GBF1 and Arf1 interact with Miro and regulate mitochondrial positioning within cells

The spatial organization of cells depends on coordination between cytoskeletal systems and intracellular organelles. The Arf1 small G protein and its activator GBF1 are important regulators of Golgi organization, maintaining its morphology and function. Here we show that GBF1 and its substrate Arf1 regulate the spatial organization of mitochondria in a microtubule-dependent manner. Miro is a mitochondrial membrane protein that interacts through adaptors with microtubule motor proteins such as cytoplasmic dynein, the major microtubule minus end directed motor. We demonstrate a physical interaction between GBF1 and Miro, and also between the active GTP-bound form of Arf1 and Miro. Inhibition of GBF1, inhibition of Arf1 activation, or overexpression of Miro, caused a collapse of the mitochondrial network towards the centrosome. The change in mitochondrial morphology upon GBF1 inhibition was due to a two-fold increase in the time engaged in retrograde movement compared to control conditions. Electron tomography revealed that GBF1 inhibition also resulted in larger mitochondria with more complex morphology. Miro silencing or drug inhibition of cytoplasmic dynein activity blocked the GBF1-dependent repositioning of mitochondria. Our results show that blocking GBF1 function promotes dynein- and Miro-dependent retrograde mitochondrial transport along microtubules towards the microtubule-organizing center, where they form an interconnected network.


GBF1 inhibition enhances retrograde movement of mitochondria.
To determine whether mitochondrial condensation upon GBF1 inhibition was the result of enhanced retrograde trafficking to the cell center, or decreased anterograde transport to the cell periphery, movements of the mitochondrial network after GCA treatment were measured and quantified from live cell imaging (Supplementary Movies S2 and S3). Because mitochondria form a network, patches of the network were identified and tracked in control and GCA-treated cells for 30 min (Fig. 2A). The movement of mitochondria was not monotonous in either condition. It consisted of alternative displacements in different directions. Mitochondria displayed retrograde movements ( Fig. 2A green traces), towards a central position (large white spot), anterograde movements ( Fig. 2A red traces), and movements perpendicular to the antero-retrograde axis ( Fig. 2A yellow traces), until their final location (small white spots). Consistent with the clustering of mitochondria, retrograde movements were dominant in GCA-treated cells ( Fig. 2A, right panel), but not in control cells ( Fig. 2A, left panel). These movements were quantified (Fig. 2B). Quite strikingly, in control cells, the duration of retrograde and anterograde movements of mitochondrial network patches was extremely similar, with no statistically significant difference observed (Fig. 2B top). As a result, the net movement was close to zero ( Fig. 2B bottom panel). In GCA-treated cells, the duration of retrograde displacements was increased by two fold, whereas anterograde or perpendicular movements were identical to control cells. This resulted in a significant net retrograde movement (Fig. 2B, bottom). In a given condition (control or GCA-treated), the speed of movements was the same in any direction, slightly, but not significantly, slower in GCA-trated cells (~0.31 µm/s) than in control cells (~0.43 µm/s). Speed was therefore without effect on the net movement. Our results strongly support the hypothesis that GBF1 inhibition markedly enhances the duration of retrograde movements of mitochondria, without affecting other movements or speed. with the dominant negative mutant of Arf1, Arf1-T31N-GFP, which traps GBF1 in an inactive complex, and disrupts Golgi structure 31 . One day post-transfection, Arf1-T31N-GFP expressing cells exhibited mitochondrial clustering in a juxta-nuclear region ( Fig. 3A top panels, B left panel). In contrast, transfection of the constitutively active form of Arf1, Arf1-Q71L-GFP, which causes an enlargement of the Golgi network, had no observable effect Figure 1. The inhibition of GBF1 induces the condensation of the mitochondrial network in the pericentriolar region without affecting cell shape. RPE1 cells were incubated with either GCA (10 μM) or DMSO alone (CT) for 1h30, were loaded with MitoTracker Orange for an additional 30 min prior to fixation and imaged using confocal microscopy. Images are maximum intensity Z-projections. All bars = 10 µm. Nuclei appear in blue (DAPI staining) and mitochondrial network in white or red. (A) Cells were stained for α-tubulin after fixation (green signal). (B) RPE1 cells were transfected with the ER marker Sec61β-GFP. (C) RPE1 cells were stained for γ−tubulin after fixation (green signal). In cells treated with GCA, the mitochondrial network is frequently condensed in the juxta-nuclear region next to the centrosome (white arrows) but not in control cells (green arrows). (D) The mitochondria area was segmented using ImageJ software.  3A lower panels, B right panel). The effect of the inactive Arf1 mutant could either be due to GBF1 cytosolic trapping by Arf1-T31N-GFP, or to GBF1 activation of Arf1. Altogether, these results indicate that GBF1 activity is required to maintain the distribution of the mitochondrial network throughout the cell. Lowering GBF1 expression in RPE1 cells by transfection with GBF1 siRNA duplexes also caused the condensation of the mitochondria into a juxta-nuclear area, around the remaining Golgi structure, where low levels of GBF1 staining were still visible ( Supplementary Fig. S1). However, in this situation, the morphology of mitochondria appeared to be somewhat different than that observed in BFA-or GCA-treated cells. In approximately 80% of GBF1 siRNA-treated cells ( Supplementary Fig. S1), mitochondria formed condensed structures around the Golgi remnant, but these condensed structures were more dispersed than in drug-treated cells.

