RIG-I recognizes metabolite-capped RNAs as signaling ligands

Abstract The innate immune receptor RIG-I recognizes 5′-triphosphate double-stranded RNAs (5′ PPP dsRNA) as pathogenic RNAs. Such RNA-ends are present in viral genomes and replication intermediates, and they activate the RIG-I signaling pathway to produce a potent interferon response essential for viral clearance. Endogenous mRNAs cap the 5′ PPP-end with m7G and methylate the 2′-O-ribose to evade RIG-I, preventing aberrant immune responses deleterious to the cell. Recent studies have identified RNAs in cells capped with metabolites such as NAD+, FAD and dephosphoCoA. Whether RIG-I recognizes these metabolite-capped RNAs has not been investigated. Here, we describe a strategy to make metabolite-capped RNAs free from 5′ PPP dsRNA contamination, using in vitro transcription initiated with metabolites. Mechanistic studies show that metabolite-capped RNAs have a high affinity for RIG-I, stimulating the ATPase activity at comparable levels to 5′ PPP dsRNA. Cellular signaling assays show that the metabolite-capped RNAs potently stimulate the innate antiviral immune response. This demonstrates that RIG-I can tolerate diphosphate-linked, capped RNAs with bulky groups at the 5′ RNA end. This novel class of RNAs that stimulate RIG-I signaling may have cellular roles in activating the interferon response and may be exploited with proper functionalities for RIG-I-related RNA therapeutics.


INTRODUCTION
RIG-I (Retinoic Acid Inducible Gene-I) is a pattern reco gnition innate imm une receptor that belongs to the DEA(X)D-box family of helicases. Its primary function is to surveil the cytoplasm for non-self RNAs arising from infecting viruses and other pathogens ( 1 ). RIG-I recognizes b lunt-ended doub le-stranded (ds) RNAs that contain either 5 -triphosphate (5 PPP) or 5 -diphosphate (5 PP) ends as pa thogen-associa ted molecular pa tterns (PAMPs) ( 2 , 3 ). These featur es ar e hallmarks of viral genomes and replica tion intermedia tes of nega ti v e-strand and positi v e-strand RNA viruses ( 4 , 5 ). Upon recognizing PAMPs, RIG-I initiates a signaling cascade culminating in a Type-I interferon response that controls the viral infection and primes the adapti v e response (5)(6)(7)(8).
In eukaryotes, the 5 PPP ends of the mRNAs are capped with 7-methyl guanosine (m 7 G), which can affect RIG-I's ability to bind cellular RNAs. Previous studies have shown that RIG-I binds m 7 G Ca p0 RN A comparabl y to 5 PPP RNA; howe v er, the combination of the m 7 G cap and a methylation e v ent at the 2 position of the first ribose, together called Cap1, reduces RIG-I affinity by a pproximatel y 200-fold ( 9 , 10 ). Structural studies show that RIG-I's Cterminal domain (CTD) contacts the triphosphate linkage of Cap0 but does not interact with the m 7 G cap moiety, and there is space in the RNA-bound RIG-I complex to accommodate the cap (Figure 1 A) ( 9 ). Both Cap0 and Cap1 RNAs are translated in the cell, but many viruses, particularly flavi viruses, hav e e volv ed strategies to chemically synthesize mRNA with Cap1 moiety for immune evasion (11)(12)(13)(14).
RN A ca ps are not limited to eukaryotes; RNAs capped with metabolites such as NAD + and Coenzyme A were initially discovered in bacteria ( 15 , 16 ). Since then, additional alternati v e caps like FAD, UDP-Glc and UDP-GlcNac have been identified in various organisms, such as yeast, plants, and human mitochondria, and cultured Dengue virus genomes (17)(18)(19)(20). These alternati v e caps are of different sizes: the NAD + is similar to the m 7 G canonical cap, dephosphoCoA has a long hydrocarbon chain, and FAD has a bulky, thr ee-member ed ring structure (Figure  1 B). Capping with metabolites occurs during transcription initiation when metabolites are incorporated instead of initiating NTP. For example, the ADP-based metabolites can compete with ATP for transcription initiation, get incorpora ted a t the 5 -end of RNA, and are thus r eferr ed to as noncanonical initiating nucleotides (NCINs) ( 21 ). While no capping enzymes for metabolites have been identified, initiation may not be the only mechanism for alternati v e ca pping. Small nucleolar RN As (snoRN As), introns generated through RNA splicing, have been identified with N AD + -ca ps, suggesting that some post-transcriptional capping mechanism may exist ( 20 ). The role of these caps is not yet well defined. The relati v e amount of N AD + -ca p versus N ADH-ca p in the mitochondria is sensiti v e to the metabolite concentrations ( 19 ). In cytoplasmic contexts, the N AD + -ca pped RN As are not translated, but N AD + ca pping is thought to promote RNA turnover ( 20 ). Because RIG-I reco gnizes Ca p0 RN As, w e w ere curious to know if these metabolite-capped RNAs found in the cell are imm uno genic through the RIG-I activation pathway.
