Three-Dimensional Neuroepithelial Culture from Human Embryonic Stem Cells and Its Use for Quantitative Conversion to Retinal Pigment Epithelium

A goal in human embryonic stem cell (hESC) research is the faithful differentiation to given cell types such as neural lineages. During embryonic development, a basement membrane surrounds the neural plate that forms a tight, apico-basolaterally polarized epithelium before closing to form a neural tube with a single lumen. Here we show that the three-dimensional epithelial cyst culture of hESCs in Matrigel combined with neural induction results in a quantitative conversion into neuroepithelial cysts containing a single lumen. Cells attain a defined neuroepithelial identity by 5 days. The neuroepithelial cysts naturally generate retinal epithelium, in part due to IGF-1/insulin signaling. We demonstrate the utility of this epithelial culture approach by achieving a quantitative production of retinal pigment epithelial (RPE) cells from hESCs within 30 days. Direct transplantation of this RPE into a rat model of retinal degeneration without any selection or expansion of the cells results in the formation of a donor-derived RPE monolayer that rescues photoreceptor cells. The cyst method for neuroepithelial differentiation of pluripotent stem cells is not only of importance for RPE generation but will also be relevant to the production of other neuronal cell types and for reconstituting complex patterning events from three-dimensional neuroepithelia.


Introduction
The nervous system develops via formation of the neural tube -a pseudostratified epithelium with the apical cell surface oriented toward a single, central lumen. To reconstitute efficient and faithful differentiation of pluripotent cells toward neural cell types, we sought to reconstitute from pluripotent stem cells in vitro organized, three-dimensional neuroepithelial structures with a single lumen. We aimed to mimic the in vivo extracellular matrix environment by applying cell culture methods originally developed for polarized mammary and kidney epithelial cysts [1,2]. Methods for differentiating embryonic stem cells (ESCs) via embryoid bodies or cell aggregates that undergo self-organization have made ground-breaking, fundamental contributions to our understanding of cellular differentiation and represent a valuable source of hardto-obtain, differentiated cell types [3,4,5,6,7]. Such cultures often contain multiple cell types, so the desired target cell type must somehow be selected. Current two-dimensional (2D) conditions, on the other hand, can only generate homogeneous populations of only very few neuronal cell types [8].
Several degenerative diseases of the retina and the retinal pigment epithelium (RPE) have been characterized where transplantation may realistically ameliorate symptoms [3,9,10,11,12,13]. Current methods to differentiate pluripotent cells to RPE include 2D differentiation of primate ESCs, spontaneous differentiation of colonies in human ESC cultures, as well as several floating aggregate methods using mouse and human ESCs (Table S1) [3,4,5,[14][15][16][17][18][19][20]. While several of these recent methods have significantly improved yield and accelerated differentiation, all methods to date result in a mixture of RPE cells and neural retina cells, thus requiring selection prior to RPE transplantation. The only method for RPE selection so far described has been manual picking and expanding pigmented colonies, limiting the relevance for large scale screening approaches and timely transplantation.
Here we demonstrate that a three-dimensional (3D) epithelial cyst culture of human pluripotent stem cells leads to the induction of polarized neuroepithelia within 5 days. This approach reconstitutes the 3D architecture of embryonic pseudostratified epithelium and the formation of a single lumen. We demonstrate the utility of this system by achieving quantitative production of RPE cells from human ESCs within 30 days. Direct transplantation of this RPE into a rat model of retinal degeneration without any selection and further expansion of the cells results in the integration of a RPE monolayer that rescues degeneration of the outer nuclear layer. Our work highlights how considering the cell biological context of pluripotent stem cells while culturing can significantly improve differentiation and the subsequent efficacy of therapeutic outcomes.

