August 2011
Volume 52, Issue 9
Free
Genetics  |   August 2011
Comparative Transcriptome and Network Biology Analyses Demonstrate Antiproliferative and Hyperapoptotic Phenotypes in Human Keratoconus Corneas
Author Affiliations & Notes
  • Matthias Macé
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
  • Stéphane D. Galiacy
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
  • Angélique Erraud
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
  • José Enrique Mejía
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
  • Heather Etchevers
    INSERM UMR910, Faculté de Médecine La Timone, Marseille, France;
  • Michèle Allouche
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
  • Laurence Desjardins
    Institut Curie, Paris, France; and
  • Patrick Calvas
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
    Service de Génétique Médicale,
  • François Malecaze
    From the INSERM (Institut National de la Santé et de la Recherche Médicale), U563, Toulouse, France;
    Université de Toulouse, Centre de Physiopathologie de Toulouse Purpan and Institut Biomédical de Toulouse, Toulouse, France;
    Service d'Ophtalmologie and Hôpital Purpan, Centre Hôpital Université (CHU) de Toulouse, Toulouse, France.
  • Corresponding author: Matthias Macé, INSERM U1043, Hôpital Purpan BP 3028, 31024 Toulouse, F-31300, France; [email protected]
Investigative Ophthalmology & Visual Science August 2011, Vol.52, 6181-6191. doi:https://doi.org/10.1167/iovs.10-70981
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Matthias Macé, Stéphane D. Galiacy, Angélique Erraud, José Enrique Mejía, Heather Etchevers, Michèle Allouche, Laurence Desjardins, Patrick Calvas, François Malecaze; Comparative Transcriptome and Network Biology Analyses Demonstrate Antiproliferative and Hyperapoptotic Phenotypes in Human Keratoconus Corneas. Invest. Ophthalmol. Vis. Sci. 2011;52(9):6181-6191. https://doi.org/10.1167/iovs.10-70981.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose.: To decipher the biological pathways involved in keratoconus pathophysiology by determining the patterns of differential gene expression between keratoconus and control corneas.

Methods.: RNA was extracted from surgically removed corneas of 10 keratoconus patients and from normal corneas of 10 control patients who had undergone enucleation of an eye for ocular melanoma. Several hundred thousand RNA transcripts were assessed using exon microarrays. Statistical comparison and identification of differentially regulated and differentially spliced RNA transcripts was performed by comparing keratoconus cases and controls. In addition, relevant biological pathways were identified by information extraction using network biology.

Results.: Eighty-seven genes showed significant differences in expression levels. Among these, 69 were downregulated in keratoconus patients, particularly partners of the transcription factor AP-1. The 18 overexpressed genes include mucins, keratins, and genes involved in fibroblast proliferation. In addition, 36 genes were shown to be differentially spliced, including 9 among those that were differentially expressed. Network biology and analysis using Gene Ontology descriptors suggest that many members of both groups belong to pathways of apoptosis and regulation of the balance between cellular differentiation and proliferation.

Conclusions.: This work constitutes the first genome-wide transcriptome analysis of keratoconus patient corneas that include all currently known genes and exons. Differential expression suggests that mechanisms of cell loss resulting from antiproliferative and hyperapoptotic phenotypes may be responsible for the pathogenesis of keratoconus. Array information, experimental design, raw intensities, and processed log2 ratios were deposited at the European Bioinformatic Institute's ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/). The accession number is E-MEXP-2777.

