EVOLVE: A Multicenter Open-Label Single-Arm Clinical and Translational Phase II Trial of Cediranib Plus Olaparib for Ovarian Cancer after PARP Inhibition Progression

This article is featured in Highlights of This Issue, p. 4169

Ovarian cancer is the second leading cause of gynecologic cancer death worldwide (1). Disease initially responds well to platinum and taxane chemotherapy, but recurs in most women diagnosed with stage III/IV ovarian cancer; 5-year survival rates are 42% and 29%, respectively (2). High-grade serous ovarian cancer (HGSOC), the most common type (2), is characterized by severe genomic instability, nearly universal TP53 mutations leading to dysfunctional p53, and defects in homologous recombination DNA repair pathways in half of the cases (3). The discovery that inhibiting PARP enzyme function in BRCA1/2-mutated cancers causes synthetic lethality (4) heralded precision targeted therapy in HGSOC. Three PARP inhibitors (PARPi; olaparib, niraparib, and rucaparib) are approved treatments for recurrent ovarian cancer (5–8), and recently olaparib was approved as first-line maintenance therapy for BRCA1/2-mutated disease (9). Clinical adoption of PARPis has significantly influenced HGSOC treatment, extending progression-free survival (PFS) while maintaining quality of life (10, 11). Contemporary investigations are moving PARPi usage earlier in the treatment paradigm, but currently there is no evidence that PARPi re-exposure is effective. The emergence of resistance raises the question of treatment options after progression on a PARPi. Several mechanisms of resistance have been described in the preclinical setting (12), but their clinical relevance has not been evaluated systematically.

Given the importance of angiogenesis in ovarian cancer (13), targeted combination treatments leveraging microenvironment modulation provide attractive opportunities to extend the benefit of PARPis. Cediranib, an oral inhibitor of vascular endothelial growth factor receptors 1–3 and c-kit tyrosine kinases, induces hypoxia and reduces angiogenesis (14). Hypoxia leads to aberrant DNA damage and repair signaling, which in turn causes genetic instability (15). Cediranib also suppresses the homologous recombination DNA repair pathway through downregulation of BRCA1/2 and RAD51 gene expression (16). In a randomized phase II trial, patients with measurable platinum-sensitive relapsed high-grade serous or endometrioid disease or with deleterious germline BRCA1/2 mutations were randomized to receive either olaparib capsules 400 mg twice a day or olaparib 200 mg twice a day combined with cediranib 30 mg daily. The combination regimen demonstrated significantly improved PFS, which was more pronounced in patients with no known BRCA mutation (17). The same effect was seen for overall survival (OS; ref. 18). However, initial reports of the randomized phase III GY004 trial (NCT02446600) indicate that the olaparib–cediranib combination therapy failed to improve PFS (primary endpoint) compared with platinum-based chemotherapy in platinum-sensitive recurrent ovarian cancer (19).

We hypothesized that adding cediranib to olaparib may overcome resistance after progression on a PARPi. EVOLVE is an investigator-initiated phase II study assessing the efficacy and safety of cediranib–olaparib in HGSOC (regardless of BRCA status) that has progressed on a PARPi. As this was designed as a hypothesis-generating study, paired (pre- and post-PARPi) tumor samples were mandatory to identify mechanisms of PARPi resistance.

This phase II study (NCT02681237), performed at two centers in Canada and Spain, enrolled women with a histologically confirmed diagnosis of recurrent ovarian, fallopian tube, or primary peritoneal cancer with high-grade serous or high-grade endometrioid histology and radiographically documented disease progression on any PARPi. Patients were required to have disease evaluable by RECIST v1.1 and amenable to a baseline biopsy. Key exclusion criteria included: active bowel obstruction, untreated or unstable central nervous system (CNS) metastases, any active significant cardiovascular event, >1+ proteinuria, and myelodysplastic syndrome/acute myeloid leukemia. Women requiring maximal doses of calcium channel blockers to stabilize blood pressure and those who were unable to discontinue CYP3A4 inhibitors and inducers were excluded. There was no limit to the number of prior treatment lines.

