| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Abteilung für Biologische Kybernetik und Theoretische Biologie, Fakultät für Biologie, Universität Bielefeld, D-33501 Bielefeld, Germany; 2 Department of Zoology, University of Cambridge, Cambridge CB2 3EJ, United Kingdom
Submitted 29 April 2003; accepted in final form 23 May 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
; Giszter et al. 1989; turtle: Mortin et al. 1985; Stein 1983) and invertebrates. Locusts groom the ventral part of their hind leg coxa with the hind leg, but slightly more anterior sites are groomed with the middle leg (Berkowitz and Laurent 1996a). Leg avoidance reflexes of locusts similarly fall into discrete categories depending on the site of stimulation (Burrows and Siegler 1985; Siegler and Burrows 1986).
In vertebrates, grooming movements within an effector workspace can be graded so that the limb can reach intermediate targets. In frogs, placing movements during the wiping reflex shift with target location (Giszter et al. 1989). In turtles (Mortin et al. 1985; Stein et al. 1986a) and dogs (Sherrington 1906), scratching movements are both localized and graded in that stimulation of a small region within a receptive field elicits a targeted limb movement toward the stimulated subregion of the receptive field. Stimulation of different subregions gives rise to different movement patterns.
In invertebrates, there is no experimental evidence that somatosensory input can cause a continuous shift of cyclic limb movements. Many forms of insect leg movements, including grooming and leg avoidance responses, are elicited by deflection of tactile hairs on the body surface. Central projections of these hairs form a somatotopic map in the segmental ganglia (Murphey et al. 1980; Newland 1991), and their postsynaptic interneurons reflect this somatotopic organization (Burrows and Newland 1993). In locusts, touching different places on the body, e.g., the first and last segments of the abdomen, elicits different grooming movements of a hind leg (Berkowitz and Laurent 1996a) and corresponding characteristic electromyographic (EMG) patterns can also be recorded in a deafferented animal (Berkowitz and Laurent 1996b). Similarly, different sites on the forewing elicit different leg movements (Matheson 1997, 1998). However, because transition zones between different patterns have not been mapped and only the kinematics of movements to distant sites have been analyzed quantitatively, the existence of a behavioral continuum within a particular form of grooming movement remains to be shown for an invertebrate.
Here we show that locust grooming movements that are elicited by tactile stimuli to adjacent sites on the forewing form a behavioral continuum. We demonstrate that a single effector point is used to groom all tested sites on the forewing and that the observed movement pattern is shifted in joint angle space by a continuous somatosensory stimulus parameter. To test whether a putative internal body surface representation is formed from both exteroceptive and proprioceptive information, we have separated the effects of these inputs by altering the target site and starting posture of the leg.
Preliminary results have been published in conference proceedings (Dürr and Matheson 2001).
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Experiments were carried out on eight adult female desert locusts (Schistocerca gregaria Forskål) taken from a crowded laboratory culture. Animals were tethered by a loop of fine wire that passed around their pronotum without obstructing movements of any of the legs. They were suspended above a light (4.5 g) styrofoam ball on which they stood or walked. The tether allowed each animal limited freedom to adjust its body posture with the exception of thorax height above the substratum, which was set to the height normally maintained by a walking locust. The eyes were covered with solvent-free typists' correction fluid, both to exclude visual cues and to reduce spontaneous struggling movements. Experiments were carried out at 22–24°C.
Grooming movements of the hind leg ipsilateral to the stimulus site were analyzed. To test the effect of start posture on the observed behavior, the hind leg was put into one of two alternative postures by placing the tarsus on a horizontal rod located at one of two fixed positions (Fig. 1A). The start locations were alternated systematically for periods of 
12 trials. Grooming behavior was elicited by a gentle touch of the ipsilateral forewing with a fine paintbrush. The site of stimulation along the wing midline was chosen in a pseudorandom order. The exact stimulus location and the resulting movements were measured from video tape as described in the following text. Stimulus locations were subsequently binned into five adjacent, equal-sized target sites on the wing. The area of stimulated wing surface was considerably smaller than the area per bin. Stimulus duration was not controlled, but controls were made to ensure that this had no detectable effect on the responses. For instance, there was no difference in limb movement velocity during or after tactile stimulation.
|
Locusts only rarely touched their own body surface during the first few movement cycles of the response. To obtain only responses that are open-loop with respect to the stimulus, responses where the leg made contact with the body surface or the stimulating brush were analyzed only until the time of first contact. In five animals, all 10 posture:target combinations were used. In 369 of 850 trials (43.4%) a grooming response was observed. In trials rated as "nongrooming responses," the animals responded with a kick or a step-like movement or did not respond at all. A kick was clearly distinguishable from grooming because it was preceded by conspicuous flexion of the tibia against the femur. In step-like movements, the tarsus either slipped off the rod or was lifted only a few millimeters above the rod before being placed on the ground. Seventy trials (8.2%) resulted in a bilateral response, involving grooming movements of both hind legs. These trials were included in the analysis, but only the movement of the leg on the stimulated side was measured. In three additional animals, only target sites 2 and 5 were stimulated and only the anterior start posture was used. In 240 of 352 trials (68%), these animals showed a grooming response.
