Anda di halaman 1dari 15

REVIEWS

Using tumour phylogenetics to identify


theroots of metastasis in humans
Kamila Naxerova and Rakesh K.Jain
Abstract | In cancer, much uncertainty remains regarding the origins of metastatic disease. Models of
metastatic progression offer competing views on when dissemination occurs (at an early or late stage
oftumour development), whether metastases at different sites arise independently and directly from the
primary tumour or give rise to each other, and whether dynamic cell exchange occurs between synchronously
growing lesions. Although it is probable that many routes can lead to the establishment of systemic disease,
clinical observations suggest that distinct modes of metastasis might prevail in different tumour types.
Gaining a more-comprehensive understanding of the evolutionary processes that underlie metastasis is not
only relevant from a basic biological perspective, but also has profound clinical implications. The tree of life
of metastatic cancer contains answers to many outstanding questions about the development of systemic
disease, but has only been reconstructed in a limited number of patients. Here we review available data on the
phylogenetic relationships between primary solid tumours and their metastases, and examine to what degree
they support different models of metastatic progression. We provide a description of experimental methods for
lineage tracing in human cancer, ranging from broad DNA-sequencing approaches to more-targeted techniques,
and discuss their respective benefits and caveats. Finally, we propose future research questions in the area of
cancer phylogenetics.
Naxerova, K. & Jain, R. K. Nat. Rev. Clin. Oncol. advance online publication 20 January 2015; doi:10.1038/nrclinonc.2014.238

Introduction

Edwin L. Steele
Laboratory for
TumourBiology,
MassachusettsGeneral
Hospital and Harvard
Medical School,
100BlossomStreet,
Cox 7, Boston,
MA02114, USA (K.N.,
R.K.J.).
Correspondence to: K.N.
knaxerova@partners.org

Gaining a deeper knowledge of heterogeneity in human


cancers is becoming increasingly important for both
basic and translational cancer research. This develop
ment is partly fueled by an urgent need to understand
and prevent or circumvent therapy resistance, 1 and
partly due to the realization that intratumour hetero
geneity offers a rare window into the complex evolu
tionary history of a cancer.2,3 The term intratumour
heterogeneity is itself subject to heterogeneity: it can
describe diversity among cells intermingling in one
localized area; refer to differences between spatially
separated tumour regions; or denote variation among
multiple noncontiguous tumours in metastatic disease,
although in this setting it would perhaps be more accu
rate to speak of intracancer heterogeneity. Intratumour
heterogeneity occurs on all levels of observation, many
of them nongenetic (such as phenotypic and epigenetic
variation);4 nevertheless, from the point of view of a
clinician considering prescribing molecularly targeted
therapies, genetic divergence between primary tumours
and metastasesthe focus of this Reviewis arguably
one of the most-relevant forms. In current practice, treat
ment strategies aimed at the eradication of metastasis
are typically guided by molecular analysis of small cell
populations from the primary tumour. However, we are
becoming increasingly aware that the genetic markers
Competing interests
The authors declare no competing interests.

that form the basis for allocation of many targeted treat


ments show considerable discordance both within and
between tumours.57 Additional biopsy of metastases is,
therefore, becoming advocated,8 but sampling of second
ary tumours is often impractical due to surgical inaccess
ibility. The genetic traits of micrometastases are even
more difficult to assess. Superimposed on this genetic
heterogeneity of cancer cells is the heterogeneity in the
tumour microenvironment, which can drive tumour pro
gression and promote treatment resistance.9 Owing to
these challenges, gaining a more-general understanding
of how diversity arises within the primary tumour and
how the bottleneck of metastatic dissemination modu
lates tumour heterogeneity is becoming increasingly
expedient. Importantly, a cancers genetic landscape also
encodes a record of its evolutionary trajectory; decod
ing this record has the potential to reveal fundamental
principles of tumour biology.
The emergence of metastasis has been surrounded
by mystery for a long time. The humoralist school of
thought regarded cancer as an inherently systemic
disease caused by abnormal accumulation of black
bile, one of the four humours believed to constitute
the human body.10 Consequently, metastases were not
considered a product of the primary tumour, but rather
another independent symptom of the same underlying
pathology. For a prolonged period during his scien
tific career, even Rudolf Virchow, the father of cellular
pathology, firmly believed the paradigm that primary

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 1


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
Key points
Whether metastasis tends to occur early or late in tumour development
remains controversial, and whether metastases descend directly from the
primary tumour or give rise to each other is unclear
In the linear progression model, metastatic precursors leave the primary
tumour at late stages of disease, after clonal evolution has given rise to a cell
with metastatic ability; consequently, primary tumours and metastases are
genetically closely related
The parallel progression model, on the other hand, assumes that metastasis
occurs in early stages of carcinogenesis, and that metastases and primary
tumour evolve independently, resulting in genetic disparity between them
Comparative genomics studies in different cancer types illustrate a variety of
possible progression trajectories for systemic disease, but analysis of more
patients is needed to arrive at generalizable conclusions
Problems complicating the interpretation of comparative genetics data include
the unknown contributions of the tissue-specific background mutational burden,
self-seeding and tumour-cell dormancy, and the extensive heterogeneity of
primary tumours
A variety of experimental approaches beyond next-generation DNA sequencing
are available for lineage tracing in human cancer

tumours and metastases arise independently. Virchow


hypothesized that the primary cancer infuses the blood
with toxins that trigger the formation of secondary
growths at distant sites, but that no cellular exchange
takes place.10 Surprisingly, at present, some of the mostimportant steps in the metastatic process remain poorly
understood. In particular, much uncertainty surrounds
the following questions. First, when do metastatic pre
cursor cells leave the primary tumour? Of note, the time
point of dissemination is probably a major determinant
of genetic divergence. Second, by what route do meta
static cells spread to form widely disseminated disease?
That is, do metastases typically emanate directly from
the primary tumour and independently of each other,
or do metastases give rise to other metastases in cas
cades?11 Third, can metastatic cells return to the tumour
from which they originated? Fourth, how does tumourcell dormancy influence when metastasis emerges?
Unfortunately, many decisive properties of metastasis
in humans, among them latency and variable time to
progression, are not recapitulated accurately in animal
models.12 Obtaining data in humans is, therefore, crucial.
The aim of this Review is to summarize pertinent
information about the lineage of metastases in humans.
We describe the theoretical models that currently domi
nate our concepts of metastatic progression, and consider
comparative data for solid primary tumours and associ
ated metastases, with an emphasis on how the available
evidence corroborates these models. In addition, we
outline experimental approaches that can be used to
study tumour lineage in humans and discuss their rela
tive advantages and disadvantages. We conclude with a
discussion of areas of inquiry for future research.

Models of metastatic progression

Metastases can be diagnosed at approximately the


same time as the primary tumour (synchronous) or
after latency periods ranging from 6months to several
decades (metachronous). The distribution of metastasisfree survival intervals in the population of patients in

whom relapse at distant sites does occur is specific to the


tumour type and can be very broad. For example, risk
of distant relapse for oestrogen receptor (ER)negative
breast cancer peaks at less than 2years after resection
of the primary tumour, whereas for ERpositive breast
tumours, the highest risk of distant recurrence occurs
more than 3years after surgery and falls off more slowly
over time.13 Currently, the time to recurrence (TTR) at
distant sites cannot be accurately predicted. Although the
size of the primary tumour at diagnosis is correlated with
the overall risk of metastasis for many cancer types, this
measure is not necessarily an indicator of TTR. In colon
cancer, for example, primary tumour characteristics
(size, stage, grade) in patients with synchronous versus
metachronous metastases are not significantlydifferent.14
Therefore, individual tumour biology seems to not
only dictate whether distant recurrence will occur, but
also determines TTR and how quickly metastatic disease
will reach lethal dimensions once a cancer has relapsed.
The biological mechanisms underlying this large vari
ation are unknown, and many combinations of inter
acting causative elements are conceivable, including
the time point of metastasis relative to the evolutionary
trajectory of the primary tumour, possible dormancy
periods and tumour growth rates at ectopic sites. All
these parameters are probably influenced by a multitude
of host factors, including the stromal1517 and immuno
logical18 components of the local9 and systemic19 envi
ronment. A discussion of the host elements that causally
determine the fate of the growing primary tumour and
the disseminated tumour cells is beyond the scope of
this Review, as is a breakdown of the specific functional
roles of (epi)genetic alterations in metastasis. Herein, we
focus on genetic alterations only as a form of forensic
evidence, taking advantage of what they can tell us about
the phylogenetic histories of metastases, without assum
ing that they have a causal role in the metastatic process.
In fact, most genetic aberrations probably have no role
in metastasis, but they can, nevertheless, provide a useful
record of past dissemination events.
Several conceptual frameworks that aim to explain
the behaviour of metastatic cancer are under ongoing
development. This fact is a reflection of not only the
diverse biology of metastasis, but also of how challenging
this biology is to study and the considerable research that
remains to be done in this area. Assessment of genetic
intracancer heterogeneity is one of few approaches
available to address these important questions in humans.

