-
Population modeling using harpacticoid
copepods
Bridging the gap between
individual-‐level effects and protection
goals of environmental risk
assessment
Elin Lundström Belleza
Doctoral thesis in Applied Environmental
Science Department of Applied
Environmental Science (ITM)
Stockholm University 2014
-
2
Doctoral Thesis, 2014 Elin Lundström
Belleza Department of Applied
Environmental Science (ITM) Stockholm
University SE-‐106 91 Stockholm,
Sweden
©Elin
Lundström Belleza, Stockholm 2014
ISBN, 978-‐91-‐7447-‐894-‐5 pp. 1-‐36
Printed in Sweden by
Universitetsservice US-‐AB, Stockholm, 2014
Distributor: Department of Applied
Environmental Science (ITM) Cover by
Gian Carlo Belleza, including artwork
by Göte Göransson.
-
3
To Carlo, Wellington and Winter
Alba
-
4
-
5
Abstract
To protect the environment from
contaminants, environmental risk assessment
(ERA) evaluates the risk of
adverse effects to populations,
communities and ecosystems. Environmental
management decisions rely on
ERAs, which commonly are based
on a few endpoints at the
individual organism level. To bridge
the gap between what is
measured and what is intended
for protection, individual-‐level effects
can be integrated in population
models, and translated to the
population level. The general aim
of this doctoral thesis was
to extrapolate individual-‐level effects
of harpacticoid copepods to the
population level by developing
and using population models. Matrix
models and individual based models
were developed and applied to
life-‐history data of Nitocra
spinipes and Amphiascus tenuiremis,
and demographic equations were used
to calculate population-‐level effects
in low-‐ and high-‐density
populations. As a basis for the
population models, individual-‐level
processes were studied. Development
was found to be more sensitive
compared to reproduction in
standard ecotoxicity tests measuring
life-‐history data. Additional experimental
animals would improve statistical
power for reproductive endpoints,
but at high labor and cost.
Therefore, a new test-‐design was
developed in this thesis. Exposing
animals in groups included a
higher number of animals without
increased workload. The number of
reproducing females was increased,
and the statistical power of
reproduction was improved.
Individual-‐level effects were more
or equally sensitive compared to
population-‐level effects, and
individual-‐level effects were translated
to the population level to
various degrees by population models
of different complexities. More
complex models showed stronger effects
at the population level compared
to the simpler models. Density
dependence affected N. spinipes
populations negatively so that
toxicant effects were stronger at
higher population densities. The
tools presented here can be
used to assess the toxicity
of environmental contaminants at
the individual and population level,
improve ERA, and thereby the
basis for environmental management.
-
6
Svensk sammanfattning För att
skydda miljön från föroreningar
utvärderar miljöriskbedömningar risken för
negativa effekter på populationer,
samhällen och ekosystem. Riskhantering
är beroende av miljöriskbedömningar,
vilka ofta är baserade på ett
fåtal mätvärden på individorganismnivå.
För att överbrygga klyftan mellan
det som mäts och vad som
är avsett att skyddas, kan
effekter på individnivå integreras
i populationsmodeller, och översättas
till populationseffekter. Det övergripande
syftet med denna avhandling var
att extrapolera effekter på
individnivå från harpacticoida copepoder
till populationsnivå genom att utveckla
och använda populationsmodeller.
Matrismodeller och individbaserade modeller
utvecklades och tillämpades på
livshistoriedata för Nitocra spinipes
och Amphiascus tenuiremis, och
demografiska ekvationer användes för
att beräkna effekter på
populationsnivå i låg -‐ och
högdensitetspopulationer. Som underlag för
populationsmodellerna studerades processer
på individnivå. Utveckling visade
sig vara känsligare än reproduktion i
standardiserade ekotoxicitetstester som
mäter livshistoriedata. Ytterligare
försöksdjur skulle förbättra den
statistiska känsligheten för reproduktion,
men med ökad arbetsinsats och
kostnad som följd. Därför
utvecklades en ny testdesign i
denna avhandling. Exponering av försöksdjur
i grupper gjorde det möjligt
att inkludera ett större antal
djur utan ökad arbetsbörda, och
en statistiska känsligheten för
reproduktion förbättrades. Effekter på
individnivå var mer eller lika
känsliga i jämförelse med effekter
på populationsnivå, och översattes
till populationsnivå i olika grad
av populationsmodeller av olika
komplexitet. Mer komplexa modeller
visade starkare effekter på
populationsnivå jämfört med de enklare
modellerna. Densitetsberoende påverkade
populationer av N. spinipes, så
att de toxiska effekterna var
starkare vid högre populationsdensitet.
De verktyg som presenteras i
denna avhandling kan användas för
att bedöma toxiciteten av
miljöföroreningar på populationsnivå,
förbättra miljöriskbedömningar, och därmed
grunden för riskhantering.
-
7
List of papers Paper I
Lundström, E.; Björlenius, B.;
Brinkmann, M.; Hollert, H.; Persson,
J-‐O.; Breitholtz, M. Comparison
of six sewage effluents treated
with different treatment technologies-‐
Population level responses in the
harpacticoid copepod Nitocra spinipes.
Aquatic Toxicology. 2010, 96 (4),
298-‐307; DOI 10.1016/j.aquatox.2009.11.011
Paper II Preuss, T.G.;
Brinkmann, M.; Lundström, E.;
Bengtsson, B-‐E.; Breitholtz, M. An
individual-‐based modeling approach for
evaluation of endpoint sensitivity in
harpacticoid copepod life-‐cycle tests
and optimization of test design.
Environmental Toxicology and Chemistry.
2011, 30 (10), 2353-‐2362; DOI
10.1002/etc.614 Paper III Lundström
Belleza, E.; Brinkmann, M.;
Preuss, T.G.; Breitholtz, M.
Population-‐level effects in Amphiascus
tenuiremis: Contrasting simple and
complex population models. Submitted to
Aquatic Toxicology. Paper IV
Lundström Belleza, E.; Breitholtz, M.
Density-‐toxicant interactions and
reproductive responses in Nitocra
spinipes. Manuscript.
Statement I made the following
contributions to the papers presented
here: Paper I I took the
lead role in planning and
carrying out the ecotoxicity tests.
Experiments were carried out
together with one of the
co-‐authors and technicians. I took
the lead role in analyzing
the data and constructing the
matrix model, and I took a
large part in simulating in the
model. I took the lead role
in writing the paper. Paper
II I took a major part
in data synthesis for the model.
Co-‐ authors programmed and simulated
in the individual based model.
I took a minor part of
writing the paper. Paper III
I took the lead role in
data synthesis for the models,
and also took the lead role
in the matrix model simulations.
Co-‐authors programmed and simulated
in the individual based model.
I took the lead role in
writing the paper. Paper IV
I took the lead role in
planning and performing the
ecotoxicity tests. Experiments were
carried out together with
technicians. I took the lead
role in analyzing the data and
writing the paper.
-
8
Contents Abstract
...............................................................................................................................................
5 Svensk sammanfattning
................................................................................................................
6 List of papers
.....................................................................................................................................
