Effect of spinal cord injury on neural encoding of spontaneous postural perturbations in the hindlimb sensorimotor cortex

Abstract

Supraspinal signals play a significant role in compensatory responses to postural perturbations. Although the cortex is not necessary for basic postural tasks in intact animals, its role in responding to unexpected postural perturbations after spinal cord injury (SCI) has not been studied. To better understand how SCI impacts cortical encoding of postural perturbations, the activity of single neurons in the hindlimb sensorimotor cortex (HLSMC) was recorded in the rat during unexpected tilts before and after a complete midthoracic spinal transection. In a subset of animals, limb ground reaction forces were also collected. HLSMC activity was strongly modulated in response to different tilt profiles. As the velocity of the tilt increased, more information was conveyed by the HLSMC neurons about the perturbation due to increases in both the number of recruited neurons and the magnitude of their responses. SCI led to attenuated and delayed hindlimb ground reaction forces. However, HLSMC neurons remained responsive to tilts after injury but with increased latencies and decreased tuning to slower tilts. Information conveyed by cortical neurons about the tilts was therefore reduced after SCI, requiring more cells to convey the same amount of information as before the transection. Given that reorganization of the hindlimb sensorimotor cortex in response to therapy after complete midthoracic SCI is necessary for behavioral recovery, this sustained encoding of information after SCI could be a substrate for the reorganization that uses sensory information from above the lesion to control trunk muscles that permit weight-supported stepping and postural control.

NEW & NOTEWORTHY The role of cortical circuits in the encoding of posture and balance is of interest for developing therapies for spinal cord injury. This work demonstrated that unexpected postural perturbations are encoded in the hindlimb sensorimotor cortex even in the absence of hindlimb sensory feedback. In fact, the hindlimb sensorimotor cortex continues to encode for postural perturbations after complete spinal transection.

INTRODUCTION

Maintaining postural stability is critical for recovery of independent locomotion after spinal cord injury (SCI). Efficient control of posture is equally important for standing and walking (1, 2) as it is for providing support of voluntary limb movements (3). Depending on the location and extent of SCI, damage to descending and ascending spinal pathways can result in an impairment of postural control (1). The behavioral effect of a complete midthoracic lesion of the spinal cord has been well studied (4–8). Notably, although postural control is reduced after SCI, brief standing episodes have been reported even after injury in some animal models (9, 10). In addition, interventions, including treadmill training (11–14) and epidural electrical stimulation (15), have led to modest improvements in this control. Thus, a more thorough understanding of how the entire neural axis encodes for posture before and after SCI can inform therapies that target the restoration of postural control after SCI.

The effects of SCI on limb responses during various postural disturbances have been studied (6, 16–18). Specifically, electromyographic (EMG) recordings in the hindlimbs show increases in response latencies, decreases in response amplitudes, and changes in muscle recruitment strategies after a complete midthoracic transection, suggesting that descending neural circuits from above the lesion are necessary for postural control (6, 16). Partial lesion studies in rabbits have suggested that the ventral spinal pathways (reticulospinal and vestibulospinal tracts) arising from the brainstem are critical for postural control whereas dorsal pathways arising from the cortex and midbrain (corticospinal and rubrospinal) are less important (17, 19–22). Although these brainstem circuits clearly provide important descending control of posture, cortical responses to periodic (predictable) rotations in the frontal plane (tilts) have also been documented in rabbits and cats, both before (23–26) and after SCI (17, 22, 27). Though cortical activity is not critical for basic postural tasks (28, 29) in intact animals, its role increases substantially after spinal cord injury. It has been shown that cortical reorganization (30, 31) and sprouting of corticospinal axons (31, 32) is associated with recovery of locomotion after injury. More importantly, lesioning the reorganized cortex results in a loss of behavioral improvement achieved after therapy, demonstrating that, in addition to other supraspinal circuits, descending information from the cortex is critical for functional recovery. Therefore, the role of cortical circuits in the encoding of posture and balance is of interest and the impact of spinal cord injury is unknown

To better understand the impact of SCI on HLSMC encoding of information about postural perturbations, we assessed the hindlimb ground reaction forces, the responses of ensembles of single neurons in the rat HLSMC, and the interaction of the two during an unpredictable tilting task both before and after a complete midthoracic spinal cord transection. In intact animals, when sensory input from the hindlimbs is removed, the response of neurons in the hindlimb sensorimotor cortex (HLSMC) to predictable tilts is greatly attenuated (33). This would suggest that cortical responses would also be attenuated after complete spinal transection, resulting in a decrease in the encoding of information about the tilt. Alternatively, it is possible that HLSMC reorganization after SCI allows for continued encoding of unpredictable tilts due to inputs from sensory afferents above the level of the lesion. As reported previously, we observed cortical response modulation to different tilt types in intact rats. Cortical neurons encoded information about the initial velocity of the perturbation within 50 ms of the start of tilt, and that the timing and magnitude of the hindlimb ground reaction forces scale with initial tilt velocity. After a complete spinal transection (PostTx), cortical activity was sufficient to determine if a tilt occurred and to provide information about the initial velocity of the tilt. Despite a reduction, this sustained cortical encoding about postural perturbations even after a complete spinal cord transection could inform therapies that target cortical reorganization. In addition, as open-loop spinal stimulation has been shown to improve functional outcome after SCI (27, 34–36), how this cortical information could be used to control a spinal prosthetic is discussed.

