|Home | About | Journals | Submit | Contact Us | Français|
Numerous vesiculation processes throughout the eukaryotic cell are dependant on the protein dynamin, a large GTPase that constricts lipid bilayers. We have combined x-ray crystallography and cryo-electron microscopy (cryo-EM) data to generate a coherent model of dynamin-mediated membrane constriction. X-ray structures of mammalian GTPase and pleckstrin homology (PH) domains of dynamin were fit to cryo-EM structures of human ΔPRD dynamin helices bound to lipid in non-constricted and constricted states. Proteolysis and immunogold labeling experiments confirm the topology of dynamin domains predicted from the helical arrays. Based on the fitting, an observed twisting motion of the GTPase, middle and GTPase-effector domains coincides with conformational changes determined by cryo-EM. We propose a corkscrew model for dynamin constriction based on these motions and predict regions of sequence important for dynamin function as potential targets for future mutagenic and structural studies.
The formation of transport vesicles during receptor-mediated endocytosis, caveolae internalization, and clathrin-mediated membrane trafficking from the Golgi and recycling endosomes requires the large GTPase dynamin (Hinshaw, 2000; Praefcke and McMahon, 2004). During vesiculation, dynamin wraps around the necks of invaginating pits and plays an active role in the final stages of vesicle fission. In support of this model, dynamin, which exists as a tetramer in solution (Binns et al., 1999; Muhlberg et al., 1997), has been shown to polymerize into large oligomeric spirals (Carr and Hinshaw, 1997; Hinshaw and Schmid, 1995) that are representative of the dynamin structures observed at the necks of budding vesicles (Evergren et al., 2004; Iversen et al., 2003; Takei et al., 1995). Furthermore, in the presence of negatively charged lipid, dynamin self-assembles into helical tubes that constrict upon addition of GTP (Danino et al., 2004; Sweitzer and Hinshaw, 1998). This conformational change is believed to mimic the constriction at the necks of coated pits during endocytosis – a constriction that may lead to membrane fission contingent on the environment of the pit (Roux et al., 2006).
The GTPase domain of dynamin is conserved among all dynamin family members and residues within G-domains (Figure 1B; yellow boxes), which form the nucleotide-binding site, are conserved between dynamin and other GTPases including Ras. Mutational analyses have targeted conserved residues known to be functionally significant based on homology to other GTPases. Specifically, mutating K44, S45 and T65 cause a negative effect on dynamin function due to defects in GTP binding and/or hydrolysis (Damke et al., 2001; Herskovits et al., 1993; Marks et al., 2001; Song et al., 2004a; van der Bliek et al., 1993). Temperature-sensitive (ts) mutants have also been localized to the GTPase domain of a homologous dynamin gene product, shibire, in Drosophila (Chen et al., 1991; van der Bliek and Meyerowitz, 1991). At non-permissive temperatures, shibire ts mutants are defective in vesicle release, and a dynamin collar is observed at the necks of accumulating coated pits (Koenig and Ikeda, 1989). Equivalent mutations in humans also exhibit a defect in endocytosis and GTPase activity at non-permissive temperatures (Damke et al., 1995; Narayanan et al., 2005). In addition to mutation studies, x-ray structures of GTPase domains from Dictyostelium discoideum dynamin A (dyn A) (Niemann et al., 2001) and Rattus norvegicus dynamin 1 (Reubold et al., 2005) demonstrate that the core architecture of the dynamin GTPase domain is conserved when compared with other GTPase structures.
Along with the GTPase domain, dynamin contains a middle domain, a pleckstrin homoiogy (PH) domain, a GTPase effector domain (GED) and a C-terminal proline-rich domain (PRD) (Figure 1, A & B). The middle domain and GED promote self-assembly, while the PH domain and PRD target dynamin to sites of vesicle scission. The PRD has been shown to interact with Src homology 3 (SH3) domains in proteins involved in endocytosis, which recruit dynamin to sites of action and may modulate dynamin activity (Schmid et al., 1998). Most dynamin-related proteins lack the PH and PRD domains, so the roles of these domains are specific for dynamin.
