The substrate as a skeleton: ground reaction forces from a soft-bodied legged animal

Articulated (stiff, jointed) and hydrostatic (soft wall, pressurized) skeletons (Chapman, 1958) are the two primary structural models used to characterize and explain the biomechanics of animal locomotion. In an articulated system, movements can be described precisely by the center of mass position and joint angles (Colobert et al., 2006; Holmes et al., 2006). Muscles work antagonistically around a joint to produce directionally constrained motions and structural levers allow the mechanical exchange of force and displacement.

A similar process occurs in many soft-bodied animals that exploit incompressible body-fluid to transmit forces and to actuate body parts (Kier, 1992; Niebur and Erdos, 1991; Quillin, 1998; Skierczynski et al., 1996; Wadepuhl and Beyn, 1989). For example, most annelids use circumferential muscles that operate antagonistically with longitudinal muscles to control extension and shortening (Kristan et al., 2005; Quillin, 1999). These animals produce different postures using a temporary stiff hydrostatic skeleton. Even animals such as the octopus, which does not have an open fluid-filled cavity, use the viscous, semi-fluid properties of connective tissues and muscles to control movements (muscular hydrostats) (Kier and Stella, 2007).

Most current models of hydrostatic skeletons make the general assumption of volume conservation and use geometric approaches to predict movements (Skierczynski et al., 1996; Smolianinov and Mazurov, 1976; Wadepuhl and Beyn, 1989), sometimes employing tissue material properties in the simulations (Chiel et al., 1992; Herrel et al., 2002; Kier and Smith, 1985; Nishikawa et al., 1999). Although these modeling approaches can be used to estimate mechanical interactions, direct measurements of locomotor forces by soft-bodied animals remain scarce, and highly constrained by technical limitations (Keudel and Schrader, 1999; McKenzie and Dexter, 1988a; McKenzie and Dexter, 1988b; Quillin, 2000).

One group of animals that makes an attractive model system for studying such locomotor forces is the larval stage of the Lepidoptera (i.e. caterpillars) because most caterpillars interact with the environment via discrete contacts (prolegs). By measuring the substrate reaction forces of each proleg simultaneously, it should be possible to determine how body deformation translates into locomotor dynamics. Caterpillars are also interesting from biomechanical and neurobiological perspectives because they are pressurized soft cylinders capable of climbing in complex three-dimensional environments. This is accomplished with only longitudinal muscles and a few short oblique muscles, making it puzzling how they manage to co-ordinate shortening and extension using a non-septate fluid and tissue-filled body cavity.

Caterpillars usually employ the abdominal prolegs for attachment to the substrate and produce anterograde waves of movement, but there is considerable species diversity in the arrangement of these appendages and in their use during locomotion (Miller et al., 2006; Snodgrass, 1993). Four pairs of abdominal prolegs (on abdominal segments 3 to 6; A3–A6) and a pair of terminal (anal) prolegs (on abdominal segment 10; TP) are thought to be the ancestral and/or dominant form for lepidopteron (as labeled in Fig. 2) (Forbes, 1910; Hinton, 1955) but prolegs in different species can be lost from any segment (Wagner, 2005). For example, most caterpillars in the family Notodontidae have modified terminal prolegs for purposes other than locomotion and in the family Noctuidae both anterior and posterior prolegs are reduced. The subfamily Hypeninae have lost functional A3 prolegs while the terminal prolegs are elongated, and members of the subfamily Catocalinae have even lost the A4 prolegs. Species such as Parallelia bistriaris (maple looper) and Caenurgina crassiuscula (clover looper) do not have prolegs on segments A3 or A4 and they move with looping gaits. Members of the subfamily Plusiinae (cabbage looper, soybeam looper, celery looper) move by ‘inching’ with prolegs on A5, A6, and the terminal segments. Of course, the inchworms (Geometridae) all perform inching locomotion with only the A6 and terminal prolegs. Some lepidopteran larvae do not need prolegs for locomotion and have reduced them into a continuous ventrum (slug caterpillars) (Epstein, 1996; Rubinoff and Haines, 2005) or transformed posterior legs into a hydraulic warning display (family Notodontidae). Case-bearer caterpillars only employ the six thoracic legs for locomotion. On top of this diversity, there is a large range of body size and cuticular adornments across lepidopteran species (Miller et al., 2006). It is therefore very important to recognize the body morphology and modes of locomotion. In this study, we only focused on large caterpillars with the ancestral body plan such as Manduca, which might inform us about some general strategies for locomotory movements.

