The n e w e ng l a n d j o u r na l of m e dic i n e

Review Article

Frontiers in Medicine

Application of Cell-free DNA Analysis

to Cancer Treatment
Ryan B. Corcoran, M.D., Ph.D., and Bruce A. Chabner, M.D.​​

From the Massachusetts General Hos­
pital Cancer Center and the Department
of Medicine, Harvard Medical School,
Boston. Address reprint requests to Dr.
Corcoran at Massachusetts General Hos­
pital Cancer Center, 149 13th St., 7th Fl.,
Boston, MA 02129, or at ­rbcorcoran@​
­partners​.­org.
N Engl J Med 2018;379:1754-65.
DOI: 10.1056/NEJMra1706174
Copyright © 2018 Massachusetts Medical Society.
T
umor biopsies represent the standard for cancer diagnosis
and the primary method for molecular testing to guide the selection of
precision therapies. Liquid biopsies, particularly those involving cell-free
DNA (cfDNA) from plasma, are rapidly emerging as an important and minimally
invasive adjunct to standard tumor biopsies and, in some cases, even a potential
alternative approach. Liquid biopsy is becoming a valuable tool for molecular test-
ing, for new insights into tumor heterogeneity, and for cancer detection and
monitoring. Here, we review the current and potential clinical applications of
cfDNA analysis in patients with cancer (see video).

L iquid Biopsie s
Although liquid biopsy has most often referred to the analysis of cfDNA from
peripheral blood, this term also encompasses the isolation and analysis of tumor-
derived material (e.g., DNA, RNA, or even intact cells) from blood or other bodily
An illustrated fluids1,2 (Fig. 1). For example, intact circulating tumor cells intravasate into the
glossary and a bloodstream at low frequency (often <10 circulating tumor cells per milliliter of
video overview of blood in patients with metastatic cancer). With specialized technology, circulating
cfDNA analysis
tumor cells can be detected and isolated from a background of normal blood cells,
in cancer are
available at facilitating molecular analysis and even implantation and growth of these cells in
NEJM.org immunocompromised mice.1-3 Subcellular particles called exosomes, or extracellular
membrane-encased vesicles, are also released by tumor cells into the bloodstream
and contain tumor-specific proteins and nucleic acids.4 Cell-free nucleic acids, in-
cluding cell-free RNA (which is less stable than DNA)5 and cfDNA (the focus of
this review), are also released into the circulation.
Blood is not the only bodily fluid that can be used for liquid-biopsy approaches.
Urine, stool, cerebrospinal fluid (CSF), saliva, pleural fluid, and ascites are all
potential sources of tumor-derived material, including cfDNA.2,6-8 Although this
review will focus on the analysis of cfDNA from blood, detection and analysis of
cfDNA from other bodily fluids may have applications for specific cancer types
(e.g., CSF for cancers of the central nervous system) or for the detection of cancers
arising in defined organ systems (e.g., stool for colorectal cancer or saliva for head
and neck cancers).

Pl a sm a-Der i v ed cf DNA
The term “cfDNA” refers to fragmented DNA found in the noncellular component
of the blood, as first reported by Mandel and Metais in 1948.9 It is thought that
cfDNA is released into the bloodstream through apoptosis or necrosis, and cfDNA
is typically found as double-stranded fragments of approximately 150 to 200 base
pairs in length, corresponding to nucleosome-associated DNA.10 Molecules of
cfDNA are rapidly cleared from the circulation, with a half-life of an hour or less.

1754 n engl j med 379;18 nejm.org  November 1, 2018

The New England Journal of Medicine

Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Cell-free DNA Analysis and Cancer Treatment

Liquid-Biopsy Sources

Cerebrospinal fluid
Tumors of the central nervous system

Saliva
Head and neck tumors

Pleural fluid
• Thoracic cancers
• Metastatic cancers

Peripheral blood

BLOOD
VESSEL
Tumor

Circulating
tumor cells

Exosomes
Apoptotic
tumor cell
cfDNA

Ascites
Metastatic cancers

Stool
Gastrointestinal tract cancers

Urine
• Urinary tract cancers
• cfDNA filtered from blood

Figure 1. Overview of Liquid-Biopsy Approaches.

Liquid­biopsy approaches involving peripheral blood include isolation of circulating tumor cells, which intravasate from tumors into the
bloodstream; exosomes, which are membrane­bound vesicles released by tumor cells; and cell­free DNA (cfDNA), which is released by
apoptotic or necrotic tumor cells. Although peripheral blood is the most common source of liquid biopsy, several other bodily fluids have
been used for specific liquid­biopsy applications, including isolation and analysis of cfDNA, as shown.

The cfDNA from normal cells is found in plas-
ma at low levels in healthy persons (approxi-
mately 10 to 15 ng per milliliter, on average),11
and the level can increase under conditions of
tissue stress, including exercise, inflammation,
surgery, or tissue injury.12
More than 40 years ago, it was observed that
patients with cancer have higher overall levels of
cfDNA than persons without cancer.11 In patients
with cancer, cfDNA that is released from tumor
cells is often referred to as circulating tumor
DNA (ctDNA) and constitutes only a portion of
the overall cfDNA. The fraction of ctDNA in
overall cfDNA in patients with cancer can vary
greatly, from less than 0.1% to more than
90%.12,13 Although the fraction of ctDNA tends
to parallel tumor burden within an individual
patient, substantial variability has been observed
among patients with the same cancer type, possi-
bly reflecting biologic differences or differences
in rates of cell death in the individual tumors.13
Moreover, patients with different tumor types

