Content uploaded by Christos C Ioannou
Author content
All content in this area was uploaded by Christos C Ioannou on Dec 25, 2019
Content may be subject to copyright.
Turbidity increases risk perception but constrains collective behaviour
during foraging by fish shoals
Alice C. Chamberlain, Christos C. Ioannou
*
School of Biological Sciences, University of Bristol, Bristol, U.K.
article info
Article history:
Received 2 March 2019
Initial acceptance 18 April 2019
Final acceptance 17 July 2019
MS. number: 19-00154
Keywords:
antipredator behaviour
collective behaviour
foraging
group decision making
refuge
stickleback
Turbidity reduces the distance that animals can detect food, predators and conspecifics. How turbidity
affects decision making in social contexts has rarely been investigated; moreover, it is unknown whether
decreased shoaling in turbid water is due to visual constraints (a mechanistic explanation) or a reduced
perception of predation risk (an adaptive explanation). Using a V-shaped decision-making arena, we
investigated the effect of turbidity on foraging in groups of three-spined sticklebacks, Gasterosteus
aculeatus. In turbid conditions, fish took longer to leave a refuge and locate the food in one of the arms
and consumed less food once it was found. This increase in risk-averse behaviour was further supported
by improved accuracy over repeated trials and a speedeaccuracy trade-off only being observed in turbid
conditions. Despite evidence of a higher perception of risk in turbid water, the first fish to choose an arm
of the maze was more likely to be alone in turbid water; thus, this individual lost the antipredator and
decision-making benefits of collective behaviour. This suggests that turbidity acts mechanistically as a
visual constraint, shifting decisions away from being made collectively to being made by individuals
separated from the group, which could have potential impacts for wild prey populations.
©2019 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.
Turbidity is caused by suspended particles in water that atten-
uate and increase the scattering of light (Utne &Utne-Palm, 2002).
The subsequent decrease in visibility reduces the availability and
reliability of visual cues for aquatic animals, limiting the private and
social information available for decision making. In addition to
some habitats being naturally highly turbid, anthropogenic activity
is driving an increase in turbidity via eutrophication and sedi-
mentation from intensified agriculture, deforestation and urbani-
zation (Davies-Colley &Smith, 2001). Accordingly, there is much
research interest in quantifying and predicting behavioural re-
sponses to increased turbidity as behaviour can mediate the impact
of turbidity on wild animal populations and their distributions
(Abrahams &Kattenfeld, 1997; Tuomainen &Candolin, 2011; Utne
&Utne-Palm, 2002).
Previous research exploring the effect of turbidity on behaviour
has focused on predatoreprey interactions, including foraging
behaviour by predators and antipredator behaviour in prey. In
turbid water, detection of prey has been shown to be impaired
(Pekcan-Hekim &Lappalainen, 2006; Quesenberry, Allen, &Cech,
2007; Utne, 1997), which can account for decreased encounter
rates between predators and prey (Turesson &Br€
onmark, 2007)
and reduced selectivity regarding prey choice (Abrahams &
Kattenfeld, 1997; Kimbell &Morrell, 2016; Sohel, Mattila, &
Lindstr€
om, 2017). A reduced predation rate in turbid water has
been used to explain a decrease in risk-averse behaviours in prey
(Lehtiniemi, Engstr€
om-
€
Ost, &Viitasalo, 2005). Examples include
shifts in habitat choice by fathead minnows, Pimephales promelas,
where the use of more risky areas increases (Abrahams &
Kattenfeld, 1997), and in three-spined sticklebacks, Gasterosteus
aculeatus, responding to the sudden appearance of a predator, the
use of shelters decreases and fewer individuals show escape re-
sponses (Sohel &Lindstr€
om, 2015). Other studies, however, have
found an increase in antipredator behaviours, possibly due to the
reduced visibility in detecting predators increasing the perception
of risk. Three-spined sticklebacks in turbid water were shown to
increase refuge use and reduce activity (Ajemian, Sohel, &Mattila,
2015); a reduction in activity in turbid water has also been found in
guppies, Poecilia reticulata (Borner et al., 2015) and zebrafish, Danio
rerio (Suriyampola, Cac
eres, &Martins, 2018). Turbidity has also
been shown to increase the rate of freezing by guppies in response
to detecting a predator (Kimbell &Morrell, 2015) and reduce
foraging in spiny damselfish, Acanthochromis polyacanthus (Leahy,
McCormick, Mitchell, &Ferrari, 2011). These mixed results may
be explained by differences in predator hunting modes and the
*Correspondence: C. C. Ioannou, School of Biological Sciences, University of
Bristol, Life Sciences Building, 24 Tyndall Avenue, Bristol, BS8 1TQ, U.K.
E-mail address: c.c.ioannou@bristol.ac.uk (C. C. Ioannou).
Contents lists available at ScienceDirect
Animal Behaviour
journal homepage: www.elsevier.com/locate/anbehav
https://doi.org/10.1016/j.anbehav.2019.08.012
0003-3472/©2019 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.
Animal Behaviour 156 (2019) 129e138
relative visibility of predators and prey in clear and turbid water;
they all indicate, however, that turbidity affects both predation and
antipredator behaviour.
A handful of recent studies have shown shoaling tendencies in
fish are reduced in turbid water (Borner et al., 2015; Fischer &
Frommen, 2013; Kimbell &Morrell, 2015). Because shoaling is a
key antipredator behaviour in many fish species (Rieucau, Fern€
o,
Ioannou, &Handegard, 2015), this reduction in shoaling in turbid
water could represent a lower perception of risk. In contrast,
turbidity could act as a visual constraint, reducing the ability of fish
to form and maintain groups as vision is a widely used and
important sensory modality in shoaling (Ioannou, Couzin, James,
Croft, &Krause, 2011). Despite extensive work on collective de-
cisions in animal groups (Conradt &List, 2009; Ioannou, 2017b), it
is unknown how decision making in social contexts is impacted by
turbid conditions. Decisions in groups can be made collectively,
where a large proportion of the group contribute to a group deci-
sion, or be led by one or a few individuals, or groups can split with
different decisions made by different subsets, or even individuals,
from the group (Ioannou, Ramnarine, &Torney, 2017). Although in
this latter case individuals are not constrained by maintaining
group cohesion after they have separated from the group (Ioannou,
Singh, &Couzin, 2015), they lose antipredator benefits of grouping.
