Summary

1. A predictive framework of community and ecosystem dynamics that applies across systems has remained elusive, in part because non-consumptive predator effects are often ignored. Further, it is unclear how much individual-level detail community models must

include.

2. Previous studies of short-lived species suggest that state-dependent decisions add little to our understanding of community dynamics. Body condition-dependent decisions made by long-lived herbivores under risk of predation, however, might have greater community level effects. This possibility remains largely unexplored, especially in marine environments.

3. In the relatively pristine seagrass community of Shark Bay, Australia, we found that herbivorous green sea turtles (Chelonia

mydas Linnaeus, 1758) threatened by tiger sharks (Galeocerdo

cuvier Peron and LeSueur, 1822) select microhabitats in a

condition dependent manner. Turtles in poor body condition selected profitable, high-risk microhabitats, while turtles in good body condition, which are more abundant, selected safer, less profitable microhabitats. When predation risk was low, however, turtles in good condition moved into more profitable microhabitats.

4. Condition-dependent use of space by turtles shows that tiger sharks modify the spatiotemporal pattern of turtle grazing and their impacts on ecosystem dynamics (a traitmediated indirect interaction). Therefore, state-dependent decisions by individuals can

have important implications for community dynamics in some situations.

5. Our study suggests that declines in large-bodied sharks may affect ecosystems more substantially than assumed when non-lethal effects of these top predators on mesoconsumers

are not considered explicitly.

Key-words

: marine conservation, non-lethal predator effects, predation risk, state-dependent behaviour, trait-mediated indirect effects (TMII).

Introduction

Interactions within communities and ecosystems

are complex, and ecologists have struggled to develop

frameworks for predicting their dynamics across

systems and taxa. Part of the difficulty may be due to a

lack of a functional understanding of relationships

within communities (Ovadia & Schmitz 2004), such as

non-consumptive effects of predators (Dill et al. 2003;

Schmitz et al. 2004; Schmitz 2006). These non-lethal

effects – which include prey changing their habitat use,

activity level or diet – are pervasive in natural communities.

A growing literature suggests that they may have

greater impacts on prey populations than lethal effects due to their propensity to affect many individuals within

prey populations simultaneously (Werner & Peacor 2003; Preisser et al. 2005). Furthermore, transmission

through food webs of non-lethal effects of predators

(trait-mediated indirect interactions, TMII) may be

stronger than the transmission of lethal effects (densitymediated

indirect interactions) (see Preisser et al. 2005

for a review). However, tests of these ideas in complex

natural communities are few. Therefore new data on the

importance of TMIIs in diverse communities should aid

in generating a general theory of community dynamics.

The development of a general predictive framework

for community dynamics has also been limited by not

knowing how much individual-level detail should be

included in community models. Ovadia & Schmitz (2002,

2004) addressed this question by investigating whether

knowledge of size-related differences in risk-taking by

grasshoppers foraging under the risk of spider predation

improved models of community dynamics. They found

that, although grasshoppers adjusted foraging effort

in a size-dependent manner, details on size-dependent

decisions did not enhance predictions about community

dynamics (grasshopper impacts on vegetation mediated

by risk from spiders) (Ovadia & Schmitz 2002, 2004).

However, the results of the grasshopper study may apply

primarily to short-lived or semelparous species (Ovadia

& Schmitz 2002, 2004). It is likely that long-lived, iteroparous

species would be more risk-averse than grasshoppers

(Warner 1998), and state-dependent decisions

made under the risk of predation are more likely to have

community-level effects. Also, state variables other than

body size influence foraging–predation risk trade-offs,

and variables that operate in all size and age classes of

a population, such as energetic state, may be of greater

importance in structuring communities (Luttbeg et al 2003).

Field studies have yet to address the effect of state dependent

risk-taking decisions on community dynamics

(but for theoretical treatments see Luttbeg et al. 2003;

Frid et al . 2006).

Behavioural ecologists have long been interested in

the trade-off between foraging opportunities and the

risk of predation. This interest stems in large part from

widespread recognition that the behaviour of consumers

under risk of predation can cause (and provide insights

into the potential importance of) top-down effects that

may shape spatial distributions and limit prey population

sizes. In general, forager populations are unlikely

to be influenced solely by bottom-up forces when individuals

are willing to give up energy in order to be safe

(through reduced activity levels, increased giving-up

densities, selection of energy-poor but safe habitats

for foraging, or restricting foraging locations or times:

Houston et al. 1993; Walters & Juanes 1993; Brown 1999;

Kotler et al. 2004; for a review see Brown & Kotler 2004).

