< Back to previous page

Project

Towards a better selection of systemic treatment for patients with ER-positive HER2-negative breast cancer: the use of prognostic and predictive markers in early and metastatic setting

Worldwide, breast cancer represents about 25% of all cancers in women. There is extensive research effort into this highly complex and heterogeneous disease comprising many different histological and molecular subtypes with luminal, oestrogen receptor (ER)-positive human epidermal growth factor receptor 2 (HER2)-negative tumours being the most frequent subtype. Even within this subtype, there is a lot of heterogeneity, which complicates accurate patient selection for therapeutic decision-making. Patients with luminal early-stage breast cancer receive endocrine treatment for at least 5 years after surgery as standard of care. Some patients also benefit from the addition of adjuvant chemotherapy (aCT). As chemotherapy has many side effects on the short- and long-term, we need to correctly identify the patients who would really benefit from aCT. Multiple demographic and clinical-pathological features are taken into account for aCT recommendation. In addition to these markers, the molecular biology of the tumour is gaining importance in the decision-making through the development of multigene signatures (MGS) tests. These MGS, such as MammaPrint® (MP) and Oncotype DX® (ODX), estimate the risk of recurrence based on the expression of genes in the tumour, and have already proven their clinical utility. However, several factors impede the broad use of these MGS, such as its high cost, the lack of consensus on preselection of a defined target population and the lack of local, decentralised tests that can be integrated in existing workflows in molecular diagnostic laboratories.

In this thesis, our general aim was to improve the selection of systemic treatment in patients with ER-positive HER2-negative breast cancer by using prognostic algorithms and specific biomarkers in early and metastatic setting. In the early setting, we retrospectively investigated the administration of aCT in Belgium, we evaluated the impact of MGS in the recommendation of aCT during the multidisciplinary meeting (MDM), and we benchmarked several statistical models that are predictive for MGS outcome to assist clinicians during the MDM discussions whether to perform an MGS.

First, we studied the difference in aCT administration in clinical low- and clinical high-risk ER-positive HER2-negative early-stage Belgian breast cancer patients, diagnosed in 2008 versus 2014. In addition, we studied the difference in aCT administration across the Belgian regions (Flemish, Brussels-Capital and Walloon region). We observed a significant reduction in aCT administration for both risk groups when comparing the two periods and were able to confirm this observation for the clinical high-risk patients diagnosed in 2013-2015 versus 2007-2009 at University Hospitals Leuven (UHL). The reduction did not change the 5-year survival in any of the patient groups. Additionally, we observed regional differences in aCT administration in the clinical low-risk group in 2014, with more aCT administration in the Brussels-Capital and Walloon regions versus the Flemish region.

The differences in aCT administration observed across Belgian regions illustrates the need for standardised criteria allowing accurate patient selection for aCT and potentially MGS testing. To this end, we evaluated the impact of different MGS tests in assisting decision-making during the MDM for aCT administration in luminal early-stage breast cancer patients with uncertain risk of recurrence based on demographic and clinical-pathological parameters. The integration of MGS results into the MDM recommendations resulted in a decisional switch in 47% of cases and a 9% absolute reduction in aCT administration. However, when we considered each MGS separately, we observed the reduction in the administration of aCT with MP and ODX but not with Prosigna®.

Unfortunately, not every country has access to MGS tests as these are quite expensive. Moreover, the analysis in a centralised test laboratory located abroad can prevent MGS use because of privacy regulations. We therefore aimed at increasing patient accessibility to MGS by prospectively validating a targeted RNA-based next-generation sequencing (NGS) kit, MP/BluePrint® (BP), in a decentralised setting. We compared raw data generated by MP/BP NGS at UHL and Curie Institute Paris, with those obtained centrally at Agendia by NGS and by the gold standard microarray technique. We also compared breast cancer subtypes by molecular MP and BP targeted RNA sequencing with immunohistochemistry (IHC). We showed high concordance between risk results obtained at two independent laboratories and the central laboratory at Agendia. We also observed a high concordance between NGS and microarray molecular subtyping, indicating a successful translation of the MP and BP microarray test to the MP and BP NGS test, in a decentralised setting.

