CELLULAR AND MOLECULAR PROFILES OF LARVAL AND ADULT XENOPUS CORNEAL EPITHELIA RESOLVED AT THE SINGLE-CELL LEVEL

Abstract

Corneal Epithelial Stem Cells (CESCs) and their proliferative progeny, the Transit Amplifying Cells (TACs), are responsible for homeostasis and maintaining corneal transparency. Owing to our limited knowledge of cell fates and gene activity within the cornea, the search for unique markers to identify and isolate these cells remains crucial for ocular surface reconstruction. We performed single-cell RNA sequencing of corneal cells from larval and adult stages of Xenopus. Our results indicate that as the cornea develops and matures, there is an increase in cellular diversity which is accompanied by a substantial shift in transcriptional profile, gene regulatory network and cell-cell communication dynamics. Our data also reveals several novel genes expressed in corneal cells and changes in gene expression during corneal differentiation at both developmental time-points. Importantly, we identify specific basal cell clusters in both the larval and adult cornea that comprise a relatively undifferentiated cell type and express distinct stem cell markers, which we propose are the putative larval and adult CESCs, respectively. This study offers a detailed atlas of single-cell transcriptomes in the frog cornea. In the future, this work will be useful to elucidate the function of novel genes in corneal epithelial homeostasis, wound healing and regeneration.

1. Introduction

The corneal epithelium is the outermost layer of the eye, and is essential for focusing light, maintaining normal vision and protecting deeper eye tissues. As it is exposed to the external environment, the corneal epithelial tissue has a remarkable capacity for self-renewal and repair, which is attributed to a population of somatic stem cells, called Corneal Epithelial Stem Cells (CESCs), and their proliferative progeny, the Transit Amplifying Cells (TACs). Therefore, the self-renewing corneal epithelium serves as an outstanding model to study the biology of epithelial stem cells, and their contributions to a highly specialized differentiated tissue.

The cellular composition and homeostatic maintenance of the corneal epithelium have been studied extensively (Cotsarelis et al., 1989; Lavker et al., 2004; Schermer et al., 1986; Thoft and Friend, 1983). In adult vertebrates, the stratified corneal epithelium consists of an innermost layer of undifferentiated basal cells, suprabasal cells, and multiple layers of differentiated, superficial (apical) squamous cells. Some studies in humans and mice (Amitai-Lange et al., 2015; Davanger and Evensen, 1971; Di Girolamo, 2011; Zhao et al., 2009) have indicated that CESCs reside in the basal epithelium (in a specialized niche located in the peripheral “limbus”). In contrast, some other studies suggest that CESCs are distributed throughout the basal corneal epithelial layer (Chang et al., 2008; Majo et al., 2008). Despite some ambiguity about their precise location, it is known that these cells divide either symmetrically for stem cell renewal, or asymmetrically to produce more highly proliferative TACs that migrate centripetally to populate the corneal epithelium (Kinoshita et al., 1981; Tseng, 1989). As these TACs divide and move superficially, their progeny progressively become more differentiated and give rise to post-mitotic, terminally differentiated cells (TDCs), which eventually undergo senescence and are sloughed off from the ocular surface. Despite the long known existence of CESCs and TACs among vertebrates, we are limited in our understanding of the comprehensive genetic profiles of these cornea cells. This has largely affected the search for definitive biomarkers to identify and isolate corneal stem cells that can be used for clinical applications.

Several studies on transcriptional profiling of corneal epithelial cells isolated from humans, rats and mice have relied on bulk-sampling techniques and cell enrichment using pre-defined markers (Adachi et al., 2006; Bath et al., 2013; Ma and Lwigale, 2019; Sartaj et al., 2017; Zhou et al., 2006). As these studies were restricted to specific subpopulations or regions of the corneal epithelium, it has been difficult to compare findings across these studies to systematically analyze corneal epithelial heterogeneity. Therefore, to facilitate the search for CESC biomarkers, a comprehensive and unbiased characterization of the transcriptome of many individual corneal epithelial cells representing progressive stages of vertebrate corneal maturation is required.

The South African clawed frog, Xenopus laevis, has been a valuable research model, particularly for studying eye tissue development, repair and regeneration, which includes both the cornea and the lens (Barbosa-Sabanero et al., 2012; Henry et al., 2008; Kha et al., 2018). The anatomy and development of the frog cornea are nearly identical to that of humans (Hu et al., 2013) making it an excellent, low-cost vertebrate model to study cornea biology and determine the dynamics of CESCs. Early during development (e.g., larval stages 48-51) (Nieuwkoop and Faber, 1956), the corneal epithelium consists of a distinct apical layer of differentiated cells, a single basal layer, and relatively few keratocytes underlying this epithelium. As the larva undergoes metamorphosis and matures into an adult (stage 66 and beyond), the cornea undergoes dramatic changes; the stromal space thickens and is filled with collagen lamellae and keratocytes, new epithelial layers are added, and limbal crypt-like palisades are observed (similar to those seen in mammals) (Goldberg and Bron, 1982; Hamilton and Henry, 2016). The presence of oligopotent epithelial stem cells in the frog cornea has been demonstrated in previous studies. For example, cells undergoing S phase DNA replication and mitosis (examined by EdU labeling and anti-phospho-Histone H3 (S10)) are scattered throughout the basal corneal epithelium of tadpoles (Perry et al., 2013; Thomas and Henry, 2014). Whereas in adult frogs, long-term BrdU or EdU label-retaining cells, that likely identify CESCs, become concentrated in the limbal area (Hamilton and Henry, 2016). Furthermore, molecular characterization of corneal epithelia of larval and adult frogs reveals that putative stem cell markers are expressed at different stages in different epithelial layers (Sonam et al., 2019). Although these markers label different cellular layers and regions of the corneal epithelium during maturation, none was found to be unique to CESCs or TACs. This largely agrees with similar conclusions reached by immunolocalization studies using corneas of other vertebrates (Kammergruber et al., 2019; Morita et al., 2015; Mort et al., 2012; Schlotzer-Schrehardt and Kruse, 2005). However, these studies provide only a glimpse into the genes expressed by these cells. Single-cell resolution molecular data is needed to provide a comprehensive, unbiased picture of tissue heterogeneity and to better understand cellular states and transcriptional profiles during corneal development and differentiation.

Single-cell RNA sequencing (scRNA-seq) is a revolutionary technique that has enabled the identification of cell subpopulations, analysis of rare cell types, reconstruction of cell lineage trajectories and the building of gene regulatory networks for tissues and organs including the skin, heart, kidney, uterine epithelium and intestinal epithelium (Combes et al., 2019; Griffiths et al., 2018; Haber et al., 2017; Haensel et al., 2020; Lescroart et al., 2018; Wu et al., 2017). It has provided key insights into the biology and transcriptional states of different types of stem cells, including hematopoietic progenitors, neural progenitors, embryonic stem cells, muscle stem cells and others (Dell'Orso et al., 2019; Klein et al., 2015; Kumar et al., 2017; Llorens-Bobadilla et al., 2015; Zhou et al., 2016). Recent scRNA-seq studies have even provided insights into the molecular and cellular heterogeneity of eye tissues, including human and mouse retinas (Macosko et al., 2015; Voigt et al., 2019), as well as human and mouse corneas (Kaplan et al., 2019; Li et al., 2021a; Li et al., 2021b). However, these studies focused on adult eye tissues, making it difficult to understand the developmental progression of gene expression changes at the single-cell level as the ocular tissue undergoes development and maturation.

Here, we use scRNA-seq to expand our understanding of the vertebrate corneal epithelium in an amphibian model at two distinct stages of development. By examining 22,481 cells, we identified eight distinct cell clusters in larvae and thirteen clusters in adult frog corneas, each having discrete transcriptional signatures. In search of novel biomarkers for corneal stem cells, we identified the distinctly enriched expression of marker genes in subsets of basal corneal cells – e.g., tspan1, gpha2 in larvae, and tgm2 in adults. Furthermore, the reconstruction of pseudotemporal trajectories of epithelial cells allowed us to better understand the gene expression changes that take place during the process of corneal differentiation in larval and adult frogs. In addition, we identified key gene regulatory networks of the corneal basal epithelium that are both developmentally conserved and unique, as well as examined the cell-cell communication dynamics that characterize the two developmental time-points. Together, this is the first study to uncover the transcriptional profile of the amphibian cornea, and this data may be extrapolated to understand the biology of cornea development in other vertebrates.

