Abstract
The ability of neurons to sense and respond to damage is crucial for maintaining homeostasis and facilitating nervous system repair. For some cell types, notably dorsal root ganglia and retinal ganglion cells, extensive profiling has uncovered a significant transcriptional response to axon injury, which influences survival and regenerative outcomes. In contrast, the injury responses of most supraspinal cell types, which display limited regeneration after spinal damage, remain mostly unknown. In this study, we used single-nuclei sequencing in adult male and female mice to profile the transcriptional responses of diverse supraspinal cell types to spinal injury. Surprisingly, thoracic spinal injury induced only modest changes in gene expression across all populations, including corticospinal tract (CST) neurons. Additionally, CST neurons exhibited minimal response to cervical injury but showed a much stronger reaction to intracortical axotomy, with upregulation of numerous regeneration and apoptosis-related transcripts shared with injured DRG and RGC neurons. Thus, the muted response of CST neurons to spinal injury is linked to the injury's distal location, rather than intrinsic cellular characteristics. More broadly, these findings indicate that a central challenge for enhancing regeneration after a spinal injury is the limited detection of distant injuries and the subsequent modest baseline neuronal response.
Significance Statement
The inability of axons to regenerate after spinal injury limits functional recovery. Efforts to improve regeneration rely on a precise understanding of the baseline transcriptional response to spinal injury. Through single-nuclei sequencing of diverse descending cell types, we find that spinal injury causes only modest changes in gene expression, whereas axon damage close to cell bodies elicits a much larger response. These findings highlight the muted detection of distant injury and the subsequent failure to initiate widespread gene expression changes, as major obstacles to axon regeneration after spinal injury.
Introduction
Axon injury can trigger diverse responses in neurons, ranging from rapid cell death, survival with abortive axonal sprouting, or robust axonal regeneration (Clemente, 1964; Bareyre et al., 2004; Watkins et al., 2013; He and Jin, 2016; Hassannejad et al., 2018; Varadarajan et al., 2022). Underlying these differences are distinct patterns of gene transcription that are triggered as neurons sense and respond to axon damage (Ma and Willis, 2015; Venkatesh and Blackmore, 2017; Zhang et al., 2023). Understanding the various transcriptional responses to injury is crucial for identifying both pro- and antigrowth pathways that can be targeted to enhance regeneration and protect vulnerable cell types. Consequently, a significant focus in regenerative neuroscience has been to clarify the transcriptional reaction of neurons to axon damage (Li et al., 2015; Fink et al., 2017; Dhara et al., 2019; Tran et al., 2019; Poplawski et al., 2020; Renthal et al., 2020; Matson et al., 2022).
For some cell types, particularly sensory neurons of the dorsal root ganglia (DRG; Li et al., 2015; Palmisano et al., 2019; Renthal et al., 2020; Xu et al., 2022) and retinal ganglion cells (RGCs; Dhara et al., 2019; Tran et al., 2019; Jacobi et al., 2022; Tian et al., 2022), extensive profiling efforts have provided deep insight. DRG neurons, which survive axotomy and mount an effective regenerative response, respond to injury with a partial loss of transcripts that distinguish subtypes and the upregulation of numerous regeneration-associated gene (RAG) transcripts that support axon growth. RGCs are highly vulnerable to cell death and display limited spontaneous axon growth after axotomy but can be induced to regenerate long axons using a variety of interventions (Park et al., 2008; Sun et al., 2011; Belin et al., 2015; Norsworthy et al., 2017; Galvao et al., 2018; Feng et al., 2023). Consequently, transcriptional profiling of RGCs has shown a significant upregulation of transcripts associated with both cell death and axon regeneration.
In contrast, very little is known about the genome-wide responses of most supraspinal neurons after spinal injury. Phenotypically, supraspinal populations generally survive spinal axotomy but fail to initiate long-distance axon regeneration, although some spontaneously grow collateral branches above the injury (Bareyre et al., 2004; Fawcett, 2020). Conceptually, it is unclear whether axon extension is constrained by an inability to activate necessary pro-growth genes or if these genes are expressed but countered by emerging growth-suppressive networks. The corticospinal tract (CST), an important regulator of sensation and fine motor control (Starkey et al., 2005; Lemon, 2008; Moreno-López et al., 2016), is among the most studied descending tracts and follows the pattern of spontaneous collateralization but not extension of the main axon (Bareyre et al., 2004; Blackmore et al., 2021). Earlier studies that examined specific transcripts in supraspinal populations, including the CST, showed minimal transcriptional responses to injury, particularly when axotomy occurred far from the cell body (Fernandes et al., 1999; Mason et al., 2003). In contrast, recent profiling of the CST after spinal injury reported a large-scale transcriptional response and a transient reversion to an embryonic state (Poplawski et al., 2020).
In this study, we utilized single-nuclei sequencing in a mouse model to extensively examine the transcriptional response of supraspinal populations to spinal injury. Thoracic injury induced only modest transcriptional changes in any supraspinal population, including the CST, with differentially expressed genes (DEGs) numbering fewer than 10% of the totals typically reported for DRG and RGC neurons. To investigate the effect of cell body-to-axotomy distance, we focused on the CST and found that intracortical injury, but not cervical injury, triggered transcriptional changes on a scale similar to those previously observed in DRG and RGC responses. These data offer comprehensive insight into the response of supraspinal neurons to spinal injury and favor a model in which limited damage response, rather than counteracting transcription or a failure to maintain an initial reversion to a growth state, likely explains the limited axon growth after spinal injury.
Materials and Methods
Animal details
Male and female (8–12 weeks old) wild-type (C57BL/6J, The Jackson Laboratory) mice were used for this study. All procedures involving animals were conducted in strict adherence to ethical guidelines provided by the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health. The Institutional Animal Care and Use Committee at Marquette University reviewed and approved all animal experimental protocols (approval numbers 3283 and 4013). No notable sex-dependent differences were observed during the analysis. Housing conditions for the mice included ad libitum access to food and water, a controlled 24 h cycle of 12 h of light followed by 12 h of darkness, temperature of 22°C ± 2°C, and relative humidity between 40 and 60%.
AAV preparation
rAAV2-retro-CAG-H2B-mGreenLantern (Addgene #177332) was produced at the University of Miami viral core facility at the Miami Project to Cure Paralysis at 1.4 × 1013 particles/ml. rAAV2-retro-H2B-mScarlet (Addgene #191093) and rAAV2-retro-tdTomato (Addgene #59462) were produced at the University of North Carolina Viral Vector Core at 6 × 1012 particles/ml and 3 × 1012 particles/ml, respectively. To create barcoding constructs, a 625 bp sequence corresponding to positions 6903–7525 of human Malat RNA (NR_002819), previously identified as responsible for nuclear localization (Miyagawa et al., 2012), was synthesized by GenScript with a 3′ BamHI and a 5′ SalI-barcode-NotI cassettes (designated Malat-BC). Using rAAV2-H2B-mGl (above), BamHI and SalI were used to replace the mGl-WPRE region with Malat-BC. Barcode sequences of 20 nt (BC0 and BC1) or 100 nt (BC102) were randomly generated and sequenced with flanking SalI and NotI sites, which were used for insertion into the CAG-Malat-BC, generating CAG-Malat-BC0, CAG-Malat-BC1, and BC102 and used for AAV production at the University of Miami Viral Vector Core. All viruses were diluted to 2.5 × 1012 with phosphate-buffered saline (PBS: Sigma-Aldrich P4474, containing 154 mM sodium chloride, 1.058 mM potassium phosphate monobasic, and 2.97 mM sodium phosphate dibasic dihydrate, pH 7.4) immediately prior to injection.