The PH domain is essential for interactions with lipid bilayers (Salim et al., 1996). Based on x-ray structures (Ferguson et al., 1994; Timm et al., 1994), the core folds of the dynamin PH domain are conserved when compared to PH domains from other proteins. Complementary NMR studies of the dynamin PH domain reveal dynamic motions in variable loops (Figure 1B; red boxes) that interact with lipid headgroups (Fushman et al., 1995). During self-assembly of dynamin onto lipid surfaces, the variable loops preferentially bind negatively charged bilayers (Zheng et al., 1996), which stimulates GTP hydrolysis (Tuma et al., 1993). Mutations in these loops result in a decrease of lipid binding, assembly-stimulated GTPase activity and endocytosis (Lee et al., 1999; Vallis et al., 1999), demonstrating that dynamin-lipid interactions are essential. Furthermore, mutations in the PH domain of dynamin-2, the ubiquitously expressed isoform, lead to a human peripheral neuropathy, Charcot-Marie-Tooth disease (Zuchner et al., 2005).
The middle domain of dynamin not only links the GTPase and PH domains, but also plays a role in regulating self-assembly (Smirnova et al., 1999). In vitro mutation studies of dynamin (Ramachandran et al., 2006) and a dynamin-related protein in yeast (DNM1) (Ingerman et al., 2005) show that the middle domain is required for efficient self-assembly into higher ordered structures. In addition, missense mutations in the middle domain of dynamin-2 are linked to another human disease, autosomal dominant centronuclear myopathy (Bitoun et al., 2005).
Unlike other GTPases that require a GTPase activator protein (GAP), dynamin’s GED promotes self-assembly and stimulates its own GTPase activity (Song et al., 2004b). Addition of isolated GED to unassembled dynamin stimulates GTPase activity in a highly cooperative manner (Sever et al., 2000), and suppressor mutants of a temperature-sensitive mutation in the GTPase domain (Shibire ts2) were identified in the GED, confirming its role in regulating GTPase activity (Narayanan et al., 2005). However, in addition to the self-regulatory properties of dynamin, phospholipase D has been proposed to act as an external GAP for dynamin and accelerates endocytosis of epidermal growth factor receptor (Lee et al., 2006).
Cryo-electron microscopy (cryo-EM) has been useful for elucidating structural features of dynamin-1, the neuronal-specific isoform (Zhang and Hinshaw, 2001). Three-dimensional density maps of a dynamin mutant, lacking its C-terminal proline-rich domain (ΔPRD), in the constricted and non-constricted states reveal a T-shaped subunit consisting of three prominent radial densities: inner, middle and outer (Figure 1C) (Chen et al., 2004; Zhang and Hinshaw, 2001). Manual docking of crystal structures for the GTPase domain from human guanylate-binding protein (a distantly related dynamin family member) and the PH domain from human dynamin-1 to the constricted ΔPRD dynamin tube suggested that these domains reside in the outer and inner radial densities, respectively (Zhang and Hinshaw, 2001).
In this study, we predict conformational changes that occur during constriction using real-space refinement methods. This automated technique places crystal structures of mammalian GTPase and PH domains from dynamin in both the constricted and non-constricted cryo-EM maps of ΔPRD dynamin. We define conformational motions of the crystal structures in the outer and inner radial densities that lead to the observed constriction. Furthermore, proteolysis and immunogold labeling studies verify the positions of dynamin structural domains observed in our fittings. From these results, we can predict interactions between and motions in the middle domain and GED based on topologic restraints. Overall, our fittings provide a model where repeating subunits in the ΔPRD dynamin helical array undergo a corkscrew motion during constriction consistent with conformational changes observed experimentally.
Dynamin readily forms ordered tubes in the presence of negatively charged liposomes. Both full-length and ΔPRD dynamin tubes constrict and twist upon GTP hydrolysis, however, only ΔPRD dynamin constricts in the presence of non-hydrolysable GTP analogs. Three-dimensional reconstructions of ΔPRD dynamin tubes in the constricted and non-constricted states reveal the conformational changes that occur upon GTP binding (Chen et al., 2004). Specifically, the middle radial density (Figure 1C) undergoes a dramatic rearrangement such that this region becomes highly kinked in the constricted state (compare red lines in Figure 2, C & D). The outer and inner radial densities (Figure 1C) also rearrange during constriction, and until now it remained unclear what these conformational changes represent at the molecular level. To address this question, the atomic structures of the rat dynamin GTPase (99% identical to human, Reubold et al., 2005) and human dynamin-1 PH (Timm et al., 1994) domains were fit to the cryo-EM reconstructions of human dynamin-1 (Chen et al., 2004) with a molecular modeling program, YAMMP (Tan et al., 1993), using rigid body Monte Carlo with simulated annealing.
