Transcriptome of the human retina, retinal pigmented epithelium and choroid

Abstract

The retina and its adjacent supporting tissues -- retinal pigmented epithelium (RPE) and choroid -- are critical structures in human eyes required for normal visual perception. Abnormal changes in these layers have been implicated in diseases such as age-related macular degeneration and glaucoma. With the advent of high-throughput methods, such as serial analysis of gene expression, cDNA microarray, and RNA sequencing, there is unprecedented opportunity to facilitate our understanding of the normal retina, RPE, and choroid. This information can be used to identify dysfunction in age-related macular degeneration and glaucoma. In this review, we describe the current status in our understanding of these transcriptomes through the use of high throughput techniques.

1. Introduction

In the past decade, multiple common genetic variants have been identified for prevalent diseases through genome-wide association studies (GWAS). Translating the results of GWAS to new treatments for disease still faces several challenges including: 1) Single nucleotide polymorphisms (SNPs) significant in GWAS are not causative and may be in linkage disequilibrium with causative SNPs that are distantly located; 2) associated genes harboring these SNPs have not been evaluated in tissues affected by disease; and 3) gene expression in cells central to the disease have not been evaluated for regulatory effects of these SNPs. While accumulating scientific evidence suggests that regulatory changes contribute to the phenotypes interrogated by GWAS, progress in defining these regulatory changes has been slow. This problem is apparent in the eye, a highly specialized structure composed of multiple layers and cells derived from mesoderm and neuroectoderm that has produced multiple GWAS hits outside coding regions for age-related macular degeneration (AMD), glaucoma, and refractive error, three widely prevalent eye conditions. Here, we focus on AMD and glaucoma, the most prevalent causes of untreatable vision loss in the elderly.

The transcriptome is a complex mixture of multiple isoforms for known genes and non-coding RNAs that will require multiple approaches for characterization. These multiple transcript types are heritable, based on evidence that gene expression is heritable between monozygotic twins [1], Centre d'Etude du Polymorphisme Humain families [2, 3], and Icelandic families [4]. The transcriptome is also cell/tissue specific, requiring cell specific characterization. For example, there is lack of significant sharing of cis-expression quantitative trait loci (eQTLs) between adipose tissue and blood [4] or between brain cortical tissue and blood mononuclear cells [5], making it necessary to characterize the transcriptome of each cell type.

Vision scientists have long recognized the importance of characterizing the transcriptome through their use of microarrays to evaluate expression. This technique persists as a mainstay among vision scientists even though microarrays have been shown to have biases in hybridization strength, as well as the potential for cross hybridization to probes with similar sequences [6]. Herein, we provide a brief overview of the eye structure with an emphasis on its posterior layers, and follow with a review of SAGE, cDNA microarray, and RNA sequencing (RNA-Seq) studies of the protein-coding transcriptomes of human retina, retinal pigmented epithelium (RPE) and choroid.

2. Structure of the human eye: retina and choroid

The retina is a specialized neural tissue lining the back of the eye that is responsible for vision. The retina originates as an outgrowth of the brain during ontogenesis and is thus part of the central nervous system. There are 21 chorioretinal layers and clinically important spaces, many of which can be visualized non-invasively at high-resolution (Figure 1) [7]. The macula has the highest overall density of neurons and is responsible for acute central vision (diameter = 6 mm = 21° of visual angle). The nasal retina (close to the nose, seeing the temporal visual field) has 1.4-3-fold more neurons, depending on the cell type, than temporal retina (close to the temple, seeing the nasal visual field.) The cellular composition of the neurosensory retina is highly organized with a variety of neurons and two glial types. Major neuron types have been historically classified based on morphology, laminar distribution, topography, brain connections, physiological responses to light stimuli, transmitter pharmacology, and molecular signatures, such as immunoreactivity to specific antibodies. These cells likely differ by their RNA transcriptional programs as well. A recent classification of retinal cell types indicates three cone photoreceptors, one rod photoreceptor, two horizontal cells, 13 bipolar cells, ≥29 amacrine cells, and ≥20 ganglion cells. Only the ganglion cells project to the brain (via the optic nerve). Ganglion cells are preferentially affected in glaucoma, a prevalent cause of vision loss for which high intraocular pressure is an important risk factor. Glial cells important for retinal function are Müller cells that span all the neuronal layers and astrocytes among ganglion cell axons en route to the optic nerve head. The RPE and choroidal vasculature constitute the photoreceptor support system, which is affected in AMD, a major cause of vision loss in the elderly of European descent worldwide. Polarized RPE has demanding dual roles serving photoreceptors apically and choroid basolaterally. Distinctive extracellular lesions that differentially confer risk for AMD progression distributes on both aspects of this key cell layer [8–10]. The retina has two vascular beds with different properties and propensity for disease. The retinal circulation serves the inner retina and is within the blood-retina barrier. The choroid serves the photoreceptors and RPE, and it is part of the systemic circulation. The RPE maintains the outer limit of the blood retina barrier with junctional complexes. The choroid is distinguished by the highest blood flow in the body, especially under the macula, and it thins markedly with aging [11]. Choroidal cells include vascular and lymphatic endothelia, smooth muscle cells, fibroblasts, melanocytes, mast cells, autonomic neuronal ganglia, and resident and transient cells of monocyte lineage.

Figure 1. Chorioretinal tissue layers in human eye.

A. Cross-section of a human eye. Green bracket shows extent of the posterior pole that is illustrated in panel B, including optic nerve head and macula with fovea. Light shines into this eye from the right. Schematic from David Fisher Designs.

B. In vivo high-resolution imaging of a normal human retina (black bar) and choroid (white bar) shows tissue layers of the 6 mm diameter macula and adjacent optic nerve clearly. Light shines into this eye from above. The neurosensory retina contains multiple bands of alternating high and low reflectivity that coincide in part with the anatomical layers. Blue bar delimits layers occupied by photoreceptors. Arrowheads indicate a dip at the fovea, shown by histology in panel C. The choroidal vasculature contains lumens of large vessels. It is bounded externally by the sclera, which has a more homogeneous reflectivity. Swept-source optical coherence tomography scan, a 58 yr old male, courtesy of R.F. Spaide, MD.

