Skip Navigation


Human Molecular Genetics Advance Access originally published online on September 30, 2003
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Supplementary Material
Right arrow All Versions of this Article:
12/22/2881    most recent
ddg326v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (23)
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Chowers, I.
Right arrow Articles by Zack, D. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Chowers, I.
Right arrow Articles by Zack, D. J.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

Human Molecular Genetics, 2003, Vol. 12, No. 22 2881-2893
DOI: 10.1093/hmg/ddg326
© 2003 Oxford University Press

Gene expression variation in the adult human retina

Itay Chowers1, Dongmei Liu4, Ronald H. Farkas1, Tushara L. Gunatilaka1, Abigail S. Hackam1, Steven L. Bernstein6, Peter A. Campochiaro1,2, Giovanni Parmigiani3,4 and Donald J. Zack1,2,3,*

1Wilmer Eye Institute, 2Department of Neuroscience and 3Department of Molecular Biology and Genetics, and the McKusick–Nathans Institute of Genetic Medicine, The Johns Hopkins University School of Medicine, Baltimore, Maryland, USA, 4Department of Biostatistics, Johns Hopkins University School of Public Health, Baltimore, Maryland, USA, 5Sidney Kimmel Comprehensive Cancer Center and 6Department of Ophthalmology, University of Maryland, Baltimore, Maryland, USA

Received June 4, 2003; Accepted September 18, 2003


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
 REFERENCES
 
Despite evidence that differences in gene expression levels contribute significantly to phenotypic variation across individuals, there has been only limited effort to study gene expression variation in human tissue. To characterize expression variation in the normal human retina, we utilized a custom retinal microarray to analyze 33 normal retinas from 19 donors, aged 29–90 years. Statistical models were designed to separate and quantify biological and technical sources of variation, including age, gender, eye laterality, gene function and age-by-gender interaction. Although the majority of the 9406 genes analyzed showed relatively stable expression levels across different donors (for an average gene the expression level value of 95 out of a 100 individuals fell within a 1.23-fold range), 2.6% of genes showed significant donor-to-donor variation, with a false discovery rate of 10%. The mean expression ratio standard deviation was 0.15±0.8, log2, with a range of 0.09–0.99. Genes selectively expressed in photoreceptors showed higher expression variation than other gene classes. Gender, age and other donor-specific factors contributed significantly to the expression variation of multiple genes, and groups of genes with an age- and gender-associated expression pattern were identified. Our findings show that a significant fraction of gene expression variation in the normal human retina is attributable to identifiable biological factors. The greater expression variability of many genes central to retinal function (including photoreceptor-specific genes) may be partially explained by the dynamics of the vision process, and raises the possibility that photoreceptor gene expression levels may contribute to phenotypic diversity across normal adult retinas. In addition, as such diversity may result in different levels of disease susceptibility, exploring its sources may provide insights into the pathogenesis of retinal disease.


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
 REFERENCES
 
DNA sequence variation within the coding regions of genes has traditionally been thought to compose the major biological factor accounting for normal phenotypic diversity. However, both intra- and inter-species variation in gene expression levels also exists, and such variation is likely to contribute significantly to phenotypic differences between individuals as well as between species (1–6). This gene expression variation appears to be at least partially genetically determined, and its regulation appears to be complex. For example, in Saccharomyces cerevisiae linkage analysis-based data suggests that most gene expression variation between strains is genetically controlled, and often involves multiple loci (7). Multiple promoter polymorphisms have been identified in humans and other species (8,9). On a molecular scale, stochastic noise at the level of transcriptional processes may be able to propagate to higher levels (10). On a broader scale, a study comparing gene expression in human lymphoblastoid cells found that individuals who are more genetically related tend to show less expression variation (11).

In addition to its biological significance, and what it can teach us about regulatory mechanisms, normal variation in gene expression is also an important factor to consider in studies focused on identifying differentially expressed genes across physiologic and pathological conditions (3,4). As there is increasing evidence that variation varies between different genes, and between different tissues, characterization of normal expression variation for individual genes is a prerequisite for identifying differentially expressed genes among populations in which relatively small magnitude expression level differences across multiple genes may exist. Such patterns of gene expression differences conceivably play a role in many of the common polygenic disease processes.

While microarray and other technologies have been used extensively to obtain expression profiles and identify individual genes whose expression varies across physiologic and/or pathological conditions, the nature and magnitude of normal gene expression variation in human tissue have received relatively little attention. To begin to address these issues, several recent studies have tried to quantitatively define the pattern of expression diversity between normal tissues, individuals, and species (2–6,12). Among the early interesting findings are that variation can be high, it is partially genetically determined, and that variation can vary by tissue, with the central nervous system demonstrating a particularly high degree of diversity (2). Efforts have also been begun to dissect out the effects of aging and gender on gene expression (13–24), but so far the information available is limited in terms of the contribution of specific biologic factors across all genes (for further information on these studies, see the Discussion).

The human retina, a part of the central nervous system, shows diversity across individuals that is reflected in epidemiological, functional, histological, and molecular biological parameters (12,13,25–27). Some of these differences can be attributed to factors such as age, gender and genetic background (26–28). For example, advanced age is associated with alteration in retinal function, as measured by electrophysiological and psychophysical tests, by rod photoreceptor and ganglion cell loss, and by increased susceptibility for common retinal diseases such as age-related macular degeneration (AMD) and glaucoma (27,29,30). These factors, together with the available broad knowledge of normal retinal biology, make the retina an attractive model for the study of human gene expression variation.

Here we describe such a study in which we utilized a large number of eyes from adult donors, a custom human retinal cDNA microarray (31), and a study design and statistical model developed to facilitate identification of subtle gene expression changes associated with biological and technical variables. We anticipate that characterization of this variation coupled with dissection of its biological sources will help provide insight into the complex pattern of retinal gene expression, a pattern that underlies phenotypic differences across tissues, across physiologic conditions within a tissue, and between health and disease. The availability of this normal gene-specific variability data will also aid in interpreting the significance of data from studies that are seeking to identify differences in gene expression between normal and diseased retinas.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
DISCUSSION
 MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
 REFERENCES
 
Overall contribution of technical and biological factors
Thirty-three human retinas from 19 donors (aged 29–90 years, 10 male and nine female, Table 1), all reported to be free of retinal disease, were analyzed in duplicate with a custom human retinal cDNA microarray using a common reference sample-based experimental design (31). The human array contains 10 034 cDNA fragments that were selected to represent genes thought to be of interest in retinal development, function and pathology, as well as other genes and expressed sequence tags (ESTs) that are expressed in the retina. In the present study, 9406 of the spots had a signal/background ratio higher than 1 in each of the two fluorescent channels in at least 75% of the microarray slides. The remaining 628 spots were omitted from the analyses.


View this table:
[in this window]
[in a new window]
  		Table 1. Retina do nors
 
To correct for technical variation, as well as to evaluate association of gene expression levels with biological factors, we analyzed the expression ratios of the 9406 accepted spots using a two-step regression model. The approach was chosen because it can separate and quantify the ‘biological’ component of variability, i.e. that which reflected true gene expression variation between donors, from the ‘technical’ components, i.e. that arising from hybridization-to-hybridization differences within a donor. The first step was used to derive donor-specific expression profiles that were adjusted for technical variation and eye laterality. The relative contribution of each factor to the overall variability of the measured gene expression levels was assessed by calculating the mean squared error (MSE) as well as the factor's MSE/total MSE ratio.

