Spatial accuracy of a rapid defense behavior in caterpillars

Caterpillars, the larvae of butterflies and moths, have diverse defense mechanisms to cope with their many enemies. When developing from hatchling to pupa, they need to avoid being eaten by birds, mammals and other insects and avoid being parasitized by wasps and flies. Generally speaking, there are three categories of first-line defense mechanisms: (1) toxicity and/or warning signals to discourage attack as seen in the Spicebush Swallowtail caterpillar (Papilio troilus), which mimics a snake (Van Zandt Brower, 1958); (2) camouflage without crypsis by mimicking plant structures such as sticks (De Ruiter, 1952) or other inedible objects (Skelhorn et al., 2010); and (3) cryptic behavior by hiding under leaves and in leaf rolls, as seen in the Brazilian Skipper (Calpodes ethlius). When this first line of defense fails, there is a second line of defense consisting of sudden, startling, movements. One of these behaviors is the strike response in which the animal rapidly swings its head and thorax toward the abdomen. More exotic behaviors include rolling downhill as seen in larva of the mother-of-pearl moth (Pleuroptya ruralis) (Brackenbury, 1997) and making whistling noises by pushing air through the spiracles, as observed in the North American walnut sphinx (Amorpha juglandis) (Bura et al., 2011).

This study focuses on the strike behavior, its range, precision and speed. The strike behavior was described first by Frings in several caterpillar species (Frings, 1945) and later studied in more detail in the tobacco hornworm (Manduca sexta, Linnaeus), the larva of a large sphinx moth (Walters et al., 1996; Walters et al., 2001). When the caterpillar is pinched or poked in the posterior abdominal body segments, it will move its head towards the location of the pinch. When pinched in the more anterior abdominal body segments and thorax, the animal moves away from the stimulus (Walters et al., 2001). There is much variety in the strike behaviors; often, the animal strikes several times after a single stimulus. These behaviors are seen more often after repeated stimuli. Another behavior seen after multiple stimuli is thrashing, in which the animal rapidly moves its head from side to side (Walters et al., 2001).

Although the strike behavior seems simple, there are several reasons why producing this behavior is not straightforward. First, caterpillars are soft-bodied animals, which means that they have a large field of motion: they are able to twist and turn into angles that are not possible for animals with a rigid skeleton. However, soft-bodied animals also have structural limitations that restrict certain movements. For example, although Manduca sexta can curl ventrally into a ring with no need for external attachments, it cannot bend itself into a tight dorsal curve unless it has appropriate anterior and posterior anchors (see Results). Caterpillar muscles are arranged in longitudinal and oblique blocks but caterpillars lack the circumferential musculature seen in earthworms (Peterson, 1912), which might explain why caterpillars cannot bend this way.

A second reason why controlling the strike behavior might be difficult is that there are no known proprioreceptive sensors that can be used to guide the body position. The response properties of the stretch receptor organs (SROs) located close to the cuticle in each body segment are not very useful for fast, precise signaling of local stretch and nor does SRO ablation affect the caterpillar's locomotion (Simon and Trimmer, 2009). Other mechanosensors, such as the alpha-type cells of the multidendritic complex that tiles the inner cuticle, have been proposed to be proprioceptive but there is no direct evidence for this suggestion in Manduca (Grueber et al., 2001). A related complex of neurons has been identified in both the larvae and adult Drosophila melanogaster (Grueber et al., 2002; Shimono et al., 2009). A subset of the neurons in this complex in Drosophila (class 1) has been assigned a proprioceptive role based on the location they target in the neuropil (Grueber et al., 2007). Other classes of neurons in this complex have sensing functions. In the larval stage of the fruitfly, the class IV neurons are necessary to evoke the rolling behavior that these larva use to escape parasitoid wasps (Hwang et al., 2007).

Third, proprioception and the control of movements are further complicated by the body plan of caterpillars. Caterpillars have distinctive body segments, but internally these segments are not separated by septa, unlike in earthworms. The caterpillar has an open hemacoel through which the gut can freely move during locomotion (Simon et al., 2010). The caterpillar's body does not necessarily function as a hydrostat because it has spiracles that can open, so it does not have a fixed volume (Lin et al., 2011). Additional evidence for this comes from pressure measurements in crawling caterpillars: increases in pressure were complex, with large peaks corresponding to large movements (Mezoff et al., 2004). In fact, the pressure changes inconsistently with the smallest motions, which makes it difficult to establish a baseline (Lin et al., 2011). This is stark contrast to the clear pressure changes seen in the earthworm, which is a true hydrostat (Quillin, 1998).