Keratoconus (KC) is a vision-threatening condition characterized by thinning and deformation of the cornea. It is one of the most common indications for corneal grafting in industrialized countries. The disease prevalence is approximately 1 in 2000, and familial aggregation, together with increased familial risk, suggests important genetic influences on its pathogenesis, but the etiology of KC is still poorly understood. Aside from genetic determinants, environmental stresses such as eye rubbing or atopy have been suggested as possible causes or aggravating factors in KC. 1 To date, several loci for familial KC have been described. 2 8 These have been mapped by genome-wide scans of varying resolutions to the chromosomal regions 2p24, 3p14-q13, 5q14.3-q21.1, and 16q22.3-q23.1, in familial studies, and to the chromosomal regions 4q31, 5q31, 9q34, 12p12, 14p11, 17q24, and 20q12, in affected-only linkage analyses. However, no genes have yet been identified as responsible for the development of most KC cases. Within these chromosomal regions, several candidate genes (COL6A1, SOD1, MMP9, MMP2, and COL8A1) have been excluded, 2,3,9,10 whereas mutations in VSX1 have been reported in a few KC patients. 11 In addition, one differential expression study pointed to AQP5 encoding aquaporin 5. 12 Taken together, these data suggest that KC is a complex disease involving multiple susceptibility loci. 6  
Holistic approaches integrating various “-omics” techniques are promising for discovering the molecular bases of genetic disorders with complex patterns of inheritance, such as KC. 13 Variation in gene expression is an important mechanism underlying the susceptibility to complex diseases, and it has been established that the steady state abundance of mRNA transcripts for many human genes is a highly heritable, quantitative phenotypic trait. 14 Recent technical advances for the profiling of virtually all human or mouse exons on genome-wide transcriptome microarrays 15,16 have made the comprehension of molecular disease mechanisms possible. 17 Thus far, most published analyses have assessed only the variations in levels of gene expression, without taking into account exon content. However, approximately 50% of all human genes are predicted to be alternatively spliced, 18 and disruptions in the balance of multiple transcript isoforms have been shown to be at play in conditions ranging from Alzheimer's disease to several types of cancer. 19,20 In this study, we therefore sought to analyze the KC corneal transcriptome in a comparison with normal corneas, to identify differentially expressed or spliced genes that would highlight the pathophysiological pathways involved in KC. Network biology allowed us to assign some these gene products to a small number of molecular cascades pertaining to proliferation, differentiation, and programmed cell death. 
Methods
Ethics Statement
This study was performed in accordance with French regulations and the Declaration of Helsinki. The samples were assigned a laboratory number and have remained anonymous throughout. The experimental protocol was approved by the relevant interregional ethics committee (CPP Sud-Ouest Outre-Mer No. 2). Written consent was obtained from all participants. 
Patients and Control Tissue Collection and Storage
Ten corneas were collected from nonrelated patients during a penetrating keratoplasty procedure for advanced KC in the Centre National de Référence du Kératocône (Toulouse, France). All KC eyes included in the study were stage III (Amsler-krumeich classification), with high refractive errors, severe loss of visual acuity, and absence of scarring. In addition all patients had contact lens intolerance, and none of them had worn contact lens since at least 6 months. Patients were 34 ± 10.3 years old; six were men and four were women. Ten control corneas were obtained in the Institut Curie (Paris, France) from unrelated patients who underwent enucleation for choroidal melanomas strictly localized to the posterior pole of the eye. These patients did not receive antineoplastic treatment before surgery. None was wearing contact lenses. The patients were 59 ± 10.1 years old; five were men and five were women. In addition, for each control cornea, anatomic or topographic abnormality was ruled out by careful examination. Excised corneal buttons were 8 mm in diameter. They were processed according to identical standard procedures in the two centers and were immediately stored in 1.5-mL microtubes in liquid nitrogen until RNA extraction. 
RNA Isolation from Whole Corneas
Corneas were transferred from liquid nitrogen to 500 μL butter (RLT; RNeasy Mini kit; Qiagen, Hilden, Germany)/β-mercaptoethanol (GE Healthcare, Pittsburgh, PA) solution, in a 2-mL microtube (Lysing Matrix D; MP Biomedicals, Irvine, CA) and kept on ice. This mix was then subjected to eight cycles of 20 seconds of shaking followed by 5 minutes of cooling at 4°C (FastPrep-24 System; MP Biomedicals). The supernatant was then retrieved, and total RNA was extracted and further purified (RNeasy Mini kit and RNase-Free DNase Set; Qiagen), according to the manufacturer's protocols. RNA quality was assessed on an RNA microarray (RNA 6000 Nano Chips; Agilent Technologies, Santa Clara, CA) on a bioanalyzer (model 2100; Agilent). RNA samples were immediately stored frozen at −80°C. RNAs with an RNA integrity number (RIN 21 ; Agilent) >9 were selected for further analysis (i.e., above the 8.5 threshold recommended by Agilent). RNA concentration and purity were determined immediately before reverse transcription through measurement of A260/A280 ratios with a spectrophotometer (NanoDrop ND-1000; NanoDrop Technologies, Wilmington, DE). 
Microarray Transcriptome Hybridization
We used an exon array platform (Human GeneChip Exon 1.0 ST Array; Affymetrix, Santa Clara, CA) designed to interrogate expression levels at both exon and gene levels, querying 1.4 million probesets. To minimize background and increase the sensitivity of the assay, ribosomal RNA was removed from 0.7 to 1 μg of total RNA from each preparation (RiboMinus Human/Mouse Transcriptome Isolation Kit; Invitrogen, Carlsbad, CA). The resulting RNA was then subjected to reverse transcription using random hexamers tagged with a T7 promoter sequence followed by second-strand cDNA synthesis using a DNA polymerase (GeneChip WT cDNA Synthesis Kit; Affymetrix). The resulting double-stranded cDNA was then used for amplification of antisense cRNA and cleaned (Gene Chip Sample Cleanup Module; Affymetrix). A second cycle of cDNA synthesis was then performed using random primers to reverse transcribe the cRNA into sense single-stranded DNA. This DNA was then fragmented, labeled, and hybridized to a human gene chip (Human Gene Chip Exon 1.0 ST Arrays; Affymetrix). Target labeling, array hybridization, washing, and staining were performed as described by the manufacturer (GeneChip Whole Transcript [WT] Sense Target Labeling; Affymetrix). Arrays were then hybridized, washed, and stained (GeneChip Hybridization, Wash and Stain Kit in a GeneChip Hybridization Oven 645 and a GeneChip Fluidics Station 450; Affymetrix). Arrays were then scanned (GeneChip Scanner 3000 7G; Affymetrix). 
Array information, experimental design, raw intensities and processed log2 ratios are all available through ArrayExpress (http://www.ebi.ac.uk/arrayexpress/). 
In Silico Data Filtering for Probes, Signal Normalization, and Summarization
Standard methods for outlier removal were used, including principal component analysis on the sample covariance matrix
( V = j = 1 n ( y i y ¯ ) ( y i y ¯ ) T / n where y ¯ = j = 1 n y i / n )
and hierarchical clustering by defining a dissimilarity between the expression signatures between two samples as 1 − |ρ|, ρ being the sample correlation 22 as implemented in the oneChannelGUI package 23 of the R programming language. A hierarchical clustering tree is thus a dendrogram representing the pairwise similarity structure between arrays. A cutoff value is given as a proportion of the tree depth to remove too divergent arrays as potential artifacts. 
After quantile normalization, 24 data from the 20 probes were fitted to a global model of expression and probe affinities (model = chip effecti + probe affinityj + ε for the jth probe on the ith array). Expression levels were summarized using the robust multichip average (RMA) model, 25 which is based on the dependence of the measured intensity on the amount of material (chip effect), the probe affinity, and a reading error (measurement error). Model fitting was performed using the fast median polish algorithm. 26 This expression summary provides an estimated value for the abundance of the transcript in the sample but does not provide a measurement of the reliability of this estimation. Exon arrays lack paired mismatch probes but instead comprise a separate pool of ∼25,000 background probes that allows the computation of a detection above background (DABG) score by matching those probes to members of the background with the same GC content. It is used to discard poorly performing probesets. In addition to this probe-level summary, gene-level summaries (expression levels averaged across probes) are calculated as median expression of probes that (1) are not multiply targeted and (2) hit an exon with all probes in the probeset. 
Mapping to Annotation
Probesets were mapped to the genome and to gene annotations using various X to Y functions included in the Exonmap R package 27 and a local MySQL instance of the X:MAP database. 28 This allowed discrimination between probesets hitting introns, transcripts, or genes as well as interrogation of known alternative splicings. 
Identification of Differentially Expressed Genes
Linear modeling and tests for differential expression, adjusted for multiple testing, were performed with the Limma R package. 29 The use of linear modeling allows the borrowing of information from all the transcripts to assist inference about each transcript individually. Briefly, Limma first adjusts a linear model over the systematic part of the data, intending to estimate its variability (lmfit function). Then, a contrast step allows the fitted coefficients to be compared regardless of their number. For single-channel microarrays (e.g., Affymetrix Exon), linear modeling is equivalent to ANOVA but with the fitting of a model for every gene. The fitted matrix and a contrast matrix are used to compute fold-changes (FCs) and t-statistics (makeContrasts function in Limma R). Change in expression (FC) was considered biologically relevant when the variation was twofold or greater (i.e., log2(FC)>1). Differential expression is then assessed by empiric Bayes statistics: the eBayes function is used to compute moderated t-statistics, moderated F-statistic, and log-odds of differential expression by empiric Bayes shrinkage of the standard errors toward a common value. Moderated means that the residual mean squares and degrees of freedom are moderated between probes. An adjustment for multiple testing was applied using the decideTests function. The Benjamini-Hochberg method for controlling the false discovery rate (FDR) 30 was used. 
The result was a list of significant probes within genes of interest. Last, we assessed whether for each of these genes all their probesets showed the same FC, thus reflecting gene-level differential expression. For this purpose, we computed for each gene the variance of the FC for its exonic probeset. This allowed us to filter differentially expressed genes, characterized by a low variance (i.e., <1). 
Identification of Differentially Spliced Variants
Detection of alternative splicing events was performed using two parallel approaches, the Splicing Index and MiDAS (microarray detection of alternative splicing). For both tests, given the much lower number of tests involved (equaling the number of probes per gene) compared with gene-level differential expression, P < 0.05 was applied. 
In the Splicing Index (SI) 15,31 method, the expression values of the probesets are first converted to log2 space. For each probeset examined, the expression value is subtracted from the mean expression value of all constitutive aligning probesets to create a constitutive corrected log2 expression difference. This difference was calculated for each individual in the study, using microarray data from that subject only, and was then used to calculate the mean expression difference in each of the patient and control groups. The probeset SI value was then derived by subtracting the KC group mean from the control sample group mean. This value represents the change in exon inclusion (dI). A t-test (two tailed, assuming unequal variance) of the means was performed for statistical significance. A dI of −1 indicated a twofold change in the expression of a probeset relative to the mean constitutive expression, with expression being higher in the KC patient group than in the control group. 
The MiDAS approach is a two-way ANOVA-based method measuring differences between the exon level and aggregating gene level signals, including an error term and possible interactions. 31 First, probe logarithmic intensity error (PLIER) normalization is performed. This generates both exon-level signals and gene-level estimates which are robust against exon-level anomalous signals across samples. Under the null hypothesis of no alternative splicing at an exon level, the expectation is to observe a constant difference between the exon and the corresponding gene across all the samples. 
Functional Association and Network Biology Analyses
The Cytoscape platform 32 was used as a network visualization and analysis tool for differentially expressed and spliced transcripts. We specifically used several plug-ins for interaction retrieval, network statistics, and Gene Ontology (GO) enrichment (below). 
Annotation and Interactome Characterization
To harvest protein–protein physical and functional interactions, we used the STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) database, which aggregates most of the available information on protein–protein interactions, organizing them by scoring and weighting. 33 The database is queried for interaction matrices derived from data and text mining, including experimental data and predicted interactions. STRING querying is characterized by a unique scoring framework based on benchmarks of the different types of associations against a common reference set. This results into a single confidence score per prediction. For each type of interaction, a different algorithm is used to provide a score (listed in Ref. 34). To recover the proteins belonging to given metabolic maps and so presumed to interact in the same metabolic pathways, we also interrogated the KEGG (Kyoto Encyclopedia of Genes and Genomes) database. 35 Networks were then represented by a cloud of nodes connected by edges. Node connectivity (number of edges of the node) was assessed using the Hubba plug-in for Cytoscape. 36  
Identification of Molecular Pathways
We used the BINGO plug-in from Cytoscape 37 to identify enriched GO terms 38 among the differentially expressed genes, as follows: When sampling K genes (set) out of R genes (an annotation reference set), we aimed to infer the probability that k or more of these genes belonged to a functional category shared by r of the R genes in the reference set. By performing a binomial test, sampling with replacement, we were able to provide an approximate P value. Only well-characterized genes (excluding hypothetical proteins) were included in the analysis. GO terms that were overrepresented in the analysis set were selected (five or more hits, binomial test Pc < 0.05 for the Benjamini-Hochberg [BH] correction 39 ). The enrichment score of each term cluster was computed as the geometric mean of each member's −Log(P). 
Reverse Transcription and Real-Time Polymerase Chain Reaction Amplification
In addition to the microarray experiment, quantitation of selected gene transcripts was performed by reverse transcription and real-time quantitative PCR amplification (RT-qPCR). Eight KC and 10 control corneas (of which five KC and five control corneas were shared with the microarray hybridizations) were used in this experiment. One microgram of each of the total RNA preparations was reverse transcribed into single-stranded cDNA using reverse transcriptase from a kit (SuperScript III; SuperScript VILO kit; Invitrogen). The following genes were selected to cover a range of FCs and gene locations within the interaction network studied: JUN, FOS, FOSB, BTG2, EGR1, MCL1, HSP90AA1, S100A6, MAT2A, ANO1, and KRT78. Three further genes, TBP, FBRS, and PIH1D1, were selected as standard baseline genes because they displayed some of the lowest differences between cases and controls in the microarray experiment. Relevant gene-specific PCR primers are listed in Supplementary Table S8. qPCR was performed (DNA Master SYBR Green I reagents in a LightCycler 480; Hoffmann-La Roche, Basel, Switzerland). Thermal cycling conditions were as follows: 95°C for 5 minutes (denaturation); 40 cycles at 95°C for 15 seconds, 60°C for 10 seconds (amplification), and 72°C for 20 seconds; and the melting curve: 95°C for 10 seconds, 70°C for 20 seconds, and 97°C for 0.1 second qPCR data were analyzed using the comparative CT method using geometric averaging of the three internal control genes. 40,41  
To assess the correlation between the FC levels from qPCR and from the microarrays (at both gene and probe levels), we computed Spearman's rank correlation coefficient (Rho, ρ) statistics that estimate a rank-based measure of association. 
Results
RNA Extraction and Data Quality Control
On average, 2.3 μg of total RNA was recovered from each of 20 individual whole corneas: 10 KC corneas and 10 control donor corneas. The quality control (QC) of microarray hybridization data, using exon and gene level signal densities (Supplementary Fig. S3), exon-level principal component analysis, and average-linkage hierarchical clustering (with 20% from the tree height as the cutoff) detected one obvious outlier sample among the controls, namely Ctlr7, that was excluded from the subsequent analysis of the data. 
Approximately 100 Transcripts Show Moderate Differential Expression between KC and Nonaffected Corneas
One hundred sixteen transcripts, represented by 794 probes, displayed significant differential expression levels at one or more probesets, with absolute log2(FC) > 1 and FDR-adjusted P < 0.05. Among these transcripts, 87 displayed a probeset FC variance low enough (<1) to be considered differentially expressed at the gene level. The most significant differential annotations are listed in Table 1, together with their FCs and levels of significance. Of note, most of the differently expressed transcripts meeting these statistical criteria were downregulated in KC (69 downregulated versus 18 upregulated genes). Interestingly, although considered to be a housekeeping gene, ACTB belongs to this list. A subset of downregulated genes—FOS, JUN, FOSB, MYC, and CDKN1A—are involved in cell cycle regulation and varied most significantly in expression between KC and control corneas. Figure 1 presents the relative distribution of transcript expression levels in KC and control corneas for overexpressed (Fig. 1A) and underexpressed (Fig. 1B) genes. 
Table 1.
 
