Precision Medicine and Imaging Clinical

Cancer
Research
The Landscape of Actionable Genomic Alterations
in Cell-Free Circulating Tumor DNA from 21,807
Advanced Cancer Patients
Oliver A. Zill1, Kimberly C. Banks1, Stephen R. Fairclough1, Stefanie A. Mortimer1,
James V. Vowles1, Reza Mokhtari1, David R. Gandara2, Philip C. Mack2, Justin I. Odegaard1,
Rebecca J. Nagy1, Arthur M. Baca1, Helmy Eltoukhy1, Darya I. Chudova1, Richard B. Lanman1,
and AmirAli Talasaz1

Abstract

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

Purpose: Cell-free DNA (cfDNA) sequencing provides a in alteration prevalence being due to patient treatment.
noninvasive method for obtaining actionable genomic infor- This highly sensitive cfDNA sequencing assay revealed
mation to guide personalized cancer treatment, but the pres- numerous subclonal tumor-derived alterations, expected as
ence of multiple alterations in circulation related to treatment a result of clonal evolution, but leading to an apparent
and tumor heterogeneity complicate the interpretation of the departure from mutual exclusivity in treatment-na€ve
observed variants. tumors. Upon applying novel cfDNA clonality and copy-
Experimental Design: We describe the somatic mutation number driver identification methods, robust mutual exclu-
landscape of 70 cancer genes from cfDNA deep-sequencing sivity was observed among predicted truncal driver cfDNA
analysis of 21,807 patients with treated, late-stage cancers alterations (FDR ¼ 5  107 for EGFR and ERBB2), in effect
across >50 cancer types. To facilitate interpretation of the distinguishing tumor-initiating alterations from secondary
genomic complexity of circulating tumor DNA in advanced, alterations. Treatment-associated resistance, including both
treated cancer patients, we developed methods to identify novel alterations and parallel evolution, was common in the
cfDNA copy-number driver alterations and cfDNA clonality. cfDNA cohort and was enriched in patients with targetable
Results: Patterns and prevalence of cfDNA alterations driver alterations (>18.6% patients).
in major driver genes for non–small cell lung, breast, and Conclusions: Together, these retrospective analyses of a
colorectal cancer largely recapitulated those from tumor large cfDNA sequencing data set reveal subclonal structures
tissue sequencing compendia (The Cancer Genome Atlas and emerging resistance in advanced solid tumors. Clin
and COSMIC; r ¼ 0.90–0.99), with the principal differences Cancer Res; 24(15); 3528–38. 2018 AACR.

Introduction obtained in the clinical setting may open avenues for learning
how to navigate these complexities.
Genomic analysis of cell-free DNA (cfDNA) from advanced
As a recently developed testing method, clinical cfDNA
cancer patients allows the identification of actionable alterations
sequencing has repeatedly been benchmarked against tissue
shed into the circulation and may provide a global summary of
sequencing, but these performance comparisons are confounded
tumor heterogeneity without an invasive biopsy (1). Plasma
by temporal and spatial heterogeneity in tumors (2–6). In addi-
cfDNA analysis can provide insights from genomic information
tion, circulating tumor DNA (ctDNA) may be undetectable when
shed from multiple lesions within a patient, but this broader level
shedding of tumor DNA is nominal, such as when therapy
of insight can introduce added complexity. Indeed, most clinical
stabilizes tumor growth (7, 8). Recent efforts to globally charac-
cfDNA sequencing is performed on patients with advanced or
terize tumor heterogeneity using both plasma cfDNA analysis and
metastatic disease, often at the second line of therapy or later.
multiregion tumor sequencing highlight the complementary
Retrospective analyses of large-scale cfDNA sequencing data
nature of the two approaches (9–11). However, there is a paucity
of cfDNA data sets large enough to evaluate the similarity of
1
tumor-initiating alterations ("truncal drivers") in solid tumor
Guardant Health, Inc., Redwood City, California. 2University of California Davis
Comprehensive Cancer Center, Sacramento, California.
cancers to those found in the cfDNA of advanced cancer patients.
Among the various methods available for cfDNA analysis, tar-
Note: Supplementary data for this article are available at Clinical Cancer
geted panel deep-sequencing assays that utilize extensive error-
Research Online (http://clincancerres.aacrjournals.org/).
correction methods provide the depth (sensitivity) and genomic
Present address for O.A. Zill: Department of Bioinformatics and Computational breadth necessary to optimally survey tumor-derived genomic
Biology, Genentech, South San Francisco, California.
alterations in plasma cfDNA, even at low allelic fractions (12–15).
Corresponding Author: Stephen R. Fairclough, Guardant Health, 505 Penobscot In order to elucidate the landscape of truncal driver mutations
Drive, Redwood City, CA 94063. Phone: 650-814-2311; E-mail: in cfDNA, we first evaluated the extent of detectable tumor
[email protected]
heterogeneity in a large cohort of patients subject to deep sequenc-
doi: 10.1158/1078-0432.CCR-17-3837 ing of cfDNA to assess how ctDNA levels across patients might
2018 American Association for Cancer Research. impact cfDNA variant detection. To distinguish truncal driver

3528 Clin Cancer Res; 24(15) August 1, 2018

 Somatic Genomic Landscape of Circulating Tumor DNA

2014 to September 2016 (data freeze at September 22, 2016).

Translational Relevance Disease stage was confirmed for each patient by the provider to be
This study describes genomic alterations from the largest advanced disease (stage III/IV). Treatment histories and survival
cell-free circulating tumor DNA cohort to date, as derived from information were generally not available. Over 50 solid tumor
regular clinical practice. The high prevalence of resistance types were represented [the most common were non–small cell
alterations found in advanced, treated cancer patients neces- lung cancer (NSCLC, 37%), breast cancer (16%), colorectal cancer
sitated accurate methods for determining mutation clonality (9%)], although some cancer types were represented by only a
and driver/resistance status from plasma. We provide such small number of samples. See Supplementary Tables S3 and S4 for
methods, thereby extending the utility of cell-free DNA comparisons of the cfDNA cohort characteristics with those of the
sequencing analysis. Our finding of an association between pan-cancer TCGA (The Cancer Genome Atlas) cohort. The clinical
estimated circulating tumor DNA (ctDNA) levels and tumor cfDNA sequencing platform (Guardant360) used in this study has
mutational burden ascertained from plasma suggests that been previously described (13), and additional information on
ctDNA level is likely an important variable to consider for the assay and variant calling is provided in the Supplementary
immunotherapy applications of ctDNA analysis. Although Methods and in the accompanying paper by Odegaard and
cell-free DNA can provide a summary of tumor heterogeneity colleagues (16). For analysis purposes, variants were categorized
across multiple metastatic sites in a patient, our findings of as clinically actionable based on whether their presence would be
high variability in ctDNA levels across patients, and its impact reasonably expected to inform standard-of-care treatment deci-

