Vitreous Modulation of Gene Expression in Low-Passage Human Retinal Pigment Epithelial Cells

Abstract

purpose. In proliferative vitreoretinopathy (PVR), retinal pigment epithelial (RPE) cells enter the vitreous and proliferate. They become fibroblast-like and participate in the formation of contractile membranes, which can lead to retinal detachment. Vitreous treatment of RPE cells in vitro results in similar morphologic changes. This study was conducted to examine vitreous-induced modulation of gene expression in RPE cells.

methods. Low-passage human RPE cell lines derived from three donors were each treated for 6, 12, 24, or 48 hours with complete medium or complete medium containing 25% vitreous. Changes in mRNA levels were examined by using microarrays. Real-time quantitative PCR (qPCR) was used to measure mRNA expression of a subset of genes in cells from three additional donors. Immunohistochemistry and immunoblot analysis were used to examine protein expression.

results. Vitreous treatment caused a progressive reprogramming of gene expression. qPCR confirmed vitreous modulation of mRNA levels of 10 of 10 genes. Changes consistent with a transition from an epithelial to a mesenchymal phenotype were observed. Downregulated genes included genes associated with differentiated RPE cells. Upregulated genes included genes associated with stress and inflammation. Pathway analysis indicated that the transforming growth factor-β/bone morphogenetic protein (BMP) pathway and the focal adhesion pathway may play a role in this process. BMP-2 protein and mRNA were increased.

conclusions. Despite the biological variation in vitreous and RPE donors, vitreous reproducibly modulated a limited number of mRNAs. Many of these changes were consistent with the more fibroblast-like appearance of vitreous-treated cells and with the pathobiology of PVR. TGF-β and BMP-2 may be important modulators of vitreous-induced changes in gene expression.

Retinal pigment epithelial (RPE) cells form a barrier epithelium between the neural retina and the choriocapillaris at the back of the eye. They play an important role in the blood–retinal barrier and control access of nutrients from the choroidal blood supply to the outer retina. RPE cells have many other functions, including reduction of light scattering, participation in the visual cycle, the removal of waste products (including shed outer segment membranes), and the maintenance of the interphotoreceptor ionic environment. ¹ RPE cells do not usually divide in adult life, but in certain diseases, such as proliferative vitreoretinopathy (PVR), they can dissociate from the Bruch’s membrane and proliferate in the vitreous or in the subretinal space. ^{2 3 4} In PVR, RPE cells are found in epiretinal membranes, which contain various cell types embedded in an extracellular matrix distinct from that of the vitreous. In what can be regarded as an aberrant wound-healing response, ^{2 3} these membranes become contractile and cause traction retinal detachment and loss of sight. RPE cells are thought to play a critical role in the formation and contraction of these membranes. ⁵ The RPE cells in PVR membranes have often undergone morphologic changes and appear more fibroblast-like than epithelial. ⁴ When exposed to vitreous in vivo ⁶ or in vitro, ⁷ they undergo similar morphologic changes. In vitro, this epithelial-mesenchymal transformation (EMT)–like process requires the presence of serum factors as well as vitreous. ⁸  

Risk factors for PVR include the presence of RPE cells in the vitreous, a tear in the neural retina (which could expose the RPE layer to vitreous components), a breakdown in the blood–retinal barrier and the presence of inflammatory components. ^{3 9 10 11} Knowledge of which mRNAs are modulated by vitreous treatment of RPE cells will help elucidate the early stages of PVR. Thus, differential display polymerase chain reaction (PCR)–based methods ^{12 13} and membrane arrays representing 588 genes ¹⁴ have been used, but these studies identified only a limited number of gene changes at a single time point. We report here the use of 21,000-gene microarrays to profile vitreous-induced changes in gene expression in low-passage human RPE cells from multiple donors at various times of treatment. Since vitreous-induced transformation of RPE cells requires serum, ⁸ all these experiments were performed in the presence of serum. Vitreous caused a progressive change in the transcriptional program of these cells. There was a downregulation of genes associated with tight junctions and a change in expression of focal adhesion genes, as would be expected if epithelial cells lost cell–cell contact, became migratory, and transformed into more fibroblast-like cells. Many of the modulated genes were associated with the transforming growth factor-β (TGFβ)/bone morphogenetic protein (BMP) pathway. 

Methods

RPE Cells

RPE cells were obtained postmortem from human donor eyes (Lifepoint, Charleston, SC, and Lions’ Eye Bank, Portland, OR). The protocol adhered to the tenets of the Declaration of Helsinki for research involving human tissue. Eyes were not used if there was a known history of retinal disease or diabetes. RPE cells were isolated as described previously ¹⁵ and cultured in F-10 medium (Invitrogen Corp., Carlsbad, CA) containing 10% fetal bovine serum (Invitrogen Corp.), 1% penicillin-streptomycin-glutamine (Invitrogen Corp.), 1 μM CaCl₂, and 1% ITS (BD Biosciences, Franklin Lakes, CA). The epithelioid nature of the RPE cells was checked by immunohistochemistry using an anti-cytokeratin primary antibody and a fluorescent secondary antibody. ¹⁶ RPE cells were used between passages 3 and 6. 

Vitreous Treatment of RPE Cells

Vitreous gel from human donor eyes was shredded by using a syringe, diluted with three volumes of complete medium and was vacuum-filtered through a 0.22-μm polyethersulfone filter (Corning Inc., Corning, NY). ¹⁵ Control medium was also filtered. At the 0 time point, the medium was removed from subconfluent RPE cells and replaced with complete, serum-containing medium or complete, serum-containing medium containing 25% vitreous. Both control and vitreous-treated cells were then incubated at 37°C for the same time (6, 12, 24, or 48 hours) before RNA extraction. Cells were still subconfluent at the time RNA was extracted. 

