| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
INNOVATIVE METHODOLOGY
1School of Applied Physiology, Center for Human Movement Studies, Georgia Institute of Technology, Atlanta, Georgia; and 2Barrow Neurological Institute, St. Joseph's Hospital and Medical Center, Phoenix, Arizona
Submitted 9 July 2004; accepted in final form 6 May 2005
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
), exertion of isometric forces (Ashe 1997), and reaching by the arm (Georgopoulos et al. 1982). These types of studies have revealed that the activity of the neuronal population of the motor cortex is related to different mechanical variables of motor tasks. For example, Evarts (1968) demonstrated in monkeys executing simple wrist movements that the activity of some neurons in the motor cortex was related to the activation of wrist flexor and extensor muscles. Motor cortex activity has also been found to be well correlated with other movement variables, e.g., the direction and velocity of movement (Georgopoulos et al. 1982; Moran and Schwartz 1999; respectively), the direction and magnitude of the static force exerted by the arm (Ashe 1997), the hand location in space (Sergio and Kalaska 1997), the arm orientation (Scott and Kalaska 1997), and the joint torque and power (Scott et al. 2001).
Mechanically constrained paradigms employed in the abovementioned studies have allowed for the accurate and detailed determination of mechanical aspects of the task and their comparison with motor cortex activity. The motor cortex may also encode some parameters of multi-joint, full-body behavior (e.g., Donchin et al. 2002; Kalaska and Drew 1993). Simultaneous recording of motor cortex activity and body biomechanics in freely moving animals is required to test this suggestion, however.
Methods for recording the activity of the motor cortex in freely moving animals have been developed (Armstrong and Drew 1984; Beloozerova and Sirota 1986; Buzsaki et al. 1989; Girman 1973). Using these methods in cats revealed that the activity of most pyramidal tract neurons (PTNs) during locomotion on a flat surface is rhythmically modulated with respect to phases of the step cycle. A detailed movement analysis is more challenging in freely moving animals than during tasks with a limited number of degrees of freedom. For this reason, motor cortex activity during unconstrained movements typically has been analyzed in conjunction with muscle activation and simple kinematics of the contralateral limb (e.g., Drew 1988).
We believe that simultaneous recordings and analyses of motor cortex activity and full-body biomechanics during unconstrained animal behavior will provide a better understanding of motor cortex functions. PTNs, which have widespread connections onto the interneurons and motoneurons of a number of muscles of the contralateral limb (Georgopoulos and Grillner 1989; Lawrence et al. 1985; Leblond et al. 2001; Shinoda et al. 1986), are likely to control different aspects of movement through complex spinal networks. In addition, extensive branching of corticospinal axons in the brain stem and spinal cord (Armand 1982; Armand et al. 1985; Canedo 1997; Futami et al. 1979) suggests that PTN activity may affect not only the motor aspects of the contralateral limb but also those of the other three limbs. Examples of such effects might include full-body postural responses to voluntary gait modifications made by one limb during obstacle overstepping (Lavoie et al. 1995) and posture correction responses on a tilting surface (Beloozerova et al. 2005). Recently, encoding and decoding of global movement variables have been reported for neurons of the dorsal spinocerebellar tract (Bosco and Poppele1997) and sensory neurons in the L6 and L7 dorsal root ganglia (Stein et al. 2004) of the cat. Thus it seems logical to examine the correlation among the activity of neurons of the major cortical descending tract, the PTNs, and global aspects of movement, including kinetic variables (i.e., joint moments and power) that appear to be important for interpreting motor cortical activity (Scott et al. 2001). Experimental methods that can provide detailed full-body movement analyses in conjunction with recordings of motor cortex activity are necessary for this.
In this report, we describe a new experimental paradigm that combines in vivo recordings of full-body kinematics and kinetics with high quality, single-unit extracellular recordings from the motor cortex in the cat. This paradigm allowed for direct comparisons between a variety of mechanical variables and cortical activity in unconstrained conditions. We hypothesized that such comparisons may reveal new relationships between the activity of the motor cortex and mechanical aspects of locomotor behaviors.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Locomotion task
Positive reinforcement (food) was used to habituate cats to the experimental situation and to engage them in locomotion behavior (Pryor 1975; Skinner 1938). During the first phase of training, cats were fed their total daily ration in the experimental walkway (Fig. 1) and were allowed to explore the walkway at their own pace. The walkway was 3.0 m long and 0.4 m wide, and a black rubberized material (NovaMat) covered the floor. The walkway had clear Plexiglas walls (height: 0.6 m, thickness: 0.005 m) along both long sides and wooden walls on the short sides at each end. A clear Plexiglas sheet in the middle of the walkway separated it into two corridors along its length (Fig. 1, A and B). When animals exhibited content and confident behavior in the walkway, the second phase of the training began. Food was associated with a brief audible signal (5 kHz) during this phase. When the cats clearly showed an orientation toward the sound, training proceeded to the third phase, in which walking from one end of the walkway to the other along one corridor and returning using the other corridor was associated with the audible signal and food reward.
