How predator hunting-modes affect prey behaviour: Capture deterrence in Drosophila melanogaster

Hunting mode, the distinct set of behavioural strategies that a predator employs while hunting, can be an important determinant of the prey organism’s behavioural response. However, few studies have considered how a predator’s hunting mode influences anti-predatory behaviours of a prey species. Here we document the influence of active hunters (zebra jumping spiders, Salticus scenicus) and ambush predators (Chinese praying mantids, Tenodera aridifolia sinensis) on the capture deterrence anti-predatory behavioural repertoire of the model organism, Drosophila melanogaster. We hypothesized that D. melanogaster would reduce overall locomotory activity in the presence of ambush predators, and increase activity with active hunters. First we observed and described the behavioural repertoire of D. melanogaster in the presence of the predators. We documented three previously undescribed behaviours-abdominal lifting, stopping and retreat-which were performed at higher frequency by D. melanogaster in the presence of predators, and may aid in capture deterrence. Consistent with our predictions, we observed an increase in the overall activity of D. melanogaster in the presence of jumping spiders (active hunter). However, counter to our prediction, mantids (ambush hunter) had only a modest influence on activity. We also observed considerable intra and inter-individual variation in response to both predator types. Given these new insights into Drosophila behaviour, and with the genetic tools available, dissecting the molecular mechanisms of anti-predator behaviours may now be feasible in this system.

predatory behaviours of a prey species. Here we document the influence of active hunters 24! (zebra jumping spiders, Salticus scenicus) and ambush predators (Chinese praying 25! mantids, Tenodera aridifolia sinensis) on the capture deterrence anti-predatory 26! behavioural repertoire of the model organism, Drosophila melanogaster. We 27! hypothesized that D. melanogaster would reduce overall locomotory activity in the 28! presence of ambush predators, and increase activity with active hunters. First we 29! observed and described the behavioural repertoire of D. melanogaster in the presence of 30! the predators. We documented three previously undescribed behaviours-abdominal 31! lifting, stopping and retreat-which were performed at higher frequency by D. 32! melanogaster in the presence of predators, and may aid in capture deterrence. Consistent 33! with our predictions, we observed an increase in the overall activity of D. melanogaster 34! in the presence of jumping spiders (active hunter). However, counter to our prediction, 35! mantids (ambush hunter) had only a modest influence on activity. We also observed 36! considerable intra and inter-individual variation in response to both predator types. Given 37! these new insights into Drosophila behaviour, and with the genetic tools available, 38! dissecting the molecular mechanisms of anti-predator behaviours may now be feasible in 39! this system. 40!

44!
Predator hunting-modes, i.e., the set of behaviours that predators employ to pursue and 45! capture their prey (Schoener, 1971;Huey & Pianka 1981;Preisser, Orrock & Schmitz 46! 2007), have been shown to induce distinct prey responses (Schmitz, 2008) that in turn 47! influence the productivity of ecological communities. In habitats dominated by active 48! hunters there is lower species evenness and higher above-ground net primary productivity 49! compared to habitats dominated by ambush hunters (Schmitz, 2008). The authors suggest 50! the observed differences in prey productivity to be driven by hunting mode specific trade-51! offs between foraging and seeking refuge. Although studies often describe the effects of 52! predators on prey traits (i.e. DeWitt, Robinson & Wilson, 2000;Reznick, Butler & Rodd, 53! 2001;Relyea, 2001), it is rare for the role of predator hunting-mode to be explicitly 54! considered.

