Identification of Differentially Expressed Long Noncoding RNAs in Bladder Cancer

This article is featured in Highlights of This Issue, p. 5145

Bladder cancer (BC) is the fourth commonest male malignancy and one of the most expensive human cancers to manage (1, 2). The majority of tumors are urothelial carcinoma in histologic type. Clinicopathological data suggest that urothelial cancers are best stratified into two distinct phenotypes characterized by low- and high-grade cellular differentiation. Low-grade urothelial cancers frequently have mutations of FGFR3 and STAG2, partial deletion of chromosome 9 (3), and rarely progress to muscle invasion or metastases following endoscopic resection (4). In contrast, high-grade BCs are aggressive tumors that may present before or after the onset of muscle invasion (5). These poorly differentiated urothelial cancers have widespread chromosomal instability, multiple mutations, and are best characterized by deficiency of p53-mediated pathways (3). High-grade tumors share many molecular alterations regardless of stage and respond poorly to chemotherapy (6).

Although many reports detail genetic events in urothelial cancer, alterations of epigenetic gene regulation are also important in this disease (7). In general, epigenetic alterations reflect urothelial cancer disease phenotypes and associated genetic events (8, 9). For example, Wolff and colleagues reported that high-grade invasive urothelial cancers had widespread aberrant hypermethylation, whereas low-grade noninvasive tumors had regional hypomethylation (10). In general, epigenetic events reflect chromosomal changes with the disease and can be used as predictor of disease progression (11). With regard to miRNA, low-grade urothelial cancers are characterized by loss of expression of many species, including miRs99a/100 that target FGFR3 (12, 13). In contrast, high-grade tumors have upregulation of many miRs, including miR21 that targets the p53 pathway (14) and miR129 (15).

Epigenetic gene regulation may occur directly or indirectly through noncoding RNA (ncRNA) species. These are classified according to size, function, and gene location (16). Most reports of ncRNA in cancer have focused upon miRNAs, which modulate gene expression through targeting complementary seed sequences within mRNA 3′ UTRs (17). The role of miRNAs in disease varies with cell type, biologic stress, species conservation, and processing (9, 18). To date, few data have reported the role of long noncoding RNA (lncRNA) in cancer and little is known of their function (16, 19). Exceptions include lncRNAs clearly implicated in carcinogenesis (such as MALAT-1 or HOTAIR, reviewed in Spizzo and colleagues; ref. 20). LncRNAs are annotated according to gene location (intergenic, exonic, intronic, anti-sense etc.) and size (21). The GENCODE consortium recently annotated 9,277 long ncRNA genes, corresponding to 14,880 transcripts (22). In contrast with protein coding genes, the majority of reported lncRNAs had few gene exons, were located in the nucleus and chromatin, and many were not conserved between species. As multiple reports suggest that lncRNAs are important in human cancer (16, 19, 23) and preliminary data suggest this is true for urothelial cancer (24), we hypothesized that they play a role in urothelial carcinogenesis. To test this hypothesis, we profiled the expression of 9,351 lncRNAs in normal and malignant urothelial samples. We identified global changes that reflect the disease phenotype and differ to patterns of mRNA alteration. We performed explorative analysis of selected species and identified those with potential prognostic roles.

We analyzed two patient cohorts in this work (Table 1). For RNA microarray profiling, we examined urothelial samples from 57 patients with urothelial cancer and 26 disease free controls (urothelium from radical prostatectomy cases). For validation, we examined a second cohort of n = 138 samples from patients with urothelial cancer. Tumors were obtained following trans-urethral resection of the first cancer within a patient, naïve to Bacillus Calmette—Guerin, chemotherapy, or other pretreatments, and classified using the 2004 WHO/ISUP criteria. Tumors were selected to reflect the urothelial cancer spectrum: low-grade non-muscle invasive (NMI), high-grade NMI, and invasive urothelial cancer. All tissues were fresh frozen in liquid nitrogen. Histologic confirmation was obtained before use. We also analyzed the human telomerase reverse transcriptase (hTERT) immortalized normal human urothelial (NHU) cells maintained in keratinocyte serum-free medium containing bovine pituitary extract, EGF (Invitrogen), and cholera toxin. The hTERT NHU cells were obtained as a gift directly from Prof. M.A. Knowles, University of Leeds (Leeds, United Kingdom), and tested by this laboratory using comparative genomic hybridization and mutational analysis of CDKN2A and TP53 as described (25). The cells were not authenticated in Sheffield before use.

