I-related enzymes on the
other. Statistically significant correlations were also observed between both PCB
and OCP levels, and GSSG, UDPGT and DNA adducts. No statistically significant
Fig. 4. Representation of six Amsterdam sites in the first factorial plane obtained from discriminant
analysis (DA) on all pollutant analyte group levels in organic matter of sediments. Ellipses include
90% of the individual sediment samples.
60 R. van der Oost et aLlAquatic Toxicology 39 (1997) 45-75
2.9
El
1.7 3.0
-1.4
A
B
C
Sk:
Die-ijk (DZ)
0
0 Lab3 Gaaperpla5 (GP)
Lab Nlewe blew (NM)
0 Amerika Harbaur (AH)
0 Enclosed rter IJ (EIJ)
0 w-%wmeerpolder P/M)
R. van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 61
Fig, 5. Representation of six Amsterdam sites in the first factorial plane obtained from discriminant
analysis (DA) on pollutant levels in eel muscle tissues: (A) all 19 analyte groups on a fresh weight ba-
sis, (B) all 19 analyte groups on a lipid weight basis and (C) a limited dataset of 10 analyte groups on
a fresh weight basis. Ellipses include 90% of the individual eel samples.
c
correlations were observed between the fresh-weight-based PAH tissue levels and
the biochemical parameters in the eel. There is, however, an important discrepancy
between the conventional correlation analyses (Pearson and Spearman) and the
REML analyses. REML analysis demonstrated significant correlations between
lipid-normalized tissue levels and those biomarkers which showed a clear response
to pollution (Table 2; e.g. CYPlA, EROD, EROD/P450, UDPGT; Van der Oost et
al., 1996b), while correlations between fresh-weight-based pollutant levels and these
biomarkers (Table 3) were less pronounced. The Pearson and Spearman assays
demonstrated significant correlations between fresh-weight-based pollutant levels
and the biomarkers which were most sensitive to contaminant stress (e.g. cyt bg,
CYPlA, EROD, EROD/P450 and UDPGT), while lipid-normalized tissue levels
also correlated with the biomarkers which did not show a clear pollution-related
response (e.g. CYP3A-like, PROD, P450 RED, AST; Van der Oost et al., 1996b).
The correlation-significance indices (CSIs) between all of the biochemical param-
eters and the fresh-weight-based levels of pollutants in eel muscle tissue were de-
termined, using both the Spearman correlation coefficients and their levels of sig-
nificance (formula (1) Section 2). The mean CSI values between individual
biomarkers and all congener groups of accumulated pollutants are displayed in
Fig. 1. With the CSI method it could be visualized that a number of phase I
enzymes (notably cyt bs, cyt P450 lA, EROD and EROD/P450) correlated strongly
with PCB and OCP tissue levels, while weaker relationships were observed between
PCB levels and P450 RED as well as between OCP levels and PROD (Fig. IA).
Relationships between phase II parameters and pollutant levels were less pro-
nounced, although UDPGT correlated with PCB and OCP tissue levels, whereas
weak correlations existed between OCP levels and GSSG, GSH/GSSG and GST
(Fig. 1B). No significant correlations were observed between any of the antioxidant
enzymes and pollutant levels (Fig. 1C).
The mean CSI values between separate congener groups and all biochemical
parameters studied are presented in Fig. 2. Virtually no correlations were found
between PAH tissue levels and biochemical parameters (Fig. 2A). Accumulated
levels of PCB congeners correlated strongly with phase I-related enzymes, while
weaker correlations were observed with phase II enzymes and cofactors (Fig. 2B).
As compared to the other PCB congener groups, di- and tri-CBs did not correlate
well with any of the biomarker enzymes. With the exception of the Zdrins and the
heptachlor congeners, OCP tissue levels correlated well with phase I-related en-
zymes and, to a lesser extent, with phase II enzymes and cofactors (Fig. 2C).
62 R. van der Oost et ul.lAyuatic Toxicology 39 (1997) 45-75
A
B
C
Sites:
5 Diimerzeedijk (IX)
0 take GaasperproS WI
Lake Nleuwe Meer (NM)
herika Harbour (AH)
0 Enclosed river IJ (ELI)
Vobermeerpolder [VM)
R van der Oost et al.lAquatic Toxicology 39 (1997) 45-75 63
Fig. 6. Representation of six Amsterdam sites in the first factorial plane obtained from discriminant
analysis (DA) on biochemical parameters in eel: (A) all 20 biomarkers, (B) a limited dataset of 6 bio-
markers and (C) a limited dataset of 9 biomarkers. Ellipses include 90% of the individual eel samples.
