Dynamins are a family of large GTPases that are involved in key cellular processes, where they mediate events of membrane fission and fusion. The dynamin superfamily is not restricted to eukaryotes but might have a bacterial origin, with many species containing an operon of two genes related to mitofusins. However, it is not clear whether bacterial dynamins promote membrane fission or fusion. The dynamin-like protein DynA of Bacillus subtilis is remarkable in that it arose from a gene fusion of two dynamins and contains two separate dynamin-like subunits and GTPase domains. We found that DynA exhibits strictly auto-regulated GTP hydrolysis, and that progress through the GTPase cycle is concerted within DynA oligomers. Furthermore, we show that DynA can tether membranes and mediates nucleotide-independent membrane fusion in vitro. This process merely requires magnesium as a cofactor. Our results provide a set of minimal requirements for membrane fusion by dynamin-like proteins and have mechanistic implications in particular for the fusion of mitochondria.
Dynamin GTPases also conduct central steps of mitochondrial membrane dynamics (Hoppins et al., 2007). Key mediators of these processes are Dnm1/DRP1, Mgm1/OPA1 and the fuzzy onions (Fzo1)/mitofusins, all of which are members of the dynamin superfamily. These components are conserved among eukaryotes, and mutations in some of the human homologues are cause of severe neurodegenerative disease (Knott et al., 2008). Dnm1/DRP1 is the single dynamin responsible for the fission of mitochondria, and is located at the cytoplasmic site of the outer mitochondrial membrane (Bleazard et al., 1999). Mitochondrial fusion on the other hand requires two distinct complexes of dynamins, with Fzo1/mitofusins acting on the outer membrane and Mgm1/OPA1 performing the fusion of inner membranes. Mitofusins are anchored to the mitochondrial surface via transmembrane helices (Hoppins et al., 2007) and interactions between mitofusin complexes on adjacent membranes are regulated by nucleotide (Ishihara et al., 2004; Koshiba et al., 2004). However, the mechanism of how membrane fusion is achieved following the tethering process is unknown. After outer membrane fusion is completed, inner membranes are tethered by Mgm1/OPA1, ultimately leading to their fusion (Meeusen et al., 2006).
Since the dynamin superfamily has diverse functions in eukaryotes, it is not surprising that it has emerged early in evolution (van der Bliek, 1999; Leipe et al., 2002). Indeed, many bacterial species contain genes encoding dynamin-like proteins, which are most closely related to the Fzo1/mitofusins class (van der Bliek, 1999; Leipe et al., 2002; Low and Löwe, 2006). However, a function of bacterial dynamins remains unknown. Nevertheless, detailed structural data exist. The structure of bacterial dynamin-like protein 1 (BDLP1) from Nostoc punctiforme has been solved in both its nucleotide-free and GDP-bound states (Low and Löwe, 2006), revealing a similar architecture to the distantly related human guanylate binding protein 1 (hGBP1) (Prakash et al., 2000). BDLP1 is able to assemble on the surface of liposomes as a helical polymer, deforming its membrane template into highly curved tubules and perturbing the outer lipid layer (Low et al., 2009). The polymer itself is stabilized by contacts between GTPase and stalk domains, respectively, reminiscent of interactions found in other dynamin family members (Chappie et al., 2010; Gao et al., 2010; Pawlowski, 2010).
Interestingly, many bacterial species – including N. punctiforme– contain more than one dynamin gene, with two of them often found in tandem (Fig. 1A, Fig. S1 and Table S1). The organization in an operon probably indicates close functional relationship, with some species even harbouring a fusion of the two dynamin genes containing two GTPase domains. One of the two-headed bacterial dynamins is the protein encoded by the ypbR locus of Bacillus subtilis which we term DynA (Fig. 1C). The roles of eukaryotic dynamin-like proteins in membrane remodelling processes led us to speculate whether the bacterial dynamin-like proteins might also be involved in membrane fusion or fission. Fusion of membranes in B. subtilis takes place during the late stages of cytokinesis and at completion of prespore engulfment during sporulation. Here, we focused on a biochemical analysis of the B. subtilis DynA protein and its enzymatic characteristics and show that it is capable of tethering and merging membranes in vitro.