Transcripts with Differentially Levels of Expression between KC and Control Corneas
Table 1.
 
Transcripts with Differentially Levels of Expression between KC and Control Corneas
HUGO Gene Symbol log2(FC) P Gene Name Change Ratio
KRT78 1.21 1.62 × 10−05 Keratin 78-keratin 5B 2.31
MUC4 0.49 7.46 × 10−06 Mucin 4 1.40
S100A6 0.27 7.92 × 10−06 S100 calcium binding protein A6 (calcicyclin) 1.21
ROR2 0.12 9.57 × 10−05 Receptor tyrosine Kinase 10-like orphan receptor 2 1.09
SQSTM1 −0.1 4.90 × 10−05 Sequestosome 1 0.93
INSIG1 −0.15 4.59 × 10−05 Insulin-induced gene 1 0.90
HSPB1 −0.16 4.87 × 10−05 Heat shock 27-kDa protein 1 0.90
NQO1 −0.2 1.51 × 10−05 NAD(P)H dehydrogenase, quinone 1 0.87
DDX3X −0.2 9.23 × 10−06 DEAD (ASP-GLU-ALA-ASP) box polypeptide 3, X-linked 0.87
KRT5 −0.24 6.86 × 10−07 Keratin 5A 0.85
HSP90AA1 −0.27 4.64 × 10−05 Heat shock protein 90-kDa alpha (cytosolic), class A member 1 0.83
FTH1 −0.32 3.45 × 10−07 Ferritin, heavy polypeptide 1 0.80
RNF39 −0.35 1.10 × 10−05 Ring finger protein 39 0.78
DDX5 −0.39 1.33 × 10−05 Dead (Asp-Glu-Ala-Asp) box polypeptide 5 0.76
ACTB −0.45 3.12 × 10−06 Actin, Beta 0.73
SAT1 −0.59 1.56 × 10−06 Spermidine/Spermine N1-Acetyltransferase 0.66
DNAJB1 −0.59 1.65 × 10−05 Dnaj (Hsp40) Homolog, Subfamily B, Member 1 0.66
ZNF750 −0.61 1.54 × 10−05 Hypothetical Protein Loc79755 0.66
THBS1 −0.63 6.60 × 10−06 Thrombospondin 1 0.65
MYC −0.64 9.27 × 10−06 V-Myc myelocytomatosis viral oncogene homolog (avian) 0.64
HES1 −0.71 1.13 × 10−05 Hairy and enhancer of split 1, (Drosophila) 0.61
CDKN1A −0.76 2.79 × 10−06 Cyclin-dependent kinase inhibitor 1a (P21, Cip1) 0.59
MAT2A −0.83 3.16 × 10−06 Methionine adenosyltransferase Ii, Alpha 0.56
ZFP36L1 −0.87 6.85 × 10−06 Zinc finger protein 36, C3h Typ.10-Like 1 0.55
MCL1 −1.01 9.20 × 10−06 Myeloid cell leukemia sequence 1 (Bcl2-related) 0.50
ID1 −1.08 3.59 × 10−05 Inhibitor of DNA binding 1, dominant negative helix-loop-helix protein 0.47
ODC1 −1.08 3.37 × 10−06 Ornithine decarboxylase 1 0.47
SLC20A1 −1.08 3.56 × 10−06 Solute carrier family 20 (phosphate transporter), Member 1 0.47
NUAK2 −1.16 9.61 × 10−05 Nuak family, Snf1-like kinase, 2 0.45
BTG2 −1.52 1.24 × 10−05 Btg family, member 2 0.35
UBC (RPS27A) −1.84 1.83 × 10−05 Ubiquitin C 0.28
ZFP36 −1.92 1.02 × 10−05 Zinc finger protein 36, C3h type, Homolog (Mouse) 0.26
DUSP1 −2.19 2.02 × 10−05 Dual specificity phosphatase 1 0.22
JUN −3.13 9.98 × 10−05 V-Jun sarcoma virus 17 oncogene homolog (Avian) 0.11
FOSB −3.69 1.87 × 10−05 Fbj Murine osteosarcoma viral oncogene homolog B 0.08
FOS −4.2 2.93 × 10−05 V-Fos Fbj murine osteosarcoma viral oncogene homolog 0.05
Figure 1.
 
Levels of expression of the differentially expressed genes. The boxplots represent dispersion among samples. KC, keratoconjunctivitis samples; ctrl, control samples. (A) Transcripts underexpressed in KC, ordered by decreasing absolute FC. (B) Transcripts overexpressed in KC, ordered by increasing absolute FC.
Figure 1.
 
Levels of expression of the differentially expressed genes. The boxplots represent dispersion among samples. KC, keratoconjunctivitis samples; ctrl, control samples. (A) Transcripts underexpressed in KC, ordered by decreasing absolute FC. (B) Transcripts overexpressed in KC, ordered by increasing absolute FC.
Only 18 genes displayed upregulation by this analysis. Among these are several genes involved in the extracellular matrix and the epithelial cell cytoskeleton (PTCH2, KRT5, KRT78, and LYPD3), in the stress response (HSP90AA1 and ALDH1A3), or encoding mucins (MUC4 and MUC16). 
Markers considered to be melanoma-specific 42 44 were absent from the differentially expressed transcripts, consistent with our screening of the control corneas. 
In parallel, a weighted correlation network analysis using the WGCNA R package 45 identified a module of 134 transcripts (Supplementary Table S4) that are co-expressed in correlation with the disease status across the microarray samples (P = 9 × 10−4) and globally decreased in KC (average log2(FC) = −0.68). The correlation network and individual gene analyses are convergent, in that 33 transcripts in the set of 69 individually underexpressed transcripts belong also to the WGCNA module. 
KC Patients Express Different Splice Isoforms of Certain Transcripts
We next set out to identify genes that were differentially spliced in the comparison between the KC and control groups. Thirty-six candidates were identified with a threshold of P = 0.05 with either the Splicing Index or MiDAS methods (Table 2, and see Methods for details on significance levels). MUC4, encoding a membrane-tethered sialomucin of the ocular surface epithelium 46 is remarkable among these, because it displays profound changes in transcript regulation at both the gene (overexpression in KC) and splice regulation levels. 
Table 2.
 
Genes Differentially Spliced between Keratoconus and Control Corneas
Table 2.
 
Genes Differentially Spliced between Keratoconus and Control Corneas
HUGO Gene Symbol SI SI P MiDAS P Description
PTGS2 1.58 2.50 × 10−05 3.93 × 10−02 Prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase)
FOSL2 1.24 8.35 × 10−05 3.71 × 10−02 FOS-like antigen 2
MYL6 1.35 1.86 × 10−04 3.80 × 10−02 Myosin, light chain 6, alkali, smooth muscle and non-muscle
HSPB1 2.72 2.32 × 10−06 1.54 × 10−02 Heat shock 27kDa protein 1
BAG1 1.18 5.88 × 10−04 4.82 × 10−02 BCL2-associated athanogene
ALDH3A1 1.16 8.26 × 10−05 3.46 × 10−02 Aldehyde dehydrogenase 3 family, memberA1
DDX5 1.40 6.13 × 10−05 3.30 × 10−02 DEAD (Asp-Glu-Ala-Asp) box polypeptide 5
ODC1 3.00 8.16 × 10−05 3.26 × 10−02 Ornithine decarboxylase 1
SDC1 1.74 2.65 × 10−04 3.60 × 10−02 Syndecan 1
SLC2A1 1.19 2.25 × 10−04 4.06 × 10−02 Solute carrier family 2 (facilitated glucose transporter), member 1
CDKN1A 1.04 2.99 × 10−05 2.77 × 10−02 Cyclin-dependent kinase inhibitor 1A (p21, Cip1)
ID1 1.31 2.26 × 10−05 2.34 × 10−02 Inhibitor of DNA binding 1, dominant negative helix-loop-helix protein
SAT1 1.30 3.23 × 10−06 2.21 × 10−02 Spermidine/spermine N1-acetyltransferase 1
LDLR 1.72 1.14 × 10−06 1.86 × 10−02 Low density lipoprotein receptor (familial hypercholesterolemia)
H3F3B 1.23 8.80 × 10−05 3.64 × 10−02 H3 histone, family 3B (H3.3B)
TPT1 1.82 1.88 × 10−04 3.37 × 10−02 Tumor protein, translationally-controlled 1
1.77 2.82 × 10−04 4.21 × 10−02 -
RRAS2 1.54 6.83 × 10−04 4.04 × 10−02 Related RAS viral (r-ras) oncogene homolog 2
BHLHB2/BHLHE40 1.15 4.97 × 10−07 1.70 × 10−02 Basic helix-loop-helix domain containing, class B, 2//class E basic helix-loop-helix protein 40
ANXA1 2.80 2.80 × 10−04 3.39 × 10−02 Annexin A1
GLUL 1.10 1.97 × 10−04 4.36 × 10−02 Glutamate 10-ammonia ligase (glutamine synthetase)
CTSL2 1.09 4.41 × 10−04 4.34 × 10−02 Cathepsin L2
JDP2 1.53 4.19 × 10−04 4.61 × 10−02 Jun dimerization protein 2
MUC4 −2.38 2.31 × 10−04 3.56 × 10−02 Mucin 4, cell surface associated
KIAA1754/ITPRIP 1.38 5.19 × 10−04 4.78 × 10−02 Inositol 1,4,5-triphosphate receptor-interacting protein
RPS27A/UBC 2.52 9.01 × 10−07 1.41 × 10−02 Ubiquitin C
2.38 1.90 × 10−07 1.20 × 10−02 Ubiquitin C
NUAK2 1.48 4.37 × 10−05 3.10 × 10−02 NUAK family, SNF1-like kinase, 2
WEE1 1.23 2.18 × 10−04 4.51 × 10−02 WEE1 homolog (S. pombe)
SERTAD3 1.46 3.36 × 10−05 3.06 × 10−02 SERTA domain containing 3
MAT2A 1.31 9.77 × 10−07 1.44 × 10−02 Methionine adenosyltransferase II, alpha
EFNA1 1.30 4.66 × 10−06 2.36 × 10−02 Ephrin-A1
OVOL1 1.18 3.77 × 10−05 3.38 × 10−02 Ovo-like 1 (Drosophila)
ACTG1 1.80 2.19 × 10−04 4.04 × 10−02 Actin, gamma 1
TACSTD2 1.49 7.17 × 10−05 3.35 × 10−02 Tumor-associated calcium signal transducer 2
INSIG1 1.39 7.39 × 10−05 3.26 × 10−02 Insulin induced gene 1
PTMA 1.41 4.21 × 10−05 3.20 × 10−02 Prothymosin, alpha
CD55 2.56 2.02 × 10−05 1.89 × 10−02 CD55 molecule, decay accelerating factor for complement (Cromer blood group)
Network Biology and GO Annotations
An interaction network was inferred to test whether underexpressed genes were involved in one or more biological signaling pathways. This network was obtained after a protein–protein interaction matrix was retrieved from a high-confidence STRING database query (score ≥0.7). The representation of weighted protein interactions from STRING provides a high-level view of functional linkage, enhancing the analysis of modularity in biological processes. Without the addition of further molecular partners, we were able to cluster most of the both differentially expressed and differentially spliced proteins into a single, highly connected network (Fig. 2). A core of those genes seems central to this network, as they are the most connected nodes. In decreasing order of connectivity, they are JUN, MYC, FOS, PTGS2, CDKN1A, ODC1, HSP90AA1, ALDH3A1, ANXA1, and NQO1
Figure 2.
 
