Mitofusin-2 stabilizes adherens junctions and suppresses endothelial inflammation via modulation of β-catenin signaling

Endothelial barrier integrity is ensured by the stability of the adherens junction (AJ) complexes comprised of vascular endothelial (VE)-cadherin as well as accessory proteins such as β-catenin and p120-catenin. Disruption of the endothelial barrier due to disassembly of AJs results in tissue edema and the influx of inflammatory cells. Using three-dimensional structured illumination microscopy, we observe that the mitochondrial protein Mitofusin-2 (Mfn2) co-localizes at the plasma membrane with VE-cadherin and β-catenin in endothelial cells during homeostasis. Upon inflammatory stimulation, Mfn2 is sulfenylated, the Mfn2/β-catenin complex disassociates from the AJs and Mfn2 accumulates in the nucleus where Mfn2 negatively regulates the transcriptional activity of β-catenin. Endothelial-specific deletion of Mfn2 results in inflammatory activation, indicating an anti-inflammatory role of Mfn2 in vivo. Our results suggest that Mfn2 acts in a non-canonical manner to suppress the inflammatory response by stabilizing cell–cell adherens junctions and by binding to the transcriptional activator β-catenin.

O ne devastating manifestation of the disassembly of endothelial adherens junctions (AJs) is acute lung injury (ALI), in which an excessive immune response triggers the disruption of the lung endothelial barrier, fluid, and protein leak into the alveolar space resulting in compromised oxygen exchange and an unhinged state of inflammatory activation 1,2 . Understanding the mechanisms leading to the breakdown of endothelial barrier function as well as the inflammatory pathways within the endothelium are essential for developing therapeutic strategies to treat ALI patients 3 . The vascular endothelium maintains an intact barrier to prevent leakage of circulating nutrients, solutes, and fluid into the tissues as well as tightly regulating the influx of immune cells 4 . However, during severe infections and subsequent inflammatory responses, the endothelial barrier is compromised due to the breakdown of endothelial AJs. The plasma membrane AJ protein complex in endothelial cells consists of VE-cadherin as well as members of the catenin family such as β-catenin and p120-catenin 5 . The influx of immune cells following AJ disassembly exacerbates the inflammatory process, leading to a feed-forward activation of inflammation and further destruction of the endothelial barrier 1 .
There are multiple mechanisms underlying the breakdown of AJs such as signaling induced by the pro-inflammatory cytokine TNFα which promotes inflammation via the generation of reactive oxygen species (ROS) 6,7 . ROS mainly mediate reversible or irreversible oxidative modification at cysteine residues 8 . Reactive cysteine thiol (SH) converts cysteine sulfenic acid (Cys-SOH), termed as protein sulfenylation, a key initial mediator of redox signaling 9 . Prior studies of inflammatory redox signaling and barrier function have understandably focused on known junctional proteins at the plasma membrane. However, recent studies on inflammatory signaling implicate the involvement of mitochondria [10][11][12] .
The mitochondrial GTPases Mitofusin-1 (Mfn1) and Mitofusin-2 (Mfn2) are key regulators of mitochondrial function by mediating mitochondrial network fusion which allows for the distribution of proteins, mitochondrial DNA, and metabolites to maintain network connectivity 13,14 . Mitofusin 1 and Mitofusin 2 are located in the outer membrane of mitochondria and thus may facilitate interactions with other organelles 13 . The function of Mfn1/2 can be regulated by post-translational modifications such as ubiquitination, acetylation, or phosphorylation 13 . Mfn1/2 also acts as tethers for mitochondria with each other or with other organelles such as ER and as mitochondrial anchoring proteins 15 . Although Mfn1 and Mfn2 demonstrate approximately 80% sequence homology, double mutant embryos die earlier than either single mutant, and the post-natal Mfn1 knockout (KO) mice do not exhibit significant pathology whereas Mfn2 KO mice show rapid lethality 16 . It has thus been suggested that nonredundant functions of Mfn1 and Mfn2 that may go beyond the traditionally ascribed mitochondrial fusion roles of these proteins. Non-fusion unique roles of Mfn2 include the mediation of mitophagy and apoptosis 13 , or regulation of contact sites with ER 14,16,17 . However, the potential roles of Mfn2 in the formation or stabilization of adherens junctions are not yet known.
In this study, we identified a non-canonical function of Mfn2 as a stabilizer of endothelial adherens junction complexes during homeostasis. Mfn2 is sulfenylated during inflammation and accumulates in the nucleus where it acts as an endogenous suppressor of inflammation by binding the transcriptional regulator β-catenin.

Results
Mfn2 binds to the adherens junction complex during homeostasis. To identify binding partners of Mfn2 in an unbiased manner, we performed a comprehensive proteomic analysis in human lung microvascular endothelial cells (HLMVECs). Mfn2 was overexpressed in HLMVECs using lentiviral GFP-Mfn2 and GFP-Mfn2 was immunoprecipitated with GFP-trap magnetic agarose to reduce non-specific binding that may be present with the use of anti-GFP antibodies. Overexpression of lentiviral GFP in HLMVECs was used as a control to assess protein partners that non-specifically bind to GFP ( Supplementary Fig. 1a, b). We focused on protein partners unique to the expression of GFP-Mfn2 versus the GFP-control. The obtained proteins were analyzed using high-affinity liquid chromatography with tandem mass spectrometry (LC-MS/MS). We considered candidates which are identified in at least two samples among three independent experiments with a >1.5-fold increase threshold. We found that 25 proteins specifically interacted with Mfn2 in ECs (see Source data) and all identified protein partners were classified based on their function using gene ontology enrichment analysis ( Supplementary Fig. 1c). We found that Mfn2 interacted with the expected mitochondrial protein partners but, we also identified non-mitochondrial protein partners. The nonmitochondrial proteins interacting with Mfn2 included cadherin binding partners and cell-cell adhesion partners such as KRT18, ENO1, or MACF1 ( Supplementary Fig. 1c). In order to understand the functional significance of this intriguing finding, we focused on studying the role of non-mitochondrial Mfn2.
