Molecular analysis of transitional cell carcinoma using cDNA microarray


The incidence of transitional cell carcinoma (TCC), the fourth most common neoplasm diagnosed in men, is rising. Despite the development of several noninvasive diagnostic tests, none have gained full recognition by the clinicians. Gene expression profiling of tumors can identify new molecular markers for early diagnosis and disease follow-up. It also allows the classification of tumors into subclasses assisting in disease diagnosis and prognosis, as well as in treatment selection. In this paper, we employed expression profiling for molecular analysis of TCC. A TCC-derived cDNA microarray was constructed and hybridized with 19 probes from normal urothelium and TCC tissues. Hierarchical clustering analysis identified all normal urothelium samples to be tightly clustered and separated from the TCC samples, with 29 of the genes significantly induced (t-test, P<10−5) in noninvasive TCC compared to normal urothelium. The identified genes are involved in epithelial cells' functions, tumorigenesis or apoptosis, and could become molecular tools for noninvasive TCC diagnosis. Principal components analysis of the noninvasive and invasive TCC expression profiles further revealed sets of genes that are specifically induced in different tumor subsets, thus providing molecular fingerprints that expand the information gained from classical staging and grading.


Transitional cell carcinoma (TCC) of the bladder is the second most common malignancy of the urinary tract (Burchardt et al., 2000; Koenig et al., 2000). It is the fourth most prevalent cancer, after prostate, lung and colorectal, accounting for 7% of all cancer cases in men. In total, 56 500 newly diagnosed cases and 12 600 resultant deaths are estimated to have occurred during 2002 (Jemal et al., 2002). More than 70% of TCC tumors are superficial tumors limited to the bladder mucosa and lamina propria – Ta- or T1-staged tumors, of which many are well-differentiated low-grade tumors. Patients presenting with less-differentiated, large, multiple high-grade tumors are at the greatest risk for recurrence and development of invasive cancer. Following initial diagnosis and tumor resection, as many as 70% of the noninvasive tumors recur and 5–30% of the recurrent tumors progress (Brown, 2000; Burchardt et al., 2000). As a result, patients require regular follow-up examinations, which are associated with a significant impact on health-care resources and the patient's quality of life (Schamhart et al., 2000). Unfortunately, the means by which TCC is detected, either initially or during routine follow-up, have not changed over the last six decades, invariably requiring invasive procedures such as cystoscopy, uretheroscopy and biopsies.

To assist in the initial diagnosis of tumor and for the follow-up of TCC patients, several potential diagnostic and prognostic markers have been studied. However, none of these is widely used by the clinicians, mainly due to their limited sensitivity or specificity (Burchardt et al., 2000; Koenig et al., 2000; Kausch and Bohle, 2002). Discovery of new diagnostic tumor markers may improve screening and early detection, as well as identification of tumor recurrence. Furthermore, molecular classification of TCC tumors into clinically relevant subgroups may also facilitate the management of the disease.

Molecular diagnostics based on assays for a single protein may be inadequate to account for the inherent complexity of cancer and for the variability of both healthy and affected populations. In fact, for many cancers, no single test (molecular, cytological or histological) is sufficient to establish a diagnosis. cDNA microarrays, containing thousands of different genes, allow the simultaneous identification of multiple potential tumor markers, which could then serve as a basis for the development of a more comprehensive and robust molecular diagnostic assay (Golub et al., 1999; van't Veer et al., 2002; Giordano et al., 2003; Liu, 2003). Large-scale molecular analysis may simultaneously reveal new relationships and hierarchies within the complex population of tumors, and define new phenotypic classification (Macgregor and Squire, 2002; Mohr et al., 2002).

In this study, we have applied the microarray hybridization approach to the molecular analysis of TCC. Using expression profiling, we have identified potential markers for noninvasive TCC and studied the subclassification of the tumor population, as a basis for better, more refined diagnostics.


Separation of normal urothelium samples from tumor samples by hierarchical clustering of expression profiles

We prepared a custom TCC microarray (see Materials and methods) and hybridized it with 19 cDNA probes, six of which originated from normal urothelium and 13 from TCC samples (Table 1). Of the 13 TCC probes, 11 were obtained from noninvasive TCC samples: six Ta (G1 or G2), of which one was mixed Ta/T1, and five T1 (G2 or G3), of which one was mixed with carcinoma in situ (CIS). Two of the tumor samples originated from invasive G3 TCC. Unlike commonly used generic microarrays, only genes expressed in human bladder were printed in our microarray, improving the likelihood of identifying relevant molecular markers for TCC. All probes were hybridized against an identical common probe similar in composition to the TCC array, enabling efficient hybridization to the entire array. This design had two advantages. First, the hybridization intensities of all common probes were highly correlated, facilitating further comparison of all hybridization results, using the common probe as a reference. Second, it allowed us to examine the full variability within and between individual healthy and pathological samples.

