UC Davis

This dissertation summarizes original work relevant to product predictions for Cytochrome P450 (CYP450) catalyzed transformations using a combination of computationally affordable methods, specifically modern force field and semi-empirical methods and protein-ligand docking. Additionally, it highlights multiple applications of Density Functional Theory (DFT) in collaboration with our synthetic chemistry colleagues to explore and explain photoisomerization, redox chemistry, and reaction mechanisms.Firstly, Cytochrome P450s (CYP450) are metabolically and synthetically important enzymes, catalyzing an array of oxidative transformations across all kingdoms of life. The prediction of oxidative products resulting from CYP450 catalyzed transformations is historically challenging and often relies only on enzyme-substrate fit and binding affinity estimates while neglecting measures of reactivity. Herein we present computationally affordable methodology for estimating epoxidation and hydroxylation barriers. When predicted hydroxylation barriers are paired with traditional protein-ligand docking, we improve on previously published prediction success rates and open the door to enzyme design in CYP450s for the purpose of achieving novel biosynthetic outcomes. In Chapter 1, epoxidation barriers were predicted using a multiple linear regression model with the fractional occupation number weighted density (FOD) and orbital weighted Fukui index (fw+) as descriptors localized to the vinylic carbon atom involved in the initial C–O bond formation event during epoxidation. Relative to previously computed epoxidation barriers using density functional theory in a panel of 36 compounds, mean absolute errors of 0.66 and 0.70 kcal/mol were achieved in the training and test sets, respectively, with coefficients of determination of ca. 0.80 were. This was done at the GFN2-xTB//GFN-FF level of theory. By performing electronic structure calculations on force field generated geometries, this approach is highly scalable. In Chapter 2, a single linear regression model was built to predict hydrogen atom transfer (HAT) barriers following the formation of Compound I, relevant to CYP450-mediated hydroxylations. The C–H bond dissociation energy involving a “frozen radical” – that is the removal of a hydrogen atom from an sp3 hybridized carbon in the substrate followed by single point energy calculations as doublets for the resulting unoptimized substrate radical and hydrogen atom – was found to correlate well with hydrogen atom transfer barriers previously computed with density functional theory. At the GFN2-xTB//GFN-FF level of theory for a panel of 24 sp3 hybridized carbon atoms across 21 substrates, mean absolute errors of ca. 1 kcal/mol were achieved in both training and test sets. By again leveraging force field and semi-empirical methods, this approach will scale to thousands of structures on even a modest computing resource. In Chapter 3, hydroxylation product predictions are made by combining enzyme-substrate docking and HAT barrier regression modelling. Hydroxylated product formation certainly relies significantly on the fit and binding affinity of a substrate with a CYP450 enzyme and not on the HAT barrier with Compound I alone. To this end, HAT barriers predicted using regression modeling were combined with Oheme–Hsubstrate constrained docking and pose clustering to make product predictions on a set of 25 substrates for the CYP101A1 camphor 5-monooxygenase enzyme. Using RxDock as an example utility used in high throughput virtual screening (HTVS), the prediction success rate for any hydroxylation product was 84% in the top two predictions when HAT barriers were included, compared to only 80% without the inclusion of HAT barriers. Combining HAT barriers and docking scores from Rosetta, any hydroxylation product was successfully predicted in the top two predictions in 92% of the 25 substrates studied. More importantly, the primary hydroxylation product prediction success rate was 84% in the top two predictions. Collectively, these findings meet or exceed the performance of previously published results in a non-parametric fashion. More importantly, the performance using Rosetta indicates our combination of docking and HAT barriers holds tremendous promise in the application of enzyme design. In the second half of this work, theoretical calculations were employed to rationalize experimental outcomes. Such retrospective analysis tends to be employed when experimental observations fail to meet our preconceived chemical intuition. By coupling wet experiment with quantum chemical theory, we can gain insight into underlying electronic structures and, in doing so, better understand and even predict spectroscopic or thermochemical properties in our systems of interest. To this end and in collaboration with the laboratory of David Olsen, we explored three series of experimental findings using Density Functional Theory in the sort of post-hoc fashion described above. These three efforts focused on the spectroscopic properties (and limitations) of azobenzene photoswitches,1 oxidation potentials relevant to Baeyer-Mills reactions,2 and the samarium-mediated rearrangement of vinyl aziridines to afford more complex heterocycles. In all three cases, computational efforts followed behind the experiment and aimed to generate models that explained the Olsen’s labs findings, as well as affording methodology that could be used to further expand their work ahead of experimental efforts. In Chapter 4, the photoisomerization of acylhydrazone-functionalized azobenzene derivatives is explored. With seemingly two photoswitchable motifs present, our collaborators only observed E to Z photoisomerization across the azobenzene substructure. Using Time-dependent Density Functional Theory (TD-DFT), the π to π* transition at approximately 380 nm is predicted to have a strongly localized electron density difference over the azo motif between the ground and excited state, with no discernable difference predicted over the acyl hydrazone functionality. Additionally, substituent effects were studied for a handful of electron withdrawing and donating cases explored synthetically, all showing no π to π* transition near 300 nm, inconsistent with other acylhydrazone photoswitches. This study rationalized the findings of our collaborators, and while retrospective analysis is useful, this study further highlights the opportunity to leverage computational techniques prospectively to guide synthetic efforts. In Chapter 5, retrospective analysis of synthetic findings was again conducted. In this application, the Bayer-Mills reaction is a traditional route to azobenzenes by way of a condensation reaction. However, azoxybenzene side products are also formed. Here, we attempted to correlate the formation of azoxybenzene with one electron oxidation potentials computed with DFT. In this work, electron-rich aniline derivatives with low oxidation potentials were found to produce undesirable levels of the azoxybenzene product, and we demonstrate that the computed oxidation potential from DFT with implicit solvation is a useful descriptor in predicting the outcome of the Baeyer-Mills reaction for given reactants. Lastly in Chapter 6, access to vinyl aziridines is explored mechanistically using traditional stationary point searching with DFT. Our collaborators discovered that vinyl aziridines could undergo ring expansion in the presence of samarium (II) iodide. While simple Lewis-acid promoted expansions are known, we explored a radical mechanism consistent with samarium (II) iodide mediated single electron transfer reactions observed in reductions and cross-couplings of ketones. From our analyses at the PBE0-D3BJ/def2-TZVP (ECP = Sm, I; SMD = toluene)//PBE0/def2-SVP (ECP = Sm, I) level of theory, a radical mechanism on the septet spin surface is achievable thermally at room temperature, with an overall free energy barrier of 25.1 kcal/mol and a strong thermodynamic driving force to favor the product-catalyst complex by 22.1 kcal/mol, both relative to the reactant-catalyst complex. These findings corroborate those of our synthetic colleagues and suggest that the transformation occurs according to a single electron transfer mechanism. This affords a mechanistically differentiated route to substituted 3-pyrrolines. In all, this work showcases multiple applications of computational chemistry that are relevant to protein engineering and medicinal chemistry, with an aim toward increased prospective use in the design of experiments.