Dengue Virus Impairs Mitochondrial Fusion by Cleaving Mitofusins

Mitochondria are highly dynamic subcellular organelles participating in many signaling pathways such as antiviral innate immunity and cell death cascades. Here we found that mitochondrial fusion was impaired in dengue virus (DENV) infected cells. Two mitofusins (MFN1 and MFN2), which mediate mitochondrial fusion and participate in the proper function of mitochondria, were cleaved by DENV protease NS2B3. By knockdown and overexpression approaches, these two MFNs showed diverse functions in DENV infection. MFN1 was required for efficient antiviral retinoic acid-inducible gene I–like receptor signaling to suppress DENV replication, while MFN2 participated in maintaining mitochondrial membrane potential (MMP) to attenuate DENV-induced cell death. Cleaving MFN1 and MFN2 by DENV protease suppressed mitochondrial fusion and deteriorated DENV-induced cytopathic effects through subverting interferon production and facilitating MMP disruption. Thus, MFNs participate in host defense against DENV infection by promoting the antiviral response and cell survival, and DENV regulates mitochondrial morphology by cleaving MFNs to manipulate the outcome of infection.


Introduction
Mitochondria, the powerhouse of cells, participate in various cellular events, such as ATP production, fatty acid synthesis, calcium homeostasis, and apoptosis induction [1,2]. Their roles in cell signaling are also emerging: mitochondria can sense perturbations of intracellular homeostasis, then regulate and transduce signaling responses, especially during stressful conditions [3,4]. The identification of mitochondrial antiviral signaling (MAVS), a mitochondrial outermembrane protein functioning as the adaptor of retinoic acid-inducible gene I (RIG-I)-like receptors (RLRs), revealed a link between mitochondria and antiviral innate immunity. Cytosolic viral RNA recognized by RLRs can activate MAVS and recruit various signaling molecules to transduce the downstream pathways, such as type I interferon (IFN) production [5][6][7][8] and cell death induction [9,10], two major cellular events controlling viral infection.
Mitochondria are highly dynamic double-membrane organelles, and their shapes change continually via the combined actions of fusion, fission and trafficking [11,12]. These dynamic events play critical roles in maintaining functional mitochondria because fusion promotes complementation of damaged mitochondria and fission creates new mitochondria [13,14]. Therefore, a balanced mitochondrial dynamics keeps mitochondria in good health and disturbing such physiological balance would contribute to diseases, such as abnormal brain development, autosomal dominant optic atrophy and Charcot-Marie-Tooth type 2A [15,16].
Two mitofusin proteins, MFN1 and MFN2, located on the mitochondrial outer membrane, mediate tethering and fusion of mitochondria [3,4,17]. Human MFN1 and MFN2 share 63% protein sequence identity and have the same relevant functional domains: a GTPase domain at the N-terminus, two coiled-coil domains (HR1 and HR2), and a transmembrane (TM) domain at the C-terminus [18]. MFN2 but not MFN1 contains a proline-rich region, and MFN2 is also present in the endoplasmic reticulum (ER) in addition to mitochondria [19]. Both MFNs mediate mitochondrial outer-membrane fusion by tethering of two adjacent mitochondria membranes with their HR2 domains, followed by GTP-required docking of both membranes before final fusion.
Accumulated findings demonstrated that antiviral RLR signaling can be regulated by the dynamics of mitochondria [4,[20][21][22]. Fibroblasts deficient in both MFNs showed impaired induction of RLR-induced antiviral responses [23]. MFN2 has been shown to interact with MAVS and suppress MAVS activating the IFNβ promoter [24]. Other reports also showed that MFN1 interacts with MAVS and mitochondrial fusion is required for efficient RLR signaling [25,26]. Thus, MFNs interact with MAVS and modulate RLR signaling, but the detailed involvement of these two MFNs in viral infection is largely unclear.
Dengue virus (DENV) is an enveloped flavivirus with a positive-sense RNA genome encoding a polyprotein. DENV infection in humans can cause diseases ranging from mild self-limited dengue fever to life-threatening dengue hemorrhagic and dengue shock syndrome [27]. The virus is transmitted by mosquitos, with possibly 390 million DENV infections every year [28]. So far, we lack an approved DENV vaccine or anti-DENV drug. Cells mainly sense DENV infection by two cytosolic RLRs, RIG-I and MDA5, then induce IFN production in a MAVS-dependent pathway [29,30]. In cells or animals with deleted MAVS, DENV-triggered IFN induction is significantly reduced, while DENV-induced mitochondrial membrane potential (MMP) disruption [10], cytopathic effects [10] and murine mortality [31] are attenuated. Thus, MAVS is involved in two cellular antiviral events, IFN induction and cell death, and contributes to the control and pathogenesis of DENV infection. To antagonize MAVS-mediated antiviral signaling, MAVS protein is cleaved by virus proteases such as that of hepatitis C virus (HCV) [6,32] and picornaviruses [33,34], and by cellular caspase in the cases of DENV infection [10], while MAVS-associated cofactor MITA/STING is cleaved by the DENV protease NS2B3 [35,36]. Thus, MAVS plays important roles in triggering the host defense against DENV infection, and DENV has evolved ways to counteract the MAVS-mediated signaling pathway.
In this study, we explored the relationship between DENV and mitochondrial dynamics and found that DENV infection can disrupt mitochondrial fusion by cleaving MFNs. Using cells with inducible expression of MFN1 or MFN2 and cells with MFN1 or MFN2 knockdown, we found that MFN1 was required for efficient RLR signaling, whereas MFN2 reduced the cell death triggered by DENV infection. Thus, we provide the first example of viral modulating of mitochondria dynamics by cleaving MFNs and identified two new cellular substrates of DENV protease. Our study shows that manipulating mitochondrial dynamics by viral protease could be one of the mechanisms contributing to DENV pathogenesis.

