Cutting Edge: Antimalarial Drugs Inhibit IFN-β Production through Blockade of Cyclic GMP-AMP Synthase–DNA Interaction

Evidence for the involvement of type I IFNs (IFN-Is) in the pathogenesis of systemic autoimmune disorders, such as systemic lupus erythematosus (SLE) (1), as well as in rare monogenic disorders called “IFNopathies,” which include Aicardi–Goutières syndrome (AGS), is demonstrated by increased expression of IFN-I–stimulated genes (ISGs) (2). How, where, and when IFN-I is initially stimulated, as well as which of the IFN-I subtypes are expressed in each disease, have been difficult to determine. In vitro studies, including our own (3), reveal that IFN-α is induced by SLE immune complexes (4). However, family studies of individuals without autoantibodies (5), direct evidence of IFN-β involvement (6, 7), and incomplete neutralization of ISGs in clinical trials using biologics targeting IFN-α suggest that other IFN-Is are involved in SLE (8). Approximately 25% of patients with AGS and 1–2% of SLE patients have mutations in the 3-5′ DNA exonuclease TREX1 (9, 10). TREX1 deficiency in mice leads to the accumulation of intracellular DNA and cell-intrinsic production of IFN-β (11).

Although many DNA sensors have been described, the recent discovery of cyclic GMP-AMP synthase (cGAS) (12) is of particular interest because it appears to be necessary and sufficient for stimulation of IFN-β in TREX1-deficient cells (13). Following binding to dsDNA, cGAS undergoes a conformational change, revealing a catalytic active site that synthesizes the cyclic dinucleotide, cyclic dinucleotide GMP-AMP (cGAMP) (14). cGAMP binds to the adapter protein, stimulator of IFN genes (STING), which triggers activation of TANK-binding kinase 1 and phosphorylation of IFN regulatory factor 3, with resulting transcription of IFN-β (12, 15).

In view of the key role that cGAS plays in DNA-stimulated IFN-β production and the suspected role of this pathway in some diseases associated with high expression of ISGs (16), we performed in silico screening of chemical and drug libraries to identify candidate drugs predicted to interact with cGAS.

cGAS inhibitors were screened based on the crystal structure of cGAS and docking of compounds predicted in silico. In silico structure-based drug screening was performed using Computational Analysis of Novel Drug Opportunities (17). Predictions made by this docking algorithm were confirmed by publicly available software, such as AutoDock Vina (18), and analyzed with the PyMOL Molecular Graphics System (version 1.5.0.4; Schrödinger) or UCSF Chimera (19).

The hexa-histidine–tagged SUMO-cGAS construct was kindly provided by Russell Vance (University of California, Berkeley, Berkeley, CA). Protein was expressed in Rosetta BL21 pLysS cells and isolated sequentially by Heparin Sepharose affinity and S200 Superdex (both from GE Healthcare) size-exclusion chromatography. Protein was ∼95% pure by Coomassie blue–stained SDS-PAGE gel.

cGAS activity was monitored in the presence of 250 μM ATP, 250 μM GTP, 0–500 μg/ml herring testes DNA, 1.4 μM cGAS, and 3.3 nM [³²P]-ATP in Tris buffer (pH 7.5). After incubation for 1 h at room temperature, samples were spotted on PEI-cellulose TLC plates, and developed using a solvent composed of 1:1.5 [v/v] saturated NH₄SO₄ and 1.5 M KH₂PO₄ [pH 3.6]. For inhibition assays, compounds were characterized in the presence of the above reaction contents at fixed 100 μg/ml HT DNA. TLC plates were exposed to a phosphor-storage screen and imaged using a Typhoon imaging system. The interaction between cGAMP and mouse STING was measured by titrating protein from 3.6 μM to 1.8 nM with constant [³²P]-cGAMP in the absence of drug or in the presence of hydroxychloroquine (HCQ; 2.5 mM) or quinacrine (QC; 100 μM).

