A Phase 2 Trial of Niraparib in Patients with Advanced Pancreatic Cancer Harboring Pathogenic Variants in ATM, BRCA1, BRCA2, PALB2, and CHEK2

Abstract

Purpose:

Poly(adenosine diphosphate-ribose) polymerase (PARP) inhibition has demonstrated efficacy in patients with platinum-sensitive pancreatic cancer with germline BRCA1/2 pathogenic variants (PV). Whether PARP inhibitors might be effective in a broader population of patients with pancreatic cancer remains under investigation.

Patients and Methods:

This multicenter, open-label phase II trial (NCT03601923) enrolled patients with advanced pancreatic cancers that harbored germline or somatic BRCA1, BRCA2, PALB2, ATM, and/or CHEK2 PVs. Patients with prior progression on platinum-based therapy were excluded. Patients were treated with niraparib 200 or 300 mg once daily, with their initial dose determined by weight and platelet count. The primary endpoint was 6-month progression-free survival (PFS).

Results:

Thirty-two patients (ten women [31%]; median age: 67 years) were enrolled. Patients had a PV in at least one of the following genes: ATM (n=14), BRCA2 (n=10), PALB2 (n=3), CHEK2 (n=4), or BRCA1 (n=2). The 6-month PFS was 25% (90% CI: 13–41%) for the overall population, exceeding the pre-established threshold of 17%. The median PFS for the entire population was 2.0 months (95% CI: 1.4–3.8), and the objective response rate was 14% (95% CI: 4–33%). All six patients with ≥ 6 months of PFS and evaluable tumor zygosity had biallelic inactivation of a DNA repair gene. Three patients with biallelic inactivation of ATM and no progression on prior chemotherapy received niraparib for more than 1 year.

Conclusions:

Niraparib demonstrated clinically meaningful benefit in a subset of patients. The prolonged PFS observed in patients with biallelic ATM loss warrants further investigation.

INTRODUCTION

Treatment for advanced pancreatic cancer has traditionally been limited to cytotoxic chemotherapy (1,2). Analyses of large scale genomic studies have identified that a subset of pancreatic cancers have germline and somatic pathogenic variants (PV) in the homologous recombination (HR) DNA repair pathway (3–10). The most common DNA repair alterations are in BRCA1/2 (~8%), ATM (~4%), PALB2 (~2%), and CHEK2 (~2%) (6,8,11–13). Pancreatic cancers harboring pathogenic BRCA1/2, PALB2, and ATM alterations exhibit heightened susceptibility to platinum-based chemotherapy regimens, reflecting the roles of these proteins in repairing double-stranded DNA breaks (3–6,14–17).

Building on the success of platinum sensitivity as a biomarker for poly(adenosine diphosphate-ribose) polymerase (PARP) inhibitor sensitivity in ovarian cancer, the landmark POLO trial demonstrated the efficacy of maintenance PARP inhibition in patients with pancreatic cancer and BRCA1/2 germline PVs who had at least stable disease on first-line platinum containing chemotherapy (18–22). While a subset of patients on the POLO trial had durable prolonged responses to the PARP inhibitor olaparib, in the overall population olaparib improved the median progression-free survival (PFS) by 3.6 months compared to placebo control (18). Maintenance PARP inhibition with the PARP inhibitor rucaparib was also efficacious in patients with pancreatic cancer who had either germline or somatic BRCA1/2 or PALB2 PVs (15). One report evaluating olaparib’s efficacy in patients with pancreatic cancer with DNA repair alterations other than germline BRCA1/2 PVs included 14 patients with ATM PVs (seven germline and seven somatic). This study found a median PFS of 3.7 months overall and 5.0 months in patients with ATM PVs (23).

The variation in PARP inhibitor sensitivity in pancreatic cancers harboring BRCA1/2 and PALB2 alterations, characterized by occasional prolonged responses contrasted with frequent resistance, poses a significant clinical challenge and underscores the urgent need for additional predictive biomarkers. Beyond platinum sensitivity, studies in multiple malignancies have suggested that complete inactivation of the HR gene, caused by genomic biallelic loss, also increases the likelihood of PARP inhibitor sensitivity (5,6,24,25). Furthermore, genomic signatures of HR deficiency, such as COSMIC signature SBS3, are also predictive of PARP inhibitor sensitivity (26–28). Biomarkers predictive of PARP inhibitor resistance include reversion mutations in BRCA1/2 and PALB2 that restore the function of the underlying altered gene (29,30).

Efforts to develop clinically-actionable biomarkers that predict PARP inhibitor sensitivity have been challenging. Functional assays, such as the RAD51 assay, that can distinguish between HR proficient and deficient tumors, typically require live tumor cells which is not currently practical in clinical settings (31). Genomic methods analyzing mutational signatures of HR deficiency, such as COSMIC signature SBS3 and the HRDetect assay, require whole exome sequencing or whole genome sequencing (27,32,33). However, recent findings indicate that genomic assays designed for clinical-grade next-generation sequencing platforms, such as HRDsig, show promising results in detecting HR deficiency (34,35).

