| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Imaging, Diagnosis, Prognosis |
Authors' Affiliations: 1 Molecular Diagnostic Laboratory, Department of Clinical Biochemistry; 2 Department of Urology, Aarhus University Hospital; 3 Department of Theoretical Statistics, Faculty of Mathematical Sciences; and 4 Bioinformatics Research Center, Aarhus University, Aarhus, Denmark
Requests for reprints: Torben F. Ørntoft, Molecular Diagnostic Laboratory, Department of Clinical Biochemistry, Aarhus University Hospital, Skejby, DK-8200 Aarhus N, Denmark. Phone: 45-8949-5100; Fax: 45-8949-6018; Email: orntoft{at}ki.au.dk.
 |
Abstract |
|---|
 Top Abstract Materials and MethodsResultsDiscussionReferences |
|---|
Experimental Design: We selected 85 patients with superficial bladder tumors, stratified into early (stage Ta/grade 1-2, n = 35) and more advanced (either stage T1 or grade 3, n = 50) developmental stages. The patients were followed prospectively, and metachronous tumors were included. We did screening for FGFR3 and TP53 mutations by direct bidirectional sequencing and for genome-wide molecular changes with microarray technology.
Results: A total of 43 of 85 cases (51%) showed activating mutations of FGFR3. The mutations were associated with papillary tumors of early developmental stage. However, after stratifying for developmental stage, FGFR3-mutated tumors showed the same malignant potential as wild-type tumors. Tumors with concomitant CIS were generally FGFR3 wild type. They were characterized by different patterns of chromosomal changes and gene expression signatures compared with FGFR3-mutated tumors, indicating different molecular pathways.
Conclusions: FGFR3 mutations seem to have a central role in the early development of papillary bladder tumors. These tumors follow a common molecular pathway, which is different from tumors with concomitant CIS. FGFR3 mutations do not seem to play a role in bladder cancer progression.
Superficial bladder tumor development follows at least two different molecular pathways; one pathway comprises papillary tumors, and one includes the occurrence of CIS (6, 7). Recently, activating mutations of the fibroblast growth factor receptor 3 (FGFR3) gene have been shown in bladder tumors with high frequency (8). They are associated with early papillary lesions with low malignant potential (9, 10). In fact, the mutations have been found in 75% of nondysplastic genuine urothelial papillomas (11), indicating that they are very early events in the papillary tumor development. FGFR3 is a receptor tyrosine kinase activating the extracellular signal-regulated kinase/mitogen-activated protein kinase and the protein kinase B/Akt pathways that both promote cell survival (12). Constitutively activating mutations might enhance proliferation and provide growth advantage (13). FGFR3-mutated tumors have some malignant potential, because around 25% of the advanced bladder cancers show a FGFR3 mutation. On the other hand, it seems that the frequency of FGFR3 mutations decreases with increasing tumor stage and grade of dysplasia (9, 10). The most probable explanation for this finding is that the majority of FGFR3-mutated tumors never progress. Most of the advanced cancers would then derive from nonmutated precursors following a different pathway of tumor development. Subsequent publications substantiate this hypothesis. Two independent studies found changes of both FGFR3 and TP53 almost mutually exclusive, as only 5% and 5.7%, respectively, showed this pattern (13, 14). FGFR3 mutations are rarely found in CIS (9), which is known to have a high frequency of TP53 mutations (7) and a high probability of progression. Consequently, FGFR3-mutated tumors are regarded as tumors with low malignant potential and a favorable prognosis (10, 15). However, the development of FGFR3-mutated and wild-type tumors has never been shown in longitudinal studies.
