Identification of Differentially Expressed Genes in Keratoconus Epithelium Analyzed on Microarrays

Abstract

purpose. The present study was conducted to investigate differential gene expression in individual samples derived from fresh-frozen human keratoconus and normal corneal epithelium, using gene microarrays.

methods. Total RNA was extracted (11 keratoconus and 8 normal samples), labeled, and hybridized to microarrays (GeneChip; Affymetrix, Inc., Santa Clara, CA). GeneChip data were validated by verifying the expression profiles of 10 genes by real-time PCR and by recalculation using dChip software (Wong Laboratory, Department of Biostatistics, Harvard School of Public Health, Boston, MA). Furthermore, 3 of the 10 encoded proteins were stained by immunohistochemistry.

results. In comparison with normal cornea, the expression of 471 of the 5600 genes on the microarrays was changed in the keratoconus samples. This number was reduced to 47 with increased expression and 9 with decreased expression when more stringent selection parameters were applied. These genes are believed to be involved in keratoconus. Two of the candidate genes, lysyl oxidase and tissue inhibitor of metalloproteinase 3, are known to be involved in other eye diseases. Expression profiles were reproduced with the software dChip (Wong Laboratory) and real-time PCR. Increases in keratin 6 and 13 were also detected at the protein level.

conclusions. Keratoconus epithelium appears to be characterized by massive changes of the cytoskeleton, reduced extracellular matrix remodeling, altered transmembrane signaling, and modified cell-to-cell and cell-to-matrix interactions. Validation of gene expression with dChip analysis and real-time PCR indicates GeneChip to be a valid technique for investigation of epithelium from single dissected corneal samples. Association between alterations at RNA and protein levels was observed for some of the tested candidates.

Keratoconus (KC; On-line Mendelian Inheritance in Man [OMIM] 148300) is a noninflammatory bradytrophic human corneal disease. The bilateral distortion induces myopia and astigmatism. At first, glasses and contact lenses are used to compensate for the optical defects, but later a corneal transplantation may become necessary. KC is one of the most common causes of penetrating keratoplasty in the Western world. No medical treatment is known today; however, if single factors could be identified, they might serve as drug targets in a medical treatment—for example, in the form of eye drops that would be cost-effective compared with transplantation. 

KC is believed to be autosomal inherited, because both eyes are affected, it occurs more often in certain families, ^{1 2 3 4} and it develops in homozygous twins. ^{5 6 7 8} However, contradictory information exists in reports of unilateral KC ^{9 10 11} and cases of recurrent KC after transplantation, ^{12 13 14} but asymmetrical progression and transplantation with KC-affected donor corneas cannot be excluded. Diagnosis is usually based on Placido disk analysis, slit lamp biomicroscopy, and computer-assisted topography, which are often difficult to interpret, especially in young patients with mild symptoms. If genomic mutations can be identified, ^{15 16} they may serve as valuable markers in clinical diagnostics. 

KC is characterized by a reduction in corneal thickness. ^{17 18} The epithelium degenerates, loosens its smoothness, and becomes irregular. ¹⁹ Stromal erosion and breaking of the Bowman’s membrane are often observed. The cause of KC is unknown; it may originate in any of the three corneal cell layers. In many investigations, corneas have been treated as a homogeneous tissue, whereas we dissected the cornea and focused on the epithelium. If one or more genes are differentially expressed between KC and normal corneal epithelium, early detection of KC could be possible by simple gene expression analysis of scraped-off epithelium. 

High-density oligonucleotide microarrays (GeneChip; Affymetrix, Inc., Santa, Clara, CA), which provide the ability to measure gene expression of approximately 5600 known genes simultaneously, were used to study gene expression in individual patients. By comparing the expression in KC samples to normal samples derived from patients with myopia, we determined those genes that were differentially expressed and thus isolated 56 candidate genes that may be related to KC. 

Microarray analysis has a great potential but should be independently validated. In this case, expression of 10 candidate genes was verified with real-time PCR and, furthermore, data were recalculated with the alternative software dChip (Wong Laboratory, Department of Biostatistics, Harvard School of Public Health, Boston, MA). Both experiments demonstrated comparable expression patterns as determined on the GeneChips. 

The process from RNA to protein is regulated in numerous ways. We wanted to determine whether the observed changes at RNA level could also be observed at protein level, using immunohistochemistry. Three of the candidate proteins were antibody-stained and for two of these, an association was observed. 

Materials and Methods

Tissue

Fresh human corneal epithelial cells were obtained from 10 patients with KC having a corneal transplantation and from 25 myopes who underwent photorefractive keratectomy. One patient with KC had surgery on both eyes and contributed two samples: Ks3 and Ks122. Patients were asked not to wear contact lenses for at least 2 days before surgery. Tissue from the myopic eyes was considered to be normal. Ten minutes before surgery, the eye was anesthetized with 0.4% oxybuprocaine (SAD, Copenhagen, Denmark). The central 8 mm of the cornea was marked, and the epithelium was scraped off with a scalpel after a brief period of coagulation with absolute ethanol followed by flushing with saline. The procedure took place immediately before the primary surgery was initiated. Cells were quickly dissolved in 4 M guanidine thiocyanate (25 mM citrate, 0.5% lauroylsarcosine, and 100 mM mercaptoethanol) and immediately frozen in liquid N₂. Samples were stored at −80°C. Permission to collect human tissue was given from The Scientific Ethics Committee of Aarhus County, and informed consent was obtained from all patients in accordance with the Declaration of Helsinki. 

