Molecular Vision 2014; 20:56-72
Received 27 August 2013 | Accepted 02 January 2014 | Published 06 January 2014
1Department of Biological Sciences, Purdue University, West Lafayette, IN; 2Department of Statistics, University of Georgia, Athens, GA; 3Department of Cell Biology, Neurobiology and Anatomy, Medical College of Wisconsin, Milwaukee, WI; 4Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA; 5Joint Shantou International Eye Center, Shantou University and the Chinese University of Hong Kong, Shantou, China; 6Department of Biochemistry and Molecular Biology, Indiana University School of Medicine Lafayette, West Lafayette, IN
Correspondence to: Yuk Fai Leung, Department of Biological Sciences, Purdue University, LILY 2-236, 915 W. State Street, West Lafayette, IN, 47907; Phone: 765-496-3153; FAX: 765-494-0876; email: firstname.lastname@example.org
Dr. Zhang is now at the Department of Ophthalmology, University of Cincinnati, Cincinnati, OH.
Purpose: The purpose of this study was to develop a framework for analyzing retinal pigment epithelium (RPE) expression profiles from zebrafish eye mutants.
Methods: The fish model we used was SWI/SNF-related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 4 (smarca4), a retinal dystrophic mutant with a previously described retinal phenotype and expression profiles. Histological and Affymetrix GeneChip analyses were conducted to characterize the RPE defects and underlying differential expression, respectively.
Results: Histological analysis revealed that smarca4 RPE was formed, but its differentiation was abnormal. In particular, ultrastructural analysis of smarca4 RPE by transmission electron microscopy demonstrated several defects in melanogenesis. The nature of these defects also suggests that the cytoskeletal dynamics, which are tightly linked with melanogenesis, were impaired in smarca4 RPE. To compare the expression profile of normal wild-type (WT) and smarca4 RPE, the gene expression profiles of microdissected retinas and RPE-attached retinas were measured with Affymetrix GeneChip analysis. The RPE expression values were then estimated from these samples by subtracting the retinal expression values from the expression values of the RPE-attached retinas. A factorial analysis was conducted using the expression values of the RPE, retinal, and whole-embryo samples. Specific rules (contrasts) were built using the coefficients of the resulting fitted models to select for three groups of genes: 1) smarca4-regulated RPE genes, 2) smarca4-regulated retinal genes, and 3) smarca4-regulated RPE genes that are not differentially expressed in the retina. Interestingly, the third group consists of 39 genes that are highly related to cytoskeletal dynamics, melanogenesis, and paracrine and intracellular signal transduction.
Conclusions: Our analytical framework provides an experimental approach to identify differentially-regulated genes in the retina and the RPE of zebrafish mutants in which both of these tissues are affected by the underlying mutation. Specifically, we have used the method to identify a group of 39 genes that can potentially explain the melanogenesis defect in the smarca4 RPE. In addition, several genes in this group are secreted signaling molecules. Thus, this observation further implicates that the smarca4 RPE might play a role in the retinal dystrophic phenotype in smarca4.
The retinal pigment epithelium (RPE) is a single layer of pigmented epithelial cells that supports the function and development of photoreceptors . RPE dysfunction leads to many retinal degenerative diseases, including age-related macular degeneration (AMD)  and retinitis pigmentosa (RP) [3,4]. Regeneration  and transplantation [6,7] of the RPE could lead to new treatments for these retinal diseases. The success of these therapeutic approaches will rely on our understanding of the genetic regulation of RPE differentiation and paracrine signaling to the retina. A major technical limitation in studying RPE signaling is the difficulty obtaining pure and intact RPE tissue from small developing embryos, which has precluded accurate expression profiling of the RPE during development. We previously addressed this issue by developing an approach for estimating the RPE gene expression profile in the zebrafish by comparing microdissected RPE-attached retinas and RPE-free retinas [8,9]. Subsequently, expression profiling of developing chick  and human  RPE has been reported. Higdon and colleagues recently demonstrated that RPE-specific gene expression could be detected in purified pigment cells from whole zebrafish embryos, using density gradient centrifugation and fluorescence-activated cell sorting (FACS) based on the pigment granule density of the cells . The availability of these data sets has substantially facilitated the study of RPE development and the identification of RPE paracrine signals to the retina.