RNA Extraction

The medium was quickly removed from the cells, which were briefly washed with PBS. RNA was extracted (RNeasy; Qiagen Inc., Valencia, CA). Column-bound RNA was treated with DNase (Qiagen, Inc.), according to the manufacturer’s directions, except that the concentration and time for DNase treatment were doubled. The amount and purity of the total RNA were determined from the absorbance at 260 and 280 nm and by nondenaturing gel electrophoresis. ¹⁵  

RNA Amplification and Hybridization to Microarrays

RNA extracted from each of three low-passage RPE donor–vitreous donor pairs at 6, 12, 24, or 48 hours of control treatment or vitreous treatment was copied into cDNA and subjected to linear amplification using an aRNA amplification kit (Amino Allyl MessageAmp; Ambion Inc., Austin, TX). The amino allyl-labeled RNA was fluorescently labeled with cyanine-3 or -5 using reactive dye packs (CyDye Post-Labeling; GE Healthcare, Piscataway, NJ) and purified on GFX columns (GE Healthcare). Fluorescently labeled RNA from the vitreous-treated and the corresponding control cells, each labeled with a different cyanine dye, were mixed and hybridized to prehybridized 70-mer oligonucleotide microarray slides (Operon Human Genome Oligo Set ver. 2, 21,318 genes; University of Cincinnati Genomics and Microarray Laboratory, Cincinnati, OH) according to the protocols of the W. M. Keck Foundation Biotechnology Resource Laboratory (New Haven, CT). All samples were analyzed twice, with the dyes reversed on the second analysis, giving a total of 24 dual-color microarray slides. Slides were scanned at the University of South Carolina Microarray Facility (ScanArray 5000; PerkinElmer Life and Analytical Sciences, Wellesley, MA) or the University of Cincinnati Genomics and Microarray Laboratory (GenePix 4000B; Molecular Devices, Sunnyvale, CA). 

In all cases, RNA from vitreous-treated cells from a particular donor was compared with RNA from untreated cells from the same donor, which were grown at the same time and incubated for the same period after medium change. Thus, genes that varied in expression from donor to donor but which were not modulated by vitreous should not be detected using this approach. 

Analysis of Gene Expression

Microarray analysis software (GeneSpring; Agilent Technologies, Palo Alto, CA) was used for the initial analysis of the digitized data and to correct for background. Gene expression was normalized using the Lowess option and the ratio of expression in vitreous:control cells determined for each gene (see Appendix A). The resultant data were analyzed using the Significance Analysis of Microarrays (SAM) statistics program to determine genes that were significantly regulated at each time point. ¹⁷ SAM uses a permutation method to determine the significance of the results and hence reports a false-discovery rate (FDR) rather than a probability. In addition, genes that gave signals at least 10 standard deviations above background (microarray scanner fluorescence values of >300 for ScanArray data [Perkin-Elmer] and >100 for GenePix data [Molecular Devices]) for both control and vitreous channels and that were regulated by at least twofold at any time point in all three donors were determined using the microarray analysis software (GeneSpring; Agilent Technologies). SOURCE ¹⁸ was used to annotate genes and to convert accession numbers to Entrez Gene IDs/Locus Link IDs or HGNC (Human Genome Nomenclature Committee) gene symbols. EASE ¹⁹ software was used to remove duplicated genes. Pathways were examined using the WebGestalt-Gene Set Analysis Toolkit ²⁰ programs to search the KEGG database. ²¹ Statistical significance was determined using the hypergeometric test option. The Gene Ontology Database ²² was searched with DAVID ²³ , EASE, Ontoexpress, ²⁴ and GOTM. ²⁰  

Reverse Transcription and Quantitative Real-Time PCR

Total RNA (1 μg) was reverse transcribed into cDNA (Iscript reagents; Bio-Rad Laboratories, Hercules, CA) and simultaneous priming from both oligo(dT) and random hexamers. The cDNA mix was diluted to 300 μL with glass-distilled water. QPCR was performed (iQ SYBR Green Supermix; Bio-Rad Laboratories, Hercules, CA) with 5 μL of the diluted DNA and 2.5 picomoles of each primer in a total volume of 25 μL. Reactions were run on a PCR system (iCycler IQ Real Time PCR; Bio-Rad Laboratories). Primer pairs (Table 1)were designed on computer (Oligo 6; Molecular Biology Insights, Cascade, CO). The specificity of all primer pairs was checked by direct sequencing of the PCR product. For validation experiments, five replicates were performed for both the target gene and the internal standard gene (ribosomal protein, large, P0; RPLP0) used for normalization. The relative expression of the target gene in RNA from vitreous-treated cells compared to the matched control cell RNA was determined using the Pfaffl analysis method ²⁵ which corrects for any differences in the expression of the internal normalization gene (RPLP0; internal loading control) as well as for any differences in the efficiency of amplification for the primer pairs for each gene (Table 1) . Statistical significance was determined using REST-XL software. ²⁶ Statistical analysis with this program in the non-normalized mode showed that vitreous treatment had no significant effect on the expression of the internal standard gene (RPLP0). 

Immunohistochemistry

RPE cells were grown for 48 hours in control medium or medium containing 25% vitreous. To inhibit secretion of extracellular proteins, Brefeldin A (Sigma-Aldrich, St. Louis, MO) was then added to the medium to a final concentration of 5 ng/mL, and the incubation was continued for a further 6 hours. The cells were fixed in 4% paraformaldehyde, permeabilized with 0.5% Triton X-100, and stained with rabbit anti-human tissue plasminogen activator antibody (Santa Cruz Biotechnology Inc., Santa Cruz, CA) or with rabbit anti-human BMP-2 antibody (Aviva Systems Biology, San Diego, CA) followed by Cy5-conjugated anti-rabbit IgG (Invitrogen). The cells were double stained with Texas Red phalloidin (Invitrogen) to detect actin microfilaments and were observed by confocal microscopy (Meta500; Carl Zeiss Meditec, Inc., Dublin, CA). 

Immunoblot Analysis

The cells were extracted with ice-cold extraction buffer (1% NP40; Sigma-Aldrich) in 50 mM Tris-HCl (pH 8.0) containing 340 mM NaCl, 1 mM phenylmethylsulfonyl fluoride (PMSF), 50 mM NaF and protease cocktail inhibitor (20 μL/mL; Pierce Biotechnology Inc., Rockford, IL). The extract was centrifuged and dissolved for SDS-polyacrylamide gel electrophoresis. After electrophoresis, the proteins were transferred to polyvinylidene difluoride (PVDF) membrane (Bio-Rad Laboratories) and protein-binding sites were blocked overnight with 5% nonfat dried milk in blocking buffer (20 mM Tris-HCl [pH 7.6], 0.8 M NaCl, and 0.1% Tween-20). Membranes were incubated overnight at 4°C with the appropriate primary antibody diluted in blocking buffer. The primary antibodies used were rabbit anti-human TFPI2 (kindly given by Walter Kisiel, Department of Pathology, University of New Mexico Health Sciences Center, Albuquerque, NM; diluted 1:1000), or rabbit anti-human BMP-2 antibody (diluted 1:100; Aviva Systems Biology). After the membrane was washed, it was incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG (GE Healthcare). Chemiluminescence was used for detection (West Pico Supersignal; Pierce Biotechnology, Inc.). 

