Author contributions: A.P.-G.: conception and design, collection and assembly of data, provision of study materials, data analysis, manuscript writing, and final approval of manuscript; S.P.-G.: collection and assembly of data, data analysis, provision of study materials, administrative support, and final approval of manuscript; G.T.: collection of data and data analysis; M.A.S.: financial support, conception and design, assembly of the data, data analysis, manuscript writing, provision of study materials, and final approval of manuscript.
Disclosure of potential conflicts of interest is found at the end of this article.
First published online in STEM CELLSEXPRESS July 20, 2012.
Goblet cells are terminally differentiated cells secreting mucins and antibacterial peptides that play an important role in maintaining the health of the cornea. In corneal stem cell deficiency, the progenitor cells giving rise to goblet cells on the cornea are presumed to arise from differentiation of cells that migrate onto the cornea from the neighboring conjunctiva. This occurs in response to the inability of corneal epithelial progenitor cells at the limbus to maintain an intact corneal epithelium. This study characterizes clusters of cells we refer to as compound niches at the limbal:corneal border in the unwounded mouse. Compound niches are identified by high expression of simple epithelial keratin 8 (K8) and 19 (K19). They contain variable numbers of cells in one of several differentiation states: slow-cycling corneal progenitor cells, proliferating cells, nonproliferating cells, and postmitotic differentiated K12+Muc5ac+ goblet cells. Expression of K12 differentiates these goblet cells from those in the conjunctival epithelium and suggests that corneal epithelial progenitor cells give rise to both corneal epithelial and goblet cells. After wounds that remove corneal epithelial cells near the limbus, compound niches migrate from the limbal:corneal border onto the cornea where K8+ cells proliferate and goblet cells increase in number. By contrast, no migration of goblet cells from the bulbar conjunctiva onto the cornea is observed. This study is the first description of compound niches and corneal goblet cells and demonstration of a role for these cells in the pathology typically associated with corneal stem cell deficiency. Stem Cells2012;30:2032–2043
The ocular surface is composed of conjunctival, limbal, and corneal epithelia that express distinct repertoires of keratins. While all three epithelial cell types express K5 and K14 keratins like the epidermis, the conjunctiva and limbal epithelia express “simple epithelial” keratins including K4, K8, K13, K18, and K19 [1–5]. Human corneal epithelial cells express the K3/K12 keratin pair . The mouse genome lacks an intact K3 gene but its corneal epithelium expresses K12, which is considered a marker of corneal epithelial cells [7, 8]. Conjunctival and corneal epithelia have been reported to derive from distinct stem cell populations [9–15]. Data from studies looking at expression of corneal-specific proteins in addition to K12 within conjunctival and corneal epithelial cell lines largely confirm that the conjunctiva and corneal epithelium represent distinct cell lineages [16, 17].
The location of the limbal stem cells (LSCs) has been the topic of numerous studies. A LSC hypothesis (LSCH) was originally proposed [18, 19], has been expanded upon [14, 20-24], challenged , and reviewed extensively [26–34]. In brief, the LSCH states that stem cells required for maintaining the corneal epithelium, called LSCs, reside within a niche in a subset of the limbal basal cells. In the human cornea, this niche takes the form of invaginations called limbal crypts . LSCs divide and give rise to progeny that migrate across the corneal surface converging at the center of the cornea. Similar to conjunctival and limbal epithelia, LSCs express simple epithelial keratins. Based on the LSCH, stem cell deficiency results from damage to the limbus which allows conjunctival epithelial and goblet cells to migrate onto the ocular surface. Clinically, the LSCH is supported by the fact that limbal transplants are highly successful treatments for stem cell deficiency caused by chemical injury .
We described previously that 4 weeks after wounds that remove 70% of the corneal epithelium but not the limbal basal cells, K8+Muc5ac+ goblet cells are seen on the central cornea [37, 38]. Here, we determine that the cells giving rise to the goblet cells on the central cornea arise from migration of progenitor cells located adjacent to the limbus within K8+ clusters. These clusters contain cells that are K12+ and Muc5ac+ and are embedded within a K12+ corneal epithelium. We refer to these clusters as compound niches. The data presented identify the corneal epithelial goblet cell for the first time. In addition, our work indicates that “corneal stem cell deficiency” in the mouse is actually the expansion of a K8+ progenitor cell population located in the limbus that can give rise to both corneal epithelial and K12+ goblet cells.
