Schwann Cells Are Key Regulators of Corneal Epithelial Renewal

Abstract

Purpose: Corneal sensory nerves protect the cornea from injury. They are also thought to stimulate limbal stem cells (LSCs) to produce transparent epithelial cells constantly, enabling vision. In other organs, Schwann cells (SCs) associated with tissue-innervating axon terminals mediate tissue regeneration. This study defines the critical role of the corneal axon-ensheathing SCs in homeostatic and regenerative corneal epithelial cell renewal.

Methods: SC localization in the cornea was determined by in situ hybridization and immunohistochemistry with SC markers. In vivo SC visualization and/or ablation were performed in mice with inducible corneal SC-specific expression of tdTomato and/or Diphtheria toxin, respectively. The relative locations of SCs and LSCs were observed with immunohistochemical analysis of harvested genetically SC-prelabeled mouse corneas with LSC-specific antibodies. The correlation between cornea-innervating axons and the appearance of SCs was ascertained using corneal denervation in rats. To determine the limbal niche cellular composition and gene expression changes associated with innervation-dependent epithelial renewal, single-cell RNA sequencing (scRNA-seq) of dissociated healthy, de-epithelized, and denervated cornea limbi was performed.

Results: We observed limbal enrichment of corneal axon-associated myelinating and non-myelinating SCs. Induced local genetic ablation of SCs, although leaving corneal sensory innervation intact, markedly inhibited corneal epithelial renewal. scRNA-seq analysis (1) highlighted the transcriptional heterogenicity of cells populating the limbal niche, and (2) identified transcriptional changes associated with corneal innervation and during wound healing that model potential regulatory paracrine interactions between SCs and LSCs.

Conclusions: Limbal SCs are required for innervation-dependent corneal epithelial renewal.

Corneal epithelial renewal^1,2 depends on the activity of limbal stem cells (LSCs)^3–6 that are located in the basal epithelium of the limbus.^5,7 Previous reports have suggested that epithelial stem cells, evenly distributed at the corneal basal layer, differentiate vertically to replenish the upper layers of the epithelium.⁸ However, recent studies using single-cell RNA sequencing (scRNA-seq) and lineage tracing approaches in a mouse model showed that LSCs are restricted to the limbal niche, the most richly innervated part of the cornea.^9–13 In the niche, the LSCs interact with mesenchymal stromal cells (MSCs) and T cells that play a critical role in the maintenance and activity of the LSCs.^9,13,14 Normally, the LSCs reside in a growth-arrested or slow-cycling state,^15–18 exhibit morphological characteristics of stem cells, and express genes associated with asymmetric cell division.^19–23 During homeostatic epithelial renewal or after injury, the LSC progeny differentiate to transient amplifying cells (TACs) that migrate centripetally to the central cornea, where they differentiate further into epithelial cells, thereby replenishing the corneal epithelium.^{9–13,24–31} 

The cornea is innervated by sensory nerves that are responsible for activation of tearing and blink reflexes and repair.³² They also play a critical role in maintaining the LSCs and/or the stem cell niche.³³ The prevailing hypothesis is that corneal epithelial wound healing and maintenance depend on direct trophic stimulation of the LSCs by the sensory axon terminals.^33–49 The axons in the basal epithelial layer of the limbus run adjacent to the LSCs, and their free nerve endings contact epithelial cells.⁵⁰ When the protective corneal sensory innervation is lost after infection, trauma, intracranial tumors, or ocular surgery or when innervation fails to develop in congenital cases, permanent blindness results from repetitive microtraumas that induce epithelial breakdown, ulceration, scarring, and opacification of the cornea.^34,36,51,52 This condition, known as neurotrophic keratopathy (NK), affects nearly 5/10,000 people worldwide.^53,54 

The only U.S. Food and Drug Administration–approved drug for treating NK thus far is topically applied recombinant human nerve growth factor (rhNGF).⁵⁵ However, whilst rhNGF showed no toxicity and improved NK symptoms, it was not effective in all patients, especially those with more severe symptoms.⁵⁵ The cellular sources of NGF and other trophic factors potentially involved in the regulation of LSC activity have not been elucidated. 

As is the case of all peripheral nerve fibers, the corneal axons interact with Schwann cells (SCs).^56,57 In addition to their role supporting axonal maintenance and conductivity, SCs in the nerve terminals play a key role in tissue regeneration. After skin wounding or digit tip amputation, mature SCs dedifferentiate into SC precursors that produce cytokines and growth factors which mediate recovery.^58,59 While the neurotrophic role of SCs in the cornea has been addressed,^56,57 their potential role in regulating the function of the limbal niche during epithelial renewal has not been investigated. 

There is an abundance of myelinating and non-myelinating SCs in the corneal limbus that are closely associated with LSCs. We hypothesize that these corneal SCs support innervation-dependent epithelial cell renewal. To examine the role of SCs in the corneal epithelial recovery directly, we induced a local genetic ablation of SCs in the cornea. Our findings demonstrated that, even though corneal axons remained intact, SC ablation prevented healing of the de-epithelialized cornea, similar to an anesthetic cornea. We used scRNA-seq analysis and computational processing performed on isolated dissociated limbi to compare healthy, de-epithelialized, and denervated corneas, and we identified multiple cell clusters including molecularly distinct subclusters of LSCs and MSCs. This analysis suggests that the corneal SCs are a potential source of multiple trophic factors, including NGF, and predicts multiple paracrine interactions between SCs and LSCs and between SCs and MSCs. These findings indicate a new role for limbal SCs in mediating innervation-dependent corneal epithelial renewal. 

