Whole-genome sequencing identifies genomicheterogeneity at a nucleotide and chromosomallevel in bladder cancerCarl D. Morrisona,1,2, Pengyuan Liub,1, Anna Woloszynska-Readc, Jianmin Zhangd, Wei Luoc, Maochun Qine,Wiam Bsharaf, Jeffrey M. Conroya, Linda Sabatinif, Peter Vedellb, Donghai Xiongb, Song Liue, Jianmin Wange, He Shend,Yinwei Lid, Angela R. Omilianf, Annette Hillf, Karen Headf, Khurshid Gurug, Dimiter Kunnevh, Robert Leache,Kevin H. Enge, Christopher Darlaka, Christopher Hoeflicha, Srividya Veerankia, Sean Glennd, Ming Youb,Steven C. Pruitth, Candace S. Johnsonc, and Donald L. Trumpi

aCenter for Personalized Medicine and Departments of cPharmacology and Therapeutics, dCancer Genetics, eBiostatistics and Bioinformatics, fPathology,gUrology, hMolecular and Cellular Biology, and iMedicine, Roswell Park Cancer Institute, Buffalo, NY 14263; and bDepartment of Physiology and the CancerCenter, Medical College of Wisconsin, Milwaukee, WI 53226

Edited* by Carlo M. Croce, The Ohio State University, Columbus, OH, and approved January 2, 2014 (received for review July 22, 2013)

Using complete genome analysis, we sequenced five bladder tumorsaccrued from patients with muscle-invasive transitional cell carcinomaof the urinary bladder (TCC-UB) and identified a spectrum of genomicaberrations. In three tumors, complex genotype changes were noted.All three had tumor protein p53 mutations and a relatively largenumber of single-nucleotide variants (SNVs; average of 11.2 permegabase), structural variants (SVs; averageof 46), or both. This groupwas best characterized by chromothripsis and the presence ofsubclonal populations of neoplastic cells or intratumoralmutationalheterogeneity. Here, we provide evidence that the process of chromo-thripsis in TCC-UB is mediated by nonhomologous end-joining usingkilobase, rather thanmegabase, fragments of DNA, which we refer toas “stitchers,” to repair this process. We postulate that a potential uni-fying theme among tumorswith themore complex genotype group isadefective replication–licensingcomplex.Asecondgroup (twobladdertumors) had no chromothripsis, and a simpler genotype, WT tumorprotein p53, had relatively few SNVs (average of 5.9 per megabase)and only a single SV. Therewas no evidence of a subclonal populationof neoplastic cells. In this group,weused a preclinicalmodel of bladdercarcinoma cell lines to study a unique SV (translocation and amplifica-tion) of the gene glutamate receptor ionotropic N-methyl D-aspertateas a potential new therapeutic target in bladder cancer.

next-generation sequencing | tumor heterogeneity | GRIN2A | replication

Transitional cell carcinoma arising in the urinary bladder (TCC-UB) is a frequent cause of morbidity and mortality, and amongpatients in the United States, it is one of the most costly cancers totreat (1, 2). The traditional somatic genetic basis of TCC-UB isa distinct division of low-grade papillary tumors from high-gradeinvasive tumors. Low-grade papillary superficial tumors are gener-ally characterized by constitutive activation of the receptor tyrosinekinase–Ras pathway, and they have activating mutations in theHRAS and fibroblast growth factor receptor 3 (FGFR3) genes(3–6). In contrast, high-grade invasive TCC-UB is characterizedby alterations in the tumor protein p53 (TP53) and retino-blastoma 1 (RB1) pathways. These genes normally regulate thecell cycle by interacting with the Ras–mitogen-activated proteinkinase signal transduction pathway (7, 8). Both low-grade papillaryand high-grade invasive tumors frequently have loss of chromosome9. This loss presumably inactivates the p16 gene and is an earlyevent in the initiation of TCC-UB (9, 10)Although TP53, cyclin-dependent kinase inhibitor 2A (p16), RB1,

HRAS, and FGFR3 abnormalities have been well described in TCC-UB, there are limited data on the more complete genomic analysisof TCC-UB (11). A recent study focusing on genome-wide copynumber analysis showed extensive heterogeneity across all sub-types of TCC-UB to such an extent that precise moleculargroupings were difficult to define (12). In this study, similar to

earlier studies in melanoma (13) and medulloblastoma (14), ev-idence of the association of TP53 mutations with specific copynumber alterations, referred to as chromothripsis, was noted. Thestudy in medulloblastoma (14) was particularly intriguing in thatidentification of a molecular subclass with TP53 mutations wasassociated with chromothripsis and a more aggressive clinicaloutcome was noted. Chromothripsis, or the shattering of two ormore chromosomes and their reassembly into derivative chro-mosomes, is different from other types of genomic instability,which tend to occur on a genome-wide basis (15, 16). Chromo-thripsis is different in that it includes one to three alternatingcopy number states across the derivative chromosome, an asso-ciation with changes in heterozygosity, and numerous genomicrearrangements in localized chromosomal regions likely occurringin condensed chromosomes (15). There is evidence to suggestthat the primary mechanism of reassembly of the derivativechromosome in chromothripsis is nonhomologous end-joining(NHEJ) (14). With the advent of next-generation sequencing(NGS) allowing for detailed genomic analysis, chromothripsis

Significance

Genetic alterations are frequently observed in bladder cancer.In this study, we demonstrate that bladder tumors can beclassified into two different types based on the spectrum ofgenetic diversity they confer. In one class of tumors, we ob-served tumor protein p53 mutations and a large number ofsingle-nucleotide and structural variants. Another characteristicof this group was chromosome shattering, known as chromo-thripsis, and mutational heterogeneity. The other two bladdertumors did not show these profound genetic aberrations, butwe found a novel translocation and amplification of the geneglutamate receptor ionotropic N-methyl D-aspertate, a poten-tially druggable target. Advancements in bladder cancertreatment have been slow. Understanding the genetic land-scape of bladder cancer may therefore help to identify newtherapeutic targets and bolster management of this disease.

Author contributions: C.D.M., W.B., K.G., M.Y., C.S.J., and D.L.T. designed research; C.D.M.,J.Z., W.L., J.M.C., L.S., A.R.O., A.H., and K.H. performed research; M.Q., C.D., C.H., andS.V. contributed new reagents/analytic tools; C.D.M., P.L., J.Z., M.Q., J.M.C., P.V., D.X.,S.L., J.W., H.S., Y.L., D.K., R.L., K.H.E., C.D., C.H., S.V., and S.G. analyzed data; and C.D.M.,P.L., A.W.-R., J.Z., J.M.C., S.C.P., C.S.J., and D.L.T. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.1C.D.M. and P.L. contributed equally to this work.2To whom correspondence should be addressed. E-mail: carl.morrison@roswellpark.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental.

E672–E681 | PNAS | Published online January 27, 2014 www.pnas.org/cgi/doi/10.1073/pnas.1313580111

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1313580111&domain=pdf&date_stamp=2014-01-30mailto:carl.morrison@roswellpark.orghttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1313580111

has been identified more frequently (17). Additionally, thepresence of these complex genomic events and their potentialassociation with TP53 mutations may contribute to a better un-derstanding of cancer, including TCC-UB.NGS technologies provide other evidence of complex genomic

heterogeneity, such as the recent identification of subclonal pop-ulations of cells with mutations distinct from the dominant clonalpopulation of cells within one tumor or between a primary, re-current, or metastatic tumor from one patient (18–20). Importantly,a recent study in chronic lymphocytic leukemia (CLL) (21) showedhow selective pressures on cancer cells, such as chemotherapy, se-lect for these subclonal populations to become the dominantclone contributing to genomic heterogeneity. It is not yet certainwhether broad measurements of genomic heterogeneity willhave an impact on molecular classification of cancer, but it islikely that they will significantly contribute to biological differ-ences, and therefore have an impact on patient outcomes.To evaluate the spectrum of genomic heterogeneity in TCC-

UB, we performed complete genome sequencing of five high-grade muscle-invasive tumors and matching germ-line blood (SIAppendix, Table S1), and validated a subset of our findings inmore than 300 bladder cancer specimens. Our overall resultsshowed a great deal of genomic heterogeneity at either extremeof a spectrum of genomic complexity.