Results

Genes Modulated by Vitreous Treatment

Treatment of low-passage human RPE cells in serum-containing medium with 25% vitreous results in reproducible morphologic changes within 48 hours, although changes can often be detected earlier. ^{12 27} Hence, changes in gene expression were examined after 6, 12, 24, or 48 hours of incubation with normal or vitreous-containing medium. Some heterogeneity of response was expected because of biological variation between vitreous or RPE cells from different donors. Because the goal was to identify general responses to vitreous, microarray analysis was performed on three different RPE donor–vitreous donor pairs (RPE donors 1–3). In all cases, both control and vitreous-treated cells were kept subconfluent throughout treatment, partly because it is assumed that as RPE cells migrate into the vitreous, they are no longer in confluent monolayers and start to divide. ^{2 4} In addition, this method avoids possible complications resulting from vitreous-treated cells’ being less subject to contact inhibition than untreated cells. ¹²  

Genes that gave microarray signals that were at least 10 SD above background and that also showed at least twofold changes in expression in all three microarray donors (RPE donors 1–3) at any time point were identified. These 200 genes are referred to as the twofold/donors 1 to 3 gene set. Thus, these were genes with mRNA that was readily detectable on arrays with reproducible changes in expression at a level that should be detectable by qPCR. However, such an analysis may miss many mRNAs with lower levels of expression for at least one donor, or those that were regulated by less than twofold in at least one donor. Therefore, the data were also analyzed with SAM statistical analysis. SAM takes into account not only the ratio of a gene’s expression in treated compared with control cells, but also the variation in expression. It can thus detect changes in regulation of genes with expression levels or expression changes less than the cutoffs used for the twofold/donors 1 to 3 gene set. Not surprisingly, SAM analysis at each time point using a false-discovery rate (FDR) of <5% detected many more regulated genes. The number of regulated genes increased at longer periods of vitreous treatment, consistent with a prolonged, progressive alteration in the gene transcription program (Fig. 1 ; see Appendix B for the gene list). The 1335 regulated genes from the SAM FDR < 5% gene set included 187 of the 200 genes in the twofold/donors 1 to 3 gene set. Because these 187 genes were likely to be robustly regulated by vitreous in all donors at levels detectable by qPCR, we refer to this subset of the SAM FDR < 5% gene set as the “stringent gene set” (Appendix C). We focused on the stringent gene set for much of the initial validation. Most (79%) of the genes in this set were regulated at multiple time points, supporting the concept that these genes were robustly regulated by vitreous. A further indication of the reliability of the data was that 96% of those genes regulated at multiple time points were regulated at adjacent time points. 

Validation of Changes in mRNA Expression

To test the reliability of the microarray analysis, the effect of vitreous-treatment on 10 of the vitreous-regulated genes was determined in three additional donor/vitreous pairs (RPE donors 4–6) at 6, 12, 24, or 48 hours of treatment. Some genes were chosen because the changes observed on the microarrays were consistent with an EMT response. These included α-5 integrin (ITGA5), ²⁸ Slug (SNAI2), ²⁹ and collagen IV (COL4A1). ³⁰ Connective tissue growth factor (CTGF) and thrombospondin-1 (THBS1) were included because others have speculated that increased expression of these genes, possibly by RPE cells, play a role in the early stages of PVR, ^{31 32} but the microarrays showed decreased expression. The remaining five genes (SAT, TFPI2, tissue plasminogen activator [PLAT], ADAMTS5, and TFRC) were chosen arbitrarily because they were up- or downregulated at multiple time points. However, recent reports indicate that TFPI2 is an RPE-specific growth factor ³³ and that SAT is involved in cell migration. ³⁴ These genes were all in the stringent gene set, except for SNAI2 which was in the SAM FDR < 5% gene set only. The qPCR results confirmed the microarray data for all 10 genes (Table 2 ; compare the average of the microarray results for RPE donors 1–3 with the qPCR data for donors 4–6). In general, the kinetics of gene regulation was similar to the microarray analysis, although the response of RPE donor 6 tended to be more transient than that of donors 4 and 5. 

We also examined expression of some of these genes at the protein level. We have reported that vitreous induces increased expression of α-5 integrin protein. ²⁷ Expression of PLAT protein was examined by immunohistochemistry of RPE cells grown in the presence or absence of vitreous for 48 hours and then treated with 5 ng Brefeldin A/mL for 6 hours to prevent secretion of PLAT. Control cells showed little cytoplasmic fluorescence for PLAT (Figs. 2A 2B 2C) . However, vitreous treatment caused increased levels of PLAT protein in virtually all cells (Figs. 2D 2E 2F) . Expression of TFPI-2 protein was examined by immunoblot analysis of extracts from cells grown in the presence or absence of vitreous and treated with 5 ng Brefeldin A/mL for the final 6 hours of treatment. TFPI-2 protein levels were increased in the vitreous-treated cells at both 24 and 48 hours (Fig. 3) . 

Changes in mRNA levels for an additional eight genes from the SAM FDR < 5% list have been confirmed by qPCR (data not shown). Most attempts to confirm changes in mRNA expression of genes from this list have been successful. Failures were usually because of problems with design of good qPCR primers or because the signal in qPCR was too low. 

We have reported that the mRNA and protein for three genes associated with inflammation, cyclooxygenase-2 (PTGS2), prostaglandin E synthase (PTGES), and heme-oxygenase 1 (HO1) are increased by vitreous ^{15 35} ; all three geneswere present in the SAM FDR < 5% gene set. Fan et al. ¹⁴ observed vitreous-induced increases in expression of genes associated with inflammation in a human RPE continuous cell line (ARPE-19). Consistent with this, gene ontology analysis of the SAM FDR < 5% gene set showed that categories associated with inflammation and stress (response to stress, cytokine activity, innate immune response) were significantly overrepresented (EASE score <0.05) in the vitreous-induced genes. However, there was little overlap with the 24 vitreous-regulated genes identified by Fan et al. ¹⁴ in ARPE-19 cells. This could be because they used cDNA arrays with only 588 genes represented, or differences in the type of cells, the exact details of the vitreous treatment, or other aspects of the array methodology. 

