In vertebrate eyes, the retinal pigment epithelium (RPE) interacts with the outer segments of photoreceptors on its apical side and with the extracellular matrix of Bruch’s membrane on its basal side (Lehmann et al., 2014; Weisz and Rodriguez-Boulan, 2009). The RPE supports the survival and functions of photoreceptors by absorbing light scattered from the outer segments of photoreceptors; by participating in the visual cycle of photopigments; by capturing toxic metabolic wastes from photoreceptors; by providing nutrients from choroidal capillaries to the retina via trans-epithelial transport; and by engulfing the outer segments of photoreceptors and dead photoreceptors (Simó et al., 2010; Strauss, 2005).

Structural and functional defects in the RPE frequently result in dysfunction and/or degeneration of the photoreceptors, causing various retinal degenerative diseases, including retinitis pigmentosa, retinal detachment, and age-related macular degeneration (Kang et al., 2009; Marmorstein, 2001; Veleri et al., 2015). Its unique functions in vision therefore require the RPE to maintain a polarized distribution of many proteins (Bonilha et al., 1999; Finnemann et al., 1997; Fujimura et al., 2009; Shimura et al., 1999). To date, however, the underlying molecular mechanisms responsible for the establishment and maintenance of a polarized protein distribution in the RPE remain incompletely understood.

Following their synthesis in the endoplasmic reticulum, membrane proteins in epithelial cells are delivered to various intracellular membranous compartments by exocytic vesicles derived from the Golgi apparatus (Rodriguez-Boulan and Macara, 2014). Later, the proteins, however, can be redistributed to other membranous compartments from initial sites by endocytic transport (Le Borgne and Hoflack, 1998; Mellman and Nelson, 2008; Rodriguez-Boulan and Macara, 2014; Shivas et al., 2010; Williams et al., 1984). Therefore, the polarized distribution of membrane proteins in epithelial cells can result from both exocytosis and endocytosis.

Proteins in the plasma membrane are loaded onto endosomes and subsequently transferred to lysosomes by endosome-lysosome fusion, resulting in the degradation or re-routing of these proteins to other cellular membrane compartments. During the course of endo-lysosomal maturation, many membrane proteins in endosomes can be sorted further into multivesicular bodies (MVBs). The intra-MVB vesicular trafficking removes endosomal membrane proteins from cytoplasm, where their functional sites are exposed. Consequently, cytoplasmic events mediated by membrane receptors, such as epidermal growth factor receptor (EGFR), are terminated by MVB trafficking (Eden et al., 2009). Proteins in the intra-MVB vesicles, also called exosomes, are often released into extracellular space following fusion of the MVBs with the plasma membrane (Grant and Donaldson, 2009; Gruenberg and Stenmark, 2004; Hurley, 2010; Schmidt and Teis, 2012). Fusion of these MVB-derived extracellular vesicles with the plasma membrane can lead to the reintegration of these membrane proteins into the plasma membrane (Clague et al., 2012), resulting in the transfer of the proteins autonomously and non-autonomously.

Therefore, the endosomal sorting complexes required for transport (ESCRT), which is responsible for the formation of MVB (Babst, 2011; Gruenberg and Stenmark, 2004), starts to receive a focus as an intercellular communication machinery in addition to its classical role in the endo-lysosomal proteins degradation. The ESCRTs have been subdivided into four major complexes: ESCRT-0, -I, -II, and -III. The components of ESCRT-0, including hepatocyte receptor tyrosine kinase substrate (Hrs)/vacuolar protein sorting 27 (Vps27) and signal transducing adapter molecule 1 (Stam1), capture ubiquitinated proteins, transferring them to ESCRT-I and subsequently to ESCRT-II and -III for intra-endosomal vesicular sorting to form MVBs (Hurley, 2010; Schmidt and Teis, 2012). Membrane proteins in the endosomes are also subjected to ESCRT-mediated intra-endosomal vesicular trafficking, although ESCRT-0 is not essential for MVB formation.

The present study examined the roles of ESCRT-mediated protein trafficking in mouse RPE by eliminating tumor susceptibility gene 101 (Tsg101), which encodes an ESCRT-I component Tsg101, also called vascular protein sorting 23 (Vps23). Deletion of Tsg101 abrogated the polarity in mouse RPE, which form an irregular multilayer structure. In the Tsg101-deficient mouse RPE, the phosphoinositide 3-kinase (PI3K)-Akt and mitogen-activated protein kinase (MAPK) pathways, which can disrupt RPE polarity (Kang et al., 2009; Kim et al., 2008), were also found to be activated at downstream of EGFR, which is subjected to the ESCRT-mediated downregulation (Eden et al., 2009). These results suggest that the ESCRT-mediated protein trafficking is necessary for the establishment and maintenance of RPE cell polarity.

