Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Activity-dependent spinal cord neuromodulation rapidly restores trunk and leg motor functions after complete paralysis

Abstract

Epidural electrical stimulation (EES) targeting the dorsal roots of lumbosacral segments restores walking in people with spinal cord injury (SCI). However, EES is delivered with multielectrode paddle leads that were originally designed to target the dorsal column of the spinal cord. Here, we hypothesized that an arrangement of electrodes targeting the ensemble of dorsal roots involved in leg and trunk movements would result in superior efficacy, restoring more diverse motor activities after the most severe SCI. To test this hypothesis, we established a computational framework that informed the optimal arrangement of electrodes on a new paddle lead and guided its neurosurgical positioning. We also developed software supporting the rapid configuration of activity-specific stimulation programs that reproduced the natural activation of motor neurons underlying each activity. We tested these neurotechnologies in three individuals with complete sensorimotor paralysis as part of an ongoing clinical trial (www.clinicaltrials.gov identifier NCT02936453). Within a single day, activity-specific stimulation programs enabled these three individuals to stand, walk, cycle, swim and control trunk movements. Neurorehabilitation mediated sufficient improvement to restore these activities in community settings, opening a realistic path to support everyday mobility with EES in people with SCI.

This is a preview of subscription content

Access options

Buy article

Get time limited or full article access on ReadCube.

$32.00

All prices are NET prices.

Fig. 1: Anatomical quantification and personalizable computational models.
Fig. 2: Optimal arrangement of electrodes.
Fig. 3: Preoperative planning and intraoperative validation.
Fig. 4: Configuration of activity-dependent stimulation programs.
Fig. 5: Configuration of trunk-specific stimulation programs.
Fig. 6: Recovery of independence in the community.

Data availability

Data that supports the findings are available in the following data depository:

https://doi.org/10.5281/zenodo.5614586Source data are provided with this paper.

Code availability

Software routines developed for the data analysis will be made available upon reasonable request to gregoire.courtine@epfl.ch.

References

  1. Ichiyama, R. M. et al. Step training reinforces specific spinal locomotor circuitry in adult spinal rats. J. Neurosci. 28, 7370–7375 (2008).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  2. Courtine, G. et al. Transformation of nonfunctional spinal circuits into functional states after the loss of brain input. Nat. Neurosci. 12, 1333–1342 (2009).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  3. Wenger, N. et al. Closed-loop neuromodulation of spinal sensorimotor circuits controls refined locomotion after complete spinal cord injury. Sci. Transl. Med. 6, 255ra133–255ra133 (2014).

    PubMed  Article  CAS  Google Scholar 

  4. Wenger, N. et al. Spatiotemporal neuromodulation therapies engaging muscle synergies improve motor control after spinal cord injury. Nat. Med. 22, 138–145 (2016).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  5. Brand, Rvanden et al. Restoring voluntary control of locomotion after paralyzing spinal cord injury. Science 336, 1182–1185 (2012).

    PubMed  Article  CAS  Google Scholar 

  6. Asboth, L. et al. Cortico–reticulo–spinal circuit reorganization enables functional recovery after severe spinal cord contusion. Nat. Neurosci. 21, 576–588 (2018).

    CAS  PubMed  Article  Google Scholar 

  7. Gill, M. L. et al. Neuromodulation of lumbosacral spinal networks enables independent stepping after complete paraplegia. Nat. Med. 24, 1677–1682 (2018).

    CAS  PubMed  Article  Google Scholar 

  8. Wagner, F. B. et al. Targeted neurotechnology restores walking in humans with spinal cord injury. Nature 563, 65–71 (2018).

    CAS  PubMed  Article  Google Scholar 

  9. Harkema, S. et al. Effect of epidural stimulation of the lumbosacral spinal cord on voluntary movement, standing, and assisted stepping after motor complete paraplegia: a case study. Lancet 377, 1938–1947 (2011).

    PubMed  PubMed Central  Article  Google Scholar 

  10. Angeli, C. A. et al. Recovery of over-ground walking after chronic motor complete spinal cord injury. N. Engl. J. Med. 379, 1244–1250 (2018).

    PubMed  Article  Google Scholar 

  11. Danner, S. M. et al. Human spinal locomotor control is based on flexibly organized burst generators. Brain 138, 577–588 (2015).

    PubMed  PubMed Central  Article  Google Scholar 

  12. Formento, E. et al. Electrical spinal cord stimulation must preserve proprioception to enable locomotion in humans with spinal cord injury. Nat. Neurosci. 21, 1728–1741 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  13. Herman, R., He, J., D’Luzansky, S., Willis, W. & Dilli, S. Spinal cord stimulation facilitates functional walking in a chronic, incomplete spinal cord injured. Spinal Cord 40, 65–68 (2002).

    CAS  PubMed  Article  Google Scholar 

  14. Barolat, G., Myklebust, J. B. & Wenninger, W. Enhancement of voluntary motor function following spinal cord stimulation: case study. Appl. Neurophysiol. 49, 307–314 (1986).

