The microRNAs miR-302b and miR-372 regulate mitochondrial metabolism via the SLC25A12 transporter, which controls MAVS-mediated antiviral innate immunity

MicroRNAs (miRNAs) are small noncoding RNAs that suppress the expression of multiple genes and are involved in numerous biologic functions and disorders, including human diseases. Here, we report that two miRNAs, miR-302b and miR-372, target mitochondrial-mediated antiviral innate immunity by regulating mitochondrial dynamics and metabolic demand. Using human cell lines transfected with the synthetic analog of viral dsRNA, poly(I-C), or challenged with Sendai virus, we found that both miRNAs are up-regulated in the cells late after viral infection and ultimately terminate the production of type I interferons and inflammatory cytokines. We found that miR-302b and miR-372 are involved in dynamin-related protein 1 (DRP1)-dependent mitochondrial fragmentation and disrupt mitochondrial metabolism by attenuating solute carrier family 25 member 12 (SLC25A12), a member of the SLC25 family. Neutralizing the effects of the two miRNAs through specific inhibitors re-established the mitochondrial dynamics and the antiviral responses. We found that SLC25A12 contributes to regulating the antiviral response by inducing mitochondrial-related metabolite changes in the organelle. Structure–function analysis indicated that SLC25A12, as part of a prohibitin complex, associates with the mitochondrial antiviral-signaling protein in mitochondria, providing structural insight into the regulation of the mitochondrial-mediated antiviral response. Our results contribute to the understanding of how miRNAs modulate the innate immune response by altering mitochondrial dynamics and metabolic demand. Manipulating the activities of miR-302b and miR-372 may be a potential therapeutic approach to target RNA viruses.

In this study, we investigated the mechanistic and functional roles of miRNAs in mitochondrial-mediated innate immunity. Although some miRNAs were previously demonstrated to be involved in antiviral innate immunity (4 -6, 15), how those miRNAs function to facilitate innate immunity by changing the mitochondrial potentials has remained unclear. We clarified that two miRNAs, miR-302b and miR-372, which are involved in the embryonic stem cell-specific cell cycle and cell reprogramming (16,17), target mitochondrial-mediated antiviral innate immunity through their regulation of mitochondrial dynamics and metabolic demand. These insights contribute to elucidate the mechanistic roles of the miRNAs essential for mitochondrial-mediated innate immunity, and manipulating these activities may be a potential therapeutic antiviral target.

Searching for microRNAs linked to antiviral innate immunity and mitochondrial dynamics
To elucidate the expression profile of miRNAs in cells that respond to a viral infection event, we first evaluated changes in miRNA expression in human embryonic kidney 293 (HEK293) cells, a widely-used human cell line that responds to a synthetic analog of viral dsRNA, poly(I-C) transfection. We followed 53 miRNAs that were up-regulated more than 3-fold in poly(I-C)transfected cells relative to those in the mock state (Table S1). To confirm that we would be observing a specific response to activation of the RLR-dependent signaling pathway, we induced another viral infection using Sendai virus (SeV), a negative-stranded RNA virus of the Paramyxoviridae family. Remarkably, nine miRNA candidates (miR-1250, miR-143, miR-302b, miR-338, miR-372, miR-410, miR-485, miR-520f, and miR-922) were well-matched between these two different stimuli ( Fig. 1A and Table S1).
We therefore evaluated whether these miRNAs with increased levels had a role in the RLR-signaling pathway. Although the delivery of poly(I-C) into HEK293 cells potently activated an IFN-␤ luciferase-based reporter (Fig. 1B, NC), cotransfection of each miRNA mimic with poly(I-C) had various effects on the activation of its reporter. Among the nine candidates mentioned above, we confirmed that the delivery of three miRNA mimics (miR-302b, miR-372, and miR-520f) with poly(I-C) was sufficient to inhibit signal transduction (Fig. 1B). Interestingly, the delivery of two miRNA mimics, miR-302b and miR-372, in HeLa cells (widely used for mitochondrial imaging) led to extensively fragmented mitochondria ( Fig. 1C and Table 1). These findings suggest that miR-302b and miR-372 affected the activity of both the RLR pathway and mitochondrial morphology. Because miR-302b and miR-372 have homologous seed sequences that can target overlapping genes (18), we mainly analyzed the function of miR-302b and confirmed the results against miR-372 in the following experiments.