C. Ex vivo high-resolution histology of the fovea, which is located in the center of macula and is responsible for high acuity vision. The neurosensory retina including thickened photoreceptor layers and choroid are indicated by black, blue, and white arrows, respectively. A foveal pit is created by inner retinal neurons, Müller cells, and accompanying retinal vasculature being swept to the side of the visual axis. This facilitates light capture by tightly packed inner segments of cone photoreceptors on the opposite aspect of the retina. The photoreceptors and Müller cells (glial cells) have long fibers that together make a thick layer that is 14% of total retinal thickness in this region (Henle fiber layer). The RPE is a simple cuboidal epithelium that sits on Bruch’s membrane. Bruch’s membrane is the inner wall of the choroid and functions as a vessel wall between choriocapillaries and RPE, as well as substrate for RPE attachment. RPE maintains health of the photoreceptors and the choroid. RPE is visible clinically by its melanosomes and lipofuscin, which is autofluorescent due to long-lasting vitamin A byproducts. The choroidal vasculature contains fenestrated capillaries adjacent to the RPE and larger arteries and veins near the sclera. A 63-year-old female recovered <3 hour after death, osmium post-fixation, epoxy embedding, 1 µm thick section, and toluidine blue stain. Figure was prepared by J.D. Messinger, DC.

3. Transcriptome analysis of the retina and RPE/choroid

3.1 High-throughput technologies: cDNA microarray, SAGE and RNA-Seq

A cDNA microarray consists of immobilized probes complementary to known transcripts on a solid substrate [12]. Isolated RNA is labeled with fluorescent dyes and hybridized to the cDNA microarrays, washed, and scanned with a laser scanner. The amount of fluorescent dye intensity is a measure of gene expression. In early versions of cDNA microarrays, biases and artifacts produced inconsistent results among the same samples. In 2006, quality control standards were developed by the MicroArray Quality Control (MAQC) to address these issues [13]. Still, microarrays are unable to identify RNA editing events or novel isoforms, and they cannot accurately measure absolute expression levels due to hybridization and background variation [14].

Serial analysis of gene expression (SAGE) can perform a quantitative analysis of transcripts without requiring prior knowledge of the transcript sequence. Compared to microarrays, SAGE represents an unbiased, comprehensive representation of the transcriptome [15, 16]. A short tag i.e., 9–14 bases of a gene transcript is linked with 20– 50 other such tags in a cloned DNA fragment. Sequencing of such a clone provides the sequence of 20–50 tags, and a set of a few thousand such clones represents a library of SAGE tags [17]. Most individual SAGE tags can be assigned to specific genes by alignment. It can identify low-abundance transcripts and detect relatively small differences in their expression. However, it is limited to simultaneously analyzing a small number of expression profiles.

RNA-Seq can measure absolute gene expression levels and identify the exact sequence of transcripts belonging to mRNA and non-coding RNA [6]. RNA-Seq can provide a measure of gene expression as well as the underlying genetic variants that influence gene expression. This additional information provided by RNA-Seq compared to microarrays and SAGE have made RNA-Seq the current method of choice for studying the transcriptome.

The major advantages of RNA-Seq include:

• Identification of known and previously unknown transcripts as opposed to array-based methodologies that depend on annotation of known transcripts.

• Accurate quantification of transcript levels by counting reads that map to transcripts as opposed to inferring levels from hybridization intensity signals as in microarrays. Background noise levels are essentially zero in RNA-Seq, making it easier to assay rare transcripts provided sequencing depth is sufficient.

• The counting methodology used to ascertain expression levels in RNA-Seq provide a more accurate assessment of expression levels throughout the dynamic range than can be obtained through array-based methods.

• Integration of RNA-Seq and whole-genome DNA sequencing data will allow determination of RNA-DNA differences and allele-specific expression within individual samples.

However, the cost of RNA-Seq, compared with microarray, is significantly higher, especially when deep depth of sequencing coverage is desired. Another challenge to the downstream data analysis of RNA-Seq is the unevenness of the depth of coverage across the expressed regions in the genome [14].

3.2 RNA qualities and methods of case ascertainment

For studies utilizing human eye tissues obtained from eye banks, the interval between donor death and RNA extraction for subsequent analysis should be considered when assessing the quality of published data. This interval can vary widely between studies. RNA deteriorates by deadenylation of the 3’ poly-A tail, decapping of the 5’ end, and degradation by 5’ and 3’ exonucleases, with a major role being played by the 5’ exonuclease. An analysis of RNA stability through a time course simulation of tissue recovery using eye bank conditions was performed by Malik et al. using pig eyes [18], which are of similar size to human eyes. At different time intervals after animal sacrifice, the authors isolated retina and RPE from the eyes and assessed the recovered RNA. They determined that RNA for housekeeping and RPE-specific genes deteriorated starting at 5 hr. Immersion into a cold RNA stabilization medium (RNA Later) delayed degradation onset at least 24 hr. Retina, with many more cells than RPE (>100M vs. 4–6M), had more stable RNA than RPE. However, RNA specific to potentially vulnerable minority cell populations was not investigated in this study. The possible reason for more rapid degradation of RNA from the RPE, a cell central to AMD, could be due to the high metabolism of native RPE and mechanical damage during dissection of this single cell layer [18]. In our studies, we examined both RNA integrity number at the time of processing after overnight shipping in cold stabilization medium and assessed expression of genes specific for numerically sparse cell types with good success [19].

In addition to RNA quality, studies utilizing diseased donor eyes should be checked for methods of case ascertainment, typically done with family report, clinical records, postmortem inspection and/or histology of the fundus, or a combination. Clinical records of eye donors can vary considerably in source and in length of time before donor death. Pathology grading systems exist for donor eyes based on the color fundus photography systems used for large epidemiologic and clinical studies [7, 20–22]. Through recent clinical experience with optical coherence tomography (OCT, Figure 1B), a high-resolution imaging technology that provides detailed cross-sectional views of retina, it is now known that important AMD pathology is not captured well by color photography. Even less detail is apparent with color photography of partly opacified post-mortem retina. Identifying and quantifying these newly discovered AMD-relevant endophenotypes (e.g., subretinal drusenoid deposits [9] and outer retinal tubulation [23]) will be important for quantitative assessment of gene expression in tissues. Ex vivo OCT of short post-mortem eyes, if validated by histology, holds great promise for the molecular pathology laboratory [7, 24, 25]. For glaucoma, clinical records are the preferred method of case ascertainment, as optic nerve damage may not be visible by macroscopic inspection until advanced stages (C.A. Girkin, MD, MSPH, personal communication, 5/9/14). In Table 1 and Table 2, we summarize the studies of retina and RPE/choroid transcriptome reviewed in this manuscript, including sample size and postmortem delay to preservation.

