See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/230664194

The influence of the choice of physical and chemistry variables on

interpreting the spatial patterns of sediment contaminants and their
relationships with benthic communities

Article  in  Marine and Freshwater Research · January 2010

DOI: 10.1071/mf09263

CITATIONS READS

36 107

4 authors, including:

Anthony A Chariton Stuart L Simpson

Macquarie University The Commonwealth Scientific and Industrial Research Organisation
86 PUBLICATIONS   806 CITATIONS    159 PUBLICATIONS   3,958 CITATIONS   

SEE PROFILE SEE PROFILE

Graeme E Batley
Commonwealth Scientific and Industrial Research Organisation, CSIRO
257 PUBLICATIONS   10,892 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Saltwater Intrusion into Kakadu's South Alligator River View project

Nanomaterials in the environment View project

All content following this page was uploaded by Stuart L Simpson on 01 June 2014.

The user has requested enhancement of the downloaded file.

CSIRO PUBLISHING

www.publish.csiro.au/journals/mfr Marine and Freshwater Research, 2010, 61, 1109–1122

Influence of the choice of physical and chemistry variables

on interpreting patterns of sediment contaminants and their
relationships with estuarine macrobenthic communities

Anthony A. CharitonA,C, Anthony C. RoachB, Stuart L. SimpsonA

and Graeme E. BatleyA
A
Centre for Environmental Contaminants Research, CSIRO Land and Water, Locked Bag 2007,
Kirrawee, NSW 2232, Australia.
B
Ecotoxicology and Environmental Contaminants Section, Department of Environment,
Climate Change and Water, Lidcombe, NSW 1232, Australia.
C
Corresponding author. Email: Anthony.Chariton@csiro.au

Abstract. A primary objective of contaminated sediment risk assessments is to identify if contaminant enrichment is
eliciting an ecological response. Using complementary environmental and biotic datasets, we examined five scenarios
with respect to: dataset complexity; metal extraction; normalisation of organics; the inclusion/exclusion of acid-volatile
sulfide data, and iron and manganese concentrations. Spatial distributions of abiotic variables were examined by principal
components analysis, with canonical correspondence analysis used to examine the total and partitioning of biological
variation. Metals were the dominant contaminant and explained the largest proportion of variation in the macrobenthic
data. Extraction procedure and carbon normalisation of organics had little influence on the overall analysis. Porewater
metal data was essential for interpretation, with excess of acid-volatile sulfide over simultaneously extractable metals
being a poor surrogate. In the canonical correspondence analyses, the inclusion of Fe/Mn accentuated the covariation
between the ecological and contaminant variables. Multimodel comparisons aided interpretation by emphasising specific
relationships among environmental variables and their interactions with the biotic data. Furthermore, for future
examinations of the described system, the findings can be used to reduce the collection of redundant environmental
variables or variables that are poorly correlated with changes in macrobenthic assemblages.

Additional keywords: benthos, bioavailability, canonical correspondence analysis, environmental risk assessment,
porewater, principal component analysis, sediment.

Introduction acid-soluble fraction is likely to better approximate the bioavail-

In assessing contaminated sediments, environmental managers
need to determine whether the observed impacts are due to
contaminants or to other physical and chemical stressors
(e.g. grain size, ammonia, sulfide). Additional evidence from
ecotoxicology or bioaccumulation can assist in interpreting
ecological data in a weight-of-evidence (WOE) approach;
however, such assessments can still be equivocal (Wenning
et al. 2005). What is required is to demonstrate the existence or
absence of inferred cause–effect relationships between mea-
surements of stressors (exposure) and ecosystem health (as
measured by parameters such as diversity and abundance).
Such relationships may, however, be masked by the fact that
not all of the contaminant forms are bioavailable and some
measure of the bioavailable fraction might be preferred
(Chapman et al. 1998; Driscoll and Burgess 2007; Simpson and
Batley 2007).
In the case of metal contaminants, many assessments use
total metal concentrations obtained by extracting sediments
using concentrated acids, despite the knowledge that a dilute
able fraction (Tessier and Campbell 1987; Chapman et al. 1998;
Simpson and Batley 2007). Metal bioavailability has also been
shown to depend on the presence of reactive sulfides (e.g. acid-
volatile sulfides (AVS)) and organic carbon, while porewater
metals are an important exposure pathway for many sensitive
benthic organisms (Chapman et al. 1998; USEPA 2005). Simi-
larly, many organic contaminants bind strongly with organic
carbon, which significantly moderates their accumulation and
toxicity (Di Toro and McGrath 2000; USEPA 2003). To
accommodate this, sediment quality guidelines are frequently
based on concentrations normalised to 1% total organic carbon
(TOC) (ANZECC/ARMCANZ 2000). Grain size is an impor-
tant control on sediment ingestion by organisms, but the impor-
tance of dilution of fine sediments in controlling contaminant
dose has been poorly studied. Grain size can itself be a stressor if
it constrains burrowing activity or makes an area undesirable for
particular taxa (Gray 1974). The role of confounding stressors
such as ammonia and sulfide also needs to be evaluated (Ankley
et al. 1990; Wang and Chapman 1999).

Ó CSIRO 2010 10.1071/MF09263 1323-1650/10/101109

1110 Marine and Freshwater Research
A range of statistical approaches can be applied to examine
the relationships between ecological effects and environmental
stressors. These relationships can be initially assessed using
pairwise correlations and principal components analysis (PCA).
Constrained techniques, in particular, canonical correspondence
analysis (CCA) and redundancy analysis (RDA), have become
increasingly popular as they maximise relationships between
the structure of benthic communities and linear combinations
of environmental variables (ter Braak and Verdonschot 1995).
Furthermore, they allow the total variation in the biological
species data to be partitioned into independent components (e.g.
space, time and environmental influences) and an undefined or
stochastic component (Borcard et al.1992). From a risk assess-
ment perspective, partitioning or partial analysis enables the
investigator to identify the amount of variance in the community
data that is explained solely by ecological variables, solely by
contaminants, and shared by ecological variables and contami-
nants, as well as the proportion of the variance that is un-
explained. The reliability of this assessment is, however,
dependent on the investigators actually measuring environmen-
tal contamination in a manner that best predicts the biotic
response.
This study applies several statistical approaches to investi-
gate the spatial relationships between a range of contaminant
measures and other non-contaminant stressors on benthic com-
munity structure, in an attempt to provide information to guide
the selection of appropriate measures in future studies. The
ultimate objective was to establish an appropriate set of statis-
tical measures that might be generically applied to contaminated
sediments to assist environmental managers in the interpretation
of ecology lines of evidence in WOE assessments.
Although a large number of permutations can be extracted
from any extensive dataset, five scenarios were considered that
were believed to be of relevance to resource managers and
environmental risk assessors. These relate to: (i) the implica-
tions of employing a simplified environmental dataset (particu-
late metals, organic contaminants and granulometry); (ii) choice
of metal extraction technique; (iii) differences attributable to
normalising organic contaminants to total organic carbon;
(iv) the inclusion or exclusion of concurrent measurements of
AVS and simultaneously extractable metals (SEM); and (v) the
inclusion or exclusion of the covariables particulate iron
and manganese, and the porewater status indicators Fe(II) and
Mn(II).
Methods and materials
Sediment sampling
In early 2005, sediments were sampled from 15 locations
encompassing two estuarine regions of Sydney Harbour
(Parramatta and Lane Cove Rivers) and two adjacent estuaries
(Hawkesbury and Georges Rivers) (Fig. S1 of the Accessory
Publication to this paper). Their choice was based on literature
and historic data, and included seven contaminated locations
(Parramatta River), four internal harbour reference locations
(Lane Cove River) and four external reference locations (two
in each of the Hawkesbury and Georges Rivers). Additional
criteria used to select the locations included the degree of
similarity in their substrate (sandy or silts) and the common
A. A. Chariton et al.
physico-chemical properties of the water column (depth,
salinity, dissolved oxygen, pH, turbidity and temperature).
To encapsulate within-location variation, four sites with
approximate dimensions of 3
3 m were nested equidistant
within each location. At each site, seven samples were collected
using a hand-operated sediment corer (100 mm internal dia-
meter), comprising four samples for benthos and one each for
contaminants, porewater and particle size analysis.
For benthos, the top 100 mm of each core was extruded in the
field and rinsed on a 500-mm sieve, with the retained organisms
fixed in 7% buffered formalin in seawater solution with a Rose
Bengal vital stain. After 7 days, the samples were washed with
water to remove the formalin, and preserved in 70% ethanol
solution. Samples were sorted under a dissecting microscope
and keyed to family, with the exception of nemerteans (phyla)
and sipunculids (phyla). Confirmation of all taxa was performed
by the Australian Museum, Sydney. The abundance of each
taxon was determined from the mean of the four replicate cores
obtained from each site.
For the environmental analyses, the top 100 mm of each core
was extruded in the field and sectioned into 0–20-mm (top) and
20–100-mm (bottom) depth fractions. Subsamples were taken
from each depth for analyses of metal contaminants (50 g in a
polycarbonate vial) and AVS (50 g in a polycarbonate vial), and
from the 0–20-mm fraction for analyses of organic contaminants
and TOC (250 g in a glass jar with Teflon lid), porewater metals
(sediment-filled, 50-mL centrifuge tube, no head space) and
particle size (50 g in a polycarbonate vial). All sediments were
chilled on ice for transport to the laboratory, and then stored
at 48C in the dark (AVS samples were frozen) until the time
of analysis. Sediment porewater was extracted under a nitrogen
atmosphere within 24 h of collection.
Analytical methods
All glassware and plasticware used for trace metal and metalloid
analyses were cleaned by soaking in 10% (v/v) HNO3 (BDH
Laboratory Supplies, Poole, England) for a minimum of 24 h,
followed by a thorough rinsing with Milli-Q water (Millipore,
Academic Water System, Sydney, Australia). Prior to use, all
glassware used for analyses of organic contaminants was
cleaned by rinsing with nanograde acetone and dichloromethane
(Suprasolv, Merck, Darmstadt, Germany).
The sediment grain size fractionations involved sieving a
known weight of wet sediment through nylon sieves followed
by gravimetric analysis to give the fractions: o63 mm (fines),
o63 mm–1 mm (sand), 41 mm (coarse). TOC analyses in sedi-
ments were performed using a high-temperature TOC analyser
(Dohrmann DC-190, Teledyne Tekmar, Mason, OH, USA)
following removal of inorganic carbon (carbonates and bicar-
bonates) by acidification with 1 M HCl until effervescence was
complete. AVS was determined on the top (0–20 mm) and
bottom (20–100 mm) sediments according to Simpson (2001).
Sediment porewater was extracted by centrifugation (5 min at
1500g, 18–228C) under a nitrogen atmosphere, then filtered
through acid-washed 0.45-mm membrane filters (Minisart, Sar-
torius, Goettingen, Germany) before analyses. Porewater ana-
lyses included: ammonia, preserved by freezing then analysed
within 1 week by colourimetric analysis (Spectroquant Kit
14752, Merck); sulfide, analysed immediately using the
Variables for contaminant–benthos relationships Marine and Freshwater Research 1111

