Conditional knockdown of DNA methyltransferase 1 reveals a key role of retinal pigment epithelium integrity in photoreceptor outer segment morphogenesis

In retinal and macular neurodegenerative diseases, such as retinitis pigmentosa (RP) and age-related macular degeneration (AMD), vision loss ultimately results from the dysfunction and/or death of photoreceptor cells (Jackson et al., 2002; Bramall et al., 2010; Wright et al., 2010). Rod and cone photoreceptors in the retina are highly polarized neurons that initiate the visual process (Lamb et al., 2007). To maximize photon capture, the photoreceptors contain unique structures, called outer segments (OSs), that utilize membrane discs to organize the visual pigment opsin and other phototransduction components for maximal efficiency (Palczewski et al., 1999; Luo et al., 2008). Each day, ∼10% of OS discs are shed and then regenerated in a circadian manner in the mammalian retina (Young, 1974; Besharse et al., 1977; Bok, 1993). The retinal pigment epithelium (RPE) serves numerous essential roles in maintaining the health and function of photoreceptors, including acting as a barrier between photoreceptors and choroidal blood circulation, removing shed OSs and mediating retinoid recycling (Strauss, 2005; Sparrow et al., 2010).

The neural retina and RPE originate from anterior neuroectoderm, and their lineages separate at the late optic vesicle stage during eye development (Chow and Lang, 2001). Distinct pools of multipotent retinal progenitor cells (RPCs) acquire competence to produce specific sets of neurons in a defined order under the combined influence of intrinsic program(s) and extrinsic factors (Livesey and Cepko, 2001; Marquardt and Gruss, 2002; Wallace, 2011). Specific cell fates are determined by combinatorial actions of a select set of transcription factors. OTX2, RORβ, BLIMP1 (PRDM1) and CRX are among the regulatory proteins that are crucial for photoreceptor development (Swaroop et al., 2010); however, two key transcription factors - NRL and TRβ2 (encoded by Thrb) - together determine the generation of three distinct types of photoreceptors (rods, S-cones and M-cones) from postmitotic precursors (Ng et al., 2011). Downstream targets of NRL and CRX, as well as signaling proteins that modify their activity, further modulate the expression of photoreceptor genes (Oh et al., 2008; Onishi et al., 2009; Onishi et al., 2010; Roger et al., 2010; Hao et al., 2011). Loss or altered function of these regulatory proteins results in photoreceptor dysfunction and retinal diseases (Swaroop et al., 2010; Wright et al., 2010). Recent studies have begun to define the transcriptional regulatory networks that dictate photoreceptor development and homeostasis (Corbo et al., 2010; Hu et al., 2010; Hao et al., 2012; Hwang et al., 2012). Despite increasing interest in the role of epigenetic mechanisms (Cvekl and Mitton, 2010; Merbs et al., 2012; Nickells and Merbs, 2012; Popova et al., 2012), the role of the epigenome, if any, in establishing retinal cell fate, differentiation and/or homeostasis is yet to be elucidated.

We have been exploring the role of DNA methylation in RPE and photoreceptor development (Nasonkin et al., 2011; Merbs et al., 2012). Specification of pigmented RPE cells is an early event in eye morphogenesis (Bharti et al., 2006; Fuhrmann, 2010). Defects in the RPE monolayer, caused by partial genetic ablation of RPE in 6-week-old mouse eyes, can lead to photoreceptor degeneration (Longbottom et al., 2009). However, a direct developmental relationship between RPE and photoreceptors has not been established. Previous studies indicated that when RPE fails to specify (for example, in Mitf mutants) or when RPE development is abrogated by ectopic expression of diphtheria toxin, retinal development and photoreceptor maturation are severely compromised (Raymond and Jackson, 1995; Bumsted et al., 2001). Similarly, in retinal organotypic cultures, RPE is required for photoreceptor differentiation and survival (Ogilvie et al., 1999; Bandyopadhyay and Rohrer, 2010). Although the role of RPE in photoreceptor homeostasis is well established, genetic or epigenetic mechanisms that guide the development of the intricate photoreceptor-RPE complex are poorly understood.

