To recover from spinal cord injury (SCI), the severed wiring of the nerves would have to be rebuilt exactly as it was before the injury—for a long time, that is what was believed. The 2019 review we read here writes that decades of basic research have challenged this premise.
This article organizes the review into two toolboxes. The biology toolbox holds “grow, attract, provide a scaffold”; the engineering toolbox holds “wake dormant circuits, train, bypass”. A third tool links the two: the clock. Which tool to use, when, and in what order. Merely expecting the effects to add up is not enough—that was the review’s bitter point.
According to the estimate of the US National SCI Statistical Center (2016 edition) cited in the review, the annual incidence in the United States is about 54 cases per million people, or about 17,000 cases a year. The World Health Organization (WHO) roughly estimates that 250,000 to 500,000 people worldwide suffer an SCI every year. Half of those affected remain paralyzed, living for decades with permanent disabilities. Lifetime costs are estimated at 1 million to 5 million US dollars per person, depending on the level and severity of the injury. What is more, the review’s introduction writes that, as of 2019, not a single randomized clinical trial (RCT) had demonstrated the efficacy of a repair strategy for improving functional recovery.
Two caveats before we begin. First, this is a 2019 review, a snapshot of its time. Statements that may change over time are marked “as of 2019”, and later developments are gathered, with sources, in the second half. Second, the evidence the review summarizes is mostly from animal experiments, and the human data came from studies with small numbers of people.
Publication Details
- Paper: Spinal cord repair: advances in biology and technology
- Type: Review Article
- Journal: Nature Medicine 2019, vol. 25, no. 6, pp. 898–908 (received 29 January 2018, accepted 1 May 2019, published online 3 June). PMID 31160817
- DOI: 10.1038/s41591-019-0475-6
- Impact Factor: 2025 JIF 52.5 (Clarivate, released June 2026)
- Access: Subscription (not open access). No author manuscript is available in PubMed Central either
- Citations: 500 in Crossref (retrieved 11 October 2026)
- Funding: European Research Council (the recipient is given as “to G.G.”, as in the original), Swiss National Science Foundation, National Center of Competence in Research (NCCR) Robotics, US National Institutes of Health, Dr Miriam and Sheldon G. Adelson Medical Foundation, Wings for Life
- Conflicts of interest: Courtine holds several patents related to the work covered in the review and is a founder and shareholder of GTX medical, a company developing treatments for spinal cord injury. Nothing is declared for Sofroniew
About the Corresponding Author, Grégoire Courtine
Grégoire Courtine is a full professor at the Swiss Federal Institute of Technology in Lausanne (EPFL) and, since 2025, director of the Neuro X Institute. Since December 2019 he has co-led the joint center ”.NeuroRestore” with the neurosurgeon Jocelyne Bloch. He earned his doctorate in 2003 under international joint supervision by the University of Burgundy in France and the University of Pavia in Italy, and after postdoctoral work at UCLA and elsewhere he moved to EPFL in 2012. With epidural electrical stimulation (EES) at the core, he researches technologies that use devices to put the remaining circuits to work, together with rehabilitation, and in 2026 he received the Queen Elizabeth Prize for Engineering (QEPrize; shared by 9 laureates, including him).
GTX medical, named in the conflict-of-interest statement, is the name that the company he co-founded with Bloch and others in 2014 adopted in 2018; today it is the ONWARD Medical group. The company’s annual reports list him as Chief Scientific Officer in fiscal year 2021, part-time Chief Scientific Officer in fiscal years 2022–2023 and scientific advisor from fiscal year 2024, and he stepped down from the board of directors on 4 July 2025. He also holds stock options (fiscal 2025 annual report).
Co-author Michael V. Sofroniew is a Distinguished Professor (MD, PhD) in the Department of Neurobiology at UCLA, known for his research on the “borders” of astrocyte scars. The pairing of the Courtine lab in engineering with the Sofroniew lab in the biology of the lesion site comes into play in the critique in the second half.
What Wasn’t Understood Before?
”Complete” Has Two Meanings
Spinal cord injuries are broadly divided into complete and incomplete, but in the legend to Figure 1 the review distinguishes two meanings of “complete”.