The study was approved by each participating site's institutional review board or ethics committee. All patients provided written informed consent according to the Declaration of Helsinki and Good Clinical Practice.

Patients were enrolled into one of three cohorts according to platinum sensitivity status as defined by the Gynecologic Cancer InterGroup (11): platinum sensitive, platinum resistant, or an exploratory cohort including patients whose disease had progressed on a PARPi and progressed again on subsequent standard chemotherapy, regardless of platinum sensitivity. Oral olaparib 300 mg tablets were administered twice a day and oral cediranib 20 mg once daily, repeated every 28 days. The cediranib dose of 20 mg has been used in other trials, such as the randomized phase II BAROCCO trial (20), and was chosen in the present study based on the safety profile, particularly at the time of PARPi rechallenge. One dose reduction of cediranib to 15 mg was suggested in the event of intolerable cediranib-related toxicities. The olaparib dose could be reduced to 250 mg twice a day, with a second dose reduction to 200 mg twice a day for the management of olaparib-related toxicities. Treatment was continued until disease progression, unacceptable toxicity, or withdrawal of informed consent, whichever occurred first.

Tumors were assessed according to RECIST v1.1 every 8 weeks until radiologic progression. Toxicity was graded according to National Cancer Institute Common Terminology Criteria for Adverse Events version 4.03. Provision of archival tissue (initial diagnostic tissue, before PARPi) and baseline biopsy (after PARPi) samples was mandatory; biopsy at the time of disease progression was optional.

The primary objective was to determine the efficacy of cediranib–olaparib in women whose ovarian cancer had progressed on prior PARPi therapy, assessed by objective response rate (RECIST v1.1) at 8 weeks and PFS rate at 16 weeks. We aimed to understand whether the addition of cediranib to olaparib overcomes resistance to PARPi and allows repeated response to PARPi. Secondary objectives were to evaluate disease control rate, safety, and mechanisms of resistance to PARPis.

DNA and RNA were coisolated using QIAGEN All-Prep DNA/RNA/miRNA Universal Co-Isolation kit from formalin-fixed paraffin-embedded tissues from archival (initial diagnostic sample before PARPi), baseline (at study entry after initial PARPi), and progression (on study after cediranib–olaparib) samples. Whole-exome sequencing, total RNA sequencing, and shallow whole-genome sequencing (s-WGS) libraries were constructed at the Princess Margaret Genomics Centre (www.pmgenomics.ca) using the methodology previously described by Lheureux and colleagues (21).

Sequencing reads were aligned to human genome reference (build hg38) using Burrows–Wheeler aligner (BWA-MEM) software (22). Somatic single-nucleotide variants were called using MuTect (version 1.1.5; ref. 23); insertions and deletions were called using VarScan2 (version 2.4.2; ref. 24). Variants were filtered to retain only nonsynonymous exonic changes present with a minor allele frequency >1% in 1000 Genomes phase 3 (release version 5.20130502), coverage >20×, and variant allele frequency >5%. Sequenza (version 2.1.0) and ichorCNA (github repo cloned May 21, 2019; https://github.com/broadinstitute/ichorCNA) were used for copy-number calling for exome and s-WGS data, respectively (25, 26). RNA sequencing reads were aligned using STAR (version 2.4.2a; ref. 27), and fusions were detected using STAR-Fusion (https://www.biorxiv.org/content/10.1101/120295v1). Gene expression analysis was performed using RSEM (version 1.3.0), and FPKM+0.01 values were log₂ transformed and quantile normalized. “Overexpression” was defined as a ≥4-fold increase in expression between archival and baseline samples.

This proof-of-concept study aimed to detect an initial efficacy signal from the cediranib–olaparib combination after PARPi failure. The target clinical benefit rate (complete response, partial response, or stable disease for ≥16 weeks) was 30% in the platinum-sensitive cohort and 10% in the platinum-resistant and exploratory cohorts, based on expectations with standard chemotherapy. We analyzed efficacy in the intention-to-treat population and separately by cohort (three cohorts of ≥10 patients with evaluable tissue samples). The 16-week PFS and 1-year OS rates were estimated using Kaplan–Meier methodology with corresponding 95% confidence intervals (CI). All statistical analyses were performed in R version 3.5.2 (28).