    CAS  PubMed  Google Scholar 

  15. Dimitrijevic, M. R., Gerasimenko, Y. & Pinter, M. M. Evidence for a spinal central pattern generator in humans. Ann. N Y Acad. Sci. 860, 360–376 (1998).

    CAS  PubMed  Article  Google Scholar 

  16. & Darrow, D. et al. Epidural spinal cord stimulation facilitates immediate restoration of dormant motor and autonomic supraspinal pathways after chronic neurologically complete spinal cord injury. J. Neurotrauma 36, 2325–2336 (2019).

    PubMed  PubMed Central  Article  Google Scholar 

  17. Gerasimenko, Y. P. et al. Spinal cord reflexes induced by epidural spinal cord stimulation in normal awake rats. J. Neurosci. Methods 157, 253–263 (2006).

    PubMed  Article  Google Scholar 

  18. Rattay, F. et al. Mechanisms of electrical stimulation with neural prostheses. Neuromodulation 6, 42–56 (2003).

    CAS  PubMed  Article  Google Scholar 

  19. Capogrosso, M. et al. A computational model for epidural electrical stimulation of spinal sensorimotor circuits. J. Neurosci. 33, 19326–19340 (2013).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  20. Moraud, E. M. et al. Mechanisms underlying the neuromodulation of spinal circuits for correcting gait and balance deficits after spinal cord injury. Neuron 89, 814–828 (2016).

    CAS  PubMed  Article  Google Scholar 

  21. Capogrosso, M. et al. A brain–spine interface alleviating gait deficits after spinal cord injury in primates. Nature 539, 284–288 (2016).

    PubMed  PubMed Central  Article  Google Scholar 

  22. Minev, I. R. et al. Electronic dura mater for long-term multimodal neural interfaces. Science 347, 159–163 (2015).

    CAS  PubMed  Article  Google Scholar 

  23. Capogrosso, M. et al. Configuration of electrical spinal cord stimulation through real-time processing of gait kinematics. Nat. Protoc. 13, 2031–2061 (2018).

    CAS  PubMed  Article  Google Scholar 

  24. Yakovenko, S., Mushahwar, V., VanderHorst, V., Holstege, G. & Prochazka, A. Spatiotemporal activation of lumbosacral motoneurons in the locomotor step cycle. J. Neurophysiol. 87, 1542–1553 (2002).

    PubMed  Article  Google Scholar 

  25. Cappellini, G., Ivanenko, Y. P., Dominici, N., Poppele, R. E. & Lacquaniti, F. Migration of motor pool activity in the spinal cord reflects body mechanics in human locomotion. J. Neurophysiol. 104, 3064–3073 (2010).

    PubMed  Article  Google Scholar 

  26. Molnar, G. & Barolat, G. Principles of cord activation during spinal cord stimulation. Neuromodulation 17, 12–21 (2014).

    PubMed  Article  Google Scholar 

  27. Minassian, K., Hofstoetter, U., Tansey, K. & Mayr, W. Neuromodulation of lower limb motor control in restorative neurology. Clin. Neurol. Neurosurg. 114, 489–497 (2012).

    PubMed  PubMed Central  Article  Google Scholar 

  28. Moraud, E. M. et al. Closed-loop control of trunk posture improves locomotion through the regulation of leg proprioceptive feedback after spinal cord injury. Sci. Rep. 8, 76 (2018).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  29. Gill, M. et al. Epidural electrical stimulation of the lumbosacral spinal cord improves trunk stability during seated reaching in two humans with severe thoracic spinal cord injury. Front. Syst. Neurosci. 14, 79 (2020).

    PubMed  PubMed Central  Article  Google Scholar 

  30. Greiner, N. et al. Recruitment of upper-limb motoneurons with epidural electrical stimulation of the cervical spinal cord. Nat. Commun. 12, 435 (2021).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  31. Neufeld, E., Szczerba, D., Chavannes, N. & Kuster, N. A novel medical image data-based multi-physics simulation platform for computational life sciences. Interface Focus 3, 20120058 (2013).

    PubMed  PubMed Central  Article  Google Scholar 

  32. Kinany, N., Pirondini, E., Micera, S. & Ville, D. V. D. Dynamic functional connectivity of resting-state spinal cord fMRI reveals fine-grained intrinsic architecture. Neuron 108, 424–435 (2020).

    CAS  Article  PubMed  Google Scholar 

  33. Landelle, C. et al. Functional brain changes in the elderly for the perception of hand movements: a greater impairment occurs in proprioception than touch. Neuroimage 220, 117056 (2020).

    Article  PubMed  Google Scholar 

  34. Courtine, G., Nunzio, A. M. D., Schmid, M., Beretta, M. V. & Schieppati, M. Stance- and locomotion-dependent processing of vibration-induced proprioceptive inflow from multiple muscles in humans. J. Neurophysiol. 97, 772–779 (2007).

    Article  PubMed  Google Scholar 

  35. Roll, J. P., Vedel, J. P. & Ribot, E. Alteration of proprioceptive messages induced by tendon vibration in man: a microneurographic study. Exp. Brain Res. 76, 213–222 (1989).