Results and discussion
DynA is a dynamin-like GTPase
In B. subtilis, DynA is present in the membrane as a 137 kDa full-length protein and is not processed into separate dynamin subunits (Fig. 1B), suggesting that the components of bacterial two-dynamin systems might generally act in close proximity. Sequence analysis of DynA suggests that it does not harbour transmembrane segments, but rather might bind to membranes as described for the N. punctiforme BDLP1 protein (Low and Löwe, 2006). The GTP-binding domains of DynA show similarity to GTPase domains from other dynamin-like proteins such as BDLP1 and eukaryotic members of the dynamin protein family (Fig. S1 and Table S1).
To investigate its molecular properties, we expressed DynA in Escherichia coli and purified it by chromatography (Fig. S2). The protein was nucleotide-free as determined by HPLC (Fig. 2A). A central motif in the GTPase domain is the P-loop, which contains an essential lysine residue that contributes to nucleotide binding (Fig. S1 and Saraste et al., 1990). We analysed GTP-binding to recombinant DynA and its P-loop mutants using UV-cross-linking of [α-32P]-GTP (Fig. 2B). Single P-loop mutants (K65A and K625A) still bind GTP, whereas GTP-binding is fully abolished when both P-loops are mutated. Purified DynA displays GTPase activity and is inactivated by introduction of mutations into both P-loops (Fig. 2C). Interestingly, GTPase activity of the single P-loop mutants DynAK56A and DynAK625A is also abolished (Fig. 2D). It therefore seems likely that both GTPase domains of a given DynA molecule are required to bind nucleotide in order to complete the hydrolysis cycle. This model is in agreement with the finding that GTP hydrolysis of wild-type protein is cooperative with respect to substrate concentration with a K0.5 of 0.12 ± 0.02 mM, a Vmax of 3.9 ± 0.14 min−1 and a Hill constant of 2.3 ± 0.6 at 0.25 µM protein (Fig. 3A). The kinetic data suggest that hydrolysis by a given GTPase domain is sensitive to the nucleotide bound state of another GTPase domain. Because activity is not reconstituted when both single P-loop mutants are mixed (Fig. 2D), cooperativity is likely exhibited between GTPase domains of the same polypeptide and cannot be provided in trans. Consistently, truncated versions of the protein lacking either the N-terminal or C-terminal GTPase subunit are also unable to perform hydrolysis (data not shown).
In addition, we noticed a weak activation of the protein in the presence of liposomes (Fig. 3A). Since the specific activity of DynA is sensitive to protein concentration (Fig. 3B), we believe that this effect might be due to accumulation of DynA on the membrane surface. In summary, DynA displays two modes of cooperativity: First, there is intramolecular cooperativity between the two GTPase domains of a given DynA molecule, and second there is intermolecular cooperativity in which DynA stimulates GTP hydrolysis of partner molecules.
GTPase subunits of DynA show self-interaction symmetry
As predicted by the kinetic analysis, we indeed observed self-interaction of the protein. DynA forms dimers in the nucleotide-free state as determined by size exclusion chromatography (Fig. 3C). Truncated versions of DynA, lacking either the C-terminal dynamin subunit (DynAD1, 71.2 kDa) or the N-terminal subunit (DynAD2, 74.4 kDa), are also able to oligomerize, with DynAD2 forming dimers and DynAD1 forming dimers and probably tetramers (Fig. 3C). Therefore, both subunits of DynA can provide homotypic contacts for complex formation.