Interaction network showing links between differentially expressed and differentially spliced genes. The network includes interactions retrieved from the STRING interactome database. HGNC (HUGO Gene Nomenclature Committee) official names are given. The colors of interaction partners describe the modifications we observed between cases and controls in the dataset. Red: downregulated genes; blue: upregulated genes; yellow: differentially spliced genes; green: upregulated and differentially spliced genes; and orange: downregulated and differentially spliced genes. Colors of edges between partner nodes show the type of interaction. Blue line: binding; green line: activation; yellow line: modified expression; and pink line: posttranslational modification. For interactions inferred from text-mining (gray), the level of confidence is proportional to line thickness. Unlinked genes are represented by three rows at the top of the figure. First row: genes that would be linked by adding one intermediary connecting node; second row: genes that would be linked by adding two intermediary connecting nodes; third row: other genes.
Figure 2.
 
Interaction network showing links between differentially expressed and differentially spliced genes. The network includes interactions retrieved from the STRING interactome database. HGNC (HUGO Gene Nomenclature Committee) official names are given. The colors of interaction partners describe the modifications we observed between cases and controls in the dataset. Red: downregulated genes; blue: upregulated genes; yellow: differentially spliced genes; green: upregulated and differentially spliced genes; and orange: downregulated and differentially spliced genes. Colors of edges between partner nodes show the type of interaction. Blue line: binding; green line: activation; yellow line: modified expression; and pink line: posttranslational modification. For interactions inferred from text-mining (gray), the level of confidence is proportional to line thickness. Unlinked genes are represented by three rows at the top of the figure. First row: genes that would be linked by adding one intermediary connecting node; second row: genes that would be linked by adding two intermediary connecting nodes; third row: other genes.
After loading this network into Cytoscape, the BINGO plug-in was used to further investigate whether specific GOs or pathways were overrepresented in the whole network. Within the first cluster of GO terms relative to biological function (enrichment score = 3.080), we filtered out several redundant terms (Supplementary Table S5), mainly within categories related to apoptosis regulation. On the other hand, within clusters 3, 4, and 5 (scores 2.43, 2.22, and 2.09, respectively), there were many cell-cycle/proliferation-oriented categories. More than half of the genes had GO terms within these categories and corresponding GO term P values, after correction, were below 10−6, meaning these categories were significantly overrepresented relative to their frequency in a randomized sample of expressed transcripts. Further examination showed that KC patients had significantly lower expression of genes that fell into a limited number of KEGG signaling pathways: ErbB (CDKN1A, JUN, and MYC), MAP kinases (FOS, DUSP1, JUN, HSPB1, and MYC), and focal adhesion (ACTG1, ACTB, CDKN1A, JUN, and THBS1) molecular cascades. 
In addition, we mapped the FCs in expression between cases and controls (Supplementary Table S6) over the loci identified in previously published linkage analyses. 2 8 In agreement with the genetic heterogeneity of these studies and their difficulty in reproducing each other's results, no obvious candidate genes for KC were distinguishable, as their FC ranged from −0.45 to +0.58. We also scrutinized the previously hypothesized candidate genes for KC (COL6A1, SOD1, MMP9, MMP2, and AQP5). None of them showed significance at the Bonferroni-corrected threshold of 10−6
Validation of Gene Microarray Data
Relative quantitative real-time polymerase chain reaction (qPCR) was conducted as an independent technique to validate the expression level changes in the microarray experiment. Results in Figure 3 show a good level of correlation between the three empiric assessment of expression as shown by rho values ∼0.9. Even considering as a replicate the samples used only for the qPCR experiment, the same levels of correlation are observed (data not shown). 
Figure 3.
 
qPCR experiment. Results are represented as a histogram, grouped by gene. For each gene, three results are presented. White, gene level expression estimated by averaging microarray probesets; gray: the level of expression at the microarray probeset corresponding to the region amplified in the qPCR experiment; black: the level of expression as estimated in the qPCR (primer pair sequences and location provided in Supplementary Table S8). The pairwise levels of correlation between the three estimated is also given as Spearman's rho statistics.
Figure 3.
 