DENV infection attenuates mitochondrial dynamics
To explore whether DENV can affect mitochondrial dynamics, we performed a mitochondrial intermixing experiment, which has been used to investigate mitochondrial fusion/fission events by fusing two individual cells [37,38]. We established two stable A549 cell lines with either mitochondria-targeted YFP (mitoYFP) or RFP (mitoCherry) and fused them by using an HVJ Envelope Cell fusion kit ( Fig 1A). In mock-infected cells, continuous mitochondrial fusion/fission led to an even redistribution and colocalization of mitoYFP and mitoCherry in the fused cells, whereas much less colocalization of mitoYFP and mitoCherry was noted in the fused DENV-infected cells ( Fig 1B) at 24 h post infection (p.i.) when DENV replicated to high titer ( Fig 1C) but without significant cytotoxicity ( Fig 1D). The delayed redistribution of mitochondria might be resulted from reduced mobility, enhanced fission or suppressed fusion. We found no obvious difference in the mitochondrial movement between mock-and DENVinfected cells by the live-confocal microscopy analysis (S1 Movie and S1 Fig). We also checked the phosphorylation level of the fission-related dynamin-related protein 1 (Drp1) at Ser616 (pDrp1), which was induced by hepatitis B virus (HBV) and HCV to promote mitochondrial fission [39,40], as well as the endogenous protein levels of several mitochondria fusion/fission regulators. Levels of fission-related proteins, Drp1, pDrp1, optic atrophy 1 (OPA1) and mitochondrial fission 1 (Fis1) were similar during the course of DENV infection (Fig 1E and S2  Fig), while reduced fusion-related MFN1 and MFN2 proteins were detected in DENV-infected cells (Fig 1E). The mRNA levels of MFN1 and MFN2 remained constant in DENV-infected cells when viral RNA and IFNβ were induced exponentially (Fig 1F-1I), so the downregulation of MFNs proteins likely occurred at post-transcriptional level. Moreover, MFN2 protein levels were reduced in the peripheral blood mononuclear cells (PBMC) isolated from STAT1-deficient mice during the course of DENV infection ( Fig 1J). Overall, despite the role of mitochondrial fission-related molecules cannot be completely excluded, our data indicated that mitochondrial fusion might be affected in DENV-infected cells. To ask whether mitochondrial fusion is blocked by DENV infection, we overexpressed MFN1 to induce mitochondria hyperfusion and aggregated/grape-like cluster morphology [41,42]. To express MFN1 in a controlled manner, we established an A549 +on cell line that expressed an advanced Tet-On transactivator, which can turn on the Tet-responsive element (TRE)-tight promoter in the presence of doxycycline (Dox). We then transduced A549 +on cells with lentivirus expressing HA-tagged MFN1 under the control of TRE-tight promoter. With Dox treatment, the stable A549 +on /HA-MFN1 cells expressed HA-MFN1 (Fig 2A) around mitochondria ( Fig 2B) as previously reported [41,42]. Furthermore, distinct mitochondria morphologic features were seen in cells with or without MFN1 induction-clumped/aggregated versus filamentous/tubular mitochondria-as expected (Fig 2A).
We then addressed the potential influence of DENV infection on MFN1-triggered mitochondrial hyperfusion. A549 +on /HA-MFN1 cells were mock-infected or infected with DENV for 36 h before Dox treatment. Clumped mitochondria aggregations were apparent within 6 h of Dox treatment in mock-infected MFN1-overexpressing cells, whereas mitochondria remained pleomorphic in DENV-infected cells during the same time frame (Fig 2C and S2 Movie). Quantification by high-content analysis confirmed the microscopy observations: the proportion of cells with clumped mitochondria decreased from 86.2% to 37.1% when compared the mock-to DENV-infected cells ( Fig 2D). Thus, our data suggested that DENV can block MFN1-mediated mitochondria fusion.

