1. Cell Biology
  2. Structural Biology and Molecular Biophysics
Download icon

The role of scaffold reshaping and disassembly in dynamin driven membrane fission

  1. Martina Pannuzzo
  2. Zachary A McDargh
  3. Markus Deserno  Is a corresponding author
  1. Carnegie Mellon University, United States
Research Article
  • Cited 0
  • Views 1,395
  • Annotations
Cite this article as: eLife 2018;7:e39441 doi: 10.7554/eLife.39441

Abstract

The large GTPase dynamin catalyzes membrane fission in eukaryotic cells, but despite three decades of experimental work, competing and partially conflicting models persist regarding some of its most basic actions. Here we investigate the mechanical and functional consequences of dynamin scaffold shape changes and disassembly with the help of a geometrically and elastically realistic simulation model of helical dynamin-membrane complexes. Beyond changes of radius and pitch, we emphasize the crucial role of a third functional motion: an effective rotation of the filament around its longitudinal axis, which reflects alternate tilting of dynamin’s PH binding domains and creates a membrane torque. We also show that helix elongation impedes fission, hemifission is reached via a small transient pore, and coat disassembly assists fission. Our results have several testable structural consequences and help to reconcile mutual conflicting aspects between the two main present models of dynamin fission—the two-stage and the constrictase model.

https://doi.org/10.7554/eLife.39441.001

eLife digest

When cells take up material from their surroundings, they must first transport this cargo across their outer membrane, a flexible sheet of tightly organized fat molecules that act as a barrier to the environment. Cells can achieve this by letting their membrane surround the object, pulling it inwards until it is contained in a pouch that bulges into the cell. This bag is then corded up so it splits off from the outer membrane. The ‘cord’ is a protein called dynamin, which is thought to form a tight spiral around the bag’s neck, closing it over and pinching it away. The structure of dynamin is fairly well known, and yet several theories compete to explain how it may snap the bag off the outer membrane.

Here, Pannuzzo et al. have created a computer simulation that faithfully replicates the geometry and the elasticity of the membrane and of dynamin, and used it to test different ways the protein could work. The first test featured simple constriction, where the dynamin spiral contracts around the membrane to pinch it; this only separated the bag from the membrane after implausibly tight constriction. The second test added elongation, with the spiral lengthening as well as reducing its diameter, but this further reduced the ability for the protein to snap off the membrane. The final test combined constriction and rotation, whereby dynamin ‘twirls’ as it presses on the neck of the bag: this succeeded in efficiently severing the membrane once the dynamin spiral disassembled. Indeed, the simulations suggested that dynamin might start to dismantle while it constricts, without compromising its role. In fact, getting rid of excess length as the protein contracts helps to dissolve any remnants of a membrane connection.

Defects in dynamin are associated with conditions such as centronuclear myopathy and Charcot‐Marie‐Tooth peripheral neuropathy. Recent research also indicates that the protein is involved in a much wider range of neurological disorders that include Alzheimer's, Parkinson's, Huntington's, and amyotrophic lateral sclerosis. The models created by Pannuzzo et al. are useful tools to understand how dynamin and similar proteins work and sometimes fail.

https://doi.org/10.7554/eLife.39441.002

Introduction

Of the three 80% homologous mammalian isoforms of dynamin, Dyn1, highly expressed in neurons, has been studied best (Hinshaw, 2000; Praefcke and McMahon, 2004). Crystallographic analysis (Chen et al., 2004; Faelber et al., 2011; Ferguson et al., 1994; Ford et al., 2011; Zhang and Hinshaw, 2001; Chappie et al., 2010) reveals five distinct domains: an N-terminal GTPase or G-domain; a ‘stalk’ region consisting of a four helix bundle; the signaling BSE domain that links G-domain and stalk; a pleckstrin homology (PH) domain binding phosphatidylinositol-4,5-biphosphate (PIP2) lipids; and a proline-rich C-terminal domain (PRD) that mediates interactions with membrane scaffolding proteins. Dynamin monomers oligomerize via their stalks in a criss-cross fashion, resulting in stable dimers (Cocucci et al., 2014) or tetramers (Ramachandran et al., 2007) in solution. Continuing this assembly path yields helical filaments (Carr and Hinshaw, 1997), which have an outer diameter of about 50 nm and a helical pitch (i.e., height increment during one complete helical turn) between 10 nm and 20 nm in the absence of GTP (Chen et al., 2004). Specific interactions among dynamin subunits are not conserved throughout the large dynamin superfamily, yet all members share similar architectural assembly properties (Praefcke and McMahon, 2004).

Dynamin helices are narrower in the presence of GTP; the strongest constriction has been seen with the point mutant K44ADyn1: it forms a stable superconstricted 2-start helix, tightening membrane tubules to a water-filled inner lumen with a diameter as narrow as 3.7 nm, but not cutting them (Sundborger et al., 2014). Since K44ADyn1 is very inefficient in GTP hydrolysis, which however is necessary for remodeling (Zhang and Hinshaw, 2001; Chappie et al., 2011), this mutant is believed to trap the system in a pre-fission state. This leaves open the role of GTP activity, with presently two major competing scenarios: the two-stage and the constrictase model.

In the two-stage model (Bashkirov et al., 2008; Mattila et al., 2015), nucleotide-loaded dynamin assembles into highly constricted helices, whose entrapped lipid tubules spontaneously undergo hemifission; GTP hydrolysis induces subsequent detachment of the dynamin scaffold from the membrane, allowing the hemifission state to proceed to complete fission. However, neither the lack of dynamin’s cooperativity with GTP during hydrolysis (Tuma and Collins, 1994), nor the short lifetime of the post-hydrolysis GDP + Pi state (Antonny et al., 2016) are compatible with key requirements of the two-stage model. It also remains unclear why the experimentally observed superconstricted state does not reach hemifission. In the constrictase model (Chappie et al., 2011), several cycles of GTP hydrolysis induce adjacent turns (or ‘rungs’) to stepwise slide past one another, actively constricting the helix and its enclosed membrane tube until the latter fissions; disassembly is then a consequence of the vanishing lipid substrate. The problems here are that the cryo-EM density of stalks in a constricted (albeit: 2-start) helix does not match the X-ray structure of dimers, that it is not known how G-G links across rungs would unbind and step cooperatively, and that the mechanics of the final disassembly is less clear (Antonny et al., 2016).

To refine these two partially conflicting scenarios into a consistent and realistic model requires information about fast changes at the nanometer-scale. This is a challenge for all structural techniques capable of reaching the necessary resolution, and it often requires the use of mutants that might introduce artifacts. However, there is also a second problem: it is by no means obvious how a highly curved bilayer responds to the forces and torques imposed by such complex geometric constraints. Theoretical calculations have provided constraints on energetics and morphology (Kozlov, 1999; Kozlovsky and Kozlov, 2003; Bashkirov et al., 2008; McDargh et al., 2016; McDargh and Deserno, 2018), but these assume high symmetries and ignore fluctuations. Recent coarse-grained (CG) simulations, in which the dynamin helix is represented as a pair of rings, have elucidated the relevance of local torques (Fuhrmans and Müller, 2015) and the possibility of long-lived hemifission intermediates (Mattila et al., 2015; Zhang and Müller, 2017), but the implied mirror symmetry differs from the actual helicoidal one. Here we go beyond these studies and investigate the constriction of fluctuating lipid bilayer tubes by geometrically realistic dynamin helices.

We employ an implicit-solvent CG membrane model (Cooke et al., 2005; Cooke and Deserno, 2005) (see Materials and methods) in which lipids are represented by three consecutive beads, each with a diameter of σ0.8nm (Figure 1a), that assemble via tail attraction and form bilayers with a bending rigidity of κ13kBT (Cooke et al., 2005; Cooke and Deserno, 2005), where the thermal energy kBT=4.1×1021J0.6kcal/mol provides the natural energy scale. The CG time scale maps to approximately τ15ns, based on lipid self-diffusion, but we stress that this needs to be interpreted cautiously, since coarse-graining might speed up different dynamic processes differently. The emergent CG membranes capture not only a wide range of mesoscopic phenomena (Deserno, 2009), but also very local bilayer properties that likely play a role during leaflet-breaking fission events, such as a well-placed pivotal plane (Wang and Deserno, 2015), a correct magnitude and correlation length for lipid tilt (Wang and Deserno, 2016), and a pore-opening scenario (Cooke et al., 2005; Deserno, 2009) in agreement with previous simulations and continuum theory (Farago, 2003; Tolpekina et al., 2004). The dynamin filament is composed of similar CG beads, arranged and elastically connected to capture filament radius r, helical radius R and pitch 2πp, and the location of an effective PH binding strip (Figure 1b and Figure 5, Materials and methods). In the present study we forgo the dynamin assembly process and start with a pre-formed helix winding around the neck of an elongated vesicle comprising about 10,000 lipids (Figure 1c). We change the scaffold geometry by tuning the equilibrium distances between its CG beads, that is by imposing internal stresses that trigger a global elastic shape relaxation, and we do this slowly enough to avoid viscous stresses in the bilayer (Figure 6, Materials and methods). All simulations were performed using the ESPResSo package (Limbach et al., 2006) and run in triplicate, leading to consistent results. Visualization was done with VMD (Humphrey et al., 1996).

Ingredients of the model.

(a) CG lipids comprising three beads assemble into membranes via tail bead attractions and head bead repulsions. (b) the dynamin helix is built as a stack of disks with 19 CG beads each; it contains an adhesive membrane-binding strip (red) that represents the PH domains and that can rotate by an angle φ away from the inward-pointing direction. (c) a CG filament constricts the neck of a CG vesicle consisting of 10,000 CG lipids.

https://doi.org/10.7554/eLife.39441.003

Results

Figure 2 with 1 supplement see all
Dynamin shape changes and membrane response.