    CAS  Article  PubMed  Google Scholar 

  36. Pierrot-Deseilligny, E. & Burke, D. The Circuitry of the Human Spinal Cord (Cambridge University Press, 2005).

  37. Lempka, S. F. et al. Patient‐specific analysis of neural activation during spinal cord stimulation for pain. Neuromodulation 23, 572–581 (2020).

    Article  PubMed  Google Scholar 

  38. Mignardot, J.-B. et al. A multidirectional gravity-assist algorithm that enhances locomotor control in patients with stroke or spinal cord injury. Sci. Transl. Med. 9, eaah3621 (2017).

    Article  PubMed  Google Scholar 

  39. Shokur, S., Mazzoni, A., Schiavone, G., Weber, D. J. & Micera, S. A modular strategy for next-generation upper-limb sensory-motor neuroprostheses. Med 2, 912–937 (2021).

    Article  Google Scholar 

  40. Edgerton, V. R. & Harkema, S. Epidural stimulation of the spinal cord in spinal cord injury: current status and future challenges. Expert Rev. Neurother. 11, 1351–1353 (2011).

    PubMed  PubMed Central  Article  Google Scholar 

  41. Courtine, G. & Sofroniew, M. V. Spinal cord repair: advances in biology and technology. Nat. Med. 25, 898–908 (2019).

    CAS  PubMed  Article  Google Scholar 

  42. Morse, L. R. et al. Meeting proceedings for SCI 2020: launching a decade of disruption in spinal cord injury research. J. Neurotrauma 38, 1251–1266 (2021).

    PubMed  Article  Google Scholar 

  43. Schiavone, G. et al. Soft, implantable bioelectronic interfaces for translational research. Adv. Mater. 32, 1906512 (2020).

    CAS  Article  Google Scholar 

  44. Barra, B. et al. Epidural electrical stimulation of the cervical dorsal roots restores voluntary arm control in paralyzed monkeys. Preprint at bioRxiv https://doi.org/10.1101/2020.11.13.379750 (2021).

  45. Bonizzato, M. et al. Brain-controlled modulation of spinal circuits improves recovery from spinal cord injury. Nat. Commun. 9, 3015 (2018).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  46. Benabid, A. L. et al. An exoskeleton controlled by an epidural wireless brain–machine interface in a tetraplegic patient: a proof-of-concept demonstration. Lancet Neurol. 18, 1112–1122 (2019).

    PubMed  Article  Google Scholar 

  47. Chen, C.-L. et al. The relationship between sitting stability and functional performance in patients with paraplegia. Arch. Phys. Med. Rehab 84, 1276–1281 (2003).

    Article  Google Scholar 

  48. Anderson, K. D. Targeting recovery: priorities of the spinal cord-injured population. J. Neurotrauma 21, 1371–1383 (2004).

    PubMed  Article  Google Scholar 

  49. Herrity, A. N., Williams, C. S., Angeli, C. A., Harkema, S. J. & Hubscher, C. H. Lumbosacral spinal cord epidural stimulation improves voiding function after human spinal cord injury. Sci. Rep. 8, 8688 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  50. Squair, J. W. et al. Neuroprosthetic baroreflex controls haemodynamics after spinal cord injury. Nature 590, 308–314 (2021).

    CAS  PubMed  Article  Google Scholar 

  51. Ditterline, B. E. L. et al. Beneficial cardiac structural and functional adaptations after lumbosacral spinal cord epidural stimulation and task-specific interventions: a pilot study. Front Neurosci. 14, 554018 (2020).

    Article  Google Scholar 

  52. Soloukey, S. et al. The dorsal root ganglion as a novel neuromodulatory target to evoke strong and reproducible motor responses in chronic motor complete spinal cord injury: a case series of five patients. Neuromodulation 24, 779–793 (2021).

    PubMed  Article  Google Scholar 

  53. Kirshblum, S. & Waring, W. Updates of the International Standards for Neurologic Classification of Spinal Cord Injury. Phys. Med. Rehabil. Clin. N. Am. 25, 505–517 (2014).

    PubMed  Article  Google Scholar 

  54. Roll, J. P. & Vedel, J. P. Kinaesthetic role of muscle afferents in man, studied by tendon vibration and microneurography. Exp. Brain Res. 47, 177–190 (1982).

    CAS  PubMed  Article  Google Scholar 

  55. Landelle, C., Ahmadi, A. E. & Kavounoudias, A. Age-related impairment of hand movement perception based on muscle proprioception and touch. Neuroscience 381, 91–104 (2018).

    CAS  PubMed  Article  Google Scholar 

  56. Kavounoudias, A. et al. Proprio-tactile integration for kinesthetic perception: an fMRI study. Neuropsychologia 46, 567–575 (2008).

    CAS  PubMed  Article  Google Scholar 

  57. Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W. & Smith, S. M. FSL. Neuroimage 62, 782–790 (2012).

    PubMed  Article  Google Scholar 

  58. Leener, B. D. et al. SCT: Spinal Cord Toolbox, an open-source software for processing spinal cord MRI data. Neuroimage 145, 24–43 (2017).