We have reported that 48 hours of vitreous treatment results in decreased levels of fibroblast growth factor 2 (FGF2) mRNA and protein and that FGF2 downregulation plays an important role in the response to vitreous. ^{12 27} FGF2 was downregulated at 48 hours on the arrays for donors 1 and 3 (ratio of vitreous to control 0.49 and 0.51, respectively) but not in donor 2 (ratio 1.25). Interpretation of the array data is complicated, because the arrays only detect the largest of the four forms of FGF2 mRNA found in RPE cells ³⁶ and thus may give misleading results. Even if donor 2 failed to downregulate FGF2, it is likely that signaling via the FGF pathway decreased, since FGF1, FGF5, and fibroblast growth factor receptor-2 (FGFR2) were all significantly downregulated (SAM analysis, FDR < 5%). In addition, vitreous contains soluble FGFRs, which would sequester FGF2 ³⁷ and opticin, which inhibits the FGF2 pathway (Le Goff MM, et al. IOVS 2005;46:ARVO E-Abstract 451). 

Pathway Analysis of Genes Modulated by Vitreous

The Genetic Information Processing, Environmental Information Processing and Cell Processes pathways from the KEGG database were analyzed with the WebGestalt program. Table 3shows the results for all pathways identified as significantly (P < 0.05) overrepresented in the SAM FDR < 5% or the stringent subset of vitreous-modulated genes compared with the total gene set on the microarrays. Of the six pathways identified as the most significantly (P < 0.01) overrepresented in the stringent gene set, five (extracellular matrix–receptor interaction, focal adhesion, TGF-β signaling, cell communication, and hematopoietic cell lineage pathways) were also among the most significantly (P < 0.01) overrepresented in the SAM FDR < 5% gene list. Because epithelioid RPE cells in culture tend to stay associated with each other after cell division, whereas vitreous-treated cells tend to migrate away from each other, ²⁷ the overrepresentation of genes associated with the focal adhesion pathway (Table 4)was of interest. The downregulation of the mRNA for genes associated with tight junctions at later times (Table 5)was also consistent with less close contact between cells. The overrepresentation of genes associated with the TGFβ/BMP pathway (Table 6 , Fig. 4 ) was noteworthy because it has been proposed that TGFβ plays an important role in PVR. ³⁸ Table 6includes six additional genes (RUNX2, POSTN1, DLX2, BAMBI, LAPTM5, and NBL1) from the SAM FDR < 5% gene set that are strongly associated with the TGFβ and/or BMP pathways but that were not in the KEGG pathway gene list. 

Effects of Vitreous on Expression of BMP-2 and Periostin mRNA

Because the potential involvement of the BMP-2 arm of the TGFβ/BMP-2 pathway (Fig. 4)was intriguing, mRNA levels for BMP-2 and periostin (a target of BMP-2 signaling, SAM [FDR < 5%] gene set) were measured by qPCR (Table 7) . Both were significantly upregulated by vitreous. Upregulation of periostin appeared to be later than that of BMP-2, as would be expected if it were a downstream target of BMP-2 signaling. 

Immunohistochemistry and Immunoblot Analysis of BMP-2

To determine whether vitreous modulates the levels of BMP-2 protein, RPE cells were grown in the presence or absence of vitreous for 48 hours and then treated with 5 ng Brefeldin A/mL for 6 hours to prevent secretion of BMP-2. Cells were stained using anti-human BMP-2 antibody to detect intracellular BMP-2 and were also stained for F-actin so that the proportion of cells expressing BMP-2 could be determined. Control cells showed little cytoplasmic fluorescence for BMP-2 (Figs. 5Aa-Ac ; the apparent nuclear fluorescence may be due to nonspecific binding). However, vitreous-treatment caused increased levels of BMP-2 protein in virtually all cells (Figs. 5Ad-f) . 

Immunoblot analysis indicated that treatment of the cells for 42 hours with vitreous followed by 6 hours with Brefeldin A resulted in the accumulation of a protein with an apparent molecular weight of ∼42 kDa that bound anti-human BMP-2 antibodies (Fig. 5B) . This size of protein was consistent with the molecular weight of pro-BMP-2 (the expected form of BMP-2, since Brefeldin A should inhibit processing of pro-BMP-2). 

Discussion

Oligonucleotide microarray technology enables mRNA levels for many genes to be examined simultaneously. Changes in mRNA levels are often (although not always) a good indication of changes in the expression of the corresponding protein. Thus, knowledge of global changes in mRNAs is valuable for our understanding of how cells respond to changes in their environment. In the present work, we examined the effects of vitreous treatment on low-passage human RPE cells, since this may help to elucidate early changes in gene expression in RPE cells in PVR. Biological variation is to be expected when cells from different donors are used. For example, variations in the repertoire of retinal gene expression between normal individuals have been reported. ³⁹ Using differential display methodology, we observed that some vitreous-induced gene changes were dependent on a particular RPE donor or a particular vitreous donor, whereas other gene changes were common to multiple RPE and vitreous donors. ¹² In the present study the use of multiple RPE donors and multiple vitreous donors served a filtering function enabling us to remove responses specific to a particular donor due to genetics, phenotypic changes, undetected disease processes, and medications. This may in part explain the high validation rate of independent RPE–vitreous donor combinations using qPCR. 

The number of vitreous-modulated genes increased from 6 to 48 hours of vitreous treatment, indicative of a gradual and major reprogramming of cell functions. Some changes were consistent with an alteration from epithelial-like to more motile and fibroblast-like cells. For example, genes in the tight junction pathway were downregulated by vitreous (Table 5) . Keratins 18 (KRT18) and 20 (KRT20), typical of simple epithelia ⁴⁰ and collagen IV (COL4A), a typical component of epithelial cell basement membranes, ⁴¹ were also downregulated. Integrin α5 (ITGA5) is typically upregulated during epithelial–mesenchymal transformation. ²⁸ We have shown integrin α5 protein is increased by vitreous treatment, ²⁷ and in this study we report that the mRNA was upregulated. Slug is a transcriptional repressor associated with EMT in development and cancer ²⁹ and epithelial cell migration during wound healing. ⁴² Slug and integrin α5 both have antiapoptotic effects, ^{43 44} and increased expression of such factors could promote survival of RPE cells after detachment from the basement membrane during the early stages of PVR. Downregulated genes with expression that is associated with the differentiated state of RPE cells included MITF (microphthalmia-associated transcription factor), which is associated with RPE development ⁴⁵ ; ezrin (VIL2), which is important in the long apical RPE microvilli ¹ ; and retinol dehydrogenase 5 (RDH5) which participates in the RPE visual cycle. ¹  

