identifying superficial, muscle-invasive, and ...transitional cell carcinoma of the bladder: use of...

Identifying Superficial, Muscle-Invasive, and MetastasizingTransitional Cell Carcinoma of the Bladder: Use of cDNAArray Analysis of Gene Expression Profiles

Olga Modlich,1 Hans-Bernd Prisack,1

Gerald Pitschke,1 Uwe Ramp,2 Rolf Ackermann,3

Hans Bojar,1 Thomas A. Vogeli,3 andMarc-Oliver Grimm3

1Department of Chemical Oncology, 2Institute of Pathology, and3Department of Urology, University of Dusseldorf, Dusseldorf,Germany

ABSTRACTPurpose: Expression profiling by DNA microarray

technology permits the identification of genes underlyingclinical heterogeneity of bladder cancer and which mightcontribute to disease progression, thereby improving assess-ment of treatment and prediction of patient outcome.

Experimental Design: Invasive (20) and superficial (22)human bladder tumors from 34 patients with known out-come regarding disease recurrence and progression wereanalyzed by filter-based cDNA arrays (Atlas Human Cancer1.2; BD Biosciences Clontech) containing 1185 genes. For 9genes, array data were confirmed using real-time reversetranscription-PCR. Additionally, Atlas array data were val-idated using Affymetrix GeneChip oligonucleotide arrayswith 22,283 human gene fragments and expressed sequencetags sequences in a subset of three superficial and six inva-sive bladder tumors.

Results: A two-way clustering algorithm using differentsubsets of gene expression data, including a subset of 41genes validated by the oligonucleotide array (Affymetrix),classified tumor samples according to clinical outcome assuperficial, invasive, or metastasizing. Furthermore, (a) aclonal origin of superficial tumors, (b) highly similar geneexpression patterns in different areas of invasive tumors,and (c) an invasive-like pattern was observed in bladdermucosas derived from patients with locally advanced dis-ease. Several gene clusters that characterized invasive orsuperficial tumors were identified. In superficial bladdertumors, increased mRNA levels of genes encoding transcrip-tion factors, molecules involved in protein synthesis andmetabolism, and some proteins involved into cell cycle pro-

gression and differentiation were observed, whereas tran-scripts for immune, extracellular matrix, adhesion, peritu-moral stroma and muscle tissue components, proliferation,and cell cycle controllers were up-regulated in invasive tu-mors.

Conclusions: Gene expression profiling of human blad-der cancers provides insight into the biology of bladdercancer progression and identifies patients with distinct clin-ical phenotypes.

INTRODUCTIONBladder cancer is the fifth most common solid tumor in

man, with an overall estimated annual incidence of 57,400 newcases in the United States (1). Of these, �70% are superficial atthe time of initial presentation, whereas 30% of patients arediagnosed with muscle-invasive disease. Despite treatment,about half of patients with muscle-invasive bladder cancer willdevelop distant cancer recurrence and ultimately succumb totheir disease, leading to 12,100 deaths annually (2).

Superficial bladder tumors, which are removed by transure-thral resection, are characterized by frequent tumor recurrences(3). Molecular analysis of multiple synchronous tumors andrecurrences reveal that the vast majority of these are clonallyrelated (4–8). Moreover, some muscle-invasive bladder carci-nomas can arise from independently transformed progenitorcells (9). A small group of superficial bladder tumors (5–30%,depending on stage and grade) will progress to muscle invasivedisease (10–14). Curative treatment of invasive tumors requiresradical cystectomy with urinary diversion. Depending on thetumor stage at cystectomy 30–70% of patients will developdistant cancer recurrence. Some patients with locally advanceddisease or lymph node metastasis may benefit from adjuvantchemotherapy to treat (micro-) metastatic disease (15–17).Thus, two major events determine the outcome of patients withbladder cancer: (a) progression from superficial to invasivedisease, and (b) the development of metastases.

The current Tumor-Node-Metastasis classification systemof solid tumors is based on the anatomical spread of the disease,which to some degree reflects the biological aggressiveness ofcancer. However, clinical experience shows a broad range ofbehavior among tumors at the same stage. Recent studies havefocused on cancer-associated molecular alterations because thegenotype might better indicate the clinical phenotype of a tumorrather than a sample taken at a single time point. More detailedmolecular insights into cancer biology are expected to eventu-ally provide a classification allowing optimized therapy of in-dividual patients.

For this purpose, several individual molecular markers,most of which are involved in cell cycle regulation, have beenstudied previously. Indeed, immunohistochemical detection ofthe tumor suppressor genes p53 and Rb, as well as cell cycle

Received 9/1/03; revised 1/28/04; accepted 2/10/04.The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely toindicate this fact.Requests for reprints: Olga Modlich, Institut fur Onkologische Che-mie, Heinrich Heine Universitat Dusseldorf, Moorenstr. 5, D-40225Dusseldorf, Germany. Phone: 49-211-811-4302; Fax: 49-211-811-5114;E-mail: omodlich@onkochemie.uni-duesseldorf.de.

3410 Vol. 10, 3410–3421, May 15, 2004 Clinical Cancer Research

regulator p21Waf1/CIP1, appeared to correlate moderately wellwith bladder cancer recurrence after radical cystectomy in ret-rospective studies (18–21). However, multiple genetic alter-ations are required for transformation of a normal cell into themalignant and finally metastatic phenotype. Therefore, assess-ment of multiple markers may better describe the biologicalphenotype of a particular cancer.

Several previous studies have applied microarray technol-ogy to gain insight into the changes in expression profilesoccurring during different stages of bladder cancer disease. Celllines and clinical tissue samples have been used for monitoringof expression of thousands of genes. These investigations werefocused on the role of histopathological stage, tumor grade, andproliferation activity in bladder cancer cell lines and tissues(22–26).

In this study, the expression of 1176 cancer-related geneswas studied using a filter cDNA array to identify genes and geneclusters involved in bladder cancer progression from superficialto invasive and metastatic disease. Several clusters of function-ally related genes where identified that were differentially ex-pressed between invasive and superficial bladder tumors. More-over, invasive tumors could be subdivided into two groups withdifferent outcomes based on their gene expression patterns.Additional elucidation of such genetic differences might im-prove the prediction of the clinical course of disease allowingoptimized treatment the bladder cancer patients.

MATERIALS AND METHODSPatients and Tumor Specimens. Overall, 42 transitional

cell carcinoma specimens derived from 34 patients who under-went surgery at the Department of Urology, Heinrich-HeineUniversity (Dusseldorf, Germany) were examined. Fourteenpatients had superficial (Ta, T1) and 20 had muscle-invasive(�pT2) disease. The tumors were staged according to the cur-rent Tumor-Node-Metastasis classification system by UnionInternational Contre Cancer and graded according to Bergkvistet al. (27). Patient and tumor characteristics, as well as clinicaloutcome, are summarized in Tables 1 and 2.

