ATP-competitive inhibitors of PI3K enzymes demonstrate an isoform selective dual action by controlling membrane binding

Abstract

PI3Kα, consisting of the p110α isoform of the catalytic subunit of PI 3-kinase (encoded by PIK3CA) and the p85α regulatory subunit (encoded by PI3KR1) is activated by growth factor receptors. The identification of common oncogenic mutations in PIK3CA has driven the development of many inhibitors that bind to the ATP-binding site in the p110α subunit. Upon activation, PI3Kα undergoes conformational changes that promote its membrane interaction and catalytic activity, yet the effects of ATP-site directed inhibitors on the PI3Kα membrane interaction are unknown. Using FRET and biolayer interferometry assays, we show that a class of ATP-site directed inhibitors represented by GSK2126458 block the growth factor activated PI3KαWT membrane interaction, an activity dependent on the ligand forming specific ATP-site interactions. The membrane interaction for hot spot oncogenic mutations that bypass normal p85α regulatory mechanisms was insensitive to GSK2126458, while GSK2126458 could regulate mutations found outside of these hot spot regions. Our data show that the effect of GSK126458 on the membrane interaction requires the enzyme to revert from its growth factor activated state to a basal state. We find that an ATP substrate analogue can increase the wild type PI3Kα membrane interaction, uncovering a substrate based regulatory event that can be mimicked by different inhibitor chemotypes. Our findings, together with the discovery of small molecule allosteric activators of PI3Kα illustrate that PI3Kα membrane interactions can be modulated by factors related to ligand binding both within the ATP site and at allosteric sites.

Keywords: PIK3CA; lipid kinase; membrane proteins; phosphoinositide 3-kinase; protein conformation; small molecules.

Figure 1.. Relationship between the ATP binding site of the PI3Kα p110α subunit membrane interaction sites and oncogenic mutations.

(A) Model of PI3Kα activation by phosphotyrosine-peptide (pY) binding to the SH2 domains of the p85 subunit, illustrating the relative positions of the active site (light blue hexagon), helix Kα12 (orange cylinder) and known membrane binding interfaces (dark blue). (B) Amino acids known to interact with the membrane, PIP₂ or ATP in the PI3Kα^WT and PI3Kα^H1047R oncogenic mutant displayed on the PI3Kα^WT PDB structure 4jps [16]. Left panel: Amino acids contributing to the interaction with ATP and PIP₂ from Ref's [5,8] are shown as magenta sticks on the 4jps structure. Regions involved in membrane binding from HDX data in Ref. [6] are coloured blue for PI3Kα^WT and cyan for additional regions in PI3Kα^H1047R. Right panel: Surface representation of the left panel (C) p110α domain organisation and locations of different p110α oncogenic mutations used in this study relative to the ATP binding site occupied by BYL-719 in 4jps.

Figure 2.. ATP-competitive inhibitors affect PI3Kα^WT membrane binding in FRET and BLI assays.

(A) Inhibitors in the PI3K ATP binding sites. A66 was modelled into the active site of apo PI3Kα by molecular docking [37], ZSTK474 was crystallised in the p110δ active site (PDB entry 2wxl [25]) and GSK2126458 was crystallised in the p110γ active site with a structural water molecule shown (red sphere; PDB entry 3l08 [38]). (B and C) Effect of ATP-competitive inhibitors on PI3Kα^WT and PI3Kα^H1047R membrane interactions. FRET I-I₀ values are expressed as a percentage of the FRET signal for each inhibitor at its lowest concentration. Inhibitors were tested from 2 μM down to 31.25 nM using 2-fold dilution steps. The experiment was repeated at least 2 times and data from a representative experiment is shown as the Mean ± SEM (n = 3). Curves were fit in Prism using a dose-response model with four parameters, including a variable Hill slope. (B) PI3Kα^WT was used at 0.5 μM with the P1pY peptide. (C) PI3Kα^H1047R was used at 0.2 μM without peptide. (D and E) Effect of ATP-competitive inhibitors on the BLI wavelength shift (λ) generated by 1 µM PI3Kα^WT binding to immobilised liposomes. (D) PI3Kα^WT liposome binding after 300 s with 5% DMSO only, 4 μM GSK2126458 (GSK) or 20 μM PIK-75 (no. of biosensors ≥4). PI3Kα^WT was activated with P2pY for the BLI experiments. (E) Representative sensogram showing association of PI3Kα^WT with liposome loaded biosensors over 300 s combined with a 300 s dissociation sensogram.

Figure 3.. PI3Kα^WT membrane binding is more sensitive to GSK2126458 than the PI3Kα^H1047R oncogenic mutant.