An anatomically complete injury is one in which the entire cross-section of the spinal cord has been severed. Experimentally, it is produced by a full transection or a complete crush, but it is uncommon in humans. A clinically complete injury, on the other hand, is defined by function: in the review’s description, it is a state in which neither motor nor sensory function can be detected below the injury (in the clinical severity classification, it is determined by the absence of any sensory or motor function in the lowest sacral segments, S4–S5). Even so, some tissue and connections may remain, and this is what later creates an opening for the engineering toolbox. Moreover, after such injuries, detrimental changes continue for decades below the injury. Even if nothing is done, the clock keeps running. In incomplete injuries, bridges of tissue spanning the injury remain, and through the reorganization of pathways and circuits, function returns to varying degrees.
The Scar Has Three Compartments
Regardless of size or severity, a mature lesion has three compartments.
- Non-neural lesion core: the central zone where the neural tissue is gone, filled by cells of the fibroblast family and by pericytes. Cysts (cavities) are also common
- Astrocyte scar border: a narrow band in which astrocytes, the cells that support neurons, multiply and surround the core. It corrals inflammatory cells within the injured area and protects the neighboring functioning tissue
- Spared but reactive neural tissue: the tissue outside it, which keeps working while reacting to the injury and rewiring its connections
To use a fire as a comparison: the burned-out center, the firebreak that stops the fire from spreading, and the surrounding streets where people go on living even as smoke drifts over them.
Two “Established Views”
Once development is complete, the capacity of neurons in the adult central nervous system (CNS) to extend axons, the cables of the wiring, declines, and after injury their growth programs restart spontaneously only to a limited extent. The cells filling the lesion core do not spontaneously support regrowth, and activated macrophages cause axons to retract. And for a long time, the main reason axons fail to grow was thought to be the “scars” formed by astrocytes. The other established view was that recovery requires the “original wiring” mentioned at the outset—that is, precise reconstitution of the connections that existed before the injury.
👦 Student: If a wire has been cut, I’d think the only option is to reconnect it the way it was.
🧬 Dr. Exotaro: For the wiring in a house, that’s true. But the spinal cord is closer to a road network than to electric wires. Even if a highway collapses, deliveries can sometimes still get through by linking local roads into a detour. What’s more, the warehouse at the destination has a skilled site manager: as long as a short instruction gets through, the manager reads the situation on the ground and organizes the rest of the work on their own.
What Did This Paper Find?
Recovery Is Possible Even Without the “Original Wiring”
Box 1 of the review rethinks the requirements for recovery. First, detours. After an incomplete injury, spinal circuits reorganize spontaneously (the evidence comes from studies in rodents and monkeys). In rodents, even when multiple incomplete lesions cut every pathway descending directly to the lumbar cord, relay neurons connecting points within the spinal cord built detour circuits and restored rudimentary voluntary walking. The review suggests that relay neurons, even if newly derived from grafted cells, may be sufficient for meaningful recovery (we also covered detours in a study of rats injured at two sites).
Next, the site manager—that is, the automaticity of the spinal cord. The spinal cord is now regarded as a “smart information-processing interface” that continuously translates sensory information and input from the brain into muscle activity suited to the goal and the environment. Even in animals that have completely lost input from the brain, when given a boost such as stimulation, spinal circuits use sensory information as cues to manage standing, walking and even climbing stairs. What steers the reorganization is proprioception, which conveys the position of the limbs; in mice whose proprioceptive circuits did not work, recovery itself was lost.
The conclusion is cautious. Delivering only a limited amount of information to the spinal cord below the injury may be enough to restore meaningful motor control.
The Biology Toolbox—Grow, Attract, Provide a Scaffold
Rewire. In incomplete injuries, the aim is to enhance the plasticity of the remaining circuits (the property that lets the wiring reorganize). In rats, intermittent hypoxia—breathing air low in oxygen in short, repeated bouts—drew this out. A clinical study suggested that, combined with rehabilitation, it may also improve walking in humans with incomplete injuries, but at that stage confirmation would still require a phase 3 trial with a large number of patients. For strategies targeting chondroitin sulfate proteoglycans (CSPGs) and NOGO (RTN4) receptors, long regarded as obstacles to growth, evidence was also accumulating to suggest that their main target is not regeneration across the lesion but synaptic plasticity in spared tissue.