7 Statement
...........................................................................................................................................
7 Abbreviations
...................................................................................................................................
9 1. Introduction
...............................................................................................................................
10 2. Aim and hypotheses of
the thesis
.......................................................................................
11 3. Background
................................................................................................................................
11 3.1 Environmental risk assessment
(ERA)
.....................................................................................
11 3.2 Test methods and
organisms
.......................................................................................................
12 3.2.1 Ecotoxicity tests
...........................................................................................................................................
12 3.2.2 Individual-‐level endpoints
......................................................................................................................
12 3.2.3 Harpacticoid copepods
.............................................................................................................................
13
3.3 Population models
..........................................................................................................................
14 3.3.1 Unstructured models
.................................................................................................................................
15 3.3.2 Biologically structured
models
.............................................................................................................
15 3.3.3 Individual based models
(IBMs)
...........................................................................................................
15 3.3.4 Population-‐level endpoints
.....................................................................................................................
16
3.4 ERA and density dependence
.......................................................................................................
17 4. Material and Methods
.............................................................................................................
19 4.1 Test organisms
.................................................................................................................................
19 4.2 Test substances
................................................................................................................................
19 4.3 Test methods
.....................................................................................................................................
19 4.3.1 Cohort experiments
...................................................................................................................................
19 4.3.2 Time-‐series experiments
.........................................................................................................................
20 4.3.3 Population models
......................................................................................................................................
20 4.3.4 Measure of adverse
effects
......................................................................................................................
22
5. Results and Discussion
...........................................................................................................
23 5.1 Model development
........................................................................................................................
23 5.2 Contrasting individual-‐ and
population-‐level effects
..........................................................
24 5.3 Statistical power and
replicates for reproductive endpoints
........................................... 25
5.4 Contrasting simple and complex
modeling approaches
.................................................... 26
5.5 Contrasting effects in low-‐
and high-‐density populations
................................................. 27
6. Conclusions
................................................................................................................................
29 7. Future perspectives
................................................................................................................
29 Acknowledgement – Tack!
........................................................................................................
30 References
......................................................................................................................................
32
-
9
Abbreviations λ Lambda, finite
rate of increase, population growth
rate
A Adult stage
CI Copepodite stage one
CV Copepodite stage five
EFSA European Food Safety Authority
EC10 Effect Concentration at 10 %
ERA Environmental (ecological) Risk
Assessment
NI Naupliar stage one
NOEC No Observed Effect Concentration
NVI Naupliar stage six
IBM Individual Based Model
LOEC Lowest Observed Effect
Concentration
LTRE Life-‐Table Response Experiment
MM Matrix Model
PCB PolyChlorinated Biphenyls
PEC Predicted Environmental Concentration
PNEC Predicted No Effect Concentration
r Intrinsic/instantaneous rate of
increase, population growth rate
-
10
1. Introduction A vast number
of anthropogenic substances are used
in society today. As an
example, there are more than
143 000 industrial chemicals
pre-‐registered for commercial use in
the European Union (ECHA, 2014).
To protect the environment from
adverse effects of environmental
pollutants, environmental risk
assessment (ERA) is used as a
tool for protecting populations,
communities and ecosystems (e.g.
European Commission, 2003; EMEA,
2006; van Leeuwen and Vermeire,
2007). The effects of
environmental pollutants are commonly
estimated from the results of
standard laboratory (eco)toxicity tests.
In order to detect adverse
effects in these tests, it
is important that the measured
endpoints have high statistical power
and that endpoints are sensitive.
Using many replicates or test
concentrations commonly increases
statistical power, but at increased
labor and cost. Ecotoxicity tests
are often performed on individually
exposed animals, and the effects
are measured on for example
survival, development and reproduction.
In ERA, it is therefore
assumed that data from simple
ecotoxicity tests can be used
to estimate risk for the
ecological entities intended for
protection (e.g. Forbes et al.,
2001). In this context, population
models are useful since they
can bridge the gap between what
is measured and what is
intended for protection (e.g.
Barnthouse et al., 2008; Forbes
et al., 2008), and can be
used to reduce uncertainty in
extrapolation of (standard) test
results to ecologically relevant
effects (Forbes et al., 2011). There
is a range of population
models available in the scientific
literature that has been used
to assess the risk of chemicals
to many different organisms (e.g.
Pastorok et al., 2002; Akcakaya
et al., 2008). Even though the
most sensitive individual-‐level endpoints
are likely to be equally or
more sensitive to stressors, such
as environmental pollutants, than
effects on the population level
(Forbes and Calow, 2002), the
relationship is sometime reversed
(Forbes and Calow, 1999).
Moreover, effects that are measured
on isolated individuals, at low
population density, ignore density
dependence, which in natural
populations may affect the responses
(e.g. Forbes et al., 2001).
Experiments carried out at low
densities may underestimate population
stress responses compared to
high-‐density populations due to the
lack of interaction between density
and toxicity (e.g. Sibly, 1999),
or overestimate the effects due
to compensation in high-‐density
populations (Forbes et al., 2001).
Models for calculating concentrations
of pollutants in the environment
have long been used in
directives relating to risk
assessment of chemicals (e.g.
Hommen et al., 2010). Ecological
models, including population models,
are however lagging behind, and
stakeholders name e.g. the lack
of guidance on how to choose
and use population models as a
reason why they are not put
to practice in ERA (Hunka et
al., 2013). Contrasting population
models of differing complexity may
aid risk assessors in choosing
what population model to use
(Meli et al., 2014). In
the last years, population models
have been included in several
directives and their related
guidance documents (e.g. EFSA, 2009,
2010; SCENIHR 2012; EFSA, 2013,
2014), bringing on the new era
in ERA.
-
11
2. Aim and hypotheses of the
thesis The overall aim of
this doctoral thesis was to
extrapolate individual-‐level ecotoxicological
effects to the population level
by developing and using population
models for harpacticoid copepods.
The hypotheses were that:
• Individual-‐ and population-‐level effects
are found in the same
concentration range for copepods exposed
to single substances and mixtures
(papers, I, III and IV).
• The number of replicate animals
can be increased without a
higher workload by grouping
of animals, which will in
turn increase fertilization success and
statistical power of reproductive
endpoints (papers II and IV).
• Simple stage-‐based matrix population
models do not translate
individual-‐level effects on
development time to the population
level, to the same degree
as individual based population models
(paper III).
• Toxic effects in harpacticoid
copepods are negatively influenced
by population density
(paper IV).
3. Background
3.1 Environmental risk assessment (ERA)
To protect the environment, risk
management decisions are based on
environmental (ecological) risk assessment
(ERA). ERA is the process
for evaluating the risk that the
environment will be impacted as
a result of exposure to
environmental pollutants. ERA is
normally a tiered process that
in lower tiers focuses on
“worst-‐case” scenarios, and, for
substances that initially resulted in
unacceptable adverse effects, proceeds
to more realistic assessments at
higher tiers (e.g. European
Commission, 2002; van Leeuwen and
Vermeire, 2007). Environmental fate
and exposure of environmental
pollutants are often estimated using
models, or when available, on
environmental measurements (e.g.