We established in 1994 a consecutive tissue bank combined with a clinical database with comprehensive prospective follow-up data for studying molecular aspects of bladder cancer development. It now contains material from almost 2,000 patients. The aim of the present study was to examine the development of FGFR3-mutated and wild-type bladder tumors, with emphasis on the development of the muscle-invasive phenotype. Furthermore, we looked for evidence for different molecular pathways and investigated the correlation of the development of the different tumors with TP53 mutations. For this, we selected patients with superficial bladder tumors and stratified them by developmental stage into early (stage Ta/grade 1-2 tumors, no previous T1 or grade 3 disease) and more advanced (either stage T1 or grade 3 disease, previous or present). The disease courses were followed prospectively. We determined genome-wide chromosomal instability patterns using 10K single nucleotide polymorphism (SNP) microarrays and gene expression changes using expression microarrays and screened the tumors for FGFR3 and TP53 mutations in parallel. Metachronous tumors were examined in a subgroup of cases to show changes in developing tumors.
|
Materials and Methods |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
The bladder tumor specimens were cleaved immediately after transurethral resection. Appropriate amounts of tissue were sent to conventional histopathologic diagnosis, the rest snap-frozen in liquid nitrogen and stored at –80°C. Tissue for RNA extraction was stored in a guadinium thiocyanate solution as described (16). Tumors were staged according to tumor-node-metastasis and graded according to the Bergkvist classification.
A total number of 85 patients with initial superficial bladder tumors (histopathologic stage pTa and pT1) were selected based on the disease course and availability of material. Our aim was to study superficial lesions (stage Ta/T1) to obtain information on the biological difference between those that progress and those that do not. We therefore included a relatively high proportion of potentially malignant (either stage T1 or grade 3, previous or present) tumors. In total, we examined 115 tumors: for 22 patients, we examined at least one subsequent tumor; for five patients, we examined two subsequent tumors; and for one patient, we examined three subsequent tumors.
Patients were followed prospectively with control cystoscopies every 4 months in the first year, every 6 months in the second, and hereafter with yearly controls until a recurrence-free period of 5 years. Disease recurrences were treated in a standardized fashion. Eleven patients received intravesical instillations of Bacillus Calmette-Guerin during the course.
The clinical disease courses were allocated to the following groups:
All patients, tumors, the previous history, stage, grade, concomitant intraurothelial neoplasia [CIS/HGIUN, low-grade intraurothelial dysplasia (LGIUN)], the mutation status, and the microarray analyses done are detailed in Table 1. For further clinical characteristics of the individual disease courses, see Supplementary Information.
|
DNA for FGFR3 and TP53 mutation screening was extracted from the corresponding formalin-fixed, paraffin-embedded tissue sections, using a standard DNA extraction kit (Gentra, Minneapolis, MN). Tumor tissue was trimmed in the microscope to ensure at least 75% tumor cells and digested with proteinase K before DNA extraction. PCR of parts of exons 7, 10, and 15 of FGFR3 and exons 5 to 8 of TP53 was done using primer sequences and PCR procedures as reported by Bakkar et al. (13) and Koed et al. (17), respectively. Mutation analysis was done by direct bidirectional sequencing using an ABI 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA).
Single nucleotide polymorphism array data transformation
Loss of heterozygosity analysis. The arrays were normalized and scored using the Microarray Suite 5.0 and Genotyping Tools software supplied by the array manufacturer (Affymetrix) to obtain allelic calls for tumor DNA and corresponding normal DNA extracted from blood. Scoring of allelic imbalance was conducted as in Koed et al. (17).
Copy number analysis. Normalization: The arrays were normalized, and signal values for the individual SNPs were extracted according to a procedure described previously (18). Briefly, an invariant set normalization method was applied to normalize all arrays at the probe intensity level to a baseline array. After normalization, a model-based method (the perfect match/mismatch difference model) was used to extract a single signal value (the "observed signal") for each SNP in each array. Subsequently, data was normalized SNP-wise to have mean = 0 and SD = 1, using the empirical mean and SD in 113 normal DNA samples extracted from blood: zij = (yij – mj)/SDJ, where yij is the "observed signal" for array i and SNPj, mi is the mean of SNPj, and SDJ the SD of SNPj. To reduce noise, data zij was further smoothened over K SNPs to obtain xij = 
zik, where the sum is over k = j – K,..., j + K. Here, K = 5.
