Predator chemical cues decrease attack time and increase metabolic rate in an orb-web spider

It is well understood that the risk of predation induces notable changes in the behaviour and physiology of animals (Lima, 1998; Lima and Dill, 1990). Theoretical (Charnov, 1976; MacArthur and Pianka, 1966) and empirical evidence (Hawlena and Schmitz, 2010) suggests that prey face a trade-off where the benefits of the anti-predator response are balanced with other fitness-improving activities, such as foraging or mating. For example, Lagos et al. (2014) showed that crickets will continue foraging in a patch until the last possible second before initiating an escape response from an approaching lizard predator. Predation risk is perhaps the strongest selection pressure on a population (Lima, 1998; Lima and Dill, 1990).

Sit-and-wait foragers, like other animals, are relatively exposed to various predators, such as birds, lizards and invertebrate predators. One way of reducing exposure to predators for sit-and-wait foragers is to prevent detection in the first place through background matching (e.g. praying mantids: Barry et al., 2015; crab spiders: Heiling et al., 2005) or hiding in a retreat (e.g. leaf curling spider: Manicom et al., 2008). Orb-web spiders, however, usually reside in the middle of their web when foraging and often display bright and conspicuous colours (Bush et al., 2008; Hauber, 2002). Thus, orb-web spiders seem even more exposed to predators and consequently might moderate either their web-building behaviour or their behaviour on the web when exposed to the threat of predation. Very few studies have investigated predator-induced changes in the foraging behaviour of orb-web spiders and these have mainly focused on the effect of predation on web-building behaviours (Bruce and Herberstein, 2006; Li and Lee, 2004), perhaps because web construction is widely assumed to be the most costly component of foraging (Peakall and Witt, 1976; Tanaka, 1989) and is directly linked to foraging success (Herberstein, et al., 2000; Herberstein and Heiling, 1998). It has been shown that building web decorations (also known as stabilimenta) is influenced by the risk of predation (Li and Lee, 2004; Seah and Li, 2001) but less is known about how orb-web spiders modify the structure of their web in response to predation risk.

While the successful capture of prey in the web will result in an energy gain for the spider, it may also impose several costs. In addition to the energy expenditure of moving, running on the web changes the arrangement of the capture threads, which may affect future prey capture efficiency (Higgins and Buskirk 1992; Herberstein and Heiling, 1998; Heiling and Herberstein, 1999). More importantly, moving in the web can attract the attention of visually hunting predators such as praying mantids (Bruce and Herberstein, 2006; Herberstein, et al., 2000). Moreover, spiders in damaged webs may suffer a higher risk of predation, as spiders use vibrations on the web as cues for evading predators (Vollrath, 1985). Considering these potential costs, we expect that spiders moderate their behaviour under the threat of predation based on the relative benefit and cost of prey capture. To our knowledge, this is the first study to focus on the variation of attack behaviour in response to a predator cue in orb-web spiders.

A well-established effect of predation risk in a variety of taxa is an elevated metabolic rate (Beckerman et al., 2007; Lagos and Herberstein, 2017), moderated by the release of hormones (Orchard, 1982). These physiological changes redirect energy to locomotory organs to prepare for a rapid escape response (Orchard, 1993; Roeder, 2005). However, under constant risk, metabolism is reduced, presumably to compensate for the loss of foraging opportunities (Barry and Syal, 2013; Handelsman et al., 2013; Steiner and Van Buskirk, 2009). The metabolic rate of spiders is generally lower than that of other ectothermic animals of the same size, which may buffer spiders against prey fluctuations and periods of low food availability (Anderson, 1970; Greenstone and Bennett, 1980). Moreover, considerable variation in resting metabolic rate (RMR) has been reported between species (Anderson, 1970) and individuals (Kasumovic and Seebacher, 2013). For more active spiders, temporal variation may correspond to circadian rhythm and diel periods of activity under natural conditions. In web-building spiders, some of the variation in metabolic rate is associated with constructing or repairing the web (Anderson, 1970), but changes in response to predation have not been reported for orb-web spiders.

