Philip Schouten

Inactivating mutations in the BRCA1 and BRCA2 genes predispose to breast and ovarian cancer. A defect in error-free DNA double strand break repair through homologous recombination is thought to contribute largely to this predisposition as genomic instability ensues that facilitates cancer cell transformation. However, this deficiency may open therapeutic options as it has been identified that treatment with double strand break inducing agents kills such BRCA deficient cells. Furthermore, synthetic lethality between BRCA deficiency and Poly(ADP)Ribose polymerase inhibitors offers another treatment option that has been introduced clinically over the last years. To treat these BRCA deficient patients it is necessary to identify and validate biomarkers that detect this defect and treatments that target this defect. This thesis describes the identification, validation (retrospective and prospective) and improvement of biomarkers and treatments for patients with BRCA deficient tumors.

Previously it was found that one such biomarker, the BRCA-like array Comparative Genomic Hybridization classifier identifies breast cancer patients that might benefit from high dose alkylating chemotherapy. This classification determines whether the aberrations in a breast cancer are similar to those found in BRCA-mutated cancer or not and thus that there may be an underlying defect in BRCA1 or BRCA2 and thus a defect in DNA repair. For this already discovered biomarker the technical validations comprised of translation from Bacterial Artificial Chromosome arrays to oligonucleotide arrays (Chapter 1) and low coverage whole genome sequencing (Chapter 2). In these analyses we used tumor samples that were hybridized on both the BAC platform as well as a new platform. We extended these in-house platforms with an analysis of the validity of classification with data obtained by 5 other copy number profiling techniques (Chapter 2). Employing the same dataset of paired samples as in Chapter 2 we further refined the processing of data between platforms correcting for differences in centering and dynamic range between platforms (Chapter 3).

In chapter 1 we observed that mapping data obtained with oligonucleotide array CGH to the BAC platform locations by averaging the oligonucleotide measurements that overlapped with the BAC clone resulted in high kappa values, which describe inter-test variability of the BRCA1-like and BRCA2-like classifier obtained by the two techniques. Also the predictive effect for the benefit of high dose alkylating chemotherapy was retained. We thus concluded that we could use the oligonucleotide-mapped data for further studies to validate that BRCA-like cancers can be treated with high dose alkylating chemotherapy.

In chapter 2 we translated low coverage whole genome sequencing to the BAC platform locations in a similar fashion as we did for the oligonucleotide to BAC conversion. This proved successful exemplified by high kappa values and the retention of the prediction of benefit of high dose alkylating chemotherapy for BRCA1-like patients. We concluded that mapping low coverage whole genome sequencing data can also be used as input for classification. Given the robust image of copy number aberration profiling in general we investigated 5 more techniques, leading to an overall analysis of 7 platforms. In this overall analysis we changed the gold standard from BAC classification to consensus classification between all techniques to include all available information and dampen the effect of lower quality profiles, in particular that of the lower resolution platforms. We observed kappa values of moderate-high to high concordance for most techniques. Therefore we concluded that data from other platforms can be used, if processing leads to good quality data that resembles the characteristics of the BAC platform.

In chapter 3 we extended the findings of robustness of classification between platforms by taking into account the centering and dynamic range of the platforms. Correction factors were obtained by fitting a linear model between sorted average data of the original platform and the new platform. This correction improves the kappa values for most classifiers, which led us to conclude that preferably samples that are re-hybridized on both techniques are employed to find these correction factors.

The biological validation comprised investigation of whether the BRCA-like classifiers identify patients with tumors that are BRCA mutated. In chapter 3 we performed a cross-tumor, cross-mutation analysis of BRCA1-mutated, BRCA2-mutated and non-mutated breast and ovarian cancers. We trained classifiers between BRCA mutated and non-mutated cases using various combinations of tumor type and mutation status as definition of BRCA-mutated. We observed that high similarity in the average profiles of BRCA1-mutated and BRCA2-mutated high grade serous ovarian cancers. In breast cancer, however, clear differences were present in the average profiles of BRCA1 and BRCA2-mutated cancers that may partially vary with hormone receptor status. Employing the prediction of benefit of high dose alkylating chemotherapy as readout of BRCA deficiency we observed that specific classifiers demonstrated expected behavior. E.g. breast cancer specific classifiers predict benefit of high dose chemotherapy and ovarian cancer classifiers do not predict in breast cancer samples. However, we also found that we were unable to predict all (BRCA-mutated) samples correctly. Whether this is a technical or biological issue is unclear. Although we trained various classifiers that predict BRCA mutation status and benefit of high dose chemotherapy, further validation is required. This validation could be aimed two-fold: 1) to clarify prediction of the BRCA deficiency, and 2) the predictive value for benefit of BRCA-deficiency-targeting agents. The latter validation is particularly relevant for ovarian cancer classifiers, because we lacked a dataset to assess predictive value.

