Mfn2-dependent fusion pathway of PE-enriched micron-sized vesicles

Significance

Membrane fusion is essential for mitochondrial physiology. Mitofusins 1 and 2 (Mfns) are GTPase proteins that perform the fusion of the outer mitochondrial membranes. From structural information of chimeric and incomplete proteins, it is conjectured that Mfns use guanine 5′-triphosphate disodium salt (GTP) to approximate and fuse apposing membranes through a conformational change. However, the mechanistic information of the Mfn-mediated fusion is not fully understood. Here, we present the functional reconstitution of the full-length Mitofusin 2 into giant lipid vesicles. High-speed video-microscopy shows that GTP-fueled Mfn2 proteins zipper apposing vesicles at the rim of the contact region, which eventually destabilizes and leads to the formation of a fusion opening. The Mfn2-mediated fusion process required the presence of physiological concentrations of phosphatidylethanolamine.

Abstract

Mitofusins (Mfn1 and Mfn2) are the mitochondrial outer-membrane fusion proteins in mammals and belong to the dynamin superfamily of multidomain GTPases. Recent structural studies of truncated variants lacking alpha helical transmembrane domains suggested that Mfns dimerize to promote the approximation and the fusion of the mitochondrial outer membranes upon the hydrolysis of guanine 5′-triphosphate disodium salt (GTP). However, next to the presence of GTP, the fusion activity seems to require multiple regulatory factors that control the dynamics and kinetics of mitochondrial fusion through the formation of Mfn1–Mfn2 heterodimers. Here, we purified and reconstituted the full-length murine Mfn2 protein into giant unilamellar vesicles (GUVs) with different lipid compositions. The incubation with GTP resulted in the fusion of Mfn2-GUVs. High-speed video-microscopy showed that the Mfn2-dependent membrane fusion pathway progressed through a zipper mechanism where the formation and growth of an adhesion patch eventually led to the formation of a membrane opening at the rim of the septum. The presence of physiological concentration (up to 30 mol%) of dioleoyl-phosphatidylethanolamine (DOPE) was shown to be a requisite to observe GTP-induced Mfn2-dependent fusion. Our observations show that Mfn2 alone can promote the fusion of micron-sized DOPE-enriched vesicles without the requirement of regulatory cofactors, such as membrane curvature, or the assistance of other proteins.

Membrane fusion is crucial for many fundamental physiological processes, including neuronal communication (1), viral infection (2), and mitochondrial dynamics (3). During the fusion event, two biological compartments must approximate and merge to mix their membranous and aqueous components. This process is energetically unfavorable and catalyzed by the action of specialized fusion proteins to overcome the energy barrier of 30 k_BT of the complete fusion pathway (4). The common fusion process usually proceeds through a series of intermediate stages that include the membrane approximation and docking, the potential formation of a hemifusion structure/stalk, the expansion of the hemifusion diaphragm, and the opening and growth of a fusion pore (5). Individual domains of the specialized fusion proteins are thought to orchestrate one or more of these intermediate stages throughout the fusion pathway (6). Whereas the core of underlying molecular mechanisms is well established for neuronal and intracellular vesicle trafficking (7), the detailed mechanism of mitochondrial membrane fusion remains elusive (8).

Mitochondrial fusion is described as a two-step process that starts with the GTP-dependent fusion of the outer mitochondrial membrane (OMM) that is subsequently followed by the fusion of the inner mitochondrial membrane (IMM) (9). The first step is directed by the mitochondrial fusion proteins Mitofusin 1 and 2 (Mfn1 and Mfn2) (10, 11), whereas the fusion of IMM is controlled by the optic atrophy type 1 protein Opa1 (12). Although both fusion events are independent, they need to strictly be coordinated to prevent mitochondrial dysfunction (13).

Mfn1 and Mfn2 share a high homology of their amino acid sequence (14–17) and tertiary structure (18, 19), with an amino-terminal GTP hydrolyzing domain (G-domain) followed by a heptad-repeat domain (HR1), two transmembrane segments (TMS), and a second carboxy terminal HR2 (8, 14, 20) (Fig. 1A). Though this might suggest a certain redundancy (21), it was shown that Mfn1 and Mfn2 exhibit differences in their expression patterns throughout different tissues (10, 15, 17, 22), nucleotide-binding affinity and hydrolysis rate (11, 19), and posttranslational modifications (23, 24). The mitochondrial fusion proteins are classified as members of the dynamin protein superfamily (25). As seen for other members of the dynamin superfamily, the modular design of Mfn suggests allocating specific functions to domains of the protein.

Fig. 1.

Expression and purification of full-length murine Mfn2 proteins. (A) Schematic representation of Mfn2 WT, Mfn2-myc, and Mfn2-myc K109A proteins with the functional domains of the molecule: GTP hydrolyzing domain (GTP domain), heptad-repeat domains 1 and 2 (HR1 and HR2), and two transmembrane segments (TMS 1 and TMS2). Domains are color-coded: pink, GTPase; blue, HR1; yellow, HR2; and white, TMSs. The arrow indicates the position of the K109A mutation. (B) Hypothesized folding knife-type transition leading to membrane docking. The dashed line represents the Mfn2–Mfn2 interaction interface. The open conformation was established according to the AlphaFold protein structure database (Q80U63) whereas the putative closed conformation is based on structural homology with the crystal structure of the chimeric Mfn1 in the presence of GTP (PDB ID: 5YEW). See SI Appendix for more details. (C) Immunoblot analysis of Mfn2-myc protein in wild-type HEK293 cells (lane 1) HEK293/Mfn2 cells (lane 2), isolated mitochondria from HEK293/Mfn2 (lane 3), and purified protein (lane 4). Immunodetection was revealed with commercial α-Mfn2 and α-Myc antibodies. (D) Titration of GTPase activity of Mfn2-myc (Mfn2) and Mfn2-myc K109A (K109A) as a function of GTP concentration. Protein concentration was 130 nM. Each point represents the average ± SD values of two independent measurements. (E) Binding affinity to GMPPCP for Mfn2-myc (Mfn2) and Mfn2-myc K109A (K109A) as measured by isothermal titration calorimetry (ITC). Dissociation constants, K_d, and stoichiometry numbers, N, are given in Insets.

The G-domain was very early hypothesized to function like the dynamin GTPase domain, which converts the chemical energy released during GTP hydrolysis into mechanical energy to exert the forces needed during membrane fusion (25). Indeed, the crystal structures of the truncated Mfn1 variant, lacking both TMSs and the second HR domain (18, 26), together with the crystal structures of the bacterial dynamin-like protein BDLP (27, 28), suggest that the GTP binding and hydrolysis promotes a large structural change from an extended conformation to a compact conformation, through a folding knife-type transition (Fig. 1B) (27, 28). Similar structures have been reported for truncated Mfn2 proteins (19). Mfn1 and Mfn2 are thus conjectured to bring the adjacent Mfn-containing membranes closer together and initiate both their adhesion and fusion upon the binding and hydrolysis of GTP (29).

The modular function of the individual Mfn domains throughout the progress of the membrane fusion can also be inferred by the membrane destabilization function of HR1 (30). The reconstitution of the HR1 and HR2 domains of Mfn1 in DOPC liposomes induced liposome fusion in vitro. This fusion occurred in the absence of nucleotide and is related to the capacity of the HR domains to insert into lipid bilayers, particularly in regions presenting lipid-packing defects (31). Thus, the HR domains might be involved in the membrane merging after docking. The importance of this finding was validated in Mfn1-KO mouse embryonic fibroblasts that were not able to be rescued in mitochondrial morphology by Mfn1 mutants lacking either HR1 or HR2 domains (30).