DNA methylation and histone modifications are key epigenetic regulators that modulate the ability of transcription factors to gain access to DNA sites at which they act (Jaenisch and Bird, 2003; Bernstein et al., 2007). DNA methyltransferase (DNMT) 3a and 3b contribute to de novo methylation, whereas DNMT1 targets hemi-methylated CpGs during DNA replication, resulting in identical methylation patterns in the daughter DNA strands. Loss of function of Dnmt1 in mice leads to embryonic lethality (Li et al., 1992), consistent with the established roles of DNMTs in organogenesis and disease (Robertson, 2005; Ma et al., 2010). DNMT1 is expressed at high levels in postmitotic neurons and is suggested to control their survival (Hutnick et al., 2009; D’Aiuto et al., 2011). However, the in vivo relevance of DNMT1 in neuronal cell fate determination and functional maturation is poorly understood.

While exploring the role of Dnmt1 in retinal development, we developed a new mouse model that exhibits appropriate lamination and cell layers in the retina but shows compromised RPE differentiation and an absence of photoreceptor OSs. Using a comprehensive set of Cre lines to knockdown Dnmt1 in RPE or neural retina, together with gene profiling and pyrosequencing of genomic DNA, we demonstrate a direct role of RPE in modulating photoreceptor OS biogenesis and identify candidate RPE genes that might play a role in this process. Our model should be useful in investigating the epigenetic control of photoreceptor-RPE architecture and in ascertaining the signaling molecules required for OS morphogenesis that are crucial for photoreceptor survival.

All procedures with mice were performed in accordance with protocols approved by the National Eye Institute Animal Care and Use Committee. The mouse lines have been described previously: Dnmt1^fl/fl (Jackson-Grusby et al., 2001), Rx-Cre (Swindell et al., 2006), Pax6-[α] Cre (Marquardt et al., 2001), VMD2-Cre (Le et al., 2008), Six3-Cre (Furuta et al., 2000) and Crx-Cre (Nishida et al., 2003).

Primers used for amplifying genomic DNA, RT-PCR and pyrosequencing are listed in supplementary material Table S1.

Retinal histology was performed by fixing retinas for 30 minutes in paraformaldehyde (4% in PBS), followed by storage in 4% glutaraldehyde until processing. Tissues were embedded in methacrylate, sectioned at 1 μm, then counterstained with Hematoxylin and Eosin (H&E). EM was performed as described (Fariss et al., 1990). IHC procedures were as described (Nasonkin et al., 2011) and antibodies are listed in supplementary material Table S2.

RNA was prepared from neural retina (NR) or RPE/choroid from four Dnmt1^fl/fl (control) and four Dnmt1^fl/fl:Rx-Cre/+ (experimental) animals at each chosen time point using the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). RNA quality and quantity were assessed using the BioAnalyzer 2100 RNA NanoChip (Agilent Technologies, Santa Clara, CA, USA) and Nanodrop (Thermo Scientific, Wilmington, DE, USA). The methods for qRT-PCR were described previously (Brooks et al., 2011). The targets for hybridization to GeneChip Mouse Exon 1.0 ST Arrays (Affymetrix, Santa Clara, CA, USA) were synthesized from 1 μg total RNA using the GeneChip Whole Transcript (WT) Sense Target Labeling Assay (Affymetrix). Hybridization, washing and scanning of GeneChips were performed according to the manufacturer’s instructions (Affymetrix). Microarray data have been deposited in Gene Expression Omnibus under accession number GSE43951.

Heads from E16.5 and P0.5 mice were cryoprotected by passage through increasing (6.75 to 25%) concentrations of ice-cold sucrose in 0.1 M phosphate buffer, equilibrated with a 2:1 mixture (volume:volume) of 25% sucrose:OCT Compound (Tissue-Tek) for 1 hour at 4°C, snap frozen on dry ice, and cryosectioned at -30°C. Cryosections (7 μm) were thaw-mounted onto PEN foil slides (Leica Microsystems), kept at -30°C for 10-30 minutes, fixed in ice-cold 70% ethanol for 30 seconds, rinsed in water, stained for 2 minutes in Meyer’s Hematoxylin (Sigma-Aldrich), dehydrated through a 70-95-100% ethanol series (30 seconds each), and air-dried for 2 minutes. Cells from NR or RPE were isolated by LCM using the Leica LMD6000 system. Tissue fragments were collected by gravity into tube caps containing lysis buffer for DNA isolation.