To test whether self-interaction might be influenced by GTPase activity, we constructed fusions of P-loop mutants to Bordetella pertussis adenylate cyclase T18 and T25 fragments and performed two-hybrid analysis in E. coli (Fig. 3D). The assay reported interaction signals for full-length proteins and also displays homotypic interactions for truncated D1 and D2 subunits, which is consistent with the observations obtained by gel filtration. Since the nucleotide binding and hydrolysis assays indicate that GTPase defective mutants of DynA are trapped in different stages of nucleotide loading, they may therefore provide a crude stop motion picture through this process. In the two-hybrid assay, P-loop mutants gave most intense signals when self-interaction was tested (visible on the diagonal from upper left to lower right in the matrix and also reported by interactions of the truncated versions), indicating that the probability of complex formation is highest when both proteins are in a symmetrical GTP bound state. Put differently, stability of the complex might be low when the molecules are not in the same state. Since heterotypic interactions between T18 and T25 constructs compete with homotypic interactions between T18 or T25 tagged molecules that do not reconstitute adenylate cyclase activity, the two-hybrid assay is sensitive to relative expression levels. T18 constructs were expressed from a high-copy vector, whereas T25 fusions were expressed from a low-copy plasmid. Wild-type DynA generally gave strong signals when it was expressed from the high-copy vector and tested against constructs expressed from the low-copy vector. Signals were drastically reduced when the expression ratio was reversed. We believe that this reflects the ability of wild-type DynA to complete its hydrolysis cycle and to enter all nucleotide bound states, enabling it to efficiently form complexes with each of the mutant proteins when it is in excess. The generally increased signal intensity of the T18 wild-type construct might be explained by a low stability of the hydrolytically active complex. This would minimize unproductive interactions between T18-tagged constructs and shift the equilibrium towards the formation of more stable, hydrolytically inactive complexes with adenylate cyclase activity. It should be noted, however, that signals reported by the two-hybrid assay might also partially reflect conformational changes other than changes in oligomerization.
DynA is able to tether membranes via its D1 subunit
In contrast to N. punctiforme BDLP1 which shows nucleotide modulated membrane affinity (Low et al., 2009), DynA displayed nucleotide-independent membrane binding in vitro (Fig. 4A). Also in live B. subtilis, GFP-tagged DynA stayed associated with the cell periphery when nucleotide binding was disrupted (Fig. 4C). Since we additionally introduced a dynA deletion it can be excluded that mutant DynA–GFP is recruited to the membrane by wild-type protein. To investigate the contributions of the N- and C-terminal subunits to membrane binding, we expressed GFP fusions of DynAD1 or DynAD2. Only DynAD1–GFP was found to be membrane associated, whereas DynAD2–GFP displayed cytoplasmic localization (Fig. 4C). This membrane binding behaviour was also observed for purified components. DynA and DynAD1 co-sedimented with liposomes, whereas DynAD2 did not (Fig. 4B). It should be noted, however, that although DynAD1 was clearly shifted to the pellet fraction in the presence of liposomes, it also partially sedimented in the absence of liposomes. The purified construct is therefore more likely reflecting membrane binding in a rather qualitative than quantitative way. Binding to the membrane surface was confirmed by electron microscopy, which showed ordered self-assembly of DynA in the absence of nucleotide (Fig. 5).
Strikingly, full-length DynA was not only able to bind to liposomes, but also tethered them into large clusters as was readily observed by light microscopy and turbidimetry (Fig. 6B and C). The D1 subunit was required and sufficient for this effect (Fig. 6C). In addition to this, we also observed the formation of extensive tethering zones between adjacent cellular compartments when DynA–GFP was expressed in yeast (Fig. 6A). Since D1 seems to be the membrane binding subunit and was sufficient for tethering, the DynA oligomer is likely in a conformation, which is able to cross-link opposing membranes via separate D1 subunits.
Because DynA was able to act as an efficient bilayer tether, we asked if it was also able to promote membrane fusion. We performed lipid mixing assays with 4-nitrobenzo-2-oxa-1,3-diazole (NBD) and rhodamine head-group-labelled lipids (Struck et al., 1981). These fluorophors form a Förster resonance energy transfer (FRET) pair, which is diluted into the membrane of unlabelled acceptor liposomes upon membrane fusion. This results in impaired FRET efficiency, which can be monitored by changes in NBD fluorescence. Surprisingly, we observed extensive DynA-mediated liposome fusion in the absence of nucleotide but dependent on the presence of magnesium ions (Fig. 6D). Fusion did not take place in the presence of magnesium when protein was omitted from the reaction, but also occurred when DynA was substituted for DynAK56A/K625A or its D1 subunit (Fig. S3A). This shows that the fusion process is not driven by a nucleotide contamination of the lipid preparation, but is truly independent of nucleotide, and that the membrane binding and tethering subunit of DynA is sufficient for the process. In addition to this, we did not observe alterations in tethering or fusion behaviour when GTP was added to the reaction (Fig. S3B and C).