Surgical procedures
All surgeries were performed under ketamine/xylazine anesthesia using adult C57/Bl6 mice 8–12 weeks of age (20–28 g). For AAV injections, mice were mounted on a custom spine stabilizer, and laminectomy was performed at C5 or T10–T12. AAV-retro particles (1 µl, 2.5 × 1012 particles/ml) were injected by a 1701 Hamilton syringe fitted with a pulled glass capillary needle and driven by a Stoelting QSI pump (catalog #53311), guided by a micromanipulator. AAV was injected at a rate of 0.04 μl/min, 0.35 mm lateral to the midline at an initial depth of 0.8 (500 nl), and then raised to 0.6 mm (additional 500 nl).
Thoracic crush injuries were performed as in a previous study (Wang et al., 2022). Mice were mounted on a custom spine stabilization device, laminectomy was performed at thoracic vertebra 10–12, and a forceps with a stopper of 0.15 mm width was used to compress the spinal cord for 15 s, flipped in orientation, and reapplied at the same site for an additional 15 s. For cervical transection, as in our previous work23, mice were mounted in a custom spine stabilizer, and a Vibroknife device was used to produce a transection that extended from slightly left of the midline to the right lateral edge of the spinal cord and to a depth of 0.85 mm. This injury severs dorsal and dorsolateral CST tracts on the right side of the cord, damaging nearly all CST neurons in the left cortex with the exception of approximately 1% of axons that could escape injury by traveling in the ventral columns (Fink et al., 2015) and a very small number of axons that remain uncrossed in some animals and travel ipsilaterally (Steward et al., 2004). In this injury model, we subsequently collected only CST cell nuclei from the left (injured) cortex.
Finally, intracortical injuries were performed with procedures modified from previously published work (Mason et al., 2003). Mice were mounted on a stereotaxic frame in earbars, and the left cortex was exposed by gentle scraping of the skull with a #11 scalpel blade. A 30 G needle was bent at 90° 2 mm from the tip and with stereotactic guidance was placed on the surface of the cortex with the bent tip in the sagittal plane, 0.5 mm to the left of the midline, with the tip aimed caudally. The needle was then lowered 0.6 mm into the brain and rotated 180° clockwise, swinging the tip of the needle beneath the left motor cortex, and then rotated back to the starting position and withdrawn.
Nuclei isolation, library preparation, and sequencing
Following previously published procedures (Beine et al., 2022), mice were isoflurane-anesthetized and decapitated, and brains were rapidly placed in ice-cold slushy artificial cerebrospinal fluid for 1 min. Working quickly on ice, brains were then sectioned in the sagittal (thoracic injury) or coronal (cervical or intracortical injury) plane at 500 μm intervals using Adult Mouse Brain Slicer Matrix on ice (Zivic Instruments BSMAS005-2). Retrogradely labeled populations were rapidly microdissected in ACSF using a stereomicroscope and fluorescence adapter (NIGHTSEA SFA-GR) and immediately frozen on dry ice; collected target regions included corticospinal for all treatments and also hypothalamic, midbrain, pontine, and medullary supraspinal populations for the thoracic injury group. Note that for cervical injury experiments, only the left (injured) CST was collected, corresponding to the transection of the right spinal cord. Samples were then stored at −80°C for up to 1 month until FANS sorting.
Cell nuclei were isolated and sorted on the day of library preparation as previously described (Venkatesh et al., 2021; Beine et al., 2022). Briefly, tissue was dounced in Nuclei EZ Lysis Buffer (Sigma-Aldrich N3408), incubated on ice for 5 min, centrifuged at 500 × g at 4°C for 5 min, resuspended in 4 ml Nuclei EZ Lysis Buffer, incubated on ice, and centrifuged at 500 × at 4°C for 5 min. The resulting pellet was resuspended in 500 µl of nuclei suspension buffer [2% BSA, 40 U/µl RNase inhibitor (Invitrogen Ref. AM2684), in PBS], passed through a 20 µm filter, and moved directly to the FACS machine. The collection tube for the sorted nuclei was coated with 5% BSA and contained the 10x Genomics master mix. Debris and doublets were gated out using side scatter area (SSC-A), side scatter width (SSC-W), forward scatter area (FSC-A), and forward scatter width (FSC-W) and then collected based on the level of fluorescence so that only the brightest were collected. With the above parameters, the FACS machine was set to collect 5,000 nuclei in 5–7 min. The collected nuclei were then prepared into libraries, using Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (PN-1000269), according to the manufacture protocol (10x Genomics, CG000204 Rev D).
Sequencing was performed at the UW-Madison Biotechnology Center using an Illumina NovaSeq 6000, yielding at least 400 reads per library. Files were processed with CellRanger using default parameters to produce a unique molecular identifier matrix for all nuclei-containing droplets. Normalization, dimensionality reduction, and cell clustering were performed using the Seurat package version 5.0. (https://www.nature.com/articles/s41467-022-35574-x#ref-CR18; Butler et al., 2018). Cell clusters were identified by performing principal component analysis on the top 30 principal components, with significant components determined by an elbow plot and jackstraw test. Cells were clustered using the Louvain method in the FindClusters function (resolution, 0.05) and visualized using Uniform Manifold Approximation and Projection (UMAP). For injured versus uninjured data comparison, the “merge” and “FindIntegrationAnchors” functions from Seurat were used to integrate the datasets. Differential expression analysis was conducted using the “FindMarkers” function in Seurat, applying a nonparametric Wilcoxon rank sum test. Cluster plots in Seurat were generated using the “scCustomize” package in R (https://github.com/samuel-marsh/scCustomize), while MA plots, histograms, and heatmaps were created using the “ggplot2” package (https://ggplot2.tidyverse.org/). We have utilized ShinyCell (Ouyang et al., 2021) in R to provide prebuilt clusters of the analyzed scRNA-seq data, enabling direct visualization and interactive exploration. The clustered data are publicly accessible and can be explored through our dedicated webpage at https://venkateshlab.github.io/CNS_response/.
Gene ontology and network analysis
Gene ontology analysis was conducted using the “ClusterProfiler” (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3339379/) package in R (Yu et al., 2012; Wu et al., 2021), with input genes showing a fold change value of ≥1 or less than or equal to −1 and an adjusted p-value of <0.05, marking them as upregulated and downregulated, respectively. The “org.Mm.eg.db” database served as the reference for ontology searches. The analysis included biological process, cellular component, and molecular function categories, setting gene count size limits between 3 and 500, and a p-value cutoff of <0.05. The top 20 pathways were visualized using the dot plot function from the enrich plot package in R, and the results were saved as a CSV file for network visualization in Cytoscape (3.10; Shannon et al., 2003; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC403769/).
