Since LSCs and their progenitors (TACs) represent only a fraction of cells that make up the epithelium, we sought to enrich for these populations to interrogate the mechanisms that regulate the mode of cell division and therefore their fate. To do this, we implemented a dynamic in vivo reporter mouse model (CycB1-GFP) that enabled us to visualize the activity of actively cycling cells in the corneas of live mice (Fig. 1A). CycB1-GFP transgenic animals express an enhanced green fluorescent protein (EGFP) that is fused to the N-terminus of the Cyclin B1 (Ccnb1) gene (Klochendler et al., 2012). Since the N-terminus is targeted for degradation in a cell-cycle dependent manner, only cells that are actively cycling (S-G2-M stages) express the CycB1-GFP protein.

We performed time-lapse imaging in the corneas of live mice to validate reporter dynamics in actively cycling cells. We confirmed that only cells in the basal layer were GFP+ as only basal-layer cells in the corneal epithelium are capable of dividing (Fig. 1B-D; Movie S1). Using real-time imaging we visualized cells in the live cornea undergoing mitoses, which validated the fast kinetics of the in vivo cell cycle reporter (Fig. 1C; Movie S2). To further test the fidelity of the reporter, we topically administered 12-O-Tetradecanoylphorbol-13-acetate (TPA) to mouse corneas, which has a strong mitogenic effect on epithelial tissues (Taketani et al, 1983). Only one day after treatment, we saw massive increases in the number of GFP+ cells across the ocular epithelium, which was more pronounced in the limbus and central cornea (Fig. 1E). One week after treatment, the number of GFP+ cells had declined and were comparable to untreated eyes. Taken together, we demonstrate here a powerful tool for visualizing cell proliferation dynamics in a living eye.

To capture possible imbalances in the rate of cell proliferation between the epithelial compartments, we performed image analysis and quantified the number of GFP+ cells in different areas of the corneal epithelium. Our data showed that actively cycling stem cells and progenitors are distributed uniformly within the basal layer of the limbus and cornea (Fig. 1F-H). This suggests that the rate of proliferation alone, does not impose a growth pressure between these compartments. To this point, we observed from our time-lapse imaging data from live corneas of CycB1-GFP mice that all cell divisions are planar with both daughter cells separating within the basal layer (Fig. 1C, Movie S2). This implies that the orientation of the mitotic spindle is likely not the mechanism by which corneal epithelial cells regulate their asymmetric fate and terminal differentiation. We then asked if the orientation of the cell divisions, in respect to the radial axis connecting the limbus to the center of the cornea, plays a role in the centripetal flux of progenitors within the basal layer. Our analysis showed that even though there were measurable biases for specific angles; on average, cell divisions in the limbus and cornea were oblique to the radial axis of the cornea (Fig. 1I). From these data, we infer that the orientation of mitosis does not directly determine the fate of LSCs and TACs in the cornea.

To develop a greater understanding of the genes and molecular pathways that regulate the fate of LSCs and TACs, we harvested corneas from adult mice and performed Fluorescence Activated Cell Sorting (FACS) to purify only GFP+ corneal epithelial cells. (Fig. 2A; Fig. S1A). Cell filtering and unbiased clustering defined 8 cell populations (Fig. 1B, C; Fig. S2A). Direct comparison with published single-cell RNA sequencing datasets of mouse corneal epithelial cells showed an enrichment in proliferative basal cells (Fig. S1B, C). Moreover, we confirmed that the majority of cells in our dataset expressed basal cell markers, were actively cycling and were healthy, according to low expression of apoptotic markers (Fig. S1D). Taken together the initial analysis highlighted our ability to specifically interrogate gene expression in single LSCs and corneal TACs with high resolution.

Unique cell identities were classified based on gene expression, which revealed clusters of conjunctival (1), limbal (1), corneal (5), immune (1) cells. Conjunctival cells were primarily identified based on the expression of Keratin 13 (Krt13), Keratin 19 (Krt19) (Altshuler et al., 2021; Ramirex-Miranda et al. 2011, Kasper et al., 1988), as well as the lack or low expression of corneal and limbal markers Keratin 12 (Krt12) (Kasper et al., 1988; Kurpakus et al., 1994; Liu et al., 1993) and Glycoprotein Hormone Subunit Alpha 2 (Gpha2) (Altshuler et al., 2021), respectively (Fig. 1C, Fig. S2B). Similarly, the limbal cluster was defined based on Krt15 and Gpha2 expression (Altshuler et al., 2021; S. Yoshida et al., 2006) and low expression, of both conjunctival and corneal markers (Fig. 1C, Fig. S2B). The corneal-specific Keratin 12 (Krt12) was detectable in 5 distinct clusters (Fig. 1C, Fig. S2B).