Table 1 Clinical data

We filtered the expression profiles based on quality criteria (for details, see Materials and methods), resulting in 6688 array elements, and hierarchically clustered the filtered expression profiles (Figure 1). The clustering revealed a major branch in which the six normal urothelium samples formed a tight cluster, and four high-grade TCC samples (either T1 or invasive) were more distant. Another major branch included all other (except one) tumor samples. Within this tree branches, all Ta TCC samples were clustered together while a higher variability was observed with other TCC samples. Similar separation of the Ta samples was obtained by principal components analysis (PCA) (see below).

Figure 1

Hierarchical clustering of expression profiles. Expression profiles were clustered by average linkage hierarchical clustering, using Pearson's correlation coefficient as the distance measure. Clustering results of 19 hybridizations to 6688 microarray clones are provided (see Table 1 for clinical data). Branch length is inversely proportional to degree of similarity

Potential diagnostic markers for noninvasive TCC

Although three of the noninvasive TCC samples clustered distantly from the rest, most of the noninvasive samples were clustered together, and all were clearly distinct from the normal urothelium ones. We therefore utilized the hybridization data to identify the genes that discriminated between noninvasive TCC and normal urothelium samples. To select such genes, we employed Student's t-test on the results of 17 hybridizations: six with normal urothelium samples and 11 with noninvasive TCC samples. In all, 6688 array elements were included in this analysis. Since clone redundancy was not evaluated prior to the statistical analysis, we anticipated a possible bias in favor of upregulated and hence over-represented genes (although we have employed the SDGI method to reduce redundancy, see Materials and methods). Indeed, postanalysis sequencing indicated some redundancy in the clone set. However, as our goal was to identify potential upregulated markers, this bias cannot be regarded as a major problem.

In all, 248 cDNA clones were identified that discriminated the normal urothelium samples from noninvasive TCC samples (P<10−5), of which 115 were increased in TCC, representing 29 genes that were upregulated in noninvasive TCC (namely, the signal of which was significantly higher in the TCC hybridizations compared to the normal urothelium hybridizations). These genes can facilitate diagnosis of noninvasive tumors in spite of the background of nonmalignant cells. These significantly upregulated genes, shown in Figure 2, were functionally classified using databases and literature searches. Included in this list are many epithelial cells' markers such as representatives of the keratins (KRT7, KRT8, KRT18 and KRT19) and cathepsin E that were already suggested for TCC diagnostics (Yamamoto et al., 1996; Southgate et al., 1999) and two genes that belong to the S100 protein family, all of which function in dedifferentiation and keratinization of epithelial cells (Mota et al., 1997; Donato, 2003). Genes related to epithelial cells' adhesion were identified (syndecan-1, HAS3, nectin-1, nectin-4 and claudin-7), some of which were already reported to enhance tumor cell growth (e.g. HAS3; Liu et al., 2001a), and others were implicated in bladder cancer (E-cadherin; Giroldi et al., 2000). Genes involved in other functions of epithelial cells were also selected (e.g. the tyrosine kinase DDR1, found to be overexpressed in brain neoplasms (Weiner et al., 2000) and the small Ras-like GTP-binding protein Rab25). Genes involved in the regulation of proliferation were identified: the FOXQ1 transcription factor, which is known to be overexpressed in colon and lung cancer (Bieller et al., 2001), CDK4, reported to be amplified in bladder tumors (Simon et al., 2002) and laminin receptor 1, which is related to tumor progression but not previously suggested as potential TCC marker (Menard et al., 1998). We also identified N33, a tumor suppression gene (Levy et al., 1999), and several apoptosis-related genes (e.g. caspase-4 and quinone-1, the latter was previously shown to be overexpressed in human bladder cancer; Choudry et al., 2001). Overall, of the 23 known genes identified as significantly upregulated in TCC, eight were previously related to bladder carcinoma. The remaining potential noninvasive TCC marker genes encode for novel proteins with as yet unknown function.

Figure 2

Potential diagnostic markers for noninvasive TCC. In all, 29 genes that showed statistically significant overexpression in noninvasive TCC (P<10−5) are shown (rows). The differential expression level for each of the hybridization samples (columns) is provided relative to the common reference probe (induced – red; unchanged – black; reduced – green). Genes' names are listed. 1genes previously reported to be related to TCC (see text for references). Accession number denotes Genebank's database accession number. Classification into functional groups was as reported in the literature and public databases. Signal (average) depicts the average of Cy5 signal level for each of the genes in hybridizations with either normal urothelium samples or noninvasive TCC

Semiquantitative RT–PCR

To further validate the microarray-based findings, we performed semiquantitative RT–PCR experiments with S100P, keratin 7, CDK4, HAI-2 and syndecan-1, representatives of different functional groups identified as noninvasive TCC markers. Cyclophilin A (peptidylprolyl isomerase A) was used as internal control (Feroze-Merzoug et al., 2002). Two normal urothelium samples and five of the noninvasive TCC tumor samples (representatives of Ta and T1 tumors of various grades) were tested. The results confirmed the pattern of expression identified by the microarray hybridization and show higher expression of the genes in most of the TCC samples compared to those of the normal urothelium (Figure 3). Importantly, consistent with the hybridization data, the RT–PCR results indicate the variability in the expression of single gene markers. This further emphasizes the need for multigene diagnostics to account for the inherent complexity of cancer, as well as for the variability of both healthy and affected populations.