Mitochondrial condensation caused by mutant
To verify that the mitochondrial phenotype observed upon GBF1 siRNA treatment was not due to off-target effects, we performed a rescue experiment using a heterologous GBF1 cDNA (from hamster). Unfortunately, the double transfection was toxic to RPE1 cells. Hence we turned to HeLa cells. First, we verified that the phenotype of GCA-treated HeLa cells was indistinguishable from that of GCA-treated RPE1 cells ( Supplementary Fig. S2). Second, HeLa cells treated with GBF1 siRNA had a condensed mitochondrial phenotype (Fig. 4A), and quantifications indicated a decrease of mitochondrial area in siRNA-treated compared to control cells (Fig. 4B). HeLa cells remained viable upon double transfection with GBF1 siRNA and the rescue plasmid from hamster. We demonstrated that hamster GBF1 expression was resistant to our GBF1 siRNA treatment (Fig. 4C), and that hamster GBF1 cDNA restored normal mitochondrial morphology (Fig. 4A,B). Together, these results show that GBF1 inhibition by either drug or GBF1 siRNA treatment causes a condensation of the mitochondrial network in cells.
In HeLa cells, as in RPE1 cells, the mitochondrial network is less condensed upon GBF1 siRNA treatment compared to drug treatments. This difference may be due to different modes of GBF1 inhibition, i.e. rapid stabilization of an Arf1-GDP-GBF1 complex for BFA and GCA, in contrast to long-term, but incomplete, reduction in the overall level of GBF1 (Figs 4C and S1), and the presence of Golgi remnants, after siRNA treatment ( Supplementary Fig. S2). Because mitochondrial condensation is observed in siRNA-treated RPE1 and HeLa cells, yet many Golgi elements remain at least partially intact ( Supplementary Fig. S2), we can conclude that the mitochondrial phenotype is not simply a secondary effect of Golgi disassembly.
Identification of Miro as a GBF1 binding partner. Using a tandem affinity purification/mass spectrometry strategy in HEK293 cells, we have identified Miro2 as one putative GBF1 binding partner ( Supplementary  Fig. S3). To verify the interaction of GBF1 with Miro1 and Miro2 in RPE1 cells, we performed pull-down experiments in cells expressing Venus-GBF1 and either myc-Miro1 ( Fig. 5A) or myc-Miro2 (Fig. 5B). The results show that a small fraction of both Myc-Miro1 and Myc-Miro2 co-immunoprecipitate with Venus-GBF1 (Figs 5A,B and S3), suggesting that GBF1 and Miro are physically associated. The relatively small level of interaction is consistent with GBF1 localizing only transiently to membranes, cycling rapidly on and off the cytosolic surfaces of organelles, as has been demonstrated for the Golgi 30 . Because Miro2 appeared to be the major Miro isoform expressed in RPE1 cells ( Supplementary Fig. S4), we will focus exclusively on Miro2 in the following experiments. In order to evaluate if Arf1 also interacts with Miro2, cells were co-transfected with Myc-Miro2 and either inactive Arf1-T31N-GFP or constitutively active Arf1-Q71L-GFP, and the Arf1 mutants were immunoprecipitated using GFP Trap. A small pool of Myc-Miro2 was immunoprecipitated with Arf1-Q71L-GFP (Fig. 5C), but not with Arf1-T31N-GFP (Fig. 5C), indicating a selective interaction between Miro2 and the active form of Arf1. These results show that Miro2 is either an effector of Arf1, or is part of an Arf1 effector complex, that interacts with GBF1 at the mitochondrial membrane.  It is well established, particularly in neuronal systems, that Miro1 and Miro2 regulate mitochondrial trafficking and positioning along microtubules 12,20,32 . To assess if Miro2 may contribute to GBF1-dependent mitochondrial network remodeling, we first expressed Myc-Miro2 in RPE1 cells. In these Myc-Miro2 transfected cells, mitochondria were condensed near the MTOC, as visualized by γ-tubulin staining (Fig. 6A), which has been reported previously 24 . This result suggests that Miro2 overexpression alone results in the same phenotype as BFA/GCA treatment (Fig. 6A). More importantly, lowering Miro2 expression by transfection with specific siRNA duplexes abolished BFA-and GCA-induced mitochondrial condensation to the pericentrosomal region (Fig. 6B). Miro2 siRNA duplexes had no effect on mitochondrial network positioning in untreated cells (Figs 6C and S5) or on Miro1 expression (Fig. 6B). Finally, transfection of Miro2 siRNA duplexes together with GBF1 siRNA prevented the effect of GBF1 siRNA treatment on the spatial organization of the mitochondrial network (Figs 6C and S5). Together, our results suggest that Miro interacts physically with GBF1 and mediates the effects of GBF1 on intracellular positioning of mitochondria. Miro-dependent trafficking of mitochondria is dependent on the interaction between Miro and the microtubule motor proteins kinesin and dynein 20,21 . To assess if the retrograde transport of the mitochondrial network upon BFA/GCA treatment in RPE1 cells is dependent on microtubule minus end directed dynein transport, we studied the effect of BFA and GCA in cells pretreated with ciliobrevin D, a cytoplasmic dynein inhibitor. As shown in Fig. 7, ciliobrevin D pretreatment abolished the GCA-induced condensation of mitochondria towards the centrosomal region, whereas ciliobrevin D alone had no detectable effect on the mitochondrial network. Similar results were obtained in cells treated with BFA (not shown). Together with the mitochondrial tracking experiments (Fig. 2), these results support the conclusion that mitochondrial retrograde movement upon GBF1 inhibition is mediated by dynein along microtubules. Mitochondrial morphology is altered by GBF1 inhibition. In order to determine if the relocalization of mitochondria upon GBF1 inhibition also induced a change in mitochondrial morphology, the ultrastructure of mitochondria was studied by electron microscopy in GCA-treated and control RPE1 cells by electron tomography on semi-thin sections. Mitochondria and microtubules were segmented in 3D (Fig. 8). In control cells, mitochondria were relatively circular and dispersed. In GCA-treated cells, mitochondria were consistently condensed in a juxta-nuclear region and exhibited a more complex morphology (Fig. 8A). A quantitative analysis after segmentation of mitochondria and microtubules in tomograms from several cells (4 control cells, 5 GCA-treated cells) showed significant alterations of mitochondrial morphology and spatial organization after GCA treatment (Fig. 8B). In GCA-treated cells, mitochondria were bigger, (75% volume increase, p < 0.05 Fig. 8B left graph) and exhibited a more complex morphology as shown by a 20% increase in the surface area of outer membrane per micron of mitochondrion length (p < 0.05 Fig. 8B, center graph). However, there was no significant difference in the dimension of mitochondria (longest length measured). These results demonstrate an increase in mitochondria connectivity, either through packing of long mitochondria into a smaller volume, or through fusion of smaller mitochondria at the site of clustering. In both control cells and GCA-treated cells, the average orientation of mitochondria was parallel to the average direction of microtubules (not shown), in support of microtubule dependent orientation of mitochondria. However, the angular spread of the mitochondrial long axis with respect to the microtubule network was smaller in GCA treated cells than in control cells (average angular ratio of 0.85 in GCA cells, and 1.10 in control cells, Fig. 8B, right graph, p < 0.05), indicating a better alignment of mitochondria with microtubules in drug-treated cells. ER-mitochondrial contact sites are believed to play a role in mitochondrial fission processes and in their metabolism. Despite the change in mitochondrial position and morphology, we did not find changes in the frequency or the morphology of ER-mitochondria contact sites by electron tomography in GCA-treated RPE1 cells (not shown). These observations suggest that upon inhibition of GBF1, mitochondria are transported along microtubules towards the MTOC, where they form a hyper-connected network.

Discussion
To date, the best-studied function of GBF1 has been in the secretory pathway, where Arf1 activation by GBF1 regulates retrograde vesicular transport from the cis-Golgi and the ERGIC to the ER. This function of GBF1 is important to maintain Golgi structure and secretory function, and is consistent with the major steady-state localization of GBF1 at the cis-Golgi. Although GBF1 is primarily found at the cis-Golgi, it also localizes to other organelles such as the trans-Golgi and lipid droplets 33,34 .
