CLINICAL CANCER RESEARCH | PRECISION MEDICINE AND IMAGING

Minimal Residual Disease Detection using a Plasma-only

Circulating Tumor DNA Assay in Patients with Colorectal
Cancer A C

Aparna R. Parikh1, Emily E. Van Seventer1, Giulia Siravegna1, Anna V. Hartwig2, Ariel Jaimovich2,
Yupeng He2, Katie Kanter1, Madeleine G. Fish1, Kathryn D. Fosbenner1, Benchun Miao3, Susannah Phillips3,
John H. Carmichael3, Nihaarika Sharma3, Joy Jarnagin1, Islam Baiev1, Yojan S. Shah1, Isobel J. Fetter1,
Heather A. Shahzade1, Jill N. Allen1, Lawrence S. Blaszkowsky1, Jeffrey W. Clark1, Jon S. Dubois1,
Joseph W. Franses1, Bruce J. Giantonio1, Lipika Goyal1, Samuel J. Klempner1, Ryan D. Nipp1,
Eric J. Roeland1, David P. Ryan1, Colin D. Weekes1, Jennifer Y. Wo4, Theodore S. Hong4,
Liliana Bordeianou5, Cristina R. Ferrone5, Motaz Qadan3, Hiroko Kunitake5, David Berger5,
Rocco Ricciardi5, James C. Cusack3, Victoria M. Raymond2, AmirAli Talasaz2, Genevieve M. Boland3, and
Ryan B. Corcoran1

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

ABSTRACT
◥
Purpose: Detection of persistent circulating tumor DNA 15 patients had detectable ctDNA, and all 15 recurred [positive
(ctDNA) after curative-intent surgery can identify patients with predictive value (PPV), 100%; HR, 11.28 (P < 0.0001)]. Of 49
minimal residual disease (MRD) who will ultimately recur. Most patients without detectable ctDNA at the landmark timepoint,
ctDNA MRD assays require tumor sequencing to identify tumor- 12 (24.5%) recurred. Landmark recurrence sensitivity and speci-
derived mutations to facilitate ctDNA detection, requiring tumor ficity were 55.6% and 100%. Incorporating serial longitudinal and
and blood. We evaluated a plasma-only ctDNA assay integrating surveillance (drawn within 4 months of recurrence) samples, sen-
genomic and epigenomic cancer signatures to enable tumor- sitivity improved to 69% and 91%. Integrating epigenomic signa-
uninformed MRD detection. tures increased sensitivity by 25%–36% versus genomic alterations
Experimental Design: A total of 252 prospective serial plasma alone. Notably, standard serum carcinoembryonic antigen levels did
specimens from 103 patients with colorectal cancer undergoing not predict recurrence [HR, 1.84 (P ¼ 0.18); PPV ¼ 53.9%].
curative-intent surgery were analyzed and correlated with recurrence. Conclusions: Plasma-only MRD detection demonstrated favor-
Results: Of 103 patients, 84 [stage I (9.5%), II (23.8%), III able sensitivity and specificity for recurrence, comparable with tumor-
(47.6%), IV (19%)] had evaluable plasma drawn after completion informed approaches. Integrating analysis of epigenomic and geno-
of definitive therapy, defined as surgery only (n ¼ 39) or completion mic alterations enhanced sensitivity. These findings support the
of adjuvant therapy (n ¼ 45). In “landmark” plasma drawn 1-month potential clinical utility of plasma-only ctDNA MRD detection.
(median, 31.5 days) after definitive therapy and >1 year follow-up, See related commentary by Bent and Kopetz, p. 5449

Introduction disease, 5-year survival for patients with regional disease is only
71% (1). Surgery alone is often curative for stage I and II disease, and
Colorectal cancer is the third most commonly diagnosed and
in higher risk disease, adjuvant chemotherapy can reduce the risk of
second leading cause of cancer death in the United States in men and
recurrence (2). However, aside from risk stratification by tumor stage,
women. While a majority of patients are diagnosed with early-stage
clinical criteria, and carcinoembryonic antigen (CEA), there are no
effective clinical tools to identify patients with postoperative minimal
1
Department of Medicine, Division of Hematology and Oncology, Massachusetts residual disease (MRD) who may be at highest risk for recurrence (2).
General Hospital Cancer Center and Harvard Medical School, Boston, An effective clinical tool to identify patients with MRD following
Massachusetts. 2Guardant Health, Inc, Redwood City, California. 3Department completion of curative-intent therapy could identify patients who may
of Surgical Oncology, Massachusetts General Hospital, Boston, Massachusetts. benefit from additional systemic therapy or allow avoidance of unnec-
4
Department of Radiation Oncology, Massachusetts General Hospital Cancer essary and potentially toxic therapy for lower risk patients with no
Center and Harvard Medical School, Boston, Massachusetts. 5Department of
evidence of MRD (2–5).
General and Gastrointestinal Surgery, Massachusetts General Hospital, Boston,
Massachusetts. ctDNA is a promising noninvasive biomarker for MRD detection
following curative-intent treatment in colorectal cancer and other
Note: Supplementary data for this article are available at Clinical Cancer
Research Online (http://clincancerres.aacrjournals.org/).
cancer types. Detection of persistent ctDNA after surgery or adjuvant
treatment effectively identifies patients with colorectal cancer with
Corresponding Author: Ryan B. Corcoran, Department of Medicine, Harvard
MRD who will ultimately recur without additional therapy (6–12).
Medical School, 149 13th St, Boston, MA 02129. Phone: 617-726-8599; Fax: 617-
724-9648; E-mail: [email protected]
Accordingly, several prospective clinical trials of ctDNA-guided adju-
vant therapy are currently underway to evaluate whether patients with
Clin Cancer Res 2021;27:5586–94
evidence of MRD through ctDNA detection following surgery and/or
doi: 10.1158/1078-0432.CCR-21-0410 adjuvant therapy may benefit from additional or more intensive
2021 American Association for Cancer Research. systemic therapy to reduce recurrence risk (13–21).