The 22 superficial papillary transitional cell carcinomasfrom 14 patients included pTaG1 (5), pTaG2 (14), pT1G2 (2),and pT1G3 (1) tumors. One superficial tumor was obtainedduring nephroureterectomy and all others by transurethral re-section between October 1999 and September 2000. Eachpapillary-shaped tumor was cut at the base to avoid coagulationof the specimen. To compare multifocal and recurrent bladdercarcinomas of 7 patients, 2 and in one case 3, tumors wereexamined. These included (a) two primary tumors and the firstrecurrence, (b) five multifocal tumors (including one multifocalrecurrence), and (c) one synchronous ureter and bladder tumor.No patient with superficial bladder cancer has progressed tomuscle-invasive disease after a mean follow-up period of 32months.

Specimens of locally advanced bladder cancer were ob-tained from 20 patients during radical cystectomy betweenAugust 1993 and December 2000. Of four muscle-invasivetumors (P1, P35, P36, P37) specimens from different areas, e.g.,superficial or deeper layer of the bladder wall, were examined.These tumors were macro-dissected into two to four counter-

parts containing different proportions of tumor to stroma cells.In three of these and one other case (P24), corresponding normalmucosa of the bladder was available. Half of the 20 patientsrecurred after a mean time period of 9 months and subsequentlydied of disease. Eight patients remain alive after a mean fol-low-up of 75 months. Two patients died perioperatively.

Tissue Sample Preparation. Immediately after surgicalresection, tumors were macro-dissected. Each bladder cancersample was confirmed as representative by adjacent fresh frozenand paraffin-embedded sections of cystectomy and transurethralresection specimens, respectively, then frozen in liquid nitrogenand stored at �80°C until RNA extraction. H&E-stained sec-tions from tumor specimens and normal mucosa samples wereexamined to assess the relative proportion of tumor cells, benignepithelium, stroma, and lymphocytes.

Bladder Cancer Cell Lines. The origin and culture oftransitional cell carcinoma cell lines T24, VM Cub 1, and VMCub 3 have been described previously (28). Cells were har-vested at 70% confluence for RNA extraction.

Total RNA Isolation and cDNA Probe Synthesis. TotalRNA from tissue specimens and cell lines was extracted by theguanidine-isothiocyanate method, followed by phenol/chloro-form extraction and DNase treatment using the recommendedprotocol for the Atlas Pure Total RNA Labeling System (BDBiosciences Clontech, Heidelberg, Germany). Total RNA wasquantified using UV spectrophotometry (Photometer ECOM6122; Eppendorf AG, Hamburg, Germany), and the quality andintegrity was confirmed by electrophoresis of 0.5 �g of isolatedRNA in 1% formamide agarose gels. Additionally, RNA spec-imens were analyzed by microcapillary electrophoresis onLabChips using the Agilent 2100 Bioanalyzer (Agilent Tech-nologies GmbH, Boeblingen, Germany) following the manufac-turer’s instructions.

For first-strand cDNA synthesis, 5–10 �g of native totalRNA were mixed with 1 �l of coding sequence primer mix(0.02 �M; BD Biosciences Clontech), incubated at 70°C for 4min followed by 4 min at 50°C. Master mix containing 2 �l of5� reaction buffer, 1 �l of 10� deoxynucleoside triphosphatemix (5 mM each of dCTP, dGTP, dTTP, and dATP), 0.5 �l of100 mM DDT, 3 �l of [�-32P]dATP (10 �Ci/�l, 3000 Ci/mmol;Amersham Biosciences Europe GmbH, Freiburg, Germany),and 1 �l of Moloney murine leukemia virus reverse tran-scriptase (100 units/�l; Sigma-Aldrich Chemie GmbH,Taufkirchen, Germany) was added to each template. After in-cubation at 50°C for 35 min, the reaction was terminated usinga mixture of glycogen and EDTA [10� Termination mix: 0.1 M

EDTA (pH 8.0), 1 mg/ml glycogen; BD Biosciences Clontech],the labeled cDNA probes were purified and probe incorporationassessed by scintillation counting.

Hybridization and Image Analysis. Atlas Human Can-cer 1.2 Array nylon membranes (BD Biosciences Clontech)containing 1185 cDNA spots corresponding to 1176 knownhuman genes and 9 human housekeeping genes, as well asgenomic DNA spots and positive and negative controls, wereprobed with labeled cDNAs by overnight hybridization at68°C using ExpressHyb Hybridization Solution (BD Bio-sciences Clontech). Briefly, radiolabeled probes were heatdenatured and then added to 5-ml aliquots of hybridizationbuffer containing 100 �g/ml heat-denatured sheared salmon

3411Clinical Cancer Research

testes DNA (Life Technologies, Inc., Karlsruhe, Germany).The final probe concentration was 1–3 � 106 cpm/hybrid-ization. For all samples, expression profiles were averagedfrom two independent hybridization experiments performedwith separate array filters. Identical batches of gene-specificprimer mix and membranes were used to exclude possible

variations in cDNA probe synthesis and hybridization. Afterextensive washing (three times with 2� SSC, 1% SDS, andonce with 0.1� SSC, 0.5% SDS each for 30 min at 68°C,once with 2� SSC for 10 min at room temperature), themembranes were exposed to PhosphoImager screens (BAS-1500; Fuji, Raytest, Germany).

Table 1 Clinicopathological information: invasive tumors

CaseAge(yrs) Gender

Stage/Nodalstatus Grade Grade

Recurrence(mos) Localizationa Follow-up (mos) Patient statusb

No. ofspecimens

1 60 M pT3a pN0 G3 No 94 Ab 224 72 F pT2b pN0 G2 No 99 A 1 (�MU)6 61 M pT3a pN0 G3 No 108 A 1

26 64 F pT2b pN0 G3 No 99 A 130 67 F pT3a pN0 G3 No 60 A 131 73 M pT3b pN0 G2 No 67 A 134 59 M pT2b pN0 G3 No 48 A 139 61 M pT2b pN0 G3 No 24 A 12 68 M pT3b pN0 G3 5 Local 9 DOD 17 52 M pT2b pN0 G3 9 Lu, Ln, L, B, Br 21 DOD 18 62 M pT2b pN0 G3 12 Ln, B 18 DOD 1

23 81 F pT2b pN0 G3 4 Ln, L 9 DOD 125 60 M pT3b pN0 G3 4 L, B, Pl 4 DOD 127 66 M pT3a pN2 G3 5 Ln, B 8 DOD 1 (�LN)28 60 F pT3b pN2 G3 37 Local 41 DOD 1 (�LN)32 54 M pT2b pN0 G2 8 Ln, B 40 DOD 133 59 M pT3a pN0 G3 4 B, Pe 13 DOD 136 59 F pT3a pN0 G3 5 Local 11 DOD 3 (�MU)35 60 M pT3b pN2 G3 n/a Dc 3 (�MU)37 72 F PT3b pN0 G3 n/a Dc 4 (�MU)

a Site of tumor recurrence; Br, brain; L, liver; Ln, lymph nodes; B, bone; Lu, lungs; Pl, pleura; Pe, peritoneum; n/a, not available; D, perioperativedeath; LN, lymph node metastasis; MU, mucosa.

b A, alive; DOD, dead of disease.