Table 1.

Studies of retinal transcriptome

Author	Samples	Platform	Results
Yoshida et al. [27](2002)	Five human retinas 2 young (ages 13–14) 3 elderly (ages 62–72) 24+ h to processing	Microarray	17 (1.4%) genes were expressed at higher levels in retinas from young, whereas 7 (0.6%) genes were designated as elderly- dominant. Northern blot analysis and qRT- PCR results confirmed the changes in expression in 8 of 10 genes examined
Hornan et al. [31](2007)	Human neural retinas (ages 40–77) 12 pooled fovea macula 2 pooled macula 10 peripheral Time to processing not given	Microarray	qRT-PCR results of eight genes enriched in fovea macula and four genes enriched in peripheral retina.
Cai et al. [28](2012)	Twelve human retinas 6 young (ages 18–43) ○ 6 macula vs. 6 peripheral 6 elderly (ages 72–79) ○ 6 macula vs. 6 peripheral 48 h to processing	Microarray	The transcriptome of the human retina is affected by age and topographic location.
Wagner et al. [35](2013)	10 tissues from 6 human eyes stored in RNAlater, time to processing not given	Microarray	Ocular Tissue Database (OTDB) for normal eye gene expression is provided.
Young et al. [30](2013)	Fifteen human eyes 9 human fetal eyes (four male and five female) 6 adult normal eyes (three male and three female) <6 h to processing, stored in RNAlater, shipped overnight	Microarray	Significant gene expression fold changes (>1.5) were found in adult versus fetal retina
Sharon et al. [17](2002)	2 human retinas (ages 88 and 44) time to processing not given	SAGE	An average of 13,779 UniGene transcripts were detected in retina.
Farkas et al. [29] (2013)	3 normal human retinas (ages 42, 44, 46, one male, two female) time to processing not given	RNA-Seq	75% of known exons were detected. Novel exons increase the number of exons by 3%.
Li et al. [19] (2014)	8 normal human retinas <6 h to processing, stored in RNAlater, shipped overnight, processed upon arrival	RNA-Seq	80% of the transcriptome is expressed in retina. There is significant differential expression between retinal layers and locations of a tissue layer.
Whitmore et al. [32] (2014)	4 normal human retinas <6 h to processing, stored in liquid nitrogen nasal, temporal, and macular regions	RNA-Seq	No genes were differentially expressed between temporal and nasal retina.

Table 2.

Studies of RPE/choroid transcriptome

Author	Samples	Platform	Results
Bowes Rickman et al. [42] (2006)	17 human eyes ages 28–66 <6h to processing, stored in RNAlater	SAGE	Three differential expression lists were reported: macula retina enriched, periphery retina enriched, and RPE enriched.
Radeke et al. [39] (2007)	15 human eyes normal eyes (ages 45–65 and 75–87) ○ 9 males, 1 female AMD eyes (ages 78–92) ○ 3 males, 2 females from each eye, one macular and one peripheral punch of the RPE/choroid complex was taken 0.5 to 7.4 h to processing, stored in RNAlater, stored up to a week	Microarray	76 genes differentially expressed between macular and peripheral regions. 29 genes selected for validation via qPCR and 21 confirmed.
Van Soest et al. [40] (2007)	6 human eyes ages 17–36 2 females, 4 males from each eye, one macular and two peripheral fragments were isolated 24+ h to processing	Microarray	438 genes differentially expressed between macular and peripheral regions. Validation of 33 genes using RT- PCR resulted in an 84% correlation.
Booij et al. [36] (2009)	6 human eyes ages 63–78 macular RPE isolated 30+ h to processing	Microarray	Attempted to identify the expression profile of macular RPE and demonstrated the lack of knowledge regarding the RPE transcriptome. Functional analysis revealed enrichment for pathways such as oxidative phosphorylation and ATP synthesis.
Booij et al. [37](2010)	5 human eyes ages 63–78 isolated RPE, choroid, and photoreceptor layers 16–22 h then frozen	Microarray	Identified 114 RPE specific genes. Selected 39 for validation. 85% validated through literature and s-QPCR confirmed RPE expression for remaining genes.
Strunnikova et al. [38](2010)	4 adult Caucasian (ages 64–89) 4 fetal native RPE samples 4 fetal native choroid samples <24 h to processing fetal RPE primary cultures ARPE-19 cell line	Microarray	154 RPE signature genes identified. Using qRT-PCR, 48 genes were highly expressed both in vivo and in vitro. Demonstrated that culturing cells can change the expression, but did not affect enrichment of signature genes.
Li et al. [19] (2014)	8 normal human RPE/choroid/sclera <6 h to processing, stored in RNAlater, shipped overnight, processed upon arrival	RNA-Seq	926 genes were differentially expressed between macular and peripheral RPE/choroid/sclera
Whitmore et al. [32] (2014)	4 normal human RPE/choroids <6 h to processing, stored in liquid nitrogen nasal, temporal, and macular regions	RNA-Seq	81 genes showed increased expression in the nasal RPE/choroid and 39 genes showed decreased expression

Finally, studies should be scrutinized for their dissection techniques, and particularly how they name samples of different regions. Tissue punches covering differing extents of retina can yield variable results, especially in the macula, which contains the singularity of densely packed cone photoreceptors. The diameter of punches and their point of centration should be specified with dimensions (e.g., a punch 8 mm in diameter centered at 13 mm temporal to the fovea), as well as names (“periphery”), because names are variably defined. “Foveal” punches intended to capture only or primarily cone photoreceptors should be ≤1 mm in diameter. An annulus remaining after removal of the fovea (as just defined) from the 6-mm diameter macula is most accurately described as perifovea. The perifovea has a 9:1 rod:cone ratio in a young adult and 6:1 in an older adult. Outside the perifovea, the rod:cone ratio ranges from 30:1 just outside the macula to 10:1 at the retinal edge. Photoreceptors are the majority cell population and dictate the organization and cellular content of both the RPE/choroid and inner retinal layers. Therefore, rod:cone ratios should be factored into experimental design.

