Single-nucleus characterization of adult mouse spinal dynorphin-lineage cells and identification of persistent transcriptional effects of neonatal hindpaw incision

1. Introduction

Neonatal tissue damage can have long-lasting effects on nociceptive processing in the central nervous system. Human infants who receive invasive medical procedures while being treated in the neonatal intensive care unit (NICU) experience enhanced pain sensitivity in response to prolonged nociceptive stimulation later in life [37; 38]. Furthermore, injury during early life ‘primes’ rodents to exhibit exacerbated hyperalgesia upon re-injury to the same dermatome in adulthood [57; 58; 83].

This aberrant pain perception following early-life tissue damage may be partially due to persistent, injury-evoked disruptions in the normal balance of synaptic excitation vs. inhibition in the mature spinal dorsal horn (DH) [51–53]. For example, neonatal injury strengthens primary afferent input onto adult lamina I spinoparabrachial neurons [50], which represent a major output of the spinal nociceptive network [78], while simultaneously weakening feedforward inhibition onto these projection neurons [45]. Lamina I spinoparabrachial neurons receive direct GABAergic input from spinal neurons derived from the prodynorphin (pDyn) lineage [17; 35], which play a critical role in dampening mechanical pain [25] and itch [43]. Importantly, the strength of this inhibitory synapse onto adult projection neurons is significantly reduced following neonatal injury, due to a prolonged reduction in the probability of GABA release [16]. The efficacy of the pDyn inhibitory circuit is further compromised by a concomitant reduction in the intrinsic membrane excitability of mature pDyn neurons as well as a reduction in the strength of their primary afferent inputs [16]. However, the molecular mechanisms underlying the long-lasting disruption to the function of this critical inhibitory network following neonatal injury have yet to be identified.

We therefore present an unbiased investigation into the persistent effects of neonatal hindpaw incision on the transcriptome of mature spinal pDyn neurons. Our previous population-level analysis of gene expression in developing spinal pDyn neurons revealed considerable heterogeneity [73], which is consistent with reports demonstrating that the dynorphin-expressing dorsal horn neurons intersect with other inhibitory populations including those characterized by expression of galanin, neuronal nitric oxide synthase (nNOS) or neuropeptide Y (NPY) [13; 18; 68]. Moreover, although the majority of pDyn dorsal horn neurons are inhibitory, this population also contains excitatory neurons [13; 25], which may respond to neonatal injury differently than their inhibitory counterparts.

The present study utilizes a single nucleus-based RNAseq (sNuc-seq) approach to identify and characterize distinct subpopulations of adult spinal pDyn neurons, to more precisely assess persistent changes in gene expression associated with neonatal surgical injury that may be unique to each of these subpopulations. Eleven transcriptionally distinct clusters of adult spinal pDyn cells were identified, and gene expression was compared within each neuronal cluster between neonatally incised mice and naïve littermate controls. Additionally, aggregate clustering of identified excitatory or inhibitory neurons allowed separate characterization of the transcriptional effects of neonatal injury specific to each neurotransmitter phenotype. These analyses revealed minimal injury-evoked transcriptional reprogramming in pDyn neurons, but nevertheless identified 15 genes that are significantly up- or downregulated in identified subpopulations of adult spinal pDyn neurons following neonatal tissue damage.

2. Methods

2.1. Animals

All animal experiments were performed in accordance with University of Cincinnati Institutional Animal Care and Use Committee policies. To selectively label spinal nuclei derived from the prodynorphin lineage, homozygous Pdyn-IRES-cre mice [45] (Jackson Stock #027958) were bred with homozygous mice expressing a cre-dependent Sun1-GFP fusion protein from the Rosa26 locus (R26-CAG-LSL-Sun1-sfGFP-Myc; Jackson Stock #021039). The offspring of this cross produced mice heterozygous for both alleles, termed pDyn-GFP mice. A total of 16 animals were used for sNuc-seq [34]. 8 of these mice (4 male and 4 female) underwent hindpaw incision at postnatal day (P) 3, while the other 8 mice (5 male and 3 female; littermates of incised mice) underwent only exposure to isoflurane as a control. To perform the hindpaw incision, P3 mice were anesthetized with isoflurane, and an incision was made in the left hindpaw skin and underlying plantar muscle [14; 52]. The incision was closed with sutures, and mice were allowed to mature under standard housing conditions until the time of tissue harvest at P63-P72.

2.2. Isolation of adult spinal pDyn nuclei via Fluorescence-Activated Nuclei Sorting (FANS)

Adult pDyn-GFP mice were euthanized with an overdose of sodium pentobarbital (30 mg/kg, Vortech Pharmaceuticals; Dearborn, MI) then decapitated. The lumbar spinal cord was rapidly dissected and placed in ice-cold 0.1M RNAse-free phosphate-buffered saline (PBS). The lumbar enlargement was split longitudinally to separate the left and right sides, then the dorsal half of segments L3-L5 was harvested from the left side (i.e. ipsilateral to the injury), snap-frozen on dry ice, and stored at −80°C until the day of sample preparation. Isolation of spinal cord nuclei was performed as previously described [21; 73]: dorsal spinal cord halves were homogenized in batches of 4, grouped by injured or naïve condition, in 1 ml of homogenization buffer (HB) containing in mM: 250 Sucrose, 25 KCl, 20 Tris-HCl (pH 8), 5 MgCl₂, plus 1 tablet/10 ml cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche; Indianapolis, IN), 40 U/ml RNAsin Plus (Promega; Madison, WI), and 1 μM DTT. Centrifugation and filtration were carried out as previously described [21; 73], generating two samples of pooled nuclei (injured and naïve), each suspended in 600 μl HB. Nuclei were stained with propidium iodide (PI; ThermoFisher; Waltham, MA) and stored on ice until fluorescence-activated nuclei sorting (FANS).

FANS was performed by the Research Flow Cytometry Core in the Division of Rheumatology at Cincinnati Children’s Hospital Medical Center, using a BD FACSAria II cell sorter equipped with a 70 μm nozzle. Intact GFP+ nuclei were recovered from each sample and sorted into a 1.5 ml microcentrifuge tube filled with 0.1M RNAse-free PBS containing 2% non-acetylated BSA (Sigma; St. Louis, MO) and 0.5U/μl RNAsin Plus (Promega). To concentrate nuclei per 10x Genomics guidelines, recovered samples were centrifuged at 300 x g for 7 minutes at 4°C, then most of the supernatant was aspirated off. Samples were centrifuged again under the same conditions for 3 minutes, then all but 22 μl of the supernatant was removed and visually examined to confirm a lack of nuclei remaining in suspension. The pellet of nuclei was gently pipetted up and down to re-suspend in the remaining 22 μl of PBS. 2 μl of this suspension was diluted 1:5 with trypan blue and visualized on a hemacytometer to obtain an approximate concentration of nuclei.