European Commission, 2003). Effects of
environmental pollutants, on the
other hand, are estimated from
the results of laboratory
(eco)toxicity tests or sometimes on
mesocosm or field studies.
Standard test data from the
laboratory are still preferred and
recommended for ERA (e.g. European
Commission 2002; 2003), even though
non-‐standard data could improve
the scientific basis by providing
relevant and more sensitive
endpoints (Ågerstrand et al.,
2013). Standard tests are performed
using established and validated
protocols, which is why they are
considered more reliable than
non-‐standard data. However, different
evaluation protocols for peer-‐reviewed
data exist, and reporting data
in a sufficiently detailed manner
would facilitate the use of
non-‐standard data for ERA
(Ågerstrand et al., 2013). To
estimate the risk for the
environment, risk-‐quotients are used.
They consist of the predicted
environmental concentration (PEC) of
a substance, divided by the
predicted no-‐effect concentration (PNEC),
which is based on
ecotoxicological tests. To reflect
uncertainties (e.g. intra-‐ and
inter-‐species variations, the extrapolation
from short-‐term toxicity to
long-‐term toxicity and the extrapolation
of laboratory test results towards
the field), PNECs are combined
-
12
with uncertainty factors (OECD, 2011a;
ECHA, 2012). Commonly, the most
sensitive endpoints derived from
ecotoxicological testing are used for
the assessment (European Commission,
2002; 2003). In ERA it is
therefore assumed that data on
direct effects (on e.g. survival,
development and reproduction) in
simple toxicity tests (combined with
uncertainty factors) reflect effects
on the population level and can
be used to protect populations,
communities and ecosystems (e.g.
Forbes et al., 2001). To
bridge the gap between test-‐endpoints
performed on individual organisms
and the ecological entities intended
to be protected by ERA,
population models have an important
role to play (Forbes et
al., 2008; EFSA, 2010). Population
models integrate potentially complex
interactions among life-‐history
traits, such as mortality, development
and reproduction. In this way,
they include ecological complexity,
and can reduce uncertainties in
extrapolation of individual-‐level test
endpoints to ecologically relevant
impacts (Forbes et al., 2011).
Ignoring endpoints above the
individual-‐level often leads to an
overestimation of risk, but sometimes
to underestimations (Forbes and
Calow, 1999). Using population models
in ERA could therefore lead
to distributing resources better and
more efficiently in environmental
risk management (e.g. Pastorok et
al., 2002; Barnthouse et al.,
2008). There is currently no
regulatory framework for ERA based
on ecological modeling, but
suggestions on how such an
approach could be structured are
given in e.g. Pastorok et al.
(2002) and Wentzel et al.
(2008). Population models are
however more and more mentioned
in European directives and guidance
documents on ERA (e.g. EFSA
2009, 2010; SCENIHR 2012; EFSA
2013), and there is also
a guidance document on good modeling
practice for risk assessment of
plant protection products (EFSA,
2014).
3.2 Test methods and organisms
3.2.1 Ecotoxicity tests Ecotoxicity
tests study the effects of
single toxicants or mixtures
(stressors) on organisms. There are
a large variety of test
methods, and tests can be
performed on different levels of
organization, from subcellular through
individual organisms to populations.
Acute toxicity tests are short
tests (hours or days) that
generally measure lethality as a
response. Concentrations of test
substance are usually higher in
acute tests compared to chronic
tests. Chronic (or long-‐term tests)
generally cover a significant
fraction of the life cycle
(weeks, months or years), and
concentrations of test substance are
lower in order to measure
sub-‐lethal endpoints. Life-‐table response
experiments (LTREs) are often used
in chronic testing, and commonly
follow animals from newborn until
they reproduce (Caswell, 2001).
Ecotoxicity tests can also be
performed directly on populations (e.g.
Sibly, 1999), or communities in
e.g. mesocosm studies or in the
field (European Commission, 2002).
3.2.2 Individual-‐level endpoints Organism
attributes commonly used as endpoints
in ERA include life-‐history rates,
which are the rates of birth,
growth, development, fertility and
mortality, and describe the
movement of individuals through the
life cycle (Caswell, 2001).
LTREs measure how single toxicants
or mixtures affect life-‐history
rates. Other individual-‐level endpoints
are e.g. body size and
physiological characteristics such as
respiration, food intake and
metabolic rate (Menzie et al.,
2008).
-
13
3.2.3 Harpacticoid copepods Invertebrates
account for approximately 95 %
of all known species on
Earth (Wilson, 1999). Crustaceans are
the second largest invertebrate
subphylum after insects, including
some 35.000 classified species.
Harpacticoid copepods are a subclass
of crustaceans, comprising 3.000
species. They usually make up
the second most abundant group of
animals in marine benthic
communities (Huys et al., 1996),
and are a primary food source
for juvenile fish (Hicks and
Coull, 1983). N. spinipes is
a harpacticoid copepod, widely
distributed in shallow waters around
the world (Lang, 1948). It
acclimatizes to fluctuations in
salinity (0-‐30 %0) and temperature
(0-‐26 0C) (Noodt, 1970; Wulff,
1972), and can therefore be
used for testing of various
environmental conditions. N. spinipes
is well suited for long-‐term
(chronic) ecotoxicity testing since
it is small (adults < 1 mm
long, Abraham and Gopalan, 1975),
reaches sexual maturity in 10-‐12
days and completes a life
cycle in 16-‐18 days at 20
0C (Dahl, 2008). N. spinipes
molts and sheds an exoskeleton
between each developmental stage.
It has six naupliar stages
(NI to NVI) and six
copepodite stages (CI to CV +
A; the reproducing adult stage)
(Figure 1). Between stages NVI
and CI the animals complete
a metamorphosis with profound changes
to body shape and segmentation.
Amphiascus tenuiremis. (e.g. Chandler
et al., 2004) is another
harpacticoid copepod, closely related to
N. spinipes. As opposed to the
more commonly used test organism,
the water flea Daphnia magna
that reproduces asexually, N.
spinipes and A. tenuiremis are
sexually reproducing species. This
introduces ecological relevance into
ecotoxicity testing since both sexes
are present, and introduced stressors
can potentially affect reproductive
behavior.
Figure 1: N. spinipes,
six naupliar stages and ovigerous
female. Illustrations by Göte Göransson,
modified by Gian Carlo Belleza.
-
14
3.3 Population models The overall
purpose of population models is
to evaluate the ecological
significance of observed or estimated
effects on individual organisms (e.g.
Pastorok et al., 2002). There
is a broad range of population
models that have been applied
to address ecotoxicological problems
(e.g. Schmolke et al., 2010),
even though they are not
commonly used in ERA. Population
models are however used for
decision-‐making in conservation ecology
and fisheries-‐management (EFSA, 2010;
SCENIHR, 2012). Figure 2 (modified
from Munns et al., 2008)
describes the three classes of
models dealt with in this
thesis; unstructured, biologically structured
and individual based models
(IBMs). Several other types of
population models also exist, including
metapopulation models that consider
many subpopulations of a species
that interact through migrations
(Munns et al., 2008). Spatially
explicit models focus on the
environment that a species inhabits,
and the variability therein (Munns
et al., 2008). These models are
demographic and study e.g. the
size, structure and distribution
of populations. Dynamic energy budget
models describe how individuals in
different classes assimilate energy
from food and use it for
maintenance, growth, reproduction, and
development (Nisbet et al., 2000).