Top-down effects of predators on prey populations

are enhanced when individuals in poor condition take

greater risks in order to obtain greater foraging gains

(condition-dependent risk taking: McNamara & Houston

1987, 1994; Anholt & Werner 1995; Sinclair & Arcese 1995;

Nelson et al. 2004). Therefore predators may influence

community dynamics indirectly through conditiondependent

risk-sensitive foraging by their prey, which

would modify the overall intensity and spatio-temporal

pattern of resource exploitation.

Since 1997, we have used Shark Bay, Western Australia

as a model system for understanding the importance of

non-consumptive effects of tiger sharks (Galeocerdo cuvier),

including TMII, in a relatively undisturbed seagrass

community. Here tiger sharks are predators on a variety

of large vertebrates, including herbivorous green turtles

(Chelonia mydas) and dugongs (Dugong dugon Müller

1776) (Heithaus 2005; Heithaus & Dill 2006; Wirsing

et al . 2007). Because there is marked temporal variation

in shark abundance while the abundance of many

of their prey species remains relatively constant, we are

able to use temporal variation in risk as a natural

experiment to assess the importance of predation risk

in shaping consumer behaviours (sensu Biro et al . 2005).

Given that green turtles are long-lived, suffer relatively

low mortality rates, and can modify seagrass communities

and nutrient composition as well as detrital cycles

(Bjorndal 1980, 1997; Aragones 2000; Aragones et al

2006), their condition-dependent variation in habitat

use when threatened by predators could influence

community and ecosystem dynamics.

Materials and methods

Natural History

The study was conducted in the relatively pristine

Eastern Gulf of Shark Bay, Western Australia (Fig. 1), which contains offshore subtidal seagrass beds (usually 4·0 m deep) covered primarily by the seagrass Amphibolis antarctica

Labillardière and occasionally Posidonia australis

Hooker 1858. The rest of the study area is deeper (generally 6·0–12·0 m) with little seagrass cover. There is a large population of green turtles in the study area that is not subject to substantial

anthropogenic influences. Tiger sharks are the only

predator on adult turtles in the area, but shark abundance

fluctuates seasonally (Heithaus et al. 2002, 2005; Wirsing

et al . 2006). Studies conducted concurrently with turtle

sampling revealed that tiger sharks are abundant in

summer (Septmber–May), but rare or absent during

the winter (June–August) (Wirsing et al. 2006).

Spatial variation in predation risk (the ‘landscape

of fear’: Laundré et al. 2001) is influenced by predator

encounter rates, intrinsic habitat risk (probability of

prey death in an encounter situation; Lima & Dill 1990),

and antipredator behaviour. Therefore it is best assessed

using consumer behaviour (Brown & Kotler 2004).

Risk from tiger sharks, to turtles and other species, is

higher in shallow habitats than in deep ones within the

study area (Heithaus & Dill 2006; Heithaus et al . 2006),

but risk also varies across microhabitats within shallow

habitat patches. Risk is highest in the middle of shallow

banks (interior microhabitats far from bank edges) due to a lack of escape routes, and long distances to deeper water where turtles can outmanoeuvre tiger sharks (Heithaus et al. 2002, 2006; Heithaus 2004); this pattern of risk is supported by predation-sensitive microhabitat shifts in both dugongs and bottlenose dolphins

(Tursiops truncatus) (Wirsing et al. 2007; Heithaus & Dill 2006), which use escape tactics similar to turtles (bursts of speed with very sharp turning manoeuvres). While foraging, both

these species increase their use of edges relative to interior

areas (corrected for spatial and temporal variation in food

availability) as the density of tiger sharks in the study

area increases (Heithaus & Dill 2006; Wirsing et al. 2007).

The biomass of A. antarctica, which is the primary

forage available to turtles in the study area, is similar

between edge and interior microhabitats of banks (Wirsing

et al. 2007).

Forage quality, however, varies spatially (Table 1).