In addition to the structural features, it was suggested that Mfns might be assisted by other regulatory factors more related to the lipid bilayer membrane itself (8, 32, 33). Lipid bilayer membranes built of native mitochondrial lipid extracts retain the mechanical properties that one would expect for highly dynamic membranes. They are essentially fluid and highly deformable upon compression and bending (34). The mechanical properties are related to the lipid composition which plays an important role in promoting and driving the progression of intermediate membrane hemifusion states (35). The presence of fusogenic lipids such as dioleoyl-phosphatidylethanolamine (DOPE) promotes the hemifusion state to the complete fusion of the lipid bilayers (36). Both OMM and IMM are enriched in diester phospholipids, but they also contain a high amount of plasmalogens (37) that are also considered to be fusogenic lipids (38). For example, in addition to diester glycerophosphatidylcholine (PC), the OMM contains up to 30% high percentage of plasmenylcholine (p-PC), and thus, it is worthy studying the role of plasmalogens in Mfn-dependent mitochondrial fusion.

Taken together, all findings lead to the question about the minimal components necessary for Mfn-dependent fusion and the pathway it occurs. Here, we report the functional reconstitution of the full-length Mfn2 protein into lipid vesicles. In vitro experiments with Mfn2 reconstituted into small unilamellar vesicles (Mfn2-SUVs) showed that GTP, but not guanine 5′-diphosphate sodium salt (GDP), induces the fusion of SUVs in the second timescale when composed of palmitoyloleoylphosphatidylcholine (POPC) and DOPE. Moreover, the protein-to-lipid ratio (L/P) for efficient fusion was determined to be L/P ≈ 10³. The direct visualization of the fusion process was tracked by reconstituting the fusion protein into giant unilamellar vesicles (GUVs). High-speed video-microscopy demonstrated that Mfn2-dependent fusion follows an uncommon pathway where the adhesion patch of apposing membranes does not progress to a hemifusion diaphragm but rather grows through a zipper mechanism at the rim of the contact interface or septum. Finally, the adhesion patch opens nearby the rim of the septum and expands fast to eventually complete the fusion of two apposing membranes. Thus, Mfn2 alone embedded in a PE-enriched lipid bilayer is able to promote membrane fusion in vitro without the presence of other regulatory factors such as membrane curvature, membrane potential, or accessory proteins.

Results

Expression and Purification of Full-Length Mouse Mfn2.

To reconstitute the Mfn-dependent membrane fusion of lipid vesicles, we used the mouse Mfn2-myc construct that was previously shown to complement mitochondrial fusion of Mfn2-deficient cells (15) (Fig. 1A). We produced the full-length mouse Mfn2 in HEK293 suspension cells and purified the protein from isolated mitochondria by affinity chromatography using anti-c-Myc agarose (SI Appendix, Materials and Methods). For convenience, we call these cells HEK293/Mfn2. Immunoblot analysis using the commercial α-Mfn2 antibody shows that the Mfn2 protein was overproduced in HEK293/Mfn2 and found to be present in the isolated mitochondrial fraction of these cells (Fig. 1C). The observed protein band of interest migrates at an apparent molecular size of 130 kDa and not at 112 kDa as expected from its primary amino acid sequence (15). Wild-type HEK293 cells did not produce the 130 kDa protein band (Fig. 1C, lanes 1). Moreover, we observed the presence of the endogenous Mfn2 protein (82 kDa) in wild-type HEK293, HEK293/Mfn2, and mitochondria isolated from HEK293/Mfn2 cells (Fig. 1 C, Left panel, lanes 1 to 3). When α-Myc antibody was used, we observed the same 130 kDa protein band that was detected with the α-Mfn2 antibody and the presence of a possible degradation product with an apparent size of 75 kDa (Fig. 1 C, Right panel, lanes 2 to 4). For both antibodies, western blot analysis of the purified protein revealed one major band at 130 kDa and a faint band below 70 kDa that might be due to cross-reactivity of the commercial antibodies (Fig. 1C, lanes 4). In addition, the SDS-page silver staining of the different elution fractions obtained during Mfn2 purification (SI Appendix, Fig. S1) also revealed the presence of the 130 kDa protein band, which was extracted from the gel to be analyzed by mass spectrometry (SI Appendix, Materials and Methods) that verified the identity of this band to be the recombinant Mfn2 protein with a probability score for protein identification of 41 fixed to a p-value less than 0.05 (SI Appendix, Fig. S2). As a control, we also expressed and purified the GTPase mutant Mfn2-K109A, which is thought to bind, but not hydrolyze, GTP (39) (Fig. 1A). Mfn2-K109A alters the nucleotide-binding pocket and lacks K109, which is necessary to stabilize the gamma-phosphate of the GTP at the transition state to complete the nucleophilic attack during GTP hydrolysis (15, 40). Finally, we also produced and purified the GTPase mutants Mfn2-E230A and Mfn2-R259A, which are known to alter the GTPase domain interface and protein dimerization (19). The high purity of proteins was demonstrated as all of them showed the same native molecular weight in SDS-page silver-stained gels (SI Appendix, Fig. S3). The presence of other interacting proteins in Mfn2 fractions, such as Mfn1 (39), was discarded by western blot of the purified proteins (SI Appendix, Fig. S4).

GTP Hydrolyzing Activity and Nucleotide-Binding Affinity of Purified Mfn2.

We further checked the GTPase hydrolyzing activity of purified Mfn2 and Mfn2-K109A proteins through the amount of phosphate release with the malachite green assay (41). Purified Mfn2 fractions with higher protein concentration were able to hydrolyze GTP, showing a maximal phosphate release of ~0.48 μM P min⁻¹ for Mfn2 and 0.05 μM P min⁻¹ for Mfn2-K109A (Fig. 1D). Data were corrected with the absorption of the reaction in the absence of either protein or nucleotide that background coloring of the buffer components. The observed GTPase hydrolyzing activity of the solubilized Mfn2 protein allows us to conclude that the purification procedure does not affect protein activity or functionality. If we compare our data with published GTPase hydrolyzing activities and divide our phosphate release of ~0.48 μM P boomin⁻¹ by the amount of Mfn2 per reaction (0.139 μM) we obtain a phosphate release of ~3.5 min⁻¹, a value that is one order of magnitude higher than the ~0.4 min⁻¹ reported previously for the truncated Mfn2_IM mutant (19). In contrast, Mfn2-K109A showed a reduced GTPase hydrolyzing activity, which is not completely abolished, but not sufficient to maintain mitochondrial physiology intact (15).

The nucleotide-binding affinity to Mfn2 and Mfn2-K109A, was then estimated by isothermal titration calorimetry with the nonhydrolyzable guanosine-5′-[(β, γ)-methyleno]triphosphate sodium salt (GMPPCP) (Fig. 1E). We have chosen GMPPCP over GTP to avoid GTP hydrolysis during the binding experiment. As reported for the truncated Mfn2_IM mutant (19), our full-length Mfn2 shows a similar affinity for GMPPCP, as characterized by the binding constant K_d value obtained from the fitting of binding curves to a standard single-site binding model. As expected, the Mfn2-K109A abolished or diminished the nucleotide binding, in agreement with previously reported data for similar mutants in Mfn1 (18, 42).

Assessment of the Fusion Activity of the Full-Length Mfn2 through the Lipid-Dilution and Volume-Mixing Assays.

To determine the lipid-to-protein ratio required for fusion activity, the lipid specificity, and the kinetics of Mfn2-mediated membrane fusion, we first performed the conventional lipid-dilution assay (43). The lipid-dilution assay is a Förster resonance energy transfer (FRET)-based method that monitors the lipid exchange between fluorescently labeled and nonlabeled lipid vesicles (Fig. 2A). The used donor and acceptor FRET-pair are lipid-specific fluorescent dyes present at high concentration on the same liposome to produce FRET and suppress the donor fluorescence signal. Upon fusion of labeled and nonlabeled liposomes, the concentration of the dye decreases due to their distribution throughout the entire liposome surface and the higher average distance between the FRET probes. As a consequence, the efficiency of the FRET-pair decreases leading to an increase of the fluorescence emission intensity of the donor dye.

Fig. 2.