Niraparib is a highly selective, low nanomolar inhibitor of PARP1 and PARP2, that has a longer half-life and improved bioavailability compared to the PARP inhibitors olaparib and rucaparib (36,37). It is approved by the Food and Drug Administration (FDA) as maintenance therapy in platinum-sensitive ovarian cancer (38,39). To enhance understanding of the role of PARP inhibition in pancreatic cancer, we conducted a multicenter phase 2 clinical trial evaluating niraparib in patients whose pancreatic cancers had not progressed on platinum-containing chemotherapy regimens. Patients enrolled on the trial had germline or somatic BRCA1, BRCA2, PALB2, ATM or CHEK2 PVs.

METHODS

Patient Population and Trial Design

This multicenter, open label, single arm phase 2 clinical trial examined niraparib in advanced pancreatic cancer (ClinicalTrials.gov identifier: NCT03601923). Eligible patients had incurable locally advanced or metastatic pancreatic cancer with a germline (identified on germline testing) or somatic (not identified on germline testing) likely pathogenic or pathogenic alteration in ATM, BRCA1, BRCA2, CHEK2, or PALB2. Circulating tumor DNA was not used to establish eligibility. A molecular review committee assessed whether the genomic alteration in the DNA repair gene was pathogenic. Patients were eligible for inclusion at any line of palliative treatment, including the maintenance setting and after progression on non-platinum chemotherapy. The only exclusionary prior treatment was a history of disease progression on platinum-containing chemotherapy. Additional eligibility requirements included age ≥18 years, measurable disease according to Response Evaluation Criteria in Solid Tumors, version 1.1 (RECIST v1.1) criteria, and an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. Key exclusion criteria included prior treatment with a PARP inhibitor, ongoing grade ≥ 2 toxicity from prior therapy, and known history of myelodysplastic syndrome or acute myeloid leukemia.

The study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice Guidelines. The study protocol and all amendments were approved by Dana-Farber Harvard Cancer Center institutional review board. All patients provided written informed consent. This investigator-initiated trial was funded by Tesaro (GSK).

Treatment and Assessments

Each patient’s initial niraparib dose was based on their baseline weight and platelet count. Patients received niraparib 200mg daily if they weighed less than 77 kg and/or had a baseline platelet count below 150 × 10³/microliter. All other patients received niraparib 300 mg once daily. In cases with no treatment interruption or dose reduction during the first two cycles, the daily dose for patients receiving 200 mg could be increased to 300 mg at the investigators’ discretion.

Outcomes

The primary endpoint was 6-month PFS, defined as a binary outcome. Enrollment of 32 participants provided 80% power to differentiate between an unacceptable 6-month PFS of 5% (40,41) and a desirable 6-month PFS of 17% at a 10% one-sided type 1 error rate. The null hypothesis 6-month PFS rate of 5% was established using the CONKO-003 trial and NAPOLI-1 trial as benchmarks, as there were no positive pancreatic cancer PARP inhibitor trials at the time of study design (40,41). For the primary endpoint, an exact 90% confidence interval (CI) was used, corresponding to a 10% type 1 error rate and the 6-month PFS rate was compared to the null value (5%) using a two-sided exact binomial test; for all other analyses, a 95% CI was used. PFS was defined as the time from registration to documented disease progression or death from any cause, whichever occurred first. Participants who had not experienced an event of interest by the time of analysis were censored at the date of the last disease assessment. The secondary endpoints included objective response rate (ORR), based on the RECIST v1.1 criteria, PFS, and overall survival (OS) in an intention-to-treat analysis. OS was calculated as the time from treatment initiation until death. The Kaplan-Meier method was used to estimate PFS and OS, and the log-rank test and Cox proportional hazards models were used for comparisons. Adverse events were graded according to the Common Toxicity Criteria for Adverse Events (CTCAE), version 5.0. Statistical analyses were performed in SAS v9.4 (SAS Institute, Cary, NC, RRID: SCR_008567).

Tumor Sequencing and Analysis

Fresh tumor biopsy or available archival samples were genomically analyzed with a next generation sequencing (NGS) panel (OncoPanel (42) or Foundation One test (43)), whole exome sequencing, or both, depending on tissue availability. Whole exome sequencing was performed at Novogene (100X coverage) or the Broad Institute (100X coverage). To assess the genomic alterations present in each patient and construct an oncoprint, we used MuTect (v1.1.4, RRID:SCR_000559) to detect single nucleotide variants, GATK Indelocator (RRID:SCR_005258) for identifying insertions and deletions, and the RobustCNV (internal tool) and BreaKmer (RRID:SCR_026898) algorithms for detecting copy number and structural variants. Germline mutational status was determined by a commercial assay or whole exome sequencing (100X coverage) of a blood sample (42).