n engl j med 379;18 nejm.org November 1, 2018 1755

The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

show considerable variation in the frequency of nucleotide barcode), and error-suppression algo-
detectable ctDNA. Thus, the detection and analy- rithms have improved the limits of detection.23-25
sis of ctDNA amidst a background of normal
germline cfDNA presents a considerable challenge.
Isol at ion a nd A na lysis of cf DNA
Owing to the low levels and short half-life of
tumor-derived cfDNA, specialized approaches are
needed for both isolation and analysis of cfDNA.
Two key issues are the stability of the cfDNA it-
self and the potential for lysis of normal blood
cells, leading to contamination with normal DNA.
To limit these effects, when cfDNA is isolated
from blood collected in standard phlebotomy
tubes, plasma must typically be centrifuged and
separated within 1 to 4 hours after collection.
This need for rapid processing creates logistic
challenges and the potential for preanalytic vari-
ability caused by fluctuations of cfDNA concen-
tration and purity due to differences in process-
ing times.14,15 Alternatively, specialized cfDNA
collection tubes containing fixatives can stabilize
both cfDNA and intact cells for up to 7 to 14
days at room temperature, allowing for easy
shipping, storage, and batched or centralized
processing.16
The fraction of ctDNA within the background
of normal cfDNA in patients with cancer is typi-
cally small and highly variable from patient to
patient. Thus, ultrasensitive methods are required
to detect mutations, copy-number changes, or other
alterations that are present in cfDNA at very low
variant-allele frequencies (i.e., the percentage
of variant alleles present among all alleles, in-
cluding wild-type alleles) (Fig. 2). For detection
of individual point mutations, mutation-specific
techniques based on polymerase-chain-reaction
(PCR) analysis — such as BEAMing (beads,
emulsion, amplification, and magnetics) or drop-
let digital PCR (ddPCR) analysis — can identify
and quantify alterations present at allele frequen-
cies of 0.01% or less in cfDNA.17,18 Next-genera-
tion sequencing methods have also been tailored
for cfDNA, ranging from whole-genome or whole-
exome sequencing to targeted sequencing of a
limited gene panel.19-22 However, the sensitivity
and specificity of these approaches are limited by
the error rate of DNA polymerase and the se-
quencing reaction. Thus, modified approaches in-
corporating deep sequencing coverage, molecular
barcoding methods (in which individual input
template DNA fragments are tagged with a unique
Cl inic a l A ppl ic at ions
The ability to analyze tumor-derived DNA from
a routine blood draw without the need for an
invasive tumor biopsy represents a critical ad-
vance with potentially transformative clinical ap-
plications (Fig. 3). In particular, the minimally
invasive nature of cfDNA analysis provides a
means of molecular profiling for tumors that
are difficult or unsafe to biopsy and allows a
practical means for monitoring tumor DNA seri-
ally over time without the risk and potential
complications of standard tumor biopsy. In ad-
dition, cfDNA analysis may better capture the
molecular heterogeneity harbored by multiple dis-
tinct clonal populations in a patient’s tumor, as
compared with a needle biopsy of a single tumor
lesion. Finally, cfDNA analysis offers the poten-
tial for tumor detection or monitoring in patients
without clinically evident disease.
Diagnosis and Molecular Profiling
Tumor molecular profiling for the selection of
therapy has become a fundamental practice in
cancer medicine.26 The potential to assess the
molecular profile of a patient’s tumor from a
simple blood draw, without the need for an inva-
sive biopsy, makes cfDNA analysis an attractive
tool. However, a key initial question is whether
the mutational profile established through cfDNA
testing reliably reproduces the mutational pro-
file derived from a direct tumor biopsy, which re-
mains the standard of care. Early studies, based
on small numbers of patient samples, suggest-
ed low concordance between DNA alterations
detected in tumor and plasma samples from the
same patient.27,28 However, the validity of these
studies was compromised by the shortcoming
that tumor and plasma samples were often not
collected at the same time, yielding potential
differences due to molecular evolution of the
tumor. In addition, many plasma samples were
collected at suboptimal times, such as during
therapy, when ctDNA levels are at their lowest.
Nonconcordance of key alterations is most
often observed in patients with low ctDNA levels,
and low levels make detection of alterations more
challenging. For example, one recent study com-
pared the performance of two commercial cfDNA
sequencing assays on parallel pairs of blood

1756 n engl j med 379;18 nejm.org  November 1, 2018

The New England Journal of Medicine

Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Cell-free DNA Analysis and Cancer Treatment

Mutations Nucleotide coverage

10%
• Whole-genome sequencing
• Whole-exome sequencing
1%

Targeted next-generation sequencing

Copy-number alterations 0.1%

Next-generation sequencing with molecular
barcodes
0.01%
• BEAMing
• Droplet digital PCR analysis
Plasma 0.001%
Limit of detection

Next-generation sequencing with molecular barcodes

Gene fusions Wild type Mutant Wild type
Buffy coat
(source of
germline DNA) Molecular Each DNA molecule
barcode tags tagged with barcode

Redundant sequencing; true

mutations represented in all
reads with given barcode

DNA methylation

CH3
Normal cfDNA
ctDNA
CH3
True mutation Polymerase error

Figure 2. Isolation and Analysis of cfDNA.

As shown on the left, cfDNA is isolated from plasma after centrifugation of peripheral blood to separate the cellular component of the
blood, including white and red cells. Circulating tumor DNA (ctDNA) is found (often at low fractions) among a background of normal
germline cfDNA released by normal cells throughout the body. The cfDNA can be analyzed to identify several common DNA­based altera­
tions observed in tumors, including mutations, copy­number alterations, gene fusions, and DNA methylation changes. As shown at the
upper right, owing to the frequently low fraction of ctDNA in a background of normal cfDNA, specialized methods for cfDNA analysis
have been developed, including sequencing and methods based on polymerase­chain­reaction (PCR) analysis (BEAMing [beads, emulsion,
amplification, and magnetics] and droplet digital PCR). Typically, those analyses providing the greatest breadth or nucleotide coverage
have lesser sensitivity and require higher fractions of ctDNA in overall cfDNA (percentages in red) for analysis. Conversely, methods that
are directed against targeted mutation panels or single mutations offer improved limits of detection. As shown at the lower right, molecu­
lar barcoding to tag individual molecules of DNA has allowed next­generation sequencing approaches for cfDNA to achieve better limits
of detection, allowing discrimination of true mutations present in the original template DNA molecule (and thus present in all sequenc­
ing reads corresponding to a specific molecular barcode) from mutations introduced by polymerase error during the sequencing reaction.