ResultsOverview of Somatic Alterations Reveals Heterogeneity BetweenPatients. At one end of the spectrum was a more complex geno-type, characterized by frequent single-nucleotide variants (SNVs)and structural variants (SVs), TP53 mutation (TP53mut), CDNK2A(p16) deletion (p16del), frequent mutations in known cancer-related genes, SV breakpoints that often precisely align withsegmental copy number states indicating chromothripsis, andevidence of subclonal intratumoral heterogeneity (Fig. 1). Wefound (i) evidence that SV breakpoints can have a unique as-sociation with copy number in the context of chromothripsisthat may be related to a process of genomic amplification, (ii)complex genomic rearrangements mechanistically use kilobasefragments of DNA that we refer to as “stitchers” as part of anNHEJ DNA repair process, and (iii) some cases of TCC-UB doshow intratumoral mutational heterogeneity. At the other end ofthe spectrum, was a simpler genotype, with few SNVs and SVs,infrequent mutations in any known cancer-related gene in theCancer Gene Census, TP53WT (TP53wt), p16WT (p16wt), and noevidence of chromothripsis. In this group, we provide (i) an un-equivocal demonstration that amplified interchromosomal trans-locations (CTXs) can be found in bladder carcinoma and (ii)evidence of rare events of translocation and, more frequently,amplification of GRIN2A in a subset of TCC-UB representinga potential therapeutic target.

Somatic Mutation Analysis Identifies Intertumoral Genomic Heterogeneityat the Nucleotide Level. We obtained 44.8-fold mean sequence cov-erage for each tumor and 39.5-fold mean sequence coverage for thematching normal tissues (SI Appendix, Table S2). To identify so-matic events, we compared the sequencing data of each tumor withthat of matched blood using multiple algorithms and filtered the listby reference to the Single Nucleotide Polymorphism database(dbSNP) build 130 and 1000 Genomes Project (SI Appendix,SI Methods and Tables S3–S6). There was a wide variation inthe number of somatic mutations per tumor, ranging from 14,256in case 16933 to 49,889 in case 19685 (average of 29,326). Tumorsfrom two patients (cases 16933 and 17802) had fewer somaticmutations (Fig. 1), including both SNVs and SVs, and wereTP53wt. The tumors of three patients (cases 18698, 18195, and19685) were TP53mut and contained a much larger number ofsomatic mutations, providing additional intertumoral genomicheterogeneity at the nucleotide level. Not unexpectedly, four or five

tumors (cases 18693, 18195, 19685, and 17802) contained mutationsin one or more chromatin remodeling genes, including mutations inNSD1, PBRM1, KDM6A, ARID2, APC, and EP300, which wereidentified in all of these tumors, consistent with prior exomic se-quencing in this tumor type (22). In this same group of four tumors,there were 11 mutated genes present in two or more samples (SIAppendix, Table S7), including TP53, CTBP2, ZFHX4, XIRP2,WDR89, PCMTD1, PABPC3, MCM4, GXYLT1, CDCA7L, andCC2D1A. Genes with a nonsynonymous mutation and coding re-gion deletion (DEL) in one or more samples included ANKRD11and CC2D1A. In both instances, the mutation and DEL occurred inthe same case. Neither of these two genes has been identifiedpreviously as mutated in TCC-UB (22), and reports of mutations inother tumor types have been reported only rarely in the CosmicMutation Database (www.sanger.ac.uk/genetics/CGP/cosmic/).

MCM4 and Replication–Licensing Complex Defects. With the excep-tion of TP53, none of the above-listed genes was previouslyreported as mutated in an Asian cohort of 97 patients usingwhole-exome sequencing (22). Among these 97 patients, MCM3was mutated in one tumor. MCM3 is part of a six-gene MCM2–7replication–licensing complex that binds chromatin during theG1-phase of the cell cycle and is required for initiation of DNAreplication in the subsequent S-phase. In our study, two tumors(cases 18195 and 19685) both showed mutations in the MCM4gene of this replication–licensing complex, which were validatedby Sanger sequencing. Both of these tumors were TP53mut, hadthe largest number of SNVs and SVs, showed chromothripsisinvolving multiple chromosomes, and demonstrated intratumorgenomic heterogeneity. As we postulate in our discussion, all ofthese events may have a single underlying association througha replication–licensing complex defect (see Fig. 7).

010

20304050

607080

90100

16933 17802 18698 18195 19685

SNV/MbNV/Mb Structural Variants

TP53mut/p16del

TP53mut/p16wt

TP53mut/p16del

TP53wt/p16del TP53wt/p16wt

Increasing genomic complexity

TP53 mutant

p16 deletion TP 53/p16 wild type

Single structural variant (SV) Fewer single nucleotide variants (SNVs) TP53 wild type Chromothripsis absent No evidence of subclonalpopulations

Frequent structural variants (SVs) Frequent single nucleotide variants (SNVs) TP53 mutant Chromothripsis involving multiple chromosomes Subclonal populations identified in 2 of 3

Fig. 1. Number of SNVs per megabase (Mb) of DNA and total number ofvalidated SVs for each of five patients with muscle-invasive TCC-UB used forwhole-genome sequencing. Three of the five tumors (cases 18195, 18698,and 19685) had many more SVs and SNVs than the other two tumors andwere also TP53mut. Two of the five tumors (cases 16933 and 17802) had veryfew SVs and SNVs, and were also TP53wt. Patient 17802, although havingonly one SV, shared in common with the TP53mut group a p16 (CDKN2A)DEL. Patient 16933 was negative for p16del and TP53wt status, had nomutations in any known cancer-related genes in the Cancer Gene Census,and had a single distinct SV represented by a CTX between the SCN8A geneat 12q14 and the GRIN2A gene at 16p13.2.

Morrison et al. PNAS | Published online January 27, 2014 | E673

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.sanger.ac.uk/genetics/CGP/cosmic/

Mutated Subclones Contribute to Intratumoral Genomic Heterogeneityat the Nucleotide Level. Clusters of mutations with dissimilar variantallele frequencies in an individual case provide evidence ofintratumoral heterogeneity and support the existence of mul-tiple clones of neoplastic cells with different genotypes withinone tumor (20, 21, 23–26). By itself, the presence of multipleneoplastic clones within a single tumor implies a more complexgenome, and recent evidence supports an evolution toward a moreaggressive phenotype (26). To assess intratumoral clonality, thefrequency data of tumor variant alleles for all identified somaticmutations were input into an R function “density” to estimate theempirical probable density function of allele frequencies (19) (Fig.2). Estimates of clonality were determined using a kernel densityanalysis of tumor variant allele frequency, which was performedseparately for each tumor. The two bladder tumors with the sim-pler genotype (cases 16933 and 17802) showed a Gaussian distri-bution of variant frequencies without evidence of subclonalpopulations of neoplastic cells. In contrast, cases 18195 and 19685,with a more complex genotype, showed an obvious skew of thenormal distribution. The kernel density analysis plot for case 18195shows at least two neoplastic clones centered on variant allelicfrequencies of 20% and 40%. Case 19685 showed a similar skeweddistribution but without obvious peaks of variant allelic frequen-cies, perhaps reflecting the resolution of coverage with whole-genome sequencing (Fig. 2). The last case, 18698, although TP53mt

and having a relatively large number of mutations, did not showsubclonal populations by this analysis, implying there is somecontinuum across these groups. Larger numbers of patients arenecessary to determine if the biological importance of these

subclonal populations of neoplastic cells portend a worse prog-nosis, as has been previously described in leukemia (21).