An expert molecular pathologist reviewed the available genomic sequencing data to determine the biallelic status of each sample. DNA repair genes were evaluated for biallelic loss based on the presence of deleterious alterations affecting that gene, including deleterious single nucleotide variants (SNVs), insertion and deletions (INDELs), copy number alterations, and structural rearrangements. Determination of biallelic loss for a gene required the presence of two deleterious alterations, one deleterious alteration plus loss of heterozygosity, or complete or partial homozygous deletion. Variant allele fractions and copy number log2 ratios were reviewed to ensure consistency with these findings in the context of the sample tumor purity.

Analysis of Genomic Signatures of HR Deficiency

Four patients (P02, P03, P06, and P21) had matched normal-tumor sample pairs with whole exome sequencing (100X), thus enabling high-confidence somatic variant calling in their tumor samples by using the default Mutect2 (RRID:SCR_026692) filters. The somatic mutational signature composition (COSMIC version 3.4, RRID:SCR_002260) of these variants was estimated with SigProfilerAssignment (v0.1.9, RRID:SCR_026899) (26). For the single base substitution (SBS) signatures, only those previously reported in pancreatic adenocarcinomas (SBS1, SBS2, SBS3, SBS5, SBS6, SBS8, SBS13, SBS17a, SBS17b, SBS18, SBS20, SBS26, SBS28, SBS30, and SBS40) were used, thereby resulting in high cosine similarities ranging from 0.89 to 0.96 (26,27).

Circulating Tumor DNA Sequencing and Analysis

Serial plasma samples were prospectively obtained from study patients and retrospectively analyzed using the Guardant Reveal assay, which uses proprietary next-generation sequencing technology and a bioinformatics classifier to evaluate thousands of differentially methylated regions for tumor fraction detection (20,44). Sequencing libraries are generated using cell-free DNA that has been partitioned based on its methylation state. Tumor fraction is estimated by normalizing cancer-specific differentially methylated regions with appropriately matched control regions within each sample. Comprehensive genomic profiling is accomplished using the Guardant360 genomic panel in samples with detectable ctDNA and includes genotyping of > 750 genes. For promoter methylation, promoter regions up to 5kb upstream of the canonical transcriptional start site were targeted, normalized to internal control regions, and binarized based on a 98% specificity threshold trained on over 2,500 ctDNA-free samples. The ctDNA results were analyzed for the presence of reversion mutations that restore the open reading frame of the HR gene. Subsequent analyses looked for evidence of microhomology-mediated end-joining (MMEJ) pathway mediated double-strand DNA repair deletions. Deletions were classified with the BSgenome.Hsapiens.UCSC.hg38 R (v1.4.5, RRID:SCR_024230) package. The genomic context of the sequence surrounding each deletion, in both 5’ and 3’ directions, was compared with the sequence of the deleted region. Deletions were categorized as follows: 1) “microhomology-mediated deletion” if the first N nucleotides, but not the entire deleted sequence, were repeated immediately after the deleted sequence, or if the last N nucleotides in the deleted sequence preceded it; 2) “repeat” if the entire deleted sequence was repeated either before or after the deletion; and 3) “unique deletion” if no repeated context surrounded the deletion. Additionally, deletions with a repeated context of less than 2 base pairs (bp) were categorized separately, given the non-negligible statistical likelihood of such deletions occurring randomly.

Data availability statement

The data generated in this study are available upon reasonable request from the corresponding author. The whole exome sequencing data is available in dbGaP under the accession number phs004405.v1.p1.

RESULTS

Between August 31, 2018 and August 19, 2022, 32 eligible patients at three institutions were enrolled (Supplementary Figure S1). The median age was 67 years (Table 1). Thirty-one percent of patients were women (n=10), and most patients were white (n=28 [88%]). Ten patients received niraparib as maintenance therapy (defined as no history of progression after 4–12 months of FOLFIRINOX-based chemotherapy), and 12 patients were experiencing disease progression on a non-platinum-based chemotherapy regimen at the time of study entry. Four patients had never been exposed to platinum chemotherapy before receipt of niraparib. Eighteen (56%) patients were treated with one line of systemic therapy before receiving niraparib in the trial, 11 (34%) patients were treated with two prior lines of systemic therapy, and three (9%) patients were treated with three or more lines of prior therapy (median number of therapies: 1; range: 1–5). The demographic representativeness of study participants is described in Supplementary Table S1.

