Tumour-specific STING agonist synthesis via a two-component prodrug system

Abstract

Pharmacological activation of STING holds promise in cancer treatment. A recent trend is the development of tumour-specific or conditionally activated STING agonists for enhanced safety and efficacy. Here we explore an unconventional prodrug activation strategy for on-tumour synthesis of a potent agonist. Leveraging the unique mechanism of MSA2, a small-molecule agonist that dimerizes non-covalently before binding to STING, we showed that its analogues bearing reactive functional groups readily and selectively form covalent dimers under mild conditions and in complex environments. We identified a reacting pair that led to a thioether-linked dimer with submicromolar potency in cell-based assays. Caging one of the reactants with a self-immolative β-glucuronide moiety resulted in a two-component prodrug system that near-exclusively formed the active compounds in tumours overexpressing β-glucuronidase. These results exemplify the use of small-molecule recognition for on-site generation of active compounds from benign precursors.

Fig. 1. Activation of a two-component prodrug system based on the unique mechanism of MSA2.

a, The unique predimerization mechanism of STING agonist MSA2 discovered by Merck. b, Our two-component prodrug strategy with enhanced tumour specificity for on-site synthesis of a potent STING agonist. Nu, nucleophile. Panels a and b created with BioRender.com.

Fig. 2. Facile formation of covalent dimers from MSA2 analogues.

a, Structures of reactive MSA2 analogues. b, HPLC traces of the dimerization reactions. Each reacting pair (50 μM) was allowed to react for 2 h under physiological conditions (37 °C, 10 mM NaPi pH 7, 150 mM NaCl). Reactions were quenched with iodoacetamide (500 μM) to block the remaining free thiol. A260, absorbance at 260 nm; a.u., arbitrary unit; abs, absorbance. See Supplementary Fig. 1 for a detailed analysis of all identified species. c, Structures of electrophiles used as non-specific competitors. d, A colorimetric assay for rate constant measurement. e, A summary of calculated rate constants (mean ± s.d.). Data were averaged from three independent experiments performed on freshly prepared compound dilutions. See Extended Data Fig. 1 for the original plots. f, Top: HPLC traces of the N1–E4 dimerization reaction (50 μM each, 37 °C, 10 mM NaPi pH 7, 150 mM NaCl, 50 µM TCEP) in the presence of competitors. Reactions were quenched with iodoacetamide after 2 h. Right: yields calculated by comparing the integration value of the dimer peak using a calibration curve constructed with pure standards. See Supplementary Figs. 3 and 4 for details. g, Left: non-covalent dimerization between the non-reactive E4 analogue (E4-ctrl) and N1. Right: dose-dependent ¹H NMR chemical shift perturbations of E4-ctrl by N1. [E4-ctrl] = 2 mM; 10 mM NaPi pH 7, 150 mM NaCl, 100 μM TCEP in D₂O; 25 °C. Source data

Fig. 3. In situ formation of a potent thioether-linked dimer from MSA2 analogues.

a, The structure of SC2S dimer formed from N1 and E4 under mild conditions. b, Left: dose–response curves of purified SC2S (red) and MSA2 (blue) on THP-1 Lucia ISG reporter cells. RLU, relative luminescence unit, reported as folds compared with DMSO-treated controls. Curves shown are averaged from three biological replicates, with cells that were split and passaged at least once. EC₅₀ values are reported as mean ± s.d. Right: representative ITC data for binding constant determination of SC2S and purified LBD of wild-type human STING (hSTINGwt, Supplementary Fig. 5). c, Phosphorylation of major STING pathway effectors in THP-1 Lucia ISG cells treated with purified SC2S dimer (5 μM). The experiment was repeated once with similar results. d, Structures of thioether-containing dimers D1–7 with systematically explored linker lengths. e, Left: dose–response curves of dimers D1–7 on THP-1 Lucia ISG reporter cells. Right: THP-1 EC₅₀ values and binding constants with hSTINGwt of D1–7. Cellular EC₅₀ values were calculated from three biological replicates and reported as mean ± s.d. Binding constants were averaged from duplicate measurements. See Extended Data Fig. 2 for all ITC data. f, Binding mode of D5 with STING LBD. Left: crystal structure of D5-bound STING (PDB: 9QVT; dark and light green) superimposed on 2′,3′-cGAMP-bound STING (PDB:4KSY; dark and light grey). Right: key interacting residues with D5. All structural representations were generated using PyMOL. See Extended Data Fig. 4 for more structural analysis. g, Top: workflow to test the in situ formation of SC2S on THP-1 cells. Analogues N1 and E4 were serially diluted in two separate 96-well plates and mixed to generate an 8 × 8 matrix of various concentration combinations. The crude mixtures were then seeded with THP-1 cells. IFN-β concentrations in the supernatants were then measured by ELISA. Bottom: IFN-β secretion levels of THP-1 cells treated with N1 and E4 combined at different concentrations. The response from cells treated with 25 μM of purified SC2S dimer was defined as 100%. Panel g created with BioRender.com. Source data