The smallest of those corneal TACs clusters displayed hallmarks of the previously described compound corneal niches, such as expression of Krt12, Krt19 and Mucin 4 (Fig. 1C, Fig. S2B) (Pajoohesh et al., 2012). These compound niches were shown to give rise to goblet cells in the corneal epithelium that are distinct from those in the conjunctiva and play a role in the pathology of limbal stem cell deficiency. Other unique genes associated with the goblet niche population are Krt6b, Krt78, Mal, Nccrp1, Plac8, Psca and the transcription factor Hopx (Fig S2B). Another distinct cell cluster within the cornea supergroup was characterized by the expression of Urah, a gene that was previously described to be highly expressed in a group of distinct corneal progenitors proposed to play a role in “adaptive regeneration” (Lin et al, PNAS 2023) (Fig S2B). Similarly, this cluster expresses Fabp5, Sparc and Krt16, consistent with previous reports. In addition to Krt12, progenitor cells in all the corneal clusters express high levels of Slurp1, Mgarp and Aqp5 (Fig. S2B).

To investigate the relationships between our identified cell clusters and resolve the differentiation trajectory of cells that maintain the corneal epithelium, we performed trajectory inference analysis. A clear trajectory emanating from the limbal cluster progressed through the TAC-peripheral, TAC-central, and TAC-terminal clusters, suggesting that these represent consecutive stages of terminal differentiation of the TACs that emerge from the limbus (Fig. 2D). The trajectory from the LSC cluster also showed distinct branching points towards the “goblet niche” and “adaptive” TAC clusters. Next, we performed differential expression analysis of genes that encode for transcription factors to identify key transcriptional regulators of the distinct cellular identities found within the stem and progenitor cell pools of the corneal epithelium. Our data showed that LSCs express high levels of Sox9, Sox4, Barx2, Nfib1 and Osr2 (Fig. 2E). Expression of these genes decreases as the TACs exit the niche, and in turn there is upregulation of Klf4, Fos, Hif, Jun, Junb, and Klf3. (Fig. 2E). While the five corneal clusters showed similar expression for the same transcription factors that differed from the limbal cells, the “goblet niche” progenitors additionally expressed high levels of Elf3, Mxd1 and Zbtb7a. Expression of all these transcription factors waned in the TAC-terminal cluster, suggesting that these cells may be the ones closest to terminal differentiation as also indicated by the trajectory analysis.

To address our original quest for genes which regulate the switch from symmetric cell fate in LSCs to asymmetric in TACs, we identified Sox9 as one of the transcription factors that are differentially expressed between the limbal and the corneal clusters (Fig 2E, F). The transcription factor Sox9 has long been known for its role as a regulator of stem and progenitor cell activity in other ectodermal-derived appendages (Kadaja et al., 2014; Poché et al., 2008; Scott et al., 2010; Vidal et al., 2005). Given the many parallels in gene expression patterns between the cornea and the hair follicle, where Sox9 has an established role in regulating hair follicle stem cell fate, we sought to explore the expression of Sox9 within our single cell dataset We confirmed that Sox9-high cells express all the typical limbal markers including Gpha2 and Krt15 (Fig. 2G). Other highly expressed genes within the Sox9-high population are Wfdc18, Tppp3, Tmem17a, Tmem17b, Igfbp2, and Lgals7 (Fig. 2G). Although transcript levels are not necessarily indicative of functional activity, we predicted that Sox9 may be functionally more important in LSCs than in corneal epithelial cells.

To interrogate the role of Sox9 in homeostatic maintenance of the corneal epithelium, we utilized a two-photon-microscopy-based system to visualize whole eyes in live mice without compromising the structural integrity of the tissue (Farrelly et al., 2021). To validate Sox9 expression within the murine cornea in vivo, we imaged the corneas of adult Sox9^IRES-EGFP(Sox9-EGFP) mice (Chan et al., 2011), which express EGFP from the endogenous Sox9 locus (Fig. 3A, B). In these animals, GFP expression is almost exclusively restricted to the limbal compartment. Moreover, Sox9-EGFP levels have been previously shown to correlate with endogenous Sox9 levels (Formeister et al., 2009). Here, we observe Sox9-EGFP is expressed at two steady-state levels whereby Sox9 activity is highest in LSCs residing in the niche and lowest in progenitors that have exited the niche (Fig, 3B).