Mouse lines

Tag101 ^fl/fl mice were generated as described previously, and used to breed with various Cre lines and R26R mice to generate the mice used in this study (Iacovelli et al., 2011; Mori et al., 2002; 2012; Rowan and Cepko, 2004; Soriano, 1999; Wagner et al., 2003). Mice were genotyped according to the published protocols. Samples were collected and analyzed between littermates. All mice used in this study were maintained in a specific pathogen-free facility of Korea Advanced Institute of Science and Technology (KAIST) Laboratory Animal Resource Center. All experiments were performed according to the Korean Ministry of Food and Drug Safety (MFDS) guidelines for animal research. The protocols were certified by the Institutional Animal Care and Use Committee (IACUC) of KAIST (KA-2014-20).

Immunohistochemistry

Immunohistochemistry of cryosection was done as described in previous report (Kim et al., 2008). Briefly, mice are perfused and then adult eyes samples are fixed overnight in 4% paraformaldehyde at 4°C, whereas embryonic heads are fixed for 4 h at 4°C. Heat-induced antigen retrieval was also performed with citrate buffer (10 mM sodium citrate, 0.05% Tween-20, pH 6.0) when necessary. Samples were incubated in blocking solutions (phosphate-buffered saline [PBS] with 0.5% Triton X-100 + 5% normal serum for embryonic eyes and PBS with 1% Triton X-100 + 10% normal serum for adult eyes). Samples are incubated with primary antibodies diluted in blocking solution (1:100 v/v) overnight at 4°C. Secondary antibodies are diluted in blocking solution (1:200 v/v) for 1 h at room temperature. Hoechst staining is done for 15 min at room temperature following secondary antibodies incubation. Antibodies used in this study are provided in Supplementary Table S1.

The immunostaining images were then acquired using Olympus FV1000 confocal microscope and manipulated by the Photoshop CS6 and Bitplane Imaris 6.3.1 softwares.

Electron microscopy

Transmission electron microscopy (TEM) analyses were done as described previously (Ha et al., 2017; Kim et al., 2008). The eyes of were isolated from adult C56BL/6J mice perfused with a solution containing 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M sodium cacodylate buffer (pH 7.2) in the morning and further fixed in the same solution for 24 h. Alternatively, embryonic mouse heads were isolated from the uterus and fixed for 24 h. The eyes and embryonic heads were then post-fixed with 1% OsO4 for 2 h on ice. The samples were stained and blocked with 0.5% uranyl acetate for 2 h, and then embedded in Epon 812 after dehydration. Ultrathin sections (80 nm) of the samples were then made and examined using an H-7000-type electron microscope (Hitachi, Japan) operated at 75 kV.

Electroretinogram (ERG) and OptoMotry

ERG measurements were done as described previously (Kim et al., 2017). In brief, mice were either dark- or light-adapted for 12 h and anesthetized with 2,2,2-tribromoethanol (Sigma, USA) in prior to dilating the pupils of the mice by 0.5% tropicamide. The mice were placed with a gold-plated objective lens on their corneas and silver-embedded needle electrodes at their foreheads and tails. The ERG recordings were performed using Micron IV retinal imaging microscope (Phoenix Research Labs, USA) and analyzed by Labscribe ERG software according to the manufacturer’s instruction.

Mouse visual acuity was measured with the OptoMotry system (Cerebral Mechanics, USA) as previously described (Prusky et al., 2004). Mice adapted to ambient light for 30 min were placed on the stimulus platform surrounded by four computer monitors displaying black and white vertical stripe patterns. An event that mice track the stripe movements with reflexive head-turn was counted as a successful visual detection. The detection thresholds were then obtained from the OptoMotry software.

Statistical analysis

All statistical analyses and graphs construction were done using IBM SPSS Statistics (ver. 20; IBM, USA). All data from statistical analysis are presented as the mean ± SD. Comparison between two groups was done by unpaired Student’s t-test, and the differences among multiple groups were determined by ANOVA with Tukey’s post-test. P values < 0.01 were considered as statistically significant results.