Methods

Experimental Design

The primary objective of this study was to define the role of SCs in mediating corneal innervation-dependent epithelial renewal. To address this objective, we (1) identified corneal myelinating and non-myelinating SCs by using in situ hybridization, in vivo genetic labeling, and in vitro immunostaining approaches; (2) defined the link between corneal innervation and limbal SCs by inducing corneal denervation in mice using an ophthalmic nerve electrocautery approach; (3) defined the effect of SC ablation on corneal epithelial maintenance and wound healing in a loss-of-function experiment whereby corneal SCs were locally ablated; and (4) using scRNA-seq, defined the cellular composition of corneal limbus and modeled paracrine interactions between SCs and LSCs and between MSCs and LSCs following denervation and injury of rat corneas. 

Experimental Animals and In Vivo Procedures

Animals

For scRNA-seq analysis, Sprague Dawley rats were used. For the experiments that used genetically modified animals, the following transgenic mouse strains were purchased from The Jackson Laboratory (Bar Harbor, ME, USA): CBA;B6-Tg(Sox10-icre/ERT2)388Wdr/J (Sox10-iCreER^T2)⁶²; tdTB6;129S6-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J (R26-LSL-tdTomato)¹⁰⁷; and B6;129- Gt(ROSA)26Sor^tm1(DTA)Mrc/J (R26-LSL-DTA).¹⁰⁸ The mice were maintained on a C57Bl/6J background. All animals were managed based on the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. 

Corneal Denervation

A single electrocautery of the ophthalmomaxillary branch of the ophthalmic nerve was performed on mice and rats as described previously.^65,66 Briefly, the rodents were mounted on a stereotactic frame, and a 1-mm burr hole was made at the coordinates anterior–posterior and medial–lateral, as previously defined for rats (∼230–260 g)⁶⁵ and mice (∼30 g). An insulated 22-gauge monopolar electrode was introduced through the burr hole. An electrical current was activated to induce the electrocautery (rats, 10 W; mice, 5 W) for 60 seconds. Then the electrode was removed, and the skin was sutured. Ablation of the corneal innervation was confirmed by absence of the blink reflex to touch and cold saline under light anesthesia having confirmed an intact reflex on the contralateral eye. 

Corneal De-Epithelialization

Removal of corneal epithelium was performed with an Amoils rotating brush (Innovative Excimer Solutions, Toronto, ON, Canada). The diameter of the rotating brush head was trimmed to 5 mm for mice and 8 mm for rats. The central epithelium was removed without affecting the underlying stroma and the limbal epithelium. 

Tarsorrhaphy

After each experimental or imaging session, the eyelids were sutured together as described previously.⁶⁵ 

Corneal Healing Assay

The experimental animals were anesthetized, and the cornea was assessed for a blink reflex with cold saline. Fluorescein visualization (DIOFLUOR Strips; Innova Medical Ophthalmics, Toronto, ON, Canada) and digital imaging (D5100 camera; Nikon, Tokyo, Japan) were performed immediately after de-epithelialization and every 24 hours up to 96 hours after injury to monitor wound size and healing of the corneal epithelium as described previously.⁶⁵ 

Schwann Cell Visualization and/or Ablation

Tamoxifen (T5648; Sigma-Aldrich, St. Louis, MO, USA) was dissolved in 60°C preheated (liquid form) oil-based eye ointment (SYSTANE eye ointment; Alcon, Geneva, Switzerland) to a final concentration of 60 mg/mL. To completely dissolve the tamoxifen, the solution was shaken at 60°C for 30 minutes. The tamoxifen-containing ointment was used for a maximum of 3 days while maintained in the dark at 4°C. To activate expression of the Cre-ERT gene, the tamoxifen-containing ointment was applied to the surface of the cornea for 10 minutes under 3% isoflurane anesthesia. This process was repeated for 3 consecutive days. In vivo microscopic observation of tdTomato-positive SCs in corneas was performed with a ZEISS Axio Zoom.V16 microscope (ZEISS Group, Oberkochen, Germany) while a mouse's head was stabilized in a stereotaxic frame under 3% isoflurane. 

Microscopy

Immunohistochemistry

Mouse corneas were harvested for immediate fixation for 1 hour in precooled (4°C) fixation solution that contained 2% paraformaldehyde (PFA) and 15% saturated picric acid, followed by two 30-minute washes in PBS and blocking in PBS containing 0.5% Triton X-100, 5% normal donkey serum, and 1% bovine serum albumin for 1 hour at room temperature. Corneas were incubated with primary antibodies dissolved in the blocking solution for 48 hours at 4°C and then washed for 1 hour in 1% Triton X-100, followed by two consecutive washes in PBS for 30 minutes at room temperature. The samples were incubated at 4°C overnight with secondary antibodies dissolved in the blocking solution that contained 4′,6-diamidino-2-phenylindole (DAPI) for nuclei staining after three constitutive washes, as above. Whole corneas were mounted on glass slides and analyzed using confocal (ZEISS LSM 880 Airyscan) or epifluorescence (ZEISS Axio Imager 2) microscopy. 

Primary Antibodies

The primary antibodies included Anti-beta III Tubulin [2G10] (1:1000, ab78078; Abcam, Cambridge, UK), Recombinant Anti-Cytokeratin 15 (1:200, ab52816; Abcam), Anti-beta III Tubulin (1:300, ab18207; Abcam), Recombinant Anti-SOX10 [EPR4007] (1:100, ab155279; Abcam), Monoclonal Anti-Myelin Basic Protein (MBP, 1:100, MAB386 Sigma-Aldrich), Monoclonal Anti-Sodium Channel, Pan antibody produced in mouse (1:500, S8809; Sigma-Aldrich), and polyclonal Anti-Caspr (1:1000, kindly provided by Elior Peles, PhD, Weizmann Institute of Science, Rehovot, Israel), Recombinant Monoclonal Anti-Mayelin Associated Glycoprotein [EPR24276] (MAG, 1:500, ab277524, Abcam). 

