Adhesive latching and legless leaping in small, worm-like insect larvae

Jumping and hopping are most often achieved with legs – either through the lever-based launches of elongate legs (Astley and Roberts, 2014; Alexander and Vernon, 1975; Zajac et al., 1981; Moore et al., 2017) or through elastically powered short legs (Burrows et al., 2008; Burrows and Sutton, 2008; Bennet-Clark, 1975; Bennet-Clark and Lucey, 1967; Henry et al., 2005; Aerts, 1998). Nevertheless, jumping mechanisms that do not involve legs have evolved in a variety of animals, such as the body-segment torque of click beetles (Evans, 1972, 1973; Ribak and Weihs, 2011), hydrostatic pressurization of a tail-like appendage in springtails (Brackenbury and Hunt, 1993), and terrestrial tail flips in bony fishes (Ashley-Ross et al., 2014). A particularly intriguing jumping strategy is employed by some small, worm-like animals that use latching, hydrostatic manipulation and elastic mechanisms to launch themselves into the air. This strategy is employed by larvae of certain groups of dipterans (Camazine, 1986; Suenaga et al., 1992; Maitland, 1992; Bonduriansky, 2002; Marinov et al., 2015; Manier and Deamer, 2014) and by nematodes (Campbell and Kaya, 1999a,b). Analogous legless-jumping systems have been discovered several times, yet a basic framework for the biomechanics of worm-like jumpers is needed in order to establish general design principles and interpret the diversity and ecology of these systems.

Legless jumpers can achieve impressive jump performance that rivals that of small, legged jumpers. The jumping flea, a famous small, legged jumper, can jump with speeds of 1.3 m s⁻¹ (Sutton and Burrows, 2011), and leafhoppers can accomplish speeds of 1.1 to 2.5 m s⁻¹ (Burrows and Sutton, 2008), metrics similar to the 1.98 m s⁻¹ takeoff speed of legless fruit fly larvae (Ceratitis capitata; Maitland, 1992). Likewise, the click beetle, a hard-bodied legless jumper, can achieve takeoff speeds of 2.4 m s⁻¹ (Evans, 1973). Soft-bodied legless jumpers, such as larvae of the piophilid fly Prochyliza xanthostoma, have been reported to jump distances 28 times their body length; assuming a takeoff angle of 45 deg, this suggests takeoff speeds of approximately 1 m s⁻¹ (Bonduriansky, 2002). Even with the documentation of these intriguing legless systems, most analyses have been limited by imaging technology such that the kinematic capabilities of these systems are uncertain; the fastest recorded frame rate, to our knowledge, of jumping soft-bodied, legless organisms is 54 frames s⁻¹ (Camazine, 1986). Recent studies of the hard-bodied click beetle system offered a significant improvement in modeling and energetic calculations through improvements in imaging (Ribak and Weihs, 2011; Ribak et al., 2012).

Small-legged jumping and legless-jumping systems are often powered by elastic mechanisms, typically through latching, elastic loading and release (Longo et al., 2019; Ilton et al., 2018; Sakes et al., 2016; Gronenberg, 1996; Patek, 2015; Patek et al., 2011). Worm-like, legless jumpers typically form a loop with their body by latching their anterior and posterior in place and then springing into the air when the latch is released (Fig. 1). Elastic energy is stored by stretching body surfaces and pressurizing body fluids, much like ‘a bent, sausage-shaped balloon’ (p. 160, Maitland, 1992). In a study of fruit fly larvae (Ceratitis capitata), Maitland (1992) determined that the stored elastic strain energy (31 J kg⁻¹) of the exoskeleton, when experimentally pressurized to 82 mmHg and a strain of 29%, provides just enough energy to launch a jump.

Latching mechanisms are diverse in small, legless systems. For example, gall midge larvae (Tricholaba barnes) wedge either the head or anal segments between ventral segments and then load the body into a taut loop (Milne, 1961). Piophilid flies and several species of fruit flies (Tephritidae) reportedly place their mouthparts into the integument on the posterior–ventral side of their body and elastically load their body in a loop (Bonduriansky, 2002; Marinov et al., 2015; Maitland, 1992). In addition, Camazine (1986) suggested that fungus gnat larvae (Mycetophila cingulum) latch the head and tail together using protolegs or ‘pegs’. Finally, microscopic entomopathogenic nematodes (Steinernema carpocapsae) use surface tension of water coating the body to adhere the head and tail to form a body loop before jumping (Campbell and Kaya, 1999a,b; Reed and Wallace, 1965). Despite the diversity of legless-jumping mechanisms that have been observed, surprisingly little is known about the morphology and mechanics of the latches that are central to successful jumping.