Linear progression
The traditional paradigm of metastatic progression is
the linear progression model, so called because it postu
lates that primary tumour development and metasta
sis occupy sequential positions on a unidirectional
timeline of events (Figure1a). A central assumption
of the linear progression model is that only genetically
advanced cancer cells can effectively colonize distant
organs. These cells are thought to arise in a step-wise
fashion, through multiple clonal expansions, during the
development of the primary tumour.20 As acquisition of

2 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
a

Primary
tumour

Liver
metastasis

Primary
tumour

Lung
metastasis

Liver
metastasis

Lung
metastasis

Liver
Lung
metastasis metastasis

Primary
tumour

TumourTumourcell
mass
dormancy dormancy

Primary
tumour

Figure 1 | Overview of human metastasis models. a | The linear progression and metastatic cascade model. A transformed
Nature Reviews | Clinical Oncology
cell divides to form an early stage primary tumour (lower panel, red cells). As the tumour proliferates (dotted lines indicate cell
divisions), it undergoes clonal evolution, eventually giving rise to a fully malignant clone that has the ability to metastasize
(yellow cells). Other clones might or might not coexist with the metastatic clone in the primary tumour (not shown in this
depiction that focuses on the genetic similarity between tumour lesions as opposed to the intratumour heterogeneity within
individual lesions). As the metastatic clone is fully malignant, it can colonize other organs efficiently (yellow cells: lung
metastasis) and spread further (yellow cells: liver metastasis) in a cascade, without having to acquire major new alterations.
The anatomical cartoon (upper panel) depicts an advanced invasive breast carcinoma initiating a metastatic cascade shortly
before clinical detection. All curved arrows indicate dissemination. b | The parallel progression model. In its early growth
stages, the primary tumour (lower panel, red cells) begins to seed metastases in other organs (brown and purple panels).
These cells proliferate in their respective ectopic microenvironments (dotted lines) in parallel with the primary tumour,
acquiring distinct sets of mutations (green and orange cells). The anatomical cartoon shows various metastases that are
seeded directly and independently from an early stage cancer. c | Tumour self-seeding. Tumour cells return from the
metastatic site to the primary tumour. Self-seeding is compatible with both parallel (green cells from a liver metastasis) and
linear progression (yellow cells returning from a lung metastasis). However, the likelihood of self-seeding could potentially be
higher in the case of parallel progression because metastasis and primary tumour exist synchronously for longer periods of
time. The primary tumour could also continually emit cells to a metastasis that was in fact seeded at early stages of
tumorigenesis, masking evidence of parallel progression (yellow cells from the primary tumour migrating to the liver
metastasis). d | Dormancy. After dissemination (curved arrows) from the primary tumour to distant organs (brown and purple
panels), single tumour cells enter a state of dormancy (tumour-cell dormancy), or persist as small microscopic lesions in
which cell birth and cell death (red cross) are balanced (tumour-mass dormancy). In this depiction, tumour-cell dissemination
occurs early, but dormancy might also be possible in cells that leave the primary tumour at late stages of disease.

metastasis-enabling mutations is a stochastic process,


the likelihood of such mutations arising increases
withthe number of cell divisions that a tumour under
goes. Therefore, metastasis is assumed to occur shortly
before a tumour becomes clinically detectable (after it
has undergone many cell divisions to form the detect
able tumour mass). The linear progression model pre
dicts a small evolutionary distance between primary
and secondary neoplasms, and therefore suggests that
the primary tumour is a good surrogate for the molecular
properties of metastases.

Metastatic cascades
Loosely associated with the linear progression model, by
virtue of placing the development of metastasis in the latest
stages of carcinogenesis,11 is the concept that metastases,
perhaps particularly those in central organs with extensive
vascularization and high blood flow, such as the lung and
liver, give rise to other metastases in a cascading manner,21
forming showers of metastases (Figure1a).22 To account
for the development of widespread metastases in a rela
tively short amount of timein many cases, metastatic
cancer emerges 23years after diagnosis of the primary

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 3


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
tumour, and according to linear progression, dissemina
tion occurs shortly before clinical detectionthe cascade
model assumes a very high growth rate of metastases.11
Aggressive tumours formed through rapid clonal expan
sions, perhaps due to infrequent cell death by apoptosis or
necrosis, would be expected to harbour comparatively low
levels of diversity (unless the mutation rate in the tumour
cells is increased). Consequently, the cascade model pre
dicts metastases that are relatively genetically uniform
and that are moreclosely related to each other than to
the primary tumour.
The invasion of the lymphatic system is a potentially
important step in the metastatic cascade. Autopsy studies
show that regional and distant lymph nodes are by far the
most-common sites of metastasis. In one large autopsy
study in patients with various cancers,23 at the time of
death, lymphatic metastases were twofold more frequent
than lesions in the next-most-common site of metastasis:
the liver. Indeed, for many cancer types, the presence
of cancer cells in regional lymph nodes is a well-known
negative prognostic indicator. Historically, lymph node
metastases were, therefore, assumed to be precursors of
distant lesions. This belief motivated aggressive surgical
interventions to eradicate locoregional disease, such as
the radical mastectomy and axillary lymph node dissec
tion approach pioneered in breast cancer by Halsted at
the end of the 19th century.24,25 However, the benefit of
axillary lymph node dissection has been questioned,26,27
and it has been argued that lymphatic lesions are unlikely
to give rise to distant metastases.28

Parallel progression
Diametrically opposed to linear progression, although
not strictly mutually exclusive, is the parallel progression
model (Figure1b). This model posits that metastasis
occurs early in cancer development, potentially already at
or before the carcinoma insitu stage,29 and that primary
and secondary tumours evolve independently.11 Parallel
progression is closely related to the alternative hypoth
esis of tumour progression (proposed by Bernard Fisher25)
that regards cancer as a systemic disease from the outset.
The parallel progression model assumes that cell dissemi
nation and ectopic survival do not necessarily require a
complex repertoire of mutations, and can be accomplished
by cancer cells with few genetic abnormalities. According
to this model, the somatic evolution of these early dissemi
nated tumour cells (DTCs) occurs predominantly at the
distant organ sites and involves extensive adaption to local
microenvironments. Therefore, substantial genetic dis
parity between the primary tumour and its metastases, as
well as between metastases in different anatomic locations,
is expected. This paradigm is in contrast to the metastatic
cascade model, which postulates that metastases arise
from each other during a rapidly fatal period of clonal
expansion that would leave less time for diversification
than slow parallel evolution of metastases, at least under
the assumption of comparable mutation rates. Under the
parallel progression model, molecular profiling of primary
tumours is considered inappropriate for selecting effective
therapeutics against (micro)metastatic disease.

Tumour self-seeding
Both the linear and the parallel progression models
regard metastasis as a unidirectional process that begins
within the primary neoplasm and terminates at a distant
site(s). Tumour self-seeding (Figure1c) is a recently
articulated hypothesis stating that bidirectional, dynamic
cell exchange exists between synchronous lesions.30 The
primary tumour is proposed to continually shed cancer
cells into the bloodstream, some of which pass through
the lung capillary network to enter the arterial circula
tion, with a highly selected subset of these circulating
tumour cells (CTCs) re-entering the primary tumour to
drive local progression. Cells that are shed or extrava
sate from proliferating metastases could similarly return
to the primary tumour, compounding intratumour
heterogeneity. At present, we do not know whether
a clinically significant degree of tumour self-seeding
occurs in humans, but if it does, this process would
obscure evidence of independent tumour evolution at
differentsites.
Dormancy
Dormancy is a loosely defined term that is used to
describe multiple distinct forms of tumour growth arrest
(Figure1d).31 In the clinical setting, dormancy is invoked
to explain ultra-late disease recurrence after 10years of
disease-free survival.12 From a cell biology perspective,
dormancy can either refer to a senescence-like state of
single DTCs after they were entrapped in foreign and
potentially hostile tissue microenvironments (that is,
tumour-cell dormancy) or to the indolent behaviour of
subclinical cancers that exhibit no net growth (tumourmass dormancy). Dormancy is an important factor
to consider when interpreting tumour phylogenies
because the mutations that are used for reconstruction
of evolutionary trees typically only occur in proliferat
ing cells. A dormant tumour cell that does not divide,
therefore, effectively stops its evolutionary clock. Thus,
whether a metastatic lesion arose after a prolonged
latency period because it disseminated late in cancer pro
gression or because it underwent a period of dormancy
at the distant site might be difficult to judge.
Biological variability
Which of these models best describes metastasis in dif
ferent patient populations is currently unknown, and a
better understanding of how systemic disease develops is
urgently needed. Personalizing treatment for metastatic
disease according to the molecular profile of the primary
tumour could be adequate for cancers that follow the
linear progression model and metastasize in cascades;
however, in the case of parallel progression, repeat biop
sies or analysis of CTCs or circulating tumour DNA
(ctDNA) might be required to obtain up-to-date infor
mation on the genetic profile of target cells. Improved
understanding of modes of metastatic evolution will also
answer fundamental biological questions, which could
have important implications for diagnosis, prognosis and
therapy; for example, whether the ability to metastasize
is present in cancer cells at early stages of tumorigenesis

4 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
(with interplay between the local microenvironment at
the distant site and genetic subclonal evolution deter
mining the survival and establishment, growth rate and
biology of the metastasis), or whether this capacity has
to be acquired over time through progressive stochastic
genetic mutation (providing the disseminated tumour
cells with an inherent ability to colonize new tissues
effectively and multiply relatively rapidly).
Different modes of metastasis might be prevalent in
different tumour types, and varying combinations of
the models presented earlier can probably also occur.
The spectrum theory has formalized this view, in an
attempt to find a harmonious middle ground between
the linear and parallel progression models, acknowledg
ing the large degree of biological variation that exists in
human cancers.24 The following section presents data
from patients with cancer on the relationship between
solid primary tumours and metastases, and discusses
how compatible the results are with different models
of metastatic progression. We focus on comparative
tumour genetics, and only briefly mention other types
of evidence.

Comparative genetics evidence

The observation that growth rates of primary tumours


and metastases are similar supports the parallel progres
sion model;11 epithelial cancers are thought to develop
over years and decades,32,33 therefore, overt synchronous
metastases should be of similar age. In other words, if
a primary tumour takes 10years to reach a detectable
volume, a metastasis that grows at similar rate should
take approximately the same period of time to become
observable. The linear progression model, on the other
hand, cannot realistically explain the emergence of overt
metastasis close to the point of primary tumour diag
nosis without assuming dramatically elevated growth
rates of secondary lesions. Most imaging studies indi
cate that metastasis doubling times are similar to those
observed for the tumour of origin.11 Thus, the growth
rates of macroscopic metastases do not seem to be
increased; however, we do not know whether their early
growth stages (that is, the growth of micrometastases) are
accelerated (perhaps because the disseminated tumour
cells have already acquired enabling mutations or due to
microenvironmental effects9), with subsequent deceler
ation as diverse factors (such as environmental and
metabolic pressures) begin to limit the high rate of cell
division. More-direct evidence of the route of metastasis
can be garnered from genetic comparisons of primary
tumours and metastatic precursors.