To understand the specific roles of Mfn2 in AJ protein complexes, we next investigated potential binding partners of Mfn2 within AJs in HLMVECs. Total Mfn2 proteins were immunoprecipitated and followed by immunoblotting with antibodies for AJ complex proteins VE-cadherin or β-catenin. Mfn2 significantly interacted with VE-cadherin and β-catenin in resting ECs (Fig. 1i, j). To verify the specificity of the Mfn2 interactions at AJs, Mfn2 was depleted in HLMVECs with a doxycycline inducible lentiviral Mfn2 shRNA. Mfn2 depletion (Mfn2-KD) decreased the interaction of Mfn2 with VE-cadherin and β-catenin ( Supplementary Fig. 1k). Importantly, these biochemical approaches independently confirmed that Mfn2 interacts with the endothelial AJ proteins VE-cadherin and β-catenin at the plasma membrane, consistent with the 3D-SIM findings (Fig. 1a).
Mfn2 localized at the plasma membrane promotes endothelial barrier integrity during endothelial homeostasis. To determine whether the presence of Mfn2 at AJs affects AJ function as manifested by endothelial barrier integrity, we next investigated endothelial barrier function in control and Mfn2-depleted HLMVECs. We confirmed that our Mfn2 depletion strategy was specific to Mfn2 and did not affect Mfn1 levels ( Supplementary  Fig. 2a, b). Mfn2 depletion disrupted adherens junctions (AJs) as visualized by VE-cadherin and β-catenin immunostaining (Fig. 2a), whereas Mfn1 depletion did not affect the endothelial barrier ( Supplementary Fig. 2c), indicating that the observed barrier stabilizing effect was specific for Mfn2. The area and fluorescence intensity of VE-cadherin or β-catenin at AJs were quantified with ImageJ ( Fig. 2b and Supplementary Fig. 2d). Next, we used a transendothelial resistance (TER) assay and a FITCconjugated albumin permeability assay to assess endothelial barrier integrity. Mfn2 depletion significantly decreased transendothelial resistance (TER) ( Fig. 2c and Supplementary Fig. 2e). Similarly, the transendothelial permeability for the FITCconjugated albumin tracer was increased in Mfn2-depleted HLMVEC monolayers (Fig. 2d). Moreover, we examined whether the loss of barrier integrity could be rescued by GFP-Mfn2 overexpression. We designed an Mfn2 shRNA to target 3′-UTR regions of Mfn2 gene instead of the coding sequences (CDS), thus making GFP-Mfn2 impervious to Mfn2 shRNA. We observed that the increased permeability in Mfn2-KD cells was significantly rescued by overexpressing GFP-Mfn2 as assessed by FITCconjugated albumin permeability (Fig. 2d). We also assessed the barrier by confocal microscopy and found that the disrupted barrier in Mfn2-depleted ECs was restored after GFP-Mfn2 overexpression (Fig. 2e, f).
It is known that filamentous actin (F-actin) plays a critical role in stabilizing cell-cell contacts at AJs 21 . Our proteomic analysis had shown that Mfn2 interacts with an actin binding protein, MACF1 (Microtubule Actin Crosslinking Factor 1) (Supplementary Fig. 1c). Thus, we examined whether the effects of Mfn2 on the endothelial AJ stability may in part relate to its interaction with F-actin. Control endothelial cells showed a regular F-actin structure, whereas Mfn2-KD ECs clearly demonstrated decreased sites of cell-cell contact and a disassembly of F-actin contact points using LifeAct-EGFP live cell imaging of actin filaments ( Supplementary Fig. 2f). Furthermore, we found that F-actin colocalization with VE-cadherin was dependent on Mfn2 (Supplementary Fig. 2g, h). These results suggest that Mfn2 may help anchor F-actin structures at AJs and thus promote endothelial barrier integrity.
In addition, we investigated the possibilities that our observed effects of Mfn2 depletion on the loss of endothelial barrier integrity may reflect a form of generalized cellular stress that would increase cell death, oxidative stress, or reduce cell proliferation. We found that Mfn2 depletion in homeostatic ECs did not decrease cell proliferation or induce apoptosis (Supplementary Fig. 3a-f). Furthermore, Mfn2 depletion did not increase mitochondrial ROS production ( Supplementary Fig. 3g, h) and also had no significant effect on mitophagy (Supplementary Fig. 3i, j), ER stress ( Supplementary Fig. 3k), and total AJs protein levels in homeostatic ECs ( Supplementary Fig. 2l). Next, we examined whether changes in the mitochondrial morphology could affect EC barrier integrity, as Mfn2 is a key regulator of mitochondrial fusion and therefore any effects on EC barrier integrity may be mediated by changing mitochondrial dynamics. We used siRNA to specifically deplete the mitochondrial fission mediator Drp1 or the mitochondrial fusion mediators Opa1, Mfn1, and Mfn2 ( Supplementary Fig. 3m, n). Even though the siRNA depletions were sufficient to modify mitochondrial network structure, as seen in the increase of mitochondrial fragmentation following Opa1 depletion or the increase in mitochondrial elongation with Drp1 depletion, EC barrier integrity was only significantly decreased with Mfn2 depletion suggesting that the observed Mfn2 effects on EC barrier integrity were not primarily related to the Mfn2 role in mitochondrial fusion. Taken together, these data suggest that Mfn2 binds to AJ complexes at the plasma membrane where it specifically stabilizes the endothelial barrier and regulates F-actin filament structures.
