Human Limbal Progenitor Cells Expanded on Intact Amniotic Membrane Ex Vivo FREE

THE STEM CELLS (SCs) of the corneal epithelium are located exclusively at the limbus, the anatomic junction between the cornea and the conjunctiva,¹ and serve as the ultimate source for corneal epithelial regeneration under normal and injured conditions.^{2 - 3} These cells were initially identified in the entire basal layer of the limbal epithelium by the lack of cornea-specific K3 keratin expression.¹ Subsequently, a fraction of these limbal-basal epithelial cells were found to have a long cell cycle⁴ and high clonogenicity,^{5 - 6} ie, general characteristics of SCs. When limbal SCs proliferate, they self-renew and/or give rise to transient amplifying cells (TACs) located in the corneal basal epithelium. Unlike SCs, TACs have a short cell cycle and a limited proliferative capacity (ie, a shorter lifespan). The mechanism governing the balance between SC self-renewal and SC differentiation into TACs remains unclear. One explanation of how SCs in general maintain their "stemness" is the fact that they are located in a microenvironmental"niche." Within such a niche, SCs lack gap junction intercellular communication(GJIC), which enables SCs to be sequestered from more differentiated TACs.^{7 - 8}

Gap junctions are specialized cell membrane structures forming intercellular channels that are composed of a variety of transmembrane proteins (polypeptides) called connexins (Cx).^{9 - 10} Gap junctions play an important role in direct cell-cell communication, which affects cell proliferation, differentiation, and apoptosis.^{9 - 13}

Connexin 43 and Cx50 are the only 2 Cx found in the human ocular surface epithelium so far.^{14 - 16} Under normal atraumatic conditions, expression of Cx43 is noted in the basal cell layer of the human corneal but not limbal epithelium, suggesting that the expression of Cx43 de notes the differentiation of SCs into corneal TACs. Wolosin and coworkers¹⁶ proposed that the apparent incongruity of Cx expression may endow limbal epithelial SCs with the property of stemness in this microenvironmental niche so that they can be segregated from further differentiated TACs.

When limbal epithelial SCs are partly or totally destroyed, the corneal surface will invariably be covered by the migrating conjunctival epithelium, a pathologic entity found in a number of ocular surface disorders.¹⁷ Clinical transplantation of limbal epithelial SCs from an autologous or allogeneic source is necessary to restore vision and a normal corneal surface.^{18 - 19} Transplantation of preserved intact amniotic membrane alone has recently been shown to restore such damaged corneal surfaces in patients with partial limbal stem cell deficiency, ie, the limbus has been partially destroyed.^{20 - 21} This result suggests that transplanted amniotic membrane helps expand residual limbal epithelial SCs in vivo. Promising results of transplanting limbal epithelial SCs expanded on amniotic membrane in culture have recently been reported for treating partial or total limbal SC deficiency in human patients.^{22 - 26} These findings prompt us to examine the hypothesis that amniotic membrane may help maintain and expand limbal epithelial SCs ex vivo by serving as a substrate mimicking their microenvironmental niche. Herein we provide experimental evidence that limbal epithelial cells ex vivo expanded on intact amniotic membrane are indeed largely devoid of Cx43 expression, lack GJIC, and are slow cycling. On xenotransplantation into nude mice, these expanded cells yield a stratified epithelium whose basal layer remains negative to Cx43 and K3 keratin expression and retained bromodeoxyuridine (BrdU) labels, resembling their in vivo counterpart.

Human tissue was handled according to the Declaration of Helsinki. Corneoscleral tissue from human donor eyes was obtained from the Florida Lions Eye Bank, Miami, directly after the central corneal button had been used for corneal transplantation. The tissue was rinsed 3 times with Dulbecco modified Eagle medium (GIBCO BRL, Grand Island, NY) containing 50-mg/mL gentamicin (GIBCO BRL) and 1.25-mg/mL amphotericin B (GIBCO BRL). After careful removal of excessive sclera, iris, and corneal endothelium, the remaining tissue was placed in a culture dish and exposed to dispase II (1.2 U/mL in magnesium- and calcium-free Hank balanced saline solution [GIBCO BRL]) at 37°C under humidified 5% carbon dioxide for 5 to 10 minutes. After 1 rinse with Dulbecco modified Eagle medium containing 10% fetal bovine serum (GIBCO BRL), the scleral rim was trimmed to obtain limbal tissue cubes of approximately 1 × 1.5 × 2.5 mm.

