Learn more: PMC Disclaimer | PMC Copyright Notice
Frequent mutations of chromatin remodeling genes in transitional cell carcinoma of the bladder
Associated Data
Abstract
Transitional cell carcinoma (TCC) is the most common type of bladder cancer. Here we sequenced the exomes of nine individuals with TCC and screened all the somatically mutated genes in a prevalence set of 88 additional individuals with TCC with different tumor stages and grades. In our study, we discovered a variety of genes previously unknown to be mutated in TCC. Notably, we identified genetic aberrations of the chromatin remodeling genes (UTX, MLL-MLL3, CREBBP-EP300, NCOR1, ARID1A and CHD6) in 59% of our 97 subjects with TCC. Of these genes, we showed UTX to be altered substantially more frequently in tumors of low stages and grades, highlighting its potential role in the classification and diagnosis of bladder cancer. Our results provide an overview of the genetic basis of TCC and suggest that aberration of chromatin regulation might be a hallmark of bladder cancer.
Bladder cancer is the ninth most common cancer worldwide1, with TCC being the predominant form (representing 90% of cases). Clinically, there are two distinct groups of TCCs: ~70% of the affected individuals have superficial non–muscle-invasive TCCs (NMI-TCCs, stage Ta or T1), which tend to recur but which are generally not life-threatening, and ~30% of individuals have muscle-invasive TCCs (MI-TCCs, stages T2–T4), which are associated with a high risk of death from distant metastases2. Previous studies based on candidate gene approaches suggested that the two subgroups of TCCs may have divergent genetic backgrounds2: NMI-TCCs often harbor mutations in FGFR3 and the Ras gene family, and MI-TCCs usually have defects in TP53 and RB1 in high-grade tumors. These discoveries provided important insights into potential diagnoses and therapeutic applications, yet no comprehensive analysis of this cancer has been performed. In this study, we aimed to screen TCC systematically to identify other previously unidentified bladder-cancer–associated genes.
We performed whole-exome sequencing of the genomic DNA from tissue samples from nine individuals with MI-TCC (stage ≥ T2; Supplementary Table 1) and their matched peripheral blood samples. In this stage, called here the ‘discovery screen’, we primarily focused on MI-TCC because of its poor prognosis and survival. We generated sequencing reads with an Illumina GAII platform (Illumina, Inc.) and aligned them to the reference human genome (hg18) using MAQ3 and BWA4 software (Online Methods). On average, we sequenced the 34-Mb targeted exome regions of each sample to a mean depth of 100× or greater (Supplementary Fig. 1a and Supplementary Table 2). Over 95% of the target regions were covered sufficiently for confident variant calling (defined as ≥ 10×; Supplementary Fig. 1b).
To identify somatic mutations, we compared the sequencing data generated from the tumor-normal sample pairs of each subject (Online Methods). To eliminate any previously described germline variants, we cross-referenced potential somatic mutations against the dbSNP130 and SNP datasets of Han Chinese in Beijing (CHB) and Japanese in Toyko (JPT) from the three pilot studies in the 1000 Genomes Project. Based on these criteria, we identified 465 predicted somatic mutations, including 105 synonymous mutations, 284 missense mutations, 45 nonsense mutations, 2 splice-site changes and 29 small coding insertions or deletions (indels) (Supplementary Table 3). Of the 221 predicted somatic substitutions and 12 indels that we randomly selected for validation, we confirmed 200 (90.5%) and 8 (66.7%) by genotyping and Sanger sequencing, respectively (Online Methods). The mutation spectrum in the nine TCC samples was dominated by C:G>T:A transitions (Supplementary Fig. 2), as has been reported in several other human cancers5.
To further evaluate the mutation prevalence of 328 genes that had at least one non-silent somatic mutation in the discovery screen, we determined the coding sequences of these genes in a prevalence screen of 88 additional individuals with TCC (Supplementary Table 1). We included both individuals with MI-TCC (51 individuals) and NMI-TCC (37 individuals) to investigate whether there are genes that may be preferentially mutated in any subgroups of affected individuals. In addition, in this stage, we also screened two other well-known bladder cancer genes (FGFR3 and RB1)2 without somatic mutation in the discovery stage. In brief, we enriched the targeted exonic sequences of matched normal and tumor DNAs from the 88 untreated subjects followed by Illumina-based resequencing (Online Methods). Overall, we achieved a mean coverage depth of ~80× for all the samples sequenced, with at least ~85% of the targeted bases being sufficiently covered (≥10×; Supplementary Table 4). We determined somatic mutations using the same method as used in the discovery screen, and we selected all the non-silent mutations for verification by genotyping or Sanger sequencing (Supplementary Table 5).
