Publication: Case Study: Impact of advanced x-ray refinement methods on our understanding of structure and function

EXPLORE DIVCON PLUGIN

Introduction

In a recent paper in which we detail the treatment of 50+ structures, we demonstrate that quantum mechanics (QM)-based x-ray refinement available in Phenix/DivCon is able to consistently generate ligand structures of lower strain vs. conventional refinement. However, is strain the only measure of success? In modern, structure based drug design, ligand strain is just one of several goals of a successful crystallography effort. Arguably, more accurate x-ray refinement should also yield actionable intelligence about how the ligand is interacting with the protein within the pocket, and what sorts of interactions are critical for continued lead optimization. While one could suggest that better ligand CIF restraints (e.g. target atom types, bond lengths, and bond angles) will lead to improved ligand structure, capturing and measuring significant protein:ligand interactions requires a more advanced energy functional. With this in mind, we recently coupled our Phenix/DivCon refinement with the ligand interaction diagram and analysis tool in MOE and explored how QM refinement leads to improved understanding of structure and ultimately the interactions that lead to function.

The Ligand Interaction diagram and analysis tool can be executed on target:ligand complexes in order to generate a 2-dimensional (2D) representation of the active site(s) of each complex. These diagrams show several interaction types including hydrogen bonds, arene interactions, ionic bonds, metal coordination, and of course, covalent bonds. By using this method directly, we can programmatically count the numbers of each of these interactions and determine how they change from one structure (e.g. initial conventional-refined) to another structure (e.g. QM refined). These diagrams can then be used to illustrate these interactions graphically, and through this analysis, we can show that QM refinement will “rescue” interactions. For example, in the figure below, we show that QM refinement will rescue the hydrogen bonds noted by the second blue comparison arrow. These sorts of interactions are missing in the original, stereochemical restraint-based refinement in large part because the function used in the initial refinement does not have the ability to capture electrostatics, hydrogen bonds, dispersion, and so on. Interestingly, the QM method is also able to filter out interactions that may not actually exist (or are very weak) as depicted by the first blue comparison arrow where the ionic bond disappears upon QM refinement. We refer to these two cases as False- and False+ respectively.

Below we show two example analyses both with a structure involving multiple ligand copies (e.g. PDBid:1HWI) and with a structure that includes a metal ion close to the ligand (e.g. PDBid:3QIY).

Example 1 – PDBid:1HWI:

The 1HWI structure has four copies of the same ligand and Phenix/DivCon will automatically refine each ligand:pocket structure using the PM6 Hamiltonian. Inclusion of this example illustrates the variability that can be observed in crystal structures of the same ligand when a more advanced functional is brought to bear on the problem. To run this example, the following basic refinement steps are performed:

% source /path/to/DivConDiscoverySuite/etc/qbenv.sh
% source /path/to/phenix-dev-release/phenix_env.sh
% qbphenix –pdbid=1HWI –resname 115 –protonation MOE –qmMethod pm6
–region-radius 3.0 –buffer-radius 2.5 –NP 4

Once the refinement has completed, the qbLigandInteractions.svl file – which is included within the DivCon Discovery Suite SVL directory – can be executed on the command line or within your own phenix scripts. The first PDB file in the argument list is the initial or pre-refined structure while the second PDB file is the QM refined structure. The optional third argument is the 3-letter code of the ligand. If this argument is not provided, then all of the ligands – as determined by MOE – will be treated. It is important to note that both the before and the after structures are identically protonated.

% qbmoebatch -exec -licwait
“qbLigandInteractions [‘1HWI.pdb’,’1HWI_refine_001.pdb’,’115′]”

The analysis takes a few seconds to run, and it will output a number of files in the directory including initial and final 2D ligand interaction diagrams, a csv file with the counts of the various interactions found within the structure, and an HTML document that can be viewed to help navigate the results. Download the analysis 1HWI.tar.gz file to visualize the data on your own computer.

The figure above shows the False- and False+ interactions found for Chain D while the table below provides all four binding pocket copies. The table also details the counts of the various interaction types, False-‘s, False+’s, and so on. Clicking on the Map for each ligand will allow you to bounce back and forth between the two states.