Table 1:

Demographics and Baseline Clinical Characteristics

Baseline characteristics	N=32
Age, years, median (range)	67 (45–80)
Female, No. (%)	10 (31)
Race, No. (%)
White	28 (88)
Black or African American	1 (3)
Asian	1 (3)
Other	2 (6)
Time from diagnosis to study start, months
Median	13.7
Range	4.3–79.5
Pathogenic variant
ATM, total No. (%)	14 (44)
Germline, No. (% of ATM)	12 (86)
Somatic, No. (% of ATM)	2 (14)
Biallelic, No. (% of known ATM allelic status)	8 (89)
Monoallelic, No. (% of known ATM allelic status)	1 (11)
Unknown allelic status, No. (% of ATM)	5 (36)
BRCA1, total No. (%)	2 (6)
Germline, No. (% of BRCA1)	1 (50)
Unknown germline, No. (% of BRCA1)	1 (50)
Biallelic, No. (% of BRCA1)	2 (100)
BRCA2, total No. (%)*	10 (31)
Germline, No. (% of BRCA2)	7 (70)
Somatic, No. (% of BRCA2)	3 (30)
Biallelic, No. (% of known BRCA2 allelic status)	4 (100)
Unknown allelic status, No. (% of BRCA2)	6 (60)
CHEK2, total No. (%)*	4 (13)
Germline, No. (% of CHEK2)	4 (100)
Somatic, No. (% of CHEK2)	0 (0)
Unknown allelic status, No. (% of CHEK2)	4 (100)
PALB2, No. (%)	3 (9)
Germline, No. (% of PALB2)	2 (66)
Somatic, No. (% of PALB2)	1 (33)
Biallelic, No. (% of known PALB2 allelic status)	2 (100)
Unknown allelic status, No. (% of PALB2)	1 (33)
ECOG performance status, No. (%)
0	19 (59)
1	13 (41)
Number of prior therapies, No. (%)
1	18 (56)
2	11 (34)
≥3	3 (9)
Prior therapies
FOLFIRINOX, No. (%)	27 (84)
Gemcitabine/nab-paclitaxel, No. (%)	12 (38)
Radiation therapy, No. (%)	11 (34)
Primary tumor resection, No. (%)	8 (25)
Niraparib maintenance therapy, No. (%)^	10 (31)
Sites of disease
Bone, No. (%)	1 (3)
Liver, No. (%)	19 (59)
Lung, No. (%)	7 (22)
Lymph node, No. (%)	9 (28)
Peritoneum, No. (%)	7 (22)

^*One patient had two pathogenic germline variants identified in BRCA2 and CHEK2.

^{^}Niraparib maintenance therapy defined as no history of progression after 4–12 months of FOLFIRINOX-based chemotherapy.

Two patients had a BRCA1 PV (one germline and one somatic), ten patients had a BRCA2 PV (eight germline and two somatic), three patients had a PALB2 PV (two germline and one somatic), 14 patients had an ATM PV (12 germline and two somatic), and four patients had a germline CHEK2 PV (including one patient [patient 31] with dual germline BRCA2 and CHEK2 PVs). The BRCA1/2, PALB2, ATM, and CHEK2 PVs are listed in Supplementary Table S2. Thirty-one patients had pancreatic ductal adenocarcinoma, and one patient had acinar cell carcinoma of the pancreas with a germline BRCA2 PV (patient 21). Thirty patients had metastatic disease at enrollment. The most common oncogenic alterations, other than HR pathway genes, identified in somatic sequencing of trial patients (n=24) were KRAS (83%), TP53 (38%), and CDKN2A (29%) (Figure 1). In agreement with previous reports showing a high prevalence of biallelic ATM inactivation in patients with germline ATM PVs, we observed biallelic ATM inactivation in six of the seven patients (86%) with germline ATM PVs and evaluable allelic status (Figure 1) (6,45).

Figure 1: Genomic alterations in pancreatic cancers from patients enrolled in the trial.

OncoPanel next generation sequencing (NGS), Foundation One commercial NGS, or whole exome sequencing (Exome) was performed on tumor samples for 24 patients enrolled in the trial. Columns represent genomic profiles of individual patients. Genomic alterations are depicted in rows, with the homologous recombination (HR) genes required for eligibility listed at the top and the most frequently altered genes listed at the bottom. The tumor mutational burden (TMB), cancer type, sample type, assay, 6 month progression-free survival (PFS) on niraparib, zygosity status, and germline status for each individual patient are displayed as tracks above the gene alteration profile. A color-coded legend depicting these tracks and the types of genomic alterations is shown to the right of the oncoprint.

Niraparib Efficacy in the Overall Study Population

As of the data cut-off date of October 30, 2023, 30 of 32 patients had discontinued study treatment. The most common reason for withdrawal was progressive disease (84%). The median follow-up for patients was 10.2 months (range: 1.2–61). At the time of the data cut-off, 26 (81%) patients had died (Supplementary Figure S1).

The study met its primary endpoint, with an observed 6-month PFS of 25% (8/32) (90% CI: 13–41; p<0.001 against the null hypothesis of 5% 6-month PFS) (Figure 2). For the entire trial population, the median PFS was 2.0 months (95% CI: 1.4–3.8), and the median OS was 10.2 months (95% CI: 6.2–13.2) (Supplementary Figure S2). The ORR was 14% (95% CI: 4–33) (n=28 evaluable patients). Two patients had a confirmed partial response, and two patients had a confirmed complete response. Both patients who experienced a complete response remain on niraparib after 5.1 years (patient 2) and 3.8 years (patient 18) (Figure 3).