SV Analysis Identifies Genomic Heterogeneity at the ChromosomalLevel. To detect chromosomal rearrangements, we searched forfragments in which the sequence from the paired-end readmapped discordantly to the reference genome and further re-fined them by de novo assembly (SI Appendix, SI Methods). Atotal of 263 putative somatic rearrangements were predicted. Toassess the accuracy of these predictions, PCR was performedacross the putative breakpoints for both the tumor and germ-lineDNA. We confirmed, by either PCR or FISH, 150 (57%) ofthese predicted SVs as true somatic rearrangements (SI Appen-dix, Table S8), 6 (2%) as germ-line, and 59 (22%) as false, andthe remaining 48 (19%) failed to produce a PCR in either thetumor or germ-line DNA. The number of somatic SVs per tumorvaried greatly (range: 1–79) with a median of 20 (average of 30)per tumor (SI Appendix, Fig. S1 and Table S9). Tumors witha higher number of SNVs also showed more SVs (Fig. 1). DELswere the most common SV identified [58 (39%) of 150], fol-lowed by inversions [INVs; 56 (37%) of 150] and CTXs [34(23%) of 150] (SI Appendix, Fig. S2 and Table S10). Intra-chromosomal translocations [2 (1%) of 150] were infrequentand only identified in two of the five tumors. Interestingly, bothtumors with the simpler genotype (cases 16933 and 17802) hadonly a single SV; in case 17802, the SV was a p16del. The re-maining tumors (cases 18698, 18195, and 19685) had the morecomplex genotype, showed an average of 49 (median of 49) SVs,and were evaluated further.

Chromothripsis Contributes to Genomic Heterogeneity at theChromosomal Level. In tumors with the more complex genotype,SVs were not evenly distributed across all chromosomes; 54% (81of 150) involved chromosomes 4, 5, and 6 in a pattern consistentwith chromothripsis (SI Appendix, Figs. S3–S6 and Table S11). Thehighest number of SVs was identified in chromosome 4 (total of31), followed closely by chromosomes 5 (total of 27) and 6 (total of23). Interestingly, when adjusted for SVs per 50 Mb of DNA (SIAppendix, Figs. S7–S11 and Table S12), chromosome 21 (8.6 SVsper 50 Mb) was the most frequently involved chromosome, fol-lowed closely by chromosomes 4 (8.3 SVs per 50 Mb), 5 (7.7 SVsper 50 Mb), and 6 (6.6 SVs per 50 Mb). Most of the SVs forchromosome 21 were the result of CTXs with chromosome 5 ina pattern of chromothripsis for case 19685. In addition, the pre-diction for SVs to involve chromosomes 4, 5, 6, and 21 was pri-marily associated with the occurrence of INVs and CTXs ratherthan DELs (SI Appendix, Figs. S8–S11). Some chromosomes, suchas chromosomes 1, 3, 8, 15, 20, and 22 and chromosome X, showedvery few SVs even when adjusted for size.

Chromothripsis Does Not Lead to Functionally Relevant Gene Fusions.Among the 33 CTX events in the three tumors with evidence ofchromothripsis, 32 had one or both breakpoints in an intergenicregion without the possibility of a gene fusion event. In a similarfashion, none of the 56 INVs resulted in a predicted functionalgene fusion event. One CTX event had juxtaposed (intronic–intronic) and appropriately aligned in-frame coding regionspredicted to result in a putative productive fusion protein. In thisevent, the first exon of CDH10 on chromosome 5, encodinga type II classical cadherin that mediates calcium-dependentcell–cell adhesion, was predicted to join with the last three exonsof CAB39L, a protein that binds and activates serine/threoninekinase STK11. A CDH10/CAB39L translocation was validated inthe index case (case 18195) by PCR but not by FISH. Additionalbreak-apart FISH studies for both CDH10 and CAB39L failed toshow any other translocations in the validation cohort of 329bladder cancer samples. We show later that this concept of PCR-positive, FISH-negative translocation events is a common event

Fig. 2. Kernel density analysis plots of tumor variant allele frequency (Freq)to assess intratumoral heterogeneity. Shown are plots for all five tumorsrepresenting either end of the spectrum of genomic complexity. Each plotgraphs the variant allelic frequency (x axis) vs. the density of the variantallelic frequencies (y axis). Plots with a single peak (cases 16933, 17802, and18698) represent clusters of mutations with similar variant allelic frequenciesand no evidence of subclonal populations of neoplastic cells. Plots with twoor more peaks (cases 18195 and 19685) represent tumors with subclonalpopulations of neoplastic cells and a more complex genome.

E674 | www.pnas.org/cgi/doi/10.1073/pnas.1313580111 Morrison et al.

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1313580111

in chromothripsis. Similar to the findings in a recent study inprostate cancer (27), chromothripsis, although a marker of ge-nomic instability, does not lead to recurrent functionally relevantfusion genes in TCC-UB.

SVs Associated with Chromothripsis Align with Segmental CopyNumber Changes. Although SVs in the tumors with the complexgenotype did not have apparent single-gene implications asdriver mutations in comparison to the simple genotype, the re-lationship of SVs to copy number changes suggests other bi-ologically relevant mechanisms. In this regard, a pattern of CTXsand INVs closely aligned with segmental copy number states wascharacteristic of the complex genotype. To define the genomicsignificance of these events better, we developed an enhancedSV viewer similar to Circos but with a linear view and the ca-pability to “zoom in” for a more detailed view. This enhancedlinear view showed frequent sharing of breakpoints for INVs andCTXs. Most interestingly, when the copy number profile acrossthese regions was added to the viewer, an alternating change ofone to a few copies with transitions aligned with shared break-points for both INVs and CTXs consistent with chromothripsiswas identified (Fig. 3). Using NGS technology, chromothripsishas been previously reported in one case of CLL (15) and threecases of colorectal cancer (17), but this precise alignment of SVbreakpoints and copy number changes was not described. Themechanism of reconstitution of these fragments of DNA intoa complex, highly rearranged fragment of DNA as either an in-tact, cytogenetically recognizable chromosome or double minutehas recently been called chromoanagenesis (28) (“chromo,”meaning chromosome, and “anagenesis,” meaning reborn). Aspart of this mechanism, we provide unique sequencing informationthat small fragments (average size of 50–100 kb) of DNA fromchromosomes other than the cytogenetically recognizable chro-mosome are used to “stitch” such chromosomes together (Fig. 4).This conclusion is based on consistent PCR validation at pre-dicted interchromosomal breakpoints, and consistently negativefindings by FISH. Paired-spectrum orange and green break-apartFISH probes were designed on either side of multiple chromo-some 4, 5, and 6 CTX and INV breakpoints with a 50- to 100-kbgap (SI Appendix, Table S13) for each probe set. Because thesensitivity of FISH with interphase nuclei is in the range of 50 kb

of DNA and our probes were purposely designed to allow fora 50- to 100-kb error in prediction of the exact breakpoint, thiswould indicate that the segment of DNA involved in thisstitching process is less than this size.

NHEJ Is the Predominant Mechanism of Genomic Rearrangement inBladder Cancer. There is evidence from earlier studies that uniqueshort stretches of an identical sequence, or microhomology (29),located near the breakpoints of DNA double-strand breaks maybe critical in a stitching process in mouse Ltk− cells (30) similarto that reported here. This leads to the creation of localizedcomplex rearrangements. Similar evidence of the importance ofmicrohomology as a general mechanistic model for chromosomalrearrangements and amplification has been provided for humanlymphoma cell lines (31, 32). In these studies, a direct relation-ship to TP53 mutation status was noted. In three of our tumorswith the complex genotype, microhomology was identified in68% (101 of 148) of the breakpoints. Remarkably, this per-centage of microhomology was quite consistent among cases[case 18195, 34 (69%) of 49 breakpoints; case 18698, 15 (79%) of19 breakpoints; and case 9685, 52 (65%) of 80 breakpoints] andfor subtypes of SV (CTX, 64%; DEL, 75%; and INV, 67%). Theresults in these three tumors with an average of 2.2 bp of micro-homology per SV were similar to those of a recent study (33) of95 complete tumor genomes of various histological subtypes, inwhich an average of 1.7 bp of microhomology per SV was identi-fied. In our study, we also identified nontemplated sequences atthe rearrangement junctions in 18 of 148 SVs (SI Appendix, TableS14), which, along with microhomology, is considered to be thesignature of a DNA double-strand break repair process (34). Only20% of all SVs displayed neither microhomology nor nontemplatedsequences, indicating that NHEJ was the predominant DNA dou-ble-strand break repair process.