The TGFβ/BMP pathway was significantly overrepresented in the vitreous-modulated genes. TGF-β and BMP belong to the TGFβ superfamily of growth factors, both have been implicated in EMT ^{46 47} and both can induce Slug. ²⁹ TGF-β is known to play an important role in wound-healing and fibrosis and it has been postulated that it plays a role in PVR. In a mouse model of PVR, signaling via SMAD3 (a downstream target of TGFβ) is necessary for PVR, for transdifferentiation of RPE cells and for RPE cell expression of the myofibroblast marker, smooth muscle actin. ⁴⁸ Normal vitreous contains TGFβ2, and levels of this cytokine have been reported to be elevated in PVR. ^{49 50 51 52 53} Hence, vitreous treatment may be expected to activate TGF-β signaling, and we have preliminary evidence that TGF-β signaling plays a role in vitreous-induced phenotypic changes in low-passage human RPE cells ³⁵ (Parapuram S, Li L, Ganti R, Hunt RC, Hunt DM, unpublished data, 2004–2006). From the array data, it appeared that the TGFβ-dependent arm of the pathway might be downregulated at later times, since expression of mRNA for TGFβ2, the major form of TGFβ made by RPE cells, ⁵⁴ and its receptor TGFβR1 was decreased at 12 to 48 and 24 to 48 hours, respectively. The decrease in thrombospondin-1 (THBS1) and increase in latent TGFβ-binding protein (LTBP1) expression could result in less activation and more sequestration of TGFβ. ⁵⁵ CTGF has been reported to be induced by TGFβ, to prolong TGFβ signaling, and to be important in some of the profibrotic effects of TGFβ. ⁵⁶ Thus, although downregulation of CTGF mRNA was unexpected in view of some of the models of the initial events in PVR, ³¹ this finding would fit with the concept of decreased TGFβ signaling at later times in vitreous-treated RPE cells. In PVR, RPE cells can acquire myofibroblast-like properties, and TGFβ and CTGF play important roles in myofibroblast differentiation, ⁵⁶ which is associated with increased smooth muscle actin (ACTA2) expression. ACTA2 mRNA expression was decreased at later times, which would also be consistent with a downregulation of the TGFβ/CTGF pathway. Thus, although the vitreous-treated RPE cells have some features of fibroblasts, they may not be differentiating into myofibroblasts in this system, at least at the times examined. NRK fibroblasts treated with TGFβ and CTGF do not differentiate into myofibroblasts if they are still proliferating ⁵⁷ and our experiments were performed with subconfluent cells in the presence of serum. Thus, even if TGFβ and CTGF were present and functional at early times, they may not be able to induce full myofibroblast differentiation in the subconfluent system. The fact that genes upregulated by vitreous did not include genes associated with fibrotic extracellular matrix production such as fibronectin or collagen I would be consistent with this. Thus, cultured, vitreous-treated RPE cells may be able to control fibrotic and myofibroblast responses induced by TGFβ by downregulation of the TGFβ pathway at later times, at least in subconfluent conditions. In vivo, additional factors, including variation between individuals in patterns of gene expression, ³⁹ may interfere with this protective mechanism and contribute to the disease progression and fibrosis. 

The array data were consistent with upregulation of the BMP-2 pathway. BMP-2 mRNA and protein were increased (Table 7 , Fig. 5 ), and the BMP pathway inhibitors BAMBI, NBL1 and smurf2 were downregulated. In addition, periostin, runx2, DLX2, and ID1 have all been reported to be targets of BMP-2 signaling ^{58 59 60} and all were upregulated. Of interest, in view of the fact that ID1 can suppress myogenesis, ⁶⁰ expression of genes related to muscle development was significantly downregulated in vitreous-treated cells. Although runx2, periostin, and ID1 are typically BMP targets, they can also be targets of the TGFβ pathway. ^{61 62 63} There is cross-talk between the TGFβ and BMP pathways, ⁶⁴ and a balance between these pathways may be important in the effects of vitreous on RPE cells. 

Our results show that many of the effects of vitreous treatment on gene expression by RPE cells are consistent with what might be expected in early PVR. These include major changes in gene expression, upregulation of genes associated with stress and inflammation, downregulation of genes associated with epithelial cells and upregulation of genes associated with migration. The TGFβ pathway has already been implicated in PVR, but from the array data, it is possible that signaling via other members of the TGFβ family such as BMP-2 may also be important. 

 Supported by National Eye Institute Grant EY12711 and a DRF grant from the University of South Carolina School of Medicine

 Submitted for publication February 23, 2006; revised August 8 and December 14, 2006; accepted February 16, 2007.

 Disclosure: R. Ganti, None; R.C. Hunt, None; S.K. Parapuram, None; D.M. Hunt, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Table 1.

 

Primers Used in qPCR

Gene	Symbol	Sense Primer	Antisense Primer	E*
Normalization gene
Ribosomal protein, large, P0	RPLP0	TTAAACCCCCTCGTGGCAATC	CCACATTCCCCCGGATATGA	92%
Validation genes
Integrin, alpha 5	ITGA5	CCCCACCCTCATTCTACTGA	GCCCCATTTGAGTTCTGATTC	90%
Snail homolog 2, Slug	SNAI2	AGTGATTATTTCCCCGTATCT	GGAGAGAGGCCATTGGGTAGC	94%
Spermidine/spermine N1-acetyltransferase	SAT	CATACTGCGGCTGATCAA	GCCAATCCACGGGTCATAGGT	98%
Tissue factor pathway inhibitor 2	TFPI2	CAACGCCAACAATTTCTACAC	CGCAGAAGCCCATACAAGTA	93%
Tissue plasminogen activator	PLAT	CGATGGCCGCATGACTT	CGGGTGTTCCTGGTCACGGTC	96%
Collagen, type IV, alpha 1	COL4A1	TCACGGGGGAAAACATAAGAC	TCGATGAATGGCGCACTT	91%
Connective tissue growth factor	CTGF	CATCTTCGGTGGTACGGTGTA	CAGGCAGTTGGCTCTAATCATAG	89%
Thrombospondin 1	THBS1	GCCGGCGTGAAGTGTACTAGC	TTGGCCGTGGCATGTTCGAC	97%
Aggrecanase-2	ADAMTS5	GGTGGTGCTAGGCGACAAG	CACATAAATCCTCCCGAGTAA	92%
Transferrin receptor	TFRC	GCCCAGCAGAAGCATTATCTT	TATAGAAATTGCCCAGTAACC	93%
Additional TGF/BMP pathway genes
Bone morphogenetic protein 2	BMP2	CCCGACACTGAGACGCTGTTC	GGGGAAGCAGCAACGCTAGAA	97%
Periostin, osteoblast specific factor	POSTN	CGGATCTTGTGGCCCAATTAG	TCCAGTATTTGCCCGTTGTAA	95%

*  Efficiency in qPCR.