MATERIAL AND METHODS

Ethical Approval

All experiments were performed under approval of the Drexel University Institutional Animal Care and Use Committee, followed established National Institutes of Health guidelines (Protocol 19786), and was conducted in accordance with the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines (37).

Surgical Procedures

Cortical implant.

Eight Long Evans rats were trained to tolerate a harness and were acclimated to the task for 1–2 wk. Once adapted to the harness and standing, animals were chronically implanted bilaterally with 16 channel (4 × 4) Teflon-insulated stainless steel microwire arrays (MicroProbes for Life Sciences) in layer V of the hindlimb/trunk sensorimotor cortex (Fig. 1A) using methods standard in our lab (40, 41). Of note, there is an almost complete overlap of hindlimb sensory and motor cortices in the rat (42). As the electrode was lowered, neural activity was preamplified, bandpass filtered between 100 Hz and 8 kHz, and digitized with data acquisition hardware (National Instruments, Austin, TX). Digitized raw signal and waveforms were displayed on a computer with Recorder software (Plexon Inc, Dallas, TX) and also monitored on an oscilloscope and made audible through speakers. The array was lowered in 20-µm increments, no faster than 60 µm/min. At each 20-µm step, sensory responsiveness was assessed by tapping a blunt tip probe over cutaneous surfaces. Neurons predominately responding to hindlimb stimulation ensured proper electrode position in the hindlimb sensorimotor cortex (38, 43). Once characteristically large amplitude layer V neurons were visualized on a majority of the channels (at a depth of 1.3–1.6 mm), the array was cemented in place. All surgical procedures were performed under general anesthesia (2%–3% isoflurane in O₂) via orotracheal intubation. Pain was managed using Buprenorphine SR LAB (0.5 mg/kg; Wildlife Pharmaceuticals Inc.), and animals were given at least a week to recover from the surgery.

Figure 1.

Experimental design. A: location of the bilaterally implanted cortical arrays [based on data from Nandakumar et al. (38)]. The array (red box) is centered over the exclusive hindlimb motor cortex (M1) with some overlap with trunk motor (M1) cortex medially and hindlimb sensory (S1) cortex laterally (noted with dashed lines). B: diagram of the tilt task. At baseline, the animal stands on a platform with left hindlimb, right hindlimb, and forelimb force sensors. The platform makes unpredictable tilts of varying speed and direction in the horizontal plane described in C. Neuron spiking activity is recorded throughout task, and an exemplar neuron’s raster plot and PSTH for a specific tilt is shown. C: platform angle (top) and velocity (bottom) profiles for the 8 different tilt types in one direction: 4 “constant duration tilts” with varying final angle and peak velocity (i), 2 “constant final angle tilts” with varying duration and peak velocity (ii), and 2 “constant peak velocity tilts” of varying duration and final angle (iii). Note that each tilt includes the increase to max speed and then comes to a complete stop to include the effects of acceleration and deceleration. D: description of the PSTH-based method for classifying trials adapted from Bridges et al. (39). PSTH templates are generated for every neuron from over at least 100 of each tilt or baseline epoch. The PSTH from an individual trial is then compared with each template bin by bin, and the single trial is classified as belonging to the template with the most similarity (in this case, Tilt Type #2). A confusion matrix is generated from the classifier performance and information is calculated using Shannon’s equation. PSTH, peristimulus time histogram.

Spinal transection.

After intact (PreTx) recordings were complete, animals underwent a complete spinal transection procedure at T8, identical to the methods in previous studies (40, 41). In brief, animals were given prophylactic antibiotics (Baytril 5 mg/kg) and were anesthetized with an induction dose of 4% isoflurane followed by maintenance at 1.5%–2% isoflurane. A laminectomy was performed at T8/T9. Microdissecting scissors were used to remove the dura, and the cord was transected with iridectomy scissors immediately followed by aspiration. Two surgeons confirmed the lesion visually under ×20 magnification. The muscle and skin were sutured in layers with 4-0 nondissolving suture. Animals were treated with an analgesic (buprenorphine 0.05 mg/kg), given 10 mL saline, and placed on a heating pad until recovery. After transection, animals were kept on a heating pad and received ongoing care including bladder expression 2–3 times daily, antibiotics, and fluids as needed. Transected animals were given 1 wk of recovery before PostTx recordings were performed. All PostTx recordings were collected within 3 wk of the injury.

Behavioral Tasks

Tilt task.

The tilt task is shown in Fig. 1B. Rats stood in a neutral position on a platform consisting of three Plasti Dip (Plasti Dip International) coated plexiglass plates (one for each hindlimb and for the forelimbs collectively) coupled to a high-performance brushless AC servo motor (J0400-301-4-000, Applied Motion Products). The animal wore a pelvic harness attached to a body weight support system at the pelvis. No vertical weight support was provided PreTx, allowing the animal to freely adjust posture, make small steps, and shift body weight. PostTx, approximately 50% of the weight was supported vertically at the pelvis.

The platform rotated in the frontal plane, remained at the peak angle for approximately 1 s, then returned to the neutral position while the neural response to the perturbation was recorded. The platform remained in a neutral position for a random intertrial interval of 2–3 s. Tilt type was randomized within a recording session using LabVIEW (2015, National Instruments), and direction was reversed during a subsequent recording session. Several tilt types were programmed using Si Programmer (v. 2.7.22, Applied Motion Products) on a digital motor drive (SV7-SI-AE, Applied Motion Products). Start of tilt was defined as the time at which the platform started to move.

Tilt types.

