Circulating Tumor DNA Dynamics Fail to Predict Efficacy of Poly(ADP-ribose) Polymerase/VEGFR Inhibition in Patients With Heavily Pretreated Advanced Solid Tumors

Abstract

PURPOSE

Cell-free circulating tumor DNA (ctDNA) has shown its potential as a quantitative biomarker for longitudinal monitoring of response to anticancer therapies. However, ctDNA dynamics have not been studied in patients with heavily pretreated, advanced solid tumors, for whom therapeutic responses can be weak. We investigated whether changes in ctDNA could predict clinical outcomes in such a cohort treated with combined poly(ADP-ribose) polymerase/vascular endothelial growth factor receptor inhibitor therapy.

MATERIALS AND METHODS

Patients with metastatic pancreatic ductal adenocarcinoma (PDAC), triple-negative breast cancer (TNBC), small-cell lung cancer (SCLC), or non–small-cell lung cancer (NSCLC) received up to 7 days of cediranib 30 mg orally once daily monotherapy lead-in followed by addition of olaparib 200 mg orally twice daily. Patients had progressed on a median of three previous lines of therapy. Plasma samples were collected before and after cediranib monotherapy lead-in and on combination therapy at 7 days, 28 days, and every 28 days thereafter. ctDNA was quantified from plasma samples using a multigene mutation–based assay. Radiographic assessment was performed every 8 weeks.

RESULTS

ctDNA measurements were evaluable in 63 patients. The median baseline ctDNA variant allele fractions (VAFs) were 20%, 28%, 27%, and 34% for PDAC, TNBC, SCLC, and NSCLC, respectively. No association was observed between baseline VAF and radiographic response, progression-free survival, or overall survival (OS). Similarly, no association was found between ctDNA decline and radiographic response or survival. However, an increase in ctDNA at 56 days of combination therapy was associated with disease progression and inferior OS in a landmark analysis.

CONCLUSION

ctDNA levels or dynamics did not correlate with radiographic response or survival outcomes in patients with advanced metastatic malignancies treated with olaparib and cediranib.

ctDNA may have limited ability to monitor treatment response in patients with heavily pre-treated malignancies

INTRODUCTION

Recent rapid developments in precision oncology have catalyzed paradigm-shifting changes in the treatment of many types of cancers. Because many therapies are only effective for subsets of patients, there is a critical need for biomarkers that allow minimally invasive, accurate, and timely evaluation of treatment responses. These biomarkers should be able to supplement information gained from radiographic assessment to predict therapeutic benefit, ideally earlier than is possible with conventional imaging. Such biomarkers should also be able to track disease progression so that an alternative therapy can be implemented as early as possible to deliver the optimal disease control and survival benefit to patients.^1,2

CONTEXT

• Key Objective

• Longitudinal changes in circulating tumor DNA (ctDNA) levels during therapy have been shown in numerous studies to be strongly predictive of radiographic response and survival outcomes. However, it is not known if ctDNA is similarly effective at predicting treatment efficacy in patients with very advanced cancer that has already progressed on multiple lines of therapy.

• Knowledge Generated

• We evaluated longitudinal ctDNA trends in patients with metastatic breast, pancreatic, or lung cancer who were treated on a clinical trial of poly(ADP-ribose) polymerase and vascular endothelial growth factor inhibitor therapy after having progressed through a median of three previous lines of therapy. In contrast to previous studies focusing on earlier lines of therapy, the present study did not find an association between ctDNA decline and radiographic response or survival outcomes in this heavily pretreated cohort.

• Relevance

• ctDNA dynamics might have limited ability to predict therapeutic efficacy in patients with heavily pretreated metastatic malignancies.

