Coupling cellular drug-target engagement to downstream pharmacology with CeTEAM

Abstract

Cellular target engagement technologies enable quantification of intracellular drug binding; however, simultaneous assessment of drug-associated phenotypes has proven challenging. Here, we present cellular target engagement by accumulation of mutant as a platform that can concomitantly evaluate drug-target interactions and phenotypic responses using conditionally stabilized drug biosensors. We observe that drug-responsive proteotypes are prevalent among reported mutants of known drug targets. Compatible mutants appear to follow structural and biophysical logic that permits intra-protein and paralogous expansion of the biosensor pool. We then apply our method to uncouple target engagement from divergent cellular activities of MutT homolog 1 (MTH1) inhibitors, dissect Nudix hydrolase 15 (NUDT15)-associated thiopurine metabolism with the R139C pharmacogenetic variant, and profile the dynamics of poly(ADP-ribose) polymerase 1/2 (PARP1/2) binding and DNA trapping by PARP inhibitors (PARPi). Further, PARP1-derived biosensors facilitated high-throughput screening for PARP1 binders, as well as multimodal ex vivo analysis and non-invasive tracking of PARPi binding in live animals. This approach can facilitate holistic assessment of drug-target engagement by bridging drug binding events and their biological consequences.

Fig. 1. CeTEAM is predicated on stability-dependent biosensors to measure drug binding.

a U-2 OS V5-MTH1 G48E cells were treated with the indicated MTH1 inhibitors for 24 hours. b HCT116 3-6 3xHA-NUDT15 R139C cells were incubated with the indicated molecules for 72 hours. c U-2 OS PARP1 L713F-GFP cells were treated with PARP inhibitors for 24 hours. Biosensors were pre-induced with doxycycline, and all blots are representative from two independent experiments. d A schematic description of CeTEAM. Stability-dependent drug biosensors (blue) containing a destabilizing mutation (yellow) accumulate in the presence of binding ligand (pink) and detection can be facilitated by protein fusion tags (orange) to measure drug-target engagement. The presence of endogenous target protein (gray) and physiological conditions enable phenotypic multiplexing and discerning of on- from off-target effects of test compounds.

Fig. 2. Delineation and expansion of CeTEAM-permissive mutations with PARP1/2.

a Leucine residues of interest (magenta) within the PARP1 HD domain (PDBID: 7KK2, made with Protein Imager). b A representative western blot of inducible C-terminal GFP-tagged PARP1 variants in U-2 OS cells incubated with DMSO, 10 nM, or 1 µM veliparib for 24 hours. Black arrow – PARP1-GFP; gray arrow – endogenous PARP1. c Quantitation of PARP1-GFP variant fold change relative to DMSO control (light gray) following 10 nM (blue) or 1 µM veliparib (orange; related to a). GFP abundance normalized to β-actin and no (–) DOX controls are shown for each variant. Means of n = 3 ± SD. Means of n = 2 shown for –DOX controls. P values shown for two-way ANOVA with multiple comparisons to each DMSO control (Dunnett’s test; F [DFn, DFd]: F_Interaction [12, 42] = 4.899, F_{Row Factor} [6, 42] = 10.15, F_{Column Factor} [2, 48] = 52.01). d Comparison of HD mutant thermal stability changes from Langelier et al. to fold change (over DMSO control) after 1 µM veliparib treatment (from c). a – denotes reported values from Langelier et al., * – thermal stability change reported for L713A. e Live-cell fluorescence fold change of functional PARP1-GFP CeTEAM variants with veliparib (solid) or 3-AB (open/dashed) dose response after 24 hours. Means shown ± SEM (for n_L713F–v: 4, n_L713F–3AB: 3); means and range for all others (n = 2). f Overlay of PARP1 (blue, PDBID: 7KK2) and PARP2 (magenta, PDBID: 3KCZ) HD domains with L713/L269 denoted (made with Protein Imager). g A representative western blot (from n = 2) demonstrating stabilization of constitutive PARP2 L269A-GFP in U-2 OS cells by various PARPi after 24 hours (3-AB – 100 µM and 1 mM, iniparib – 20 µM, all others – 10 nM and 1 µM). A psuedocolor density depiction in RFU is also shown. h Example GFP fluorescence micrographs of PARP2 L269A-GFP after 24-hr PARPi treatment. Nuclei are demarcated by outlines and scale bar = 100 µm. i Live-cell PARP2 L269A-GFP fluorescence following 24-hr dose-response with either veliparib (orange), 3-AB (gray), or iniparib (black). Means of n = 5 ± SEM. FC fold change, RFU relative fluorescence units.