Isolation of Total RNA

Epithelial cells were lysed in a 100-μL glass homogenizer (Jencons, Leighton, UK). Total RNA was isolated with extraction reagent (RNAzol B; Wak Chemie Medical, Bad Soden/Ts, Germany) as described by the manufacturer. The amount of RNA was quantified photospectrometrically and checked for degradation by agarose electrophoresis. Samples labeled Ks and Ns are single-patient samples whereas Np18 is a pool of equal amounts of total RNA from 18 patients with myopia. 

cRNA Synthesis

Reverse transcription was performed on 9 μg total RNA for 1 hour at 42°C using a T7-oligo(dT₂₄)-primer and reverse transcriptase (SuperScript II; Life Technologies-Invitrogen, Carlsbad, CA). Second-strand cDNA synthesis was performed for 2 hours at 16°C using Escherichia coli DNA polymerase I, DNA ligase, and RNase H (Life Technologies) followed by incubation in 50 mM NaOH and 0.1 mM EDTA for 10 minutes at 65°C leading to degradation of the RNA. After phenol-chloroform extraction and ethanol precipitation, in vitro transcription was performed for 6 hours at 37°C using biotin-16-UTP and biotin-11-CTP with an RNA transcript labeling kit (BioArray; Enzo Diagnostics, Farmingdale, NY). cRNA was purified on spin columns (RNeasy; Qiagen, Valencia, CA), followed by fragmentation for 30 minutes at 95°C. Finally, spikes necessary for image adjustment after scanning were added. 

GeneChip Performance

Before analysis of 19 samples on Hu6800FL GeneChips (Affymetrix, Inc.), the quality of each sample was checked on Test3 arrays (Affymetrix). GeneChips were hybridized with 15 μg biotin-labeled cRNA at 45°C in a hybridization oven (model 640; Affymetrix) overnight, followed by a first staining with streptavidin-phycoerythrin in a fluidics station (model 400; Affymetrix) and a second staining with biotinylated anti-streptavidin and streptavidin-phycoerythrin. Finally, the GeneChips were scanned (GeneArray Scanner; Hewlett Packard, Palo Alto, CA). 

Data Mining

Data were extracted from antibody-stained microarray images and analyzed using the Microarray Suite 5.0 (MAS), Data Mining Tool 3.0, and MicroDB software (Affymetrix). GeneChips were globally scaled to the arbitrary value of 150. ²⁰ Probe sets for spikes were removed before further analysis of the remaining 7070 probe sets. 

Two different approaches were used for data analysis. One strategy was statistical and based on the nonparametric Mann-Whitney test of the arbitrary gene expression values (termed Signal by Affymetrix) in KC (n = 11) and normal (n = 8) epithelium. The other strategy was a comparison analysis of KC samples versus normal samples, based on MAS software interpretations of Detection (transcript scored as Present or Absent), Change (scored as Increased, Decreased, or Not Changed) and Signal Ratio. We emphasize that the Signal Ratio is not the simple ratio between two gene expression levels but is a complex analysis of individual probe cells. 

Two-way (gene and sample) hierarchical cluster analysis (GeneCluster 2.11, Michael Eisen, Lawrence Berkeley National Laboratory, University of California at Berkeley) ²¹ was performed on log-transformed, median-centered, and normalized (sum of squares similar to 1.0 in both directions) Signals on the 19 samples. Results were visualized with TreeView 1.5 (http://taxonomy.zoology.gla.ac.uk; developed by Roderic D. M. Page and provided in the public domain by the Institute of Biomedical and Life Sciences, University of Glasgow, Glasgow, Scotland, UK). ²²  

Gene expression in the 19 samples was recalculated using the software dChip 1.2 (Wong Laboratory, Dept. of Biostatistics, Harvard School of Public Health), ^{23 24} which is based on alternative algorithms to MAS. 

Real-Time PCR

One microgram total RNA was reverse transcribed into 20 μL cDNA for those samples with excessive RNA, using 50 pmol oligo dT₂₁-primer, 1 mM dNTP mix and 200 U reverse transcriptase (SuperScript II; Life Technologies-Invitrogen). Samples tested negative for DNA contamination, by amplification of microsatellite marker D2S119. 

RNA expression was determined with a sequence detection system (Prism 7000; Applied Biosystems, Foster City, CA). Gene-specific primers (Table 1) for 11 genes were designed to be located in the same region as the Affymetrix probe sets, and, if possible, to be intron spanning. Real-time PCR was performed in a volume of 25 μL containing 2 μL cDNA (×20), 7.5 pmol of each primer, and 12.5 μL master mix (SYBR Green PCR Master Mix; Applied Biosystems). The program was 50°C for 2 minutes and 95°C for 10 minutes, with amplification in 40 cycles of 95°C for 15 seconds and 60°C for 1 minute. Fluorescence was monitored after each elongation period. The program terminated with a melting curve analysis by elevating the temperature from 60°C to 95°C while monitoring the fluorescence. A standard dilution (×1, ×10, ×100, ×1,000, and ×10,000) composed of a large reference pool of cDNA, was amplified for each run. Samples were quantified compared with the standard curve created from those dilutions. 

The housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was confirmed to be constantly expressed in corneal epithelium from the GeneChip analysis. Samples were amplified with GAPDH primers for determination of the relative starting amount of cDNA in each sample, and all genes were normalized to that amount. Negative controls without template were produced for each run. Tissue samples were amplified in triplicate and from these replicates, averages were calculated. 

Immunohistochemistry

Eight archival corneas were fixed in 4% formalin and embedded in paraffin. Four samples were KC corneas, independent of the samples used for GeneChips and real-time PCR, and four were from eyes with ocular tumors (three with retinoblastoma and one with orbital tumor). In this study, the tumor samples were considered to be normal and served as a reference. Tissue sections (5 μm) were transferred to electrostatic slides (Menzel-Gläser, Braunschweig, Germany). Sections were deparaffinized in mineral oil for 10 minutes at 60°C and washed in ethanol. Antigens were retrieved by boiling in a 900-W microwave for 12 minutes in buffer (10 mM tris(hydroxymethyl)aminomethane, 0.5 mM EGTA [pH 9.0]). Endogenous peroxidase activity was quenched in 1.4% hydrogen peroxide for 20 minutes. Sections were blocked in 1:25 horse serum (Sigma, St. Louis, MO) for 20 minutes and incubated with primary antibody at 4°C overnight. Antibodies were cytokeratin 6 (K6, 1:2; Research Diagnostics, Flanders, NJ), cytokeratin 10/13 (K13, 1:100; Dako, Glostrup, Denmark), and vimentin (1:120; Santa Cruz Biotechnology, Santa Cruz, CA). They were rinsed in phosphate-buffered saline (PBS) buffer before the corresponding secondary antibody (biotinylated anti-mouse and anti-rabbit [both from Oncogene Research Products, San Diego, CA], or anti-goat [Santa Cruz Biotechnology]) was applied for 30 minutes. The sections were then rinsed in PBS, incubated in an avidin DH-biotinylated horseradish peroxidase H complex solution (Oncogene Research Products) for 30 minutes, and stained in 3-amino-9-ethyl-carbazole (Sigma) for 10 minutes. Finally, the slides were counterstained in hematoxylin and mounted. Control staining was produced by omitting the primary antibody. All antibodies were diluted in PBS containing 1% bovine serum albumin (Calbiochem, San Diego, CA). 

Results

Selection of Candidate Genes

Total RNA was extracted from human corneal epithelium. To ensure comparability of the arrays, we used pair-wise scatterplots. Figure 1 shows three scatterplots representative of the 171 possible plots. Most of the expression pairs were within a narrow range. The correlation coefficients were typically between 0.88 and 0.96 and for the two corneas from the same patient, it was 0.9485. The uniform array performance illustrated by the scaling factors and Present probe sets (Table 2) and by the scatterplots allowed us to proceed with comparative analyses, based on two different methodological approaches. 

The first selection method was based on MAS software (Affymetrix, Inc.) interpretations. KC and normal samples were compared each to each, resulting in 88 comparisons. When the criteria of Increase in expression in at least 50% of the comparisons was used, 78 increased probe sets were selected and, analogous to this, 23 decreased probe sets. When the parameter Signal Ratio was used instead, 41 probe sets showed an more than twofold increase and 18 probe sets a less than twofold decrease. The second method was based on selection of altering genes, using the nonparametric Mann-Whitney test on the Signal Ratios, and 414 probe sets were found to change significantly (P < 0.05). 

A total of 471 altered probe sets were selected. The MAS software analysis was the most conservative, selecting only 101 and 59 probe sets, respectively, compared with 414 selected by the statistical method. Probe sets that were selected by one of the two software-mediated tests and that were also significantly changed according to the nonparametric test were extracted. Those noninformative probe sets that were either increasing to Absent or decreasing from Absent were eliminated, which reduced the number of probe sets from 88 to 62. Some of the genes were detected with more than one probe set, indicating a good reproducibility of the method, and because of this, the final set of candidate genes consisted of 56 different genes. 

When we looked at all genes, the two KC samples from the same individual differed as much as two KC samples from different individuals. The two samples from the same individual were further studied. Figure 2 shows the average Signal Ratio between Ks122 and the eight normal samples plotted against similar values for Ks3. Despite the global dissimilarity between the two samples, a strong relation was observable when focusing on the candidate genes. Comparing random samples from different individuals showed much less relation (data not shown). 