Figure 2: Taxonomy of population
models for population-‐level ERA.
Modified from Munns et al.
(2008). Models are
simplifications of real systems, and
it is important to understand
their limitations and applicability.
Specificity and testability of
predictions from simple models is
low (Topping et al., 2005),
which makes it difficult to
define how well they describe
the system they are supposed to
simulate. Depending on the research
question, and the available data,
simple models can still be
useful (Topping et al., 2005).
Predictions from complex models
have higher specificity and testability,
and can often be tested
to make sure they are realistic
enough to meet their intended
purpose (Augusiak et al., 2014;
EFSA, 2014).
!"#$%!&$!%'()*+(',#)
-"(-.-(!/,#)-('"0&/,)
!"#"$%&'%()&$&*)+,$%
,**-"*,.&/%
-"(-.-(!/,#)!"-1!')
!"#"$%&'%01,.,$%
,**-"*,.&/%
#2/0/,,3)&+*2,'4)
#2/0/,,3))!"-5+%*)
-"(-.-(!/,)6/#'()*+(',#)
6-+,+7-&/,,3)#$%!&$!%'()*+(',#)
-
15
3.3.1 Unstructured models The simplest
form for assessing population-‐level
risk is by using unstructured
models. Individuals within the
population are treated identically in
terms of their life history
rates which usually only consists
of births and deaths, measured
as population size (Munns et
al., 2008). Data for unstructured
models can be obtained from
time-‐series experiments in which
population size is sampled over
time (Sibly, 1999; Moe, 2008).
These models can be either
deterministic (no randomness) or
include stochasticity. Stochastic models
are founded on the properties
of probability so that given
input produces a range of
possible outcomes due to random
effects (Pastorok et al., 2002).
Unstructured models can also either
assume exponential growth, which is
defined as density-‐independent growth
under no limitation of
resources, or include environmental
carrying capacity. The concept of
carrying capacity describes the
maximum size (density) of a
population that the environment they
live in can sustain (Pastorok
et al., 2002). At the
environmental carrying capacity,
density-‐dependent processes will affect
births and deaths in the
population (Moe et al., 2008).
Unstructured models can also be
continuous or discrete, where time
is treated incrementally. These
kinds of models have been used
to investigate adverse effects of
chemicals. As an example,
Hendriks and Enserink (1996)
investigated the change in abundance
of D. magna populations in
response to polychlorinated biphenyls
(PCBs). These models are very
generalized and have low data
requirements, why they are useful
mostly in lower tiers (screening)
of ERA (Munns et al., 2008).
3.3.2 Biologically structured models
Biologically structured models divide
individuals within the population
into distinct classes, and
incorporate biological structure by
assigning those classes with
life-‐history rates of mortality,
development and reproduction (Munns
et al., 2008). Classes can
be identified on the basis
of age, developmental stage or
size. Biologically structured models
are commonly matriarchal, meaning that
once in the reproductive state,
only females are included since
they are the only ones
contributing to population growth in
future generations (Caswell et al.,
2001). Life-‐history rates from
different age-‐ or stage classes
are commonly obtained from LTREs
where animals are exposed to
control treatments and to stressors
(e.g. chemicals). Biologically structured
models are often density-‐independent
and deterministic, but can
include density dependence (e.g. Grant
1998) as well as environmental
(e.g. Hamda et al., 2014) and
demographic stochasticity Munns et
al., 2008). One of the most
commonly used biologically structured
models is the Euler-‐Lotka equation
(e.g. Calow and Sibly, 1990;
Sibly, 1999), which has been
used to investigate contaminant
effects on population growth rate.
A few examples include effects
on population growth rate in D.
magna from titanium oxide
nanoparticles (Jacobasch et al.,
2014) and effects of synthetic
musks for N. spinipes (Breitholtz
et al., 2003). Matrix models
(MMs) have been used for
studying the effects of increasing
environmental copper concentrations to
the earthworm Lumbricus rubellus
(Klok and de Roos, 1996), and
in A. tenuiremis for studying
population consequences of the
insecticide fipronil (Chandler et al.,
2004) and crude oil (Bejarano
et al., 2006). Biologically
structured models can be almost
as general as unstructured models,
or more specific depending on
the research question addressed
and relevant detailing regarding e.g.
environmental and demographic
stochasticity and density dependence.
These models can therefore be used
for screening as well as
higher-‐tiers of ERA (Munns et
al., 2008).
3.3.3 Individual based models (IBMs)
IBMs, also called agent based
models, focus on the individual
as the basic element of
populations. These models track
the characteristics of each individual
(all sexes and stages) through
time and assume that individuals
can differ with respect to
their behavioral and physiological
responses to the environment
(Munns et al., 2008; SCENIHR,
2012). Individual
-
16
variability is key in IBMs,
and different individuals have
different probabilities of e.g.
survival, growth and reproduction (Munns
et al., 2008). IBMs are
mechanistic in their nature and
implement simple behavioral rules
that give rise to complex
behavior. In this way, effects
from e.g. toxicant stress on
physiological processes and individual
behavior are modeled. In more
aggregated models, such as
unstructured and biologically structured
models, these effects are indirectly
measured as effects on e.g.
survival and reproduction (Munns et
al., 2008). Individual variability
is often modeled as probability
distributions from which individual
events and their realizations are
drawn (Munns et al., 2008).
IBMs can have specific assumptions
related to the life cycle of
the species being modeled, which
can result in high specificity
and realism (Munns et al.,
2008). IBMs have been produced
for a large variety of
organisms, both animals and plants,
and Grimm (1999) reviews some
50 IBMs for animal populations
alone. IBMs can range from
spatially uniform such as the
IBM for D. magna (Preuss et
al., 2009) or to spatially
explicit such as the IBM for
Skylark (Topping et al., 2005).
IBMs have a broad range
of applications and have been
used to study e.g. how soil
contamination of different spatial
heterogeneity affects population dynamics
of soil invertebrates (Meli et
al., 2013), population-‐level effects
of PCBs on largemouth bass
(Micropterus sulmoides) (Jaworska et
al., 1997), and to predict the
population capacity and extinction
probability of D. magna exposed
to 3.4-‐dichloroaniline at laboratory
conditions (Preuss et al., 2010).
IBMs are well suited for
higher-‐tier ERA because of their
high level of ecological realism
and their flexibility to include
e.g. various environmental conditions
(Forbes et al., 2011).
3.3.4 Population-‐level endpoints There
are several important population
attributes that can be measured
and used as endpoints in ERA
(Menzie et al., 2008). Population
abundance is the size of
the population, measured as the
number of individuals or the
biomass of the population. Population
density is a related term,
which describes the size of
the population per unit of
habitat (area or volume).