Tissue clearing and imaging
Three weeks after cervical AAV injection and 1–14 d after sham or intracortical injury, animals underwent transcardial perfusion with 0.9% saline and 4% paraformaldehyde (PFA) solutions in 1× PBS (15710, Electron Microscopy Sciences). Whole brains were dissected and fixed overnight in 4% PFA at 4°C and washed three times in PBS pH 7.4, and the brains were cleared using a modified version of the 3DISCO (Soderblom et al., 2015; Wang et al., 2018). Whole mouse brains were incubated on a rotating shaker at room temperature in 50% and 80% and twice with 100% peroxide-free tetrahydrofuran (Sigma-Aldrich, 401757) for 12 h each for a total of 2 d. Samples were transferred to BABB solution [1:2 ratio of benzyl alcohol (Sigma-Aldrich, 305197) and benzyl benzoate (Sigma-Aldrich, B6630)] for 3 h. After clearing, whole brains were imaged the same using a high-speed confocal, Andor Dragonfly 202-2540 (Oxford Instruments), based on a Leica microscope. We used a 10× Plan Apo, NA 0.45, and W.D 2.8. The laser was a solid-state 488 nm diode laser at 150 mW. Laser power was set at 15%, and exposure time for each plane was between 60 ms. The camera used was a Sona sCMOS 4.2B-6 set at an ROI size (W × H) of 2,048 × 2,048. All images were stitched using Imaris Stitcher Vx64 10 (Oxford Instruments). After stitching, 3D rendering and spot detection were performed using Imaris 10.01. The number of spots detected in cortical tissue dorsal to the region of intracortical injury was quantified and then normalized to an identically sized region in the contralateral (uninjured) cortex.
Tissue processing and immunohistochemistry
For assessment of spinal injuries, at the time of brain microdissection, the spinal cord was rapidly dissected and immersed in 4% PFA in 1× PBS (15710, Electron Microscopy Sciences). For assessment of phospho-c-JUN and ATF3 expression, animals were perfused with 4% PFA, after which brains and spinal cords were removed and postfixed overnight in 4% PFA at 4°C. Spinal cords and cortical tissues were embedded in 12% gelatin in PBS and cut via vibratome to yield 100 μm sections. Sections were incubated overnight with primary antibodies GFAP (Dako Z0334, 1:500, RRID:AB_10013482), p-cJUN (Cell Signaling 9261, 1:500, RRID:AB_2130162), ATF3 (Atlas Antibodies, catalog #HPA001562, RRID:AB_1078233), or Iba1 (Wako 019-19741 1:500, RRID: AB_839504) and rinsed and then incubated for 2 h with appropriate Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific R37117, 1:500). Fluorescent images were acquired using a Nikon A1R+ laser scanning confocal microscopy system on a Nikon Ti2-E inverted microscope. For quantification of cJUN and ATF3, ROIs were first manually drawn in NIS-Elements software using the mScarlet label within retrogradely transduced CST cell bodies. The average pixel intensity of the detection channel was then determined for each ROI. All visible CST cell bodies from at least two replicate cortical sections for each animal were selected by observers blinded to treatment.
Data availability
There are no restrictions on data availability. All the datasets generated and analyzed during the current study are available at Figshare: https://doi.org/10.6084/m9.figshare.c.7238734.v1. The clustered data are publicly accessible and can be explored through our dedicated webpage at https://venkateshlab.github.io/CNS_response/.
Code availability
There was minimal custom code development, and all software used in data analyses are previously published, open access, and have been cited under the relevant methods section. Links to relevant software repositories/documentation are listed as follows:
Cytoscape: https://cytoscape.org/
CellRanger: https://github.com/10XGenomics/cellranger
Seurat: https://github.com/satijalab/seurat;
clusterProfiler: https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
ggplot: https://ggplot2.tidyverse.org/
scCustomize: https://github.com/samuel-marsh/scCustomize
Manual plots: https://github.com/RegenerationLab/Supraspinal_Injury
scRNA cluster: https://venkateshlab.github.io/CNS_response/
ShinyCell: https://github.com/SGDDNB/ShinyCell
Experimental design and statistical analyses
Statistical difference in gene expression was determined by nonparametric Wilcoxon rank sum tests performed within Seurat. For quantification of immunofluorescence, experimental replicates are indicated in figure legends, and significant differences were tested by one-way ANOVA with post hoc Sidak's using Prism software (GraphPad version 10.2.3). The number of experimental replicates and animals pooled per replicate are indicated for each sequencing experiment in the relevant Results section. Statistical details and replicate numbers for each experiment can be found in the figure legends. All quantification of images was conducted by experimenters blind to the experimental group.
Results
Profiling supraspinal transcriptional responses to thoracic injury
We used retrograde labeling, cell nuclei isolation, and single-nuclei analyses to examine gene expression in supraspinal populations after spinal injury. Cell nuclei of supraspinal neurons were fluorescently labeled by injection of AAV2-retro-H2B-mGl into the lumbar spinal cord of adult mice, which we previously found to produce a bright, nuclear-localized signal in tens of thousands of supraspinal neurons throughout the brain (Beine et al., 2022; Wang et al., 2022; Fig. 1A). One week later, following successful viral transduction of supraspinal neurons, we performed a severe crush of the spinal cord at thoracic level 10 (T10), using a method that we showed previously to result in complete interruption of descending axon tracts (Wang et al., 2022). After another week, fluorescently labeled brain regions were microdissected and flash-frozen. Supraspinal nuclei were purified by fluorescence-activated nuclei sorting (FANS) and single-nuclei libraries prepared on a 10x Genomics platform with three animals pooled in each of three replicate samples. Spinal tissue was collected at the time of dissection and subjected to GFAP immunohistochemistry. Examination confirmed complete interruption of axon tracts and the absence of any astrocytic tissue bridges across the injury site (Fig. 1A, inset).
Single-nuclei analyses detect minimal changes in gene expression in supraspinal neurons following thoracic spinal injury. A, Experimental design including retrograde labeling of supraspinal neurons, thoracic spinal injury, and fluorescent-activated nuclei sorting 1 week postinjury. The inset shows GFAP verification of spinal injury. B, UMAP visualization of uninjured (blue) and injured (orange) nuclei, both of which cluster into six main cell types: corticospinal tract (CST), dorsal pons (DP), red nucleus (RN) neurons, medulla (MED) neurons, hypothalamus (HYP) neurons, and “other” cell types. C, A histogram indicating the number of differentially expressed genes in injured versus uninjured CST neurons. D, MA plot showing the relationship between transcript abundance and differential gene expression in injured versus uninjured CST neurons, with significant up- and downregulation indicated by green and red, respectively (nonparametric Wilcoxon rank sum test, p < 0.05). E, Creb5 and (F) Atf3 gene expression across five cell types under uninjured and injured states, indicating varying responses to injury. G, A heatmap indicating the identity of the top twenty up- and downregulated transcripts in injured CST neurons. Extended Data Figure 1-1 provides the full set of differentially expressed genes for all cell types.
Figure 1-1
Full list of fold change and significance for all detected transcripts in supraspinal populations one week after injury to the thoracic spinal cord. Download Figure 1-1, XLSX file.
To identify injury-sensitive transcripts, we used a Seurat-based pipeline to merge thoracic-injured samples with uninjured control samples from our prior work (Beine et al., 2022). The timing of viral delivery and methods of dissection, FANS purification, and library preparation were identical between uninjured and injured samples. UMAP clustering of the merged datasets revealed 16 distinct clusters, which were assigned supraspinal cell identity based on our previously established markers (Beine et al., 2022; Fig. 2A,B). These clusters were then grouped into five main brain regions: corticospinal (CST), hypothalamus (HYP), red nucleus (RN), dorsal pons (DP), and medullary neurons (MED, predominantly reticular; Fig. 2C). Notably, injured cells were found in the same UMAP clusters as uninjured cells (Fig. 1B) and continued to express cell-specific transcripts (Fig. 2B). For example, injured CST neurons clustered separately from all subcortical cell types, intermingled with uninjured CST neurons, and remained readily identifiable by markers such as Crym (Fink et al., 2015; Fig. 2B). Thus, unlike in peripherally injured DRG neurons (Renthal et al., 2020), axotomy does not trigger widespread dedifferentiation in supraspinal neurons.