Candidate Genes

Various functional groups were identified among the candidate genes (Table 3) of which the cytoskeleton, the extracellular matrix (ECM), transmembrane signaling, and cell-to-cell–cell-to-matrix interaction–related genes are some of the most well represented. Intermediate filaments such as keratins, are cytoskeletal building blocks. K6 is highly represented among the candidate genes, with four probe sets increasing 5.2-fold, on average. This gene shows the largest increase in gene expression. K13 and type III vimentin, which aggregate into homopolymers, were also increased, and as were the two small proline-rich proteins SPR2-1 and SPRR2B. They are small cross-linked molecules attached to the cytoplasmic side of the plasma membrane in keratinocytes. ²⁵ The cytoskeleton is, in addition to intermediate filaments, formed of microfilaments and microtubules as well. Gene expression of two major components of the microfilaments, α1-actinin and α1-tropomyocin increased as did the tropomyosin-binding protein troponin T1. Microtubules are involved in positioning chromosomes to the equatorial plan before cytokinesis is initiated. One element of the microtubules, α3-tubulin, showed increased expression. 

We observed upregulation of different signal proteins in the data set (Table 3) . The protein inhibitor of neuronal nitric oxide synthase, also known as the light chain of the 1.2 MDa dynein complex, is thought to control multiple events through the control of the neurotransmitter nitric oxide. ²⁷ Other selected genes involved in signaling are the sortilin-related receptor, and two G protein-coupled receptors GPR37 and RDC1. RDC1 acts as a coreceptor for immunodeficiency viruses. ²⁸  

Candidate genes encoding proteins related to the ECM included several protease inhibitors—secretory leukocyte protease inhibitor, two serine (or cysteine) proteinase inhibitors, tissue inhibitor of metalloproteinase 3 (TIMP3), and cystatin E/M—all of which were upregulated. No changes in expression of matrix proteins, such as collagen, elastin, and proteoglycan were observed. Only the lysyl oxidase, involved in cross-linking of collagens was increased. Four of the five inhibitors are characterized as being secreted from the cell, ^{29 30 31} with TIMP3 being bound to the ECM. ³² Cystatin M/E, an inhibitor of cysteine proteinases, is found both intracellularly and extracellularly. ³³ The 200-kDa cathepsin C was the only proteinase among the 56 candidate genes. It is an intracellular enzyme involved in the activation of lysosomal serine proteinases ³⁴ and thus is not in direct contact with the ECM in living cells. 

Genes encoding proteins involved in cell–cell interactions and cell–matrix adhesion also account for many of the probe sets showing increasing expression. These include the bullous pemphigoid antigen 1, carcinoembryonic antigen-related cell adhesion molecule 6, desmoglein 3, γ2-laminin, and lectin galactoside-binding soluble 7 protein. Desmogleins are components of the desmosomes that help the cells to adhere to each other. Intracellularly, they are linked to keratins. Apart from these sets showing increased expression, mesothelin was the only gene with decreased expression. Some of the candidate genes have multiple functions, and others are less characterized. 

Cluster Analysis

The 471 candidate genes were analyzed with a two-way hierarchical cluster analysis (Fig. 3A) . The 19 samples formed two groups on the horizontal axis, clearly separating all 11 KC samples from the eight normal samples (Fig. 3B) . Sample Ks3 and Ks122, which originated from the same patient, cluster close together but not on the same branch. In the gene cluster (vertical axis), only the keratins form a distinct biological group (Fig. 3C) . Figure 3D shows an example of a downregulated cluster involving several lipid-associated genes. 

Validation of Data with Real-Time PCR

Gene expression was validated with real-time PCR. Often the low yield of epithelial RNA is adequate only for analysis on a single GeneChip, but excessive material (>9 μg) was available from 13 samples. Expression of 10 candidate genes, showing some variation (low/high and increased/decreased expression), was tested on these samples. Figure 4 shows the expression profiles determined by the two techniques. The curve for real-time data is very similar to the GeneChip data, indicating a good reproducibility of the GeneChips, even for genes with low expression, such as cyclin D2 and 5′-nucleotidase. A striking finding was the rather heterogeneous expression level of some of the genes among the KC samples. When single genes were the focus, they appeared to be heterogeneously expressed, but when the total gene expression intra- and intergroup was examined with scatterplots, no difference in expression were observable between the group of KC and normal (data not shown). 

Validation of Data with dChip

Various software programs provide different algorithms for calculation of gene expression levels. Expression data were validated for some of the candidates by comparing them with data calculated with dChip. Profiles for the 10 candidate genes are included in Figure 4 . The expression profiles appear similar, although the dChip expression levels are much higher than those calculated by MAS, but, because all values are in arbitrary units, the profiles are not influenced by the different levels. 

Immunohistochemistry

A possible relation between alterations at the RNA and protein level was examined. Three of the verified candidate genes were antibody stained (Fig. 5) . The anti-keratin 10/13 antibody recognized only K13 on formalin-fixed, paraffin-embedded tissue sections (information from supplier) which was also confirmed by the absence of staining in the negative control. 

Four KC and four reference corneas were immunohistochemically stained. Anti-K6 antibody stained the epithelium in three fourths of the KC samples and anti-K13 antibody halves of the samples in a heterogeneous way, along the longitudinal axis. One sample (Fig 5B) showed staining of only the superficial epithelium with K13. The two keratin antibodies did not stain the corneal epithelium of the four eyes with ocular tumor. No systematic difference was detectable for vimentin, which weakly stained both KC and reference. 

