Ventral Epidural Stimulation Restores Forelimb Function and Reduces Microglial Activation after Subacute Cervical Spinal Cord Injury

Abstract

Neuromodulation has recently garnered attention as a potent approach to improve sensorimotor function following neural injury. After spinal cord injury, electrical stimulation targets sensorimotor circuitry to enhance volitional movement; however, motor neurons are located in the ventral gray matter. Previously, we developed a versatile wireless stimulation system and surgical approach to selectively activate motor neurons caudal to a cervical spinal cord injury. Here, we empirically tested if ventral epidural stimulation is sufficient to promote lasting recovery of skilled motor function. Ventrally stimulated animals exhibited significantly improved forelimb function associated with electrically stimulated motor pathways. Further, stimulation produced increases in neuronal Nr4a1 expression, a subacute reduction in microglia Nr4a1 expression, and a shift from activated toward ramified microglial morphologies in the stimulated ventral gray matter. This study reports the first successful application of ventral spinal stimulation to improve function in an animal model and the first sustained improvement in skilled forelimb reaching when behavioral testing was decoupled from the stimulation.

Graphical Abstract

Introduction

Spinal cord injury (SCI) remains a debilitating condition despite decades of research. However, a promising treatment path is the incorporation of electrical stimulation strategies targeting the spinal cord, which have yielded exciting preliminary results in motor and sensory recovery.^1–4 Indeed, studies involving electrical stimulation of the spinal cord have been linked to improvements in standing,^5–7 walking,^4,8–10 cardiovascular function¹¹ and upper limb function.^9,11–13 Despite continued advances in the field, many of the mechanisms behind the improvements related to electrical stimulation of the central nervous system remain a mystery. Why, for instance, are not all subjects responders^3,11,14? Much work has been done to uncover the neural circuits involved in spinal neuromodulatory approaches to restore or enable motor function.^8,15–23 It is generally accepted that epidural and non-invasive transcutaneous spinal stimulation activate sensorimotor networks through large-caliber A1 afferents.^8,24,25 Researchers have observed differences in the neural circuits and motor outputs activated via intraspinal, epidural and transcutaneous spinal stimulation modalities;^{17,18,26–31} however, the functional and biological distinctions between each approach remain unclear.

More investigation into the potential mechanisms and different applications of electrical stimulation is needed to determine its clinical relevance, and mechanistic understanding is necessary to advance the field. In the case of people who suffer from quadriplegia, hand function is a major therapeutic target, and while there are a limited number of promising trials focusing on upper limb function, ultimately there have been limited improvements reported. For example, transcutaneous spinal stimulation or dorsal epidural stimulation leads to enhanced strength and kinematics after stroke in non-human primates and in humans.^9,32,33 In rats and humans, electrophysiological mapping studies in the cervical spinal cord have revealed that sensorimotor stimulation at the segmental level yields fused contractions and that isolated motor pools are not individually accessible.^34,35 While dorsal epidural stimulation has demonstrated efficacy, we hypothesize that electrical stimulation from the ventral surface could offer advantages for discrete activation of motor pools of the upper limb.

The need for novel approaches is exemplified in recent studies using dorsal epidural stimulation where neuroenabling stimulation improved function but did not lead to clear sustained functional recovery independent of stimulation.³⁶ Intraspinal microstimulation (ISMS) involves microwires implanted into the ventral horn of the spinal cord and enables discrete motor unit activation.³⁷ The use of ISMS to target the ventral gray motor units of the paw and wrist alone or combined with electromyogram (EMG) activity-dependent stimulation was able to sustainedly enhance forelimb function independent of the electric stimulation.^38–40 ISMS, unlike epidural and transcutaneous stimulation, can independently and discretely activate ensembles of motor neurons, which may contribute to sustained plasticity and functional motor recovery.^41,42 ISMS is a highly promising treatment approach given its capability to activate discrete upper limb motor pathways and improve skilled motor function; but it requires invasive surgery and, should electrodes fail, removal and/or replacement would represent a significant challenge.

We previously developed a device designed to target the ventral surface of the cervical spinal cord and identified unique electrophysiological signals elicited from the targeted circuits using a ventral approach (Fig. 1).⁴³ Utilization of a multi-electrode array or bipolar wires deployed in the ventral space enables discrete activation of motor units with similar fidelity to that of our previous ISMS studies.^37,39 We refer to our approach as epidural ventral spinal stimulation (VSS). In the current study, we implanted a VSS electrode at C6, two segments below a C4 spinal hemi-contusion lesion in a rat model of spinal cord injury. We applied epidural stimulation starting in the subacute period at 20 Hz and 80% motor threshold for 1 h per day. Our results indicate that VSS can produce significant improvement in upper limb motor recovery. Recovery was associated with reduced expression of the immediate early gene Nuclear Receptor Subfamily 4 Group A Member 1 (Nr4a1) in microglia and a morphological return of microglia to a ramified state. Together, these data establish VSS as a less invasive alternative to ISMS and illustrate its role as an efficacious treatment given the persistence of recovery when the VSS stimulator is turned off.

FIG. 1.

Model design and approach. (A) Visual graphic of the stimulator implant and surgical insertion at T4/5 and targeting of electrodes to C4 spinal level (B) Stimulator device design and layout (C) Example of upper limb EMG data showing typical thresholds achieved with an electrode implanted at C6 after C4 hemicontusion. Targeted muscles include infraspinatus, triceps and wrist flexors. EMG, electromyogram.

Results

Initial electrode placement and muscle EMG

At C6, in subacute preparations 1 week after injury, VSS optimally activated triceps followed by wrist flexors and infraspinatus. Activation thresholds ranged from 20 to 140 μA (Fig. 1C). All evoked potentials represented fused contractions, as we did not observe single muscle activation from any location at a given current input. During placement for therapeutic stimulation, we targeted electrodes to the C6 spinal level and confirmed proper placement (ipsilateral to the lesion) by physical observation of targeted forelimb twitch. Sham control animals received spinal injuries and spinal stimulators but did not receive spinal stimulation. Regular thresholding of therapeutic animals confirmed stable stimulation of the targeted forelimb during longitudinal behavior studies. We observed partial activation of the contralateral limb tactilely at higher thresholds; however, the dominant responses remained in the muscles ipsilateral to the stimulation and lesion through the longitudinal study. Over the course of the 8-week behavioral study, stimulation thresholds ranged from a minimum of 80 μA to a maximum of 500 μA in order to achieve a muscle twitch. The surgical approach combined with the anatomy of the ventral epidural space restricted electrode movement laterally with physiology and post-mortem explants confirming stable electrode localization.⁴³

Behavioral data rater training and inter-rater reliability

Prior to study onset, rats were pseudo-randomly assigned to treatment groups with pre-injury reaching performance as the balance variable, ensuring groups started out at similar baseline levels using an iterative balanced assignment approach.⁴⁴ Sham control rats reached with an average pre-training success of 61.7 ± 13.4% and stimulated rats achieved an average pre-training success of 62.7 ± 12.3% as a cohort. For reaching assessment, we utilized a group training and blinded data quantification approach to maximize the objectivity and rigor of forelimb reaching data.⁴⁵ Kappa Cohen inter-rater reliability assessment revealed an average agreement of 0.72 ± 0.11, with a maximal single scorer agreement of 0.76 ± 0.12 and a minimal single scorer agreement of 0.64 ± 0.11. The maximal total agreement between two scorers was 0.89 and the minimal agreement was 0.56 (Supplementary Fig. S1). While a Kappa Cohen score cannot be universally interpreted as “good”,⁴⁶ scores in the range of 0.60–0.75 and scores over 0.75 have been classified historically as good and excellent agreement between independent scorers in subjective scoring assessments, respectively.⁴⁷

Spinal cord injury for longitudinal studies

We performed a C4 hemicontusion injury for all experiments presented in this article. For the 8-week stimulation study, we assessed kinematic injury data between treatment groups and observed no significant difference in injury level based on maximum displacement, mean force (kDyne), and peak force (kDyne) values (Supplementary Fig. S2, n = 6 sham control, n = 5 stimulated). The mean force was 26.8 ± 13.2 kDyne for the sham control group and 27.5 ± 9.28 kDyne for the stimulated group. The peak force was 58.7 ± 4.80 kDyne for the sham control group and 65.7 ± 2.49 kDyne for the stimulated group. Finally, the max displacement was −0.779 ± 0.0036 kDyne for the sham control group and −0.780 ± 0.0035 kDyne for the stimulated group.