In yeast and C. elegans, Arf1 and GBF1/Gea were also shown to be required for mitochondrial morphology, through genetic interactions with Miro/Gem1 15 . Our data reveal that in human cells, a pool of GBF1, and active Arf1-GTP, physically interact with Miro, and we show that inhibition of GBF1 function causes condensation of mitochondria near the MTOC, where they become more connected. Strikingly, inhibition of GBF1 results in a doubling of the time mitochondria engage in retrograde-directed movement, without changing the amount of time spent in movements in other directions. We demonstrate that the movement of mitochondria in the absence of GBF1 function is dependent on the minus end-directed motor dynein and on Miro2 in RPE1 cells. Therefore, our data support the conclusion that GBF1 regulates dynein-mediated retrograde transport of mitochondria along microtubules towards the cell center, with an increase in retrograde movement occurring when GBF1 is inhibited (Fig. 9). We also demonstrate a physical association between the Miro2 dynein adaptor complex and either GBF1 or constitutively active Arf1. However, our results do not demonstrate a direct interaction, or whether this interaction is directly responsible for the movements of mitochondria. In the future, it will be interesting to determine how these physical interactions might be involved in GBF1/Arf1 regulation of retrograde mitochondrial movement. Much is known about Miro regulation of kinesin motors, but recent studies have shown that Miro also interacts with dynein through TRAK/Milton in neurons 35 , and through CenpF in dividing cells 36 . The yeast Miro orthologue Gem1 is a regulator of endoplasmic reticulum-mitochondria contacts and is also involved in mitochondrial fission 37,38 . There is also evidence that Miro plays a role at ER-mitochondria contact sites in mammalian cells (Kornmann, et al., 2011;Lee, et al., 2016), and may play a role in mitochondrial fission in Drosophila 39 . We have not observed a change in overall ER distribution, or in ER-mitochondria contacts after mitochondria repositioning induced by GBF1 inhibition. However, we found an increase in mitochondrial volume and network complexity, consistent with increased fusion upon GBF1 inhibition. In any case, the role of Miro and Gem1 at ER-mitochondria membrane contact sites and their regulation by Arf1 and its GBF/Gea GEFs are evolutionarily conserved features, and provide a new perspective to explore the links between mitochondrial positioning and fission/fusion processes mediated by ER-mitochondrial membrane contact.
Although GBF1 itself has not yet been shown to regulate microtubule motor activity at other organelles, active Arf1 has been implicated in dynein-mediated movement of the early Golgi 40 . However, in contrast to our results at mitochondria, Arf1-GTP has been shown to promote the centrosomal localization of the Golgi, through recruitment of the dynein receptor golgin160 to Golgi membranes 40 . Taken together, these results lead to the prediction that inhibition of Arf1 activation would cause opposite movement of mitochondria and Golgi elements along microtubules, with mitochondria becoming more concentrated at the minus ends and the Golgi becoming more dispersed due to net plus end movement. It is interesting to note that during mitosis, when both the mitochondrial network and the Golgi apparatus undergo dramatic structural changes 41 , GBF1 activity is inhibited through phosphorylation by AMPK 42 . Therefore, given that GBF1 has a known role in mitosis 43 , another attractive hypothesis to investigate is whether GBF1 might coordinate the reorganization of the Golgi apparatus and the mitochondrial network during cell division.

Methods
Plasmids and primary antibodies. All constructs were confirmed by DNA sequencing. pTKN101 (pVenus-GBF1) has been described previously 30  Drugs. Brefeldin A (BFA), the primary targets of which are the high molecular weight Arf1 GEFs GBF1, BIG1 and BIG2, and the highly specific GBF1 inhibitor, golgicide A (GCA), were obtained from Sigma-Aldrich and used as described previously 15,27,29 . The cytoplasmic dynein inhibitor, ciliobrevin D, was purchased from EMD Millipore (Billerica, MA, USA). Stock solutions were prepared in DMSO. Immunoprecipitation. Cells cultured in 6-well plates were disrupted with 300 μl of cold lysis buffer (50 mM Tris-HCl (pH 7.5), 100 mM NaCl, 1 mM EDTA, 0.5% NP40) supplemented with a protease inhibitor mix (GE Healthcare). Lysates were preclarified by centrifugation (14,000 g, 4 °C, 15 min). Immunoprecipitations were performed using purified GFP Binding Protein (GBP) 46 bound covalently to AminoLink resin, according to the manufacturer's instructions (Thermo Scientific Pierce), after pre-clearing with AminoLink resin alone. Planegg, Germany). After extensive washing with cold W100 buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1 mM EDTA), immunoprecipitates were finally eluted from the beads by boiling for 5 min in Laemmli sample buffer, then were submitted to Western blot analysis.

