Cytosolic RNA:DNA hybrids activate the cGAS–STING axis

Using in vitro transcribed RNA molecules (e.g. T7 RNA polymerase) as templates for reverse transcriptase, RNA:DNA hybrids of different lengths can easily be produced enzymatically. However, reverse transcriptase also possesses DNA‐dependent DNA polymerase activity and as such can generate dsDNA from the newly synthesized ssDNA (Hsieh et al, 1993). Therefore, to faithfully exclude dsDNA contamination in assessing the immunostimulatory capacity of DNA:RNA hybrids, we made use of enzymatically generated homopolymers of poly(rA) and poly(dT). Annealing of poly(rA) and poly(dT) resulted in the formation of a poly(rA):poly(dT) RNA:DNA hybrid with a predominant band of around 1,300 bp (Fig 1A). To confirm that this product indeed represented a RNA:DNA hybrid, we performed dot‐blot assays wherein membrane‐immobilized nucleic acids were probed with a monoclonal antibody (S9.6) specific for RNA:DNA hybrids (Boguslawski et al, 1986; Hu et al, 2006). We observed a specific detection of the hybrid by the antibody, correlating with the amount of nucleic acids immobilized (Fig 1B). Digestion of the hybrid with RNase A which digests dsRNA, DNase I which digests dsDNA or ssDNA and RNase H which digests RNA part of RNA:DNA hybrids provided further confirmation regarding the purity of the hybrid. Thus, while the RNA:DNA hybrid was completely digested by RNase H and DNase I enzymes, it was resistant, as expected, to RNase A activity (Fig 1C). Control experiments confirmed the specificity of these enzymes on dsDNA or dsRNA, respectively (A.K. Mankan, unpublished observations). In order to assess the biological activity of RNA:DNA hybrids, we then transfected the poly(rA):poly(dT) hybrids or the single components poly(rA) or poly(dT) into murine bone marrow‐derived macrophages. Transfection of the hybrids resulted in a robust induction of type I IFNs and pro‐inflammatory genes (Ifna, Ifnb, Il6 and Ip10) (Supplementary Fig S1A). Of note, a significant activity of poly(dT), by itself, was also observed for some readouts studied (see discussion below). To assess the relevance of hybrids in the human system, we next transfected hybrids or the individual polynucleotide components into peripheral blood mononuclear cells. RT–PCR confirmed enhanced expression of IFNB gene in PBMCs transfected with hybrids (Fig 1D). Synthetic poly(dA):poly(dT), which was transfected as a dsDNA control, showed higher activity with regard to IFNB induction. We also observed a significant secretion of IP‐10 in response to the RNA:DNA hybrids, while only a minimal IP‐10 induction was seen in cells transfected with the single polynucleotides (Fig 1E). We next tested the activity of RNA:DNA hybrids in differentiated human THP‐1 cells. In line with the PBMC data, RNA:DNA hybrids induced robust transcription of IFNB in THP‐1 cells, again with dsDNA being more active (Fig 1F). In order to have a sensitive readout for antiviral gene expression, we equipped THP‐1 cells with Gaussia luciferase (GLuc) under the promoter of IFIT1, a well‐characterized ISG that is also directly transcribed upon PRR stimulation (Supplementary Fig S2A). Studying pIFIT1‐GLuc THP‐1 in response to these different stimuli paralleled the data obtained by measuring IFNB (Fig 1G). However, as observed in the previous set of experiments, the transactivation of the IFIT1 promoter in response to RNA:DNA hybrid was never as strong as upon dsDNA transfection, also when compared over a broad range of different ligand concentrations (Fig 1H). Interestingly, assessing the production of the antiviral chemokine IP‐10 in response to these stimuli revealed a similar plateau for both hybrids as well as dsDNA at higher ligand concentrations, most likely due to the fact that this chemokine is induced by both PRR‐dependent as well as type I IFN‐dependent mechanisms (Supplementary Fig S1B). Altogether, these data indicated that intracellular RNA:DNA hybrids are indeed sensed by the innate immune system, leading to strong antiviral immune responses.

We next focused on the identification of the intracellular receptor that was essential for mediating the RNA:DNA hybrid‐induced innate immune response. As RNA:DNA hybrids contain a single polyribonucleotide molecule, we first assessed the involvement of the RLR system as sensors of RNA‐DNA hybrids. MAVS constitutes the critical signaling adapter downstream of both RIG‐I and MDA‐5 and as such, MAVS‐deficient cells are devoid of both RIG‐I and MDA5 signaling. We made use of the CRISPR/Cas9 gene editing system to knock out MAVS in THP‐1 cells. As such, we targeted a critical coding exon to disrupt the reading frame and therefore the expression of MAVS (Supplementary Fig S2B). Transfection of differentiated wild‐type (WT) THP‐1 cells with RNA:DNA hybrids resulted in increased IFNB, and a similar response was also observed in MAVS knockout cells (Fig 2A). As expected, MAVS‐deficient THP‐1 cells failed to respond to 5′‐triphosphate RNA, whereas dsDNA‐mediated activation of THP‐1 cells was not affected. To study the involvement of the cGAS–STING axis in RNA:DNA hybrid recognition, we next generated THP‐1 cells deficient in either STING or cGAS (Supplementary Fig S2C and D). Interestingly, we did not observe any induction of IFNB in response to transfection of RNA:DNA hybrids in differentiated STING or cGAS KO THP‐1 cells (Fig 2B–D). As anticipated, dsDNA‐mediated IFNB induction was also completely abrogated in cGAS/STING KO cells, whereas 5′‐triphosphate RNA detection was still intact. Similar results were also observed for IL‐6 (Supplementary Fig S3). These data suggested that, next to dsDNA, cGAS also serves as the receptor for cytosolic RNA:DNA hybrids.