VSS improves skilled forelimb recovery

We analyzed skilled forelimb recovery using two quantitative approaches: 1) recovery to baseline of pellet retrieval success rate and 2) recovery of individual sub-movements. Data was collected weekly for 3 weeks prior to spinal cord injury to establish baseline functional performance. Data for functional assessment of recovery was collected in weeks 2 through 10 post injury. All data was collected at least 4 h after therapeutic stimulation was halted to allow for a washout period. Rats receiving C6 stimulation plateaued at an overall reaching recovery rate of 84 ± 6.2% of baseline compared to rats not receiving stimulation where reaching recovery rate plateaued at 47 ± 12% of baseline when averaged as a cohort over weeks 6, 7, and 8 (Supplementary Fig. S3, n = 5 stimulated, n = 6 sham control, p = 0.19). At week 7, we observed improved forelimb reaching success rate recovery (Supplementary Fig. S3, 85 ± 5.55% stimulated vs. 33 ± 9.50% sham control, p = 0.045). This effect lost statistical significance at week 8 (Supplementary Fig. S3, 83 ± 7.16% stimulated vs. 46 ± 12.0% sham control, p = 0.22). When we combined all quantified features of skilled reaching into an Aggregate Functional Score, C6 stimulated animals exhibited a statistical improvement in reaching over their sham control counterparts as early as two weeks post-stimulation where the aggregate functional score returned to 70.05 ± 7.2% of baseline for sham control animals and 86.8 ± 5.9% for stimulated animals (Fig. 2C, n = 5 stimulated, n = 6 sham control, p = 0.023). At week 6 the sham control group demonstrated a 84.6 ± 6.7% restoration of aggregate function whereas the sham control group remained stable at 91.3 ± 5.4% (Fig. 2C, n = 5 stimulated, n = 6 sham control, p = 0.48); however, in week 7, sham controls demonstrated an aggregate functional recovery of 63.0 ± 5.2% whereas stimulated animals returned to 96.6 ± 4.9% of baseline aggregate functional score (Fig. 2C, n = 5 stimulated, n = 6 sham control, p = 0.000027), and in week 8 sham control returned to 75.6 ± 6.8% and stimulated animals to 93.4 ± 3.2% of baseline levels (Fig. 2C, n = 5 stimulated, n = 6 sham control, p = 0.031). In a review of individual movements during reaching, we noted that C6 stimulation produced statistically significant improvements in limb lift, digits open, advance and aim (Fig. 3).

FIG. 2.

VSS improves skilled reaching in a stimulation-independent manner. (A) Experimental timeline, stimulation protocol and outcome measures. (B) Visual representation of the pellet grasp phase of the skilled forelimb reaching task (C) Animals receiving VSS stimulation significantly outperform sham stimulation controls in analysis of a combined average subject score that includes all 11 functional subscores represented as mean ± standard error (p < 0.05, n = 6 stimulated, n = 5 control). VSS, ventral spinal stimulation.

FIG. 3.

VSS Improves Multiple Phases of Skilled Reaching. Individual subscores for each component of a successful reaching motion represented as mean ± standard error (p < 0.05, n >= 20) accompanied by a representative image of each component of a successful reach. Early phases of recovery are statistically driven by enhanced opening and closure of the digits, while later phases of recovery are punctuated by improvements in limb advancement and aim. VSS, ventral spinal stimulation.

VSS drives increased neuronal but decreased microglia Nr4a1 mRNA

Nr4a1 is an immediate early gene involved in cell plasticity and known to be regulated by acute changes in cell activity.⁴⁸ We sought to map subacute and chronic Nr4a1 as a tool to probe how VSS changes spinal cord plasticity. We compared the expression of Nr4a1 in animals receiving a C4 hemi-contusion with or without epidural spinal stimulation at C6 (Fig. 4). Specifically, we visualized Nr4a1 probes by RNAscope with a separate subacute cohort following 1 week of stimulation after SCI (n = 8 per group) or in the functional cohort following 8 weeks of stimulation at C6 (n = 5 stimulated, n = 6 sham control). Comparing sham control and stimulated animals, we observed a significant increase in Nr4a1 RNA in neurons ipsilateral to the lesion on the side of the spinal cord most proximal to the stimulator leads following 1 week of VSS where the normalized Nr4a1 expression was determined to be 0.31 ± 0.091 for the stimulated animals compared to 0.16 ± 0.081 in the unstimulated control (Fig. 4C, n = 8 stimulated, n = 8 sham control, p = 0.024). No such effect was observed on the contralateral side (p = 0.56). These data indicate that neuronal Nr4a1 expression is highly sensitive to activity changes in the injury setting. In contrast to the expression of Nr4a1 in neurons, we detected a significant reduction in Nr4a1 mRNA expression in microglia ipsilateral to the lesion on the stimulated side of the spinal cord following 1 week of VSS where the normalized Nr4a1 expression was determined to be 0.22 ± 0.082 for the stimulated animals compared to 0.40 ± 0.098 in the unstimulated control (Fig. 4C, n = 8 stimulated, n = 8 sham control, p = 0.0053). No such effect was observed on the contralateral side (p = 0.054). At 8 weeks of stimulation, there was no difference in Nr4a1 expression in microglia and neurons between the stimulated and non-stimulated groups (Fig. 4D). These data imply that microglial and neuronal Nr4a1 expression returns to basal levels during prolonged VSS (i.e., 8 weeks of VSS) (Fig. 4D). Our findings indicate that during the subacute injury period, VSS facilitates neuronal activity-dependent Nr4a1 expression. We hypothesized that the 1 week, VSS-dependent decrease in microglial Nr4a1 expression could indicate a change in microglial activity or phenotype, which we sought to confirm by assessing microglial morphology.

FIG. 4.

