- To try and identify a molecular signature for pathological staging and/or grading. through microarray analysis.
Bladder cancer is the sixth most common cancer among 23 sites  and can be lethal (3% of all cancer deaths) in some cases. There is a wide range of ways in which urothelial tumours can evolve: some tumours may remain localized, will never progress and, after resection, will never recur, whilst some will develop, invade the different layers of the bladder wall and eventually metastasize. Although histological analysis can provide much information, this is not sufficient, and a recent multicentre study reported that in so-called non-muscle-invasive (NMI) T1 urothelial carcinomas, as many as 16.2% had lymph node involvement and one third were found to be non-organ-confined at cystectomy . Because of these discrepancies, the need to find better tools to define bladder tumour classification has arisen. Some authors have already used gene profiling to predict the clinical outcomes of bladder cancer [3-5]. Although this was certainly our aim, we thought that a first reasonable step would be to assess whether gene profiling may help to better define the different groups of patients with bladder cancer, in addition to standard histology. We therefore designed a retrospective study to establish a molecular signature for pathological subgroups (Ta, T1a T1b and >T1b tumours) . In addition to this analysis, we studied whether cytological grading information added any further value.
Patient and tumour characteristics of our study cohorts are shown in Table 1. The cohort for microarray analysis comprised 108 patients treated in three medical centres (Centre Hospitalier Universitaire Lyon-Sud, Centre Hospitalier Universitaire de Lille and AP-HM Marseille, France). Informed consent was obtained from patients to use their surgical specimens for research purposes, as required by the French Committee for the Protection of Human Subjects. Patients were selected for inclusion if they had a primary tumour with no evidence of distant metastasis at the time of diagnosis. The tumours were snap-frozen by the surgeons immediately after endoscopic resection or cystectomy. Only significant tumours with a large resection (resection with muscle fragments observed) were selected for the study.
|Characteristic||Microarray Cohort||Validation Cohort|
|N = 108||N = 95|
|Median age (range) at diagnosis, years||73 (36–95)||73 (47–92)|
|Gender, n (%)|
|Female||18 (16.7)||17 (17.9)|
|Male||90 (83.3)||78 (82.1)|
|Stage (UICC 2004), n (%)|
|Ta||45 (41.7)||62 (65.3)|
|T1||35 (32.4)||13 (13.7)|
|>T1||28 (25.9)||20 (21.0)|
|Histological grade (WHO 1987), n (%)|
|G1||11 (10.2)||20 (21.0)|
|G2||27 (25.0)||36 (37.9)|
|G3||70 (64.8)||39 (41.1)|
|Histological grade (WHO 2004), n (%)|
|Low||31 (28.7)||46 (48.4)|
|High||77 (71.3)||49 (51.6)|
|Histological grade/NMI stage* (WHO 2004), n (%)|
|Low||31 (38.8)||46 (61.3)|
|High||49 (61.2)||29 (38.7)|
|Median (range) tumour cell percentage||82 (70–95)||81 (70–95)|
|Median (range) RNA quality, RIN†||8.7 (7.5–10.0)||8.2 (6.5–10.0)|
For validation analysis using reverse transcriptase quantitative (RT-q)PCR, 95 patients from a fourth centre (Clinique Beau-Soleil, Montpellier) were included to constitute a new cohort.
At the beginning of the study, all the cases were reviewed (P.P.B. and M.D.) and a tumour grade and stage were agreed upon. The tumours were prospectively classified using the TNM classification and graded using the WHO classification systems from before (G1, G2, G3) and after 2010 (low and high grades). All bladder muscle-invasive (MI) samples were cystectomy specimens and NMI samples were transurethral resection specimens. Before inclusion of the NMI samples in the study, we waited a period of 1 year to ensure the absence of progression to a more invasive stage.
Tumour biopsies, stored in liquid nitrogen until RNA extraction, were included in polyethylene glycol and cut-frozen. RNAs were prepared from 25–30 sections of 25-μm thickness. Frozen tumour sections were reviewed before analysis. Total RNAs were extracted using TRI Reagent (Sigma, St Louis, MO, USA). To remove any genomic DNA contamination, total RNAs were treated with RNAse-free DNAse I and purified using RNeasy micro columns (Qiagen, Hilden, Germany). RNA quality was verified using an Agilent Bioanalyser 2100 (Agilent, Santa Clara, CA, USA).
