The neurons that restore walking after paralysis

The study addresses how epidural electrical stimulation (EES) combined with rehabilitation (EESREHAB) restores walking after chronic spinal cord injury (SCI). Although prior clinical reports showed EES can immediately enable stepping and that training with EES improves function even when stimulation is off, the neural mechanisms and specific spinal neurons engaged and remodeled by EESREHAB were unknown. The authors hypothesized that EESREHAB drives activity-dependent selection of a specific lumbar spinal cord neuronal subpopulation that becomes essential for walking after paralysis. They tested this in humans and in a mechanistically tractable mouse model, aiming to map cellular and spatial transcriptional changes and identify neurons necessary and sufficient for recovery.

Previous studies demonstrated that EES can acutely reactivate nonfunctional lumbar spinal networks and enable stepping in people with complete or severe SCI, and that activity-based rehabilitation with EES further improves motor recovery, sometimes persisting without stimulation. Foundational work established the role of descending brainstem and cortical pathways in locomotion and showed that SCI disrupts these supraspinal inputs, rendering lumbar circuits nonfunctional. Computational and experimental studies indicated EES recruits large-diameter afferents and must preserve proprioceptive signaling to enable locomotion. Prior rodent work implicated reticulospinal pathways and various spinal interneuron classes (including V2a) in locomotor control and plasticity after injury. However, no prior work had identified the specific spinal neurons selected by EESREHAB that organize recovery of walking.

Human component: A first-in-human clinical trial (STIMO; NCT02936453) enrolled nine individuals with chronic SCI (six with severe/complete motor paralysis with some sensation, three with complete sensorimotor paralysis). Participants were implanted with an epidural neurostimulator connected to a multi-electrode paddle lead (first six: repurposed pain paddle; last three: purpose-built lead targeting thoracic–lumbosacral dorsal roots). Closed-loop, biomimetic EES protocols were configured, and participants underwent 5 months of EESREHAB (standing, walking, and exercises with EES on, 4–5 sessions/week) using a multidirectional body-weight support robotic system. Outcomes included lower limb motor scores, 6-minute walk test distance, and FDG-PET imaging of spinal cord metabolic activity during walking before and after EESREHAB. Muscle responses to single EES pulses were also analyzed to assess aberrant long-latency responses.

Mouse model: Researchers established a translational platform replicating EESREHAB features. Adult mice received a severe mid-thoracic contusion SCI (T8) producing permanent paralysis; tract tracing plus tissue clearing (CLARITY) and light-sheet imaging confirmed interruption of corticospinal and depletion of glutamatergic reticulospinal fibers caudal to injury. A custom robotic body-weight support system for overground walking and epidural electrodes were engineered with stimulation protocols avoiding off-target ventral root recruitment. Optogenetics enabled volitional modulation via motor cortex neurons; reticulospinal vGi neurons served as a relay for cortical control. EESREHAB protocols were applied to restore walking with and without ongoing stimulation.

Activity mapping: Whole-spinal-cord cFos immunolabeling following walking, combined with CLARITY and automated detection, quantified neuron activation patterns. FDG-PET in humans and cFos in mice assessed changes in lumbar activity after EESREHAB.

Single-cell and spatial genomics: The team performed single-nucleus RNA-seq (snRNA-seq) on lumbar spinal cord across eight experimental conditions capturing EESREHAB’s features (including 30-minute terminal conditions) from 24 mice, yielding 82,093 nuclei (20,990 neurons) and identifying 36 neuronal subpopulations by unsupervised clustering. Spatial transcriptomics on 61 sections (22,127 barcodes) was registered to a common coordinate framework to map gene expression and deconvolve subpopulation locations. Multiplexed RNAscope validated spatial localization.

Computational prioritization: To identify perturbation-responsive cell types, they applied Augur (cell type prioritization) to snRNA-seq across key experimental comparisons and developed Magellan (spatial prioritization) to localize perturbation responses within tissue. They embedded single-nucleus data in the spatial framework to cross-validate cell type and spatial prioritization.