Ventral spinal stimulation (VSS) leads to an early increase in Nr4a1 expression in neurons and a decrease in Nr4a1 in microglia acutely after SCI. Regions of interest (ROIs) occupying the ventral horn gray matter of stimulated C6 segment were quantified for the expression of Nr4a1 in Neuronal Nuclei (NeuN)-labeled neurons and ionized calcium-binding adapter molecule 1 (Iba1)-labeled microglia. (A) Microglial staining and segmentation co-registered with Nr4a1 RNAscope labeling and segmentation; scale bars = 10 μm i. Iba1 + cell (magenta) with 4′,6-diamidino-2-phenylindole (DAPI) (blue) and Nr4a1 RNAscope staining (red) ii. Iba1 segmentation iii. Nr4a1 segmentation overlapping Iba1 segmentation (cyan). iv. Overlay of iii. on original staining. (B) Neuronal soma staining and segmentation co-registered with Nr4a1 RNAscope labeling and segmentation; scale bars = 20 μm i. NeuN + cell (green) with DAPI (blue) and Nr4a1 RNAscope staining (red) ii. NeuN segmentation iii. Nr4a1 segmentation overlapping Neun segmentation (cyan). iv. Overlay of iii. On original staining. (C) Quantification of Nr4a1 RNA in neurons demonstrates that stimulation increased the amount of Nr4a1 expression in neurons and decreased the amount of Nr4a1 expression (dots·μm³/cell·μm³) in microglia/macrophages ipsilateral to stimulation at 1-week post-stimulation when compared with sham-control SCI-injured rats (*p < 0.05, n = 8 per group). (D) The decrease in microglial and increase in neuronal Nr4a1 expression observed after 1 week of stimulation was transient and returned to control levels at 8 weeks post-stimulation. (*p < 0.05, n ≥ 4 per group).

Microglia distribution and shape after VSS

Microglial morphology has proven itself to be a significant predictor of activation state.⁴⁹ To define microglial morphology, we imaged microglia expressing Iba1 and applied unbiased, single-cell reconstruction using Neurolucida (MicroBrightfield, Inc) at 8 weeks post-stimulation. We segmented hundreds of microglia in triplicate for each subject from coronal spinal cord sections. Specifically, we placed a region of interest within the C6 ventral horn gray matter of the site of stimulation or contralateral to the electrode side (Fig. 5A). Iba1 reactive microglia appeared to have similar distributions surrounding neural elements (Map2) and motor neurons (ChAT, Fig. 5B). At low magnification, stimulated spinal cord microglia appeared to have morphologies that are more ramified compared to sham controls (Fig. 5C,D). Confocal images of typical microglia morphologies indicate that microglia in the unstimulated spinal cord tend to have ameboid cell bodies with short, unbranched processes (Fig. 5D). A predominant feature of microglia in gray matter at the site of stimulation is of a ramified morphotype with long, branching processes (Fig. 5D). These data indicate that early changes in microglial plasticity, resulting from epidural stimulation, manifest in altered cellular morphology.

FIG. 5.

Ventral spinal stimulation (VSS) increases morphological complexity of microglia at 8 weeks post-stimulation. (A) Epifluorescent tiled images (10×) at C6 spinal level of sham (left) and stimulated (right) spinal cords. Region of interest (ROI) for high-resolution examination of microglial morphology (white box). (B) Individual channels probed for DAPI, ChAT (motor neuron marker), Map2 (neuronal marker), and Iba1 (microglia/macrophage marker), 20 um thick confocal stack imaged with 40× oil objective. (C) Representative ROI of Iba1+ cells from the hemisphere that is ipsilateral to the injury/stimulator. (D) i. Representative sham and stimulated microglia/macrophages labeled for DAPI (blue) and Iba1 (red) showing a reduction in the caliber of unstimulated microglia processes (arrows) and an increase in fine, complex processes in stimulated microglia. ii. Iba1+ cells segmented using Neurolucida. iii. Segmented Iba1 signal with a pixel count map. iv. Labeled and segmented microglia. All scale bars are 20 μm. (E) Example of stimulated and unstimulated Iba1+ cells with orthogonal views centered on the white cross hair.

VSS drives a morphological transformation of microglia

To quantify the changes in microglial morphology, we reconstructed gray matter microglia using Neurolucida. Three-dimensional microglial reconstructions generated in Neurolucida360 were interrogated in Neurolucida Explorer to extract morphological features of each microglia (Fig. 6C). We recorded several morphological parameters (e.g., number of intersections or arbors crossing at each spherically increasing distance from the microglia soma, number of terminals and number of nodes) to infer changes in microglial morphology associated with spinal stimulation after spinal cord injury (Fig. 6A,C,D, Supplementary Fig. S4). Microglia present in the injured rats receiving stimulation displayed significantly longer arbors measuring 167.4 ± 13.8 μm compared to those seen in the unstimulated sham control group whose dendrites measured an average of 125.8 ± 19.9 μm (Fig. 6A, n = 6 sham control, n = 5 stimulated, p = 0.0061). We observed no significant changes in dendrite thickness as the dendrite volume was observed to be 345.7 ± 19.9 μm³ for stimulated rats and 266.5 ± 176.9 μm³ for the sham control group (n = 6 sham control, n = 5 stimulated, p = 0.42). There was an increased number of terminals per microglia in stimulated animals averaging 10.68 ± 0.54 terminals/microglia and 8.10 ± 1.35 in sham control (Fig. 6A, n = 6 sham control, n = 5 stimulated, p = 0.0062). Microglia covered a larger region following spinal stimulation indicated by an increased convex hull volume with stimulated animals covering a territory of 5470 ± 940 μm³, whereas sham control animals covered an average territory of 3535 ± 845 μm³ (Fig. 6A, n = 6 sham control, n = 5 stimulated, p = 0.010). Reconstructed microglia were visibly more complex in stimulated animals (Fig. 6B, Supplementary Fig. S5) and significantly more complex as determined by the formula:

$$\begin{gathered} \mathit{\text{Complexity}}=[\mathit{\text{Sum of the terminal orders}}+\mathit{\text{Number of terminals}}] \\ *[\mathit{\text{Total dendritic length}}/\mathit{\text{Number of primary dendrites}}] \end{gathered}$$

FIG. 6.

Spinal stimulation restores microglial complexity and reduces the number of microglia with an activated morphology at the site of stimulation. (A) Quantification of significantly altered features of microglial segmentations ipsilateral to the lesion in stimulated and sham control C6 spinal sections. VSS significantly increases dendrite length, number of terminals, microglial territory (convex hull volume), and dendrite complexity (*p < 0.05, n = 6 stimulated, n = 5 control). (B) Principal component (PC) analysis of 16 extracted microglial features between stimulated and sham-control microglia demonstrates a clear bias along the PC1 axis. Representative branch reconstructions of individual microglia along the PC1 axis (below) along with histological examples of microglia in each canonical state of activation demonstrate features of reduced activation with an increase in PC1 values. Quantification of the difference in average PC1/PC2 location for microglia from stimulated and sham control animals reveals a positive shift along the first principal component following spinal stimulation (right), which correlates with a morphological trend toward increased complexity and morphological indications of reduced activation. (C) Representative microglia with overlaid skeleton segmentation utilized for Sholl analysis. (D) Sholl analysis of microglial intersections, nodes, and terminals at 3 μm radial increments. Graphs are represented as means with a 95% confidence interval (*p < 0.05, n = 465 microglia per group). VSS, ventral spinal stimulation.

Microglia from stimulated animals averaged a complexity value of 12334.1 ± 2624 versus 7063.8 ± 2957 in sham control animals (Fig. 6A, n = 6 sham control, n = 5 stimulated, p = 0.021). There was no significant change in these features on the contralesional side (Supplementary Fig. S4). A principal component analysis of 16 morphological features of segmented microglia (Table 1) revealed that the first principal component was the primary axis along which microglia were generally distributed with an explained variance ratio of 0.969 (Fig. 6B). The variance along this axis was primarily explained by convex hull volume (μm³) and dendrite complexity. Random sampling of microglia along the axis demonstrated a trend from less complex to more complex morphology from left to right indicating this axis may be correlated with activation state. Stimulation induced a left-right shift along this axis with stimulated microglia biasing along the first principal component with an average position of 0.83 ± 0.38, whereas microglia from unstimulated sham control animals averaged a position of −0.45 ± 0.61 (Fig. 6B, n = 6 sham control, n = 5 stimulated, p = 0.0050).