2. Materials and Methods

Animals

Adult Xenopus laevis were obtained from Nasco (Fort Atkinson, WI). Fertilized eggs were collected, and larvae were raised to Stage 49-51 based on developmental staging by Nieuwkoop and Faber (1956). All animal care and experiments performed in this study were approved and monitored by the Institutional Animal Care and Use Committee (IACUC) and the Division of Animal Resources (DAR) at the University of Illinois.

Single-cell dissociation

Larvae:

Tadpoles were anesthetized at room temperature (22°C) by incubation for 1-2 min in a 1:2000 dilution of MS222 (ethyl 3-aminobenzoate methanesulfonate, Sigma, St. Louis, MO) diluted in 1/20X Normal Amphibian Media, NAM (Slack, 1984), while adult frogs were incubated for 30-35 mins in 2g/L MS222. Anesthetized tadpoles were transferred to clay-lined dishes containing Ca⁺⁺- Mg⁺⁺ free Dulbecco’s Phosphate Buffered Saline (1X DPBS; Corning, NY, #21-031-CV). The corneal epithelial tissue was dissected using fine microscissors (Henry et al., 2018), taking care not to include any surrounding skin. The skin is demarcated by the outer edge of the orbit, a lack of transparency, and the obvious presence of pigmented melanophores. To limit the amount of time needed to collect these tissues, three operators simultaneously collected a total of 80 corneas in three separate glass depression dishes containing 400 μL 1X DPBS, on ice. Care was taken to ensure that corneas were carefully removed from the tip of the forceps and remained submerged in the 1X DPBS solution. Using a 200 μL pipette tip, most of the DPBS was gently removed without disturbing the corneas at the bottom of each glass dish. To dissociate the corneal epithelial tissue, ~350 μL of Accutase (CellnTec Advanced Cell Systems, Bern, Switzerland) was added to each dish, and the dishes were covered with plastic petri dish lids (35 x 10 mm, Corning, #351008), flat side down, and wrapped tightly with lab tape to secure them and prevent desiccation. Next, the dishes were incubated at 37°C for 35 min with intermittent trituration using a 200 μL pipette tip. To stop the reaction, 200 μL of media containing 61% Leibovitz's L-15 (Invitrogen, Carlsbad, CA), 0.5 μM EDTA (Invitrogen) and 10% Fetal Bovine Serum (Invitrogen) was added to each dish. For mechanical dissociation, the suspension was finally triturated 8-10 times using a 200 μL pipette tip. The three cell suspensions were pooled in a 2mL LoBind microcentrifuge tube (Eppendorf, Enfield, CT; #022431048), spun at 325g for 5 min in a fixed angle centrifuge, and resuspended in 200 μL L-15 with 10% FBS and 0.5 μM EDTA. Finally, the cells were passed through a 40 μm FlowMi cell strainer (Bel-Art, Wayne, NJ; #136800045) and collected in a 2mL LoBind tube. Cell viability was assessed using Acridine Orange/Propidium Iodide (AO/PI) staining solution (Nexcelom Bioscience, Lawrence, MA, #CS2-0106) that labels live and dead cells respectively, and counted using an automated cell counter (Cellometer K2, Nexcelom Bioscience).

Adults:

Similarly, for experiments with adult animals, 7-8 mature corneas were dissected out of the animals for each sample replicate. For harvested tissues collected in a glass dish, dissociation was performed using ~600 μL of Accutase for 60 min at 37°C. While holding the tissue submerged in dissociation media, the stroma was carefully peeled from the epidermis, and the epithelial tissue was gently trimmed into small fragments using microscissors. Further dissociation was done by trituration, and the reaction was quenched by adding 61% L-15 + 0.5 μM EDTA + 10% FBS. The suspension was pooled, centrifuged at 325g for 5 min, resuspended in 400 μL 61% L-15 + 0.5 μM EDTA + 10% FBS media and strained through 40 μm FlowMi cell strainers. Finally, the cell viability (as mentioned above) was determined before proceeding with library preparation.

Library preparation and sequencing

Single-cell sequencing libraries were prepared individually from each replicate/sample using the 10X Genomics Chromium Single Cell 3' Kit v3 and v3.1 (10X Genomics, San Francisco, CA) and sequenced with Illumina NovaSeq 6000 on a SP/SP4 flow cell to obtain 150bp paired reads. Library preparation and sequencing was conducted at the High-Throughput Sequencing and Genotyping Unit of the Roy J. Carver Biotechnology Center at the University of Illinois at Urbana-Champaign.

scRNA-seq data processing

The RNA sequencing reads were processed using the CellRanger v3.1.0 pipeline by 10X Genomics. For generating the CellRanger reference, the Xenopus laevis v9.2 reference genome and gene models were downloaded from Xenbase (http://ftp.xenbase.org/pub/Genomics/JGI/Xenla9.2/XENLA_9.2_Xenbase.gff3) (Karimi et al., 2018). Further downstream processing was performed in R Studio v3.6.1. The output from CellRanger was uploaded into an R Studio session, and processed using the Seurat v3.1 pipeline (Butler et al., 2018; Satija et al., 2015), wherein a standard filtering and preprocessing pipeline was employed. Batch-effects across samples were removed through a harmony algorithm using sample source (2 time points X 2 replicates -> 4 samples) as the different batches (Korsunsky et al., 2019). Genes expressed in less than 3 cells, and cells expressing < 200 or > 3,000 genes were removed from downstream analysis. We used a 5% cutoff for mitochondrial reads to remove dead/apoptotic cells. Post-normalization, we scaled data using the ScaleData function and performed this for all features/genes. We used FindNeighbours and FindClusters functions at parameter values of dims 1:20 and resolution 0.2, respectively. Significantly enriched genes within clusters were identified using the FindMarkers functionality in Seurat, that uses the MAST algorithm and the ratio of number of genes expressed/detected in a cell as a variable to regress. Cell type identities were assigned using expression patterns of previously established marker genes from the available literature (Castro-Muñozledo, 2015; Guo et al., 2018) and our previous study (Sonam et al., 2019). The frog X. laevis is allotetraploid, and gene expression data can be accessed for each allele from both the Large (Gene.L) and/or Short (Gene.S) chromosomes (Session et al., 2016). Taking into consideration that these alleles may have different functions, we reported cluster-specific allele expression patterns for each gene.

Trajectory Analysis

To infer the trajectory of single cells, we used Monocle2 (Trapnell et al., 2014) and applied the DDRtree algorithm for dimensionality reduction. Differentially expressed genes were calculated using comparison across the different clusters found by Seurat v3.1 above. We used the Top 3,000 genes to order cells and determine the pseudo-time trajectory. Relevant functions within Monocle2 were used to plot a heatmap for genes varying along the pseudo timeline. To define Gene Ontology (GO) for the regulated genes along the trajectory, we humanized the X. laevis gene names and converted them to human Ensembl GeneIDs, and used DAVID 6.8 for GO analysis (Dennis et al., 2003). GO option GOTERM_BP_ALL was selected, and the terms with a P-value < 0.05 were chosen as significant categories.

Cell Cycle Analysis

For cell cycle analysis, we used a previously defined core set of 84 G1/S and 98 G2/M cell cycle genes (Aztekin et al., 2019). We identified the approximate cell cycle state of each cell with the average expression (called the “cycle score”) for the two gene sets using the Cell cycle scoring function in Seurat v3.1, and used default classification by Seurat.

Gene Regulatory Network (GRN) analysis

We used the single-cell regulatory network inference and clustering (SCENIC) pipeline (Aibar et al., 2017), to estimate the AUCell Score activity matrix from the scRNA-seq dataset (Chembazhi et al., 2021). Briefly, since current relevant databases do not hold gene regulatory information specific to X. laevis, we collapsed raw counts from .L and .S gene forms in our data. This collapsed data was processed in the Seurat v3.1 pipeline, as explained above. A normalized count table was extracted from the Seurat object, and Xenopus genes with a known human ortholog were retained for downstream analysis, and the gene names were humanized. Unlike the standard SCENIC workflow where this AUCell score activity matrix is binarized by thresholding to generate a binary regulon-activity matrix, we retained the full AUCell score for all further analysis. The matrix of AUCell score containing regulons (rows) x cells (columns) was used as input to Seurat object and used to cluster and identify important regulons for each previously identified cluster.