Table 2 Clinicopathological information: superficial tumorsa

CaseAge(yrs) Gender Type Stage Grade

Time torecurrence(months) Recurrence

Follow-up(months) Note

10 68 M Primary Ta G2 No 3511 75 M Primary Ta G2 No 4012 48 M Primary Ta G2 5 TaG2 3913_1 61 M Recurrence Ta G2 13 TaG2 39 Multifocal13_2 Recurrence Ta G214_1 64 M Primary Ta G2 No 37 Multifocal14_2 Ta G215_1 75 M Primary Ta G1 15 TaG2 33 Multifocal15_2 Ta G116_1 79 M Primary Ta G2 No 37 Multifocal16_2 T1 G217_1 87 M Recurrence Ta G2 No 2 (Loss to FU) Multifocal17_2 Ta G218 73 M Primary Ta G1 31 TaG1 3119 65 F Primary Ta G2 4 TaG2 37 Tumor recurrence of case

no. 1929 66 Recurrence Ta G2 22 TaG220 90 F Primary T1 G2 No 18 Unrelated death21 71 M Recurrence Ta G1 No 3622 68 M Primary T1 G3 7 TaG2 34 Multifocal tumor recurrence

of case no. 2240_1 69 Recurrence Ta G2 7 T1G240_2 Ta G238 77 M Recurrence Ta G1 8 TaG2 26a There was no patient with tumor progression to muscle-invasive disease.

3412 Classes of Bladder Carcinoma with cDNA Arrays

Atlas Array Data Analysis. Radioactive intensity ofeach spot on the membrane was linearly digitized to gray levelswith a pixel size of 100 �m (PhosphoImager) and recordedusing a commercially available image analysis software (TINA;Fuji). Data were obtained as a list of intensity values for allmeasured positions on the array. All values derived from imageanalysis were corrected for background using Excel 97 software(Microsoft). Normalization of signal intensities obtained fromdifferent hybridization experiments was based on the sum ofbackground-subtracted signal data of all expressed genes andwas performed to account for differences between labeling andquality of RNA samples. Such total intensity normalizationmethod relies on the assumption that quantities of initial mRNAare the same for all samples. Under this assumption, a normal-ization factor can be calculated and used to re-scale the intensityfor each gene in the array (29). Absolute values correspondingto abundance of RNA in the probe were normalized according tomean value (after normalization the mean value of signal inten-sities in each array was equal to 1.0). The resulting data table andraw data are available from our web site (see data files 1, 2).4

Reproducibility of experiments was verified by comparingone hybridization with the same probe on two independentarrays. Identical batches of gene-specific primer mix and mem-branes were used to exclude possible variations in cDNA probesynthesis and hybridization. The results showed good reproduc-ibility with correlation coefficients from 0.9 to 0.98. Moreover,the notch group protein gene represented by two different clones(A02g and A05 h) displayed similar expression profiles for thetwo clones in all samples.

Hierarchical Clustering. Mean values were calculatedfor each gene from the invasive and superficial groups ofsamples. Weak signals on the membrane close to backgroundresult in an unfavorable signal-to-noise ratio and show lowreproducibility between experiments. Ratios in gene expressionbetween different samples based on such values may be veryhigh even when there is no significant difference in the expres-sion level of the corresponding genes (29). Therefore, onlygenes with a mean expression value of �0.7, either in theinvasive or in the superficial group, were used for additionalanalysis (n � 270). The normalized values of 270 genes werelog transformed, and data for each gene were mean-centered andnormalized. Data for each sample were also mean-centered. Thehierarchical clustering program (version 3.1) developed byEisen et al. (30) was applied to group samples on the basis ofoverall similarities in their gene expression patterns, as well asto group genes on the basis of similarity of their expressionlevels in all samples. Results were displayed with the TreeViewprogram, version 1.6.6 (Ref. 30; Fig. 1). Red color was used toshow genes with expression levels greater than the mean andgreen for expression levels lower than the mean.

Oligonucleotide Array Hybridization and Analysis.Aliquots of total RNA of 9 tumors were subjected to additionalanalysis using the Affymetrix HG-U133A array. Three of thesesamples were from superficial tumors (p10, 11, and 12), 3 frominvasive tumors of patients with long-term disease-free survival

(p1, 30, and 34), and 3 from invasive tumors of patients withrapid tumor progression and cancer-related death (p23, 27, and36). The HG-133A array contains 22,283 human gene fragmentsand expressed sequence tag sequences.

Total RNA (5 �g) was used in Affymetrix oligonucleotidearray hybridization. The whole experimental procedure wasperformed according to standard Affymetrix protocols, and hy-bridizations were done in the core facility at the Department ofChemical Oncology, Heinrich-Heine University. Synthesizedand fragmented cRNAs were analyzed by electrophoresis usingLabChips and the Agilent 2100 bioanalyzer.

Arrays were scanned at 560 nm using a laser-scanningmicroscope (Agilent GeneArray Scanner G2500A; AgilentTechnologies GmbH). To normalize for sample loading andstaining variation, the average of the fluorescent intensities of allprobe sets on an array were scaled to a constant target signalintensity (target signal factor 100) for all arrays used. Scalingfactor between arrays was 2.

Expression profile data were collected and analyzed usingAffymetrix Microarray Suite 5.0. To identify differentially ex-pressed genes between superficial and invasive tumors, the ratioof average expression level for each gene between the twogroups (invasive to superficial) was calculated. For this analysis,raw data as scaled by the Affymetrix software were used (seedata file 4 from our web site).4 Low values corresponding tosignal intensities close to background and defined as ‘A’ (ab-sent) were not changed. When the intensity of the detected genesis close to background, it is not accurate to calculate the ratio.Therefore, ratios for genes that are not expressed in one samplegroup but expressed in the second one should be considered asrelative. Because the empirical minimum for detection of asignal on the Affymetrix array by use of target signal factor �100 is 30, the cutoff value for the average gene expression levelswas set to 100.

To compare Atlas array hybridization data obtained for 34patients with 22 superficial and 20 invasive bladder cancers,data obtained from Affymetrix platform for the subset of 9bladder cancers was filtered to include only genes in commonwith both arrays. For filtering, GenBank accession numbershave been used. Of 1185 genes listed on the Clontech array,1164 corresponding genes could be identified on the AffymetrixHG-U133A array. There are 8 genes on the Human Cancer 1.2array, which are present twice, and one gene with three posi-tions. This data file is available from our web site (see data files4, 5).4

Statistical Analysis. Standard statistical procedures, in-cluding the Wilk-Shapiro and the unpaired Student’s t test, wereused. Correlation in gene expression between data obtainedfrom Affymetrix and Clontech platforms were analyzed bySpearman’s rank test. All reported tests were two-sided, andP 0.05 was considered significant.

Beissbarth et al. (31) systematically analyzed array datafrom radioactive hybridization array experiments with regard tocomparability of different experiments and reproducibility. Theauthors concluded that a 2-fold change can be considered sig-nificant given a high reproducibility with correlation coeffi-cients of 0.9 as we did observe (see above). According to thisand other studies (32–37), a change of at least factor 2 wasconsidered significant.4 Internet address: http://www.cancertoday.com.