The crystal structures are fit to the cryo-EM maps using the vector lattice (VLAT) component of YAMMP, which defines the electron density as a 3-D potential. To sample possible orientations, a reduced-representation (Cα atoms) GTPase domain was iteratively moved at random as a rigid unit using Monte Carlo with simulated annealing. The score of the fitting, based on the VLAT term, improves until an orientation that best matches the experimental data is found. The GTPase domain was initially placed at an arbitrary location relative to the cryo-EM map. After several iterations of rigid body refinement, the structure was determined to have the best fit in the outer radial density of the cryo-EM maps. A reasonable fit was not found in the density near the lipid bilayer or in the middle radial density.
The reconstruction of ΔPRD dynamin tubes in the nucleotide-bound (constricted) state identified 13.2 subunits per turn of the helix (Chen et al., 2004). Scanning transmission electron microscopy (STEM) analysis of dynamin tubes revealed ~30 dynamin molecules per turn (Zhang and Hinshaw, 2001), suggesting each repeating subunit contains a dimer. This dimer is evident when the density is viewed at higher thresholds (yellow density boxed in Figure 2, C & D) and appears to be asymmetric. Two GTPase monomers were initially fit to the asymmetric repeat in the density (one of the 13 equivalent subunits) using rigid body Monte Carlo with simulated annealing. The best fit was determined for each monomer in the dimer pair, and 13 dimers (26 GTPase monomers) with this configuration were then fit to a complete turn of the constricted ΔPRD dynamin helix (Figure 2; green ribbons). The final model provides the optimal fit of dynamin GTPase domains to the helical density in the constricted state (Figures 2, B & D).
We used a similar method to define the best fit of the GTPase domain to the non-constricted helical density, which contains 14.2 repeating subunits per turn corresponding to ~28 dynamin monomers. Briefly, we again fit two monomers to a repeating subunit of the helix using rigid body refinement. To prevent bias, GTPase monomers were placed at initial orientations distinct from the dimer orientation determined for the constricted state. The best orientation for the monomers in the repeating subunit was determined after several rounds of refinement, and from this model, 14 dimer subunit orientations were then fit to one turn of the non-constricted helix (Figures 2, A & C; 28 GTPase monomers).
In both fittings, the final placements for the GTPase domain position the N- and C-terminal helices toward the cleft of the dynamin structure (Figure 3; light and dark purple helices) where they form a hydrophobic cleft that has been proposed to bind the GED (Niemann et al., 2001). In addition, the Shibire ts1 mutation (G273D) resides immediately upstream of the C-terminal helix of the GTPase domain (Figure 3; red spheres). The C-terminal helix is also in an orientation where contiguous sequence would continue towards the middle radial density. This is consistent with the placement of the middle domain within the middle radial density.
For comparison, a nearly identical placement of the GTPase domain was observed when the crystal structure from Dictyostelium dyn A (60% identity to human, Niemann et al., 2001) was fit to the density (Figure 3, insert). Dyn A, which is involved in mitochondrial division, has additional sequence between the G2 and G3 regions of the GTPase when compared with rat and human dynamin. Not surprisingly, this additional loop (circled in red) cannot be accommodated when fit to the ΔPRD dynamin structure and protrudes away from the density.
The proximity of GTPase monomers due to helical packing are close enough to allow intermolecular interactions. Specifically, adjacent dynamin GTPases along the ridge of outer radial density are in a position to interact with one another, while connections across the cleft of the helical array are likely maintained via middle-GED interactions (Figure 4, A & B). Based on the fittings of the rat GTPase crystal structure to the constricted and non-constricted densities, the interface within the GTPase dimer remains largely unchanged (Figure 4C; Interface #1). This interface is comprised of a highly conserved sequence near the switch 2 region (Figure 4, A & B; gold ribbon), and the Shibire ts2 mutation (G146, colored orange in Figure 4C) resides near this interface (Narayanan et al., 2005).