MATERIALS AND METHODS
Animals and Wounding
All experiments described in this article were conducted in voluntary compliance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research, and the George Washington University Animal Care and Use Committee approved all procedures. Wild-type mice (BALB/c) were obtained from NCI-Frederick. Eight- to ten-week-old BALB/c mice (22–25 g) were anesthetized as previously described . Once the animals were anesthetized, a topical anesthetic (proparacaine ophthalmic solution, Dublin, OH, http://www.butlerschein.com) was applied to their ocular surface until the blink sensation was lost. Corneas were demarcated with a trephine and scraped with a dulled blade to remove the corneal epithelium. Various wound sizes were studied: small wound (1.5 mm of central cornea), medium wound (2 mm involving the central and peripheral cornea), large wound (2.5 mm of central and peripheral cornea), and zonal wound (removing epithelial cells between the two marks made by 1.5 and 2.5 mm trephines, leaving a corneal button and the limbal rim). After wounding, all eyes were treated with erythromycin ophthalmic ointment to keep the ocular surface moist while mice were under anesthesia. For studies involving large wounds, corneas were allowed to heal in vivo for 1, 2, 7, and 10 days as well as 2 and 4 weeks. Animals were sacrificed at the appropriate time points. A minimum of three corneas per time point were evaluated for each antibody combination used. Eyes were sutured prior to enucleation and stored in groups of left or right eyes to allow orientation by quadrant. Eyes were fixed for 2 hours in 1:4 dilution DMSO/methanol and stored long-term in 100% methanol at −20°C. Since data are similar at 1 and 2 days after 2.5 mm wound, only day 1 data are presented.
Whole Mounts and Confocal Microscopy
Eyes stored in 100% methanol were dissected to remove the lens, iris, and retina and four equidistant incisions were made to flatten the corneas. Corneas were stained with a rat anti-mouse antibody against keratin-8 (K8-Troma-I; Developmental Hybridoma Data Bank, University of Iowa, Iowa City, IA, http://dshb.biology.uiowa.edu), rat anti-mouse antibody against keratin-19 (K19-Troma-III; Developmental Hybridoma Data Bank, University of Iowa, Iowa City, IA, http://dshb. biology.uiowa.edu), goat anti-mouse keratin-12 (Santa Cruz Immunology #sc-17101, Santa Cruz, CA, http://www.scbt.com), mouse anti-Muc5ac Ab-1 (MS 145-P1; NeoMarkers, Freemont, CA, http://www.thermoscientific.com), rat anti-mouse β4 integrin (BD Pharmingen, 346-11A, San Jose, CA, http://www.bdbiosciences.com/home.jsp), rabbit monoclonal against Ki67 (Abcam, #ab16667, Cambridge, MA, http://www.abcam.com), and mouse monoclonal anti-BrdU (Roche #347580, Indianapolis, IN, http://www.rocheusa.com) followed by appropriate Alexa Fluor 488 or Alexa Fluor 594 from Molecular Probes or DyLite 488 or 649 from Jackson Immunobiologicals. Corneas were stained with 4′,6-diamidino-2-phenylindole (DAPI, Invitrogen, Grand Island, NY, http://www.invitrogen.com) before flat mounting to reveal nuclei. To achieve the best flattening, the corneas were placed epithelial side-up with mounting medium and coverslipped.
Confocal microscopy was performed at the center for microscopy and image analysis (CMIA) at the George Washington University Medical Center. A confocal laser-scanning microscope (Zeiss 710) equipped with a krypton-argon laser was used to image the localization of Alexa Fluor 488 (488 nm laser line excitation; 522/35 emission filter) and Alexa Fluor 594 (568 nm excitation; 605/32 emission filter). Optical sections (z = 0.5 μm for ×40 images and 1 μm for ×10 images) were acquired sequentially with a ×10 or ×40 objective lens. Typically, eight optical sections were merged and viewed en face. The 3D images presented were generated using Volocity software (Version 5.0, Perkin Elmer).