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

on variant detection, highlight the need for an improved sions as judged by the clinical oncologists and/or molecular
understanding of factors influencing ctDNA levels and safe pathologist participating in the study. Descriptions of the meth-
methods for maximizing them at the time of ctDNA testing. ods for cfDNA clonality estimation, cfDNA copy number driver
analysis, comparisons of cfDNA to TCGA alterations, resistance
alterations and longitudinal analysis, and statistical analysis can
be found in the Supplemental Methods. The reported cfDNA
mutations from secondary resistance mutations, we developed
alterations data and associated analysis code have been deposited
methods to infer clonality and driver status of tumor mutations
in an open-access GitHub repository and are available at https://
from cfDNA. We then examined the similarity of cfDNA patterns
github.com/guardant/Zill_2018.
of common driver alterations in a large cohort of advanced,
previously treated solid tumor patients to those found in treat-
ment-na€ve tumor tissue compendia. Finally, we explored the Results
landscape of resistance variants to targeted therapies in the large Somatic genomic alterations in cfDNA across 21,807 patients
cfDNA cohort. Somatic cfDNA alterations were detected in 85% (18,503/
21,807) of patients across all cancer types, ranging from 51% for
Materials and Methods glioblastoma to 93% for small cell lung cancer (Fig. 1A). Half of
Characteristics of the cfDNA cohort the reported somatic cfDNA alterations had VAF <0.41% (range,
A summary of the cfDNA cohort used in this study is provided 0.03%–97.6%; Fig. 1B). Alteration-positive samples had on aver-
in Table 1. The cohort was assembled from 21,807 consecutive age three or four alterations detected (median ¼ 3; mean ¼ 4.3;
cancer patients (25,578 total samples) tested on the Guardant range, 1–166), including copy-number amplifications (CNA;
Health cfDNA sequencing platform as part of clinical care (in a Supplementary Fig. S1B). For subsequent analyses, we assumed
Clinical Laboratory Improvement Amendments (CLIA)-certified, that the fraction of tumor-derived cfDNA molecules within the
College of American Pathologists (CAP)-accredited, New York total population of circulating cfDNA molecules ("ctDNA level")
State Department of Health-approved clinical laboratory at Guar- in a sample was proportional to the copy-number–adjusted
dant Health, Inc.). As such, this was an observational, noninter- maximum somatic VAF. We then examined the distribution of
ventional study and was conducted in accordance with recognized estimated ctDNA levels per indication. Although most of the
ethical guidelines. All patient data were deidentified as per an major cancers (bladder, liver, prostate, gastric, NSCLC, melano-
institutional review board-approved protocol (Quorum Review ma, breast) had similar average estimated ctDNA levels, brain
IRB Protocol 30-001: Research Use of De-Identified Specimens cancers had significantly lower levels (Wilcoxon P ¼ 0.006 in
and Data), which waived the requirement for individual patient comparison with renal) putatively owing to the blood–brain
consent. Tests were ordered by 3,283 oncologists across the barrier, whereas colorectal cancer and SCLC had significantly
United States, Europe, Asia, and the Middle East from June higher levels than all other indications examined (Wilcoxon
P < 0.008 in comparison with bladder). (Similar patterns have
Table 1. Summary of the cfDNA cohort used in this study been previously reported by Bettegowda and colleagues (17), but
Clinical characteristic Statistics
without sufficient sample sizes to determine statistical signifi-
Number of patients 21,807 cance.) This variation in ctDNA levels suggested the possibility of
Number of samples 25,578 interindication variability in variant detection and, therefore, in
Number of patients with alterations 18,503 ability to estimate tumor mutational burden from cfDNA.
Number of patients with multiple tests 2,222
Number of cancer types >50
Relationship between estimated ctDNA level and tumor
Alterations per sample 3 (median); 0–166 (range)
Days from diagnosis to blood draw 738 (mean); 335 (median) mutational burden
Gender proportion 56% female; 44% male We tested whether ctDNA levels affected tumor mutation
Age range 23–92 (median, 64) burden estimates by examining the number of SNVs per sample
NOTE: No treatment, follow-up, or outcome data were available. across cancer indications in the cfDNA cohort. Notably, when

www.aacrjournals.org Clin Cancer Res; 24(15) August 1, 2018 3529

 Zill et al.

A 100
B

4
(n = 21,807 patients; 25,578 samples)
Somatic alteration detection rate (%)

80

Percentage of total variants

3
Min: 0.03%
60 Median: 0.41%
Mean: 3.67%

2
Max: 97.62%
40

1
20

0
0 20 40 60 80 100

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

0
VAF for Reported variants (cfDNA %)
Pancreatic
Liver

Ovarian

Head & Neck

SCLC

Bladder
Esophageal

Prostate

NSCLC
Colorectal
CUP
Cholangio
Breast

Renal

Sarcoma
Melanoma

GBM
Endometrial

Gastric

C D
15
50.00

20.00
ns
10.00
*
Somatic SNVs per sample

**
10

5.00 ***
ctDNA Level (%)

2.00 ***
1.00

0.50
5

0.20

0.10

0.05
0

0.1−0.5 0.5−1 1−5 5−10 10−20 20−50

Pancreas
Melanoma
Colorectal

Prostate

Ovarian

Glioma*
Bladder

Gastric

NSCLC

Breast
SCLC

Renal
Liver

(n = 2,628) (n = 1,144) (n = 2,209) (n = 740) (n = 675) (n = 569)

ctDNA Level (%) per sample

Figure 1.
CfDNA alteration detection and estimated ctDNA levels in 21,807 advanced-stage cancer patients. A, Somatic cfDNA alteration detection rates per cancer
type in the 21,807-patient cfDNA cohort. Percentages of alteration-positive samples are indicated. Note that the last 16,939 consecutive samples (November
2015–September 2016) were analyzed with version 2.9 of the cfDNA test, whereas the previous 8,639 samples were analyzed with earlier versions of the
cfDNA test (see Supplementary Table S3). SCLC, small cell lung cancer; CUP, cancer of unknown primary; GBM, glioblastoma. B, VAF distribution for all somatic SNVs,
indels, and fusions detected by the cfDNA test. C, Distributions of estimated ctDNA level per indication. CtDNA levels were significantly higher in colorectal
cancer and SCLC and significantly lower in glioma/GBM ("Glioma ") than in the other cancers shown (Wilcoxon rank sum test). Numbers of SNV/indel/fusion-positive
samples per indication are colorectal, 1,991; SCLC, 267; bladder, 210; liver, 210; prostate, 909; gastric, 260; NSCLC, 8,078; melanoma, 410; breast, 3,301;
ovarian, 594; pancreas, 867; renal, 220; glioma/GBM, 107. D, Number of somatic cfDNA SNVs per sample versus estimated ctDNA level, which is binned on
the x-axis, in alteration-positive NSCLC samples (n ¼ 8,078). Asterisks indicate significance levels from pairwise comparisons using the Wilcoxon rank sum
test ("ns," not significant).

tumor mutation burden in this cfDNA cohort was examined, the from TCGA/ICGC whole-exome sequencing. These differences
indication ordering established with TCGA/International Cancer were likely due to several combined factors: the relatively narrow
Genome Consortium (ICGC) data of average tumor mutation cfDNA panel being heavily biased toward exons with known
burden was not recapitulated (Supplementary Fig. S1B; refs. 18, cancer mutations, differences in cohort demographics such as
19). Additionally, across all cancer indications, the mutation disease stage and treatment, and differences in variant detection
burden distributions were shifted upward, and the dynamic between tumor tissue sequencing and cfDNA sequencing. When
ranges were compressed, relative to mutation burdens derived the mutations from TCGA NSCLC cases were filtered to those