Disseminated/circulating tumour cell analyses


CTCs and DTCs are defined as cells found in the blood
or other tissues of patients with cancer, respectively,
that are positive for the expression of cytokeratin34 or
epithelial cell-adhesion molecule (EpCAM). 35 The
presence of DTCs in the bone marrow (the site where
DTCs are most-commonly sampled) is associated with
a higher risk of relapse in many common epithelial
cancers.36 In addition, CTC numbers are predictive of

survival in multiple forms of metastatic cancer,3739 and


the potential for analysis of CTCsas well as circulating
free DNA (cfDNA)40as liquid biopsies is becoming
established.41 As key targets of adjuvant therapy in all
cancers, there has been a long-standing interest in the
genomics of DTCs. Data from many studies indicate that
DTCs have fewer genomic aberrations than the corre
sponding primary tumours,4244 suggesting that DTCs
disseminate from the primary tumours at early stages
of their development, before more-complex genomic
aberrations were acquired; however, the possibility that
dissemination selects for subclones with relatively intact
genomes cannot be excluded. Whether the genetic diver
gence between DTCs and primary tumours can be con
firmed with high-resolution techniques remains to be
determined. Genomic analyses of DTCs often relied on
comparative genomic hybridization (CGH),45,46 a tech
nique with relatively low resolution. Technologies for
analysis of single-cell genomes are improving,4749 and
will probably become applicable to DTC genomes soon.
A potential caveat is that it remains unclear whether
these genomically intact cells are truly metastatic pre
cursors, or rather represent indolent and thus clinically
irrelevant remnants of early evolutionary stages of the
primary tumour.28 The most-instructive data at present,
therefore, come from the direct comparison of primary
tumours and macroscopic metastases.

Genome-wide comparisons in solid tumours


Relatively few genome-wide analyses of solid primary
tumours and their metastases have been conducted to
date (Table1), but the available data provide crucial
empirical feedback on the current metastatic progres
sion models. Therefore, we discuss these studies in detail,
summarizing the major findings and reporting the con
clusions drawn by the original investigators. As intro
duced previously, close genetic ties between primary
tumours and metastases are typically considered indica
tive of linear progression, whereas genetic divergence is
interpreted as evidence of parallel progression. However,
no defined cutoff has been established to separate these
two categories. Rather, they represent two ends of a
spectrum of possibilities, beginning with metastases
that arise in the earliest stages of cancer development
(during atypical or insitu growth) and ending with dis
semination immediately before surgical excision of the
primary tumour. Some important caveats complicate
this classification into early and late metastasis (and
potentially, therefore, linear and parallel progression)
based on genetic distance, and are discussed throughout
the text and examined in greater depth in the section,
Interpreting comparative genetics data.
In a study dating to the early years of next-generation
sequencing (2008), an index lesion approach was used
to compare metastases to their respective primary
tumours in 10 patients with metastatic colorectal
cancer (mCRC).32 In index-lesion sequencing, exonic
mutations discovered in the index lesion (which was a
metastasis inall cases in the aforementioned study) are
evaluatedin other lesions (here, the primary tumour)

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 5


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
Table 1 | Genome-wide comparisons of solid primary tumours and their metastases
Study

Primary cancer

Number
of
patients

Time between resection


of primary tumour
andmetastasis

Genetic relationship
between primary tumour
and metastases

Evidence
of possible
metastatic
cascade*

Jones etal. (2008)32

Colon

10

Ranged from synchronous


to 20months

High similarity

NA

Liu etal. (2009)52

Prostate

24

Synchronous

High similarity; primary only


available in 5 cases

Yes

Shah etal. (2009)57

Breast

9years

Divergent

NA

Campbell etal.
(2010)51

Pancreas

13

Synchronous

High similarity in most


patients; primary tumour not
available in some cases

Yes, in
some
patients

Breast

8months

High similarity

NA

Pancreas

Synchronous

Similarity between metastases


and localized area of primary

No

Breast

Not specified

High similarity

NA

Gerlinger etal. (2012)

Kidney

Synchronous

Divergent

Yes

Wu etal. (2012)58

Medulloblastoma

Not specified

Divergent

Yes

Haffner etal. (2013)54

Prostate

17years

Similarity between metastases


and localized area of primary

Yes

Ding etal. (2010)55


Yachida etal. (2010)

33

Navin etal. (2011)49


5

*That is, metastasis giving rise to metastasis; NA for studies that did not assess multiple metastases. Abbreviation: NA, not applicable.

in a locus-specific manner (that is, without obtaining


a full exome sequence).32 The results indicated that a
vast majority (97%) of mutations were present in both
the metastasis and the primary tumour.32 The investi
gators created a mathematical model to translate the
mutation data into chronological time and estimated
that, whereas the development of the primary tumour
took approximately 25years, metastasis occurred only
3years before diagnosis. 32 The findings were thus
interpreted by the investigators as consistent with the
linear progression model.32 More recently, high muta
tion concordance (79%) in a set of cancer-associated
genes in mCRC lesions was reported by Brannon and
colleagues.50 In addition, this study showed that many
seemingly divergent mutations were actually present in
both the primary tumour and metastasis, but restricted
to subclones that only existed in spatially defined
regions of either lesion;50 these subclonal mutations were
not included in the previously stated concordance rate.
These results further support a linear progression model
for colorectal cancer.
Another index lesion sequencing study of seven
patients with pancreatic carcinoma initially found genetic
divergence between primary tumours and metastases.33
On average, 36% of mutations present in the metastatic
index lesion could not be detected in the primary tumour
or other metastases.33 This initial finding suggested that
metastasis in pancreatic cancer must occur earlier than
in colorectal cancer. However, similarly to the study by
Brannon and co-workers,50 further analysis of spatially
distinct regions of the primary tumour in two patients
revealed large areas that were closely related to the
metastases, demonstrating that the relevant clone was
missed in the first round of testing.33 The authors con
cluded that, as in colorectal cancer, metastasis occurs
late in pancreatic cancer progression.33 This study also

showed that different metastatic lesions corresponded


to different subclones present in the primary tumour,33
indicating that multiple genetically distinct cell popula
tions metastasized independently of each other. Astudy
that reconstructed pancreatic cancer phylogenies from
chromosomal rearrangements also found evidence of
independent dissemination events in some patients;51
however, in other patients studied, different metastases
recovered from the same organ were highly similar to
each other, supporting cascading progression.51
When interpreting these results, it is important to
remember that the index lesion sequencing approach that
was common early in the next-generation sequencing
era neglects the evolutionary trajectory of the primary
cancer. All mutations that arise in the primary tumour
after departure of the metastatic clone remain obscure
if variant discovery does not encompass both tumours
(in the index lesion approach, only the metastasis is
used for variant discovery). Owing to this limitation,
a scenario in which the metastatic clone disseminates
early, enters a period of dormancy (freezing the muta
tional profile of the primary tumour in time), and finally
undergoes a rapid clonal expansion without acquiring
many new alterations, could also explain genetic con
cordance observed in index-lesion sequencing studies
that did not assess mutations specific to the primary
tumour; all mutations that the primary tumour acquired
while the metastasis was dormant would remain invis
ible, therefore, late dissemination could erroneously
beconcluded.
In a study of metastatic prostate cancer, Liu etal.52
compared the copy number profiles of multiple metasta
ses in 24 autopsy cases. Metastases and primary tumours
shared most copy-number alterations in a majority of
patients, leading the researchers to conclude a mono
clonal origin of metastatic prostate cancer.52 The primary

6 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
tumour was available for comparison in only five
patients, but in these patients, no significant divergence
was observed, a finding that would seem to support the
linear progression model. Furthermore, Liu etal.52 did
not report any recurrent genetic adaptations of metasta
ses to ectopic microenvironments in different organs,
a key feature of the parallel progression model. This
finding could conceivably be a consequence of the small
sample numbers. However, some subclonal alterations
did exist in metastases.52 Subsequent re-analysis of the
data confirmed this finding and reported several other
interesting observations; for example, that metastases
excised from the liver were more similar to each other
than to other metastases, possibly indicating intrahepatic
cascading metastasis.53
A more-recent analysis of metastatic prostate cancer by
whole-genome sequencing also observed that metasta
ses in different anatomical locations were highly similar
to each other at the time of death,54 indicating cascad
ing progression. However, in apparent contrast to the
findings of Liu and colleagues,52 the bulk of the primary
tumour did not share the hallmark mutations of the
metastases.54 These alterations could only be identified in
a single small patch of low-grade tumour tissue, suggest
ing that this isolated area gave rise to the metastatic clone
that led to the patients death 17years after resection of
the primary tumour. A full genome sequence of this area
of the primary tumour could not be obtained, as only a
limited number of cells could be microdissected from
the 17-year-old paraffin block. Therefore, it remained
unclear whether the metastatic precursor area con
tained a substantial number of mutations that were
not present in the metastases (indicating some degree
of parallel progression), or whether the metastasis had
inherited all mutations present in the primary tumour
(indicating linear progression). Interestingly, a lymphnode metastasis that was resected along with the primary
tumour also did not contain the mutations present in the
metastaticclone.
That tumour cells can thrive in dramatically differ
ent microenvironments without undergoing extensive
genetic adaptation was demonstrated further by a deepsequencing analysis of a triple-negative breast carci
noma (TNBC), an associated cerebellar metastasis and
a pretreatment biopsy that was propagated as a murine
xenograft.55 48 out of 50 detected somatic point muta
tions were present in all three tumours.55 Interestingly,
mutant allele frequencies were broadly distributed in the
primary tumour (ranging from <10% to 89%), whereas a
heterogeneity reduction took place in both the metasta
sis and the xenograft, with more than 50% of mutations
showing enrichment in the metastasis and/or xenograft.55
This similar enrichment pattern showed that competing
for survival in ectopic microenvironments as different as
human cerebellum and a murine organism might select
for similar sets of tumour-propagating subclones. In both
ectopic samples, the narrowing of the mutant allele fre
quency distribution was not accompanied by outright
loss of any mutations,55 raising the possibility that the
cerebellar metastasissimilar to thexenograftwas

seeded by more than one cell from the primary tumour.