miR-302b acts as a negative regulator of the mitochondrialmediated antiviral response
Having identified that miR-302b (also miR-372) might link antiviral cellular innate immunity with mitochondrial potentials, we evaluated the functional role of the miRNA involved in mitochondrial-mediated antiviral innate immunity. The kinetic profile of IFN-␤ expression in A549 cells (widely used in viral research) infected with SeV peaked at 6 h post-infection (p.i.) as detected by quantitative PCR (qPCR), and the IFN-␤ mRNA level was sharply decreased at 12 h p.i. (Fig. 1D, top), which is a typical antiviral immune response (19). Expression of miR-302b, which was delayed relative to that of IFN-␤, was induced 12 h after infection and increased for up to 36 h p.i.; this lagging trend in the expression was also observed in cells transfected with poly(I-C) (Fig. 1D, bottom; see also Fig. S1A). To verify that the observed RLR-mediated up-regulation of miR-302b was not due to a specific characteristic of the cell type we used, we performed the same miRNA induction experiments in other human cell lines, MRC-5 (fetal lung fibroblasts) and HAP-1 (widely used for biomedical and genetic research), and confirmed that both cell types also exhibited increased expression of miR-302b ( Fig. 1E and Fig. S1B).
Because altering the level of miR-302b in cells would be consistent with an antiviral immune response (Fig. 1, B and D), we hypothesized that the miRNA acts as a modulator of the RLRsignaling pathway. To test this hypothesis, we treated HEK293 cells with the miR-302b mimic and assessed their immune response against either viral infection or poly(I-C) stimulation. The viral-triggered induction of IFN-␤ and other pro-inflammatory cytokines such as RANTES or TNF␣ was dramatically suppressed in cells transfected with the miR-302b mimic ( Fig.  1F and Fig. S1, C and D). In addition, we confirmed that an abundant presence of the miR-302b mimic in the cells similarly attenuated the phosphorylation of endogenous interferon regulatory factor 3 (IRF-3; Fig. 1G), a hallmark of IRF-3 activation, and the production of endogenous IFN-␤ protein ( Fig. 1H and Fig. S1, E and F). Treatment of HEK293 cells with the miR-302b mimic similarly reduced the activation of both IFN-␤ and NF-B reporters in response to overexpression of either the N-terminal CARD domain of RIG-I (designated as RIG-I 1-250 ) or MAVS, the MOM protein that acts downstream of RIG-I ( Fig. S1G) (9). Most importantly, interfering with either endogenous miR-302b or miR-372 by applying its specific inhibitors to poly(I-C)-stimulated HEK293 ( Fig. 2A) or HAP-1 (Fig. 2B) cells sufficiently increased the production of IFN-␤. Taken together, these observations indicated that miR-302b (also miR-372) acts as a negative regulator of the mitochondrial-mediated antiviral response and suggests that it would act in the late stage of viral infection.

miR-302b regulates the DRP1-dependent pathway in the mitochondrial fission process
Because the accumulation of miR-302b in HeLa cells leads to remarkable mitochondrial fragmentation that is restored by the presence of its specific inhibitors (Fig. 3, A and B, and Fig. S2, A and B), we next attempted to elucidate whether certain molecules involved in mitochondrial dynamics are associated with MicroRNAs regulate mitochondrial-mediated antiviral response the phenomenon. To reveal its mechanistic actions, we performed microarray analysis in HEK293 cells transfected with the miR-302b mimic. Gene expression profiling of the cells revealed an association of miR-302b with the mitochondrial dynamics (Fig. 3C, left heat