Fluorescent In Situ Hybridization

Rat corneas were fixed for 1 hour in precooled (4°C) 2% PFA and 15% saturated picric acid. The fluorescent in situ hybridization (FISH) procedure was performed as per the Molecular Instruments (Los Angeles, CA, USA) protocol (available at https://files.molecularinstruments.com/MI-Protocol-RNAFISH-Mouse-Rev8.pdf), using a Sox10 transcript-recognizing probe (Accession NM_019193.3). The samples were mounted on glass slides and observed using confocal microscopy (ZEISS LSM 880 Airyscan). 

Single Cell Expression Analysis

Limbus Harvesting

Limbal areas of healthy, de-epithelialized, and denervated rat corneas (n = 8 per experimental condition, two separate experiments) were harvested with 50-µm margins into the conjunctival and central cornea. 

Tissue Dissociation

Limbi of every experimental condition were pooled together, cut into small pieces, and exposed to EDTA-free trypsin (1×) solution (15090046; Thermo Fisher Scientific, Waltham, MA, USA) for 15 minutes at 37°C, followed by an intensive trituration and supernatant collection. The remaining undissociated tissue was centrifuged in 1000 rpm for 1 minute at 4°C and exposed to four more steps of digestion: (1) one more 15-minute exposure to the EDTA-free trypsin solution; (2) two steps of 10-minute exposures at 37°C to 0.25% trypsin-EDTA (25200056; Thermo Fisher Scientific); and (3) one exposure to 10× trypsin (15090046; Thermo Fisher Scientific) for 10 minutes at 37°C. The consolidated supernatant of every experimental condition was filtered using a 40-µm cell filter (Fisherbrand Sterile Cell Strainers, 22363547; Thermo Fisher Scientific) to guarantee a homogeneous single-cell solution, precipitated at 1000 rpm for 5 minutes, and resuspended in Dulbecco's Modified Eagle Medium (DMEM; Thermo Fisher Scientific). 

Single-Cell RNA Sequencing

A droplet-based, high-throughput, scRNA-seq10× Genomics technology was used to profile dissected and dissociated healthy, de-epithelialized, and denervated rat limbus cells. The single-cell transcriptomes were analyzed as before.⁷¹ To construct a cell–cell communication network, we used a database in which ligands and receptor genes were defined based on the annotations from the set of gene ontology terms, as described previously.⁵⁸ The cell–cell communication network was defined by the expression of ligands and receptor genes on the cell types of interest. For example, if cell A expresses ligand X and cell B expresses the receptor for ligand X, then an arrow shows the direction from cell A to cell B. Using this principle, a network was constructed and visualized in Cytoscape 3.8.1 (Cytoscape Consortium, San Diego, CA, USA). 

Statistical Analysis

JMP 16 (JMP Software, Cary, NC, USA) was used for statistical analysis. The Shapiro–Wilk W-test and quantiles plot were used for the normality test of continues variables. The pairwise comparison Student's t-test was used. A significance level of 5% was used (P < 0.05). Descriptive statistics were presented as means with standard deviation (SD). Six animals were required for each group to reach a statistical power of 0.8 (or 80%), assuming a SD of the mean of 30% for all groups. 

Results

Mature Myelinating and Non-Myelinating Schwann Cells Occupy the Limbal Niche

FISH with a probe recognizing the terminally differentiated SC-specific promotor gene Sox10^60,61 was performed to observe mature SCs in rat corneas. It showed an enrichment of Sox10-positive nuclei in the corneal limbus (Fig. 1A; Supplementary Fig. S1). To further characterize the limbal SCs in vivo, we used a genetic mouse model of tamoxifen-inducible CreERT2 knocked into the Sox10 promotor and the tdTomato reporter gene Sox10-CreER^T2/+; R26-LSL-tdT (Sox10-tdT).^62–64 Topical administration of tamoxifen revealed abundant tdT-positive SCs that specifically occupied the limbal area (Fig. 1B; Supplementary Fig. S2). Note that no detectable expression of tdTomato was observed in cells in other subcorneal eye compartments (not shown). Whole-mount immunofluorescence staining of fixed corneas of Sox10-tdT mice and rats with neuronal and glial markers demonstrated abundant βIII Tubulin–positive neurons. The neurons densely innervated the limbal area and were colocalized with both non-myelinating and myelinating SCs; that were positive for MAG and MBP, as well as for markers of the node of Ranvier, respectively (Figs. 1C–1F). 

Figure 1.

 

Myelinating and non-myelinating SCs are abundant in the limbus. (A) In situ hybridization demonstrates enrichment of Sox10-positive cells in the corneal limbus. Scale bar: 100 µm. (B) Live image of tamoxifen-induced tdTomato (tdT)-positive fluorescent SCs in Sox10-tdT mouse whole eye (arrowheads, upper panel) and part of limbus (magnified in lower panel). Scale bars: 1 µm and 200 µm for upper and lower panels, respectively. (C) Immunofluorescent 10 µm extended-depth focus immunofluorescent image of Sox10-tdT mouse fixed cornea demonstrates a 1:1 association between by (βIII)-tubulin–positive (green) neurons and tdT-positive SCs (red), which abundantly populate the limbal area. Scale bar: 250 µm. (D, E) Immunofluorescent image of whole-mount rat cornea, demonstrating non-myelinated, MBP-negative (indicated by arrowhead) and myelinated, MBP-positive (red, indicated by arrow) βIII-tubulin–positive (green) neurons in the limbal area. Image in D is positive for the paranodal protein Caspr (green), together with red sodium channel NaCh⁺ (indicated by asterisks) at the nodes of Ranvier in the left and right panels, respectively (E). Scale bars: 50 µm (D); 2 µm (E). (F) Immunofluorescent image of sagittal section of central rat cornea at the limbal area demonstrating a cross-section of single βIII-tubulin–positive axon enwrapped by a single membrane layer of MAG-positive non-myelinating Schwann cells, presented in a sequence of constitutive magnifications of the defined area. Scale bars (top to bottom): 100 µm, 25 µm, and 2.5 µm.

