Identification of tumour molecular markers that predict PARP inhibitors response

The process of cellular duplication occurs in all living organisms and forms the basis for biological inheritance. This mechanism involves DNA replication that, being crucial for the preservation of genetic material, counts a great number of control mechanisms to verify its correctness. Discontinuities in a strand of the DNA double helix, known as single-strand breaks (SSBs), can sometimes occur and, if not repaired appropriately, could pose a serious threat to genetic stability and cell survival. Consequently, cells have evolved efficient mechanisms for their repair, involving Poly ADP-ribose polymerase (PARP), a family of proteins involved in DNA repair and apoptosis whose main role is to detect and signal SSBs to the enzymatic machinery involved in their repair (SSBR). The importance of this process is highlighted by the fact that unrepaired SSBs could cause in proliferating cells the blockage or collapse of DNA replication forks, leading to the formation of double-strand breaks (DSBs). Although cells possess a remarkable capacity to accurately repair such DSBs using homologous recombination (HR), acute increases in cellular levels of SSBs could saturate this pathway, leading to genetic instability and/or cell death. Defects in HR repair mechanism, caused by mutations in the pathway, confer the so-called "BRCAness" phenotype. Inactivation of these genes causes HR deficiency (HRD), resulting in high levels of genomic abnormalities. Recent studies have demonstrated the importance of a good predictor of the biological status of an HR-deficient tumour: the set of tumours that show BRCAness and that can be selectively sensitive to PARP inhibitors includes a wide range of sporadic breast, ovarian and also colorectal, cancers. All these features led Nik-Zainal's research group to use a weighted model, called "HRDetect", to identify mutational signatures predictive of BRCA deficiency. This type of analysis requires the tumour and the matched normal samples to correctly compare and extract the somatic variations caused by the tumour, discarding the germinal ones. The work carried out within this thesis is aimed at overcoming this limitation, being able to correctly predict the score provided by the HRDetect algorithm in situations in which the normal sample is not available (e.g. cell lines). Through the introduction of a new '’metanormal’' sample it has been possible to unlink the tumour sample from the matched normal, replacing it in the comparison. It follows that, since the tumour is not "matched" with the respective normal, series of germline mutations are not properly filtered as such appearing in the set of extracted somatic variations. To correctly emulate the expected HR scores, obtained by direct comparison between the tumour sample and the normal matched sample of the considered patient, different strategies were developed. These latter made it possible to extract somatic mutations increasingly accurately, reducing the prediction errors related to the presence of germline mutations. Each strategy was applied to all the breast cancer samples by running the algorithms via the Linux shell, supported by the Python programming language and the interpreted language 'AWK'. The application of the final strategy correctly predicted HR deficiency/proficiency in 98.6% of the 77 samples considered. The consistency of these results, through future implementation of the strategy to a larger number of samples, could contribute significantly to biomedical research.