Eight tilt types were classified based on their duration, final tilt angle, and peak velocity. A set of four “constant duration tilts” varied in peak velocity and final tilt angle while maintaining the same duration from tilt onset to maximum angle (Fig. 1Ci). In addition, two “constant angle tilts” modified the peak velocity and duration of the tilt, but they reached the same maximum angle (Fig. 1Cii). Finally, two “constant velocity tilts” modified the final angle and duration of the tilt with the same peak velocity (Fig. 1Ciii). For five of the animals, all tilt types were recorded in the same session with the same population of neurons. For a subset of three animals, only the two “constant final angle tilts” were recorded.

Ground reaction forces.

Ground reaction forces were measured for the subset of three animals. Original Equipment Manufacturer (OEM)-style single point load cells (LCAE-600 G; Omega) positioned underneath the platform plexiglass plates quantified hindlimb and forelimb ground reaction forces (GRFs). Data were acquired using LabVIEW software (1,000 samples/s), which was filtered offline using a 2nd order Butterworth zero-phase low-pass filter. Sensor data were normalized to a period found in the 200-ms time window before the start of each tilt. Positive values indicate additional loading onto the sensor.

Single Neuron Recordings

Populations of single units were recorded simultaneously during the tilting task using methods standard in our lab (40, 41). In brief, before every session, neurons were sorted online (Sort Client, Plexon Inc.). First, the neural activity was played over speakers to identify whether neurons were present. An oscilloscope was used to confirm neural activity. The oscilloscope, waveforms, and the first two principal components were used to sort the cells (typically 1 or 2 per channel, occasionally 3). A Multichannel Acquisition Processor (MAP, Plexon Inc, Dallas, TX) was used to record from multiple single neurons during each recording session and to record event timestamps. Offline, neurons were categorized as single units if less than 0.5% of spiking occurred in the first 1 ms of the interspike interval histogram (Offline Sorter, Plexon Inc.). All other units were discarded. Waveforms were checked for consistency over the course of the experiment.

To determine cell responsiveness, peristimulus time histograms (PSTHs) were built around the start of tilt (±300 ms) with a 2 ms bin in a manner similar to our previous work (41, 44). Background activity was defined in the 300 ms before the start of tilt, when the platform had been in the neutral position for a minimum of 1 s. Threshold was defined as the average background activity plus 1.65 times the standard deviation. If five bins (10 ms), with no more than 10 ms separating them, crossed the threshold, then response was compared to an equivalent background window. If they were significantly different (paired t test, P < 0.05), the cell was classified as responsive, and the characteristics of the response were found.

Data Analysis

Ground reaction forces.

For three animals, differences in GRFs were compared between fast and slow tilts and between injury states using a two-way analysis of variance. Peak forces generated during the tilt were averaged for different tilt types before and after SCI.

Neuron response profiles.

The number of responsive neurons and the response characteristics of those neurons were evaluated between each tilt type and before and after spinal cord injury using two-way analysis of variance or t tests as appropriate. Using the PSTHs, response characteristics were defined by both the timing and magnitude of the response. The first and last bins to cross threshold were defined as the “first and last bin latencies,” respectively. “Response duration” was the difference between the last and first bin latencies. “Response magnitude” was defined as the sum of the spikes in all the bins between the first and last bin latencies, divided by the total number of trials after subtracting the average background activity. “Response firing rate” (in Hz) was therefore the response magnitude divided by the duration of the response in seconds. If a neuron responded to multiple tilt types, only the case with the largest response magnitude was used such that each neuron only contributed one observation to analyses.

Neuron classification.

For the four constant-duration tilts (Fig. 1Ci), neurons were classified based on the number of tilts which elicited a significant response. A “selective” neuron responded to a single tilt type only, a “nonselective” neuron responded to more than one but less than four tilts, and a “scaled” neuron responded to all four tilt types.

Relating ground reaction forces to neural responses.

To summarize the effect of SCI, changes in the average firing rate of neurons during a tilt were correlated to the change in GRF during the tilt. The data were normalized to the start of tilt and the change in firing rate was plotted against the change in GRF for different tilt types and injury conditions in the clockwise and counterclockwise direction (Fig. 5).

Figure 5.

Relating average neural firing rate to hindlimb ground reaction forces throughout tilts. Trial-averaged left hindlimb ground reaction forces and neural firing rates (calculated as average spikes across all responsive neurons per 10 ms bin) are plotted for tilts in the counterclockwise (left plots) and clockwise (right plots) directions from the time of tilt onset (black dot) to the time of maximum tilt angle (red dot) for a single animal. A second animal’s results are inset for comparison. Force and neural firing rate were zeroed at the start of tilt. A: comparison between fast and slow velocity tilts in the PreTx condition in the counterclockwise (left) and clockwise (right) directions. B: comparison between fast tilts in the Pre- and PostTx condition for the same animals in the clockwise (left) and counterclockwise (right) directions. Tx, spinal transection.

Information analysis.