|
After each cat was trained to walk in the walkway, surgery was performed under aseptic conditions using isoflurane anesthesia. The methods for introduction of recording and stimulation electrodes into the brain have been, in part, reported earlier (Beloozerova and Sirota 1986, 1993; Beloozerova et al. 2003; Sirota et al. 1988). After immobilization of the cat's head in a stereotaxic head holder, a longitudinal incision was made in the skin on the top of the head. The skin was then retracted to the sides of the head. The fascia was removed from much of the dorsal surface of the skull. At 10 points around the circumference of the head, holes of 1.5 mm diam were drilled approximately orthogonal to the surface of the skull, and stainless steel screws were inserted into these holes (Fig. 2, A and B). To protect the dura from penetration, the screws had a restraining shoulder 4 mm from the tip, and the tips of the screws were rounded (Fig. 2C). The screws protruded below the inner surface of the skull by <1 mm. The heads of the screws extended 15 mm above the surface of the skull and were soldered to a copper wire; this apparatus was used as an electrical zero point while recording the activity of cortical neurons and limb muscles. The screw heads and the wire were inserted into a plastic cast to form a circular base (Fig. 2A, 1). This base later allowed the head of an awake animal to be fixed in a head-restraining device so that neurons could be isolated before recording their activity during locomotion. Such fixation does not damage the skull or surrounding tissues. The firmness of the fixation of the base to the skull was mainly determined by the radial location of the screws around the skull, which minimize the stress applied along the axis of any one screw. The top portion of the base was also used to affix connectors, miniature microdrives, preamplifiers, and a protective and electrically shielding cap (Fig. 2D).
|
1.0 cm2 were removed (Fig. 2A, 5). The motor cortex was visually identified by surface features and photographed. The aperture was then covered by an acrylic plate (Fig. 2E, 1) 1 mm thick in which 200 holes, 0.3 mm in diameter, had been drilled and filled with a mixture of sterile bone wax and petroleum jelly. The plate was fixed to the surrounding bone with acrylic cement. Recording microelectrodes were later inserted into the motor cortex of the awake animal through the holes in the plate. To access the pyramidal tract with stimulating electrodes, two 26-gauge hypodermic tubes were implanted above the medullary pyramids at the Horsley and Clarke coordinates (P 10; L 0.5 and P 9.3; L 1.2; Fig. 2A, 6) at a depth of H 0. Stimulating electrodes (200 µm OD), constructed of platinum-iridium wire insulated with Teflon to within 0.4 mm of the tip were introduced into the medullary pyramid through the guide tubes several days after the surgery in awake animals under physiological guidance (see following text).
Electromyographic (EMG) electrodes were constructed from Teflon-insulated multi-strand stainless-steel wire (AS633, Cooner Wire). A pair of these wires was inserted into each of the following muscles of the right fore- and hindlimbs (contralateral to the motor cortex): m. brachialis (BR), the lateral head of the m. triceps brachii (TR), m. tibialis anterior (TA), and m. soleus (SO). Electrode placement was verified by stimulation through the implanted wires before closure of the incision and also later during the terminal experiment. The electrode leads were led subcutaneously to the inside of the head base.
All exposed surfaces of the skull were covered with a thin layer of orthodontic resin (Densply Caulk). The space between the skull, the retracted skin, and the bottom of the base was filled with silicone elastomer (Factor II) to form a soft barrier between the inside and outside of the base. A protective and electrically shielding cap was fixed on the top portion of the base by bolts (Fig. 2D). The cat was shaved. Permanent ink marks were made on the skin above all major limb joints. The animal was then allowed to recover for several days.
Sampling of neuronal and muscle activity
Several days after the surgery, a cat was placed on a table equipped with a comforting pad and encouraged to take a "sphinx" position. After the cat rested in this posture for several minutes, the base attached to the skull during surgery was fastened to an external frame so that the resting position of the head was approximated. This procedure minimized stress on the neck while the head was immobilized. After a few training sessions, all cats sat quietly with their head restrained. They did not seem to be disturbed as they frequently fell asleep.