Cell-free circulating tumor DNA (ctDNA) has emerged as a promising biomarker that could fulfill these requirements. The utility of ctDNA for noninvasive tumor mutation profiling to guide therapy selection is well-established, and ctDNA tests are now commonly used for this purpose in clinical practice.^3-6 There is also now substantial evidence supporting the use of ctDNA as a longitudinal biomarker for monitoring of therapeutic responses and disease progression.^7-11 Indeed, ctDNA changes within 4-8 weeks of treatment have been shown to predict radiographic responses and clinical outcomes in a number of studies, including patients with non–small-cell lung cancer (NSCLC)^12,13 and urothelial,¹⁴ prostate,¹⁵ breast,³ and colorectal cancers.⁹ These observations were made in studies of immunotherapy, surgery, chemotherapy, and/or targeted therapies, including poly(ADP-ribose) polymerase (PARP) inhibitors.^15-17 However, ctDNA dynamics can be variable for different types of cancers and in different clinical scenarios. In particular, little is known about how ctDNA performs as a longitudinal biomarker for assessment of response to later lines of therapy in patients with heavily pretreated advanced cancers.

In the current study, we used a multigene mutation assay to measure ctDNA levels before and during therapy in patients treated with olaparib (a PARP inhibitor) in combination with cediranib (a vascular endothelial growth factor receptor [VEGFR] inhibitor)¹⁸ as part of a phase II clinical trial (NCI-9881, ClinicalTrials.gov identifier: NCT02498613). The trial enrolled patients with advanced small-cell lung cancer, non–small cell lung cancer, pancreatic ductal adenocarcinoma (PDAC), or triple-negative breast cancer (TNBC) whose disease was progressing after having received a median of three previous lines of therapy. We evaluated whether ctDNA dynamics could predict radiographic response, disease progression, progression-free survival (PFS), and overall survival (OS).

MATERIALS AND METHODS

Patients and Plasma Samples

NCI 9881 was a multicenter phase II clinical trial in which patients were treated with approximately 7 days of cediranib 30 mg orally once daily monotherapy lead-in, followed by addition of olaparib 300 mg orally twice daily. Serial blood samples were collected at the following timepoints: (1) at baseline before cediranib lead-in, (2) 4-7 days after cediranib lead-in and before starting olaparib, (3) 7 days (ie, C1D15) after olaparib was added to cediranib, (4) C2D1 (study day 36), and (5) every 28 days thereafter (Fig 1A). Up to 10 ml of blood was collected at each time point in EDTA tubes. Plasma was separated by centrifugation at 1,000×g for 10 minutes, which was performed within 4 hours of blood draw and then stored at –80°C. Blood sample collection and ctDNA purification/analyses were performed independently by research personnel who were blinded to clinical data, and treating physicians were not aware of ctDNA data. Informed written consent for the study was obtained from all patients as part of the consent to participate in the NCI-9881 trial. The study was reviewed and approved by the Human Investigation Committee of Yale University and was conducted according to established ethical guidelines outlined in the Declaration of Helsinki.

FIG 1.

(A) Schema of the trial design and treatment plan. Cediranib was given on C1D1, followed by adding olaparib on C1D4-7, and the combined treatment continued for at least 28 days. Therefore, cycle 1 had a total of 35 treatment days. Treatment was then continued with both cediranib and olaparib on a 28-day cycle. Red arrows and teal arrows indicate the scheduled times when imaging scans were performed and blood samples were collected for cfDNA analysis, respectively. cfDNA, cell-free DNA; NSCLC, non–small-cell lung cancer; PDAC, pancreatic ductal adenocarcinoma; SCLC, small-cell lung cancer; TNBC, triple-negative breast cancer. (B) A consort diagram showing the distributions of enrolled patients in each type of cancer and the different stages of ctDNA analysis. cfDNA, cell-free DNA; NSCLC, non–small-cell lung cancer; PDAC, pancreatic ductal adenocarcinoma; SCLC, small-cell lung cancer; TNBC, triple-negative breast cancer.

Cell-Free DNA Isolation and Measurement of Mutations

Cell-free DNA (cfDNA) was extracted from 1 mL of plasma using a QIAamp Circulating Nucleic Acid kit or a QIAamp MinElute Virus Vacuum kit (QIAGEN, Santa Clarita, CA) and processed according to the manufacturer's protocol. DNA was eluted in 25 μL. Tumor-derived somatic mutations within cfDNA were identified and quantified using the error-suppressed deep sequencing method previously published by our group.¹³ The assay simultaneously queries point mutations and insertions or deletions within 43 mutation-prone regions of 24 cancer-associated genes (Data Supplement, Table S1). The targeted amplicon library was subjected to next-generation sequencing on an Illumina HiSeq 2500 instrument (Illumina, San Diego, CA) in 75 base-pair, paired-end mode. Analysis of the sequence data, which includes suppression of background polymerase chain reaction-amplification errors and sequencer errors to enhance the sensitivity and accuracy of mutation calling, has been described previously.¹³ Tumor tissue mutation testing was performed in an independent laboratory using the BROCA-HR next-generation sequencing assay, which uses a hybrid capture panel which targets mutations, gene rearrangements, copy number variations that are pertinent to the Fanconi Anemia-BRCA HR, non-homologous end joining, and DNA mismatch repair pathways.¹⁹

