Abstract
Recent Genome-wide Association Studies (GWASs) for eye diseases/traits have delivered a number of novel findings across a diverse range of diseases, including age-related macular degeneration (AMD), glaucoma and refractive error. However, despite this astonishing rate of success, the major challenge still remains to not only confirm that the genes implicated in these studies are truly the genes conferring protection from or risk of disease but also to define the functional roles these genes play in disease. Ongoing evidence is accumulating that the single nucleotide polymorphisms (SNPs) used in GWAS and fine mapping studies have causal effects through their influence on gene expression rather than affecting protein function. The biological interpretation of SNP regulatory effects for a tissue requires knowledge of the transcriptome for that tissue. We summarize the reasons to characterize the complete retinal transcriptome as well as the evidence to include an assessment of differences in regional retinal expression.
Introduction
All living unicellular and multicellular organisms constantly communicate with the environment and must be ready to adapt to environmental changes that include nutrition, exercise, pollution and smoking, etc. The ability to rapidly sense and integrate signals and elaborate a response to them is a signature of the plasticity and robustness of a biological system (1). The response to external changes occurs through the activation of specific cellular programs that respond rapidly through alterations in gene expression. Gene expression responses to the environment can include modifications in RNA isoform expression influenced by alternative splicing enzymes and changes in RNA concentrations regulated by microRNAs (miRNAs), long non-coding RNAs (lncRNAs) and histone modifications. For example, alternative splicing is known to cause isoform heterogeneity in 92–94% of human genes that can rapidly respond to the environment (2). Therefore, it is critical to map the alternative splicing events of a tissue in order to understand how this mechanism can react to the changes in environment and in some cases, result in disease.
In order to advance our understanding of eye disease the vision community needs to develop an ocular tissue database for normal gene expression that includes RNA concentrations for coding and noncoding RNAs, alternative splicing events for that tissue, and potential epigenetic changes, i.e. methylation and histone modifications, that occur in response to the surrounding environment. Such a database will be a valuable resource for assessing the genomic landmarks specific to diseased tissue and elucidate the role of these altered genomic components on normal cellular function (Fig. 1). This approach of characterizing the genome on normal tissue before analyzing disease tissue follows the opposite path of the genome-wide association study (GWAS) which has been successful in identifying potential biological pathways but less successful in identifying causative variants within genes. Alternative analysis methods are especially needed to advance our understanding of diseases like age-related macular degeneration (AMD), glaucoma and diabetic retinopathy.
Alternative approach to identifying the cause of a complex disease trait. Genomic components are first characterized in the normal tissue affected by disease. These genomic characteristics are then packaged together using an array of bioinformatic programs to understand the normal interactions between DNA, RNA and proteins. Perturbations in these interactions can then be evaluated in the diseased tissue.
The human retina is a complex structure consisting of millions of cells packed together in a tightly knit network spread over the surface of the back of the eye as a thin film only 1/4 millimeter thick with a vascular bed (the choroid) of similar thickness behind it. The retinal cells can be divided into six basic cell classes: photoreceptor, retinal ganglion, horizontal, bipolar, amacrine and glial cells (3). Within each cell type there are multiple subtypes that are defined by morphological, physiological and molecular criteria. Masland has estimated >60 distinct cell types in the mammalian retina, thus increasing the complexity of the retina beyond the six basic classes (4). Additionally, neurons and glia in the human macula, a region specialized for detailed vision, differ both in number and type from their counterparts in the peripheral retina. Similarly, the choroidal vasculature has numerous cell types including vascular and lymphatic endothelium, fibroblasts, melanocytes, mast cells, autonomic neuronal ganglia, and resident and transient cells of monocyte lineage. Further refinements in the functional identities of these various cell types will eventually require single-cell RNA-Sequencing to understand the role of these cells in normal physiology and how they respond to the environment through changes in gene expression. Such knowledge will provide the foundation for cell-based therapies that are likely to be more specific with fewer side effects compared to current therapies directed at eye tissues.
Herein, we provide our views on some important aspects of retinal gene expression that are in progress and require more emphasis to advance our understanding of retinal diseases and eventually lead to new and more efficient therapies. This review will not discuss the complete list of previous retinal gene expression studies which have been well-summarized previously (5–7).