This analysis showed that technical variables had the largest MSE, followed by donor-related factors and eye laterality (Fig. 1 and Table 2). It should be noted that the seemingly large contribution of technical and unknown factors (residual) to the expression variation in the first regression is small in overall magnitude, since the expression levels of most genes were stable (see ‘gene expression variation’ below). The residual (unexplained) variability probably represents the combination of a variety of factors, such as differences among RNA preparations and among individual microarray slides.



View larger version (13K):
[in this window]
[in a new window]
  		Figure 1. Distributions of components of variation attributable to microarray slide batch and dye effects (solid line), variation across donors (biological variation; dashed line), and eye laterality (left versus right eye difference; dotted line). For each gene, we consider the distribution of the ratios between the MSE attributable to the factor, and the total MSE. This ratio is expressed in the log2 scale to facilitate visualization. Results are based on the first stage regression, in which the microarray slide-specific gene expression was used as the response variable, and microarray batch and dye effects, biological factors and eye laterality were used as predictors.

 

View this table:
[in this window]
[in a new window]
  		Table 2. Variation attributed to factors assessed in the first regression step
 
The component of variation that reflects ‘true’ differences between individual donors can be further broken down into genetic, environmental, stochastic and demographic factors such as age, gender, post-mortem handling of tissue, and others. This was accomplished through the second regression step (Table 3; see below for the results of this analysis, and Materials and Methods for details on how the regression analysis and modeling were performed).


View this table:
[in this window]
[in a new window]
  		Table 3. Variation attributed to factors assessed in the second regression step
 
Gene expression variation
The gene expression variation results arising from the first regression were evaluated in two ways. We first determined whether variation attributed to biological factors is likely to reflect true biologic phenomena rather than random fluctuations. This was performed by comparing the distribution of biological factor effects in the first regression step (Fig. 1) to that of a reference distribution obtained by permutation of donor IDs, followed by calculation of the false discovery rate (FDR) (32) (see Materials and Methods section for a description of FDR calculations). The analyses indicated that donor-related expression variation quantified in the first regression step significantly differed from variation expected by chance alone (Fig. 2A and B). For example, for 252 genes, more than 60% of the expression variation was contributed by biological factors. Based on the FDR shown in Figure 2B, we estimated that only 10% of these are false discoveries, i.e. 90% of the genes identified truly vary across subjects.



View larger version (21K):
[in this window]
[in a new window]
  		Figure 2. Expected (solid line) and observed (histogram) distribution of donor associated expression fraction from the first regression stage (A). The expected distribution was obtained by permuting the donor IDs and repeating the first regression step. The difference between the curves reflects the non-random character of the distribution. Quantitative assessment of the difference between expected and detected donor effect distribution was performed using false discovery rate estimation (B). The x-axis shows again fraction of variation contributed by donor effect. The y-axis shows the estimated false discovery rate for each expression fraction cut-off level. Distribution of standard deviations is shown in (C) for expression ratios of all genes of known function on the array (solid line), photoreceptor genes (dashed line), and genes involved in cell proliferation (dotted line). The y-axis shows the density of genes from each of the groups at each standard deviation level.

 
After determining that the detected gene expression variations were not random, we measured gene expression variation magnitude across donors. The standard deviation (SD) of the retina/reference sample expression ratios for each gene in each donor, after normalizing with respect to all technical variables, was used for this calculation. While most genes had a relatively small SD, and therefore similar expression levels across donors, some genes showed marked variation among individuals. The mean expression ratio SD for all genes on the array was 0.15±0.8 (log2), and the range was 0.09–0.99 (log2). The distribution of these SDs was skewed towards higher values (Fig. 2C). Table 4 lists the 30 genes with the highest expression variation.


View this table:
[in this window]
[in a new window]
  		Table 4. Top 30 genes by expression ratios standard deviation
 
To explore whether functional classes of genes are associated with differences in expression variation across individuals, we visualized variation by functional group and also used two formal statistical analyses. Figure 2C shows the empirical distributions of the expression ratios SDs for all genes of known function (genes were classified according to the Gene Ontology database; www.geneontology.org). Some classes include an excess of highly variable genes, represented by a right tail above the overall distribution. The evidence is strongest for photoreceptor genes, whose distribution appears bimodal. As the centres of the distributions are similar, testing for differences between these distributions requires an analysis that can capture differences in any quantile of the distribution. To this end, we divided the SD by deciles of the overall distribution, and performed binomial tests for differences in proportions between observed and expected counts in each bin/functional class combination. To account for the multiplicity of tests performed, we chose a significance cut-off based on false discovery rates. Compared to the average, photoreceptor genes showed significantly higher representation in bins of higher SD (P<0.0001) and lower representation in bins of lower SD (P<0.0001); both P-values were with a FDR cut-off level of 1.5%.

Given that known photoreceptor genes tend to be highly expressed, the observed increase in photoreceptor gene variability could have been an artifact if the SD across individuals varied with overall spot fluorescence intensity (which depends on abundance of the transcript in addition to other biological and technical factors). To rule out this possibility, we performed a linear regression analysis using class indicators and average fluorescence intensity as predictors, and SD as response. Of the 20 functional classes evaluated, classes including photoreceptors (P<0.0001) and cell proliferation genes (P=0.019) showed significant effects, indicating that the increased photoreceptor gene variation across individuals is not due to fluorescence intensity alone (Fig. 2C and Supplementary Material). Genes involved in stress response, metabolism, energy, developmental processes and production of extracellular matrix showed borderline P-values of 0.05–0.1 (Supplementary Material). Of these functional classes, only genes involved in proliferation showed lower expression variation compared with the average (Fig. 2C).

In order to help validate the microarray results, the expression variation of nine genes representing different degrees of variation was also measured by quantitative real-time RT–PCR (QPCR) (Table 5 and Fig. 3). The genes demonstrating the least variation had SDs of approximately 0–0.2 by array analysis and 0.6 by QPCR (log2). Above these lower thresholds both methods detected a similar trend of increased expression variation, although the microarray, as previously reported, often tends to underestimate expression levels differences compared to QPCR (33–35) (Fig. 3 and Supplementary Material).


View this table:
[in this window]
[in a new window]
  		Table 5. Primers used for QPCR experiments
 


View larger version (8K):
[in this window]
[in a new window]
  		Figure 3. Comparison of expression variation detected by QPCR and microarray. Nine genes were evaluated by both microarray and QPCR in six donors (see Materials and Methods section for details). The standard deviations of expression ratios for each method are presented. Actual values for each gene are presented in the Supplementary Material.

 
Age-associated gene expression variability
As noted above, in the second regression step ANOVA was used to assess the relative contributions of age, gender, age- by-gender interaction, and death-to-enucleation time interval on gene expression. We also quantified the magnitude of gene expression changes associated with each of the factors using each gene's regression coefficient. To assess the overall contribution of age, gender and age-by-gender interaction, each factor's MSE as well as the factor's MSE/total MSE ratio were calculated (Table 3 and Fig. 4). Age, gender and death-to-enucleation time interval had a similar contribution to the expression variation, while age-by-gender interaction and the residual variation had a smaller contribution. The residual of the second regression step includes genetic, environmental and stochastic factors along with potentially additional unidentified modifiers of gene expression.