Before the first recording session, connectors for neuronal and EMG preamplifiers, sockets for stimulating and EMG electrodes, and the main output connector were positioned permanently on the caudal portion of the base (Fig. 2D). One or two manual microdrives (Fig. 2E) with dimensions 5 x 5 x 30 or 2 x 3 x 10 mm were also attached to the base. Each micromanipulator had an arm (Fig. 2E, 4), the position of which over the motor cortical plate (Fig. 2E, 1) could be adjusted by bending it or by changing its position on the moving nut (Fig. 2E, 10) of the micromanipulator.
A conventional extracellular tungsten varnish insulated microelectrode (50–125 µm OD, FHC) or a quartz insulated microelectrode with platinum-tungsten core (40 µm OD) (Reitboeck 1983) connected to a micromanipulator was inserted into the motor cortex through a hole in the plastic plate and advanced into the cortical tissue (Fig. 2E). After the electrode reached the cortical depth where clear responses of many neurons to limb movements could be observed (presumably layer V), two platinum-iridium wires (200 µm OD, insulated with Teflon to within 0.4 mm of the tip) were slowly lowered into the medullar pyramid through the guide tubes implanted during surgery. Pulses of graded intensity (in the range of 0.1–0.5 mA, 0.2-ms duration), were delivered through this bipolar electrode. The wires were fixed at the positions that most effectively elicited antidromic responses in motor cortex neurons, and they served as the pyramidal tract-stimulating electrodes for the remainder of the experiments with the subject. The criterion for identification of antidromic activation was the test for collision of spontaneous and evoked spikes (Bishop et al. 1962; Fuller and Schlag 1976).
Neuronal and muscle activity were preamplified using miniature preamplifiers positioned on the head base. The activity was additionally amplified and filtered (300- to 10,000-Hz band-pass for neurons and 30- to 1,000-Hz band-pass for EMG) using a CyberAmp 380 (Axon Instruments) amplifier. Neuronal activity was sampled at 30 kHz and muscle activity at 3 kHz. All signals were displayed on a computer monitor and archived on a computer hard drive in a single file. A waveform analysis was employed on- and off-line to discriminate between and identify the spikes of the neurons using the Power1401/Spike2 system (Cambridge Electronic Design, Cambridge, UK) waveform-matching algorithm. The identities of the neurons were periodically verified during locomotion using the collision test. A representative record of the discharges of two neurons projecting to the pyramidal tract (PTNs) along with the EMG activity of selected muscles during six walking steps is shown in Fig. 3A.
|
Light-reflecting markers (diameter: 6–10 mm) were placed on the following previously marked spots on the cat's body using sticky double sided tape: ventral iliac spine, greater trochanter, approximate knee joint center, lateral malleolus, base of the fifth metatarsal, end of the distal phalanx of the two hindlimbs, vertebral border of the scapula, the greater tubercle, approximate elbow joint, lateral styloid process, base of the fifth metacarpal, and end of the distal phalanx of the two forelimbs (Fig. 4). Also two markers were placed on each side of the head at approximate intersections of the long axis of the head and perpendiculars from the lateral corner of the eye and the anterior edge of the ear (Fig. 4). The midpoint between the two markers corresponds to the center of mass location of the head (Hoy and Zernicke 1985).
|
Ground reaction force recording
Two miniature force platforms (Broker and Gregor 1990; Fowler et al. 1993; Gregor et al. 2001) were concealed in the center of the right side of the walkway (Fig. 1A). Mechanically separated from the surrounding walkway floor, the force platforms measured the vertical, anterior-posterior, and medial-lateral components of the ground reaction force vector and the coordinates of its point of application. This information was used as input to the inverse dynamics analysis of fore- and hindlimb movements. Prior to the measurements, the force platforms were calibrated using known weights. The sampling rate of the ground reaction forces was 1,000 Hz.
Video recordings, ground reaction forces, and cortex and muscle activity recordings were synchronized by a common electronic impulse that appeared on both video and analog records.
Terminal experiment and histological procedures
At the termination of the experiments, the cats were deeply ansthetized with pentobarbital sodium, and several reference electrolytic lesions were made in the recording area in the left motor cortex. Positions of EMG electrodes in the muscles were verified. Segments' lengths as defined in Hoy and Zernicke (1985) were measured using an anthropometer. The cats were killed by overdose of pentobarbital sodium and perfused with isotonic saline followed by a 10% Formalin solution. Frozen sections of 50-µm thickness were cut in the regions of the motor cortex and medullary pyramid. The tissue was stained for Nissl substance with cresyl violet. Positions of tracks in the cortex were estimated in relation to the reference lesions. Stimulating electrode locations in the medullary pyramid were verified.