Recurrent Breakpoints Are Often Amplified in Bladder Cancer. Someof the CTXs and INVs in the group with a more complex genotypewere further defined by a clustering of breakpoints both withinand between different samples at chr4:180 Mb, chr5:29 Mb,chr5:40 Mb, chr6:10 Mb, chr6:18 Mb, and chr6:24 Mb. Thesebreakpoints were of interest as potential recurrent genomicevents in TCC-UB and were subsequently examined in a validation

Fig. 3. Utilization of SV viewer to demonstrateprecise alignment of breakpoints for CTXs and INVswith the change in segmental copy number states ina tumor in case 18195. (A) SV viewer with a com-plete genome view highlighting chromosome 4(chr4; blue zone) for a tumor in case 18195. CTXsand intrachromosomal translocations (ITXs) arerepresented as horizontal ticks for each breakpoint,with arcs representing the partner breakpoint. INVsare represented as solid yellow bars, with each endof the bar representing the two breakpoints. DELsare represented as solid green bars. Small redsquares represent tier 1 SNVs. (B) SV viewer high-lighting 69 Mb of chr4 from chr4:118,403,673–187,070,340 and illustrating six CTXs, two INVs, oneDEL, and seven SNVs. (C) Copy number is illustratedand shows that all six CTX and three of four INVbreakpoints precisely align with five of nine seg-mental copy number states.

Morrison et al. PNAS | Published online January 27, 2014 | E675

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdf

cohort of 343 patients. Using a FISH break-apart approach, we didnot identify translocations at any of these sites in the validationcohort; however, surprisingly, amplification (Fig. 4 B–D) was acommon event at chr5:40 Mb [32 (9%) of 343 patients], chr6:18Mb [53 (16%) of 332 patients], and chr6:24 Mb [53 (16%) of 332patients]. The gains at chr6:18 Mb and chr6:24 Mb were high-leveltandem amplifications typical of known oncogenes, such as HER-2,MYC, CCND1, or MDM2, whereas the gain at chr5:40 Mb wasone of low to intermediate amplification with a copy numberconsistently ranging from 5 to 10. FISH with additional BAC clonesperformed in the validation cohort across the chr18–24 Mb at 1- to2-Mb intervals (SI Appendix, Table S13) showed this was onecontinuous amplicon. The highest level of amplification and mini-mal region of copy number gain was seen at chr6:19.8 Mb (RP11-93O13) to chr6:21.4 Mb (RP11-204E9), including the genes E2F3and SOX4. Previous studies in multiple tumor types (35), includingbladder cancer (36), have identified increased copy number in a6p22 amplicon that is centered between the recurrent break-points with a segmental copy number change at chr6:18–24 Mb.Recent release of data by the Cancer Genome Atlas Network(www.cbioportal.org) shows this 6p22 amplicon containing theE2F3, SOX4, PRL, and CDKAL1 genes to be the most commonamplification in TCC-UB. The amplification at chr5:40 Mb wasnot studied further, but the overall evidence supports the ob-servation that the large number of SVs in the complex genotypeis not simply a reflection of random chromosomal instability.

SV Analysis Identifies a SCN8A-GRIN2A Translocation. As previouslydiscussed, the single SV in case 17802 was a DEL involving the p16gene at chromosome 9p21. The single SV identified in case 16933was unique in that it predicted an in-frame fusion protein involvingthe SCN8A gene at 12q13 and the GRIN2A gene at 16p13.2 (Fig.5). This fusion variant was predicted to result in an in-frame fusionof the SCN8A 5′ UTR and exon 1 with the GRIN2A completecoding sequence (CDS) and 3′ UTR (Fig. 5A). Subsequent ge-nomic PCR and capillary sequencing of this tumor using primersfor SCN8A 5′ UTR and GRIN2A exon 1 demonstrated the pre-dicted fusion variant in tumor DNA and not in the correspondinggerm line (SI Appendix, Figs. S12 and S13). Although RT-PCRwith a GRIN2A exon 2 and 4 primer set using case 16933 tumorcDNA demonstrated expression of CDS GRIN2A, the SCN8A

exon 1 andGRIN2A exon 4 primer set did not delineate this fusionvariant for unexplained reasons. Further analysis of this SV incase 16933, using our enhanced linear SV viewer (Fig. 5 B andC), showed an unexpected finding. For the SCN8A breakpointat 12q13, the corresponding data from Illumina SNP chips(HumanOmni1-Quad_v1-0 containing 1,140,419 dbSNP) showedcopy number gain centromeric to the breakpoint, corresponding tothe 5′UTR and exon 1 region of SCN8A, and a diploid state on thetelomeric side. For the GRIN2A breakpoint at 16p13.2, the cor-responding results showed copy number gain telomeric to thebreakpoint, corresponding to the CDS and 3′ UTR of GRIN2A,and a diploid state on the centromeric side. Translocation withsubsequent amplification of the involved genes is unique in cancergenetics and has been identified frequently only in the COL1A1and PDGFB translocations in dermatofibrosarcoma protuberans(37); to the best of our knowledge, this has not been previouslyreported in a carcinoma.

FISH Confirms an Amplified Reciprocal SCN8A-GRIN2A Translocation.To interrogate the chr12:52,049,200 breakpoint for SCN8A, wedesigned a break-apart FISH probe set (SI Appendix, Table S13),with a SpectrumOrange-labeled probe (RP11-923I11, orange)centromeric to the breakpoint representing the translocated 5′UTR and exon 1 of SCN8A and an FITC-labeled probe (RP11-285E4, green) telomeric to the breakpoint representing the non-translocated portion of this gene. Likewise, to interrogate thechromosome 16:10,035,762 breakpoint for GRIN2A, we designeda break-apart FISH probe set in the reverse fashion with an FITC-labeled probe (RP11-895K13, green) telomeric to the breakpointrepresenting the translocated CDS and 3′ UTR of GRIN2A anda SpectrumOrange-labeled probe (RP11-297M9, orange) centro-meric to the breakpoint representing the nontranslocated portionof this gene. Based on our sequencing results, we predicted evi-dence of orange-green break-apart at both sites with amplificationof the orange probe for SCN8A and amplification of the greenprobe for GRIN2A. Fig. 5D shows the FISH results using theSCN8A break-apart probe set for case 16933, and, as anticipated,multiple single orange signals with no associated green signalswere identified. Fig. 5E shows the FISH results using the GRIN2Abreak-apart probe set for case 16933 and, as anticipated, multiplegreen signals with no associated orange signals were identified.

B Chr5:40,062,440 Chr6:18,155,619-18,365,654C D Chr6:23,292,635-24,467,711

A

Fig. 4. Mechanism of creation of complex genomicrearrangement by NHEJ using chromosome-specificstitcher DNA fragments. (A) Chromosome shatteringfor the 6p amplicon is shown, resulting in sevendifferent megabase pairs in size fragments of DNAwith one of two segmental copy number states. NGSresults predicted a CTX or INV, or both, at each ofthese changes in segmental copy number state. FISHat each of these breakpoints was consistently PCR-positive but negative for rearrangement. In theprocess of rejoining these fragments, the resultingreformed chromosome can be linear or circular, maycontain inverted segments, and often shows ampli-fication at the breakpoints. (B–D) FISH shows am-plification but not translocation at chr5:40 Mb,chr6:18 Mb, and chr6:24 Mb breakpoints in the val-idation cohort. (Magnification: B–D, 1,000×.)

E676 | www.pnas.org/cgi/doi/10.1073/pnas.1313580111 Morrison et al.

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.cbioportal.orghttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1313580111

For both SCN8A and GRIN2A, the amplified signals consisted ofmicroclusters indicating high-level tandem duplication. To confirmour findings, we then designed a fusion FISH probe set using theamplified member of the two break-apart FISH probe sets, orSpectrumOrange-labeled RP11-923I11 (orange) for SCN8A rep-resenting the translocated 5′ UTR and exon 1 of this gene andFITC-labeled probe RP11-895K13 (green) for GRIN2A repre-senting the translocated CDS and 3′ UTR of this gene. Fig. 5Fshows the results of this fusion probe set, displaying multipleclustered green-orange fusion signals representing an amplifiedSCN8A-GRIN2A translocation. The configuration of these am-plified signals is most consistent with translocation and amplifi-cation within a ring chromosome.