Information was quantified using a PSTH-based method (45). In short, PSTHs were generated to find the average response profile (100 trials) of each neuron to each event. In a leave-on-out manner, individual trials were compared with the average response (generated without the single trial) and the difference between the single trial and the average profiles was calculated in a bin-bin comparison (see Fig. 1D). The single trial was classified as either a particular tilt response or a background response by identifying the profile with the smallest difference from the single trial. Performance was expressed as the percentage of correctly classified trials. The information was calculated using Shannon’s information formula, formally defined as:

$$I\left(s;r\right)={\displaystyle {\sum }_{s,r}P(r,s){\log \hspace{0.17em}}_2\left[\frac{P(r,s)}{P(r)P(s)}\right]}$$ (1)

where P(r), P(s), and P(r,s) correspond to the probability of the tilt-perturbation response r, the tilt perturbation stimulus s, and their joint probability, respectively. I(s;r), which is measured in bits, was calculated for each neuron using the actual and predicted tilt type confusion matrix generated when applying the classifier. Residual bias for I(s;r) was then estimated using a bootstrapping procedure by pairing the trial response and tilt types in a randomized order—effectively eliminating their associations. This bootstrapping procedure was performed 20 times and the calculated bias was subtracted from I(s;r) such that 0 bits is chance.

To establish the bin size that resulted in the maximal PSTH classifier performance, a range of bin sizes between 2 and 280 ms were considered. The optimal bin size was determined to be 20 ms across animals, consistent with our previous work (39), so this bin size was used for all information analyses.

Tilt detection.

Trials from tilt profiles in which the severity of the perturbation (changing velocity and peak angle but keeping duration constant, Fig. 1Ci) were used for detection and discrimination analyses. To determine if neurons could detect any tilt severity from standing in a neutral position, responses of neurons to each of the four constant duration tilts (Fig. 1Ci) were compared with their firing rates at stance (baseline) and thus single trials were classified as either a tilt or baseline. To account for differing neurons numbers between animals and conditions, information was calculated 20 times with a random sampling of 34 neurons. This number was chosen as it represented the total number of neurons for the animal with the least number of discernable neurons. The effects of tilt severity and injury on information were compared using a two-way analysis of variance.

Tilt discrimination.

An important feature of proper postural control is the ability to discriminate different perturbation types. For tilt discrimination analyses, single trials were classified as being from one of the four tilt profiles. As with tilt detection analyses, information and performance were calculated from the averages obtained from twenty iterations using 34 randomly selected neurons.

To further explore the encoding for discrimination, the following analyses were performed. First, a set number of randomly selected neurons (from 4 to 34 neurons) was used to determine how population size affected information (neuron dropping). To determine how quickly information was conveyed to the hindlimb sensorimotor cortex, information was calculated with an incremental increase in the event window, from the first 20 ms after tilt onset to 280 ms after tilt onset. For both analysis types, information and performance were calculated 20 times using a different sampling of neurons.

For information and neuron analysis, five animals were tested PreTx condition with two sessions each for a total of 10 recordings. Four of those animals were then evaluated similarly PostTx for a total of eight recordings. For ground reaction force analysis, three animals were tested PreTx and PostTx. All reported values are means ± standard error.

RESULTS

Ground Reaction Force Magnitude and Timing Scale with Tilt Severity and Injury

To assess the impact of SCI on the response of the hindlimbs to unexpected tilts in the lateral plane, ground reaction forces (GRF) in response to each hindlimb and to both forelimbs together were collected during the two tilts of constant final angle (varying duration and peak velocity; refer to Fig. 1Cii, constant angle tilts) before and after the SCI (Fig. 2A). As described in previous work (39), before SCI, counterclockwise tilts in healthy animals begin with the platform pushing into the right hindlimb and away from the left hindlimb in the first 100 ms, leading to an unloading of force from the left hindlimb sensor. The next 100 ms is characterized by an active correction of the animal’s center of mass, during which the animal extends the left hindlimb and flexes the right hindlimb to shift the center of mass over the base of support. This causes a subsequent increase in the ground reaction forces measured in the left hindlimb sensor. The converse is true for clockwise tilts (data not shown). As expected, the measured GRF of the combined forelimbs is much less than that observed in a single hindlimb since the shifting of the weight is conserved. Nonetheless, the weight on the forelimbs, as measured by the GRF, is unloaded as the tilt is initiated, resulting in an overall reduction in total GRF at the peak of the tilt, suggesting a shift of the center of mass to the hindlimbs.

Figure 2.

Ground reaction forces during tilts. Negative forces imply unloading whereas positive forces imply loading. A: average force measured in the left hindlimb (left) and both forelimbs (right) by the sensor for fast (270 ms) and slow (620 ms) constant-angle tilts from tilt onset to the maximum tilt angle in the counterclockwise direction over 100 trials for a single animal. Forces are normalized such that each trial begins with 0 N of force. Shaded areas represent the standard error of the mean. Dotted vertical lines represent the time at which the platform reaches the maximum tilt angle for fast (270 ms) and slow (620 ms) tilts. Inset shows the same animal’s forces for tilts in the clockwise direction. B: the average maximum force applied to the left hindlimb force detector between approximately 100–200 ms during counterclockwise tilts. C: the timing of maximum force applied to the hindlimb force detector in B. D: the magnitude of the force applied to the hindlimb force detector during tilt at the end of the tilt (when the tilt platform achieves its maximum angle). E: forelimb forces were also evaluated. The maximum loading force applied to the forelimb force detector was calculated and compared across tilt types and Tx status. Note: only one sensor was used for forelimb force measurements, and animals’ weight was partially supported by a trunk harness PostTx. Means and standard deviations are plotted over 99 individual trial data points both Pre- and PostTx. ****P < 0.001, ***P < 0.001, **P < 0.01 (Bonferroni corrected). Tx, spinal transection.

After SCI, the initial response to the tilt, as the platform moves away from the paw, is similar to the response PreTX. As the platform moves away from the paw, the GRF is reduced. Presumably, spinal circuits below the level of the lesion contribute to a reflex response where the limb is extended resulting in a restoration of GRF. However, unlike in the PreTx response, the GRF in the PostTx condition simply returns to the same force seen at stance, such that the extended position of the paw does not exert any force greater than what was exerted in the neutral position as a harness provided additional support and some of this weight shifted to the forepaw.