The tobacco hornworm (Manduca sexta L.) is a well-studied model system in neurobiology, physiology and ecology. Because their muscle anatomy is known so well (Barth, 1937; Eaton, 1988; Kopec, 1919; Libby, 1959; Snodgrass, 1961; Snodgrass, 1993) and each muscle is generally innervated by a single motoneuron (Levine and Truman, 1985; Taylor and Truman, 1974; Weeks and Truman, 1984) these caterpillars show great promise for electromyographic studies of soft-bodied locomotion (Belanger and Trimmer, 2000; Dominick and Truman, 1986; Johnston and Levine, 1996a; Johnston and Levine, 1996b; Mezoff et al., 2004; Simon and Trimmer, 2009). Manduca is a relatively large insect (which helps force measurements) using four pairs of abdominal prolegs and a pair of terminal prolegs to generate most of its motion. A crawl typically consists of anterograde waves of body muscle contractions with the prolegs releasing their grip (Belanger et al., 2000; Mezoff et al., 2004) to be carried forward for re-attachment. Several segments are in swing phase simultaneously and the wave of motion usually continues into the thorax and head as the next cycle begins from the terminal prolegs (Trimmer and Issberner, 2007). Manduca caterpillars weighing 2 g move at an average speed of 3.10±0.24 mm s⁻¹ during horizontal locomotion with low acceleration and hence very small center of mass dynamics (van Griethuijsen and Trimmer, 2009).

In preliminary experiments it was found that Manduca are able to crawl horizontally even when the thoracic legs are removed. Apart from the unsupported thorax sliding across the substrate and slower movements, these crawls appeared normal with appropriately sequenced steps in the abdomen. Another interesting finding is that the hemocoele pressure of restrained (Mezoff et al., 2004) and crawling Manduca (H.T.L., unpublished data) is not isobarometric nor do the fluid pressure changes correlate well with movements. This is supported by phase-contrast synchrotron X-ray videography showing that internal tissues such as the gut, hemolymph and trachea are in constant motion (M. A. Simon, personal communication), presumably generating, and responding to, pressure gradients. During crawling, the gut movements in particular are quite large; this means we should be very careful interpreting static loads from inertial forces. In addition, the internal air cavity constitutes between 3 and 10% of body volume and air can be expelled during underwater locomotion, suggesting that the internal volume is not necessarily constant (H.T.L., unpublished preliminary data).

These prior observations suggest that caterpillar locomotion is biomechanically distinct from the current models of soft-bodied terrestrial crawling. In this study, the central questions we wanted to address were: (1) during crawling, how does a caterpillar extend its body without circumferential muscles?; (2) within a crawl cycle, how much weight shift occurs, and what are the roles of different prolegs?; (3) given that the body is not necessarily incompressible, how does Manduca transmit forces for posture support and propulsion?

A major finding is that caterpillars use the substrate to transmit forces from each contact point to other parts of the body. Instead of contact points propelling the center of mass forward, we propose that soft animals can employ an ‘environmental skeleton’. In this view, posture generated by stiff hydrostatic control has a smaller secondary role in locomotion.