Inverse dynamics analysis
Fourteen markers on each side of the body were digitized using the Peak Performance Motion Analysis System and its software. We developed software for a full-body inverse dynamics analysis. The marker coordinates were filtered using a fourth order, zero phase lag, Butterworth low-pass filter. The cutoff frequency of the filter was determined separately for the horizontal and vertical coordinates of each marker using Fourier (harmonic) analysis and the method described by Lanczos (1956), which is based on the different properties of the signal (true coordinates of a marker) and noise. The cutoff frequencies found were between 4 and 10 Hz. After filtering marker coordinates, the coordinates of the knee and elbow joints were calculated using the coordinates of the ankle and hip, and the wrist and shoulder markers, respectively, and the lengths of the forming segments that were measured during the terminal experiment. This was done to correct the recorded knee and elbow marker coordinates, which are significantly influenced by skin movements (Goslow et al. 1973). Velocity, acceleration, and jerk of the markers were calculated using numerical differentiation (Lanczos 1956).
Other kinematic variables also calculated for each body segment included the coordinates of the segments' center of mass locations; their velocities and accelerations; segment angles with respect to the vertical axis and their velocities and accelerations; and joint angles and their velocities. To calculate the center of mass location of each segment, the relative location of the center of mass reported by Hoy and Zernicke (1985) for major segments of the cat fore- and hindlimbs were used. The position, velocity, and acceleration of the general center of mass of the cat body were subsequently calculated.
Resultant joint forces, moments, and their power at four joints of each limb during motor behaviors were calculated using an inverse dynamics analysis typically employed for kinetic analyses of a single limb (Fowler et al. 1993; Gregor et al. 2001; McFadyen et al. 1999; Prilutsky and Zatsiorsky 1994; Scott et al. 2001). A two-dimensional model of the cat body consisting of 21 rigid-body segments (Fig. 4) was used for the calculations. The segments were assumed to have constant inertial properties and to be connected by frictionless hinge joints. The mass and moment of inertia of each segment were calculated from the total body mass and segments' length using the empirical regression equations of Hoy and Zernicke (1985). Knowing the position and acceleration of the limb at each time instant, the length, mass, moment of inertia, center of mass location of each segment, and the magnitude, direction, and point of application of the external force, the resultant joint forces and moments at the metatarsophalangeal/metacarpalphalangeal, ankle/wrist, knee/elbow, and hip/shoulder joints were calculated (Fowler et al. 1993; Gregor et al. 2001; McFadyen et al. 1999; Prilutsky and Zatsiorsky 1994). Joint power at each joint was calculated as the product of joint moment and the corresponding joint velocity. In total, 295 mechanical variables were computed as a function of time.
Comparison between motor cortex activity and movement biomechanics
Many computational methods have been used to investigate whether and how cortex activity is related to movement characteristics (e.g., Drew and Doucet 1991; Evarts 1968; Georgopoulos et al. 1982; Kakei et al. 1999; Moran and Schwartz 1999 and other authors). One of these methods, the cross-correlation analysis (for a recent review of its neurophysiological applications, see Hamm et al. 2001), in particular allows for assessing the strength of relationships between time-dependent activity of the PTN and time-dependent mechanical variables and a time delay between them (Miller and Houk 1995; Schwartz and Adams 1995). To demonstrate how motor cortex activity can be compared with full-body mechanics, we employed a related statistical analysis, a multivariate linear regression analysis, which has several advantages. First, it takes into account possible correlations among the large number of mechanical variables obtained in this study. Second, it allows for potential predictions of PTN activity as a function of biomechanical variables (neural encoding) or for reconstructions of biomechanical variables from the PTN ensemble activity (neural decoding) (e.g., Stein et al. 2004).
To conduct the multivariate regression analysis on neural and biomechanical data, histograms of the activity of 43 single PTN neurons were calculated for each analyzed step cycle of locomotion using a 10%-step cycle bin width. The onset of the swing phase of the right forelimb determined from ground reaction force and/or video recordings was taken as the beginning of the step cycle. The duration of each step cycle was divided into 10 equal bins, and the number of spikes in each bin was counted. The discharge frequency in a bin was derived according to the method of Udo et al. (1982), which averages the instantaneous frequency of interspike intervals that fall within the bin and also accounts for those intervals that overlap with the bin's beginning and end. Subsequently, the histograms were averaged over 20–40 step cycles (see, for example, Fig. 3B) to obtain a resultant histogram. All kinematic, kinetic, and EMG variables of cat locomotion (299 in total) were also averaged over the same 10 bins as PTN activity.
Given the number of the PTN histograms (43) and averaged biomechanical variables (299), it was expected that some variables might be inter-correlated, i.e., the data set could be redundant. The degree of redundancy was determined by a principal component analysis of the 342 10-bin patterns with a variance maximizing rotation, which also classified the variables according to their patterns. The eigenvectors (principal components) were extracted according to Cattell's criterion using STATISTICA (StatSoft) software. The correlations calculated between each variable and the extracted factors (factor loadings) were considered significant when they exceeded a value of 0.7.