MicroRNAs regulate mitochondrial-mediated antiviral response
and MFF, adaptors for dynamin-related protein 1 (DRP1) (20), and significant down-regulation of Ras-related protein 32 (RAB32), all of which participate in mitochondrial fission (21)(22)(23), were further confirmed by gene expression analysis with qPCR in the miR-302b-transfected cells (Fig. 3C, right graph). Indeed, the search for target genes against miR-302b using the program TargetScan revealed that RAB32 had a putative binding sequence in the 3Ј-UTR of the mRNA. To confirm the predicted results experimentally, we constructed a reporter plasmid in which the 3Ј-UTR region of RAB32(110 -140) was cloned into a Renilla luciferase reporter gene. The results of the reporter assay confirmed that the luciferase activity in HEK293 cells transfected with the miR-302b mimic was greatly attenuated when the WT RAB32 plasmid was co-transfected, whereas co-transfection with its mismatch-induced mutant plasmid had no effect ( Fig. 3D and Fig. S2C). Consistent with these results, delivery of the miR-302b mimic into the cells significantly affected the abundance of endogenous RAB32 at both the mRNA (Fig. S2, D and E) and protein (Fig. 3E) levels. In addition, we observed that eliminating RAB32 through the action of miR-302b suppressed the phosphorylation of DRP1 at Ser 637 (Fig. 3F), a hallmark of DRP1 activation (24), as seen in HEK293 cells treated with siRNA against RAB32 (Fig. S2F).

MicroRNAs regulate mitochondrial-mediated antiviral response
In contrast, HEK293 cells transfected with the miR-302b mimic also exhibited increased expression of the DRP1 adaptors MID49 and MID51, as confirmed by the increase in the amounts of their proteins in the cells (Fig. 3G). DRP1 recruitment in the mitochondrial fraction was sufficiently increased in an miR-302b-dependent manner (Fig. 3H), further indicating that our morphologic phenotype resulted in mitochondrial fragmentation (25). Together, these results highlighted the mechanism by which miR-302b is involved in mitochondrial fission via the DRP1-dependent pathway.

miR-302b also targets a mitochondrial transporter, SLC25A12, which modulates MAVS-mediated signaling
The presence of excess fragmented mitochondria ultimately leads to cellular signaling dysfunction, including the mitochondrial-mediated antiviral response (14). Therefore, we next evaluated whether arresting DRP1 activity in cells treated with an miR-302b mimic (Fig. S2G) would affect RLR-mediated signal transduction. As reported previously (14), depleting endogenous DRP1 in HEK293 cells through its specific siRNA enhanced the induction of both IFN-␤ and RANTES in response to poly(I-C) stimulation relative to that in control siRNA-transfected cells (Fig. 3I). However, the loss of DRP1 in cells treated with the miR-302b mimic also showed potential induction of immune responses compared with controls, but the recovery was only partial in the cells transfected with a negative control mimic (Fig. 3I, NC/siDRP1 versus miR-302b/ siDRP1). These findings suggest that some other factors besides mitochondrial morphology are involved in the RLR pathway, leading us to explore the direct involvement in signal transduction.
Thorough analysis of our microarray data revealed that SLC25A12, a member of the solute carrier family, was intensively down-regulated in miR-302b-transfected HEK293 cells (Fig. 4A). As expected, SLC25A12 possessed an miR-302bbinding sequence in its 3Ј-UTR ( Fig. 4B and Fig. S3A), whose low abundance of protein was experimentally confirmed in cells transfected with the miRNA mimic (Fig. 4C). Strikingly, the antiviral immune responses detected on the basis of the expression of IFN-␤ and RANTES in HEK293 cells stimulated by poly(I-C) were significantly decreased in an SLC25A12-depleted condition (Fig. 4D). Knockdown of endogenous SLC25A12 in the cells similarly reduced IFN-␤ reporter activation in response to the overexpression of either RIG-I 1-250 or MAVS constructs (Fig. S3B). We also verified the antiviral immune function of SLC25A12 using SLC25A12-deficient HAP-1 cells (Fig. 4E). Because SLC25A12, an aspartateglutamate transporter localized in the mitochondrial inner membrane (MIM) (26), is required for regulating antiviral signaling through MAVS, it is likely that SLC25A12 acts downstream of MAVS in the pathway.