Three requirements. To get axons beyond the lesion core in a complete injury, three things are probably needed, the review concludes from experiments in mice and rats: (1) grow = activate the growth programs inside neurons, (2) scaffold = provide a growth substrate that supports axons, and (3) attract = supply growth factors that draw axons in. Each of the three was necessary, and only when all were present did sensory axons and the axons of relay neurons (propriospinal) regrow robustly across the core. This, however, concerns regrowth, not functional recovery. What stands out is that the chemoattractive factors required differed between sensory neurons and relay neurons.
Rethinking the scar. The review argues that there is strong evidence that astrocyte scar borders are not a primary cause of regeneration failure. In mice, eliminating the borders did not make axons regenerate, and regrowth stimulated by treatment actually decreased; chemoattracted sensory axons grew in mice and rats even with the borders present. However, the key experiments (refs. 65 and 75) are both papers from the Sofroniew lab (with him as last author; Courtine is also a co-author on 75). This part should be read as the authors’ own position.
Provide a scaffold. In animals, grafts of Schwann cells and other cells helped host axons cross the core, and as of 2019 clinical trials of autologous Schwann cell grafts were under way. Grafts of neural progenitor cells built new relay circuits in rodents and restored partial communication across complete injuries, but functional recovery remained limited, the review acknowledges. In animal biomaterials research, placing hydrogels that slowly release growth factors in sequence, at staggered times, drew axons beyond the injury. Here, too, there is a clock.
👦 Student: Do all three requirements really have to be in place?
🧬 Dr. Exotaro: If you want a climbing vine to cross a river, you need a vigorous seedling (the growth program), a support for it to climb along (the scaffold) and light showing it where to head (the attraction). If even one is missing, it won’t reach the far bank, and different kinds of plants prefer different light. But just getting across does not make it bear fruit.
The Engineering Toolbox—Wake, Train, Bypass
Wake. The sensorimotor circuits below the injury are often intact, yet having lost modulation and drive from the brain, they cannot work. Neuromodulation makes up for this with electricity or drugs. The mechanism was still debated at the time, but there was evidence suggesting that electricity applied from the dorsal side, as in EES, spreads to the dorsal roots—the entry point for sensation—first exciting the large-diameter proprioceptive fibers, which conduct current easily, and thereby awakening the circuits. As of 2019, it had been reported that a small number of people with motor complete paralysis could voluntarily move parts of their legs during spinal cord stimulation. The review’s reading is that, even in clinically complete injuries, pathways from the brain often survive but are too weak on their own, and stimulation brings them into play. Spatiotemporal stimulation targets specific dorsal roots and delivers short bursts timed to coincide with the natural activation of motor neurons.
Train. Treadmill training in animals works without any involvement of the brain, and it did not restore interaction with the brain. When training in a robotic body-weight support system that encourages this interaction was combined with neuromodulation, voluntary control of walking returned in rats and humans (for humans, in small numbers), and voluntary movement without stimulation returned as well. In the animals, cortical commands had been rerouted through new brainstem pathways and relays within the spinal cord. The mechanisms, in the review’s words, “remain enigmatic”.
Bypass. Brain–computer interfaces (BCI), which operate machines with signals from the brain, were at the time shifting from substitution to restoration, and “neural bypasses” linking cortical signals directly to electrical stimulation of muscles enabled functional upper-limb movements in 2 people with tetraplegia (see also the later study of a double neural bypass). For a brain–spine interface (BSI) linking the leg motor cortex to spinal cord stimulation, what the review cited were studies in rodents and monkeys.
The barriers are high, too. As of 2019, the Utah array, then the only intracortical probe approved for clinical use, lost signal quality rapidly, and the functional benefit of BCIs was modest compared with their costs and technical challenges. Exoskeletons were cumbersome to put on and operate, and slow, and people with SCI chose wheelchairs to get around—the review is frank about this.