Figure 3

Validation of overexpression in noninvasive TCC tumors. Semiquantitative RT–PCR was used to amplify fragments of S100P, keratin 7, CDK4, HAI-2 and syndecan-1, from mRNAs of normal urothelium (TC36 and TC46) and five noninvasive TCC samples (TC29, TC31, TC32, TC33, TC43). Cyclophilin A served as internal control in the PCR reactions of S100P, keratin 7 and CDK4, while in the cases of HAI-2 and syndecan-1, a separate PCR reaction was performed for each of the genes and for cyclophilin A, due to PCR–primer interference

Inner tumor classification

We next focused on the variability between the different tumor samples. To this end, we employed PCA to classify the tumors and to identify the genes that underlie this classification. PCA reduces the complex expression profiles to a few linear combinations of factors, where each factor represents a prominent feature in the data set and each of the hybridizations receives a specific coefficient (factor loading) for each of the factors.

PCA identified four main factors. The factor loadings for each of the hybridizations were examined in order to characterize the different factors and identify the underlying classification (Table 2). The four Ta lower-grade tumors had particularly high loadings (0.56–0.77) for factor 1, while other tumors (higher-grade T1 or invasive ones) did not have. In addition to this separation of Ta low-grade tumors (also evident in hierarchical clustering), the three other factors provided a detailed intratumor classification. Factors 2 and 3 were specific for high-grade tumors of either T1 or invasive type. Three such tumors were included in factor 2 (loadings of 0.57–0.80) while two high-grade, either T1 or invasive tumors (one of which was a T1 mixed with CIS), were included in factor 3 (loadings of 0.73–0.76). Factor 4 represented three noninvasive tumors of Ta, Ta/T1 or T1 stages with either G1 or G2 grading (with loadings of 0.59–0.76). One of the TCC samples had no conclusive loading and could not be classified into any of the factors. We concluded that factor 1 represents one ‘class’ of low-grade noninvasive tumors, factor 4 a different subclass of noninvasive tumors, while factors 2 and 3 represent two subclasses of high-grade TCC.

Table 2 Tumor classification by pathological information and by PCA

To identify the contribution of each gene to this classification, we used a regression procedure and calculated factor scores for each of the genes. A gene's factor score depicts the contribution of that gene's expression to the sample ‘class’ represented by this factor, and allows us to identify the most essential genes. Overall, 376 elements on the TCC array highly contributed (2.5 loading in absolute values) to at least one of the factors: 93 to factor 1 (low-grade noninvasive tumors), 106 to factor 2 (high-grade tumors), 101 to factor 3 (high-grade tumors) and 76 to factor 4 (non-invasive tumors). Note that 108 of these array elements significantly contributed to more than one factor. Detailed description of these results is available in the web supplement (

The identified elements were sequenced and found to represent 103 distinct genes. We then focused on the genes that received the highest positive score (2.5 and above) in each of the factors most of which were also highly expressed in the tumors related to their factor, and are more likely to serve as effective markers (unlike negatively scoring, downregulated, genes). Results are presented in Figure 4. Genes involved in dedifferentiation of epithelial cells were the major functional group contributing to factor 1 (Ta low-grade tumors). Three of them were also selected as potential markers for noninvasive tumors (keratin 7, cathepsin E and S100P). Other genes related to epithelial cells' functions (aquaporin 3 and laminin S B3) as well as genes related to immune response and proliferation (annexin 8 and phospholipase A2) were also identified to contribute to factor 1. The major positively contributing genes in factor 2 were already related to cancer, some of which were also previously associated with bladder cancer (e.g. G3BP, Liu et al., 2001b; GSTM1, Engel et al., 2002; VEGF, Turner et al., 2002). Factor 3-related tumors showed high scores (and vast overexpression) of immune response-related genes along with increase in the expression of detoxification and extracellular matrix degradation enzymes (e.g. GSTP1 and MMP2). Factor 4, which represented a potential subclass of noninvasive tumors (of low–medium grade), included GAPDH, Nrf2 and MUC1, which are known to be connected to cancer progression (Vila et al., 2000; Simms et al., 2001; Kwak et al., 2002). Note that nearly all the genes that contributed highly to factor 4 were not essential for any other factor, suggesting that a distinct genes' set underlies the formation of this factor representing a subset of the noninvasive TCC tumors.