An important difference between the metabolite-capped RNAs and a comparable Cap0 RNA is the phosphate linkage between the cap and the first nucleotide. In each alternati v e cap, the metabolite moiety is connected to adenosine with a diphosphate linkage. In contrast, in Cap0 and Ca p1, the guanosine ca p is connected to the first nucleotide through a triphosphate linkage (Figure 1 B). RIG-I predominantly interacts with the ␣and ␤-phosphates of Cap0 RN A w hile making a single contact with the ␥ -phosphate of Cap0 RNA ( 3 , 9 ). The 5 PP RNAs are also recognized as PAMPs ( 2 ), suggesting tha t ␥ -phospha te recognition is dispensab le. Howe v er, the difference in length between the di-and tri-phosphate linkages may be relevant to how the cap moiety positions itself in the CTD binding pocket. The short di-phosphate link would bring the cap moiety closer to the CTD and potentially hinder phosphate interactions.
Our initial guess was the shortened phosphate linkage, combined with the bulk y cap , would inhibit RIG-I recognition. Howe v er, this study shows that RIG-I recognizes the metabolite-ca pped dsRN As deri v ed from various metabolites similar to 5 PPP and Cap0 RNAs. Using in vitro transcribed metabolite-ca pped RN As, we demonstra te tha t RIG-I binds to metabolite-capped RNAs in an ATPasecompetent manner and utilizes these RNAs as oligomerization scaffolds and ultimately as signaling ligands in cell signaling reporter and antiviral signaling assays. That RIG-I reco gnizes metabolite-ca pped RN As of significantl y different geometries demonstrates its tolerance for a wide variety of 5 phosphate-linked moieties and constrains how such RN As can a ppear in the cytoplasm without generating an immune response.
The 39-mer ssRNA products were purified by gel electrophoresis on 1.5 mm × 30 cm 40% acrylamide and 6 M urea gel in 1.5 × TAE. Electrophoresis was conducted at 500 V for 4 h, and the resulting ssRNA was identified by UV shadowing, the band was excised, and RNA was extracted from the gel using a W ha tman Elutrap electroelution system in 1.5 × TAE buffer run at 150 V for 6 h. The resulting ssRNA was concentrated with ethanol precipitation and stored at -80 • C.
In vitro, synthesized ssRNAs were analyzed for purity and metabolite cap incorporation using L C-MS (Nov atia, LLC; Newton, PA). Samples were run at a 200 • C source temper ature r ather than 350-375 • C to pre v ent RNA 5' caps detachment.

Protein expression and purification
The RIG-I gene was cloned in the protein expression vector pET28 SUMO and expressed as SUMO fusion proteins in Esc heric hia coli strain Rosetta (DE3) (Novagen). The protein was purified using a series of chromato gra phy columns as published previously ( 3 ). The soluble lysate was fractionated through a HisTrap HP (GE Healthcare), followed by Ulp1 protease digestion to remove 6 × His-SUMO tag, hy droxy apatite (CHT-II, Bio-Rad), and heparin Sepharose (GE Healthcare). The purified protein was dialyzed into 50 mM HEPES pH 7.5, 50 mM NaCl, 5 mM MgCl 2 , 5 mM Oxygen atoms ar e color ed r ed, phosphorus orange, nitrogen blue, and carbon white, while the CTD is transparent red. Six members of the asymmetric unit are aligned and overlaid. The triphosphate linkage demonstrates little differences between asymmetric units, while the Ca p0 moiety, w hich is m uch bulkier, twists in multiple dif ferent conforma tions without contacting RIG-I residues. ( B ) The structure of ATP and the metabolites incorporated at the 5 -end of RNA as 5 -PPP adenosine, 5 NAD + , 5 dephospho-CoA, and 5 FAD are shown. The red color highlights the ADP-based moieties serving as the initiating nucleotide for transcription. All metabolite cap moieties used here have been identified in bacterial RNA transcripts, w hile onl y N AD+ or N ADH-ca pped RN As have been identified in humans and yeast. The eukaryotic m7G Ca p1 structure has the methyl groups of G-cap and 2 -position of the first ribose highlighted in blue. Combining the G-cap and the 2 methylation is essential for RIG-I evasion.