4. Exploring the human retina transcriptome

4.1 Number of genes detected in retina

There are over 60 functionally different neuronal cell types present in the retina. It is estimated that a complete transcriptome of the mammalian retina might consist of approximately 25,000 transcripts [26]. Early attempts to quantitate the retina transcriptome were done with focused arrays. Yoshida et al. used an array containing 2400 probes for human neuronal genes to assay RNA from two young and three elderly postmortem retinas [27]. Fifty-two percent of the 2400 gene probes reportedly hybridized strongly with the retinal cDNA. Cai et al. assessed differential gene expression of young maculas and old maculas, and young and old peripheral retinas from 12 healthy donor eyes on an Affymetrix Human Genome U133 plus 2.0 array, containing 54,600 gene probes [28]. On average, 49% of the probes were detected in human retinas. There was no significant difference in the number of genes detected between human young macula (26686 ± 319), old macula (26956 ± 275), young peripheral retina (27122 ± 108), or old peripheral retina (26533 ± 490). Sharon et al. used the SAGE technique to assess the genes expressed in the human peripheral and macular retina from two individuals [17]. They detected an average of 13,779 UniGene transcripts from two peripheral retinas and one macular retina.

Recently, RNA-Seq has revealed that the retinal transcriptome is more complex than previously reported. Farkas et al. reported RNA-Seq results from three normal human retinas [29]. They were able to detect 75% to 82% of all exons annotated in eight annotation reference tracks (UCSC, RefSeq, CCDS, Vega, Ensembl, Aceview, Gencode, and LincRNAs) at an average read depth of five or greater. Li et al. performed a more comprehensive analysis of eight normal human retinas [19]. They estimated the expression levels of 23,569 RefSeq protein-coding genes using the Fragments Per Kilobase of gene per Million mapped fragments (FPKM) metric. With coverage depth ranging from 66 to 133 million paired-end reads per sample, they detected the expression (i.e., FPKM > 0) of the majority of known protein-coding genes. The average number of expressed RefSeq genes was 18,318 (78%) in macular retina (MR), and 18,850 (80%) in peripheral retina (PR). This is a significant increase over the SAGE results reported by Sharon et al. [17]. The reported RNA-Seq results reflect the complexity of the retinal transcriptome that was not appreciated with cDNA microarrays and SAGE.

4.2 Effect of aging on gene expression

Yoshida et al. identified 17 (1.4%) genes expressed at higher levels in young retinas and 7 (0.6%) genes expressed at higher levels in older retinas [27]. Northern blot analysis and qRT-PCR confirmed the changes in expression in 8 of 10 genes examined and included increases in IFN-responsive transcription factor subunit (ISGF3G), creatine kinase B (CKB), and pancreatic amylase (AMY2A), and a decrease in TGF-beta receptor interacting protein 1 (TRIP1), LPS-induced TNF-alpha factor (PIG7), alpha-1 (E)-catenin (CTNNA1), ubiquitin hydrolase (USP9X), GABA receptor beta-3 subunit (GABRB3), and alpha-1 Type VII collagen (COL7A1).

Cai et al. also reported changes in the retina transcriptome with age [28]. Macular and peripheral retina RNA from six young and six older human donor eyes were analyzed by microarray. They found 85 genes expressed at higher levels in young macula and 55 genes expressed at higher levels in older maculas. Fifty-two genes were more highly expressed in young peripheral retina and 34 genes more highly expressed in older peripheral retina. Although there was evidence of differential expression in this study, none of the differences were statistically significant (F-test, p >0.05).

The studies of Cai et al. [28] and Yoshida et al. [27] have little overlap for differentially expressed genes. Even for genes identified in both, the directions of enrichment might be different; for example, ENO3 was differentially expressed in both studies, but Yoshida et al. [27] reported enrichment in young retina while Cai et al. [28] found the gene enriched in the older macula.

Another study by Young et al. compared gene expression patterns in normal human fetal (N=9) and adult retinas (N=6) using Illumina HumanHT-12 v4 Expression BeadChips [30]. A significant number of differentially expressed genes (N=1185, fold change ≥1.5) were found between adult and fetal retina. These differentially expressed genes were involved in development, cell death/growth, cancer functions, and signaling pathways.

4.3 Effect of retina anatomic location on gene expression

Although Cai et al. found no significant difference in expression between regions of eyes of different age groups, 105 out of 26,700 gene probes were differentially expressed between macula and periphery if all age groups are combined [28]. They found 24 gene probes expressed at higher levels in the peripheral than the macular retina. A second study by Hornan et al. compared the expression profile of macular vs. peripheral retina using 2–4 mm retinal punches from ten peripheral and two macular retinas and found 8 genes enriched in macula and 4 genes enriched in peripheral retina [31]. The eight fovea macula enriched genes are: OPN1LW, HDAC9, NPIP, SRGAP3, SRGAP2, FLJ, CRP, and BHM2. Cai et al. also reported two genes, HDAC9 and SRGAP2, to be enriched in the macula [28]. Hornan et al. reported 4 peripheral retina enriched genes - RP2, YAP, RHO, and SAG [31]. One gene, RHO, was also enriched in peripheral retina by Cai et al. [28].

In a recent transcriptomic analysis of human neural retina and RPE/choroid by RNA-Seq, Whitmore et al. reported that 1) no genes were differentially expressed between temporal and nasal retina; 2) 30 genes showed increased expression in nasal retina and 128 genes showed decreased expression when comparing nasal retina vs. macular retina; and 3) 37 genes showed increased expression in temporal retina and 323 genes showed decreased expression when comparing temporal retina vs. macular retina [32]. Li et al. showed that 1) 2051 genes were differentially expressed between macular and peripheral retina; and 2) 60–70% of them had higher expression levels in macula than in periphery [19].