Immunostaining of corneas harvested from tamoxifen-administrated Sox10-tdT mice for the LSC-specific marker Krt15^9,18 was performed to identify potential interactions between corneal SCs and LSCs. This analysis demonstrated that limbal SCs are located adjacent to Krt15-positive LSCs (Fig. 2). Collectively, these findings demonstrate an abundance of myelinating and non-myelinating SCs that are closely localized with LSCs in the limbus, suggesting a potential for interaction between the two cell populations. 

Figure 2.

 

Close localization of SCs and LSCs at limbus. (A) A 10-µm extended-depth focus immunofluorescent image of Sox10-tdT mouse cornea demonstrates that Krt15-positive LSCs (green) and tdT-positive (red) SCs abundantly occupy the limbal area. Scale bars: 100 µm. (B) Immunofluorescent one-focal-plate image of Sox10-tdT mouse fixed cornea limbal area shows close localization (arrowheads) of Krt15-positive LSCs (green) and tdT-positive (red) SCs. Scale bar: 100 µm.

Local Genetic Ablation of SCs Inhibits Corneal Epithelial Recovery

Based on recent reports demonstrating a critical role of nerve-associated SCs in the repair of injured skin and digit tip,^58,59 in addition to the observed adjacent localization of SCs with LSCs in the limbus (Fig. 2), we hypothesized that an analogous SC-dependent process of epithelial renewal occurs in the cornea. Previously, using our NK animal model, we observed that chronic corneal sensory denervation prevented corneal epithelial wound healing.⁶⁵ We examined how the corneal SCs respond to acute corneal denervation to determine whether corneal denervation is linked to a reduction or loss of limbal SCs. In Sox10-tdT mice pretreated with tamoxifen, we denervated the cornea by performing a single stereotactic electrocautery of the ophthalmomaxillary branch of the trigeminal nerve.⁶⁵ In vivo serial observations of the denervated cornea demonstrated a complete loss of SCs by 48 hours after denervation (Fig. 3A). Thereafter, to directly assess the role of SCs in corneal epithelial renewal, we crossed Sox10-tdT with R26-LSL-DTA to generate a Sox10-tdT;DTA mouse in which topical administration of tamoxifen activates Diphtheria toxin A (DTA)-induced apoptosis of corneal tdT-positive SCs. A tamoxifen-containing hydrogel was applied to the left eye of every experimental mouse to ablate limbal SCs locally, and the phenotype of the Sox10-tdT;DTA, tamoxifen-treated corneas was compared to that of the vehicle-treated right corneas, as well as with those of Sox10-tdT tamoxifen-treated littermates in which the activation of Cre induced only tdTomato labeling (Figs. 3B, 3D). As expected, tamoxifen induced expression of tdTomato in SCs several hours after Cre activation (Figs. 3B, 3D). Following expression of the Diphtheria toxin, the number of tdTomato-positive SCs in Sox10-tdT;DTA corneas decreased in the limbus by 58% to 69% (n = 3, P < 0.005) at 10 days after tamoxifen administration (Figs. 3B, 3D, 3G; Supplementary Fig. S2). This observation was supported by immunostaining of harvested Sox10-DTA fixed corneas that demonstrated an absence of Sox10-positive nuclei at 10 days after tamoxifen administration (Fig. 3C). Immunostaining of harvested Sox10-tdT;DTA tamoxifen-treated corneas demonstrated that the SC ablation had no effect on corneal innervation, as both the limbal and the epithelial axons remained intact, comparable to those of normal, SC-non-ablated corneas (Fig. 3D). 

Figure 3.

 

Both denervation-induced and genetic ablation of corneal SCs caused the NK phenotype. (A) Corneal denervation completely ablated SCs in 48 hours, as shown by in vivo fluorescent imaging of tdT-positive SCs in Sox10-tdT mouse denervated cornea. The magnified area shows 3 and 48 hours after corneal denervation. Scale bar: 200 µm. (B) Fluorescent image of the limbal area of tamoxifen-induced tdT-positive SCs in Sox10-tdT (left panel) and Sox10-tdT;DTA mouse fixed corneas 10 days after tamoxifen administration (right panel). Scale bar: 500 µm. (C) Immunofluorescent image of the limbal area of whole-mount fixed cornea harvested from Sox10-DTA mouse shows an abundance of Sox10-positive SC nuclei (green, indicated by arrowheads) attached to βIII-tubulin–positive axons (red) at 4 hours (left panel), but SCs disappeared 10 days after tamoxifen administration (right panel). Scale bar: 200 µm. (D) Immunofluorescent 10-mImextended-depth focus image of Sox10-tdT (left panel) and Sox10-tdT;DTA mouse fixed cornea shows unmodified limbal and epithelial innervation by βIII-tubulin–positive (green) neurons in the SC-ablated, comparable to the SC-non-ablated corneas despite a significant reduction in volume of tdT-positive SCs (red) 10 days after tamoxifen administration, as shown in the magnified lower panel images. Scale bars: 100 µm and 250 µm in the main and magnified images, respectively. (E) In vivo photos of mouse corneas demonstrate opacification of the SC-ablated left (L) eye (arrows) compared to a clear cornea in the untreated right (R) eye. (F) In vivo photographs of de-epithelialized mouse corneas demonstrate the progress of corneal epithelial healing for 4 days after de-epithelialization in SC-ablated (SC abl, Sox10-DTA mice) and SC-present (SC pr) tamoxifen treated corneas. “0h” indicates the cornea condition immediately after epithelial removal, 10 days after tamoxifen administration. Fluorescein (yellow) stains the corneal area lacking epithelium. (G, H) Quantitative representation of SCs ablation in B to D (n = 3 per condition) and the corneal healing in F (n = 6 per condition), respectively. Error bars represent standard error. *P < 0.05; **P < 0.01; ***P < 0.005 (t-test in G and ANOVA in H).