From each sample, 10 × 10 μm thick sections were microdissected to extract normal and malignant urothelial cells (>90% content). Total RNA was extracted using the mirVana kit (Ambion) according to manufacturer's protocol (26), and measured using a 2100 Bioanalyzer (Agilent). The expression of long ncRNAs and protein coding mRNAs was determined using microarrays (NCode Human Non-coding RNA Microarrays, Invitrogen). Each sample was prepared according to Agilent's one-color microarray protocol. Briefly, 200 ng RNA samples were mixed with spike in control RNA (Agilent), labeled with cyanine 3-CTP (Low Input Quick Amp Labeling Kit, one-color, Agilent), and hybridized to the microarray (Gene Expression Hybridization Kit, Agilent) according to the manufacturer's protocol. After washing (Gene Expression Wash Buffer 1 and 2, Agilent), the microarray slides were scanned using the Agilent Microarray Scanner platform (High Resolution Microarray Scanner C) and raw probe fluorescence extracted. The microarray data are deposited with Gene Expression Omnibus (GSE55433). The NCode microarray contains duplicate probes to 17,112 ncRNAs and 22,074 mRNAs. The lncRNAs were identified by various strategies without annotation (27). To annotate the array, we converted all probes to hg 19.0 loci (using LiftOver, UCSC genome browser), matched to target gene, structure and the Gencode v7.0 annotation (22), and the nearest CpG island/differentially methylated region (28). For microarray validation, individual lncRNA expression was measured in triplicate using qRT-PCR. Total RNA was transcribed using random hexamer primers (Applied Biosystems) and the High Capacity Reverse Transcription Kit (Applied Biosystems) before diluting 10-fold in nuclease-free water. cDNA was mixed with SyberGreen MasterMix (Applied Biosystems) and PCR primers, and analyzed using the ABI 7900HT real-time PCR system. cDNA expression was calculated using ΔC_t values normalized by subtraction of the mean of TEGT and HSP90AB1 expression (as reference genes) and fold change (FC = 2^−ΔΔCt) calculated (29).

We investigated the protein coding potential of selected lncRNAs using geneID (30) and CPC against UniProt Reference Clusters (31). For both, negative scores suggest low protein coding potential.

We selected representative lncRNAs for further analysis and modulated expression using custom siRNA. All experiments were performed in triplicate using hTERT immortalized NHU cells at 70% confluence transfected with siRNAs (synthesized with LifeTechnologies BLOCK-iT RNAi designer) specific to lncRNAs and controls (scrambled RNA sequence). Cells were seeded into 12-well dishes and incubated for 3 hours before transfection with 80 nmol/L RNAi using 2 μL Lipofectamine RNAiMAX (Invitrogen) per well. Transfection efficiency was determined 72 hours later by qRT-PCR (Applied Biosystems).

The growth characteristics of siRNA-transfected cells were analyzed for cell-cycle regulation (propidium iodide flow cytometry), apoptosis (caspase-3 activation), and proliferation (MTT assay). For cell-cycle analysis, cells were harvested after 24, 48, and 72 hours of knockdown, centrifuged (3 minutes at 1,400 g) and washed in PBS. Cells (1 × 10⁶) were resuspended in 1 mL PBS, mixed with 3 mL ice cold absolute ethanol, and fixed over night at 4°C before washing twice in PBS. Cells were then mixed with 5 μL RNase solution (2 mg/mL) and 300 μL propidium iodide solution (50 μg/mL), and incubated over night at 4°C. The next day, cells were analyzed on a FACSCalibur flow cytometry analyzer (Becton Dickinson). For apoptosis, we determined caspase-3 activity in 3 × 10⁵ cells, 24 hours after transfection using the CaspGLOW Fluorescein Active Caspase-3 Staining Kit (MBL). Briefly, we resuspended the cells in 300 μL PBS, added 1 μL FITC-DEVD-FMK, and incubated them for 1 hour at 37°C. After pelleting the cells and washing them in washing buffer, they were resuspended in 300 μL wash buffer and analyzed on a FACSCalibur flow cytometry analyzer (Becton Dickinson). For proliferation, cells were seeded into 96-well plates following siRNA-mediated gene knockdown. The culture medium was replaced with 50 μL MTT-PBS solution (3 mg MTT/1 mL PBS) 24, 48, and 72 hours after knockdown, respectively. Cells were incubated at 37°C for 3 hours before 200 μL MTT dissolvent (DMSO) was added to each well. MTT absorbance was measured at 570 nm using a microplate reader.