To confirm the direct recognition of RNA:DNA hybrids by cGAS, we took advantage of the fact that cGAS‐mediated activation can be studied in vitro. To this effect, we incubated the single polynucleotides (poly(rA) and poly(dT)) or the synthetic RNA:DNA hybrids (poly(rA):poly(dT)) with recombinant cGAS in the presence of ATP and GTP. Moreover, we included synthetic dsRNA (poly(rA):poly(rU)) and synthetic dsDNA (poly(dA):poly(dT)) as controls. The samples were incubated for 45, 90 or 180 min. Analysis of the RNA:DNA/cGAS sample via HPLC revealed a peak correlating with the expected peak for cGAMP(2′–5′) (Fig 3A). However, no such product was observed upon the incubation of cGAS with dsRNA or any of the single polynucleotide preparations (Fig 3C and Supplementary Fig S4A). Compared to dsDNA, RNA:DNA hybrids induced less cGAMP, also when studying cGAMP production over time (Supplementary Fig S4B). Several groups have recently used shorter length synthetic ssRNA and ssDNA to generate hybrids and to study the immune cell activation by these shorter hybrids. To confirm whether such short hybrids can also activate cGAS–cGAMP pathway, we used 60‐bp synthetic poly(rA) ssRNA and poly(dT) ssDNA oligos and generated RNA:DNA hybrids by annealing them (Fig 4A). To verify whether these shorter oligos were functionally active, we transfected them into pIFIT1‐GLuc THP‐1 cells and observed a robust secretion of luciferase only in response to hybrids and dsDNA (Fig 4B). As above, incubation of these shorter hybrids with recombinant cGAS clearly resulted in the formation of cGAMP (Fig 4C).

To understand the structural basis for the interaction of cGAS with RNA:DNA hybrids, we generated protein–nucleic acid interaction models in silico. Modeling of the complexes of cGAS with either RNA:DNA hybrid, dsDNA or dsRNA revealed that RNA:DNA hybrid could bind in the cleft of cGAS (regardless of the orientation of RNA and DNA strands) exactly in the same mode as dsDNA. At the same time, the dsRNA helix could not be accommodated in this cleft (Fig 5A–D). Structural alignment of dsDNA, RNA:DNA hybrid and dsRNA helixes showed that dsDNA and RNA:DNA hybrid molecules have a similar conformation of their double‐stranded helix with similar shapes of their minor and major groves. While there was an obvious overlap between the orientation of the bases of RNA:DNA hybrids and dsDNA, the shape of a dsRNA helix differs from both the dsDNA and hybrid helices with the dsRNA helix being wider (Fig 5E and F). It is likely that this prevents its proper binding into the cGAS cleft and as such the conformational switch of cGAS that is required for its nucleotidyltransferase activity. As RNA:DNA hybrids can bind to cGAS in both orientations of the RNA chain, its binding to cGAS should lead to the same switch‐like conformational changes of the activation loop, which are induced by dsDNA binding (Zhang et al, 2014) (Supplementary Fig S5). Altogether, these data proved that RNA:DNA hybrids can serve as ligands initiating cGAS activity.

Equal amounts of synthetic poly(dT) (Sigma #P6905) and poly(rA) (Sigma # P9403) were mixed in 5× annealing buffer (50 mM Tris–HCl pH 7,6, 250 mM NaCl, 5 mM EDTA), for example, 40 μl poly(rA) (300 ng/μl) + 40 μl poly(dT) (300 ng/μl) in 20 μl 5× annealing buffer. The samples were incubated in a PCR block starting at 95°C with decreasing gradient of −1°C per 100 seconds until 20°C. 60‐bp poly(rA) was purchased from Biomers. 60‐bp poly(dA) and 60‐bp poly(dT) were purchased from Integrated DNA Technologies.

Nitrocellulose membrane was marked and spotted with the RNA:DNA hybrids or poly(dA:dT) (dsDNA) as a negative control. The membrane was allowed to air dry for 2 h and then blocked in 0.5% milk in PBST buffer. The membrane was then incubated with the anti‐hybrid S9.6 antibody overnight at 4°C, washed and probed with anti‐mouse secondary antibody.

Murine BMDMs were isolated and cultured as per standard protocols. THP‐1 cells were maintained in RPMI medium. The cells were differentiated for 3 h with PMA (330 ng/ml), the medium was removed, and the cells were washed with 1× PBS three times. The cells were detached in PBS, and 7 × 10⁴ cells were plated in 96‐well flat‐bottomed plate. Isolation of PBMC from buffy coat was performed using standard protocol.