DTT and 10% glycerol overnight at 4 • C, frozen in liquid nitrogen and stored at -80 • C.
RNA K D, app measurements using ATP h y drolysis-based titrations ATP hydrolysis was measured at constant RIG-I (15 nM) and increasing RNA concentration (1-400 nM) in the presence of 2 mM ATP (spiked with [ ␥ -32 P] ATP). The AT-P ase r eaction was measur ed after 20, 40, and 60 min of r eaction times in ATPase buffer (50 mM MOPS pH 7.4, 5 mM DTT, 5 mM MgCl 2 , 0.01% Tween20) at 25 • C. Reactions were stopped at each time point using 4 N formic acid and analyzed by PEI-Cellulose-F TLC (Merck) de v eloped in 0.4 M potassium phosphate buffer (pH 3.4). TLC plates were exposed to a phosphorima ger plate, ima ged on a Typhoon phosphor-imager, and quantified using the Image-Quant software. (1)

Fluorescence polarization assays
Fluor escence polarization measur ements wer e performed on a Tecan Spark microplate reader in a 384-well black plate a t 25 • C . A monochroma tor set the e xcitation wav elength at 485 nM and the emission wavelength at 535 nm, with a 20 nm bandwidth. Purified RIG-I protein was serially titrated in 1 × ATPase buffer (50 mM MOPS pH 7.4, 5 mM DTT, 5 mM MgCl 2 , 0.01% Tween20) and incubated with a constant 20 nM fluorescein-labeled dsRNA for 15 min at 25 • C; 500 M ATP was added prior to taking measurements. To obtain K D values for each RIG-I's binding affinity to each RNA, polarization values from triplicate data sets were plotted as a function of protein concentration. The data were fit to equations 2 and 3. The observed fluorescence polarization ( F ) from the initial fluorescence polarization ( F 0 ) is proportional to the amount of protein-RNA complex ( PR ) and modified by a coefficient of complex formation ( f c ). Reported errors for K D are fitting errors.
To measure the affinity of ssRNA, we used a competition binding assay where the unlabeled ssRNA was serially titrated in 1 × ATPase buffer and incubated with a constant, pr e-mix ed 50 nM RIG-I and 20 nM fluorescent dsRNA for 15 min at 25 • C. 500 M ATP was added prior to taking measurements under the same conditions previously stated. Polarization values were plotted as a function of ssRNA concentration. Unlabeled dsRNA was used as a control to demonstrate competition for RIG-I binding against the fluorescent dsRNA. This data was fit to equations (2) and (3) to estimate the IC 50 value rather than K D .

Electrophoretic mobility assays (EMSA)
EMSAs were performed by incubating RIG-I (75 nM) and ds39 RNAs (25 nM) in ATPase buffer for 60 min at 4 • C. 2 mM ATP was added to each RIG-I containing reaction 15 min before loading the sample into the gel. Loading buffer (10 × concentration of 1.5% Ficoll 400 in Tris-borate buffer, pH 8.0) was added to the samples loaded on a 4-16% gradient Nati v e PAGE gel (Invitrogen) a t 4 • C . Gels were scanned at 532 nm using a Typhoon FLA 9500 laser-based scanner (GE).

IFN-␤ reporter cell reporter signaling assays
HEK293T cells were grown in 5% CO 2 and 37 • C, in DMEM with 10% FBS in 6-well plates to 60% confluence and cotransfected with firefly luciferase reporter plasmid (pLuc125 / 2.5 g), Renilla luciferase reporter plasmid (pRL-TK / 500 ng), and a plasmid carrying myc-tagged WT RIG-I gene under the constituti v ely acti v e CMV promoter (pcDNA-3.1 / 2 g). The firefly luciferase gene is under the control of the interferon ␤ promoter, and the Renilla luciferase plasmid is under the control of the constituti v ely acti v e TK promoter. The plasmid transfections were carried out with X-tremeGENE HP DNA Transfection Reagent (Roche). Cells wer e r epla ted in 96-well pla tes the next day at 2 × 10 4 cells / well density and transfected with indicated single-stranded and double-stranded RNAs (700 nM, or as indicated, in 110 l volume) using Lyovec transfection reagent (InvivoGen). After 20 h, the activities of firefly and Renilla luciferases wer e measur ed sequentially with the Dual-Luciferase reporter assay (Promega). Each trial's number of replicates are shown in figure legends. Error bars r epr esent the standard error of the mean (SEM).