Several paracrine signaling molecules are known to play a role in mediating the interactions between the RPE and the retina during development. For example, pigment epithelium-derived factor (PEDF) is a glycoprotein that has been shown to mediate normal photoreceptor development in the frog  and chicken . Two additional signaling molecules, glial cell-derived neurotrophic factor (GDNF) and brain-derived neurotrophic factor (BDNF) are released by cultured human RPE cells, and their presence in culture medium enhances the survival of dopaminergic neurons . Similar to PEDF, GDNF also regulates photoreceptor development in chickens [14,16]. Another class of signaling molecules, Bone morphogenetic protein (Bmp), is also involved in the development of the retina and the RPE. For instance, Bmp4 and Bmp7 expressed in the surface ectoderm overlying the optic vesicle in the chick are essential and sufficient for RPE specification . In addition, Bmp2 and Bmp4 in adult bovine RPE act as potential negative growth regulators and are downregulated during injury in the retina and the RPE . Finally, the canonical wingless (Wnt) signal transduction pathway was recently implicated in RPE development and retinal degeneration. In particular, Wnt activity is essential for transcriptional activation of Mitf and Otx2, two genes that are crucial for RPE specification in mice [19,20]. Interestingly, the conditional knockout of β-catenin in RPE not only affects RPE differentiation but also disrupts retinal morphogenesis and lamination. These findings suggest that Wnt likely mediates signaling interactions between the RPE and the retina.
Although the study of the signaling molecules described above has contributed to our understanding of the interactions between the RPE and the retina, it is clear that only a fraction of the vast genetic network that underlies RPE and retinal development has been described. The purpose of this study was to establish a genomic approach for identifying genes that control RPE differentiation and potential paracrine signals specifically secreted by the RPE. The model we used was a zebrafish SWI/SNF-related, matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 4 (smarca4; also known as yng) mutant . This mutant has a null mutation in smarca4, and was originally described as a model with developmental problems in several organs, including the eye, ear, and heart . Smarca4 codes for the ATPase of the SWI/SNF chromatin remodeling complex, which is critical to regulation of gene expression during development. As a result of dysregulated gene expression, the retinal structure is disorganized, and appropriate retinal lamination is disrupted in the smarca4 mutants. The retinal cells are specified but do not fully differentiate. We previously conducted an expression profiling experiment to identify candidate genes that underlie the retinal differentiation phenotype with microarray analysis . The true discovery rate of the study was more than 90% and was highly concordant with the theoretical rate of 95% . Subsequent functional characterizations of the candidate genes have also begun to reveal the gene network regulated by Smarca4 and its role in normal retinal differentiation [25-27].
In addition to the effect on retinal development, the smarca4 mutation also plays a role in RPE differentiation. In fact, one of the earliest hallmarks used to identify smarca4 was abnormal RPE pigmentation. We observed the integrity of the RPE layer and its adhesion to the retina were also compromised during our microdissection of smarca4 retinas . We hypothesized that these defects were caused by abnormal gene expression in smarca4 RPE. In this study, we first characterized the defects in smarca4 RPE differentiation with histological analysis. Then, we used our RPE expression analysis approach [8,9] to obtain RPE gene expression in smarca4. To identify differential gene expression and potential paracrine signal transduction that might underlie the smarca4 eye phenotype, an extension of our factorial microarray array analysis  was established. In particular, we used the fitted statistical models to build sequential rules (contrasts) to narrow down the list of possible candidate genes.
The following zebrafish lines were used in this study: wild-type (WT) AB and smarca4a50/+ (yng), which was originally generated from the AB line [21,22]. The WT parents were the genotyped WT siblings of the smarca4a50/+, and are isogenic other than the smarca4a50 allele. The adult breeders were maintained according to standard procedures . Embryo collection, staging, and incubation were performed according to standard procedures. The embryos used for an individual experiment were collected from the same parents and spawned within a 15 min interval. All protocols were approved by the Purdue Animal Care and Use Committee.
Microdissection of retinas and RPE-attached retinas from WT and smarca4 a50/ a50 embryos at 52 h post-fertilization (hpf) was conducted as described . WT retinas (WR52) and WT RPE-attached retinas (WRR52) were originally collected and reported in , smarca4/yng retinas (YR52) were originally collected and reported in , and smarca4 RPE-attached retinas (YRR52) were collected in this study. Three biologic replicates were collected for each condition, and the number of tissues in each replicate is shown in Table 1.
