Structural mechanism of a drug-binding process involving a large conformational change of the protein target

Abstract

Proteins often undergo large conformational changes when binding small molecules, but atomic-level descriptions of such events have been elusive. Here, we report unguided molecular dynamics simulations of Abl kinase binding to the cancer drug imatinib. In the simulations, imatinib first selectively engages Abl kinase in its autoinhibitory conformation. Consistent with inferences drawn from previous experimental studies, imatinib then induces a large conformational change of the protein to reach a bound complex that closely resembles published crystal structures. Moreover, the simulations reveal a surprising local structural instability in the C-terminal lobe of Abl kinase during binding. The unstable region includes a number of residues that, when mutated, confer imatinib resistance by an unknown mechanism. Based on the simulations, NMR spectra, hydrogen-deuterium exchange measurements, and thermostability measurements and estimates, we suggest that these mutations confer imatinib resistance by exacerbating structural instability in the C-terminal lobe, rendering the imatinib-bound state energetically unfavorable.

Fig. 1. Simulations of Abl-imatinib binding.

A Top: A diagram describing the scheme of Abl-imatinib binding, starting with the spontaneous DFG flip of the apo kinase with an open activation loop (AL) (i.e., from the DFG-in/AL-open to the DFG-out/AL-open conformation). A binding intermediate is then formed with imatinib (DFG-out/AL-open·I), which resolves to the native imatinib-bound state (DFG-out/AL-closed·I). Bottom: The five generations of simulations (G0–G4) that together span the binding process. Each gray line represents an individual simulation, and each blue arrow represents a group of simulations. The connected red lines represent a set of four sequential simulations that, when concatenated, describe the entire binding process (together, we refer to these four simulations as the “concatenated binding simulation”). The red dots represent simulation snapshots that are representative of conformational states denoted by the corresponding circled numbers below (this labeling scheme is also used in the other panels). States 0, 1, 3, and 5 respectively correspond to the four states illustrated in the binding diagram (these are the more stable states), while States 2 and 4 represent the more transient states in the binding process. B The imatinib molecule moving from the free state to the bound state in the concatenated binding simulation; the native pose is reached in ~10 μs. C Imatinib occupancy derived from all G0 simulations (green) and derived from all G1 simulations (red). The imatinib molecules primarily interacted with Abl kinase at the following five labeled regions: (1) the ATP-binding site and its extension between the N- and the C-lobes; (2) the interface with the SH2 domain used in Abl autoinhibition; (3) the myristoyl-binding site; (4) the αG helix region of the C-lobe; and (5) the PIF-binding site of the N-lobe. D Snapshots of the imatinib-binding site corresponding to the numbered states from the concatenated binding simulation. E Imatinib root-mean-square deviation (RMSD) with respect to the native pose (red) and A-loop RMSD with respect to the closed conformation (blue) as functions of time in the concatenated binding simulation. The five states (States 1–5) are shown. F Imatinib-bound pose from the simulation superimposed on the imatinib-bound crystal structure pose (PDB ID: 1OPJ). G Starting and final conformations of the A-loop from the concatenated binding simulation.

Fig. 2. Simulations starting from the active Abl kinase conformation.

A G0 simulations, which start from the active conformation (State 0). The DFG flip occurred in most of these simulations (i.e., State 1 was reached) and two simulations reached State 2. The binding diagram and G1–G4 simulations are shown and explained in Fig. 1A. B Imatinib RMSD with respect to the native pose (red) and Asn368-Asp381 distance indicative of DFG conformation (green) as functions of time in one of the G0 simulations that reached State 2. The DFG flip occurred at ~67 µs. C Left: A representative snapshot after the DFG flip in the simulation superimposed on the DFG-out/AL-open crystal structure (PDB ID: 1OPK). Right: A representative snapshot from a G0 simulation superimposed on State 2 from the concatenated binding simulation shown in Fig. 1A.

Fig. 3. Cracking at the C-lobe in the Abl-imatinib binding simulation.

A Top: Root mean square fluctuation (RMSF) of every residue as a function of time from the concatenated binding simulation (shown in Fig. 1A). The cracking site at the C-lobe is highlighted with the red rectangles. Bottom: Cracking site (residues 412–419, 477–483) RMSF over time. (RMSF values are shown as transparent lines for each frame; smoothing was applied and shown as a solid line to highlight the general trends in the data.) Time points 3–5 refer to the states marked in Fig. 1. B Snapshots of the cracking site in State 1 and State 4 in the concatenated binding simulation. C Similar analysis as shown in panel (A), but for a dasatinib-binding simulation. D Similar analysis as shown in A, but for a simulation of apo Abl kinase starting from the imatinib-bound crystal structure (i.e., with the imatinib removed). As shown, the cracking subsides after A-loop closing. E A-loop RMSD with respect to the active conformation (DFG-in/AL-open state; PDB ID: 2F4J) as a function of time in the simulations starting from the crystal structure of the Abl-imatinib complex (PDB ID: 1OPJ), with (red) or without (black) removing the imatinib. The events of A-loop opening in the apo simulations are marked with arrows.