Dynamic changes on the distribution of RPE polarity markers during development

Unlike the fully polarized characteristics of mature RPE, less is known about the polarity of embryonic RPE, which does not have recognizable microvilli on its apical side but forms focal contacts with adjacent retinal progenitor cell (RPC) (Fig. 1). The finding suggests that the expression and subcellular distribution of proteins responsible for interaction with adjacent retinal cells may differ in embryonic and mature RPE. Therefore, the distribution of various cell adhesion and polarity markers were assessed in mouse RPE from embryo to adult (Fig. 2, Supplementary Fig. S1, Table 1).

P/E-cadherin

Previously, placental and epithelial (P/E) cadherin protein and mRNA were shown to be expressed in adult RPE of adult mouse eyes and in culture (Burke and Hong, 2006; Burke et al., 1999), whereas only P-cadherin mRNA was identified in RPE of embryonic mouse eyes (Xu et al., 2002). We found that P/E-cadherin was expressed evenly in the RPE from embryonic day 12.5 (E12.5) to post-natal day 0 (P0) and then to be enriched at apical and basal sides of RPE at P7, when photoreceptors start to develop their outer segments (Figs. 2A-2D, Table 1). In mature RPE (i.e., P14 and P30), P/E-cadherin was found to be expressed strongly in the baso-lateral membrane but was not expressed in the apical membrane, as reported previously (Burke and Hong, 2006; Burke et al., 1999) (Figs. 2E and 2F, Table 1).

N-cadherin

N-cadherin was reported to be expressed in the baso-lateral membrane of mature RPE (Cachafeiro et al., 2013; Imamura et al., 2006), but on both sides in RPE of E10.5 mice (Fujimura et al., 2009). We also found that N-cadherin was detectable throughout the membrane areas of RPE in mice at E12.5 and E14.5 (Figs. 2G and 2H, Table 1). During the first post-natal week, N-cadherin expression was significantly reduced on the baso-lateral side but its expression was maintained strongly on the apical side (Figs. 2I and 2J, Table 1). From P14 onward, the level of N-cadherin expression was significantly reduced in RPE and was expressed only in the basal membrane at a low level (Figs. 2K and 2L, Table 1).

_β -catenin

In agreement with previous findings on the expression of β-catenin in embryonic and mature RPE (Cachafeiro et al., 2013; Fujimura et al., 2009; Imamura et al., 2006), we found that β-catenin was highly expressed in RPE junctional areas at all stages (Figs. 2M-2R, Table 1). However, unlike its exclusive distribution in the adherens junctions (AJs) at the baso-lateral sides of mature RPE, β-catenin expression was not polarized in embryonic mouse RPE (Figs. 2M and 2N, Table 1). The results suggest that β-catenin may be involved in the apical contacts between the RPE and RPC as well as the AJs between the RPE in embryonic mouse eyes, later becoming concentrated at the inter-RPE AJs and basal membrane in mature mouse eyes.

Integrin αvβ5

We found that integrin αvβ5 was expressed on both the apical and basal sides of mature mouse RPE (Figs. 2W and 2X, Table 1), as reported previously (Finnemann et al., 1997). During the embryonic period, however, integrin αvβ5 was expressed only on the basal side of RPE (Figs. 2S and 2T), whereas integrin αvβ5 expression in post-natal RPE was shifted to the apical membrane (Figs. 2U and 2V, Table 1).

Na ⁺ /K ⁺ -ATPase α 1

Similar to its expression in the apical membrane of cultured rat RPE (Gundersen et al., 1991), Na⁺/K⁺-ATPase α1 was found to be expressed in the apical membrane of post-natal mouse RPE (Figs. 2A’-2D’, Table 1). In RPE of embryonic mice, however, Na⁺-K⁺/ATPase α1 was expressed on the basal side (Figs. 2Y and 2Z, Table 1), similar to findings in other types of epithelial cells (Amerongen et al., 1989; Sztul et al., 1987).

Ezrin

Expression of ezrin, a marker for microvilli (Bonilha et al., 1999), was detected in the apical membranes of post-natal and mature mouse RPE (Figs. 2G’-2J’, Table 1). In embryonic mouse RPE, however, ezrin expression was not polarized (Figs. 2E’ and 2F’, Table 1). These patterns were similar to those of F-actin, which is recruited by ezrin to the microvilli (Figs. 2K’-2P’).