It is not clear whether the strike behavior in caterpillars is a general defense or is directed at a particular enemy. In a previous description of the behavior, two functional roles were suggested: (1) the movement could be used to startle predatory birds or to cause the bird to release the caterpillar after picking it up from a plant; and (2) the movement could be used to remove parasitoid insects (Walters et al., 2001). During striking movements, Manduca opens its mandibles and scrapes them along the stimulated side, often repeatedly. This could be used to remove parasitoid wasps (Cotesia congregata) (Gilmore, 1938) or parasitoid flies (Winthemia manducae) that lay eggs in the caterpillar's body cavity (DeLoach and Rabb, 1971; DeLoach and Rabb, 1972).

The precision of the strike might be important for the ability of the caterpillar to remove a parasitoid wasp, and the caterpillar body needs to curve to various degrees for the mandibles to reach different locations on the body. Our hypothesis was that the precision of the strike at a point stimulus varies along the body, with lower accuracy for anterior and dorsal stimulus locations. We also predicted that there is a tradeoff between speed and precision, with slower strikes being more precise, as predicted by Fitts' Law. Such a trade-off has also been seen in human reaching movements and is explained by noise in sensory feedback and motor commands, which increases for faster movements (Fitts, 1954). Our study had three main goals: (1) to make an inventory of the range of behaviors seen after noxious stimulation, (2) to determine the precision with which the head reaches the stimulation site and how this varies for different locations on the body and, (3) to measure the speed of the strike behavior.

Experiments were carried out on Manduca caterpillars from a colony at Tufts University. The caterpillars were raised on an artificial diet and on a 17 h light/7 h dark cycle (Bell and Joachim, 1978). On the day of recording, the caterpillars, both male and female, were in their third day of the fifth instar. The caterpillars weighed 1.776±0.325 g (mean ± s.d.) and were about 47 mm in length (43.6±6.6 mm). In total, four caterpillars were used for recordings. To aid in the visibility of the different locations on the animal's body, 2nd day fifth instar larvae were chilled for at least 20 min on ice then marked using a black ultrafine Sharpie pen (Sanford, L.P., Oak Brook, IL, USA). The caterpillars were allowed to recover from the handling and anesthesia for a full day before the experiments were started.

Video recordings were made using a Casio Ex-F1 HD camera at 59.94 frames s⁻¹ (Casio, Shibuya, Tokyo, Japan). The M. sexta caterpillars were placed on a short wooden dowel (66 mm length, 8 mm diameter) facing a mirror at 45 deg (Fig. 1A,B). The dowel was covered in graph paper so that strikes landing on the dowel could be categorized. The camera was placed behind the animal so the caterpillars' dorsal side could be seen in the mirror. The dowel was turned 22 deg to one side or the other so that the stimulus location was in view.

Stimuli were applied only to body segments posterior of abdominal segment 2 (A2) because more anterior stimuli generally lead to avoidance behavior rather than strikes (Walters et al., 2001). To record strike precision, the caterpillars' body was mapped as a grid system based on anatomy. This allowed animals of different sizes to be compared and also corrected for differences in posture such as degrees of stretch along the body or between individuals. The same grid system was used to describe both the stimulus site and the strike site. The animals were marked with several horizontal lines (dorsal midline, lateral midline at the spiracles, a line between the dorsal midline and the lateral midline and a line above the prolegs) as visual guide for the radial categories. Radially, the animal was divided into 13 categories, starting on top with the dorsal midline (0) and then with three categories between each horizontal pen line (1–3, 4–6, 7–9). A strike on the proleg was categorized as 10, one below the proleg (on the wooden dowel) was marked as 11, while low strikes on the dowel, more than 0.5 cm below the proleg, were marked as category 12. Stimuli were applied to both left and right sides of the caterpillar, but because caterpillars are symmetrical the results from the left and right sides were pooled. There were 23 lateral categories, four for each body segment A3–A7 and three categories for the terminal proleg (TP). Each body segment has seven skin folds dividing the segment up into eight annuli. Each lateral category contained two annuli. The anatomy of the caterpillar did not allow the drawing of guidelines as done for the radial categories because the circular folds may obscure these guidelines. The TP proleg was divided into three lateral categories: from A6 to the dorsal horn, from the horn to the edge of the anal flap, and the terminal prolegs, which are located under or behind the anal flap depending on how much the caterpillar has rotated these legs (Fig. 1C–E).

