Generation of a foveomacular transcriptome

Human fovea tissue and RNA

Cadaver eyes were obtained from the National Disease Research Interchange (NDRI, Philadelphia, PA) and as such are exempt from IRB approval. Human eyes of Caucasian origin were obtained from the National Disease Research Interchange and from individual eye banks. Tissue samples were excluded if they were enucleated from donors with any reported ocular diseases or genetic abnormalities. To obtain human foveal tissue, human donor eyes were dissected on ice. The posterior pole of the eyes, containing the retina, was dissected free of the overlying vitreous. The retinal tissue surrounding the central foveola to a radius of 0.75 mm centered over the foveolar umbo was dissected essentially free of RPE and choroid, free of sclera, and flash frozen on dry ice. In all cases, dissected tissues were stored at −70 °C until RNA extraction. For RNA extraction, foveae tissue from ten pairs of donor eyes was pooled, total RNA was extracted, and poly(A⁺) RNA was prepared using standard methodologies (Tel-test, Friendswood, TX; Oligotex, Qiagen, Chatsworth, CA).

Raw data analysis

Nylon membrane macroarrays bearing 248,832 human cDNA clones were purchased from Research Genetics (Huntsville, AL) and screened as recommended by Research Genetics. Clones on the Research Genetics arrays are double spotted on a set systematic grid pattern so that the location of hybridization signal for a pair of spots as defined by Research Genetics was used to elucidate gene identity. A second set of arrays, consisting of duplicate spotted 18,300 partially sequenced cDNA clones, all between 500 and 700 nucleotides in length, was purchased from Genome Systems (GDA version 1.3, St. Louis, MI) and screened as recommended by Genome Systems. Both sets of macroarrays were hybridized with a radioactively labeled mixed cDNA probe derived from pooled human fovea RNA. Mixed cDNA probes synthesized from poly(A+) RNA prepared from a pool of human fovea RNA were random-primed labeled using a combined mixture of ³²P nucleotides or ³³P nucleotides in a standard labeling reaction (Prime-It II, Stratagene, La Jolla, CA). Minimum activity used per hybridization was 5×10⁶ CPM/ml of hybridization buffer (Hybrisol II, Oncor, Gaithersburg, MD). Hybridizations were performed using the protocol recommended by the manufacturer of the hybridization buffer. Blots were subjected to stringent hybridization and washes, followed by autoradiography at −80 °C [16] or scanned on a Molecular Dynamic scanner. The RNA used to make the cDNA probes for the two screens were derived from different sets of donors; therefore, each screen represents an independent experiment. The fovea cDNA probe used for screening the Research Genetics arrays were derived from a fovea tissue pool obtained from donors whose age ranged from 2 to 79 years; the mean donor age was 42.6 years. The fovea cDNA probe used for screening the Genome Systems arrays was derived from a separate fovea tissue pool obtained from donors spanning 12–80 years of age; the mean donor age was 50.3 years.

The resulting autoradiographs for the Research Genetics arrays were analyzed by four individuals according to the manufacturer’s protocol. An autoradiograph signal was considered to represent a true fovea-expressed gene if all four individuals concurred. In the absence of consensus, a potential signal was considered negative, thus limiting false-positive results at the consequence of increasing false-negative results for this data set. Each confirmed positive hybridization signal address was retrieved from the Research Genetics EST database, which converted each identified signal into an IMAGE number. IMAGE IDs were then translated into GenBank accession numbers.

The molecular dynamic scans of screened GDA array filters were analyzed by Genome Systems using imaging software specialized for high-density array analysis (Array Vision; IMAGING Research, St. Catharines, Ontario, Canada). Accession numbers were collected for positive signals (the cut-off for a positive signal was twofold over the minimum value detected on the array) using the GDA software package. Accession numbers for ESTs derived from single pass sequencing of human macular clones from library 10,282 were obtained directly from the NCBI.

Data utilities

Perl scripts were written to search for each accession number in the DataBase of Expressed Sequence Tags (dbEST), UniGene, and Gene, and to harvest selected data from each database. Although GenBank accession numbers were our primary data type and are highly stable identifiers, Gene database identifiers are the central data type our database is modeled on, as this is a gene-centric effort. UniGene clusters were deemed too variable over time and were collected only to link ESTs identified to other ESTs thought to belong to the same transcript. Our Perl scripts formatted the retrieved results into a series of distinct files, corresponding to separate tables of our relational database. Figure 1 indicates the basic activities of each of our scripts. Figure 2 describes our MySQL relational database structure. A relational database structure is one in which related subtypes of data are organized into distinct tables, which are linked to one another by sharing common elements. These common elements, unique identifiers for a specific data point in a set, are the connections between tables, and allow for variably complex queries of the data.

Identification of positive clones

From our Research Genetics macroarray screen, we identified 16,646 positive clones representing foveally expressed genes. Some IMAGE numbers have no associated accession numbers (indicating that they had not been sequenced), and lacked a dbEST entry. The 16,646 positive clones corresponded to 10,281 GenBank accession numbers that were used for further analysis. Our GDA arrays identified 3962 positive signals indicative of foveally expressed transcripts. At the time of most recent analysis (spring 2010), the foveomacular library 10,282 contained 6,279 ESTs, and data regarding this library is freely available through the NCBI. In total, 20,522 ESTs were examined (10,281 identified by Research Genetics macroarrays, 3,962 by GDA array, and 6,279 from library 10,282).