Cell-cell communication analysis

To determine the cell-cell communication network, we first collapsed raw counts from .L and .S gene forms, and the Xenopus genes with a known mouse ortholog were retained for further analysis. To systematically construct the cell-cell communication networks and perform statistics of interactions, we utilized methods described in detail by Farbehi et al. (2019). Briefly, we used a directed and weighted network with four layers of nodes: source cell populations expressing the ligands, the ligands that are expressed by the source populations, the receptors directed by the ligands, and the target cell populations. Weights of edges that link ‘source to ligand’ and ‘receptors to target’ were calculated as log₂ (fold change) in expression of ligand/receptor in source/target in comparison to other cells. We determined the ligand-receptor interactions using a mouse-specific ligand-receptor interaction dataset published earlier by Farbehi et al. (2019). The sum of weights along the path was used to determine path weights connecting a source to target through a ligand:receptor interaction. We chose all ligand:receptor connections with a minimum path weight of 1.5 and calculated the overall weight, w_s:t, as the sum of all path weights between the corresponding source and target. Only edges with Benjamini-Hochberg adjusted P-values, P_w<0.01 were regarded as significant. Next, we constructed ligand:receptor interaction dot plots using ggplot2.

Immunofluorescence staining and imaging

Immunostaining of larval corneas was performed using a previously published protocol (Sonam et al., 2019). Briefly, the larval specimens were fixed in Dent’s fixative (80% methanol, 20% dimethyl sulfoxide, DMSO), or 3.7% Paraformaldehyde (PFA). Next, the eyes were excised from each sample, taking care to keep the corneas intact. The excised eyes were washed six times for 10 min each in 1X PBST (Phosphate buffered saline; 1.86 mM NaH₂PO₄, 8.41 mM Na₂HPO₄, 175 mM NaCl, pH7.4, containing 0.5% Tween-20). Blocking was done using 5% Bovine Serum Albumin/10% normal goat serum/1X PBST for 2 h at room temperature. Finally, the eyes were incubated with primary antibodies diluted in the blocking solution overnight at 4°C. Following primary incubation, six more washes were conducted with 1X PBST prior to incubating the eyes with secondary antibodies. Goat anti-mouse Alexa-Fluor 488 or 546, and goat anti-rabbit Alexa-Fluor 488 or 546 (Invitrogen, Rockford, IL) were used as secondary antibodies at a concentration of 1:300 diluted in blocking solution for 2 h at room temperature. Finally, eyes were washed twice with 1X PBST and incubated with 1μg/mL solution of Hoechst 33342 (Molecular Probes, Eugene, OR) in 1X PBS for nuclear counter-staining. RainX (ITW Global Brands, Houston, TX) treated slides were used for mounting the corneas, and SlowFade antifade reagent (Life Technologies, Eugene, OR) was used for the mounting media. To flat mount corneas on slides, labeled corneas were carefully removed from the whole eyes, and small, peripheral, radial incisions were made to facilitate flattening during the mounting process. Confocal imaging of immunolabeled larval corneas was performed using an inverted LSM 700 microscope (Carl Zeiss, Munich, Germany). Z-stacks were processed with Zen software (Carl Zeiss, Munich, Germany), and images were compiled using Adobe Photoshop.

3. Results

Single-cell RNA sequencing identifies cellular diversity in developing Xenopus cornea

To study cellular and transcriptional heterogeneity during amphibian corneal development, we isolated corneal cells from frog larvae and adults. We developed an Accutase enzyme-based digestion method for dissociating viable single cells from corneal tissue (Fig. 1A). Using the 10X Genomics-based scRNA-seq platform, we sequenced 24,000 cells in total, with ~139,000 mean reads per cell, ~1,700 median genes per cell, and a median of ~6,900 unique transcript molecules per cell. After quality assessment and filtering, a total of 22,481 high-quality cells were retained for downstream computational analysis. To allow cross-sample comparisons, we merged datasets from all samples and corrected the batch effects using the harmony algorithm (Fig. 1B and Fig. S1).

Figure 1. Overview depicting workflow for isolation of corneal cells from Xenopus for single-cell RNA sequencing (scRNA-seq).

(A) Isolation of corneal epithelial cells from Xenopus larvae and adults. Corneal epithelia were dissected, pooled and dissociated to isolate single cells. After determining high cell viability, a single-cell library was prepared using the 10X Chromium Single Cell 3' Reagent Kit (V3). (B) Computational workflow demonstrating data processing and analysis pipeline for scRNA-seq data. Cell Ranger was used to align raw reads and generate feature-barcode matrices. Seurat v3.1 was used to perform basic quality check (QC) and normalization, followed by use of Harmony to remove batch-specific effects.

Next, an unbiased, graph-based clustering algorithm of the Seurat software was used to group cells according to their gene expression profiles. We identified eight and thirteen distinct cell clusters in larval and adult frog corneas, respectively, and these clusters were visualized in two-dimensional space using Uniform Manifold Approximation and Projection (UMAP) (Becht et al., 2018) (Fig. 2A). Furthermore, we projected harmony-corrected data using tSNE (van der Maaten and Hinton, 2008) and PHATE embedding (Moon et al., 2019), and found cells annotated as a cell type to cluster together in these projections (Fig. S2).

Figure 2. Identification of cell types in scRNA-seq dataset of corneas from Xenopus larvae and adults.

(A) Unbiased clustering of 10,659 high-quality cells from larval corneas after QC cutoffs, visualized by Uniform Manifold Approximation and Projection (UMAP). (B) Clustering of 12,155 filtered cells from adult frog corneas, visualized by UMAP. Each dot represents a single cell, and cells from the same cluster are similarly colored. Color codes in A, B represent unique clusters in each data set, and are not homologous. (C, D) Dot plot showing expression of known cell-type marker genes for each cluster. Dot diameter depicts the percentage of cluster cells expressing that marker and intensity encodes average expression of a gene among cells within that cluster. Abbreviations: Ap-L: Apical Larval; Ba-L: Basal Larval; Im-L: Immunocytes Larval; Ke-L: Keratocytes Larval; Un-L: Unknown Larval, and Ap-A: Apical Adult; Ba-A: Basal Adult; En-A: Endothelial Adult; Im-A: Immunocytes Adult; Ke-A: Keratocytes Adult; Ne-A: Neuronal Adult.

Cell-type identity was assigned based on a wide panel of cell-type-specific marker genes previously reported in various vertebrates, including Xenopus laevis (Castro-Muñozledo, 2015; Chen et al., 2004; Davies et al., 2009; Sonam et al., 2019; Yam et al., 2020; Yi et al., 2000). This included marker genes tp63.L, tp63.S, krt19.L, krt19.S, krt15.2.L, itgb1.L, itgb1.S and itgb4.L for basal epithelial cells; tjp3.L, tjp3.S, ocln.L, ocln.S, gjb1.L and gjb2.S for differentiated apical epithelial cells; nfatc1.L, nfatc1.S, gpha2.L for stem cells; col8a2.L, col8a2.S, pitx2.L, pitx2.S, acta2.L, acta2.S for endothelial cells; vim.L, vim.S, cd34.L, cd34.S, fn1.S for stromal cells/keratocytes; ccl5.L, ccl5.S, cxcr6.L, spi1.L, spi1.S, lcp1.L, lcp1.S, mmp7.L for immunocytes; tac1.L, tac1.S, npy.L, atoh1.L, atoh1.S, prox1.L, prox1.S for neuronal cells, and hba1.L, hba1.S, hbg1.L, lmo2.L, lmo2.S for blood cells.

Based on the expression pattern of these known markers, we were able to assign clusters (C) 0 as apical epithelial cells (Ap-L), C1/C2/C6/C7 as basal epithelial cells (Ba_I-L, Ba_II-L, Ba_III-L and Ba_IV-L, respectively), C3 as stromal keratocytes (Ke-L), C4 as unknown (Un-L), and C5 as immunocytes (Im-L) in the larval cornea. In adult corneas, C1/C3 are apical cells (Ap_I-A, Ap_II-A), C2/C4/C7 are basal cells (Ba_I-A, Ba_II-A and Ba_III-A, respectively), C8 are endothelial cells (En-A), C5/C9/C11 are immune cells (Im_I-A, Im_II-A, Im_III-A), C0 are stromal keratocytes (Ke-A), C12 are neuronal cells (Ne-A), C10 are blood cells (Bl-A), while identity of cluster C6 was unknown (Un-A). Given that we excluded the loose endothelial layer during cornea dissections in larvae, we did not identify a discrete cluster representative of the endothelial cell population. Despite our attempt to peel off the stroma and endothelial layer located beneath the adult cornea epithelium, we detected cell clusters expressing markers of stromal keratocytes and endothelium (C0 and C8, respectively).