Isolation of RNA and quantitative real-time RT-PCR analysis
HEK293T RIG-I KO cells were grown in 5% CO 2 at 37 • C in DMEM cell culture medium (Gibco) with 10% FBS in 6-well plates to 80% confluence and transfected with empty vector and myc-tagged WT RIG-I gene under the constituti v ely acti v e CMV promoter (pcDNA-3.1 / 2 g). Cells were replated in 12 well plates the next day at 1 × 10 5 cells / well density and transfected with indicated RNAs (50 nM in 110 l volume) using 50 l Lyovec transfection reagent (InvivoGen).
A549 and A549 RIG-I KO cells were maintained in RPMI cell culture Medium (Gibco) with 10% FBS. Cells were plated in 12-well plates at 1 × 10 5 cells / well density and transfected with indicated RNAs (100 nM in 110 l volume) using 50 l Lyovec transfection reagent (InvivoGen).
Cells were washed with PBS, and total RNA was extracted from each well immediately using RNeasy micro kit (Qiagen, Hilden, Germany) 20 h post RNA transfection for HEK293T RIG-I KO and 25 h post RNA transfection for A549 and A549 RIG-I KO respecti v ely. Cells were lysed by adding 350 l of cold RLT lysis buffer containing 1% (v / v) 2-mer captoethanol (Millipor eSigma) and passing the lysate through 1 ml Syringe 5-6 times. Total RNA was extracted from the lysate supernatant following the manufacturer's specification. The purification included in-column DNase treatment using the RNase-free DNase Set. The yield and purity of the RNA were measured using a NanoDrop 2000c spectrophotometer (ThermoFisher Scientific, Waltham, MA). The total RNA extracted ranged between 708 ng / l to 1540 ng / l concentration (Supplemental Figure S6B).
Complementary DN A (cDN A) was pr epar ed using 1.5 g of RNA using the High-Capacity cDNA Re v erse Transcription Kit (Applied Biosystems, Carlsbad, CA) in a total volume of 20 l reaction mixture following the manufacturer's specification.
The RT-qPCR analysis was performed following the MIQE guidelines http://rdml.org/miqe.html ( 23 ). Quantitati v e real-time PCR of specific genes was performed using the Platinum SYBR Green qPCR SuperMix-UDG with ROX (ThermoFisher Scientific, Waltham, MA) and primers specific for human IFNB1, ISG15, OAS1 or GAPDH (Supplemental Table S1) on a QuantStudio ™ 3 Real-Time PCR System (96-well, 0.2 ml Block, Applied Biosystems, Waltham, MA). The incorporation of SYBR Green dye into the PCR products was monitored in real-time after each PCR cycle, resulting in the calculation of the threshold cycle or Ct value that defines the PCR cycle number at which exponential growth of PCR products begins. PCR cycle conditions were as follows: 2 min at 50 • C, 10 min at 95 • C, 40 cycles of 15 seconds at 95 • C and 1 min at 60 • C. To ensure no contamination, each PCR procedure included a negati v e control reaction without a templa te. Real-time PCR da ta were analyzed using the QuantStudioTM Design and Analysis Software V1.5.1 (Applied Biosystems, Waltham, MA). The GAPDH housekeeping gene was used as a r efer ence for the relati v e quantification of the gene of interest, which was expressed as the ratio of the 'concentration of the target' to the 'concentration of GAPDH'. Fold changes for RT-qPCR were determined by the CT method. Data r epr esent the mean of triplicate of one r epr esentati v e analysis.
Two biolo gicall y independent experiments were performed. Data r epr esent the mean of triplicate of one representati v e analysis.

Generation of pure metabolite-capped RNAs through in vitro transcription using T7 RNA polymerase
RIG-I is potently stimulated by 5 PPP RNAs. Therefore, it is critical to ensure that metabolite-capped RNAs synthesized enzymatically from in vitro transcription reactions are not contaminated with 5 PPP RNA-ends. Previously published techniques of T7 RN A pol ymerase (T7 RNAP)-based transcription relied on purifying ADP-based metabolite-ca pped RN As from 5 PPP RN A ( 24 ). Because the two RNAs migrate close on gel electrophoresis, the purified metabolite-ca pped RN As may still contain 5 PPP RN A, w hich would be reco gnized by RIG-I and lead to false positi v e results ( 25 ).