The stimuli were applied with a blunt fiber-reinforced plastic pipette filler (0.35 mm outer diameter, World Precision Instruments, Inc., Sarasota, FL, USA). Because this tube is flexible it produces a constant maximal force when bending of 0.108 N. The duration of each stimulus was measured from the video and averaged 0.39±0.192 s. Each animal received stimuli at 20 locations on their bodies: eight locations were used to determine radial precision in A4 (anterior) and TP (posterior). The other 12 locations were used to determine longitudinal precision: two per body segment and all distributed slightly above the spiracle. The sequence of these 20 stimuli was randomized using a random number generator (www.random.org) for each trial because of sensitization to stimuli (Walters et al., 2001). Stimuli to the left and right sides were also randomized, except for stimuli on the dorsal midline. If the animal did not respond to the first few stimuli at a particular location, this location was stimulated again at the end of the trial. When a caterpillar did not return the anterior part of its body to the starting position, it was gently pushed back onto the dowel before receiving another stimulus. Each animal was used for one trial only. Strikes were omitted when the caterpillar came into contact with the pipette filler. Trials with strikes to fewer than 16 different stimuli were omitted.

The strike site was defined as the location where the mandibles first touched the cuticle. To determine the precision of the strikes, still images of the stimulus and the first contact of the mandibles with the cuticles were taken from the video using video software (VirtualDub by A. Lee, http://www.virtualdub.org/). Both the radial and longitudinal positions of the stimulus and strike location were recorded. The number of frames from the beginning of the stimulus, end of the stimulus, beginning of the strike and first contact between mandible and cuticle allowed for calculation of the duration of the stimulus and the duration of the strike. Because timing was based on a video with a frame rate of 59.94 frames s⁻¹, the time measurements are in 0.01668 s increments. Often, strikes were followed with scraping of the mandibles over the cuticle. To determine whether this movement brought the mandibles closer to the stimulus site, the trajectory was calculated from the location of the end of strike (the beginning of the scrape) and that of the end of the scrape.

Strike behavior changes slightly over time as the caterpillar becomes sensitized to the stimulus. For this reason, we recorded the number of stimuli and strikes so these could serve as covariables in the statistical analysis. Other variables recorded were the presence of additional strikes after a single stimulus, and the sex and mass of the caterpillar. In addition, behaviors before and after the strike were noted: the position of the head (curled under the body or not), pulling of the head in the contralateral direction before striking, and thrashing (rapidly moving the head from side to side, no contact of the mandibles with the cuticle). Any other notable behavior was recorded and described.

The body plan of the caterpillar can be described as a pressurized cylinder. When a pressurized cylinder is bent into a tight curve, the body wall undergoes extreme deformation by stretching on the outer surface and compressing on the inner surface, potentially leading to buckling (Lin et al., 2011). Because this effect is more extreme for small radius curves, we expected posterior and ventral strikes to be more precise than dorsal and anterior strikes. Strike responses were expressed as the number of radial or longitudinal categories (see ‘Experimental design’) between the stimulus location and the location at which the mandibles would first touch the cuticle. Lower precision for dorsal and anterior stimuli would mean that the strike response would show higher variance when compared with the data from more ventral and posterior stimulus locations. In other words, when plotting the strike response data against their stimulation sites (numbered categories) on the x-axis, we expected to see a funnel shape with data points clustered tightly for posterior and ventral stimulation sites. However, this was not the case (see Figs 2, 3), so we decided to test for accuracy (see Fig. 1F for a definition of precision versus accuracy). Accuracy was tested using a mixed model regression, which tests the slope of the data points (in Figs 2, 3) against the expected slope (no difference between stimulus and strike locations: gray dashed line in Figs 2, 3). Our tests were done on four different animals; to control for differences among animals, animal identification number was included as a random variable in each model. In addition, previous research reported sensitization after repeated stimulation (Walters et al., 2001), so the number of received stimuli before each strike was included as a covariate. Three tests were done for accuracy: longitudinal data, radial data for A4, and radial data for TP. For each test we plotted the residuals to look for non-random patterns.