Video acquisition and analysis
Animals were videotaped from the left side and, in some experiments, from above using synchronized color CCD cameras (JVC TK-C1380) operated at a shutter speed of 1/500 s. The VHS video signals were combined on a multiviewer (For-A MV-40PS) and time stamped using a video timer (For-A VTG-33). Images were taped on an sVHS video recorder (Panasonic NV-HS900), displayed on a monitor and played back for capture by a personal computer video interface card (miroVIDEO DC30 plus, Pinnacle Systems).
Interlaced video frames were captured at a size of 768 x 576 pixels using miroVIDEO Capture software and saved as AVI files with Microsoft video1 compression. Each sequence was subsequently deinterlaced into two separate AVI files (Adobe Premiere 4.2), each containing the sequence of either odd or even half frames, giving an overall frame rate of 50 s–1. A custom-written program (Borland Delphi) was used to display the pairs of AVI files and to digitize manually the coordinates of 8 points per frame and camera view (Fig. 1B). The marked points included four joints on the hind leg and a point on the second tarsus segment to allow measurement of all joint angles. Flexion of the unguis was ignored as tarsus position and orientation were determined solely by the proximal segments. Tarsus length was standardized to the length of 6.75 mm (see straight line segments in Fig. 1D). The line between the hind leg coxa and the front leg coxa defined the x axis of the body-centered coordinate frame with the thorax-coxa-joint of the hind leg set to the origin. The proximal and distal ends of the wing were used to assign the target site number of the stimulus. Stimulus location and position of the tarsus rod were digitized in the first frame of each behavioral sequence.
Spatial resolution of the side-view-only video images was 0.1 mm/pixel, allowing manual digitizing accuracy of 0.5 mm as determined from the SD of all digitized points in repeated analyses of three sequences. Spatial resolution of the side-and-top-view video images was 0.2 mm/pixel. Grooming movements of the hind leg are executed without notable protraction of the coxa, so the angle between the leg and the sagittal body plane changes only little (5 and 95% percentiles are –11 and 20°, respectively). As a result, the difference between the joint angles measured from the side view only and those determined from a three-dimensional reconstruction based on both camera views was negligible (1.2 ± 4.5° for thorax-coxa joint, –0.6 ± 4.8° for coxa-trochanter joint, –0.4 ± 1.4° for femur-tibia joint, 0.4 ± 1.7° for tibia-tarsus joint; means ± SD) Because of this, only the two-dimensional information is used for the presented results, i.e., only the rostrocaudal and dorsoventral location of the digitized points (x axis and z axis of body-centered coordinate system, respectively).
Data analysis
Analysis of individual grooming responses started with the frame before tarsal movement began and ended with the occurrence of one of the following four events: the tarsus touched the ground, the tarsus hit the stimulus brush, the leg completed three cycles of movement, or the tarsus stopped moving for >120 ms. The digitized joint coordinates, angles, and trajectories were individually inspected (Fig. 1, C and D) and analyzed by means of a custom-written program (Borland Delphi) and Origin (MicroCal). ANOVA statistics were calculated with the Statistics Package for the Social Sciences (SPSS).
Two components of each grooming response were analyzed: a short initial component in which the trajectory of the tarsus was rather straight and a second cyclic component. The initial movement component was determined for a period of F frames, and quantified by the average direction vector vav =1/F 
v(t), with 0 < t 
F*20 ms. Each (transposed) direction vector v = (x, z)T is the normalized direction vector pointing from the start coordinates [x(t0), z(t)]T 0 of the tibia-tarsus joint to the trajectory point at time t, [x(t), z(t)]T. The angle between vav and the target direction vtarget at t = t0 is a measure of targeting accuracy (error angle). The length of vav is a measure of targeting consistency and was used to determine the average duration of the trajectory until the beginning of the cyclic component (see Fig. 5 and corresponding text for further details).
|
The cyclic component was quantified by means of a probability distribution PS, describing the likelihood p(X|S) of a particular part of the leg moving across point X in the leg's workspace, given the tactile stimulus location S. Ten distributions PS were determined for both the tarsus and an equally long part of the distal tibia. They were calculated as follows. The workspace was divided by an orthogonal lattice of 1-mm grid width. The movement of the leg segment was interpolated between each pair of consecutive video frames, and the nodes of the lattice that were covered by the interpolated area were determined (see also Fig. 6 and corresponding text). Repetition of this interpolation for all video frames gave a two-dimensional (2D) frequency histogram on the lattice, indicating the number of observations in which the leg crossed a given lattice node. Histograms were calculated for each trajectory, spatially smoothed by a 2D-Gaussian filter of 5 x 5 mm area and 
2 = 2 mm2, normalized to a standard volume, and averaged across trials node by node. Using these histograms as empirical likelihood functions, Bayes' rule was applied to decide which one of two sites Si had most likely been stimulated to cause the leg to move across point X during a single video frame. Assuming p(Si) = 0.5 for two equally probable stimulus sites Si as a prior, a probabilistic decision was made according to p(Si|X) = p(X|Si)p(Si)/p(X). The probability of making a wrong choice between stimulus sites i and j was 
, with g grid nodes Xg. Finally, assuming a binomial distribution of n independent observations with an error probability of 
, the critical number of observations was calculated that was necessary to allow no more than n/2 – 1 wrong decisions in 95% of cases. If the critical number was smaller than the average number of video frames per measured trajectory, the two distributions were considered statistically different. In this case, a single response trajectory would allow reliable distinction between two possible target sites (see Tables 1 and 2 and corresponding text).
