ИСТИНА

Trypanosoma cruzi (T. cruzi) is a parasitic protozoan responsible for Chagas disease, or American trypanosomiasis. This parasite is primarily transmitted through blood-sucking insects inhabiting rural regions of South America, and it can lead to fatal consequences if left untreated. Currently, two drugs, nifurtimox and benznidazole, are known to be effective against T. cruzi. However, their application is limited by severe side effects, debatable efficacy, and developed resistance. The final step of vitamin C biosynthesis in trypanosoma is catalyzed by the enzyme TcGAL, or L-galactonolactone dehydrogenase from T. cruzi (EC 1.3.2.3). Unlike humans, who get vitamin C with food, the parasite relies on vitamin C synthesized by TcGAL and is unable to uptake it from external sources. Therefore, TcGAL might be an attractive target for novel drugs. However, TcGAL is a membranotropic enzyme, and its inherent instability in vitro is a substantial challenge. To overcome this obstacle, in our group TcGAL was successfully stabilized within micelles, mimicking its natural membrane environment. This approach enables the exploration of TcGAL functions and kinetic properties using various substrates and inhibitors. The limited throughput of the experiment makes in silico screening essential. Since the experimental structure of TcGAL is unknown, we employed AlphaFold and classical homology modeling to construct a full holoenzyme structure, including the cofactor FAD. Subsequent preliminary molecular dynamics (MD) simulations revealed a tunnel inside the protein, and ensemble docking with known inhibitors demonstrated a correlation with experimental kinetic data [1]. Specifically, competitive inhibitors, which are close analogs of vitamin C, bind in the active site, while other inhibitors (such as lycorine, apiol and its analogs, allylbenzenes, and chalcones) bind differently. However, the stability and observed conformational diversity of the enzyme in bulk raised concerns. Therefore, the objective of this work is a more rigorous investigation of the dynamic properties of the enzyme. We conducted several microsecond-scale MD simulations utilizing both conventional and accelerated approaches to gather comprehensive structural data. We took into account different protonation states. The resulting trajectories were characterized by a standard MD analysis framework. Additionally, we performed clustering of the trajectories to determine representative enzyme conformations and estimate their relative populations. The contacts between the cofactor and the protein were identified to understand the realizable protonation states. We employed dimensionality reduction techniques to visualize a conformational space to compare conventional and accelerated MD exploration power. As a more advanced method, we trained a distance-preserving autoencoder neural network on the structural data in an effort to construct an interpretable low-dimensional space, and the results were compared to other methods to assess its efficacy. The enzyme dynamics demonstrated that the model is stable and suitable for application in structure-based drug design. The representative protein conformations we obtained will be employed in comprehensive ensemble docking within a dataset of drug-like molecules.