(a) Top and (b) Side-view of dynamin dimers assembled into an unconstricted helical filament. (c) CG representation of a constricted filament. (d) Constricted and elongated filament. (e) Constricted and rotated filament. (f) Radius of gyration Rg of the membrane neck (see Materials and methods section) as a function of filament radius R under four protocols: constriction only (blue) constriction+rotation (cyan), constriction+rotation+elongation (dark cyan), constriction+elongation (dark blue). (g) An unconstricted filament resting on a membrane of matching radius (left) creates a Darboux-torque once the (red) adhesion strip is rotated (right), inducing the membrane to asymmetrically bulge; the two arrows indicate the torque couple. (h) Cross-sectional view of a 1- and 1.5-turn helical scaffold at R=10.5 σ and a 3.5-turn scaffold at R=10σ. The filament was constricted and rotated, only the adhesion strip is shown. Hemifission seeds are small pores, visible as breaks in bilayer continuity (arrows). (i) Cross-cut illustration of membrane shape changes triggered by a simultaneous filament constriction, rotation, and gradual disassembly (Video 5), leading to hemifission. (j) Continuation of the previous sequence from hemifission to complete fission.

https://doi.org/10.7554/eLife.39441.004

Pure constriction

We begin by exploring a process in which the filament constricts its helical radius R but maintains both the pitch and the orientation of the PH domain with respect to the substrate (Figure 2c). At a confinement approximately corresponding to the superconstricted state, the enclosed membrane tube remains stable. The hemifission intermediate only forms at constriction levels that eliminate the inner lumen (Figure 2f, blue curve, and Video 1). This seems to contradict the findings by Kozlovsky and Kozlov (2003) that an inner lumen diameter smaller than approximately 2 nm should spontaneously proceed towards hemifission, but this is not so: in their study an approximately catenoidal neck fissions by the application of bending moments at the upper and lower edge of the neck, while in our case the neck is created by external constriction—a mechanically different scenario. It is of course conceivable that our bilayer tube is merely kinetically stable, but nothing analytical is known about the barrier towards the topologically distinct hemifission state. Indeed, the dynamin-coated superconstricted lipid tubule is stable in experiment (Sundborger et al., 2014).

Video 1
Pure filament constriction.

A dynamin filament, initially with radius R=18σ, pitch 2πp=12σ and rotation angle φ=0 is constricted by stepwise reduction of the radius R down to the critical radius R=Rc=10.5σ, following the blue time-series in Figure 6. The right part of the video shows a side view, in which the 10,000 lipids of the vesicle are only rendered in stick-representation such as to permit some amount of transparency. The dynamin filament is shown as a transparent gray tube, and only the adhesive PH domain strip is explicitly shows as a sequence of red beads. The left part of the video shows a view along the helical axis of the filament, and lipids are only shown if their z-distance from the filaments center of mass is within ±1σ, such as to follow the extent of constriction and the diameter of the luminal region.

https://doi.org/10.7554/eLife.39441.014

Constriction and elongation

Some experiments have shown that GTP hydrolysis may increase the filament’s pitch at fixed radius (Figure 2d), which has led to the suggestion that dynamin might act as an extension spring (‘poppase model’) (Stowell et al., 1999). However, since dynamin changes its shape slowly compared to a membrane’s viscoelastic time scale, believed to be shorter than about 10 ms (Camley and Brown, 2011), this putative mechanism cannot rely on viscous stresses but instead only on the reversible work being performed during elongation of the neck. To test this, we simultaneously decrease the radius and increase the pitch (Materials and methods, Figure 6). Contrary to expectation, helical elongation impedes fission: at the same helical scaffold radius R, a larger pitch results in a larger effective radius Rg of the enclosed membrane tubule, requiring further constriction to achieve hemifission. This happens because a large pitch prevents the helical filament from symmetrically confining the membrane, allowing it to ‘bulge out’ at the open groove (Figure 2f, dark blue curve). In nature this could be prevented by accessory proteins, or—if the two-start helix is physiologically relevant—by the second interlocking filament. Compared to the pure constriction case, which in all three runs transitioned into the hemifission step at the same constriction step, we here observe a slight scatter of the precise transition points between two consecutive steps—see Figure 2—figure supplement 1d. We attribute this to the reduced confinement of a membrane inside a helical scaffold with a widening groove, permitting more fluctuations that render the transition point less definite.

Constriction and rotation

Beyond changes of radius and pitch, the set of macroscopic motions available to a helical filament contains a third possibility: a ‘twirling’ motion around the local filament axis (Figure 1b and Figure 2e). Its relevance for dynamin derives from a growing number of studies that provide experimental evidence for the tilting of dynamin’s PH domains. Mehrotra et al. (2014) have shown that a change in PH domain orientation may regulate fission. Sundborger et al. (2014) fitted dynamin’s crystallographic structure to EM density maps and noticed that one of the two PH domains per dimer tilts out of the membrane, thereby breaking the symmetry of the dimer. And a very recent 3.75 Å resolution Cryo-EM reconstruction of human Dyn1 in the GMPPCP-bound state (Kong et al., 2018) shows that the bundle signaling element (BSE) is asymmetrically bent, presumably due to forces generated from the GTPase dimer interaction. These forces are further transferred across the stalk to the PH domain and onto the lipid membrane.

On mesoscopic scales, this local rearrangement of the tertiary structure can effectively be described as a net rotation of the helical scaffold around its local longitudinal axis, even if the core of the protein filament does not actually co-rotate. If the PH domains of each monomer were to tilt in the same fashion, then due to dynamin’s criss-cross assembly half of the PH domains would tilt ‘up’ while the other half would tilt ‘down’, which on average does not displace the filament’s binding surface. However, if only one of these two sets of PH domains tilts, this breaks the up-down symmetry and—at the level of our CG model—effectively rotates the average position of the adhesive strip away from the substrate. This creates a tangential torque on the membrane, because adhesion pins the filament’s local material frame to its Darboux frame on the surface (Guven et al., 2014), and therefore it has appropriately been called ‘Darboux torque’ (Fierling et al., 2016)—see Figure 2g and Box 1.

Considering that the PH domain is connected to the stalk rather flexibly, it is not obvious that it can actually transmit such a torque. However, flexibility does not preclude the transmission of stresses, provided other contacts are in place that help acting as a pivot, and these might simply be steric interactions with other parts of the assembled scaffold. We recall that Kong et al. (2018) provide structural evidence that forces are transmitted all the way from dynamin’s G-domain to the PH domain and onto the membrane, even though this is indirectly deduced by aligning the coordinates of dynamin with an unbent BSE to the structure of a bent dynamin in the DynGMPPCP map at the GTPase domain. Still, neither this new structure, nor the previously observed tilting of the PH domain (Sundborger et al., 2014) unambiguously proves the existence of torques. But while elucidating the mechanical nature of this process will have to await more detailed molecular modeling, the existence of a Darboux torque as been posited and exploited in another model for dynamin-driven fission, which simplifies the complex scaffold geometry to two counter-rotating apposing rings (Fuhrmans and Müller, 2015; Shnyrova et al., 2013), a geometry that remote-pinches the membrane between the rings. Notice, though, that both geometry and symmetry are different in the latter case: counter-rotating rings have a mirror symmetry and hence exert oppositely acting Darboux torques that constructively interfere in the middle of the scaffold. Since a helix is one contiguous filament, it will rotate everywhere in the same sense, rendering torque amplification much less straightforward.

In our own simulations, which account for the full helical geometry, we find that rotation is necessary for a constricted membrane tube with a remaining aqueous lumen to transition into the hemifission intermediate. Consistent across all three simulation runs, the constriction plus rotation sequence (Figure 2f, cyan curve, Figure 2—Figure Supplement 1a, and Video 2) transitions at R=Rc10.5σ8.4nm, when the inner lumen is small but has not yet vanished; we will subsequently refer to Rc as the ‘critical constriction radius’.

Video 2
Filament constriction and rotation.

Same as Video 1, but this time the filament is additionally rotated, following a combination of the blue and red time series in Figure 6. The hemifission event towards the end happens at R=Rc, at which the non-rotation protocol shown in Video 1 does not lead to hemifission. It results in a stable cylindrical micelle spanning the two daughter vesicles.

https://doi.org/10.7554/eLife.39441.015

Since rotation gradually diminishes binding between membrane and filament, a non-rotating scaffold maintains adhesion while constricting, which could hold the membrane tube open and thereby prevent it from transitioning into the hemifission case. To test this, we considered a constricted but not rotated filament at R=10σ (which is even below the critical radius) and artificially switched off the adhesion between scaffold and membrane. As shown in Figure 3, the neck did not progress towards hemifission; in fact, the response to this change is a reduction in Rg which, considering the error bars (determined via blocking (Flyvbjerg and Petersen, 1989)) is so small that it is not statistically significant. This suggests that the functionally important consequence of rotation is not merely a reduction of binding energy, but active stresses, likely due to the above-mentioned Darboux torque.

Box 1

Illustration of a tangential membrane torque.

Illustration of a tangential membrane torque.

An object in contact with a membrane can exert a force on it by locally pushing or pulling. But it can also exert a tangential torque (i.e., a torque whose rotation axis is parallel to the membrane plane). This can be conceptualized as a force couple—two forces of equal magnitude but opposite direction, which do not share the same line of action. As an illustration, panel (a) shows a pen clipped to a flat piece of paper and held tangentially to its surface. Rotating the pen around its axis, as in (b), exerts a horizontal torque on the paper pinned between pen and clip, resulting in an S-shaped curvature deformation. The rotation is illustrated as a blue curve, the corresponding force couple as two red arrows. Darboux torques are tangential torques that are not triggered by an explicit external rotation (as in the pictured example) but through a mismatch between the direction of a spontaneously curved filament’s normal curvature and the direction of its adhesive strip (Fierling et al., 2016), which in our case is quantified by the angle φ (see Figure 1b). Mechanical equilibrium then also requires that in the absence of any external forcing the total Darboux torque integrates to zero.