55!
Here we investigate segregating differences in the anti-predatory behavioural 56! repertoire of the fruit fly, Drosophila melanogaster, in response to two predator species 57! differing in hunting modes. Based on (Schmitz, 2008), we predicted that fruit flies, in the 58! presence of a familiar predator, would exhibit hunting-mode specific modifications in 59! activity levels. We used D. melanogaster because, although it is one of the most well-60! studied model organisms, there is a relative paucity of information regarding D. 61! ! 9! other states (e.g. individuals cannot simultaneously walk and run). Behavioural events 178! are discrete behaviours that occur instantaneously and are also mutually exclusive with 179! each other (e.g. turning versus jumping) but not always mutually exclusive with 180! behavioural states. For example, an individual could perform a wing display (event) while 181! simultaneously walking (state), but it could not jump (event) while simultaneously 182! running (state). In this study we treated flying as an event because the structure of the 183! experimental chamber constrained flight duration. Attempted flight by D. melanogaster 184! could result in landing due to contact with a wall of the petri dish. We also recorded when 185! a fly was not visible (occluded) to the observers analysing video. We recorded a total of 6 186! discrete events and 5 behavioural states in D. melanogaster in response to predation by 187! spiders and mantids (Table 1). In order to interpret an individual fly's behaviour in the 188! context of predatory encounters, we designated two keys to describe the location of the 189! predator in regard to its interactions with the fly. As flies might alter their behaviour 190! when a predator is within striking distance, we recorded predator location based on 191! whether or not it was within striking distance of the fly (~ 5mm from the spider/mantid, 192! also see Spider location/ Mantid location in Figure 1 behavioural differences were not confounded with seasonal differences in behaviour, we 206! performed 6 additional assays (alternating between spider and mantid treatments) within 207! the span of one week. Following a spider assay, the plates were wiped down with 30% 208! ethanol followed by a rinse with RO water before a mantid assay was conducted.

209!
Additionally, the process of adding a predator to the arena invariably resulted in a 210! disturbance that likely startled the fly (unrelated to the presence of a predator). To 211! confirm that behaviours induced by this disturbance were not confounded with predator 212! induced behavioural differences, we performed 3 control assays. Here, after 5 minutes of 213! acclimatization without a predator (see above for more details), the arena containing the 214! fruit fly was disturbed gently (~ magnitude of disturbance caused by the addition of a 215! predator). For all controls, video processing and behaviours recorded were identical to 216! mantid and spider treatments described above. See Supplement b, S1 for a detailed 217!
Where y is a vector of time spent in a behavioural state. β 1 is the regression coefficient for 233! predator state, β 2 is for duration in each predator state, β 3 is for age of the fly, β 4 is for 234! temperature, β 5 is for time at which assay was started, β 6 is for sex of the fly and β 7 is for 235! date on which the assay was performed. We estimated random effects for individuals 236! including variation in response to predator state and duration of assay, and we fit an 237! independent random effect for date. Thus we fit a repeated effects (longitudinal) mixed 238! effects model allowing for variation among individuals for the influence of predator 239! presence and duration of assay where for the i th individual 240! the intervals were more difficult to estimate given the complexities of the random effect 258! structure of the model, and some caution is warranted for their interpretation.

259!
To test for non-random associations in the temporal structure of behavioural 260! patterns we constructed transition frequencies using the "msm" library (version 1.2) 261! (Jackson, 2011) in R. To test for both for first order Markov processes between 262! behaviours (transition probabilities), as well as the influence of predator presence on 263! these transition probabilities, we fit log-linear models (assuming poisson distributed data) 264! with the transition frequency matrices (Crawley, 2012) using glm in R. As advocated by 265! (Crawley, 2012;Bakeman & Gottman, 1997) we fit a saturated log-linear model (with 266! lag0, lag1 and predator state as the effects in the model) and tested the influence of 267! deleting the terms (i.e. third order interaction) on change in deviance. We used modified 268! "Z-scores", adjusted using sequential Bonferroni to assess the deviation of particular cells 269! in the transition frequency matrix from expected values (assuming independence). For the 270! visual transition probability matrices, we combined the behavioural event "pause" with 271! the behavioural state "stop" because 1) we wanted to reduce the complexity of the matrix 272! and 2) the main difference between the two behaviours is that pause is instantaneous and 273! stop has duration. All transition diagrams were constructed in Inkscape (version 0.48.2, 274! Harrington, 2004(version 0.48.2, 274! Harrington, -2005. 275!

277!
From pilot observations (not included in analysis), we (I.D., A.P. and C.P.) 278! catalogued and described Drosophila melanogaster behaviours observed in the presence 279! of a predator (Table 1). Among the behaviours listed in Table 1, abdominal lifting (ab, 280! supplement b, video 1) and retreat (supplement b, video 3), to our knowledge, have not 281! been previously described in D. melanogaster literature.

