Abstract
Spinal cord injury (SCI) is devastating, with limited treatment options and variable outcomes. Most in vivo SCI research has focused on the acute and early post-injury periods, and the promotion of axonal growth, so little is understood about the clinically stable chronic state, axonal growth over time, and what plasticity endures. Here, we followed animals into the chronic phase following SCI, to address this gap. Male macaques received targeted deafferentation, affecting three digits of one hand, and were divided into short (4–6 months) or long-term (11–12 months) groups, based on post-injury survival times. Monkeys were assessed behaviorally, where possible, and all exhibited an initial post-injury deficit in manual dexterity, with gradual functional recovery over 2 months. We previously reported extensive sprouting of somatosensory corticospinal (S1 CST) fibers in the dorsal horn in the first five post-injury months. Here, we show that by 1 year, the S1 CST sprouting is pruned, with the terminal territory resembling control animals. This was reflected in the number of putatively “functional” synapses observed, which increased over the first 4–5 months, and then returned to baseline by 1 year. Microglia density also increased in the affected dorsal horn at 4–6 months and then decreased, but did not return to baseline by 1 year, suggesting refinement continues beyond this time. Overall, there is a long period of reorganization and consolidation of adaptive circuitry in the dorsal horn, extending well beyond the initial behavioral recovery. This provides a potential window to target therapeutic opportunities during the chronic phase.
Significance Statement
Most preclinical studies of spinal cord injury focus on the early phases of recovery, during which the greatest behavioral improvements occur and there is significant sprouting of spared fibers. Here, we extended these observations into the chronic phase, in a primate model of spinal injury affecting hand function, to see if these changes were maintained long term. We show that following an early period of corticospinal and spared primary afferent sprouting, afferents remain stable while exuberant corticospinal tract sprouts are pruned back to their baseline range. The presence and activation of microglia demonstrate that this process is driven partly by inflammation. Our findings provide important new insight into the chronic phase of recovery and the potential for longer-term plasticity.
Introduction
Spinal cord injury (SCI) is severely disabling, with huge personal, and societal costs. Not surprisingly, patients with upper limb involvement list improving hand function as a top priority (Anderson, 2004). Most SCIs are incomplete with an opportunity for at least partial recovery. However, the process is complex, and our current lack of understanding of the underlying mechanisms provides a significant barrier to rehabilitation.
Once the early acute phase responses (Tang et al., 2022) recede and relative stability is achieved, upregulated gene expression triggers the regenerative growth response (including axon sprouting). Most behavioral recovery occurs during this “subacute” or “sub-chronic” period, and this has traditionally been when in vivo studies have probed spontaneous recovery and tested therapeutic interventions. However, this completely overlooks the chronic period, which is lifelong and most accessible. Stability is thought to be established clinically at ∼2 years in humans (Ziegler et al., 2018), yet there is compelling evidence that functional improvements can continue well beyond this point (Corbetta et al., 2002; Gill et al., 2018).
Axon sprouting has often been correlated with behavioral recovery in SCI (Hagg, 2006; Rosenzweig et al., 2010; Jin et al., 2015; Zareen et al., 2017; Jiang et al., 2019; Rosenzweig et al., 2019; Cao et al., 2021; Martin, 2022). The corticospinal tract (CST) is often the focus, as the dominant descending pathway mediating hand function in primates. Extensive CST sprouting has been reported in many species during the early stages of recovery, predominantly in the motor CST (Fouad et al., 2001; Rosenzweig et al., 2010; Oudega and Perez, 2012; Friedli et al., 2015; Jin et al., 2015; McCann et al., 2020; Cao et al., 2021). Few studies have addressed what happens to the sprouts long term, particularly in primates, and none have examined long-term changes to the somatosensory (S1) CST despite known clinical problems that can arise from uncontrolled sprouting of sensory fibers (e.g., spasticity and pain).
Here, we focus on the S1 CST, which has received little attention yet is critical to function and contributes ∼25% of fibers to the macaque CST (Galea and Darian-Smith, 1994). Their terminals overlap with afferent inputs in the spinal dorsal horn (Fisher et al., 2020), making them well placed to mediate sensory gating and movement control (Seki et al., 2003; Seki and Fetz, 2012). The S1 CST sprouts and significantly expands its terminal territory in the months following a combined dorsal root/dorsal column lesion (DRL/DCL) that targets hand function (Darian-Smith et al., 2014; Fisher et al., 2018, 2020), and behavioral recovery also occurs during this period (Crowley et al., 2021). However, nothing is known about what happens to the dorsal horn circuitry over the longer term.
Here, we use the same spinal deafferentation model (DRL/DCL) as in earlier reports (Darian-Smith et al., 2014; Fisher et al., 2020, 2022), to ask questions about the chronic phase. While this model does not simulate typical clinical SCIs, its focus and reproducibility make it uniquely suited for studying the recovery process. As with our earlier investigations, the DRL/DCL only affected digits 1–3 of one hand and caused an initial deficit in a precision grip task requiring sensory feedback. Specifically, we asked the following: (1) Is the extensive S1 CST sprouting that is observed in DRL/DCL animals 4–5 months post-lesion maintained into the chronic recovery period, or is it pruned to refine and consolidate the circuitry? (2) What happens to the spared primary afferents of chronically deafferented digits, which are also known to sprout within the dorsal horn over the early post-injury months? Finally, (3) how do microglia (a major component of the immune response) contribute to the chronic phase of recovery?
Our findings demonstrate that despite early behavioral recovery, anatomical changes within the spinal dorsal horn continue for at least 1 year post-injury, with the ongoing microglial response contributing to this reorganization. The timeline approximates clinical accounts and offers insight into chronic rehabilitation strategies.
Materials and Methods
Data from 10 young adult macaque monkeys (Macaca cynomolgus) were used. Three are the main focus of this work (M1703, 1704, and 1705; 4.8–5.64 kg), and the others (two controls, 1601 and 1702, and five DRL/DCL animals, 1402, 1602, 1603, 1804, and 1805), reported in previous studies (Fisher et al., 2020, 2022; Crowley et al., 2021), were used for a critical time point comparison. All the animals described were male (due to vendor availability). However, there is no evidence of gender differences in either hand function or sensorimotor recovery following SCI, so this was not seen as a confound. We otherwise followed the ARRIVE guidelines for animal research.