7 × 10⁴ cells were incubated overnight in flat‐bottomed 96‐well plate. A transfection mix with Lipofectamine and 1 μg/ml of poly(rA), poly(dT), poly(rA):poly(dT) was prepared as per standard protocol. The cells were incubated with the nucleic acids overnight. The supernatants were collected and subjected to further assays. For RT–PCR, the cells were stimulated with the nucleic acids for 6 h.

1 μg of hybrid was digested with 1 μl of the RNase H, RNase A or DNase I (Fermentas) enzyme in a final reaction volume of 10 μl. The sample was incubated at 37°C for 30 min. The enzyme was heat‐inactivated by incubating the samples at 65°C for 10 min, and the efficiency of digestion was checked by running the samples in 0.8% agarose gel.

RNA isolation was performed using the QIAcube system (Qiagen), and equal amounts of RNA were used for cDNA synthesis. Real‐time PCR analysis with Eva Green PCR Master Mix (Biobudget) was performed on Real‐time PCR system (Roche). The expression of target genes was normalized to HPRT or GAPDH expression and plotted as arbitrary units on a linear scale. Primer sequences are available on request.

Cell culture supernatants were assayed for human IP‐10 and human IL‐6 (BD Biosciences) according to the manufacturer's instructions.

THP‐1 knockout cells were generated using the CRISPR/Cas9 system. A plasmid containing mCherry Cas9 and a U6 promoter‐driven gRNA against STING, MAVS or cGAS was electroporated into THP‐1 cells under following conditions: 5 μg plasmid were mixed with 250 μl cell suspension (10 × 10⁶/ml in OptiMEM) and electroporated at 950 μF and 250 V. Cells were FACS sorted for mCherry expression 24 h after electroporation. Positive cells were diluted under limiting conditions and plated in 96‐well plates to obtain single cell clones. The genotype of THP‐1 clones was analyzed by deep sequencing (Illumina, MiSeq) (Schmid‐Burgk et al, 2014).

To study induction of interferon‐stimulated genes at the natural gene expression level using a reporter system, we generated a IFIT1 reporter cell line. Using the CRISPR/Cas9 system, the Gaussia luciferase gene was knocked into the IFIT1 gene locus in THP‐1 cells. For this purpose, a reporter insert cassette, harboring flanking 50‐bp homology arms, a 2A peptide and the luciferase gene, was synthesized by a two‐step PCR. THP‐1 cells were co‐electroporated with a CRISPR targeting the IFIT1 locus and the Gaussia luciferase insert cassette. As the CRISPR construct contained a mCherry gene, electroporated cells were sorted for mCherry expression by FACS 24 h after electroporation. For isolation of reporter cell clones, cells were seeded under limiting dilution conditions. Single cell clones were tested for knockin of the luciferase gene by stimulation with synthetic triphosphate RNA (pppRNA) and subsequent measurement of luciferase induction. Positive reporter clones were validated by sequencing of the insert locus.

For in vitro reactions of cGAS in the presence of varying nucleic acids, 2 mM recombinant cGAS was mixed with 0.2 μg/μl poly(rA):poly(dT), poly(dA):poly(dT), poly(rA):poly(rU) or the corresponding single‐stranded controls and 0.1 M ATP and 0.1 mM GTP in buffer A (100 mM NaCl, 40 mM Tris pH 7.5, 10 mM MgCl₂). After 45, 90 or 180 min of incubation at 37°C, the reaction mixture was analyzed by RP‐HPLC. Samples were prepared in 0.3 M triethylammonium acetate, applied to a Waters XBridge C18 column (4.6 × 50 mm, 2.5 μm particle size) and separated in an isocratic gradient of 100 mM ammonium acetate for 5 min at a flow rate of 1 ml/min. Chemically synthesized cGAMP(2′–5′) (Ablasser et al, 2013a) served as a positive control.

Complexes of cGAS with three short double‐stranded nucleic acids (NA) were modeled by docking of corresponding NA (dsDNA (PDB ID 4KB6) (Civril et al, 2013) or (PDB ID 3V6T) (Deng et al, 2012), DNA:RNA hybrid (PDB ID 4GG4) (Yin et al, 2012) and dsRNA (PDB ID 3KS8) (Kimberlin et al, 2010) into the structure of pig cGAS (PDB ID 4KB6 chain A) (Civril et al, 2013)). Docking was done on the HADDOCK server (de Vries et al, 2010). HADDOCK permits position restrained docking of NA into proteins. Sets of residues for docking in cGAS and dsDNA that should interact with NA have been taken from the original crystal structure of the cGAS–dsDNA complex (PDB ID 4KB6) (Civril et al, 2013). Visualization of the results was performed using PyMol.

Statistical tests were performed using GraphPad Prism program. For column statistics, one‐way ANOVA with Bonferroni's or Tukey's multiple comparison test was performed. P ≤ 0.05 was considered significant.