Table 1.

Microglia Features Measured in Stimulated and Sham Control Conditions Ipsilateral to the Spinal Cord Injury and Stimulator

Soma	Dendrites	Terminals	Nodes	Territory
• Soma volume (μm3) • Soma surface (μm2)	• Dendrite length (μm) • Dendrite surface (μm2) • Dendrite volume (μm3) • Dendrite complexity	• Number of terminals • Average (AVG) terminal (μm) • Standard devation (SD) Terminal distance (μm)	• Number of nodes • AVG node distance (μm) – Linear or along dendrite • SD node distance (μm) – linear or along dendrite	• Convex hull surface (μm2) • Convex hull volume (μm3)

Discussion

This work establishes that a novel, ventral epidural stimulation approach produces significant, rapid enhancement of skilled motor function in rodents with a cervical spinal cord injury that remains present at least 4 h after the cessation of therapeutic stimulation. Previous studies have demonstrated a stimulation-dependent effect where there is a temporary enhancement in skilled motor function which eventually washes out.³⁶ Persistent recovery of upper limb function following spinal stimulation which is stable for weeks following stimulation halt has been reported in rats³⁸ and humans,⁵⁰ but only when combined with rehabilitation during the stimulation phase. Here, we demonstrate stimulation-induced persistent recovery where positive impacts of stimulation remain stable for at least 4 h and are not associated with targeted rehabilitation. These data also push the potential therapeutic window to as early as 7 days after SCI. We utilized a versatile, self-contained wireless spinal stimulation device that will allow future work to examine common or additive advantages of multi-site spinal cord stimulation to deliver enhanced motor recovery after spinal cord injury. We also quantified the expression of an immediate early gene, Nr4a1, known to play a role in activity-dependent plasticity, as a harbinger of neuronal and glial plasticity.^48,51–53 Together, our research provides a strong rationale for future exploration of site of stimulation and immediate early gene expression analysis to illuminate new applications and mechanisms of epidural stimulation for the restoration of sensorimotor function in the treatment of paralysis.

Targeting select muscle groups and motor pools within the spine has been an area of emphasis for therapeutic neuromodulation. Prior cervical stimulation approaches include epidural, intraspinal, optogenetic and transcutaneous stimulation. In each approach, relative specificity of the targeted movements and muscles contrasts with level of invasiveness and clinical feasibility. We previously reported a higher degree of specificity elicited from the cervical ventral spinal surface in rats when compared with transcutaneous or dorsal epidural approaches.⁴³ Additionally, a ventral lumbar approach generated more biomimetic lower limb movements.⁵⁴ Physiologically, we also observed a preferential targeting of the triceps, wrist flexors and infraspinatus muscles from the ventral surface of the C6 spinal level, which corresponded to improvements in limb lift and advance as well as digit extension (Figs. 1C and 3). While intraspinal stimulation corresponds to a higher degree of specificity and has previously been reported to induce stimulation-independent persistent motor functional recovery, intraspinal approaches require highly invasive and percutaneous wire placement. Conversely, non-invasive stimulation strategies such as transcutaneous spinal stimulation are highly translational and minimally invasive; yet, selectivity, particularly in cervical regions, is generally lower than what is seen in more invasive methods where the position of the electrode and current density within the spinal cord can be better controlled.³⁵ Our data suggest a ventral approach as an intermediate between the relative invasiveness/specificity of intraspinal approaches and transcutaneous or epidural dorsal approaches given similarities to functional outcomes present with intraspinal stimulation in this model and in our previous work where we demonstrated selectivity of motor activation with a ventral approach and wireless stimulator in a rat model of spinal cord injury.^30,36,43,50

With a ventral approach targeting the C6 spinal level, we observed > 15% improvement of a skilled forelimb reaching, even when the task was decoupled temporally by 4 h from the ventral stimulation. Multiple aspects of the reaching behavior were significantly improved over an 8-week period, indicating the therapeutic potential of a ventral approach for targeted neuromodulation of the upper limb. While Edgerton et al. observed post-stimulation functional improvements from dorsal cervical neuromodulation, those improvements were reduced following a 24 h washout period.³⁶ In our studies, a 4-h washout may be insufficient to completely abrogate stimulation-dependent effects on associated motor circuitry. Further, the circuits targeted via a ventral approach may be unique to those recruited with other stimulation methods. Indeed, we previously reported physiological evidence of motor output signatures elicited from stimulation of the ventral epidural surface (e.g., higher motor pool specificity).⁴³ Still, the 4-h washout period was a minimum requirement, extending as long as 6 h. We observed no correlation with time from stimulation to relative functional performance. Current evidence therefore suggests stable changes to the circuits impacted by ventral epidural stimulation; however, more studies are required to confirm the long-term stability of this recovery.

Mechanism of neuromodulation induced recovery

The mechanisms behind neuromodulation mediated recovery after spinal cord injury remain opaque. A better understanding of said mechanisms will serve to boost the utility and application of neuromodulation for recovery of neurological injury. Previous works have correlated functional and morphological outcomes of various neuromodulation methods with neuronal branching, angiogenesis, sprouting and myelin remodeling.^55–59 Still, gain and loss of function studies have yet to be carried out, and the full mechanistic landscape remains unclear. In the current work, we sought to add to the early identification of potential mediators of epidural stimulation. In an early genetic screen, we identified Nr4a1 as a transcriptional responder to VSS (data not shown). We therefore decided to explore Nr4a1 expression in cell subtypes relevant to function and recovery after spinal cord injury. Aberrant signaling and translocation of transcripts of various orphan receptors like Nr4a1 have been identified as mediators of neuronal cell death in neurodegenerative disorders of the spinal cord^52,60 and as critical regulators of immune responses through signaling in microglia and macrophage after insult.^53,61–64 Perhaps most exciting and unique was our finding of a short-term repression of Nr4a1 mRNA in microglia after 1 week of VSS (Fig. 4C). We were able to correlate this early change in Nr4a1 in stimulated microglia with a chronic shift in microglial morphology where ipsilesional stimulated microglia displayed morphological features consistent with reduced activation compared to unstimulated sham control microglia (Fig. 6).^51,53,63,64 Nr4a1 expression in microglia has been described as a key mediator of microglial and monocyte activation level⁶² and complete knockout of the Nr4a1 receptor in microglia results in a significantly higher activation state.⁵³ Still, it is unclear whether an early shift in Nr4a1 expression in microglia is a direct effector of long-term microglial activation state. Given that N4a1 levels in microglia were observed to be unchanged following 8 weeks of electrical stimulation compared to sham control microglia (Fig. 4D), the impacts of Nr4a1 transcription may be more relevant in the subacute period. Stimulation was linked with reduced Nr4a1 expression in microglia following 1 week of stimulation ipsilateral to the lesion and stimulator but remained unchanged on the contralateral side (Fig. 4C). Using a Sholl analysis, microglia were observed to have a significantly increased number of intersections contralateral to stimulation (Fig. 6D); however, key morphological feature changes in microglia on the stimulated, ipsilesional side (Fig. 6A) were not observed to be significantly different contralateral to stimulation (Supplementary Fig. S4). It may be that current targeted to the ipsilesional side is still sufficient to impact microglia contralateral to stimulation given the morphological shift on the contralateral side.⁶² More work remains to be done to elucidate the mechanisms behind the impact of electric fields and Nr4a1 transcription on microglia. Nr4a1 transcription was also altered in stimulated neurons following 1 week of stimulation (Fig. 4C). Although Nr4a1 activity in neurons is associated with synaptic rearrangement,^51,52 we found no gross changes to neuronal morphology (Supplementary Fig. S6). Nevertheless, in the future, performing gain- or loss-of-function in neurons during stimulation may help uncover the morphological and functional impact of this activity-dependent transcription pathway.