Discussion

Gene expression in KC epithelium was examined with microarrays. Epithelium from myopes was considered to be normal and used as the reference. Alternatively, corneas can be obtained from a tissue bank, but the RNA may be degraded by postmortem RNase activity, and gene expression may change when tissue is stored in culture medium. Consequently, we developed a protocol to snap freeze fresh corneal epithelium and thereby also prevent induction of new transcripts. 

Corneal epithelium is separated from the stroma by the Bowman membrane, a rough layer of collagen that prevents contamination with keratocytes during isolation. The cornea is a unique tissue because it has no vascular system and therefore has no immune cells. Our samples can be regarded as very homogeneous, only infiltrated by nerve cells. The purity of the tissue samples is also implied by the absence of immunoglobulin gene expression findings in the data. 

The samples included in this study were epithelium from 9 KC corneas from single patients, a left and right cornea from one patient with KC, 7 single normal reference corneas, and a pool of 18 reference corneas. The left and right samples were investigated in a comparison analysis that showed 12% changed probe sets. The proportion was 10% to 20% when the two samples were compared with any of the normal samples (data not shown). Based on this information, the samples were analyzed independently of each other. 

Selection of candidate genes was performed in multiple steps. Low-stringency parameters were used to select 471 potential candidate probe sets. When stringency was increased, only 56 genes fulfilled the stated requirements. We assume that these candidates are likely to be involved in the KC disease; however, it cannot be ruled out that some of the excluded genes may be involved in the pathogenesis. The close relationship, illustrated in Figure 2 , between the left and right eye from the same individual supports the consistency of the candidates. Most of the differentially expressed genes are changing in the same direction and with similar magnitude for these two samples. 

Cytokeratins are mostly expressed in epithelium. They are strong markers of differentiation. ³⁵ Keratins combine in pairs of a type I and II subunits. K5 and K14 are expressed in nonkeratinized stratified epithelium but when cells detach from the basement membrane and differentiate in the suprabasal layers, K4 and K13 expression is switched on. ^{36 37} Superficial K13 staining was observed in one of the samples. K6 and K16 are markers for hyperproliferation and are expressed during wound healing and in skin with psoriasis. ³⁸ K6 (type II) and K13 (type I) were upregulated in KC but none of their partners was coexpressed. Perhaps, and this is strictly speculation, the keratins dimerize in KC into an atypical K6/K13 pair, a combination that has been shown in vitro. ³⁹ At present it is unknown whether such dimerization could have an impact on epithelial cell function. 

Cytokeratins are involved in several diseases (reviewed by Corden and McLean ⁴⁰ ) but have so far only been related to a single corneal disease, Meesmann corneal dystrophy (OMIM 122100) where mutations in the K3 or K12 gene prevent heterodimerization of keratin. ^{41 42} Two of the other selected candidate genes lysyl oxidase and TIMP3 are involved in the eye diseases Ehlers-Danlos syndrome type V (OMIM 305200) and Sorsby fundus dystrophy (OMIM 136900), respectively. 

We observed changes, not only in the keratins but in all parts of the cytoskeleton: microfilaments, intermediate filaments, and microtubules. They all showed an increase in gene expression. This suggests the internal cell structure to be heavily reinforced in KC epithelium. 

The finding of the two PDZ/LIM domain-containing proteins is most interesting, because they are able to construct a linkage between the cytoskeleton and intracellular kinases as part of signal transduction through the cell membrane. The extracellular part of a signal transduction complex is the ligand-recognizing receptor. Different receptors were selected: the LDL-binding sortilin-related receptor 1 suspected to be involved in Golgi-endosome sorting, ^{43 44} and the uncharacterized G protein-coupled receptors GPR37 and RDC1. Information about their functions is very limited. 

Homeostasis of the ECM is regulated by the processes of synthesis and degradation, the latter being tightly controlled by balancing the activity of proteinases and proteinase inhibitors. The observed increase of secreted proteinase inhibitors does not support the traditional idea of corneal protease degeneration (reviewed by Collier ⁴⁵ ), but it should be kept in mind that the data presented in this article originate from the epithelium and cannot be extrapolated to the entire cornea. If corneal degradation is caused by increased proteinase activity, it does not appear to be directed from the epithelium. On the contrary, our data suggest a reduced ECM remodeling. At present, a protocol is being developed for analysis of the stroma. 

Cell-to-matrix and cell-to-cell interacting proteins were also well represented. Two of them, desmoglein 3 and bullous pemphigoid antigen (BP230), are involved in severe skin diseases. ^{46 47} Another hemidesmosomal component besides BP230, the bullous pemphigoid antigen BP180, is suspected to be involved in KC, but Cheng et al. ⁴⁸ found no difference at the protein level. Mutations in γ2-laminin cause the skin disease junctional epidermolysis bullosa. ⁴⁹ Galectin 7 and γ2-laminin have also been found to increase in whole corneas from mice during wound healing. ⁵⁰  

One of the differentially expressed transcripts between normal and KC tissues has not previously been described in the human cornea. Nagase et al. ⁵¹ cloned KIAA0095 from a human myeloid cell line in 1994. We recently blasted the protein sequence to GenBank (http://www.ncbi.nlm.nih.gov/Genbank; provided in the public domain by the National Center for Biotechnology Information, Bethesda, MD) and found similarity to nucleoporin-interacting component NIC96 in yeast, An4 in frog, and dye (dead eye) in zebrafish; but now, 8 years after its cloning, it remains uncharacterized. Mutations in the dye gene were found to cause necrosis in eyes of zebrafish embryos. ⁵²  

Most of the 56 altered candidate genes were in the direction of an increase. KC appeared to be characterized by relative small changes in gene expression with changes ranging from −2.7 to +6.6-fold (averages of 88 comparisons). Similar low magnitudes of change have been observed in other quantitative studies of KC, ^{53 54 55} which may be related to the bradytrophic character of the disease. 

