α9-Integrin–positive basal cells are regionally distributed around the circumference of the adult BALB/c mouse eye (6–8 weeks old) at the limbus, as shown in Figure 1B. The number of α9-integrin–positive cells varies from one quadrant of the eye to the next, such that, at the nasal region, α9-integrin–positive cells can make up as many as 70% of the basal cells, whereas, at the inferior region, fewer than 15% of the cells were α9-integrin positive. The restriction of α9-integrin to the limbus occurs over time during eye development, and the width of the limbal epithelium containing the α9-integrin–positive cells varies by quadrant from a minimum of 150 to a maximum of 300 μm at the inferior and nasal regions, respectively . This pattern is consistent and invariant unless the cornea is subjected to wounds involving removal of >75% of the ocular surface epithelium . In the current study, after first investigating the pattern of expression of β1-, α3-, β4-, and α6-integrins, we identified the α9-, β1-, and β4-integrin expression profiles of the slow-cycling LRCs on the mouse ocular surface.
Unlike α9-Integrin, β1-, β4-, α3-, and α6-Integrin Are More Abundant in the Central Cornea
Each panel in Figure 2 shows a stack of images 10-μm thick and viewed from the basal to apical aspect using the 3D Constructor module of Image Pro Plus software. Each stack was then rotated 90° to show the cross section (CS), indicated by the white dashed lines, presented below each image. Figure 2A and 2B shows the staining for β1-integrin and PI in the limbal basal cells and in the central corneal epithelial basal cells, respectively. The cross-sectional view shows that β1-integrin was present at the basolateral and apical aspects of the basal cells and appeared more abundant in the central cornea than in the limbus. It is interesting to note that β1-integrin was not present immediately under the nuclei of the basal cells. We could confirm that antibody penetration was not responsible for the lack of staining beneath the nuclei because we observed integrin staining around cells of the vasculature deep in the stroma (supplemental online movie).
Figure 2C and 2D shows staining for β4-integrin and PI in the limbus and the central cornea, respectively. β4-integrin forms heterodimers with the α6 subunit, and the staining obtained for α6-integrin was identical to that of β4-integrin (data not shown). Viewing the tissues from their basal aspect showed that, whereas β4-integrin was present at the basement membrane, there was also variability in the amount and intensity of this integrin among individual basal cells. This variability was not easily observed in the cross-sectional view, where the staining seemed to be linear and uniform. β4-integrin was present primarily in the basolateral, apical, and basal aspects of the basal cells, and, viewing cross sections, appeared to be more abundant in the central cornea than at the limbus.
We then assessed the distribution of α3-integrin, the primary β1 family integrin on the mouse ocular surface (Fig. 2E, 2F). We found that the localization of α3-integrin was similar to that observed for β1-integrin; it was primarily present at the lateral aspect of the basal cells and appeared to be more abundant in the cornea than in the limbus. From these data, we concluded that β1-, β4-, and α3-integrins are more abundant in the central and cornea, with their expression decreasing toward the limbus. Even though there was variability observed in the amount and intensity of these integrins within individual cells at the limbus, there were no differences in the regional distribution of these integrins in the manner shown previously for α9-integrin. β1-, β4-, α3-, and α6-integrins were distributed uniformly among the four quadrants of the mouse eye.
Triple-Labeling Corneas for BrdU, Integrin Antibody, and PI Positively Identified LRCs and Their Integrin Profiles In Situ in the Limbal Epithelium
En face images were taken from each of the four different quadrants of each eye with a ×60 oil objective of a confocal microscope as described; each region acquired represented approximately 484 basal cells. Typical data for the α9-integrin and BrdU from two regions of the cornea are shown in images presented in Figure 3. Figure 3A and 3F shows images viewed basally, which were then rotated 90° to show the cross section (Fig. 3B, 3G). Using the 3D Constructor module of Image Pro Plus software, the white rectangular region indicated in Figure 3B and 3G was digitally magnified and is presented in Figure 3C–3E and 3H–3J, respectively, showing the α9-integrin, BrdU, and PI triple-stained tissues in Figure 3C and 3H, the α9-integrin and BrdU double-stained tissues in Figure 3D and 3I, and α9-integrin alone stained tissues in Figure 3E and 3J.
The limbal region presented in Figure 3A–3E shows two α9-integrin–positive cells surrounded by several α9-integrin–negative basal cells. There are two BrdU-positive cells in the limbal basal cell layer, indicated by arrowheads, next to and behind one of the α9-integrin–positive cells. A second cell layer has one BrdU-positive cell as indicated by the arrow. To approximate the boundaries of the BrdU-positive cell nuclei, it was necessary to rotate and use handles, a tool in the Image Pro Plus software that allowed us to move through the stack of optical sections while viewing the PI-stained nuclei and BrdU staining simultaneously.