View larger version (15K):
[in this window]
[in a new window]
  		Figure 4. Distributions of components of variation attributable to age (solid line), gender (dashed line) and their interaction (dotted line). For each factor, we consider the distribution of the ratios between the MSE attributable to the factor, and the residual MSE. This ratio is expressed in the log2 scale to facilitate visualization. Results are based on the second stage regression, in which the donor-specific gene expression was used as the response variable, and age, gender and their interaction were used as predictors, with adjustment for the death to enucleation time interval effect. All three factors show substantial variability, indicating that some genes are strongly associated with these factors while other are not affected by them. For all factors we can identified a substantial fraction of genes (between 36 and 46%) for which the variation by factor is greater than the residual variation, indicating that the factors are important in affecting expression.

 
Age-associated gene expression can be identified with greater accuracy than gender-related gene expression according to the FDR calculation. Seven-hundred and eighteen genes have 25% or more of their expression attributable to age, and the FDR for this list was 41%. The confidence in identifying genes with ‘true’ age-associated expression increased markedly when genes with higher age-associated ANOVA expression fractions were selected. For example, there are 32 genes with more than 53% of expression contributed by age and these had an FDR of 10% (Table 6)—14 and 18 of these genes showed decreased and increased expression with age, respectively. The mean fold change over 60 years of age for the 14 genes down-regulated with age was 1.4 (range 1.2–2), while the 18 genes up-regulated with age showed a mean fold change of 1.75 (range 1.3–3) for the same time period. Although these 32 age-associated genes belong to different functional classes (Table 6), genes related to transcriptional regulation (five of the 22 genes of known function) and synaptic transmission (also five of 22 genes) are the most prevalent of the group.


View this table:
[in this window]
[in a new window]
  		Table 6. Age-associated genes
 
To assess the relationship between gene function and age on the broader transcriptome, we compared the expression fraction explained by age across all functional groups on the array using ANOVA analysis. The extracellular matrix genes showed a trend towards higher expression levels in older age (average fold change over 60 years was 1.3, P=0.08, F-test). Other functional groups did not show an age-associated expression pattern on average. In particular, none of the genes with larger than 1.5-fold expression change over 60 years of age are known to be selectively expressed in photoreceptors.

To confirm the age-identified correlations, we used QPCR to compare the expression profile of 9 genes in 6 donors to those obtained with the microarrays (Table 5). The methods were consistent regarding the trend of the age-related expression changes, although the magnitude of the effects tended to be larger when measured by QPCR (Fig. 5 and Supplementary Material).



View larger version (9K):
[in this window]
[in a new window]
  		Figure 5. Comparison of age associated regression slopes according to microarrays and QPCR. Nine genes were evaluated by both microarray and QPCR in six donors (see Materials and Methods section for details). The regression slopes, estimating the change in expression associated with a one-year increase in age, are presented for each method. Actual values for each gene are presented in the Supplementary Material.

 
Gender-associated gene expression variability
Six-hundred and eight genes had more than 25% of their expression level variation attributable to gender according to the ANOVA calculation; 296 and 312 of these genes were more highly expressed in females and males, respectively. The average gender-associated fold change for these genes was 1.3±1.2 (range 1.05–3.5 fold; see Supplementary Material for list of the top 30 genes with a gender-associated expression pattern). The FDR of genes with more than 25% of the expression levels explained by gender was 58%. Unlike with the age-associated genes, the specificity did not increase when genes with higher gender-associated ANOVA fractions were selected—the group with a gender-associated expression fraction of more than 50% still had an FDR greater than 50%.

By applying ANOVA analysis to compare gender-associated expression fractions across different functional groups, we found that genes with a vision-related function (almost all of these are selectively expressed in photoreceptors) show a trend toward higher expression levels in males (average fold difference male/female=1.3, P=0.076, F-test). Seven of 32 genes of known function that showed 1.5-fold higher expression in males compared to females were photoreceptor-specific genes, including ABCA4, arrestin, recoverin and rod-specific phosphodiesterase. By contrast, none of the 72 genes expressed 1.5-fold or more in females are known to have a specific vision-related function.

Transporter genes also showed higher average expression levels in males compared with females (average fold difference male/female=1.3, P=0.041, F-test). Other functional groups did not show a gender-associated expression pattern.

We also examined the chromosomal location of the gender-associated genes since one might predict that X chromosome genes located in areas that escape inactivation, other than the pseudoautosomal region (36,37), would be more highly represented in the female enriched gene list compared to the male enriched list. Although the numbers are small, this does indeed seem to be the case, providing support for the overall validity of our assignment of gender-associated genes. Of the 231 female-enriched genes with known chromosomal location and more than 25% of their expression contributed by gender, 13 are X linked; of these, four are known to escape X-inactivation, one is located in the pseudoautosomal region, four are thought to undergo inactivation, and the status of the remaining four is unknown. However, since the X-inactivation status for a particular gene can be tissue-specific, the four genes thought to be inactivated may escape inactivation in the retina (38). Of the equivalent 198 male enriched genes, only five are located on the X chromosome, with one reported to show X-inactivation and the other four of unknown status.

Clustering using age- and gender-associated genes
We used cluster analysis as an alternative algorithm to assess the validity of the second regression step analysis-derived age- and gender-associated gene sets. Values obtained from the first regression step were used as input for both cluster and second step regression analyses. Hierarchical clustering utilizing all 9406 genes from 17 donors (of the 19 donors included in the study, second step ANOVA data were obtained on 17, and data from the two remaining donors were used for the regression steps) did not classify donors according to age or gender (Fig. 6A). When only the age-associated gene set was used for clustering (Table 6), donors were correctly classified into two major age groups with a cut off point at 60 years, except one 58-year-old donor who was classified with the older age group (Fig. 6B). When a similar ANOVA cut-off was used for clustering donors by gender (more than 53% contributed by gender), correct gender classification was obtained (Fig. 6C), indicating consistency between cluster analyses and the second regression step.



View larger version (24K):
[in this window]
[in a new window]
  		Figure 6. Hierarchical clustering of 17 donors using donor-specific effects estimated by the first regression step (see text for details). Clustering was performed using all genes on the array (A), and using only age (B) and gender (C) associated genes (see Results section for details). M=males; F=females; numbers are donors age in years.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
 REFERENCES
 
We have used a custom cDNA microarray to assess the expression level of 9406 genes in 33 normal retinas obtained from 19 human donors. The expression level of most of these genes was similar among the different individuals, but a substantial minority of the genes showed a high degree of variability in expression level. The average SD of expression ratios was 0.15 (log2), with a range of 0.09–0.99 (log2), meaning that for an average gene the value of approximately 95 out of 100 normal individuals will fall within a 1.23-fold range. Certain characteristics correlated with expression variability—a subset of the genes selectively expressed in photoreceptors showed increased variation, while genes involved in cell proliferation tended to show decreased variation. Age and gender also contributed significantly to gene expression variation, with gender effects tending to be small but distributed over a relatively large number of genes.