This cluster analysis demonstrated that cellular heterogeneity increases as the frog cornea undergoes development and maturation from larval to adult stages. In addition, very low to no expression of mlana.L, tyr.S, mitf.L, mitf.S, tyrp1.L, dct.L, helped us rule out the presence of any skin melanophores in our data. Likewise, we did not detect the presence of cryaa.S, cryab.L, cryab.S marker genes, that confirmed the absence of any contaminating lens cells (Brahma and McDevitt, 1974; Henry et al., 2002) (Fig. S3).

Proliferative cells characterize the corneal epithelium in frogs

To distinguish the presence of transit amplifying cells (TACs) in our dataset, we looked for the expression of proliferation marker genes. However, these genes including mki67.L, mki67.S, pcna.L, and pcna.S, showed distributed expression across all clusters at both developmental time-points (Fig. 3A, B). Cell cycle analysis corroborated these findings (Fig. 3C, D), demonstrating that a considerable percentage of cells exist in the S and/or G2/M phases. To test the presence of proliferative cells in the bi-layered larval epithelium, we immunostained the cornea for the ubiquitous cell cycle marker PCNA. Using a Xenopus specific antibody (Agathocleous et al., 2009), we detected PCNA labeled nuclei in both basal and apical epithelia (Fig. 3E, F) of larval cornea. The presence of proliferative cells in both epithelial layers differs from previous Xenopus corneal studies (Hu et al., 2013; Perry et al., 2013), that reported the presence of dividing cells primarily in the larval basal epithelium. The proliferative cells in the corneal epithelium appear to be those involved in continuous growth and renewal of the larval cornea. To support our observation, we immunostained the corneal tissue with an anti-phospho-histone-H3 (S10) antibody and detected the presence of labeled nuclei in both epithelial layers (Fig. 3G, H). As phosphorylated Histone H3 (S10) is a marker of mitotic cells (Hans and Dimitrov, 2001), the positive staining further confirms cell division activity throughout this bi-layered epithelium.

Figure 3. Proliferative cells are distributed throughout the corneal epithelium in Xenopus.

(A, B) Gene expression UMAP overlay with proliferation genes mki67.L, mki67.S, pcna.L, and pcna.S in larval (A) and adult (B) corneas. Color intensity correlates with the relative transcript level for the gene. (C, D) UMAP plot showing the cell cycle status of each cell in (C) larval, and (D) adult clusters, determined using the “CellCycleScoring” module in Seurat v3.1. Color key indicates the cell cycle state. (E-F) Confocal images showing immunofluorescent staining for Proliferating Cell Nuclear Antigen (PCNA) (green) in the tadpole corneal epithelium. (E’-F’) Merged images for E-F with Hoechst labeled nuclei (magenta). (E, E’) Nuclear staining for PCNA is detected in some apical epithelial cells. (F, F’) PCNA labeled nuclei are detected in the basal epithelium. (G-H) Confocal images showing anti-phospho-Histone H3 (H3S10P) (green) in the tadpole corneal epithelium. (G’-H’) Merged images for G-H with Hoechst labeled nuclei (magenta). (G, G’) Mitotic nuclei (H3S10P) are present in a few apical epithelial cells. (H, H’) H3S10P labeled nuclei are detected in the basal epithelium. Scale bar in H’ equals 50 μm for E-H’.

In addition, the presence of proliferative cells along the entire length of sectioned corneal epithelium in adult frogs was previously shown by EdU labeling studies carried out by Hu and colleagues (2013). However, we were unable to successfully immunostain the adult corneal cross-sections to examine the presence of PCNA-positive or phospho-histone-H3 (S10)-positive cells in the multilayered adult epithelium. Altogether, our observations indicate that unlike recent scRNA-seq corneal studies in humans and mice (Kaplan et al., 2019; Li et al., 2021a; Li et al., 2021b; Ligocki et al., 2021), a single, discrete proliferative cell cluster (TAC population) could not be clearly identified in frog corneas.

Basal epithelial cells subcluster into distinct subtypes with unique transcriptional signatures, providing novel markers to identify corneal cell types

In organisms where detailed reference datasets are limited for select tissues and developmental stages, a notable advantage of scRNA-seq analysis is the unbiased identification of new marker genes for various cell types. Genes with Differential Expression (DE) in a given cluster, compared to all other clusters (average log fold change > 0.25; adjusted P-value <0.01), were identified using Seurat’s “FindMarkers” function. Next, we filtered DE genes using the difference between pct.1 and pct.2, along with a higher overall expression fold change value. For our analysis, we focused on genes expressed in more than 50% of the cells within that cluster. Fig. 4A, B depicts differentially expressed genes in individual corneal cell clusters of larval and adult frogs (the top 5 differentially expressed genes from each cluster is visualized by a heatmap in Fig. S4).

Figure 4. Identification of new marker genes for different cell types at two developmental time-points.

(A) Violin plots showing the cluster-specific expression of the top-ranking candidate marker genes for each identified cell type in larvae. (B) Violin plots with cluster-specific expression of the top-ranking candidate marker genes for each identified cell type in the adult frog. The highlighted (grey) marker genes are differentially enriched in basal cell clusters of larvae and adults.

As the basal epithelium is believed to harbor the corneal stem cells and highly proliferative TAC population in vertebrates (Gonzalez et al., 2017), we focused our investigation on the heterogeneity in basal corneal clusters. A closer look at the tadpole corneal basal cells revealed that some marker genes, including col14a1.L, col11a1.L, col17a1.L, col5a3.L, itgb4.L and krt19.L, were highly expressed and shared among the basal clusters, while other genes, including cnfn. 1.L, arhgdib.L and tspan1.S, were exclusively expressed in individual basal clusters (Fig. 4A). Many of these biomarker genes are known to play important roles in stem cells of individual organs and tissues in different organisms (Deng et al., 2021; Michel et al., 1996; Shcherbina et al., 2020; Zeng et al., 2018), and can be further investigated for their potential function as CESC markers at early stages of frog development. For instance, the gene Col17a1 encodes a hemidesmosomal transmembrane collagen protein that is expressed in basal cells of the epidermis. A higher level of Col17a1 expression is considered to be a marker of long-term epidermal stem cells and is crucial for skin homeostasis (Liu et al., 2019).

Among the four clusters annotated as basal epithelial cells in our larval dataset, cluster Ba_IV-L comprises a small percentage of basal corneal cells (~1.3% in our study) (Fig. S5A). Notably, cluster Ba_IV-L was also differentially enriched in tspan1.S and gpha2.L genes. Gpha2 (glycoprotein hormone subunit alpha 2) has been recently identified as a novel marker for limbal stem cells in both mouse and humans (Altshuler et al., 2021; Collin et al., 2021). Likewise, the cell-surface protein-coding gene, tspan1 (tetraspanin 1) was shown to be a marker of adult pluripotent stem cells (neoblasts) that underlie regeneration in planarians (Zeng et al., 2018). In addition, another member of tetraspanin family, TSPAN7, was identified as a novel marker of limbal stem cells in humans (Li et al., 2021a). Cumulatively, these observations suggest that cluster Ba_IV-L represents a putative CESC population in larval frogs.