MFNs protect cells from DENV-induced cell death
We next examined the effects of MFN1 overexpression on DENV infection. DENV-infected A549 +on /HA-MFN1 cells with Dox pre-treatment showed higher IRF3 phosphorylation (pIRF3), an indicator of RLR signaling activation, and lower viral production when compared to cells without Dox treatment (Fig 3A and 3B). Higher IFNβ mRNA expression was noted in Dox-treated MFN1-overexpressing cells (Fig 3C). We further used an IFN-sensitive recombinant sindbis virus containing a firefly luciferase reporter gene (dSinF-Luc/2A) [35,43] to assess the antiviral activity. The culture supernatants from MFN1-overexpressing cells suppressed the replication of dSinF-Luc/2A to a greater extent than the Dox (-) control (Fig 3D), indicating higher antiviral activity in the MFN1-overexpressing cells. MFN1-overexpression also blocked the DENV-triggered activation of apoptotic marker caspase 3 (Fig 3A), probably because of reduced DENV replication. To ascertain whether MFN1 and MFN2 share the same features in DENV infection, A549 +on /HA-MFN2 cells were established by lentivirus transduction as described for MFN1. Interestingly, MFN2 overexpression did not affect IRF3 phosphorylation, DENV viral protein expression (Fig 3E) or viral progeny production ( Fig 3F), but DENVinduced cytopathic effects, measured by caspase 3 activation (Fig 3E), LDH release ( Fig 3G) and MMP disruption, which initiates intrinsic route of apoptosis ( Fig 3H; attenuated from 16 + 12.2 = 28.2% to 10.2 + 9.4 = 19.6%), were less evident in Dox-treated A549 +on /HA-MFN2 cells. Therefore, MFN1 facilitated efficient RLR signaling and MFN2 alleviated MMP RT-qPCR (F to I) for the indicated genes. Data are mean ± SD (n = 3 per group). The data for LDH release was compared by one way ANOVA and Bonferroni multiple-comparison test with use of Prism 5 (GraphPad; La Jolla, CA, USA). ns, no significance; *, p<0.05. Nocodazole treatment (NOC; 100 ng/ml for 16 h) served as positive control to induce Drp1 phosphorylation at S616 residue. RQ, relative quantification. (J) STAT1 -/mice were inoculated with DENV (serotype 2, strain NGC-N) by an intraperitoneal plus intracerebral route. The peripheral blood mononuclear cells (PBMC) were isolated on day 0, 1, and 2 as indicated (n = 2 for each time point) and then sampled for western blot analysis with the indicated antibodies.  disruption during DENV infection. These data suggested that MFN1 and MFN2 play diverse roles in DENV infection despite that they share structural and functional similarity.