K8+ Goblet Cell Cluster Quantitation
Two variables were evaluated for each quadrant: (a) the longest distance traveled by the K8+ cluster toward the center of the cornea from the anatomical limbus (defined by the limbal blood vessels) and (b) the number of goblet cells present within representative ×20 fluorescent images as identified by Muc5ac and K8 staining and taken with a fluorescent microscope. The distance from the limbus and the number of goblet cell clusters for each quadrant were averaged and Graphpad Prism was used with one-way ANOVA and Bonferroni post-test to analyze the data. To simplify the quantitation process, images were taken of those sites showing the most clusters within each of the quadrants for all corneas; these images were considered representative of goblet cell induction. We averaged the data (four images per cornea) from each quadrant of all corneas assessed. Forty images from 10 corneas for control, small, medium, and large wounds, and 24 images from six corneas for zonal wounds were used for quantitation. We acknowledge that this approach could bias results toward showing more severe pathology. However, alternative approaches for quantitation involving counting 100% of the clusters present in each cornea or determining the percent of normal corneal epithelium were not feasible because of the incisions required for flattening the corneas for imaging, and the increased number and sizes of the clusters in corneas after large wounds. K8+ clusters were counted if they had more than four cells. If a cluster had more than ∼ 12–15 cells, morphology based on DAPI staining was used to determine whether more than one cluster was present.
BALB/c pups between 1 and 3 days of age were injected with BrdU as previously described  and allowed to mature to 6–8 weeks of age. After sacrifice, eyes were fixed in methanol and DMSO and stored in 100% methanol. Eyes were stained with the primary and secondary antibodies to localize β4 integrin, K8, or Muc5ac. After fixation, they were processed for detection of BrdU. Details of this technique have been published .
Tissues from control and wounded corneas (2 weeks large wound) were collected by scraping the corneal surface with a dulled blade, frozen in liquid nitrogen, and stored at −80°C. Each sample included epithelia from two corneas; 12 wounded and an equal number of control corneas were used for these studies. Tissues were extracted in 50 μl 1× RIPA buffer. Prior to use, 100 μl of 200 mM o-phenathroline hydrochloride in methanol, 25 μl Pefabloc SC Plus (Roche # 11 873 601 001, Indianapolis, IN, http://www.rocheusa.com), 286 μl 7× complete mini protease inhibitor (Roche Complete Mini, # 04 693 124 001, Indianapolis, IN, http://www.rocheusa.com), and 20 μl HALT phosphatase inhibitor cocktail (Pierce/ThermoFisher #78420, Rockford, IL, http://www.piercenet.com) were added to 2 ml of RIPA. Protein assays were performed and equal amounts of total protein were prepared and run onto gels by mixing extracts with keratin sample buffer. Keratin sample buffer was prepared by mixing 300 μl 3× Laemmli sample buffer with 200 μl β-mercaptoethanol . Samples were boiled for 15 minutes before running onto 10% Tris-glycine gels at 140 V. The proteins were transferred to nitrocellulose membrane at 400 mA for 1.5 hours and blots blocked in 5% milk in TBS with 0.1% Tween 20 (TBST) for 1 hour at RT. Primary antibody was added overnight at 4°C followed by washes and secondary antibody incubation for an hour. Blots were subjected to chemiluminescence reaction (Pierce # 34076, Rockford, IL, http://www.piercenet.com) and signal detected using x-ray film. The following primary antibodies were used for immunoblots: goat anti-mouse keratin-12 (Santa Cruz Immunology # SC-17101, Santa Cruz, CA, http://www.scbt.com), mouse antikeratin-8 (Research Diagnostics Inc., # RDI-CBL195C, Flanders, NJ, http://www.cyto.purdue.edu), mouse antikeratin-18 (Research Diagnostics Inc., # RDI-CBL236C, Flanders, NJ, http://www.cyto.purdue.edu), and mouse antiactin (Millipore, # MAB1501R, Billerica, MA, http://www.millipore. com/). Secondary antibodies conjugated with HRP included anti-mouse IgG (GE Healthcare, # NA931V, Buckinghamshire, UK, http://www.gelifesciences.com) and anti-goat IgG (Jackson ImmunoResearch, # 805-035-180, West Grove, PA, http://www.jacksonimmuno.com).
