People with late-stage Parkinson’s disease (PD) often suffer from debilitating locomotor deficits that are resistant to currently available therapies. To alleviate these deficits, we developed a neuroprosthesis operating in closed loop that targets the dorsal root entry zones innervating lumbosacral segments to reproduce the natural spatiotemporal activation of the lumbosacral spinal cord during walking. We first developed this neuroprosthesis in a non-human primate model that replicates locomotor deficits due to PD. This neuroprosthesis not only alleviated locomotor deficits but also restored skilled walking in this model. We then implanted the neuroprosthesis in a 62-year-old male with a 30-year history of PD who presented with severe gait impairments and frequent falls that were medically refractory to currently available therapies. We found that the neuroprosthesis interacted synergistically with deep brain stimulation of the subthalamic nucleus and dopaminergic replacement therapies to alleviate asymmetry and promote longer steps, improve balance and reduce freezing of gait. This neuroprosthesis opens new perspectives to reduce the severity of locomotor deficits in people with PD.
This is a preview of subscription content, access via your institution
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$209.00 per year
only $17.42 per issue
Rent or buy this article
Prices vary by article type
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data that support the findings and software routines developed for the data analysis will be made available upon reasonable request to the corresponding authors at: firstname.lastname@example.org or email@example.com. Third party datasets used in our study are available under the Creative Commons Public License: VerSe 2019 (https://osf.io/nqjyw/) and VerSe 2020 (https://osf.io/t98fz/).
Code developed for the data analysis will be made available upon reasonable request to the corresponding authors at: firstname.lastname@example.org or email@example.com.
Ge, H. L. et al. The prevalence of freezing of gait in Parkinson’s disease and in patients with different disease durations and severities. Chin. Neurosurg. J. 6, 17 (2020).
Stolze, H. et al. Prevalence of gait disorders in hospitalized neurological patients. Mov. Disord. 20, 89–94 (2005).
Fasano, A., Canning, C. G., Hausdorff, J. M., Lord, S. & Rochester, L. Falls in Parkinson’s disease: a complex and evolving picture. Mov. Disord. 32, 1524–1536 (2017).
Snijders, A. H. et al. Physiology of freezing of gait. Ann. Neurol. 80, 644–659 (2016).
Marsden, C. D. & Parkes, J. D. Success and problems of long-term levodopa therapy in Parkinson’s disease. Lancet 1, 345–349 (1977).
Funkiewiez, A. et al. Long term effects of bilateral subthalamic nucleus stimulation on cognitive function, mood, and behaviour in Parkinson’s disease. J. Neurol. Neurosurg. Psychiatry 75, 834–839 (2004).
Fasano, A., Aquino, C. C., Krauss, J. K., Honey, C. R. & Bloem, B. R. Axial disability and deep brain stimulation in patients with Parkinson disease. Nat. Rev. Neurol. 11, 98–110 (2015).
Collomb-Clerc, A. & Welter, M. L. Effects of deep brain stimulation on balance and gait in patients with Parkinson’s disease: a systematic neurophysiological review. Neurophysiol. Clin. 45, 371–388 (2015).
Fuentes, R., Petersson, P., Siesser, W. B., Caron, M. G. & Nicolelis, M. A. Spinal cord stimulation restores locomotion in animal models of Parkinson’s disease. Science 323, 1578–1582 (2009).
Santana, M. B. et al. Spinal cord stimulation alleviates motor deficits in a primate model of Parkinson disease. Neuron 84, 716–722 (2014).
Pinto de Souza, C. et al. Spinal cord stimulation improves gait in patients with Parkinson’s disease previously treated with deep brain stimulation. Mov. Disord. 32, 278–282 (2016).
de Lima-Pardini, A. C. et al. Effects of spinal cord stimulation on postural control in Parkinson’s disease patients with freezing of gait. eLife 7, e37727 (2018).
Nishioka, K. & Nakajima, M. Beneficial therapeutic effects of spinal cord stimulation in advanced cases of Parkinson’s disease with intractable chronic pain: a case series. Neuromodulation 18, 751–753 (2015).
Opova, K., Limousin, P. & Akram, H. Spinal cord stimulation for gait disorders in Parkinson’s disease. J. Parkinsons Dis. 13, 57–70 (2023).
de Andrade, E. M. et al. Spinal cord stimulation for Parkinson’s disease: a systematic review. Neurosurg. Rev. 39, 27–35 (2016). discussion 35.
Fuentes, R., Petersson, P. & Nicolelis, M. A. Restoration of locomotive function in Parkinson’s disease by spinal cord stimulation: mechanistic approach. Eur. J. Neurosci. 32, 1100–1108 (2010).
Thevathasan, W. et al. Spinal cord stimulation failed to relieve akinesia or restore locomotion in Parkinson disease. Neurology 74, 1325–1327 (2010).
Hassan, S., Amer, S., Alwaki, A. & Elborno, A. A patient with Parkinson’s disease benefits from spinal cord stimulation. J. Clin. Neurosci. 20, 1155–1156 (2013).
Prasad, S. et al. Spinal cord stimulation for very advanced Parkinson’s disease: a 1-year prospective trial. Mov. Disord. 35, 1082–1083 (2020).
Capogrosso, M. et al. A computational model for epidural electrical stimulation of spinal sensorimotor circuits. J. Neurosci. 33, 19326–19340 (2013).
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).
Wenger, N. et al. Spatiotemporal neuromodulation therapies engaging muscle synergies improve motor control after spinal cord injury. Nat. Med. 22, 138–145 (2016).
Capogrosso, M. et al. A brain–spine interface alleviating gait deficits after spinal cord injury in primates. Nature 539, 284–288 (2016).
Wagner, F. B. et al. Targeted neurotechnology restores walking in humans with spinal cord injury. Nature 563, 65–71 (2018).
Rowald, A. et al. Activity-dependent spinal cord neuromodulation rapidly restores trunk and leg motor functions after complete paralysis. Nat. Med. 28, 260–271 (2022).
Kathe, C. et al. The neurons that restore walking after paralysis. Nature 611, 540–547 (2022).
Bezard, E., Imbert, C., Deloire, X., Bioulac, B. & Gross, C. E. A chronic MPTP model reproducing the slow evolution of Parkinson’s disease: evolution of motor symptoms in the monkey. Brain Res. 766, 107–112 (1997).
Yin, M. et al. Wireless neurosensor for full-spectrum electrophysiology recordings during free behavior. Neuron 84, 1170–1182 (2014).
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).
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).