Box 1—Figure 1
Illustration of a tangential membrane torque.
https://doi.org/10.7554/eLife.39441.016
https://doi.org/10.7554/eLife.39441.017
Time evolution of Rg.

The neck’s gyration radius is shown in the end stages of a pure constriction protocol. The time series starts when the scaffold radius is R=Rc=10.5σ, which leads to Rg/σ=4.187±0.02 (as measured over the subsequent 600τ, with an error determined via blocking (Flyvbjerg and Petersen, 1989)). After that, a further constriction of the filament to R=10σ reduces the neck’s gyration radius to Rg/σ=3.890±0.04 but does not trigger hemifission (which corresponds to the much smaller value Rg2σ—see Figure 2f). Turning off the adhesion between scaffold and membrane only reduces the gyration radius by a very minor amount, Rg/σ=3.841±0.035, a change that is not statistically significant (p=0.36). Once we additionally let the scaffold disassemble into (non-adhesive) dimers (see also Video 4), the neck very rapidly doubles its radius within about 200τ, after which the definition of its location becomes ambiguous.

https://doi.org/10.7554/eLife.39441.018

If in addition to constriction and rotation we also elongate the helix, hemifission is still reached before the inner lumen disappears, but this requires a further filament radius reduction (Figure 2f, dark cyan curve). Consistent with our observations on constriction plus elongation, this case with additional rotation also shows a greater variability across our three runs, transitioning over three consecutive constriction steps, which however fall between the bounds of constriction plus rotation and constriction only (see Figure 2—figure supplement 1b). These findings again endorse the view that rotation supports hemifission, while elongation opposes it.

Pre-hemifission disassembly of the dynamin coat

In the two-stage model the constricted state assembles from GTP-loaded dynamin, while subsequent hydrolysis-triggered depolymerization of the filament induces fission (Bashkirov et al., 2008; Mattila et al., 2015). Two pathways are conceivable: pre-hemifission coat disassembly would need to drive the membrane tube towards (at least) hemifission, while post-hemifission disassembly would only need to destabilize this hemifission state. To test the first pathway, we start with a filament at R=Rc (induced only via constriction) and then cut it into approximately dimer-sized fragments. Irrespective of whether these retain their adhesion capability to the membrane (Video 3) or lose it and hence immediately unbind (Video 4), the membrane neck rapidly widens following scaffold rupture (see also Figure 3). Of note, even binding-capable fragments are individually not strong enough to impose their curvature on a substrate whose geometry becomes progressively unfavorable, and they ultimately detach from it. This also documents that our scaffold only binds weakly, and that rotation-driven hemifission is not the trivial result of massive forces resulting from large adhesion, or even of pulling lipids out of the membrane.

Video 3
Filament breakage with retained adhesion.

A filament constricting a vesicle down to the critical radius Rc is suddenly cut into pieces comprising five discs each (corresponding roughly to dynamin dimers). Even though the membrane curving fragments retain their adhesion energy with respect to the membrane substrate, this leads to a re-expansion of the neck radius, followed later by fragment unbinding once the substrate curvature becomes too unfavorable.

https://doi.org/10.7554/eLife.39441.020
Video 4
Filament breakage with simultaneously lost adhesion.

Same as Video 3, except that the fragments also lose their binding affinity to the membrane, causing them to immediately detach and stop curving the substrate.

https://doi.org/10.7554/eLife.39441.021

Observe that our simulation setup does not exert external membrane tension, since our implicit solvent enforces no volume constraint and permits the membrane—even though closed and constricted—to relax the area per lipid. Nevertheless, it is unlikely that a nonzero tension would prevent the widening of the neck: the value that would be needed to maintain a cylindrical membrane at the critical radius Rc is given by κ/2Rc2, or a few mN/m (Deserno, 2015; Bukman et al., 1996Hochmuth et al., 1996), which is very high. Even disregarding the question how such a large tension would arise in a physiological context, the mere fact that this approaches a membrane’s rupture stress and would hence place the system dangerously close to unspecific bilayer failure renders a recourse to such a high tension implausible.

Non-axisymmetric pathway to initiate hemifission via pores

Recent experiments have observed that fission of dynamin-coated membrane tubules occurs within the coated region (Dar et al., 2015), although earlier studies had suggested that fission occurs at the edge between the dynamin helix and the uncoated membrane (Morlot et al., 2012). We always observe that an initial hemifission seed appears in the middle of the neck for short filaments (up to 1.5 turns), or two seeds appear near the edges for longer ones (more than two turns), see Figure 2h. This suggests that two regions of highly localized stress appear slightly inwards of either edge, which overlap and may mutually amplify for sufficiently short filaments. Furthermore, since longer filaments trigger hemi-fission only at an even smaller scaffold radius (Rc,long10σ), both findings indicate that short filaments induce hemifission more efficiently than long ones.

That hemifission is seeded by a small pore is unexpected, not only because fission is believed to be non-leaky (Bashkirov et al., 2008), but also because the existence of a hemifission intermediate is generally taken to exclude the need for (or even possibility of) pores. However, a tube’s inner leaflet must change topology, posing the question how a filament only in contact with the outer leaflet could promote this transition. Our simulations suggest an intriguing non-axisymmetric pathway: the filament seeds a small pore that widens around the circumference of the neck, while its top and bottom edges pull radially inward and fuse with the apposing intact membrane. This creates two defect lines at which three bilayers meet, and contracting them results in the top and bottom point singularity of the hemifission state’s cylindrical micelle. This sequence of events never involves a large pore, provided the filament stays short enough, a condition that imposes constraints on the constriction process itself.

Concomitant constriction and disassembly of the scaffold

If dynamin constricts due to GTP hydrolysis, as posited in the constrictase model, the initially unconstricted helix has to be long enough for the two rungs to meet, so that apposing G-domains can cross-catalyze hydrolysis. Constricting the radius by a factor of two then necessarily doubles the number of turns if the helical length stays fixed. We find that such long scaffolds induce rather sizable pores (Figure 2h) that would likely result in leakage. But the ends of a filament that in the constricted state only takes one turn would not be able to meet up in the preceding unconstricted state, and GTP driven constriction could not commence. This dilemma could be resolved if constriction and disassembly happen in short succession: when the two ends of the initially relaxed helical filament first meet, they hydrolyze GTP and trigger a first constriction step. This brings a new pair of GTP loaded G-domains in contact that drive the next constriction step, and so forth until full constriction. But since GTP hydrolysis enhances disassembly, a filament end comprising ever longer stretches of spent dynamin can shed its monomers concurrently with the ongoing constriction, especially if its PH domain retracted from the membrane. The helix would tighten but never extend much beyond one turn, and hence never induce large pores (Video 5). This hypothesis is compatible with the experimental evidence that, upon GTP hydrolysis, the scaffold seems to adjust to a length optimal for fission (about one full turn) (Shnyrova et al., 2013).

Video 5
Constriction, rotation, and concurrent filament depolymerization.

Same as Video 2, except that the filament also concurrently depolymerizes during constriction and rotation. As a result, once hemifission sets in, the two point-defects at the ends of the resulting cylindrical hemifission micelle are no longer kept far apart by the now much shorter enclosing scaffold, permitting the defects to annihilate and thus fission to complete.

https://doi.org/10.7554/eLife.39441.022

Transition from hemifission to complete fission

Without concurrent filament shortening, our hemifission state resembles a short cylindrical micelle (Figure 2f, background image), a transition state explored in two other recent investigations (Mattila et al., 2015; Zhang and Müller, 2017). These studies find that this micelle is remarkably stable, suggesting that completing fission faces a very sizable second free energy barrier. Additional tension might reduce its height, and indeed in vitro experiments by Roux et al. (2006) and Morlot et al. (2012) have found that longitudinal tension assists in producing fission (though it is interesting to note that our simulations present a tension-free pathway to fission). However, Zhang and Müller (2017) estimate that the barrier might remain as large as 30kBT even at the largest biologically justifiable tensions (close to uncontrolled rupture), which leads them to speculate that other factors help to complete fission, such as the high curvature in the region where the micelle merges with the bilayer.

Here we propose that this second barrier might indeed not be biologically relevant, since it is a consequence of the simulation setup in these studies, which prevents the micelle from shrinking and instead requires it to rupture mid-length. In our simulations we also never observe the micelle to break while being enclosed by the dynamin scaffold (from which it has already unbound), but we argue that this is due to the scaffold preventing the two point defects at the end of the micelle from merging. Once the scaffold disassembles (whether gradually or abruptly), they are being pulled together by a force that can be estimated to be Fπκ/2z0𝒪(100pN). This is the energy per unit length of creating a cylindrical micelle of curvature radius z0 (the pivotal plane distance) out of lipids that have a monolayer leaflet rigidity of κ/2. This force pulls the two daughter membranes together, whereupon bilayer contact catalyzes micellar fission (see Video 6). This hypothesis, even though derived from a fairly coarse-grained model, is nevertheless plausible, because splitting the cylindrical micelle somewhere along its length would further raise the free energy by creating two spherical caps of even more unfavorable packing geometry (Zhang and Müller, 2017), while the contact between the daughter membranes annihilates the two already existing point defects at the micellar ends.

Video 6
Fission via unbinding and defect merger.