hESC-derived cysts are composed of polarized neural progenitors
To induce hESCs to faithfully reproduce neuroepithelial cell architecture, we embedded hESC clusters in the proteinaceous matrix Matrigel that had been reported to support 3D epithelial cyst formation [21] in the presence of the neural induction medium N2B27 ( Figure 1A) [22]. Within 24 hours, all hESC clumps organized into neural tube-like structures with a smooth basal edge and an apical lumen (Movie S1, Figure 1B). We refer to these pseudostratified neuroepithelial structures as cysts, since they have a single lumen, similar to kidney and mammary epithelial cysts. The presence of a single luminal structure also distinguishes them from embryoid bodies or floating aggregates [3,4,23,24]. Immunostaining of whole-mount preparations followed by confocal microscopy showed that hESC-derived cysts were composed of polarized neural progenitors. At Day 1, all formed cysts expressed pluripotency-related genes OCT4, SSEA4 and SOX2, as well as neural lineage-associated gene NESTIN ( Figure 1C). The neuroectodermal marker PAX6 was gradually up-regulated so that it was strongly expressed by Day 5 in nearly all cysts, confirming a neural lineage ( Figure 1D). The apicobasal polarity of the cysts was reflected by the apically restricted distribution of CD133 and ZO-1 at apical junctions from Day 1 on ( Figure 1C and 1D). Apicobasal polarity was further demonstrated by luminal cell mitosis ( Figure 1E), which is an important feature of polarized neuroepithelial cells. These observations indicate that the Matrigel-based 3D cyst model supports efficient generation of polarized neural progenitors from hESCs within 5 days.

hESC-derived cysts acquire retinal identity
We next identified the regional identity of hESC-derived cysts. During 2D in vitro hESC differentiation, hESC-derived neuroectodermal cells adopt anterior neuroepithelial characteristics in the absence of extrinsic patterning cues [24,25]. Consistent with an anterior phenotype observed in other protocols, the cysts were BF-1 and PAX2 positive, as well as HOX B1 and HOX C5 negative ( Figure 2A). Since expression of BF-1 and PAX2 suggests a possible retinal identity [26,27], we examined the expression of eye field transcription factors (EFTFs), including PAX6, RX and SIX3 [28]. All three transcripts were up-regulated from Day 5 ( Figure 2A). When cysts were further cultured in Matrigel, the RPE-specific transcription factor MITF and the neural retina-associated transcription factor CHX10 were strongly expressed from Day 10, suggesting that by Day 10 the cultures represented a mixture of RPE and neural retina-biased cells. Weak expression of CRX, which is highly expressed in photoreceptor precursors, was also detected from Day 10 ( Figure 2A). Taken together, human ESCderived cysts formed in Matrigel supplemented with neural induction media naturally express retinal progenitor genes as a population.
To quantitatively assess the efficiency of retinal induction within our cyst cultures, we immunostained the cysts for the optic vesicle stage-related markers PAX6, RX, OTX2 and the optic cup stagerelated marker CHX10 ( Figure 2B and 2C). The specificity of each antibody was verified on cryosections of appropriately staged embryonic mice ( Figure S1). By Day 5, nearly all of the human ESC-derived cysts were PAX6 + (98.063.3%), OTX2 + (96.363.2%), and RX + (91.967.7%) (means6SD, n independent experiment = 3, more than 100 cysts counted for each experiment using H9 cell line). Within the positive cysts, a nearly uniform expression of PAX6, OTX2, and RX was observed. CHX10 was rarely expressed in Day 5 cysts ( Figure 2B). At Day 15, most of the cysts maintained expression of PAX6, RX, and OTX2 while also becoming CHX10 + (61.964.6%) (means6SD, n independent experiment = 3, more than 100 cysts counted for each experiment using H9 cell line) ( Figure 2C). To confirm the results of the antibody staining, we performed in situ hybridization to detect the corresponding messenger RNA of cryosectioned cysts. These cryosectioned cysts showed strong signals for PAX6 and RX, especially apically, ( Figure S2A) suggesting that the stronger basal signal in the whole-mount immunofluorescence of large cysts ( Figure 2B and 2C) might be due to the hindered penetrance of the antibodies into the center of the cyst.
We conclude from these results that the vast majority of hESCderived cysts at Day 5 in the 3D differentiation model acquire an optic vesicle cell phenotype. Different batches of Matrigel may vary in the efficiency of generating optic vesicle stage cells. We tested three different lots of Matrigel, two of which showed no significant difference in generating PAX6 + RX + OTX2 + cysts. The third lot of Matrigel showed weaker PAX6 induction in formed cysts.