SLC25A12 associates with MAVS in mitochondria
While exploring the mechanisms of action of SLC25A12, we found that the transporter co-immunoprecipitated with endogenous MAVS in HEK293 cells under physiologic conditions (Fig. 5A). We therefore performed a structurefunction analysis to fill the topologic gap between SLC25A12 and MAVS. Using a co-immunoprecipitation approach, we mapped the region of SLC25A12 that interacts with MAVS via the C-terminal region (Fig. 5B) that faces the intermembrane space (IMS) with an unknown function (27). Consistent with the immunoprecipitation results, fluorescence microscopy confirmed the co-localization of both SLC25A12 and MAVS in HeLa cells (Fig. 5C). We explored additional molecules that might bridge SLC25A12 and MAVS at the IMS and found that prohibitins (PHB1 and PHB2), mitochondrial scaffolds localized in the MIM (28), associated with SLC25A12 (Fig. 5, A and D). We assume that SLC25A12, as a part of the PHB interactome (28), interacts with MAVS in mitochondria.
Interestingly, decreasing SLC25A12 in cells affected the structural features of the MAVS activation accompanied by RLR-mediated signal transduction, i.e. homotypic oligomerization (29,30) at the MOM (Fig. 5, E and F), which further highlights the importance of the transporter in the activation of the signaling pathway. In addition, SLC25A12-specific siRNA had no effect on IFN-␤ reporter activation in

MicroRNAs regulate mitochondrial-mediated antiviral response
response to the overexpression of TANK-binding kinase 1 (TBK-1) (Fig. S3B), a kinase that targets IRF-3, despite the fact that the effector acts downstream of MAVS (9). On the basis of these findings, we propose a model in which SLC25A12 inhibits the RLR pathway downstream of MAVS (and/or at the same level) and upstream of the kinase TBK-1.

miR-302b regulates mitochondrial metabolism via SLC25A12, which controls antiviral responses
The results described above prompted us to investigate the functional role of SLC25A12 in the regulation of the antiviral response. Because SLC25A12 is a member of the malateaspartate NADH shuttle (31,32), we assessed its ability to transfer reducing equivalents. Treatment with the miR-302b mimic

MicroRNAs regulate mitochondrial-mediated antiviral response
significantly decreased the level of NADH in HEK293 cells without changing NAD, resulting in an ϳ50% greater NAD/ NADH ratio (Fig. 6A), which indicated impaired mitochondrial metabolism. Metabolome analysis of the miRNA-transfected cells further revealed that the amounts of aspartate, malate, and pyruvate were similarly decreased in cells treated with the miR-302b mimic, while the amount of lactate was increased (Fig.  6B). Consistent with our observations, a decrease in the oxygen consumption rate (OCR) indicated that the miR-302btargeted mitochondria in the cells became less active ( Fig. 6C and Fig. S4A), and the cellular metabolism shifted to glycolytic conditions (Fig. 6D). We confirmed that the observed regulation of mitochondrial-related metabolites was mainly due to the function of SLC25A12 on the basis of diminishing SLC25A12 in cells with a lower OCR (Fig. 6E and Fig. S4B). To explore whether these metabolic changes in the mitochondria correlated with RLR pathway activation, we performed a signaling assay. As expected, the immunodeficiency observed in the siSLC25A12treated HEK293 cells was significantly recovered by adding aspartate to the media (Fig. 6F), similar to the OCR patterns in SLC25A12-deficient cells (Fig. S4C). In addition to the substantial contribution of SLC25A12 to mitochondrial metabolism, we identified that miR-302b targets mitochondrial pyruvate carrier 1 (MPC1) (Fig. S4, D and E) (33,34), and changes in the amount of its protein partially affect mitochondrial respiration (Fig. S4B). Taken together, these findings indicate that the absence of mitochondrial metabolites leads to defective mitochondrial-mediated antiviral signaling.

