Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Molecular characterization of early-stage bladder carcinomas by expression profiles, FGFR3 mutation status, and loss of 9q

Abstract

We used gene expression profiling, mutation analyses of FGFR3 and TP53, and LOH analyses of chromosome 9 and the TP53 region on chromosome arm 17p, to molecularly characterize 75 Ta and T1 bladder carcinomas. We identified four major cellular processes related to cell cycle, protein synthesis, immune response, and extra cellular components that contribute to the expressional heterogeneity of early-stage urothelial cell carcinoma (UCC). Activating FGFR3 mutations were found at the highest frequency in G1 tumors (80%), and showed a strong correlation with FGFR3 expression. In contrast, G3 tumors displayed mutations in less than 10% of the cases and a low level of FGFR3 expression. Even though LOH on chromosome 9 was not associated with any specific expression pattern, our data indicate that loss of chromosome 9 is associated with tumor development rather than initiation. The combined analyses suggest the existence of two types of UCC tumors, one which is characterized by FGFR3 mutation or expression, high expression of protein synthesis genes, and low expression of cell cycle genes. Furthermore, the presented data underscore FGFR3 receptor involvement in urothelial cell transformation as the presence of FGFR3 mutations has a major impact on the global gene expression profile of bladder carcinomas.

Introduction

Urothelial cell carcinoma (UCC) is a heterogeneous neoplasm that either presents as superficial or muscle invasive at diagnosis. Superficial tumors, corresponding to 70% of the cases, are typically papillary and confined to the urothelial mucosa (stage Ta) or to the lamina propria (stage T1) (Haukaas et al., 1999). Apart from displaying a high risk of recurrent disease, most superficial tumors are of low grade (G1 or G2), rarely progress, and are associated with a highly favorable prognosis (Prout et al., 1992; Haukaas et al., 1999). In contrast, high-grade Ta tumors (TaG3), and T1 tumors, represent a significant risk of future tumor progression and death from the disease (Herr, 2000). Even though factors such as tumor stage, grade, size, multifocality, recurrence rate and presence of concomitant carcinoma in situ (Tis) contribute with prognostic information, the course of the individual disease is hard to predict. As morphological features and growth patterns have a limited predictability on outcome, many investigations have aimed at identifying more reliable molecular markers. Although several single markers have been proposed, very few have been implemented in clinical practice (Duggan and Williamson, 2004). A deeper understanding of the biology of urothelial carcinomas and the biological processes operating in this tumor type would further facilitate the identification of relevant and informative marker systems.

At the genomic level, UCC is characterized by complex chromosomal changes with several recurrent aberrations. A highly characteristic change is loss of chromosome 9, cytogenetically seen at high frequencies in Ta/T1, but to a lesser extent in muscle invasive tumors (Höglund et al., 2001). Cytogenetic loss of chromosome 9 is also reflected by the recurrent loss of heterozygosity (LOH) on this chromosome. The region showing most frequent LOH on 9p includes CDKN2A, which is homozygously lost in a substantial proportion of the cases (Baud et al., 1999). At least three genomic segments in 9q have been implicated by LOH analyses but no definite target genes have been assigned to these regions (Knowles, 1999). Because loss of chromosome 9 is frequently seen as the sole anomaly, –9 have been considered to be an early event in UCC progression (Cordon-Cardo, 1998; Richter et al., 1998; Höglund et al., 2001). UCC is also characterized by activating mutations in the fibroblast growth factor receptor 3 gene (FGFR3) (Cappellen et al., 1999). FGFR3 belongs to a tyrosine kinase receptor family that regulates diverse cellular processes including growth, differentiation and angiogenesis. An important clinical finding is that bladder tumors with FGFR3 mutations represent a subset of low grade/stage tumors that rarely progress, and hence associates with a favorable prognosis (Billerey et al., 2001; van Rhijn et al., 2003). In contrast to Ta tumors, with a FGFR3 mutation frequency close to 70%, mutations of FGFR3 are rarely found in Tis or muscle invasive tumors (Billerey et al., 2001). Tis is a likely precursor to invasive tumors and frequently demonstrate TP53 mutations (Spruck et al., 1994). It has therefore been proposed that two alternative genetic pathways or disease entities are present in urothelial tumors; one characterized by FGFR3 mutations and the other by TP53 mutations (Spruck et al., 1994; Bakkar et al., 2003; van Rhijn et al., 2004). However, the biological understanding and molecular details of such suggested pathways are at present far from clear.

In recent years, genome-wide expression profiling by the use of microarray technology has contributed to extensive and new insights into the complex gene expression patterns and dysregulation of genes occurring in urothelial neoplasias (Thykjaer et al., 2001; Dyrskjøt et al., 2003, 2004, 2005; Mor et al., 2003; Sanchez-Carbayo et al., 2003; Modlich et al., 2004; Blaveri et al., 2005; Kim et al., 2005; Wild et al., 2005). Microarray analyses have not only delineated gene expression profiles distinct for different histological subtypes (Thykjaer et al., 2001; Dyrskjøt et al., 2003, 2004; Sanchez-Carbayo et al., 2003; Blaveri et al., 2005; Wild et al., 2005), but also gene expression signatures of importance for tumor recurrence and progression (Dyrskjøt et al., 2003, 2005; Modlich et al., 2004; Wild et al., 2005), and patient survival (Blaveri et al., 2005; Kim et al., 2005). Whereas previous microarray studies of bladder cancer mostly have been dedicated to isolate gene expression profiles and predictors that classify tumors according to stage, recurrence and outcome, we focus on a comprehensive molecular characterization of Ta and T1 tumors. In the present study we analyse a large cohort of early-stage bladder tumors, 74 Ta and T1 tumors, on 27k cDNA microarrays. To obtain a more refined molecular description, we combine gene expression data with both mutational analyses of FGFR3 and TP53, and LOH analyses of chromosome 9 and 17p. We demonstrate that the global gene expression profiles subdivide low-stage urothelial carcinomas into four clusters that correlate with FGFR3 mutational status and LOH on chromosome 9. Apart from providing a biological understanding and critical insights into the molecular mechanisms operating in early-stage urothelial carcinomas, the identified molecular features may be important for both diagnostic and prognostic evaluations as well as to identify molecular targets for tumor treatment.