AACRJournals.org | 5586
 Plasma-only ctDNA-guided MRD Detection in Patients with CRC

2016 to May 2019 at the Massachusetts General Hospital Cancer

Translational Relevance Center (Boston, MA). The study was approved by the Dana-Farber/
Detection of persistent circulating tumor DNA (ctDNA) after Harvard Cancer Center Institutional Review Board and was conducted
curative-intent surgery to identify patients with minimal residual in accordance with the Declaration of Helsinki. All patients provided
disease (MRD) who will ultimately recur has emerged as a poten- written informed consent. All patients received treatment according to
tially transformative approach in oncology. Early identification of standard of care as per the treating medical oncology and surgical
patients with MRD through ctDNA detection could identify oncology teams. Data on neoadjuvant therapy, adjuvant therapy, and
patients in whom additional therapy might salvage the chance of clinicopathologic information were collected from the electronic
cure. To date, ctDNA MRD assays have employed a tumor- health record (EHR).
informed approach, requiring initial sequencing of tumor tissue
to guide ctDNA detection, and thus cannot be used when a patient Sample collection
has insufficient tumor tissue for analysis. Here, we evaluate the first We collected 20 mL of peripheral blood in two 10 mL Streck tubes
tumor-uninformed, plasma-only ctDNA assay integrating geno- preoperatively (when available), approximately 4 weeks after surgery
mic and epigenomic signatures to detect MRD in patients with and approximately 4 weeks after completion of adjuvant therapy for
postoperative colorectal cancer, without requiring parallel tumor patients who received additional treatment. “Landmark” timepoint
sequencing, which produced favorable sensitivity and specificity, was defined as the plasma specimen drawn approximately 1 month
comparable with tumor-informed approaches. These data high- after completion of definitive therapy (surgery alone or completion of

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

light the feasibility and potential clinical utility of plasma-only adjuvant therapy for patients who received adjuvant treatment).
ctDNA-guided MRD detection. Longitudinal timepoints were defined by patients who had subsequent
draws after their “landmark” timepoint, and, based on the methods
employed by Reinert and colleagues, we also assessed performance in
patients with evaluable “surveillance” draws, defined as a draw
To date, most ctDNA assays designed for MRD detection rely on obtained within 4 months of clinical recurrence. Because of the
initial genomic profiling of tumor tissue to identify tumor-derived variability of follow-up, a window of up to 6 weeks after recurrence
alterations specific for each individual patient, so that these precise scan was allowed for some serial draws to allow them to be collected at
alterations can be evaluated in ctDNA (6, 9, 22, 23). The rationale the patient’s scheduled visit, provided that no intervening therapy was
behind such “tumor-informed” approaches is that prior knowledge of received. CEA at the landmark timepoint ( 9 days) was obtained from
the tumor-specific mutations may allow increased sensitivity for the EHR. The timing of CEA draws was not mandated by the study and
ctDNA detection and may improve specificity by enabling determi- was performed according to the treating physician’s discretion accord-
nation of which alterations are tumor-derived versus potential false ing to appropriate standard-of-care guidelines. Abnormal CEA was
positives arising from nontumor origins, such as clonal hematopoiesis defined as >3.4 ng/mL (28). Serial blood collections from eleven
of indeterminate potential (CHIP; refs. 24–26). However, a tumor- patients who recurred but were negative at their landmark timepoint
informed approach may pose several limitations. Importantly, in many were analyzed. Plasma was separated within 1–4 days of collection
cases, tumor cellularity may be limited, which would preclude the use through two different centrifugation steps (the first at room temper-
of a tumor-informed approach. In one series, up to 9% of molecular ature for 10 minutes at 1,600  g and the second at 3,000  g for the
analysis may be inadequate for tissue sequencing given low tumor same time and temperature). Plasma was isolated and stored at 80
cellularity, DNA yield, or quality (27). This issue is particularly until extraction. In a subset of samples (N ¼ 72), cell-free DNA
relevant alongside the increasing use of neoadjuvant therapy in many (cfDNA) was extracted from plasma using QIAamp Circulating
tumor types, where surgical specimens may have insufficient tissue for Nucleic Acid Kit (QIAGEN) with 60 minutes of proteinase K incu-
molecular analysis due to treatment response. In contrast, a plasma- bation at 60 C.
only ctDNA assay for MRD detection could offer several advantages,
including more rapid turnaround time due to the need to analyze a Plasma genomic and epigenomic based analysis of cfDNA
single plasma sample, potential cost savings, and decreased logistical Plasma (N ¼ 180 samples; median, 4 mL; range, 1–4 mL) and
complexity. However, no studies have evaluated if a plasma-only MRD extracted cfDNA (N ¼ 72 samples; median, 60 ng; range, 4.4–100 ng)
assay can detect ctDNA with clinically meaningful specificity and were transferred to a single site for analysis (Guardant Health). cfDNA
sensitivity. was extracted from plasma as described previously (29). Extracted
We report results from a prospective, observational study in patients cfDNA was analyzed using the plasma-only Guardant Reveal test
with stage I to IV colorectal cancer treated with curative-intent therapy (formerly called LUNAR-1), which is a single sample next-generation
to assess the ability of a plasma-only ctDNA assay to identify patients sequencing test validated in early-stage colorectal cancer that inte-
with MRD who would ultimately recur. In addition to standard grates assessment of somatic alterations with an epigenomic cancer
detection of tumor-derived genomic alterations employed by most signature to identify the presence of methylation signatures associated
MRD assays, this tumor-uninformed assay (Guardant Reveal, Guar- with cancer versus normal DNA (Fig. 1A). The Guardant Reveal assay
dant Health) also integrates epigenomic signatures related to aberrant was designed to detect the presence of MRD without prior knowledge
DNA methylation facilitating detection of ctDNA without requiring of the specific molecular alterations present in an individual patient’s
parallel assessment of tumor tissue. tumor.
For analysis, ctDNA fragments are partitioned and individual
molecules within each partition are barcoded and then pooled and
Materials and Methods processed together through the rest of the library preparation. The
Study population libraries are enriched with an approximately 500 kb panel targeting
This single-institution prospective study recruited patients with both somatic and epigenomic regions using biotinylated bait oligo-
stages I–IV colorectal cancer treated with curative intent from August nucleotides and sequenced on a NovaSeq 6000 System. Enriched

AACRJournals.org Clin Cancer Res; 27(20) October 15, 2021 5587

 Parikh et al.