Supraspinal neurons maintain subtype-specific gene expression and display only modest transcriptional changes after thoracic injury. A, UMAP clustering of combined injured and uninjured samples shows 16 discrete supralumbar cell types. B, A dot plot shows specific expression of marker genes within the identified clusters and with similar levels of expression in injured and uninjured samples. C, MA plots show changes in gene expression within various brain regions and cell types following thoracic injury (pooled). These regions include the corticospinal tract (CST), dorsal pons (DP), hypothalamus (HYP), red nuclei (RN), and medulla (MED), as well as a composite overview of all regions (ALL). Upregulated genes are marked in green and downregulated genes in red, and the total counts of differentially expressed genes are indicated for each plot, with significance established at a p-value of <0.05 via a nonparametric Wilcoxon rank sum test.
Indeed, supraspinal neurons typically showed limited transcriptional responses to thoracic injury. When all nuclei were pooled for a global comparison between injured and uninjured neurons, only 147 transcripts changed significantly above a threshold of twofold (log2 fold change ±1; 41 increasing, 106 decreasing; Fig. 2C). Furthermore, ∼90% of the 147 DEGs came from transcripts detected in fewer than 10% of all nuclei, thus minimizing the effect on the overall population. In contrast, many hundreds or thousands of transcripts typically surpass the same threshold following axotomy in other neuronal cell types (Fig. 10; Dhara et al., 2019; Tran et al., 2019; Renthal et al., 2020; Xu et al., 2022). Comparisons performed population by population also revealed few injury-regulated transcripts (Fig. 2C). Of all supraspinal types, CST neurons showed the largest number of transcripts significantly changed, with 45 and 151 up- and downregulated more than twofold, respectively (Fig. 1C,D). As in the all-neuron analysis, nearly all the CST DEGs were low-abundance transcripts (Fig. 1D). Gene ontology (GO) enrichment analysis of up- and downregulated gene sets revealed no overrepresented terms, possibly due to the limited size of the input lists. Interestingly, Creb5, Hrk, and Dusp16 emerged as some of the most upregulated transcripts, all of which have been previously associated with axon damage (Fernandes et al., 2013; Yao et al., 2023; DeVault et al., 2024; Fig. 1E,G). On the other hand, other transcripts typical of axotomized or otherwise stressed cells, such as Atf3, Ddit3, Ecel1, and Casp3, were not upregulated (Fig. 1G). Likewise, transcripts associated with neural differentiation, including Sox2 and Ascl1 (Poplawski et al., 2020), remained unaffected by injury (all thoracic injury DEGs are available in Extended Data Fig. 1-1). Examination of the list of downregulated transcripts showed several markers of neural activity (Npas4, Arc, and Egr1) but no transcripts linked to neural differentiation or axon growth (Fig. 1G, Extended Data Fig. 1-1). In summary, profiling following thoracic injury revealed a transcriptional response characterized by the activation of a limited set of damage-responsive transcripts and a decrease in several activity-dependent transcripts. However, there was no widespread activation of regenerated-associated genes or transcripts associated with cellular stress, embryonic development, or axon growth.
The response of CST neurons to cervical axotomy
A potential explanation for the muted transcriptional response to thoracic injury could be the significant distance between the injury site and the originating cell bodies (Fernandes et al., 1999; Mason et al., 2003). To test if a closer injury site elicits a more pronounced response, we moved the injury site to the cervical spinal cord. This experiment focused exclusively on CST neurons, which were selected based on their abundance, functional importance, and the feasibility of transecting nearly all descending CST axons via dorsal hemisection while maintaining animal survival. Neurons were retrogradely labeled by C5 injection of AAV2-retro-H2B-mGl followed by unilateral dorsal hemisection at the C4/5 level. One week later, layer V cortical tissue was microdissected and flash-frozen, replicate libraries were prepared from FANS-purified CST nuclei from three pooled animals, and data from injured and uninjured libraries were merged and analyzed in Seurat. Spinal tissue from injured animals was imaged with GFAP immunohistochemistry to confirm complete transection of the right CST (Fig. 3A, inset). In an initial check of cell purity, small clusters were identified that expressed markers for superficial neurons (Cux2) or intracortically projecting neurons (Slc30a2); these comprised 1.7% of total nuclei and were removed (Zhang et al., 2021; Beine et al., 2022). The remaining cluster contained a mixture of nuclei from injured and uninjured samples, both of which expressed similar levels of CST and layer V markers Crym, Bcl11b, and Fezf2 (Fig. 3B,C). These data indicate high purity of CST nuclei and, similarly to the thoracic injury, an absence of injury-triggered dedifferentiation that was reported in axotomized DRG neurons (Renthal et al., 2020). Consistent with this, only 32 DEGs exceeded a twofold threshold (Fig. 3D,E). Although the list was small, it showed a highly significant overlap with the set of thoracic DEGs (46.8%, 32.2-fold enrichment, p < 7.409 × 10−20, hypergeometric test) including shared upregulation of Creb5, Hrk, and Dusp16 and shared downregulation of Npas4, Arc, and Egr1 (Fig. 3E,F). This consistency across different injury locations enhances confidence in the snRNA-seq method's ability to detect a shared biological signal, albeit small. However, even with axotomy moved to the cervical level, it still produced limited transcriptional changes in CST neurons.
Single-nuclei profiling detects minimal responses in CST neurons to cervical spinal injury. A, Experimental design: CST neurons were retrogradely labeled with nuclear-localized H2B-mGl, the right spinal cord was transected or left uninjured, and 1 week later nuclei were harvested for single-nuclei sequencing. Inset shows GFAP signal, verifying injury completeness. B, UMAP visualization shows a predominant overlap of injured and uninjured nuclei, with the exception of one injured-enriched group (dotted circle). C, Feature plots show expression of Crym, Bcl11b, and Fezf2, verifying CST identity, and higher levels of Creb5 in the injured-enriched region (dotted line). D, A histogram indicating the number of differentially expressed genes in injured versus uninjured CST neurons. E, MA plot showing the relationship between transcript abundance and differential gene expression in injured versus uninjured CST neurons. In both D and E, transcripts that are significantly up- and downregulated are colored green and red, respectively (nonparametric Wilcoxon rank sum test, p < 0.05). F, A heatmap indicating the identity of the top 20 up- and downregulated transcripts in CST neurons 1 week after cervical spinal injury. Extended Data Figure 3-1 provides the full set of differentially expressed genes for cervically injured CST neurons.
Figure 3-1
Full list of fold change and significance for all detected transcripts in corticospinal tract neurons after cervical spinal injury or intracortical injury. Download Figure 3-1, XLSX file.
We next explored the idea that while CST neurons as a collective exhibit a modest response to injury, there might be subpopulations within them that show greater responsiveness. Indeed, although the injured and uninjured samples mostly intermingled in UMAP clustering, closer examination revealed a region of nonoverlap that was comprised almost entirely of injured nuclei. Interestingly, the expression of Creb5, previously associated with neuronal injury responses (Yao et al., 2023), was highly concentrated in this region (Fig. 3C, dotted circle). When injured nuclei within this region were compared separately to uninjured neurons they displayed somewhat larger differences, with 210 transcripts exceeding a threshold of ±2-fold change and >551 at a relaxed threshold of ±50% (±0.58 log2FC; Fig. 4). We, therefore, designated injured nuclei within this region, which comprised 19.1% of all injured nuclei, as CST-high Creb5 (CST-hCreb5). Among the genes significantly upregulated in this subgroup were Bbc3 (PUMA)/Casp3, Ddit3 (CHOP), Eif2ak3, and Trib3, well-known markers of ER stress. Transcripts downregulated in the CST-hCreb5 cluster were enriched for synaptic functions, consistent with prior work in other cell types linking axon injury to altered synaptic function (Tedeschi et al., 2016; Kiyoshi and Tedeschi, 2020; Hilton et al., 2022; Fig. 4; all GO terms and associated genes are available in Extended Data Fig. 4-1). Thus, after cervical axotomy, a subset of CST neurons initiate a damage response that includes some activation of ER stress pathways and downregulation of synapse function. It should be emphasized that fewer than 20% of CST neurons displayed this response. In addition, this subset lacked other markers of cellular stress, notably Atf3, and did not upregulate regeneration-associated transcripts (Ma and Willis, 2015) such as Sox11, Tuba1a, Klf6, Gap43, and others (Extended Data Fig. 3-1). Overall, although some signs of cellular stress can be detected in a subset of CST neurons, cervically injured CST neurons did not show widespread transcriptional changes or activation of axonal growth pathways.