Results

Tumor subclusters as determined by expression profiles

To identify molecular subtypes of early stage UCC we applied hierarchical cluster analyses (HCA) and multidimensional scaling (MDS) to cDNA array expression data. HCA using Euclidean distances clustered the 75 tumors into three groups; Cluster I, II and III of which Cluster II could be divided in two further subgroups, Clusters IIa and IIb (Figure 1a: a two-way HCA is given in Supplementary Figure S1). Hierarchical cluster analyses with 1-Pearson correlation as a distance measure yielded essentially the same cluster outcome. A polarized pattern of tumor grade was apparent in the HCA where Cluster I contained the highest proportion of G1 (73%, P=0.0029, χ2-test), and Cluster III the highest proportion of G3 tumors (75%, P=0.0000). Cluster II was more heterogeneous regarding tumor grade. However, 63% of the G2 tumors localized to this cluster, a frequency significantly higher than in Cluster I and III, respectively (P=0.0116 and P=0.0400, χ2-test).

Figure 1
figure1

Classification of tumors based on expression profiles. The MDS and HCA analyses are based on the expression of 6749 unique LocusLink entities. Euclidian distances and Wards' algorithm for cluster formation were used in the HCA analysis. The MDS was based on Euclidean distances. (a) HCA identifying four subgroups of early-stage urothelial carcinomas, Cluster I, Cluster IIa, Cluster IIb and Cluster III, respectively. The presence of FGFR3 and TP53 mutations are indicated with +. (b) MDS colored according to Cluster I, Cluster IIa, Cluster IIb and Cluster III cases as revealed by the HCA. (c) The same MDS colored according to FGFR3 and TP53 mutational status.

To further evaluate the relationship between tumors, a MDS analysis was performed (Figure 1b). Cases within Clusters I, II and III, were separated along the first dimension, and since Clusters I and III represent low- and high- grade tumors, respectively, the first MDS dimension represents an axis of increasing pathological grade. Notably, Clusters IIa and IIb are separated in the second MDS dimension. The MDS also revealed a higher level of inter-tumor variation among Cluster III than among Cluster I tumors, suggesting a link between higher grade/stage and an increased variability at the expression level. Thus, based on expression patterns of individual tumors, unsupervised HCA and MDS grouped tumors into three major clusters where Cluster I contained the largest proportion of TaG1 tumors, Cluster II was heterogeneous but dominated by G2 tumors, and Cluster III was characterized by G3 tumors.

We then performed SAM analyses to identify gene expression profiles characteristic for the respective tumor clusters. This was achieved by comparing each tumor cluster with the remaining tumors, setting the d-values equivalent to false discovery rates (FDRs) of zero, and subsequently subjecting the lists of significant genes to Gene Ontology (GO) analyses. All gene lists obtained through SAM analyses are given in Supplementary Table S2. Among the 121 unique genes (LocusLink ID's) found with significantly lower expression in Cluster I relative the other tumors (Clusters II and III), GO-analysis revealed significant enrichment for the GO category cell cycle and a number of its related subcategories (Table 1). In contrast, Cluster III tumors showed a significant enrichment of cell cycle genes among the highly expressed genes (Table 1). Genes that showed high expression in Cluster III and low in Cluster I included, CCNE, E2F1, CDC6, three members of the MCM gene family, TOP2A, RFC3 and -4, CCNB1 and -B2, CENPA, -E, -F, and BUB1. In addition, several cell cycle genes were among the 25 top discriminatory genes for Clusters I and III (Supplementary Table S2). Furthermore, FGFR3 expression was significantly lower in Cluster III and was among the 10 most discriminating genes for this tumor group. The upregulated genes in Cluster IIb showed enrichment for the GO categories immune response, cell communication and extra cellular. Genes associated with the GO categories cell communication and extracellular included CAV1, eight members of the collagen gene family (collagens 1A2, 5A2, 6A1, 6A2, 6A3, A5A1, 16A1 and 18A1), ELN, FBLN2, FYN and the FYN binding protein FYB, ITGA4, LAMA4 and -B2, NID2, TNC, and TNS. Genes associated with immune response included EBF, FCGR2B, HHEX, IL16, IL7R ITGAL, LCK, LEF1, LNK, MYADM, NFKBIA, NFKBIE, NFKBIZ, TCF4, and TRD@.

Table 1 Gene Ontology analysis of differentially expressed genes in the tumor Clusters I, IIa, IIb, and IIIa

Clinical data were reviewed with respect to the HCA Clusters (Table 2). The proportion of tumors with a recurrence free follow-up was significantly higher (P=0.0193, χ2-test) in Cluster I as compared to Cluster II tumors. In line with this, the frequencies of multiple tumors increased with tumor cluster (27, 48, and 68% for Clusters I, IIa and IIb, respectively). Furthermore, none of the tumors in Cluster I and II were diagnosed with concomitant Tis, whereas three out of eight (38%) of the tumors in Cluster III were. Even though the tumors in Cluster III are few (n=8) this difference is highly significant (P=0.0000, χ2-test). The association of Tis with Cluster III tumors persisted when presence of Tis in previous or recurrent tumors was included in the analysis; 12% of the Cluster I and II cases and 75% of the Cluster III showed a history of Tis.