Lipid-dilution and volume-mixing assays of proteo-SUVs. (A) Schematic illustration of experimental design. Lipid-dilution assay: proteo-SUVs composed of POPC, POPC:p-PC (70:30 mol%), or POPC:DOPE (70:30 mol%) were prepared. One of the populations also contained the FRET NBD-PE donor and Rho-PE acceptor at 0.6 and 0.1 mol%, respectively. After mixing of the two populations at a 1:6 molar ratio, the initial FRET efficiency decreases as a result of the dilution of lipid probes into the fused membrane. The fusion kinetics is traced through the increase of the fluorescence emission intensity of the donor dye. The lipid concentration was 2 mM and the lipid-to-protein ratio ranged from 3,000:1 to 10,000:1. Volume dilution assay: proteo-SUVs composed of POPC:DOPE (70:30 mol%) were prepared. Two separate vesicle populations are loaded with the volume probe HPTS or the fluorescence quencher DPX. After mixing of the two populations at a 1:1 molar ratio, the initial HPTS fluorescence emission decreases as a result of the volume mixing in the fused vesicle. The fusion kinetics is traced through the quenching of the fluorescence emission intensity of HPTS. The lipid concentration was 2 mM and the lipid-to-protein ratio was 3,000:1. (B) Mfn2-mediated membrane fusion is shown as a function of time at different lipid-to-protein ratios. Lipid composition was POPC:DOPE (70:30 mol%). GTP was added at time 0 (final concentration was 0.5 mM). The NBD emission was monitored at 530 nm, with the excitation wavelength set at 465 nm. Fusion yield (%) was calculated as described in Materials and Methods. (C) FRET efficiency ratio of vesicles for different lipid-to-protein ratios. Data were calculated from spectra obtained before and after 5 min of incubation with GTP (mean values ± SD, with N = 3). (D) Mfn2-mediated membrane fusion is shown as a function of time in different lipid compositions. The experimental conditions are the same as in B. (E) FRET efficiency ratio of vesicles with and without protein. Data were calculated from spectra obtained before and after 5 min of incubation with GTP (mean values ± SD, with n = 3). (F) Mfn2-mediated membrane fusion is shown as a function of time using the volume mixing assay. GTP was added at time 0 (final concentration was 0.5 mM). The HPTS emission was monitored at 512 nm, with the excitation wavelength set at 452 nm. Fusion yield (%) was calculated as described in Materials and Methods. Statistical differences were examined by Student's t test. Symbols denote statistical significance (**P < 0.01).

Mfn2-SUVs were labeled with both NBD-PE (donor) and Rho-PE (acceptor) at a molar ratio of 1:6, respectively. At this molar ratio, we obtained a good FRET signal at λ_em = 580 nm upon the excitation of the sample at λ_ex = 465 nm. The fusion kinetics was assessed by the incubation of labeled Mfn2-SUVs with unlabeled Mfn2-SUVs in the presence and absence of GTP. A first set of experiments consisted of the titration of Mfn2 determining the minimal lipid-to-protein that leads to membrane fusion. This was done with lipid-to-protein ratios (L/P) ranging from 10,000:1 to 3,000:1 in Mfn2-SUVs with a lipid composition of POPC:DOPE (70:30 mol%) as the OMM is enriched in PE species (34).

Remarkably, the addition of GTP to Mfn2-SUVs led to a rapid increase of the donor fluorescence intensity within seconds for lipid-to-protein ratios up to L/P = 8,000 (Fig. 2B). The fusion kinetics observed here are much faster than seen for other in vitro mitochondrial fusion assays that either used the isolated HR1 domain (30) or complete mitochondria (9). Interestingly, the experimental data could be described by a sigmoidal growth function, where a short lag phase was followed by a sudden growth to finally reach an asymptotic value at long times. From the multiple sigmoidal growth models capable of capturing the fusion evolution trajectory with accuracy, the Gompertz model (44) fitted better with the empirical data as, in contrast to the simple logistic function, both asymptotes are approached asymmetrically. From the fit, we could obtain the maximum asymptotic fusion yield $\left(F_{POPC:DOPE}^{\mathit{max}}\right)$ and the fusion rate ($k_{POPC:DOPE}$) of different lipid-to-protein ratios: $F_{POPC:DOPE}^{\mathit{max}}=25\%,20\%,12\%$, $k_{POPC:DOPE}=0.03 {\mathrm{s}}^{-1},0.02 {\mathrm{s}}^{-1},0.01 {\mathrm{s}}^{-1}$ for L/P = 3,000, 5,000, and 8,000, respectively (Fig. 2C). Unlike high-protein content conditions, Mfn2-SUVs with a L/P = 10⁴ rather followed a hyperbolic model, where a fusion rate constant $k_{10,000}=0.1 {\mathrm{s}}^{-1}$ and a maximum asymptotic fusion yield of $F_{10,000}^{\mathit{max}}=5\%$ were obtained. Note that at the optimal L/P = 3,000, complementary experiments using the mutant Mfn2-K109A did not promote lipid mixing upon GTP incubation (Fig. 2B).

The role of lipid composition in fusion activity was evaluated in a second set of experiments using Mfn2-SUVs made of POPC or POPC:p-PC (70:30 mol%) and with the optimal L/P = 3,000. Likely to DOPE-containing Mfn2-SUVs with a L/P =10,000; Mfn2-SUVs lacking DOPE exhibited a modest decrease of the FRET efficiency below 5% upon GTP incubation (Fig. 2D). The absence of a high fusion yield for those lipid compositions confirms the essential role of specific lipids for completing Mfn2-mediated membrane fusion, in agreement with a previous report using the HR1 domain (45). Moreover, their fusion time trajectory followed also a hyperbolic model, where similar fusion rate constants were obtained for both lipid compositions ($k_{\mathit{POPC}}=0.1 {\mathrm{s}}^{-1},k_{POPC:p-PC}=0.5 {\mathrm{s}}^{-1}$), with maximum asymptotic fusion yields of $F_{\mathit{POPC}}^{\mathit{max}}=3\%$ and $F_{POPC:p-PC}^{\mathit{max}}=5\%$. We also monitored, as a control experiment, the spontaneous membrane fusion of bare SUVs made of POPC, POPC:DOPE (70:30 mol%), and POPC:p-PC (70:30 mol%) in the presence of GTP (Fig. 2E). As expected, in the absence of DOPE, the donor fluorescence intensity remained unaltered over the measured time interval of 5 min and the FRET efficiency did not decrease significantly. However, the addition of 30% mol of DOPE to the lipid composition produced a small variation of the FRET signal (up to 5% after 5 min), which confirms this lipid as a helper species for spontaneous membrane fusion (4).

Although the lipid-mixing assay indicated the Mfn2-mediated membrane fusion in SUVs, we additionally monitored the complete membrane fusion through the volume-mixing assay (46) (Fig. 2A). This was done to discard that our results did not represent hemifusion or stable fusion intermediates. Content mixing assays typically rely on the detection of fluorescence quenching upon the intermixing of soluble compounds trapped in previously separate liposome populations. Here, one population of Mfn2-SUVs was loaded with the volume fluorescent marker pyranine [8-hydroxypyrene-1,3,6-trisulfonic acid trisodium salt (HPTS)], whereas the second population of Mfn2-SUVs was loaded with the complementary fluorescence quencher p-xylene-bis-pyridinium bromide (DPX). Again, we assessed the fusion kinetics by monitoring the fluorescence quenching of HPTS after incubation of both populations in the presence of GTP using Mfn2-SUVs made of POPC:DOPE (70:30 mol%) with the optimal L/P = 3,000.