Between June 2016 and October 2018, 34 women were enrolled (platinum-sensitive cohort n = 11; platinum-resistant cohort n = 10; exploratory cohort n = 13). All patients had HGSOC, and 62% harbored somatic and/or germline BRCA1/2 mutations (Table 1). Patients had received a median of five prior treatment lines. Six patients (18%) had received two lines of the same prior PARPi (olaparib n = 4; niraparib n = 2), and 18% had received prior antiangiogenics (bevacizumab n = 5, cediranib n = 1).

At the data cutoff (June 24, 2019), the median duration of follow-up was 11.6 months; 31 patients had completed therapy and 3 remained on treatment (Fig. 1). The most common reason for study discontinuation was disease progression (n = 27); 5 patients were in follow-up at the data cutoff and withdrew consent for follow-up [one was being treated as standard of care outside the study site and did not want to travel for follow-up; the other withdrew during cycle 1 as she found the toxicities intolerable, although most adverse events (AE) were considered unrelated to study medications]. Objective response rates were 0% in the platinum-sensitive cohort, 20% in the platinum-resistant cohort, and 8% in the exploratory cohort (Table 2). Corresponding 16-week PFS rates were 55%, 50%, and 39%, respectively.

Treatment was well tolerated; AEs were as expected and manageable. The most common all-grade AEs were diarrhea (68%), nausea (53%), fatigue (44%), and vomiting (44%), typically grade 1/2 (Table 3). Overall, 38% of patients experienced grade 3 AEs, most commonly diarrhea (12%) and anemia (9%). Grade 3 thrombocytopenia and neutropenia were both absent. There was one grade 4 AE (myelodysplastic syndrome in a patient with germline BRCA1 mutation; prior treatments included two lines of platinum-based chemotherapy and maintenance olaparib for 21 months) and no grade 5 AEs. Supplementary Table S1 shows cediranib-related AEs.

Cediranib dose reductions and interruptions were implemented in 12% and 53% of patients, respectively. Corresponding percentages for olaparib were 15% and 47%. The median duration of dose interruption was 14.5 days for cediranib and 21 days for olaparib. There were no toxicity-related treatment discontinuations.

Exome data were generated from archival samples (before PARPi) for 33 patients, baseline samples (after progression on PARPi) for 32 patients, and from paired archival and baseline samples for 31 patients. This paired-sample cohort was used to interrogate acquired resistance mechanisms. Paired archival and baseline RNA sequencing data (n = 27) were used to examine changes in gene expression.

TP53 variants were detected in 30 of 33 archival samples (29/31 paired; Fig. 2), consistent with HGSOC histology. Pathological review by gynecologic pathology experts confirmed high-grade serous morphology for the TP53-negative cases. The TP53 locus was also reviewed manually to rule out false negatives. BRCA1/2 alterations were present in 21 patients (Table 1); of these, 20 harbored truncating BRCA1/2 variants and 5 (two platinum-sensitive, three platinum-resistant) had secondary reversion mutations in matched progression tumors (Fig. 2; Supplementary Fig. S1). Across a set of 17 homologous recombination deficiency genes covered by exome sequencing (ATM, ATR, EMSY, BRIP1, CHEK1, CHEK2, RAD50, RAD51C, RAD51D, BRIP1, BARD1, RAD50, FAM175A, NBN, PALB2, PTEN, and MRE11), we observed a single RAD51B mutation reversion whereby an 18 bp in-frame deletion detected in the archival specimen was absent in the baseline sample. Overall, patients with homologous recombination gene (BRCA1/2 or RAD51B) reversions showed markedly worse PFS (median 1.9 months) than those with mutated or wild-type homologous recombination genes (4.8 and 6.4 months, respectively; Fig. 3). Patients with BRCA1/2 reversions invariably stopped treatment early (within 2 months from initiation) because of disease progression (Fig. 4). We also observed BRCA1/2 amplification or overexpression in three baseline samples, suggesting functional restoration of homologous recombination deficiency; this was not associated with treatment response (Fig. 4). Furthermore, expression of other homologous recombination genes changed over time (Supplementary Fig. S2). Examination of other known intrinsic or acquired mechanisms of PARPi resistance (31–34) revealed CCNE1 amplification/overexpression in 11 of 31 paired cases (three in both archival and baseline, seven in baseline alone, one in archival alone; Fig. 2; Supplementary Fig. S3). We also reviewed loss of TP53BP1 function in BRCA1/2-mutant cases and perturbations in ABCB1, SLFN11, and DYNLL1 expression in baseline versus archival samples. Although no TP53BP1 and DYNLL1 loss was detected, ABCB1 was overexpressed in four baseline samples and amplified in one, whereas SLFN11 was significantly downregulated in two baseline samples.