Reaction forces are usually measured with a force plate larger than the animal's total contact area (Goldman et al., 2006; Heglund, 1981; Zumwalt et al., 2006) but Manduca moves so slowly that it exerts very weak inertial forces (forces associated with acceleration). In other words, the overall acceleration of the caterpillar does not adequately reflect the size or direction of local ground reaction forces, some of which cancel out when measured collectively. Furthermore, because caterpillars' bodies deform massively between contacts it is not desirable to measure forces on one leg at a time as can be done for animals with stiffer skeletons (Blickhan et al., 2007; Dutto et al., 2004; Roberts and Belliveau, 2005). Instead, reaction forces from each proleg should be measured independently and simultaneously. Recent breakthrough in developing micro force sensor units has made insect locomotor dynamics much more accessible (Bartsch et al., 2007; Lemmerhirt et al., 2006; Reinhardt et al., 2009). However, these microelectromechanical fabricated or precision-machined sensors are very expensive and require specialized maintenance. For our application, we constructed a robust, low cost, sensory array to resolve forces from each contact point simultaneously. Our observations of Manduca with amputated thoracic legs suggested that the prolegs are sufficient for crawling. Therefore, the number and spacing of sensors was arranged to optimally capture data from the prolegs. Reaction forces from the thoracic legs can occasionally be resolved with this array (usually a single combined thoracic force, rarely as separate segments) so they are reported where available. The sensor array was paved over most of the caterpillar length so that the center of mass and weight redistribution could be tracked during locomotion.

The device consisted of an array of cantilevered strain-gage based force beams each with four sensors arranged to detect both normal forces (i.e. perpendicular to the crawl, aligned with gravity in horizontal crawling; nGRF) and axial forces (i.e. fore–aft forces along the axis of a crawl; fGRF; Fig. 1A). The beams were arranged with a pitch of 3.81 mm per sensor, which is less than half the caterpillar average stride length (8.52±0.22 mm) (van Griethuijsen and Trimmer, 2009). Preliminary data showed that each pair of prolegs takes up to one-third of the body weight (<7 mN for a 2 g fifth instar) and three times the body weight in the fore–aft direction (<60 mN). A custom built control system provided AC excitation, analog amplification and basic signal processing. A Basler 206A high speed camera (Basler, Ahrensburg, Germany) was used for real-time two-dimensional (2-D) kinematics tracking and a Sony SSC S20 color camera (SONY, Tokyo, Japan) for detailed behavioral observation. Data collection was controlled by a custom program running in LabView 8.5 (National Instrument, Austin, TX, USA; Fig. 1B). The program took images from the kinematics camera and identified markers on the animal, transformed the coordinates according to the location of the substrate and marked the touchdown of prolegs in a force data spreadsheet. The GRF data were displayed in real-time to help monitor experimental progress. The program saved five data files for every experiment. This included a reference recording for noise and offset analysis, the raw force data, kinematics raw data, processed force data with kinematics markings, and the synchronized behavioral video. Because this device is unique and suitable for use with a variety of species it will be described in more technical detail in a separate publication.

Healthy second day fifth instar Manduca sexta caterpillars were collected from our colony, with a body mass of between 1.5 and 3 g, corresponding to body length between 26.5 mm and 45.3 mm (based on a body dimension statistics for our colony, N=25). All animals were reared on an artificial diet and kept in a 17 h:7 h L:D cycle at 27°C. The colony maintenance and rearing protocol were as described by Bell and Joachim (Bell and Joachim, 1976). Animal subjects were weighed, sexed, and labeled before being placed in a large transparent container (conditioning chamber) for 20 min of free roaming. Eight animals were used in this study.

During the development of the sensor device data was collected from two sets of beam arrays. A 1-D array consisting of 12 unidirectional force beams was use to collect weight shift information (range=45.72 mm, or ~100% resting length of a fifth instar Manduca). A second array consisting of five two degrees of freedom (2-D) force beams (range=19.05 mm, or ~40% resting length) was used for force correlation analysis. Prolegs were tracked using a bead (0.50±0.05 mm diameter) coated with clear red UV fluorescent paint (Risk Reactor, Dallas, OR, USA) placed on each proleg on one side of the animal (Fig. 1B). The wet paint allowed the red fluorescent beads to adhere to the proleg cuticle. A Black-Ray BL-15 UV lamp (UVP Inc., San Gabriel, CA, USA) was installed above the 2-D array to excite the red fluorescent paint.