However, it currently remains unclear which patients benefit most from MGS due to the lack of consensus on preselection of a defined target population. In order to assist the recommendation for MGS testing of the clinicians during the MDM, we aimed to analyse the possibility of preselecting patients for MGS testing. We determined the concordance between the outcomes of existing inexpensive statistical models, developed for predicting the outcome of MGS, and MGS risk outcomes. All cases predicted as being high risk by either the Magee average equation or the breast cancer recurrence score estimator, had a high risk outcome with the MGS. This could result in a 5% reduction in MGS testing. We further investigated the possibility of developing our own predictive model for risk assessment by the MP/BP MGS using machine learning. As the MP/BP MGS is the most widely used in Belgium, preselection of patients through such a model could further assist the selection of patients during the MDM who could benefit from MGS testing. We considered including the risk outcome of existing predictive statistical models based on multiple clinical-pathological features, mainly developed for ODX, into our own predictive model. We also considered using the clinical-pathological features themselves, and the combination of the existing models with the clinical-pathological features into a single model. These three feature sets were tested as input for a new model based on machine learning. Support vector machine (SVM), random forest (RF), linear discriminant analysis (LDA) and decision tree (DT) machine learning models were tested. The best performance was obtained through DT, resulting in accuracies up to 77.8% with a sensitivity of 83.3% (15/18 true high risk) and a specificity of 72.2% (13/18 true low risk). Our model still needs some fine-tuning before it can be validated in a larger cohort. In addition, these results should be tested further to prove its clinical applicability.

In the metastatic setting, chemotherapy is not the treatment of choice for patients with luminal breast cancer and is typically only used as a last resort when there are no other options. Cyclin-dependent kinases 4 and 6 (CDK4/6) inhibitors, in combination with endocrine therapy in first or second line of treatment, are currently standard of care in this setting. With the approval by the United States food and drug administration (FDA) and the European medicines agency (EMA) of alpelisib, an α-specific phosphatidylinositol 3-kinase (PI3K) inhibitor which showed benefit in patients with a PIK3CA mutation, testing for these mutations has gained importance. However, in the metastatic setting, a tissue biopsy is not always feasible or available. Therefore, we implemented a highly sensitive droplet digital PCR assay (ddPCR) for the detection of PIK3CA hotspot mutations in circulating cell-free DNA (cfDNA) from plasma. We showed a good performance of ddPCR, especially when 4ml plasma was used in combination with a semi-automated cfDNA extraction method (Maxwell). A good correlation between mutation status obtained by plasma-based ddPCR and tissue-based NGS analysis was obtained, assuming that this technology can be safely used in diagnostic practice when tumour tissue is not available.

In conclusion, our results demonstrated the need for standardised criteria to determine aCT administration for luminal early-stage breast cancer patients in Belgium. In addition, we showed that MGS testing assists in therapeutic decision-making, and can reduce aCT administration in clinical high-risk patients. The decentralisation of MGS testing in local laboratories equipped with state-of-the-art technologies for molecular diagnostics can improve access to MGS testing while complying with privacy regulations and reducing the costs of the test. This decentralisation, and the use of inexpensive statistical models predictive for MGS outcome, could further improve the selection for aCT administration and MGS testing in patients with luminal early-stage breast cancer. Finally, we provided evidence for the clinical application of the highly sensitive ddPCR as a useful screening method to detect PIK3CA hotspot mutations in patients with metastatic luminal breast cancer for which no representative tumour tissue is available.

Date:1 May 2017 →  14 Oct 2021
Keywords:Breast cancer, Prognostic algorithms, Predictive biomarkers
Disciplines:Endocrinology and metabolic diseases, Gynaecology and obstetrics, Nursing
Project type:PhD project