Previously described or proposed mechanisms of PARPi resistance were detected in 55% of evaluable progression tumors in this study. Mechanisms were similarly distributed across all three cohorts (Fig. 2). Upregulation of multidrug efflux pump gene ABCB1 was indicative of worse PFS and OS (Fig. 3). Baseline samples were enriched with gene fusions versus archival samples (Supplementary Fig. S4), consistent with increased genome instability following treatment cycles administered before enrollment into EVOLVE.

To supplement the gene-centric approach, we used pathway enrichment methods to identify change over time in expression patterns. We focused on DNA repair and angiogenesis gene sets using gene set variation analysis to calculate an enrichment score for each sample. Comparison of archival and baseline samples revealed a significant increase in the angiogenic pathway during treatment, except in patients previously treated with antiangiogenic therapy (Supplementary Fig. S5). Progression biopsies from 2 patients showed upregulation of the angiogenesis pathway compared with the corresponding archival specimen.

Cediranib–olaparib combination therapy showed some activity after progression on prior PARPi in all three cohorts, despite heavy pretreatment in most patients. The combination was well tolerated, and the safety profile was consistent with previous findings (18). AEs were manageable with early intervention, using loperamide and antihypertensive therapy as required. Only 4 patients required a cediranib dose reduction and 5 required an olaparib dose reduction; no patients discontinued treatment because of toxicity. The observed tolerability of the combination may be related to the cediranib starting dose of 20 mg. In the phase II trial of cediranib–olaparib in PARPi-naïve ovarian cancer, patients received a lower olaparib dose (200 mg twice daily capsules) but a higher cediranib dose (30 mg once daily) than in our study (18). However, at this dose, the drug-related AEs were more common in the cediranib–olaparib arm than with olaparib alone; 70% of patients experienced a grade ≥3 AE, particularly diarrhea and fatigue (18). In addition, 77% of patients in the cediranib–olaparib treatment group required dose reductions compared with 24% receiving olaparib alone. In our trial evaluating PARPi rechallenge, we were cautious about potential AEs and used a cediranib dose of 20 mg, which demonstrated some activity in the recently reported BAROCCO trial (20).

One described mechanism of drug resistance involves reversion of the original truncating mutations in BRCA1/2 genes, restoring functional protein expression and the ability of cancer cells to repair DNA by homologous recombination (35, 36). Reversion mutations in BRCA1/2 and RAD51C/D are associated with primary and acquired resistance to the PARPi rucaparib (37, 38). In our study, 5 patients had BRCA1/2 reversion mutations at the time of progression on a PARPi, which correlated with poor outcome. However, based on our study design, we cannot discern whether this resistance mechanism was acquired during platinum treatment or de novo because of PARPi as the pre-PARPi sample was the archival tissue. In the ARIEL2 study, secondary somatic mutations restoring RAD51C and RAD51D were associated with acquired resistance to the PARPi rucaparib (37), but this was not detected in EVOLVE. Instead, we observed RAD51B reversion mutation in one post-PARPi progression sample. We hypothesize that during PARPi and/or platinum treatment, a RAD51B wild-type clone emerged selectively. Interestingly, RAD51C was significantly overexpressed in the baseline biopsy sample after progression on a PARPi (Supplementary Fig. S2). Preclinical evidence shows that RAD51C-deficient cancer cells are sensitive to PARPi (39), and loss of RAD51C promoter hypermethylation may lead to PARPi resistance (40). Recently, a study showed that homozygous or hemizygous BRCA1 methylation predicts rucaparib clinical response, and that loss of BRCA1 methylation after exposure to chemotherapy disables a therapeutic mechanism of response (41). Therefore, the RAD51C overexpression seen in our study may contribute to acquired PARPi resistance.