To encourage crawling, animals were lined up on a linear elevated wooden substrate (standby track) leading to the force beam array. Caterpillars were found to move forward when naturally probed by another conspecific. On the 1-D beam array, data logging was started when the animal entered the buffer zone and stopped when either the animal stopped crawling or had covered the entire force bean array. For the shorter 2-D array, data logging did not begin until the thoracic legs triggered one of the active beams. After each trial, the animal was put back into the conditioning chamber for at least 3 min before it was returned to the standby track for another trial. All the data presented in this paper were from Manduca crawling horizontally in a straight-line, thus the normal GRFs (nGRFs) represent the body weight loading. A successful recording was one in which the caterpillar traveled continuously over the entire sensing zone. Since Manduca caterpillars tend to pause unpredictably, most animals produced only one successful trial during the experimental session. To be consistent, we analyzed only one successful trial from each of the eight experimental animals, resulting in five sets of nGRF data and three sets of 2-D (nGRF and fGRF) data.

Because the caterpillars moved slowly, force data were sampled at 100 Hz, kinematics image analysis at 10 Hz, and behavioral videos at 30 frames per second. In preliminary experiments (nGRF recorded at 1 kHz) it was found that stepping forces changed over about 0.5 s and that most high frequency signals were due to electromagnetic noise. Occasional drift due to thermal fluctuation was too slow to interfere with the force recordings. The raw data was filtered using an RC type low-pass filter at 2 Hz and then smoothed with a second order polynomial over every 0.1 s. These curves were compared with the raw data to check consistency for every recording. Signal offsets were controlled under 6 mV and subtracted out using the reference recording. Raw voltages were converted to forces using calibration curves made for each beam.

The resulting GRF data were normalized to the individual caterpillar body weight (BW). The normalized data were then checked with the behavioral videos for single steps that occasionally occurred across two adjacent force beams and these readings were summed. Crawls with several such overlapping steps were rated as unsuccessful and removed from the analysis.

Video tracking provided the first indication of proleg touchdown timing and location. Individual proleg GRFs were then extracted by identifying the points at which stance occurred and by using threshold detection (normal load crosses 1% of the animal's body mass). The 2-D array captured at least one step per proleg in each trial. We therefore compiled at least one complete crawl cycle for each experimental animal for qualitative comparison. Although contact by a single pair of thoracic legs on the 2-D force beam was rare, some occurrences were identified from the videos. Examples of these 2-D GRFs have been included for comparison with the proleg GRFs.

Using the extensive kinematics data collected from our research group (von Griethuijsen and Trimmer, 2009; Trimmer and Issberner, 2007), the 2-D GRF data was scaled to the Manduca standard kinematics template representing the average step length and relative timing of proleg movements during a full crawl cycle (Fig. 2). The timing of the thoracic leg contacts are also noted on this template. A crawl cycle was defined as from the beginning of terminal prolegs swing phase to the beginning of A3 stance phase. We took the average stance phase time over a complete crawl time as the duty cycle of that particular proleg. By scaling all the corresponding GRF data to this template, we could analyze force interactions between different prolegs for different crawl cycles and animals. To match up thoracic legs in a crawl cycle, we analyzed the videos frame by frame to place the GRF data of the sample thoracic legs into the appropriated time points of a crawl cycle (as indicated in Fig. 2).

Manduca caterpillars do not change length more than 5% during normal locomotion, so we could evaluate the overall weight shift by summing nGRFs from all contacts (including those from thoracic legs). This procedure produced an oscillating trace with each oscillation corresponding to one crawl cycle. Because inertial forces were so small these oscillations directly reflected the weight placements of the caterpillar. By measuring the loading change over each cycle, we could estimate how much mass was shifted forward and backward in one average step length (8.52 mm) per cycle period. Instead of tracking the center of mass, we assumed that only part of the body was accelerated forward in each crawl and that mechanical work associated with this weight shift could be evaluated.