Table 2 Clinical and histopathological characteristics of the four HCA clustersa

Clusters of co-expressed genes

We identified eight clusters of co-expressed genes with the QT clust algorithm that showed significant enrichment of GO categories (gene clusters A–H, Supplementary Table S3; individual genes for each gene cluster are given in Supplementary Table S4). The most prominent gene cluster, gene cluster A, was significant for the GO categories immune response, cell communication and extracellular. We noted a functional relationship between this cluster and gene clusters D (immune response), E, (extracellular) and G (immune response). In addition, clusters A, D and E, showed a substantial correlation (ρ>0.7, Spearman's rank correlation). The second most prominent cluster, gene cluster B, revealed a strong association to the cell cycle. A functional relationship could also be observed between the cell cycle cluster and the small cluster H, associated with chromatin assembly and nucleotide metabolism. The third largest cluster, gene cluster C, were specific for protein biosynthesis and ribosome, and showed negative correlation with the cell-cycle cluster (ρ<−0.7). Finally, one minor gene cluster, significant for mitochondrion, was identified.

To relate the major gene clusters A, B and C to tumor Cluster assignment obtained through the HCA, tumors were rank ordered according to median gene expression for each gene cluster (Figure 2). A continuous increase or decrease of the median expression levels were seen in all rank plots. The median gene expression values in gene cluster A varied eightfold across the tumors and Cluster IIb tumors were associated with the highest expression (Figure 2a). The cell-cycle cluster (gene cluster B) showed a strong correlation with tumor Cluster (ρ=0.75, Spearman's rank correlation) and varied fivefold across the tumors. The protein synthesis/ribosome cluster (gene cluster C), varied 3.6-fold across the tumors with the highest expression in Cluster I and lowest in Cluster III tumors and showed a negative correlation with tumor cluster (ρ=−0.70). Furthermore, as determined by a normal probability plot, the protein synthesis/ribosome showed a non-normal distribution where tumors above rank position 53 exhibit a marked decrease in expression of protein synthesis/ribosome-related genes.

Figure 2
figure2

Expressional variation of the co-expressed gene clusters across the tumors. Rank plots displaying the median log 2 expression ratios (y-axis) for genes in gene clusters A, B and C. Cases are rank ordered (x-axis) according to median expression level and color coded according to the HCA Cluster assignment (Figure 1a). Below each rank plot, an expression heatmap of all genes within the corresponding gene cluster is shown. Rows, individual genes; columns, tumors rank ordered by median gene expression; high relative expression levels, red; low relative expression, green; gray, missing values. Lists of individual gene cluster members are found in Supplementary Table S3. (a) Gene cluster A significant for GO-categories immune response, cell communication and extracellular (b) Gene cluster B significant for the GO category cell cycle. (c) Gene cluster C specific for the GO categories protein biosynthesis and ribosome. Cases in (a) and (b) are ranked according to increasing, and in (c) according to decreasing expression levels.

FGFR3 and TP53 mutation analyses

Activating mutations of FGFR3 were found in 46 (61%) of the 75 tumor samples (Supplementary Table S1), with S249C being the most common (31 out of 67 cases with mutation, 67%). One case demonstrated two FGFR3 mutations, G382R and N542S. The highest mutation rates were seen in G1 and G2 cases (80 and 64%, respectively), whereas only one of the 11 G3 tumors demonstrated a mutation. TP53 mutation analysis identified mutations in nine tumors of which three were G3, four G2, and two G1 (Supplementary Table S1). To investigate FGFR3 expression across the tumors, expression levels were rank ordered and plotted (Figure 3a). This revealed that FGFR3 mutations were associated with high FGFR3 expression and that a subset of cases with low relative expression and no FGFR3 mutation were present. Furthermore, a subsequent t-test revealed a significantly higher expression in mutated than in non-mutated tumors (P=0.0000).

Figure 3
figure3

Expressional variation of FGFR3 and the co-expressed gene clusters across the tumor samples. (a) Expression levels of FGFR3 as determined by the mean value of the two reporters for FGFR3 present on the array. (b) Gene cluster B significant for the GO category cell cycle. (c) Gene cluster C specific for the GO categories protein biosynthesis and ribosome. Tumors are colored according to FGFR3 and TP53 mutational status. Rank plots are analogous to the plots in Figure 2.

To correlate the mutational status of FGFR3 to the global gene expression profiles, cases with FGFR3 mutations were indicated in the HCA dendogram (Figure 1a) and a marked polarized pattern was seen: 14 of 15 Cluster I (93%), 18 of 27 Cluster IIa (67%), 14 of 26 Cluster IIb (64%), and none of the Cluster III cases displayed FGFR3 mutations. The reverse was seen for TP53 mutations: five occurred in Cluster III (63%) and four in Cluster II tumors (8%) of which two cases also showed FGFR3 mutations. The similar expression profiles of FGFR3 mutated tumors suggested by the HCA is also evident in the MDS analysis (Figure 1c). In addition, the MDS revealed that FGFR3wt/TP53wt tumors were located between FGFR3mut/TP53wt and FGFR3wt/TP53mut tumors. Therefore, we investigated the association of mutational status of FGFR3 and TP53 with the activity of the cell-cycle gene cluster (gene cluster B). Figure 3b shows that the genotypes are ranked in order FGFR3mut/TP53wt < FGFR3wt/TP53wt < FGFR3wt/TP53mut with respect to cell-cycle gene activity. The association of genotype with protein synthesis/ribosome expression (gene cluster C) was investigated in a similar way. A high expression of protein synthesis/ribosome genes was seen in FGFR3mut/TP53wt cases (Figure 3c), whereas almost all low expressing cases, that is, with rank values exceeding 53, were either FGFR3wt/TP53wt or FGFR3wt/TP53mut. The proportion of FGFR3 mutated tumors in the former group (42 out of 53) deviated significantly from the latter (4 out of 23, P=0.0000, χ2-test). In addition, the low-ranking (>53) tumors were more advanced as 10 out of 11 G3 tumors included in the analysis were low ranking on the protein synthesis/ribosome variable.