Addition Is Not Enough—the “Clock” That Links the Two Boxes
As of 2019, both biology and engineering had shown potential in animals but had not reached clinically important improvement in severe injuries. If they target different mechanisms, combining them should produce synergy. Logical as that is, in practice it proved more complex than expected, the review writes.
In studies in rats and other animals that combined physical training with treatments that enhance plasticity, which movements were practiced, and in which time frame (temporal window) each therapy was given, determined whether plasticity tipped toward the beneficial or the detrimental. The review also writes that many failed combinations remain buried in laboratory notebooks. And yet, it warns, dubious treatments combining unproven cell therapies, immature engineering methods and intense exercise programs have unfortunately flourished in clinics offering medical tourism.
The review admits that combinations so far have been designed “algebraically”—that is, expecting simple addition without considering the interactions between the mechanisms. Even the vision of stacking up all the tools “may be simplistic”. Yet with well-targeted approaches, it also writes, the two may catalyze each other. On that basis, it argues that combinations will need to obey strict temporal windows.
👦 Student: I thought that if you combined two drugs that work, they would work even better.
🧬 Dr. Exotaro: Not necessarily. A treatment that enhances plasticity softens neural circuits like clay, and training kneads that clay into shape. If a different movement is practiced while it is soft, it sets in a different shape. Thinking in terms of addition leaves out this interaction and this timing.
Animal models come in for self-criticism, too (Box 2). Most interventions are tested in rodents with a partial cut of the spinal cord, but partial cuts recover well spontaneously even in humans, and they lack many features of the injuries common in humans. What resembles humans, permanent paralysis included, is severe contusion or crush (a squashing injury). The authors argue that interventions should be tested in both models, and they also hold out hope for small, well-designed studies in monkeys (nonhuman primates).
How Will the Future Change? (The Path to the Clinic)
The review’s prescription is to combine biology and engineering over time, based on the interactions between their mechanisms. That, the review concludes, can only come from partnership between biologists and engineers.
What happened between 2019 and 2026
- Demonstrations in small numbers: 3 people whom the paper describes as having “complete sensorimotor paralysis” were able, with new electrodes and activity-specific stimulation, to stand, walk, cycle, swim and control their trunks within a single day (Rowald, A., et al., Nat Med 2022;28:260–271). BSI also reached humans: 1 person with chronic incomplete cervical spinal cord injury became able to walk overground with crutches even with the BSI turned off (Lorach, H., et al., Nature 2023;618:126–133). The device on one side was removed because of infection and later replaced
- From a single-arm trial (no control group) to marketing authorization: The multicenter Up-LIFT trial, which combined transcutaneous spinal cord stimulation with rehabilitation for the arms and hands in chronic tetraplegia, enrolled 65 people, of whom 60 completed it, and 43 of the 60 (71.7%) were classed as responders in both of the two assessment domains defined by the trial (Moritz, C., et al., Nat Med 2024;30:1276–1283). On 19 December 2024, the US Food and Drug Administration (FDA) granted De Novo authorization (DEN240014) to ARC-EX. Its indication is to improve hand sensation and strength, in conjunction with task practice in the clinic, in people aged 18 to 75 with chronic, non-progressive deficits resulting from incomplete injuries at C2–C8. The FDA’s own decision summary lists as limitations that the probable benefit has been observed only in combination with intensive outpatient rehabilitation; that with a single-arm, within-subject-controlled, open-label design it is unclear how much of the improvement is due to the device; that no functional benefit from the device has been shown, nor any improvement in neurological level of injury or in severity grade (AIS); and that the durability of the effect has not been verified
- Biology turns to the “destination”: In 2023, Courtine, Sofroniew and colleagues reported that in mice with complete injuries, guiding axons all the way to their natural target regions led to substantial recovery of walking, whereas merely getting them across the lesion had no effect (Squair, J. W., et al., Science 2023;381:1338–1345)
- From Japan: In an open-label phase 1 trial that transplanted iPS cell-derived neural stem/progenitor cells into 4 people with complete paralysis (AIS grade A) in the subacute phase (14 to 28 days after injury), a group at Keio University met its primary endpoint of safety (Sugai, K., et al., Nat Med 2026;32:3438–3448). There was no control group, the motor improvement rests only on an exploratory comparison with historical patient cohorts, and engraftment could not be directly confirmed. The investigators’ conflicts of interest are also disclosed. It is a cautious sequel, in humans, to the neural progenitor cell grafts discussed in the review
Of these, 4 papers—Rowald, Lorach, Up-LIFT and Squair—are studies with Courtine among the authors, and ONWARD is both the sponsor of Up-LIFT and the applicant for ARC-EX.
