First-in-human study of epidural spinal cord stimulation in individuals with spinal muscular atrophy

a-b Rheobase for WT and SMA motoneuron biophysical models. To compute the rheobase we slowly increased the input current injected to the motoneuron’s soma (a) until a spike was triggered (b). The black and purple lines indicated the rheobase, the smallest amplitude that triggers an action potential. The SMA motoneuron biophysical model reproduces the active hyperexcitability characterized by reduced rheobase. c, Example of 2 second isometric force produced by a SMA-affected motoneuron pool simulated with the biophysical model. From bottom to top: Raster plot of 11 Ia-afferent fibers recruited by SCS at 40 Hz. Note that spikes are completely synchronized because they are triggered by SCS pulses. Supraspinal neurons with natural firing rate (Poisson neurons); for visualization propose, we only show 11 neurons but the supraspinal population has 200 neurons. Motoneuron pool firing rate with 30% of motoneuron loss and 50% SMA-affected motoneurons. Simulated force produced by the motoneuron pool normalized by the maximum force produced by an intact motoneuron pool. d, SMA-affected motoneuron pool firing rate for different simulated percentage of motoneuron loss and SMA-affected motoneurons. e, Simulated force for different percentages of motoneuron loss and SMA-affected motoneurons. f, Motoneuron pool firing rate as a function of the supraspinal firing rate with SCS ON and OFF. The biophysical model predicted that the effect of SCS should be more prominent on low than high supraspinal inputs indicating that if all motoneurons are recruited at their maximum firing rate, there should be no effect of SCS. Thus, the effect of SCS at maximum voluntary contraction will depend on the participants’ ability to recruit the resources of their motoneuron pool. SCS could give access to motoneurons that are not recruited or recruited with a submaximal firing rate. g, Simulated force with a low supraspinal input (5 Hz) and very high supraspinal input (50 Hz). h, Firing rate for a SMA-affected pool with all SMA-affected motoneurons and 40% rescued which is the percentage of rescued motoneurons for SMA01 Right Knee Extension. i, Predicted forces by the biophysical model given the percentage of motoneurons rescued in SMA01 Right Knee Extension and SMA03 Right Knee Flexion. First, we simulated the torque produced with a SMA-affected pool with 70 motoneurons all affected by SMA (light purple). In purple we simulated the torque rescuing the same percentage that we found experimentally: 40% for SMA01 Right Knee Extension and 25% SMA03 Right Knee Flexion. Finally, we used a bootstrap method to generate multiple predictions of the percentage of rescued units per participant, to estimate a maximum possible upper limit that could occur by random variations in the data (95th percentile). Using this method we found that the maximum percentages of rescued motoneuron consistent with our data were 80% and 50% for SMA01 Right Knee Extension and SMA03 Right Knee Flexion respectively. We simulated the torques with these upper limits (dark purple) and we found that the experimental values (dash black line) fell below the upper boundary. For SMA01 our biophysical model predicted an increase of 24% and up to 69% which is consistent with the experimental increase in right knee extension +55%. For SMA03 our model predicted an increase in torque of 15% and up to 42% which is remarkably consistent with the +21% increase that we found in right knee flexion. All simulated torques are normalized by the torque produced by a SMA-affected motoneuron pool with 70 SMA-motoneurons. j, Simulated peak to peak of the TMS MEP for two hypotheses: reversion of the maladaptive changes produced by SMA (SMA rescued) and reduction of the supraspinal synaptic weight. k, Examples of simulated MEP. Both hypotheses are consistent with experimental findings of smaller MEP. l,m, Simulated firing rates as a function of the supraspinal inputs (l) and simulated torque (m) for the two hypotheses. Only the hypothesis of rescued motoneurons is consistent with the experimental findings of higher firing rates and torques.

a, d, g, For each study participant demographic information, including ethnicity, age, score at the enrollment on the Hammersmith Functional Motor Scale Expanded (HFMSE), and detail on any treatment used. b, e, h. Left: pre implant T2 MRI, with the approximate lead ___location in red. Right: X-ray post-implant, with the bilateral leads in black. All three participants had the conus medullaris under the vertebra T12, a vertebra above what is normally observed (L1) c, f, i, SCS parameters used during walking. We used a program targeting the right and a program targeting the left leg muscles. Each program applied a fixed amplitude, frequency and pulse width for each SCS pulse. j, l, n. Configurations of the experimental set ups for passive knee joint movement for all participants and the SCS parameters used (anode blue, cathode red). k, m, o. Muscle responses were used to compute the peak to peak of each bin. Plots are reporting the mean and 95% CI of the peak-to-peak responses. All error bars indicate the confidence interval, computed with bootstrap (N = 10,000). p. The cycle of joint oscillations was divided into 10 bins of equal duration during which the muscle responses were extracted and regrouped. q. Configuration of the experimental set up for proprioception testing (Supplementary Methods). Randomly selected flexion or extension movement were imposed to the knee joint of each subject at a speed of 0.5 degree per second. r. Configurations of SCS parameters used (anode blue, cathode red) and the detection angle (the difference between the starting angle and final angle displacement) plotted for each participant under varying SCS conditions of the proprioception testing: SCS OFF (black), SCS ON (light blue). The SCS ON condition was continuous stimulation at 40 Hz. Statistical differences were assessed with two-sided bootstrap (N = 10,000). *, **, and *** denote significant differences with p-values of p < 0.05, p < 0.01, and p < 0.001, respectively.