As with shoaling in general, groups may split during decision
making in turbid water because of a mechanistic constraint in vi-
sual information or this may be an adaptive change in behaviour if
the perception of risk is reduced.
Three-spined sticklebacks are facultatively group living and
widely distributed across a range of habitats, including in clear
and turbid water, making them an ideal system for studying the
effect of turbidity on decision making in a social context
(Quesenberry et al., 2007). In this study we used a V-shaped
decision-making arena, where the fish started in a refuge at the
base of the V and had to choose between two arms, only one of
which contained food. Shoals were tested in either clear or turbid
water, and given multiple trials per day, to test how turbidity af-
fects foraging and social interactions. In addition to exploring how
foraging decisions in a social context are affected by turbidity, we
also measured the social cohesion of the individual that first
entered an arm of the maze, that is, the deciding, or ‘initiator’,
individual (Bevan, Gosetto, Jenkins, Barnes, &Ioannou, 2018).
Whether turbidity is predicted to have a positive or negative effect
on the speed of decision making, the accuracy of choosing and
cohesion between individuals in the group will depend on
whether turbidity increased or decreased the perception of pre-
dation risk. By investigating group cohesion in parallel with
foraging, we can also infer whether changes in social cohesion are
due to altered perception of risk in turbid water (as revealed by
the response in foraging) or are more likely to be due to altered
visibility of other individuals. Repeated testing within a day
allowed us to test how the cumulative effect of habituation,
training in the presence and location of food, and satiation
(McDonald, Rands, Hill, Elder, &Ioannou, 2016) differs in clear
versus turbid environments. We predicted that due to the reduced
spatial information in turbid water, for example from landmarks
(Odling-Smee &Braithwaite, 2003), decision making would
improve more slowly in turbid water over repeated trials. We also
varied shoal size to test whether the effect of turbidity was sen-
sitive to the number of individuals in the group, which can have
multiple effects on decision making such as reducing the
perception of risk in individuals, increasing the likelihood that
bolder or more cognitively able individuals are included in the
group, and allowing for information transfer and feedback be-
tween individuals which results in swarm intelligence (Ioannou,
2017a).
METHODS
Study Subjects and Husbandry
Three-spined sticklebacks were caught from the River Cary,
Somerset, U.K. (ST 469 303) in September 2014. The collection site
is surrounded by arable land and turbidity varies frequently
throughout the year due to variation in agricultural activity and in
the weather. Fish were transported to the University of Bristol by
car where they were placed in glass stock tanks (40 x 70 cm and
34 cm deep) in a temperature-controlled room (water tempera-
ture: 15e16
C; photoperiod: 12:12 h). The tanks housed 60e90
fish and were enriched with plastic plants and dark plastic tubing
the fish could use as refuges. Fish were not sexed as the water
temperature and light cycle prevented breeding. They were fed
defrosted bloodworms and tropical flake food daily. At the time of
testing, the fish in the study had a mean standard body length of
42.3 mm (SD 3.9 mm).
Trials were carried out in January to March 2016. The day before
they were tested, up to three groups consisting of two, four or eight
individuals were formed by transferring randomly chosen in-
dividuals to three separate breeding nets (16.5 x 12.7 cm and
12.7 cm deep) within a glass stock tank, so that up to three groups
were tested each day. Fish within each group came from the same
stock tank so were familiar with one another and were held over-
night in the breeding nets and between trials on the test day that
followed. Each group was tested in only one water treatment and
individual fish were not tested in more than one group; 132 in-
dividuals were exposed to the turbid treatment (groups of two,
N¼20; groups of four, N¼9; groups of eight, N¼7) and 116 in-
dividuals to the clear treatment (groups of two, N¼20; groups of
four, N¼7; groups of eight, N¼6). At the start of testing, fish had
not been fed since the previous day to standardize hunger before
the start of trials.
Experimental Apparatus and Protocol
Experiments took place in a V-shaped maze constructed from
white Perspex (Ioannou &Dall, 2016), ca.130 cm in length (Fig. 1).
Each day, one arm of the maze (left: 130 trials; right: 133 trials) was
randomly chosen to contain a bloodworm reward inside a Petri dish
(two bloodworms per fish per trial). The arm containing blood-
worms was kept constant throughout that day of testing. Opaque
white plastic covered the front half of the petri dishes to prevent
fish seeing the reward prior to swimming into the arm of the maze.
Without this visual barrier, the food would have been visible from a
greater distance in clear versus turbid water; once past the visual
barrier, the bloodworms were visible to the fish even in the turbid
water treatment. Trials were recorded with a Panasonic SD800
camera at a resolution of 1920 1080 pixels, mounted 1 m above
the maze on a tripod.
The maze was filled with aged water to a depth of 8 cm (42 litres
in total). On the morning of turbid treatment days, 4.5 g (0.11 g/
litre) of kaolin clay (Trustleaf, Cambridge, U.K.) was added to the
maze water and mixed thoroughly, producing a measurement of
30.98 ±2.24 NTU (mean ±SD) in line with previous studies (Leahy
et al., 2011; Meager &Batty, 2007). Suspended clay particles are
frequently a large component of turbidity in natural aquatic habi-
tats (Horppila, Eloranta, Liljendahl-Nurminen, Niemist€
o, &Pekcan-
Hekim, 2009; Rhoton &Bigham, 1997), so is an ecologically relevant
method for manipulating turbidity. Clear water was measured at
0.36 ±0.21 NTU (Appendix Fig. A1).
For each trial, the fish were transferred to the refuge (an area
covered with 5 mm black plastic mesh; Fig. 1) and acclimatized for
2 min; the door was then raised remotely using fishing line,
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138130
allowing the fish to enter the maze. Trials ended when all blood-
worms had been consumed or after 20 min if the bloodworms were
not all eaten. The group was then returned to their breeding net.