To elucidate the jumping mechanisms of small, worm-like animals, we studied larvae of an undescribed species of gall midge (Asphondylia sp.) that jump when removed from the galls they form on goldenrod plants. We employed high-speed imaging, scanning electron microscopy and mathematical analyses to address the following goals: (1) analyze jump preparation, launch and takeoff dynamics, horizontal travel distance and jump energetics; and (2) establish the morphology and mechanics of latching. We place these specific findings in the broader context of a nascent and phylogenetically diverse field of small, hydrostatic jumpers.

Larvae of Asphondylia sp. (Diptera: Cecidomyiidae) (Fig. 2A) induce multi-chambered rosette galls in the stem apices of several species of goldenrods (Solidago, Asteraceae) (Fig. 2B) (Dorchin et al., 2015). Usually, one larva (but up to three larvae) develop within each chamber, where they develop through three larval instars and pupation, emerging from the galls as adults in early autumn. Mature, third-instar larvae are bright orange, with a fully developed spatula. The spatula is a small, anterior structure near the mouth which, in other species, is used for digging. Second-instar larvae are not only smaller but also generally pale yellow in color and they do not have a spatula. Only the fully grown third-instar larvae are able to jump when removed from their galls.

Cuttings of S. bicolor plants containing Asphondylia sp. larvae were collected in Bedford County, VA, USA, in mid-August. In the lab, we extracted larvae from their galls by first peeling away the stiff leaves making up the rosette that surrounded the gall chambers. A hole was then poked into the top of an exposed chamber, whereupon the gall midge larva inside (if healthy) would immediately squeeze its way out. The larvae were quickly placed on a damp paper towel in a covered plastic Petri dish to keep them hydrated until testing. Healthy full-grown third-instar larvae would begin jumping immediately after their exit from a gall.

Jump preparation (100 frames s⁻¹) and launch (20,000 frames s⁻¹) were visualized using high-speed imaging (1024×672 pixel resolution; Fastcam SA-X2, Photron, San Diego, CA, USA). Each larva was placed on a plastic platform that was illuminated for high-speed imaging (75 W LED, Varsa, Nila, Inc., Altadena, CA, USA). We moved the platform to keep the larva in the camera's focal plane with its lateral side facing the camera. When the larva prepared to jump, we stopped moving the platform and filmed from that position.

Two phases of jump preparation were measured: loop formation (Fig. 3A) and loop contraction (Fig. 3B) (phases are defined in Results). Because of the long duration of jump preparation, we down-sampled the original videos filmed at 20,000 frames s⁻¹ to 100 frames s⁻¹, and report these durations with a temporal resolution of 10 ms.

Launch kinematics were autotracked and calculated from high-speed images sampled at the full 20,000 frames s⁻¹. Each image in a video sequence was thresholded to generate a binary image. The thresholded image sequence was then auto-tracked by calculating the center of the body (centroid function, default B&W threshold settings, ImageJ v.1.52a, National Institutes of Health, Bethesda, MD, USA). Tracking began when the larva was in a motionless, curled, pressurized loop position and ended when the larva either was out of focus or left the field of view (Fig. 3C,D). Because most jumps exceeded the field of view, only the initial ascent was tracked in most analyses. For one video sequence, we repeated the auto-tracking process 10 times to calculate digitizing error. Auto-tracking generated a standard error of the mean that constituted 1% of the reported distance and duration measurements.

We measured launch and takeoff kinematics during two distinct time periods. We defined launch as the time period when the body is accelerated before leaving the ground. Therefore, launch began at the first video frame of body movement (latch release) and ended at the first video frame when the body was off the ground (Fig. 3C). Each video was coded by the investigator for the first frame at which the latch released and the first frame during which the body left the ground. We measured the total duration (t) and distance (d) traveled by the centroid during launch. From this, we calculated average launch power (P=md²t⁻³) where m is the mass of the whole animal. For mass-specific power output [P(0.3m)⁻¹], we conservatively assumed that the larva used one-third of its body mass to launch the system, based on the proportion of the body used for launching (Fig. 3).

