Associations with response to Poly(ADP-ribose) Polymerase (PARP) inhibitors in patients with metastatic breast cancer

PARP inhibitors (PARPi) have modest antitumor activity in patients with advanced breast cancer and mutation in BRCA. It is unclear whether some subgroups derive greater benefit from treatment. MEDLINE and EMBASE were searched from inception to March 2021 to identify trials of PARPi in patients with metastatic breast cancer. Objective response rate (ORR) and clinical benefit rate (CBR) to PARPi were extracted and pooled in a meta-analysis using the Mantel Haenszel random effects model. Meta-regression explored the influence of patient and tumor characteristics on ORR and CBR. For randomized trials, hazard ratio comparing PARPi to control therapy were pooled using inverse variance and random effects. Analysis included 43 studies comprising 2409 patients. Among these, 1798 (75%) patients had BRCA mutations and 1146 (48%) were triple negative. In 10 studies (28%; n = 680 patients), the PARPi was given in combination with platinum-based chemotherapy. Weighted mean ORR was 45%; 64% when combined with platinum vs 37% with PARPi monotherapy (p < 0.001). Previous platinum-based chemotherapy was associated with lower ORR (p = 0.02). Compared to standard chemotherapy, progression-free survival was improved (HR 0.64, p < 0.001), but there was no difference in overall survival (HR 0.87, p = 0.06). There were no differences in ORR or CBR between BRCA1 and BRCA2 mutations. PARPi are more active in combination with platinum than as monotherapy, with lower response if given as monotherapy after platinum exposure. Significant improvements in ORR translated to modest improvement in progression-free, but not overall survival. There was no association between ORR and BRCA mutations.


INTRODUCTION
Over the last decade, poly-ADP ribose polymerase (PARP) inhibitors (PARPi) have emerged as a new treatment option for metastatic breast cancer (MBC) 1 . PARP is a family of enzyme involved in base excision repair of single-strand DNA breaks 2 . When PARP is inhibited, single-strand breaks persist and result in stalled replication forks and double-strand breaks 3 . There are currently 5 PARPi being investigated in clinical trials for MBC: olaparib, rucaparib, niraparib, talazoparib and veliparib. They all differ in their potency for catalytic inhibition and their ability to trap PARP 4 . Currently, only olaparib and talazoparib are approved by the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) for use in MBC with a germline breast and ovarian cancer susceptibility gene (BRCA) mutation.
In MBC clinical trials, PARPi have been studied in 2 settings. The first setting is as monotherapy in cancers with impaired DNAdamage repair pathway. MBC with a germline or somatic mutation in BRCA1 or BRCA2 presents with deficiency in homologous recombination repair of double-stranded DNA breaks. Treatment of these BRCA-mutant cancers with PARPi leads to an accumulation of DNA damage and results in cell-cycle arrest and apoptosis. This effect is called synthetic lethality 3,5,6 . The second strategy is in combination with DNA-damaging chemotherapy, such as alkylators and/or topoisomerase I inhibitors. PARPi seem to sensitize tumor cells to damage caused by cytotoxic agents and may potentiate their effects.
PARPi when given as monotherapy or in combination with chemotherapy, are emerging treatment options in the metastatic setting for patients with breast cancer. Clinical activity is modest and selecting patients most likely to benefit from treatment is challenging. Here, we present a meta-analysis of the pooled response rates of PARPi for MBC. We explored the impact of tumor and patient level variables such as receptor status, type of combination agent and BRCA status on cancer outcomes in an attempt to better understand predictors of response in MBC.

Study selection
The search identified 2141 citations of which a total of 43 studies were included in the meta-analysis (see Fig. 1 for study selection schema). Data on risk of bias of each included study are presented in SUPPLEMENTARY Table 1.

Characteristics of the included studies
Included studies 7-49 comprised 2409 patients. Among these, 1798 (74.6%) patients had a germline (n = 1783) or a somatic (n = 15) BRCA mutation. Forty-eight percent of included patients were triple-negative (TNBC; n = 1146). In 10 of the included studies (28%, n = 680 patients), the PARPi was given in combination with a platinum-based chemotherapy. Sixty-four percent (n = 1550) of patients were treatment-naive in the metastatic setting and 20% (n = 472) were previously exposed to a platinum-based chemotherapy either in the early-stage or metastatic setting (see Table 1). Five studies were randomized comparative trials whereas the remaining 38 studies were non-comparative studies. Formal assessment of publication bias was limited by a small number of comparative trials. As such funnel plots (see Supplementary Figure  1) were not informative.