Guiho, T., Baker, S. N. & Jackson, A. Epidural and transcutaneous spinal cord stimulation facilitates descending inputs to upper-limb motoneurons in monkeys. J. Neural Eng. 18, 046011 (2021).
Minev, I. R. et al. Biomaterials. Electronic dura mater for long-term multimodal neural interfaces. Science 347, 159–163 (2015).
Barra, B. et al. Epidural electrical stimulation of the cervical dorsal roots restores voluntary upper limb control in paralyzed monkeys. Nat. Neurosci. 25, 924–934 (2022).
Lorach, H. et al. Walking naturally after spinal cord injury using a brain–spine interface. Nature 618, 126–133 (2023).
Milekovic, T. et al. Stable long-term BCI-enabled communication in ALS and locked-in syndrome using LFP signals. J. Neurophysiol. 120, 343–360 (2018).
Capogrosso, M. et al. Configuration of electrical spinal cord stimulation through real-time processing of gait kinematics. Nat. Protoc. 13, 2031–2061 (2018).
Milekovic, T., Truccolo, W., Grun, S., Riehle, A. & Brochier, T. Local field potentials in primate motor cortex encode grasp kinetic parameters. Neuroimage 114, 338–355 (2015).
Milekovic, T. et al. An online brain–machine interface using decoding of movement direction from the human electrocorticogram. J. Neural Eng. 9, 046003 (2012).
Burciu, R. G. & Vaillancourt, D. E. Imaging of motor cortex physiology in Parkinson’s disease. Mov. Disord. 33, 1688–1699 (2018).
Yu, H., Sternad, D., Corcos, D. M. & Vaillancourt, D. E. Role of hyperactive cerebellum and motor cortex in Parkinson’s disease. Neuroimage 35, 222–233 (2007).
Sabatini, U. et al. Cortical motor reorganization in akinetic patients with Parkinson’s disease: a functional MRI study. Brain 123, 394–403 (2000).
Eckert, T., Peschel, T., Heinze, H. J. & Rotte, M. Increased pre-SMA activation in early PD patients during simple self-initiated hand movements. J. Neurol. 253, 199–207 (2006).
Catalan, M. J., Ishii, K., Honda, M., Samii, A. & Hallett, M. A PET study of sequential finger movements of varying length in patients with Parkinson’s disease. Brain 122, 483–495 (1999).
Turner, R. S., Grafton, S. T., McIntosh, A. R., DeLong, M. R. & Hoffman, J. M. The functional anatomy of parkinsonian bradykinesia. Neuroimage 19, 163–179 (2003).
Milekovic, T., Ball, T., Schulze-Bonhage, A., Aertsen, A. & Mehring, C. Detection of error related neuronal responses recorded by electrocorticography in humans during continuous movements. PLoS ONE 8, e55235 (2013).
Gilron, R. et al. Long-term wireless streaming of neural recordings for circuit discovery and adaptive stimulation in individuals with Parkinson’s disease. Nat. Biotechnol. 39, 1078–1085 (2021).
Bruel, A. et al. Investigation of neural and biomechanical impairments leading to pathological toe and heel gaits using neuromusculoskeletal modelling. J. Physiol. 600, 2691–2712 (2022).
Prasanth, H. et al. Wearable sensor-based real-time gait detection: a systematic review. Sensors 21, 2727 (2021).
Movement Disorder Society Task Force on Rating Scales for Parkinson’s Disease. The Unified Parkinson’s Disease Rating Scale (UPDRS): status and recommendations. Mov. Disord. 18, 738–750 (2003).
Nutt, J. G. et al. Freezing of gait: moving forward on a mysterious clinical phenomenon. Lancet Neurol. 10, 734–744 (2011).
Grabli, D. et al. Gait disorders in parkinsonian monkeys with pedunculopontine nucleus lesions: a tale of two systems. J. Neurosci. 33, 11986–11993 (2013).
Thenaisie, Y. et al. Principles of gait encoding in the subthalamic nucleus of people with Parkinson’s disease. Sci. Transl. Med. 14, eabo1800 (2022).
Ahmed, M. R. et al. Lentiviral overexpression of GRK6 alleviates l-dopa-induced dyskinesia in experimental Parkinson’s disease. Sci. Transl. Med. 2, 28ra28 (2010).
Bezard, E. et al. The mGluR5 negative allosteric modulator dipraglurant reduces dyskinesia in the MPTP macaque model. Mov. Disord. 29, 1074–1079 (2014).
Bezard, E. et al. Attenuation of levodopa-induced dyskinesia by normalizing dopamine D3 receptor function. Nat. Med. 9, 762–767 (2003).
Fasano, S. et al. Inhibition of Ras-guanine nucleotide-releasing factor 1 (Ras-GRF1) signaling in the striatum reverts motor symptoms associated with l-dopa-induced dyskinesia. Proc. Natl Acad. Sci. USA 107, 21824–21829 (2010).
Porras, G. et al. PSD-95 expression controls l-DOPA dyskinesia through dopamine D1 receptor trafficking. J. Clin. Invest. 122, 3977–3989 (2012).
Shen, W. et al. M4 muscarinic receptor signaling ameliorates striatal plasticity deficits in models of l-DOPA-induced dyskinesia. Neuron 88, 762–773 (2015).
Urs, N. M. et al. Targeting β-arrestin2 in the treatment of l-DOPA-induced dyskinesia in Parkinson’s disease. Proc. Natl Acad. Sci. USA 112, E2517–E2526 (2015).
Bezard, E. et al. Relationship between the appearance of symptoms and the level of nigrostriatal degeneration in a progressive 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine-lesioned macaque model of Parkinson’s disease. J. Neurosci. 21, 6853–6861 (2001).
Meissner, W. et al. Time-course of nigrostriatal degeneration in a progressive MPTP-lesioned macaque model of Parkinson’s disease. Mol. Neurobiol. 28, 209–218 (2003).
Rosenblad, C. et al. Vector-mediated l-3,4-dihydroxyphenylalanine delivery reverses motor impairments in a primate model of Parkinson’s disease. Brain 142, 2402–2416 (2019).
Courtine, G. et al. Kinematic and EMG determinants in quadrupedal locomotion of a non-human primate (Rhesus). J. Neurophysiol. 93, 3127–3145 (2005).
Tass, P. A. et al. Coordinated reset has sustained aftereffects in Parkinsonian monkeys. Ann. Neurol. 72, 816–820 (2012).
Schiavone, G. et al. Soft, implantable bioelectronic interfaces for translational research. Adv. Mater. 32, e1906512 (2020).