Quantitation of DNA methylation was performed by bisulfite conversion followed by pyrosequencing of genomic DNA from P6.5 mouse tail and NR, or from DNA isolated from E16.5 and P0 NR and RPE using LCM. Bisulfite conversion was performed on 200 ng genomic DNA using the EZ DNA Methylation-Gold Kit (Zymo, Orange, CA, USA) following the manufacturer’s protocol. After bisulfite conversion, a 20 μl PCR reaction was carried out using Hot Start Taq polymerase (Qiagen) as per manufacturer’s instructions. Methylation was examined at LINE1 repeats (211 bp), in the differentially methylated region of H19 (-3810 to -3557 bp), within exon 1 of Opn1sw (+25 to +226 bp), and in the 5′ UTR of Tuba3a (-481 to -331 bp). Primers (supplementary material Table S1) were designed using PyroMark Assay Design 2.0 software (Qiagen). PCR cycling conditions were 95°C for 15 minutes, followed by 45 cycles of 95°C for 20 seconds, 20 seconds at the annealing temperature, and 72°C for 20 seconds, with a final extension at 72°C for 5 minutes. The biotinylated PCR product was purified and made single stranded to act as a template in the pyrosequencing reaction, as recommended by the manufacturer, using the PyroMark Q24 Vacuum Prep Tool (Qiagen). Briefly, the PCR product was bound to Streptavidin Sepharose High Performance beads (Amersham Biosciences, Uppsala, Sweden) followed by purification, washing, denaturation using 0.2 M NaOH, and an additional wash. A pyrosequencing primer (0.3 μM; supplementary material Table S1) was annealed to the purified single-stranded PCR product. Pyrosequencing and methylation quantification were performed using the PyroMark Q24 Pyrosequencing System (Qiagen).

To examine the role of DNMT1 in retinal development, we crossed homozygous mutant mice (Dnmt1^fl/fl, with exons 4 and 5 flanked by loxP sites; Fig. 1A) (Jackson-Grusby et al., 2001) with a line carrying the Rx-Cre transgene [Rx, retina and anterior neural fold homeobox (Rax) - Mouse Genome Informatics] (Medina-Martinez et al., 2009), which results in specific excision of Dnmt1 exons 4 and 5 during early development of the neural retina (NR), RPE and posterior pituitary/ventral hypothalamus (Muranishi et al., 2012). The mutant mice were runted compared with their littermate controls and had small eyes, with eyelids opening only partially (and never completely) and later than in controls (Fig. 1B,C). PCR analysis of genomic DNA from NR and RPE/choroid (RPE-Ch) samples, obtained from at least 30 control (Dnmt1^fl/fl) and mutant (Dnmt1^fl/fl:Rx-Cre/+) mice at different developmental stages, revealed the predicted excision of exons 4 and 5 (Jackson-Grusby et al., 2001); the excision is more efficient in NR (>95% in most animals at all ages) than in RPE-Ch (Fig. 1D), presumably because the Rx-Cre transgene is expressed in RPE but not in the contaminating choroidal tissue (Swindell et al., 2006) (a representative blot is shown in Fig. 1D). We noted that the choroid remained associated with RPE and could not be removed during the dissection quickly enough to avoid RNA degradation. Sequencing of RT-PCR products using primers in exon 1 and exon 9 demonstrated the presence of two Dnmt1 transcripts in mutant retinas: one with a reported out-of-frame mutation resulting from the deletion of exons 4 and 5 (Δ4-5) (Jackson-Grusby et al., 2001) and another (Δ4-6) with an in-frame deletion due to the skipping of exon 6 during splicing (Fig. 1A; data not shown).

qRT-PCR analysis using RNA from NR and RPE-Ch revealed that Dnmt1 expression in mutants was almost half that in controls at P0.5 and even at E14.5 (supplementary material Table S3; data not shown). As a measure of the efficiency of Dnmt1 deletion, we performed co-immunoprecipitation with PCNA because the DNMT1 domain encoded by exons 4-6 is known to be important for PCNA binding (Spada et al., 2007); thus, the deletion event in mutant mice should lead to decreased PCNA co-immunoprecipitation. Reduced levels of DNMT1 were evident by immunoblot analysis in P0.5 mutant NR compared with control NR (data not shown). Consistent with excision, our co-immunoprecipitation analysis using NR extracts revealed a 3- to 4-fold reduction in the amount of PCNA pulled down by mutant DNMT1 compared with the wild type (Fig. 1E).