Additionally, we observed several significant changes to microglial morphology, indicating a shift from activated to a more resting or ramified morphology induced by VSS. According to the literature, microglia can morphologically respond to electrical fields by entering a resting state.^65–68 Activated microglia, in particular, are responsive to the effects of an electric field.⁶⁷ Activated microglia are reported to engulf and destabilize synapses.^52,69 Following VSS, microglia exhibited smaller somas and covered larger territories, which is consistent with a resting state.^49,70–72 Further, Sholl analysis revealed more intersections radial distances from 9 to 24 μm, suggesting that stimulation modulates microglial morphology towards a morphotype consistent with reduced activation state.^73–75 A recent study demonstrated that ISMS targeting the ventral horn reduced nociception in a thoracic model of spinal cord injury in rats.⁷⁶ Allodynia is not a feature of cervical lesions of the rat, so we were not able to test whether microglial activation state and nociception are linked.⁷⁷ Altogether, our observations suggest that Nr4a1 may be a mediator of microglial phenotypic changes induced by electrical stimulation. Future studies should focus on a broader gene expression profile of glia as well as downstream targets of Nr4a1 to better understand how electrical activity is translated into a morphological shift in microglial activation state. Given that early changes in Nr4a1 expression were linked with reduced microglial activation following VSS, we sought to examine changes in motor neuron morphology as neurons also exhibited increases in Nr4a1 expression within the ventral horn in response to ventral stimulation (Figs. 4 and 6 and Supplementary Fig. S6). While gross analysis of motor neuron morphology did not reveal significant differences between treatment groups, soma size may be an insufficient metric to infer the subacute impacts of Nr4a1 expression in neurons after spinal cord injury.

Target of VSS

Our data is consistent with the idea that the primary neural target of ventral epidural stimulation is the motor neuron via antidromic activation through the ventral root. However, it is possible that cells in the ventral horn are directly recruited at the applied therapeutic currents. Indeed, other researchers have demonstrated recruitment of cells in the ventral horn of the lumbar enlargement with similar levels of current.⁷⁸ Still, given the relatively low transverse conductivity of white matter, the physical distance of ventral gray matter to the electrodes and the relatively small soma size of ventral interneurons in the cervical spinal cord, direct activation of a significant number of other neural circuits from the ventral aspect seems unlikely; though, more work is needed to confirm whether cells in the ventral horn are depolarized at the currents used.⁷⁹ We therefore examined the motor neurons for morphological indicators of health and connectivity. While observed features were sensitive to the injury, there were no differences associated with neuromodulation (Supplementary Fig. S6). More research is needed to uncover the full impact of the electric field on motor neuron populations. For example, does VSS drive changes in synaptic number, type, or strength? Or could VSS change overall motor neuron excitability, allowing for improved access from spared descending excitatory pathways? What is the relation to altered microglial state and the structure/function of local motor neuron populations? Further, beyond the motor neuron itself, it will be important to characterize the exact populations of neurons recruited from the ventral surface, including ascending sensory systems, descending neuromodulators, and local inter- or propriospinal systems.³¹ Surprisingly, few deep investigations into the mechanisms of epidural stimulation have been performed despite the potential value of such data.⁸⁰

Previously, we have shown that VSS allows the activation of muscles that are distinct from dorsal spinal stimulation (DSS) when applied in an injury model.⁴³ However, our experiments do not directly confirm the advantage of a ventral approach over a dorsal approach in terms of motor function. Future questions need to explore the distinct advantages that discrete sites of stimulation or combinations of sites may have on function and how the timing and therapeutic regimen in ours and other previously published protocols will have overlapping, distinct or combinatorial advantages. Of note, we pushed the therapeutic window to a subacute time not typically explored, and it is possible that dorsal stimulation within this early timepoint could have equally potent sensorimotor benefits. What’s more, given the subacute intervention applied, findings may not be generalizable to the chronic spinal cord injury condition. Further, we used ventral Nr4a1 expression in microglia and neurons as a sentinel for changes in cell activity, but we do not provide an empirical connection between Nr4a1 expression specifically and the microglial cell morphological transformation. Despite these limitations, Nr4a1 has been reported to be a mediator of microglial activation, and the findings argue for exploration of Nr4a1 function in an injury setting.⁵³ Altogether, we believe our data are the first to show persistent stimulation-induced recovery of upper limb function, an exciting pre-clinical proof-in-principle.

While we do observe persistent recovery where improved performance is observed at least 4 h after therapeutic stimulation, the washout period for stimulation-dependent improvement of skilled motor function has not been definitively established. All histological analysis was focused at the C6 spinal segment where stimulation was delivered. The C4 spinal tissues at the level of injury were damaged beyond use during processing, and so the impacts of VSS on the injury zone cannot be assessed for the longitudinal study. We were therefore unable to correlate VSS stimulation two segments caudal to the lesion with any morphological changes to the injury zone. Also, we do not directly correlate subscoring improvement to the optimally targeted muscle groups at motor threshold for each subject. It may be that electrode position shifts and initial placement account for some of the variation observed in the recovery of the measured subsets of reaching (Fig. 3).

Electrical stimulation shows exciting potential to reverse some of the devastating effects of spinal cord injury. In the case of injuries to the cervical spinal cord, our data show that a ventral epidural approach has the potential to enhance skilled motor function that is persistent for 4 h after stimulation cessation. Our data further add to the limited but growing set of observations on how electrical stimulation elicits changes to the spinal cord. Specifically, we show that a stimulation-induced reduction in microglial activation may play a role in the mechanism of recovery. A better understanding of how electrode position and changes to the underlying substrate mediate recovery of function is needed. While much research remains to empirically establish the mechanisms of neuromodulation, application of electrical current to the ventral surface shows great promise to promote recovery of function after cervical spinal cord injury.

Materials and Methods

Experimental design

The primary objective of the study is to assess whether ventral cervical spinal neuromodulation provides therapeutic motor benefit in a cervical hemicontusion model of spinal cord injury. We pre-trained rats on a skilled-reaching behavioral task prior to study initialization until skill plateau and baseline function were assessed for each rat over the ensuing 3 weeks. At study onset, we applied a cervical C4 moderate hemicontusion to all animals in both the stimulated and sham control groups. Sham control groups are defined as animals that receive a spinal cord injury as well as an electrode and stimulator implant but not electrical stimulation. At 1 week after SCI, we implanted a stimulator at the C6 level for all animals in both stimulated and sham control groups. Subjects in the stimulated group received therapeutic stimulation applied daily for a period of 8 weeks while sham control subjects received no stimulation. We also assessed forelimb function during those 8 weeks. At study termination, rats were perfused, and tissues were prepared for histological examination. Experimenters were blinded for all analyses.