Baufreton, J. et al. Inhaling xenon ameliorates l-dopa-induced dyskinesia in experimental parkinsonism. Mov. Disord. 33, 1632–1642 (2018).
Zampieri, C. et al. The instrumented timed up and go test: potential outcome measure for disease modifying therapies in Parkinson’s disease. J. Neurol. Neurosurg. Psychiatry 81, 171–176 (2010).
Palmerini, L., Mellone, S., Avanzolini, G., Valzania, F. & Chiari, L. Quantification of motor impairment in Parkinson’s disease using an instrumented timed up and go test. IEEE Trans. Neural Syst. Rehabil. Eng. 21, 664–673 (2013).
McLachlan, G. J. & Peel, D. Finite Mixture Models (Wiley, 2000).
Postuma, R. B. et al. MDS clinical diagnostic criteria for Parkinson’s disease. Mov. Disord. 30, 1591–1601 (2015).
Stanslaski, S. et al. A chronically implantable neural coprocessor for investigating the treatment of neurological disorders. IEEE Trans. Biomed. Circuits Syst. 12, 1230–1245 (2018).
Kremen, V. et al. Integrating brain implants with local and distributed computing devices: a next generation epilepsy management system. IEEE J. Transl. Eng. Health Med. 6, 2500112 (2018).
Pistohl, T., Ball, T., Schulze-Bonhage, A., Aertsen, A. & Mehring, C. Prediction of arm movement trajectories from ECoG-recordings in humans. J. Neurosci. Methods 167, 105–114 (2008).
Moreau, C. et al. STN-DBS frequency effects on freezing of gait in advanced Parkinson disease. Neurology 71, 80–84 (2008).
Delp, S. L. et al. An interactive graphics-based model of the lower extremity to study orthopaedic surgical procedures. IEEE Trans. Biomed. Eng. 37, 757–767 (1990).
Ong, C. F., Geijtenbeek, T., Hicks, J. L. & Delp, S. L. Predicting gait adaptations due to ankle plantarflexor muscle weakness and contracture using physics-based musculoskeletal simulations. PLoS Comput. Biol. 15, e1006993 (2019).
Geyer, H. & Herr, H. A muscle-reflex model that encodes principles of legged mechanics produces human walking dynamics and muscle activities. IEEE Trans. Neural Syst. Rehabil. Eng. 18, 263–273 (2010).
Song, S. & Geyer, H. Predictive neuromechanical simulations indicate why walking performance declines with ageing. J. Physiol. 596, 1199–1210 (2018).
Thelen, D. G. Adjustment of muscle mechanics model parameters to simulate dynamic contractions in older adults. J. Biomech. Eng. 125, 70–77 (2003).
Seth, A. et al. OpenSim: simulating musculoskeletal dynamics and neuromuscular control to study human and animal movement. PLoS Comput. Biol. 14, e1006223 (2018).
Häkkinen, K. & Häkkinen, A. Muscle cross-sectional area, force production and relaxation characteristics in women at different ages. Eur. J. Appl. Physiol. Occup. Physiol. 62, 410–414 (1991).
Goodpaster, B. H. et al. Attenuation of skeletal muscle and strength in the elderly: the Health ABC Study. J. Appl Physiol. 90, 2157–2165 (2001).
Cotofana, S. et al. Correlation between single-slice muscle anatomical cross-sectional area and muscle volume in thigh extensors, flexors and adductors of perimenopausal women. Eur. J. Appl. Physiol. 110, 91–97 (2010).
Geijtenbeek, T. SCONE: open source software for predictive simulation of biological motion. J. Open Source Softw. 4, 1421 (2019).
Hansen, N. & Kern, S. Evaluating the CMA evolution strategy on multimodal test functions. In Parallel Problem Solving from Nature - PPSN VIII (eds Yao, X. et al.) 282–291 (Springer, 2004).
Wang, J. M., Hamner, S. R., Delp, S. L. & Koltun, V. Optimizing locomotion controllers using biologically-based actuators and objectives. ACM Trans. Graph. 31, 25 (2012).
Altini, N. et al. Segmentation and identification of vertebrae in CT scans using CNN, k-means clustering and k-NN. Informatics 8, 40 (2021).
Sekuboyina, A. et al. VerSe: a vertebrae labelling and segmentation benchmark for multi-detector CT images. Med. Image Anal. 73, 102166 (2021).
Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J. & Maier-Hein, K. H. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat. Methods 18, 203–211 (2021).
De Leener, B. et al. SCT: Spinal Cord Toolbox, an open-source software for processing spinal cord MRI data. Neuroimage 145, 24–43 (2017).
Patterson, K. K., Gage, W. H., Brooks, D., Black, S. E. & McIlroy, W. E. Evaluation of gait symmetry after stroke: a comparison of current methods and recommendations for standardization. Gait Posture 31, 241–246 (2010).
Horak, F. B. & Mancini, M. Objective biomarkers of balance and gait for Parkinson’s disease using body-worn sensors. Mov. Disord. 28, 1544–1551 (2013).
Lord, S., Galna, B. & Rochester, L. Moving forward on gait measurement: toward a more refined approach. Mov. Disord. 28, 1534–1543 (2013).
Franchignoni, F., Horak, F., Godi, M., Nardone, A. & Giordano, A. Using psychometric techniques to improve the Balance Evaluation Systems Test: the mini-BESTest. J. Rehabil. Med. 42, 323–331 (2010).
Dal Bello-Haas, V., Klassen, L., Sheppard, M. S. & Metcalfe, A. Psychometric properties of activity, self-efficacy, and quality-of-life measures in individuals with Parkinson disease. Physiother. Can. 63, 47–57 (2011).
Powell, L. E. & Myers, A. M. The Activities-specific Balance Confidence (ABC) Scale. J. Gerontol. A Biol. Sci. Med. Sci. 50A, M28–M34 (1995).
Franchignoni, F., Giordano, A., Ronconi, G., Rabini, A. & Ferriero, G. Rasch validation of the Activities-specific Balance Confidence Scale and its short versions in patients with Parkinson’s disease. J. Rehabil. Med. 46, 532–539 (2014).