We observed a modest reduction of total DNA methylation, as measured by methylation of the LINE1 element (usually 90-100%) and methylation of the imprinted gene H19 (usually ∼50%) in mutant NR (Fig. 1F), as indicated previously in cells carrying the in-frame Δ3-6 Cre-mediated excision (Spada et al., 2007). LINE1 methylation in control NR was 93.5±0.2% compared with 82.6±5% in mutant NR (n=3 each) with marginal significance (P=0.0951), whereas tail DNA of mutants and controls showed similar methylation (92.2±0.2% controls versus 92.4±0.6% mutants, n=3 each). H19 methylation at six CpGs in mutant NR (43.9±2.8%, n=4) was reduced compared with control NR (50.9±0.9%, n=3) (P=0.0993). In tail DNA, H19 methylation was similar in control and mutant samples.

Our data thus point to a knockdown of DNMT1 function (hypomorph) in Dnmt1^fl/fl:Rx-Cre/+ mice caused by moderately reduced expression of mutant DNMT1 with an in-frame deletion of the PCNA-interacting domain.

The Dnmt1^fl/fl:Rx-Cre/+ mice usually died 2-4 weeks after birth, probably reflecting the secondary neuroendocrine complications due to Rx-Cre expression in the ventral hypothalamus/posterior pituitary (Nasonkin et al., 2004). Nonetheless, the mutant mice survived long enough to allow investigation of retinal differentiation. Histological analysis of Dnmt1^fl/fl:Rx-Cre/+ retinas demonstrated normal lamination, but there were clearly a number of abnormalities. Retinal layers, especially the outer nuclear layer (ONL) and the inner nuclear layer (INL), showed variable thickness and were uniformly thinner than in wild-type retina (Fig. 2A). The RGC layer also displayed reduced cell numbers (Fig. 2A; supplementary material Fig. S1A), which correlated with noticeable thinning of the optic nerve (supplementary material Fig. S1B).

A key aspect of the mutant retina phenotype was consistent detachment of the NR from the RPE at E16.5 (Fig. 2A, arrows), or even earlier at E12.5-14.5. Indeed, poor adhesion between the NR and RPE in the mutant, but not in the control, eyes provided a robust way to sort the mutant and control mice even before genotyping. Additional electron microscopy (EM) studies demonstrated the absence of photoreceptor OSs in P15.5 mutant retina (Fig. 2B,C) and poorly developed microvilli in the RPE (Fig. 2C, arrowheads). Only mitochondria-rich inner segments could be observed in between photoreceptor nuclei and the RPE (Fig. 2C, asterisks).

As DNMT1 can modulate chromatin conformation (Jaenisch and Bird, 2003), we explored the chromatin patterns in mutant retina using antibodies against heterochromatin- and euchromatin-specific histone modifications (H4K20Me3 and H3K4Me3, respectively). The Dnmt1^fl/fl:Rx-Cre/+ retina showed significant and consistent changes in the typical heterochromatin, and to a lesser extent euchromatin, staining patterns in the ONL and INL (Fig. 3A). Specifically, distinct rod- and cone-specific heterochromatin patterns observed in control mouse retina (Carter-Dawson and LaVail, 1979; Nasonkin et al., 2011) were not apparent, and the typical strong nuclear heterochromatin-specific signal was absent in the outer half of the INL layer (Fig. 3A). The DNMT1 staining that colocalized with the euchromatin marker H3K4Me3 in control cone photoreceptors (Nasonkin et al., 2011) was undetectable in most P15.5 mutant retinas (Fig. 3B), pointing to a likely change in chromatin conformation. Mutant RPE did not exhibit major changes in heterochromatin and euchromatin staining patterns within the nuclei, even though the cells were abnormal (Fig. 3B).

To evaluate the effect of Dnmt1 knockdown on retinal cell fate, we performed high-resolution confocal microscopy using antibodies that identify distinct cell types. Immunostaining for photoreceptor-specific markers - rhodopsin, PRPH2 and recoverin - revealed their near normal expression in inner segments (Fig. 4A,B; supplementary material Fig. S2) of Dnmt1^fl/fl:Rx-Cre/+ retina, even though OSs were absent (see Fig. 2). Surprisingly, there was a significant underrepresentation of cones in P15.5 mutant retina as shown by peanut agglutinin (PNA) staining (Fig. 4B,C), a reduced number of M-cones as measured by M-opsin antibody (Fig. 4B), and a complete absence of S-opsin immunostaining (Fig. 4B). In concordance, Opn1sw (S-opsin) RNA expression was reduced by at least 50% in mutant retina as assessed by qRT-PCR analysis (Fig. 4D).