Likewise, we investigated the three basal clusters in adult frog corneas. Among them, cells of cluster Ba_III-A constitute approximately 2.9% of cells in total (Fig. S5B). While cluster Ba_III-A expresses many of the stem-cell related marker genes such as collagens (col14a1, col17a1), integrins (itgb4, itga6), and keratins (krt19), cluster Ba_I-A was differentially enriched for adult-type keratins (krt12.6.L, krt78.2.S, krt57.L) (Suzuki et al., 2017). Based on this, we postulate that Ba_III-A identifies the “least differentiated basal cells” population, and potentially characterizes the putative CESC population in mature frogs. On the other hand, cluster Ba_I-A is most advanced developmentally and represents the “most differentiated basal cells”. As one of the most enriched and novel marker genes, we detected the transcript of tgm2 (transglutaminase 2) in cluster Ba_III-A. This is a protein-coding gene related to the maintenance of stemness (Kang et al., 2018), and an increased expression of tgm2 points towards its possible role as a unique marker and key regulator of the CESC population in adult frogs. A third group of basal cells, namely cluster Ba_II-A, co-express genes of the later stage (e.g., krt12.6.S, cldn1.L, cldn1.S) while still retaining expression of stem cell-related marker genes (e.g., itgb4, col17a1, itga6), suggesting that these cells are transitioning from Ba_III-A to Ba_I-A clusters. Therefore, we posit this cluster as “transitional basal cells”.

Pseudo-temporal analysis reveals the differentiation trajectory of larval corneal epithelium and associated changes in gene expression

To investigate the differentiation trajectory of corneal epithelial cells, we leveraged one of the most powerful features of scRNA-seq datasets — single cells can be ordered in an unbiased way along paths to resolve their progressions. Taking into consideration the bi-layered organization of larval corneas, we expected our data to follow a rather linear trajectory (basal → apical) with some possible branching; therefore, we utilized the trajectory inference tools best suited for such analyses. Using Monocle2 (Trapnell et al., 2014), we performed pseudo-time-based ordering of single cells from Ap-L, Ba_I-L, Ba_II-L, Ba_III-L and Ba_IV-L clusters (Fig. 5A). The additional clusters — corneal keratocytes and immune cells — were omitted from this analysis, as they were transcriptionally disconnected and presumably lineage restricted from the epithelial cell types.

Figure 5. Pseudo-temporal ordering of epithelial cells reconstructs the larval corneal differentiation process.

(A) Pseudo-time plot indicating the cellular trajectory of all larval corneal basal and apical cells (Clusters Ap-L, Ba_I-L, Ba_II-L, Ba_III-L and Ba_IV-L). Single-cell trajectories were constructed, and pseudo-time values were calculated using Monocle 2. (B) Trajectories are colored by cluster identity that corresponds to the key in Figure 2A. (C) Heatmap depicting the classes of genes that vary along the pseudo-time plot during larval corneal differentiation. Pseudo-time is indicated by the color key similar to that shown in (A). Gene Ontology (GO) analysis was performed using DAVID and the enriched GO terms (P < 0.05) in each class are listed. Relative expression is indicated by the color key. (D) Gene expression kinetics along the pseudo-time progression of representative genes belonging to different pathways and processes, as indicated. Genes shown belong to the Retinoic Acid pathway (RA pathway), the Fibroblast Growth Factor pathway (FGF pathway), the Wnt pathway, Transforming Growth Factor (TGF) beta pathway, Epithelial- to-Mesenchymal Transition (EMT), Extracellular Matrix (ECM) deposition, and the Thyroid Hormone pathway (TH pathway).

In the pseudo-temporal trajectory derived for these epithelial cells, we found that while Ba_IV-L cells were located at the beginning, the cluster of apical cells (Ap-L) was concentrated primarily towards the trajectory terminus. Cells of Ba_I-L and Ba-III-L reside mostly on the left arm of the trajectory, while Ba_II-L cluster cells are distributed throughout the trajectory (see Fig. 5B), and likely represent the population of basal cells that are undergoing a transition to the apical layer. This ordering of cells along the axis originating from basal cells towards the terminally differentiated apical state is in agreement with the known corneal stratification in tadpoles (Hu et al., 2013; Sonam et al., 2019).

Next, we identified expression dynamics of the top 2500 genes that change as a function of progress through pseudo-time. These genes can be categorized into three distinct gene classes (namely Class I, II and III; Fig. 5C). To better understand the biological significance of these classes, we performed Gene Ontology (GO) analysis. Genes whose expression peaks late along this trajectory (Class I) are enriched in GO terms associated with cell adhesion, cell migration, keratan sulfate biosynthesis, inflammatory response and apoptotic process. On the other hand, genes predominantly expressed at the beginning of the developmental trajectory show an abundance of GO terms associated with ECM organization, integrin signaling and skin development (Class III). In addition, the smallest class (Class II) with genes that peak transiently along the pseudo-time was enriched for translational initiation and cell adhesion.

The pseudo-time trajectory can also be used to gain insight into the temporal expression changes of individual genes. As signaling mechanisms (such as Retinoic Acid, FGF and Wnt/β-catenin), processes involved in ECM deposition and EMT-related processes are crucial for corneal formation and differentiation (Dhouailly et al., 2014; Nakatsu et al., 2011), we closely investigated the expression changes of genes implicated in these pathways/processes along the pseudo-time (Fig. 5D). Retinoic acid signaling genes (cyb26b1, aldh1a1, rara), show an overall increasing trend along this pseudo-time, while genes in the FGF (fgfbp2, fgfr2, fgfr1), Wnt/β-catenin (wnt3a, wnt4 and others), and TGF/β pathway (dcn, id3), show a reduction in expression. We also noted a reduction in expression of genes that are involved in Epithelial-to-Mesenchymal (EMT) transition process (pcdh7, snai), and those constituting the corneal ECM (lum, sparc, itga2, col12a1 and others), along the trajectory. Furthermore, we noted a gradual increase in expression levels of thyroid hormone-controlled genes (thdl17, thdl18 and thdl20), along the pseudo-time axis. Changes in thyroid hormone gene expression dynamics represent a novel observation in the context of Xenopus eye development, and these changes appear to be similar to those observed in frog skin (Suzuki et al., 2009; Yoshizato, 2007). Furthermore, the corneal epithelium in anurans has been described to undergo changes during the process of metamorphosis (Hu et al., 2013; Slansky et al., 1970).

Differentiation trajectory of adult corneal epithelium involves key cell path decisions

The adult cornea in frogs has higher cellular complexity compared to pre-metamorphic stages, with a well-formed multilayered, stratified corneal epithelium, stroma and endothelium. The epithelium is composed of approximately 13 cellular layers at the center and 10 layers of cells on the periphery, and contains cuboidal cells at the basal side cells and flat squamous epithelial cells at the apical side (Hu et al., 2013). To examine the developmental relationship among basal cells, we initially focused on clusters Ba_I-A, Ba_II-A and BA_III-A and performed pseudo-time ordering based on a Monocle2 algorithm. We observed one branch point in the trajectory (Fig. 6A), with cells of cluster Ba_III-A concentrated at the start of the trajectory. Cells of cluster Ba_I-A largely constituted cell path 1, while Ba_II-A cells exist along both cell paths 1 and 2 (Fig. 6B). This pseudo-temporal trajectory deciphers major cell fate decisions of basal cells along the adult corneal differentiation process. We can conclude two key points here: (i) the data supports our hypothesis that the Ba_I-A cluster is representative of a “mature basal cell state”, and (ii) the cells of cluster Ba_II-A cluster exhibit an “transitional basal cell state,” between the immature/undifferentiated basal cell type and the mature basal cell type.

Figure 6. Differentiation trajectory in adult corneal epithelium involves key cell path decisions.

(A) Pseudo-temporal ordering of all cells from the basal epithelium present in the adult cornea (Ba_I-A, Ba_II-A and Ba_III-A). The trajectory has two different paths distributed around branch point 1, namely, cell paths 1 and 2. The arrows indicate the developmental directionality of the cell clusters, based on our inference. (B) Distribution of cells from three basal clusters along the pseudo-time shown in A. While cells of Ba_I-A are present towards the end of path 1, cells of Ba_II-A are distributed throughout the trajectory, and cells of Ba_III-A are located at start of trajectory. (C) Heat map showing bifurcation of gene expression programs executed along the pseudo-time after branching. Three distinct classes of genes (with top GO terms enriched in each class) were identified. (D) Pseudo-time plot to further resolve the cellular trajectories from basal clusters (Ba_I-A and Ba_II-A) to apical clusters (Ap_I-A and Ap_II-A). The branch point 3, which is under evaluation has two different paths, namely cell paths 3 and 4, around it. (E) Pseudo-time plots showing distribution of each of the four clusters along combined cellular trajectories shown in D. (F) Heatmap of gene expression along the pseudo-time plot in D, with the top GO terms enriched in each class of genes, as indicated.