2.3. 10x Genomics library preparation and sequencing

Single nucleus RNA-sequencing was performed by the Gene Expression Core Facility in the Division of Developmental Biology at Cincinnati Children’s Hospital Medical Center, using 10x Genomics Chromium Single Cell 3’ Gene Expression Assay v3 (10x Genomics; Pleasanton, CA). Nuclei from the Naïve and Injured groups were run as separate samples on the 10x Chromium instrument and constructed as separate libraries in the same run, which were assessed for quality using a 2100 Bioanalyzer High Sensitivity DNA Assay (Agilent; Santa Clara, CA) then pooled for sequencing on a NovaSeq 6000 sequencer (Illumina; San Diego, CA) using an SP flow cell. The combined samples were subjected to a total of ~450M reads (paired-end 75 bp). Mean reads per cell for the Naïve sample = 53,301; mean reads for the Injured sample = 65,886.

2.4. Bioinformatic analysis

Bioinformatic analysis was performed in collaboration with the Bioinformatics Collaborative Services (BCS) in the Division of Biomedical Informatics (BMI) at Cincinnati Children’s Hospital Medical Center. Raw base call files were de-multiplexed with Cell Ranger [93] v3.0.2 mkfastq (10x Genomics). Reads were aligned to mouse reference genome mm10 (modified to include intronic sequences [7]) and gene expression quantified using Cell Ranger count. Further data analysis was carried out with Seurat [19; 76] v3.0–3.1 in R v3.5.0–4.0.1 [64]. Nuclei displaying more than 0.3% mitochondrial gene expression, more than 20,000 UMI, or less than 500 total expressed genes were excluded from the analysis. A total of 4,031 (42 excluded) and 3,438 (112 excluded) nuclei in naive and injured samples, respectively, survived the cutoffs and were considered for further analysis. Gene expression counts were normalized with the NormalizeData function in Seurat, which uses a logarithmic normalization method where gene counts for each nucleus are divided by its total counts and natural log-transformed using log1p and multiplied by a scale factor of 10,000. The two samples were integrated together using FindIntegrationAnchors and IntegrateData functions from Seurat. This integrated dataset was used for principal component analysis, variable gene identification, Shared Nearest Neighbor (SNN) clustering analysis, and Uniform Manifold Approximation and Projection (UMAP).

2.5. Clustering and Hierarchical Analysis (Dendrogram)

Using the Louvain modularity optimization in Seurat, we explored clusters (unsupervised) at several resolution values, ranging from 0 to 1. Resolution is a parameter within the Louvain algorithm that controls the number of generated clusters. A tree displaying clusters at different resolutions was generated using Clustree [86] v0.4.0 to aid in the exploration and selection of the appropriate resolution value. Marker genes for each cluster were identified by comparing gene expression in nuclei from the target cluster against all other nuclei in the remaining clusters using the Wilcoxon rank sum test. Marker genes met the following criteria: 1) Log₂ fold-change (Log2FC) ≥ 0.50 in the target cluster, 2) minimum of 25% of nuclei expressing the gene in both compared groups of nuclei, and 3) Bonferroni-corrected p-value < 0.05. A resolution value of 0.06 was ultimately selected, as distinct gene expression profiles were present in each cluster, while higher resolutions produced adjacent clusters with a high degree of overlapping marker genes. To evaluate the predictive power of clustering at resolution 0.06, a random forests classifier algorithm (Python) using 10,000 trees was trained on 80% of the dataset (randomly selected using random state 42) and then predicted the cluster of the remaining 20% of nuclei.

The BuildClusterTree function in Seurat was used to generate a dendrogram based on a distance matrix at resolution 0.06, showing hierarchical linkage between 11 clusters. Genes governing each node “split” were identified by comparing gene expression on one side of the split against the other side of the split using the Wilcoxon rank sum test with the following inclusion criteria: 1) minimum enrichment Log2FC ≥ 0.50 on either side; 2) minimum percentage of cells expressing the gene ≥ 25 on either side; and 3) adjusted p-value < 0.05. Genes displayed on the dendrogram were selected on the basis of relative enrichment and the relative percentage of cells expressing the gene across both sides of the node. A full list of genes differentially expressed across each dendrogram node is attached as supplemental material (supplemental digital content 2).

2.6. Analysis of Injury-evoked Differential Gene Expression and classification of clusters

Differentially expressed genes (DEGs) with average fold change (Log2FC) ≥ 0.25 and adjusted p-value < 0.05 were identified between Naïve and Injured samples for each cluster. For aggregate excitatory/inhibitory neuron population DEG analysis, each of the 11 clusters was first identified as either neuronal or non-neuronal based on the relative expression of neuronal markers Snap25, Syn1, and Syp, or glial markers Mog, Mbp, Sox10, Aqp4, and S100b. Cluster 10 was identified as non-neuronal and therefore excluded from either aggregate neuronal population. Next, the remaining neuronal clusters were classified as either excitatory or inhibitory, based on relative mean expression of inhibitory markers Slc32a1, Gad1, Gad2 and Slc6a5, or excitatory marker Slc17a6 [36]. Cluster 9 was composed of both excitatory and inhibitory neurons in similar proportions. Attempts to split this cluster by increasing resolution (up to 0.6) did not result in separation of excitatory and inhibitory neurons, so Cluster 9 was classified as “Mixed” and excluded from either aggregate neuronal population. Remaining clusters 1–6 and 11 were classified as Inhibitory, while clusters 7 and 8 were classified as Excitatory.

2.7. Identification of Functional Classes of Genes and DotPlot Generation

Gene lists of enzymes, G protein-coupled receptors (GPCRs), ion channels, and transporters were obtained from the IUPHAR Database [6]; neuropeptides from the NeuroPep Database [84]; transcription factors from the Riken Transcription Factor Database [42]; and pain-related genes from the Pain Genes Database [46]. All lists were obtained as the most recent version of the databases as of 30 Mar 2020. Each functional class list was cross-referenced with the list of 2,000 most highly variable genes in our dataset, as identified by the FindVariableFeatures function in Seurat. Genes on this cross-referenced list with a minimum percentage of cells expressing the gene ≥ 25 in at least one cluster were included on the dot plot for a given functional class. All plots were generated using various libraries in R v3.5.0–4.0.1, including Seurat, ggplot2 v3.2.1 [86], and Clustree. Figures were assembled in Photoshop (Adobe; San Jose, CA).

2.9. In situ hybridization and Immunohistochemistry (IHC)

Two male and two female adult (P63) pDyn-GFP mice were euthanized by IP injection of sodium pentobarbital (30 mg/kg; Vortech Pharmaceuticals) and transcardially perfused with 0.01 M phosphate buffer (PB) followed by 4% paraformaldehyde (PFA) dissolved in PB. Lumbar spinal cords were dissected out and post-fixed for an additional 2 hours in 4% PFA at room temperature. Lumbar spinal cords were then transferred to 30% sucrose in RNAse-free PBS and stored overnight at 4°C. 14 μm transverse sections were cut on a CM 1860 cryostat (Leica; Wetzlar, Germany) and mounted onto SuperFrost Plus slides (Fisher) such that sections from all four animals were arranged on each of the slides to eliminate variability in subsequent processing.