Fig. 1 A two-way clustering analysis of bladder invasive and superficial bladder cancer tissues and cancer cell lines. Thirty-four samples and threecell lines are lined up on the horizontal axis, and 270 genes prefiltered from 1185 genes listed on Clontech Atlas human cancer 1.2 filter array areon the vertical axis (see “Materials and Methods”). Superficial and invasive tumors, as well as cell lines, were distinguished. Each cell in the colormatrix represents the relative expression level of a single gene transcript in a single sample. Expression level was normalized per gene, and the relativevalue to the mean among 37 arrays is shown by color: red; transcript levels above the mean for the given gene across the samples, green; transcriptlevels below the mean for the given gene across the samples. The magnitude of deviation from the mean value is represented by color saturation. A,matrix representation of expression levels and the dendrogram of samples. f from a–g to the right indicate the locations of gene clusters of interest.B, expanded view of several gene clusters of interest named from top to bottom: a, integrin cluster; b–d, clusters of genes involved in nucleotide andnuclear acid metabolism, protein metabolism and synthesis, repair processes, cell growth and differentiation, and cell cycle control; e, stromal cluster;f, extracellular matrix remodeling proteins; g, cluster contains genes controlling cell cycle, and cell proliferation and growth.

Quantitative Real-Time Reverse Transcriptase-PCR(rtRT-PCR). rtRT-PCR was performed using the ABIPRISM 7900 Sequence Detection System (Applied Biosystems,Foster City, CA). Total RNA (3 �g) from 24 bladder tumorsamples (superficial: 10, 12, 13I, 13II, 14I, 15I, 17I, 29, 38, 40I,40II; invasive: 1, 2, 23, 24, 25, 26, 27, 34, 35I, 36I, 36II, 37I,37II) were reverse transcribed using random hexamer primerand Superscript II (Life Technologies, Inc.) as recommended bythe manufacturer. Each cDNA mixture was diluted 1: 8 indistilled H2O for subsequent PCR amplification using Assays-on-Demand Gene expression products (Applied Biosystems).Expression of the gene for cyclophilin A served as an internalcontrol. Data were extracted and amplification plots generatedwith the ABI Sequence Detection System software. All ampli-fications were done in duplicates and threshold cycle (Ct) scoreswere averaged for subsequent calculations of relative expressionvalues. Quantification of rtRT-PCR data was performed asfollows. The Ct scores for genes of interest for each sample werenormalized against Ct scores for the corresponding endogenouscontrol gene (cyclophilin A; PPIA): Ct � Ct Taget gene �CtPPIA. Values of Ct detected by rtRT-PCR were comparedwith ln of expression obtained from array experiments (Clon-tech and/or Affymetrix) by calculating of Pearson correlationcoefficient (r) and its corresponding P.

RESULTSGene Expression Profiling of Bladder Cancer. RNA

isolated from 42 different human bladder tumors from 34 pa-tients and from three bladder cancer cell lines were used inhybridization experiments with cDNA arrays containing 1176cancer-associated and 9 housekeeping genes. Of these genes,270 were expressed at significant levels either in the invasive or

in the superficial tumor group and were subjected to hierarchicalcluster analysis.

The unsupervised clustering algorithm groups both genesand tumors according to similarities in expression levels. Over-all, samples divided into two major clusters which separated all42 bladder cancers according to histopathological features (Fig.1). Thus, invasive tumors were placed separately from thesuperficial tumors branch of the dendrogram. Expression pro-files of three cancer cell lines examined were grouped togetheron a subbranch of the sample dendrogram within the superficialtumors branch. This branch also included a high-grade superfi-cial invasive tumor (pT1G3) and a locally advanced tumorrecurrence after an initial pT1G3 disease. A similar separationof tumor groups was also obtained using expression data fromthe full gene expression dataset (1185 genes; Fig. 2A).

Notably, unsupervised clustering also divided the muscle-invasive tumors into two major subgroups with disproportionalrepresentation of adverse clinicopathological features. SubgroupII (Fig. 2A, P8-P23, right) mainly contained patients with locallyadvanced disease or lymphatic spread and those who rapidlyrecurred after radical cystectomy. In contrast, 5 of 9 patients (56%;Fig. 2A, P39-P24, left) in subgroup I experience long-term recur-rence-free survival (mean, 66 months). Considering only thosepatients with organ-confined disease (pT2bpN0G3), 1 of 5 com-pared with 4 of 4 tumors in the two different branches progressed.

We also compared gene expression profiles of differentsamples of multifocal superficial tumors and samples fromdifferent parts of invasive tumors (Fig. 2B). The same set of 270genes, described above, was used in the hierarchical clusteringalgorithm. Although all 5 multifocal superficial bladder tumorswere each clustered together, the synchronous ureter and blad-der cancer (p14I & p14II, both pTaG2) were separated on the

Fig. 2 Dendrograms of tissuesamples and cell lines based onclustering of different data sets.A, dendrogram of samples, in-cluding invasive and superficialbladder cancer tissues, andthree bladder cancer cell linesbased on the whole data set forthe expression of 1185 geneslisted on Clontech Atlas HumanCancer 1.2 array. Mean valuesof gene expression were usedif several tumor specimens ortwo different tumors originatedfrom 1 patient were studied. B,dendrogram of all tissue sam-ples studied obtained by use ofthe subset of 270 genes filteredfrom the whole set of 1185listed genes: mean expressionin either invasive or superficialgroup of tumor samples � 0.7.Mucosa tissue samples are des-ignated as MU.

dendrogram. Expression profiles of differently localized tumorspecimens from invasive tumors (P1, P35, P36, and P37) weremore similar to themselves than to samples from differenttumors. The only exception was a part of tumor P36 containinga smaller proportion of tumor cells (data not shown). Expressionprofiles of samples corresponding to normal mucosa of patientswith locally advanced tumors were clustered together but sep-arately from their corresponding tumor specimens as a sub-branch in one invasive tumors branch.

There are 124 genes in the Clontech dataset with foldchange of 2, 63 down, and 61 up. By use of an additionalcriterion for P of 0.05 (Student’s t test), 106 genes showeddifferential expression in invasive versus superficial tumors.Therefore, the false discovery rate (32) of the 106 probe set is5.8%. For 48 genes, mRNA levels were increased in invasiveversus superficial tumors, and 58 genes were decreased ininvasive versus superficial tumors (see Supplementary Materi-als; Table 1). The clustering algorithm identified several clustersof genes that were differentially expressed between invasive andsuperficial tumors. These clusters included genes involved innucleotide and nucleic acid metabolism, protein metabolism andsynthesis, DNA repair, cell growth and differentiation, and cellcycle control. Clusters b, c, and d (Fig. 1) were expressed athigher levels in superficial tumors. In contrast, invasive tumorsdemonstrated higher expression of genes (clusters e and f; Fig.1) encoding stroma proteins, proteins involved in extracellularmatrix remodeling, as well as immune system proteins.

Clusters a and g consisted of genes that were expressedsignificantly higher in 6 invasive tumors compared with otherinvasive samples. These specimens, derived mainly from high-risk patients, were clustered together on a distinct branch of thesample dendrogram (Fig. 1). Cluster a contained genes related tocell cycle regulation, transcription, cell-matrix, and cell-celladhesion. Cluster g contained genes controlling cell cycle, andcell proliferation and growth.