|
|
|
To compare our results with those of Berkowitz and Laurent (1996a), the trajectories were also analyzed in joint angle space. Measurements of the thorax-coxa joint angle and the coxa-trochanter joint angle were smoothed with a sliding window having a width of five frames, weighted by binomial coefficients. For each trajectory, the center of gravity of the cyclic component in joint angle space was calculated as the average posture vector aav = 1/n ai, where the components of each (transposed) posture vector ai = (aTC, aCT, aFT, aTT)T are the angles of the thorax-coxa, coxa-trochanter, femur-tibia and tibia-tarsus joints, respectively, and n is the number of observed postures.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
). If a front wing is stimulated, the grooming response involves movement of the ipsilateral hind leg in which, typically, the tarsus is moved toward the stimulus site and repeatedly moved in cycles around or close to the stimulated area (e.g., Fig. 1, C and D) (Matheson 1997, 1998). To test for continuous modulation of a hind leg movement pattern, we analyzed grooming responses to mechanical stimulation at five target sites (Fig. 1A). Two natural start postures were used to test for an effect of proprioceptive sensory signals. What is being targeted?
A prerequisite for a functional interpretation of limb targeting behavior is knowing what part of the effector limb is being aimed. Whereas this may seem to be a trivial problem in the case of human reaching movements, it is less obvious whether locusts aim the tarsus (foot), the distal tip of the tibia (heel), or another part of their leg. We divided the leg into 24 units and measured the targeting performance of each (Fig. 2). For the most distal target (site 5), the tarsus approached the target most closely. The tarsus and the three distal units of the tibia sometimes hit the target perfectly (white lines in Fig. 2, A and B), but the average distance for any tibial unit was larger than for any part of the tarsus. The proximal femur performed worst because it is too short to approach distal targets. For increasingly proximal target sites, the minimum of the histogram shifted gradually in a proximal direction. Critically, however, there was no sudden shift in the effector point and therefore no evidence for switch in movement strategy. A second local minimum appeared for target sites that required the whole leg to be rotated forward for the distal tibia to reach the target. This forward rotation meant that the femur had to approach or even cross the target site (see femur movement in Fig. 1C). As a result, the distance histogram for responses to target 2 in Fig. 2A has two minima: one for the mid femur and another for the distal tibia. Location of the minima in histograms for responses to target sites 3 and 4 (not shown) are intermediate to the two examples shown in Fig. 2. It is important to note that the femoral minimum is the inevitable result of forward rotation of the femur that is required to bring the distal tibia and tarsus close to the target. The tibia crossed the stimulus site repeatedly during the cyclic part of grooming, whereas the femur was typically rotated forward so that it passed the target only once and was then held anterior to the target throughout the response (see Fig. 1, C and D). This indicates that although the femur can pass over proximal targets, it is not specifically aimed at them. In contrast, the tibial minimum in Fig. 2 indicates the effector site as it is due to repeated close approaches to the target with the distal part of the tibia. The same results were obtained for responses from the posterior start posture (Fig. 2B), indicating that the same part of the leg is aimed regardless of start position. To determine the part of the leg that was most reliably aimed at the stimulus, irrespective of target site, the average distance histogram was calculated from all 10 posture: target combinations (Fig. 2C). It had a global minimum at a point on the distal tibia, 
4 mm proximal to the tibia-tarsus joint. This location on the tibia will subsequently be called the "effector point." Locusts therefore aim most reliably with their distal tibia rather than with their foot.
|
Targeting accuracy
Having determined the point on the leg that most effectively reaches the target site, we used the movement trajectory of this effector point to locate the position of closest approach to each target for the anterior (Fig. 3A) and the posterior (Fig. 3B) start postures. The minimal distance to the target served as a measure of targeting accuracy that was suitable for statistical analysis. Targeting accuracy was best for the three middle targets (sites 2–4) and deteriorated proximally and distally (Fig. 3, A and B). Responses to the most anterior target (site 1) on average did not reach forward far enough to hit the target. Similarly, responses to the most posterior target (site 5) typically did not reach back far enough. With the posterior start posture the average point of closest approach to target 5 was located slightly in front of the average start position, showing that the femur-tibia joint tended to be flexed, even if this caused a movement away from the target. A two-factorial ANOVA revealed that the effects of target site and start posture on the x coordinate of the closest point of approach explained 85% of the total variance [F(9, 6090) = 380.9, P < 0.001]. Whereas the target site had a strong and significant effect on the closest point [F(4, 6090) = 505.7, P < 0.001], the start posture had no significant effect [F(1, 6090) = 1.757, P = 0.814]. Interaction of both factors was weaker than the effect of target site, but also significant [F(4, 6090) = 9.135, P < 0.001]. This suggests that the accuracy of hind limb targeting depends strongly on a somatosensory input and that the strength of this influence is modulated by proprioceptive input.