Retinal cyst formation involves IGF-1/insulin signaling
Our results indicate that growth in Matrigel alone with N2B27 medium is sufficient to induce retinal neuroepithelium from hESCs. Matrigel represents a mixture of not only extracellular matrix components but also soluble factors. In previous 2D induction protocols, IGF-1 (10 ng/ml) contributed to directing hESCs to a retinal progenitor identity [7]. Our Matrigel-based 3D model could potentially favor the retinal differentiation of hESCs due to the levels of IGF-1 reported to be in Matrigel (15.6 ng/ml) (www.bdbiosciences. com/external_files/dl/doc/manuals/live/web_enabled/356230_ Guidelines.pdf). To test the possible role of IGF-1, we added the pharmacological inhibitor, AG 1024 that inhibits insulin receptor kinase and insulin-like growth factor-1 receptor kinase signaling, to the N2B27 medium at Day 0. In the presence of AG 1024, the percent of PAX6 + cysts at Day 5 was reduced from 98% to 33.963.3%. The PAX6 2 cysts failed to achieve a pseudostratified architecture, and instead formed single cell layered epithelial cysts ( Figure 2D and 2E). In addition, the expression of RX and OTX2 was also extremely low or absent in these PAX6 2 cysts. ZO-1 expression was still mainly seen at the apical-lateral border of the cysts indicating that cell polarity was preserved ( Figure 2E). These results suggest that IGF-1/insulin signaling is required for the formation of a pseudostratified, retinal neuroepithelium during hESC differentiation in Matrigel.

Quantitative differentiation of polarized retinal progenitors to RPE or neural retina
Given the high efficiency with which we could induce retinal progenitor cells, we sought to quantitatively differentiate these to RPE, one of the earliest differentiated retinal cell types during retinogenesis. Turning to another classical epithelial culturing method, we dispersed Day 5 cysts and plated them onto transwell filters coated with growth factor-reduced-Matrigel to obtain polarized epithelial cells under 2D culture conditions ( Figure 3A). The culture medium was changed from neural induction medium (N2B27) to a standard RPE-supporting medium containing knockout serum replacement. In the presence of 100 ng/ml Activin A, a known inducer of RPE fate [29], we observed the appearance of pigmented cells starting at Day 18 and the characteristic polygonal cell shape emerging by Day 25. By Day 30, 95.761.0% of the cells were pigmented, whereas in the absence of Activin A only isolated colonies of pigmented cells appeared ( Figure 3B and white arrows). To characterize the Activin A-dependent differentiation process, we performed a timecourse analysis of some key features during RPE development ( Figure 3C). Expression of PAX6 remained stable throughout the RPE differentiation process in vitro. The RPE-specific marker MITF started to be expressed in the majority of the cells by 22 days ( Figure 3C). Importantly, Activin A-dependent RPE induction was highly sensitive to the initial cell seeding density on transwell filters ( Figure S3A).
We assessed the differentiation of the RPE sheets post-Day 30 using a variety of molecular, ultrastructural, and functional assays. Several indicators showed that the cultures formed a tight epithelial barrier including robust ZO-1 immunofluorescence ( Figure 4A). Ultrastructural analysis at Day 50 confirmed the presence of tight junctions, desmosomes, microvilli and abundant melanin granules enriched toward the apical surface, which displayed the characteristic dome shape ( Figure 4B-4D). To functionally test the epithelial barrier function, we measured the transepithelial resistance (TER) which gradually increased to a TER of 31364 V?cm 22 at Day 50 ( Figure 4E). This value is within the upper range of TER measured from cultured human fetal RPE (150 and 330 V?cm 22 ) [30,31], suggesting that the epithelial barrier function has been established. To further characterize RPE differentiation, we immunostained at Day 40 for mature RPE cell markers RPE65 and BESTROPHIN ( Figure 4F). Finally, we confirmed that the proliferating RPE progenitor cells exited the cell cycle over time based on 5-ethynyl-29-deoxyuridine (EdU) uptake at representative time points. As shown in Figure 4G, the vast majority of pigmented cells were negative for EdU by Day 30. We asked whether the Day 5 retinal progenitor cells could alternatively be differentiated into neural retina cells with similar efficiency and speed. Plating of cysts onto transwell filters and further culturing in neural induction medium rather than RPE medium ( Figure S4A) resulted in a uniform culture of CHX10 + PAX6 + cells where .99% of cells were CHX10 + PAX6 + by 20 days ( Figure S4B). The expression of CRX on protein level was not detectable by 20 days, although weak CRX expression was detected in RT-PCR ( Figure 2A) by 10 days of differentiation. By 40 days post induction, rosette-like structures formed on the transwell filters with the appearance of a small number of CRX + cells. By 60 days, cells had self-organized into stratified, rosette-like structures with a high percentage of CRX + cells surrounded by CHX10 + PAX6 +weak and CHX10 2 PAX6 +strong cells, suggesting their neural retina cell identity. This stratified structure resembled the layered structure seen in the intact mouse retina ( Figure S4C).
These findings demonstrate that the sequential 3D and 2D epithelial culture of hESCs efficiently directs them toward RPE or neural retina depending on the media conditions applied. From the same hESC-derived cysts we have generated either pigmented cells that have morphological properties highly characteristic of native RPE cells and form a confluent RPE cell layer without the need for clonal expansion or neural retina cultures that have a high percentage of CRX + cells.