Meta-regression analysis
Univariable and multivariable meta-regression results are shown in the Table 3. When adjusted for covariates, having received previous platinum chemotherapy was negatively associated with ORR to PARPi among all MBC (p = 0.02). Similarly, there was a statistically significant negative association between ORR and platinum refractory disease (p = 0.03). There was a statistically significant positive association between ORR to PARPi and the proportion of patients with TNBC in individual trials (p = 0.03). However, no association was found for the proportion of patients with hormone receptor positive MBC (p = 0.33). While there was an association between BRCA mutation and ORR in univariable analysis, this association was not maintained after adjustment for other variables. While combination platinum, age, and treatment for first line MBC were associated with CBR in univariate metaregression, there were no statistically significant association with CBR when adjusting for covariates using multivariate metaregression.

DISCUSSION
In unselected MBC, our study shows that PARPi are associated with a pooled ORR of 45% and a CBR of 66%. This is slightly better than the reported ORR of 35-40% with first-line single agent chemotherapy such as anthracyclines 51 or taxanes 52 , 30% with carboplatin for BRCA mutated TNBC 53 and significantly better than the 13% with eribulin in the thirdline setting 54 . We also found a higher ORR (64%) and CBR (84%) when PARPi were combined with platinum-based chemotherapy versus when it was not (37% and 59%, respectively). This is concordant with the principles of synthetic lethality. Of note, meta-regression confirmed a significant association between previous platinum exposure and platinum refractory disease with lower response rate. This is an important observation especially as platinum resistance was an exclusion criteria for most of the PARPi RCTs performed in metastatic disease. Of note, this observation is concordant with the adjuvant Olaparib randomized phase III trial 55 , where the improvement in invasive disease-free survival was substantially lower among patients with prior exposure to platinum-based chemotherapy (n = 486, HR 0.77, 95% CI 0.49-1.21) compared to those without prior exposure to a platinum (n = 1350, HR 0.52, 95% CI 0.39-0.69).
Veliparib and talazoparib had the numerically highest response rates (ORR 51% and 51%, CBR 72% and 74%, respectively) and rucaparib the lowest (ORR 7% and CBR 38%), although these differences were not statistically significant. Response rates in patients with a BRCA1 and/or BRCA2 mutation were similar (ORR 57% vs 45% and CBR 73% vs 66%, respectively) and we found no association between mutation status and response rate in meta-regression analysis. These results might reflect the fact that vast majority of included patients had a BRCA mutation.
Of note, for the five largest trials included in this metaanalysis, pooled data did not demonstrate statistically significant PFS and/or OS improvement in unselected patients. The PFS result remained non-significant even after a sensitivity analysis was performed, excluding the one trial performed in non-BRCA mutated patients 34 . However, statistically significant PFS improvement was observed if trials were restricted to those with standard chemotherapy (i.e. excluding one study using temozolomide as the chemotherapy backbone). There was no improvement in OS in any analysis. This likely reflects the relatively modest incremental benefits of PARPi in MBC.
To our knowledge this is the largest pooled analysis of PARPi and outcomes in unselected MBC. Poggio et al 56 .
reported on a pooled analysis of only two RCTs and while results were similar to those reported here, the current Table 1 continued