LOH analyses

We used 17 microsatellite markers for chromosome 9 and two for 17p to perform LOH analysis. LOH on 9q were seen in 22 of the 50 investigated cases (44%). LOH of 9p segments were less common (24%) and associated with the concomitant LOH of 9q. LOH of the TP53 region on 17p was seen in three of the cases (Figure 4). No losses of chromosome 9 or 17p segments were seen among the 13 analysed Cluster I tumors. No difference in LOH 9q frequency was seen between FGFR3mut and FGFR3wt cases, 55 and 47%, respectively. To identify an expression profile associated with LOH of 9q we performed a SAM analysis comparing cases with and without LOH of 9q. This analysis identified 118 genes of which 27 were not located on chromosome 9 (Supplementary Table S5). No significant GO categories could be ascribed to these latter genes. However, the gene cluster as a whole was highly significant for the GO category Homo sapiens 9q (P=6.71 × 10−100) indicating that the major transcriptional effect of 9q LOH is a decrease in expression of a large number of 9q genes.

Figure 4
figure4

Results from the LOH analysis. Tumor samples are grouped into Cluster I, Cluster IIa, Cluster IIb and Cluster III according to the results obtained in the HCA (Figure 1a). Microsatellite marker positions given in Mb were obtained from the NCBI Map Viewer (http://www.ncbi.nlm.nih.gov/mapview/, Build 34.3). Open squares, retention of both alleles; gray squares, not informative; black squares, allelic imbalance. +, presence of mutation; −, absence of mutation.

Markers for recurrence

With the aim to identify genes that predict recurrence at follow-up in patients with Ta G1/G2 disease, a SAM analysis was performed where tumors from patients showing recurrence within 8 months were tested against tumors from patients with either no recurrence or a recurrence after more than 24 months. This produced a gene list comprising 49 unique genes with significantly higher expression (FDR=10%) among the high recurrence rate tumors. A subsequent supervised HCA demonstrated that 14 of the 18 tumors with high recurrence rate clustered within two of the four observed subclusters (Figure 5). When the 49 identified genes were subjected to a GO-analysis, significant enrichment of the GO categories cell adhesion, extracellular matrix, and cell communication was observed (P=5.98 × 10−7, P=3.37 × 10−4, P=1.15 × 10−2, respectively).

Figure 5
figure5

Supervised HCA of genes identified though SAM analysis as differentially expressed in TaG1/G2 tumors with short recurrence-free follow-up (<8 months) compared to tumors with either no recurrence or more than 24 months to recurrence. Bar on top represents tumor recurrence status; gray, no recurrence or >24 months to recurrence; black, recurrence within 8 months. In heatmap: rows, individual genes; columns, tumors; green, low relative expression; red, high relative expression.

Discussion

We performed gene expression profiling in conjunction with LOH and mutation analyses to characterize 75 early-stage urothelial carcinomas. The initial analysis of the expression data identified three tumors clusters of which one could be divided in two further subclusters. The obtained clusters were designated Cluster I, IIa, IIb and III, respectively. Pathological data revealed that Cluster I was dominated by TaG1, Cluster II (IIa and IIb) by G2 and Cluster III by G3, and thus that the molecular clusters defined by the expression profiles corresponds to increasing grade. Genes distinguishing Cluster I and III were enriched for cell-cycle genes, which showed low expression in Cluster I and high in Cluster III. Cluster IIb tumors showed significant expression of genes related to immune response and extra cellular components. The subsequent MDS analysis organized the cases along the first dimension according to tumor cluster that is, grade/stage and cell-cycle activity, but did not separate Cluster IIa and IIb tumors. These were, however, separated in the second dimension indicating features other than pathological grading/cell-cycle gene expression as important for their difference. In addition, the MDS analysis revealed Cluster I as a dense cluster of tumors with low expressional variation, whereas Cluster III tumors showed a high-level of inter-tumor heterogeneity. This implies that a transition from Cluster I to Cluster III is linked to an increased expressional diversity. We suggest that this diversity may contribute to the recognized unpredictability of the clinical behavior of TaG3/T1 tumors (Herr, 2000).

To identify genes of importance for the development and heterogeneity of UCCs we performed non-hierarchical clustering of genes to identify groups of co-expressed genes varying across the tumors. Eight gene clusters ascribed with significant GO categories were identified. Two were associated with cell-cycle activity (gene clusters B and H), one with protein synthesis (cluster C), four with immune response and extracellular components (clusters A, D, E, and G) and one significant for mitochondrial genes (cluster F). Correlation analyses revealed that gene clusters B and C showed significant and substantial negative correlation and clusters A, D and E substantial positive correlations. Thus, we identify four major processes/components, cell cycle, protein synthesis, immune response and extra cellular components, that contribute to the expressional heterogeneity observed in early-stage UCC.

The cell-cycle cluster (gene cluster B) contained several genes involved in chromatin replication such as H2A histone family members, MCM family members, the origin recognition complex protein ORC1L, PCNA, replication factor C members (RFC) as well as TOP2A. It also contained cell-cycle regulatory genes such as CCNE, E2F1, CDC2, CCNB1 and -B2, and CDCD25A, and the mitosis regulating genes BUB1, MAD2L1, PLK1 and -4, PRC1 and three CENP gene family members. Thus, genes important for the G1, S, and G2 phases as well for mitosis contributed to this cluster. The gene cluster varied fivefold across the tumors and showed substantial correlation with tumor Cluster, with low expression in Cluster I and high in Cluster III, as well as with tumor grade and stage. Furthermore, tumors in the present investigation with increased levels of Ki67 immunohistochemical staining (Liedberg et al., unpublished data) also showed significantly higher cell-cycle gene expression levels (P=0.0000, t-test). Consequently, we used the activity of the cell-cycle genes as a marker for proliferation. The protein synthesis cluster (gene cluster C) included the genes RPL5 and -L23, RPS8, -S12, -20, -S23 and -S25, as well as the translation initiation factor EIF3S6. This gene cluster showed negative correlation with tumor Cluster as well as with the cell-cycle gene cluster. Other investigators have also noted the low expression of cell-cycle genes and high expression of ribosomal genes among superficial tumors (Thykjaer et al., 2001; Dyrskjøt et al., 2003). This indicates that the progression of early stage UCC is characterized by two major processes: increased cell-cycle activity and decreased protein synthesis. The decrease in protein synthesis may be attributed to dedifferentiation as cell structures and cell–cell interactions involved in preserving the normal cell and tissue organization are present to a lesser extent in undifferentiated cells, which in turn may lead to a decrease in the complexity of proteins that have to be maintained and hence a decrease in protein synthesis.