The clinical validation was aimed to falsify the hypothesis that BRCA1-like tumor status predicted benefit from high dose alkylating chemotherapy in breast cancer. In chapter 4 we investigated a balanced cohort of high risk breast cancer patients that were treated with a conventional chemotherapy regimen (cyclophosphamide-methotrexate-5-fluorouracil or epirubicin-cyclophosphamide or 5-fluorouracil-epirubicin-cyclophosphamide) or tandem high dose alkylating chemotherapy (induction with ifosfamide-epirubicin, followed by 2 cycles ifosfamide-epirubicin-carboplatin and autologous stem cell transplantation). We observed that BRCA1-like patients that were treated with high dose chemotherapy had 6 times lower death rates than BRCA1-like patients that were treated with conventional chemotherapy (hazard rate: 0.15, 95% CI: 0.03-0.83, p=0.03). This difference was not observed in non-BRCA1-like patients (interaction test p: 0.045). We concluded that BRCA1-like status was a predictive biomarker for this regimen of high dose chemotherapy.

In chapter 5 we investigated the biomarker-treatment interaction of BRCA1-like patients in the randomized trial WSG-AMON, which randomized patients between dose dense (epirubicin-cyclophosphamide followed by cyclophosphamide-methotrexate-5-fluorouracil), and tandem high dose chemotherapy (epirubicin-cyclophosphamide followed by 2 cycles epirubicin-cyclophosphamide-thiotepa and autologous stem cell transplantation). BRCA1-like patients treated with tandem high dose chemotherapy had a 5 times lower risk of events than BRCA1-like patients treated with dose dense chemotherapy (HR 0.2, 95% CI 0.07-0.63, p=0.006). This difference was not observed in non-BRCA1-like patients (p for interaction: 0.003). We concluded that BRCA1-like status was a predictive biomarker for high dose chemotherapy, and that high dose alkylating chemotherapy yields improved outcomes, even if compared to a dose-dense regimen.

In chapter 6 we aimed to better delineate the group of patients that benefit from high dose chemotherapy by incorporating mechanisms of resistance to therapy that were identified in preclinical model systems. XIST and 53BP1 expression were identified as resistance markers in BRCA1-deficient mouse (cell) models. Instead of testing these biomarkers in the general population, as was done previously, we employed them within patients with a BRCA1-like tumor, which represents the modeling conditions better. We found that BRCA1-like patients with low expression of 53BP1 or high expression of XIST did not derive benefit from the high dose alkylating chemotherapy regimen. In adjusted survival analysis we found that patients predicted to be sensitive had 7 times less chance of an event compared to patients that were predicted to be resistant (hazard rate 0.14, 95% CI 0.03-0.71, p=0.018, p for interaction: 0.22). These markers may be used as predictive biomarkers in BRCA1-like breast cancer, although validation is required due to the small number of patients.

Double strand break inducing chemotherapy is one class of agents that may be used to target BRCA-deficient breast cancers. Poly(ADP)Ribose Polymerase inhibitors are a more recently identified class of agents that was found to be synthetically lethal with BRCA1 or BRCA2 deficiency. The combination of PARP-inhibition with olaparib and carboplatin (a double strand break inducer) was found to be feasible and also led to long term survival in early clinical studies. In chapter 7 we describe the trial protocol for a phase 1 followed by randomized phase 2 trial which will compare carboplatin-olaparib followed by olaparib monotherapy against the currently registered standard capecitabine as first line treatment in BRCA1 or BRCA2 mutated metastatic breast cancer. Phase 1 is a dose escalation study of the tablet formulation of olaparib combined with 2 cycles carboplatin, to determine the maximum tolerated dose of this combination that likely has bone marrow depression as dose limiting toxicity. Subsequently, in phase 2 patients will be randomized in first line to 2 cycles carboplatin-olaparib dosed according to the findings in phase 1 followed by the monotherapy dose olaparib versus capecitabine. In second line, patients will be administered capecitabine or vinorelbin or paclitaxel or eribulin, by physician’s best choice. The aim of this study is to determine whether carboplatin-olaparib should become the new standard first line option in BRCA-mutated breast cancer.

In conclusion, the studies in this thesis describe biomarkers to identify patients with a BRCA-deficient cancer and treatments that could be employed to improve outcomes. These studies covered the various aspects of biomarker discovery: technical validation, biological validation and clinical validation. Furthermore, various stages in biomarker discovery are covered, from the initial discovery phase to clinical trial. As such, these studies have led to increased level of evidence and prospective validation of BRCA1-like-high dose alkylating chemotherapy biomarker-treatment combination and olaparib-carboplatin in BRCA-deficient breast cancer, insights in various biological and technical aspects of BRCA-like classifiers, ovarian cancer BRCA-like classifiers and the application of additional preclinical identified biomarkers. In the larger picture, these studies are examples of precision oncology, the school of thought aiming to improve treatment outcomes by matching the treatment of cancer to the molecular aberrations present.