PTN activity encoding and decoding during walking was analyzed using multivariate linear regression models similar to those of Stein et al. 2004. The average firing rate (fi) during a step cycle of the ith PTN was calculated as a linear function of averaged mechanical variables (gj, j = 1,2,...,n) (PTN activity encoding): fi = a0i + a1ig1 + a2ig2 +... + anign. The averaged magnitude of selected biomechanical variables (gj) in a step cycle was calculated as a linear function of ensembles of PTN firing rates (PTN activity decoding) : gj = b0j + b1jf1 + b2jf2 +... + bmfm. The regression coefficients a and b were calculated using the forward entry method of stepwise linear regression, which finds the best subsets of independent variables (of 299 for encoding and 43 for decoding) based on a correlation coefficient value. No time lag between PTN activity and mechanical variables was incorporated in these regression models because the bin duration (60–75 ms) exceeded the minimum time required for microstimulation of the motor cortex to modify EMG of the forelimb muscles in the walking cat (11–14 ms) (Armstrong and Drew 1985) or the time required to initiate the early PTN responses to unexpected perturbation of locomotion (20 –100 ms) (Marple-Horvat et al. 1993).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Full-body biomechanics of walking
SELECTED KINEMATICS.
During walking, all cats demonstrated typical patterns of stance-swing phase sequences (2 phases for each leg in a cycle, 8 phases in total; Figs. 5–7, 1 and 8). Depending on walking speed, the number of feet on the ground at the same time during the eight phases of a cycle was 2-3-4-3-2-3-4-3 at speeds 0.5 m/s and below (Figs. 5–7) and was 3-2-3-2-3-2-3-2 at faster walking speeds (not shown) (see Prilutsky et al. 2001).
|
|
; Kuhtz-Buschbeck et al. 1994; McFadyen et al. 1999; Shen and Poppele 1995; Smith et al. 1998). For example, the ankle, knee, and shoulder joints yielded in the first half of stance and extended in the second half of stance. In swing,these joints flexed first and extended second (Figs. 5 and 6, 2). Thus joint angle patterns at the ankle, knee, and shoulder joints had two flexion and extension peaks in the step cycle. Metatarsophalangeal (MTP), metacarpophalangeal (MCP), and wrist joint angles had one flexion peak at the end of stance and one extension peak in the middle of swing. Hip joint angles reached their maximum flexion in the middle of swing and maximum extension at the end of stance (Figs. 5 and 6, 2).
|
The jerk (the 3rd time derivative of displacement) was also calculated. For example, the horizontal jerk of the forelimb paw had two positive and negative peaks in the step cycle. The largest, negative peak approaching –300 to –400 m/s3 occurred approximately in the middle of the swing phase and coincided with joint angle peaks (Figs. 5 and 6, 3).
Positive peaks of vertical velocity of the general center of mass appeared to coincide with mid-stance of hindlimbs, whereas negative peaks occurred in late stance (Fig. 7, 4 and 8). Patterns of horizontal velocity of the general center of mass were less stereotyped and dependent on whether the cat maintained a constant average speed of progression.
SELECTED KINETICS.
The horizontal and vertical components of the ground reaction forces applied to the hind and fore digits had patterns typical for mammalian locomotion (Fig. 7, 5). During the entire stance phase the vertical force was positive and reached values of 13–16 N (for hindlimbs) and 18–22 N (for forelimbs). The horizontal force component was negative during the first half of stance (i.e., contributed to deceleration of the body) and positive in the second half. Horizontal force peaks were between ±2 and ±6 N depending on locomotion speed. The higher forelimb vertical forces compared with the hindlimb forces (Fig. 7, 5) are consistent with literature reports and result from the general center of mass being located closer to the forelimbs (see also Alexander 1980; Manter 1938; Pandy et al. 1988).
Because of higher forelimb forces, the absolute values of joint moment peaks were generally larger for forelimbs than hindlimbs (Figs. 5 and 6) (see also Manter 1938; Pandy et al. 1988). Patterns of moments and powers at hindlimb joints (Fig. 6) are also consistent with those reported for locomotion in cats by Fowler et al. (1993), McFadyen et al. (1999), and Gregor et al. (2001). In particular, values of moments and power at the hindlimb joints (Fig. 6, 4 and 5) were relatively small during the swing phase of locomotion. During initial stance, all joints exerted extensor moments, and the MTP, ankle and knee joints absorbed mechanical energy (joint power was negative). In the second part of stance, all joints except the hip generated positive power contributing to cat propulsion.