3530 Clin Cancer Res; 24(15) August 1, 2018 Clinical Cancer Research
 Somatic Genomic Landscape of Circulating Tumor DNA

lying within the cfDNA panel regions (107 kb of sequence for colorectal cancer but had middling amplification levels in lung
reported variants), this cohort's average mutation burden was 18 and prostate cancers.
mutations/Mb rather than the value of 9 mutations/Mb derived
from whole-exome analysis, consistent with the notion that Estimated cfDNA clonality and driver alteration prevalence in
higher frequencies of mutations were expected per base pair of cfDNA versus TCGA
the cfDNA panel relative to the whole exome. The abundance of advanced, treated cancer cases in the cfDNA
Nonetheless, the median tumor mutation burden estimated cohort was expected to contribute additional subclonal variants
from cfDNA steadily increased from 18.7 mutations/Mb (2 SNVs when variant detection in cfDNA was not limited by low ctDNA
per sample) in low-ctDNA NSCLC samples to 37.4 mutations/Mb levels. The trend toward increased numbers of cfDNA variants in
(4 SNVs per sample) in high-ctDNA samples (Fig. 1D; Supple- high-ctDNA samples (Fig. 1D) and the observation of frequent
mentary Fig. S1D). (The overall median was 28 mutations/Mb, or resistance alterations (Fig. 2B) suggested that comparisons of this
3 SNVs per sample.) The positive and statistically significant cfDNA cohort with large tumor tissue cohorts such as TCGA
relationship between the number of cfDNA alterations per sample should account for a higher level of mutational heterogeneity in
and ctDNA level held in breast cancer and colorectal cancer the cfDNA cases. In order to compare the prevalences of common
(Supplementary Fig. S1E and S1F), likely reflecting improved alterations between the large tissue cohorts of earlier-stage tumors
detection of genomic alterations when tumors shed more (TCGA) with those of the advanced-stage tumors in the cfDNA
DNA into circulation. As described below, at least part of the cohort, accounting for the potentially increased mutational het-

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

observed increase in mutation burdens in high-ctDNA samples erogeneity in cfDNA, we derived an estimated cfDNA clonality
was due to increased detection of subclonal variants (Supplemen- metric using the VAF/maximum VAF ratio that would allow us to
tary Fig. S2). infer the likely cancer-cell fraction of mutations present in the
tumor (see Materials and Methods). We noted that mutated
Comparisons of alteration patterns across alteration types in oncogenes such as EGFR could be subsequently amplified, which
cfDNA versus TCGA could inflate the cfDNA VAF, leading to an inaccurate clonality
To determine whether alteration patterns found in cfDNA estimate. Closer examination of the VAF/CN relationship revealed
recapitulated those found in published tissue sequencing studies, two separate nonlinear behaviors: log linearity of amplified driver
the frequencies of SNVs and indels in commonly mutated driver mutations at high VAF and high copy number, and clearly sub-
genes were compared with the frequencies found in TCGA. Highly clonal alterations with low VAF that occurred subsequent to, or in
similar mutation patterns were observed for TP53 and EGFR a separate subclone from, the amplification (Fig. 3A; Supplemen-
(Pearson r ¼ 0.94 and r ¼ 0.78, respectively; Fig. 2A and B), as tary Figs. S6–S9). Our model therefore takes into account these
well as for KRAS, BRAF, and PIK3CA (r ¼ 0.99, 0.99, and 0.94, nonlinearities by normalizing the VAF by log-transformed copy
respectively). EGFRT790M and EGFRC797S, treatment-induced resis- number for driver variants and by holding out variants that
tance mutations, were more frequent in the heavily pretreated initially appear subclonal from the normalization procedure. We
cfDNA NSCLC cases (10%) relative to the untreated TCGA NSCLC then examined the cfDNA clonality distributions for the most
cases (0.3%). Excluding T790M and C797S resistance alterations, frequently mutated genes in LUAD, breast cancer, and colorectal
the Pearson correlation for mutation frequencies in the tyrosine cancer to understand how this metric related to well-known
kinase domain of EGFR (exons 18–24) rose from 0.78 to 0.90. biological properties among cancer mutations (Supplementary
Breast cancer often harbors therapeutically targetable CNAs in Fig. S4).
ERBB2 (HER2). We compared the ranks of amplification frequen- EGFR mutations were among the most prevalent alteration
cies in breast cancer patients for the 18 CNA genes assayed by the across the cfDNA cohort, but had different expected cohort-level
cfDNA assay to the same genes in TCGA (amplification status behaviors in LUAD versus colorectal cancer. In LUAD, EGFR-
determined by GISTIC), and found high rank correlation (r ¼ activating mutations should occur frequently as drivers and these
0.86; Fig. 2C). Similarly, 5% to 10% of lung adenocarcinoma mutations should tend to be clonal. In colorectal cancer, recurrent
(LUAD) is driven by targetable kinase gene fusions. To compare EGFR extracellular domain mutations would generally be
the patterns of gene fusions found in cfDNA with those found in expected to be acquired resistance alterations in patients treated
tissue, we determined the frequencies per intron of breakpoints in with anti-EGFR antibodies such as cetuximab and, therefore,
the three most commonly observed fusions among lung cancer should tend to be subclonal. As predicted, cfDNA EGFR altera-
patients in the cfDNA cohort: EML4–ALK, CCDC6–RET, and tions in LUAD were predominantly clonal, whereas in colorectal
KIF5B–RET. Breakpoint locations for all three fusions were strong- cancer they were predominantly subclonal (Fig. 3B). Direct com-
ly correlated with the frequencies of breakpoints found in pub- parison of the clonality distributions of EGFRL858R in LUAD
lished tissue data (r ¼ 0.98; Fig. 2D; Supplementary Table S5). versus EGFR ectodomain mutations in colorectal cancer showed
A more detailed analysis of cfDNA CNAs of 18 genes across four an even more striking dichotomy (Fig. 3B). In colorectal cancer,
major cancer indications (lung, breast, colorectal, prostate) alterations in the common driver genes APC, TP53, and KRAS
revealed amplification patterns consistent with known driver were predominantly clonal (Supplementary Fig. S4). Strikingly,
alterations in each indication (Supplementary Fig. S3). For exam- nonsense mutations in APC had a strong tendency to be clonal
ple, EGFR was the most commonly amplified gene in lung cancers, (median clonality ¼ 0.72) and had significantly higher average
MYC and FGFR1 were the most commonly amplified gene in clonality than APC missense or synonymous alterations (median
breast cancer, and AR was the most commonly amplified gene in clonality ¼ 0.07; Wilcoxon P < 106), further confirming that the
prostate cancer. Notably, some established driver genes tended to cfDNA clonality metric reflected the expected behaviors of tumor-
have higher amplification levels per sample than other genes that derived alterations (Fig. 3C). In breast cancer, alterations in
reflected indication-specific biology. For instance, ERBB2 (HER2) the common driver genes PIK3CA, AKT1, and TP53 showed
had the highest average amplification levels in breast cancer and strong tendencies toward clonality (Supplementary Fig. S4).

www.aacrjournals.org Clin Cancer Res; 24(15) August 1, 2018 3531

 Zill et al.

A B
0.08 TP53 R273 EGFR

0.3
L858R
TCGA mut freq

R248

TCGA mut freq

0.06

R175

0.2
E746_A750del
0.04

R213
L861Q

0.1
0.02

G719X S768I
0

0
TAD DBD Tet Tyrosine kinase domain
R273X MAF
0

0
Min: 0.09%

ctDNA mut freq

ctDNA mut freq
0.02

Median: 0.74%
Mean: 6.4%

0.1
Max: 83.4%
0.04
0.06

Mutation type
Missense In-frame del

0.2
Mutation type 0 10 20 T790M
0.08

Nonsense Missense Cell-free DNA %

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

C D
12
20

ρ = 0.87 FGFR1
10
CNA Frequency rank (cfDNA)