Phagocytic function of hESC-derived RPE cells
One of the most important functions of RPE cells in vivo is phagocytosis of shed photoreceptor outer segments, supporting the daily renewal of photoreceptor cells [32]. To analyze whether our hESC-derived RPE cells phagocytose shed outer segment disks, we endeavored to recapitulate the in vivo photoreceptor/RPE interaction by co-culturing the hESC-derived RPE cells with neural retinal explants from transgenic mice expressing human RHODOPSIN-GFP [33] (Figure 5A), similar to a previously described co-culture system [9]. We prefer this system over using purified outer segments or latex bead incubation assays where it can be difficult to distinguish truly endocytosed disks from material adherent to the surface. After 24 hours of co-culture, the hESCderived RPE cells were directly juxtaposed with the mouse retina cells ( Figure 5B). Immunofluorescence to specifically detect human nuclei in the pigmented cell layer excluded the possibility of contamination from mouse RPE cells. Abundant GFP + outer segment structures were observed within the pigmented cells ( Figure 5C, Movie S2). Immunostaining of co-cultures to detect the early endosome marker EEA1 specifically in human cells showed EEA1 signal surrounding GFP + particles, suggesting that outer segments had been truly phagocytosed into the RPE cells ( Figure 5D).

Differentiation of RPE cells from other human pluripotent stem cell lines
The bulk of our analysis was performed on the hESC line H9, but we confirmed that our hESC-to-RPE protocol also worked efficiently on different human pluripotent cell lines. The data represented in Figures 1, 2, 3, and 4 were reproducible with the H1 cell line, except for the requirement of exogenous Activin A, as described below. In our H9 work, the addition of Activin A significantly enhanced the appearance of pigmented cells ( Figure 3B) and efficient RPE induction was highly sensitive to cell seeding density in the presence of Activin A ( Figure S3A). When using the H1 cell line, the RPE differentiation efficiency and timeline was very similar compared to that of the H9, but exogenous Activin A was not required for RPE differentiation. However, the pharmacological Activin/TGF-b inhibitor SB 431542 completely blocked pigmentation in H1 cells, indicating that Activin-like signaling is indeed required and is presumably made endogenously during RPE differentiation from H1 cells ( Figure S3B).
We also determined if our method was effective on the human induced pluripotent stem cell (iPSC) line O27-08, which was reprogrammed from human foetal neural stem cells (CB660) [34]. The timescale of cyst formation and RPE differentiation were indistinguishable from those of hESCs ( Figure S5A and S5B). During differentiation, immunostaining revealed that the human iPSC-derived cysts were also positive for retinal progenitor markers PAX6, RX and OTX2, with the tight junction marker ZO-1 toward the apical side ( Figure S5C). However, we found Pax6 expression in O27-08 cell-derived cysts was weaker compared to that in hESC-derived cysts. This coincides with previous reports in which considerable variation was found in the ability of different iPSC cell lines to differentiate into PAX6 + neuroectodermal cells [16,35]. Although the percentage of pigmented cells differentiated from O27-08 cells was lower compared to the H9 and H1 hESCs, we obtained a relatively high percentage of RPE cells in the 30-day timescale (48.066.2% pigmented cells), still representing a significant improvement to other RPE protocols using ESCs ( Figure S5B, Table S1). It is likely that boosting PAX6 induction by other exogenous factors, or using iPSC cell lines that are biased toward RPE formation [36] might rescue this lower efficiency.
Collectively, these results demonstrate that the optic vesicle stage cells can be derived from different human pluripotent cell lines and can be further directed toward RPE cells with exogenous or endogenous Activin A signals.