Discussion
MicroRNAs are known to be involved in various cellular functions, and several miRNAs are triggered by an immune response to induce IFNs (35,36). Here, we report that two miR-NAs, miR-302b and miR-372, potentially attenuate virus-triggered induction of IFN-␤ and pro-inflammatory cytokines by manipulating mitochondrial functions, such as mitochondrial dynamics and metabolism.
Although much attention has been focused on the identification and characterization of signaling molecules that activate and/or inactivate mitochondrial-mediated antiviral innate immunity, our study explored the mechanistic basis of how these molecules are regulated at the mRNA level and the involvement of the translated products in the signaling event.
On the basis of our findings, we propose that miR-302b and miR-372 regulate mitochondrial-mediated antiviral responses by controlling the expression of mitochondrial proteins (such as SLC25A12) to prevent excessive progression of the antiviral response (Fig. 6G). In addition to revealing the mechanistic action of these miRNAs, the findings of this study may pave the way for understanding an unexpected role of these miRNAs as a therapeutic target for attenuating excess inflammation, such as a cytokine storm because of their high potency for downregulating pro-inflammatory cytokines.
Finally, a previous study revealed that miR-302a, a member of the miR302 family encoded in the human miR302-367 cluster (37), suppresses influenza A virus, an RNA virus of the Orthomyxoviridae family, stimulates interferon regulatory factor-5 (IRF-5) expression, and inhibits the production of IFNs and inflammatory cytokines (38). The seed sequences of miR-302b and miR-372 are homologous to that of miR-302a, indicating that both miR-302b and miR-372 would likely act as negative regulators of IRF-5 to prevent the induction of a cytokine storm, as in the case of miR-302a. Therefore, miR-302b and miR-372 may function as potential regulators of the influenza A virus-induced cytokine storm and provide a candidate target for the treatment of influenza A virus infection.

Reagents
Poly(I-C) was purchased from InvivoGen (San Diego, CA), and enzyme-linked immunosorbent assay (ELISA) kits for human IFN-␤ were supplied by Kamakura Techno-Science Inc. (Kanagawa, Japan) and R&D Systems (Minneapolis, MN). All other reagents were biochemical research grade.

Reverse transcription qPCR
Total RNA was isolated from cells with an miRNeasy mini kit (Qiagen, Hilden, Germany) or a Maxwell 16 LEV simplyRNA cell kit (Promega) according to the manufacturer's protocol, and 500 ng of total RNA was reverse-transcribed with a QuantiTect reverse transcription kit (Qiagen) or TaqMan MicroRNA reverse transcription kit (Thermo Fisher Scientific) to generate cDNAs. Real-time PCR assays were performed with a TaqMan Gene Expression assay using the indicated probes.

ELISA
Production of IFN-␤ was measured with species-specific ELISA reagents for human IFN-␤ from Kamakura Techno-Science Inc. and R&D Systems.

Confocal microscopy
Cells were plated on coverslips in 12-well plates (5 ϫ 10 4 cells/well). The following day, the cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.2% Triton X-100 in 1ϫ PBS (pH 7.2), and blocked with 5% fetal bovine serum. Mitochondria were visualized by using either mitochondrial-targeted red fluorescent protein (12) (mito-RFP; Fig.  1C) or staining with anti-mtHSP70 ( Fig. 3A and Fig. S2, A and G) or MAVS (Fig. 5C) pAbs followed by Alexa Fluor 568conjugated polyclonal and Alexa Fluor 488 -conjugated polyclonal secondary Abs, respectively. Cells were imaged with a C2ϩ confocal microscope (Nikon Instruments Inc.). In this process, we assume that SLC25A12, as a part of the PHB interactome (inset, highlighted with blue background) (28), interacts with MAVS in mitochondria that have a critical role in MAVS oligomerization. At longer times post-infection (ϳ24 h), miR-302b (and also miR-372) is up-regulated in the cells, which promotes mitochondrial fission through the DRP1-dependent pathway and also impairs mitochondrial metabolism (➁). We propose that the miRNAs undergo sequential negative regulations following viral infection, a process that is critical for terminating excess RLR signaling.