If at the endpoint of Video 2 the filament depolymerizes, the cylindrical hemifission micelle pulls the two daughter vesicles together and permits the point defects to merge and annihilate, thus completing fission.

https://doi.org/10.7554/eLife.39441.023

The latter scenario seems to conflict with the widely held belief that highly curved vesicles are fusogenic. However, recent experiments by François-Martin et al. (2017) paint a more nuanced picture. These authors measured the free energy barrier towards fusion between 60nm diameter POPC (16:0–18:1 PC) and DOPC [18:1 (Δ9–Cis) PC] vesicles at 37°C. On the one hand, the observed free energy barrier of ΔF30kBT is at the lowest end of theoretical estimates, indicating that protein-free fusion between such vesicles is easier than previously thought. On the other hand, even when incubating such vesicles at the unusually high concentrations of 18 mM PC, only 2% had completed fusion after 30 min, which does not support the notion of spontaneous fusion—as these authors indeed conclude.

Nevertheless, the free energy pathways for events that change membrane topology remain challenging, for at least four unrelated reasons: first, the choice of boundary conditions and constraints in theory or simulations can strongly affect the outcome (for instance, whether cylindrical micelles can shrink, hemifusion diaphragms can expand, or tensions can relax); second, the key physics happens at the nanometer scale, and hence the results of highly coarse-grained models such as ours or continuum theory must be interpreted cautiously; third, a topological barrier to fusion exists, whose height depends on the Gaussian curvature modulus and is thus not well known (see the contribution of Deserno in Bassereau et al. (2018)); and fourth, additional complexity due to accessory proteins or lipid mixtures may qualitatively change the energetics, for instance by lipid sorting in very high leaflet curvature gradients (Cooke and Deserno, 2006; Tian and Baumgart, 2009). For the pathway that naturally arises in our simulations—a shrinking of the hemifission micelle post-disassembly, followed by autocatalytic cutting—the free energy barriers are not obvious, especially for solvent free models like ours (or those in the above-mentioned studies by Mattila et al. (2015) and Zhang and Müller (2017)), and hence this subject warrants more dedicated future studies.

Results and discussion

Taken together, our studies support a number of conclusions. To begin with, the poppase model (Stowell et al., 1999) faces not only kinetic but also equilibrium obstacles: extension of the helical scaffold disfavors fission, because a widening groove offsets constriction. In contrast, hemifission is promoted by tilting one of the two symmetry subsets of PH domains (Mehrotra et al., 2014; Sundborger et al., 2014; Kong et al., 2018), which effectively rotates the filament’s adhesive strip away from normal contact. The resulting Darboux torque (Fierling et al., 2016; Guven et al., 2014) has been previously invoked to explain fission (Fuhrmans and Müller, 2015; Shnyrova et al., 2013), but these earlier models pictured the dynamin scaffold as two rings, whose counter-rotation-induced torque remote-pinches the enclosed membrane tube. The effect is particularly intuitive under mirror symmetry, a geometry at odds with the actual helical one. Surprisingly, this difference appears not to be central.

We show that pre-hemifission scaffold breakage aborts fission by allowing the enclosed membrane tube to re-expand. But disassembly is still essential to complete fission, as argued in the two-stage model, because even if the severed bilayer is energetically preferable to the hemifission state (Kozlovsky and Kozlov, 2002), a substantial energy barrier is required to explicitly break the hemifission micelle (Mattila et al., 2015; Zhang and Müller, 2017). Our simulations lead us to suggest that this can be circumvented by merging and annihilating the micelle’s two endpoint defects, a process that commences once the two daughter membranes are pulled into close proximity and hence critically depends on the scaffold getting out of the way by disassembling. Notice that this mechanism is not scaffold specific and may hence apply to any other fission machinery that leads to a hemifission state involving a small stretch of a cylindrical micelle.

How hemifission is induced is experimentally less clear (Antonny et al., 2016). In our studies constriction alone, down to the smallest experimentally observed luminal radii, does not suffice. We believe this to be independent of whether the constricted state is reached passively (two-stage model) or actively (constrictase model), since our dynamic protocol could simply be viewed as a means to adiabatically prepare a highly constricted state. PH domain tilting thus emerges as a way to catalyze hemifission, and since applying a Darboux torque costs energy, we posit that the energy of GTP hydrolysis is at least in part used to drive this conformational change. This is in accord with experimental findings by Dar and Pucadyil (2017) that replacing the PH domain by a simple binding motif strongly slows down the fission rate. It is also supported by the recent Cryo-EM reconstruction by Kong et al. (2018), who suggest that forces generated from the GTPase dimer interaction are transferred across the stalk to the PH domain and from there onto the membrane. It is worth noting that in their reconstruction Dyn1 was bound to the nonhydrolyzable guanosine triphosphate analogue GMPPCP, and so the question whether GTP hydrolysis would create additional forces or torques that then could be transduced to the membrane remained open. To better understand the mechanical basis and viability of such a force transmission, it will be essential to structurally resolve the connection between the PH domain to the stalk domain.

Finally, we consistently observe that hemifission is initiated by a transient pore puncturing the constricted neck (see arrows in Figure 2h, the time sequence in Figure 4, and Video 2). This is reminiscent of likewise non-axisymmetric pathways seen in fusion simulations (Noguchi and Takasu, 2001Noguchi and Takasu, 2002Müller et al., 2002; Müller et al., 2003), whose implied transient leakage currents were soon after confirmed to be common in electrophysiological measurements on hemagglutinin-mediated fusion (Frolov et al., 2003). We estimate that our pores open to a diameter of no more than 2 nm, which under physiological ionic strength equips them with a conductivity on the order of 1 nS, while their lifetime is a few microseconds (mapped very roughly from our coarse-grained model via lipid self-diffusion). This would not be trivial to detect experimentally, but similarly fast transients (few microseconds) have been observed in the gating currents of Shaker potassium channels embedded in Xenopus oocytes, using an eight-pole Bessel filter and achieving a bandwidth of about 200 kHz (Bezanilla, 2018; Sigg and Bezanilla, 2003). Increasing the ionic strength in reconstituted systems to 1 M, a further gain in sensitivity by up to an order of magnitude could be obtained, provided this does not interfere with dynamin’s operation. But notice also that our leakage pores are partially covered by the dynamin scaffold (see again Figure 2h) and, under physiological conditions, possibly by additional proteins proposed to assist in dynamin-driven fission, such as BAR domains (Takei et al., 1999; Farsad et al., 2001; Itoh et al., 2005; Mim et al., 2012). Hence, a pore’s effective conductance might be significantly lower than that of a freely accessible membrane pore or channel.

Hemifission via a transient pore.

This time sequence of simulation snapshots shows slices through the membrane neck with a width of 2σ, placed symmetric around the center of mass of the scaffold (not shown for clarity). The continuity of the tail region (yellow; blue are the head groups) is emphasized by using VMD’s ‘QuickSurf’ rendering on all tail beads (Humphrey et al., 1996), which creates an isosurface extracted from a volumetric Gaussian density map. In panel (a), chosen to be time-point 0τ in this sequence, the tail region is still continuous and encloses an inner lumen with a diameter of about 1\textnm. In panel (b) a small pore opens (red arrow) that connects the inner lumen to the exterior region of the vesicle. As it widens through panels (c) and (d), the pore rim above and below the imaged plane fuses to the inner leaflet of the lumen, which finally leaves in panel (e) a cylindrical hemifission micelle connecting two closed vesicles. See also Video 2.

https://doi.org/10.7554/eLife.39441.024

Our simulations show that pores are larger—and hence potential leakage more severe—if hemifission is triggered at larger radii, or if the filament is longer, suggesting that a close coordination of constriction, rotation, and possibly concomitant disassembly renders fission not only more efficient but also more tight. We hence expect mutants disrupting this coordination (for instance the ΔPH mutant of Dar and Pucadyil (2017)) to have larger pores. This would exacerbate leakage problems and could be experimentally observed.

While we have explicitly focused on the case of classical dynamin, several of our findings have implications beyond this particular protein and offer lessons for membrane topology remodeling that go beyond Dyn1. For instance, all members of the dynamin superfamily are believed to oligomerize (Praefcke and McMahon, 2004), likely into helices, for which our generic basic model holds. Furthermore, the mismatch between the normal curvature direction of such a helical filament and the direction of its adhesive domain (measured by φ in our case) is a generic degree of freedom for such a structure. Whether directly present during assembly or only later triggered by a (possibly GTP-dependent) conformational change, it will give rise to Darboux torques on the membrane. We have explicitly shown that these support the transition into the hemifission state, and hence they provide a means to trigger a topological transition that is different from the notions of constriction or elongation. This matters because other topology remodeling proteins exist that are unlikely to work by either. Consider in particular the reverse-geometry scission driven by the ESCRT-III complex (Wollert et al., 2009), which involves the adsorption and polymerization of a helical filament on the inside of the neck to be cut. Present models (for a recent review, see Schöneberg et al. (2017)) rely on adhesion and/or geometric changes of the spiraling filaments, which result in either direct forces or boundary-induced stresses at the contact site (arising from, for instance, a contact curvature condition (Deserno et al., 2007)). But geometric rearrangements of a curved elastic filament with a finite twist rigidity, whose material frame is pinned to the membrane (Guven et al., 2014), almost invariably result in additional Darboux torques, as well as twist-induced analogs (Fierling et al., 2016), the relevance of which is only beginning to emerge (Quint et al., 2016). In this work we have explicitly demonstrated, for the particular example of Dyn1, that this geometrically elementary mechanism is remarkably effective, which suggest that it might be more common than so far realized: traces of the essential action could date back to the earliest bacterial FtsZ ancestor, which shares many of the key geometric (Erickson, 2000) and biochemical (Lu et al., 2000) features. This might hence suggest novel functional models also for other membrane remodelers (such as ESCRT-III), for which the mode of operation is much less well understood than for classical dynamin.