Inorganic carbon appears to be avoided by green turtles,

while organic C and nitrogen are nutritionally important

(Thayer et al. 1984). In our study area, the leaves of

A. antarctica plants found along bank edges have higher

inorganic C content, but lower organic C content, than

those in the middle of banks. Thus turtles must select

whether to forage in microhabitats far from bank edges

that are high in both food quality and predation risk, or

in low-quality but low-risk microhabitats along the edges

of banks. Based on state-dependent theory, we predicted

that, during months when tiger sharks were abundant,

turtles in relatively good condition would forage along

bank edges, while turtles in relatively poor condition

would accept higher predation risk far from the edges

of banks in order to forage on higher-quality seagrass.

Furthermore, we predicted that when shark abundance

was lower, this pattern would be relaxed and turtles in

good condition would move further from bank edges

to take advantage of high-quality resources.

Field Methods and Analysis

Turtles were captured by hand from March 1999–

August 2006 during routine transect surveys of the

study area and during searches of relatively shallow

waters (Heithaus et al. 2002, 2005). Turtles were

brought alongside the capture boat and their curved

carapace length (CCL) was measured before they were

individually tagged. Turtles were assigned to one of

four condition classes based on external characteristics

(Fig. 2). Turtles were considered to be in ‘poor’ condition

when their undersides were extremely concave.

These individuals also tended to exhibit sunken eyes

and reduced body mass around the neck and base of

the foreflippers, and were easily captured due to their

slow escape responses. Turtles in ‘fair’ condition had

concave or slightly concave undersides, but were usually

active during capture, and other body parts were

not obviously in deteriorating condition. Turtles in

‘good’ condition had flat undersides; turtles in ‘very

good’ condition had convex undersides. This condition

index has been validated by comparing field measurements of body condition index (mass/CCL3; Bjorndal et al. 2000) to our condition classifications, assigned before masses were measured, for a subset of 38 green turtles (J.T., unpublished data).

During the winters of 1999–2005, tiger sharks were

caught in low numbers during the austral winter, but

none was captured in the winter of 2006 (Wirsing et al.

2006). We did not analyse the winter months of 2006

separately because of relatively low sample sizes (n<10)

for turtles in good, poor and fair condition during the

months when no sharks were present. Therefore we

defined two seasons based on monthly average shark

catch rates (‘low’, 0·03

sharks per bait h). Because we had an adequate

sample size for turtles in very good condition, which

should be most responsive to changes in the intensity

of predation risk (Clark 1994; Clark & Mangel 2000;

Luttbeg et al. 2003), we conducted a companion analysis

of their microhabitat selection with seasons broken into

high (>0·03 sharks per bait h), low (>0, <0·03 sharks per

bait h), and very low (0 sharks per bait h) shark abundance.

Analyses included only turtles that were captured

within the shark study area, excluding turtles captured

in near-shore shallows primarily covered by sand. Also, we

used only the first capture of an individual within the

restricted study area in analyses of microhabitat use, leaving

138 turtles for analysis (15 adult males and 123 females and

juvenile males, which cannot be easily distinguished in

the field). We mapped the location of each turtle into a

GIS (MapInfo, MapInfo Corp., Troy, NY) and measured

the distance between the turtle’s position and the bank

edge, defined as the point at which water depth exceeded

4·5 m. Condition was treated as an ordinal variable, and we

used ANOVA on log(x+1)-transformed distance-from-edge

data to determine the effects of condition and relative shark

density on the distance turtles were found from edges.

Results

Turtle condition was dynamic over periods of weeks to

months. Of 34 turtles recaptured, the condition of 18

did not change, nine had lost condition, and seven had

improved their condition. None of six turtles recaptured

within 14 days had changed in condition, but two

of seven (28·6%) recaptured after 15–30 days were in a

different condition category, as were seven of 10 (70%)

and seven of 11 (63·6%) turtles recaptured after 31–180

and 181–2197 days, respectively.

Turtles in better condition were found closer to bank

edges than those in poorer condition (F3,137 = 32·3, P<

0·0001), but the strength of this trend was much

stronger when tiger sharks were abundant than when

they were relatively scarce (condition×shark density,

F3,137 = 4·4, P= 0·005). Turtles in fair and poor condition

did not greatly change their relative distance from

the edges of banks as shark density increased, while

turtles in very good and good condition moved closer

to edges when sharks were more abundant (Fig. 3).