    PubMed  Article  Google Scholar 

  59. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for the robust and accurate linear registration and motion correction of brain images. Neuroimage 17, 825–841 (2002).

    PubMed  Google Scholar 

  60. Gros, C. et al. Automatic spinal cord localization, robust to MRI contrasts using global curve optimization. Med. Image Anal. 44, 215–227 (2018).

    PubMed  Article  Google Scholar 

  61. Cohen-Adad, J. et al. Venous effect in spinal cord fMRI: insights from intrinsic optical imaging and laser speckle. Neuroimage 47, S186 (2009).

    Article  Google Scholar 

  62. Power, J. D. et al. Methods to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage 84, 320–341 (2014).

    PubMed  Article  Google Scholar 

  63. Gros, C. et al. Automatic segmentation of the spinal cord and intramedullary multiple sclerosis lesions with convolutional neural networks. Neuroimage 184, 901–915 (2019).

    PubMed  Article  Google Scholar 

  64. Eippert, F., Kong, Y., Jenkinson, M., Tracey, I. & Brooks, J. C. W. Denoising spinal cord fMRI data: approaches to acquisition and analysis. Neuroimage 154, 255–266 (2017).

    PubMed  Article  Google Scholar 

  65. Brooks, J. C. W. et al. Physiological noise modelling for spinal functional magnetic resonance imaging studies. Neuroimage 39, 680–692 (2008).

    PubMed  Article  Google Scholar 

  66. Kong, Y., Jenkinson, M., Andersson, J., Tracey, I. & Brooks, J. C. W. Assessment of physiological noise modelling methods for functional imaging of the spinal cord. Neuroimage 60, 1538–1549 (2012).

    PubMed  Article  Google Scholar 

  67. Behzadi, Y., Restom, K., Liau, J. & Liu, T. T. A component-based noise correction method (CompCor) for BOLD and perfusion-based fMRI. Neuroimage 37, 90–101 (2007).

    Article  PubMed  Google Scholar 

  68. Kasper, L. et al. The PhysIO Toolbox for modeling physiological noise in fMRI data. J. Neurosci. Meth 276, 56–72 (2017).

    Article  Google Scholar 

  69. Woolrich, M. W., Ripley, B. D., Brady, M. & Smith, S. M. Temporal autocorrelation in univariate linear modeling of FMRI data. Neuroimage 14, 1370–1386 (2001).

    CAS  PubMed  Article  Google Scholar 

  70. Woolrich, M. W., Behrens, T. E. J. & Smith, S. M. Constrained linear basis sets for HRF modelling using variational Bayes. Neuroimage 21, 1748–1761 (2004).

    PubMed  Article  Google Scholar 

  71. Gomez‐Perez, S. L. et al. Measuring abdominal circumference and skeletal muscle from a single cross‐sectional computed tomography image. Jpen-parenter. Enter. 40, 308–318 (2016).

    Article  Google Scholar 

  72. Schirmer, C. M. et al. Heuristic map of myotomal innervation in humans using direct intraoperative nerve root stimulation. J. Neurosurg. Spine 15, 64–70 (2011).

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

We thank our study participants for their commitment and trust. All participants gave their informed consent for publication of their images. We thank A. Curt for support; A. van der Kolk and F. Visser for support in imaging data collection and inspection; and many students, interns and former employees for various contributions, including R. Wang, C. Puffay, Y. L. R. Wan, I. Perret, E. Revol, M. Van Campenhoudt, I. Youssef, I. Turcu, F. Sellet, G. Carparelli, C. Moerman, D. Scherrer-Ma, F. Magaud, M. Damiani and N. Regazzi. Investigational implantable stimulators and paddle leads were donated by Medtronic and ONWARD Medical. This work was supported by Wings for Life, the Defitech Foundation, the International Foundation for Research in Paraplegia, Rolex for Enterprise, Carigest Promex, Riders4Riders, ALARME, the Panacée Foundation, the Pictet Group Charitable Foundation, the Firmenich Foundation, ONWARD Medical, European Union’s Horizon 2020 (785907 Human Brain Project SGA2), RESTORE: Eurostars E10889, CONFIRM!: Eurostars E!12743, the Swiss National Science Foundation (NCCR Robotics), the European Research Council (ERC-2015-CoG HOW2WALKAGAIN 682999), the Commission of Technology and Innovation Innosuisse (CTI 41871.1 IP_LS and CTI 25761.1) and the H2020-MSCACOFUND-2015 EPFL fellows program (grant 665667 to F.B.W.).