Also, the addition of GTP to Mfn2-SUVs led to a rapid increase of the fusion signal as quantified through the decrease in the fluorescence signal of HPTS (Materials and Methods). The fusion reaction was completed in 5 min (Fig. 2F), in reasonable agreement with the lipid dilution kinetics observed under the same experimental conditions of L/P ratio and lipid composition (Fig. 2B). Moreover, control experiments using the Mfn2-K109A mutant or pure lipid SUVs did not present a significant protein-mediated fusion yield but rather exhibited a spontaneous fusion pattern also in agreement with data obtained with the lipid-dilution assay (Fig. 2B).

The Mfn2-mediated Fusion of GUVs Requires GTP Hydrolysis and PE.

To extend our observations made in SUVS and directly visualize that Mfn2 proteins are capable of mediating complete membrane fusion, we evaluated the protein fusion activity in GUVs. Mfn2-GUVs were formed by detergent-mediated incorporation (47) of the purified Mfn2 protein into previously electroformed GUVs composed of POPC:DOPE (70:30 mol%). The incorporation of Mfn2 into the lipid bilayers was confirmed through fluorescent labeling of Mfn2 with Alexa-555. Confocal fluorescence microscopy images show the presence of Alexa-555 fluorescent signal in the GUV membrane (SI Appendix, Fig. S5). The lipid-to-protein ratio corresponded to intermediate protein contents ranging from L/P = 1,000 to L/P = 5,000 (N = 40), as measured from the calibration curves using an Alexa-555 bioconjugated lipid especially synthesized for this quantification (see SI Appendix, Fig. S6 and SI Appendix for details). Note that Mfn2 exhibits fusion activity in SUVs at this L/P. Interestingly, after protein reconstitution, the Mfn2-GUVs exhibited a tendency to adhere to each other (SI Appendix, Fig. S7). This observation suggested that purified Mfn2 might reside in a GDP-bound state and, similarly to Mfn1, induce membrane adhesion (48). Indeed, the reconstitution of Mfn2 mutants that alter the GTPase dimerization interface (E230A, R259A) diminished the adhesion of GUVs. The direct interaction of Mfn2 in trans might be then responsible for the docking of Mfn2-bearing membranes (SI Appendix, Fig. S8). However, additional experiments are needed to quantitatively address this issue.

To monitor the nucleotide-induced Mfn2-dependent fusion of GUVs, we first incubated our sample with GDP and MgCl₂ at a final concentration of 0.5 mM and 1 mM, respectively. Apparently, no difference was observed when compared to the sample in the absence of nucleotide and MgCl₂ (Fig. 3A). However, the incubation of Mfn2-GUVs with 0.5 mM GTP in the presence of 1 mM MgCl₂ induced the fusion of GUVs, progressing from two or more adhering vesicles into one spherical vesicle (Fig. 3B). In contrast to the wild-type Mfn2 protein, the K109A mutant did not promote the fusion of GUVs in the presence of GTP. This control experiment verifies that GTP binding and hydrolysis are needed for Mfn2-mediated fusion (Fig. 3C).

Fig. 3.

Mfn2-mediated fusion of GUVs. Membrane remodeling of Mfn2-GUVs upon the addition of either GDP (A) or GTP (B) at room temperature. Control experiments with Mfn2-K109A were also run in the presence of GTP (C). GUVs are composed of POPC and DOPE (70:30 mol%) and labeled with the green membrane dye ATTO-PE. Images of representative experiments were taken at indicated intervals (in seconds) after the addition of the nucleotides and MgCl₂ at a final concentration of 0.5 mM and 1 mM, respectively. Vesicles remained adhered for several minutes in A and C. Scale bars correspond to 10 µm.

To confirm the essential role of PE in promoting fusion, we removed the fusogenic lipid DOPE from the lipid composition of GUVs and tested Mfn2-POPC-GUVs for fusion events in the presence of GTP and MgCl₂. Under these conditions, the incubation of GTP did not promote the fusion of vesicles. As we did not observe any fusion event from nine independent assays, therefore we assume that the presence of PE is crucial for Mfn2-mediated fusion. To rule out that the observed fusion events were mainly driven by the presence of DOPE in the GUV lipid composition, we repeated our fusion assay with bare GUVs (POPC:DOPE; 70:30 mol%). As expected, the presence of GDP or GTP did neither promote the adhesion nor the fusion of GUVs.

The spontaneous fusion of small liposomes (<100 nm in diameter) is associated with a free energy barrier of 30 k_BT (4). The fraction of vesicles leading to complete fusion at room temperature can be estimated through the equilibrium Boltzmann distribution as $f\propto e^{\frac{30k_{B}T}{k_{B}T}}\approx {10}^{-7}$. Unlike highly curved liposomes, where spontaneous fusion is mainly driven by the relaxation of membrane tension, flat and fluctuating surfaces as those from GUV models prevent their spontaneous fusion, which might be associated with an even higher energy barrier. Our observation of multiple fusion events in the same field of observation (SI Appendix, Fig. S9 and Movie S1) confirms the functional reconstitution of the full-length mouse Mfn2 into lipid bilayers and represents additional evidence for the specific interaction between Mfn2 proteins upon addition of GTP.

Fusion Pathway by the Mitochondrial Mfn2.

As the dynamics of the membrane fusion event occurs in the submillisecond timescale (49), a closer assessment of Mfn2-GUV fusion was obtained by high-speed video-microscopy with an acquisition rate of 10 to 25 fps. To visualize and track the sample during the fusion event we labeled the Mfn2-GUVs with the photostable lipid dye DOPE-ATTO488 (ATTO-PE). For a quantitative evaluation, the tracking of fusion events was recorded upon the addition of GTP to adhering Mfn2-GUV doublets. We identified two different dynamic regimes (Fig. 4A). In a first step, the contact region between Mfn2-GUVs gradually increased leading to the growth of the contact area that was simultaneously accompanied by the accumulation of lipid material at the rims of the septum. The accumulation of lipid material can be inferred from an increase of the fluorescence intensity of the lipid channel in those regions. In a second regime, one edge of the septal plane opened and retracted very rapidly. Eventually, the relaxation of the contact membrane usually produced a small spherical vesicle that was trapped inside the lumen of the fused GUV (SI Appendix, Fig. S10).

Fig. 4.

Mfn2-dependent fusion pathway as visualized by high-speed video microscopy. (A) After GTP addition, the contact region between two adhering Mfn2-GUVs gradually enlarged without the apparent increase of the contact angle. Simultaneously, an accumulation of the fluorescent lipid at the rims of the septum occurs. In a second step, the destabilization of the rim leads to the formation and expansion of a membrane opening and the adhesion patch retracts. (B) Schematic representation of the lipid rearrangement during membrane fusion for the standard hemifusion model (Left) and the partition breakage model (Right). (C) Mean fluorescence intensity ratio; $\rho =\frac{I_{\mathit{septum}}}{I_1+I_2}$ for Mfn2-GUVs undergoing fusion previous to membrane destabilization. Average value ± SD is ρ = 0.9 ± 0.2 (N = 25). Values close to 1 are indicative of the absence of a hemifused patch. (D) Snapshot of adhering Mfn2-GUVs. Red and yellow dashed lines circumscribe the spherical caps with radii R₁ and R₂ of the two GUVs. A white solid line marks the adhesion patch of length, L. The contact angles θ₁ and θ₂ are measured using Eq. 2 (Materials and Methods). (E, Left) Contact angles of adhering Mfn2-GUVs made of POPC:DOPE (70:30 mol/mol) and POPC prior to the addition of nucleotides. Average value ± SD are ${\theta }_{POPC:DOPE}=30\pm 15$ (N = 27) and ${\theta }_{\mathit{POPC}}=30\pm 15$ (N = 27). (E, Right) Variation of contact angles as a function of the variation of length of the adhesion patch upon addition of GTP. (F) Retraction rate of the septum after its destabilization at the rim. $v=-\frac{\mathit{dS}}{\mathit{dt}}=23\pm 21 \mathrm{\mu }\mathrm{m} {\mathrm{s}}^{-1}$ (N = 8) as experimentally measured from the video-micrographs. (G) Variation of fluorescence intensity at the rims upon addition of GTP in Mfn2-GUVs made of POPC:DOPE and POPC. Two fluorescent lipid analogues were tested. Average value ± SD are ${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{POPC/DOPE}^{ATTO-PE}=1.2\pm 0.3$, N = 20; ${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{\mathit{POPC}}^{ATTO-PE}=1.0\pm 0.3$, N = 5 and ${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{POPC/DOPE}^{ATTP-PE}=0.9\pm 0.2$, N = 9. (H) Volume conservation during vesicle fusion as measured by the volume ratio between the final GUV (V₃) and the volumes of initial GUVs (V₁ + V₂). The average value ± SD is $\frac{{\mathrm{V}}_3}{{\mathrm{V}}_1+{\mathrm{V}}_2}=1.1\pm 0.1$ (N = 25). Boxplot elements are as follows: black square—mean; center line—median; box limits—upper and lower quartiles; whiskers—SD; symbols—experimental data. The scale bar corresponds to 10 µm.