To investigate the effect of SC ablation on corneal epithelial renewal, we assessed both corneal maintenance and recovery after epithelial removal with an Amoils brush. As expected, the cornea of the SC non-ablated right eye remained clear during the experiment (Fig. 3E). Despite intact corneal innervation (Fig. 3D) and intact blink reflexes in response to saline eye drops (not shown), the SC-ablated corneas of the left eye opacified within 2 weeks after commencing tamoxifen administration of the Sox10-tdT;DTA mice (Fig. 3E). As before, we used a fluorescein absorption assay as an indicator of corneal healing following de-epithelialization.⁶⁵ The SC-ablated de-epithelialized corneas of the Sox10-tdT;DTA mice failed to heal the epithelial wound (Figs. 3F, 3H), similar to the anesthetized corneas of NK animal models.^65,66 No inhibition in wound healing was observed in the SC non-ablated control corneas of Sox10-tdT tamoxifen-treated mice (Figs. 3F, 3H). Collectively, these results demonstrate that (i) limbal SCs require the cornea-innervating neurons for their survival, (ii) corneal SCs are necessary for corneal wound healing, and (iii) corneal axons, otherwise intact but lacking adequate numbers of SCs, are insufficient for corneal wound healing. 

Identification of Multiple Cell Types Composing the Limbal Niche

Previous studies have used scRNA-seq to identify cells types^9,57,67,68 and their interactions at the limbal niche.^69,70 However, these reports focused on LSCs and their subpopulations and did not clearly differentiate between the stem and mesenchymal stromal cells at the niche or they did not identify SCs among the limbal cell populations in rodents.^9,67–69 To more comprehensively characterize the cell populations in the limbal niche, we performed droplet-based, high-throughput 10× Genomics analysis at 10,554 to 7000 median unique molecular identifier counts per cell and 24,000 to 7000 cells per experiment, with a depth of 4700 to 29,000 mean reads per cell. Limbal areas (eight rat corneas per experiment in two separate experiments) were harvested with ∼100-µm margins into the conjunctival and central cornea. Given that different cell types are differentially sensitive to tissue dissociation conditions, the harvested limbi were subjected to multiple steps of trypsinization and trituration with a gradual increase in exposure time and intensity of trypsin solution, and EDTA concentration (see Methods). To perform the comparative scRNA-seq analysis, two separate biological samples (eight limbi each) per experimental condition were collected. During the expression analysis, particular attention was given to expression events known to be associated with stem cell self-renewal and differentiation, adhesion, and migration in other tissues. The single-cell transcriptomes were analyzed with a modified analysis pipeline that incorporated extensive data quality analysis and visualization and clustering methods that used an evidence-based parameter selection process.⁷¹ The predicted cell doublets and cells with high relative mitochondrial gene transcripts, likely corresponding to dying cells (10.5%, not shown), were removed. Genes with high variance were used to compute principal components as input to visualize the cells in two dimensions using Uniform Manifold Approximation and Projection (UMAP) for dimension reduction (Fig. 4A). The graph-based clustering method from the Seurat package for R (R Foundation for Statistical Computing, Vienna, Austria) was employed to identify cell types. 

Figure 4.

 

The scRNA-seq identification of cell types populating the limbal niche in rat. scRNA-seq analysis was performed on dissociated limbi harvested from healthy rat corneas. The clusters of LSCs and MSCs were extracted and reanalyzed. (A) UMAP of resultant cell clusters, with colors and numbers denoting distinct clusters: LSCs (11); TACs (1 and 18); MSCs (14 and 15); epithelial cells (4, 6, 7, 8, 9, 12, and 13); conjunctival cells (0, 2, 3, 5, 10, and 18); Langerhans cells, macrophages, lymphocytes, mast cells, and NK cells (16 and 17); endothelial cells of lymphatic vessels (20); and SCs (19). The clusters of LSCs, MSCs, TACs, and SCs are highlighted. (B) Violin plot showing relative expression of the representative cell-type–specific gene markers as per A: Sox10 and Mbp for SCs; Abcg2 and Sox17 for LSCs; Pdgfra and Pdgfrb for MSCs; and Birc5 and Ki67 for TACs. (C) UMAP of colors and numbers represent transcriptionally distinct clusters of LSCs and MSCs which, as in A, were combined and reanalyzed. Subclusters 0, 3, 5, 8 and 10 outlined in black represent LSCs, and subclusters 1, 2, 6, 7, and 9 outlined in blue represent MSCs. Subcluster 4 shares mRNA representatives for both cell populations. (D) Violin plot showing relative expression of the selected genes in the indicated subclusters shown in C.

Cell clusters of previously defined limbal cell populations were detected by the cell population-specific signature genes. Abcg2 and Sox17 co-localized with expression of C/EBPδ, Notch1, and Tspan7, and missin Connexin4 for LSCs (5.2%; cluster 11). Birc5 co-localized with the proliferation markers Ki67 and Krt12 for TACs (9.3%; clusters 1 and 18), and Pdgfrα co-localized with CD105, CD73, CD90, and Pdgfrβ for MSCs (5.4%; clusters 14 and 15). Krt12 co-localized with Hes1, Gjb2, and Areg for epithelial cells (38.8%, clusters 4, 6, 7, 8, 9, 12, and 13). Krt13 and Krt19 co-localized for conjunctival cells (38.4%; clusters 0, 2, 3, 5, 10, and 18); Cd207 for Langerhans cells; Cd45 (PTPRC) for macrophages; Cd4 for lymphocytes; Icam1 and Cxcl2 for mast cells; Ncr1(Nkp46) and Klrk1(Nk1.1) for NK cells (2.4%; clusters 16 and 17); and Pdpn, Lyve1, Prox1, and Vegfr-3(Flt4) for endothelial cells of lymphatic vessels (0.34%; cluster 20). Sox10, Mbp, Mpz, Gfap, and Mag defined the cluster of SCs (0.35%; cluster 19). The cell clusters are shown in Figures 4A and 4B and Supplementary Fig. S3.^{4,5,67,70,72–74} 