In order to put our results in context, it is useful to compare them with a variety of recent studies that have also compared expression across normal cells, tissues, and species (2–6,12). However, it should first be noted that quantitative comparison with these other studies is complicated by the fact that the various studies utilized different technologies [e.g. microarray, serial analysis of gene expression (SAGE), or GeneTag], experimental designs and approaches for statistical analysis. Nonetheless, overall there are a number of similarities between the studies, and also some interesting differences.

Oleksiak et al. (4) used custom cDNA microarrays to examine heart gene expression variation in Fundulus (a type of teleost fish), both within and between populations, and employed ANOVA methods to determine the fraction of genes that were statistically significantly differentially expressed between individuals. They reported that 18% of the 907 genes evaluated showed significantly different expression levels within the same population (nominal P-value of 0.01), and the magnitude of the observed changes was generally in the order of 1.5-fold. By comparison, we found that only 252 of the 9406 genes analysed (2.6%) showed significant variation in expression level at an FDR level of 10%. Technical and analytical factors probably play a role in accounting for these different estimates of variation, but differences between fish and human and between heart and retina may also be important. In addition, it should be noted that when more stringent criteria that accounted for multiple testing were employed, the number of differentially expressed genes in the Fundulus study dropped to 4.1%.

Cheung et al. (5) used cDNA microarrays to study gene expression variation in lymphoblastoid cells from normal donors. As their primary measure they calculated a variance ratio, defined as the intra-observer variance divided by the inter-observed variance based on array replicates. For the 813 genes analyzed, this ratio had a median value of 2.5, with a range of 0.4–0.64. A common finding between their study and ours is limited expression variation of genes related to cell proliferation, underscoring the importance of tight transcriptional regulation in maintaining normal cell cycle activity. Cheung et al. (5) also analyzed the genetic contribution to their observed variance, and found that for five highly variable genes the degree of variation was least amongst the genetically most similar individuals (identical twins versus siblings versus unrelated volunteers). It turns out that four of these five genes are also present on our retinal array, but, rather than showing more variation, three of the four showed less variation than average in the retina, suggesting that gene expression variation patterns can be tissue-specific. It would clearly be interesting to determine whether gene expression within the retina is also less amongst genetically similar individuals, but the obvious constraints in obtaining donor eyes make this kind of analysis difficult.

Whitney et al. (6) carried out a similar study of expression variation in normal human peripheral blood mononuclear cells, and also found examples of highly variable genes, and genes whose expression level was highly affected by gender (see below). However, as their method of analysis was largely based on clustering, it is difficult to quantitatively compare their results to ours or those of Cheung et al. (5). Nonetheless, an interesting observation is that, while interferon regulated genes showed high variation in the blood, 22 of the 33 such genes represented on our array showed lower variation than the mean, a finding that again suggests that gene expression variation patterns may be tissue-specific.

The possibility that tissues are characterized by different degrees of expression variation is supported by several recent studies. Pritchard et al. (3), using a 5406-clone spotted cDNA microarray and ANOVA to analyze tissue expression in six individual C57BL/6 mice, reported that 0.8, 1.9 and 3.3% of genes showed significant expression level variation (at the 0.05 level) in liver, testis and kidney, respectively. In another study, Enard et al. (2) used a combination of oligonucleotide- and membrane-based arrays to analyze intra- and inter-species expression differences between primates. Although their tree-based method of analysis makes direct comparison with our results difficult, they observed significant expression variation between humans, and suggested that evolutionary changes in expression pattern are greater in the nervous system than in other tissues. In the Pritchard study, the authors found that expression variation among individuals is different depending upon the tissue examined; the variance of gene expression ratios was 0.054, 0.038 and 0.018 in the testis, liver and kidney, respectively. The finding that these values are in the same range as the variance we observed, which was 0.03, seems counterintuitive in that one might expect to see more variation in an outbred human population than between inbred mice raised in a controlled environment.

GeneTag, a variant of amplified fragment length polymorphism (AFLP) technology, has been applied to assess variation in expression in normal human liver samples (39). The coefficient of variation (CV) was used as the measure of inter-sample variability, and was found to have a mean of 25.5%, with a range from 1% to greater than 100%. Although this study had a limited sample size (triplicate analysis of only four liver profiles), the results are fairly comparable to ours and the other human studies. The value is also similar to that of Hsiao et al. (40), who used oligonucleotide arrays to study a number of different human tissues. Analysis of four kidney samples revealed a mean CV of 31%.

Blackshaw et al. (12) have used microSAGE data to directly compare expression between normal human retina samples (a 41-year-old versus a 44-year-old, and the same 44-year-old versus a 88-year-old). Although the limited number of samples they studied and the correlation coefficient-based statistical analysis that they employed make quantitative comparison of their results with ours difficult, their general conclusion that expression variation of individual genes can be significant and important is consistent with ours.

In order to identify some of the definable determinants of gene expression variation within the human retina, we used our data to estimate the relative contribution of donor age, gender and age-by-gender interaction. The analysis indicated that age and gender have an overall significant and similar contribution to gene expression; this contribution was characterized by consistent small magnitude changes in multiple genes. Again, it is helpful to view these results in the context of other recent studies that have examined the effects of age and gender on gene expression (13–16,18–25,41,42). Bernstein et al. (25) used cDNA membrane arrays to identify age-associated gene expression in the human retina, and reported down-regulation of the heat shock cognate protein 70 in primate retinas. Unfortunately, since HSC70 was not on our cDNA array, we could not directly compare this result. Yoshida et al. (13) applied a commercial cDNA microarrays with 2400 genes to assess age-related expression changes in five human retinas, two of them from young donors (age 13 and 14 years) and three from older donors age 62–72 years. Twenty-four genes were identified as having a differential expression pattern across the age groups. QPCR was used to test nine of these genes in an independent set of eyes, and the results in seven of the nine were interpreted to be consistent with the array results. Our retina custom microarray includes 13 of the 24 genes identified by Yoshida et al., but none of them showed a statistically significant age-related expression pattern in our study. Although the cause(s) for this discrepancy are unknown, a probably important difference between the studies is that Yoshida et al. compared small groups of eyes that differed widely in age (13–14 versus 62–72) whereas we compared a relatively large number of eyes that constituted an older range of ages (29–90). A second potentially important difference between the studies is the method of data analysis—we employed a statistical model that takes into account multiple factors including gene expression variation, while Yoshida and colleagues utilized an expression ratio-based threshold criterion (greater than 2-fold expression difference in at least three of four hybridizations).

Other studies have also looked at aging and gender. In a microarray-based study in which 1- and 6-week-old Drosophila melanogaster were compared, age-associated effects were identified, but they were of limited magnitude (similar to the ones we identified) and were smaller than gender-associated effects (15). Array studies of murine brain have also identified groups of genes that demonstrate an aging effect, and the interesting finding that energy restriction can reduce the magnitude of the changes has been reported (14,23,43). The different studies tend to find different groups of age-associated genes, perhaps because of the small magnitude of most age-related changes, and the different tissues and different organisms studied. It is thus noteworthy that both our study and Jiang et al. (14) identified genes related to neuronal transmission, consistent with known age-related alterations in neuronal function that occur in brain and retina.