Indeed, evidence indicates that tumour cells can metasta
size as oligoclonal clusters.56 These clusters had a mark
edly increased metastatic potential compared with
single CTCs in a mouse model, and repeated detection
of CTC clusters in the blood of patients with breast or
prostate cancer was correlated with a worse progno
sis.56 Oligoclonal metastasis might be associated with
less genetic divergence between primary tumours and
metastases because the unmasking of mutations during
the expansion of a single metastatic cell at an ectopic site
would not occur. Instead, the distribution of mutant
allele frequencies in the primary tumour might be
preserved to some degree.
By contrast, a single cell sequencing study of a
TNBC and an associated liver metastasis concluded
that metastasis was monoclonal; Navin and colleagues
compared genome-wide copy-number alterations of
100individual cells derived from the two lesions.49 Cells
from the metastasis and the primary tumour were largely
similar to each other, but nevertheless separated into two
distinct branches in an unsupervised neighbour-joining
analysis.49 This result indicated that one cell (or possibly
a cluster of genetically similar cells) from the dominant
clonal population of the primary tumour had founded
the metastasis and that no further mingling had taken
place since then. Again, these results were considered
indicative of linear progression and late dissemination.
Relatively few genome-wide studies have reported
genetic divergence between primary tumours and
metastases. In one example, an index-lesion sequencing
approach was used to compare the mutational spectrum
of a pleural effusion metastasis and a primary lobular
breast carcinoma that was resected 9years earlier. 57
Only 11 of the 32 mutations discovered in the metasta
sis were also present in the primary tumour, indicating
independent somatic evolution of the metastatic clone.
If this result was obtained in synchronously resected
paired tumours, it would support parallel progression.
However, given the long TTR, and the fact that the full
genome sequence of the primary tumour was not avail
able, it remains possible that dissemination occurred
late and that the primary tumour and metastasis would
have been found identical at the time of surgerywith all
the metastasis-specific mutations being acquired as the
metastasis proliferated independently from the primary
tumour for 9years. Therefore, we cannot distinguish
between linear and parallel progression in this case. As
discussed above for the study of Haffner and colleages54
only a full genome sequence of the primary tumour
could resolve this question.
In a more-recent case study of a renal clear-cell carci
noma, Gerlinger and colleagues5 documented more com
prehensively the genetic differences between a primary
tumour and its metastases; the exomes of nine spatially
distinct portions of the primary renal cancer, two syn
chronous metastases, and two pretreatment biopsies
were sequenced.5 The data allowed for several impor
tant observations: firstly, the evolutionary branches
of primary and metastatic clones were found to have

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 7


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
diverged early in the development of the tumour; sec
ondly, a discrete region of the primary tumour harboured
a precursor of the metastatic clone that contained some
mutations that were otherwise metastasis-specific (that
is, found in the metastases and the precursor area of
the primary tumour, but not other parts of the primary
tumour); and finally, the pretreatment biopsies of the
primary tumour and the metastasis clustered with their
post-treatment counterparts, suggesting that treatment
with everolimus had not substantially affected clonal
compositions (although the effectiveness of treatment
was not reported, and could have influenced this obser
vation).5 In samples from a second patient, Gerlinger
etal.5 found that a synchronous metastasis also formed
a distinct phylogenetic branch that separated from the
primary tumour at an early stage of cancer development.
These results are consistent with parallel progression
of the primary tumours and metastases. As in pros
tate cancer, individual metastases were similar to each
other, favouring a cascade scenario. Similar findings
divergence between the primary tumour and associated
metastases, but similarity among metastaseshave been
obtained in paediatricmedulloblastoma.58
In summary, a majority of the genome-wide compari
sons of paired primary tumours and metastases con
ducted to date seem to support the linear progression
model of metastasis, with several notable exceptions. The
extent to which these findings are generalizable remains
to be determined. The clinical course of metastatic cancer
can be extremely variable, and how advanced a cancer is
at the point of analysis and how aggressively it developed
probably considerably influence its genetic landscape. For
example, as far as can be inferred from the clinical infor
mation provided, many of the studies that found genetic
concordance between primary tumours and metastases
involved patients with metastases that were diagnosed
synchronously or within a short period of time after
primary tumour detection, underwent extensive treat
ment and rapidly succumbed to aggressive disease.33,52,55
Samples for comparison are often easier to obtain from
such patients than from patients who develop metachro
nous metastases, but might not accurately represent the
trajectory of more-indolent cancers.

Smaller-scale genetics studies


Although high-resolution genome-wide or whole-exome
comparisons of primary tumours and metastases remain
rare, hundreds of such studies have been conducted
using more-restricted panels of markers or metaphase
CGH. Patient numbers in these studies are typically
larger than in genome-wide analyses, potentially improv
ing understanding of the generalizability of the findings,
and cases supporting linear and parallel progression are
usually found in varying proportions in the same study.
Thought-provoking examples are: deep sequencing of a
cancer mini-genome in primary colorectal cancers and
matched metachronous liver metastases that revealed
vast differences in the number of concordant mutations
among patients; 59 a CGH analysis of primary breast
carcinomas and matched metachronous metastas es

demonstrating close clonal relationships in 69%, and


almost completely unrelated genomic alterations in
31% of patients;60 and, reports of varying frequencies of
discordant mutations in therapeutically or prognosti
cally important genes in lung adenocarcinoma,61 mela
noma,62 and colorectal63 and breast cancers.8 Stoecklein
and Klein64 expertly reviewed many more examples.
These studies suggest that the mode of metastasis varies
between individual patients.

Interpreting comparative genetics data


The diverse results presented above illustrate that we
have yet to arrive at a definitive and coherent under
standing of metastasis in humans. Developing moreaccurate models of metastatic progression will not only
require more-comparative genetics analyses in the meta
static setting, but also resolution of several issues that
complicate interpretation of the data from these studies.
First, meaningful comparison of cancers arising in
different tissues might not always be straightforward, as
the number of mutations common to all cells within a
tumour could vary widely depending on the mutational
history of the tumour founder cell (Figure2). The muta
tional burden of any normal cell increases continually as
it divides and is exposed to environmental stresses over
a patients lifetime. Current estimates are that 50%65 or
more66 of the mutations found in a cancer represent the
fossil record of the cell-division history of the tumour
founder cell; therefore, depending on how frequently
the founder cell divided before it underwent transfor
mation and the extent of other stresses it was exposed to,
the genetic alterations that accumulated during tumour
growth could represent different fractions of the total.
Even if two tumours and their metastases evolved in the
exact same way, the percentage of mutations that are dis
cordant between the primary tumour and its metastasis
will be smaller for the tumour that arose from a founder
cell with a higher baseline mutational burden (Figure2).
These effects should be taken into consideration when
comparing results from different studies, particularly
when the tissues under consideration have different pro
liferative histories. Gaining a greater understanding of
mutation prevalence in normal cells located in different
human tissues would be helpful in this regard.
Second, determining whether tumour self-seeding
has a role in human cancer progression will be impor
tant, because a substantial exchange of cells between
synchronously growing lesions would make genetic
reconstruction of the evolutionary history a tumour very
challenging. Tumour self-seeding seems to be a plausible
explanation for some phenomena observed in genomewide comparisons. For example, in cases in which an
isolated patch in the primary tumour corresponds to a
distant lesion, but is distinct from the dominant clone
in the primary tumour, retrograde metastasis of cells
derived from a disseminated lesion formed early in the
evolution of the disease (with subsequent parallel pro
gression) might be a more parsimonious explanation
than late metastasis of this specific subclone (Figure1c).
Moreover, gene-expression profiles of bulk tumour

8 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
a

100
3

3
6
9

3
6
9

64% divergent
(3+6 out of 14
total mutations)

8% divergent
(3+6 out of 109
total mutations)

Primary tumour
Metastasis

Figure 2 | The mutational burden of the tumour founder


might| Clinical
affect Oncology
Nature cell
Reviews
interpretation of data from comparative genetics studies. a | In the example of a
founder cell with a low mutational burden, the tumour arises from a cell that is
distinguished from the zygote (grey cell) by five mutations (cell divisions are implied
by dotted lines with the number of mutation acquired in the process given
alongside). The tumour follows the parallel progression route of metastasis and
metastasizes at an early stage of disease (curved arrow). When a sample from the
primary tumour (yellowcells) is analysed, 11 mutations are identified with
reference to the germline (five mutations acquired before transformation plus six
mutations acquired during primary tumour development). Three different mutations
that are not found in the primary tumour are detected in the metastasis (green
cells). Of a total of 14mutations found in the tumours (five shared, three
metastasis-specific, six primary-tumour-specific), nine (64%) are different between
primary tumour and metastasis, indicating a high degree of divergence and
therefore parallel evolution. b| If the founder cell has a high mutational burden,
thedegree of divergence might differ, suggesting a different metastasis model.
Forexample, if the same tumour as in panel a arose from a founder cell that has
acquired 100 mutations since the zygote, a total of 109 mutations would be
detected in this cancer (100 shared, threemetastasis-specific, and six primarytumour-specific); therefore, only 8% of mutations differ between the primary tumour
and the metastasis, and linear progression could be erroneously concluded based
on the low degree of divergence.

tissue predict the risk of metastasis,67 and this finding


is difficult to reconcile with metastatic properties being
confined to a very small portion of cells in the primary
tumour. The tumour-self seeding model predicts that
returning seeds are more likely to inhabit the surface
of the primary tumour,68 with the surface defined as
the boundary between the tumour mass and the stroma.
In addition, blood-borne cells returning to the tumour
might be depleted in areas with collapsed vessels.69 In
the future, detailed spatially stratified analysis could
show whether this model holds true in human tumours.
Alternatively, the existence of self-seeding could be
corroborated by demonstrating that a metastasis con
tains definitive precursors of a clone that is also found in
the primary tumour, with an absence of such precursors
in the primary tumour itself (Figure3). The possibility

that the precursor clones were initially present in the


primary tumour, but underwent elimination by purify
ing selection cannot be excluded; however, this type of
analysis could at least provide tentative support for selfseeding. If a relevant degree of self-seeding does indeed
occur, it will be difficult to demonstrate early metastasis
and, therefore, parallel progression, or to discover genetic
variants that are required for early adaptation to specific
microenvironments in late-stagedisease.
Other challenges limit our ability to definitively prove
parallel progression. Although close genetic ties between
a primary tumour and a metastasis (corrected for the
mutational burden of the founder cell and in the absence
of self-seeding) provide credible evidence of late dissemi
nation, genetic divergence does not necessarily imply
early metastasis (Figure4). Because we cannot sample
every single part of a primary tumourin the clinical
setting, analysing all of the tumour is de facto impos
sible because some parts are required for diagnostic
purposesone cannot exclude the possibility that the
area harbouring the pre-metastatic clone was missed,
as shown in previous studies. 33,50 Similarly, if a pre-
metastatic clone is intermingled with other subclones in
the sampled area, but contributes only a small fraction
to the total number of tumour cells, it could escape the
detection limit of the assay.