Fig. 3. Parsing differential pharmacology of equipotent MTH1 inhibitors with a V5-G48E drug biosensor.

a MTH1 inhibitors tested and their reported biochemical IC₅₀ values^,–. b A representative blot (n = 2) of induced V5-MTH1 WT or V5-MTH1 G48E in U-2 OS cells following incubated with DMSO, 2.5 µM TH588, 1 µM AZ19, 1 µM IACS-4759, or 1 µM BAY-707 for 24 hours. Black arrow – V5 MTH1; gray arrow – endogenous, WT MTH1. c A schematic depicting multiparametric CeTEAM analysis by flow cytometry of TH588 and AZ19 pharmacology by V5 (drug binding, AF-488), pHH3 Ser10 (mitotic marker, AF-647), and Hoechst (cell cycle) readouts using V5-MTH1 G48E clone 6 cells. d V5-G48E saturation profiles of TH588 (blue) and AZ19 (orange) after 24 hours by western blot. Mean of n = 3 ± range. Comparison made by two-sided extra sum-of-squares F Test (F [DFn, DFd] = 12.67 [1, 36]). Interpolated data points representing occupancy designations are shown in red and defined adjacent to data plots. e Median V5-G48E fold change (left axis; TH588 – blue, AZ19 – orange) and percent pHH3 Ser10⁺ cells (right axis; yellow) of described occupancy designations by flow cytometry (mean of n = 3 ± SD). Reference line at V5 fold change = 1. P values shown for one-way ANOVA with multiple comparisons to the DMSO control (Dunnett’s test; F_HH3 [DFn, DFd] = 122.3 [6, 14], F_V5 [DFn, DFd] = 1.625 [6, 14]). f Proportion of cells in sub-G1 (light gray), G0/G1 (black), S (white), and G2/M (dark gray) phases by Hoechst intensity. Mean of n = 3 ± SD. Red highlight – P < 0.05. P_{Pre,TH588,SubG1} > 0.9999, P_{Pre,TH588,G0/G1} = 0.9998, P_Pre,TH588,S > 0.9999, P_{Pre,TH588,G2/M} = 0.9997, P_{Sat,TH588,SubG1} > 0.9999, P_{Sat,TH588,G0/G1} = 0.9251, P_Sat,TH588,S = 0.9978, P_{Sat,TH588,G2/M} = 0.9963, P_{Lit,TH588,SubG1} > 0.9999, P_{Lit,TH588,G0/G1} = 0.1700, P_Lit,TH588,S = 0.0934, P_{Lit,TH588,G2/M} = 0.0001, P_{Pre,AZ19,SubG1} > 0.9999, P_{Pre,AZ19,G0/G1} = 0.9992, P_Pre,AZ19,S=0.9948, P_{Pre,AZ19,G2/M} = 0.9950, P_{Sat,AZ19,SubG1} > 0.9999, P_{Sat,AZ19,G0/G1} = 0.9997, P_Sat,AZ19,S = 0.9997, P_{Sat,AZ19,G2/M} = 0.9995, P_{Lit,AZ19,SubG1>}0.9999, P_{Lit,AZ19,G0/G1=}0.7657, P_Lit,AZ19,S = 0.8923, P_{Lit,AZ19,G2/M} > 0.9999 by ordinary two-way ANOVA with multiple comparisons to the DMSO control (Dunnett’s test; F [DFn, DFd]: F_Interaction [18, 56] = 3.103, F_{Row Factor} [6, 56] = 0.01093, F_{Column Factor} [3, 56] = 552.0).

Fig. 4. Leveraging the NUDT15 pharmacogenetic variant, R139C, to decipher thiopurine pharmacology in cellulo.