Amniotic membrane with the epithelial side facing up was fastened onto a culture insert (Milipore Corp, Bedford, Mass) as previously reported.²⁷ On the center of the amniotic membrane an explant was placed and cultured in a medium made of an equal volume of HEPES-buffered Dulbecco modified Eagle medium (GIBCO BRL) containing bicarbonate and Ham F12 (GIBCO BRL). The medium was supplemented with 0.5% dimethylsulfoxide (Sigma-Aldrich Corp, St Louis, Mo), 2-ng/mL mouse epidermal growth factor (Sigma-Aldrich Corp), 5-µg/mL insulin, 5-µg/mL transferrin, 5-ng/mL selenium(Sigma-Aldrich Corp), 0.5-µg/mL hydrocortisone (Sigma-Aldrich Corp), 30-ng/mL cholera toxin A subunit (Sigma-Aldrich Corp), 5% fetal bovine serum(GIBCO BRL), 50-µg/mL gentamicin (GIBCO BRL), and 1.25-µg/mL amphotericin B. Cultures were incubated at 37°C under 5% carbon dioxide and 95% air, and the medium was changed every 2 to 3 days. When human limbal epithelium cultures almost reached confluent growth, they were subjected to qualitative dye transfer assay or incubated with 10µM BrdU (Boehringer-Mannheim Corp, Indianapolis, Ind) for 24 hours and fixed in cold methanol for immunostaining.

All procedures were performed according to the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research. Details of this technique have been described previously.²⁸ Briefly, nearly confluent human limbal epithelial cultures were labeled with BrdU for 7 days and transplanted to the subcutaneous plane of the abdomen of NIH-bg-nu-xidBR mice. After 5 days, mice were killed and the tissue including implanted amniotic membrane was removed and embedded in optimal cutting temperature compound for cryosectioning. A total of 6 cultures was transplanted.

Immunostaining was performed as previously described.²⁸ Briefly, frozen sections were fixed and preincubated with 5% bovine serum albumin (Sigma-Aldrich Corp) to block nonspecific staining. Sections were incubated with a mouse anti-Cx43 (1:200) (Chemicon International Inc, Temecula, Calif), AE-5 (anti-K3) (1:100) (ICN Pharmaceuticals, Costa Mesa, Calif), or anti-BrdU (1:1000) (Boehringer-Mannheim Corp) monoclonal antibody for 45 minutes followed by a fluorescein isothiocyanate–conjugated secondary antibody(goat anti–mouse IgG at 1:200) (Sigma-Aldrich Corp), mounted with an antifade solution (Vectashield; Vector Laboratories, Burlingame, Calif), and analyzed with a fluorescence microscope (Axiophot; Carl Zeiss Inc, Oberkochen, Germany).

For BrdU and Cx43 double labeling, confluent cultures were incubated with 10µM BrdU in the same culture medium for 24 hours. These cultures on amniotic membrane were prepared as flat-mount samples. After samples were treated with 2N hydrochloric acid at 37°C for 45 minutes and neutralized in boric acid (pH 8.5), incorporated BrdU and Cx43 expression were detected by immunostaining with a mouse anti–BrdU antibody (1:1000) and a mouse anti–Cx43 antibody (1:200) followed by a diaminobenzidine-peroxidase staining protocol (Vectastain Elite Kit; Vector Laboratories). Samples were counterstained with hematoxylin. Under magnification of ×400, positive nuclei were counted among the total nuclei within the entire field, and a total of 16 fields (within the major outgrowth area) were counted per specimen. The labeling index for BrdU was expressed as the number of positive-labeled nuclei divided by the number of all nuclei multiplied by 100%. We defined 1 U of Cx43 expression as all cells in one ×400 field expressing Cx43. That is, 0.5 U was defined as 50% of cells expressing Cx43. We counted 100 fields per sample for a total of 5 samples and reported their mean and SD.