mRNA expression was determined in transfected cells using HG-U133 Plus 2.0 microarrays (Affymetrix, Cal.; ref. 32). Briefly, RNA was prepared with the Affymetrix protocol and annealed to an oligo-d(T) primer with a T7 polymerase-binding site. cDNA was generated using superscript II and Escherichia coli DNA ligase and polymerase I, before the reaction was completed with T4 DNA polymerase and EDTA. Amplified cDNA was cleaned, biotin-labeled, and fragmented, before hybridizing to the microarray for 16 hours (45°C in a rotating oven at 60 rpm). After washing and staining, the arrays were scanned (GC3000 scanner) and data processed using Gene Chip Operating System software. mRNA expression was determined using Microarray Analysis Suite 5 (Affymetrix) and defined as expressed (perfect match probeset intensity greater than mismatch intensity) or absent (mismatch probeset intensity greater or equal to perfect match intensity). ANOVA analysis was then performed using Partek Genomic Suite 6.5 β and differentially expressed transcripts were defined as ≥2 relative fold change using a cutoff of P < 0.05. qRT-PCR was used to confirm the expression of individual mRNAs of interested in the second patient cohort.

Raw intensity values for each microarray probe were used to calculate lncRNA expression. The NCode microarray includes multiple probes for each target RNA. We excluded RNAs whose targeting probes were non-concordant (defined as signal ratio less than 0.5 or more than 2.0) in all the arrays. Normalization was achieved according to the 1-color default normalization of Agilent's Gene Spring software (version 7), by dividing each raw intensity value by the median of the chip and the median expression of that RNA in all samples. Changes in lncRNA expression and statistical significance were calculated and illustrated using Volcano plots. Significant differences in expression between malignant versus normal urothelium, or between urothelial cancer phenotypes were defined using the Significance Analysis of Microarray software (33) as a t test P value of <0.05, a false discovery rate of <0.01, and an expression FC of ±2. Hierarchical clustering was performed using Cluster 3.0 and visualized in Tree view (Eisen Lab 9). Correlation between variables was assessed using Pearson correlation coefficient within SPSS (version 14.0, SPSS, Inc.). Area proportional to Venn diagrams were produced using BioVenn (34). lncRNA expression and tumor outcome were investigated using the log-rank test and plotted by the Kaplan–Meier method within SPSS. Tumor progression was defined as the presence of pathologic, radiologic, or clinical evidence of an increase in tumor stage and measured from the time of surgery to the time of proven event.

We investigated the expression of 17,112 lncRNAs and 22,074 mRNAs. The data are accessible through Gene Expression Omnibus Series accession number GSE55433. We filtered to 9,351 lncRNAs and 7,922 mRNAs, for which the microarray probe signals were concordant (Fig. 1A and Supplementary Table S1). Comparison between urothelial cancer and normal urothelium revealed 2,075 differentially expressed lncRNAs. In general, there was an increased expression of long ncRNAs in urothelial cancer [1,788 (86%) were upregulated (FC > 2) and 287 (14%) were downregulated (FC < 0.5)] when compared with normal urothelium. Fewer protein coding mRNAs (n = 1,410) were differentially expressed between the malignant and normal tissues, and their distribution was more balanced [836 (59%) upregulated and 574 (41%) downregulated in urothelial cancer] than for lncRNAs (t test P < 0.001). We filtered the microarray transcripts to those in the Gencode v7 catalogue [ref. (22); n = 3,885, Supplementary Table S2], of which 355 were lncRNAs differentially expressed in urothelial cancer. Although the majority were long intergenic ncRNA (lincRNA, n = 225), we identified 130 lncRNAs located within protein coding genes (termed genic). These included RNAs coded in the sense [intronic (n = 20) and overlapping (n = 3)] and antisense [exonic (n = 50), intronic (n = 51), and overlapping (n = 6)] direction with respect to the mRNA gene (Supplementary Fig. S1). The proportion of aberrantly expressed lncRNAs in urothelial cancer, when compared with normal urothelium, did not vary with gene location (range from 15% for antisense exonic to 22% for overlapping antisense lncRNAs; Fig. 1B).