The incidence of KC in patients with Down syndrome is higher than in the normal population ^{56 57} and thus has been speculated to be related to chromosome 21 in some way. We examined the chromosomal localization of the 56 candidate genes, but none of them was positioned on chromosome 21. 

This year, Fullerton et al. ²⁶ published the results of a genome scan study involving eight patients with KC, based on the founder population in Tasmania, Australia. We examined the loci of the candidate genes with the inherited regions from the study and found six of the candidates to be located within these conserved regions. Although the 10-cM genome region comprises several genes of interest, these six candidates are obvious subjects for further studies. Most recently, a Finnish study comprising 20 KC-affected families linked the disease to locus 16q22-q23, ⁵⁸ but none of our candidate genes located to this position. 

The candidate genes represent multiple functional groups, whereas our focus was on the larger changes. Summarizing the results from the screening of 5600 genes, we observed increases in all three parts of the cytoskeleton and in proteins closely related thereto, and we saw signs of reduced ECM remodeling in the form of secreted proteinase inhibitors. KC causes changes in transmembrane signaling, together with altered cell-to-matrix cellular properties of the corneal epithelium. No indication of apoptosis was registered, as described by Kim et al. ⁵⁹ and recently by Kaldawy et al. ⁶⁰ in a TUNEL assay. 

Analyses of microarray data are dependent on the algorithms and integrated cutoff values of the software. The MAS-calculated data were recalculated with dChip, which provides an alternative normalization and calculation of gene expression. First, dChip normalizes to a group of microarrays, in contrast to MAS, which normalizes microarrays separately; and second, the conservative MAS normalization uses the scaling factor across the entire expression set, whereas dChip uses a running median. Finally, the calculation of expression levels is diverse. ^{23 24} Despite the differences in software, the two analyses yielded similar results for the 10 candidate genes. We also verified the expression with real-time PCR. The expression profiles were well reproduced, confirming that it is possible to determine gene expression in dissected corneas of individual patients with GeneChip microarrays. It also showed that when candidate genes are identified with the use of GeneChip microarrays, high throughput and low-cost real-time PCR is appropriate for subsequent and more extensive investigations. 

High myopic refraction, corneal thinning, and opacification suggest an advanced stage of KC, but these cannot explain the heterogeneous gene expression among the KC samples observed with real-time PCR. The differences may be an indication of subgroups, but more samples must be investigated. Unsupervised cluster analysis was performed on the candidate probe sets, and the 19 samples were correctly separated into KC and normal groups, but no new clinical subgroups of KC were detected. That the two samples derived from the same patient did not cluster to the same node suggests that the samples are too few, compared to the variation in gene expression, to be divided into subgroups. 

Changes at the mRNA level are not necessarily reflected at the protein level. We examined whether the alterations found on the GeneChip array would also be observed at protein level. Cytoskeletal changes were indicated by an increase of K6, K13, and vimentin at the RNA level. These three genes had been verified by real-time PCR, and commercial antibodies were available. From antibody staining of a small series of samples 4+4, we registered an overall relation between RNA and protein level for the two keratins. The heterogeneous staining of KC samples possibly demonstrates the natural variation that was also observed on the GeneChip array. Unfortunately, we did not have sets of cornea for both microarray and immunohistochemistry. 

Microarray technology is rapidly entering the field of vision, as recent publication of studies involving cell cultures, whole corneal buttons, and knockout animals demonstrate. ^{15 50 61 62} The current study is, to our knowledge, the first differential microarray study on human corneal epithelium that identifies genes differentially expressed between KC and normal tissue. We have proposed 56 candidate genes that changed expression level between 11 KC and 8 normal samples, based on different selection strategies. Each gene deserves further study in larger sample collections, with characterization of the respective corneal proteins. Instead of considering the cornea as a homogeneous tissue, we dissected the tissue and examined the epithelium. In the future we will use microarrays to investigate the stroma. 

 Supported by the Danish Eye Research Foundation (Copenhagen, Denmark), the Danish Eye Health Society (Copenhagen, Denmark), Aarhus University Hospital Research Initiative (Aarhus County, Denmark), and the Institute of Experimental Clinical Research (Skejby University Hospital, Denmark).

 Submitted for publication July 5, 2002; revised November 6 and November 27, 2002; accepted January 30, 2003.

 Disclosure: K. Nielsen, None; K. Birkenkamp-Demtröder, None; N. Ehlers, None; T.F. Orntoft, 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.