8
ERBB2
MYC 6

PIK3CA 4
15

CCNE1 2
EGFR
KRAS 0

CCND1
FGFR2
10

CDK4
RAF1
BRAF
CCND2
PDGFRA
KIT
5

ctDNA Breakpoints COSMIC Breakpoints

MET
AR EML4 intron (n = 69) (n = 375)
CDK6
13 41% 47%
0

0 5 10 15 20 6 35% 35%
CNA Frequency rank (TCGA) 20 10% 14%
other 14% 4%

Figure 2.
Comparison of cfDNA alteration patterns to tumor tissue alteration patterns in TCGA and COSMIC. A, Per-codon mutation frequencies for SNVs in the TP53
coding sequence [cfDNA n ¼ 14,696 SNVs (10,574 samples); tissue/TCGA n ¼ 1,951 SNVs (1,845 samples)]. B, Per-codon mutation frequencies for the
EGFR tyrosine kinase domain (exons 18–24) [cfDNA n ¼ 3,098 SNVs/indels (2,095 samples); tissue/TCGA n ¼ 112 SNVs/indels (96 samples)]. C, Rank-by-rank
comparison of amplification frequencies in breast cancer from the cfDNA cohort (1,010 patients with amplifications out of 2,808 patients) versus the tissue/TCGA
cohort (413 samples with amplifications out of 816 profiled samples). D, Comparison of EML4–ALK fusion breakpoints for cfDNA versus tissue (COSMIC).
Top, Schematic showing breakpoints versus VAF, expressed as cfDNA percentage, for EML4–ALK fusions detected in cfDNA. Bottom, Breakpoint frequency
per EML4 intron; tissue data were compiled by the COSMIC database (http://cancer.sanger.ac.uk/cosmic) from various literature sources.

Additionally, we compared the clonality distributions in LUAD of samples (14%) than in TCGA samples (0.5%), and a substantially
known EGFR driver (e19 del, L858R, etc.) and EGFR resistance higher frequency of TP53 mutations in cfDNA colorectal cancer
(T790M, C797S) alterations. Again, as expected, the driver altera- samples.
tions showed a strong tendency toward clonality and the resis-
tance alterations showed a strong tendency toward subclonality Mutual exclusivity analysis of driver alterations in cfDNA
(Fig. 3D; Supplementary Fig. S4). To determine whether truncal driver alterations followed pat-
Comparisons of mutation prevalence per gene between the terns of mutual exclusivity established in early-stage disease (i.e.,
cfDNA and TCGA cohorts, accounting for cfDNA clonality, TCGA studies), we performed mutual exclusivity analysis on
revealed that the prevalences of most major driver alterations in common cfDNA alterations in LUAD, breast cancer, and colorec-
NSCLC, breast cancer, and colorectal cancer were overall similar tal cancer samples. Because driver status for CNAs is often unclear
(Pearson correlation, r ¼ 0.85; Supplementary Fig. S5 and Sup- or ambiguously reported across studies, we developed and
plementary Table S6). Some genes had significant differences in applied a cohort-level CNA driver identification method that
mutation prevalence between cohorts (c2 test), which largely retains statistical outliers relative to background aneuploidies to
reflected differences in patient demographics (i.e., prior treat- enrich the initial set of CNA calls for likely driver alterations
ment). Notable differences included EGFR and KRAS alterations (Supplementary Fig. S10, see Materials and Methods). Similarly,
in NSCLC (cfDNA: 43% EGFR-mutant, 16% KRAS-mutant; cfDNA SNVs, indels, and fusions were filtered to clonal alterations
TCGA/tissue: 14% EGFR-mutant, 33% KRAS-mutant), a much (clonality > 0.9) to enrich for likely truncal drivers (see Materials
higher frequency of ESR1 mutations in cfDNA breast cancer and Methods).

3532 Clin Cancer Res; 24(15) August 1, 2018 Clinical Cancer Research
 Somatic Genomic Landscape of Circulating Tumor DNA

A C

1.0
25

0.8
20

cfDNA Clonality
0.6
Copy number
15

0.4
10

0.2
0.0
L858R
5

e19 del
0 200 400 600 800
T790M Missense
C797S
Sample number Nonsense
0 20 40 60 80 100 Synonymous
VAF (cfDNA %)

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

B D

1.0
250

LUAD
CRC

0.8
200

cfDNA Clonality
0.6
150
Frequency

0.4
100

0.2
50

0.0
0

0.0 0.2 0.4 0.6 0.8 1.0 L858R T790M C797S

cfDNA Clonality

Median clonality Clonality > 0.9 Clonality < 0.1

LUAD
0.66 40% 16%
(n = 1562)
CRC
0.04 9% 62%
(n = 416)
LUAD L858R
1.0 75% 2%
(n = 383)
CRC ecto
0.02 2% 84%
(n = 89)

Figure 3.
Estimated cfDNA clonality reveals trends consistent with indication-specific biology. A, Copy number versus VAF for mutant EGFR alleles with
amplifications in NSCLC. Note the distinct population of low-VAF/high CNAs, and the log-linear behavior among high-VAF alterations. B, Estimated cfDNA
clonality is plotted for all EGFR mutations in LUAD (blue) or colorectal cancer (pink). Thresholds for clonality filtering are indicated by vertical gray lines. The median
clonality and percentage of mutations that were clonal (clonality > 0.9) or subclonal (clonality < 0.1) for all EGFR mutations, for L858R alone in LUAD, or for recurrent
ectodomain mutations ("ecto") in colorectal cancer are shown below the histogram. C, CfDNA clonality for all APC SNVs in colorectal cancer cases.
Nonsense variants are colored red, missense variants are green, and synonymous are gray. D, CfDNA clonality of a canonical EGFR driver alteration (L858R)
and two resistance alterations (T790M, C797S) in cfDNA from 1,119 NSCLC samples. Note the relationship between variant clonality and presumed order of
treatment with a given therapy is consistent with the sequential emergence of each resistance alteration (erlotinib is given to patients with EGFRL858R; EGFRT790M
confers resistance to erlotinib, and patients with EGFRL858R,T790M can then be given osimertinib; EGFRC797S confers resistance to osimertinib).

In LUAD, strong evidence for mutual exclusivity was observed both cases, but with a 30 drop in the proportion of double-
in cfDNA across several pairs of genes (Fig. 4). Importantly, the mutant [KRAS-alt; EGFR-alt] genotypes after filtering to clonal
tendency for mutual exclusivity increased when comparing the alterations. For MET and EGFR, a tendency toward alteration
post-clonality-filtering alterations to the pre-filtering alterations co-occurrence pre-filtering (FDR ¼ 6  106, log OR ¼ 0.6)
(Supplementary Figs. S11 and S12; Supplementary Tables S7– was flipped to one of exclusivity after filtering (FDR ¼ 0.03, log
S12). Of note, KRAS and EGFR were highly mutually exclusive in OR ¼ 1.0), suggesting that mutation co-occurrence before

www.aacrjournals.org Clin Cancer Res; 24(15) August 1, 2018 3533

 Zill et al.

All reported cfDNA alterations

EGFR 53%

KRAS 27%

NF1 16%

ERBB2 12%

BRAF 15%

MET 14%

ALK 5%

RET 3%

ROS1 2%

CDK4 5%

Amplification Fusion Truncating mutation Inframe mutation Missense mutation

n = 2,418

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

Truncal cfDNA alterations

EGFR 51%

KRAS 24%

NF1 7%

ERBB2 5%

BRAF 5%

MET 5%

ALK 3%

RET 1%

ROS1 1%

CDK4 1%

Amplification Fusion Truncating mutation Inframe mutation Missense mutation

n = 1,551

Figure 4.
Mutual exclusivity analysis of cfDNA alterations for LUAD before and after filtering for clonality and CNA driver status. Numbers of patients are indicated
at bottom right of each plot. The top oncoprint shows all alterations in alteration-positive patients, whereas the bottom oncoprint shows truncal SNVs, indels,
and fusions (clonality > 0.9), and likely driver CNAs across patients that have at least one clonal driver alteration. Gray boxes indicate the absence of
alterations, and the color/shape combinations corresponding to the various alteration types are indicated below each oncoprint. Frequencies of gene
alterations within each plot are indicated at left (samples lacking clonal alterations in the selected genes were omitted).