Figure 4

Genes that obtained the highest positive factor scores within the four identified PCA factors. Differential expression levels of 32 genes (rows) in 12 hybridizations with TCC samples (columns) are shown (induced – red; unchanged – black; and reduced – green), as calculated compared to the common reference probe used in all hybridizations. Samples and genes (having the highest positive factor scores) in each factor are marked with a yellow border. Genes' names, their accession numbers (from Genebank's database) and the factor scores are shown. All highest scores (2.5 in absolute numbers) are marked in bold letters; positive scores connected to each of the factors are marked with yellow border; *Genes that were also suggested as potential markers for noninvasive TCC


Patients with noninvasive TCC, the most common type of TCC tumors, are subjected to a lifetime follow-up program including frequent invasive examinations. For these patients, noninvasive diagnostic tools are particularly important. Several such tests are available (Koenig et al., 2000), all plagued by low specificity and/or low sensitivity. To identify genes that could serve as molecular markers for noninvasive TCC, and to further extend the classification of TCC tumors at the molecular level, we conducted an expression profiling experiment employing normal urothelium and TCC samples. A TCC microarray that contained bladder-specific and tumor-derived sequences was constructed. We used 19 hybridization probes extracted from either normal urothelium-derived samples or TCC tumors, at different stages and grades. Hierarchical clustering identified a tight cluster of the normal samples, while more variability was observed between the different tumors, of which the Ta TCC hybridizations formed a separate cluster. The expression profiles obtained clearly demonstrate the feasibility of molecular analysis of bladder cancer based on the hybridization results.

We next identified potential TCC markers suitable for early diagnostics and patients' surveillance. Since the majority of TCC cases are diagnosed at the noninvasive stage, we compared the gene expression profiles of normal urothelium samples and noninvasive TCC samples (excluding invasive tumors also because their global gene expression pattern was deviant from most noninvasive samples). In total, 17 hybridizations were subjected to statistical analysis and 29 genes were identified to be significantly (P<10−5) upregulated in noninvasive TCC compared to normal urothelium (Figure 2).

The identified genes vary in their diagnostic potential. Some represent secreted or membranal proteins while others code for intracellular proteins. Proteins that are secreted, like HAI-2 and syndecan-1 (both of which are overexpressed in several cancers; Bayer-Garner et al., 2000; Dobra et al., 2000; Itoh et al., 2000; Kataoka et al., 2000), are suitable for the development of a noninvasive urine-based diagnostic test for TCC. Alternatively, an RT–PCR-based test on urine samples may be developed for genes coding for intracellular proteins, since the transitional epithelium cells are exfoliated in the urine in both normal and disease conditions (Rotem et al., 2000). Many of the genes that were shown to be overexpressed in noninvasive TCC compared to normal urothelium are characteristic of dedifferentiated epithelial cells, some of which (keratins 7, 8, 18 and 19) have already been suggested as TCC markers (Southgate et al., 1999). A diagnostic assay for keratin 20 based on RT–PCR of urine samples has been reported, confirming the potential of this technical approach for diagnosis (Rotem et al., 2000). Genes connected to other epithelial cells' functions (e.g. HAS3 that is related to cell adhesion) were also significantly upregulated in noninvasive TCC. Following individual validation in a large number of TCC patients, these identified markers can be utilized individually or in combination, as the basis for the production of a diagnostic test for noninvasive TCC tumors.

Note that some of the identified markers were connected to proliferation and cell cycle regulation (e.g. CDK4), while others were associated with apoptosis and tumor suppression processes (e.g. caspase-4). The overexpression of such seemingly contradictory factors probably contributes to the progression of the normal transitional urothelium to carcinoma.

Although most of the identified genes are already known as being associated with cancer, the majority of them have not been reported as induced in TCC. We therefore validated representatives of different functional subgroups by semiquantitative RT–PCR. This independent validation supported the hybridization results, while highlighting the variability between samples and the importance of a diagnostic assay based on multiple markers.

A different diagnostic and prognostic challenge is the classification of the tumors into accurate subgroups. Despite the existence of clear histopathological criteria, the grade and stage of TCC is currently still somewhat subjectively determined by the pathologist. To gain insight into the classification of TCC tumors and its underlying molecular causes, we applied PCA to the gene expression profiles of different TCC samples. The results obtained were in general concordance with the classical histological staging and grading of the disease while revealing potential subclasses (Table 2). Four main factors were identified: two represented the lower-grade tumors (either Ta low-grade or Ta/T1 medium-grade tumors), and two factors represented the high-grade tumors (either T1 or invasive tumors). Importantly, we scored the genes for their contribution to each of the factors. We focused on genes with better diagnostic potential, namely those that were upregulated in the tumors represented by each factor and received high positive factor-specific scores (Figure 4).

Genes involved in epithelial cells' functions (e.g. differentiation, adhesion) received the highest scores in factor 1, of which five showed an opposite expression pattern in the other factors. Examples are aquaporin 3 and cathepsin E, both mark epithelial cells, which were reduced in factors 2- and 3-associated tumors, respectively. The identified opposite regulation pattern of these genes in the different tumors may specifically facilitate their use as molecular classification markers. Moreover, three of the epithelial cells' markers – keratin 7, cathepsin E and S100P that distinguished between factor 1 Ta low-grade tumors and other factors (i.e. tumors of higher grade and stage) – also differentiated between normal urothelium and noninvasive TCC samples. Thus, these genes may be used as diagnostic tools for both population screening and follow-up of patients with previously identified Ta low-grade tumors.