|
Preferred directions of movement during grooming
In grooming, the predominant movement direction of each leg segment indicates the most likely direction that it will be moving in when it contacts an obstacle. We investigated the functional significance of the combined action of all four leg joints by relating the frequency of observed movement direction of the tibia (Fig. 4, A and C) and tarsus (Fig. 4, B and C) to the location and orientation of spines and spurs on the tibia (Fig. 4D). To illustrate the most likely contact angle, frequency distributions of movement direction were determined for each segment. The polar plot histograms in Fig. 4 show the movement direction of the distal end of each segment relative to that segment's own long axis. Frequencies were averaged across all 10 posture: target combinations, revealing two peaks for each limb segment that corresponds to the predominant movement directions (Fig. 4C). The small values for directions around zero for both tibia and tarsus (Fig. 4C) show that neither segment is likely to hit an object with its tip (much as a soccer player avoids hitting a ball with the tips of his toes). The tibia was most likely to push outward (260°) or downward at an angle of 40° (Fig. 4A). The tarsus predominantly moved downward relative to its long axis (90°, Fig, 4B). This implies that it would most likely contact an object with its ventral surface, which would allow it to grasp the object. The second smaller peak for the tarsus, at 300°, corresponds to upward movements, a direction which is appropriate for kicking an object away.
|
The locust tibia is armed with two rows of sturdy spines on its dorsal surface and two pairs of spurs at its distal end. The orientation of these spines and spurs can be related to the observed limb movements. In 10 animals examined, the tibial spines closest to the effector point were oriented at an angle of 44° with respect to the most likely contact direction of the distal tibia (304° for the spines vs. 260° for the movement, Fig. 4, C and D). The spurs also formed an angle of 41–62° with respect to the most likely contact direction (301–322 vs. 260°, Fig. 4D) and furthermore coincided with the observed limit of extension of the tarsus. The orientations of the spines and spurs, which lie proximal and distal to the tibial effector point respectively, are therefore ideally suited to prevent the tibia or tarsus slipping off any object that they contact during grooming.
Effect of target site and start posture on initial targeting
To show how movement trajectories depend on stimulus site or start posture (e.g., Figs. 1D and 6A), each trajectory was divided into two parts: the initial 200 ms of the response in which the tarsus moved toward the target and the subsequent cyclic component. Figure 5A shows the average direction vectors vav of the initial movement component (see METHODS) from each start posture toward each target. Both target site and start posture had a statistically significant effect on the initial movement direction (test for common median direction after Fisher 1993; target site: P 0.001 for both start postures; start posture: P 0.001 for targets 1 to 3, P 0.025 for target 4, P 0.005 for target 5), indicating that the response contained directional information about the target right from movement onset. The temporal development of movement direction is shown in Fig. 5B. Although the direction of vav changed over time, this amounted to <20° during the first 400 ms. Responses to different target sites differed from the start of movement. The choice of a 200-ms window to calculate vav was based on the observation that the length of vav reached a saturation level after that period of time (Fig. 5C). The length of this direction vector is a measure of targeting consistency (0 indicates random directions, 1 indicates a straight trajectory). During the initial part of the response, this value increased as the average trajectory direction became increasingly consistent with time. As the majority of trajectories have entered the cyclic component, the length of vav saturates at a value <1 because all trajectories remain within the triangle spanned by the start position and the proximal and distal borders of the cyclic trajectory. The error angle between vav and the target direction systematically deviated from zero (Fig. 5, D and E), indicating that, in spite of the systematic effect on initial movement direction, initial targeting was not very accurate.
Targeting of the cyclic grooming phase
Because initial targeting does not reflect the grooming accuracy of entire trajectories (Fig. 3), we further analyzed the subsequent cyclic grooming phase, in which the tarsus and distal tibia were repeatedly moved across an area near the target. The total workspace of the tarsus during this part of grooming was large (dashed line in Fig. 6A) but the area covered by each individual response was only a small subset of this region (2 examples are shown in Fig. 6A). To capture the natural variability of the animal's performance in grooming the target site, we quantified the likelihood that the effector point moved across any given point in the workspace (see Fig. 6B and METHODS). This resulted in two-dimensional probability distributions that we represent as density plots like those shown in Fig. 6, C and D. The central tendency of these distributions was characterized by the location of their peak value and their center of gravity (white squares and circles in Fig. 6, C and D, respectively).