qPCR experiment. Results are represented as a histogram, grouped by gene. For each gene, three results are presented. White, gene level expression estimated by averaging microarray probesets; gray: the level of expression at the microarray probeset corresponding to the region amplified in the qPCR experiment; black: the level of expression as estimated in the qPCR (primer pair sequences and location provided in Supplementary Table S8). The pairwise levels of correlation between the three estimated is also given as Spearman's rho statistics.
Discussion
Using full human transcriptomic microarray, our study is the first comparison of expression of all known genes between control and KC whole corneas. The microarray chosen is a multitarget array including several probesets per exon and four neighbor replicates per probeset, allowing a better confidence in gene-level expression estimates and as such also provides the possibility of exploring alternative RNA splicing. As a matter of fact, quantitative RT-PCR validated the microarray data, even though this last method reflects the expression of only one portion of the whole transcript (i.e., the amplicon). Moreover, the advantage of using network biology in transcriptome analysis is to make use of available limited sampling and to combine it with extant knowledge (interactome) for enhancing meaningful extraction, the global pathway being more meaningful in a first approximation than individual gene expression. Previous comparative transcriptomic studies were performed on human KC tissue such as corneal epithelium, 47 in vivo 48 and in vitro 49 keratocytes. Two studies used restricted microarray chips targeting 5600 genes with a limited number of probesets, 47,48 and one study directly targeted the expression of 164 apoptotic genes. 49 However, none of these studies replicate the results of the others, perhaps due to a limited number of corneal samples, as well as the limited probeset distribution and redundancy on microarray chips used in those studies compared with currently used microarrays (e.g., the Affymetrix Exon array). 
Our own study on human KC tissue was conducted on whole corneas. As controls, we used corneas from patients who underwent enucleation for posterior pole (i.e., choroidal) melanoma. They were considered to be a proper control since (1) collection was achieved in conditions similar to those of KC corneas excluding previously treated subjects; (2) the anterior segment of the eye was free of neoplastic extensions; and (3) melanoma extension is blood borne, 50 cornea is avascular, and melanoma transmission was not observed after grafting cornea from a melanoma donor. 51 Consistently, we did not observe overexpression of any melanoma-specific transcript. 
In this context, our aim was to reassess the genetic component of the KC, taking into account the following hypotheses: a condition characterized by genetic heterogeneity, with various loci linked in different populations, and a complex phenotypic trait whose transmission appears to deviate from a Mendelian model. Because transcriptomics have helped identify molecular pathways and alternatively spliced variants involved in other complex diseases (e.g., in oncology 15,20,52 and autoimmunity 53 ), we have transposed this approach to KC. The interplay of both experimental and predicted interactions of the annotation database we used (STRING), is supposed to offer a more accurate approximation of the interactome than GO alone. Nevertheless, this analysis infers statistical links between KC phenotype and gene expression and requires experimental validation, perhaps in animal models. We identified mice mutants for 29 of the differentially expressed genes and 27 of the differentially spliced ones (Supplementary Table S7) in the Mouse Genome Informatics database (http://www.informatics.jax.org/ Jackson Laboratory, Bar Harbor, ME). Some of them display corneal clinical phenotypes: The Jun mutant shows eye opacity and increased incidence of corneal inflammation, significant in the light of a recent study also implicating a downstream JUN effector, JNK2, in abnormal corneal barrier response to dry eyes. 54 The THBS1 (thrombospondin 1) mutant displays abnormal corneal epithelial morphology, and the Ldlr (low density lipoprotein receptor) mutant also has dry eyes. However, their relevance remains to be tested, especially in human disease. JUN is a human proto-oncogene implicated in carcinogenesis, whereas LDLR mutations lead to autosomal dominant hypercholesterolemia. 
The majority (69/87) of the genes highlighted were downregulated in KC corneas. They include several genes globally involved in cellular proliferation and the prevention of differentiation (e.g., the AP-1 transcription factor partners FOS, JUN, and FOSB) 55 or MYC, required for proliferation or stem cell mobilization. 56 Confirming previous studies, we did not find any significant change regarding putative candidate genes previously proposed (MMP2, MMP9, COL6A1, SOD1, and AQ5). Another original finding is that the β-actin ACTB is downregulated, thus confirming that, at least in pathologic conditions, it cannot be considered a good reference gene for expression in cornea. GO terms enrichment analysis demonstrated, independent of the algorithms used, that these downregulated genes are more widely involved in so-called (cellular) developmental processes or cell differentiation. First, we observed a redundancy of GO terms in the first three annotation clusters (GO:0032502, developmental process; GO:0048869, cellular developmental process; and GO:0030154 cell differentiation). Their congruence strengthens the conclusion that some KC corneal cells are maintained in a less differentiated, proliferative state. 57,58 Simultaneously, decreased protection from apoptosis may be implied by the terms GO:0043066 (negative regulation of apoptosis) or GO:0048468 (cell development). The “gene population background”–based enrichment analysis from DAVID's web-based annotation tool, 59 which uses GO terms and the Kyoto Encyclopedia of Genes and Genomes (KEGG 60 ), as well as GO pruning and filtering of WikiPathways using GO-Elite 61 gave similar results. The weighted correlation network analysis showed significant co-expression of modules containing intermediaries consistent with the highlighted pathways. Finally, GO enrichment analysis gave roughly the same results when applied to the P < 0.01 and P < 0.05 differentially expressed gene lists. Moreover, the qPCR results show that for all genes selected for validation, the direction and magnitude of changes were consistent with the results obtained from the microarray analysis. 
Finally, we failed to replicate previous differential expression experiments on KC versus normal corneas 47 49 : None of the differentially expressed genes found in these studies overlapped each other and with our results. Nevertheless, we performed a network biology analysis combining the highlighted genes from these and our studies, as well as three genes involved in communication between the epithelium and the stroma: HGF, KGF, and EGF 62,63 (Supplementary Fig. S9). We found that although we did not highlight the same genes, the same biological pathways seem to be involved. The dataset from Stachs et al. 48 comes from freshly isolated keratocytes, while the one of Nielsen et al. 47 comes from fresh epithelial cells. Our own dataset composed of both epithelial and stromal components is linked with both datasets, showing that these different studies are congruent. In addition, the links observed between the majority of our highlighted genes and Nielsen et al.'s ones by means of EGF, KGF, and EGF, considered to be involved in epithelium/stroma signaling, 63 are a further argument that there is probably an interplay between both cellular layers in KC. 
Our results implicate both known and novel pathways as perhaps playing a key role in the pathophysiology of KC. It is noteworthy to mention here that many of the selected genes have elsewhere been associated with cancer. Nevertheless, they are also primarily involved in developmental processes and cell proliferation. Thus, is not surprising that a change in mRNA expression could be associated with a disease suspected to derive from cell loss. In fact, apoptosis has been observed in KC corneas by direct confocal microscopy and histology showing decreased cell densities in the three corneal layers, 58,64 as well as biochemical or expression studies. 9,65 68 This is in full agreement with the overrepresentation of GO term annotation in our dataset. The primary cause of apoptosis in KC remains unknown, but it may be a secondary consequence of another molecular defect. Several reports have suggested a decreased resistance to environmental aggressions by hypersensitivity to oxidative stress in cultivated keratocytes from KC corneas, 69 as well as in vivo. 57 Cultured KC fibroblasts also may have an inherent, hypersensitive response to oxidative stressors that involves mitochondrial damage and the expression of intermediate filaments such as vimentin or tenascin or the proinflammatory TGF-β and IL-1. 70 Apoptosis in KC may also result from mechanical trauma or be secondary to the production of cytokines after atopy. 
At this point, the cellular population affected by these network alterations remains uncertain. This lack of hypothesis led us to perform, in a first approach, a full-thickness corneal study. The recovered amounts of mRNAs were largely in favor of epithelial cells, and this could have biased the results and consequently their interpretation. Nevertheless, one can imagine that both epithelial cells and keratocytes are key players. Keratocytes are specialized fibroblasts, derived from neural crest mesenchyme in common with the cartilaginous stromal layer behind the eyeball, 71 that produce collagen fibers and proteoglycans constituting the corneal stroma. Throughout life, most keratocytes are in a quiescent state. 72 By the end of eye development, a keratocyte network, interconnected through dendritic processes, is established. Keratocyte apoptosis, either of quiescent or actively dividing cells, is a process of great interest for corneal growth and remodeling. Previous reports have speculated that keratocytes may be involved in KC development through increased catabolism, either misregulation of metalloprotease activity 67 or a modification of collagen subtype composition. 73 We suggest here the novel hypothesis that keratocyte implication would be mediated by a tissue-specific misregulation of apoptosis that may be due to less redundant antiapoptotic pathways in the cornea. In a healthy cornea, programmed cell death is a rare occurrence, but immediately after an injury, keratocytes directly below the injury site undergo apoptosis when the basal membrane is broken. In the following steps of the healing and scarring process, this cell loss is counteracted by mitoses among the remaining adjacent keratocytes. 62 In KC corneas, impairment of apoptotic signaling pathways on minor corneal injuries sustained through life may affect appropriate keratocyte proliferation, itself necessary for the recovery of the initial normal cell density. Keratocyte-restricted deficiencies in the regulation of apoptosis may lead to the gene expression differences we have observed in this study. This could provide an explanation for the loss of cell density and corneal thinning observed in KC. Interestingly, a recent study underlined the role of TWIST2, a bHLH transcription factor, in keratocyte proliferation in mouse leading to a corneal thinning. 74 The epithelium, with cells that are continuously renewed, may also be affected by the antiproliferative and hyperapoptotic phenotypes. The 18 overexpressed genes we observed to be significantly differentially expressed included KRT78, ROR2, S100A6, MUC4, and KRT78 is a structural protein of epithelial cells whose cross-linking helps withstand mechanical and chemical stresses. 75 ROR2 encodes a nuclear orphan receptor in the noncanonical Wnt pathway that appears to trigger, among other effects, the maintenance and proliferation of stem cells. 76 S100A6 (calcyclin) is involved in the fibroblast cell cycle 77 and overexpressed during wound healing after corneal injury. In addition, sialomucin (MUC4) is expressed by the corneal epithelium, which raises the question of how epithelial cells may also be involved in the pathogenesis of KC. We found increased expression of the whole transcript population of MUC4 in KC corneas. The apical location of mucins could suggest their general involvement in mechanisms of response to epithelial damage. 78 In addition, among mucins, MUC4 displays tissue-specific expression patterns—notably, a conjunctival-type expression pattern is observed on the corneal surface in limbal stem cell deficiency. 79 Both mucin-4 (the product of the MUC4 gene) and calcyclin are implicated in corneal protection and wound healing by the constitution of the lacrimal film. 80,81 Mutations in other mucin genes are already implicated in human pathology, notably the dry eye syndromes 82 (Supplementary Table S6). Alternatively sialomucin is also an activator partner of the epidermal growth factor ErbB2 receptor, potentially implicating its overexpression in growth factor signaling pathways leading to either differentiation or increased cell proliferation. 78 Each of these overexpressed transcripts therefore could be either pathogenic or simply deregulated as a consequence of other causal molecular mechanisms. Overexpression of MUC4 may be a protective response or could be causative in KC, and a corneal-specific mouse overexpression model would be welcome to resolve this question. 
Another differentially spliced molecule is annexin A1 (ANXA1). This calcium-dependent phospholipid-binding protein has no allelic variants described, regulates phospholipase A2 activity, and promotes membrane fusion. It has been shown to be upregulated (×1.5) in KC corneal epithelial cells relative to normal epithelial cells. 47 ANXA1 is located within the chromosomal interval 9q21, which has been suggestively linked to KC using nonparametric linkage. 83 However, it is the only differentially expressed gene that showed consensus with the chromosomal regions linked to KC so far. This observation confirms that expression analysis is an interesting complementary approach to linkage analyses, and that our having included the keratocyte population from whole KC versus normal corneas not only found this difference but has implicated broad functional pathways in addition to individual molecules. 
Altogether these results cannot exclude that, whatever the pathways involved, KC could result from a distortion of the cross-talk between epithelial and stromal cells of the cornea. In the future, more precise, layer specific expression studies should be conducted. Such expression results may be combined with the latest possibilities for genome-wide association studies to identify genomic variants that correlate with both expression and KC phenotype. 
Supplementary Materials