colourimetic method of Cline (1969); and dissolved metals,
acidified to 2% HNO3 (v/v) (Tracepure, Merck) then analysed
using inductively coupled plasma atomic emission spectrometry
(ICP-AES).
The ICP-AES instrument (Spectroflame EOP, Spectro Ana-
lytical Instruments, Kleve, Germany) was calibrated with
matrix-matched standards and had limits of determination of
3 mg L1 for Cd, Co, Cr, Cu, Mn, Ni and Zn, and 5–20 mg L1 for
Al, As, Fe and Pb for all marine porewater and sediment
acid extracts. The levels of detection for total particulate
metals (TPM) and dilute acid-soluble metals (ASM) were 0.2–
1 mg kg1. All acid extracts were filtered through 0.45-mm
membrane filters and analysed by ICP-AES.
TPM concentrations of sediments were measured following
an aqua regia extraction (25 mL of 2 : 1 concentrated HCl :
HNO3, 20 min microwave at 100 W in closed 50-mL vessels)
using 0.25 g of oven-dried (808C), homogenised sediment.
Dilute ASM concentrations of total sediments and o63 mm
sediment ASM fractions (ASMf) were measured following
extraction of sediments (0.25 g dry weight) with 25 mL of cold
1 M HCl for 1 h. The ASM extraction is similar to the acid
extraction used to determine SEM for AVS minus SEM (AVS-
SEM) studies (US EPA 2005). The terms ASM or ASMf are used
to describe this metal fraction, except when comparing to AVS,
in which case the term SEM is used. As a check on analytical
quality, a certified reference material, PACS-2 (National
Research Council Canada, Ottawa, Canada), was analysed in
each batch of sediment samples. For the TPM analyses, recov-
eries of the certified reference material were 80–105% of all
metals, and within expected ranges for the ASM analyses
(Scouller et al. 2006). Spike recoveries of metals in TPM and
ASM extracts were 85–105% for all metals.
Analyses of organic contaminants comprising a wide range
of hydrophobic organic contaminants (e.g. polycyclic aromatic
hydrocarbons (PAH)) and organochlorine and organophosphate
pesticides followed methods 8260/8015 (USEPA 1996). A full
list of the organic contaminants analysed and the detection limits
is provided in Table S1 of the Accessory Publication. As a check
on analytical quality for organic contaminant analyses, surro-
gate compounds were spiked into all samples and recoveries
were (mean  s.d., range) 96  9%, 75–130% for semivolatiles
(e.g. PAH), 99  6%, 80–120% for volatiles (e.g. benzene),
91  17%, 57–125% for organochlorines, and 118  30%,
52–149% for organophosphates. Sediments collected from the
top 0–20 mm and lower 20–100 mm are suffixed ‘top’ and
‘bottom’.
Statistical analysis
Seven environmental data matrices were created for the purpose
of comparing different combinations of environmental vari-
ables. The major constituents for each matrix are provided in
Table 1, with environmental variables of grain size (fines, sand,
gravel) and the percentage of TOC being common to all datasets.
The most rudimentary matrix was the ‘Basic’ matrix, which
only contained aqua regia-extracted TPM, organic con-
taminants, grain size and organic carbon data, with no porewater
data. The matrices TPM, ASM and ASMf differed only in the
metal extraction techniques (aqua regia, 1 M HCl and 1 M HCl
of the fine fraction), with all three matrices containing porewater
data (Cr, Cu, Ni, Zn, total sulfide and ammonia). Matrix AVS-
SEM was the same as matrix TPM with the addition of the
variables AVS-SEM (top) and AVS-SEM (bottom). Matrix
%TOC was the same as matrix TPM with the exception that
the organic contaminant concentrations were normalised to 1%
TOC. Matrix Fe/Mn was the same as matrix TPM except for the
addition of particulate and porewater iron and manganese.
Using these seven matrices, the following comparisons were
made: (i) extensive environmental v. basic environmental data
(TPM v. Basic); (ii) the influence of metal extraction technique
(TPM v. ASM v. ASMf); (iii) stratified measurements of AVS
minus SEM (AVS-SEM v. TPM) as a surrogate for porewater
metal measurements; (iv) normalised v. unnormalised organic
contaminants (TPM v. %TOC); and (v) the inclusion/exclusion of
particulate and porewater iron and manganese (TPM v. Fe/Mn).

Table 1. Main components of the seven environmental matrices used for statistical analyses
Dotted cells indicate environmental variables included in each matrix. ASM, cold 1 M HCl extraction; ASMf, cold 1 M HCl extraction of theo63-mm sediment;
AVS, acid-volatile sulfide (mmol kg1); organic contaminants (mg kg1); Fe/Mn, Fe and Mn concentrations in total particulate (mg kg1) and porewater
(mg L1); grain size, the proportion of sediment in each of three size classes fines (o63 mm), sands (63 mm–1 mm) and gravel (41 mm); organic contaminants,
total concentrations (mg kg1) of chlordanes; organic contaminants normalised, each class of organic contaminant was normalised to 1% total
Porganic carbon;
porewater metals, Cd, Cu, Ni, Pb, Zn (mg L1); porewater NH3 and HS, ammonia and sulfide (mg L1); SEM, 1 M HCl extractable metal ( Cd, Cu, Ni, Pb,
Zn); TOC, total organic carbon (%); TPM, hot concentrated 2 : 1 HCl : HNO3 extraction

Matrix Environmental variables

name TPM ASM ASMf Fe/Mn Organic Organic TOC Grain Porewater Porewater AVS-SEM
contaminants contaminants size metals NH3 and HS
(normalised)

Basic    

TPM      
ASM      
ASMf      
AVS-SEM      
%TOC      
Fe/Mn       
1112 Marine and Freshwater Research A. A. Chariton et al.