Endothelial-specific deletion of Mfn2 increases inflammatory injury in vivo. We next studied whether Mfn2 regulates the endothelial barrier integrity in vivo and used a tamoxifeninducible EC-specific conditional Mfn2 knockout (Mfn2 EC−/− ) mice ( Supplementary Fig. 4a, b). First, we investigated mRNA or protein expression levels of Mfn2 in whole lungs from Mfn2 fl/fl (littermate controls) and Mfn2 EC−/− mice. The mRNA levels of Mfn2 showed about 50% knockdown efficiency in whole lungs from Mfn2 EC−/− mice ( Supplementary Fig. 4b). To verify specific deletion of Mfn2 in ECs, we examined protein levels of Mfn2 in ECs isolated from the lungs of Mfn2 fl/fl and Mfn2 EC−/− mice. ECs from Mfn2 EC−/− mice demonstrated greater than 80% reduction of Mfn2 levels compared to ECs from Mfn2 fl/fl (Fig. 3a,  b). Next, we examined the functional role of endothelial Mfn2 in the regulation of in vivo vascular permeability. Mfn2 EC−/− and Fig. 1 Non-canonical localization of Mfn2 at the plasma membrane in endothelial cells. Confluent HLMVECs were fixed with 4% PFA for 10 min and permeabilized with 0.25% Triton X-100 for 10 min. a Mfn2 cellular localization was examined by Mfn2 immunostaining with co-staining of the mitochondrial membrane protein Tom20 and the AJ protein VE-cadherin. The images were taken using three-dimensional illumination microscopy (SIM, DeltaVision OMX, Olympus 60X/1.42 NA Plan Apo objective). Images represent maximum intensity projection of 15-17 Z-planes (125 nm step size) acquired in full frame (1024 × 1024 pixel) structured illumination mode. Mfn2, Tom20, and VE-cadherin were pseudocolored with green, blue, and red, respectively. The white arrow heads represent non-mitochondrial Mfn2 at the AJs along the plasma membrane. b Co-localization of VE-cadherin, Mfn2, and Tom20 was analyzed by comparing the fluorescence intensity for each protein. The white lines, R1 and R2 of (a), indicate the area where the distance between Mfn2 and VE-cadherin, or Mfn2 and Tom20 was analyzed using ImageJ, respectively. The black arrow heads represent their co-localized regions. c The co-localization of Mfn2 and VE-cadherin was statistically estimated by Manders' overlap coefficient analysis. The images are representative of n = 9 biologically independent samples. Data are mean values ± SEM. p = 0.0048 by paired, two-tailed t-test. d HLMVECs expressing GFP-Mfn2 were used to determine localization of Mfn2 and Tom2 by immunostaining with a specific antibody for Tom20 using confocal microscopy (Zeiss LSM880, Plan 1.46NA, ×63 magnification). Green color indicates GFP-Mfn2 and red color indicates Tom20. The white arrow heads represent non-mitochondrial Mfn2 in the cytosol. e Co-localization analysis of GFP-Mfn2 and Tom20 at R1 of (d) using ImageJ. The black arrows represent non-mitochondrial Mfn2. The images are representative of at least 3 independent experiments. f Mfn1 cellular localization was examined by Mfn1 immunostaining with co-staining of Tom20 and VE-cadherin using 3-D SIM. Images represent maximum intensity projection of 15-17 Z-planes (125 nm step size) acquired in full frame (1024 ×1024 pixel) structured illumination mode. Mfn1, Tom20, and VE-cadherin were pseudocolored with green, blue, and red, respectively. g Co-localization analysis of VEcadherin and Mfn1 at R1 and R2 of (f) using ImageJ. R1 and R2 was used for analyzing co-localization of Mfn1 and VE-cadherin or of Mfn1 and Tom20, respectively, and black arrows represent their co-localized regions. h The co-localization of Mfn1 and VE-cadherin was statistically estimated by Manders' overlap coefficient analysis. The images are representative of n = 5 biologically independent samples. Non-significance (ns), p = 0.1577 by unpaired, twotailed t-test. i Confluent resting HLMVECs were used to determine interaction of Mfn2 and AJ proteins, VE-cadherin and β-catenin. Mfn2 were immunoprecipitated with a specific antibody followed by Western blotting with VE-cadherin or β-catenin antibodies. #1 and 2 indicate two independent experiments. Uncropped blots can be found in the Source Data file. j Quantification of band intensities in (I) using Image J. Data are mean values ± SEM for n = 4 independent experiments. *p = 0.0407 for VE-cadherin, *p = 0.0261 for β-catenin by paired, two-tailed t-test.
Mfn2 fl/fl mice were injected intravenously with the Evans bluealbumin dye to assess lung vascular permeability 22 . The lungs from Mfn2 EC−/− mice demonstrated significantly higher leakiness than lungs from Mfn2 fl/fl control mice (Fig. 3c, d). We also examined the expression levels of pro-inflammatory genes in the lungs of Mfn2 fl/fl or Mfn2 EC−/− mice. As shown in Fig. 3e, the expression of IL-β, IL-6, TNFα, or IFNγ genes was significantly increased in lungs of Mfn2 EC−/− mice than in those of Mfn2 fl/fl control mice (Fig. 3e).
Furthermore, the lungs from Mfn2 EC−/− mice demonstrated significant accumulation of immune cells in the proximity of venous blood vessels but not arterial blood vessels when compared to the lungs of control Mfn2 fl/fl mice (Fig. 3f). Finally, to determine the phenotype of infiltrating immune cells in the  23 , and performed flow cytometry after staining for cell typespecific markers. We found that the lungs of Mfn2 EC−/− mice were significantly enriched for myeloid cells such as interstitial macrophages (IMs) and neutrophils (Neu), as well as lymphoid B + T cells and natural killer T (NKT) cells when compared to the lungs of Mfn2 fl/fl mice (Fig. 3g, h). However, B cells and T cells were decreased in the lungs from Mfn2 EC−/− mice, suggesting an activation of the innate immune response in resting lung endothelial cells upon endothelial Mfn2 deletion ( Supplementary  Fig. 4c, d). These data indicate that endothelial Mfn2 is a critical regulator of the endothelial barrier integrity during homeostasis and that its absence induces spontaneous vascular leakiness as well as inflammation.
Mfn2 interaction with β-catenin is enhanced following their disassociation from adherens junctions. After establishing a key role of Mfn2 in stabilizing the endothelial barrier during homeostatic conditions, we next examined the role of Mfn2 during inflammation because inflammatory stimulation promotes AJ disassembly 24 . HLMVECs were stimulated with TNFα and the AJ complex proteins VE-cadherin or β-catenin were immunoprecipitated, followed by immunoblotting. We found that inflammatory stimulation increased the interaction between Mfn2 and β-catenin, but not between Mfn2 and VE-cadherin (Fig. 4a, b and Supplementary Fig. 4e, f). We confirmed these results by using a GFP-tagged Mfn2 construct that was expressed in ECs and found a six-fold increase in the interaction between Mfn2 and β-catenin following inflammatory activation with TNFα ( Fig. 4c, d). We also used an in situ proximity ligation assay (PLA), which creates a spatial fluorescent signal within a 30-40 nm maximum distance and can thus establish close proximity interactions between protein partners 25,26 . The known interaction between VE-cadherin and β-catenin was used as a positive control. The interaction between VE-cadherin and β-catenin (red dots) was present under homeostatic baseline conditions but rapidly decreased following TNFα stimulation (Fig. 4e, f). However, the binding of Mfn2 and β-catenin (green dots) was markedly increased in the cytosol (Fig. 4g, h). Moreover, EC barrier impairment in Mfn2-KD ECs was further increased by TNFα stimulation, and rescued by overexpressing GFP-Mfn2 ( Supplementary Fig. 4g). Based on multiple lines of inquiry, we concluded that Mfn2 disassociates from VE-cadherin and there is a concomitant increase of Mfn2 interaction with β-catenin in the cytosol following inflammatory stimulation.