Figure 1.

 

The number of genes whose expression is significantly upregulated (▪) or downregulated ( ) at the mRNA level by vitreous (SAM analysis, < 5% FDR) at each time point are shown.

Table 2.

 

Comparison of Vitreous-Induced Changes in mRNA Expression Determined by Microarray Analysis or by qPCR Analysis

GenBank Accession and Symbol*	Gene Description	Microarray Average (Donors 1–3), †							QPCR Donors 4–6, ‡
				6 h	12 h	24 h	48 h	6 h	12 h	24 h	48 h
Upregulated genes
NM_002205	Integrin, alpha 5	1.8	2.3	2.1	1.6	2.4	2.6	3.2	1.9
ITGA5						2.4	2.1	1.7	1.1
						1.7	1.1	2.3	1.4
BC014890	Snail homolog 2 (slug)	1.3	1.7	2.4	1.7	2.1	4.7	5.6	3.3
SNAI2						2.0	2.1	2.3	4.1
						0.6	0.9	0.9	2.4
NM_002970	Spermidine/spermine N1-acetyltransferase	2.1	2.1	2.1	1.1	1.2	2.7	2.6	0.9
SAT						3.2	3.5	2.8	3.4
						1.9	2.5	2.5	1.0
NM_006528	Tissue factor pathway inhibitor 2	2.1	2.7	3.4	3.1	2.5	3.0	3.3	3.6
TFPI2						5.3	5.9	5.5	3.0
						4.0	5.9	5.8	7.2
NM_000930	Plasminogen activator, tissue	2.1	3.0	2.3	1.6	4.2	11.6	13.0	2.6
PLAT						4.1	7.8	4.7	2.9
						3.5	5.5	2.7	5.6
Downregulated genes
NM_001845	Collagen, type IV, alpha 1	0.8	0.9	0.5	0.5	0.8	0.7	0.5	0.4
COL4A1						0.9	0.9	0.6	0.3
						1.0	0.7	0.7	0.6
NM_001901	Connective tissue growth factor	1.1	0.7	0.6	0.4	1.3	0.6	0.3	0.3
CTGF						1.1	0.8	0.6	0.4
						1.2	0.7	0.7	1.4
BC015134	Thrombospondin 1	1.0	0.8	0.6	0.4	0.7	0.5	0.3	0.3
NM_003246		0.9	0.5	0.5	0.4	0.7	0.4	0.3	0.3
THBS1						0.9	0.4	0.5	1.0
NM_007038	ADAMTS5 (aggrecanase-2)	0.4	0.2	0.2	0.1	0.5	0.2	0.1	0.1
ADAMTS5						0.3	0.2	0.2	0.1
						0.5	0.2	0.3	0.1
NM_003234	Transferrin receptor	0.3	0.5	0.5	0.6	0.2	0.4	0.3	0.3
TFRC						0.2	0.4	0.5	0.6
						0.2	0.4	0.5	1.0

*  The GenBank accession number used to design the oligomer on the microarray is given, followed by the HGNC symbol. In the case of THBS1, there were two clones on the array for this gene.

†  The ratio of mRNA levels in vitreous and control cells determined from microarray expression data for RPE donors 1 to 3 after 6, 12, 24, or 48 hours of treatment. The data were analyzed with SAM; results found to be significant (FDR < 5%) are shown in bold.

‡  qPCR was used to determine the ratio of mRNA levels in vitreous and control cells for the additional RPE donors 4 to 6 (top, middle, and bottom lines, respectively, for each gene) after 6, 12, 24, or 48 hours of treatment. The data were analyzed using REST-XL; results found to be significant (P < 0.05) are shown in bold.

Figure 2.

 

Effect of vitreous on protein expression of tissue plasminogen activator (PLAT). RPE cells (donor 7) were grown for 54 hours in either control medium (A–C, G–I) or medium with 25% vitreous (D–F), with Brefeldin A (5 ng/mL) present from 48 to 54 hours. The cells were then stained with anti-human PLAT antibodies or phalloidin (to stain F-actin) (A, D) staining for PLAT; (G) staining as for PLAT but with the primary antibody omitted; (B, E, H) phalloidin staining; (C, F) merged images of anti-human PLAT and phalloidin-stained cells; (I) merged images in (G) and (H).

Figure 3.

 

Effect of vitreous on protein expression of TFPI-2. RPE cells (donor 8) were grown for 24 or 48 hours in either control medium (C24, C48) or medium with 25% vitreous (V24, V48) with Brefeldin A (5 ng/mL) present for the last 6 hours. The extracts were subjected to immunoblotting with anti-human TFPI-2 antibodies.

Table 3.

 

Pathways Significantly Enriched in the Vitreous-Modulated Gene Sets

Pathway*	Gene Set Examined
		Stringent			SAM (FDR < 5%)
		Genes	P	Genes	P
ECM-receptor interaction	6	0.000	14	0.003
Hematopoietic cell lineage	6	0.000	13	0.007
Cytokine-cytokine receptor interaction	9	0.001	23	0.097, †
TGF-beta signaling pathway	5	0.002	12	0.009
Cell communication	5	0.002	16	0.002
Focal adhesion	7	0.007	34	0.000
Complement and coagulation cascades	3	0.038	10	0.024
MAPK signaling pathway	—	—	41	0.000
Circadian rhythm	—	—	5	0.002
Long-term potentiation	—	—	12	0.002
Wnt signaling pathway	—	—	18	0.014
Tight junction	—	—	15	0.016
Hedgehog signaling pathway	—	—	8	0.026
Adherens junction	—	—	10	0.038

*  KEGG Genetic Information Processing, Environmental Information Processing, and Cell Processes pathways were analyzed by using WebGestalt. All pathways that were represented by more than two genes and found to be significantly (P < 0.050, hypergeometric test) overexpressed in the gene set of interest compared with all the genes on the microarray are shown. (—) P > 0.100.