1.0 1.0
(a) Basic (b) TPM
Co

PW-Cu Ni

PW-Ni
PW-Cr
Heptochlor Cr
TPH Fines
As
Gravel
TOC Chlordane
PAH PCBs Cd

PCBs Ag PW-NH3 PW-HS

Dieldrin Dieldrin
Zn Zn
DDT Cu Cu
PW-Zn TOC Ag
DDT Gravel PAH
Cd
Chlordane
Heptochlor TPH
As
Fines
Cr

Ni
Co

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

1.0 1.0
(c) ASM (d ) ASMf
PW-Cu PW-Ni
Co Cu Ni
As PW-Cr
Co

Cr
PCB Ni
Cd
Zn

PW-Zn Cr
Fines
PW-NH3
DDT PW-HS Cd
Fines Heptachlor PAH Gravel
Dieldrin TOC
Dieldrin Heptachlor Zn
Gravel DDT PAH
Chlordan As TPH
PCB Chlordane
PW-HS TPH Cu
PW-Cr PW-NH3 Ag
PW-Cu
TOC
PW-Ni
PW-Zn

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

Fig. 1. Two-dimensional principal components analysis ordination biplots from the seven environmental datasets: (a) Basic, (b) total particulate metals
(TPM), (c) acid-soluble metals (ASM), (d) the o63-mm sediment ASM fractions (ASMf), (e) AVS minus simultaneously extractable metals (SEM),
(f) percentage of total organic carbon (TOC) and (g) Fe/Mn. Hawkesbury River (up-triangles), Georges River (down-triangles), Lane Cove River (diamonds)
and Parramatta River (circles). Dotted line in (b) TPM highlights the four sites from the Parramatta River with elevated concentrations of the porewater metals
Cu, Cr and Ni. All horizontal and vertical axes represent PC1 and PC2, respectively. The prefix PW indicates that the variable was obtained from the porewater.
T, top layer of sediment 0–20 mm; B, bottom layer of sediment (20–100 mm).

To examine the differences in efficiencies between the three
metal extraction techniques, correlations (Pearson’s product-
moment) were performed on the particulate metal concentrations
using NCSS statistical software (NCSS, Kaysville, UT, USA).
Spatial patterns in environmental variables
To assess the influence of variable selection on the spatial pat-
terns of environmental variables, PCA was performed on all
seven environmental matrices (Table 1). Prior to computation,
plots were created to visualise the normality of the abiotic
variables, with product-moment correlations between all abiotic
variables being simultaneously produced. Subsequent to the
appropriate transformations (log10 for dissolved and particulate
metals; arcsine for grain size measurements), highly correlated
variables (r240.95) were removed from the analysis (Clarke
and Ainsworth 1993). PCA were scaled to focus on the inter-
species correlations with the data centred and standardised.
All PCA were performed using CANOCO 4.5 (Biometris,
Wageningen, The Netherlands). In all illustrated PCA
ordination plots (see Fig. 1), the horizontal and vertical axes
Variables for contaminant–benthos relationships Marine and Freshwater Research 1113

1.0 1.0
(e) AVS-SEM (f ) %TOC
DDT Chlordane

Heptochlor
TPH Dieldrin
Heptochlor
Gravel TOC
Chlordane
PAH PCB
PW-NH3 Cu
PW-HS
PCB Ag TPH Cr
Zn Zn
Cu Gravel
Dieldrin Cd
(T)AVS-S PW-Cr
DDT Fines PW-Zn
Cd Ag
(B)AVS-S PW-NH3
As
Ni
PW-HS
As PW-Cu Co
Fines
Cr
TOC
PW-Ni
Ni
Co

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

1.0
(g) Fe/Mn

PW-Fe

DDT Heptochlor
Chlordane
Gravel TPH
PW-Zn
PW-NH3
PAHs Zn
Cu
PW-Mn TOC Ag
PW-HS Dieldrin
PCBs
PW-Cr
Cd
Fines Cr

PW-Ni
As
PW-Cu

Mn

Fe Ni
Co
⫺1.0
⫺1.0 1.0

Fig. 1. (Continued)

correspond to PC1 and PC2, respectively. The length of each
arrow represents how well the environmental variable is repre-
sented in the ordination plot, with longer length indicating
greater representation. The angles between arrows approx-
imates the correlation between environmental variables, with
information regarding the observed values of a particular site,
e.g. relative concentrations of contaminants, being obtained by
projecting a perpendicular line from a site to a specific envir-
onmental arrow.
To aid interpretation of the PCA numerical output, the
broken-stick criterion was used to assess the number of compo-
nents (axes) to retain (Frontier 1976). The broken-stick criterion
is a simple and heuristic approach that assumes that total
variation is randomly distributed among components (Legendre
and Legendre 1998). Put simply, observed eigenvalues derived
from the PCA are considered meaningful if their value exceeds
that predicted by the broken-stick model. This approach can be
used as a stopping mechanism when the observed eigenvalue is
exceeded by the predicted eigenvalue. Comparisons between
observed and predicted values were restricted to the first four
components.
Relationships between benthic communities
and environmental matrices
CCA was performed on all seven environmental datasets
(Table 1) using the same biological (species data) matrix
(Table S2 of the Accessory Publication). Prior to multivariate
analysis, taxa represented by one or two individuals summed
across all samples were removed from the biological matrix,
1114 Marine and Freshwater Research A. A. Chariton et al.

with the sampled abundance data of the remaining thirty- A summary of the macrobenthic data is provided in Table S2
five taxa log(x þ 1) transformed to reduce the weight of (Accessory Publication).
highly abundant taxa (Clarke and Warwick 1994). Preliminary
detrended correspondence analysis on the species data found Simple v. complete environmental data (Basic v. TPM)
that a majority of the species responded in a unimodal manner
An examination of all PCA ordination plots clearly shows that
to the environmental variables (gradient length 43 standard
sites from both the Georges River and Hawkesbury River
deviations), indicating that CCA was suitable (Lepš and
had considerably lower concentrations of contaminants than
Šmilauer 2003). For each environmental matrix, an initial CCA
either the Lane Cove River or the Parramatta River (Fig. 1). As
was run to identify variables with a high collinearity, as these
indicated by the size and direction of the variable arrows, sites
interfere with the analysis. Variables with the highest variation
from the Parramatta River were aligned along several strong
inflation factors (VIF) were sequentially removed until all
contamination gradients. It should be noted that particulate lead
variables had a VIF  20 (Peeters et al. 2001). Details of the
and the proportion of sand were removed from all PCA analyses
environmental variables used in each CCA are provided in
before computation as they were strongly highly correlated
Table S3 (Accessory Publication).
with other environmental variables, most notably zinc, and silt,
For each CCA, a global permutation test on the significance
respectively.
of the combined canonical axes was performed to examine
The simplest PCA (Basic), which contained just variable sets
whether there was a significant relationship between the biotic
for total particulate metals/metalloids and organic contami-
and environmental matrices. Variance partitioning was per-
nants, organic carbon, and particle size, provided the greatest
formed using the procedure described by Borcard et al.
sum of explained variance across the first four components
(1992). In this procedure, four separate CCA analyses are
(principal component (PC) 1–4 ¼ 72.0%) (Table 2). In all the
performed: (1) CCA-1, constrained by the ecological data;
presented PCA, the primary interpretable information was
(2) CCA-2, constrained by the contaminant dataset; (3) partial-
not restricted to the first four PCs, as calculated eigenvalues
CCA-1, constrained by the ecological dataset with the contami-
exceeded those predicted by the broken-stick criterion (Frontier
nant dataset used as the covariables; and (4) partial-CCA-2,
1976) (Table 2). The ordination plot derived from TPM data
constrained by contaminant dataset with the ecological dataset
matrix (PCA-TPM, Fig. 1b) showed a strong relationship
used as the covariables. From these analyses, variances were
between particulate metals, polychlorinated biphenyls (PCBs),
partitioned into: (i) total explained variance ¼ (1) þ (4); (ii)
dieldrin and PAH along the first PC, with porewater Cu, Ni and
solely ecological ¼ (3); (iii) solely contaminant ¼ (4); (iv) ecol-
Cr, and particulate Co being along the second PC. Although the
ogy shared by contaminants ¼ total explained – (3 þ 4); and
total sum of explained variation was 9% lower in PCA-TPM
(v) undefined ¼ 1 – total explained. For each CCA, the effect of
than in PCA-Basic, the inclusion of porewater contaminants not
each subgroup (e.g. particulate metal contaminants, porewater
only aided separation of the four estuaries, but also highlighted
metals, organic contaminants, particulate and porewater iron
several Parramatta River sites with elevated concentrations of
and manganese, grain size, TOC and AVS) was examined in
porewater metals and hydrogen sulfide, a feature not common to
separate analyses by individually constraining each subgroup,
all sites (Fig. 1b).
with all other variables used as the covariable dataset. Monte
For all seven CCA, both ecological variables and contami-
Carlo permutations tests were used to identify if each environ-
nants were found to significantly influence benthic community
ment subgroup significantly (P  0.05) contributed to variations
structure (Monte Carlo permutation test, Po0.05) (Table 3).
in the species data.
The addition of porewater (CCA-Basic v. CCA-TPM) had a
In all CCA biplots, the horizontal and vertical axes correspond
pronounced effect on the analyses, with the percentage of
to the first and second axis (Fig. 2). Arrows represent environ-
explained variance in the biotic data increasing 54.7% (CCA-
mental variables, with the direction of each arrow indicating its
Basic) to 67.9% (CCA-TPM). As indicated by the partial CCA,
trajectory of increasing influence. Arrow length signifies the
where biotic data were constrained by the subgroups of con-
importance of the environmental variable. Sample points, in this
taminant variables (e.g. particulate and porewater metals),
case up and down triangles, circles and diamonds, represent
differences between the CCA-Basic and CCA-TPM were pre-
macrobenthic assemblages, with the distance between samples
dominately due to the contribution of porewater contaminants
increasing with the dissimilarity in species composition.
(metals, ammonia and hydrogen sulfide). The influence of
porewater contaminants on the structure of macrobenthic com-
Results munities was also evident in qualitative comparisons between
the biplots derived from the two analyses (Fig. 2a, c). For
Metal contamination of the sediment was at higher levels than
example, the biotic composition of several sites from the
organic contamination, although some sediments had envir-
Parramatta River correlated with elevated concentrations of
onmentally relevant concentrations of organic contaminants
porewater chromium, with the arrow position and length of
such as PAH. The concentrations of As, Cd, Co, Cr, Cu, Ni, Pb
the chromium porewater vector indicating its direction and
and Zn measured by the three different metal extraction tech-
importance (e.g. Fig. 2b).
niques are provided in Table S4 (Accessory Publication). The
mean concentrations of the major organic contaminants and key
sediment properties (TOC and particle size) in the sediments Metal extraction techniques (TPM v. ASM v. ASMf)
from the Parramatta, Lane Cove, Hawkesbury and Georges Correlations between metals extracted using TPM (concentrated
Rivers are provided in Table S5 (Accessory Publication). acid digest on bulk sediment), ASM (dilute acid digest on bulk
Variables for contaminant–benthos relationships Marine and Freshwater Research 1115

