Severe spinal cord contusions interrupt nearly all brain projections to lumbar circuits producing leg movement. Failure of these projections to reorganize leads to permanent paralysis. Here we modeled these injuries in rodents. A severe contusion abolished all motor cortex projections below injury. However, the motor cortex immediately regained adaptive control over the paralyzed legs during electrochemical neuromodulation of lumbar circuits. Glutamatergic reticulospinal neurons with residual projections below the injury relayed the cortical command downstream. Gravity-assisted rehabilitation enabled by the neuromodulation therapy reinforced these reticulospinal projections, rerouting cortical information through this pathway. This circuit reorganization mediated a motor cortex–dependent recovery of natural walking and swimming without requiring neuromodulation. Cortico–reticulo–spinal circuit reorganization may also improve recovery in humans.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We are grateful to Kissei Comtec (Japan) for providing the KinemaTracer software; J. Bloch for comments on the manuscript; S. Arber for various advices and providing plasmids; D. Trono and S. Offner for their support for the production of the viruses used in this study; E. Bezard for providing us with DREADD plasmids; T. Isa for providing us with plasmids; A. Karpova from the Janelia Virus Service Facility for providing the retro-AAV vectors; J. Courtine for the voiceover in movies; T. Laroche, R. Guiet and O. Burri at the EPFL BioImaging and optics Core Facility for their assistance; F. Pidoux, V. Padrun and A. Aebi for AAV vector production; and G. Ulrich, A. Nguyen, J.H. Ghelman, P.-Y. Helleboid, S. Ghazanfari, E. de Saint-Exupéry, M. Decroux and M.-C. Ung for their help with animal care and data analysis. Financial support was provided by the International Paraplegic Foundation (IRP), a Consolidator Grant from the European Research Council (ERC-2015-CoG HOW2WALKAGAIN 682999) and the Swiss National Science Foundation including an individual grant (subside 310030A_146925), a Bonus of Excellence (310030B_166674), R’Equip (subside 316030_145035), Sinergia (CRSII3_160696) and the National Center of Competence in Research (NCCR) Robotics.
Integrated supplementary information
(Timeline) Summary of the experimental procedures and timeline for the main groups of animals. (Step 1) Surgical implantation of chronic epidural electrodes over the midline of L2 and S1 spinal cord segments to deliver electrical neuromodulation therapies. Bipolar electrodes are inserted into a pair of flexor (tibialis anterior) and extensor (medial gastrocnemius) muscles of the ankle to record electromyographic activity. (Step 2) A severe contusion was performed at the mid-thoracic level. Leg kinematics and muscle activity are shown for a rat tested on a treadmill 1 week after contusion, both without and with electrochemical neuromodulation. (Step 3) Design of the task-specific training regimen throughout the period of recovery, including the transition from automatic stepping on a treadmill to overground walking with robotic assistance, to stair climbing. The features and time-dependent adaptations of the electrochemical neuromodulation therapy are shown. Briefly, the type and concentration of administered chemicals is constantly adjusted to the current motor performance of the rats. (Step 4) Behavioral tasks to evaluate leg motor control. (Step 5) Schematic overview of the functional inactivation and anatomical experiments that were used to evaluate the reorganization of neuronal pathways.
(a) Photograph of a coronal section through the contusion epicenter (GFAP, Glial fibrillary acidic protein), which was used to trace the contour of the contusion cavity, as illustrated below. Scale bar, 250 µm. (b) The contours of the lesion cavity at the epicenter are shown for all the main experimental rats used for behavioral and anatomical evaluations. (c) The bar graph reports the area of spared tissue at the lesion epicenter for pooled non-trained (n = 14) and trained (n = 13) rats. ns, not significant; Non-paired Student’s t-test.
