Endogenous Opioid Signaling Regulates Spinal Ependymal Cell Proliferation

Summary

After injury, mammalian spinal cords develop scars to confine the lesion and prevent further damage. However, excessive scarring can hinder neural regeneration and functional recovery^1,2. These competing actions underscore the importance of developing therapeutic strategies to dynamically modulate scar progression. Previous research on scarring has primarily focused on astrocytes, but recent evidence suggests that ependymal cells also participate. Ependymal cells normally form the epithelial layer encasing the central canal, but they undergo massive proliferation and differentiation into astroglia following certain injuries, becoming a core scar component^3–7. However, the mechanisms regulating ependymal proliferation in vivo remain unclear. Here, we uncover an endogenous kappa (κ) opioid signaling pathway that controls ependymal proliferation. Specifically, we detect expression of the κ opioid receptor, OPRK1, in a functionally under-characterized cell type known as cerebrospinal fluid-contacting neuron (CSF-cN). We also discover a neighboring cell population that expresses the cognate ligand, prodynorphin (PDYN). Surprisingly, whereas κ opioids are typically considered inhibitory, they excite CSF-cNs to inhibit ependymal proliferation. Systemic administration of a κ antagonist enhances ependymal proliferation in uninjured spinal cords in a CSF-cN-dependent manner. Moreover, a κ agonist impairs ependymal proliferation, scar formation and motor function following injury. Altogether, our data suggest a paracrine signaling pathway whereby PDYN⁺ cells tonically release κ opioids to stimulate CSF-cNs and suppress ependymal proliferation, revealing an endogenous mechanism and potential pharmacological strategy for modulating scarring after spinal cord injury.

Cerebrospinal fluid (CSF) flows through the central canal of the spinal cord, transporting metabolites into and away from the nervous system. The canal, which traverses the length of the cord along its midline, is lined by a layer of ciliated epithelial cells that retain proliferative capacity in adulthood. Following spinal cord injury, these ependymal cells divide and differentiate into astrocytes, which assemble into scar tissue that limits secondary damage^3–7. Currently, it remains unknown how ependymal cell proliferation is regulated endogenously under healthy or pathological conditions, although the inhibitory neurotransmitter GABA is believed to be an instructive signal⁸.

One potential source of GABA is a group of morphologically distinct cells called CSF-contacting neurons (CSF-cNs) that are highly conserved across vertebrate species. In mice, the somata of these cells are scattered within the ependyma, from which they extend a thin projection into the central canal, terminating in a bulbous enlargement that contacts the CSF⁹. This characteristic morphology is reminiscent of olfactory sensory neurons and, indeed, CSF-cNs have been suggested to possess sensory functions, such as detecting changes in pH or osmolarity within the CSF^10,11 or, in zebrafish, detecting the presence of meningitis bacteria¹². Zebrafish CSF-cNs have also been proposed to mechanically detect spinal curvature and provide motor feedback for controlling tail movement^13,14. Nonetheless, whether mammalian counterparts have preserved these same functions or acquired new physiological roles is unclear.

In this study, we focus on the role of CSF-cNs in injury by asking whether and how they regulate ependymal cell proliferation under normal or pathological conditions. We find that CSF-cNs are constitutively activated by kappa (κ) opioids, which are produced by a neighboring cell type within the ependyma. Paracrine opioid signaling tonically suppresses ependymal cell proliferation. Importantly, after partial spinal cord transection, we show that an exogenously administered κ agonist can utilize the same cellular pathway to reduce ependymal proliferation and thus impair motor function, providing proof-of-principle evidence that pharmacological modulation of this system may be leveraged to tune scar formation and recovery.

Genetic labelling of CSF-cNs

The dearth of knowledge regarding mammalian CSF-cNs reflects, in large part, the lack of specific genetic access to these cells. Mouse CSF-cNs express the PKD2L1 ion channel, and thus most previous studies have used a Tg(PKD2L1-Cre) mouse line¹⁰ to fluorescently label these cells. This method, in conjunction with the characteristic shape and location of CSF-cNs, has permitted their identification for electrophysiological recording and imaging^{10,11,15–17}. However, we found that this transgenic line also marks many other cells in the spinal cord (Extended Data Fig. 1a), making it difficult to trace and manipulate CSF-cNs selectively. This problem does not appear to be the result of transient Pkd2l1 expression during development since non-CSF-cNs were also labelled when a fluorescent reporter was introduced in adulthood using an AAV vector (Extended Data Fig. 1b). To bypass this issue, we asked whether particular AAV serotypes would show specific tropism for CSF-cNs when introduced into the CSF via the brain ventricles (i.e., intracerebroventricular (i.c.v.) injection, see Methods), which are connected to the central canal. Remarkably, AAV serotype 2/2 infected a set of cells around the central canal in a highly specific manner, even without the need to utilize a genetic Cre driver line. The AAV2-labelled cells express the CSF-cN marker, PKD2L1 (Fig. 1a) and demonstrate the stereotypical morphology of CSF-cNs (Fig. 1b). Using this method, we delivered an alkaline phosphatase (PLAP) reporter into CSF-cNs to reveal their detailed morphology. We observed that CSF-cNs send projections to the ventral white matter, in addition to the central canal (Fig. 1c). The same projection pattern was observed for CSF-cNs at all levels of the spinal cord up to the brainstem (Extended Data Fig. 1c). In sagittal sections, these ventral projections were seen to form a bundle that ran parallel to the central canal (Extended Data Fig. 1d). By titrating the concentration of injected AAV to achieve sparse labelling, the long projection of a CSF-cN could be seen to first extend ventrally from the soma, then turn 90° and extend rostrally for at least 400 μm (Extended Data Fig. 1e and g). These morphological features are comparable to those described in zebrafish¹⁸ and, more recently, in mice using a similar but independently developed AAV injection method employing AAV2/1, which is reportedly less specific and preferentially labels CSF-cNs that are located ventrally in the ependyma¹⁹.

Fig. 1. κ-opioid receptor and ligand are expressed in the ependymal region of the mouse spinal cord.

a, Genetic labelling of CSF-cNs by intracerebroventricular AAV injection. A representative coronal spinal cord section from a mouse injected with AAV2/2-tdTomato is shown. Some cells around the central canal (boxed area) are specifically labelled (magenta). Punctate signals below are ventral projections from these cells. Dotted line demarcates the boundary between gray and white matter. b, Magnified view of the boxed area in panel a, showing the overlap between AAV labelling (magenta) and immunohistochemical signal for PKD2L1 (green), a known marker of CSF-cNs. Nuclear stain by DAPI is in blue. c, Morphology of labelled CSF-cNs. Somata of the labelled cells (arrows) extend fine projections into the central canal (dotted line). Arrowheads indicate the terminal enlargement, a stereotypical morphological feature of CSF-cNs. Ependymal cells are identified by their strong DAPI signal (blue). d, Projections from CSF-cNs to ventral white matter, as revealed by AAV-mediated PLAP expression. Morphological features shown in a-d are representative of altogether >100 cells in >5 mice. e, Heat map showing the transcript expression level of known CSF-cN markers (Pkd2l1, Gad1, Gad2) and that of the κ opioid receptor, Oprk1, obtained by bulk mRNA sequencing. Color scale is based on median-of-ratios calculation by DESeq2. Each column represents the CSF-cN-enriched preparation from one mouse. The expression of the housekeeping gene Actb remains constant across samples. f, Immunohistochemical detection of OPRK1 (magenta) in CSF-cNs (green). Merged signal (white) on top; images from individual fluorescence channels below. g, Expression pattern of Pdyn, which encodes OPRK1 ligands, in the spinal ependymal region of a Pdyn^Cre;Rosa26^LSL-tdTomato mouse. Pdyn⁺ cells (magenta) are located close to CSF-cNs (green). h, Immunostaining confirmation of protein PDYN expression (green) in Pdyn^Cre-labelled cells (magenta). Scale bars: 100 μm (a) and 20 μm (b,c,f-h), with each experiment repeated three times.

κ receptor and ligand in the ependyma

Gaining specific genetic access to CSF-cNs enabled us to obtain clues about their potential function by transcriptome profiling. We expressed tdTomato in CSF-cNs of 4 wildtype C57BL6/J mice by AAV injection, then captured the tdTomato⁺ cells by fluorescence-activated cell sorting (FACS), generated independent cDNA libraries, and performed bulk sequencing. As reflected by the presence of glial transcripts (Extended Data Fig. 2a), our cell preparations were apparently contaminated to differing degrees by glial fragments that likely adhered to the surface of tdTomato⁺ cells. Notably, tdTomato transcripts were also represented at different levels across the 4 samples (Fig. 1d, top row), which we used to our advantage as a reference to identify transcripts specifically enriched in CSF-cNs. Indeed, known CSF-cN markers, such as Pkd2l1, Gad1 and Gad2^10,20, were among the identified genes (Fig. 1d), together with those of relevance to GABA metabolism and neurotransmission (Extended Data Fig. 2b and c). Intriguingly, expression of the κ opioid receptor gene, Oprk1, correlated strongly with that of tdTomato transcript (Fig. 1d), suggesting that it is reasonably enriched in CSF-cNs. Consistent with this, OPRK1 protein was readily detected in CSF-cNs by immunohistochemical staining (Fig. 1e).