filtering was caused by subclonal resistance alterations, as colorectal cancer samples. We also noted that filtered FGFR1
opposed to co-occurring truncal mutations (the pre-filtered data amplifications and ERBB2 (HER2) amplifications showed a weak
had a high prevalence of MET amplifications and EGFRT790M, trend toward exclusivity in breast cancer and colorectal cancer,
both of which are associated with resistance to erlotinib). A although in the latter case, significance could not be readily
similar pattern of flipping from co-occurrence (nonsignificant) assessed owing to the small number of certain genotype classes.
to exclusivity was seen for ERBB2 and EGFR alterations (FDR
¼ 2  105, log OR ¼ 1.9). The landscape of actionable resistance alterations in cfDNA
In breast cancer, five driver genes (ERBB2, FGFR1, BRCA1, The large cfDNA cohort provided a unique opportunity to
BRCA2, and AKT1) showed tendencies toward mutual exclusivity explore qualitatively and quantitatively the evolution of resis-
with PIK3CA after clonality filtering. Exclusivity was not neces- tance alterations in patients who have progressed on targeted
sarily expected for PIK3CA alterations except with respect to AKT1 therapies in regular clinical practice. To estimate the frequency of
mutations (20), reflecting the more complementary nature of actionable resistance alterations (defined as alterations that might
driver alterations in this disease (Supplementary Figs. S11 and influence a physician's choice of therapy at disease progression) in
S12; Supplementary Tables S7–S12). In colorectal cancer, mutual advanced, previously treated cancer patients, we identified known
exclusivity was observed between KRAS and BRAF, ERBB2, and resistance alterations in the cfDNA cohort across six cancer types:
NRAS, similar to reports in tumor tissue. In the pre-filtered data, NSCLC, breast cancer, colorectal cancer, prostate cancer, mela-
KRAS and PIK3CA tended to co-occur, but in the post-filtering noma, and GIST. A total of 3,397 samples of the 14,998 samples
data they showed a weak trend toward exclusivity, suggesting the analyzed (22.6%) had at least one of the 134 known resistance
presence of subclonal KRAS resistance alterations in the cfDNA alterations that were identified by the cfDNA test. The proportion

3534 Clin Cancer Res; 24(15) August 1, 2018 Clinical Cancer Research
 Somatic Genomic Landscape of Circulating Tumor DNA

A NSCLC CRC Breast Prostate

100 100 100 100
ALKmut EGFRmut ERBB2mut ARmut
EGFRmut NRASmut ESR1mut None
EGFR C797S KRASmut None
80 EGFR e20ins 80 KRAS amp
80 80
Percentage of samples

Percentage of samples
MET amp None

Percentage of samples

Percentage of samples
EGFR T790M
None

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
All samples Actionable samples All samples Actionable samples All samples Actionable samples All samples Actionable samples
n = 8,190 n = 1,906 n = 2,002 n = 465 n = 3,385 n = 665 n = 970 n = 423

B Diagnosis Samples harboring putative resistance alteration Altered gene Mediates resistance to
NSCLC 11 ALK, MET Crizotinib
NSCLC 3 ALK Crizotinib/Alectinib
NSCLC 7 ALK Crizotinib/Ceritinib/Alectinib
NSCLC 1198 EGFR Erlotinib/Gefitinib
NSCLC 40 EGFR Osimertinib
CRC 1097 EGFR, KRAS, NRAS Panitumumab/Cetuximab

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

Breast 41 ERBB2 Lapatinib
Breast 646 ESR1 Letrozole/Anastrozole/Exemestane
Prostate 62 AR Abiraterone
Prostate 154 AR Abiraterone/Enzalutamide
Prostate 26 AR Bicalutamide
Prostate 9 AR Enzalutamide
Melanoma 89 KRAS, NRAS Vemurafenib
GIST 19 KIT Imatinib

C D E
50.0

50.0

50.0
TP53 H214R
EGFR L858R

KRAS G12S
10.0

10.0

10.0
cfDNA VAF (%)
cfDNA VAF (%)
cfDNA VAF (%)

ROS1 A1921S
5.0

5.0

5.0
TP53 H168R

PIK3CA E542K

ESR1 L536N PIK3CA V344M

1.0

1.0

1.0
EGFR T790M
0.5

0.5

0.5
ESR1 D538G
BRCA2 V733I
EGFR R932C NOTCH1 P2097P MET L515L
ESR1 Y537N
NTRK1 F383F
NTRK1 R314C
MET AMP
0.1

0.1

0.1

CDK6 AMP CCND1 AMP *CCND1 AMP

0 100 200 300 400 0 20 40 60 80 0 10 20 30 40 50

Time (days) Time (days) Time (days)

Figure 5.
CfDNA landscape of resistance to on-label therapies across cancer types. A, Landscape of resistance alterations in cfDNA. Numbers of patients with various
resistance alterations (y-axis, left) to targeted therapies (x-axis, bottom) in 6 common cancer types (top) are plotted. Note that some patients harbored
multiple distinct resistance alterations. Cancer type/genotype categories are AR-mutant prostate cancer, ALK-fusion–positive NSCLC, EGFR-mutant NSCLC,
breast cancer with ESR1 or ERBB2 (HER2) mutations, colorectal cancer, BRAF-mutant melanoma, and KIT-mutant gastrointestinal stromal tumor (GIST).
The "EGFRmut" category for NSCLC includes variants A722V, L747P, L747S, V769M, T854A, T854S (18 mutations in total). For each indication, the "Actionable
samples" group showed a highly significant enrichment for resistance alterations (P < 106, c2 test). B, The numbers of samples harboring putative resistance
mutations found in cfDNA and corresponding on-label targeted therapies to which these mutations would confer resistance across six cancer types. "Diagnosis"
indicates the patient diagnosis provided by ordering clinician on the test requisition form. See Supplementary Table S13 for the complete details of the
candidate resistance alterations. C, Longitudinal monitoring analysis of an NSCLC patient with emerging resistance (T790M) in the third draw after presumptive
EGFR inhibitor therapy indicated by the L858R mutation. Colored lines track the VAF (cfDNA %) of each mutation across three consecutive blood draws.
Note that the y-axis is log scale. D, Monitoring analysis showing polyclonal ESR1 mutations in a breast cancer patient, which confer resistance to aromatase
inhibitors, and possible patient response to therapy over time. Asterisk indicates four additional amplifications (not in PIK3CA) were detected in the
first sample. E, Monitoring analysis showing stability of the cfDNA clonal structure (relative VAF) over consecutive draws in an NSCLC patient lacking
on-label-therapy-indicating mutations.

of samples harboring likely resistance alterations increased when Although some resistance mutations can either occur as pri-
each indication was limited to samples harboring driver altera- mary, truncal drivers or emerge secondarily upon treatment (e.g.,
tions with associated FDA-approved targeted drugs (hereafter, KRASG12X/G13X in colorectal cancer can be a truncal driver or
"on-label targetable driver alterations"), consistent with the resis- emerge upon treatment with cetuximab), estimated cfDNA clon-
tance alterations having arisen due to therapy (Fig. 5A and B; ality helped distinguish resistance alterations whose emergence
Supplementary Table S13). The most common resistance altera- was likely caused by therapy pressure (Supplementary Fig. S13). A
tions were EGFRT790M and MET CNA in NSCLC, AR ligand- conservative estimate, focusing on clearly subclonal SNVs (clon-
binding-domain SNVs in prostate cancer, KRASG12/G13/Q61 in ality <0.1), was that at least 18.6% (ranging 10%–34% across
colorectal cancer, and ESR1L536/Y537/D538 in breast cancer cancer types) of samples with on-label targetable alterations
(Fig. 5B; Supplementary Table S13). (381/2,053) had emerging secondary resistance alterations to

www.aacrjournals.org Clin Cancer Res; 24(15) August 1, 2018 3535

 Zill et al.