In vivo integration and function of transplanted RPE
We then determined if our RPE could be directly transplanted without manual selection into Royal College of Surgeons (RCS) rats to rescue retinal degeneration in vivo. RCS rats are the primary animal model currently used for RPE replacement studies. The rats are characterized by a phagocytosis defect in the RPE due to a mutation in the receptor tyrosine kinase Mertk gene [37]. This defect leads to an enrichment of outer segment fragments in the sub-retinal space. Accumulation of debris causes degeneration of photoreceptors with an almost complete loss of the outer nuclear layer (ONL) at an age of six weeks.
Human H1 and H9 ESC-derived RPE cells were transplanted into the sub-retinal space of 19-21 day-old RCS rats and identified by human cell-specific markers, i.e. human nuclei and b2microglobulin, 5-6 weeks after grafting ( Figure 6A, 6B, Figure S6A). Directly at the injection site, human ESC-derived RPE cells were found in clusters and obvious disruption of the retinal structure was observed ( Figure S6A). However, next to the injection site, the donor RPE formed a monolayer with direct contact to adjacent outer segments ( Figure 6A, Figure S6A). Correspondingly, the ONL structure overlying the donor RPE layer was well preserved ( Figure 6B, Figure S6A). Apicalbasolateral polarization of the transplanted RPE cells was confirmed at the ultrastructural level with a basally located nucleus and apically distributed melanosomes containing melanin ( Figure 6C). Additionally, electron-microscopic analysis of junctions between human ESC-derived RPE cells, identified by pigmentation and morphology, showed the formation of tight junctions and/or desmosomes between donor cells ( Figure S6B), a typical feature of endogenous RPE monolayers.
To determine the extent of ONL rescue, we compared ONL thickness in regions with organized structure overlying transplanted RPE monolayers, versus peripheral regions lacking transplanted cells. Fluorescence images of tissue sections revealed that ONL associated with transplanted human ESC derived RPE contained 5-6 rows of nuclei ( Figure 6D). In contrast, ONL in peripheral regions lacking transplanted cells contained 1-2 rows of photoreceptor nuclei ( Figure 6E), similar to untreated control retinas ( Figure S6C, S6D). The ONL thickness of experimental retinas (n = 6) was quantified and compared with untreated RCS rats (n = 6), animals injected with human fibroblasts (n = 5), or PBS (sham injection) (n = 5). In all experimental groups, the ONL in peripheral regions lacking transplanted cells displayed a thickness of <8-9 mm, corresponding to 1-2 rows of nuclei ( Figure 6E). In contrast, in areas containing donor RPE the ONL was significantly thicker (29.7069.34 mm) compared to untreated controls (5.9162.38 mm) or sham-injected animals (12.9968.39 mm) ( Figure 6F). Interestingly, in accordance with previous studies that also demonstrated photoreceptor rescue following injection of several other, non-RPE cell-types [13,19], transplantation of human fibroblasts had an almost similar protective effect on ONL thickness as human ESC-derived RPE cells reaching 25.3869.49 mm in central regions of experimental retinas ( Figure 6F).
To further evaluate the protective effects of RPE versus fibroblast transplantations, we quantified the total retinal area that showed increased ONL thickness for both experimental groups ( Figure 6G). Although similar numbers of cells were transplanted, retinas that received human ESC-derived RPE showed a significantly increased area of rescued ONL (855.636513.91 mm 2 with a maximum of 1900 mm 2 ) compared to human fibroblast recipients (178.906148.85 mm 2 ) ( Figure 6G). Given that only the RPE cells were able to correctly integrate into the host RPE monolayer, these results suggest better functionality or survival of donor RPE cells over fibroblasts. A potential explanation for the improved rescue effect by donor RPE cells might be the production and secretion of RPE-specific factors that are essential for photoreceptor development, survival, maintenance and function. The identification of such factors will have to be evaluated in detail in future studies.
Taken together, our results indicate that RPE cells that were differentiated in vitro from neuroepithelial cysts and transplanted without any pre-selection efficiently form mature RPE in vivo. We observed no tumorigenesis within the timescale of our experiments.