Fig. 4. β-Glucuronidase-specific N1 decaging and subsequent in situ STING agonist formation.

a, Cleavage of the glycosidic bond of β-GlcA-N1 (caged N1) releases a hemithioacetal, which collapses in aqueous environments and liberate the free N1. Liberated N1 could then react with E4 to form the active SC2S dimer. b, HPLC traces of reactions between β-GlcA-N1 and E4 (250 µM each) in the absence or presence of E. coli β-glucuronidase (250 Sigma units as defined by the manufacturer). 37 °C, 75 mM potassium phosphate pH 6.8, 1-h incubation. See Supplementary Fig. 6 for a detailed analysis of all identified species. c, Kinetics of β-GlcA-N1 cleavage mediated by human β-glucuronidase (5 μg ml⁻¹, 67 nM). 37 °C, 0.2 M sodium acetate pH 4.5. V, velocity of the reaction. Each data point was averaged from duplicate measurements. d, Cytokine secretion assays for β-glucuronidase-specific STING activation of THP-1 Lucia ISG cells. Luciferase: reporter gene triggered by interferon pathway; E4-control: E4 substitute without the reactive chloride (see Fig. 2g for structure). The final concentration of each compound was 25 μM. The final concentration of β-glucuronidase was 250 units ml⁻¹. Data are represented as mean ± s.d. from three biological replicates with cells that had been split and passaged at least once. Significance was analysed by two-tailed Welch’s t-test. Source data

Fig. 5. β-GlcA-N1/E4 showed efficacy in the Hs578T zebrafish xenograft model in the presence of β-glucuronidase.

a, Experimental design. Hs578T cells were injected into the PVS of 2 dpf zebrafish embryos. At 1 dpi, xenografts were randomly distributed into five treatment groups: DMSO, MSA2 (15 μM), SC2S (0.62 μM), β-GlcA-N1/E4 (20 μM each) and β-GlcA-N1/E4 + enzyme (20 μM each + 250 units ml⁻¹ of β-glucuronidase) with daily renewal of the compounds. At 4 dpi (3 dpt), xenografts were fixed and analysed for apoptosis, tumour size, phagocytosis and macrophage infiltration. Macrophage polarization analysis was performed at 2 dpi (1 dpt) and 4 dpi (3 dpt) by live imaging. All compounds were administrated at concentrations below their MTCs (Extended Data Fig. 7a). mac, macrophages; enz, enzyme (β-glucuronidase). b, Treated xenografts at 4 dpi. Blue, DAPI; white, activated Caspase-3. c, Quantification of Caspase-3 activation. Act., activated. d, Quantification of tumour sizes (number of tumour cells). For c and d, the results of two independent experiments were averaged and are presented as mean ± s.e.m. Each dot represents one xenograft. e, Representative confocal projection images of macrophages (red) in Hs578T xenografts in Tg(mpeg1:mcherry-F) at 4 dpi. f, Fold induction of TAM (defined by the ratio of the total number of macrophages and that of tumour cells) normalized to control treatments at 4 dpi. Data were averaged from two independent experiments and are presented as mean ± s.e.m. g, Representative confocal images of Hs578T xenografts injected in Tg(mpeg1:mcherry-F, tnfa:GFP-F) at 2 and 4 dpi. Red, macrophages; green, tnfa-positive cells; yellow, macrophages expressing TNF (that is, M1-like macrophages). h, Ratios of M1- to M2-like macrophages in the TME at 2 dpi. i, Ratios of M1-to M2-like macrophages at 4 dpi. For h and i, data are presented as mean ± s.e.m. Images are maximum intensity projections. All images are anterior to the left, posterior to right, dorsal up and ventral down. Scale bars, 50 μm. Dashed lines delineate the tumours. All datasets were challenged by D’Agostino and Pearson and Shapiro–Wilk normality tests. Those with a Gaussian distribution were analysed by parametric unpaired t-test, and those that did not pass the normality tests were analysed by non-parametric unpaired Mann–Whitney test. All tests were two-sided. Source data