The gene expression profile of epithelial bladder tumor celllines grown in vitro and the human bladder tumor samples didnot show major differences. There were, however, several genesthat were preferentially expressed in cell lines but not in tumorsamples (see Supplement Cell Lines; web site as above).4 Thesegenes were not included in the 270 gene set further used forclassifying of tumors. The genes up-regulated in the examinedcell lines were mainly involved with proliferation processes andcell growth.

Affymetrix GeneChip Hybridization. To further ex-plore results from Atlas arrays, expression data obtained byAffymetrix GeneChip HG-U133A, containing 22,283 humangenes and expressed sequence tags, were compared between 3randomly selected superficial and 6 invasive tumors. Of 45–50% elements expressed, a total of 215 genes was differentiallyexpressed in all 19 pairwise comparisons, as defined by criteriain the Affymetrix software. The expression levels of 105 mR-NAs were increased in invasive tumors and of 110 mRNAs wereincreased in superficial tumors. Applying a more stringent 10-fold change as the cutoff, we generated a subset of 77 differen-tially expressed genes (list provided in Supplemental Table 2).

Comparison of Gene Expression Data for Clontech andAffymetrix Platforms. On the basis of the 270 genes identi-fied by analysis of the Atlas data, a group of 117 genes whose

Fig. 3 Average signals: Affymetrix versus Clontech Atlas platform.The differential up- or down-regulation between invasive and superficialbladder tumors based on Affymetrix (A) and Clontech (B) platformswere compared. The analysis was applied to genes that were in commonwith both platforms. Expression values were averaged per gene sepa-rately for 3 superficial and 6 invasive tumors (tested with Affymetrixarrays) using already normalized data sets for 1164 genes obtained withClontech and Affymetrix platforms. The genes on the Affymetrix plat-form were defined as differentially expressed if there was a 2-folddifference. The up-regulated genes are shown in red and the down-regulated genes in green. Notably that many red and green crosses,which are still on their true positions, confirm a particular gene differ-ential expression, although there are many red and green colors that arespread between blue (no regulation). Notice the disconnection for dif-ferential expression between two platforms.

corresponding measurements using the Affymetrix GeneChipsignificantly correlated in a Spearman’s rank test (P 0.05)were identified (for detailed list see our web site “Affymetrixversus Clontech Data”).4 For the majority of 153 genes, how-ever, the direct comparison of Atlas and the Affymetrix datasetsrevealed a lack of agreement between the two platforms. Toillustrate the discrepancy between both platforms, scatterplotswere used (Fig. 3). As illustrated in Fig. 3, some genes thatappear up-regulated on the Clontech platform did not change oreven show down-regulation on the Affymetrix platform and viceversa. A detailed comparison of Clontech and Affymetrix dataare subject of a separate article.5

Of the 117 corresponding genes, 41 were also differentiallyexpressed between superficial and invasive tumors based on

data from both platforms. We again applied the hierarchicalclustering algorithm to group tumor samples and these 41 geneson the basis of their overall similarities (Fig. 4). Gene clusters,as well as sample groups, were similar to those obtained for the270 genes from the Atlas arrays and clearly separated superficialand invasive tumors (Fig. 1). Moreover, the majority of muscle-invasive tumors of subgroup II (Fig. 2 A), characterized byadverse clinicopathological features, remained as a separatebranch.

Validation Using rtRT-PCR. To confirm the array data,24 of the initial 42 bladder cancers were randomly chosen, andrtRT-PCR was performed for the 11 genes listed in Table 3.With the exception of only 2 of the 11 genes, there was asignificant correlation between the Clontech and/or Affymetrixexpression data. Some gene transcripts were represented bymore than one probe set on Affymetrix array and each detectedsimilar relative levels of expression, e.g., different variants ofthe decorin gene (Table 3). However, the extent of changedetected by the different methods varied.

Two of the genes validated, decorin and stratifin, are

5 O. Modlich, H-B. Prisack, M-O. Grimm, and H. Bojar. Comparativeanalysis of gene expression datasets obtained by use of Clontech cDBAnylon arrays and Affymetrix oligonucleotide arrays for transitional cellcarcinoma of the bladder, manuscript in preparation.

Fig. 4 Hierarchical clustering analysis of 41 genes differentially expressed between superficial and invasive bladder cancers based on Clontech arraydata analysis (Student’s t test; P 0.05) and validated by Affymetrix platform. Gene expression data from Clontech Atlas and Affymetrix GeneChiphybridization analysis were compared using Spearman’s rank test (P 0.05). Mean values of gene expression were calculated if syn- or metachronoussuperficial tumors or different invasive tumor specimens derived from one patient were available (M). Values were log-transformed, normalized, andmedian-centered for cluster analysis.

differentially expressed between the two major subgroups ofmuscle-invasive bladder cancer. Fig. 5 illustrates the expressionof decorin and stratifin as determined by either rtRT-PCR orAtlas array analysis. Despite the small sample size, the differ-ence in expression between the subgroups was highly significantfor stratifin and approached statistical significance for decorin.

The rtRT-PCR expression data of two genes, p21Waf1/CIP1

and SOCS3, did neither correlate with Clontech nor with Af-fymetrix expression data for the selected samples. Despite threedifferent probe sets, the Affymetrix GeneChip failed to detectexpression of signal transducers and activators of transcription-induced signal transducers and activators of transcription inhib-itor 3 (SOCS3) in any of the tumor samples tested.

DISCUSSIONGlobal gene expression profiling has been used for molec-

ular classification of several cancers, including breast, prostate,colon, non-small cell lung cancers, lung adenocarcinoma,medulloblastoma, and melanoma, as well as of large B-celllymphoma (38–43). This classification has been shown to cor-relate with clinical phenotype as defined by tumor progression

and survival. Regarding bladder cancer, only limited data ongene expression profiles are available. The group of Ørntoftapplied hierarchical cluster analysis to pooled suspensions ofbladder cancer cells and identified gene clusters covarying withstage and grade (44). Using a cross-validation approach, thisgroup proposed a 32-gene molecular classifier predicting dis-ease progression of Ta tumors (45).

In this study, microarray analysis was used to identifydistinct gene expression profiles of superficial, muscle-invasive,and metastasizing transitional cell carcinoma of the bladder. Toclearly define the phenotype of a given tumor, only patients withuniform treatment (transurethral resection for superficial andradical cystectomy for invasive bladder cancer) and long-termfollow-up were studied. Using unsupervised hierarchical clusteranalysis, a clear cut separation of superficial and invasive tu-mors was observed. This separation was remarkably stable usingdifferent subsets of expression data. Furthermore, several geneclusters related to superficial or invasive bladder cancer wereobserved. For example, an up-regulation of certain genes en-coding transcription factors, factors involved in protein synthe-sis and metabolism, as well as some proteins controlling cell

Table 3 Correlation of Clontech/Affymetrix expression data with real-time PCR derived valuesa

Gene Array platformClontech/Affymetrix probe set

GenBank accession no.

