Paired Exome Analysis Reveals Clonal Evolution and ......Philippe Lamy1, Iver Nordentoft1, Karin Birkenkamp-Demtroder€ 1, Mathilde Borg Houlberg Thomsen1, Palle Villesen2, Søren

Tumor and Stem Cell Biology

Paired Exome Analysis Reveals Clonal Evolutionand Potential Therapeutic Targets in UrothelialCarcinomaPhilippe Lamy1, Iver Nordentoft1, Karin Birkenkamp-Demtr€oder1,Mathilde Borg Houlberg Thomsen1, Palle Villesen2, Søren Vang1,Jakob Hedegaard1, Michael Borre3, Jørgen Bjerggaard Jensen3,Søren Høyer4, Jakob Skou Pedersen1, Torben F. �rntoft1, and Lars Dyrskjøt1

Abstract

Greater knowledge concerning tumor heterogeneity and clon-ality is needed todetermine the impact of targeted treatment in thesetting of bladder cancer. In this study, we performed whole-exome, transcriptome, and deep-focused sequencing of meta-chronous tumors from 29 patients initially diagnosed with ear-ly-stage bladder tumors (14 with nonprogressive disease and 15with progressive disease). Tumors from patients with progressivedisease showed a higher variance of the intrapatient mutationalspectrum and a higher frequency of APOBEC-related mutations.

Allele-specific expression was also higher in these patients, par-ticularly in tumor suppressor genes. Phylogenetic analysisrevealed a common origin of the metachronous tumors, with ahigher proportion of clonal mutations in the ancestral branch;however, 19 potential therapeutic targets were identified as bothancestral and tumor-specific alterations. Few subclones werepresent based on PyClone analysis. Our results illuminate tumorevolution and identify candidate therapeutic targets in bladdercancer. Cancer Res; 76(19); 5894–906. �2016 AACR.

IntroductionBladder cancer is a major cause of mortality worldwide (1).

More than 90% of bladder tumors are classified as urothelialcarcinoma, and around 25% of patients are initially diagnosedwithmuscle-invasive bladder cancer (MIBC). The remaining 75%are diagnosedwith non-muscle invasive bladder cancer (NMIBC),with high recurrence rates and moderate progression rates(�15%) to MIBC (1, 2). Identifying clinically informative altera-tions in bladder cancer may help stratifying patients to follow-upregimens, chemotherapy, and targeted treatment, which mayultimately increase survival. However, knowledge on tumor het-erogeneity and clonality of potential therapeutic targets is neededto administer optimal treatment.

Bladder cancer has a heterogeneous clinical presentation, andrecurrent tumors are often observed at shifting locations in thebladder. Furthermore, patients withmultiple tumors are at higher

risk of recurrence, progression, and cancer-related death (1).Multifocal tumors are hypothesized to originate from a fieldcancerization of the bladder, where tumors develop from pre-neoplastic fields of transformed urothelial cells (3). The fieldcancerization model is supported by frequently observed geneticinstability and mutations in normal-appearing urothelium (4)and dysplasia or carcinoma in situ (CIS) in the surroundingurothelium (5). In addition, analysis of gene expression in tumorsand adjacent normal-appearing urothelium has shown expres-sion patterns in the normal urothelium similar to those observedin the tumor lesions (5).

The recent identification ofmutational signatures characteristicfor each cancer type has provided insight into the mutationalheterogeneity and mutational mechanisms as well as repairdeficiencies that lead to a given cancer (6). The continued acqui-sition of mutations may explain the shift from benign to malig-nant tumors over time and could in theory guide preventivemeasures. The mutational landscape in NMIBC has recently beendescribed (7–9). Furthermore, the APOBEC enzyme family thatcreates specificmutational signatures in single-stranded DNAwasshown to be prevalent in bladder cancer, and it may fuel theoncogenic transformation (10, 11). Finally, deep sequencing hasrecently been used to identify the clonal composition of driveralterations in several cancers (7, 12, 13).

For optimal targeted treatment, it is important to define theancestral mutations, as these may be ideal therapeutic targets(present in all cancer cells). Here, we defined the mutationalevents occurring over time in multiple paired bladder tumors toidentify phylogenetic relations between metachronous tumors. Acommon origin was found for all tumor pairs, but driver muta-tions in known tumor suppressors and oncogenes were notpredominantly found to be early events. Potential therapeutic

1Department of Molecular Medicine, Aarhus University Hospital, Aar-hus, Denmark. 2Department of Bioinformatic Research, Aarhus Uni-versity, Aarhus, Denmark. 3Department of Urology, Aarhus UniversityHospital, Aarhus, Denmark. 4Department of Pathology, Aarhus Uni-versity Hospital, Aarhus, Denmark.

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

P. Lamy and I. Nordentoft contributed equally to this article.

Corresponding Author: Lars Dyrskjøt, Department of Molecular Medicine,Aarhus University Hospital, Palle Juul-Jensens Boulevard 99, 8200 Aarhus N8200, Denmark. Phone: 45-78455320; Fax: 45-86782108; E-mail:[email protected]

doi: 10.1158/0008-5472.CAN-16-0436

�2016 American Association for Cancer Research.

CancerResearch

Cancer Res; 76(19) October 1, 20165894

on September 3, 2021. © 2016 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

Published OnlineFirst August 3, 2016; DOI: 10.1158/0008-5472.CAN-16-0436

http://cancerres.aacrjournals.org/

targets were identified as both ancestral (shared) and tumor-specific alterations. We found a higher intrapatient variation ofthe tumor mutation spectrum and frequent APOBEC-relatedmutation signatures in patients with progressive disease (PD).In addition, we found high temporal and low spatial mutationalheterogeneity, suggesting that targeted treatment decisionsshould ideally be based on analysis of biopsies from severaltumors.

Materials and MethodsSample preparation

This study includes data from a previous study in whichmetachronous tumors from 4 patients were analyzed togetherwith single tumors from 20 patients (8). In total, 75 UCC samplesfrom 38 patients were obtained fresh from surgery, embedded inTissue-Tek, and stored at �80�C. Two or more metachronoustumors were included for 29 patients. Inclusion criteria aredefined in the Supplementary Data. In addition, for one patient(p39), DNA was extracted from eight laser microdissected sub-regions froma single tumor. LeucocyteDNA from initial visitswassequenced as reference. Genomic DNA and total RNA wereextracted from serial cryosections. All patients gave their writteninformed consent, and the study was approved by the NationalCommittees on Health Research Ethics (#1300174).

NGS library construction and sequencingLibraries for whole-exome sequencing (WES) were made using

the Illumina TruSeq DNA Kit and NimbleGen SeqCap EZ v3.0.Whole-transcriptome RNA sequencing (RNA-seq) was preparedfrom rRNA-depleted total RNA using the Ribo-Zero and Script-Seq (Epicentre). WES for patient p39 was made using theIllumina Nextera Rapid Exome Capture Expanded Kit. Librariesfor ultradeep targeted resequencing were constructed using theIllumina TruSeq Custom Amplicon v1.5 Kit and contained 1,530amplicons. Amplicons were designed using DesignStudio soft-ware v1.5 (Illumina). Sequencing was performed on the Illu-mina HiSeq2000 and NextSeq 500 platforms.

Sequence processingDNA samples were mapped to hg19, marked for duplicates,

and the alignments were recalibrated and realigned using thepicard (http://picard.sourceforge.net/) and GATK suites (14).SNPs were called using GATK HaplotypeCaller and somaticmutations (SNV) and indels were called using MuTect (15) andthe Somatic Indel Detector from GATK. Indels shared bymultiple patients were excluded. Functional annotation ofmutations were performed using SnpEff (16), and SNVs wereclassified in two categories reflecting the quality of the calls (Cat1, high quality; Cat 2, low quality) and four tiers reflecting theimpact of the call (Tier 0, nonsense or splice site mutations; Tier1, missense mutations; Tier 2, silent mutations; Tier 3, non-coding mutations). RNA-seq data were mapped and quantifiedusing the Tuxedo Suite (17).

Mutational signaturesFor all samples, the sequence context (30 and 50 nucleotides)

was extracted for all somatic mutations (Cat 1–2 Tier 0–3),leading to 96 classes of mutations. Each sample mutationalprofile was projected onto 22 known signatures (18), and onlysignatures contributing to more than 7% of the mutations were

retained. This was done iteratively until no more signatures wereremoved. The six most frequent signatures were used, and theircontributions were recalculated.

Allele-specific expressionWe analyzed DNA and RNA sequencing data combined to

investigate allele-specific expression (ASE) in the tumor samples.ASE was measured as the absolute difference in alternative allelefrequency between RNA and DNA.

Phylogenetic tree analysisFor all patients withmultiple samples, we investigated whether

a mutation was present (minimum two reads). This was done forall mutations called as Cat 1 in any of the samples from a givenpatient.We then used the status of eachmutation in all samples tofind the most probable phylogenetic tree (see SupplementaryData for details). Confidence for internal nodes was estimatedusing bootstrap analysis.

Copy number estimationThe sequenza package was used to define the copy number

variation from WES data (19).

Cancer cell fraction calculationCancer cell fraction (CCF) of Cat 1 Tier 0–1 mutations was

estimated by integrating copy number estimates and normal cellcontamination (histology) with the variant allele frequency asdescribed in ref. 20. For each autosomal mutation with at least 30reads, we obtained an estimate of the CCF and a 95% confidenceinterval (CI). We defined mutations as being clonal if the 95% CIincluded 0.95 and subclonal otherwise.

Statistical inference of tumor cell populations using deepexome sequencing

PyClone 0.12.7 (12) was used to infer subclonal populations.

Statistical analysist test statistics was used unless otherwise indicated.

Data accessThe sequencing data reported were deposited to the European

Genome-Phenome Archive (EGA; www.ebi.ac.uk/ega/) underaccession nr. EGAS00001001686.

ResultsDNA and RNA sequencing

WES was performed on 75 bladder tumors from 38 patients(Supplementary Table S1A and S1B). We analyzed two to fivemetachronous tumors from 29 patients and a single tumorfrom 9 patients. In total, 17 patients (35 tumors) had nonpro-gressive disease (NPD, patients with recurrent disease and nostage progression) and 21 patients (40 tumors) had PD(patients with transitions from Ta!T1–4 or from T1!T2–4).Tumor and matched germline DNA was sequenced to anaverage depth of 90� (28-450);Fig. 1A; Supplementary TableS1C). We identified an average of 711 (113-3331) SNVs persample within the exome target regions, categorized by strengthof evidence and expected functional impact (SnpEff; ref. 16) asdescribed previously (Supplementary Table S2; ref. 8) as well asan average of 26 (5-104) indels per sample (Supplementary

Mutational Heterogeneity in Bladder Cancer

www.aacrjournals.org Cancer Res; 76(19) October 1, 2016 5895




ADNA−seq read depth (blue line) and RNA−seq total read (green bars)

0

10

100

1000

0

100

200

300

DN

A c

over

age M

illion reads

****B Mutation/Mb for Cat 1 Tier 0 (Splice site + Nonsense), Tier 1 (Missense), Tier 2 (Silent + Other) mutations and indels (Tier 0 + 1).

0

2.5

5

7.5

10

0

0.5

1

Mut

atio

ns/M

b

Splice SitesNonsenseMissenseSilentOther

IndelsSS+NS

C Clinical information and signature classification65

P

NC

NP

71

P

NC

NP

67

P

NC

NP

68

P

NC

P

39

P

NC

NP

41

P

NC

NP

66

P

NC

NP

68

P

NC

NP

71

P

NC

NP

75

P

NC

NP

84

P

NC

P

87

P

NC

NP

75

P

NC

NP

78

P

NC

NP

55

P

57

P

NC

NP

64

P

NC

P

65

P

NC

NP

74

P

NC

NP

75

P

NC

NP

76

P

NC

NP

44

P

NC

NP

47

S

NC

NP

50

P

NC

NP

52

P

NC

P

53

P

NC

NP

73

P

NC

NP

75

P

C

P

73

P

NC

P

74

P

C

NP

71

P

NC

NP

57

P

C

NP

63

P

C

NP

62

P

C

P

65 68

P

NC

NP

75

P

NC

P

67

P

NC

NP

68

P

NC

NP

68

P

NC

P

79

P

NC

NP

85

C

P

80

P

NC

NP

81

P

61

P

NC

NP

67

P

NC

NP

68

S

C

P

74

P

C

P

75

P

NC

P

75

P

C

P

73

P

C

P

75

P

C

P

67

P

C

NP

71

P

C

NP

72

P

C

P

70

P

C

NP

73

P

C

P

63

P

C

NP

58

P

C

P

61

C

P

78

P

C

P

63

P

C

P

63

P

C

NP

76

P

C

P

80

S

C

P

80

P

74

S

C

NP

83

P

C

P

83

S

C

NP

84

S

C

P

75

P

C

P

76

P

C

P

89

P

C

P

89

C

NP

74

S

C

P

Cis signatureTumor typeGradeStageProgressionSmokingSexAge

p01_

T6p0

1_T1

5p0

2_T1

p02_

T3p0

3_T1

p03_

T6p0

4_T1

p04_

T9p0

5_T1

p05_

T7p0

6_T1

p06_

T7p0

7_T1

p07_

T8p0

8_T1

p08_

T4p0

9_T2

p09_

T7p1

0_T1

p10_

T5p1

0_T8

p11_

T2p1

1_T8

p11_

T12

p11_

T14

p11_

T15

p12_

T1p1

2_T5

p13_

T1p1

3_T4

p14_

T1p1

5_T3

p16_

T2p1

7_T1

p17_

T7

p18_

T1p1

8_T6

p19_

T1p1

9_T2

p19_

T3p2

0_T1

p20_

T4p2

1_T1

p21_

T3p2

2_T1

p23_

T1p2

3_T4

p24_

T1p2

4_T3

p24_

T5p2

5_T1

p25_

T2p2

6_T4

p26_

T6p2

6_T9

p27_

T1p2

7_T7

p28_

T5p2

9_T1

p29_

T6p3

0_T1

p31_

T2p3

1_T3

p32_

T1p3

3_T1

p33_

T2p3

4_T1

p35_

T1p3

5_T2

p35_

T4p3

6_T1

p36_

T2p3

7_T1

p37_

T2p3

8_T2

D Most frequently hit genes by mutations (Cat 1 Tier 0 + 1 deleterious) and indels

KDM6A *

ARID1A *

FGFR3 *

TP53 *

FAT1*

EP300

RBM10*

STAG2*

ZFP36L1*

PIK3CA *

MLL3

MLL2

ELF3*

FSIP2

MACF1

TSC1*

DMXL2

RB1

CRYBG3*

20 10 5 0 0 5 10 20

DNA Mutation information

RNA Expression information

Copy number variation information

Black box means

that the SNV was

found in RNA−seq

Indel found

DNA Mutation information:

No deleterious SNV found

0 0.5 1SNV allele frequency

RNA Expression information:

No RNA expression (FPKM=0)

0 5 10+Log2(FPKM+1)

Copy number variation information:

No copy number change

0 copies

1 copy

3 copies

4+ copies

SNV Tier distribution:

0 5 10 20

Number of SNVs

Tier 0Tier 1 deleterious

Color code for panel CSex:

Male

Female

Smoking:Never

Former

Current

Unknown

Progression:Nonprogressing

Progressing

Stage:Ta

T1

T2–4Grade:

Grade I

Grade II

Grade III

Grade IV

Tumor type:SolidS

PapillaryP

UnknownCIS signature:

Non-CISNC

CISC

Unknown

PDNPD

Lamy et al.

Cancer Res; 76(19) October 1, 2016 Cancer Research5896




Table S2B). Of these, an average of 110 (5-1,259) SNVs (Cat 1Tier 0þ 1) and 8 (1–42) indels per tumor had a predicted effecton protein function (Fig. 1B). The number of mutations wasnot correlated to sequencing depth. Whole-transcriptomesequencing was performed for 71 of the 75 tumors. Sampleswere sequenced to a mean of 119.6 (7.8–234.6) millionmapped reads (Fig. 1A; Supplementary Table S1D). Clinicaldata for all patients are summarized in Fig. 1C.

Mutational spectrum of tumors from NPD and PD patientsTo evaluate the mutational landscape in bladder cancer, we

identified themost frequentlymutated genes (Supplementary Fig.S1). A subset of genes with mutations predicted as deleterious(SIFT, PROVEAN, and Polyphen2) is presented in Fig. 1D(21, 22). In agreement with previous studies including NMIBC,themost frequentlymutated genes in the total patient cohortwereKDM6A (53%),ARID1A (32%), FGFR3 (26%),TP53 (24%), FAT1(18%), EP300 (16%), RBM10 (16%), and STAG2 (16%; refs. 8, 9,23). Compared with studies of MIBC, we observed higher muta-tion rates of KDM6A, ARID1A, FGFR3, FAT1, RBM10, and a lowermutation rate of TP53 (Fig. 1D and Supplementary Fig. S1;refs. 23–26). There was no significant difference in the mutationrates when comparing the NPD and PD groups (10.3 vs. 14.6mutations/Mb; t test with unequal variances, P ¼ 0.20, SNVsCat 1–2 þ indels). Furthermore, no significant difference wasobserved when considering only the first Ta tumor from eachpatient in each group (8.8 vs. 12.1 mutations/Mb; t test withunequal variances, P ¼ 0.37, SNVs Cat 1–2 þ indels). Thevariation in the PD group was significantly higher for bothcomparisons (Bartlett test, P ¼ 0.001 and P ¼ 0.004).

We observed higher intrapatient variation of the mutationalspectrum for metachronous tumors from patients with PD com-pared with tumors from patients with NPD (c2 test, P ¼0.0013; Fig. 2A), pointing toward additionalmutationalmechan-isms in the PD group, or more variation in repair capacity. UsingRNA-seq data, we classified all samples according to a previouslyreported CIS signature (5) using ConsensusClusterPlus (Fig. 1C;ref. 27). The CIS signature is associated with high-risk NMIBC(concomitant CIS) and disease progression (28). The CIS signa-ture showed a significant correlation to progression in our study(c2 test; P < 0.02). When stratifying themutational analysis basedon samples with similar gene expression signatures, we did notobserve a significant difference in SNVs or indels between thegroups (Fig. 1D).

Analyzing the proportions of the individual mutation types(Cat 1) in the initial versus the most recent tumor showed asignificant decrease (P < 0.017) of C!T mutations in the PDpatients (Supplementary Fig. S2A–S2D). When comparing thechange in proportion (mean of absolute changes) of the individ-ual mutation types from the initial to the most recent tumor, we

found a larger change in C!G and C!T (P < 0.05) in the PDpatients compared with the NPD patients (Supplementary Fig.S2E). Taking the direction into account, only the proportion ofC!T mutations remained significant (P < 0.01; SupplementaryFig. S2F).

To gain insight into the mutational processes and signatures,we included the 50 and 30 nucleotides and grouped the muta-tions (Cat 1 þ 2) by context (Supplementary Fig. S3; ref. 6). Thefrequency of the 22 signatures described by Alexandrov andcolleagues (6) was estimated in our 75 tumors (Fig. 2B). Inmost tumors, we observed several signatures (on average, 3.72signatures). The six most abundant signatures (1A, 1B, 2, 6, 12,and 13) were used to define the mutational processes in thesamples (Fig. 2C). Signature 1A was found with a higherfrequency in the NPD group compared with the PD group(P < 0.0001), whereas no differences between groups wereobserved for the remaining 5 signatures (Supplementary Fig.S4). When focusing only on the first Ta tumors from the twogroups, signature 1A was found with a higher frequency in theNPD group compared with the PD group (Mann–Whitney test,P < 0.0011), and signatures 1B and 6 were more frequent in thePD group (Mann–Whitney test, P ¼ 0.0270 and P ¼ 0.0295,respectively; Fig. 2D). Signatures 1A and 1B (characterized byC!T substitutions at NpCpG trinucleotide) are related to age(5-methyl-cytosine deamination). However, there was no agedifference between patients at the time of the first Ta tumors inthe NPD and PD groups [Mann–Whitney test, P¼ 0.12, medianage 66 (39–84) vs. 67 (58–80)].

When comparing the initial versus themost recent tumor in thetwo groups, signature 1A was found more frequently in theprogressed tumors in the PD group (Wilcoxon signed-rank test,P ¼ 0.0048), whereas no difference was observed in the NPDgroup (Supplementary Fig. S5). Thefinding that signature 6 (C!Tat NpCpG) is present in 60% of the tumors and is more abundantin the PD group is interesting, as this signature is a hallmark ofDNA mismatch repair deficiency, not commonly described inbladder cancer.

Samples were classified according to the APOBEC signature,where "high APOBEC"was defined as a contribution of signatures2 and 13 above 30% and low otherwise. About two thirds of thePD patients were classified as "high APOBEC" in at least one oftheir tumor samples compared with one third in NPD patients(Fig. 2E; 66% vs. 35%; Fisher's exact test, P¼ 0.04). Interestingly,intrapatient shift in APOBEC classification was observed in onlyone NPD patient (1/14), while it was seen in 53% (8/15) of PDpatients (Fig 2E; Fisher's exact test, P¼ 0.009). In PDpatients witha classification shift, the "high APOBEC" classification wasobserved in the initial tumor for half of the patients. Consequent-ly, the data indicate that APOBEC mutational activity is a lateevent, often tumor specific.

Figure 1.Mutational landscape andassociated clinical andhistopathologic information.A,DNAsequencing readdepth (blue lines) andRNA-seq total read counts (greenbars).B, mutation rate subdivided in types of mutations [SNV mutations: splice sites (SS) and nonsense (NS; Tier 0), missense (Tier 1), silent (Tier 2), or other(Tier 2þ); indels]. Samplesmarkedwith anasteriskwere scaleddownbya factor of ten. OnlyCat 1mutationswere considered.C, clinical information and classificationof the sample using RNA-seq signatures. Samples are sorted according to the CIS signature. Note that p24 and p26 show stage reversion following initialprogression, with no grade differences. D, frequently mutated expressed genes across the patients with information about mutation frequency (Cat 1 Tier0þ deleterious Tier 1), RNA expression, and copy number variation. Supplementary Fig. S1 shows all genes affected in at least four patients, and this figureshows the subset of genes that have a median FPKM value above 1 in either the tumor samples or the normal samples. Orange boxes were added to the DNAinformation level when the alternate allele was observed and not called in the same patient. � , significantly mutated genes (MutSigCV).






0

10

20

30

40

50

60

p01_

T6p0

1_T 1

5p0

2_T1

p02_

T3p0

3_T1

p03_

T6p0

4_T1

p04_

T9p0

5_T1

p05_

T7p0

6_T1

p06_

T7p0

7_T1

p07_

T8p0

8_T1

p08_

T4p0

9_T2

p09_

T7p1

0_T1

p10_

T5p1

0_T8

p11_

T2p1

1_T8

p11_

T12

p11_

T14

p 11_

T15

p12_

T 1p1

2_T5

p13_

T1p1

3_T4

p14_

T1p1

5_T3

p16_

T2p 1

7_T1

p17_

T7p1

8_T 1

p18_

T6p1

9_T1

p19_

T2p1

9_T3

p20_

T 1p2

0_T4

p21_

T1p2

1_T3

p22_

T 1p 2

3_T1

p23_

T4p2

4_T1

p24_

T3p2

4_T5

p25_

T1p2

5_T2

p26_

T 4p 2

6_T 6

p26_

T9p2

7_T1

p27_

T 7p2

8_T5

p 29_

T1p2

9_T6

p30_

T1p 3

1_T 2

p31_

T3p3

2_T1

p33_

T1p 3

3_T2

p34 _

T1p3

5_T1

p35_

T2p3

5 _T4

p 36 _

T1p3

6_T2

p37_

T1p3

7_T 2

p38 _

T2

NPD PD

Cat 1

+2 S

NV/

Mb Sig 6

Sig 12Sig 1BSig 1ASig 13Sig 2

0

5

10

15

20

25

30

35

40

45

50

Sign

atur

e.2

Sign

a tur

e.1 B

S ig n

atur

e .1 2

Sig n

a tu r

e.6

Sig n

atur

e.1A

Sign

atur

e.13

Sign

atur

e.4

S ign

atur

e.20

Sig n

a tur

e.5

Sign

atur

e.16

Sign

atur

e.11

S ign

atur

e.19

Sign

atur

e.21

Sign

atur

e.3

Sign

atur

e.15

Sig n

atur

e .7

Sign

a tur

e.8

Sign

atur

e.18

Sig n

a tur

e.9

Sign

atu r

e.10

S ign

atur

e.14

Sign

atur

e.17

Num

ber o

f sam

ples

hav

ing

the

sign

atur

e

p01 _

T6p 0

2 _T1

p03_

T1p0

4_T1

p05_

T 1p0

6_T1

p07_

T1p0

8_T 1

p 09 _

T2p1

0_T1

p 11 _

T 2p1

2 _T1

p 13 _

T1p1

4_T 1

p 15 _

T3p1

6_T 2

p17 _

T1p1

8_T 1

p19_

T1p2

0_T1

p21_

T1p2

2_T1

p23_

T1p2

4_T1

p25_

T1p2

6_T4

p27_

T 1p2

8_T5

p29_

T 1p3

0_T1

p31 _

T2p3

2_T1

p01_

T 6p0

2_T1

p03_

T1p0

4_T1

p05_

T1p0

6 _T1

p07_

T1p0

8 _T1

p09_

T 2p1

0_T1

p11_

T2p1

2 _T1

p13_

T1p1

4_T1

p15_

T3p1

6_T 2

p17_

T1p1

8_T1

p19_

T1p2

0 _T1

p21_

T1p2

2_T1

p23_

T1p 2

4_T1

p25 _

T1p2

6_T4

p27_

T 1p2

8_T5

p29_

T1p3

0_T1

p31_

T2p 3

2_T1

p01_

T6p 0

2 _T1

p03_

T1p 0

4_T1

p05_

T1p0

6 _T1

p07 _

T1p0

8_T1

p09_

T2p1

0_T1

p11_

T2p1

2_T1

p 13_

T1p1

4_T1

p 15 _

T3p1

6_T2

p17 _

T1p1

8_T1

p 19 _

T1p2

0_T1

p 21_

T 1p2

2_T1

p23_

T1p2

4_T1

p25 _

T1p2

6_T4

p27_

T1p2

8_T5

p29_

T1p3

0_T1

p31_

T2p3

2_T1

B

A

D

E

C NPD PDAll NPD samples NPD First Ta samples PD First Ta samplesAll PD samples

0%

20%

40%

60%

80%

100%

p01_

T 15

p01 _

T 6p0

2_T1

p02 _

T3p 0

3_T1

p03 _

T 6p 0

4_T1

p 04_

T9p0

5_T 1

p05_

T7p0

6_T1

p 06_

T7p0

7_T 1

p07_

T8p 0

8_T1

p08_

T 4p0

9_T2

p09_

T7p1

0_T1

p10_

T5p1

0_T8

p11_

T12

p11_

T14

p11_

T15

p11_

T2p 1

1_T8

p12 _

T 1p1

2_T5

p 13_

T1p1

3_T4

p14_

T1p1

5_T3

p16 _

T2p 1

7_T1

p 17_

T7p 1

8_T1

p18 _

T 6p 1

9_T1

p 19_

T2p1

9_T3

p20_

T1p2

0_T4

p 21_

T1p2

1 _T 3

p22_

T1p 2

3_T1

p23_

T4p2

4_T 1

p24_

T3p2

4 _T 5

p 25_

T1p2

5_T2

p26_

T4p2

6_T6

p26_

T9p2

7_T1

p27_

T7p2

8_T 5

p29_

T1p 2

9_T6

p30_

T1p3

1_T2

p 31_

T3p 3

2_T 1

p33_

T1p3

3_T2

p34_

T 1p3

5 _T1

p 35_

T2p3

5_T4

p36_

T 1p3

6_T2

p37_

T1p3

7 _T 2

p38_

T2

NPD PD

APO3B

APO3A

0%

20%

40%

60%

80%

100%

APO3BAPO3A

F

G

0.0

0.2

0.4

0.6

0.8

1.0Signature 1A Signature 1B Signature 6

NPD

PD

0%

20%

40%

60%

80%

100%

p01_

T6p0

1_T1

5p 0

2_T 1

p02_

T3p0

3_T1

p03_

T6p0

4_T 1

p04_

T9p0

5_T1

p05_

T7p 0

6_T 1

p06_

T7p 0

7_T1

p07_

T8p 0

8_T1

p08_

T4p0

9_T2

p09_

T7p1

0_T1

p10_

T5p 1

0_T8

p11_

T12

p11_

T14

p11_

T15

p11_

T2p1

1_T 8

p12_

T1p1

2_T5

p13_

T1p1

3 _T4

p14_

T1p1

5 _T3

p16_

T 2p1

7 _T1

p 17_

T7p1

8_T1

p18_

T6p1

9_T1

p19_

T 2p1

9 _T3

p20_

T 1p2

0_T4

p 21_

T 1p2

1_T3

p 22_

T 1p2

3_T1

p23_

T4p2

4 _T1

p24_

T3p2

4 _T5

p25_

T1p2

5 _T 2

p26_

T4p2

6_T6

p26_

T9p2

7_T1

p 27_

T7p2

8_T5

p 29_

T1p2

9_T6

p30_

T1p3

1_T2

p31 _

T3p3

2_T1

p 33_

T1p3

3_T 2

p34_

T1p3

5_T 1

p35_

T2p3

5_T 4

p36_

T1p3

6_T 2

p37_

T1p3

7_T 2

p38_

T2

NPD PD

T→G

T→C

T→A

C→T

C→G

C→A

p 01

p02

p03

p04

p05

p06

p07

p08

p 09

p10

p11

p12

p 13

p14

p 15

p 16

p17

p18

p19

p 20

p 21

p 22

p23

p24

p25

p26

p 27

p28

p29

p 30

p31

p32

p33

p34

p 35

p 36

p37

p 38

Tumor 1Tumor 2Tumor 3Tumor 4Tumor 5

↑ ↓ ↑ ↓ ↓ ↑ ↑ ↓ ↑

APOBEC LowAPOBEC High

Lamy et al.





No correlation between the APOBEC score and the mutationrate (Cat 1 þ 2 SNVs/Mb) was detected (Supplementary Fig.S6). Previous studies have shown a correlation between theAPOBEC signature and APOBEC expression, especially APO-BEC3B (10, 29). Here, transcript expression values of nineAPOBEC genes did not show a correlation between expressionand APOBEC signatures 2 þ 13. Comparing APOBEC expres-sion in Ta tumors from NPD versus PD patients showed nosignificant difference; however, APOBEC3A expression wasborderline significantly higher in the PD group (Mann–Whit-ney test, P ¼ 0.065; Supplementary Fig. S7). In conclusion, weidentified no correlation between mutational load and expres-sion of APOBEC transcripts. The reason for this could be thatmRNA abundance in tumors does not represent abundance atthe time of mutagenesis.

Recently, Chan and colleagues (30) showed that APOBEC3Aand APOBEC3B can be further separated from the TCA motifbased on the nucleotide 50 of the consensus trinucleotide, whereAPOBEC3A (A3A) prefers a pyrimidine base and APOBEC3B(A3B) a purine base. Conducting this analysis showed that thesignature of A3A contributed twice as much as A3B (Fig. 2F).Interestingly, we found the opposite pattern for the mRNA levelsof A3A and A3B (Fig. 2G), in concert with the A3A enzyme beingmoremutagenic thanA3B.Weobserved nodifferences inA3A andA3B across groups.

ASEWe performed a combined analysis of DNA and RNA to

investigate ASE. First we compared our tumor samples withsamples from normal bladder urothelium from TCGA (31).Overall, cancer samples showed a higher ASE compared withnormal samples [Wilcoxon rank sum test with continuitycorrection (WRST), P ¼ 1.14e�96], especially for tumorsuppressor genes (WRST, P ¼ 0.00169) but also more broadly("other genes": WRST, P ¼ 3.89e�100; Supplementary TableS3A). Five genes, PLEC, ACTG2, SVIL, COL5A1, and MMP2,were found to have a higher ASE in bladder cancer than innormal bladder with an FDR below 0.25 (SupplementaryTable S3B). We also observed that genes that were hit by amutation had a significantly higher ASE (mean ¼ 0.183) thannonmutated genes (mean ¼ 0.165; WRST, P ¼ 2.37e�07).Interestingly, the ASE was directed toward the reference allelefor Tier 0 mutations (exact binomial test, P < 0.039), while itwas directed toward the alternate allele for all other tiers(exact binomial test, Tier 1: P < 8.8e�12; Tier 2: P < 2.6e�5).Next, we compared the NPD and the PD groups and foundhigher ASE in the initial Ta tumors from PD patients com-pared with those from NPD patients (WRST, P ¼ 1.18e�56),with tumor suppressor genes showing a larger ASE thanoncogenes (WRST, P ¼ 0.0164 vs. P ¼ 0.417; SupplementaryTable S3C and S3D).

Group and pairwise comparison of frequently mutatedgenes

Comparison of the initial Ta tumors from each group iden-tified possible early drivers of progression (LRP1B, RYR3,PCDHGA12, TLN1, RP1L1, BBX, ZNF717, ADAMTS18, TENM1,and SSPO) only mutated in Ta tumors from the PD group(Supplementary Fig. S8A). A comparison of the initial Tatumors from NPD patients to the T1 or T2 tumors from PDpatients (most recent tumor) is found in Supplementary Fig.S8B. A comparison of paired Ta tumors from NPD patientsrevealed frequent mutations in PIK3CA, RBM10, KDM6A,RANBP2, and FGFR3 were early events present in the initialTa, whereas mutations in MKI67, CCDC168, MIA3, andOPN1LW primarily occurred late in tumorigenesis of recur-rent Ta tumors (Supplementary Fig. S8C). Finally, comparingmetachronous tumors from PD patients showed frequentmutations in genes like RYR3, RP1L1, KDM6A, FGFR3,PIK3CA, and ZNF717 in the initial tumor (SupplementaryFig. S8D). Interestingly, only 35% (7/20) of the mutationsin these genes seemed to be retained in the progressedsamples. However, visual inspection resulted in detection ofadditional recurrent low frequency SNVs (marked by orangein Supplementary Fig. S8D) in early (n ¼ 7) and late PDsamples (n ¼ 8), increasing the retained mutations to 50%(10/20) in these genes.

Temporal analysis of mutations in individual patientsTo gain further insight into the accumulation of genetic altera-

tions, we constructed phylogenetic trees of disease evolutiontaking all mutations in Cat 1 Tier 0–2 into account. We hypoth-esized that mutations found in the ancestral branch, that is,observed in all samples from the same patient, probably charac-terized the field disease. We observed a tendency for longerancestral branches [60.8% (31%–85%) vs. 46.5% (14%–87%)of the SNVs; P ¼ 0.06] for the NPD patients (Fig. 3 and Supple-mentary Fig. S9). Furthermore, in 4 of 6 patients with three ormore tumors analyzed, we observed that new mutations are onlyfound in the latest tumors (Fig. 4 and Supplementary Fig. S10;Supplementary Table S4).

Next, we hypothesized that a larger fraction ofmutations in theancestral branch is clonal (present in all carcinoma cells) com-pared with the sample-specific branches, as they arose early (20).We calculated the CCF for all Cat 1 Tier 0–1 mutations in allsamples as in ref. 20. Estimations were made only for mutationson autosomal chromosomes and with enough coverage (>30reads). The clonal status is shown in Figs. 3 and 4 for all drivergenes in patients with paired samples. When looking at patientswith two tumors analyzed, we observed as expected a higherproportion of clonal mutations in the ancestral branch versussample-specific branches (Wilcoxon signed-rank test one-tailed,P ¼ 0.002; Supplementary Figs S11–S14).

Figure 2.Mutational signatures identified in the patient cohort. A, sample-specific mutation spectrum. The relative frequency of each mutation type is shown. B, numberof samples affected by the 22 signatures sorted in descending order of prevalence. C, contribution of the six most prevalent signatures colored as in A.Vertical axis, somatic mutation prevalence; pie charts, average contribution of each signature.D, graphs display the contributions of signature 1A, 1B, and 6 across allinitial Ta tumors from the NPD and PD patients. E, APOBEC signature classification. Arrow, intrapatient shifts. F, relative contributions of APOBEC3A andAPOBEC3B signatures. The number of mutations in RTCA (associated to APOBEC3B) or YTCA (associated to APOBEC3A) context was calculated, and thecontribution from each signature is plottedwithout correction for background composition. Here, R corresponds to A or G and Y to C or T.G, relative contributions ofAPOBEC3A and APOBEC3B expression.






p01

(n = 117)

45 %

15 %

39 %Ta 1

Ta 2

ATMNOTCH2

FGFR3TXNIP

0150Years after first tumor

Ta 1Ta 2

p02

(n = 294)

65 %

31 %

4 %

Ta 1

Ta 2

COPS2CHEK2PIK3CA

TRIOSTAG2

KDM6ASMC1A

ANK3CREBBP

CDH1FGFR3KDM6A


Ta 1Ta 2

p03

(n = 206)

59 %

21 % 20 %

Ta 1 Ta 2

ARID2FUS

FGFR3


Ta 1Ta 2

p04

(n = 233)

84 %

6 %10 %

Ta 1

Ta 2

BRCA1BRCA1MED12


Ta 1Ta 2

p05

(n = 441)

78 %

10 % 11 %

Ta 1 Ta 2

ZMYM2RB1

ADAM10PIP5K1A

MLH1CCT5CASTMLL3MLL3

MYH10


Ta 1Ta 2

p06

(n = 300)

45 %

5 %

50 %Ta 1

Ta 2

BAP1BAP1

FGFR3ERBB2IP


Ta 1Ta 2

p07

(n = 472)

58 %

5 %

37 %Ta 1

Ta 2

EP300PIK3CA

PCDH18FGFR3

SMARCA4NCOR2CASP8

MLL3


Ta 1Ta 2

p08

(n = 186)

39 %

9 %

52 %Ta 1

Ta 2

NRASBAP1MLL2

PIK3CABAP1

SMC1AATRX


Ta 1Ta 2

p09

(n = 160)

81 %

4 %

14 %Ta 1

Ta 2

FGFR3MLL3

KDM6ARBM5


Ta 1Ta 2

p12

(n = 157)

85 %

3 %

12 %Ta 1

Ta 2

BCLAF1


Ta 1Ta 2

p13

(n = 366)

60 %

33 %

7 %

Ta 1

Ta 2

ZMYM2CREBBP

BRCA1ARID1A

MLL3KDM6AACTG1ACTG1MTOR

FGFR3


Ta 1Ta 2

p17

(n = 279)

31 %

54 %

14 %

Ta 1

Ta 2

TP53AFF4


Ta 1Ta 2

p18

(n = 525)

29 %

36 % 34 %

Ta T1

ARID1AKDM6A

THRAP3PIK3CA

MAP3K1ACTB


TaT1

p20

(n = 310)

78 %

10 % 11 %

Ta T1

TP53ARHGAP35ARHGAP35

ASXL1PIK3CAKDM6A


TaT1

p21

(n = 737)

14 %

37 %49 %

Ta

T2

SOS2CHD9

SMARCA4EP300RBM5

FGFR3STAG2

AQRARID1A

FAT1


TaT2

p23

(n = 745)

61 %

12 %27 %

Ta

T2

ANK3AHNAK

TP53ELF3

H3F3ASETD2

CLTCAHR


TaT2

p25

(n = 236)

32 %

5 %

63 %

Ta

T1

ARID1ABCLAF1ERBB3

TP53TSC1


TaT1

p27

(n = 178)

56 %

22 % 22 %

Ta T2

PIK3CAFGFR3

TP53


TaT2

p29

(n = 117)

42 %

10 %

48 %Ta

T1

APCCASP8CASP8


TaT1

p31

(n = 174)

59 %

19 % 22 %

TaT2

CREBBPZNF814ASXL1

CICFGFR3FBXW7


TaT2

p33

(n = 1,077)

87 %

12 %

1 %

T1

T2

CBLCBL

BRCA2BLM

TP53SMARCA4

ELF3JAK1

FUBP1RUNX1

CLASP2TET2CCT5TRIO

MAP3K1HSP90AB1

ATMERBB2

DIS3


T1T2

p36

(n = 1,209)

40 %

37 %22 %

T1

T2

HRASRB1

HSP90AA1HSP90AA1HSP90AA1

DICER1PCSK6ERBB2

TP53JAK1

ARHGAP26ATRX

CBLPCSK6

ADAM10EP300

EIF4A2ACTBTBX3

RHOA


T1T2

p37

(n = 560)

14 %

72 %

14 %

T1

T2

ERBB2AHNAKBRCA2BRCA2BRCA2

AQRCLSPNEP300

TBL1XR1


T1T2

NPD

Pat

ient

sN

P Pa

tient

s

***

*

***

****

**

**

**

****

**

** *

*

****

**

**

*

*

*

Lamy et al.





Spatial heterogeneityThe analysis of metachronous tumors showed a high degree of

intrapatientmutational heterogeneity. We therefore askedwheth-er this reflected high intratumor heterogeneity, potentially relatedto tissue sampling bias, or whether it mainly reflected changesbetween metachronous tumors. To evaluate the degree of intra-tumor heterogeneity, we performed multiregional exomesequencing. Eight regions from a single muscle-invasive tumorbiopsy were lasermicrodissected and exome sequenced to ameandepth of 61.1� (31.4–73.6; Fig. 5A). The regions were evolution-arily similar (Fig. 5B andC), with relatively fewprivatemutations.However, allele frequencies of the SNVs in four adjacent regionswere consistently higher compared with the remaining fourregions (Fig. 5B, dark red), indicating a higher normal tissuecontribution in these regions. A pool of sequences from all eightregions was analyzed with PyClone. Two main subclones wereobserved reflected by the twomain clusters (Fig. 5D), underlininga lowdegree of heterogeneity. However, no biclonal pattern couldbe observed in the heatmap from the eight regions (Fig. 5B andC),pointing toward a similar intermix of the two subclones in allregions. This suggests that the heterogeneity observed at the bulklevel is also present at the local level. Consequently, a minorfraction of the intrapatient variation appears to be due to intra-tumor heterogeneity.

Subclonal selectionTo understand patterns of clonal selection in patients with

metachronous tumors, we performed ultradeep sequencing ofselected SNVs (30–105 for each patient) in 20 patients with twotumor samples (n ¼ 15), three tumors (n ¼ 4) and five tumors(n ¼ 1). For all 1,800 SNVs, we achieved a median coverage of6,809� (2,650–87,455) per sample (Supplementary Table S5).The clonal subpopulations were inferred using PyClone (12)withcopy numbers derived fromexome sequencing data (19).Overall,we observed relatively few subclones; in most cases, we observedan ancestral clone (defined as a set ofmutations) present in all thecells from all tumors and one or two private clones in each of thetumors. Themutations from the private clones were either presentin all cells (e.g., patient p09) or only in a subset of the cells (e.g.,patient p02; Supplementary Figs. S15–S18).

Next, we combined subpopulation analyses with the muta-tional signature results. In patient p26, the phylogenetic treeshowed that the initial Ta tumor deviated from the two mostrecent tumors. The mutational profile supported this, as weobserved a strong APOBEC signature only in the two most recenttumors (Fig. 6A). Furthermore, the expression of APOBEC3B wasapproximately two times higher in the later tumors comparedwith the initial (Fig. 4A and Supplementary Fig. S19). Interest-ingly, the PyClone analysis showed the presence of a mainsubclone present in half of the cells and only found in the twomost recent tumors (Fig. 6B). This could be interpreted as thedevelopment of a new clone, after the resection of the initial

tumor, characterized by a strong APOBEC signature. Patient p19showed a different pattern. Both the APOBEC signature and thePyClone analysis suggested that the most recent tumor wasdifferent from the first two. The T1 tumor was characterized bya strong APOBEC signature not seen in the two previouslysampled tumors (Fig. 6C), and the expression of APOBEC3B wasincreased compared with the Ta tumors (Fig. 4A and Supplemen-tary Fig. S19). Furthermore, the PyClone analysis suggested thedevelopment of a new subclone in the T1 tumor present in abouthalf of the tumor cells (Fig. 6D).

Finally, we looked for potential therapeutic targets in the NPDand the PD group. We searched the DGIdb (The Drug GeneInteraction database; ref. 32), TARGET (tumor alterations relevantfor genomics-driven therapy) available from theBroad Institute aswell as the IntOGen database and focused our analysis on FDA-approved drugs. In total, we identified 19 altered potential ther-apeutic target genes, and the majority of the targets were found inthe ancestral branches (Figs. 3 and 4; Supplementary Table S6). Atleast one potential therapeutic target in the ancestral branch wasobserved in 72%(21/29) of patients.Whether the proposed drugsare applicable to the different variants identified is beyond thescope of this analysis. Activating FGFR3mutations were observedin 11 patients (S249C, Y373C, and the less common G370Cmutation), and in PIK3CA, eight mutations were found, of whichfive are known to be activatingmutations (E542K, E545K,Q546E,and H1047L). In total, 11 activating ancestral FGFR3 and PIK3CAmutations across 9 patients [31%, (9/29)] were identified, indi-cating that targeted treatments of these genes and pathways maybe highly relevant in NMIBC.

DiscussionDetailed knowledge on tumor heterogeneity and early clonal

mutations with disease-driving potential is needed to implementtargeted treatment. Bladder cancer has one of the highest muta-tion rates, which entail increased possibility of mutational het-erogeneity and increased likelihood of identifying potential ther-apeutic targets (6, 31).Here, we characterized genomic alterationsin paired tumors from 29 patients initially diagnosed withNMIBC. We found a higher intrapatient variation of the tumormutation spectrum in patients with PD, suggesting that newmutational processes are acquired during progression. In allpatients, we observed shared ancestral mutations representingthe field disease, documenting the clonal nature of bladder canceras previously observed (8).

Tumor evolution has been assessed previously in bladdercancer, showing a clonal origin of metachronous tumors(8, 33, 34). Here, we observed a higher genomic diversity in thetumors from PD patients, consistent with the change in cellularproperties acquired during disease progression. This diversitymight not be explained only by a tissue sampling procedure, aswedidnot observe large differences in themutational profile fromeight regions within a single tumor biopsy. The observation needs

Figure 3.Disease evolution in patients with two tumors analyzed. The total number of mutations and the percentages of SNVs belonging to each of the branches are shown.Long ancestral branches were shortened for graphical purpose (two crossing lines). A full version is presented in Supplementary Fig. S9. Tumor suppressorgenes, oncogenes, and bladder intOGen genes with Tier 0–1 SNVs are highlighted on the right side [ancestral, black font; sample specific, first tumor (green font)and second tumor (blue font) together with their clonal status as defined in ref. 20 (red, clonal; blue, subclonal; and gray, not defined)]. �, genes potentiallyactionable using FDA-approved drugs. Time scale, time between the tumor removals.






S2 S8 S12

S14

S15

S1 S5 S8 S1 S3 S5 S1 S2 S4 S1 S2 S3 S4 S6 S9

NPD PDP11 P10 P24 P35 P19 P26

KDM6A(T)

ARID1A(T)

SF3B1(O)

MLL5

KDM6A(T)

HRAS(O)

ATRX(T)

NOTCH2(T)

KDM6A(T)

KDM6A(T)

KDM6A(T)

MLL3(T)

FBXW7(T)

SHPRH

NF1(T)

BCOR(T)

SF3B1(O)

NF1(T)

MLL2(T)

PIK3CA(O)

PIK3CA(O)

FGFR3(O)

FLT3(O)

AR(O)

KDM6A(T)

KDM6A(T)

MSH2(T)

NFE2L2(O)

NF1(T)

SMAD2(T)

TSHR(O)

MTOR

CLSPN

CLTC

ERBB3

CHEK2

FAT1

ACTG1

DDX3X

BMPR2

Moderate impactHigh Impact

0.00

0.50

1.00

1.50

2.00

2.500.0

1.0

2.0

3.0

0.0

1.0

2.0 0.0

1.0

2.0

0.0

1.0

2.0

0.0

1.0

2.0

APOBEC3B

p10

HRAS

ERBB3

CLTC

KDM6A

CHEK2

Ta 1

Ta 2

Ta 3

1

1

p11

MTOR

ARID1A

CLSPN

SF3B1

KDM6A

Ta 1

Ta 2

Ta 5

Ta 3

Ta 4

0.961

0.962

0.550.587

p24

NOTCH2

MLL3

KDM6A

KDM6A

KDM6A

ATRX

Ta 2

Ta 1 T1

0.999

0.999

p35

NF1

NF1

SF3B1

FAT1

BCOR

T2 2T1

T2 1

0.999

0.999

p19

MLL2

ACTG1

PIK3CA

FGFR3

FLT3

Ta 1

Ta 2 T1

0.8510.851

p26

PIK3CA

DDX3X

KDM6A

AR

MSH2

TSHR

NF1

SMAD2

KDM6A

NFE2L2

BMPR2

Ta 1 T1

Ta 2

1

1

B

A

**

*

* **

**

*

*

*

*

**

*

*

*

*

*

*

*

*

0.999

0.999

0.961

0.5500.587

0.962

0.999

0.999

0.8510.851

1

1

1

1

(n = 204)

Ancestral

80 %

Ta 1

2 %

Ta 20 %

7 %

Ta 50 %

4 %

Ta 30 %

2 %

Ta 4

4 %


Ta 1 Ta 2 Ta 3 Ta 4Ta 5

(n = 334)

38 %

2 %

26 %

31 %3 %

Ancestral

Ta 1

Ta 3

Ta 2


Ta 1 Ta 2 Ta 3

(n = 313)

26 %

40 %

10 %

4 %

20 %

Ancestral

Ta 2

T1

Ta 1


Ta 1 T1 Ta 2

(n = 287)

60 %

0 %9 %

2 %28 %

Ancestral

T2 2

T2 1

T1


T1 T2 T2

(n = 443)

30 %

16 %3 %

30 %21 %

Ancestral

Ta 1

T1

Ta 2


Ta 1 Ta 2 T1

( n = 416 )

31 %

34 %

19 %

10 %6 %

Ancestral

Ta 1

Ta 2T1


2 aT1T1 aT


Lamy et al.




to be confirmed in additional samples. Inferring tumor subpo-pulations by integrating deep sequencing data and copy numberinformation can be done for a single sample, but it is moreinformative using multiple metachronous samples, where sub-clones can be traced over time (12). Single-cell sequencing per-formed on breast cancer xenografts has validated the technicalprocedure, as the main subclones found by PyClone analysis ofbulk tumor biopsieswere confirmedby the clonal genotypes fromsingle cells (7, 12). Few subpopulations were inferred in triple-negative breast cancer (35) and in tumors froma fewpatientswith

bladder cancer (8, 36). In this study, we also observed fewsubclones and no statistical difference between patients withNPD and PD. However, our analysis may underestimate thenumber of clones, as it was conducted on a selected set of SNVs.To compensate for this, we also investigated the clonal status ofthe mutations, as described earlier (20, 35). This approachrequires reliable estimations of the tumor cellularity and the copynumber. Here, we showed a general higher proportion of clonalmutation in the ancestral branch that may characterize the fielddisease.

0.0 0.2 0.4 0.6 0.8 1.0Allele frequency

0%

20%

40%

60%

80%

100%

T-G T-C T-A

C-T C-G C-A

Sub1

Sub4

Sub3

Sub5

Sub6

Sub8

Sub2

Sub7

-

-

A

PIK3CA

TP53

B C

D

Sub 1Sub 4

Sub 2Sub 5Sub 3Sub 6Sub 8Sub 7

P39 Tumor top section P39 Tumor bottom section

Sub 3

Sub 2

Sub 1

Sub 6

Sub 5

Sub 4

Sub 8Sub 7

Sub 3

Sub 2

Sub 1

Sub 6

Sub 5

Sub 4

Sub 8 Sub 7

Sub1

Sub4

Sub3

Sub5

Sub6

Sub8

Sub7

Sub2

0

20

40

60

80

Mea

n re

ad d

epth

0 0.8

Clo

ne 1

Clo

ne 2

Figure 5.

Spatial heterogeneity in bladder cancer. A,WES was applied to DNA from eight regions of a tumor from patient P39. B, heatmap of alterations in the eight regions.Bottom, boxplot of the distribution of allele frequencies. C, top, the regions were clustered (Spendall t correlation) using frequencies for all SNVs found inthe tumor; bottom,mutational spectrumandmean target read depth.D, similaritymatrix showing the cellular prevalence of all SNVs called in the tumor bulk (unionofall reads from all regions). The cellular prevalence was inferred using PyClone, and a similarity matrix was created showing the clustering of the differentSNVs with similar prevalence in 9,000 iterations.

Figure 4.Disease evolution in patientswith three ormore tumors analyzed.A, heatmaps of alterations in samples from patients havingmore than twometachronous samples.Red boxes designate mutations and intensity reflects their frequencies, black boxes indicate no mutation, and gray boxes no classification. Known tumorsuppressors (T), oncogenes (O), and bladder intOGen genes are highlighted (red box, Tier 0; orange, Tier 1). �, genes potentially actionable using FDA-approved drugs. The APOBEC3B expression is shown below each heatmap. Heatmaps showing all gene names, allele frequencies, and gene expression levelsare displayed in Supplementary Fig. S10 and Supplementary Table S4. B, phylogenetic trees (top, NPD patients; bottom, PD patients). The total number ofmutations and the percentage of SNVs belonging to each branch are indicated. Long ancestral branches were shortened for graphical purpose (two crossing lines).Tumor suppressor genes, oncogenes, and intOGen genes with Tier 0–1 SNVs are highlighted (ancestral, black font; sample specific, colored fonts; internbranches, gray fonts) together with their clonal status as defined in ref. 20 (red, clonal; blue, subclonal; and gray, not defined). � , genes potentially actionableusing FDA-approved drugs. Time scale, time between the tumor removals. Bootstrap P values are displayed to the left of the trees.






Themutational landscape is the result of multiple mutationalprocesses caused by different endogenous or exogenous dam-aging agents, as well as defects in the DNA repair or replicativemechanisms (37). The mutational signatures resulting fromthese processes have been characterized mathematically inhuman cancers (6, 35) and associated to known chemical,physical, or endogenous agents, such as UV, smoking, andongoing deamination. Especially, the APOBEC mutational sig-nature has been studied intensively (29), and APOBEC-relatedmutations have been observed at an increasing rate over time inlung cancer (38). Furthermore, the expression of APOBECproteins has previously been associated with poor prognosis inbreast cancer (39) and bladder cancer (8). The APOBEC muta-tional signature was recently shown to be significantly associ-ated with high-risk NMIBC (11), and APOBEC mutations maybe responsible for driving the tumor diversification and adap-tion under selective pressure (10). We observed that APOBECmutational signatures were more frequent in PD patients; how-ever, heterogeneity in signature frequency during disease devel-opment was observed, indicating late tumor-specific (nonfielddisease) APOBEC mutational activity.

Exome sequencing has identified genomic alterations in theNotch and Hedgehog signaling pathways to be important fortumor progression (40, 41). Furthermore, TP53 and FGFR3muta-tions are present in tumors of different stages and partly mutuallyexclusive pathways (42). We found no significant difference inmutational load when comparing early-stage tumors frompatients with different outcome. This may be caused by therelatively few samples analyzed, but it may also indicate a signif-icant heterogeneity at the genomic level, and several pathways ofprogression may exist.

Cancer tissue generally shows transcriptional bias comparedwith normal tissue, due to a highermutational load, leading somealleles to become transcriptionally silent. Five genes (PLEC,ACTG2, SVIL, COL5A1, and MMP2) showed significant ASE.These genes encode proteins associated with cell structure andmotility. Plakins (PLEC) and actins (ACTG2) are important forcell motility and cytoskeleton integrity. Supervillin (SVIL) istightly associated with actin filaments, and COL5A1 encodes acollagen subunit present in extracellular matrix (43). Finally,matrix metallopeptidases (MMP2) have the ability to cleavecomponents of the extracellular matrix and are important for

0

0

P26

(n = 416)

31%

34%

19%

10%6%

Ancestral

Ta 1

Ta 2T1

0 5 10Years after first tumor

Ta 1 T1 Ta 2

(n = 443)

30%

16%3%

30%21%

Ancestral

Ta 1

T1

Ta 2

0 5 10Years after first tumor

Ta 1 Ta 2 T1

A

C

Ta 1 T1 Ta 2

Ta 1 Ta 2 T1

B

D

P26_T4

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G

1,104 Mutations252 Mutations

P26_T6

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G

276 Mutations160 Mutations

P26_T9

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G


P19_T1

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G


P19_T2

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G


P19_T3

0.00

0.05

0.10

0.15

0.20 C > A C > G C > T T > A T > C T > G


P19

APOBEC

APOBEC

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

p26_T4 Cellular prevalence

p26_

T6 C

ellu

lar p

reva

lenc

e

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0


p26_

T9 C

ellu

lar p

reva

lenc

e

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0


p26_

T9 C

ellu

lar p

reva

lenc

e

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0


p19_

T2 C

ellu

lar p

reva

lenc

e

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0


p19_

T3 C

ellu

lar p

reva

lenc

e

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0


p19_

T3 C

ellu

lar p

reva

lenc

e

ARKDM6A

MSH2

NF1

BMPR2

DDX3X

PIK3CA

ARKDM6A

MSH2

NF1BMPR2

DDX3XPIK3CA

ARKDM6A

MSH2

NF1 BMPR2

DDX3X

PIK3CA

MLL2ACTG1

FLT3

MLL2ACTG1

FLT3

MLL2ACTG1

Figure 6.

Temporal and clonal heterogeneity in two bladder cancer samples.A, sample case: patient P26. Trinucleotidemutational context for each sample from P26. Each barrepresents the relative ratio of a given mutation in a given trinucleotide context (dark color, Cat 1 SNVs; light colors, all SNVs). B, the phylogenetic tree isshown (top left) with an arrow indicating where a newmutational process occurred in the tumor development. The time span between tumors is shown (top right),and an interpretation in terms of main subclones is given (middle right). Cellular prevalence of individual SNVs was inferred with PyClone and plottedfor each comparison (bottom): first tumor versus second tumor (left plot), first tumor versus third tumor (middle plot), and second tumor versus third tumor (rightplot). C and D, as A and B for sample case: patient P19.


Lamy et al.




metastasis (44). Nonsense and splice site mutations showedincreased wild-type allele expression, while missense and silentmutations showed increased expression of the alternate allele. Theincreased ASE for nonsense mutations is in agreement withlowered steady-state level of cytoplasmic mRNA by nonsense-mediated decay (45). Furthermore, the higher ASE of Ta tumorsfrom PD patients compared with NPD patients indicates biolog-ical differences even at this early stage.

Efficient targeted therapy should preferably target alterations inall tumor cells. Thus, it is important to clearly identify the clonalmutations. However, observing clonality within a single tumormay not necessarily imply clonality across synchronous or meta-chronous tumors. Here, we showed that analysis of paired tumorsidentified potential therapeutic targets that were of clonal origin,and consequently, targeted therapy may be able to eradicate thefield disease and multiple tumors.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: P. Lamy, I. Nordentoft, T.F. �rntoft, L. DyrskjøtDevelopment ofmethodology:P. Lamy, I.Nordentoft, P. Villesen, J.S. Pedersen

Acquisition of data (provided animals, acquired and managed patients, pro-vided facilities, etc.): I.Nordentoft, K. Birkenkamp-Demtr€oder,M.B.H. Thomsen,J. Hedegaard, M. Borre, J.B. Jensen, S. Høyer, T.F. �rntoft, L. DyrskjøtAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): P. Lamy, I. Nordentoft, M.B.H. Thomsen, P. Villesen,S. Vang, J.S. Pedersen, T.F. �rntoft, L. DyrskjøtWriting, review, and/or revision of the manuscript: P. Lamy, I. Nordentoft,K. Birkenkamp-Demtr€oder, M.B.H. Thomsen, P. Villesen, S. Vang, J. Hedegaard,M. Borre, J.B. Jensen, S. Høyer, J.S. Pedersen, T.F. �rntoft, L. DyrskjøtAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): I. Nordentoft, K. Birkenkamp-Demtr€oder,M. Borre, S. Høyer, T.F. �rntoftStudy supervision: I. Nordentoft, J.S. Pedersen, T.F. �rntoft, L. Dyrskjøt

Grant SupportThis work was supported by The John and Birthe Meyer Foundation, The

Lundbeck Foundation, TheDanish Cancer Society, TheDanishCancer Biobank,and The Danish Research Council.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received February 17, 2016; revised May 31, 2016; accepted June 26, 2016;published OnlineFirst August 3, 2016.

References1. Babjuk M, Burger M, Zigeuner R, Shariat SF, van Rhijn BW, Comperat E,

et al. EAU guidelines on non-muscle-invasive urothelial carcinoma of thebladder: update 2013. Eur Urol 2013;64:639–53.

2. Knowles MA, Hurst CD. Molecular biology of bladder cancer: newinsights into pathogenesis and clinical diversity. Nat Rev Cancer 2015;15:25–41.

3. Hoglund M. On the origin of syn- and metachronous urothelial carcino-mas. Eur Urol 2007;51:1185–93.

4. Cianciulli AM, Leonardo C, Guadagni F, Marzano R, Iori F, De NC, et al.Genetic instability in superficial bladder cancer and adjacent mucosa: aninterphase cytogenetic study. Hum Pathol 2003;34:214–21.

5. Dyrskjot L, KruhofferM, Thykjaer T,MarcussenN, Jensen JL,Moller K, et al.Gene expression in the urinary bladder: a common carcinoma in situ geneexpression signature exists disregarding histopathological classification.Cancer Res 2004;64:4040–8.

6. Alexandrov LB,Nik-Zainal S,WedgeDC,Aparicio SA, Behjati S, BiankinAV,et al. Signatures of mutational processes in human cancer. Nature2013;500:415–21.

7. Eirew P, Steif A, Khattra J, Ha G, Yap D, Farahani H, et al. Dynamics ofgenomic clones in breast cancer patient xenografts at single-cell resolution.Nature 2015;518:422–6.

8. Nordentoft I, Lamy P, Birkenkamp-Demtroder K, Shumansky K, Vang S,Hornshoj H, et al. Mutational context and diverse clonal development inearly and late bladder cancer. Cell Rep 2014;7:1649–63.

9. Sfakianos JP, Cha EK, Iyer G, Scott SN, Zabor EC, Shah RH, et al. Genomiccharacterization of upper tract urothelial carcinoma. Eur Urol 2015;68:970–7.

10. Swanton C, McGranahan N, Starrett GJ, Harris RS. APOBEC enzymes:mutagenic fuel for cancer evolution and heterogeneity. Cancer Discov2015;5:704–12.

11. Hedegaard J, Lamy P, Nordentoft I, Algaba F, Høyer S, Ulhøi BP. Com-prehensive transcriptional analysis of early stage urothelia carcinoma.Cancer Cell 2016;30:27–42.

12. Roth A, Khattra J, Yap D, Wan A, Laks E, Biele J, et al. PyClone: statisticalinference of clonal population structure in cancer. Nat Methods2014;11:396–8.

13. Gerlinger M, Catto JW, Orntoft TF, Real FX, Zwarthoff EC, Swanton C.Intratumour heterogeneity in urologic cancers: frommolecular evidence toclinical implications. Eur Urol 2015;67:729–37.

14. DePristo MA, Banks E, Poplin R, Garimella KV, Maguire JR, Hartl C, et al. Aframework for variation discovery and genotyping using next-generationDNA sequencing data. Nat Genet 2011;43:491–8.

15. Cibulskis K, LawrenceMS,Carter SL, SivachenkoA, JaffeD, SougnezC, et al.Sensitive detection of somatic point mutations in impure and heteroge-neous cancer samples. Nat Biotechnol 2013;31:213–9.

16. Cingolani P, Platts A,Wang lL, CoonM,Nguyen T,Wang L, et al. A programfor annotating and predicting the effects of single nucleotide polymorph-isms, SnpEff: SNPs in the genome of Drosophila melanogaster strainw1118; iso-2; iso-3. Fly 2012;6:80–92.

17. Trapnell C, Roberts A, Goff L, Pertea G, KimD, Kelley DR, et al. Differentialgene and transcript expression analysis of RNA-seq experiments withTopHat and Cufflinks. Nat Protoc 2012;7:562–78.

18. Alexandrov LB, Jones PH, Wedge DC, Sale JE, Campbell PJ, Nik-Zainal S,et al. Clock-like mutational processes in human somatic cells. Nat Genet2015;47:1402–7.

19. Favero F, Joshi T, Marquard AM, Birkbak NJ, Krzystanek M, Li Q, et al.Sequenza: allele-specific copy number and mutation profiles from tumorsequencing data. Ann Oncol 2015;26:64–70.

20. McGranahan N, Favero F, de Bruin EC, Birkbak NJ, Szallasi Z, Swanton C.Clonal status of actionable driver events and the timing of mutationalprocesses in cancer evolution. Sci Transl Med 2015;7:283ra54.

21. Adzhubei I, JordanDM, Sunyaev SR. Predicting functional effect of humanmissense mutations using PolyPhen-2. Curr Protoc Hum Genet 2013;Chapter 7:Unit7.20.

22. Choi Y, Chan AP. PROVEAN web server: a tool to predict the functionaleffect of amino acid substitutions and indels. Bioinformatics 2015;31:2745–7.

23. Balbas-Martinez C, Sagrera A, Carrillo-de-Santa-Pau E, Earl J, Marquez M,Vazquez M, et al. Recurrent inactivation of STAG2 in bladder cancer is notassociated with aneuploidy. Nat Genet 2013;45:1464–9.

24. Ross JS, Wang K, Al-Rohil RN, Nazeer T, Sheehan CE, Otto GA, et al.Advanced urothelial carcinoma: next-generation sequencing revealsdiverse genomic alterations and targets of therapy. Mod Pathol 2014;27:271–80.

25. Guo G, Sun X, Chen C, Wu S, Huang P, Li Z, et al. Whole-genome andwhole-exome sequencing of bladder cancer identifies frequent alterationsin genes involved in sister chromatid cohesion and segregation. Nat Genet2013;45:1459–63.






26. Iyer G, Al-Ahmadie H, Schultz N, Hanrahan AJ, Ostrovnaya I, Balar AV,et al. Prevalence and co-occurrence of actionable genomic alterations inhigh-grade bladder cancer. J Clin Oncol 2013;31:3133–40.

27. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery toolwith confidence assessments and item tracking. Bioinformatics 2010;26:1572–3.

28. Dyrskjot L, Zieger K, Real FX, Malats N, Carrato A, Hurst C, et al. Geneexpression signatures predict outcome in non-muscle-invasive bladdercarcinoma: a multicenter validation study. Clin Cancer Res 2007;13:3545–51.

29. Burns MB, Lackey L, Carpenter MA, Rathore A, Land AM, Leonard B, et al.APOBEC3B is an enzymatic source of mutation in breast cancer. Nature2013;494:366–70.

30. Chan K, Roberts SA, Klimczak LJ, Sterling JF, Saini N, Malc EP, et al. AnAPOBEC3A hypermutation signature is distinguishable from the signatureof background mutagenesis by APOBEC3B in human cancers. Nat Genet2015;47:1067–72.

31. Comprehensive molecular characterization of urothelial bladder carcino-ma. Nature 2014;507:315–22.

32. Griffith M, Griffith OL, Coffman AC, Weible JV, McMichael JF, Spies NC,et al. DGIdb: mining the druggable genome. Nat Methods 2013;10:1209–10.

33. Kim J, Lee IH, Cho HJ, Park CK, Jung YS, Kim Y, et al. Spatiotemporalevolution of the primary glioblastoma genome. Cancer Cell 2015;28:318–28.

34. Saldanha AJ. Java Treeview–extensible visualization of microarray data.Bioinformatics 2004;20:3246–8.

35. Shah SP, Roth A, Goya R, Oloumi A, Ha G, Zhao Y, et al. The clonal andmutational evolution spectrum of primary triple-negative breast cancers.Nature 2012;486:395–9.

36. Morrison CD, Liu P, Woloszynska-Read A, Zhang J, Luo W, Qin M, et al.Whole-genome sequencing identifies genomic heterogeneity at a nucleo-tide and chromosomal level in bladder cancer. Proc Natl Acad Sci U S A2014;111:E672–81.

37. Helleday T, Eshtad S, Nik-Zainal S. Mechanisms underlying mutationalsignatures in human cancers. Nat Rev Genet 2014;15:585–98.

38. de Bruin EC, McGranahan N, Mitter R, Salm M, Wedge DC, Yates L, et al.Spatial and temporal diversity in genomic instability processes defines lungcancer evolution. Science 2014;346:251–6.

39. Sieuwerts AM, Willis S, Burns MB, Look MP, Meijer-Van Gelder ME,Schlicker A, et al. Elevated APOBEC3B correlates with poor outcomes forestrogen-receptor-positive breast cancers. Horm Cancer 2014;5:405–13.

40. Rampias T, Vgenopoulou P, Avgeris M, Polyzos A, Stravodimos K, Vala-vanis C, et al. A new tumor suppressor role for the Notch pathway inbladder cancer. Nat Med 2014;20:1199–205.

41. Shin K, Lim A, Zhao C, Sahoo D, Pan Y, Spiekerkoetter E, et al. Hedgehogsignaling restrains bladder cancer progression by eliciting stromal produc-tion of urothelial differentiation factors. Cancer Cell 2014;26:521–33.

42. van Rhijn BW, Vis AN, van der Kwast TH, KirkelsWJ, Radvanyi F, Ooms EC,et al. Molecular grading of urothelial cell carcinomawith fibroblast growthfactor receptor 3 and MIB-1 is superior to pathologic grade for theprediction of clinical outcome. J Clin Oncol 2003;21:1912–21.

43. Fang Z, TakizawaN,WilsonKA, Smith TC,Delprato A,DavidsonMW, et al.The membrane-associated protein, supervillin, accelerates F-actin-depen-dent rapid integrin recycling and cell motility. Traffic 2010;11:782–99.

44. Wieczorek E, Jablonska E, Wasowicz W, Reszka E. Matrix metalloprotei-nases and geneticmousemodels in cancer research: amini-review. TumourBiol 2015;36:163–75.

45. Bidou L, Allamand V, Rousset JP, NamyO. Sense from nonsense: therapiesfor premature stop codon diseases. Trends Mol Med 2012;18:679–88.


Lamy et al.




2016;76:5894-5906. Published OnlineFirst August 3, 2016.Cancer Res Philippe Lamy, Iver Nordentoft, Karin Birkenkamp-Demtröder, et al. Therapeutic Targets in Urothelial CarcinomaPaired Exome Analysis Reveals Clonal Evolution and Potential

Updated version

10.1158/0008-5472.CAN-16-0436doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2016/08/03/0008-5472.CAN-16-0436.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/76/19/5894.full#ref-list-1

This article cites 44 articles, 8 of which you can access for free at:

Citing articles

http://cancerres.aacrjournals.org/content/76/19/5894.full#related-urls

This article has been cited by 7 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cancerres.aacrjournals.org/content/76/19/5894To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/lookup/doi/10.1158/0008-5472.CAN-16-0436

http://cancerres.aacrjournals.org/content/suppl/2016/08/03/0008-5472.CAN-16-0436.DC1

http://cancerres.aacrjournals.org/content/76/19/5894.full#ref-list-1

http://cancerres.aacrjournals.org/content/76/19/5894.full#related-urls

http://cancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://cancerres.aacrjournals.org/content/76/19/5894


Documents

Paired Exome Analysis Reveals Clonal Evolution and ......Philippe Lamy1, Iver Nordentoft1, Karin Birkenkamp-Demtroder€ 1, Mathilde Borg Houlberg Thomsen1, Palle Villesen2, Søren