A subset of cervically injured CST neurons display elevated transcriptional responses. A, UMAP plots of CST nuclei that were uninjured (purple), injured, and expressing low levels of Creb (light blue) or injured and expressing higher levels of Creb (orange). B, MA plots showing upregulated (green) and downregulated (red) transcripts at a log2 fold change threshold of ±0.5 and p-value of <0.5 (nonparametric Wilcoxon rank sum test). C, Gene ontology (GO) enrichment analysis of genes differentially expressed between uninjured and high-Creb5 nuclei identify terms associated with ER stress, apoptosis, and axon regeneration in upregulated transcripts and terms related to synaptic function in downregulated transcripts. Extended Data Figure 4-1 provides the full set of statistically enriched GO terms.
Figure 4-1
Full list of GO terms that are statistically enriched in up- or down-regulated genes in the high-Creb5 subset of corticospinal tract neurons one week injury to the cervical spinal cord. Download Figure 4-1, XLSX file.
Intracortical injury, but not spinal injury, upregulates ATF3 and phospho-c-JUN in CST neurons
DRG and RGC neurons, which display a much larger transcriptional response to axotomy than what we detected after spinal injury in CST neurons, both undergo a retrograde stress response that is marked by rapid phosphorylation of cJUN and upregulation of ATF3 protein (Tsujino et al., 2000). We therefore compared these indicators in CST neurons at 1 and 7 d after cervical axotomy to sham-injured controls. ATF3 was undetectable in any CST neuron at either time point (Fig. 5E,F), and phospho-c-JUN was barely visible in most CST neurons (Fig. 5D,E). We noted, however, that at 7 d postinjury some CST neurons showed faint phospho-c-JUN signal. Quantification of signal intensity showed no significant difference between cervically injured and sham-injured controls (p > 0.05, one-way ANOVA with post hoc Sidak's), but a small population with elevated phospho-c-JUN signal can be seen in the scatterplot at 7 d (Fig. 5D). In summary, in line with the snRNA-seq findings, detection of phospho-c-JUN and ATF3 suggests a minimal cell body response to cervical axotomy in CST neurons.
Intracortical but not spinal axon injury causes phosphorylation of cJUN and upregulation of ATF3 in CST neurons. A, Experimental procedure: CST neurons were retrogradely labeled by AAV2-retro-mGl and then injured in the cervical spinal cord or within the cortex. B, A coronal section of cortex 1 week after intracortical injury with injury (arrow, dotted line) and nearby expression of p-cJUN, including in CST neurons (mGl, green). C, Immunohistochemistry for p-cJUN 1 or 7 d postinjury. D, Quantification of p-cJUN shows significant elevation after intracortical injury but not after spinal injury. E, Immunohistochemistry for ATF3 (red) in CST neurons (green) 1 or 7 d postinjury. F, Quantification of ATF3 signal shows significant elevation 7 d after intracortical injury. Scale bars: 500 µm (B), 200 µm (top panels C and E), 10 µm (bottom panels C and E). ***p < 0.001, ANOVA with post hoc Sidak's. N = 4 animals per group, 100 neurons quantified per animal.
Does this subdued response reflect an inherent cellular difference in CST neurons, or does it arise from differences in the anatomical proximity to the site of injury? To distinguish these possibilities, we employed an intracortical injury model to sever descending CST axons within 1 mm of their cell bodies and proximal to any collateral synapse formation (Fig. 5A,B). In striking contrast to the prior spinal injury, phospho-c-JUN was strongly upregulated at both 1 and 7 d postlesion, indicating rapid engagement of a retrograde stress response (Fig. 5C,D). In addition, proximal injury resulted in strong expression of ATF3 by 7 d postlesion (Fig. 5E,F; p < 0.001, ANOVA with post hoc Sidak's). Importantly, all tissue processing and IHC procedures were performed simultaneously including both spinally and intracortically injured tissue, thereby strengthening the conclusion of a large difference in expression between the two conditions. Immunohistochemistry for Iba1 showed high microglial reactivity in the vicinity of the intracortical injury, including in close proximity to CST cell bodies, indicating a local inflammatory response (Fig. 6B–D). We also noticed that by 7 d after intracortical injury, the number of retrogradely labeled CST neurons appeared diminished (Fig. 5C, compare 1 and 7 d panels; Fig. 6B, compare the injured and uninjured cortex). This visual impression was confirmed in a second cohort of animals in which CST neurons were imaged by tissue clearing and 3D imaging, showing a significant reduction in cell number in the weeks following intracortical injury (Fig. 6E,F). Thus, reminiscent of the RGC response to optic nerve injury and consistent with prior observations of CST cell loss after intracranial axotomy (Giehl and Tetzlaff, 1996), CST neurons appear to undergo significant cell death after proximal axotomy. Taken together, these data indicate that axotomies located near CST cell bodies and proximal to collateral contacts with brain targets (Sinopoulou et al., 2022) result in strong activation of cellular stress pathways and substantial cell death. These data rule out the possibility of cell-type insensitivity as the reason for the limited activation of phospho-c-Jun and Atf3 in CST neurons after spinal injury and instead indicate injury location as a determining factor.
Intracortical axon injury causes inflammation and loss of CST neurons. A, Experimental design: CST cell nuclei were retrogradely labeled by cervical injection of AAV2-retro-H2B-mGl followed by intracortical injury and immunohistochemistry or whole-brain imaging after 7 or 14 d. B, A coronal section of mouse cortex 1 week after intracortical injury shows elevated Iba1 signal near the site of intracortical injury (dotted line) compared with the contralateral, uninjured cortex. C, D, Higher magnification views shows numerous Iba1-reactive microglia in close proximity to CST neurons (green) after intracortical injury, but not in the contralateral and uninjured cortex. E, Dorsal views of the cortex after whole-brain clearing with CST cell nuclei labeled in green and approximate injury location shown in red. C, Quantification of detected CST nuclei overlying the area of cortical injury, normalized to the number detected in the corresponding area in the contralateral cortex, reveals a decline of approximately 40% by 1 week and 85% by 2 weeks. N = 3 animals per time point. **p < 0.01, ANOVA with post hoc Sidak's. Scale bars in B are 3 mm (top), 1 mm (middle), and 30 µm (bottom).