Materials and methods

General modeling aspects

Our investigation focuses on the mesoscopic effects of a constricting helical dynamin filament on a tubular lipid membrane, expressed primarily by the interplay between (i) filament geometry and binding affinity and (ii) membrane and filament elasticity. A long-standing aim in the dynamin field has been to develop an explanatory model for dynamin’s membrane fission mechanism within the framework of such mesoscopic emergent properties. By creating a coarse-grained (CG) model of dynamin fission that captures precisely these mesoscopic aspects, our goal is to explore the consequences of this interplay, while either disregarding much of the microscopic or chemical detail, or implicitly accounting for it in terms of effective interactions.

The fundamental degrees of freedom in our model are mesoscopic ‘beads’ with a size (diameter) σ0.8nm, which hence correspond to 𝒪(10) heavy atoms. Physics below this scale leaves its trace at larger dimensions through the collective action of effective potentials—much like the quantum mechanics of correlated electron clouds re-emerges classically as effective van der Waals interactions, or the configurational distributions of charged moieties in a molecule yield effective dipole moments and polarizabilities. A rich literature exists outlining the technique of coarse graining in a soft-matter and biophysics context (Deserno, 2009; Ingólfsson et al., 2014; Izvekov and Voth, 2005a; Izvekov and Voth, 2005b; Brini et al., 2013; Müller-Plathe, 2002; Murtola et al., 2009; Noid, 2013; Noid et al., 2008; Peter and Kremer, 2009; Riniker et al., 2012; Saunders and Voth, 2013; Voth, 2008). In the terminology of Noid (2013), we employ a ‘top down’ modeling approach.

Since we are not concerned with hydrodynamic effects, we account for the embedding water implicitly via effective attractive interactions between hydrophobic species (such as CG lipid tail beads).

The reduction of the number of degrees of freedom allows not only for a computational speed-up and a better statistical sampling; it also offers insight into the nature of the original problem: physical effects that are correctly represented in a CG model prove to be largely independent of microscopic specifics, rendering them important players in an emergent mesoscopic explanatory model.

All simulations were performed using the ESPResSo package (Limbach et al., 2006), using an integration time step Δt=0.01τ (where τ is the simulations CG time unit). A Langevin thermostat (Grest and Kremer, 1986) with friction constant Γ=1.0τ1 was used to keep the temperature constant. All simulations have been performed in an NVT ensemble.

Lipid model

We use an implicit solvent lipid model (Cooke et al., 2005Cooke and Deserno, 2005) in which a single lipid molecule is replaced by three consecutive beads, one for the hydrophilic head and two for the hydrophobic tails, which are not individually resolved (Figure 1a). In the absence of solvent, the hydrophobic effect, and hence aggregation of lipids into bilayer membranes, is driven by attractive interactions between the CG tail beads of depth ε (the simulations’ energy unit) and range wc (the precise potential forms are detailed by Cooke et al. (2005)). The magnitude of wc and the temperature T determine whether lipids aggregate into fluid membranes, and if so, what their material properties are. We work at the frequently employed state point wc=1.6σ and kBT=1.1ε (which also sets the energy scale). Under these conditions lipids aggregate into fluid membranes with an area per lipid of about a=1.2σ20.77nm2, an average head-bead distance from the midplane of about dH=2.2σ1.8nm, and a separation between the half-maximum density points (a possible proxy for the Luzzati thickness (Luzzati and Husson, 1962)) of about d1/2=5.6σ4.5nm. Considering the limitations of the underlying microscopic basis of nanometer-sized CG beads, these numbers are reasonably close to typical biologically relevant lipids, such as POPC (a0.643nm2 and d1/23.91nm at 30°C (Kučerka et al., 2011); still, sub-nanometer resolution should not be taken too literally in a model like this.

The membrane has a bending rigidity of κ12.8kBT (Cooke et al., 2005Cooke and Deserno, 2005; Harmandaris and Deserno, 2006; Hu et al., 2013), purposefully chosen smaller than a typically value of 20kBT (Kučerka et al., 2005), since this affords a significantly entropy-deprived coarse-grained system an alternative opportunity to undergo fluctuations. Wang and Deserno (2016), recently also measured the tilt modulus of this model, finding it to be κt7kBT/nm2. This results in the characteristic tilt decay length t=κ/κt1.3nm, in good agreement with experimental values (Jablin et al., 2014), indicating that the model does not merely capture overall fluidity and curvature elasticity, but also local lipid reorientation physics, which matters for the intermediate states of fission and fusion (Kozlovsky and Kozlov, 2002).

Protein

We model the right-handed helical dynamin scaffold by CG beads of the same diameter σ as the lipid CG beads (represented by a purely repulsive Weeks-Chandler-Andersen potential that only acts between filament and lipid beads), placing them such as to represent the shape of a CG filament (Figure 1b). This first requires setting up a coordinate system that embodies helical symmetry—see Figure 5 for the following discussion.

Definition of helicoidal coordinate system.

The black vertical z-axis is surrounded by a cylinder, around which a helical dynamin tube winds. Its green center line with radius R and pitch 2πp is given by (Equation 1). On this space curve we define the (right-handed) Darboux-frame {T,N,L}, consisting of a tangent vector T (black), normal vector N (blue) and co-normal vector L (red), which are given by (Equation 2). In the N-L-plane we can then define the circular cross-section of the filament and place rings of CG beads at the correct distance from the green filament axis. The direction along N points towards the enclosed cylinder, but it is easy to rotate it by an angle φ around the T axis. This is how one may rotate the beads representing the PH domain (red) off the underlying substrate.

https://doi.org/10.7554/eLife.39441.025

The arc-length (s) parametrization of a right-handed helix of radius R and pitch 2πp, whose axis coincides with the z-axis of the coordinate system, is given by

(1) X(s)=(Rcos(s/R2+p2)Rsin(s/R2+p2)ps/R2+p2) .

This has a length 2πR2+p2 per turn and a local (Frenet) curvature of R/(R2+p2). The local unit tangent vector T(s)=X(s) is then

(2a) T(s)=1R2+p2(Rsin(s/R2+p2)Rcos(s/R2+p2)p) .

We need to define a finite-thickness filament for which X(s) is the central axis. Moreover, since we have to specify the location of an adhesive strip relative to a central cylinder that is being wrapped by the helix, it is convenient to extend the tangent vector T along the helical curve into a local basis by defining two additional vectors N and L as follows: both have unit length and are perpendicular to T, the normal vector N coincides with the local surface normal of the inscribed cylinder, and the remaining co-normal vector L is given by L=T×N:

(2b)N(s)=(cos(s/R2+p2)sin(s/R2+p2)0)(2c)L(s)=1R2+p2(psin(s/R2+p2)pcos(s/R2+p2)R) .

The thus defined triplet {T,N,L} of vectors constitutes a right-handed orthonormal basis at every point along the filament. It is called the ‘Darboux frame’ (with respect to the underlying cylindrical surface on which the filament rests).

On the {N,L} plane perpendicular to the filament we now place CG beads that will form one of the cross-sectional circular discs from which we build the filament slice by slice. Each disk consists of a central bead placed directly onto the filament axis, a first ring of 6 beads and radius R6, and a second ring of 12 beads and radius R12. The coordinates of the beads sitting on the 1-, 6- and 12-ring can be parametrized as

(3a)X1(s)=X(s)(3b)X6(s,n)=X(s)+R6[N(s)cosφn(6)+L(s)sinφn(6)](3c)X12(s,n)=X(s)+R12[N(s)cosφn(12)+L(s)sinφn(12)]

where n{0,,5} for the 6-ring and n{0,,11} for the 12-ring. The angles are given by

(4) φn(6)=2πn6+φ,φn(12)=2πn12+φ,

and the overall phase shift φ denotes the extent to which bead 0 on the 6- and 12-ring is rotated around the local filament axis, see Figure 1b. Since we will subsequently equip bead 0 on the 12-ring with an additional adhesion towards lipid head groups, representing the inner binding region of the dynamin filament due to the PH domains, this phase angle φ describes a rotation of the CG filament with which we effectively capture the asymmetric tilting of half of the PH domains. Notice that due to the way the Darboux frame is set up, φ=0 corresponds to an adhesive strip that exactly sits on the enclosed cylindrical surface (and hence does not exert any Darboux torque).

We typically consider filaments consisting of 9 dynamin dimers, each comprising five discs, for a total of 45 discs. We take R6=2σ and R12=4σ, making for a filament diameter of about 8σ6.4nm. The spacing Δs between discs along the central helix is set so that a full turn in the unconstricted state (R=20σ and 2πp=11σ) comprises 40 discs, leading to Δs=2πR2+p2/40=3.1536σ. The location of every bead in the filament at a given shape triplet {R,p,φ} is now completely specified.

The shape is elastically fixed by introducing harmonic bonds between all beads within a cutoff distance of rcut=5σ, whose rest length equals the equilibrium distance between the beads according to the chosen shape triplet {R,p,φ}, and whose spring constant is K=200ε/σ2, which is similar to the choice for other elastic networks for coarse-grained protein models (Periole et al., 2009). This renders the filament sufficiently stiff so that it can impose its shape on an underlying membrane tube, and not vice versa.