Our finding that genes selectively expressed in photoreceptors tend to show high expression variation may be related to their role in vision. Some photoreceptor-specific genes, particularly those encoding genes expressed in the outer segment such as rhodopsin and arrestin, demonstrate a light/dark cycle-dependent expression pattern (44–46), and this may account for some of the variability. However, additional factors are likely to be involved because, for example, the greater variability between genders for photoreceptor-specific genes compared to other genes is not likely to be light/dark cycle related. Whatever the mechanism, it is interesting to speculate that retinal gene expression variation may partially account for the wide range of electrophysiological and psychophysical behaviour observed amongst normal individuals. Given that, at least in transgenic mice, altered expression of wild-type retinal genes can lead to retinal degeneration (47), a speculation of perhaps even greater interest is that differences in the normal pattern of retina gene expression may act as modifiers of susceptibility for retinal disease, for example by modifying individual thresholds for light toxicity.

In conclusion, our study provides quantification of normal gene expression variation in the adult human retina. We demonstrate that expression variation is gene specific and is frequently associated with age and gender and with gene function. Our findings combined with recent reports also suggest that gene expression variation is tissue-specific, and that tissue-specific genes may have a tendency for high expression variation. The data on the range of normal variation should help in the interpretation of ongoing studies directed at identifying gene expression changes associated with retinal disease. Further exploration of the sources and consequences of gene expression variation will hopefully also lead to additional insights into normal retina function and into the mechanisms by which normal expression variation can affect the development of retinal pathology.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
SUPPLEMENTARY MATERIAL
 REFERENCES
 
Donors and tissue processing
Human donor eyes were obtained from the National Disease Research Interchange (NDRI, Philadelphia, PA, USA). Eyes were enucleated 1–12 hours after death (median=4 h), placed on wet ice and shipped to our laboratory. Upon receipt, globes were dissected at the equator, the anterior part was removed, and the retina was dissected after examination under magnification to exclude retinal pathology. RNA was extracted using Trizol (Invitrogen Corporation, Carlsbad, CA, USA) according to the manufacturer protocol, and stored at -80. RNA quantification by spectrophotometer, and assessment of quality on a gel were performed prior to storage.

Microarrays
A human retina custom microarray constructed in our lab in order to reflect the predicted human retina gene expression profile based on ESTs databases was used for this study. A detailed description of array construction, probe labeling technique and hybridization conditions, was recently reported (31). Briefly, PCR products were spotted on SuperAmine coated slides (TeleChem, Sunnyvale, CA, USA) using MicroGrid II (Biorobotics, Cambridge, MA, USA) and GeneMachine (Genomic Solutions, Ann Arbor, MI, USA) arrayers. RNA labeling was performed by incorporating aminoallyl-UTP (Sigma-Aldrich Corp., St Louis, MO, USA) during first-strand cDNA synthesis followed by coupling to mono-reactive cy3 or cy5 fluorescent dye (Amersham Biosciences Inc., Piscataway, NJ, USA). Slide post processing, hybridization, and post hybridization washings were performed according to a previously published protocol (48).

Slides were scanned using the ScanArray 5000 confocal laser scanner (Perkin Elmer, Wellesley, MA, USA). Spot finding and fluorescent intensity quantification were performed using the Imagene software (Biodiscovery Inc., Marina Del Rey, CA, USA). Imagene export file were used for further statistical analyses.

Study design
We employed a common reference sample-based experimental design in which replicates of every donor retina RNA were hybridized in combination with a standard RNA reference. RNA from left and right eyes were kept separate and studied independently. Although other experimental paradigms, like direct comparison ‘loop’ designs, have been suggested to reduce variability and increase information content (49,50), we chose the common reference design because of the large number of samples to be employed, our interest in performing a number of different types of comparisons, and our plans to compare these retinas to additional normal and diseased retinas in the future. Each retina RNA was studied on two microarray slides along with the reference RNA sample, which was composed of 80% cortex RNA, 15% retina RNA and 5% retinal pigment epithelium RNA (total of 66 slides). In the first hybridization of each retina, retina RNA was labeled with cy3 and reference sample RNA with cy5, and on the second hybridization reaction the retina was labeled with cy5 and the reference sample with cy3 (dye swap). This design enabled control for cy3/cy5 related variations as well as artifacts introduced by individual hybridization reactions.

Statistical analysis
Combining data from three different laser power scans.
Commonly, microarrays are scanned using different laser powers and the scan that shows no saturated spots is used for analyses. Based on our experience, while this approach avoids incorporation of data from saturated spots, it leads to loss of useful information from lower-intensity spots, where discrimination of signal from background can be difficult (data not shown). To circumvent this problem we have developed a novel algorithm for combining different laser power scans to maximize data extraction from microarray experiments. Recently, the development of a similar algorithm and the advantage of its use where reported by another group (51).

Following quantification we separated spots into saturated and not saturated using the high power readings. For genes with saturated intermediate and high-power readings, we used the low-power reading for analysis. For genes with saturated high power readings only, we combined low and intermediate power reading for analysis. For genes not saturated at high power, we combined all three readings. For combination, we made use of the fact that the relationship between readings at different power was linear, with the exception of some outliers and the saturated points (data not shown). We therefore used robust linear regression, after exclusion of the saturated points, to predict lower-level reading given higher-level readings. The predicted reading was used for subsequent analysis. For genes that were saturated at high level only, this involves two regressions, the predictions of which were averaged.

Missing data.
We have used the nearest neighbour method to fill in missing data spots (52). Briefly, a matrix containing all data points from the 66 slides used in the study was composed and genes were separated into those with all data points present (complete set) and those with missing data points (incomplete set). A gene from the incomplete set was then picked (target gene), and its distance from all genes in the complete set was calculated. The first five genes with closest distance to it were identified and defined as its closest neighbourhood. The average of these five nearest neighbourhood genes readings was used to fill in the target gene's missing data. Genes with more than 25% missing data points were excluded from further analyses.

Two-step regression model.
To normalize expression ratios we first inspected MVA plots for each slide and then applied loess smoothing. The residual between each original value and the loess smooth was used for further analyses (53). Throughout, gene expression was measured by the normalized log ratio of signal in the retina channel to signal in the reference channel. Background was not subtracted because we found that in our data set background subtraction introduced significant noise at the low intensity range (data not shown). However, to avoid introduction of artefacts to the data matrix, we omitted all spots in which the background level was higher than the spot signal itself. As described above in the Results section, we employed a two-step regression model for analysis. This approach implements, albeit approximately, a multilevel, or hierarchical model (54), since standard approaches to this type of model are computationally prohibitive in the situation at hand due to the high dimensionality of the data and the complexity of the interactions investigated.

In the first regression step we assessed and corrected for the effects of technical variables, including replication, dye assignment and the use of different microarray batches. We also assessed the contribution of eye laterality (right versus left eye) to gene expression variation in this first step in order measure the contribution of this factor before combining data from both eyes in the second regression. Combining data from both eyes allowed for further control on technical variables, as most donors in the study had data from four replicate arrays (two from each eye) and as we were primarily interested in detecting variation across individuals.

In this first step, each gene was considered separately. We fit a linear regression model with gene expression as the response, and dye effect, microarray lot effect, eye laterality and donor effect as covariates. Adjustments for the technical effects were made and all slides from one donor were combined into a single gene expression value. The regression coefficient of each donor (donor effect) and the corresponding standard error were then used as input for the second regression step. In this second step weighted linear regression was used with donor effect as response, and age, gender, age-by-gender interaction, and death-to-enucleation time interval as covariates. The donor effect was weighted by the reciprocal of its variance from the first step.