Here, we describe an in vitro transcription method that would generate ADP-based capped RNAs without contaminating 5 PPP RNAs. We designed a T7 DNA promoter with an initiating adenosine and 39 bp coding sequence that lacked internal A's ( Figure 2 , Supplemental Figure S1). The 39 bp RNA length was chosen because it robustly stimulates RIG-I signaling activity compared to shorter RNAs (Supplemental Figure S1D). The initiating adenosine allows ADP-based metabolite incorpora tion a t the + 1 position to form an alternati v e cap (Figure 1 B). Because there are no internal A's in the coding sequence, we could exclude ATP from the transcription reaction cocktail, preventing 5 PPP contamination. We also excluded internal C's and ended the RNA transcript with a terminal GG to avoid copyback synthesis, where T7 RNAP utilizes the nascent RNA transcript as a template to generate a duplex ed RNA structur e ( 26 , 27 ). Excluding CTP from the r eaction cocktail ensur es a pr ecise 3 end and no contaminating, self-annealing RN As. Additionall y, we methylated the terminal CC of the promoter DNA template strand, which promotes T7 RNAP dissociation, again minimizing the copyback tr anscription ( 28 ). The tr anscribed RNA pr oducts were gel-purified, electr o-eluted, and concentrated using ethanol precipitation.
The RNA sample purity was assessed by mass spectroscop y (MS, Figur e 2 , Supplemental Figur e S2). The results confirmed the absence of 5 PPP or 5 PP contamination in the transcribed 5 metabolite RNAs. The LC-MS analysis shows that 5 PPP and 5 NAD + dsR-NAs were nearly 100% pure. The 5 dephosphoCoA and 5 FAD ca pped RN As contained small amounts of other mass species but not 5 PPP or 5 PP. The 5 dephospho-CoA ca pped RN A sample contained 12% of a species twice the expected molecular weight, corresponding to an RNA dimer from disulfide-linkage between the terminal sulfur groups of two dephosphoCoA moieties. The 5 FAD capped RNA contained a peak at 12646.4 Da comprising 11% of the total peak area, close in mass to a 5 monophosphate RNA (12639.3 Da), resulting from loss of the flavin moiety and one phosphate from the diphosphate linkage. These species could have been generated during the mass spectrometry analysis steps. RIG-I does not recognize the Cytosines are excluded from the transcript sequence, and CTP is not added to the reaction to pre v ent copy-back transcription. The dsDNA promoter is incubated with T7 RN A pol ymerase and other reagents described in the Methods to generate the desired 39-mer RNA transcript, purified from remaining NTPs and aborti v e pr oducts using Urea-PAGE, gel electr oelution, and ethanol precipitation. The purity of these ssRNAs was assessed using LC-MS, shown in the table, with percentages of each RNA species in parenthesis. If no percentage is shown, the purity is 100%. The ssRNAs were annealed to a chemically synthesized, complementary RNA with a 3-nt DNA overhang at the 5 -end to pre v ent RIG-I binding at the opposite end from the 5 PP-cap end, as well as having a fluorescein moiety on the second DNA position for biochemical assays. Created with Biorender.com. 5 monophosphate RNA end, and if present in the sample, it should not generate a false positi v e ( 29 ). Aside from a disulfide-bridge mediated dimer with dephosphoCoAca pped RN A, we did not observe species whose sizes would correspond to incorrect 3 extension by T7 RNAP copyback, indicating our transcription method produced precise 3 ends.
Transcribed RNAs were annealed to a chemically synthesized, complementary ssRNA to generate 39-bp 5 PPP and metabolite-ca pped dsRN As. The complementary ssRN A contained a 5 DNA overhang of 3-nt, distal to the capped or PPP 5 end of interest, to pre v ent RIG-I from binding at the opposite RNA end (Figure 2 , Supplemental Figure S1). For biochemical studies, a fluorescein moiety was added to the DNA overhang. An additional control RNA was chemically synthesized with a 2-nt overhang and a 5 OH instead of a 5 PPP or a cap. The 5 OH, 5 OVG RNA-end has a weaker affinity for RIG-I and shows a low signaling activity in a reporter assay ( 9 ), serving as a negati v e control for RIG-I binding.