Total RNA extraction was performed with an optimized procedure that combined TRIzol (Life Technologies, Grand Island, NY) and column-based purification (Qiagen, Valencia, CA) [8,29]. The yield (Table 1) and the quality of the purified total RNAs were evaluated with NanoDrop spectrophotometry (Thermo Scientific, Wilmington, DE) and Bioanalyzer electrophoresis (Agilent Technologies, Santa Clara, CA), respectively. Total RNAs were amplified and labeled using a two-cycle target-labeling protocol (Affymetrix, Santa Clara, CA), and hybridized to GeneChip Zebrafish Whole Genome arrays (Affymetrix). The input RNA amounts are specified in Table 1. Hybridization, washing, and scanning were performed according to the manufacturer’s standard procedure.
Histological analysis on 1 μm plastic sections was conducted as previously described  and imaged with a SPOT-RT3 color slider camera (Diagnostic Instruments, Sterling Heights, MI) mounted on a BX51 fluorescence compound microscope (Olympus, Center Valley, PA). To analyze the ultrastructural changes in the RPE, transmission electron microscopy (TEM) was used. The samples were fixed and processed as previously described . Ultrathin sections (100 nm) through the optic nerve region were collected for TEM analysis with a Philips CM-10 transmission electron microscope (FEI Company, Hillsboro, OR). The resulting images were merged in Adobe Photoshop CS6 (Adobe, San Jose, CA).
In situ hybridization was conducted as described . The following riboprobes were used in this study: dopachrome tautomerase (dct; GenBank accession number: NM_131555) and retinal pigment epithelium-specific protein 65a (rpe65a; NM_200751). To prepare the probes for these genes, a fragment of each gene was amplified from a cDNA library prepared as described . The primer sequences are as follows: dct-1F: 5′-ACT TCT TCG TCT GGC AGC AT-3′, dct-1R: 5′-CGG CTT ATC ATA TCC CTC CA-3′, rpe65–1F: 5′- GCT TCG AGT CGG ATG AAG AG-3′ and rpe65–1R: 5′-CAG GGA CGA AAT GGT TGA GT-3′. The resulting PCR fragment was cloned into the pGEM-T easy vector (Promega, Madison, WI) for propagation. The riboprobes were synthesized according to standard procedures .
Melanosome and RPE parameters were measured and extracted from the images with i-Solution 10.1 (IMT i-Solution, Burlington, Canada). The morphology of the melanosomes was measured by area and roundness. The latter parameter is defined as 4A/f2π (A = area, f – max Feret diameter (the longest diameter along the region-of-interest boundary). Thus, 1 = round, 0 = elongated). Standard error propagation was used to combine the measurement errors of the variables. All standard descriptive statistics and data analyses were performed in the R statistical environment version 2.15.2. Melanosome numbers, area and roundness, as well as RPE area were analyzed with standard ANOVA. The melanosomes along the apical/basal axis was demarcated by the center of the RPE nuclei and their number counted. The resulting data were analyzed with logistic regression. An alpha level of 0.05 was used for all general statistical tests.
The analysis conducted in this study used the WR52, WRR52, YR52, and YRR52 data as described, as well as the whole embryo data obtained from WT (WA52) and smarca4 (YA52) previously collected . The microarray data were deposited at the Gene Expression Omnibus (GEO) under the accession number (GSE50241).
The probe-level data of these sample groups were background-adjusted, normalized, and summarized with a robust multiarray average (RMA) algorithm  implemented in the affy library of the Bionconductor  in R statistical environment version 2.15.2, using default parameters. RPE expression values were estimated from the comparison between the retinal samples and the RPE-attached retinal samples of the same genotype based on a method we previously developed . First, the RMA-normalized expression values in RPE-attached retinal sample were adjusted according to the yield and number of tissues with the following equations:
adjExprWRR52ij=ExprWRR52ij x (mean(yieldWRR52j/tissueNoWRR52j)/mean(yieldWR52j/tissueNoWR52j))
ExprWR52ij (i.e., no adjustment)
adjExprYRR52ij=ExprYRR52ij x (mean(yieldYRR52j/tissueNoYRR52j)/mean(yieldYR52j/tissueNoYR52j))
ExprYR52ij (i.e., no adjustment)
where Expr is the expression value of a gene i in replicate j, adjExpr is the adjusted expression value of a gene i in replicate j, yield is the total RNA yield of replicate j, and tissueNo is the number of tissue used in replicate j; i=1,…,n; j=1,2,3.