After each trial, the water was thoroughly mixed in both water
treatments to maintain turbidity. Up to three groups were tested
per day in up to 10 trials per group; no further trials for a group
were conducted if no bloodworms were eaten within 20 min for
two sequential trials. This was to avoid testing groups that were
already satiated, and hence unlikely to be motivated to find food. At
the end of each day all fish were returned to stock tanks and not
reused. The water in the maze was replaced with aged water at the
end of each day and filtered overnight using an external Eheim
2213 filter pump, which was turned off during trials.
Trials were conducted over 23 days (turbid, N¼12; clear, N¼11
days). Water treatment (clear or turbid) was randomized between
days. On each day, each group received its first trial, then all groups
received their second trial, then their third, etc. The mean time
between consecutive trials starting was 24.64 min (SD 18.46 min).
Typically, three groups were tested each day, consistingof one large
group of four or eight individuals and two groups of two in-
dividuals. More trials of smaller groups were conducted as these
were expected to be more variable than larger groups (as they are
less likely to contain a representative sample of the overall
population).
Video Analysis
Data were extracted from video recordings per trial rather than
per individual fish because tracking individual identities in turbid
water was too unreliable (Kimbell &Morrell, 2015). The time taken
to leave a refuge is a commonly used method of assessing risk-
taking behaviour (Webster, Atton, Ward, &Hart, 2007) and was
measured as the latency from when the refuge door was raised
until the whole body of the first fish to leave was past the door. The
time taken to make the first decision was the time that elapsed
between the first fish leaving the refuge and the first fish to touch
19 cm
20 cm
128 cm
48 cm
13 cm
48 cm
Figure 1. The experimental V-maze used to determine the effects of turbidity on decision making in three-spined sticklebacks. The shaded box represents the refuge where the
sticklebacks were released by raising a door on a pulley at the right-hand side of the refuge. The dashed line is the point that determined when a decision had been made (the
decision boundary). The circles at the ends of both arms represent the petri dishes where food was placed prior to each trial, and the red line represents the opaque plastic used to
cover the front half of the dishes to prevent fish seeing the contents of the arm. To minimize disturbance and prevent light reflection on the water surface, a white curtain screened
the maze during trials. The camera was connected to a monitor beside the maze to allow fish behaviour to be observed remotely without causing disturbance.
0
0.2
0.4
0.6
0.8
1
13579
0
0.2
0.4
0.6
0.8
1
13579
Decision-making accuracy
Trial number
(a) (b)
Figure 2. The relationship between water treatment and trial number on decision-making accuracy. (a) Turbid and (b) clear water. Accuracy is measured as whether the arm
containing food is chosen first. The line for each treatment represents the fitted values calculated from the GLMM coefficients, while controlling for group size at its mean value. The
dotted lines around each line of best fit shows 2 predicted standard error of each line. Jitter has been added to the raw data to allow overlapping points to be visualized.
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138 131
the ‘decision boundary’of either arm with its snout (Fig. 1). This
determined the point in the trial at which a fish first decided which
arm to enter, and this fish was considered the deciding fish by being
the first to initiate a movement into an arm. Accuracy of decision
making was determined by whether this arm contained the food
reward. The time to locate food was measured as the time taken
from first crossing a decision boundary to when the first blood-
worm was consumed, and foraging success was the proportion of
bloodworms consumed within 20 min of the trial starting.
The spatial distribution of fish was measured to quantify the
social cohesion of the first fish to make a decision (Ioannou et al.,
2015). Coordinates were manually recorded from the centre of
every fish using ImageJ (Rasband, 2012) at the time of the first
decision to enter an arm in each trial. Any fish in the refuge were
given the same coordinate, fixed at the centre of the refuge area, as
their exact position was unknown. Position coordinates were used
to calculate the distance between the first fish that made a decision
and its nearest neighbour (Ioannou et al., 2015). The proportion of
individuals outside the refuge was measured from the number of
individuals inside and outside the refuge when the first decision
was made.
Statistical Analysis
Time taken response variables (latencies for the first fish to leave
the refuge, the first decision to enter an arm and the first time food
was consumed) and nearest-neighbour distance (the NND of the
first individual to choose an arm, i.e. the deciding fish) were each
analysed using a negative binomial generalized linear mixed model
(GLMM; the analyses did not meet the assumptions of normal
(Gaussian) or Poisson linear models). Decision-making accuracy,
the proportion of fish outside the refuge when the decision was
made, whether at least one other fish was out of the refuge when
the decision was made and the proportion of bloodworms
consumed were each analysed using a binomial GLMM. However,
due to convergence warnings in the models with the proportion of
1
10
100
1000
Incorrect Correct
1
10
100
1000
Incorrect Correct
Accurac
y
Time taken to make first decision (s)
(a) (b)
Figure 3. The relationship between speed and accuracy of decision making in (a) turbid and (b) clear water treatments. Speed is measured as the time taken to make the first
decision. Medians are illustrated by thick black lines, the interquartile range (IQR) is shown within the boxes and the whiskers represent data points within 1.5 x IQR. The open
circles represent data points outside of the whiskers. The y-axes show natural log transformed time taken to make the first decision for visual clarity in plotting.
10
100
1000
13579
10
100
1000
13579
10
100
1000
13579
Nearest-neighbour distance (pixels)
Trial number
(a) (b) (c)
Figure 4. The effect of group size and trial number on the nearest-neighbour distance of the fish making the first decision at the time the first decision is made in clear (dash line,
filled circles) and turbid (solid line, crosses) water treatments. (a) Two, (b) four and (c) eight fish. Lines are the predicted trends calculated from GLMM coefficients. The y-axes show
natural log transformed nearest-neighbour distance for visual clarity in plotting.
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138132
bloodworms consumed as the response variable, whether at least
50% of the bloodworms were eaten in each trial was analysed as an
alternative measure of food consumed. In 79% of trials (271 of 343),
all or none of the bloodworms were consumed, suggesting that
much of the variation in the proportion of bloodworms consumed
was captured by a simple threshold of whether 50% were
consumed. To analyse the speedeaccuracy trade-off (i.e. the rela-
tionship between the decision-making time and whether the
rewarding arm was chosen first), the time taken to make the first
decision (i.e. cross the arena) was used as the response variable and
decision-making accuracy was included as an additional explana-
tory variable in a negative binomial GLMM.