Figure 2: Kaplan-Meier estimates of progression-free survival (PFS) in patients treated with niraparib across the entire trial cohort.

Abbreviations: Number (No.) and confidence interval (CI).

Figure 3: Swimmer plot indicating duration of time on niraparib.

The color-coded columns to the left indicate the germline and zygosity status of each patient’s DNA repair gene(s) with the pathogenic variant (PV), as well as whether the patient’s tumor previously progressed on chemotherapy (the legend for the color coding is shown at right).

To better understand the differences between niraparib sensitive and resistant cancers, we investigated additional clinical and genomic features to assess their association with outcomes. We examined whether the zygosity status of the tumor (monoallelic versus biallelic) was associated with 6-month PFS. In this subset of patients who remained on niraparib for at least 6 months, all six patients with evaluable tumor zygosity had biallelic inactivation of the HR gene (Figures 1 and 3). We next analyzed the impact of progression on prior non-platinum containing chemotherapy regimens on survival outcomes. Patients whose tumors had not previously progressed on chemotherapy exhibited significantly longer PFS (p=0.02) (Supplementary Figure S3A) and OS (p=0.003) (Supplementary Figure S3B) than those whose tumors previously progressed on non-platinum containing chemotherapy. We also investigated whether a lower baseline tumor burden might be associated with improved outcomes, as observed in a recently reported trial (15). However, we observed no difference in baseline tumor burden (overall sum in millimeters of measurable tumor burden, as measured by independent radiology review for RECIST measurements) and PFS benefit. Additionally, we found no significant association between the time from diagnosis to enrollment and PFS or OS in this patient population. Clinical characteristics of those patients who remained on niraparib for at least 6 months is summarized in Supplemental Table S3.

Niraparib Efficacy in Genomically Defined Subtypes of Pancreatic Cancer

In patients with a core HR gene (BRCA1, BRCA2, or PALB2) PV (n=15), the 6-month PFS was 20% (3/15) (95% CI: 4–48), the ORR was 13% (2/15), and the median PFS was 3.6 months (95% CI: 1.8–5.5). No significant differences were observed in the 6-month PFS rate, PFS, or ORR among tumors with BRCA1, BRCA2, or PALB2 PVs. Similarly, no differences were observed in the 6-month PFS rate, ORR, PFS, or OS when comparing germline to somatic altered tumors. A patient with biallelic loss of BRCA1 (patient 2; germline BRCA1 p.E143* alteration and BRCA1 single copy deletion) had a CR to niraparib and continues in the trial after 5 years of therapy.

In patients with tumors harboring ATM PVs (n=14), the 6-month PFS was 36% (5/14) (95% CI: 13–65), the ORR was 14% (2/14), and the median PFS was 1.6 months (95% CI: 0.9–9.4). Three patients with biallelic loss of ATM (patients 18, 22, and 30), whose tumors had not progressed on prior lines of chemotherapy, received niraparib for more than 1 year. Given the variation in PFS among patients with ATM PVs, we assessed whether the type or location of ATM alterations might affect patient outcomes. However, we observed no clear differences in the type or location of ATM alterations between patients with PFS longer than 6 months or shorter than 6 months (Supplementary Figure S4). Notably, among patients with tumors harboring ATM PVs, those whose cancer had not previously progressed on chemotherapy had a significantly longer median PFS than those showing previous progression on non-platinum containing chemotherapy (median PFS 9.4 versus 1.2 months, p=0.03) (Supplementary Figure S3C).

There were 4 patients who had a germline CHEK2 PV. All patients with CHEK2 PVs experienced rapid tumor progression in the trial, and the 6-month PFS rate was 0% (0/4). The ORR was 0%, and the median PFS was 1.6 months (95% CI: 1.4; unreached).

Safety

A total of 41% (13/32) of patients experienced at least one adverse event of grade 2 or higher, and events of grade 3 or higher occurred in 6/32 patients (Supplementary Table S4). Grade 3 adverse events included ALT, AST, and GGT elevation (n=1 patient) along with anemia (n=3 patients), and hypertension (n=2 patients). One patient had grade 4 neutropenia, which resolved after holding therapy for 7 days and did not recur after a niraparib dose reduction. No patients discontinued the study because of niraparib-related toxicity, and no treatment-related deaths occurred.

Analysis of Genomic Signatures of HR Deficiency and Clinical Outcomes

Genomic analysis of tumors with HR deficiency in multiple malignancies has demonstrated a specific genomic signature, COSMIC signature SBS3, that is characteristic of HR-deficient tumors (26,27,46). We investigated the presence of signature SBS3 in four patients (patients 2, 3, 6, and 21) who underwent whole exome sequencing on both tumor and germline DNA samples (Figure 4A). Two patients (patients 2 and 6), with BRCA1 and BRCA2 germline mutations, respectively, who had a confirmed partial response to niraparib and a PFS greater than 6 months, exhibited signature SBS3 positive tumors. Additionally, the tumor from patient 21, which had biallelic loss of BRCA2, also showed signature SBS3 positivity. Patient 21 benefitted from 33 cycles of FOLFIRINOX/FOLFIRI before enrolling in the study, remained on niraparib for 5.5 months, and showed a maximal tumor response of −24% according to RECIST criteria. Notably, in agreement with findings in both pancreatic cancer and other malignancies, signature SBS3 positivity was not observed in patient 3, whose tumor had biallelic loss of ATM (Figure 4A and 4B) (6,32,46,47).