Funding was obtained from the Defitech Foundation, the Roger de Spoelberch Prize, ONWARD Medical, CAMS Innovation Fund for Medical Sciences grant 2021-1-I2M-034, National Natural Science Foundation of China grants 81941012 and 82161138027, PDWALK ERANET JP cofunND 2-NT (ANR, FNS, ZonMw), the Parkinson Schweiz Foundation, the European Community’s Seventh Framework Program (NeuWalk), a Consolidator Grant from the European Research Council, the Wyss Center for Bio- and Neuroengineering, the Bertarelli Foundation, a Marie Curie fellowship to D.A.B., Marie Curie COFUND EPFL fellowships to T.M. and G.S., a Morton Cure Paralysis Fund fellowship to T.M., a Whitaker Foundation fellowship to M.G.P. and the Swiss National Science Foundation, including the National Center of Competence in Research in Robotics, the Sino-Swiss Science and Technology Cooperation (IZLCZ3_156331), the NanoTera.ch program (SpineRepair) and the Sinergia program (CRSII3_160696).
The authors declare the following competing financial interests: G.C., J.B., R.D., S.M., S.L., T.M., E.M.M. and M.C. hold various patents or applications in relation to the present work. G.C. and J.B. are consultants for ONWARD Medical. G.C., J.B., S.M., S.L., H.L. are founders and minority shareholders of ONWARD Medical, a company with potential commercial interest in the presented work. E.B. reports personal fees from Motac Neuroscience Ltd UK and is a shareholder of Motac Holding UK and Plenitudes SARL France. The remaining authors declare no competing interests.
Peer review information
Nature Medicine thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editor: Jerome Staal, in collaboration with the Nature Medicine team.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended Data Fig. 1 Objective quantification of gait impairments and balance problems in the NHP MPTP model of Parkinson’s disease and in people with PD.
a, Scatter plot shows the rounded mean of PD scores for each monkey across sessions recorded after MPTP administration. Photographs show the dopaminergic projections labelled with tyrosine hydroxylase (TH) in the putamen and caudate of a healthy monkey and in M8. The bar plots show the density of TH-labelled projections in the putamen and caudate (n = Healthy: M6: 6, M7: 6, M8: 4, M9: 6) and count of dopaminergic cells in substantia nigra (n = Healthy: 3, M5: 1, M6: 2, M7:2, M8: 2, M9: 2) in healthy monkeys and in monkeys after the MPTP administration. b-c, The NHPs were trained to walk across a 3m-long corridor (M1-9, n = 9) and along the rungs of a 3m-long horizontal ladder (M3-8, n = 6). Both runways were embedded within Plexiglass enclosures that allowed the NHPs to behave freely and untethered while anatomical landmarks painted on the joints were filmed using 4 or 6 cameras in order to reconstruct whole-body kinematics in 3D. We used these kinematic recordings to compute 83 variables from each gait cycle that quantified kinematic features of monkeys’ locomotor patterns (Supplementary Table 3). This dataset was arranged in a matrix with variables as the matrix columns and each row representing one gait cycle. Data collected from different conditions (before and after MPTP administration) and different monkeys were pooled together in a single matrix and z-scored across columns. Two different tasks, corridor (a) and ladder (b), were analysed separately. We then applied PCA on this dataset and visualized the outcome by plotting the dataset in a new space spanned by the three leading PCs. The data for each monkey and condition is represented by a balloons – ellipsoids with the centre and principal semi-axis as the mean and standard deviation calculated across all the gait cycles for that condition and monkey (number of gait cycles in corridor: before MPTP: M1: 4, M2: 6, M3: 18, M4: 13, M5: 14, M6: 22, M7: 14, M8: 95, M9: 49; after MPTP: M1: 28, M2: 8, M3: 16, M4: 22, M5: 11, M6: 6, M7:9, M8: 48, M9: 59; ladder: before MPTP: M3: 12, M4: 25, M5: 23, M6: 19, M7: 9, M8: 40; after MPTP: M3: 18, M4: 7, M5: 24, M6: 31, M7: 24, M8: 14). Since the variance in the dataset is driven by the changes in gait parameters between the healthy and MPTP conditions consistent across monkeys, the parameters that best capture gait and balance deficits after MPTP administration have the highest loading factors in leading principal components (PCs). The colorplot shows the loading factors for the three leading PCs. The bar plots report the mean values of the parameters with the highest factor loadings. These parameters reflect gait and balance deficits commonly observed in people with PD. d, Healthy subjects (H; n = 9) and subjects with PD (PD; n = 25) walked straight overground as we recorded their full-body kinematics in 3D using the Vicon multi-camera system. We used these kinematic recordings to compute 35 variables from each gait cycle that quantified kinematic features of human locomotor patterns (Supplementary Table 5). As for the monkeys, we arranged this dataset in a gait parameters x gait cycles matrix, applied PCA on this dataset and visualized the outcome by plotting the distribution balloons for each subject in a space spanned by the three leading PCs (number of gait cycles: H1: 37, H2: 36, H3: 33, H4: 44, H5:42, H6: 38, H7: 45, H8: 33, H9: 39; PD1: 28, PD2: 30, PD3: 54, PD4: 53, PD5: 81, PD6: 47, PD7: 69, PD8: 37, PD9: 8, PD10: 100, PD11: 32, PD12: 22, PD13: 48, PD14: 25, PD15: 69, PD16: 70, PD17: 33, PD18: 61, PD19: 40, PD20: 29, PD21: 82, PD22: 8, PD23: 66, PD24: 33, PD25: 29). The colorplot shows the loading factors for the three leading PCs. *, ** significant difference at p < 0.05 and p < 0.01, respectively, using two-sided Wilcoxon signed rank test. Error bars, sem.
a, Colorplots showing the average spatiotemporal map of motor neuron activity underlying locomotion in M6 (n = 12 gait cycles), M7 (n = 10), M8 (n = 20), M9 (n = 32) and M10 (n = 13) before MPTP administration (Healthy). We identified the hotspots of motor neuron activity using Gaussian Mixture Modelling. The spatial maps of motor neuron activity corresponding to the time at which each hotspot reached a maximum (centre) are laid over the schematics of the spinal cord. b, Colorplots show the spatiotemporal maps of motor neuron activity underlying locomotion in M6 (n = 55 gait cycles), M7 (n = 44), M8 (n = 17) and M9 (n = 11) after MPTP administration (MPTP). Bar plots compare the surface correlation between two maps calculated before MPTP administration and between a map calculated before MPTP administration and a map calculated after the MPTP administration. **, *** significant difference at p < 0.01, p < 0.001, respectively, using non-parametric one-sided Monte Carlo permutation test. Error bars show sem.
