A core assumption of structure-based drug discovery (SBDD) is that the 3D structure of an enzyme complexed with its ligand(s) and cofactor(s) is an important driver in the exploration and discovery of novel drug and drug-like compounds. This structure can be experimentally determined using X-ray crystallography, which works by “shinning” a beam of X-rays through a crystal and using the resulting diffraction pattern to determine atom positions. However, when it comes to protons, these ubiquitous atoms are all but invisible to X-rays. Therefore, it is extremely hard if not impossible to experimentally demonstrate where a proton is found through diffraction – especially at the lower resolution ranges so often used in SBDD. In fact, at these lower resolutions, knowledge of any atom position rests on a mixture of experimental diffraction data and computational/modeling methods, and this synergy is referred to as X-ray refinement. Since the diffraction data rarely has explicit information about proton positioning, often the conventional tools used in the field do not even take protons into account during the refinement process.
On the other hand, quantum mechanics (QM) based methods – such as those found in QuantumBio’s Phenix/DivCon package – are considered “all atom” methods which means that these more accurate methods are more sensitive to protonation, and unlike conventional methods, QM methods readily capture electrostatics and quantum effects whenever they are used to study structure. Therefore, QM methods are a natural fit for X-ray refinement since they more accurately compensate for deficiencies in the experimental model. To refine a structure using Phenix/DivCon, various tools can be used to predict and ooptimize the protonation of a complex prior to refinement. In addition to tools within the Phenix package, tools such as Protonate3D from Chemical Computing Group Inc. or the open-source Reduce software can also be used. These tools generally do a good job and they will usually give the SBDD practitioner a solid start. In the present case study, we will explore what happens when one of these tools yields a surprising prediction, and we will see how we can use advanced QM methods coupled with X-ray crystallographic methods and experimental data to answer the question: Where’s the Proton?
While exploring the PDBid:1AZM structure depicted in the figure to the right, we happened upon an interesting case surrounding the possible protonation states found within the active site of the Human Carbonic Anhydrase I (HCAI). The active site of HCAI utilizes a zinc ion along with a number of coordinated histidine amino acids as part of its catalytic function. Conventional X-ray tools require a thorough, user-supplied description of the properties of the coordination sphere around the metal. QM methods however treat metals such as zinc much more transparently, and 
therefore HCAI is a good “test case” for treatment. The QM-augmented X-ray refinement method calculates characteristics (e.g. energies and gradients) for the entire active site including any ligand, the histidine residues, and of course the coordinated zinc and the tool automatically updates these characteristics in “real-time” during each step of the refinement process. These gradients are then used by the software to “push” each atom towards a lower energy which leads to a lower energy structure.
For this validation, the protonation software used predicts that the ND1 of HIS94 would not be protonated and that the histidine therefore should carry a negative charge. As depicted in the figure, the HIS94 is situated immediately next to and coordinated with a positively charged Zn++ which could explain why the software chose this state. However, this predicted protonation state is not expected since negative histidines are considered quite unusual. Further, this same negative histidine is observed regardless of the pH setting.
In order to determine whether or not this protonation made sense energetically, we explored the various possible states. First, we calculated the total energy of both protonation states using the Merck Molecular Force Field (MMFF), and we find that the HIS64- structure actually has a slightly – albeit insignificantly – lower energy then the protonated-HIS94 by 0.2kcal/mol. Since the difference is insignificant, this result is not really enough information to make a truly informed decision. With that in mind, we hypothesized that the proton (or the absence thereof) should in some way impact that structure of this HIS94 and that we could demonstrate this dependency crystallographically. Another change that the preparation software predicts is a nitrogen/oxygen “flip” of the GLN92 and this therefore presents us with a third option to explore. So QM-based refinements were undertaken using Phenix/DivCon on the three possible configurations:
- GLN92 flipped from published 1AZM configuration along with the unprotonated HIS94-.
- GLN92 not flipped along with the unprotonated HIS94-.
- GLN92 not flipped along with the HIS94 protonated on the ND1 atom.
Within an hour or so, we had a definitive answer. The results of these refinements are displayed in the figure below where the left side depicts refinements (1) and (2) and the right shows the results of refinement (3). Both refinements (1 – pdb;mtz) and (2 – pdb;mtz) display significant negative density (red netting) and positive density (green netting) associated with the movement of the HIS94- away from the GLN92 and toward the Zn++. These densities are often indicators of a problem or disagreement between the predicted model (e.g. atomic coordinates) and the experimental diffraction data. Specifically, negative density corresponds to areas where the model places atoms that don’t agree with experiment, and positive density is where the diffraction experiment suggests there should be atoms that are missing from the model. However, when we consider refinement (3 – pdb;mtz), which has the protonated histidine variation, this simulation demonstrated no negative or positive density associated with the movement of HIS94.
These results are quite encouraging, but could we have answered the “Where’s the proton?” question without using the QM functional? We again started with the original three configurations listed above, and we ran the corresponding conventional refinements using the same default Phenix settings. In this case, no movement one way or the other is observed. In other words, as far as the conventional refinement is concerned, the three structures are identical to one another. Therefore, it is clear that in this example, if one wants to determine which structure is correct to some level of certainty one must use a more advanced functional then those available in conventional refinement. These three configurations are chemically distinct, and it makes sense that without the proton, the negatively charged HIS64- should gravitate towards the positively charged Zn++ ion. The fact that the conventional refinement did not show any movement shows not only that it does not capture the fact that a proton has been placed on the HIS64, but that it isn’t able to model the electrostatic component of the interaction.
For more information on the method and its use please visit the Phenix/DivCon product page. If you would like to evaluate the method for your own structure based drug discovery efforts, feel free to contact us at sales@quantumbioinc.com for more information.
About the authors:
Lance M. Westerhoff, Ph.D.
President and General Manager
QuantumBio Inc.
Oleg Y. Borbulevych, Ph.D.
Research Fellow: Crystallography Methods
QuantumBio Inc.
Thanks to Drs. Dhruva Chakravorty and Kennie Merz at the University of Florida for helpful discussion!