Analysis of the isolated LSCs revealed five subclusters (0, 3, 4, 8, and 10) (Fig. 4C) characterized by expression of Lgals7, Sox17, Cd63, Gpha2, and lfitm3 and which contained proliferating cells as defined by Ki76 expression. A sixth subcluster (5), had a high expression level of Gpha2 and Ifitm3 but lacked expression of Ki76 (Figs. 4C, 4D) and displayed a lower value of RNA reads per cell (mean of 1345.4 vs. 6385.3). This suggests that the two groups of subclusters (0, 3, 4, 8, and 10, and 5) are active and quiescent LSCs, respectively (Figs. 4C, 4D).⁹ The analysis of the isolated MSCs revealed five subclusters (1, 2, 6, 7, and 9) (Fig. 4C) that all expressed Pdgfrb, and three of the subclusters (1, 6, and 7) also expressed Pdgfra. Subclusters 2, 4, and 9 expressed the embryonic markers Jag1, and subclusters 1, 4, and 7 expressed the embryonic marker Osr2. Other embryonic markers that were expressed were Sox9 (cluster 4) and Foxc2 (cluster 7) (Fig. 4D),⁵⁸ suggesting the potential of MSCs to differentiate into other mesenchymal tissues.⁷⁵ Intriguingly, subcluster 4 of the LSCs also shared the markers of MSCs and embryonic stem cells, indicating an earlier stage of their differentiation status and suggesting that they are a potential early source for both cell populations. Our scRNA-seq analysis highlights the transcriptional heterogenicity of cell populations in the limbal niche and differentiates between LSC and MSC populations and their subclusters, in addition to identifying a cluster of SCs. 

Transcriptome-Based Modeling of Paracrine Interactions Predicts Factors Expressed by SCs That May Regulate Limbal Cell Activity

To define the corneal nerve–induced expression events in the limbal niche required for epithelial renewal, we compared the expression profiles of healthy rat cornea cells (as above) with those of de-epithelialized (wounded, 12 hours after de-epithelialization) central and acutely denervated corneas (5 days after single denervation). To achieve corneal de-epithelialization, the epithelium was removed with an Amoils brush. Acute corneal denervation was achieved by a single stereotactic electrocautery of the ophthalmomaxillary branch of the trigeminal nerve.⁶⁵ The rationale for this comparison was that the corneal innervation-induced processes, usually occurring during homeostatic epithelial renewal, will be accelerated during corneal healing but inhibited after corneal denervation. Five days after denervation was chosen for the analysis, as it was found to be the latest time point at which the damaged corneal epithelium was still capable of healing 24 hours later (Fig. 5A), although forming scar tissue (Supplementary Fig. S4). 

Figure 5.

 

Expression changes following corneal de-epithelization or denervation allow prognosing of paracrine interactions regulating LSCs activity. (A) In vivo photographs of injured rat corneas demonstrate the progress of corneal epithelial healing for 3 days after de-epithelialization in denervated (den) and normally innervated (norm) corneas. “0h” indicates cornea condition immediately after epithelial removal. Fluorescein (yellow) stains the corneal area lacking epithelium. In the denervated cornea, de-epithelization was induced 5 days after a single electrocautery of the ophthalmomaxillary branch of the trigeminal nerve.⁶⁵ (B–F) A scRNA-seq–based quantitative representation of the indicated gene expression in clusters of SCs (B), MSCs (D), and LSCs (E), comparing dissociated limbi harvested from healthy (norm), de-epithelized (injured [inj]), and denervated (den) corneas. Note that, due to the absence of SCs in denervated corneas, no gene expression for this condition is presented in B or C. Expression of the Sox10/Sox2 ratio in limbal SCs comparing healthy (norm) versus de-epithelialized (inj). (F) Expression level of Ki67 in the three conditions in LSCs (upper chart) and MSCs (lower chart). (G) Transcriptome-based network model of computationally predicted paracrine communication between SCs and LSCs and between MSCs and LSCs, based on scRNA-seq analysis of the three experimental conditions in B to F. Arrows indicate the direction of communication. Green and yellow nodes indicate the expression of genes encoding for ligands of interest in both SCs and MCSs. Green nodes indicate genes with a notable expression change, and yellow nodes indicate genes with no or minor expression changes between normal, de-epithelialized (24 hours after de-epithelization), or denervation (5 days after denervation) conditions, as shown in A. The purple node indicates de novo expression of Osm by MSCs following corneal denervation of (inj) corneas.

Following de-epithelialization, transcriptional changes in the gene expression of several trophic factors by SCs were identified by the scRNA-seq analysis, including upregulation of Ngf, Ptn, Pdgfa, and Wnt4 (Fig. 5B). The ligands encoded by these genes were previously shown to regulate the activity of stem cells, including LSCs,^14,76–81 and were implicated in regulation of the processes of self-renewal, differentiation, and migration.^82–86 Notably, unlike in other regenerating organs, where Sox10-positive mature SCs dedifferentiate to Sox2-positive SC precursors producing cytokines and growth factors that mediate recovery,^58,59 we did not observe any changes in the Sox10/Sox2 ratio in SCs following de-epithelialization (Fig. 5C). This finding suggested a different mechanism responsible for the SC trophic activity. Interestingly, a high similarity in expression of most of those genes was observed between SCs and MSCs (Figs. 5B, 5D). Following denervation, mesenchymal cells began to express oncostatin M (Osm) (Fig. 5D), which, together with platelet-derived growth factor alpha (PDGFα), have been reported to induce digit tip regeneration.⁵⁸ There was a significant reduction in the proliferation of MSCs following denervation as assessed by expression of the cell proliferation marker Ki67 (Fig. 5F), suggesting involvement of corneal nerves in supporting MSCs proliferative activity. As expected, the expression of Ki67 was upregulated in LSCs following de-epithelialization (Fig. 5F). A decrease in Ki67 expression in LSCs correlated with its decrease in MSCs following denervation (Fig. 5D). Following denervation, the expression of genes directly associated with regeneration, including corneal healing, such as Mapk1 (Erk2) and Frizzled (Frz), was downregulated (Fig. 5E).^85,87–90 