those on-label therapies. Further, 24% of those resistance- amplification frequency ranks in breast cancer and locations of
harboring samples (91/381) had >1 alteration associated with intronic fusion breakpoints in NSCLC were similarly high (Fig. 2C
resistance to the same therapy, suggesting independent evolution and D). The high sensitivity of the cfDNA assay combined with the
in distinct tumor lesions (21) or sequential treatment with dis- more evolved advanced cancers tested at progression, which have
tinct therapies targeted to the same gene. For example, one NSCLC greater numbers of mutations than earlier-stage, treatment-na€ve
patient had an EML4–ALK fusion (VAF of 7.1%) and ALK SNVs cancers, may contribute to differences in estimated tumor muta-
reported to confer resistance to crizotinib (L1196M, 2.5%), tion burden versus TCGA (10, 28). Importantly, we show that
crizotinib/alectinib (I1171T, 0.1%), and crizotinib/ceritinib/alec- accurate estimation of tumor mutation burden from plasma
tinib (G1202R, 5%). In another example, the treatment history of cfDNA will require taking ctDNA level into account, as the two
certain patients harboring EGFRL858R or EGFRe19del driver altera- factors are correlated (Fig. 1D). Our estimates of ctDNA levels are
tions was immediately apparent by the combined presence of based on the copy-number-adjusted VAF of cfDNA somatic
secondary EGFRT790M and tertiary EGFRC797S resistance altera- alterations, and future studies should also consider allele-specific
tions (24 patients had both EGFRT790M and EGFRC797S—21 molecule counts (germline allele imbalance) in estimates of
patients had these two variants in cis, the other 3 were in trans). tumor DNA in circulation.
The cfDNA clonality of EGFRC797S was generally lower in those Our inference of tumor mutation clonality based on copy-
cases than that of EGFRT790M (Fig. 3C), consistent with tumor number-adjusted relative cfDNA VAF (Fig. 3) enabled a recapit-
evolution following sequential lines of treatment with erlotinib/ ulation of mutual exclusivity among truncal driver mutations and

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

afatinib/gefitinib, followed by osimertinib at progression. facilitated identification of subclonal emerging resistance altera-
Novel resistance alterations were also identified in this clinical tions (Figs. 4 and 5C; Supplementary Figs. S6–S8). These results
cohort, including ERBB2T798I (analogous to EGFRT790M), which suggest that mutation clonality, as it exists in tumor tissue, can be
causes resistance to an ERBB2 tyrosine kinase inhibitor; estimated from properly normalized relative cfDNA VAFs, as has
METD1228N, METY1230H, and METG1163R (analogous to ALKG1202R been previously hypothesized (29). High accuracy in VAF esti-
and ROS1G2302R) causing resistance of MET exon 14-mutated mation is likely key to the success of this approach, and notably,
NSCLC to a next-generation MET inhibitor; and five FGFR2 VAFs measured by the cfDNA NGS assay used in this study show
mutations (V564F, N549H, K641R, E565A, and L617V) shown good agreement with digital droplet PCR (30, 31). This approach
to drive resistance to a selective pan-FGFR inhibitor (22–26); and points to the possibility of analyzing the clonal structures of
the recurrent EGFR ectodomain mutations V441D/G, which arise tumors from cfDNA sequencing data, unencumbered by the
in the setting of cetuximab resistance in colorectal cancer but are complications of tumor heterogeneity and tumor impurity intro-
not yet characterized as functional (25). These putative resistance duced by single-region tissue sampling. However, the estimation
alterations were consistently subclonal relative to the original of tumor mutation clonality from cfDNA is subject to several
driver alteration and many were missed by single-metastatic-site sources of inaccuracy, including absence of sequence coverage
tissue biopsy but confirmed by repeat biopsy or biopsy of mul- from the cfDNA panel, nonuniform shedding of cfDNA across
tiple metastases at autopsy (22, 23, 26). tumor subclones, and low ctDNA levels. Future studies of under-
To illustrate the temporal dimension of the cfDNA landscape, explored biological factors, such as the variability of cfDNA
we identified patients with multiple tests and significant clonal shedding via tumor-cell death across patients and the uniformity
structure apparent in their ctDNA. These longitudinal cases illus- of cfDNA shedding across distinct tumor sites harboring genet-
trated emerging or polyclonal resistance after presumptive tar- ically distinct clones, could enable statistical modeling of tumor
geted therapy (Fig. 5C and D), as well as stability of VAF estimates clonal structures using cfDNA VAFs and cfDNA molecule counts
and clonality estimates over time (Fig. 5E). per locus or per allele.
The most notable differences in prevalence of driver alterations
between cfDNA and tissue cohorts were EGFR and KRAS altera-
Discussion tions in LUAD (whose prevalences were flipped), and the higher
Much of our understanding of cancer genomes is derived from frequency of ESR1 mutations in cfDNA breast cancer samples and
early-stage, treatment-na€ve cancers via consortia efforts such as of TP53 mutations in the cfDNA colorectal cancer samples. The
TCGA. However, the desire to increase treatment efficacy in higher EGFR alteration prevalence in LUAD cfDNA was likely due
advanced cancers that likely have evolved considerably from to a population bias resulting from clinicians ordering the cfDNA
baseline has led to a recent shift to "real world" cancer genomics test at progression on an EGFR TKI (median time between
studies focused on the realities of the clinic, yet grounded in diagnosis and plasma collection of 335 days). This is supported
lessons from earlier-stage cancers (27, 28). It is becoming increas- by EGFRT790M being the one of the most common EGFR variants
ingly clear that obtaining comprehensive genomic assessments, in the cfDNA cohort, second only to EGFR exon 19 deletion driver
across heterogeneous tumor subclones, will be necessary for mutations. Screening known EGFR-mutant NSCLC patients at
tailoring effective therapies for advanced cancer patients (9, 10). progression for resistance mutations is routine practice whereas
We have provided the largest cohort-level snapshot of genomic re-profiling KRAS-mutant NSCLC patients would generally not be
alterations in advanced cancer patients by cfDNA analysis in real- done, leading to an overrepresentation of EGFR driver mutations,
life clinical practice. Our results demonstrate that patterns and and the concomitant underrepresentation of KRAS mutations, in
frequencies of truncal driver alterations in advanced cancers reflect this cohort. Similarly, the higher frequency of ESR1 mutations
patterns found in early-stage disease, but also reflect the increased (a documented resistance mechanism to aromatase inhibitors)
complexity of advanced, treated cancers. We found that cfDNA likely reflects the clinical application of ctDNA assays at progres-
alterations (SNVs and small, activating indels) in TP53, EGFR, sion. There are several possible explanations for the higher TP53
KRAS, PIK3CA, and BRAF strongly correlated with TCGA tissue prevalence in cfDNA colorectal cancer samples relative to TCGA:
alterations (r ¼ 0.90–0.99, Fig. 2A and B), and that correlations for stage III/IV tumors, which predominate the cfDNA cohort, may