To quantify these differences, the forces generated by the hindlimb at the start and end of the faster tilts were compared with those for slower tilts before and after injury. As expected, the initial force was significantly greater for faster tilts [effect of tilt type: F(1,390) = 113.6, P < 0.0001] and this force was reduced after injury [effect of injury: F(1,390) = 160.6, P < 0.0001) regardless of the tilt type (Fig. 2B). However, the difference in force between tilt types was attenuated after SCI [interaction: F(1,390) = 23.87, P < 0.0001]. Furthermore, regardless of injury status, rats initiated a correction of their center of mass more quickly during faster tilts [effect of tilt type: F(1,392) = 142.2, P < 0.0001; interaction F(1,392) = 0.41, P = 0.53] but these corrections were significantly delayed after injury [effect of injury: F(1,392) = 272.2, P < 0.0001; Fig. 2C].

Thus, to stabilize the center of mass within the first 200 ms of the onset of the tilt, the response of the hindlimbs to faster tilts was quicker and more robust than those to slower tilts whereas the response after SCI was delayed and attenuated regardless of tilt type. To assess the impact of tilt type and injury on the final postural adjustment, the forces applied at the maximum angle of the tilt was compared (Fig. 2D).

As expected, as the final angle was the same, there was no effect of tilt type on the final force generated by the hindlimbs [F(1,392) = 0.71, P = 0.40], but there was a significant attenuation of this force with injury [F(1,392) = 115.7, P < 0.0001]. Thus, despite differences in the magnitude and timing of the initial force applied during the tilts PreTx, the animal ultimately applies the same force at the maximum angle of the tilt. After injury, these relationships remain, but the force is attenuated. In an effort to stabilize their center of mass, some of this force is transferred to the forelimbs [Fig. 2E; effect of tilt type: F(1,338) = 10.81, P = 0.001; effect of injury: F(1,338) = 14.83, P = 0.0001; interaction: F(1,338) = 0.12, P = 0.12], whereas the remainder is taken up by the harness. Thus, after spinal cord injury, the center of mass of the animal shifts forward as more weight is on the forelimbs at the peak of the tilt.

An Increasing Number of Neurons Respond to Tilts of Increasing Severity

Neuronal responses to a larger set of tilt types were evaluated to assess the impact of SCI on the encoding of the postural shifts described in Fig. 2. Populations of single neurons were recorded while animals were subjected to eight different rotations in the frontal plane at random intervals. These tilts maintained either constant duration, constant final angle, or constant peak velocity (Fig.1C). For neurons with a significant response to at least one tilt type, their response profiles (response magnitude, duration of response, and latency of response) were compared across tilt types and between Pre and PostTx. The number of cells that responded, or responsive neurons, were different depending on the tilt type [F(7,63) = 22.00, P < 0.0001]. This change in responsive neurons was observed if the peak velocity was changed but not if the peak velocity was held constant (Fig. 3A). Thus, more neurons were recruited into the response primarily when the severity of the tilt was increased.

Figure 3.

Neuron response profiles for all tilt types Pre- and PostTx. For tilts of different duration, final angle, and peak velocity profiles (see Inset table), the number of neurons and their average response metrics were calculated and compared. If a neuron responded to more than one tilt type, it was only included in the tilt type that led to its largest response. A: average number of neurons per animal responsive to each tilt type Pre- (n = 10) and PostTx (n = 8). B: average neuron firing rate (in Hz) throughout response. C: average duration of neuronal response. D: average latency of response onset, defined as the time at which the first bin surpasses threshold. Inset: table defining the tilt types for A–D. Means and standard deviations are plotted over individual data points representing animals in A or individual neurons in B–D. Total neuron numbers for each tilt type (in B–D): A1 (PreTx: 120; PostTx: 36), A2 (PreTx: 116; PostTx: 63), A3 (PreTx:132; PostTx: 71), A4 (PreTx: 152; PostTx: 90), B1 (PreTx: 147; PostTx: 69), B2 (PreTx: 155; PostTx: 90), C1 (PreTx: 115; PostTx: 37), C2 (PreTx: 117; PostTx: 30). ***P < 0.001, *P < 0.05 (Bonferroni corrected). Tx, spinal transection.

After complete spinal transection (PostTx), there was a trend toward an overall reduction in responsive neurons [F(1,9) = 4.986, P = 0.0524]. This was observed despite an unchanged number of discriminable neurons PostTx [paired t test, t(3) = 1.140, P = 0.34]. However, just as in the PreTx condition, neuron recruitment remained tuned to tilt severity (changes in peak velocity).

Duration of Neuronal Responses Also Increased with Increasing Severity of Tilt

The magnitude of the response was similarly modulated by tilt type [F(7,1524) = 0.014, P = 0.0142] with increasing spikes per tilt with increasing severity but no change in number of spikes when severity was held constant. Interestingly, there was no effect of injury on the magnitude of the response, suggesting that information about the tilt is reaching the brain despite the injury [F(1,1524) = 0.394, P = 0.536]. However, for both Pre- and PostTx conditions, the change in response magnitude for more severe tilts was not due to changes in firing rates [F(7,1524) = 0.81, P = 0.815; Fig. 3B] but rather to increases in response durations [F(7,1524) = 0.749, P = 0.0009; Fig. 3C]. Therefore, increases in the severity of the tilt resulted in an increase in the duration of the response both pre- and post-SCI, regardless of the duration of the tilt.

