Exploration of Cryptic Pockets Using Enhanced Sampling Along Normal Modes: A Case Study of KRAS G12D

Abstract

Identification of cryptic pockets has the potential to open new therapeutic opportunities by discovering ligand binding sites that remain hidden in static apo structures of a target protein. Moreover, allosteric cryptic pockets can become valuable for designing target-selective ligands when the natural ligand binding sites are conserved in variants of a protein. For example, before an allosteric cryptic pocket was discovered, KRAS was considered undruggable due to its smooth surface and conservation of the GDP/GTP binding pocket across the wild type and oncogenic isoforms. Recent identification of the Switch-II cryptic pocket in the KRAS^G12C mutant and FDA approval of anticancer drugs targeting this site underscores the importance of cryptic pockets in solving pharmaceutical challenges. Here, we present a newly developed approach for the exploration of cryptic pockets using weighted ensemble molecular dynamics simulations with inherent normal modes as progress coordinates applied to the wild type KRAS and the G12D mutant. We performed extensive all-atomic simulations (>400 μs) with and without several cosolvents (xenon, ethanol, benzene), and analyzed trajectories using three distinct methods to search for potential binding pockets. These methods have been applied as a proof-of-concept to KRAS and have shown they can predict known cryptic binding sites. Furthermore, we performed ligand-binding simulations of a known inhibitor (MRTX1133) to shed light on the nature of cryptic pockets in KRAS^G12D and the role of conformational selection vs induced-fit mechanism in the formation of these cryptic pockets.

Figure 1

Structures of KRAS:GDP showing regions important for cryptic pocket formation. (A) The crystal structure of the KRAS^WT:GDP complex (Protein Data Bank (PDB) ID: 4OBE) is shown with the Switch-I and Switch-II regions highlighted in magenta and yellow colors, respectively. The Gly12 residue, associated with oncogenic mutations in over 10% of all human cancers, is colored in green and rendered as a ball-and-stick. (B) Crystal structure of KRAS^G12D:GDP complex is shown in surface representation with MRTX1133 bound at the Switch-II pocket (PDB ID: 7RPZ), where both FDA approved drugs also bind (Sotorasib and Adagrasib). (C) Crystal structure of the KRAS^G12D:GDP complex bound to an analog of MRTX1133 (Compound 15) in the Switch-I/II pocket (PDB ID: 7RT1) is shown with surface representation. (D) The 2D chemical structure of the highly potent KRAS^G12D inhibitor, MRTX1133, is shown, which binds to the Switch-II pocket and was used in this study.

Figure 2

Normal mode analysis of KRAS^WT. (A) ANM mode 14, (B) ANM mode 15, and (C) ANM mode 12 calculated for the wild type KRAS (PDB ID:4OBE). The KRAS protein is shown in the light blue ribbons and the black arrows indicate the mode shape point from the positions of the Cα atoms (not shown) to the (tangent) direction of the motion. (D) Mode fractional variances for the top 20 slowest modes. Fractional variance measures the percentage of the overall dynamics explained by a given mode (see Methods section for details). (E) Mode collectivity of the top 20 slowest modes. Modes are sorted in descending order by the fractional variance in (D) and (E). Mode 12, 14, and 15 are highlighted in orange.

Figure 3

Exposon analysis and dynamic probe-binding analysis of KRAS^G12D. (A) Residues from an exposon that predicted MRTX1133 binding site as a potential cryptic pocket are shown in spheres. MRTX1133 is shown in orange stick. (B) Structural overlap of apo KRAS^G12D:GDP complex (shown in light blue) and KRAS^G12D:GDP complex bound to MRTX1133 (light yellow) shows that Tyr71 becomes solvent exposed in the inhibitor (MRTX1133) bound state. (C) Residues from a dynamic probe-binding site that predicted MRTX1133 binding site as a potential cryptic are shown in spheres.

Figure 4

Probe map analysis of KRAS^G12D. (A) Xenon-binding grid points that have a binding free energy less than −2.5 kcal/mol are shown as magenta spheres. (B) Xenon-binding cluster-1 and -2 overlap with the binding site of MRTX1133 and the secondary binding site of an analog of MRTX1133, respectively. MRTX1133 and its analog are shown as green sticks, while the MRTX1133-binding protein. (C) Residues from an exposon are shown as cyan spheres. The exposon residues surround two adjacent xenon-binding clusters that overlap with inhibitor-binding sites shown in (B). (D) Xenon-binding clusters -1, -3, and -4 overlap with a peptide inhibitor (purple ball and stick) site. In all panels, the simulation protein structure is shown in gray, while the reference structure is shown in either light yellow (PDB ID: 7RPZ) or light blue (PDB ID:5XCO).

Figure 5

SiteHopper similarity analysis of the WT and G12D isoform of KRAS. (A) Free energy-like distribution of SiteHopper similarity score computed from simulations of KRAS:GDP complex in various conditions is plotted. Red dotted line corresponds to the SiteHopper similarity score computed for the Apo structure of KRAS^G12D:GDP complex. The black dotted line corresponds to the SiteHopper similarity score computed for the equilibrated structure of KRAS^G12D:GDP complex used for launching WE MD simulations. The free energy is calculated as -k_BTln[P(SiteHopper similarity score)]. (B-D) Overlap of MRTX1133 pose observed in crystal structure (PDB ID: 7RPZ) with the lowest RMSD bound pose of MRTX1133 observed in each ligand-binding simulation. Bound pose in crystal structure is shown in gray-colored sticks. (B) The lowest RMSD pose observed in ligandRMSD-mode12 simulations is shown in violet (C) The lowest RMSD pose observed in ligandRMSD-mode14 simulations is shown in dark purple; and (D) The lowest RMSD pose observed in logRMSD-site-distance simulations is shown in magenta. RMSD of bound pose of MRTX1133 with reference to the pose observed in the crystal structure is also mentioned.