Fig. 6. In vivo proof of concept of the two-component prodrug on a β-glucuronidase-overexpressing mouse tumour model.

a, Experimental design. CT26 cells overexpressing the murine β-glucuronidase (CT26mβGUS) were inoculated into the right flank of BALB/c mice. Overexpression of β-glucuronidase was confirmed by western blotting (Extended Data Fig. 9a). Treatments began once the tumour volume reached about 90 mm³. MSA2: 5 mg kg⁻¹ (IT); β-GlcA-N1: 12.5 mg kg⁻¹ (IT or IP); E4: 5 mg kg⁻¹ (IT); SC2S dimer: 8.6 mg kg⁻¹ (IT, matching molar concentration of the β-GlcA-N1/E4 treatment group). n = 6 for each treatment group. IP, intraperitoneal; IT, intratumoral. b, Tumour volume curves. Data were analysed by one-way ANOVA, followed by multiple comparisons adjusted with the Šidák–Holm correction. c, Body weight monitoring. d, Survival curves. Differences between the compound-treated groups and the vehicle-treated control group were analysed by the log-rank test. e, Tissue distribution of SC2S dimer formed after the β-GlcA-N1 (IP) + E4 (IT) treatment. Here, 0.1 ng g⁻¹ indicates that [SC2S] was below the detection limit. All data points are presented as mean ± s.e.m. n = 3 for each timepoint. Panel a created with BioRender.com. Source data

Extended Data Fig. 1. Original absorbance data for deriving second order alkylation rate constants.

a Absorbance time-course data. b Linearized second-order rate plots. [A]₀ = 100 μM, initial concentration of the nucleophile; [B]₀ = 250 μM, initial concentration of the electrophile. Each data point is presented as mean ± STD from three independent measurements. Validity of the assay was confirmed by LC-MS analysis of selected reactions at the end of the assay, which showed the alkylated thiol as the sole product. See Supplementary Fig. 2 for details. Source data

Extended Data Fig. 2. Isothermal titration calorimetry (ITC) data for binding constant measurements.

Purified hSTINGwt138-379 (250 μM of homodimer) was titrated into each dimer (25 μM) dissolved in a matching buffer (25 mM HEPES pH 7.5, 150 mM NaCl) at 25 °C and 750 rpm stirring. Binding constants were reported as the average from duplicate measurements.

Extended Data Fig. 3. Molecular dynamics (MD) simulations of STING complexed with different dimers.

a Representative frame from 1 µs MD simulations. The key interacting residues (S162, T263, Y167, R232, and R238) are highlighted. Hydrogen bonds are depicted as yellow dashed lines. b Theoretical binding energy (relative to compound D7) of each dimer calculated using the MM-PBSA method. See the Method section for details. c Distance between the aromatic ring center of Y167 and the carbon atom of the methylene group (CO-CH₂-) of the dimers over the course of the simulation, reflecting the stability of this interaction. d Population of hydrogen bonds observed in the MD simulation trajectory of each complex.

Extended Data Fig. 4. Additional structural analysis of D5-bound hSTING LBD.

a Cartoon representation of the dimeric crystal structure of STING LBD bound to compound D5 at 2.31 Å resolution, followed by a 90° rotated view. The displayed distance was measured between the α-carbon (Cα) of Y186 in both chains (as spheres). Superposition of apo (PDB: 4EMU) and D5-bound (PDB: 9QVT) STING structures, aligned by chain A and shown in cartoon representation. The distances displayed were calculated between the Cα atoms of residue Y186 in chains A and B (represented as spheres). b Electron density map (2Fo-Fc, contoured at 1.0 σ) of the ligand, overlaid with its stick representation. c Superposition of D5- and 2’,3’ cGAMP-bound STING LBD crystal structures, emphasizing the angle formed by the C–S–C bonds in the linker region of D5. In both views, it is apparent that D5 penetrates deeper into the binding pocket and is also narrower at the base of the site compared to 2’,3’-cGAMP. d Superposition of D5- (PDB: 9QVT) and MSA2-bound (PDB: 6UKM) STING LBD crystal structures, highlighting conformational differences in chain B when chains A are aligned.

Extended Data Fig. 5. LC-MS analysis of the reaction between N1 and E4 in the presence of glutathione (1 mM).

N1 (50 μM) and E4 (50 μM) were reacted at neutral pH (25 mM HEPES pH 7.0, 150 mM NaCl) and in the presence of reduced glutathione (1 mM). The reaction mixture was incubated at 37 °C for 2 h. No glutathione-adduct of E4 was identified.