Although we often observed LRCs in clusters adjacent to patches of α9-integrin–positive cells (Fig. 3C–3E), this was not always the case. Figure 3I and 3J shows a single LRC isolated away from other LRCs; this LRC is α9-integrin–positive and is surrounded by several α9-integrin–positive cells that did not retain BrdU. This cell is shown closely apposed to BrdU-positive stromal fibroblasts (asterisk). To define the boundary between the epithelium and the corneal stroma and the BrdU-positive cells contained there, again, it was necessary to rotate the 3D images and move through the tissue from its apical to basal aspect and observe the abrupt change in organization and orientation of the nuclei of cells within the stroma. Whereas the nuclei of the epithelial basal cells were either cuboidal or columnar oriented perpendicular to the basal cell surface, those of the stromal fibroblasts were oriented parallel to the basal cell surface, as shown by the asterisk shown in Figure 3H (a movie showing how LRCs were identified in the mouse cornea is provided in the online supplemental data). Because of this intimate association between epithelial LRCs and underlying stromal fibroblasts, care had to be taken with each epithelial LRC cell evaluated to carefully assess the border between the epithelial LRCs and the stromal fibroblasts.
Images such as those shown in Figure 3 provide compelling evidence that most α9-integrin–positive cells are not slow-cycling LRCs. However, they also show that there were cells that were both BrdU- and α9-integrin–positive. Not all of the LRCs express the same repertoire of proteins, and some LRCs lose their label faster than others. To better assess the relationship between the LRCs and integrin expression, an experimental approach was required that would be suitable for statistical analysis allowing for assessment of sufficient numbers of cells so that we could get an idea of integrin profiles within populations of slow-cycling limbal basal cells.
Assignment of Scores for Statistical Analyses
As shown in Figure 3, BrdU could be present in either single or multiple discrete patches in the nucleus; only by careful analysis of 3D images and visual inspection of PI-stained nuclei could we approximate the boundaries of the nucleus for each BrdU-positive cell. Integrin staining was restricted to the cytoplasm and cell membranes of cells. Using confocal sections to assign regions of interest and determine pixel intensities was not possible since optimal BrdU and integrin staining were never found in the same confocal layer of these 3D images; rather, 3D images had to be rotated in different planes to obtain optimal staining for integrin and BrdU. For this reason and to allow us to obtain data on populations of cells, a scheme for assignment of scores by visual inspection was developed that minimized subjectivity and maximized the numbers of cells that could be assessed. To test for reliability in score assignments, data were analyzed statistically for consistency from region to region and eye to eye.
The scheme that was developed for assigning BrdU and integrin scores is shown in Figure 4. Each of the 36 regions obtained for each BrdU-integrin pair was analyzed using the 3D Constructor module of Image Pro Plus software, every BrdU-positive epithelial cell within each region was viewed individually, and a value was assigned to it based on both extent and intensity of its BrdU staining. Cells that had retained more BrdU were considered to be more stem-like (slower cycling) than those with less BrdU. BrdU was never found in the central corneal epithelial cells. After BrdU scores were assigned, integrin levels in each of the BrdU-positive cells were assigned a score. Scores assigned to BrdU levels were 0.5, 1, 2, or 3, and scores for integrins were 0, 1, 2, or 3.
BrdU Retention Is Positively Correlated with β1- and β4-Integrin Expression but Negatively Correlated with α9-Integrin Expression
Each eye was divided into four quadrants based on the location of sutures placed on the sclera of the eyes before enucleation. Three regions were randomly sampled from each quadrant (Fig. 1A); a total of 36 regions was assessed per integrin from a minimum of three eyes. Figure 5A–5C shows the number of BrdU-positive cells and their integrin expression profiles based on the assigned scores in eyes stained for α9-integrin (Fig. 5A), β1-integrin (Fig. 5B), and β4-integrin (Fig. 5C). To confirm that the assignment of BrdU scores was performed consistently, we analyzed the data for variations in the distribution of BrdU scores (0.5, 1, 2, 3) across eyes, regions, and quadrants, using the Mixed Linear Model as described. Based on the p-values obtained (eye, .054; region, .08; quadrant, .60), we concluded that the distribution of BrdU was not significantly different across eyes, regions, or quadrants. The consistency with which BrdU scores were assigned among the different eyes, regions, and quadrants analyzed provides further assurance that these data are as accurate as technically possible.