Flies perform a range of anti-predatory behaviours in response to a zebra spider 284!
To visualize each individual fruit fly's response to the presence of a zebra 285! jumping spider, we generated ethograms (see Figure 1a and Supplement a). For the two 286! predator states (spider present and spider absent) we measured the mean proportion of indicates whether or not the predator was within striking distance of the fruit fly at that time point. This information is relevant only after the predator was added to the chamber (~ 300 s into the assay). Dark grey bars in Predator location indicate that the spider was within striking distance and light grey regions indicate that the spider was out of striking distance. Predator location is white when the predator is absent from the arena or after successful capture. If capture did not occur, predator location remains light grey in colour.   Error bars are ± 95% CI.

14!
while grooming 60% less (95% CI: 43-77% decrease). This is shown in Figure 2 and 291! Supplement b Figure S1 (treatment contrasts with 95% CI in figures are provided to 292! enable assessment of significance). While they were observed at low frequencies prior to 293! the addition of spiders, D. melanogaster substantially increased the frequency of pauses, 294! jumps and flights (per minute) in the presence of spiders ( Figure S2). However in the presence of spiders, 300! the average total time spent "stopping" increased to ~25.8 seconds (95% CI: 10.1 -41.7 301! seconds). When interacting with spiders, flies were only observed to perform the 302! "retreat" behaviour once (of 30 individuals). Interestingly, we did not see significant sex 303! specific differences in either frequencies of occurrence (Supplement b Figure S3) or 304! proportion of time allocate (Supplement b Figure S1) to the majority of measured 305! behaviours (But see S3 panels "pause" and "turn").

306!
Given the design of our experiment, we were able to model the degree to which 307! individuals varied in their responses to the jumping spiders. Individuals varied greatly 308! both in their baseline activity levels as well as in their propensities to respond to jumping 309! spiders. The among-individual coefficient of variation for time spent grooming in the 310! absence of predators was 57.7% (40.1-74.2%). While most individuals reduced their 311! grooming activity in the presence of predators, the degree to which they did so varied 312! substantially, with the among individual coefficient of variation for the decrease being 313! ! 15! 67.8% (26.5-94.6%), as shown in Figure 3a. For walking, the among-individual 314! coefficient of variation was 80.3% (50-105%) in the absence of the spider, and 135% (1-315! 181%) for the magnitude of increase in the presence of the spider (Figure 3b).

316!
Performance of the stopping behaviour by D. melanogaster in the presence of spiders 317! varied substantially among individuals, with the among-individual coefficient of variation 318! being 168% (95% CI: 123-214%). This is driven in part by the fact that 40% of 319! individual flies never performed stopping, even in the presence of the spider. There was 320! a negative correlation (-0.84), between the amount of time individuals spent grooming 321! before and after the addition of the spiders (Table 2). That is, on average, individuals who 322! were more active prior to the addition of the spider reduced their activity to a greater 323! extent in the presence of the spider. A similar negative correlation (-0.66) for among 324! individual activity for locomotion, was observed (Table 2).

325!
To visualise the temporal associations among behavioural sequences, we 326! constructed transition matrices (Supplement b Tables S1, S2, S5 and S6) and transition 327! probability diagrams for all pairs of behaviours in the presence ( Figure 4a) and absence 328! (Supplement b Figure S7) of spiders. In response to jumping spiders, transitions among 329! behaviours are somewhat more dispersed (with many connections between behaviours), 330! suggesting that there is weak temporal association between fruit fly behaviours. Indeed 331! these qualitative conclusions are supported based on the Z-scores. In the absence of 332! spiders 8 possible transitions were significant (after controlling for multiple comparison, 333! Supplement b Table S2), while 13 transitions were significant in the presence of the 334! spider (Supplement b Table S1). Most of these differences were due to the increase in Estimates are derived from the predicted values for each individual from the mixed models.  Although the presence of a mantid had a small effect on fly behaviour, flies did 359! vary considerably in their grooming and walking activities. Indeed, the among-individual 360! variability in proportion of time spent grooming and walking is greater in magnitude in 361! the presence of the mantids than spiders (Figure 3). Evidence for negative co-variation 362! for intra-individual behaviour before and after the addition of the predator was not 363! strongly supported (i.e. 95% CIs for covariances included zero) ( Table 2).   indicates mean number of occurrences per minute of that behaviour. To reduce the complexity of the web we combined the behaviours "pause" with the behaviour "stop".
Behavioural transitions that occurred less than 10 times have not been shown in the figure.