DENV protease NS2B3 cleaves MFNs
We noted protein reduction of endogenous and overexpressed MFNs with DENV infection (Figs 1E, 3A and 3E), so we explored whether MFN proteins were downregulated by DENV. We found several potential DENV protease-cleavage recognition sequences, two basic residues followed by a small amino acid [44][45][46][47], in human MFN1 and MFN2 proteins ( Fig 4A). To test whether DENV protease can cleave MFNs, we cotransfected cells with Flag-tagged DENV NS2B3 plus C-terminal V5-tagged MFN1 or MFN2 for western blot analysis. Besides the fulllength MFNs (~100 kDa), smaller protein bands of~28 kDa were detected by anti-V5 antibody in cells with wild-type (WT) but not protease-dead (S135A) DENV NS2B3 (Fig 4B). The protease from another flavivirus, Japanese encephalitis virus (JEV), failed to cleave MFNs (Fig 4B, lane 3 and 7), even though they share the same cleavage recognition sequences [48] and both of them can physically interact with MFNs ( Fig 4C), for unknown reasons.
Based on the sizes of the cleavage products ( Fig 4B), we predicted the positions of 538 PRN#A 541 in MFN1 and 562 SRR#A 565 in MFN2 are the most likely cleaved sites by DENV protease NS2B3 (Fig 4A). To verify that the smaller protein bands are indeed the cleavage products at the predicted sites, we created the MFN1-ST mutant with potential cleavage sequences changed from 538 PRN#A 541 to 538 STN#A 541 and the MFN2-AV mutant with cleavage consensus sequences 562 SRR#A 565 changed to 562 SRR#V 565 . Despite the mutant MFNs still bound with DENV protease (Fig 4D), much less cleavage products were seen in the MFN1-ST and MFN2-AV mutants ( We next addressed whether MFNs can be cleaved by DENV infection. In cells expressing N-terminal HA-tagged MFN1-WT but not the -ST mutant, anti-HA antibody revealed a cleavage product (~72 kDa) when MG132 was added to block proteasome-mediated protein degradation ( Fig 6A, lanes 5-6). Similarly, the~72 kDa cleaved fragment of MFN2 was evident in cells with MFN2-WT but not -AV mutant upon DENV infection with MG132 treatment ( Fig  6A, lanes 11-12). Thus, our results suggested that DENV targets both MFNs by its viral protease and the cleaved MFN fragments were further degraded by host proteasome machinery. Moreover, endogenous MFN1 and 2 can be cleaved by four serotypes of DENV, as smaller protein bands were noted in western blotting with anti-MFN1 and anti-MFN2 antibodies in cells infected with either one of the four DENV serotypes (Fig 6B). To examine whether expression of DENV protease alone is sufficient to cleave the endogenous MFNs, we checked the protein pattern of MFNs in cells with WT or protease-dead DENV protease whose IFN response was impaired in a protease activity dependent manner (S4 Fig) as reported in our previous study [35]. Smaller MFNs protein bands were detected in cells with WT but not the protease-dead S135A mutant of DENV NS2B3, especially in the presence of MG132 (Fig 6C). Since only a small portion of MFNs was cleaved, we further studied whether this cleavage caused mitochondrial morphology change. Short/fragmented and tubular mitochondria were noted in cells stably expressing WT and protease-dead NS2B3, respectively (Fig 7A). Furthermore, mitochondrial dynamics measured by the intermixing experiment of A549 cells stably expressing WT or S135A-mutated NS2B3 labeled with mitoYFP or mitoCherry (outlined in Fig 7B) was hampered in cells with DENV protease. The signals of mitoYFP and mitoCherry were less colocalized in cell hybrid harboring wild type NS2B3, while even redistribution of these fluorescent signals were found in the hybrid with mutated NS2B3 (Fig 7C). Thus, even though the endogenous MFNs were not completely cleaved by NS2B3, DENV protease affects mitochondrial morphology by governing its dynamics.

