Single cell RNA-seq characterization of non-fibrotic stromal wound repopulation in the rabbit

Abstract

Following injury or surgery, quiescent stromal keratocytes can transition into fibroblasts or myofibroblasts leading to either transient or protracted corneal haze. In this study, we investigate the transcriptional changes associated with non-fibrotic wound healing using a transcorneal freeze injury (FI) in the rabbit, which induces full-thickness stromal cell loss without inducing keratocyte-myofibroblast transformation.

In control corneas, scRNA-seq revealed multiple clusters expressing markers associated with keratocyte identity (e.g. KERA, LUM, DCN, and ALDH1A1), suggesting heterogeneity in stromal keratocytes in the uninjured stroma. On day 7 after FI, in vivo imaging revealed elongated cells with increased backscatter, consistent with fibroblast migration into the wounded region. Using scRNA-seq, two additional clusters expressing fibroblast markers were also identified. These clusters retained many markers consistent with keratocyte identity, and trajectory analysis demonstrated a continuous progression from quiescent keratocytes to fibroblasts. Both fibroblast clusters had elevated expression genes encoding tenascin C (TNC), claudin 5 (CLDN5), developmental proteoglycans (e.g., BGN, ASPN, VCAN), and cytoskeletal genes (MYL9, MYH10, CDH11), but did not express markers of myofibroblast transformation. Together these genes suggest a mechanically active but non-fibrotic phenotype. One of the two fibroblast clusters also expressed genes related to cell proliferation. By day 28, fibroblastic gene expression was reduced, consistent with resolution of wound healing.

These findings define the transcriptional dynamics of intrastromal cell migration following FI and reveal a transient fibroblastic state that supports wound repopulation without fibrosis. Understanding this non-fibrotic repair mechanism could inform strategies to prevent scarring following corneal surgery or injury.

1. Introduction

The corneal stroma, which makes up 90% of corneal thickness, is a highly organized structure consisting of collagen lamellae with specific packing and spacing that is critical to the maintenance of corneal transparency. Corneal stromal cells (keratocytes) reside between the collagen lamellae and are responsible for secreting extracellular matrix (ECM) components required to maintain normal corneal structure and function (i.e. transparency and refractive power). While keratocytes are normally quiescent, following injury, surgery, or other insults, corneal keratocytes can become activated by growth factors in the wound environment and transform into fibroblast and myofibroblast phenotypes (Blalock et al., 2003; Garana et al., 1992; Huh et al., 2009; Jester et al., 1999b; Singh et al., 2014; Stramer et al., 2003; Tomasek et al., 2002). Myofibroblasts generate strong mechanical forces on the matrix and synthesize a disorganized fibrotic extracellular matrix. Together, these processes can affect corneal shape and reduce visual acuity through a reduction in transparency (Boote et al., 2012; Dupps and Wilson, 2006; Jester et al., 2012; Moller-Pedersen et al., 2000; Møller-Pedersen et al., 1998; Ruberti et al., 2011), and corneal fibrosis and scarring are leading causes of blindness worldwide. Even routine clinical procedures, such as photorefractive keratectomy (PRK) and laser assisted in situ keratomileusis (LASIK) can lead to fibrosis in about 2–4 % of eyes, with the chance of developing haze being proportional to the correction level needed (Hersh et al., 1997; Kuo et al., 2004; Lipshitz et al., 1997; Maurer et al., 1997; Møller-Pedersen et al., 1998; Shah et al., 1998; Siganos et al., 1999).

In addition to fibrosis which develops on top of the stromal wound bed, most refractive surgeries induce a region of keratocyte death beneath the laser-treated area (Mohan et al., 2000; Møller-Pedersen et al., 1998; Wilson, 2002). Stromal cell death can also be induced by toxic injury (Jester et al., 1998; Maurer et al., 1997), as well as UV cross-linking (CXL) of the cornea in keratoconus patients (Knappe et al., 2011; Mencucci et al., 2010; Wollensak et al., 2004). Ideally, repopulation following these insults should occur via intra-stromal migration of keratocytes from the surrounding stromal tissue, without generation of strong contractile forces or fibrosis that could disrupt the collagen architecture and reduce transparency. Transcorneal freeze injury (FI) is an established model for studying intra-stromal migration, as it induces a full-thickness region of cell death within the stroma (Ichijima et al., 1993; Petroll et al., 1997, 2015). Importantly, FI allows for characterization of non-fibrotic wound repopulation, as myofibroblast transformation is not induced by this injury.

Our lab has developed and applied high resolution multidimensional imaging techniques for assessing the cornea both in vivo and in situ, including quantitative in vivo confocal microscopy, and in situ multiphoton and second harmonic generation imaging (Kivanany et al., 2016, 2018; Petroll et al., 2015, 2020). Importantly, we have established a correlation between the changes in keratocyte morphology and backscatter observed in vivo, and the cytoskeletal organization and cell/ECM patterning observed in situ during different stages of wound healing in the rabbit model (Kivanany et al., 2018). We previously used these approaches to study changes in cytoskeletal organization and cell morphology following FI (Petroll et al., 2015), and observed a unique pattern of keratocyte alignment and connectivity that was highly correlated with the structural organization of the lamellae, suggesting topographic guidance of intra-stromal cell migration. While biophysical cues have been shown to impact fibroblast behavior using in vitro models (Karamichos et al., 2014; Myrna et al., 2012; Teixeira et al., 2004), this was the first direct demonstration that ECM structure mediates the pattern of corneal fibroblast migration during in vivo corneal wound healing. However, the gene expression patterns and associated signaling pathways that underlie this mechanism of intra-stromal migration are still unknown.