Figure 4: Mutational signature analysis is patient samples.

(A) Bar charts indicating the proportions of COSMIC single base substitution (SBS) signatures across four patient samples (patients 2, 3, 6, and 21). The y axis represent the ratio of each SBS signature to the total number of variants in each sample. Each color represents a distinct SBS signature (legend on the right). (B) Bar charts showing SBS mutational spectrums of reference and patient samples. The top panel displays the reference COSMIC SBS3 mutational signature. The lower two panels display the SBS mutational spectrums observed in patient 21 (middle) and patient 3 (bottom).

Analysis of Circulating Tumor DNA

We analyzed ctDNA with available banked plasma samples to see if the inclusion PV in ATM, BRCA1, BRCA2, CHEK2, or PALB2 could be detected via ctDNA. Of the 31 patients with ctDNA available from either an archival sample or a sample obtained anytime during the trial, we detected the inclusion DNA repair gene PV in ctDNA in 90% of samples (28 patients). In the three patients where the DNA repair gene PV was not detectable, one had a germline deletion involving the 5’ untranslated region and exon 1 of BRCA2 (patient 4), one had a germline Alu insertion in ATM (patient 9) which was not technically detectable via the ctDNA assay, and one had a homozygous deletion of BRCA1 (patient 11) with a tumor fraction below the threshold for copy number loss calling on the ctDNA assay. We then analyzed the number of patients with detectable baseline tumor fraction in ctDNA. In the 22 patients with baseline (cycle 1, day 1) plasma samples evaluable for ctDNA analysis, 19 patients (86%) had detectable baseline tumor fraction. Baseline tumor fraction detection in ctDNA was associated with significantly worse PFS (1.8 months vs not reached, p=0.006) and OS (8.6 months vs not reached, p=0.03) compared to patients with no tumor fraction detected (Supplementary Figure S5A and B). Separation of patients based on detectable tumor fraction above versus below the median did not result in significant differences in PFS and OS (Supplementary Figure S5C and D).

Reversion mutations that restore the open reading frame of the altered HR gene are known to cause resistance to PARP inhibitors (20,29). To investigate reversion mutations in patients enrolled in our trial, we analyzed the ctDNA from serially collected plasma samples. No reversion alterations were observed in patients with PVs in BRCA1, CHEK2, PALB2, or ATM. However, reversion mutations were observed in three of the nine patients with BRCA2 PVs who had evaluable ctDNA. All patients developing reversion mutations had BRCA2 frameshift mutations. Two of these patients had biallelic loss of BRCA2, according to tumor NGS (patients 16 and 32), and the third patient’s allelic status was unevaluable by tumor NGS, although ctDNA suggested biallelic loss of BRCA2 (patient 24). While the BRCA2 reversion mutations observed in patients 24 and 32 occurred 3.7 months (patient 24) and 5 months (patient 32) after the start of niraparib, the BRCA2 reversion mutation in patient 16 was present in the patient’s pre-niraparib (baseline) blood sample. Genomic analysis of these reversion mutations revealed that the majority of reversion mutations were generated by the microhomology-mediated end joining (MMEJ) pathway of double-strand DNA repair (Figure 5). A genomic depiction of how representative BRCA2 reversion mutations restore the BRCA2 open reading frame are outlined in Supplementary Figure 6.

Figure 5: Summary of deletion types in circulating tumor DNA samples from three patients who developed reversion mutations in BRCA2.

Figure 5A is a segmented bar chart illustrating the number and type of deletions observed in each patient with a BRCA2 reversion mutation. The legend indicating the color code for each type of deletion is at the top left corner. Figure 5B shows genomic examples of the different types of deletions occurring in niraparib-treated patients. The nucleotides listed in the top row are the deleted sequence, and the nucleotides in the bottom row are the reference sequence from BRCA2.

Promoter methylation in HR genes has been identified as a mechanism of PARP inhibitor sensitivity (48,49). In addition to analyzing genomic alterations, the Guardant Infinity assay also examines promoter methylation for select genes. Examination of promoter methylation of DNA repair genes in ctDNA did not identify any “second hit” HR gene promoter methylation events that would have led to biallelic inactivation of the DNA repair gene. The only DNA repair gene promoter methylation event observed was methylation of the BRCA1 promoter in a patient with a CHEK2 mutated tumor.