Data mining

All three data sets were subjected to the same subsequent analyses. For our initial data collection, we harvested data from three databases maintained by the National Center for Biotechnology Information (NCBI): dbEST (DataBase of Expressed Sequence Tags), UniGene, and Gene. These databases were chosen because they are well established, their interfaces lend themselves to automated searching, and all of the information contained within belongs to the public domain.

The data types we chose to collect and their organization are described in Figure 2. Since our three data sets are EST-based, dbEST was a natural starting point for information retrieval. dbEST provides several basic points of data for our genes of interest, namely, library of origin, available sequence, and other identifiers unique to the corresponding clone of interest (such as IMAGE number and GenBank gi (gene identifier)) that are useful for further data mining. The Gene database provides information regarding mapping, genomic organization, known disease states and phenotypes linked to that gene, and functional data in the form of Gene Ontology (GO) annotations.

In our examination of the UniGene database, we found that 17,437 (84%) of ESTs identified were associated with a UniGene cluster. Typically, multiple accession numbers are associated with a given UniGene cluster; when multiple occurrences of UniGene clusters were parsed out, the non-redundant list of positive clusters was limited to 6162, for an average EST-to-cluster ratio of approximately 3:1. In most cases, a single UniGene cluster is associated with a single Gene database entry, but occasionally multiple clusters associate with a single gene, or vice versa. In the end, 6,056 unique GenedB entries were identified. Appendix 1 contains additional database schema and database tables described in this publication.

Foveomacular transcriptome

The 6,056 identified human foveomacular genes varied, ranging from well-characterized genes to transcribed loci with little additional information. Identification of the same gene in independent data sources provides internal experimental confirmation of foveomacular gene expression status. There is a 3% overlap of genes identified (190 genes) in all three experimental data sources surveyed (Figure 3, Appendix 2). Twenty-two percent of the genes (1,316 genes) were identified in two of the three data sources surveyed.

In our collection, 5,979 genes (99%) had a defined nucleotide mapping position in the human genome. The distribution of these mapped genes by chromosome is shown in Figure 4A. A broad measure of normal distribution of foveomacular expressed genes can be obtained by comparing the gene collection of the current study against estimates of whole genome gene distribution as described by the human genome build used, 37.1. A chi-square test comparing the two sets indicated that there was a significant difference in the number of genes per chromosome for the fovea and whole genome sets (p>0.005). We calculated the ratio of foveomacular expressed genes to total genes per chromosome and compared this to the overall average (16.5% with a standard deviation of 2.2; Figure 4B). Regarding our collection of foveomacular expressed genes, there is a slight overrepresentation of these genes on chromosomes 12, 16, 17, 20, and 22, and an under representation on chromosome 15 and the sex chromosomes.

Functional annotation was available for 5,355 of our identified genes. Of these, 69% had ten or fewer GO terms annotated. The transforming growth factor beta gene (TGFB, Gene ID 7040) was the most annotated gene with 108 annotations. The distribution of GO annotation frequencies for our genes is provided in Figure 5, showing a steady decrease in annotation frequency. A minority of genes identified in this study are extremely well characterized, while the majority of genes are only lightly annotated if at all. To further organize genes regarding GO terms, we applied the clustering utility DAVID [17,18] to those 190 genes identified by all three of our data sources (Appendix 3). Clusters formed around common functions, including signaling, oxidative response, and metabolism. Our Perl scripts are included in Appendix 4 and are released under a Creative Commons Attribution-Non Commercial license (scripts previously described in [19,20]).

We previously demonstrated that there is a strong correlation of retinal disease genes to be expressed in the foveomacula as well as the peripheral retina [21]. We compiled a list of 125 genes implicated in retinopathic or maculopathic conditions from RetNet (Appendix 5). We reidentified 57 (46%) of these retinopathic genes in our list of 6,056 foveomacularly expressed genes. We also compared our data to other published collections of foveomacularly expressed genes. Library 420 represents the first human fovea cDNA library made [16,22]; 100 ESTs defining 32 known human genes have been deposited in GenBank. Twenty of these genes were reidentified in our study. In an independent study, analysis of Incyte arrays identified 5,702 foveomacularly expressed genes [7]; 2,576 were reidentified in our data sets.

Comparing our EST-based foveomacular transcriptome with other published studies provides independent confirmation of foveomacular gene expression for 2,576 of the 6,056 genes collected in our current study; 3,480 genes were newly identified as foveomacularly expressed. Between the foveome and our two external comparisons with Library 420 and Incyte arrays, a total of 9,197 human foveomacular expressed genes can be defined (Appendix 2). Of these genes, 2,582 were found in at least one additional data set. Moreover, the published data sets (Incyte and Lib.420) identified 3,140 foveomacularly expressed genes that were not identified in our data set. The finding that there is incomplete overlap between experimental data sets indicates that none of the data sets considered define a full complement of human foveomacular expressed genes, and that more foveomacular genes remain to be discovered. Another possibility is that we have observed bias in gene expression due to individual genetics, age, and environmental backgrounds from donor samples used to generate our cDNA probe. If this is the case, a great deal of variation may be detected between individuals, primarily as low expressed transcripts.