Muc5ac Containing Goblet Cells Are Present Within Clusters of Cells Located at the Limbus of the Unwounded Mouse Cornea
While studying goblet cell induction after corneal debridement wounds, we observed K8+Muc5ac+ clusters of cells at the limbus of the unwounded BALB/c cornea. To simplify our terminology, we refer to the clusters at the limbus as compound niches (Fig. 1A). To determine the number of compound niches per unwounded BALB/c cornea, 12 BALB/c 8-week-old unwounded mouse corneas were used for whole mount imaging after localizing K8 and Muc5ac. We found that the mouse cornea has 11–71 compound niches with a mean of 35. Compound niches are not distributed uniformly around the cornea. As shown in Figure 1B, there are more compound niches present in the temporal and inferior quadrants compared to the nasal and superior quadrants. By counting the number of DAPI-stained nuclei, we found 4–20 cells per niche with a mean of nine cells.
Since compound niches have never been described at the limbus of the healthy mouse, we confirmed that other strains of mice besides BALB/c also have compound niches. The corneas from two to five mice from seven different strains were used to assess K8 and Muc5ac staining at the limbus including pigmented (C57BL6, DBJ, and NZB) and albino (nude, 129SVJ, FVB, and C57BL6-albino) strains. Pigmented and albino C57BL6 mice have fewer and smaller compound niches at their limbus (only one in 10 corneas assessed). The number of compound niches at the limbus in other strains was similar to those observed in BALB/c mice. Thus, the presence of compound niches at the limbus is a common feature of the healthy mouse cornea.
The Goblet Cells Present Within the Compound Niches at the Limbal:Corneal Border Are Both K12+ and Muc5ac+
To determine whether compound niches are part of the corneal or conjunctival epithelium, we colocalized K12, K8, and Muc5ac. As presented in Figure 1C, compound niches express K12, Muc5ac, and K8; Muc5ac is present in the apical aspect of the compound niche. To further characterize the location of compound niches relative to the anatomical limbus, defined by the limbal vasculature, we colocalized K8 and K12 using whole mounts and generated confocal 3D image reconstructions. Typical 3D images are presented in Figure 1D. The compound niches shown are both K8+ and K12+ as indicated by the yellow color in the merged images. Cells at the periphery of the niches express more K12 and K8 than those in the center. The compound niches also express K19, which is more abundant on cells located at the periphery of the niche, as shown in Figure 1E. Goblet cells terminally differentiate from tissue-specific progenitor cells, do not proliferate, and are Muc5ac+. K12 is a marker for cells committed to the corneal epithelial phenotype [7, 8]. These data show that compound niches reside within the corneal epithelium adjacent to the limbus. Further, expression of K12 and Muc5ac in compound niches suggests that these goblet cells derive from LSCs.
Compound Niches Contain β4-Integrin-Bright Label-Retaining Cells Located Adjacent to Muc5ac+ Cells
To identify the cells that give rise to the K12+ goblet cells present within the compound niche, we used the fact that progenitor cells are slow cycling. Neonatal pups were injected with BrdU and allowed to mature. At 8 weeks of age, mice were sacrificed and their corneas were used for assessment of BrdU in slow-cycling cells within morphologically distinct compound niches. In addition to K8 and BrdU, β4 integrin (Fig. 2A) and Muc5ac (Fig. 2B) expression were assessed. Figure 2A shows several compound niches consisting of K8+BrdU+β4integrin+ cells en face on the left and two different clusters of cells, indicated by asterisk and arrowhead, shown both en face and in cross-section at higher magnification on the right. The slow-cycling cells are K8+ and retain variable amounts of BrdU; cells with more BrdU express high levels of β4 integrin. The slow-cycling cells with high level of β4 integrin and BrdU (identified here by asterisk) were characterized previously as LSCs .
In Figure 2B at the upper left, an en face view of several compound niches is shown along with adjacent K8+BrdU+ slow-cycling cells. A 3D image reconstruction was generated and rotated to present the compound niches more clearly; the asterisks in all images indicate the site rotated to generate the 3D image. Two to three BrdU label-retaining cells are seen at the periphery of the clusters; the Muc5ac+ cells are not BrdU+. In addition, there are numerous cells adjacent to the clusters that retain less BrdU, express less K8, and are Muc5ac−. The data presented in Figure 2A and B are consistent with the compound niches containing slow-cycling progenitor cells that give rise to K12+Muc5ac+ goblet cells.
