Chemical Screen of novel GCN2-specific ISR-activating molecules.

(A) Schematic of ATF4::NanoLuc reporter, exploiting the upstream open reading frames (uORFs). Short uORFs 1, 2 and 3 precede the ATF4 translation start site, with uORF3 overlapping the ATF4 coding sequence, but out-of-frame. During the ISR, skipping of uORF3 leads to increased translation initiation at the ATF4 initiator AUG. (B) Relative Light Unit (RLU) of ATF4 signal in CHO stable cell-line overexpressing ATF4::NanoLuc treated with 1mM histidinol, 2.5 µg/mL tunicamycin, or 1 µM latrunculin A (n=3; ± SEM). (C) Chemical screen of 6,600 small molecules tested with CHO stable cell-line overexpressing ATF4::NanoLuc. The molecules were treated alone or with submaximal ISR activation with 0.3mM histidinol. All molecules were displayed along the x-axis based on their z-score of relative reporter signal when treated with compound alone (pale symbols). Z-score for treatment with compound and histidinol is also plotted at the same x-axis position (dark symbols). Molecules with z-score > 3 (i.e., over 3 standard deviations from the mean reporter signal) were identified as hits: ISR activator in orange and ISR augmenter (exaggerating ISR activation by histidinol) in green. (D) Pipeline of molecule short-listing, primary screening using CHO ATF4::NanoLuc reporter (153 hits), NanoLuc alone control (134 hits), selection based on cytotoxicity (130 hits), orthogonal CHO ATF4::FLuc reporter (89 hits), and subsequent GCN2 specificity in GCN2 null reporter cells (30 hits) and compound quality control via LC-MS (16 hits), finally validation by GCN2-specific FACS to a short list of 3 compounds.

Shortlist and characterisation of three top hits (compound 18, 20, and 21)

(A) Representative dot plot of flow cytometry data in CHO CHOP::GFP and XBP1::Turquoise dual-reporter wildtype, Eif2ak4Δ, and Ppp1r15aΔ cells under 18-hour treatment of 0.75mM histidinol (blue), 2μg/mL tunicamycin (orange), and untreated control (gray) (n=2). Bi-exponential display was used for the plots. The x-axis represents fluorescence signals at 530nm for GFP (CHOP reporter) and y-axis represents fluorescence signals at 450nm for turquoise (XBP1/IRE1 reporter). (B) Representative bi-exponential displayed dot plot of flow cytometry data in CHO CHOP::GFP and XBP1::Turquoise dual-reporter wildtype, Eif2ak4Δ, and Ppp1r15aΔ cells under 18-hour treatment of 30µM compound 18, 20µM compound 20, and 7.5µM compound 21 alone (magenta) or in combination with 0.75mM histidinol (yellow) (n=6). Histidinol alone is shown in blue and untreated in grey. The x-axis represents fluorescence signals at 530nm for GFP (CHOP reporter) and y-axis represents fluorescence signals at 450nm for turquoise (XBP1/IRE1 reporter). Structures of compounds 18, 20, and 21 shown on the right. (C) Quantification of B (n=6); Two-way ANOVA with Tukey’s multiple comparisons test was performed and the CHOP-GFP signals (530nm) in different treatment groups were compared. *: p ≤ 0.05, **: p ≤ 0.01, ***: p ≤ 0.001, ****: p ≤ 0.0001. (D) Box and whisker plot displaying mRNA fold change of ISR effector CHOP and PPP1R15A after 4-hour treatment of 30µM compound 18, 10µM compound 20, and 10µM compound 21 as detected by qPCR (n=3). ± SEM, Two-way ANOVA was performed, * p≤0.05, ** p≤0.01.

Compound 18, 20, and 21 display cell type-specific differences in ISR activation.

(A) Normalised dose-dependent fold change in ATF4 expression by neratinib in HeLa WT, HeLa EIF2AK4Δ, and CHO WT ATF4::NanoLuc reporter cells after 16-hour treatment (n=3; mean ± SEM). DMSO was used as vehicle. (B) Normalised dose-dependent fold change in ATF4 expression by compound 18, 20, and 21 in CHO WT ATF4::NanoLuc reporter cells after 16-hour treatment (n=3; mean ± SEM). EC50 was estimated using 4-parameter non-linear regression. DMSO treatment was used as vehicle. (C) Normalised dose-dependent fold change in ATF4 expression by compound 18, 20, and 21 in HeLa (black) and 293T (blue) WT cells transiently transfected with ATF4::NanoLuc reporter after 16-hour treatment (n=3, ± SEM). DMSO treatment was used as vehicle. (D) Relative Light Unit (RLU) of of ATF4 signal in CHO WT ATF4::NanoLuc reporter cells treated with 1, 3, 10, 30µM Compound 18, 20, and 21 alone (black) or in combination with 1µM GCN2iB (pink) for 18 hours (n=3; ± SEM) . 0.1, 0.3, 1, 3% DMSO served as vehicle and 3mM histidinol as positive control. Two-way ANOVA was performed *** p≤0.001, **** p≤0.0001.

Compound 20 displays GCN1 dependence.