Monkeys were healthy (with no damage to hands/digits), colony bred (Charles River), and housed individually at the Stanford Research Animal Facility. Each had access to four-unit cages (64 × 60 × 77 cm, depth × width × height per each unit) and was housed in a room (12 h light/dark cycle, at 72–74°F), with other monkeys.
Animals always had access to water and a primate diet, which was supplemented with fresh fruit, vegetables, and novel foods and drinks. Enrichment was provided daily and included toys and puzzles in home cages, as well as videos and music in the housing area, and behavioral training in an adjacent room.
Animal procedures were carried out in accordance with National Institutes of Health guidelines and the Stanford University Institutional Animal Care and Use Committee (IACUC).
Experimental sequence and timeline
Three groups of animals are discussed in this report—controls, short-term DRL/DCLs (4–6 months recovery from lesion), and long-term (chronic) DRL/DCLs (11–12 months recovery from lesion). Table 1 provides details.
Details of the animals used in this study
Long-term DRL/DCL animals underwent a laminectomy to make the lesion, which affected digits 1–3 of one hand. Monkeys recovered for 10–11 months before undergoing a unilateral craniotomy to map and place anterograde tracer injections. Approximately 5 weeks later, animals were sedated, and cholera toxin subunit B (CT-B) was injected bilaterally into the finger pads of digits 1–3 to label primary afferents. After allowing 1 week for retrograde transport, monkeys were perfused, and tissue was retained for histological analysis.
The experimental sequence for short-term DRL/DCL animals was the same but with only 3–5 months of recovery between the lesion and craniotomy. Control monkeys received a craniotomy only.
Surgery
Clinical care followed the same pattern for all surgical procedures. Animals were initially sedated with ketamine hydrochloride (10 mg/kg) and subsequently maintained under gaseous anesthesia (isoflurane, 1–2% in/1% O2) for surgery, using a standard open circuit anesthetic machine. The surgical site was line blocked (bupivacaine) pre-incision, and atropine sulfate (0.05 mg/kg), buprenorphine (0.015 mg/kg), and the antibiotic cefazolin (20 mg/kg) were administered. A Normosol-R infusion (i.v.) was adjusted as necessary throughout surgery to maintain fluid balance. For craniotomy procedures, dexamethasone (0.25 mg/kg) was given to minimize brain edema.
Physiological signs (including blood pressure, heart rate, oxygen saturation, and capnography) were monitored throughout surgery to ensure deep, stable anesthesia. Core temperature was maintained using a thermostatically controlled heating pad (circulating water) and an air blanket.
Sustained release buprenorphine (0.01 mg/kg, ∼48 h coverage) was given at the end of surgery for postoperative analgesia. Monkeys were returned to their home cages for recovery and monitored closely until fully alert, and it was clear that there were no postoperative sequelae. In the following 3–5 d, oral meloxicam (0.1–0.2 mg/kg) was administered as necessary.
Laminectomy and DRL/DCL lesions
These methods have been described extensively in earlier papers (Darian-Smith et al., 2014; Fisher et al., 2018, 2020). Briefly, animals in the lesion groups underwent a laminectomy to expose spinal segments C5–C8. Receptive fields were mapped using electrophysiological recordings from the dorsal roots, which let us construct a micro-dermatome map. Rootlets with detectable inputs from digits 1–3 were then transected (the DRL component of the injury), using iridectomy scissors, and a gap was left to prevent nerve regeneration. The DCL was made at the rostral border of the DRL (i.e., thumb input), using a micro scalpel blade (Feather micro scalpel, 15°), to make a 2 mm incision into the cuneate fasciculus.
Craniotomy
A unilateral craniotomy was made in all monkeys to access the “hand” region of the sensorimotor cortex (contralateral to the lesion side), as described in detail in previous publications (Darian-Smith and Brown, 2000; Darian-Smith et al., 2013, 2014; Fisher et al., 2018, 2020). Briefly, S1 (areas 3b/1) was mapped electrophysiologically to determine the location of inputs from the partially deafferented digits and inform the placement of anterograde tracer injections (Fig. 1). Upon completion, the bone flap was secured using bone wax and Vetbond adhesive, and the opening was closed.
Spinal cord lesions and cortical injection sites in control (A), short-term (B), and long-term (C) animals. A, Schematics of the sensorimotor cortex viewed from above, which show S1 cortical recording locations (black dots) and BDA injection sites (orange circles) for each monkey. B, Illustrations of the dorsal root and DCL positioning (green lines) in short-term animals, drawn from an image of the laminectomy taken during surgery. Individual dorsal roots are outlined in orange (those with D1–D3 receptive fields) or gray (those with receptive fields on D4–5 or elsewhere on the hand/wrist). Cross sections of dorsal column lesions show that these were confined to the cuneate fasciculus and in some animals involved the medial dorsal horn. The green dotted lines outline the region affected by the DCL. Note that laminectomies and lesions are always shown on the left here for ease of viewing, and schematics are shown as these images have been published previously (Fisher et al., 2020, 2022). The actual lesion laterality for each animal is indicated in Table 1 for reference. S1 cortical recording (black dots) maps with BDA (orange circles) or LYD (yellow circles) injection sites are also indicated. C, Experimental photographs for the long-term animals, starting on the left in each case with images of the spinal cord superimposed with schematics of dorsal roots related (orange) or unrelated (gray) to D1–D3, as well as lesion locations (green lines). Adjacent micrographs show cross sections of the cord at the level of the DCL with the affected area outlined by a green dotted line. Also shown are images of the cortex during the craniotomy with recording sites (black dots) and injection sites (orange-filled circles) superimposed and coronal brain sections showing the BDA injection sites in S1. Lamina IV and gray matter boundaries were identified in adjacent Nissl-stained sections. The scale bars are consistent for each image type.