The immune response/extracellular components gene cluster (gene cluster A) varied eightfold across the tumors showing the highest expression in Cluster IIb tumors. This gene profile was uncorrelated with the cell cycle and protein synthesis gene clusters and thus represents an expression signature attained by tumors independent of tumor progression, hence the separation of Cluster IIa and IIb in the second MDS dimension (Figure 1b). The cluster included several collagen genes, laminins A4 and B2, NID2, ITGA4, CAV1 and FYN. Nidogen (NID2) links laminins and collagens, which ultimately signals though integrins (Belkin and Stepp, 2000) and CAV1 may function as a membrane adaptor that connects the integrin alpha subunit to the tyrosine kinase FYN, thereby coupling integrins to the RAS-ERK pathway (Wary et al., 1998). Hence, both extracellular matrix components as well as signaling systems responding to the ECM co-varied across the tumors and showed high expression in Cluster IIb cases. The fact that gene cluster A showed substantial correlations with gene clusters D and E, significant for the GO categories immune response and extracellular matrix, respectively, indicates that these clusters varies along the second MDS dimension as well. Consequently, Cluster II tumors show a high heterogeneity with regards to immune response and extracellular matrix signatures. Furthermore, genes specific for Cluster IIb were significant for the GO categories defense response, response to wounding, and inflammatory response indicating that Cluster IIb tumors may be subjected to increased inflammatory response possibly owing to tissue damage related to tumor growth. Upregulated genes significant for the GO category categories extra cellular, extracellular matrix, and plasma membrane could in this context be a consequence of extracellular matrix remodeling as a part of a wound healing process. Even though it can not be excluded that these expressional events take place in cells adjacent to the neoplastic cells, they still may provide critical insights into the tumor microenvironment and bladder cancer biology, and be of value for classification of relevant clinical types of tumors (Blaveri et al., 2005).

A characteristic feature of all gene clusters was the gradual increase/decrease in expression across the tumors. This was clearly revealed by the continuous distributions of the rank-ordered median expression levels of the gene clusters. Graphs produced by rank-ordered data would display distinct gaps in the distribution if groups of tumors with markedly different expression patterns had been present. Therefore, Ta tumors do not differ by discrete activation/inactivation of groups of genes. This was also shown by the finding that Clusters I and II overlapped in the MDS indicating a continuous change in expression levels, even though the HCA revealed Cluster I tumors as a distinct subtype. Furthermore, and as previously shown (Wild et al., 2005), we could not find any genes that distinguished G1 from G2 tumors, again underscoring the continuous transition from one state to another. These findings may limit the possibility to define distinct clinical categories of Ta tumors based on expression patterns only. Hence, a robust molecular classification of Ta tumors most probably requires information at more than one molecular level.

Activating point mutations in FGFR3 frequently occur in low grade/stage, bladder carcinomas (Billerey et al., 2001). In the present investigation, mutations in FGFR3 were detected in 61% of the tumors. As expected, mutations were more frequent in G1/G2 than in G3 tumors, underscoring the low propensity of urothelial carcinomas with FGFR3 mutations to progress or to become invasive (Bakkar et al., 2003; van Rhijn et al., 2003). Virtually all FGFR3 mutations seen in bladder cancer also occur in the lethal disease thanatophoric dysplasia (Vajo et al., 2000). Two mutations, G382R and N542S, associated with the non-lethal achondroplasia and hypochondroplasia, respectively, were nevertheless identified, both found in the same tumor. FGFR3 mutations were confined to Clusters I and II, and were particularly characteristic for Cluster I tumors showing mutations in more than 90% of the cases.

FGFR3 expression showed an inverse relationship with grade/stage, also noted by other investigators (Dyrskjøt et al., 2005; Wild et al., 2005). The distribution of the expression levels was, however, not normally distributed suggesting the presence of two categories of UCCs, one showing high and one low FGFR3 expression. All but one case with FGFR3 mutation belonged to the high-expression group suggesting a strong association between FGFR3 mutation and expression. There are two possible explanations for this finding. One is that only cells that show expression of FGFR3 acquire mutations to constituently activate a signaling pathway normally a part of the response repertoire of the cell. This would indicate that UCCs may originate from different urothelial cell types, FGFR3 expressing and non-expressing. Alternatively, mutational activation of FGFR3 induces increased expression of the receptor gene in an autocrine fashion. In this latter scenario, increased expression in the absence of mutation has to be accounted for by other mechanisms. Irrespective of the relationship between mutations and high expression of FGFR3 there exist distinct high and low FGFR3-expressing tumors underscoring the importance of FGFR3 receptor function in a large subset of urothelial carcinomas.

Nine TP53 mutations were detected and these were most frequent among Cluster III cases. Of the four Cluster II tumors with TP53 mutations, two also showed FGFR3 mutations. This suggests that the FGFR3 mutation status is more indicative for the overall expression profile than the TP53 status. Interestingly, the two tumors in Cluster II with FGFR3wt/TP53mut genotype displayed high FGFR3 expression and mutations that, in contrast to the other identified TP53 mutations, only partially reduce TP53 activity (67 and 51% activity, respectively, as determined by WAF1 promoter activating assays; Professor Thierry Soussi, personal communication). This may explain why these cases do not cluster with the other FGFR3wt/TP53mut cases.