During a crawl, a sequence of ground reaction forces was detected on each of the sensor beams. Frame-by frame analysis of the synchronized video allowed these forces to be attributed to different segmental appendages (Fig. 3A). Although the features of these forces varied in magnitude and duration each crawl, there were consistent differences in the nGRF profiles of each segment (Fig. 3B).

For the mid body segments (A4–A6) the ground impact and leg detachment were visible as two peaks in the nGRF data which overshot the weight-baring portion of the proleg stance phase. Behavioral observations suggested that these two peaks correspond to crochet protraction and retraction; both attachment and detachment involved momentary extension of the prolegs before the crochets grip or release. This may explain why there was no negative normal load in the recording of horizontal crawling. The initial loading and unloading rates were similar for each proleg in segments A3–A6 but differed for the terminal prolegs (TPs). After a more gradual loading phase, the TPs immediately started to unload as the animal shifted its weight forward. The vertical reaction forces due to crochets were less pronounced for the TPs, but the weight-supporting section appeared highly variable just like the A3.

Overall weight shift can be observed from the net nGRF across all force beam channels. As the animal entered the sensor zone, forward and backward weight shifts appeared as nGRF fluctuation in each crawl cycle (Fig. 4A). When the animal was completely supported in the sensor zone the net nGRF was stable and equivalent to the weight of the animal (e.g. for Fig. 4 the average nGRF=16.94 mN, animal mass=1.72±0.01 g, or weight 16.9 mN). The 1-D array was accurate to within 10 μN for most recordings.

Although complete tracking of the center of mass throughout a crawl required a force beam array longer than the current device (one body length plus two additional steps), it was possible to perform a preliminary assessment of the overall weight shift during each step. The peak-to-peak net nGRF changes were measured for about 10 steps in each of five animals, standardized to the crawl cycle and body mass, and combined to show the overall weight shift in a crawl (Fig. 4B). During the first 64.5% of a step, Manduca shifted 30% of its body mass forward and then sharply arrested the transfer. Then it moved 15% of body mass back again in the last 35.5% of the step.

Although the sensor array was optimized for proleg reaction forces it was occasionally possible to detect GRFs from the thoracic legs collectively (Fig. 3) or segmentally (Fig. 5). From this sparse data it was observed that: (1) in all the thoracic GRF fragments, fGRF is almost always positive; (2) the nGRF results suggest that each thoracic leg supports as much body weight as each proleg; (3) force profiles for each thoracic leg overlap for much of the crawl cycle. Although the stiff articulated thoracic legs could withstand compressive forces they showed no indication of decelerating fGRFs (Fig. 5A), which suggests that the posterior segments do not push the thorax while it is in stance. The timing of the thoracic maximum pull also coincides with periods of proleg ‘drag’ and this is presumably responsible for body stretching (at least from T1–A3) during locomotion. There is some redundancy in the mechanism of body extension since abdomen extension (A3–A10) can be achieved using proleg interactions alone although it may be less effective. Further examination of this unexpected role for the thoracic legs will require modification to the senor array in future studies.

The reaction forces in the direction of locomotion (fGRF) are almost one magnitude higher than the normal loads (Fig. 5A). There are typically two negative impulses (decelerating forces) one at the start of stance and the other before the onset of swing, associated with proleg attachment and detachment. Terminal prolegs initiate each cycle with a decelerating impact and then rock briefly into a neutral state at about 20% of the stance phase. They then drag (negative fGRF) for most of the stance phase. A6 prolegs also decelerate during the terminal prolegs dragging phase, but they pivot into a positive pull as the terminal prolegs release from the substrate. Successive prolegs in segments A5–A3 share this force profile but the magnitude of the mid-stance fGRF is progressively less negative in more anterior segments and is generally positive (i.e. a pulling force) in segments A4 and A3.

For analytical purposes, we divided a complete crawl into two stages and six phases (Fig. 5B) as follows.

In the beginning of a crawl cycle, anterior prolegs are detached and pulled forward. At the same time, the thoracic legs starts to walk the thorax forward, stretching the frontal abdomen in the process (preliminary GRF data for thoracic legs). This initiates the overall mass shift.