EMSAs were performed to measure the DNA-binding ability, with or without QC, in vitro. For the DNA-binding studies, 0.2 μM the 100-bp annealed IFN-stimulatory dsDNA (Forward: 5′-ACATCTAGTACATGTCTAGTCAGTATCTAGTGATTATCTAGACATACATCTAGTACATGTCTAGTCAGTATCTAGTGATTATCTAGACATGGACTCATCC-3′, Reverse: 5′-GGATGAGTCCATGTCTAGATAATCACTAGATACTGACTAGACATGTACTAGATGTATGTCTAGATAATCACTAGATACTGACTAGACATGTACTAGATGT-3′) was mixed with 1.5 mM cGAS protein and incubated at room temperature for 1 h. QC (serial 2-fold dilution from 250 to 16 μM) was then added and incubated for an additional 2 h at room temperature. The mixtures were resolved on 6% PAGE gel, analyzed with the fluorescence-based EMSA kit (Molecular Probes), and visualized using a Gel Doc XR System (Bio-Rad).

THP1 cells or 20 nM PMA–differentiated THP1 cells (0.2 × 10⁶) were transfected with 0.5 μg/ml herring testis DNA, 2 μg/ml polyinosinic-polycytidylic acid, or 1 μg/ml retinoic acid–inducible gene 1 (RIG-I) ligand using Lipofectamine 2000 (Invitrogen), according to the manufacturer’s instruction. Before transfection, cells were incubated with different concentration of QC and HCQ or other antimalarial drugs (AMDs) for 5 min. Cells were stimulated for 3 or 16 h, and then harvested for RNA extraction.

For cGAMP stimulation of STING, THP1 cells (0.5 × 10⁶) were permeabilized with digitonin (10 μg/ml) and stimulated with cGAMP (10 μg/ml) for 5 min. Cells were washed and resuspended in fresh medium with different concentrations of QC or HCQ for 4 h before harvesting for RNA extraction.

Quantitative PCR was performed on an ABI StepOne Plus machine using the IFN-β primers from QIAGEN. Threshold cycle values were set as a constant threshold at 0.2, and fold changes in gene expression were calculated using the 2^−ΔΔCT method.

To specifically examine the effect of AMDs on the cGAS/STING pathway, 293T cells were transfected with 25 ng ISRE-luciferase reporter plasmid (Takara Bio), 100 ng pcDNA3-expressing hemagglutinin-tagged human STING (R71H, G230A, R293Q alleles), and 25 ng pcDNA3-expressing Myc-tagged human cGAS. For controls, 293T cells were transfected with ISRE-luciferase reporter plasmid and pcDNA3-expressing hemagglutinin-tagged human STING only, as well as with a plasmid encoding RIG-I CARD. 293T cells were treated with AMDs 5–10 min prior to transfection of plasmids. Cells were lysed 16 h after plasmid transfection, and luciferase activity was assayed with the Luciferase Reporter Assay System (Promega), according to the manufacturer’s instructions.

To specifically examine the effect of AMDs on the AIM2 inflammasome, a pyroptosis assay using THP1 control (Cas9) or THP1 cells in which AIM2 was disrupted by CRISPR-Cas9 was performed. THP1 cells were differentiated with PMA, primed with human IFN-β (10 U/ml), and transfected with 4 μg/ml calf thymus DNA in the presence or absence of AMD. Cell death was assayed with a two-color IncuCyte ZOOM in-incubator imaging system (Essen Biosciences, Ann Arbor, MI). All results included in the supplemental data are from experiments that were performed at least twice, with very similar results in each experiment.

DNA binding to cGAS results in a pronounced conformational change, allowing for substrates ATP and GTP to bind to the catalytic pocket of the enzyme and the formation of a higher-order catalytically active complex of 2:2 cGAS/DNA (Fig. 1A). The formation of this complex is cooperative and results in contacts between the two protein monomers, as well as DNA binding at two distinct regions of the protein (Fig. 1B): Sites A and B. The more pronounced Site A interactions are defined through extensive contact with a positively charge surface of the protein flanked by an α-helical spine and Zn-thumb (20). Site B is structurally located opposite from Site A on the adjacent protein monomer (21).