Figure sf01, PDF - Figure sf01, PDF 
Figure sf02, PDF - Figure sf02, PDF 
Table st1, DOC - Table st1, DOC 
Table st2, DOC - Table st2, DOC 
Table st3, DOC - Table st3, DOC 
Table st4, DOC - Table st4, DOC 
Table st5, DOC - Table st5, DOC 
Table st6, DOC - Table st6, DOC 
Table st7, DOC - Table st7, DOC 
Footnotes
 Supported by the Rétina France Foundation and the French Ministère de la Santé.
Footnotes
 Disclosure: M. Macé, None; S.D. Galiacy, None; A. Erraud, None; J.E. Mejía, None; H. Etchevers, None; M. Allouche, None; L. Desjardins, None; P. Calvas, None; F. Malecaze, None
The authors thank Nathalie Marsaud (Genotoul Biopuces, Toulouse Biochip Facility) for help and advice about the experimental protocol and Didier Laborie and the Bioinfo-Genotoul team at the Toulouse Bioinformatics Facility. 
References
Karseras AG Ruben M . Aetiology of keratoconus. Br J Ophthalmol. 1976;60:522–525. [CrossRef] [PubMed]
Brancati F Valente E Sarkozy A Feher J . A locus for autosomal dominant keratoconus maps to human chromosome 3p14–q13. J Med Genet. 2004;41:188–192. [CrossRef] [PubMed]
Fullerton J Paprocki P Foote S Mackey DA Williamson R Forrest S . Identity-by-descent approach to gene localisation in eight individuals affected by keratoconus from north-west Tasmania, Australia. Hum Genet. 2002;110:462–470. [CrossRef] [PubMed]
Hughes AE Dash DP Jackson AJ Frazer DG Silvestri G . Familial keratoconus with cataract: linkage to the long arm of chromosome 15 and exclusion of candidate genes. Invest Ophthalmol Vis Sci. 2003;44:5063–5066. [CrossRef] [PubMed]
Hutchings H Ginisty H Le Gallo M . Identification of a new locus for isolated familial keratoconus at 2p24. BMJ. 2005;42:88.
Li X Rabinowitz YS Tang YG . Two-stage genome-wide linkage scan in keratoconus sib pair families. Invest Ophthalmol Vis Sci. 2006;47:3791–3795. [CrossRef] [PubMed]
Tang YG Rabinowitz YS Taylor KD . Genomewide linkage scan in a multigeneration Caucasian pedigree identifies a novel locus for keratoconus on chromosome 5q14.3-q21.1. Genet Med. 2005;7:397–405. [CrossRef] [PubMed]
Tyynismaa H Sistonen P Tuupanen S . A locus for autosomal dominant keratoconus: linkage to 16q22. 3-q23. 1 in Finnish families. Invest Ophthalmol Vis Sci. 2002;43:3160. [PubMed]
Rabinowitz YS Maumenee I Lundergan M . Molecular genetic analysis in autosomal dominant keratoconus. Cornea. 1992;11:302–308. [CrossRef] [PubMed]
Udar N Atilano S Brown D Holguin B . SOD1: a candidate gene for keratoconus. Invest Ophthalmol Vis Sci. 2006;47:3345–3351. [CrossRef] [PubMed]
Bisceglia L Ciaschetti M De Bonis P . VSX1 mutational analysis in a series of Italian patients affected by keratoconus: detection of a novel mutation. Invest Ophthalmol Vis Sci. 2005;46:39–45. [CrossRef] [PubMed]
Rabinowitz YS Dong L Wistow G . Gene expression profile studies of human keratoconus cornea for NEIBank: a novel cornea-expressed gene and the absence of transcripts for aquaporin 5. Invest Ophthalmol Vis Sci. 2005;46:1239–1246. [CrossRef] [PubMed]
Ostrowski J Wyrwicz LS . Integrating genomics, proteomics and bioinformatics in translational studies of molecular medicine. Expert Rev Mol Diagn. 2009;9:623–630. [CrossRef] [PubMed]
Monks SA Leonardson A Zhu H . Genetic inheritance of gene expression in human cell lines. Am J Hum Genet. 2004;75:1094–1105. [CrossRef] [PubMed]
Gardina PJ Clark TA Shimada B . Alternative splicing and differential gene expression in colon cancer detected by a whole genome exon array. BMC Genomics. 2006;7:325. [CrossRef] [PubMed]
Licatalosi DD Darnell RB . RNA processing and its regulation: global insights into biological networks. Nat Rev Genet. 2010;11:75–87. [CrossRef] [PubMed]
Clark TA Schweitzer AC Chen TX . Discovery of tissue-specific exons using comprehensive human exon microarrays. Genome Biol. 2007;8:R64. [CrossRef] [PubMed]
Johnson JM Castle J Garrett-Engele P . Genome-wide survey of human alternative pre-mRNA splicing with exon junction microarrays. Science. 2003;302:2141–2144. [CrossRef] [PubMed]
Faustino NA Cooper TA . Pre-mRNA splicing and human disease. Genes Dev. 2003;17:419–437. [CrossRef] [PubMed]
Venables JP . Unbalanced alternative splicing and its significance in cancer. BioEssays. 2006;28:378–386. [CrossRef] [PubMed]
Schroeder A Mueller O Stocker S . The RIN: an RNA integrity number for assigning integrity values to RNA measurements. BMC Mol Biol. 2006;7:3. [CrossRef] [PubMed]
McLachlan GJ Do KA Ambroise C . Analyzing microarray gene expression data. Wiley Series in Probability and Statistics: Applied Probability and Statistics Section Series. Vol. 422. Hoboken, NJ: Wiley-IEEE. 2004:320.
Sanges R Cordero F Calogero RA . oneChannelGUI: a graphical interface to Bioconductor tools, designed for life scientists who are not familiar with R language. Bioinformatics. 2007;23:3406–3408. [CrossRef] [PubMed]
Bolstad BM Irizarry Ra Astrand M Speed TP . A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics. 2003;19:185–193. [CrossRef] [PubMed]
Irizarry RA Bolstad BM Collin F Cope LM Hobbs B Speed TP . Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. 2003;31:e15. [CrossRef] [PubMed]
Mosteller F Tukey J . Data Analysis and Regression. Reading, MA: Addison-Wesley; 1977.
Okoniewski MJ Miller CJ . Comprehensive analysis of Affymetrix Exon Arrays using BioConductor. PLoS Comput Biol. 2008;4:e6. [CrossRef] [PubMed]
Tai TYT Damani MR Vo R . Keratoconus associated with corneal stromal amyloid deposition containing TGFBIp. Cornea. 2009;28:589–593. [CrossRef] [PubMed]
Smyth GK . Limma: linear models for microarray data. In: Gentleman R Carey V Huber W Irizarry R Dudoit S eds. Bioinformatics and Computational Biology Solutions using R and Bioconductor. New York: Springer; 2005:397–420.
Benjamini Y Hochberg Y . Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B. 1995;57:289–300.
Srinivasan K Shiue L Hayes JD . Detection and measurement of alternative splicing using splicing-sensitive microarrays. Methods. 2005;37:345–359. [CrossRef] [PubMed]
Cline MS Smoot M Cerami E . Integration of biological networks and gene expression data using Cytoscape. Nat Protoc. 2007;2:2366–2382. [CrossRef] [PubMed]
Jensen LJ Kuhn M Stark M . STRING 8: a global view on proteins and their functional interactions in 630 organisms. Nucleic Acids Res. 2009;37:D412–416. [CrossRef] [PubMed]
von Mering C Huynen M Jaeggi D Schmidt S Bork P Snel B . STRING: a database of predicted functional associations between proteins. Nucleic Acids Res. 2003;31:258–261. [CrossRef] [PubMed]
Kanehisa M Goto S Kawashima S Okuno Y Hattori M . The KEGG resource for deciphering the genome. Nucleic Acids Res. 2004;32:D277–280. [CrossRef] [PubMed]
Lin CY Chin CH Wu HH Chen SH Ho CW Ko MT . Hubba. hub objects analyzer–a framework of interactome hubs identification for network biology. Nucleic Acids Res. 2008;36:W438–W443. [CrossRef] [PubMed]
Maere S Heymans K Kuiper M . BiNGO: a Cytoscape plugin to assess overrepresentation of gene ontology categories in biological networks. Bioinformatics. 2005;21:3448–3449. [CrossRef] [PubMed]
Ashburner M Ball CA Blake JA . Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000;25:25–29. [CrossRef] [PubMed]
Gilbert PB . A modified false discovery rate multiple-comparisons procedure for discrete data, applied to human immunodeficiency virus genetics. J R Stat Soc Ser C. 2005;54:143–158. [CrossRef]
Livak KJ Flood SJ Marmaro J Giusti W Deetz K . Oligonucleotides with fluorescent dyes at opposite ends provide a quenched probe system useful for detecting PCR product and nucleic acid hybridization. PCR Methods Applications. 1995;4:357–362. [CrossRef] [PubMed]
Vandesompele J De Preter K Pattyn F . Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol. 2002;3:research0034.0031–0034.0011. [CrossRef]
Böni R Heizmann CW Doguoglu A . Ca(2+)-binding proteins S100A6 and S100B in primary cutaneous melanoma. J Cutan Pathol. 1997;24:76–80. [CrossRef] [PubMed]
Indrei A Cianga P Florea ID Haba D Foia L Cianga CM . A rare case of double recurrent choroidal melanoma, with distinctive immunohistochemical features. Rom J Morphol Embryol. 2010;51:187–193. [PubMed]
Kan-Mitchell J Liggett PE Taylor CR . Differential S100 beta expression in choroidal and skin melanomas: quantitation by the polymerase chain reaction. Invest Ophthalmol Vis Sci. 1993;34:3366–3375. [PubMed]
Langfelder P Horvath S . WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. [CrossRef] [PubMed]
Govindarajan B Gipson IK . Membrane-tethered mucins have multiple functions on the ocular surface. Exp Eye Res. 90:655–663. [CrossRef] [PubMed]
Nielsen K Birkenkamp-Demtröder K Ehlers N Orntoft TF . Identification of differentially expressed genes in keratoconus epithelium analyzed on microarrays. Invest Ophthalmol Vis Sci. 2003;44:2466–2476. [CrossRef] [PubMed]
Stachs O Bochert A Gerber T Koczan D Thiessen HJ Guthoff RF . The extracellular matrix structure in keratoconus (in German). Ophthalmologe. 2004;101:384–389. [CrossRef] [PubMed]
Ha NT Nakayasu K Murakami A Ishidoh K Kanai A . Microarray analysis identified differentially expressed genes in keratocytes from keratoconus patients. Curr Eye Res. 2004;28:373–379. [CrossRef] [PubMed]
Zakka KA Foos RY Omphroy CA Straatsma BR . Malignant melanoma. analysis of an autopsy population. Ophthalmology. 1980;87:549–556. [CrossRef] [PubMed]
Harrison DA Hodge DO Bourne WM . Outcome of corneal grafting with donor tissue from eyes with primary choroidal melanomas. a retrospective cohort comparison. Arch Ophthalmol. 1995;113:753–756. [CrossRef] [PubMed]
Dong L Jensen R De Rienzo A . Differentially expressed alternatively spliced genes in malignant pleural mesothelioma identified using massively parallel transcriptome sequencing. BMC Med Genet. 2009;10:149. [CrossRef] [PubMed]
Gillett A Maratou K . Alternative splicing and transcriptome profiling of experimental autoimmune encephalomyelitis using genome-wide exon arrays. PloS One. 2009;4:e7773. [CrossRef] [PubMed]
De Paiva CS Pangelinan SB Chang E . Essential role for c-Jun N-terminal kinase 2 in corneal epithelial response to desiccating stress. Arch Ophthalmol. 2009;127:1625–1631. [CrossRef] [PubMed]
Eferl R Wagner EF . AP-1: a double-edged sword in tumorigenesis. Nat Rev Cancer. 2003;3:859–868. [CrossRef] [PubMed]
Arnold I Watt FM . c-Myc activation in transgenic mouse epidermis results in mobilization of stem cells and differentiation of their progeny. Curr Biol. 2001;11:558–568. [CrossRef] [PubMed]
Kenney MC Chwa M Atilano SR . Increased levels of catalase and cathepsin V/L2 but decreased TIMP-1 in keratoconus corneas: evidence that oxidative stress plays a role in this disorder. Invest Ophthalmol Vis Sci. 2005;46:823–832. [CrossRef] [PubMed]
Uçakhan OO Kanpolat A Ylmaz N Ozkan M . In vivo confocal microscopy findings in keratoconus. Eye Contact lens. 2006;32:183–191. [CrossRef] [PubMed]
Huang DW Sherman BT Lempicki Ra . Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 2009;4:44–57. [CrossRef] [PubMed]
Kanehisa M Goto S . KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. 2000;28:27–30. [CrossRef] [PubMed]
Pico AR Kelder T van Iersel MP Hanspers K Conklin BR Evelo C . WikiPathways: pathway editing for the people. PLoS Biol. 2008;6:e184. [CrossRef] [PubMed]
Wilson SE He YG Weng J . Epithelial injury induces keratocyte apoptosis: hypothesized role for the interleukin-1 system in the modulation of corneal tissue organization and wound healing. Exp Eye Res. 1996;62:325–327. [CrossRef] [PubMed]
Wilson SE Liu JJ Mohan RR . Stromal-epithelial interactions in the cornea. Prog Retin Eye Res. 1999;18:293–309. [CrossRef] [PubMed]
Ali Javadi M Kanavi MR Mahdavi M . Comparison of keratocyte density between keratoconus, post-laser in situ keratomileusis keratectasia, and uncomplicated post-laser in situ keratomileusis cases. a confocal scan study. Cornea. 2009;28:774–779. [CrossRef] [PubMed]
Kaldawy RM Wagner J Ching S Seigel GM . Evidence of apoptotic cell death in keratoconus. Cornea. 2002;21:206–209. [CrossRef] [PubMed]
Kim WJ Rabinowitz YS Meisler DM Wilson SE . Keratocyte apoptosis associated with keratoconus. Exp Eye Res. 1999;69:475–481. [CrossRef] [PubMed]
Matthews FJ Cook SD Majid MA Dick AD Smith VA . Changes in the balance of the tissue inhibitor of matrix metalloproteinases (TIMPs)-1 and -3 may promote keratocyte apoptosis in keratoconus. Exp Eye Res. 2007;84:1125–1134. [CrossRef] [PubMed]
Nielsen K Vorum H Fagerholm P . Proteome profiling of corneal epithelium and identification of marker proteins for keratoconus, a pilot study. Exp Eye Res. 2006;82:201–209. [CrossRef] [PubMed]
Chwa M Atilano S Hertzog D . Hypersensitive response to oxidative stress in keratoconus corneal fibroblasts. Invest Ophthalmol Vis Sci. 2008;49:4361–4369. [CrossRef] [PubMed]
Zhou L Yue BY Twining SS Sugar J Feder RS . Expression of wound healing and stress-related proteins in keratoconus corneas. Curr Eye Res. 1996;15:1124–1131. [CrossRef] [PubMed]
Le Lièvre CS Le Douarin NM . Mesenchymal derivatives of the neural crest: analysis of chimaeric quail and chick embryos. J Embryol Exp Morphol. 1975;34:125–154. [PubMed]
West-Mays JA Dwivedi DJ . The keratocyte: corneal stromal cell with variable repair phenotypes. Int J Biochem Cell Biol. 2006;38:1625–1631. [CrossRef] [PubMed]
Newsome DA Foidart JM Hassell JR Krachmer JH Rodrigues MM Katz SI . Detection of specific collagen types in normal and keratoconus corneas. Invest Ophthalmol Vis Sci. 1981;20:738–750. [PubMed]
Weaving L Mihelec M Storen R . Twist2: role in corneal stromal keratocyte proliferation and corneal thickness. Invest Ophthalmol Vis Sci. 2010;51:5561–5570. [CrossRef] [PubMed]
Coulombe PA Omary MB . ‘Hard’ and ‘soft’ principles defining the structure, function and regulation of keratin intermediate filaments. Curr Opin Cell Biol. 2002;14:110–122. [CrossRef] [PubMed]
Willert K Brown JD Danenberg E . Wnt proteins are lipid-modified and can act as stem cell growth factors. Nature. 2003;423:448–452. [CrossRef] [PubMed]
Breen EC Tang K . Calcyclin (S100A6) regulates pulmonary fibroblast proliferation, morphology, and cytoskeletal organization in vitro. J Cell Biochem. 2003;88:848–854. [CrossRef] [PubMed]
Carraway K Ramsauer V Haq B . Cell signaling through membrane mucins. Bioessays. 2002;25:66–71. [CrossRef]
Pauklin M Kakkassery V Steuhl KP Meller D . Expression of membrane-associated mucins in limbal stem cell deficiency and after transplantation of cultivated limbal epithelium. Curr Eye Res. 2009;34:221–230. [CrossRef] [PubMed]
Bazan HE Allan G Bazan NG . Enhanced expression of the growth-regulated calcyclin gene during corneal wound healing. Exp Eye Res. 1992;55:173–177. [CrossRef] [PubMed]
Pflugfelder SC Liu Z Monroy D . Detection of sialomucin complex (MUC4) in human ocular surface epithelium and tear fluid. Invest Ophthalmol Vis Sci. 2000;41:1316–1326. [PubMed]
Paulsen F Langer G Hoffmann W Berry M . Human lacrimal gland mucins. Cell Tissue Res. 2004;316:167–177. [CrossRef] [PubMed]
Bisceglia L De Bonis P Pizzicoli C . Linkage analysis in keratoconus: replication of locus 5q21.2 and identification of other suggestive Loci. Invest Ophthalmol Vis Sci. 2009;50:1081–1086. [CrossRef] [PubMed]
Figure 1.
 