To change the filament’s shape, we adjust the rest lengths of all elastic springs so that they represent distances in a filament at a new shape triplet, {R,p,φ}{R,p,φ}. This introduces local stresses which the filament relaxes by transforming into the new equilibrium shape. We consider parameter ranges within R/σ[6,18], 2πp/σ[11,20] and φ[0,80]. Figure 6 illustrates the time series of these parameter changes, which we use individually or in combination, as outlined in the main text. For instance, during constriction we reduce the helical radius by 1σ every 300τ, corresponding to about 0.18nm/μs. Albeit much faster than in reality, this is still effectively quasistatic.

Changes of filament geometry.

Time sequence followed when driving the dynamin filament through changes in its radius R, pitch 2πp, and rotation angle φ. As detailed in the results section, we conduct sets of simulations in which various combinations of these observables are adjusted, while others stay at their original value. For instance, in a constriction+elongation simulation the radius is successively decreased from 18σ to 9σ, while the pitch is simultaneously tuned up from 11.5σ to 20σ in a way documented by the blue and black curves, respectively, while the rotation angel remains at φ=0.

https://doi.org/10.7554/eLife.39441.026

Beads on the red adhesive strip (accounting for the PH domains) additionally experience an adhesion of strength ε towards lipid head beads, represented by a standard Lennard-Jones potential that is truncated and shifted to zero at r=2.5σ. For the beads on the two immediately adjacent neighboring strips (i.e., numbers 1 and 11 on the 12-ring) we turn off the hard core repulsion with respect to lipid head (but not tail) beads, in order to permit the adhesive domain to embed into the head group region.

The free energy of binding for dynamin dimers or larger fragments depends not only on the local chemistry, but also on the curvature of both dynamin and the membrane. We are not aware of measurements that are precise enough to help parametrizing this interaction, but in order to avoid driving fission by overly strong interactions (which could, for instance, exert unrealistically large torques or even pull lipids out of the membrane), we have opted for a lower-bound scenario, in which we made the interaction between dimer-equivalents (blocks of 5 discs) and a highly curved membrane neck just strong enough to trigger binding; but when substrate curvature decreases due to membrane tube widening, the interplay between curvature energy and adhesion increasingly disfavors binding (McDargh et al., 2016), and fragments detach (Video 3).

Measuring membrane constriction

Several possibilities exist to quantify the extent of tubular membrane constriction, but the two most obvious ones have significant drawbacks: the midplane radius cannot be defined once the membrane is in the hemifission state, and the radius of the inner lumen cannot distinguish between a completely closed bilayer tube and a hemifission micelle.

We hence use as a metric for constriction the cross-sectional gyration radius Rg of beads in the vicinity of the constriction point, defined as follows:

(5) Rg2=1Mi=1M[(rir0)xy]2 ,

where the sum extends over all lipids that are at most a radial distance of 20σ and an axial distance of ±1σ away from the filament’s center of mass, r0 is the center of mass of these selected lipids, and the subscript 'xy' indicates that we first take the projection of rir0 into the xy-plane.

If Rm is the midplane radius of a cylindrical lipid tube and w the monolayer width, then in continuum approximation we get

(6) Rg2=RmwRm+wdr2πrr2RmwRm+wdr2πr=Rm2+w2 ,

showing that Rg=Rm[1+12(w/Rm)2+] is close to Rm, with quadratic higher order corrections in w/Rm. Moreover, a membrane cylinder of vanishing inner lumen has Rm=w and hence Rg=2w, while the hemifission micelle has Rg=w/2. Hence, the jump in Rg is w/2=12×5.6σ/22σ, using the Luzzati width of our CG membrane. This is very close to our observed jump distance in the constriction only case (Figure 2f, blue curve), which indeed only transitions when the inner lumen disappears. For the other cases we investigate the jump tends to be higher, because hemifission occurs while the membrane tube is still water-filled.

References

  1. 1
  2. 2
  3. 3
  4. 4
    Gating currents
    1. F Bezanilla
    (2018)
    The Journal of General Physiology 150:jgp.201812090.
    https://doi.org/10.1085/jgp.201812090
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
    Activation of dynamin GTPase is a result of positive cooperativity
    1. PL Tuma
    2. CA Collins
    (1994)
    The Journal of Biological Chemistry 269:30842–30847.
  84. 84
    Coarse-Graining of Condensed Phase and Biomolecular Systems
    1. GA Voth
    (2008)
    CRC Press. ISBN 978-1-4200-5955-7.
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89

Decision letter

  1. Michael M Kozlov
    Reviewing Editor; Tel Aviv University, Israel
  2. Vivek Malhotra
    Senior Editor; The Barcelona Institute of Science and Technology, Spain

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "The role of scaffold reshaping and disassembly in dynamin driven membrane fission" for consideration by eLife. Your article has been reviewed by Vivek Malhotra as the Senior Editor, a Reviewing Editor, and three reviewers. The following individuals involved in review of your submission have agreed to reveal their identity: Vadim Frolov (Reviewer #1); Aurélien Roux (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The article describes the results of state-of-the-art CG simulations of membrane fission by dynamin, which, potentially, represent a considerable advance in understanding the mechanism of this phenomenon. The simulations allowed the authors to check, within the framework of their computational model, the feasibility of the previously suggested two-stage and constrictase scenarios of the process, to test whether the early suggested modes of the conformational changes of the dynamin helix can drive membrane fission, and to conclude that (i) the Darboux torque applied by the dynamin helix to the membrane is necessary for fission, (ii) the precursor of the hemi-fission stage is formation of a membrane nano-pore; (iii) the transition from the hemi-fission to the complete fission happens spontaneously after disintegration of the dynamin scaffold.

The reviewers found the results interesting and, potentially, suitable for publication in eLife provided that their comments are addressed.

Essential revisions:

1) One of the major concerns of the reviewers is a lack of the statistical validation of the results. In the current manuscript only one fission scenario is presented without any indication of the probability of its realization in the computational experiments.

Further, the statement that other scenarios, which include different combinations of the dynamin spiral constriction, elongation and rotation, are ineffective in terms of driving fission, needs a statistical substantiation.

A quantitative comparison of fission probability in all possible scenarios is necessary to validate the major conclusion of the study.

More specific questions would be: How variable is the fission pathway? How many hemi-fissions (in all cases? All with pores? Do pores ever go to complete fission directly?)? All of them are stable cylindrical micelles? All micelles shrink spontaneously to 0 length and then break upon the protein filament disassembly?

2) The quantitative characterization of the structural details of the suggested and alternative pathways has to be strengthened. Specifically,

- It is claimed that pores nucleating before hemi-fission are few nm wide and live for few microseconds, too small to be detected in real life. Of note, few nm pores are huge, not small (typical transient pores in electroporation are close/below 1 nm). Even if a pore opens for few microseconds, the integral charge transferred through it under a typical voltage bias of 100mV would be within detection limits. To substantiate the claim that the pores predicted by the proposed pathway are too small to be detected experimentally, determination of the size/open time distributions would be necessary.

In this context, the authors should comment on the validity of their strategy to simulate pores. In the case the computational method used is known to cause difficulties to simulate pore formation, it is requested that the authors perform simulations to estimate the pore formation at different tensions and compare the results to the experimental data (for example, Evans et al., 2003).

- It is claimed that pure constriction produces fission only upon complete closure of the lumen. What are the criteria for "no lumen"? How big is the difference (quantitatively, e.g. in nm) between "no lumen" and "visible" lumen seen in fissions caused by rotation+constriction? What are the variations in Rc? How Rc/its variations depend of the rotation angle/effective torque?

3) The agreement between the suggested mechanism and the existing structural data has to be further elaborated. Specifically:

- The Darboux torque is suggested to be produced when the dynamin adhesion line moves along a fixed/stable filament, e.g. via asymmetric displacement of the PH domains. This movement assumes synchronous tilting of many domains to produce the torque, yet such cooperative actions were seemingly ruled out. Then how does the torque build up? Further, can the PH domains support the torque providing that their connections to the protein stalk are rather flexible? More broadly, the torque creates stresses in the protein filament itself – is it realistic (can it be estimated) that the helical filament sustains the stress without changing shape?

Along the same lines, the authors state that tilting of the PH domain upon GTP hydrolysis could create a Darboux torque only if an asymmetric tilt occurs. In currently available cryo-EM maps (Sundborger et al., 2014), only 1 of the 2 PH domains of dimers is tilted, ensuring the required asymmetry. However, when looking at the tetrameric level, the titled PH domains are on each side of the tetramers, restoring symmetry. The authors should carefully check these points and explain how the data are compatible with their findings about the importance of the Darboux torque.

Finally, the tilting of the PH domains has been discussed in detail by biochemists, in particular, in the context of its effect on the molecular interactions between the membrane and the dynamin helix. Two levels of changes have been proposed: (i) tilting helps insertion of an amphipathic helix that helps to promote membrane curvature and thus constriction; (ii) titling breaks the specific lipid (PIP2)-dynamin bond or pulls the lipid out of the membrane. While the first change is unlikely because the amphipathic helix is located on the side of dynamin that is not in interaction with the membrane when titled (Harvey McMahon, private communication), the latter is essential for transmission of the Darboux torque to the membrane. It is thus required that the authors propose an estimate of the energy put on the dynamin/membrane bond when PH domains are tilting and compare it to the affinity values of dynamin PH domain for PIP2, and to forces required to pull off lipids from membrane.