While the dimer interface is preserved after constriction, differences between adjacent GTPase dimers are apparent in the outer radial density (Figure 4C, Interface #2). In the non-constricted state, GTPase dimers are further away from each other than in the constricted state (Figure 4C; green vs. blue ribbons). Upon nudeotide binding, the dimers undergo a subtle rotation (~10°) that allows the subunits to pack more closely when constricted. Notably, the sequence at interface #2 is unique to dynamin family members (Figure 4, A & B; blue and red ribbon, and also underlined in Figure 1B), and the amino acids in this region are more charged when compared with the rest of the GTPase sequence. The Shibire ts1 mutation, G273D (Figure 4C; Cα colored red), is found adjacent to this interface and has been proposed to uncouple the GTPase domain from the C-terminal half of dynamin (Damke et al., 1995). The position of this conserved glycine is proximal to the proposed GED binding site (the aforementioned hydrophobic cleft), and therefore may act as a hinge between two sites where GTPase-GTPase and GTPase-GED interactions may sense nucleotide binding and drive helical constriction, respectively.
The human dynamin-1 PH domain crystal structure (Ferguson et al., 1994) was fit to the cryo-EM density in both the non-constricted and constricted states using rigid body Monte Carlo with simulated annealing (Figure 2, E & F). In the non-constricted map, the molecular fitting places the PH domain in the inner radial density at the lipid bilayer interface. The density in this region is well defined with a contour that matches the shape of the X-ray structure, where the density is more tapered near the middle radial density and is wider near the lipid interface. Therefore, the variable loops are positioned near the lipid bilayer with the N- and C-terminal ends adjacent to the middle radial density (Figure 3A). This N-terminal orientation is consistent with placing the middle domain in the middle radial density between the GTPase and PH domain structures. The C-terminal sequence of the PH domain likely continues into the middle radial density as well, allowing the GED to interact with the middle and GTPase domains as predicted previously (Narayanan et al., 2005; Ramachandran et al., 2006; Sever et al., 2000; Smirnova et al., 1999; Song et al., 2004b). The inner radial density for the constricted map is weaker when compared to the non-constricted map, but a similar contour exists (Figure 3B). Therefore, the best fit to the constricted density is also near the lipid interface and places the N- and C-terminal ends near the middle radial density and the variable loops near the lipid bilayer (Figure 3B). Again, the starting orientations for the refinement were distinct from those determined for the non-constricted map.
The relative arrangements of the PH domains in the non-constricted and constricted maps shift during constriction (Figure 2, E & F; dashed boxes), and this motion coincides with the conformational changes observed in the outer and middle radial densities. The PH structures are also closer together in the constricted state as expected based on the tighter packing. However, the domains appear as distinct densities that do not interact with one another. Therefore, PH domain interactions are limited to the membrane, and any motions during constriction are dictated by interactions between the GTPase, middle and GED domains.
The variable loops in the PH domain interact with the lipid headgroups (Burger et al., 2000) and are dynamic, as shown by NMR (Fushman et al., 1995). Our fittings place the variable loops (VL1, VL2 and VL3) near the lipid bilayer, in a position where they can interact with negatively charged lipids. Loop VL1 buries deeper into the bilayer when compared with the other loops (Figure 3; red). This loop contains residue K535, which when mutated to an alanine destroys interactions with negatively charged lipids and inhibits endocytosis (Vallis et al., 1999). VL2 has been identified as a binding site for acidic phospholipids (Zheng et al., 1996) and contains residues associated with Charcot-Marie-Tooth disease (Zuchner et al., 2005). Inherent flexibility of this loop would allow VL2 to insert into the lipid bilayer in a manner similar to VL1. VL3 is near the lipid bilayer in our model, and mutations in this loop also affect lipid interactions (Klein et al., 1998).