Radiographic Assessment and Clinical Outcomes

Patients underwent restaging scans with computed tomographic (CT) scans or magnetic resonance imaging (MRI) of abdomen and pelvis and CT of chest within 31 days before treatment initiation, after 8 weeks of combined cediranib and olaparib treatment, then every 8 weeks during the first 24 weeks of treatment, and every 12 weeks thereafter. Radiographic response to therapy was assessed on the basis of imaging studies including CT scans, positron emission tomography/CT, and MRI according to the RECIST, version 1.1²⁰ and was performed by clinical trial investigators. Treatment duration was determined on the basis of the off-treatment date designated by the treating investigators. PFS was defined as the interval between treatment initiation and the date of disease progression or death, whichever occurred earlier. OS was defined as the time interval from treatment initiation to death.

Statistical Analysis

Comparisons of ctDNA variant allele fractions (VAFs) between groups were performed using one-way ANOVA. Agreement between radiographic response and ctDNA response was assessed with Cohen's kappa coefficient.²¹ Survival analyses were performed using the Kaplan-Meier analysis and the Cox regression. Landmark analysis was used to examine the relationship between early ctDNA changes at C1D15 (day 15), C2D1 (day 36), and C3D1 (day 64) and clinical outcomes. In each landmark analysis, a fixed time after the baseline was selected as a landmark for conducting the Kaplan-Meier survival analysis. Only patients surviving at the landmark time were included in the analysis, and patients were divided into two groups of either early ctDNA increase (ie, increase of ctDNA VAF, ie, ≥50% from baseline to landmark time) or ctDNA decrease/stable (ie, increase of ctDNA VAF by <50% from baseline or any decrease from baseline to landmark time). Statistical analyses were performed using SAS 9.4 (Cary, NC) or GraphPad Prism 9.2 (San Diego, CA). All P values were based on two-sided testing, and differences were considered statistically significant at P < .05.

RESULTS

Baseline ctDNA Levels Do Not Correlate With Radiographic or Clinical Outcomes

This study included data from 63 patients whose cfDNA samples and radiographic assessment of treatment responses were available at the time of analysis. The baseline patient characteristics are shown in Table 1. Of these 63 patients, three had a tumor mutation allele fraction of <0.5% of total cfDNA at baseline (pretreatment) and were excluded from further analysis (Fig 1B). Of the remaining 60 patients, the median baseline tumor DNA VAF was 28.5% (IQR, 10.7%-36.6%). The median baseline VAFs for each disease-specific cohort categorized by radiographic response are listed in the Data Supplement (Table S2).

TABLE 1.

Baseline Characteristics and Clinical Outcomes of the 63 Patients With RECIST Radiographic Assessment and Circulating Tumor DNA Data

Cohort	No.	Age, Years, Median (range)	Female, No. (%)	Previous Lines of Systemic Tx, No., Median (range)	Best Overall Response	Median PFS: Days	Median OS: Days
PDAC	11	64 (45-80)	4 (36.4%)	3 (2-6)	PD: 6 SD: 5 PR: 0	PD: 52 SD: 143 PR: -	PD: 106 SD: 220 PR: -
TNBC	24	49 (32-68)	24 (100%)	4 (2-10)	PD: 6 SD: 14 PR: 4	PD: 55 SD: 113 PR: 186	PD: 84 SD: 247 PR: 653
SCLC	13	67 (50-79)	5 (38.5)	2 (1-4)	PD: 3 SD: 6 PR: 4	PD: 61 SD: 173 PR: 175	PD: 100 SD: 314 PR: 455
NSCLC	15	66 (47-78)	5 (33.3)	4 (1-7)	PD: 1 SD: 13 PR: 1	PD: 59 SD: 118 PR: 226	PD: - SD: 345 PR: 249
Overall	63	61 (32-80)	37 (61.7)	3 (1-10)	PD: 15 SD: 39 PR: 9	PD: 56 SD: 118 PR: 179	PD: 99 SD: 269 PR: 486

Abbreviations: NSCLC, non–small-cell lung cancer; PDAC, pancreatic ductal adenocarcinoma; PD, progressive disease; PFS, progression-free survival; PR, partial response; SCLC, small-cell lung cancer; SD, stable disease; TNBC, triple-negative breast cancer.