Argiope keyserlingi (Araneae: Araneidae) is a medium sized orb-web spider common along the East Coast of Australia, where they primarily construct their orb webs in Lomandra bushes (Blamires, et al., 2007; Rao, et al., 2007). These diurnal spiders, like other orb-web spiders, are sit-and-wait predators, resting at the centre of the web and waiting for prey, usually an insect, to intercept the web. The size and structure of the web is highly flexible and affected by various physiological factors, such as spider condition and age, and environmental factors such as prey availability and light intensity (Heiling and Herberstein, 2000). In this study, we hypothesised that chemical cues from predators influence attack behaviour, web-building behaviour and metabolic rate in the female orb-web spider Argiope keyserlingi. We predicted that under the influence of a predator cue, spiders spend less time forging, resulting in delayed responses to prey and, when attacking, shorter attack time. We also predicted that female spiders invest less in web-building behaviour, which may result in smaller webs and less frequent rebuilding of damaged webs. We predicted that the metabolic rate of female spiders becomes elevated when individuals are exposed to a predator chemical cue.

A group of 45 juvenile Argiope keyserlingi Karsch 1878 females were collected from the West Pymble area in Sydney, NSW, Australia (33°45′48″S, 151°08′10″E) during June and July 2016. A group of 20 juvenile praying mantids, Archimantis latistylus (Mantodea: Mantidae), were collected from the same area at the same time. All animals were transferred to the laboratory for further tests. Female spiders were transferred into 3D Perspex frames (40×40×10 cm), where they could build their typical orb-web. A strip of masking tape was attached to the inner side of the frames to aid the attachment of the silk. All animals were sprayed with water and fed twice a week with 2–3 small crickets, Acheta domestica. All animals were housed in a temperature-controlled room (26±2°C; 70% relative humidity) on a 12 h:12 h light:dark cycle. All experiments were conducted under the same environmental conditions. These spiders capture prey during daylight and repair damaged webs overnight; therefore, all tests were conducted between 11:00 h and 18:00 h to match the time of day the spiders normally forage.

In this experiment, we tested the effect of chemical cues from a predator on two aspects of the spiders' foraging behaviour: attack behaviour and web-building behaviour. Newly moulted adult females were transferred to a clean Perspex frame and allowed to build a web. Perspex frames were cleaned with 70% ethanol to remove any chemical cues. Spiders were randomly allocated to one of three treatments (N=15): control, predator cue and non-predator cue.

For the control treatment, spiders foraged under normal laboratory conditions where no chemical cues were introduced to them. For the predator cue treatment, spiders were allowed to forage while being exposed to the predator cue. The predator we used was praying mantids, A. latistylus, which are known to co-occur with A. keyserlingi in Lomandra bushes and to regularly attack spiders (Blamires et al., 2007; Bruce and Herberstein, 2006). For predator cue collection, a penultimate or adult stage mantid was placed in a 200 ml ventilated plastic cup containing a piece of filter paper (7 cm diameter), and fed on medium-sized A. domestica; the cue was collected on the filter paper over a period of 24 h. In trials, the filter paper containing mantid exuvia was fixed to the lid inside the spider frame perpendicular to the web hub where the spider resides. The distance from the cue filter paper to the spider was approximately 50 mm, providing an airborne cue with no physical contact. To control for the presentation of a chemical cue, we included a non-predator cue treatment. We chose a young leaf of the Sydney blue gum tree, Eucalyptus saligna, which was crushed on filter paper and placed in the same position in the frame as the predator cue. Predator and non-predator cue filter paper was replaced every 48 h. Spiders were exposed to cues for 24 h before beginning the trials. The trials were conducted from September to October 2016 and spiders were kept food deprived for 4 days prior to commencing the trials to reduce any previous feeding effect.

Using an electric toothbrush (Spinbrush Pro Clean Powered Toothbrush), we generated vibrations to simulate prey impact in the web. The frequency of the vibration was 21 Hz with an amplitude of 3.52 mm, which represents natural vibration generated by prey (Hoffmaster, 1982; Masters, 1984). Because the spiders were repeatedly exposed to this stimulus, we rewarded them by placing a small cricket (0.035±0.003 g, n=15) on the web. This prevented spiders from ignoring the stimulus with repeated exposure. The cricket was narcotised using carbon dioxide gas, and gently placed alternately along the left and right horizontal radial thread of the web to control for any gravity effect on attack speed (Herberstein and Heiling, 1999). The toothbrush was placed adjacent to the top or bottom radial non-sticky thread, which is responsible for transferring vibration to the web hub where the spider resides. The response of the spider to the stimulus was recorded using a high-speed camera (CASIO Exilim EX-F1, CASIO Computer Co., Tokyo, Japan) at 300 frames s⁻¹. We used the videos to time the spiders' behaviour: initial response time (time taken to react to the stimulus) and total attack time (time taken from leaving the web hub to move back to the initial position). Each trial was repeated seven times because this would give a better estimate of short-term behavioural changes than one single trial, which may produce biased results (Marshall and Uller, 2007). The trials were repeated every second day, allowing spiders to rest and digest prey to avoid overfeeding.