Tracer injections
Cortical injections
Biotinylated dextran amine (BDA, 10–15% aqueous, Sigma B9139) was injected across the D1–D3 span of S1 in long-term DRL/DCL animals. For the other monkeys used in this study, either BDA or Lucifer yellow dextran (LYD, 15% aqueous, Thermo Fisher D1825) was used (Table 1). A constant-pressure Hamilton syringe with a secured glass micropipette tip (diameter ≤ 30 µm) was mounted in a micromanipulator and used to make the injections. Multiple injections (each 0.3 µL, 0.8–1 mm depth) were made to flood the D1–D3 region (Fig. 1). The syringe remained in place for 2 min before moving to the next site. Six to 7 weeks were allowed for tracer transport.
Digit pad injections
A week before perfusion, animals were sedated with ketamine (5 mg/kg), and CT-B (10 µL, Sigma C9903%, 1% concentration) was injected subcutaneously into the distal and middle pads of digits 1–3 of both hands, using a Hamilton syringe.
Note that animals 1804 and 1805 had CT-B injected into intrinsic hand muscles instead of the digit pads; this data is not discussed here and is the subject of a separate publication (Fisher et al., 2022).
Perfusion
Animals were sedated (ketamine, 10 mg/kg), intubated, and transferred to isoflurane anesthesia. A lethal dose of sodium pentobarbital (Beuthanasia, minimum 100 mg/kg) was given (i.v.), and perfusion was transcardial with heparinized saline followed by 4% paraformaldehyde (in 0.1 M phosphate buffer solution; PB). The brain and spinal cord were postfixed for 24 h, before transfer to 20% sucrose (in 0.1 M PB) for cryoprotection. Tissue was then blocked, flash frozen using isopentane, and stored at −80°C.
Tissue processing
CNS tissue was sectioned using a freezing microtome (brain–coronal plane, 50 µm thickness; spinal cord–transverse plane, 40 µm), and free-floating sections were collected in series for immunohistochemistry.
For reactions involving permanent chromogens [3,3′-diaminobenzidine (DAB) or VIP substrates], processing began with an initial block of endogenous peroxidases using Bloxall (Vector Laboratories, SP-6000) followed by three washes in 0.1 M PB.
BDA was visualized using a metal-enhanced DAB reaction (producing a black reaction product). For better penetrance, we used 0.1% Triton X-100 in our 0.1 M PB buffer. Sections were incubated in ABC elite (Vector, PK-6100, 1 h, room temperature), washed three times in buffer, and reacted with nickel-intensified DAB with urea peroxidase (Sigmafast, D0426, Sigma-Aldrich) for 5–12 min.
For CT-B processing, sections were blocked in 10% horse serum (2 h, room temperature) and incubated in anti-CT-B made in goat (List Biological Laboratories #703; 1:4,000 dilution) in TBS-TX/3% horse serum for 2–3 d at 4°C. After thorough washing, sections were further incubated in ImmPRESS reagent (HRP anti-goat IgG; Vector Laboratories, MP-7405; 2 h, room temperature) and washed again. Finally, VIP substrate (Vector Laboratories, SK-4600) was used to visualize CT-B terminals. Reactions were rapid and arrested with PBS after 1–2 min before the background became too dark. We took care not to dehydrate sections processed in this way as VIP is soluble in alcohol. Alternatively, they were mounted onto slides, dried overnight, transferred directly to xylene, and coverslipped immediately.
Immunofluorescence protocols had the same workflow regardless of the antibody combination. Sections were first thoroughly washed in PB with 0.5% Triton X-100 (PB-TX) and then blocked with a 10% solution of normal serum in PB-TX (2 h, room temperature). They were then transferred to a solution of primary antibodies (anti-CT-B, #703, List Biological, 1:4,000; anti-Iba-1, 019-19741, Wako, 1:6000; anti-NeuN, MAB377, Millipore, 1:100; anti-synaptophysin, MA5-14532, Invitrogen, 1:500; anti-VGLUT1, AB5905, Millipore, 1:5,000) diluted in PB-TX with 3% normal serum; this incubation lasted 48–60 h (4°C). Following multiple washes in buffer, secondary antibodies in PB-TX were added to the wells for 30 min at room temperature (conjugated secondaries were from Invitrogen or Jackson ImmunoResearch, 1:400). If BDA visualization was required, sections were subsequently washed and incubated in ExtrAvidin CY3 (E4142, Sigma-Aldrich, 1:200) in PB-TX overnight. Sections were then mounted using 0.5% gelatin, air dried for 30–45 min, and coverslipped using antifade mounting media with or without DAPI (ProLong Diamond Antifade Mountant, Invitrogen).
LYD visualization required immunostaining, and we refer the readers to the previous publications since this was not carried out in long-term DRL/DCL animals (Fisher et al., 2018, 2020).
Light microscopy
We used Neurolucida software in combination with a Lucivid projection (MBF Bioscience) and a Zeiss Axioscope microscope to map terminal distributions of BDA-, LYD-, and CT-B-labeled axons. Sections spanning C1–T2 were mapped at a frequency of 400–1,600 µm (depending on analysis). The number of sections used per animal varied depending on where the tracer was present in the spinal cord (CST analysis, 34–82 sections/animal; afferent analysis, 15–39 sections/animal). We outlined the distribution territory within each section and calculated its area. Outlying boutons (<5) were not included since they constituted <1% of the total population (Darian-Smith et al., 2014; Fisher et al., 2018).
Retraction bulb counts
We counted the number of retraction bulbs in sections from C5–6 processed for BDA (S1 CST label). This was done over a region of ∼10 mm centered on the DCL site and corresponded to 12–14 sections per animal. Since there was no clear definition of a corticospinal retraction bulb in the literature, we chose to count only the labeled boutons for which the area exceeded 12 µm. This was done because we have previously not identified S1 CST terminals with areas >10 µm (Fisher et al., 2020), most of them being substantially smaller. The mean terminal size we found for retraction bulbs was 19.88 µm (range = 12.191–49.869 µm).
Confocal microscopy
Immunofluorescent sections were scanned using a Nikon A1R confocal microscope, and resulting 3D images were analyzed offline using Nikon Elements software.
Neuron density and dorsal horn gray matter area analyses
To assess changes in the gray matter area (n = 9 sections/animal) and neuronal density (n = 5–8 sections/animal) in the dorsal horn, we analyzed 2D fluorescent NeuN images from the lesion zone (±2 mm relative to DCL). These confocal images (10×) were analyzed in ImageJ. An outline was drawn around the dorsal horn to measure area, and a cell counter plugin was used to mark and count individual neurons.