DISCUSSION

PARP inhibitors are effective in a subset of patients with pancreatic cancer who have DNA repair pathway alterations (15,18,23,50). Our study aimed to extend these findings by evaluating whether there might be benefit to PARP inhibition in patients with pancreatic cancer that have not progressed on platinum chemotherapy and harbor pathogenic germline or somatic PVs in BRCA1, BRCA2, PALB2, ATM, or CHEK2 genes. We observed that 25% (8/32) of patients achieved a PFS of 6 months or longer, which met our preestablished primary endpoint (40,41). Highlighting the importance of tumor zygosity in determining PARP inhibitor sensitivity, all six patients with zygosity evaluable tumors and at least 6-month PFS had biallelic loss of the DNA repair gene.

The anti-tumor efficacy observed in our study was modest compared to the 6-month PFS of 53%–60% reported in other trials evaluating maintenance PARP inhibition (15,18). The observed differences in efficacy between our trial and others are largely attributable to differences in trial design. To identify an expanded population of patients who might benefit from PARP inhibition, we enrolled patients with PVs in CHEK2 and ATM, as well as patients who had previously experienced disease progression on non-platinum containing chemotherapy regimens. While the heterogeneity of our trial design allowed us to explore multiple different populations, a weakness of this design is that this heterogeneity makes it difficult to compare our overall findings with pancreatic PARP inhibitor trials that had a more homogenous population.

In line with the observation that disease progression on platinum-based chemotherapy predicts resistance to PARP inhibitors, we found that progression on non-platinum containing chemotherapy was also associated with primary resistance to PARP inhibition (22,51). The median PFS in patients with prior progressive disease on non-platinum containing chemotherapy was significantly lower than in those without previous disease progression on chemotherapy. This finding aligns with earlier trials of PARP inhibitors in pancreatic cancer, which also demonstrated limited efficacy in patients with previously treated pancreatic cancer (51–53).

While the median PFS in patients with tumors that had ATM PVs was only 1.6 months, five patients with germline ATM PVs stayed on niraparib for over 6 months without disease progression. Three of these patients, whose cancers never previously progressed on chemotherapy and had biallelic loss of ATM, remained on niraparib for more than 1 year, with one patient continuing in the trial after 3.8 years. The effectiveness of PARP inhibitors in ATM-deficient tumors remains a major unanswered question. In their trial of the olaparib PARP inhibitor, Javle et al. observed several patients with ATM-altered pancreatic cancer who had stable disease and reported a median PFS of 5.0 months in the subgroup of pancreatic cancers harboring ATM PVs (23). While ATM activates the HR pathway, ATM-deficient tumors lack evidence of the characteristic genomic changes of HR-deficient tumors such as COSMIC signature SBS3, which our analysis confirms as well (6,31,32,46,54). However, PARP inhibitors have been speculated to promote cytotoxicity in ATM-deficient tumors through the accumulation of toxic DNA alterations generated by non-homologous DNA end joining (NHEJ) and by induction of replication fork instability (55,56). In prostate cancer, although PARP inhibitors are FDA-approved for tumors with ATM PVs, the efficacy of PARP inhibitors in this population has been disappointing in most patients (57). However, molecular analysis in the TOPARP-B and TALAPRO-1 prostate cancer trials suggests that biomarkers associated with complete inactivation of ATM function, including loss of ATM protein expression and biallelic inactivation of ATM, are associated with PARP inhibitor sensitivity Further complicating the assessment of PARP inhibitor efficacy in ATM-deficient tumors, is emerging research from several groups indicating that pancreatic cancers harboring ATM PVs have prolonged overall survival and heightened sensitivity to platinum-based chemotherapy (6,14,58). These findings suggest that pancreatic cancers harboring ATM PVs might possibly have a more indolent and platinum-sensitive biology compared to those without ATM PVs. One possibility is that the favorable outcomes experienced by a subset of patients with ATM-altered tumors is a “carry-over” effect from their previous chemotherapy. This “carry-over” effect was observed in some placebo-treated patients on the POLO study who experienced a partial response and/or prolonged PFS following platinum chemotherapy (18). Hence, further study is needed to establish whether PARP inhibitors are efficacious in patients with ATM-altered tumors. Although a randomized trial would be ideal, the relative rarity of ATM-altered pancreatic cancer would make accrual to such a study challenging. Before initiating a randomized study, an individual patient-level meta-analysis of trials evaluating PARP inhibitors in this specific population could help assess if there is an efficacy signal.

To explore whether CHEK2 PVs might serve as biomarkers of PARP inhibitor sensitivity, we included four patients with germline CHEK2 PVs (59,60). Unfortunately, none of these four patients derived clinical benefit from niraparib monotherapy. A similar lack of efficacy has also been observed in trials evaluating PARP inhibitors in CHEK2-altered tumors across other malignancies, including the TBCRC 048 breast cancer trial (n=8) and the Belgian Precision tumor-agnostic trial (n=14) (61–63). Similarly, in PARP inhibitor trials in prostate cancer, the vast majority of patients with CHEK2 PVs have not appeared to benefit from PARP inhibition (64–66).