We then measured a second time period, which we term takeoff. Takeoff is defined as the time period beginning when the body leaves the ground. We measured the distance traveled by the centroid between each frame before and after the animal leaves the ground. We termed the resulting calculations takeoff speed (v=dt⁻¹) and takeoff kinetic energy (KE=0.5v²). In addition, we calculated the angle of takeoff based on the starting and ending position of the centroid relative to the horizontal ground at the start and end of a 1 ms time period after the frame at which the animal leaves the ground.

For each image sequence, body length was measured along the body's midline, and diameter was measured across the body's center (ImageJ v.1.51n). Measurements were calibrated using videos of a millimeter-scale ruler filmed in the plane of focus. The ruler was re-calibrated to 0.1 mm using a 0.02 mm scale under a microscope [KR-814 (1×3) stage micrometer, Klarmann Rulings, Inc., Litchfield, NH, USA]. We weighed the larvae immediately after jumping data were collected (resolution: 0.1 µg; XPE56, Mettler Toledo, Pleasant Prairie, WI, USA).

We performed error (uncertainty) propagation analyses for the launch calculations based on the uncertainty of the mass (balance: 1×10⁻¹⁰ kg), distance (calibrated ruler: 2×10⁻⁵ m) and time measurements (20,000 frames s⁻¹ imaging: 5×10⁻⁵ s). The resulting uncertainty in the reported launch calculations ranged from 10% to 21% (speed: 10%; acceleration: 12%; kinetic energy: 19%; power: 21%; mass-specific power: 21%) and is represented via the significant digits of the results for both takeoff and launch.

We measured horizontal jump travel distance from jump origin to landing location using a ruler (resolution: 1 mm) on the benchtop. Larvae bounced slightly upon landing, potentially changing measured jump distance by a few body lengths. We did not film these jumps, because their travel distance was too far to capture simultaneously with the takeoff kinematics using extremely high-speed imaging.

Cost of transport, in terms of energy per distance per mass, was estimated for gall midge jumps. We combined average horizontal jump distance from the benchtop ruler measurements with the average takeoff angle measured from high-speed imaging (Table 1), and validated that takeoff speeds measured from the high-speed imaging were congruent with takeoff speeds predicted from the standard ballistics equation d=2v²sin(θ)cos(θ)g⁻¹, where θ is the elevation angle of the jump and g is the acceleration due to gravity (Meriam and Kraige, 2010). We did not incorporate drag or spin losses. We then calculated the kinetic energy required to generate this takeoff velocity, divided that by the average jump distance, again yielding a conservative estimate of energy per unit distance (J m⁻¹). Lastly, we divided the calculated energy per unit distance (J m⁻¹) by the average larval mass (Table 1) to yield an estimate of the energy density required (J m⁻¹ kg⁻¹) to traverse a given distance.

We performed linear mixed models (lme4) in R to assess the associations among body size and kinematics (v.1.1.442, RStudio, Boston, MA, USA; http://www.R-project.org/; Bates et al., 2015). R code, including the statistical models, is available in the associated archival data (Dryad deposition, dryad.jc5745v). Using linear mixed modeling, we compared models to identify how launch distance, launch duration and body mass influence takeoff speed. All models included takeoff speed as the dependent variable and individual ID as the random effect. The null model was a reduced model including takeoff speed as the dependent variable. The other models used different independent variables: launch duration, launch speed, interaction of launch distance and launch duration, and interaction of launch distance and body mass.

We compared a second set of linear mixed models to assess the effect of body size and launch duration on launch distance, specifically through comparison of a reduced model as the null model, a second model with launch duration as the independent variable, and a third model with the interaction of body size and launch duration as the independent variable.

The sample sizes for all of the statistical models were 31 observations across 10 groups (individual ID) (Table 1). We modeled slopes and intercepts as random for all linear mixed models. Before comparing model fits, we tested each model's residuals for normality (shapiro.test) and if the residuals were not normal, then that model was not included in the final model comparison. The model that best fitted the data was determined through relative ranking based on the Akaike information criterion (AIC; lower AIC score indicates better model fit).