The main portion of the distribution for movements aimed at the posterior target site 5 from the anterior start posture lay anterior to the target (Fig. 6C). The distribution for the anterior target site 2 lay predominantly posterior and dorsal to the corresponding target (Fig. 6D). The two measures of central tendency occupied similar but not identical locations. Both distributions covered a considerable part of the workspace, yet they are clearly distinct. We used four measures to compare the location, size and overlap of the 10 different distributions that resulted from all possible posture:target combinations: their contours at certain probability levels, their statistical separability, their centers of gravity, and transects taken along the line connecting the centers of gravity.
Distributions for movements aimed at the same target from different start postures were similar (compare correspondingly color-coded contour lines between the left and right columns in Fig. 7), indicating that start posture had no effect on the cyclic phase of the grooming response. There was an anterior (rightward) shift with increasingly anterior target site showing that the cyclic part of the response was significantly shifted to correspond with the target. The centers of gravity were located dorsal to the wing surface for the three most anterior target sites (Fig. 7, 1st row), revealing that the leg spent most of its grooming effort above the animal. The 10% contour lines for each target site (Fig. 7, top) had virtually no overlap. At the 75% level (Fig. 7, 4th row) all distributions had common overlap. Some movements made in response to stimulation of the anterior target site 1 (Fig. 7, magenta triangle, top row) extended far anterior, above the animal's head, leading to an elongation of the corresponding (magenta) probability distribution that is evident in the bottom three rows of Fig. 7.
|
To test whether the target site affected the grooming distributions, we determined how many observations were required to deduce reliably the correct target site from pairs of distributions (Table 1, also see METHODS). This number is related to the overlapping volume of the pair of distributions (Table 1). Grooming movements made in response to stimulation of next-but-one target sites could be discriminated reliably in all cases (Table 1). In addition, movements made in response to target sites 3 and 4 could be discriminated even though the targets are next to each other, as could movements aimed at target sites 1 and 2 in the case of the anterior start posture (Table 1). In no case could the start posture be determined from observations of the cyclic phase of the grooming response (Table 2). Furthermore because each distribution can be discriminated from two to four others and none of the distributions can be constructed by scaling and summing of the others, there is no indication of site-dependent switching between different forms of response. We conclude that the tested target sites lie within the receptive field of a single form of response, i.e., the anterior-posterior shift is driven by a continuous somatosensory input.
To characterize how the exteroceptive stimulus parameter, i.e., position on the wing surface, shifts the behavioral output, we connected the centers of gravity of each probability distribution (Fig. 8A). The resulting axis followed a curved path from posterior-ventral (Fig. 8A, 0 mm) to anterior-dorsal (Fig. 8, 70 mm), suggesting that a shift of the stimulus site along the linear wing axis shifts the center of the grooming distribution along a compressed and curved "output" axis. To test whether this nonlinear transformation was dependent on the choice of the leg segment, the analysis was repeated for the tarsus. The resulting density distributions and their centers of gravity greatly resembled those for the tibial effector site except that they were broader and shifted ventro-caudally by a few millimeters (Fig. 8A, dashed lines). Transects through the distributions taken along the output axis showed that the peaks were larger for the tibial effector distributions (Fig. 8B) than for the tarsus distributions (Fig. 8C), indicating that movements of the distal tibia were more focused than movements of the tarsus. A distal shoulder in the transects that became more pronounced with increasingly anterior target sites reflects return trajectories from short responses. The proximal shoulder of the distributions for target site 1 (magenta) are due to occasional far forward reaching movements.
|
Behavioral continuum in joint-angle space
Because none of the five transects in Fig. 8, B and C, could be explained by a weighted sum of the remaining four, we hypothesized that the observed shift was not due to blending of a small set of distinct forms of motor patterns but rather due to a continuous modification of a single form of motor pattern within the same somatosensory receptive field. The kinematics of locust hind leg movements are redundant, so it is necessary to demonstrate this continuous modification not only within the Cartesian workspace of the leg, but also in joint angle space. The homogeneity of all 610 individual grooming movements was therefore examined by plotting their average posture vectors in joint angle space, following the approach used by Berkowitz and Laurent (1996a) to compare different forms of grooming motor patterns. If the behavior resulted from blending of discrete motor patterns, then the responses would consist of movement cycles caused by different motor patterns (e.g., Mortin et al. 1985). As a result, the average posture vectors would cluster around the average postures of the pure patterns and few intermediate vectors. Given the small number of cycles in a single response (1–3), intermediate responses should be few. If, on the other hand, there was continuous modulation of a single form of movement, there should be no clustering because the average posture vector would vary smoothly.