TNFα-induced ROS increase the binding of Mfn2 to β-catenin. We next investigated the mechanism by which inflammatory activation with TNFα could affect the interaction between Mfn2 and β-catenin. ROS are key mediators of inflammatory signaling by increasing reversible oxidative modifications 8 . As TNFα has been shown to increase ROS production 7 , we posited that ROSinduced cysteine modifications may post-translationally regulate interactions of Mfn2 with partner proteins. ROS production increased within 15 min of inflammatory stimulation with TNFα (Fig. 5a). Exogenous ROS (H 2 O 2 ) increased the interaction of Mfn2 and β-catenin (Fig. 5b), consistent with the notion that Mfn2 modified by inflammation-induced ROS promotes the interaction of Mfn2 with cytosolic β-catenin when both are disassociated from AJ complexes, but that Mfn2 no longer interacts with VE-cadherin. Moreover, the complex of Mfn2 and β-catenin exhibited band shifts that were dependent on Mfn2 in a nonreducing SDS-PAGE gel ( Supplementary Fig. 5a) and decreased in Mfn2-KD ECs in reducing SDS-PAGE gels ( Supplementary  Fig. 5b). These data suggest that their interaction may be mediated via disulfide bond formation that was dependent on the presence of ROS. Then, we performed a rescue experiment using the ROS scavenger N-Acetyl-L-cysteine (NAC) 27 which reduces disulfide bond formation 28 . As shown in Fig. 5c, d, the increased interaction of Mfn2 and β-catenin by H 2 O 2 was reversed by NAC treatment. These data suggest that during inflammation, ROS mediate the interaction of Mfn2 and β-catenin. Moreover, we investigated whether the complex formation of Mfn2 and βcatenin was associated with cysteine sulfenylation, a key initial step for cysteine oxidation. To examine whether TNFα induced-ROS modulate cysteine sulfenylation of Mfn2 and β-catenin, TNFα stimulated-ECs were lysed with lysis buffer containing DCP-Bio1, a cell permeable biotin-labeled Cys-OH trapping probe (DCP-Bio1) 29 to capture all sulfenylated-proteins. The captured proteins were used to determine sulfenylation of Mfn2 or β-catenin by Western blotting with their specific antibodies. Importantly, sulfenylation of both Mfn2 and β-catenin was significantly increased 1 h after TNFα stimulation and then subsequently decreased in a time-dependent manner (Fig. 5e, f). We then addressed the role of sulfenylation in the binding of Mfn2 and β-catenin. Control and Mfn2-depleted ECs were stimulated with 500 µM H 2 O 2 for 30 min and subjected to a DCP-Bio1 assay. H 2 O 2 treatment of control ECs induced cysteine sulfenylation of Fig. 2 Mfn2 at the plasma membrane promotes endothelial barrier integrity. HLMVECs were infected with doxycycline inducible lenti-Mfn2 shRNA virus for 48 h and treated with doxycycline for 72 h to deplete endogenous Mfn2 (Mfn2-KD). DMSO was used as a vehicle for control ECs. a Barrier integrity of control and Mfn2-KD ECs was examined by VE-cadherin and β-catenin immunostaining using confocal microscopy (Zeiss LSM880, Plan 1.46NA, ×63 magnification). b The levels of VE-cadherin and β-catenin at the cell surface in (a) were determined by analyzing area (upper panel) or fluorescence intensity (lower panel) using ImageJ and represented as % of control. Data are mean values ± SEM for n = 5 biologically independent samples. *p = 0.0184, **p = 0.0024 vs control by paired or unpaired, two-tailed t-test. c EC barrier resistance (Ω) of control and Mfn2-KD ECs was measured with TER. The data are presented without normalizing to show baseline difference. Data are presented with box and whiskers plot and whiskers are Min to Max. n = 5 biologically independent experiments (TER was measured every 8 s for 40 min). ****p = 0.0001 vs control by paired, two-tailed t-test. d Control and Mfn2-KD ECs were grown on transwell plates (0.4 µm pore) and treated with a FITC-albumin tracer (0.    4 Mfn2 binding to the AJ complex protein β-catenin is increased following inflammatory stimulation. Confluent HLMVECs were treated with TNFα (10 ng/mL) for 6 h. a Mfn2 was immunoprecipitated with Mfn2 specific antibody followed by Western blotting for β-catenin or Mfn2 antibodies. Uncropped blots can be found in the Source Data file. b The band intensities for Mfn2 and β-catenin interactions in (a) were quantified with ImageJ. Data are mean values ± SEM for three independent experiments. *p = 0.0403 by unpaired, one-tailed t-test. c HLMVECs expressing GFP or GFP-Mfn2 were stimulated with TNFα (10 ng/mL) for 6 h. The GFP-Mfn2 was immunoprecipitated with GFP-trap magnetic agarose followed by Western blotting for βcatenin and Mfn2 antibodies. Uncropped blots can be found in the Source Data file. d The band intensities of the Western blot were quantified with ImageJ. Data are mean values ± SEM for three independent experiments. *p = 0.0097 by unpaired, one-tailed t-test. e-h Proximity ligation assay (PLA). e The interaction of endogenous VE-cadherin and β-catenin in HLMVECs was examined using mouse VE-cadherin antibody and rabbit β-catenin antibody for positive control. The red dots show an interaction between VE-cadherin and β-catenin. f Quantification of the number of interactions (red dots) per field in (e). Data are mean values ± SEM from n = 6 for basal and n = 7 for TNFα treatment in independent biological replicates samples. **p = 0.0018 by paired, two-tailed t-test. g The interaction of endogenous Mfn2 and β-catenin in HLMVECs was examined with mouse Mfn2 antibody and rabbit β-catenin antibody. The green dots show interaction between Mfn2 and β-catenin, and DAPI (blue) indicates nuclei. h Quantification of the number of interactions (green dots) per cell in (g). Data are presented with box and whiskers plot and whiskers are Min to Max. n = 8 for basal and n = 14 for TNFα. ****p < 0.0001 by paired, two-tailed t-test. The images in (e) and (g) are representative from at least six independent biological replicates.