In the second regression step, we assessed the association of gene expression with age, gender and age-by-gender interaction. First, we used analysis of variance (ANOVA) to separate gene expression variation across the donors into five parts: variation explained by age, gender, age-by-gender interaction, death-to-enucleation time interval, and residual (the variation that could not be explained by any of the other listed sources) (55). The association of gene expression with different biological effects was then measured by the variation that could be explained by each effect. Then, we used each gene's regression coefficient for each of these factors to measure gene expression changes that can be attributed to the corresponding factor.

The overall significance of the associations identified between gene expression patterns and biological effects was assessed by calculating the FDR at different cut-off levels (32). To generate the null distribution, donor IDs were permuted and the regression analysis was repeated. The false discovery rate for a certain cut-off level was approximated by the upper bound given by the number of false positive genes (genes identified after donor's ID permutation) to observed genes ratio at a particular cut-off level.

All analyses were performed using the R package (56).

Quantitative real time RT–PCR
cDNAs were synthesized from 1ug (mixture of 0.5 µg each from right and left eyes) of DNase-treated retinal RNA from six donors (donor numbers 2, 8, 9, 14, 16 and 17 from Table 1) using Superscript II (Invitrogen Corporation, Carlsbad, CA, USA). Primers are listed in Table 5. QPCR reactions were performed using a Light-Cycler (Roche, Nutley, NJ, USA). PCR products were quantified using the second derivate maximum values calculated by the Light-Cycler software. Expression levels were normalized to the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA level. Each gene was tested in duplicate, and the mean of the two reactions was used for calculating the expression levels.


   SUPPLEMENTARY MATERIAL
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
REFERENCES
 
Supplementary Material is available at HMG Online.


   ACKNOWLEDGEMENTS
 
The authors wish to thank Drs Jeremy Nathans and Amir Rattner for providing one of the retina cDNA libraries that was used for the construction of the microarray. This study was supported in part by grants from the National Eye Institute, Macula Vision Foundation, Fight For Sight, and Santen Pharmaceutical, and by generous gifts from Mr and Mrs Marshall and Stevie Wishnack and from Mr and Mrs Robert and Clarice Smith. S.L.B. is funded by the V. Kann Rasmussen Foundation (Denmark), a Career Development award from the Research to Prevent Blindness, and a macular degeneration research grant from the American Health Assistance Foundation (AHAF). P.A.C. is the George S. and Dolores Dore Eccles Professor of Ophthalmology and Neuroscience; D.J.Z. is the Guerrieri Professor of Genetic Engineering and Molecular Ophthalmology, and the work was performed in the Guerrieri Center.


   FOOTNOTES
 
* To whom correspondence should be addressed at: 809 Maumenee Bldg, Johns Hopkins University School of Medicine, 600 N. Wolfe Street, Baltimore, MD 21287, USA. Tel: +1 4105025230; Fax: +1 4105025382; Email: dzack{at}bs.jhmi.edu Back


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 RESULTS
 DISCUSSION
 MATERIALS AND METHODS
 SUPPLEMENTARY MATERIAL
 REFERENCES
 

  1. King, M.C. and Wilson, A.C. (1975) Evolution at two levels in humans and chimpanzees. Science, 188, 107–116.[Free Full Text]

  2. Enard, W., Khaitovich, P., Klose, J., Zollner, S., Heissig, F., Giavalisco, P., Nieselt-Struwe, K., Muchmore, E., Varki, A., Ravid, R. et al. (2002) Intra- and interspecific variation in primate gene expression patterns. Science, 296, 340–343.[Abstract/Free Full Text]

  3. Pritchard, C.C., Hsu, L., Delrow, J. and Nelson, P.S. (2001) Project normal: defining normal variance in mouse gene expression. Proc. Natl Acad. Sci. USA, 98, 13266–13271.[Abstract/Free Full Text]

  4. Oleksiak, M.F., Churchill, G.A. and Crawford, D.L. (2002) Variation in gene expression within and among natural populations. Nat. Genet., 32, 261–266.[CrossRef][Web of Science][Medline]

  5. Cheung, V.G., Conlin, L.K., Weber, T.M., Arcaro, M., Jen, K.Y., Morley, M. and Spielman, R.S. (2003) Natural variation in human gene expression assessed in lymphoblastoid cells. Nat. Genet., 33, 422–425.[CrossRef][Web of Science][Medline]

  6. Whitney, A.R., Diehn, M., Popper, S.J., Alizadeh, A.A., Boldrick, J.C., Relman, D.A. and Brown, P.O. (2003) Individuality and variation in gene expression patterns in human blood. Proc. Natl Acad. Sci. USA, 100, 1896–1901.[Abstract/Free Full Text]

  7. Brem, R.B., Yvert, G., Clinton, R. and Kruglyak, L. (2002) Genetic dissection of transcriptional regulation in budding yeast. Science, 296, 752–755.[Abstract/Free Full Text]

  8. Cowles, C.R., Joel, N.H., Altshuler, D. and Lander, E.S. (2002) Detection of regulatory variation in mouse genes. Nat. Genet., 32, 432–437.[CrossRef][Web of Science][Medline]

  9. Hall, D., Ybazeta, G., Destro-Bisol, G., Petzl-Erler, M.L. and Di Rienzo, A. (1999) Variability at the uridine diphosphate glucuronosyltransferase 1A1 promoter in human populations and primates. Pharmacogenetics, 9, 591–599.[Web of Science][Medline]

  10. Blake, W.J., M, K.A., Cantor, C.R. and Collins, J.J. (2003) Noise in eukaryotic gene expression. Nature, 422, 633–637.[CrossRef][Medline]

  11. Chung, C.H., Bernard, P.S. and Perou, C.M. (2002) Molecular portraits and the family tree of cancer. Nat. Genet., 32 (suppl.), 533–540.[Medline]

  12. Blackshaw, S., Kuo, W.P., Park, P.J., Tsujikawa, M., Gunnersen, J.M., Scott, H.S., Boon, W.M., Tan, S.S. and Cepko, C.L. (2003) MicroSAGE is highly representative and reproducible but reveals major differences in gene expression among samples obtained from similar tissues. Genome Biol., 4, R17.[CrossRef][Medline]

  13. Yoshida, S., Yashar, B.M., Hiriyanna, S. and Swaroop, A. (2002) Microarray analysis of gene expression in the aging human retina. Invest. Ophthal. Visual Sci., 43, 2554–2560.[Abstract/Free Full Text]

  14. Jiang, C.H., Tsien, J.Z., Schultz, P.G. and Hu, Y. (2001) The effects of aging on gene expression in the hypothalamus and cortex of mice. Proc. Natl Acad. Sci. USA, 98, 1930–1934.[Abstract/Free Full Text]

  15. Jin, W., Riley, R.M., Wolfinger, R.D., White, K.P., Passador-Gurgel, G. and Gibson, G. (2001) The contributions of sex, genotype and age to transcriptional variance in Drosophila melanogaster. Nat. Genet., 29, 389–395.[CrossRef][Web of Science][Medline]

  16. Jiang, M., Ryu, J., Kiraly, M., Duke, K., Reinke, V. and Kim, S.K. (2001) Genome-wide analysis of developmental and sex-regulated gene expression profiles in Caenorhabditis elegans. Proc. Natl Acad. Sci. USA, 98, 218–223.[Abstract/Free Full Text]