1.0 1.0
(a) Basic (b) TPM
PW-Cr
PW-Cu

PW-Ni

Cu
PAH
TPH PW-NH3 PW-Zn
Zn Dieldrin
Fines
Ag
PW-HS
Heptochlor Heptochlor Co
Gravel As
Gravel
Cr
Dieldrin TPH TOC
Cr Zn
Ag As
Co PAH
PCB Cu

TPH

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

1.0 1.0
(c) ASM (d ) ASMf
PW-Cr PW-Cr
PW-Cu PW-Cu

PW-Ni
PW-Ni

PW-Zn PW-NH3 Co
PW-Zn
Fines Fines PW-NH3
PCB
PW-HS PW-HS
Heptachlor Heptachlor Ni
Gravel Gravel
Cd TOC Dieldrin Cd
Co Dieldrin Ag
Zn Cu
TPH PAH
PAH
As As

Cu

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

Fig. 2. Canonical correspondence analysis (CCA) ordination biplots derived from the seven environmental datasets: (a) Basic, (b) total particulate metals
(TPM), (c) acid-soluble metals (ASM), (d) the o63-mm sediment ASM fractions (ASMf), (e) AVS minus simultaneously extractable metals (SEM),
(f) percentage of total organic carbon (TOC) and (g) Fe/Mn. Hawkesbury River (up-triangles), Georges River (down-triangles), Lane Cove River (diamonds)
and Parramatta River (circles). To aid visual interpretation, all variables with a fit of less than 15% were removed from the ordination maps. The prefix PW
indicates that the variable was obtained from the porewater. T, top layer of sediment 0–20 mm; B, bottom layer of sediment (20–100 mm).

sediment) and ASMf (dilute acid digest on o63-mm fraction)
are shown in Table 4. The concentrations of metals extracted
using ASM and ASMf were significantly correlated with those
extracted using the TPM procedure, although this was not the
case for the metalloid arsenic. Comparisons between the mean
TPM and ASM concentrations showed that the ASM procedure
extracted from 11 to 58% of the TPM fraction. Copper and zinc
concentrations were similar in the ASMf and TPM extracts.
Although lower, ASMf concentrations of Co, Cr, Ni were still
between 77 and 84% of those extracted using TPM. Both cad-
mium and lead mean concentrations were higher in the ASMf
than the TPM fraction at 156 and 130%, respectively. As with
ASM, the relative proportion of ASMf arsenic was low. For all
metals, concentrations obtained from the bulk ASM fraction
(o2 mm) were lower than those extracted from the fine fraction
(o63 mm). In the case of copper, differences between the bulk
and fine fraction were marked, with concentrations being up to
38 times greater in the fine fraction.
PCA for all three metal extraction procedures explained
similar amounts of variation (Table 2). Visually, the ordination
plots derived from the TPM and ASMf datasets were very
similar, with the first PC being driven by particulate
1116 Marine and Freshwater Research A. A. Chariton et al.

1.0 1.0
(e) AVS-SEM (f ) %TOC
PW-Cr
PW-Cu

TPH
Heptochlor
Gravel TOC PW-Ni
Chlordane
PW-NH3 PAH PCB
PW-NH3 PW-Zn
PW-HS
PCB Ag
Heptochlor Fines
Zn PW-HS
Cu Gravel Co
Dieldrin
(T)AVS-S Dieldrin
DDT Cr TPH
Cd TOC
(B)AVS-S As
Zn Ag
Cu PAH
As
fines
Cr

Ni
Co

⫺1.0 ⫺1.0
⫺1.0 1.0 ⫺1.0 1.0

1.0
(g) Fe/Mn
PW-Cr

PW-Cu

PW-Ni
PW-NH3
PW-Zn
PW-HS Fines
Heptochlor Co
Cr
Gravel
Dieldrin TPH TOC
Zn
Ag
As
PW-Fe PAH

⫺1.0
⫺1.0 1.0

Fig. 2. (Continued)

contaminants and the second PC by dissolved ammonia, metals
and sulfide (Fig. 1b, d). In the ordination plot derived from PCA-
ASM, separation along the PC1 axis was predominately due to
increasing concentrations of organic contaminants, with separa-
tion due to contaminant levels occurring in both directions along
the PC2 axis, e.g. As, Cu, Co, Ni in the positive direction, and
porewater Cu, Ni, Cr and Zn in the opposite direction. This does
not imply that concentrations of these elements were all of
environmental significance, but rather, separation occurred due
to differences in their concentrations. The four Parramatta River
sites identified in PCA of TPM and ASMf (as indicated by the
circles in Fig. 1b, d) with clearly elevated concentrations of
porewater metals were less obvious in the PCA-ASM (Fig. 1c),
as was the separation of sites from the Georges and Hawkesbury
Rivers.
CCA derived from all three metal extraction procedures
explained similar amounts of variation in the biotic data, with
the proportion of variance explained solely by the ecological
variables, contaminants and their shared variance also varying
little among the three analyses (Table 3). The CCA biplots
derived from these analyses were comparable (Fig. 2b–d), with
macrobenthic communities being more strongly correlated with
metals, both dissolved and particulate, than with organic con-
taminants, grain size, dissolved ammonia or dissolved sulfide.
The amount of variance explained by particulate metals was
greater in CCA-TPM (13.8%) and CCA-ASMf (12.7%) than
CCA-ASM (9.1%) (Table 3). To a lesser extent, these trends
were also present in the proportions of variation that could
be attributed to porewater metals and organic contaminants.
In CCA TPM and ASMf, grain size and porewater toxicants
Variables for contaminant–benthos relationships Marine and Freshwater Research 1117