RNAscope Multiplex Fluorescent Kit v2 (Advanced Cell Diagnostics; Newark, CA) was used to carry out fluorescent in situ hybridization experiments according to manufacturer’s directions for fresh-frozen tissue pretreatment without the on-slide fixation step. The following probes were used: Cdh3 (514591-C3), Fam19a1 (520731), Nos1 (437651-C3), Pde11a (481841), Slc17a6 (319171-C3), Tac2 (446391-C2), and Mobp (431728-C2) in conjunction with TSA Plus Cyanine 3 and Cyanine 5 systems (NEL744E001KT and NEL745E001KT, PerkinElmer; Waltham, MA) for visualization. Because RNAscope pretreatment steps destroyed endogenous Sun1-GFP fluorescence, anti-GFP immunohistochemistry staining was then performed using rabbit anti-GFP (ab6556, Abcam; Cambridge, UK) at a dilution of 1:500 and AlexaFluor 488 Goat anti-Rabbit IgG (A11034, Thermofisher) as previously described [73]. After IHC, slides were washed 3 times with PBS, DAPI from the RNAscope kit was applied for 30 seconds, and slides were coverslipped with VectaShield mounting medium (Vector Labs; Burlingame, CA).

2.10. Image Acquisition and Analysis

Images were captured on a BX63 upright fluorescent microscope (Olympus; Center Valley, PA) using CellSens Dimension Desktop Software (Olympus). Multichannel Z-stack images were obtained using the 20X or 40X objectives and projected as Extended Focal Images. Four sections from each of the four mice were captured such that adjacent sections were not evaluated (to avoid double-counting the same cells). To determine the percentage of cells which expressed a given mRNA transcript, all GFP+ nuclei in each image were marked using CellSens Count and Measure function, then each marked nucleus was separately examined for Cy3 and Cy5 fluorescence. A cell was considered positive for expression of a given target if 3 or more puncta in the appropriate fluorescent channel were within or immediately adjacent to (touching) the GFP-immunoreactive nucleus. The percentage of cells in which mRNA for multiple targets co-localized was then determined by counting the number of cells which were positive for both markers.

2.11. Data Availability

Raw sequence data for all samples in this study, have been deposited in the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) under study accession number GSE149527. Seurat object containing the integrated dataset is available on the Open Science Framework at project URL: https://osf.io/k6bvs/?view_only=69e27a9fee854e55be45696f1b51986b.

3. Results

3.1. Transcriptional profiling of spinal dynorphin-lineage neurons with single-nucleus RNAseq

pDyn-GFP mice, in which nuclei derived from the prodynorphin lineage (pDyn) selectively express the Sun1-GFP fusion protein, were generated by breeding Pdyn-IRES-cre mice [45] with cre-dependent Sun1-GFP reporter mice designed for use with the Isolation of Nuclei in TAgged Cell Types (INTACT) method [56]. Our previous characterization of the GFP-tagged population in these animals demonstrated that about 80% of GFP+ cells in lamina I-III of the spinal cord actively express Pdyn mRNA in adulthood [73]. At postnatal day (P) 3, 8 mice underwent an incision to the left plantar hindpaw [14; 52] while 8 littermate controls were exposed to isoflurane for an equivalent duration but did not undergo an incision. Although both experimental groups included male and female pups, individual animal samples were pooled and sex differences were not examined. Incised or control pups were allowed to mature until P63–72, whereupon they were euthanized, and the dorsal portion of the ipsilateral side of spinal cord segments L3-L5 was harvested (Fig. 1A).

Figure 1: Single-nucleus RNA sequencing of dynorphin-lineage (pDyn) spinal nuclei reveals 11 distinct clusters.

(A) The dorsal half of the left lumbar spinal cord (L3-L5) was harvested and homogenized to liberate intact nuclei (1). Fluorescence-activated nuclear sorting (FANS) selectively isolated GFP-labeled nuclei of pDyn cells (2), which were subjected to single-nucleus capture (3) and RNA-sequencing (4) using 10X Chromium Single Cell Gene Expression assays (n = 7,469 nuclei pooled from 16 mice: 8 naïve and 8 with P3 hindpaw incision). (B) Heat map showing Z-score of scaled and log-transformed normalized expression per cell of the top 10 differentially expressed genes in each cluster, ranked by adjusted p-value. (C) UMAP dimensional reduction plot depicting 11 clusters of spinal pDyn nuclei. (D) UMAP plot showing mean expression of neuronal marker genes Snap25, Syn1, and Syp. Expression was robust in all clusters except Glia10. (E) UMAP plot showing mean expression of inhibitory neuron markers Slc32a1, Gad1, Gad2, and Slc6a5. (F) UMAP plot showing mean expression of excitatory marker Slc17a6. In general, inhibitory and excitatory markers segregate into spatially distinct clusters.

Intact nuclei were isolated from the harvested tissue, pooled according to experimental group (Naïve or Injured) and stained with propidium iodide (PI) to facilitate fluorescence-activated nuclei sorting (FANS). FANS selectively isolated nuclei marked by both GFP and PI fluorescence (supplemental digital content 1, Fig. S1A), which were then subjected to single-nucleus RNA sequencing using 10x Genomics Chromium 3’ Gene Expression Assay v3 (Fig. 1A). Quantification of a sample of sorted nuclei revealed that 97/101 nuclei exhibited both GFP and PI fluorescence, while 4/101 nuclei exhibited PI fluorescence only. Therefore we cannot exclude the possibility that a small percentage of the sequenced nuclei included those from non-pDyn cells. The use of nuclei rather than whole cells allowed for the use of stored frozen tissue such that all experimental samples from multiple litters of mice could be homogenized, sorted, and submitted for 10x sequencing all on the same day, thereby eliminating variations in preparation conditions which could confound differential gene expression analysis. Moreover, nuclei can be easily isolated from the adult spinal cord without the use of heat or protease incubation which may induce immediate-early gene expression, as described in sequencing studies utilizing whole cells [2; 33; 81]. Although nuclear RNA represents only a fraction of total cellular RNA, comparisons between whole-cell and nuclear RNA sequencing demonstrate high concordance in the overall transcriptional profile between the two approaches [7; 8; 33].

Injured and naïve nuclei were run as separate samples on the 10x Chromium instrument and constructed as separate cDNA libraries which were sequenced and de-multiplexed for analysis. A total of 4,073 nuclei were sequenced from the Naïve sample while 3,551 nuclei were sequenced from the Injured sample. Cells expressing less than 500 genes/cell, greater than 20,000 UMI/cell, or greater than 0.3% mitochondrial genes were excluded from further analysis, leaving 4,031 Naïve nuclei and 3,438 Injured nuclei which were integrated for clustering (supplemental digital content 1, Fig. S1B–C).