GWAS Signals are Enriched for Regulatory Variations
Multiple GWAS’s for complex eye disease have suggested multiple biological pathways contributing to these disorders (www.ebi.ac.uk/gwas) (8). Currently, there are 83 GWAS reported for eye disorders. Despite the impressive success rate for discovering disease susceptibility loci, few, if any, GWAS results have delivered new therapies. This can be explained by the design of the GWAS that provides a list of associated genomic loci but not directly identify the causative gene. Therefore, GWAS association peaks identify a handful of gene candidates that require additional studies to find the causal gene. Additionally, it now seems that most GWAS signals result from regulatory variation-alleles that impact gene expression-rather than amino acid changes that would impact protein structure (9). Because regulatory elements, like enhancers, often act over very long genomic distances, a large number of candidate genes may need to be further studied to determine the gene target of the enhancer. Because gene regulation can result in tissue specific expression of coding and noncoding RNAs, alternative splicing and epigenetic changes, the transcriptome needs to be characterized for each tissue involved in a specific disease in order to efficiently use it to find a causative gene. Some complex tissues, like the retina with over 60 different types of cells, may require characterization of single cell transcriptomes in order to capture marker genes for specific cell types that may not be detected at the tissue level.
Recently, Li et al. (10) analyzed the action of DNA variants on eight lymphoblastoid cell phenotypes that included chromatin modification, open chromatin, methylation levels, transcription rate, steady-state mRNA levels, RNA decay rates, ribosome occupancy and steady-state protein levels. They found that SNPs located within transcription start sites, enhancers and promoters regulate expression levels while SNPs affecting splicing were more likely to affect protein function rather than overall expression. Such detailed accounting for SNP effects on gene expression requires a comprehensive understanding of these variants on gene regulation from chromatin to proteins. Currently, much of this information is not available for the retina.
Recently, the Genotype-Tissue Expression (GTEx) network undertook a huge effort to assess the influence of GWAS-identified SNPs on gene expression in multiple human tissues and published transcriptomes for 1,641 tissues from 175 individuals (9,11,12). Eye tissue transcriptomes are not represented in this network. Some of these SNPs have been shown to affect expression levels (eQTLs) or splicing (sQTLs). Given that some eQTLs and sQTLs are tissue-specific and their effects on disease phenotypes can be limited to those tissues, GTEx recommends identifying QTLs in disease-relevant tissue. Systematically generated eQTL and sQTL information can provide immediate insight into the biological basis for disease associations identified in GWAS through the identification of gene networks involved in disease pathogenesis (13). For example, eQTLs from adipose tissue and lymphoblastoid cell lines provided leads into risk loci related to obesity (14) and immunity-related diseases (15). Li et al. (10) also found immune related QTLs in lymphoblastoid cells that were predicted to contribute risk for rheumatoid arthritis, multiple sclerosis, and height. Looking forward it will be important for the vision community to develop a GTEx-like database of eQTLs and sQTLs for each chorioretinal tissue layer and region, a critical need to expand our understanding of complex eye disorders like AMD, glaucoma and diabetic retinopathy.
Gene Expression Variation in Human Retina
RNA-seq is the direct sequencing of transcripts by next-generation technologies with strong potential to replace microarrays for whole-genome transcriptome profiling. Coupled with DNA variations, RNA-seq has altered our view of the functional diversity and regulation of the eukaryotic transcriptome. It is now apparent that transcript isoforms, alternative splicing, and non-coding RNAs (ncRNAs) such as long non-coding RNAs (lncRNAs), increase gene and phenotypic diversity and makes one cell differ from another.
The first RNA-seq study of human retina was reported by Farkas et al. (16). From three normal human retinas, they identified nearly 80,000 novel alternative splicing events and ∼30,000 novel exons. Additionally, they identified 116 potential novel genes that had not been identified previously. The investigators validated almost 15,000 of these novel transcript features and found that more than 99% of them could be reproducibly detected. Several thousand of the novel exons appear specific for the retina. In total, the novel mRNA sequence increased the number of exons identified in the human genome by 3%. This work will add to our understanding of what makes the retina unique.