The MDS indicated not only a clear segregation of tumors with FGFR3mut and TP53 inactivating mutations, but also that FGFR3wt/TP53wt tumors were located at the border between FGFR3 and TP53 mutated cases. The more detailed analysis initiated by this obervation revealed that FGFR3 mutated tumors showed low expression of cell-cycle genes and high expression of protein synthesis genes, whereas cases with the FGFR3wt/TP53wt genotype show the opposite features, even more pronounced in cases with the FGFR3wt/TP53mut genotype. This finding further strengthens the proposal that two alternative pathways operate in UCC (Bakkar et al., 2003; van Rhijn et al., 2004). The present analysis suggests, in line with Bakkar et al. (2003) and van Rhijn et al. (2004), one pathway characterized by FGFR3 involvement and a second genetically less well-characterized pathway that show TP53 mutations at higher frequencies. The suggested important role of FGFR3 in early-stage UCC and the recent finding of specific FGFR3 inhibitors (Trudel et al., 2005) make it possible to explore FGFR3 as a molecular target for tumor treatment.

As loss of chromosome 9 is one of the most frequent chromosomal events in UCC (Richter et al., 1998; Höglund et al., 2001) we determined the chromosome 9 LOH status in cases for which DNA from tumor and peripheral blood was available. As expected, LOH of 9q was more frequent than of 9p, LOH was, however, not uniformly distributed among the tumor clusters. Particularly, all of the investigated Cluster I tumors showed absence of LOH, whereas LOH was frequent in Cluster II. Using the activities of the cell cycle and the protein synthesis gene clusters as indicators of tumor development, loss of 9q seems to be associated with the more advanced Cluster II than Cluster I tumors and hence with tumor development. Cluster I tumors also showed recurrences to a lesser extent underscoring the view that the acquisition of LOH on chromosome 9 is related to a more advanced tumor type. The loss of chromosome 9 was, however, not linked to any specific expression profile that could explain this association; most of the genes with altered expression in –9 tumors were located on chromosome 9 and not associated with any specific cellular process or component. Hence, the impact of –9 on the overall expression profile is subtle, particularly when only the non-9q genes are considered. Nevertheless, this does not rule out a possible biological importance of 9q loss in tumor progression.

We identified a set of 49 genes that were significantly highly expressed in samples from patients that received a recurring tumor within a short period after the examined tumor. None of the identified genes in the present study, however, coincided with the predictor genes previously described by Dyrskjøt et al. (2003). Several factors might contribute to the fact that no common and distinct expression profile was found. For example, incomplete resection may be a non-biological factor affecting tumor recurrence rate (Brausi et al., 2002). In addition, there might be more than one molecular pathway contributing to the development of recurrent tumors. Nevertheless, a high significance for the GO-category cell adhesion observed among the 49 genes, as well as in the gene predictor described by Dyrskjøt and co-workers, suggests that the identified genes may be compatible with processes involved in tumor recurrences (Hafner et al., 2002).

In conclusion, we used expression profiling, mutation analysis and LOH analysis to molecularly characterize a large cohort of early-stage UCC. The presented data thus suggests the existence of two types of UCC tumors. The first consisting of low-grade tumors characterized by FGFR3 activity, either by FGFR3 mutation or by expression, high protein synthesis and low cell-cycle activity. In contrary, the second group show less, or no, dependence of the FGFR3 receptor, low levels of protein synthesis and high cell-cycle gene activity. The presented data thus suggests that low grade/stage UCC are characterized by FGFR3 receptor involvement at a more critical level than previously may have been appreciated.

Materials and methods

Patients and tissues

Urothelial tumors were acquired by cold cup biopsies from patients undergoing transurethral resection for bladder cancer at the University Hospital of Lund, Sweden, between 2001 and 2003. One or several cold cup biopsies were taken from the exophytic portion of the tumor, and the tumor sample was immediately put into a transportation medium, rapidly transferred to the laboratory, and frozen. Histopathological staging and grading were reviewed according to the 2002 TNM (Sobin and Wittekind, 2002) and 1999 WHO (Mostofi et al., 1999) classification systems by one single pathologist (GC) and Ta–T1 tumors of grade G1–G3 were included. Clinical records were reviewed in April 2005 and at this time 70 of the 75 included patients were alive with a median follow-time of 35.6 months. Of the five registered deaths, one was because of bladder cancer, whereas the other four patients died of other causes. The study was approved by the local ethical committee and informed consent was obtained from all patients. Individual tumor characteristics are listed in Supplementary Table S1.

RNA extraction and amplification

Prior to RNA extraction, samples were snap-frozen in liquid nitrogen and ground with a micro-dismembrator II (B. Braun Biotech Inc., Allentown, PA, USA). Total RNA was isolated using Trizol reagent (Invitrogen, Carlsbad, CA, USA) followed by purification on RNeasy columns (Qiagen, Valencia, CA, USA) according to the manufacturer's instructions. The integrities of all RNA samples were assessed using an Agilent 2100 Bioanalyzer (Agilent technologies, Palo Alto, CA, USA) and samples showing indication of RNA degradation were excluded. T7-based RNA amplification of all tumor samples was performed using the MessageAmp aRNA Kit (Ambion, Austin, TX, USA). In this reaction, 1 μg of total RNA was used as template, yielding 30–50 μg of aRNA. As reference for the microarray hybridizations, the Universal Human Reference RNA (Stratagene, La Jolla, CA, USA) was amplified as above.