(A) Relative Light Unit (RLU) of ATF4 signal in 293T WT (black) and GCN1Δ (magenta) cells transiently transfected with ATF4::NanoLuc reporter for 16-hour treatment with compound 20 (10 and 20µM), GCN2iB (5, 10, 35, 75nM), and neratinib (75, 200, 500, 1000nM) (n=3; ± SEM). Two-way ANOVA was performed * p≤0.05, ** p≤0.01, *** p≤0.001, **** p≤0.0001. (B) Normalised fold-change in ATF4 signal in CHO WT (black) and Gcn1-/- (blue) ATF4::NanoLuc reporter cells for 18-hour treatment with compound 20 (1, 3, 10 and 30µM) and neratinib (75, 200, 500nM) (n=3; ± SEM). Two-way ANOVA * p≤0.05, ** p≤0.01. (C) Normalised fold-change in ATF4 signal in CHO WT (black) and Gcn1-/- (blue) ATF4::NanoLuc reporter cells treated for 19-hours with a panel of ATP-competitive kinase inhibitors reported to activate GCN2 (0.1, 0.3, 1, 3, and 10µM) (n=3-9; mean ± SEM). Two-way ANOVA was performed * p≤0.05, ** p≤0.01, *** p≤0.001, **** p≤0.0001.

Compound 18 and 20 directly binds to GCN2 in cellulo and in vitro

(A) Illustration of the NanoBRET assay. Left panel displays BRET signal at baseline with tracer binding to GCN2; right panel displays reduced BRET signal due to the addition of a competitive compound that outcompetes the tracer from binding to GCN2. (B) BRET ratio in CHO and 293T WT cells with 2-hour treatment of 3% DMSO, 100µM compound 18, 20, and 21 or 0.5µM GCN2iB. (n=3; ± SEM). Two-way ANOVA was performed. *: p ≤ 0.05, **: p ≤ 0.01, ***: p ≤ 0.001, ****: p ≤ 0.0001. (C) Schematic of GST-GCN2 fusion protein (residues 585–1005) and representative immunoblot of in vitro kinase assay. Purified GST-GCN2 was incubated with eIF2α-NTD (residues 2–187) using a ramp of ATP concentrations (0, 0.01, 0.03, 0.1, 0.3, 1, 3, 10mM) (n=3). Total and serine51 phosphorylated eIF2α were detected. (D) Representative immunoblot of in vitro kinase assay showing inhibition of eIF2α-NTD (residues 2–187) phosphorylation. Purified GST-GCN2 and eIF2α-NTD was incubated with neratinib (0.075, 0.68, 6.67, 10, 20µM) and compound 18 (100-500µM), compound 20 (100, 200, 500µM), and compound 21 (100-500µM) (n=3). 5% DMSO used as vehicle control. Total and serine51 phosphorylated eIF2α were detected. Molecular size in kDa. (E) Representative immunoblot of in vitro kinase assay (n=3) showing inhibition of eIF2α-NTD (residues 2–187) phosphorylation by 300µM compound 18 using a ramp of ATP concentrations (0.01, 0.03, 0.1, 0.3, 1, 3, 10mM); 3% DMSO used as vehicle control. Bottom panel showed Michaelis-Menten kinetics analysis of DMSO control (black) vs compound 18 (green). Molecular size in kDa. (F) Representative immunoblot of in vitro kinase assay (n=3) showing inhibition of eIF2α-NTD (residues 2–187) phosphorylation by 100µM compound 20 using a ramp of ATP concentrations (0.01, 0.03, 0.1, 0.3, 1, 3, 10mM); 3% DMSO used as vehicle control. Bottom panel showed Michaelis-Menten kinetics analysis of DMSO control (black) vs compound 20 (green). Molecular size in kDa. (G) Schematic of full-length GCN2 protein and representative immunoblot of in vitro kinase assay. Purified full-length GCN2 was incubated with full-length eIF2α using a ramp of compound 20 or tRNA concentrations (0, 0.001, 0.005, 0.01, 0.05, 0.13, 0.4, 1.2, 3.7, 11, 33, 100µM) (n=3). Total and serine51 phosphorylated eIF2α, as well as total and threonine899 phosphorylated GCN2 were detected. On the right showed quantification of serine51 phosphorylated eIF2α with a ramp of tRNA (black) vs compound 20 (green). Molecular size in kDa.

Structural modelling

(A) Illustration of GCN2 kinase domain from up (N-terminus) to down (C-terminus) with illustrative ligand (yellow) in ATP-binding pocket. (B) 2D ligand interaction diagram of neratinib, (D) compound 18, and (F) compound 20. Violet, green, and red colored lining indicate hydrogen bond, π-π, and cation-π interactions respectively. Polar and hydrophobic residues are highlighted in blue and green respectively. Mesh representation of the ATP-binding pocket in GCN2 kinase domain was shown with ligand (yellow): (C) neratinib, (E) compound 18, and (G) compound 20. Key residues surrounding the ATP-binding pocket, including the DFG motif (D866-G868) and EYC residues (E803-C805) following the gatekeeper M802 were labelled red. Figures adapted from crystal structure 6N3N.