3536 Clin Cancer Res; 24(15) August 1, 2018 Clinical Cancer Research
 Somatic Genomic Landscape of Circulating Tumor DNA

have higher frequencies of TP53 alterations than stage I/II tumors; body of primary driver alterations characterized by the TCGA,
more subclonal TP53 mutations may have been detected due to GENIE, and other projects. As such, a portion of this database has
the high sensitivity of the cfDNA assay; or some TP53 mutations been included in the Blood Profiling Atlas in Cancer, a National
may stem from somatic myeloid malignancies known as clonal Cancer Moonshot Initiative (35). Improved detection of resis-
hematopoiesis of indeterminate potential (32, 33). The most tance mutations may facilitate enrollment in clinical trials and
likely explanation is that the TCGA cohort (< 300 samples) enable the development of more accurate biomarkers of response
underestimates TP53 mutation prevalence in colorectal cancer to therapy (22, 36, 37). Therefore, cfDNA and other minimally
relative to the larger 1,374 sample GENIE cohort, the latter invasive techniques address a real and unmet need, as it is
reporting a 68.5% mutation prevalence for this gene (34). essential to provide real-time tumor genotyping at the time of
The prevalence of resistance alterations that are informative for progression to guide subsequent therapeutic strategies.
FDA-approved therapies or clinical trials of novel targeted agents
is a key, and clinically important finding of this cfDNA study. Disclosure of Potential Conflicts of Interest
Nearly 1 in 4 cfDNA alteration–positive patients (22.6%) across 6 O.A. Zill, S.R. Fairclough, and D.R. Gandara have ownership interests
cancer indications had one or more alterations previously sug- (including patents) in Guardant Health. P.C. Mack is a consultant/advisory
gested to confer resistance to an FDA-approved on-label therapy board member for Apton Biosystems, AstraZeneca, Celgene, and Guardant
Health and reports receiving commercial research support from Boehringer
(Fig. 5A and B), which would inform clinical decision-making.
Ingelheim. A.M. Baca is a consultant/advisory board member for and has
The significant enrichments for these candidate resistance altera- ownership interests (including patents) at Guardant Health. D.I. Chudova

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

tions when cohorts were subset to patients with therapeutically has ownership interests (including patents) in Guardant Health. R.B. Lanman
informative driver alterations suggested that they were indeed has ownership interests (including patents) at Guardant Health. A. Talasaz has
linked to prior patient treatment. Our estimates of the frequencies ownership interests (including patents) at Guardant Health. No potential
of secondary, rather than primary, resistance alterations conflicts of interest were disclosed by the other authors.
(10%–34% of patients across the 6 cancer types) were likely
conservative, as we examined only low-level subclonal SNVs Authors' Contributions
(clonality < 0.1), in part because accurate assessment of clonality Conception and design: O.A. Zill, K.C. Banks, S.R. Fairclough, D.R. Gandara,
P.C. Mack, J.I. Odegaard, D.I. Chudova, R.B. Lanman, A. Talasaz
for CNAs remains difficult. As expected, the prevalence of resis-
Development of methodology: O.A. Zill, K.C. Banks, S.R. Fairclough,
tance alterations was higher in cfDNA than TCGA/tissue, as these S.A. Mortimer, J.V. Vowles, R. Mokhtari, D.R. Gandara, P.C. Mack, J.I. Odegaard,
alterations would not be present in early-stage tissue biopsies. For H. Eltoukhy, R.B. Lanman, A. Talasaz
instance, the high frequency of cfDNA ESR1 mutations in breast Acquisition of data (provided animals, acquired and managed patients,
cancer patients likely reflected prior treatment with aromatase provided facilities, etc.): K.C. Banks, J.V. Vowles, D.R. Gandara, J.I. Odegaard,
inhibitors. Additionally, EGFRT790M was one of the most common R.J. Nagy, H. Eltoukhy, R.B. Lanman, A. Talasaz
Analysis and interpretation of data (e.g., statistical analysis, biostatistics,
EGFR mutations found in the cfDNA NSCLC cohort (8% of
computational analysis): O.A. Zill, K.C. Banks, S.R. Fairclough, R. Mokhtari,
patients), but was seen in only two patients from the TCGA tissue D.R. Gandara, P.C. Mack, J.I. Odegaard, A.M. Baca, D.I. Chudova, R.B. Lanman,
NSCLC cohort (0.3%). A. Talasaz
Although our cohort of clinically ordered cfDNA tests is uncon- Writing, review, and/or revision of the manuscript: O.A. Zill, K.C. Banks,
trolled, its large size provides a realistic cross-section of patients S.R. Fairclough, S.A. Mortimer, J.V. Vowles, D.R. Gandara, P.C. Mack,
with advanced disease at the forefront of cancer care. Interpreta- J.I. Odegaard, R.J. Nagy, D.I. Chudova, R.B. Lanman, A. Talasaz
Administrative, technical, or material support (i.e., reporting or organizing
tion of our findings should take into consideration the selection
data, constructing databases): S.R. Fairclough, J.V. Vowles, D.R. Gandara,
biases related to cfDNA test ordering patterns in clinical practice P.C. Mack, J.I. Odegaard, R.J. Nagy, A.M. Baca, H. Eltoukhy, D.I. Chudova
and other potential limitations. Genomic alteration prevalence Study supervision: D.R. Gandara, J.I. Odegaard, H. Eltoukhy, R.B. Lanman
may be biased by preferential ordering of the cfDNA test for
patients with certain demographic characteristics, such as non- Acknowledgments
smoking females with NSCLC (thereby enriching for EGFR muta- We thank the patients and the physicians who submitted samples that were
tion over KRAS mutation). Plasma-based genotyping is often deidentified and used in this analysis. We thank all our colleagues at Guardant
ordered at progression in advanced cancer and thus is biased Health for their support, inspiration, and helpful discussions. The results
published here are in part based upon data generated by the TCGA Research
toward higher prevalence of resistance alterations, as discussed
Network: http://cancergenome.nih.gov/.
above. Although the cfDNA panel (70 genes) was focused pri- The study was funded by and conducted at Guardant Health, Inc. No
marily on the therapeutically informative portion of the cancer additional grant support or administrative support was provided for the study.
genome, a tradeoff of its relatively small size may be somewhat
reduced accuracy for estimating mutation clonality relative to an The costs of publication of this article were defrayed in part by the payment of
assay with a larger genomic footprint. However, it is currently page charges. This article must therefore be hereby marked advertisement in
impractical to perform whole-exome sequencing of cfDNA at accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
>15,000 coverage.
This clinical cohort represents the largest sequencing landscape Received December 27, 2017; revised March 19, 2018; accepted May 11, 2018;
of resistance in advanced cancer patients and builds upon the published first May 18, 2018.

References
1. Lebofsky R, Decraene C, Bernard V, Kamal M, Blin A, Leroy Q, et al. 2. Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E,
Circulating tumor DNA as a non-invasive substitute to metastasis biopsy et al. Intratumor heterogeneity and branched evolution revealed by multi-
for tumor genotyping and personalized medicine in a prospective trial region sequencing. N Engl J Med 2012;366:883–92.
across all tumor types. Mol Oncol 2015;9:783–90. 3. Govindan R. Cancer. Attack of the clones. Science 2014;346:169–70.

www.aacrjournals.org Clin Cancer Res; 24(15) August 1, 2018 3537

 Zill et al.