Population growth rate is generally
thought of as the key
intervening variable linking individual
level effects to effects on
populations (e.g. Calow et al.,
1997; Caswell, 2001), and integrates
effects on survival, development
and reproduction, (Forbes and Calow,
1999). Population growth rate is
best expressed on a per capita
basis, and there are two ways
in which population growth rate
can be reported. Expressed as
the finite rate of increase (λ)
the population growth rate describes
how much the population has
potential to grow or shrink
in the next time step. Multiplying
λ with the population size
projects the population size in
the next time step. In practice
λ > 1 indicates a
growing population, λ < 1 a
shrinking population and λ =
1 a stable population. Expressed
as the intrinsic or instantaneous
rate of increase (r) population
growth rate describes the potential
of the population, in each
instant, to contribute to how
much the population grows or
shrinks. In practice r > 0
indicates a growing population, r
< 0 a shrinking population
and r = 0 a stable
population. These two measures of
population growth are related so
that the natural logarithm of
λ is r (!"# = !; !! = !)
(Sibly, 1999; Menzie et al.,
2008). In this thesis, λ
and r are both called the
population growth rate, and
distinguished by their symbols when
necessary. Population structure is
commonly the distribution of
individuals with respect to age
or developmental stages, sex,
reproductive status and so on.
Population dynamics describes how
population structure varies over
time. Population structure can both
influence and be an indicator
of the dynamics of the
population since life-‐history rates
such as survival and reproduction
often vary across the life
cycle (Menzie et al., 2008).
Other population attributes could be
related to, for example,
extinction and recovery of populations
or their spatial distribution.
-
17
3.4 ERA and density dependence
Standard tests used for ERA
are commonly performed on
isolated individuals in low-‐density
populations (Sibly, 1999), and in
ERA, the concept of density
dependence offers some challenges.
Density dependence is a
fundamental concept in population
biology, affecting the responses of
most animals and plant species
(Moe, 2008). Crowding is a
concept that can create
density-‐dependent effects in
populations (e.g. Gergs et al.,
2014), especially at laboratory
conditions. Crowding can lead to
decreased feeding rate at higher
animal density, due to
inter-‐individual interactions. Most
natural populations are likely to
be in steady state (i.e. not
growing or declining) whereas
populations used in toxicity testing
often grow exponentially (Forbes et
al., 2001). An important problem
for ERA is, therefore, that
populations under density dependence
(high-‐density populations) may respond
differently to toxicants than
those growing exponentially (low-‐density
populations) (Forbes et al.,
2001; Forbes and Calow, 2002).
Sibly (1999) suggests that high-‐density
populations are likely to be
more sensitive to toxicants than
low-‐density populations, because of
generally lower fitness, due to
increased competition over resources.
Forbes et al. (2001), on
the other hand, suggest that
compensation in high-‐density populations
could make the populations less
sensitive to toxicants. Compensation
is a process where density
reductions caused by increasing
toxicant concentrations would be
compensated by an amelioration of
density-‐dependent effects. Interactions
between density dependence and toxic
stress can be broadly categorized
in antagonistic (effects of the
toxicant is weaker at higher
population densities), additive
(toxicant effects are not affected
by density) and synergistic
(effects of the toxicant are
stronger at higher densities) (Forbes
et al., 2001; Moe, 2008)
(Figure 3).
-
18
Figure 3: Possible interactions
between density-‐dependent effects and
toxicant exposure on population
growth rate. Low and High
refers to populations of low
and high densities, respectively. The
slopes of the “low” curves are
held straight since they represent
no density dependence but only
toxicant effect. Modified from Forbes
et al. (2001). The
interactions between density dependence
and toxicity are not straightforward,
and may be affected by, for
example, the initial age-‐or stage
structure of the population (Stark
and Banken, 1999), and whether
populations are growing or declining
(Forbes and Calow, 1999). Since
there are so many factors
affecting the density-‐toxicant interactions,
they are difficult to foresee,
which is why experimental approaches
have to be taken (Forbes et
al., 2001).
!"#$
%&'"
()*+",
-.)+&
-/)
0"(1/(-+&'"()
2334'5/)
!"#$
%&'($
!"#$
%&'"
()*+",
-.)+&
-/)
0"(1/(-+&'"()
2(-&*"(46'1)
!"#$
%&'($
!"#$
%&'"
()*+",
-.)+&
-/)
0"(1/(-+&'"()
78(/+*46'1)
!"#$
%&'($
-
19
4. Material and Methods
4.1 Test organisms Harpacticoid copepods
are well suited for long-‐term
testing due to their small size
and relatively short life cycle
(Dahl, 2008). They are also
relevant test-‐species since they are
abundant in many different ecosystems
around the world (Lang, 1948).
In all studies, the harpacticoid
copepods N. spinipes (ecotoxicity
tests in papers I and IV,
modeling in paper II) and A.
tenuiremis (modeling in paper III)
were used. Culturing conditions and
handling of N. spinipes as
laboratory animals has been
published elsewhere (e.g. Breitholtz
and Bengtsson, 2001; Breitholtz et
al., 2003). A. tenuiremis as
a test species has been
described in e.g. Coull and
Chandler (1992) and Chandler et
al. (2004). In all laboratory
tests performed for this thesis,
the animals were fed with
the red microalga Rhodomonas salina.
Test medium was natural brackish
water filtered through 0.03-‐mm,
pre-‐heated to 80 °C and GF/C
(glass microfiber)-‐filtered.
4.2 Test substances In paper I
municipal sewage effluent from
Henriksdal sewage treatment plant in
Stockholm was tested on N.
spinipes. The sewage was treated
with conventional and novel sewage
treatment technologies aimed at
removing pharmaceuticals. The IBM
for N. spinipes in paper II
was developed using control data
from paper I, an OECD
validation report (OECD, 2007), and
Dahl and Breitholtz (2008).
The IBM for A. tenuiremis in
paper III was developed using
control data on A. tenuiremis
from Chandler et al. (2004) and
an OECD validation report (OECD,
2011b). Effects from lindane (OECD,
2011b) were simulated in the
model. Lindane is a
gamma-‐hexachlorocyclohexane used mainly as
an insecticide. In paper
IV, lindane was used as model
substance and tested on N.
spinipes since it previously showed
clear effects in harpacticoid
copepods (e.g. Dahl and Breitholtz,
2008).
4.3 Test methods Two main types
of experiments were performed
for this thesis; cohort and
time-‐series experiments.
4.3.1 Cohort experiments Cohort data
is obtained from experiments that
follow even-‐aged groups of
organisms through (parts of) their
life cycle (Moe, 2008). LTREs
and life-‐cycle tests are two
terms often used to describe
experiments that produce cohort
data, and they are performed on
low-‐density populations (commonly
isolated individuals) (Sibly, 1999).