  17. Han, E., Hilsenbeck, S.G., Richardson, A. and Nelson, J.F. (2000) cDNA expression arrays reveal incomplete reversal of age-related changes in gene expression by calorie restriction. Mech. Ageing Dev., 115, 157–174.[CrossRef][Web of Science][Medline]

  18. Roth, S.M., Ferrell, R.E., Peters, D.G., Metter, E.J., Hurley, B.F. and Rogers, M.A. (2002) Influence of age, sex, and strength training on human muscle gene expression determined by microarray. Physiol. Genomics, 10, 181–190.[Abstract/Free Full Text]

  19. Weindruch, R., Kayo, T., Lee, C.K. and Prolla, T.A. (2001) Microarray profiling of gene expression in aging and its alteration by caloric restriction in mice. J. Nutr., 131, 918S–923S.[Abstract/Free Full Text]

  20. Weindruch, R., Kayo, T., Lee, C.K. and Prolla, T.A. (2002) Gene expression profiling of aging using DNA microarrays. Mech. Ageing Dev., 123, 177–193.[CrossRef][Web of Science][Medline]

  21. Welle, S., Brooks, A. and Thornton, C.A. (2001) Senescence-related changes in gene expression in muscle: similarities and differences between mice and men. Physiol. Genomics, 5, 67–73.[Abstract/Free Full Text]

  22. Shelton, D.N., Chang, E., Whittier, P.S., Choi, D. and Funk, W.D. (1999) Microarray analysis of replicative senescence. Curr. Biol., 9, 939–945.[CrossRef][Web of Science][Medline]

  23. Prolla, T.A. (2002) DNA microarray analysis of the aging brain. Chem. Senses, 27, 299–306.[Abstract/Free Full Text]

  24. Cristofalo, V.J. (2000) A DNA chip off the aging block. Nat. Med., 6, 507.[CrossRef][Web of Science][Medline]

  25. Bernstein, S.L., Liu, A.M., Hansen, B.C. and Somiari, R.I. (2000) Heat shock cognate-70 gene expression declines during normal aging of the primate retina. Invest. Ophthal. Visual Sci., 41, 2857–2862.[Abstract/Free Full Text]

  26. Ogueta, S.B., Schwartz, S.D., Yamashita, C.K. and Farber, D.B. (1999) Estrogen receptor in the human eye: influence of gender and age on gene expression. Invest. Ophthal. Visual Sci., 40, 1906–1911.[Abstract/Free Full Text]

  27. Yoo, S.H. and Adamis, A.P. (1998) Retinal manifestations of aging. Int. Ophthal. Clin., 38, 95–101.[Medline]

  28. Friedman, D.S., Katz, J., Bressler, N.M., Rahmani, B. and Tielsch, J.M. (1999) Racial differences in the prevalence of age-related macular degeneration: the Baltimore Eye Survey. Ophthalmology, 106, 1049–1055.[CrossRef][Web of Science][Medline]

  29. Jackson, G.R., Owsley, C. and Curcio, C.A. (2002) Photoreceptor degeneration and dysfunction in aging and age-related maculopathy. Ageing Res. Rev., 1, 381–396.[CrossRef][Web of Science][Medline]

  30. Smith, W., Assink, J., Klein, R., Mitchell, P., Klaver, C.C., Klein, B.E., Hofman, A., Jensen, S., Wang, J.J. and de Jong, P.T. (2001) Risk factorsfor age-related macular degeneration: Pooled findings from three continents. Ophthalmology, 108, 697–704.[CrossRef][Web of Science][Medline]

  31. Chowers, I., Gunatilaka, T., Farkas, R., Qian, J., Hackam, A., Duh, E., Kageyama, M., Wang, C., Vora, A., Campochiaro, P. et al. (2003) Identification of novel genes preferentially expressed in the retina using a custom human retina cDNA microarray. Invest. Ophthal. Visual Sci., 44, 3732–3741.[Abstract/Free Full Text]

  32. Tusher, V.G., Tibshirani, R. and Chu, G. (2001) Significance analysis of microarrays applied to the ionizing radiation response. Proc. Natl Acad. Sci. USA, 98, 5116–5121.[Abstract/Free Full Text]

  33. Ramakrishnan, R., Dorris, D., Lublinsky, A., Nguyen, A., Domanus, M., Prokhorova, A., Gieser, L., Touma, E., Lockner, R., Tata, M. et al. (2002) An assessment of Motorola CodeLink microarray performance for gene expression profiling applications. Nucl. Acids Res., 30, e30.[Abstract/Free Full Text]

  34. Yuen, T., Wurmbach, E., Pfeffer, R.L., Ebersole, B.J. and Sealfon, S.C. (2002) Accuracy and calibration of commercial oligonucleotide and custom cDNA microarrays. Nucl. Acids Res., 30, e48.[Abstract/Free Full Text]

  35. Livesey, F.J., Furukawa, T., Steffen, M.A., Church, G.M. and Cepko, C.L. (2000) Microarray analysis of the transcriptional network controlled by the photoreceptor homeobox gene Crx. Curr. Biol., 10, 301–310.[CrossRef][Web of Science][Medline]

  36. Carrel, L., Cottle, A.A., Goglin, K.C. and Willard, H.F. (1999) A first-generation X-inactivation profile of the human X chromosome. Proc. Natl Acad. Sci. USA, 96, 14440–14444.[Abstract/Free Full Text]

  37. Sudbrak, R., Wieczorek, G., Nuber, U.A., Mann, W., Kirchner, R., Erdogan, F., Brown, C.J., Wohrle, D., Sterk, P., Kalscheuer, V.M. et al. (2001) X chromosome-specific cDNA arrays: identification of genes that escape from X-inactivation and other applications. Hum. Mol. Genet., 10, 77–83.[Abstract/Free Full Text]

  38. Carrel, L., Hunt, P.A. and Willard, H.F. (1996) Tissue and lineage-specific variation in inactive X chromosome expression of the murine Smcx gene. Hum. Mol. Genet., 5, 1361–1366.[Abstract/Free Full Text]

  39. Wong, L.Y., Hafeman, A., Boyd, V.L., Bodeau, J., Lazaruk, K.D., Liew, S.N., Casey, P., Belonogoff, V., Bit, S., Sumner, C. et al. (2003) Assessing gene expression variation in normal human tissues using GeneTag, a novel, global, sensitive profiling method. J. Biotechnol., 101, 199–217.[CrossRef][Web of Science][Medline]

  40. Hsiao, L.L., Dangond, F., Yoshida, T., Hong, R., Jensen, R.V., Misra, J., Dillon, W., Lee, K.F., Clark, K.E., Haverty, P. et al. (2001) A compendium of gene expression in normal human tissues. Physiol. Genom., 7, 97–104.[Abstract/Free Full Text]

  41. Han, E. and Hilsenbeck, S.G. (2001) Array-based gene expression profiling to study aging. Mech. Ageing Dev., 122, 999–1018.[CrossRef][Web of Science][Medline]

  42. Welle, S. (2002) Gene transcript profiling in aging research. Exp. Gerontol., 37, 583–590.[CrossRef][Web of Science][Medline]