Spiders were checked daily and we recorded whether they built a new web. A photograph of the newly built web was taken to measure the web capture area (the area where spiral sticky threads are placed) and capture thread length. Capture thread length (CTL) was estimated using a modified formula (Heiling et al., 1998): CTL=[(C_hub+C_web)/2]×[(sp₁+… sp₈)/8]; where C_hub and C_web are the circumference of the hub (central area of the web) and the overall circumference of the web, respectively, and sp_1–8 is the number of spiral threads in eight sectors of the web, from the upper vertical sector moving 45 deg clockwise around the entire web. All photographs were measured using ImageJ 1.49v software (U.S. National Institutes of Health, Bethesda, MD, USA).

In this experiment, we tested the effect of predator cues on the spiders' metabolic rate, estimated through CO₂ production. Following the behavioural trials, spiders were allowed to rest for 2–3 weeks under a standard diet (one medium-sized cricket per week). Metabolic rate was measured under two treatments (N=15), non-predator and predator cue. Individuals were allocated to the same predator and non-predator cue treatments as in the foraging trial, and the cues were obtained in the same way.

We used open-system respirometry (LI-COR Li-6400XT respirometer, LiCor, Lincoln, NE, USA), in which the air that passed through the sample container was compared with reference air and additional CO₂ produced in the sample container was measured. The respirometer was set to the pre-installed ‘insect respiration’ configuration to measure the amount of CO₂ released when the spiders were inactive, where the air flow rate was set to 500 µmol min⁻¹ and soda lime and silica gel were used for CO₂ and humidity scrubber, respectively. The final computed metabolic rate was given in μg CO₂ min⁻¹, with a precision of 0.2 μmol CO₂ per mol of air. During respirometry measurements, the temperature and humidity were 27.46±0.27°C and 50.49±3.25%, respectively.

To measure metabolic rate, spiders were transferred into a cylindrical container (8 cm height×6.5 cm diameter) and allowed to habituate for 7 min. Plastic mesh was placed on the inner surface of the container to provide a grip for the spiders, thereby reducing movement and stress. During the control period, metabolic rate was measured every 5 s for a period of 8 min. Then, the container was opened for a few seconds and either the predator cue (filter paper containing the Archimantis cue) or the non-predator cue (filter paper containing the Eucalyptus cue) was placed inside and the container was closed. The respirometry machine was used to remove atmospheric CO₂ from the container for 2 min. Once the CO₂ had stabilised, the metabolic rate was measured for another 8 min. As the spiders were motionless, the released amount of CO₂ is indicative of the RMR of each individual. The spiders were observed during the respirometry trials, and if they were moving or spinning silk, indicating that they were not in a resting state, the trial was repeated to ensure that the lowest respiratory rate measurements were recorded.

The order in which the predator and non-predator cue was applied was not randomised for three reasons: first, the aim was to test the effect of a chemical cue on spiders that were initially unaware of the presence of a predator; second, if the spiders were introduced to the predator cue first, it would take too long for the metabolic rate to return to normal before applying the control treatment; third, by comparing post-experimental stages (predator versus non-predator cue), this method allowed us to ascertain whether the elevation in metabolic rate in response to the predator cue is necessarily a predator-specific response or instead is a generic response to disturbance.

All data were analysed with R 3.3 (http://www.R-project.org/). The distribution of the variables was checked before further analysis by calculating quantiles of each variable in the dataset and then plotting them against theoretical quantiles of different distributions (q–q plot) to determine whether the empirical quantiles were within the borders of a theoretical distribution. The foraging data were analysed with a generalised linear mixed model using family=Gamma with log link function for attack behaviour data, and Gaussian family with identity link function for web construction data. The different cues (predator, non-predator and no cue control) and trial number were added to the model as fixed factors. The lme4 package was used to fit models for each response variable, then models that excluded each fixed effect were compared with the model that included all fixed factors using the F-test (Bates et al., 2014). For all analyses, spider ID was added to the model as a random factor to control the variation between individuals contributing repeated data.

For metabolic rate analyses, the average CO₂ production for each individual was calculated for the control period and the experimental period (predator cue or non-predator cue) and then tested using analysis of covariance (ANCOVA), where body mass of the spiders was added to the models as a covariation factor. First, the control period was compared with the experimental period for each predator or non-predator cue (within subject test). Second, spiders in the predator cue treatment were compared against spiders in the non-predator cue treatment for each control and experimental period (between subjects).