Studying metastatic lineage

A phylogenetic marker suitable for somatic lineage


tracing should ideally have several important properties:
selective neutrality; a high mutation rate; and acquisi
tion of mutations that is coupled to cell division, such
that thetotal mutational burden of a cell is proportional
to the number of cell divisions since the zygote. The
following section will outline experimental approaches
for somatic lineage reconstruction that meet some or all
of these demands.

Histopathology
We begin with the inspection of a cancers histopatho
logy, which perse is not a phylogenetic method, but
de facto constitutes the oldest, and remains the mostwidely used, method for lineage determination. Even
in molecular-biology-empowered clinical practice,
pathologists use morphological examination to deter
mine whether a lesion is a metastasis or a new primary
tumour, for example, in multifocal lung or breast cancer
cases. Prognosis and treatment can vary widely based
on the outcome of such assessments. The advantage of
this lineage tracing by eye is mainly its convenience.
However, morphological comparison might not always
reliably determine common descent; owing to pheno
typic convergence, comparative approaches become
more reliable as they move from phenotype toward
genotype. For instance, in a pathological evaluation of
lungsquamous-cell carcinoma associated with head
and neck squamous-cell carcinoma, 86% of cases were
diagnosed as metastases by histopathological evaluation,
whereas a molecular assay based on loss of heterozygosity
(LOH) indicated that, in fact, only 43% of lung tumours

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 9


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS

Mut A

Mut AB

Mut ABC

Mut D

Mut DE

Mut DEF

Figure 3 | Using intermediate subclones to searchNature


for evidence
Reviewsof| tumour
ClinicalselfOncology
seeding. In this hypothetical scenario suggesting the existence of self-seeding,
aprimary tumour contains a subclone characterized by mutations in genomic
locations D and E and F (dark blue cells) that are highly similar to those found in
ametastasis; whereas the metastasis contains definite precursors cells that have
mutations in D, or D and E (lighter blue cells), the primary tumour contains no such
precursors. Although loss of the precursors in the primary tumour through clonal
selection cannot be excluded, this scenario predominantly indicates self-seeding of
the primary tumour by a cell(s) from the metastasis. Abbreviation: mut, mutation(s).

were related to the primary head and neck tumour, with


the remainder representing independent transformations
(second primary cancers).70 Furthermore, some tumours
can undergo profound histological changes in response
to treatment.71 Therefore, visual inspection of tumour
morphology is not a preferred method for lineage
tracing, lacking both reliability and resolution (Table2).

Somatic copy-number alterations


A rich literature documents the use of somatic copynumber alterations (SCNAs) to study clonal relation
ships in cancer. SCNAs can be detected relatively easily
and occur in almost all cancer types (Table2). However,
whether they represent good lineage markers is debat
able because many SCNAs probably have selective
effects.72 Convergent evolutionthat is, the indepen
dent occurrence of similar alteration patterns in two
unrelated cellscannot be excluded, unless breakpoints
are mapped in fine detail. For example, amplifications
of chromosome7 and deletions of chromosome10 are
present in >80% of primary glioblastomas,73 despite the
fact that these tumours in different patients are obviously
not related by descent. Some genomic rearrangements
can be induced by exogenous stimuli, such as the sharp
increase of gene fusions between TMPRSS2 and ERG
upon dihydrotestosterone exposure in prostate cancer
cells.74 Some caution in the use of SCNAs for tumour
phylogenetic studies is, therefore, advocated.

Single-nucleotide variants
The problem of selective forces potentially causing arte
facts in lineage reconstruction is relevant when consid
ering not only chromosomal alterations, but also exonic
single-nucleotide variants (SNVs). The exome is the 1% of
the genome that is under the most-intensive evolutionary
pressure and is, therefore, arguably one of the least suit
able targets for lineage analysis; analogous emergence (or
disappearance) of SNVs that provide a selective advantage
(or disappearance of those that are disadvantageous), could
easily be misinterpreted as homology. The impact this issue
has on lineage analysis probably varies depending on the
number of divergent mutations. If several dozens of SNVs
differ between a primary tumour and associated metasta
ses, as has been reported in renal-cell carcinoma,5 many
of the variants are probably neutral and are more likely to
reflect lineage relationships correctly. However, in a hypo
thetical scenario in which only a few mutations (perhaps
limited to cancer-related genes) are shared by multiple
metastases, but not by the primary tumour, convergent
evolution of early disseminated lesions (parallel progres
sion) cannot be as easily dismissed. Nevertheless, SNVs are
correlated with cell division and genome-scale assessment
holds the potential for detailed lineage tracing (Table2).75
Epigenetic approaches
X-chromosome inactivation
Epigenetic modifications have a long history as lineage
markers; Xchromosome inactivation, the random
silencing of one Xchromosome in females during early
embryonic development, has been used extensively to test
clonality both within a tumour mass76 and between differ
ent lesions.60,77 Xchromosome inactivation is a random
and presumably neutral event (Table2), thus fulfillingthe
first criterion of a good lineage marker: in bulk tissue,
theratio between silenced alleles is about 1:1, arguing
against strong selective effects.78 A further advantage is
that silencing is stably heritable. If two cell populations do
not share the same pattern of Xchromosome inactiva
tion, it can be concluded that they have not intermixed
since embryogenesis. However, the static nature of this
marker is also its main limitation, because it cannot
provide any information on evolutionary events that
occurred after the transformation of a tumour founder
cell. However, the inactive Xchromosome acquires muta
tions at an increased rate,79 and therefore might be excep
tionally useful for lineage tracing by providing a reservoir
of presumably selectively neutral somatic variation that
could be tapped for phylogenetic analysis.
CpG methylation analysis
Methylation status of CpG dinucleotides fulfils most of
the criteria for a good phylogenetic marker. A majority
of CpG loci are unmethylated in early development and
acquire heritable cytosine methylation marks with succes
sive rounds of cell division at a rate that is several orders
of magnitude higher than the somatic nucleotide substi
tution rate.80 Neutrality can be assumed when CpGs in
promoters that are not expressed in the tissue of interest
(for example, heart-specific loci such as CSX in colonic

10 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
a

Result:
Divergence

Result:
Divergence

Result:
Divergence

Primary tumour
Metastasis

Nature Reviews
Clinical
Oncology
Figure 4 | The challenge of proving the parallel progression
model |of
metastasis.
When a primary tumour and its metastasis are compared (black boxes indicate the
region sampled) and divergent alterations are found, multiple explanations could
exist. a | Firstly, true parallel progression: dissemination (curved arrow) occurred
inearly stages of tumour development (grey cells), and the primary tumour and
metastasis evolved separately, acquiring distinct genetic alterations (indicated by
different cell colours). b | Secondly, omission of the metastatic precursor cell during
sampling: a clone that is the proximate source of the metastasis (blue cells) does
exist in the primary tumour, but it occupies a spatially constrained region and was
missed during sampling. c | Finally, limited assay sensitivity: the clone that is the
source of the metastasis (blue cells) does exist in the primary tumour, but at such a
low frequency that it falls below the detection limit of the assay used for comparison.

tissue) are examined (Table2). CpG methylation repre


sents a useful molecular clock,81 and has been used exten
sively to study stem cell82,83 and tumour 8486 lineages in
humans. However, an important concern is that cytosine
methylation is a reversible mark and could potentially be
affected by genome-wide methylation changes that occur
during tumorigenesis.87,88 Apermanent change in DNA
sequence would, therefore, be preferable to methylation
for purposes of lineagetracing.

Microsatellite analysis
Microsatellites, short tandem repeats of 14bp units in
DNA, arguably come as close as possible to being optimal
somatic lineage markers. Most are noncoding and show
high levels of polymorphism.89 Mutations typically occur
in the form of insertions or deletions of one or multiple
units through slippage of DNA polymerase during DNA
replication,90 and are thus tightly coupled to cell division
(Table2). Mutation rates vary depending on the size and
length of the repeat, but can be much higher than the
average genome-wide mutation rate of approximately 109
per base per division;91 for example, the mutation rate of a
typical (CA)17 dinucleotide repeat in human cells is on the
order of 107100 times higher.92
Microsatellites first entered the spotlight in cancer
genetics when frequent somatic length-polymorphisms
in these sequences were discovered in familial colo
rectalcancer in patients with hereditary nonpolyposis
colorectal cancer (HNPCC; Lynch syndrome).93,94 The
phenomenon, coined microsatellite instability (MSI), was

also observed in 1015% of sporadic colorectal cancers


and was associated with an improved prognosis compared
with cancers lacking MSI.95 Subsequently, MSI was found
to be associated with mutations in DNA mismatch repair
(MMR)genes.96,97
Microsatellites have been used as molecular clocks
of tumour evolution in MSI-positive human cancers.
Shibata and colleagues 81 showed that dinucleotide
repeat-length distributions vary across tumour regions
in patients with HNPCC and suggested that intratumour
heterogeneity is related to mitotic age, with older regions
displaying more diversity. Interestingly, this group found
that mitotic ages in MMR-deficient adenomas and carci
nomas were similar.98 Microsatellite mutations have also
been used to reconstruct the phylogenetic relationships
between single cells in MMR-deficient mice99102 and
human leukaemia cells.103
In 2006, Salipante and Horwitz104 introduced a novel
methodology for somatic lineage tracing that relied on
a class of particularly mutable guanine mononucleotide
repeats: polyguanine (polyG) tracts can reach mutation
rates of 106 per base per division in human cells,92 and thus
mutate approximately 10100-times faster than dinucleo
tide repeats and 1000-times faster than nonrepetitive DNA
sequences. Analysis of polyG tracts was subsequently
used to study various aspects of murine development in
MMR-proficient animals.105108 Importantly, this technique
was also shown to be applicable in the human setting when
pre-neoplastic clonal expansions marked by polyG muta
tions were identified as a prodrome of cancer development
in patients with ulcerative colitis.109
We have recently demonstrated that polyG profil
ing can be used for lineage tracing in metastatic colon
cancer.110 Interrogation of only 20 markers yielded suf
ficient information to build robust phylogenetic trees
that reflected the evolution of metastasis in individual
patients.110 Many of the biological insights gleaned from
these trees recapitulated observations from exome or
whole-genome sequencing studies: isolated areas of
large, diversified primary tumours giving rise to wide
spread metastasis, or genetic divergence between lymph
node and distant metastases.110 PolyG tract profiling is
a simple and cost-effective PCR-based methodology that
works well with formalin-fixed and paraffin-embedded
tissues. As such, it could easily be used to study tumour
lineage in large numbers of samples that are collected as
part of routine clinical care. Importantly, in contrast to
whole-genome or exome sequencing, polyG tract analysis
does not produce large amounts of personal genetic infor
mation that might be associated with ethical concerns.
Most institutional guidelines prohibit whole-genome
or even exome sequencing of archival tissues without
explicit consent due to such ethical issues; hence, the large
numbers of valuable cancer samples stored in pathology
departments worldwide cannot be used for these purposes
because patients would have to be contacted retrospec
tively to gain consent for new analyses. By contrast, polyG
tract profiling can be performed under most discarded
tissue protocols, thereby opening up additional tissue
resources for analysis of intratumourheterogeneity.