Pearson correlation

Decorin Clontech D11b �0.8117 9.65E-07M14219

Decorin, variant A Affymetrix 201893_x_at �0.9473 3.01E-04AF138300.1

Decorin, variant D Affymetrix 211813_x_at �0.9288 8.12E-04AF138303.1

Decorin, variant C Affymetrix 211896_s_at �0.9285 8.24E-04AF138302.1

Tenascin C (cytotactin) Affymetrix 201645_at �0.9751 2.39E-05NM_002160.1

Actin, � 2 Affymetrix 200974_at �0.9707 4.15E-05NM_001613.1

Filamin A, � Affymetrix 200859_x_at �0.9108 1.69E-03NM_001456.1

Myosin, heavy chain polypeptide 11 Affymetrix 201497_x_at �0.9789 1.36E-05NM_022844.1

Calponin 1 Affymetrix 203951_at �0.9674 5.99E-05NM_001299.1

WAF1/p21 Clontech A091 0.0915 0.6702U09579

WAF1/p21 Affymetrix 202284_s_at 0.1556 0.7350NM_000389.1

STAT-induced Clontech B121 �0.2769 0.1891STAT inhibitor 3 (SOCS3)b AB00490414-3-3 protein �; stratifin Clontech B03m �0.9252 3.38E-11

AF02908214-3-3 protein �; stratifin Affymetrix 209260_at �0.9848 4.43E-06

BC000329.1IFN-induced transmembrane protein 9–27 Affymetrix 1) 201601_x_at 1) �0.9615 1) 1.05E-04

NM_006341.12) 214022_s_at 2) �0.9763 2) 2.02E-05

AA749101IFN-induced transmembrane protein 9–27 Clontech A01h �0.6791 2.23E-04

J04164a Correlations significantly positive for 9 of 11 genes.b There are three probe sets for the STAT-induced STAT inhibitor 3 on the Affymetrix GeneChip (206359_at; 206360_s_at; 214105_at). All

probe sets did not detect any expression of this gene.

growth and differentiation were characteristic of superficialtumors. Similar observations were reported by the group ofØrntoft (46) who correlated gene expression patterns with stageand grade but not clinical outcome.

The dendrogram obtained by cluster analysis also identi-fied a group of patients with muscle-invasive disease who are athigh risk for bladder cancer recurrence after radical cystectomy.This was even more evident for those with organ-confineddisease: while all of the poor risk branch of patients recurred thisoccurred only in 1 of 5 patients in the good-risk group. How-ever, this data need to be confirmed in a larger cohort ofpatients.

Bladder cancer frequently occurs as multifocal disease withseveral simultaneous tumors scattered over the bladder. In thisstudy, multifocal superficial bladder cancers always clusteredtogether, suggesting a common genetic background. We cannoteliminate the possibility that this relationship is based on thegenotype of the respective patient. However, clonality of mul-tifocal bladder cancer has also been reported by several otherstudies based on loss of heterozygosity analysis (7, 8).

Invasive bladder tumors are histologically complex tissuesconsisting of a variety of cell types besides carcinoma cells. Aswith other methods, measuring tissue RNA expression microar-rays determine expression levels irrespective of cell type. How-ever, tumor development and progression, e.g., growth, angio-genesis, tumor cell detachment, and invasion, is dependent oninteractions between tumor cells and the host microenvironment(stroma). Thus, expression of factors involved in tumor-cell

stroma interactions might reflect biological properties independ-ent of cellular origin (47–51). Although microdissection wouldallow the examination of highly pure tumor cell populations ofa given specimen, it might be of little use in a clinical setting.Thus, identification of a molecular classifier gene set, which isto some degree independent of tumor tissue heterogeneity, isrequired.

To determine the influence of different tumor-stroma rela-tions and tumor heterogeneity, specimens from different areas ofinvasive tumors, e.g., infiltrating, superficial, or deeper layers ofthe bladder wall, were included in the cluster analysis. With oneexception (P36.3), all samples from the same tumor weregrouped together on the dendrogram (Fig. 2B), suggesting thateach part of the tumor was representative for the whole, basedon multiple gene expression analysis. Notably, the only excep-tion was found to contain the lowest proportion of tumor cells(data not shown). The clonality of different parts of invasivetumors does, however, not reflect the genetic identity of a givenpatient because corresponding mucosa samples appeared on aseparate branch of the dendrogram.

Expression profiles of all histologically verified normalmucosa specimens, derived during cystectomy for invasivebladder cancer, were clustered together on the invasive branchof the sample dendrogram; the respective (sub-) branch con-tained several good-risk patients. This is in line with otherreports of genetic alterations (e.g., LOH on chromosome 9 loci)in normal mucosa of bladder cancer patients, suggesting a fielddefect (49, 50). However, our observation that these mucosa

Fig. 5 Comparison betweencDNA array and real-time re-verse transcriptase-PCR (rtRT-PCR) results for 2 selectedgenes. Expressions of decorinand stratifin were examined us-ing RT-PCR in two subgroupsof invasive tumors: poor-out-come group of samples 1, 2, 23,25, and 37I&II, and the secondsubgroup of invasive tumor sam-ples (called invasive) 24, 26, 27,34, and 36I & II. All RT-PCR-derived values have been normal-ized to cyclophilin A. All arraydata were globally normalized(see “Materials and Methods”),and mean expression values werecalculated for each of subgroup ofinvasive tumors. � and � col-umns represent poor-outcome andinvasive subgroups of tumor sam-ples, respectively. Differential ex-pression between two subgroupswas confirmed by unpaired Stu-dent’s t test. P values are shown.

samples cluster into the invasive, but not the superficial (or anown distinct) branch, suggests that this mucosa is genotypicallydetermined not only as malignant but also as potentially inva-sive. This raises the intriguing possibility that a gene expressionprofile of bladder mucosa may exist that characterizes patientswith a high risk of developing invasive disease.

Because the major focus of this study was the identificationof differentially expressed genes, we aimed at independentconfirmation of the whole data set. Therefore, we comparedgene expression data obtained by the cDNA nylon array (Atlas;Clontech) with those from an oligonucleotide GeneChip (Af-fymetrix). The direct comparison revealed a limited agreementbetween the two different array platforms (Fig. 3). A lack ofagreement between gene expression data from different com-mercial microarray platforms has also been reported by othergroups emphasizing that independent confirmation of completegene sets for molecular classification is required (52–58).

In this study, Clontech expression data were confirmed for117 of 270 significantly expressed genes by GeneChip analysis.Of these, 41 were significantly differentially expressed betweensuperficial and invasive tumors on both array platforms. Thus, aset of 41 validated genes is provided, which using unsupervisedhierarchical cluster analysis, resulted in a clear separation ofsuperficial and invasive bladder cancers and defined a subgroupof tumors with adverse clinicopathological features (Fig. 4).

In summary, we used hierarchical cluster analysis on tumorspecimens of a cohort of uniformly treated patients with well-defined clinical outcome. Distinct gene expression profiles ofsuperficial and invasive tumors were observed. Furthermore, wecould identify distinct prognostic groups of invasive bladdercancer. The pattern of gene and tumor clusters was remarkablystable using different subsets of gene expression data, includinga set of 41 genes, which was validated using two different arrayplatforms. We also observed (a) a clonal origin of superficialtumors, (b) highly similar gene expression patterns in differentareas of invasive tumors, and (c) an invasive type of pattern inbladder mucosa derived from patients with locally advanceddisease. Our data provide additional insight into the molecularpathogenesis of bladder cancer offering the opportunity to detectnovel prognostic markers for superficial, invasive, and metas-tasizing disease.