Table 2. Summary of principal component analysis of the environmental variables

Eigenvalues for the first four principal components (PCs), cumulative variation of PC1–2 are presented. Although not shown, in all cases calculated
eigenvalues were above those estimated using the broken-stick criterion. ASM, acid-soluble metals; ASMf, theo63-mm sediment ASM fractions; AVS, acid-
volatile sulfides; SEM, simultaneously extractable metals; TOC, total organic carbon; TPM, total particulate metals

Matrix No. of variables Eigenvalues Sum of explained variance

PC1 PC2 PC3 PC4 PC1–2 PC1–4

Basic 18 0.34 0.16 0.12 0.1 50.5 72

Full (TPM) 24 0.28 0.14 0.12 0.09 42 63
ASM 23 0.22 0.19 0.13 0.1 40.1 62.6
ASMf 24 0.31 0.15 0.1 0.1 45.9 65.4
AVS-SEM 22 0.33 0.14 0.1 0.09 47 65.9
%TOC 23 0.24 0.16 0.15 0.1 40.1 64.8
Fe/Mn 28 0.28 0.18 0.11 0.08 45.6 64.8

Table 3. Summary table from the canonical correspondence analysis (CCA) of each matrix and the partitioning of variance in the species data
Sum of variances explained by subgroups of variables (e.g. grain size and total organic carbon) do not necessarily equal the total sum explained by its
source of variation (e.g. ecological variables). ASM, acid-soluble metals; ASMf, the o63-mm sediment ASM fractions; AVS, acid-volatile sulfides; SEM,
simultaneously extractable metals; TOC, total organic carbon; TPM, total particulate metals

Source of variance CCA matrices

Basic TPM ASM ASMf AVS-SEM %TOC Fe/Mn

Explained 54.7 67.9 66 66.8 57.8 65.8 68.2

Ecological 7.4 6.7 6 5.5 7.8 6.8 8.8
Grain size 5.4 5 4.3 4.7ns 5.4 4.3 4.2
Total organic carbon 2 1.7 1.8 0.8 2.4 2.4 1.3
Fe (particulate þ porewater) n/a n/a n/a n/a n/a n/a 3.5
Contaminant 45 58.2 56.3 57.1 48.1 58.3 47.2
Particulate metals 26.3 13.8 9.1 12.7 16.7 14.1 6.8
Porewater metals n/a 9.4 6.9 8.3 n/a 9.6 8.8
AVS-SEM n/a n/a n/a n/a 2.7ns n/a n/a
NH3 and HS n/a 2.7 3.4 2.5ns 2.9ns 3.1 3.2
Organic contaminants 14 8.8 5.4 8.9 13 10.3 11
Shared 2.3 3 3.7 4.2 1.9 2.9 14
Unexplained 45.3 32.1 34 33.2 42.2 32 31.4
ns
Subgroup did not significantly (Monte Carlo P40.05) contribute to variation in the species data.

(ammonia and sulfide) were found to significantly contribute to
changes in benthic community structure; however, this was not
the case for CCA-ASM (Table 3).
AVS-SEM as a surrogate for porewater metals
(AVS-SEM v. TPM)
In both the top (0–20 mm) and bottom (20–100 mm) of the
sediment column, differences in the relative concentrations of
‘AVS minus SEM’ (AVS-SEM) were strongly correlated with
the main particulate contamination gradient along the Parra-
matta River sites (Fig. 1e), with the ordination biplot having a
strong resemblance to that derived from PCA-Basic (Fig. 1a).
In contrast to the PCA for the TPM dataset (Fig. 1b), there was
no evidence to suggest porewater metal concentrations were
negatively correlated with AVS-SEM (Fig. 1b, e), and conse-
quently, the sites with high concentrations of porewater metals
were not clearly separated from the remaining sites with the
Parramatta River.
Using stratified AVS-SEM measurements as a surrogate for
porewater metals (CCA-AVS-SEM compared with CCA-TPM)
considerably reduced the total amount of variance explained in
the macrobenthic community data (Table 2; Fig. 1b, f), with this
stemming from a 10% reduction in the variance in the biotic
data, which could be solely attributed to contaminants. Stratified
measurements of AVS-SEM did not significantly (P40.05)
influence the variation in biotic data, with the CCA analysis of
the AVS-SEM matrix producing comparable results to CCA-
Basic.
Normalisation of organic contaminants (%TOC v. TPM)
The normalisation of organic contaminant concentrations to TOC
(i.e. 1% TOC) did not considerably increase the total amount
of variance explained by the first four components in the PCA
(Table 2). Qualitatively, normalisation increased the influence of
some organic contaminants, most notably dichlorodiphenyl-
trichloroethane (DDT), and to a lesser extent heptachlor and
1118 Marine and Freshwater Research A. A. Chariton et al.

Table 4. Correlations (r2-values) and ratios of metals extracted using three different techniques
Ag is not presented as acid-soluble metal (ASM) concentrations were consistently below the detection limit. ASMf, the o63-mm sediment ASM fractions;
TPM, total particulate metals

Metal Correlations (r2)A Relative proportion of metal extracted

B
TPM v. ASM TPM v. ASMf ASM v. ASMf ASM to TPM (%) ASMf to TPM (%) ASM to ASMf (%)B

As 0.122ns 0.147ns 0.071ns 11  2 (2–40) 34  2 (9–94) 36  3 (3–137)

Cd 0.811*** 0.853*** 0.583*** 58  2 (18–100) 156  11 (30–407) 45  4 (7–189)
Co 0.572*** 0.709*** 0.365** 22  1 (8–62) 77  3 (45–188) 30  2 (10–78)
Cr 0.854*** 0.859*** 0.660*** 17  1 (4–40) 80  5 (26–268) 25  3 (7–126)
Cu 0.529*** 0.972*** 0.460*** 16  2 (1–73) 103  5 (53–278) 17  2 (1–81)
Ni 0.674*** 0.727*** 0.314* 19  1 (8–50) 84  4 (46–202) 25  1.7 (6.7–75)
Pb 0.773*** 0.930*** 0.621*** 51  2 (22–114) 130  7 (60–271) 46  4 (13–132)
Zn 0.938*** 0.966*** 0.885*** 44  2 (18–101) 102  5 (45–200) 48  1 (17–170)
A
*Po0.05, **Po0.01, ***Po0.001, ns P40.05. BMean  1 s.e. (min–max).