We evaluated whether there was an association between baseline VAF and radiographic response. As shown in Figure 2A, the distribution of baseline VAFs did not appear to be correlated with radiographic Best Overall Response (BOR) in the full group of 60 patients. When the baseline VAFs were compared between patients who had radiographic partial response (PR), stable disease (SD), or progressive disease (PD) in each cohort, no statistically significant association was found (one-way ANOVA, P = .4189).

FIG 2.

Baseline ctDNA VAF versus radiographic BOR or patient clinical outcomes. (A) A bar graph showing the level of baseline ctDNA VAF in each patient. The radiographic BOR of each case is indicated by the color of each bar. The type of cancer of each case is labeled with a color-coded dot below each bar. The horizontal line indicates a level of ctDNA VAF at 10%. Kaplan-Meier curves showing (B) PFS and (C) OS of patients with baseline ctDNA VAF equal to/above the median versus those with baseline ctDNA VAF below the median. The PFS and OS in this analysis were measured from the beginning of treatment. The numbers of patients at risk with each category at specified time points are listed under each graph. BOR, best overall response; ctDNA, circulating tumor DNA; NSCLC, non–small-cell lung cancer; OS, overall survival; PD, progressive disease; PDAC, pancreatic ductal adenocarcinoma; PFS, progression-free survival; PR, partial response, SCLC, small-cell lung cancer; SD, stable disease; TNBC, triple-negative breast cancer; VAF, variant allele fraction.

Next, we assessed whether baseline VAFs were associated with clinical outcomes. Previous studies have shown that higher baseline VAF can be correlated with higher baseline disease burden and predict poorer clinical outcomes.^22-24 We separated patients into two groups on the basis of their baseline VAF measurement, using the median value as the cut off point. We found that there was no significant difference in PFS or OS between these two groups (Figs 2B and 2C). A similar analysis of each of the four separate disease cohorts failed to show a significant difference in PFS or OS on the basis of baseline VAF value, but the analyses were inconclusive because of the limited number of patients in each cohort (data not shown).

Early Change in ctDNA Level After Initiating Therapy Does Not Robustly Predict Radiographic Response

Early ctDNA changes have previously been shown to be predictive of radiographic disease response in studies of immune checkpoint inhibitors.^12,13,22,23 We asked whether early ctDNA trends in our study population could similarly predict radiographic outcomes. Similar to previous studies,^13,25,26 we categorized patients who had a >50% reduction in VAF from baseline as having a ctDNA response and those who had an increase in VAF by >50% from baseline as having ctDNA progression. Patients with VAF changes between –50% and 50% from baseline were categorized as having stable ctDNA.

We used Cohen's kappa metric to evaluate the degree of agreement between radiographic assessment and early ctDNA changes.²⁷ In this analysis, we defined ctDNA VAF measured at C3D1 (on treatment day 63) compared with baseline as early ctDNA changes. Of note, only 45 patients with available ctDNA VAF at C3D1 and corresponding radiographic assessment were included in this analysis (Fig 3).

FIG 3.

Treatment course of 45 patients with radiographic and ctDNA responses. Patients are grouped on the basis of their respective radiographic BOR as PR, SD, or PD cases. Case numbers are listed on the y-axis. The time in days from the beginning of treatment is shown on the x-axis. The durations of treatment in each case are plotted as light gray–colored horizontal bars. Within each bar, the horizontal line on the top half of each bar indicates radiographic responses, with the hash marks denoting the time points when imaging scans were performed. The colors of the lines and hash marks indicate the results of radiographic assessment (the red double hash marks label new lesions on scans). The horizontal line on the bottom half of each bar indicates ctDNA responses, with the filled circles denoting the time points when cfDNA samples were collected. The colors of the lines and filled circles indicate the results of ctDNA analysis. The black diamonds indicate the time of known death events when the data were analyzed. C1D15 and C3D1 time points are marked by two dotted vertical lines across the bars. BOR, best overall response; cfDNA, cell free DNA; ctDNA, circulating tumor DNA; PD, progressive disease; PR, partial response, SD, stable disease.