4) The relationship with the results of other simulation models has to be thoroughly discussed. Specifically:

The conclusion that the scission of the stalk/micelle, may not be physiological relevant sounds problematic. Other simulation works show that a stalk formed between fusing vesicles is highly long living even for symmetric, membrane forming lipids such as POPC (Risselada, 2014). Those systems essentially do not differ from the ones simulated here after dynamin disassembly. In support of those results, it was recently found by one of the reviewers (using a string method) that a 'dimple' stalk formed between POPC membranes is thermodynamically stable and that scission would require a ~20 kBT barrier, quite in contrast to the here-envisaged fast rupture after Dynamin assembly. The question is whether the Crooks model underestimates the stalk stability and thereby falsely advocates the conclusion that the coined "second barrier" may not be physiological relevant. In flat membranes POPC stalks can become metastable and even stable under membrane dehydration (stabilization of the stalk is hydration repulsion driven in that case). The solvent free Crooks model cannot reveal such a behavior since it has a strong inherent tendency to form lamellar structures. Furthermore, the stalk may equally well expand after Dynanim disassembly, i.e., progression of fusion, since progression of fission and fusion are competitive pathways at this stage (they share the shame intermediate). The statement that small hemifused vesicles - which is essentially the structure one obtains after Dynamin disassembly - are poised to undergo fission is the exact opposite of the widely accepted observation that highly curved vesicles are fusogenic (even when being protein-free). In fact, there is evidence that completion of fission relies on feedback mechanisms and may involve several constriction cycles. This could suggest that scission attempts may fail, and that the mechanism perhaps relies on a dynamically imposed stress.

A related issue is the lack of movement of the (centers of mass of) daughter vesicles in Video 5 and Video 6 (comparable in size "protein pieces" move)? Is there a constraint? If so, does it affect the hemifission stability? Breakage of the cylindrical micelle upon shortening (Video 6) looks puzzling, one would rather expect formation of a stable stalk-like structure.

https://doi.org/10.7554/eLife.39441.033

Author response

Summary:

The article describes the results of state-of-the-art CG simulations of membrane fission by dynamin, which, potentially, represent a considerable advance in understanding the mechanism of this phenomenon. The simulations allowed the authors to check, within the framework of their computational model, the feasibility of the previously suggested two-stage and constrictase scenarios of the process, to test whether the early suggested modes of the conformational changes of the dynamin helix can drive membrane fission, and to conclude that (i) the Darboux torque applied by the dynamin helix to the membrane is necessary for fission, (ii) the precursor of the hemi-fission stage is formation of a membrane nano-pore; (iii) the transition from the hemi-fission to the complete fission happens spontaneously after disintegration of the dynamin scaffold.

The reviewers found the results interesting and, potentially, suitable for publication in eLife provided that their comments are addressed.

We thank the reviewers for their positive assessment of our work and the many detailed and informed comments they have offered. We have made a number of significant changes to our manuscript, some of which required extensive additional simulations. Below we outline in detail how we have addressed the individual points that have been brought up.

Essential revisions:

1) One of the major concerns of the reviewers is a lack of the statistical validation of the results. In the current manuscript only one fission scenario is presented without any indication of the probability of its realization in the computational experiments.

Further, the statement that other scenarios, which include different combinations of the dynamin spiral constriction, elongation and rotation, are ineffective in terms of driving fission, needs a statistical substantiation.

A quantitative comparison of fission probability in all possible scenarios is necessary to validate the major conclusion of the study.

We agree with the reviewer that additional statistical support of the simulation trajectories will indeed be beneficial. We have therefore repeated each of the four investigated studied fission scenarios (pure constriction, constriction+rotation, constriction+elongation,

constriction+rotation+elongation) in two additional simulation runs. The repeated runs confirm our overall conclusion but provide additional information on the variability of the process (see below). For the sake of visual clarity, we have not included all resulting 12 runs in a single figure but restricted to the original 4 representatives. However, we have included an additional supplementary figure which in each case show all executed runs and the resulting fission pathway.

Since in all cases we drive the system until it undergoes hemifission, the key question is not the “fission probability” but something more akin to an efficiency, namely, the question how far one has to constrict the neck to trigger the transition. In our work we find that the four scenarios we study rank in efficiency (from high to low) according to:

1) Constriction plus rotation

2) Constriction plus rotation plus elongation

3) Pure constriction

4) Constriction plus elongation

This ranking is consistent across all simulation runs, with cases (1) and (3) showing essentially no variation, and cases (2) and (4) some variation, which however does not switch the ranking. See also the discussion on variability below.

More specific questions would be: How variable is the fission pathway? How many hemi-fissions (in all cases? All with pores? Do pores ever go to complete fission directly?)? All of them are stable cylindrical micelles? All micelles shrink spontaneously to 0 length and then break upon the protein filament disassembly?

Let us begin with the reproducibility of our ranking: our repeated simulations show that the two cases of pure constriction and constriction+rotation show no statistically quantifiable variation. In each case hemifission happens at exactly the same amount of overall constriction (or constriction+rotation) as in the case we had previously documented. In contrast, the remaining two cases of constriction+elongation and constriction+elongation+rotation show some variability. In the first case, the additional two simulations transitioned even later into the hemifission state (meaning, requiring even more constriction and elongation than the case we had previously shown), while in the latter case one of the two simulations transitioned in the constriction step before and the other one in the step after the originally presented case.

All these findings are in accord with our previous claims: constriction+elongation is the least efficient pathway, and the two new simulations support this even more strongly. The other scenario, constriction+elongation+rotation is less efficient than constriction+rotation, but still more efficient than pure constriction, since the spectrum of variation still falls between these two bounds. We attribute the larger variability of these two pathways to the involvement of elongation, because this reduces the uniformity of overall confinement by opening the helical “grooves” along the dynamin scaffold. This enables larger conformational fluctuations of the enclosed lipid bilayer and hence leads to a larger variability. We now address all of this in our revised manuscript.

Next, the reviewers asked more specific questions on our pathways. Based on all available simulations we now have, based not just on three times as many constriction protocols as before, but also not forgetting that we have a variety of other protocols (such as disassembly, or stepwise disassembly), we can conclude the following:

1) We can only rigorously talk about pore formation in processes that have an inner lumen left, because otherwise it becomes somewhat ambiguous to talk about a pore (and, at any rate, it would not be experimentally measurable). That being said, that essentially means the case constriction+rotation, which in fact is the case that is physically realized. And in that case, we always see a pore, and that pore never directly proceed towards complete fission; we always first get a cylindrical micelle connecting the daughter vesicles.

2) None of our micelles ever break mid-length.

3) Micelles can only shrink spontaneously to zero length when the scaffold either has gotten out of the way, or concomitantly disassembles. But in these two cases, the micelle does spontaneously shrink. We would like to point out that anything else would be extremely surprising, since the energy per lipid in a micelle is significantly larger than the energy per lipid in a bilayer (this is why lipids make bilayers, not micelles), and so our observation that, given the choice, the lipids in a micelle physically in contact with a bilayer try to merge into the bilayer, is precisely what’s expected.

2) The quantitative characterization of the structural details of the suggested and alternative pathways has to be strengthened. Specifically,

- It is claimed that pores nucleating before hemi-fission are few nm wide and live for few microseconds, too small to be detected in real life. Of note, few nm pores are huge, not small (typical transient pores in electroporation are close/below 1 nm). Even if a pore opens for few microseconds, the integral charge transferred through it under a typical voltage bias of 100mV would be within detection limits. To substantiate the claim that the pores predicted by the proposed pathway are too small to be detected experimentally, determination of the size/open time distributions would be necessary.

We believe this comment might have resulted from a misunderstanding, based likely on an unclear presentation of the relevant text in our manuscript. We expressly do not want to claim that these pores cannot be experimentally detected. Our intention was to give a rough idea what sizes and time scales we expect from these pores. Notice that given the nanometer scale of our observed pores, as well as the resolution of our coarse-grained model, it will be difficult in any case to pinpoint a precise length-scale. Moreover, the coarse-grained nature of the model also renders time-scale mapping problematic, and our estimated scale, based on lipid self diffusion as the scale-bar to translate between coarse-grained and real units, may be off (for instance, if a different dynamic process is more important). And finally, both the dynamin scaffold itself, as well as any other proteins, such as BAR domains, which will almost certainly be present in more realistic experimental situations, could “patch up” some of the leakage sites.

Given this situation, our primary concern was to suggest that the resulting pores are, in our estimation, small and short lived enough to render them not immediately visible in ordinary run-of-the-mill electrophysiology experiments. We would agree that more dedicated state-of-the-art experiments should be able to see these types of transient electrophysiological breakdowns. In fact, we would strongly hope that this is possible, since it would render our claim accessible to experimental tests.

We have significantly extended our discussion of the detectability of pores within the anticipated length- and time-scale window, suggesting that they should, indeed, be detectable, unless they are too well covered by the filament and other accessory proteins. Of course, our simple model cannot make any quantitative comment on the latter.

In this context, the authors should comment on the validity of their strategy to simulate pores. In the case the computational method used is known to cause difficulties to simulate pore formation, it is requested that the authors perform simulations to estimate the pore formation at different tensions and compare the results to the experimental data (for example, Evans et al., 2003),

We have in fact studied pore formation with the Cooke model in the past Cooke and Deserno, (2005). Specifically, we checked the pore opening scenario as a function of applied strain, which leads to an informative balance between membrane tension and the energy of an open pore (due to its rim). This particular setup is computationally more sensitive than controlling the applied tension, because the crucial transition at pore opening can be sampled without immediately thereafter destroying the membrane. One can make precise analytical predictions for the stress-strain relation of this pore opening scenario, which had previously already been discussed by both Farago, (2003) and Tolpekina, den Otter and Briels, (2004). This relation permits one to fit not only the area expansion modulus but also, crucially, the free energy per length of the open edge. Our results yield an edge tension of about 15pN and an (adiabatic!) rupture tension of about 8 mN/m, both of which are in very good agreement with typical values quoted for these observables. For full disclosure, this required fixing an analysis error in the original publication which led to an under-estimation of these two parameters by a factor of three - the correct numbers cited here were published in Deserno and Macromol. (2009). It is also worth noting that Tolpekina et al., had not just proposed the theory but also performed the same type of simulations we did in their work, using a slight better resolved (and explicit solvent) coarse-grained model due to Goetz and Lipowsky. Their results are qualitatively in agreement with ours, even though the value for their edge tension is about 2.7 times higher.