We compared RNA expression across urothelial cancer phenotypes (low-grade NMI, high-grade NMI, and invasive) with normal urothelium. Unsupervised hierarchical clustering revealed three different expression profiles fitting the disease-free, the low-grade, and the high-grade/invasive phenotypes (Fig. 1C). For all tumor phenotypes, we observed more up than downregulated lncRNAs, although the extent varied between groups (91% in invasive tumors, 61% in low-grade, and 53% in high-grade; Fig. 2A and Supplementary Fig. S2a). Accordingly, the magnitude of differential lncRNAs expression varied between the invasive cohort (average fold change 2.57), and the low-grade and high-grade cohort (average fold change 1.75 and 1.73, respectively, Supplementary Fig. S2b). There was less difference in the magnitude of change in expression for mRNAs than lncRNAs (average fold change of 2.11 in invasive, 1.59 in low-grade, and 1.45 in high-grade, t test P < 0.01). The majority of aberrantly expressed lncRNAs and mRNAs belonged to the invasive subset [1,800/2,034 lncRNAs (88%) and 1,050/1,410 mRNAs (74%)], followed by the low-grade cohort [n = 560 lncRNAs (28%) and n = 757 mRNAs (54%)] and the high-grade cohort [n = 416 lncRNAs (20%) and n = 661 mRNAs (47%), (Fig. 2B)]. We defined a specific signature for each tumor phenotype by identifying RNAs significantly differentially expressed between each tumor type. Of these, 75% (1,356/1,800) of lncRNAs in the invasive phenotype, compared with 19% [109/560] of the low-grade cohort and only 3% [15/416] of the high-grade cohort were phenotype specific. Similarly for mRNAs, 45% in the invasive cohort, 17% and 4% of the low-grade and high-grade cohort, respectively, were phenotype specific. We identified n = 188 (9%) lncRNAs and n = 374 (26%) mRNAs that were altered in urothelial cancer regardless of tumor phenotype.

To identify lncRNAs with roles in urothelial carcinogenesis, we searched for those related to disease progression. We performed univariate log-rank analysis and identified 32/2,075 (1.5%) lncRNAs associated with tumor progression (Bonferonni adjusted log-rank P value < 0.05; Table 2, Fig. 3A). More of these tumor progression-related lncRNAs were phenotype specific (n = 16) than shared between all urothelial cancer (n = 6). Eight were annotated within GENCODE. LncRNAs associated with tumor progression were balanced between up (n = 17) and downregulation (n = 15). Protein coding potential was calculated using two algorithms (Table 2; refs. 24, 25). The resultant scores were closely correlated (Pearson r = 0.81) and suggested only one member was likely to be a misclassified mRNA (CR611332).

For functional validation, we investigated an aberrantly expressed lncRNA using siRNA in normal urothelial (hTERT-NHU) cells. We selected AB074278, as it was (i) associated with disease progression, (ii) upregulated in all urothelial cancer phenotypes (we were interested in lncRNAs generic to urothelial cancer not subtype specific), (iii) had low predicted protein-coding scores (thus likely to be a ncRNA), (iv) worse outcomes with high expression (thus a potential oncogenic role), (iv) also upregulated in urothelial cancer, and (v) appeared of particular interest as it was intronic (sense direction) to a protein coding host gene (sense to TANC2; as were most validated ncRNAs in GENCODE) also upregulated in urothelial cancer (thus potentially regulated by the lncRNA; ref. 35). siRNA Transfection in hTERT-NHU cells reduced expression by 80% after 24 hours and significantly increased cell apoptosis/death (Fig. 3B). Knockdown also significantly reduced proliferation (Fig. 3C) when compared with scrambled RNA controls. For comparison, we also knocked down two aberrantly expressed lncRNAs not associated with tumor progression (namely, G36639 and U50531). Proliferation was reduced for both these lncRNAs, but significantly less dramatically than for AB074278 (Supplementary Fig. S3).