t
3.2. Principal component analysis (PCA)
The biomonitoring dataset was also subjected to principal component analysis
(PCA) in order to characterize the relationships between different parameters and
their variations in all sites. In Fig. 3, the results of PCA are shown by the position
of all variables on the first factorial plane (factors Fl and F2). The first factorial
plane is a two-dimensional projection of the multi-dimensional hyper-cube of all
variables included in the PCA analysis (i.e., biochemical parameters, tissue pollu-
tant residues, physiological/morphological parameters). PCA was performed on
datasets of tissue pollutant levels on the basis of both fresh weight (Fig. 3A) and
lipid weight (Fig. 3B). The first factors (Fl) corresponded to 35% and 26% of the
global variance of the matrix and the first planes (Fl and F2) to 44% and 41%, for
fresh weight and lipid-weight-based parameters, respectively. High correlation co-
efficients between variables and axes are indicative of a good representation of these
variables with PCA. For fresh-weight-based tissue levels, Fl is correlated with all
PCB congener groups, HCB, CHCH, XDDT, XOCP and to a lesser extent with
PAH 5 and PAH 6. Very strong correlations were found between Fl and some
biomarkers (i.e., cyt bs, CYPlA, EROD and EROD/P450 and, to a lesser extent,
GSSG, UDPGT and DNA adducts), while an inverse correlation was observed
between Fl and GSH/GSSG. Since the tissue pollutant levels and biomarkers
that are most relevant for pollution assessment both appear in the same region
of the graph they are most probably correlated with one another. Fl is also corre-
lated with fish weight, fish length, liver weight, MIC protein and muscle lipids. F2
shows a very strong negative correlation with muscle lipids. Lipid normalization of
tissue residues led to a weaker correlation with Fl, especially for PCBs and PAHs
(Fig. 3B). As a consequence, relationships between phase I-related biomarkers and
lipid-weight-based tissue levels were found to be less pronounced then those ob-
tained with fresh-weight-based tissue levels. A strong correlation, however, was
observed between UDPGT activity and lipid-weight-based PCB and OCP levels
(Fig. 3B), which was in line with the univariate correlation analyses (Table 2).
3.3. Discriminant analysis (DA)
Discriminant analysis was carried out with different datasets in order to compare
the site classification based upon the various monitoring methods. The datasets
subjected to DA analysis were all analyte group levels in sediments (Fig. 4) and
in eel tissue (Figs. 5A and 5B), selected analyte group levels in eel tissue (PCB 4,
PCB 5, PCB 6, PCB7/8, XPCB, XPCB.i, HCB, XHCH, CDDT and COCP; Fig.
5C), all biochemical parameters in eel (Fig. 6A) and selections of six (cyt bg,
CYPlA, EROD, EROD/P450, UDPGT, DNA adducts; Fig. 6B) and nine (cyt
64
Table 4
R. van der Oost et allAquatic Toxicology 39 (1997) 45G75
Comparison of the results of environmental monitoring (EM), biological monitoring (BM) and biolog-
ical effect monitoring (BEM) on the classification of six freshwater sites
Sediment pollutant levels
18 organic matter levels
Eel pollutant levels (BM)
19 fresh weight levels
Figure
(EMi
Fig. 4
Fig. 5A
19 lipid weight levels Fig. 5B
10 fresh weight levels Fig. 5C
Eel biochemical parameters (BEM)
19 biomarkers Fig. 6A
6 biomarkers
9 biomarkers
Fig. 6B
Fig. 6C
Distinct groups
0, = 0.05) C’
_ ._
Distinct groups
o,=O.Ol) h
1: DZ
2: GP
3: AH
4: EY
5: NM+VM
1: DZ
2: GP
3: AH+EY
4: NM+VM
1: DZ+GP 1: DZ+GP
2: AH+EY 2: AH+EY
3: NM 3: NM
4: VM 4: VM
1: DZ+GP I: DZ+GP
2: AH+EY 2: AH+EY
3: NM 3: NM
4: VM 4: VM
1: DZ+GP 1: DZ+GP
2: AH+EY 2: AH+EY
3: NM 3: NM
4: VM 4: VM
1: DZ
2: GP
3: NM
4: AH+EY
5: VM
1: DZ
2: GP
3: NM
4: AH
5: EY
6: VM
1: DZ
2: GP
3: NM
4: AH
5: EY
6: VM
_~ ..___~
I: DZ
2: GP
3: NM
4: AH+EY
5: VM
I: DZ
2: GP+NM
3: AH
4: EY
5: VM
1: DZ
2: GP
3: NM
4: AH+EY
5: VM
Interclass/total
variance ratio
a Significantly different sites, using Student’s t test with p=O.O5 on DA data.
h Significantly different sites, using Student’s t test with p = 0.01 on DA data.