Synapse counts
To determine the proportion of presumed functional synapses in the S1 CST terminal region, we analyzed 3D images of C6 dorsal horn sections labeled with NeuN, BDA, and synaptophysin (Fig. 6A; 60×; n = 4–5 per animal). These were viewed in 3D by two people (C.D.-S. and K.M.F.). The image area was restricted to 210 × 210 × 10 µm to ensure fair comparison between samples. For each image, viewers manually marked and counted both (1) individual BDA boutons (representing S1 CST terminals) and (2) BDA boutons co-labeled with synaptophysin (to indicate functionality). We then normalized the co-labeled synapses to the actual number of BDA terminals.
Microglia analyses
To probe whether microglia were upregulated and activated in lesioned animals, we analyzed 3D scans of tissue from C5–6 labeled with Iba-1 (60×; three sections per animal). As before, all images had the same area (210 × 210 × 10 µm). We (1) counted the number of microglia present in each image and (2) measured the soma area at the largest point. Both measures would be expected to increase in line with inflammation and activation of immune cells. While Iba-1 is known to label macrophages in addition to microglia, we are confident microglia were being counted, based on the morphology of the cells, their location in the CNS, and the longer timeline of our analysis post-lesion (Uff et al., 2022).
Statistical analysis
As described in previous work, we used a repeated measures approach to control for systematic variation wherever possible (Darian-Smith et al., 2014; Fisher et al., 2018, 2020). This is particularly useful in nonhuman primate studies that involve small numbers of animals, as it greatly increases the statistical power. All analyses were performed in JMP Pro 15/16 Pro and SAS 9.4 for Windows.
While there was no blinding during data collection (both K.M.F. and C.D.-S. were aware of the lesion status of each animal due to the involved nature of primate studies), our statistician (J.P.G.) was blinded to treatment and expected outcome.
Note that the factorial design of our analyses utilizes least-square means (LSMs), not arithmetic ones. These represent group averages that control for all other variables (e.g., animal). Least-square and arithmetic means typically differ so it would be misleading to plot individual datapoints alongside LSMs. While we do not plot the raw data in any of our graphs, it is available on request should anyone want to verify our methods.
Neuron density and dorsal horn gray matter area analyses
Ratios (i.e., lesioned side to contralateral side) were calculated for neuronal density, and the gray matter area, with each section acting as its own control. We then performed a repeated measures REML mixed model, with subjects nested within a group (i.e., control vs short-term DRL/DCL vs long-term DRL/DCL groups) as a random effect. Section distance from the DCL lesion was included as a control variable. Tukey’s post hoc tests were used to test for differences between groups, and Bonferroni-corrected contrasts were used to test for differences from a 1:1 ratio. Appropriate error terms were used (Newman et al., 1997). Ratios were log-transformed to meet the assumptions of linear methods (Grafen and Hails, 2002).
S1 CST and primary afferent terminal area analysis
The terminal territory was analyzed using a repeated measures REML Mixed Model, following best practices (Newman et al., 1997; Grafen and Hails, 2002; Littell et al., 2002; Thabane et al., 2013). The subject was Monkey, treated as a random effect, and nested within the three groups, or “treatments” (control, n = 2; short-term DRL/DCL, n = 2; and long-term DRL/DCL, n = 3). Treatment was therefore a between-subject experimental fixed effect and was further nested within lesioned versus control animals. Each territory area was normalized to the corresponding gray matter area to produce a ratio. The ratio for any given section was assigned to its spinal cord segment, and the segment was treated as a within-subject experimental fixed effect, further nested within rostral versus caudal segments to the lesion. Positional effects between segments were controlled for by an autoregressive covariance structure. Interactions between variables of interest were then tested with the appropriate error terms for mixed models, including Kenward–Roger-corrected degrees of freedom (Newman et al., 1997; Littell et al., 2002). This design represents a hierarchical linear model implemented with an autoregressive repeated measures design. The hierarchical approach explicitly tests for higher-level differences (i.e., control vs lesion; rostral vs caudal) before finer-grained distinctions (i.e., short-term vs long-term or individual segments), which is a higher power approach than testing for fine-grained effects and then examining higher-level differences as post hoc tests. Following a significant interaction, we used post hoc planned contrasts to dissect the result, correcting for multiple comparisons. We confirmed the assumptions of linear models (normality of error, homogeneity of variance, and linearity) post hoc and square-root transformed the terminal territory ratio to meet these assumptions. The same analysis was performed for both afferent and efferent datasets except that for the efferent dataset, the hierarchy for treatment involved grouping controls with long-term lesions versus short-term lesions.
Synapse analysis
To analyze synapse counts, we performed a repeated measures REML mixed model, with monkey nested within treatment as a random subject effect and treatment as a fixed effect. Appropriate error terms for mixed models were used in all subsequent tests (Newman et al., 1997). Co-labeled synapse counts (BDA+/synaptophysin+) were expressed as proportions, with the number of co-labeled synapses divided by the total number of BDA-labeled boutons in the section. Proportions are more accurate when more items are counted, and the total bouton count varies considerably between sections. Therefore, to avoid less accurate proportions having greater sway over the data, we calculated the variance of each proportion (Woodward, 1999) and used the inverse of that variance as a weighting factor (Neter et al., 1996). We have used this approach to solve similar problems (Madrid et al., 2017; Parker et al., 2019; Oztan et al., 2020). Further assumptions of linear models were confirmed post hoc. Tukey’s tests were used to correct for multiple comparisons when examining significant effects.
Microglia analysis
For the glial soma area, we first calculated the ratio of the mean soma area (lesioned to the contralateral side), so each section acted as its own control. We then used a repeated measures REML mixed model with subjects nested within a group as a random effect and with a group as the experimental variable of interest. The ratio was log-transformed to meet the assumptions of linear methods (Grafen and Hails, 2002). Again, Tukey’s tests were used post hoc to examine significant differences.
For glial count, we initially adopted the same approach. However, the within-subject variance in count violated the assumption of homogeneity of variance and could not be met by transformation, so we calculated the mean and variance of the log of the ratio for each subject. This allows each section to be its own control, and the log transformation is required to meet underlying assumptions (particularly linearity). We then performed a weighted least-square GLM (WLS-GLM) test for group effects. WLS-GLM controls for non-homogeneity of variance, if the variance of a data point can be estimated and the inverse of the variance is used as the weight (Neter et al., 1996). Tukey’s tests were again used post hoc to examine significant differences.