The initial model for each response variable included all two-
way interactions between the water treatment (clear or turbid),
the number of fish in the group (a continuous variable: 2, 4 or 8)
and trial number (also a continuous variable: 1st to the 10th) terms.
Using a stepwise backward elimination approach, models were
simplified to remove nonsignificant interactions (P>0.05) and the
final models always included all main effects. Group identity was
also always included as a random factor. The dispersion parameter
for each statistical model was checked to be approximatelyequal to
one using an equivalent generalized linear model without a
random term. Statistical analyses were carried out using R version
3.14.2 (The R Foundation for Statistical Computing, Vienna, Austria,
http://www.r-project.org).
Ethical note
All procedures were approved by the University of Bristol
Ethical Review Group (UIN UB/11/042 and UB/14/045). Manipu-
lating turbidity using dissolved clay powder is a common
method in laboratory experiments (Ferrari, Lysak, &Chivers,
2010), and has been used in previous behavioural studies of
fish with no observed ill effects (Vollset &Bailey, 2011). Turbidity
less than 40 NTU was used which is within ecologically realistic
limits in the habitats of sticklebacks where turbidity can change
rapidly due to increased precipitation and runoff from sur-
rounding land. A number of studies have shown the use of clay
powder to induce turbidity has no significant effect on water pH
(Johannesen, Dunn, &Morrell, 2012; Leahy et al., 2011). After use
in the laboratory, fish were rehomed in a pond unconnected to
other water bodies, in accordance with the U.K.‘s Environment
Agency regulations.
RESULTS
Latency to Leave the Refuge
The latency of the first fish to leave the refuge was significantly
longer in turbid water and increased over repeated trials (negative
binomial GLMM: turbidity:
c
21
¼4.93, P¼0.026; trial number:
c
21
¼44.62, P<0.0001; Appendix Fig. A2). As expected, the latency
to leave the refuge decreased with increasing group size
(
c
21
¼28.22, P<0.0001).
Time Taken to Make First Decision
The time taken to make the first decision, that is, the time taken
from a fish first leaving the refuge to a fish first crossing one of the
decision lines into an arm, depended upon a significant interaction
between group size and trial number, where repeated exposure
over sucessive trials had no effect on groups of two but groups of
four and eight became slower over repeated trials (negative bino-
mial GLMM: group size*trial number:
c
21
¼9.26, P¼0.002;
Appendix Fig. A3). Independent of this effect, the time taken to
make the first decision was higher in turbid water (
c
21
¼27.60,
P<0.0001; Appendix Fig. A3).
Decision-Making Accuracy
Confirming that the food was adequately concealed from the
fish when they first chose an arm, the fitted probability of choosing
the rewarding arm at the start of testing was approximately 0.5
(Fig. 2). The probability of the rewarding arm being chosen first
increased over successive trials in the turbid treatment but
remained close to chance levels (0.5) in clear water (binomial
GLMM: water treatment*trial number:
c
21
¼7.60, P¼0.006; Fig. 2).
Group size had no significant effect on decision-making accuracy
(
c
21
¼0.80, P¼0.37).
We then assessed whether the time taken to make the first
decision was related to accuracy (choosing the rewarding arm or
not), to determine whether the fish showed a speedeaccuracy
trade-off during decision making. The time taken to make the
first decision (the response variable) depended significantly upon
the interaction between water clarity and accuracy of decision
making (negative binomial GLMM:
c
21
¼4.80, P¼0.029; Fig. 3). In
0
0.2
0.4
0.6
0.8
1
13579
0
0.2
0.4
0.6
0.8
1
13579
0
0.2
0.4
0.6
0.8
1
13579
Trial number
Probability 50% bloodworms consumed
(a) (b) (c)
Figure 5. The effect of group size and trial number on whether 50% of the bloodworms were eaten in clear (dash line, filled circles) and turbid (solid line, crosses) water treatments.
(a) Two, (b) four and (c) eight fish. The lines are the predicted values calculated from GLMM coefficients. Jitter has been added to the raw data to allow overlapping points to be
visualized.
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138 13 3
turbid water, decisions took longer to make if they were accurate
(Fig. 3a) whereas in clear water, the time taken to make the first
decision was not associated with decision accuracy (Fig. 3b). In
contrast to the statistical model where accuracy was not included
as a main effect, group size and trial number had nonsignificant
effects on time taken to make the first decision (group size:
c
21
¼2.07, P¼0.15; trial number:
c
21
¼3.14, P¼0.77).
Social Cohesion of the First Fish to Choose an Arm
The NND of the first individual to make a decision regarding
which arm to choose (the deciding fish) depended significantly on
water turbidity, group size and trial number (Fig. 4), although there
were no significant interaction terms. The NNDs increased as the
trials progressed (negative binomial GLMM:
c
21
¼24.43,
P<0.0001) and, as expected, were lower in larger groups
(
c
21
¼27.23, P<0.0001). They were significantly greater in the
turbid water treatment (
c
21
¼5.06, P¼0.025), that is, the first fish
to make a decision was further from its nearest neighbour in turbid
water.
In the turbid water treatment, there was a significantly smaller
proportion of the group out of the refuge when the first decision
was made compared to the clear water treatment (binomial GLMM:
water treatment:
c
21
¼19.25, P<0.0001; Appendix Fig. A4). With
increasing group size and trials, the proportion of fish out of the
refuge also decreased (group size:
c
21
¼14.90, P<0.0005; trial
number:
c
21
¼52.20, P<0.0001). Although the proportion of fish
per group declined with group size, the probability that at least one
other fish was out of the refuge increased with group size, as ex-
pected (
c
21
¼11.98, P¼0.001). There was no effect of trial number
(
c
21
¼2.98, P¼0.084). Reflecting the trend of the proportion of the
group out of the refuge, the probability that another fish was out of
the refuge when the first decision was made was lower in turbid
water (
c
21
¼4.77, P¼0.029).
These results demonstrate the increased isolation of the fish
making a decision in turbid water, controlling for the number of fish
in each trial. To explore these trends further, we repeated the
analysis of the deciding fish's NND but only including cases when at
least one other fish was out of the refuge (285 trials remained in
this analysis, compared to 331 trials in the initial analysis of NND).