Next, we asked whether there is any κ opioid ligand in the vicinity of CSF-cNs. Endogenous OPRK1 peptide ligands, including dynorphin A (DYNA), dynorphin B and big dynorphin, are derived from enzymatic cleavage of the precursor peptide, prodynorphin (PDYN)²¹. We examined spinal cord sections of Pdyn^Cre;Rosa26^LSL-tdTomato mice injected i.c.v. with AAV2/2-GFP, in which Pdyn⁺ cells were labelled by tdTomato and CSF-cNs by GFP. Interestingly, in addition to previously reported Pdyn⁺ cells in the dorsal horn²² (Extended Data Fig. 1h, left), we observed prominently labelled cells at the two poles of the ependyma, immediately apposing CSF-cNs (Fig. 1f). Immunostaining with an antibody against DYNA also highlighted the same region around the central canal (Fig. 1g), confirming expression at the protein level. We made use of the MORF3 mouse line²³ to label the ependymal Pdyn⁺ cells sparsely for morphological tracing. In coronal spinal cord sections, Pdyn⁺ cells located at the dorsal pole of the ependyma extended a long process dorsally across the border between gray and white matters (Extended Data Fig. 1f and g), sometimes reaching all the way to the edge of the spinal cord. Ventral Pdyn⁺ cells projected ventrally instead, arborizing in the same field where CSF-cNs’ projections were found (Extended Data Fig. 1f and g). Both Pdyn⁺ cell types morphologically resembled radial glia but appeared to be negative for the astroglial marker GFAP. Instead, they were positive for the transcription factor SOX2 and the intermediate filament protein NESTIN (Extended Data Fig. 1h), two common markers of neural stem cells and progenitors previously shown to mark a subset of ependymocytes²⁴.

In all, we have identified a novel population of PDYN⁺ cells in the ependymal region that express κ opioids and reside near CSF-cNs that express the κ opioid receptor, hinting at some neurochemical communication between the two cell types.

κ opioids excite CSF-cNs

Activation of OPRK1 is typically believed to trigger an inhibitory cellular response through a Gα_i or Gβγ-mediated pathway^25,26. However, there are a few suggestions of excitatory responses such as in cultured astrocytes involving L-type voltage-gated calcium channels (Ca_v channels)^27,28, as well as in the rat nucleus raphe magnus involving the mobilization of internal calcium stores²⁹, although these are generally less well-characterized. We were curious to see whether OPRK1 stimulation would inhibit or enhance the activity of CSF-cNs. We started by performing current-clamp recording on individual tdTomato-labelled CSF-cNs in acutely harvested spinal cord slices. At near resting membrane potential (−60 mV), CSF-cNs exhibited spontaneous action potentials at baseline as previously reported, owing to the spontaneous opening of PKD2L1 channels^11,15. Interestingly, local application of the κ opioid ligand, DYNA_1–17 (1 μM), increased the firing frequency above baseline (Fig. 2a, black). In addition to OPRK1, κ opioid ligands have been suggested to act through bradykinin receptors³⁰, NMDA receptors^31,32 and/or delta and mu opioid receptors³³. Although none of these other candidate receptors are highly enriched in CSF-cNs based on our transcriptomic data (Extended Data Fig. 2d), we nevertheless evaluated whether OPRK1 is required for the response of CSF-cNs to κ opioid ligands. Indeed, the DYNA-evoked response was blocked when a selective OPRK1 antagonist, Nor-BNI (0.1 μM), was bath applied (Fig. 2a, red).

Fig. 2. Activation of κ-opioid receptor excites CSF-cNs.

a, OPRK1-dependent electrical responses of CSF-cNs to the κ opioid ligand, DYNA. Representative current-clamp recording traces showing spontaneous and DYNA-evoked firing in CSF-cNs in the absence (black) and presence (red) of an OPRK1 inhibitor, Nor-BNI. Collective data on right. Individual cells are in light colors and averages in dark colors. Repeated measures two-way ANOVA: DYNA × Nor-BNI (F_1,14=10.99, p=0.0051); Šidák posthoc pairwise comparisons of baseline versus +DYNA: −Nor-BNI (p=0.0002) and +Nor-BNI (p=0.8172, not significant); n = 8 and 8 cells. b, OPRK1-dependent Ca²⁺ responses of CSF-cNs to DYNA. Representative ΔF/F traces showing the responses of GCaMP5G-expressing CSF-cNs to local DYNA application in the absence (black) or presence (red) of Nor-BNI. Local application of a high K⁺ solution was used to reveal all responsive neurons in spinal cord slices. Each trace is from a single cell. c, ΔF/F images for the cells in panel b (arrowheads). Images are temporal averages over 10 sec of baseline or for the duration of the stimuli. CC: central canal. Scale bars are 20 μm. d, Collective data comparing responses of CSF-cNs to DYNA in the absence (black) and presence (red) of Nor-BNI. Each dot shows the integral DYNA response of a single cell normalized to the high-K⁺ response (see Methods). Mean ± SD; Kruskal-Wallis non-parametric test and Dunn’s post-hoc pairwise comparisons: p=0.0425; n = 34 and 13 cells. e, DYNA-evoked Ca²⁺ responses under bath application of TTX or synaptic blockers. Control is same as –Nor-BNI in d. Mean ± SD, Kruskal-Wallis non-parametric test combined with d and Dunn’s post-hoc pairwise comparisons with control: TTX (p=0.9689, not significant) and synaptic blockers (p>0.9999, not significant); n = 34, 39, and 25 cells.

We confirmed the excitatory effect of OPRK1 activation by performing calcium imaging of GCaMP5G-expressing CSF-cNs in spinal cord slices from Tg(PKD2L1-Cre);Polr2a^{GCaMP5G-tdTomato} mice. In response to DYNA (1 μM), a considerable number of CSF-cNs increased their calcium activity significantly above baseline (Fig. 2b). Two other OPRK1 agonists, Nalfurafine (1 μM) and BRL-52537 (10 μM), also elicited similar calcium responses in many CSF-cNs (Extended Data Fig. 3). We noticed that not all CSF-cNs responded, and those that did respond continued to spike even after stimulus withdrawal, likely reflecting inefficient perfusion of agents into and out of the tissue slices. However, it was clear that the responses were OPRK1-dependent since they were potently blocked by the selective OPRK1 antagonists, Nor-BNI (0.1 μM, Fig. 2b and Extended Data Fig. 3b) or DIPPA (1 μM, Extended Data Fig. 3a). Moreover, agonists for bradykinin, delta or mu opioid receptors did not produce any significant Ca²⁺ responses (Extended Data Fig. 4a), further corroborating that κ agonists act specifically through OPRK1 in CSF-cNs.

To ascertain whether the response to DYNA is intrinsic to CSF-cNs or synaptically driven, we included the broad-spectrum voltage-gated sodium (Na_v) channel blocker, tetrodotoxin (TTX), in the bath to block action potential firing in upstream neurons. The response of CSF-cNs to DYNA persisted in the presence of TTX (Fig. 2c). Previously, CSF-cNs were found to express functional AMPA/kainate-type ionotropic glutamate receptors¹⁵, GABA_A receptors^11,34, nicotinic acetylcholine receptors^15,35 and glycine receptors¹¹. We therefore also applied a cocktail of synaptic blockers to inhibit fast glutamatergic, GABAergic, cholinergic and glycinergic neurotransmission. Again, no significant effect was seen on DYNA-evoked responses (Fig. 2c). Moreover, DYNA did not influence the rate of spontaneous postsynaptic events in voltage-clamp recordings of CSF-cNs (Extended Data Fig. 5e). Together, our experiments suggest that a functional excitatory OPRK1 signaling pathway is intrinsic to CSF-cNs.

We next used pharmacological tools to gain initial insights into how OPRK1 activation regulates CSF-cN excitability. A canonical GPCR-mediated transduction pathway that promotes a rise in intracellular calcium involves activation of G_q-phospholipase C (PLC) signaling. Indeed, a Gα_q inhibitor (YM254890, 10 μM) diminished DYNA-evoked calcium responses in CSF-cNs (Extended Data Fig. 4a, blue region). Inhibition was also observed with the PLC inhibitor, U73122 (10 μM) but not with its inactive analog (U73343, 10 μM) (Extended Data Fig. 4b, blue region). PLC catalyzes the hydrolysis of phosphatidylinositol 4,5-bisphosphate (PIP₂) to inositol 1,4,5-triphosphate (IP₃) and diacylglycerol (DAG). IP₃ can elicit a release of calcium from intracellular stores, which subsequently opens calcium-sensitive membrane ion channels. In fact, in rat nucleus raphe neurons, regulation of HCN channels by internal calcium has been proposed to explain the excitatory response triggered by OPRK1 stimulation²⁹. In CSF-cNs, however, DYNA responses were not significantly affected by depletion of internal calcium stores with thapsigargin (4 μM) or the HCN blocker, Ivabradine (10 μM) (Extended Data Fig. 4b, yellow region). An alternative pathway downstream of PLC involves the activation of TRPC channels by DAG. CSF-cNs express Trpc1 and Trpc6 transcripts (Extended Data Fig. 2e) and are particularly abundant in PKD2L1, a TRPP channel that may also be positively regulated by PLC³⁶. Nevertheless, by voltage-clamp recording of CSF-cNs (at −80 mV, Extended Data Fig. 5a–d), we did not detect any macroscopic current or an increase in single-channel opening events upon DYNA application, arguing against membrane depolarization through these non-selective cation channels as a mechanism of DYNA-mediated excitation.

Finally, we considered the possibility of DAG activating protein kinase C (PKC), which is known to modulate the activity of Ca_v channels via phosphorylation, consequently increasing their open probability or changing their voltage dependence and/or kinetic properties³⁷. Indeed, the cell-permeable PKC inhibitor, chelerythrine chloride (10 μM), reduced DYNA-evoked calcium responses in CSF-cNs (Extended Data Fig. 4b, green region). Further experiments involving individual Ca_v channel blockers suggest that external calcium enters CSF-cNs through P/Q-, N- and T-type Ca_v channels (i.e., Ca_v2.1, 2.2 and 3, respectively) (Extended Data Fig. 4b, green region). Consistent with this, our bulk sequencing data indicate that certain Ca_v2 and 3 channel subunits are enriched in CSF-cNs (Extended Data Fig. 2e). Previous pharmacological studies have also suggested roles for T-type and HVA calcium channels (which include P/Q- and N-type) in mediating spontaneous and evoked spiking in CSF-cNs¹⁶.