Both the observed fusion pathway and the requirement of PE for membrane fusion relatively match with the partition (septum) breakage model (Fig. 4B) (50). Unlike the conventional fusion model, the partition breakage model does not go through the formation of a hemifusion diaphragm intermediate at the septum region, but it rather proceeds through the fusion between the membranes at the edge of the adhesion patch and its destabilization at one site on the edge is favored by the presence of negative curvature-inducing lipids such as DOPE. A first assessment of this fusion pathway was obtained by analyzing the putative formation of a hemifusion state at the adhesion patch. The identification of a hemifusion intermediate is based on a change in the mean fluorescence intensity ratio; $\rho =\frac{I_{\mathit{septum}}}{I_1+I_2}$; where $I_{\mathit{septum}}$ is the mean fluorescence intensity at the septum and I₁ and I₂ are the mean fluorescence intensities at the poles of adhering Mfn2-GUVs. For docked membranes, ${\rho }_{\mathit{docked}}\le 1$. Deviations from ρ ≈ 1 are explained by the spatial resolution (Fig. 4C). In contrast, ${\rho }_{\mathit{hemi}}\approx 0.5$ for hemifused vesicles with similar radii, as the contact zone is only sustained by a single bilayer (Fig. 4C). Here, the intensity ratio, ρ, remained constant as a function of time, with typical numerical values of ρ = 0.9 ± 0.2 (N = 25) (Fig. 4C). The absence of changes in ρ suggests that the apposing membranes remain in an adhered state until the opening at the rim of the adhesion patch takes place without the formation of a hemifusion intermediate at the adhesion patch.

To derive the underlying mechanism of the Mfn2-dependent membrane fusion, we then analyzed the process of the membrane remodeling at the adhesion patch by monitoring the time evolution of geometrical parameters such as the length of the septum, the radii of adhering vesicles and their contact angle (Fig. 4D). The observed increase of the contact area upon the addition of GTP may be due to an increase of the adhesion energy between GUVs. The adhesion energy between two apposing GUVs follows, in a first approximation, the Young equation for liquid droplets:

$$E_{\mathit{adh}}={\sigma }_1\left(1-\cos {\mathrm{\theta }}_1\right)+{\sigma }_2\left(1-\cos {\mathrm{\theta }}_2\right),$$

where σ₁ and σ₂ are the surface tensions, and θ₁ and θ₂ are the contact angles. Unlike liquid droplets, the resulting contact angle of adhering vesicles is constrained by both the constant membrane area and volume vesicle. Under mechanical equilibrium, in the absence of nucleotide, the contact angles directly depend on the membrane tensions of membranes. It is known that electroformation produces vesicles with very different initial tensions and degree of deflation in the same batch. Therefore, tenser vesicles will present lower contact angles than deflated vesicles as shown by the disperse distribution of the contact angles before incubation with nucleotides where ${\theta }_{POPC:DOPE}=30\pm 15$ (N = 27) and ${\theta }_{\mathit{POPC}}=30\pm 15$ (N = 27) (Fig. 4E). Upon addition of GTP, an increasing number of protein contacts might promote enlarged adhesion patches, which are accompanied by an increase in the contact angle (Fig. 4E). As expected, this effect was also observed in POPC GUVs although only PE-containing vesicles led to membrane fusion. Interestingly, the size of the adhesion patch sometimes increased without a significant change in the contact angle in many cases. This apparent paradox might be explained at first by considering a change in the surface tension through an elastic deformation of the membrane. The transition from the open to the close conformation of Mfn2 upon GTP hydrolysis might induce a visible stretching of membranes during their approximation (Fig. 1B). This would lead only to small membrane area increases, as area expansions of 10% are known to reach the lysis tension of lipid bilayers, ${\sigma }_{\mathit{lys}}\approx {10}^{-3} \mathrm{N}/\mathrm{m}$ (51), following the equation $\sigma =K\frac{A-A_0}{A}$, where K is the stretching modulus of the lipid bilayer. This is inconsistent also with the larger expansions of the septum (Fig. 4E). Again, electroformed GUVs are also known to frequently exhibit hidden membrane reservoirs, which might contribute to enlarge the contact area during adhesion. The molecular details of how the adhesion patch grows remain elusive in our experimental setup, but the increased fluorescence intensity at the rims of the contact zone suggests a local accumulation of membrane material in those regions. As the dynamic changes at the septum occur upon GTP incubation, it is likely that the increased contact area is produced through a zipper mechanism provided by conformational changes of proteins at the rim (Fig. 1B), where the disruption of the membrane eventually occurs (8).

An additional insight into the fusion pathway is provided by the assessment of the membrane tension, which sustains the pulling force that produces the contraction of the membrane septum after breaking. In close analogy with the formation and closing of pores in stretched lipid bilayers, the kinetics of the opening size, r_o(t), might result from the competition between the line tension, λ, and surface tension, σ, as described previously (52);

$$\frac{d{r}_o\left(t\right)}{\mathit{dt}}=\frac{r_o\left(t\right)}{2{\eta }_m}\left[\sigma \left(t\right)-\frac{\lambda }{r_o\left(t\right)}\right],$$

where η_m is the dominant dissipation of the system corresponding to the viscosity of lipid bilayers. Whereas surface tension favors the formation and expansion of the opening through the stretching of membranes, the line tension favors their closure to reduce the interface between the hydrophobic moiety of lipids and the aqueous solvent (Fig. 4B). For single bilayer pores, the opening expands as soon as the membrane tension overcomes the line tension and reaches values close to the lysis tension for lipid bilayers, ${\sigma }_{\mathit{lys}}\approx {10}^{-3} \mathrm{N}/\mathrm{m}$ (51). Under these conditions, the growth of the opening might be very fast and given by the linear approximation (52):

$$r_o\left(t\right)=\frac{r_n\sigma }{2{\eta }_m}t,$$

where r_n = λ/σ is the nucleation size of the opening at t = 0, with typical values of 10⁻⁸ m (52). Here, the expansion rate, $v=-\frac{\mathit{dS}}{\mathit{dt}}=23\pm 21 \mathrm{\mu }\mathrm{m} {\mathrm{s}}^{-1}$ (N = 8), was experimentally measured from the video-micrographs (Fig. 4F) and used to directly estimate the membrane tension as $\sigma =\frac{2v{\eta }_m}{r_c}\approx {10}^{-5} \mathrm{N}/\mathrm{m}$, with ${\eta }_m={10}^{-9} \mathrm{N} \mathrm{s} {\mathrm{m}}^{-1}$ (53). This value for the surface tension is lower than the lysis tension, thus discarding a pure stretching mechanism for membrane destabilization (54).