A Methylated
cfDNA Signal
Methylation
processing
Plasma
isolation from
whole blood Target capture ctDNA
collected in digital sequencing detection
(500 kb panel)
StreckBCT
Genomic Biological
Non- alterations noise
Methylated filter
cfDNA

B 103 Curative-intent surgery

9 Excluded
• Not NED post-op

94 NED post-op

10 Excluded

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

• Limited sample available

84 Evaluable for MRD analysis

Landmark analysis Longitudinal analysis

4 Excluded 4 Excluded
• Patients without blood • Patients without blood
draw landmark timepoint post-definitive treatment
80 with draws at 80 with blood draws
landmark timepoint post-definitive treatment

10 Excluded 8 Excluded
• Insufficient cfDNA • Insufficient cfDNA
quantity and/or quality quantity and/or quality

70 evaluable for 72 evaluable for

landmark analysis longitudinal analysis

Figure 1.
Guardant Reveal plasma-only ctDNA assay schema (A) and patient enrollment and analysis groups (B).

samples are then amplified and sequenced. Sequencing data files rank test was used for HRs and all P values were based on two-sided
containing raw data are analyzed using a proprietary bioinformatics testing with statistically significant differences at P ≤ 0.05. Statistical
pipeline software, trained to detect the presence of ctDNA based on analysis was performed using Graphpad PRISM version 8.0 for
multiple analytic features, including genomic variation (single- Windows, GraphPad Software.
nucleotide variants and insertion-deletion alterations) and epigenomic
signals, and to exclude common sources of interference such as CHIP.
On the basis of this analysis, the plasma-only ctDNA test returns a Results
result of either ctDNA detected or ctDNA not detected. For the current We evaluated the feasibility of tumor-uninformed MRD detection
study, ctDNA analysis was performed blinded to the clinical data. with a plasma-only MRD ctDNA assay in 103 patients with stage I–IV
Neither treating physicians nor patients were informed regarding the colorectal cancer undergoing curative-intent surgery. This assay
results of the ctDNA analyses. (Guardant Reveal, Guardant Health) evaluates epigenomic signatures
related to aberrant DNA methylation in addition to “standard”
Statistical analysis detection of tumor-derived genomic alterations employed by most
The primary outcomes of the study were detection of ctDNA and MRD assays, without requiring parallel assessment of tumor tissue
recurrence-free survival (RFS) as assessed by standard radiographic (Fig. 1A). Genomic alterations detected are filtered to remove variants
imaging. RFS was measured from the day of completion of definitive of likely benign origin (e.g., CHIP). Overall, 84 patients were evaluable
treatment to first radiographic recurrence or death from colorectal (Fig. 1B), and patient characteristics (9.5% stage I, 23.8% stage II,
cancer. For patients whose only treatment was surgery or surgery was 47.6% stage III, and 19% stage IV) are shown (Table 1; Supplementary
the final intervention, definitive treatment was defined as day of Tables S1 and S2). A total of 45% of patients received neoadjuvant
surgery. For patients who received adjuvant chemotherapy, RFS was therapy and 53.6% received adjuvant therapy. A total of 16 of 84 (19%)
measured from the day of completion of adjuvant therapy. Patients had surgery alone with no neoadjuvant or adjuvant therapy. Overall,
were censored at the date of last follow-up or non-colorectal cancer– 30 of 84 (35.7%) patients recurred with a median time to recurrence
related death. Patients without clinical follow-up available were from surgery of 348.5 days (range, 35–887) and median time to
excluded from the study. Analysis was completed for patients with recurrence from completion of definitive treatment of 211 days (range,
at least 1 year of follow-up and for the overall eligible cohort. The 7–887). Blood was drawn a median of 30 days (range, 11–148)
Kaplan–Meier method was used to describe survival outcomes. A log- postoperatively and median of 32 days (range, 0–193) after completion

5588 Clin Cancer Res; 27(20) October 15, 2021 CLINICAL CANCER RESEARCH
 Plasma-only ctDNA-guided MRD Detection in Patients with CRC

Table 1. Baseline patient and disease characteristics. specificity for recurrence was 55.6% (95% confidence interval, 35.3–
74.5) and 100% (95% confidence interval, 90.5–100). Negative pre-
Overall cohort dictive value was 75.5% (95% confidence interval, 61.1–86.7). ctDNA
Characteristic N ¼ 84 % detection predicted recurrence regardless of stage, neoadjuvant, or
Age (years) – median (range) 60 (35–84) adjuvant therapy (Supplementary Table S4).
Sex
Female 33 39.3 Longitudinal and surveillance ctDNA analyses
Male 51 60.7 Prior studies have shown that sensitivity for recurrence prediction
Stage at surgery can be improved with longitudinal plasma monitoring. Overall, 9 of 14
I 8 9.5 (64%) of patients who recurred despite no detectable landmark ctDNA
II 20 23.8 or who lacked landmark draws had at least one evaluable longitudinal
III 40 47.6
specimen at a later timepoint (median, 3 per patient). Integrating
IV 16 19.0
Sidedness
longitudinal specimens increased sensitivity from 55.6% to 69.0%
Right 18 21.4 [HR, 12.26 (P < 0.0001)], with specificity remaining 100% (Fig. 3B;
Transverse 5 6.0 Supplementary Fig. S1B). Based on the methods employed by Reinert
Left 31 36.9 and colleagues (6), we assessed performance in patients with evaluable
Rectal 30 35.7 “surveillance” draws, defined as a draw obtained within 4 months of