All other lower-grade (G1 or G2) noninvasive tumors (except one sample that could not be classified to any of the four factors) were included in factor 4. This factor was characterized by genes associated with tumor progression and hypoxia (e.g. MUC1, GAPDH). Unlike factor 1, epithelial cells' markers were not significant in this factor. The higher contribution of genes associated with differentiation and keratinization of urothelial cells to factor 1-related tumors is in concordance with their histological typing (low-grade papilloma tumors). We concluded that the segregation of the lower-grade noninvasive tumors into two different factors defines a molecular separation of all these TCC samples, possibly implying a more advanced state of the factor 4-related tumors.

The high-grade T1 and the two invasive tumors were represented by two factors, again indicating that tumors with identical pathological diagnosis may fall into two separate molecular subclasses. Genes related to the immune response (e.g. IgG family, α-2-macroglobulin) were prominent in factor 3-related tumors. Also included were genes associated with degradation of extracellular matrix (MMP2 and TIMP1) and detoxification (GPX2 and GSTP1). The connection of these gene families to cancer progression and invasiveness is well established. The increased expression of immune response-related genes is characteristic of inflammation, a phenomenon that cannot be deduced solely from the tumors' histopathological report.

Different genes contributed to the formation of factor 2. The highest positive score was allocated to HREV-H- and IL18-binding protein, both showing negative PCA scores in factor 3 tumors. These two genes may define a molecular subset of advanced TCC tumors. Note that different overexpressed detoxification enzymes (e.g. GSTM1 and GSTP1) obtained high scores in different factors (2 and 3, respectively) representing advanced tumors. Since the analysis described here was not designed to address the relationship of expression profiling and clinical outcome, the information on the treatment regimen for these patients was unfortunately not available. Thus, a more elaborate analysis of the possible drug effects on the expression of detoxification genes was impossible and will require further studies.

Our analysis shows that some of the high-grade noninvasive T1 tumors exhibited common molecular features with the invasive TCC, which were not characteristic of noninvasive Ta tumors (as suggested by the contribution of some T1 tumors along with invasive tumors to the formation of factors 2 and 3). This identified subclassification of high-grade tumors from different tumors stages may indicate differences in their prognosis and may hence influence their therapeutic strategy.

Some of the gene families differentiating between bladder tumors (e.g. genes related to cell proliferation, cell adhesion, immune response, growth factors) have already been reported in a study based on hybridization of pools of tumor samples each representing different stages and grades (Thykjaer et al., 2001). Although supportive, our work strongly indicates that pooling of several tumor samples for microarray analysis, based on classical staging and grading, may be misleading and hence may reduce the value and reliability of information obtained from expression profiling experiments. Moreover, our results suggest that tumors that were identically classified by histopathological analysis may be characterized by a set of completely different molecular fingerprints. A classification of bladder tumors based on microarray hybridization data of independent samples has been recently published (Dyrskjot et al., 2003). Clustering analysis identified three major groups of hybridization probes, as suggested by the pathological stages of the different samples used (Ta, T1 and invasive) with further classification of the Ta tumors into molecular subgroups. Although the number of tumors studied by us was smaller, we have also observed a major separation of TCC Ta tumors from more advanced tumors with subclassification of the Ta tumors. Unlike this previous study, our analysis also identified molecular subgroups within the noninvasive tumors, which classifies the different T1 tumors mainly according to their grade. Moreover, by combining PCA with regression procedures, we identified the major gene groups that contribute to each tumor subclass, an approach most suitable for the analysis of large number of comparable microarray expression profiling experiments.

The potential TCC marker genes and tumor classification presented in this study are the first step towards the development of diagnostic and prognostic assays for TCC. As the specific groups of genes were clearly associated by PCA with certain tumor phenotypes, assessment of their expression in a larger set of tumors will be indicative for the diagnostic aspects of interest (invasiveness, grade, prognosis, etc). In the next step, the identified genes will be used for the construction of a small diagnostic microarray and their expression level will be monitored in a large group of urine samples from TCC and non-TCC patients. This study – with a large number of samples, but a smaller number of genes – will allow us to select subsets of genes that are particularly suitable for noninvasive diagnostics and prognostics of TCC.

Materials and methods

Sample collection and RNA preparation

TCC tumors and normal urothelium samples were obtained with informed consent from patients who underwent surgery (either transurethral resection of bladder tumor for TCC samples or total nephrectomy for normal urothelium samples) in the Carmel or the Bnei Zion Hospitals, Haifa, Israel. The collected urothelium samples were immediately frozen in liquid nitrogen until used for RNA extraction. Tumor samples with marked inflammation or necrosis were excluded from further analysis. All tumor samples included in this study had accompanying histopathological data. Total RNA was extracted using the Eazy RNA isolation kit (Beit H'aemek). PolyA RNA was purified using the Oligotext Poly A RNA isolation kit (Qiagen) from the nondegraded total RNA samples. Of these samples, 20 were used for construction of the TCC microarray and 47 for the preparation of hybridization probes (see below).

