Targeted, activity-dependent spinal stimulation produces long-lasting motor recovery in chronic cervical spinal cord injury

Abstract

Use-dependent movement therapies can lead to partial recovery of motor function after neurological injury. We attempted to improve recovery by developing a neuroprosthetic intervention that enhances movement therapy by directing spike timing-dependent plasticity in spared motor pathways. Using a recurrent neural-computer interface in rats with a cervical contusion of the spinal cord, we synchronized intraspinal microstimulation below the injury with the arrival of functionally related volitional motor commands signaled by muscle activity in the impaired forelimb. Stimulation was delivered during physical retraining of a forelimb behavior and throughout the day for 3 mo. Rats receiving this targeted, activity-dependent spinal stimulation (TADSS) exhibited markedly enhanced recovery compared with animals receiving targeted but open-loop spinal stimulation and rats receiving physical retraining alone. On a forelimb reach and grasp task, TADSS animals recovered 63% of their preinjury ability, more than two times the performance level achieved by the other therapy groups. Therapeutic gains were maintained for 3 additional wk without stimulation. The results suggest that activity-dependent spinal stimulation can induce neural plasticity that improves behavioral recovery after spinal cord injury.

Keywords: recurrent neural–computer interface; rehabilitation; spike timing-dependent plasticity; spinal cord injury.

Fig. 1.

TADSS reinforces volitional activation of spared motor circuits. In TADSS, the Neurochip detects increased EMG activity in an ipsilesional forelimb extensor and triggers ISMS at a functionally related spinal site below the lesion. Solid black lines indicate pathways spared by SCI; dashed lines indicate pathways damaged by SCI. The goal of TADSS is to strengthen synapses along the spared pathways (red synaptic terminals).

Fig. 2.

Implementation of TADSS. (A) C4–C5 hemicontusion damages corticospinal tracts, resulting in persistent motor deficits in the ipsilesional forelimb. Intraspinal microwires are implanted into ipsilesional elbow and wrist extensor motor regions spanning C6–C8 spinal segments and deliver either TADSS or TOLSS. (B) The Neurochip continuously records ipsilesional forelimb extensor EMG, discriminates salient increases in activity, and delivers ISMS to functionally related spinal motor regions.

Fig. S1.

Discrimination of salient EMG activity and delivery of TADSS. (A) Amplitude–time discrimination process; streaming EMG waveforms must meet all criteria to trigger TADSS. Thin gray lines indicate discrimination-aligned EMG activity highlighting a small motor unit action potential. The bold black line indicates average discriminated waveform. The horizontal dashed line indicates the minimum initial amplitude threshold chosen to distinguish EMG activity from baseline. The vertical blue lines indicate amplitude ceiling criterion on the left and amplitude return criterion on the right. EMG is on the y axis, and elapsed time (milliseconds) is on the x axis. (B) A 10-ms refractory period is imposed after each stimulation event. Blue bars indicate successful discrimination events, red bars indicate triggered stimulation, and horizontal gray bars indicate refractory periods. EMG is on the y axis, and elapsed time (milliseconds) is on the x axis. Note that the rightmost discrimination event does not result in stimulation, because it fell within the previous refractory period. (C) Example time record of EMG activity and triggered stimulation during free behavior. Red circles indicate TADSS stimuli, which are visible only during periods of robust activity. EMG is on the y axis, and elapsed time (milliseconds) is on the x axis. Data presented in B and C are taken from the same treatment session in one TADSS animal.

Fig. S2.

Generation of the TOLSS stimulation profile. (A) Histogram of 1.65 million ISIs from 18 activity-dependent stimulation (i.e., TADSS) therapy sessions. The mean ISI for all ISIs less than 250 ms is 0.048 s; mean ISI for all possible ISIs (up to 30 s) is 0.151. Count is on the y axis, and ISI duration (seconds) truncated at 500 ms is on the x axis. (B) Distribution of 1.65 million ISIs generated from inverse transform sampling of the data in A, which was the basis of our TOLSS approach. The mean open-loop ISI for all ISIs less than 250 ms is 0.048 s, with an overall mean ISI (up to 30 s) of 0.159 s. Count is on the y axis, and ISI duration (seconds) truncated at 500 ms is on the x axis.