Although more than 19,000 genes were detected in this analysis, the top 2,000 most variable genes across this dataset were used for principal component analysis (PCA), dimension reduction (UMAP), and shared nearest neighbor (SNN) clustering analysis, ultimately yielding 11 transcriptionally distinct clusters of nuclei (Fig. 1B, C). The predictive power of these cluster identities was evaluated using a random forests classifier trained on 80% of the dataset, which predicted the cluster identity of the remaining 20% with an overall accuracy of 95.78% (supplemental digital content 1, Fig. S1D). The high rate of correct prediction may be partially due to the size of the dataset, as even the smallest cluster contained over 100 cells on which the algorithm could be trained. Clusters 1–9 and 11 were characterized by high expression of neuronal marker genes Snap25, Syn1, and Syp; therefore these were classified as neuronal clusters (Fig. 1D). Cluster 10 lacked strong expression of these neuronal genes, while expressing high levels of glial markers Gja1, Mog and Mbp, and were therefore classified as non-neuronal. For each neuronal cluster, mean expression of inhibitory markers Slc32a1 (encoding the vesicular GABA transporter VGAT), Gad1, Gad2, and Slc6a5 (encoding the glycine transporter GlyT2) was compared to that of excitatory marker Slc17a6 (encoding the vesicular glutamate transporter VGLUT2). On this basis, clusters 1–6 and 11 were designated inhibitory (Fig. 1E) and clusters 7–8 were designated excitatory (Fig. 1F). Interestingly, cluster 9 contained both excitatory and inhibitory neurons, but attempts to split this cluster did not result in division between the excitatory and inhibitory phenotypes. Therefore this cluster was designated as mixed-neuronal. Importantly, few cells exhibited co-expression of inhibitory and excitatory markers (supplemental digital content 1, Fig. S2).

A dendrogram showing the phylogenetic relationship between the 11 clusters also identified genes governing the various node splits; i.e., highly differentially expressed genes across a given branch point (Fig. 2A; see also supplemental digital content 2). An overview of cell type-specific marker gene expression in each cluster is given in Fig. 2B, and expression of selected cluster-specific marker genes is presented in Fig. 2C (see also supplemental digital content 3).

Figure 2. Dendrogram and transcriptomic characterization of spinal pDyn nuclei clusters.

(A) Linkage analysis between clusters is shown in a dendrogram. Each node of the dendrogram is identified with a boxed number, and genes best defining the left (blue text) and right (red text) sides of the split were identified on the basis of differential gene expression (DEG) analysis across the two sides of the node (supplied in Supplementary Digital Content 2). Clusters were named according to their inhibitory, excitatory, or other classification, and select marker genes identified by DEG analysis are shown in the colored box associated with a given cluster. Marker genes for a given cluster met the following criteria: AvgLog2FC ≥ 0.50 compared to all other clusters, at least 25% of cells in the given cluster expressed the gene, and p-adj < 0.05 using Wilcoxon Rank Sum Test adjusted with Bonferroni correction. (B) Violin plots represent scaled and log-transformed normalized gene expression (gene UMIs/total cell UMIs) of established neuronal, inhibitory, excitatory, glial, and central canal-contacting cell markers. (C) Violin plots of cluster marker genes, as identified in (A). All violin plots depict scaled and log-transformed normalized expression; y-axis labels removed for clarity, minimum = 0, maximum = 4.

3.2. Characterization of inhibitory pDyn neuron clusters

We hypothesized that the transcriptional effects of neonatal injury may vary across different subpopulations of pDyn neurons, so we first identified and characterized these subpopulations to facilitate the investigation of cell type-selective effects of early life tissue damage. While it is established that the majority of spinal pDyn cells are inhibitory neurons, this analysis identifies, for the first time, 7 transcriptionally distinct pDyn inhibitory subpopulations in the dorsal horn. Cluster Inhib1 was enriched for many of the markers commonly associated with dynorphin-expressing spinal cord neurons: Gal (Log2FC +0.98), Rorb (Log2FC +1.49), and the recently identified Gucy2d (Log2FC +0.95) [68; 73]. However, the most highly enriched gene in this cluster was Pde11a (Log2FC +1.98), encoding Phosphodiesterase 11A (Fig. 2C). Cluster Inhib2 was most highly enriched for Nos1 (Log2FC +1.26), encoding neuronal nitric oxide synthase (nNOS), which has been previously identified in dynorphin-expressing spinal neurons in lamina I-II [69]. Other notable marker genes of cluster Inhib2 were Slit3 (Log2FC +1.06) and Cdh6 (Log2FC +1.02).

Because the Inhib1 and Inhib2 clusters were sister groups in the cluster dendrogram (Fig. 2A), the extent of overlap between these two clusters was investigated using multiplex in situ hybridization with Pde11a as a marker for cluster Inhib1 and Nos1 as a marker for cluster Inhib2 (Fig. 3A–B). Quantification of cells with Sun1-GFP immunoreactive nuclei (i.e., pDyn-GFP cells) confirmed that these two clusters indeed overlapped only minimally, with 33.91 ± 1.36% of pDyn-GFP cells expressing Pde11a mRNA and 36.82 ± 1.60% expressing Nos1 mRNA, while only 9.84 ± 0.91% of pDyn-GFP cells expressed both markers (Fig. 3C). Among only those pDyn-GFP+ cells which expressed either Pde11a or Nos1, 16.07 ± 1.46% displayed co-localization for both markers. For this and all following in situ hybridization experiments, only mRNA expression within pDyn-GFP+ cells was quantified, and the incidence of wider co-localization in unidentified spinal cord cells was not investigated.

Figure 3. In situ hybridization validation of marker gene expression in clusters Inhib1–4.

(A) UMAP plots showing expression of markers genes Pde11a for cluster Inhib1 (left) and Nos1 for Inhib2 (right). (B) mRNA transcript expression of Pde11a (magenta) and Nos1 (yellow) in the lumbar dorsal horn. Scale bar = 50 μm. (C) High magnification view of Pde11a (magenta) and Nos1 (yellow) mRNA expression within pDyn-GFP nuclei (green). Open arrowheads indicate pDyn-GFP cells positive for Pde11a expression while filled arrowheads indicate pDyn-GFP cells positive for Nos1 expression. Scale bars = 20 μm, nuclei stained with DAPI (blue). (D) UMAP plots showing expression of marker genes Cdh3 for cluster Inhib4 (left) and Tac2 for cluster Inhib3 (right). (E) mRNA transcript expression of Cdh3 (magenta) and Tac2 (yellow) in the lumbar dorsal horn. Scale bar = 50 μm. (F) High magnification view of Cdh3 (magenta) and Tac2 (yellow) mRNA expression within pDyn-GFP nuclei (green). Open arrowheads indicate pDyn-GFP cells positive for Cdh3 while filled arrowheads indicate pDyn-GFP cells positive for Tac2 expression. Asterisks indicate pDyn-GFP cells that are negative for both markers. Scale bar = 20 μm, nuclei stained with DAPI (blue).