We originally expected a trade-off between the duration of the movement and the precision of the strike (Fitts' law) (Fitts, 1954; Guigon et al., 2008). We used duration of the movement instead of speed because it is difficult to measure the distance that the head moves during a strike and we believe the time between the start of the movement and when the mandibles touch the cuticle is biologically more relevant than the speed. We did note that the caterpillar slowed down near the end of the strike, but have not quantified this. Because plots of the data showed no differences in precision among strike sites (see above), we instead tested for duration of the movement versus accuracy; we hypothesized that there would be a trade-off between the accuracy and duration of the movement. We used a mixed model regression with animal identification number as a random variable. Because accuracy is influenced by the location of the stimulus, this was also included as a covariate. Three tests were done: longitudinal data, and radial data for A4 and TP. Analyses were run using SAS (version 9.2, SAS Institute Inc., Cary, NC, USA).

To test for an increase in accuracy between the first and second strikes, we compared the number of categories between stimulus location and first contact of the mandibles for the first and second strike using a mixed model approach in SPSS (version 17.0, IBM, Armonk, NY, USA). Comparisons were made for longitudinal accuracy and radial accuracy. The data for A4 and TP in the radial dataset were combined because of the small sample size.

There was a considerable variation seen in the responses to stimuli. Defensive behaviors that we observed in our caterpillars include: (1) a single strike, often followed by searching or grooming behavior (locally scraping the mandibles over the cuticle from anterior to posterior); (2) strikes towards the abdomen that were preceded by a movement to the contralateral side, placing the body into a curve; (3) multiple strikes following a single stimulus with searching or grooming behavior near the stimulation site; and (4) thrashing, which includes violent alternating left and right movements. These behaviors have been mentioned before (Frings, 1945; Walters et al., 2001), but we also saw a novel response that we describe as a barrel roll. The barrel roll was only seen when a persistent stimulus was applied to the dorsal mid-line of one of the more posterior body segments. After the start of the stimulus, the caterpillar moved toward the posterior body segment either left or right and then switched sides without bringing its body back into the rest position (see supplementary material Movie 1). The barrel roll was not seen without the stimulus object being pressed against the cuticle, suggesting that this was needed, perhaps mechanically, for the animal to make the movement. Another novel movement we observed was throwing. Throwing occurred when the caterpillar found the stimulus object during search and subsequently grasped it with its mandibles or wrapped the anterior body around it. The caterpillar then made a fast, forceful movement with the head and thorax back to the rest position as if trying to throw the stimulus object from its body.

To look at how accurately the caterpillar can reach a stimulus location, we focused on single strikes and the first strike in those instances where there were multiple strikes. We expressed the accuracy as the number of radial or longitudinal categories on the body (Fig. 1E) between the stimulus location and the location where the mandibles first touched the body.

Radial accuracy of the strike was lower when the animals responded to more dorsal stimuli. This was the case for stimuli applied to both body segment A4 (d.f.=1,10, F=6.02, P=0.03) and body segment TP (d.f.=1,7, F=248.49, P<0.0001) (Fig. 2, Fig. 4A). No pattern could be seen in the residual plots, which suggests there is no difference in variance and that precision is uniform along the range of radial stimulus locations. Interestingly, the number of stimuli previously received had no effect on the radial accuracy of the strike. This was true in both A4 (d.f.=1,10, F=0.01, P=0.92) and TP (d.f.=1,7, F=0.00, P=0.95).

The longitudinal accuracy of the strike was lower when the caterpillars responded to a more anterior stimulus (d.f.=1,29, F=35.68, P<0.0001) (Fig. 3, Fig. 4B). The number of stimuli the caterpillars had received previously significantly improved the accuracy of their strike (d.f.=1,29, F=4.98, P=0.03). This is in contrast to the data for radial accuracy. No pattern could be seen in the residual plots, which suggests there is no difference in variance and that precision is uniform along the range of longitudinal stimulus locations.

We hypothesized a trade-off between the duration of the movement and its accuracy for each of the three datasets (longitudinal accuracy, and radial accuracy in A4 and TP). As can be seen from Fig. 5A, the duration of the movement appears to negatively influence the accuracy of the strike for the radial data in TP, which is opposite to our prediction. However, the change in radial accuracy in TP cannot be explained by the duration of the strike (TP: d.f.=1,7 F=0.21, P=0.66). The radial accuracy in A4 does show a trade-off between duration and accuracy (A4: d.f.=1,10, F=10.66, P=0.01) (Fig.5B). No trend is visible for the longitudinal data (d.f.=1,29, F=0.79, P=0.38). In all cases there is an effect of the location of the stimulus on the duration of the strike (A4: d.f.=1,10, F=22.82, P=0.01; TP: d.f.=1,7, F=175.59, P<0.0001; longitudinal data: d.f.=1,29, F=30.36, P<0.0001).