Discussion
This study demonstrates that the application of a Matrigelbased 3D epithelial cyst culture to human pluripotent stem cells results in an efficient reconstitution of neuroepithelial structures including the formation of a single lumen. Recently, mouse ESC aggregates cultured in laminin and Nodal were shown to form three-dimensional structures where a major part of the cluster organized into multi-layered optic vesicle-like tissue that formed neural retina and RPE; a ground-breaking system, important for studying complex, inductive events between different cell types [4]. Modified conditions placing floating aggregates in Matrigel plus WNT inhibitor, FBS, and SHH activators have also been utilized to obtain self-organizing retinal structures from human ESCs [6]. There are several important distinctions between such methods and the one we introduce here. In our system, pluripotent cells are not pre-aggregated, but rather directly placed in neural induction conditions in Matrigel. This leads to an unprecedented speed of neural induction compared to other methods [4,23,24,38]. Human ESCs form a PAX6 + neuroepithelium within 5 days representing a significant enhancement in the speed of neural induction compared to previously described methods. In contrast to self-organizing embryoid bodies or neural aggregates, our human ES-derived cysts have a single lumen and are uniform in cellular identity. The uniformity and patternability of the neuroepithelia was evidenced in the human ESC system where cysts uniformly achieve retinal identity within 5 days. From these cysts we could quantitatively generate neural retina epithelium or RPE under directed culture media conditions, rather than relying on self-organization of the optic vesicle tissue. These experiments open the path for future experiments reconstituting neuroepithelial patterning directly in neuroepithelial cysts using defined soluble factors.
The RPE cells from our human ESC-derived cysts have morphological and functional properties highly characteristic of native RPE cells and, importantly, form a confluent RPE cell layer without the need for isolation and further expansion. Therefore our method provides an unprecedentedly high efficiency and speed of human RPE cell sheet generation, which could be applied toward drug screening, transplantation approaches for the replacement of RPE in retinal degeneration, and for studying human RPE biogenesis at the biochemical level. All previous protocols required manual sub-selection and expansion [3,9,14,15,18]. This is important since in vitro passaging of primary RPE cells can result in an epithelial to mesenchymal transition and loss of RPE phenotype [39].
Our in vivo experiments demonstrate that direct transplantation of hESC-derived RPE without selection or expansion resulted in integration and monolayer formation, confirming the robust epithelial structural integrity of these cells. The transplanted monolayers significantly rescued overlying photoreceptor cells from degeneration in RCS rats, the standard model for RPE transplantation studies. Whether the observed protection of the ONL is caused by a paracrine effect or phagocytic function of donor RPE has to be further evaluated given the finding that human fibroblasts showed similar effects on the ONL thickness of RCS rats. However, hESC-derived RPE protected a significantly larger area of ONL than fibroblasts, pointing to a superior effect of these cells. In light of recent, promising transplantation approaches for the replacement of dysfunctional or lost RPE in pre-clinical models [3,9,10,12,19] and human patients [11] with ES or iPS cell-derived RPE [40,41,42], this rapid protocol for directly differentiating pure, human RPE cell populations from an unlimited in vitro cell source has high medical relevance. Regarding the possibility of using such cells for human transplantation, further caution on the purity of transplanted RPE cells should be considered. Before use of these cells in clinical applications, their functionality following transplantation in vivo has to be further  to evaluate their potential for e.g. phagocytosis of photoreceptor outer segments or recycling of retinoids, besides analyzing their rescue capacity in other preclinical models characterized by RPE loss or dysfunction.

Maintenance of hESCs and human iPSCs
hESC lines H9, H1 (WISC Bank) [43] and one iPSC line O27-08 (gift of Yasuhiro Takashima and Austin Smith, University of Cambridge, Cambridge, UK) were cultured in mTeSR1 medium (StemCell Technologies) on BD Matrigel hESC-qualified Matrix (BD Biosciences)-coated plates [44]. The hESCs and human iPSCs expressed OCT4 and SSEA4 confirming the undifferentiated state. Cells were passaged every 5-6 days using Dispase (StemCell Technologies). Morphologically distinguishable differentiated cells were mechanically removed at each passage. To improve cell survival during passaging, the Rho kinase inhibitor, Y-27632 (Calbiochem), was added in the culture medium during the first 24 hours after plating. For storage, the hESC colonies were suspended in mFreSR TM (StemCell Technologies), frozen in an isopropanol freezing container at 280uC overnight and transferred into liquid nitrogen the next day.