In the foraging trials, the initial response time to a vibratory cue increased over the seven trials under the control treatment, while exposing the spiders to predator or non-predator cues significantly changed their responses (Table 1, Fig. 1). When exposed to a non-predator cue, the spiders were slower in their initial response from the third to the sixth trials, with a sudden drop in time to respond on the seventh trial. However, when exposed to a predator cue, the spiders maintained their quick response over the seven trials, without a noticeable change (Fig. 1A). Overall, the lowest average initial response time over seven trials was seen under the predator cue treatment (0.07±0.01 s), followed by the control (0.08±0.01 s) and non-predator cue (0.12±0.01 s) treatments. Total attack time was similarly influenced by the presence of different cues (Table 1). The spiders did not vary the total attack time over seven trials under the control treatment. When experiencing a non-predator cue, total attack time decreased over time. However, under a predator cue, the spiders did not vary their behaviour over the seven trials but spent much less time attacking the prey than spiders experiencing the control or non-predator cues (Fig. 1B). Overall, the lowest average total attack time over seven trials was seen under the predator cue treatment (5.26±0.72 s), followed by the non-predator cue (5.92±1.28 s) and control (6.67±1.28 s) treatments. The interaction between the cues and trials was significant for both initial response time and total attack time (Table 1).

The spiders were able to rebuild their web up to 5 times (six webs in total) during the 7 day experiment. However, the fifth and sixth webs were excluded from the web structure analysis because of low sample size. The number of times the spiders rebuilt their web was significantly lower under the predator cue than under the control or non-predator cue (Table 2, Fig. 2). The spiders doubled the capture area of their web during the experiment irrespective of treatment, while the length of the capture thread per area remained almost constant. Neither the capture area nor the capture thread length varied between the three treatments (Table 2, Fig. 3). There was no significant interaction between different cues and the number of trials for capture area and capture thread length (Table 2).

During the control period, there was an initially high rate of CO₂ production (3.5±1 µg min⁻¹), which gradually decreased to around 2 µg min⁻¹ after 8 min (Fig. 4). There was no significant difference in the average CO₂ production between the two control groups over the control period (Table 3). Once the cues were added, spiders exposed to the predator cue released around 2 times more CO₂ than those exposed to the non-predator cue. Metabolic rate did not change relative to the control period for spiders exposed to the non-predator cue (Fig. 4, Table 3).

Our findings revealed that the presence of predator cues altered the attack behaviour of female orb-web spiders and elevated their metabolic rate. As predicted, when the risk of predation was increased, the spiders spent less time foraging – they responded and attacked faster and spent less time with the prey. While the structure of the web was not influenced by the presence of a predator cue, spiders were less likely to rebuild the web under predator cue conditions.

Argiope keyserlingi females were able to detect the chemical cue from praying mantid predators and distinguish it from a non-predator cue, and adjusted their foraging behaviour accordingly. Argiope spiders have limited eyesight and it is unlikely they detect predators via visual cues. Therefore, chemical substances are more reliable cues indicating the approach of predators (Li and Lee, 2004). This response is predicted to be adaptive and has been reported for the congener Argiope versicolor (Li and Lee, 2004) and other taxa (Amo, et al., 2006; Dixson et al., 2010; Hoefler et al., 2012). The ability to distinguish a non-predator cue from a potential predator avoids the costs of inappropriate anti-predator responses, e.g. relocating the web (Nakata and Ushimaru, 2013), and initiates an appropriate reaction that may increase fitness (Buxton, et al., 2017).

There were no notable differences in initial response time between treatments in early trials. In later trials, the spiders reacted with a delay to the vibratory stimulus under control and non-predator cue conditions, while the spiders maintained their initial response time under the predator cue treatment. The initial rapid response time at the start of the experiment across all treatments may reflect a degree of food deprivation experienced by spiders before the start of the trials, which diminished as spiders were fed at each trial. A faster initial response time is beneficial as it would enable the spiders to catch intercepted prey before they escape from the web (Nentwig, 1982). Considering the benefit of capturing prey, further experiments are needed to explore why spiders in the control and non-predator cue treatments slowed their initial response time in trials 3–6. This shows that foraging behaviour in spiders is highly variable over the short term, suggesting a complex decision-making process that might be influenced by a variety of external and internal factors (Nelson and Jackson, 2011).