The relative lack of overlap between two other sister inhibitory clusters, Inhib3 and Inhib4, was also validated using in situ hybridization. These two clusters had distinct transcriptional profiles, where Inhib3 was characterized by enrichment for Fbxl7 (Log2FC +1.75), Il1rapl2 (Log2FC +1.76), and Tac2 (Log2FC +0.89) encoding the neuropeptide Neurokinin B, while Inhib4 was enriched for Mgmt (Log2FC +0.93), Cdh3 (Log2FC +0.87), and Nppc (Log2FC +0.61) encoding Natriuretic Peptide C. Using Tac2 as a marker for Inhib3 and Cdh3 as a marker for Inhib4 (Fig. 3D) revealed that these clusters were primarily composed of cells in deeper dorsal horn laminae where dynorphin-lineage cells are less prevalent than in laminae I-II (Fig. 3E). Additionally, a lack of overlap between these two populations was observed within the dynorphin lineage: while 3.98 ± 0.82% of pDyn-GFP cells expressed Tac2 mRNA and 17.93 ± 1.22% of cells expressed Cdh3 mRNA, no pDyn-GFP cells were observed which expressed both targets (Fig. 3F).

Clusters Inhib5 and Inhib6 were much less closely related to Inhib1–4, with cluster Inhib5 branching off relatively early in the dendrogram (Fig. 2A; node 14). Both lacked strong expression of Kcnt2, encoding the Slick K⁺ channel, which was expressed at a much higher level in clusters Inhib1–4. While Inhib5 expressed each of the four identifying inhibitory marker genes – Gad1, Gad2, Slc32a1, and Slc6a5 – it was notably enriched for Slc6a5 (GlyT2) in comparison to the other inhibitory clusters (Fig. 2B). Despite the rare incidence of co-localization with dynorphin documented in the existing literature [13; 43], modest enrichment for Pvalb (encoding parvalbumin; Log2FC +0.34) was also detected in this cluster, consistent with previous reports localizing parvalbumin expression to primarily glycinergic neurons in laminae I-III [48]. Additionally, expression of Atp10a (Log2FC +1.38), transcription factor Ebf2 (Log2FC +1.91), and Fam20a (Log2FC +1.13) were also highly specific to this cluster. Cluster Inhib6 was marked by expression of Npy (Log2FC +0.54), encoding Neuropeptide Y, the transcription factor Pax5 (Log2FC +0.53), and Rxfp2 (Log2FC +0.53), which encodes the receptor for insulin-like peptide 3. However, Cacna2d3, encoding a voltage-gated calcium channel subunit highly expressed in all other inhibitory clusters, was notably reduced in Inhib6 (Fig. 2A, node 15).

Dynorphin-expressing cells in the spinal dorsal horn have been extensively characterized in lamina I-III, as the majority of this lineage resides in these most superficial layers [13; 25; 43]. However, the dramatic enrichment of highly selective markers of cerebrospinal fluid-contacting neurons (CSF-cN), including Pkd1l2 (Log2FC +4.08) and Pkd2l1 (Log2FC +3.05), within cluster CC11 strongly suggested that the dynorphin lineage may also include a deeper subpopulation residing close to the central canal [32; 41]. Histological analysis of spinal cord sections from neonatal (P7) and adult (P74) pDyn-GFP mice confirmed the presence of spinal pDyn cells contacting the central canal; moreover the number and distribution of cells in this area appeared to be age-dependent. In the neonate, densely packed pockets of pDyn-GFP nuclei contacted both the dorsal and ventral aspects of the central canal and appeared to extend dorsally and ventrally along the midline (Fig. 4A, left). However, these pockets appeared to diminish in size and density in the adult, and very few adult pDyn-GFP nuclei were observed distal to the few cell layers directly surrounding the canal (Fig. 4A, right; Fig. 4B). Like inhibitory clusters 1–6, CC11 is enriched for neuronal (Fig. 1D) and inhibitory neuron marker genes (Fig. 1E), but the high specificity of marker genes identified in this cluster suggest that this population may be both spatially and functionally distinct from other pDyn neurons detected in this analysis. Other notable enriched genes include Slc26a4 (Log2FC +2.24) encoding the Cl⁻/HCO₃⁻ exchanger pendrin, as well as Myo3b (Log2FC +3.46) and Espn (Log2FC +1.48), both of which play a role in stereocilia organization and development and have been previously detected in ciliated neurons lining the central canal [26; 32].

Figure 4. A population of cerebrospinal fluid-contacting neurons (CSF-cNs) are derived from the dynorphin lineage.

(A) Transverse sections of neonatal (P7, left) and adult (P74, right) lumbar spinal cord showing cells tagged with pDyn-GFP nuclei located at the dorsal and ventral poles of the central canal. Scale bars = 50 μm. (B) A sagittal view of the adult spinal cord reveals that these populations of pDyn-GFP cells extend along the length of the lumbar central canal. Scale bar = 20 μm. All panels: nuclei stained with DAPI (blue).

3.3. Characterization of excitatory dynorphin-lineage clusters

Excitatory clusters Excit7 and Excit8 (and a subset of nuclei within the Mixed9 cluster) were characterized by the expression of the glutamate transporter Slc17a6 (VGLUT2), voltage-gated calcium channel subunit Cacna2d1, and Fam19a1, a chemokine-like neuropeptide which is the putative ligand of GPCR1 (GPR1) [92]. While Fam19a1 expression has been characterized in the brain and DRG [49; 90; 92], spinal cord expression has not been specifically investigated. The expression of Fam19a1 coincides with Slc17a6 expression in the present analysis (Fig. 5A), and in situ hybridization analysis using these two markers confirms substantial co-localization within the spinal dynorphin lineage (Fig. 5B). Fam19a1 was detected in 36.05 ± 2.22% of pDyn-GFP cells, and 89.86 ± 2.36% of these co-expressed Slc17a6 (Fig. 5C). It should be noted that Fam19a1 is also modestly expressed in inhibitory cluster Inhib5 (Fig. 5A, right; Fig. 2C), which may account for some of the 10% of Fam19a1+ pDyn-GFP cells that do not express Slc17a6. Moreover, although the incidence of Fam19a1 and Slc17a6 co-localization in cells lacking pDyn-GFP expression was not quantified, visible co-localization of these transcripts strongly suggests that many Fam19a1+ cells, especially in the deeper laminae of the dorsal horn, are excitatory neurons (Fig. 5B).

Figure 5. Fam19a1 is enriched in excitatory pDyn-GFP neurons in the spinal dorsal horn.

(A) UMAP plots showing expression of excitatory marker gene Slc17a6 (left) and Fam19a1 (right), which is highly concentrated in clusters Excit7–8 and the excitatory portion of Mixed9. (B) mRNA transcript expression of Slc17a6 (magenta) and Fam19a1 (yellow) in the lumbar dorsal horn. Scale bar = 50 μm. (C) High magnification view of Slc17a6 (magenta) and Fam19a1 (yellow) mRNA expression within pDyn-GFP nuclei (green). Open arrowheads indicate pDyn-GFP cells positive for both Slc17a6 and Fam19a1 expression while filled arrowheads indicate pDyn-GFP cells that do not express either marker. Scale bar = 20 μm, nuclei stained with DAPI (blue).