To determine how DYNA regulates the properties of Ca_v channels, we examined the Ca_v current responses of CSF-cNs to voltage steps by recording under conditions where sodium and potassium currents were inhibited (see Methods). We found that DYNA increases the peak Ca_v current amplitude without an obvious shift in the current-voltage relationship (Extended Data Fig. 6), suggesting an increase in channel open probability or single-channel conductance. Notably, although DYNA increases the firing rate of CSF-cNs (Fig. 2a), the amount of Ca_v current developed during DYNA application is rather insignificant at the resting membrane potential of ~ −55 mV^11,15 (Extended Data Fig. 6). In other words, additional depolarization is needed to drive Ca_v channel activation. This depolarization may be provided by the spontaneous activity of PKD2L1 channels.

In sum, our pharmacological data suggest that OPRK1 activation of CSF-cNs mainly involves a G_q-PLC-PKC cascade, leading ultimately to increased calcium influx through Ca_v channels and action potential firing (Extended Data Fig. 4c).

CSF-cNs suppress ependymal proliferation

What action does κ signaling exert in the ependymal region? As noted above, ependymal proliferation can be inhibited by GABA⁸. CSF-cNs are well poised to be the cellular origin of this GABA-mediated regulatory mechanism because they are GABAergic^20,38,39 and reside near ependymal cells. Hence, we tested whether ependymal cell proliferation would be affected by interfering with κ signaling of CSF-cNs in vivo. We injected healthy adult wildtype C57BL6/J mice with a κ agonist (Nalfurafine, 20 μg/kg, i.p.) or antagonist (Nor-BNI, 10 mg/kg, i.p.) daily for 8 consecutive days. Control mice were injected with the respective vehicles. We then subjected the mice to an overlapping but staggered week of treatment with EdU (50 mg/kg, i.p.), a thymidine analog that is incorporated into newly synthesized DNA during cell division (Fig. 3a). After that, we prepared spinal cord sections from these animals and counted the number of EdU⁺ ependymal cells. We found that the average EdU⁺ cell count was not significantly affected by the κ agonist, although a slight decreasing trend was seen when cell counts from each section were tallied as a distribution histogram (Fig. 3bi). Most likely, the low number of proliferating ependymal cells at baseline makes it difficult to detect small reductions. On the other hand, the κ antagonist significantly upregulated the number of EdU⁺ cells in the spinal ependymal region, as evident from both representative histological images and collective data (Fig. 3bii). Thus, we conclude that a constitutively active κ signaling pathway normally suppresses ependymal cell proliferation.

Fig. 3. Constitutive κ signaling via CSF-cNs suppresses ependymal proliferation in vivo.

a, Drug injection scheme for labelling proliferating cells after systemic treatment with κ agonist (20 μg/kg Nalfurafine), antagonist (10 mg/kg Nor-BNI) or respective vehicles. b, Representative images of EdU⁺ cells (white) in the ependyma (blue) in 4 sections from a mouse of each treatment group. (Right) Distribution histogram showing the frequency of sections containing the indicated number of EdU⁺ cells. (Inset) Bar graph summary. Each dot represents the cross-sectional average from a mouse. Mean ± SD; Two-sided Welch’s t-test on log-transformed data: p=0.0881 (not significant); n = 7 and 7 mice. c, Same as b but for saline and antagonist-treated mice. Mean ± SD; Two-sided Welch’s t-test on log-transformed data: p=0.0021; n = 9 and 9 mice. d, Nor-BNI’s effect after CSF-cN ablation. Representative images from DTX-injected mice immunostained with anti-PKD2L1 to reveal CSF-cNs (top left) and collective data (bottom left) are shown. Mean ± SD; Two-sided Welch’s t test: p<0.0001; n = 9 and 13 mice. Number of EdU⁺ cells is also shown (right). Mean ± SD; Two-way ANOVA on log-transformed data: Nor-BNI × DTR (F_1,47=5.512, p=0.0232); Šidák posthoc pairwise comparisons: saline versus Nor-BNI in DTR− mice (p=0.0059), saline versus Nor-BNI in DTR+ mice (p=0.9843, not significant), and saline-treated DTR− mice versus DTR+ mice (p=0.0005); n = 15, 15, 11, and 10 mice. e, Nor-BNI’s effect in ependyma-specific Pdyn knockout mice. Immunohistochemical detection of PDYN in the ependyma and the dorsal horn of tamoxifen (TAM)-injected mice (left), and the number of EdU⁺ cells (right) are shown. Mean ± SD; Two-way ANOVA on log-transformed data: Nor-BNI × KO (F_1,32=8.768, p=0.0057); Šidák posthoc pairwise comparisons: saline versus Nor-BNI in controls (p=0.0041), saline versus Nor-BNI in knockout mice (p=0.8757, not significant), and saline-treated control versus knockout mice (p=0.0056); n = 10, 8, 8, and 10 mice. All scale bars are 20 μm.

To determine whether the robust antagonist effect depends on CSF-cNs, we ablated these neurons by generating Tg(PKD2L1-Cre);Rosa26^iDTR mice in which the diphtheria toxin receptor (DTR) is selectively expressed in CSF-cNs. Diphtheria toxin (DTX, 50 μg/kg, i.p.) was administered for 5 consecutive days prior to the same antagonist and EdU injection scheme described above. Immunostaining with a PKD2L1 antibody verified that most CSF-cNs in these animals were eliminated compared to DTX-treated control Rosa26^iDTR mice that did not carry the Cre allele (Fig. 3ci). As in wildtype mice, the κ antagonist Nor-BNI enhanced ependymal proliferation in control Rosa26^iDTR animals (Fig. 3cii). When CSF-cNs were ablated, vehicle-injected animals showed significantly more EdU⁺ ependymal cells, demonstrating that CSF-cNs exert constant suppression on ependymal proliferation under normal physiological conditions. Moreover, the proliferation-enhancing effect of the κ antagonist was lost after CSF-cN ablation (Fig. 3cii), further supporting the notion that CSF-cNs are necessary for κ opioid-mediated suppression of proliferation.

We wondered whether the tonic κ signaling originates from the PDYN⁺ cells we identified in the ependymal region, or from other PDYN⁺ cells in the spinal cord, most notably those in the dorsal horn. We noticed that SOX2 is expressed only in the former PDYN⁺ population (Extended Data Fig. 1h). Thus, we generated Sox2^CreERT2;Pdyn^fl/fl mice to specifically knock out Pdyn expression in the ependymal area (Fig. 3di) and repeated the Nor-BNI experiment. The result recapitulated that observed with CSF-cN ablation. Specifically, when treated only with vehicle, knockout animals exhibited a higher number of EdU⁺ ependymal cells than controls, consistent with a loss of proliferation suppression, which was further confirmed by an abrogation of the Nor-BNI effect in the knockouts (Fig. 3dii). These results suggest that CSF-cNs indeed receive constant κ opioid stimulation from their neighboring PDYN⁺ cells, plausibly through the CSF.

κ agonist modulates post-injury recovery

Injury to the midline of the spinal cord has been shown to elicit extensive ependymal cell proliferation. Most ependyma-derived cells differentiate into astrocytes that migrate to the lesion site and form the center of glial scar tissues^3–7. At 4 months after injury, they contribute to ~50% of newly generated astrocytes in the spinal cord⁶, with the remaining half derived from resident astrocytes that constitute the periphery of the scar^4–6. Glial scars are a double-edged sword: they help to prevent further injury to the spinal cord but also present a barrier to axonal regeneration^1,2. We wondered whether systemic administration of κ opiate drugs can be used as a means for controlling the degree of ependymal cell proliferation, which could potentially shift the balance between beneficial and detrimental consequences of scar formation.

To address this question, we used a common injury model where the dorsal aspect of the spinal cord is transected (i.e., dorsal hemisection) (Fig. 4a). To best capture the early phase of injury-induced proliferation, we began EdU labelling on the day after injury (Fig. 4b). We first examined the effect of the κ antagonist Nor-BNI. In sham operated controls, we saw a significant increase in the number of EdU⁺ ependymal cells in antagonist- versus vehicle-treated mice (Fig. 4c), which mirrors the phenomenon observed in healthy animals and indicates a functional tonic κ signaling pathway. Consistent with earlier reports^3–6, dorsal hemisection caused a substantial increase in the number of EdU⁺ ependymal cells in mice injected with vehicle relative to sham controls. Importantly, Nor-BNI did not further enhance this proliferation (Fig. 4c), suggesting that the constitutive κ signaling had now been interrupted by injury. Both CSF-cNs and PDYN⁺ cells remained intact after injury and are non-proliferative (Extended Data Fig. 7a and b). Expression of PDYN and OPRK1 in the ependymal region was also sustained, at least qualitatively, over the same period (Extended Data Fig. 7c and 8b). Although we were not able to measure peptide release directly, these observations suggest that an injury-evoked decrease in κ signaling is due to a reduction in κ opioid release.

Fig. 4. Systemic administration of κ agonist reduces ependymal proliferation induced by spinal cord injury.

a, DAPI-stained coronal sections of spinal cords from a sham-operated mouse and a mouse with dorsal hemisection (DH) to demonstrate the degree of injury (dotted line). b, Injury and drug treatment regimen. c, Number of EdU⁺ cells in sham-operated and DH animals with or without systemic treatment with κ antagonist (10 mg/kg Nor-BNI) or agonist (20 μg/kg Nalfurafine). Mean ± SD; Two-way ANOVA on log-transformed data: Nor-BNI × DH (F_1,27=13.84, p=0.0009); Šidák posthoc pairwise comparisons: saline versus Nor-BNI in sham-operated mice (p=0.0006), saline versus Nor-BNI in injured mice (p=0.9640, not significant), saline-treated sham control versus injured mice (p<0.0001); n = 5, 6, 10, and 10 mice. Two-way ANOVA on log-transformed data: Nalfurafine × DH (F_1,30=14.81, p=0.0006); Šidák posthoc pairwise comparisons: vehicle versus Nalfurafine in sham-operated mice (p=0.6439, not significant), vehicle versus Nalfurafine in injured mice (p=0.0002), vehicle-treated sham control versus injured mice (p<0.0001); n = 7, 7, 9, and 11 mice. All scale bars are 20 μm. d, Number of EdU⁺ cells in vehicle versus Nalfurafine-treated mice after CSF-cN ablation (by DTX injection to Tg(PKD2L1-Cre);Rosa26^iDTR mice) and subsequent dorsal hemisection. Mean ± SD; Two-sided Welch’s t-test on log-transformed data: p=0.1382, not significant; n = 5 and 7 mice.