a Representative microscopy images (from n = 2 independent experiments) of doxycycline-induced HCT116 3-6 3xHA-NUDT15 R139C cells treated with DMSO or 10 µM 6TG for 72 hours and stained with indicated markers. Hoechst staining is shown in the merged image. Scale bar=200 µm. b NUDT15 inhibition by thiopurine metabolites (n = 2 with lines of best fit). c Structures of NSC56456, TH8234, and TH8228 with moieties of interest highlighted in red. d NUDT15 inhibition by TH8228 (gray), NSC56456 (batch ID: BV122529; orange), and TH8234 (blue). n = 2 with lines of best fit. e Melting temperatures of NUDT15 WT (blue) and R139C (orange) with 50 µM NUDT15i by DSF assay compared to DMSO (gray). Means of n = 2. f A schematic depicting a high-content microscopy assay for simultaneous detection of target engagement (HA) and phenotypes (DNA damage response – γH2A.X, cell cycle – Hoechst) of potential NUDT15 inhibitors -/+ low-dose 6TG. g, h Representative per-cell three-dimensional analysis of γH2A.X (y-axis), Hoechst (x-axis), and HA intensities (white-orange-red gradient) following treatment with DMSO, 3.67 µM NSC56456, 3.67 µM TH8228, or 3.67 µM TH8234 ± 200 nM 6TG and compared to 3.33 µM 6TG alone. n = 500 cells per condition, except n_6TG = 399. i Binning of NUDT15i into non-responder/∅ (gray), stabilizer (yellow), potentiator or 6TG mimetic (red; NUDT15 binding-related 6TG potentiation), and non-specific (blue; NUDT15 binding-independent DNA damage) based on HA-R139C intensity and DNA damage induction. Stabilizers may reclassify to potentiators in the presence of 6TG. j Per-drug analysis of median HA (y-axis), γH2A.X (x-axis), and Hoechst intensities (symbol size) for NSC56456, TH8228, and TH8234 at multiple concentrations (white-magenta gradient) either alone (circles) or combined with 6TG (squares) and compared to DMSO (gray). RFU relative fluorescence units.

Fig. 5. Profiling of PARPi engagement and trapping in live cells with PARP1 L713F-GFP.

a Chemical structures and PARP1 inhibitory potencies of PARPi studied (SelleckChem and). b A representative blot (n = 2) of induced GFP-PARP1 WT and L713F-GFP in U-2 OS cells treated with PARPi for 24 hours. Black arrow – GFP-tagged PARP1; gray arrow – endogenous PARP1. c Experimental schematic for live cell tracking of PARP1 target engagement (GFP) and PARPi-induced replication stress (cell cycle, Hoechst) by high-content microscopy. Trapping depends on PARP1 engagement and replication stress. d Curve fitting of median GFP (solid) and Hoechst (DNA content; open/dashed) intensities in live, PARP1 L713F-GFP clone 5 cells incubated with talazoparib (blue), olaparib (purple), niraparib (red), veliparib (orange), 3-AB (gray), or iniparib (black) for 24 hours following DOX induction. Means from n = 2. e Summary of observed L713F-GFP stabilization and median DNA content EC₅₀ values for tested PARPi (in nM). f Concentration-dependent dynamics of PARP1 target engagement and DNA trapping in live cells after PARPi. Median GFP (y-axis) and Hoechst (x-axis) intensities are shown. Representative of n = 2, replotted from d. Light gray circle – –DOX control; dark gray circle – +DOX control; blue gradient circles – PARPi concentration gradient (3-AB – 12.8 nM to 1 mM; all other PARPi – 0.128 nM to 10 µM), red areas – PARP trapping phenotype. g Representative single cell, 2D plots comparing GFP intensity (y-axis) and Hoechst intensity (x-axis) following DMSO (gray) and PARPi treatment (orange; replotted from d). Inferred G1 and G2/M cell cycle phases demarcated by gray columns. Overview data in f representative of n = 500 cells per group; individual cell plots in g are n = 1000 cells per group. RFU – relative fluorescence units.

Fig. 6. A PARP1 biophysical perturbagen screen enabled by a L713F-nLuc dual luminescence assay format.

a The PARP1 L713F-nLuc biosensor (em: 460 nm) was paired with akaLuc (em: 650 nm) to enable sequential dual luciferase analyses. b Time-resolved detection of PARP1 L713F-nLuc stabilization following veliparib treatment and normalized to akaLuc signal. n = 2 with line of best fit shown. c Dose-dependent veliparib stabilization of different PARP1 L713F-nLuc abundances (DOX gradient) after 24 hours and normalized to akaLuc. n = 2 with line of best fit shown. d The MedChemExpress Epigenetics and Selleck Nordic Oncology libraries were screened (10 µM, 24 hours) with the L713F-nLuc/akaLuc system. Compounds was excluded if akaLuc intensity differed > 4 SDs from controls, leaving 840 compounds for further analysis. e Ranked, log₂-transformed L713F-nLuc/akaLuc ratios from 840 screening compounds (dark gray). Negative (light gray, DMSO) and positive controls (blue, 10 µM veliparib) are shown for reference. Hits were defined as at least 2 (orange) or 3 standard deviations (σ, red) from the screening library mean. Annotated PARPi are indicated with black borders and trapping DNMT compounds are labeled. f Detailed overview of positive screening hits (n = 47). Non-PARPi were triaged by target class, contextualized by hit rate within the general target class, and by anecdotally defined primary target/compound class. g Hit rates of PARPi within the screening library by increasing stringency (general PARPi → PARP1i → PARP1i [IC₅₀ < 1 µM]) and numbers of qualifying compounds. Hit proportions are shown in blue, while non-hits are gray. * – PJ34 missed the akaLuc cut-off. h Hit confirmation of PARPi (orange) and non-PARPi (yellow) positive screening hits. Identical positive (blue) and negative controls (gray) are used from the screen, and means of n = 24 (negative, positive control), n = 3 (linifanib to fluzoparib), or n = 6 (pamiparib to AZD5305) data points are shown ± SD. Names of statistically significant compounds in red, and confirmed non-PARPi are summarized by primary target hit rate (final target share). P values are shown for one-way ANOVA analysis with comparisons to DMSO control (Dunnett’s test; F_Treatment [DFn, DFd] = 200.9 [48, 202]). FC fold change.