Manipulating MFNs expression governs consequences of DENV infection
To validate the role of endogenous MFNs in DENV infection, we knocked down the expression of MFN1 or MFN2 in A549 cells by transduction with a lentivirus expressing short hairpin RNA (shRNA) targeting MFN1 or MFN2. The resulting A549-shMFN1 and -shMFN2 cells showed reduced mRNA and protein expression of MFN1 and 2, respectively (Fig 8A and 8B). A549 cells with MFN1 and MFN2 double knockdown harbored severe fragmented mitochondria and showed constitutive death, and cannot be included in this study. DENVinduced LDH release was attenuated in A549-shMFN1 cells but enhanced in cells with shMFN2 ( Fig 8C). DENV viral production measured by plaque-forming assay was higher in A549-shMFN1 cells, whereas A549-shMFN2 cells produced similar levels of DENV as the shLacZ control (Fig 8D). Both of the antiviral activity against dSinF-Luc/2A in culture supernatants and IFNβ mRNA in cell lysates were reduced in samples from DENV-infected A549-shMFN1 cells, whereas those with MFN2 knockdown showed similar levels as the shLacZ control (Fig 8E and 8F). Consistently, lower level of pIRF3 was noted in cells with MFN1 knockdown (Fig 8G). Although MFN2 did not affect RLR signaling, knockdown of MFN2 deteriorated DENV-induced cytopathic effects measured by LDH release (Fig 8C), caspase 3 activation (Fig 8G), and MMP disruption in DENV-infected A549-shMFN2 cells ( Fig  8H). Furthermore, vesicular stomatitis virus (VSV), an IFN-sensitive virus, formed enlarged plaques on A549-shMFN1 cells, and to a lesser extent on MFN2-knockdown cells, as compared to the shLacZ control ( Fig 8I). Thus, MFN proteins participate in host defense against virus infection probably through promoting antiviral response and/or cell survival.

Different functions played by MFN1 and MFN2 in MAVS signaling
To clarify the role of MFN1 and MFN2 in MAVS signaling, we established A549 +on /myc-MAVS inducible cell line with MFN1 or MFN2 knockdown. MAVS overexpression induced by Dox treatment triggered IRF3 phosphorylation, IFNβ induction and caspase 3 activation in cells with shLacZ control (Fig 9A, lane 4; and Fig 9B); however, different responses were noted in cells with shMFN1 or shMFN2 expression. Silencing MFN1 attenuated MAVS-triggered pIRF3, IFNβ induction and caspase 3 activation (Fig 9A, lane 6; and Fig 9B), but enhanced caspase 3 activation was seen in cells with reduced MFN2 (Fig 9A, lane 8). Furthermore, MFN1 alone cannot turn on IFNβ promoter but can enhance MAVS-triggered IFNβ promoter activation. However, MFN2 and the N-terminal or C-terminal cleaved products of MFN1 failed to facilitate MAVS-mediated IFNβ promoter activation (Fig 9C). Thus, MFN1 and MFN2 interplay with the antiviral IFN and cell death pathways in RLR-MAVS signaling differently. MFN1

Cells with dominant-negative MFN1 show reduced RLR signaling and enhanced DENV infection
To further address the importance of MFN1 in DENV infection, we established a stable cell line overexpressing a mutated MFN1-T109A [42], which dominant-negatively blocked mitochondrial fusion and resulted in fragmented mitochondria (Fig 10A). Compared to the GFP control cells, DENV-infected MFN1-T109A cells showed lower pIRF3, higher caspase 3 activation ( Fig 10B) and more MMP disruption (Fig 10C). Consistently, MFN1-T109A cells induced  lower IFNβ mRNA (Fig 10D), and resulted in higher DENV RNA replication ( Fig 10E) and severer cytotoxicity measured by LDH release and cell viability (Fig 10F and 10G). Thus, mitochondria fusion contributes to host innate defense against DENV infection.