We analyzed the time course of change of three leg joint angles (defined in Fig. 9A) during grooming responses (Fig. 9B). The thorax-coxa joint rotated the entire leg, the coxa-trochanter-joint levated or depressed the femur about the coxa (because the trochanter is fused to the femur in an immovable joint), and the femur-tibia joint extended or flexed the tibia about the femur. These three joints therefore allow an infinite number of postures for the hind leg to reach any target within its workspace. Because the average thorax-coxa angle during grooming correlates with stimulus position, however, it is possible to use the current thorax-coxa angle to calculate the corresponding optimal angles of the other two joints for the leg to reach the target (Fig. 9B). The angles attained by these joints during grooming repeatedly approached their optimal values as the leg moved in the cyclic part of the behavior (Fig. 9B). By using the average recorded thorax-coxa angle (40°), the optimal angles of the coxa-trochanter and femur-tibia joints could be plotted for targets on the wing midline or the leading or trailing edges of the wing (Fig. 9C, thick and thin solid lines). An increase of thorax-coxa angle would shift the optimal angle lines to the left, whereas a decrease would cause a shift to the right (not shown). The angles measured at the point of closest approach to the target (see Fig. 3) formed a continuum along the wing midline when plotted in this joint angle space (Fig. 9C). There are more data points near the end and in the middle of the wing because there were more responses to target sites 2 and 5 than for other sites. The joint angles that correspond to limb postures at the center of gravity of each response also form a continuum in joint space with no sign of clustering (Fig. 9D). As single responses cannot be assigned to distinct clusters in joint angle space, there is no indication of a small set of distinct movement cycles that the animal can combine depending on stimulus site. This suggests strongly that the underlying motor pattern was varied smoothly, i.e., was shifted by a continuum of stimulus positions. For comparison, we superimposed on this distribution published data points for movements aimed at targets on the first and last abdominal segments (labeled "ear" and "posterior abdomen" by Berkowitz and Laurent 1996a; see our Fig. 9A). These responses fall neatly at the ends of our ellipsoid distribution for movements aimed at targets on the wing, suggesting that all of these responses are the same form of grooming.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Effects of proprioception and tactile exteroception
The location of the target site (i.e., the tactile stimulus) on the wing has a highly significant effect on both the initial targeting direction (Fig. 5) and the cyclic grooming phase of the response (Figs. 7 and 8). The location of the closest point on the trajectory is also significantly related to the location of the tactile stimulus. This is caused by a systematic shift of the movement pattern in joint angle space: the mean femur-tibia angle decreases (joint flexion) with a rostral shift of the target site, whereas there is an increase in both thorax-coxa angle (forward rotation) and coxa-trochanter angle (levation).
Start posture has a significant effect on the direction of movement during the initial phase but does not affect the area covered by the leg in the grooming phase. Start posture also significantly modulates grooming accuracy (Fig. 3), indicating that there is interaction between proprioceptive and exteroceptive afferent information in the sensory control of these leg movements. We have shown elsewhere that loading the leg has no significant effect on the cyclic phase of grooming (Matheson and Dürr 2003), which provides further evidence that proprioception plays a central role in closed-loop control of grooming.
Behavioral, physiological, and computational aspects of the sensorimotor transformation
In insects, the central projections of exteroceptive afferents from the legs and lateral thorax are organized into a continuous somatotopic map (Murphey et al. 1980; Newland 1991; Newland et al. 2000). Our finding of a continuous shift of the cyclic component of grooming with target site does not address directly the nature of the neuronal mechanism underlying limb targeting, but it shows that the somatotopic representation of the body surface can be read out to produce a continuum of behavioral responses. In stick insects, targeted stepping movements also form a behavioral continuum (Cruse 1979; Dean and Wendler 1983), but this aiming of one limb toward another involves a different sensorimotor transformation to that required for aiming a limb toward a point on the body surface. In the case of stepping, Brunn and Dean (1994) have described a set of three interneurons that together encode the tarsus position of the adjacent leg, each one signaling the angle of a single joint. This suggests a purely proprioceptive implementation of the inverse kinematics transformation, which cannot be the case for the grooming movements we describe. In the locust, Matheson (2002) described spiking interneurons that are likely to be involved in the sensorimotor transformation underlying grooming. Some of these encode both wing exteroceptive signals and leg proprioceptive signals, so they could carry out the computations underlying the sensorimotor transformation of a somatotopically coded tactile stimulus into a rhythmic motor sequence.
A computational framework based on a simple artificial neural network, which was developed to model the control of targeted leg movements in walking insects (Cruse et al. 1998), also lends itself to modeling locust grooming. This network incorporates sensorimotor feedback loops and inter-joint coupling. As a model for stick insect leg movements, it explains the transition from aimed protraction into cyclic searching (Dürr 2001); a movement sequence that is similar to hind leg grooming in locusts. Current versions of the network use proprioceptive information from one leg to signal the target posture for an adjacent leg. An analogous feed-forward network could also transform exteroceptive information about locations on the wing midline into target postures suitable to move the effector point along the map axis in Fig. 8 A. Our data provide a behavioral basis to study and model the computational nature of this transformation in a tractable insect model system, allowing us to link cellular neurophysiology with behavioral performance.