BLI concentration response curves for P2pY activated PI3Kα^WT (A) and PI3Kα^H1047R (B) binding to liposomes in the presence of DMSO only or GSK2126458. An inhibitor to protein concentration ratio was maintained at 4:1. Each data point is shown as Mean ± SEM (no. of biosensors ≥3). The data were modelled in Prism using a one-site, specific binding model.

Figure 4.. PI3Kα^WT membrane binding is more sensitive to GSK2126458 than the PI3Kα^H1047R oncogenic mutant, and the influence of GSK2126458 is unaffected by anionic lipid composition.

BLI wavelength shift (λ) data for 1 µM of P2pY activated (A) PI3Kα^WT and (B) PI3Kα^H1047R binding to mixed lipid liposomes with different amounts of PIP₂ and phosphatidylserine (PS) in the presence and absence of 4 µM GSK2126458; data is shown as Mean ± SEM (no. of biosensors ≥3).

Figure 5.. Benzene sulphonamide containing ATP competitive inhibitors are more effective at disrupting PI3Kα^WT membrane binding.

(A) FRET data for 0.5 μM P2pY activated PI3Kα^WT in the presence of compounds 1 (blue), 2 (red), 3 (black) and 5 (magenta) across a concentration range from 125 nM to 8 μM. FRET I-I₀ values are expressed as a percentage of the FRET signal for each inhibitor at its lowest concentration. The experiment was repeated at least twice and representative data are shown as the Mean ± SEM (n = 3). Curves were fit with Prism using a four-parameter dose response model with a variable Hill slope. (B) BLI wavelength shift (λ) data for P2pY activated PI3Kα^WT (1 µM) binding to immobilised liposomes in the presence or absence of inhibitor (GSK2126458, 4 µM; A66, 20 µM; compound 1, 50 µM; compound 2, 50 µM; compound 4, 50 µM; compound 5, 50 µM; n = ≥4 biosensors; data shown as Mean ± SD). (C) Molecular docking model of 5 [37] in PI3Kα^WT (PDB code 2rd0 [56]) superimposed on the crystal structure of p110γ with GSK2126458 bound (PDB code 3l08 [38]). PI3Kα^WT amino acids are displayed. The red sphere represents a water molecule included in the docking model of 5, and the cyan sphere represents a water molecule observed in the crystal structure for GSK2126458. (D) BLI wavelength shift (λ) data for P2pY activated PI3Kα^WT (1 µM) membrane binding in the presence of GSK2126458 (GSK; 4 µM), 6 (4 µM), 7 (10 µM). Results are shown as the Mean ± SD; no. of biosensors used: DMSO n = 16; GSK n = 10; 6 n = 9; 7 n = 8.

Figure 6.. GSK2126458 only affects receptor activated PI3Kα^WT.

(A) Effect of P2pY (pY) activation on the PI3Kα^WT membrane binding BLI response in the presence and absence of GSK2126458 (GSK). The data are shown as Mean ± SD; PI3Kα^WT +pY n = 7; PI3Kα^WT +pY + GSK n = 5; PI3Kα^WT −pY n = 4; PI3Kα^WT −pY + GSK n = 3. (B) Effect of GSK on the BLI wavelength shift (λ) on membrane binding by PI3Kα^WT, the PI3Kα^E545K oncogenic mutant, class IA PI3Kβ^WT and class IA PI3Kδ^WT enzymes. In all experiments inhibitors were tested at 4 µM with 1 µM of the different enzymes and membrane binding was followed by BLI for 300 s. The graph shows the wavelength shift in the presence or absence of compound as Mean ± SD; PI3Kα^WT +pY n = 14; PI3Kα^WT +pY + GSK n = 8; PI3Kα^E545K +pY n = 10; PI3Kα^E545K +pY + GSK n = 4; PI3Kβ^WT + pY n = 4; PI3Kβ^WT + pY + GSK n = 4; PI3Kδ^WT + pY n = 4; PI3Kβ^WT + pY + GSK n = 5.

Figure 7.. Nucleotide binding increases PI3Kα^WT membrane binding.

Effect of GSK2126458 (GSK) and the non-hydrolysable ATP analogue AMP-PNP on the BLI wavelength shift (λ) for membrane binding by PI3Kα^WT in the presence and absence of 5 mM Mg²⁺. GSK was used at 4 µM with 1 µM of P2pY activated PI3Kα^WT and AMP-PNP was used at 2 mM. Membrane binding was followed by BLI for 300 s. The graph shows the wavelength shift in the presence or absence of compound as Mean ± SD; PI3Kα^WT DMSO −Mg²⁺ n = 14, DMSO + Mg²⁺ n = 7, GSK −Mg²⁺ n = 4, GSK+Mg²⁺ n = 4, AMP-PNP −Mg²⁺ n = 3, AMP-PNP +Mg²⁺ n = 5.