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

Neoadjuvant treatment 38 45.2 clinical recurrence, and observed that sensitivity improved to 91%.
Adjuvant treatment 46 54.8
Type of adjuvant treatment Integration of genomic and epigenomic assessment of ctDNA
FOLFOX 31 67.4
Because this is the first MRD assay to leverage ctDNA methylation
CAPOX 7 15.2
analysis in addition to genomic alterations, compared with other
FOLFOX þ chemoxRT 3 6.5
5FU/LV 3 6.5 ctDNA MRD assays that detect genomic alterations only, we assessed
Other 2 4.3 the relative contributions of integrating genomic and epigenomic
Days on adjuvant treatment – 134.5 (28–463) signatures for ctDNA detection. ctDNA methylation and other epi-
median (range) genomic markers, such as ctDNA fragmentation, show promise in
Experienced disease recurrence 330 335.7 ctDNA-based early cancer detection approaches, which are tumor
Days from surgery to recurrence – 348.5 (35–887) uninformed, and thus might also improve MRD detection (30–32).
median (range) Across all ctDNA-positive specimens, 47% of samples were positive by
Days of clinical follow-up from 632.5 (33–1,246) both epigenomic and genomic calls, while 28% and 25%, respectively,
surgery – median (range)
were positive by either genomic or epigenomic calls alone (Fig. 3C).
For the landmark analysis, using genomic calls alone, recurrence
sensitivity would have been only 40.7%, and 48.2% using epigenomic
of definitive therapy (Table 1). Four patients were excluded from the calls alone. By integrating genomic and epigenomic calls, sensitivity
landmark analysis as they did not have an appropriate blood draw at improved to 55.6%. Similarly, in the longitudinal and surveillance
after completion of definitive therapy. Of the 80 remaining patients, 10 analyses, respectively, sensitivity would have been only 55.2% and
patients were excluded from the primary analysis as after curative- 72.7% with genomic calls alone or 48.3% and 63.6% with epigenomic
intent specimens yielded extracted cfDNA that was of insufficient calls alone, but sensitivity improved to 69.0% and 90.9% by integrating
quantity or failed to pass sequencing quality control thresholds, leaving genomic and epigenomic calls. Thus, incorporating epigenomic calls
70 (88% success rate) patients evaluable for the primary landmark with standard genomic calls increased relative MRD detection sensi-
analysis. Timing and receipt of neoadjuvant therapy, surgery, adjuvant tivity by 25%–36% across analysis groups, suggesting ctDNA meth-
therapy, and serial blood collections for each patient are shown ylation may be an effective modality for MRD detection.
(Fig. 2).
Analysis of CEA levels and recurrence
Primary “landmark” analysis As a basis of comparison, we evaluated the ability of the standard-of-
For the primary analysis, a single “landmark” plasma specimen care colorectal cancer serum tumor marker CEA to predict recurrence
drawn approximately 1 month after completion of definitive therapy in this same cohort. CEA values after definitive therapy were available
(median, 31.5 days) was assessed, as early MRD detection is critical to for 56 patients at their landmark timepoint. Notably, CEA values failed
enable therapeutic decisions during the standard window for adjuvant to predict recurrence (Fig. 4; Supplementary Fig. S2).
therapy initiation. Of 70 landmark evaluable patients, 17 of 70 (24%)
patients had detectable ctDNA after completion of definitive therapy
and 15 of 17 (88%) of these patients recurred (specificity 95.4%; Discussion
Supplementary Fig. S1A; Supplementary Table S3). However, the 2 In this prospective cohort study, we report initial findings for one of
patients with detectable ctDNA but no recurrence both had clinical the first plasma-only assays designed for tumor-uninformed ctDNA-
follow-up of less than 1 year. In patients with at least 1 year of clinical based MRD detection, which demonstrated favorable sensitivity and
follow-up available at the data cutoff, 15 of 15 patients with detectable specificity, comparable with previously reported tumor-informed
ctDNA (ctDNA-detected) recurred at a median of 162 days after approaches. This study also represents one of the first assessments
definitive therapy [positive predictive value of recurrence of 100% of a ctDNA assay incorporating both genomic and epigenomic mar-
(95% confidence interval, 78.2–100); HR, 11.28 (P < 0.0001)] (Fig. 3A). kers for MRD detection which also increased sensitivity.
Of the 49 patients without detectable landmark ctDNA, 12 of 49 (24%) One of the primary concerns with plasma-only assays is that
recurred (median time to recurrence, not reached). Sensitivity and specificity and sensitivity might be limited if the assay is not guided

AACRJournals.org Clin Cancer Res; 27(20) October 15, 2021 5589

 Parikh et al.

A Surgery only B Surgery + adjuvant therapy

TPS489 TPS1715

TPS880
TPS485 = ctDNA not detected
TPS296
TPS186
TPS1613
= ctDNA detected
TPS493
TPS480 = Failed QC
TPS521 TPS1047
{ = Adjuvant start
TPS236 TPS626

TPS796 } = Adjuvant end

TPS644

TPS751
TPS906 = Recurrence
TPS1363
TPS733
TPS216
= Clinical follow-up
TPS065 TPS896 = Neoadjuvant treatment
TPS460 TPS1473

TPS848
TPS067
TPS1425
TPS113
TPS443
TPS406
TPS1355

TPS725 TPS618

TPS865 TPS527

TPS1279
TPS742
TPS165
TPS120

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

TPS444
TPS306 TPS1040

TPS854 TPS1170

TPS999
TPS849
TPS1019
TPS182
TPS1025
TPS666
TPS658
TPS550 TPS899

TPS430 TPS446

TPS822
TPS718
TPS792
TPS218
TPS621
TPS641
TPS497

TPS461 TPS411

TPS259 TPS529

TPS483
TPS575
TPS278
TPS440
TPS392
TPS290
TPS416

TPS486 TPS265

TPS285 TPS172

TPS300
TPS425
TPS331
TPS330
TPS254
TPS149
TPS246

–50 50 100 150 200 250 300 350 400 450 500 550 600 –50 50 100 150 200 250 300 350 400 450 500 550 600
y

y
1, 00
0
er

1, 0

1, 00
0
er

1, 0
25

Days from surgery

75

25
Days from surgery

75
0
rg

rg

0
Su

Su

Figure 2.
ctDNA assay results and timing of treatment for each patient. Patients are ordered by total days of clinical follow-up from surgery for patients that received curative-
intent surgery alone (A) and for patients that received curative-intent surgery followed by adjuvant therapy (B). Patients receiving neoadjuvant therapy are
designated by a yellow square. Specimens obtained on the day of surgery were obtained preoperatively.