We first divided the cases into three categories of progression, stable, or response on the basis of either radiographic assessment or early ctDNA changes (Data Supplement, Table S3). The calculated kappa was 0.306 (95% CI, 0.098 to 0.514), which is consistent with a fair or slight agreement (Data Supplement, Table S4).

We also grouped the cases into two categories: progression versus stable/response, again on the basis of either radiographic or ctDNA assessment (Data Supplement, Table S5). We reasoned that in this heavily pretreated population where robust treatment responses are less common, clinical benefit may be better captured by combined analysis of patients with SD/response. With this approach, the calculated Cohen's kappa was 0.461 (95% CI, 0.192 to 0.730), which implies a moderate agreement between ctDNA and radiographic assessment (Data Supplement, Table S6).

Longitudinal Trends in Radiographic and ctDNA Measurements Are Heterogeneous

The clinical courses of 45 patients with available longitudinal ctDNA data and corresponding radiographic assessment are summarized in a swimmer's plot in Figure 3. The patients are grouped by their radiographic BOR.

Among patients with radiographic PD as their radiographic BOR (n = 14), 10 patients had ctDNA progression that preceded or occurred concurrently with radiographic disease progression (Fig 3 and Data Supplement, Fig S1). In comparison, among patients with radiographic SD (n = 25; Fig 3 and Data Supplement, Fig S2), only four (cases 012, 091, 080, and 085) had ctDNA progression before or around C3D1 (16.0%). The remaining 21 patients with radiographic SD had either stable (n = 13, 52.0%) or decreased ctDNA (n = 8, 32.0%) by C3D1 (Fig 3 and Data Supplement, Fig S2). For patients with radiographic PR (n = 6; Figure 3 and Data Supplement, Fig S3), there were three cases (056, 042, and 046) that had early ctDNA progression (50.0%). In all three cases, new lesions were found on scans at the time of progression despite the continuous decrease in the volume of existing measurable lesions (Fig 3 and Data Supplement, Figs S3A-S3C). Of the remaining PR cases, two patients had decreased ctDNA more than 50% from baseline by C3D1 (cases 065 and 095) and one had stable ctDNA (case 050; Fig 3 and Data Supplement, Fig S3D-S3F). The longitudinal ctDNA VAF measurements and radiographic sum of target lesion diameters for individual patients are presented in detail in the Data Supplement (Figs S1-S3), further illustrating the significant heterogeneity among the cases.

Early ctDNA Increase Was Associated With Inferior OS Outcomes

We next asked whether an increase in ctDNA levels within the first 2 months of initiating treatment could predict survival outcomes in our study. We adopted a landmark analysis approach for this analysis.

At a landmark of C3D1 (on treatment day 63), a significantly worse OS was observed for patients with a ctDNA increase (median OS: 64 days) when compared with those with stable or decreased ctDNA (median OS, 86 days; with Kaplan-Meier P = .0166). But a significant difference was not found in PFS between patients with an early ctDNA increase (median PFS, 61.5 days) and those with stable or decreased ctDNA (median PFS, 61 days; with Kaplan-Meier P = .3339; Fig 4 and Data Supplement, Table S7).

FIG 4.

Correlation between early ctDNA changes and clinical outcomes in a landmark analysis. Kaplan-Meier curves showing (A) PFS and (B) OS. The PFS and OS in this analysis were measured from the designated landmark (C3D1). The numbers of patients at risk with each category at specified time points are listed under each graph. ctDNA, circulating tumor DNA; OS, overall survival; PFS, progression-free survival.