†   P = 0.050–0.010.

Table 4.

 

Vitreous-Modulated Genes in the Focal Adhesion Pathway

GenBank Accession	Gene Description	Symbol	Ratio Vitreous/Control*
						6 h	12 h	24 h	48 h
Upregulated genes
NM_001166	Baculoviral IAP repeat-containing 2	BIRC2	1.59	1.59	1.30	0.77
AK056838	Calpain 2, (m/II) large subunit	CAPN2	0.87	1.04	1.87	2.02
NM_002203	Integrin, alpha 2 (CD49B, alpha 2 subunit of VLA-2 receptor)	ITGA2	1.53	1.60	2.33	1.42
NM_002205	Sorting nexin 11	ITGA5	1.76	2.25	2.08	1.55
NM_002290	Laminin, alpha 4	LAMA4	1.07	1.10	1.28	1.59
NM_002291	Laminin, beta 1	LAMB1	1.27	1.61	1.87	1.48
NM_000228	Laminin, beta 3	LAMB3	2.65	1.63	1.43	1.14
NM_002293	Laminin, gamma 1 (formerly LAMB2)	LAMC1	1.31	1.54	1.03	0.71
AF022375	Vascular endothelial growth factor	VEGF	1.98	1.61	0.60	0.48
Downregulated genes
NM_001102	Actinin, alpha 1	ACTN1	0.83	0.84	0.72	0.65
AF070674	Baculoviral IAP repeat-containing 3	BIRC3	2.30	1.47	0.51	0.27
NM_001749	Calpain, small subunit 1	CAPNS1	1.11	1.47	0.62	1.15
NM_001759	Cyclin D2	CCND2	1.13	1.07	0.84	0.48
NM_001854	Collagen, type XI, alpha 1	COL11A1	0.80	0.58	0.58	0.74
NM_001845	Collagen, type IV, alpha 1	COL4A1	0.83	0.88	0.48	0.46
AK023345	Diaphanous homolog 1 (Drosophila)	DIAPH1	0.85	0.87	0.48	1.83
AK055601	FERM, RhoGEF and pleckstrin domain protein 2	FARP2	1.13	1.08	0.77	0.66
AY014272	Actin-like protein	FKSG30	0.97	0.87	0.61	0.85
AK056699	FYN oncogene related to SRC, FGR, YES	FYN	0.67	0.58	0.50	0.70
AF109681	Integrin, alpha 11	ITGA11	1.24	0.88	0.80	0.35
NM_000419	Integrin, alpha 2b (CD41)	ITGA2B	0.74	0.61	0.40	0.28
NM_002206	Integrin, alpha 7	ITGA7	1.04	0.84	0.59	0.82
NM_053025	Myosin, light polypeptide kinase	MYLK	0.88	0.82	0.66	0.78
NM_002524	Neuroblastoma RAS viral (v-ras) oncogene homolog	NRAS	0.93	0.87	0.89	0.68
NM_016205	Platelet derived growth factor C	PDGFC	0.80	0.79	0.78	0.59
NM_002609	Platelet-derived growth factor receptor, beta polypeptide	PDGFRB	0.75	0.56	0.60	0.95
NM_014308	Phosphoinositide-3-kinase, regulatory subunit 5, p101	PIK3R5	0.67	0.70	0.54	0.45
NM_002480	Protein phosphatase 1, regulatory (inhibitor) subunit 12A	PPP1R12A	0.85	0.74	0.82	0.65
AK055431	Protein kinase C, alpha	PRKCA	0.82	0.75	0.71	0.67
NM_002738	Protein kinase C, beta 1	PRKCB1	0.72	1.36	0.92	0.49
NM_012250	Related RAS viral (r-ras) oncogene homolog 2	RRAS2	1.02	1.05	0.80	0.63
NM_003246	Thrombospondin 1	THBS1	0.90	0.47	0.47	0.43
AL049689	Tenascin N	TNN	0.74	0.88	1.01	0.66
AF022375	Vascular endothelial growth factor	VEGF	1.98	1.61	0.60	0.48
NM_003461	Zyxin	ZYX	1.07	0.87	0.63	0.76

*  Ratio of vitreous and control mRNA levels for each gene from microarray analysis. The values for each time point are the average for RPE donors 1 to 3. Data for time points for which the results were found to be significant by SAM statistical analysis (FDR < 5%) are shown in bold.

Table 5.

 

Vitreous-Modulated Genes in the Tight-Junction Pathway

GenBank Accession	Gene Description	Symbol	Ratio Vitreous/Control*
						6 h	12 h	24 h	48 h
NM_001102	Actinin, alpha 1	ACTN1	0.83	0.84	0.72	0.65
NM_020770	Cingulin	CGN	0.88	0.63	0.71	1.03
AF101051	Claudin 1	CLDN1	0.67	0.82	0.71	0.47
NM_001306	Claudin 3	CLDN3	0.66	0.67	0.47	1.02
NM_005231	Cortactin	CTTN	0.90	0.91	0.64	0.57
AY014272	Actin-like protein	FKSG30	0.97	0.87	0.61	0.85
NM_002069	Guanine nucleotide binding protein, alpha inhibiting activity polypeptide 1	GNAI1	0.68	0.61	0.67	0.56
NM_005799	InaD-like (Drosophila)	INADL	0.94	0.80	0.55	0.81
NM_002524	Neuroblastoma RAS viral (v-ras) oncogene homolog	NRAS	0.93	0.87	0.89	0.68
NM_016948	Par-6 partitioning defective 6 homolog alpha (C.elegans)	PARD6A	1.02	0.81	0.65	0.75
NM_004156	Protein phosphatase 2 (formerly 2A), catalytic subunit, beta isoform	PPP2CB	1.04	1.02	0.90	0.67
AK055431	Protein kinase C, alpha	PRKCA	0.82	0.75	0.71	0.67
NM_002738	Protein kinase C, beta 1	PRKCB1	0.72	1.36	0.92	0.49
AK002107	RAB3B, member RAS oncogene family	RAB3B	0.98	1.03	0.96	0.59
NM_012250	Related RAS viral (r-ras) oncogene homolog 2	RRAS2	1.02	1.05	0.80	0.63

*  See Table 4 .