Single stimuli were followed by multiple strikes in about 1/3 of the responses analyzed (22 out of 64). After the first strike, the animal would move towards the rest position, but before reaching it, initiate another strike. These second strikes did not significantly change in longitudinal accuracy (d.f.=1,0.1, F=11.06, P=0.74, 11 s strikes). On average, the second strike landed about one category posterior to the first strike. Radial accuracy (11 s strikes) increased, but not significantly (d.f.=1,5.368, F=3.616, P=0.112). On average, those strikes landed two categories more dorsal than in the first strike. In a small number of trails the caterpillars responded with a third (7 out of 64), fourth (3 out of 64) and even fifth strike (2 out of 64). Because there were very few examples of more than two strikes, we did not include them in our analysis.

Often, strikes were also followed by grooming or scraping movements (30 times out of 35 strikes following a longitudinal stimulus, 14 times out of 29 strikes following a radial stimulus, A4 and TP combined). Although the scraping movements brought the mandibles closer to the stimulus site by moving them more anterior, radially there was very little improvement and only one observation was made in which the mandibles moved through the stimulation site (see Fig. 6).

Manduca sexta caterpillars have a diverse set of behaviors when responding to noxious stimuli applied to the abdominal body segments. The ‘basic’ behavior is a strike in which the caterpillar rapidly turns its head toward the stimulus site. This behavior has many variations including repeated strikes ipsilateral and contralateral as well as interactions with the stimulus applicant and continued scraping of the skin with the mandibles.

Manduca sexta caterpillars do not reach all areas on their body equally well during the strike behavior. Radial accuracy is significantly lower when caterpillars respond to stimuli that are more dorsal (Fig. 2). Longitudinal accuracy is significantly lower when caterpillars respond to stimuli that are more anterior (Fig. 3). In general, strikes landed consistently more ventral and posterior to the site of the stimulus (Fig. 4).

As predicted, slower strikes were more accurate (Fig. 5), but only for radial accuracy and only in A4. These strikes were responses to more anterior stimuli and we suspect that the trade-off reflects a mechanical or muscular limitation related to the tight curving of the body when responding to stimuli in A4. Second strikes following a single stimulus were not more accurate than the first response.

Although this paper focuses on the first strike and its accuracy, the behavior displayed can be quite variable and includes multiple strikes and curving of the body before striking. In addition, the caterpillar sometimes ‘grooms’ the stimulus site by opening and closing the mandibles while scraping them over the cuticle. This scraping is almost exclusively done by moving the head from posterior to anterior, which may explain why the strike lands posterior to the stimulus. Plotting the trajectory of the scraping movement revealed that generally this movement brought the mandibles closer to the stimulus site, but only longitudinally (see Fig. 6). We found that the second strikes were radially closer to the stimulus site than the first strikes. Because the caterpillar does not return to the rest position before starting a second strike, there might be a biomechanical reason (less momentum) for the increased radial accuracy in the second strike.

Radial accuracy is lower when responding to more dorsal stimuli; this was true when testing in A4 and TP. We suspect a limitation in anatomy can explain this phenomenon, perhaps in the layout of the muscles. When responding to a stimulus the caterpillar does not always follow the shortest route. For example, to reach a stimulus on the dorsal midline in A4, the caterpillar does not do a back bend. Instead, it lifts its head and thoracic legs off the substrate and then strikes either left or right, curving the thoracic segments upward. A direct backbend is presumably more difficult to execute because it requires the dorsal muscles to stretch the large ventral muscles. Manduca can bend itself into this shape but it usually does so by gripping a dorsal substrate to provide bending leverage. Another reason might be that the caterpillar's goal is to get not just the head near the stimulation site but the mandibles in particular. As the mandibles are located ventrally, a backbend would require even more curvature of the body than bending the body left or right.

This ventral bias in strike location was also seen in the tests of radial accuracy where all the strikes (N=17 strikes for A4, N=13 for TP) landed ventral to the stimulus site. In many cases the animal touched the substrate at the end of the strike. Under natural conditions, Manduca are found hanging underneath leaves in 78% of cases (Madden and Chamberlin, 1945); we have not tested strike accuracy with the caterpillar in starting orientations other than right-side up on top of a dowel. If gravity has an effect on the accuracy of the strike, we would expect the strikes to be more accurate radially when the caterpillar is hanging upside down. In those circumstances the part of the body that is not in contact with the substrate during the strike (head and thorax) might be pulled more dorsally by its weight, counteracting the ventral bias we see in the caterpillar that is on top of a dowel.