382!
Prey organisms can alter their behaviour to reduce the likelihood of detection, 383! capture or encounter with a predator (Lima, 1998). For example, when predators are 384! present, ground squirrels dedicate more time to vigilance behaviours (like scanning for a 385! predator, see (Bachman, 1993) and some aquatic insects spend more time in refuges 386! (Kohler & McPeek, 1989). These changes in behaviour may alter the use of resources, 387! and potentially the fitness of an organism. However, the nature and intensity of non-388! consumptive effects of a predator on its prey are a function of several predator specific 389! factors, one of which is the predator's hunting mode (Preisser, Orrock & Schmitz, 2007).

390!
Predator hunting mode, i.e., the set of behavioural strategies that a predator employs to 391! pursue and capture its prey (Schoener, 1971;Huey & Pianka, 1981;Schmitz, 2008) can 392! be an important determinant of a prey organism's anti-predatory behavioural response 393! (Schmitz, 2008). In this study, we describe the anti-predatory behavioural repertoire of a 394! natural population of Drosophila melanogaster in response to predation by the zebra 395! jumping spider (Salticus scenicus) and juvenile Chinese praying mantids (Tenodera 396! aridifolia sinensis). Among other characteristics, zebra spiders and praying mantids differ 397! in their hunting mode. While we discuss our findings with respect to hunting mode 398! differences, we recognize that other attributes differing among the predators may 399! contribute to the observed differences in prey behavioural repertoires. However, as our 400! experimental design was meant to minimize the effects of many possible confounding 401! factors (e.g. time of day, temperature, humidity) it seems likely that, in part, our results 402! reflect hunting mode differences. 403! !

19!
In response to active hunters (those that constantly patrol for prey), we predicted 404! that fruit flies would increase their overall activity levels (including flight) in order to 405! maintain maximum distance from the predator at all times; To reduce the likelihood of an 406! encounter with an ambush predator however (i.e., a predator that only attacks when a 407! prey organism wanders in to its strike zone), we predicted that D. melanogaster would 408! respond by decreasing locomotory activities (Schmitz, 2008). Our results, however, were 409! only partially in line with these predictions. While the actively hunting jumping spiders 410! induce a clear increase in overall activity, we found the presence of juvenile mantids-our 411! ambush predators-to have minimal influence on fruit fly activity levels ( Figure 2, 412! Supplement b Figure S2). It has been previously argued that ambush predators might be a 413! predictable source of threat to prey organisms (Preisser, Orrock & Schmitz, 2007;414! Schmitz, 2008) as opposed to the diffuse and variable threat imposed by active hunters 415! (Schmitz, 2008). Therefore, it is perhaps surprising that fruit flies show a stronger 416! behavioural response to the threat of active hunters (zebra jumping spiders). However, 417! our predictions are based on studies on a grasshopper and its two predatory spider species 418! that differ in hunting mode. Given that selection pressures faced by adult diptera are 419! different from those experienced by grasshoppers (orthoptera), such predictions may not 420! having experienced a longer evolutionary history with small jumping spiders, are better 427! able to recognize these spiders as a threat. In addition, the disturbance created by a 428! constantly patrolling zebra spider may be partly responsible for the increased activity 429! levels seen in D. melanogaster (either due to actual mechanical disturbance or because 430! flies are able to detect moving objects quicker than stationary ones). In this study, we are 431! unable to tease apart the effects of evolutionary recognition versus constant mechanical 432! disturbance on the differences in flies' activity levels. Further experimentation with 433! harmless but constantly moving heterospecifics (such as field crickets) or immobilized 434! active hunters might be useful in addressing these issues.