These mechanisms of acquired resistance in HGSOC with reversion mutations in the homologous recombination genes BRCA1/2 and RAD51 effectively restore homologous recombination DNA repair and confer clinical resistance to PARPis (37, 38, 42) and platinum (43). These mechanisms can be shared between platinum and PARPi resistance. Our study showed that these homologous recombination gene reversion mutations confer poor prognosis and cannot be overcome by adding cediranib to olaparib. This finding leads us to suggest incorporating detection of reversion mutations in homologous recombination DNA repair genes for patient selection or as a stratification factor for future post-PARPi clinical trials. With the successful incorporation of liquid biopsies into clinical practice, circulating tumor DNA analysis of reversion mutations and potentially promoter methylation of homologous recombination deficiency genes should be a valuable tool for disease monitoring and early management of emerging resistant clones.

Another acquired mechanism of resistance to PARPis involves upregulation of drug efflux pumps encoded by genes such as ABCB1 [multidrug resistance 1 (MDR1)]. Overexpression of MDR1 often leads to cross-resistance to structurally unrelated drugs (32). In our study, we observed four cases of ABCB1 overexpression (15% of 27 evaluable for RNA sequencing) and one case of DNA amplification that did not lead to transcript upregulation. With our technical methodology, we were unable to determine whether ABCB1 upregulation was a consequence of structural rearrangements, a recently described mechanism of ABCB1 upregulation in HGSOC (44, 45). Interestingly, 2 patients with ABCB1 upregulation also had other mechanisms of PARPi resistance, suggesting that multiple cellular pathways act in parallel to confer clinical drug resistance at progression on PARPis.

Our study also shows an increased burden of gene-fusion transcripts in the post-PARPi setting, suggesting enhanced genomic instability in these tumors. This preliminary observation warrants further investigation in a larger dataset as targeting these changes may present a new therapeutic opportunity. CCNE1 amplification has been observed at progression on PARPis, and although not considered as a drug resistance mechanism per se (32), given trends toward mutual exclusivity with BRCA1/2 alterations due to synthetic lethality (46), it is commonly considered a surrogate biomarker of poor response to platinum (32, 44). However, we observed more CCNE1 amplifications in BRCA1/2-mutated than BRCA1/2 wild-type tumors (33% vs. 15%, respectively), and in platinum-sensitive than platinum-resistant patients (36% vs. 20%, respectively). The most likely explanation for these discrepancies versus previous reports is the extensive pretreatment and selection criteria in our study. CCNE1 amplifications were detected in only 12% of archival (treatment-naïve) samples, probably because eligible patients had previously responded to platinum therapy, as required for access to prior PARPi. Intriguingly, when baseline samples were examined, BRCA1/2-deficient patients lacking a subsequent reversion mutation but who also had CCNE1 amplifications experienced a significant improvement in PFS versus all other patients (median, 7.6 vs. 2.5 months, respectively; Supplementary Fig. S3). This suggests that these patients have at least partial activity of this synthetic lethality phenotype, despite their long treatment history. Resistance to PARPis can also be driven by SLFN11 inactivation, which may open specific treatment options, as suggested in a preclinical model where resistance to PARPis was overcome by ATR inhibition (47).

We observed increased enrichment of the angiogenesis pathway in samples taken after progression on a PARPi, whereas patients who had received prior antiangiogenic therapy showed the opposite effect, with post-PARPi samples showing decreased pathway enrichment (Supplementary Fig. S4). Although we observed no significant relationship between angiogenesis pathway enrichment and on-trial response, this could be due to potential confounders and small sample sizes. Thus, findings from this study would benefit from a randomized design and larger sample size in the postcombination setting to understand better how the various involved mechanisms affect response or benefit from this combination.