Next generation sequencing (NGS) can provide a comprehensive assessment of gene expression patterns from both cultured cells and living tissues (Collin et al., 2021; Ma and Lwigale, 2019). Bulk (total) RNA sequencing allows for high sensitivity whole transcriptome analysis of a sample and comparison between varying experimental conditions or normal versus disease states. Bulk RNA analyses have been performed for normal human cornea (Ligocki et al., 2021; Tokuda et al., 2020), sulfur mustard injured rabbit cornea (Horwitz et al., 2019), and post-PRK surgery in human cornea (Kumar et al., 2019). However, total RNA sequencing is limited in that expression patterns specific to keratocytes cannot be determined. In contrast, single cell RNA sequencing (scRNA-seq) analysis allows for the elucidation of gene patterns specific to and compared with each cell population within a heterogenous tissue. Prior studies of scRNA-seq have been conducted using normal human adult cornea (Català et al., 2021; Collin et al., 2021; Dou et al., 2021; Gautam et al., 2021; Li et al., 2021a, 2021b, 2022; Ligocki et al., 2021; Maiti et al., 2022), normal rabbit cornea (Bargagna-Mohan et al., 2021), normal human fetal cornea (Collin et al., 2021), human cornea with keratoconus (Collin et al., 2021; Dou et al., 2022), Fuchs endothelial corneal dystrophy (Wang et al., 2022), or limbal dysplasia (Collin et al., 2021) as well as human derived corneal organoids (Maiti et al., 2022). However, human tissue undergoing wound healing is not generally available, and to our knowledge, a comprehensive assessment of the keratocyte gene expression patterns associated with specific phases of wound healing has not yet been performed.

In vivo models such as the rabbit afford access to the full corneal thickness, allow for longitudinal study of wound healing, and tissue isolation at discrete, controlled timepoints, overcoming limitations of acquiring living patient tissue. In addition, the rabbit is a standard, accepted model to study the corneal stroma, with corneal diameter, cellular and ECM composition, and geometry comparable to humans (Zernii et al., 2015; Peiffer et al., 1994). In this study, we combine in vivo confocal imaging with scRNA-seq after FI, with the goal of identifying gene expression patterns that may drive non-fibrotic repopulation of corneal stromal wounds.

2. Materials and methods

2.1. Animals

Experiments in this study were performed using 8 young (3–6 months old at preoperative imaging) New Zealand White Rabbits (8 females; 2.9–3.6 kg; Charles River Laboratories, Wilmington, MA, USA). All animal procedures adhered to the ARVO (Association for Research in Vision and Ophthalmology) Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the UT Southwestern Institutional Animal Care and Use Committee (IACUC).

2.2. Transcorneal freeze injury (FI) model

Rabbits were anesthetized with intramuscular ketamine (50 mg/kg body weight) and xylazine (5 mg/kg body weight), and Buprenorphine SR (slow release) analgesic was administered subcutaneously (0.3 mg/kg body weight). Local anesthesia was administered topically to the left eye of each rabbit with one drop of preserved 0.5 % proparacaine hydrochloride (Alcon Laboratories, Fort Worth, TX, USA), and a speculum was placed into the eye. A 3 mm diameter stainless steel probe was cooled using liquid nitrogen and immediately applied to the anterior, central corneal surface for 10 s – this was repeated two additional times (Petroll et al., 2015). This procedure consistently creates a region of cell death through the full thickness of the cornea without significantly altering ECM structure (Ichijima et al., 1993; Petroll et al., 1997, 2015). One drop of 0.3 % gentamicin sulfate ophthalmic solution (Bausch + Lomb, Bridgewater, NJ, USA) was applied twice a day to the left eye for 5 days.

2.3. In vivo imaging

Anterior segment optical coherence tomography (OCT) and in vivo confocal microscopy through focusing (CMTF) was performed one week preoperatively (Pre-Op), and at day 7 and day 28 postoperatively – timepoints selected based on our previous studies characterizing different phases of wound healing following FI (Petroll et al., 2015). For OCT, rabbits were anesthetized with intramuscular injection of ketamine (50 mg/kg body weight) and xylazine (5 mg/kg body weight) and positioned with the central cornea aligned to a Heidelberg Spectralis OCT (HRA-OCT with an OCT2 Advanced Unit; Heidelberg Engineering, GmBH, Dossenheim, Germany). Cross-sectional radial scans of the cornea were obtained with the corneal curvature set to 7.3 mm. Scans were viewed in the accompanying Heidelberg Eye Explorer software (Version 1.10.4.0; Heidelberg Engineering, GmBH, Dossenheim, Germany).

Following OCT, in vivo confocal microscopy with a custom modified Heidelberg Retinal Tomograph with Rostock Corneal Module (HRT-RCM; Heidelberg Engineering, GmBH, Dossenheim, Germany) was used to qualitatively and quantitatively characterize phenotypic changes over time following FI as previously described (Petroll and Robertson, 2015; Petroll et al., 2013). Briefly, Systane lubricant night gel (Alcon Laboratories, Fort Worth, TX, USA) was applied to the lens cap and surface of the eye. Consecutive images were acquired at a constant speed of 60 μm/s starting from the anterior chamber and finishing past the corneal epithelium. Each scan consists of an image field of view of 400 μm × 400 μm, with a 2 μm step size between each image. All images were acquired with gain at 10, which is set by unchecking the “Auto Brightness” box in the Heidelberg Engineering software and moving the slider 10 steps to the right. At least 3 scans were collected in the central cornea of the FI eyes at each time point, and a minimum of 8 total scans were collected across the wounded and unwounded areas of the corneas. Additional images were acquired in the contralateral control eyes. All 3D image stacks were saved as “.vol” files.

The “.vol” files were then loaded in our custom CMTF software (Ophthalmology Dept. UT Southwestern Medical Center © 2011), which calculates the average pixel intensity in the central region (a user selectable 200:200 pixels, x:y) of each image within the sequential acquisition, and plots it versus depth (z) (Li et al., 2000). Using this intensity curve and the corresponding image, we labelled the position of the epithelium, basal lamina, and endothelium. For a control eye, these positions correspond to the maxima of three intensity peaks along the curve. The software then calculated the thicknesses of the epithelium, stroma, and total cornea as the distance from the epithelium-to-basal lamina, basal lamina-to-endothelium, and epithelium-to-endothelium, respectively. A relative estimate of haze in the stroma was calculated by measuring the area under the intensity curve from the basal lamina-to-endothelium positions, using a baseline intensity of 10 (the value just above the intensity of the anterior chamber).