Compound Niches Also Contain Ki67+ Cells at Their Periphery
The Muc5ac+ goblet cells themselves are not label-retaining. These data suggest that the goblet cell progenitors, located within the compound niche, undergo cell division and lose their BrdU label before differentiating into goblet cells. Therefore, we expect to find proliferating cells within the compound niche. We next evaluated Ki67 expression as a marker for actively proliferating cells; data are presented in Figure 2C. A low magnification image with K8 (red), Ki67 (green), Muc5ac (magenta), and cell nuclei (DAPI, blue) is shown on the upper left. The compound niche identified by arrowhead is magnified in the center and presented in 3D on the right. The lower panel in Figure 2C shows the localization of K8, Ki67, and Muc5ac individually. Data indicate that Ki67+ cells are present at the periphery of the compound niche. Since we show in Figure 1D and 1E that K12 expression is enriched at the periphery of the niche, these proliferating cells are likely K12+. A cartoon summarizing the data presented thus far describing the morphology and cellular composition of these clusters is shown in Figure 2D. We refer to these clusters as compound niches because of their proximity to the limbus, their complexity, and the fact that they contain stem-like cells.
Compound Niches Migrate onto the Cornea After Large Wounds
We previously described a model to study mouse stem cell deficiency in which K8+ goblet cells are induced on the cornea beginning 2 weeks after 2.5 mm wounds and persist for 8 weeks [37, 38]. Here, we compare differences in healing responses induced by small (1.5 mm), medium (2.0 mm), large (2.5 mm), and zonal wounds. Zonal wounds close beginning 18 hours after wounding with cells migrating from the periphery and from the center of the cornea. Figure 3 presents data showing the sites and sizes of the wound areas and representative low magnification images of the corneal peripheries from corneas 4 weeks after wounding. Clusters were identified based on K8 staining and morphology. The corneas were evaluated for the number of K8+ clusters and the average distance the K8+ clusters travel from the limbus defined anatomically as the middle of the 100 μm area overlying the limbal vasculature. Once the compound niche moves away from the limbus onto the peripheral and central cornea, we no longer refer to them as compound niches but as K8+Muc5ac+ clusters.
Medium, large, and zonal wounds induce more K8+Muc5ac+ clusters 4 weeks after wounding. The data for the distance migrated by the goblet cells from the limbus are significant for both large and zonal but not medium wounds. Zonal and large wounds remove cells closer to the limbus than medium and small wounds implicating factors related to the proximity of the debridement wound to the limbus in the induction of K8+ clusters on the cornea.
Our next question was whether the goblet cells present on the cornea several weeks after wounding arise from migration and proliferation of compound niches onto the cornea. Figure 4 shows the localization of K8 and Muc5ac and expression of Ki67 at the limbal region 18 hours after small wounds (left) and 24 hours after large wounds (right). These time points were chosen to ensure that the epithelial sheet was actively migrating. After small wounds, the cells that make up the compound niches do not show changes in cellular composition or morphology. However, we do note the upregulation of K8 within linear stripes of cells that correlate with the location of compound niches near the limbus. The area indicated by asterisk is magnified in the images on the lower panels. The proliferating cell marked by Ki67 (green) in the compound niche is present next to the cell which is Muc5ac+ (magenta); they are both K8+. By contrast, 24 hours after large wounds, the morphology of the compound niches located at the limbus is dramatically altered. Cells within each cluster assume a migratory phenotype with Muc5ac+ goblet cells and Ki67+ cells at the front of the migrating cluster. Clusters migrate toward the center of the wound but take a circumferential rather than a direct route. K8 expression appears elevated in the epithelium adjacent to the compound niches.
After cell migration is complete, K8+ cells continue to proliferate and expand so that at 7 days (Fig. 5A), 10 days (Fig. 5B), and 2 weeks (Fig. 6) after wounding, there are large numbers of K8+Ki67+ and K8+Muc5ac+ cells on the corneal surface. K8+ cells are maximal at 2 weeks and take on varied morphologies including K8+Ki67+Muc5ac− cells (Fig. 6, asterisk) associated like beads on a string as well as groups of multiple clusters of cells that include K8+Ki67+Muc5ac+ cells (Fig. 6, arrowhead). K8+ cells are also present near the recurrent erosion site in the central cornea 2 weeks after large wound (Fig. 6).