We used scanning electron microscopy (SEM) to view the external morphology of the two body surfaces that form the latch. The larvae were difficult to preserve effectively for SEM, so we tried two different methods and report both here so that future investigators can benefit from these efforts. One group of larvae was placed in a 2.5% glutaraldehyde and 0.1 mol l⁻¹ saline buffer solution and then transferred stepwise to a 100% ethanol solution. Specimens were then dried using a critical point dryer (CPD3, Ladd Research Industries, Williston, VT, USA), gold sputter coated (Desk IV, Denton Vacuum, Moorestown, NJ, USA), and imaged using a scanning electron microscope (XL30 FEG-SEM, FEI, Hillsboro, OR, USA) (n=2 larvae). The second group was placed in boiling water and preserved in 70% ethanol. Specimens were dehydrated stepwise through hexamethyldisilazane (HMDS) and 100% ethanol solutions (15 min each in this order: 1:2 HMDS:ethanol; 1:1 HMDS:ethanol; 2:1 HMDS:ethanol; 100% HMDS). The specimens were placed in a fume hood and the fluid was evaporated for 1–24 h. Specimens were then mounted on stubs and imaged using SEM (n=2 larvae).

The boiling water/70% ethanol/HMDS preservation method yielded superior SEM images compared with the glutaraldehyde/100% ethanol/CPD preservation. Glutaraldehyde evidently failed to fully penetrate the larvae, such that the interior decayed, turning the larvae black/brown in color. This, or perhaps the stepwise transfer into 100% ethanol, also caused shriveling, such that the SEMs were not reliable. Therefore, all images and results reported here are based on the boiling water method.

The SEMs revealed that the latching areas of the body are covered with microstructures. We measured the dimensions of the microstructures and the area of the regions covered by these structures in order to assess the role of adhesion and friction in this latch mechanism. Microstructure width was measured from SEM images in ImageJ (ImageJ v.1.51n). To obtain an average size of the microstructures, 10 microstructures were measured on both the third body segment and the posterior-most end of the larva's body. The areas of these two microstructure regions were measured by tracing the area using the polygon selection tool in ImageJ. Because measurements of a three-dimensional surface were taken using a two-dimensional image, measurements are approximate.

Larvae were tested for the presence of resilin (GFPSirius M205FA/M165FC filters, excitation nm AT350/50x, emission nm ET420/50m, band-pass; L6000, Leica Microsystems Inc., Buffalo Grove, IL, USA) (n=3 larvae). Resilin is an elastomeric protein in arthropods that autofluoresces under ultraviolet light (Anderson, 1963, 1966).

Following previous terminology (Campbell and Kaya, 1999a,b; Maitland, 1992), jump preparation occurred in two phases: loop formation and loop contraction (Fig. 3A,B; Movie 1). Loop formation began when one end of the body was positioned on the substrate. Then, the other end slid along the ground until the tip of the posterior end connected with the ventral side of the head segment (Fig. 4, yellow square), thereby forming a latch. Once latched, the body formed a circular shape, balancing on the dorsal side of the tail and the tip of the head (Fig. 3A). Loop formation averaged 1.79±0.34 s (n=2 larvae, 3 jumps, 1–2 jumps per larva).

Loop contraction (Fig. 3B) began when the body started to compress, distorting the circular shape. During this phase, a hinge or kink (Fig. 3B, white dashed line) appeared between the middle third and posterior third of the body, forming what we term a ‘transient leg’ (Fig. 3C, white bracket). Simultaneously, the ‘leg’ swelled, evidenced by smoothing of the body's ridges and folds, even as the ridges and folds were still visible on the anterior and middle thirds of the body. Loop contraction averaged 1.60±0.81 s (n=6 larvae, 9 jumps, 1–2 jumps per larva). Loop contraction ended when the latch was released, simultaneously pushing the transient ‘leg’ against the surface of the substrate and propelling the anterior end of the larva upward (Fig. 3C; Movie 2). The larvae spun as they followed a ballistic trajectory through the air, retaining a slight bend at the hinge, with the example in Fig. 5 spinning at 40 Hz (Fig. 3D, Fig. 5; Movie 3). Fig. 6 shows the trajectory of each jump until the larva was out of focus or left the field of view.

Power density during launch was on average 910 W kg⁻¹ and reached a maximum of 2420 W kg⁻¹ (Table 1). Launch distance was marginally better explained by the interaction between launch duration and body mass [log likelihood estimate (LL)=230, AIC=−444], when compared with the model with only launch duration (Fig. 7A) (LL=223, AIC=−433) and the null model (LL=221, AIC=−431). The residuals of the takeoff speed model containing the interaction between launch distance and duration were non-normally distributed; therefore, we did not include that model in the final model comparison. Among the remaining models, takeoff speed was best explained by launch distance (Fig. 7B) (LL=15, AIC=−17), followed by the interaction of launch distance and body mass (LL=14, AIC=−13). The remaining models had AIC scores above 0.