samples drawn from 40 patients with metastatic
prostate cancer.29 Low concordance rates were
observed, raising questions about the reproduc-
ibility of cfDNA analysis. However, the study
had major limitations, including the facts that
at least half the patients had normal levels of
prostate-specific antigen and that participants
were not restricted to those not receiving therapy.
Thus, much of the variability may be attributable
to low cfDNA levels at or below the limit of de-
tection of these assays. In addition, the way in
which true positive or negative results are scored
in patients with low levels of cfDNA remains an
important challenge.
Larger and more carefully controlled studies
have shown high concordance rates of 80 to 90%
between plasma and tissue samples obtained
simultaneously, particularly for alterations in key

n engl j med 379;18 nejm.org November 1, 2018 1757

The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Localized Disease Metastatic Disease

Clinically Clinically Minimal Metastatic Treatment Disease Subsequent

undetectable detectable residual disease response progression therapy
disease

Surgical resection Systemic therapy 1 Systemic therapy 2

Resection
location

Clinical Identification
Residual-disease Molecular Response Monitoring of
Early detection of resistance
Application: detection profiling monitoring mechanisms clonal dynamics

Normal germline Baseline clonal alteration Subclonal resistance Subclonal resistance

cfDNA in ctDNA alteration A alteration B
Abundance of ctDNA

driver genes.20,30-32 Similarly, analyses of large
cfDNA sequencing databases have yielded mo-
lecular landscapes that closely match mutation
frequencies produced by large-scale tumor tissue
sequencing compendia, such as the Cancer Ge-
nome Atlas.33
Although these data suggest that cfDNA test-
ing could be a potential surrogate for or adjunct
to standard-of-care tumor biopsy testing, for now
tumor biopsy remains the standard for initial
pathological diagnosis and molecular testing.
Tumor biopsies allow for histologic interpreta-
tion and the assessment of non–DNA-based altera-
tions, such as in the expression of hormone re-
ceptors or other proteins, which can be important
for diagnosis and treatment decisions. Further-

1758 n engl j med 379;18 nejm.org November 1, 2018

The New England Journal of Medicine

Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Cell-free DNA Analysis and Cancer Treatment

Nonetheless, cfDNA testing can play an im-

Figure 3 (facing page). Clinical Applications of cfDNA
Analysis. portant role in initial molecular testing, par-
ticularly for patients in whom standard tumor
Analysis of cfDNA has potential applications at multi­
biopsies yield insufficient material for clinical
ple points throughout the natural course of cancer de­
sequencing, which can occur in as many as 20 to
velopment, diagnosis, and treatment. Early-detection
methods that screen for evidence of nascent tumors in
25% of needle biopsies.30,34 In such circumstances,
cfDNA are currently under development. A liquid-biopsy
cfDNA testing for treatment selection is increas-
test capable of identifying early-stage cancers in asymp­
ingly used as an alternative to repeat invasive
tomatic persons may allow cancers to be identified at
biopsy and may reveal actionable mutations that
a stage when they are more likely to be curable. After
guide treatment decisions in these patients.30,35
surgery with curative intent, cfDNA from postoperative
plasma drawn during the weeks after surgery can be
Technological advances are likely to allow more
analyzed for the persistent presence of mutations or
rapid and inexpensive testing of cfDNA for rela-
other alterations known to exist in the patient’s resected
tively rare abnormalities that could identify ac-
tumor. Since the half-life of cfDNA is very short (an hour
tionable mutations and predict the likelihood of
or less), any evidence of persistent tumor-derived muta­
response (e.g., the presence of microsatellite in-
tions in cfDNA from postoperative plasma can provide
direct evidence of residual disease that may ultimately
stability as a predictor of response to T-cell check-
lead to tumor relapse. Detection of residual disease in
point inhibition).
cfDNA shortly after surgery may allow patients to be
For repeat or serial testing after one or more
stratified according to risk of recurrence and may offer
lines of therapy, minimally invasive cfDNA analy-
an opportunity for early intervention to salvage cure. In
sis offers many apparent advantages over repeat
the context of metastatic disease, clinical sequencing
of cfDNA can identify potentially targetable genetic al­
invasive tumor biopsies, which are less practical,
terations to select precision therapies. Sequencing of
less safe, and less cost-effective. In particular,
cfDNA can identify many of the same target alterations
liquid biopsy may identify emergent genetic al-
identified by tumor sequencing, which can be particu­
terations driving acquired resistance to therapy
larly useful when insufficient tumor material is available
that can be targetable with newer-generation
for clinical sequencing. Studies have suggested that
ctDNA levels closely parallel overall tumor burden and
therapies.22,36-43 One of the earliest and best
can be used as an accurate means of monitoring treat­
examples is the use of cfDNA testing to identify
ment response and the development of resistance. On
the emergence of the epidermal growth factor
disease progression, cfDNA analysis has proved effec­
receptor (EGFR) T790M gatekeeper mutation
tive in identifying emergent genetic alterations that drive
after EGFR inhibitor therapy in EGFR-mutated
therapeutic resistance, which can guide subsequent
therapy choice. Owing to the potential for extensive tu­
non–small-cell lung cancer. Key studies have
mor heterogeneity in the context of acquired resistance,
shown a high degree of concordance of cfDNA
cfDNA analysis may identify multiple concurrent resis­
analysis with tissue testing in identifying the
tance alterations residing in distinct tumor metastases
presence of the T790M mutation, which responds
that would not be captured by a single tumor biopsy.
to third-generation EGFR inhibitors, such as
Analysis of cfDNA has been used to monitor the clonal
dynamics of multiple tumor subclones during subse­
osimertinib.35,44 Indeed, the cobas EGFR Muta-
quent therapy, an approach that can provide a molecu­
tion Test can identify T790M and other EGFR
lar basis for mixed clinical responses to therapy.
driver mutations in plasma and has been ap-
proved by the Food and Drug Administration as
more, large series have shown that approxi- a companion diagnostic for choosing specific
mately 15% of patients with metastatic can- EGFR therapies.45
cer may not have sufficient ctDNA levels to allow
for mutational profiling from plasma, and these Tracking of Therapeutic Response
numbers vary depending on tumor type and tu- Although ctDNA levels can vary greatly among
mor burden.13,31,33 Although improving the tim- patients, ctDNA levels in an individual patient
ing of blood draws for cfDNA analysis (e.g., over time correlate well with changes in tumor
before the initiation of therapy) may increase the burden and treatment response. The short half-
yield of cfDNA testing, confidence in the pres- life of cfDNA in circulation (approximately 1 hour)
ence or absence of key alterations decreases as can be advantageous for measuring real-time
the ctDNA fraction approaches the limit of de- tumor burden in response to therapy, in contrast
tection of current technologies. This factor must to many standard serum tumor markers in cur-
be taken into account when interpreting clinical rent clinical use (e.g., carcinoembryonic antigen
cfDNA tests. [CEA] and cancer antigen 125 [CA-125]), which

