Molecular mechanisms and targets of new anticancer treatments

Abstract: The work presented in this thesis is an effort to decipher and understand the mechanism of action (MOA) of anticancer agents by building on and complementing chemical proteomics methods. The backbone of the thesis relies on a recent method called Functional Identification of Target by Expression Proteomics (FITExP) developed in Zubarev lab, where drug induced proteomic signatures are analyzed in various cell lines and top differentially regulated proteins with consistent behavior are determined, among which the drug target and mechanistic proteins are usually present. FITExP relies on the assumption that proteins most affected with a perturbation have a higher probability of being involved in that process. In this regard, Paper I aimed to enhance the performance of FITExP analysis by merging proteomic data from drug-treated matrix attached and detached cells. This is while the majority if not all proteomics and molecular biology experiments are performed in matrix attached cells, as the general belief is that detached cells lose their structural integrity and do not harbor valuable information. However, detached cells are those that are more sensitive to chemotherapeutics and might reflect the proteome changes better. The comparative proteomics of living and dying cells improved FITExP performance with regards to identification of targets and provided insight about proteins involved in cellular life and death decisions. Furthermore, the orthogonal partial least squares-discriminant analysis (OPLS-DA) paradigm presented in this study, was used throughout the thesis for contrasting and visualizing the proteomic signature of a molecule against others, to reveal targets and specific proteins changing in response to the molecule of interest. In Paper II, as a further development of FITExP and to demonstrate its applicability in a broader context, we built a proteome signature library of 56 clinical and experimental anticancer agents in A549 lung adenocarcinoma cell line. This resource called ProTargetMiner can be used for different purposes. The proximity of compounds in hierarchical clustering or t-SNE could be used for prediction of the mechanism of new compounds. Contrasting each molecule against other treatments using the OPLS-DA scheme presented in Paper I, revealed drug targets, mechanistic proteins, resistance factors, drug metabolizing enzymes and effects on protein complexes. Representative examples were used to demonstrate that the specificity factors extracted from the OPLS-DA models can help identify subtle but biologically significant processes, even when such an effect is as low as 15% fold change. Furthermore, we showed that the inclusion of 8-10 contrasting molecules in the OPLS-DA models can produce enough specificity for drug target deconvolution, which offered a miniaturization opportunity. Therefore, we built three deeper datasets using 9 compounds that showed the most diverse proteome changes in the orthogonal space in three cell lines from major cancer types: A549 lung, MCF-7 breast and RKO colon cancers. These datasets provide a unique depth of 7398, 8735 and 8551 respectively, with no missing values. Subsequently, a Shiny package was created in R, which can employ these datasets as a resource and merge it with user data and provide OPLS-DA output and target deconvolution opportunity for new compounds. Finally, using the original ProTargetMiner data, we also built a first of its kind proteomic correlation database which can find applications in deciphering the function of uncharacterized proteins. Moreover, the resource helped to identify a set of core or untouchable proteins with stable expression across all the treatments, revealing essential functions within the cells. Such proteins could be used as house-keeping controls in molecular biology experiments. In paper III, we combined FITExP with other chemical proteomics tools Thermal Proteome Profiling (TPP) and multiplexed redox proteomics, to study the target and mechanism space of auranofin. This would also allow to assess the power, orthogonality and complementarity of these techniques in the realm of chemical proteomics. TPP is a recently developed technique that can monitor changes in the stability of proteins upon binding to small molecules. Redox proteomics is a method by which the oxidation level of protein cysteinome can be quantitatively analyzed. Auranofin is an FDA-approved anti-inflammatory drug for treatment of rheumatoid arthritis, but due to its potent antitumor activity, it is currently in clinical trials against cancer. Although several MOAs have been suggested for auranofin, uncertainties exist regarding its cellular targets; therefore, this molecule was chosen as a challenging candidate to test the chemical proteomics tools. A combination of the above mentioned tools confirmed thioredoxin reductase 1 (TXNRD1) (ranking 3rd) as the cognate target of auranofin and demonstrated that perturbation of oxidoreductase pathway is the main route of auranofin cytotoxicity. We next showed that changes in the redox state of specific cysteines can be linked to protein stability in TPP. Some of these cysteines were mapped to the active sites of redox-active enzymes. In Paper IV, using quantitative multiplexed proteomics, we helped to show that b-AP15, a bis-benzylidine piperidone compound inhibiting deubiquitinases USP14 and UCHL5, produces a similar perturbation signature as bortezomib in colon cancer cells. However, in comparison with bortezomib, b-AP15 induces chaperone expression to a significantly higher level and leads to a more extensive accumulation of polyubiqutinated proteins. The polyubiqutinated proteins co-localize with mitochondrial membrane and subsequently reduce oxidative phosphorylation. These results help define the atypical cell death induced by b-AP15 and describe why this molecule is effective against apoptosis resistant cells in variety of tumor models. Finally, in Paper V, we extended the applications of TPP and combined it with specificity concept for proteome-wide discovery of specific protein substrates for enzymes. We developed a universal method called System-wide Identification of Enzyme Substrates by Thermal Analysis (SIESTA) that relies on the hypothesis that enzymatic post-translational modification of substrate proteins can potentially change their stability against thermal denaturation. Furthermore, we applied the concept of specificity similar to the above papers, to reveal potential substrates using OPLS-DA. SIESTA was applied to two enzyme systems, namely TXNRD1 and poly-(ADP-ribose) polymerase-10 (PARP10), identifying known and putative candidate substrates. A number of these candidate proteins were validated as PARP10 substrates by targeted mass spectrometry, chemiluminescence and other assays. SIESTA is an unbiased and system wide approach and its broad application can improve our understanding of enzyme function in homeostasis and disease. In turn, specific protein substrates can serve as readouts in high throughput screening and facilitate drug discovery. Taken together, in this thesis, FITExP methodology was improved in two directions. In paper I, we improved the performance of FITExP by combining the proteomics data from detached and attached cells. In Paper II, we demonstrated how the proteomics data on a multitude of drugs in a single cell line enables the discovery of compound targets and MOA. Furthermore, we built an R Shiny package which can serve as a resource for the cancer community in target and MOA deconvolution. In Papers III and IV, we applied an arsenal of chemical proteomics tools for characterization of two anticancer compounds. In Paper V, we expanded the applications of TPP to identification of specific protein substrates for enzymes in a system-wide manner.