Then, the RPE expression in WT and smarca4 was estimated as follows:
RPEWT52=adjExprWRR52ij – ExprWR52ij
RPEyng52=adjExprYRR52ij – ExprYR52ij
A factorial analysis  was conducted using the adjusted RPE expression values, as well as the retinal and whole-embryo values. The overall design is a 3×2 model (Figure 1A) that determines the effect of tissue type (T) and mutation (M) on the expression level. The corresponding levels of each factor are listed below:
Mutation (M) – two levels: WT and smarca4
Tissue (T) – three levels: whole embryo, retina, and RPE
For any gene g, its expression (yg) in the six experimental conditions in Figure 1A was modeled with the following equations:
WTembryo: yg = μ (Eq. 2.1)
WTretina: yg = μ + TR (Eq. 2.2)
WTRPE: yg = μ + TRPE (Eq. 2.3)
smarca4embryo: yg = μ + M (Eq. 2.4)
smarca4retina: yg = μ + M + TR + M*TR (Eq. 2.5)
smarca4RPE: yg = μ + M + TRPE + M*TRPE (Eq. 2.6)
The candidate genes for a specific biologic question were selected by building contrast with the coefficients in Equations 2. In other words, two conditions were statistically compared to identify differentially expressed genes among the conditions (see  for further discussion and examples). Depending on whether the two-way interaction term in the Equations 2 was significant or not, the resulting contrasts had to be built differently. In other words, there were first-order (two-way interaction term is not significant) and second-order models (two-way interaction term is significant). Multiple hypothesis testing was corrected by calculating the false discovery rate (FDR). A gene was inferred as differentially expressed when the contrast in consideration had a q value less than 0.001 unless specified otherwise. By combining multiple contrasts, specific rules were built to select 1) smarca4-regulated RPE genes, 2) smarca4-regulated retinal genes, and 3) smarca4-regulated RPE genes that are not differentially expressed in the retina (see the corresponding Result sections for the list of contrasts used). A schematic diagram that shows the relationship between these selections and the resulting candidate genes is shown in Figure 1B. General gene annotations were adopted from an annotation file from Affymetrix (Zebrafish annotation release 30). The selected RPE genes were also analyzed with the Database for Annotation, Visualization, and Integrated Discovery (DAVID) v6.7 [34,35] using default parameters. Further grouping was aided by the fold change of a comparison. If an estimated expression value was less than zero, indicating the gene was not expressed in the condition, the value was artificially set as one.
One distinctive feature of smarca4 mutant was the lower level of pigmentation in pigment cells, including the RPE. The smarca4 RPE cells developed pigmentation at around 24 hpf, a stage when the WT RPE cells began to differentiate and form pigment. However, the pigmentation level was always less intense in smarca4. By 52 hpf, the whole eye was covered with RPE and appeared dark in the WT (Figure 2A,E), while the smarca4 RPE cells were noticeably less pigmented, particularly on the ventral side (Figure 2B,F, red arrow). At this stage, there is differentiation of several retinal cell types in the WT eye, and retinal lamination is present [30,36,37]. In contrast, the smarca4 retina at 52 hpf was dystrophic and would gradually decrease [21-23] (Figure 2F). As a result, the smarca4 eyes were noticeably different from the WT eyes at 72 hpf (Figure 2C,D,G,H). Although the smarca4 RPE was more pigmented at this stage, the RPE cells were still abnormal with holes detected in the RPE layer (Figure 2G,H, red arrow).
To analyze the differentiation problem of the smarca4 RPE, we first conducted an in situ hybridization analysis at 52 hpf with dct and rpe65a, a pigment cell and an RPE-specific marker, respectively. Dct is a key enzyme in the melanogenesis pathway melanin synthesis , which gives the pigment cells the distinctive black color. The dct signal generally covered the eye surface of the WT (Figure 2I,K) and smarca4 embryos (Figure 2J,L), suggesting that the smarca4 RPE cells were still committed to the pigment cell lineage. However, the smarca4 staining was irregular, which indicates abnormal RPE differentiation. Indeed, this was supported by the staining of rpe65a, an RPE-specific enzyme that mediates a critical step of isomerization of all-trans-retinol to 11-cis-retinal in the visual cycle . In WT embryos, the staining was intense in the anterior region of the eye and highlighted the RPE cell shape clearly (Figure 2M,O), while in the smarca4 embryos, the staining was irregular (Figure 2N,P), and the extent of staining was far less comprehensive than that observed with dct expression (Figure 2J,O).