We harvested epithelium from control and corneas 2 weeks after large wounds for immunoblots to evaluate keratins biochemically. We detect elevated expression of K8 and K18, K8's partner, but only K18 expression was significantly higher. K12 expression decreased slightly but the difference was not significant (Fig. 7A). Continued expression of K12 by corneal epithelial cells is inconsistent with corneal epithelial stem cell deficiency. By 4 weeks, there is a mosaic pattern of K8 and K12 expression on the central cornea (Fig. 7B). The K8+Muc5ac+ cells that appear on the mouse cornea 4 weeks after large debridement wounds arise from migration and proliferation of K8+ cells derived from or associated with compound niches and are present along with K12+ corneal epithelial cells.
Conjunctival Goblet Cells do not Migrate Across the Limbus on to the Cornea During Re-Epithelialization After Large Wounds
Conjunctival epithelial and goblet cells also express K8. Presented in Figure 7C are data showing K8 and Muc5ac localization at the limbus and adjacent bulbar conjunctiva in unwounded corneas and in corneas 1 day and 4 weeks after large wounds. In unwounded corneas, numerous K8+Muc5ac+ conjunctival goblet cells are seen in the bulbar conjunctiva beginning 500–1,000 μm from the limbus, but no K8+Muc5ac+ goblet cells are detected between the conjunctival:limbal border and 400–500 μm away from the limbus toward the fornix. Both at 1 day and 4 weeks after wounding, K8 expression increases within the conjunctival epithelial cells but migration of goblet cells over the limbus is not observed. Occasional goblet cells are seen closer to the limbus 4 weeks after wounding than in unwounded corneas.
Our data show for the first time, clusters of cells that we call compound niches expressing K19, K8, K12, and Muc5ac located at the limbal:corneal border in the unwounded mouse cornea. Contained within the compound niche are slow-cycling, label-retaining progenitor cells, and Ki67+ proliferating cells that are K8+ and K12+ and can give rise to both the corneal goblet cells and K12+ corneal epithelial cells. When 2.5 mm wounds are made to the cornea, conjunctival goblet cells do not migrate over the limbus but compound niches migrate from the limbus onto the cornea where the K8+ cells proliferate to generate more goblet cells. K8+ progenitor cells present within and around the compound niche are the likely sources of the goblet cells that persist on the mouse cornea after large or zonal debridement wounds. Unwounded BALB/c mice have more compound niches compared to C57BL6 mice. We have shown previously that 2.5 mm wounded C57BL6 mice generate more goblet cells on their corneas than BALB/c mice 4 weeks after wounding . We conclude that K8+ progenitor cells at the limbus have the capacity to differentiate into goblet cells in C57BL6 mice when wounds occur close to the limbus.
Beginning in the mid-1980s, a number of reports appeared describing a process called “conjunctival transdifferentiation” [41–44]. These studies reported that after complete removal of the corneal epithelium and limbus in rabbits and rats, conjunctival epithelial and goblet cells migrated onto the cornea where they differentiated into corneal epithelial cells. While experimental details varied, the primary evidence for transdifferentiation was the presence of a clear corneal epithelium and stroma several months after injury. The initial healing that took place in response to these injuries followed a pattern similar to what we report here. Within 2–3 weeks, there were abundant goblet cells on the corneas. Then, in contrast with what we see in the mouse, the goblet cells disappeared over time. Using newly characterized antibodies against K12 and varying the time, corneas were treated with n-heptanol to create the wounds [45, 46], it was shown that injuries to the limbus where transdifferentiation was observed did not remove all the corneal stem cells. Over time, the progeny of those stem cells repopulated the ocular surface with K12+ corneal epithelium.
The data generated by these studies in the 1980–1990s contribute to the LSCH by focusing attention on the LSCs and their ability to resist removal and survive harsh chemical and mechanical insults to permit the re-establishment of the corneal epithelium. In the mouse, we observe K12+ goblet cells on the cornea after wounding but the goblet cells do not disappear, at least not for time periods of up to 8 weeks. They appear without removal of the LSCs but only when the cornea is injured near the limbus. If we reinterpret the conjunctival transdifferentiation studies in light of the fact that LSCs give rise to K8+K12+ goblet cells, the goblet cells that transiently appear on the corneas of rabbits and rats 2–3 weeks after “complete” limbal debridement are likely the progeny of K8+ progenitor cells derived from the LSCs that go on to repopulate corneas with K12+ corneal epithelium.