17%
bs, CYPlA, EROD, EROD/P450, UDPGT, GSSG, GSH/GSSG, GST and DNA
adducts; Fig. 6C) relevant biomarkers in eel. The results of the DA analyses are
presented in graphs displaying the individual eels or sediment samples, connected
with their site values. Each mean site value is presented as the center of gravity of
all individuals from the same site. The ellipses of inertia, centered around the mean
Table 5
R. van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 65
Correlation coefficients between initial biomarker variables and the two first factors from discriminant
analysis (DA) on the complete dataset and on two limited datasets
19 biomarkers 9 biomarkers 6 biomarkers -
Fl F2 Fl F2 Fl F2
Phase I-related enzymes
cyt P450
cyt bs
CYPlA
CYP3A-like
EROD
PROD
ERODIP450
P450 RED
Phase II enzymes and cofactors
GSH
GSSG
GSH+GSSG
GSH/GSSG
GST
UDPGT
Antioxidanf enzymes
GPOX
SOD
CAT
PAH parameters
1 -OH pyrene
DNA adducts
Serum transaminases
ALT
AST
-0.28 0.13
-0.66 a 0.32
-0.88
0.00
-0.90
-0.45
-0.90
0.09
-0.14
0.70
0.09
0.23
-0.13
0.52
0.25 0.10
0.50 -0.12
0.04 0.04
0.45 0.10
-0.28 0.11
0.38 0.23
0.05 0.02
-0.19 -0.18
0.14 -0.31
b
-0.47
b
0.37
-0.09
0.07
0.03
-0.01
-0.72
-0.90
_
-0.91
_
-0.90
-0.52
_
0.46
-0.34
-0.42
_
_
_
-0.51
_
_
0.26 -0.71
-0.24 -0.87
_
0.02 -0.92
_ _
-0.26 -0.93
_ _
_ _
-0.20 -
_ _
0.13 ~
-0.02
0.14 -0.42
_
_
_
_ _
0.47 -0.56
_ _
_
_
-0.64
0.14
_
-0.36
_
-0.34
_
_
_
_
_
-0.33
_
_
_
0.62
_
R Bold values are correlation coefficients with absolute values above 0.5.
h l-OH pyrene was not included in the multivariate analyses because of to many missing values.
site values, contain about 90% of the individuals to illustrate a virtual border
between the sites. A comparison of the discriminating potencies of different param-
eters can be made by observing Figs. 46. The results of all DA analyses on distinct
datasets, showing the distinct groups obtained after discrimination, are summarized
in Table 4.
A clear discrimination between reference and polluted sites was observed in the
first factorial plane (Fl-F2) of DA on sediment pollutant levels (Fig. 4). Analyses
of the DA results with Student’s t test revealed 5 distinct groups that were signifi-
cantly different from each other (Table 4). No significant discrimination, however,
was observed between the moderately polluted NM and the heavily polluted VM
sites.
When DA was carried out on eel tissue pollutant levels, similar graphs were
obtained with fresh weight and lipid-normalized levels (Figs. 5A and 5B). A clear
66 R. van der Oost et al.lAquatic Toxicology 39 (1997) 45-75
ured PCB tot 1
GP NM AH EIJ VM
~ 2000
5
t
s
t” 1500 I-$!zzz]
P
E
c
0 1000
0
jj
500 I
B
Fig. 7. Comparison between measured analyte group tissue levels in eel (on a fresh weight basis) and
those predicted by means of multiple regression on factors from PCA on a limited dataset of 9 bio-
markers: (A) total PCB levels and (B) total OCP levels. The individual eel are presented per site on
the abscissa. Abbreviations of sites are explained in Section 2.
discrimination was observed between the heavily polluted VM site and all other
sites. No clear discrimination, however, was observed between the reference sites
and the moderately polluted AH and EY sites, although analyses with Student’s t
test still revealed a significant difference between the two reference sites and the
moderately polluted AH and EY sites. DA on a restricted dataset of 10 fresh-
weight-based eel tissue pollutant levels (without PCB 213 and all PAH analyte
groups) did not reveal a better discrimination among sites (Fig. 5C; Table 4).