While the NND still decreased when larger groups were tested
(negative binomial GLMM:
c
21
¼14.89, P<0.0005) and increased
as trials progressed (
c
21
¼27.56, P<0.0001), the effect of water
turbidity became nonsignificant (
c
21
¼1.13 , P¼0.29). This suggests
that the effect of water turbidity on NND was mainly driven by
whether or not there were any other fish out of the refuge, and
when two or more fish were out, the deciding fish could maintain a
similar NND in both clear and turbid water.
Time to Locate Food and Proportion of Bloodworms Eaten
The time taken to locate food, that is, the latency between first
entering an arm and initiating feeding, was significantly longer in
turbid water (negative binomial GLMM: water treatment:
c
21
¼11.42, P¼0.001; Appendix Fig. A5a). This included latencies
from the incorrect arm being chosen first, although this cannot
explain the longer times taken in turbid water as choosing the
incorrect arm first was not more frequent in the turbid treatment
(Fig. 2). The time taken to locate food also increased significantly
with trial number (trial number:
c
21
¼8.35, P¼0.004; Appendix
Fig. A5b). Group size had no significant effect (group size:
c
21
¼0.026, P¼0.87).
Whether at least half of the bloodworms were eaten in each trial
was significantly affected by an interaction between group size and
trial number (binomial GLMM:
c
21
¼12.45, P<0.0005) and an
interaction between the water treatment and trial number (GLMM:
c
21
¼4.020, P¼0.045). Fewer bloodworms were eaten as trials
progressed, as expected if the fish became satiated over repeated
trials. In earlier trials per group, larger groups were more likely to
eat at least 50% of the food compared to smaller groups, although
this probability declined to close to zero for all group sizes as trials
progressed (Fig. 5). Similarly, fish tended to consume fewer
bloodworms in earlier trials in turbid versus clear water, but the
difference also declined as the trials progressed.
DISCUSSION
Our results demonstrate effects of water turbidity at multiple
stages of foraging in a social context. The latencies of the first fish to
leave the refuge, make the first decision and consume the food after
choosing an arm were all slower in turbid water, with these effects
being independent of group size and trial order (i.e. there were no
interactions with these variables and the turbidity treatment). The
higher latencies are consistent with reduced activity in turbid
water. This suggests that the perception of predation risk was
higher in turbid water in our experiment; the latency to leave a
refuge in particular is commonly used to assay perception of risk in
fish (Ajemian et al., 2015; Bevan et al., 2018). The higher perception
of risk in the turbid treatment could be a consequence of turbidity
being a more novel environment, rather than turbidity being
perceived as a higher risk habitat per se. However, we found no
evidence that the water treatment affected how the latencies
changed over repeated trials, suggesting novelty of a turbid envi-
ronment, which would decrease with more experience, was not
driving the effects on perceived risk. The fish may have reduced
their activity to acquire adequate information in the visually con-
strained turbid water, but this would not explain why fewer of the
available bloodworms were eaten in the turbid water treatment,
which also supports the interpretation that the turbid treatment
condition was perceived as riskier.
Olfactory cues from the bloodworms did not appear to influence
the first decision made; the probability of choosing the rewarding
arm would be expected to be greater than 0.5 if nonvisual cues
were being used. Evidence of improved decision making over time,
possibly due to learning, came from the fish increasing their
probability of choosing the correct (i.e. rewarding) arm as trials
progressed, but only in the turbid water treatment. Fish in the clear
water treatment tended to choose arms randomly throughout
testing. This is particularly surprising because sticklebacks are
known to use landmarks during spatial learning (Odling-Smee &
Braithwaite, 2003), and the ability to detect landmarks would be
constrained in more turbid water (Utne &Utne-Palm, 2002).
Together with the apparent increased perception of risk in the
turbid water treatment, this result suggests that improved decision
making over time occurred only in the turbid water treatment due
to the increased cost of making a wrong decision. An increased cost
could arise from moving over a greater area than is necessary,
which would increase the chance of encountering a predator the
fish are unable to detect from a distance in turbid water. This would
also explain the reduced activity in turbid water (Ajemian et al.,
2015), and could be studied further by measuring the movement
of the fish under the different conditions (Ioannou, Ruxton, &
Krause, 2008). The cost of making an inaccurate decision is
further supported by there being a speedeaccuracy trade-off only
in turbid water, where choosing the rewarding arm took longer
than choosing the incorrect arm. When the risk of making an
inaccurate decision is high, speed is typically reduced to maximize
accuracy (Chittka, Skorupski, &Raine, 2009). Thus, our results
suggest the fish made more careful (i.e. risk-averse) decisions that
showed patterns typical of other decision-making studies (e.g. a
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138134
speedeaccuracy trade-off) when perceived risk was high from
being in a turbid environment.
Despite these multiple lines of evidence that turbid water con-
ditions were perceived as higher risk, the analysis of the spatial
distribution of fish at the point when a fish first chose an arm of the
maze shows that a reduced proportion of the group was outside the
refuge in turbid water and the deciding fish was more likely to be
alone, and this drove an increased NND for the deciding fish.
Increased shoaling after exposure to predation risk has been
demonstrated previously (Hoare, Couzin, Godin, &Krause, 2004);
thus, the increased perception of risk in turbid water should result
in greater cohesion with groupmates. One possibility is that
grouping may be a less beneficial antipredator strategy for in-
dividuals in turbid environments, a key issue often ignored in
studies comparing antipredator behaviour in turbid versus clear
water conditions. Of the common antipredator grouping mecha-
nisms (Ioannou, 2017a, 2017b), the confusion effect is less likely to
be effective in turbid conditions as predators targeting one of
multiple prey from a group should be less confused when nearby
prey are less visible. However, the increased uncertainty regarding
the presence of a predator should increase reliance on group
(rather than individual) vigilance, and hence make groups safer
than solitary individuals in turbid water. The effect of turbidity is
distance dependent, having a greater effect when objects are
further away. Thus, larger groups will not be as conspicuous as they
are in clear water, and encounter rates between predator and prey
will instead be more important. As aggregation in prey reduces
encounter rates with predators (Ioannou, Bartumeus, Krause, &
Ruxton, 2011), turbid water should thus favour larger groups.