All of these results are summarized in Table 1 and low magnification images are provided in Supplementary Fig. S1.

Tsg101-deficent embryonic RPE cells form random cell aggregates

Dynamic changes on the distribution of the adhesion and polarity markers in embryonic, post-natal, and mature RPE indicated that the localization of these proteins may be regulated by stage-specific environments, including changes in interactions of the RPE with the retina and ECM and in soluble factors produced by the RPE, retina, and choroid. These signals might not only induce the exocytic trafficking of proteins to the target sites, but may also induce their redistribution by endocytic down-regulation at unstable sites and/or their redistribution to stable sites (Rodriguez-Boulan and Macara, 2014; Shivas et al., 2010).

ESCRT complexes play important roles in the down-regulation of ubiquitinated proteins through the endo-lysosomal pathway (Luzio et al., 2009; Saksena et al., 2007). These complexes are also involved in the remobilization of membrane proteins via the fusion of MVBs to the plasma membrane. Furthermore, Tsg101, a component of ESCRT-I, was previously shown to establish the polarity of Drosophila embryonic cells (Moberg et al., 2005), suggesting that Tsg101 may also regulate developmental changes in RPE cell polarity in mice by remobilizing membrane proteins.

This hypothesis was tested by generating Tsg101 ^fl/fl ;TRP1-Cre mice, in which the floxed Tsg101 gene (Tsg101 ^fl ) was deleted in tyrosinase-related protein 1 (TRP1)-Cre affected RPE, ciliary epithelium, and cells in the peripheral retina (Mori et al., 2002; Wagner et al., 2003). Gross anatomical examination showed that eyes of Tsg101 ^fl/fl ;TRP1-Cre mice, beginning at E14.5, were significantly smaller than in their Tsg101 ^fl/+ ;TRP1-Cre littermates (Fig. 3A, Supplementary Fig. S2). Moreover, RPE cells were depigmented in the eyes of Tsg101 ^fl/fl ;TRP1-Cre mice (Fig. 3A). These depigmented RPE cells, however, still expressed microphthalmia factor 1 (Mitf1; Fig. 3B), a key transcription factor that induces the expression of melanin producing enzymes, such as tyrosinase and TRP1 (Fang and Setaluri, 1999; Martinez-Morales et al., 2004). These results therefore suggest that depigmentation of the RPE may be associated with the involvement of Tsg101 in melanogenesis (Truschel et al., 2009), not with the trans-differentiation of RPE to other ocular cell types.

Mitf1-positive cells in Tsg101 ^fl/fl ;TRP1-Cre mouse eyes, however, failed to form a monolayer epithelial sheet; rather, they formed irregular cell aggregates (Fig. 3B, Supplementary Fig. S2). These Mitf1-positive cells were positive for _β-galactosidase (Fig. 3A), which was expressed from the lacZ Cre reporter knocked in the ROSA26 gene locus (R26R) after the Cre-mediated deletion of the loxP-STOP-loxP gene cassette (Soriano, 1999), suggesting that impaired layer formation was an autonomous characteristic of Tsg101-deficient RPE. The numbers of Mitf1-positive cells were significantly lower in E14.5 Tsg101 ^fl/fl ;TRP1-Cre mouse eyes than those in Tsg101 ^fl/+ ;TRP1-Cre littermate eyes (Figs. 3B and 3C). In contrast, the numbers of RPCs, which are positive for expression of visual system homeobox 2 (Vsx2) and SRY-Box transcription factor 2 (Sox2), and of post-mitotic retinal neurons, which are positive for expression of tubulin_β3 (Tuj1) and brain-specific homeobox 3b (Brn3b), did not differ markedly in the retinas of these mice (Supplementary Figs. S3A and S3B). The numbers of BrdU-labeled proliferative cells and pH3-positive mitotic cells at E14.5 also did not differ significantly in eyes of E14.5 Tsg101 ^fl/+ ;TRP1-Cre and Tsg101 ^fl/fl ;TRP1-Cre mice (Supplementary Figs. S3C and S3D), suggesting that cell proliferation was not affected by the deletion of Tsg101 from the RPE. The reduced numbers of RPE cells were not due to cell death, either, as there were no significant differences in the numbers of TUNEL-positive apoptotic cells in eyes of Tsg101 ^fl/fl ;TRP1-Cre and Tsg101 ^fl/+ ;TRP1-Cre mice at E14.5 (Supplementary Figs. S3C and S3D). In contrast, the numbers of TUNEL-positive cells were significantly higher in E14.5 Tsg101 ^fl/fl ;Chx10-Cre mouse eyes, in which Tsg101 had been eliminated in RPCs and their descendent retinal neurons (Rowan and Cepko, 2004), than those in Tsg101 ^fl/+ ;TRP1-Cre littermate eyes (Supplementary Fig. S4). These results suggest that Tsg101 is essential for the survival of RPC and/or retinal neurons but not for the RPE, which requires it for monolayer formation and pigmentation.