Grin2A Is Often Amplified in Bladder Cancer. FISH validation in anadditional 333 tumors from patients with bladder cancer usinga GRIN2A break-apart probe (SI Appendix, Table S15) showedinfrequent GRIN2A translocation (two of 333 tumors), but,

surprisingly, identified a high-level tandem duplication of theGRIN2A gene in 8% (26 of 333) of tumors. Additional probe setsspanning the region telomeric (chr16:8,558,071–8,349,774) andcentromeric (chr16:11,180,357–11,439,054) to GRIN2A witha CEP16 probe showed the most frequent region of amplificationwas centered within the region containing the GRIN2A gene.Neither of the two additional samples in the validation cohortwith GRIN2A translocation showed SCN8A translocation usinga GRIN2A-SCN8A fusion probe set, and the translocation part-ner of these two GRIN2A translocation-positive samples was notdetermined due to the lack of a high-quality tumor sample.GRIN2A amplification was identified in none of the 41 TCC-UBs of the low-grade superficial type. Among high-grade su-perficial TCC-UBs, GRIN2A amplification was nearly as fre-quent [7 (8%) of 87] as in the muscle-invasive bladder cancercohort [19 (9%) of 205]. The lack of GRIN2A amplification inlow-grade vs. high-grade superficial or muscle-invasive TCC-UB

Copy number Chr12:51,049,200-53,049,200 CTX breakpointFISH probe

chr1

2SCN8A

RP11-895K13 RP11-297M9FISH probe

Chr

16GRIN2A

Chr

12GRIN2A

Chr

16SCN8A

A

F

E

D

C

SCN8A

GRIN2A

Copy number Chr16:9,035,762-11,035,762 CTX breakpoint

B

GRIN2A

SCN8A-GRIN2A gene fusion

Chr12:52,049,200 breakpoint

Chr16:10,036,019 breakpoint

SCN8A

SCN8A 5’UTR and exon 1 Complete GRIN2A CDS and 3’UTR

Translocation

Chr12:52,049,200

Chr16:10,036,019

RP11-285E4RP11-923I11

RP

11-2

97M

9R

P11

-895

K13

RP

11-9

23I1

1R

P11

-285

E4

RP

11-8

95K

13R

P11

-923

I11

Fig. 5. Details of the SCN8A-GRIN2A for the tumor in case 16933. (A, Upper) Illustration of the chr12:52,049,200 breakpoint between the 5′ UTR and exon 1of SCN8A. (A, Middle) Illustration in reverse orientation of the chr16:10,036,019 breakpoint between the 5′ UTR and exon 1 of GRIN2A. (A, Lower) Illustrationof the SCN8A-GRIN2A in-frame translocation using the SCN8A 5′ UTR and exon 1 and the GRIN2A CDS and 3′ UTR. (B) SV viewer highlighting the chr12breakpoint for SCN8A using a copy number profile of 1 Mb on either side of the breakpoint. Centromeric to the breakpoint copy number gain is identified,whereas telomeric to the breakpoint copy number is diploid. (C) SV viewer highlighting the chr16 breakpoint for GRIN2A using a copy number profile of 1 Mbon either side of the breakpoint. Centromeric to the breakpoint copy number gain is identified, whereas telomeric to the breakpoint copy number is diploid.(D) Break-apart FISH probe for the SNC8A gene shows amplification of the orange probe but not the green probe, consistent with the prediction by the SVviewer. (Magnification: 1,000×.) (E) Break-apart FISH probe for the GRIN2A gene shows amplification of the green probe but not the orange probe, consistentwith the prediction by the SV viewer. (Magnification: 1,000×.) (F) Fusion design FISH probe using SCN8A orange probe and GRIN2A green probe showsa green-orange fusion signal indicative of translocation and highly amplified for both partner genes. (Magnification: 1,000×.)

Morrison et al. PNAS | Published online January 27, 2014 | E677

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdf

implies GRIN2A amplification may be an early event in theprogression of bladder cancer to a lethal phenotype. Hence,GRIN2A amplification may be a driver “event” in bladder cancerand appears to be independent of the more common well-knownevents, such as loss of p16 and TP53 mutations. GRIN2A ampli-fication was more common in node-positive TCC-UB [11 (14%)of 81] compared with node-negative TCC-UB [12 (7%) of 166],further suggesting a role for GRIN2A amplification in the meta-static phenotype. Of interest, a higher frequency of GRIN2A am-plification was noted in patients with evidence of cancer at lastfollow-up [12 (9%) of 127]) vs. those with no evidence of cancer [10(5%) of 183]. However, there was no apparent association betweenGRIN2A amplification and survival (P = 0.24) (SI Appendix, TableS15). To evaluate the biological mechanisms and significance ofGRIN2A amplification further, we compared the mRNA level ofGRIN2A for 20 GRIN2A-amplified patient samples and 20 non-amplified patient samples (SI Appendix, Fig. S14). Tumors withGRIN2A amplification showed 12-fold increased expression ofGRIN2A mRNA (P = 0.005), supporting the hypothesis thatoverexpression of GRIN2A occurs by gene amplification.

GRIN2A as a Potential Oncogene in Bladder Cancer. GRIN2A enc-odes the e-1 subunit of the NMDA receptor, which has beenreported to confer growth advantage to glioma implants and isassociated with glutamate release (38). Therefore, we hypothe-sized that GRIN2A expression may also contribute to a growthadvantage in TCC cells. To test this hypothesis, we examined themRNA expression of GRIN2A in a collection of 17 human TCCcell lines (SI Appendix, Fig. S15). From this list, we chose twohigh Grin2A mRNA-expressing cell lines, 253J and HT-1376,and developed a shRNA lentiviral construct specifically to targetthe expression of GRIN2A in these TCC cell lines. Knockdownof GRIN2A was successfully achieved, and reduced expression ofGRIN2A decreased cell proliferation of both the 253J and HT-1376cell lines (Fig. 6A). Using the HT-1376 tumor model in mice, weimplanted s.c. HT-1376/shGRIN2A and HT-1376/shGFP constructsinto the right and left flanks of SCID mice (6–8 wk of age, five miceper group). As shown in Fig. 6B, we observed a reduction of HT-1376/shGRIN2A tumor growth in mice compared with HT-1376/shGFP tumors (P < 0.01), where expression of GRIN2A was de-creased in HT-1376/shGRIN2A tumor cells compared with HT-1376/shGFP tumor cells at the time of injection. Similarly, asa marker of proliferation, Ki-67 was decreased in tumor sectionstaken from HT-1376/shGRIN2A tumor cells compared with con-trols (Fig. 6B). These results indicate that that the silencing ofGRIN2A inhibits proliferation in vitro and in vivo in a bladdertumor cell line model. We also evaluated the oncogenic effect ofGRIN2A using a SV40 immortalized human urothelial cell line(SV-HUC), which is an SV40 immortalized, nontransformed, hu-man uroepithelial cell line (39). GRIN2A was overexpressed in theSV-HUCs as shown by mRNA and protein levels (Fig. 6C), andconsistent with the data in Figs. 6 A and B, overexpression ofGRIN2A increased the proliferation and migration of SV-HUCs,suggesting that overexpression of GRIN2A promotes an increase incell proliferation and migration of bladder epithelial cells.

