Florentina Troger

Title of the Doctoral Thesis: Computational prediction of mitochondrial toxicity with a special focus on structure-based methods.

Publishing year: 2020

Tags: mitochondria / mitochondrial toxicity / toxicity / structure-based methods

Follow me on Twitter

I agree that content from Twitter will be displayed to me.

 


Abstract

Mitochondria are known as the power-plants of cells, since their major role is to produce a vast amount of energy in the form of adenosine triphosphate (ATP). Therefore, compounds which impair mitochondrial function are known to cause severe adverse effects which can result in various illnesses such as Alzheimer’s or Parkinson’s disease, cancer and liver injury. For this reason, mitochondrial toxicity is increasingly recognized as a safety concern in drug development. Due to the increasing amount of high resolution crystal-structures of mitochondrial proteins, structure-based approaches are promising techniques to investigate and to predict mitochondrial toxicity. The knowledge of protein-ligand interactions, as well as the understanding of the underlying molecular binding mechanisms, can assist hazard analysis and toxicity predictions. The aim of this thesis is, on one hand, to identify distinct chemical substructures that are related to mitochondrial toxicity. On the other hand, to decipher the protein-ligand-interactions of mitochondrial proteins such as complex I and III. Finally, the gained knowledge will be employed for toxicity predictions. In the first study we focused on analyzing a dataset for mitochondrial toxicity regarding their physicochemical properties and their sub-structures. In this study, we not only created machine learning models based on the compiled dataset, but also identified structural alerts to obtain mechanistic information on toxicity for specific sub-structures. The second and the third study focused on proteins belonging to the mitochondrial respiratory chain. The mitochondrial respiratory chain is responsible for ATP-synthesis, which is essential for most living organisms. It is the target of many pesticides which, due to the high sequence identity of the respiratory chain among different species, also cause hazardous effects in humans. The second study is based on an adverse outcome pathway, which links complex I inhibition to parkinsonian motor deficiency. By docking the two known complex I inhibitors and pesticides rotenone and deguelin, pharmacophore models were obtained. The generated pharmacophore models were then further used for virtual screening of DrugBank and Chemspace. Together with predictions from the machine learning models published in study one, we were able to create a workflow for predicting complex I toxicity. This workflow was validated by experimental testing. We showed that three out of twelve chosen and experimentally tested compounds were specific complex I inhibitors. The third study focused on the mitochondrial respiratory complex III. Due to the availability of crystal structures containing co-crystallized ligands, we used the bovine structures for the structure-based modeling. Here, a similar workflow as created in the second study was applied and slightly adapted. The pharmacophore models were created based on the co-crystallized ligands in the structures. Subsequently, these pharmacophores were used for a virtual screening of DrugBank and, in addition, all compounds of DrugBank were also docked into the distinct structures. For refining the hit list, two machine learning models of the first study were implemented. By applying consensus scoring based on the pharmacophore models, the docking results and the predictions by the machine learning models, we were able to select a set of compounds ready for experimental testing. In the fourth study we focused on a member of the SLC25 family, the mitochondrial carriers. SLC25A12 is an aspartate/glutamate transporter which is known to be overexpressed in hepatocellular carcinoma. Furthermore, mutations on SLC25A12 are linked to hypomyelination. Due to the lack of a crystal structure the first part of this study was to create a homology model based on the homologous crystal structures PDB 1Ds 1OKC and 6GCI. Further, the homology models were used to characterize key structural determinants of binding of glutamate and aspartate by performing docking experiments. In this way, insights into the binding-mode of these compounds could be obtained, which can guide the design of new inhibitors. To conclude, we were able to show in all our studies that structure-based methods are not only capable of predicting mitochondrial toxicity, but also suited to obtain new insights into the underlying molecular mechanisms.