Construction of TCC microarray

A pool of RNA extracted from 10 different normal urothelium samples and a pool of 10 different TCC samples was used as a source for cDNA libraries. Two different cDNA libraries were prepared in order to optimize the composition of the microarray. One library was based on the proprietary Sequence Dependent Gene Identification technique (SDGI library, USA patent no. 6 468 749, see web supplement Briefly, the method is based on the digestion of cDNA with type II restriction enzymes that produce a four-base overhang (e.g. BbsI). The digested cDNA pool is then ligated to a mix of 64 adaptors that cover the entire repertoire of three of the four nucleotides in the overhang. The resulting cDNA clones are next amplified in 64 sets using each of the adapters and an oligo-dT-based anchor as amplification primers. The complexity of the amplified pools is examined by sample sequencing. For printing on the microarray, more clones are taken from highly complex pools, while from pools of low complexity (in which highly expressed genes are represented) only few clones are taken. The end result is a set of clones with low redundancy in which rare genes are also represented. Since normal urothelium and TCC samples (10 of each) were pooled and used for the preparation of the cDNAs, this SDGI library contained sequences representatives of genes expressed in both cancer and normal bladder tissue. The second library was prepared using the Suppression Subtractive Hybridization method (SSH method) using a PCR-Select cDNA subtraction kit (Clontech) by a unidirection subtraction of the cDNA pool of 10 normal urothelium samples from the cDNA pool of 10 TCC samples that yielded a library enriched with TCC-specific cDNA clones. A total of 9930 clones derived from both libraries were printed on the TCC microarray (5000 clones from the SSH library and 4930 clones from the SDGI library). In all, 52 ‘cancer control’ cDNA clones derived from the genes known to be involved in cancer in general and in TCC specifically, all selected from the IMAGE database and sequence verified, were also printed on the TCC microarray. A list of these ‘cancer control’ genes is provided (see web supplement However, due to the low printing quality of these specific plasmid DNA-derived cDNA clones, they could not be used for hybridization analysis.

Printing was carried out as published (Yue et al., 2001). The cDNA clones derived from the libraries were not sequenced prior to the printing and only the clones displaying a statistical relevance were sequenced following the analysis of the hybridization data.

Probe preparation and hybridization scheme

A total of 19 probes derived from 32 normal urothelium, noninvasive and invasive TCC were analysed by hybridization on the TCC microarray (Table 1). Six ‘normal’ probes were derived from 19 normal urothelium samples, of which two probes were obtained from a single donor, three probes from a pool of four different donors and another probe from a pool of five donors. The pooling of normal urothelium samples was due to the limited amount of available material. A total of 13 ‘TCC’ probes were derived from individual tumor samples, representing various TCC tumors: six were Ta (G1 or G2), of which one was mixed Ta/T1 with G2 grade; five were T1, of which one was mixed with CIS with either G2 or G3 grade; two were invasive G3 tumors. In total, 200 ng of polyA RNA for each of these probes was used for cDNA preparation using Superscript Reverse Transcriptase (Life Technologies), primed with 18-mer oligo-dT. The cDNA probes were labelled with Cy5-dCTP as previously described (Yue et al., 2001).

mRNA prepared from a pool of RNA extracted from 15 samples (10 TCC samples and five normal urothelium samples) was used for preparation of a reference probe, which was labelled with Cy3-dCTP and introduced in all the hybridizations. Hybridization, washing and scanning of the slides were carried out as previously described (Yue et al., 2001).

Data acquisition and statistical analysis

All differential expression values were log transformed and balanced, essentially as previously described (Yue et al., 2001). Briefly, the differential values were calculated only if the signal to background ratio exceeded 2.5, the signal intensity was above 200 U in at least one of the two channels and the element diameter exceeded 40% of the mean element diameter for the array. In all, 6688 array elements have met these criteria and were included in subsequent analysis. It is worth noting that while one of our library preparation methods (SDGI, above) was designed to reduce redundancy, some of the clones were expected to be redundant. As the printed clones were not sequenced prior to hybridization analysis, this redundancy could have produced some biasing effect in favor of upregulated genes.

Reproducibility of the signal of the common reference probe in all hybridizations was measured by Pearson's correlation matrix. All the hybridizations showed a correlation coefficient above 0.95, suggesting good reproducibility between the hybridizations, thus enabling analysis and comparison of data within the whole set of hybridizations. The hybridization results as well as the correlation matrix results of the common probe hybridizations are available in web supplement (

A t-test was used to compare the hybridization results of the noninvasive TCC samples to the normal urothelium samples. To overcome the multicomparison problem, a threshold of significance of P<10−5 was used, which corresponds to P<0.05 following the Bonferroni adjustment. To assess the molecular classification of the bladder tumors, PCA was performed (Statistica, StatSoft). This unsupervised method reduces the data to a few linear combinations of the original variables without any a priori assumptions. The new composite variables, termed factors, represent major features of the hybridization set. The original hybridization results of all tumor samples (prefiltered 6688 array elements) were expressed in terms of the new PCA factors, and factor loadings (the relative contribution of each of the hybridizations to a given factor) were calculated. Using a regression procedure (Statistica), the factor loadings and the original expression raw data, we further computed a factor score for each gene, as a quantitative measure of the importance of the gene to the factor. Thus, genes with the highest correlation to a given factor and showing extreme positive or negative differential value in the hybridizations forming the factor received a high positive or negative score, respectively.