by specific alterations identified in the resected tumor. Notably, the loss approximately 40%–50% when specifically assessing a single landmark
of specificity is a critical concern, as noncancer-derived mutations are timepoint obtained approximately 1-month after therapy (6, 9), which
frequently present in the blood which could lead to false positives (24). is comparable with the landmark sensitivity of 55.6% produced by this
In the current study, specificity was 100% in patients with at least plasma-only tumor-uninformed assay. Furthermore, recent evidence
1-year minimum clinical follow-up, which aligns with specificity of has demonstrated that serial monitoring for recurrence can increase
other tumor-informed MRD approaches for colorectal cancer (6, 8). In detection sensitivity. For example, Reinert and colleagues found that
the overall cohort, 2 patients with ctDNA detected had not yet recurred for patients with surveillance draws, sensitivity improved to 88% (6).
by the cut-off date. However, both patients had clinical follow-up of Similarly, when evaluating patients with analogous surveillance draws
less than a year. Further analysis of larger cohorts is needed as high in our cohort, defined as at least one draw within 4 months of clinical
specificity of MRD detection will remain critical if MRD assays are to recurrence, we observed that sensitivity improved to 91%. Thus, these
be used to select patients for additional or more intensive therapy. This initial data suggest that this plasma-only assay can achieve perfor-
would avoid situations in which patients who are cured are erroneously mance characteristics comparable with currently approved tumor-
identified as MRD positive and subjected to potentially unnecessary informed approaches.
therapy. Importantly, we also evaluated whether assessment of epigenomic
In addition, we found that sensitivity was comparable with the markers (specifically, changes in DNA methylation patterns) in
performance of many tumor-informed assays at the landmark time- ctDNA might increase the effectiveness for MRD detection. Indeed,
point. Landmark detection sensitivity is of central importance, as this methylation and other epigenomic markers, and DNA fragmentation
assessment occurs within the time window in which treatment deci- patterns show promise in ctDNA-based early detection approaches,
sions are typically made. Previously reported “fixed panel” and which are tumor uninformed, and thus may also help with MRD
“bespoke” tumor-informed MRD assays produced sensitivities of detection (30–34). Our data suggest that epigenomic methylation

5590 Clin Cancer Res; 27(20) October 15, 2021 CLINICAL CANCER RESEARCH
 Plasma-only ctDNA-guided MRD Detection in Patients with CRC

A Landmark analysis P < 0.0001

100 ctDNA detected n = 49 n = 15

Recurred
Probability of survival

ctDNA not detected 100%

Percentage of patients
90% Not recurred
80%
70%
60% 37 Sensitivity = 55.6%
50% Specificity = 100%
50 15
ctDNA Positive mRFS = 168 days 40% PPV = 100%
ctDNA Negative mRFS = Not 30% NPV = 75.5%
Reached 20%
10% 12
Log rank: P < 0.0001
0%
Hazard ratio: 11.20 ctDNA ctDNA
0 not detected detected
0 500 1,000 1,500
Epigenomic Genomic
Days post-completion of therapy 4 9 2 positives
positives

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

B
1
2 1
7 ctDNA detected
7
9 ctDNA not detected
8
12 2 2 No surveillance draw
Failed landmark draw
4 Included in analysis

20 20
15
Epigenomic Genomic
4 10 6
positives positives

Landmark analysis Longitudinal analysis Surveillance analysis

Sensitivity: 55.6% (15/27) Sensitivity: 69.0% (20/29) Sensitivity: 90.9% (20/22)

C Sensitivity (%) by ctDNA calling methods

Genomic only Epigenomic only Genomic & epigenomic
100
90.9%

Epigenomic 15 28 17 Genomic
positives positives
80 72.7%
69.0%
63.6%
Sensitivity (%)

60 55.6% 55.2%

48.2% 48.3%

40.7%
40

20

0
Landmark Longitudinal Surveillance

Figure 3.
A, RFS by landmark 1-month posttherapy ctDNA detection for patients with >1 year follow-up. Bar graph displays recurrence rates by ctDNA detection status with
Venn diagram of ctDNA detection by calling methods (epigenomic only, genomic only, or both). B, Sensitivity analysis for landmark, longitudinal, and surveillance
cohorts and Venn diagram of ctDNA detection for longitudinal and surveillance analyses by calling method. C, Recurrence sensitivity and Venn diagram of ctDNA
detection by individual calling methods.

AACRJournals.org Clin Cancer Res; 27(20) October 15, 2021 5591

 Parikh et al.

P = 0.3241
Landmark timepoint
(1 year minimum Follow-Up) n = 13 n = 38
Recurred
100%

Percentage of patients
100 CEA Abnormal 90% Not recurred
Probability of survival

CEA Normal 80% 6

70% Sensitivity = 35.0%
25
60% Specificity = 80.7%
50 50% PPV = 53.9%
40% NPV = 65.8%
CEA Abnormal mRFS = 400 days
30%
CEA Normal mRFS = Not reached 7
Log rank: P = 0.1846 20% 13
Hazard ratio: 1.840 10%
0 0%
0 500 1,000 1,500 Abnormal Normal
Days post-completion of therapy

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

Figure 4.
RFS by landmark 1-month posttherapy CEA analysis for patients with >1 year follow-up. Bar graph displays recurrence rates by CEA status.