The release of A6 immediately leads to more drag on the A5, indicating that the body tension anterior to A5 is initially higher than that between A6 and A5. This is also when the caterpillar reaches the shortest body length.

With the detachment of A5, the nGRF in A4 declines and that in A3 increases. This demonstrates that Manduca can lift its abdomen with only the A3 proleg and thoracic legs attached. Thoracic legs begin a new crawl cycle during this phase.

A4 detachment releases the muscle strain energy between A3 and A4, but this contraction is soon anchored by the TP. This initiates the stretching stage where anterior muscles stretch the posterior muscles between two bridging anchors.

A6 attaches to the substrate to hold the stretch and propagates this antagonist stretching forward. By now the caterpillar has returned to the original length, using the thoracic legs to maintain it. Body mass starts to rebound backward, probably through movements of internal organs and fluid.

The last phase of a crawl involves stretching the anterior abdomen back to its resting length using the thorax. Most abdominal muscles must do negative work in the process.

From detailed observations (Snodgrass, 1961) and kinematic analyses (Belanger and Trimmer, 2000; Mezoff et al., 2004; Trimmer and Issberner, 2007; van Griethuijsen and Trimmer, 2009) we proposed that caterpillars such as Manduca do not use the prolegs as actuated propulsive limbs, but instead use them as support and to generate controllable grip. For the first time it has now been possible to measure the GRFs during Manduca crawling and to show that the prolegs are indeed anchors rather than levers. An analysis of the GRF profiles and segmental timing reveals several new and unexpected details.

Most vertebrates exploit their stiff skeleton and joints to transfer muscle forces directly into limb displacements. They can apply large compressive forces to the skeleton and typically use their hind legs to push off and the front legs to decelerate (Biewener, 2003). In contrast, the caterpillar prolegs are pouches of soft cuticle on the abdomen with no discrete articulation or structures for gaining a lever advantage. Moreover, because stress (and indirectly, stiffness) in the body wall is largely a function of hydrostatic pressure and curvature (Mezoff et al., 2004; Wainwright, 1988), it is impossible for the small radius prolegs to be stiffer than the body itself without additional chitin cross-linking or other structural changes. This means that prolegs are relatively soft and unsuitable for resisting large compressive loads. This is borne out by the fGRF results in which caterpillars mostly load their prolegs in tension and only momentarily experience compressive fGRF when TP and A6 ‘rock’ the body mass over them. Caterpillars use the anterior prolegs to pull and the posterior prolegs to drag, effectively reversing the roles seen in most legged animals (Fanchon et al., 2006; Schmidt, 2005). Although the nGRFs during horizontal crawling are positive and therefore compress the prolegs (weight-bearing), these forces are an order of magnitude smaller than fGRFs and easily supported by baseline body pressure.

It is clear from the fGRFs that the substrate takes the compressive load during multi-legged stance phases. This was verified using a simple behavioral test. Manduca was suspended from the head capsule and the dorsal horn and curved substrates of different stiffness but similar curvature (caterpillars prefer rode-like substrates) were placed between the prolegs as the caterpillar tried to crawl. As expected, relatively stiff substrates such as balsa wood were translocated backwards by the anterograde segment movements. However, as the substrate became softer these movements were less effective. When presented with a soft wire or a silk braid that could only take tensile loads, Manduca was unable to extend and its locomotory movements failed to move the wire progressively (Fig. 6). The substrate is therefore an essential component of this mechanical system, and Manduca is incapable of normal locomotion on surfaces that are softer and more compliant than its own body. Presumably, body stiffness is tunable within a narrow range through changes in muscle tone and internal pressure. This aspect of caterpillar locomotion could play an interesting ecological role since plant stiffness will affect the mobility of different caterpillar species. Unfortunately, we are not aware of any data correlating host plant foliage flexural stiffness to the mechanics of caterpillars feeding on them. We are therefore developing a numerical model of caterpillar mechanics based on tissue material properties (Lin et al., 2009), morphology and pressure (H.T.L., unpublished data). We intend to check this model against some ecological and morphological data in the near future.