Strikes were more accurate longitudinally when the stimulus was applied to more posterior body segments. Again, we suspect this is related to the increased difficulty and larger forces needed to curve the body around a small radius. Perhaps more behaviorally interesting is the finding that few of the strikes in our dataset (5 out of 35 strikes for longitudinal accuracy) were located anterior to the stimulus. It is possible that striking posterior to the stimulus is intrinsically beneficial; a parasitoid wasp at the strike location could still be hit by other body parts and even removed by the anterograde scraping movements that often follow a strike. We also observed that the longitudinal accuracy (but not radial accuracy) improved as the stimulus number increased. This difference probably reflects the spatial resolution of our procedure (there were only 12 radial segments defined compared with 23 longitudinal categories), which could mask any dependency of radial accuracy on the number of preceding stimuli. There is no evidence for a difference in absolute spatial accuracy in orthogonal planes. Although the increased accuracy with repeated stimuli could represent a form of spatio-temporal integration (‘learning’ the location by successive estimations), it could also arise from a progressive decrease in mechanical performance.

The strikes are less accurate when the caterpillar responds to stimuli that are more dorsal or anterior. Because of the mechanical limitations mentioned earlier, we expected a decrease in precision for these areas. However, mechanical limitations are not the only explanation because the caterpillar can reach most locations on its body but strikes are made with a posterior–ventral offset. For example, strikes can land near the spiracles in A5 and A6 when the caterpillar is responding to stimuli applied to dorsal locations in A4, but they fall posterior and ventral when stimuli are applied near the A5 and A6 spiracles themselves (see Fig. 4). One possible explanation for these offsets and changes in strike accuracy could be variations in the accuracy of the mechanosensory information (touch and proprioception) along the body. This can be examined by measuring the sensory fields of the mechanosensors by directly recoding their spike activity.

We predicted that accuracy would decline with increased strike speed (i.e. shorter strike duration). This relationship was only observed for radial accuracy in A4. For TP, it appears that slower strikes are instead less accurate; however, this effect was not significant and probably was much influenced by one observation that had a relatively long duration. The duration of strikes can be explained by the stimulation site for radial accuracy in A4 and TP, and longitudinal accuracy according to our statistical tests. However, judging from our results, the influence of stimulation site (represented by the color of the data points in Fig. 5) on duration does not seem to be very large and might not be biologically relevant.

The strike behavior in caterpillars has strong parallels with aimed movements studied in other animals, in particular with scratching movements in locusts (Matheson, 1997), frogs (Giszter et al., 1989), turtles (Mortin et al., 1985) and cats (Deliagnina et al., 1975). In all cases, the animal needs to identify the location of the stimulus on the body (sensory and proprioceptive information) and navigate the scratching limb towards this location (motor commands and proprioceptive information). The caterpillar, however, moves a relatively large part of its body towards the stimulus location. In addition, to our knowledge there are no studies describing movements aimed at a location on the body in soft-bodied animals. In the caterpillar we see an offset between the stimulus location and the location where the mandibles land, and the size of the offset is dependent on the location of the stimulus. None of the studies mentioned above reports an off-set between the stimulus location and the location where the limb scratches. Like the caterpillars, the species mentioned above also show repeated movements towards the stimulation site, but because the scratching is by a limb, the repetition is often achieved by only moving the most distal part of the limb and therefore is not directly comparable to the movement that a caterpillar makes with its head and thorax.

Although mechanical and sensory constraints may explain some of the features of the strike, the behavior overall is probably determined by its primary role, defense against natural enemies. Manduca sexta's main enemies as a caterpillar are wasps and birds.