n engl j med 379;18 nejm.org  November 1, 2018 1759

The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l
have half-lives of days to weeks.38 Many standard
clinical tumor markers also have limited sensi-
tivity and specificity, whereas tumor-specific
clonal alterations (i.e., those alterations present in
the original tumor clone and thus present in all
tumor cells) can be monitored in plasma with
high sensitivity and are unique to a patient’s
tumor. Some studies have suggested that ctDNA
levels may actually increase transiently after the
initiation of therapy, as tumor-cell death leads to
increased release of ctDNA.6 However, within 1 to
2 weeks after the initiation of therapy, ctDNA
levels drop dramatically in patients who have a
response to therapy.
Indeed, some studies have suggested that
changes in ctDNA can outperform standard
tumor markers in the prediction of treatment
response.38,42 Moreover, a rise in ctDNA levels
may precede radiographic progression by weeks
to months.36-38,46 Thus, as technologies for cfDNA
monitoring become increasingly available and
cost-effective, their potential to detect early evi-
dence of response or progression may become
important for clinical management.
Monitoring of Resistance and Tumor
Heterogeneity
The clinical benefit of precision cancer therapies
is limited by the eventual emergence of acquired
resistance.26 In general, acquired resistance arises
from one or more tumor subclones that harbor
preexisting resistance alterations, which emerge
under the selective pressure of therapy. Analysis
of cfDNA has become an effective tool for the
early detection and identification of molecular
alterations leading to clinical resistance to ther-
apy. A key advantage of cfDNA is the ability to
capture molecular heterogeneity associated with
resistance (Fig. 4). Acquired resistance is often
characterized by the clonal outgrowth of mul-
tiple resistant subclones in an individual pa-
tient.36,39,40,43,47,48 These resistant subclones may
coexist in the same lesion or in distinct meta-
static sites. Thus, a single-lesion tumor biopsy
may dramatically underestimate the molecular
heterogeneity present, whereas the presence of
multiple resistance mechanisms can often be
captured by cfDNA, which is shed from tumor
cells throughout the body.
Indeed, studies incorporating multiple tumor
biopsies or autopsy specimens have shown that
multiple unique resistance alterations frequently
coexist in different metastatic sites in an indi-
1760 n engl j med 379;18 nejm.org  November 1, 2018
The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
of m e dic i n e
vidual patient but can be detected collectively in
cfDNA from a single plasma sample.39,40,49 A
study involving patients with colorectal cancer
after anti-EGFR antibody therapy showed that
patients may harbor as many as 13 different re-
sistance alterations, as detected in cfDNA, with
fewer than 10% of patients having only a single
resistance alteration.33 Studies comparing matched
tumor biopsies and cfDNA tests at the time of
disease progression have suggested that cfDNA
testing may reveal additional alterations not
identified by a single tumor biopsy in as many
as two thirds of cases.50 Thus, targeting a sin-
gle resistance mechanism on the basis of the
results of a single tumor biopsy may lead to
mixed clinical responses as a result of the out-
growth of distinct resistant subclones not ob-
tained in the biopsy specimen,39 and cfDNA test-
ing may help guide treatment decisions.
In addition to emerging as an important dis-
covery tool, cfDNA analysis can be applied clini-
cally for the management of therapeutic resis-
tance. The use of cfDNA to identify the presence
of the T790M resistance mutation in EGFR-mutated
lung cancer was discussed above. Other studies
have shown that cfDNA analysis can identify the
coexistence of T790M with other resistance altera-
tions, such as MET amplification, and that pa-
tients with such coexisting alterations may de-
rive less benefit from subsequent therapy with
third-generation EGFR inhibitors, such as osimer­
tinib.44 Similarly, detection of one or more ESR1
mutations in the cfDNA of patients with estrogen-
receptor–positive breast cancer predicted poorer
outcome of subsequent hormonal therapy.51 Other
studies show that resistance to various targeted
therapies can be observed in cfDNA before the
clinical detection of progression by standard
imaging or tumor markers.36,37,40
Analysis of cfDNA has also been used to
track the clonal dynamics of distinct resistant
subclones during sequential therapy and even
after the discontinuation of therapy. Siravegna
et al. found that the RAS mutations emerging in
patients with colorectal cancer during EGFR
blockade can fall to undetectable levels in cfDNA
after withdrawal of EGFR-directed therapy, and
many of these patients may then benefit from
reinstitution of anti-EGFR therapy.46 Integration
of real-time cfDNA analysis into clinical trials
and eventually into standard clinical manage-
ment has the potential to become a valuable tool
for precision medicine.
 Cell-free DNA Analysis and Cancer Treatment

First-line therapy Second-line therapy

Tumor
biopsy

Clonal target Resistance alteration A Resistance alteration C

alteration Resistance alteration B Resistance alteration D
Clonal abundance
Abundance of ctDNA

Figure 4. Tumor Heterogeneity and cfDNA.