In a complementary set of experiments, we investigated the effect of the κ agonist Nalfurafine. We confirmed that sham operated controls behaved similarly to uninjured wildtype mice in that Nalfurafine had no impact on ependymal proliferation (Fig. 4c). Remarkably, injury-induced proliferation could be lowered by ~ 55% with Nalfurafine treatment (Fig. 4c), indicating that κ agonists indeed have a pronounced suppressive effect on ependymal cell proliferation and that this sensitivity to opioids remains after injury. Although previous studies have noted transient Oprk1 expression in astrocytes after spinal cord contusion⁴⁰ (Extended Data Fig. 8a), we did not observe OPRK1 immunosignal in cell types other than CSF-cNs (Extended Data Fig. 8b). Importantly, the effect of Nalfurafine was abolished after the CSF-cN ablation, demonstrating that it indeed acts through CSF-cNs (Fig. 4d).

Genetic suppression of ependymal proliferation has been shown to cause defective scar formation and secondary enlargement of the lesion site in injured spinal cords⁵. We asked whether pharmacological suppression of ependymal proliferation by long-term application of a κ agonist would produce a similar effect. For this, we implanted mice subcutaneously with either a placebo or Nalfurafine pellet (0.027 mg drug per pellet, 60-day release rate) and examined the spinal cord 5–8 weeks after dorsal hemisection (Fig. 5a). Indeed, the injured segment of spinal cords from Nalfurafine-treated animals showed less buildup of GFAP⁺ astrocytic scars (Extended Data Fig. 7d) as well as a decrease in the number of cells expressing SOX9, a transcription factor that is found in the ependymal lineage. In contrast, there was no significant change in the populations of OLIG2⁺ oligodendrocyte or CD68⁺ activated microglia/macrophages^4,6 (Extended Data Fig. 7d), two cell types that scatter around the border of scar tissues. Strikingly, Nalfurafine-treated spinal cords demonstrated very severe thinning compared to those from controls (Fig. 5a), consistent with the notion that opioid regulation of ependymal proliferation is critical for supporting tissue integrity after injury. To assess functional deficits, we subjected placebo or Nalfurafine-treated mice to the rotarod test 5–8 weeks after sham or dorsal hemisection surgeries. Nalfurafine did not affect the performance of sham controls but significantly reduced the amount of time injured mice were able to remain on the rotarod (Fig. 5bi and Extended Data Fig. 9). A decline in rotarod performance was most apparent in mice whose lesions exceeded half the thickness of the spinal cord and extended into the motor area (Fig. 5bii), consistent with the idea that activation of κ opioid signaling impairs both anatomical and functional recovery.

Fig. 5. κ agonist exacerbates tissue damage and locomotor deficit after spinal cord injury.

a, Effect of long-term κ agonist administration (Nalfurafine, 0.027mg/pellet with 60-day release rate, 5–8 weeks of treatment) on thickness of hemisected spinal cords (minimal value normalized to thickness in uninjured regions in the same section). Mean ± SD; Two-sided Welch’s t-test: p=0.0098; n = 17 and 13 mice. b, Effect of Nalfurafine treatment on rotarod performance. (Left) Percentage of mice that stayed on the rotarod over time. Solid or dotted lines (DH or sham, respectively) are actual data from the averages of 3 highest scores for each mouse. Cutoff time was 10 min. Shadow curves are survival curves estimated by Cox regression analysis: Time on rotarod ~ Surgery + Surgery : Drug, β₁ = 1.764 with p = 0.0158, β₂ = 1.032 with p = 0.0496; n = 17 (Sham Placebo), 19 (Sham Nalfurafine), 13 (DH Placebo) and 19 (DH Nalfurafine) mice. (Bottom Right) Average time on rotarod plotted compared to spinal cord thickness measured at the lesion site as in a. Each dot is from an injured mouse (black: Placebo, n = 9; green: Nalfurafine, n = 6). A sharp decline in rotarod performance is observed when the lesion extended past midline. Spinal cord image above indicates the target injury level relative to midline. (Top Right) Spinal cord image indicating the target injury level relative to the midline. c, Schematic summary of the proposed signaling pathway. (Top) In healthy conditions, PDYN⁺ cells (red) constitutively release κ opioids to activate CSF-cNs (green), which in turn suppress ependymal proliferation. (Bottom Left) Injury reduces κ signaling and subsequent suppression by CSF-cNs, contributing to increased ependymal proliferation (purple). This de-repression may be accompanied by the action of other yet-to-be identified injury-associated factors (orange arrow). (Bottom Right) Systemic administration of a κ agonist (blue arrow) partially restores CSF-cN-mediated suppression, leading to reduced scarring.

Conclusions

Our findings provide insight into how the proliferation of spinal cord ependymal cells is regulated in vivo (Fig. 5c). We propose that in normal healthy conditions, CSF-cNs receive constant κ opioid stimulation from neighboring PDYN⁺ cells. Activation of the κ opioid receptor in CSF-cNs excites these neurons, which release GABA and/or other inhibitory signals that suppress ependymal cell proliferation. Upon spinal cord injury, constitutive κ signaling halts, possibly due to physical damage to the radial processes of PDYN⁺ cells. As a result, suppression by CSF-cNs is relieved, contributing to increased ependymal proliferation critical for facilitating scar formation.

The implication of radial glia-like PDYN⁺ cells in this process may explain why ependymal proliferation is triggered most prominently when the injury occurs close to the midline, rather than laterally or by a general crush^4,41. The neurotransmitter GABA has been shown to elicit GABA_A receptor-mediated electrical responses in ependymal cells⁴² and restrict their proliferation⁸, thus representing a likely suppressive signal released by GABAergic CSF-cNs. Interestingly, CSF-cNs express functional GABA_A receptors themselves^11,34 and hence they may possess a feedback mechanism for autoregulating GABA release. CSF-cNs also express several other neurotransmitter receptors^11,15,35, although their upstream synaptic partners are largely uncharacterized. It remains to be determined whether ependymal proliferation is also modulated by synaptic input to CSF-cNs under specific physiological circumstances. Moreover, CSF-cNs have been reported to be part of some motor circuits^13,17,19 and it will therefore be interesting to see whether these neurons coordinate locomotion readjustment after spinal cord injury. We also note that injury-induced ependymal proliferation was more pronounced than that induced by κ antagonist in healthy mice. This suggests that the application of opiate drugs may not fully reproduce the dynamics of the endogenous opioid regulatory pathway, or that κ signaling constitutes just one component of the injury response, which may involve other, as-yet to be identified molecular signals.

Given that scar tissues have dual functions (both protective and inhibitory) in the recovery from injury^1,2, our work raises the important question of whether κ opioid receptor drugs can be employed to control ependymal cell proliferation dynamically, thus modulating scar formation and wound healing in certain types of spinal cord injuries. We have provided proof-of-principle evidence that tissue integrity and functional recovery after injury can be modulated by Nalfurafine, a κ agonist that has received approval in multiple countries as an antipruritic with minimal adverse effects⁴³. Whether this strategy has potential therapeutic value will require further assessment of the temporal contributions of the ependymal lineage and κ signaling to the scar formation process. Additionally, over-proliferation of ependymal cells also occurs in clinical cases of spinal ependymoma, which is associated with pain and weak muscle control. It will be interesting to investigate whether the endogenous κ signaling pathway is involved in tumor formation in such conditions. In any case, this pathway likely represents just one of several controlling ependymal proliferation, and further studies will determine its contribution and therapeutic potential relative to other mechanisms.

Methods

Animals

All animal experiments were conducted in accordance with protocol AN192533 approved by the Institutional Animal Care and Use Committee, University of California, San Francisco. Mice were raised under regular diurnal (12:12) light-dark cycles at a temperature of 68–79°F and a humidity of 30–70% with ad libitum access to food and water. We used mice of both sexes between the age of 1–6 months. Tg(PKD2L1-Cre) mice (MGI: 5691401)¹⁰ and Pdyn^fl/fl mice (MGI: 7407169)⁴⁴ were kindly provided by Dr. Charles Zuker (Columbia University) and Dr. Richard Palmiter (University of Washington, Seattle), respectively. Wildtype C57BL6/J mice as well as Rosa26^LSL-tdTomato (MGI: 3813512)⁴⁵, Pdyn^Cre (MGI: 5562873)⁴⁶, Polr2a^{GCaMP5G-tdTomato} (MGI: 5560331)⁴⁷, MORF3 (MGI: 6441965)²³, Rosa26^iDTR (MGI: 3772576)⁴⁸ and Sox2^CreERT2 (MGI: 5295990)⁴⁹ mice were acquired from Jackson Laboratory. For cell ablation, 1 mg/ml stock solution of DTX (List Biological Laboratories, #150) was prepared with water, aliquoted and stored at −20°C until use. When needed, the DTX solution was diluted to 5 ng/μl with sterile 0.9% saline and 50 μg/kg was administered i.p. daily for 5 consecutive days. For CreERT2 induction, tamoxifen (Sigma-Aldrich, T5648) was dissolved at a concentration of 20 mg/ml in corn oil at 37°C overnight and then stored at 4°C for no longer than 4 days. Mice were i.p. injected with 2 mg tamoxifen/day for 3 consecutive days. To avoid any direct effect of tamoxifen on cell proliferation, mice were used for experiments after >3 weeks. In all experiments, littermates were used as genotype controls whenever possible; when not, age-matched animals were used. Mice of each genotype were randomly selected and assigned to different groups in all experiments. Where possible, experimenter was blinded from the genotype or drug treatment information when performing surgeries and quantifications, including event counting, cell counting and intensity measurements.