We next combined intravital imaging with an inducible Cre-recombinase-based in vivo lineage tracing strategy to resolve the activity and stemness potential of Sox9-expressing limbal cells in corneal homeostasis by longitudinal live imaging. These mice express a tamoxifen-inducible Cre recombinase from the endogenous Sox9 locus and a ROSA26LoxP-STOP-LoxP-tdTomato (R26-tdTom) Cre-reporter (Sox9CreERT2;tdTom) (Fig. 3A). To directly track the activity of Sox9CreERT2;tdTom lineage-traced cells, we acquired high resolution, full-thickness serial optical sections of the whole eye, including the limbus and central cornea, and then reimaged the same eyes longitudinally with the same acquisition parameters (Fig. 3A, C). By 1-month post-induction, we observed the emergence of long-lived lineages emanating from the limbus (Fig. 3A). Given that these lineages can only be sustained by LSCs (Amitai-Lange et al., 2015; Di Girolamo et al., 2015; Dorà et al., 2015), we concluded that Sox9 marks a population of LSCs as evidenced by their ability to contribute to long-term corneal homeostasis.

To explore the tissue-specific role of Sox9 within the corneal epithelium, we implemented an inducible conditional gene targeting strategy (Fig. 4A). Sox9 floxed mice (Akiyama et al., 2002) were crossed to p63CreERT2 reporter animals (p63CreERT2;tdTom) to selectively delete Sox9 from basal-layer epithelial cells, the only cells that retain proliferative capacity and stemness potential (Cotsarelis et al., 1989), in adult corneas. By 2 months post induction (MPI), a proportion of Sox9 conditional knockout (cKO) corneas presented with hyperplastic growths at the central cornea, reminiscent of squamous metaplasia, and displayed extensive epithelial and stromal thickening at the limbus compared to control corneas (Fig. 4B-E’’). Compared to controls, Sox9-cKO corneas were thicker in the central cornea and limbal compartments (Fig. 4H). While stromal thickness was increased within the central cornea and limbal compartments of Sox9-cKO mice, changes in epithelial thickness were significant at the limbus, but not at the central cornea (Fig. 4.4F-G). This is likely due to the observed variance in plaque development/severity across Sox9-deleted animals. When Sox9-cKO corneas were scored on the severity of epithelial and/or stromal thickening, only about one-third of corneas (5/11 animals [7/21 corneas]) presented with moderate to severe epithelial thickening (Fig. 4I-M). This likely explains why epithelial thickening between experimental and control corneas is not significantly different (Fig. 4F). Based on these initial observations, we hypothesized Sox9 may play an important role in regulating the integrity of the corneal epithelium.

To begin characterizing our phenotype, we first validated the efficiency of our Cre/LoxP gene ablation system by performing immunostaining for Sox9 (Fig. S3). In control eyes, nuclear Sox9 staining was restricted to the basal layer cells within the limbus, whereas in the central cornea Sox9 faint staining could be observed in both the basal and suprabasal layers (Fig. S3A/A’, C/C’). In Sox9-cKO animals, Sox9 was not detected in either the limbus or central cornea, indicating efficient deletion of the Sox9 gene (Fig. S3B/B’, D/D’). Furthermore, we validated the epithelial specificity of our system by staining for Sox9 in retinal Müller glial cells, which are not of epithelial origin (Fig. S3E/E’, F/F’) (Poché et al., 2008).

Unlike control corneas, which exhibited a single K14+ basal layer, Sox9-cKO corneas displayed an expanded K14 layer in which K14 could be detected ectopically in the suprabasal and superficial layers of the epithelium (Fig. S4A-D’). Keratin 14 (Krt14) is a classic marker of basal-layer keratinocytes, the proliferative layer of the epithelium (Coulombe et al., 1993; Moll et al., 1982). Unlike the epidermis of the skin, the corneal epithelium is non-keratinized and mostly defined by the expression of Keratin 12 (Krt12) (Liu et al., 1993). In control animals, Krt12 expression can be detected throughout the suprabasal and superficial layers of the epithelium; however, it is not expressed within the limbus. In Sox9-cKO corneas, Krt12 is noticeably absent within the hyperplastic growths, while its expression can still be detected in the periphery of the corneal epithelium, outside the boundaries of the growths, where its expression is comparable to control eyes (Fig. S4E-H’). As Krt12 expression is unique to the corneal epithelium, its absence within the corneal plaques indicated a loss of corneal identity or cell fate switch.