Author information

Authors and Affiliations

Authors

Contributions

E.B., S.D.H.-C. and E. Paoles contributed equally. A.R., S.K., R.D., E.B., F.B., J.R., M.D., C.V., L. McCracken, N. Hankov, M.V., L.B.-F., H. Lorach, A.G., E. Pralong, M.R., K.M., Q.B., L.A., F.B.W., J. Bloch and G.C. performed experiments and analyzed data. A.R., S.K., R.D., H.M., A.C., B.L., T.N., M.D., N. Hankov, M. Caban, L.B.-F., C.H., S.B., S.C., N.G., B.F., N.B., T.D., D.G., J.B., K.M., E.K., N.K., E.N., M. Capogrosso, F.B.W., J. Bloch and G.C. designed, developed and/or fabricated hardware and/or software. A.R., S.D.H.-C., E. Paoles, H.M., A.C., B.L., T.N., S.B., S.C., N.G., N.K., E.N. and M. Capogrosso performed simulations. A.R., E.B., S.D.H.-C., E. Paoles, F.B., N.K., J.-B.L., E.F., S. Mandija, L. Mattera, R.M., B.N., M.F., A.K., S. Mandija, C.A.T.v.d.B. and D.V. acquired medical imaging datasets. E.B., A.P., M.T., N. Herrmann, M.W., L.G., I.F., V.R., K.K. and G.E. conducted physical therapy. R.D., M.V., A.W., C.J., L.B-F., R.B., V.D., H. Lambert and L.A. managed regulatory affairs. C.W. handled intellectual property. A.R., S.K., R.D., E.B., S.D.H.-C., J.R., L.A. and G.C. prepared illustrations. J. Bloch performed neurosurgical interventions. G.C. and J. Bloch conceived and supervised the study. G.C. wrote the paper, and all the authors contributed to its editing.

Corresponding authors

Correspondence to Jocelyne Bloch or Grégoire Courtine.

Ethics declarations

Competing interests

G.C., J. Bloch, S.M., K.M., F.B.W. and M.Capogrosso hold various patents in relation with the present work. V.D., D.G., J. Bakker, H.L., A.W., C.J., M.D., M. Caban and E. Paoles are ONWARD Medical employees. G.C. is a consultant with ONWARD Medical. G.C., J. Bloch, S.M. and V.D. are shareholders of ONWARD Medical, a company with direct relationships with the presented work. N.K. and E.N. are shareholders of ZMT Zurich MedTech AG, which produces the Sim4Life software. C.W. handles intellectual property for ONWARD Medical. The remaining authors declare no competing interests.

Peer review

Peer review information

Nature Medicine thanks Bruce Dobkin, Philip Star, Blair Calancie and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Jerome Staal was the primary editor on this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Personalizable computational models of the interactions between EES and the spinal cord.

Step 1, High-resolution MRI images enable clear-cut visualization of spinal tissues, including individual dorsal roots. Step 2, CT images enable reconstructing the tridimensional geometry of vertebral bodies. Step 3, Elaboration of highly realistic anatomical models from MRI and CT scans. Step 4, Automatic generation of rootlets based on the identification of the uppermost rootlet (shown in red) in high-resolution MRI acquisitions. Rootlet trajectories are interpolated from this rootlet, using the measured segment length as a reference. Step 5, Splines representing the nerve fibers are automatically generated inside the rootlets. For this purpose, points are initialized in each cross-section based on a weighted superposition of the points constituting the cross-section itself. These points are connected to generate splines. Step 6, Conductivity maps are imposed on each voxel of the discretized finite element models. The white matter and rootlets require an anisotropic conductivity map. Functionality has been implemented in Sim4Life for that purpose that generates anisotropic conductivity maps by solving a diffusion problem with suitable boundary conditions in the tissues of interest and locally aligning conductivity tensors with the gradient of the obtained solution. Step 7, Compartmental cable models are initialized along each spline to integrate the nerve fibers.

Extended Data Fig. 2 Importance of modeling rootlet bundles.

Step 1, Models of the same spinal cord wherein the dorsal roots are modelled as single tubular structures (left) versus multiple tubular structures mimicking the topology of rootlet bundles observed in humans (right), as shown in Step 2, side by side comparison of the rootlet bundles in the model and in a real spinal cord. To create the model of the rootlets, we determined the entry point of the uppermost rootlet for each spinal segment, and then populated the space from the uppermost rootlet of a given dorsal root to the uppermost rootlet of the next dorsal root by distributing rootlets homogeneously across this space. Step 3, A pulse of EES was delivered with increasing intensities through the electrode depicted in step 1, over the L3 dorsal root. The plots show the resulting recruitment curve of each dorsal root. The explicit models of rootlets led to pronounced differences in the recruitment curves of each dorsal root. Step 4, Performance of the new paddle lead evaluated in 15 computational models of the atlas. The top left electrode of the paddle lead was positioned over the dorsal root innervating the L1 spinal segment, as depicted in the model on the left. The plot on the left reports the selectivity of this electrode for each model, organized laterally based on the length of the spinal cord (as reported in Fig. 1). The plot on the right reports the selectively of the bottom left electrode to recruit the dorsal root projecting to the S1 spinal segment. Lower Panel, Horizontal bar plots on the left report the variability of the width of the dorsal root entry zone (n = 15 healthy volunteers). Horizontal bar plots on the right report the variability of length of each spinal segment (n = 27 spinal cords). The bar plot between these two plots reports the variability of the width of the dorsal root entry zones and of the length of spinal segments. p = 0.000035, ***, P < 0.0001, two-tailed t-test.