To investigate whether sulfenylation of Mfn2 also occurs during inflammation in vivo, we used the experimental model of endotoxemia which induces profound inflammatory injury. C57/ BL6 mice were intraperitoneally injected with the bacterial endotoxin lipopolysaccharide (8 mg/kg sublethal LPS) or PBS (control) and we monitored Mfn2 sulfenylation at 6 h and 24 h in whole lungs. LPS induced inflammation significantly increased Mfn2 sulfenylation at 6 h (Fig. 5g, h), indicating that Mfn2 sulfenylation is a post-translational modification that also occurs in vivo. However, we found that Mfn2 depletion or TNFα stimulation did not induce phosphorylation at serine 33 and 37 residues of β-catenin which are known to promote β-catenin  Fig. 5d, e). TNFα (10 ng/mL) stimulation had no effect on total protein levels of AJs complex until 24 h (Supplementary Fig. 5f). Taken together, these data suggest that inflammation-induced Mfn2 sulfenylation increases its interaction with sulfenylated-β-catenin.
Transcriptional activity of β-catenin is negatively regulated by Mfn2 during inflammation. To address the potential functional roles of the Mfn2-β-catenin interaction during inflammation, we investigated whether the presence of Mfn2 affects the expression of key pro-inflammatory genes such as Intercellular cell adhesion molecule-1 (ICAM-1), Interleukin-6 (IL-6), and −18 (IL-18). Control and Mfn2-KD ECs were stimulated with TNFα and mRNA expression levels of pro-inflammatory genes were evaluated by quantitative real-time PCR (qRT-PCR). Mfn2 depletion in ECs significantly increased TNFα-induced mRNA levels of ICAM-1, IL-6, and IL-18, indicating that Mfn2 suppresses the TNFα induced pro-inflammatory response in ECs (Fig. 6a, b).
Upon disassociating from AJs complexes at plasma membrane, β-catenin translocates to the nucleus where it acts as a key cofactor for the transcription factor, T-cell factor (TCF) and thereby mediates Wnt signaling 30 . Since we had found that Mfn2 binds βcatenin after inflammation-induced disassociation from AJs, we next investigated whether inflammation affects β-catenin transcriptional activity using a β-catenin luciferase assay. TNFα stimulation significantly increased β-catenin transcriptional activity in a time-dependent manner ( Supplementary Fig. 6a). Interestingly, β-catenin transcriptional activity was significantly increased by Mfn2 depletion and further enhanced by TNFα stimulation in ECs (Fig. 6c), thus suggesting that Mfn2 acted as a suppressor of β-catenin-mediated transcriptional activation during inflammation. The exaggerated β-catenin transcriptional activity was reset by restoring Mfn2 expression (Fig. 6d).
We then examined whether β-catenin directly regulates the expression of pro-inflammatory genes. HLMVECs were transfected with siRNA for β-catenin (Supplementary Fig. 6b) and mRNA levels of pro-inflammatory genes were examined with or without TNFα stimulation. The TNFα-induced expression of ICAM-1, IL-β, or IL-18 genes was significantly inhibited in βcatenin knockdown ECs (Fig. 6e). Taken together, these data indicate that Mfn2 functions as a suppressor for β-cateninmediated transcriptional activation during inflammation.
TNFα triggers nuclear accumulation of Mfn2. Next, we investigated whether Mfn2 accumulates in the nucleus during inflammation. The precise subcellular localization of Mfn2 was determined by a subcellular fractionation assay as well as by confocal microscopy at baseline and following TNFα stimulation. We first confirmed nuclear translocation of the pro-inflammatory transcription factor NF-kB after TNFα stimulation 29,31 to validate the degree of inflammatory activation induced by TNFα in ECs ( Supplementary Fig. 6d). Mfn2 was mainly localized in the cytosolic/mitochondrial fraction but a portion of Mfn2 clearly accumulated in the nuclei at 6 h after TNFα stimulation (Fig. 7a,  b), whereas Mfn1 showed no such accumulation ( Supplementary  Fig. 6d). It came as a surprise because Mfn2 does not have a nuclear localization sequence (NLS) even though the presence of mitochondrial proteins in the nucleus has been recently described 32,33 . To verify nuclear accumulation of Mfn2 during inflammation, we investigated whether exogenously overexpressed GFP-Mfn2 is also found in the nucleus along with endogenous Mfn2 during inflammation using confocal microscopy with three-dimensional Z-stacking. GFP-Mfn2 accumulated in the nucleus along with endogenous Mfn2 after 6 h of TNFα stimulation (Fig. 7c, d). Moreover, it was confirmed by 3dimensional image analysis using all sections (average 91 sections) and ortho analysis using one section to show colocalization of GFP-Mfn2, Mfn2 (red color), and DAPI (Fig. 7e). We found that β-catenin translocated into the nucleus at 6 h after TNFα stimulation, mirroring the nuclear accumulation pattern we had observed for Mfn2 (Fig. 7f, g and Supplementary  Fig. 6d, e). The persistence of the Mfn2/β-catenin interaction in the nucleus by TNFα stimulation was further demonstrated by a proximity ligation assay (PLA) (Fig. 7h). We next examined whether Mfn2 affected nuclear accumulation of β-catenin using confocal microscopy. The TNFα-induced nuclear accumulation of β-catenin was not affected by Mfn2 depletion ( Supplementary  Fig. 6f, g). Taken together, these results indicate that Mfn2 and βcatenin both accumulate in the nucleus and that nuclear Mfn2 may suppress the transcriptional activity of β-catenin.