Microarray hybridization and data processing

Microarray slides were obtained from the Swegene DNA microarray resource center at Lund University, Sweden (http://swegene.onk.lu.se). The 25 648 cDNA clones spotted on the array represent 13 737 UniGene clusters and 11 592 unique LocusLink IDs according to Unigene Build 180. For hybridization, 1.5 μg of tumor and reference aRNA were differentially labeled with Cy3 and Cy5, respectively, using the CyScribe Post-Labeling Kit (Amersham Biosciences, Uppsala, Sweden). Probes were purified with Cyscribe GFX Purification Kit (Amersham), lyophilized and dissolved in 42 μl of Universal Hybridization Solution (Corning, Acton, MA, USA). Hybridizations were carried out in a CMT-Hybridization Chamber (Corning) for 18 h at 42°C. Pre-hybridization of slides and post-hybridization washes were performed as described in the Universal Hybridization Kit (Corning) and slides were scanned with an Agilent G2565AA microarray scanner (Agilent Technologies). Raw images were analysed using the GenePix Pro 4.0 software (Axon Instruments, Union City, CA) and the quantified data matrices were subsequently loaded into BioArray Software Environment (BASE) (Saal et al., 2002). Within BASE, background corrected Cy3 and Cy5 intensities were used to obtain Cy3/Cy5 ratios. Spots of poor quality, identified through the GenePix analysis, were removed followed by normalization of each individual array using an implementation of the lowess fit (Yang et al., 2002). Data were further filtered to remove spots showing signal-to-noise ratios (SNR) below 2 and only spots varying more than twofold in at least five arrays with respect to the median expression level were selected for further analysis. In addition, spots with more than 10% missing values across experiments were removed. Expression ratios for duplicated spots were averaged and the data matrices, containing log2 Cy3/Cy5 ratios for each gene, were median centered across spots. The final data set, consisting of 8772 spots corresponding to 6479 unique LocusLink entities, is available as supplemental data.

Statistical analyses

For analysis and visualization of differences between tumor samples caused by global changes in gene expression, unsupervised HCA, principal component analyses and MDS were performed using the STATISTICA 6 software (StatSoft Inc., Tulsa, OK, USA). HCA was performed using 1-Pearson correlation or Euclidic distances and Wards' algorithm or complete linkage for cluster formation. For MDS, Euclidean distances were used. The SAM (Tusher et al., 2001) analyses were carried out with 1000 random permutations and the cut-off level for delta-values chosen so that the median number of false significant genes was zero, that is, the median FDR was 0%. Co-varying genes were identified through the QT clust algorithm (Heyer et al., 1999) using 1-Pearson correlation, a cluster diameter of 0.35, and the minimum cluster size restricted to 20 clones. SAM analyses and QT-clustering were performed within the TIGR Multi Experiment Viewer (Saeed et al., 2003).

To analyze the behavior of gene clusters across the samples, the median log 2 expression ratio based on all genes in a given cluster was calculated for each tumor case. Hence, each tumor was assigned with a value corresponding to the median expression of all genes within the given cluster. The values were then either rank ordered and plotted, or used to investigate the correlation between different gene clusters or between gene clusters and tumor cluster assignment as defined by HCA. As a tool for finding significantly enriched gene ontology terms, the EASE module within the TMEV software was used. A Bonferroni step-down-corrected P-value <0.05 using EASE-score statistics were considered significant. Apart from the commonly used GO organizing principles, EASE may also be used to identify significant enrichment of genes within a list with regard to, for example, chromosomal localization.

FGFR3 and TP53 mutation analyses

Amplified RNA was primed by random nonamer primers and reversely transcribed into cDNA using Superscript II reverse transcriptase (Invitrogen) according to the manufacturer's instructions. For FGFR3 mutation analysis, three separate PCRs was performed covering all activating mutations in exons 7, 10, 13, and 15, previously described in thanatophoric dysplasia, achondroplasia, hypochondroplasia, crouzon syndrome with acanthosis nigricans, and urothelial carcinoma (Vajo et al., 2000; van Rhijn et al., 2002). In addition, exons 4–9 of TP53 were amplified in four separate PCRs. All primer sequences are available upon request. Direct sequencing of all fragments was carried out with an ABI PRISM 3100-Avant Genetic Analyzer automated sequencer (Applied Biosystems, Foster City, CA, USA) using Big Dye terminator kit (Applied Biosystems), and analysed with the SeqScape v2.5 software (Applied Biosystems). In cases where genomic DNA was available, identified mutations in FGFR3 or TP53 were sequence verified at the genomic level.

LOH analysis

LOH analysis was performed on 50 tumors from which matching blood samples were available. A total of 17 highly polymorphic microsatellite markers distributed over both arms of chromosome 9 and two markers close to TP53 at chromosome 17 were used. Primer information was obtained from The Genome Database (http://www.gdb.org). HEX or 6-FAM labeled forward primers were used in standard PCR amplifications, followed by size separation on an ABI PRISM 3100-Avant (Applied Biosystems) and subsequently analysed using the GeneScan Analysis Software (Applied Biosystems). Allelic imbalance was defined as a more than 50% reduction in allele signal intensity and LOH was defined as the presence of at least two consecutive markers showing allelic imbalance.

References

  1. Bakkar AA, Wallerand H, Radvanyi F, Lahaye JB, Pissard S, Lecerf L et al. (2003). Cancer Res 63: 8108–8112.

  2. Baud E, Catilina P, Bignon YJ . (1999). Int J Oncol 14: 441–445.

  3. Belkin AM, Stepp MA . (2000). Microsc Res Tech 51: 280–301.

  4. Billerey C, Chopin D, Aubriot-Lorton MH, Ricol D, Gil Diez de Medina S, Van Rhijn B et al. (2001). Am J Pathol 158: 1955–1959.