Alternatively, a curvature-driven mechanism might induce the opening of the septum at the boundaries through the formation of a hemifusion stalk. Hemifusion stalks are under a high curvature stress and the constituting lipids are tightly packed. The increase of the fluorescence intensity of ATTO-PE at the rims during expansion of the septum, $\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}$, is consistent with such a high lipid density structure (${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{POPC/DOPE}^{ATTO-PE}=1.2\pm 0.3$, N = 20) (Fig. 4G). However, the same probe did not accumulate at the rims of Mfn2-GUVs made of POPC upon the addition of GTP (${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{\mathit{POPC}}^{ATTO-PE}=1.0\pm 0.3$, N = 5) (Fig. 4G). This observation suggests that the formation of the hemifusion stalk might be specifically mediated by the presence of a high content of PE. Furthermore, the lipid analogue C12-NBD-PE (labeled at one hydrocarbon chain) did not exhibit any increase of fluorescence in DOPE-containing vesicles (${\left(\frac{I_{\mathit{rim}}}{I_{\mathit{rim}}^0}\right)}_{POPC/DOPE}^{ATTP-PE}=0.9\pm 0.2$, N = 9) undergoing fusion (Fig. 4G). The bulky moiety of the probe might prevent the high packing of the analogous lipid in the interstitial hydrocarbon region of the hemifusion stalk. A high curvature stress would induce the membrane to disrupt by the formation of membrane defects, which might further be established under the form of a double-bilayer structure between the two adhering membranes (Fig. 4B). Taking together, the Mfn2-dependent fusion pathway would prevent the uncontrolled burst of vesicles upon the membrane destabilization that is required to complete the fusion process.

Although the structural details about how Mfn2 destabilizes the rim of the septum remain to be elucidated, the gradual retraction of the contact region usually led to the formation of a small vesicle that was either completely absorbed or trapped inside the lumen of the fused vesicle (SI Appendix, Fig. S10) (55). The intraluminal vesicle is considered a fingerprint of the partition fusion model (50, 55), which preserves volume conservation. Indeed, during the standard hemifusion pathway, there is a significant loss of volume in the final GUV in comparison to the GUVs at the beginning of the fusion process (50), as leakage is associated with the formation of the hemifusion diaphragm (56). To further discern between different fusion pathways, we calculated the ratio of volume between the final GUV (V₃) and the volumes of initial GUVs (V₁ + V₂). The assessment of the volume ratio is $\frac{{\mathrm{V}}_3}{{\mathrm{V}}_1+{\mathrm{V}}_2}=1.1\pm 0.1$ (N = 25), indicating that the volume of the final GUV is equivalent to the sum of the volume of the two initial GUVs and thus compatible with the partition fusion pathway (Fig. 4H).

Overall, our results show that the full-length Mfn2 protein is able to fuse PE-enriched lipid vesicles upon GTP incubation. The Mfn2-dependent fusion pathway follows the partition breakage model, where the fusion between the membranes occurs at the rim of the adhesion patch and its destabilization proceeds at one edge of the septum. The Mfn2 fusion activity is independent of the presence of other proteins or regulatory factors such as membrane curvature or membrane potential.

Discussion

The direct visualization of the membrane contact area of adhering vesicles upon GTP incubation provides us with enough elements for a simple model of Mfn2-dependent membrane fusion at the molecular scale, compatible with the partition fusion pathway (Fig. 5). This model has been proposed previously for vacuolar lysosome fusion (56). Here, a first stage would be the adhesion of apposing membranes by the GDP-bound state of Mfn2 (48). The incubation with GTP would promote the transition from the tethered state to a docked state. As the distance between apposing membranes is on the order of few nanometers, only the Mfn2 proteins at the outermost region of the adhesion patch would be accessible to GTP. Mfn2 proteins are then conjectured to bring the adjacent Mfn2-containing membranes closer together at the edge of the docking membrane through the folding knife-type transition (25). This membrane zippering leads to a larger contact over an extended area as visualized by an increase of the adhesion patch. The accumulation of lipid material at the edge of the docking patch suggests the formation of a hemifusion stalk, which requires a particular rearrangement of lipids at those high curvature sites. The presence of PE at this stage might be essential as discussed below. A high curvature stress would cause the local breaking of the membrane at the edges of the adhesion patch through the formation of membrane defects capable of forming double-bilayer structures between the two adhering membranes. Certainly, the low surface tension measured upon the expansion of the adhesion patch $\sigma \approx {10}^{-5} \mathrm{N}/\mathrm{m}$ suggests that membrane destabilization is curvature-driven and thus prevents the uncontrolled rupture of vesicles (54). The progress of the membrane opening is driven by the relaxation of the docking membrane that is absorbed eventually within the lipid membranes of adhering vesicles or forming an intraluminal vesicle. The depicted model matches strikingly well with a recent study where the authors visualize the tethering, docking, and the local fusion of mitochondria isolated from Saccharomyces cerevisiae with electron cryotomography (57). Each fusion intermediate displayed a particular arrangement of proteins within the fusion site. Thus, a regularly spaced interstitial protein density at the contact zone supports the tethered state whereas the docked state is sustained by the formation of docking ring structure around extended areas of contact. As here, the opening of the membrane is locally formed at the edge of the docking area.

Fig. 5.

Scheme of the proposed mechanism for Mfn2-induced membrane fusion of lipid vesicles including DOPE in their lipid composition. 1) Membrane Tethering: The presence of Mfn2 tethers adjacent Mfn2-containing membranes and forms an adhesion patch, the contact interface between two vesicles. The addition of GTP promotes a transition from the tethered state to a docked state, where only the Mfn2 present at the outermost region of the adhesion patch would be accessible to GTP. 2) Zippering: GTP hydrolysis brings in a zipper-like fashion the adjacent Mfn2-containing membranes closer together at the edge of the docking membrane resulting in a larger contact interface and the extension of the adhesion patch. 3) Curvature-driven destabilization: A high curvature stress causes the local breakage of the membrane at the edges of the adhesion patch resulting in membrane defects capable of rearranging and forming double-bilayer structures between the two adhered membranes. 4) Retraction of the adhesion patch: The local breakage of the membrane produces the relaxation of the adhesion patch that is eventually absorbed within the larger vesicle membrane bilayer of the fused vesicle or also results in the formation of an intraluminal vesicle. See body text for more details.

Another interesting observation from our results is the direct interplay between Mfn2 and PE for a successful fusion of lipid bilayers. First, PE is known to reduce the short-range repulsion that results when lipid bilayers are forced into proximity, as in the case of membrane fusion (58, 59). This interaction describes the resistance of bilayers to dehydration at the level of the polar headgroups and/or their entropic repulsion upon their approximation and it can be estimated from the pressure profile, P(d), of compressed lamellar phases as a function of the interbilayer distance, d as P(d)= P₀ e^{− d /λ}, where the amplitude P₀ and the decay length λ depend on the lipid composition. For small interbilayer distances (≈0.9 nm), the lamellar phase is no longer stable, and transition to hexagonal phases might occur (60). A direct connection between the transition distance and membrane fusion can be established (61). Different reports have shown that PE reduces the transition dehydration pressure leading to the formation of hexagonal phases and hemifusion intermediates, thus favoring membrane fusion (60, 62). For example, the surface energy for the transition of DOPC bilayers can be estimated to be $4 k_{B}T/{\text{nm}}^2$ (60). An accumulation of the nonbilayer lipid DOPE up to 50% of the lipid composition at the rim might reduce the surface energy to $\text{below 3} k_{B}T/{\text{nm}}^2$ (60). For the typical homogeneous lipid-to-protein ratio L/P = 10³ used here, the metabolic surface energy put into play in a single GTP hydrolysis cycle by an interbilayer Mfn2–Mfn2 dimer can be estimated as $\mathrm{\Delta }{G}_{Mfn2}^{\mathit{GTP}}=\frac{2 ·\mathrm{\Delta }{G}^{\mathit{GTP}}}{a ·(L/P)}=0.04 k_{B}T/{\text{nm}}^2$, where, α = 0.7 nm² is the typical mean molecular area of a phospholipid and $\mathrm{\Delta }{G}^{\mathit{GTP}}=15 k_{B}T$ of the GTP hydrolysis reaction to GDP and P_i. This suggests that local and higher Mfn2 densities might be required for the transition dehydration at the contact region between the membranes (57).