NATURE REVIEWS | CLINICAL ONCOLOGY

ADVANCE ONLINE PUBLICATION | 11


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
Table 2 | Experimental approaches for studying cell lineage in human cancer
Lineage marker

Advantages

Disadvantages

Histopathology

Fast and convenient


Assessed as part of routine clinical care

Can be unreliable
Limited resolution

Somatic
copy-number
alterations

Present in many cancers


Specific alterations are easy to assess
(forexample, FISH for HER2 amplification)

Selective effects might confound interpretation (due to


convergent evolution) unless precise breakpoints are known
Unknown relationship with cell division

Single-nucleotide
variants

Comprehensive and unbiased when


assessed on a genome-wide scale
At least partially correlated with cell division

Selective effects might confound interpretation if only a limited


number of mutations in cancer-related genes are considered

X-chromosome
inactivation

Selectively neutral
Relatively easy to assess, also insitu

Limited resolution
Might be affected by DNA methylation alterations in cancer

DNA methylation

Selectively neutral loci available


Correlated with cell division

Might be affected by DNA methylation alterations in cancer

Microsatellites

Selectively neutral loci available


Correlated with cell division
Permanent change

Hypermutable sequences can be difficult to evaluate

Abbreviation: FISH, fluorescence insitu hybridization.

Whole-genome studies
Ultimately, of course, if informed consent has been given,
whole-genome sequencing is the most-comprehensive
technique for somatic lineage tracing in human tissue
samples. Whole-genome analyses of paired primary
tumours and metastases remain rare, but the insights
into clonal evolution that we have already gained from
sequencing the genomes of single tumour samples
provide a glimpse of future possibilities. Whole-genome
sequencing is also the only way to assess comprehensively
and comparatively how driver and passenger mutations
evolve during cancer development and metastasis. From
a phylogenetic perspective, passenger mutations are of
great interest because they constitute a record of cell
division and other mutagenic processes. Driver muta
tions, on the other hand, carry the bulk of functional
and clinical relevance. These two mutation classes might
behave very differently during disease progression. For
example, important driver mutations might remain con
cordant (or evolve convergently), while large numbers of
passenger mutations diverge between primary tumours
and metastases. It is important to point out here that
although experimental evidence suggests that replica
tion slippage mutations in normal, MMR-competent
cells are correlated with cell division, this is likely not
the case for genome-wide mutation patterns because a
variety of influences (carcinogen exposure,111 APOBEC
activity, 111,112 or drug-treatment effects, 113 among
others) shape the mutational landscape in addition to
replication-inducederrors.

Future directions

As our knowledge of genetic heterogeneity in cancer


expands, the question of what concrete clinical advantages
we can expect to gain from an improved understanding
of the lineage of metastases remains. Contemporary
clinicians might have a theoretical interest in knowing
which cancers undergo linear or parallel progression,
but they will still want to confirm, by all possible means
(grid sampling of primary tumours, additional biopsy

of metastases or liquid biopsy), what actionable genetic


alterations are present in a specific tumour in individual
patients. Indeed, insights from cancer phylogenetics
might only manifest in the clinic in the mid-to-long
term, but once established are likely to have an important
impact on patient care. Parallel progression postulates
that the ability to metastasize is an inherent qualitative
trait that manifests in the earliest stages of tumour pro
gression. This view de-emphasizes the importance of
clonal evolution at the primary site, as mutations that
are acquired later on in tumour development are likely
to be mere instigators of local proliferation with limited
relevance to overall disease outcome. From the perspec
tive of linear progression, these mutations arethe ultimate
cause of death. As we sequence the DNA of more and
more tumour samples in search of mechanistic insights
and novel drug targets, deriving a phylogenetically
informed model that provides clear predictions of which
mutationsthose in the trunk of the phylogenetic tree
of cancer, or those in the branches or leavesare the root
cause of systemic spread will beimportant.
Many more clinically relevant questions related to
tumour lineage await resolution. For example, a system
atic analysis of how the degree of intratumour hetero
geneity changes at different stages of disease progression
would be instructive; few studies have been conducted
in this area. High levels of heterogeneity are connected
with increased malignant potential in the early stages of
tumorigenesis. For instance, clonal diversity in Barrett
oesophagus, a premalignant condition, is linked with a
higher likelihood of progression to cancer.114 The final
step of systemic disease advancement, metastasis, on the
other hand, seems to go hand-in-hand with a decrease
in heterogeneity. This pattern makes intuitive sense, as
metastasis represents an evolutionary bottleneck, and the
reduced heterogeneity in systemic disease is evident on
multiple levels of observation: narrowing of the mutantallele frequency distribution in a breast cancer metasta
sis;55 monoclonality of metastases at the time of death in
prostate cancer (despite the fact that localized prostate

12 | ADVANCE ONLINE PUBLICATION

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
cancer is often multifocal and highly heterogeneous);52,54
increased homogeneity of genomic rearrangements in
DTCs derived from patients with overt metastasis versus
those with minimal residual disease; 43,46 and homo
geneity of metastases in comparison with the primary
tumour in mCRC.110 Parallel comparison of cancers at
different stages of progression with standardized tech
niques will elucidate further whether changes in diversity
accompany progression.
In addition, how the decline in heterogeneity that
potentially occurs in metastatic disease relates to treat
ment response remains unclear. That genetic hetero
geneity leads to decreased treatment response is a widely
held belief,115,116 which should at least hold true in the
case of driver mutation heterogeneity. According to
the linear progression model, metastases are younger
clonal expansions than primary tumours; hence,
metastases should be less diverse and more responsive
to therapy. This hypothesis seems probable for some
cancer types. Clinical studies have shown that primary
lung carcinomas, for example, are much less likely to
respond to chemotherapy than synchronous metasta
ses (response rates of 11.8% versus 32.8%).117 For breast
cancer, however, this pattern is reversed (40% versus
19.8% response rate for primary tumours compared
with metastases).117 Microenvironmental cues probably
have an important influence on the differential thera
peutic responses of primary tumours and metastases,118
but more-rigorous investigation of how intratumour
heterogeneity relates to treatment outcomes will also be
important in the future.
Another interesting and unanswered question is how
spatial diversificationthe presence of distinct clones
1.

2.
3.

4.

5.

6.

7.

8.

9.

Holohan, C., Van Schaeybroeck, S.,


Longley,D.B. & Johnston, P.G. Cancer drug
resistance: an evolving paradigm. Nat. Rev.
Cancer 13, 714726 (2013).
Navin, N.E. & Hicks, J. Tracing the tumor
lineage. Mol. Oncol. 4, 267283 (2010).
Swanton, C. Intratumor heterogeneity: evolution
through space and time. Cancer Res. 72,
48754882 (2012).
Marusyk, A., Almendro, V. & Polyak, K.
Intratumour heterogeneity: a looking glass for
cancer? Nat. Rev. Cancer 12, 323334 (2012).
Gerlinger, M. etal. Intratumor heterogeneity
andbranched evolution revealed by multiregion
sequencing. N.Engl. J.Med. 366, 883892
(2012).
Sankin, A. etal. The impact of genetic
heterogeneity on biomarker development
inkidney cancer assessed by multiregional
sampling. Cancer Med. http://dx.doi.org/
10.1002/cam4.293 (2014).
Simmons, C. etal. Does confirmatory tumor
biopsy alter the management of breast cancer
patients with distant metastases? Ann. Oncol.
20, 14991504 (2009).
Niikura, N. etal. Latest biopsy approach for
suspected metastases in patients with breast
cancer. Nat. Rev. Clin. Oncol. 10, 711719
(2013).
Jain, R.K. Normalizing tumor microenvironment
to treat cancer: bench to bedside to
biomarkers. J. Clin. Oncol. 31, 22052218
(2013).

in different regions of a tumourrelates to strictly local


heterogeneity (intermingling of clones), and how these
forms of diversity associate with clinical behaviour.
More-invasive and motile cells might generate moredispersed forms of genetic heterogeneity (as proliferation
of subclones would be accompanied by migration). It has
furthermore been suggested that locally coexisting clones
might be able to develop commensal relationships.119
One could, therefore, speculate that high levels of local
heterogeneity are a hallmark of aggressive disease,
whereas regional clonal expansions that stay delineated
from each other might signify more-indolent pheno
types. Currently, no data exist to corroborate or refute
this hypothesis, necessitating further studies.

Conclusions

Gaining a more-comprehensive and instructive under


standing of cancer progression includes furthering our
knowledge of when metastases are seeded and their
genetic relationships to the primary tumour. Different
cancers probably metastasize along different paths and
timelines, and much remains to be learned about the
biology underlying these behaviours. Reconstructing
thetree of life of metastatic cancer in representative
patient populations will be important in this regard.
Amajority of genome-wide comparisons document
similar mutation profiles in primary tumours and
metastas es, but thought-provoking counterexamples
also exist. We have outlined a variety of experimental
techniques for the purpose of lineage tracing in human
cancer. Owing to these approaches, the answers to
many crucial questions in the field of intratumour
heterogeneity are now within our reach.