In another recent RNA-seq study, Li et al. (17) identified most of the mRNAs produced in the human retina from eight normal eyes and concluded that the majority of the more than 20,000 human genes are expressed in the retina. This study also performed between-region comparisons within the same tissue as well as between-tissue comparisons within the same region. About 59–83% of their results were replicated in an independent set of eight individuals from the same population. This study provided the first evidence of regionally specific gene expression, with >2,000 genes being differentially expressed between macular and peripheral retinas and >900 genes differentially expressed between macular and peripheral RPE/choroid/sclera. The differences between macula and periphery were attributed to differences in neuronal/glial populations like the macular Henle fiber layer, and the high expression of cellular receptors in peripheral regions. An important strength of Li et al.’s study is that the eyes were recovered from donors and preserved within 6 h postmortem.
More recently, Pinelli et al. (18) performed RNA-seq analysis of 50 retinal samples from non-visually impaired post-mortem donors. They performed both de novo transcript assembly as well as gene expression quantification of known transcripts. Through de novo assembly, they identified 77,623 transcripts from 23,960 genes covering 137 Mb of the human genome. Most of the transcripts (92%) were multi-exonic: 81% had known isoforms, 16% had unique isoforms and 3% were novel genes. For known transcripts, the authors quantified expression levels across 94,521 known transcripts in 13,792 genes. This study found retinal expression for 65% of the protein-coding genes in Gencode.
Variations in Alternative Splicing
Alternative splicing is an important mechanism for providing variation in gene expression. It occurs in a tissue-specific manner in >95% of multi-exon human genes (2). It can be disrupted by genetic variants affecting the ‘splicing code’ as well as the addition of actual splice sites during exon inclusion. Because 15–60% of known disease-causing mutations affect alternative splicing, it is crucial to catalogue normal retina splicing to interpret the effect of alternative splicing in retinal disease. In a recent study by Li et al. (17), numerous alternative splicing events in the normal human retina were detected suggesting a role for splicing to increase the transcriptional complexity of retinal gene expression. Farkas et al. (16) also studied alternative splicing events, although with a relatively smaller sample size, but similarly concluded that the prevalence of alternative splicing in human retina was high. While alternative splicing has been shown to occur in normal aging eye tissue of RPE/choroid/sclera (17) and AMD retina (19), it remains an understudied area.
Expression Variations in Non-Coding RNAs
There is now ample evidence that most transcribed RNA is not translated into protein and these ncRNAs have critical functional and regulatory roles in health and disease. One well-recognized class of ncRNAs important in mammals is miRNAs, which are generally 19–24 nucleotides in length and are important for posttranscriptional regulation of mRNA transcripts. It is estimated that they target up to a third of all human mRNAs, and are known to mediate their effect on mRNAs through translational repression or mRNA degradation (20). miRNAs are largely derived from introns of protein coding or noncoding genes. While miRNAs may share the regulation of their host gene, they may also have separate promoters and be independently regulated (21). Mutations in the 2–8 bp ‘seed’ region of the mRNA binding site, located in the 3’ UTR of the target gene, can prevent miRNA binding to target mRNA, likely resulting in increased target expression (22). Moreover, differential expression of miRNAs has been linked to disease; for example, complement factor-H (CFH), a protein that plays a key role in AMD, is negatively regulated by both miRNA-146a and miRNA-125b, both of which are up-regulated in Alzheimer’s Disease (23). Moreover, given their role in AMD pathogenesis, several miRNAs have been proposed as potential therapeutic targets (24–27).
A separate class of ncRNAs, the lncRNAs, also have incredibly diverse roles in regulation. Many lncRNAs are expressed from intronic regions of genes, sometimes with a separate promoter, but others are found outside existing coding genes and have the special designation ‘long intergenic noncoding RNAs’ or lincRNAs. The broad class of lncRNAs are distinguished from microRNAs in being considerably longer (> 200 nt), and more widely engaged in various interactions, akin to proteins in their functional and structural diversity.
