Subjects
Our data were collected from two male rhesus (Macaca mulatta) monkeys (Monkey H and Monkey V, 4–6 years old) to identify the cortical areas related to lower limb voluntary movements. The animal behavioural training and the actual experimental protocols had been described in detail in our previous paper [34]. The experiments were complied with NIH policy on Humane Care and Use of laboratory animals, and were approved by the institutional Animal Care and Use Committee, Arizona State University, USA.
Experimental protocol
Each monkey was trained to perform a visually-cued stand and squat task using a well-designed primate chair. The chair was made of stainless steel and acrylic plastic, equipped with a movable pedal. While the head and body were restrained in the chair, the subject can still perform stand and sit movement, via pushing down the pedal with adjustable weight. The pedal was connected to the chair with two inner-set Axletrees, which made it possible to move up and down easily with negligible friction. When the subject wanted to stand up, the pedal can be pushed down to achieve the upright standing posture. When the subject intended to sit down and release the pushing force, the pedal was pulled up by the counter weight through the sliding axles, which were set on the back part of the chair.
The subjects performed the sit and stand task in a virtual reality environment. Figure 1 showed the procedure of the behaviour task in a typical successful trial. A LED marker placed at the ankle was represented as a red ball in the animation environment. Each trial was preceded by an inter-trial interval (varied randomly from 5 to 10 seconds) when the screen was illuminated with bright blank scene to prevent dark adaptation since the room was in low light condition. A trial was commenced with appearance of a green box representing the starting position (Center On). The subject was required to move his ankle position to match the green box (Center Hit) for 100 msec. After that, another green ball (Target) was displayed on the top of the screen (Target on). The subject was required to push down the pedal to match the ankle cursor with target position. Then the box turned back to green (Center Release). The subject was trained to fully extend his legs to hit the green ball (Target hit) for up to 400 ms and the counterweight matches the body weight to simulate the standing. Then, the subject was required to retract both legs back to the starting position (Target release). When animal fully retracted both legs, the cursor hit the centre box again, and the box turned into red (Center hit again). After another 100 ms, both of the center box and the target ball disappeared indicating the end of a trial (Center and Target off). The subject received his reward. The process entered into another inter-trial interval period. For this experiment, we expected to record kinematics from the legs through the whole stand and sit task.
Surgery
Under deep general anaesthesia, 3 headpieces were surgically implanted to allow head restraint. At a second surgery, a single 23×20 mm recording chamber was mounted so as to give access to the primary motor area for recordings in the left hemisphere. After the craniotomy surgery had been made, the monkeys received a full course of antibiotics (20 mg/kg oxytetracycline, i.m.) and analgesic (10 μg/kg buprenorphine, i.m).
Separate surgical procedures were performed to implant fine-wire intramuscular electrodes for electromyogram (EMG) into 6 selected lower limb muscles from each leg, including right and left soleus (RS and LS), right and left tibialis anterior (RTA and LTA), right and left semitendinosus (RST and LST), right and left recutus femoris (RRF and LRF), right and left extensor digitorum longus (REDL and LEDL), and right and left flexor digitorum longus (RFDL and LFDL).
Acute cortical activity and lower-limb electromyogram recordings
Single-unit activities were recorded using a microdrive recording system (Thomas Recording system, Germany). The location of each electrode penetration was determined by triangulation on fiducial markers on the chamber lid, which allowed the stereotaxic position of each new penetration to be calculated. The actual coordination of M1 area in stereotaxic apparatus can be obtained accurately according to the atlas of monkeys’ brain anatomy. Before every penetration, the positions of each electrode were calculated precisely with the method above to ensure that the recording area was within the right range. Lower-limb EMGs were recorded synchronously with a sampling rate of 1000 Hz.
Signal analysis
Neural signals corresponding to extracellular action potentials were conditioned and amplified, filtered, digitized and stored using a 64-channel neuron recording system (Plexon Inc., Dallas, TX). About 20 successful trials formed a recording session. Spike sorting was performed to isolate individual neuronal unit on each data session based on the clustering of detected action potential waveforms in principal components (PC) feature spaces using the software tool of Offline Sorter (Plexon Inc., Dallas, TX). Sorted data from each session were imported to Neuroexplorer (Plexon Inc.) for statistical analysis. Peri-event histogram analysis was applied to examine patterns of multiunit activity associated with events of lower limb kinematics in motor tasks, with special attention on: initiation of stand, stand up, stand hold on and sit down during the position control task.
A one-way analysis of variance (ANOVA) with p < 0.05 was applied to classify neurons based on their statistical properties of neural activity (firing rate) in each epoch, compared to the average firing rate during the whole trial. The firing rate of each neuron was computed in a 200 ~ 500 milliseconds time window with 30-millisecond bins.
Density contour of neuronal distribution was calculated at each penetration point as the number of neurons related to event divided by the total number of cells recorded under that penetration. Then the densities for all the penetrations were smoothed by a median filter, and the density contour for neuronal distribution related to events was plotted. The hottest colour represented the highest density.
We also investigated the innervating relationship between cortical neuronal units and lower-limb muscles. A multi-input-single output (MISO) linear model was built to depict the correlation of the EMG activities of each muscle in lower limbs using primary motor cortical (M1) neurons. This MISO model utilized the current and historical spiking rates of multiple motor neurons as input to predict the current rectified EMG signal:
$$ {\mathrm{y}}_m(t)={\displaystyle \sum_{i=1}^N{\displaystyle \sum_{\tau =0}^T{f}_{im}\left(\tau \right){x}_i\left(t-\tau \right)+\varepsilon }} $$
(1)
In (1), each spiking rates input x is convolved with its finite impulse response (FIR) function fim. The subscription im stands for the FIR function of neuron i on muscle m, and ε stands for the residue error, and N is the number of total neuron units involved into the prediction. The values of the FIR function fim need to be estimated from the system inputs and outputs in a train dataset. In model implementation, the spiking rate of each neuron was computed in non-overlapping 20 ms time windows; the EMG signals were detrended, full-wave rectified and then filtered by an order 8 Butterworth low-pass filter with a cut-off frequency of 4 Hz. For each session, we used the first 2/3 trials to train the model, and test it on the remaining 1/3 trials.