CST neurons show large transcriptional responses to intracortical injury
Because CST neurons upregulate damage-responsive phospho-c-JUN and ATF3 after intracortical injury, we tested for a larger genome-wide response using snRNA-seq. As previously, CST cell nuclei were retrogradely labeled from the cervical spinal cord, and 7 d later, a sharp blade was passed through subcortical white matter, severing CST axons approximately 1 mm from their cell bodies (Fig. 7A). At 1, 3, or 7 d after injury, layer V cortex was microdissected under fluorescence, taking tissue only from brain sections in which visual inspection confirmed an injury located deep to layer V. These experiments also employed DNA barcoding, delivered using AAV to CST neurons along with the nuclear reporter, which allowed pooling of injured and uninjured tissue and subsequent deconvolution. In UMAP plots, unlike the prior thoracic and cervical injury responses, proximally injured CST nuclei clustered separately from sham-injured nuclei and showed strong and near-universal upregulation of Creb5 (Fig. 7B,C). Across the three time points, >3,000 DEGs significantly exceeded a twofold threshold at either 1, 3, or 7 d (Extended Data Fig. 3-1). Moreover, unlike in the prior spinal injury datasets, these DEGs included many transcripts that were detected in a significant proportion of nuclei, indicating broad population-level participation in the transcriptional response (Fig. 7D). Gene ontology analyses of upregulated transcripts revealed enrichment of terms associated with wound healing, various inflammatory processes, adhesion, and actin cytoskeleton remodeling (Fig. 7E). Conversely, downregulated transcripts were enriched for terms associated with synapse organization, membrane potential, and mitochondrial function (Fig. 7F; all GO terms and associated genes are presented in Extended Data Fig. 7-1). Comparison of our data to a prior in situ hybridization-based characterization of CST gene expression after intracortical injury showed perfect correspondence across eight transcripts (Atf3, Jun, Gap43, L1cam, Chl1, and Stmn2 strongly upregulated in both, Basp1/Cap-23 and Krox-24/Egr1 upregulated in neither; Mason et al., 2003). In summary, the large number of DEGs detected after intracortical injury confirms the sensitivity of the snRNA-seq approach in detecting transcriptional changes. This reaffirms the notion that the subdued response seen after spinal injury is indicative of a genuine biological signal rather than technical limitations.
Single-nuclei profiling detects a large transcriptional response to intracortical injury in CST neurons. A, Experimental design in which CST neurons were retrogradely labeled by AAV2-retro-H2B-mGl injection, axotomized in the lower cortex or left uninjured, and harvested for single-nuclei profiling 1, 3, or 7 d later. B, UMAP clustering shows that injured samples at various postinjury intervals (1, 3, and 7 d) segregate from one another and from uninjured samples. C, A feature plot showing elevated levels of Creb5 in intracortically injured CST neurons. D, Histograms and MA plots show numerous transcripts significantly upregulated (green) and downregulated (red) in CST neurons after intracortical injury (p < 0.05, nonparametric Wilcoxon rank sum test). E, Gene ontology enrichment analysis indicating biological processes that are significantly overrepresented in CST-upregulated gene sets at 1, 3, or 7 d postinjury (p < 0.05, Benjamini–Hochberg test). F, A gene ontology and pathway interaction network shows gene ontologies and pathways affected by intracortical injury, with the intensity of the node fill color representing the timeline [1 DPI (violet), 3 DPI (green), and 7 DPI (yellow); p < 0.05, Benjamini–Hochberg test]. Extended Data Figure 7-1 provides the full set of statistically enriched GO terms and Extended Data Figure 3-1 provides the full set of differentially expressed genes.
Figure 7-1
Full list of GO terms that are statistically enriched in up- or down-regulated genes in corticospinal tract neurons one, three, or seven days after intracortical axon transection. Download Figure 7-1, XLSX file.
Examination of stress-associated transcripts illustrates this phenomenon. CST-IC cells strongly upregulated ER stress markers that were previously detected in the cervically injured CST-hCreb5 subset, notably Bbc3 (PUMA), Casp3, Ddit3 (CHOP), Eif2ak3, and Trib3 (Extended Data Fig. 3-1). Besides these, intracortical injury caused strong upregulation of additional stress and apoptosis markers that were not affected by cervical injury, including Atf3, Atf5, Pp1r15a, Gadd45g, Bcl2l11/BIM, Casp9, and Bcl6. Thus, spinal injury appears to trigger only partial activation of the stress response; we emphasize again that only a subset of ∼20% of cervically injured neurons display even this partial response. Intracortical injury also caused upregulation of well-known regeneration-associated genes, including transcription factors Sox11, Stat3, Klf6, and Smad1, and other RAGs such as Gap43, L1cam, Sprr1a, Stmn2/SCG10, Gal, Inppk5, and Itga9 (Fig. 7D, Extended Data Fig. 3-1). Of these, only Itga9 was significantly upregulated after cervical spinal injury, and again only in the hCreb5 subset.
The difference in scale raises the question of whether the responses are distinct at the molecular level or if they stem from a shared axotomy program that is activated less intensely by distant injury. The former model predicts minimal overlap between spinal and intracortical injury DEGs, at the level of chance, while the latter predicts DEGs that are shared by both injuries but with fold changes that are larger in the intracortical set. To distinguish these possibilities, we compared significantly regulated transcripts, using a relaxed 25% threshold (log2FC > 0.32) to sensitize detection. Of the 137 transcripts upregulated >25% in cervically injured CST neurons, 64% were also upregulated by intracortical injury, a significant enrichment (p < 0.001, hypergeometric test; Fig. 8A). Overlap was even larger for the 317 transcripts upregulated in the hCreb5 group, with 78% upregulated by intracortical injury (p < 0.001, hypergeometric test). Fold change values in the common upregulated genes were positively correlated (R2 = 0.4 and 0.6 for all CST and hCreb5 CST, respectively) with a slope significantly <1 (Δcervical / Δintracortical), indicating larger changes in the intracortical than spinal injury (Fig. 8B). Consistent with this, the intracortical fold change was larger than the spinal injury change in 97 and 85% of transcripts shared with the all-CST or hCreb subset, respectively. Overall, these data favor the existence of an axotomy response that is shared but differs in the degree of activation, with spinal injury causing smaller changes in a subset of genes comprising ∼10% of intracortical-responsive genes.
Transcripts upregulated by spinal injury mostly overlap with intracortical-upregulated transcripts but show a lower fold change. A, Comparison of transcripts significantly increased >25% (log2FC > 0.32) in CST neurons after spinal injury to transcripts increased after intracortical injury shows significant overlap (64%, p < 0.0001, hypergeometric). B, A similar comparison using only the stronger-responding subset of cervically injured CST neurons (CST-hCreb5) shows 79% overlap (p < 0.0001, hypergeometric). C, D, The fold change values in commonly upregulated transcripts are positively correlated with a slope significantly <1 (p < 0.001, F-test), indicating larger fold changes after intracortical than cervical injury.
Finally, because these analyses relied on distinct sequencing libraries, we performed an additional experiment to ensure that variability in injury timing, library preparation, or sequencing was not driving the difference in DEG detection between SCI and IC samples. To enable pooling of different injury conditions in the same library, CST neurons in adult mice were retrogradely labeled by cervical injection of H2B-mGl combined with one of three distinct DNA barcodes. One week later, the barcoded subgroups were subjected to cervical hemisection and intracortical injury or remained uninjured, followed 3 d later by microdissection and flash freezing. Tissue from all three groups was then pooled such that nuclei isolation, FANS sorting, library preparation, and sequencing were identical. In the resulting data, 94.2% of nuclei could be classified by barcode, with the remainder excluded for low or conflicting barcoding reads. The IC sample segregated from the SCI and uninjured nuclei in UMAP clustering (Fig. 9A) and in feature plots displayed strong elevation of stress and regeneration-associated genes (RAGs) including Creb5, Atf3, Sox11, Stat3, Klf6, and Sprr1a compared with both uninjured and SCI samples (Fig. 9B). Indeed, although SCI displayed only 33 DEGs compared with uninjured at a twofold threshold, the IC sample had 1,846 DEGs (Fig. 9C,D; Extended Data Fig. 9-1). Moreover, consistent with the prior IC datasets these DEGs included numerous additional RAGs including Tubb2b, L1cam, and Gap43 (Fig. 9D,E, Extended Data Fig. 9-1). In contrast, SCI nuclei did not exhibit significant upregulation of most RAGs. Thus, under identical conditions of FANS, library preparation, and sequencing, snRNA-seq detects a robust transcriptional response to IC injury but a muted response—approximately 50-fold lower—to SCI. Overall, these data show that when CST neurons experience a nearby injury stimulus, they initiate a strong transcriptional response including upregulation of transcripts associated with axon regeneration. However, this response is largely absent following more distal axon injury.