signatures in ctDNA can allow MRD detection with high specificity samples had plasma input volumes of 4 mL or less, versus the
and may improve performance compared with detection of genomic recommended input of 8–10 mL, which may have affected overall
calls alone, which is the method utilized by most MRD assays. performance characteristics. Finally, this study focused primarily on
Interestingly, we observed that while most cases showed ctDNA analysis of a single landmark timepoint and did not systematically
positivity by both genomic and epigenomic signatures, many cases incorporate serial longitudinal draws for all patients. While effective
were detected as positive by genomic or epigenomic calls alone detection of MRD at a landmark timepoint early after completion of
(Fig. 4A and B), suggesting that integrating these two modalities may curative-intent therapy is clinically relevant, recent data suggest a
augment sensitivity for MRD detection. Specifically, integration of potential value for serial monitoring during surveillance (6). Although
epigenomic signatures for landmark MRD detection increased sensi- incorporation of longitudinal and surveillance draws available for
tivity by a relative 36% compared with genomic alterations alone. some patients did improve overall sensitivity from 55.6% to 69% and
Overall, a plasma-only ctDNA assay for MRD detection may hold 91%, respectively, the lack of systematic longitudinal and surveillance
several advantages relative to tumor-informed approaches. These draws across all patients precluded a comprehensive assessment.
advantages include a more rapid turnaround time as only a single In summary, we show that plasma-only, tumor-uninformed
plasma draw is required, potential cost savings, and decreased logis- ctDNA-based detection of MRD is feasible and can produce compa-
tical complexity. Moreover, a plasma-only assay would offer the rable sensitivity and specificity to previously reported tumor-informed
potential for MRD detection even in situations where tumor tissue MRD approaches. These data also suggest that integration of epige-
is not available or sufficient for use in tumor-informed ctDNA nomic markers, such as DNA methylation analysis, may enhance
detection. Although sufficient tumor tissue is often available following detection sensitivity over standard genomic alteration detection meth-
a surgical procedure, some studies have suggested that in a subset of ods alone and the integration of genomic and epigenomic assessment
patients available tissue may be insufficient for molecular analysis, improves performance. The Guardant Reveal test is currently being
which would preclude the ability to utilize a tumor-informed approach utilized in several prospective clinical trials to assess the impact of
for MRD detection (27). As neoadjuvant therapy becomes increasingly ctDNA-guided adjuvant therapy (13, 17, 21). Ongoing prospective
utilized as standard of care for many tumor types, a growing number of interventional studies will further evaluate the performance of this
surgical specimens may yield insufficient tissue for molecular analysis assay for MRD detection and to help guide treatment decisions.
after surgery, due to low tumor cellularity following a favorable
treatment response (35). Thus, in some situations, a plasma-only Authors’ Disclosures
ctDNA assay might offer the only means of assessing for MRD in A.R. Parikh reports personal fees from Foundation Medicine, Natera, Checkmate
certain situations. Pharmaceuticals, Eli Lilly, Pfizer, and Roche; and other from Puretech, PMV Pharma,
While this study supports the potential of plasma-only ctDNA BMS, Novartis, Plexxicon, Takeda, Macrogenics, C2I genomics outside the submitted
detection of MRD, this study also has several key limitations. First, the work. E.E. van Seventer reports personal fees and other from Blueprint Medicines
sample size is modest, and further evaluation in larger patient cohorts outside the submitted work. A.V. Hartwig reports being an employee and shareholder
of Guardant Health. A. Jaimovich reports other from Guardant Health during the
will be important to better define the assay performance character-
conduct of the study. Y. He reports being an employee and shareholder of Guardant
istics. Second, while this study suggests the potential for effective MRD Health. J.W. Franses reports personal fees from Foundation Medicine during the
detection across multiple stages of colorectal cancer and across conduct of the study. L. Goyal reports receiving research funding (to institution) from
different treatment pathways (neoadjuvant and/or adjuvant therapy), Agios, Adaptimmune, Bayer, Eisai, Merck, Macrogenics, Genentech, Novartis, Incyte,
a more focused and in-depth analysis will be important to understand Eli Lilly, Loxo Oncology, Relay Therapeutics, QED, Taiho Oncology, Leap Thera-
the true performance in specific patient populations. Third, while this peutics, Bristol Meyers Squibb, and Nucana; scientific advisory committee (to self)
from Agios Pharmaceuticals Inc, Alentis Therapeutics AG, H3Biomedicine, Incyte
cohort provides a potentially valuable initial assessment the study
Corporation, QED Therapeutics, Sirtex Medical Ltd, and Taiho Oncology Inc.;
design utilizing banked, isolated plasma samples, showed that in some consulting (to self) from Agios Pharmaceuticals Inc, Alentis Therapeutics, Genentech,
cases the extracted ctDNA quantity or quality was below the recom- Exelixis, Incyte Corporation, QED Therapeutics, Sirtex Medical Ltd, and Taiho
mended and optimal input levels for the assay. In particular, all of our Oncology Inc.; and DSMC (to self) from AstraZeneca. S.J. Klempner reports personal

5592 Clin Cancer Res; 27(20) October 15, 2021 CLINICAL CANCER RESEARCH
 Plasma-only ctDNA-guided MRD Detection in Patients with CRC

fees from Merck, BMS, Eli Lilly, Astellas, Daiichi-Sankyo, Natera, Inc, and Pieris I.J. Fetter: Data curation, formal analysis, writing–review and editing.
Oncology, and other from Turning Point Therapeutics outside the submitted work. H.A. Shahzade: Data curation, formal analysis, writing–review and editing.
E.J. Roeland has served as a consultant for Mitobridge Inc., Asahi Kasei Pharma- J.N. Allen: Data curation, formal analysis, writing–review and editing.
ceuticals, DRG Consulting, Napo Pharmaceuticals, American Imaging Management, L.S. Blaszkowsky: Data curation, formal analysis, writing–review and
Immuneering Corporation, and Prime Oncology; additionally, E.J. Roeland has editing. J.W. Clark: Data curation, formal analysis, writing–review and
served on recent advisory boards for Heron Pharmaceuticals, Vector Oncology, and editing. J.S. Dubois: Data curation, formal analysis, writing–review and
Helsinn Pharmaceuticals and has also served as a member on data safety monitoring editing. J.W. Franses: Investigation, writing–review and editing. B.J. Giantonio:
boards for Oragenics, Inc, Galera Pharmaceuticals, and Enzychem Lifesciences Data curation, formal analysis, writing–review and editing. L. Goyal: Data curation,
Pharmaceutical Company. D.P. Ryan reports personal fees and other support from formal analysis, writing–review and editing. S.J. Klempner: Data curation, formal
MPM; other support from Acworth Pharmaceuticals; personal fees from Iteos, analysis, writing–review and editing. R.D. Nipp: Data curation, formal analysis,
Uptodate, McGraw Hill, and Boehringer Ingelheim; non-financial support from writing–review and editing. E.J. Roeland: Data curation, formal analysis, writing–
Exact Sciences; and grants and personal fees from SU2C during the conduct of the review and editing. D.P. Ryan: Data curation, formal analysis, writing–review and
study; personal fees and other support from MPM; other support from Acworth editing. C.D. Weekes: Data curation, formal analysis, writing–review and editing.
Pharmaceuticals, Exact Sciences; personal fees from Iteos, Uptodate, McGraw Hill, J.Y. Wo: Data curation, formal analysis, writing–review and editing. T.S. Hong: Data
and Boehringer Ingelheim, and grants and personal fees from SU2C outside the curation, formal analysis, writing–review and editing. L. Bordeianou: Data curation,
submitted work. T.S. Hong reports personal fees from Merck, Novocure, and formal analysis, writing–review and editing. C.R. Ferrone: Data curation, formal
Synthetic Biologics outside the submitted work. V.M. Raymond reports being an analysis, writing–review and editing. M. Qadan: Data curation, formal analysis,
employee and shareholder of Guardant Health. A. Talasaz reports other support from writing–review and editing. H. Kunitake: Data curation, formal analysis, writing–
Guardant Health, Inc. outside the submitted work. G.M. Boland reports grants from review and editing. D. Berger: Data curation, formal analysis, writing–review