2.4. Tissue digestion for single cell suspension

At both 7 and 28 days postoperatively, two rabbits (four total) were euthanized with intravenous injection of sodium pentobarbital (120 mg/kg body weight). Central corneal buttons were excised from both the FI procedure and control eye of each rabbit using an 8 mm diameter trephine. For the FI eye, care was taken to center the 3 mm diameter wound within the 8 mm diameter button. The corneal buttons were sliced into five vertical strips. These strips were digested in a 37 °C incubator (5 % CO₂) with 2D rocking for approximately 3 h in 1 mL of media containing 2 mg/mL collagenase (Gibco, Grand Island, NY, USA), and 0.5 mg/mL hyaluronidase (Worthington Biochemicals, Lakewood, NJ, USA) in minimum essential medium (MEM; Gibco, Grand Island, NY, USA) supplemented with 2 % penicillin/streptomycin/amphotericin (Lonza, Walkersville, MD, USA). This media solution was previously optimized for keratocyte extraction from rabbit corneas for primary cell culture (Jester et al., 1996). The total corneal incubation time is similar to that used in previously published scRNA-seq studies (Bargagna-Mohan et al., 2021; Català et al., 2021; Gautam et al., 2021; Maiti et al., 2022; Wang et al., 2022).

Following incubation, the solution underwent centrifugation (1500 rpm, 4 min), the supernatant was discarded, and the pellet was resuspended in 1 mL 0.05 % trypsin (Gibco, Grand Island, NY, USA) for incubation at 37 °C for 3 min. The trypsinization reaction was halted with 4 mL of trypsin inhibitor (Gibco, Grand Island, NY, USA). After centrifugation (1500 rpm, 4 min), the supernatant was discarded, and the pellet was washed in 1 mL 4 °C chilled PBS (MP Biomedicals, Solon, OH, USA). The cell suspension was filtered sequentially with 70 μm and 40 μm porosity Flowmi^® cell strainers (Bel-Art, Wayne, NJ, USA). Following centrifugation (1500 rpm, 4 min), the supernatant was discarded, and the single cells were resuspended in 500 μL 4 °C chilled PBS enriched with 0.4 % bovine serum albumin (Sigma, St. Louis, MO, USA).

2.5. Single cell RNA sequencing

Ice-cold, freshly prepared single cell suspensions were submitted to the UT Southwestern McDermott Center Next Generation Sequencing Core for single cell RNA sequencing (scRNA-seq), where sample viability was assessed via a Countess III cell counter following trypan blue staining (Thermo Fisher Scientific, Bothell, WA, USA). Suspensions of ~10,000 cells per sample were loaded into a 10X Genomics Chromium controller for transcript barcoding and sequenced on an Illumina Next-Seq sequencer (Illumina, San Diego, CA, USA).

2.6. Bioinformatic analysis

Cell Ranger version 5.0.1 (10 × Genomics) was used to process the raw sequencing data. Briefly, raw BCL files were converted to FASTQ files and aligned to a rabbit RabGTD reference transcriptome (Zhou et al., 2018) built in-house to fit the 10x Genomics format. Raw feature–barcode matrices were first denoised using CellBender (remove-background; CellBender, version 0.3.0) to remove ambient RNA and technical background. For each sample, the raw 10x Genomics feature–barcode matrix was used as input, and a corrected count matrix (HDF5 format) was generated for downstream analyses. The CellBender-corrected gene-by-cell expression matrices were then loaded into the R package Seurat version 4.0.5 for downstream analyses (Hao et al., 2021). To identify potential doublets, we used scDblFinder on the CellBender-corrected matrices and removed predicted doublets prior to downstream analyses. Cells with low quality were filtered out based on nUMI ≥500, nGene ≥250, mitoRatio <0.20, and a novelty score where log10GenesPerUMI >0.80. Only genes detected in more than 10 cells were retained. These thresholds were chosen based on inspection of the distributions of these QC metrics in our data (to remove clear low-complexity or high-mitochondrial outliers) (Supplemental Figs. 1–4) and are consistent with commonly used cutoffs in single-cell RNA-seq studies. The “LogNormalize” global-scaling normalization method in Seurat was performed only for cell-cycle scoring, while all integration and downstream analyses were performed using SCTransform. With these methods in place, 32,406 cells were obtained from the four contralateral control eye samples, 16,616 cells from the two FI procedure samples at Day 7, and 14,330 cells from the two FI procedure samples at Day 28.

Samples were normalized and integrated using Seurat’s SCT-based integration workflow, with integration features selected by SelectIntegrationFeatures and anchors identified using FindIntegrationAnchors. The integrated assay was used for downstream analyses. Highly variable genes were identified with FindVariableFeatures. The data was scaled, and dimensional reduction was performed with principal component analysis followed by UMAP. A shared nearest neighbor (SNN) graph was constructed using FindNeighbors, and clusters were identified with FindClusters, with specific parameters utilized shown in Supplemental Table 1. We obtained 14 clusters from the control samples (resolution 0.6), 16 clusters from the Day 7 procedure samples (resolution 0.6), and 15 clusters from the Day 28 procedure samples (resolution 0.6). Specific UMAP plots, feature plots, violin plots, and dot plots were generated using Seurat version 4.3.1.

2.7. Single cell trajectory analysis

The output from Seurat was fed into the Monocle3 tool (Qiu et al., 2017). From the UMAP visualization, a root node state was identified within the keratocyte clusters. A trajectory graph was generated to order cells within the UMAP by pseudotime.