Cell culture and transient transfection.
Western Blotting. Samples in Laemmli's buffer were separated by SDS-PAGE. Proteins were then transferred onto nitrocellulose blotting membranes (GE Healthcare). Membranes were blocked with TBS-Tween containing 5% skim milk for 1 h and then incubated 2h30 with primary antibody (anti-GFP, 0.4 μg/ml; anti-c-myc, 1 μg/ml; anti-Miro1, 1 μg/ml and anti-Miro2, 0.8 μg/ml) at 4 °C with stirring. After washing with TBS-tween, membranes were incubated with the suitable secondary antibody coupled to peroxidase (GE Healthcare, Cat. No NA934V, NA931V), and immune complexes were revealed by enhanced chemiluminescence (ECL Select Western Blotting Detection Reagent, GE Healthcare) and visualized using the LAS-3000 Imaging System from Fujifilm (Tokyo, Japan). Mitochondrial network area and length quantification. To measure the surface area of mitochondria on maximum intensity Z-projections, an image processing algorithm was developed in ImageJ to quantify the ratio between the total fluorescence intensity of mitochondria and the mitochondrial area for each cell. All the cells were processed with the same parameters. Total fluorescence intensity of mitochondria was calculated by thresholding the image with the default ImageJ thresholding tool, and the total mitochondrial area contour was determined by thresholding the low pass filtered image. The result was obtained by computing the ratio between the two values. To measure the length of the network, another approach was also used in ImageJ. The mitochondrial area contour was determined as above, and the longest dimension of the network was measured and divided by the perimeter of the nucleus for normalization.

Mitochondrial network tracking.
Live cell imaging of mitochondria was performed on a spinning disk microscope after labeling cells with mitotracker. Cells were treated with DMSO alone or GCA in DMSO and images were recorded at 30 s intervals for 30 min at a resolution of 10 pixels per micron. An image processing algorithm was developed for tracking of the mitochondrial network. For each time point, the average intensity projection of 3 confocal slices was calculated. Cells of interest were outlined manually and drift correction was applied in ImageJ on the whole cell. Then, segmentation of the cell mitochondrial network was performed for each time-point of the movie by background substraction of median-filtered images. An automatic threshold was applied to generate a mask for the network and this mask was applied to the grey level images. Then for each cell, the first image of the movie was cut into a mosaic of 8 × 8 pixels to identify mitochondria-positive ROIs. Each ROI was then expanded to a 64 × 64 pixel patch, corresponding to a small fragment of the mitochondrial network. Each of the patches was tracked across the movie by cross correlation: the new position of the positive patch was identified in the next image and a new reference image, taking into account changes in the mitochondrial network shape, was taken for correlation with the next image. Then, the center of mass of the mitochondrial network at the end of the movie was used as a central reference. Displacements towards this central reference were referred to as "retrograde" and movements opposite to that direction were referred to as "anterograde". Finally, movements within 10° of the perpendicular to the antero-retrograde axis were named "perpendicular". The time between changes of direction was measured for each of the patches. The "net retrograde movement" is the difference between the averages of retrograde and anterograde displacements. Electron microscopy and tomography. RPE1 cells were grown in 6 well plates, treated with GCA (10 µM) or DMSO alone as a control for 1 h30min. The cells were fixed with 1% glutaraldehyde and 3% formaldehyde in PBS, followed by post fixation in reduced osmium containing potassium ferrocyanide and 1% osmium tetroxide for 1h. The cells were stained pre-embedding with uranyl acetate prior to dehydration in ethanol and embedded in epoxy resin (Durcupan). 300 nm-thick semi-thin sections were cut, counter-stained in lead-citrate and uranyl acetate and imaged by electron tomography (Tecnai F30, 300 kV, FEI, the Netherlands), using a Gatan US-1000 camera and the SerialEM acquisition software 49 . Electron tomograms were reconstructed using the IMOD software 50 . The segmentation of mitochondria was performed manually in IMOD on computer-generated sections, whereas microtubule segmentation was performed in Amira using automated segmentation, followed by manual validation 51 . Quantitative parameters from EM models were extracted using IMOD imodinfo prior to statistical analysis.
Quantification and statistical analysis. For statistical analysis, values are expressed as mean ± S.D.
(mean ± S.E.M. for electron tomography) of at least three independent experiments. The significance of differences between groups was tested using the Student's t test on unpaired samples. A value of p ≤ 0.05 was considered significant. Statistical analysis was performed using the Sigmastat software.

Data Availability
The data generated and analysed during the current study are available from the corresponding authors on reasonable request.