Discussion
Viruses depend on host cells to replicate and cellular factors involved in viral replication might be modulated during infection; these dysregulated cellular proteins/pathways may then contribute to viral pathogenesis. Various strategies have been adapted by viruses to modulate cellular proteins, for example DENV NS2B3, a protease essential for viral polyprotein processing, can cleave the IFN adaptor protein MITA/STING to subvert innate immunity [35,36]. Here, we identified two new cellular targets of DENV protease, MFN1 and MFN2, and also reveal their involvements in efficient RLR signaling and maintaining MMP in DENV-infected cells (as outlined in Fig 11). Downregulation of MFNs can impair mitochondrial dynamics and IFN signaling, thus promoting DENV replication and inducing cell death. Thus, to counteract the cellular antiviral responses, DENV protease not only cleaves IFN adaptor MITA/STING but also targets MFNs to blunt the mitochondria-mediated host defense.
Viruses affect mitochondrial dynamics, but the understanding of detailed mechanisms are limited to only a few cases [39,40,49]. Infection of alphaherpesviruses, such as herpes simplex virus type 1 and pseudorabies virus, blocks mitochondrial motility by reducing the recruitment of kinesin-1 to mitochondria via a viral glycoprotein B-dependent event [49]. Both HBV and HCV infection can promote mitochondrial fission by inducing the expression, phosphorylation and mitochondrial translocation of Drp1 [39,40]. Given that the dynamics and morphology of mitochondria are critical in innate immune response [20,22,23], the demonstration of particular viruses can manipulate mitochondrial dynamics has just begun to elucidate. Previously, DENV infection is known to cause apoptosis with mitochondrial involvement and DENV M protein has been shown to disrupt mitochondrial membrane potential [50,51]. Here, we provide the first example of viral modulating of mitochondria dynamics by cleaving MFNs and open a new research avenue on the interplay of viruses and cellular mitochondria.
Even though these two MFNs are similar in protein structure and function, compared with MFN1, MFN2 has lower GTPase activity [52] but has other specific functions such as oxidative metabolism [53,54], cell death involvement [55,56] and ER location in addition to mitochondria [19]. Previous findings on the roles of MFN1 and MFN2 on antiviral signaling are also not unified: MFN2, but not MFN1, interacts with MAVS and plays a suppressive role on the MAVS-triggered IFNβ promoter activation [24]; whereas subsequent reports showed that MFN1 interacts with MAVS and mitochondrial fusion is required for efficient RLR signaling [25,26]. Our findings that MFN1 and MFN2 play distinct roles in DENV infection, RLR signaling for MFN1 and MMP maintaining for MFN2, support that these two similar proteins are not functionally redundant in viral infection. MFNs may trigger mitochondria fusion to provide platforms for molecules in RLR signaling and anti-apoptotic pathways to interact with each other. However, it is not clear why MFN1 and MFN2 play different roles in DENV infection; possibilities, such as stronger GTPase activity of MFN1 and ER location of MFN2, might contribute to these phenomena and are awaited to be tested. Furthermore, DENV protease could cleave MFN1 and 2 at a region between the HR1 and TM domains, thus separating the GTPase domain from the HR2 domain and abolishing the proper function of MFNs. Our findings that DENV protease can cleave these two MFNs reinforce the importance of MFN-mediated cellular events in host defense against viral infection. The specific mechanisms contributed by MFN1/2 and the roles of these two MFNs in other viral infections remain to be further studied.
Malfunction of MFNs may cause severe diseases [15]. For example, mutations in the human Mfn2 gene are responsible for the inherited disorder Charcot-Marie-Tooth neuropathy type 2A [57]. Downregulation of MFN protein is a common mechanism modulating the function of MFNs, in that ubiquitination of MFNs has been found in several organisms such as mammals, yeast and fly [58][59][60]. Parkin, an E3 ubiquitin ligase involved in mitochondrial integrity, induces the ubiquitination of MFN1 and 2, which leads to their degradation and prevents or delays mitochondria fusion [61]. JNK phosphorylation of MFN2 leads to the recruitment of the large HECT-domain E3 ubiquitin ligase Huwe1 and results in ubiqitination-mediated proteasomal degradation of MFN2 [62]. The ubiquitin-proteasome system also participates in downregulating MFNs in DENV-infected cells because the proteasome inhibitor MG132 could block the subsequent degradation and increase the visibility of the cleavage protein bands. Thus, the protease-cleaved MFNs fragments were further targeted by proteasome-mediated degradation. Murine MFN1, whose sequences at the putative DENV protease cleavage site are similar to that of human MFN1-ST mutant, was also slightly downregulated in DENV-infected murine PBMC, supporting the involvements of cellular protein degradation pathways. Moreover, only a portion of MFN1 and 2 was targeted by DENV protease, which suggests that specific subcellular location and/or certain stimuli are required for the cleavage of MFNs by DENV.
Mitochondria constantly undergo fusion and fission, and this dynamic behavior is critical for maintaining the structure and function of healthy mitochondria. The shift in mitochondrial dynamics toward fission and mitophagy to attenuate apoptosis may facilitate persistent HBV and HCV infection [39,40]. Cleavage of MFN1 and 2 during DENV infection may also contribute to the pathogenesis of DENV-related diseases. Our findings that DENV cleaves MFNs to modulate mitochondria-related events such as antiviral RLR signaling and induction of apoptosis open up new research directions on the interplay of DENV with various cellular functions related to mitochondrial dynamics.