2.8. Immunocytochemistry

Additional eyes were collected for sectioning and staining on day 7 and day 28. Following euthanasia with intravenous injection of sodium pentobarbital (120 mg/kg body weight), corneas were immediately excised. The corneas were then dissected into central or peripheral regions using a single edge blade, embedded with a sagittal plane presentation within Tissue Plus^® O.C.T. Compound (Fisher Healthcare, Houston, TX, USA), snap-frozen in liquid nitrogen and stored at −80 °C. Sections of 10 μm thickness were cut using a Leica Cryostat (CM3050 S; Leica, Deer Park, IL, USA), and mounted on Superfrost^™ Plus microscope slides (Fisher Scientific, Pittsburgh, PA, USA). Samples were fixed in aqueous 4 % paraformaldehyde (Electron Microscopy Services, Hatfield, PA, USA) for 10 min at 4 °C, washed twice with PBS, then permeabilized with 0.5 % Triton X-100 in PBS for 30 min, and then washed once with PBS. The samples were then blocked with 4 % bovine serum albumin fraction V (Equitech-Bio, Kerrville, TX, USA) overnight at 4 °C. The samples were then incubated overnight at 4° with primary antibody solution. The following primary antibodies were used: anti-α-SMA (1:200 dilution; A5228; Sigma-Aldrich, St. Louis, MO, USA) and anti-tenascin C conjugated to Alexa Fluor 488 (1:150 dilution; sc-25328 AF488; Santa Cruz Biotechnology, Dallas, TX, USA). Samples were washed with PBS 4 times. Samples labelled for tenascin C were then incubated with Alexa Fluor 633 phalloidin (1:200 dilution; Invitrogen, Waltham, MA, USA) for 2 h at 37 °C. Samples labelled for α-SMA were incubated with fluorescein-conjugated goat anti-mouse antibodies (1:200 dilution; Jackson Laboratories, West Grove, PA, USA) for 2 h at 37 °C. After washing 4 times, the samples were incubated at room temperature with 4′6-diamidino-2-phenylindole (DAPI; Invitrogen, Waltham, MA, USA) for 30 min. Imaging was performed on a laser scanning confocal microscope (Leica SP8, Heidelberg, Germany). A 488-nm laser was used for imaging antibodies, a 633-nm laser was used for imaging f-actin, and a UV laser (405-nm) was used to image DAPI. Z-series were acquired using a X25 water immersion objective (0.95NA, 2.4-mm free working distance). Sequential scanning was used to avoid crosstalk between channels.

2.9. Statistics

Statistical analysis of CMTF data was performed in GraphPad Prism 10 for macOS.

3. Results

3.1. In vivo confocal imaging

Prior to surgery (Pre-Op), the superficial epithelium, stroma, and endothelial monolayer were easily distinguishable in CMTF scans. Keratocytes appeared quiescent, as indicated by the primarily nuclear source of backscatter (Fig. 1A, Pre-Op). Unoperated contralateral (OD) control eyes looked similar to Pre-Op eyes, without evident structural defects (Not shown), and there were no significant changes in thickness (Not shown) or stromal haze observed over time (Fig. 1C).

Fig. 1.

Images and measurements from in vivo confocal scans. (A) Representative images from the pre-operative (Pre-Op) central cornea, 7 days after FI within the central wounded region (Day 7 – W), 7 days after FI outside the wound (Day 7 - OW) region, and 28 days after FI within the central cornea (Day 28 – W). Top row shows 3-D reconstructions of CMTF scan. Bottom rows show in vivo 2-D confocal images from the location indicated by the letters on the corresponding 3-D reconstruction. (B) Measurements of epithelial and stromal thickness at each time point in wounded eyes. (C) Stromal backscatter (haze) calculated at each time point. Tukey’s multiple comparisons test between groups, *P < 0.05, ***P < 0.0005.

At Day 7 following FI, the central cornea was edematous, as indicated by a significant increase in stromal layer thickness compared to Pre-Op (Fig. 1B). Throughout the full thickness of the central stroma, elongated cells with higher reflectivity were observed, suggesting fibroblast migration into the FI wound area (Fig. 1A, Day 7 – W). This increase in cellular reflectivity corresponded to the maximum observed stromal haze (Fig. 1C). When observed with OCT imaging, this haze was also visible in the wound area (Not shown). Keratocytes outside the FI wound area at Day 7 were more quiescent (Fig. 1A, Day 7 – OW), with predominately nuclear backscatter as observed in contralateral controls (Not shown). Both the epithelium and endothelium had resurfaced by day 7 (Fig. 1A and B).

By Day 28, keratocytes appeared to be more quiescent in the wound area as compared to Day 7, and elongated cells indicative of the migratory fibroblastic phenotype were no longer observed (Fig. 1A, Day 28 – W). Haze was closer to normal inherent backscatter levels and was significantly less than that observed at Day 7 (Fig. 1C). Overall, the results at both Day 7 and Day 28 were consistent with previously published studies characterizing the cell and extracellular matrix (ECM) changes following FI (Petroll et al., 2015).

3.2. scRNA results

3.2.1. Identification of normal expression patterns within control corneas

Following in vivo imaging, rabbits were sacrificed for immediate central corneal excision. Central 8 mm buttons from both injured (OS) and contralateral control (OD) eyes were isolated from 2 rabbits at Day 7 after FI, and 2 rabbits at day 28 after FI. Each corneal button was disassociated into a single cell suspension and submitted for sequencing. The four samples from the control eyes were individually filtered for quality control metrics and were combined for a total cell count of 32,406. These cells were embedded in a uniform manifold approximation and projection (UMAP), where unbiased clustering yielded 14 clusters when visualized with a resolution of 0.6 (Fig. 2A). Analyzing within these clusters, we find that Clusters 2, 3, 5, 6, and 12 expressed markers consistent with keratocyte identity (Table 1), such as proteoglycans, stromal collagens, corneal crystallins, and intermediate filaments (Hassell and Birk, 2010; Jester, 2008; Jester et al., 1999a; Meek, 2009; Meek and Knupp, 2015). Specifically, expression of keratocan (KERA), lumican (LUM), decorin (DCN), mimecan (OGN), collagen type I (COL1A1), collagen type V (COL5A1), and vimentin (VIM) were elevated as compared to other clusters captured in our dataset (Fig. 2B). Corneal crystallin transketolase (TKT) and aldehyde dehydrogenase 1 Family Member A1 (ALDH1A1) were highly expressed in both keratocyte clusters and other cell types (Fig. 2B). The expression patterns observed are consistent with prior scRNA-seq results from human corneal tissue (Català et al., 2021; Collin et al., 2021; Ligocki et al., 2021; Maiti et al., 2022). Note that ALDH3A1 is normally expressed in human tissue whereas ALDH1A1 is expressed in the rabbit (Jester et al., 1999a).