So does that sentence from the review’s introduction still hold today? In neurotechnology, for the review’s core technologies (spinal cord stimulation and brain–spine interfaces), no multicenter confirmatory RCT showing efficacy with activity-level function as the primary endpoint can be found as of October 2026. For transcutaneous spinal cord stimulation, a multicenter, triple-blind, sham-controlled RCT (eWALK) in 50 people with chronic thoracic spinal cord injury, with its statistical power set in advance, found that adding stimulation to walking training did not improve the primary endpoint of walking ability compared with sham stimulation (Bye, E. A., et al., eClinicalMedicine 2026;93:103802). For non-invasive brain stimulation, there are also small single-center positive RCTs. In biology, no biological repair therapy that has shown functional efficacy in a multicenter, blinded, controlled trial can be found as of October 2026 (NISCI, VX-210, RISCIS and others failed to show efficacy on the primary endpoint in the overall population. Some post hoc or subgroup analyses suggested improvement, and RISCIS was terminated before reaching its planned enrollment). On the other hand, small randomized trials, such as single-center or open-label ones, have also produced positive reports, and confirmation is awaited.
What the review called “pushing toward clinical translation” took shape first on the neurotechnology side, and a device reached the market on the evidence of a single-arm trial. On the biology side, some drugs have advanced as far as phase 2–3 trials, but an effect on functional recovery has not yet been established. The iPS cell-derived neural stem/progenitor cell transplantation introduced here is at the early clinical stage.
How to Critically Read This Research (Limitations and Paths to Improve It)
What Deserves Credit
Dividing the lesion into three compartments provides a common language for where repair should aim. The reinterpretation in Box 1 is an approach that puts biology and engineering on the same map. The writing is also careful to avoid overstatement, and the review is honest in acknowledging that many failures go unpublished, in warning about medical tourism with the engineering side included, and in criticizing the biases of animal models in Box 2.
Points to Read With Caution
1. It is a narrative review. Unlike a systematic review, it reports no prespecified protocol, search strategy, PRISMA-compliant record of study selection or grading of study quality. What gets picked up is the authors’ call.
2. Reference selection and self-citation. Of the 174 references, 31 (17.8%) include Courtine or Sofroniew as an author (Courtine 23, Sofroniew 11, both 3). This count is exact as far as the author lists go, checked against all authors in PubMed/Crossref (papers by lab members or collaborators alone are not counted). Their names are visible in only 10; in 21, they are abbreviated to “et al.” by the journal’s style and cannot be seen from the printed reference list alone. In the section on neuromodulation, 11 of 37 citations (29.7%) are papers with Courtine as an author. It is natural for a leading lab to cite its own work; the problem is that, in a review with conflicts of interest, readers find it hard to check whether the selection is free of bias.
Some assertions should also be read as the authors’ own position. The claim that the scar border is not a primary cause rests mainly on experiments from the Sofroniew lab, and the sentence that “there is at present no compelling evidence” that growth-inhibiting molecules play major roles in regeneration across lesions rests on his single-author review (ref. 5). There are other views: a 2020 review writes that “NogoA and CSPGs are strongly inhibitory to the regeneration of adult axons” (Fawcett, J. W., Neurochem Res 2020;45:144–158), and there is also an animal study in which blocking the interaction between astrocytes and type I collagen prevented scar formation and improved regrowth and function (Hara, M., et al., Nat Med 2017;23:818–828). The debate is not settled.