A more likely explanation for reduced shoaling is that the
reduced visibility in turbid water acted as a constraint to social
interactions. Vision is a key sensory modality used in the formation
and maintenance of shoals in many fish species (Ioannou, Couzin,
James, Croft, &Krause, 2011), and the optical effects of turbidity
mean that while individuals in close proximity are visible to one
another, visual contact is quickly lost once interindividual distances
increase. This optical property of turbidity is consistent with the
results from the NND analysis: with at least one other fish out of the
refuge, the first fish to make a decision was able to maintain visual
contact with another fish and maintain similar NNDs as in clear
water. Although sticklebacks can shift reliance from visual to ol-
factory cues (Suriyampola et al., 2018; Webster et al., 2007), our
results suggest that they are not able to fully compensate for
reduced visual cues in turbid water during shoaling. Previous
studies have shown reduced shoaling in turbid conditions (Borner
et al., 2015; Fischer &Frommen, 2013; Kimbell &Morrell, 2015);
by examining effects of turbidity on other behaviours such as refuge
use and foraging in parallel with the social cohesion of the deciding
fish, our study provides evidence that turbidity constrains shoaling
behaviour, rather than the reduced tendency to shoal being an in-
direct response to lower perceived risk in turbid environments.
We have demonstrated the impact of turbidity on foraging
behaviour via increased perception of risk which influences deci-
sion making, but that turbidity also constrains greater shoal cohe-
sion in response to increased risk. The multiple potential
mechanisms underlying our results demonstrate the complexity of
understanding behaviour in environments where sensory systems
are limited, as both proximate (the reduced ability to detect other
group members) and adaptive (the effects of reduced ability to
detect, and be detected by, predators) factors operate simulta-
neously. Improving our understanding of the social responses to
turbidity and other changes driven by anthropogenic activity, such
as ocean acidification (Duteil et al., 2016) and increasing noise
(Herbert-Read, Kremer, Bruintjes, Radford, &Ioannou, 2017; Tidau
&Briffa, 2019), is essential in implementing appropriate mitigation
strategies for natural populations, given the predicted increase in
anthropogenic disturbance in aquatic habitats and potential
widespread implication for individual fitness and population
viability (Nel et al., 2009). Our study suggests that sticklebacks can
respond plastically to increased turbidity by slowing their foraging
and making more accurate decisions, which may minimize short-
term negative impacts.
Acknowledgments
We thank Sean A. Rands, Amy S.I. Wade and two anonymous
referees for comments. This work was supported by a Natural
Environment Research Council Independent Research Fellowship
(NE/K009370/1) and responsive mode standard grant (NE/
P012639/1) awarded to C.C.I. The authors declare they have no
competing interests.
References
Abrahams, M., & Kattenfeld, M. (1997). The role of turbidity as a constraint on
predator-prey interactions in aquatic environments. Behavioral Ecology and
Sociobiology, 40(3), 169e174. https://doi.org/10.1007/s002650050330.
Ajemian, M. J., Sohel, S., & Mattila, J. (2015). Effects of turbidity and habitat
complexity on antipredator behavior of three-spined sticklebacks (Gasterosteus
aculeatus). Environmental Biology of Fishes, 98(1), 45e55. https://doi.org/10.
1007/s10641-014-0235-x.
Bevan, P. A., Gosetto, I., Jenkins, E. R., Barnes, I., & Ioannou, C. C. (2018). Regulation
between personality traits: Individual social tendencies modulate whether
boldness and leadership are correlated. Proceedings of the Royal Society B: Bio-
logical Sciences, 285(1880). https://doi.org/10.1098/rspb.2018.0829.
Borner, K. K., Krause, S., Mehner, T., Uusi-Heikkil€
a, S., Ramnarine, I. W., & Krause, J.
(2015). Turbidity affects social dynamics in Trinidadian guppies. Behavioral
Ecology and Sociobiology, 69(4), 645e651. https://doi.org/10.1007/s00265-015-
1875-3.
Chittka, L., Skorupski, P., & Raine, N. E. E. (2009). Speed-accuracy tradeoffs in animal
decision making. Trends in Ecology &Evolution, 24(7), 400e407. https://doi.org/
10.1016/j.tree.2009.02.010.
Conradt, L., & List, C. (2009). Group decisions in humans and animals: A survey.
Philosophical Transactions of the Royal Society B: Biological Sciences, 364(1518),
719e742. https://doi.org/10.1098/rstb.2008.0276.
Davies-Colley, R. J., & Smith, D. G. (2001). Turbidity, suspended sediment, and water
clarity: A review. Journal of the American Water Resources Association, 37(5),
1085e110 1. https://doi.org/10.1111/j.1752-1688.2001.tb03624.x.
Duteil, M., Pope, E. C., P
erez-Escudero, A., de Polavieja, G. G., Fürtbauer, I.,
Brown, M. R., et al. (2016). European sea bass show behavioural resilience to
near-future ocean acidification. Royal Society Open Science, 3(11), 160656.
https://doi.org/10.1098/rsos.160656.
Ferrari, M. C. O., Lysak, K. R., & Chivers, D. P. (2010). Turbidity as an ecological
constraint on learned predator recognition and generalization in a prey fish.
Animal Behaviour, 79(2), 515e519. https://doi.org/10.1016/j.anbehav.2009.12.
006.
Fischer, S., & Frommen, J. G. (2013). Eutrophication alters social preferences in
three-spined sticklebacks (Gasterosteus aculeatus). Behavioral Ecology and So-
ciobiology, 67(2), 293e299. https://doi.org/10.1007/s00265-012-1449-6.
Herbert-Read, J. E., Kremer, L., Bruintjes, R., Radford, A. N., & Ioannou, C. C. (2017).
Anthropogenic noise pollution from pile-driving disrupts the structure and
dynamics of fish shoals. Proceedings of the Royal Society B: Biological Sciences,
284(1863), 20171627. https://doi.org/10.1098/rspb.2017.1627.