Matched within-library profiling confirms pronounced transcriptional changes in CST neurons after intracortical but not spinal injury. Animals received retrograde nuclear labeling along with a molecular barcode followed by no injury, spinal injury, or intracortical injury. Three days later, tissue was pooled for library preparation, followed by postsequencing deconvolution. A, A UMAP plot shows the distribution of nuclei from uninjured (purple), cervical spinal cord injured (SCI, blue), and intracortical injury (orange) samples. B, Feature plots illustrating strongly elevated expression of stress and regeneration-associated transcripts (Creb5, Atf3, Sprr1a, Sox11, Stat3, Klf6) in CST neurons after intracortical injury. C, D, MA plots show fewer upregulated (green) and downregulated (red) in CST neurons that received spinal injury (C) compared with intracortical injury (D; p < 0.05, nonparametric Wilcoxon rank sum test). E, A heatmap compares the log fold change in gene expression between spinal and intracortical injury, with a scale from green (upregulation) to red (downregulation). Bottom, Bar graphs comparing the log fold change of specific growth-associated genes between intracortical (IC) and cervical SCI conditions. Extended Data Figure 9-1 provides the full set of differentially expressed genes.
Figure 9-1
Full list of fold change and significance for all detected transcripts in in a pooled library of spinally and intracortically injured corticospinal tract neurons. Download Figure 9-1, XLSX file.
Proximally injured CST neurons share gene changes with injured RGC and DRG neurons
We next examined the extent to which the response of CST neurons to nearby axotomy resembles that of RGC or DRG neurons. To match our data, we assembled sets of transcripts that are significantly up- and downregulated at a twofold threshold within the first week after axon injury from optic nerve crush (Tran et al., 2019; RGC) or spinal axotomy (DRG; Renthal et al., 2020). The intersection of the datasets showed that of 2,285 transcripts upregulated by CST neurons between 1 and 7 d postinjury, 569 and 329 were shared with axotomy-upregulated transcripts in DRG or RGC neurons, respectively, significantly more than chance (p < 3.495 × 10−106 and p < 2.670 × 10−92, hypergeometric test; Fig. 10A, Extended Data Fig. 10-1). Notably, despite the differences in cell types, injuries, and experimental techniques, 145 transcripts were upregulated in all three datasets, >10 times the number predicted by chance. Thus, although many axotomy-upregulated transcripts are unique to the cell type and injury, there also exists a shared set of transcripts that represent a common response to injury.
Key growth-related genes are activated in response to injuries across various cell types. A, A Venn diagram shows the overlap of differentially expressed genes in CST neurons after intracortical injury (IC), dorsal root ganglia (DRG) after spinal injury, and retinal ganglion cells (RGC) after optic nerve crush, indicating unique and shared transcriptional responses. B, Interaction network (STRING) networks of the genes shared among the three injury subtypes reveal five distinct functional categories: (1) cellular stress and axon growth, (2) actin interaction, (3) tRNA synthesis, (4) one-carbon metabolism, and (5) amino acid import. Each network node represents a gene, with edges indicating interactions or functional associations of all the 145 genes found in all three groups. C, A heatmap displaying the gene expression changes across different pathways in the DRG, RGC, and CST after intracortical (IC) or spinal injury (SCI). The color scale reflects the fold change from downregulation (red) to upregulation (green). Specific genes of interest are annotated on the side, highlighting their expression patterns across the conditions. Extended Data Figure 10-1 provides the full set of differentially expressed genes across cell types, and Extended Data Figure 10-2 provides manually curated evidence to link common injury-responsive genes to axon growth and/or cell death.
Figure 10-1
Full gene sets that compare the transcriptional responses to axons injury across corticospinal tract neurons, dorsal root ganglion neurons, and retinal ganglion cells. Download Figure 10-1, XLSX file.
Figure 10-2
Manually curated list of publications that link common injury-responsive genes to axon growth and/or cell death outcomes. Download Figure 10-2, XLSX file.
For insight into the function of the 145 commonly upregulated (CU) transcripts, we generated a functional interaction network using STRING (Szklarczyk et al., 2015; Fig. 10A,B). Five main clusters emerged, four of which contained between four and eight gene products based on actin organization, tRNA synthesis, amino acid import, and one-carbon metabolism. The fifth and largest clusters contained numerous RAGs including Sox11, Smad1, Klf6, Stat3, Cdkn1a, and Sprr1a and stress-related transcripts such as Ddit3 (Chop), Hrk, Eif2ak3 (Perk), and Bbc3 (Fig. 10B). To systematically identify transcripts that could contribute functionally to axon growth, each of the 145 CU gene symbols was combined with “axon” in PubMed, and the resulting manuscripts manually examined for evidence that knock-out or overexpression in any neuronal cell type resulted in a change in axon length (Extended Data Fig. 10-2). Interestingly, 32 of 145 (22%) transcripts met these criteria, with all but 5 playing a positive role in axon growth (evidence for increased growth upon overexpression and/or decreased growth after knockdown). Similarly, all gene symbols were combined with “neuron” and “apoptosis,” and the results were manually screened for experimental evidence of neuron-intrinsic effects on cell survival, yielding 40 (27.6%) with a functional role in cell survival and/or integrated stress responses. For comparison, a background set of 145 transcripts upregulated in the CST-IC set but not shared by DRG or RGC neurons was generated. Of these, only 11 could be manually linked to axon growth or neuronal survival, significantly lower than detection in the CU set (p < 0.0001, chi-squared; Extended Data Fig. 10-2). Finally, we examined the behavior of CU transcripts in the prior spinal injury sets. Across all CST neurons after cervical injury, only one CU transcript, Hrk, was significantly upregulated more than twofold, with nine additional transcripts showing upregulation that was statistically significant but below this fold threshold (Fig. 10C). Within the hCreb5 subset, 14 CU transcripts were upregulated more than twofold, including four axon-linked terms (Jun, Arid5a, Rhoq, and Abca1) and five stress-related terms (Hrk, Trib3, Bbc3, Eif2ak3, Casp3, and Ddit3; Extended Data Fig. 3-1, Fig. 10C). Besides these, nine CU transcripts showed an injury-triggered increase that was significant but below a twofold threshold. It is notable, however, that 122 (84%) of CU transcripts were unaffected by spinal injury even in this higher-responding subset. Overall, these data identify a common set of axon- and stress-relevant transcripts that are strongly upregulated by CST neurons after proximal axon injury but minimally after spinal injury.