AAV packaging and intracerebroventricular injection

AAV2 or PHP.eB viral preparations made from pAAV-CAG-tdTomato (#59462) or pAAV-CAG-GFP (#37825) plasmids were purchased from Addgene. AAV2 was generated from pAAV-FLEX-PLAP plasmid (Addgene, #80422) by triple transfection into HEK293FT cells (Thermo Fisher Scientific, R70007), followed by purification of AAV particles with the iodixanol gradient ultracentrifugation method. Viral titer was between 1E12 – 5E13 GC/ml for in vivo delivery. We did not find any information about authentication and mycoplasma contamination from the vendor regarding the HEK293FT cell line.

Under isoflurane anesthesia, a mouse was positioned on a heating pad on a stereotaxic instrument (David Kopf Instruments, Model 1900). Hairs were removed from the top of the head and the surgical site was disinfected with alternating scrubs of 7.5% Povidone Iodine (Betadine surgical scrub) and 70% Ethanol. Lidocaine (Vedco, NDC 50989-417-12, 1.6 mg/ml in sterile 0.9% saline, 100 μl/mouse) was injected subcutaneously prior to making a ~1-cm long mid-sagittal skin incision over the skull. The bregma and lambda points were identified, and the skull was levelled by using a stereotaxic alignment indicator (Kopf Instruments, Model 1905) according to the manufacturer’s instructions. A small hole was opened on the skull at the coordinates (X = +1.7, Y = −0.9) or (X = +1.0, Y = −0.4) above the right lateral ventricle by using a stereotaxic drill (Kopf Instruments, Model 1911) coupled to a 0.027” ball drill bit (CircuitMedic, 115–6050). The cranial window was irrigated with sterile PBS immediately after opening to prevent dehydration. A Wiretrol II glass pipette (Drummond Scientific Company, 5–0002010), pulled to give a sharp tip (Sutter, P-97 micropipette puller) before the experiment, was mounted onto the stereotaxic frame and was loaded with 5 μl of AAV solution. The pipette tip was slowly advanced to 2.2 mm beneath the brain surface, and the AAV solution was dispensed gradually by manually controlling the Wiretrol plunger. Tissues were allowed to take up the AAV solution for 10 min before the pipette was retrieved slowly. Finally, the skin was closed by suturing, and the mouse was allowed to recover on a warm pad. Buprenorphine (Par Pharmaceutical, NDC 42023-179-05, 0.1 mg/kg) and Meloxicam (Pivetal, NDC 46066-937-13, 10 mg/kg) were administered i.p. during and after the surgery for analgesia.

Immunohistochemistry and Click-iT EdU imaging

Mice were euthanized by CO₂ asphyxiation according to American Veterinary Medical Association’s guidelines. Transcardiac perfusion with phosphate-buffered saline (PBS, Quality Biological, 119-069-491) followed by 4% paraformaldehyde (PFA, Electron Microscopy Sciences, 15714) was performed immediately after. Spinal cords were harvested into 4% PFA and fixed overnight. After 3 washes with PBS, spinal cords were transferred to 30% sucrose (Sigma-Aldrich, S7903) and allowed to settle at 4°C. The tissues were then cryopreserved in Tissue-Tek O.C.T. Compound (Sakura Finetek USA) and sectioned at a thickness of 25–50 μm on a Leica CM3050 S cryostat. Sagittal sections were collected directly on glass slides and stored at −80°C until use. Coronal sections were collected in PBS and were stored in freezing medium (30% w/v sucrose [Sigma-Aldrich, S7903], 1% w/v polyvinylpyrrolidone [Sigma-Aldrich, PVP40], 30% v/v ethylene glycol [Sigma-Aldrich, 102466] in 50 mM phosphate buffer) at −20°C until use. On the day of experiment, sections were retrieved into PBS and were washed with PBST (i.e., PBS containing 0.5% Triton X-100 [Sigma-Aldrich, T8787]) for multiple times. Subsequently, sections were incubated with blocking buffer (PBST containing 10% normal goat serum [Thermo Fisher Scientific, 16-210-064] or donkey serum [Sigma-Aldrich, D9663]) at room temperature for 1 hr. Overnight primary antibody incubation was done at 4°C with one or more of the following antibodies in blocking buffer: rabbit anti-DsRed (Takara Bio, 632496, 1:500), goat anti-mCherry (Sicgen, AB0081–200, 1:500), rabbit anti-PKD2L1 (Sigma-Aldrich, AB9084, 1:500–700), rabbit anti-OPRK1 (Abcam, ab183825, 1:500), chicken anti-GFP (Abcam, ab13970, 1:500), rabbit anti-DYNA (Peninsula Laboratories, IHC8676, 1:500), rabbit anti-SOX2 (Abcam, ab97959, 1:500), chicken anti-NESTIN (Aves Labs, NES, 1:500), goat anti-SOX9 (R&D Systems, AF3075, 1:500), chicken anti-GFAP (Abcam, ab4674, 1:500), mouse anti-MBP (BioLegend, #808401, 1:500), mouse anti-CD68 (Cell Marque, Kp-1, 1:500) and rabbit anti-OLIG2 (Sigma-Aldrich, AB9610, 1:500). On the following day, sections were washed with PBST for 3 times and incubated with the appropriate secondary antibodies at 1:500 in blocking buffer. Secondary antibodies include donkey anti-goat IgG-Alexa Fluor 488 (Thermo Fisher Scientific, A-11055), donkey anti-goat IgG-Alexa Fluor 568 (Thermo Fisher Scientific, A-11057), donkey anti-goat IgG-Alexa Fluor 647 (Abcam, ab150131), donkey anti-rabbit IgG-Alexa Fluor 488 (Thermo Fisher Scientific, A-21206), donkey anti-rabbit IgG-Alexa Fluor 568 (Thermo Fisher Scientific, A-10042), donkey anti-mouse IgG-Alexa Fluor 647 (Thermo Fisher Scientific, A-31573), goat anti-rabbit IgG-Alexa Fluor 488 (Thermo Fisher Scientific, A-11034), goat anti-rabbit IgG-Alexa Fluor 568 (Thermo Fisher Scientific, A-11036), goat anti-chicken IgY-Alexa Fluor 488 (Thermo Fisher Scientific, A-11039), goat anti-chicken IgY-Alexa Fluor 568 (Thermo Fisher Scientific, A-11041) and goat anti-mouse IgG-Alexa Fluor 488 (Thermo Fisher Scientific, A-11029). The nuclei stain, 4′,6-diamidino-2-phenylindole (DAPI, 0.5 μg/ml, Thermo Fisher Scientific, D1306), was included in one of the final PBST washes. Lastly, sections were laid on a glass slide and mounted with Fluoromount-G (SouthernBiotech, 0100–01). Z-stack images were taken with a Nikon CSU-W1 spinning disk confocal microscope or a Nikon Ti Inverted Microscope (UCSF Center for Advanced Light Microscopy) with the Micro-Manager (v2.0)⁵⁰ or NIS-Elements software (Nikon). Image stitching was done by NIS-Elements software. Maximal intensity projections were generated by the Fiji software (NIH). Cell counting was done by eyes or on projected images by using the Cell Counter plugin in Fiji. For quantifying GFAP and CD68 immunosignal intensity, a region of interest was drawn on a projected image, and integrated pixel intensity was measured by the Fiji software.

To detect EdU⁺ cells, 25-μm cryosections of the spinal cord were made as above and processed through the aforementioned immunostaining procedure when necessary. After the final PBST wash, the freely-floating sections were incubated with shaking in the Click-iT reaction cocktail (Thermo Fisher Scientific, C10340) in the dark at room temperature for 1 hr. Following one PBST wash, the sections were incubated with DAPI, then further washed, mounted, and imaged as described above. For each mouse, EdU⁺ cells were counted and averaged across >3 sections.

PLAP staining and tissue clearing

For PLAP staining of tissue sections, we prepared 50-μm cryosections of the spinal cord and brainstem from perfused mice as described above. To inactivate endogenous alkaline phosphatase activity, the samples were heated up in PBS in a 72°C water bath for 1 hr. The sections were washed twice in Buffer 1 (100 mM Tris base, 150 mM NaCl, pH 7.5) and twice in Buffer 2 (100 mM Tris base, 100 mM NaCl, 50 mM MgCl₂, pH 9.5) before being transferred to the staining buffer (2% NBT-BCIP stock solution [Sigma-Aldrich, 11681451001] in Buffer 2). Samples were allowed to be stained for 30 min in the dark and were then dehydrated through a series of increasing methanol concentration (20%, 40%, 60%, 80%, 100%, room temperature, 30 min each). After mounting, images were taken with a Nikon Ti Inverted Microscope (UCSF Center for Advanced Light Microscopy). Stitched images were generated by NIS-Elements (Nikon Instruments).

For wholemount PLAP staining, spinal cords were harvested from perfused mice and fixed overnight in 4% PFA as described above. They were then bisected along the midline to expose the ependymal region. Heat inactivation and PLAP staining were carried out as above except that the incubation in staining buffer was 1 hr long. After serial dehydration, the tissues were stored in 100% methanol at room temperature until being transferred to 66% dichloromethane (DCM)/33% methanol on the following day for clearing at room temperature for 3 hrs on a rotator. Subsequently, methanol was washed away by two 15-min incubations in 100% DCM, and the samples were allowed to stand and clear in a glass bottle filled completely with dibenzyl ether (DBE). The tissues were submerged in DBE in a home-made chamber when being imaged on an Olympus FV3000 microscope using the FV31S-SW software (v2.5.1.228).