Luciferase reporter assay
HEK293 cells were plated in 96-well plates (3 ϫ 10 4 cells/ well). On the same day, the cells were co-transfected with 40 ng of luciferase reporter plasmid (p125luc or pELAM), 4 ng of Renilla luciferase internal control vector phRL-TK (Promega), and each of the indicated expression plasmids or poly(I-C) (500 ng) using Lipofectamine 2000 reagent (Invitrogen) following the manufacturer's protocols. The cells were harvested 24 h after transfection and analyzed using a Dual-Glo Luciferase Assay System (Promega). Relative luciferase activity was calculated by normalizing luciferase activity by the Renilla luciferase activity expressed in internal control plasmids and dividing by the normalized value of the control in which only an empty vector (pcDNA3.1) was transfected. Each experiment replicated at least three independent reproducible results. The p125luc and pELAM reporter plasmids were provided by T. Taniguchi (University of Tokyo, Japan) and T. Seya (Hokkaido University, Japan), respectively.

BRET assay
BRET experiments were performed as described previously (30) with minor modifications. HEK293 cells were transfected with miRNA mimics. At 48 h after the treatment, the miRNA mimic-transfected cells were plated in a 96-well microplate and transfected with 5 ng of Nanoluc-tagged MAVS plasmid and increasing amounts of Venus-tagged MAVS constructs using Lipofectamine 2000 at the same time. Twenty hours after transfection, 0.1 l of Nanoluc substrate was added to each well, followed by 1 min of gentle mixing, and luminescence was measured simultaneously for the donor ( em ϭ 460 nm; short wavelength) and acceptor ( em ϭ 515 nm; long wavelength).

Immunoprecipitation
Co-immunoprecipitation experiments were performed as described previously (39) with minor modifications. HEK293 cells at 80% confluence were transiently transfected with the appropriate plasmids (1.25 g each) in a six-well plate using the FuGENE HD transfection reagent (Promega). The following day, the cells were lysed with 1 ml of radioimmunoprecipitation assay (RIPA) buffer with a protease inhibitor mixture, and the clarified supernatants were incubated overnight at 4°C with anti-c-Myc-agarose beads (Sigma). After four washes with 1ϫ PBS (pH 7.2), the immunoprecipitates were resolved by 10% SDS-PAGE and analyzed by Western blotting with a mAb against HA followed by a horseradish peroxidase-conjugated antibody against mouse IgG (Cell Signaling Technology). To immunoprecipitate endogenous MAVS, PHB1, or PHB2, HEK293 cells in a 10-cm dish were washed once with 1ϫ PBS (pH 7.2), lysed with 2 ml of RIPA buffer containing protease inhibitor mixture, and the clarified supernatants were incubated with 10 g of anti-SLC25A12 rabbit mAb followed by incubation overnight at 4°C with 25 l of protein A-Sepharose beads. The beads were washed four times with 1ϫ PBS (pH 7.2), and the immunoprecipitates were resolved by 10% SDS-PAGE and immunoblotted with each Ab.

Mitochondrial fractionation
The cells were harvested and washed once with cold 1ϫ PBS (pH 7.2) and resuspended in 700 l of homogenization buffer (20 mM HEPES (pH 7.4), 70 mM sucrose, and 220 mM mannitol) by 30 strokes in a Dounce homogenizer. The homogenate was centrifuged at 800 ϫ g for 10 min, and the resulting supernatant was further centrifuged at 10,000 ϫ g for 10 min at 4°C to precipitate the crude mitochondrial fraction. The resultant supernatant was centrifuged at 20,000 ϫ g for 30 min to obtain the cytosol fraction. The crude mitochondrial fraction was resuspended in homogenization buffer and applied to a discontinuous Percoll gradient comprising 80, 52, and 26% (w/v) Percoll diluted in homogenization buffer. Centrifugation was performed at 43,000 ϫ g for 45 min. The purified mitochondrial fraction was recovered from the interface between the 52 and 26% (w/v) Percoll.