Second, PE is also prone to induce negative curvature in lipid bilayers and known to favor the formation of hexagonal and cubic phases (63), which are present in the hemifusion stalk structure. Similar to the biomimetic-induced membrane fusion of GUVs, PE might participate here in decreasing the packing energy of the acyl chains in a putative hemifusion stalk produced at the edge of the septum membrane (50, 55), where the fusion between the outer monolayers of apposing bilayers might occur by the action of Mfn2 proteins. The increased fluorescence intensity of the head-group labeled ATTO-PE (but not the tail labeled C12-NBD-PE) at this site supports the local accumulation of highly packed DOPE lipids at the interstitial hydrocarbon region of the hexagonal phase (Figs. 4B and 5); eventually favoring a controlled destabilization of the septum at the edge (50, 55).

Third, DOPE probably participates in the curvature-driven mechanism for membrane destabilization and the formation of membrane defects of double-bilayer nature. Unexpectedly, a significant decrease in the line tension for lipid pores in DOPC membranes has been found upon addition of DOPE (64). The ability of PE to form interlipid hydrogen bonds (62, 65) might produce its enrichment at the edge of the opening and the stabilization of hydrophilic edges.

Although deeper investigations using complex living systems are required to further connect molecular-level details to biomembrane behavior, the essential role of PE in mitochondrial fusion has been reported so far. Mitochondrial PE (mtPE) is synthesized de novo via the cytidine diphosphate ethanolamine pathway [or Kennedy pathway (66)] or more predominantly by the phosphatidylserine decarboxylase (PSD) pathway (67). Notably, a decrease of mtPE levels in mice or cultured cells leads to an alteration of the mitochondrial morphology, where mitochondria display significant fragmentation and the mitochondrial ultrastructure is compromised (68, 69). Surprisingly, the expression levels of mitochondrial fusion proteins, including Mfn2, remain unaltered or slightly increase in mtPE-deficient cells (69). Also, yeast cells lacking PSD exhibit fragmented mitochondria with impaired mitochondrial fusion and reduced fusion kinetics (70). The metabolism of PE lipids also includes its catalytic conversion to PC by PE N-methyltransferase (PEMT). Mice lacking PEMT display an accumulation of PE and the PC/PE ratio decrease in liver cells. Again, despite the normal levels of mitochondrial fusion proteins in those cells, the mitochondrial morphology is significantly modified by PEMT deficiency (71). Together with our results, a very specific and critical interaction between PE and Mfn2 for mitochondrial fusion can be suggested.

Our experiments also show that Mfn2 can promote membrane fusion of zero-curvature vesicles. Unlike SUVs, GUVs do not reach such high curvatures and therefore are assumed to remain essentially flat. Mitochondrial fusion mainly occurs at their tips (72), where previous studies localized Mfn2 proteins (73). However, mitochondrial fusion events can occur either side-to-tip or side-to-side (74), which is compatible with a spread localization of Mfn2 (14). In consequence, the membrane curvature at the tip might not be an essential requisite for mitochondrial outer membrane fusion, in agreement with our observations in GUVs. As the length of mitochondrial tubules is on the micron scale, one of the mean curvatures at the side is basically zero. The width of mitochondrial tubules depends both on the organism and the cell type but in general, mitochondrial tips have an approximate curvature radius of 0.5 µm (75). Thus, the observation that ∼80% mitochondrial fusion events involve the tip of tubules (74) might be of physiological relevance in a cellular environment related to mitochondrial sorting and organization. Indeed, actin filaments are believed to orchestrate the tip-to-tip localization of the fusion site (76).

Finally, a remarkable experimental result is that Mfn2 alone can lead to membrane fusion of lipid bilayers. However, the formation of Mfn1–Mfn2 heterodimers is essential for the tight control of the mitochondrial fusion activity (19). Interestingly, the overexpression of wild-type Mfn2 cannot complement the mitochondrial fusion in Charcot–Marie–Tooth syndrome type 2A (CMT2A) cells (77). Nevertheless, wild-type Mfn1 rescues the mitochondrial phenotype through the formation of hetero-oligomeric Mfn1–Mfn2 complexes, including complexes that form in trans between mitochondria (39). This suggests that the functional importance of the hetero-oligomeric complexes might lie in the spatial or time regulation of mitochondrial fusion rather than the intrinsic fusion activity. Our experimental approach opens the way to combined reconstitutions of Mfn1 and Mfn2 for a better insight into the mitochondrial fusion process but also creates a synthetic platform to assess the functional modulation of CMT2A Mfn2 variants in the context of drug screening.

Conclusions

In summary, we provide experimental evidence for the expression, purification, and functional reconstitution of full-length Mfn2 into lipid bilayers. Our experiments demonstrate that Mfn2 is able to fuse micron-sized vesicles built up as giant vesicles made of POPC:DOPE (70:30 mol%). The Mfn2-dependent fusion requires the hydrolysis of GTP and the presence of DOPE as an essential fusogenic lipid, which might be involved in the formation of the hemifusion stalk that leads to membrane destabilization. Remarkably, any other regulatory factor, such as membrane curvature, membrane potential, or the presence of other protein partner, is not necessary for completing the fusion of two membranous compartments. Moreover, Mfn2-dependent fusion follows the partition fusion pathway, where the membrane-docking patch grows by a zippering mechanism and then proceeds through the controlled opening and expansion of a membrane defect at the rim of the docking patch. Our results point out the fundamental role of Mfn2 in mitochondrial membrane fusion, which is essential for our deeper understanding of mitochondrial dynamics in vivo.

Materials and Methods

Chemicals.

Potassium chloride (KCl), magnesium chloride (MgCl₂), sodium chloride (NaCl) glucose, sucrose, tris(hydroxymethyl)aminomethane (Tris), 4-(2-hydroxyethyl)−1-piperazineethanesulfonic acid (HEPES), guanine 5′-triphosphate disodium salt (GTP), GDP, GMPPCP, and β-casein were supplied by Sigma-Aldrich. n-dodecyl-β-d-maltoside (DDM) was purchased from VWR. Alexa Fluor™ 555 NHS Ester, Alexa Fluor™ 555-C2 maleimide, DPX, and HPTS were acquired from Thermofisher (Molecular Probes). Ultrapure water was produced from a Milli-Q unit (Millipore, conductivity lower than 18 MΩ cm⁻¹).

Lipids.

1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl) (ammonium salt) (Rho-PE), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1,2-dipalmitoyl-sn-glycero-3-phosphothioethanol, 1,2-dioleoyl-sn-glycero- 3-phosphoethanolamine (DOPE), 1-oleoyl-2-{12-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]dodecanoyl}-sn-glycero-3-phosphoethanolamine (18:1-12:0 C12-NBD-PE), 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl) (ammonium salt) (16:0 NBD-PE) and 1-(1Z-octadecenyl)-2-oleoyl-sn-glycero-3-phosphocholine (p-PC) were purchased from Avanti Polar. 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine fluorescently labeled with ATTO 488 (ATTO-PE) was acquired from ATTO-TEC. Lipids were suspended in chloroform at 1 mg mL⁻¹ and stored at −20 °C.

Electroformation of GUVs.

GUVs were prepared using the standard electroformation protocol using indium-tin-oxide (ITO)-covered slides (78). GUVs made of POPC or POPC/DOPE (70:30 mol%) were prepared by transferring 10 µL of the lipid solution (0.5 mg/mL) on each ITO slide. Then, lipid films were rehydrated with a sucrose solution (200 mM, pH 6), and the electrodes were connected to an AC power supply (500 Hz, 1.1 V; Agilent) for at least 3 h.

Detergent-Based Reconstitution of the Mouse Mfn2 Proteins in GUVs and In Vitro Proteo-GUV Fusion Assay.