a, We developed a fabrication process to manufacture the non-human primate epidural spinal arrays used in M11 using e-dura technology (see Methods). Step 1: Substrate creation. We used 4’ silicon wafers as substrate to prepare the arrays. (top) A polystyrene sulfonic acid layer is spin coated on the carrier to provide a water-release layer for the substrate stack. A PDMS layer is subsequently cast on the substrate until reaching a thickness of approximately 500 µm. (middle) A laser-machined, 50 µm thick polyimide mask is then manually laminated on the PDMS surface, and the carriers are then mounted in a thermal evaporation chamber. A stack of chromium and gold is thermally evaporated on the carriers through the polyimide masks, at a thickness of 5 nm (Cr) and 55 nm (Au). The chromium acts as adhesion promoter for the gold interconnect on PDMS. (bottom) The polyimide shadow mask is then peeled off the surface, revealing the interconnect design patterned in the metal stack. Step 2: Passivation. (top) A 3 mm thick PDMS handling layer is cast in a Petri dish. Once cured, the top surface is exposed to an oxygen plasma and a vapour-phase 1H,1H,2H,2H-perfluorooctyltriethoxysilane layer is applied in a vacuum chamber. This process inhibits the adhesion of the silicone to subsequent PDMS layers deposited on the surface. Two subsequent 20 µm thick PDMS layers are spin-coated and cured on the thick PDMS, separated by the same adhesion inhibiting layer. A slab of this triple PDMS stack (3 mm, 20 µm, 20 µm in cross section) is then cut with a blade and mounted on a glass slide. (bottom) A mechanical catheter puncher is used to make holes through the two thin PDMS layers and into the thick handling layer, in order to machine the passivation stack with through-vias. Each via is created by punching a series of 4 round holes of 690 µm diameter with 400 µm centre-to-centre spacing. Step 3: Assembly. (top) The top surfaces of the substrate and triple stack encapsulation are exposed to oxygen plasma, then mounted on an alignment rack, with the two treated surfaces facing one another. The vias machined in the encapsulation are aligned with the interconnect patterned on the substrate, and the two parts are then put into contact in order to form a covalent bond between the silicone layers. (bottom) Once bonded, the thick PDMS handling layer is peeled off the substrate, leaving the interconnect encapsulated by two 20 µm thick PDMS layers with openings corresponding to the position of the electrodes. Step 4: Electroactive coating. (top) The electroactive coating is prepared as a composite material obtained by dispersing microscale platinum particles (3.5 µm maximum particle size) within a polydimethylsiloxane (PDMS) matrix. This creates a conductive paste that offers a balance between the charge injection properties of platinum and the mechanical properties of PDMS. The composite paste is applied on the encapsulation through the screen print PDMS layer, filling the openings to make an electrical contact with the interconnect. (bottom) The screen print layer is then peeled off to remove the excess coating and define the active stimulation sites. Step 5: Packaging. (top) The assembled implant is manually cut while still on wafer to the desired shape using a blade. Electrical wires are threaded in a PDMS guiding piece through holes that are machined at the same pitch as the gold tracks on the substrate. This soft connector is aligned and placed onto the interconnect with wires close to the ends of the gold tracks. Conductive Silver paste is then pressure dispensed to form individual electrical connections between the wires and the gold tracks. Once all electrical connections are made, a bolus of room temperature vulcanisation sealant (one component silicone sealant 734, Dow Corning) is applied over the connector to mechanically secure the assembly. (bottom) After the sealant is cured, the implant is released from the silicon carrier by dissolving the PSS layer under the PDMS substrate with DI water. All silicone layers are prepared by mixing polydimethylsiloxane using a weight ratio of 10:1 between pre-polymer and cross-linker. The deposited layers are cured for a minimum of 3 hours in a temperature-controlled oven set to 80 °C. Photographs show a fabricated e-dura spinal implant, including a zoom on the electrode contacts. b, We exploited our fabrication process to produce epidural spinal arrays that embedded laterally-located electrodes targeting the left and right posterior roots of the lumbar spinal cord, as well as midline-located electrodes targeting the ascending fibres within the dorsal column. The shown spinal cord was reconstructed from a magnetic resonance imaging scan of a rhesus macaque onto which we represented the planed locations of the rostral and caudal spinal arrays. c, Circular plots reporting the amplitude (grey scale) of muscle responses recorded from leg muscles when delivering single-pulse EES at increasing amplitudes (radial axis). Red circles highlight the optimal amplitude while the polygon quantifies the muscular selectivity at this amplitude. Spatial map of motor neuron activity corresponding to optimal EES amplitudes are laid over the schematics of the spinal cord for each hotspot. Bar plots report the correlation between each maximal-selectivity spatial map of motor neuron activity and the spatial map corresponding to the targeted hotspot. Muscle responses were normalized to the maximum amplitudes observed across all the recording sessions. d, The scheme illustrates the similarity between non-human primate and human implementation of the spinal cord neuroprosthesis. In both implementations, sensors collected physiological signals that are wirelessly acquired by the control computer running a stimulation control software. This software processed the sensor signals and used a rLDA algorithm to detect hotspot initiation events. On detection of an event, the stimulation control software sent a command to the implanted pulse generator via a wireless communication pipeline that featured electromagnetic induction though the users’ skin. On reception of the command, the implanted pulse generator modified the EES sequence to promote the activity of the detected hotspot and, therefore, reinforce the intended movements. Modified EES was delivered over the posterior spinal cord by the epidural spinal arrays. Between non-human primate and human implementations, only the sensors and the spinal arrays differed. Non-human primate implementation relied on recordings from neurosensors featuring microelectrode arrays implanted into the leg area of the motor cortex; and on custom spinal arrays designed for Rhesus macaque spinal anatomy. Human implementation relied on recordings from wearable non-invasive IMU sensors distributed across major anatomical landmarks; and on clinically-approved epidural spinal arrays. The remainder of the spinal cord neuroprosthesis implementation was identical.