435!
We also identified a number of (to our knowledge) undescribed behaviours of D. 436! melanogaster, potentially relating to its interactions with predators. The behaviour we 437! called "stopping" (Table 1) was observed numerous times after a direct (but failed) attack 438! by a spider (Supplement 3 video 1). While D. melanogaster will spend time without any 439! ambulatory activity (walking, running), they are almost always observed to be active 440! (generally grooming) during these periods. However, when fruit flies performed the 441! stopping behaviour, there was a complete lack of movement on the part of the fly, even 442! when video was viewed at a few frames/second. When a fruit fly was "stopped", the 443! spider had to search for the fly, irrespective of the physical proximity between the spider 444! and the fly. In salticids, while the principal eyes have high spatial acuity, secondary eyes 445! are primarily used to detect moving objects (Harland, Jackson & Macnab, 1999;Land, 446! 1971). Because salticids are unable to accommodate by changing the shape of their lens, 447! they need to extensively sample their visual field to see details in object shape and form 448! (Harland, Jackson & Macnab, 1999;Land, 1971;Blest, Hardie & McIntyre, 1981 in function to fin flicking in tetras, (Brown, Godin & Pedersen, 1999). Finally it may be 473! an indication of some sort of physiological priming of the fly in preparation for a fight-474! or-flight response. Determining whether it is a specific anti-predator behaviour, as well as 475! the details of its function need to be a focus of future work.

487!
We also investigated how the presence of the different predators may influence 488! non-random associations among behaviours. We observed that in the presence of both 489! predators there was an increase in the number of behavioural transitions that deviated 490! from expectations under independence (from 12 to 23 with the mantid, and 8 to 13 with 491! the spider). Despite this, the log-linear model (analysing the whole transition frequency 492! matrix) did not support the influence of predator state on the frequencies of transitions.

493!
This may be partly due to the relatively modest sample sizes (in terms of both individuals 494! ! 23! and transitions among behaviours). Further work is necessary to validate and extend this 495! sequential analysis.

496!
While we show that there are some predator hunting-mode specific behavioural 497! differences in D. melanogaster's anti-predator response, we reiterate two important 498! caveats. First, although the primary distinction between the zebra jumping spider and 499! juvenile Chinese praying mantids as predators is their hunting-mode, other factors 500! between these species (for example, size, colour, odour) may also influence differences in 501! fruit fly behaviours. Replicating the observations with other predator pairs that differ in 502! hunting-mode is necessary to confirm hunting-mode's influence on anti-predatory 503! repertoires. Secondly, our assay chambers are an artificial environment and do not 504! resemble the conditions under which D. melanogaster face predators in the wild. Due to 505! the nature of our assay chamber, D. melanogaster were unable to employ behavioural 506! strategies that may reduce encounters with predators (e.g., utilizing a refuge). Therefore 507! we were only able to describe the capture-deterrence repertoire of D. melanogaster 508! behaviour. We believe that our study is a necessary first step to describing and 509! documenting the complete anti-predatory behavioural repertoire of D. melanogaster and 510! we foresee future work to be conducted in a modified chamber, under more "natural" 511! conditions. Doing so will allow us to take this premier model genetic system and make it 512! into an ecological model as well.

Differential response to spiders versus mantids 7!
Because spider and mantid population densities vary by season, we had to temporally 8! segregate the spider assays from the mantid assays. We conducted all spider observations 9! between October and December 2012 and all the mantid observations between March and May 10! 2013. Comparing time allocation and frequencies of occurrences in the predator absent state 11! between the two predator treatments suggest that behavioural modifications were predator 12! induced, and not due to seasonal effects ( Figure S5 and S6). Although the assays were carried 13! out under highly controlled conditions, to confirm that predator species-specific behavioural 14! differences were not confounded with seasonal differences in behaviour, we performed 6 15! additional assays (alternating between spider and mantid treatments) within the span of one 16!
week. The control experiments show no evidence of confounding effects of season with D. 17! melanogaster's anti-predator behavioural repertoire (Table S9, S11 and S12 below). Ethograms 18! are shown in Supplement a. Furthermore, to confirm that the disturbance we caused (to the assay 19! ! 2! chamber) during the addition of a predator did not confound behavioural responses to the 20! predator, we did 3 "no predator" control assays. For these "no predator" controls, instead of 21! adding a predator to the arena, we caused a mild disturbance (~ to intensity of disturbance caused 22! while adding the predator) without actually adding any predator. We found that disturbance 23! caused during predator addition was not responsible for observed behavioural modifications 24! (Table S10 and S13). Finally, "no predator" controls also ruled our temporal differences in fruit 25! fly activity levels (Table S10 and