Two different kinds of cohort
experiments were performed in this
thesis: In paper I,
newborn nauplii (NI) were
isolated in wells in 96-‐well
micro plates, and their development
to the first copepodite stage
(CI), and adulthood (A) was
recorded as the number of days
the development took. Force-‐mating
pairs of one male and one
female were constructed in 24-‐well
micro plates, and the number of
offspring and the fertilization
success of mating pairs
-
20
were recorded. Observations were
performed on a daily basis and
also endpoints such as time to
mating and time between clutches
of viable offspring were recorded.
Mortality was pooled from the
life stages NI to CI, from CI
to A and for parent animals,
as well as over all
life-‐stages. The same test method,
based on the “Harpacticoid Copepod
Development and Reproduction Test for
Amphiascus tenuiremis” (OECD, 2013)
was also used to obtain the
data modeled in papers II and
III. Here, this test is termed
LTRE. In paper IV, a
low-‐density LTRE was started by
rearing N. spinipes in groups
of 6 on 24-‐well micro plates.
Population density was 32
mm2/animal and given as per area
since N. spinipes are
bottom-‐dwellers. Development and
survival from NI to CI was
closely monitored, and when
animals reached the first
copepodite stage, they were isolated
on 96-‐well micro plates, and
development and survival was observed
until animals reached the adult
stage. Reproduction was studied
separately: Newborn nauplii were reared
in groups of 24 on
6-‐well micro plates (population density
was 40 mm2/animal) until ovigerous
females (females with egg-‐sacks)
were discovered. Ovigerous females
were then isolated on 24-‐well
micro plates and the time to
first reproduction, number of
offspring, fertilization success and
survival was recorded. Here, this
test is termed separated LTRE.
4.3.2 Time-‐series experiments Time-‐series
data is obtained from
experiments on whole populations that
are followed (preferably) through
several generations (Moe, 2008).
Time-‐series experiments are commonly
performed at high population-‐density
(Sibly, 1999). In paper IV,
populations were started with 3
individuals from each developmental
stage (nauplii, copepodites, males,
females), 12 individuals in total.
Populations were kept in 20
ml glass vials with an
initial population density of 42
mm2/animal. As the number of
individuals in each experimental unit
was growing, population density
increased, and the mean population
density over the test-‐period for
all replicates was 9.3 mm2/animal.
The number of animals in each
life stage was recorded once a
week for seven weeks. Here,
this test is termed population
test.
4.3.3 Population models Four types
of population models, ranging from
simple to complex, were used
in this thesis. All models used
assumed exponential growth, i.e.
included no density dependence, or
limitation of resources such as
food. Unstructured model From
the population test (paper IV),
population size was sampled over
time and deterministic population
growth rate was calculated by
applying an equation for exponential
growth (1),
! ! = ! 0 !!" (1) where
r = population growth rate, N
= population size, t =
time. The natural logarithm of the
population size was plotted against
time. The slope of the
regression was r, and λ was
calculated using: r = lnλ.
Endpoints obtained from the equation
only included population growth rate.
-
21
Biologically structured models
Equations In paper IV, life-‐history
rates of survival, development and
reproductive output were used in
an equation for calculating
deterministic relative finite rate of
increase (population growth rate)
(2),
! = 1/(!!!!)!/!! (2) where
λ = relative population growth
rate, lt = survival probability
from NI to A (mean for
each population), bt = reproductive
output, which is the product of
sr (sex ratio, 50 %), fs
(fertilization success, mean for
each population) and n (mean
number of nauplii over two
clutches, per female), t = time
to first reproduction in days
(mean for each population). In
the traditional Euler Lotka model
(3),
1 = Σ!!!!!!! (3)
lx is the probability of
surviving to age x, mx is
the age-‐specific fecundity, and
x is the time between reproductive
events. Population growth rate was
termed “relative” in equation 2
since x was exchanged for time
to first reproduction t. Using
the Euler-‐Lotka equation routinely
in ERA for a sexually
reproducing species such as N.
spinipes would be very expensive
and time-‐consuming (Breitholtz et al.,
2003). Endpoints obtained from the
equation only included relative
population growth rate.
MMs Stage-‐based MMs (Lefkovitch
MM) were used for the
copepods in papers I and III.
The MMs include life stage
transitions (the proportions of
animals at the start of the
test that survive to and
reach, each development stage) and
fecundity. Stochastic matrixes were
generated from the distributions
defined by the test data.
Population-‐level endpoints were calculated
using matrix algebra by multiplying
a vector with the MM (4),
!!!!!!!!! !
=
!!! ! ! !!!" !!! ! !! !!" !!! !! ! !!!! !
×
!!!!!!!!! !!!
(4)
matrix
vector where N
represents the number of
individuals in a certain stage
class and P the life stage
transition rate. Indices represent
the different stage classes: n,
nauplius; c, copepodite; f, female;
fo, ovigerous female, F represents
the number of offspring per
female. Population growth rate was
obtained as λ. The MMs were
matriarchal, meaning that once
sexually mature, males were excluded
from the calculations. Time in
the MM was treated as discrete
time-‐steps, which were not
equidistant, meaning that different
time-‐steps correspond to different
lengths of time. Endpoints from the
MMs included population growth rate
and population dynamics.
-
22
IBMs IBMs for the copepods
(Figure 4) were developed and
used in papers II and III,
and were parameterized using
control data for the copepods.
Population-‐level endpoints were obtained
using stochastic simulation techniques
of individual-‐level effects. The
IBMs included 7 input variables;
stage specific mortality, development
time to reach the copepodite
and adult stage, sex ratios,
interclutch time (time between
consecutive clutches), latency (time
from mating to first clutch,
minus interclutch time), clutch
size and fertilization success. The
IBM models included both sexes
during the simulations (assuming a
1:1 sex ratio), and the
instantaneous rate of increase
(population growth rate) was obtained
as r. Endpoints from the
IBMs included population growth rate
and population dynamics.
Figure 4: Conceptual diagram of
the harpacticoid copepod life cycle
implemented in the individual based
models. Rectangles indicate processes
on the individual level and
queries are expressed in rhombs,
whereby “Dev?” indicates if
development is finished. y =
yes, n = no.
4.3.4 Measure of adverse effects
Individual-‐level effects At the
individual level, effect values
were given as the Lowest
Observed Effect Concentration (LOEC)
in paper I. LOEC is the
lowest concentration tested that
is statistically different to the
control. The No Observed Effect
Concentration (NOEC) is the highest
concentration tested that is not
statistically different from the
control, i.e. the next lower
concentration after the LOEC. In
papers III and IV, the
effect values were given as
the Effect Concentration at 10 %
(EC10). EC10 values are estimated
by fitting a curve to the
test data points over the
tested concentration interval. The value
at which there is a 10 %
effect compared to the control
treatment is termed EC10. NOEC
and EC10 are considered equivalent
to each other (European Commission,
2003). In ERA, NOECs or
EC10 values are considered the
PNEC value, which is first
combined with
Die? Nauplii
Development
Copepodite
Adult Mate Initiate brood
Offspring Development
Die?
Dev?
Die?
Development Dev?
Dev?