We investigated whether altered neuroendocrine function in Dnmt1^fl/fl:Rx-Cre/+ mice might contribute to the striking difference in M-cones and almost complete absence of S-opsin-expressing cones, as cone opsin expression is controlled by thyroid hormone in developing and mature retina (Roberts et al., 2006; Glaschke et al., 2011). We examined the cones in adult hypopituitary Ames dwarf (Prop1^df/df) mice. PNA staining (which labels both M- and S-cone matrix sheath) of the Prop1^df/df retina revealed a normal density of cones (per field), as compared with age-matched wild-type littermates (supplementary material Fig. S3).

Consistent with a thinner (about half that of controls) and pallid optic nerve, the number of RGCs was reduced in Dnmt1^fl/fl:Rx-Cre/+ mutants, as evidenced by BRN3A and BRN3B (POU4F1 and POU4F2 - Mouse Genome Informatics) staining at E16.5 and P0.5 (see supplementary material Fig. S1). We did not observe a significant difference in immunostaining using specific markers of other retinal cell types - ON bipolar neurons (PKCα), horizontal cells (CALB1 and PROX1), amacrine (CALB2) and Müller glia (cyclin D3 and glutamine synthetase) - in P15.5 Dnmt1^fl/fl:Rx-Cre/+ retina (supplementary material Figs S4-S6; data not shown).

To understand the molecular basis of NR abnormalities, specifically the lack of photoreceptor OSs, we generated expression profiles of Dnmt1^fl/fl control and Dnmt1^fl/fl:Rx-Cre/+ retina at different developmental stages using exon arrays (GeneChip, Affymetrix) (supplementary material Table S4). Surprisingly, we did not observe significant changes in the expression of a number of genes that encode transcription factors, phototransduction (except a few cone genes, including S-opsin) and/or cilia proteins associated with homeostasis, OS biogenesis or intracellular trafficking (e.g. Nrl, Crx, Mef2c, Esrrb, Nr2e3, Gnat1, Prph2, Rom1, Cnga3, Aipl1, Rpgr, Rpgrip1, Tulp1, Prom1, Mkks, Bbs2, 4, 5, 7, 9) (data not shown). A diverse set of genes showed aberrant (generally higher) expression in the mutant retina at early stages of development (E16.5 and P0.5; supplementary material Tables S3, S4), a likely result of hypomethylation at distinct genomic loci. However, the expression of most genes was unaltered at later stages of photoreceptor development (e.g. at P6.5 and P10.5; supplementary material Tables S3, S4). The greatest increase in expression was detected for Tuba3a, the expression of which is strongly affected by methylation (Borgel et al., 2010).

Extensive retinal detachment, together with striking abnormalities in photoreceptors, led us to examine the RPE in Dnmt1^fl/fl:Rx-Cre/+ retina. As early as E16.5, a highly disorganized RPE monolayer contained cells of variable height and pigmentation, abnormally sized nuclei and variable compaction of heterochromatin (Fig. 5A). This disorganized monolayer was particularly evident in flat-mounted RPE stained with phalloidin to decorate F-actin and with DAPI to identify nuclei (Fig. 5B). Differential interference contrast (DIC) or brightfield microscopy revealed homogeneous pigmentation of control RPE cells but variable degrees of pigmentation in mutant RPE (Fig. 5B, bottom panels).

Analysis of 3D confocal stacks demonstrated the variable position of junctional actin in mutant RPE, in striking contrast to the uniform height and position of junctional actin in control RPE (green signal; Fig. 5B, bottom; 5C, top) and less F-actin on the apical surface of mutant RPE cells, indicative of underdeveloped apical microvilli (green signal; Fig. 5C, middle). β-catenin (a lateral membrane marker) and adherens junctions were also highly disorganized in mutant RPE (red signal; Fig. 5C, bottom). Confocal imaging of P15.5 mutant and control retina vibratome sections demonstrated shorter microvilli in mutants (Fig. 5D) and a highly disorganized distribution of ezrin (green signal; Fig. 5C, bottom), a scaffolding protein that plays a key role in the development of apical microvilli and basal infoldings of RPE (Bonilha et al., 1999). In agreement, EM analysis at P15.5 revealed highly organized microvilli and basal infoldings in control RPE and poorly developed microvilli and basal infoldings in mutant RPE (Fig. 5E). Notably, labeling with antibody to rootletin (Crocc) in the RPE of Dnmt1^fl/fl:Rx-Cre/+ mice identified irregular ciliary rootlets (supplementary material Fig. S7), providing further evidence of disruption of RPE organization and polarity.