To explore interactions for AB074278, we compared changes in mRNA expression between siRNA knockdown and control NHU cells using HG-U133 Plus 2.0 microarrays in triplicate. We identified 471 mRNAs that were significantly differentially expressed (359 upregulated and 112 down regulated; Fig. 4A). As the cellular phenotype of cells with AB074278-siRNA is increased apoptosis and reduced proliferation, we selected the 87 of 471 mRNAs with roles in proliferation, cell death, and apoptosis (gene functions identified using DAVID Bioinformatics Resources, Vsn. 6.7; ref. 36; bold font in Supplementary Table S3). We identified 3 of 87 mRNAs (EMP1, CKS2, and PTGS2) correlated to AB074278 expression in the discovery NCode microarray dataset (Pearson r < −0.35 or >0.2, P < 0.001; Supplementary Table S4). We measured the expression of these three mRNAs and TANC2 (the host protein-coding gene for AB074278) in the validation cohort of 138 urothelial samples (Supplementary Fig. S4) to look for associations with disease phenotype. We identified that low expression of epithelial membrane protein 1 (EMP1) in urothelial cancer was associated with increased risk of progression and BC-specific mortality (Fig. 4B, 20% vs. 38% progression rate for tumors with high and low expression, respectively, log-rank P < 0.03) and EMP1 was down regulated in urothelial cancer (Fig. 4C; ANOVA P = 0.005), in contrast with the changes seen for AB07428 [worse outcomes with high expression (log-rank P < 0.05) and upregulation in urothelial cancer (ANOVA P = 0.02)]. To support a direct regulation, we also saw increased EMP1 expression (3.8 × fold) following AB074278 knockdown in NHU-TERT cells (Fig. 4D; P < 0.01). No significant changes were seen with respect to phenotype for TANC2, PTGS2, and CKS2.

Technological improvements have revealed the importance of ncRNA in cellular function (20). Although the recent GENCODE annotated catalogue details an abundance and the distribution of many lncRNAs, little is known about their function. Here, we have performed the first large comprehensive screen of lncRNAs in urothelial cancer to identify those likely to play roles in the biology of this cancer. Putative lncRNAs were identified using an algorithm scoring characteristics of protein-coding genes, including open reading frame length, synonymous/nonsynonymous base substitution rates, and similarity to known proteins (27). We identified differential expression of many lncRNAs, without selection for genetic location, and found that the overall pattern of altered expression (mostly upregulation) was more extensive to that seen for protein coding mRNAs. Recently Wang and colleagues reported balanced changes in lncRNA expression in 12 urothelial cancer and preliniary data to suggest lncRNAs have malignant roles in mTOR and p53 signaling, and other cancer pathways (24). Although our observations do not suggest such a balanced alteration in expression, data in breast and neurologic cancers (23) and in human primary keratinocytes (35) support our findings. Given a lack of knowledge about lncRNAs and that our panel was designed computationally, we focused upon the minority (17%) of lncRNAs catalogued within GENCODE (22). This percentage compares with the 12% overlap for lincRNAs reported by Kapranov and colleagues (37), and suggests that much work is needed to fully map the lncRNA transcriptome. To date, various profiling strategies have been used to identify differential lncRNA expression in cancer [e.g., analysis of candidate ncRNA (38, 39), microarrays (40, 41), and transcriptome sequencing (23, 42)]. These reports identify that lncRNA expression is often tissue or disease specific (23). Our findings support these data and also identify that the majority of lncRNAs have tumor phenotype-specific expression (71% of all aberrant expressed lncRNAS were phenotype specific vs. 9% were common to all urothelial cancer). This reflects (but varies in extent) changes in mRNA expression (44% phenotype-specific expression vs. 26% commonly expressed), suggesting coordinated anatomical dysregulation (e.g., through regional chromosomal instability or regional epigenetic silencing) or direct interaction (43). To explore this, we mapped adjacent RNAs in our dataset. We identified many (n = 3,048) neighboring pairs where the expression was correlated (Pearson correlation r > ±0.5 and P < 0.05). This relationship was strongest for genes within 3,000 base pairs (Supplementary Fig. S5a). There were 803 lncRNA/mRNA pairs (26%), where both members were significantly differentially expressed in urothelial cancer. Expression was directly correlated for the vast majority of these pairs (82%; Supplementary Fig. S5b), suggesting common transcriptional control or the epigenetic mediation of mRNA expression by nearby lncRNAs (22, 35), antisense RNAs (44), or promoter-associated lncRNAs (45).

