Identifying Mechanisms of Resistance by Circulating Tumor DNA in EVOLVE, a Phase II Trial of Cediranib Plus Olaparib for Ovarian Cancer at Time of PARP Inhibitor Progression

Abstract

Purpose: To evaluate the use of blood cell-free DNA (cfDNA) to identify emerging mechanisms of resistance to PARP inhibitors (PARPi) in high-grade serous ovarian cancer (HGSOC).

Experimental design: We used targeted sequencing (TS) to analyze 78 longitudinal cfDNA samples collected from 30 patients with HGSOC enrolled in a phase II clinical trial evaluating cediranib (VEGF inhibitor) plus olaparib (PARPi) after progression on PARPi alone. cfDNA was collected at baseline, before treatment cycle 2, and at end of treatment. These were compared with whole-exome sequencing (WES) of baseline tumor tissues.

Results: At baseline (time of initial PARPi progression), cfDNA tumor fractions were 0.2% to 67% (median, 3.25%), and patients with high ctDNA levels (>15%) had a higher tumor burden (sum of target lesions; P = 0.043). Across all timepoints, cfDNA detected 74.4% of mutations known from prior tumor WES, including three of five expected BRCA1/2 reversion mutations. In addition, cfDNA identified 10 novel mutations not detected by WES, including seven TP53 mutations annotated as pathogenic by ClinVar. cfDNA fragmentation analysis attributed five of these novel TP53 mutations to clonal hematopoiesis of indeterminate potential (CHIP). At baseline, samples with significant differences in mutant fragment size distribution had shorter time to progression (P = 0.001).

Conclusions: Longitudinal testing of cfDNA by TS provides a noninvasive tool for detection of tumor-derived mutations and mechanisms of PARPi resistance that may aid in directing patients to appropriate therapeutic strategies. With cfDNA fragmentation analyses, CHIP was identified in several patients and warrants further investigation.

Figure 1.

Summary of patients with HGSOC. Summary of patient timelines; diamonds indicate time points at which samples were collected; empty points indicate the sample failed to provide sufficient cfDNA for sequencing, whereas each X indicates patients for which cfDNA was not obtained.

Figure 2.

Estimated tumor content of cfDNA. A, Estimated tumor-derived DNA fraction in cfDNA at each timepoint, based on allele fraction of known TP53 mutations (or maximum VAF of high-confidence somatic variants); estimates for three patients could not be obtained, as their tumors had no known TP53 mutations, and no somatic variants were called in the cfDNA. For each patient, points are colored according to best RECIST1.1 response. B, Change in estimated tumor fraction of circulating DNA between baseline and on-treatment samples (ΔctDNA); calculated as the relative percent difference between baseline and cycle 2 of treatment and was available for 18 of 30 patients. EOT, end of treatment.

Figure 3.

Fragmentation profiles of cfDNA during treatment. A, Distribution of fragment size across all patient-derived and healthy control cfDNA samples. Each line represents the density (y-axis) of fragments at each size (x-axis; measured in nucleotide counts); each grey line represents a distinct patient-derived sample, whereas the red line shows the median profile across healthy controls. Insets show the increased frequency of short (100–150 bp) or long (250–320 bp) fragments in patient-derived samples. B, Patient-derived samples collected at baseline and time of progression show a higher proportion of short (<150 bp) fragments than healthy controls (two-sample Wilcoxon test: *, P < 0.1; **, P < 0.01; ***, P < 0.001), whereas samples collected at cycle 2 of treatment did not. C, Fragment size of mutation-containing reads is significantly different [frequently shorter (<150 bp) or longer (>230 bp), representing increased fragmentation beyond single- and dinucleosomal DNA sizes of 167 and 334 bp] than reads containing the reference allele at known somatic mutation sites (paired Wilcoxon rank-sum test). D, Kaplan–Meier curve showing disease-free survival for patients with a significant difference in mutation-specific fragmentation as compared with patients with no difference in mutation-specific fragmentation profiles [Cox proportional hazards test: HR, 4.8 (1.8–12.5), P = 0.001].

Figure 4.

Mutation summary of tumor-derived cfDNA. Somatic mutations were detected in TP53, BRCA1, BRCA2, and PALB2 using an ensemble approach, whereas CCNE1 amplifications were detected using panelCN.mops; symbols indicate mutation was expected on the basis of previous WES of the baseline tumor (filled = successfully detected; empty = not detected; diamonds represent previously identified reversion mutations). Top plot shows estimated ctDNA level for each sample with dashed lines to represent limits of detection for mutations (black) and copy-number changes (red).

Figure 5.

Fragment size distribution of tumor and nontumor variants. Fragment size distributions of reference- and variant-containing reads for (A) germline SNP from a cohort of 22 healthy controls as well as (B) known somatic TP53 mutations, (C) suspected CHIP TP53 mutations, and (D) novel somatic TP53 mutations from our patient cfDNA. Distributions for reference- and variant-containing reads were compared using a one-sided KS test [alternative hypothesis = the cumulative distribution function (CDF) of variant fragment size is above, and therefore consists of shorter fragments, than the CDF of reference reads].