4.4 RNA-Seq assesses alternative splicing

Farkas et al. performed RNA-Seq from three normal human retinas and detected 79,915 novel alternative splicing events, including 29,887 novel exons, 21,757 3’ and 5’ alternative splice sites, and 28,271 exon skipping events [29]. They also identified 116 potential novel genes in the human retina.

4.5 Comparison between RNA-Seq and Microarray

Previous studies have shown RNA-Seq to be superior at quantifying transcript levels through read counts in comparison to microarrays that infer levels from hybridization intensity signals [6, 33]. Background noise levels are essentially zero in RNA-Seq, which enables one to assay rare transcripts provided sequencing depth is sufficient [34]. The counting method of RNA-Seq also provides a more accurate assessment of expression levels throughout the dynamic range compared to array based methods. A comparison of microarray data from Wagner et al. [35] with RNA-Seq data by Li et al. [19] confirms that RNA-Seq is more sensitive for low abundance transcripts. Li et al. found 9864 genes with a FPKM value lower than 1 [19]; whereas, only 94 genes having a Plier number lower than 1 were found using microarray [35] (Figure 2). The difference in sensitivity between microarray and RNA-Seq for rare transcripts makes RNA-Seq the preferred method for transcriptome evaluation.

Figure 2. Distributions of gene expression values from microarray (Plier) and RNA-Seq (FPKM).

X axis shows the log2 value of the Plier number; Y axis shows a histogram of the number of genes.

5. Exploring the human RPE/choroid transcriptome

The RPE is a simple cuboidal epithelium apposed to the neurosensory retina (Figure 1C). It is best understood for its function in providing vitamin A derivatives for phototransduction and the daily phagocytosis of spent photoreceptor outer segment tips. However, the RPE also transfers from the systemic vasculature all the nutrients required by photoreceptors, acts as a waste-management system, maintains the choriocapillaris vascular endothelium and Bruch’s membrane, regulates the outer retinal component of the blood-retina barrier, and orchestrates the secretion of cytokines, among many other functions. The RPE is involved in many macular diseases especially AMD.

5.1 Number of genes detectable by microarray

Booij et al. used microarray to assess the expression profile of the RPE from six eyes of normal donors between the ages of 63 and 78 [36]. Using six 22k custom arrays on laser dissected macular RPE cells they found that the highly expressed genes in the RPE represented functional categories in oxidative phosphorylation, ribosome pathway, and ATP synthesis. A subsequent study, carried out by the same group, assessed the purity of the RPE samples through microarray analysis of RPE, adjacent photoreceptors, and choroid cell layers [37]. The justification for this design is based on the difficulty of isolating RPE cells free of contamination from the retina and/or choroid. Genes with 2.5× higher expression in the RPE relative to either the retina or choroid was limited to 114 RPE-specific genes. Those genes with high expression levels in two or more tissues were not analyzed further. Thirty-nine of the top highly expressed genes in the RPE were selected for validation and confirmed by a literature review and semi-quantitative PCR. Pathway analysis of the 114 genes revealed an overrepresentation of genes in categories such as membrane transport, vision, and glycoproteins. Three canonical pathways RAR-activation, retinol metabolism, and GABA receptor signaling were identified.

In 2010, Strunnikova et al. reported a microarray study of adult RPE, fetal RPE, and the RPE-derived cell line ARPE-19 [38]. A gene was designated as RPE specific when its RPE cell expression was 10-fold higher than its mean relative expression in 18 non-ocular cell lines. One hundred fifty-four genes fulfilled this criterion and were labeled as signature RPE genes. Subsequent validation via qRT-PCR of these 154 genes confirmed 100 genes were RPE signature genes. This standard chosen by these authors is good for capturing large effects. However, it is not complete, as low abundance genes known to be expressed in RPE by other methods were not included in the set.

5.2 Effect of anatomic location on gene expression

To uncover expression differences present in the macula that confer vulnerability to macular diseases, Radeke et al. assessed gene expression for the macula and extramacular regions of RPE/choroid from middle aged (45–65 year old) normal eyes, older (75–87 year old) normal eyes, and AMD afflicted eyes [39]. Using the Agilent Human 1A microarray platform for the AMD samples and the Agilent Whole Human Genome array for the middle-aged and older normal samples, differentially expressed genes were determined between macular and extramacular regions of the RPE/choroid complex. Of the 201 genes identified as differentially expressed in the RPE/choroid, 76 genes remained as differentially expressed after retinal contaminants were removed. Twenty-nine genes were chosen for validation by qRT-PCR, and 21 of the 29 genes were confirmed. The differentially expressed genes grouped into categories including inflammatory response or regulation, angiogenesis, and those associated with extracellular matrix (ECM).

A second study by van Soest et al. that sought to identify differential expression between locations was done on normal human macula RPE/choroid and peripheral RPE with a 22K custom oligonucleotide microarray [40]. The custom array contained RPE and neural retina transcripts identified through public databases and other experimental data from total RNA hybridizations of RPE/choroid, retina, and brain samples. They reported 438 genes differentially expressed between macular and peripheral RPE. Thirty-three were chosen for validation using RT-PCR and 84% were validated.

Whitmore et al. demonstrated that 1) in the nasal vs. macular RPE/choroid comparison, 81 genes showed increased expression in nasal RPE/choroid and 39 genes showed decreased expression; 2) in the temporal vs. macular RPE/choroid comparison, 70 genes were increased in temporal RPE/choroid and 44 genes were decreased; and 3) in the comparison of temporal vs. nasal RPE/choroid, 3 genes were increased in the nasal RPE/choroid and 11 were decreased [32].

It is important to relate gene expression to gene products where possible, because the former is not necessarily a predictor of the latter. Skeie and Mahajan performed proteomic analysis of RPE/choroid in three anatomic regions called fovea, macula, and periphery from three normal human eyes [41]. They defined the foveal region as 4 mm in diameter, thus including both cone-dominant and rod-dominant sub-regions of the macula (see above). In RPE/choroid, they identified 1) 4204 unique proteins in peripheral, 4595 in macular, and 4409 in foveal punches; 2) high expression of 66 proteins in the foveal RPE/choroid only and 251 in the foveal and macular RPE/choroid compared to the peripheral RPE/choroid. Gene ontology analysis showed that proteins with the highest expression in the periphery were enriched in the category of metabolites/energy. Proteins with the highest expression in the macula and fovea were enriched in the response to stimulus, reproduction, homeostasis, or immune system process categories, whereas these categories were absent in the periphery.