Innate immune receptor RIG-I recognizes metabolite-capped dsRNAs
We le v eraged RIG-I's ability to hydrol yze ATP w hen bound to RNA to determine the apparent K D of RIG-I for each RNA. The ATPase assay informed whether metaboliteca pped RN A could effecti v el y bind and stim ulate ATP hydrolysis by RIG-I, which is important for RIG-I signaling activity ( 30 ). Five dsRNAs were tested, each with a different 5 end modification: 5 triphosphate (5 PPP), a PAMP RNA positi v e control; 5 OH, 5 2-nt overhang (5 OH, 5 OVG), a negati v e control; and 5 NAD + , 5 FAD, and 5 dephospho-CoA (5 dpCoA), the metabolite-capped RNAs (Figure 3 A, Supplemental Figure S3). The ATPase activity increased hyperbolically with the increase in RNA concentra tion, and the da ta fit a quadra tic binding equation (Equation 1) to provide the apparent RNA K D values. The 5' PPP and 5' N AD RN As have an almost identical binding affinity with K D, app ∼ 2 nM. The 5' dpCoA and FAD capped RNAs bind 2-3-fold more weakly than 5 PPP but still bind with sub-nanomolar affinity ( K D, app ∼ 6-7 nM). The 5 OH 5 OVG binds with the weakest affinity ( K D, app ∼ 52 nM), 8-fold worse than the weakest binding metabolite-capped dsRNA, FAD. Additionally, each metabolite-ca pped dsRN A showed similar rates of ATP hydrolysis and comparable V max values as 5 PPP (Supplemental Figure S3), indicating that these are producti v ely bound RIG-I / RNA complexes. These results indicate that metabolite-ca pped RN As ar e r ecognized by RIG-I, like the Ca p0 dsRN A ( 9 ), and with affinities similar to the wellestablished RIG-I PAMP 5 PPP dsRN A. Additionall y, metabolite-ca pped RN As stim ulate RIG-I's ATPase to the same extent as the 5 PPP dsRNA.
Because the ATPase-based assay is an indirect method to measure RNA binding affinity, we also used fluorescence polariza tion assays tha t directly estima te the RNA K D values (Figure 3 B, Supplemental Figure S4). The same RNA panel was tested, and like the ATP hydrolysis assay results, RIG-I bound all metabolite-capped RNAs comparably to 5 PPP dsRNA and between 3.5-and 6-fold worse than 5 OH 5 OVG dsRNA. Thus, two RNA binding methods confirm that RIG-I binds metabolite-capped dsRNAs indistinguishably from 5 PPP dsRNAs.
Noticeabl y, all tested RN As a ppeared to bind more weakly in polarization assays than in the ATP hydrolysis assays (compare the K D and K D, app of 5 PPP of ∼50 and 1.5 nM, respecti v ely). In the ATPase assay, we titrate a small amount of RIG-I with increasing RNA, conditions that support mostly RIG-I monomers on RNA. On the other hand, we titrate a small amount of RNA in the polarization assay with increasing RIG-I concentrations, conditions that support RIG-I oligomerization (Figure 3 C). Thus, the RNA K D values from the polarization assay are composite values of multiple RIG-I binding e v ents rather than a single binding e v ent in the ATPase assay. This e xplains the difference in K D s' between the two assays.
With a footprint of a pproximatel y 10 bp per monomer ( 3 ), the 39-bp dsRNA can accommodate at least three RIG-I molecules, as shown by the EMSA results (Figure 3 C). With a threefold excess RIG-I over RNA, all RNAs supported RIG-I dimers, and trimers were observed on the 5 PPP and metabolite-capped dsRNAs. A higher-order disulfide-linked RNA dimer was observed with 5 dpCoA dsRNA, deduced from the minus RIG-I lane and detected by MS. The long 5 OH 5 OVG dsRNA also oligomerized RIG-I, but the amount of trimer was modest compared to the other RNAs tested.
We used an RNA competition assay to determine if RIG-I binds to 5 PPP or 5 N AD + ssRN As. A complex of RIG-I and fluorescein-labeled 5 PPP ds39 was mixed with increasing concentrations of unlabeled 5 PPP ds39 or 5 PPP / 5 N AD + ds39 ssRN A in the presence of ATP. While the ds39 competed effecti v ely with the labeled RNA, the ssRNAs did not, demonstra ting tha t RIG-I does not bind metaboliteca pped ssRN As noncanonicall y (Figure 3 D).