chlordane, with these changes increasing the dispersal of sites
within the Parramatta and Lane Cove Rivers (Fig. 1a, e).
The total proportion of explained variance in the biotic data
and the partitioning of the main components (ecological, con-
taminants and shared) was similar in the normalised analysis to
that obtained from unmodified concentrations of organic con-
taminants (Table 3). Normalisation marginally increased (1.5%)
the proportion of variance that was solely due to organic
contaminants, with this being reflected as minor increases in
the lengths of some vectors, e.g. heptachlor (Fig. 2b, f).
Addition of porewater and particulate iron and manganese
(Fe/Mn v. TPM)
The inclusion of particulate and porewater iron and manganese
(PCA-Fe/Mn) marginally increased (1.8%) the percentage of
explained variance summed across the first four PC (Table 2).
The environmental vectors on the ordination plot indicated that
the concentrations of both particulate and porewater manganese
were greater in the external reference locations (Fig. 1f). Parti-
culate iron and manganese also aided separation of the external
locations, with greater concentrations of particulate manganese
being characteristic of sediments sourced from the Hawkesbury
River (Fig. 1f and Table S5, Accessory Publication).
Both particulate and porewater manganese were removed
before CCA computations due to their high VIF. The inclusion
of particulate and porewater iron (CCA-Fe/Mn) had a percep-
tible and quantifiable effect on the canonical analysis (Table 3).
Although the total amount of explained variance was similar to
that produced from CCA-TPM, the proportion of variance solely
due to contaminants was decreased by 11%, with this being
predominately due to a reduction in variance explained by
particulate metals. The addition of iron also appreciably
increased the collinearity between the ecological and contami-
nant variables (shared variance), indicating that iron influences
both the ecological and contaminant datasets simultaneously.
Although the relationship between the biotic data and iron
(combined porewater and particulate) was shown to be signifi-
cant (Monte Carlo permutation tests Po0.05), with the vector
length of porewater iron also affirming this relationship, overall
the biplot derived from CCA-Fe/Mn was akin to that derived
from CCA-TPM (Fig. 2b, g).
Multimodel comparisons
In all seven models, global permutation tests of the CCA were
statistically significant (Po0.05), indicating that each model
adequately fitted the data. The inclusion of porewater metals
was found to be essential for obtaining good correlations
between environmental variables and benthic community
structures; therefore, models that lacked these variables (CCA
Basic and AVS-SEM) can be considered to be suboptimal. Of
the five remaining models (CCA-TPM, ASM, ASMf, %TOC
and Fe/Mn), the total amount of explained variance differed
little (2.2%). Some differences in the way the total variance was
partitioned were evident. This was especially true for CCA-
Fe/Mn, which showed a greater proportion of shared variance
between the ecological and contaminant variables than the other
analyses. The results for the remaining four analyses (TPM,
ASM, ASM and %TOC) were congruent, including the parti-
tioning of variances and number of environmental variables
required in the analyses (Table 3). However, a marginally
greater percentage of the total variation in the biotic data and that
solely attributable to contaminants was extracted from CCA-
TPM (Table 3).
Discussion
Examination of the PCA biplots clearly demonstrates that con-
taminant enrichment was appreciably more pronounced in the
Parramatta River locations. This area of the harbour has been
subjected to a diverse range of anthropogenic activities (e.g.
industries, general urbanisation and stormwater runoff) (Irvine
and Birch 1998). Elevated concentrations of some contaminants
were also present in the Lane Cove River, a system that is also
physically linked to Sydney Harbour; however, the diversity and
concentrations of contaminants were markedly lower than in
the Parramatta River. Combined with the two external reference
estuaries, Georges and Hawkesbury, the study encapsulated a
diverse suite of contaminants, with the concentrations of some
contaminants varying by several orders of magnitude.
Regardless of approach, a large and significant proportion
(45%) of the variation in the structure of benthic communities
could be attributed to environmental contaminants, most nota-
bly, particulate-associated metals. The lower contribution of the
Variables for contaminant–benthos relationships
ecological variables, e.g. grain size and TOC, was most likely
because of the low variation in these variables. Indeed, the site-
selection criteria used in the sampling program led to minimal
variation in the ecological variables, as grain size and organic
carbon have been frequently reported as major factors in
determining the structure of benthic communities (Snelgrove
and Butman 1994; Kraan et al. 2009). The strong relationship
between contaminants and benthic community structure may
also be the result of the number of contaminant variables
exceeding the number of ecological variables, with the addition
of each variable reducing the degrees of freedom (ter Braak
1996). It is emphasised, however, that the relationships between
biotic data and contaminants were statistically significant,
affirming the correlative relationship between benthic commu-
nities and environmental contaminants within the study region.
Simple v. complete environmental data (Basic v. TPM)
The PCA derived from the Basic matrix explained a greater total
sum of variation than all other analyses. As the results of this
analysis failed to constrain the interpretable information down
to the axes presented in the biplots, i.e. observed eigenvalues
always exceeded those expected by the broken-stick criterion
(Frontier 1976), this finding provides no additional information
to assist in interpreting the spatial distribution of the environ-
mental variables. Conversely, the addition of porewater con-
taminants, especially metals, identified several Parramatta
River sites of potential concern that otherwise would have been
overlooked as they were not directly associated with sites of
high particulate contaminant concentrations.
The results from the CCA (Basic v. TPM) clearly demon-
strate that there is a strong association between benthic com-
munity structure and porewater contaminants within Sydney
Harbour, especially within the Parramatta River. Dissolved
metal contaminants in the porewaters are expected to be bio-
available to many organisms and can cause both chronic toxicity
and sublethal effects (e.g. impairment of growth and reproduc-
tion), and potentially cause significant effects at a community
level (Ankley et al. 1991; Wang et al. 1999; Millward et al.
2001). Although there are several issues associated with the
collection and analysis of porewaters (Chapman et al. 2002), as
evident in this study, the addition of these variables appears
essential when examining the potential effects of contaminants
on benthic communities.
Metal extraction techniques (TPM v. ASM v. ASMf)
The overall difference between the bulk and fine (o63 mm)
sediment fraction was as expected, as much of the bulk sediment
mass consisted of sand and other large particles. As larger par-
ticles have a lower surface area to volume ratio, they will have a
lower binding affinity for metals, their presence in a sample
effectively diluting the concentrations of metals. The compar-
isons between the three extraction procedures suggest that
studies using ASMf-extracted metals may produce relatively
similar contaminant patterns to TPM. This is, however, a feature
of this particular study and there is no reason why this should be
a generic feature of data from other sites. For example, we know
that in nearby estuaries, e.g. Lake Macquarie, NSW, consider-
able fractions of the total metal concentration can be in coarse
Marine and Freshwater Research 1119
mineralised forms (Batley 1987). The ecological significance
of ASMf v. ASM deserves further consideration. There is good
evidence to suggest that only small particles are able to be
ingested by amphipods and similar benthic invertebrate feeders
(Tessier et al. 1984). This would suggest that the ASMf fraction
is the most ecotoxicologically important fraction, but the
organism sees this in a sediment diluted by coarse particles;
therefore, the exposure scenario is different from that in which
there is no coarse fraction.
Here we found that the CCA derived from the TPM data
(CCA-TPM) explained a similar amount of the total variance to
that explained solely by the other two metal extraction proce-
dures (ASM and ASMf). Quantification of metal concentrations
by total particulate metals is the traditional approach used in
environmental studies, including the development of sediment
quality guidelines (e.g. Long et al. 1998), although, at least
in Australia and New Zealand, the assessment hierarchy
then considers ASM if total metal guidelines are exceeded
(ANZECC/ARMCANZ 2000). In all three procedures (TPM,
ASM and ASMf), the relative proportion of variance that could
be explained purely by the ecological variables (grain size and
organic carbon) was relatively low, emphasising the overriding
influence of the contaminants. This is a case where statistical
interpretations need to be informed by our knowledge of con-
taminant bioavailability. As noted earlier, the insensitivity to
chemical form may be system-specific, simply reflecting the
observed correlations between TPM, ASM and ASMf. Irrespec-
tive of the final metal concentrations obtained by the three
extraction techniques, metals extracted using TPM (excluding
arsenic) were always significantly correlated with those