We performed in silico screening of chemical and drug libraries to identify compounds that interact at these regulatory sites. The murine cGAS-DNA crystal structure (PDB: 4LEZ) was used as a target, because it captures the protein in a catalytically active 2:2 complex with the productive protein–DNA interactions and the active site open. These studies predicted that HCQ would bind in two sites in the 2:2 dimer: Drug Sites I and II. In each site, HCQ sits in the minor groove of the DNA between the protein–DNA interface, interacting with DNA binding Sites A and B on two adjacent monomers (Fig. 1C, 1D). Intriguingly, HCQ is predicted to be within 4 Å of the residues K372 in Site A, which interacts with the phosphate backbone of the DNA, and R342 in Site B, which reaches into the minor groove. Mutation of K372 abolishes enzyme activity with minimal effects on DNA binding. R342 mutations disrupt both DNA binding and enzyme activity (21). As such, small molecule interactions with these residues could lead to changes in the stability of the protein/nucleic acid complex and/or activation of the enzyme by DNA that are mediated by residues R342 and K372, respectively.

We next performed an extended computational analysis of related compounds with an altered quinoline core and functional side chains (Supplemental Fig. 1A). Each compound was predicted to bind in the two sites defined by the cGAS–DNA interface, analogous to HCQ (Supplemental Fig. 1B), albeit with different affinity (Table I) and with Drug Site II binding more strongly than Drug Site I. Together, these observations predict that HCQ and related AMDs interact with the cGAS/DNA complex at a site required for binding to, and subsequent activation of, cGAS by DNA.

We first examined the effect of HCQ (data not shown) and QC on DNA binding to cGAS using an EMSA. Even when DNA and cGAS were preincubated, and complexes were allowed to form, QC disrupted the complex in a dose-dependent manner and led to increased amounts of free DNA at the bottom of the gel (Fig. 2A), indicating that QC blocked dsDNA/cGAS binding.

Having demonstrated that HCQ and QC disrupted DNA/cGAS binding and that there were substantial differences in computational docking affinity among AMDs, we next tested the functional effects and potencies of these compounds. For this analysis, we performed in vitro dose-titration experiments with recombinant cGAS and quantified cGAMP production using TLC (Fig. 2B). Each compound yielded dose-response curves similar to QC but with different inhibitory activities (Fig. 2C, Table I). As shown in Fig. 2C, QC and 9-amino-6-chloro-2-methoxyacridine (ACMA) were the most potent inhibitors of cGAMP production (low μM range); quinine (QN) had very low inhibitory activity, and primaquine (PQ), chloroquine (CQ), and HCQ had intermediate inhibitory activities. The computational predicted binding affinities correlated well with the IC₅₀ of AMDs, validating the prediction of our computational analysis (Table I). A striking correlation between predicted binding energies and pIC₅₀ values was observed for the drug binding Site I (data not shown) and Site II (Fig. 2D), with the exception that QN does not belong to the aminoquinoline or aminoacridine family. Several of the compounds also exhibited a cooperative inhibition profile, consistent with multiple binding sites, perhaps as a consequence of disrupting higher-order oligomer formation and cGAS activation or as a consequence of the presence of two predicted binding sites within the active 2:2 protein/nucleic acid complex. Together, these observations establish that aminoquinoline- and aminoacridine-based AMDs impair DNA-stimulated cGAS activity.

To determine the potential physiological relevance of our computational analysis and in vitro observations, we examined whether AMDs could inhibit IFN-β production within the cell. As demonstrated previously, cGAS is the key cytosolic DNA sensor required for IFN-β production in THP1 cells in response to DNA (12, 15). Therefore, we transfected THP1 cells with dsDNA in the presence or absence of different AMDs and quantified IFN-β expression by quantitative PCR. As shown in Fig. 2E, QC, ACMA, HCQ, and CQ all inhibited IFN-β production by THP1 cells with an IC₅₀ dose range of 3 to 25 μM. At the IC₅₀ for QC and HCQ (∼3 and 25 μM, respectively), both cell viability and transfection efficiency were unaffected compared with the no-drug control, although some reduction in cell viability and/or transfection efficiency was observed at higher doses (data not shown). In contrast, PQ and QN exhibited 10-fold lower potency. Of note, the rank order of AMD inhibitory activity in vitro and in the cell were very similar, consistent with DNA/cGAS interaction as the target (Fig. 2C, 2E). Furthermore, when 293T cells were transfected with STING and cGAS plasmids and ISRE stimulation was quantified by luciferase production, HCQ and QC, at IC₅₀ doses, attenuated IFN-I production (Supplemental Fig. 2A). As further evidence that the AMDs blocked the DNA/cGAS interaction rather than the cGAMP/STING interaction, we showed that the AMDs did not inhibit cGAMP stimulation of STING in cells or the in vitro interaction between the two (Supplemental Fig. 2C, 2D). QC inhibited IFN-β production even when cells were incubated with the drug overnight and the drug was washed away before DNA transfection (data not shown).