The microarray gene expression profiling was performed using Agilent long oligonucleotide (60 mers) technology based on a dual colour analysis method in which probes from samples and the reference (a pool of RNA extracted from tumours of Ta, T1 and >T1 stages) were differentially labelled using cyanine-5 and cyanine-3 dyes, respectively. Probes were prepared according to the manufacturer instructions using a low fluorescent low input linear amplification kit (Agilent) from 500 ng total RNA. Hybridization was performed on Human Whole Genome 4 × 44K oligonucleotide microarrays (design 014 850, # G4112F; Agilent) using reagents and protocols provided by the manufacturer. Feature extraction software provided by Agilent (Version 10.2.1.3) was used to quantify the intensity of fluorescent images and to normalize the results using a linear lowness method according to the manufacturer's instructions. All the data were imported into Resolver software (Rosetta Biosoftware, Kirkland, WA, USA) for database management, quality control and analysis.
The logRatios of probes (samples vs pool) from normalized data were imported into the Biometric Research Branch (BrB) Arrays Tools software version 3.8.1 (February 2010) (http://linus.nci.nih.gov/BRB-ArrayTools.html) . Only data with a minimum intensity of 100 were imported. A total of 12 986 probes met that criterion.
The genes that were differentially expressed among the two classes were identified using a random-variance t-test (an improvement over the standard separate t-test as it permits sharing information among genes about within-class variation without assuming that all genes have the same variance). Genes were considered statistically significant if their P value was <0.001. A stringent significance threshold was used to limit the number of false-positive findings. A global test was also performed to determine whether the expression profiles differed between the classes by permuting the labels of which arrays corresponded to which classes. For each permutation, the P values were re-computed and the number of genes significant at the 0.001 level was noted. The proportion of the permutations that gave at least as many significant genes as did the actual data was the significance level of the global test.
The lists of probes resulting from the class comparison was further used in the dChip software (http://www.dchip.org)  for clustering (distance defined as 1 − r where r is the Pearson correlation coefficient and centroid linkage). During sample clustering, the sample information is used to calculate the sample clusters enriched by samples of a certain description.
All data obtained from the microarray analysis have been submitted to Array Express at the European Bioinformatics Institute (http://www.ebi.ac.uk/arrayexpress/; accession number E-MTAB-993) in accordance with Microarray Gene Expression Data recommendations (http://www.mged.org).
A total of 500 ng RNAs were reverse transcribed using M-MLV RT RNase H Minus reverse transcriptase and oligo(dT)15 primer, according to the manufacturer's instructions (Promega, Madison, WI, USA). All cDNA amplifications were performed using 1/20th of the reverse transcription products and the LC Fast Start DNA Master SYBR Green kit (Roche Applied Science, Basel, Switzerland). Quantitative PCR was run on a LightCycler® instrument (Roche) with the following parameters: 10 min at 95 °C for the initial denaturation step then 15 s at 95 °C, 6 s at 60 °C and 12 s at 72 °C per cycle for a total of 40 cycles. The amplified cDNA concentration was evaluated using an external curve of standard samples and the specific amplification was checked using a melting curve. The PCR kinetics and quantitative data were determined using LightCycler software 4.05 (Roche). Gene expression was normalized to glyceraldehydes-3-phosphate dehydrogenase expression. To maintain quality control, two internal controls were regularly used. The interassay variations were <5% (data not shown).
Sample quality was evaluated using two criteria: the tumour cell percentage of the biopsy and the quality of RNA extract. To be included in the present study specimens were required to contain at least 70% tumour cells. Among these biopsies, the RNA quality was estimated by using the RNA integrity number (RIN) calculated using Bioanalyser 2100 Expert V2.0. RNA samples were retained when the RIN was >7.5 for microarray analysis and >6.5 for RT-qPCR analysis (Table 1). These criteria meant that one third of specimens from the four tumour banks was excluded.