 

Primer Sets Used for Real-Time PCR, Designed According to the Manufacturer’s Description

Gene	Upstream Primer	Downstream Primer
CCND2	gctggctaagatcaccaacaca	cctcaatctgctcctggcaa
DSG3	ggcagtctggaaccatgagaa	tcctggccatcgtcttcct
GAPDH	tgccaaatatgatgacatcaagaa	ggagtgggtgtcgctgttg
HBP17	gcctgggattgcactggat	aaggagagcagggtgaggcta
KIAA00095	gaggaccgcgactctcaact	aaaggtaatcagagtgcgggc
KRT6A	ctgaatggcgaaggcgtt	ctgccgacaccactggc
KRT13	aacgtggagatggatgcca	tggcgtggaaccattcct
NT5	tggagatgggttccagatgataa	ggataaattactttcattttggagatatatgta
PCK1	actcgaggttctgcacccct	aggcagcatcaatgatggg
SLPI	cctggatcctgttgacaccc	cacttcccaggcttcctcct
VIM	ccaaacttttcctccctgaacc	gtgatgctgagaagtttcgttga

Figure 1.

 

Three representative scatterplots from six randomly chosen microarrays. The correlation coefficient (R ²) is shown in each plot.

Table 2.

 

Information on Participants

	Normal Samples															Keratoconus Samples
		Ns50	Ns58	Ns64	Ns78	Ns107	Ns108	Ns121	Np18	Ks3*	Ks7	Ks11	Ks42	Ks76	Ks122*	Ks187	Ks200	Ks313	Ks400	Ks401
Sex	F	F	F	F	M	M	M	9M + 9F	F	M	F	M	M	F	F	M	M	M	M
Age	28	30	25	28	39	50	24	29	43	28	76	25	32	44	20	59	25	50	42
Corneal power (D)	46	51	47	46	51	47	48		62	>63	>63	>63	54	>63	57	63	59	69	49
Maculation (+/−)	−	−	−	−	−	−	−	−	+	+	+	+	+	+	+	−	−	+	+
Corneal thickness (mm)									0.36	0.47	0.20	0.38		0.31	0.35	0.38	0.48	0.37	0.32
Scale Factor	0.153	0.171	0.141	0.099	0.162	0.130	0.130	0.209	0.142	0.130	0.146	0.147	0.171	0.182	0.175	0.135	0.124	0.135	0.105
Present Probe Sets (%)	45	42	45	49	44	45	45	44	47	47	47	45	46	42	43	46	42	46	47

Figure 2.

 

Scatterplot from one individual showing Signal Ratios from sample Ks122 (left eye) plotted against Ks3 (right eye) for the 62 candidate probe sets. The Signal Ratios of both Ks3 and Ks122 were calculated using the average of the eight normal samples as a reference.

Table 3.

 