Similar patterns of joint moments and power were observed at the forelimb joints (Fig. 5, 4 and 5). The MCP, wrist, and elbow exerted mostly extensor moments with their peaks around mid-stance. The shoulder moment changed direction from extensor to flexor in the middle of stance, whereas the elbow moment switched from extensor to flexor at the end of stance. Among the forelimb joints, the elbow contributed the most to positive power generation and thus to forward propulsion (Fig. 5, 5).
EMG ACTIVITY. Increased activity of an elbow extensor, the lateral head of the m. triceps brachii (TR), coincided with an extensor elbow moment during stance, suggesting its agonist role at this joint in stance. An elbow flexor, m. brachialis (BR), was active at the end of stance and early swing, which also coincided with a flexor elbow moment (Fig. 5, 6). The ankle extensor m. soleus (SO) was primarily active during the generation of the extensor (plantar flexor) moment at the ankle in stance (Fig. 6, 6).
Patterns of neural and biomechanical variables during walking
The principal component analysis conducted on PTN activity histograms and on biomechanical variables averaged over 10 bins of a step cycle revealed four factors that accounted for 91% of the variance (Table 1). As a result, many neural and biomechanical variables were classified by the analysis into one of the four groups. Variables within each group had high factor loadings (>0.7), i.e., high negative or positive correlations with the corresponding eigenvectors. The first group accounted for 34% of the variance (Table 1) and included histograms of 13 PTNs and 106 biomechanical variables (e.g., segments' angular orientations, BR EMG, forelimb joint angles and moments, etc.) the patterns of which had one broad maximum/minimum in swing/stance (Fig. 8A). The second factor accounted for 27% of the variance (Table 1) and consisted of histograms of 8 PTNs and 69 biomechanical variables (e.g., ground reaction force, joint velocities, etc.) with a maximum/minimum at the stance-swing transition (Fig. 8B). Variables corresponding to the third factor [2 PTNs and 16 biomechanical variables: e.g., toe jerks, displacement of the general center of mass (GCM), etc.] accounted for 16% of the variance and had a more complex pattern: two maximums and two minimums in early/late swing and stance (Fig. 8C). The fourth factor included 1 PTN and 28 mechanical variables (e.g., joint velocity, linear velocity of the GCM) and also had two minimums and maximums in the step cycle (Fig. 8D). This factor accounted for 14% of the variance (Table 1).
|
|
Encoding and decoding mechanical variables in/by PTN activity
The ensemble activity of different groups of a relatively small number of PTNs (typically between 2 and 8) could predict patterns of any biomechanical variable with very high accuracy (R2 = 0.99). Selected examples of predicted variables (toe jerk of the right forelimb, left shoulder moment, triceps EMG) are shown in Fig. 9, A–C. Neuron ensembles decoding biomechanical variables typically included at least one neuron from the group to which the biomechanical variable belonged (see Fig. 8). Each PTN contributed to decoding of at least one biomechanical variable. No other trends in neural decoding were apparent.
|
Thus the 10-bin pattern of any biomechanical variable of walking obtained in this study could be predicted by ensemble activity of a relatively small number of PTNs. Likewise, the 10-bin firing rate pattern of any of the 43 PTNs recorded here could be predicted by a relatively small number of biomechanical variables.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Biomechanics of full-body locomotion
Although the inverse dynamics analysis employed here is well established, in studies of quadrupedal locomotion, it has been primarily performed either on one hindlimb (Fowler et al. 1993; Gregor et al. 2001; McFadyen et al. 1999) or on the hind- and forelimb of one side of the body (Manter 1938; Pandy et al. 1988). Only a limited number of biomechanical variables from the 299 obtained in this study can be found in and compared with the published data. This comparison demonstrated a general agreement with the published data (see RESULTS), but some differences exist.
In accordance with the results of Manter (1938), our positive and negative peaks of horizontal accelerations of hindlimb segments occurred in late and early stance, respectively. However, contrary to Manter's data, our results suggest a clear dependence of acceleration peaks on segment proximity (the more distal the segment, the higher the acceleration peak; Fig. 7, 3, and Manter's Fig. 5D). Also, Manter's horizontal acceleration peaks (±10 m/s2 for the shank and thigh and ±20 m/s2 for the hind foot) are substantially higher than ours (±3 to ±12 m/s2, Fig. 7, 3) for about the same walking speeds. Different methods of estimating the time derivatives probably cause these differences. Manter used graphic differentiation, which is not an accurate method for obtaining velocities and accelerations (Lanczos 1956). This also can explain differences in peak values of segments' angular velocities (see our data in Fig. 7, 2; Manter's are in Fig. 6A of his paper). The patterns of angular segment velocities were similar, however; peak positive values occurred in the middle of swing.