a,Maximum strength in hip and knee for all participants at pre-implant, post-explant and follow up (6 weeks after post explant). To reduce the day-to-day variability in our measures of maximum torque, we used data from all session’s pre-implant and post-explant (see Extended Data Fig. 4b–f). We corrected for multiple comparison using Bonferroni with N = 3 (pre-implant vs post-explant, pre-implant vs follows up and post-explant vs follow up) b, Maximum strength with SCS ON and OFF for all sessions normalized by the mean maximum torque with SCS OFF across repetitions for of each session. Error bars show the mean and 95% confidence interval of the mean, computed with bootstrap (N = 10,000). Each dot represents the torque of a single repetition. All statistical significance was assessed with two-sided bootstrapping (N = 10,000): p < 0.05 (*), p < 0.01 (**), p < 0.001(***).

a, Experimental set up to measure isometric torques for knee extension and flexion. b-f Maximum voluntary contraction in all muscles, sessions and subjects that we could reliably measure (torque above 4 N): SMA01 Hip (b), SMA01 Knee (c), SMA02 Hip (d), SMA03 Hip (e) and SMA03 (f). Each point corresponds to one repetition and the square represents the mean across repetitions. g, We used a multivariable linear regression analysis to study the effect of fatigue in our measurements of torque (see Methods). For SMA01, we found that the largest regressor was the repetition indicating that SMA01 was fatiguing after only a few repetitions of maximum voluntary contractions. Given that we always started with SCS OFF trials, our results could be underestimating the assistive effect of SCS (Fig. 3a-c). Indeed, we found for SMA01 right knee extension, the interaction between the SCS and the repetition number was significant indicating that SCS increased maximum torque when SMA01 was fatigued. In SMA03 we reduced the trial duration time from 5 to 2 seconds and we found that the repetition number had no impact indicating that direct comparison between SCS OFF and ON trial is probably not affected by the repetition number. h,Torque traces for SCS ON and SCS OFF fatigued repetition (after more than 12 repetitions of maximum voluntary contraction) in day 12 (See supplementary methods for more information). i, Maximum torque at right knee extension produced by SMA01 in fatigued trials for each session. j, Data from all sessions pooled where we tested the maximum torque in fatigued condition. All error bars show the mean and 95% confidence interval of the mean, computed with bootstrap (N = 10,000).

a. Gait variables (step length, step height, and stride velocity) for SCS ON and OFF for all sessions normalized by the median value with SCS OFF across repetitions for each session. P-values corrected with Bonferroni correction (N = 2, 4 and 4 for SMA01, SMA02 and SMA03 respectively) b, Same gait variables for each session with SCS OFF. We used Kruskal Wallis ANOVA to find significance differences between groups (all p-values < 0.001). To study if the improvements saturated, we compared each week with the following week. In most gait quality variables, the improvements do not saturate (significance difference between week 3 and week 4). P-values corrected using Bonferroni correction (N = 3, 4 and 4 for SMA01, SMA02 and SMA03 respectively) c, Therapeutics effects in gait quality variables (week 1 vs week 4). All gait quality variables improved after the SCS intervention. d. Left, Examples of isometric torque traces during maximum voluntary knee extension contraction with SCS ON, SCS OFF, and with sham SCS parameters. The sham consisted of an SCS configuration chosen not to target the specific muscles involved in the joint movement tested changing the active contacts or the amplitude or frequency of the stimulation. Right, Maximum torque during knee extension for SCS ON, OFF, and sham SCS parameters. Dots represent individual repetitions, while error bars indicate mean and the 95% CI of the mean (CI). e. Maximum torque during Hip Flexion for SMA01 using the same program with different stimulation amplitudes. Dots represent individual repetitions, while error bars indicate mean and the 95% CI of the mean (CI). f. Left, Mean heel marker trajectory across gait cycles with SCS ON using same program and different frequencies. Right, gait quality variables computed under the same conditions. P-values corrected using Bonferroni correction (N = 7). g, Setup for testing exaggerated hip flexion walking on the anti-gravity treadmill (Supplementary Methods). h. Left, Examples of metatarsal marker trajectory during exaggerated hip flexion walking with SCS ON and OFF. Right, The maximum hip flexion increased with SCS ON. i, same to h, considering SCS OFF in week 1 vs week 4. Again, the maximum hip flexion increased from week 1 to week 4 even without SCS. Box plots represent the median, 25th and 75th percentile and minimum and maximum data points without outliers. All statistical significance was assessed with two-sided bootstrapping (N = 10,000) and corrected for multiple comparisons: p < 0.05 (*), p < 0.01 (**), p < 0.001(***).