To model the paracrine crosstalk and downstream intercellular signaling pathways between SCs and the limbal niche cell populations, we extracted potential ligand–receptor interactions using the transcriptome–cell surface communication model approach, as we previously described.^58,91 We consolidated the data from the three experimental conditions. This analysis indicated receptor–ligand communication between SCs and LSCs and between MSCs and LSCs (Fig. 5G). There was a substantial overlap in the paracrine crosstalk between SCs and MSCs with LSCs (Fig. 5G). Among the predicted interactions, we highlighted several implicated in tissue regenerating processes, including corneal epithelial renewal (Fig. 5G).^14,76–81 Subsequently, these highlighted interactions were segregated into two groups: (i) ligand transcriptional expression by SCs and/or MSCs upregulated following de-epithelialization (e.g., Pdgfa, Wnt4, Ptn, Ngf) (Figs. 5B, 5D, 5G), and (ii) where no obvious expression changes were detected (e.g., Jag1, Tgfb1, Cntf) (Fig. 5G; Supplementary Fig. S5B). 

Collectively, following a refined tissue processing method, our scRNA-seq–based comprehensive expression analysis of the corneal limbus harvested from corneas exposed to normal, de-epithelialized, and denervated conditions demonstrated expression changes in limbal cell populations that are associated with corneal sensory innervation and, specifically, with SCs in the limbus. These findings suggest that limbal SCs play a key regulatory role in mediating corneal epithelial renewal. 

Discussion

While the corneal axons are currently viewed as the key drivers of epithelial maintenance and regeneration,^65,66,92 corneal SCs had not been examined for their role in activity of the limbal niche. Indeed, the results of our study indicate a central role of the corneal nerve–associated SCs in the regulation of the renewal of the corneal epithelium. The nerve fibers may be viewed as SC “carriers.” We report that (i) myelinating and non-myelinating corneal nerve-ensheathing SCs are present in abundance in the limbus, (ii) SC survival requires the presence of corneal nerves, and (iii) ablation of corneal SCs alone, even in the presence of corneal nerve fibers, is sufficient to recapitulate all aspects of the NK phenotype, including the development of corneal opacification and failure of epithelial wound healing (Figs. 1, 2). These findings extend those of previous studies that have demonstrated that terminal axon-associated SCs are key regulators of regeneration and stem cells homeostasis in other tissues.^58,59,93 

Although, neural crest cells transcription factor Sox10 specifically characterizes terminally differentiated SCs,^60,61 it has been suggested that it is also expressed by MSCs in the mouse cornea.⁹⁴ Because MSCs are involved in LSCs activity,^13,14 it was critical for us to ensure that the genetic ablation of Sox10-expressing cells was restricted to SCs. The morphological analysis of tdTomato-expressing cells together with immunostaining (Fig. 1) suggested that SCs are the only Sox10-experssing cell population in the cornea. This suggestion was later supported by scRNA-seq analysis of healthy and wounded corneas that indicated that SCs are the only cell cluster containing a detectable level of Sox10 mRNA (Fig. 4B; Supplementary Fig. S5). No detectable expression of tdTomato was observed in SCs or other cell populations located in subcorneal eye compartments, indicating that there was no cross-corneal diffusion of the topically applied tamoxifen. These observations suggest that, in Sox10-tdT;DTA mutant mice, topical eye administration of tamoxifen specifically induces ablation of SCs in the treated cornea without affecting other tissues implicated in the corneal epithelial renewal. The discrepancy between previously suggested Sox10 expression by MSCs⁹⁴ and our results could be explained by conceptual differences in the chosen experimental assays. While we analyzed in vivo gene expression, Su et al.⁹⁴ based their assumption on immunostaining of isolated cultured cells expressing MSCs markers that might affect their expression profile in ex vivo conditions. 

In mouse models of skin and digit tip regeneration, upon injury of the innervating terminal axons, Sox10-positive SCs de-differentiate into Sox2-positive SC precursors. These SC precursors migrate from the damaged tissue and infiltrate the wound bed to trophically induce regeneration by stimulating the wound site–residing tissue progenitor cells.^58,59 It was suggested that, as in skin, cornea-wide homogeneously distributed epithelial stem cells differentiate vertically to replenish the upper layers of epithelial cells.⁸ However, recent reports using scRNA-seq and lineage tracing approaches in mice show unequivocally that LSCs, being the only source of the epithelial cells, are restricted to the limbal niche, which is also the mostly richly innervated part of the cornea (Fig. 3D).^9–13 In contrast to injury of the skin and digit tip that involves damage of the tissue innervating axon terminals, physical ablation of the epithelium at the central cornea does not affect the integrity of the limbal nerves. Also, unlike in skin and digit tip, which recover in several weeks,^58,59 the epithelial wound closure in the cornea occurs within 48 hours (Figs. 3F, 5A). Our findings suggest a different mechanism of SC-dependent corneal epithelial recovery not dependent on SC de-differentiation and migration. Our scRNA-seq analysis of limbi dissected from de-epithelialized corneas indicated no change in the Sox10/Sox2 ratio compared to that of healthy corneas (Fig. 5C). Also, after de-epithelialization we did not detect any migration of limbal SCs during the epithelial recovery (not shown). Although further research is required to define the type of limbal axon-ensheathing SCs (non-myelinating and/or myelinating) involved in regulating corneal epithelial renewal and to experimentally demonstrate the paracrine interaction between SCs and LSCs, we propose a different mechanism for the role of SCs in the mammalian cornea. According to this model, unlike in other organs, the terminally differentiated SCs associated with limbal axons stimulate LSCs in the limbus distant from the wound site to support normal homeostatic epithelial turnover and induce its recovery after injury. 