In contrast to the inhibitory clusters identified in this analysis, excitatory clusters Excit7 and Excit8 appeared to be somewhat less transcriptionally distinct from one another. Excit7 was notable in that it is enriched for Meis2 (Log2FC +1.75), Fam19a2 (Log2FC +1.47), and Tac1 (Log2FC +1.32) encoding substance P. Excit8 was primarily enriched in the canonical excitatory markers mentioned above, in addition to Lmx1b (Log2FC + 0.80), Tmem163 (Log2FC +1.16), and Tacr1 (Log2FC +0.61) encoding the neurokinin 1 receptor for substance P.

3.4. Non-neuronal Pdyn-lineage cells

Although overwhelmingly neuronal, a small percentage of the sequenced nuclei appears to be non-neuronal. Cluster Glia10 lacked robust expression of neuronal marker genes but was highly enriched in glial-specific genes Mbp (Log2FC +2.72), Mobp (Log2FC +1.84) and Apoe (Log2FC +2.12) [4]. While these findings are consistent with previous reports which show that immature spinal astrocytes and cultured oligodendrocytes can express dynorphin [44; 82], it is unclear whether the inclusion of non-neuronal nuclei in our dataset is due to Pdyn expression in these cells at some point during development, or simply due to a small amount of GFP-negative nuclei inadvertently isolated by FANS. Notably, although it might be reasonably suspected that FANS contamination with non-pDyn nuclei would also include microglia, expression of microglial marker genes Tmem119, Cx3cr1, and Cd11b [10] was either minimal or undetected in all 11 clusters. In situ hybridization was performed on spinal cord sections from adult pDyn-GFP mice in an attempt to confirm oligodendrocyte marker Mobp expression in pDyn-GFP cells, but no such co-labeled cells were observed (n = 16 sections, data not shown). Therefore, further study will be required to definitively determine whether the dynorphin lineage within the spinal dorsal horn contains non-neuronal cells.

3.5. Functional classes of genes expressed in dynorphin-lineage clusters

Although marker genes of any type are informative for the characterization of transcriptional heterogeneity within spinal nociceptive circuits, G protein-coupled receptors (GPCRs), ion channels, neuropeptides, and other categories of actionable targets are more relevant for the treatment of pain and other pathological states. Therefore, the list of the top 2,000 most highly variable genes identified in the present study was filtered against curated databases [6; 42; 46; 84] to produce plots showing the expression of several functional classes of genes across each of the 11 clusters (Fig. 6A). Genes belonging to each functional class were only included on their respective plot if they were expressed by a minimum of 25% of cells in at least one cluster, eliminating those minimally expressed genes which may be less viable targets for further study or potential intervention.

Figure 6. Expression of select functional classes of genes in each cluster.

(A) Genes of selected functional classes detected among the 2,000 most highly variable genes in this analysis. (B) Listed genes are among the 50 most highly enriched genes detected in our previous transcriptional profile of spinal pDyn nuclei at the population level. Italic numbers indicate enrichment (Log2FC) of the given gene in spinal pDyn nuclei compared to unidentified non-pDyn spinal nuclei as described in our previous study. All plots: Color scale indicates Z-score of scaled and log-transformed gene expression, normalized across each row. Size of dot indicates percentage of cells in a given cluster expressing the listed gene. Only genes expressed by at least 25% of cells in at least one cluster are included on plots.

Unsurprisingly, given the established role of dynorphin and dynorphin-expressing spinal cord cells in nociceptive processing, 70 genes from the present study have documented relevance to pain according to the Pain Genes Database [46]. It is interesting to note that some of these genes are predominantly expressed in a few clusters (for example, Nos1 in cluster Inhib2 and Pnoc encoding pronociceptin primarily in clusters Inhib1 and Inhib2), while others are expressed at a high level in several clusters (such as Asic2, Hcn1, Trpm3).

Many voltage-gated ion channels are expressed throughout all neuronal clusters, with notable exceptions Kcnh8, Trpc4, Trpc7, and Pkd2l1. Ligand-gated ion channels exhibit a similar pattern of distribution, though it is notable that the expression of different GABA receptor subunits varied across clusters – for example, Gabra5 appears somewhat selectively expressed in cluster Inhib5, whereas Gabra2 expression appears lower in this particular cluster compared to other clusters. The plots of GPCRs and neuropeptides also call attention to the selectivity of these targets for particular subpopulations of spinal pDyn cells. Dopamine receptor D2 (encoded by Drd2) is more strongly expressed in cluster Inhib5 in comparison to other clusters, while the substance P receptor Tacr1 is localized to cluster Excit8 and the neurokinin B receptor Tacr3 is most strongly expressed in clusters Inhib1 and Inhib2. It is also notable that while μ-opioid receptor expression was detected throughout most neuronal clusters, δ- and κ-opioid receptors were not highly expressed within spinal dynorphin-lineage neurons. With the exception of Nxph1 (encoding Neurexophilin 1), which was expressed by a high percentage of cells in the majority of clusters, most other neuropeptides detected in this assay were selectively expressed in 1–2 clusters. Strikingly, Penk (encoding proenkephalin) was most highly expressed in cluster Excit8 but also, albeit more minimally, in clusters Inhib2 and Inhib3. In addition to the functional classes shown in Fig. 6, additional plots of catalytic receptor, transporter, transcription factor and enzyme expression are provided as supplemental material (supplemental digital content 1, Fig. S3).

Despite the ability to localize expression of selected genes to particular subpopulations of the dynorphin lineage, the present study does not take into account the specificity of any of the detected genes to the dynorphin population. Indeed, the most highly differentially expressed marker for cluster Inhib1, Pde11a, is also highly enriched in excitatory neurons of the somatostatin lineage [21] and therefore not a distinct marker of dynorphin-expressing spinal cord cells. To determine whether genes known to be enriched in spinal dynorphin-lineage cells were preferentially expressed by specific clusters, we compared the list of the top 2,000 most highly variable genes in the present study to the 50 most highly enriched genes detected in our previous population-level transcriptional study of pDyn-GFP spinal nuclei [73]. Genes expressed in inhibitory clusters (such as Gucy2d, Sstr2, Hcrtr2, and Pax8) dominated this list (Fig. 6B), which is unsurprising since inhibitory neurons make up the bulk of the pDyn-GFP population and would therefore drown out genes highly expressed by lesser-represented populations such as excitatory neurons or glia. Nevertheless, Pou4f1 was significantly enriched in the pDyn-GFP population compared to unidentified non-pDyn spinal cells despite its selectivity for excitatory neuron cluster Excit8. Additionally, Pdk2l1, Myo3b, and Slc26a4 were highly enriched in the pDyn-GFP population despite their exclusive expression in the putative central cerebrospinal fluid-contacting neurons of cluster CC11.

