179P - Organ-independent prediction of BRCA1 mutation-like status based on copy number gains and losses in breast and ovarian cancer

Date 28 September 2014
Event ESMO 2014
Session Poster Display session
Topics Ovarian Cancer
Breast Cancer
Translational Research
Presenter Philip Schouten
Citation Annals of Oncology (2014) 25 (suppl_4): iv58-iv84. 10.1093/annonc/mdu326
Authors P. Schouten1, D.J. Vis2, E. van Dijk2, C. van Deurzen3, A. Jager4, E.M. Berns5, H.H. van Boven6, H. Hereditary Breast And Ovarian Cancer Research Group Netherlands7, M. Rookus8, P.M. Nederlof9, L.F. Wessels2, S. Linn10
  • 1Molecular Pathology, The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, 1066 CX - Amsterdam/NL
  • 2Molecular Carcinogenesis, The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, 1066 CX - Amsterdam/NL
  • 3Pathology, Erasmus MC – Cancer Institute, 3015CE - Rotterdam/NL
  • 4Medical Oncology, Erasmus MC – Cancer Institute, 3015CE - Rotterdam/NL
  • 5Medical Oncology, Erasmus MC – Cancer Institute, Rotterdam/NL
  • 6Pathology, Antoni van Leeuwenhoek Hospital, Amsterdam/NL
  • 7Hebon, Hereditary Breast And Ovarian Cancer research group Netherlands, HEBON/NL
  • 8Psychosocial And Epidemiological Research, The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, 1066 CX - Amsterdam/NL
  • 9Diagnostic Oncology, The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, 1066 CX - Amsterdam/NL
  • 10Division Of Medical Oncology, The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, 1066 CX - Amsterdam/NL

Abstract

Aim

BRCA1 is important in maintaining genomic stability by error-free repair of DNA double strand breaks by homologous recombination. Lack of a functional BRCA1 gene results in genomic instability with characteristic gains and losses in DNA copy number variation (CNV) profiles in breast tumors. We previously developed and used this breast cancer BRCA1-like signature to identify patients that benefit from double strand break inducing chemotherapy. In the current study we set out to train an ovarian and ovarian + breast classifier for BRCA1-like tumor status.

Methods

We used our own and public CNV profiles of human BRCA1 mutated and non-mutated primary breast and ovarian cancer. We trained logistic regression models for prediction of BRCA1-like status in breast and ovarian cancers separately and in both types combined. For this step we used the area under the curve (AUC) of the receiver operating characteristic curve to analyze predictive capacity. We identified regions of CNVs that associate with BRCA1 mutation status by comparing CNV profiles of BRCA1 mutated and non-mutated cancers within and between the two organs of origin.

Results

CNV profiles contained significantly different regions of gains and losses, both within and between (combinations) of organ and mutation status. In a 10-fold double cross validation run, the organ-specific and organ-independent classifiers predicted BRCA1 status based on CNV profiles with a mean AUC (+/- standard error of the mean) of 0.81 (+/- 0.05) for breast and 0.74 (+/- 0.07) for ovarian and 0.83 (+/- 0.04) independent of organ. Organ dependent classifiers predicted worse on the other organ. Organ-independent differences between BRCA1 mutated and non-mutated CNVs were present on chromosomes 3 to 7, 9, 10-17, 20 and X.

Conclusions

We identified specific CNVs, both gains and losses, associated with BRCA1 mutation status, that overlap in breast and ovarian cancer. We trained an ovarian and breast + ovarian BRCA1 classifier which we will further validate as predictive biomarker for double strand break inducing chemotherapy and Poly(ADP)Ribose Polymerase 1 inhibitor targeted therapy in randomized trials.

Disclosure

P.M. Nederlof: named inventor on patent application for BRCA1 like breast cancer classifier; S. Linn: named inventor on patent application for BRCA1-like breast cancer classifier. All other authors have declared no conflicts of interest.