Discussion
We have analyzed the transcriptional responses of mixed supraspinal populations to thoracic spinal injury and the response of CST neurons to thoracic, cervical, and intracortical injury. The main finding is that spinal injury led to minor transcriptional changes, as evidenced by the number of affected transcripts and the degree of up- or downregulation in affected transcripts. Conversely, after axotomy located close to CST cell bodies, we detected a much larger transcriptional response that was comparable in magnitude to that observed in other cell types. Thus, the low response to spinal injury cannot be explained by technical limitations or inherent CST insensitivity to injury. Instead, it is more likely that the cellular response to spinal injury is limited by anatomical factors, such as the intervening distance to the injury site or the preservation of proximal collateral connections. Furthermore, when comparing the CST response to proximal injury with that of other cell types, we identified a central network of injury response involving many transcripts linked with cellular stress and axon growth. However, this network is only minimally activated by spinal injury. Thus, the limited regenerative response of descending neurons to spinal injury is more likely caused by the failure to detect or react to distant injury, rather than inherent cellular constraints or an inability to maintain a transient reversion to an embryonic state.
This conclusion correlates closely with findings from decades past, emphasizing the critical role of injury proximity in the process of axon regeneration. Seminal work in the 1980s established that although some supraspinal populations can regenerate axons into peripheral nerve grafts placed into the cervical spinal cord, this ability was mostly lost when grafts were placed in thoracic regions (Richardson et al., 1984). Similarly, the ability of RGCs to regenerate axons into peripheral nerve grafts requires that injury occur within ∼3 mm of the cell body (You et al., 2000). Likewise, a series of manuscripts in the 1990s showed that in rubrospinal neurons, both regenerative ability and the expression of regeneration-associated genes depend on injury proximity (Tetzlaff et al., 1991; Jenkins et al., 1993; Fernandes et al., 1999). Finally, research in CST neurons identified six RAGs that increase in expression after intracortical but not spinal injury; our work not only replicates this result but also extends it across the genome (Mason et al., 2003). Therefore, our findings of relatively modest transcriptional responses to spinal injury reinforce and expand upon an earlier concept that the proximity of injury plays an important role in regulating the regenerative response of neurons.
Our conclusions, however, are unexpected considering a recent study examining postinjury transcriptional patterns in CST neurons (Poplawski et al., 2020). Although this study primarily examined CST neurons regenerating into embryonic tissue, which may provide transcription-altering signals not present in the current experiment, it also included nongrafted controls for more direct comparison. Within CST neurons that were cervically injured but not provided with embryonic grafts, Poplawski et al. reported approximately 10 times more DEGs at similar fold change thresholds than were detected here. This led to a conclusion that even without embryonic grafting or other treatment, spinal injury triggers in CST neurons a spontaneous, albeit transient, reversion to an embryonic pattern of gene expression (Poplawski et al., 2020). We did not detect evidence for embryonic reversion nor for any strong transcriptional response. Technical differences between the studies might contribute to the discrepancy, such as the 10 d postinjury timepoint in the previous study compared with 7 d in our study and, crucially, the use of a TRAP approach (Heiman et al., 2014) versus our single-nuclei analysis centered approach. The TRAP approach could potentially detect injury-induced changes in RNA translation not reflected in altered mRNA levels within the nuclear compartment. Interestingly, one area of agreement between the two datasets concerns the injury-responsive transcripts that we found to be shared across cell types (Fig. 5). Of the 145 transcripts upregulated twofold in intracortical CST, DRG, and RGC datasets, Poplawski et al. found only 12 (8%) to be upregulated after cervical injury, none of which have previously been linked to axon growth. Thus, both our data and those from Poplawski et al. converge on the critical point that after spinal injury, CST neurons fail to strongly upregulate a core set of proregenerative transcripts (e.g., Atf3, Sox11, Klf6, Smad1, Stat3, Gap43, L1cam, and Gal).
Indeed, the lack of ATF3 expression and the very low level of c-JUN phosphorylation observed in CST neurons after spinal injury, as confirmed at the protein level through immunohistochemistry, may play a significant role in understanding the low regenerative response. ATF3 plays a crucial role in regulating axon regeneration, as evidenced by experiments across multiple cell types showing that genetic ablation of Atf3 substantially reduces regenerative axon growth (Renthal et al., 2020; Jacobi et al., 2022; Katz et al., 2022). More broadly, JUN phosphorylation and ATF3 upregulation are markers of damage-sensing pathways, notably the DLK pathway, which is critical for initiating axon growth in various cell types (Watkins et al., 2013; Asghari Adib et al., 2018; Shin et al., 2019). Studies across a diverse range of species and cell types suggest that inhibiting the DLK pathway nearly eliminates injury-triggered axon growth, indicating that this pathway is likely essential for neurons to sense and respond to injury (Shin et al., 2012; Watkins et al., 2013; Saikia et al., 2022). In this context, our findings that DLK indicator genes are either low or absent in CST neurons following spinal injury hint that a muted activation of damage-sensing pathways may limit their potential for a regenerative response.
Important caveats apply. First, the large transcriptional response to intracortical injury cannot be interpreted entirely as a response to axotomy. The nearby injury also acts directly on cell bodies, for example, by altering blood supply and by triggering local activation of microglia (Fig. 6). Thus, the transcriptional response is most likely a mixture of retrograde injury signaling and stimuli detected directly by the cell body; parsing these effects would require some form of proximal injury with smaller inflammatory and vascular consequences, for example, laser axotomy. Another important caveat is that the detection of nuclear RNAs provides a snapshot of the RNA currently in production and early processing but does not report on the final levels of RNA within the cytoplasm and within the axons. It is therefore possible that although spinal injury does not change the nuclear abundance of most transcripts, mechanisms related to RNA quality control, rate of translation, localization, or others could shape a response not detected here. However, this limitation is less relevant for RAG transcripts such as Atf3, Sprr1a, and Tubb2b which remain essentially undetectable even after spinal injury. In these cases, downstream mechanisms would have limited capacity to reverse extreme scarcity in the nucleus, indicating that these RAGs are likely unavailable to support any attempt at regeneration.
In summary, our thorough analysis of transcriptional responses in mixed supraspinal populations and CST neurons following different spinal injuries reveals modest changes in gene expression, particularly in response to distal spinal injuries. This redirects attention to an earlier hypothesis that anatomical factors, particularly the proximity of the injury to the neuron's cell body, play a crucial role in determining the extent of transcriptional response and regeneration. Additionally, our data confirm that CST neurons do not spontaneously regress to a growth-competent state following spinal injury, underscoring the need to therapeutically stimulate a transcriptional program supportive of growth (Sun and He, 2010; Blackmore, 2012; Venkatesh and Blackmore, 2017; Mahar and Cavalli, 2018). Furthermore, the minimal activation of essential damage-sensing pathways, such as the DLK pathway—as indicated by the sparse presence of DLK pathway genes—suggests a potential mechanistic explanation for the restrained regenerative responses observed in CST neurons after spinal injuries. Together, these insights enhance our understanding of the regulatory mechanisms affecting axon regeneration and lay the groundwork for further research into the biological processes that limit neuronal recovery following injury.
Footnotes
This work was supported by grants from NINDS R01 NS083983, the Bryon Riesch Paralysis Foundation, and the Christopher and Dana Reeve Foundation (awarded to M.G.B.) and funds from CSIR-CCMB, SERB-SRG (BT/PR51467/MED/122/358/2024; awarded to I.V.), and CSIR (HCP532401; awarded to I.V.) and Department of Biotechnology (DBT, awarded to I.V.).
↵*Z.W. and M.K. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Murray G. Blackmore at murray.blackmore{at}marquette.edu or Ishwariya Venkatesh at ishwariya{at}ccmb.res.in.