Deficiency of Tsg101 in embryonic RPE leads to the loss of polarity

We next assessed whether the expression of the proteins, which exhibit the polarized distributions in embryonic RPE (Fig. 2, Table 1), was altered in Tsg101-deficient mouse embryonic RPE to form the cell aggregates. P/E-cadherin and β-catenin were detectable evenly in E14.5 Tsg101 ^fl/fl ;TRP1-Cre mouse RPE, whereas those are accumulated in the apical and basal RPE of their Tsg101 ^fl/+ ;TRP1-Cre littermates (Figs. 3D [two left columns] and 3E). Integrin αvβ5, which was expressed on the basal side of RPE in Tsg101 ^fl/+ ;TRP1-Cre mice, was expressed evenly on both sides of the RPE in their Tsg101 ^fl/fl ;TRP1-Cre littermates (Figs. 3D [third column from left] and 3E). Na⁺/K⁺-ATPase, a basal RPE marker of embryonic RPE (Fig. 2, Table 1), was found to be diffused throughout the entire RPE area of Tsg101 ^fl/fl ;TRP1-Cre mice (Figs. 3D [third column from right] and 3E). Ezrin was also detected in the entire RPE cell area of Tsg101 ^fl/fl ;TRP1-Cre mice, but it was enriched in the apical and basal RPE of Tsg101 ^fl/+ ;TRP1-Cre mice (Figs. 3D [second column from right] and 3E). Consequently, F-actin was failed to enrich in the apical side but diffused into the entire RPE area in Tsg101 ^fl/fl ;TRP1-Cre mice (Fig. 3D, rightmost column). These results suggest that Tsg101 is necessary for the polarized distribution of these cell adhesion proteins in mouse embryonic RPE.

Tsg101-deficient adult mouse RPE fails to form a monolayer sheet

Because the RPE aggregates could not be maintained in the eyes of Tsg101 ^fl/fl ;TRP1-Cre embryos (Supplementary Fig. S2), we could not determine the polarity of mature RPE in these mice. We therefore bred Tsg101 ^fl/fl mice with Bestropin 1 (BEST1)-Cre mice to generate Tsg101 ^fl/fl ;BEST1-Cre mice, which lose Tsg101 in the RPE at the adult stage (Iacovelli et al., 2011). Anatomical and histological analyses showed no significant differences in eyes between P60 Tsg101 ^fl/+ ;BEST1-Cre and Tsg101 ^fl/fl ;BEST1-Cre mice, except for depigmentation of the eyes of the latter (Fig. 4A). However, the depigmented RPE cells in P60 Tsg101 ^fl/fl ;BEST1-Cre mouse eyes failed to form a monolayer structure (Fig. 4B). Furthermore, the Otx2-positive RPE cells in the Tsg101 ^fl/fl ;BEST1-Cre mouse eyes were present in higher numbers and stacked in multiple layers (Figs. 4B and 4C). However, the numbers of major retinal cell types, including Otx2-positive bipolar cells, in the eyes of Tsg101 ^fl/fl ;BEST1-Cre and Tsg101 ^fl/+ ;BEST1-Cre mice did not differ significantly (Figs. 2B and 2C, Supplementary Fig. S5).

Analysis of the polarity of mouse RPE revealed that E-cadherin was expressed evenly on RPE membranes of Tsg101 ^fl/fl ; BEST1-Cre mice, but was expressed strongly only at the AJs in Tsg101 ^fl/+ ;BEST1-Cre mice (Fig. 4D; leftmost column). Consequently, β-catenin was expressed throughout the entire RPE area in Tsg101 ^fl/fl ;BEST1-Cre mice (Fig. 4D, second column from left). It was also shown that integrin αvβ5, Na⁺/K⁺-ATPase, and ezrin, which were detectable only in the apical regions of Tsg101 ^fl/+ ;BEST1-Cre mouse RPE, were expressed throughout the RPE in Tsg101 ^fl/fl ;BEST1-Cre mice (Fig. 4D, three center columns). Similarly, F-actin, which was concentrated in the cell cortex of Tsg101 ^fl/+ ;BEST1-Cre mouse RPE, was expressed throughout the RPE in Tsg101 ^fl/fl ;BEST1-Cre mice (Fig. 4D, rightmost column).