Bulk mRNA-sequencing and expression data analysis

Mice received intracerebroventricular injection of AAV2-CAG-tdT as described above. After 3–4 weeks, a mouse was perfused with ice-cold NMDG-aCSF (92 mM NMDG, 2.5 mM KCl, 10 mM MgSO₄, 0.5 mM CaCl₂, 1.2 mM NaH₂PO₄, 30 mM NaHCO₃, 20 mM HEPES, 5 mM Na-ascorbate, 2 mM Thiourea, 3 mM Na-pyruvate, 25 mM Glucose, pH 7.4) and its spinal cord was freshly removed into the same solution by applying hydraulic pressure through a syringe to the open spine. The spinal cord was bisected along the midline with a stab knife (Surgical Specialties, 72–2201) on a sylgard (Dow, DC4019862) plate and embedded in 2% low-melting agarose. 200-μm thick sagittal slices containing the ependymal region were collected from each half of the spinal cord by using a Leica VT1000 S vibratome. Slices were cut into smaller pieces and transferred into Enzyme Mix 1 (955 μl of Buffer Z, 25 μl of Enzyme P [Miltenyi Biotec, 130-094-802]) in a 1.5-ml tube. After 15-min incubation in a 37°C water bath, Enzyme Mix 2 (15 μl of Buffer Y, 7.5 μl of Enzyme A [Miltenyi Biotec, 130-094-802]) was added for another 5-min, 37°C incubation. The digestion solution was then replaced with 1 ml HABG (97.75% Hibernate A Low Fluorescence [BrainBits, HALF], 2% B27 supplement [Thermo Fisher Scientific, 17504044], 0.25% GlutaMAX [Thermo Fisher Scientific, 35050061]). The digested tissues were triturated multiple times by using fire-polished glass pipettes with decreasing tip diameters to mechanically dissociate the cells. The cell suspension was then laid onto 4 ml of 20% Percoll (in HABG) and was centrifuged at 300× rcf for 15 min in a cold room. The resulting supernatant was removed carefully, and the cell pellet was resuspended in 0.8 ml HABG containing Ribolock RNase inhibitor (Thermo Fisher Scientific, EO0382, 0.4 U/μl). A far-red LIVE/DEAD dye (Thermo Fisher Scientific, L34973, 1:1000) was added to the suspension for excluding dead cells during FACS. The cell suspension was sorted based on forward and side scatters, as well as tdTomato and LIVE/DEAD fluorescence in a BD Biosciences FACSAria II Cell Sorter (UCSF Laboratory for Cell Analysis). Fifty cells were collected in a tube containing ice-cold lysis buffer and processed for cDNA synthesis according to the Smart-Seq v4 manual (Takara Bio, 634890) with a PCR cycle number of 11. Subsequently, cDNA library was prepared by using Nextera XT DNA Library Preparation Kit (Illumina, FC-131–1024). We followed the manufacturer’s manual for tagmentation, amplification and cleanup steps, but estimated the concentrations of our libraries by comparing their SYBR green (Thermo Fisher Scientific, S7563) fluorescence to a standard curve made from λ DNA (Thermo Fisher Scientific, 25-250-010). Library quality was confirmed by using the Bioanalyzer High Sensitivity DNA Kit (Agilent Technologies, 5067–4626). Four libraries were pooled for pair-end sequencing with 100 cycles on a S1 flow cell lane in an Illumina NovaSeq 6000 System (UCSF Center for Advanced Technology). A technical replicate was included in a separate lane. About 40–60 million reads were obtained per sample and the sequencing quality was examined by using FASTQC. Adapters and low-quality reads were trimmed or removed by TrimGalore (v0.6.0). Reads were then aligned to the annotated mouse reference genome (mm10) containing an additional exogenous sequence for AAV-CAG-tdT by using RSEM. Gene counts were imported into RStudio (v4.2.0) for analysis by using the DESeq2 module (v1.36.0, Bioconductor v3.16). Replicates were collapsed, and the DESeq function was applied to normalize read counts. Results were displayed as heatmaps.

To analyze single-cell sequencing dataset published by Matson et al.⁴⁰, we imported the expression matrices and cluster identities (GSE172167) into RStudio and computed the average expression value of Oprk1 in each cluster by the AverageExpression function in Seurat (v5.0.1). Results were displayed as heatmaps. Of note, this study employed a thoracic contusion model and collected tissues from the lumbar level, which may not be directly comparable to our immunohistochemical analyses done at the lesion level in a lumbar dorsal hemisection model.

Calcium imaging of spinal cord slices

A Tg(PKD2L1-Cre);Polr2a^{GCaMP5G-tdTomato} mouse was euthanized by CO₂ asphyxiation, and its spinal cord was acutely dissected into ice-cold NMDG-aCSF (92 mM NMDG, 2.5 mM KCl, 10 mM MgSO₄, 0.5 mM CaCl₂, 1.2 mM NaH₂PO₄, 30 mM NaHCO₃, 20 mM HEPES, 5 mM Na-ascorbate, 2 mM Thiourea, 3 mM Na-pyruvate, 25 mM Glucose, pH 7.4). After removing the meninges, the spinal cord was embedded in 2% low-melting agarose, and the thoracic-lumbar region was sectioned coronally at a thickness of 300 μm in ice-cold NMDG-aCSF by using a Leica VT1000 S vibratome. Each spinal cord slice was adhered to a small piece of filter membrane (Whatman Nuclepore Track-Etched Membrane, Sigma-Aldrich, WHA110657) for convenient handling. Slices were allowed to recover at 34°C in NMDG-aCSF for 30 min and then transferred to aCSF (124 mM NaCl, 2.5 mM KCl, 2 mM MgSO₄, 2 mM CaCl₂, 1.2 mM NaH₂PO₄, 24 mM NaHCO₃, 5 mM HEPES, 12.5 mM Glucose, pH 7.4) for storage at room temperature until being used within 3 hrs. All solutions were constantly bubbled with 95% O₂/5% CO₂ to provide oxygen and maintain pH. When needed, a piece of filter membrane carrying a spinal cord slice was immobilized by vacuum grease in an imaging chamber and was placed under a Leica SP8 confocal microscope. The ependymal region was located under brightfield, and CSF-cNs were identified by their tdTomato signal. Image acquisition was done by using the LAS X software (v3.5.5.19976, Leica Microsystems) at 4 Hz with a confocal pinhole equivalent to a depth of ~220 μm of tissues. The tissues were bath perfused with bubbled room-temperature aCSF at a rate of ~1 ml min⁻¹. Pharmacological reagents were included in the bath and/or applied locally at a rate of ~140 ul min⁻¹ through a 100-μm needle connected to a pressure-controlled dispensing system (Smartsquirt, Automate Scientific). To avoid including responses due to mechanical stimulation of the cells, we applied aCSF through the needle and regarded this as the baseline before switching to pharmacological reagents. Pharmacological reagents used in this study are listed in Table 1.

As a note, U73122 and U73343 stock solutions were prepared by first dissolving in chloroform to 10 mM. 250 μg aliquots were made in glass vials, and chloroform was evaporated away under a stream of Argon gas. The drugs were stored at −20°C for <1 week before re-dissolving in DMSO for use.

At the end of each imaging session, high K⁺ solution (2.5 mM NaCl, 124 mM KCl, 2 mM MgSO₄, 2 mM CaCl₂, 1.2 mM NaH₂PO₄, 24 mM NaHCO₃, 5 mM HEPES, 12.5 mM Glucose, pH 7.4) was squirted onto the slice to reveal all responsive neurons within the imaging field.

Fluorescence images were imported into the Fiji software (NIH), and regions of interest (ROIs) were drawn around labelled CSF-cNs. ΔF/F traces were extracted for each ROI over three time periods: (1) 30 sec immediately before DynA application, (2) 30 sec immediately following DynA application, and (3) 10 sec at the peak of the high K⁺ response. Since the response of CSF-cNs to DynA is typically composed of multiple transient spikes, we computed the area under ΔF/F traces for comparisons instead of using a single peak amplitude value. To account for the basal spontaneous activity of CSF-cNs and the difference in fluorescence due to the variable depths of the cells within the tissue slices, we subtracted away the baseline value and further normalized to the high K⁺ response (except for Extended Data Fig. 4a). In other words, the DynA response is reported as (Area under ΔF/F for period 2 – Area under ΔF/F for period 1) / Area under ΔF/F for period 3. Slices that showed substantial movements during imaging or abnormal high K⁺ response were not analyzed. Cells with strong basal fluorescence (indicative of poor health) or large or extremely frequent spontaneous fluorescence fluctuations were also excluded from analyses due to difficulty in defining the baseline. ΔF/F images were generated by using the “F div F0” function in the T-functions plugin in Fiji.

Patch-clamp recording

Spinal cord slices were prepared from Tg(PKD2L1-Cre);Polr2a^{GCaMP5G-tdTomato} mice by using the vibratome as above except at a thickness of 150 μm for better visualization under epifluorescence microscope (BX50WI, Olympus). During recording, slices were bath perfused with room-temperature aCSF bubbled with 95% O₂/5% CO₂ at a rate of ~70 μl min⁻¹. Patch electrodes (3–5 MΩ) were pulled from borosilicate capillaries (BF-150-110-10, Sutter Instrument) and filled with internal solution (10 mM NaCl, 135 mM KCl, 2 mM MgCl₂, 1 mM CaCl₂, 10 mM HEPES, 5 mM EGTA, 10 mM Phosphocreatine-Tris, 4 mM Mg-ATP, pH 7.4 with KOH). Under brightfield, the ependymal region was located at 4× magnification and CSF-cNs were identified at 40× by their tdTomato signal. We targeted only cells that were exposed or beneath a single cell layer. After the whole-cell patch-clamp configuration was established, voltage steps were given to confirm excitability of the cell. Recording was made through the pClamp software (v10.7, Molecular Devices) by using the Multiclamp 700A amplifier (Molecular Devices) connected to a digitizer (Digidata 1550B, Molecular Devices) and was sampled at 10 kHz. For voltage-clamp recording, a low pass filter of 3 kHz (software single-pole RC filter) was applied. Cells were held at −80 mV without correcting for liquid-junction potential. As soon as a stable baseline was achieved, we locally perfused aCSF with a 100-μm Smartsquirt needle placed close to the ependymal region, followed by DYNA (1 μM), at a rate of ~50 μl min⁻¹. In experiments involving κ antagonist, 0.1 μM Nor-BNI was included in the bath. For current-clamp recording, low pass filter was set at 10 kHz and liquid-junction potential was corrected. When necessary, cells were brought to about −60 mV by current injection before DYNA application. Drugs were applied as above.