Die
Die
Die
y
n n
y
y
n
n
y
y n
n
y
-
23
uncertainty factors, before the PEC
for the environment is divided
by the PNEC to obtain a
risk-‐quotient. Statistical power
(or sensitivity) is a concept
that describes how likely it
is that a type II error
occurs. A type II error is
to obtain a false negative
response, meaning that the
statistical test says there is
no effect, when there actually
is an effect. High sensitivity
makes it less likely to make
a type II error. The use
of NOECs promotes the use
of many replicates in order to
obtain high sensitivity in
hypothesis testing (Landis and Chapman,
2011). For curve fitting/regression
analysis, the use of many test
concentrations allows for better
curve fitting and therefore more
reliable estimates of EC10
(Landis and Chapman, 2011). The
number of replicates and
test-‐concentrations are however often
limited due to logistic-‐ or
economic reasons. Population-‐level
effects A commonly used measure
of adverse effects at the
population level is the concentration
at which population density is
stable, i.e. when λ = 1
and r = 0. Declining
populations are defined by λ
values < 1 and r values
< 0 (e.g. Sibly, 1999), and
this approach was used in
papers I, III and IV. Other
methods for assessing population-‐level
effects are by calculating NOECs
(e.g. Lin et al., 2005) or
EC10 (e.g. Beaudouin and Péry,
2013) from population-‐level endpoints.
In papers III and IV, EC10
were calculated from λ. In
paper I, the 95 % confidence
limits for population growth rate
were used in a way that
non-‐overlapping confidence limits between
control and treatment were
interpreted as a statistically
significant effect (Environment Canada,
2005). To determine the type of
interaction between density and
toxicant from the low-‐and
high-‐density population tests in
paper IV, linear regressions were
used.
5. Results and Discussion This
thesis:
• Developed and used four different
kinds of population models for
harpacticoid copepods, ranging from
simple equations, through MMs and
complex IBMs.
• Compared toxic effects at the
individual-‐ and population level.
• Developed a new test-‐design where
animals were grouped, to increase
the number of
replicate animals without a higher
workload, and to increase
fertilization success and statistical
power of reproductive endpoints.
• Contrasted population models of
different complexity for how they
translate individual-‐level effects to
the population level.
• Compared toxic effects at different
population densities.
5.1 Model development Four different
types of population models were
developed and used in this
thesis. MMs were applied to N.
spinipes (paper I) and A.
tenuiremis (paper III). IBMs were
developed for both species of
harpacticoid copepods, in paper II
for N. spinipes and in paper
III for A. tenuiremis.
Unstructured and biologically structured
demographic equations were applied to
N. spinipes in paper IV. The
MM in paper I was used
to project long-‐term effects of
sewage treatment technologies aimed
at removing pharmaceuticals. The
IBM in paper II was used
to study endpoint sensitivity and
test-‐design of a draft OECD
guideline for harpacticoid copepods.
The MM and the IBM in
paper III were used to project
individual-‐level effects of lindane
to the
-
24
population level. In paper IV,
the unstructured equation was
used to calculate population growth
rate from population sizes of
populations exposed to lindane
over time. Finally, the biologically
structured equation in paper IV
was used to calculate population
growth rate from life-‐history rates
measured in a LTRE (separated
LTRE). The A. tenuiremis MM
was tested by plotting the
projected abundance of females in
the different test concentrations
after four time steps, against
the measured abundance of females
in the experimental data (paper
III, Figure 1, appendix A). The
projected abundances correlate well
with the measured abundances. The
IBMs were tested against the
data used to parameterize them
(N. spinipes: paper II, Figure
3; A. tenuiremis: paper III,
Figure 2, appendix A). The
model structures were concluded
appropriate to simulate the
experiments and the models were
well implemented.
5.2 Contrasting individual-‐ and
population-‐level effects ERA of
today is commonly based on
individual-‐level endpoints. The most
sensitive individual-‐level endpoints
are likely to be equally or
more sensitive to stressors than
effects on the population level
(Forbes and Calow, 2002). Analyzing
effects by integrating key
life-‐history rates in population
models is however a more robust
approach for assessing ecological
risk of stressors (Forbes and
Calow, 2002). In this thesis,
comparisons of individual-‐ and
population-‐level endpoints were therefore
compared. As an example,
development time, which was the
most sensitive individual-‐level endpoint,
was significantly affected already at
3 % conventionally treated effluent
(paper I). At the population
level, however, population growth
rate indicated a significant
population decline (λ < 1)
only at 75 % effluent (Table
1). The EC10 values of the
most sensitive individual-‐level endpoints
were in the same range (paper
III) or more sensitive (paper I
and IV) than the
population-‐level endpoints (Table 1).
The results from these studies
were therefore in agreement with
the view of Forbes and Calow
(2002), and the most sensitive
individual-‐level endpoint would hence
in most cases be protective of
population-‐level effects. Table 1:
Examples of effect concentrations at
the individual-‐ and population
level.
Test substance Individual-‐level
endpoint
Population-‐level endpoint
paper I N. spinipes
Conventionally treated effluent (%)
Development time (NI-‐A) effect
at 3 %
λ effect at 75 %
paper III A. tenuiremis
Lindane (μgL-‐1)
Brood size EC10 of 2.8
EC10 5.6 λ, MM 2.8 r, IBM
paper IV N. spinipes
Lindane (μgL-‐1)
Brood size EC10 of 2.6
EC10 94.7 λ, LTRE 13.7 λ,
POP
NI = naupliar stage I, A =
adult stage, LTRE =separated LTRE,
POP= population test The
importance of population-‐level data is
however not that it should
be more sensitive than
individual-‐level data. Instead,
population-‐level data can be used
to reduce uncertainty in
extrapolation of (standard) test
results to ecologically relevant
effects (Forbes et al., 2011).
For example, in paper IV,
effect on brood size was 46.9
% in the highest lindane
concentration in the separated LTRE
(low population-‐density test) (Table
1, paper IV), whereas the
effect on λ for the same
lindane concentration was only 6.2 %
(Figure 2, paper IV). This
indicates that effects on brood
size were much stronger than
effects on population growth rate,
which is the more ecologically
relevant endpoint. Effects on λ
in the high-‐density population
test were however
-
25
larger (47.5 %, Figure 2, paper
IV), indicating that population
density influenced the effects of
lindane. In paper I,
combining individual-‐level effects with
population-‐level effects resulted in
different conclusions than conclusions
that would be reached using
either of the measures of
effect on its own. In this
case, juvenile development and
survival allowed for a closer
monitoring of the molting process.
Novel treatment technologies were
evaluated, and the ecotoxicity
tests were used to observe effects
at the individual level, as a
way of discriminating between the
animals exposed to different
treatments. The population modeling was
useful for studying potential
long-‐term effects from the effluents
at the population level. Life-‐cycle
test or LTREs are the basis
for many models used in
effect modeling (Caswell, 2001), where
individual-‐level effects are analyzed
in order to parameterize the
models. Meli et al. (2014)
conclude that “two pairs of
eyes are better than one”, and
by that they mean using both
simple and complex population models
to assess toxicant-‐ induced
effects. This is also true for
combining individual-‐ and
population-‐level effects to assess the
risk of a toxicant. Effects on
the individual level that are
not translated to population-‐level
effects are still important for
understanding the mechanisms involved,
to design testing procedures and
build alternative models.