How do Corneal SCs Regulate Epithelial Renewal?

It has been suggested that the release of several growth factors from corneal axonal terminals, stroma, and tears regulates corneal epithelial renewal by stimulating LSC proliferation and migration.^{42,45,92,95–98} Thus far, rhNGF is the only clinically approved compound shown to improve corneal clarity in patients with NK.^48,55,99 However, inconsistent results of rhNGF treatment, the need for extremely high doses, the inconvenience of needing to dose every 2 hours, and the duration of treatment required of the factor^48,99 suggest the existence of a more complex mechanism of regulation of LSC activity-dependent corneal epithelial cell renewal. 

Tissue regeneration processes in other organs, coupled with our SC ablation results (Fig. 3), suggest that SCs are a potential source of the multiple trophic factors that regulate this process. To address this hypothesis, we performed scRNA-seq analysis of dissociated rat limbi, comparing healthy, de-epithelialized, and denervated corneas. Previous studies that used scRNA-seq to define limbal cell populations and their potential interactions failed to either identify SCs in rodents or clearly differentiate between LSC and MSC clusters and subclusters.^9,67,69 Therefore, we used a multistep digestion process to maximize the survival of the limbal cell populations. Among the identified limbal cell clusters (Figs. 4C, 4D), we focused on the MSCs that are thought to regulate LSC maintenance and self-renewal^14,100–102 and the SCs. We used the data to model the paracrine crosstalk between SCs and LSCs. Among those ligand-encoding genes, Ngf, Wnt4, Ptn, and Pdgfa were upregulated transcriptionally in SCs following corneal de-epithelialization (Fig. 5), suggesting roles in the regulation of regenerative epithelial renewal. Interestingly, a high overlap in potential paracrine interactions was predicted between SCs and LSCs and between MSCs and LSCs (Figs. 5B, 5E, 5G). However, unlike in SCs, we found no transcriptional upregulation of Ngf, Wnt4, Ptn, or Pdgfa in MSCs after de-epithelialization (Fig. 4E). Based on these observations, we hypothesize that SCs and MSCs play different roles in the regulation of LSCs after injury. While MSCs are the limbal niche cells responsible for stem cell maintenance and self-renewal,^100,102,103 SCs are likely involved in the regulation of their differentiation and epithelial renewal, as they are in other regenerating tissues.^58,59,93 

The failure to detect SC clusters during scRNA-seq analysis of dissociated limbi harvested from denervated corneas (Supplementary Fig. S5A) was confirmed by our in vivo finding that SCs are gradually lost after denervation (Fig. 3A). Following corneal denervation and loss of SCs, the levels of Erk2 and Fzd associated with LSC proliferation and self-renewal⁹⁰ were reduced, as well as that of cycling LSCs (Figs. 5F, 5E), correlating with a delayed epithelial recovery (Fig. 5A). The levels of cycling MSCs after denervation were also reduced. These findings suggest that the corneal innervation required for limbal niche maintenance is associated with the proliferative cues provided by limbal SCs. On the other hand, the upregulation in transcriptional expression of NGF and Osm by MSCs shortly after denervation suggests that MSCs induce axonal regeneration and support LSC activation until corneal re-innervation has occurred. 

LSCs are rare, slowly cycling cells that reside in the opaque limbus, a ring-shaped zone at the corneal–conjunctival boundary.^7,15,104,105 LSC clusters from healthy corneas were identified in five Ki67-expressing dividing subclusters (0, 3, 4, 8, and 10) and one subcluster (5) with no detectable expression of Ki67, along with a high expression of the quiescent LSC-specific markers Lgals7 and Gpha2 (Figs. 4C, 4D) and the lowest value of total RNA reads per cell. These findings support the recently introduced model of two subpopulations of LSCs: (i) faster proliferating LSCs that maintain homeostasis, and (ii) quiescent LSCs that serve as a boundary between the conjunctiva and the LSC reservoir, composing the inner and the outer limbal ring, respectively.^9,12,106 Intriguingly, among the LSC subclusters, subcluster 4 contained the highest level of Lgals7 and Gpha2 mRNA and was also positive for MSC-expressing Pdgfra and Pdgfrb, as well as embryonic stem cell markers Jag1, Osr2, Sox9, Etv1, and Foxc2.⁵⁸ This cluster also expressed the highest levels of Ki67 (Figs. 4C, 4D). These findings raise the question of the existence of a possible subpopulation of early stromal mesenchymal cells serving as a progenitor source of both LSCs and MSCs. 

In summary, these loss-of-function observations illustrate the importance of limbal SCs in innervation-dependent corneal epithelial renewal. Our single-cell mRNA expression analysis of the limbal cell populations suggests that terminally differentiated SCs are a potential source of multiple trophic factors, including NGF and predicts a model of trophic interactions between SCs and LSCs associated with LSC activity in epithelial maintenance and/or regeneration after injury. 

Acknowledgments

Supported by the Foundation for Fighting Blindness Canada (KF and GHB) and by the Canadian Institute for Health Research (DRK). 

Disclosure: K. Mirmoeini, None; K. Tajdaran, None; J. Zhang, None; T. Gordon, None; A. Ali, None; D.R. Kaplan, None; K. Feinberg, None; G.H. Borschel, None