The differentiation defects of the smarca4 RPE were further analyzed with TEM at 52 hpf. The smarca4 RPE had fewer melanosomes compared with WT (Figure 3A), which is consistent with the pigmentation defects observed in the semithin sections (Figure 2F,H). In addition, there appeared to be fewer melanosomes on the basal side of the cell (red arrow). To quantify these changes, three independent transverse sections of WT and smarca4 eyes were collected and several RPE attributes measured. First, the number of melanosomes per RPE cell area was analyzed (Figure 3B and Table 2). The RPE on the dorsal and ventral sides of the optic nerve (ON) was analyzed separately. A two-way ANOVA indicated that there were fewer melanosomes/RPE cell area in smarca4 (p value=1.43e-4) and in the RPE ventral to the ON in both genotypes (p value=4.91e-4; see detailed descriptions of the ANOVA output in Appendix 1). In addition, the total RPE area in the smarca4 embryos was not smaller than that in the WT embryos (WT: 1176.36±58.24 μm−2; smarca4:1078.35±133.05 μm−2; Mann–Whitney test; p value=0.7). Together, these results suggest that there was a general decrease in the melanosome number at this stage. Second, the distribution of the melanosomes in the RPE was evaluated by calculating the apical/basal ratio, using the center of the flattened nuclei to demarcate the apical and basal sides of the RPE cells (Figure 3C and Table 2). A logistic regression analysis of the counts demonstrates that the melanosomes in the smarca4 RPE had an apical bias (p value=3.47e-12) and that there was a general apical bias in the ventral RPE compared with the dorsal RPE in both genotypes (p value=6.15e-07; Appendix 2).
Melanosomes mature through distinctive phases from the circular immature type to the ellipsoidal mature type . To assess the maturity of the melanosomes, their roundness (1: circular; 0: elongated; Figure 3D and Table 2) and area (Figure 3E and Table 2) were separately evaluated with two-way ANOVA. The analysis shows that the roundness of the melanosomes was not affected by genotype (p value=0.67; Appendix 3). There was also no difference in melanosome roundness between the dorsal and ventral RPE (p value=0.07). These observations suggest that the melanosome maturity, as assessed by shape, was similar between the smarca4 and WT genotypes. However, there was a small but statistically significant decrease in the melanosome area in smarca4 RPE (p value=1.96e-7), and in the ventral RPE for both genotypes (p value=2.37e-6; Appendix 4). In addition, the decrease in melanosome size in smarca4 RPE was less drastic in the ventral side (p value=0.0066). Together, these histological data suggest that smarca4 RPE cells were formed, but the differentiation was abnormal at 52 hpf. This observation is consistent with the overall impairment of melanogenesis observed in the smarca4 mutants.
The differentiation defects in the smarca4 RPE suggest that the absence of smarca4 disrupted the regulation and expression of genes that were key to normal RPE differentiation. Furthermore, terminal differentiation was also impaired in the smarca4 retina, resulting in abnormal retinal lamination (Figure 2F) [21-23]. Since the RPE and the retina interact closely during development, we were interested in elucidating whether the retinal phenotype might be secondary to disruption of normal signals between the developing RPE and retina. To this end, we extended a factorial analysis framework  that was previously used to study smarca4 retinal differentiation. An overview is described (also see the Methods section for the details of the statistical framework).
First, the RPE expression values were estimated by comparing the retinas and the RPE-attached retinas . Then, a 3×2 factorial design (Figure 1A) was used to analyze the expression values of three kinds of tissue (T: whole embryo, retina, and RPE) in two mutation backgrounds (M: WT and smarca4). The expression value of a gene in the individual conditions (yg) was deconstructed into coefficients that represented the contribution to the expression level by the presence of that factor (Figure 1A, equations under the condition name). Using these coefficients, contrasts were built to identify candidate genes that fulfilled specific criteria. The false discovery rate (FDR) q value cutoff for a contrast was 0.001. Several contrasts were ultimately combined to select genes that were regulated by smarca4 in the retina and the RPE (Figure 1B) and will be discussed below. In addition, a fold-change between smarca4 and WT was calculated to aid ranking of the resulting gene list.
The contrasts selected genes that were differentially expressed in the WT or smarca4 RPE and at the same time were regulated by smarca4, either directly or indirectly (Figure 1B, black circle; Table 3). A total of 591 genes were selected (Appendix 5). Among them, 432 and 162 genes were under- and overexpressed in the smarca4 RPE when compared with the WT RPE, respectively. Seven genes had negative RPE expression value in WT and smarca4 and were not biologically meaningful. DAVID analysis of these genes revealed two functional annotation groups that were highly enriched (Appendix 6; Benjamini-adjusted p value=2.7e-2). These groups include two gene ontology (GO) terms for cellular components: non-membrane-bounded organelle (32 genes/6.1% of total annotated genes) and intracellular non-membrane-bounded organelle (32 genes/6.1% of total annotated genes). These terms refer to organized structures that are not bound by a lipid bilayer membrane including the cytoskeleton, suggesting that the cytoskeletal dynamics of the smarca4 RPE were impaired. Further selection of these RPE candidate genes is described in the third subsection.