RIG-I can effectively utilize metabolite-capped dsRNAs as signaling ligands in HEK293T cells
Cellular reporter assays were used to assess whether the metabolite-ca pped RN As could induce the expression of a luciferase reporter gene under the control of IFN ␤ promoter and thus function as a RIG-I signaling ligand. A mammalian e xpression v ector with the RIG-I gene under a constituti v ely acti v e promoter was transfected into HEK293T cells lacking endogenous RIG-I ( 31 ). RIG-I expression was verified by western blotting (Supplemental Figure S6A). RIG-I expressing cells were transfected with 5 PPP dsRNA, 5 OH 5 OVG dsRNA, and the above-described panel of metabolite-capped dsRNAs. The metabolite-ca pped ssRN As did not produce an interferon response above the background (Figure 4 A). This is consistent with the weak affinity of RIG-I to ssRNA, reported earlier for 5 PPP ssRNA ( 32 ) and demonstrated here for metabolite-ca pped ssRN A by the competition assays (Figure 3 D). This result also demonstrates that there was no copyback dsRNA in our sample generated from the in vitro tr anscription. In contr ast to the ssRN As, the dsRN As with alternati v e caps signaled at le v els comparab le to 5 PPP dsRNA. The 5 OH 5 OVG dsRNA, our negati v e contr ol, signaled appr oximately threefold less than the capped or the 5 PPP dsRNA, consistent with its poor binding to RIG-I.
A dose-dependent study, where 5 PPP, 5 NAD + and 5 OH 5 OVG dsRNAs were titrated from 5 to 700 nM, was conducted to assess the relati v e signaling potency of these RNAs. The signaling activity of all tested RNAs re-mained constant between 25 and 700 nM, with 5 OH 5 OVG consistently signaling approximately 2.5-fold lower than 5 PPP or 5 NAD + (Figure 4 B). At 5 nM RNA concentration, the 5 OVG-dependent signaling decreased by 4fold relati v e to higher RNA concentrations, consistent with its weaker affinity for RIG-I, while 5 PPP and 5 NAD + dsRNA-dependent signaling did not significantly decrease. These results indicate that RIG-I recognizes NAD + capped RNA with a similar sensitivity as the well-established PAMP, 5 PPP RNA, which agrees with their similar RNA binding properties.
To determine if the metabolite-capped RNAs can activate the expression of endogenous interferon and antiviral r esponse-r elated transcripts, we used RT-qPCR to measure the mRNA le v els of IFNB1 (Interferon beta 1), OAS1 (2 -5 -oligoadenylate synthetase 1) and ISG15 (Interferon stimulated gene 15) after ectopic expression of RIG-I in HEK293T RIG-I KO cells (Figure 4 C-E ). In all tested cases, 5 PPP and 5 NAD + ds39 RNAs stimulated comparable OAS1, IFNB1 and ISG15 transcription le v els, 8-25-fold higher than the no RN A background. Additionall y, 5 OH 5 OVG ds39 signaled a pproximatel y 2-3-fold worse than 5 PPP or 5 NAD+, consistent with the reporter assay r esults. These r esults demonstra te tha t RIG-I ef fecti v ely reco gnizes metabolite-ca pped RN As as imm uno genic ligands in cellular contexts.

Endogenous RIG-I in A549 cells generates an innate immune response against metabolite-capped RNAs
The reporter signaling assay, shown above, relied on overexpressed, ectopic RIG-I. To determine if endogenous levels of RIG-I would recognize the metabolite-capped RNAs, we used A549 cells. In agreement with our reporter assay in HEK293T, the 5 PPP ds39 and all the metabolite-capped RNAs generated a robust antiviral signaling state in A549 cells, showing expression of IFNB1, OAS1, ISG15, and MX1 (MX dynamin-like GTPase 1) genes ( Figure 5 A-D ). In addition, Western blotting confirmed that RIG-I expression significantly increased in the presence of 5 PPP and metabolite-ca pped RN As (Supplemental Figure S7). Interestingly, with a lower le v el of RIG-I in A549 cells, the 5 OVG 5 OH dsRN A signaled very poorl y, comparabl y to no RNA transfected controls, and did not stimulate endogenous RIG-I expression. In contrast, when RIG-I was ectopically expressed in HEK293T experiments, this RNA generated an immune signal greater than the no RNA control but still less than the rest of the panel.
To confirm that the immune stimulation observed in A549 cells was RIG-I-dependent, we transfected A549 RIG-I KO cells with our RNA panel, monitoring the antiviral immune response using qPCR targeting IFNB1 (Figure 5 E). Agreeing with our hypothesis, none of the RNAs could stimulate antiviral signaling in RIG-I KO cells, demonstrating that the recognition of these RNAs is RIG-I dependent. Our study re v eals that metabolite-ca pped dsRN As are RIG-I ligands that activate the interferon and antiviral immune signaling pathways, r epr esenting a new class of RIG-I stim ulatory RN As.