Extended Data Fig. 4 Calibration of neural decoders for real-time detection of hotspot initiation events.
a. Step 1: Hindlimb kinematics and MI activity were recorded during locomotion. The neural signals were band-pass filtered (0.5-7.5 kHz), and multiunit spike events were collected based on a threshold set at 3.5 times the standard deviation. Step 2: We marked video frames containing left and right foot off and foot strike events. We estimated multiunit spike rates from overlapping 150 ms bins that were updated every 0.5 ms. Step 3: We identified the right weight acceptance and right leg lift hotspots initiation events from the spatiotemporal map of motor neuron activity by aligning the gait events to the derived map of spatiotemporal motor neuron activity. The left hotspot initiation events were derived using the same process, assuming symmetry between both legs. The hotspot events were adjusted to account for the stimulation latency of 105 ms. Step 4: We extracted feature vectors that originated at hotspot events and assigned them to respective hotspot classes. We assigned all other feature vectors to the ‘neither’ class. Step 5: We used these feature vector classes to calibrate a regularized linear discriminant analysis decoder. Step 6: The decoder was uploaded into our real-time analysis software application running on the control computer. Neural data was collected in real-time, processed into multiunit spike rates, and passed through the decoder that calculated the probabilities of hotspot events. When one of the hotspot event probabilities crossed a threshold of 0.8, a wireless command was sent to the implanted pulse generator to trigger the respective stimulation sequences. These sequences were composed of one or more stimulation protocols, each designed to reinforce one of six hotspots: left and right weight acceptance, propulsion and leg lift hotspots. b. Two-step decoder calibration. Step 1: Data are acquired without EES to calibrate the first-step decoder as shown in a. Step 2: An additional set of data is acquired during which the first step decoder trigger composite stimulation sequences once per gait cycle. This sparse triggering mitigates the ability of EES to influence the neural activity used to detect the hotspot events that trigger EES. The composite EES sequences contain either left weight acceptance, left propulsion and right leg lift; or right weight acceptance, right propulsion and left leg lift EES protocols. Step 3: A second step decoder is then calibrated using all the acquired datasets. Since the decoder is calibrated on neural activity that is non-affected and affected by EES, the decoder maintains high accuracy regardless of the presence or absence of EES.
Extended Data Fig. 5 The brain-controlled spinal cord neuroprosthesis improves basic and skilled locomotion after MPTP administration.
a, Examples of locomotor execution along the corridor (3.3 s) and ladder (2.4 s) without stimulation (left columns) and when using the brain-controlled spinal cord neuroprosthesis in M8 after MPTP administration. From top to bottom: stick diagram decompositions of left and right leg movements; neural recording from a single channel; probability of left and right weight acceptance events; detected hotspot events (broken vertical lines), periods of stimulation through the electrodes targeting the left and right weight acceptance, propulsion and leg lift hotspots; electromyographic signals; whole-limb extension calculated as distance from the hip to the ankle joint. The white, light grey and dark grey backgrounds correspond to double stance, left and right swing gait phases, respectively. b, The histogram plots show the distributions between hotspot initiation events measured from kinematic recordings (ground truth) and hotspot initiation events decoded during locomotion with the brain-controlled spinal cord neuroprosthesis (n = 516, 618 and 612 events for M8, M9 and M11, respectively). Bar plots report key parameters associated with gait and balance deficits commonly observed in people with PD (n = 26, 51, 27, 81, 50 and 45 steps for M8, 63, 62, 45, 140 and 55 steps for M9, and 25, 87, 33 and 19 steps for M11 across conditions from left to right). M8’s gait was recorded in two days before the MPTP treatment (days 1 and 2) and two days after the treatment (days 3 and 4). M9’s gait was recorded in one day before the MPTP treatment (day 1) and two days after the treatment (days 2 and 3). M11’s gait was recorded in two days after the treatment (days 1 and 2). The statistical significance is shown only for comparison of between the MPTP EESOFF dataset and MPTP EESON datasets recorded on the same day. c, Changes in EES frequency between 30 and 80 Hz modulate gait parameters but has minimal impact on the efficacy of the therapy. The plots report the mean stride length, endpoint velocity and lateral hip displacement during locomotion along the corridor (n = 50, 45, 26, 16, 17, 18, 27, 26, 22 and 33 steps for M8, 55, 63, 62, 45, 58, 43 and 39 steps for M9, and 25, 11, 29, 10, 19, 26, 33 and 19 steps for M11 across conditions from left to right) under different EES frequencies with the brain-controlled spinal cord neuroprosthesis. Small circles, individual gait cycles; lines, mean across all gait cycle for each condition. Recording days and presentation of statistical significance same as in (c). d, Body posture reconstructed from body kinematics using a whole-body skeletal model. Bar plots show the spine curvature measured from these reconstructions (n = 12, 12 and 12 samples for M8, 12, 12 and 12 samples for M9, and 10 and 10 samples for M11 across conditions from left to right). e, Brain-controlled spinal cord neuroprosthesis immediately improves locomotor performance when traversing a horizontal ladder. Stick diagrams show right leg kinematics during walking along the horizontal ladder of M11 after MPTP administration without and with brain-controlled spinal cord neuroprosthesis. Pie charts report the temporal accuracy of the decoder (n = 135 and 103 events for M8 and M11 respectively) measured during the online use of the neuroprosthesis. The histogram plots show the distributions between hotspot initiation events measured from kinematic recordings (ground truth) and hotspot initiation events decoded during locomotion with the brain-controlled spinal cord neuroprosthesis (n = 135 and 103 events for ladder for M8 and M11, respectively). Bar plots repot ‘task’ time needed to complete the task (n = 11, 9, 30 trials for M8 and 6, 14 trials for M11 across conditions from left to right) and the occurrence of falls (n = 30 trials). Balloons show mean ± standard deviation of all gait cycles for a given condition in space defined by PC1 and PC2, which explained 41% of all the variance. Bar plots report the Euclidean distance in the full 83-dimensional space of gait parameters between each gait cycle and the mean values across all the gait cycles recorded during two independent sessions before MPTP administration; as well as key parameters associated with gait and balance deficits commonly observed in people with PD (Ladder: n = 10, 13, 14, 19 and 40 steps for M8, and 20 and 12 steps for M11 across conditions from left to right). Changes in EES frequency between 30 and 80 Hz modulates gait parameters but has minimal impact on the efficacy of the therapy in the ladder. The plots report the stance duration and velocity during locomotion along the ladder (n = 40, 10, 8, 5, 14, 14 and 5 steps for M8, and 20 and 12 steps for M11 across conditions from left to right) under different EES frequencies with the brain-controlled spinal cord neuroprosthesis. Small circles, individual gait cycles; thick lines, mean across all gait cycle for each condition. Recording days and presentation of statistical significance same as in (c). *, **, *** significant difference at p < 0.05, p < 0.01 and p < 0.001, respectively using two-sided Wilcoxon rank sum test or the one-sided Monte Carlo permutation test. n.s., not significant (p ≥ 0.05) according to the same tests. Error bars show sem.