The pattern of vertical velocity of the center of mass presented in Fig. 7, 4, is similar to that reported by Manter, but he did not determine the coordinates of the center of mass directly. He used a marker on the trunk near the assumed center of mass instead.
Patterns of neuronal activity and biomechanical variables and their correlations
Which mechanical or physiological variables are encoded by motor cortex activity has been a central question in motor cortex physiology for decades. Many researchers have correlated cortex activity with different mechanical parameters and muscle activity (Evarts 1968; Georgopoulos et al. 1982; Holdefer and Miller 2002; Houk et al. 1987; Moran and Schwartz 1999; Reina et al. 2001; Scott and Kalaska 1997; Scott et al. 2001; Sergio and Kalaska 1997). Because many physiological (e.g., EMG) and mechanical variables (e.g., joint moments, limb velocity and acceleration) are correlated with each other (as also demonstrated in this study; Fig. 8), it is not easy to elucidate the primary variables encoded in motor cortex activity. This might partly explain why this problem remains unresolved despite significant effort. In addition, the motor cortex contains groups of cells that appear to encode different aspects of motor behavior, including nonmotor aspects, in both extrinsic and different intrinsic (joint angle, muscle) coordinates (Alexander and Crutcher 1990; Carpenter et al. 1999; Kakei et al. 1999). Some of these cells might be among the 19 PTNs not classified into the four groups by the principal component analysis in this study. Still another problem is that the discharge of some PTNs appears to encode different parameters under different conditions (e.g., Beloozerova et al. 2005; Kakei et al. 2003; Sergio and Kalaska 2003; Thach 1978).
Despite their limitations, correlation and in particular multivariate regression analyses (Hamm et al. 2001; Holdefer and Miller 2002; Houk et al. 1987; Schwartz and Adams 1995; Stein et al. 2004) have been used extensively and have provided valuable information regarding representations of movement variables in motor cortex activity.
The developed method of simultaneously recording PTN activity and full-body mechanics provided new information on possible relationships between motor cortex activity and movement biomechanics. The fact that 73% of the full-body mechanical variables were highly correlated with one of the four eigenvectors, as revealed by the principal component analysis (Fig. 8), indicates that these patterns are redundant, i.e., many of the variables are closely correlated. As a consequence, a comparison of a limited number of mechanical variables with neural activity might yield misleading results. On the other hand, activation patterns of only 56% of sampled PTNs could be related to the four eigenvectors. The remaining 44% of neurons have patterns that differ from the majority of full-body mechanical variables and might encode information that is not directly related to motor patterns. These PTN activation patterns could also be related to muscles not sampled in this study.
Given the great number of mechanical variables obtained in this study, perhaps it is not surprising that it was possible to find subsets of variables that could accurately predict the firing rate of each sampled PTN. Accurate predictions of mechanical variables by the ensemble activity of a small number of PTNs could also be explained by the majority of these variables having relatively simple patterns (with 1–3 peaks per cycle). However, whether or not these predictions are real and not caused by chance will require further studies.
Limitations and advantages of the method
The main limitations of the described experimental approach are the high complexity of the employed experimental procedures and the huge volume of information to be analyzed. The experimental procedures involve a combination of two innovative methods, analysis of full-body mechanics, and recording of PTN activity in unrestrained animals and thus require expertise in both areas of research. The motion analysis requires digitizing the coordinates of 28 markers on the animal body and is time consuming, but it can be substantially simplified by using motion analysis systems with automatic digitizing capabilities, which are commercially available. The results presented in this report show that despite the limitations, the new experimental approach is manageable and can yield unique and useful data on the functions of PTNs.
In particular, the developed method allowed us for the first time to compare the kinetics (ground reaction forces, joint forces, moments, power) of the contralateral limb with PTN activity during unconstrained cat locomotion, to obtain and compare PTN activity with the kinematics and kinetics of the other three limbs, and to obtain and compare PTN activity with movement characteristics of the general center of mass in the cat. These new developments have the potential to reveal new information about the functions of the motor cortex. We wish to stress, however, that any single method must be complemented with other techniques to obtain any definite information regarding physiological functions.
|
GRANTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: B. I. Prilutsky, School of Applied Physiology, Georgia Institute of Technology, 281 Ferst Dr., Atlanta, GA 30332-0356 (E-mail: boris.prilutsky{at}ap.gatech.edu)
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Alexander RM. Optimum walking techniques for quadrupeds and bipeds. J Zool 192: 97–117, 1980.