Differentiation of hESCs and human iPSCs in the neuroepithelial cyst model
Undifferentiated cell colonies were partially lifted by Dispase and scraped off with a cell scraper. Detached cell aggregates (diameter: 50-100 mm) from one confluent well of a six-well plate were embedded in 150-300 ml of Matrigel. 50 ml of Matrigel containing the cell clumps was added per square centimeter of growth surface. After gelling at 37uC for 10 min, cells embedded in the Matrigel layer were cultured in neural induction medium N2B27 which consisted of DMEM/F12+GlutaMAX TM (GIBCO), neurobasal medium (GIBCO), 0.56B27 supplement (GIBCO), 0.56N2 supplement, 0.1 mM b-mercaptoethanol and 0.2 mM L-Glutamine, prepared as described [22]. In separate experiments, AG 1024 (Calbiochem, 10 mM, from Day 0 to 5) or DMSO was added to differentiating hESC cultures.

Differentiation of RPE or neural retina cells on transwell filters
hESC-or human iPSC-derived cysts at Day 5 were taken out of the Matrigel using Cell Recovery Solution (BD Biosciences) and treated with TrypLE TM (Invitrogen), followed by gentle trituration to achieve single cell suspension. To get rid of the few non-cyst cells, large cysts were permitted to sink down to the bottom of the tube by gravity. 40 mm-cell strainer (BD Biosciences) was used to exclude big clumps after TrypLE TM treatment. Dispersed cells were resuspended in N2B27 medium and plated onto growth factor-reduced-Matrigel (BD Biosciences)-coated 6.5 mm Trans-wellH with 0.4 mm Pore Polyester Membrane Insert (Corning Costar) at a density of 2-4610 5 cells/well. At Day 6, attached cells were washed twice with RPE medium-DMEM+GlutaMAX TM medium supplemented with 20% knockout TM serum replacement (Gibco), Non-Essential Amino Acids, 1 mM L-Glutamine and 0.1 mM b-mercaptoethanol. The cells were then kept in RPE medium with or without 100 ng/ml human Activin A (Peprotech).
In the experiments of inhibiting TGF-b signaling, SB 431542 (Sigma, 8 mM) was added in cultures from Day 6 to 25. For neural retina differentiation, the cells were kept in N2B27 medium when culturing on transwell filters. The medium was changed every 2-3 days.

Animals and transplantation
Host animals were 19 to 21 days old RCS albino rats, which were maintained in a 12 h light/dark cycle, with food and water ad libitum. For transplantation, human ESC-derived RPE cells were washed with PBS and incubated 20 minutes at 37uC with 0.5 mg/ ml trypsin (Invitrogen). Then 2 mg/ml trypsin inhibitor (Roche Diagnostics) and 0.1 mg/ml DNase I (Invitrogen) were added. Additional mechanical dissociation was performed. The cells were centrifuged 5 minutes at 300 g and finally resuspended in PBS.

Immunocytochemistry
Cells or cysts were fixed with freshly made 4% paraformaldehyde (PFA) for 20 min at room temperature (RT), followed by antigen retrieval (if necessary, Citrate buffer, 70uC, 30 min), quenching (16PBS, 100 mM Glycine, 0.3% Triton X-100) and blocking (16PBS, 0.5% BSA, 0.3% Triton X-100). Immunostaining was performed in blocking buffer and each antibody reaction was done overnight at 4uC. The primary antibodies and their working dilutions were as follows: (1:1000, Millipore), mouse anti-human nuclei (1:50, Millipore), mouse anti-EEA1 (1:1000, gift of Marino Zerial, Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany), and goat anti-GFP (1:400, Rockland). Antibody against PAX6 was obtained as previously described [45]. Antibodies against RX and CRX were raised in rabbits immunized with a portion of human RX (residues 216-258) or human CRX (residues 112-259) tagged with GST and affinity-purified against the corresponding MBP-RX or MBP-CRX fusion proteins respectively. The immunoreactivity of each antibody was confirmed by immunostainings on mouse tissues of corresponding stage as a positive control. The secondary antibodies used were as follows: Alexa Fluor 488

RT-PCR analyses
Total RNA was extracted from undifferentiated or differentiating hESCs using a RNeasy kit (Qiagen), treated with RNase-free DNase I (Invitrogen) and reverse-transcribed with SuperScript II (Invitrogen). The synthesized cDNA was amplified with gene-specific primers. PCR was carried out with Taq DNA Polymerase with denaturation 94uC-20 s, annealing 55uC-30 s, extension 72uC-60 s for 35 cycles. The primers used are available in Table S2.