Histologically, Sox9-cKO corneas displayed evidence of a keratinized surface, which had a striking resemblance to the stratum corneum of the epidermis (Fig. S5). Staining for Loricrin (Lor), a major structural protein of the epidermal cornified cell envelope, confirmed our suspicion that Sox9-cKO corneas were starting to take on an epidermal-like fate (Fig. S5). In Sox9-cKO corneas that displayed severe thickening and plaque growth (n = 5 corneas [4 mice]), Lor could be detected within the plaque growths, but not at the corneal periphery or limbus. This is consistent with the observation that K12 expression was properly patterned within Sox9-cKO corneas outside of the plaques. Collectively, these findings suggest Sox9 plays an important role in regulating corneal epithelial cell fate and identity as its loss resulted in a corneal-to-epidermal cell fate switch.

Within the corneal epithelium, cell fate decisions are largely influenced by the position of the cell within the tissue. At the limbus, LSCs divide symmetrically to maintain the stem cell pool but also to give rise to TAC progeny. As TACs exit the niche, they start to divide asymmetrically with increasing frequency giving rise to progenitors that will either terminally differentiate or divide into daughter cells with asymmetric fates where one terminally differentiates and the other is retained as a progenitor with proliferative capacity (Farrelly et al., 2021). Compared to controls, Sox9-cKO corneas with severe plaque development showed extensive hyperproliferation (Fig. S6). Staining for Ki67 revealed almost all basal cells, as well as cells in the suprabasal layers, were actively proliferating throughout the limbus and corneal epithelium compared to controls (Fig. S6 A-D’). We postulated that the increased proliferative activity throughout the corneal epithelium in the absence of Sox9 explains, at least in part, the observed thickening of the epithelium.

To this end, we provided evidence that the transcription factor Sox9 is preferentially expressed in LSCs and that conditional ablation of Sox9 from the basal layer of the corneal epithelium leads to the development of squamous metaplasia at the central cornea. Although we hypothesized that loss of Sox9 would primarily affect the limbus, the most prominent phenotype in Sox9 cKO mice is the aberrant growths developing at the center of the cornea. Given that cell divisions switch from mostly symmetric in the limbus to increasingly asymmetric in the central cornea, with an increased propensity to terminally differentiate (Farrelly et al. 2021), we hypothesized that the transition from Sox9-high to Sox9-low expression as the TACs exit the limbal niche is required for their normal differentiation. Thus, we predicted that after conditional deletion of Sox9 TACs are unable to switch to asymmetric fates, without which they continue to divide symmetrically leading to cell crowding at the center of the cornea.

To test this hypothesis, we developed an inducible dual Cre and Tet system that would enable us to suppress differentiation (Differentiation-OFF) in single clones that we could visualize with a co-dependent H2BGFP reporter and track their behavior over time by live imaging (Fig. 5) (Farrelly et al., 2019). The premise of this genetic system is that active Notch signaling is required for the commitment to terminal differentiation of progenitors in stratified epithelia. Therefore, by enforcing total inhibition of canonical Notch, using the dominant negative MAML1 allele (Differentiation-OFF), we predicted that cells would be locked in a symmetric fate and undergo perpetual self-renewal. While H2BGFP labelled clones in control corneas could be observed differentiating normally and moving from the basal to the suprabasal layers of the epithelium over time, labelled clones in “Differentiation-OFF” corneas remained trapped in the basal layer (Fig. 5B). To further validate that suppressing asymmetric cell fate and differentiation of TACs leads to unchecked self-renewal, we performed quantitative clonal analysis. As expected, basal clones from normal corneas decayed over time, consistent with the short-lived nature of corneal TACs (Fig. 5C). In contrast, “Differentiation-OFF” clones continued to grow. While delamination of “Differentiation-OFF” cells was predictably suppressed for many days, it eventually started increasing after several days. This is likely due to crowding of cells which compete for space leading to cell extrusion from the basal layer (Fig. 5B see right panel, Movie S3).

To assess changes in migratory potential of “Differentiation-OFF” cells, we followed their fate by imaging the corneas of our “Differentiation-OFF" mice over time in vivo. Despite being unable to differentiate, “Differentiation-OFF” clones retained their ability to clonally expand and transit centripetally, leading to increasing crowding in the central cornea (Fig. 5D; Movie S4).

Moreover, several weeks after induction, we observed plaques developing in the central cornea that phenocopied those we previously saw in the Sox9-cKO mice. Altogether, these data provide support for a model whereby plaque growths in Sox9-cKO corneas are the result of TACs failure to switch to an asymmetric cell fate (cells trapped in basal cell identity), leading to their unchecked growth and crowding at the central cornea (Fig. 6). Therefore, Sox9 is a key transcription factor that regulates progenitor cell fate and is required to maintain corneal epithelial homeostasis.