6. Transcriptome differences between human retina and RPE/choroid

The first attempt to catalogue the genes of retina and RPE used postmortem retina and RPE from a 44-year-old male and 88-year-old female [17]. Using SAGE analysis, the catalogue of expressed tags in both retina and RPE was completed without measurements of differential expression.

In a second study, Bowes Rickman et al. did a more extensive SAGE analysis of macula and peripheral retina genes from five patients and a second SAGE analysis of macula and peripheral RPE/choroid from another five patients [42]. This data was combined with that of Sharon et al. [17] to identify genes enriched in macular retina, peripheral retina, and RPE. They reported three differential expression lists; macula retina enriched, periphery retina enriched, and RPE enriched. The macular retina enriched genes included 270 tags that corresponded to 209 genes. The periphery retina enriched list included 200 tags corresponding to 173 genes and the RPE enriched list contained 143 tags or 123 genes.

Using a different approach, Booij et al. set out to find differentially expressed genes between the macula regions of the retina, RPE and choroid from three eyes aged from 63–78 years old [37]. The authors performed two microarrays to compare retina vs. RPE expression and RPE vs. choroid expression. The intersection of the RPE enriched genes from both assays was used as the RPE-Specific genes. Even though they combined RPE enriched genes in both microarrays to find RPE specific enriched genes, results from the separate microarray experiments were briefly analyzed. From the 33,712 features on the array, 1904 probes were found to have 2.5-fold higher expression in the macular RPE compared to the macular retina, corresponding to 1792 genes. In the RPE vs. choroid assay, 1127 probes were expressed 2.5-fold higher in RPE than choroid, corresponding to 1066 genes.

6.1 Comparison of differential expression

We compared the enriched genes from all three above studies to our RNA-Seq results [19]. For macular retina, 65/148 genes reported enriched by Bowes Rickman et al. [42] were enriched in our RNA-Seq gene list (44% agreement). For the peripheral retina, 75/135 genes reported by Bowes Rickman et al. [42] were enriched in both studies (55.6% agreement).

Comparing the RPE enriched genes between the studies by Booij et al. [37] and Bowes Rickman et al. [42], we found a small overlap of five genes. When comparing microarray results from Booij et al. [37] with our RNA-Seq data, we found 1021/1358 genes enriched in our RNA-Seq (75.2% agreement). When comparing our RNA-Seq data to the RPE data of Bowes-Rickman et al. [42], we find agreement for only ten genes (9.7% agreement). This discrepancy in the degree of overlap between our data and previous studies could be due to the small number of RPE-enriched genes in the SAGE experiments, or because the SAGE analysis required RPE enriched genes to fulfill stricter criteria, a requirement not used in the microarray or RNA-Seq studies. Whitmore et al. [32] also compared their dataset with Li et al. [19] for 91 genes implicated by GWAS in eye diseases. They found that the expression values roughly follow the diagonal, and FPKM values were consistently higher for genes in their RPE/choroid samples compared to the RPE/choroid/sclera samples of Li et al. [19]; the authors attributed this finding to enrichment with RPE/choroid in their samples that lacked sclera.

6.2 Comparison of Pathway Analysis

Bowes Rickman et al. [42] reported that individual macula-enriched genes were involved in pathways of metabolite precursors and energy substrates including glucose, alcohol, and monosaccharaides, but the only pathway found to be enriched was glycolysis. Our list of macula genes from RNA-Seq gave a slightly different result, being significantly enriched in psychiatric, neurological, and chemical dependency disease classes. Most of the pathways enriched were signaling pathways (MAPK signaling, Wnt), as well as glycosphingolipid biosynthesis. The biological processes enriched by RNA-Seq in macular retina included neuron generation/differentiation, ion transport, cell signaling, and vision.

Both macular and peripheral retina is dominated by rod photoreceptors, which are very sensitive to light and contribute minimally to color vision [43]. Bowes Rickman et al. reported overrepresentation of genes involved in detection, response, perception of light stimulus, cognition, and eye/structural development for peripheral retina [42]. This overlaps with the results of our RNA-Seq dataset.

The list of genes enriched in RPE reported by Bowes Rickman et al. [42] provided no significant pathways, most likely because of the small number of reported genes. On the other hand, Booij et al. reported that RPE enriched genes were represented in cardiovascular, vision, metabolic, and immune diseases [37]. The biological processes for these genes include cell adhesion, vasculature development, immune response, response to wounding, and cell motility. Genes from our RNA-Seq dataset gave similar functional enrichment and biological processes as Booij et al. [37]. The RNA-Seq enriched pathways were similar to the results of Booij et al. [37] with the addition of glycosaminoglycan degradation. In summary, our RNA-Seq dataset demonstrated that 1) for the comparison in retina, the differentially expressed genes that had higher expression levels in macula were significantly enriched for psychiatric and neurological disorders, for chemical dependency in disease category, and for ion transport, transmission of nerve impulse, and synaptic transmission in GO; genes that had higher expression levels in periphery were significantly enriched for translational elongation, sensory perception of light stimulus, and visual perception in GO, and for ribosomes in KEGG; and 2) The overall pattern was different for the comparison in the RPE/choroid layer. Differentially expressed genes that had higher expression levels in macula were significantly enriched for muscle contraction and multicellular organismal process in GO and for dilated cardiomyopathy, hypertrophic cardiomyopathy, and calcium muscle contraction in KEGG; genes that had higher expression levels in peripheral RPE/choroid were enriched for developmental in disease category and for biological adhesion, cell adhesion, and cell development in GO [19].

7. The transcriptome in AMD and glaucoma

In an effort to identify the underlying genetic architecture of AMD and glaucoma, multiple GWAS have been done on both diseases. In addition, direct approaches such as RNA-Seq or microarray transcriptome analysis of the diseased eye have been initiated.