Continuous shift of a movement pattern within a single form of movement
Adaptive behavior requires that movements of a limb are changed in a context-dependent way. These changes can be abrupt switches between different "forms" of a movement, blends of different forms or graded variations within a form (reviewed in Stein et al. 1986b). Complex movements can be built up from chains of simpler movement "modules" (Berridge 1989; Berridge et al. 1987). Insects can use any given limb in a wide variety of behaviors, including walking in different orientations, searching, righting, grooming and perhaps stridulation (e.g., cockroach: Delcomyn 1987; Reingold and Camhi 1977; locust: Duch and Pflüger 1995; Pflüger and Burrows 1978). Because many of these motor patterns occur in mutually exclusive behavioral contexts (e.g., running and grooming) they are switched on and off rather than modulated by a simple sensory cue. In a singing grasshopper, for example, the patterns of cyclic hind leg movements that produce the sound are controlled by sequential activation of command neurons in the brain (Hedwig and Heinrich 1997). Grooming movements in insects have generally been considered to be stereotyped, cyclic movement patterns that are released by stimulation of a distinct area on the body surface (e.g., Eaton and Farley 1969; Honegger et al. 1979; Zack 1978). In locust grooming, the movement sequence and underlying muscle activity differ considerably between responses elicited by stimulation of different sites on the body surface (Berkowitz and Laurent 1996a,b). For example, the animal uses the hind leg to groom the posterior surface of the hind leg coxa but switches to using the middle leg to scratch the anterior surface only a few millimeters away. This corresponds to a switch in movement form analogous to that seen in turtles (e.g., Mortin et al. 1985) and frogs (Berkinblit et al. 1989; Giszter et al. 1989). Leg avoidance movements in locusts, which result from the activation of tactile hair afferents on the legs themselves, also fall into perhaps as few as four discrete patterns as revealed by mapping the receptive fields of activated motoneurons (Siegler and Burrows 1986) and interneurons (Burrows and Siegler 1985). Grooming movements aimed at targets on the forewing differ between anterior and posterior stimulation sites (Matheson 1997, 1998), but because the boundaries of the somatosensory receptive fields were not determined, it was not clear whether the differences were due to switching between a small set of movement patterns or due to a fine-graded and spatially continuous shift of a single movement pattern.
In the present study, we found no evidence for receptive field boundaries across the forewing. Rather our quantitative kinematic analyses revealed that shifting the somatosensory input signal results in a graded modification within a single form of output pattern. First, the animal aims a single effector point—the distal end of the tibia—at all targets on the wing, which is the key criterion that has been used to define a movement form in vertebrates (Stein et al. 1986b). Second, both target site and start posture significantly affect initial targeting, showing that responses are appropriately directed right from the beginning of the response (Fig. 5). Third, the likelihood with which a particular region in the hind leg's workspace will be groomed changes systematically with target site (Figs. 7 and 8). Fourth, the responses form a continuum within the joint angle space (Fig. 9), indicating that there are only smoothly graded modifications of a movement pattern and not combinations of distinct movement patterns. Finally, the observation that hind leg movement patterns that are used to groom the first and the last segment of the abdomen ("ear" and "posterior abdomen" in Berkowitz and Laurent 1986a) fall at the ends of the range spanned by movements used to groom the wing (Fig. 9), indicates that there is only one grooming form for both the forewing and the abdomen beneath it. This implies that somatotopic information from the wing is mapped to that of the abdomen.
We therefore conclude that the sensorimotor transformation underlying grooming in locusts shares two features with similar transformations in vertebrates: first, it is continuous within a receptive field that is larger than the area reached by a single grooming response. Second, it integrates somatotopic information from different body surfaces with proprioceptive information from the limb. As our kinematic analysis shows that these features also hold true in an invertebrate, we argue that they are key features of targeted movements of multi-jointed limbs in general. The locust provides a powerful model system, in which the combination of a computational framework (Dürr 2001), identifiable neuronal circuitry (Matheson 2002) and quantitative behavioral analyses allows us to study these universally important sensorimotor transformations at various levels.
|
DISCLOSURES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Address for reprint requests: V. Dürr, Abt. 4, Fakultät Biologie, Universität Bielefeld, Postfach 10 01 31, D-33501 Bielefeld, Germany (E-mail: volker.duerr{at}uni-bielefeld.de).
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Berkowitz A and Laurent G. Local control of leg movements and motor patterns during grooming in locusts. J Neurosci 16: 8067–8078, 1996a.
Berkowitz A and Laurent G. Central generation of grooming motor patterns and interlimb coordination in locusts. J Neurosci 16: 8079–8091, 1996b.
Berridge KC. Progressive degradation of serial grooming chains by descending decerebration. Behav. Brain Res. 33: 241–253, 1989.[ISI][Medline]
Berridge KC, Fentress JC, and Parr H. Natural syntax rules control action sequence in rats. Behav Brain Res 23: 59–68, 1987.[ISI][Medline]
Brunn DE and Dean J. Intersegmental and local interneurons in the metathorax of the stick insect Carausius morosus that monitor middle leg position. J Neurophysiol 72: 1208–1219, 1994.