We combined mutations obtained from the discovery and validation screens to prioritize the significantly mutated genes that are likely to be implicated in TCC tumorigenesis. We considered a gene to be a potential significantly mutated gene if the gene harbor confirmed non-silent mutations in at least two tumors and if its non-silent mutation rate was significantly higher (P < 0.05) than the background (Supplementary Table 6 and Online Methods). In total, we identified 54 significantly mutated genes in our dataset (Supplementary Table 7). Five of the significantly mutated genes have previously well-established roles in TCC (Table 1), including TP53 (ref. 6) (altered in 21% of TCCs), RB1 (ref. 7) (altered in 11%), HRAS (ref. 8) (altered in 10%), FGFR3 (ref. 9) (altered in 9%) and KRAS (ref. 8) (altered in 6% of TCCs). The other 49 significantly mutated genes were all new candidates with no previously well-defined roles in TCCs, and 33% (16 out of 49) of these genes were frequently mutated genes that showed mutations in more than 5% of TCCs (Table 1).
Table 1
Frequently mutated genes identified in 97 TCCs
| Gene | Non-silent somatic changes | Synonymous mutations | Total mutations | N:S ratio | Number (%)of subjects harboring non-silent mutations | P | |
|---|---|---|---|---|---|---|---|
|
| |||||||
| Missense | Nonsense, splice site or indel | ||||||
| UTXa | 5 | 17 | 1 | 23 | 22:1 | 20 (21) | 2.22 × 10−37 |
| TP53 | 13 | 8 | 1 | 22 | 21:1 | 20 (21) | 1.01 × 10−49 |
| CREBBPa | 4 | 11 | 2 | 17 | 15:2 | 13 (13) | 3.42 × 10−26 |
| EP300a | 6 | 7 | 0 | 13 | - | 13 (13) | 6.58 × 10−15 |
| ARID1Aa | 4 | 14 | 0 | 18 | - | 13 (13) | 1.99 × 10−32 |
| RB1 | 3 | 8 | 1 | 12 | 11:1 | 11 (11) | 1.00 × 10−11 |
| HRAS | 10 | 0 | 0 | 10 | - | 10 (10) | 7.89 × 10−7 |
| ERBB3 | 9 | 0 | 2 | 11 | 9:2 | 8 (8) | 3.28 × 10−18 |
| CHD6a | 6 | 1 | 2 | 9 | 7:2 | 7 (7) | 6.00 × 10−9 |
| LRP2 | 8 | 1 | 0 | 9 | - | 9 (9) | 4.47 × 10−10 |
| FGFR3 | 9 | 0 | 0 | 9 | - | 9 (9) | 1.69 × 10−18 |
| MLLa | 5 | 3 | 1 | 9 | 8:1 | 7 (7) | 2.00 × 10−11 |
| LAMA4 | 8 | 0 | 1 | 9 | 8:1 | 7 (7) | 4.47 × 10−9 |
| ANK2 | 7 | 0 | 1 | 8 | 7:1 | 7 (7) | 3.07 × 10−8 |
| NF1 | 5 | 2 | 1 | 8 | 7:1 | 7 (7) | 8.76 × 10−10 |
| ESPL1 | 5 | 1 | 1 | 7 | 6:1 | 6 (6) | 2.76 × 10−11 |
| NCOR1a | 5 | 2 | 0 | 7 | – | 6 (6) | 3.21 × 10−6 |
| KRAS | 6 | 0 | 1 | 7 | 6:1 | 6 (6) | 3.29 × 10−26 |
| MLL3a | 2 | 4 | 1 | 7 | 6:1 | 5 (5) | 1.58 × 10−6 |
| ANK3 | 5 | 0 | 1 | 6 | 5:1 | 5 (5) | 9.10 × 10−4 |
| ZFHX3 | 5 | 0 | 0 | 5 | – | 5 (5) | 5.52 × 10−3 |
Most notably, we detected frequent non-silent mutations in eight genes that are involved in the chromatin remodeling process (Table 1). Of these genes, the most frequently altered were the histone demethylase gene (HDMT) UTX (mutated in 21% of TCCs), two histone acetyltransferase (HAT) genes, CREBBP and EP300, and the SWI/SNF-related chromatin remodeling gene ARID1A (with all these genes being mutated in 13% of TCCs). We also detected aberration of each of the remaining four genes in more than 5% of TCCs, including the histone methyltransferase (HMT) genes MLL and MLL3 and two other chromatin remodeling genes, NCOR1 (encoding a constitutive subunit for the N-coR-HDAC3 complex that possesses histone deacetylation (HDAT) activity10) and CHD6 (encoding a component of SNF2/RAD54 helicase family that remodels chromatin to allow cell-type–specific gene expression11).