To further examine how cellular gene expression changes while traversing through branchpoint 1 into either cell path 1 or 2, we used BEAM analysis within the Monocle2 pipeline. This led to identification of ~2500 genes that could largely be distinguished into three main classes based on their behavior across the branchpoint (Fig. 6C). Class I and II contained genes that had low expression in the pre-branch state, but post-branching, they were reciprocally up- or down-regulated along cell paths 1 and 2, respectively. These classes were enriched for genes representing cell adhesion, FGFR pathway, Notch pathway, cell proliferation (Class I), and ECM organization, cell migration, integrin signaling, and Wnt signaling (Class II). Alternatively, Class III represented genes that were relatively highly expressed pre-branching followed by an increased or decreased expression in cell paths 2 and 1, respectively. These were enriched for GO terms including cell migration, cell proliferation, BMP signaling and activated MAPK signaling.

Having explored the course of progression among different basal cell types, we sought to understand the cellular transitions involved in further differentiation of basal cells into the outermost epithelial cells, including acquisition of positional cell states (i.e. basal or apical). At first, we included cells belonging to apical epithelial clusters (Ap_I-A and Ap_II-A), and basal clusters, Ba_I-A and Ba_II-A (excluding the cells of cluster Ba_III-A) for constructing the pseudo-time trajectory. The ordering analysis revealed bifurcation into two distinct cell paths (namely, cell path 3 and 4) around branch point 3 in the trajectory (see Fig. 6D). The cells of cluster Ba_II-A, mainly concentrated at the start of the trajectory, can convert into either Ba_I-A or Ap_I-A/Ap_II-A (traveling right or left around branch point 3, respectively). The right arm of the trajectory (denoted by cell path 4), indicates that a subset of “transitional basal cells” (Ba_II-A), transform to “mature basal cells” along the differentiation course. On the other hand, cells of apical clusters Ap_I-A and Ap_II-A primarily constitute cell path 3 (the trajectory’s left arm around branch point 3). This path represents the progression of “transition basal cells” to terminally differentiated cell states, which potentially involves intermediate cell states and can be experimentally tested in future studies (Fig. 6E). Furthermore, we investigated the differentiation course of Ba_I-A cluster cells. Our Monocle pseudo-time analysis (including clusters Ba_I-A, Ap_I-A and Ap_II-A), shows that the mature basal cells (cluster Ba_I-A) ultimately differentiate to become cells of the apical epithelia (see Fig. S6). To further identify signature genes between the different cell populations we performed differential gene expression analysis between the two branches (Fig. 6F). For cell path 3, the differentially expressed genes (DEGs) enriched in the GO terms of cell-cell adhesion, negative regulation of apoptotic process, inflammatory response, positive regulation of I-kappaB kinase/NF-kappaB signaling (Class VI), and mitochondrial electron transport, mRNA splicing, response to cAMP (Class V), while for cell path 4, DEGs were enriched for GO terms of extracellular matrix organization, cell proliferation, integrin signaling, and Notch signaling (Class IV).

Gene regulatory network analysis reveals key differences in the regulation of corneal basal epithelial cells during development

To characterize gene regulatory networks (GRNs) that govern the developmental trajectory of basal corneal cells in vertebrates, we used the SCENIC pipeline to create a gene regulatory framework (Aibar et al., 2017). As transcription factors (TFs) are known to be highly conserved despite the allotetraploidization event in Xenopus laevis (Watanabe et al., 2017), we converted the X. laevis gene names to their human orthologs (see Materials and Methods). SCENIC calculates the activity of TFs from individual cells by integrating co-expression data with transcription factor motif enrichment analysis to generate a “regulon.” A regulon is comprised of an expressed transcription factor and all of its co-expressed target genes. We determined the regulon activities using Area Under Cell (AUCell), which ranks targets of an individual regulon among the expressed genes, thereby generating a regulon-by-cell activity matrix.

To explore regulon activities distinguishing the cellular features of larval and adult corneal basal cells, we analyzed the AUCell score activity matrix of basal clusters obtained from the two developmental time-points (i.e., clusters Ba_I-L, Ba_II-L, Ba_III-L, Ba _IV-L, and B_I-A, Ba_II-A, Ba_III-A). Upon 2D UMAP projection — constructed from the AUCell scores — we noted that basal cells from larval and adult stages formed distinct nonoverlapping clusters, representing clear differences in their GRNs (Fig. 7A). Cells from Ba_III-L and Ba_IV-L overlapped with Ba_I-L, demonstrating that mostly similar regulons are active in a majority of larval basal cells. However, few basal cells from Ba_II-L overlapped with the Ba_I-L cluster. Interestingly, for basal cells of Ba_II-A and Ba_III-A origin, the clusters overlapped and were located at the far end of the UMAP plot, indicating their GRNs are substantially discrete from those active in larvae; however, the cells from the Ba_I-A grouped more closely to larval basal cells (Fig. 7B).

Figure 7. Distinct gene regulatory networks are active in basal epithelial cells of the larval and adult frog corneas.

(A) UMAP projection of all basal cells based on the AUC scores for each regulon calculated using SCENIC. Cells belonging to the same regulon are colored, according to the key. (B) AUC score based UMAP projection, grouped according to the origin of basal cells. Regulons of basal cells (Ba_I-L, Ba_II-L, Ba_III-L and Ba_IV-L) belonging to larval frogs group together and overlap, whereas regulons from adult basal cells (Ba_I-A, Ba_II-A and Ba_III-A) are separated from larval regulons. (C, D) Dot plot showing unique and shared regulons that are active in different basal cell states. Dot diameter depicts the percentage of cluster cells expressing that regulon and color intensity encodes average expression of a regulon among cells within that cluster, as indicated in the keys.

Furthermore, we explored the regulons characterizing each of the basal cell clusters to identify the main TFs and their target genes that drive those cell states (Fig. 7C, D). To our interest, we found that distinct developmental regulons are activated in basal cells of tadpole and adult corneas. For the tadpole basal clusters, MEIS1, NFATC1, ARID3A, XBP1, and CREB3L1 regulons were active while these were muted in adult clusters. Whereas, for adult basal clusters exclusively activated regulons included E2F4, CEBPA, NFIA, NRF1, NFE2L2 and JUND (Fig. 7C). On the other hand, certain regulons associated with ETS-like factors (e.g., ELF1, ELF3), Kruppel-like factors (e.g., KLF6), EGR (e.g., EGR1), MYC, IRF (e.g., IRF8, IRF9), and CEBP (e.g., CEBPD, CEBPG) were detected in basal corneal clusters of both the larvae and adults (Fig. 7D), indicating significant conservation of regulons through development. Among these, many of the proteins and target genes belonging to TF families such as Nuclear Factor of Activated T Cells (e.g., NFATC1), homeobox (e.g., MEIS1), activator protein-1 (e.g., JUND), and basic region-leucine zipper (e.g., CEBPA, ATF4) proteins are known to play critical roles in stem cell maintenance of different tissues (Horsley et al., 2008; Unnisa et al., 2012; Zhao et al., 2015). For instance, a regulon comprising the CCAAT enhancer binding protein delta (CEBPD) transcription factor, detected in all corneal basal cells, is known to be instrumental in regulating self-renewal and cell cycle length of limbal stem cells in humans (Barbaro et al., 2007). Likewise, FOS which is a key factor in regulating cell cycle entry, proliferative expansion, and adult muscle stem activation (Almada et al., 2021), has an enriched expression among all basal cells. The identification of these regulons, both well-established and novel, will open new windows to explore TF markers and pathways for stem cell maintenance and differentiation at key developmental time-points.