Discussion
Our goal was to identify binding partners of Mfn2 and potential non-mitochondrial roles of Mfn2 in the endothelium. The non-mitochondrial roles for Mfn2 we identified in the Fig. 5 TNFα induced ROS production mediates the interaction of Mfn2 and β-catenin. a HLMVECs were stimulated with TNFα (10 ng/mL) for indicated durations (0, 15 min, 3 h, and 6 h). TNFα-induced total ROS production was measured using a DCFH-DA probe which detects intracellular H 2 O 2 . The ROS images in upper panel were taken with confocal microscope (Zeiss LSM880, Plan 1.45NA, ×63 magnification). The relative fluorescence of DCF was quantified with ImageJ (lower panel) and presented fold change to basal. Data are mean values ± SD from n = 4. b Left panel, HLMVECs were stimulated with H 2 O 2 (500 µM) for 30 min and immunoprecipitated with Mfn2 antibody followed by Western blotting with β-catenin antibody under reducing conditions (with β-mercaptoethanol, ME). Mouse normal IgG was used as a negative control for immunoprecipitation. Uncropped blots are provided in the Source Data file. Right panel, the quantification of band intensities for Mfn2 and β-catenin interactions with ImageJ. Data are mean values ± SEM for three independent experiments. *p = 0.050 by unpaired, one-tailed t-test. c HLMVECs were pretreated with 20 mM NAC for 30 min. The cells were stimulated with H 2 O 2 (500 µM) for 30 min and immunoprecipitated with Mfn2 antibody followed by Western blotting with β-catenin antibody. Uncropped blots are in the Source Data file. Mfn2 is typically found in the outer mitochondrial membrane and its structural motifs contain a cytosolic N-terminal GTPase domain, a proline-rich region (PR), two coiled-coil heptad-repeat domains (HR1 and HR2), and a transmembrane domain (TM) which allows Mfn2 anchorage in the outer mitochondrial membrane 34 . Mfn1 and Mfn2 are key mediators of mitochondrial fusion but only Mfn2 has a PR domain, thought to be responsible for specific protein-protein interactions 14 suggesting the possibility for additional roles beyond mitochondrial fusion for Mfn2. In line with this, it has been recently reported that the mitochondria-ER-cortex anchor (MECA) interacts directly with mitochondria and the plasma membrane via core protein component, Num1 in budding yeast 35 . Moreover, it has been appreciated that many organelles communicate by using molecular tethers 36 and the function of the mitochondria-plasma membrane contact extends beyond the mitochondrion itself 35,37 .
Our proteomic analysis revealed unexpected Mfn2 binding partners located outside of the mitochondria such as cell junction proteins. We took advantage of super-resolution microscopy which resolves individual protein complexes 18,19 to validate the extra-mitochondrial localization of Mfn2. Super-resolution imaging identified the expected mitochondrial localization of the bulk of Mfn2 but also clearly visualized Mfn2 localization at the plasma membrane. Biochemical immunoprecipitation assays identified the plasma membrane binding partners of Mfn2 which included the adherens junction proteins VE-cadherin and βcatenin in homeostatic ECs. Loss of Mfn2 resulted in the disruption of the endothelial barrier, thus indicating that Mfn2 is not only localized at the junctions but also plays a functional role by stabilizing barrier integrity. We also found that Mfn2-depleted ECs showed an impaired F-actin structure which is important for cell-cell interaction. Importantly, during inflammatory stimulation, Mfn2 was disassociated from the junctions and this likely contributes to the disruption of barrier integrity that is typically observed during inflammation 3,38 . It is plausible that Mfn2 may play a role as an anchor between AJs proteins at plasma membrane and F-actin at cytosol to maintain EC barrier integrity in homeostatic ECs. We also did not find any broad effects on cell stress, cell death or cell proliferation following Mfn2 deletion, which suggested that our observations were most likely due to a specific role for Mfn2 in AJs and endothelial barrier integrity that is independent of generalized cellular health functions of Mfn2.
We next investigated the mechanism by which Mfn2 disassociated from the adherens junctions during inflammation. Post-translational modifications are regulatory switches which modify the activity of proteins, and oxidation is one of the most frequently occurring post-translational modification 39 . Especially during inflammation, ROS levels can acutely increase and result in oxidative modification of proteins 6 . Cysteine (Cys) is an amino acid that is susceptible to several types of oxidative posttranslational modification including sulfenylation, disulfide formation, S-glutathionylation, and S-nitrosylation 9,40 . Oxidative modifications of proteins are critical mediators of compartmentalized ROS signaling 41 . Recently, it has been shown that increasing concentrations of xanthine oxidase, a cytosolic source of ROS inhibits Mfn2 activity by inducing disulfide linked oligomerization via oxidation of C-terminal Cys 684, 700 residues in vitro 42,43 . However, it is unknown whether redox-dependent modifications of Mfn2 could affect its non-mitochondrial roles as a stabilizer of AJ protein complexes at the plasma membrane in ECs. We found that TNFα-induced ROS increased Mfn2 sulfenylation in ECs. Our results suggest that the sulfenylation step might be required for the disassociation of Mfn2 from adherens junctions and that Mfn2 sulfenylation may constitute a form of post-translational regulation of Mfn2 activity during inflammation. Future studies could identify specific cysteine residues that serve as sulfenylation targets and whether such sulfenylation would also impact other aspects of Mfn2 function such as its GTPase activity and mitochondrial fusion.
Interestingly, we also found Mfn2 presence in the nucleus during inflammatory activation. This surprising nuclear localization of Mfn2 was independently confirmed by several different approaches including biochemical subcellular fractionation, immunofluorescence with 3D analysis, and proximity ligation assays. We also confirmed TNFα-induced accumulation of Mfn2 using exogenously overexpressed GFP-Mfn2. However, there are important questions regarding the mechanisms of Mfn2 nuclear accumulation during inflammation that still need to be addressed in future studies. Recent work suggests that several mitochondrial enzymes can translocate into the nucleus and constitute a form of mitochondria-to-nucleus communication 29 and that mitochondria-derived vesicles or chaperone proteins and nuclear transcription factors may promote the entry of selected mitochondrial proteins into the nucleus via the nuclear pores 33,44 . It is possible that the Mfn2 interaction with β-catenin or other proteins may facilitate the entry via nuclear pores during inflammation but this will need to be addressed in future studies targeting nuclear transport mechanisms. Although Mfn2 lacks a nuclear localization sequence (NLS), we found that its binding to the adherens junction protein β-catenin, which is known to translocate to the nucleus where it acts as a transcriptional coregulator 30 , was increased following inflammatory activation. Sequence alignment analysis indicated that Mfn2 has a putative β-catenin binding motif such as "SxxSSLSxLS" or "DxθθxΦx 2-7 E" Fig. 6 Mfn2 suppresses β-catenin transcriptional activity which requires TNFα induced pro-inflammatory gene expression. a, b Control and Mfn2-KD ECs were stimulated with or without TNFα (10 ng/mL) for 6 h and total RNA was extracted. a The knockdown efficiency of Mfn2 mRNA was determined under basal and TNFα stimulation by qRT-PCR with its specific primers. Data are mean values ± SEM for n = 6 independent biological replicates. ****p < 0.0001 for Mfn2 in basal control vs Mfn2-KD. b The mRNA levels of pro-inflammatory genes, ICAM-1, IL-6, and IL-18 were determined by qRT-PCR with their specific primers. ***p = 0.0001 for ICAM-1 of TNFα 6 h in control vs Mfn2-KD, *p = 0.016 for IL-6 of TNFα 6 h in control vs Mfn2-KD, ****p < 0.0001 for IL-18 in basal control vs Mfn2-KD by unpaired, two-tailed t-test. Data are mean values ± SEM for n = 4-6 independent biological replicates. c Control and Mfn2-KD ECs were transfected with 1 µg Topflash (β-catenin reporter containing the TCF promoter) and 35 ng of pRL/TK for 48 h. The cells were stimulated with or without TNFα (10 ng/mL) for 6 h. Firefly and renilla-luciferase activity were determined by the dual luciferase reagent assay system and the firefly luciferase activity was normalized by renilla-luciferase activity. Transcriptional activity of β-catenin was presented for fold change by further normalizing with value of control basal (TNFα 0 h). Data are mean values ± SEM for n = 3 independent experiments. ***p = 0.0008 for control in TNFα 0 h vs 6 h, *p = 0.0256 for basal in control vs Mfn2-KD, *p = 0.0394 for TNFα 6 h in control vs Mfn2-KD by paired, two-tailed t-test. motifs (θ and Φ are hydrophobic and aromatic residues, respectively) between its heptad repeat (HR) domains 45 .