Electron microscopic analyses
Cells were fixed and processed on the transwell filters. They were fixed with 2% glutaraldehyde and 2% PFA in 50 mM HEPES overnight at 4uC. After several washes in 100 mM HEPES and PBS they were postfixed with 1% OsO 4 /PBS for 2 hours on ice, washed with PBS and water, and en bloc contrasted with 1% uranyl acetate in water, again for 2 hours on ice. Samples were dehydrated in a graded ethanol series, infiltrated in ethanol/ epon mixtures (1:3, 1:1, 3:1) and in pure epon (26) and cured at 60uC for 24 hours. Samples were cut to small pieces (1-2 mm), remounted for cross-sectioning of the RPE layer on empty epon dummy blocks and cured for another 24 hours at 60uC. Ultrathin sections were stained with lead citrate and uranyl acetate and inspected at a FEI Morgagni 268 D transmission electron microscope at 80 kV.

Measurement of transepithelial resistance
The transepithelial resistance of the hESC-derived RPE cells growing on transwell filters and the empty transwell filters only coated with growth factor-reduced-Matrigel as a control was measured by using an EVOM epithelial voltohmmeter (World Precision Instruments, Hamden, CT) with a pair of chopstick electrodes.

Phagocytosis analyses using retinal explant co-culture model
The transwell filter (diameter: 6.5 mm) with hESC-derived RPE growing on it was excised and transferred to bigger transwell filter (diameter: 24 mm) with the cells facing up, for easier handling. Neural retinal explants from transgenic mice expressing human RHODOPSIN-GFP [33] were dissected out in PBS and oriented on top of the RPE cell sheet to make the outer segments localize adjacent to the apical surface of hESC-derived RPE. By adding one small drop of low-melting-point agarose gently on top of the retinal explant, the outer segment layer better contacted the RPE. 1 ml of N2B27 medium supplemented with 10% FBS was added to the lower chamber of the transwell so that the medium just touched the surface of the transwell filters. The hESC-derived RPE plus mouse retina composites were fixed in 4% PFA for 1 hour at RT after 24 hours of co-culture, followed by immersion in 30% sucrose at 4uC overnight and subsequent embedding in Tissue-Tek O.C.T. TM Compound (Sakura). The thickness of cross-sections was 10 mm.
Sense and antisense probes of OCT4 and RX for in situ hybridizations were obtained by RT-PCR from total RNA of undifferentiated or differentiated hESCs. Primers used to amplify the fragments of OCT4 (database accession no. NM_203289) and RX (database accession no. NM_013435) cDNA were shown in Table S2. Sense and antisense probes of PAX6 were prepared from the Full Length cDNA Clone IRATp970B0617D (imaGenes, database accession no. BC011953).

Statistical analyses
Values are expressed as means6s.e.m.. For quantification of cysts, 2 to 3 dishes were used for staining per marker at each experiment. Around 100 randomly chosen fields, 100-200 cysts in total, were scored for each marker under confocal microscope. The total number of the cysts was determined by Hoechst nuclear staining. Nuclear localized staining for each marker on all Hoechst-stained nuclei within cysts was manually scored as positive. Both uniform nuclear staining signal of each marker within the cysts or strong signal at the basal-lateral side and weaker signal at the apical side were scored as positive. All experiments were repeated at least 3 times. For quantification of pigmented cells, 1-2 wells were used at each experiment. 3-4 randomly chosen fields, 500-1000 cells in total, were manually counted. The total number of the cells was determined by Hoechst nuclear staining. Experiments were performed in 3 replicates. For the ONL thickness and protected area analyses, 6 RCS rats transplanted with approximately 120,000 human ESC-derived RPE cells, 5 with human fibroblasts, 5 sham-injected and 6 untreated controls were quantified in central (protected) and peripheral regions using approximately every 8 th serial section. Areas of experimental retinal sections that showed an ONL with more than three rows of nuclei were defined as rescued allowing calculating the total ''protected area'' per eye. Statistical analyses were performed with IGOR PRO 6.1. Figure S1 Specificity of antibodies. The antibodies used for immunostaining were tested on mouse embryonic sections in the eyefield at representative stages. Inset: higher magnification confirming the nuclear staining of Mitf. Nuclei were counterstained with Hoechst. Scale bar, 50 mm (TIF)