  43. Lee, C.K., Weindruch, R. and Prolla, T.A. (2000) Gene-expression profile of the ageing brain in mice. Nat. Genet., 25, 294–297.[CrossRef][Web of Science][Medline]

  44. Korenbrot, J.I. and Fernald, R.D. (1989) Circadian rhythm and light regulate opsin mRNA in rod photoreceptors. Nature, 337, 454–457.[CrossRef][Medline]

  45. Farber, D.B., Danciger, J.S. and Organisciak, D.T. (1991) Levels of mRNA encoding proteins of the cGMP cascade as a function of light environment. Exp. Eye Res., 53, 781–786.[CrossRef][Web of Science][Medline]

  46. McGinnis, J.F., Whelan, J.P. and Donoso, L.A. (1992) Transient, cyclic changes in mouse visual cell gene products during the light-dark cycle. J. Neurosci. Res., 31, 584–590.[CrossRef][Web of Science][Medline]

  47. Olsson, J.E., Gordon, J.W., Pawlyk, B.S., Roof, D., Hayes, A., Molday, R.S., Mukai, S., Cowley, G.S., Berson, E.L. and Dryja, T.P. (1992) Transgenic mice with a rhodopsin mutation (Pro23His): a mouse model of autosomal dominant retinitis pigmentosa. Neuron, 9, 815–830.[CrossRef][Web of Science][Medline]

  48. Hegde, P., Qi, R., Abernathy, K., Gay, C., Dharap, S., Gaspard, R., Hughes, J.E., Snesrud, E., Lee, N. and Quackenbush, J. (2000) A concise guide to cDNA microarray analysis. Biotechniques, 29, 548–550, 552–554, 556 passim.[Web of Science][Medline]

  49. Kerr, M.K. and Churchill, G.A. (2001) Statistical design and the analysis of gene expression microarray data. Genet. Res., 77, 123–128.[CrossRef][Web of Science][Medline]

  50. Churchill, G.A. (2002) Fundamentals of experimental design for cDNA microarrays. Nat. Genet., 32 (suppl.), 490–495.[CrossRef][Medline]

  51. Dudley, A.M., Aach, J., Steffen, M.A. and Church, G.M. (2002) Measuring absolute expression with microarrays with a calibrated reference sample and an extended signal intensity range. Proc. Natl Acad. Sci. USA, 99, 7554–7559.[Abstract/Free Full Text]

  52. Hastie, T., Tibshirani, R., Sherlock, G., Eisen, M., Brown, P. and Botstein, D. (1999) Imputing missing data for gene expression arrays. Stanford University Technical Report.

  53. Yang, Y.H., Dudoit, S., Luu, P., Lin, D.M., Peng, V., Ngai, J. and Speed, T.P. (2002) Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucl. Acids Res., 30, e15.[Abstract/Free Full Text]

  54. Carlin, B. and Louis, T. (2000) Bayes and Empirical Bayes Methods for Data Analysis. 2nd edn. Chapman and Hall/CRC Press, Boca Raton, FL.

  55. Kerr, M.K., Martin, M. and Churchill, G.A. (2000) Analysis of variance for gene expression microarray data. J. Comput. Biol., 7, 819–837.[CrossRef][Web of Science][Medline]

  56. Ihaka, R. and Gentleman, R. (1996) A language for data analysis and graphics. J. Comput. Graph. Stat., 5, 299–314.[CrossRef]


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 Toxicol SciHome page
Z. Li, F. A. Wright, and J. Royland
Age-Dependent Variability in Gene Expression in Male Fischer 344 Rat Retina
Toxicol. Sci., January 1, 2009; 107(1): 281 - 292.
[Abstract] [Full Text] [PDF]

Home page
 IOVSHome page
R. Ganti, R. C. Hunt, S. K. Parapuram, and D. M. Hunt
Vitreous Modulation of Gene Expression in Low-Passage Human Retinal Pigment Epithelial Cells
Invest. Ophthalmol. Vis. Sci., April 1, 2007; 48(4): 1853 - 1863.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Respir. Cell Mol. Bio.Home page
M. P. Gruber, C. D. Coldren, M. D. Woolum, G. P. Cosgrove, C. Zeng, A. E. Baron, M. D. Moore, C. D. Cool, G. S. Worthen, K. K. Brown, et al.
Human Lung Project: Evaluating Variance of Gene Expression in the Human Lung
Am. J. Respir. Cell Mol. Biol., July 1, 2006; 35(1): 65 - 71.
[Abstract] [Full Text] [PDF]

Home page
 IOVSHome page
C. Bowes Rickman, J. N. Ebright, Z. J. Zavodni, L. Yu, T. Wang, S. P. Daiger, G. Wistow, K. Boon, and M. A. Hauser
Defining the Human Macula Transcriptome and Candidate Retinal Disease Genes Using EyeSAGE
Invest. Ophthalmol. Vis. Sci., June 1, 2006; 47(6): 2305 - 2316.
[Abstract] [Full Text] [PDF]

Home page
 IOVSHome page
I. Chowers, R. Wong, T. Dentchev, R. H. Farkas, J. Iacovelli, T. L. Gunatilaka, N. E. Medeiros, J. B. Presley, P. A. Campochiaro, C. A. Curcio, et al.
The Iron Carrier Transferrin Is Upregulated in Retinas from Patients with Age-Related Macular Degeneration
Invest. Ophthalmol. Vis. Sci., May 1, 2006; 47(5): 2135 - 2140.
[Abstract] [Full Text] [PDF]

Home page
 IOVSHome page
M. Saghizadeh, A. A. Kramerov, J. Tajbakhsh, A. M. Aoki, C. Wang, N.-N. Chai, J. Y. Ljubimova, T. Sasaki, G. Sosne, M. R. J. Carlson, et al.
Proteinase and Growth Factor Alterations Revealed by Gene Microarray Analysis of Human Diabetic Corneas
Invest. Ophthalmol. Vis. Sci., October 1, 2005; 46(10): 3604 - 3615.
[Abstract] [Full Text] [PDF]

Home page
 Hum Mol GenetHome page
J. Liu, Q. Huang, J. Higdon, W. Liu, T. Xie, T. Yamashita, K. Cheon, C. Cheng, and J. Zuo
Distinct gene expression profiles and reduced JNK signaling in retinitis pigmentosa caused by RP1 mutations
Hum. Mol. Genet., October 1, 2005; 14(19): 2945 - 2958.
[Abstract] [Full Text] [PDF]

Home page
 IOVSHome page
S. Zareparsi, A. Hero, D. J. Zack, R. W. Williams, and A. Swaroop
Seeing the Unseen: Microarray-Based Gene Expression Profiling in Vision
Invest. Ophthalmol. Vis. Sci., August 1, 2004; 45(8): 2457 - 2462.
[Full Text] [PDF]

Home page
 IOVSHome page
R. H. Farkas, J. Qian, J. L. Goldberg, H. A. Quigley, and D. J. Zack
Gene Expression Profiling of Purified Rat Retinal Ganglion Cells
Invest. Ophthalmol. Vis. Sci., August 1, 2004; 45(8): 2503 - 2513.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Supplementary Material
Right arrow All Versions of this Article:
12/22/2881    most recent
ddg326v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (23)
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Chowers, I.
Right arrow Articles by Zack, D. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Chowers, I.
Right arrow Articles by Zack, D. J.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?