Responses to Tilts Are Delayed, but Not Attenuated, After Transection

Interestingly, response latencies were dependent on the severity of the tilt, similar to the duration of the response (Fig. 3D). Furthermore, despite SCI having no impact on the firing rate or magnitude and duration of response, injury delayed the timing of the response (Fig. 3D). In fact, after injury, the response onset, was significantly delayed by about 10 ms [first bin latency: PreTx: 54.1 ± 62.3 ms and PostTx: 63.7 ± 67.1 ms, F(1,1524) = 11.22, P = 0.0008].

In summary, as the severity of the tilt increased, the number of responding neurons and duration of the response increased while the latency of the response shifts earlier without a change in firing rate. These findings emphasize the importance of the severity of the tilt on the neural response. Post-SCI, the severity of tilt had similar impact on the neural response, but the response was shifted later without impact on the magnitude of the response. Therefore, in an unexpected postural perturbation, neurons in the hindlimb sensorimotor cortex respond to the severity of the tilt even in the absence of sensory feedback from the hindlimbs. The shift in latency after SCI suggests this response in the hindlimbs is due to sensory information about the tilt coming from the upper trunk and forelimbs.

Some Neurons Scale their Responses to Tilts of Increasing Severity

To further explore the responsiveness of neurons to tilts of increasing severity (peak velocity), neurons were classified into three classes depending on the range of tilts they responded to (out of four possible tilts): 1) selective, neurons responsive to only one tilt type; 2) nonselective, neurons responsive to two to three tilt types; or 3) scaled, responsive to all four tilts (examples seen in Fig. 4, A–C, respectively).

Figure 4.

Neuron response classification. For the four tilts of constant duration but increasing peak velocity and tilt angle, neurons were categorized as being selective, nonselective, or scaled. A–C: exemplar peristimulus time histograms of neurons for each classification type, where A represents a neuron responsive to only one tilt type, B represents a neuron responsive to 2–3 tilt types, and C represents a scaled neuron, or a neuron responsive to all four tilt types. D: distribution of each neuron type before and after Tx. E: distribution of preferred event for all selective neurons before and after Tx. ***P < 0.001, **P < 0.01. Tx, spinal transection.

In addition to delayed response to tilt after transection, there was a shift in the classification of neurons (Fig. 4D), with a decrease in the proportion of scaled neurons (39.6%–17.0%), an increase in the proportion of selective neurons (21.4%–39.7%), and a modest increase in nonselective neurons (39.1%–43.3%) [χ²(2) = 16.06, P = 0.0003]. This shift occurred due to a reduced response to the least severe tilts PostTx. In fact, when classifying the selective neurons by the tilt to which they responded (Fig. 4E), there was a significant change in these proportions [χ²(3) = 12.69, P = 0.0054], with the biggest change being a reduction in the number of cells uniquely responsive to the lowest severity tilt (33% v. 13%). Thus, this reduction in scaled neurons is likely due to the fact that neurons previously responsive to all tilts may no longer respond to the mildest tilts after injury, leading to a classification as either a selective (responding to one tilt type) or non-selective (responding to 2–3 tilt types) neuron.

Impact of SCI on Neuronal Dynamics during Tilt

To visualize the impact of SCI on the change in firing rate of the population of neurons during tilt, the average response of neurons during each tilt was plotted against the change in GRF. Examining the left hindlimb during clockwise and counterclockwise tilts, the GRF changes first, before the increase in average firing rate (Fig. 5A). Then, the firing rate peaks at about the same time as the peak unloading of the limb due to the movement of the platform. As the limb begins to exert a restorative force to stabilize the animal’s center of mass, the firing rate of the population starts to decline but does not quite return to baseline before the GRF reaches its maximum restorative force. For less severe tilts, this trajectory pattern is simply scaled down. As the population of cells are the same for the clockwise and counterclockwise tilts, these data suggest that in each hemisphere, a subset of cells are responding to the contralateral limb’s extension while others are responding to the contralateral limb’s flexion. After injury, the loading, unloading, and final restorative forces are attenuated, likely due to the harness support and the shift of weight to the forelimbs, and the neuronal firing rate is also reduced (Fig. 5B), resulting in a similar shaped trajectory, albeit reduced in size.

Neurons Encode for The Detection of Tilt

As neurons respond to the tilt even after a complete spinal transection and changes in the severity of the tilt (peak velocity) are the greatest drivers of that response, we wanted to gain insight into how these neurons encode information about the tilt. Specifically, we investigated how neurons encode for the occurrence of a tilt (i.e., is the platform stationary or has it tilted?) and the magnitude of the severity of the tilt (i.e., what was the peak velocity?) as well as the effect, if any, of SCI on that encoding. The first step to evaluate the encoding of tilt was to quantify the information that a tilt occurred (tilt detection) using a fixed number of neurons (Fig. 6A).

Figure 6.