For analysis, we selected recordings that were made at an access resistance <25 MΩ and showed no drifting nor excessive noise. To produce an amplitude histogram, traces were further low-pass filtered digitally at 1 kHz (8-pole Bessel), and single channel opening events were identified by the Single-Channel Search function in the Clampfit software (v11.2.2.17, Molecular Devices) with open levels set at −4 pA and −12 pA. The Gaussian function was fitted by using the Prism software (GraphPad). To study the effect of DYNA on open probability, the original 3 kHz filtering was maintained and single channel opening events were searched over a 30-sec stretch of recording with a single open level at −12 pA in Clampfit and then confirmed manually by eye. Postsynaptic events were distinguished based on kinetics over the same period. Open probability was calculated by dividing the total dwell time by the recording duration. To calculate the firing rate, spikes were searched over a 30-sec stretch of recording by setting an amplitude threshold of 25 mV in Clampfit. Events were confirmed manually, and firing rate was computed by dividing the number of spikes by the recording duration.

For recording Ca_v current, we used a different external solution (140 mM NMDG, 10 mM BaCl₂, 10 mM HEPES, 12.5 mM Glucose, 0.5 μM TTX, pH 7.4) and internal solution (140 mM CsMeSO₄, 2 mM MgCl₂, 0.1 mM CaCl₂, 10 mM HEPES, 5 mM EGTA, 10 mM Phosphocreatine-Tris, 4 mM Mg-ATP, pH 7.4 with CsOH). NMDG and TTX were used to limit sodium current while Cesium was included to inhibit potassium current. Barium instead of Ca²⁺ was used as the charge carrier for a larger current. Cells were initially voltage-clamped at −100 mV. 200-ms voltage steps to different voltages were then applied at 20 mV interval up to +60 mV. The resulting current was confirmed to be through Ca_v channels by adding 500 μM Cadmium in the bath to block Ca_v channels. DYNA was applied as above. In this experiment, we were able to washout DYNA, likely because the application needed for voltage steps was much shorter than in other imaging and recording experiments. We report the peak inward current, normalized to the cell capacitance.

Drug administration

Uninjured mice received daily injections of the κ opioid receptor agonist Nalfurafine (i.p. injected at a concentration of 4 μg/ml in sterile 0.9% saline with 0.1% DMSO to give a final dose of 20 μg/kg body weight, once/day, AdooQ Bioscience, A12579) or antagonist Nor-BNI (i.p. injected at a concentration of 2 mg/ml in sterile 0.9% saline to give a final dose of 10 mg/kg body weight, once/day, Abcam, ab120078) or the respective vehicle control over a course of 8 days. Starting from day 5 of the treatment, 50 mg/kg EdU (at a concentration of 10 mg/ml in sterile 0.9% saline, Cayman Chemical Company, 20518) was additionally i.p. injected once everyday for 7 days. Mice were sacrificed on the day after the EdU injections ended. All injections were done at roughly the same time of the day (~ 5 PM) for consistency.

In spinal cord injury experiments, mice received the same 8-day agonist/antagonist treatment as above starting from the afternoon of the day of injury. EdU injections began on the following day and lasted for 7 days. Mice were sacrificed on the next day. For long-term drug administration, Nalfurafine powder was mixed with a matrix comprising cholesterol, cellulose, lactose, phosphates and stearates. The mixture was made into a pellet (0.027 mg Nalfurafine per pellet, 60-day release rate, Innovative Research of America). A placebo pellet containing only the matrix was used as a control. Five days before dorsal hemisection of the spinal cord, a pellet was implanted subcutaneously at the neck between the ears under isoflurane anesthesia, and the wound was closed with sutures and wound clips.

Spinal cord hemisection

Under isoflurane anesthesia, hairs were removed from the dorsal surface of the mouse, and the surgical site was disinfected. An incision was made along the dorsal spine, and the skin was held back by a retractor to expose the vertebrae. The muscles covering one of the lower thoracic or lumbar vertebral disks was cut in a manner perpendicular to the disk space. Laminectomy was then performed at L1-L2 by using small vannas scissors to cut through the spinous processes and the lateral sides of the vertebral lamina. After lifting off the dorsal aspect of the vertebrae, a pair of microscissors was inserted into the exposed spinal cord segment at a depth of ~0.8 mm to remove the dorsal column and dorsal horns. The muscle layer was closed by suturing and the skin was stapled. The mouse was supplemented with Lactated Ringer’s with 5% Dextrose (Baxter Healthcare Corporation, NDC 0338-0125-03) to regain homeostasis and was allowed to recover on a heating pad. Buprenorphine (Par Pharmaceutical, NDC 42023-179-05, 0.1 mg/kg) and Meloxicam (Pivetal, NDC 46066-937-13, 10 mg/kg) were administered i.p. before and after the surgery for analgesia. Sham-operated mice underwent the same procedure except that no spinal segment was removed. All surgeries were performed in the morning so that drug treatment could begin in the same afternoon.

During the first week after surgery and whenever necessary thereafter, we manually expressed the bladder of the animals by using the Crede maneuver twice daily. Both EAE and BCS scores were recorded daily. None of the operated mice used in this study had a >15% drop in body weight.

Rotarod test

Mice were implanted with placebo or Nalfurafine pellets 5 days before dorsal hemisection as described above. Rotarod test was performed 5–8 weeks after the surgery. Mice were trained on the rotarod instrument (Ugo Basile, 47650) for at least two days prior to the test and were then tested with a uniform rotating speed of 25 rpm and a cutoff time of 10 min. The time to fall was registered, and the 3 highest scoring trials (out of 3–5 replicates) were analyzed for each mouse. Averages across these trials were used to calculate the percentage of mice remaining on the rotarod over time. Mice that reached the cutoff time in >2 trials were included in the analysis but statistically censored and not shown in the plot of survival curves. We applied Cox regression analysis with the following model: Time on rotarod ~ Surgery + Surgery : Drug. This model is preferred over the more complicated model 2: Time on rotarod ~ Surgery + Drug + Surgery : Drug with an AIC difference of −1.723, i.e., probability that the model is correct is 70.29% (model 1) versus 29.71% (model 2). The further simplified model 3: Time on rotarod ~ Surgery is rejected with an AIC difference of 2.369, i.e., probability that the model is correct is 76.57% (model 1) versus 23.43% (model 3). Goodness of fit is examined with 3 tests: Partial likelihood ratio test, Wald test and Score test. All gave p<0.0001 for rejecting the null hypothesis that the parameter coefficients for all predictor variables are zero, or in other words, the predictor variables have no effect on the hazard rate.

Data analysis and statistics

Unless otherwise specified, collective data are reported as mean ± standard deviation (SD). Statistical analyses were done by using the Prism software (GraphPad), and data are reported following the ARRIVE guidelines. For statistical comparisons, sample size was selected based on power calculations performed with reference to the means and variances observed in previous or present experiments carried out in our laboratory and in the field. Parametric tests were performed only if the data passed at least two of the following four normality tests: D’Agostino-Pearson, Anderson-Darling, Shapiro-Wilk and Kolmogorov-Smirnov tests. Parametric tests include Welch’s t-test, mixed-effect analyses, two-way ANOVA, and repeated measures two-way ANOVA, with posthoc pairwise comparisons done using Šidák correction. Non-parametric tests include Mann-Whitney test, Kruskal-Wallis test, Wilcoxon matched-pairs signed rank test, and repeated measures Friedman test, with posthoc pairwise comparisons done using Dunn’s correction. Table 2 summarizes the statistical tests and post hoc comparisons used for data shown in Figs. 1 to 5.

Extended Data

Extended Data Fig. 1. Characterization of CSF-cNs and PDYN⁺ cells in the spinal cord.

a, Labelling of spinal cord cells by the Tg(PKD2L1-Cre);Rosa26^LSL-tdTomato mouse line. The tdTomato reporter (magenta) was expressed in many other spinal cord cells in addition to the CSF-cNs (magnified on right), which were identified by their strong immunostaining signal for PKD2L1 (green), particularly in their bulbous projections within the central canal (CC). b, Labelling of spinal cord cells by the Tg(PKD2L1-Cre) mouse line when tdTomato reporter was delivered by i.c.v. AAV injection after adulthood. The PHP.eB serotype labels cells beyond the ependymal region (e.g., arrow in magnified view on right). c, Projection pattern of CSF-cNs at different levels of the spinal cord. Coronal spinal cord and brain sections from a mouse that had received i.c.v. injection of an AAV carrying the PLAP transgene. Sections were stained with NBT/BCIP to reveal the ventral projections of CSF-cNs. d, Longitudinal projection of CSF-cNs. Sagittal section of the spinal cord showing the nerve bundle formed by CSF-cNs’ projections that travel rostrocaudally within the ventral white matter. Cell bodies of CSF-cNs are found in the ependyma. e, Morphology of a single PLAP-labelled CSF-cN. Spinal cord was cleared by the iDisco method⁵² after NBT/BCIP staining. f, Morphology of Pdyn⁺ cells. Arrowheads trace the long processes of sparsely labelled Pdyn⁺ cells situated at the dorsal or ventral pole of the ependyma. Dotted lines mark the boundary between gray and white matters. g, Schematic depicting the projection pattern of CSF-cNs (green) and PDYN⁺ cells (magenta). h, Spinal cord sections from Pdyn^Cre;Rosa26^LSL-tdTomato mice immunostained with antibodies against known markers (cyan) for various subsets of ependymal cells. SOX2 is expressed in PDYN⁺ cells (magenta) in the ependymal region but not in PDYN⁺ cells in the dorsal horn. Ependymal PDYN⁺ cells are also positive for NESTIN but not GFAP. All experiments have been repeated for at least 3 times with similar results.

