Abstract
purpose. To reconstruct the transcriptional profile of the human adult retina
and the genomic map of the genes expressed in this tissue.
methods. Original software was used for the retrieval and analysis of records
from UniGene (http://www.ncbi.nlm.nih.gov/UniGene/) pertaining to
selected cDNA libraries from adult human retina.
results. The 4974 genes reported so far to be expressed in retina were included
in a catalog available on the Internet. For each entry, an estimation
of the level of expression of the corresponding gene in the retina was
provided. A high-resolution genomic map of the human retina was built
up by inclusion of 3152 genes showing a precise and unique map
assignment. The correspondence was established between 53 gene–orphan
retinal diseases and clusters of genes expressed in the retina.
conclusions. The in silico reconstruction of the transcriptional profile of the
adult human retina provides preliminary information on the pattern of
genomic expression in this tissue. The chromosomal location of many
retinal genes, combined with their expression data, should speed up the
identification of genes involved in retinal
diseases.
Catalogs of genes expressed in human retina have been published
based on 209, 736, 755, and 137 gene signatures.
1 2 3 4 Presently, more than 16,000 expressed sequence tags (ESTs; 6300 genes)
from human retina cDNA libraries are reported in UniGene, the
largest existing database of expressed genes
(http://www.ncbi.nlm.nih.gov/UniGene/, release January 14,
2000), thus potentially enabling an in silico
analysis
5 of the transcriptional pattern of the human
retina.
We report on an attempt to computationally reconstruct the
transcriptional profile of the adult human retina and to produce the
first on-line high-resolution genomic map of genes expressed in this
tissue.
Data pertaining to four unbiased UniGene cDNA libraries obtained
from normal adult retina (Lib.177; Lib.178; Lib.228; and Lib.313), were
mined from UniGene by specialized software developed in our laboratory.
Each UniGene record corresponds to a UniGene cluster and includes all
the available ESTs, genomic clones, and/or complete mRNA pertaining to
the corresponding human gene. In addition, gene name and description,
function of the corresponding protein, best SwissProt (database of
protein sequences) hits, mapping data, and tissue expression
are provided.
Records were collected from UniGene by the software and merged in a
single data set, removing redundancy. Different fields of each record
were analyzed, and entries were sorted out according to given criteria
(e.g., mapping information or EST content). Details of the process have
been described elsewhere.
6
The basic assumption underlying the reconstruction of the
transcriptional profile of the retina was that the larger the number of
retinal ESTs reported per entry, the more active the corresponding gene
should be in the tissue. Thus, to estimate the abundance of each
transcript in the retinal mRNA population,
1 6 7 8 9 10 we used
the number of ESTs reported for each given entry. The relative
contribution of a given gene to the retinal transcriptional activity
(expression level) was obtained by the ratio between the number of ESTs
corresponding to that gene and the total number of ESTs corresponding
to all the genes included in the data set. The expression of each
retinal gene in other human tissues was evaluated by recording the
number and the type of nonretinal ESTs reported for that gene.
The genomic map of genes expressed in human retina was constructed by
collecting the chromosomal localization of all the mapped UniGene
clusters included in the retina dataset. Genes showing multiple map
locations were excluded. The detailed procedure for building the
transcript map has been reported elsewhere.
11
To calculate the expected distribution of genes by chromosome, we used
the Human Gene Map database,
12 which includes more than
30,000 human genes. The statistical significance of the deviation from
the expectation was tested by a χ
2 goodness-of-fit test with 22 and 1
df, when analyzing the
genomic distribution of genes and the deviation of a given chromosome,
respectively. The level of significance was set to 0.002, according to
the Bonferroni correction
13 for multiple tests.
The RETNet Web site (release of January 25, 2000) was searched for
accurately mapped gene–orphan retinal diseases. The megabase interval
corresponding to the cytogenetic localization of each disease was
established according to the LDB database.
14 UniGene
clusters corresponding to genes expressed in retina and mapped to the
locus of a retinal disease were then identified.
The retrieval of UniGene entries belonging to the selected cDNA
libraries resulted in 4974 individual records, which were included in a
catalog available on the Internet
(http://telethon.bio.unipd.it/GETProfiles/retina/). In the catalog,
2284 entries corresponded to known human genes and 2690 (54.1%) to EST
clusters. More than two thirds of the EST clusters (1958 ESTs) showed
no similarity to already known DNA sequences.
The catalog entries were classified, according to the level of
expression of the corresponding genes: 443 corresponded to highly
expressed genes (more than 0.05% of the detected transcriptional
activity, more than eight reported ESTs per entry), 1162 to moderately
expressed genes (0.04%–0.015%), and 3369 to weakly expressed genes
(less than 0.015%).
Approximately 89% of the retinal genes were found to be expressed in
at least one additional tissue, and approximately 74% in more than
four additional tissues. Of note, approximately 70% of the retinal
genes were found to be expressed also in adult brain.
Taking together the 536 putative retina-specific genes and the 443
genes highly expressed in retina, a set of 979 transcripts may be
obtained that characterizes the adult human retina.
More than 60% of the genes included in the catalog (3152 genes) showed
a precise and unique map assignment and were used to build a
comprehensive high-resolution genomic transcript map of human retina
also available on the Internet
(http://telethon.bio.unipd.it/GETMaps/retina/). The confidence interval
of each gene localization resulted less than 2 megabases (Mb) for 1126
entries and less than 3 Mb for 1422 entries.