At landmarks of C1D15 (day 15) and C2D1 (day 35), no significant difference was observed between patients with the ctDNA increase and those with stable or decreased ctDNA, in terms of PFS or OS (Data Supplement, Fig S4 and Table S7). The median PFS for patients with the early ctDNA increase at the landmark of C1D15 was 74.5 days, and that for those with stable or decreased DNA was 95 days; the Kaplan-Meier P value for the difference between two groups was .9874. The median OS was 131 days and 98.5 days for those with the early ctDNA increase and those with stable or decreased ctDNA, respectively (with Kaplan-Meier P = .8384). Similarly, at C2D1, the median PFS was 88 days versus 85 days (with Kaplan-Meier P = .6594) and the median OS was 132.5 days versus 95 days (with Kaplan-Meier P = .1863). Of note, for both landmarks at C1D15 and C2D1, fewer than 10 patients had an early ctDNA increase, which reduces the reliability of such an analysis.

Clinical Disease Progression in Patients With Initial Radiographic PR or SD is Not Consistently Accompanied by Rising ctDNA Levels

We investigated whether a rising ctDNA level could predict radiographic disease progression among patients whose tumors initially shrank after initiating therapy. We examined 12 patients who had reduction in measurable lesions during the earlier course of treatment and plotted the relative changes of the existing measurable lesions over time (Fig 5A). Ten of these patients had a sustained reduction in size of existing measurable lesions but eventually had disease progression because of the emergence of new lesions. The other two patients had progression of existing lesions eventually. However, when their ctDNA trends were similarly plotted over time, no clear pattern was discernible (Fig 5B).

FIG 5.

Longitudinal radiographic measurements and ctDNA VAF measurements over the course of treatment. Twelve patients with any reduction in size of existing lesions at the first radiographic assessment (C3D1), including those designated as SD by the RECIST criteria, are shown in the graphs. The percentage change over time from baseline level is indicated for (A) radiographic measurements and (B) ctDNA VAF measurements. Cases with eventual progression on the basis of emergence of new lesions are indicated with solid lines, and cases that progressed on the basis of increasing size of existing lesions without emergence of new lesions are indicated with dashed lines. ctDNA, circulating tumor DNA; VAF, variant allele fraction.

DISCUSSION

In this study, we evaluated longitudinal changes in ctDNA levels in patients with advanced solid tumors treated with cediranib and olaparib. Most patients in the study had very advanced disease and had already experienced disease progression on multiple previous lines of cancer therapy (median three previous lines). The high tumor burden among the patients is reflected in the high concentration of tumor-derived mutant DNA in the baseline plasma samples, with a median VAF of 28.5%. In these patients with TNBC, PDAC, small cell lung cancer (SCLC), and NSCLC treated with a PARP inhibitor–based combination therapy, we found that baseline ctDNA levels did not correlate with radiographic or clinical outcomes. We also found that in this heavily pretreated population, an early decline in ctDNA level after initiating therapy did not robustly predict radiographic response or survival. This contrasts with several independent studies (generally in less heavily pretreated populations) that have shown that an early drop in ctDNA level is predictive of radiographic response and/or survival.^12,13,22-24 Using a landmark approach, we did find that an increase in ctDNA of 50% or more after two cycles of treatment was associated with worse OS (but not PFS). Earlier landmark timepoints of C1D15 or C2D1 did not show a statistically significant difference in OS or PFS between patients with versus without a 50% or greater rise in ctDNA level.

Our observation that baseline ctDNA levels or an early drop in ctDNA levels was not predictive of radiographic or clinical outcomes may be potentially explained by the exceptionally poor prognosis of most patients in this study. In this patient population, the baseline ctDNA VAFs were much higher than those that have been reported in previous studies of ctDNA-based treatment monitoring (the average VAF is typically <10% at baseline in most cohorts with metastatic disease).^22,23 As shown in Figure 2A, 46 of the 60 patients (76.7%) had baseline mutant ctDNA VAFs >10%, likely reflecting a high tumor burden and high rate of cancer cell death causing more shedding of tumor DNA into the bloodstream. At such late stages of disease where tumor burden and ctDNA levels are very high, baseline ctDNA concentrations may be less effective at predicting relatively small differences in generally poor survival outcomes. Consistent with our finding, in the TOPARP-A trial, there was also no correlation between the baseline ctDNA level and clinical responses to olaparib in metastatic prostate cancer.¹⁵ Furthermore, treatment responses to PARP and vascular endothelial growth factor (VEGF) inhibition are expected to be relatively modest and short-lived in patients who had already progressed on multiple lines of systemic therapy, and who were not preselected for having a known mutation conferring DNA damage repair deficiency. In the setting of such modest therapeutic responses, a clear association between longitudinal ctDNA change and treatment efficacy can be more difficult to discern. A post hoc analysis of somatic alterations in tumor tissue using the BROCA-HR next-generation sequencing assay,¹⁹ which uses a hybrid capture panel that targets mutations, gene rearrangements, and copy number variations that are pertinent to the Fanconi Anemia-BRCA HR, nonhomologous end joining, and DNA mismatch repair pathways, will be reported in a separate publication.