extracted using ASM and ASMf. Even though marked differ-
ences in the concentrations between ASM and TPM extractable
metals may have occurred, the correlative relationships between
the biotic and abiotic datasets were preserved, emphasising that
CCA is a correlation-based procedure.
AVS-SEM as a surrogate for porewater metals
(AVS-SEM v. TPM)
As previously indicated, the inclusion of porewater data pro-
vided critical information regarding the spatial distributions of
contaminants within the Parramatta River. When compared with
analyses that included porewater variables (e.g. CCA-TPM), the
substitution of these data with stratified AVS-SEM did little to
aid interpretation, as these variables were correlated with par-
ticulate contaminants, and not with concentrations of porewater
Cr, Cu and Ni. The redundancy in AVS-SEM measurements
was evident in the CCA analyses, with these variables having
no significant influence on the composition of macrobenthic
communities. Even though there is considerable evidence
that supports the AVS/SEM model in experimental treatments
(USEPA 2005), the ability of the model to predict metal toxicity
in sulfide-rich sedimentary environments, e.g. poorly biotur-
bated estuarine sediments, is questionable (Simpson and Batley
2007). In this study, benthic communities from several sites
within the Parramatta River were found to be strongly modified
by elevated porewater concentrations of copper and chromium.
The observation of environmentally significant porewater cop-
per and chromium concentrations despite an excess of AVS may
1120 Marine and Freshwater Research
be attributed, in part, to the existence of oxic/suboxic sediment
microenvironments (Motelica-Heino et al. 2003; Naylor et al.
2006) or the oxidation of metal sulfide phases in surface sedi-
ments (Ward et al. 2009). Note that when there is an excess of
AVS, chromium is predicted to be present as the less toxic
Cr(III) species (USEPA 2005). Furthermore, in cases where
metals may be deposited in other forms, e.g. metallic oxides
or salts, only the surficial layer of the metal will be sulfidised
(Simpson et al. 2000). In such cases, no consistent molar rela-
tionship between metals and reactive sulfides can be derived.
The direct relationship between the AVS and SEM is less
predictable in estuarine sediments than in sulfate-limited fresh-
water sediments as AVS concentrations are generally higher in
contaminated locations due to a decrease in infaunal activity
(bioturbation and bioirrigation), which in turn leads to more
reducing conditions and the production of more AVS. Conse-
quently, AVS-SEM is not a suitable surrogate for porewater
measurements in systems such as Sydney Harbour and its
surrounding estuaries.
Normalisation of organic contaminants (%TOC v. TPM)
Qualitative examination of the PCA for normalised TOC high-
lighted a within-estuary difference in the Parramatta and Lane
Cove Rivers, although this did not profoundly affect the inter-
pretation of the environmental data. The proportion of TOC
within sediments influences their sorption of non-polar organic
contaminants. Normalisation is designed to correct for this
phenomenon, enabling sediments with varying concentrations
of organic carbon to be compared (Di Toro and McGrath 2000).
While normalisation can assist in highlighting sites that may
have bound or potentially labile non-polar organic con-
taminants, its significance is predominately ecotoxicological.
Nevertheless, normalisation had no perceivable effect on the
amount of explainable variation in macrofauna data. This is
most likely attributable to two factors: first, organic con-
taminants were not the dominant contaminant, and second, there
was a low level of spatial variation in TOC. Consequently, only
for sites with relatively high or low proportions of TOC was the
influence of organic contaminants affected by these analyses.
Whilst it should not be inferred that normalisation of organic
contaminant concentrations is not important, in this case, its
application neither aids nor hinders the interpretation of either
the environmental or ecological findings.
Influence of particulate and porewater iron
and manganese
The inclusion of particulate and porewater iron and manganese in
the PCA assisted in discriminating the two external estuaries,
Hawkesbury and Georges Rivers. As manganese was excluded
from the CCA due to its high VIF, the relationships between the
particulate and dissolved forms of this metal and benthic com-
munity structure could not be examined. The addition of iron
appreciably increased the collinearity between the ecological and
contaminant variables, emphasising how this variable influences
both the ecological and contaminant datasets simultaneously.
Particulate iron and manganese oxyhydroxide phases are con-
sidered to be important metal-binding phases in sediments
(Chapman et al. 1998; Simpson and Batley 2007). The relative
A. A. Chariton et al.
proportions of Fe(II) and Mn(II) in the porewaters provides
information on the redox status of the sediments and how iron and
manganese phases may be affecting metal partitioning (Simpson
and Batley 2003). In this study, although iron and manganese may
provide an insight into geological and hydrological attributes of
some of the sites, their addition adds considerable noise to ana-
lysis and little ecotoxicological information.
Multimodel comparisons
The primary goal of both the principal component analyses and
the canonical correspondence analyses was to identify which
environmental dataset best described the spatial distribution
of contaminants with the study region, and the relationship
between these variables and the sampled macrobenthic inver-
tebrate communities. The findings of this study emphasise that it
is not the number of environmental variables, but rather, their
relevance in a particular system that is of greatest importance.
In some cases, the addition of variables can even reduce the
explanatory power of the analysis, e.g. iron. Conversely, the
exclusion of other variables, e.g. porewater metals, significantly
limits ecotoxicological interpretation.
The results from this specific study suggest that the choice of
metal extraction had little influence on the interpretation of the
data, even though both ASM and ASMf are more ecologically
relevant measurements than TPM. In addition, the relationship
between benthic communities and contaminants was not notice-
ably strengthened by the normalisation of organic contaminant
concentrations to TOC concentrations. Consequently, in this
study, any one of these four approaches (TPM, ASM, ASMf or
%TOC) could be considered suitable for evaluating the envir-
onmental data and their interactions with biota. There is,
however, several caveats associated with the generalisation of
these findings. First, in the present study, variation in grain size
was deliberately constrained by the experimental design; there-
fore, similarities between metal extraction techniques may not
be applicable to studies in sandy environments or where grain
size varies greatly among samples. Second, the influence of
TOC normalisation on the biotic data can be expected to be more
pronounced in systems where non-polar organic contaminants
are of primary concern. Nevertheless, the findings of the present
study provide evidence that contaminants, especially particulate
and porewater metals, are modifying the composition of macro-
benthic communities. Such evidence, albeit correlative, is an
essential component of a WOE approach.
Conclusion
The selection of which stressor variables to measure is critical to
understanding the spatial distribution of environmental vari-
ables and how these variables influence the structure of benthic
communities. Although not novel, partial CCA and similar
procedures are underutilised in routine environmental risk
assessments. The application of approaches such as those
described not only decrease the cost of any repeated monitoring
by reducing the collection of redundant or poorly correlated
variables, but also increases the potential for collecting data that
are the most relevant, thereby aiding interpretation and the
management options that ensue. For example, the contribution
of specific contaminants to observed changes in ecological data
Variables for contaminant–benthos relationships
can be examined in isolation, clarifying the relevance of eco-
logical lines of evidence within a WOE framework.
The approach and ultimate selection of variables will differ
among studies; however, the current study makes the following
conclusions:
 Environmental risk assessments of contaminated sediments
should contain porewater data. Such data not only aid the
identification of sites potentially contaminated with bioavail-
able metals, but also, as demonstrated here, can maximise the
relationships between abiotic and biotic matrices.
 AVS-SEM should not be used as a surrogate for measurements
of porewater metals, especially in sulfidic environments.
 In cases where strong correlations exist between metals
extracted using different procedures, e.g. TPM and ASM,
differences in the spatial patterns of environmental variables
and relationships between abiotic and biotic matrices may be
insignificant. However, this relationship may not be repre-
sentative of all studies and requires testing.
 Normalisation of organic contaminants is potentially impor-
tant; however, its influence on the biotic data should be
investigated on a case-by-case basis.
 The ad hoc inclusion of all variables in an analysis may hinder
interpretation. For example, we found that the inclusion of
particulate and dissolved iron increased the collinearity
between the ecological and contaminant variables.
 Multimodel comparisons using permutations of the same
dataset can be an invaluable tool for identifying both key