When data for good-condition turtles were partitioned

into seasons when sharks were absent, present in low

numbers, and present in high numbers, there was a

continuous pattern of increasing distance from bank edges

as shark abundance increased (F2,44 = 8·4, P= 0·0008;

Fig. 4). There was no relationship between turtle CCL

and distance at which turtles were found from banks

(F1,135 = 0·02, P= 0·89), or between CCL and turtle

condition (F1,137= 1·7, P= 0·19).

Two individuals were recaptured in the same condition,

but in different seasons of shark abundance. In both

cases, the turtles were further from bank edges when

shark abundance was low than when sharks abundance

was high. Two other individuals were recaptured during

periods of high shark abundance, but in different conditions.

The first was captured in fair condition 0·35 km

from the edge, but was in good condition and 0·2 km

from the edge 80 days later. After another 164 days, the

turtle was in poor condition and 0·58 km from the

edge. The second individual was captured twice in fair

condition 0·24 and 0·25 km from the edge. Twenty-seven

days later, it was captured in poor condition and 0·44 km from the edge.

Discussion

The use of microhabitats by green turtles in Shark Bay

was consistent with a predicted condition-dependent

response to the trade-off between foraging rewards and

predation risk, and not consistent with alternative

explanations (Appendix S1, see Supplementary material).

When tiger sharks were abundant, turtles in poor condition

were found far from bank edges, where both seagrass quality and the risk of tiger shark predation are greater, while more turtles, especially those in very good condition, were found in safer habitats with lower forage quality. Moreover, turtles in good condition moved into areas of higher food quality, far from bank edges,

when shark abundance dropped: turtles in good condition

were found more than twice as far from bank edges

during months of low shark density than they were when

shark abundance was high, and five times as far from

bank edges during months when sharks were at very low

densities. Also, all four individuals recaptured during

the study shifted their distances from the edges of banks

in accordance with predictions. Finally, changes in the

body condition of individual turtles occurred over relatively

long periods (weeks) rather than over individual

foraging bouts, suggesting that changing body condition

is due to long-term behavioural tactics rather than shortterm

ones (e.g. vigilance over the course of a foraging

bout). Recently developed GPS tags, which allow finescale

tracking of individual movements (Schofield et al. 2007), could provide even more detailed information on condition-dependent behaviour (e.g. residence times in microhabitats) and how individual turtles modify their use of microhabitats in response to changes in their body condition.

Condition-dependent predation risk-sensitive decisions

by turtles suggest that predation risk from tiger sharks

influences turtle population sizes, and thus the overall

intensity of turtle herbivory. The predation-sensitive

food hypothesis (McNamara & Houston 1987, 1994)

predicts that both food and predators will limit prey

populations because, as food becomes limiting, individuals

in poor condition will take greater risks when

foraging and suffer high predation rates (e.g. wildebeest,

Connochaetes taurinus, Sinclair & Arcese 1995; vertically

migrating copepods, Metridia pacifica , Hays et al. 2001).

This may lead to differential effects of predation within

turtle populations. For example, green turtles, especially

females, returning to foraging grounds after breeding,

are likely to be in poorer condition than individuals that remained in feeding areas (Hays et al. 2002). Thus these post-reproductive turtles may be forced to take greater risks to improve their condition, and suffer higher mortality from predators as a result. However, turtles need not incur higher predation rates when in poor condition for tiger sharks to influence their population size

through non-lethal mechanisms. If access to food limits

turtle populations, as has been assumed implicitly in

previous studies (Jackson et al. 2001), then turtles

giving up foraging opportunities or high-quality food

to be safe would lead to reduced population sizes through

non-lethal (behavioural) mechanisms (Walters & Juanes

1993; Nelson et al. 2004; Creel et al. 2007).

Green turtles are long-lived and suffer low predation

rates, while tiger sharks have diverse diets and low feeding

rates. Thus predator-inflicted mortality and possible

density-mediated indirect effects of tiger sharks are

probably modest. In contrast, condition-dependent

decisions by green turtles probably result in TMII playing

a prominent role in community and ecosystem dynamics

in Shark Bay. Turtle cropping can alter the community

composition of seagrasses, and alters ecosystem dynamics

by short-circuiting the detrital cycle and modifying

seagrass chemical composition (Bjorndal 1980; Aragones

2000; Aragones et al. 2006). The reduced organic C

content and increased inorganic content of segrasses

along the edges of banks, relative to interior areas, may

be attributable to intense herbivory by turtles in very

good and good condition (Bjorndal et al. 2000). In

contrast, seagrass communities in the middle of banks

experience relatively low grazing pressure, at least when

sharks are present. Grazing in these areas during high-risk

periods is by turtles in poor condition, which represent

only 7% of individuals in the study area (Heithaus et al.