Retraction bulb analysis
The total number of retraction bulbs observed was divided by the number of sections observed for each subject. The resulting data were highly skewed, so to meet the underlying assumptions of linear methods, the data were rank-transformed and analyzed in GLM as a hierarchical linear model (GLM-HLM), with group nested with lesioned versus control animals. An HLM approach is more powerful for small sample sizes than a simple one-way GLM, because it restricts the hypotheses tested to only those specified and does not rely on correction for multiple comparisons.
Results
Data were examined in the dorsal horn of monkeys at two timepoints following a DRL/DCL and compared to control animals. The projections examined related specifically to digits 1–3 and therefore underpin fine-directed finger movements, unique to higher primates and of particular interest for rehabilitation in SCI (Anderson, 2004).
Extent and effect of lesions
The size and placement of spinal lesions were assessed histologically (Fig. 1). Monkeys 1703 and 1705 had very similar lesions, consistent with the short-term DRL/DCL animals. Monkey 1704 had a slightly smaller DRL and DCL (Fig. 1), which corresponded to less impairment. However, there was still sufficient evidence of weakness and clumsiness in digits 1 and 2 on behavioral testing to consider this injury broadly consistent with all previous animals. DCLs were confined to the cuneate fasciculus of the dorsal column, though monkeys 1804 and 1805 showed some additional partial involvement of the adjacent medial dorsal horn.
All lesioned animals had a region of minor distortion immediately adjacent to the cuneate fasciculus, due to the loss of input fibers and neuropil following deafferentation (Fig. 1). However, the dorsal horn ipsilateral to the lesion was distinctly elongated, and its area significantly decreased in the three long-term DRL/DCL monkeys at 1 year post-lesion (F2,2.951 = 14.69; p = 0.0292; Fig. 2A), though this did not amount to a change in neuronal density (F2,3 = 1.33; p = 0.3859; Fig. 2B). The reduction in gray matter area was not significant at 4–5 months indicating that the change was part of the chronic reorganization.
Changes in dorsal horn morphology following chronic reorganization. The bar graphs show the ratio (ipsi- vs contralateral to the lesioned side of the cord) of the dorsal horn gray matter area (A) and the corresponding neuronal density for each experimental group (B). The asterisks indicate significance.
Tracer injections
Cortical injection sites were examined and found to be consistent across animals, with no contamination either across the central sulcus or of underlying white matter tracts (Fig. 1C, right column). Given electrophysiological guidance during surgery, the tracers were localized to the D1–D3 region of the S1 cortex.
Behavioral recovery is achieved early
Immediately after the lesion, animals exhibited a clear sensorimotor deficit in the ipsilateral hand, with poor precision grip, inaccurate targeting, and relative weakness of the affected digits. This did not preclude the use of the hand, but the lack of tactile and proprioceptive feedback from the digits often forced monkeys to pay close visual attention to fine hand movements where this was unnecessary pre-lesion.
While we were unable to collect sufficient data for statistical analysis in the long-term monkeys used in this study, they followed the same pattern of deficit and recovery observed in other monkeys with the same lesion, or a DRL alone (Darian-Smith and Ciferri, 2005), as recently reported (Crowley et al., 2021). Performance of a reach retrieval task was initially impaired post-injury, but over ∼8 weeks, monkeys learned new compensatory strategies to restore these parameters to pre-lesion efficiency. Importantly, the behavioral recovery observed was compensatory and not the restoration of a normal precision grip (Fisher et al., 2020; Crowley et al., 2021).
Figure 3 shows examples of pre- and post-lesion precision grip strategies from long-term monkeys used in this study. Pre-lesion, both animals opposed the distal pads of their thumb and index fingers to precisely grasp and retrieve target pellets from a manipulandum. Post-lesion, this was far more challenging, and they had to adopt alternative strategies. We noted that the strategies developed crudely related to the individual's hand structure, and the relative distance between their thumb and index fingertips (Fig. 3B,D). For example, Monkey 1704 had a smaller thumb–index distance, which enabled the use of a retrieval strategy reasonably close to pre-lesion even though the candy was typically mispositioned on the side of the distal or middle fingerpad of the thumb. In contrast, monkey 1705 had a longer thumb–index distance, which meant that post-lesion, he found it easier to use the index finger as a winkling tool.
Altered retrieval strategies following DRL/DCL. A,C, Images from videos taken of two of the long-term animals while performing a reach retrieval task. In each case, a pre-lesion example shows the typical precision grip strategy employed to retrieve the candy pellet. Following the lesion, we show two new strategies used by the different animals. Monkey 1704 tries to continue performing a precision grip, but inadequate sensory feedback likely affected the force applied, and the candy pellet rolls to rest against either the side or lower part of digits 1 and 2. Monkey 1705, however, completely changed strategy and chose to winkle the candy out with digit 2, using the base of the other digits for support. B,D, Images of the hand of each monkey, and the considerable inter-animal variability of the D1:D2 length ratio, which is likely to account for differences in digit strategies used across individuals (Crowley et al., 2021).
Consolidation of exuberant S1 CST sprouts occurs by 1 year post-injury
Example S1 CST terminal distributions from the three groups are shown in Figure 4A. The overall terminal territory expanded during the initial months following the DRL/DCL (Darian-Smith et al., 2014; Fisher et al., 2020) and then retracted to a normal level over the post-lesion year (control and long-term mean vs short-term; F1,3.92 = 35.46; p = 0.0043). The rostrocaudal S1 CST distribution in control animals was not statistically different from that in long-term animals (F7,19.1 = 0.40; p = 0.8888; Fig. 4B). Nor was there a statistical difference between segments located rostral and caudal to the DCL site.