DISCUSSION
The cellular machinery of the higher eukaryotes adds an m 7 G RNA cap to a mature mRNA with Cap1 modification to evade RIG-I recognition and pre v ent an immune response ( 9 , 10 ). Alternati v e, metabolite-based 5 RN A ca ps, such as NAD + , FAD and dephosphoCoA, are found in various organisms, including human cells where RIG-I is present (18)(19)(20). Whether RIG-I recognizes these newly identified caps as immunogenic or these caps confer protection against RIG-I recognition has important implications for the functions of metabolite-deri v ed RN A ca ps, as well as illustrating the mechanisms behind RIG-I's interaction with such RNAs. This study shows that RIG-I recognizes the metabolite-ca pped RN As as imm uno genic, similar to 5 triphosphorylated and Cap0 RNAs ( 9 , 10 ). RIG-I binds to, hydrolyzes ATP on, and can oligomerize on these metabolite-ca pped RN As --all essential biochemical functions for activating the RIG-I signaling pathway ( 30 ). Cellular assays show that endogenous and ectopically expressed RIG-I can utilize these metabolite-capped RNAs as signaling ligands to stimulate the antiviral immune signaling pathway. Animal studies are required to confirm the immunogenicity of metabolite-capped RNAs in vivo .
This study also describes a strategy to effecti v ely transcribe metabolite-ca pped RN As using the T7 RN AP system without purifying out contaminating 5 PPP RNA and copyback transcripts. First, the initiating nucleotide, A, is only present at the +1 position, which removes the need to add ATP. Secondly, the terminal two nucleotides of the template strand are methylated to promote T7 RNAP dissociation and reduce nonspecific 3 end extension and copyback transcription, resulting in clean and precise RNA ends. Finally, NTPs complementary to the terminal dinucleotide are excluded from the transcription reaction to inhibit copyback transcription. While these sequence constraints could be prohibiti v e for transcribing long RNAs or RN As w hose sequences are meant to mimic those found in vivo , this strategy does allow clean and convenient transcription of smaller metabolite-capped RNAs. The ADP-based metabolites system demonstrated here could be adapted to other metabolites with different NDP substructures, like UDP-GlcN Ac, or to artificiall y synthesized NDP-based moieties, with a concomitant change in RN A sequence w hile preserving the key design elements.
The study of RIG-I cap recognition so far has been limited to the role of the canonical m 7 G cap. This study demonstra tes tha t RIG-I has a broad tolerance for bulky groups with disparate geometries at the 5 end of the RNA with at least a diphosphate backbone between the cap and the first RNA nucleotide. A pre viously pub lished crystal structure of RIG-I bound to Cap0 dsRNA (m 7 G-capped RNA) ( 9 ), can be used to understand a potential mechanism of metabolite-ca pped RN A reco gnition by RIG-I. In this structure, CTD of RIG-I interacted e xclusi v ely with the triphosphate linkage without making noticeable interactions with the Cap0 moiety. By overlaying multiple members of the asymmetric unit, it was inferred that the m 7 G cap is fle xib ly accommodated in the CTD binding pocket without specific contacts (Figure 1 A). Likely, this is also occurring with the metabolite-capped dsR-N As. While metabolite-ca pped RN As contain a diphospha te ra ther than triphospha te linkage between RNA and cap moiety, the predominant interactions between 5 PPP and CTD are with the ␣and ␤-phosphates ( 3 , 9 ). The potential change in spatial orientation of the alternati v e cap due to the lack of a third phosphate linkage is not as crucial as the diphosphate linkage. Thus, it is likely that CTD binds to the diphosphate linkage without making important contacts with the metabolite ca p moieties. Interestingl y, the tested caps have a range of geometries: NAD + is similar in size to m 7 G canonical cap, dephosphoCoA has a long hydrocarbon chain, and FAD has a bulky, thr ee-member ed ring structure. The finding that large functional groups can be added to the 5 end of RNAs and still be recognized by RIG-I can be le v eraged for RN A technolo gies that rel y on RIG-I, either avoiding it (requiring additional modifications) or triggering it (5 end modifications imparting new functionalities to the RNA) ( 33 ).
The role of the metabolite-capped RNAs in mammals is not well understood, but if such RNAs accumulate in the cytoplasm of healthy cells, RIG-I will recognize them as imm uno genic. Sensiti v e RIG-I recognition of metaboliteca pped RN As that are endo genousl y made needs to be demonstrated. Still, if they are imm uno genic, additional mechanisms would be r equir ed to pre v ent deleterious autoimmune responses without viral infection. Such mechanisms could include internal RNA modifications, including methylation to inhibit RIG-I signaling ( 30 , 34 ) or methylating the 2 -O -ribose like in Ca p1 RN A. Howe v er, whether