Cloud-based simulations on Google Exacycle reveal ligand modulation of GPCR activation pathways

Abstract

Simulations can provide tremendous insight into the atomistic details of biological mechanisms, but micro- to millisecond timescales are historically only accessible on dedicated supercomputers. We demonstrate that cloud computing is a viable alternative that brings long-timescale processes within reach of a broader community. We used Google's Exacycle cloud-computing platform to simulate two milliseconds of dynamics of a major drug target, the G-protein-coupled receptor β2AR. Markov state models aggregate independent simulations into a single statistical model that is validated by previous computational and experimental results. Moreover, our models provide an atomistic description of the activation of a G-protein-coupled receptor and reveal multiple activation pathways. Agonists and inverse agonists interact differentially with these pathways, with profound implications for drug design.

Figure 1. MSM activation trajectories on the sub-millisecond timescale

(a) The key differences between active and inactive structures of β₂AR. The root mean squared deviation (RMSD) of the connector region (I121-F282), the H5 bulge (S207-F208) RMSD from active crystal structure (3P0G), the distance between Helix 3 and Helix 6 (measured as R131-L272 distance) and the RMSD of NPxxY region (N322-C327) in Helix 7 from the active crystal structure (3P0G). Active crystal structure values are marked with dashed lines. 150 μs MSM activation trajectories were built using the data from simulations of GPCR bound to (b) agonist BI-167107, (c) inverse agonist carazolol as well as (d) the apo receptor. Agonist bound simulations stabilize active-like conformations (highlighted with boxes) throughout the trajectory and deactivate in 2.5 ± 0.05 μs. Inverse agonist bound simulations deactivate from active states slightly faster, at 1.3 ± 0.1 μs and Apo simulations deactivate in 0.67 ± 0.02 μs (See Supplementary Figures S5 and S6). These timescales corroborate previous simulations of deactivation (5) and experiments when keeping in mind that only G protein binding can truly stabilize the active states.(30)

Figure 2. Markov State models and high flux activation pathways for agonist and inverse agonist bound simulations

(a) Network representation of the 3000 state MSM built from the simulations of agonist bound GPCR with each circle representing an individual conformational state. (b) 10 State MSMs build from the 3000 state MSMs using spectral clustering methods to identify kinetically relevant states. The circles in the 3000 state MSM are colored according to their membership in the coarse-grained 10 state MSM. The weight of arrow indicates the transition probability between states. (c) Pathways are shown as states (circles) connected along the 3-D reaction coordinate used, in part, to build the MSM. Pathway connections are scaled by the path flux relative to the highest flux in black; for inverse agonist pathways, red is 61% and orange is 51% of the max; for agonist red: 48% and orange: 35%. (d) Mutual information networks of dynamically correlated residues. Black lines indicate connected residue pairs, and only helices 3-7 are shown in the image for clarity. Agonist bound simulations reveal a network of residues that connect the extra and intra-cellular parts of the receptor to stabilize active states, whereas inverse agonist eliminates these connections and blocks activation.

Figure 3. Structural details of the activation pathways

Representative structures from states along the activation pathways in Figure 2c labeled by number in four panels. The transition from inactive conformations in panel 1 proceed via Helix 3-6 outward movement (Panel 2), the switching of M215 interactions from connector I121 to F282 as these residues flip conformation (Panel 3), and changes in the H5 region around F208, S207, and S203 (Panel 4) that form ligand-mediated interactions for stabilizing active conformations that can be selected by G proteins. Residues in grey are the aligned inactive crystal (2RH1) conformations; residues in yellow are from the active crystal structure (3P0G).

Figure 4. Examples of GPCR Ligand Chemotypes Enriched at MSM States Along Activation Pathways

MSM states from high flux activation pathways were assigned a progress score ξ based on structural metrics in Figure 1 and range from the inactive crystal structure (2RH1) at ξ=0 to the active crystal structure (3P0G) at ξ=1. Top ranking compounds from a retrospective virtual screen of known β₂AR ligands at each MSM state, and both crystal structures, were clustered according to their 3D shape and chemistry overlap. Four examples of chemotype clusters enriched by select MSM states along the activation pathways are shown, with the percentage of a chemotype represented in the total ligands enriched at a given ξ. (a) Example agonist chemotypes are catecholamine derivatives (1, 2 and 3) and ethanolamine (4) derivatives. b) Example antagonist chemotypes (5, 6, 7 and 8) share a 2-hydroxy propyl amino core and include a carbostyril substituted pyridazinone (5), benzhydryl-amine (6), and pyridone nitriles (7, 8). The complete distributions along ξ for all agonist and antagonist chemotype clusters are shown in Supplementary Figures S21 and S22. These results show that docking to intermediates identified by MSM Transition Path Theory analysis enriches more diverse chemotypes that could be missed by screens of only a few structures.