Test for difference between FGFR3 wild-type and FGFR3-mutated tumors (relative copy number changes): The average of xij was calculated for FGFR3 wild-type (group 1) and FGFR3-mutated (group 2) tumors for each SNP, and the difference dj between the two averages was found. High and low values of dj were indicative of group differences. Low values indicated that group 1 had higher mean than group 2 (corresponding to gains in FGFR3 wild-type tumors or losses in FGFR3-mutated tumors), and high values indicated that group 2 had higher mean (gains in FGFR3-mutated tumors or losses in FGFR3 wild-type tumors). Long segments of SNPs with high (or low) values of dj were more likely to be due to true differences; thus, the number of segments with length exceeding a fixed threshold (here, 10) and the maximum segment length were also calculated. Significance of the observed statistics was calculated using a permutation test. New groups of the same sizes were formed by permuting group labels 50,000 times, and the number of times (in %) a higher value (or lower) was observed was calculated.
Loss and gains: The pattern of "losses" and "gains" in the two groups were derived for each SNP in each array using the xij values. The empirical distribution of all SNPs in 113 normal samples was calculated. If xij is in the upper 1% tail of the distribution, "gain" is reported; if xij is in the lower 1% tail of the distribution, "loss" is reported. Whether the number of losses (gains) in the two groups for a particular SNP deviates from random was assessed, assuming losses (gains) were drawn without replacement from the pooled samples of the two groups (i.e., a test statistic based on the hypergeometric distribution was applied).
Gene expression array data
Preparation of cRNA from total RNA and subsequent hybridization and scanning of the Affymetrix GeneChip arrays was done as described previously (19). The biotinylated targets were hybridized to Eos Hu03–customized Affymetrix GeneChip oligonucleotide arrays and to U133A Affymetrix GeneChip oligonucleotide arrays. For further details on the Eos Hu03 array design, expression measure, and normalization procedure, see Dyrskjot et al. (20). The U133A array expression measures were generated and normalized using the GC-RMA procedure implemented in the ArrayAssist software (Stratagene, La Jolla, CA).
Filtered gene sets were generated for each array type. Before analyzing the expression data from the Eos Hu03 GeneChips, all control probes were removed. Furthermore, we added 300 to all values to avoid negative values. Values were log 2 transformed, and only probes with expression levels above 8.64 (log2 400) in at least three experiments and with max-min of 
0.8 were selected. This filtering generated a gene set consisting of 9,045 probes. For the U133A expression data, we also removed the Affymetrix control probes before analysis. Values were log 2 transformed, and only probes with expression levels of >7 in at least three experiments and with max-min of 2 were selected. This filtering generated a gene set consisting of 8,858 probes.
The average linkage hierarchical cluster analysis of the tumor samples was carried out using a modified Pearson correlation as similarity metric. Genes and arrays were median centered and normalized to the magnitude of 1 before clustering using the Cluster software (21). Data were visualized using the Treeview software (21).
We linked the filtered probe sets from both microarray platforms by gene IDs (LocusLink), resulting in 6,257 linked probes. Then, we compared the sample groups in both data sets by computing common t values for all linked probes. For further details, see Supplementary Table 6.
Carcinoma in situ classifier
To investigate relations of FGFR3 mutations to the "CIS pathway," we used a 76-gene molecular "CIS classifier" based on the 100 significant genes, which we published previously (16). We applied an in-house fabricated microarray platform as described in Dyrskjot et al. (20).
|
Results |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
The analysis of metachronous tumors in 22 patients showed identical mutation pattern in 21 patients. In a single patient (#154), no mutation was found in the subsequent tumor, in contrast to two previous tumors harboring an activating mutation. We did not observe any patient who acquired a FGFR3 mutation during the disease course (Supplementary Table 3).
At analysis time, 35 tumors were early tumors with expected benign disease course (stage Ta/grade 1-2 tumors, no previous T1 or grade 3 disease). Twenty-six of these tumors showed an activating FGFR3 mutation (74%). Among tumors with either previous or present stage T1 or grade 3 disease, FGFR3 mutations were less frequent, as only 17 of 50 (34%) presented this feature. This difference was highly significant (P = 0.0003, 
2).