One of the main limitations of our study is the small sample size, albeit this is, to the best of our knowledge, the largest prospective translational study with paired archival and baseline biopsies characterizing resistance mechanisms after PARPi. These results should be considered as hypothesis generating and merit further exploration in a larger prospective trial. Translational findings from this study may also be affected by tumor heterogeneity. As described by Patch and colleagues (44), there is considerable tumor heterogeneity in HGSOC within and between metastatic sites. The site for biopsy post PARPi was the progressive lesion, but this was not necessarily the same site as the archival (pre-PARPi) tissue.

In conclusion, our study suggests that the cediranib–olaparib combination is tolerable with some activity in women with ovarian cancer following PARPi failure. Our data also hint that patients with reversion mutations in homologous recombination genes, and/or upregulated ABCB1, should perhaps be considered for other options to overcome their aggressive disease transformation. This observation in a small study warrants further evaluation. Several additional mechanisms of PARPi resistance have been observed, leading to potential therapeutic opportunities targeting these acquired new vulnerabilities.

S. Lheureux is a paid consultant for AstraZeneca and GlaxoSmithKline and reports receiving speakers bureau honoraria from Roche and Merck. A. Oaknin reports receiving speakers bureau honoraria from Roche, AstraZeneca, PharmaMar, Clovis Oncology, Immunogen, and Genmab. J.P. Bruce is an unpaid consultant/advisory board member for Bowhead Health. N.C. Dhani reports receiving speakers bureau honoraria from AstraZeneca. V. Bowering reports receiving speakers bureau honoraria from AstraZeneca. L. Fariñas-Madrid reports receiving speakers bureau honoraria from MSD, Roche, and AstraZeneca, and is an unpaid consultant/advisory board member for Tesaro. A.M. Oza is an unpaid consultant/advisory board member for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Lheureux, T.J. Pugh, A.M. Oza

Development of methodology: S. Lheureux, S. Accardi, T. Li, L. Fariñas-Madrid, T.J. Pugh, A.M. Oza

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Lheureux, A. Oaknin, N.C. Dhani, V. Bowering, S. Accardi, S. Pedersen, T. Li, L. Fariñas-Madrid, Y.C. Lee, A.M. Oza

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Lheureux, A. Oaknin, S. Garg, J.P. Bruce, A. Madariaga, N.C. Dhani, Q. Tan, M. Braunstein, Y.C. Lee, Z. (Amy) Liu, T.J. Pugh, A.M. Oza

Writing, review, and/or revision of the manuscript: S. Lheureux, A. Oaknin, S. Garg, J.P. Bruce, A. Madariaga, N.C. Dhani, V. Bowering, Q. Tan, M. Braunstein, K. Karakasis, Y.C. Lee, Z. (Amy) Liu, T.J. Pugh, A.M. Oza

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Lheureux, S. Garg, J. White, S. Accardi, Y.C. Lee, Z. (Amy) Liu, T.J. Pugh, A.M. Oza

Study supervision: S. Lheureux, A. Oaknin, J. White, S. Accardi, L. Fariñas-Madrid, A.M. Oza

We would like to thank the patients who participated in this study. We also extend our thanks to the staff of the Department of Gynaecological Drug Development and the Princess Margaret Genomics Centre (Troy Ketela, Neil Winegarden, Julissa Tsao, Nick Khuu, and Gurbaksh Basi) and the Bioinformatics and High-Performance Computing Core (Carl Virtanen, Zhibin Lu, and Natalie Stickle) for their expertise in generating the sequencing data used in this study. The authors would like to acknowledge Princess Margaret Genomics Centre (www.pmgenomics.ca) for sequencing support, Princess Margaret Cancer Foundation for funding, AstraZeneca for support through an investigator-initiated trial grant and drug supply, and Ontario Institute for Cancer Research through the Government of Ontario for supporting translational analyses. Editorial support was provided by Jennifer Kelly, MA (Medi-Kelsey Ltd.).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.