The most altered transcripts were found in invasive urothelial cancer. This was especially true for upregulated lncRNAs (∼64%), making the invasive phenotype distinguishable from the low- and high-grade tumors. This observation is in contrast with miRNA profiles within urothelial cancer, which often share differences between high-grade and invasive tumors (12), and suggests an exciting role to aid pathologic disease staging of high-grade tumors. Although our data require validation, we noted fewer progression events in our high-grade NMI tumors than typical (usually around 25%; ref. 5), suggesting that a chance enrichment for more indolent disease may have affected our comparisons. We focused upon lncRNAs that play potential roles in urothelial carcinogenesis, through selecting those associated with tumor progression. Most of those identified were phenotype specific, reflecting global trends within urothelial cancer, and have not been reported previously in cancer. We used loss-of-function studies for AB074278 in NHU cells to explore carcinogenic roles. This cell line has more intact cellular processes (e.g., epigenetic regulation) and fewer genetic events than most malignant cell lines, and so is better to model the subtle impact of epigenetic changes on gene expression (changes from genetic events, such as chromosomal loss/amplification, may dominate subtle modifications arising from epigenetic alterations). AB074278 was chosen because it was significantly upregulated in all tumor phenotypes (its overexpression in urothelial cancer samples could be confirmed by qPCR, data not shown), strongly associated with tumor progression, and has a low protein coding potential score. Furthermore, lncRNA AB074278 is intronic to TANC2, which was also upregulated in urothelial cancer, suggesting the potential for direct regulation. Although the expression of lncRNA AB074278 and TANC2 was not correlated in urothelial cancer cases, TANC2 expression did drop significantly following AB0724278 knockdown (Supplementary Fig. S6), allowing for the speculation that lncRNA AB074278 may regulate (by enhancing) the expression of TANC2 in cis (35). Little is known about the function of TANC2, although it is believed to play a role in embryonic development (46). Although these data are promising, we were keen to select mRNAs whose expression was abnormal in urothelial cancer and correlated to AB074278. Using two microarray screens, we identified a dynamic correlation between AB074278 and EMP1, and selected this gene due to its potential involvement in disease progression and its role in proliferation consistent with our observations. Our expression data of EMP1 matched its functional description in the literature (overexpression of EMP1 was found to inhibit the proliferation of EC9706 cells; ref. 47) and could explain the observed decrease in proliferation in transfected NHU-TERT cells. EMP1 is a putative tumor suppressor gene whose decreased expression is associated with advanced clinical stage and metastasis in oral squamous cell carcinoma (48) and reported to be directly involved in the inhibition of proliferation (47). To explore the link between AB074278 and EMP1, we compared the sequence of both RNAs and EMP1-associated genes (identified through Ingenuity Pathway Analysis, Ingenuity Inc. and EpiTect ChIP qPCR Primers by Transcription Factor search algorithm by SABiosciences) using BLAT (UCSC, GRCh37/hg19) to investigate potential direct interactions. We found a 21nt stretch of identical sequence for AB074278 and SND1 (Staphylococcal nuclease domain-containing protein 1; Supplementary Table S5). SND1 is a transcription cofactor that regulates EMP1 through its interaction with EMP1's transcription factor STAT5A (Supplementary Fig. S7 and Supplementary Materials S1–S3). These data suggest but do not confirm the exact relationship between AB074278 and EMP1, which now requires further investigation.

In summary, we have identified many lncRNAs significantly altered in urothelial cancer and associated with disease progression and tumor subtypes. We specifically implicate AB074278 in apoptosis avoidance and cell proliferation, potentially through regulating the expression of EMP1.

No potential conflicts of interest were disclosed.

Conception and design: S. Peter, J.W.F. Catto

Development of methodology: S. Peter, R.M. Drayton, C.P. Rakhit, A.P. Noon, J.W.F. Catto

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Peter, E. Borkowska, C.P. Rakhit, A.P. Noon, W. Chen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Peter, E. Borkowska, R.M. Drayton, C.P. Rakhit, J.W.F. Catto

Writing, review, and/or revision of the manuscript: S. Peter, J.W.F. Catto

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Peter

Study supervision: A.P. Noon, J.W.F. Catto

The authors thank Ian Sudbery, University of Oxford and Magnus Rattray, Manchester University for help with microarray annotation, and Dr. Paul Heath University of Sheffield for help with microarray studies.

J.W.F. Catto was supported by a GSK Clinician Scientist fellowship and project/program grants from Yorkshire Cancer Research (Grant number S305PA), Astellas Educational Foundation, and the European Union (European Community's Seventh Framework Programme; grants FP7/2007-2013 and HEALTH-F2-2007-201438). E. Borkowska was supported by the Ministry of Science and Higher Education, Poland (grant 617/MOB/2011/0).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.