4. Gremel G, Lee RJ, Girotti MR, Mandal AK, Valpione S, Garner G, et al. 20. Network TCGA. Comprehensive molecular portraits of human breast
Distinct subclonal tumour responses to therapy revealed by circulating cell- tumours. Nature 2012;490:61.
free DNA. Ann Oncol 2016;27:1959–65. 21. Misale S, Di Nicolantonio F, Sartore-Bianchi A, Siena S, Bardelli A.
5. Murtaza M, Dawson S-J, Pogrebniak K, Rueda OM, Provenzano E, Grant Resistance to anti-EGFR therapy in colorectal cancer: from heterogeneity
J, et al. Multifocal clonal evolution characterized using circulating to convergent evolution. Cancer Discov 2014;4:1269–80.
tumour DNA in a case of metastatic breast cancer. Nat Commun 22. Goyal L, Saha SK, Liu LY, Siravegna G, Leshchiner I, Ahronian LG, et al.
2015;6:8760. Polyclonal secondary FGFR2 mutations drive acquired resistance to FGFR
6. Sacher AG, Paweletz C, Dahlberg SE, Alden RS, O'Connell A, Feeney N, et al. inhibition in patients with FGFR2 fusion-positive cholangiocarcinoma.
Prospective validation of rapid plasma genotyping for the detection of Cancer Discov 2017;7:1–12.
EGFR and KRAS mutations in advanced lung cancer. JAMA Oncol 23. Hanker AB, Brewer MR, Sheehan JH, Koch JP, Sliwoski GR, Nagy R, et al. An
2016;2:1014–22. Acquired HER2 T798I gatekeeper mutation induces resistance to neratinib in
7. Vallee A, Audigier-Valette C, Herbreteau G, Merrien J, Tessonnier L, a patient with HER2 mutant-driven breast cancer. Cancer Discov
Theoleyre S, et al. Rapid clearance of circulating tumor DNA during 2017;7:575–85.
treatment with AZD9291 of a lung cancer patient presenting the resistance 24. Lu X, Peled N, Greer J, Wu W, Choi P, Berger AH, et al. MET exon 14
EGFR T790M mutation. Lung Cancer 2016;91:73–4. mutation encodes an actionable therapeutic target in lung adenocarcino-
8. Marchetti A, Palma JF, Felicioni L, De Pas TM, Chiari R, Del Grammastro M, ma. Cancer Res 2017;77:4498–4505.
et al. Early prediction of response to tyrosine kinase inhibitors by quan- 25. Strickler JH, Loree JM, Ahronian LG, Parikh AR, Niedzwiecki D, Pereira
tification of EGFR mutations in plasma of NSCLC patients: J Thorac Oncol AAL, et al. Genomic landscape of cell-free DNA in patients with colorectal
2015;10:1437–43. cancer. Cancer Discov 2017;8:164–173.
9. Abbosh C, Birkbak NJ, Wilson GA, Jamal-Hanjani M, Constantin T, Salari R, 26. Russo M, Siravegna G, Blaszkowsky LS, Corti G, Crisafulli G, Ahronian LG,

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/24/15/3528/2047369/3528.pdf by guest on 25 April 2025

et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolu- et al. Tumor heterogeneity and lesion-specific response to targeted therapy
tion. Nature 2017;545:446–51. in colorectal cancer. Cancer Discov 2016;6:147–53.
10. Jamal-Hanjani M, Wilson GA, McGranahan N, Birkbak NJ, Watkins TBK, 27. Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, Diaz LA, Kinzler
Veeriah S, et al. Tracking the evolution of non-small-cell lung cancer. N Engl KW. Cancer genome landscapes. Science 2013;339:1546–58.
J Med 2017;376:2109–21. 28. Yates LR, Knappskog S, Wedge D, Farmery JHR, Gonzalez S, Martincorena I,
11. Pectasides E, Stachler MD, Derks S, Liu Y, Maron S, Islam M, et al. Genomic et al. Genomic evolution of breast cancer metastasis and relapse. Cancer
heterogeneity as a barrier to precision medicine in gastroesophageal Cell 2017;32:169–184.e7.
adenocarcinoma. Cancer Discov 2017;8:37–48. 29. Piotrowska Z, Niederst MJ, Karlovich CA, Wakelee HA, Neal JW, Mino-
12. Kim ST, Lee W-S, Lanman RB, Mortimer S, Zill OA, Kim K-M, et al. Kenudson M, et al. Heterogeneity underlies the emergence of EGFRT790
Prospective blinded study of somatic mutation detection in cell-free DNA wild-type clones following treatment of T790M-positive cancers with a
utilizing a targeted 54-gene next generation sequencing panel in metastatic third-generation EGFR inhibitor. Cancer Discov 2015;5:713–22.
solid tumor patients. Oncotarget 2015;6:40360–9. 30. Hong DS, Morris VK, El Osta B, Sorokin AV, Janku F, Fu S, et al. Phase IB
13. Lanman RB, Mortimer SA, Zill OA, Sebisanovic D, Lopez R, Blau S, et al. study of vemurafenib in combination with irinotecan and cetuximab in
Analytical and clinical validation of a digital sequencing panel for quan- patients with metastatic colorectal cancer with BRAFV600E mutation.
titative, highly accurate evaluation of cell-free circulating tumor DNA. PloS Cancer Discov 2016;6:1352–65.
One 2015;10:e0140712. 31. Ulrich BC, Nagy RJ, Odegaard JI, Lanman RB, Oxnard GR, Paweletz CP.
14. Newman AM, Lovejoy AF, Klass DM, Kurtz DM, Chabon JJ, Scherer F, et al. Cross-platform detection and quantification of actionable mutations in
Integrated digital error suppression for improved detection of circulating cell-free DNA shows high concordance and correlation between next-
tumor DNA. Nat Biotechnol 2016;34:547–55. generation sequencing and droplet digital PCR. Cancer Res 2017;77:5692.
15. Thompson JC, Yee SS, Troxel AB, Savitch SL, Fan R, Balli D, et al. Detection 32. Steensma DP, Bejar R, Jaiswal S, Lindsley RC, Sekeres MA, Hasserjian RP,
of therapeutically targetable driver and resistance mutations in lung cancer et al. Clonal hematopoiesis of indeterminate potential and its distinction
patients by next-generation sequencing of cell-free circulating tumor DNA. from myelodysplastic syndromes. Blood 2015;126:9–16.
Clin Cancer Res 2016;22:5772–82. 33. Zhang BM, Aleshin A, Lin CY, Ford J, Zehnder JL, Suarez CJ. IDH2 mutation
16. Odegaard JI, Vincent JJ, Mortimer S, Vowles JV, Ulrich BC, Banks KC, et al. in a patient with metastatic colon cancer. N Engl J Med 2017;376:1991–2.
Validation of a plasma-based comprehensive cancer genotyping assay 34. AACR Project GENIE Consortium. AACR Project GENIE: powering preci-
utilizing orthogonal tissue- and plasma-based methodologies. Clin Cancer sion medicine through an international consortium. Cancer Discov
Res 2018, Apr 24. doi: 10.1158/1078-0432.CCR-17-3831. 2017;7:818–831.
17. Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, Agrawal N, et al. 35. Grossman RL, Abel B, Angiuoli S, Barrett JC, Bassett D, Bramlett K, et al.
Detection of circulating tumor DNA in early- and late-stage human Collaborating to compete: blood profiling atlas in cancer (BloodPAC)
malignancies. Sci Transl Med 2014;6:224ra24–224ra24. consortium. Clin Pharmacol Ther 2017;101:589–592.
18. Lawrence MS, Stojanov P, Polak P, Kryukov GV, Cibulskis K, Sivachenko A, 36. Gainor JF, Dardaei L, Yoda S, Friboulet L, Leshchiner I, Katayama R, et al.
et al. Mutational heterogeneity in cancer and the search for new cancer- Molecular mechanisms of resistance to first- and second-generation ALK
associated genes. Nature 2013;499:214–8. inhibitors in ALK-rearranged lung cancer. Cancer Discov 2016;6:1118–33.
19. Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin 37. Mok TS, Wu Y-L, Ahn M-J, Garassino MC, Kim HR, Ramalingam SS, et al.
AV, et al. Signatures of mutational processes in human cancer. Nature Osimertinib or platinum-pemetrexed in EGFR T790M-positive lung
2013;500:415. cancer. N Engl J Med 2016;376:629–640.

3538 Clin Cancer Res; 24(15) August 1, 2018 Clinical Cancer Research