10. Weiss, L. Concepts of metastasis.


CancerMetastasis Rev. 19, 219234 (2000).
11. Klein, C.A. Parallel progression of primary
tumours and metastases. Nat. Rev. Cancer 9,
302312 (2009).
12. Klein, C.A. Framework models of tumor
dormancy from patient-derived observations.
Curr. Opin. Genet. Dev. 21, 4249 (2011).
13. Hess, K.R., Pusztai, L., Buzdar, A.U.
&Hortobagyi, G.N. Estrogen receptors
anddistinct patterns of breast cancer relapse.
Breast Cancer Res. Treat. 78, 105118 (2003).
14. Tsai, M.S. et al. Clinicopathological features
andprognosis in resectable synchronous and
metachronous colorectal liver metastasis.
Ann.Surg. Oncol. 14, 786794 (2007).
15. Bragado, P., Sosa, M.S., Keely, P., Condeelis, J.
& Aguirre-Ghiso, J.A. Microenvironments
dictating tumor cell dormancy. Recent Results
Cancer Res. 195, 2539 (2012).
16. Talmadge, J.E. & Fidler, I.J. AACR centennial
series: the biology of cancer metastasis:
historical perspective. Cancer Res. 70,
56495669 (2010).
17. Duda, D.G. etal. Malignant cells facilitate lung
metastasis by bringing their own soil. Proc. Natl
Acad. Sci. USA 107, 2167721682 (2010).
18. Teng, M.W., Swann, J.B., Koebel, C.M.,
Schreiber, R.D. & Smyth, M.J. Immunemediated dormancy: an equilibrium with cancer.
J. Leukoc. Biol. 84, 988993 (2008).
19. McAllister, S.S. & Weinberg, R.A. The tumourinduced systemic environment as a critical

NATURE REVIEWS | CLINICAL ONCOLOGY

20.
21.

22.

23.

24.

25.

26.

27.

regulator of cancer progression and metastasis.


Nat. Cell Biol. 16, 717727 (2014).
Cairns, J. Mutation selection and the natural
history of cancer. Nature 255, 197200 (1975).
Bross, I.D., Viadana, E. & Pickren, J.
Dogeneralized metastases occur directly
fromthe primary? J. Chronic Dis. 28, 149159
(1975).
Weinberg, R.A. Mechanisms of malignant
progression. Carcinogenesis 29, 10921095
(2008).
Disibio, G. & French, S.W. Metastatic patterns
of cancers: results from a large autopsy study.
Arch. Pathol. Lab. Med. 132, 931939 (2008).
Hellman, S. Karnofsky Memorial Lecture. Natural
history of small breast cancers. J.Clin. Oncol.
12, 22292234 (1994).
Fisher, B. Laboratory and clinical research
inbreast cancera personal adventure:
theDavid,A. Karnofsky memorial lecture.
CancerRes. 40, 38633874 (1980).
Giuliano, A.E. etal. Axillary dissection vs no
axillary dissection in women with invasive breast
cancer and sentinel node metastasisa
randomized clinical trial. JAMA 305, 569575
(2011).
Giuliano, A.E. etal. Locoregional recurrence
after sentinel lymph node dissection with or
without axillary dissection in patients with
sentinel lymph node metastases: the American
College of Surgeons Oncology Group Z0011
randomized trial. Ann. Surg. 252, 426432
(2010).

ADVANCE ONLINE PUBLICATION | 13


2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
28. Klein, C.A. Selection and adaptation during
metastatic cancer progression. Nature 501,
365372 (2013).
29. Husemann, Y. etal. Systemic spread is an early
step in breast cancer. Cancer Cell 13, 5868
(2008).
30. Kim, M.Y. etal. Tumor self-seeding by circulating
cancer cells. Cell 139, 13151326 (2009).
31. Aguirre-Ghiso, J.A. Models, mechanisms and
clinical evidence for cancer dormancy. Nat. Rev.
Cancer 7, 834846 (2007).
32. Jones, S. etal. Comparative lesion sequencing
provides insights into tumor evolution. Proc. Natl
Acad. Sci. USA 105, 42834288 (2008).
33. Yachida, S. etal. Distant metastasis occurs late
during the genetic evolution of pancreatic
cancer. Nature 467, 11141117 (2010).
34. Meng, S. etal. Circulating tumor cells in patients
with breast cancer dormancy. Clin. Cancer Res.
10, 81528162 (2004).
35. Nagrath, S. etal. Isolation of rare circulating
tumour cells in cancer patients by microchip
technology. Nature 450, 12351239 (2007).
36. Riethdorf, S., Wikman, H. & Pantel, K.
Review:Biological relevance of disseminated
tumor cells in cancer patients. Int. J. Cancer 123,
19912006 (2008).
37. Cristofanilli, M. etal. Circulating tumor cells,
disease progression, and survival in metastatic
breast cancer. N.Engl. J.Med. 351, 781791
(2004).
38. Danila, D.C. etal. Circulating tumor cell number
and prognosis in progressive castration-resistant
prostate cancer. Clin. Cancer Res. 13,
70537058 (2007).
39. Krebs, M.G. etal. Evaluation and prognostic
significance of circulating tumor cells in patients
with nonsmallcell lung cancer. J. Clin. Oncol. 29,
15561563 (2011).
40. Crowley, E., Di Nicolantonio, F., Loupakis, F.
&Bardelli, A. Liquid biopsy: monitoring cancergenetics in the blood. Nat. Rev. Clin. Oncol. 10,
472484 (2013).
41. Alix-Panabires, C. & Pantel, K. Circulating tumor
cells: liquid biopsy of cancer. Clin. Chem. 59,
110118 (2013).
42. Schmidt-Kittler, O. etal. From latent
disseminated cells to overt metastasis: genetic
analysis of systemic breast cancer progression.
Proc. Natl Acad. Sci. USA 100, 77377742
(2003).
43. Weckermann, D. etal. Perioperative activation
ofdisseminated tumor cells in bone marrow of
patients with prostate cancer. J.Clin. Oncol. 27,
15491556 (2009).
44. Schardt, J.A. etal. Genomic analysis of single
cytokeratin-positive cells from bone marrow
reveals early mutational events in breast cancer.
Cancer Cell 8, 227239 (2005).
45. Klein, C.A. etal. Comparative genomic
hybridization, loss of heterozygosity, and DNA
sequence analysis of single cells. Proc. Natl
Acad. Sci. USA 96, 44944499 (1999).
46. Klein, C.A. etal. Genetic heterogeneity of single
disseminated tumour cells in minimal residual
cancer. Lancet 360, 683689 (2002).
47. Wang, Y. etal. Clonal evolution in breast cancer
revealed by single nucleus genome sequencing.
Nature 512, 155160 (2014).
48. Xu, X. etal. Single-cell exome sequencing reveals
single-nucleotide mutation characteristics of a
kidney tumor. Cell 148, 886895 (2012).
49. Navin, N. etal. Tumour evolution inferred by
single-cell sequencing. Nature 472, 9094
(2011).
50. Brannon, A. etal. Comparative sequencing
analysis reveals high genomic concordance
between matched primary and metastatic

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

colorectal cancer lesions. Genome Biol. 15, 454


(2014).
Campbell, P.J. etal. The patterns and dynamics
of genomic instability in metastatic pancreatic
cancer. Nature 467, 11091113 (2010).
Liu, W. etal. Copy number analysis indicates
monoclonal origin of lethal metastatic prostate
cancer. Nat. Med. 15, 559565 (2009).
Letouz, E., Allory, Y., Bollet, M.A., Radvanyi, F.
&Guyon, F. Analysis of the copy number profiles
of several tumor samples from the same patient
reveals the successive steps in tumorigenesis.
Genome Biol. 11, R76 (2010).
Haffner, M.C. etal. Tracking the clonal origin
oflethal prostate cancer. J. Clin. Invest. 123,
49184922 (2013).
Ding, L. etal. Genome remodelling in a basal-like
breast cancer metastasis and xenograft. Nature
464, 9991005 (2010).
Aceto, N. etal. Circulating tumor cell clusters
areoligoclonal precursors of breast cancer
metastasis. Cell 158, 11101122 (2014).
Shah, S.P. etal. Mutational evolution in a
lobular breast tumour profiled at single
nucleotide resolution. Nature 461, 809813
(2009).
Wu, X. etal. Clonal selection drives genetic
divergence of metastatic medulloblastoma.
Nature 482, 529533 (2012).
Vermaat, J.S. etal. Primary colorectal cancers
and their subsequent hepatic metastases are
genetically different: implications for selection
ofpatients for targeted treatment. Clin. Cancer
Res. 18, 688699 (2012).
Kuukasjarvi, T. etal. Genetic heterogeneity
andclonal evolution underlying development
ofasynchronous metastasis in human breast
cancer. Cancer Res. 57, 15971604 (1997).
Schmid, K. etal. EGFR/KRAS/BRAF mutations
inprimary lung adenocarcinomas and
corresponding locoregional lymph node
metastases. Clin. Cancer Res. 15, 45544560
(2009).
Colombino, M. etal. BRAF/NRAS mutation
frequencies among primary tumors and
metastases in patients with melanoma. J. Clin.
Oncol. 30, 25222529 (2012).
Baldus, S.E. etal. Prevalence and heterogeneity
of KRAS, BRAF, and PIK3CA mutations in primary
colorectal adenocarcinomas and their
corresponding metastases. Clin. Cancer Res. 16,
790799 (2010).
Stoecklein, N.H. & Klein, C.A. Genetic disparity
between primary tumours, disseminated tumour
cells, and manifest metastasis. Int. J. Cancer
126, 589598 (2010).
Tomasetti, C., Vogelstein, B. & Parmigiani, G.
Half or more of the somatic mutations in
cancers of self-renewing tissues originate prior
to tumor initiation. Proc. Natl Acad. Sci. USA 110,
19992004 (2013).
Welch, J.S. etal. The origin and evolution of
mutations in acute myeloid leukemia. Cell 150,
264278 (2012).
vant Veer, L.J. etal. Gene expression profiling
predicts clinical outcome of breast cancer.
Nature 415, 530536 (2002).
Comen, E., Norton, L. & Massagu, J. Clinical
implications of cancer self-seeding. Nat. Rev.
Clin. Oncol. 8, 369377 (2011).
Jain, R.K., Martin, J.D. & Stylianopoulos, T.
Therole of mechanical forces in tumor growth
and therapy. Annu. Rev. Biomed. Eng. 16,
321346 (2014).
Geurts, T.W. etal. Pulmonary squamous cell
carcinoma following head and neck squamous
cell carcinoma: metastasis or second primary?
Clin. Cancer Res. 11, 66086614 (2005).

