On analysis of protein function and variation

Over the last fifteen years, next-generation sequencing has overcome time and cost constraints to be widely used to generate a plethora of biological data. Around 200 million sequences are currently stored in one of the biggest protein databases, UniProt. However, less than 1% of UniProt sequences have functions supported by experimental evidence. This thesis focuses on increasing knowledge about protein functions and demonstrating why it is essential to do so. The first aim is fulfilled by a project that used computational approaches to assign functions to the proteins of unknown function within the minimal bacterial genome. Synthesising the minimal cell was completed in 2016, and it revealed that 149 genes out of 473 had unknown functions. Our study demonstrated that using several protein function prediction methods that each explores different protein properties is an effective way to annotate unknown genes of the minimal cell. As a result, 133 out of 149 genes were assigned a function. This included 66 proteins, for which we identified more informative functions than predicted by Hutchison et al. in the initial study from 2016 (Hutchison et al., 2016). This thesis's second goal is to show how important it is to expand our knowledge about protein functions and have access to good protein function prediction methods to apply them where there is not experimental functional information. This thesis focuses specifically on applying protein function prediction methods to study the impact of DNA mutations on the proteins, which would hopefully lead to a better understanding of acquired resistance to anti-cancer therapies, which remains one of the biggest obstacles in treating cancer patients. Acquired drug resistance is developed through alterations in different molecular mechanisms such as drug efflux or binding of a drug to the target, which can sometimes be caused by a mutation in a single protein. Here, whole-exome sequencing data of the acute myeloid leukaemia cell lines Molm13 and four Molm13 sub-lines adapted to nutlin-3 was studied to identify potential drivers of resistance. Additionally, 41 UKF-NB-3 (a neuroblastoma cell line) sub-lines adapted to tubulin-binding agents were analysed with a focus on the clonal composition of cancer cells and its impact on developing different resistance mechanisms. The analysis of de novo variants in the Molm13 sub-lines adapted to nutlin-3 revealed that three out of four sub-lines acquired loss-of-function mutations in TP53 commonly associated with resistance to MDM2 inhibitors. Additionally, all four Molm13 sub-lines demonstrated an increased sensitivity to cytarabine that may be connected to likely deleterious de novo mutations identified in three out of four sub-lines in SAMHD1, which cleaves the triphosphorylated active form of cytarabine, causing its deactivation, while its natural function is to cleave deoxynucleoside triphosphates (dNTPs) into deoxyribonucleosides and inorganic triphosphate with the main goal of restricting viral infections by reducing dNTPs’ cellular levels. These results identify SAMHD1 mutations as a candidate biomarker for cytarabine sensitivity after the failure of MDM2 targeted therapies, which is consistent with studies demonstrating that low SAMHD1 activity, likely caused by lower SAMHD1 expression or deleterious mutations, generally tends to be associated with an increased cytarabine sensitivity in AML cells (Schneider et al., 2017). Subsequently, the analysis of de novo variants in the UKF-NB-3 sub-lines adapted to tubulin-binding-agents demonstrated that different sub-lines adapted to the same drug can share many of the same de novo variants, which shows that they may have come from a similar clone. However, this is not always the case. The results revealed that different subpopulations could be selected upon the repeated adaptation of the same cancer cell line to the same drug. This emphasises the heterogeneity of processes underlying acquired resistance to anti-cancer therapies and demonstrates the need to identify these processes' biomarkers.