Levels of expression of the differentially expressed genes. The boxplots represent dispersion among samples. KC, keratoconjunctivitis samples; ctrl, control samples. (A) Transcripts underexpressed in KC, ordered by decreasing absolute FC. (B) Transcripts overexpressed in KC, ordered by increasing absolute FC.
Figure 1.
 
Levels of expression of the differentially expressed genes. The boxplots represent dispersion among samples. KC, keratoconjunctivitis samples; ctrl, control samples. (A) Transcripts underexpressed in KC, ordered by decreasing absolute FC. (B) Transcripts overexpressed in KC, ordered by increasing absolute FC.
Figure 2.
 
Interaction network showing links between differentially expressed and differentially spliced genes. The network includes interactions retrieved from the STRING interactome database. HGNC (HUGO Gene Nomenclature Committee) official names are given. The colors of interaction partners describe the modifications we observed between cases and controls in the dataset. Red: downregulated genes; blue: upregulated genes; yellow: differentially spliced genes; green: upregulated and differentially spliced genes; and orange: downregulated and differentially spliced genes. Colors of edges between partner nodes show the type of interaction. Blue line: binding; green line: activation; yellow line: modified expression; and pink line: posttranslational modification. For interactions inferred from text-mining (gray), the level of confidence is proportional to line thickness. Unlinked genes are represented by three rows at the top of the figure. First row: genes that would be linked by adding one intermediary connecting node; second row: genes that would be linked by adding two intermediary connecting nodes; third row: other genes.
Figure 2.
 
Interaction network showing links between differentially expressed and differentially spliced genes. The network includes interactions retrieved from the STRING interactome database. HGNC (HUGO Gene Nomenclature Committee) official names are given. The colors of interaction partners describe the modifications we observed between cases and controls in the dataset. Red: downregulated genes; blue: upregulated genes; yellow: differentially spliced genes; green: upregulated and differentially spliced genes; and orange: downregulated and differentially spliced genes. Colors of edges between partner nodes show the type of interaction. Blue line: binding; green line: activation; yellow line: modified expression; and pink line: posttranslational modification. For interactions inferred from text-mining (gray), the level of confidence is proportional to line thickness. Unlinked genes are represented by three rows at the top of the figure. First row: genes that would be linked by adding one intermediary connecting node; second row: genes that would be linked by adding two intermediary connecting nodes; third row: other genes.
Figure 3.
 
qPCR experiment. Results are represented as a histogram, grouped by gene. For each gene, three results are presented. White, gene level expression estimated by averaging microarray probesets; gray: the level of expression at the microarray probeset corresponding to the region amplified in the qPCR experiment; black: the level of expression as estimated in the qPCR (primer pair sequences and location provided in Supplementary Table S8). The pairwise levels of correlation between the three estimated is also given as Spearman's rho statistics.
Figure 3.
 
qPCR experiment. Results are represented as a histogram, grouped by gene. For each gene, three results are presented. White, gene level expression estimated by averaging microarray probesets; gray: the level of expression at the microarray probeset corresponding to the region amplified in the qPCR experiment; black: the level of expression as estimated in the qPCR (primer pair sequences and location provided in Supplementary Table S8). The pairwise levels of correlation between the three estimated is also given as Spearman's rho statistics.
Table 1.
 