and redundant environmental variables, strengthening lines
of evidence between chemical and biotic responses, minimis-
ing costs, and for refining the experimental design of sub-
sequent sampling designs.
Acknowledgements
We wish to thank Cheryl Tang, Steve Jacobs, Richard Gardiner, Tim
Ingleton, Max Carpenter, Chris Rush and Joanne Ling who assisted in bio-
logical survey work and biota identification. David Spadaro and Ian
Hamilton undertook most of the physico-chemical characterisation, che-
mical analyses and assistance with biological survey work, and we thank
Edwin Peeters (Wageningen University) for his statistical advice. We also
wish to thank our CSIRO colleagues Jenny Stauber and Merrin Adams for
editorial comments and advice. Funding for this research was provided by
the NSW Environmental Trust (Research Project 22003/RD/G0002).
References
Ankley, G. T., Katko, A., and Arthur, J. W. (1990). Identification of
ammonia as an important sediment-associated toxicant in the lower
Fox River and Green Bay, Wisconsin. Environmental Toxicology and
Chemistry 9, 313–322. doi:10.1002/ETC.5620090307
Ankley, G. T., Schubauer-Berigan, M. K., and Dierkes, J. R. (1991).
Predicting the toxicity of bulk sediments to aquatic organisms with
aqueous test fractions: pore water vs. elutriate. Environmental Toxicol-
ogy and Chemistry 10, 1359–1366. doi:10.1002/ETC.5620101015
ANZECC/ARMCANZ (2000). ‘Australian and New Zealand Guidelines
for Freshwater and Marine Water Quality Vol. 1.’ (Australian and New
Zealand Environment and Conservation Council and Agriculture and
Resource Management of Australia and New Zealand: Canberra.)
Batley, G. E. (1987). Heavy metal speciation in waters, sediment and biota
from Lake Macquarie, New South Wales. Australian Journal of Marine
and Freshwater Research 38, 591–606. doi:10.1071/MF9870591
Marine and Freshwater Research 1121
Borcard, D., Legendre, P., and Drapeau, P. (1992). Partialling out the
spatial component of ecological variation. Ecology 73, 1045–1055.
doi:10.2307/1940179
Chapman, P. M., Wang, F. Y., Janssen, C., Persoone, G., and Allen, H. E.
(1998). Ecotoxicology of metals in aquatic sediments: binding and
release, bioavailability, risk assessment, and remediation. Canadian
Journal of Fisheries and Aquatic Sciences 55, 2221–2243. doi:10.1139/
CJFAS-55-10-2221
Chapman, P. M., Wang, F. Y., Germano, J. D., and Batley, G. E. (2002). Pore
water testing and analysis: the good, the bad and the ugly. Marine
Pollution Bulletin 44, 359–366. doi:10.1016/S0025-326X(01)00243-0
Clarke, K. R., and Ainsworth, M. (1993). A method of linking multivariate
community structure to environmental variables. Marine Ecology
Progress Series 92, 205–219. doi:10.3354/MEPS092205
Clarke, K. R., and Warwick, R. M. (1994). ‘Changes in Marine Commu-
nities: An Approach to Statistical Analysis and Interpretation.’
(Plymouth Marine Laboratory: Plymouth, UK.)
Cline, J. D. (1969). Spectrophotometric determination of hydrogen
sulfide in natural waters. Limnology and Oceanography 14, 454–458.
doi:10.4319/LO.1969.14.3.0454
Di Toro, D. M., and McGrath, J. A. (2000). Technical basis for narcotic
chemicals and polycyclic aromatic hydrocarbon criteria. II. Mixtures
and sediments. Environmental Toxicology and Chemistry 19, 1971–1982.
doi:10.1002/ETC.5620190804
Driscoll, S. B. K., and Burgess, R. M. (2007). An overview of the develop-
ment, status, and application of equilibrium partitioning sediment
benchmarks for PAH mixtures. Human and Ecological Risk Assessment
13, 286–301. doi:10.1080/10807030701226707
Frontier, S. (1976). Etude de la décroissance des valuurs propres dans une
analyse en composantes principales: comparaison avec le modéle du
‘bâton brisé’. Journal of Experimental Marine Biology and Ecology 25,
67–75. doi:10.1016/0022-0981(76)90076-9
Gray, J. S. (1974). Animal-sediment relationships. Oceanography and
Marine Biology Annual Reviews 12, 223–261.
Irvine, I., and Birch, G. F. (1998). Distribution of heavy metals in surficial
sediments of Port Jackson, Sydney, New South Wales. Australian Journal
of Earth Sciences 45, 297–304. doi:10.1080/08120099808728388
Kraan, C., van der Meer, J., Dekinga, A., and Piersma, T. (2009). Patchiness
of macrobenthic invertebrates in homogenized intertidal habitats: hidden
spatial structure at a landscape scale. Marine Ecology Progress Series
383, 211–224. doi:10.3354/MEPS07994
Legendre, P., and Legendre, L. (1998). ‘Numerical Ecology.’ 2nd edn.
(Elsevier: Sydney.)
Lepš, J., and Šmilauer, P. (2003). ‘Multivariate Analysis of Ecological Data
Using CANOCO.’ (Cambridge University Press: Melbourne.)
Long, E. R., Field, L. J., and McDonald, D. D. (1998). Predicting toxicity of
marine sediments with numerical sediment quality guidelines. Environ-
mental Toxicology and Chemistry 17, 714–727. doi:10.1002/ETC.
5620170428
Millward, R. N., Carman, K. R., Fleeger, J. W., Gambrell, R. P., Powell, R.,
and Rouse, M. A. M. (2001). Linking ecological impact to metal
concentrations and speciation: A microcosm experiment using a salt
marsh meiofaunal community. Environmental Toxicology and Chemis-
try 20, 2029–2037. doi:10.1002/ETC.5620200923
Motelica-Heino, M., Naylor, C., Zhang, H., and Davison, W. (2003).
Simultaneous release of metals and sulfide in lacustrine sediment.
Environmental Science & Technology 37, 4374–4381. doi:10.1021/
ES030035þ
Naylor, C., Davison, W., Motelica-Heino, M., van den Berg, G. A., and
van der Heijdt, L. M. (2006). Potential kinetic availability of metals in
sulphidic freshwater sediments. The Science of the Total Environment
357, 208–220. doi:10.1016/J.SCITOTENV.2005.04.034
Peeters, E. T. H., Dewitte, A., Koelmans, A. A., van der Velden, J. A., and
den Besten, P. J. (2001). Evaluation of bioassays versus contaminant
1122 Marine and Freshwater Research
concentrations in explaining the macroinvertebrate community structure
in the Rhine-Meuse Delta, The Netherlands. Environmental Toxicology
and Chemistry 20, 2883–2891. doi:10.1002/ETC.5620201231
Scouller, R. C., Snape, I., Stark, J. S., and Gore, D. B. (2006). Evaluation
of geochemical methods for discrimination of metal contamination in
Antarctic marine sediments: A case study from Casey Station. Chemo-
sphere 65, 294–309. doi:10.1016/J.CHEMOSPHERE.2006.02.062
Simpson, S. L. (2001). A rapid screening method for acid-volatile sulfide in
sediments. Environmental Toxicology and Chemistry 20, 2657–2661.
doi:10.1002/ETC.5620201201
Simpson, S. L., and Batley, G. E. (2003). Disturbances to metal partitioning
during toxicity testing of iron(II)-rich porewaters and whole sediments.
Environmental Toxicology and Chemistry 22, 424–432. doi:10.1002/
ETC.5620220225
Simpson, S. L., and Batley, G. E. (2007). Predicting metal toxicity in
sediments: A critique of current approaches. Integrated Environmental
Assessment and Management 3, 18–31. doi:10.1002/IEAM.5630030103
Simpson, S. L., Apte, S. C., and Batley, G. E. (2000). Effect of short term
resuspension events on the oxidation of cadmium, lead and zinc sulfide
phases in anoxic sediments. Environmental Science and Technology 34,
4533–4537.
Snelgrove, P. V., and Butman, C. A. (1994). Animal-sediment relationships
revisited: cause versus effect. Oceanography and Marine Biology: an
Annual Review 32, 111–177.
ter Braak, C. J. (1996). ‘Unimodel Models to Relate Species to Environ-
ment.’ (DLO-Agricultural Mathematics Group: Wageningen.)
ter Braak, C. J., and Verdonschot, P. F. M. (1995). Canonical correspon-
dence analysis and related multivariate methods in aquatic ecology.
Aquatic Sciences 57, 255–289. doi:10.1007/BF00877430
Tessier, A., and Campbell, P. G. C. (1987). Partitioning of trace metals
in sediments: relationships with bioavailability. Hydrobiologia 149,
43–52. doi:10.1007/BF00048645
Tessier, A., Campbell, P. G. C., Auclair, J. C., and Bisson, M. (1984).
Relationships between the partitioning of trace metals in sediments and
their accumulation in the tissues of the freshwater mollusc Elliptio
http://www.publish.csiro.au/journals/mfr
View publication stats
A. A. Chariton et al.
complanata in a mining area. Canadian Journal of Fisheries and Aquatic
Sciences 41, 1463–1472. doi:10.1139/F84-180
US Environmental Protection Agency (1996). ‘Test Methods for Evaluat-
ing Solid Waste, Physical/Chemical Methods.’ Methods and Risk
Analysis Division SW-846. (US Environmental Protection Agency:
Washington, DC.)
US Environmental Protection Agency (2003). ‘Procedures for the Deriva-
tion of Equilibrium Partitioning Sediment Benchmarks (ESBs) for the
Protection of Benthic Organisms: PAH Mixtures.’ Office of Research
and Development Report EPA-600-R-02-013. (US Environmental
Protection Agency: Washington, DC.)
US Environmental Protection Agency (2005).‘Procedures for the Derivation
of Equilibrium Partitioning Sediment Benchmarks (ESBs) for the
Protection of Benthic Organisms: Metal Mixtures (Cadmium, Copper,
Lead, Nickel, Silver and Zinc).’ Office of Research and Development
Report EPA-600-R-02-011. (US Environmental Protection Agency:
Washington, DC.)
Wang, F., and Chapman, P. M. (1999). Biological implications of sulfide
in sediment – a review focusing on sediment toxicity Environmental
Toxicology and Chemistry 18, 2526–2532.
Wang, W. X., Stupakoff, I., and Fisher, N. S. (1999). Bioavailability of
dissolved and sediment-bound metals to a marine deposit-feeding poly-
chaete. Marine Ecology Progress Series 178, 281–293. doi:10.3354/
MEPS178281
Ward, W., Strom, D., Simpson, S. L., and Jolley, D. F. (2009). Dietary
toxicity sulfide-bound copper to the benthic amphipod Melita plumu-
losa. Environmental Science & Technology.
Wenning, R. J., Batley, G. E., Ingersoll, C. G., and Moore, D. W. (2005).
‘Use of Sediment-quality Guidelines and Related Tools for the Assess-
ment of Contaminated Sediments.’ (Society of Environmental Toxicol-
ogy and Chemistry: Pensacola, FL.)
Manuscript received 19 October 2009, accepted 14 April 2010