The observed distribution of retinal genes by chromosome significantly
deviated from the expectation (χ
2 = 70.5, 22
df,
P = 6 × 10
−7;
Table 1 ). In particular, a highly significant excess of genes was
observed for chromosomes 17 (χ
2 = 25.1, 1
df,
P = 5 × 10
−7) and 19
(χ
2 = 6.8, 1
df,
P =
0.0089).
RetNet (http://www.sph.uth.tmc.edu/Retnet/home.htm) provides a list of
cloned and/or mapped genes that cause retinal diseases. So far, 122
diseases have been reported there. We excluded 55 conditions in which
the involved gene was already known or for which the disease locus was
mapped to a very large interval. For the remaining loci, the
correspondence between cytogenetic band boundaries and megabase
distances was established, according to the integrated“
gmaps” of LDB (location database).
Table 2 reports the list of 53 disease loci, along with information on the
number of genes and EST clusters mapped in the corresponding interval.
The source of data in the present work was UniGene,
virtually containing all the ESTs sequenced so far from a number of
different human tissues and pertaining to more than 90,000 human genes.
The correspondence between UniGene entries and genes is generally
assumed.
The quality of the data analyzed and produced by the present study is
partially dependent on the quality of UniGene data. Although UniGene is
the less redundant among the gene indexing databases,
15 some UniGene clusters can include sequences of chimeric clones
belonging to different genes, and very large genes can be represented
in UniGene by two or more clusters. For this reason, before
constructing the genomic map of transcripts, we decided to discard all
the UniGene clusters showing multiple chromosomal locations.
The complementary situation (different UniGene clusters for a unique
human gene) is also possible, especially when dealing with different
UniGene clusters showing no similarity with any known sequence and
mapped to the same position. However, this situation is expected to be
rather rare.
The total number of genes expressed in retina is probably between
10,000 and 30,000.
2 Presumably, fewer than 70,000 genes
are active in a differentiated tissue. Therefore, the sample of 4974
individual transcripts considered by the present study corresponds, at
worst, to approximately 7% of the presumed total number of genes
expressed in retina, an adequate size for statistical inference.
The computational approach to the analysis of transcriptional profiles
is based on the assumption that the level of activity of a given gene
may be inferred from the information regarding the number of the
corresponding ESTs. The impossibility to detect differences in gene
expression resulting from posttranscriptional regulatory processes is a
strong limitation of this approach, but the same bias is shared by all
the present methods for estimating individual gene expression on a
large scale.
EST sequencing, from which the in silico approach is derived, is
intrinsically inadequate to identify truly rare genes. However, when
the sample size is sufficiently large, a fairly good quantitative
estimation of the transcription level of highly or moderately expressed
genes is possible.
16 On the contrary, quantitative
hybridization on high-density cDNA array fails to detect 80% of the
transcripts identified in a given tissue by EST
sequencing.
17
The transcriptional profile of the retina is characterized by a low
number of highly expressed genes, accounting for less than 10% of the
total number of genes, but for approximately 50% of the detected
transcriptional activity. The percentage of tissue-specific genes
(9.8%) is higher than previously reported in human adipose
tissue
10 and five times higher than observed in skeletal
muscle.
6 The difference could be ascribed to the high
specialization of the retinal tissue and/or to the relative absence of
contaminant tissues in the sample from which the cDNA libraries were
prepared. The subsample including highly expressed genes and genes
found exclusively in retina could be profitably used for monitoring
major changes in the transcriptional profile, after physiological or
pathologic modifications.
Data on the transcriptional profile of the human retina have been
reported in two independent studies. However, the small size of the
sample reported in BodyMap
3 and the different type of
source cDNA library
4 make it impossible to perform a
statistical comparison with the present data.
More than 60% of genes included in the catalog showed a precise and
unique map assignment. The distribution of genes by chromosome
significantly differed from the expectation. In a preliminary study, we
observed a peculiar concentration of retinal genes on chromosome
17.
18 This finding is confirmed by the present
investigation, based on a much larger number of map entries, which
suggests also a concentration of retinal genes on chromosome 19. The
short arm of chromosome 17 is known to be a hot spot to which several
phenotypically distinct retinal disorders have been
mapped.
19 The most recent version of the Human Gene Map
(HGM)
13 show that chromosome 17 is richer in genes than
expected under the hypothesis of a constant gene density along the
human genome. In the present study the expected distribution of genes
by chromosome was calculated according to HGM data. Therefore, the
hypothesis of a particularly high concentration of retinal genes on
chromosomes 17 and 19 is even stronger. A selective concentration of
genes in specific chromosomes was observed also in human skeletal
muscle.
11 18
Chromosomal maps showing the location of genes expressed in retina
represent a novel and important resource for positional cloning of
genes involved in retinal disorders. Most mapping information obtained
from UniGene resulted from radiation hybrids (RH) mapping,
which is, at present, the most precise method of gene mapping, if we
exclude the direct localization on the genomic DNA sequence. The
presence of several genes mapped to the same megabase distance from the
p-telomere probably reflects the actual concentration of genes in short
chromosomal segments. However, the width of the interval corresponding
to a single point of the present map is variable, because the physical
distance between the two flanking markers may be different. In spite of
this, the present genomic map of retinal genes is, at the moment, the
most precise representation of their cluster formations along each
human chromosome. Moreover, the resolution of the map is adequate to
detect which group of genes corresponds to a given interval to which a
human disease was mapped by linkage studies.
The in silico reconstruction of the transcriptional profile of the
adult human retina and the building of the genomic map of genes
expressed in this tissue provides for the first time a resource in
which functional and structural genomics of retina are integrated on a
large scale. It is hoped that this resource will speed up the
identification of genes involved in retinal disorders and will enable
innovative approaches to the study of retinal development, physiology,
and diseases.