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

Takeda Oncology, Olink Proteomics, and Palleon Pharmaceuticals; grants and other and editing. R. Ricciardi: Data curation, formal analysis, writing–review and editing.
from Novartis and Merck; and other from Nektar Therapeutics outside the submitted J.C. Cusack: Data curation, formal analysis, writing–review and editing.
work. R.B. Corcoran reports personal fees from Abbvie, Pfizer, Astex Pharmaceu- V.M. Raymond: Conceptualization, data curation, formal analysis, supervision,
ticals, Chugai, Elicio, Fog Pharma, Guardant Health, Ipsen, Mirati Therapeutics, validation, investigation, methodology, writing–original draft, project administration,
Natera, Navire, Qiagen, Roivant, Shionogi, Tango Therapeutics, Taiho, and Zikani writing–review and editing. A. Talasaz: Resources, data curation, formal analysis,
Therapeutics; grants and personal fees from Asana Biosciences and AstraZeneca; supervision, project administration, writing–review and editing. G.M. Boland:
personal fees and other from Avidity Biosciences, C4 Therapeutics, Kinnate Bio- Resources, data curation, formal analysis, supervision, investigation, methodol-
pharma, nRichDx, Remix Therapeutics, and Revolution Medicines; and other from ogy, writing–original draft, project administration, writing–review and editing.
Erasca outside the submitted work. No disclosures were reported by the other authors. R.B. Corcoran: Conceptualization, resources, data curation, formal analysis,
supervision, funding acquisition, validation, investigation, writing–original draft,
Authors’ Contributions writing–review and editing.
A.R. Parikh: Conceptualization, data curation, formal analysis, supervision, valida-
tion, investigation, methodology, writing–original draft, writing–review and editing. Acknowledgments
E.E. Van Seventer: Conceptualization, data curation, formal analysis, writing– This study was supported by a NIH/NCI Gastrointestinal Cancer SPORE (P50
original draft, writing–review and editing. G. Siravegna: Conceptualization, data CA127003, to R.B. Corcoran) and Stand Up To Cancer Colorectal Dream Team
curation, formal analysis, supervision, validation, investigation, writing–original Translational Research Grant (SU2C-AACR-DT22-17, to R.B. Corcoran). Stand Up
draft, project administration, writing–review and editing. A.V. Hartwig: Data To Cancer is a division of the Entertainment Industry Foundation. Research grants
curation, formal analysis, writing–review and editing. A. Jaimovich: Data curation, are administered by the American Association for Cancer Research, the Scientific
formal analysis, writing–review and editing. Y. He: Data curation, formal analysis, Partner of SU2C. A.R. Parikh was supported by a Conquer Cancer Foundation of
writing–review and editing. K. Kanter: Data curation, formal analysis, writing–review ASCO Career Development Award and an American Cancer Society Institutional
and editing. M.G. Fish: Data curation, formal analysis, writing–review and editing. Research Grant.
K.D. Fosbenner: Data curation, formal analysis, writing–review and editing. B. Miao:
Data curation, formal analysis, writing–review and editing. S. Phillips: Data curation, The costs of publication of this article were defrayed in part by the payment of page
formal analysis, writing–review and editing. J.H. Carmichael: Data curation, formal charges. This article must therefore be hereby marked advertisement in accordance
analysis, writing–review and editing. N. Sharma: Data curation, formal analysis, with 18 U.S.C. Section 1734 solely to indicate this fact.
writing–review and editing. J. Jarnagin: Data curation, formal analysis, writing–
review and editing. I. Baiev: Data curation, formal analysis, writing–review and Received February 1, 2021; revised March 23, 2021; accepted April 26, 2021;
editing. Y.S. Shah: Data curation, formal analysis, writing–review and editing. published first April 29, 2021.

References
1. Siegel RL, Miller KD, Goding Sauer A, Fedewa SA, Butterly LF, Anderson JC, 8. Tie J, Cohen JD, Wang Y, Christie M, Simons K, Lee M, et al. Circulating tumor
et al. Colorectal cancer statistics, 2020. CA Cancer J Clin 2020;70:145–64. DNA analyses as markers of recurrence risk and benefit of adjuvant therapy for
2. Chakrabarti S, Peterson CY, Sriram D, Mahipal A. Early stage colon cancer: stage III colon cancer. JAMA Oncol 2019;5:1710–7.
Current treatment standards, evolving paradigms, and future directions. World J 9. Tie J, Wang Y, Tomasetti C, Li L, Springer S, Kinde I, et al. Circulating tumor
Gastrointest Oncol 2020;12:808–32. DNA analysis detects minimal residual disease and predicts recurrence in
3. Wan JCM, Massie C, Garcia-Corbacho J, Mouliere F, Brenton JD, Caldas C, et al. patients with stage II colon cancer. Sci Transl Med 2016;8:346ra92.
Liquid biopsies come of age: towards implementation of circulating tumour 10. Tie J, Cohen JD, Lo SN, Wang Y, Li L, Christie M, et al. Prognostic
DNA. Nat Rev Cancer 2017;17:223–38. significance of postsurgery circulating tumor DNA in nonmetastatic
4. Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, Agrawal N, et al. Detection colorectal cancer: individual patient pooled analysis of three cohort studies.
of circulating tumor DNA in early- and late-stage human malignancies. Int J Cancer 2021;148:1014–26.
Sci Transl Med 2014;6:224ra24. 11. Dasari A, Grothey A, Kopetz S. Circulating tumor DNA-defined minimal
5. Siravegna G, Marsoni S, Siena S, Bardelli A. Integrating liquid biopsies into the residual disease in solid tumors: opportunities to accelerate the development
management of cancer. Nat Rev Clin Oncol 2017;14:531–48. of adjuvant therapies. J Clin Oncol 2018;36:JCO2018789032.
6. Reinert T, Henriksen TV, Christensen E, Sharma S, Salari R, Sethi H, et al. 12. Coakley M, Garcia-Murillas I, Turner NC. Molecular residual disease and
Analysis of plasma cell-free DNA by ultradeep sequencing in patients with stages adjuvant trial design in solid tumors. Clin Cancer Res 2019;25:6026–34.
I to III colorectal cancer. JAMA Oncol 2019;5:1124–31. 13. U.S. National Library of Medicine. Circulating tumor DNA testing in predicting
7. Tarazona N, Gimeno-Valiente F, Gambardella V, Zu~ niga S, Rentero-Garrido P, treatment for patients with stage IIA colon cancer after surgery. Available from:
Huerta M, et al. Targeted next-generation sequencing of circulating-tumor DNA https://clinicaltrials.gov/ct2/show/NCT04068103.
for tracking minimal residual disease in localized colon cancer. Ann Oncol 2019; 14. Circulating tumor DNA based decision for adjuvant treatment in colon cancer
30:1804–12. stage II. Available from: https://clinicaltrials.gov/ct2/show/NCT04089631.