Reconstitution of mouse Mfn2 proteins into GUVs was performed by a rapid dilution below the critical micelle concentration of the protein–solubilizing detergent (cmc_DDM = 0.17 mM). This promotes the protein incorporation into the membranous environment (47). For this, 40 µL of the GUVs suspension was transferred onto a homemade eight-well observation chamber with a glass bottom surface, which was previously treated with a 10 mg/mL β-casein solution to prevent the rupture of GUVs onto the slide. Then, 0.4 µL of purified Mfn2 proteins (0.3 µg/µL) were added to the well. After gentle mixing, the suspension was incubated in the dark for 1 h at room temperature. Finally, glucose solution (200 mM) was added to reach a final volume of 240 µL. Before the addition of any nucleotide, MgCl₂ was added into the well to a final concentration of 1 mM and incubated for 2 min to achieve a homogeneous distribution (200 mM Glucose, 0.5 mM GTP, 1 mM MgCl_2, and 5 mM Tris HCl pH 7.4). In order to monitor the nucleotide-induced Mfn2-dependent membrane fusion, doublets of adhering proteo-GUVs were focused and tracked upon the addition of GDP or GTP (0.5 mM final concentration).

Fluorescence Microscopy of Proteo-GUVs.

Confocal microscopy images of proteo-GUVs were collected with a Nikon Ti-E inverted microscope equipped with a Nikon C2 confocal scanning confocal module, 488-nm and 561-nm continuous lasers, emission band-pass filters (525/ 50 and dichroic 561LP for the green and red channel, respectively) and a Nikon CFI Plan Apo DM Lambda 100× ~NA 1.45 oil immersion objective. Membrane fusion events were tracked using the Nikon Ti-E inverted microscope in wide-field fluorescence mode equipped with a Niji LED light source (Blue Box Optics Ltd) and an ultrafast Zyla 4.2 sCMOS camera (Andor Technology). Images were processed with the Fiji/Image J image analysis software (Wayne Rasband, NIH, USA).

Image Analysis.

A region of interest containing the equatorial image of adhering doublets was selected (Fig. 4D). Each vesicle was fitted to circles C_i of radii R_i (red and yellow circles in Fig. 4D) and the length L of the adhesion patch (white line in Fig. 4D) was measured as the intersection between C₁ and C₂. Considering GUVs as water-impermeable objects, we assumed constant volume, and the total area of vesicles, including the nonadhering portion and the adhesion area, was approximated to:

$$A_i=2\pi R_i^2\left(1+\text{cos}\left({\theta }_i\right)\right)+\pi {\left(\frac{L}{2}\right)}^2,$$ [1]

where θ_i are the contact angles given by

$${\theta }_i=\text{arcsin}\left(\frac{L}{2R_i}\right).$$ [2]

SUVs Preparation.

POPC, POPC/p-PC (70/30 mol%), and POPC/DOPE (70/30 mol%) SUVs were fabricated by ultrasonication. Appropriate aliquots of lipids were mixed in a vial glass and placed for 30 min in the dark under a constant flow of N₂ for solvent evaporation. The dried lipid films were then hydrated by adding 1 mL of buffer HEPES (5 mM, pH 7.4) to reach the desired final lipid concentration, vortexed for 1 min, and sonicated in an ice-water bath, to avoid heating of the sample with an ultrasonic tip sonicator for 10 min set to 30% power cycle with a pulse length time 5 min alternating 3 s on and 5 s off cycles. Membrane or volume fluorescent probes were added when needed.

Reconstitution of the Mouse Mfn2 and Mfn2-K109A Proteins in SUVs.

The protein reconstitution protocol in SUVs consisted of a two-step dilution process. Initially, 16 μL were taken from SUVs suspension (2 mM) and mixed with 20 μL of Mfn2 or Mfn2-K109 (3.5 μM) in an Eppendorf tube at 4 °C for 45 min with continuous agitation. As a result, DDM concentration (0.28 mM) was above of its CMC (cmc_DDM = 0.17 mM). Subsequently, the mixture was transferred to ultracentrifuge tubes and 500 μL of Tris buffer (50 mM, pH 7.4) was added to reduce the DDM concentration below its CMC (0.018 mM) facilitating the incorporation of the protein into the membranous environment. The suspension was then centrifuged for 30 min at 346,000 g. The pellet was resuspended in 100 μL of Tris buffer (50 mM, pH 7.4).

In Vitro Proteo-SUV Fusion Assay/Total Lipid-Dilution Fluorescence Assay.

To monitor the lipid mixing occurring between Mfn2 or Mfn2-K109A reconstituted SUVs upon incubation with GTP, two sets of proteo-SUVs (fluorescent and nonfluorescent) were mixed in a final volume of 100 µL. The ratio between fluorescent and nonfluorescent proteo-SUVs was 1:6, respectively, leading to a total lipid concentration of 2 mM. Prior to mixing, the set of fluorescent vesicles was labeled with NBD-PE and Rho-PE at a 1:6 molar ratio. At this ratio, the NBD group is quenched by the Rhodamine dye through FRET. Protein-mediated membrane fusion was followed as a function of time with a plate reader (Spectra MAX Gemini XS, Molecular Devices). GTP was added at time 0 (final concentration ~500 µM). The NBD emission was monitored at 530 nm, with the excitation wavelength set at 465 nm. Percentage of fusion was calculated after adding Triton X-100 to the vesicle suspension as $\frac{\left({\text{FI}}_{\mathrm{post} \mathrm{GTP}}-{\text{FI}}_{\text{pre} \text{GTP}}\right)}{{\text{FI}}_{\mathrm{post} \mathrm{Triton} \mathrm{X}-100}-{\text{FI}}_{\text{pre} \text{Triton} \mathrm{X}-100}}\times 100$. The fluorescence emission spectra (480 to 700 nm) were recorded for an excitation wavelength of 460 nm in Spectramax plate Gemini XS plate reader before and after 5 min incubation with 0.5 mM of GTP. The FRET efficiency before and after GTP incubation was calculated through the ratio between the acceptor emission (at 593 nm) and the donor emission (at 523 nm). Control experiments were performed using SUVs in the absence of proteins.

In Vitro Proteo-SUV Fusion Assay/Vesicle Volume-Mixing Assay.

The intermixing of the aqueous volume of vesicles upon GTP incubation was assessed with the HPTS/DPX fluorophore quencher pair. Mfn2 or Mfn2-K109A were reconstituted in SUVs composed of POPC:DOPE (70.30) at a lipid-to-protein ratio of 3,000 mol/mol. During protein reconstitution, two separate vesicle populations are loaded with 25 mM HPTS in 40 mM NaCl, 50 mM Tris (pH 7.4), or 90 mM DPX in 50 mM Tris (pH 7.4). The nonencapsulated probes were washed out with two centrifugation runs at 100,000 g for 30 min. The two sets of vesicles were then mixed in a 1:1 ratio with a final volume of 100 µL (final lipid concentration was 2 mM). GTP and MgCl₂ were added to a final concentration of 0.5 mM and 1 mM, respectively. HPTS was excited at 452 nm and its fluorescence emission intensity at 512 nm was tracked after addition of GTP. The mixing of aqueous compartments was monitored as the quenching in the HPTS fluorescence emission intensity. The fluorescence level corresponding to 100% of volume mixing was determined from SUVs entrapping 12.5 Mm HPTS, 45 mM DPX in 20 mM NaCl, 50 mM Tris (pH 7.4). All fluorimetric measurements were performed in a Spectramax plate Gemini XS plate reader.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Adhered proteo-GUVs composed of POPC and DOPE at 70:30 molar ratios, labeled with a membrane marker (DOPE-ATTO488, green) and containing Alexa555-labeled Mfn2 were followed by high-speed video microscopy after GTP addition at room temperature. Nucleotide and MgCl₂ final concentrations were 1 mM and 0.5 mM, respectively. Arrows indicate multiple fusion events occurring in the same microscope field.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.