We noticed that liposome complexes did not undergo significant shape transitions during the fusion process, but stayed in the rather compact form also seen without magnesium. This shape might be generated by DynA coating. To visualize fusion products, we therefore treated fused liposome complexes with proteinase K to remove the DynA coat. Strikingly, under these conditions large vesicular fusion products could be observed (Fig. 6E), showing that DynA catalyses complete fusion of both membrane leaflets.
Since free intracellular magnesium has been estimated to be in the low millimolar range in bacteria (Alatossava et al., 1985), the DynA fusion reaction may operate efficiently in a cellular environment. This is reminiscent of the magnesium-assisted membrane fusion of plant Golgi membranes, which has been shown to depend on an unidentified protein factor (Takeda and Kasamo, 2002). Indeed, magnesium ions are known to facilitate membrane fusion in areas of high membrane curvature and proximity (Wilschut et al., 1981). This effect might be explained by reduced charge repulsion between bilayers or an influence on lipid phase behaviour. Membrane proximity and potentially also curvature might be provided by DynA to overcome the kinetic barrier of membrane fusion. Another possibility we cannot yet exclude is that magnesium might bind to DynA directly and trigger the transition to a fusogenic conformation.
Our findings are likely relevant for fusion processes mediated by other dynamin-like proteins, like the fusion of mitochondrial membranes. For these processes, GTP hydrolysis might not be needed to energize fusion by means of mechanical force, but probably plays a regulatory role that determines if and how a dynamin complex tethers membranes and enters its fusogenic state. In their fusogenic state, dynamin-like proteins might then merely act as passive catalysts, promoting fusion simply by lowering the activation energy of bilayer merging. DynA might be an example for a dynamin-like protein that is – at least under the conditions used in this study – always in its active state.
DynA localizes to the sites of septation
In B. subtilis, expression of DynA–GFP under control of the native promoter showed a preferred localization of DynA to the sites of septation (Fig. 7A and B). In order to corroborate the localization results we analysed the localization of DynA–GFP in absence of the division protein MinJ. We have chosen MinJ because it plays an important role in the mature divisome and makes protein–protein contacts to many membrane integral and membrane-associated division proteins (Bramkamp et al., 2008; van Baarle and Bramkamp, 2010). Strikingly, we observed a dramatic change in DynA localization in a ΔminJ strain background, showing a dispersed DynA localization along the entire cell membrane (Fig. 7C). Thus, DynA is very likely localized to the sites of septation making essential protein–protein contact with other cytokinetic proteins. These results might point to a role of bacterial dynamin-related proteins in cytokinesis, but since we did not observe a morphological phenotype for the ΔdynA strain, either surrogate systems might be expressed or bacterial dynamins might be important under specific environmental conditions. Of course at this stage we cannot entirely rule out that bacterial dynamins may be involved in other cellular functions.
In addition to their homology to mitofusins, structural arguments (Low et al., 2009) and our experimental findings suggest that bacterial dynamins might be involved in a membrane fusion process, maybe taking place at the sites of septation. In vitro membrane fusion mediated by DynA, however, is in contrast to that performed by atlastins or mitofusins for which GTP hydrolysis is a prerequisite (Meeusen et al., 2004; Orso et al., 2009). This suggests that GTP hydrolysis is not generally needed to energize a dynamin-mediated fusion reaction and points to the existence of an ion-supported lipid mixing step. Our data suggest that dynamin-like proteins might not act as molecular machines but may act as regulated fusion catalysts. These molecules might have a fusogenic conformation, which can be switched on or off. Due to the evolutionary relationship between bacteria and mitochondria and the homology of bacterial dynamins to mitofusins, we think that these findings might be of particular relevance to the mechanism of mitochondrial fusion.