BRCA1/2 reversion mutations that restore the open reading frame of the altered HR gene are a validated marker of PARP inhibitor resistance and can be detected with ctDNA (20,29). We observed that three patients with BRCA2 frameshift mutations, all of whom appeared to have BRCA2 biallelic loss, developed a BRCA2 reversion mutation on ctDNA analysis. The majority of the BRCA2 reversion mutations appear to have been generated by the MMEJ pathway of double-strand DNA repair. This observation is consistent with emerging data suggesting that the addition of a polymerase theta (Polθ) inhibitor, which block MMEJ-mediated DNA repair, could enhance the efficacy of PARP inhibition (20,67–69). Notably, one patient (patient 16) had a BRCA2 reversion mutation present at baseline and their tumor progressed rapidly on niraparib, suggesting that analysis of ctDNA before initiation of PARP inhibitor therapy might aid in identifying patients with PARP inhibitor resistant tumors who should not be treated with PARP inhibitors. Beyond the three patients with BRCA2 frameshift mutations, we did not observe any other patients with reversion mutations. The infrequent occurrence of reversion mutations in our trial is consistent with a study of maintenance rucaparib in pancreatic cancer, which also found that the detection of BRCA1/2 and PALB2 reversion mutations in ctDNA was uncommon (29). These ctDNA findings may indicate that, in contrast to breast and ovarian cancer, alternative PARP inhibitor resistance mechanisms may play a predominant role in pancreatic cancer (20,70). However, another potential explanation for the rarity of reversion mutations is the relatively low ctDNA shed rate of pancreatic cancers and the presence of monoallelic BRCA1/2 PVs (71,72). Of note, in an effort to identify other ctDNA-based biomarkers of PARP inhibitor sensitivity and resistance, we assessed DNA repair gene promoter methylation. We did not observe any evidence of “second hit” promoter methylation that contributed to biallelic inactivation of the DNA repair gene, in agreement with the results of two previous reports showing that BRCA1/2 and ATM promoter methylation is very uncommon in pancreatic cancer (48,73).

Limitations

The heterogeneity of our patient population, along with the relatively small sample size, makes it difficult to draw definitive conclusions about niraparib’s efficacy in this patient population. In addition, the trial’s single-arm design, without a comparator arm, further hinders the interpretation of niraparib’s efficacy. Hence, the efficacy results observed in the population with ATM PVs should be interpreted with caution, as they require validation in future randomized clinical trials. An additional limitation of the trial is molecular heterogeneity, given that some patients had biallelic loss and others had monoallelic loss of the DNA repair gene, which hampers the interpretation of the results. Given the importance of biallelic HR gene loss as a biomarker of PARP inhibitor sensitivity, future trials should aim to limit the trial population to patients with tumors that are confirmed to be HR deficient. While restricting eligibility to patients with tumors that have biallelic loss of the HR gene would be ideal, the limited cellularity of many pancreatic cancer tissue samples can make it challenging to detect biallelic loss using NGS assays. Instead, future efforts are needed to identify genomic assays that can identify HR deficiency using clinical-grade NGS platforms.

A final weakness of this trial was that interpretation of niraparib’s efficacy was complicated by differences in patients’ previous response to chemotherapy. The diminished efficacy seen in patients who had previously progressed on chemotherapy strongly supports limiting future PARP inhibitor trials to the maintenance population of patients who have not experienced progressive disease on any chemotherapy regimen.

Conclusion

In agreement with previous studies, this trial demonstrated that PARP inhibitors have efficacy in a subset of patients with BRCA1/2 and PALB2 PVs (15,18). This efficacy is enhanced in patients without prior disease progression on chemotherapy and in patients with biallelic loss of the HR gene. Interestingly, we also observed improved outcomes in patients with pancreatic cancer with germline ATM PVs, who had not previously shown disease progression on chemotherapy, and had evidence of biallelic loss of ATM in the tumor. Future randomized trials are needed to rigorously assess whether PARP inhibitors are an effective treatment in ATM-deficient pancreatic cancers, or if the observed improved clinical outcomes are due to biological differences driving the behavior of ATM-deficient tumors.

TRANSLATIONAL RELEVANCE.

PARP inhibition is FDA-approved as maintenance therapy for patients with pancreatic cancer who have germline BRCA1 or BRCA2 pathogenic variants (PVs). However, data on the efficacy of PARP inhibitors in patients with pancreatic cancer with germline or somatic PVs in other DNA repair genes are limited. This phase II trial evaluated the PARP inhibitor niraparib in patients with advanced pancreatic cancer who had no prior progression on platinum chemotherapy. Prolonged progression-free survival was observed in a subset of patients with BRCA1, BRCA2, and ATM PVs, including three patients with biallelic ATM inactivation who received niraparib for more than 1 year without disease progression. BRCA2 reversion mutations, which were generated predominantly by microhomology-mediated DNA repair, were observed in cell-free DNA in patients with both primary and acquired resistance to niraparib.