A clear discrimination between the reference sites, the moderately polluted sites
and the heavily polluted site was observed when DA was run on the complete set of
biomarkers (Fig. 6A). Except for the moderately polluted AH and EY sites, all sites
were significantly different from each other (Table 4). The individual variances
within sites increased when DA was performed on the above-mentioned datasets
R. van der Oost et al.lAyuatic Toxicology 39 (1997) 45-75 67
restricted to six or nine relevant biomarkers (Figs. 6B and 6C), although reference
sites, moderately polluted sites and the heavily polluted site were still separated
from each other. The discriminating potency increased when DA was performed
on both the restricted datasets, since all six sites appeared to be significantly differ-
ent from one another (Table 4), although the graphs suggest otherwise (Fig. 6).
Among the correlations between the first factorial axes (Fl and F2) of DA and
the complete set of biochemical parameters and the limited datasets of the nine and
six most relevant biomarkers, the highest correlations were observed between the Fl
axis and the CYPlA-related biomarkers (CYPlA, EROD and EROD/P450; Table
5). No discrepancies were found between the discriminant parameters of these three
biomarker datasets. The best discriminating potencies were observed when DA was
performed on the limited dataset of nine biomarkers. The absolute values of the
correlation coefficients between factor Fl of the DA and six of the nine parameters
investigated (cyt bs, CYPlA, EROD, EROD/P450, GSSG and DNA adducts) were
higher than 0.5 (Table 5). The activity of UDPGT, which was found to be relevant
by PCA analysis (especially when lipid-normalized residues were used), showed a
relatively weak discriminating potency in DA analysis (Table 5).
4. Discussion
In the present study, uni-, bi- and multivariate statistical relationships were ex-
amined between the bioaccumulation of PCB, OCP, and PAH congener groups in
eel and biochemical responses associated with these pollutants. The bioaccumula-
tion of organochlorines and polyaromatic hydrocarbons in eel from six Amsterdam
freshwater sites with different pollution levels was discussed in two previous papers
(Van der Oost et al., 1996a and Van der Oost et al., 1994b, respectively). Twenty-
one parameters of biochemical responses, possibly induced by these pollutants, were
measured in the same eel (Van der Oost et al., 1996b). The primary aim of the
present study was to compare the various monitoring methods in order to make a
selection of the most suitable biomarkers for the assessment of inland water pollu-
tion, possibly in support of previously reported trends (Van der Oost et al., 1994b,
1996a,b). In addition, multivariate analyses on the datasets of chemical and bio-
chemical parameters was used to classify the pollution status of the different fresh-
water sites.
4.1. Relationships between bioaccumulation and biomarkers
In the present study, various statistical methods were used in order to examine
possible relationships between the levels of organic trace pollutants (in both sedi-
ments and muscle tissue of feral eel) and a suite of biochemical parameters in the
eel. Univariate correlations between analyte levels and biomarkers were determined
with Pearson and Spearman correlation coefficients, and by REML analyses. Sim-
ilar trends were found when the three methods were used to determine the relation-
ships between environmental, bioaccumulation and biomarker data of the present
68 R. vun der Oost et al.lAyuatic Toxicology 39 (1997) 45-75
study (Tables l-3). The REML analysis, however, is assumed to be the most
suitable method for determining correlations between parameters in an unbalanced
sampling design, like the one in the present study, since the other two analyses
might be too liberal in terms of identifying significant correlations (Searle et al.,
1992). Most of the statistically significant correlations were found between the
phase I enzymes and OCPs and PAHs in sediments (Table 1) on the one hand,
and the PCBs and OCPs in eel muscle tissue (Tables 2 and 3) on the other. How-
ever, care should be exercised when interpreting significance levels of correlation
analyses on large datasets, such as in the present study. Even if there is no causal
association between the variables, some of them would be expected to correlate
significantly by chance alone (Norusis, 1993).
Spearman rank correlation coefficients and their levels of significance were used
to determine the “correlation-significance indices” (CSIs) of bivariate relationships
between the various biochemical parameters and fresh-weight-based tissue levels of
organic pollutants measured in these studies. The mean CSI values were used to
visualize the correlations between individual eel biomarkers and summed tissue
pollutant levels (Fig. 1) and between individual levels of analyte groups in eel tissue
and summed biomarker groups (Fig. 2). The graphs clearly illustrate statistically
significant relationships between several biomarkers (notably cyt b5, CYPlA,
EROD, EROD/P450, UDPGT) and tissue pollutant levels (mainly PCB and OCP
analyte groups). No statistically significant correlations were observed between
PAH tissue levels and biomarkers, nor between the antioxidant enzyme activities
and tissue pollutant levels. The lack of correlation between muscle tissue levels of
PAHs and the measured biochemical responses is not surprising considering the
extensive metabolism of parental PAHs to nonparent compounds in teleost fish
(Varanasi et al., 1985). In an earlier study it was demonstrated that PAH muscle
tissue levels in eel did not reflect the exposure to these compounds (Van der Oost et
al., 1994b). Although parental PAHs hardly accumulate in eel tissues they may be
partly responsible for the observed biochemical effects.