Larvae jumped as far as 121 mm (36.2 times their body length), with an average of 77.2±12.2 mm (range: 49–121 mm), which is equivalent to 23.1±3.6 times their body length (n=5 larvae, 26 jumps, 2–11 jumps per larva). Average takeoff speed calculated using the ballistics equation and these jump distances was 0.97 m s⁻¹, and is similar to the average takeoff speed measured from high-speed imaging (0.85 m s⁻¹) (Table 1). Using average jump distance, takeoff angle and body mass (Table 1), we calculated that an average of 6.6 J kg⁻¹ m⁻¹ transport cost is required for gall midge jumps.

The larvae were delicate and did not respond well to preservation techniques. The boiling water/70% ethanol/HMDS method yielded superior results, but still had drawbacks. Boiling water may have killed bacteria inside the gut (Martin, 1978), thus preventing the decay found in glutaraldehyde-preserved larvae. About 20% of larvae preserved with this method were in excellent condition for SEM, with minor cracking (which could be due to handling while mounting larvae on SEM stubs) and little to no shriveling. However, 80% of larvae preserved with this method were shriveled, deflated and generally in an unusable condition. It is unclear why the success rate was so low and what caused the differences in preservation outcome.

Even with these technical challenges, we were able to visualize the latch surfaces in multiple specimens. The latch is formed between the posterior-most region of the body and the ventral side of the third body segment. Bands of micrometer-scale, finger-like projections spanned the ventral body surface at each body segment, including the third body segment where the latch is located. The projections were oriented posteriorly in the SEM (Fig. 8). These microstructures were an average of 1.12±0.03 μm wide on the third body segment (Fig. 8, top row), and 1.14±0.08 μm on the posterior-most end of the body. The total area of the field of microstructures was 2.73×10⁴ μm² on the third segment and 1.69×10⁴ μm² on the posterior-most segment. The tip of the tail exhibited two large, semi-spherical pads dorsal to the finger-like microstructures (Fig. 8, bottom row). Autofluorescence was not observed during the test for resilin.

Larvae of Asphondylia sp. are small, worm-like jumpers that use hydrostatic body deformation and associated behaviors that are similar to those of previously described larval insect jumpers. Our study revealed the presence of an intriguing adhesive latching mechanism that mediates elastic energy storage and releases impressive launch kinematics and horizontal travel distances. Improved high-speed imaging capabilities yielded greater resolution of the energetics and dynamics of these jumps and place this class of locomotor system as being as effective as that of equivalently sized legged jumpers. Small, soft-bodied jumpers ultimately achieve an effective and energetically efficient locomotor system through a combination of hydrostatic manipulation, elastic energy storage and latching mechanics.

The mass-specific power outputs of gall midge larval jumps are equivalent to or exceed the known capabilities of high-power vertebrate flight muscle (e.g. 400 W kg⁻¹; Askew and Marsh, 2002), and therefore support the hypothesis that they are not muscle powered and are instead driven by an elastic mechanism. The larvae generate impressive mass-specific power output during launch (Table 1; mean: 910 W kg⁻¹; maximum: 2420 W kg⁻¹). Comparison with the high-power muscle of vertebrate flyers may not be the appropriate threshold for testing for the presence of elastic energy in these larvae: extreme ‘power-amplified’ systems in arthropods, such as those found in trap-jaw ants and mantis shrimp, use force-modified, slowly contracting muscle to load stiff springs rather than high-power muscle (Ilton et al., 2018; Patek et al., 2006; Blanco and Patek, 2014; Gronenberg and Ehmer, 1996).

Therefore, given that they are unlikely to be using high-power muscles, gall midge larvae are almost certainly using elastic energy storage to power their jumps to achieve such high power densities. They appear to store elastic energy in their exoskeleton through manipulation of hydrostatic pressure and associated body deformation (Kier, 2012). We did not detect resilin during external tests of the cuticle. However, Maitland (1992) was able to experimentally store appreciable elastic strain energy in the body walls of another larval jumping insect species, suggesting that either resilin is present in deeper layers of the cuticle or that stiff, deformable cuticular or composite structures are used for elastic energy storage (e.g. Burrows et al., 2008).