Based on the fittings of the GTPase and PH domains to the cryo-EM helical densities, we are able to predict interactions between, and accessibility of, different regions in dynamin. Therefore the model was tested using limited proteolysis of dynamin in the presence and absence of lipid (Figure 5A). After trypsin digestion, a significant portion of dynamin bound to lipid is cleaved at its C-terminal end, removing the PRD (Figure 5; ΔPRD). It is well known that the PRD is readily accessible to proteases when dynamin is free in solution (Carr and Hinshaw, 1997; Muhlberg et al., 1997). Cleavage of the PRD in the presence of lipid further suggests that this domain is peripherally located after self-assembly and is therefore in a position to interact with other endocytic factors.
When dynamin exists as a monomer/tetramer in solution, trypsin proteolysis occurs at residue 393 in the middle domain as determined by mass spectrophotometry (Figure 5; Fragment 3). This cleavage site is accessible to trypsin even though this region is essential for dynamin tetramer assembly (Ramachandran et al., 2006). In the presence of lipid, the middle domain tryptic site (residue 393) is protected and cleavage is seen at residue 629, which is near the PH-GED boundary (Figure 5; Fragment 1). The protection in the presence of lipid coincides with strong middle domain interactions with the GED. Regardless of whether dynamin is bound to lipid or not, another population of digested fragments is cleaved at residue 465, near the C-terminal end of the middle domain (Figure 5; Fragment 2). The cryo-EM density connecting the middle and inner radial densities is relatively weak, which suggests that an extended, flexible region connects these two domains and would be accessible to protease.
In addition to proteolysis protection, we used immunogold labeling to determine the accessibility of the GTPase domain of dynamin after self-assembly in the presence of lipid. A polyclonal antibody that targets amino acids 223–248 (MC65) of dynamin 1 (Henley et al., 1998) labeled to the periphery of the dynamin helical array confirming the fitting of the GTPase domain in the outer radial density (Figure 6A). In addition to labeling dynamin tubes, background dynamin monomers and tetramers were also labeled (Figure 6B). The labeling on the tubes was 24.1 ± 5.8 gold particles per 0.01 µm², while only 6.4 ± 2.1 gold particles per 0.01 µm² were counted for the background dynamin monomers and tetramers. The background gold labeling is specific to dynamin since it is equivalent to labeling observed when dynamin monomers and tetramers alone (no liposomes added) are applied to the grids (7.0 ± 0.9 particles per 0.01 µm²). Furthermore, 2° antibody added alone did not result in significant labeling of dynamin monomers, tetramers or tubes (Figure 6C; 0.05 ± 0.04 particles per µm²) and MC65 labeling was not observed with BSA alone (not shown).
The ability of dynamin to assemble into helices and constrict the underlying lipid bilayer is essential for vesiculation events in the cell. This study provides new insight into the mechanism of membrane vesiculation by examining conformational changes apparent in ΔPRD dynamin upon GTP addition. As observed previously, when comparing the non-constricted and constricted helical structures, the decrease in the number of subunits per turn (14.2 to 13.2, giving a ratio of 1.07) does not account for the change in circumference (a ratio of 1.25, 50 nm * π divided by 40 nm * π). If dynamin acted as a simple sliding ratchet upon constriction, 11.3 (14.2 divided by 1.25) subunits would be found in one turn of the constricted ΔPRD dynamin helix. Therefore, conformational changes within and between repeating subunits in the helical arrays of ΔPRD dynamin occur during constriction.
The computational fittings described here confirm the initial prediction placing the GTPase domain in the outer radial density, which is also confirmed by immunogold labeling, and the PH domain in the inner radial density. Moreover, the current fittings determine the packing within the repeating dimer subunit and the relative orientations of multiple dimers within the helical array. The remaining two domains, middle and GED, are positioned in the middle radial density, which undergoes the most dramatic conformational change upon helical constriction (red line in Figure 2C, 2D, and Figure 7). Here, we show that upon assembly, a region of the middle domain of dynamin is protected from proteolysis due to interactions in the helical array. This finding confirms the tight packing proposed for the middle domain and GED in the helical structure when lipid is present. The site of protection is in an area that recently has been shown to be essential for dynamin tetramer formation in solution (Ramachandran et al., 2006). Protection of the middle domain from cleavage in the dynamin lipid tube suggests that a conformational rearrangement occurs upon lipid binding and/or assembly of the helical array.