To decipher the mechanism of defective photoreceptor OS biogenesis, we used different Cre transgenic lines to selectively knockdown Dnmt1 in RPE and/or NR. We bred Dnmt1^fl/fl mice with those expressing a doxycycline (DOX)-inducible VMD2-Cre transgene (Le et al., 2008), which is specifically expressed in RPE (Fig. 6A-E). The genomic PCR assay confirmed the predicted ablation of Dnmt1 exons 4 and 5 in RPE but not in the NR of Dnmt1^fl/fl:VMD2-Cre DOX⁺ mice (Fig. 6A).

About 12% of the mutant, but not control, newborn pups demonstrated microphthalmia and RPE hypopigmentation (Fig. 6B). The eyes of mutant pups were frequently smaller (Fig. 6C). Strikingly, continuous administration of DOX in pregnant mice to excise exons 4 and 5 as early as VMD2-Cre transgene expression (Le et al., 2008) resulted in shortening of the OSs in Dnmt1^fl/fl:VMD2-Cre retina but not in control Dnmt1^fl/fl or heterozygous retina (Fig. 6D). The dynamics of OS elongation was dependent on the efficiency of DOX-induced VMD2-Cre transgene activation, which can be inherently variable (Le et al., 2008). As DOX was included in the diet, the timing and extent of its effect on the OS in different mice, even within a litter, were different; yet, the defects in OS morphogenesis were clearly evident by P15.5. Mutant retina displayed areas of NR/RPE detachment as observed in Dnmt1^fl/fl:Rx-Cre/+ mice (Fig. 6D, arrows). The RPE flatmount preparations also revealed a disrupted actin cytoskeleton (Fig. 6E), in accordance with the data from Dnmt1^fl/fl:Rx-Cre/+ pups.

We then used Pax6-[α]Cre (Marquardt et al., 2001), Six3-Cre (Furuta et al., 2000) and Crx-Cre transgenes (Nishida et al., 2003) to excise Dnmt1 exons primarily in the developing NR. We independently validated the effectiveness of Six3-Cre and Crx-Cre transgenes using the ROSA26-lacZ reporter (data not shown). The Pax6-[α]Cre mice express robust Cre activity in the retinal periphery but not in the central retina (Marquardt et al., 2001; Bäumer et al., 2002). As predicted, we observed rapid degeneration of the peripheral but not central retina in Dnmt1^fl/fl:Pax6-[α]Cre/+ mice (Fig. 6F), and rapid degeneration of predominantly central NR but not RPE in Dnmt1^fl/fl:Six3-Cre/+ retina (Fig. 6G). The mutant retina in both instances had normal OS elongation and RPE morphology before NR degeneration. The only phenotype displayed by Dnmt1^fl/fl:Crx-Cre/+ retina was retinal degeneration that noticeably developed by P15.5 (data not shown). OSs, however, elaborated normally.

Based on the gene profiling data (supplementary material Table S4), we selected 75 genes that might impact photoreceptor differentiation and/or maturation and performed qRT-PCR analysis in control and mutant RPE-Ch (supplementary material Table S5). We observed a substantial reduction in the expression of Ihh [which is expressed in choroid (Dakubo et al., 2008)] at P0.5 (5.2-fold) and P6.5 (3.3-fold), Ptch2 (2.2-times less at P0.5 and 2.3-times less at P6.5), Wnt3a (1.5-times less at P6.5) and Sfrp5 (2-times less at P0.5). The expression of two Notch pathway genes, Hes1 and Dll1, was also more than halved at P0.5 and P6.5, whereas Fzd5 expression was increased by 2.3-fold at P0.5. Notably, high expression of Tuba3a was detected at E14.5 (80-fold) and P6.5 (132-fold). However, a majority of early RPE development genes showed no significant change in expression in qRT-PCR analysis using the criteria of fold change ≥2.0 with P≤0.05 in a set of four biological replicates at each time point and for each genotype (Dnmt1^fl/fl and Dnmt1^fl/fl:Rx-Cre/+).