Genetic alterations of the genes involved in the chromatin remodeling process were also reported in various other tumors in recent studies. Inactivating mutations in UTX were observed in multiple types of cancers12; frequent mutations of ARID1A were identified in ovarian clear cell carcinoma13; and truncating mutations in the HMT gene SETD2, the HDMT gene JARID1C and another member of the SNF/SWI complex, PBRM1, were identified in renal cell carcinoma14,15. Aberrations of the chromatin remodeling genes may directly lead to the misregulation of multiple downstream effector genes, consequently promoting the tumorigenesis process16. Nevertheless, to our knowledge, except for UTX, which showed alterations in 2 of 14 TCC cell lines12, genetic mutations in the genes involved in the chromatin remodeling process have not yet been reported in the primary tumors of TCC. In our study, we that found 57 (59%) subjects harbored non-silent mutations in chromatin remodeling genes, indicating that disruption of the chromatin remodeling machinery may be one of the main mechanisms that leads to TCC.
In addition to the frequency of mutations in a gene, the nature of the mutations can provide valuable information for classifying the mutated genes as oncogenes (dominant genes) or as tumor suppressors (recessive genes)17. Our observations clearly distinguished UTX as a tumor suppressor gene in TCC, as evidenced by its mutation pattern of significant enrichment (16 of 22 mutations; P < 0.001) with truncating mutations. These included 11 nonsense mutations, 4 frameshift indels and 1 splice site change (Fig. 1a). All but one of these were predicted to truncate the JmjC domain, which is essential for the demethylase activity of the protein product. We observed the same mutation pattern in ARID1A (Fig. 1b), in which truncating mutations accounted for 14 (78%) of 18 mutations (P < 0.001) and which generated defective proteins lacking the ARID DNA-binding domain or LXXLL motif. The patterns of mutations in CREBBP and EP300 also suggested their potential tumor suppressing roles in TCC. Recent studies showed that they behave as tumor-suppressor genes in various tumors18,19. In our screen, 73% and 54% of the mutations in these two genes are predicted to disrupt the key HAT functional region in CREBBP and to truncate the CH3 domain that mediates the interactions between p300 and other cellular proteins (for example, p53 and E2F1)20 in EP300, respectively (Fig. 1c). Of note, we also identified recurrent truncating mutations in other frequently mutated chromatin remodeling genes (MLL, MLL3 and NCOR1). In previous studies, MLL was shown to be a dominant cancer gene affected by recurrent translocations in leukemias21. However, our data, as well as the mutation data recently deposited in COSMIC database (see URLs), supported that MLL and its homolog MLL3 are likely to act recessively in some solid tumors22,23.
Somatic mutations in UTX, ARID1A and CREBBP-EP300. The types and relative positions of confirmed somatic mutations are shown in the transcripts of UTX (a), ARID1A (b) and CREBBP-EP300 (c) using the following symbols: red stars, nonsense mutations; bullets, missense mutations; red triangles, frame-shift indels; green triangles, in-frame indels; and diamond, mutations at splice sites. Domains and motifs in each encoded protein product, as well as the key region responsible for the histone acetyltransferase activity of CREBBP, are also indicated. TPR, tetratricopeptide repeat; JmjC, transcription factor jumonji/aspartyl beta-hydroxylase; ARID, AT-rich interactive domain; LXXLL, C-terminal leucine-rich LXXLL motif; CH1, CH2 and CH3, cysteine–histidine-rich domains; KIX, CREB-binding domain; Bromo, bromodomain.