(Step 1) Rats were evaluated during overground locomotion in a bipedal posture with robotic assistance. They were tested without neuromodulation, with electrical neuromodulation, and with electrochemical neuromodulation. Rats were trained to step overground with the gravity-assist during 9 weeks. Stepping performance (kinematics, muscle activity and ground reaction forces) was evaluated for different testing conditions at the end of the training period and compared to the motor output of non-trained rats. (Step 2) A PC analysis was applied on a total of 129 parameters characterizing global and fine details of leg motor control (Supplementary Table 2) table S1). The gait cycles of each rat under each neuromodulation condition (single dot, averaged of 10 to 20 gait cycles per rat) and averaged per group and conditions (large dots) are represented in new space defined by PC1 and PC2 (%, percent of explained variance). PC1 differentiated the effects of training, distinguishing the ability to move forward (walking) versus movements in place or falling backward (not able to walk). PC2 distinguished the effects of the neuromodulation conditions on leg motor control. (Step 3) The score of each rat on PC1 was extracted to quantify motor performance, i.e. the relative difference to intact (no injury) rats. The parameters that correlated highly with PC1 were extracted and regrouped in functional clusters that we named for clarity (# refers to Supplementary Table 2). The bar plots report the mean values for one parameter per each cluster, highlighted in yellow in the cluster. (Step 4) The same procedure was applied for parameters correlating with PC2, to identify the specific effects of neuromodulation conditions. *P < 0.05; **P < 0.01; ***P < 0.001. One-way ANOVA followed by Bonferroni’s post-hoc test.
(Step 1) To identify corticospinal tract neurons projecting to lumbar spinal segment, the retrograde tracer FastBlue was injected in the lumbar segments of Thy1:ChR2-YFP mice. Photographs show a sagittal and coronal view of FastBlue-labelled cells in the motor cortex. Scale bar 500 µm. The scheme shows a 3D reconstruction of all the labelled cells. (Step 2) Short bursts of optical stimulation were delivered using a mapping grid positioned over the motor cortex region identified in Step1, as shown in the photograph. The amplitude of the motor evoked potentials recorded in the contralateral flexor muscles of the ankle was measured for each site of stimulation in order to elaborate a motor map for each tested mouse (bottom, squares indicate hotspots), and all mice combined (top). (Step 3) Optic fibers were implanted chronically in the region identified in step 2. Stick diagram decomposition of leg movements, amplitude of leg oscillations, and electromyographic activity of the tibialis anterior (flexor) and gastrocnemius medialis (extensor) of the left and right legs in response to optical stimulation of the right motor cortex in suspended, awake mice. (Step 4) Mice received a contusion of the spinal cord at the T9 spinal segment level. For each experimental mouse, the lesion was reconstructed in order to visualize spared tissue (black) and lesion cavity (grey). (Step 5) Scheme showing the experimental conditions to test the mice with contusion. Bar plots reporting the mean speed and relative travelled distance under the different experimental conditions (n = 6 mice). Note the absence of behavioral effects when a non-specific yellow light was delivered over the motor cortex. ***P < 0.001. One-way ANOVA followed by Bonferroni’s post-hoc test. Bottom left, the latency between the delivery of optical stimulation and the onset of locomotion was measured as the duration between the onset of the optical stimulation and the first vertical elevation of the foot. Bottom right, the latency between the termination of optical stimulation and the cessation of locomotion was measured as the duration between the end of the optical stimulation and the zero crossing of the filtered hip velocity profile.
Supplementary Figure 5 The motor cortex regains adaptive control of leg movements during neuromodulation.
Leg kinematics were recorded during continuous stepping on a treadmill with different levels of optical intensities (laser output, %). A total of 133 parameters characterizing gait patterns were measured and submitted to a principal component (PC) analysis. The plot shows each gait cycle (small dot) in the new, reduced space created by PC1 and PC2. Large spots represent average per stimulation intensity. The bar plot on the right reports mean PC1 values, averaged over individual values for each gait cycle (individual dots). The number of gait cycles varies for each light intensity: n = 17 steps (40%), n = 11 (50%), n = 17 (60%), n = 13 (70%), n = 13 (80%), n = 14 (90%), n = 10 (100%). Laser output is expressed in percentage of the maximal light intensity (60mW). (b) The bar plots show average values of single gait parameters for each stimulation intensity. *P < 0.05; **P < 0.01; ***P < 0.001. One-way ANOVA followed by Bonferroni’s post-hoc test.