Extended Data Fig. 2. Expression of transcripts of interest in preparations of acutely dissociated and FACS-enriched CSF-cNs.

CSF-cNs were labelled by tdTomato via i.c.v. AAV injection and were fluorescently sorted for bulk mRNA sequencing. Columns represent 4 separate preparations with increasing level of enrichment in CSF-cNs, as reflected by the expression level of tdTomato transcript. Color scale is based on median-of-ratios calculation by DESeq2. a, Marker genes for oligodendrocytes (Plp1 and Mbp), migroglia (P2ry12) and astrocytes (Gfap), showing the degree of glial contamination. b, Genes typically involved in GABA metabolism. c, Genes related to neurotransmission. d, Genes for receptors proposed to be sensitive to κ opioid ligands. e, TRP and Ca_v channel genes.

Extended Data Fig. 3. Ca²⁺ responses of GCaMP5G-expressing CSF-cNs to OPRK1 agonists.

ai, Example ΔF/F traces showing the responses of CSF-cNs to local application of the κ agonist, Nalfurafine, in the absence (black) or presence (red) of the antagonist, DIPPA, in the bath. Local application of a high K⁺ solution was used to reveal all responsive neurons. Each trace is from a single cell. aii, ΔF/F images for the spinal cord slices in ai. Images are temporal averages over 10 sec of baseline or for the duration of the stimuli. CC: central canal. Scale bars are 20 μm. aiii, Collective data comparing CSF-cNs’ responses to Nalfurafine in the absence (black) and presence (red) of DIPPA. Each dot shows the integral DYNA response of a single cell normalized to the high-K⁺ response. Two-sided Mann-Whitney test: p=0.0445; n = 29 and 28 cells. b, Same as a except that BRL-52537 was used as the agonist and Nor-BNI as the antagonist. Two-sided Mann-Whitney test: p=0.0150; n = 20 and 12 cells. In all bar graphs, data are mean ± SD.

Extended Data Fig. 4. Pharmacological experiments to delineate the downstream pathway of OPRK1 signaling.

a, CSF-cNs showed significant responses only to κ agonist (DYNA, 0.5 μM, 1 min) but not to agonists of the bradykinin receptors (bradykinin, 0.5 μM, 1 min), the delta opioid receptor (SNC162, 0.5 μM, 1 min) and the mu opioid receptor (Endomorphin-1, 0.5 μM, 1 min). Calcium imaging of acutely harvested spinal cord slices in the presence of TTX. Each color represents one cell. Agonists were applied without gaps in the order displayed on the graph. Because comparisons were done within the same cell, responses were raw area under ΔF/F traces and were not baseline-subtracted nor normalized to high-K⁺ responses as in other figures. Repeated measures Friedman non-parametric test and Dunn’s posthoc pairwise comparisons with baseline: Bradykinin (p=0.5466), SNC162 (p>0.9999), Endomorphin-1 (p>0.9999), and DYNA (p<0.0001); n = 13. b, Normalized integral DYNA response (mean ± SD) as in Fig. 2c, but in the presence of various inhibitors or in different ionic conditions. Molecular targets of the drugs are indicated in brackets. Routes of drug application are detailed in Methods. Kruskal-Wallis non-parametric test and Dunn’s posthoc pairwise comparisons with control, which is same as –Nor-BNI in Fig. 2biii and Fig. 2c: YM254890 (p<0.0001), U73122 (p=0.0133), U73343 (p=0.3874, not significant), Thapsigargin (p>0.9999, not significant), Ivabradine (p>0.9999, not significant), Chelerythrine Cl (p=0.0149), Nifedipine (p=0.4256, not significant), ω-Agatoxin (p<0.0001), ω-Conotoxin (p=0.0005), SNX482 (p>0.9999, not significant) and NNC 55–0396 (p=0.0004). Numbers of cells analyzed are in brackets above bars. c, Proposed signaling pathway downstream of OPRK1 in CSF-cNs.

Extended Data Fig. 5. Voltage-clamp recording on CSF-cNs during DYNA application.

ai, Representative voltage-clamp recordings of CSF-cNs. Membrane potential was held at −80 mV in the absence or presence of the κ antagonist, Nor-BNI, in the bath. No macroscopic current was observed during local DYNA application (line above trace). aii, Expanded view of the boxed regions of traces in ai, showing single channel openings at baseline or during DYNA application. b, Example of a single channel opening event and a spontaneous postsynaptic event to show the clear distinction between the two waveforms. c, Amplitude histogram of spontaneous single channel opening events detected at baseline. Two peaks at amplitude ~5 pA and ~11 pA were detected. The ~11 pA events resembled those described in earlier reports^11,15, which were shown to originate from PKD2L1 channels¹⁵. d, Open probability of the ~11 pA channel before and after DYNA application with or without Nor-BNI in bath. Each pair of light-colored dots is from a single cell. Group averages are in dark colors. Repeated measures two-way ANOVA: DYNA × Nor-BNI (F_1,10=0.01981, p=0.8909, not significant); n = 6 and 5. We did not analyze the ~5 pA events because they were not clearly discernable from background noise. e, Rate of postsynaptic events before and during DYNA application in normal aCSF bath. Each pair of light-colored dots is from a single cell. Group averages are in dark colors. Wilcoxon matched-pairs signed rank non-parametric test: p=0.8750, not significant; n = 7 and 7.

Extended Data Fig. 6. Ca_v current recording from CSF-cNs during DYNA application.

a, Representative voltage-clamp recordings of CSF-cNs. Membrane potential was held at −100 mV, and 20 mV voltage steps from −100 mV to 60 mV were applied before, during, and after DYNA application. Voltage-gated sodium and potassium channels were inhibited by NMDG and TTX in the external solution and by cesium in the internal solution. Ba²⁺ (10 mM) was supplemented to the external buffer to increase the conductance of Ca_v channels. b, Current-voltage relationship recorded under the 3 conditions in a. The general Ca_v channel blocker, Cadmium (Cd²⁺), was added to one cell to verify that the recorded current was through Ca_v channels. Current responses were normalized to membrane capacitance, represented as mean ± SD. Mixed-effect analyses with Šidák correction for pairwise comparisons: statistical significance reported in table; n = 10 for baseline and DYNA conditions and n = 8 for washout (data from 2 cells were excluded due to unstable recording).

Extended Data Fig. 7. CSF-cNs, Pdyn⁺ cells and scar formation in injured spinal cords.

Following dorsal hemisection, no EdU signal (white) was detected in (a) CSF-cNs labelled in the Tg(PKD2L1-Cre) mouse line and (b) Pdyn⁺ cells labelled in the Pdyn^Cre line on Day 9. EdU injection scheme as in Fig. 4b. c, Immunohistochemical detection of PDYN in the ependymal region of sham-operated and injured mice 9 days after surgery. d, Effect of long-term κ agonist treatment (Nalfurafine, 0.027mg/pellet with 60-day release rate, 5–6 weeks of treatment) on scar components. (Left) Representative images of the lesion sites in spinal cords harvested from placebo- or Nalfurafine-treated mice 5–8 weeks after dorsal hemisection. Sections were immunostained with antibody against GFAP (astrocytes), SOX9 (expressed in ependymal lineage), OLIG2 (oligodendrocytes) and CD68 (activated microglia and macrophages), which label components of scar tissues. (Right) Intensity of immunosignal or cell number was quantified and normalized to the volume of the region of interest. Two-sided Welch’s t-test: GFAP (p=0.0005; n = 9 and 6), SOX9 (p=0.0358; n = 12 and 10), and OLIG2 (p=0.5756, not significant; n = 12 and 10). Mann-Whitney test: CD68 (p=0.9546, not significant; n = 9 and 6).

Extended Data Fig. 8. Expression of the κ opioid receptor in various cell types.

a, Transcript expression of Oprk1 in various cell types that are known to participate in the injury response. Expression data and cluster identities were from Matson et al.⁴⁰. The authors used contusion instead of dorsal hemisection as their injury model. The injury was inflicted at the thoracic level and spinal cord tissues were collected at the lumbar level. Color scale shows the average expression value. b, Protein expression of OPRK1 in various cell types identified by immunohistochemical markers. OPRK1 expression was detected only in CSF-cNs in the ependymal region (arrows) but not in other cell types in both sham-operated and injured animals 9 days after dorsal hemisection. Similar results were obtained from 3 experiments.

Extended Data Fig. 9. Effect of Nalfurafine (0.027 mg/pellet with 60-day release rate, 5–8 weeks of treatment) on rotarod performance.

Amount of time individual mice from each treatment group stayed on rotarod. Datapoints are the 3 highest scores; corresponding averages are represented by bars. The cutoff time is 10 min. Pie charts (black) indicate the fraction of animals that reached the cutoff time in at least 2 trials (i.e., censored). Number of animals tested are given in brackets.

Data availability

All data generated or analyzed during this study are included in the manuscript and its extended data. The sequencing results have been deposited at NCBI under accession number GSE255883. Source data are provided with this paper. Reference genome is built based on the annotated mouse reference genome (mm10) available as Genome assembly GRCm38 on NCBI at: https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001635.20/.