In preparation for a jump, the larvae deform their body to form a transient leg, and then launch the body when this effective lever arm is pushed against the ground, similar to previously studied jumping of Mediterranean fruit fly larvae (Maitland, 1992). Once latched (Fig. 3B), the transient leg appears to swell, either through pressurized compartmentalized fluid in this region or by the use of a hydraulic mechanism that moves fluid into the posterior end of the body. This pressurization is visible as the smoothing of ridges and folds in the transient leg (Fig. 4A) during the loop contraction phase (Fig. 3B). A transient hinge, formed along the posterior third of the body, creates this transient leg that launches the jump. At first glance, this transient hinge appears to be a mechanical instability, like a ‘kink’, that ought to be avoided in cylindrical systems (Wainwright, 1988). However, given that this ‘kink’ remains visible throughout the jump (Fig. 5), it is possible that there are controlled internal structures that form this constriction and that the transient hinge is not an instability. Future studies would benefit from examining the timing of the contraction phase and its effect on the energetics of the jump.

Given that elastic energy storage and latch-mediated control enable larvae to travel horizontally from 20 to 30 times their body length, how do the energetic costs of locomoting via legless jumping compare with those of worm-like crawling (Walton et al., 1990)? Here we consider a back-of-the-envelope comparison of the energetic costs of these two modes of locomotion. Assuming no energy losses in the musculature, the energetic cost of moving via jumping in Asphondylia sp. larvae is approximately 8 J kg⁻¹ m⁻¹ (i.e. it takes 8 J of kinetic energy for the animal to move 1 kg of mass 1 m). Given that muscles are typically between 10% and 50% efficient in transferring metabolic energy to mechanical work depending on the species and the mechanics of muscle contraction (Smith et al., 2005), we can multiply larval jumping energy costs by 10 to account for the metabolic energy costs of the underlying muscle contraction, giving a jumping energy cost on the order of 80 J kg⁻¹ m⁻¹. A previous analysis of the cost of transport of larval blowfly locomotion yielded crawling costs of the order of 2300 J kg⁻¹ m⁻¹ (Berrigan and Lighton, 1993). This difference, 80 J kg⁻¹ m⁻¹ for larval, legless jumping compared with 2300 J kg⁻¹ m⁻¹ for larval crawling, suggests one hypothesis for the advantages of small, worm-like jumping: jumping is far more energetically efficient than crawling an equivalent distance.

Based on high-speed imaging (Movie 2), we found that the latch does not involve mouthparts, as has been reported in piophilid or tephritid larvae (Bonduriansky, 2002; Marinov et al., 2015; Maitland, 1992). Additionally, microscopy showed no evidence of legs or protolegs that could be used to aid in latching, as hypothesized in larval fungus gnats (Camazine, 1986). Related species of gall midge outside the Asphondylinii tribe use a hard structure adjacent to the mouth, called the spatula, to exit the gall and burrow into the substrate just prior to the pupal stage (Gagné, 1994; Milne, 1961; Pitcher, 1957). Contrary to early conjectures on the function of the spatula found in third-instar larvae of most families of gall midges (Milne, 1961), the latch is not the sternal spatula (Figs 2 and 4, dashed white ellipse). The spatula is located anterior to the latching mechanism (Fig. 4, yellow square).

Instead, the latch in Asphondylia sp. larvae appears to operate via micro-scale, surface interactions. Hairy adhesive pads function by increasing surface area contact between the organism and the substrate by using micrometer- or nanometer-scale protrusions that can achieve close contact with even-textured surfaces (Federle, 2006; Clemente et al., 2017; Biewener and Patek, 2018). The finger-like microstructures seen on both the posterior-most section of the larva and the third segment are at the same scale as adhesive structures found in independently evolved arthropod larvae, adult insects and even gecko foot pads that use van der Waals interactions to facilitate attachment (Gao and Yao, 2004; Autumn et al., 2002). These microstructures are most likely acanthae, ‘hairs’ that originate from a single cell. This type of hair is common in Diptera, as opposed to the socketed setae more common in Coleoptera and Dermaptera (Arzt et al., 2003; Scherge and Gorb, 2001).