Variable cell-cell signaling dynamics dictate cornea development in frog larvae and adult

Next, we examined the intracellular crosstalk that exist among different cell types in larval and adult frog cornea by constructing potential cell-cell communication networks (Farbehi et al., 2019) from our dataset. The edges of the network point from source to target cells, which express specific ligands and their corresponding receptors, respectively. The thickness of edges denote weights representing fold-changes in the expression of ligand-receptor pairs (see Material and Methods). Using this we generated a weighted and directed network of potential cell-cell interactions within the frog cornea. In larval frogs (Fig. 8A), the keratocytes presented a unique interaction landscape with strong inbound and outbound connections with most other cell types, whereas apical epithelial cells were refractory to significant crosstalk. The basal cluster, representing the immature putative CESC population (Ba_IV-L), actively communicates with keratocytes, as the ligands and receptors identified in these epithelial cells display a significant number of interactions (Fig. 8A, C and Fig. S7). Previous studies in humans have demonstrated that the communication between limbal epithelial stem cells and stromal cells is crucial for maintaining the stem cell phenotype and thereby regulates corneal homeostasis (Notara et al., 2010; Wong et al., 2021; Xie et al., 2011). Our current findings in the Xenopus larval cornea supports these observations of epithelial-stromal interactions in the vertebrate cornea. Interestingly, the immunocytes appear to constitute another prominent communicating cell population, although it possessed low cell number. On the other hand, in corneas of adult frogs (Fig. 8B) we noticed an overall increase in interactions that accompanies the rise in cellular diversity. The endothelial cell population have the highest outbound interactions with other cell types, while the apical epithelial cells showing minimal interactions exhibit a pattern similar to that seen in larvae. Surprisingly, we did not detect the stromal keratocytes as an actively communicating cell population within corneas of the adult frog. Of note, keratocytes (in larvae) and endothelial cells (in adults) also seemed to target the respective cell types themselves, suggesting an autocrine mode of regulation.

Figure 8. Dynamics of cell-cell communication networks active in the larval and adult frog corneas.

(A, B) Network diagrams depicting significant cell-cell interactions marked by arrows (edges) pointing in the source-to-target direction. Thickness of arrows shows the sum of weighted paths between cell types, and the color of arrows corresponds to the source. Network diagrams for (A) larval and (B) adult frog corneas are shown. (C, D) Dot plot of representative outbound signals to different cell types in (C) larval and (D) adult frog corneas. Size of each dot signifies the weight of the corresponding ligand-receptor interaction, and the color indicates negative log₁₀ P-value of the source-to-target interaction, as indicated in the keys. The ligands of the source cell are in black and the receptor present on the target cell follows in red. Empirical P-values were calculated and Benjamini-Hochberg correction was performed. The detailed ligand-receptor interactions are shown in Figure S7.

Furthermore, we analyzed the individual ligand-receptor interactions among various cell types. We created dot plots for corneas from both developmental time points demonstrating all ligand-receptor interactions with a minimum path weight of 1.5, for all significant cell-cell relationships (P_adj < 0.01) (Fig. 8C, D and Fig. S7). This helped resolve a comprehensive network of potential cell-cell interactions, revealing prominent ligand-receptor interactions that characterize the frog cornea. In line with reports from humans and other vertebrates (Robertson et al., 2021; Zhang et al., 2015), we detected the expression of ligands and receptors of key signaling pathways, such as the Wnt/β-catenin, TGF/β, FGF and Notch pathways in the frog cornea. For instance, Fibroblast Growth Factor 7 (FGF7), has been implicated in promoting limbal stem cell proliferation and modulating renewal and homeostasis of the corneal epithelium (Hayashi et al., 2005). In our study, Fgf7 was detected in basal epithelial cells, apical epithelial cells, as well as immune cells, while their corresponding receptors (Fgfr1, Fgfr2, Fgfr3) were mainly identified in keratocytes (in larvae) and endothelial cells (in adults). Here, we detected Tenascin C (Tnc) expression exclusively in basal cells (Ba_IV-L), and its corresponding receptors, including Itgb1, Itga9, Itga5 and Egfr were present both in basal cells and keratocytes of larvae, indicating intra-and inter-cellular communications. Similarly, in adult corneas Tnc ligands were mostly expressed in endothelial cells, while their respective receptors were heterogeneously detected in apical, basal and endothelial cells. Similar to recent single-cell reports for the human limbus (Collin et al., 2021), we observed multiple chemokine ligands, such as Ccl5, Cxcl12, Cxcl16 and their receptors in the frog cornea, indicative of potential regulatory interactions exhibited by immune cells.

4. Discussion

In this study, we characterized over 22,000 single-cell transcriptomes to generate the first comprehensive cellular atlas of the cornea in Xenopus larvae and adults. Recent single-cell transcriptomic studies in humans (Collin et al., 2021; Dou et al., 2021) and mouse (Kaplan et al., 2019) focused on fully formed mature corneas; however, both early and adult developmental stages of corneal transcriptional heterogeneity have not been analyzed in any single species. We leveraged the transition from the simple bi-layered corneal tissue organization of larvae, to the anatomically more advanced organization of adult frog corneas, which is similar to that of humans, to understand the gene profile dynamics, differentiation program, and transcription factor regulatory networks for in vivo corneal development and maturation.

Cluster analysis revealed increasing cellular complexity with the progression of corneal development. This includes 4 clusters of basal epithelia, 1 cluster of apical epithelia, 1 keratocyte cluster, and 1 immunocyte cluster found in larvae (cells of the loose larval endothelial cell layer were not captured here). Whereas in the adult, 3 clusters of basal cells, 2 clusters of apical cells, 1 keratocyte cluster, 1 endothelial cluster, 1 neuronal cluster, 3 immunocyte clusters and 1 blood cell cluster were identified (Fig. 2). Traditionally, studies have considered proliferation marker genes such as Mki67 and PCNA, to characterize TACs (Guzman et al., 2013; Lehrer et al., 1998). In addition, a more recent study by Li et al. (2021b) described additional cell cycle-dependent genes (e.g., RRM2, TK1 and CENPF, CDC20) as signature markers for identifying TACs. Surprisingly, none of the basal subclusters (either in larvae or adult) were differentially enriched for these marker genes. The presence of a TAC cell type has been challenged using the mouse tail epidermis model (Jones et al., 2007), and our observations could align with arguments in the field that TACs are not a real entity; rather simply a functional concept. On the other hand, the lack of a unique TAC cluster in the frog corneas could be a species-specific feature. The heterogeneous distribution of proliferation marker genes, and cells existing in S and G2/M phases bolster the observation that a specific TAC cell type may not exist in frogs (Lavker et al., 2020; Mort et al., 2012).

The heterogeneity present among limbal basal epithelial cells has been previously postulated (Nowell and Radtke, 2017), and was recently characterized at single-cell level in humans. While work by Li et al. (2021a) highlighted five major cell types spanning stages of limbal SC differentiation in the basal epithelium of humans, another study by Dou et al. (2021) reported four cell states in limbal epithelial stem cells that vary in properties of stemness and differentiation. In the present study, we identified four distinct transcriptional states in basal cells of larval frogs (Ba_I-L, Ba_II-L, Ba_III-L and Ba_IV-L) and three in adult frogs (Ba_I-A, Ba_II-A and Ba_III-A). In our attempt to identify genes that specifically characterize basal populations, we detected a high transcript level expression of corneal marker genes (e.g., tp63, krt19, krt15.2, itga6, itgb4) known from studies in other vertebrates. Of significance, the smallest basal clusters found in larvae (Ba_IV-L) and adults (Ba_III-A) exhibited differential expression of known stem-cell related genes (e.g., col17a1, col14a1, nfatc1, gpha2), leading us to postulate that these clusters represent the putative stem cell pool in frogs.

To discover the gene expression patterns along the differentiation axis, we modeled a pseudo-temporal trajectory of epithelial cells at both developmental stages. As illustrated by the pseudo-temporal plot in larvae (Fig. 5), we propose a trajectory of epithelial cells originating from the least differentiated cluster, Ba_IV-L, transitioning through three discrete basal cell states, Ba_I-L, Ba_II-L and Ba_III-L, and then finally to the terminally differentiated epithelial cluster, Ap-L. The underlying cellular and molecular transitions represent a complex differentiation and migratory process of epithelial cells in early-stage frog corneas. Although it cannot be ruled out that the temporal progressions of individual cells during differentiation might be more dynamic than captured in this analysis, the data adds to our current understanding of earlier stages of corneal development and differentiation. We are particularly intrigued by our new finding of the role of thyroid hormone genes in frog corneal differentiation, and will seek to explore thyroid hormone function in future studies. Furthermore, a number of signaling pathways emerged as being involved in regulating cornea differentiation (e.g., the Retinoic Acid, FGF, and Wnt/β-catenin pathways), that have also been shown to play crucial roles in supporting cornea-lens regeneration in larvae (Day and Beck, 2011; Fukui and Henry, 2011; Hamilton et al., 2016; Thomas and Henry, 2014).