3. Conflicts of interest. Courtine’s patents and his relationship with GTX medical (now the ONWARD Medical group) are disclosed. The issue is how to read the engineering section. The review portrays epidural spatiotemporal stimulation favorably, while judging that intraspinal stimulation, if similar outcomes are confirmed, would become “less attractive than other strategies” because it requires inserting electrodes into the gray matter. There is no basis for saying this assessment is wrong, but the fact that the person making it holds related patents and shares in a related company is a premise readers should know.
4. Animal-centered evidence, and a citation mismatch. The authors themselves caution that it is “not yet certain” whether the mechanisms in Figure 1 also operate in humans. Note also that the references (105, 129, 133) for the sentence saying spatiotemporal stimulation enabled full weight-bearing locomotion “in animal models and humans” are all animal studies; the paper on human walking (ref. 111, Wagner 2018) is cited in the preceding sentence and elsewhere.
5. A motor-centered view, and a clock with no face. The authors themselves state that they mainly discuss motor recovery, and acknowledge that the recovery of autonomic and sexual functions is among the top priorities of people with SCI. The bladder, bowel and pain are not addressed in the text (outcome measures like those in a study on bladder recovery lie outside this map). And the clock has no face. The phrase “temporal windows” appears 4 times, but how many days after injury is never specified (it is left as a task for the future), and the words “chronic” and “subacute” do not appear in the text either.
6. A 2019 snapshot. BSI, for which the review cited only animal results, has already moved on to humans (previous section).
How Could Its Quality Have Been Improved Further?
- A preregistered protocol, PRISMA-compliant reporting, risk-of-bias assessment and GRADE
- A synthesis tabulating the sample size, the presence or absence of controls and the effect sizes of the human studies
- Adding a co-author without conflicts of interest to the neurotechnology section
- Also including failed combinations and negative studies
- Outcome measures that people with SCI prioritize, such as bladder, bowel, sexual and autonomic function
- Preregistered multicenter preclinical studies in contusion models
Dr. Exotaro’s Perspective
I myself do research on treating spinal cord injury with mesenchymal stem cells (MSC) and the extracellular vesicles (EV) they release. Please read what this article says about cell therapy and about MSCs and EVs with that in mind.
My Field Is Not on the Map
MSC or EV therapies do not appear even once in this review. The word “mesenchymal” occurs only once in the paper, in the legend to Figure 1, describing the scar of the lesion core; the words MSC, exosome and extracellular vesicle appear zero times. Placed in the two toolboxes, what MSC and EV research targets is the environment of the injury site, such as inflammation, macrophages and microvessels. I do not consider them tools that reconnect the wiring on their own, either.
The Clock Was in My Data, Too
In a rodent model of severe spinal cord injury, we reported that intravenously administered MSCs themselves did not reach the injury site, whereas small extracellular vesicles (sEV) released by MSCs migrated to the injury site and associated with M2-type macrophages. Under these experimental conditions, the group given the same total amount of sEVs as a single dose showed no significant motor improvement, while the group in which it was divided over 3 days achieved motor recovery close to that with MSCs themselves (Nakazaki, M., et al., Journal of Extracellular Vesicles 2021;10(11):e12137). Over the same 6 days and with the same total amount, continuous administration by osmotic pump produced greater motor recovery than daily injections (Nakazaki, M., et al., Neurotherapeutics 2026;23:e00892; explainer article). Both are preclinical results, but not only “what” but also “when, and how to deliver it” made a difference. It is a question that connects to the review’s clock.
The Warning Applies to My Field, Too
Combinations including unproven cell therapies had spread in clinics offering medical tourism. As a physician who researches cell therapy, I take this warning head-on. Clinics that offer such treatments without adequate verification may deprive patients not only of their money but also of time that could be spent on appropriate care and rehabilitation.
Designing From the Clock
If MSCs or EVs are considered one piece of a combination, their partners would likely be rehabilitation and neuromodulation. Would an intervention like sEVs, which acts on the environment, make the circuit reorganization that those partners draw out more likely to occur—or, if the temporal window is wrong, would it strengthen detrimental plasticity? This is a hypothesis, and testing it would start with animal contusion models. So that combinations are not designed “algebraically”, as addition, the timing of administration itself should be made an experimental variable. That is the next piece of homework for this field.
Note: this article was written with the assistance of AI.