We used the scrape-loading dye transfer assay originally described by El-Fouly et al^{29 - 30} and discussed further by Trosko et al.³⁰ For positive control we cultured human limbal epithelium from an explant on plastic dishes for 14 days. Human limbal epithelium on plastic or amniotic membrane was rinsed with sterile phosphate-buffered saline. One milliliter of lucifer yellow plus rhodamine-dextran (0.5 mg/mL) in phosphate-buffered saline was added to the culture dish. A sterile scalpel blade was applied with gentle pressure to cut the cells. Six scrape lines were placed in different areas per culture. Dishes were left in a dark room for 3 minutes. Cells were rinsed extensively with phosphate-buffered saline to prevent high background fluorescence. Cultures were fixed in 4% formalin and epifluorescence was examined with a microscope(Axiophot; Carl Zeiss Inc) equipped with a UV light source. A rhodamine filter set was used to identify red fluorescence of the primary loaded cells along the scrape line (absorbency, 555 nm; emission, 580 nm). Fluorescence filter sets were used to detect green fluorescence of lucifer yellow, which was transferred through gap junctions (absorbency, 428 nm; emission, 536 nm). We analyzed a total number of 18 scrape lines (6 scrape lines per culture for 3 separate cultures). The percentage of the entire length of all 6 scrape lines per culture was measured where we observed dye transfer in more than 4 cell rows away from the initial loaded cells.

The limbus, ie, the transitional zone between human conjunctiva and cornea, had a multilayered epithelium with the basal layer arranged in a palisade pattern (Figure 1A). The limbal epithelium was situated on top of a loose and vascular connective tissue (Figure 1A, arrowheads), while the corneal epithelium lay on top of the dense Bowman layer (ie, a thick basement membrane; Figure 1A, black arrows) with a subjacent dense avascular stroma. Immunostaining showed that the expression of Cx43 was absent in the basal layer, but positive in the suprabasal layers, of the limbal epithelium (Figure 1B and C). In central cornea sections, Cx43 was predominantly expressed in the basal layer and with a less intensity in more superficial layers (Figure 1D). This finding confirmed that reported by Matic et al¹⁴ and Wolosin and coworkers.¹⁶

Human limbal epithelium outgrowth was detected after 1 week from the border of a limbal explant (Figure 2A) and reached confluence (ie, 22 mm in diameter) after an average duration of 3 to 4 weeks (Figure 2B, nearly reaching the insert edge). The outgrowth consisted of a sheet of small, compact, and uniform cells with an approximate 1:1 nucleus-cytoplasm ratio in the majority of cells (Figure 2C); the leading edge of the outgrowth built a bulge consisting of both limbal epithelial cells and amniotic epithelial cells (Figure 2B, arrows). On cross section, the majority of the outgrowth consisted of a monolayer of cuboidal cells (Figure 1 2D, black arrows) growing on top of amniotic epithelial cell debris (Figure 2D, black arrowheads).

After 3 weeks of culturing on amniotic membrane, the majority of expanded human limbal epithelium did not express Cx43 regardless of whether the final outgrowth area was 70%, 90%, or confluent (Figure 3A). Cells expressing Cx43 were calculated as 12.4% ± 14.5% positive units and were found in focal areas predominantly adjacent to the explant and randomly scattered among the outgrowth area (Figure 3B). Positive Cx43 staining appeared in a punctate pattern confined to the cell membrane of adjacent cells, compatible with the formation of gap-junction channels (Figure 3B, inset). To correlate Cx43 expression with the proliferative activity at the same time, we labeled the S-phase of the cell cycle with BrdU, a thymidine analogue, for 24 hours in nearly confluent cultures. The labeling index was low, in the range of 2.4% ± 0.9% (n = 5) (Figure 3A and C, arrows). Areas with high BrdU uptake were found predominantly near the explant or at the leading edge of the outgrowth and were devoid of Cx43 expression (Figure 3C, inset). To confirm that the nonlabeled cells are indeed slow cycling and not postmitotic differentiated cells, we continuously incubated a set of 6 cultures with BrdU for 5 days. As shown in Figure 3D, the BrdU labeling index increased to 62.0% ± 9.5% (n = 6). This result indicated that the majority of the expanded human limbal epithelium on amniotic membrane was indeed slow cycling.