Association between FGFR3 mutations and the development of bladder cancer disease
Recurrence frequency. Five of nine FGFR3 wild-type tumors with expected benign disease course developed frequent recurrences and so did a similar fraction (15 of 26) of FGFR3-mutated tumors (Fig. 1). Of the tumors from patients with T1 or grade 3 disease, 12 of 24 (50%) FGFR3 wild-type tumors recurred frequently, and in mutated tumors, 13 of 15 (87%). This difference was significant (P = 0.02, Mann-Whitney; Fig. 1A).
|
In the 37 disease courses where intraurothelial neoplasias were observed, 12 of 22 patients (55%; one LGIUN) progressed when the tumors were FGFR3 wild type, and only 2 of 15 (13%, both LGIUN) progressed when the tumors were mutated (P = 0.01).
Clinical course/progression. After initial analysis, most of the early tumors followed the benign courses 1 and 2 (31 of 35, 89%); two developed a high risk course (recurrent T1 or grade 3 disease), and two progressed (stage T2-T4). Importantly, there was no difference in outcome between FGFR3-mutated and wild-type tumors (P = 0.46, Mann-Whitney; Fig. 1C). Among tumors with previous or actual T1 or grade 3 disease, 13 followed the benign course 2 (no stage T1 or grade 3 recurrences), 13 followed the high-risk course 3 (recurrent stage T1 or grade 3 disease), and 24 followed the progressive course 4 (stage T2-T4). Again, we observed no difference in outcome between FGFR3-mutated and wild-type tumors (P = 0.35, Mann-Whitney; Fig. 1C).
Of the 16 muscle-invasive cancers analyzed (four at initial analysis and 12 during the disease course; see Table 1), four (25%) had a FGFR3 mutation. All these tumors already had the same mutations at the superficial stage; the four tumors that were progressed at initial analysis were wild type.
Correlation with TP53 mutation
We screened tumors from 69 patients for both FGFR3 and TP53 mutations. Bidirectional sequencing of exons 5 to 8 of TP53 was done, as TP53 mutations are most frequently reported in these exons (23). We observed TP53 mutations in tumors from 12 patients (17%), FGFR3 mutations in tumors from 35 patients (51%), and both TP53 and FGFR3 mutations in tumors from four patients (5.8%); no mutations were found in tumors from 27 patients (39%; Table 1). The mutations are listed in detail in Supplementary Table 1.
As expected, TP53 mutations were associated with a progressive disease course, because nine of the 12 patients (75%) progressed. Tumors with expected benign disease course (stage Ta/grade 1-2) at analysis had no TP53 mutations. However, we observed an acquired TP53 mutation in one of the later tumors of this group, with subsequent progression (#1058). In the group of stage T1 or grade 3 tumors, TP53 mutations were found in 3 of 17 (18%) FGFR3-mutated tumors and 8 of 32 (25%) FGFR3 wild-type tumors. This difference was not significant and did not support previous publications (13, 14).
Correlation with chromosomal instability pattern
Tumors from a total of 19 patients were analyzed using 10K SNP arrays and screened for FGFR3 and TP53 mutations. Four tumors with expected benign disease course at the time of the first analysis all developed at least T1 tumor during the course; thus, no tumors with the low-risk course 1 were included. For detailed information, see Table 1 and Koed et al. (17).
The results of the loss of heterozygosity (LOH) study were published previously (17); however, the addition of FGFR3 mutation analysis and computation of DNA copy number changes are novel. The LOH study revealed few regions of LOH in most nonprogressing tumors, predominantly on chromosome 9q (66% LOH). LOH of >33% of the informative SNPs throughout the genome, which indicated general chromosomal instability, was seen in progressing tumors. This pattern was characterized by LOH at distinct areas of chromosome 17p13.1 (the TP53 locus) and at two loci at chromosomes 8p12 and 8p23.1. The tumors from three patients (30%) retained chromosomal stability throughout the cancer development [i.e., also in the progressed tumors (stage T2-T4)]. This observation illustrated the complexity of the bladder cancer development.
We computed DNA copy number changes in tumors from 17 patients. The data transformation procedure (see Materials and Methods) enabled an impression of gained and reduced DNA copy number regions in addition to the LOH analysis. See Supplementary Fig. 1 for a survey of all observed copy number changes at late superficial stages.