Fig. 4. Mapping of imatinib-resistance mutations to the cracking site.

A Top: Previously reported imatinib-resistance mutations mapped to an Abl-imatinib complex structure (PDB ID: 1OPJ). Bottom: A close-up view of the cracking site. The mutations are colored according to their positions in a manner consistent with C. B Observed binding rate constant (k_obs) from stopped-flow experiments with increasing concentrations of imatinib (top panel, WT, n = 4; M343T, M472I, n = 3; I502M, E509D, n = 2 independent experiments) and dasatinib (bottom panel, n = 4 independent experiments). Data are presented as mean values ± SD (SD is shown only where the sample size is >3). Only the kinase domain of Abl was used. In vitro data of F486S is absent in this study because we failed to express and purify this mutant. Source data are provided as a Source Data file. C The FoldX estimates of the effects of the imatinib-resistance mutations on the folding free energy of Abl kinase (n = 3 independent estimations; data are presented as mean values ± SD). The values of mutations of different regions of Abl kinase are shown in different colors. Source data are provided as a Source Data file. D Normalized histograms of the RMSFs of the cracking region with respect to the averaged conformation of the region. The N-lobe and the αE and αF helices were aligned for the RMSD calculations. The top panel refers to simulations of Abl kinase bound with imatinib (PDB ID: 1OPJ), and the bottom panel refers to simulations of Abl kinase bound with both imatinib and GNF-2 (PDB ID: 3K5V). E HDX differences (D_M472I – D_wt) between the WT and M472I mutant of Abl kinase, as measured by MS, mapped to the Abl kinase structure according to the color scheme shown (raw data is presented in Supplementary Fig. 4B). The alpha carbon of M472I is represented as a red sphere. F Melting temperatures of WT Abl kinase and seven imatinib resistance mutants (for WT, M372I, and I502M, n = 3; for all other constructs n = 2 independent experiments). Source data are provided as a Source Data file. G Cell proliferation IC₅₀ values of imatinib (left panel: WT, n = 19; T315I, n = 12; M343T, n = 3; M351T, n = 6; M472I, n = 7; F486S, n = 11; I502M, n = 3 independent experiments) and dasatinib (right panel: WT, n = 8; T315I, n = 1; M343T, n = 3; M351T, n = 6; M472I, n = 5; F486S, n = 9 independent experiments) with cells expressing either WT Abl or the indicated imatinib-resistance mutants. Data are presented as mean values ± SD (SD is shown only when sample size is >3).

Fig. 5. Melting-temperature and NMR analyses of the M472I mutation and the binding of imatinib, dasatinib, and GNF inhibitors.

A Melting-temperature changes resulting from GNF-2 binding to apo, imatinib-bound, and dasatinib-bound WT Abl kinase and the M472I mutant (n = 3 independent experiments; ** indicates P ≤ 0.01, *** indicates P ≤ 0.001, and **** indicates P ≤ 0.0001; two-tailed t-test with 95% confidence interval was performed; apo: P = 0.001, degrees of freedom (df) = 5.52, effect size (d) = 5.9; imatinib bound: P = 0.003, df = 8.2, d = 4.1; dasatinib bound: P = 0.00004, df = 6.62, d = 9.5). Data are presented as mean values ± SEM. Source data are provided as a Source Data file. B Chemical shifts of WT Abl kinase due to imatinib or dasatinib binding. C NMR spectra of R457 and G250 in apo, imatinib-bound, dasatinib-bound, and imatinib+GNF-5-bound Abl kinase. D Comparison of imatinib-bound WT and M472I Abl kinase. The residues that underwent large chemical shifts due to M472I mutation are shown in the right panel in a ribbon diagram. The yellow, orange, and red dashed lines mark shifts of more than one, two, and three standard deviations, respectively. The residues that displayed chemical shifts above the red or the orange threshold are shown in the right panel in the same color. E Chemical shifts associated with GNF-5 binding to WT and M472I Abl kinase. The residues for which WT and M472I behave differently on GNF-5 binding (with differences in chemical shifts of more than one standard deviation) are shown in the image to the right. F NMR spectra of D381 of the DFG motif in imatinib-bound and imatinib+GNF-5-bound WT (upper) and M472I (lower) Abl kinase.