Supplementary Figure 6 Characterization of projection neurons with residual connection below the contusion in mice.
To visualize the interruption of the corticospinal tract in 3D, an AAV-Cre was co-injected with an AAV-flex-Synaptophysin-GFP into the motor cortex of Flex-tdTomato mice to label corticospinal tract fibers in the spinal cord. A large block of spinal cord containing the contusion was cleared using CLARITY. A representative 3D rendering of a CLARITY-processed spinal segment rostral to the contusion shows corticospinal tract fibers. Longitudinal photographs of spinal segments rostral and caudal to contusion show the synapses from corticospinal tract neurons. Scale bar 500 µm. (b) To verify the interruption of corticospinal tract projections independent of viral injections, spinal cord sections rostral (T6) and caudal (L1) to the contusion were stained with Protein Kinase C gamma (PKC-Gamma) antibodies, which label corticospinal tract axons. The bar plot reports the mean relative (arbitrary unit, a.u.) intensity of the PKC-Gamma signal in segments caudal and rostral to the contusion. Scale bars 500 µm, 50 µm insets. *P < 0.05. Paired Student’s t-test. (c) To quantify the interruption of corticospinal tract projections from the leg region of the motor cortex, fluorescence of the main corticospinal tract component in the dorsal column, the lateral component and the ventral component was measured in spinal cord sections rostral (T6) and caudal (L1) to the contusion. Scale bars 500 µm, 50 µm insets. Bar plots show the mean relative (arbitrary unit, a.u.) intensity of corticospinal tract axons projecting in the different regions of the white matter. (d) Diagram illustrating the injections of the retrograde tracer FastBlue in the lumbar spinal cord to identify projection neurons with residual connections below the contusion. The diagram illustrates the various regions where neurons were found and analysed. Bar plots report the number of neurons in these regions (n = 4 each for intact and contused mice). 3D cervical (e) and (f) thoracic spinal cord reconstructions in intact and contused mice, including quantifications. *P < 0.05; **P < 0.01. Mann-Whitney test. (g) The phenotype of brainstem neurons with residual projections below the contusion was studied with immunohistochemistry. Photographs show neurons retrogradely labelled from lumbar segments and the co-localisation with glutamate or serotonin (white arrows). Scale bars 25 µm. The polar plots display the mean percentage (+- s.e.m) of neurons co-localising with glutamate or serotonin with respect to the entire population of neurons retrogradely labelled from lumbar segments in intact mice (n = 4 mice).
(Step1) Strategy to express DREADDs specifically in glutamatergic neurons of the vGi, while manipulating the activity of motor cortex projection neurons with light (n = 6 mice). Photograph showing expression of hM4Di in vGluT2ON vGi neurons. Scale bar, 500 µm. (Step 2) Intact (no injury) mice were tested before and after CNO during quadrupedal locomotion along a flat surface and a horizontal ladder. A PC analysis was applied on 104 gait parameters (see Supplementary Table 2) characterizing the walking pattern. Gait cycles (small dot, individual gait cycles; large dots, average per mouse and condition) measured along the flat surface are represented in the new space created by PC1 and PC2 (%, explained variance). No significant difference was detected between both conditions. The stick diagram decomposition of leg movements shows the progression along the horizontal ladder. The polar plot reports the mean relative percent of hits, misses and slips onto the rungs of the ladder for both conditions (+- s.e.m). Silencing vGluT2ON vGi neurons did not alter skilled leg movements. (Step 3) The same mice received a severe contusion. For each experimental mouse, the lesion was reconstructed in order to visualize spared tissue (black) and lesion cavity (grey). (Step 4) Experimental conditions to evaluate the cortical control of leg movements before and after silencing vGluT2ON vGi neurons. Mice were tested bipedally under chemical neuromodulation (5HT agonists). The bar plots represents the mean distance travelled and the mean stepping frequency when delivering optical motor cortex stimulation without and with CNO (n = 6 mice). The snapshots illustrate the forward tilt of the trunk and whole leg extension when delivering optical stimulation. The bar plots report the mean values of these parameters (n = 6 mice). **P < 0.01; ***P < 0.001. Two-tailed paired Student’s t-test. (Step 5) Post-mortem evaluation of hM4Di positive neurons in the vGi and potential infection in neighbouring regions for all experimental animals (n = 6 mice), shown as mean +- s.e.m. The coronal and dorsoventral snapshots of the brainstem show the 3D volume containing vGluT2ON hM4Di expressing neurons. Post-mortem evaluation of hM4Di positive neurons in the lateral vestibular nuclei (n = 4 mice) shown as mean +- s.e.m., and 3D volume containing vGluT2ON hM4Di expressing neurons is shown in the right panel.