ACKNOWLEDGMENTSWe thank Professor Wolfgang Schulz and Dr. Tracy Krage for

critical reading and discussion of the manuscript.

REFERENCES1. Jemal A, Murray T, Samuels A, Chafoor A, Ward E, Thun MJ.Cancer statistics, 2003. CA - Cancer J Clin 2003;53:5–26.2. Amling CL. Diagnosis and management of superficial bladder can-cer. Curr Probl Cancer 2001;25:219–78.3. Kurth KH, Bouffioux C, Sylvester R, van der Meijden AP, Ooster-linck W, Brausi M. Treatment of superficial bladder tumors: achieve-ments and needs. The EORTC Genitourinary Group. Eur Urol 2000;37(Suppl 3):1–9.4. van Tilborg AA, de Vries, A, de Bont M, Groenfeld LE, van derKwast TH, Zwarthoff EC. Molecular evolution of multiple recurrentcancers of the bladder Hum Mol Genet 2000;9:2973–80.5. Volante M, Tizzani A, Casetta G, Zitella A, Pacchioni D, BussolatiG. Progression from superficial to invasive carcinoma of the bladder:

genetic evidence of either clonal heterogeneous events. Hum Pathol2001;32:468–74.

6. Halachmi S, DeMarzo AM, Chow NH, et al. Genetic alterations inurinary bladder carcinosarcoma: evidence of a common clonal origin.Eur Urol 2000;37:350–7.

7. Hartmann A, Rosner U, Schlake G, et al. Clonality and geneticdivergence in multifocal low-grade superficial urothelial carcinoma asdetermined by chromosome 9 and p53 deletion analysis. Lab Investig2000;80:709–18.

8. Sidransky D, Frost P, Von Eschenbach A, Oyasu R, Preisinger AC,Vogelstein B. Clonal origin bladder cancer. N Engl J Med 1992;326:737–40.

9. Cheng L, Gu J, Ulbright TM, et al. Precise microdissection of humanbladder carcinomas reveals divergent tumor subclones in the sametumor. Cancer (Phila.) 2002;94:104–10.

10. Zieger K, Wolf H, Olsen PR, Hojgaard K. Long-term follow-up ofnoninvasive bladder tumours (stage Ta): recurrence and progression.BJU Int 2000;85:824–8.

11. Zieger K, Olsen PR, Wolf H, Hojgaard K. Long term follow-up ofsuperficial invasive bladder carcinoma with or without concomitantepithelial atypia: recurrence and progression. Scand J Urol Nephrol2002;36:52–9.

12. Chen SS, Chen KK, Lin AT, et al. The significance of tumour gradein predicting disease progression in stage Ta transitional cell carcinomaof the urinary bladder. Br J Urol 1996;78:209–12.

13. Brausi M, Collette L, Kurth K, et al. Variability in the recurrencerate at first follow-up cystoscopy after TUR in stage Ta T1 transitionalcell carcinoma of the bladder: a combined analysis of seven EORTCstudies. Eur Urol 2002;41:523–31.

14. Giuliani L, Giberti C, Martorana G, Bonamini A, Natta GD, RovidaS. Results of radical cystectomy for primary bladder cancer. Retrospec-tive study of more than 200 cases. Urology 1985;26:243–8.

15. Jimenez RE, Gheiler E, Oskanian P, et al. Grading the invasivecomponent of urothelial carcinoma of the bladder and its relationshipwith progression-free survival Am J Surg Pathol 2000;24:980–7.

16. Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in thetreatment of invasive bladder cancer: long-term results in 1054 patients.J Clin Oncol 2001;19:666–75.

17. Yu DS, Chang SY. The expression of oncoproteins in transitionalcell carcinoma: its correlation with pathological behavior, cell cycle anddrug resistance. Urol Int 2002;69:46–50.

18. Primdahl H, von der Maase H, Sorensen FB, Wolf H, Orntoft TF.Immunohistochemical study of the expression of cell cycle regulatingproteins at different stages of bladder cancer. J Cancer Res Clin Oncol2002;128:295–301.

19. Cordon-Cardo C, Zhang ZF, Dalbagni G, et al. Cooperative effectsof p53 and pRB alterations in primary superficial bladder tumors.Cancer Res 1997;57:1217–21.

20. Korkolopoulou P, Christodoulou P, Konstantinidou AE, Thomas-Tsagli E, Kapralos P, Davaris P. Cell cycle regulators in bladder cancer:a multivariate survival study with emphasis on p27Kip1. Hum Pathol2000;31:751–60.

21. Jankevicius F, Goebell P, Kushima M, Schulz WA, Ackermann R,Schmitz-Drager BJ. p21 and p53 Immunostaining and survival follow-ing systemic chemotherapy for urothelial cancer. Urol Int 2002;69:174–80.

22. Chen CC, Shieh B, Jin YT, et al. Microarray profiling of geneexpression patterns in bladder tumor cells treated with genistein.J Biomed Sci 2001;8:214–22.

23. Maxwell SA, Davis GE. Differential gene expression in p53-medi-ated apoptosis-resistant vs. apoptosis-sensitive tumor cell lines. ProcNatl Acad Sci USA 2000;97:13009–14.

24. Sanchez-Carbayo M, Socci ND, Charytonowicz E, et al. Molecularprofiling of bladder cancer using cDNA microarrays: defining histogen-esis and biological phenotypes. Cancer Res 2001;62:6973–80.

25. Sanchez-Carbayo M, Capodieci P, Cordon-Cardo C. Tumor sup-pressor role of KiSS-1 in bladder cancer: loss of KiSS-1 expression isassociated with bladder cancer progression and clinical outcome. Am JPathol 2003;162:609–17.

26. Sanchez-Carbayo M. Use of high-throughput DNA microarrays toidentify biomarkers for bladder cancer. Clin Chem 2003;49:23–31.

27. Bergkvist A, Ljungqvist A, Moberger G. Classification of bladdertumours based on the cellular pattern. Preliminary report of a clinical-pathological study of 300 cases with a minimum follow-up of eightyears. Acta Chir Scand 1965;130:371–8.

28. Grimm MO, Jurgens B, Schulz WA, Decken K, Makri D, Schmitz-Drager BJ. Inactivation of tumor suppressor genes and deregulation ofthe c-myc gene in urothelial cancer cell lines. Urol Res 1995;23:293–300.

29. Quackenbush J. Computational analysis of microarray data. Nature(Lond.) 2001;2:418–27.

30. Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysisand display of genome-wide expression patterns. Proc Natl Acad SciUSA 1998;95:14863–8.

31. Beissbarth T, Fellenberg K, Brors B, et al. Processing and qualitycontrol of DNA array hybridization data. Bioinformatics 2000;16:1014–22.