To elucidate how Dnmt1 knockdown affected NR versus RPE methylation and consequently photoreceptor development, we examined global DNA methylation [LINE1 elements (Poage et al., 2011)] and the methylation of two genes, Tuba3a and Opn1sw, that exhibited significantly altered expression in Dnmt1^fl/fl:Rx-Cre/+ mutant retina (Fig. 7). For LINE1 and Tuba3a, genomic DNA was isolated from NR and RPE, obtained by laser capture microdissection (LCM) from E16.5 and P0.5 retina of control and mutant mice, and used for pyrosequencing (Ronaghi et al., 1998). A significantly greater reduction in global DNA methylation was detected in mutant RPE as compared with mutant NR (Fig. 7A). Methylation of Tuba3a, which showed dramatically higher expression in both mutant RPE and NR by qRT-PCR, was reduced in mutants compared with controls at five CpG sites located near the transcription start site; however, RPE exhibited more pronounced demethylation than NR (Fig. 7C,D). Our results imply a direct effect of hypomethylation at the locus on Tuba3a transcription.

Pyrosequencing of three CpG sites at Opn1sw (S-opsin, expressed in S-cones) using P0.5 NR genomic DNA identified a substantial decrease in methylation in mutants (61.9±6.8% versus 87.0±0.6% in controls, n=5; Fig. 7B). Reduced CpG methylation, yet decreased expression of S-opsin, in the mutant retina (see Fig. 4D) suggests an indirect effect via concurrent demethylation at an unidentified suppressor of Opn1sw.

Here, we report for the first time that Dnmt1 knockdown in RPE, but not in NR, results in specific changes in RPE structure, and possibly function, and is associated with aberrant photoreceptor development and lack of OS morphogenesis. The complete absence of OSs is not due to major changes in the expression of phototransduction or cilia transport genes. Yet, reduced DNMT1 function in early retinal development does not affect cell type specification and lamination. Consistent with our findings, a recently reported NR-specific knockdown of Dnmt1 using a Chx10-Cre driver (Chx10 is also known as Vsx2 - Mouse Genome Informatics) revealed the generation of all classes of retinal cell types; however, this mutant exhibited abnormal expression of rhodopsin and M-opsin and a progressive loss of all nuclear layers (Rhee et al., 2012). In addition to revealing a novel role of RPE in photoreceptor maturation, our Dnmt1 knockdown mutants provide useful models with which to dissect the epigenetic control mechanisms and signaling molecules that guide the differentiation of the photoreceptor-RPE interface.

Epigenetic reprogramming is established in early stages of development, and recent studies indicate a continuous dynamic control of the epigenome (Branco et al., 2008; Illingworth et al., 2008; Wu and Zhang, 2011; Yaish et al., 2011). DNMT1 is primarily responsible for maintaining DNA methylation and is necessary for survival; however, its interaction with PCNA does not appear to be crucial for enzymatic function (Egger et al., 2006; Chen et al., 2007; Spada et al., 2007; Hirasawa et al., 2008). Interestingly, the loss of the PCNA-binding domain in the Dnmt1 knockdown mutant reported here did not broadly impact eye development, acquisition of NR and RPE cell fate or gene expression patterns. Nonetheless, subtle changes in the expression of genes in the differentiating RPE-Ch can be associated with altered cell shape, size and polarity. Of specific importance are alterations in several signaling molecules and transcription factors, including Bmp2, Bmp4, Fzd5, Hes1, Ihh, Mitf, Ptch2 and Sfrp5. Further investigations are necessary to delineate their precise roles in RPE development.

The genesis of OSs that include the complete machinery for phototransduction is a hallmark of ciliated photoreceptors and is initiated at ∼P9 in mice. The relatively small changes in the expression of photoreceptor genes that we observed seems unlikely to account for RPE detachment and the complete absence of OSs in the Dnmt1 mutant mice. We were intrigued by a possible direct link between RPE integrity and photoreceptor differentiation, as reflected in the phenotypes of the various Dnmt1 mutants created with RPE and/or NR-Cre lines. RPE secretes a many chemokines and cytokines (Shi et al., 2008); however, our data also suggest a key role for cell-cell interaction in producing the requisite inductive signals for OS biogenesis.