The first lncRNA to be widely studied is XIST, which plays a key role during early development in the inactivation of one of the pair of X chromosomes inherited in female mammals (28). An increasing number of lncRNAs have been discovered which function by binding to a genomic recognition site and recruit protein partners. These include lncRNAs capable of operating fully in trans, with multiple binding sites at widely separated locations. An example of epigenetic regulation operating in trans is MEG3, which acts as a guide for the polycomb repressive complex PRC2 (29). MEG3 was found to recognize binding sites near many genes associated with the Transforming Growth Factor-Beta (TGF-beta) pathway, which has been suggested as an important factor in AMD (30). LncRNAs can also act in cis, such as enhancer lncRNAs (eRNAs) that are transcribed near enhancer loci, and modulate the enhancer by interacting with transcriptional proteins (31,32).
The advent of high-throughput sequencing and sensitive methods for gene expression measurement has led to the recognition of ncRNAs as an important component of the transcriptome. However, despite growing interest in this class of molecules, the importance of noncoding RNAs in normal retina and disease is understudied.
Example: Differential Gene Expression in AMD and Fine Mapping of GWAS Signals
Recent GWAS of AMD have identified significantly associated SNPs but translating these GWAS findings into clinical benefit requires knowledge about the functional consequences of these SNPs. A key feature of the AMD GWAS is the location of common variants within genomic regions with no apparent function. In some cases, e.g. CFH, the effects of genetic variants can be interpreted at the cellular level. Still, it is difficult to understand their effect at the whole organism level owing to the large number of direct and indirect interactions between DNA variants and AMD phenotypes. Information about gene expression can provide insight into biological pathways that influence the development of AMD. In a pilot study, we generated RNA-seq data for macula and peripheral retina (MR, PR) and macula and peripheral RPE/choroid/sclera (MRCS, PRCS) from five AMD and seven normal eyes. For retina, we detected 958 differentially expressed (DE) MR genes between AMD and normal eyes while comparisons between MR versus PR, PRCS or MRCS resulted in only 35, 39 and 142 DE genes, respectively (Fig. 2). Strikingly, >95% of the DE genes identified in MR were upregulated in AMD eyes compared to normal eyes. Similar patterns were observed in PR, PRCS and MRCS, although upregulated genes were over-represented to a lesser degree. Among the 958 DE MR genes, 37 overlapped with previous AMD GWAS loci in which 5 genes (CFH, CFD, CFHR1, C2and C3) are from the complement pathway. The relationship between the CFH gene and AMD was the first association identified by GWAS (33–36). Many proteins of the complement cascade including CFH localize to drusen (extracellular material deposited below RPE). Importantly, the CFH differential expression is localized to the macula retina where it has a 32-fold higher expression in AMD versus normal. The geographic localization for elevated CFH expression could help explain why AMD is a macula disease. Secondly, we found that CFH over-expression is exclusively found in geographic atrophy but not in early AMD, raising the intriguing possibility that CFH inhibitors may be efficacious in AMD with GA.
Distribution of calculated fold-change (log2) of differentially expressed protein coding genes with FDR ≤ 0.05, one or the other FPKM ≥ 1, and fold difference ≥ 2, based on CuffDiff analysis against HG38. Calculated fold-changes for upregulated genes are plotted in red and downregulated genes plotted in blue. Note the significant RNA overexpression in the comparison between AMD and normal macula retina. A, comparison of AMD and normal macula retina; B, comparison of AMD and normal rpe/choroid/sclera; C, comparison of AMD and normal peripheral retina; D, comparison of AMD and normal peripheral rpe/choroid/sclera.
Results from gene expression analysis can also help prioritize GWAS findings. The AMD Genomics Consortium recently identified 52 independently associated common and rare variants (P < 5 × 10−8) distributed across 34 loci (37). However, most variants are located in non-coding regions of the genome, and some of them are embedded among many genes, making it difficult to directly link any one gene to disease. One technique of narrowing the candidate gene list in GWAS loci is to identify genes within the loci that are differentially expressed. ‘Looking under the lamppost’ in this way prioritizes genes for further functional studies. For example, SNP chr6:31930462 is located in a gene-rich region, harboring >100 genes in its credible set (30). From our pilot RNA-seq data, we identified three genes within this locus that are DE between AMD and normal eyes (major histocompatibility complex, class I (HLA-C) and class II (HLA-DRA, HLA-DRB), suggesting that the GWAS hit (SNP chr6:31930462) may associate with AMD risk by influencing gene expression of these three genes. These three genes are thus candidates for directly impacting AMD risk and will require functional studies to determine their importance.