Our study has several limitations. Each cohort (NSCLC, SCLC, TNBC, and PDAC) consisted of a relatively small number of patients, limiting the statistical power of our analyses. Only a few patients exhibited a radiographic PR to therapy, further limiting the power to detect a difference in likelihood of PR between patients with versus without a ctDNA decline. Another limitation of our study is that our ctDNA assay was designed to probe a set of mutation-prone regions in a predefined panel of cancer-associated genes. Such an approach is unlikely to detect mixed responses among metastatic tumor subclones, some of which might have acquired new mutations that confer resistance to therapy while other subclones are responding. Because of the heterogeneity of potential resistance mechanisms to PARP/VEGF inhibition among four cancer types and the predefined coverage of our ctDNA mutation panel, we were unable to track discordant (mixed) responses among different tumor subclones in the same patient. Tracking several mutations simultaneously could provide a more accurate overall assessment of tumor burden, especially in scenarios where some mutations are lost and new mutations arise in subclonal tumor cell populations. Such mutation losses or gains could lead to inaccurate estimation of tumor-derived cfDNA levels. However, even if a cfDNA assay measured multiple mutations (or multiple epigenetic features), it would still be unable to account for variable shedding of DNA from different metastatic sites because of the aggregated nature of the cfDNA signal, which may contribute to a discrepancy between radiographic assessment and ctDNA measurement. This is a general limitation of ctDNA-based assessments of therapy response. Finally, this was a retrospective analysis of prospectively collected plasma samples, which should be considered as hypothesis-generating. The study's findings should be validated in an independent cohort.

In conclusion, in this study of patients with heavily pretreated, advanced solid tumors treated with a combination of PARP and VEGFR inhibitors, we found that baseline ctDNA levels or longitudinal changes in ctDNA levels could not robustly predict therapeutic efficacy. These results suggest that ctDNA may be a less reliable marker of response in patients with advanced malignancies that have progressed on several previous lines of therapy and in whom vigorous and durable treatment responses are not generally expected.

DISCLAIMER

The funding sources did not have any role in the decision making, study design, data collection, data interpretation, and writing of this text.

EQUAL CONTRIBUTION

Y.H. and A.N. contributed equally to this work.

PRIOR PRESENTATION

Presented in part at the annual meeting of the American Society of Clinical Oncology in Chicago, IL, June 4-8, 2021.

SUPPORT

Supported by the following grants: UM01-CA186689 to P.L., A.A.P., and J.W.K. and U01-CA233364 and R01-CA197486 from the National Cancer Institute (NIH) to A.A.P.

AUTHOR CONTRIBUTIONS

Conception and design: Yiduo Hu, Azeet Narayan, Thi Trinh, S. Percy Ivy, Joseph Paul Eder, Patricia LoRusso, Joseph W. Kim, Abhijit A. Patel

Financial support: Patricia LoRusso, Abhijit A. Patel

Administrative support: Patricia LoRusso

Provision of study materials or patients: Yunshan Xu, S. Percy Ivy, Joseph Paul Eder, Patricia LoRusso, Abhijit A. Patel

Collection and assembly of data: Yiduo Hu, Azeet Narayan, Julia Wolfe, Dennis Vu, Joseph Paul Eder, Joseph W. Kim, Abhijit A. Patel

Data analysis and interpretation: Yiduo Hu, Azeet Narayan, Yunshan Xu, Chaitanya Kantak, S. Percy Ivy, Joseph Paul Eder, Yanhong Deng, Patricia LoRusso, Joseph W. Kim, Abhijit A. Patel

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).