Fig. 2.

Single cell RNA sequencing data of the central cornea of control eyes. An integrated dataset from four rabbits, consisting of 32,406 single cells. (A) UMAP visualization of 14 clusters identified by unbiased clustering with resolution of 0.6. (B) Feature plots of expression level of markers previously associated with keratocyte identity. (C) UMAP visualization of the four distinct cell populations identified using the markers shown in D. (D) Violin plots showing expression level of genes KERA (keratocyte marker gene), KRT3 (epithelial marker gene), SLC4A11 (endothelial marker gene), and IL6 (immune cell marker gene) across the four major cell populations.

Table 1.

Table of marker genes utilized for annotation of keratocyte, epithelial, endothelial, and immune cells.

Major Cell Types
Cell Type	Marker	Key References
Epithelium	KRT3	(Bargagna-Mohan et al., 2021; Català et al., 2021; Collin et al., 2021; Dou et al., 2021, 2022; Li et al., 2021a; Ligocki et al., 2021; Maiti et al., 2022; van Zyl et al., 2022)
	KRT5
	KRT12
Keratocyte	KERA	(Hassell and Birk, 2010; Jester, 2008; Jester et al., 1999a; Meek, 2009; Meek and Knupp, 2015)
	LUM
	DCN
	OGN
	COL1A1
	COL5A1
	VIM
	TKT
	ALDH1A1	Jester et al. (1999a)
Endothelium	SLC4A11	(Català et al., 2021; Ligocki et al., 2021; Van den Bogerd et al., 2019; Wang et al., 2022)
	COL8A1	(Van den Bogerd et al., 2019; Wang et al., 2022)
Immune Cell	IL6	(Collin et al., 2021; Dou et al., 2022)
	CD74	(Català et al., 2021; Gautam et al., 2021; Li et al., 2022)

In addition to keratocytes, clusters expressing markers for epithelium, endothelium, and immune cells were identified using markers previously utilized within single cell corneal characterizations (Table 1, Fig. 2C). Cluster 10 is likely the endothelial cell population, with upregulation of collagen type VIII (COL8A1) (Van den Bogerd et al., 2019; Wang et al., 2022) and SLC4A11 (Català et al., 2021; Ligocki et al., 2021; Van den Bogerd et al., 2019; Wang et al., 2022) (Fig. 2D). Cluster 13 can be identified as a resident immune population with markers IL6 (Collin et al., 2021; Dou et al., 2022) (Fig. 2D) and CD74 (Català et al., 2021; Gautam et al., 2021; Li et al., 2022). Clusters 0, 1, 4, 7, 8, 9, and 11 exhibit higher expression of keratin 3 (KRT3, Fig. 2D), keratin 5 (KRT5), and keratin 12 (KRT12), indicating epithelial identity (Bargagna-Mohan et al., 2021; Català et al., 2021; Collin et al., 2021; Dou et al., 2021, 2022; Li et al., 2021a; Ligocki et al., 2021; Maiti et al., 2022; van Zyl et al., 2022). Prior publications have identified different expression patterns between epithelial sublayers (Català et al., 2021; Collin et al., 2021; Dou et al., 2021; Li et al., 2021a, 2021b; Ligocki et al., 2021; Maiti et al., 2022), which may explain the differential expression observed between our 7 epithelial subclusters. In addition, Cluster 10 exhibits a small subset of cells with expression of additional markers consistent with non-myelinated Schwann cells (Bargagna-Mohan et al., 2021; Dou et al., 2021), but these do not comprise a distinct cluster.

3.2.2. Expression patterns during active phase of wound healing

The two central corneal buttons from OS eyes collected at day 7 were also disassociated into single cell suspensions and submitted for sequencing. The two samples were individually filtered for quality control metrics and were combined (integrated) for a total cell count of 16,616. These cells were embedded in a UMAP, where unbiased clustering yielded 16 clusters when visualized with a resolution of 0.6 (Fig. 3A). Clusters 2, 3, 4, 5, 13, and 15 all had expression patterns consistent with keratocyte identity, including high expression of KERA (Fig. 3C). Clusters 2 and 15 are distinct, however, in that they also express markers consistent with fibroblast differentiation (Fig. 3D). These two clusters are still theorized to be of a keratocyte lineage, based on expression of KERA among other markers previously associated with a quiescent keratocyte identity (Figs. 3C and 4A–C). Overall, six major cell populations were identified as keratocyte, epithelium, endothelium, immune cells, fibroblast cluster 1, and fibroblast cluster 2 (Fig. 3B). Fibroblast cluster 1 is approximately 27 % and fibroblast cluster 2 is approximately 1 % of the keratocyte marker expressing cells. Note that the central 3 mm diameter wound volume is 14 % of the total 8 mm diameter button volume.

Fig. 3.

Single cell RNA sequencing data of the central cornea from eyes collected 7 days after FI. An integrated dataset from two rabbits, consisting of 16,616 single cells. (A) UMAP visualization of 16 clusters identified by unbiased clustering with resolution of 0.6. (B) UMAP visualization of the six distinct cell populations identified using the markers shown in C. (C) Violin plot showing expression level of genes KERA (keratocyte marker gene), KRT3 (epithelial marker gene), SLC4A11 (endothelial marker gene), and IL6 (immune cell marker gene) across the six major cell populations. (D) Violin plots showing expression of ALDH1A1, FAP, MKI67, TNC, CLDN5, and BGN among keratocyte and fibroblast clusters. (E) Dot plot showing expression level of selected positive differentially expressed gene features over an average Log2fc threshold of 0.5 for the keratocyte cluster (left-most 8 features), fibroblast cluster 1 (middle 24 features), and fibroblast cluster 2 (right-most 13 features) across all keratocyte marker positive clusters. Features with an average Log2fc change >1 are indicated by beige shading within the graphical plot.