Mitochondrial respiration and glycolysis analysis
Mitochondrial respiration and glycolysis analysis were performed using the Seahorse XFe96 Analyzer (Agilent Technologies, Santa Clara, CA). HEK293 cells were seeded on XFe96 spheroid microplates (6 ϫ 10 4 cells/well) the day before the experiment and analyzed using the Seahorse XF Cell Mito Stress Test kit or the Seahorse XF Glycolysis Stress Test kit according to the manufacturer's protocols. Before measuring mitochondrial OCR, the culture medium was replaced with XF base medium supplemented with 1 mM pyruvate, 10 mM glucose, and 2 mM L-glutamine. First, the baseline values were measured three times, and the mitochondrial OCR in each state was then measured three times after adding the mitochondrial respiratory chain inhibitors to the cells. Oligomycin (2 M), FCCP (1.5 M), and rotenone/antimycin A (0.5 M) were used as mitochondrial respiratory chain inhibitors. The obtained data were analyzed using Wave 2.4.0 software (Agilent Technologies).

Microarray analysis
Total RNA was isolated from cells transfected with the negative control miRNA mimic or miR-302b mimic 72 h later using an miRNeasy mini kit (Qiagen) according to the manufacturer's protocol. RNA samples were quantified using an ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE), and the quality was confirmed with an Experion System (Bio-Rad). The cRNA was amplified, labeled (low-input Quick Amp labeling kit), and hybridized to the SurePrint G3 Human GE version 3 8 ϫ 60K microarray according to the manufacturer's instructions (Agilent Technologies). All hybridized microarray slides were scanned by an Agilent scanner (Agilent Technologies). Relative hybridization intensities and background hybridization values were calculated using Agilent Feature Extraction Software (version 9.5.1.1). Raw signal intensities and flags for each probe were calculated from the hybridization intensities (gProcessedSignal) and spot information (gIsSaturated, etc.) according to the procedures recommended by Agilent. The raw signal intensities of four samples were normalized by a quantile algorithm with the "prepro-cessCore" library package (40) of the Bioconductor software (41). To identify up-regulated or down-regulated genes, we cal-MicroRNAs regulate mitochondrial-mediated antiviral response culated the Z-scores (42) and ratios (nonlog scaled fold-change) from the normalized signal intensities of each probe to compare between control and experimental samples, and we established criteria for regulated genes: Z-score Ն2.0 and ratio Ն1.5-fold for up-regulated genes, and Z-score ՅϪ2.0 and ratio Յ0.66 for down-regulated genes, respectively. The GEO accession number for the microarray data reported in the paper is GSE129615.

Metabolite measurement for LC-MS and GC GC-MS
Metabolites were extracted from cultured cell pellets using water/methanol/chloroform (1:1:2), and the aqueous phase was preserved. For LC/MS experiments, the dried metabolites were dissolved in 50 l of LC/MS-grade water (Wako Pure Chemical Industries, Osaka, Japan) and then filtered with a 0.45-m Millex filter unit (Merck Millipore). For GC-MS experiments, derivatization of the samples was carried out in two steps using methoximation and N-methyl-N-trimethylsilyltrifluoroacetamide ϩ 1% trimethylchlorosilane (Pierce) as described previously. Levels of NAD and NADH were determined by multiple reaction monitoring using an Agilent 6460 Triple Quad mass spectrometer coupled with an Agilent 1290 HPLC system. The MS settings and chromatographic conditions used were described previously (43). Tricarboxylic acid cycle intermediates were analyzed by selected ion monitoring using an Agilent 5977 MSD Single Quad mass spectrometer coupled with Agilent 7890 gas chromatograph as described previously (44). Metabolite amounts were calculated by the integrated sum of the area using Mass Hunter Quantitative software (Agilent Technologies).

Statistical analysis
An analysis of variance test (GraphPad Prism) was used for the statistical analyses. We considered different populations of cells to be biologic replicates; aliquoting or repeated measurements of a cell population was considered to represent technical replicates. We performed at least three independent reproducible results for most key experiments, although we did not perform explicit power calculations. Data are presented as the mean Ϯ S.D. Statistical significance was assessed by Student's t test. A p value of less than 0.05 was considered statistically significant.