We found no correlation between FGFR3 mutation status and general chromosomal instability (defined as LOH of >33% of the informative SNPs). We did, however, observe a distinct pattern of chromosomal changes that correlated with the FGFR3 mutation status. This was shown for the three most significantly affected chromosomes in bladder cancer: chromosomes 8, 9, and 17 (Fig. 2). Chromosome 8 is known to exhibit complex chromosomal instability in invasive bladder cancer, typically with losses of 8pter-8p12 and gains of the rest of the chromosome. Changes at chromosome 17 typically include loss of 17p13 (the TP53 locus), also preferentially found in invasive tumors. Figure 2 shows that these patterns also were found in superficial, FGFR3 wild-type tumors, whereas mutant tumors had only few and random changes at chromosomes 8 and 17 (P < 0.01). In contrast, losses on chromosome 9, the most frequent chromosomal change observed in bladder cancer, were more frequent in FGFR3 mutant than wild-type tumors. This was significant (P < 0.01) at 9p21 (the MTS1/INK4 locus) and at broad regions of the 9q chromosomal arm.
|
In general, significant copy number gains were only seen in FGFR3 wild-type tumors. A survey of the significantly (P < 0.01) gained regions is provided in Supplementary Table 4. Supplementary Data also includes figures of differences in copy numbers, gains, losses, and LOH between FGFR3 wild-type and FGFR3-mutated tumors for all chromosomes (Supplementary Figs. 3 and 4). Furthermore, the number of 10 SNP segments that change in each chromosome and the length of these changes are listed together with the associated probabilities (Supplementary Table 5). The reported results were not tested for relation to tumor stage, because only late superficial tumors (mostly stage T1) were included in the analysis.
Correlation with gene expression pattern
We did genome-wide expression profiling using tumors from 64 patients and two different microarray platforms: Customized Affymetrix GeneChips (34 tumors) and Affymetrix U133A GeneChips (33 tumors). Three tumors were analyzed on both arrays. FGFR3 mutations were found in 31 tumors, whereas 33 were wild type. Twenty tumors progressed to muscle-invasive stages; five (25%) had FGFR3 mutations. The FGFR3-mutated tumors were about equally distributed on the two microarray platforms (17 versus 14; Fig. 3).
|
We identified differentially expressed genes on both platforms in a FGFR3 mutation–dependent manner. We found 135 of 6,257 probe sets being differentially expressed with a significance threshold of 0.005 (t test; expected number: 31). One of the most significant differentially expressed genes was the FGFR3 gene itself, whose expression was increased in mutated tumors but also in benign, wild-type tumors (Fig. 4). Most of the differentially expressed genes were down-regulated in FGFR3-mutated tumors compared with FGFR3 wild-type tumors. For example, Fig. 4 shows the differential expression of the CXCR4 gene. The differentially expressed genes were involved in cell-matrix interaction, invasion, angiogenesis, apoptosis, and metastasis and thus seemed related to the oncogenic process associated with tumor development. However, our results did not provide evidence that FGFR3 mutations are driving this process. A complete list of significantly altered genes is found in Supplementary Table 6.
|
2) correlation between the tumors with FGFR3 mutations and the tumors classified as "no CIS." The performance of the classifier and the correlation with FGFR3 mutations is illustrated in Table 2. Notably, none of the benign FGFR3 wild-type tumors showed the molecular "CIS" signature.
|
|
Discussion |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
FGFR3-mutated tumors have malignant potential, although FGFR3 mutations apparently are not directly involved in cancer progression. It was speculated that the mutation may alter cellular proliferation and provide a growth advantage, but that cell cycle and/or apoptosis mechanisms and genomic stability may still be maintained (13). Recent results support this by suggesting that FGFR3 mutations may be involved in bypassing the G1 checkpoint of the cell cycle, leading to replicative stress and activation of the DNA damage response mechanisms (24). This would result in precursor lesions of benign appearance. However, as a key event of malignant development due to sustained replicative stress, the DNA damage response mechanisms may become impaired, leading to genomic instability and cancer progression (24). To investigate the relative malignant potential of FGFR3 mutations, we stratified the tumors in this study into early stage (Ta) well-differentiated (grade 1-2) tumors (with an expected benign disease course) and tumors presenting features of malignant development (stage T1 or grade 3). After stratification, neither the recurrence rate, the frequency of progression, nor TP53 mutation frequency differed between FGFR3-mutated and FGFR3 wild-type tumors, in contrast to previous reports (10, 13–15). This suggested that precursor lesions harboring a FGFR3 mutation have about the same malignant potential as wild-type lesions at comparable developmental stages. However, the FGFR3 mutations are observed very early in development of papillary tumors and could be a driving force in the formation of these tumors. As papillary tumors are diagnosed at early stages due to bleeding and other symptoms, it may seem that FGFR3-mutated tumors have a low malignant potential. A closer examination of these early lesions, however, shows that there is no difference in the likelihood of progression between FGFR3-mutated and nonmutated cases.