The contrasts selected genes that were differentially expressed in the WT or smarca4 retina and at the same time regulated either directly or indirectly by smarca4 (Figure 1B, dark blue circle; Table 4). A total of 412 genes were selected (Appendix 7). Among them, 252 and 160 genes were under- and overexpressed in the smarca4 RPE when compared with the WT RPE, respectively. Many gene candidates were previously identified from a microarray analysis of the smarca4 retinas  and were validated in subsequent follow-up experiments [24,25,27]. Furthermore, we identified the smarca4 gene itself as under-expressed in the smarca4 retina, which validated the array experiment and our statistical selection. This is because the null-mutation was expected to result in nonsense-mediated decay of the mRNA and reduce its gene expression. DAVID analysis of these genes revealed several similar functional annotation groups that were enriched (Appendix 8; Benjamini-adjusted p value <0.05; see the table for the gene count and percentage of total annotated genes). For example, eight terms are related to cytoskeletal dynamics, nine terms are related to the cell cycle, and four terms are related to neuron/eye morphogenesis.
Comparing the final list in groups (i) and (ii) showed that 51 genes were commonly expressed in both groups (Figure 1B, the white region between the black and blue circles), while 540 and 361 genes were uniquely expressed in (i) and (ii), respectively. Thus, a total of 952 genes were identified. Among them, 293 (30.78%) had no annotation.
We theorized that paracrine signaling genes would likely be differentially expressed in one of the two interacting cell-types (i.e., the retina and the RPE), either as secreted ligands or the intracellular signal transducers. To identify these candidate genes in smarca4 RPE, additional selection criteria were imposed on the group (i; Table 5). In particular, these genes should not be differentially expressed in WT or smarca4 retinas. In other words, we selected for genes with RPE-specific differential expression. Specific contrasts were first built to identify retinal-specific genes in WT and smarca4 retinas with a relatively inclusive cutoff (q value <0.05; Figure 1B, light blue circle). Then, these genes were removed from the candidate gene group (i) to select for the genes that were regulated by smarca4 in the RPE but were not differentially expressed in the retina.
A total of 39 genes were selected based on these criteria (Appendix 9). Among them, 31 were under-expressed in the smarca4 RPE, while the remaining eight were overexpressed. Based on their annotations and previous investigations in humans [41-44], mice [44-46], chicks , and zebrafish , we found that many of the known genes are expressed in the RPE and potentially play an important role in RPE signaling and physiology (Appendix 9; supporting evidence columns). For example, previous studies had detected expression of 31 out of 39 (79.5%) genes in the RPE. Among these data, the zebrafish data represent carefully selected RPE genes that were consistently expressed in our previous investigation  (Appendix 9). Even if the zebrafish data are excluded from this calculation, studies in the other systems still support the expression of 22 out of 39 (56.4%) genes in RPE. Therefore, these selected Smarca4-regulated RPE genes represent a group of highly validated RPE-specific genes. DAVID analysis of these genes also revealed two enriched functional annotation groups with more than two genes in each group (Appendix 10; p value<0.05). These functional annotation groups include the focal adhesion pathway (four genes; 12.5% of total annotated genes) and the melanogenesis pathway (three genes; 9.375% of total annotated genes).
This study established an approach for analyzing differential gene expression between the RPE and retina in zebrafish, as well as identifying a group of candidate genes that may underlie the developmental defects observed in the smarca4 RPE and the retina. Although the analytical emphasis was on the RPE, the same approach could be used to analyze the retina and its secreted signals. This approach focuses on tissue-level expression, and thus may complement cell-level approaches, including expression profiling of FACS-purified cells from the RPE-specific reporter lines. An alternative approach to RPE cell purification for expression study is by density gradient centrifugation followed by FACS using pigment granule density as a sorting criterion . This alternative approach may alleviate the need to use RPE-specific reporter lines for RPE expression profiling, which may not be readily applicable to all experimental conditions. However, FACS requires enzymatic dissociation of individual cells, which may affect cell physiology, as well as gene expression. Furthermore, the preparation time from embryos to sorted cells can take several hours. As a result, using FACS for RPE expression analysis may not reveal the desired expression profile at a specific stage. We believe that our approach, which measures at a precise developmental stage, can complement the FACS approach and that together the approaches will facilitate RPE developmental analysis.