Fig. 3.

TADSS enhances motor recovery over TOLSS and RT-only therapies. Over 13 wk of intervention, animals receiving TADSS-based therapy (red) recovered to a significantly greater extent than animals receiving either TOLSS (blue) or RT-only (black) therapies. Over the last 3 wk of intervention, TADSS animals had recovered 63% of their pre-SCI skilled reaching performance, whereas the TOLSS and RT-only cohorts averaged only 30% and 31% over this period, respectively. Performance of TOLSS and RT-only cohorts was statistically indistinguishable; y axis shows the weekly average reaching score across all rats in each cohort expressed as a percentage of each rat’s pre-SCI maximum. Error bars represent SEM. Statistical analyses were by a linear mixed model; detailed model parameters and results are available in SI Materials and Methods.

Fig. S3.

Quantile–quantile (QQ)/normality plots for TADSS, TOLSS, and RT-only longitudinal recovery datasets. To establish the validity of using a linear mixed model analysis for longitudinal reaching recovery/performance data, the normality of each cohort’s data was assessed using a combination of QQ plots, the Kolmogorov–Smirnov test, and quantitative assessments of kurtosis and skewness. Data were determined to be suitable for linear modeling using these metrics. For all other statistical tests (as noted in SI Materials and Methods), a Kolmogorov–Smirnov assessment of normality was performed; distributions failing to meet normality criteria were analyzed using nonparametric statistical techniques.

Fig. S4.

Spinal stimulation does not directly cause the intended movements. Representative catch trial data from the (A) TOLSS and (B) TADSS animals exhibiting the best recovery in each group. We assessed reaching performance with (red) and without (black; catch) stimulation at random times throughout the 13 wk of intervention, and we observed no significant differences in reaching performance between reach retraining sessions with stimulation and those without stimulation. This finding indicates that ISMS itself did not cause the intended movements. Thus, behavioral recovery is not attributable to direct effects of spinal stimulation on reaching performance. The y axis indicates reaching performance expressed as a literal number of food pellets successfully acquired out of a maximum of 20 (binary scoring), and the x axis indicates the week of intervention.

Fig. 4.

Therapeutic benefits persist after stimulation ends. Mean reaching performance during the last 3 wk of intervention and over 3 wk of additional observation with reduced physical retraining frequency and no stimulation. TADSS animals averaged 63% ± 12% and 57% ± 12% (P = 0.3), TOLSS animals averaged 30% ± 5% and 28% ± 8% (P = 0.59), and RT-only animals averaged 31% ± 12% and 33% ± 13% (P = 0.76), respectively. Paired t tests revealed no significant changes in performance within each group over the two time periods (SI Materials and Methods).

Fig. 5.

Timing of TADSS relative to STDP. (A) Single biphasic pulses (0.1 ms per phase) of ISMS (160 μA) were delivered at 0.5 Hz to a spinal motor region ipsilateral and caudal to SCI. Onset of stimulus-evoked responses in triceps occurs at a 1.3-ms latency and returns to baseline by 5 ms. Thin gray lines indicate individual stimulus-evoked EMG responses. The bold black line indicates stimulus-triggered average EMG activity. The red patch indicates stimulus artifact; the blue patch indicates stimulus-evoked response. Triceps EMG is on the y axis, and elapsed time from stimulus onset (milliseconds) is on the x axis. (B) Elapsed time from motoneuron action potential to triggered ISMS (i.e., TADSS) occurs in ∼10 ms or less, well within the optimal window for STDP facilitation. MEP, motor evoked potential.