The CSI results are consistent with the results of the PCA analysis (Fig. 3A),
since variables that appear in the same region of the graph (e.g. PCBs, OCPs, cyt
bg, CYPlA, EROD and EROD/P450) are linearly correlated. The close relation-
ships between biomarkers and pollutant levels in eel could also be illustrated when
measured levels of PCBs (Fig. 7A) and OCPs (Fig. 7B) were compared to those
predicted by means of a multiple regression on the factors of PCA performed on a
limited dataset with nine of the most relevant biomarkers (cyt bg, CYPlA, EROD,
EROD/P450, UDPGT, GSSG, GSHIGSSG, GST and DNA adducts). A good
similarity was observed between predicted and measured values of both PCBs
and OCPs. These results show that the use of these biomarkers can indeed be a
substitute for chemical residue analysis and can provide a semiquantitative indica-
tion of the bioaccumulation of persistent organochlorine pollutants.
With regard to the chemical residues in fish tissues, a comparison can be made
between lipid normalized and fresh-weight-based tissue levels. The lipid content of
eel muscle tissue was clearly correlated with both factors in the first factorial plane
of the PCA (Fig. 3A), thus moving lipid normalized values far from the fresh-
R. van der Oost et aLlAquatic Toxicology 39 (1997) 45-75 69
weight-based values on the opposite side of the Fl-F2 plane. Recently, Hebert and
Keenleyside (1995) recommended that tissue residue levels should be lipid normal-
ized only when contaminant concentrations vary in direct proportion to lipid con-
tents. This was the case in the present study, since the highest lipid contents were
observed in eel from the heavily polluted VM site, where the largest eel were caught.
However, the multivariate analysis allowed us to examine both fresh-weight-based
values and lipid-normalized values. As far as classification of the sites is concerned,
this led to very similar results (Figs. 5A and 5B; Table 4). Lipid normalization
reduces the discriminating potency of PAHs and of the majority of PCBs, when
compared with a DA run on the respective fresh-weight-based tissue levels. Since
the effects of the size of fish, caught at different sites, are reduced by lipid normal-
ization, this might be an argument to use lipid-normalized rather than fresh-weight-
based tissues levels in bioaccumulation studies. On the other hand, fresh-weight-
based tissue levels seem to correlate better with biochemical responses induced by
hydrophobic organic pollutants, although this was not confirmed with REML anal-
ysis in the present study (Tables 2 and 3).
4.2. Suitable biomarkers to assess inland water pollution
Generally, a biomarker is suitable to assess inland water pollution if it has a
sensitive and selective response which is related to the exposure to or the effects of
(a certain class of) environmental pollutants (Mercier and Robinson, 1993). Bio-
markers can facilitate quantitative environmental risk assessment, in particular by
increasing understanding of factors contributing to inter-individual variability in
response to toxic xenobiotics (Perera and Whyatt, 1994). In the present study, a
relatively high correlation between accumulated chemical residues in eel and various
biochemical parameters was demonstrated. It confirmed the fact that inducible
phase I enzymes are responsive to organic trace pollutants (Vindimian and Garric,
1989; Vindimian et al., 1993; Stegeman and Hahn, 1994; Van der Oost et al.,
1996b). Phase II enzymes and other biomarkers were found to be less suited to
assess this particular type of aquatic pollution (George, 1994), which was in line
with the trends observed in part II of the present study (Van der Oost et al., 1996b).
Physical and morphological parameters correlated well with factor Fl of the PCA,
but this may be due to the fact that the eel caught at the most polluted VM site
were larger than those from the other sites. This illustrates certain limitations of the
correlation analyses, since it is not clear whether or not the larger size of eel at the
VM site is a consequence of the existing level of pollution. The observation that
variables are statistically linked does not necessarily imply that there is a causal
mechanism underlying these relationships. It may, however, be an additional argu-
ment for a further investigation of the mechanisms or for using the knowledge on
enzymatic induction in order to derive causeeffect relationships within ecosystems.
An example of a significant correlation that probably has no causal relevance is
the observed linear relationship between DDT tissue levels and the CYPl A-related
biomarkers (Fig. 2C). DDTs are known to be potent CYP2B inducers that hardly
affect CYPlA activities (Stegeman and Hahn, 1994). The significant relationship
70 R. van der Oost rt al.lAquatic Toxicology 39 (1997) 45-75
nevertheless observed between DDT levels and CYPlA induction (Spearman’s
~~0.57; p<O.Ol), may be explained by the