Stimulator design

We previously created a wireless stimulation module in collaboration with Indus inc. designed to be a flexible platform for multi-site electrical stimulation of the rodent spinal cord.⁴³ To accomplish the current experiments, we performed iterative design improvements to our original stimulator design.⁴³ Firstly, we incorporated a larger battery (45 milliamp-hour [mAh] vs. 22 mAh) and increased the overall form factor to ensure a redundant power supply and allow flexibility for higher power draw stimulation protocols. To accommodate the larger volume, the Printed Circuit Board (PCB) and components were re-designed to fit in a circular shape, thereby maximizing the inductive charge receive coil size per device footprint. In addition, the device was (1) rounded to reduce irritation to the animal⁸¹ as well as to fit the new copper receive coil optimally. The device was then (2) connectorized with eight wire coils fabricated from 7-strand A-M Systems stainless steel (A-M Systems 793500). Finally, the device was (c) encapsulated in poly-ether-ether ketone (PEEK) as before.⁴³ The wires were externalized from the device prior to complete filling with UV-cured epoxy to ensure device stability and reduce potential damage to implanted electronics. Further, the externalized wires can be connected to an electrode system or be used as electrophysiology wires independently, allowing for device flexibility while maintaining long-term durability. All devices manufactured to said specifications have maintained functionality for 2 years and were reconditioned as needed through 30 min in a sonicating bath followed by submersion in Oakton Electrode cleaner. Devices were sterilized for 10 min in an ethanol bath followed by UV irradiation prior to implantation.

Subjects

Six-week-old female Long-Evans rats were purchased from Charles River and housed at the Houston Methodist Research Institute animal vivarium in a 12-hour light and dark cycle. Animals were double housed and provided access to food and water ad libitum. All procedures were approved by the Houston Methodist Institutional Animal Care and Use Committee under the issued protocol IS00006709 and IS00008648 and performed in accordance with guidelines from the National Institute of Health regarding the care and use of laboratory animals. For the forelimb reaching task, to detect an effect size of 10% with a power of 80% and a p < 0.05 cutoff, $\left[z_{1-\beta }=0.84,z_{1-\alpha }=1.96\right]$, we estimated n = 8 animals per group. As this was the lowest presumed effect size, we determined a minimum of 8 animals per group for all studies. For the 8-week stimulation study, we initially assigned 10 animals per group. 20% were non-learners and excluded from the study. Additionally, 3 animals, 1 from the sham control and 2 from the stimulation group, were lost due to surgical complication post-group assignment yielding our final group numbers of n = 6 sham control and n = 5 stimulation.

Behavior training and data capture

Rats were acclimated to behavior cages (Fig. 2B) for one week prior to the onset of training as previously described.⁴⁵ Rats were then trained daily to grasp a chocolate pellet during the dark cycle in 10-min session intervals. Over the first two weeks, rats were allowed to freely grasp using either paw. The displayed paw preference was tracked for each animal. Once the dominant paw was established, a plexiglass divider was inserted to limit the grasp to the preferred paw. Rats were trained until a cohort average success rate of > 50% was established and individual rat success rates had plateaued. A minimum of 50% cohort success rate was previously demonstrated to be a sufficient baseline for forelimb reaching assessment after a mild cervical spinal cord injury.⁴⁵ Rats were determined to have plateaued if they did not exhibit increased performance over a 3-trial period. Daily training was halted, and rats were assessed weekly. Attempts were captured using two 120 fps CV508 cameras (Marshall Electronics Inc.) positioned perpendicularly to the rat, and video was obtained on a capture card (Magewell Inc, cat. 11080). Baseline attempts were gathered over a three-week period prior to study onset. Rats were given five warm-up attempts prior to scoring assessment, and each scored session consisted of ten grasps followed by a complete reset. A reset was defined as all paws returning to the ground followed by a 360-degree rotation prior to the next scored attempt. Baseline behavior scores were established as an average success rate over three sessions. Following the onset of therapeutic stimulation, rats were assessed at least four hours after a stimulation session on a weekly basis. Stimulation occurred at the onset of the light cycle change (10 am) for 1 h and all subsequent behavioral testing took place from 3 to 5pm, allowing for a minimum of 4 h between the cessation of therapeutic stimulus and the beginning of behavioral testing.

Behavior scoring and inter-rater reliability

All scoring was performed in a blinded fashion by seven individuals using a modified subscoring assessment outlined by Wishaw et al.⁸² Scorers were subjected to a training workshop prior to scoring sessions per our previously published protocol to ensure low subjective variability in assessment.⁴⁵ Scores were collected using a custom behavioral code map in Behavioral Observation Research Interactive Software. Collected scores were compared using a Kappa Cohen Inter-Rater Reliability assessment to ensure good agreement between raters.

Spinal cord injury

Spinal contusion injuries were conducted using a custom, electromechanical device (Ohio State University [OSU] Impactor) as previously described.⁸³ A C4 laminectomy was performed, and the dorsal surface of the spinal cord was exposed and cleaned. The animal was suspended from a stereotactic mounting frame by the dorsal and ventral lamina to the laminectomy site. The animal was given a 0.8 mm contusion by the OSU impactor device,⁸³ and data were collected using a data acquisition board (DAQ) in LabView software. Animals were divided semi-randomly into groups with matched output measures to balance groups. All animals presented in this study received a C4 hemicontusion injury.

VSS electrode implantation

Insertion of the electrode along the ventral spinal cord was achieved through a technically challenging procedure designed to prevent damage to the dorsal roots of the cervical spinal cord and has been previously detailed.⁴³ Briefly, the T4/T5 laminae were exposed, and a partial lateralized laminectomy was performed on a caudal portion of T4 and a rostral portion of T5 on the contralateral-handedness side. Sufficient bone was removed such that the lateral surface of the spinal cord dural surface could be visualized. The electrode was inserted between the T4 and T5 spinal roots with the electrodes facing the spinal surface (Fig. 1). The electrode was marked intermittently to allow for measurement of the traversed insertion distance to allow for approximate placement of the electrode tip at C6 on the ipsilesional side. Electrode placement is confirmed by visual and tactile assessment of target musculature stimulation or by EMG. Contralateral muscles were examined to ensure unilateral placement of the electrodes at the time of implantation. Electrodes were placed to optimally target infraspinatus, triceps, and wrist flexors (Fig. 1D). All animals in this study received an implant; however, sham control animals received dummy electrodes (i.e., missing internal electronic components) weighted to match functional stimulators.