Circuit mapping and physiology: Monosynaptically restricted trans-synaptic viral tracing identified inputs to prioritized neurons (large-diameter parvalbumin+ DRG afferents and ventral gigantocellular reticulospinal neurons) and outputs to ventral spinal neurons. Optotagging and single-unit recordings tested functional inputs. Synaptic appositions from prioritized neurons to glutamatergic, GABAergic, and cholinergic ventral neurons were quantified across conditions. Muscle responses to EES pulses assessed aberrant long-latency components and effects of manipulating prioritized neurons.

Causal manipulations: Optogenetic and chemogenetic silencing/activation targeted excitatory SCVsx2::Hoxa10 neurons (a subset of V2a interneurons) to assess necessity and sufficiency for walking restoration with EES and for recovery without EES. A custom implant combined epidural electrodes and red-shifted microLEDs for in vivo optogenetic silencing during EES-enabled walking. Chemogenetic tools (hM4Di/hM3Dq) manipulated activity acutely and chronically. Targeted ablation (flex-DTR) of SCV2a::Hoxa10 neurons tested their role in spontaneous recovery after lateral hemisection SCI. Chronic silencing during EESREHAB assessed impact on remodeling and recovery.

• Clinical efficacy and remodeling: All nine participants immediately improved or regained walking with EES in a robotic interface; two with complete sensorimotor paralysis exhibited volitional modulation of step amplitude with EES on. After 5 months of EESREHAB (4–5 sessions/week), weight-bearing improved and outdoor walking with EES and assistive devices became possible. Participants with residual function showed marked increases in lower limb motor scores, restoring walking without EES in four participants. FDG-PET during walking showed decreased lumbar spinal cord metabolic activity after EESREHAB, indicating reduced neuronal activity (metabolic activity mixed-effects model: t = −3.2, P = 0.002; lower limb motor scores: paired t-test t = 3.7, P = 0.0063; 6-min walk distance: paired t-test t = 3.5, P = 0.0076).
• Mouse model recapitulation: EES immediately restored supported walking after severe contusion SCI; EESREHAB produced lasting recovery even with EES off (walking performance improvement: one-way ANOVA, Tukey HSD SCI vs EESREHAB EESoff P = 3.3 × 10^−11). Whole-cord cFos mapping revealed fewer active neurons during walking after EESREHAB versus SCI mice walking with EES (independent t-test: t = −5.7, P = 0.001).
• Identification of recovery-organizing neurons: Augur prioritized two excitatory interneuron populations expressing Vsx2 and Hoxa10 (SCVsx2::Hoxa10) across all EESREHAB features. Magellan localized perturbation responses to intermediate laminae (SCVsx2::Hoxa10 location) and ventral laminae. Immediate early genes were enriched in SCVsx2::Hoxa10 during EES-enabled walking; longer-term remodeling genes modulated after EESREHAB.
• Circuit properties: SCVsx2::Hoxa10 neurons received monosynaptic inputs from large-diameter PV+ DRG afferents and vGi reticulospinal neurons; optotagged single units showed short-latency responses to both inputs. EESREHAB increased synaptic input density from large-diameter afferents (vGluT1 synapses: t = 4.9, P = 0.002) and reticulospinal fibers (t = 4.8, P = 0.0029) onto SCVsx2::Hoxa10 neurons. These neurons projected exclusively ventrally, forming dense synaptic appositions onto ventral neurons (52% glutamatergic, 77% GABAergic, 56% cholinergic). SCI reduced appositions onto ventral GABAergic neurons; EESREHAB reversed this (GABAergic appositions: uninjured vs SCI P = 5.2 × 10^−4; SCI vs EESREHAB P = 1.8 × 10^−4). Despite overall reduced cFos+ neuron counts post-EESREHAB, cFos activity in Vsx2 neurons doubled (t = 5.7, P = 0.0013).
• Functional impact on EES responses: SCI induced aberrant long-latency muscle responses to EES; EESREHAB abolished these in mice and humans. Chemogenetic silencing of SCVsx2::Hoxa10 after EESREHAB reintroduced aberrant responses; activation in chronic SCI abolished them.
• Causal role in recovery: Optogenetic silencing of SCVsx2::Hoxa10 instantly suppressed EES-enabled walking (repeated-measures ANOVA, Tukey HSD P = 0.0023); walking resumed upon light cessation. Chemogenetic silencing also suppressed EES-enabled walking (paired t-test: t = −21.3, P = 0.0002). Activation of SCVsx2::Hoxa10 in chronic SCI phenocopied EESREHAB-like recovery with or without EES (paired t-test: t = 5.3, P = 0.0013). Chronic silencing during EESREHAB prevented recovery (independent t-test: t = −3.5, P = 0.008) and blocked remodeling of inputs/outputs. Ablation of lumbar SCV2a::Hoxa10 neurons prevented full spontaneous recovery after lateral hemisection SCI (independent t-test: t = 5.9, P = 0.0004) and reduced reticulospinal sprouting below injury.
• Nonessential in uninjured walking: Silencing or ablating SCVsx2::Hoxa10 in healthy mice did not impair basic locomotion.