Further literature searches revealed potential candidates from the 39 genes in group (iii) that could affect RPE/melanosome differentiation, cytoskeletal dynamics, and cell-cell signaling (Table 6). These observations are also supported by the functional annotation enrichment revealed by the DAVID analysis (Appendix 10). These genes belong to the under-expression category (i.e., smarca4/WT RPE expression fold change <1). The first category contains 12 genes (si:ch211–87l2.1, fancl, magi2, map2k1, myl12.2, nav2b, pacrg, pdcl3, ppp1r12a, sec23ip, smarca2, and tubb5) that may provide insight into the abnormal RPE development in the smarca4 mutant, particularly regarding two interdependent processes: cytoskeletal dynamics and melanogenesis . For example, si:ch211–87l2.1 is a novel gene that is predicted to be a Rab GTPase activator by sequence homology (ZDB-GENE-030131–4497; this is the ID used in the zebrafish community website for the gene). Rab GTPases are small GTPases that regulate membrane trafficking. Functional impairment in specific Rab proteins including Rab32/38 and Rab27A causes pigmentation disorders . Rab27A forms a complex with myosin VIIa and another interacting protein MyRIP, and this complex mediates local trafficking of retinal melanosomes to actin cytoskeleton . Mutations in myosin VIIa in patients with Usher syndrome cause abnormal melanosome distribution in the RPE , a phenotype that mimics that seen in the smarca4 RPE. Together, these observations suggest that the abnormal expression of si:ch211–87l2.1 might partially underlie the abnormal melanogenesis defects observed in the smarca4 RPE. Another gene, fancl, is an enzyme implicated in Fanconi anemia . One of the clinical symptoms of Fanconi anemia is pigmentation abnormalities , which suggests that fancl may play also a role in melanogenesis. Map2k1 is related to melanosome transport (Entrez Gene ID:5604) and pigmentation disorders . Myl12.2 was identified in a proteomic study of human RPE blebs, abnormal cell membrane structures that may contribute to drusen formation in age-related macular degeneration . Nav2b is involved in actin dynamics  and plays a role in cell migration and the outgrowth of cellular processes, including axons . Pacrg forms a molecular chaperone complex called chaperonin containing TCP1 complex (CCT) . One of its components, CCT4, has been shown to bind to melanosome in a proteomic characterizations . CCT is also involved in the biogenesis of many cytoskeletal proteins, including actins and tubulins. Another selected gene in the category, Pdcl3/ PhLP2A, physically interacts with CCT and modulates its folding activity . Thus, the defects in melanogenesis and cytoskeletal dynamics in the smarca4 RPE might be secondary to dysregulated CCT activity. Furthermore, ppp1r12a (also known as MYPT1 (Entrez Gene ID: 4659)) is a myosin phosphatase. Smarca2, another family member of smarca4, interacts with the intermediate filament ( ENSG00000080503). Magi2 has been shown to be a component of tight junction  and is regulated by the planar cell polarity pathway in glomerular podocytes . Together, the under-expression of these genes in smarca4 RPE might contribute to defects in cytoskeleton and cell-cell adhesion. Consistent with a defect in cell-cell adhesion, we noticed in our experiments that the smarca4 RPE cells did not adhere to each other well during retinal dissection.
Our statistical design also allowed for the detection of RPE-secreted signals, including adam9, gdnf, and bmp8a, which may play a role in retinal development and degeneration (Table 6). Adam9 is a protein secreted by the RPE located within the inter-photoreceptor matrix . Adam9 is believed to mediate photoreceptor outer segment (POS) attachment to the RPE. ADAM9 mutations in humans, mice, and dogs cause cone-rod dystrophy [44,62]. In these cases, the apical processes of the RPE are disorganized, and the adhesion between RPE and POS is compromised. We also noticed that the smarca4 RPE did not tightly adhere to the retina. Gdnf is secreted by cultured RPE cells, and its presence in the culture medium enhances the survival of dopaminergic neurons . Gdnf also regulates proper photoreceptor development in chickens [14,16]. Bmp8a, a member of the Bmp signal transduction pathway, protects mouse osteoblasts from glucocorticoid-induced apoptosis ; however, the role of Bmp8a in RPE and retinal development has not been fully characterized. The attenuation of secreted signals in the smarca4 RPE might affect the surrounding tissues, including the retina and the RPE. Furthermore, this observation suggests that other tissues, including paraocular tissues, may play a signaling role in the developmental defects of the smarca4 retina and RPE, as the smarca4 mutation affects multiple organ systems.