Subacute measurements of C6 spinal evoked potentials from the ventral epidural surface

Rats received a C4 hemicontusion injury and recovered for one week prior to terminal electrophysiological mapping of the cervical spine from the ventral surface. Rats were anesthetized under ketamine/xylazine (80 and 20 mg/kg, respectively) for all electrophysiology experiments. The skin of the targeted forelimb was removed to accurately visualize the boundaries of the upper limb muscles. Subdermal 13 mm paired needles (Rhythmlink) were placed in pairs along fibers of 7 muscles in the injured forelimb and secured with cyanoacrylate. No electrodes were placed in the contralateral limb as we never observed muscle twitching. Muscles included: infraspinatus, trapezius, biceps, triceps, wrist flexors, wrist extensors, and flexor/extensor digitorum (paw). Subdermal needles were connected to a Tucker Davis Technology (TDT) PZ5–32 Neurodigitizer amplifier, and recruitment curves were generated at the C6 spinal level and measured using an RZ5D BioAmp Processor. A mapping electrode probe was created by 3D printing 2.85 mm polylactic acid (PLA) filament (Tough PLA, Ultimaker) to ~100 μm thickness with axial geometries of 0.9 mm in width and 80 mm in length. The same physiology wire used in electrode construction was threaded onto the PLA probe, and 0.2 mm of paraformaldehyde (PFA) coating was stripped using a scalpel blade under a stereoscope (Nikon SMZ800N) to expose the conducting tips. Anode and cathode wires were secured along the surface of the PLA probe using cyanoacrylate with 2 mm longitudinal spacing to ensure uniform separation of anode and cathode during mapping experiments. Opposing ends of each wire were stripped and connected to an Analog Stimulus Isolator Model 2200 (A-M Systems). The stimulating PLA probe was inserted into the ventral epidural space as outlined above and threaded to the C6 spinal level as previously described.⁴³ Current was delivered as a biphasic pulse with 300 μs pulse width and a 60 μs inter-stimulus interval as a constant current impulse to match the VSS stimulation protocol. Recruitment curves were generated by increasing the stimulation in 20 μA intervals from 20 to 200 μA to assess typical muscle activation at the C6 spinal level. Evoked muscle potentials were analyzed in MATLAB and Python as previously described⁴³ (Fig. 1C).

VSS stimulation protocol

Each animal received a spinal cord injury and was implanted with a spinal stimulator. Stimulated groups received regular spinal stimulation, while sham control animals received no spinal stimulation. Stimulators for sham control animals were identical in form and weight to those of the stimulated animals; yet they lacked internal electronic components. Current was delivered as a biphasic pulse with 300 μs pulse width and a 60 μs inter-stimulus interval as a constant current impulse. Muscle activation threshold was established daily during the first week and once per week after the first week of stimulation as thresholds stabilized over time. Thresholds were established by determining the minimum current of activation of the muscles of the forelimb. The implanted stimulator was adjusted to establish the level of electrical stimulation below the level to initiate muscle twitch. Penetrating EMG electrodes were not used to establish thresholds for longitudinal studies to minimize potential damage or interference with the muscles of the targeted limb. The muscles of the arm were palpated, and stimulation current was increased in 10 μA increments until a muscle twitch was observed by touch. Therapeutic stimulation levels were set to 80% of the minimum current required to initiate a muscle twitch. Stimulus was applied pseudo-randomly at a frequency of 20 Hz with up to a 5% variation per pulse, averaging to the 20 Hz mean. Stimulation was applied for 1 hour per 24-h cycle beginning at light-cycle change at the onset of the night cycle with a 10 min on 5 min off interval. Thresholds were interrogated weekly through the course of the experiment and adjusted to 80% muscle activation threshold. Stimulation thresholds ranged from 80–500 μA over the 8-week study.

Tissue preservation and preparation

Animals were pericardially perfused on a down-draft perfusion table with heparinized phosphate-buffered saline (PBS) followed by 4% PFA. Spinal cords were extracted and post-fixed overnight, then cryoprotected using 10%, 20%, and 30% sucrose solutions. Spinal cords were embedded in Tissue-Tek OCT compound (Sakura, Holland) and snap-frozen in liquid nitrogen. Samples were sectioned axially with a 20 μm thickness on a Cryostar NX50 cryomicrotome (ThermoFisher) and mounted on positively charged slides for staining. Tissues were sectioned in 1:6 series with 3 sections per slide, each 180 μm apart. Depth-matched sections were always compared for all imaging and analysis.

Histology

Sections were washed 3× in PBS solution under agitation for 15 min and then blocked in 10% donkey serum with 0.1%-0.2% Triton X-100 with agitation for 1–2 h. The block was removed, and primary antibodies were added to the sections that were then incubated with gentle agitation at 4°C overnight. Primary antibody was removed, and the sections were again washed in PBS 3× or 15 min each, followed by the addition of secondary antibody. Then, they were incubated at room temperature with gentle agitation for 1 h. Secondary antibody was removed, and sections were washed in PBS once; then DAPI was added for 5 min at a concentration of 1:1000, then removed, and sections were washed 3 × 15 min PBS each with agitation. Slides were dried, and cover-slips were mounted with Fluoromount-G Mounting Medium (ThermoFisher Scientific) and allowed to cure overnight prior to imaging. Concentrations and catalog numbers for staining are listed in Table 2.

Table 2.

List of Antibodies Used for Immunohistochemistry

Species	Antigen	Concentration	Cat no.
Primary antibodies
Rabbit	Iba1	1:500	Wako 019-19741
Mouse	Map2 (2a + 2b)	1:500	Sigma M1406
Goat	ChAT	1:500	EMD AB144p
Rabbit	Synapsin1	1:500	Ab64581
Guinea pig	Homer1	1:100	Synaptic Systems 160004
Species	Antigen	Concentration	Type
Secondary antibodies
Donkey	Goat Ab	1:500	Alexa 647, 488
Donkey	Mouse Ab	1:500	Alexa 488
Donkey	Rabbit Ab	1:500	Alexa 568, 647
Donkey	Guinea Pig Ab	1:500	Alexa 647, 568

RNAscope method

Sections were stained for DAPI, NeuN antibody, Iba1 antibody, and Nr4a1 RNAscope probe (ACD Bio cat no. 414571-C3) using a modified protocol from ACD Bio and co-detection kit (ACD Bio 323180). Briefly, tissues were baked at 60°C in PBS for 30 min. The tissues were serially dehydrated and, prior to staining, 2–4 drops of hydrogen peroxide were incubated on each slide for 10 min followed by washing. Antigen retrieval was performed using co-detection target retrieval solution for 5 min at 100°C. Slides were immediately placed in a hot slide rack with distilled water. Primary antibodies were incubated on sections diluted in ACD Bio co-detection antibody diluent, followed by a 30 min post incubation fixation in 4% PFA. RNAscope hybridization was performed per the ACD Bio protocol, followed by secondary antibody stain in co-detection antibody diluent. Slides were washed and counterstained with DAPI and cover slipped with Fluoromount-G (Sigma-Aldrich).

Microscopy

All images were captured using a Leica DMi8 microscope with an epifluorescence and SPE-point scanning module. Images were captured using LAS X (Leica) software with the LAS Navigator extension for region selection and microscope experimental set-up. All images were exported as raw image files for analysis in Python or MATLAB (MathWorks). For microglial and synapse morphological analysis, all images were captured with a voxel size of 0.179 μm × 0.179 μm × 0.42 μm (X, Y, Z). All other z-stacks were captured with a voxel size of 0.179 μm × 0.179 μm × 1 μm. The injured/stimulated and uninjured/unstimulated ventral horn of the gray matter was captured for all sections in a rectangle of 684 μm wide by 364 μm tall, extending to just cover the bottom-most and lateral-most aspects of the ventral horn as visualized by 10× phase/DAPI imaging of the entire cross-section.

Quantification of Nr4a1

Each cross-section of spinal cord was imaged, quantified, and averaged in triplicate per subject. DAPI-stained nuclei were visualized in bilateral 20 μm thick, 684 μm wide by 364 μm tall areas in the ventral horn. Pyclesperanto and Napari were used to segment and label each nucleus by Otsu thresholding followed by Voronoi Otsu labeling with a spot sigma of 12. A scikit-learn neural network classifier was trained on over 600 labeled nuclei classified as Neurons, Microglia/Macrophage, or Other to an accuracy of > 95%. The trained classifier was applied to the identified cells in all visualized samples. Nr4a1 was thresholded using the Otsu method and subsequently labeled using a connected components algorithm with base settings. Nr4a1 “dots” were quantified within each nucleus using the Regionprops function from the scikit-image package. Relative Nr4a1 dot volume (dots·μm³/cell·μm³) was collected and presented as normalized Nr4a1 expression per cell (Fig. 4C, D). Signal intensity, dot number, and dot location were also collected, and intensity and dot number were each found to be proportional to normalized expression.