Hoare, D. J., Couzin, I. D., Godin, J. G. J., & Krause, J. (2004). Context-dependent group
size choice in fish. Animal Behaviour, 67(1), 155e164. https://doi.org/10.1016/j.
anbehav.2003.04.004.
Horppila, J., Eloranta, P., Liljendahl-Nurminen, A., Niemist€
o, J., & Pekcan-Hekim, Z.
(2009). Refuge availability and sequence of predators determine the seasonal
succession of crustacean zooplankton in a clay-turbid lake. Aquatic Ecology,
43(1), 91e103. https://doi.org/10.1007/s10452-007-9158-3.
Ioannou, C. C. (2017a). Grouping and predation. InT. K. Shackelford, & V. A. Weekes-
Shackelford (Eds.), Encyclopedia of Evolutionary Psychological Science (pp. 1e6).
Cham, Switzerland: Springer. https://doi.org/10.1007/978-3-319-16999-6_
2699-1.
Ioannou, C. C. (2017b). Swarm intelligence in fish? The difficulty in demonstrating
distributed and self-organised collective intelligence in (some) animal groups.
Behavioural Processes, 141, 141e151. https://doi.org/10.1016/j.beproc.2016.10.
005.
Ioannou, C. C., Bartumeus, F., Krause, J., & Ruxton, G. D. (2011). Unified effects of
aggregation reveal larger prey groups take longer to find. Proceedings of the
Royal Society B: Biological Sciences, 278, 2985e2990. https://doi.org/10.1098/
rspb.2011.0003.
Ioannou, C. C., Couzin, I. D., James, R., Croft, D. P., & Krause, J. (2011). Social orga-
nisation and information transfer in schooling fish. In C. Brown, K. Laland, &
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138 13 5
J. Krause (Eds.), Fish Cognition and Behavior (2nd ed., pp. 217e239). New York,
NY: Wiley https://doi.org/10.1002/9781444342536.ch10.
Ioannou, C. C., & Dall, S. R. X. (2016). Individuals that are consistent in risk-taking
benefit during collective foraging. Scientific Reports, 6, 33991. https://doi.org/
10.1038/srep33991.
Ioannou, C. C., Ramnarine, I. W., & Torney, C. J. (2017). High-predation habitats affect
the social dynamics of collective exploration in a shoaling fish. Science Advances,
3(5). e1602682 https://doi.org/10.1126/sciadv.1602682.
Ioannou, C. C., Ruxton, G. D., & Krause, J. (2008). Search rate, attack probability, and
the relationship between prey density and prey encounter rate. Behavioral
Ecology, 19(4), 842e846. https://doi.org/10.1093/beheco/arn038.
Ioannou, C. C., Singh, M., & Couzin, I. D. (2015). Potential leaders trade off goal-
oriented and socially oriented behavior in mobile animal groups. American
Naturalist, 186(2), 284e293. https://doi.org/10.1086/681988.
Johannesen, A., Dunn, A. M., & Morrell, L. J. (2012). Olfactory cue use by three-
spined sticklebacks foraging in turbid water: Prey detection or prey location?
Animal Behaviour, 84(1), 151e158. https://doi.org/10.1016/j.anbehav.2012.04.
024.
Kimbell, H. S., & Morrell, L. J. (2015). Turbidity influences individual and group level
responses to predation in guppies, Poecilia reticulata.Animal Behaviour, 103(0),
179 e185. https://doi.org/10.1016/j.anbehav.2015.02.027.
Kimbell, H. S., & Morrell, L. J. (2016). Turbidity weakens selection for assortment in
body size in groups. Behavioral Ecology, 27(2), 545e552. https://doi.org/10.1093/
beheco/arv183.
Leahy, S. M., McCormick, M. I., Mitchell, M. D., & Ferrari, M. C. O. (2011). To fear or to
feed: The effects of turbidity on perception of risk by a marine fish. Biology
Letters, 7(6), 811e813. https://doi.org/10.1098/rsbl.2011.0645.
Lehtiniemi, M., Engstr€
om-
€
Ost, J., & Viitasalo, M. (2005). Turbidity decreases anti-
predator behaviour in pike larvae, Esox lucius.Environmental Biology of Fishes,
73(1), 1e8. https://doi.org/10.1007/s10641-004-5568-4.
McDonald, N. D., Rands, S. A., Hill, F., Elder, C., & Ioannou, C. C. (2016). Consensus
and experience trump leadership, suppressing individual personality during
social foraging. Science Advances, 2(9), e1600892. https://doi.org/10.1126/sciadv.
1600892.
Meager, J. J., & Batty, R. S. (2007). Effects of turbidity on the spontaneous and prey-
searching activity of juvenile Atlantic cod (Gadus morhua). Philosophical
Transactions of the Royal Society B: Biological Sciences, 362(1487), 2123e2130.
https://doi.org/10.1098/rstb.2007.2104.
Nel, J. L., Roux, D. J., Abell, R., Ashton, P. J., Cowling, R. M., Higgins, J. V., et al. (2009).
Progress and challenges in freshwater conservation planning. Aquatic Conser-
vation: Marine and Freshwater Ecosystems, 19(4), 474e485. https://doi.org/10.
1002/aqc.1010.
Odling-Smee, L., & Braithwaite, V. (2003). The influence of habitat stability on
landmark use during spatial learning in the three-spined stickleback. Animal
Behaviour, 65(4), 701e707. https://doi.org/10.1006/anbe.2003.2082.
Pekcan-Hekim, Z., & Lappalainen, J. (2006). Effects of clay turbidity and density of
pikeperch (Sander lucioperca) larvae on predation by perch (Perca fluviatilis).
Naturwissenschaften, 93(7), 356e359. https://doi.org/10.1007/s00114-006-0114-
1.
Quesenberry, N. J., Allen, P. J., & Cech, J. J. (2007). The influence of turbidity on three-
spined stickleback foraging. Journal of Fish Biology, 70(3), 965e972. https://doi.
org/10.1111/j.1095-8649.2007.01350.x.
Rasband, W. (2012). ImageJ: Image processing and analysis in java. Astrophysics
Source Code Library, 1, 06013.