Fig. 7. In vivo-optimized CeTEAM PARPi-GFP biosensors for ex vivo detection of drug-target engagement.

a The PARP1 L713F-GFP biosensor was paired with mCherry for normalization. b Representative fluorescent micrographs (from n = 3) of U-2 OS PARP1 L713F-GFP/mCherry cells treated with DMSO or 1 µM veliparib for 24 hours. Scale bar = 50 µm. c mCherry-normalized L713F-GFP signal from 24-hour veliparib by flow cytometry. Modal normalization is shown. Representative of n = 2. Gray-blue gradient – veliparib gradient. d Graphical overview of in vivo experiments with HCT116 subcutaneous xenografts constitutively expressing PARP1 L713F-GFP and mCherry treated with either niraparib or vehicle control (2x, qd). n_Vehicle: 4, n_15mg/kg: 3, n_60mg/kg: 4 mice per treatment group. e Representative flow cytometry histograms of PARP1 L713F-GFP/mCherry tumors. Modal normalization is shown. Veh₄: vehicle (mouse #4; gray), 15mg₃: 15 mg/kg (mouse #3; blue), and 60mg₄: 60 mg/kg (mouse #4; orange). f mCherry-normalized L713F-GFP intensity of tumors by flow cytometry. Means with 95% confidence intervals are shown from n = 3 (15 mg/kg) or n = 4 mice (Vehicle, 60 mg/kg). g Gross L713F-GFP signal from individual tumors by western blot. h mCherry-normalized L713F-GFP abundance from western blots in g and Supplementary Fig. 19e. Means with 95% confidence intervals are shown from n = 3 (15 mg/kg) or n = 4 mice (Vehicle, 60 mg/kg). i Representative L713F-GFP micrographs from tumor sections with fire LUT pixel density depiction. Scale bars=100 µm. j Floating histogram of L713F-GFP intensities across tumors and treatment groups (gray – vehicle, blue – 15 mg/kg, orange – 60 mg/kg). Blue region represents an arbitrary cut-off of GFP intensity ≥ 0.075 RFU. 16,482 total cells per treatment group. k Distribution of individual cell L713F-GFP intensities ≥0.075 RFU. Violin plots with median (thick line) and quartiles (thin lines) are overlayed onto individual datapoints. In all cases, P values are shown for one-way ANOVA (Dunnett’s test; f and h; F_{Treatment (f)} [DFn, DFd] = 32.64 [2, 8], F_{Treatment (h)} [DFn, DFd] = 17.384 [2, 8]) or Kruskal-Wallis test (Dunn’s test; k; Kruskal-Wallis statistic = 160.0) with multiple comparisons to the vehicle control. RFU relative fluorescence units.

Fig. 8. An in vivo-compatible CeTEAM PARPi-nLuc biosensor for multiplexed tracking of drug binding in live animals.

a Graphical overview of in vivo experiments with constitutive expression PARP1 L713F-nLuc/akaLuc subcutaneous HCT116 xenografts treated with either vehicle or 60 mg/kg niraparib. n = 7 total mice per group. b Representative bioluminescence (radiance) overlays of mice treated as in a following administration of fluorofurimazine (nLuc) or AkaLumine HCl (akaLuc). Radiance intensity in psuedocolor representation. c akaLuc-normalized quantification of in vivo L713F-nLuc signals following vehicle (gray) or niraparib (orange) treatment from two experimental arms. d Ex vivo L713F-nLuc (blue gradient; 1,149,410 to 17,154,040 RLU) and akaLuc (circle size; 137,106 to 401,261 RLU) luminescence intensity representations of each tumor. e Quantitation of ex vivo, akaLuc-normalized L713F-nLuc bioluminescence following vehicle (gray) or niraparib (orange) treatment. For c and e, means (n_c=7; n_e = 4) and 95% confidence intervals are shown, as well as P values from unpaired, two-tailed t tests (t, df_c = 4.969, 12; t, df_e = 4.871, 6).