The self-assembly of dynamin results in enhanced stability within or between adjacent GTPase domains, which may lead to more efficient hydrolysis within higher-ordered structures. In support of this hypothesis, the switch 2 region of the dynamin GTPase is adjacent to interface #1 (Figure 4C). Near this region, the shibire ts2 mutation (G146, orange sphere) results in a defect in GTP binding at non-permissive temperatures (Narayanan et al., 2005). Furthermore, mutations of another residue (T141) near the switch 2 region have been shown to enhance (T141A) or inhibit (T141D) assembly-stimulated GTPase activity (Song et al., 2004a). Separately, a neighboring mutant (K142A) inhibits endocytosis despite having no affect on GTPase activity, and therefore may uncouple GTP hydrolysis from dynamin’s conformational change (Marks et al., 2001). We pursued the possibility that the GTPase domain alone may form weak oligomers (see Experimental Procedures), however, no oligomerization was found unless a GED fragment was present (data not shown), which is consistent with previous studies (Muhlberg et al., 1997). Therefore, any interactions between adjacent GTPase monomers are dependent upon assembly promoted by other regions of the dynamin sequence, specifically the middle domain and GED.
After constriction, adjacent GTPase dimers are in a position to interact. The sequence in Interface #2 (Figure 4C) is specific to dynamin, and several residues adopt different conformations when comparing the rat and Dictyostelium structures. The peripheral sequence (purple loop in figure 4C) in the rat GTPase domain (Reubold et al., 2005) is largely unstructured, while an additional α-helix is observed for the same region in the Dictyostelium structures (Niemann et al., 2001). The crystallographic B-factors for this region are higher in both structures indicating potential flexibility. This flexibility may be stabilized by self-assembly of the loop region, while elasticity in this region may be important for constriction to occur upon GTP addition.
Immediately downstream from the shibire ts1 mutation, GTPase-GED interactions have been proposed at the hydrophobic cleft between the N- and C-terminal helices of the GTPase domain (Niemann et al., 2001). Changes in the structure and orientation of the GTPase domain, due to GTP binding (and later hydrolysis), are likely propagated to the GED/middle domain leading to the kinked structure observed for the middle density in the constricted state. The observed kinked pattern of middle radial density coincides with the twisting of the GTPase domains relative to one another, which allows for tighter packing of the GTPase domains in the constricted state. Overall, the combined GTPase/middle/GED subunit motions act like a corkscrew normal to the helical axis (Figure 7). This motion along the helical array changes the subunit packing leading to compaction of the structures parallel to the helical axis and constriction normal to the lipid bilayer.
Comparing the PH domain fittings in the non-constricted and constricted maps also reveals a shift relative to the helical axis upon constriction (Figure 2, E & F). However, adjacent PH domains are not in a position to interact or stabilize the oligomeric state through intermolecular interactions between adjacent dynamin monomers. Rather, the PH domain provides an affinity for lipid membranes with a negative potential, which concentrates and anchors dynamin to the membrane. As a result, the lipid bilayer provides a backbone for dynamin to assemble and undergo conformational changes upon nucleotide binding/hydrolysis. Without this backbone, dynamin spirals formed in vitro immediately disassemble when GTP is added. Most members of the dynamin family do not contain a PH domain, so the function of the PH domain is specific for its role in vesicle fission.
The relative intensities for the inner radial densities are different between the non-constricted and constricted maps of ΔPRD dynamin tubes. The dynamin PH domain (inner radius) and lipid bilayer densities are stronger, and therefore more uniform, in the non-constricted map. Constriction of ΔPRD dynamin due to nucleotide binding may weaken stable interactions between the dynamin PH domain and the lipid bilayer, and subsequent GTP hydrolysis results in release from the membrane. Accordingly, over time dynamin falls off the lipid bilayer in the presence of GTP (Danino et al., 2004).
Based on the fittings of dynamin domains to the cryo-EM reconstructions, we propose that the GTPase, middle and GED domains work in concert to drive helical constriction. The PH domain acts as a separate entity during vesicle fission, tethering dynamin to lipid. Concurrent with dynamin recruitment to regions of negatively charged lipid, the middle domain and GED drive self-assembly of dynamin. In a cooperative manner, the GTPase domain is then stimulated, and upon nucleotide binding, undergoes a conformational change that is sensed and propagated through the middle/GED interaction. Overall, the conformational change resembles a corkscrew motion that acts to constrict the membrane through a combination of sliding (going from 14.2 to 13.2 subunits per turn) and twisting of each repeating subunit in the array (Figure 7). Interactions between GTPase, middle and GED domains are more tightly packed, while PH domain interactions with the membrane may be placed under stress due to changes in lipid substrate curvature. Ultimately, fission relieves the strain placed on the membrane, and dynamin is released.