Overall, spiders restricted their foraging time budget under the predator cue treatment as they responded quicker to web vibration and spent less time handling the prey. This is likely to be an adaptive anti-predator tactic specifically against visually oriented hunters like praying mantids. Theoretical models predict that orb-web spiders should reduce activity and remain undetected in the presence of predator cues (Venner et al., 2006). Consistent with this prediction, Nakata and Mori (2016) showed that two species of orb-web spiders, Cyclosa argenteoalba and Eriophora sagana, reduce the time they spend building their web in the presence of predator cues.

Overall, web architecture was not influenced by the presence of a predator cue. This is surprising, as spider web-building behaviour is quite flexible and changes quickly in response to environmental factors (Eberhard, 1988; Heiling and Herberstein, 2000), including airborne predator cues (Nakata, 2008; Nakata and Mori, 2016). For example, A. versicolor (Li and Lee, 2004) responded to the spider predator Portia labiata by trading-off web and decoration size. However, in A. keyserlingi, predatory mantid cues did not induce spiders to vary web architecture, including web decorations (Bruce and Herberstein, 2006). This may indicate species-specific strategies in response to predator cues where modifying the structure of the web is not favoured by A. keyserlingi spiders.

While our spiders did not change the overall web architecture, they did rebuild their webs less frequently when exposed to the predator cue. This is more evidence that female A. keyserlingi are capable of perceiving and responding to predator cues. The data also suggest that it might be a more effective anti-predator strategy to refrain from rebuilding the web all together rather than to adjust web architecture. This would be beneficial for the spiders as they can save energy associated with web rebuilding and stay undetected by predators. While not rebuilding a web may entail some costs such as drying out of the capture silk or a reduction in the prey capture efficiency of the damaged web (Heiling and Herberstein, 1999), lower initial response time in the presence of a predator cue may compensate for reduced web-rebuilding frequency.

Female A. keyserlingi exposed to a predator cue doubled their metabolic rate (estimated through CO₂ production) compared with the control group. Similarly, visually oriented jumping spiders, Hasarius adansoni, were able to detect visual cues from a predator and consequently respond with an increase in metabolic rate (Okuyama, 2015). The proximate mechanism underlying this response is the release of stress hormones, such as octopamine (invertebrates) or noradrenaline (vertebrates). The neurotransmitter octopamine facilitates a quick escape response by redirecting energy to locomotory organs (Orchard, 1993; Orchard, 1982). If physical contact occurs, female A. keyserlingi jump off the web to escape from the potential threat; however, this escape response of spiders was not investigated in the present study. This could be explored in further studies to reveal whether there is a relationship between elevated metabolic rate and spider escape responses. It should be noted that the metabolic rate of the spiders showed a downward trend during the control phase before cue application (Fig. 4), suggesting they might not be in a physiologically resting state despite the fact that they were motionless during the trials.

While modification of metabolic rate under predation stress would be evolutionarily advantageous, it may also incur some fitness costs. Janssens and Stoks (2013) showed that damselfly larvae (Enallagma cyathigerum) produce significantly more reactive oxidative species as a by-product of elevated metabolic rate in the presence of predator cues. These oxidative factors redirect energy towards locomotory organs for anti-predator responses, resulting in energy deficiency in other functions such as growth and body maintenance (Benard, 2004; Slos and Stoks, 2008). Moreover, these damaging effects may be extended to subsequent generations through altering the reproductive strategy of the females, resulting in a lower investment in eggs and reduced offspring quality (Fontaine and Martin, 2006; Travers, et al., 2010). For example, under predation risk, the females of an egg-carrying spider, Scytodes pallida, laid eggs that hatched sooner and the spiderlings developed into smaller individuals (Li, 2002).

In conclusion, our study demonstrates that a cue of a potential predator affects the foraging behaviour and metabolic rate of A. keyserlingi females. The female spiders responded to a vibratory stimulus relatively quicker with a reduced total attack time, which may have been facilitated by an elevation in metabolic rate, although increased mobility is also likely to increase an individual's metabolic rate. While web architecture did not vary, the spiders were also less likely to rebuild their web in the presence of predator cues, which may allow them to remain undetected from visually oriented predators. This may lead to a reduction in prey capture success; however, the spiders may compensate for this by reducing their initial response time. In addition to the short-term effects on behaviour, predation risk may also have long-lasting consequences on the life history of animals (Benard, 2004). Further tests to explore the long-term effects of predator cues on individuals and subsequent generations will elucidate the evolutionary importance of predator-induced responses and facilitate a better understanding of ecological processes.

We would like to thank Justin McNab and Alexis Diodati for their help with housing the spiders and help with capturing the footage during the experiments. We also thank Jutta M. Schneider and Lizzy Lowe for their helpful comments the manuscript.