32. Hornberg JJ, de Haas RR, Dekker H, Lankelma J. Analysis ofmultiple gene expression array experiments after repetitive hybridiza-tions on nylon membranes. Biotechniques 2002;33:108–13.

33. Wang E, Miller LD, Ohnmacht GA, et al. Prospective molecularprofiling of melanoma metastases suggests classifiers of immune re-sponsiveness. Cancer Res 2002;62:3581–6.

34. Troyanskaya OG, Garber ME, Brown PO, Botstein D, Altman RB.Nonparametric methods for identifying differentially expressed genes inmicroarray data. Bioinformatics 2002;18:1454–61.

35. Reiner A, Yekutieli D, Benjamini Y. Identifying differentially ex-pressed genes using false discovery rate controlling procedures. Bioin-formatics 2003;19:368–75.

36. Irizarry RA, Hobbs B, Collin F, et al. Exploration, normalization,and summaries of high density oligonucleotide array probe level data.Biostatistics 2003;4:249–64.

37. Zhang L, Miles MF, Aldape KDA. model of molecular interactionson short oligonucleotide microarrays. Nat Biotechnol 2003;21:818–21.38. Korenberg MJ. Gene expression monitoring accurately predictsmedulloblastoma positive and negative clinical outcomes. FEBS Lett2003;533:110–4.39. van de Vijver MJ, He YD, van’t Veer LJ, et al. A gene-expressionsignature as a predictor of survival in breast cancer N Engl J Med2002;347:1999–2009.40. Beer DG, Kardia SL, Huang CC, et al. Gene-expression profilespredict survival of patients with lung adenocarcinoma. Nat Med 2002;8:816–24.41. Wigle DA, Jurisica I, Radulovich N, et al. Molecular profiling ofnon-small cell lung cancer and correlation with disease-free survival.Cancer Res 2002;62:3005–8.

42. Shipp MA, Ross KN, Tamayo P, et al. Diffuse large B-cell lym-phoma outcome prediction by gene-expression profiling and supervisedmachine learning. Nat Med 2002;8:68–74.43. Dhanasekaran SM, Barrette TR, Ghosh D, et al. Delineation ofprognostic biomarkers in prostate cancer. Nature (Lond.) 2001;412:822–6.44. Dyrskjot L, Thykjaer T, Kruhoffer M, et al. Identifying distinctclasses of bladder carcinoma using microarrays. Nat Genet 2003;33:90–6.45. Christensen M, Jensen MA, Wolf H, Orntoft TF. Pronounced mi-crosatellite instability in transitional cell carcinomas from young pa-tients with bladder cancer. Int J Cancer 1998;79:396–401.46. Thykjaer T, Workman C, Kruhoffer M, et al. Identification of geneexpression patterns in superficial and invasive human bladder cancer.Cancer Res 2001;61:2492–9.47. Primdahl H, von der Maase H, Sorensen FB, Wolf H, Orntoft TF.Immunohistochemical study of the expression of cell cycle regulatingproteins at different stages of bladder cancer. J Cancer Res Clin Oncol2002;128:295–301.48. Keleg S, Buchler P, Ludwig R, Buchler MW, Friess H. Invasion andmetastasis in pancreatic cancer. Mol Cancer 2003;2:14.49. Hsu MY, Meier F, Herlyn M. Melanoma development and progres-sion: a conspiracy between tumor and host. Differentiation 2002;70:522–36.50. Liotta LA, Kohn EC. The microenvironment of the tumour-hostinterface. Nature (Lond.) 2001;411:375–9.51. Seripa D, Parrella P, Galluci M, et al. Sensitive detection of tran-sitional cell carcinoma of the bladder by microsatellite analysis of cellsexfoliated in urine. Int J Cancer 2001;95:364–9.52. Rogojina AT, Orr WE, Song BK, Geisert EE Jr. Comparing the useof Affymetrix to spotted oligonucleotide microarrays using two retinalpigment epithelium cell lines. Mol Vis 2003;9:482–96.53. Kuo WP, Jenssen TK, Butte AJ, Ohno-Machado L, Kohane IS.Analysis of matched mRNA measurements from two different microar-ray technologies. Bioinformatics 2002;18:405–12.54. Li J, Pankratz M, Johnson JA. Differential gene expression patternsrevealed by oligonucleotide versus long cDNA arrays. Toxicol Sci2002;69:383–90.55. Yuen T, Wurmbach E, Pfeffer RL, Ebersole BJ, Sealfon SC. Ac-curacy and calibration of commercial oligonucleotide and custom cDNAmicroarrays. Nucleic Acids Res 2002;30:E48.56. Rhodes DR, Barrette TR, Rubin MA, Ghosh D, Chinnaiyan AM.Meta-analysis of microarrays: interstudy validation of gene expressionprofiles reveals pathway dysregulation in prostate cancer. Cancer Res2002;62:4427–33.57. Lee JK, Bussey KJ, Gwadry FG, et al. Comparing cDNA andoligonucleotide array data: concordance of gene expression across plat-forms for the NCI-60 cancer cells. Genome Biol 2003;4:R82.58. Tan PK, Downey TJ, Spitznagel EL Jr, et al. Evaluation of geneexpression measurements from commercial microarray platforms Nu-cleic Acids Res 2003;31:5676–84.

2004;10:3410-3421. Clin Cancer Res Olga Modlich, Hans-Bernd Prisack, Gerald Pitschke, et al. Array Analysis of Gene Expression ProfilesTransitional Cell Carcinoma of the Bladder: Use of cDNA Identifying Superficial, Muscle-Invasive, and Metastasizing

Updated version

http://clincancerres.aacrjournals.org/content/10/10/3410

Access the most recent version of this article at:

Cited articles

http://clincancerres.aacrjournals.org/content/10/10/3410.full#ref-list-1

This article cites 54 articles, 9 of which you can access for free at:

Citing articles

http://clincancerres.aacrjournals.org/content/10/10/3410.full#related-urls

This article has been cited by 11 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

SubscriptionsReprints and

.pubs@aacr.orgDepartment at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

Permissions

Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's

.http://clincancerres.aacrjournals.org/content/10/10/3410To request permission to re-use all or part of this article, use this link

identifying superficial, muscle-invasive, and ...transitional cell carcinoma of the bladder: use of...

Documents

genomic and cdna libraries

cdna library construction

· pdf fileliquid bladder liquid bladder hydril pipehugger...

cdna microarrays on glass slides

smarter® pico pcr cdna synthesis kit user manual · cdna...

cdna synthesis - home - gene target solutions · 2016. 6....

pcr dependent cdna library.pdf

cdna synthesis

orf cdna expression clones

cdna synthesis system - sigma-aldrich · cdna synthesis...

muscle invasive bladder cancer : bladder conservation...

thermostable cdna synthesis supermix (+gdna...

manual: cdna synthesis kit, zap-cdna synthesis … synthesis...

bladder-sparing trimodality therapy for muscle-invasive...

cdna microarrays

cdna cloning, heterogeneous expression and biochemical...

blueberry prevents the bladder dysfunction in bladder

kidney & bladder -...

smart race cdna amplification kit user manual - genex.cl ·...

sequenceanalysis of the cdna liver · liver glycogen...