Extended Data Fig. 6 EES protocols must be synchronized precisely with hotspot initiation events for maximum efficacy of the brain-controlled spinal cord neuroprosthesis.
a, Examples showing 3.3 s of locomotion across the corridor using the brain-controlled spinal cord neuroprosthesis with three different decoding models (M8). From left to right: hotspot stimulation protocols initiated 200 ms before their initiation, hotspot stimulation protocols synchronized with hotspot initiation, hotspot stimulation protocols initiated 200 ms after their initiation. Conventions are the same as in Extended Data Fig. 5a. b, Dot plots showing the task time and crossing time without stimulation and when using the brain-controlled spinal cord neuroprosthesis delivering synchronized, advanced or delayed EES (task time: n = 7, 4 and 4 trials for M8 and 5, 5 and 11 trials for M9; crossing time: n = 7, 4 and 5 trials for M8 and 5, 5 and 11 trials for M9 for conditions from left to right, respectively). c, Example showing 7.16 s of locomotion across the corridor during the randomly triggered stimulation protocols. Conventions are the same as in Extended Data Fig. 5a. d, Dot plots showing the task time and crossing time without stimulation, when using the brain-controlled spinal cord neuroprosthesis to deliver synchronized EES, and random delivery of EES (task time: n = 7, 4 and 5 trials for M8 and 5, 5 and 5 trials for M9; crossing time: n = 7, 4 and 5 trials for M8 and 5, 5 and 5 trials for M9 for conditions from left to right, respectively). *, ** significant difference at p < 0.05 and p < 0.01, respectively, using two-sided Wilcoxon rank sum test.
a, We implanted monkey M9 with mini-DBS electrodes in the left and right subthalamic nucleus after the MPTP treatment to test the effect of low frequency (20 Hz) and high frequency (125 Hz) DBS during corridor walking. We evaluated the locomotor performance for the following conditions: before MPTP administration (Healthy) and after MPTP administration with stimulation off (MPTP), using 20 Hz DBS (DBSON 20 Hz), and using 125 Hz DBS (DBSON 125 Hz). Balloons show mean ± SD of all gait cycles for each condition in the space spanned by two leading PCs (number of gait cycles: Healthy: 39; MPTP: 47; DBSON 20 Hz: 27; DBSON 125 Hz: 37). The bar plot inset reports the Euclidean distance in the full 83-dimensional gait space between each gait cycle and the mean values across all the gait cycles recorded before MPTP administration. To identify the MPTP-induced locomotor deficits affected the most by the DBS, we identified the parameters with the highest loading factors on PC1. This analysis revealed a strong influence of DBS on parameters related to gait velocity and size, but reduced impact of limb configuration values (leg lift, propulsion) during gait. B, As reported in Parkinson’s disease patients, high frequency DBS increases the overall mobility and mediates a moderate increase on gait speed, while the low frequency DBS fails to improve locomotion and impairs awareness. The bar plots show the number of corridor crossings within 10 minutes (number of trials: Healthy: 24; MPTP: 19; DBSON 20 Hz: 12; DBSON 125 Hz: 17), percentage of uncompleted trials, and gait cycle duration (number of gait cycles: Healthy: 50; MPTP: 160; DBSON 20 Hz: 49; DBSON 125 Hz: 137). C, High frequency DBS moderately improves the balance locomotor deficits. The bar plot shows the mean lateral displacement of the hip during gait (number of gait cycles same as in a). d, DBS failed to correct for the lack of propulsion and leg lift induced by MPTP. The plots show examples of three successive gait cycles recorded before MPTP (left column), and after MPTP without using stimulation (middle column) or using 125 Hz DBS (right column). The plots show mean ± SD of left leg step height, limb length and limb angle across the gait cycle (number of gait cycles same as in a). *, **, *** reflect a significant difference at p < 0.05, p < 0.01, p < 0.001 respectively, using two-sided Wilcoxon ranksum test or the one-sided Monte Carlo permutation test. Error bars show sem.
Extended Data Fig. 8 EcoG signals collected from the surface of the motor cortex of people with Parkinson’s disease enable accurate detection of hotspot initiation events during locomotion.
a, Two people with Parkinson’s disease, P2 and P3, were implanted with quadripolar cortical paddles inserted subdurally over the motor cortex. Each paddle was connected to an implanted Medtronic Summit RC + S device to acquire epicortical ECoG signals wirelessly. b, Examples of locomotor execution along the treadmill (5 s) and corridor (5 s) of participant P2. From top to bottom: stick diagram decompositions of left and right leg movements; neural features (low-pass filtered ECoG signal) from all four recorded channels; probability of left and right leg lift events with detected hotspot events (black dots); left and right limb length calculated as distance from the hip to the ankle joint. The white, light grey and dark grey backgrounds correspond to double stance, left and right swing gait phases, respectively. c, Decoding remains accurate for both tasks in P2 and for P3. The pie charts show the mean ± sem accuracy of the detections for P2 (events: left: 64; right: 62) and P3 (events: left: 70; right: 74) calculated by offline analysis using cross-validation. d, Histogram plots show the distribution of the temporal differences between real and detected events (events: P1 treadmill: left: 92; right: 118; P1 corridor: left: 64; right: 62; P2 corridor: left: 70; right: 74) calculated by offline analysis using cross-validation. Median temporal difference is provided and marked by a vertical black line.
a, In order to estimate the target for the immediate effects of our therapy, we sought to simulate the gait of P1 given his anatomy but in the absence of neurodegeneration. To this end, we generated a personalized neurobiomechanical model of P1. Step1: We personalized the Lower Limb model1 to P1’s anatomy using morphological and physiological scaling. We performed the morphological scaling based on full-body motion tracking using the Vicon system. We then implemented the physiological scaling based on segmentation of muscles’ cross-sectional area (CSA) from CT images. Step2: We optimized the reflex-based gait controller using the SCONE software2,3. This controller is composed of phase dependent reflexes providing muscle excitation based on muscle length, velocity or force feedback. Step3: We simulated P1 gait in the absence of neurodegeneration using Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES)4 to optimize controller parameters. About 500 generations of CMA-ES were necessary to reach a stable gait from initialization. Once the parameters of the controller have converged, we generated 200 steps of this model and extracted full kinematics of lower limbs and muscle activity. The graphs show the mean with tubes showing the mean +/- standard deviation range of values. b, We used the P1’s CT and MRI scans to generate a three-dimensional anatomical model of the spine, which we then used to plan the surgical placement of the epidural spinal array over the entire lumbar spinal cord. Step1: P1 underwent a CT and a structural 1.5 T MRI scan of his spine. Step 2: We segmented the vertebral bones and disks from the CT scan, and segmented the spinal cord tissues (spinal cord, spinal cord roots and cerebrospinal fluid) from the MRI scan. We co-registered the tissues segmented from the two different scans, and combined them into a 3D anatomical model of the P1 spine. Step 3: We loaded a 3D model of the spinal array and placed it centred over the dorsal side of the spinal cord covering the L1-L5 spinal segments. Position of the array with respect to the segmented vertebral column determined the insertion point of the array to be between L1 and T12 vertebra. During the surgery, we opened the access to the surface of the dura by small incisions and a T12/L1 flavectomy, placed the tip of the array over the midline of the exposed dura and advanced the array rostrally to the target location. We accurately adjusted the medial and segmental position of the paddle array by monitoring the muscle responses to single-pulse EES delivered by different array electrodes. Step 4: After the surgery, we performed a post-operative CT scan to reconstruct the position of the spinal array with respect to the patient spine. The actual placement of the array was within 1 cm of the preoperative plan, as expected due to segmentation and co-registration inaccuracies.