Extended Data Fig. 6. Stability test of β-GlcA-N1 (1 mM) in human plasma.

At the indicated time point, an 80 µL aliquot was loaded onto an Agilent Bond Elut solid phase extraction column (C18, 3 mL bed volume). The collected eluent was lyophilized, redissolved in a 50% (v/v) acetonitrile in water (0.1% v/v formic acid), and analyzed by LC-MS.

Extended Data Fig. 7. Additional zebrafish experiment data.

a MTC of β-GlcA-N1/E4 treatments on non-tumor zebrafish larvae. Groups of 30 noninjected zebrafish larvae were exposed to different concentrations of the compounds (MSA2, SC2S, β-GlcA-N1/E4 and β-GlcA-N1/E4 + β-glucuronidase) for 3 consecutive days, with daily renewal of the E2 media containing the drugs. The tested concentrations of the compounds, in different combinations, were the following: MSA2: 15 μM, 75 μM and 150 μM; SC2S: 0.62 μM, 3.1 μM and 6.2 μM; β-GlcA-N1/E4: 5 μM + 5 μM, 10 μM + 10 μM, 20 μM + 20 μM and 100 μM + 100 μM; β-GlcA-N1/E4 + β-glucuronidase: 20 μM + 20 μM + 166 U/mL, 20 μM + 20 μM + 250 U/mL and 20 μM + 20 μM + 500 U/mL. Toxicity was assessed daily by counting the total number of dead larvae and checking for the presence of morphologic changes such cardiac edemas or curved tails. Control groups were treated with the highest matched DMSO concentration (0.04%, v/v). b Left: representative image of phagocytosis in a xenograft treated with β-GlcA-N1/E4 and β-glucuronidase at 2dpi in Tg(mpeg1:mcherry-F). Each yellow arrow indicates a phagocytic event. Right: n° of phagocytic events at 2dpi. Data are presented as mean ± SEM. Statistical analysis follows that of Fig. 5. Source data

Extended Data Fig. 8. β-GlcA-N1/E4 tested on a MC38 syngeneic mouse model.

a Experimental design. MC-38 cells were inoculated into the right flank of C57BL/6 mice and treated once tumors reached about 90 mm³. MSA2: 5 mg/kg. β-GlcA-N1: 12.5 mg/kg; E4: 5 mg/kg; N1 + E4: N1 12.5 mg/kg and E4 5 mg/kg; β-GlcA-N1/E4 + enz: 12.5 mg/kg of β-GlcA-N1 and 5 mg/kg of E4, preincubated with 500 units of E. coli β-glucuronidase at 37 °C for one hour. N1 was administered intraperitoneally (IP); all other compounds were administered intratumorally (IT). b Tumor growth curves. Data were analyzed by one-way ANOVA, followed by multiple comparisons adjusted with the Sidák-Holm correction. c Survival curves. Differences between the compound-treated groups and the vehicle-treated control group were analyzed by the Log-Rank test. d Body weight monitoring. e Cytokine profiling of sera obtained on Day 9. Source data

Extended Data Fig. 9. Additional data of CT26mβGUS mouse tumor model experiment.

a Confirmation of β-glucuronidase overexpression in CT26mβGUS cells by Western blots. b Haematotoxicity data. MCV: Mean Corpuscular Volume; MCH: Mean Corpuscular Haemoglobin; WBC: White Blood Cell Count; PLT: Platelet Count; RDW-CV: Red Cell Distribution Width - Coefficient of Variation; MPV: Mean Platelet Volume. RBC: red blood cell count; HGB: haemoglobin; HCT: haematocrit; MCV: Mean Corpuscular Volume; MCH: Mean Corpuscular Haemoglobin; WBC: White Blood Cell Count; PLT: Platelet Count; RDW-CV: Red Cell Distribution Width - Coefficient of Variation; MPV: Mean Platelet Volume. Data were analyzed by two-way ANOVA, followed by multiple comparisons adjusted with the Sidák-Holm correction. Source data

Extended Data Fig. 10. Histopathology analysis of non-targeted tissues.

Organs were harvested, fixed in 10% formalin, embedded in paraffin, sectioned into 3 μm-thick sections, and stained with Hematoxylin and Eosin (H&E). Whole sections were acquired with the NanoZoomer-SQ Digital slide scanner (Hamamatsu) and pathology assessment was performed by an experimental pathologist (PF) at GIMM’s Histopathology Facility. Scale bars: 100 μm.