Effects of tautomerization on computed binding affinities

One central task of computer aided molecular design is to identify potentially active compounds against specific molecular targets, which have been identified as relevant for the pathobiology of an illness. Only those compounds predicted to have sufficient activity then need to be synthesized and tested experimentally, thus saving time and cost. For a few years now, so-called free energy simulations have been in widespread use, and, in principle, provide an accurate method to predict the binding free energy of ligands/inhibitors to a target molecule, typically a protein. Knowing relevant free energy differences is important as they are the fundamental criterion whether a reaction, such as binding of a ligand to a target molecule, takes place voluntarily or not. Free energy simulations are the most costly computational methods among the tools available for screening large numbers of compounds and are used as the last step before experimental tests are carried out. Therefore, it is quite important that results obtained in such calculations are accurate. The predictive value (i.e., accuracy of results) of such calculations depends on a number of factors. One possible source of error is the presence of what chemists denote as multiple tautomeric forms of a molecule. Two tautomeric forms of a molecule contain identical types and numbers of atoms; however, their molecular structure typically differs by the location of a hydrogen atom. In parallel, the position of a double bond changes as well. Tautomeric states of a molecule may have different physico-chemical properties, which may also affect their binding affinities. Thus, if the presence of an alternative tautomeric form is overlooked when carrying out a free energy simulation, a systematic error may occur, and available evidence suggests that said error may be considerable. Interestingly, this potential source of error in free energy simulations has to date not been investigated in much detail. This project aims to address the influence of tautomeric forms on binding affinities computed by free energy simulations in three steps. First, we want to compute free energy differences of binding for different tautomeric states to a number of systems, which will allow us to understand the size of this potential error more systematically. Second, at the methodological level, there are different approaches that can be used to account for the presence of tautomeric states in free energy simulations, and we want to explore which is the most efficient one. Finally, it is not clear that the classical mechanical force fields used to describe interactions between atoms and molecules in free energy simulations are sufficient to account for the influence of tautomeric states. Therefore, we will also use hybrid methods in which a small region of particular interest is studied by quantum chemical methods (specifically the compound existing in two or more tautomeric forms), whereas the rest of the systems would be described by the usual force fields.

Whether a chemical or biological reaction will occur voluntarily is determined by the free energy needed or liberated. For this reason, it is of utmost interest to predict the free energy balance of a process. One relevant practical application is the determination of the free energy difference of the binding of a ligand to a receptor. Potential drugs must bind with high specificity and affinity to their intended site of action; i.e., the free energy difference of binding should be highly favorable. The research project focused on one detail of such calculations, which had so far attracted little interest. Many chemical compounds, which are potential drugs, can exist in more than one form; this phenomenon is known as tautomerism. To compute the binding free energy difference correctly, one must carry out the calculation with the tautomeric form of the ligand that is predominant under the conditions of the reaction. It is also possible that several such states are present simultaneously; in this case, they have to be weighted accordingly. The starting questions were the following. Does the use of an incorrect tautomeric form strongly influence the result of the calculations? Are the approximate descriptions of intra- and inter-molecular interactions, the so-called force fields, sufficient? Our results, as well as those obtained by other researchers, clearly indicate that using the correct tautomeric state(s) is essential, but the challenges go beyond predicting the correct one. Tautomers differ in the position of one polar hydrogen atom, which influences the interactions between the ligand and the receptor and may affect the location and orientation of the ligand in the binding pocket. The simulation of such reorientations is very costly, and we needed to develop tools to carry out such calculations efficiently. The program package "transformato," freely available on GitHub, is a central result of the project. Results obtained with "transformato" also highlight the limited utility of force fields. To avoid this source of error, one has to refine the calculations of binding free energy differences by multi-scale methods. We worked on this methodology prior to the start of the project. Another key result of the project was making such calculations efficient enough to be useable in real-world applications. We are currently working on applying these techniques to questions of tautomerism. The project results confirm that one has to use the correct tautomeric form(s) of a ligand in binding free energy calculations. Doing so poses considerable challenges. The program package "transformato" and the development of correct and efficient methods to refine results obtained with force fields by multi-scale techniques provide the foundation to understand the influence of tautomerism on the results of free energy simulations and to handle it correctly in the future.