Extended Data Fig. 10 Spinal cord neuroprosthesis delivering spinal EES in synchrony with attempted movements alleviates gait deficits of PD alone and synergistically with DBS of the subthalamic nucleus.
a, Examples show 3 s of P1’s locomotor execution along the corridor in four combinations of using or not the kinematically-controlled spinal cord neuroprosthesis (marked ‘EES’) and DBS of the subthalamic nucleus, and of the personalized neurobiomechanical model of P1. From top to bottom: stick diagram decompositions of left and right leg movements; four IMU features used to control the EES; probability of left and right weight acceptance events; detected hotspot events (broken vertical lines), periods of stimulation using a combination of 10 EES protocols targeting the left and right weight acceptance, propulsion and leg lift hotspots; and left and right knee angles. The white, light grey and dark grey backgrounds correspond to double stance, left swing and right swing gait phases, respectively. b, The plots showing EMG of left and right hip (iliopsoas), knee (gastrocnemius medialis) and ankle (vastus lateralis) leg muscles when transitioning from DBSON - EESOFF into DBSON - EESON conditions illustrate the change in muscle activation. Bar plots show the burst amplitude of left and right gastrocnemius medialis muscle during the propulsion phase of gait in DBSON - EESOFF (gait cycles: left: 79; right: 72) and DBSON - EESON (gait cycles: left: 136; right: 134) conditions. c, With the progressive use of spinal cord neuroprosthesis and DBS therapies, the motor neuron activation dynamics became more similar to that of the P1 neurobiomechanical model. The colorplots show the Target spatiotemporal spinal map derived from the P1 neurobiomechanical model, and the left and right leg spatiotemporal spinal maps of P1 in four combinations of using or not the kinematically-controlled spinal cord neuroprosthesis and DBS of the subthalamic nucleus (number of gait cycles: Target: 168; left leg: DBSOFF - EESOFF: 130; DBSON - EESOFF: 67; DBSOFF - EESON: 119; DBSON - EESON: 95; right leg: DBSOFF - EESOFF: 131; DBSON - EESOFF: 66; DBSOFF - EESON: 118; DBSON - EESON: 91). Surface correlation between the Target and therapy spatiotemporal spinal maps are shown on Fig. 5d. d, Bar plots show measures of gait quality, efficacy and symmetry, as well as balance: participants walk score (number of trials: DBSOFF - EESOFF: 2; DBSON - EESOFF: 2; DBSOFF - EESON: 5; DBSON - EESON: 5), stride length (number of gait cycles: P1 target model: 335, DBSOFF - EESOFF: 239; DBSON - EESOFF: 120; DBSOFF - EESON: 297; DBSON - EESON: 209), max knee angle (number of gait cycles: Target: 336; DBSOFF - EESOFF: 328; DBSON - EESOFF: 175; DBSOFF - EESON: 431; DBSON - EESON: 339), gait phase asymmetry (number of gait cycles: DBSOFF - EESOFF: 119; DBSON - EESOFF: 60; DBSOFF - EESON: 148; DBSON - EESON: 103), step length asymmetry as measured by the ratio between lengths of the left and right steps (number of gait cycles: P1 Target model: 167; DBSOFF - EESOFF: 119; DBSON - EESOFF: 60; DBSOFF - EESON: 148; DBSON - EESON: 103), stride time coefficient of variability (number of gait cycles same as for stride length), and arm swing angle (number of gait cycles: DBSOFF - EESOFF: 328; DBSON - EESOFF: 175; DBSOFF - EESON: 431; DBSON - EESON: 339). e, Decoding of hotspot initiation events from IMU signals to control the spinal cord neuroprosthesis remains accurate both when DBS is on or off. Histogram plots show the distribution of the temporal differences between real and detected events (events: DBSOFF - EESON: 462; DBSON - EESON: 328) when using the brain-controlled spinal cord neuroprosthesis. Median temporal difference is provided and marked by a vertical line. f, The bar plots show improvements in endurance, as measured by the distance covered during a 6-minute walking test, after the three-month rehabilitation supported by the spinal cord EES and 1-year after (n = 1 test in each condition). g, The bar plots show the gains in balance, as measured using the Mini-BESTest, after the three-month rehabilitation supported by the spinal cord EES (n = 1 test in each condition). h, Improvements in balance and freezing of gait, as measured by the ABC questionnaire and FoG questionnaire, prior and post-rehabilitation (n = 1 filled-out questionnaire in each condition). i, The bar plots show the sub-categories of the quality of life PDQ-39 questionnaire scores before and after the three-month rehabilitation supported by the spinal cord EES (n = 1 filled-out questionnaire in each condition). *, **, *** significant difference at p < 0.05, p < 0.01, p < 0.001, respectively, using two-sided Wilcoxon rank sum test or the one-sided Monte Carlo permutation test. n.s., not significant (p ≥ 0.05) according to the same tests. Error bars show sem.
A 31-page document with additional methods and 10 supplementary tables
NHP MPTP model of PD accurately replicates kinematic and muscle activity gait deficits of PD
A brain-controlled spinal cord neuroprosthetic that alleviates locomotor deficits due to PD
Real-time detection of hotspot events
About this article
Cite this article
Milekovic, T., Moraud, E.M., Macellari, N. et al. A spinal cord neuroprosthesis for locomotor deficits due to Parkinson’s disease. Nat Med 29, 2854–2865 (2023). https://doi.org/10.1038/s41591-023-02584-1
This article is cited by
Nature Medicine (2023)