3.6. Persistent transcriptional effects of neonatal hindpaw incision

After having identified and characterized 11 clusters of pDyn cells in the adult mouse spinal cord, we next investigated the effects of neonatal hindpaw incision on gene expression within this population. No obvious differences between neonatally injured and naïve littermate samples were apparent in UMAP visualization (Fig. 7A), suggesting a lack of profound transcriptional reprogramming that persists into adulthood. Nevertheless, we hypothesized that subtle yet potentially important injury-mediated differential gene expression may still occur, and that some effects may be specific to certain subpopulations of spinal pDyn neurons, such as those with excitatory or inhibitory phenotypes. Therefore, on the basis of the relative expression of Gad1, Gad2, Slc32a1, Slc6a5, and Slc17a6, each of the 10 previously identified neuronal clusters was classified as either inhibitory or excitatory and assigned to either an inhibitory or excitatory aggregate cluster (Fig. 7B). Thus the inhibitory aggregate was composed of clusters Inhib1–6 and CC11, and the excitatory aggregate was composed of Excit7–8. Because cluster Mixed9 exhibited similar expression of both sets of markers and contained both excitatory and inhibitory neurons (Fig. 1E, 1F; Fig. 2B), it was excluded from either of these aggregates. Since it remains unclear whether the non-neuronal cells which compose cluster Glia10 are truly derived from the dynorphin lineage, this cluster was not considered for injury-evoked differential expression analysis.

Figure 7. Effects of neonatal surgical injury on the transcriptome of pDyn cells in the adult spinal cord.

(A) UMAP plot of spinal pDyn nuclei from both naïve (blue) and neonatally incised (yellow) adult mice did not result in the emergence of any injury-specific clusters. (B) To investigate the effects of neonatal hindpaw incision across all inhibitory or all excitatory neurons, each cluster was evaluated for relative mean expression of inhibitory markers Gad1, Gad2, Slc32a1, and Slc6a5 vs. excitatory marker Slc17a6 and combined accordingly to form an inhibitory aggregate (red) and an excitatory aggregate (blue) phenotype. Mixed cluster 9 and non-neuronal cluster 10 (gray) were not included in either aggregate population. (C) Violin plots representing scaled log-transformed normalized gene expression (gene UMIs/total cell UMIs) reveal persistent upregulation of long non-coding RNA Gm26848 and downregulation of Fth1, Pcsk1n, and Ubb in inhibitory pDyn neurons following neonatal incision (see also supplemental digital content 1, Table ST1). (D) Similar trends were observed in excitatory pDyn neurons, with the addition of Oxr1, which was upregulated, and Gm42418, which was downregulated after injury (see also supplemental digital content Table ST2). (E) To investigate the long-term effects of neonatal injury on each identified cluster of pDyn nuclei, DEG analysis was performed between naïve and P3-incised groups on a cluster-by-cluster basis instead of the aggregate clusters described in panels B-D. In addition to those genes up- or downregulated as shown in panels C-D, the expression of the indicated genes was significantly altered following neonatal hindpaw incision (see also supplemental digital content 1, Table ST3). All violin plots depict scaled and log-transformed normalized expression. Box-and-whisker overlay indicates median and 1^st/3^rd quartiles. P-adj < 0.05 for all listed genes, Wilcoxon Rank Sum Test adjusted with Bonferroni correction.

Differential gene expression analysis of the inhibitory aggregate revealed that long intergenic non-coding RNA Gm26848 was significantly upregulated in adulthood following neonatal injury while Fth1, Pcsk1n, and Ubb were downregulated (Fig. 7C; see also supplemental digital content 1, Fig. S4 and Table ST1). Similarly, differential gene expression analysis in the excitatory aggregate revealed that along with Gm26848, Oxr1 was also upregulated following neonatal injury, while Gm42418 was downregulated in addition to Fth1, Pcsk1n and Ubb (Fig. 7D; see also supplemental digital content 1, Fig. S4 and Table ST2). That a strikingly similar set of genes (Gm26848, Fth1, Pcsk1n, Ubb) was significantly modulated by injury in both inhibitory and excitatory aggregates suggests a broad, pan-neuronal transcriptional response to neonatal injury rather than discrete effects on inhibitory or excitatory subpopulations.

However, given the transcriptional heterogeneity of both of these aggregates, we hypothesized that cluster-specific, injury-dependent differential gene expression may also occur. Incision-mediated differential gene expression analysis on a cluster-by-cluster basis revealed a small number of other targets that were significantly up- or downregulated in specific neuronal clusters, in addition to those identified in the aggregate analysis (Fig. 7E; see also supplemental digital content 1, Fig. S4 and Table ST3). Notably, Gal expression was decreased following injury in cluster Inhib1, Dync1h1 (encoding the dynein heavy chain) was decreased in both clusters Inhib2 and Inhib3, and Tmem181a was upregulated in clusters Inhib3, Excit8, and Mixed9.

4. Discussion

Dynorphin neurons are key components of a spinal inhibitory circuit which is persistently compromised by neonatal tissue damage via multiple mechanisms, including a reduction in the intrinsic excitability of pDyn neurons and a weakening of their GABAergic synapses onto spinoparabrachial neurons [16]. Using single-nucleus RNAseq to characterize, for the first time, the molecular profile of a genetically-defined subpopulation of dorsal horn (DH) neurons following tissue damage, the present results reveal modest changes in gene expression within adult spinal pDyn neurons following neonatal incision, which occurred across both inhibitory and excitatory subtypes. In addition, the data point to the existence of eleven transcriptionally distinct subpopulations of dynorphin-lineage cells within the dorsal spinal cord, consistent with a highly complex functional organization of somatosensory networks in the CNS.

4.1. Transcriptional heterogeneity of spinal dynorphin-lineage neurons

Single-cell transcriptomic studies have illustrated the diversity of neurons that comprise the central and peripheral nervous systems [36; 55; 70; 91] and identified subpopulations of functionally distinct neurons within a wider neurochemically-defined population [59]. Our current single-nucleus RNAseq analysis of spinal dynorphin lineage (pDyn) nuclei revealed 10 transcriptionally distinct neuronal subpopulations (or clusters), of which 7 were inhibitory, 2 were excitatory, and one was composed of both inhibitory and excitatory neurons.

Inhibitory DH neurons can be classified into four largely non-overlapping subpopulations based on expression of galanin, nNOS, NPY and parvalbumin [13; 77], which may modulate discrete sensory modalities related to pain and/or itch [25; 30; 43; 62]. The spinal dynorphin lineage contains likewise non-overlapping subpopulations of inhibitory neurons expressing the marker genes Gal in cluster Inhib1, Nos1 in Inhib2 and Npy in Inhib6. Additionally, Pvalb expression appeared selective for cluster Inhib5, although the enrichment was not sufficient to be considered a marker gene. We also identified inhibitory clusters enriched for neurokinin B (Tac2; Inhib3) and natriuretic peptide C (Nppc; Inhib4). A final cluster of inhibitory neurons, named CC11 due to its enrichment of known cerebrospinal fluid-contacting neuron (CSF-cN) marker genes, is extremely dissimilar to those neurons included in the first 6 clusters. CSF-cNs are GABAergic [24; 32; 40], but the present results are the first to reveal that at least some of these cells are derived from the dynorphin lineage. While additional work is required to elucidate the functional role of each transcriptional cluster in somatosensation, prior studies suggest that the clusters that shape pain sensitivity may include Inhib1 (expressing Gal), Inhib2 (Nos), Inhib5 (Pvalb), Excit7 (Tac1) and Excit8 (Cck), with Inhib6 (Npy) potentially important for both pain and itch [12; 27; 31; 39; 62; 75; 87]. The finding that the spinal dynorphin lineage encompasses such a diversity of inhibitory interneurons, in addition to excitatory neurons and possibly a small percentage of astrocytes and oligodendrocytes, suggests that genetic strategies designed to manipulate dynorphin neurons using Pdyn^Cre mice likely affect a wide variety of cell types in the spinal cord. This highlights the value of using intersectional genetic approaches to more precisely target certain cellular subpopulations within the dynorphin lineage.