We further investigated ultrastructural alterations in the Tsg101-deficient mouse RPE by TEM. Intercellular adhesions of Tsg101 ^fl/fl ;BEST1-Cre mouse RPE were disrupted and their basal folds became irregular (Fig. 4E, iii), findings likely due to the lack of basolateral enrichment of E-cadherin and _β-catenin. However, microvilli were found to extend from the RPEs of both Tsg101 ^fl/+ ;BEST1-Cre and Tsg101 ^fl/fl ;BEST1-Cre mice (arrows in Fig. 4E, ii and iv). However, the apical borders, from which the microvilli branched out, were detectable deeper in the Tsg101 ^fl/fl ;BEST1-Cre mouse RPE (Fig. 4E, iv, blue dotted lines), suggesting that the apical-lateral borders do not form properly in the RPE of these mice, resulting in the growth of the microvilli even from the lateral sides.

Activation of the EGFR signaling pathway in Tsg101-deficient RPE

Receptor tyrosine kinases (RTKs) are transported to the basolateral side of the RPEs, whereas their ligands are secreted from their apical side (Lehmann et al., 2014; Weisz and Rodriguez-Boulan, 2009). Consequently, the autocrine activation of these RTKs, as represented by their phosphorylation, is suppressed in healthy epithelium. We found that the phosphorylation of epidermal growth factor receptor 1 (pEgfr1) was markedly increased in the RPE of Tsg101 ^fl/fl ;BEST1-Cre mice, but was suppressed in the RPE of Tsg101 ^fl/+ ;BEST1 mice (Fig. 5), suggesting the autocrine activation of Egfr1 in Tsg101-deficient RPE. This finding was supported by results showing that the activities of the downstream PI3K-Akt and MAPK pathways, which were shown to disrupt inter-RPE adhesion (Kang et al., 2009; Kim et al., 2008), were also elevated in the RPE of Tsg101 ^fl/fl ;BEST1-Cre mice (Fig. 5). However, there was no evidence for the proliferation of RPE, which resumed the cell cycle and incorporated BrdU for one week (data not shown).

Visual impairment of Tsg101 ^fl/fl ;BEST1-Cre mice

The polarized RPE contributes to visual functions of the retina by maintaining the structures of photoreceptor outer segment and exchanging various materials with the photoreceptors (Simó et al., 2010; Strauss, 2005). We found that visual acuity was significantly impaired in P60 Tsg101 ^fl/+ ;BEST1-Cre mice (Fig. 6A), of which RPE had lost polarity but maintained contacts with the outer segments of photoreceptors (Fig. 4E). However, their electrophysiological activities, both the dark-adapted scotopic and light-adapted photopic responses measured by ERG, were normal (Figs. 6B-6G). The results suggest that the retinas adjacent to the depolarized Tsg101-deficient RPE could sense and process light but could not have spatial resolution. Collectively, our data suggest that Tsg101 should be expressed in adult mouse RPE not only to maintaining RPE structures but also to supporting normal vision of the mice.