Supplementary Figure 8 Characterization of projection neurons with residual connection below the contusion in rats, and reorganization of motor cortex projections in the brainstem.
(a) Diagram illustrating the injections of the retrograde tracer FastBlue in the lumbar spinal cord to identify projection neurons with residual connections below the contusion in rats. Coronal and dorsoventral snapshots of 3D brain and brainstem reconstructions in intact and contused (non-trained) rats. Each neuron is represented by a single dot. (b) The bar plots report the relative number of identified neurons in the whole brainstem (n = 4 each for intact and contused rats) and selected regions from the brainstem (n = 6 each for intact and contused). (c) 3D cervical and thoracic spinal cord reconstructions in intact and contused rats, including quantifications. *P < 0.05;**P < 0.01; ***P < 0.001. Non-paired Student’s t-test (b) or Mann-Whitney test (c). (d and e) Scheme illustrating motor cortex injections and regions for the quantification of motor cortex projections. Representative photographs and corresponding heat maps of motor cortex projections in the lateral vestibular nuclei (d) and red nuclei (e) of intact (n = 6), non-trained (n = 6) and trained (n = 6) rats. Scale bars, 200 µm. *P < 0.05, One-way ANOVA followed by Bonferroni’s post-hoc test.
Supplementary Figure 9 Silencing of leg motor cortex neurons abolishes leg motor control in trained rats.
(Timeline) Timeline of experimental procedures in intact (n = 6) and contused rats (n = 5). (Step 1) Intact rats were evaluated during locomotion along the irregularly spaced rungs of a ladder, before and after CNO administration. The circular plots report the percent of hits, misses and slips onto the rungs of the ladder for both conditions. (Step 2) The rats received a 250 kdyn contusion. (Step 3) They followed the standard neurorehabilitation protocols as described in Supplementary Fig. 2. (Step 4) Stick diagram decomposition of leg movements during swimming. Successive leg endpoint trajectories, including velocity vector at return stroke onset, together with the oscillation of the limb and occurrence of left and right power strokes during swimming. A representative sequence is shown for intact and contused rats, both without and with CNO. Contused rats were tested without neuromodulation. ***P < 0.001. Paired Student’s t-test. (Step 5) 3D reconstruction of the cortical region with neurons expressing hM4Di, including a photograph through the coronal plane marked (1). Scale bar, 500 µm.
Supplementary Figure 10 Inactivation of vGi neurons projecting below the contusion abolishes leg motor control in trained rats.