7.1 Age-related macular degeneration

The AMD Gene Consortium recently reported a meta GWAS analysis of 17,000 patients and 60,000 controls and found 19 loci associated with advanced AMD [44]. Newman et al. completed transcriptome profiling of age-matched normal and AMD donor eyes with the Agilent Whole Human Genome in situ oligonucleotide array [45]. They identified a number of differentially expressed genes in both the RPE/choroid and retina using a permuted p-value < 0.1 and fold change ≥ 1.5 as cutoffs. We compared the genes found near the 19 AMD disease loci by the AMD Gene Consortium with the genes found differentially expressed between diseased and normal eyes by Newman et al. [45] and found an overlap of 24 genes (Table 3). To assess the relationship among these 24 genes, we performed a DAVID analysis. Concentrating on known disease phenotypes, we found an enrichment of terms related to age-related macular degeneration, cardiovascular disease, type 2 diabetes, Alzheimer’s disease, and other neurological diseases (Figure 3). A second smaller cluster was enriched for cancer terms (specifically, breast and colorectal cancer). Functional enrichment analysis of these 24 genes showed enrichment of terms for immune response, regulation of immune processes, and regulation of apoptosis, as well as negative regulation of apoptosis (Figures 4,5). The overlap of these susceptibility loci and the identified differentially expressed genes may be coincidental. However, it may also suggest there is yet another layer of complexity to AMD.

Table 3.

The 24 overlapping genes found between GWAS signals and the Newman et al. study. The related SNP ID from the GWAS study is listed as well as the direction of expression in the tissue.

Up-Regulated in AMD retina
Gene Symbol	Gene Name	Related SNP
DDR1	discoidin domain receptor tyrosine kinase 1	rs3130783
GTF2H4	general transcription factor IIH, polypeptide 4	rs3130783
STK19	serine/threonine kinase 19	rs429608
CFI	complement factor I	rs4698775
C3	complement component 3	rs2230199
APOE	apolipoprotein E	rs4420638
VEGFA	vascular endothelial growth factor A	rs943080
C4B	complement component 4B (Chido blood group)	rs429608
APOC1	apolipoprotein C-I	rs4420638
CCDC109B	coiled-coil domain containing 109B	rs4698775
B3GALTL	Beta 1,3-galactosyltransferase-like	rs9542236

Up-Regulated in AMD RPE
Gene Symbol	Gene Name	Related SNP
DDR1	discoidin domain receptor tyrosine kinase 1	rs3130783
IER3	immediate early response 3	rs3130783
GTF2H4	general transcription factor IIH, polypeptide 4	rs3130783
APOE	apolipoprotein E	rs4420638
VEGFA	vascular endothelial growth factor A	rs943080
APOC1	apolipoprotein C-I	rs4420638

Down-Regulated in AMD retina
Gene Symbol	Gene Name	Related SNP
SYN3	synapsin 3	rs5749482
HSPH1	heat shock 105kDa/110kDa protein 1	rs9542236
TNXB	tenascin XB	rs429608
EHMT2	euchromatic histone-lysine N-methyltransferase 2	rs429608

Down-Regulated in AMD RPE
Gene Symbol	Gene Name	Related SNP
CFB	complement factor B	rs429608
TNFSF14	tumor necrosis factor superfamily, member 14	rs2230199
TNFRSF10A	tumor necrosis factor receptor superfamily, member 10a	rs13278062
STK19	serine/threonine kinase 19	rs429608
ADAMTS9	ADAM metallopeptidase with thrombospondin type 1 motif, 9	rs6795735
VAV1	vav 1 guanine nucleotide exchange factor	rs2230199
NLRC5	NLR family, CARD domain containing 5	rs1864163
HSPH1	heat shock 105kDa/110kDa protein 1	rs9542236
TUBB	tubulin, beta class I	rs3130783
TNXB	tenascin XB	rs429608
C2	Complement component 2	rs429608

Figure 3. Disease term enrichment of 24 genes near AMD disease loci.

X axis shows the disease terms; Y axis shows the number of genes enriched in the specific disease term.

Figure 4. Functional enrichment analysis of 24 genes near AMD disease loci.

X axis shows the biological function categories; Y axis shows the number of genes enriched in the specific function category.

Figure 5. Pathway analysis of 24 genes near AMD disease loci.

X axis lists top 3 pathways; Y axis shows the number of genes enriched in the specific pathway.

Cellular immune and inflammatory response appears to be integral to the AMD phenotype. The pathway analysis also reflects a link between AMD and the genetic response that causes inflammation (Figure 5). While a definitive transcriptome of the AMD disease state has not yet been established, several groups are trying to define its expression profile and underlying genetic cause(s).

Whitmore et al. assessed gene expression in the RPE/choroid in nine early AMD and seven normal postmortem samples using the Affymetrix GeneChip Human Exon 1.0 ST array [46]. They found 75 genes differentially expressed between AMD and normal RPE/choroid with 52 downregulated and 23 upregulated in AMD. A DAVID analysis of the downregulated genes in the AMD samples revealed enrichment in terms associated with vision, sensory perception, and the plasma membrane. We compared the set of differentially expressed genes from normal vs. diseased RPE/choroid samples as reported by Newman et al. [45] and Whitmore et al. [46] and found an overlap of only 15 genes. There are important differences between the two studies. First, the AMD donors selected by Newman et al. included both pre-AMD as well as late AMD [45]. Whitmore et al. [46] focused on early AMD (defined as eyes without GA or CNV) as determined by clinical ophthalmoscopy [44]. We did a DAVID functional annotation analysis of the 15 overlapping genes and found enrichment for terms such as differentiation, developmental protein, glycoprotein, and transport. A number of these genes are also transcriptional-factor targets. This suggests that there are early gene expression changes in early AMD that could serve as potential biomarkers. Whitmore et al. analyzed the top 35 genes reported as RPE-specific by Booij et al. [37] and found 13/35 genes to be increased in AMD samples [46].

7.2 Glaucoma

Glaucoma is a neurodegenerative disease resulting in damage to the optic nerve. The trabecular meshwork (TM) in the anterior segment of the eye is a series of canals that drain aqueous fluid. These canals narrow with age causing an increase in intraocular pressure (IOP) that is associated with glaucoma. The change in gene expression associated with the narrowing of TM canals has been studied to identify potential causes of glaucoma.