The differential classification of 108 bladder tumours by stage that was provided by the DNA profile was first analysed using unsupervised hierarchical clustering, combined with a univariate t-test. Differential analysis was performed using the class comparison tool included in the BrB Arrays Tools between the logRatio profiles of NMI and MI samples. This class comparison analysis of the compared subgroups provided a differential signature of 976 probes, with many genes displaying a P value <0.001 and a fold-change >2.0 in univariate testing.
A hierarchical clustering of these 976 probes, using the dChip tool, correctly classified 92.9% (26/28) of MI stages and only 66.3% (53/80) of NMI stages (Fig. 1). It was observed that Ta and T1a tumours were mostly classified in the ‘NMI-stage’ cluster (86.7% of Ta and 68.8% of T1a tumours), whilst T1b tumours were classified mostly (16/19 tumours [84.2%]) in the ‘MI-stage’ cluster.
An Ingenuity Pathways Analysis (http://www.ingenuity.org) functional approach was used to complete these histological data so as to find target genes to study in the future. There are basically three groups of genes, defined by their function in the cell: (i) cell-cycle genes overexpressed in MI stages; (ii) overexpressed immunological signal genes in MI stages; and (iii) the genes involved in xenobiotic metabolism that are underexpressed in MI stages (Table S1).
Analysis of NMI stages was conducted to verify whether a specific molecular signature was able to distinguish low from high grade tumours. The BrB class comparison analysis, based on low and high histological grade, showed a differential list of 322 probes (at a level of P = 0.001 and a fold-change >2). A clustering analysis of data with the dChip tool showed for the group of Ta stages that low grade tumours were correctly classified (28/31 tumours [90.3%]) in the low grade cluster while only four of 14 high grade Ta tumours (28.6%) were correctly classified in the high grade cluster (Fig. 2).
The cluster ‘low vs high histological grade’ was analysed in the group of NMI tumours (n = 80). A molecular signature permitted the classification of 90.3% of Ta low grade tumours in the low grade cluster and 89.5% of T1b high grade and 68.8% of T1a high grade tumours in the high grade cluster (Fig. 3). We tested whether the molecular classification would be found to support the pre-2010 WHO classification system (G1, G2, G3), but no reliable sub-classification was found; possibly because the sub-groups were, by definition, smaller.
An Ingenuity Pathways Analysis approach was used to complete this grading study, so as to find target genes to study in the future. From the 565 probe sets, this analysis showed four subsets of genes based on their role in the cell: (i) the cell-cycle genes overexpressed in the high grade cluster; (ii) the overexpressed immunological signal genes in the high grade cluster; (iii) the overexpressed genes involved in cell development in the high grade cluster; and (iv) the genes involved in xenobiotic metabolism which were underexpressed in the high grade cluster (Table S2).
To validate the microarray results, we replicated these findings in another group of samples using another method. The RNA expression of genes was confirmed by RT-qPCR in a validation cohort of 95 tumours (Table 1). From the lists of ranked genes provided by the BrB class comparison analyses, we selected 38 differentially expressed genes. These genes discriminated tumour stage and/or histological grade (Table 2). For all genes, we found differential expression according to tumour stage or histological grade, either in all tumour samples or in the NMI tumour subset. These genes could be classified into three groups: (i) the genes informative of tumour stage and histological grade; (ii) those only informative of stage; and (iii) those only informative of grade. Most of the genes were overexpressed in MI tumours and in high grade NMI tumours, as shown by microarray analysis. By contrast, ANXA10, BCl2L14, CCND1 and PGAP1 were overexpressed in NMI tumours and, more specifically, in low grade NMI tumours. For the majority of the genes, we observed a continuum in gene expression except for BCl2L14 (Fig. 3, Table 3).