The reconstruction of ΔPRD dynamin in the constricted state represents a transition state between non-constricted and supercoiled states observed with both wild type and ΔPRD tubes upon GTP hydrolysis (Danino et al., 2004; Roux et al., 2006). In vivo, if the dynamin helix were anchored at two positions (plasma membrane and coated pit), a coiled strain would be placed on the membrane possibly causing fission, which is consistent with the observed fragmentation of dynamin tubes due to GTP-induced twisting (Roux et al., 2006; Sweitzer and Hinshaw, 1998). Dynamin’s PRD-binding partners, such as amphiphysin (Grabs et al., 1997), intersectin (Evergren et al., 2007) and endophilin (Gad et al., 2000; Ringstad et al., 1999), also play a role in dynamin-dependent endocytosis. Amphiphysin has been shown to target dynamin to clathrin-coated pits (Shupliakov et al., 1997) and increases dynamin’s rate of GTP hydrolysis and conformational change (Takei et al., 1999). SH3-containing cofactors may bind the PRD of dynamin at late stages of endocytosis and prevent self-association of the PRD with the rest of the dynamin molecule, thereby allowing constriction to proceed. Therefore, the role of the PRD is limited to targeting and regulating dynamin constriction, while the GTPase, middle and GED domains, which are conserved throughout the dynamin family, comprise the mechanochemical core of dynamin that drives membrane fission. Overall, this work addresses how the conformational changes observed for dynamin in vitro are essential for constriction during endocytosis in vivo. Using a multifaceted approach, we are able to present the first model for the organization and conformational changes of the five distinct domains of dynamin during membrane fission.
Cryo-EM density was incorporated as a structural restraint for refinement of positions for the GTPase and PH domains with the YAMMP molecular modeling package (Tan et al., 1993). The VLAT force field term defines the cryo-EM density as a three-dimensional potential, providing a score for fitting the model to the density. Refinements were performed using rigid body Monte Carlo with simulated annealing, which allows for exhaustive sampling of conformational space while the structure moves to an orientation that best match the cryo-EM data.
Three-dimensional reconstructions of constricted ΔPRD dynamin were previously determined in the presence of GMP-PCP, a non-hydrolysable GTP analog, using helical reconstruction (Zhang and Hinshaw, 2001) and iterative helical real-space refinement (IHRSR) methods (Chen et al., 2004; Egelman, 2000). In addition, a 3D reconstruction of non-constricted ΔPRD dynamin was determined using the IHRSR method (Chen et al., 2004). The images used for the reconstructions were taken with a LaB6 tip, which made CTF correction beyond the first CTF zero difficult. Therefore, only data within the first Thon ring were used in the reconstructions, resulting in a resolution of ~20 Å.
To initiate fittings, several reduced-representation models (Cα atoms) of dynamin GTPase monomers (Reubold et al., 2005) were placed at random orientations near the helical structures. A non-bond term was also added to the energy calculation to prevent interpenetration of GTPase structures:
where Eij is the non-bond interaction energy between atoms i and j; kij is the non-bond force constant (100 kcal/mol•Å²) for the atom pair ij; rij is the distance between atoms i and j; and rijo is the minimum distance allowed between the two atoms. The non-bond term prevents overlap between adjacent structures, and a value of rijo = 7.5 Å was used. The initial simulation performed 2,000,000 steps of Monte Carlo refinement with simulated annealing starting at 1,000 K and cooling to 10 K while treating each rat GTPase monomer (299 atoms) as a rigid unit. We determined the orientations of dimer interactions from this fit (see interface #1, Figure 4).