Burrows M and Newland PL. Correlation between the receptive fields of locust interneurons, their dendritic morphology, and the central projections of mechanosensory neurons. J Comp Neurol 329: 412–426, 1993.[ISI][Medline]
Burrows M and Siegler MVS. Organization of receptive fields of spiking local interneurons in the locust with inputs from hair afferents. J Neurophysiol 53: 1147–1157, 1985.
Cruse H. The control of the anterior extreme position of the hind leg of a walking insect, Carausius morosus. Physiol Entomol 4: 121–124, 1979.
Cruse H, Kindermann T, Schumm M, Dean J, and Schmitz J. Walknet—a biologically inspired network to control six-legged walking. Neural Networks 11: 1435–1447, 1998.[ISI][Medline]
Dean J and Wendler G. Stick insect locomotion on a walking wheel: interleg coordination of leg position. J Exp Biol 103: 75–94, 1983.
Delcomyn F. Motor activity during searching and walking movements of cockroach legs. J Exp Biol 133: 111–120, 1987.
Duch C and Pflüger H-J. Motor patterns for horizontal and upside-down walking and vertical climbing in the locust. J Exp Biol 198: 1963–1976, 1995.[Abstract]
Dürr V. Stereotypic leg searching-movements in the stick insect: kinematic analysis, behavioral context and simulation. J Exp Biol 204: 1589–1604, 2001.[Abstract]
Dürr V and Matheson T. Functional somatotopy of targeted limb movements in an insect (Abstract). Proc Göttingen Neurobiol Conf 28: 319, 2001.
Eaton RC and Farley RD. The neural control of cercal grooming behaviour in the cockroach, Periplaneta americana. J Insect Physiol 15: 1047–1065, 1969.[ISI][Medline]
Fisher NI. Statistical Analysis of Circular Data. Cambridge, UK: Cambridge Univ. Press, 1993.
Giszter SF, McIntyre J, and Bizzi E. Kinematic strategies and sensorimotor transformations in the wiping movements of frogs. J Neurophysiol 62: 750–767, 1989.
Hedwig B and Heinrich R. Identified descending brain neurons control different stridulatory motor patterns in an acridid grasshopper. J Comp Physiol [A] 180: 285–294, 1997.
Honegger H-W, Reif H, and Müller W. Sensory mechanisms of eye cleaning behavior in the cricket Gryllus campestris. J Comp Physiol [A] 129: 247–256, 1979.
Matheson T. Hindleg targeting during scratching in the locust. J Exp Biol 200: 93–100, 1997.[Abstract]
Matheson T. Contralateral coordination and retargeting of limb movements during scratching in the locust. J Exp Biol 201: 2021–2032, 1998.[Abstract]
Matheson T. Metathoracic neurons integrating intersegmental sensory information in the locust. J Comp Neurol 444: 95–114, 2002.[ISI][Medline]
Matheson T and Dürr V. Load compensation in targeted limb movements of an insect. J Exp Biol 206: 3175–3186, 2003.
Mortin LI, Keifer J, and Stein PSG. Three forms of the scratch refllex in the spinal turtle: movement analyses. J Neurophysiol 53: 1501–1516, 1985.
). The body-centered coordinate frame of reference had its origin on the hind leg coxa, with the x axis passing through the front leg coxa. Grooming of a hind leg was elicited by brushing the ipsilateral front wing at arbitrary stimulus locations that were subsequently binned into 5 equal-sized, adjacent target sites and numbered from proximal to distal. The mean stimulus location for each bin is marked (
). Error bars depict SD across all trials having a given target site and start posture. The coordinates of 8 points on the body were measured from video recordings (B, 
, connected by —). The coordinates of the limb segments coxa, femur, tibia, and tarsus were analyzed to describe the grooming response. The wing axis was used to obtain the stimulus location along the wing. Grooming responses were analyzed with a time resolution of 20 ms (C, superimposed stick figures, each 1 containing the digitized points of a single frame) and the trajectories of all hind leg joints were measured (D, same sequence as in C). Straight line segments in D show the tarsus position in each frame.
, center of gravity; 
, maximum. D: same as C but for anterior target site 2. The greyscale codes the likelihood with which a particular point in the workspace will be groomed by the distal tibia, given a particular stimulus. Maximum indicates the most likely point in space to be groomed, and the center of gravity gives the best estimate of the center of the distribution.
, femur-tibia joint. Solid black and gray lines without symbols, the optimal angles of the coxa-trochanter and femur-tibia joints, respectively, based on the corresponding thorax-coxa angle. C: joint angles at the time of point of closest approach to the target for all grooming responses. The gray level of each symbol indicates the corresponding target location along the wing and corresponds to the shading of the wing in A. Solid lines depict wing midline and edges in joint-angle space. D: centers of gravity of the cyclic phase of individual grooming responses expressed in joint angle space. Open symbols show corresponding data from Fig. 5E of Berkowitz and Laurent (1996a