Bioinformatics analysis of the selected clones

All clones that were selected by the statistical analysis were fully sequenced and annotated using BLAST ( searches to Genebank's nonredundant genomic and nongenomic nucleotide databases, the nonredundant protein database and the EST database; the most informative annotation was chosen for each clone. Genes were grouped based on their function by searching PubMed, SwissProt ( and GeneCards (

Semiquantitative RT–PCR

Total RNA (1 μg) from two normal and five tumor samples (see legend to Figure 3) was reverse transcribed with oligo-dT and ThermoScript polymerase (Invitrogen). RT product (0.5 μl) was amplified by using 1 U Super-Therm polymerase (JMR-Holding) in a final volume of 10 μl. Keratin 7, HAI-2, CDK4, S100P, syndecan-1 and cyclophilin A (latter chosen as control) were analysed. PCR reactions were cycled for either 20 (KRT7, S100P and cyclophilin A) or 23 (CDK4, HAI-2 and syndecan-1 each together with cyclophilin A) times under the following conditions: 94 for 30 s, 57 for 30 s and 72 for 1 min, followed by final extension at 72 for 10 min. PCR products were visualized on 1.8–2% agarose gels stained with ethidium bromide. The sequences of the PCR primer pairs of these genes were: keratin 7 (forward 5′-IndexTermGCAGCTGCGTGAGTACCAG-3′, reverse 5′-IndexTermTCAGTCGCGGGCACTCCTG-3′); syndecan-1 (forward 5′-IndexTermTGGTCGGGAGACAGCATCAG-3′, reverse 5′-IndexTermGCCACAGGAGCTAACGGAG-3′); HAI-2 (forward 5′-IndexTermAAGGTGGTGGTTCTGGCGG-3′, reverse 5′-IndexTermTTCCC CAGTCCTCTTGGCG-3′); CDK4 (forward 5′-IndexTermTCCCAA TGTTGTCCGGCTGATG-3′, reverse 5′-IndexTermCTGGAGGCAGC CCAATCAGGTC-3′); S100P (forward 5′-IndexTermAGAAGGAGCTACCAGGCTTC-3′, reverse 5′-IndexTermTGGGAAGCCTGGGACCATG-3′) and cyclophilin A (forward 5′-IndexTermCGTCTCCTT TGAGCTGT-3′, reverse 5′-IndexTermTCGAGTTGTCCACAGTCA-3′).


  1. Bayer-Garner IB, Dilday B, Sanderson RD and Smoller BR . (2000). Am. J. Dermatopathol., 22, 119–122.

  2. Bieller A, Pasche B, Frank S, Glaser B, Kunz J, Witt K and Zoll B . (2001). DNA Cell Biol., 20, 555–561.

  3. Brown FM . (2000). Urol. Clin. N. Am., 27, 25–37.

  4. Burchardt M, Burchardt T, Shabsigh A, De La Taille A, Benson C and Sawczuk I . (2000). Clin. Chem., 46, 595–605.

  5. Choudry GA, Stewart PA, Double JA, Krul MR, Naylor B, Flannigan GM, Shah TK, Brown JE and Phillips RM . (2001). Br. J. Cancer, 85, 1137–1146.

  6. Dobra K, Andang M, Syrokou A, Karamanos NK and Hjerpe A . (2000). Exp. Cell Res., 258, 12–22.

  7. Donato R . (2003). Microsc. Res. Technol., 60, 540–551.

  8. Dyrskjot L, Thykjaer T, Kruhoffer M, Jensen JL, Marcussen N, Hamilton-Dutoit S, Wolf H and Orntoft TF . (2003). Nat. Genet., 33, 90–96.

  9. Engel LS, Taioli E, Pfeiffer R, Garcia-Closas M, Marcus PM, Lan Q, Boffetta P, Vineis P, Autrup H, Bell DA, Branch RA, Brockmoller J, Daly AK, Heckbert SR, Kalina I, Kang D, Katoh T, Lafuente A, Lin HJ, Romkes M, Taylor JA and Rothman N . (2002). Am. J. Epidemiol., 156, 95–109.

  10. Feroze-Merzoug F, Berquin IM, Dey J and Chen YQ . (2002). Biotechniques, 32, 776–778.

  11. Giordano TJ, Thomas DG, Kuick R, Lizyness M, Misek DE, Smith AL, Sanders D, Aljundi RT, Gauger PG, Thompson NW, Taylor JM and Hanash SM . (2003). Am. J. Pathol., 162, 521–531.