Extended Data Fig. 3 Identification of the projectome from propriospinal neurons.

Step 1, Acquisition of functional MRI from the spinal cord in response to the recruitment of proprioceptive afferents from specific leg muscles. The muscle spindles are recruited either by stretching the muscles in which they are embedded (the limb is mobilized by a physiotherapist, aided with audio cues), or by applying muscle tendon vibration using MR-compatible pneumatic vibrators (synchronized with MRI triggers). Two runs are acquired for each muscle. Only the right leg muscles are tested. 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 7 minutes. Step 3, A two staged motion correction (3D and then 2D slice-by-slice) is applied for each run. First, the volumes are registered to their respective averaged-in-time image using 3D rigid body realignment. Secondly, taking as reference the averaged-in-time corrected volume, a slice-by-slice 2D realignment is applied thus accounting for the nonrigid property of the spinal cord. Step 4, The motion-corrected series are spatially smoothed, volume by volume with 3D gaussian kernel with full width at half maximum (FWHM) of 2×2×6 mm3. Step 5, The motion-corrected series are again averaged through time. The cerebrospinal fluid and white matter are segmented from this mean functional image. Step 6, Physiological signals (heart rate and respiratory) acquired concomitantly to the fMRI volumes are used to model physiological noise (RETROICOR based procedure). If no signals are available, noise regressors are built with component based noise extraction (aCompCor). Step 7, Acquisition timings corresponding to the task-design, pre-processed (motion corrected, smoothed) fMRI volume series and physiological noise regressors are submitted to a specific first level generalized linear model. A second level fixed effects analysis (subject level, task specific) is performed by combining the two runs. Whenever possible, multiple comparison corrections are performed (Z > 2, pcorr < 0.05). Step 8, Spinal segments are identified from high-definition T2-ZOOMit structural images that allow visualization of the dorsal roots. Spinal segments are then reported in the T2 anatomical image acquired in each run. Step 9, Using non-rigid transformations, the mean functional images are co-registered to the T2 anatomical image. Step 10, Thresholded activity patterns resulting from the generalized linear model are coregistered to the anatomical image. The projectome of proprioceptive neurons innervating the mobilized muscles are extracted and mapped to the anatomical model. Step 11, Projectomes from the three participants, and for comparison, averaged myotome distribution measured electrophysiologically in a large population of patients undergoing surgery. The color dots represent the reconstructed projectome from key leg muscles. Vertical color bars represent mean population distribution of muscular motor hotpots. The projectomes differed across the participants. In particular, the projectome identified in P3 revealed an unexpected inversion of the projectome from ankle antagonists. This rostrocaudal inversion was confirmed electrophysiologically.

Extended Data Fig. 4 Preoperative planning for optimal placement of the new paddle lead.

Step 1, CT, structural MRI and functional MRI acquisitions allow to personalize a computational model of the interactions between EES and the spinal cord for each participant. Step 2, The insertion of the new paddle lead within the spinal canal is visualized in the model to anticipate bony structures or ligaments that could deviate the trajectory. Step 3, The new paddle lead is positioned at 6 locations separated by 2 mm, thus covering the entire region of the spinal cord targeted by the therapy. The same procedure was applied to the Specify 5-6-5 lead, except that 2 additional locations were necessary to cover the entire region since this lead is shorter than the new paddle lead. Step 4, The plot shows the recruitment of each dorsal root when simulating the delivery of EES at increasing intensities through the top left electrode of the paddle lead. The same simulations were performed for the electrodes located at each corner of the paddle lead. Step 5, The recruitment of dorsal roots is translated into the recruitment of motor pools based on a transformation matrix that maps the recruitment of afferents to the recruitment of motor pools. The transformation matrix was either based on the averaged location of motor pools across the human population72, or the projectome of proprioceptive neurons from key leg muscles identified from functional MRI. Step 6, Applying the transformation matrix depicted in Step 5 allows to convert the predicted recruitment of dorsal roots shown in Step 4 into a prediction of motor pool recruitment. Step 7, For each position of the lead, the recruitment of the targeted motor pools compared to the non-targeted motor pools is measured to obtain a selectivity index. For example, the recruitment of the dorsal root projecting to the L1 spinal segments intends to recruit the motor neurons innervating the iliopsoas muscle to elicit hip flexion. The relative recruitment of the iliopsoas muscle versus the rectus femoris or vastus lateralis muscles is transformed into a selectivity index. For each position of the paddle lead, the selectivity index for the tested electrode is color coded, and the selectivity between the tested locations interpolated to obtain a continuum. Step 8, The selectivity indices obtained for the electrodes located at each corner of the paddle lead (from left to right, targeting motor neurons eliciting hip flexion or ankle extension) are aggregated into a combined selectivity index that defines the performance of the paddle lead at the tested position. The optimal position for the paddle lead was defined as the position for which the highest combined selectivity index was obtained (most yellow rectangle). Step 9, Optimal position of the new paddle lead predicted based on a personalized computational model but a generic distribution of motor neuron locations. Step 10, Intraoperative quantification of the combined selectivity index, and thus identification of the optimal position of the new paddle lead. The predicted optimal position of the paddle lead based on a personalized model with the identified projectomes of proprioceptive neurons matched the optimal position validated intraoperatively, whereas simulations based on the averaged location of motor pools across the human population failed to predict the optimal position. Step 11-13, The procedures described in Steps 8-10 were repeated for the Specify 5-6-5 paddle lead. Note that the intraoperative validation of the optimal position of the Specify 5-6-5 was restricted to one position to minimize the duration of the surgical intervention.