DiscussionThese data reveal a spectrum of heterogeneity among sequencedbladder tumors. Based on our whole-genome sequencing analysis,we show evidence that SV breakpoints can have a unique associa-tion with copy number in the context of chromothripsis, possiblyrelated to a process of genomic amplification. Additionally, wedemonstrate that complex genomic rearrangements mechanisticallyuse kilobase fragments of DNA that we call stitchers as part of anNHEJ DNA repair process. Furthermore, our results support thepresence of intratumoral mutational heterogeneity in TCC-UB.Finally, although not related to smoking in our study, we provideevidence of a subset of tumors similar to lung non–small-cell

carcinoma in those who have never smoked (40), with a few mu-tations that are likely driven by one or a few driver mutations.Chromothripsis appears to be a relatively common event in

TCC-UB, but its role as a “passenger” or “driver” in bladdercancer progression is not yet determined. A recent study in leu-kemia involving a patient with multiple recurrences provides anexample of chromothripsis as a passenger event (41). In that study,chromothripsis was identified in a specimen from the time of re-currence, presumably as part of tumor progression, but it was notpresent in additional relapses after subsequent intervening che-motherapy. It is possible that only a subset of rearrangements inchromothripsis confers a selective single-cell advantage, much likesubclonal populations of mutations that are selected throughtherapeutic interventions (21). Analyses of multiple samples ofTCC-UB from one patient, preferably primary and metastatictumors with some period of months to years between the twoevents, will be required to decipher this potential mechanism. Ifthis process is merely a passenger event, it is more difficult toexplain how evidence in both medulloblastoma (14) and mela-noma (13) would suggest that chromothripsis is associated witha more aggressive clinical course. We provide some evidence thatcomplex localized genomic rearrangements may result in somecompetitive advantage for neoplastic cells via gene amplification inat least one tumor (case 18195), where breakpoints precisely linedup with the well-known 6p22 amplicon containing the E2F3 andSox4 genes (42).Another intriguing question is whether chromothripsis evolves

through the same mechanism in different tumor types or is uniqueto TCC-UB in this regard. In our study, we showed that NHEJ isthe predominant mechanism, whereas in prostate cancer, TelesAlves et al. (27) showed no evidence of microhomology involvingchromosome 5 for the vertibral cancer prostate cell line. This invitro finding contrasts to the finding of Drier et al. (33) in 95matched tumor/normal samples that included 46 breast carcinomasamples, 23 multiple myeloma samples, 9 colorectal carcinomasamples, 7 prostate adenocarcinoma samples, 5 melanoma sam-ples, 3 CLL samples, and 2 head and neck carcinoma samples. Inthese cohorts, chromothripsis was associated with all cancer typesexcept CLL. This group gave additional evidence that chromo-thripsis is associated with replication time, proximity to transcribedgenes, and guanine-cytosine content. The association of chromo-thripsis with microhomology, replication time, and proximity totranscribed genes could result from deficiencies in the replication–licensing complex that is loaded onto chromatin during the G1-phase of the cell cycle and is required for initiation of DNA rep-lication in the subsequent S-phase. In two of our tumors withMCM4 mutations, this association could be defined mechanisticallyby genomic alteration of the family of MCM2-7 genes (43). Con-sistent with this hypothesis, prior studies have shown that Mcmprotein deficiencies result in high rates of cancer in mouse models(44, 45), catastrophic chromosomal rearrangements in humanlymphoblasts in culture (46), and complex chromosomal alterationsat discrete locations that are consistent with chromothripsis (43).Previously, we have suggested that the frequent, short intra-

chromosomal DELs spanning 500 kbp or less that occur inmouse tumors resulting from deficient Mcm2 protein levelscould result from failure to rescue stalled replication forks withinindividual replication factories (43). This mechanism may alsohelp to explain the frequent localization of multiple trans-location events seen in individual tumors in the present study.For example, Fig. 7 shows the location of all structural alter-ations occurring within a single tumor that additionally harborsa nonsynonymous point mutation within the Mcm N domain ofMCM4. Multiple translocation events occur between approxi-mately five and six sites on each of chromosomes 4, 5, and 6,where the size of sites involved is ∼500 kbp or less. To accountfor the number of DNA replication forks generated during theS-phase, a single replication factory must contain 20–200 DNA

E678 | www.pnas.org/cgi/doi/10.1073/pnas.1313580111 Morrison et al.

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1313580111

replication forks (47), which, assuming ∼50,000 bp betweenreplication origins, span 0.5–5 Mb of DNA. Although it is typi-cally assumed that replication factories assemble around domainswithin an individual chromosome, it is also possible that a singlefactory contains origins from different chromosomes (48), asshown in Fig. 7B. In this case, failure to rescue stalled replicationforks could lead to the observed complex recombination eventsinvolving multiple chromosomes. It is interesting to note that theother case in our study with the highest number of both SVs andSNVs (case 19685) also had an MCM4 mutation in the samedomain as the tumor in case 18195.Nonhomologous reciprocal translocations have been identified

in lymphomas and sarcomas; however, these complex rearrange-ments, at least in TCC-UB, are different from the ones seen inlymphomas and sarcomas. Although this is not surprising, giventhe definition of chromothripsis and current knowledge of thisgenomic event, the evidence we present of a stitching processusing 50- to 100-kb fragments to reconstitute these interchro-mosomal events is intriguing and raises questions about currenttheories regarding this process. The prior concept that chromo-thripsis results in exchange of megabase fragments of DNA from

two or more chromosomes to form a highly complex derivativechromosome may be incomplete. Our study suggests that althoughone chromosome provides megabase fragments of DNA, other in-volved chromosomes provide only 50- to 100-kb fragments that werefer to as stitchers. Although our findings do not fully define thismechanism at the current time, a plausible explanation could be thattumors use stitchers in stalled replication forks in the replication–licensing complex during the G1-phase of the cell cycle (43). Ourfindings provide a framework for further mechanistic investigations.Although chromothripsis and the complex process underlying

this event may not lead to a driver mutation, intratumoral muta-tional heterogeneity leading to driver events is likely, at least inleukemia (19, 21). In a comparative sense, it could be possible thatintratumoral mutational heterogeneity is a marker of underlyinggenomic events, much as we postulate that chromothripsis is relatedto a defective replication–licensing complex. Intratumoral muta-tional heterogeneity may be associated with resistance to chemo-therapy and/or advanced stage at the time of presentation (21). Inour study, both cases that showed evidence of intratumoral muta-tional heterogeneity were also TP53mut, whereas among the threecases with no evidence of intratumoral mutational heterogeneity,

0 0.2 0.4 0.6 0.8

1 1.2

shGFP shGRIN2A

Rel

ativ

e Ex

pres

sion

0 0.2 0.4 0.6 0.8

1 1.2

shGFP shGRIN2A

Rel

ativ

e Ex

pres

sion

0 1 2 3 4 5 6 7 8

1 2 3 4 5

Prol

ifera

tion

Inde

x

Time (Days)

shGFP shGRIN2A

0 1 2 3 4 5 6 7

1 2 3 4 5 6

Prol

ifera

tion

Inde

x

Time (Days)

shGFP shGRIN2A

A C

B

HT-1376

SV-HUC HT-1376 253J

0 500

1000 1500 2000 2500 3000

vecto

r

GRIN

2A

Rel

ativ

e Ex

pres

sion

0

1

2

3

4

1 2 3 4 5 6 7

Prol

ifera

tion

Inde

x

Time (days)

Vector Grin2A

0

10

20

30

40

50

60

70

vector Grin2A

Cel

l Mig

ratio

n pe

r Fie

ld

0

0.2

0.4

0.6

0.8

1

1.2

shGF

P

shGR

IN2A

Rel

ativ

e Ex

pres

sion

0

400

800

1200

4 8 12 16 20 24

Tum

or V

olum

e (m

m3)

Day Post Implantation

shGFP shGRIN2A

*

***

**

***

*****

*

Ki-67 shGFP shGRIN2A

Fig. 6. In vitro and in vivo models using a shRNA lentiviral construct specifically to target the expression of GRIN2A in the 253J and HT-1376 TCC lines. (A)In-vitro model using a shRNA lentiviral construct targeting GRIN2A in the 253J and HT-1376 bladder cancer cell lines by real-time RT-PCR and the3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. (B) In vivo model using the HT-1376 cell line with HT-1376/shGRIN2A andHT-1376/shGFP constructs that were validated for GRIN2A expression before injection s.c. into the right and left flanks of SCID mice (6–8 wk of age, fivemice per group), with Ki-67 staining of these tumors examined on day 24. (Magnification: 20×; magnification of Insets, 40×.) (C ) Nontransformed bladderepithelial SV-HUCs were transfected with a shRNA lentiviral construct that targets GRIN2A, and expression was determined in RT-PCR and immunoblotassays. SV-HUCs that overexpressed GRIN2A were examined for a change in proliferation and migration using the MTT assay and transwell migrationassay, respectively.

Morrison et al. PNAS | Published online January 27, 2014 | E679

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

5, 2

021

one was TP53mut and the other two were TP53wt. Early clonal ex-pansion of TP53-mutated cells would be predicted to lead to in-creased genetic heterogeneity through lack of sufficient DNA repairprocesses. It is feasible that the final evaluation of this topiccan be done with exomic sequencing and will not require se-quencing of the entire genome. Progress in developing moretargeted therapies in TCC-UB will be informed by furtheranalysis of this genomic event.In the more “genomically simple” subset of TCC-UB, we iden-

tified unique events that included (i) an unequivocal demonstrationthat amplified CTXs can be found in carcinomas and (ii) infrequenttranslocation but frequent amplification of GRIN2A in a subset ofTCC-UB. Furthermore, our preliminary functional studies inbladder carcinoma cell lines support GRIN2A as a candidate on-cogene in TCC-UB. Although the first of these unique findings wasonly demonstrated in the index case (case 16933), it does demon-strate the utility of NGS as a discovery tool and likely portends thediscovery of additional such examples in bladder cancer and othercarcinomas as this technology expands. In our study, this initialdiscovery of an SCN8A-GRIN2A translocation was further noted inthe validation cohort. However, this observation led us to the dis-covery of a second unique finding, amplification of GRIN2A, ina subset of TCC-UB. Previous findings of GRIN2A as a frequently

mutated gene in melanoma (49) and a frequently overexpressedgene in ALK-positive lung cancer (50), as well as the recentrecognition that glutamate transport and intermediary metab-olism may be important in the etiology of other tumors (glio-blastoma) (38), provide convincing evidence that GRIN2A is ofimportance in cancer.To conclude, we have provided additional insight into the

genomic landscape of muscle-invasive bladder cancer and de-veloped a framework for future whole-genome sequencingstudies of TCC-UB to use as a comparison. We have shown a greatdeal of genomic diversity in a small sample set of TCC-UB that willprovide important information in planning for additional studies.

MethodsSamples and Clinical Data.We studied five tumor samples (cases 16933, 17802,18195, 18698, and 19685) of chemotherapy-naive muscle-invasive TCC-UBAmerican Joint Committee on Cancer stage III or IV with whole-genomesequencing (SI Appendix, Table S1). There were three males and two females(all Caucasian non-Hispanic), with an average age of 67 y. Three weresmokers, two were nonsmokers, and none had a prior history of superficialTCC-UB. We identified tumor-specific somatic DNA alterations by comparingeach tumor with its corresponding normal germ-line DNA derived frommatching blood. The validation cohort consisted of 333 patients with a his-tory of TCC-UB of the bladder that spanned the gamut of clinical scenariosranging from low-grade superficial bladder carcinoma to high-grade in-vasive and noninvasive TCC-UB with and without a prior history of superfi-cial disease (SI Appendix, SI Methods).

DNA Library Preparation and Massively Parallel Sequencing. Whole-genomesequencing was done using a 500-bp library with 100-bp paired-end readsand, additionally, a 5-kb library with 36-bp paired-end reads (mate pair).

Sequencing was carried out for the prepared DNA libraries with a HiSeq2000 sequencing system (Illumina) following the manufacturer’s standardprotocol using the Illumina cBot and HiSeq paired-end cluster kit, version 1(SI Appendix, SI Methods).

Read Mapping and Alignment and Variant Analysis. We recently developed anin-house analysis pipeline for cancer genome sequencing data that includes(i) mapping and alignment, (ii) SNV and indel discovery, and (iii) SNV andindel filtering and annotation (SI Appendix, SI Methods).

Detecting SVs. BreakDancer was used to detect SVs from paired-end Illuminasequencing data. Then, the de novo assemblywas performed for all filteredDELs,insertions, and INVs using the newly developed sensitive assembler TIGRA_SV(http://genome.wustl.edu/software/tigra_sv) and for translocations using Phrap(www.phrap.org/), followed by extraction of mapped reads using SAMtools(http://samtools.sourceforge.net/) (SI Appendix, SI Methods).

PCR Validation of SVs. Putative SVs were validated by PCR using R script toselect genomic sequences around the de novo assembly-determined break-points for each SV from the University of California, Santa Cruz genomebrowser (http://genome.ucsc.edu/) (SI Appendix, SI Methods).

Detection of Somatic Copy Number Alteration. To identify somatic copynumber alterations, each tumor and its matched normal DNA were geno-typed using IlluminaHumanOmni1-Quad BeadChips, which contain 1,140,419SNPs, with a median SNP spacing of 1.2 kb (SI Appendix, SI Methods).

FISH for SVs. All SVs in this study were evaluated by FISH using RP11 clonesfrom the Roswell Park Cancer Institute BAC library. A complete list of all BACclones and probe designs is provided in SI Appendix, Table S13. Bothbreakpoints of a given SV were evaluated separately using a break-apartprobe FISH design (SI Appendix, SI Methods).

RT-PCR Analysis for GRIN2A mRNA Expression. Real-time RT-PCR analysis wasdone using SYBR Green I as a reporter and ROX (Applied Biosystems) asa reference dye for GRIN2A mRNA expression (SI Appendix, SI Methods).

Statistical Analysis. The association between clinical/histological covariatesand 6p22 amplification was tested using two-sample t tests for the equalityof proportions. Survival time associations were tested with a log-rank test.Statistical analysis of data was performed using the SPSS Statistics softwarepackage (IBM). All results are expressed as mean ± SD.

Fig. 7. Model of microhomology-mediated translocation events occurringbetween two or more chromosomes in an individual replication factory. (A)Illustration of multiple translocation events occurring between chr4, chr5,and chr6 for case 18195, with each line denoting an individual CTX. Notethat many of the translocations between any two chromosomes often showa second breakpoint within a few thousand base pairs as part of a differenttranslocation, with the third chromosome resulting in this complex web-likepattern of rearrangement. (B) DNA replication factory involving portions ofchr4, chr5, and chr6, with stalled replication forks indicated by red x marks.Dashed lines indicate translocation occurring at stalled replication forks,often with closely adjacent breakpoints involving multiple chromosomes ina complicated web-like fashion.

E680 | www.pnas.org/cgi/doi/10.1073/pnas.1313580111 Morrison et al.

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://genome.wustl.edu/software/tigra_svhttp://www.phrap.org/http://samtools.sourceforge.net/http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://genome.ucsc.edu/http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1313580111

In Vitro Tumor Assays. Human bladder cancer cell lines 253J andHT-1376, as wellas SV40 immortalized human uroepithelial SV-HUCs, were cultured, and trans-fectionwasperformedusingX-tremeGENE9DNATransfectionReagent followingthe manufacturer’s protocol (Roche). Packaging of retrovirus and lentivirus,cell transduction, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromideassays, Western blotting, shRNA knockdown experiments, and migration assayswere performed following standard protocols (SI Appendix, SI Methods).

ACKNOWLEDGMENTS. Biospecimens or research pathology services forthis study were provided by the Pathology Resource Network, which isfunded by the National Cancer Institute and is a Roswell Park CancerInstitute Cancer Center Support Grant shared resource. Clinical datadelivery and honest broker services for this study were provided by theClinical Data Network, which is funded by the National Cancer Instituteand is a Roswell Park Cancer Institute Cancer Center Support Grantshared resource.

1. Siegel R, Naishadham D, Jemal A (2013) Cancer statistics, 2013. CA Cancer J Clin63(1):11–30.

2. Carter AJ, Nguyen CN (2012) A comparison of cancer burden and research spendingreveals discrepancies in the distribution of research funding. BMC Public Health 12:526.

3. Wesche J, Haglund K, Haugsten EM (2011) Fibroblast growth factors and their re-ceptors in cancer. Biochem J 437(2):199–213.

4. Beenken A, Mohammadi M (2009) The FGF family: Biology, pathophysiology andtherapy. Nat Rev Drug Discov 8(3):235–253.

5. Billerey C, et al. (2001) Frequent FGFR3 mutations in papillary non-invasive bladder(pTa) tumors. Am J Pathol 158(6):1955–1959.