(a) Timeline of experimental procedures and strategy for the reversible inactivation of vGi neurons with projections to lumbar segments, both in intact (n = 5 rats) and contused rats (n = 5 rats). (b) Intact rats (no injury) were evaluated during locomotion along the irregularly spaced rungs of a ladder, before and after repeated doxycycline administration. The circular plots report the percent of hits, misses and slips onto the rungs of the ladder for both conditions (c) PC analysis using the same conventions as in Fig. 1. All the individual gait cycles are shown, as well as the average for each rat. The bar plot near the PC analysis reports the mean values of the score on PC1, which differentiated trials with forward progression (walking) from trials without initiation (not walking). The other bar plots report the mean values of basic gait parameters illustrating the robust deterioration of leg motor control during inactivation of vGi neurons with residual projections to lumbar segments in trained rats. **P < 0.01; ***P < 0.001. One-way ANOVA followed by Bonferroni’s post-hoc test. (d) 3D reconstruction of all neurons expressing both transgenes, including photographs showing the morphology of these neurons. Scale bars, 25 µm.
(a) Scheme showing the location at which the density of 5HT axons was evaluated. Density plot of 5HT axons along the dorsoventral extent of the spinal cord (bold line represents the mean distribution for all animals, shaded region represents s.e.m.), and bar graphs reporting the mean axon density of 5HT axons and individual animals within L4/L5 spinal segments. (b) Representative heat maps of 5HT axons, visualized at L4/L5. (c) Representative images of 5HT axons in the lamina 9 of L4 segment. Scale bar, 25 µm (d) Bar graph reporting the mean caliber of 5-HT axons at L4/L5. Scale bars, 25 μm.*P < 0.05; **P < 0.01. One-way ANOVA followed by Bonferroni’s post-hoc test.
Supplementary Video 1: Neurorehabilitation procedures restore supraspinal control of the paralyzed legs.
To model a severe spinal cord contusion in rats, we delivered a robotically controlled impact to thoracic segments. One week after injury, all rats failed to activate motor neurons below the injury, which resulted in flaccid paralysis of both legs. To reactivate lumbar circuits, we delivered an electrochemical neuromodulation therapy consisting of agonists to serotonergic and dopaminergic receptors, and epidural electrical stimulation applied to lumbar and sacral segments. These stimulations instantly restored automated movements of the paralyzed legs in response to treadmill belt motion. After completion of training, all the rats regained the ability to transform contextual information into task-specific motor commands that allowed the execution of natural motor behaviors, even in the absence of neuromodulation therapies.
Supplementary Video 2: Motor cortex instantly regains control over the paralyzed legs during spinal cord neuromodulation.
We manipulated the activity of motor cortex projection circuits with optical stimulation in mice with severe spinal cord contusion. Increase in stimulation intensity during locomotion induced a proportional augmentation of leg muscle activation. These experiments showed that neuromodulation therapies instantly enabled motor cortex projection circuits to trigger, sustain and modulate locomotor movements of the paralyzed legs.
Supplementary Video 3: Glutamatergic neurons in the ventral gigantocellular relay cortical information to lumbar circuits.
To identify projection neurons that maintained connectivity across the contusion and could be responsible for relaying the cortical command downstream, we conducted anterograde and retrograde neuroanatomical tracings. CLARITY-optimized light-sheet microscopy showed the depletion of corticospinal tract synapses within lumbar segments and emphasized that the largest number of spared synaptic projections in lumbar segments originated from ventral gigantocellular (vGi) glutamatergic neurons. We then used DREADD to specifically and reversibly silence glutamatergic neurons during optical activation of motor cortex projection neurons. Locomotion triggered by optical stimulation of the motor cortex during neuromodulation was abolished during silencing of vGi neurons.
Supplementary Video 4: Motor recovery relies on the reorganization of motor cortex and vGi projection circuits.
We first tested the contribution of vGi neurons in the recovery of leg motor control in rats using a doxycycline-inducible tetanus toxin technique that allowed the reversible inactivation of vGi neurons with synaptic projections to lumbar segments. The inactivation of vGi neurons with residual projections below injury suppressed the recovery of motor function in trained rats. We then evaluated whether the anatomical reorganization of motor cortex projection neurons in trained rats is associated with a functional contribution of leg motor cortex to motor control. To this end, we silenced this brain region bilaterally using DREADD. After contusion and training, the rats regained the ability to swim without neuromodulation. Silencing the motor cortex abolished volitional leg movements, preventing the rats from performing the task.