Figure 7D shows a schematic for the location of the LSCs and the compound niches. Figure 7E shows a proposed model to explain what we believe is taking place on the mouse ocular surface during homeostasis and after wounds that cause goblet cell induction. K8+ LSCs are present within stem cell niches. They give rise to progenitor cells capable of generating corneal epithelial cells and compound niches. These progenitor cells remain at the limbus during normal corneal development and homeostasis and give rise to compound niches infrequently generating fewer than 40 compound niches on the entire mouse cornea. Strains of mice with fewer compound niches still possess K8+ progenitor cells with the potential to generate corneal goblet cells. For instance, despite having fewer and smaller compound niches, C57BL6 mice have more goblet cells on their corneas in response to large wounds compared to BALB/c mice .
After small wounds, the demand on the corneal stem cells is minimal and the number and location of the compound niches at the limbus do not change. By contrast, after large or zonal wounds, compound niches move away from the limbus along with migrating sheets of K12+ corneal epithelial cells. Whether or not compound niches are seen prior to wounding, K8+K12+ goblet cells persist on the cornea. We hypothesize that after wounds near the limbus, K8+ progenitor cells, some of which are localized within compound niches, differentiate into goblet cells with increased frequency (Fig. 7E). The cause for this apparent cell fate change is not clear. Since only large and zonal wounds induce goblet cells on the cornea, we are currently evaluating whether release of NGF and other neurotrophic factors by damaged nerves could play a role. The cornea is densely innervated and the sub-basal nerves are injured when corneal epithelial debridement wounds are created . The nerves enter the cornea at the limbus and are most abundant at this site.
One key assumption we make in interpreting our data is that compound niches are the progeny of stem cells that also give rise to corneal epithelial cells. We base this assumption on (a) the proximity of the compound niche to the LSCs and (b) the fact that K12 is considered a marker for committed corneal epithelial cells. Yet studies have appeared where noncorneal epithelial cells including oral mucosal and dental pulp cells are used to treat corneal epithelial stem cell deficiency . These studies show that noncorneal derived cells can be induced, presumably by local factors present at the cornea, to express K12 . From our data, we cannot exclude the possibility that the compound niche could be derived from a small number of conjunctival progenitor cells retained at the limbal:corneal border during corneal development. If this is the case, then conjunctival progenitor cells are present at the limbus and give rise to K12+ corneal epithelial and goblet cells. While the results of Majo et al.  argue that conjunctival and corneal epithelial progenitor cells coexist on the entire mouse ocular surface, the failure of wounds made to the central cornea to induce K8+K12+ goblet cell expansion suggests that the progenitor cells for the goblet cells reside near the limbus.
The data presented here are derived from mice; we have not shown compound niches in other mammals. The anatomy of the mouse limbus differs from the human limbus [3, 49]. Unlike humans, the mouse limbal epithelium is thinner than the corneal epithelium. In the tissues studied thus far (including conjunctiva, trachea, and gut), terminally differentiated goblet cells are generated from progenitor cells whose basal surface rests on a basement membrane [50, 51]. Our data showing 3D reconstructions of the compound niche at the limbus (Fig. 1D) indicate that this is also the case for the corneal goblet cell. If compound niches are present at the human limbus, they would need to be organized into crypt-like structures so their basal surface could rest on a basement membrane and mucins could be secreted from the apical surface into a lumen. The goblet cells present in the human eyelid within the thickened conjunctival epithelium that make up the lid wiper at the lid margin were recently shown to have this type of organization which the authors refer to as “goblet crypts” . LSCs in the human limbal epithelium are present within limbal crypts [35, 49]. Additional studies will be needed to determine whether compound niches and K12+ goblet cells are located adjacent to the stem cells in the human limbus.
We want to thank numerous colleagues for helpful conversations that contributed to the data contained in this article. Drs. Stuart Yuspa, Maria Morasso, Liowei Li, Anjali Shukla, Jim Zieske, Mary Rose, and A. Sue Menko all contributed to this study at different stages of its development. In addition, the GWU CMIA and its director Dr. Anastas Popratiloff were critical to the success of these studies. This work was funded by NIH Grants EY13559, EY08512, and OD007996 (M.A.S.).
DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST
The authors indicate no potential conflicts of interest.