|Unique ID||Gene symbol||GenBank # accession||Description||Function||NMI vs MI stage class (80 vs 28)||Low vs high grade class (NMI stages) (31 vs 49)|
|A_23_P58328||ANXA10||NM_007193||Annexin A10||Calcium-dependent phospholipid binding||<1.0 E-07||0.10||2.5 E-04||0.19|
|A_23_P34915||ATF3||NM_001040619||Activating transcription factor 3||Gene regulation, transcription||<1.0 E-07||4.28||6.4 E-04||2.23|
|A_23_P128050||BCL2L14||NM_030766||Apoptosis facilitator Bcl-2-like 14 protein||Apoptosis regulator Bcl-G||7.6 E-05||0.44|
|A_23_P118815||BIRC5||NM_001012271||Baculoviral IAP repeat-containing 5, survivin||Inhibitor of apoptosis||2.1 E-06||2.74||<1.0 E-07||3.7|
|A_23_P122197||CCNB1||NM_031966||Cyclin-B1||G2/M checkpoint, cyclins and cell cycle regulation||1.3 E-05||2.02||<1.0 E-07||2.42|
|A_23_P65757||CCNB2||NM_004701||Cyclin-B2||Cell cycle||6.3 E-06||2.23||1.0 E-07||2.69|
|A_24_P124550||CCND1||NM_053056||Cyclin-D1, PRAD1 oncogene||G1/S checkpoint, cyclins and cell cycle regulation||1.7 E-04||0.48|
|A_23_P209200||CCNE1||NM_001238||Cyclin-E1||CDK regulation of DNA replication, G1/S checkpoint, cyclins and cell cycle regulation||4.0 E-07||2.17|
|A_23_P138507||CDC2||NM_001786||Cell division control protein 2 homologue||Regulatory pathway in response to DNA damage, G1/S checkpoint, G2/M checkpoint, Cyclins and cell cycle regulation||2.9 E-05||2.22||<1.0 E-07||2.96|
|A_23_P104651||CDCA5||NM_080668||Cell division cycle-associated 5||Regulator of sister chromatid cohesion||1.0 E-07||2.79||<1.0 E-07||3.13|
|A_24_P413884||CENPA||NM_001809||Centromere protein A||Nucleosome||1.0 E-06||2.58||1.0 E-07||2.86|
|A_23_P401||CENPF||NM_016343||Centromere protein F – mitosin||Chromosome segregation||4.0 E-06||2.50||3.0 E-07||2.86|
|A_24_P402242||COL3A1||NM_000090||Collagen, type III, α1 (Ehlers-Danlos syndrome type IV)||Cell communication, ECM-receptor interaction, focal adhesion||<1.0E-07||2.81|
|A_23_P155765||HMGB2||NM_002129||High mobility group box 2||Apoptotic DNA fragmentation and tissue homeostasis, Granzyme A-mediated||1.0 E-07||2.36||1.0 E-07||2.41|
|A_23_P133263||HMGCS1||NM_002130||3 hydroxy 3 methylglutaryl CoA synthase 1||SREBP control of lipid synthesis, butanoate metabolism, PPAR 9 signalling pathway, synthesis and degradation of ketone bodies, valine, leucine and isoleucine degradation||2.0 E-7||2.32|
|A_23_P34788||KIF2C||NM_006845||Kinesin family member 2C||Microtubules turnover regulation||1.1 E-06||2.56||<1.0 E-07||3.09|
|A_23_P258493||LMNB1||NM_005573||Lamin B1||Cell communication, caspase cascade in apoptosis, TNFR1 signalling pathway||2.1 E-06||2.34||2.9 E-05||2.06|
|A_23_P92441||MAD2L1||NM_002358||MAD2 mitotic arrest deficient-like 1||Cell cycle||<1.0 E-07||2.70|
|A_32_P103633||MCM2||NM_004526||DNA replication licensing factor, minichromosome maintenance complex component 2||Cell cycle, DNA replication||1.0 E-07||2.42||<1.0 E-07||2.40|
|A_23_P161474||MCM10||NM_182751||Minichromosome maintenance complex component 10||Cell cycle, DNA replication||6.7 E-05||2.03||<1.0 E-07||2.63|
|A_23_P40174||MMP9||NM_004994||Matrix metallopeptidase 9, gelatinase B||Inhibition of matrix metalloproteinases, angiogenesis||1.5 E-05||3.03|