2005). The apparent TMII between sharks and seagrasses

would be enhanced in Shark Bay by predation-sensitive

foraging by dugongs (another long-lived megaherbivore),

which shift their foraging locations from interior to

edge microhabitats as shark density increases (Wirsing

et al. 2007). TMII involving sharks could be transmitted

further through the ecosystem because changes in the

structure of seagrass communities and detrital cycles may

affect the community of resident small-bodied consumers

(Hyndes et al. 2003).

Previous studies have sought to determine the level

of individual detail necessary for understanding community

dynamics. It has become increasingly apparent

that community models can benefit greatly from the

inclusion of non-consumptive effects of predators (Bolker

et al. 2003). Indeed, the similarity in the dynamics of the

predator–herbivore systems in Shark Bay, and terrestrial

systems involving interactions between wolves and moose,

and between grasshoppers and spiders (Schmitz 2006),

is striking and suggests that simple behavioural rules

apply across systems and are important to the dynamics

of communities in general. However, the general importance

of state-dependent decisions in community and

ecosystem dynamics is still a matter of debate. Studies

of a short-lived species (grasshoppers) suggest that

state-dependent decisions made under risk of predation may

not add significantly to our understanding of communities

(Ovadia & Schmitz 2002, 2004). However, the brief

period available for grasshopper development and

reproduction may reduce the community-level effects

of their state-dependent decisions. Our results suggest

that condition-dependent behaviour by long-lived species

may be more likely to influence community dynamics,

and that the importance of understanding individual

decisions for making predictions about community

dynamics may vary with life-history traits (e.g. longevity,

parity). Consequently, further consideration is needed

of the types of state variables, life-history characteristics

and general ecological conditions that produce community

and ecosystem-level effects.

Our results have important implications for the

conservation and management of marine ecosystems.

Recent studies have revealed that marine communities

have been greatly modified by human activities in modern

and historical times (Jackson et al. 2001; Lotze et al. 2006).

One of the most conspicuous changes is the dramatic

decline, and sometimes ecological extinction, of ‘highly

interactive species’, including large herbivores and top

predators such as sharks (Jackson et al. 2001; Baum et al.

2003). Yet there are few empirical data regarding the

ecological role of large sharks (but see Heithaus 2004;

Myers et al. 2007), and few empirical studies of the possible

impacts of predators on population sizes of large

marine herbivores and their spatial patterns of grazing.

Here, taking advantage of a rare opportunity to study

interactions between relatively intact populations of

sharks and a large marine herbivore, we have demonstrated

the importance of tiger sharks in shaping turtle

behaviour and, probably, turtle populations. Given the

antipredator responses of turtles, dugongs and other

tiger shark prey species in Shark Bay (Heithaus 2005;

Heithaus & Dill 2006; Wirsing et al. 2007), it is likely

that tiger sharks play a critical role in marine ecosystems

through multiple TMII. By extension, the reductions

of large-bodied shark populations in other areas of the

world would be predicted to have substantial community level

consequences. Indeed, some of the recent increases

in turtle populations around the world (Balazs & Chaloupka 2004) may have been facilitated by reductions in shark populations that would allow turtles to forage free of energy intake–predation risk trade-offs. While such rebounds in populations may appear to be beneficial for turtle populations in the short term, the community

and ecosystem consequences of unrestricted grazing without the spatial variation in herbivory that might be imposed by predation risk remain unclear.

The loss of turtle predators could result in negative

impacts on seagrass beds. Indeed, recent declines in

seagrasses offshore of Bermuda may be linked to increases

in turtle populations (Murdoch et al. 2007) that coincide

with steep declines in tiger shark populations in the same

region (Baum et al. 2003). Together, these findings

suggest that recent declines in large shark populations

may have more widespread ecological impacts than are generally appreciated, and conservation strategies should

explicitly consider the potential lethal and non-lethal

effects of sharks, both direct and indirect.