The correlations among mutations in the frequently altered genes could provide insights into their functional interactions. We also confirmed in our study the well-known examples of the negative correlations of mutations in TP53 and FGFR3 (ref. 24) and of mutations in TP53 and HRAS (ref. 2) (P < 0.05; Supplementary Table 8), with no sample here having mutations in both TP53 and FGFR3 or HRAS (Fig. 2). We also found significantly negative correlations between mutations in UTX and the Ras genes (KRAS and HRAS). In addition, we observed positive correlations of mutations between the following pairs of genes: ANK2 and CHD6, LRP2 and ZFHX3, and CHD6 and LRP2 (all P < 0.05). Notably, we did detect concurrent mutations in multiple genes within the chromatin remodeling process (Fig. 2), although these correlations were not statistically significant. The eight frequently mutated chromatin remodeling genes can be classified into five subgroups according to their functional specificities: HMT (MLL3 and MLL), HDMT (UTX), HAT (CREBBP and EP300), HDAT (NCOR1) and members of the nucleosomal remodeling complex (ARID1A and CHD6). We found that 16% of TCCs had mutations in at least two of the five subgroups (Fig. 2). We also observed concurrent mutations within the HAT subgroup between CREBBP and the highly homologous gene EP300. Distinct chromatin remodeling processes are often co-regulated and are functionally dependent25. The identification of concurrent genetic alterations in either the same or different subgroups of chromatin remodeling genes probably reflects that they can function synergistically or independently to promote tumorigenesis. We next investigated the relationship between the prevalence of each frequently mutated gene and tumor stage and/or grade. As previously shown2, incidence of TP53 mutations increased in TCCs of high stages or grades, whereas the frequency of FGFR3 mutations decreased with increasing tumor stage and grade of TCC (Fig. 3). Similar to the frequency distribution of FGFR3 mutations across different tumor stages and grades, our analysis indicated that mutations in UTX were likely to be associated with tumors of early developmental stage. The mutation frequency of UTX was significantly associated with tumor grade (P = 0.008), with grade 1 accounting for 35%, grade 2 accounting for 15% and grade 3 accounting for 4% of tumors with mutations in UTX. We also found a negative correlation between mutations in UTX and tumor stage. The incidences of UTX mutations in stage Ta and T1 and in ≥stage T2 tumors were 32% and 13%, respectively (P = 0.04). For the other chromatin remodeling genes (ARID1A, CREBBP and EP300), we found no significant correlations between their mutation frequencies and tumor stage or grade (Fig. 3). In summary, our study provides a comprehensive catalog of genetic alterations in TCC, and we discovered 49 new significantly mutated genes associated with TCC. Eight of these are chromatin remodeling genes that harbor frequent mutations in the majority of NMI-TCCs and MI-TCCs. Most of these remodeling genes are likely to act as tumor suppressors and have been reported to play pivotal roles in various tumors other than TCC. These findings further our current understanding of bladder cancer and other human cancers and indicate that genetic alterations and epigenetic deregulations may cooperatively contribute to tumor genesis and progression. Our data can serve as a valuable basis for future studies on TCC and suggest the necessity of epigenomics research in the field of cancer studies.
Concurrent and mutually exclusive mutations observed in the frequently mutated genes. For each gene (row) indicated, tumors (columns) with or without mutations are labeled in red or blue, respectively. P values for the occurrence of concurrent and mutual exclusive mutations in two genes across tumors are provided in Supplementary Table 8. We selected only genes harboring non-silent mutations in at least five subjects for this analysis.
Frequencies of mutations in highlighted genes across different tumor stages and grades. T, the stage of tumors under the TNM (tumor, lymph node and distant metastasis) classification system; G, grade of tumors. For each tumor stage or grade, we calculated the frequency of mutations in a given gene as the proportion of tumors harboring no silent mutations in the gene among all tumors of the indicated stage or grade. We determined the significance of the correlation between mutations in each gene (altered in at least 10% of TCCs) and tumor grade or stage using χ2 tests.