The present study showed that cell adhesion proteins were differentially distributed in embryonic, perinatal, and mature RPE. The differences may be associated with differences in RPE functions during each state of development. RPE in embryos is not involved in vision as structural and functional supporters for the photoreceptors. The embryonic RPE does not develop the microvilli, which are projected from RPE apical membrane to interact with photoreceptor outer segments (Fig. 1). Instead, it forms junctional complexes with adjacent RPC to regulating retinal neurogenesis (Fig. 1) (Ha et al., 2017). Therefore, cadherins and β-catenin are not only expressed on the basolateral sides, but were also detected on the apical sides of the embryonic RPE, forming the junctional complexes with RPC. The transition from embryonic RPE-RPC interaction to mature RPE-photoreceptor interaction, which is less strong than embryonic RPE-RPC apical contacts, may therefore cause the redistribution of cadherins and β-catenin to the lateral side, where they form AJs (Fig. 2, Table 1). The transition on RPE-retinal interaction also moves ezrin exclusively on the apical sides to support the extension of microvilli (Fig. 2, Table 1). Integrin αvβ5 is also localized at the apical side due to its function in the phagocytosis of photoreceptor outer segments (Finnemann et al., 1997). Na⁺-K⁺/ATPase-α1 was also found to localize to the apical surface of mature RPE. On the contrary, Na⁺-K⁺/ATPase-α1 is enriched on the baso-lateral sides of embryonic RPE as it is in other epithelial cell types (Marmorstein, 2001; Shimura et al., 1999). This difference may be associated with the function of Na⁺-K⁺/ATPase-α1 in phototransduction. Na⁺-K⁺/ATPase-α1 on the apical side of the RPE exports Na⁺ ions into the subretinal space for the depolarization of unstimulated photoreceptors (Gallemore et al., 1997). These apical markers were still observed in the apical membrane of Tsg101-deficient mouse RPE, whereas E-cadherin and β-catenin cannot be maintained in the baso-lateral sides (Fig. 4D). Consequently, Tsg101-deficient mouse RPE maintains microvilli (Fig. 4E) and ERG responses (Figs. 6B-6G). The results suggest that Tsg101-dependent protein downregulation and/or redistribution are more critical for the proteins on the basolateral than on the apical side of the RPE.

ESCRT components in Drosophila were reported to have tumor suppressor properties (Herz et al., 2006; Moberg et al., 2005; Vaccari and Bilder, 2005). Notch activity in dTsg101-deficient cells was elevated, leading to the proliferation of neighboring cells through activation of the JAK-STAT pathway (Moberg et al., 2005). Unlike dTsg101, mouse Tsg101 did not show tumor suppressor activity, but was necessary for cell survival (Wagner et al., 2003). The anti-apoptotic characteristics of mouse Tsg101 were also seen in Tsg101-deficient retinas, which lost cells by massive apoptosis (Supplementary Fig. S4). However, Tsg101 in mouse RPE showed neither tumor suppressor nor cell survival factor activity (Figs. 3 and 4, Supplementary Figs. S3C and S3D). Moreover, unlike dTsg101-deficient cells (Moberg et al., 2005), loss of Tsg101 did not increase Notch receptor levels in the RPE (data not shown). In contrast, the level of pEGFR, which accumulates in endocytic vesicles upon activation (Adamson and Rees, 1981; Bakker et al., 2017), was increased in Tsg101-deficient mouse RPE, activating the downstream anti-apoptotic PI3K-Akt and MAPK signaling pathways (Fig. 5). Consequently, it might suppress the death of the cells. Despite the activation of the EGFR-PI3K-Akt signaling pathway, which enhances cell proliferation, RPE did not proliferate actively but lost the polarized epithelial characteristics. This is reminiscent to Pten-deficient RPE, which was depolarized and exhibited hyperactive Akt but did not proliferate in adult mice (Kim et al., 2008). These results suggest that the distributions of RTKs and their ligands should be regulated tightly in RPE to prevent the overactivation of PI3K-Akt pathway.

Despite the loss of cell polarity, retinal structures, as determined by retinal cell composition, and functions, as measured by ERG, appeared normal in Tsg101 ^fl/fl ;BEST1-Cre mouse retinas (Figs. 6B-6G, Supplementary Fig. S5). However, their spatial resolutions were significantly compromised (Fig. 6A). The marked reduction in light-absorbing melanin granules in the RPE of Tsg101 ^fl/fl ;BEST1-Cre mice may impair the absorption of scattered light from the photoreceptors. Consequently, the spatial differences between the photoreceptors might interfere with the scattered light, resulting in the reduced visual acuity of the mice.

This work was supported by the National Research Foundation of Korea (NRF) grants (NRF-2017R1A2B3002862 and NRF-2018R1A5A1024261) funded by Korean Ministry of Science and ICT (MSIT), South Korea.

We appreciate for Drs. Mark, Dunaief, and Cepko for the generous gifts for TRP1-Cre, BEST1-Cre, and Chx10-Cre mice. We appreciate for technical services of KAIST Laboratory Resource Center and BioCore Center.

D.L. and S.L. conceived and performed experiments, and wrote the manuscript. K.W.M., J.W.P., Y.K., T.H., and K.H.M. performed experiments. K.U.W. provided the Tsc1-flox mice. J.W.K. conceived and supervised the experiments, wrote the manuscript, and secured funding.

The authors have no potential conflicts of interest to disclose.