Rhoton, F. E., & Bigham, J. M. (1997). Natural ferrihydrite as an agent for reducing
turbidity caused by suspended clays. Journal of Environmental Quality, 38,
1887e1891. https://doi.org/10.2134/jeq2008.0454.
Rieucau, G., Fern€
o, A., Ioannou, C. C., & Handegard, N. O. (2015). Towards of a firmer
explanation of large shoal formation, maintenance and collective reactions in
marine fish. Reviews in Fish Biology and Fisheries, 25(1), 21e37. https://doi.org/
10.1007/s11160-014-9367-5.
Sohel, S., & Lindstr€
om, K. (2015). Algal turbidity reduces risk assessment ability of
the three-spined stickleback. Ethology, 121(6), 548e555. https://doi.org/10.1111/
eth.12370.
Sohel, S., Mattila, J., & Lindstr€
om, K. (2017). Effects of turbidity on prey choice of
three-spined stickleback Gasterosteus aculeatus.Marine Ecology Progress Series,
566,159e16 7. https://doi.org/10.3354/meps12014.
Suriyampola, P. S., Cac
eres, J., & Martins, E. P. (2018). Effects of short-term turbidity
on sensory preference and behaviour of adult fish. Animal Behaviour, 146,
105e111. https://doi.org/10.1016/j.anbehav.2018.10.014.
Tidau, S., & Briffa, M. (2019). Anthropogenic noise pollution reverses grouping
behaviour in hermit crabs. Animal Behaviour, 151,113e120. https://doi.org/10.
1016/j.anbehav.2019.03.010.
Tuomainen, U., & Candolin, U. (2011). Behavioural responses to human-induced
environmental change. Biological Reviews, 86(3), 640e657. https://doi.org/10.
1111/j.1469-185X.2010.00164.x.
Turesson, H., & Br€
onmark, C. (2007). Predator-prey encounter rates in freshwater
piscivores: Effects of prey density and water transparency. Oecologia, 153(2),
281e290. https://doi.org/10.1007/s00442-007-0728-9.
Utne, A. C. W. (1997). The effect of turbidity and illumination on the reaction dis-
tance and search time of the marine planktivore Gobiusculus flavescens.Journal
of Fish Biology, 50(5), 926e938. https://doi.org/10.1111/j.1095-8649.1997.
tb01619.x.
Utne, A. C. W., & Utne-Palm, A. C. (2002). Visual feeding of fish in a turbid envi-
ronment: Physical and behavioural aspects. Marine and Freshwater Behaviour
and Physiology, 35(1e2),111e128. https://doi.org/10.1080/10236240290025644.
Vollset, K. W., & Bailey, K. M. (2011). Interplay of individual interactions and
turbidity affects the functional response of three-spined sticklebacks Gaster-
osteus aculeatus.Journal of Fish Biology, 78, 1954e1964.
Webster, M. M., Atton, N., Ward, A. J. W., & Hart, P. J. B. (2007). Turbidity and
foraging rate in threespine sticklebacks: The importance of visual and chemical
prey cues. Behaviour, 144(11), 1347e1360. https://doi.org/10.1163/
156853907782418222.
Appendix
0
5
10
15
20
25
30
5101520
Mean turbidity (NTU)
Da
y
Figure A1. Turbidity levels in the experimental V-maze remained stable between the front (open circles) and back (crosses) of the arena, across each day in both the turbid (high
NTU values) and clear (low NTU values) treatments. Water samples were taken from both ends of the maze at the start and end of each day to monitor turbidity, which was
determined using a calibrated spectrophotometer. Turbidity did not vary significantly within each treatment at different locations or at the start or end of the test days (two-way
ANOVA for each turbid and clear treatments: start or end of day: turbid: F
1,45
¼1.8, P¼0.18; clear: F
1,40
¼0.25, P¼0.62; position in maze: turbid: F
1,45
¼0.001, P¼0.98; clear:
F
1,40
¼0.14, P¼0.72).
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138136
0
1
2
3
4
5
6
0
1
2
3
4
5
6
Clear Turbid
Water treatment
Latency to leave refuge (In(s))
13579
Trial number
(a) (b)
Figure A2. The latency to leave the refuge (a) in the clear and turbid water treatments and (b) across trials. Medians are illustrated by thick black lines, the interquartile range (IQR)
is shown within the boxes and the whiskers represent data points within 1.5 x IQR. The circles represent data points outside the whiskers. The y-axes show natural log transformed
latency to leave the refuge for visual clarity in plotting.
1
2
3
4
5
6
7
8
13579
1
2
3
4
5
6
7
8
13579
1
2
3
4
5
6
7
8
13579
Time taken to make first decision (In(s))
Trial number
(a) (b) (c)
Figure A3. The effect of group size and trial number on the time taken to make the first decision from the first fish leaving the refuge for both clear (dash line, filled circles) and
turbid (solid line, crosses) water. (a) Two, (b) four and (c) eight fish. The fitted lines are calculated from the coefficients of the GLMM. The y-axes show natural log transformed
latency to cross the arena for visual clarity in plotting.
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138 137
0
0.2
0.4
0.6
0.8
1
13579
0
0.2
0.4
0.6
0.8
1
13579
0
0.2
0.4
0.6
0.8
1
13579
Trial number
Proportion of fish out of refuge
(a) (b) (c)
Figure A4. The effect of group size and trial number on the proportion of fish outside the refuge at the moment th e first decision was made, across clear (dash line, filled circles) and
turbid (solid line, crosses) water treatments. (a) Two, (b) four and (c) eight fish. The fitted lines are calculated from the coefficients of the GLMM.
2
3
4
5
6
7
2
3
4
5
6
7
Clear Turbid 1357
Water treatment Trial number
Time taken to find food (In(s))
(a) (b)
Figure A5. The effect of (a) water treatment and (b) trial number on the time taken to find food after an arm of the maze was chosen for the first time. Medians are illustrated by
thick black lines, the interquartile range (IQR) is shown within the boxes and the whiskers represent data points within 1.5 x IQR. The circles represent data points outside the
whiskers. The y-axes show natural log transformed latency to leave the refuge for visual clarity in plotting.
A. C. Chamberlain, C. C. Ioannou / Animal Behaviour 156 (2019) 129e138138