The third gene category among these 39 selected genes in Table 6 includes intracellular signal transducers that were specifically affected in the RPE and not differentially expressed in the retina. A total of seven genes are in this category, including ghdc, guk1a, gsk3b, map2k1, ppp1r12a, spdya, and tcf7l1a. These transducers are involved in different cellular processes, and two, gsk3b and tcf7l1a, are related to Wnt signal transduction. Gsk3β modulates cytoskeletal dynamics  and cell-cell adhesion  in the RPE, and in turn helps to establish the epithelial phenotype. Tcf7l1a/Tcf3 is a transcriptional repressor and a target of the Wnt signaling pathway . Under the influence of high levels of Wnt signaling, Tcf7l1a degrades and relieves the transcriptional repression on downstream genes . Therefore, under-expression of these genes in smarca4 RPE implicates over-activation of the Wnt signaling pathway and overexpression of target genes that would normally be repressed. These events might contribute to the abnormal RPE phenotype.
The phyh gene was put in a separate gene category in Table 6 because the gene’s function as a peroxisomal enzyme does not seem to be directly related to the other genes in our study. Nonetheless,phyh mutation in humans causes Refsum disease, and one of the clinical symptoms of this disease is RP . Thus, a decrease inphyh in the smarca4 RPE might also play a role in the smarca4 phenotype.
The functional annotation analysis of the group (ii) Smarca4-regulated retinal genes also revealed enrichment of annotations that are consistent with the smarca4 retinal phenotype. For example, Link and colleagues showed that there was a cell-cycle withdrawal delay in the smarca4 retinas . Our analysis identified nine cell-cycle terms enriched in the gene set (Appendix 8). The deregulation of the genes related to these terms in smarca4 retinas might cause the cell-cycle withdrawal defect. In addition, four and eight terms are related to neuron/retinal differentiation and cytoskeletal dynamics, respectively. This is also consistent with the terminal differentiation defects of the smarca4 retinas [21,22].
To validate the expression results from this study, quantitative reverse-transcription PCR (qRT-PCR) or northern blot validation should be performed with purified RPE cells rather than estimating the RPE values from dissected eye tissues. This requires generation of RPE reporter lines in WT and smarca4 for FACS isolation of RPE cells. In Higdon and colleagues’ FACS approach, pigment cells were isolated from the whole embryos using pigment granule density . Even though this approach detected an RPE expression signature, it probably requires further optimization before it is applicable to the purification RPE cells from dissected eyes for validation experiments. Laser capture microdissection can be used to obtain some RPE cells but is not feasible for obtaining the whole RPE layer for comparison. Thus, this approach is also not appropriate for validating the findings in this study.
In summary, identifying the smarca4-regulated RPE genes with our analysis has highlighted novel and intriguing relationships between RPE cytoskeletal dynamics, membrane trafficking, and intra- and inter-cellular signaling. This new knowledge may ultimately facilitate our understanding of the pathogenesis of related retinal degenerative diseases and development of new therapies.
Appendix 1. ANOVA output for melanosome number per RPE cell area analysis in Figure 3B.
Appendix 2. Logistic regression output for melanosomes apical-basal distribution analysis in Figure 3C.
Appendix 3. ANOVA output for melanosome roundness analysis in Figure 3D.
Appendix 4. ANOVA output for melanosome area analysis in Figure 3E.
Appendix 5. Smarca4-regulated RPE genes.
Appendix 6. Functional annotation enrichment of Smarca4-regulated RPE genes.
Appendix 7. Smarca4-regulated retinal genes.
Appendix 8. Functional annotation enrichment of Smarca4-regulated retinal genes.
Appendix 9. Smarca4-regulated RPE genes that are not differentially expressed in the retina.
Appendix 10. Functional annotation enrichment of Smarca4-regulated RPE genes that are not differentially expressed in the retina.
We thank Phillip San Miguel and Ann Feil at the Purdue Genomic Core Facility for their excellent assistance on the Affymetrix experiments. We thank Skye Brown and Pin-Chao Liao for their technical assistance. We also thank Brian Link, Woody Walls and members of the Leung laboratory for helpful discussions. This study was partially supported by a Pediatric Ophthalmology Research Grant from the Knights Templar Eye Foundation and a Charles D. Kelman, M.D. Scholar award from the International Retinal Research Foundation to LZ; NSF grants NSF DMS-1222718 to PM and NSF DMS-1120256 to WZ.