Digit 1–3 related S1 CST fibers in the spinal cord initially sprout exuberantly following DRL/DCL but then retract by 1 year post-lesion. A, The segmental termination patterns of S1 CST fibers are shown for representative monkeys from each experimental group (control, 1601; short-term DRL/DCL, 1603; long-term DRL/DCL, 1703). Tracer injections were made in one cortical hemisphere, so terminals are only plotted contralateral to this (S1 CST fibers only terminate contralaterally in the spinal cord). In each spinal section, a colored line encircles the terminal area, and the gray-shaded region indicates the lesioned side for DRL/DCL animals. For consistency, the lesion is shown on the left side regardless of which side the tracer injections were made. The line graphs at the bottom of each column summarize the different rostrocaudal terminal areas between C1 and T2, to highlight how the DRL/DCL affects this projection. The rostrocaudal location of each section (a–c) is indicated with arrows in corresponding graphs, and the gray line indicates the approximate location of the DCL. Note that data from monkeys 1601 and 1603 have been published previously (Fisher et al., 2020). B, Population-level rostrocaudal profiles for S1 CST terminal areas (averaged for control, n = 2; short-term DRL/DCL, n = 2; and long-term DRL/DCL, n = 3). Data are shown as the percentage of the gray matter area occupied by S1 CST terminal boutons. The gray-shaded area indicates the DRL zone. The DCL was always located at the rostral end of the DRL. The LSM ± standard error was plotted, and the asterisks indicate significance.
Primary afferent inputs stabilize before descending projections
Following the lesion, primary afferent terminal distributions on the affected side of the spinal cord were reduced. Individual examples from control (left column) and lesioned animals (middle and right columns) are illustrated in Figure 5A, and the population-averaged effect is shown in Figure 5B. The magnitude of this reduction in the afferent terminal area depended on the spinal segment (F8,38.1 = 2.54; p = 0.0251), and after correction for multiple comparisons, only C5 (F1,16.2 = 17.09; p = 0.0008; Bonferroni critical alpha = 0.005) and C6 (F1,19,1 = 17.36; p = 0.0005) showed significant differences. Previous work has shown that there is some early local afferent sprouting within the dorsal horn (Darian-Smith, 2004). However, we saw no difference in the afferent distributions between 4 and 5 months and 1 year post-lesion (F8,34.1 = 0.76; p = 0.6361), indicating that the spared afferent population stabilizes very early and any changes in the chronic window are likely synaptic. No significant changes were observed contralateral to the lesion (F8,20.1 = 0.51; p = 0.8350; Fig. 5B, inset).
Cutaneous afferent inputs to the spinal cord sprout and stabilize within the first few months of injury. A, Primary afferent terminal maps are shown in select spinal sections for representative animals from each experimental group (control, 1601; short-term DRL/DCL, 1603; long-term DRL/DCL, 1703). Bilateral populations are illustrated since CT-B was injected into D1–D3 in both hands (see inset schematic for injection locations). The line graphs underneath summarize the rostrocaudal distributions for both the intact/normal (dotted line; right side for control) and lesioned (solid line; left side for control) sides. Note that data from monkeys 1601 and 1603 have been published previously (Fisher et al., 2020). B, Population profiles of cutaneous primary afferent terminal input territories on the lesioned side (averaged for control, n = 2; short-term DRL/DCL, n = 2; long-term DRL/DCL, n = 3). Data are shown as the percentage of the gray matter area occupied by primary afferent terminal boutons. The inset graph shows the same data for the contralateral (intact) side of the cord. The grey-shaded area indicates the DRL zone. The LSM ± standard error was plotted, and the asterisks indicate significance.
S1 CST functional synapses increase initially but are then pruned by 1 year post-injury
Given that the primary afferent terminal domain had stabilized by 4–5 months and remained unchanged into the chronic phase, we focused on changes to the descending S1 CST fibers for the rest of our analyses. Figure 6A shows examples of putatively functional S1 CST synapses (BDA/SYN co-labeled terminals) in the dorsal horn. We quantified these across our animals and found that the numbers increased almost twofold at 4–5 months post-lesion but then reduced to within baseline limits by 1 year (Fig. 6A, iv). This effect was significant (F2,3.24 = 17.97; p = 0.0176), and there was no difference between the control and long-term groups.
Modulation of co-labeled S1 CST/synaptophysin synapses and retraction bulbs in the affected dorsal horn indicates plasticity during recovery from injury. A, Series of confocal images showing the analysis approach taken to count functional S1 CST synapses in the dorsal horn. i, S1 CST terminals (red) within the dorsal horn. ii, 3D rendered z-stack taken from the center of the BDA-labeled region (indicated by the yellow box in i). iii, Regions of interest taken from the overall 3D scan (ii) with the relative position indicated by the colored boxes. White circles indicate examples of co-labeled terminal boutons (BDA and synaptophysin), which are presumed functional synapses. iv, The bar graph shows the proportions of BDA terminals, which were co-labeled with synaptophysin for each of the three experimental groups (control, n = 2; short-term DRL/DCL, n = 2; long-term DRL/DCL, n = 3). A compact letter display indicates the results of Tukey’s significance test. Bi, 3D image of a putative retraction bulb in the lesioned dorsal horn. The individual color sequence underneath (ii) indicates that this large structure, while being heavily labeled with BDA, does not co-register with either synaptophysin (green) or VGLUT1 (blue). iii, The bar chart quantifies the number of retraction bulbs counted per section. The error bars represent standard errors.
In conjunction with the reduction of functional S1 CST synapses in the long-term DRL/DCL group, we also observed numerous large BDA-labeled terminal swellings (Fig. 6B). These never co-labeled with synaptophysin or VGLUT1, which indicates a lack of functionality. These were likely to be retraction bulbs and indicative of degeneration of exuberant synapses. The occurrence of these retraction bulbs increased with time post-lesion (Fig. 6B, iii), and significant differences were found both between control and lesioned animals (F1,4 = 14.91; p = 0.0112) as well as between the short- and long-term groups (F1,4 = 7.500; p = 0.0341).
Increased microglial presence persists into the chronic phase of recovery
Since there was a clear switch between synapse formation and elimination over time, we were interested in whether inflammation, which has been shown to occur chronically in SCI patients, might play a role in modulating synapses. Figure 7 illustrates the change in microglial profiles in the dorsal horn over the course of the first post-lesion year. Example images from a short-term lesioned monkey in Figure 7A show qualitatively that more microglia are present on the lesioned side of the spinal cord 4–5 months post-injury. The population analysis confirmed this increase was significant and that it was maintained for 1 year post-injury (Fig. 7B, left panel; F2,7 = 23.35; p = 0.0008). Note that the relative difference between sides for controls was noise, as we randomly assigned the left or right sides to be ipsi- or contralateral for each animal.