6. Tomlinson DC, Lamont FR, Shnyder SD, Knowles MA (2009) Fibroblast growth factorreceptor 1 promotes proliferation and survival via activation of the mitogen-activatedprotein kinase pathway in bladder cancer. Cancer Res 69(11):4613–4620.

7. Erill N, et al. (2004) Genetic and immunophenotype analyses of TP53 in bladdercancer: TP53 alterations are associated with tumor progression. Diagn Mol Pathol13(4):217–223.

8. Miyamoto H, Shuin T, Torigoe S, Iwasaki Y, Kubota Y (1995) Retinoblastoma genemutations in primary human bladder cancer. Br J Cancer 71(4):831–835.

9. Miyao N, et al. (1993) Role of chromosome 9 in human bladder cancer. Cancer Res53(17):4066–4070.

10. Williams SV, et al. (2002) Molecular genetic analysis of chromosome 9 candidate tu-mor-suppressor loci in bladder cancer cell lines. Genes Chromosomes Cancer 34(1):86–96.

11. Sjödahl G, et al. (2012) A molecular taxonomy for urothelial carcinoma. Clin CancerRes 18(12):3377–3386.

12. Hurst CD, Platt FM, Taylor CF, Knowles MA (2012) Novel tumor subgroups of ur-othelial carcinoma of the bladder defined by integrated genomic analysis. Clin CancerRes 18(21):5865–5877.

13. Hirsch D, et al. (2013) Chromothripsis and focal copy number alterations determinepoor outcome in malignant melanoma. Cancer Res 73(5):1454–1460.

14. Rausch T, et al. (2012) Genome sequencing of pediatric medulloblastoma links cata-strophic DNA rearrangements with TP53 mutations. Cell 148(1-2):59–71.

15. Stephens PJ, et al. (2011) Massive genomic rearrangement acquired in a single cata-strophic event during cancer development. Cell 144(1):27–40.

16. Kim TM, et al. (2013) Functional genomic analysis of chromosomal aberrations ina compendium of 8000 cancer genomes. Genome Res 23(2):217–227.

17. Kloosterman WP, et al. (2011) Chromothripsis is a common mechanism driving ge-nomic rearrangements in primary and metastatic colorectal cancer. Genome Biol12(10):R103.

18. Navin N, et al. (2011) Tumour evolution inferred by single-cell sequencing. Nature472(7341):90–94.

19. Ding L, et al. (2012) Clonal evolution in relapsed acute myeloid leukaemia revealed bywhole-genome sequencing. Nature 481(7382):506–510.

20. Shah SP, et al. (2012) The clonal and mutational evolution spectrum of primary triple-negative breast cancers. Nature 486(7403):395–399.

21. Landau DA, et al. (2013) Evolution and impact of subclonal mutations in chroniclymphocytic leukemia. Cell 152(4):714–726.

22. Gui Y, et al. (2011) Frequent mutations of chromatin remodeling genes in transitionalcell carcinoma of the bladder. Nat Genet 43(9):875–878.

23. Carter SL, et al. (2012) Absolute quantification of somatic DNA alterations in humancancer. Nat Biotechnol 30(5):413–421.

24. Gerlinger M, et al. (2012) Intratumor heterogeneity and branched evolution revealedby multiregion sequencing. N Engl J Med 366(10):883–892.

25. Nik-Zainal S, et al.; Breast Cancer Working Group of the International Cancer GenomeConsortium (2012) The life history of 21 breast cancers. Cell 149(5):994–1007.

26. Welch JS, et al. (2012) The origin and evolution of mutations in acute myeloid leu-kemia. Cell 150(2):264–278.

27. Teles Alves I, et al. (2013) Gene fusions by chromothripsis of chromosome 5q in theVCaP prostate cancer cell line. Hum Genet 132(6):709–713.

28. Holland AJ, Cleveland DW (2012) Chromoanagenesis and cancer: Mechanisms andconsequences of localized, complex chromosomal rearrangements. Nat Med 18(11):1630–1638.

29. Kidd JM, et al. (2010) A human genome structural variation sequencing resourcereveals insights into mutational mechanisms. Cell 143(5):837–847.

30. Lin Y, Waldman AS (2001) Promiscuous patching of broken chromosomes in mam-malian cells with extrachromosomal DNA. Nucleic Acids Res 29(19):3975–3981.

31. Zhu C, et al. (2002) Unrepaired DNA breaks in p53-deficient cells lead to oncogenicgene amplification subsequent to translocations. Cell 109(7):811–821.

32. Difilippantonio MJ, et al. (2002) Evidence for replicative repair of DNA double-strandbreaks leading to oncogenic translocation and gene amplification. J Exp Med 196(4):469–480.

33. Drier Y, et al. (2013) Somatic rearrangements across cancer reveal classes of sampleswith distinct patterns of DNA breakage and rearrangement-induced hypermutability.Genome Res 23(2):228–235.

34. Hefferin ML, Tomkinson AE (2005) Mechanism of DNA double-strand break repair bynon-homologous end joining. DNA Repair (Amst) 4(6):639–648.

35. Santos GC, Zielenska M, Prasad M, Squire JA (2007) Chromosome 6p amplification andcancer progression. J Clin Pathol 60(1):1–7.

36. Bruch J, et al. (2000) Delineation of the 6p22 amplification unit in urinary bladdercarcinoma cell lines. Cancer Res 60(16):4526–4530.

37. Llombart B, Serra-Guillén C, Monteagudo C, López Guerrero JA, Sanmartín O (2013)Dermatofibrosarcoma protuberans: A comprehensive review and update on di-agnosis and management. Semin Diagn Pathol 30(1):13–28.

38. Takano T, et al. (2001) Glutamate release promotes growth of malignant gliomas. NatMed 7(9):1010–1015.

39. Christian BJ, Loretz LJ, Oberley TD, Reznikoff CA (1987) Characterization of humanuroepithelial cells immortalized in vitro by simian virus 40. Cancer Res 47(22):6066–6073.

40. Kim SC, et al. (2013) A high-dimensional, deep-sequencing study of lung adenocar-cinoma in female never-smokers. PLoS ONE 8(2):e55596.

41. Bassaganyas L, et al. (2013) Sporadic and reversible chromothripsis in chronic lym-phocytic leukemia revealed by longitudinal genomic analysis. Leukemia 27(12):2376–2379, 10.1038/leu.2013.127.

42. Oeggerli M, et al. (2004) E2F3 amplification and overexpression is associated withinvasive tumor growth and rapid tumor cell proliferation in urinary bladder cancer.Oncogene 23(33):5616–5623.

43. Rusiniak ME, Kunnev D, Freeland A, Cady GK, Pruitt SC (2012) Mcm2 deficiency resultsin short deletions allowing high resolution identification of genes contributing tolymphoblastic lymphoma. Oncogene 31(36):4034–4044.

44. Shima N, et al. (2007) A viable allele of Mcm4 causes chromosome instability andmammary adenocarcinomas in mice. Nat Genet 39(1):93–98.

45. Pruitt SC, Bailey KJ, Freeland A (2007) Reduced Mcm2 expression results in severestem/progenitor cell deficiency and cancer. Stem Cells 25(12):3121–3132.

46. Orr SJ, et al. (2010) Reducing MCM levels in human primary T cells during the G(0)→G(1)transition causes genomic instability during the first cell cycle. Oncogene 29(26):3803–3814.

47. Berezney R, Dubey DD, Huberman JA (2000) Heterogeneity of eukaryotic replicons,replicon clusters, and replication foci. Chromosoma 108(8):471–484.

48. Meister P, Taddei A, Gasser SM (2006) In and out of the replication factory. Cell 125(7):1233–1235.

49. Wei X, et al.; NISC Comparative Sequencing Program (2011) Exome sequencingidentifies GRIN2A as frequently mutated in melanoma. Nat Genet 43(5):442–446.

50. Okayama H, et al. (2012) Identification of genes upregulated in ALK-positive andEGFR/KRAS/ALK-negative lung adenocarcinomas. Cancer Res 72(1):100–111.

Morrison et al. PNAS | Published online January 27, 2014 | E681

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

5, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1313580111/-/DCSupplemental/sapp.pdf