To evaluate whether immunohistochemically detected Cx43 was indeed assembled into functioning gap-junction channels, we performed a qualitative dye transfer assay using a scrape-loading technique previously described.²⁹ Human limbal epithelium expanded on intact amniotic membrane did not show any dye transfer from the scraped area to the adjacent cells in most (83%) of the scrape lines performed (n = 6 per sample) (Figure 4A and B). These areas did not express Cx43 after counterimmunostaining(Figure 4C). In a total of 18 scrape lines (3 cultures and 6 scrapes per sample) we found 3 patches, representing 17% of the entire length of all scrape lines, of dye transfer to neighboring cells (Figure 4D and E). These areas were also found to express Cx43 when subsequently counterimmunostained (Figure 4F, arrows). As a positive control, we scrape-loaded the outgrowth of human limbal epithelium grown on plastic and found marked dye transfer to adjacent cells in 94% of the entire length of all 18 scrape lines (Figure 4G and H). This difference was statistically significant (P<.002, analysis of variance).

We transplanted amniotic membrane with expanded human limbal epithelium as a composite graft (n = 6) into the subcutaneous plane of NIH-bg-nu-xidBR mice after 7 days of continuous BrdU labeling. Five days later, the resultant epithelium was stratified to an average of 5 cell layers. Basal cells were small and compact, whereas superficial cells appear more flat and squamous(Figure 5A). Expression of Cx43 was absent throughout the entire epithelium (Figure 5B). Within the same section, a positive control could be found in the mouse epidermis, which expressed Cx43 in large amounts (Figure 5B, inset). As basal cells did not express Cx43, this phenotype resembled a limbal basal epithelial cell phenotype in vivo; thus, we examined the expression of K3 keratin, which has been reported to be absent also in the limbal basal epithelium.¹ Our result showed that K3 keratin was indeed absent in the basal layer but markedly expressed in the suprabasal and superficial layers of the resultant stratified epithelium (Figure 5C). Label-retaining (BrdU-positive) cells were exclusively identified in the basal layer in direct contact with the underlying amniotic membrane (Figure 5D).

In this study, we provide experimental evidence to support the hypothesis that intact amniotic membrane preferentially preserves and expands human limbal epithelial progenitor cells. After stratification in nude mice, the basal layer of the resultant epithelium was devoid of Cx43 and K3 keratin expression and retained a fraction of slow-cycling cells, resembling the phenotype of the SC-containing human limbal basal epithelium in vivo. In the aggregate, these data support the notion that amniotic membrane mimics the in vivo stromal niche to maintain SC characteristics. Collectively, these data also explain why amniotic membrane and ex vivo expanded limbal epithelium have been successful as a new surgical strategy to reconstruct the corneal surface in patients with limbal SC deficiency.^{22 ,24 - 26 ,31}

Matic et al¹⁴ and Wolosin and coworkers¹⁶ proposed the theory that noncommunication of limbal basal epithelial cells in vivo is one feature of the microenvironmental niche in which limbal SCs lie. We observed only 12.4% ± 14.5% of Cx43-positive units in our expanded cell population after culturing for an average of 3 to 4 weeks on intact amniotic membrane (Figure 3A and B). We further proved that the lack of Cx43 expression indeed reflected the lack of gap junction formation and GJIC by means of the well-established scrape-loading dye transfer technique (Figure 4).^{29 - 30} Except for 3 localized areas composing up to 17% of the total length of all 18 scrape lines, the majority of human limbal epithelium (83%) expanded on intact amniotic membrane showed no GJIC. This value was significantly less than that observed for human limbal epithelium on plastic (17 [94%] of 18 scrape lines) (Figure 4).

We noted that the lack of Cx43 expression was neither dependent on the stage of confluence nor restricted to the edge of the outgrowth. Wolosin and coworkers¹⁶ found Cx43 expression and GJIC in 9-day-old rabbit limbal epithelial cells cultured on a 3T3 fibroblast feeder layer, and a shift from Cx43 to Cx50 expression after raising the culture to the air-liquid interface to promote stratification. The fact that most of our culture was actually devoid of Cx43 expression even after 3 weeks of culturing underscores the striking difference between the amniotic membrane culture and the 3T3 fibroblast feeder layer system.

It has been reported that gap junctions and Cx expression are dramatically decreased in late G₁ and S phases and reappear throughout the rest of the cell cycle in regenerating hepatocyte cultures, and it has been suggested that the down-regulation of GJIC might be an effective way to allow cell division without interfering with the homeostatic balance within the nonproliferative cell population.^{32 - 33} By the use of BrdU labeling to identify rapid-cycling cells, we noted that Cx43 was not expressed by both BrdU-labeled and nonlabeled cells (Figure 3A). Therefore, we ruled out that the lack of Cx43 expression was a result of rapid proliferation. We actually noted that the labeling index was overall low, in the range of 2.4% ± 0.9% after 24 hours of BrdU labeling (Figure 3C). To further prove that such a low labeling index was not caused by terminal differentiation, we continuously labeled these cells for 6 days and found an approximately 30-fold increase of the labeling index, ie, to an average of 62% ± 9.5% (Figure 3D). Collectively, these data confirmed our assertion that the amniotic membrane culture system predominantly maintains and expands slow-cycling human limbal epithelium.

The data discussed so far were obtained from cultured human limbal epithelium monolayers. To further investigate the phenotype of expanded human limbal epithelium, we performed xenotransplantation of nearly confluent human limbal epithelium on intact amniotic membrane to promote epithelial stratification. This time we labeled human limbal epithelium on amniotic membrane continuously for 7 days before xenotransplantation to identify slow-cycling, label-retaining cells after a chasing period of 5 days. The resultant stratified epithelium was devoid of Cx43 expression throughout all layers (Figure 5B), was devoid of keratin K3 expression in the basal layer(Figure 5C), and retained BrdU labels in the basal layer (Figure 5D). All of these characteristics are found in the SC-containing basal limbal epithelium in vivo.

In the present study we used intact amniotic membrane and found out that most expanded human limbal epithelium was growing on top of devitalized amniotic epithelial cells without direct contact with the underlying basement membrane (Figure 2). This finding was consistent with what has been recently reported by Koizumi et al³⁴ with the use of rabbit limbal epithelial cells. Our further study indicated that the separation of expanded human limbal epithelial cells by amniotic epithelial cells was important to maintain such a phenotype without Cx43 expression and GJIC, because denudation of amniotic epithelium to expose amniotic basement membrane will promote a corneal epithelial phenotype.²⁸ Future studies are also needed to elucidate the mechanism by which the cell contact with the basement membrane may affect the cell cycle and the expression of differentiation markers, imitating the differentiation of limbal SCs to corneal TACs, of which the latter lie on a thick corneal basement membrane. Furthermore, investigation into culturing conditions that may optimize limbal epithelial SC expansion on amniotic membrane should be fruitful for devising a clinical protocol of this new surgical procedure for treating limbal stem cell deficiency.

Submitted for publication July 17, 2001; final revision received February 3, 2002; accepted February 28, 2002.

This study was supported in part by Public Health Service Research Grant EY 06819 from the Department of Health and Human Services, National Eye Institute, National Institutes of Health, Bethesda, Md (Dr Tseng); in part by an unrestricted grant from Research to Prevent Blindness Inc, New York, NY; and in part by research fellowship grant GR 1814/1-1 from the Deutsche Forschungsgemeinschaft, Bonn, Germany (Dr Grueterich).

Preserved human amniotic membrane was kindly provided by Bio-Tissue(Miami, Fla).

Corresponding author: Scheffer C. G. Tseng, MD, PhD, Ocular Surface Center and Ocular Surface Research & Education Foundation, 8780 SW 92nd St, Miami, FL 33176 (e-mail: stseng@ocularsurface.com).