5.3 Statistical power and replicates
for reproductive endpoints The IBM
for N. spinipes was in paper
II used as a virtual
laboratory, where experiments were carried
out to evaluate endpoint
sensitivity and to optimize test
design in the draft guideline
“Harpacticoid Copepod Development and
Reproduction Test for Amphiascus
tenuiremis“ (OECD, 2013). The
guideline test-‐design was used in
paper I (experiment, using 72
replicates) and in paper III
(data collection). The test-‐design
in the draft guideline is
work-‐intensive, which limits the
number of replicates (or test
concentrations) possible to include. As
an example, paper I included
10 different treatments in 72
replicates, which required two
person’s attention every day for
46 days, aided by a third
person when sex-‐determinations and
counting of offspring was performed.
At least five test
concentrations and 60-‐120 replicates
are suggested in the draft
guideline (OECD, 2013). The
impact of the number of
replicates on the statistical power
of different endpoints in the
guideline was investigated (paper
II). As an example, using only
25 instead of 72 replicates
resulted in no reliable
detection of adverse effects.
Increasing the number of replicates
from 72 to 144 did
surprisingly not make it easier
to detect effects on developmental
endpoints. To statistically detect
effects on reproductive endpoints
when using 72 replicates, the
effect had to be a minimum
of 40-‐50 %, whereas developmental
effects were detected at 20
% effect. Increasing the number
of virtual replicates to 144
only increased sensitivity of
reproductive effects by 10 %,
meaning that effects on reproduction
could be detected at 30-‐40
% effect. Developmental endpoints
therefore had higher statistical
power compared to reproductive endpoints
in the guideline test design.
Also the inspection interval of
the draft guideline, which is
daily observations (OECD, 2013), was
investigated in paper II. The
results from the virtual experiments
concluded that it is possible
to reduce inspection to every 3
days without losing statistical
power. Using IBMs to evaluate
endpoint sensitivity and to optimize
test design for guidelines under
development could greatly speed up
the process and be of good
cost-‐benefit. For instance, the
number of replicates and the
inspection interval required to obtain
reliable data can be
investigated before validation of the
test method is initiated.
The results from paper II
resulted in the development of a
new test-‐design with a revised
inspection regime in the next
study (paper IV). There were
two main differences in the
test-‐design between the guideline
(used in the experiments in paper
I -‐ 72 replicates, and for
data
-
26
collection in paper III – 60
replicates) and paper IV: The
first difference was that the
number of animals used for
reproductive endpoints was doubled from
72 to 144, and that
animals were grouped (24 animals
were grouped in each of 6
replicates). The second difference was
that males and females could
mate freely since they were
grouped, instead of using the
force-‐mating pairs suggested in
the guideline. The aim of the
study was to increase
statistical power for reproductive
endpoints by increasing the number
of replicates, without increasing
workload, as compared to the
draft guideline. Sensitivity of brood
size was in paper IV
increased by 10 % by the use
of 144 replicates, as predicted
in paper II (Table 2).
The number of replicates for
reproductive endpoints in paper
IV was also increased due to
higher fertilization success, compared
to fertilization success for N.
spinipes in paper I (Table
2). Another study, which allowed
for free mating of 10-‐15
animals, yielded fertilization success
of 70-‐99 % in three separate
controls (Breitholtz and Bengtsson,
2001). It seems that
fertilization success varies substantially
for N. spinipes, and that
force-‐mating males and females may
result in lower fertilization
success than when they are
allowed to mate freely. A.
tenuiremis do not seem to be
affected in the same way, but
are more of “love the one
your with” kind of animals
(Table 2). In treatments where
endpoints are affected also by
a toxicant, low fertilization success
can further reduce the number
of replicates for reproductive
endpoints substantially. In paper I
(Table 5, paper I) the
proportion of the force-‐mating pairs
producing two viable clutches in
effluent C2 (75 %) was only
0.10. In reality, this means
that there were only two
replicates for reproductive endpoints in
this treatment. High statistical
power of reproductive endpoints
is important in traditional ERA
so that false negatives, or type
II errors, are avoided. Higher
numbers of replicates reduces
uncertainty in measurements, also when
individual-‐level effects are
extrapolated to the population level.
Uncertainty of population model
output was mentioned as one
important problem relating to the
use of population modeling for
ERA (Hunka et al., 2013).
Table 2: Fertilization success
of controls and % effect on
brood size statistically detected.
Statistically detected effect, brood
size Fertilization success paper
II N. spinipes
40 % (virtual experiment)
paper I N. spinipes
63 and 54 %; force-‐mating
pairs
paper III A. tenuiremis
90 %; force-‐mating pairs
paper IV N. spinipes
30 %
paper IV N. spinipes
96%; free mating
5.4 Contrasting simple and complex
modeling approaches Contrasting population
models of differing complexities may
aid risk assessors in choosing
what population model to use
(Meli et al., 2014). In paper
III, an IBM and a MM
were contrasted for their ability
to translate individual-‐level effects
to the population level. The MM
was very simple and included
only life stage transitions and
brood size as input variables,
whereas the IBM used 7
(including time-‐dependent) parameters
(Table 1 paper III). The number
of parameters needed to run the
model is lower in the MM
compared to the IBM, but the
experimental work needed to derive
these values was similar. For
A. tenuiremis exposed to lindane,
IBM-‐derived population growth rate
showed stronger effects compared to
the MM (Figure 5). Individual-‐level
effects in this data set
included time-‐dependent effects, such
as shifts in development time
(Figure 2, paper III). These effects
were not translated to the
population level response to the
same degree
-
27
in the MM output, which,
therefore, showed lower population-‐level
effects compared to the IBM,
especially at the highest lindane
concentration. Development time was
strongly affected at the individual
level for A. tenuiremis exposed
to lindane (Figure 2, paper
III), and since the MM did
not account for delay in
development, this was the probable
reason behind the differences in
effects at the population level.
Figure 5: Population
growth rates relative to control
(mean values) from a MM and
an IBM for A. tenuiremis
exposed to lindane. Error bars
represent 95 % confidence intervals.
Other studies have
compared simple and more complex
population models. Topping et al.
(2005), used population growth
rate to contrast a life-‐history
model (MM) and an individual
based landscape model for Skylark
populations exposed to pesticide.
They found that the two
models gave largely the same
results. Meli et al. (2014),
on the other hand, found
that a MM was less sensitive
compared to an IBM for
detecting different spatial patterns
of exposure of F. candida to
copper sulfate. The conclusion from
paper III was that the IBM
should be used for analyzing
datasets where time-‐dependent effects
are included. The simpler MM is
in its current form sufficient
for analyzing datasets including
effects on mortality and/or
reproduction. Effects of toxicants
measured at the individual level
can with these models be
projected to the population level,
and provide information of
population-‐level consequences, which are
important in ERA.
5.5 Contrasting effects in low-‐
and high-‐density populations Due to
density dependence, high-‐density
populations may respond differently to
stressors compared to low-‐density
populations.