Adhesive mechanisms can attach organisms long term or temporarily to surfaces or other organisms, or create interactions between body parts (Gorb, 2002). The gall midge larval latch system may operate like the rarely observed interlocking fields of microstructures found in the dragonfly and damselfly head-arresting system (Gorb, 2002). The orientation of their microstructures is such that, when the larvae curl in preparation to jump, the microstructures could oppose and interlock with each other. Similar mechanisms have been found in wing-locking devices used to hold forewings and hindwings in place during flight, or to lock wings to the side of the insect's body during rest (Gorb, 2002; Gorb and Goodwyn, 2003). Another possibility is that the larvae excrete adhesive fluids that further enhance interaction between surfaces of the latch by filling in gaps between microstructures, a common adhesion mechanism in insects (Arzt et al., 2003).

The angle and orientation of these microstructures could also be important in determining attachment strength. Changes in gross body position can create a change in microstructure angle, by which the organisms can easily attach or detach (Federle, 2006). The larvae may release the latch mechanism (whether frictional or interlocking) by hydrostatically deforming the body, changing the angle or density of the microstructures, and thereby decreasing the interaction between the third segment and posterior segment surfaces. When the latching forces are overcome, stored energy in the cuticle is released, and the larva launches into the air. The dynamic control and release of latching forces is increasingly recognized as a key part of the control of high-acceleration systems (Ilton et al., 2018).

Finger-like microstructures are found not just on the third body segment but across multiple body segments. The larvae may use them for adhesion to surfaces during locomotion, as has been implicated in other insect larvae (Hasenfuss, 1999). However, Hassenfus (1999) focused on larvae with protolegs (unlike Asphondylia sp. larvae). In addition, unlike the patches of finger-like protrusions in the Asphondylia sp. larvae, the adhesive structures were hooked and arranged in rings around the ends of protolegs. It may be that the outgrowths on Asphondylia sp. larvae aid in crawling locomotion. Uniformly sloped outgrowths allow for directional motion, as the angle of the outgrowths enables forward motion while simultaneously preventing backwards motion using peristaltic body movements (Gorb, 2002). The role of these finger-like microstructures in adhesion-based locomotion is intriguing, and may provide further insight into the locomotor abilities of soft-bodied, legless organisms.

The latch location of Asphondylia sp. larvae can be combined with their jump kinematics to estimate how much force the latch requires to hold everything in place. During a jump, the distal end of the transient leg (Fig. 3C, white bracket) generates an average ground reaction force of 0.60 mN (470 m s⁻² multiplied by an average body mass of 1.27 mg; see Table 1). Just prior to the jump, the muscles must be able to build up that amount of force while the body is held by the latch. The latch is located at the distal tip of the transient leg, allowing for a simple estimate of latch adhesive forces using lever mechanics. While the body muscles contract, the latch must hold everything in place, resisting a maximum of 0.60 mN of force before it releases. The posterior-most groups of microprojections had an area of 16,900 μm² (Fig. 8, bottom row). To hold the body in place, the microstructures would have to generate 3.7×10⁴ N m⁻² of adhesion. Similar microstructures on the feet of leafhoppers (the tibial platellae) generate 1.25×10⁶ N m⁻² of adhesive force during a jump (Clemente et al., 2017), indicating that microstructures are able to generate more than enough force to be used as a latch in the gall midge larvae.

Despite their lack of legs and rigid exoskeleton, the gall midge larvae jump with speeds comparable to those of fleas (Sutton and Burrows, 2011), and they achieve launch accelerations far exceeding those of another ‘legless’ jumper, the click beetle, which flexes its body to release energy stored in a hinge mechanism located between the body segments (Evans, 1972, 1973; Ribak and Weihs, 2011; Ribak et al., 2012).

Soft-bodied jumpers have been observed in two phyla and eight families: Steinernematidae (Campbell and Kaya, 1999a,b; Reed and Wallace, 1965), Tephritidae (Maitland, 1992; Suenaga et al., 1992), Mycetophilidae (Camazine, 1986), Piophilidae (Bonduriansky, 2002), Ichneumonidae (Day, 1970; Saeki et al., 2016), Thyrididae (Humphreys and Darling, 2013), Drosophilidae (Marinov et al., 2015) and Cecidomyiidae. There are over 6203 known species and 736 genera in the family Cecidomyiidae distributed throughout the world (Gagné and Jaschhof, 2014), several of which have been reported to jump (Tokuhisa et al., 1979; Manier and Deamer, 2014; Milne, 1961), suggesting an abundance of subjects for studying hydrostatic jumping mechanisms (Table 2). However, only a handful of studies have examined the jumping biomechanics of soft-bodied, legless jumpers (Campbell and Kaya, 1999a,b; Reed and Wallace, 1965; Bonduriansky, 2002; Maitland, 1992; Milne, 1961; Table 2). Our study thus offers the most inclusive dataset on larval jumping kinematics, including launch speed, acceleration, power and energy. Most other studies report only a few of these metrics (Table 2). Additionally, by filming at an appropriately fast frame rate, we are able to report more accurate kinematics than previous studies. Even so, at these small scales and fast kinematics, even our calculations have error uncertainties of up to 20%, largely arising from the resolution limits of the scale bar.