|A_23_P57417||MMP11||NM_005940||Matrix metallopeptidase 11, stromelysin 3||Matrix metalloproteinases, migration||<1.0 E-07||2.88|
|A_23_P100344||ORC6L||NM_014321||Origin recognition complex subunit 6 homologue-like||Cell cycle||1.0 E-07||2.26|
|A_24_P942850||PGAP1||NM_024989||Glycosylphosphatidylinositol deacylase||GPI biosynthesis||<1.0 E-07||0.34|
|A_23_P16469||PLAUR||NM_001005377||Plasminogen activator, urokinase receptor||Complement and coagulation cascades, angiogenesis||<1.0 E-07||3.11|
|A_23_P130194||PYCR1||NM_006907||Pyrroline-5-carboxylate reductase 1||Arginine and proline metabolism, urea cycle and metabolism of amino groups||2.4 E-04||2.54||<1.0 E-07||4.80|
|A_23_P88731||RAD51||NM_002875||DNA repair protein RAD51 homologue 1||ATM signalling pathway, DNA damage, DNA replication||<1.0 E-07||2.58|
|A_23_P74115||RAD54L||NM_003579||DNA repair and recombination protein RAD54-like||Folate biosynthesis, DNA damage||4.8 E-06||2.22||<1.0 E-07||2.88|
|A_23_P71558||RECQL4||NM_004260||RecQ protein-like 4||ATP-dependent DNA helicase Q4||<1.0 E-07||2.47|
|A_23_P434809||S100A8||NM_002964||S100 A8 – calgranulin A||Calcium-binding protein||1.2 E-07||9.42|
|A_23_P200866||STMN1||NM_203401||Stathmin – métablatine||Resistance to antimicrotubule agents, MAPK signalling pathway||1.3 E-06||2.19|
|A_23_P107401||TIMP2||NM_003255||TIMP metallopeptidase inhibitor 2||Inhibition of matrix metalloproteinases, angiogenesis||<1.0 E-07||3.06|
|A_23_P107421||TK1||NM_003258||TK cytosolique||Pyrimidine metabolism, gene regulation||1.0 E-07||2.77||<1.0 E-07||2.89|
|A_23_P259586||TTK||NM_003318||Phosphotyrosine picked threonine protein kinase||Cell cycle, cell signalling, signal transduction||1.4 E-05||2.03||<1.0 E-07||2.45|
|A_23_P50096||TYMS||NM_001071||Thymidylate synthetase||Pyrimidine metabolism||4.9 E-06||2.12||1.1 E-05||2.03|
|A_24_P297539||UBE2C||NM_181803||Ubiquitin conjugating enzyme E2C||Ubiquitin-mediated proteolysis||4.7 E-06||2.67||<1.0 E-07||3.61|
|A_23_P161194||VIM||NM_003380||Vimentin||Cell communication, cell signalling, Metastasis||<1.0 E-07||2.75|
|A_23_P212284||WDR51A||NM_015426||WD repeat domain 51A||<1.0 E-07||2.46|
|NMI vs MI stage (75 vs 20)||Low vs high grade* (46 vs 29)|
|Genes obtained from stage cluster and grade cluster (NMI stages)|
|Genes obtained from grade cluster (NMI stages)|
|Genes obtained from stage cluster|
The classification of bladder tumours is difficult and there are limited means of identifying tumours that will develop into invasive forms [9, 10]. Two classification systems are currently used. The first describes the infiltration by the tumour of the various bladder layers. In the 2004 TNM classification system, NMI tumours (70%) of all tumours include papillary tumours without invasion of the lamina propria (stage Ta, 50%) and tumours invading the lamina propria (stage T1, 20%); however, in 2010, the sub-classification of T1 tumours (namely T1a and T1b, depending on whether the invasion of lamina propria is superficial or deep) was considered to be non-relevant and was therefore suppressed, but this suppression is still controversial. MI tumours (30%), by contrast to NMI tumours, invade the bladder muscle layer (stage T2) and beyond.