Information in the hindlimb cortex about tilts. Information about tilt detection (A) and tilt discrimination (B) was calculated using the PSTH classifier (see material and methods) for the four constant duration tilts of increasing peak velocity and final angle. A: tilt detection, or the ability to distinguish stance from the tilt, increased as a function of tilt severity (P < 0.0001) in both the Pre- and PostTx condition, even with a significant decrease in information PostTx (P < 0.0001). This was despite standardizing the number of neurons in all tilt and Tx states. Means and standard deviations plotted over individual data points. B and C: tilt discrimination, or the ability to differentiate one tilt type from the other three, increased with adding more neurons (P < 0.0001; B) as well as with using a set number of neurons but increasing the length of recording supplied to the classifier (P < 0.0001; C). Both instances demonstrated a reduction in information PostTx (P < 0.0001 and P < 0.0001), respectively. For both B and C, solid line represents the change in mean for each condition, plotted over data points for individual animals (PreTx n = 10; PostTx n = 8). PSTH, peristimulus time histogram; Tx, spinal transection.

As expected from the increase in the number of responsive neurons and the increase in the magnitude of their response with increasing severity, the information that a tilt occurred within the HLSMC increased with the severity of the tilt both PreTx (0.27 for least severe to 0.60 bits for most severe), and PostTx [0.14 bits for least severe to 0.39 bits for most severe tilts; F(3,64)=18.93, P < 0.0001]. After transection, the information about tilt detection was significantly reduced [F(1,64) = 31.03, P < 0.0001, interaction: F(3,64) = 0.42, P = 0.7406]. As the number of neurons used to compare PostTx to PreTx was the same, this reduction in information is unlikely to be due exclusively to the trend toward fewer responsive neurons PostTx.

Neurons Convey Considerable Information about Tilt Type Both Pre- and PostTx

Though the detection that a tilt occurred is the first step in understanding how neurons encode for postural responses, of greater interest is discriminating between different types of tilts as the severity of the tilt increased, which would suggest the information necessary to determine the postural adjustments that need to be made to maintain balance. As expected, the ability of populations of neurons to discriminate between tilt type was dependent on both the number of neurons used [Fig. 6Bi; F(12,208) = 8.22, P < 0.0001] and the amount of time that passed from tilt onset [Fig. 6C; F(13,224) = 7.56, P < 0.0001]. For the same sized population, a significant reduction in information was observed after transection compared to before [neuron dropping: F(1,208) = 19.56, P < 0.0001; window size: F(1, 224) = 27.60, P < 0.0001]. When comparing the number of neurons used (Fig. 6B), as few as twelve neurons were able to convey at least 0.2 bits of information about tilt detection in the PreTx condition. PostTx, 20 neurons were needed to convey comparable levels of information, likely arising from the decreases in cells with responses to less severe tilts after spinal cord transection. Thus, the nervous system can compensate for the loss in information about postural perturbations by recruiting more neurons into the task after SCI. As expected, as the window of time after the tilt onset used to calculate the information increased (Fig. 6B), information about tilt discrimination increased. PreTx, the population of neurons conveyed a considerable amount of information (more than 0.2 bits) about the tilt type within the first 60 ms of the tilt. After injury, an additional 60 ms were needed (120 ms) after tilt onset to reach similar information levels, consistent with the delayed neuronal responses seen PostTx. Information continued to increase over time for both PreTx and PostTx conditions, with information reaching 0.40 bits PreTx and 0.29 bits PostTx within 300 ms.

Therefore, as with tilt detection, both the reduction in the number of responding neurons PostTx and differences in the firing patterns of neurons contribute to the loss of information about the discrimination between different types of tilts after SCI.

DISCUSSION

Although weight-bearing and stereotypic locomotor movements can be restored following a complete spinal injury through activation of spinal circuits below the lesion (34, 46–51), there is a need to ensure adequate postural stability (39, 44, 52, 53). An understanding of how the brain encodes for this stabilization before and after spinal cord injury could be used for the design of therapeutic interventions that aim to enhance postural responses and decrease the morbidity and mortality associated with fall incidence. The data presented here suggest that the hindlimb sensorimotor cortex (HLSMC) encodes information about the severity of an unexpected perturbation by changing the number of responsive neurons and altering the duration, but not firing rate, of their responses. Moreover, after a midthoracic spinal cord transection, which prevents sensory information from the hindlimbs reaching supraspinal levels, information about the tilt continues to be encoded in the HLSMC with similar, albeit delayed, neural response dynamics.

Role of Afferent Feedback and Effects of Spinal Cord Injury on the Encoding of Postural Responses in Unexpected Perturbations

The work presented here, in which animals were subjected to perturbations that were unexpected in timing and severity both before and after spinal cord injury, can be compared with previous studies on behavioral and neuronal responses to predictable tilts in the frontal plane (24, 25). In the previous studies using intact rabbits or cats, the animals compensate for the tilt by extending the limb when the platform moves down and flexing the limb when it moves up. Although we recorded ground reaction forces (GRFs) and not limb kinematics or trunk movements from a subset of animals, our animals behaved similarly to shift their center of mass over their base of support, maintaining an upright position and preventing falling. The use of kinematics in future studies will allow for a better assessment of post-transection behavior, as weight-supporting harnesses undoubtedly affected the ground reaction forces after SCI. With respect to neural responses, we observed unique responses in the HLSMC to unexpected tilts, just as cortical modulation was previously observed in response to tilts in both rabbits (25) and cats (24).

Despite these similarities, there are key differences that distinguish our work. First, in cats (24), the activity of pyramidal tract neurons in the hindlimb motor cortex was almost exclusively correlated with extension of the contralateral limb. In the present study, when the tilt was unexpected, cells responded to both extension and flexion, suggesting a more robust response to unexpected tilts. Second, the peak firing rate in response to unexpected tilts for intact rats was over 30 Hz, about double the rate of those previously reported (24). This would suggest that unexpected perturbations create greater cortical responses. This is consistent with EEG and TMS studies in humans showing that the magnitudes of evoked cortical responses and postural muscle activations are greater during expected compared to expected postural perturbations (54–58).