Compared to larvae, the adult frogs have a significantly complex cornea with a multilayered epithelia. Therefore, to uncover gene expression dynamics that regulate differentiation of adult corneal epithelial cells, we conducted independent pseudotime analyses involving only basal cells, followed by basal to apical cells. We inferred cellular trajectories going from the least differentiated cell type (Ba_III-A) to a transitional basal cell type (Ba_II-A) or directly leading to a more mature basal cell type (Ba_I-A). This is followed by cells of cluster Ba_II-A committing along either apical cell paths (i.e., cell path 3; terminally differentiated superficial cell type, Ap_I-A and Ap_II-A) or the mature basal cell path (i.e., cell path 4; Ba_I-A) (Fig. 6), which in turn convert to Ap_I-A and Ap_II-A cells (Fig. S6). However, to comprehensively understand the precise progression of cells through various layers in the adult cornea, we need to integrate spatial information with scRNA-seq data. In the future, lineage tracing along with scRNA-seq (Kester and van Oudenaarden, 2018) will help reconstruct a more refined hierarchical trajectory of corneal cells in these anurans — similar to that reported in adult mice corneal epithelium and skin epithelium (Altshuler et al., 2021; Guerrero-Juarez et al., 2019; Haensel et al., 2020). Moreover, live imaging of these fate transitions would help shed light on their behaviors.

At the transcriptomic level, the expression patterns and functions of several key genes involved in corneal epithelial development are known, but regulators of these genes are still being identified (Stephens et al., 2013; Swamynathan, 2013). In addition, a recent single-cell transcriptomic study investigating the inter-species (humans, pigs, mouse and others) conservation of TFs and target genes (regulons) in different compartments of the eye, including the cornea, reported that key regulons are highly conserved (Gautam et al., 2021). Therefore, the study of expressed TFs in frog corneal epithelium may provide insight into the machinery that maintains the various cell types during the vertebrate cornea developmental window. Here, we leveraged our scRNA-seq data to comparatively analyze the dynamic transcriptional regulons active in different cell types (i.e., basal and apical epithelial cells, etc.) at discrete stages. In our analysis, we identified various regulons to be conserved between larvae and adult basal cells, indicating that certain TFs and target genes in these modules globally regulate these basal cell states. Based on the identification of regulons that are unique and enriched in basal cells of larval vs. adult corneas (Fig. 7), we propose that biomarkers of basal CESCs may be developmental stage specific. This is in line with conclusions from our previous study where we demonstrated that protein level expression of various biomarkers, including TFs (e.g., p63, Pax6, Tcf7l2) varies through the developmental time course in Xenopus (Sonam et al., 2019). Similar conclusions were drawn by earlier studies examining anatomy and stem cell activity in developing corneas in humans and rats (Chung et al., 1992; Davies et al., 2009). Taking this into account, it is imperative that future studies on the regulation of corneal cell types should particularly focus on multiple, distinct stages of corneal morphogenesis rather than only the mature adult corneas. This will facilitate identification of many previously uncharacterized TFs and other genes that can be potentially explored as excellent CESC biomarker candidates for a given developmental stage.

How do cell-cell interactions change as frog corneas undergo development and maturation? To comprehensively probe underlying cell communication that drives heterogeneity and cell state transitions, we first constructed the communication map between all corneal cell types and dissected all putative interactional links across corneal cell types. During stages 48-50, keratocytes are actively migrating to form the stromal layer beneath the epithelium (Hu et al., 2013). We found that during the larval stage, the stromal keratocytes constitute the cell type with the highest number of interactions. However, in adult frogs with a fully mature cornea, the endothelial cells exhibit the greatest number of communication links. Recent single-cell studies have unraveled the communication dynamics of limbal stem cells with surrounding niche cells and surveyed the ligand-receptors that regulate the human limbus (Dou et al., 2021). Our work to identify multiple ligand-receptor pairs in the basal cells and other cell types of larval and adult frog corneas provides important clues about the regulation of the corneal epithelia in yet another vertebrate model and adds to current knowledge in the field of corneal biology.

One of the key features of Xenopus development is the metamorphosis process of tadpoles into juvenile adults (Dodd and Dodd, 1976), which involves an extensive remodeling process of various tissues and organs (such as the skin, digestive tract and nervous system), and this is accompanied by a dramatic loss of regenerative competence of organs, including the lens (Filoni et al., 1997). It is fascinating that while larval stages of Xenopus can regenerate many of their body parts, including the lens (e.g., tail, spinal cord, limb bud) (Beck et al., 2009), adult frogs dramatically lose this ability. A fundamental and rather unresolved question is why the same tissue from different stage of development exhibits such markedly different abilities for regenerative growth. Earlier work on Xenopus lens regeneration had identified some candidate genes, including pax6 and fgfr2 (bek isoform), that may contribute to the cornea’s competence to regrow a lens (Arresta et al., 2005; Gargioli et al., 2008). Understanding regenerative processes at an unprecedented single-cell resolution has been possible in various vertebrates, including spinal cord and tail regeneration in Xenopus (Aztekin et al., 2019; Kakebeen et al., 2020), as well as limb regeneration in axolotls (Leigh et al., 2018). Future single-cell transcriptomic profiling encompassing the complete time course of Xenopus cornea development, leading up to the adult frog cornea, will help us comprehensively understand the molecular and transcriptional heterogeneity that contributes to the cornea’s capacity to regenerate a lens. Furthermore, a time course-based single-cell analyses at different stages of lens reformation will shed light on gene expression profiles and the precise developmental trajectory of individual corneal cells and discern their spatio-temporal progression, as the cornea reforms a lens.

In summary, the transcriptional atlas of the Xenopus cornea described here provides detailed insights into corneal epithelium development and differentiation. One can extrapolate the findings of this study to test a number of newly identified enriched basal corneal epithelial genes for their potential application as specific markers in other vertebrates, including mice and humans. An extensive molecular and cellular characterization of corneal epithelia — with a key emphasis on transcriptional composition of basal cells and their regulation — will serve as a baseline for future studies in regenerating corneal tissues in vertebrates. This will also be crucial for designing and improving therapeutic treatments for diseases and dystrophies that affect corneal epithelial integrity.

Supplementary Material

Supplementary Data

Figure S1. Quality control (QC) metrics of Xenopus cornea scRNA-seq data. (A) Table showing the QC cutoffs applied on raw reads from scRNA-seq data prior to analysis. (B) Violin plots showing the distribution of gene counts, UMI counts and % mitochondrial content in larval and adult samples, after applying QC cutoffs.

Figure S2. (A, B) tSNE and PHATE projections of all larval corneal cells. (C, D) tSNE and PHATE projections of all adult corneal cells. Cells are colored by annotated cell types, as shown in Figure 2.

Figure S3. The absence of skin and lens cell types in (A) larval and (B) adult frog cornea was verified based on the expression of marker genes. Feature plots of cell-type-specific marker genes. Skin cell marker genes (highlighted in blue): melanocyte inducing transcription factor (mitf.L and mitf.S), tyrosinase (tyr.S), melan-A (mlana.L), dopachrome tautomerase (dct.L). Lens cell marker genes (highlighted in red): crystallin alpha A (cryaa.S), crystallin alpha B (cryab.L and cryab.S).

Figure S4. Heatmap demonstrating the cluster-specific expression of the top 5 ranking novel candidate marker genes in each cluster of (A) larvae and (B) adult cornea cells identified in the scRNA-seq datasets. Cells are colored by annotated cell types, as shown in Figure 2.

Figure S5. Proportion of cell populations in each cluster for (A) larval and (B) adult scRNA-seq datasets. Numbers on each bar indicate the number of cells.

Figure S6. (A) Pseudo-temporal ordering of all cells from basal cluster (Ba_I-A) and apical clusters (Ap_I-A and Ap_II-A) in adult cornea to resolve the path of transition basal cells (Ba_I-A) to apical layers. (B) Distribution of clusters along the cellular trajectory identified in (A).

Figure S7. Dot plot representing all outbound signals to different cell types in (A) larval and (B) adult frog corneas. Size of each dot signifies the weight of the corresponding ligand-receptor interactions, and the color indicates negative log₁₀ P-value of the source-to-target interaction, as indicated in the keys.

Data Availability

All raw RNA-seq data files are available on the NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE154896.