Microglia migrate to the lesion site and become activated. A, Representative images taken from 3D confocal scans in the dorsal horn S1 CST terminal hotspot of the short-term DRL/DCL monkey M1603. Images show microglia (Iba-1, green). B, The bar graphs show changes in microglia numbers (left) and microglia soma area (right) in the dorsal horn across the three experimental groups (control, n = 2; short-term DRL/DCL, n = 5; and long-term DRL/DCL, n = 3). In both histograms, the bars represent the ratios of ipsilateral to contralateral values and are expressed as percentages. Significance was tested using Tukey’s test (for multiple comparisons), and the results are indicated using a compact letter display. C, Examples of microglia (Iba-1, white) phagocytosing BDA-labeled (green) corticospinal terminals in the dorsal horn on the lesioned side of monkey M1703 (C5–6), 12 months after a DRL/DCL. The arrowheads (yellow) indicate BDA-labeled material internalized into the phagocytosing microglia. The example to the right shows a 3D confocal image of a tissue slab. The thicker slab (behind) contains an entire microglial cell, and this slab has then been sliced in half (along the yellow dotted line), to definitively reveal that BDA neuronal tissue (green) is inside (front lower right). Scale bar = 20 µm.
Microglial cell body size was significantly increased for the short-term DRL/DCL group (Fig. 7B, right panel), a factor that has been linked to the proinflammatory-activated state (Davis et al., 2017). This returned to baseline level by 1 year (F2,7 = 33.03; p = 0.0003). Together, these findings suggest that activated microglia move into and stay in the area affected by the lesion, and the differing levels of activation may indicate a changing role from the sub-chronic to the chronic phase of recovery.
Evidence of phagocytosis by activated microglia is shown in Figure 7C. This was clearest in the short-term DRL/DCL group but was also evident at the 12-month timepoint, which was not entirely unexpected, given that it occurs even in healthy tissue. Debris was observed internalized into the cell body, as well as attached or adjacent to phagocytic cups or microglial processes, as has been described in previous work (Jiang et al., 2019; Rotterman and Alvarez, 2020).
Discussion
While SCI is variable and debilitating, even minimal pathway sparing can lead to significant functional recovery. The injury model used here was, by design, purely sensory and relatively minor compared with most clinical SCIs. This simplified the injury, and reduced the variables at play, making it easier to determine the role of specific neuronal pathways in spontaneous recovery.
Our findings indicate that following one of these targeted DRL/DCLs, (1) the S1 CST projection fibers, which regulate inputs from the affected afferents, sprout exuberantly in the dorsal horn for many months, but then undergo substantial pruning and consolidation. (2) Afferent fibers directly affected by the injury sprout locally during the immediate weeks/months, but then stabilize and do not change significantly between 4 and 5 months and 1 year post-lesion. (3) There is also an influx of activated microglia to the deafferented dorsal horn. While microglia do not remain chronically activated, they persist in large numbers at 1 year post-lesion, suggesting they continue to contribute to reorganization beyond this time.
Behavioral recovery of hand function in all of our animals with the targeted DRL/DCL occurs within the first ∼8 weeks (Fisher et al., 2020; Crowley et al., 2021), while the anatomical reorganization occurs over a much more protracted period, and potentially well into the chronic phase. Of course, while the changes reported in this study are key to understanding the process of recovery and reorganization, additional changes occur at all levels of the neuraxis, from the periphery to the cortex (Darian-Smith and Fisher, 2020).
Consolidation of sprouted S1 CST axon terminals
The initial expansion of the S1 CST terminals in the first 4–5 months post-injury reflects compensatory sprouting, as previously reported (Darian-Smith et al., 2014; Fisher et al., 2020). Here, we now show that between 5 months and 1 year, these terminals retract to an area in the dorsal horn similar to that observed pre-lesion. Since only limited timepoints were examined, the precise timing of the switch between sprouting and dieback remains unclear. It is also not known if further pruning occurs beyond 1 year, perhaps to resemble the smaller terminal distribution pattern observed following a DRL alone (Darian-Smith and Ciferri, 2005; Darian-Smith et al., 2013; Crowley et al., 2021).
There are few if any reports that describe a sprouting–retraction pattern in the sensorimotor system over a protracted period following SCI, though the process is evident in the visual cortex following injury (Yamahachi et al., 2009), and it bares at least some resemblance to early development (Hilton and Bradke, 2017). There are, however, non-human primate (NHP) studies that report an initial sprouting of spared pathways in response to SCIs (Rosenzweig et al., 2010; Darian-Smith et al., 2014; Fisher et al., 2020), and a number of rodent studies (Hill et al., 2001; Bareyre et al., 2004; Ghosh et al., 2010) describe at least some of the elements. For example, Lang et al. (2012) described three phases of CST remodeling after SCI in mice, including growth, exuberant branching, and stabilization of connections (by 12 weeks post-lesion), but they didn't observe a retraction in the timeframe examined. Further work is clearly needed to understand chronic timepoints and species differences.
Retraction bulbs, which can be an indication of pruning (Hill, 2017), were observed in S1 CST terminals in all our DRL/DCL monkeys, and they increased with time post-lesion (Fig. 6B). They have been reported in rodents (Hill et al., 2001) and primate (Freund et al., 2007) SCI studies, but usually with respect to axonal dieback during the acute phase. However, retraction bulbs can arise throughout the recovery process, from the early hours post-injury (dieback), through Wallerian degeneration (weeks later), to the chronic phase of reorganization (Hill, 2017; Kulkarni et al., 2022). Here, their high numbers at the chronic timepoint implicate them in the ongoing refinement of CST connections, a process that clearly outlasts the period of study and may continue for years.
Consolidation of sprouted primary afferent terminals
We have previously shown that spared primary afferents sprout following a sensory SCI (Darian-Smith, 2004). Here, we extended this observation to show that sprouting peaks before 4–5 months post-injury (which correlates with behavioral recovery) and then remains at this level through the first year. The lack of synaptic pruning observed in primary afferents in the dorsal horn in the chronic phase is not surprising, given that they are from the few remaining spared axons and are crucial for relaying sensory information post-injury.