By applying RNAscope probes followed by counter-staining with DAPI in triplicate sections, we labeled Iba1 and NeuN antibodies as well as Nr4a1 transcripts. Cell nuclei were segmented in Napari using Pyclesperanto and scikit-image packages using a Threshold Otsu transformation, and Nr4a1 puncta were segmented using an N-Connected Components. Features related to Iba1 and NeuN localization within segmented DAPI were used to classify cell type and quantify the distribution of Nr4a1 puncta within each cell at 1 and 8 weeks of electrical stimulation after spinal cord injury. Data were collected and analyzed using Pandas.

Quantification of microglia

All imaged Iba1 + cells in the ventral gray matter were traced and analyzed in Neurolucida 360 (MBF Biosciences, Inc). Iba1 + cells with any component crossing outside of the visible field were excluded. Serial sections of Iba1 staining were loaded, and raw histograms were adjusted from 0–255 to 30–150 for all images to remove background and enhance signal for segmentation. Soma detection was performed automatically with a minimum soma size of 2 μm and a detection sensitivity of 70%. A blinded scorer removed mislabeled somas. Seeds, markers placed by the software to inform possible connecting components of a positively labeled cell, were automatically detected with 70% sensitivity and validated. Unconnected seeds were manually removed. Automated tree detection was performed using the Directional Kernels Method with base settings. Disconnected trees were manually removed, and unlabeled connected structures were semi-manually labeled with 80% sensitivity and a minimum dendrite width of 0.5 μm, as smaller dendrites could not reliably be extracted given optical resolution limits. Connected structures were grouped, and labels were exported to Neurolucida Explorer for analysis. A Dendrite-Sholl analysis was performed on the segmented sequences, and data were collated and analyzed using Python with the packages Seaborn, for statistical visualization, Pandas for multimodal data handling and processing, and Sklearn, for unbiased clustering and feature identification/extraction. A Sholl analysis was performed for each cell to assess the radial density of nodes (branches), endings (terminus), dendrite length (μm), and dendrite diameter (μm). From the reconstructed data, 16 total features were extracted per microglia and aggregated using Pandas to probe feature correlations with treatment groups and microglia locations.

Quantification of motor neurons

All imaged double-positive ChAT + and Map2 + cells in the ventral gray matter were traced and outlined in ImageJ. Somas were manually segmented in cross-section, and soma area was averaged per window. Segmented sequences and area data were collated and analyzed using Python with the packages Seaborn, for statistical visualization and Pandas, for multimodal data handling and processing. From the reconstructed data, soma area, intensity of Map2 signal, roundness of the soma, and solidity of the soma were extracted as features per motor neuron and aggregated using Pandas to probe feature correlations with treatment groups and motor neuron locations.

Statistical analysis

Significant differences between raw data sets for behavior were performed using one-way analysis of variance. Bonferroni correction was applied for subscoring in the behavioral data to account for 11 measured aspects of a reach. Subscores at each timepoint were collected with ≥ 20 observations per condition. Aggregate functional recovery is represented as a mean functional score where each subscore is equally weighted, and negative shift from baseline was presented as % functional loss.

$$\mathit{\text{Aggregate Score}}=\frac{\sum i_t-\sum i_b}{\sum i_b}$$

Where individual subscores: $i$ are a set of scores [0, 0.5, 1] averaged per subject per timepoint, over five trials. The sum of all averaged subscores $\mathrm{\Sigma }{i}_t$ minus the baseline sum $\mathrm{\Sigma }{i}_b$ over the baseline sum gives a representation of peper-subjectecovery to baseline over all subscores aggregated. For microglial morphology and Nr4a1 analyses, individual categorical differences were statistically analyzed via pair-wise $t$-test. A Bonferroni correction was applied where appropriate to account for 16 measured morphometric components. Imaging data were collected in triplicate per subject, and similar areas were analyzed. Areas where tissue was irregular/damaged, or staining was not present, were excluded. All analyses include at least two similar imaged areas per subject, per condition. Data are presented as means ± SD of the number (n) of observations. Two-tailed significance was determined using a p value of < 0.05. When designing all studies, we perform a power analysis for each presumed experiment using the formula outlined below.

$$N=\frac{{\sigma }^2{\left(z_{1-\beta }+z_{1-\alpha }\right)}^2}{E^2}$$

$N$ = predicted animal number per group to detect a mean population difference

$\alpha$ = Probability of a type 1 error

$\beta$ = Probability of a type II error

$\sigma$ = variance/standard deviation

$z$ = critical effect size given $\beta$ or $\alpha$

$E$ = Effect size (mean of treatment-mean control)

For the forelimb reaching task, to detect an effect size of 10% with a power of 80% and a p < 0.05 cutoff, $\left[z_{1-\beta }=0.84,z_{1-\alpha }=1.96\right]$, we estimated n = 8 animals per group. As this was the lowest presumed effect size, we initially assigned 10 animals per group. Groups for 8-week stimulation experiments were n = 6 sham control and n = 5 stimulated following exclusion and complications. Subacute 1-week studies were n = 8 sham control and n = 8 stimulated. All data were anonymized prior to analysis and interpretation and unblinded post hoc. All curve fittings and statistical analyses were performed using MATLAB and the SciPy statistics package in Python.

Transparency, Rigor, and Reproducibility

The study design and analytic plan were not registered prior to the study. Sample size was eight rats per group based on a power analysis, which indicated > 80% power to detect an effect size of 10%, estimated to be sufficient for all presented analyses. For longitudinal studies, we initially assigned n = 10 animals per group. 30% were non-learners and excluded from the study (n = 6). Prior to group assignment, rats were rank-ordered based on injury kinematics and pre-injury reaching success and divided into treatment and control groups using an iteratively nearly balanced assignment of groups to ensure an even distribution of the two primary performance variables was achieved. Post-assignment, 3 animals, 1 from the sham control and 2 from the stimulation group, were lost due to surgical complications, yielding our final group numbers of n = 6 sham control and n = 5 stimulation for longitudinal behavior studies. For subacute Nr4a1 assessment following 1 week of stimulation, animals were assigned to treatment groups randomly. We assigned 8 animals per group, and no animals were lost, resulting in n = 8 stimulated and n = 8 sham animals. Surgeons were blinded to treatment groups for all procedures including the sham stimulator implant. Therapeutic administration of VSS was delivered 2 weeks after initial injury and 1 week after device implant. Sham control animals received non-functional stimulators 1 week after injury. Stimulation was delivered for 1 h per day at light cycle change for 8 weeks. Histological and behavioral data were collected and processed prior to unblinding for assessment. Planned follow-up studies comparing dorsal and ventral stimulation are ongoing. Data from this study will be made available in a FAIR data repository (odcsci.org). The wireless stimulators used for these studies have been made available for purchase through custom order at Indus Instruments (https://indusinstruments.com/).

Supplementary Material

Supplementary Figures

Data and Materials Availability

Please direct all inquiries to M.K.H., mkhogan@houstonmethodist.org. All data are available in the main text or the Supplementary Data.