Acquired resistance to therapy is often thought to arise from rare tumor subclones that harbor preexisting resistance
alterations. Although all cells may harbor the original clonal target alteration, preexisting subclonal alterations that
provide a fitness advantage under the selective pressure of therapy may exist in some cells. As therapy is initiated,
it may exert a cytotoxic effect on most tumor cells, but an outgrowth of resistant subpopulations may occur, leading
to dynamic shifts in clonal abundance and eventual disease progression. Because preexisting resistant subclones
may reside in different metastatic lesions or in different subpopulations within a single lesion, the emergence of re­
sistance can be characterized by extensive molecular heterogeneity. Thus, a single tumor biopsy specimen that is
obtained during disease progression may reveal only a subset (i.e., green only) of the resistant clones present, and
subsequent therapies that are directed against this resistance mechanism only may lead to mixed clinical response
and treatment failure due to outgrowth of other coexistent clones. Analysis of cfDNA has the potential to identify
multiple concurrent mechanisms of resistance and can monitor clonal dynamics during therapy. Clonal alterations
(present in the original tumor clone and thus in all cells) are represented in gray, and resistant subclonal alterations
in other colors.

Detection of Postsurgical Residual Disease
One of the most transformative potential appli-
cations of cfDNA analysis may be its ability to
detect the presence of tumor in patients with no
clinically evident disease — for example, as a
screening tool in the early detection of new can-
cers or for detection of relapse after surgery or
adjuvant therapy. In general, the primary means
by which solid tumors can be cured is surgical
resection. If any tumor cells remain postsurgi-
cally, they can lead to eventual tumor relapse. In
high-risk patients, adjuvant chemotherapy can
reduce the risk of relapse. However, it is cur-
rently not possible to determine which patients
harbor residual disease immediately after sur-
gery and which patients have been cured of their
disease. In particular contexts, such as patients
with stage II colorectal cancer who have low

n engl j med 379;18 nejm.org November 1, 2018 1761

The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l
clinical risk and no evidence of nodal or distant
metastases, adjuvant therapy is not routinely
given, although approximately 15% of these pa-
tients may eventually have a recurrence. An ef-
fective method for the detection of postoperative
residual disease could spare patients who have
been cured the need to undergo potentially toxic
adjuvant chemotherapy or could identify those
patients who have residual disease and might
benefit from adjuvant therapy (Fig. 5).
In a landmark study, Diehl et al. found that
cfDNA analysis just a few weeks after surgery for
colorectal cancer could accurately predict those
patients who had residual disease and would
eventually relapse.12 First, tumor-specific muta-
tions were identified in each patient by standard
sequencing of resected tumor. Next, highly sen-
sitive methods (BEAMing), capable of detecting
mutant-allele frequencies as low as 0.01%, were
used to determine whether one or more of these
tumor-specific mutations persisted in that patient’s
plasma when collected approximately 4 weeks
after surgery. Detection of even trace levels of
these mutations would provide direct evidence
of residual tumor cells. In a follow-up study in-
volving 178 patients with stage II colon cancer
who did not receive adjuvant chemotherapy, the
detection of residual tumor-specific mutations
in postoperative plasma was associated with a
risk of tumor recurrence that was 18 times as
high as that among patients with undetectable
ctDNA (P<0.001). In addition, ctDNA markedly
outperformed clinical risk factors and a standard
tumor marker (CEA) for predicting relapse.52
Similarly, several key studies have shown that
the postoperative detection of tumor-specific
mutations in cfDNA can predict residual disease
and tumor relapse in breast cancer, lung cancer,
and pancreatic cancer.53-56 This approach has the
potential to become a critical tool in the postop-
erative management of the care of patients with
cancer and is currently being tested in prospec-
tive clinical trials that will assess the usefulness
of residual postoperative ctDNA detection to guide
adjuvant chemotherapy (Fig. 5).
Early Cancer Detection
The holy grail of liquid-biopsy applications is the
potential for early cancer detection through a
simple blood test in otherwise healthy, asymptom-
atic persons. At present, no mature technology
exists to achieve that goal, but such an approach
1762 n engl j med 379;18 nejm.org  November 1, 2018
The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
of m e dic i n e
Figure 5 (facing page). Use of cfDNA for the Detection
of Residual Disease and Management of Postoperative
Therapy.
After surgery with curative intent, there is currently no
way to determine which patients are cured of their dis­
ease and which have subclinical residual disease that
may ultimately lead to recurrence. Thus, decisions about
adjuvant chemotherapy are based on clinical-risk strati­
fication. In the example of colon cancer, patients with
stage II disease who are at low clinical risk undergo
­observation alone and do not receive adjuvant chemo­
therapy, although approximately 10 to 15% will eventu­
ally have recurrent disease. Conversely, patients with
stage III disease are treated with adjuvant chemother­
apy, even though more than half are cured by surgery
alone. Current efforts are focused on using postoper­
ative cfDNA to identify patients with residual disease,
which may allow more precise decisions about adju­
vant treatment. Briefly, tumor-specific mutations or
other alterations are identified in the resected tumor
specimen of each patient, and plasma specimens ob­
tained a few weeks after surgery are assessed for the
persistent presence of these alterations as direct evi­
dence of residual tumor. With further improvement of
this approach, it may be possible in future studies to
assign patients who would not have received adjuvant
therapy as the standard of care (i.e., with stage II dis­
ease) to receive adjuvant therapy if evidence of residual
disease is detected in cfDNA. Similarly, as the sensitiv­
ity and reliability of these tests improve, it may even be
possible to assign patients who would receive adjuvant
therapy as the standard of care (i.e., those with stage
III disease) to a low-risk group on the basis of cfDNA
testing; active surveillance could be considered in lieu
of adjuvant therapy, thus sparing many patients who are
cured by surgery alone the toxic effects of chemother­
apy. Studies are currently ongoing (e.g., the DYNAMIC
trial [Australian New Zealand Clinical Trials Registry
number, ACTRN12615000381583]).
would require a highly sensitive method — to
detect trace amounts of cfDNA or other material
released by precancerous lesions or early-stage
cancers — and would also require high specific-
ity to minimize false positive results in the large
unaffected population undergoing screening.
Additional challenges complicate this undertak-
ing. First, since many cancer types share com-
mon mutations in genes such as KRAS, BRAF, or
TP53, localizing a cancer to a specific organ after
a positive liquid-biopsy test may be difficult.57 In
addition, benign lesions may harbor the same
mutations seen in cancer and have the potential
to shed cfDNA into the circulation. Indeed,
benign nevi may harbor the same BRAF muta-
tions observed in advanced cancers.58 Mutations
 Cell-free DNA Analysis and Cancer Treatment