Interaction E Analysis Maps
PDBid Ligand ID Chain Initial Final Initial Final
1HWI 115 2 A -280.416 -354.826 I-115_2_A.png F-115_2_A.png
115 1 B -251.57 -294.351 I-115_1_B.png F-115_1_B.png
115 4 C -270.923 -341.24 I-115_4_C.png F-115_4_C.png
115 3 D -289.224 -367.805 I-115_3_D.png F-115_3_D.png
hbond metal
I F F+ F- I F F+ F-
8(55) 10(54) 1(13) 3(12)
8(55) 10(56) 0(8) 2(9)
7(50) 9(47) 1(13) 3(10)
7(57) 9(51) 1(13) 3(7)
ionic arene
I F F+ F- I F F+ F-
4(4) 5(5) 0(0) 1(1)
4(4) 4(4) 0(0) 0(0)
4(4) 3(3) 1(1) 0(0)
4(4) 3(3) 1(1) 0(0)
covalent distance
I F F+ F- I F F+ F-
134(134) 135(135) 15(15) 16(16)
146(146) 143(143) 11(11) 8(8)
123(123) 132(132) 10(10) 19(19)
136(136) 144(144) 13(13) 21(21)

Legend

  • Interaction Energy = MOE Amber12EHT (in kcal/mol)
  • Value(Value) = count of interactions between target:ligand (count of interactions within pocket)
  • F+ = False+ = interactions observed in the Initial model that are removed in the Final model
  • F- = False- = interactions rescued in the Final model but not captured in the Initial model
  • F = Final or the interaction count associated with the final model.
  • I = Initial or the interaction count associated with the initial model.
  • Distance = counts of atom-atom distances less then or equal to 4.5 Å.

Example 2 – PDBid:3QIY:

The 3QIY structure has a metal ion in addition to the ligand. Phenix/DivCon, with its use of PM6, is able to handle metal ions – such as the zinc in 3QIY – transparently and without special configuration or coordination sphere definitions. To run this example, the following basic refinement steps are performed:

% source /path/to/DivConDiscoverySuite/etc/qbenv.sh
% source /path/to/phenix-dev-release/phenix_env.sh
% qbphenix –pdbid=3QIY –resname QI1 –protonation MOE –qmMethod pm6
–region-radius 3.0 –buffer-radius 2.5 –NP 4

Once the refinement has completed, again the the qbLigandInteractions.svl file is executed on the command line or within your own phenix scripts. To recap, the first PDB file in the argument list is the initial or pre-refined structure, the second PDB file is the QM refined structure, and the optional third argument is the 3-letter code of the ligand. If this argument is not provided, then all of the ligands – as determined by MOE – will be treated. As before, both structures are the protonated versions.

% qbmoebatch -exec -licwait
“qbLigandInteractions [‘3QIY.pdb’,’3QIY_3.pdb’,’QI1′]”

The analysis takes a few seconds to run, and it will output the initial and final 2D ligand interaction diagrams for the QI1 ligand, a csv file with the counts of the various interactions found within the structure, and an HTML document that includes all of the same information. Download the analysis 3QIY.tar.gz file to visualize the data on your own computer.

For 3QIY, the initial refinement missed interactions both with the ligand and around the Zn ion. For a more detailed breakdown, the table below shows the False+ and False- counts to each interaction type. Simply scroll left or right to see these counts. For brevity, the Legend has been removed. If required, please use the Legend provided under the table for 1HWI.

InteractionE Analysis Maps
PDBid Ligand ID Chain Initial Final Initial Final
3QIY QI1 432 A -90.0462 -125.145 I-QI1_432_A.png F-QI1_432_A.png
hbond metal
I F F+ F- I F F+ F-
1(19) 2(30) 0(0) 1(11) 1(5) 2(5) 0(1) 1(1)
ionic arene
I F F+ F- I F F+ F-
0(2) 3(5) 0(0) 3(3) 1(1) 2(2) 1(1) 2(2)
covalent distance
I F F+ F- I F F+ F-
215(215) 214(214) 20(20) 19(19)

Conclusion:

The interactions between a protein and its ligand(s) are very difficult to correctly deduce prior to the refinement since they are not dependent upon stereochemical restraints. Since the QM functional in Phenix/DivCon does not rely on target atom types, bond lengths, and bond angles, it is able to capture interactions and allow the entire protein:ligand complex to optimize together with the experimental data to produce structures that are both experimentally correct and chemically accurate.

For additional examples, details of the method, and licensing information, visit the Phenix/DivCon product page.

EXPLORE DIVCON PLUGIN