Although we did not observe a regulation of DynA-mediated fusion in vitro, we think that this might likely be the case in vivo due to the kinetic and self-interactive properties of the molecule. Since membrane binding and self-interaction are not operating uniformly in the dynamin superfamily, these properties are probably adapted to specific cellular roles. Fusion activity may depend on the membrane architecture and composition prevalent in the dynamin's cellular environment. Factors determining when, if and where a dynamin enters its fusogenic state might therefore define its function.
Cloning and purification of DynA
The dynA gene was cloned from B. subtilis 168 and inserted by NcoI and XhoI into pET16b (Novagen) with a C-terminal His6 tag. Expression was performed overnight at 18°C with E. coli BL21(DE3) in Luria–Bertani (LB) with 0.5 mM IPTG and 50 µg ml−1 carbenicillin. Cells were disrupted in 50 mM Tris, 200 mM NaCl, 20 mM imidazole, 10% glycerol, 1% Triton X-100, pH 8.0/4°C, and the protein was bound to Ni-NTA agarose (Qiagen). After extensive washing with 50 mM Tris, 500 mM NaCl, 20 mM imidazole, 10% glycerol, pH 8.0/4°C, protein was eluted in 50 mM Tris, 500 mM NaCl, 1 M imidazole, 10% glycerol, pH 8.0/4°C, reduced with 1 mM DTT and gel filtrated on Superdex 200 against 50 mM Tris, 500 mM NaCl, 10% glycerol, pH 8.0/4°C. Purification of DynAD1 (residues 1–609) and DynAD2 (residues 561–1193) was analogous, except that lysis buffer of DynAD1 contained 500 mM NaCl. All constructs contained an additional glycine in position 2 and a C-terminal GSS linker.
Bacterial two-hybrid analysis
For two-hybrid analysis, a system based on reconstitution of adenylate cyclase activity was used (Karimova et al., 2005). Sequences were cloned into pUT18C and pKT25 and transformed into E. coli BTH101. Cells were grown on LB with 160 mg ml−1 X-Gal, 0.5 mM IPTG, 50 µg ml−1 kanamycin and 100 µg ml−1 carbenicillin at 30°C.
Generation of DynA–GFP strains and microscopy
A list of strains can be found in Table S2. For genomic deletion of dynA, sequences approximately 250 bp up- and downstream of the gene were fused to a tet cassette via overlap extension PCR (Heckman and Pease, 2007). The construct was adenylated with Taq and cloned into pDRIVE (Qiagen). The targeting vector was linearized with ClaI and transformed into B. subtilis. For introduction of GFP-fusions into B. subtilis, sequences were cloned into pSG1154 via KpnI and XhoI and inserted into the amyE locus. Cells were induced in LB with 0% (DynA–GFP), 0.25% (DynAD1–GFP) or 0.5% xylose (DynAD2–GFP). The construct for expression of DynA–GFP under control of its native promoter was constructed by cloning the dynA gene including the promoter region into pSG1154. Expression from the pSG1154 intrinsic Pxyl promoter was blocked by a transcriptional terminator upstream of PdynA. A strain expressing DynA–GFP in absence of MinJ was constructed by transformation with chromosomal DNA of strain 3865 (Bramkamp et al., 2008). Expression of GFP fusions in S. cerevisiae W303-1A(a) used the pYX223 vector and was performed in SC-His with 2% galactose. Vacuolar staining with FM4-64 was performed as described (Baars et al., 2007). Microscopy was performed on a Zeiss AxioImager M1 equipped with an EC Plan-Neofluar 100x/1.3 Oil Ph3 objective and a Zeiss AxioCam HRm camera. Green fluorescence (GFP, NBD) was monitored using filter set 38 HE eGFP, and red fluorescence (FM4-64, Nile red) was monitored by using filter 43 HE Cy3.
GTPase activity measurement
GTPase measurements were performed at 37°C in 50 mM Tris, 200 mM NaCl, 5 mM MgSO4, 10% glycerol, pH 7.1/37°C by a coupled enzyme assay and corrected for spontaneous GTP hydrolysis (Ingerman et al., 2005). Briefly, the indicated protein and GTP concentrations were incubated in the presence of 1 mM PEP, 0.6 mM NADH, 20 U ml−1 pyruvate kinase and 20 U ml−1 lactate dehydrogenase and a change in NADH concentration was monitored by absorbance at 340 nm.