A second round of refinement optimized the local fit of each GTPase dimer. Using one turn of the helical density, 13 and 14 rigid GTPase dimers were refined to the constricted and non-constricted helical densities, respectively, using 1,000,000 steps of Monte Carlo with simulated annealing starting at 1,000 K and cooling to 10 K. The final placement in the density was visually confirmed using the program “O” (Jones et al., 1991). From this final refinement, additional turns of the helix could be modeled. Finally, the all-atom structures of the GTPase domain were superposed onto the refined Cα positions.
A similar protocol was used to fit a reduced-model of the human PH domain of dynamin (Ferguson et al., 1994) to the helical densities. For the non-constricted map, the PH domain was fit to the inner radial density with the GTPase structures present in the outer radial density, thereby limiting the conformational freedom of the PH domain. For the constricted map, the inner radial density is relatively weak, so the Cα structure was initially placed manually at random orientations near the inner radial density, and additional pseudoatoms were placed in the middle radial density to occlude this region. 1,000,000 steps of rigid body Monte Carlo with simulated annealing were performed from 1,000 K to 10 K for both the non-constricted and constricted fittings. These structures provide models for pairs of PH domain structures, and multiple pairs were then positioned in a full turn for each dynamin helix. Additional refinement using Monte Carlo with simulated annealing was performed from 500 K to 10 K without the additional atoms in the outer or middle radial densities. From this final refinement, additional turns of the helix could be modeled. The all-atom structures of the PH domain were superposed onto the refined Cα positions. The cryo-EM density and ribbon images were generated using RIBBONS (Carson, 1997).
Wild-type dynamin in 20 mM HEPES, pH 7.2, 100 mM NaCl, 2 mM MgCl2, 1 mM EGTA (HCB100) at 0.25 mg/ml was incubated with 1:500 w/w trypsin (Sigma) at room temperature in the presence and absence of phosphatidylserine (Avanti Polar Lipids) extruded liposomes. When lipid was added, the protein was incubated at room temperature for 2 hours to allow for tube formation. Small aliquots from the mixture were retained at intervals of 15 minutes after addition of trypsin to examine the amount of proteolysis that occurred over the course of 1 hour. Samples were run on 4–12% NuPage/MES gels and stained with Colloidal Blue (Invitrogen) for visualization. The gel samples were then given to the Proteomics and Mass Spectrophotometry Facility (PMSF) at NIDDK for mass analysis to determine the sequence of the proteolyzed fragments.
ΔPRD dynamin in 20 mM HEPES, pH 7.2, 100 mM NaCl, 2 mM MgCl2, 1 mM EGTA (HCB100) at 0.25 mg/ml was incubated with phosphatidylserine extruded liposomes in the above buffer to form non-constricted tubes. Samples were applied to a carbon coated Maxtaform 400 mesh Cu/Rh (Ted Pella) grids for one minute and subsequently blocked with HCB100 containing 1% BSA for 45 minutes. The grids were then incubated for 1.5 hours with an Anti-Dyn antirabbit antibody (MC65, provided by Dr. Mark McNiven) whose epitope is against amino acids 223–249 in the GTPase domain of human dynamin-1 (Henley et al., 1998). After batch washing with HCB100, the grids were incubated with 6 nm Colloidal Gold Goat anti-rabbit secondary antibody (Jackson ImmunoResearch) for 1.5 hrs and batch washed with HCB100. The grids were then stained with 1% uranyl acetate for 30 seconds. As controls, we examined dynamin alone (no PS lipid) to observe background labeling of tetramer structures. Additionally, secondary antibody alone revealed minimal labeling.
Images were taken on a Philips (FEI Company) CM120 transmission electron microscope using low dose conditions at 100 kV with a LaB6 tip. Images were recorded using a Gatan 1 K × 1 K CCD camera. Gold particles were quantified using Image J software (available online at http://rsb.info.nih.gov/ij/).
The authors thank Drs. Robert K.-Z. Tan and Stephen C. Harvey for helpful discussions about the implementation and use of YAMMP, Dr. Mark McNiven for providing antibodies, and Dr. Eric Anderson at the NIDDK Proteomics and Mass Spectrophotometry Facility for help with analysis of proteolyzed samples. We thank Dr. Shunming Fang for computational support, Dr. Paula Flicker for comments on the manuscript, and Dr. Edward Egelman for his collaboration in determining the cryo-EM structures used in this study. This research was supported by the Intramural Research Program of the NIH, NIDDK.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.