Current standard of care Potential cfDNA-guided postoperative management

Stage II disease and low clinical risk

No adjuvant
chemotherapy

Stage II disease and low clinical risk Active

(~10–15% with residual disease) surveillance
No adjuvant
chemotherapy Serial cfDNA
for all monitoring

~10–15% have
recurrence and Normal germline cfDNA
may have
Tumor-specific mutation in ctDNA
benefited from
chemotherapy No evidence of residual disease in post-
operative cfDNA
Evidence of residual disease in post-
Cure by surgery alone operative cfDNA

Residual disease after surgery

Adjuvant
chemotherapy

Sequence resected
tumor, identify
clonal mutations
Detection of tumor-specific
mutations in postoperative plasma

Stage III disease No adjuvant

Adjuvant chemotherapy
chemotherapy
for all Stage III disease
Active
(~45% with residual disease)
>50% cured surveillance
by surgery
alone; Serial cfDNA
chemotherapy monitoring
unnecessary

~30% have
recurrence
despite
chemotherapy
and ~15%
cured by
chemotherapy
Adjuvant
chemotherapy

n engl j med 379;18 nejm.org November 1, 2018 1763

The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

detectable in cfDNA can also originate from
aberrant and often benign clonal populations in
the bone marrow, through a process referred to
as clonal hematopoiesis of indeterminate poten-
tial (CHIP). The frequency of CHIP increases
exponentially with age, with a rate of more than
10% among persons older than 70 years of age.59
CHIP is thus a common source of diverse muta-
tions detectable in cfDNA, posing a major chal-
lenge for this approach. For this reason, methods
other than simple mutation detection in cfDNA
— including tumor-associated viral sequences,
such as in tumors associated with Epstein–Barr
virus (EBV) infection and human papillomavirus
infection, and DNA methylation changes — are
being explored for early detection.7,60,61
Two recent studies have illustrated the poten-
tial of liquid biopsy for the early detection of
cancer. Chan et al. used detection of EBV DNA in
plasma to screen for nasopharyngeal carcinoma
in 20,174 Chinese patients.60 A total of 309 pa-
tients (1.5%) had detectable EBV DNA in plasma,
as confirmed on two sequential tests, and 34 of
these patients (0.17% of the original population)
had nasopharyngeal carcinoma on endoscopic
evaluation. Conversely, only 1 of the EBV DNA–
negative patients presented with nasopharyngeal
carcinoma within a year after screening. Overall,
the sensitivity and specificity of this approach
An audio were 97.1% and 98.6%, respectively. Recently,
interview with
Dr. Corcoran
Cohen et al. developed a screening method inte-
is available grating mutation detection in cfDNA with circu-
at NEJM.org lating protein markers, termed CancerSEEK.57 In
1005 patients with nonmetastatic, clinically de-
tectable tumors across eight common tumor
types, the test was positive in a median of 70%
of patients, with sensitivities ranging from 69 to
98%. Overall specificity was more than 99%,
with only 7 of 812 healthy controls testing posi-
tive. Moreover, using supervised machine learn-
ing, the investigators used the profile of detected
mutations and proteins to localize the cancer to
its tissue of origin in 83% of cases.
Though these results are encouraging, evalu-
ation in a more representative screening popula-
tion of asymptomatic patients will be critical.
Although further improvement of liquid-biopsy
screening approaches and validation in prospec-
tive clinical trials are needed, cfDNA analysis as
an early-detection tool offers a potentially trans-
formative advance for cancer medicine.
Sum m a r y a nd F u t ur e Dir ec t ions
Analysis of cfDNA has rapidly emerged as a tech-
nology with many promising clinical applications
in oncology. Effective clinical integration of cfDNA
analysis will require a careful understanding of
the advantages and limitations of this approach
for proper interpretation of results to guide clini-
cal decision making. Although further prospec-
tive study is needed, cfDNA analysis harbors the
potential to have a transformative effect on can-
cer medicine.
Disclosure forms provided by the authors are available with
the full text of this article at NEJM.org.

References
1. Haber DA, Velculescu VE. Blood-based
analyses of cancer: circulating tumor
cells and circulating tumor DNA. Cancer
Discov 2014;​4:​650-61.
2. Siravegna G, Marsoni S, Siena S,
Bardelli A. Integrating liquid biopsies into
the management of cancer. Nat Rev Clin
Oncol 2017;​14:​531-48.
3. Alix-Panabières C, Pantel K. Circulat-
ing tumor cells: liquid biopsy of cancer.
Clin Chem 2013;​59:​110-8.
4. Ruivo CF, Adem B, Silva M, Melo SA.
The biology of cancer exosomes: insights
and new perspectives. Cancer Res 2017;​
77:​6480-8.
5. Schwarzenbach H, Nishida N, Calin
GA, Pantel K. Clinical relevance of circu-
lating cell-free microRNAs in cancer. Nat
Rev Clin Oncol 2014;​11:​145-56.
6. Husain H, Melnikova VO, Kosco K,
et al. Monitoring daily dynamics of early
tumor response to targeted therapy by de-
tecting circulating tumor DNA in urine.
Clin Cancer Res 2017;​23:​4716-23.
7. Wang Y, Springer S, Mulvey CL, et al.
Detection of somatic mutations and HPV
in the saliva and plasma of patients with
head and neck squamous cell carcinomas.
Sci Transl Med 2015;​7:​293ra104.
8. De Mattos-Arruda L, Mayor R, Ng CK,
et al. Cerebrospinal fluid-derived circulat-
ing tumour DNA better represents the
genomic alterations of brain tumours than
plasma. Nat Commun 2015;​6:​8839.
9. Mandel P, Metais P. Les acides nu­
cléiques du plasma sanguin chez l’homme.
C R Seances Soc Biol Fil 1948;​142:​241-3.
10. Fan HC, Blumenfeld YJ, Chitkara U,
Hudgins L, Quake SR. Analysis of the size
distributions of fetal and maternal cell-
free DNA by paired-end sequencing. Clin
Chem 2010;​56:​1279-86.
11. Leon SA, Shapiro B, Sklaroff DM,
Yaros MJ. Free DNA in the serum of can-
cer patients and the effect of therapy.
Cancer Res 1977;​37:​646-50.
12. Diehl F, Schmidt K, Choti MA, et al.
Circulating mutant DNA to assess tu-
mor dynamics. Nat Med 2008;​ 14:​
985-
90.
13. Bettegowda C, Sausen M, Leary RJ,
et al. Detection of circulating tumor DNA
in early- and late-stage human malignan-
cies. Sci Transl Med 2014;​6:​224ra24.
14. El Messaoudi S, Rolet F, Mouliere F,
Thierry AR. Circulating cell free DNA:
preanalytical considerations. Clin Chim
Acta 2013;​424:​222-30.
15. Swinkels DW, Wiegerinck E, Steegers
EA, de Kok JB. Effects of blood-process-
ing protocols on cell-free DNA quantifica-
tion in plasma. Clin Chem 2003;​49:​525-6.
16. Alidousty C, Brandes D, Heydt C, et al.
Comparison of blood collection tubes from
three different manufacturers for the col-
lection of cell-free DNA for liquid biopsy
mutation testing. J Mol Diagn 2017;​19:​
801-4.
17. Diehl F, Li M, He Y, Kinzler KW, Vogel-
stein B, Dressman D. BEAMing: single-
molecule PCR on microparticles in water-in-
oil emulsions. Nat Methods 2006;​3:​551-9.

1764 n engl j med 379;18 nejm.org  November 1, 2018

The New England Journal of Medicine

Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Cell-free DNA Analysis and Cancer Treatment
18. Hindson BJ, Ness KD, Masquelier DA,
et al. High-throughput droplet digital PCR
system for absolute quantitation of DNA
copy number. Anal Chem 2011;​83:​8604-10.
19. Forshew T, Murtaza M, Parkinson C,
et al. Noninvasive identification and moni-
toring of cancer mutations by targeted deep
sequencing of plasma DNA. Sci Transl Med
2012;​4:​136ra68.
20. Adalsteinsson VA, Ha G, Freeman SS,
et al. Scalable whole-exome sequencing of
cell-free DNA reveals high concordance
with metastatic tumors. Nat Commun
2017;​8:​1324.
21. Leary RJ, Kinde I, Diehl F, et al. Devel-
opment of personalized tumor biomarkers
using massively parallel sequencing. Sci
Transl Med 2010;​2:​20ra14.
22. Murtaza M, Dawson SJ, Tsui DW, et al.
Non-invasive analysis of acquired resis-
tance to cancer therapy by sequencing of
plasma DNA. Nature 2013;​497:​108-12.
23. Kinde I, Wu J, Papadopoulos N, Kinzler
KW, Vogelstein B. Detection and quantifi-
cation of rare mutations with massively
parallel sequencing. Proc Natl Acad Sci
U S A 2011;​108:​9530-5.
24. Newman AM, Bratman SV, To J, et al.
An ultrasensitive method for quantitating
circulating tumor DNA with broad patient
coverage. Nat Med 2014;​20:​548-54.
25. Newman AM, Lovejoy AF, Klass DM,
et al. Integrated digital error suppression
for improved detection of circulating tu-
mor DNA. Nat Biotechnol 2016;​34:​547-55.
26. Garraway LA, Jänne PA. Circumvent-
ing cancer drug resistance in the era of
personalized medicine. Cancer Discov
2012;​2:​214-26.
27. Kuderer NM, Burton KA, Blau S, et al.
Comparison of 2 commercially available
next-generation sequencing platforms in
oncology. JAMA Oncol 2017;​3:​996-8.
28. Chae YK, Davis AA, Carneiro BA, et al.
Concordance between genomic alterations
assessed by next-generation sequencing in
tumor tissue or circulating cell-free DNA.
Oncotarget 2016;​7:​65364-73.
29. Torga G, Pienta KJ. Patient-paired sam-
ple congruence between 2 commercial
liquid biopsy tests. JAMA Oncol 2017 De-
cember 14 (Epub ahead of print).
30. Zill OA, Greene C, Sebisanovic D,
et al. Cell-free DNA next-generation se-
quencing in pancreatobiliary carcinomas.
Cancer Discov 2015;​5:​1040-8.
31. Schrock AB, Pavlick DC, Klempner SJ,
et al. Hybrid capture-based genomic pro-
filing of circulating tumor DNA from pa-
tients with advanced cancers of the gastro-
intestinal tract or anus. Clin Cancer Res
2018;​24:​1881-90.
32. Lanman RB, Mortimer SA, Zill OA,
et al. Analytical and clinical validation of
a digital sequencing panel for quantita-
tive, highly accurate evaluation of cell-free
circulating tumor DNA. PLoS One 2015;​
10(10):​e0140712.
33. Strickler JH, Loree JM, Ahronian LG,
et al. Genomic landscape of cell-free DNA
n engl j med 379;18 nejm.org  November 1, 2018
The New England Journal of Medicine
Downloaded from nejm.org at GOTEBORGS UNIVERSITETSBIBL. on November 29, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
in patients with colorectal cancer. Cancer
Discov 2018;​8:​164-73.
34. Meric-Bernstam F, Brusco L, Shaw K,
et al. Feasibility of large-scale genomic
testing to facilitate enrollment onto ge-
nomically matched clinical trials. J Clin
Oncol 2015;​33:​2753-62.
35. Oxnard GR, Thress KS, Alden RS, et al.
Association between plasma genotyping
and outcomes of treatment with osimer-
tinib (AZD9291) in advanced non-small-
cell lung cancer. J Clin Oncol 2016;​ 34:​
3375-82.
36. Diaz LA Jr, Williams RT, Wu J, et al.
The molecular evolution of acquired resis-
tance to targeted EGFR blockade in colo­
rectal cancers. Nature 2012;​486:​537-40.
37. Misale S, Yaeger R, Hobor S, et al.
Emergence of KRAS mutations and ac-
quired resistance to anti-EGFR therapy in
colorectal cancer. Nature 2012;​486:​532-6.
38. Dawson SJ, Tsui DW, Murtaza M,
et al. Analysis of circulating tumor DNA
to monitor metastatic breast cancer. N Engl
J Med 2013;​368:​1199-209.
39. Russo M, Siravegna G, Blaszkowsky
LS, et al. Tumor heterogeneity and lesion-
specific response to targeted therapy in
colorectal cancer. Cancer Discov 2016;​6:​
147-53.
40. Goyal L, Saha SK, Liu LY, et al. Poly-
clonal secondary FGFR2 mutations drive
acquired resistance to FGFR inhibition in
patients with FGFR2 fusion-positive chol-
angiocarcinoma. Cancer Discov 2017;​ 7:​
252-63.
41. Schiavon G, Hrebien S, Garcia-Muril-
las I, et al. Analysis of ESR1 mutation in
circulating tumor DNA demonstrates evo-
lution during therapy for metastatic breast
cancer. Sci Transl Med 2015;​7:​313ra182.
42. Corcoran RB, André T, Atreya CE,
et al. Combined BRAF, EGFR, and MEK in-
hibition in patients with BRAFV600E-mutant
colorectal cancer. Cancer Discov 2018;​8:​
428-43.
43. Hazar-Rethinam M, Kleyman M, Han
GC, et al. Convergent therapeutic strate-
gies to overcome the heterogeneity of ac-
quired resistance in BRAFV600E colorec-
tal cancer. Cancer Discov 2018;​8:​417-27.
44. Chabon JJ, Simmons AD, Lovejoy AF,
et al. Circulating tumour DNA profiling
reveals heterogeneity of EGFR inhibitor
resistance mechanisms in lung cancer pa-
tients. Nat Commun 2016;​7:​11815.
45. Malapelle U, Sirera R, Jantus-Lewintre
E, et al. Profile of the Roche cobas EGFR
mutation test v2 for non-small cell lung
cancer. Expert Rev Mol Diagn 2017;​ 17:​
209-15.
46. Siravegna G, Mussolin B, Buscarino M,
et al. Clonal evolution and resistance to
EGFR blockade in the blood of colorectal
cancer patients. Nat Med 2015;​21:​795-801.
47. Gerlinger M, Rowan AJ, Horswell S,
et al. Intratumor heterogeneity and
branched evolution revealed by multi­
region sequencing. N Engl J Med 2012;​
366:​883-92.
48. Piotrowska Z, Niederst MJ, Karlovich
CA, et al. Heterogeneity underlies the
emergence of EGFRT790 wild-type clones
following treatment of T790M-positive
cancers with a third-generation EGFR in-
hibitor. Cancer Discov 2015;​5:​713-22.
49. Blakely CM, Watkins TBK, Wu W, et al.
Evolution and clinical impact of co-occur-
ring genetic alterations in advanced-stage
EGFR-mutant lung cancers. Nat Genet
2017;​49:​1693-704.
50. Parikh A, Goyal L, Hazar-Rethinam M,
et al. Systematic liquid biopsy identifies
novel and heterogeneous mechanisms of
acquired resistance in gastrointestinal (GI)
cancer patients. Ann Oncol 2017;​28:​Suppl 3:​
mdx262. abstract.
51. Chandarlapaty S, Chen D, He W, et al.
Prevalence of ESR1 mutations in cell-free
DNA and outcomes in metastatic breast
cancer: a secondary analysis of the
­BOLERO-2 clinical trial. JAMA Oncol 2016;​
2:​1310-5.
52. Tie J, Wang Y, Tomasetti C, et al. Cir-
culating tumor DNA analysis detects
minimal residual disease and predicts re-
currence in patients with stage II colon
cancer. Sci Transl Med 2016;​8:​346ra92.
53. Garcia-Murillas I, Schiavon G, Weigelt
B, et al. Mutation tracking in circulating
tumor DNA predicts relapse in early breast
cancer. Sci Transl Med 2015;​7:​302ra133.
54. Chaudhuri AA, Chabon JJ, Lovejoy AF,
et al. Early detection of molecular residual
disease in localized lung cancer by circu-
lating tumor DNA profiling. Cancer Dis-
cov 2017;​7:​1394-403.
55. Sausen M, Phallen J, Adleff V, et al.
Clinical implications of genomic altera-
tions in the tumour and circulation of
pancreatic cancer patients. Nat Commun
2015;​6:​7686.
56. Beaver JA, Jelovac D, Balukrishna S,
et al. Detection of cancer DNA in plasma
of patients with early-stage breast cancer.
Clin Cancer Res 2014;​20:​2643-50.
57. Cohen JD, Li L, Wang Y, et al. Detec-
tion and localization of surgically resect-
able cancers with a multi-analyte blood
test. Science 2018;​359:​926-30.
58. Shain AH, Yeh I, Kovalyshyn I, et al.
The genetic evolution of melanoma from
precursor lesions. N Engl J Med 2015;​373:​
1926-36.
59. Jaiswal S, Fontanillas P, Flannick J,
et al. Age-related clonal hematopoiesis
associated with adverse outcomes. N Engl
J Med 2014;​371:​2488-98.
60. Chan KCA, Woo JKS, King A, et al.
Analysis of plasma Epstein–Barr virus
DNA to screen for nasopharyngeal cancer.
N Engl J Med 2017;​377:​513-22.
61. Chan KC, Jiang P, Chan CW, et al.
Noninvasive detection of cancer-associated
genome-wide hypomethylation and copy
number aberrations by plasma DNA bisul-
fite sequencing. Proc Natl Acad Sci U S A
2013;​110:​18761-8.
Copyright © 2018 Massachusetts Medical Society.
1765