a. Scheme depicting the placement of the EMG sensor during overground walking (Supplementary Methods). b-c Assistive effects in range of motion (ROM) for SMA03 during week1. b, Top: Mean of hip flexion/extension across gait cycles with SCS OFF (black) and SCS ON (light blue). Bottom: Mean envelope of Semitendinosus (Semi) across the gait cycle with SCS OFF (black) and SCS ON (light blue). The gold shades highlight semitendinosus activation in the pre-swing, contributing to hip extension during the same phase of the cycle. c. Top: Mean change in ROM for the left hip and knee across gait cycles (n = 18, SCS ON; n = 17 SCS OFF). Bottom: Change in EMG activity, computed as the root mean square, for Semitendinosus (Semi), Vastus Lateralis (VastLate), and Biceps Femoris (BicepFem). Improvements in hip ROM should be correlated with the EMG activity at Semi and BicepFem (muscles involved in hip extension). Improvements in knee extension should be related to higher activation on the knee extensors. No change of EMG or ROM. d-e Mean therapeutic improvement in ROM during a gait cycle for SMA03 (n = 17, week 1; n = 16 week 4). Improvements in hip and knee ROM are paralleled by an increase in EMG activity at hip and knee extensors. f-g. Mean assistive improvement in ROM during a gait cycle (n = 60 SCS OFF; n = 28 SCS ON) for SMA02 in week 1. The gray shades highlight Rectus Femoris proximal activation in the swing phase, contributing to hip flexion during the same phase of the cycle. An increase in hip ROM can be explained by an increase in EMG activity at the Rectus Femoris proximal (a proxy of EMG activity of the hip flexors such as the iliopsoas). h-j. Therapeutic improvement in ROM during a gait cycle (n = 60, week 1; n = 18 week 4) for SMA02. Improvements in hip and knee ROM are correlated with increases in EMG activity of hip flexors (rectus femoris proximal) and knee flexors (Semi) No data is reported for SMA01 due to EMG acquisition corruption by noise. Error bars show the mean and 95% confidence interval of the mean, computed with bootstrap (N = 10,000).

Comparison with all exercise-only control participants who completed three months of aerobic and strength training, including 3 minors. Changes for the control group were computed by comparing pre- and post- 3 months exercise intervention data, while changes for the SCS group were computed by comparing pre-implant and post-explant data, except otherwise specified. a, Clinical assessments. HFMSE and 6MWT measured pre-implant, post-explant and 6 weeks follow up. In parenthesis changes from baseline measures. All tests executed with SCS OFF. 6MWT dropped back to pre-implant at 6 weeks follow up in 2/3 participants. b, Relationship between age and age of onset for exercise-only control group (black) and the participants in our study (blue). c, Histogram of HFMSE scores for the control group (black), blue lines represent the HFMSE scores for each SCS participant. d, Histogram of aerobic exercise time for the control group (black), and exercise time of each SCS participant (blue lines). e, left: Relationship between the change in 6MWT score and the HFMSE score for the controls (R² = 0.001) and the SCS participants (R² = 0.98). Right: Mean and 95%CI of the mean change in 6MWT score for the controls (black) and changes for SCS participants (blue). f, Left: Linear regression analysis between the change in fatigue index during the 6MWT and the severity for both the control group (R² = 0.15) and the SCS participants (R² = 0.7). Right: Mean and 95% CI of the mean change in fatigue index across controls (black) and changes for SCS participants (blue) g, left: Regression analysis between the change in stride length and severity for controls (R² = 0.01), blue dots indicate SCS participants. Right: Mean change in stride length for the controls (black), and changes for SCS participants (blue). h, Same as g for stride velocity (R² = 0.07). For the SCS participants, changes were computed during self-paced overground walking, comparing SCS OFF trials from week 1 and week 4. i, Left: Regression analysis between the change in knee flexion strength for the controls (R² = 0.05). Improvements for SMA01 and SMA03 and their confidence intervals are shown in blue. Right: Mean knee flexion strength for the controls (black) and SCS participants (blue). j, Same as i for knee extension (R² = 0.48). Error bars represent the 95% confidence interval of the mean, computed with bootstrapping (N = 1000).