Rab noted that 6–10% of caterpillars in a tobacco field were killed by Polistes fuscatus, a paper wasp 15–20 mm in size (Rabb, 1950; Rabb, 1953). First instar caterpillars were eaten at site, second or third instars were sometimes partially brought back to the wasp nest. At times, P. fuscatus would kill larger caterpillars by critically wounding them; in those cases, the dead caterpillar was only partially eaten. Second and third instar Manduca are also attacked by parasitoid wasps of which the most prevalent is the braconid wasp C. congregatus. These 3 mm wasps inject eggs into the caterpillars. There is some variation in how long it takes the wasps to lay eggs: according to Fulton, and Beckage and Riddiford, it takes 2–3 s to complete oviposition (Fulton, 1940; Beckage and Riddiford, 1978); however, Gilmore lists 20–30 s, but this includes multiple oviposition attemps (Gilmore, 1938). Nonetheless, the shorter the duration of oviposition, the fewer eggs are injected. A quick removal of the parasitoid wasp thus increases the survival chances of the caterpillar.

Cotesia congregatus could increase its success rate by injecting the eggs mainly in the more anterior body segments. However, C. congregatus does not seem to have a preference for more anterior or posterior segments, although attempts at oviposition in more anterior segments do seem to be more successful (C. I. Miles, personal communication). In contrast, Gilmore mentions that oviposition ‘was principally within the posterior segments of the host’ (Gilmore, 1938). We found that accuracy was higher when a caterpillar received a stimulus in the posterior segments; if Gilmore's finding is correct, this higher accuracy would result in more wasps being removed by the caterpillar. Indeed, Gilmore also notes that ‘attempted oviposition in the anterior portion of the host frequently resulted in the parasite being brushed off or killed before the act was completed’ (Gilmore, 1938). Although this is opposite to what one would expect based on our findings, we have observed the caterpillar hitting the anterior part of its body against the substrate during thrashing, which could serve to dislodge a wasp.

There is no information about a possible preference by C. congregatus for dorsal or ventral locations on the caterpillar's body. However, Manduca caterpillars tend to conform closely to the contours of their food source and are typically found on the underside of leaves. They are often shaded by leaves and by other foliage such that their dorsal surface is the most exposed part of the abdomen. It is likely that parasitic wasps have better access to the dorsal surface than to the ventral surface of the caterpillar's body.

Oviposition by C. congregatus takes place when caterpillars are in their second or third instar, but oviposition can also occur in the larger fourth and even the first few hours of the fifth instar; the later fifth instar cuticle is too thick to be penetrated for oviposition (Beckage and Riddiford, 1978). Our observations were made on the third day of the fifth instar; however, we have no reason to believe that the caterpillar's behavior has dramatically changed in those few hours. In addition, there are also parasitoids that oviposit in M. sexta in the fifth instar, such as the tachniod fly Winthemia manducae (DeLoach and Rabb, 1971). This fly lays eggs specifically in the fifth instar so the adults hatch when the caterpillar has buried itself in the soil and has turned into a pupa.

Although our studies indicate that the caterpillar's strike accuracy varies depending on the location of a stimulus on its body, it remains difficult to draw any conclusions about the consequences of this variation on the caterpillar's ability to ward off predators and parasites. Early studies indicate that in unsprayed tobacco fields there is high predation particularly in the early instars (Lawson, 1959). Although it is clear that strike behavior could limit the success rate of parasitoid wasps or larger enemies like birds (Walters et al., 2001), we do not know how many losses are prevented by the striking behavior and how this varies with attacks aimed at anterior or posterior segments. If enemies do have a preference, how much of that is influenced by the red horn on the caterpillar's abdomen?

Additionally, a final strike position ventral and posterior to the site of stimulation doesn't necessarily mean that the caterpillar missed its target. Any ecologically relevant target will be three-dimensional, and the caterpillar would presumably make contact with the target before contacting its own body wall. In fact, because the caterpillar appears to slow down before the end of the strike, it may be better to overshoot the source of the stimulation to increase the impact of the strike on the target. Kinematics studies of the movement could be used to follow the entire trajectory path and corresponding velocities.

Manduca uses the strike behavior for defense, but a similar behavior has been seen in the carnivorous caterpillars of Hawaii (Montgomery, 1982; Montgomery, 1983). These caterpillars catch and eat flies. To catch flies, these caterpillars imitate twigs by holding onto the substrate with several of the most posterior body segments and lifting the rest of the body away from the substrate. When a fly touches the posterior body segments, the caterpillar turns around and grabs the fly with its thoracic legs and mandibles, a movement very similar to Manduca's strike behavior. More research into this hunting behavior might give an idea of how widespread the strike behavior is in Lepidoptera.

We would like to thank L. M. Reed for help with statistical analysis and C. I. Miles and J. J. Yun for helpful discussion on Cotesia congregates.