We further investigated differences in the papillary/FGFR3-mutated and CIS pathways, using a selection of tumors previously screened for genome-wide molecular changes by SNP and gene expression microarrays and screened the tumor samples for TP53 and FGFR3 mutations in parallel. In fact, we found differences both in the patterns of chromosomal imbalance and in gene expression profiles. Only advanced developmental stages (stage T1) were analyzed with SNP microarrays. FGFR3 mutations were correlated with LOH and homozygous deletions of the well-known regions on chromosome 9 (7), whereas the wild-type tumors had chromosomal features usually related to CIS, including the well-documented patterns of gains and losses on chromosome 8 with the characteristic breakpoints at fragile sites (25–27), LOH of 17p13 (17), and extensive copy number gains at characteristic regions of the genome (26). Because the progression frequencies were the same for both groups, the chromosomal patterns were characteristic for pathways rather than developmental stage. In addition, the different chromosomal patterns remained unchanged in the corresponding progressed tumors, indicating molecular relationship within the pathways. However, the extent of chromosomal instability and the frequency of TP53 mutations increased in progressing/progressed tumors of both groups (FGFR3 mutated and wild type). General chromosomal instability has previously been associated with TP53 changes (28). This is consistent with the above-cited model (24) and the assumption that both chromosomal instability and TP53 changes, in contrast to FGFR3 mutations, are markers of advanced cancer development.
We also found differences in gene expression profiles between FGFR3-mutated and FGFR3 wild-type tumors. However, the analysis was biased by different developmental stages (20). This is illustrated in Fig. 3. To further validate the different pathways, we applied a recently published molecular CIS classifier (16). This classifier showed good classification results for FGFR3-mutated tumors, which further supported the existence of different molecular pathways. Interestingly, early-stage FGFR3 wild-type tumors did not express the CIS signature. Moreover, the FGFR3 gene was up-regulated in these tumors as observed in FGFR3-mutated tumors but in contrast to tumors of the CIS pathway (Fig. 4). This confirmed an earlier reverse transcription-PCR–based report (8). Thus, FGFR3 wild-type tumors with expected benign disease course probably belong to a different pathway.
In conclusion, this study addressed the role of FGFR3 mutations in bladder cancer development and the correlation with specific molecular pathways. Our results suggest a central role of FGFR3 mutations in the early development of papillary bladder tumors. We present evidence of different molecular pathways of FGFR3-mutated tumors and CIS-associated tumors by showing different chromosomal and gene expression changes. However, the progression of advanced bladder cancer was unaffected by FGFR3 mutations.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
Received 5/24/05; revised 7/14/05; accepted 8/10/05.
|
References |
|---|
|
TopAbstractMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  F. J. Sorensen, C. L. Andersen, and C. Wiuf SNPTools: a software tool for visualization and analysis of microarray data Bioinformatics, June 15, 2007; 23(12): 1550 - 1552. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Aaboe, K. Birkenkamp-Demtroder, C. Wiuf, F. B. Sorensen, T. Thykjaer, G. Sauter, K. M.-E. Jensen, L. Dyrskjot, and T. Orntoft SOX4 Expression in Bladder Carcinoma: Clinical Aspects and In vitro Functional Characterization. Cancer Res., April 1, 2006; 66(7): 3434 - 3442. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Cell Growth & Differentiation |
1 recurrence/year; frequent, 