a. Pipeline to extract single motor units spike trains from high density surface electromyography (HDEMG). We used an 8 × 8 channel flexible grid to record muscle activity from both knee extensor (Rectus Femoris (RF)) and flexor (Biceps Femoris (BF)). Participants performed two sets of three isometric Maximum Voluntary Contractions (MVC) of knee extension and flexion. We used the convolution kernel compensation (CKC) method to decompose the EMG signals in single motor units spike trains. b. Rescued motoneuron identification: a motor unit was defined as rescued if its mean peak firing rate (across trials) was higher than the 99.73 % CI (red line) of mean peak firing rate across motor units recorded during the first session. The 99.73% CI was computed with bootstrap (N = 10,000). c. Example of motor units smoothed discharge rate (SDR) obtained during the 3rd MVC. Each line represents the SDR of every motoneuron. Light and dark colors delineate the data from the first and last session where we performed the experiment d. Circles illustrate the number of identified motoneurons over the different phases of the study. (that is, PI = pre-implant; Wi = i-th week after the implant; PE = post-explant). e, Example (SMA02) of the mean and 95% CI of post-explant single motoneurons firing rates during MVC. In gray regions indicate the start and end windows used to compute firing rate at the beginning and the end of the MVC. f, Percentage change in the mean firing rate between rescued and dysfunctional motoneurons at the start (first 150 ms) and end (last 150 ms) of the MVC for each participant. 95% CI were computed using bootstrapping (n = 10,000). We assessed statistical significance using a two-sided test with bootstrapping (n = 10000). Asterisk *, **, and *** denote significant differences with p-values of p < 0.05, p < 0.01, and p< 0.001, respectively.

Step 1, Acquisition of functional MRI from the spinal cord in response to the recruitment of motor neurons from specific leg muscles. The motor neurons are recruited by actively extending and flexing the knee joint. Three runs are acquired for each participant. Only the right leg muscles are tested. Instructions were displayed on a screen (fixation cross ‘+’ during rest blocks and text indicating activity blocks), while auditory cues were used to signal changing phases. In addition to the functional volume series, T2 anatomical images and physiological (heart rate, respiratory) signals are acquired. Step 2, Raw fMRI volume series are repeatedly acquired every 2.5 s (TR) in functional runs lasting about 6 minutes. Step 3, Spinal segments are identified from high-definition T2-ZOOMit structural images, T2 images and SCS recruitment curve responses. Spinal segments are then reported in the T2 anatomical image acquired in each run. Step 4, Then motion correction is applied to each functional run in two stages. First, volumes are aligned to the averaged images using 3D rigid body realignment. Second, a slice-by-slice 2D realignment is applied to each volume. Step 5, The spinal cord (including gray and white matter) is automatically segmented from the motion corrected mean functional image, and then manually corrected. Step 6, Functional images from runs two and three are aligned to functional images of run one using the motion corrected mean images by landmarking the conus medullaris and any visible vertebra, using rigid transformations. Step 7, The motion corrected mean images are then coregistered to the PAM50 template using the conus medullaris landmark to extract the transformation matrices, using non-rigid transformations. Step 8, Physiological signals (heart rate and respiratory) acquired concomitantly to the fMRI volumes are used to model physiological noise (RETROICOR based procedure). Acquisition timings corresponding to motion corrected fMRI volume series and physiological noise regressors are submitted to a specific first level generalized linear model (GLM). Residuals from this GLM are then spatially smoothed, volume by volume with 3D gaussian kernel with full width at half maximum (FWHM) of 2x2x6mm3. Acquisition timings corresponding to the task-design and residuals resulting from noise removal are submitted to a specific first level GLM. A second level fixed effects analysis (subject level, task specific) is performed by combining the three runs. Activity pattern maps are uncorrected and thresholded (Z > 2.3, p < 0.01). Step 9, Transformations matrices found in step 7 are then applied to the resulting activity pattern maps of step 8 to arrive in PAM50 space. Step 10, Activity patterns from the three participants, are then compared and analyzed across scanning sessions using the spinal segments found in step 3.

a, Quantification of percent change in area under the curve of peak-to-peak TMS responses post-explant vs six week follow up across all intensities tested. Abbreviations: ST, Semitendinosus; BF, Bicep Femoris; VL, Vastus Lateralis; RP, Rectus Femoris Proximal; RD, Rectus Femoris Distal; G, Gastrocnemius; TA, Tibialis Anterior. Error bars are the mean and 95% CI of the mean peak-to-peak responses (bootstrap N = 10,000). b, The SCS parameters used and the resulting plots representing the normalized peak-to-peak response as a function of SCS amplitude. Shaded areas correspond to standard error across responses, while thick lines are the means across responses.