In addition to degree of infiltration, bladder tumours are also classified according to their histological grading. The 1973 WHO three-grade system was replaced in 2004 by the WHO/International Society of Urological Pathology classification system [11-13]. In this new system, non-invasive papillary tumours may be described as papillary urothelial neoplasms of low malignant potential (PUNLMPs), or low or high grade tumours. Papillary tumours which would previously have been defined as G1 tumours now fall into the categories of PUNLMP or low grade papillary tumour, depending on their morphology. The value of the PUNLMP designation is that it avoids the carcinoma label. Tumours formerly defined as G2 tumours are now categorized as low or high grade papillary tumours, depending on their morphology, and all former G3 tumours are now high grade papillary tumours in the new system.
Although histological classification has evolved only marginally in the past 20 years, there are several well-known drawbacks of this new system. Firstly, it needs well-trained histopathologists and, even then, interobserver variation is high [14-20]. There have been attempts to use several histological markers (such as receptor type 3 fibroblast growth factor, p53, RB, E and N cadherins) to reduce this interobserver variability, but these have had little success.
Several publications have reported convergent molecular signatures of histological type and evolutionary potential, with some genes common to several of these signatures [21-25]. Aaboe et al.  showed that the expression of 86 genes can identify high grade Ta tumours and those patients who are at risk of recurrence (P = 0.006). In another publication  that included MI tumours, an unsupervised approach enabled tumours with and without muscle invasion and tumours with good and poor prognosis to be distinguished. The discrimination capabilities of a large proportion of genes identified was validated in an analysis of a group of tumours by Dyrskjot et al. . The discrimination between NMI and MI tumours reported by Modlich et al. , using a signature of 44 genes, suggested NMI tumours have a clonal origin.
The present study is original in several respects. It is the first study of bladder tumours that has used Agilent 44K microarrays. Previous studies have used Affymetrix chips or homemade chips. Moreover, its prospective design, which allowed us to define strict criteria for the quality of the samples, is an important strength: all tumour sections were analysed by the same pathologist and only samples with at least 70% tumour cells were retained. In addition, we focused our study on a simple, albeit not yet demonstrated, aim: to confirm that in a large population (n = 108) of bladder tumours, two molecular signatures could be identified to distinguish bladder tumour stage or histological grade.
We found that the cluster ‘NMI vs MI stage’ correctly classified 92.9% of MI tumours (Fig. 1). The detailed molecular analysis by stage showed that there was a continuous progression of deregulation (overexpression or repression of these genes) from Ta stage to invasive stage through the stages T1a and T1b. Among the NMI tumours, the stages Ta and T1a were correctly classified in 86.7 and 68.8% of tumours, respectively. This molecular signature may provide new perspectives in future, as a possible complement to pathological classification. The cluster ‘NMI vs MI stage’ showed that tumours with T1b stage were closer to the MI stage tumours (84.2%). Although these results were obtained from a small number of patients, they reinforce data from previous publications underlining the prognostic value of the muscularis musoae invasion by urothelial tumours [27-29]. This sub-classification is possible in almost (90%) all the cases in centres with experienced urologists .
We did not find that molecular classification added any value to either the new WHO 2004 grading classification or the old WHO 1973 system. This could be interpreted as either a lack of value of the molecular classification, or the fact that the new histological classification is imperfect.
Gene signatures with >150 members were required to obtain robust predictors. A possible explanation for this is that the large gene signatures are less likely to be affected by fluctuations related to the techniques used in the study population, or by fluctuations related to tumour heterogeneity .
In conclusion, the present study shows that a molecular profile can be identified to characterize tumours based on their stage. By contrast, in a homogeneous group of NMI urothelial tumours, it appears to be more difficult to obtain a molecular signature to classify tumours according to their histological grade. Prospective follow-up could eventually provide new information to reduce the number of genes needed to correctly classify these patients.
This study was supported by the French Ministry of Health (PHRC National 2006). Tissues used in this work were provided by: the CHRU Lille tumour bank, the AP-HM tumour bank, the HCL tumour bank and the CRLCC Val d'Aurelle – Montpellier tumour bank.
Biometric Research Branch
RNA integrity number
papillary urothelial neoplasms of low malignant potential