Third, we observed a sustained cortical response to tilts even after removal of hindlimb afferents. This is in contrast with previous work that showed that suspending a hindlimb during platform tilts (and thus removing its sensory inputs) led to a strong attenuation of the response in the contralateral hindlimb motor cortex (33). In the current study, however, a complete midthoracic spinal transection that removed all sensory input from below the lesion simply caused a delayed neural response to unpredictable tilts in the HLSMC, but had no effect on the magnitude of the response. The unexpected tilts in this work better reflect instantaneous loss of balance and provide insight into the role of the cortex during a more complex balance task. Furthermore, this work clarifies that even after a complete spinal transection, neurons in the cortex still organize to convey information about the perturbation. it is important to note that, while our microelectrode arrays were chronically implanted in the HLSMC, we do not assume that the same neurons are being recorded pre-and post-SCI. Therefore, our conclusions are limited to the average responses of neurons in the HLSMC at these two time points.

This responsiveness in the absence of ascending sensory feedback from the hindlimbs likely originates from sensory signals from forelimb and trunk afferents caudal to the injury as well as inputs from the visual and vestibular system. It has been shown that limb somatosensory afferents are processed and even converge subcortically in the spinal cord and brainstem before reaching the cortex (59–62). Activity from the forelimbs is likely transmitted to the deafferented hindlimb cortex producing sufficient activation to discriminate the severity of the tilt, with the delay reflecting the additional time needed to reach firing thresholds due to the loss of hindlimb afferent inputs. Although the majority of hindlimb sensorimotor cortex cells respond to stimulation of the contralateral hindlimb, forelimb afferents also send inputs to the hindlimb sensorimotor cortex. Therefore, neurons in the “hindlimb” representation have been shown to respond to forelimb stimulation and vice versa (43). In addition, a network of HLSMC neurons that are normally active in response to forelimb movements could be contributing to the responses we observed in the deafferented cortex. This would be in line with the work by Karayannidou et al. (33), in which a subset of cortical neurons followed the forelimbs more closely when the forelimbs and hindlimbs were tilted out of phase.

Neural Encoding of Tilt

Neurons have been shown to encode for multiple sensory and motor events by scaling their firing rate to a parameter of the movement. For example, using multiple linear regressions, cells in the motor cortex have been shown to modulate their firing rate depending on the speed, direction, position and acceleration of arm trajectory (63). Moreover, neurons in the motor cortex have been shown to increase their firing rate with finger velocity in a center-out task (64). In sensory systems, cortical firing rate has been shown to increase with speed of whisker deflection (65). Our findings extend this scaling of neuronal activity to the hindlimb cortex during postural events and show that the magnitude of the response is scaled to the severity of the tilt, encoding not only that a tilt occurred, but the severity of the tilt.

Therefore, not surprisingly, the information that a tilt occurred increased as the severity of the tilt increased. We show that this is certainly due to the increase in neuronal response magnitude as intensity of tilt increases (e.g., scaling), as holding the number of responsive neurons constant resulted in more information for greater intensity tilts. Of course, the central nervous system (CNS) has access to the information from additional neurons that respond, and it is likely that this information is used by the CNS to make appropriate postural adjustments in response to the tilt. A likely explanation for the greater number of neurons responding as the intensity of the tilt increases is the need to activate more motor neurons, in turn activating more muscle groups, to maintain balance in response to larger perturbations.

Spinal cord injury reduces the amount of information about tilt detection and tilt discrimination, even when neuron numbers were held constant. This reduction in information suggests that recruiting more neurons into the task is one strategy to ameliorate any loss of information after SCI. After complete midthoracic spinal cord transection, neurons in the HLSMC are more likely to both respond to forelimb stimulation and also activate trunk musculature after regular physical rehabilitation (31). This cortical plasticity has been shown to improve behavioral outcome, as lesioning this reorganized cortex reduced gains in weight-supported stepping achieved by animals that received therapy (31). Therefore, therapeutic interventions that support reorganization would be expected to further improve outcome, and therapy along the entire neural axis, including the cortex – is likely necessary to optimize outcome after SCI (66–70).

Finally, the sustained encoding of postural information after SCI observed in thus study (with respect to both stimulus detection as well as discrimination) has important translational implications for the field of neuroengineering. Despite playing a less significant role in postural control than other supraspinal centers in the brainstem, this study has demonstrated that the cortex can serve as a source of information about postural perturbations after SCI. Brain-machine interfaces can be developed using cortical signals to augment postural control, such as through spinal or peripheral nerve stimulation to support functional recovery after spinal cord injury.

GRANTS

This work was supported by the National Institutes of Health Grant R01 NS096971; National Science Foundation Grant DARE – 1933751; National Center for Advancing Translational Sciences, National Institutes of Health—UL1 TR001860 and linked award TL1 TR001861.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

J.B.D., G.D.D., N.R.B., and K.A.M. conceived and designed research; J.B.D., G.D.D., N.R.B., and K.A.M. performed experiments; J.B.D., G.D.D., and K.A.M. analyzed data; J.B.D., G.D.D., and K.A.M. interpreted results of experiments; J.B.D., G.D.D., and K.A.M. prepared figures; J.B.D., G.D.D., and K.A.M. drafted manuscript; J.B.D., G.D.D., and K.A.M. edited and revised manuscript; J.B.D., G.D.D., and K.A.M. approved final version of manuscript.