(13)
A. Desnoyers et al. analysis provides greater precision for the estimates of the impact of PARPi in MBC while also supplementing the analysis with data from non-comparative, earlier phase trials. Another analysis by Wang et al 57 . aimed to compare two different PARPi rather than report on a comparison of PARPi to standard of care. This study has limitations. First, there was substantial heterogeneity in patient population and treatment exposures. Metaanalytic methods such as weighted pooling, subgroup and sensitivity analyses and meta-regression were utilized to explore these differences. As noted in the results, a few quantitative or statistically significant differences were observed including the impact of concurrent and prior platinum. Important negative findings were also identified including the lack of quantitative or statistically significant differences between different BRCA mutations, ER-expression and the different PARP inhibitors. Beyond these known potential sources of heterogeneity, we also explored other (and potentially unmeasured) confounders. Sensitivity analysis excluding studies which contributed to statistical heterogeneity showed similar effect sizes. While the strategy we utilized to address heterogeneity is robust and provides clinically useful and potentially actionable information, the potential for residual heterogeneity remains and this might impact the generalizability of our results. Second, we relied on summary trial data and metaregression rather than individual patient data. This will also add uncertainty to the results. Third, while we included many early phase trials, our reliance on published data will result in the potential for publication bias. Finally, we were unable to analyze publication bias robustly as the number of comparative trials was small. Second, we relied on summary trial data and metaregression rather than individual patient data. This will also add uncertainty to the results. Third, while we included many early phase trials, our reliance on published data will result in the potential for publication bias. Finally, we were unable to analyze publication bias robustly as the number of comparative trials was small.
In summary, PARPi have modest activity in MBC and this activity seems greater in combination with platinum than as monotherapy. There is lesser response to PARPi monotherapy in patients with prior platinum exposure, especially platinum resistance. Significant improvements in ORR translated to modest improvement in progression-free survival in trials using standard chemotherapy. However, overall survival was not improved. There was no association between ORR and type of BRCA mutation status. Further clinical trials are needed to better characterize which patient population could benefit most from these drugs.

Study inclusion criteria and strategy
A systematic review consistent with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria was performed 58 . We searched EMBASE and MEDLINE databases from inception to March 18, 2021. A targeted search was performed to identify clinical studies that reported response rate for PARPi in patients with metastatic breast cancer. Inclusion criteria comprised adult women with inoperable locally advanced or metastatic breast cancer treated with a PARPi in the palliative setting. Studies included in the current analysis met additional predefined eligibility criteria (see Table 4). The detailed search strategy is reported in SUPPLEMENTARY INFORMATION 1. Citations retrieved through the literature search were screened initially for inclusion based on their title and abstract. Full texts were obtained for citations that met the inclusion criteria where it was not possible to determine whether the study fulfilled inclusion criteria based on the abstract alone.

Risk of bias
Individual study risk of bias were assessed using the Cochrane risk of bias tools; RoB 2.0 tool for randomized trials and ROBINS-I tool for nonrandomized studies of interventions. Bias domains included the following: pre-intervention bias (selection and randomization), bias due to deviation from intended intervention, due to missing data, due to outcome measurement and bias due to selection of reported results (see SUPPLEMENTARY Table 1).

Study objectives
The objectives of this study were to pool and compare the activity and efficacy of PARPi for MBC in the overall population. Exploratory subgroups analyses were performed to investigate if activity differed based on hormone receptor status, BRCA mutation status, prior exposure to platinum-based chemotherapy, platinum sensitivity and choice of PARPi.

Statistical analysis
Data were reported descriptively as pooled proportion, mean, and range as appropriate. Analyses were weighted by sample size to compensate for the variability in sample sizes across studies. Odd ratios (OR) were calculated between subgroups: receptor status, BRCA mutations and prior platinum exposure. Analyses were performed using SAS Studio 3.8 (Institute Inc., Cary, NC). Meta-regression was used to explore the influence of previous chemotherapy and patient and tumor characteristics on ORR and DCR as reported in individual studies. Meta-regression comprised a linear regression weighted by individual study sample size and using a random effects model. The correlation coefficient was used to describe the relationship and direction of association between ORR and CBR with clinical variables. For the cohort of randomized trials, response rates and survival data from randomized controlled trials (RCTs) were combined in a meta-analysis using RevMan Software 5.3 (Cochrane Collaboration, Copenhagen, Denmark). In view of substantial clinical heterogeneity, estimates for ORs and HRs were pooled and weighted by generic inverse variance and computed by random effects modeling irrespective of statistical heterogeneity. Statistical heterogeneity was assessed using the Cochran's Q and I 2 statistics. Statistically significant heterogeneity was defined as a Cochran Q P0.10 or I 2 > 50%. Sensitivity analysis was performed to evaluate the robustness of the results by excluding studies with high risk of bias. Statistical significance was defined as P < 0.05. No corrections were applied for multiple significance testing.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

DATA AVAILABILITY
The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.