  5. Blaveri E, Simko JP, Korkola JE, Brewer JL, Baehner F, Mehta K et al. (2005). Clin Cancer Res 11: 4044–4055.

  6. Brausi M, Collette L, Kurth K, van der Meijden AP, Oosterlinck W, Witjes JA et al. (2002). Eur Urol 41: 523–531.

  7. Cappellen D, De Oliveira C, Ricol D, de Medina S, Bourdin J, Sastre-Garau X et al. (1999). Nat Genet 23: 18–20.

  8. Cordon-Cardo C . (1998). Cancer Surv 32: 115–131.

  9. Duggan B, Williamson K . (2004). Curr Opin Urol 14: 277–286.

  10. Dyrskjøt L, Kruhøffer M, Thykjaer T, Marcussen N, Jensen JL, Møller K et al. (2004). Cancer Res 64: 4040–4048.

  11. Dyrskjøt L, Thykjaer T, Kruhøffer M, Jensen JL, Marcussen N, Hamilton-Dutoit S et al. (2003). Nat Genet 33: 90–96.

  12. Dyrskjøt L, Zieger K, Kruhøffer M, Thykjaer T, Jensen JL, Primdahl H et al. (2005). Clin Cancer Res 11: 4029–4036.

  13. Hafner C, Knuechel R, Stoehr R, Hartmann A . (2002). Int J Cancer 101: 1–6.

  14. Haukaas S, Daehlin L, Maartmann-Moe H, Ulvik NM . (1999). BJU Int 83: 957–963.

  15. Herr HW . (2000). J Urol 163: 60–61; discussion 61–62.

  16. Heyer LJ, Kruglyak S, Yooseph S . (1999). Genome Res 9: 1106–1115.

  17. Höglund M, Säll T, Heim S, Mitelman F, Mandahl N, Fadl-Elmula I . (2001). Cancer Res 61: 8241–8246.

  18. Kim JH, Tuziak T, Hu L, Wang Z, Bondaruk J, Kim M et al. (2005). Lab Invest 85: 532–549.

  19. Knowles MA . (1999). Electrophoresis 20: 269–279.

  20. Modlich O, Prisack HB, Pitschke G, Ramp U, Ackermann R, Bojar H et al. (2004). Clin Cancer Res 10: 3410–3421.

  21. Mor O, Nativ O, Stein A, Novak L, Lehavi D, Shiboleth Y et al. (2003). Oncogene 22: 7702–7710.

  22. Mostofi FK, Davies CJ, Sesterhenn I . (1999). Histological Typing Of Urinary Bladder Tumors. International Classification Of Tumors, 2nd edn. Springer: Berlin.

    Book  Google Scholar 

  23. Prout GRJ, Barton BA, Griffin PP, Friedell GH . (1992). J Urol 148: 1413–1419.

  24. Richter J, Beffa L, Wagner U, Schraml P, Gasser TC, Moch H et al. (1998). Am J Pathol 153: 1615–1621.

  25. Saal LH, Troein C, Vallon-Christersson J, Gruvberger S, Borg A, Peterson C . (2002). Genome Biol 3: SOFTWARE0003.

  26. Saeed AI, Sharov V, White J, Li J, Liang W, Bhagabati N et al. (2003). Biotechniques 34: 374–378.

  27. Sanchez-Carbayo M, Socci ND, Lozano JJ, Li W, Charytonowicz E, Belbin TJ et al. (2003). Am J Pathol 163: 505–516.

  28. Sobin LH, Wittekind CL . (2002). TNM Classification Of Malignant Tumors, 6th edn. John Wiley & Sons: New York.

    Google Scholar 

  29. Spruck III CH, Ohneseit PF, Gonzalez-Zulueta M, Esrig D, Miyao N, Tsai YC et al. (1994). Cancer Res 54: 784–788.

  30. Thykjaer T, Workman C, Kruhøffer M, Demtroder K, Wolf H, Andersen LD et al. (2001). Cancer Res 61: 2492–2499.

  31. Trudel S, Li ZH, Wei E, Wiesmann M, Chang H, Chen C et al. (2005). Blood 105: 2941–2948.

  32. Tusher VG, Tibshirani R, Chu G . (2001). Proc Natl Acad Sci USA 98: 5116–5121.

  33. Vajo Z, Francomano CA, Wilkin DJ . (2000). Endocr Rev 21: 23–39.

  34. van Rhijn BW, van der Kwast TH, Vis AN, Kirkels WJ, Boeve ER, Jobsis AC et al. (2004). Cancer Res 64: 1911–1914.

  35. van Rhijn BW, van Tilborg AA, Lurkin I, Bonaventure J, de Vries A, Thiery JP et al. (2002). Eur J Hum Genet 10: 819–824.

  36. van Rhijn BW, Vis AN, van der Kwast TH, Kirkels WJ, Radvanyi F, Ooms EC et al. (2003). J Clin Oncol 21: 1912–1921.

  37. Wary KK, Mariotti A, Zurzolo C, Giancotti FG . (1998). Cell 94: 625–634.

  38. Wild PJ, Herr A, Wissmann C, Stoehr R, Rosenthal A, Zaak D et al. (2005). Clin Cancer Res 11: 4415–4429.

  39. Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J et al. (2002). Nucleic Acids Res 30: e15.

Download references

Acknowledgements

This study was supported by the Swedish Cancer Society, The Gunnar, Arvid and Elisabeth Nilsson Foundation, The Crafoord Foundation, the John and Augusta Persson Foundation, the IngaBritt and Arne Lundberg Foundation, the Maud and Birger Gustavsson Foundation, and the Petrus and Augusta Hedlund Foundation. The microarray facility was supported by the Knut and Alice Wallenberg Foundation via the Swegene Program.

Author information

Affiliations

Authors

Corresponding author

Correspondence to D Lindgren.

Additional information

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Lindgren, D., Liedberg, F., Andersson, A. et al. Molecular characterization of early-stage bladder carcinomas by expression profiles, FGFR3 mutation status, and loss of 9q. Oncogene 25, 2685–2696 (2006). https://doi.org/10.1038/sj.onc.1209249

Download citation

Keywords

  • expression profiling
  • FGFR3
  • chromosome 9
  • urothelial carcinoma
  • bladder cancer

Further reading

Search

Quick links