Transcripts with Differentially Levels of Expression between KC and Control Corneas
Table 1.
 
Transcripts with Differentially Levels of Expression between KC and Control Corneas
HUGO Gene Symbol log2(FC) P Gene Name Change Ratio
KRT78 1.21 1.62 × 10−05 Keratin 78-keratin 5B 2.31
MUC4 0.49 7.46 × 10−06 Mucin 4 1.40
S100A6 0.27 7.92 × 10−06 S100 calcium binding protein A6 (calcicyclin) 1.21
ROR2 0.12 9.57 × 10−05 Receptor tyrosine Kinase 10-like orphan receptor 2 1.09
SQSTM1 −0.1 4.90 × 10−05 Sequestosome 1 0.93
INSIG1 −0.15 4.59 × 10−05 Insulin-induced gene 1 0.90
HSPB1 −0.16 4.87 × 10−05 Heat shock 27-kDa protein 1 0.90
NQO1 −0.2 1.51 × 10−05 NAD(P)H dehydrogenase, quinone 1 0.87
DDX3X −0.2 9.23 × 10−06 DEAD (ASP-GLU-ALA-ASP) box polypeptide 3, X-linked 0.87
KRT5 −0.24 6.86 × 10−07 Keratin 5A 0.85
HSP90AA1 −0.27 4.64 × 10−05 Heat shock protein 90-kDa alpha (cytosolic), class A member 1 0.83
FTH1 −0.32 3.45 × 10−07 Ferritin, heavy polypeptide 1 0.80
RNF39 −0.35 1.10 × 10−05 Ring finger protein 39 0.78
DDX5 −0.39 1.33 × 10−05 Dead (Asp-Glu-Ala-Asp) box polypeptide 5 0.76
ACTB −0.45 3.12 × 10−06 Actin, Beta 0.73
SAT1 −0.59 1.56 × 10−06 Spermidine/Spermine N1-Acetyltransferase 0.66
DNAJB1 −0.59 1.65 × 10−05 Dnaj (Hsp40) Homolog, Subfamily B, Member 1 0.66
ZNF750 −0.61 1.54 × 10−05 Hypothetical Protein Loc79755 0.66
THBS1 −0.63 6.60 × 10−06 Thrombospondin 1 0.65
MYC −0.64 9.27 × 10−06 V-Myc myelocytomatosis viral oncogene homolog (avian) 0.64
HES1 −0.71 1.13 × 10−05 Hairy and enhancer of split 1, (Drosophila) 0.61
CDKN1A −0.76 2.79 × 10−06 Cyclin-dependent kinase inhibitor 1a (P21, Cip1) 0.59
MAT2A −0.83 3.16 × 10−06 Methionine adenosyltransferase Ii, Alpha 0.56
ZFP36L1 −0.87 6.85 × 10−06 Zinc finger protein 36, C3h Typ.10-Like 1 0.55
MCL1 −1.01 9.20 × 10−06 Myeloid cell leukemia sequence 1 (Bcl2-related) 0.50
ID1 −1.08 3.59 × 10−05 Inhibitor of DNA binding 1, dominant negative helix-loop-helix protein 0.47
ODC1 −1.08 3.37 × 10−06 Ornithine decarboxylase 1 0.47
SLC20A1 −1.08 3.56 × 10−06 Solute carrier family 20 (phosphate transporter), Member 1 0.47
NUAK2 −1.16 9.61 × 10−05 Nuak family, Snf1-like kinase, 2 0.45
BTG2 −1.52 1.24 × 10−05 Btg family, member 2 0.35
UBC (RPS27A) −1.84 1.83 × 10−05 Ubiquitin C 0.28
ZFP36 −1.92 1.02 × 10−05 Zinc finger protein 36, C3h type, Homolog (Mouse) 0.26
DUSP1 −2.19 2.02 × 10−05 Dual specificity phosphatase 1 0.22
JUN −3.13 9.98 × 10−05 V-Jun sarcoma virus 17 oncogene homolog (Avian) 0.11
FOSB −3.69 1.87 × 10−05 Fbj Murine osteosarcoma viral oncogene homolog B 0.08
FOS −4.2 2.93 × 10−05 V-Fos Fbj murine osteosarcoma viral oncogene homolog 0.05
Table 2.
 
Genes Differentially Spliced between Keratoconus and Control Corneas
Table 2.
 
Genes Differentially Spliced between Keratoconus and Control Corneas
HUGO Gene Symbol SI SI P MiDAS P Description
PTGS2 1.58 2.50 × 10−05 3.93 × 10−02 Prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase)
FOSL2 1.24 8.35 × 10−05 3.71 × 10−02 FOS-like antigen 2
MYL6 1.35 1.86 × 10−04 3.80 × 10−02 Myosin, light chain 6, alkali, smooth muscle and non-muscle
HSPB1 2.72 2.32 × 10−06 1.54 × 10−02 Heat shock 27kDa protein 1
BAG1 1.18 5.88 × 10−04 4.82 × 10−02 BCL2-associated athanogene
ALDH3A1 1.16 8.26 × 10−05 3.46 × 10−02 Aldehyde dehydrogenase 3 family, memberA1
DDX5 1.40 6.13 × 10−05 3.30 × 10−02 DEAD (Asp-Glu-Ala-Asp) box polypeptide 5
ODC1 3.00 8.16 × 10−05 3.26 × 10−02 Ornithine decarboxylase 1
SDC1 1.74 2.65 × 10−04 3.60 × 10−02 Syndecan 1
SLC2A1 1.19 2.25 × 10−04 4.06 × 10−02 Solute carrier family 2 (facilitated glucose transporter), member 1
CDKN1A 1.04 2.99 × 10−05 2.77 × 10−02 Cyclin-dependent kinase inhibitor 1A (p21, Cip1)
ID1 1.31 2.26 × 10−05 2.34 × 10−02 Inhibitor of DNA binding 1, dominant negative helix-loop-helix protein
SAT1 1.30 3.23 × 10−06 2.21 × 10−02 Spermidine/spermine N1-acetyltransferase 1
LDLR 1.72 1.14 × 10−06 1.86 × 10−02 Low density lipoprotein receptor (familial hypercholesterolemia)
H3F3B 1.23 8.80 × 10−05 3.64 × 10−02 H3 histone, family 3B (H3.3B)
TPT1 1.82 1.88 × 10−04 3.37 × 10−02 Tumor protein, translationally-controlled 1
1.77 2.82 × 10−04 4.21 × 10−02 -
RRAS2 1.54 6.83 × 10−04 4.04 × 10−02 Related RAS viral (r-ras) oncogene homolog 2
BHLHB2/BHLHE40 1.15 4.97 × 10−07 1.70 × 10−02 Basic helix-loop-helix domain containing, class B, 2//class E basic helix-loop-helix protein 40
ANXA1 2.80 2.80 × 10−04 3.39 × 10−02 Annexin A1
GLUL 1.10 1.97 × 10−04 4.36 × 10−02 Glutamate 10-ammonia ligase (glutamine synthetase)
CTSL2 1.09 4.41 × 10−04 4.34 × 10−02 Cathepsin L2
JDP2 1.53 4.19 × 10−04 4.61 × 10−02 Jun dimerization protein 2
MUC4 −2.38 2.31 × 10−04 3.56 × 10−02 Mucin 4, cell surface associated
KIAA1754/ITPRIP 1.38 5.19 × 10−04 4.78 × 10−02 Inositol 1,4,5-triphosphate receptor-interacting protein
RPS27A/UBC 2.52 9.01 × 10−07 1.41 × 10−02 Ubiquitin C
2.38 1.90 × 10−07 1.20 × 10−02 Ubiquitin C
NUAK2 1.48 4.37 × 10−05 3.10 × 10−02 NUAK family, SNF1-like kinase, 2
WEE1 1.23 2.18 × 10−04 4.51 × 10−02 WEE1 homolog (S. pombe)
SERTAD3 1.46 3.36 × 10−05 3.06 × 10−02 SERTA domain containing 3
MAT2A 1.31 9.77 × 10−07 1.44 × 10−02 Methionine adenosyltransferase II, alpha
EFNA1 1.30 4.66 × 10−06 2.36 × 10−02 Ephrin-A1
OVOL1 1.18 3.77 × 10−05 3.38 × 10−02 Ovo-like 1 (Drosophila)
ACTG1 1.80 2.19 × 10−04 4.04 × 10−02 Actin, gamma 1
TACSTD2 1.49 7.17 × 10−05 3.35 × 10−02 Tumor-associated calcium signal transducer 2
INSIG1 1.39 7.39 × 10−05 3.26 × 10−02 Insulin induced gene 1
PTMA 1.41 4.21 × 10−05 3.20 × 10−02 Prothymosin, alpha
CD55 2.56 2.02 × 10−05 1.89 × 10−02 CD55 molecule, decay accelerating factor for complement (Cromer blood group)
Figure sf01, PDF
Figure sf02, PDF
Table st1, DOC
Table st2, DOC
Table st3, DOC
Table st4, DOC
Table st5, DOC
Table st6, DOC
Table st7, DOC
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×