Findings support the hypothesis that EESREHAB drives activity-dependent selection of a specific excitatory interneuron population, SCV2a::Hoxa10, in lumbar intermediate laminae, which becomes necessary and sufficient for regaining walking after paralysis. EESREHAB remodeled inputs to and outputs from these neurons, integrating proprioceptive (large-diameter afferents) and brainstem (reticulospinal) signals and broadcasting commands to ventral locomotor circuits. The unexpected reduction in global lumbar neuronal activity during walking after EESREHAB likely reflects a more efficient, reorganized circuit wherein SCV2a::Hoxa10 neurons coordinate the activity of downstream networks, including inhibitory interneurons that also showed prioritization in long-term recovery comparisons. The cross-species approach—clinical outcomes and PET in humans, with single-cell and spatial cartography plus causal manipulations in mice—provides mechanistic insight into how neuromodulation and rehabilitation restore function and offers a generalizable framework (Augur and Magellan) to pinpoint cellular substrates of complex behaviors in response to perturbations.

EESREHAB restored walking and improved neurological function in nine people with chronic SCI and reduced lumbar spinal activity during walking. A mechanistic mouse framework identified SCV2a::Hoxa10 excitatory interneurons as recovery-organizing cells: they are not required for normal walking but are necessary and sufficient for recovery after paralysis. EESREHAB strengthens convergent proprioceptive and reticulospinal inputs onto these neurons and preserves their outputs to ventral locomotor networks, abolishing aberrant spinal responses and enabling efficient gait. The study introduces Augur and Magellan as powerful tools to prioritize cell types and spatial regions underlying behavioral recovery. Future research should validate analogous neuronal populations and remodeling in humans, optimize stimulation protocols to target these circuits, expand clinical trials to larger, diverse cohorts, and explore combinatorial therapies (e.g., brain–spine interfaces, regenerative strategies) that engage SCV2a::Hoxa10 networks.

• Clinical sample size was small (n = 9) and heterogeneous in injury severity; outcomes were not compared to a randomized control group.
• Mechanistic identification of recovery-organizing neurons was performed in mice; direct confirmation of the same cell population and circuit remodeling in humans remains indirect (e.g., PET signals are not cell-type specific).
• Some mechanistic experiments used small group sizes (e.g., n = 4–6), which, despite significant effects, may limit precision of effect estimates.
• EES technologies and paddle leads differed between early and later participants; generalizability to other devices and settings requires further validation.
• The framework focused on one excitatory interneuron subset; additional neuronal subpopulations likely contribute to recovery, and their roles were not exhaustively dissected.