  12. Giroldi LA, Shimazui T, Schalken JA, Yamasaki H and Bringuier PP . (2000). Morphologie, 84, 31–38.

  13. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, Bloomfield CD and Lander ES . (1999). Science, 286, 531–537.

  14. Itoh H, Kataoka H, Tomita M, Hamasuna R, Nawa Y, Kitamura N and Koono M . (2000). Am. J. Physiol. Gastrointest. Liver Physiol., 278, G635–G643.

  15. Jemal A, Thomas A, Murray T and Thun M . (2002). CA Cancer J. Clin., 52, 23–47.

  16. Kataoka H, Itoh H, Uchino H, Hamasuna R, Kitamura N, Nabeshima K and Koono M . (2000). Cancer Lett., 148, 127–134.

  17. Kausch I and Bohle A . (2002). Eur. Urol., 41, 15–29.

  18. Koenig F, Jung K, Schnorr D and Loening SA . (2000). Clin. Chim. Acta, 297, 191–205.

  19. Kwak MK, Wakabayashi N, Itoh K, Motohashi H, Yamamoto M and Kensler TW . (2002). J. Biol. Chem., 27, 27.

  20. Levy A, Dang UC and Bookstein R . (1999). Genes Chromosomes Cancer, 24, 42–47.

  21. Liu ET . (2003). Curr. Opin. Genet. Dev., 13, 97–103.

  22. Liu N, Gao F, Han Z, Xu X, Underhill CB and Zhang L . (2001a). Cancer Res., 61, 5207–5214.

  23. Liu Y, Zheng J, Fang W, You J, Wang J, Cui X and Wu B . (2001b). Chin. Med. J. (Engl.), 114, 35–38.

  24. Macgregor PF and Squire JA . (2002). Clin. Chem., 48, 1170–1177.

  25. Menard S, Tagliabue E and Colnaghi MI . (1998). Breast Cancer Res. Treat., 52, 137–145.

  26. Mohr S, Leikauf GD, Keith G and Rihn BH . (2002). J. Clin. Oncol., 20, 3165–3175.

  27. Mota F, Kanan JH, Rayment N, Mould T, Singer A and Chain BM . (1997). Am. J. Pathol., 150, 1223–1229.

  28. Rotem D, Cassel A, Lindenfeld N, Mecz Y, Sova Y, Resnick M and Stein A . (2000). Eur. Urol., 37, 601–604.

  29. Schamhart DH, de Boer EC, de Reijke TM and Kurth K . (2000). Eur. Urol., 37, 16–23.

  30. Simms MS, Murray A, Denton G, Scholfield DP, Price MR, Perkins AC and Bishop MC . (2001). Urol. Res., 29, 13–19.

  31. Simon R, Struckmann K, Schraml P, Wagner U, Forster T, Moch H, Fijan A, Bruderer J, Wilber K, Mihatsch MJ, Gasser T and Sauter G . (2002). Oncogene, 21, 2476–2483.

  32. Southgate J, Harnden P and Trejdosiewicz LK . (1999). Histol. Histopathol., 14, 657–664.

  33. Thykjaer T, Workman C, Kruhoffer M, Demtroder K, Wolf H, Andersen LD, Frederiksen CM, Knudsen S and Orntoft TF . (2001). Cancer Res., 61, 2492–2499.

  34. Turner KJ, Crew JP, Wykoff CC, Watson PH, Poulsom R, Pastorek J, Ratcliffe PJ, Cranston D and Harris AL . (2002). Br. J. Cancer, 86, 1276–1282.

  35. Van't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven RM, Roberts C, Linsley PS, Bernards R and Friend SH . (2002). Nature, 415, 530–536.

  36. Vila MR, Nicolas A, de Morote JI and Meseguer A . (2000). Cancer, 89, 152–164.

  37. Weiner HL, Huang H, Zagzag D, Boyce H, Lichtenbaum R and Ziff EB . (2000). Neurosurgery, 47, 1400–1409.

  38. Yamamoto S, Yonezawa S, Ichinose M, Miki K, Masui T, Fukushima S, Inoue H and Tatematsu M . (1996). Virchows Arch., 427, 589–594.

  39. Yue H, Eastman PS, Wang BB, Minor J, Doctolero MH, Nuttall RL, Stack R, Becker JW, Montgomery JR, Vainer M and Johnston R . (2001). Nucleic Acids Res., 29, E41–E51.

Download references


We are grateful to Dr Paz Einat for helpful discussions, Dr Hagar Kalinsky and Dr Tania Fucs for their contribution to the bioinformatic analysis, Ms Ofra Oron for her excellent technical assistance and Ms Carol Borowitz for her kind help in preparing the paper.

Author information



Corresponding author

Correspondence to Orna Mor.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Mor, O., Nativ, O., Stein, A. et al. Molecular analysis of transitional cell carcinoma using cDNA microarray. Oncogene 22, 7702–7710 (2003).

Download citation


  • transitional cell carcinoma
  • cDNA microarray
  • molecular markers
  • expression profiling
  • tumor classification

Further reading