Fig. 4.

Dot plots of average expression showing common and differentially expressed genes across fibroblast, keratocyte, epithelial, endothelial, and immune cells. (A) Genes associated with fibroblast identity. (B) Genes associated with collagen deposition and degradation, and ECM modulation. (C) Genes associated with common and atypical proteoglycans. (D) Genes associated with myofibroblast identity and function.

We performed further differential gene expression analysis on the subset of clusters that expressed markers commonly associated with keratocytes. The resultant three clusters, keratocyte, fibroblast 1 and fibroblast 2, had distinct top gene features (Supplemental Fig. 5, Supplemental Files 1–3). Selected genes of interest upregulated within keratocytes versus the fibroblast clusters included proteoglycans (DCN, OGN), collagens normally attributed to basement membrane formation (COL4A3, COL4A4), corneal crystallins responsible for cell transparency (including ALDH1A1), and growth factors (IGFBP2) (Fig. 3D and E, Supplemental Fig. 5).

Fibroblast clusters 1 and 2 both showed upregulation of proteoglycans seen within development [biglycan (BGN)] (Fig. 3D, E, 4B, Supplemental Fig. 5) or increased fibrillogenesis [collagen types (COL1A1, COL5A1)], collagen matrix metalloproteases (MMP2, MMP3, MMP14), TIMP metallopeptidase inhibitors (TIMP2, TIMP1) (Figs. 3E and 4B, Supplemental Fig. 6), and ECM modulators (HAS2) (Supplemental Fig. 6). Markers commonly attributed to fibroblasts, including expression of fibroblast activation protein alpha (FAP), filament protein vimentin (VIM), and glycoprotein (CD44), also had upregulated expression in both fibroblast clusters as compared with keratocytes (Fig. 4A, Supplemental Fig. 6). Tenascin C (TNC), an extracellular matrix protein associated with wound healing in cornea (Kamil and Mohan, 2021; Saika et al., 2016; Sumioka et al., 2023) and other tissues (Bhattacharyya et al., 2022; Midwood et al., 2016), was also upregulated (Fig. 3D, E, 4A). Through immunocytochemistry, we observe increased tenascin C protein expression in the central stromal wound in parallel with increased f-actin labelling at Day 7 (Fig. 5). The elongated cell morphology and increased f-actin, indicated with arrows in Fig. 5, are consistent with a fibroblast phenotype. Tenascin C labeling was obsereved along the full length of migrating fibroblasts, suggesting a possible role in cell-ECM adhesion. One novel gene expressed by stromal fibroblasts was claudin 5 (CLDN5), a tight junction protein normally expressed by other cell types (Fig. 3D and E). Genes that differentiate between the two fibroblast clusters include fibronectin type III domain containing 1 (FNDC1), which was most upregulated in fibroblast cluster 1 followed by fibroblast cluster 2 (Fig. 3E). Fibroblast cluster 2 is further defined by genes suggesting a proliferative capacity, including marker of proliferation Ki-67 (MKI67) (Fig. 3D, E, 4A) and proliferating cell nuclear antigen (PCNA) (Figs. 3E and 4A, Supplemental Fig. 6). A more comprehensive analysis of fibroblast cluster 2 shows increased expression of regulatory genes involved in mitosis and cell cycle progression, including cyclins (CCNB1, CCNB2, CCNA2), cyclin dependent kinases (CDK1, CDK4), cell division cycle protein (CDC20), and centromere proteins (CENPF, CENPU) (Fig. 3E).

Fig. 5.

Images from immunocytochemistry of the central cornea. Top row shows tenascin C (green), f-actin (red), and a composite including DAPI (blue) 7 days after FI within the central wounded region (Day 7 – W). Bottom row shows tenascin C (green), f-actin (red), and a composite including DAPI (blue) 7 days after FI outside the wound (Day 7 - OTW) region. White arrows indicate increased f-actin and antibody labelling of tenascin C.

Proteoglycans KERA, LUM, DCN, OGN, and fibromodulin (FMOD) all exhibit high average expression in keratocyte, fibroblast 1, and fibroblast 2 clusters (Fig. 4C, Supplemental Fig. 6). Additional proteoglycans not commonly expressed in quiescent, adult cornea that are upregulated in the fibroblast clusters include biglycan (BGN), asporin (ASPN), and versican (VCAN). Proteoglycans with higher average expression in the keratocytes, presumably downregulated following fibroblastic transformation, include KERA, DCN, OGN, and osteomodulin (OMD).

Although markers between fibroblasts and myofibroblasts can often overlap, we did not detect upregulation of α-smooth muscle actin (ACTA2) (Fig. 3D, Supplemental Fig. 7A) or fibrotic collagen type III in the fibroblast clusters (Fig. 3B and D, Supplemental Fig. 7A). Consistent with previous studies (Ichijima et al., 1993), we see endothelial fibrosis indicated by white arrows in Supplemental Fig. 7B in a subset of rabbits following FI, which labels positively for α-smooth muscle actin protein, and exhibits prominent f-actin and multiple cell layers. No labeling of α-smooth muscle actin protein was observed within the stroma or other cellular layers of the wound (Supplemental Fig. 7B). Expression of extracellular matrix glycoprotein fibronectin (FN1) was also unchanged in fibroblasts as compared to keratocytes (Not shown). Other genes implicated with myofibroblast differentiation showing a higher average expression in both fibroblast clusters as compared to keratocyte clusters include leucine rich repeat containing 15 (LRRC15); however, the level of LRRC15 expression was still low (Fig. 3D). Additional genes shown to have an increase in this study include myosin chains (MYL9, MYH10) which are associated with cellular force generation, and cadherins (e.g. CDH11) which are associated with cell-cell interactions (Fig. 3D, Supplemental Fig. 6).