Coimmunoprecipitation
Cells were lysed with protein lysis buffer containing a cocktail of protease and phosphatase inhibitors (Roche). Cell lysates were immunoprecipitated with mouse anti-Flag IgG beads, control IgG beads (both Sigma-Aldrich) or rabbit anti-NS3 antibody (1:250; GTX124252; Gene-Tex) for overnight at 4°C. The immunocomplex was captured by use of Protein G Mag Sepharose (GE Healthcare) at 4°C for 1 h, washed with protein lysis buffer, resuspended in sample buffer with 2-mercaptoethanol, and then examined by Western blot analysis.

Cytotoxicity assay
Trypan blue exclusion test was performed for cell viability. The release of the cytoplasmic enzyme lactate dehydrogenase (LDH) was measured using a cytotoxicity detection kit (Roche).

Antiviral activity assay
As previously described [35], conditioned medium was UV-inactivated and then serial-diluted with fresh medium. Vero cells were cultured with the conditioned medium for 18 h, then infected with dSinF-Luc/2A virus for 24 h. Cell lysates were harvested for luciferase activity assay (Promega).

Flow cytometry
For MMP and annexin V analyses, cells were detached and resuspended in 1 ml warm medium. After 15 to 30 min of JC-1 (1 μM) labeling, cells were stained with annexin V-Cy5 (BioVision), then analyzed by use of the FACSCanto flow cytometer and FACSDiva software (Becton Dickinson) as described [10].
in PBS-T, antibodies against HA-tag (Cell Signaling) and NS3 [35] were added for overnight. Alexa Fluor-488-conjugated secondary antibody (Molecular Probes) was added for 1 h at room temperature, and cell nuclei were counterstained with DAPI. For samples examined by confocal laser-scanning microscopy (LSM 510, Zeiss), cells were seeded in μ-Slides chamber slides (ibidi) overnight before treatments. For live confocal microscopy, samples were examined by use of the UltraVIEW VoX live cell imaging system with Volocity software (PerkinElmer). For high-content analysis of mitochondrial morphology, we used the ImageXpress Micro Image XL System (Molecular Devices).

Mitochondrial dynamics assays
For the mitochondria intermixing experiment, cell fusion was done by using HVJ Envelope Cell Fusion Kit (Cosmo Bio) with previously described modifications [38]. Briefly, A549/ mitoYFP cells were detached and treated with HVJ-E at 4°C for 5 min, then overlaid onto semi-confluent A549/mitoCherry cells. HVJ-E-mediated cell membrane fusion was initiated in the presence of translation inhibitor cyclohexamide (50 μg/ml) by increasing the temperature to 37°C. For DENV-infection group, A549/mitoYFP and A549/mitoCherry cells were infected with DENV (moi 10) for 24 h, then HVJ-E fusion kit was used as the mock-infection group.

Ethics statement
Animal studies were conducted according to the guidelines outlined by Council of Agriculture, Executive Yuan, Republic of China. The animal protocol was approved by the Academia Sinica Institutional Animal Care and Utilization Committee (Protocol ID 11-11-245). All surgery was performed under sodium pentobarbital anesthesia and every effort was made to minimize suffering.

Statistical analysis
Data are presented as mean ± SD and were compared by two-tailed Student's t test, or analyzed by one way ANOVA and Bonferroni multiple-comparison test as mentioned in figure legends. The live confocal microscopy of S1 Movie was composed of photos taken every 0.255 sec. Since the difference between two sequential images within identical field can be subtracted as the movement [61], the sequential mitochondrial movements of DENV-infected cells within 30 sec were analyzed frame-by-frame (117 frames) by Volocity and MetaMorph software and normalized to that of mock. The result is shown as mean ± SD (n = 116, per group) and present in percentage.