UV-reactive cross-linking of [α-32P]-GTP
The GTP binding assay was performed as described before (Yue and Schimmel, 1977). Samples of 20 µl contained 3 µg protein, 1 µCi [α-32P]-GTP in 50 mM Tris, 200 mM NaCl, 5 mM MgSO4, 10% glycerol, pH 8.0/4°C. If required, 1 mM of unlabelled (cold) GTP was added. After incubation on ice for 10 min, UV irradiation was carried out at room temperature for 10 min at 0.1 J cm−2 in a UV cross-linker (LTF Labortechnik). The samples were made up with 4 × SDS sample buffer and loaded onto an SDS-polyacrylamide gel (Tris/glycine 10% acrylamide gels were used throughout the study). Gels were stained for protein, dried under vacuum and exposed onto an imager plate (Raytest) for 2 days. The plate was scanned in a phosphorimager (Fujifilm BAS-1800) and images were processed using AIDA Image Analyzer software.
Liposome sedimentation assay
Escherichia coli lipids (Avanti Polar Lipids) dissolved in chloroform were dried under nitrogen, exsiccated for 1 h and swollen at 37°C in 50 mM Tris, 10% glycerol, pH 7.1/37°C. The suspension was submitted to at least five freeze and thaw cycles, diluted to 2 mg ml−1 in assay buffer and extruded through a 400 nm pore size filter. Then 2 µM protein was incubated for 20 min at 25°C with 1 mg ml−1 liposomes in 50 mM Tris, 200 mM NaCl, 10% glycerol, pH 7.4/25°C, and the sample was fractionated by ultracentrifugation. Where indicated nucleotides were used at 1 mM with 5 mM MgSO4.
An overnight culture of B. subtilis was diluted 1:50 into LB and grown for 2 h at 37°C. Cells were harvested by centrifugation, resuspended in 50 mM Tris, 200 mM NaCl, 10% glycerol, pH 8.0/4°C and lysed in a tissue homogenizer. Cell debris was removed by centrifugation at 12 000 g for 10 min. Membranes were collected by centrifugation in a TLA 120.2 rotor at 80 000 r.p.m. for 20 min.
Liposome tethering and fusion assays
Liposome tethering was observed at 20°C by sample turbidity changes at 350 nm with 0.2 µM protein and 0.2 mg ml−1 liposomes in 50 mM Tris, 200 mM NaCl, 10% glycerol, pH 7.5/20°C. Liposome fusion was assayed at 37°C with 0.18 mg ml−1 unlabelled acceptor liposomes and 0.02 mg ml−1 donor liposomes labelled with 1.7% (w/w) NBD-PE and 2.4% (w/w) rhodamine-PE. NBD fluorescence was monitored with λex = 460 nm and λem = 538 nm, and fluorescence change (F − F0) was normalized to values after addition of 1% sodium dodecyl sulphate (Ff − F0). For visualization of fusion products, unlabelled liposomes were fused in the presence of 2 µM DynA and 5 mM MgSO4 and subsequently treated with 100 µg ml−1 proteinase K. Products were stained with 10 µg ml−1 Nile red immediately before imaging.
The 0.2 mg ml−1 preformed 400 nm liposomes made from E. coli total lipids (Avanti) were incubated with 25 µM DynA for 10 min at 37°C. Liposomes were subsequently negatively stained with uranyl acetate. The images were taken on a Philips CM100 Compustage Transmission Electron Microscope at the Newcastle EM Research Services.
Antibodies against DynAD2 were raised in rabbits and affinity purified with DynAD2 coupled to CNBr sepharose. They were used at 1:3000 for Western blot analysis.
We want to thank Gerrit Praefcke and Julia Fres (University of Cologne) for constructive criticism on the manuscript and for their help with HPLC measurements. We thank Reinhard Krämer (University of Cologne) for continuous support and the staff of the Newcastle Research Services (Newcastle University, UK) for help with electron microscopy.