To further investigate the cellular dynamics of the two fibroblast clusters, we used pseudotime analysis projections (Supplemental Fig. 8) which indicate a connective relationship between the keratocyte clusters and the two fibroblast clusters. Of particular interest is the singular branch point leading to the fibroblast clusters. This pseudotime analysis supports speculation that stromal fibroblasts are derived from keratocytes.

Though the focus of this work is upon keratocytes, analysis of global features across the total 16,616 cells at Day 7 in comparison to Control eyes and Day 28 following FI yields top gene features associated with immune cell expression including interleukin proteins IL1A and IL18BP (Supplemental Fig. 9). Additionally, the immune cell population, Cluster 7, make up 4.6 % of the cells at this timepoint after FI compared with 0.4 % of the cell population in control eyes.

3.2.3. Expression patterns during late stages of wound healing

The two central corneal buttons collected from OS eyes at day 28 were also disassociated into a single cell suspension and submitted for sequencing. The two samples were individually filtered for quality control metrics and were integrated for a total cell count of 14,330. These cells were embedded in a UMAP, where unbiased clustering yielded 15 clusters when visualized with a resolution of 0.6 (Fig. 6A). Clusters 1, 3, 4, and 11 had expression patterns consistent with keratocyte identity, including elevated expression of KERA (Fig. 6C). With the markers previously described, four major cell populations were identified as keratocyte, epithelium, endothelium, and immune cell (Fig. 6B). There were no longer clusters defined by expression patterns consistent with a fibroblast identity. Overall, there was a marked reduction in fibroblastic gene expression at Day 28 as compared to Day 7 (Fig. 6D), suggesting a transition to a more quiescent phenotype. There was still no upregulation of ACTA2 (Fig. 6D) or COL3A1 (Not shown). Genes previously upregulated at Day 7, including LRRC15 and myosin chains (MYL9 and MYH10) were not noted within the keratocyte population (Not shown). Expression of specific genes of interest we identified within our fibroblast subpopulations, including FAP, TNC, BGN, ASPN, and MKI67 (Fig. 6D), showed a lower percent expression as compared to day 7.

Fig. 6.

Single cell RNA sequencing data of the central cornea from eyes collected 28 days after FI. An integrated dataset from two rabbits, consisting of 14,330 single cells. (A) UMAP visualization of 15 clusters identified by unbiased clustering with resolution of 0.6. (B) UMAP visualization of the four distinct cell populations identified using markers shown in C. (C) Violin plots showing expression level of genes KERA (keratocyte marker gene), KRT3 (epithelial marker gene), SLC4A11 (endothelial marker gene), and IL6 (immune cell marker gene) across the four major cell populations. (D) Dot plot showing expression level gene features associated with fibroblasts within a more quiescent central cornea.

4. Discussion

Single cell RNA sequencing has been used previously to elucidate novel features of normal and disease states of the human cornea (Català et al., 2021; Collin et al., 2021; Dou et al., 2021, 2022; Gautam et al., 2021; Li et al., 2021a, 2021b, 2022; Ligocki et al., 2021; Maiti et al., 2022; Wang et al., 2022); however, this technology has not yet been applied to study the active wound healing response following injury or surgery. Animal models allow for temporal study of wound healing following specific injury types, with tissue isolation at specific timepoints of interest. The current study uses both in vivo imaging and single cell transcriptomics to define the changes in cell phenotype and gene expression that occur during stromal repopulation after FI injury in a rabbit. Since FI induces full-thickness stromal cell loss without triggering myofibroblast transformation of corneal keratocytes (Ichijima et al., 1993; Petroll et al., 1997, 2015), our goal was to identify gene expression patterns associated with non-fibrotic repopulation of corneal stromal wounds.

4.1. Keratocyte - fibroblast transformation following FI

In vivo confocal microscopy revealed temporally dependent structural and cellular changes in the stroma following FI. At Day 7 post-FI, a significant increase in stromal thickness was observed, consistent with edema attributed to endothelial function loss (Ichijima et al., 1993; Petroll et al., 1997, 2015) and the associated disruption in collagen fibril spacing. Concurrently, a population of elongated, highly reflective cells with increased f-actin expression appear in the wounded region, consistent with a fibroblastic morphology and suggestive of active migration. Previous publications suggest that these elongated fibroblasts migrate in alignment with the collagen lamellae (Petroll et al., 2015). These changes were associated with the development of stromal haze at Day 7, which decreased by Day 28 as stromal thickness and cellular backscatter returned to near normal levels. This reduction in haze after Day 7 suggests that fibroblasts may revert to a phenotype more consistent with quiescent keratocytes following wound repopulation.

To assess gene expression patterns in both normal and injured corneas, scRNA-seq was used. In the control corneas, keratocyte clusters comprised 34 % of all captured cells and expressed markers associated with keratocyte identity including KERA, LUM, DCN, and corneal crystallins such as ALDH1A1 (Hassell and Birk, 2010; Jester, 2008; Jester et al., 1999a; Meek and Knupp, 2015). These gene signatures were distributed across multiple clusters, consistent with prior publications (Català et al., 2021; Collin et al., 2021; Dou et al., 2021, 2022; Maiti et al., 2022), suggesting spatial and/or functional heterogeneity in stromal keratocytes even in the uninjured stroma. Future studies using spatial transcriptomics could reveal the distribution of these distinct keratocyte sub-populations. At day 7 after FI, two additional clusters expressing fibroblast markers were also identified. These clusters retained many markers consistent with keratocyte identity but had distinct transcriptional profiles. As discussed in more detail below, these features suggest a coordinated program of matrix adhesion, cell-cell interaction, and motility that underlie the fibroblastic response to stromal injury.

Trajectory analysis revealed a continuous path from keratocyte clusters to both fibroblast subtypes, suggesting that the wound-healing fibroblasts are derived from the surrounding quiescent keratocyte population. Notably, the singular branching point along the trajectory suggests a conserved transformation event rather than multiple activation pathways. Both fibroblast clusters express genes associated with cell motility, matrix interaction and adhesion, but the second additionally exhibits markers of cell proliferation including MKI67, PCNA, and numerous cell cycle regulators. Although the localization of these cells remains to be determined, their presence suggests that both proliferation and migration contribute to intra-stromal wound repopulation.