Relatively little is known about the role of jumping in the life history of these larvae. Larvae of most cecidomyiid species exit their galls at the end of their third instar, fall to the ground, and use their sternal spatula to burrow into the soil, where they then pupate (Sen, 1939; Milne, 1961; Pitcher, 1957; Dorchin et al., 2015). Prior to burrowing, larvae jump in order to locate a suitable location for pupation. Jumping has been shown in a variety of taxa to help larvae avoid desiccation, escape predation and parasitization, and avoid lethal temperatures (Humphreys and Darling, 2013; Manier and Deamer, 2014; Maitland, 1992; Day, 1970; Saeki et al., 2016). The cecidomyiid groups that do not pupate in the soil develop through all three larval instars and pupate in the gall, emerging only as adults. These include Asphondyliina (the subtribe to which Asphondylia sp. belongs), as well as a smattering of other species throughout the family Cecidomyiidae (Gagné, 1994; Uechi and Yukawa, 2006). Current understanding of this group suggests that these larvae never jump, as they are confined in the gall during the only point in development when jumping behavior is exhibited. Therefore, it is possible that Asphondylia sp.’s jumping ability is an evolutionary relic that served a purpose in their ancestors (pupation in soil is the ancestral state for Cecidomyiidae: Tokuda and Yukawa, 2007; Gagné, 1994; Tokuda, 2012).

However, we have observed that Asphondylia sp. removed from their galls during the third larval instar are able to jump upon the substrate, burrow into soil, pupate and emerge as adults (M.J.W., personal observation). Additionally, their jumping behavior is robust, as evidenced by their impressive kinematics and ability to jump multiple times. In combination, these observations suggest that the sternal spatula, latch mechanism and jumping ability are not mere evolutionary remnants from when they diverged from a burrowing ancestor. It remains uncertain, however, under what conditions larvae would leave their galls, thereby necessitating pupation in the soil. Studies on other gall midges of goldenrods have shown that rosette galls like those induced by Asphondylia sp. are ecologically busy locations (Crawford et al., 2007). Gall chambers are invaded by insect predators, parasitoids, inquilines and chewing herbivores that feed on plant tissue that make up the galls (Dorchin et al., 2007, 2009, 2015). Thus, it is not uncommon for the inhabitants of galls to be disturbed as their chambers are breached. If released from their galls, the ability to jump to escape predators or to quickly find a suitable burrowing site for pupation would therefore be advantageous. To better understand the natural history of this gall midge species and the potential function of its jumping, further behavioral analysis and field observations would be of great value.

With dramatic growth in the fields of small, high-acceleration systems (e.g. Ilton et al., 2018), soft robotics (e.g. Cianchetti et al., 2018; Rich et al., 2018; Sitti, 2018; Trimmer, 2018) and biological adhesion (Autumn et al., 2002; Gorb, 2002, 2005; Gorb and Goodwyn, 2003; Federle, 2006; Labonte and Federle, 2015; Labonte et al., 2016), the realm of small, worm-like, legless jumpers offers notable insights at the interface of these fields. These small larvae combine elastic energy storage and adhesive latching to achieve impressive launch kinematics and horizontal travel distances that rival those of legged jumpers. Not only is there potential for discovery in future studies of the natural history and performance of jumping in Asphondylia sp. but also these findings bridge previous research across an array of similar, independently evolved systems and point toward an untapped and rich diversity of mechanically interesting biological mechanisms in soft-bodied, legless jumpers.

We greatly appreciate the help of Michelle Plue in using equipment at Duke University's Shared Materials Instrumentation Facility, Netta Dorchin for help with methodology for preservation of larvae for SEM imaging, Fred Nijhout and Walter Federle for their input on insect adhesive mechanisms, and the Patek Lab, especially Sarah Longo, for feedback and support.