The nuclear translocation of β-catenin is essential for canonical Wnt signaling pathway because β-catenin binds DNA and acts as a transcriptional regulator during embryonic development and adult tissue homeostasis 30 . It has been shown that β-catenin translocates into the nucleus despite the lack of an NLS although the underlying mechanisms are not well understood 30 . It is possible that the nuclear accumulation of Mfn2 utilizes a similar mechanism because we found that Mfn2 remains bound to βcatenin when they both disassociate from adherens junctions. This raised the intriguing question whether Mfn2 could functionally regulate gene expression, possibly by affecting the transcriptional activity of β-catenin. We found that nuclear Mfn2 indeed inhibits β-catenin transcriptional activity for key proinflammatory cytokines such as IL-6 or IL-18 as well as an adhesion molecule ICAM-1 during inflammation. Interestingly, β-catenin depleted ECs demonstrated significant decreases in the gene expression of selected pro-inflammatory genes such as ICAM-1, IL-1β, or IL-18, but β-catenin depletion did not impact other pro-inflammatory genes such as IL-6, caspase1, or VCAM-1, indicating a specificity of the inflammatory genes regulated by βcatenin. A non-canonical role for transcriptional regulators in endothelial barrier regulation has also recently been reported for Notch signaling 46 . Possible future studies could address whether Mfn2 also modulates the role of Notch in regulating the endothelial barrier.
Although mitochondria contain an independent genome, the vast majority of mitochondrial proteins are encoded in the nuclear genome 32 . However, less is known about how mitochondria engage in retrograde communication with the nucleus and could potentially modulate nuclear gene expression. Even though it came as a surprise that Mfn2 accumulated in the nucleus, recent studies are increasingly finding mitochondrial proteins in nuclei. Multiple proteins that were previously thought to be exclusively mitochondrial have recently been shown to also localize within nuclei, yet their nuclear function is not fully understood 32,33,44 .
Importantly, we demonstrate here that Mfn2 and β-catenin are both necessary for EC barrier integrity in vitro and in vivo. Genetic in vivo deletion of Mfn2 specifically in the endothelium resulted in a disruption of vascular homeostasis, as evidenced by increased lung vascular permeability and an increase in inflammatory gene expression along with increasing inflammatory immune cells even in the absence of any additional pathogenic stimuli. The lung endothelium may be especially vulnerable to inflammatory activation as a recent comparative analysis of vascular endothelial cells in distinct organs found that the lung endothelium had the highest expression of inflammatory and immune response genes even during homeostasis when compared to the endothelium of other organs such as the brain or the heart 47 . The association between Mfn2 and vascular permeability may thus represent promising therapeutic targets, to reduce the influx of inflammatory cells 3,48 as well as to reduce tumor metastasis or tumor growth 49 . A recent study also identified an important link between angiogenesis and endothelial barrier maturation by demonstrating that the transcriptional regulator YAP/TAZ was involved in both processes 41 . Our study established the barrier stabilization role for Mfn2 in the endothelium, but it is possible that Mfn2 also may regulate angiogenesis.
In summary, our findings reveal intriguing non-mitochondrial roles for Mfn2 as a stabilizer of the vascular barrier and an endogenous suppressor of inflammatory gene expression. These insights could help us better understand the role of "mitochondrial" proteins in the nucleus and pave the way for future antiinflammatory therapies.