4.2. Gene expression patterns during fibroblast migration

Two genes of particular interest that were upregulated in corneal fibroblasts at Day 7 were TNC, which encodes tenascin C and CLDN5, which encodes claudin 5. Tenascin has been long documented to play a significant role in the migratory and metastatic capacity within cancers and developing fibrotic environments (Bhattacharyya et al., 2022; Midwood et al., 2016), while also serving as a readout for tissue repair. Tenascin has also been observed in the wounded stroma (Sumioka et al., 2023; Tervo et al., 1989, 1991), with tenascin C deficiency in a mouse model resulting in impaired wound healing (Sumioka et al., 2013). Both fibroblast clusters express tenascin C, uniquely as compared with the remodeling epithelial and endothelial cellular layers and the infiltrating immune cells. Previously, in situ imaging (Petroll et al., 2015) after FI revealed interconnected streams of cells approaching the acellular wound. We speculate that cells at the leading edge of the migratory front may deposit tenascin to create tracks or conduits that trailing cells follow. We observed tenascin labeling along the length of migrating fibroblasts, which is consistent with this model. The formation of interconnected streams of cells may be further promoted by cell-cell adhesion mediated through claudin 5, a tight junction protein not expressed by quiescent keratocytes.

Changes in proteoglycan expression also distinguished fibroblasts from quiescent keratocytes. During the wound response proteoglycans involved in transparency and collagen fibril maintenance continue to be expressed, such as keratocan (KERA), decorin (DCN), lumican (LUM), and mimecan (osteoglycin, OGN). This suggests that keratocyte functions related to the maintenance of stromal ECM may be conserved throughout fibroblastic transformation. However, unlike quiescent keratocytes, fibroblasts exhibit marked upregulation of proteoglycans associated with development and ECM assembly, such as biglycan (BGN), asporin (ASPN), and versican (VCAN). These molecules may facilitate fibroblast migration in tracks between the collagen lamellae.

The expression of the corneal crystalline ALDH1A1 was decreased in fibroblasts as compared to keratocytes, consistent with their hyper-reflective cell bodies evident in vivo (Jester et al., 1999a). Genes encoding the proteases MMP2 and MMP14 were also upregulated, and these modulate extracellular components including collagen types I, V, and fibronectin (Zhou and Petroll, 2014). Despite multiple cytoskeleton proteins and proteases suggesting interactions with fibronectin, we did not see a significant increase in extracellular fibronectin expression compared with normal keratocytes. Similar alterations in crystallin, proteoglycan, and enzyme expression are observed in cultured corneal fibroblasts (Kumar et al., 2024; Poole et al., 2025).

Transcriptional changes in genes associated with cytoskeletal organization and cellular force generation were also observed during the active phase of repopulation. Fibroblast clusters showed elevated expression of myosin genes (MYL9, MYH10) and cadherin family members such as CDH11, consistent with increased intracellular tension and cell-cell mechanical communication during migration. These components are known to support actomyosin contractility and adherens junction formation, key features of collective cell migration (Kozole and Beningo, 2024; Theveneau and Mayor, 2012). Interestingly, while these fibroblasts showed upregulations of genes typically associated with traction force generation, they did not express ACTA2 or other canonical myofibroblast markers, highlighting the distinction between transient migratory activation and pathological contractile differentiation. This intermediate phenotype may allow for sufficient mechanical engagement with the ECM to support migration without promoting fibrosis or disorganized matrix deposition.

4.3. Resolution of wound healing

By Day 28, fibroblast-associated clusters were no longer distinguishable, and keratocyte clusters exhibited gene expression profiles more similar to controls. Genes linked to matrix remodeling, cellular force generation, and proliferation were lower as compared to Day 7. Very few cells exhibited sustained expression of TNC, ASPN, and BGN. The absence of persistent fibroblasts without further induction of myofibroblast populations, combined with the restoration of near-normal inherent backscatter as observed via confocal imaging, supports a model of a transient fibroblast activation during wound repopulation.

4.4. Limitations

One potential limitation of the current study is the small sample size used for scRNA-seq analyses. However, we saw consistent scRNA-seq results in the individual samples analyzed for each condition (Control, Day 7 FI and Day 28 FI). For example, both of the day 7 FI scRNA-seq samples had fibroblast 1 and fibroblast 2 clusters in addition to the normal keratocyte clusters, and each sample had similar differential gene expression patterns. Another limitation is that experiments were only performed on female rabbits. Previous studies establishing the FI wound healing response were performed on both male and female rabbits, and no differences were observed. Finally, while scRNA-seq clusters cells with similar transcriptional profiles, the location of these clusters within the tissue is not known. In the current study, tenascin C and f-actin labeling of tissue sections suggest that the fibroblasts clusters identified at day 7 are from the central wound area where intrastromal migration and cell proliferation occur. However, spatial transcriptomics could provide further validation of these results. It would be particularly interesting to determine whether the quiescent keratocyte subclusters are spatially regulated, due to differences in lamellar organization, cytokine exposure, or mechanical tension within the tissue.

5. Conclusions

Collectively, these findings provide a high-resolution view of the molecular events underlying non-fibrotic corneal stroma healing. In the FI model, keratocytes can undergo transient phenotypic changes to support migration, matrix remodeling, and proliferation, without progressing to a fibrotic or contractile myofibroblast phenotype. This migratory mechanism is reflected in the expression of genes encoding cytoskeletal regulators, contractile machinery, proteoglycans, ECM components, and adhesion molecules that collectively facilitate orderly repopulation. Understanding how these programs are regulated in vivo could provide a foundation for future wound healing studies aimed at promoting wound repopulation while avoiding scarring following surgical, traumatic, or inflammatory injury to the cornea. This work also lays the foundation for future studies using injury models with fibrotic, regenerative and/or remodeling phases of healing.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.exer.2025.110816.

Data availability

Data will submitted to a shared data repository upon publication.