COSMIC database, http://www.sanger.ac.uk/perl/genetics/CGP/cosmic (v52 release); NCBI Consensus Coding Region dataset, http://www.ncbi.nlm.nih.gov/CCDS/CcdsBrowse.cgi; SAMtools, http://samtools.sourceforge.net/.
ONLINE METHODS
Sample description and preparation
The primary tumor samples and matched peripheral blood were obtained from individuals with TCC newly diagnosed at the member institutions of the Urinogenital Cancer Genomics Consortium (UCGC) in China. A signed written consent from each subject was obtained before recruitment for the study according to the regulations of the institutional ethics review boards. Detailed clinical information for the subjects is summarized in Supplementary Table 1. All the specimens were snap frozen in liquid nitrogen upon collection and immediately stored at −80 °C for further study. The hematoxylin eosin–stained sections prepared using the cancerous tissues were microscopically evaluated by two independent pathologists. In the present study, only TCCs with malignant cell purities of over 85% were selected for DNA extraction and subsequent sequencing.
Illumina based whole-exome sequencing and targeted exon sequencing
Genomic DNA for all tumors and blood samples from both the discovery and prevalence cohort were fragmented and hybridized to commercially available capture arrays for enrichment. In the discovery stage, the exome capture procedure was performed with NimbleGen 2.1M Human Exome Arrays, which are capable of enriching exonic sequences of more than 18,000 protein-coding genes deposited in the NCBI Consensus Coding Sequence Region database (see URLs). In both the discovery and prevalence screens, the enriched DNA fragments were sheared to 200 bp on average and subjected to standard Illumina Genome Analyzer library preparation according to Illumina’s protocol. The shotgun libraries were sequenced with the Illumina GAII platform, and 80-bp single ended reads were generated. Image analysis and base calling were performed by the Genome Analyzer Pipeline version 1.3 using default parameters.
Read mapping and detection of somatic mutations
After removing reads containing sequencing adaptors and low quality reads with more than five unknown bases, the high quality single-end reads were aligned to the NCBI human reference genome (hg18) using MAQ3 with the default options. To identify indels, the high-quality reads were gap aligned to the reference sequence using BWA4. We then performed local realignment of the BWA-aligned reads using the Genome Analysis Toolkit (GATK)26. The raw lists of potential somatic substitutions were called by VarScan27 (v2.2) based on the MAQ alignments. In this process, several heuristic rules were applied: (i) both the tumor and matched normal samples should be covered sufficiently (≥10×) at the genomic position being compared; (ii) the average base quality for a given genomic position should be no less than 15 in both the tumor and normal samples; (iii) the variants should be supported by at least 10% of the total reads in the tumors, and no high-quality variant-supporting reads were allowed in normal controls; and (iv) the variants should be supported by at least five reads in the tumors. Using the same criteria, the preliminary list of somatic indels was called out by GATK based on the local realignment results. After these two steps, germline variants could be effectively removed. To further reduce the false positive calls, variations including single nucleotide variants and indels were called with the SAMtools software package (see URLs) in the tumors. We eliminated all somatic variants that fulfilled any one of the following filtering criterion: (i) variants with Phred-like scaled consensus scores or SNP qualities <20; (ii) variants with mapping qualities <30; (iii) indels represented by only one DNA strand; or (iv) substitutions located 30 bp around the predicted indels. To deal with false positives associated with pseudo-gene issues or repeat sequences, simulated reads (80 bp in length) containing the putative mutations were generated and aligned to the reference genome. For a given variant, if more than 10% of the simulated variant-containing reads could not be uniquely mapped to the reference genome, this variant was discarded.
Validation of somatic mutations by mass spectrum or Sanger sequencing
Validation of the non-silent somatic substitutions by mass spectrum was performed with the MassArray platform of Sequenom by determining their genotypes in the tumors and matched blood. The genotyping assay and base-calling procedures were performed as previously described28. We considered the genotyping assay to have failed if the Sequenom software was unable to design primers for PCR amplification or base extension at the primer design stage or if the observed peak for a given assay was not significant enough for a confident call at the base-calling stage. To validate somatic non-silent substitutions and indels with Sanger sequencing, PCR primers designed for the putative somatic variants were initially used to amplify the source DNA from the tumors. If the mutations were successfully confirmed in the tumors, the same primer pairs were used to amplify the normal DNA from the same subjects to determine the somatic statuses of the observed mutations.