14 | ADVANCE ONLINE PUBLICATION

71. Sequist, L.V. etal. Genotypic and histological


evolution of lung cancers acquiring resistance
toEGFR inhibitors. Sci. Transl. Med. 3, 75ra26
(2011).
72. Davoli, T. etal. Cumulative haploinsufficiency
and triplosensitivity drive aneuploidy patterns
and shape the cancer genome. Cell 155,
948962 (2013).
73. Beroukhim, R. etal. Assessing the significance
ofchromosomal aberrations in cancer:
methodology and application to glioma. Proc.Natl
Acad. Sci. USA 104, 2000720012 (2007).
74. Mani, R.S. etal. Induced chromosomal proximity
and gene fusions in prostate cancer. Science
326, 1230 (2009).
75. Behjati, S. etal. Genome sequencing of normal
cells reveals developmental lineages and
mutational processes. Nature 513, 422425
(2014).
76. Going, J.J., Abd El-Monem, H.M. & Craft, J.A.
Clonal origins of human breast cancer. J.Pathol.
194, 406412 (2001).
77. Katona, T.M. etal. Genetically heterogeneous
and clonally unrelated metastases may arise in
patients with cutaneous melanoma. Am. J. Surg.
Pathol. 31, 10291037 (2007).
78. Novelli, M. etal. Xinactivation patch size
inhuman female tissue confounds the
assessment of tumor clonality. Proc. Natl Acad.
Sci. USA 100, 33113314 (2003).
79. Jger, N. etal. Hypermutation of the inactive X
chromosome is a frequent event in cancer. Cell
155, 567581 (2013).
80. Shibata, D. & Tavar, S. Counting divisions in
ahuman somatic cell tree: how, what and why?
Cell Cycle 5, 610614 (2006).
81. Shibata, D., Navidi, W., Salovaara, R., Li, Z.H.
&Aaltonen, L.A. Somatic microsatellite
mutations as molecular tumor clocks. Nat. Med.
2, 676681 (1996).
82. Nicolas, P., Kim, K.M., Shibata, D. & Tavare, S.
The stem cell population of the human colon
crypt: analysis via methylation patterns.
PLoSComput. Biol. 3, e28 (2007).
83. Yatabe, Y., Tavare, S. & Shibata, D. Investigating
stem cells in human colon by using methylation
patterns. Proc. Natl Acad. Sci. USA 98,
1083910844 (2001).
84. Woo, Y.J., Siegmund, K.D., Tavare, S. &
Shibata,D. Older individuals appear to acquire
mitotically older colorectal cancers. J.Pathol.
217, 483488 (2009).
85. Siegmund, K.D., Marjoram, P., Woo, Y.J.,
Tavare,S. & Shibata, D. Inferring clonal expansion
and cancer stem cell dynamics from DNA
methylation patterns in colorectal cancers. Proc.
Natl Acad. Sci. USA 106, 48284833 (2009).
86. Siegmund, K.D., Marjoram, P., Tavare, S.
&Shibata, D. High DNA methylation pattern
intratumoral diversity implies weak selection
inmany human colorectal cancers. PLoS ONE 6,
e21657 (2011).
87. Baylin, S.B. & Jones, P.A. A decade of exploring
the cancer epigenomebiological and
translational implications. Nat. Rev. Cancer 11,
726734 (2011).
88. Feinberg, A.P. & Vogelstein, B. Hypomethylation
distinguishes genes of some human cancers
from their normal counterparts. Nature 301,
8992 (1983).
89. Ellegren, H. Microsatellites: simple sequences
with complex evolution. Nat. Rev. Genet. 5,
435445 (2004).
90. Strand, M., Prolla, T.A., Liskay, R.M. &
Petes,T.D. Destabilization of tracts of simple
repetitive DNA in yeast by mutations affecting
DNA mismatch repair. Nature 365, 274276
(1993).

www.nature.com/nrclinonc
2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
91. Lynch, M. Evolution of the mutation rate.
TrendsGenet. 26, 345352 (2010).
92. Boyer, J.C. etal. Sequence dependent instability
of mononucleotide microsatellites in cultured
mismatch repair proficient and deficient
mammalian cells. Hum. Mol. Genet. 11,
707713 (2002).
93. Ionov, Y., Peinado, M.A., Malkhosyan, S.,
Shibata, D. & Perucho, M. Ubiquitous somatic
mutations in simple repeated sequences
reveala new mechanism for colonic
carcinogenesis. Nature 363, 558561
(1993).
94. Aaltonen, L.A. etal. Clues to the pathogenesis
of familial colorectal cancer. Science 260,
812816 (1993).
95. Samowitz, W.S. etal. Microsatellite instability
insporadic colon cancer is associated with an
improved prognosis at the population level.
Cancer Epidemiol. Biomarkers Prev. 10, 917923
(2001).
96. Bonadona, V. etal. Cancer risks associated
withgermline mutations in MLH1, MSH2,
andMSH6 genes in Lynch syndrome. JAMA 305,
23042310 (2011).
97. Fishel, R. etal. The human mutator gene
homolog MSH2 and its association with
hereditary nonpolyposis colon cancer. Cell 75,
10271038 (1993).
98. Tsao, J.L. etal. Genetic reconstruction of
individual colorectal tumor histories. Proc. Natl
Acad. Sci. USA 97, 12361241 (2000).
99. Wasserstrom, A. etal. Reconstruction of cell
lineage trees in mice. PLoS ONE 3, e1939
(2008).
100. Reizel, Y. etal. Colon stem cell and crypt
dynamics exposed by cell lineage
reconstruction. PLoS Genet. 7, e1002192
(2011).
101. Reizel, Y. etal. Cell lineage analysis of the
mammalian female germline. PLoS Genet. 8,
e1002477 (2012).

102. Frumkin, D. etal. Cell lineage analysis of a


mouse tumor. Cancer Res. 68, 59245931
(2008).
103. Shlush, L.I. etal. Cell lineage analysis of acute
leukemia relapse uncovers the role of
replication-rate heterogeneity and microsatellite
instability. Blood 120, 603612 (2012).
104. Salipante, S.J. & Horwitz, M.S. Phylogenetic
fate mapping. Proc. Natl Acad. Sci. USA 103,
54485453 (2006).
105. Salipante, S.J., Kas, A., McMonagle, E.
&Horwitz, M.S. Phylogenetic analysis of
developmental and postnatal mouse cell
lineages. Evol. Dev. 12, 8494 (2010).
106. Salipante, S.J., Thompson, J.M. &
Horwitz,M.S. Phylogenetic fate mapping:
theoretical and experimental studies applied to
the development of mouse fibroblasts. Genetics
178, 967977 (2008).
107. Zhou, W. etal. Use of somatic mutations
toquantify random contributions to mouse
development. BMC Genomics 14, 39 (2013).
108. Salk, J.J. & Horwitz, M.S. Passenger mutations
as a marker of clonal cell lineages in emerging
neoplasia. Semin. Cancer Biol. 20, 294303
(2010).
109. Salk, J.J. etal. Clonal expansions in ulcerative
colitis identify patients with neoplasia. Proc. Natl
Acad. Sci. USA 106, 2087120876 (2009).
110. Naxerova, K. etal. Hypermutable DNA
chronicles the evolution of human colon cancer.
Proc. Natl Acad. Sci. USA 111, E1889E1898
(2014).
111. de Bruin, E.C. etal. Spatial and temporal
diversity in genomic instability processes
defines lung cancer evolution. Science 346,
251256 (2014).
112. Nik-Zainal, S. etal. Mutational processes
molding the genomes of 21 breast cancers.
Cell149, 979993 (2012).
113. Johnson, B.E. etal. Mutational analysis reveals
the origin and therapy-driven evolution of

NATURE REVIEWS | CLINICAL ONCOLOGY

recurrent glioma. Science 343, 189193


(2014).
114. Maley, C.C. etal. Genetic clonal diversity
predicts progression to esophageal
adenocarcinoma. Nat.Genet. 38, 468473
(2006).
115. Burrell, R.A. & Swanton, C. Tumour
heterogeneity and the evolution of polyclonal
drug resistance. Mol. Oncol. 8, 10951111
(2014).
116. Shibata, D. Cancer. Heterogeneity and tumor
history. Science 336, 304305 (2012).
117. Slack, N.H. & Bross, I.D. The influence of site
of metastasis on tumour growth and response
to chemotherapy. Br. J. Cancer 32, 7886
(1975).
118. Kodack, D.P. etal. Combined targeting of HER2
and VEGFR2 for effective treatment of HER2amplified breast cancer brain metastases.
Proc.Natl Acad. Sci. USA 109, E3119E3127
(2012).
119. Merlo, L.M. F., Pepper, J.W., Reid, B.J.
&Maley,C.C. Cancer as an evolutionary and
ecological process. Nat. Rev. Cancer 6, 924935
(2006).
Acknowledgements
The work of the authors is supported in part by
funding from the US Department of Defence
(grantW81XWH111-0146 to K.N.), the Breast
Cancer Innovator Award (grant W81XWH-10-1-0016
toR.K.J.), the US National Institutes of Health
(grantsP01CA080124 and R01CA163815 to
R.K.J.),the Proton Beam/Federal Share Program
(R.K.J.), andthe National Foundation for Cancer
Research (R.K.J.).
Author contributions
K.N. and R.K.J. made substantial contributions
todiscussion of content and review/editing of the
manuscript before submission. K.N. researched
thedata for the article and wrote the manuscript.

ADVANCE ONLINE PUBLICATION | 15


2015 Macmillan Publishers Limited. All rights reserved

Anda mungkin juga menyukai