Liu et al. analyzed the expression profile of TM from 13 normal and 15 glaucomatous eye samples, ages 40 to 86 years old, with Illumina Human WG-6 BeadChips [47]. A total of 483 differentially expressed genes existed between normal and glaucomatous TM. Gene ontology analysis revealed enrichment for terms such as signal peptide, secreted, glycoprotein, extracellular region, cell adhesion, phosphate transport, and calcium ion binding. Also reported in this paper was one Primary Open Angle Glaucoma (POAG) case with a Myocilin (MYOC) mutation, a gene known to be associated with POAG. Differential expression between the POAG eye with MYOC mutation and the POAG eyes without MYOC mutation revealed 55 genes with fold changes ≥ 2-fold. Myocilin was not found to be differentially expressed between the MYOC and non-MYOC POAG cases. We compared the genes located in the POAG GWAS loci [48] with the 483 genes differentially expressed in the TM [47] and found no overlap.

8. Bioinformatic databases for human retina and RPE/choroid transcriptomes

It is becoming common practice to make expression information available via databases. Such resources, when maintained, can be of significant value for storing and querying genomic data. Examples of genomic databases that are commonly used include NCBI, dbSNP, Genome Browser, OMIM, GEO, UniProt, and others. Lesser known are the databases for a specific tissue or cell type.

The RetinaCentral database (http://www.retinacentral.org), a repository of information on the mammalian retinome, was created to collect and store information on genes and proteins expressed in the retina. It encompasses data from 27 studies that employed a range of techniques including cDNA sequencing, microarray, and SAGE [49]. The web interface allows querying and filtering based on chromosomal location, category (such as retinal localization or retina/RPE specific), study, and phenotype. Although last updated in 2005, it still serves as a useful reference for retina/RPE gene expression profiling and the selection of candidate genes for further study.

Alternative splicing adds another layer of complexity to the human transcriptome and interest in isoforms has grown substantially. The Human Alternative Splicing Database (HASDB) is a database of tissue-specific alternative splicing for human genes and includes 46 different human tissues including retina [50]. Bayesian statistics are used to create a scoring function of tissue specificity (TS), as well as a measure of robustness (r_TS) to assess stability of the TS value. Extensive validation was performed for tissue specificity. Although the HASDB itself can no longer be found online, HASDB data are incorporated into a successor database known as the Alternative Splicing Annotation Project (ASAP).

Katsanis et al. have published an excellent resource detailing the design of several computational tools for the manipulation of dbEST to predict clusters in specific tissues [51]. In addition to providing a mechanism to manipulate dbEST to find ESTs specific to a tissue, the study focused on the retina and by association positional candidates for disorders of the retina. They combined information from the identified retina-specific ESTs with the “genome-mapping locations for retinopathy linked loci” to identify positional candidates. This information was incorporated into a database called RETBASE. While this repository was useful, it is no longer available online.

RETINOBASE (http://www.lbgi.fr/wikili/index.php/RetinoBase) is a database designed for the analysis and visualization of microarray gene expression profiles of the retina summarizing datasets from 28 different microarray experiments. Five different model systems are represented in these datasets including drosophila, zebra fish, rat, mouse, and human [52]. It appears that this database is no longer active.

EyeSAGE (http://neibank.nei.nih.gov/EyeSAGE/index.shtml) is a relational database that provides a comprehensive picture of gene expression in the human macula and includes transcription profiles of the retina and RPE/choroid [42]. The database was generated from SAGE (serial analysis of gene expression) libraries, both long and short SAGE, and also includes some microarray data. Validation was carried out for several known retina-specific genes to corroborate differential expression that was detected by either SAGE or microarray. The database is available for download in Microsoft Access format on NEIBank’s website. In addition, the NEIBank’s EyeBrowse can be used to visualize expression levels for various genes or candidate disease regions from the EyeSAGE datasets.

Data from an Affymetrix GeneChip Human Exon 1.0 ST analysis of ocular tissues is available [35]. Based on six postmortem eyes, the tissue profiling includes retina, optic nerve head, optic nerve, ciliary body, TM, sclera, lens, cornea, RPE/choroid, and iris. Normalization was done using the Probe Logarithmic Intensity Error (PLIER) method. The web interface allows users to query both transcript and exon level PLIER expression estimates from various ocular tissues.

Finally, RetNet (https://sph.uth.edu/retnet/) is a database based on published literature was created for the purpose of providing genes and loci associated with retinal diseases [53]. It is a collection of accepted retinal disease genes/loci found in the literature.

9. Conclusions

The transcriptome encompasses all RNA molecules within a cell or a population of cells. It represents the link between an organism’s genome and its phenotype. Tools for studying RNAs have been available for decades and include Northern blots, reverse-transcription PCR (RT-PCR), and expressed sequence tags (ESTs). The transition to a rapid and high-throughput quantification of the transcriptome became possible with the development of serial analysis of gene expression (SAGE), gene expression cDNA microarrays, and RNA-Seq.

Recently, RNA-Seq has shown a more complex retinal transcriptome than previously reported. Farkas et al. [29] and our group [19] performed comprehensive analysis of the retina and RPE gene expression profiles using RNA-Seq. Both studies revealed large differences in expression between these layers of the posterior eye and a smaller, but significant number of differences between macula and periphery of each layer. With even larger scale RNA-Seq datasets, it will be possible to assess the impact of gene expression on molecular networks that will lead to an understanding of how RNA/protein variation results in disease.

Highlights.

• We provide a brief overview of the eye structure with an emphasis on its posterior layers

• We summarize the current state of the human retina and retinal pigment epithelium/choroid transcriptome

• We compare high throughput techniques including serial analysis of gene expression (SAGE), gene expression cDNA microarrays and RNA-Seq

List of abbreviations

AMD

age-related macular degeneration

eQTL

expression quantitative trait loci

GWAS

genome-wide association studies

FPKM

Fragments Per Kilobase of gene per Million mapped fragment

MAQC

MicroArray Quality Control

MR

macular retina

PLIER

Probe Logarithmic Intensity Error

PR

peripheral retina

RPE

retinal pigment epithelium

RNA-Seq

RNA Sequencing

SAGE

serial analysis of gene expression