Although Manduca is capable of extending its body using hydrostatic turgor (e.g. exploratory reaching/casting behavior) during horizontal crawling on a (relatively) stiff substrate, it instead uses the substrate to stretch out. In this process, the thorax walks forward but does not bring the heavy abdomen with it, thereby stretching the anterior segments. Abdominal segments shorten partly by releasing muscle strain energy between adjacent prolegs. To stretch the abdomen back out, the caterpillar anchors its two ends and pulls the posterior muscles with anterior muscles. Forces in the axial direction are responsible for extending the body and restoring muscle length which explains why fGRFs are much larger than the nGRFs. This is similar to the antagonist muscle stretching in vertebrates or stiff skeleton arthropods and it may convey several biomechanical advantages over a hydrostatic skeleton. First, by using the substrate as a skeleton, Manduca avoids the need to strongly pressurize its fluid filled body; the body can be just stiff enough for self support. Second, by conforming to the substrate directly there is no need to maintain posture through local pressure control; the body does not need to be compartmentalized and the demands for neural-computation and proprioception can be minimized. Third, by avoiding hydrostatic control of locomotion, internal organ systems that interface with the exterior (e.g. gas exchange and digestion) can be decoupled from movements (in contrast to the hydrostatic limb extension system of many spiders that makes fast running and breathing incompatible (Parry and Brown, 1959; Prestwich, 1988). These physiological advantages could also be evolutionary constraints that have contributed to the unique locomotion of caterpillars in general.

From lepidopteron homeotic gene studies (Suzuki and Palopoli, 2001; Warren et al., 1994), comparative anatomy (Hinton, 1955) and X-ray induced mutation (M. A. Simon, personal communication), we know that caterpillars are capable of generating prolegs in any abdominal segments. It is therefore interesting that some caterpillars share the same body plans whereas others differ so much. The new findings on caterpillar GRFs may provide insights into some of the biomechanical constraints favoring the evolution of different proleg configurations.

According to the GRF analysis in this study, prolegs anchor a caterpillar for body deformation. The posterior legs are responsible for holding the rear end while the anterior muscles stretch the abdominal muscles. During this process, the substrate takes on compressive loads and transmits the necessary forces across the body, analogous to a rigid bone in a skeletal system. This strategy is methodologically different from that of hydrostatic and articulated skeletons. Since the animal does not carry its own skeleton and simply adopts whatever it sits on, we call it an ‘environmental skeleton’. In a typical caterpillar body plan, the lack of prolegs in A7–A9 provides clearance for the terminal prolegs (TP) to swing forward and thus limits of the step size. Manduca's center of mass is in A4, so the adjacent prolegs A3–A5 are critical for securing the body. Species that have a reduced number of prolegs in these segments may have an increased dependence on a hydrostatic skeleton. For example, we predict that caterpillars with looping gaits (and few proleg attachments) will have biomechanics dominated by a hydrostatic skeleton. However, in both Manduca casting behavior and inchworm looping locomotion, the A6 prolegs work closely with the terminal prolegs to produce leverage via the hydrostatic skeleton.

It is interesting to speculate on the evolutionary transition from ancestral crawling that relies so much on the substrate, to looping gaits. Since the cost of maintaining a hydrostatic skeleton is determined by the pressure required to inflate the body it is conceivable that as the caterpillars adapted to narrower environmental niches, their body size decreased and body stiffness increased. Under these conditions, the use of a hydrostatic skeleton may be more favorable than the environmental skeleton. The findings reported here provide a new basis for exploring the relatively neglected role of locomotory mechanics in the evolutionary ecology of plant–insect interactions.

We would like to thank Caroline Blavet for collecting some nGRF data and Linnea van Griethuijsen for providing assistance in constructing the Manduca standard kinematics template. The Manduca cost of transport data was provided by Dr William Woods. This project was funded by NSF grant IOS 0718537 to B.A.T.