Extended Data Fig. 5 Impact of model abstractions to determine the optimal position of the paddle lead.

Step 1, Generalized computational model of the interaction between EES and the spinal cord, including the location of motor neurons from key leg muscles. Step 2, Personalized computational model of the interaction between EES and the spinal cord for the three participants. The models are aligned with the generalized model depicted in Step 1. Step 3, Simulations predict the optimal position of the new paddle lead for each participant, following the procedures explained in Extended Data Fig. 4, but based on various model abstractions, as explained in the boxes above each prediction. Step 4, The optimal position of the new paddle lead was validated intraoperatively, as explained in Extended Data Fig. 4, and is shown on a CT scan reconstruction. The horizontal yellow line passing through the top electrodes of the paddle lead highlights the optimal position, thus allowing a direct comparison between the various predictions and the optimal position. The fully personalized models achieved the best performance.

Extended Data Fig. 6 Configuration of activity-specific stimulation protocols.

Step 1, The participant is lying supine in a relaxed posture. Wireless sensors are positioned over selected leg muscles to monitor electromyographic signals in conjunction with leg kinematics using an optical 3D motion capture system. Step 2, Intraoperative imaging of the final paddle lead position guides the realignment of the paddle lead with respect to the personalized model of the interactions between EES and the spinal cord. The optimal cathode to target specific motor neurons are inferred based on the location of the electrodes with respect to the dorsal roots and location of motor neurons identified from fMRI measurements. Step 3, The performance of the preselected optimal cathode is assessed using trains of pulses delivered with predefined frequency ranges that are optimal for the targeted motor neurons. Step 4, The muscle responses are quantified from 40 to 500 ms after stimulation onset, and then normalized with respect to a baseline window selected 500 ms before stimulation onset. The relative amplitudes of muscle responses are represented in a polar plot that allows to appreciate the relative recruitment of each muscle. Step 5, A physiotherapist grades the precision of the elicited movements and muscle activity based on a simple clinical scale that enables the quick adjustment of anode and cathode configurations to achieve the most optimal selectivity. Step 6, This procedure enables the rapid elaboration of a library of anode and cathodes targeting specific muscles and motor hotspots, which are then implemented in preprogrammed stimulation templates that aim to reproduce the natural activation of muscles during the desired activity.

Extended Data Fig. 7 Configurations of frequency-specific EES trains to elicit functional muscular and kinematic activity.

Step 1, Configuration of electrodes to target the hotspots associated with weight acceptance (top) and whole-leg flexion (bottom). Example from participant P3. EES bursts are delivered at 20 Hz. (weight acceptance, optimal frequency for motor neurons innervating extensor muscles) and 100 Hz (whole-leg flexion, optimal frequency for motor neurons innervating flexor muscles) to elicit muscle responses, recorded from the Iliopsoas (Il), Rectus Femoris (RF), Vastus Lateralis (VLat), Semitendinosus (ST), Tibialis Anterior (TA), Gastrocnemius Medialis (MG), and Soleus (Sol) muscles (mean response, n = 5 repetitions). The muscles associated with the targeted hotspot are color-coded. Polar plots report the normalized muscle responses, using the same convention as in Extended Data Figure 7. Polar plot units are normalized with respect to the baseline (n-fold). Bar plots report the amplitude of associated kinematic responses from each joint, and the selectivity indexes for targeted and non-targeted muscles (n = 5 repetitions for each stimulation configuration). Step 2, Similar representations are shown for participant P1.

Extended Data Fig. 8 Immediate recovery of independent stepping with EES.

Step1, Kinematic and muscle activity underlying stepping on a treadmill without and with EES on the very first day of stimulation for the 3 participants. Bar plots report quantification of the muscle activity, and the range of motion for the hip, knee and ankle in both conditions (n = 10 steps for each condition, two-tailed Mann–Whitney test, *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001). Muscular activities are quantified as Mean Absolute Value over their expected phase of activity. Step 2, Number of independent steps performed during the very first day of stimulation. Step 3, Chronophotographies showing normal and voluntarily exaggerated steps while stimulation parameters remain otherwise unchanged. Bar plots report the mean step length quantified during normal and exaggerated steps (n = 12 normal and 8 exaggerated steps for P1, n = 15 normal and n = 10 exaggerated steps for P2; two-tailed t-test, P1: p = 0.0073, P3: p < 0.0001; **, p < 0.01; ****, p < 0.0001).

Extended Data Fig. 9 Selective recruitment of trunk muscles.