AACRJournals.org Clin Cancer Res; 27(20) October 15, 2021 5593

 Parikh et al.

15. DYNAMIC-II: Circulating tumour DNA (ctDNA) analysis informing adjuvant 26. Hu Y, Ulrich BC, Supplee J, Kuang Y, Lizotte PH, Feeney NB, et al. False-
chemotherapy in stage II colon cancer. Available from: https://trials.cancervic. positive plasma genotyping due to clonal hematopoiesis. Clin Cancer Res
org.au/details.aspx?ID=vctl_actrn12615000381583. 2018;24:4437–43.
16. DYNAMIC-III: Circulating tumour DNA analysis informing adjuvant 27. Zehir A, Benayed R, Shah RH, Syed A, Middha S, Kim HR, et al. Mutational
chemotherapy in stage III colon cancer: a multicentre phase II/III rando- landscape of metastatic cancer revealed from prospective clinical sequencing of
mised controlled study. Available from: https://gicancer.org.au/clinical-trial/ 10,000 patients. Nat Med 2017;23:703–13.
dynamic111/. 28. Nicholson BD, Shinkins B, Pathiraja I, Roberts NW, James TJ, Mallett S, et al.
17. Lonardi S, Montagut C, Pietrantonio F, Elez E, Sartore-Bianchi A, Tarazona N, Blood CEA levels for detecting recurrent colorectal cancer. Cochrane Database
et al. The PEGASUS trial: Post-surgical liquid biopsy-guided treatment of Syst Rev 2015;2015:Cd011134.
stage III and high-risk stage. II colon cancer patients J Clin Oncol 38:15s, 29. Lanman RB, Mortimer SA, Zill OA, Sebisanovic D, Lopez R, Blau S, et al.
2020 (suppl; abstr TPS4124). Analytical and clinical validation of a digital sequencing panel for quantitative,
18. BESPOKE study of ctDNA guided therapy in colorectal cancer. Available from: highly accurate evaluation of cell-free circulating tumor DNA. PLoS One 2015;
https://clinicaltrials.gov/ct2/show/NCT04264702. 10:e0140712.
19. Circulating tumor DNA analysis to optimize the operative and Postoperative 30. Aravanis AM, Lee M, Klausner RD. Next-generation sequencing of circu-
Treatment for Patients With Colorectal Cancer - Intervention Trial 2. Available lating tumor DNA for early cancer detection. Cell 2017;168:571–4.
from: https://clinicaltrials.gov/ct2/show/NCT03637686. 31. Cristiano S, Leal A, Phallen J, Fiksel J, Adleff V, Bruhm DC, et al. Genome-
20. Tracking mutations in cell free tumour DNA to predict relapse in early colorectal wide cell-free DNA fragmentation in patients with cancer. Nature 2019;570:
cancer. Available from: https://clinicaltrials.gov/ct2/show/NCT04050345. 385–9.
21. Identification and treatment of micrometastatic disease in stage III colon cancer. 32. Xie H, Mahoney DW, Foote PH, Burger K, Doering KA, Taylor WR, et al. Novel

Downloaded from http://aacrjournals.org/clincancerres/article-pdf/27/20/5586/3087069/5586.pdf by guest on 18 August 2022

Available from: https://clinicaltrials.gov/ct2/show/NCT03803553. methylated DNA markers in plasma detect distant recurrence of colorectal
22. Schøler LV, Reinert T, rntoft M-BW, Kassentoft CG, Arnad  ottir SS, Vang S, cancer. J Clin Oncol 38:15s, 2020 (suppl; abstr 4088).
et al. Clinical implications of monitoring circulating tumor DNA in patients with 33. Taylor WC. Comment on ‘Sensitive and specific multi-cancer detection and
colorectal cancer. Clin Cancer Res 2017;23:5437–5445. localization using methylation signatures in cell-free DNA’ by M. C. Liu et al.
23. Diehn M, Alizadeh AA, Adams H-P, Lee JJ, Klassen S, Palma JF, et al. Early Ann Oncol 2020;31:1266–7.
prediction of clinical outcomes in resected stage II and III colorectal cancer 34. Shen SY, Singhania R, Fehringer G, Chakravarthy A, Roehrl MHA, Chadwick D,
(CRC) through deep sequencing of circulating tumor DNA (ctDNA). J Clin et al. Sensitive tumour detection and classification using plasma cell-free DNA
Oncol 35:15s, 2017 (suppl; abstr 3591). methylomes. Nature 2018;563:579–83.
24. Abbosh C, Swanton C, Birkbak NJ. Clonal haematopoiesis: a source of biological 35. McDonald BR, Contente-Cuomo T, Sammut S-J, Odenheimer-Bergman A,
noise in cell-free DNA analyses. Ann Oncol 2019;30:358–9. Ernst B, Perdigones N, et al. Personalized circulating tumor DNA analysis to
25. Gibson CJ, Steensma DP. New insights from studies of clonal hematopoiesis. detect residual disease after neoadjuvant therapy in breast cancer. Sci Transl Med
Clin Cancer Res 2018;24:4633–42. 2019;11:eaax7392.

5594 Clin Cancer Res; 27(20) October 15, 2021 CLINICAL CANCER RESEARCH