Cellular Heterogeneity of Gene Expression in Retina
The retina is a complex tissue comprising more than 60 cell types (4). However, nearly all transcriptome analyses in human retina to date utilize intact tissue samples, yielding RNA profiles averaged from a mixture of cells. Advancing our treatment of retinal diseases will require a more complete characterization of the cell types that should include identification of individual cell transcriptomes. Because disease modifies the relative proportions of cell types, it is important to know if expression changes result from altered cell numbers or from altered transcription in individual cells. Recent technological breakthroughs in single-cell RNA sequencing (scRNA-seq) have made it possible to measure gene expression at the single-cell level, thus paving the way for exploring gene expression heterogeneity among retina cells. ScRNA-seq can now be applied to measure the transcriptomes of a large number of cells through the use of microfluidics device or other techniques. Since 2009, scRNA-seq has succeeded in many applications and revealed numerous insights into diverse biological systems (38).
ScRNA-seq requires the capture and lysis of single cells, reverse transcription of mRNAs to obtain cDNA, amplification of small amounts of cDNA by PCR, and finally preparation of a sequencing library. There are many different approaches to capture cells and amplify cDNA. A recent study compared 15 protocols computationally and four protocols experimentally to assess the ability to detect and accurately quantify gene expression (39). This publication concluded that whereas different protocols vary widely in their detection sensitivity, with lower limits between 1 and 1,000 molecules per cell, their accuracy in quantifying gene expression is generally high. Protocols with high sensitivity are more suitable for analyzing weakly expressed genes, or for gaining additional insights into subtle gene expression differences affecting individual cell states, but may be less suitable for other scenarios.
Variability in gene expression levels across cells can arise from both technical and biological factors. Current scRNA-seq protocols are complex and introduce technical biases that vary across cells, which can bias downstream analysis without proper adjustment. To interpret biological variability, it is critical to accurately estimate and account for technical variability (40). Several recent studies have used scRNA-seq to group retinal cells into types based on gene expression signatures (41,42). With scRNA-seq data, it is also possible to compare expression distribution between the two alleles of a diploid organism and the characterization of allele-specific bursting (43). Wills et al. (44) reported evidence that many heritable variations in gene function, such as burst size, burst frequency, and gene expression correlation between cells, are masked when expression is averaged over many cells in bulk expression analysis, and more importantly, these cellular level variations are regulated by bursting eQTLs. Given our finding of more DE genes between AMD and normal eyes in MR than in other tissue layers, comparing normal and AMD transcriptomes by cell-types would provide insights into the mechanistic and network effects of gene expression variability, and uncover novel candidates for future functional studies.
Conclusion
Genomics research has yielded many interesting findings that have improved our knowledge of common diseases. Yet 15 years after the first human genome sequence map, it is clear that studying DNA alone is insufficient to uncover the complete genetic basis of complex diseases. The transcriptome represents a vast layer of variations beyond DNA sequence that is now well studied in non-ocular tissues. It is now time to complete a detailed characterization of the human retina transcriptome to generate new biological hypotheses that will drive the vision science community forward for decades. Such advances will stimulate research linking genetic variation to cellular pathology, development of high-fidelity model systems, translation of molecular knowledge to diagnostic clinical imaging, and use of systems biology to assess gene networks.
For example, a comprehensive database of retinal RNA expression could be referenced to histopathological phenotypes that are directly translatable to clinical imaging. Identified genes together with eQTL and sQTL SNPs can be expected to provide new targets for therapy, generate novel ideas about known pathways, and motivate future studies using mass spectrometry and immunohistochemistry.
Acknowledgements
This research was supported by R01EY023164 to D.S. and R01GM108600 to M.L. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflict of Interest statement. None declared.
References