Step 1, Trunk and abdominal muscles are primarily innervated by motor neurons located in the thoracic region of the spinal cord. The new paddle array enables targeting the dorsal roots projecting to the T12 spinal segment, allowing the recruitment of trunk and abdominal muscles. Step 2, Single pulses of EES at increasing amplitude are delivered over electrodes to evaluate their ability to recruit trunk and abdominal muscles. Muscle responses are calculated, normalized, and then represented in a polar plot. The selectivity of trunk/abdominal versus leg muscle activation is calculated with a selectivity index formula. Side-specific recruitment of trunk and abdominal muscles is obtained with the upper electrodes of the new paddle lead. Step 3, Polar plots reporting the activation of trunk/abdominal muscles versus leg muscles when delivering EES through various electrodes of the new paddle lead, as indicated by the number referring to the electrodes depicted in step 2. Step 4, Trains of EES are delivered through the same electrodes as in Step 2 to elicit kinematic responses. For each tested electrode, the panels depict the mean time-dependent trajectory of trunk and knee movements in the plane perpendicular to the direction of gravity, and bar plots reporting the mean amplitude of trunk and knee movement in abduction or adduction. Electrodes 1 and 4, which are located over the top row of the new paddle lead, elicited side-specific trunk movement without disturbing knee movements.

Extended Data Fig. 10 Immediate recovery of trunk control.

Step 1, Participant P2 performing repeated front pull movement on a medicine ball without stimulation (black/EES OFF) and with EES targeting the T12 dorsal root (red/EES ON). Radius of curvature of the lumbar region is measured at position 3, which is the most difficult position for the participants to stabilize. Exercises were repeated 4-5 times in each condition (EES OFF/ON). Step 2, Representation of the trunk muscles engaged in the execution of the task (gray) and EES targeted muscles (red), together with the electrode configuration to target the subset of these muscles affected by the SCI. Step 3, Bar plots reporting the radius of curvature of the lumbar region at position 3 and the execution time of the whole exercise for each participant (n = 5 repetitions per participant, two-tailed Mann–Whitney test, Lumbar curvature (p = 0.0079 for all three participants), Execution time (P1: p = 0.0159, P2: p = 0.0079, P3: p = 0.0079), *, p < 0.05; **, p < 0.01). Step 4, Participant P2 performing repeated lumbar lordosis correction in four-point kneeling position in the absence of stimulation (black/EES OFF) and with a stimulation program that targeted trunk, abdominal and gluteus muscles to stabilize the four-point kneeling position (red/EES ON). Radius of curvature of the lumbar region is measured at the time of maximal contraction and maximal relaxation of the lower back. Exercises were repeated 4-5 times in each condition (EES OFF/ON). Step 5, Same as Step 2. Step 6, Bar plots reporting the lumbar curvature without and with stimulation (n = 6 (P1), n = 4 (P2), n = 6 (P3) repetitions, two-tailed Mann–Whitney test, P1: p = 0.0022, P2: p = 0.0286, P3: p = 0.0022, *, p < 0.05; **, p < 0.01). Step 7, Participant P2 performing repeated front shoulder raise in the absence of stimulation (black/EES OFF) and with EES (red/EES ON). Exercises were repeated 4-5 times in each condition (EES OFF/ON). Step 8, Same as Step 2. Step 9, Changes in position of the wrist in the vertical plane during the front shoulder raise movement, showing improved symmetry and range of motion with EES turned on. The bar plot reports the execution time of this task with (n = 7) and without EES (n = 6), and in 5 healthy individuals for comparison (n = 5 repetitions, two-tailed Mann–Whitney test, p = 0.0082, **, p < 0.01). Step 10, Dips lifting hip. In the absence of stimulation, the participant (P1) is able to lift his own body-weight but is not able to lift his pelvis (black). With EES, he is able to activate his lower abdominal and oblique muscles to lift his pelvis on both sides. Step 11, The participant (P1) is using a torso rotation machine at the gym. In the absence of stimulation, he is able to rotate to both sides lifting 10 kg. EES enables him to perform this exercise with twice this weight as represented on the bar plot.

Supplementary information

Supplementary Information

Supplementary Figures 1 and 2, Table 1 and methods.

Reporting Summary

Supplementary Video 1

Precise preoperative planning and intraoperative validation of model predictions.

Supplementary Video 2

Immediate and long-term recovery of walking after SCI.

Supplementary Video 3

EES during a variety of motor activities.

Supplementary Video 4

Recovery of independence in ecological settings.

Source data

Source Data Fig. 1

Spinal cord segment length and width (b), and raw recruitment curves data (g).

Source Data Fig. 2

Text file with link to repository of computational spinal cord models.

Source Data Fig. 3

Statistical source data.

Source Data Fig. 5b

Motor responses to trunk stimulation (Fig. 5b).

Source Data Fig. 6

Statistical source data.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Rowald, A., Komi, S., Demesmaeker, R. et al. Activity-dependent spinal cord neuromodulation rapidly restores trunk and leg motor functions after complete paralysis. Nat Med 28, 260–271 (2022). https://doi.org/10.1038/s41591-021-01663-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41591-021-01663-5

Further reading

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing