ArticlePDF Available

Genomic Evolution of Breast Cancer Metastasis and Relapse

August 2017
Cancer Cell 32(2):169-184.e7

DOI:10.1016/j.ccell.2017.07.005

License
CC BY 4.0

Authors:

Lucy R. Yates

Wellcome Sanger Institute

Stian Knappskog

University of Bergen

David C Wedge

The University of Manchester

Show all 30 authorsHide

Patterns of genomic evolution between primary and metastatic breast cancer have not been studied in large numbers, despite patients with metastatic breast cancer having dismal survival. We sequenced whole genomes or a panel of 365 genes on 299 samples from 170 patients with locally relapsed or metastatic breast cancer. Several lines of analysis indicate that clones seeding metastasis or relapse disseminate late from primary tumors, but continue to acquire mutations, mostly accessing the same mutational processes active in the primary tumor. Most distant metastases acquired driver mutations not seen in the primary tumor, drawing from a wider repertoire of cancer genes than early drivers. These include a number of clinically actionable alterations and mutations inactivating SWI-SNF and JAK2-STAT3 pathways.

Phylogenetic Trees Describe Evolution of 17 Primary Breast Cancers to Metastasis or Local Relapse

…

Figures - uploaded by Per Eystein Lønning

Content may be subject to copyright.

Content uploaded by Per Eystein Lønning

Content may be subject to copyright.

Article

Genomic Evolution of Breast Cancer Metastasis and

Relapse

Graphical Abstract

Highlights

dMetastases mostly disseminate late from primary breast

tumors, keeping most drivers

dDrivers at relapse sample from a wider range of cancer genes

than in primary tumors

dMutations in SWI-SNF complex and inactivated JAK-STAT

signaling enriched at relapse

dMutational processes similar in primary and relapse;

radiotherapy can damage genome

Authors

Lucy R. Yates, Stian Knappskog,

David Wedge, ..., Andrew Tutt,

Per Eystein Lønning, Peter J. Campbell

Correspondence

per.eystein.lonning@helse-bergen.no

(P.E.L.),

pc8@sanger.ac.uk (P.J.C.)

In Brief

By sequencing primary, locally relapsed,

and metastatic breast cancers, Yates

et al. show that clones seeding

metastasis or relapse disseminate late

from primary tumors but continue to

acquire mutations, including clinically

actionable alterations and mutations

inactivating the SWI/SNF and JAK2-

STAT3 pathways.

Molecular time

Cancer burden

Distant

metastasis

Loco-

regional

relapse

Synchronous

lymph node

metastasis

Canonical cancer genes -

TP53, GATA3, PIK3CA, AKT1, ERBB2

MRCA

Normal

cell

Genome-

doubling

Rarer cancer genes -

SWI/SNF, JAK-STAT

Drug resistance mutations -

ESR1

Driver

mutations

Radical treatment

of primary breast cancer

Yates et al., 2017, Cancer Cell 32, 169–184

August 14, 2017 ª2017 The Authors. Published by Elsevier Inc.

http://dx.doi.org/10.1016/j.ccell.2017.07.005

Cancer Cell

Article

Genomic Evolution of Breast Cancer

Metastasis and Relapse

Lucy R. Yates,

1,2,23

Stian Knappskog,

3,4,23

David Wedge,

1,5

James H.R. Farmery,

Santiago Gonzalez,

1,7

Inigo Martincorena,

Ludmil B. Alexandrov,

8,9,10

Peter Van Loo,

11,12

Hans Kristian Haugland,

13,14

Peer Kaare Lilleng,

13,14

Gunes Gundem,

1,15

Moritz Gerstung,

1,7

Elli Pappaemmanuil,

1,15

Patrycja Gazinska,

Shriram G. Bhosle,

(Author list continued on next page)

SUMMARY

Patterns of genomic evolution between primary and metastatic breast cancer have not been studied in large

numbers, despite patients with metastatic breast cancer having dismal survival. We sequenced whole ge-

nomes or a panel of 365 genes on 299 samples from 170 patients with locally relapsed or metastatic breast

cancer. Several lines of analysis indicate that clones seeding metastasis or relapse disseminate late from pri-

mary tumors, but continue to acquire mutations, mostly accessing the same mutational processes active in

the primary tumor. Most distant metastases acquired driver mutations not seen in the primary tumor, drawing

from a wider repertoire of cancer genes than early drivers. These include a number of clinically actionable

alterations and mutations inactivating SWI-SNF and JAK2-STAT3 pathways.

INTRODUCTION

Metastatic breast cancer is almost universally fatal within 5–10

years, a dismal statistic that has not changed much in the past

20–30 years (Tevaarwerk et al., 2013). Breast cancer recurrence

can take two forms: distant metastasis (commonly bone, brain,

liver, lung, and distant lymph nodes) and locoregional relapse

(recurrence in breast, chest wall, or regional lymph nodes). Lo-

coregional relapse occurs in about 10% of patients despite

optimal management of the primary tumor and is associated

Wellcome Trust Sanger Institute, Hinxton CB10 1SA, UK

Department of Clinical Oncology, Guys and St Thomas’ NHS Trust, London SE1 9RT, UK

Section of Oncology, Department of Clinical Science, University of Bergen, Bergen, Norway

Department of Oncology, Haukeland University Hospital, Bergen, Norway

Big Data Institute, University of Oxford, Oxford OX3 7BN, UK

Cancer Research UK Cambridge Institute, University of Cambridge, Li Ka Shing Centre, Robinson Way, Cambridge CB2 0RE, UK

European Bioinformatics Institute EMBL-EBI, Wellcome Genome Campus, Hinxton CB10 1SD, UK

Theoretical Biology and Biophysics (T-6), Los Alamos National Laboratory, Los Alamos, NM 87545, USA

Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, NM 87545, USA

University of New Mexico Comprehensive Cancer Center, Albuquerque, NM 87102, USA

The Francis Crick Institute, 1 Midland Road, London NW1 1AT, UK

Department of Human Genetics, University of Leuven, 3000 Leuven, Belgium

Department of Pathology, Haukeland University Hospital, Bergen, Norway

The Gade Laboratory for Pathology, Department of Clinical Medicine, University of Bergen, Bergen, Norway

Computational Oncology, Epidemiology and Biostatistics Memorial Sloan Kettering Cancer Institute, New York, NY 10065 USA

Division of Cancer Studies, Faculty of Life Sciences and Medicine, King’s College London, London SE1 9RT, UK

(Afﬁliations continued on next page)

Signiﬁcance

These ﬁndings have implications for personalized therapy of breast cancer. The late dissemination of cells that seed metas-

tasis or local relapse suggests that the primary tumor genome can proxy for the genome of disseminated cells at the time of

ﬁrst diagnosis, supporting the use of genome sequencing to aid decisions about adjuvant therapy for primary breast cancer.

Biopsy and sequencing of metastases may be helpful in some patients because most distant metastases have acquired

additional driver mutations not seen in the primary; these often involve potentially actionable genes and cellular pathways.

Sequencing local recurrences can distinguish a genuine relapse from a second primary cancer, two scenarios with very

different care pathways.

Cancer Cell 32, 169–184, August 14, 2017 ª2017 The Authors. Published by Elsevier Inc. 169

This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

with concomitant or future distant metastatic disease in 30%

and 60% of cases, respectively. In contrast, regional lymph

node metastasis found at the time of primary diagnosis is often

cured with surgery and radiotherapy but is a well-established

poor prognostic factor, associated with a higher risk of subse-

quent cancer recurrence.

Molecular proﬁling of breast cancer has typically focused on

the primary breast lesion. Gene expression proﬁles classify

breast cancers into different subtypes, with clinical trials

showing that these transcriptional signatures can be used to

support therapeutic decisions in primary breast cancer (Harris

et al., 2016). Large-scale genomics analyses have now been per-

formed in thousands of primary breast cancers, revealing the

complex mutational landscape of the disease (Banerji et al.,

2012; Cancer Genome Atlas Network, 2012; Ciriello et al.,

2015; Ellis et al., 2012; Nik-Zainal et al., 2016; Shah

et al., 2012; Stephens et al., 2012). General patterns to emerge

from these studies include that estrogen receptor (ER)-positive

primary breast cancer has a characteristic ‘‘luminal’’ transcrip-

tional proﬁle with frequent somatic mutations activating PI3K-

AKT signaling and inactivating GATA3 and the JUN kinase

pathway. Breast cancers with ampliﬁcation and/or overexpres-

sion of ERBB2 (also known as HER2) have a distinct transcrip-

tional and genomic proﬁle, conﬁrming the central role that

ERBB2 plays in the pathogenesis of this subtype of breast can-

cer. Breast cancers negative for ER, the progesterone receptor

(PR), and HER2, so-called triple-negative breast cancers, are

characterized by a ‘‘basal-like’’ transcriptional proﬁle, frequent

TP53 mutation, and extensive copy number variation. A number

of studies have revealed extensive genomic heterogeneity within

primary breast tumors and changes in subclonal structure during

systemic therapy (Balko et al., 2014; Gellert et al., 2016; Miller

et al., 2016; Ng et al., 2015; Shah et al., 2012; Wang et al.,

2014; Yates et al., 2015).

While the genome of primary breast cancer has been well

characterized, there has been considerably less analysis of

relapsed or metastatic breast cancer. Those studies that have

been performed have revealed that metastases are clonally

related to the primary tumor, sharing many of the driver muta-

tions, but nonetheless have typically acquired additional variants

not detectable in the primary lesion (Brastianos et al., 2015; De

Mattos-Arruda et al., 2014; Ding et al., 2010; Hoadley et al.,

2016; Juric et al., 2015; Savas et al., 2016; Shah et al., 2009;

Yates et al., 2015). Due to small sample sizes, however, it has

proved difﬁcult to extract general patterns of evolution between

primary and recurrence, leaving a number of unanswered ques-

tions with important biological and clinical implications. We con-

ducted this study to address some of these questions, including

how closely related a metastasis is to its primary lesion; whether

there are differences in evolution across locoregional relapse,

axillary metastases seeded by lymphatic spread, and distant

metastases seeded by hematogenous spread; whether the

driver landscape of metastases differs from primary cancers;

and whether there are cancer genes speciﬁc to metastases.

Since the survival of patients with metastatic breast cancer is

so poor, it is particularly important to establish whether newly

emerging driver mutations in the metastasis might offer opportu-

nities for personalized therapy.

RESULTS

Patient Cohort

The study comprises two major aims. In the ﬁrst, to deﬁne pat-

terns of genomic evolution between the primary cancer and dis-

ease progression, we performed whole-genome sequencing of

40 tumor samples from 17 patients to an average coverage of

423, together with matched germline DNA samples (Tables S1

and S2). These 17 patients encompassed three clinical sce-

narios: synchronous axillary lymph node metastasis; distant

metastasis and local relapse subsequent to deﬁnitive treatment

for the primary tumor. In all but one case (PD11458), primary tu-

mor samples were treatment naive and sampled at diagnosis.

Metachronous recurrence samples were obtained 8–158 months

after the primary tumor diagnosis. Distant metastatic samples

were obtained from tumor deposits in lung (n = 1), liver (n = 1),

distant skin regions (n = 2), contralateral breast (n = 1), and

distant lymph nodes (n = 2) (Figure 1). All patients underwent

standard management, including curative surgery with local

radiotherapy, adjuvant anthracycline-containing chemotherapy,

and/or endocrine therapies where appropriate (Table S1).

The second aim was to study the distribution of driver muta-

tions in distant metastatic or locoregionally relapsed breast can-

cer. To achieve this, we analyzed 227 recurrence samples from

163 patients for point mutations and copy number changes in

365 known cancer genes to an average coverage of 4673

(Tables S1 and S2). For 46 patients, 2–5 recurrence samples

David Jones,

Keiran Raine,

Laura Mudie,

Calli Latimer,

Elinor Sawyer,

2,16

Christine Desmedt,

Christos Sotiriou,

Michael R. Stratton,

Anieta M. Sieuwerts,

Andy G. Lynch,

John W. Martens,

Andrea L. Richardson,

19,20

Andrew Tutt,

16,21,22

Per Eystein Lønning,

3,4,23,

*and Peter J. Campbell

1,23,24,

Breast Cancer Translational Research Laboratory, Universite

´Libre de Bruxelles, Institut Jules Bordet, Bd de Waterloo 121, 1000 Brussels,

Belgium

Erasmus MC Cancer Institute and Cancer Genomics Netherlands, Erasmus University Medical Center, Department of Medical Oncology,

Rotterdam, the Netherlands

Department of Pathology, Brigham and Women’s Hospital, Boston, MA 02115, USA

Dana-Farber Cancer Institute, Boston, MA 02215, USA

Breast Cancer Now Research Unit, King’s College London, London SE1 9RT, UK

The Breast Cancer Now Toby Robins Research Centre, The Institute of Cancer Research, London SW3 6JB, UK

These authors contributed equally

Lead Contact

*Correspondence: per.eystein.lonning@helse-bergen.no (P.E.L.), pc8@sanger.ac.uk (P.J.C.)

http://dx.doi.org/10.1016/j.ccell.2017.07.005

170 Cancer Cell 32, 169–184, August 14, 2017

were sequenced, allowing heterogeneity of the recurrence land-

scape to be explored. For 51 individuals, the matched primary

tumor was available for sequencing.

Samples were obtained in clinically relevant scenarios

including ﬁrst relapse or metastasis and following systemic treat-

ment interventions. All samples therefore represent clinically

progressing disease. Progression during a documented sys-

temic therapy exposure occurred for 126 samples, including

endocrine therapy (n = 43), anthracyclines (n = 45), taxanes

(n = 12), and other chemotherapeutic regimens. Samples were

obtained following a median of 2 systemic treatment exposures

(range, 0–5) and after 39 months (range, 0–196) from primary

cancer diagnosis (Table S1). Tumors were classiﬁed according

to the primary tumor TNM stage, histological type, grade, and

presence of estrogen receptor (ER), progesterone receptor

(PR), and ERBB2 (HER2) ampliﬁcation.

Evolution between Primary Breast Cancer and

Metastasis/Relapse

Using whole-genome sequencing, we explored the patterns of

genomic evolution in three clinical scenarios across 17 patients:

local lymph node involvement at the time of primary tumor diag-

nosis (8 patients); locoregional relapse after apparently deﬁnitive

primary tumor treatment (n = 4); and subsequent development of

distant metastasis (n = 7) (Figure 1). We identiﬁed an average of

9,594 substitutions (range, 1,792–25,471), 1,098 indels (range,

60–12,786), and 245 structural variants (range, 6–786) within

each individual’s cancer genome (Table S3). We performed vali-

dation on 1,480 somatic substitutions and indels by custom cap-

ture pull-down or capillary sequencing (Table S4), conﬁrming

1,436 (97%) were truly present and somatically acquired. We en-

riched our validation experiment with mutations that were private

to one of the samples to enhance our ability to identify subclonal

populations. Rearrangements were validated by the visual

conﬁrmation of breakpoint-associated copy number changes.

To reconstruct the phylogenetic structure underlying disease

progression, we applied bioinformatic and deductive reasoning

approaches, as described previously (Nik-Zainal et al., 2012;

Yates et al., 2015)(Figures 1,S1, and S2A; Table S5). We used

multi-dimensional Bayesian Dirichlet processes to cluster so-

matic substitutions from multiple related samples according to

their respective mutation burden, corrected for tumor cellularity,

allele-speciﬁc copy number, and regions of differential chromo-

somal deletion between samples. We identiﬁed an average of 2.8

distinct clusters per patient (48 in 17 patients), with 94% of these

reproduced by independent clustering of high-coverage tar-

geted validation data (Figure S1 and Table S5). Individual clus-

ters inform on the structure of the phylogenetic tree, typically

Figure 1. Phylogenetic Trees Describe Evolution of 17 Primary Breast Cancers to Metastasis or Local Relapse

Each tree represents an individual patient’s breast cancer inferred from the analysis of a matched normal sample and 2–4 tumor samples per case (total of 40

tumor samples). Trees are derived from genome-wide substitutions. Trees are grouped according to scenario: distant metastasis (red panel), locoregional relapse

(blue panel), or synchronous axillary lymph node metastasis (green panel). Branches private to the metastasis or relapse follow the same color theme, while

branches representing clones that are speciﬁc to the primary tumor are gray. The black trunk represents clonal mutations that are present in 100% of cells in every

sample. Purple branches represent mutations within the metastasis or relapse that are subclonal within the primary tumor. Branch lengths reﬂect the proportio n of

clustered somatic mutations attributed to that subclone. The whole tree is scaled to the maximum length of a tree that would be inferred from mutations identiﬁed

in the primary tumor. Red circles identify the point of divergence between the metastasis/relapse-seeding clone and the primary tumor. The estimated whole-

genome doubling (WGD) time is indicated by 95% conﬁdence intervals. Numbers in brackets reﬂect the months elapsed between primary tumo r and metastasis

sample acquisition.

See also Figures S1 and S2 and Tables S1,S2,S3,S4, and S5.

Cancer Cell 32, 169–184, August 14, 2017 171

ABC

(legend on next page)

172 Cancer Cell 32, 169–184, August 14, 2017

enabling a single ‘‘tree solution’’ to be derived for each case. In

16 of the 17 cases, all samples studied were clonally related,

as demonstrated by thousands of shared somatic mutations,

with the trunk of the phylogenetic tree representing 12%–

98% of all clustered somatic substitutions (Figures 1 and 2A,

Table S5).

One critical observation emerges: the genomic landscape of

the primary breast cancer at diagnosis is a good surrogate for

the somatic mutations present in disseminated cells at that

moment in time. This conclusion derives from several aspects

of the data. First, the metastatic or relapsing clones branch

late from the phylogenetic lineage of the primary breast lesion,

with relatively few mutations private to the primary tumor. On

average, metastatic divergence occurs at 87% of molecular

time within the primary tumor, estimated from phylogenetic anal-

ysis of base substitutions in regions of the genome with the same

copy number across all lesions (Figure 2A and Table S5). Sec-

ond, as expounded in more detail in the next section, the excess

mutational burden of metachronous metastasis or relapse

clones exceeded that of synchronous axillary lymph node me-

tastases (p = 0.02, F test) and the one synchronous distant

metastasis (PD11458) (Figure 2B). Indeed, synchronous lymph

node metastases are typically very similar to the primary breast

lesion (green branches, Figure 1). Third, driver mutations tend to

be concentrated on the trunk of the phylogenetic tree, notwith-

standing the 1 or 2 additional driver events acquired by relapse

or metastasis clones (explored in considerable detail in later

sections) (Figure 1). Furthermore, whole-genome duplication,

when present, precedes the branching of the recurrence clone

(Figures 1 and S2B). Finally, as we shall see, the mutational pro-

cesses active on the trunk of the phylogenetic tree tend to persist

in the metastasis, suggesting that inferences (and therapeutic

decisions) based on mutational signatures in the primary will

extend to the unseen disseminated cells.

One patient (PD8948), a germline BRCA1 mutation carrier,

was diagnosed with a triple-negative cancer of the left breast,

and over the next 10 years, treated for two apparent local re-

lapses of this lesion and a distant metastasis to the contralateral

breast. In fact, our genomic analyses revealed that the three le-

sions affecting the left breast were clonally unrelated, completely

independent primary cancers, with the second of them seeding

the contralateral breast metastasis (Figure S3). This is important

clinically as the management and prognosis of a second primary

cancer and a local relapse are distinct. This case demonstrates

that genome sequencing can clarify the nature of presumed local

‘‘relapses,’’ especially important in individuals with a genetic pre-

disposition to breast cancer.

Taken together, then, these patterns of disease evolution

strongly support the use of genome sequencing of the primary

breast cancer lesion to underpin decisions about systemic ther-

apy in the adjuvant setting. In modern breast cancer treatment,

the major aim of chemotherapy or estrogen suppression is to

kill those cells that have already spread from the primary lesion,

since surgery and local radiotherapy are usually sufﬁcient to cure

the primary. If it were the case that relapsing or metastatic clones

disseminated early from the primary breast cancer with exten-

sive parallel evolution, as has been suggested previously (Klein,

2009), then targeting somatic mutations found in the primary

would not necessarily have much relevance to disseminated

cells without those changes.

Additional Burden of Mutations in Relapse Samples

For patients with synchronous axillary lymph node metastases,

the number of mutations private to the metastasis was broadly

equivalent to the number private to the primary cancer (Figures

1and 2B). This is perhaps not surprising since, by virtue of being

synchronous lesions, the major lineages in the primary and the

metastasis had the same time period in which to accrue muta-

tions after divergence. In contrast, for the local relapse and

metachronous distant metastasis samples, the relapse carried,

on average, 63% more mutations than the primary tumor,

albeit with considerable variability among patients (range,

24%–244% extra). The number of additional mutations in the

relapse only loosely correlated with the time elapsed between

diagnosis of the primary cancer and relapse (Pearson’s correla-

tion R = 0.29) (Figure 2C).

The additional mutation burden in the later relapse sample was

substantially greater than the chronological time elapsed be-

tween primary and metastasis would suggest, implying that the

rate at which mutations accumulate has typically increased dur-

ing breast cancer evolution. Strikingly, we ﬁnd that the fraction of

additional substitutions, indels, and structural variants in the

Figure 2. Genome-wide Somatic Mutation Timing in 16 Breast Cancers

(A) For each of 17 primary tumor samples, the bar height reﬂects the point in molecular time that the recurrence seeding clone is estimated to diverge from the

primary tumor (relates to phylogenetic trees in Figure 1). Molecular time is determined from the number of base substitutions.

(B) The recurrence-speciﬁc mutation excess is reported in a barplot for each of 18 recurrence samples and in a boxplot split by synchronous (S) and meta-

chronous (M) cases, where the box represents the interquartile range (IQR) bisected by the median, whiskers represent the maximum and minimum range of the

data that do not exceed 1.53the IQR while outlier data points extend beyond this. The recurrence -speciﬁc mutation excess indicates the base substitution load

in branches private to the recurrence minus those in branches private to the primary tumor, presented as a percentage of all substitutions identiﬁed in the primary

tumor. The p value is generated by an F test.

(C) The recurrence-speciﬁc mutation excess as presented in (B) according to the time from primary tumor diagnosis and acquisition of the relapse sample, each

dot represents a patient. R = Pearson’s correlation coefﬁcient.

(D) Scatterplots compare the proportion of each the major mutation types, indels (insertions and deletion), substitutions (Subs), and structural variants (SVs),

localized to the recurrence. Unlike (A) and (B), these ﬁgures include variants in regions that were variable in copy number across samples.

(E) Radiation mutation signature at relapse following external beam radiation. The mutation spectrum of an outlier sample (PD11461) highlighted by a dashed gray

circle in (D) is shown in detail. The overall contribution of indels and structural variants (SVs) outweighs that of substitutions at relapse (top left barplot). Wi thin this

sample, indels of greater lengths (bottom left barplot) and inversions and translocations (bottom, middle bar plot) are relatively more common after relapse.

Cohort-wide, the relative contribution of deletions as opposed to insertions (top right barplot) and of deletions of 5 base pairs (bp) or longer (bottom right barplot)

are reported.*p < 0.0001 (Fisher’s exact test) for enrichment in the relapse sample. Cases exposed to prior external beam radiotherapy are indicated by a star

symbol indicating that other samples do not seem to carry the same signature.

See also Figures S5 and S6.

180 Cancer Cell 32, 169–184, August 14, 2017

C D

(legend on next page)

Cancer Cell 32, 169–184, August 14, 2017 181

disease (Demeulemeester et al., 2016). Although the genome of

a metastatic clone may be similar to the primary tumor at ﬁrst

diagnosis, by the time it has expanded to be clinically detectable,

extensive further genomic changes have occurred.

Whether patients presenting with distant metastatic disease

should have that metastasis biopsied or not to decide on thera-

peutic interventions is a controversial question (Arnedos et al.,

2015). Many sites of metastatic disease are challenging and

invasive to sample, demonstrated by the bias seen in our cohort

toward sites of disease that are easy to access. Our data indicate

that metastases seeded by hematogenous spread do continue

to evolve after dissemination, acquiring many new somatic mu-

tations and key driver mutations. A recent study from a large ter-

tiary referral unit has shown that many patients with metastatic

breast cancer are willing to undergo biopsy of recurrent lesions

for molecular proﬁling (Zehir et al., 2017).

In its restless search for a genome ideally suited to autono-

mous life in far-ﬂung regions of the body, a breast cancer can ac-

cess many different mutational processes and a wide repertoire

of cancer genes. The result is considerable patient-to-patient

variability in genomic proﬁles, even more pronounced than the

already daunting levels seen in primary breast cancer. Mapping

this complexity will require recruitment of large, prospective co-

horts of patients with metastatic disease and integration with

transcriptional, epigenomic, and clinical readouts. Our data

show that such an endeavor would have potential clinical

impact, providing insights into patterns of clonal evolution,

mechanisms of therapy failure, and pathways that could repre-

sent new therapeutic targets.

STAR+METHODS

Detailed methods are provided in the online version of this paper

and include the following:

dKEY RESOURCES TABLE

dCONTACT FOR REAGENT AND RESOURCE SHARING

dEXPERIMENTAL MODEL AND SUBJECT DETAILS

BSubjects, Samples and Consent

BWhole Genome ‘Triplet’ Cohort

BRelapsed Breast Cancer Cohort

BPrimary Breast Cancer (Comparison) Cohort

dMETHOD DETAILS

BSample Size

BTumor Specimen Processing

BDNA Extraction

BMulti-Sample Whole Genome Sequencing

BMulti-Region Targeted Gene Screen

BMulti-Sample Mutation Calling

BMutation Validation

BCancer Gene Discovery

BDriver Mutation Annotation

BGenome-Wide Subclonal Copy Number Analyses

BGenome-Wide Multi-Sample Clonality Analyses

BMutation Timing in Multi-Sample Analyses

BWhole Genome Duplication Timing Analysis

BDriver Mutation Enrichment Analyses

BMutational Signature Analysis

BTelomere Length Estimates

BCase PD8948 and Whole Genome Sequencing and

Analysis of an FFPE Sample

dQUANTIFICATION AND STATISTICAL ANALYSIS

dDATA AND SOFTWARE AVAILABILITY

SUPPLEMENTAL INFORMATION

Supplemental Information includes six ﬁgures and ﬁve tables and can be found

with this article online at http://dx.doi.org/10.1016/j.ccell.2017.07.005.

AUTHOR CONTRIBUTIONS

Conceptualization, L.R.Y., P.J.C., P.E.L., S.K., and A.T.; Methodology, L.R.Y.,

P.J.C., P.E.L., S.K., D.W., I.M., L.B.A., P.V.L., M.G., S.G., M.R.S., E.S., E.P.,

A.G.L., and J.H.R.F.; Software, D.W., I.M., L.B.A., P.V.L., L.R.Y., G.G., M.G.,

S.G.B., K.R., D.J., A.G.L., and J.H.R.F.; Histopathological analysis, H.K.H.,

P.K.L.; Validation, L.R.Y., S.K., L.M., C.L.; Resources, A.M.S., J.W.M.,

A.L.R., P.E.L., C.S., C.D., P.G., L.R.Y., and A.T.; Data curation, L.R.Y., S.K.,

P.E.L.; Visualization, L.R.Y.; Writing – Original Draft, L.R.Y. and P.J.C. Writing –

Review and Editing, all authors; Supervision, P.J.C. and P.E.L.

ACKNOWLEDGMENTS

This work was supported by the Wellcome Trust. Work within the project was

also supported by the Bergen Research Foundation, the Norwegian Cancer

Society, the Norwegian Research Council, Belgian Cancer Plan-Ministry of

Health, the Breast Cancer Research Foundation and the Brussels Region,

The University of Cambridge, Cancer Research UK, and Hutchison Whampoa

Limited. L.R.Y. is funded by a Wellcome Trust research fellowship. P.J.C. is a

Wellcome Trust Senior Clinical Fellow. Dr. Papaemmanuil is a Josi e Robertson

Investigator and the recipient of a European Hematology Association early

career fellowship. A.M.S was supported by Cancer Genomics Netherlands

(CGC.nl) through a grant from the Netherlands Organisation for Scientiﬁc

Research (NWO). A.G.L. and J.H.R.F. were supported by a Cancer Research

UK Program Grant to Simon Tavare

´(C14303/A17197) We thank Beryl Leirvaag

and Dagﬁnn Ekse of Haukeland University Hospital, Bergen, Norway, for tech-

nical assistance. We also acknowledge and thank all members of the labora-

tory and IT teams within the Cancer Genome Project at the Wellcome Trust

Sanger Institute. The laboratory work performed in Bergen was conducted

in the Mohn Cancer Research Laboratory. This research used resources pro-

vided by the Los Alamos National Laboratory Institutional Computing Pro-

gram, which is supported by the U.S. Department of Energy National Nuclear

Security Administration under contract no. DE-AC52-06NA25396. Research

performed at Los Alamos National Laboratory was carried out under the aus-

pices of the National Nuclear Security Administration of the United States

Department of Energy. This work was supported by the Francis Crick Institute,

Figure 7. JAK-STAT Inactivating Mutations Are Enriched at Relapse

(A) Pencil plots of JAK2 and STAT3 genes annotated with non-synonymous mutations identiﬁed in the relapse cohort (n = 163) and the primary cohort (n = 705).

(B) A case (PD8709) of parallel evolution involving four truncating mutations in JAK2. Response to treatment exposures are documented. SD, stable disea se; PR,

partial response; PD, progressive disease.

(D) Scatter and boxplot of the number of mutations identiﬁed in samples (within the relapse cohort, 163 cases) that do or do not contain a JAK2 or STAT3

truncating mutation. The box represents the interquartile range (IQR) bisected by the median, whiskers represent the maximum and minimum range of the data

that do not exceed 1.53the IQR. p value generated using an F test.

182 Cancer Cell 32, 169–184, August 14, 2017

which receives its core funding from Cancer Research UK (FC001202), the UK

Medical Research Council (FC001202), and the Wellcome Trust (FC001202).

The results published here use data generated by the TCGA Research

Network: http://cancergenome.nih.gov/.

Received: December 21, 2016

Revised: May 13, 2017

Accepted: July 14, 2017

Published: August 14, 2017

REFERENCES

Aas, T., Borresen, A.L., Geisler, S., Smith-Sorensen, B., Johnsen, H., Varhaug,

J.E., Akslen, L.A., and Lonning, P.E. (1996). Speciﬁc P53 mutations are asso-

ciated with de novo resistance to doxorubicin in breast cancer patients. Nat.

Med. 2, 811–814.

Alexandrov, L.B., Nik-Zainal, S., Wedge, D.C., Aparicio, S.A., Behjati, S.,

Biankin, A.V., Bignell, G.R., Bolli, N., Borg, A., Borresen-Dale, A.L., et al.

(2013a). Signatures of mutational processes in human cancer. Nature 500,

415–421.

Alexandrov, L.B., Nik-Zainal, S., Wedge, D.C., Campbell, P.J., and Stratton,

M.R. (2013b). Deciphering signatures of mutational processes operative in hu-

man cancer. Cell Rep. 3, 246–259.

Arnedos, M., Vicier, C., Loi, S., Lefebvre, C., Michiels, S., Bonnefoi, H., and

Andre, F. (2015). Precision medicine for metastatic breast cancer–limitations

and solutions. Nat. Rev. Clin. Oncol. 12, 693–704.

Balko, J.M., Giltnane, J.M., Wang, K., Schwarz, L.J., Young, C.D., Cook, R.S.,

Owens, P., Sanders, M.E., Kuba, M.G., Sanchez, V., et al. (2014). Molecular

proﬁling of the residual disease of triple-negative breast cancers after neoad-

juvant chemotherapy identiﬁes actionable therapeutic targets. Cancer Discov.

4, 232–245.

Balko, J.M., Schwarz, L.J., Luo, N., Estrada, M.V., Giltnane, J.M., Davila-

Gonzalez, D., Wang, K., Sanchez, V., Dean, P.T., Combs, S.E., et al. (2016).

Triple-negative breast cancers with ampliﬁcation of JAK2 at the 9p24 locus

demonstrate JAK2-speciﬁc dependence. Sci. Transl. Med. 8, 334ra53.

Banerji, S., Cibulskis, K., Rangel-Escareno, C., Brown, K.K., Carter, S.L.,

Frederick, A.M., Lawrence, M.S., Sivachenko, A.Y., Sougnez, C., Zou, L.,

et al. (2012). Sequence analysis of mutations and translocations across breast

cancer subtypes. Nature 486, 405–409.

Behjati, S., Gundem, G., Wedge, D.C., Roberts, N.D., Tarpey, P.S., Cooke,

S.L., Van Loo, P., Alexandrov, L.B., Ramakrishna, M., Davies, H., et al.

(2016). Mutational signatures of ionizing radiation in second malignancies.

Nat. Commun. 7, 12605.

Berns, E.M., Klijn, J.G., van Putten, W.L., de Witte, H.H., Look, M.P., Meijer-

van Gelder, M.E., Willman, K., Portengen, H., et al. (1998). p53 protein

accumulation predicts poor response to tamoxifen therapy of patients with

recurrent breast cancer. J. Clin. Oncol. 16, 121–127.

Beroukhim, R., Mermel, C.H., Porter, D., Wei, G., Raychaudhuri, S., Donovan,

J., Barretina, J., Boehm, J.S., Dobson, J., Urashima, M., et al. (2010). The land-

scape of somatic copy-number alteration across human cancers. Nature 463,

899–905.

Brastianos, P.K., Carter, S.L., Santagata, S., Cahill, D.P., Taylor-Weiner, A.,

Jones, R.T., Van Allen, E.M., Lawrence, M.S., Horowitz, P.M., Cibulskis, K.,

et al. (2015). Genomic characterization of brain metastases reveals branched

evolution and potential therapeutic targets. Cancer Discov. 5, 1164–1177.

Brown, A.M. (2017). Phylogenetic analysis of metastatic progression in breast

cancer using somatic mutations and copy number aberrations. Nat. Commun.

8, 14944.

Cancer Genome Atlas Network. (2012). Comprehensive molecular portraits of

human breast tumours. Nature 490, 61–70.

Chandarlapaty, S., Chen, D., He, W., Sung, P., Samoila, A., You, D., Bhatt, T.,

Patel, P., Voi, M., Gnant, M., et al. (2016). Prevalence of ESR1 mutations in cell-

free DNA and outcomes in metastatic breast cancer: a secondary analysis of

the BOLERO-2 clinical trial. JAMA Oncol. 2, 1310–1315.

Ciriello, G., Gatza, M.L., Beck, A.H., Wilkerson, M.D., Rhie, S.K., Pastore, A.,

Zhang, H., McLellan, M., Yau, C., Kandoth, C., et al. (2015). Comprehensive

molecular portraits of invasive lobular breast cancer. Cell 163, 506–519.

Costello, M., Pugh, T.J., Fennell, T.J., Stewart, C., Lichtenstein, L., Meldrim,

J.C., Fostel, J.L., Friedrich, D.C., Perrin, D., Dionne, D., et al. (2013).

Discovery and characterization of artifactual mutations in deep coverage tar-

geted capture sequencing data due to oxidative DNA damage during sample

preparation. Nucleic Acids Res. 41, e67.

De Mattos-Arruda, L., Weigelt, B., Cortes, J., Won, H.H., Ng, C.K., Nuc iforo, P.,

Bidard, F.C., Aura, C., Saura, C., Peg, V., et al. (2014). Capturing intra-tumor

genetic heterogeneity by de novo mutation proﬁling of circulating cell-free tu-

mor DNA: a proof-of-principle. Ann. Oncol. 25, 1729–1735.

Demeulemeester, J., Kumar, P., Moller, E.K., Nord, S., Wedge, D.C., Peterson,

A., Mathiesen, R.R., Fjelldal, R., Zamani Esteki, M., Theunis, K., et al. (2016).

Tracing the origin of disseminated tumor cells in breast cancer using single-

cell sequencing. Genome Biol. 17, 250.

Ding, L., Ellis, M.J., Li, S., Larson, D.E., Chen, K., Wallis, J.W., Harris, C.C.,

McLellan, M.D., Fulton, R.S., Fulton, L.L., et al. (2010). Genome remodelling

in a basal-like breast cancer metastasis and xenograft. Nature 464, 999–1005.

Ellis, M.J., Ding, L., Shen, D., Luo, J., Suman, V.J., Wallis, J.W., Van Tine, B.A.,

Hoog, J., Goiffon, R.J., Goldstein, T.C., et al. (2012). Whole-genome analysis

informs breast cancer response to aromatase inhibition. Nature 486, 353–360 .

Futreal, P.A., Coin, L., Marshall, M., Down, T., Hubbard, T., Wooster, R.,

Rahman, N., and Stratton, M.R. (2004). A census of human cancer genes.

Nat. Rev. 4, 177–183.

Gao, R., Davis, A., McDonald, T.O., Sei, E., Shi, X., Wang, Y., Tsai, P.C.,

Casasent, A., Waters, J., Zhang, H., et al. (2016). Punctuated copy number

evolution and clonal stasis in triple-negative breast cancer. Nat. Genet. 48,

1119–1130.

Gellert, P., Segal, C.V., Gao, Q., Lopez-Knowles, E., Martin, L.A., Dodson, A.,

Li, T., Miller, C.A., Lu, C., Mardis, E.R., et al. (2016). Impact of mutational pro-

ﬁles on response of primary oestrogen receptor-positive breast cancers to

oestrogen deprivation. Nat. Commun. 7, 13294.

Gibson, W.J., Hoivik, E.A., Halle, M.K., Taylor-Weiner, A., Cherniack, A.D.,

Berg, A., Holst, F., Zack, T.I., Werner, H.M., Staby, K.M., et al. (2016). The

genomic landscape and evolution of endometrial carcinoma progression

and abdominopelvic metastasis. Nat. Genet. 48, 848–855.

Harris, L.N., Ismaila, N., McShane, L.M., Andre, F., Collyar, D.E., Gonzalez-

Angulo, A.M., Hammond, E.H., Kuderer, N.M., Liu, M.C., Mennel, R.G., et al.

(2016). Use of biomarkers to guide decisions on adjuvant systemic therapy

for women with early-stage invasive breast cancer: American Society of

Clinical Oncology clinical practice guideline. J. Clin. Oncol. 34, 1134–1150.

He, F., Li, J., Xu, J.F., Zhang, S., Xu, Y.P., Zhao, W.X., Yin, Z.Y., and Wang,

X.M. (2015). Decreased expression of ARID1A associates with poor prognosis

and promotes metastases of hepatocellular carcinoma. J. Exp. Clin. Cancer

Res. 34, 47.

Hoadley, K.A., Siegel, M.B., Kanchi, K.L., Miller, C.A., Ding, L., Zhao, W., He,

X., Parker, J.S., Wendl, M.C., Fulton, R.S., et al. (2016). Tumor evolution in two

patients with basal-like breast cancer: a retrospective genomics study of mul-

tiple metastases. PLoS Med. 13, e1002174.

Howie, B.N., Donnelly, P., and Marchini, J. (2009). A ﬂexible and accurate ge-

notype imputation method for the next generation of genome-wide association

studies. PLoS Genet. 5, e1000529.

Juric, D., Castel, P., Grifﬁth, M., Grifﬁth, O.L., Won, H.H., Ellis, H., Ebbesen,

S.H., Ainscough, B.J., Ramu, A., Iyer, G., et al. (2015). Convergent loss of

PTEN leads to clinical resistance to a PI(3)Kalpha inhibitor. Nature 518,

240–244.

Kandoth, C., McLellan, M.D., Vandin, F., Ye, K., Niu, B., Lu, C., Xie, M., Zhang,

Q., McMichael, J.F., Wyczalkowski, M.A., et al. (2013). Mutational landscape

and signiﬁcance across 12 major cancer types. Nature 502, 333–339.

Katagiri, A., Nakayama, K., Rahman, M.T., Rahman, M., Katagiri, H.,

Nakayama, N., Ishikawa, M., Ishibashi, T., Iida, K., Kobayashi, H., et al.

(2012). Loss of ARID1A expression is related to shorter progression-free

Cancer Cell 32, 169–184, August 14, 2017 183

survival and chemoresistance in ovarian clear cell carcinoma. Mod. Pathol. 25,

282–288.

Klein, C.A. (2009). Parallel progression of primary tumours and metastases.

Nat. Rev. Cancer 9, 302–312.

Lawrence, M.S., Stojanov, P., Mermel, C.H., Robinson, J.T., Garraway, L.A.,

Golub, T.R., Meyerson, M., Gabriel, S.B., Lander, E.S., and Getz, G. (2014).

Discovery and saturation analysis of cancer genes across 21 tumour types.

Nature 505, 495–501.

Martincorena, I., Roshan, A., Gerstung, M., Ellis, P., Van Loo, P., McLaren, S.,

Wedge, D.C., Fullam, A., Alexandrov, L.B., Tubio, J.M., et al. (2015). Tumor

evolution. High burden and pervasive positive selection of somatic mutations

in normal human skin. Science 348, 880–886.

Miller, C.A., Gindin, Y., Lu, C., Grifﬁth, O.L., Grifﬁth, M., Shen, D., Hoog, J., Li,

T., Larson, D.E., Watson, M., et al. (2016). Aromatase inhibition remodels

the clonal architecture of estrogen-receptor-positive breast cancers. Nat.

Commun. 7, 12498.

Ng, C.K., Martelotto, L.G., Gauthier, A., Wen, H.C., Piscuoglio, S., Lim, R.S.,

Cowell, C.F., Wilkerson, P.M., Wai, P., Rodrigues, D.N., et al. (2015). Intra-tu-

mor genetic heterogeneity and alternative driver genetic alterations in breast

cancers with heterogeneous HER2 gene ampliﬁcation. Genome Biol. 16, 107.

Nik-Zainal, S., Davies, H., Staaf, J., Ramakrishna, M., Glodzik, D., Zou, X.,

Martincorena, I., Alexandrov, L.B., Martin, S., Wedge, D.C., et al. (2016).

Landscape of somatic mutations in 560 breast cancer whole-genome se-

quences. Nature 534, 47–54.

Nik-Zainal, S., Van Loo, P., Wedge, D.C., Alexandrov, L.B., Greenman, C.D.,

Lau, K.W., Raine, K., Jones, D., Marshall, J., Ramakrishna, M., et al. (2012).

The life history of 21 breast cancers. Cell 149, 994–1007.

Patch, A.M., Christie, E.L., Etemadmoghadam, D., Garsed, D.W., George, J.,

Fereday, S., Nones, K., Cowin, P., Alsop, K., Bailey, P.J., et al. (2015). Whole-

genome characterization of chemoresistant ovarian cancer. Nature 521,

489–494.

Petitjean, A., Mathe, E., Kato, S., Ishioka, C., Tavtigian, S.V., Hainaut, P., and

Olivier, M. (2007). Impact of mutant p53 functional properties on TP53 muta-

tion patterns and tumor phenotype: lessons from recent developments in

the IARC TP53 database. Hum. Mutat. 28, 622–629.

Purdom, E., Ho, C., Grasso, C.S., Quist, M.J., Cho, R.J., and Spellman, P.

(2013). Methods and challenges in timing chromosomal abnormalities within

cancer samples. Bioinformatics 29, 3113–3120.

Robinson, D.R., Wu, Y.M., Vats, P., Su, F., Lonigro, R.J., Cao, X., Kalyana-

Sundaram, S., Wang, R., Ning, Y., Hodges, L., et al. (2013). Activating ESR1

mutations in hormone-resistant metastatic breast cancer. Nat. Genet. 45,

1446–1451.

Savas, P., Teo, Z.L., Lefevre, C., Flensburg, C., Caramia, F., Alsop, K.,

Mansour, M., Francis, P.A., Thorne, H.A., Silva, M.J., et al. (2016). The subclo-

nal architecture of metastatic breast cancer: results from a prospective com-

munity-based rapid autopsy program ‘‘CASCADE’’. PLoS Med. 13, e1002204.

Shah, S.P., Morin, R.D., Khattra, J., Prentice, L., Pugh, T., Burleigh, A.,

Delaney, A., Gelmon, K., Guliany, R., Senz, J., et al. (2009). Mutational evolu-

tion in a lobular breast tumour proﬁled at single nucleotide resolut ion. Nature

461, 809–813.

Shah, S.P., Roth, A., Goya, R., Oloumi, A., Ha, G., Zhao, Y., Turashvili, G., Ding,

J., Tse, K., Haffari, G., et al. (2012). The clonal and mutational evolution spec-

trum of primary triple-negative breast cancers. Nature 486, 395–399.

Stephens, P.J., Greenman, C.D., Fu, B., Yang, F., Bignell, G.R., Mudie, L.J.,

Pleasance, E.D., Lau, K.W., Beare, D., Stebbings, L.A., et al. (2011). Massive

genomic rearrangement acquired in a single catastrophic event during cancer

development. Cell 144, 27–40.

Stephens, P.J., Tarpey, P.S., Davies, H., Van Loo, P., Greenman, C., Wedge,

D.C., Nik-Zainal, S., Martin, S., Varela, I., Bignell, G.R., et al. (2012). The land-

scape of cancer genes and mutational processes in breast cancer. Nature 486,

400–404.

Tevaarwerk, A.J., Gray, R.J., Schneider, B.P., Smith, M.L., Wagner, L.I.,

Fetting, J.H., Davidson, N., Goldstein, L.J., Miller, K.D., and Sparano, J.A.

(2013). Survival in patients with metastatic recurrent breast cancer after adju-

vant chemotherapy: little evidence of improvement over the past 30 years.

Cancer 119, 1140–1148.

Toy, W., Shen, Y., Won, H., Green, B., Sakr, R.A., Will, M., Li, Z., Gala, K.,

Fanning, S., King, T.A., et al. (2013). ESR1 ligand-binding domain mutations

in hormone-resistant breast cancer. Nat. Genet. 45, 1439–1445.

Turner, K.M., Deshpande, V., Beyter, D., Koga, T., Rusert, J., Lee, C., Li, B.,

Arden, K., Ren, B., Nathanson, D.A., et al. (2017). Extrachromosomal onco-

gene ampliﬁcation drives tumour evolution and genetic heterogeneity.

Nature 543, 122–125.

Turner, N., Pearson, A., Sharpe, R., Lambros, M., Geyer, F., Lopez-Garcia,

M.A., Natrajan, R., Marchio, C., Iorns, E., Mackay, A., et al. (2010). FGFR1

ampliﬁcation drives endocrine therapy resistance and is a therapeutic target

in breast cancer. Cancer Res. 70, 2085–2094.

Van Loo, P., Nordgard, S.H., Lingjaerde, O.C., Russnes, H.G., Rye, I.H., Sun,

W., Weigman, V.J., Marynen, P., Zetterberg, A., Naume, B., et al. (2010). Allele-

speciﬁc copy number analysis of tumors. Proc. Natl. Acad. Sci. USA 107,

16910–16915.

Wang, Y., Waters, J., Leung, M.L., Unruh, A., Roh, W., Shi, X., Chen, K.,

Scheet, P., Vattathil, S., Liang, H., et al. (2014). Clonal evolution in breast can-

cer revealed by single nucleus genome sequencing. Nature 512, 155–160.

Yates, L.R., Gerstung, M., Knappskog, S., Desmedt, C., Gundem, G., Van Loo,

P., Aas, T., Alexandrov, L.B., Larsimont, D., Davies, H., et al. (2015). Subclonal

diversiﬁcation of primary breast cancer revealed by multiregion sequencing.

Nat. Med. 21, 751–759.

Zaretsky, J.M., Garcia-Diaz, A., Shin, D.S., Escuin-Ordinas, H., Hugo, W., Hu-

Lieskovan, S., Torrejon, D.Y., Abril-Rodriguez, G., Sandoval, S., Barthly, L.,

et al. (2016). Mutations associated with acquired resistance to PD-1 blockade

in melanoma. N. Engl. J. Med. 375, 819–829.

Zehir, A., Benayed, R., Shah, R.H., Syed, A., Middha, S., Kim, H.R., Srinivasan,

P., Gao, J., Chakravarty, D., Devlin, S.M., et al. (2017). Mutational landscape of

metastatic cancer revealed from prospective clinical sequencing of 10,000 pa-

tients. Nat. Med. 23, 703–713.

184 Cancer Cell 32, 169–184, August 14, 2017

STAR+METHODS

KEY RESOURCES TABLE

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulﬁlled by the Lead Contact, Peter J

Campbell (pc8@sanger.ac.uk).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Subjects, Samples and Consent

All samples included in this project were obtained with informed patient consent and handled in line with the wider framework and

approval for the Breast Cancer Genome Analyses for the International Cancer Genome Consortium Working Group led by the Well-

come Trust Sanger Institute, Cambridgeshire, UK, REC reference: 09/H0306/36. We performed MPS and analysis of a total of 299

tumor samples collected from 170 individual’s breast cancers and 87 matched normal, germline samples (Table S1). Three patients

were male and the remainder female, the average age at primary tumor diagnosis was 53 years (range 30-85 years). Clinical details

including tumor stage, histological features and hormone receptor status are summarized in Table S1. Clinical follow-up data was

available for 96% of patients. The cohort reﬂects a very poor prognostic group of patients whereby 96% of these patients were diag-

nosed during their disease course witheither distant metastatic disease (86%), very poorly controlled locoregional disease not

amenable to surgical resection (10% of cases) or both (7%).

Whole Genome ‘Triplet’ Cohort

To permit phylogenetic analysis of the progression from primary tumor to metastasis a total of 39 fresh frozen tumor samples were

collected from 17 females and subjected to whole genome sequencing. Samples were obtained from Dana-Farber Cancer Institute,

Boston, US (7 cases); Kings College Hospital, London, UK (4 cases); The Erasmus MC Cancer Center, Rotterdam, The Netherlands

(4 cases); The Institute Jules Bordet, Brussels, Belgium (2 cases) in line with local ethics committee approvals (project SHARE’’ #93-

085, approved by the Dana-Farber Harvard Cancer Center institutional review board; UK, REC reference: 10/H0804/33, approved by

Guy’s and St Thomas’ NHS Trust ethics committee; MEC 02.953, approved by the medical ethical committee of the Academisch

Ziekenhuis Rotterdam (EUR/ AZR) for ‘The retrospective assessment of cell biological factors in archival tumor tissues’; Protocol

1698 and 1634, approved by the Institut Jules Bordet local ethics committee). For each individual, in addition to a primary tumor

and a matched normal sample at least one sample from a distinct metastatic scenario was included in the experiment. For 7 individ-

uals, where the metastasis scenario sample was limited to a synchronous lymph node deposit these samples were only included in

the whole genome analysis where they form a comparison cohort. For one patient (PD8948) where the apparent relapse samples

were identiﬁed as distinct primary tumors we collected two additional tumor samples from formalin ﬁxed parafﬁn embedded

(FFPE) tissue blocks and performed targeted capture on both samples and whole genome sequencing on one (Figure S3).

Relapsed Breast Cancer Cohort

The number of cases of locally relapsed and distant metastatic breast cancer was extended from 10 to 163 by including a second

cohort of patients for whom 365 cancer related genes were sequenced using a targeted capture pulldown approach. Samples from

these individuals form the ‘relapsed breast cancer cohort’. The additional 153 patients were drawn from a single centre study at the

Department of Oncology, Haukeland University Hospital, performed with the aim of identifying genetic alterations in advanced and

REAGENT or RESOURCE SOURCE IDENTIFIER

Deposited Data

Targeted and whole genome

sequence data

https://www.ebi.ac.uk/ega/ Accession numbers: Targeted (2939stdy)

EGAD00001002698

Exome (492stdy): EGAD00001002697

Whole genome (2040stdy): EGAD00001002696

Somatic Mutation Calls Mendeley Data http://dx.doi.org/10.17632/g7kpzkhz8c.1

Software and Algorithms

The Sanger’s Cancer Genome Project core

somatic calling workﬂow from

the ICGC PanCancer Analysis of Whole

Genomes (PCAWG) project

https://dockstore.org/containers/quay.io/

pancancer/pcawg-sanger-cgp-workﬂow

Cancer Cell 32, 169–184.e1–e7, August 14, 2017 e1

metastatic breast cancer deposits. Between March 1996 and October 2004, a total of 206 patients with non-operable primary breast

cancers, local relapse and / or metastatic deposits suitable for biopsy were recruited to the study. All samples were snap-frozen in

liquid nitrogen in the operating room immediately upon removal from the patient. We analysed a total of 259 tumor samples from 153

patients. Patients included in the analysis had at least one sample from a distant metastatic or locoregional relapse deposit that con-

tained sufﬁcient material for DNA extraction, allowing MPS. For 41 patients included in the study we were able to identify a primary

tumor sample and extract sufﬁcient DNA for MPS (FFPE, n = 29; fresh frozen, n = 12). In addition, DNA was retrieved from FFPE-

blocks from four metastatic deposits undergoing routine biopsy in the time period between primary and fresh frozen metastatic tissue

collection, 2 primary tumors after neo-adjuvant chemotherapy and a synchronous lymph node deposit. The study was approved by

the regional ethics committee of the Norwegian Health Region West (218/97 – 77.97; REK Vest), and all patients provided written

informed consent.

In total, 163 individuals were therefore included within the relapsed breast cancer cohort and for each patient at least one sample

(total number of relapse samples = 227) was obtained from a distant metastatic deposit (n = 79) or a metachronous loco-regional

relapse (n = 148) (Table S1). Multiple relapse samples were collected for 46 individuals (range 2-5 samples per individual). A matched

primary tumor sample was collected in 51 cases and a matched germline sample was collected for 80 cases (adjacent normal breast

tissue, n = 6; blood, n = 74). The distribution of relapse sample sites is presented in Table S1. Most (177/227) relapse and metastasis

samples were pre-treated, being exposed to an average of 1.7 (range 0-5) lines of systemic therapy Table S1. A total of 80 samples

from 57 individuals were also obtained after exposure to external beam radiotherapy.

Primary Breast Cancer (Comparison) Cohort

The primary tumor comparison cohort consisted of previously published exome data from 705 individual’s primary breast cancers,

freely available from The Cancer Genome Atlas (TCGA). We included properly matched samples that were available for download

from CGHub on December 2015. We excluded variants where the matched normal coverage was lower than 10-fold and samples

for which less than 50% of the mutations detected by our calling pipeline were present in the somatic mutation calls released by

TCGA. To minimize bias in our comparisons we applied the same mutation calling algorithms, post-processing ﬁlters and driver

annotation processes as were used for in in-house generated data for the relapse cohort. Annotated mutation data for these samples,

within the scope of the cancer gene panel is available in Table S3. Clinical information for the 705 patients in the primary cohort was

downloaded from https://tcga-data.nci.nih.gov/docs/publications/brca_2012/ (ﬁle = BRCA_Clinical.tar.gz ). A comparison of the

clinical characteristics of the primary cohort and relapse cohort at the point of diagnosis is provided in Table S1. Cancer staging in-

formation for each dataset was determined using the American Joint Committee on Cancer (AJCC) Staging Manual, 7

edition. When

nodal status was recorded as ‘Nx’ within clinical information this is assumed to reﬂect node negative disease (‘N0’).

METHOD DETAILS

Sample Size

The sample size of 163 recurrent breast cancers has 99% power that a cancer gene mutated in 5% of breast cancer recurrences

would be seen in at least 3 patients in the cohort.

Tumor Specimen Processing

All samples within the whole genome and relapse breast cancer cohorts were histopathologically assessed to ensure adequate tu-

mor cellularity (>=70%) and if necessary macrodissection was performed. Where possible for both primary tumor and relapse

samples ER and PgR expression was determined by local pathologists as Allred scores of 4 or above. Where available, HER2

over-expression was determined by IHC scores of 3+ or 2+ conﬁrmed by ﬂuorescent in-situ hybridisation. Due to the historical nature

of the Haukeland University Hospital sample set, HER2 expression data however, is scarce and HER2 ampliﬁcation was determined

from sequence data using the criteria for identifying ampliﬁcations in targeted capture data as described below. We have previously

shown our approach to yield results that are highly consistent with clinical HER2 status results (Yates et al., 2015).

DNA Extraction

DNA from fresh frozen tumor tissue specimens and blood samples was isolated, using spin columns from the QIAamp DNA mini kit

(Qiagen). The procedure was performed according to the manufacturer’s instructions with the exception that 400ul sample (instead of

200ul) was used as input in the cases where full blood on EDTA were used instead of leukocyte concentrates (Haukeland University

Hospital cases). DNA from formalin ﬁxed parafﬁn embedded tissue (FFPE) was isolated, using spin columns from the QIAamp DNA

FFPE Tissue Kit (Qiagen). The procedure was performed according to the manufacturer’s instructions, with the following exceptions:

The de-parafﬁnization step with xylene was repeated three times and the subsequent washing step with ethanol was repeated twice.

Lysis of tissue was performed using 540 ml buffer ATL and 60 ml proteinase K per samples, for 2-4 hours at 56C, before addition of a

further 180 ml buffer ATL and 20 ml proteinase K and an over-night incubation at 56C.

Multi-Sample Whole Genome Sequencing

Genomic libraries with insert sizes of 300bp-600bp were derived from native DNA from 39 tumor and 17 matched normal fresh frozen

samples using Illuminapaired end sample preparation kits according to manufacturers instructions. Following cluster generation,

e2 Cancer Cell 32, 169–184.e1–e7, August 14, 2017

100bp paired-end sequence data was generated using Illumina HiSeqs and was subsequently aligned to the reference human

genome (NCBI build37) using BWA. Whole genome libraries from a single FFPE tumor (PD8948c) and matched fresh frozen normal

sample (PD8948b) were prepared using Agilent Technologies Sure Select library preparation kit (Custom library kit (cat no. 930075)

http://www.agilent.com/search/?Ntt=930075 following manufacturers instructions. 150bp paired end sequence data (with average

insert sizes of 319bp and 481bp respectively) was generated using Illumina X10. The average genome wide sequence coverage of

tumors and matched normal samples was 42 and 31 fold respectively (Table S2).

Multi-Region Targeted Gene Screen

For targeted capture pulldown experiments we used a bait design that consisted of over 8,000 targets of which almost 6,000 covered

the exons of 365 genes. To facilitate copy number analyses baits were also included to target over 2,000 SNPs outside of exonic

locations. Genomic DNA from tumor and matched normal samples, was fragmented using Covaris(average insert size 150bp)

and subjected to IlluminaDNA sequencing library preparation using Agilent’sBravo Automated liquid handling platform. Tumor

and normal samples were indexed with unique barcodes using PCR. Libraries were then hybridised to custom ribonucleic (RNA) baits

according to the AgilentSureSelectprotocol. Samples were multiplexed on average 16 samples per lane and ﬂow-cell clusters

created. Paired-end, 75bp sequence reads were generated using Illumina HiSeq 2000. Sequence data was re-aligned to the human

genome (NCBI build 37) using BWA. Unmapped reads, PCR duplicates and those outside of the target region were excluded from

analysis. The average sequence coverage of tumors and matched normal samples was 467 and 505 fold respectively (Table S2).

Multi-Sample Mutation Calling

Substitutions, indels and structural variant breakpoints were called independently in each tumor sample using mutation calling algo-

rithms (CaVEMan, Pindel and BRASS) and post-processing ﬁlters as previously described (Yates et al., 2015). Mutation calling algo-

rithms used in the analysis are freely available at https://github.com/cancerit/. Where an individual had more than one tumor sample

we performed a comparative analysis of SNP and indel variant data for union of sites from all related samples in an unbiased manner

using in-house software – vafCorrect, that is freely available at https://github.com/cancerit/vafCorrect. For substitutions unbiased

pileup results were obtained using Bio::DB::HTS (https://github.com/Ensembl/Bio-DB-HTS). For indels the approach includes un-

mapped reads whose pair is mapped within the vicinity (deﬁned by library insert size) of the indel site and resulting reads were aligned

using exonerate to original reference sequence and alternate reference sequence (created by inserting the indel variant at the given

reference location). Exonerate output was then parsed to count the fraction of reads aligned to original reference and alternate refer-

ence sequence. Reads that were mapped with equal identity scores to reference and alternate sequence were reported as

ambiguous reads while reads that were present at the variant location but did not map to either of the reference sequences were

categorized as unknown reads. Data quality was ensured and the impact of germline SNP contamination minimized by ﬁltering

against an extended unmatched normal panel of over 200 samples, cross-referencing with available germline SNP databases, using

a matched normal sample where available and visually inspecting local alignments for all reported coding mutations.

Comprehensive lists of all somatic substitutions, indels and structural variants from whole genome analysis are available for down-

load at review@sftpsrv.sanger.ac.uk. All high conﬁdence mutation calls within the scope of the cancer gene panel are presented in

Table S3.

Mutation Validation

For the whole genome experiment, native DNA where available (25 tumor and 14 normal samples) or whole genome ampliﬁed (WGA)

DNA when necessary (samples PD13596a, PD13596b, PD13596c, PD4243a, PD4252c, PD48102a, PD8948d, PD8948e) was sub-

jected to custom capture pulldown and high depth re-sequencing to a target depth of 1000-fold. Probes were designed for 6,534

genome-wide substitutions and indels using Agilent Technologies SureSelect Standard DNA Design Wizard. High-stringency repeat

masking, a tiling density of 2X and balanced boosting were applied to the design. DNA capture (paired-end, average insert size

150bp) libraries were multiplexed and sequenced using Illumina MiSeqto an average coverage of 1,076-fold. We have previously

published validation data for case PD9771 (Yates et al., 2015). To determine an experimental validation rate, all coding indels (n=144)

and substitutions (n=1,498) were included in the experiment. A true positive validation of 94% was identiﬁed for both coding indels

and substitutions independently. Amongst substitutions the most common reason for failure to validate was low coverage (4%) and

this was usually associated with the use of WGA material. Excluding WGA validation experiments was associated with a validation

rate of 97% and this is believed to be a more reliable reﬂection of the true positive rate (Table S4).

The remaining variants included in the high-depth pulldown design were selected to enable validation and reﬁnement of phyloge-

netic tree structures. The approach was biased towards subclonal events and mutations that contradicted the consensus tree.

Mutations close to each other or within 200bp of germline snps were also enriched to permit reconstruction of the subclonal structure

through phasing approaches. Using this approach, across 13 cases (cases PD13596, PD4252, PD4820, PD114780 excluded for rea-

sons stated above), 45 out of 48 tree branches were identiﬁed through an independent clustering experiment, re-capitulating and

therefore validating the basic tree structure in each case. A total of 3 small branches failed to validate and consisted of 2-3% of

the overall mutation burden in each case. See Figure S1 and Table S5 for details of phylogenetic tree construction and validation

in each case.

Regarding multi-sample targeted capture experiments we have previously demonstrated a 99% consistency rate in reporting

mutation presence and absence (Yates et al., 2015) using custom pull down duplicate experiments. Furthermore, we validated

Cancer Cell 32, 169–184.e1–e7, August 14, 2017 e3

non-synonymous mutations in ESR1, JAK2 and PIK3R1 using capillary sequencing (Table S4). One ESR1 mutation was validated in

an independent exome experiment (Brown, 2017).

Cancer Gene Discovery

To identify recurrently mutated driver genes we used dNdScv, as previously described in detail (Nik-Zainal et al., 2016). This method

uses dN/dS and covariates to detect genes with higher density of non-synonymous coding mutations than expected by chance. The

method considers the trinucleotide mutation spectrum, the sequence of each gene, the impact of coding mutations (synonymous,

missense, nonsense, splice sites substitutions and indels) and the variation of the mutation rate across genes. Multiple testing correc-

tion (Benjamini-Hochberg FDR) was applied across analyzed genes and a q value < 0.1 was used to determine statistical signiﬁcance.

For the relapse cohort signiﬁcance was tested across the 311 genes for which at least one mutation was called. The approach was

performed across all relapse samples and across the subset of samples with a matched germline sample. STAT3 was signiﬁcantly

mutated in the matched sample analysis only. Within the exome analysis over 20,000 genes are analyzed.

Driver Mutation Annotation

Each coding variant was manually curated with a likely driver status following a systematic approach. Firstly, likely cancer genes were

identiﬁed as either those found within the dNdS cancer gene discovery approach described above, from published reference mate-

rials consisting of the Cancer Gene Census (Futreal et al., 2004), the Cancer5000 series (Lawrence et al., 2014) or from literature re-

view of breast cancer sequencing studies (Banerji et al., 2012; Cancer Genome Atlas Network, 2012; Ellis et al., 2012; Shah et al.,

2012; Stephens et al., 2012). Subsequently, oncogenic mutations were annotated within these cancer genes. Oncogenic mutations

were deﬁned as those falling into one of the following categories: 1) A canonical oncogenic mutation in a recurrent mutation hotspot;

2) A lower frequency recurrent mutation in a known oncogene with 3 or more conﬁrmed somatic non-synonymous substitutions or in-

frame deletions previously reported at this locus in COSMIC or conﬁrmed through experimental models or special cohorts (i.e. ESR1

resistance mutations); 3) Likely damaging events in a known tumour suppressor that include truncating (nonsense), frame-shift,

essential splice variants or those within a mutation hotspot (>=2 somatic mutations); 4) Silent mutations in a known recurrent splice

site hotspot.

Genome-Wide Subclonal Copy Number Analyses

Segmental copy number information was derived from all targeted capture and whole genome data using the Allele Speciﬁc Copy

Number Analysis of Tumors (ASCAT) algorithm (Van Loo et al., 2010). The Battenberg algorithm was used to identify clonal and sub-

clonal copy number changes in whole genome sequence data as previously described (Nik-Zainal et al., 2012; Yates et al., 2015) and

was also used to challenge and conﬁrm copy number and ploidy estimates derived from ASCAT. The approach phases germline

SNPs within MPS data using Impute2 (Howie et al., 2009) that uses a well characterized panel of polymorphic SNPs.

Within whole genome data, copy number segments are reported as ampliﬁed when present at more than twice the estimated

average ploidy across the whole genome. Homozygous deletions are identiﬁed as segments where total copy number equals

zero or equivalent in an area of subclonal copy number. Within targeted capture data the mean logR and 95% conﬁdence interval

was calculated across known cancer driver genes. Potential ampliﬁcations in common breast cancer genes were identiﬁed based

on a mean logR of > 1, equating to 6 alleles in a diploid genome and tumor cellularity of 50%. For related samples where heterogeneity

of ampliﬁcation events was called logR and BAFs across all genes and point mutation data were reviewed manually in each related

sample to determine if heterogeneity is likely a consequence of low aberrant cell fraction as opposed to true driver heterogeneity. This

conservative approach was adopted to minimize the risk of over-calling heterogeneity.

Genome-Wide Multi-Sample Clonality Analyses

For the 17 patients with multi-sample whole genome sequencing data, to model the subclonal structure across multiple related sam-

ples previously described bioinformatics and deductive reasoning approaches were adopted. The approach follows 3 main steps

including the identiﬁcation of large-scale subclonal copy number changes using the Battenberg algorithm (Nik-Zainal et al., 2012),

clustering of subclonal somatic substitutions using a Bayesian Dirichlet process in multiple dimensions across related samples

and hierarchical ordering across multiple samples using the ‘pigeon hole principle’. Strict quality control is applied to the mutations

included in clustering analysis to avoid the generation of false positive clusters of mutations:

dDuring evolution, copy number losses may result in the loss of mutations in the affected regions, resulting in clusters of muta-

tions found uniquely in the unaffected sample(s). In order to avoid falsely calling such mutations as arising from a clonal expan-

sion in the unaffected samples, such mutations are excluded from Dirichlet process clustering.

dSome mutations may be present in multiple samples, but only called in a subset of samples, due to low allele frequency in the

other sample(s). To avoid false negatives, allele frequencies of all mutations found in any sample from a patient are therefore re-

called, with a minimum mapping quality and base quality of 10.

dThe allele frequencies of all mutations are adjusted to cancer cell fraction using purity and copy number information. Copy num-

ber segments have start and end points deﬁned by heterozygous SNP locations, so somatic variants that fall between these

boundaries have undeﬁned copy number and are excluded from clustering.

e4 Cancer Cell 32, 169–184.e1–e7, August 14, 2017

A median of 95% (range 77 – 99%) of mutations are included in clustering. Using this approach each substitution that passed qual-

ity control was assigned to a speciﬁc cluster (Tables S5). For each individual case, data including cluster size (equating to phyloge-

netic tree branch length), cluster ‘position’ (reﬂecting the proportion of cells containing the mutation cluster in each related sample)

and posterior conﬁdence intervals are presented for both discovery and validation experiments in Table S5 and Figure S1.

Mutation Timing in Multi-Sample Analyses

The relative contribution of the different mutation types during evolution (Figure 2) was estimated by comparing the proportion of

mutations that were shared, private to the primary tumor sample or private to the metastasis/ relapse sample. Each individual point

mutation was assigned to one of these categories by calculating for each mutation, in all related samples independently, R’s pbino-

mial statistic based upon a conservative, expected error rate of 1 in 200. A mutation was deemed to be present or absent from an

individual sample based upon a p value of <=0.05 or > 0.05 respectively. All structural variants reconstructed in silico were deter-

mined to be shared or private to the primary/ metastasis samples based upon either reconstruction in related samples or the pres-

ence of 4 or more split reads supporting the breakpoint using BRASS1. Substitution branch timing (Figures 1 and 2) was calculated

using mutation clustering where the cluster size dictates the branch length.

Whole Genome Duplication Timing Analysis

In this study we have estimated the prevalence of 3 different developmental stages for 22 of the breast cancer samples. The ﬁrst one

corresponds to the diploid stage previous to whole genome duplication. The second one is the tetraploid cell stage after the whole

genome duplication was acquired and previous to the subclonal diversiﬁcation. Lastly, the timing between the last selective sweep

and the emergence of the detected subclones. The duration of each of the stages in molecular time is estimated via the fraction of

mutations having arisen in each of the phases. To estimate the proportions of mutations in each stage we employ a strategy similar to

that of Purdom et al. (Purdom et al., 2013) and extend it to subclonal mutations.

Let rdenote the purity of the sample. The expected variant allele frequency f

for a mutation arising in state I depends on the number

mutated alleles m

, the total copy number c (4 in our case) and the prevalence of the subclone p

. For early clonal mutations we have

= 1 and m

= 2, for late clonal mutations we have m

= 1. For subclonal mutations we have p

< 1 and m

=1.

f=rmipi

4r+2ð1rÞ

We model the number of reads X arising from a mutation in stage I as a binomial with coverage n.

XjiBinom ðn;fiÞ

The probability that a mutation occurs in stage I is p

. This gives rise to a binomial mixture model.

PðX;iÞ=PðXjiÞ3pi

Using Bayes’ formula we can compute the probability of being in state I given X as

PðijXÞ=PðX;iÞ

PðXÞ=PðXjiÞ3pi

PiPðXjiÞ3pi

For a series of k observed mutations with variant reads x

,.x

, we can estimate the mixture proportions p

using and EM

algorithm.

Knowing the probabilities p

, for early (p

) and late (p

) stages we can calculate an estimate the relative time of WGD as:

t=2pe

2pe+pl

To assess the robustness of the above estimator and to calculate conﬁdence intervals we use bootstrapping, subsampling 100

times from the number of observed mutations with replacement and calculating t for each of the subsamples.

Presented analyses were ﬁrst applied to all mutations within individual samples with results being consistent with duplication

arising prior to primary-relapse divergence. A more accurate estimate of the timing of whole genome duplication was then deter-

mined by restricting the analysis to shared, clonal mutations allocated to the trunk of the phylogenetic tree.

Driver Mutation Enrichment Analyses

The frequency with which each cancer gene (ESR1 or genes signiﬁcantly mutated in the driver discovery experiment) was altered by a

driver mutation was compared between the relapsed and primary breast cancer cohorts using a two-sided Fishers’s exact test. A

Benjamini-Hochberg correction for multiple testing was applied to generate false discovery rates (q). A total of 7 genes were signif-

icantly enriched in the relapsed compared to the primary cohort (deﬁned by q < 0.1) while no genes were enriched in the primary

cohort.

Cancer Cell 32, 169–184.e1–e7, August 14, 2017 e5

Mutational Signature Analysis

We assessed the relative activity of mutational processes over time by allocating somatic mutations to their speciﬁc branch of the

phylogenetic tree and subjecting individual branches (composed of more than 20 mutations) to mutational signature analysis (Fig-

ure 3). Mutational signatures were detected in two independent ways: (i) de novo extraction based on somatic substitutions and their

immediate sequence context and (ii) reﬁtting of previously identiﬁed consensus signatures of mutational processes. The de novo

extraction was performed using a previously developed theoretical model and its corresponding computational framework (Alexan-

drov et al., 2013b). Brieﬂy, the algorithm deciphers the minimal set of mutational signatures that optimally explains the proportion of

each mutation type in each mutational catalogue and then estimates the contribution of each signature to each sample. Within this

dataset the computational framework identiﬁed ﬁve reproducible mutational signatures that closely resembled previously identiﬁed

breast cancer signatures.

In the second stage, 27 distinct consensus mutational signatures previously identiﬁed from examining 7,042 samples across 30

different cancer types were ‘reﬁtted’ (Alexandrov et al., 2013a). All possible combinations of up to seven mutational signatures

were evaluated for each sample. This resulted in 1,285,623 solutions per sample and a model selection was applied to select the

optimal solution. The model selection framework excludes any solution in which a mutational signature contributes less that 2%

of the somatic mutations or less than 50 somatic mutations. Exceptions were made for Signatures 1 and 5 as these are believed

to reﬂect on-going endogenous mutational processes that continuously contribute very low numbers of somatic mutations (Alexan-

drov et al., 2013a). Further, the model selection framework selects the solution that optimizes the Pearson correlation between the

original pattern of somatic mutations and the one based on reﬁtting the sample with consensus mutational signatures such that each

additional signature should improve the Pearson correlation with at least 0.02. The ﬁnal solution for each sample contained between

3-6 mutational signatures and these signatures were consistent with the ones previously identiﬁed by the de novo analysis: Signature

1, Signature 2, Signature 3, Signature 5, Signature 8, and Signature 13.

Telomere Length Estimates

Telomerecat is a de novo method for the estimation of telomere length from whole genome sequencing samples. The algorithm works

by comparing the ratio of complete telomere reads to reads on the boundary between telomere and subtelomere. The ratio is trans-

formed to a measure of length using a simulation approach that takes into account the fragment length distribution of the sample. By

considering the ratio of complete telomere reads to boundary reads, Telomerecat estimates coverage over the telomere without

interface from the affects of aneuploidy, a common occurrence in cancer. Telomerecat also corrects for error in sequencing reads

by modeling the observed distribution of phred scores associated with mismatches to the telomere sequence.

Case PD8948 and Whole Genome Sequencing and Analysis of an FFPE Sample

For all but one cancer in the dataset we found that thousands or tens of thousands of somatic substitutions were shared by the pri-

mary and metastasis sample. In one case (PD8948) however, we determined from the clonal mapping of over 16,000 somatic sub-

stitutions, indels and structural variants that the two fresh frozen DNA samples from tumors in the left and right breast (samples

PD8948d and PD8948e respectively) sampled 1 year apart were clonally unrelated cancers. Only 95 (0.6%) point mutations were

detected in both samples, none of which fell within coding regions, and validation through visual inspection and/ or targeted capture

pulldown failed to identify any mutation as a true positive in both samples. Copy number proﬁles and structural variant proﬁles from

the two cancers were also distinct. In the absence of shared somatic events we conclude that these samples are derived from 2 in-

dependent primary tumors. The samples however, shared thousands of germline SNPs and a BRCA1 frame-shift mutation conﬁrm-

ing that they are derived from the same individual who was a known germline BRCA1 mutation carrier.

To further explore the clonal evolution of this patient’s cancers we identiﬁed 2 additional FFPE samples from earlier tumor deposits

within the left breast (PD8948a and PD8948c). These samples were subjected to targeted gene panel sequencing and the likely

phylogenetic relationships between the four samples were then inferred from coding non-synonymous mutations as demonstrated

in Figure S3A. The ﬁndings were consistent with the patient having developed 3 separate primary tumors during her lifetime, each

containing a distinct TP53 mutation. Two samples (PD8948c and PD8948e) harbored identical TP53 (p.Y220C) and KDM6A muta-

tions suggesting that the later sample represented distant relapse. Genome-wide analysis could provide conclusive evidence to

conﬁrm the relatedness of two such samples, however one sample (PD8948c) was derived from a 7 year old FFPE sample and there

is little experience of whole genome sequencing of FFPE derived tumors. Weand others have previously shown low error rates for

gene capture and whole exome sequencing of FFPE samples but to date we are only aware of the results from a single tumor sample

sequenced to whole genome level and the widespread applicability of this single case is unclear. We predicted that the process of

ﬁxation and storage of such material could result in the introduction of technical artifacts and could compromise mutation-calling

sensitivity. However, for the purposes of this experiment identifying a signiﬁcant overlap of mutations called in the later sample

(PD8948e) would conﬁrm relatedness of the samples.

Library preparation of the FFPE sample was performed following our standard protocol and the tumor sample and a matched blood

derived normal sample were sequenced to 31X and 38X respectively using Illumina X10. Mutations were called using the same

algorithms as previously described. Conﬁrming the clonal relationship of the two samples, a signiﬁcant proportion of somatic muta-

tions of all classes – substitutions (25%), indels (18%) and structural variants (14%), were shared. This equates to almost 2,000 com-

mon somatic mutations, genome-wide (Figure S3B).

e6 Cancer Cell 32, 169–184.e1–e7, August 14, 2017

Analysis of other whole genome triplet cases within the cohort identiﬁed that all metachronous samples contain a signiﬁcant

excess point mutation burden compared to the primary tumor. However, this was inconsistent in this case, where the presumed pri-

mary tumor (PD8948c, FFPE) and the relapse sample obtained 3 years later (PD8948e, fresh frozen) contained a similar private point

mutation burden (Figure S3B). We investigated whether the unexpected excess of FFPE speciﬁc mutations was a likely biological or

technical phenomenon by comparing the mutation spectra in the 2 samples. The substitution proﬁle within the FFPE sample indi-

cated an excess of C>A base changes and these tended to occur in the context of one or two 5 prime cytosine nucleotides (Fig-

ure S3C). To investigate this further we applied formal mutational signature analysis to the private and shared branches from an

inferred phylogenetic tree derived from these samples (Figure S3D). The analysis conﬁrmed that the shared mutations were drawn

from 3 mutation signatures – two clock-like signatures and a dominant signature associated with homologous recombination deﬁ-

ciency (signature 3), consistent with the known BRCA1 mutation carrier status. Private mutations identiﬁed in the fresh frozen sample

(PD8948e) followed an almost identical signature distribution. In contrast, none of the mutations that were private to the FFPE sample

were assigned to these signatures, but rather were purely assigned to a mutation signature (R2) - a known sequencing artifact that

arises due to oxidative damage and has previously been described in relation to exome library preparation (Costello et al., 2013). In

constructing the phylogenetic tree in Figure 1 we therefore omit a private to primary tumor branch, although it is conceivable that a

small number of true private mutations were undetected due to the presence of an overwhelming artifact.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed and graphics produced using R version 3.0.1: A language and environment for statistical

computing (R Foundation for Statistical Computing, Vienna, Austria. Alignment viewing was performed using Gbrowse, Jbrowse,

Samtoolstview and IGV. All hypothesis tests were 2-sided when appropriate and statistical tests used are speciﬁed in Results

and ﬁgure legends.

DATA AND SOFTWARE AVAILABILITY

Targeted and whole genome sequence data has been deposited at the European Genome-Phenome Archive (https://www.ebi.ac.

uk/ega/ at the EBI) with accession numbers:

dTargeted (2939stdy) EGAD00001002698;

dExome (492stdy): EGAD00001002697;

dWhole genome (2040stdy): EGAD00001002696.

Full somatic mutation calls (substitutions, indels and structural variants) for each individual cancer analyzed by whole genome

sequencing are available for download from Mendeley Data. The link for the dataset is: http://dx.doi.org/10.17632/g7kpzkhz8c.1

The most recent version of our whole genome sequencing mutation pipeline is available as a Docker image. This, together with

documentation, can be accessed from https://dockstore.org/containers/quay.io/pancancer/pcawg-sanger-cgp-workﬂow.

Cancer Cell 32, 169–184.e1–e7, August 14, 2017 e7

Table S1. Clinical Data, Related to Figure 1

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Table S5. Somatic Mutation Clusters, Related to Figure 1

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Table S4. Validation Experiments, Related to Figure 1

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Table S2. Sequencing Metrics, Related to Figure 1

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Document S1. Figures S1–S6

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Document S2. Article plus Supplemental Information

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Supplementary Material 4

Data

August 2017

Lucy R. Yates · Stian Knappskog · David C Wedge · James H.R. Farmery · Peter J. Campbell

Treatment of de novo Bone-Only Oligometastatic Breast Cancer

Preprint

Full-text available

Oct 2024

Introduction It is estimated that 10–25% of breast cancer patients have metastatic disease upon diagnosis with roughly 20% having limited metastatic sites. The optimal management of patients with de novo bone-only oligometastatic breast cancer (OMBC), particularly curative-intent approaches, continues to evolve given extremely limited evidence in survival outcomes with this strategy. Methods This was a single-center retrospective review of the survival outcomes and details of treatment of patients with de novo OMBC who received tri-modality treatment with systemic therapy (chemotherapy with or without endocrine treatment and/or HER2-directed therapy), surgery, and radiotherapy to local and metastatic site/s. Progression-free survival (PFS) was defined as the time from initiation of treatment (either surgery, or neoadjuvant systemic therapy) to date of clinical or radiologic progression. Results From January 2014-March 2024, we identified 10 women who fit the inclusion criteria. Seven had isolated bone metastasis, and none had more than 3 metastatic sites. The included cohort had a mean age of 44.2 years. Nine had hormone receptor-positive disease, and 5 were HER2-positive. All patients were discussed in a multidisciplinary meeting. Median PFS by Kaplan-Meier analysis was 40.2 months. Nine patients were still alive at the time of analysis, and 6 of them remain disease free with a median follow-up duration of 30.8 months. Conclusion Patients with de novo bone-only oligometastatic breast cancer seem to benefit from the standard curative-intent tri-modality approach with the addition of ablative radiation to metastatic sites. These patients have a long median PFS and can be rendered disease-free for many years.

Advances in the Understanding of the Pathogenesis of Triple‐Negative Breast Cancer

Article

Full-text available

Nov 2024

Background Triple‐negative breast cancer (TNBC) is a heterogeneous disease characterized by high aggressiveness, high malignancy, and poor prognosis compared to other breast cancer subtypes. Objective This review aims to explore recent advances in understanding TNBC and to provide new insights and potential references for clinical treatment. Methods We examined current literature on TNBC to analyze molecular subtypes, genetic mutations, signaling pathways, mechanisms of drug resistance, and emerging therapies. Results Findings highlight key aspects of TNBC's molecular subtypes, relevant mutations, and pathways, alongside emerging treatments that target drug resistance mechanisms. Conclusion These insights into TNBC pathogenesis may help guide future therapeutic strategies and improve clinical outcomes for patients with TNBC.

Molecular tumor board in patients with metastatic breast cancer

Article

Full-text available

Oct 2024
BREAST CANCER RES TR

Purpose Comprehensive genomic profiling is becoming increasingly important in the management of patients with metastatic breast cancer (mBC). Real-world clinical outcomes from applying molecular tumor boards (MTBs) recommendations in this context remain limited. Accordingly, we conducted a retrospective, single-institution analysis to evaluate the clinical impact of discussing patients affected by mBC at the MTB. Methods Clinicogenomic data of patients affected by mBCs referred to the European Institute of Oncology MTB between August 2019 and December 2023 were reviewed. Genomic alterations were classified by ESCAT framework. Clinical outcomes of patients showing actionable alterations and receiving molecular-matched therapy (MMT) were compared to those receiving standard therapy (ST). Results Ninety-six patients were included. Following MTB discussion, genetic counseling was recommended in 27% (n = 26) of patients, while additional molecular analyses were requested in 25% (n = 24) cases. Fifty-six patients (58%) displayed at least one actionable alteration. For patients with available follow-up (n = 50), 32 (64%) received MMTs and 18 (36%) ST. No differences in real-world progression-free survival (rwPFS) (4.07 months [95% CI 2.14–8.28] vs. 3.12 months [95% CI 1.51-NE], P = 0.8) and 12-month overall survival (OS) (58% [95%CI 43–78] vs. 57% [95%CI 34–97), P = 0.9) were observed between the MMT- and ST-group. Level I ESCAT alterations yielded longer rwPFS (5.82 months [95% CI 3.12–8.41]) compared to ESCAT II (2.14 months [95%CI 1.61-NE]) and ESCAT III (2.10 months [95% CI 2.04-NE]; P = 0.03). Twenty-four percent of patients showed a PFS2/PFS1 ratio > 1.3 from MMT. Conclusion Molecular tumor boards can provide additional treatment options for patients affected by mBC. Besides treatment recommendations, MTBs also have the utility to assess the validity of discussed genomic reports and to identify alterations worthy of genetic counseling.

Proteogenomic Landscape of Breast Ductal Carcinoma Reveals Tumor Progression Characteristics and Therapeutic Targets

Article

Full-text available

Oct 2024

Multi‐omics studies of breast ductal carcinoma (BRDC) have advanced the understanding of the disease's biology and accelerated targeted therapies. However, the temporal order of a series of biological events in the progression of BRDC is still poorly understood. A comprehensive proteogenomic analysis of 224 samples from 168 patients with malignant and benign breast diseases is carried out. Proteogenomic analysis reveals the characteristics of linear multi‐step progression of BRDC, such as tumor protein P53 (TP53) mutation‐associated estrogen receptor 1 (ESR1) overexpression is involved in the transition from ductal hyperplasia (DH) to ductal carcinoma in situ (DCIS). 6q21 amplification‐associated nuclear receptor subfamily 3 group C member 1 (NR3C1) overexpression helps DCIS_Pure (pure DCIS, no histologic evidence of invasion) cells avoid immune destruction. The T‐cell lymphoma invasion and metastasis 1, androgen receptor, and aldo‐keto reductase family 1 member C1 (TIAM1‐AR‐AKR1C1) axis promotes cell invasion and migration in DCIS_adjIDC (DCIS regions of invasive cancers). In addition, AKR1C1 is identified as a potential therapeutic target and demonstrated the inhibitory effect of aspirin and dydrogesterone as its inhibitors on tumor cells. The integrative multi‐omics analysis helps to understand the progression of BRDC and provides an opportunity to treat BRDC in different stages.

Unraveling the metastasis‐preventing effect of miR‐200c in vitro and in vivo

Article

Full-text available

Oct 2024
MOL ONCOL

Advanced breast cancer, as well as ineffective treatments leading to surviving cancer cells, can result in the dissemination of these malignant cells from the primary tumor to distant organs. Recent research has shown that microRNA 200c (miR‐200c) can hamper certain steps of the invasion–metastasis cascade. However, it is still unclear whether miR‐200c expression alone is sufficient to prevent breast cancer cells from metastasis formation. Hence, we performed a xenograft mouse experiment with inducible miR‐200c expression in MDA‐MB 231 cells. The ex vivo analysis of metastatic sites in a multitude of organs, including lung, liver, brain, and spleen, revealed a dramatically reduced metastatic burden in mice with miR‐200c‐expressing tumors. A fundamental prerequisite for metastasis formation is the motility of cancer cells and, therefore, their migration. Consequently, we analyzed the effect of miR‐200c on collective‐ and single‐cell migration in vitro, utilizing MDA‐MB 231 and MCF7 cell systems with genetically modified miR‐200c expression. Analysis of collective‐cell migration revealed confluence‐dependent motility of cells with altered miR‐200c expression. Additionally, scratch assays showed an enhanced predisposition of miR‐200c‐negative cells to leave cell clusters. The in‐between stage of collective‐ and single‐cell migration was validated using transwell assays, which showed reduced migration of miR‐200c‐positive cells. Finally, to measure migration at the single‐cell level, a novel assay on dumbbell‐shaped micropatterns was performed, which revealed that miR‐200c critically determines confined cell motility. All of these results demonstrate that sole expression of miR‐200c impedes metastasis formation in vivo and migration in vitro and highlights miR‐200c as a metastasis suppressor in breast cancer.

The evolutionary history of 2,658 cancers

Article

Full-text available

Feb 2020

Yulia Miller

Cancer progression is a process of somatic evolution1,2. Sequencing data from a single biopsy represent a snapshot of this process, revealing the timing of specific genomic aberrations and the changing impact of mutational processes3. Here, by analyzing whole-genome sequencing of 2,658 cancers as part of the Pan-Cancer Analysis of Whole Genomes (PCAWG) consortium of the International Cancer Genome Consortium (ICGC) and The Cancer Genome Atlas (TCGA)4, we reconstruct the life history and mutational processes of 38 cancers and the sequence of driver mutations. Early tumorigenesis is characterized by mutations in a limited set of driver genes and specific copy number gains, such as trisomy 7 in glioblastoma and chromosome 17q in medulloblastoma. In 40% of samples, the mutational spectrum changes dramatically throughout tumor evolution. Advanced stages are characterized by a nearly fourfold increase in driver gene diversification and increased genomic instability. Copy number variations often occur during mitotic crisis and result in concurrent gains in chromosomal fragments. Temporal analysis shows that driver mutations often precede diagnosis by many years, even decades. Together, these results determine the evolutionary trajectory of cancer and highlight opportunities for early cancer detection.

A multi-modal single-cell and spatial expression map of metastatic breast cancer biopsies across clinicopathological features

Article

Full-text available

Oct 2024
NAT MED

Although metastatic disease is the leading cause of cancer-related deaths, its tumor microenvironment remains poorly characterized due to technical and biospecimen limitations. In this study, we assembled a multi-modal spatial and cellular map of 67 tumor biopsies from 60 patients with metastatic breast cancer across diverse clinicopathological features and nine anatomic sites with detailed clinical annotations. We combined single-cell or single-nucleus RNA sequencing for all biopsies with a panel of four spatial expression assays (Slide-seq, MERFISH, ExSeq and CODEX) and H&E staining of consecutive serial sections from up to 15 of these biopsies. We leveraged the coupled measurements to provide reference points for the utility and integration of different experimental techniques and used them to assess variability in cell type composition and expression as well as emerging spatial expression characteristics across clinicopathological and methodological diversity. Finally, we assessed spatial expression and co-localization features of macrophage populations, characterized three distinct spatial phenotypes of epithelial-to-mesenchymal transition and identified expression programs associated with local T cell infiltration versus exclusion, showcasing the potential of clinically relevant discovery in such maps.

Reconstructing evolutionary trajectories of mutation signature activities in cancer using TrackSig

Conference Paper

Full-text available

Feb 2020

The type and genomic context of cancer mutations depend on their causes. These causes have been characterized using signatures that represent mutation types that co-occur in the same tumours. However, it remains unclear how mutation processes change during cancer evolution due to the lack of reliable methods to reconstruct evolutionary trajectories of mutational signature activity. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, which aggregated whole-genome sequencing data from 2658 cancers across 38 tumour types, we present TrackSig, a new method that reconstructs these trajectories using optimal, joint segmentation and deconvolution of mutation type and allele frequencies from a single tumour sample. In simulations, we find TrackSig has a 3–5% activity reconstruction error, and 12% false detection rate. It outperforms an aggressive baseline in situations with branching evolution, CNA gain, and neutral mutations. Applied to data from 2658 tumours and 38 cancer types, TrackSig permits pan-cancer insight into evolutionary changes in mutational processes.

Reconstructing evolutionary trajectories of mutation signature activities in cancer using TrackSig

Conference Paper

Full-text available

Feb 2020

Yulia Miller

Three-dimensional breast cancer tumor models based on natural hydrogels: a review

Article

Sep 2024
J ZHEJIANG UNIV-SC B

Breast cancer is the most common cancer in women and one of the deadliest cancers worldwide. According to the distribution of tumor tissue, breast cancer can be divided into invasive and non-invasive forms. The cancer cells in invasive breast cancer pass through the breast and through the immune system or systemic circulation to different parts of the body, forming metastatic breast cancer. Drug resistance and distant metastasis are the main causes of death from breast cancer. Research on breast cancer has attracted extensive attention from researchers. In vitro construction of tumor models by tissue engineering methods is a common tool for studying cancer mechanisms and anticancer drug screening. The tumor microenvironment consists of cancer cells and various types of stromal cells, including fibroblasts, endothelial cells, mesenchymal cells, and immune cells embedded in the extracellular matrix. The extracellular matrix contains fibrin proteins (such as types I, II, III, IV, VI, and X collagen and elastin) and glycoproteins (such as proteoglycan, laminin, and fibronectin), which are involved in cell signaling and binding of growth factors. The current traditional two-dimensional (2D) tumor models are limited by the growth environment and often cannot accurately reproduce the heterogeneity and complexity of tumor tissues in vivo. Therefore, in recent years, research on three-dimensional (3D) tumor models has gradually increased, especially 3D bioprinting models with high precision and repeatability. Compared with a 2D model, the 3D environment can better simulate the complex extracellular matrix components and structures in the tumor microenvironment. Three-dimensional models are often used as a bridge between 2D cellular level experiments and animal experiments. Acellular matrix, gelatin, sodium alginate, and other natural materials are widely used in the construction of tumor models because of their excellent biocompatibility and non-immune rejection. Here, we review various natural scaffold materials and construction methods involved in 3D tissue-engineered tumor models, as a reference for research in the field of breast cancer.

Phylogenetic analysis of metastatic progression in breast cancer using somatic mutations and copy number aberrations

Article

Full-text available

Apr 2017

Several studies using genome-wide molecular techniques have reported various degrees of genetic heterogeneity between primary tumours and their distant metastases. However, it has been difficult to discern patterns of dissemination owing to the limited number of patients and available metastases. Here, we use phylogenetic techniques on data generated using whole-exome sequencing and copy number profiling of primary and multiple-matched metastatic tumours from ten autopsied patients to infer the evolutionary history of breast cancer progression. We observed two modes of disease progression. In some patients, all distant metastases cluster on a branch separate from their primary lesion. Clonal frequency analyses of somatic mutations show that the metastases have a monoclonal origin and descend from a common ‘metastatic precursor’. Alternatively, multiple metastatic lesions are seeded from different clones present within the primary tumour. We further show that a metastasis can be horizontally cross-seeded. These findings provide insights into breast cancer dissemination.

Mutational signatures of ionizing radiation in second malignancies

Article

Full-text available

Sep 2016

The Subclonal Architecture of Metastatic Breast Cancer: Results from a Prospective Community-Based Rapid Autopsy Program “CASCADE”

Article

Full-text available

Dec 2016
PLOS MED

Background Understanding the cancer genome is seen as a key step in improving outcomes for cancer patients. Genomic assays are emerging as a possible avenue to personalised medicine in breast cancer. However, evolution of the cancer genome during the natural history of breast cancer is largely unknown, as is the profile of disease at death. We sought to study in detail these aspects of advanced breast cancers that have resulted in lethal disease. Methods and Findings Three patients with oestrogen-receptor (ER)-positive, human epidermal growth factor receptor 2 (HER2)-negative breast cancer and one patient with triple negative breast cancer underwent rapid autopsy as part of an institutional prospective community-based rapid autopsy program (CASCADE). Cases represented a range of management problems in breast cancer, including late relapse after early stage disease, de novo metastatic disease, discordant disease response, and disease refractory to treatment. Between 5 and 12 metastatic sites were collected at autopsy together with available primary tumours and longitudinal metastatic biopsies taken during life. Samples underwent paired tumour-normal whole exome sequencing and single nucleotide polymorphism (SNP) arrays. Subclonal architectures were inferred by jointly analysing all samples from each patient. Mutations were validated using high depth amplicon sequencing. Between cases, there were significant differences in mutational burden, driver mutations, mutational processes, and copy number variation. Within each case, we found dramatic heterogeneity in subclonal structure from primary to metastatic disease and between metastatic sites, such that no single lesion captured the breadth of disease. Metastatic cross-seeding was found in each case, and treatment drove subclonal diversification. Subclones displayed parallel evolution of treatment resistance in some cases and apparent augmentation of key oncogenic drivers as an alternative resistance mechanism. We also observed the role of mutational processes in subclonal evolution. Limitations of this study include the potential for bias introduced by joint analysis of formalin-fixed archival specimens with fresh specimens and the difficulties in resolving subclones with whole exome sequencing. Other alterations that could define subclones such as structural variants or epigenetic modifications were not assessed. Conclusions This study highlights various mechanisms that shape the genome of metastatic breast cancer and the value of studying advanced disease in detail. Treatment drives significant genomic heterogeneity in breast cancers which has implications for disease monitoring and treatment selection in the personalised medicine paradigm.

Tracing the origin of disseminated tumor cells in breast cancer using single-cell sequencing

Article

Full-text available

Dec 2016
GENOME BIOL

Background Single-cell micro-metastases of solid tumors often occur in the bone marrow. These disseminated tumor cells (DTCs) may resist therapy and lay dormant or progress to cause overt bone and visceral metastases. The molecular nature of DTCs remains elusive, as well as when and from where in the tumor they originate. Here, we apply single-cell sequencing to identify and trace the origin of DTCs in breast cancer. Results We sequence the genomes of 63 single cells isolated from six non-metastatic breast cancer patients. By comparing the cells’ DNA copy number aberration (CNA) landscapes with those of the primary tumors and lymph node metastasis, we establish that 53% of the single cells morphologically classified as tumor cells are DTCs disseminating from the observed tumor. The remaining cells represent either non-aberrant “normal” cells or “aberrant cells of unknown origin” that have CNA landscapes discordant from the tumor. Further analyses suggest that the prevalence of aberrant cells of unknown origin is age-dependent and that at least a subset is hematopoietic in origin. Evolutionary reconstruction analysis of bulk tumor and DTC genomes enables ordering of CNA events in molecular pseudo-time and traced the origin of the DTCs to either the main tumor clone, primary tumor subclones, or subclones in an axillary lymph node metastasis. Conclusions Single-cell sequencing of bone marrow epithelial-like cells, in parallel with intra-tumor genetic heterogeneity profiling from bulk DNA, is a powerful approach to identify and study DTCs, yielding insight into metastatic processes. A heterogeneous population of CNA-positive cells is present in the bone marrow of non-metastatic breast cancer patients, only part of which are derived from the observed tumor lineages. Electronic supplementary material The online version of this article (doi:10.1186/s13059-016-1109-7) contains supplementary material, which is available to authorized users.

Tumor Evolution in Two Patients with Basal-like Breast Cancer: A Retrospective Genomics Study of Multiple Metastases

Article

Full-text available

Dec 2016
PLOS MED

Background: Metastasis is the main cause of cancer patient deaths and remains a poorly characterized process. It is still unclear when in tumor progression the ability to metastasize arises and whether this ability is inherent to the primary tumor or is acquired well after primary tumor formation. Next-generation sequencing and analytical methods to define clonal heterogeneity provide a means for identifying genetic events and the temporal relationships between these events in the primary and metastatic tumors within an individual. Methods and findings: We performed DNA whole genome and mRNA sequencing on two primary tumors, each with either four or five distinct tissue site-specific metastases, from two individuals with triple-negative/basal-like breast cancers. As evidenced by their case histories, each patient had an aggressive disease course with abbreviated survival. In each patient, the overall gene expression signatures, DNA copy number patterns, and somatic mutation patterns were highly similar across each primary tumor and its associated metastases. Almost every mutation found in the primary was found in a metastasis (for the two patients, 52/54 and 75/75). Many of these mutations were found in every tumor (11/54 and 65/75, respectively). In addition, each metastasis had fewer metastatic-specific events and shared at least 50% of its somatic mutation repertoire with the primary tumor, and all samples from each patient grouped together by gene expression clustering analysis. TP53 was the only mutated gene in common between both patients and was present in every tumor in this study. Strikingly, each metastasis resulted from multiclonal seeding instead of from a single cell of origin, and few of the new mutations, present only in the metastases, were expressed in mRNAs. Because of the clinical differences between these two patients and the small sample size of our study, the generalizability of these findings will need to be further examined in larger cohorts of patients. Conclusions: Our findings suggest that multiclonal seeding may be common amongst basal-like breast cancers. In these two patients, mutations and DNA copy number changes in the primary tumors appear to have had a biologic impact on metastatic potential, whereas mutations arising in the metastases were much more likely to be passengers.

Impact of mutational profiles on response of primary oestrogen receptor-positive breast cancers to oestrogen deprivation

Article

Full-text available

Dec 2016

Pre-surgical studies allow study of the relationship between mutations and response of oestrogen receptor-positive (ER þ) breast cancer to aromatase inhibitors (AIs) but have been limited to small biopsies. Here in phase I of this study, we perform exome sequencing on baseline, surgical core-cuts and blood from 60 patients (40 AI treated, 20 controls). In poor responders (based on Ki67 change), we find significantly more somatic mutations than good responders. Subclones exclusive to baseline or surgical cores occur in B30% of tumours. In phase II, we combine targeted sequencing on another 28 treated patients with phase I. We find six genes frequently mutated: PIK3CA, TP53, CDH1, MLL3, ABCA13 and FLG with 71% concordance between paired cores. TP53 mutations are associated with poor response. We conclude that multiple biopsies are essential for confident mutational profiling of ER þ breast cancer and TP53 mutations are associated with resistance to oestrogen deprivation therapy.

Mutational Landscape of Metastatic Cancer Revealed from Prospective Clinical Sequencing of 10,000 Patients

Article

May 2017
NAT MED

Tumor molecular profiling is a fundamental component of precision oncology, enabling the identification of genomic alterations in genes and pathways that can be targeted therapeutically. The existence of recurrent targetable alterations across distinct histologically defined tumor types, coupled with an expanding portfolio of molecularly targeted therapies, demands flexible and comprehensive approaches to profile clinically relevant genes across the full spectrum of cancers. We established a large-scale, prospective clinical sequencing initiative using a comprehensive assay, MSK-IMPACT, through which we have compiled tumor and matched normal sequence data from a unique cohort of more than 10,000 patients with advanced cancer and available pathological and clinical annotations. Using these data, we identified clinically relevant somatic mutations, novel noncoding alterations, and mutational signatures that were shared by common and rare tumor types. Patients were enrolled on genomically matched clinical trials at a rate of 11%. To enable discovery of novel biomarkers and deeper investigation into rare alterations and tumor types, all results are publicly accessible.

Extrachromosomal oncogene amplification drives tumor evolution and genetic heterogeneity

Article

Feb 2017

Human cells have twenty-three pairs of chromosomes. In cancer, however, genes can be amplified in chromosomes or in circular extrachromosomal DNA (ecDNA), although the frequency and functional importance of ecDNA are not understood. We performed whole-genome sequencing, structural modelling and cytogenetic analyses of 17 different cancer types, including analysis of the structure and function of chromosomes during metaphase of 2,572 dividing cells, and developed a software package called ECdetect to conduct unbiased, integrated ecDNA detection and analysis. Here we show that ecDNA was found in nearly half of human cancers; its frequency varied by tumour type, but it was almost never found in normal cells. Driver oncogenes were amplified most commonly in ecDNA, thereby increasing transcript level. Mathematical modelling predicted that ecDNA amplification would increase oncogene copy number and intratumoural heterogeneity more effectively than chromosomal amplification. We validated these predictions by quantitative analyses of cancer samples. The results presented here suggest that ecDNA contributes to accelerated evolution in cancer.

Corrigendum: Whole-genome characterization of chemoresistant ovarian cancer

Article

Jan 2015
NATURE

Impact of mutational profiles on response of primary oestrogen receptor-positive breast cancers to oestrogen deprivation

Article

Nov 2016

Pre-surgical studies allow study of the relationship between mutations and response of oestrogen receptor-positive (ER+) breast cancer to aromatase inhibitors (AIs) but have been limited to small biopsies. Here in phase I of this study, we perform exome sequencing on baseline, surgical core-cuts and blood from 60 patients (40 AI treated, 20 controls). In poor responders (based on Ki67 change), we find significantly more somatic mutations than good responders. Subclones exclusive to baseline or surgical cores occur in ∼30% of tumours. In phase II, we combine targeted sequencing on another 28 treated patients with phase I. We find six genes frequently mutated: PIK3CA, TP53, CDH1, MLL3, ABCA13 and FLG with 71% concordance between paired cores. TP53 mutations are associated with poor response. We conclude that multiple biopsies are essential for confident mutational profiling of ER+ breast cancer and TP53 mutations are associated with resistance to oestrogen deprivation therapy.

Genomic Evolution of Breast Cancer Metastasis and Relapse

Abstract and Figures

Supplementary resources (7)

Recommended publications

Recombination drives genome evolution in outbreak-related Legionella pneumophila isolates

The Genome of Herpes Simplex Virus: Structure, Replication and Evolution

Comparative genomics for non O1/O139 Vibrio cholerae isolates recovered from the Yangtze River Estua...

Excess of genomic defects in a woolly mammoth on Wrangel island