Lentivirus production and purification. HEK293T (CRL-11268, ATCC) cells were used to produce lentivirus and the cells were transfected with DNAs (2.5 µg pMD2. G, 5 µg of psPAX2, and 7.5 µg of DNA expression vector) with 30 µg polyethylenimine (PEI, Polysciences, 23966, USA) in DMEM media containing 10% FBS without antibiotics overnight and changed with fresh DMEM media supplemented with 10% FBS and 1% Pen/Strep and then incubated for 48 h at 37°C. The media containing secreted lentivirus was collected and the virus was purified using Lenti-X concentrator (CloneTech, 631232).  TNFα triggers translocation of Mfn2 into the nucleus. HLMVECs were stimulated with TNFα (10 ng/mL) for the indicated times (0, 1, 3, or 6 h). a The cellular localization of Mfn2 was examined by a subcellular fractionation assay. The same amount of cytosolic and nuclear fraction was loaded in the SDS-PAGE gel. GAPDH and p84 were used as controls for cytosolic or nuclear fraction, respectively. The images are representative of at least 3 independent experiments. A red asterisk (*) indicates the nuclei translocated Mfn2. Uncropped blots are provided in the Source Data file. b Quantification of Mfn2 protein levels in (a) with ImageJ. Data are mean values ± SEM for 3-5 independent experiments. **p = 0.0054 for TNFα 0 h vs 6 h in nucleus by paired, two-tailed t-test. c HLMVECs expressing GFP-Mfn2 were stimulated with TNFα (10 ng/mL) for 6 h and fixed with 100% cold (−20°C) methanol on ice for 5 min and permeabilized with 0.25% Triton X-100 for 10 min. Cellular localization of exogenous GFP-Mfn2 was determined by GFP and endogenous Mfn2 was examined by Mfn2 immunostaining. DAPI (blue) indicates nuclei. The images were obtained by Z sections (>87 sections) with x2.4 zoom using confocal microscopy (Zeiss LSM770, Plan Apo 1.46NA, ×63 magnification) and represented by one section. The white boxes are the enlarged nucleus and white lines in merged images indicate the area where the distances between GFP-Mfn2, Mfn2, and DAPI were analyzed using ImageJ (d). e 3D image analysis with Imaris software in (c) and ortho analysis to show co-localization of GFP-Mfn2, Mfn2, and DAPI. The images are representative of at least three independent experiments. f HLMVECs stimulated with TNFα (10 ng/mL) were fixed with the same method in (c), and β-catenin cellular localization was examined by β-catenin immunostaining and the images were taken using confocal microscopy (Zeiss LSM880, Plan 1.46NA, ×63 magnification). g The fluorescence intensity of β-catenin in nuclei in immunostained images of (f) was quantified using ImageJ. Data are presented with box and whiskers plot and whiskers are Min to Max. n = 27-31 biologically independent cells. *p = 0.0155 by unpaired, two-tailed t-test. h Interaction of endogenous Mfn2 and β-catenin in HLMVECs was examined with a mouse Mfn2 antibody and rabbit β-catenin antibody with or without TNFα (10 ng/mL) for 6 h by proximity ligation assay. Quantification of number of interactions within the nucleus per cell was determined by counting the number of dots by interacting of Mfn2 and β-catenin. Data are presented with box and whiskers plot and whiskers are Min to Max. n = 8-13 biologically independent samples. ****p < 0.0001 by unpaired, two-tailed t-test.
Transendothelial resistance (TER). 40,000 cells of control or Mfn2-KD ECs were plated on a gold microelectrode (8W1E + PET, ECIS Cultureware) for TER 1 day before measurement. 250 µL fresh media was added to the cells and the resistance of the EC barrier was measured with an ECIS ® -1600 R system (Applied Biophysics, Troy, NY, USA) in accordance with the manufacturer's instructions 22 .
Evaluation for total ROS. HLMVECs were stimulated with TNFα (10 ng/mL) for the indicated time (0, 15 min, 30 min, 3 h or 6 h) and then incubated with 20 µM CM-H2DCFDA (Invitrogen, C6827) for 6 min at 37°C 41 . DCF fluorescence was measured by confocal microscopy (Zeiss LSM880) with ×63 magnification using the same exposure conditions in each experiment. Relative DCF fluorescence was measured by ImageJ and presented as fold change.
DCP-Bio1 assay. To determine sulfenic acid (Cys-OH) of target protein, we used an innovative cell permeable biotin-labeled Cys-OH trapping probe (DCP-Bio1) 29 . Briefly, TNFα stimulated-control or Mfn2-KD ECs were lysed with degassedspecific lysis buffer (50 mM HEPES, pH7.0, 5 mM EDTA, 50 mM NaCl, 50 mM NaF, 1 mM Na 3 VO 4 , 10 mM sodium pyrophosphate, 5 mM IAA, 100 μM DPTA, 1% Triton-X-100, protease inhibitor, 50 unit catalase, 200 μM DCP-Bio1), and then pull-downed with streptavidin beads overnight. All steps were performed in the dark. DCP-Bio1 conjugated sulfenylated-proteins were measured with specific antibodies (1:1000 dilution) for Mfn2, β-catenin, or actin using Western blotting. Subcellular fractionation. Confluent HLMVECs in 100 mm dishes were stimulated with TNFα for the indicated times (0, 1, 3, 6 h) followed by subcellular fractionation 51 . Briefly, the cells were lysed with 500 µL lysis buffer (10 mM HEPES, pH7.9, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, and EDTA free protease inhibitor cocktail), collected in Eppendorf tubes, and incubated in ice for 10 min. The lysate was freshly added to 60 µL 10% NP-40 and then vortexed for 10 s followed by centrifugation at 16,000g for 1 min at 4°C. The supernatant (cytosolic fraction) was carefully transferred to a second tube. The pellet was washed once with lysis buffer and 200 µL nuclear extract buffer (25 mM HEPES, pH7.9, 0.4 M NaCl, 0.5 mM EDTA, 0.5 mM EGTA, and EDTA free protease inhibitor cocktail) was added by tapping gently and incubated in ice for 30 min with vortexing every 5 min. The nuclear fraction was collected by centrifugation at 16,000g for 5 min at 4°C . The same amount of cytosol and nuclear fraction was loaded in a SDS-PAGE gel followed by Western blotting. GAPDH (1;1000 dil, 10494-1-AP, rabbit) and p84 (nuclear matrix protein, 1:1000dil, sc-514123, mouse) were used as controls for the cytosolic or nuclear fraction, respectively. β-catenin reporter luciferase assay. Control and Mfn2-KD ECs were transfected with 1 µg of a β-catenin reporter (M50 super 8x TOPflash containing TCF/LEF sites upstream of a luciferase reporter) 52 and 35 ng of pRL/TK using PEI transfection reagent (Polyethylenimine). At 48 h after transfection, the cells were stimulated with TNFα (10 ng/mL) for 6 h and then 100 µL of cell lysate from each sample was used to measure reporter gene expression. Firefly and Renilla luciferase activity were determined by the dual luciferase reagent assay system (Promega). The relative luciferase activity represents the mean value of the firefly/Renilla luciferase.
Quantitative real-time PCR. Total RNA was isolated by using phenol/chloroform and TriZol Reagent (Invitrogen, 15596026) as described 41 . Reverse transcription was carried out using high capacity cDNA reverse transcription kit (Applied Biosystems, 4368814) using 2 µg of total RNA. Quantitative PCR was performed with fast start universal SYBR Green master (ROX) PCR kit (Roche, 04913914001) using QuantStudio7 (Thermofisher). Samples were all run in triplicates to reduce variability. Expression of human genes of Mfn2, ICAM-1, IL-6, IL-18, or IL-1β was determined using the following primers; Mfn2, sense 5′-CATCCCCAGTTGTCCT