Clinical studies of SCI patients report an early period of intense functional recovery during the first 3 months after injury, followed by a much longer period of more attenuated functional improvement that plateaus at ∼18 months–2 years (Ziegler et al., 2018; Kirshblum et al., 2021). This aligns with the findings here, though the absolute timeframe is species specific. Rodent studies from our lab (McCann et al., 2020) and others (Bareyre et al., 2004; Lang et al., 2012; Granier et al., 2020) report peak CST sprouting in the dorsal horn within the first 3–4 weeks after a partial cervical SCI, with a subsequent maturing of connections between 1 and 3 months post-injury. However, no retraction of rat CST terminals was observed in the dorsal horn at 10 weeks post-injury, and later timepoints were not assessed.
What is clear from the present findings in monkeys, and supported in rodent work, is that the dorsal horn experiences ongoing changes into the chronic period post-injury, with activity-dependent processes (Bradley et al., 2019; Gu et al., 2020) likely driving both short- and longer-term consolidation of connections.
Inflammation
Microglia have long been associated with CNS injury, though our understanding of their exact role in recovery remains incomplete. They are thought to contribute to secondary damage by releasing proinflammatory cytokines, while also conferring protective effects such as debris clearance and provision of trophic support (Bellver-Landete et al., 2019; Kroner and Rosas Almanza, 2019; Brockie et al., 2021; Brennan et al., 2022). Most studies concentrate on the early weeks following injury and involve rodent models, so information about the long-term functionality of microglia in primates is limited.
In agreement with previous studies in NHPs (Nagamoto-Combs et al., 2007, 2010) and humans (Fleming et al., 2006), our results show that microglia increase in number in the initial months following SCI (which likely supports axonal sprouting) and remain elevated above baseline for at least the first post-injury year. The observed decrease in microglial cell size between 4 and 5 months and 1 year, may also reflect a change in function. This aligns with previous work in a TBI model, suggesting that the major inflammatory response subsides beyond 6 months post-injury, giving way to a restorative phenotype (Nagamoto-Combs et al., 2007, 2010). In this study, the coincident reduction of S1 CST terminal boutons and synapses between the 4–5 month and 1-year timepoints suggests a similar role for microglia during this period. That microglia remain elevated so long, and are paired with increased retraction bulbs in the CST, indicates a long period of synaptic pruning of redundant connections and the potential for therapeutic intervention into the chronic phase.
Timelines of recovery
Figure 8 summarizes the approximate timelines (across acute, subacute, sub-chronic, and chronic phases), for the changes identified in the current study, to provide context and to highlight relationships. Importantly, the greatest behavioral recovery occurred during the sub-chronic phase, within 2–3 months after injury (Fisher et al., 2020; Crowley et al., 2021). At the same time, S1 CSTs and spared primary afferents sprouted. Primary afferent sprouting stabilized quickly within the first few months, while the extensive S1 CST sprouting peaked by 4–5 months, after most of the behavioral recovery had occurred, and was then slowly pruned back to baseline by 12 months. Microglia density increased in the early weeks and months, peaked around 4–5 months, and then diminished but remained above baseline at 12 months (within the chronic phase). These findings suggest that fine-tuning continues at least through the first year, and clinical findings suggest it may continue for much longer (Corbetta et al., 2002).
Schematic showing the relative timelines (in the context of recovery phases), for the changes examined over the first year, following the cervical DRL/DCL. The time of injury (O) results in an acute phase of terminal sprouting of both the spared primary afferents and S1 CST inputs to the deafferented region of the spinal gray matter. Spared primary afferent sprouting plateaued quickly (within the first 2–3 months) and tracked with the most significant behavioral recovery. This suggests that new circuitry is largely in place by this time. In contrast, the S1 CST retraction/consolidation process occurred more slowly over at least the first post-injury year and possibly for much longer. This suggests that following even this small lesion (and most likely after a larger, more impactful clinical injury), an extended window of opportunity remains open well into the chronic phase, during which further recovery of function is possible.
Summary
Most studies in SCI patients are conducted in the chronic phase when patients are stable and accessible. In contrast, most animal work focuses on the earliest timepoints, due in large part to cost and the assumption that this is the best window for intervention. To bridge the gap between animal and patient research, a deeper analysis of the chronic phase of recovery in a clinically relevant animal model is needed.
Our findings show several overlapping phases of recovery in the first-year post-injury, beginning with an early period of axonal growth, and followed by a period of consolidation when exuberant growth in the S1 CST is pruned back to the pre-lesion level. This process presumably optimizes the integration of the very limited spared peripheral sensory inputs. The prolonged period of plasticity and continued microglial elevation suggest that potential remains in the chronic phase for further intervention and functional recovery.
Abbreviations
BDA, biotinylated dextran amine; CST, corticospinal tract; CT-B, cholera toxin subunit B; D1–D3, digits 1, 2, and 3; DAB, 3,3′-diaminobenzidine; DCL, dorsal column lesion; DRL, dorsal root lesion; DRL/DCL, dorsal root and dorsal column lesion combined; HRP, horseradish peroxidase; LYD, Lucifer yellow dextran; NeuN, neuronal nuclear antigen; NHP, non-human primate; PBS, phosphate buffered saline; PBS-TX, phosphate buffered saline with Triton X-100; SCI, spinal cord injury; S1, primary somatosensory cortex (includes Brodmann areas 3b/1); S1 CST, somatosensory corticospinal tract; TBI, traumatic brain injury; TBS-TX, Tris-buffered saline with Triton X-100; WLS-GLM, weighted least-square-generalized linear model.
Footnotes
We thank Matt Crowley, Gunnar Felt, and Alayna Lilak for their technical help with the animals and Cholawat Pacharinsak, Sam Baker, and Ben Franco for their primate and veterinary expertise. We also thank Cesar Veloz for the daily care of our monkeys. This work was supported by the National Institute of Neurological Disorders and Stroke (R01 NS048425 and R01 NS091031 to C.D.-S.).
The authors declare no competing financial interests.
- Correspondence should be addressed to Corinna Darian-Smith at cdarian{at}stanford.edu.