Although most of the cells in each region analyzed were BrdU negative, we were able to identify 1889 limbal epithelial cells as label-retaining. All 1,889 LRCs were scored for both BrdU and integrin expression: 762 for α9-integrin, 715 for β1-integrin, and 412 for β4-integrin. To determine if there was an association between the retention of BrdU within the LRCs and the expression of integrins in these cells, mean integrin scores were determined for each BrdU score level (0.5, 1, 2, and 3), and Pearson correlation coefficients were calculated and tested. All bivariate associations were checked with scatter plots for distributional assumptions. Data showed significant positive correlation for both β1- and β4-integrin expression with BrdU label retention (r = 0.26, p < .0001 and r = 0.24, p < .0001, respectively). For α9-integrin expression, the correlation with BrdU label retention was negative (r = −0.155, p < .0001). For BrdU-α9-integrin-positive cells with a score of 0.5, the mean α9-integrin score was lower than that of the BrdU-α9-integrin-positive cells, with a score of 1. To test whether this apparent nonlinearity was significant, a mixed model statistical approach was used to evaluate for significant quadratic curvature. The p-value for this test was .43; thus, the nonlinear trend of the mean α9-integrin score for cells with BrdU scores of 1 was not statistically significant. The data for the correlations between β1-, β4-, and α9-integrin expression and BrdU retention are presented graphically in Figure 5D. The mean integrin scores for β1- and β4-integrin increase as the BrdU scores increase, whereas the mean integrin scores for α9-integrin increase as the BrdU scores decrease.
The amount of BrdU that each cell retained varied; of all the cells evaluated, there was a nearly equal distribution among the four possible BrdU scores assigned: 501 scored 0.5 (27%), 528 scored 1 (28%), 369 scored 2 (20%), and 491 scored 3 (26%). Because cells with higher BrdU scores had proliferated fewer times after incorporating label, they were considered to be the most stem cell-like; cells with scores of 0.5, although still retaining label, were closer to losing label than those with higher BrdU scores. Although it was possible that a slow-cycling stem cell might have acquired less label during the initial labeling period, it was unlikely that such cells made up a large percentage of the total LRCs counted in this study given the significant positive correlation between the BrdU score and the β1- and β4-integrin scores.
The image shown in Figure 3 suggests that LRCs were often present in close proximity to α9-integrin–positive cells. To assess how likely it was for α9-integrin–positive cells to be near LRCs, we identified and analyzed nine cells with BrdU scores of 3 that were negative for α9-integrin. This was straightforward because 75% of the cells that scored 3 for BrdU expressed no α9-integrin (Fig. 5A). For all nine of the LRCs assessed, we observed at least oneα9-integrin–positive cell within 10–20 μm of the LRC nucleus; for seven of nine LRCs, there was an α9-integrin–positive cell within 12.5 μm of the LRC nucleus. Since six of the nine LRCs assessed were from the superior or inferior region, where α9-integrin–positive cells are rare, these data imply a close relationship between α9-integrin–positive cells and the LRCs at the limbus. In our previous studies, we showed that α9-integrin–positive cells were present at the anatomical niche at the limbus, but we could not prove that the LRCs were found at the same site. Although we now show that the α9-integrin–positive cells were not CESCs, our results show that a close relationship exists between the LRCs and the α9-integrin–positive cells.
To approximate the percentage of cells that were label retaining, first, the total number of cells in each region was estimated directly from images such as those in Figure 4A and 4F; this number was found to be approximately 484 cells per region. Because each region was 167 × 167 μm, this estimate yielded an average cross-sectional cell size of 7.8 μm2. To calculate the total number of cells in all the random regions sampled, 484 was multiplied by 108 (12 regions per eye × 3 integrins × 3 eyes per integrin, or 108 regions) to yield 52,272 cells. Using the total number of estimated BrdU-positive cells from all regions (1,889; Fig. 5A–5C), the percentage of LRCs was calculated to be 3.6%. If only the cells with BrdU scores of 3 were considered to be the most stem cell–like (Fig. 5A–5C), 0.94% of the cells were LRCs. Since we assume that not all of the CESCs were labeled initially with BrdU, this estimate must be considered lower than the value in vivo. Therefore, our data suggest that the percent CESCs at the limbus ranges from 0.94 to 3.6%, which is in accordance with the literature stating that stem cells make up a small percentage of the total cells in the limbus [19–21].