Table 6.

 

Vitreous-Modulated Genes in the TGF-β/BMP Pathway

GenBank Accession	Gene Description	Symbol	Ratio Vitreous/Control*
						6 h	12 h	24 h	48 h
Upregulated genes
NM_001200	Bone morphogenetic protein 2	BMP2	2.13	2.43	1.38	1.02
NM_004405	Distal-less homeobox 2	DLX2	3.07	1.89	0.62	0.51
NM_002165	Inhibitor of DNA binding 1	ID1	1.11	1.40	1.86	1.69
NM_000627	Latent transforming growth factor beta binding protein 1	LTBP1	1.11	1.23	1.65	1.23
NM_006475	Periostin, osteoblast specific factor	POSTN	0.97	1.22	1.78	1.75
AL353944	Runt-related transcription factor 2	RUNX2	0.80	1.38	1.92	1.59
Downregulated genes
NM_012342	BMP and activin membrane-bound inhibitor homolog	BAMBI	0.49	0.60	0.54	0.82
NM_001719	Bone morphogenetic protein 7 (osteogenic protein 1)	BMP7	0.84	0.75	0.59	0.75
NM_004405	Distal-less homeobox 2	DLX2	3.07	1.89	0.62	0.51
BC014469	Cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4)	CDKN2B	0.99	0.95	0.56	0.76
BC014941	Inhibitor of DNA binding 4	ID4	0.37	0.41	0.68	0.51
NM_006762	Lysosomal associated multispanning membrane protein 5	LAPTM5	0.86	0.79	0.48	0.55
NM_002467	V-myc myelocytomatosis viral oncogene homolog	MYC	1.30	1.07	1.09	0.68
NM_005380	Neuroblastoma, suppression of tumorigenicity 1 (DAN)	NBL1	1.22	1.15	1.01	0.67
NM_004156	Protein phosphatase 2 (formerly 2A), catalytic subunit, beta isoform	PPP2CB	1.04	1.02	0.90	0.67
AK002019	SMAD specific E3 ubiquitin protein ligase 2	SMURF2	1.31	1.24	0.62	0.42
NM_022739	SMAD specific E3 ubiquitin protein ligase 2	SMURF2	1.38	1.56	0.94	0.39
AK021874	Transforming growth factor, beta 2	TGFB2	0.64	0.43	0.42	0.51
NM_003238	Transforming growth factor, beta 2	TGFB2	0.81	0.69	0.66	0.67
AK002171	Transforming growth factor, beta receptor	TGFBR1	0.80	0.68	0.52	0.45
NM_003246	Thrombospondin 1	THBS1	0.90	0.47	0.47	0.43
BC015134	Thrombospondin 1	THBS1	0.96	0.82	0.56	0.41

*  See Table 4 .

Figure 4.

 

Vitreous-modulated genes in the TGFβ/BMP pathway. Orange: upregulated by vitreous; blue: downregulated by vitreous; orange/blue: upregulated at earlier times, downregulated at later times. Adapted from KEGG pathway hsa04350.

Table 7.

 

Effect of Vitreous Treatment on Expression of BMP-2 and Periostin mRNAs

GenBank Accession and Symbol*	Gene	Microarray Average (Donors 1–3)†							QPCR Donors 4–6‡
				6 h	12 h	24 h	48 h	6 h	12 h	24 h	48 h
NM_001200	Bone morphogenetic protein 2	2.1	2.4	1.4	1.0	3.1	2.6	2.8	0.7
BMP2						3.3	4.3	1.1	1.7
						3.0	3.0	0.9	0.9
NM_006475	Periostin, osteoblast specific factor	1.0	1.2	1.8	1.7	1.0	1.0	1.2	1.9
POSTN						2.3	4.6	4.1	6.8
						1.1	1.9	1.4	1.4

 Footnote symbols are as described in Table 2 .

Figure 5.

 

Effect of vitreous on BMP protein expression. (A) RPE cells (donor 7) were grown for 54 hours in either control medium (A a–A c, Ag–A i) or medium with 25% vitreous (Ad–A f), with Brefeldin A (5 ng/mL) present from 48 to 54 hours. The cells were then stained using anti-human BMP-2 antibodies or phalloidin (to stain F-actin). (A a, A d) Staining for BMP-2; (A g) staining as for BMP-2 but with the primary antibody omitted; (A b, A e, A h) phalloidin staining; (A c, A f) merged images of anti-BMP-2 and phalloidin-stained cells; (i) images in (g) and (h) merged. (B) RPE cells (donor 9) were grown for 24 or 48 hours in either control medium (c24, c48) or medium with 25% vitreous (v24, v48) with Brefeldin A (5 ng/mL) present from for the last 6 hours. A549 cell extract was the positive control (m). The extracts were subjected to immunoblot analysis with anti-human BMP-2 antibodies.

Supplementary Materials

Appendix A - 19.3 MB (.xls) 

The ratio of vitreous/control mRNA expression levels for all genes on the microarray determined using Genespring software. The first worksheet contains the data, the second worksheet has a key explaining the column headings on the first worksheet. 

Appendix B - 313 KB (.xls) 

The SAM (FDR<5%) gene list, including the GenBank Accession Number (column A), the Gene Description and HGNC symbol (columns B and C), whether vitreous caused up- or down-regulation of the gene (column D), the times at which the gene was found to be regulated (SAM analysis, 5% False Discovery Rate) (column E), the average of the ratio of expression in vitreous/control cells for all three donors at 6, 12, 24, or 48 hrs (columns F-I) as determined from the microarray data, and whether the gene is included in the stringent gene list (Column J). 

Appendix C - 63.5 KB (.xls) 

The stringent gene list, including the GenBank Accession Number (column A), the Gene Description and HGNC symbol (columns B and C), whether vitreous caused up- or down-regulation of the gene (column D), the times at which the gene was found to be regulated (SAM analysis, 5% False Discovery Rate) (columns E-H), and the average of the ratio of expression in vitreous/control cells for all three donors at 6, 12, 24, or 48 hrs (columns I-L) as determined from the microarray data. 

The authors thank Saravan Chaturvedi for help with database construction.