To determine the selectivity of HCQ and QC inhibitory responses at the IC₅₀ and IC₇₀ doses found to inhibit cGAS, we examined the effects of these drugs on different nucleic acid ligands, as well as on the production of cytokines other than IFN-β. AMDs demonstrated inhibition of polyinosinic-polycytidylic acid– and RIG-I–stimulated induction of IFN-β in THP1 cells (data not shown), as well as IFN-β induced by a plasmid expressing the RIG-I CARD domains (Supplemental Fig. 2B). Although we also observed that AMDs could inhibit DNA stimulation of other cytokines, such as TNF and IL-6, by THP-1 cells (data not shown), these results are not surprising considering that these cytokines are poorly induced in STING- or cGAS-deficient mice (21), whereas a partial inhibitory effect of HCQ and QC on pyroptosis and IL-1β (Supplemental Fig. 2E–H; data not shown) suggests that AMDs also may interfere with DNA stimulation of AIM2. Together, these observations indicate that, in addition to a role for AMD in direct inhibition of cGAS function in response to DNA stimulation, AMD, at the concentrations tested, can attenuate other ligand sensor interactions. Whether these effects are due to intercalation with dsDNA and/or interference with endosomal maturation and trafficking remains to be determined.

AMDs, such as HCQ, CQ, and QC, are effective in the treatment of milder manifestations of SLE, such as skin rash and arthritis. They are known to have multiple modes of action, but precisely which mechanism(s) are responsible for their beneficial action is uncertain (22). It was shown that AMDs can attenuate IFN-α production by plasmacytoid dendritic cells due to a reduced activation of TLR associated with AMDs' known ability to alter endosomal maturation (23). Our results indicate that a number of aminoquinoline- and aminoacridine-based AMDs also strongly inhibit IFN-β production in response to DNA stimulation and that this inhibition is due, at least in part, to a blockade of the interaction of DNA and cGAS, although interference with other pathways also was noted. Although the concentration of HCQ to achieve 50% inhibition of IFN-β production was higher than the 1–2 μM detected in the blood of patients, it was reported that HCQ is concentrated 10–100-fold in the cell during long-term treatment (24). Interestingly, QC was reported to be more effective than HCQ, and these drugs take weeks to months to exert their clinical effect in SLE (24), suggesting that concentration within the cell is required.

The source of IFN-I in IFNopathies is also unclear. An important clue to the potential generation of IFN-I in autoimmunity comes from the disease AGS. This disease is characterized by severe neurologic deficits early in childhood, systemic autoimmunity, and evidence of IFN-I production in the cerebral spinal fluid and systemic circulation (25). In 25% of patients, AGS is caused by loss of function mutations of the 3′–5′ DNA exonuclease TREX1 (26). Because cells deficient in TREX1 trigger IFN-β selectively through cGAS (13), we predict that AMDs with high affinity, such as QC, will be effective for the treatment of AGS caused by TREX1 deficiency, especially if egress of QC from the brain can be slowed, as shown in mice treated with this drug (27).

Our studies identify the cytosolic DNA sensor cGAS as a target of AMD activity. This observation, together with decades of experience in human diseases, suggests that this widely used family of drugs with a strong safety profile could be repurposed to target IFNopathies and possibly other autoimmune disorders related to cGAS overactivity. The preliminary study activity relationship also suggests that the rational design of more potent and specific cGAS inhibitors of the aminoacridine and aminoquinoline families may be possible.

We thank Ram Samudrala, Toni Kline, and Elizabeth E. Gray (University of Washington) for helpful advice and/or assistance; Xizhang Sun, Joel Nelson, and Lena Tanaka for technical assistance; and Russell Vance for the mouse STING and cGAS expression constructs.

The authors have no financial conflicts of interest.