Armand J. The origin, course and terminations of corticospinal fibers in various mammals. Prog Brain Res 57: 329–360, 1982.[ISI][Medline]
Armand J, Holstege G, and Kuypers HGJM. Differential corticospinal projections in the cat. An autoradiographic tracing study. Brain Res 343: 351–355, 1985.[CrossRef][ISI][Medline]
Armstrong DM and Drew T. Discharges of pyramidal tract and other motor cortical neurons during locomotion in the cat. J Physiol 346: 471–495, 1984.
Armstrong DM and Drew T. Electromyographic responses evoked in muscles of the forelimb by intracortical stimulation in the cat. J Physiol 367: 309–326, 1985.
Ashe J. Force and the motor cortex. Behav Brain Res 86: 1–15, 1997.[CrossRef][ISI][Medline]
Beloozerova IN and Sirota MG. Assembly for neuronal activity recording in freely moving cat [in Russian]. Zhurnal Vysshei Nervnoi Deyatelnosti 6: 1114–1117, 1986.
Beloozerova IN and Sirota MG. The role of motor cortex in control of locomotion. In: Stance and Motion. Facts and Concepts, edited by Gurfincel VS, Ioffe ME, Massion J, and Roll JP. New York: Plenum, 1988, p. 163–176.
Beloozerova IN and Sirota MG. The role of the motor cortex in the control of accuracy of locomotor movements in the cat. J Physiol 461: 1–25, 1993.
Beloozerova IN, Sirota MG, and Swadlow HA. Activity of different classes of neurons of the motor cortex during locomotion. J Neurosci 23: 1087–1097, 2003.
Beloozerova IN, Sirota MG, Orlovsky GN, and Deliagina TG. Activity of pyramidal tract neurons in the cat during postural corrections. J Neurophysiol 93: 1831–1844, 2005.
Bishop PO, Burke W, and Davis R. The identification of single units in central visual pathways. J Physiol 162: 409–431, 1962.
Bosco G and Poppele RE. Representation of multiple kinematic parameters of the cat hindlimb in spinocerebellar activity. J Neurophysiol 78: 1421–1432, 1997.
Broker JP and Gregor RJ. A dual piezoelectric force pedal for kinetic analysis of cycling. Internat J Sports Biomech 6: 394–403, 1990.
Buzsaki G, Bickford RG, Ryan LJ, Young S, Prohaska O, Mandel RJ, and Gage FH. Multisite recording of brain field potentials and unit activity in freely moving rats. J Neurosci Methods 28: 209–217, 1989.[CrossRef][ISI][Medline]
Canedo A. Primary motor cortex influences on the descending and ascending systems. Progr Neurobiol 51: 287–335, 1997.[CrossRef][ISI][Medline]
Carpenter AF, Georgopoulos AP, and Pellizzer G. Motor cortical encoding of serial order in a context-recall task. Science 283: 1752–1757, 1999.
Donchin O, Gribova A, Steinberg O, Mitz AR, Bergman H, and Vaadia E. Single-unit activity related to bimanual arm movements in the primary and supplementary motor cortices. J Neurophysiol 88: 3498–3517, 2002.
Drew T. Motor cortical cell discharge during voluntary gait modification. Brain Res 457: 181–187, 1988.[CrossRef][ISI][Medline]
Drew T. Motor cortical activity during voluntary gait modifications in the cat. I. Cells related to the forelimbs. J Neurophysiol 70: 179–199, 1993.
Drew T and Doucet S. Application of circular statistics to the study of neural discharge during locomotion. J Neurosci Methods 38: 171–181, 1991.[CrossRef][ISI][Medline]
Evarts EV. Relation of pyramidal tract activity to force exerted during voluntary movement. J Neurophysiol 31: 14–27, 1968.
Fowler EG, Gregor RJ, Roy RR, and Hodgson JA. Relationship between ankle muscle and joint kinetics during the stance phase of locomotion in the cat. J Biomech 26: 465–483, 1993.[CrossRef][ISI][Medline]
Fuller JH and Schlag J. Determination of antidromic excitation by the collision test: problems of interpretation. Brain Res 122: 283–298, 1976.
Futami T, Shinoda Y, and Yokota J. Spinal axon collaterals of corticospinal neurons identified by intracellular injection of horseradish peroxide. Brain Res 164: 279–284, 1979.[CrossRef][ISI][Medline]
Georgopoulos AP and Grillner S. Visuomotor coordination in reaching and locomotion. Science 245: 1209–1210, 1989.
Georgopoulos AP, Kalaska JF, Caminiti R, and Massey JT. On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. J Neurosci 2: 1527–1537, 1982.
, positions of reflective markers on each side of the body. Arcs connecting adjacent segments define joint angles.