Summary of the 56 Candidate Genes

Probe Set	Gene Name	Function	Signal Ratio	Locus
	Cytoskeleton
L42583	Keratin 6A (KRT6A)	Intermediate filament	6.6	12q12-q13
L42601	Keratin 6C (KRT6C)	Intermediate filament	6.3	12
V01516	Keratin 6A (KRT6A)	Intermediate filament	5.9	12q13
L42611	Keratin 6A (KRT6A)	Intermediate filament	1.9	12q12-q13
X52426	Keratin 13 (KRT13)	Intermediate filament	3.4	17q21-q23
M19309	Troponin T1 (TNNT1)	Cytoskeleton	2.5	19q13.4
Z19554	Vimentin (VIM)	Intermediate filament	2.0	10p13
X53065	Small proline rich protein (SPR2-1)	Cytoskeleton	1.7	—
L05188	Small proline-rich protein 2 (SPRR2B)	Cytoskeleton	1.6	1q21-q22
X01703	Tubulin alpha 3 (TUBA3)	Cytoskeleton	1.5	12
M19267	Tropomyosin 1 alpha (TPM1)	Cytoskeleton	1.5	15q22.1
Z24727	Tropomyosin 1 alpha (TPM1)	Cytoskeleton	1.4	15q22.1
M95178	Actinin alpha 1 (ACTN1)	Cytoskeleton	1.3	14q24*
	ECM-related
U62800	Cystatin E/M (CST6)	Proteinase inhibitor	2.2	11q13
J03764	Serine (or cysteine) proteinase inhibitor (SERPINE1), plasminogen activator inhibitor type 1 (PAI-1)	Proteinase inhibitor	2.2	7q21.3-q22
M93056	Serine (or cysteine) proteinase inhibitor (SERPINB1), elastase inhibitor	Proteinase inhibitor	1.6	6p25
U14394	Tissue inhibitor of metalloproteinase 3 (TIMP3)	Proteinase inhibitor	1.6	22q12.1-q13.2
L16895	Lysyl oxidase (LOX)	ECM synthesis	1.5	5q23.3-q31.2
X04470	Secretory leukocyte protease inhibitor (SLPI), antileukoproteinase 1	Proteinase inhibitor	1.5	20q12*
X87212	Cathepsin C (CTSC)	Proteinase	1.3	11q14.1-q14.3
	Cell–matrix and cell–cell interaction
U06643	Lectin galactoside-binding soluble 7 (LGALS7), galectin 7	Cell interaction	1.8	19q13.2
M76482	Desmoglein 3 (DSG3), 130-kDa pemphigus vulgaris antigen	Cell/cell, desmosome	1.7	18q12.1-q12.2
M18728	Carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6)	Cell interaction	1.6	19q13.2
U31201	Laminin gamma 2 (LAMC2)	Cell–matrix adhesion	1.5	1q25-q31
M69225	Bullous pemphigoid antigen 1 (BPAG1), BP240	Hemidesmosome	1.4	6p12-p11
U40434	Mesothelin (MSLN)	Cell adhesion	−1.6	16p13.3
	Signaling
U67784	Orphan G protein-coupled receptor (RDC1)	Receptor	2.3	2
L42176	Four and a half LIM domains 2 (FHL2)	Signal transduction	1.6	2q12-q14
U87460	G protein-coupled receptor 37 (GPR37)	Receptor	1.4	7q31*
U32944	Protein inhibitor of neuronal NOS (PIN), dynein light chain 1	Signal transduction	1.3	14q24*
U90878	PDZ and LIM domain 1 (PDLIM1)	Signal transduction	1.2	10q22-q26.3
U60975	Sortilin-related receptor (SORL1)	Receptor	−1.3	11q23.2-q24.2*
	Miscellaneous
AJ001421	(RER1)	Golgi/ER	−1.3	1
U32907	37-kDa leucine-rich repeat protein (P37NB)	—	2.1	7
X55740	5′-nucleotidase (NT5)	Nucleotide degradation	4.5	6q14-q21
U69611	A disintegrin and metalloproteinase domain 17 (ADAM17)	Multiple	1.3	2p25
M74491	ADP-ribosylation factor 3 (ARF3)	Multiple	−1.2	12q13
X05196	Aldolase C (ALDOC)	Glycolysis	1.4	17cen-q12
X05908	Annexin I (ANXA1), lipocortin, phospholipase A2 inhibitory protein	Exocytosis/anti-inflammation	1.5	9q12-q21.2
U22178	Beta-microseminoprotein (MSMB)	FSH-related	1.8	10q11.2*
M34376	Beta-microseminoprotein (MSMB)	FSH-related	1.8	10q11.2*
Y08682	Carnitine O-palmitoyltransferase 1 (CPT1B)	β-Oxidation	−1.5	22q13.33
L01664	Charcot-Leyden crystal protein (CLC), galectin 10	—	−2.7	19q13.1
D13639	Cyclin D2 (CCND2)	Cell cycle	2.7	12p13
U09770	Cysteine-rich intestinal protein 1 (CRIP1)	—	1.5	7q11.23
D42123	Cysteine-rich intestinal protein 2 (CRIP2)	—	1.5	14q32.3
L24774	Dodecenoyl-Coenzyme A delta isomerase (DCI)	β-Oxidation	−1.2	16p13.3
M60047	Heparin-binding protein (HBP17)	—	1.8	4
M13560	1a-associated invariant gamma-chain	Immune response	1.5	5q32
D42085	KIAA0095	—	−1.5	16q13
J04456	Lectin galactoside-binding soluble (LGALS1), galectin 1	Multiple	1.8	22q13.1
X00274	Major histocompatibility complex class II, DR alpha (HLA-DRA1)	Immune	1.7	6p21.3
D90070	Phorbol-12-myristate-13-acetate–induced protein 1 (PMAIP1)	—	1.6	18q21.31
L12760	Phosphoenolpyruvate carboxykinase 1 (PCK1)	Gluconeogenesis	−1.4	20q13.31
L05144	Phosphoenolpyruvate carboxykinase 1 (PCK1)	Gluconeogenesis	−1.6	20q13.31
K02574	Purine nucleoside phosphorylase (PNP)	Nucleoside degradation	1.5	14q13.1
X55954	Ribosomal protein L23 (RPL23)	Ribosome	1.2	17q
X57959	Ribosomal protein L7 (RPL7)	Ribosome	1.2	8q13.3
M31520	Ribosomal protein S24 (RPS24)	Ribosome	1.3	10q22-q23
Y07755	S100 calcium-binding protein A2 (S100A2)	Multiple	2.6	1q21
U67171	Selenoprotein W (SEPW1)	—	1.4	19q13.3
X76534	Transmembrane glycoprotein NMB (GPNMB)	—	2.5	7p15

Figure 3.

 

Hierarchical cluster analysis based on 19 samples and 471 candidate probe sets. (A) Gene and microarray cluster, (B) dendrogram of samples, (C) keratin-cluster, and (D) downregulated cluster.

Figure 4.

 

Gene expression profiles from MAS and dChip compared with real-time PCR in arbitrary values. dChip values were reduced 10-fold to fit the axis. Real-time PCR values are averages of triplicate analyses. Left axis (□), real-time PCR; right axis (○), dChip; (▵) MAS.

Figure 5.

 

Immunostaining of corneas with keratin 6 (A, D), keratin 13 (B, E), and vimentin (C, F). Top: KC samples; bottom: reference samples (tumors). Bar, 50 μm.

The authors thank Jesper Hjortdal, MD, for his contribution to the collection of cornea samples, and Casper Moller Frederiksen, MSc, for the dChip analysis.