Statistical analysis of the significantly mutated genes
To estimate the background mutation rates, we conservatively treated all the predicted mutations identified in the discovery screen as passenger mutations. These predicted mutations were classified into 19 different categories according to their sequence context and mutation types. For each mutation category, i, we let the observed number of mutations of the category be mi and the total number of successfully sequenced nucleotides (≥10×) of this category in the nine tumors be ni; the background mutation rate for this category, bi, was calculated as mi/ni. The estimated background mutation rates for each category are listed in Supplementary Table 6. To test whether the non-silent mutation rate of a gene was significantly higher than the background, the confirmed mutation data for the gene obtained from the discovery and prevalence screens were combined. Then we estimated the passenger probability for each gene in turn as previously described23. To be specific, the probability (pgi) of obtaining the observed number of mutations of each category (i) in gene g was estimated from a binominal distribution, with bi as the success probability. The number of available nucleotides for each category was the total number of sufficiently covered (≥10×) bases for that particular category in all the 97 TCCs. The passenger probability (pg) for gene g was calculated to be the product of the 19 category-specific probabilities, or . We then determined the P value for each gene by the likelihood-ratio test as previously described29.
Statistical analysis of inactivating mutations in chromatin remodeling genes
To assess whether nonsense mutations were over-represented in the eight chromatin remodeling genes, we calculated the probability (Pn) of single base changes that would result in nonsense mutations by chance. To this end, the coding sequence of each gene was represented by its longest transcript, and every single base in the coding region was changed into the other three different bases. We obtained Pn for each gene by dividing the number of nonsense mutations observed in each gene by the length of the coding region. The significance of nonsense mutation enrichment was determined by a binomial test with the parameter Pn as the hypothesized probability of success.
Concurrence and mutual exclusion analysis
We performed the concurrence and mutual exclusion analysis on the significantly mutated genes that showed non-silent mutations in at least five tumors by permutation test as previously described30, with minor modifications. Briefly, we kept the number of samples mutated in a given gene the same as that observed in our data and permutated the mutations across samples. For a given sample, s, we let the number of the significantly mutated genes showing mutations in that particular sample be gs, and the probability (ps) that the sample s was labeled as mutated in the permutation test was defined as . For each permutation, we recorded the number of samples with concurrent and exclusive mutations (Pc and Pe, respectively) for each pair of genes and compared them with numbers observed in original data (Nc and Ne). We repeated this process 10,000,000 times and summarized the frequencies of Pc ≥ Nc and Pe ≥Ne, respectively. These frequencies were used as the empirical P values under the null hypothesis (that is, assuming no correlation between genes).
Acknowledgments
This work was supported by grants from the National High Technology Research and Development Program of China (863 Program, 2006AA02A302 to H.Y. 2009AA022707 to X.Z.), the Promotion Program for Shenzhen Key Laboratory, Shenzhen, China (CXB200903090055A and CXB201005250016A to Z.C.), and Research Fund for the Doctoral Program of Higher Education of China (20100001110100).
Footnotes
Accession codes. All sequencing data from this study are deposited in NCBI Sequence Read Archive under the accession number SRA038181.
Note: Supplementary information is available on the Nature Genetics website.
AUTHOR CONTRIBUTIONS
Jun Wang, Z.C., Jian Wang, H.Y., S.L. and Y.G. managed the project. A.T., X. Li, L.Z., Z. Li, F.Z., X. Zhao, C. Liang, C. Liu, Y.W., L.S., Z.J., Jing Chen, S. Wu, Z.Z., R. Yang, J. Zhao, C.X., Z.G., J.Y., H. Zhang and W.Y. prepared the samples. X.H., R.W., P.H., H.J., J.L. and X. Zhang performed the sequencing. Y.G., G.G., Y.H., S.G., C.C., M.H., W.J., R. Ye, Z. Liu, S. Wan, H. Zheng, K.K., M.L.N. and Y.L. performed the bioinformatic analysis. Y.H., X.H., Jinnong Chen, S.Y., X. Liu, D.F. and J. Zou performed the validation of somatic mutations. G.G. and Y.H. wrote the paper. Y.G., Jun Wang, Z.C., X.H., Y.L., D.T. and X.S. revised the paper.COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.