Excitatory pDyn neuron clusters were characterized by expression of Cacna2d1, Grid1, Cntnap5b and Fam19a1. Although Fam19a1 has been detected in various brain regions where it may modulate anxiety, long-term memory, and fear conditioning [90], as well as in a subset of dorsal root ganglia (DRG) neurons [23; 80], its role in the spinal cord is unknown. Although clusters Excit7, Excit8, and Mixed9 were less transcriptionally distinct from one another compared to the molecular diversity of the inhibitory clusters, Excit7 exhibited notable enrichment for Fam19a2 (encoding a chemokine-like secreted protein of unknown function) and for Tac1 (encoding substance P). Lamina I projection neurons, many of which are characterized by expression of the substance P receptor [20; 79], receive inhibitory synaptic input from pDyn neurons in the DH [15; 32], but the existence of functional excitatory pDyn neuronal input to projection neurons has not yet been investigated.

4.2. Effects of neonatal surgical injury on gene expression in adult spinal pDyn neurons

Both inhibitory and excitatory pDyn neurons showed increased expression of Gm26848 and decreased expression of Fth1, Ubb and Pcsk1n in neonatally incised mice compared to naïve controls. Excitatory neurons additionally expressed increased levels of Oxr1 and Gm42418 following early tissue damage. Notably, several of these injury-evoked differentially expressed genes (DEGs) are related to cellular stress responses. For example, Fth1 encodes the ferritin heavy chain (H-ferritin), which sequesters free iron contributing to oxidative damage [3]. Meanwhile, the Oxr1 gene, which encodes a protein that protects against oxidative stress-induced DNA damage and is upregulated in the both the human and rodent spinal cord in the setting of amyotrophic lateral sclerosis [60], was more highly expressed in adult excitatory pDyn neurons after neonatal injury. Ubb, encoding ubiquitin B, is another gene which is more highly expressed under stressful conditions [11], likely reflecting the increased need for degradation of misfolded or damaged proteins [74]. However, in our analysis, Ubb was downregulated in adult pDyn neurons following neonatal injury, which seems counterintuitive given the upregulation of Oxr1. Importantly, the membrane density and stability of several ion channels and neurotransmitter receptors, including AMPARs and GABA_ARs, are regulated in part by ubiquitylation [1; 47; 67; 71; 72], which could contribute to changes in synaptic integration and network excitability.

Pcsk1n encodes proSAAS, a prohormone which undergoes posttranslational cleavage to produce short peptides which are widely distributed throughout the brain and may play a role in food intake behavior [85; 89]. Two of these derivative peptides, SAAS and LEN, have been detected in the rat DH by immunohistochemistry [28]. Meanwhile, proSAAS in its unprocessed form inhibits proprotein convertase 1 (PC1), an enzyme involved in the cleavage of other propeptides including pro-neuropeptide Y, proinsulin, proopiomelanocortin, proenkephalin and prodynorphin [15; 61; 63; 95]. Pcsk1n is enriched within pDyn neurons and exhibits a significant increase in expression between P7 and P80 [73], raising the possibility that neonatal injury may have persistently disrupted this developmental increase in Pcsk1n expression. Finally, two long intergenic non-coding RNAs (lincRNAs), Gm26848 and Gm42418, were differentially expressed in adult pDyn neurons following neonatal injury. Although the function of these specific lincRNAs has not been investigated, lincRNAs can modulate gene expression through epigenetic mechanisms and via interactions with cellular machinery that processes nascent transcribed mRNA [5; 65; 66]. However, we note that this study was not optimized to evaluate changes in lncRNA expression, which may require higher sensitivity (i.e., increased read depth and read length) due to the lower prevalence of lncRNA transcripts compared to protein-coding transcripts [54; 88; 94].

4.3. Limitations

The present analysis identified only a relatively small number of genes within adult spinal pDyn neurons whose expression was significantly modulated by neonatal tissue damage. While this may indicate a surprisingly limited role for long-term alterations in neuronal gene expression in the priming of developing spinal nociceptive circuits by early life injury, it is also possible that increased read depth or sample size would have revealed additional genes that were regulated by neonatal incision. Indeed, we were surprised to note the low expression of Pdyn in our dataset despite the high percentage of adult pDyn-GFP cells which expressed Pdyn mRNA in our in situ hybridization experiments [73], suggesting that many of the sampled nuclei failed to express sufficient levels of Pdyn to be detected given the per-nucleus read depth obtained in the present study. It is therefore possible that, especially in the smaller clusters, our analysis lacked the statistical power to detect the full complement of injury-evoked DEGs. In addition, only a fraction of the analyzed pDyn cells receive input from primary afferents innervating the damaged region of the hindpaw, which could lead to a dilution effect whereby selective changes in this subset of spinal pDyn cells become harder to detect.

The transcriptional nature of this study also does not capture the myriad post-transcriptional regulations which ultimately contribute to functional gene expression in pDyn neurons. Furthermore, many of the injury-evoked DEGs identified in this analysis exert their effects through post-translational mechanisms, such as ubiquitination (Ubb), nuclear translocation (Fth1), and proprotein cleavage (Pcsk1n), and the downstream consequences of changes in the expression of these genes cannot be determined from the current data. Finally, it remains feasible that changes in network connectivity, or alterations occurring in spinal cord cells derived from outside the dynorphin lineage, critically contribute to the functional deficits in dynorphin inhibitory circuits within the adult DH following early life injury. For example, spinal microglia are absent from the dynorphin lineage, but are well known to compromise the efficacy of synaptic inhibition within the DH [22; 29] and are critical for the priming of developing nociceptive pathways by neonatal injury [9; 58].

5. Conclusions

In summary, this study yields new insight into the transcriptional diversity of spinal cord cells derived from the dynorphin lineage and identifies differential gene expression within adult pDyn neurons resulting from neonatal incision, which could contribute to the previously documented changes in the intrinsic and synaptic properties of these key interneurons following early tissue damage.

Supplementary Material

Supplementary Materials: figures, tables_2

Supplemental Data: Differentially expressed genes across dendrogram nodes

Supplementary Materials: figures, tables_3

Supplemental Data: Complete list of marker genes for each cluster