The expression of mutant Dnmt1 resulted in small mice with variably small eyes. The small size of the mice could be due to hormonal insufficiency (Nasonkin et al., 2004) resulting from Rx-Cre activity in the ventral hypothalamus/posterior pituitary in addition to the developing eye (Swindell et al., 2006; Medina-Martinez et al., 2009). However, the small eye phenotype of Dnmt1^fl/fl:Rx-Cre/+ mice is likely to be a consequence of the expression of mutant Dnmt1 in the NR and RPE rather than a result of hypothalamic insufficiency, as hypopituitary Prop1^df/df mice have normal sized eyes (data not shown) and a normal rod:cone ratio (supplementary material Fig. S3).

Our studies suggest an instructive function of RPE in the biogenesis of photoreceptor OSs, which are elaborated from a uniquely modified primary cilium (Liu et al., 2007). The primary cilium in RPE develops early, but disappears once the retina is mature (Nishiyama et al., 2002). In the adult, RPE microvilli are in close contact with photoreceptors. We hypothesize that if the RPE cilium or associated signaling is disrupted, it could prevent photoreceptor OS biogenesis. Disorganization of ciliary rootlets (supplementary material Fig. S7) and reduced Ihh expression in postnatal RPE-Ch, as observed in Dnmt1^fl/fl:Rx-Cre/+ mice, might indicate defective signaling through the RPE cilium. Misregulation of Ihh has been linked to aberrant photoreceptor development (Dakubo et al., 2008). Significant overexpression of Tuba3a, a microtubule protein present in primary cilia (Arikawa and Williams, 1993), in mutant NR and RPE is consistent with this hypothesis. The Tuba3a promoter in NR and RPE indeed showed a significant decrease in proximal promoter methylation, in agreement with the drastic upregulation of Tuba3a expression. Additionally, we observed expression changes in cilia-associated genes in both NR and RPE and in signaling molecules that might control RPE polarization (Ihh, Wnt/Frizzled pathways). Hence, misregulation of cilia formation in RPE might be a contributing factor to altered RPE-photoreceptor interaction during postnatal retinal development.

To our surprise, statistically significant gene expression changes in the NR and RPE of mutants were limited to a few genes, including Ihh, Tuba3a, Opn1sw and Dnmt1 itself (supplementary material Tables S3-S5). We believe that a single cause-effect mechanism for the failure of OS elongation is unlikely. Instead, epigenetic (DNA methylation) mechanisms tightly modulate the expression of a number of key NR and RPE genes, and these expression changes cumulatively lead to a pronounced retinal phenotype associated with the lack of photoreceptor OSs. Our findings are consistent with reports of DNMT1 mutations that are likely to affect hearing and neurological phenotypes via pleiotropic mechanisms (Klein et al., 2011; Winkelmann et al., 2012).

Our studies have translational implications for cell-based therapies of retinal degenerative diseases. The photoreceptors produced from embryonic or induced pluripotent stem cells do not generate OSs, which are crucial for phototransduction (Osakada et al., 2008; Meyer et al., 2009). In an in vitro three-dimensional model that self-organized into laminated retina from embryonic stem cell aggregates, no photoreceptor OSs were observed (Eiraku et al., 2011), probably because RPE was removed before culturing the optic vesicle. Interestingly, when early (embryonic weeks 8-17) human fetal retina or mouse photoreceptor precursors are transplanted into the subretinal space immediately next to the RPE layer, such grafts undergo lamination and graft-derived photoreceptors show well-developed OSs (Sagdullaev et al., 2003; MacLaren et al., 2006; Lamba et al., 2009). The availability of stem cell-derived RPE (Klimanskaya et al., 2004; Idelson et al., 2009; Salero et al., 2012) and efficient culture protocols for polarized RPE (Sonoda et al., 2009; Bharti et al., 2011) will allow us to test whether developing and/or mature RPE is needed for photoreceptor differentiation. The identification of RPE-derived signals that instruct developing photoreceptors to produce OSs will greatly assist the design of treatment paradigms for neurodegenerative diseases involving photoreceptor dysfunction or death.

We thank Rudolf Jaenisch for Dnmt1^fl/fl, Guillermo Oliver for Six3-Cre, Peter Gruss for [α]Cre and Michael Masternak for Ames Prop1^df/df mice; and Tiansen Li, Chi-Chao Chan, Mary Alice Crawford, Tudor Badea, Harsha Rajasimha and Dustin Hambright for assistance and advice.