In the brief description of our lipid model at the end of the introduction we had listed a number of “local” properties which the Cooke membrane model reproduces well. We have included the pore opening scenario among that list and provided the references discussed above.

- It is claimed that pure constriction produces fission only upon complete closure of the lumen. What are the criteria for "no lumen"? How big is the difference (quantitatively, e.g. in nm) between "no lumen" and "visible" lumen seen in fissions caused by rotation+constriction? What are the variations in Rc? How Rc/its variations depend of the rotation angle/effective torque?

By “no lumen”, we mean a situation in which the distance Ri between the center of mass of the neck section (1nm wide along the z-direction, measured from the center of mass of the scaffold) and the top of the inner lipid head groups equals zero - see Author response image 1:

This distance is very challenging to quantify precisely, because it is very noisy, of the order of the CG bead size, and characterized by non-Gaussian fluctuations (due to the boundary nature of the lowest value “0”). Moreover, it does not permit an easy distinction between the state of zero lumen in an intact highly curved bilayer versus in a cylindrical micelle (in both cases there is no water channel left). This is why we instead use the radius of gyration, Rg, as the cleaner observable to monitor the transition into hemifission.

For the case of rotation+constriction, the only one with a visible lumen, that visible lumen just before fissions measures about 1nm, and it is thus not vanishingly small. At this point a pore forms - by which we mean that the connection separating the inner and the outer lumen is broken. Figure 4 illustrates this transition. This sequence of images (with the time points relative to snapshot (a) as indicated) shows sections of the constricted neck, with the lipid headgroups rendered in blue and the lipid tails in yellow (the surrounding scaffold is not shown). The continuity of the tail region is emphasized in this image by using VMD’s “QuickSurf” rendering on all tail beads, which creates an isosurface extracted from a volumetric Gaussian density map; this visually emphasizes that all tails are part of a joint condensed aggregate, i.e., the bilayer region is intact. Snapshot (a) is taken moments before a pore opens, and a finite inner lumen is clearly visible. After the opening of a pore, visible in panel (b), the tail continuity is broken and a connection between the inner lumen and the space outside the neck opens up (indicated by the red arrow). Notice, however, that the inner lumen is still clearly distinct. In the subsequent snapshots (c) and (d) the pore widens circumferentially, but its upper and lower edge fuse with the inner leaflet of the inner lumen, thus again closing off the connection between outside volume and the interior of both daughter vesicles. The remaining snapshot (e) visualizes a cut through the cylindrical hemifission micelle. We have added this figure to our manuscript and discussed it in our text, in order to provide better visuals of the process.

In contrast, the transition into hemifission via constriction alone only occurs in our simulations when constriction eliminates the inner lumen (as defined above).

3) The agreement between the suggested mechanism and the existing structural data has to be further elaborated. Specifically:

- The Darboux torque is suggested to be produced when the dynamin adhesion line moves along a fixed/stable filament, e.g. via asymmetric displacement of the PH domains. This movement assumes synchronous tilting of many domains to produce the torque, yet such cooperative actions were seemingly ruled out. Then how does the torque build up?

While it is true that many PH domains along the filament need to work together to create a large torque, synchronicity is not needed – it is certainly not a requirement for the geometric transformation we are proposing to be feasible. Our comments on a lack of cooperativity instead referred to the ATP hydrolysis, not the PH domain tilting. It is worth picturing the difference between the well-known putative mechanism in which neighboring rungs “step” along each other by some hydrolysis driven process, versus the tilting of PH domains on subsequent monomers. In the first case, cooperativity is crucial, because the filament can only move longitudinally if all pieces of the filament step together. Conversely, nothing requires the transverse tilting of PH domains to necessarily be synchronized. A maybe helpful image is this: if many people stand behind one another in a line, they can only move forward if they step together; but it is perfectly possible for every other person in the line to step sideways without requiring any such cooperativity.

How the torque is ultimately created is of course a crucial biochemical question, but one which we cannot answer in our coarse-grained model. It is interesting, though, that a recent paper from the Hinshaw lab (2018), which appeared after submission of our manuscript, explicitly describes the intrinsic post-assembly asymmetry in the dynamin dimer along the helical filament, which appears to be triggered by the interactions of G-domains (specifically following GTP binding, but prior to hydrolysis) and which is transmitted along one of the BSEs (by bending it) all the way down to the PH domain, triggering a substantial asymmetric force onto the membrane. We believe these findings further support our proposal of both asymmetry and the existence of Darboux torques, and we have added a discussion of this into our manuscript.

Further, can the PH domains support the torque providing that their connections to the protein stalk are rather flexible? More broadly, the torque creates stresses in the protein filament itself – is it realistic (can it be estimated) that the helical filament sustains the stress without changing shape?

It is obvious that the filament has to sustain some stress in order to fission the membrane. But then, this is no different in our proposal than in any previously proposed mechanism that involves constriction or elongation, or in particular the torque scenario advanced by Shnyrova, (2013), and for which the sustainability of a sufficient stress has, to our knowledge, been largely taken for granted - presumably out of necessity. Of course, this does not mean it is obvious, but it evidently can only be answered with a more atomistic underpinning of the dynamin filament as the basis. If we for instance were to envision an atomic-level elastic network model of the filament, this would almost by construction lead to the conclusion that a dynamin filament is a helical structure made from a material with a Young modulus approximately equal to that of hard plastic (as is true for virtually all compact globular proteins). Moreover, since for such materials the bending and the torsion moduli along cylindrical stretches are basically the same (up to minor Poisson ratio corrections), the filament would be able to withstand the kind of torsional stresses we describe, if it could withstand the bending stresses underlying the more traditional models.

If the reviewers are worried in particular about the rather flexible connection between the PH domain and the rest of the dynamin protein, then we would like to remark that flexibility does not necessarily exclude the ability to transmit torque - provided that other connections are also in place, which could simply be steric in nature. We reiterate that the recent cryo EM reconstruction from the Hinshaw lab (2018) provides very strong evidence that the asymmetric bending of the BSE transmits a force to the stalk, from there to the PH domain, and from there onto the membrane.

We now offer some of this discussion in the manuscript, to explain that the kind of stress resilience we require, while not obvious, is no more exotic than the corresponding mechanical assumptions underlying more traditional constriction or elongation models. Better numbers have to await dedicated atomistic modeling, which however is beyond the aim of our paper.

Along the same lines, the authors state that tilting of the PH domain upon GTP hydrolysis could create a Darboux torque only if an asymmetric tilt occurs. In currently available cryo-EM maps (Sundborger et al., 2014), only 1 of the 2 PH domains of dimers is tilted, ensuring the required asymmetry. However, when looking at the tetrameric level, the titled PH domains are on each side of the tetramers, restoring symmetry. The authors should carefully check these points and explain how the data are compatible with their findings about the importance of the Darboux torque.

When the reviewers refer to the “tetrameric level”, we assume they refer to the structure published by Reubold et al., (2015). We wish to emphasize that this structure has been determined in crystals without lipid, and it is not an assembled structure. If lipids are the cause (or at least contribute) to the asymmetry, that would not have been seen in this study. These authors then “copy-paste” that structure to the full helix, by matching the stalks. Again, if assembly should lead to something that breaks the symmetry, that would be lost in such a process. Hence, caution should be exercised when extrapolating from such crystal structures. Moreover, the full structural analysis by Sundborger on a helix assembled onto a lipid tube sees the persistent asymmetry, and so it is unclear why results derived from the lipid-less tetramer would be particularly pertinent. One might object that the analysis of their data might impose that each dimer is identical, and that would “lock in” the asymmetry. Still, this would nevertheless show that such a structure is compatible with the data. Moreover, the very recent paper from the Hinshaw lab already cited also sees the asymmetry very clearly, even though it is more obvious in the BSE. Hence, we believe that the existence of an asymmetry, on which the Darboux torque argument rests, is sufficiently well experimentally established.

Finally, the tilting of the PH domains has been discussed in detail by biochemists, in particular, in the context of its effect on the molecular interactions between the membrane and the dynamin helix. Two levels of changes have been proposed: (i) tilting helps insertion of an amphipathic helix that helps to promote membrane curvature and thus constriction; (ii) titling breaks the specific lipid (PIP2)-dynamin bond or pulls the lipid out of the membrane. While the first change is unlikely because the amphipathic helix is located on the side of dynamin that is not in interaction with the membrane when titled (Harvey McMahon, private communication), the latter is essential for transmission of the Darboux torque to the membrane. It is thus required that the authors propose an estimate of the energy put on the dynamin/membrane bond when PH domains are tilting and compare it to the affinity values of dynamin PH domain for PIP2, and to forces required to pull off lipids from membrane.

For the case of a neck between two vesicles (i.e., a locally approximately catenoidal geometry), we cannot easily produce an estimate of the energetic cost due to PH domain tilting, or the rotation of our helical scaffold. Such analyses have been performed for a much simpler system, namely a helical filament deforming a cylindrical membrane neck (McDargh and Deserno, 2018; Fierling et al., 2016); we wish to emphasize that these calculations made up the content of entire publications in their own right. Solving this problem for our twolobed membrane geometry would be much more difficult and is beyond the scope of this publication.

That being said, we wish to emphasize the following: the obvious worry in modeling this situation is that we overestimate the strength of scaffold binding, and therefore the extent to which the filament could transduce non-compressive forces (pull or shear) onto the underlying substrate. For this reason, when designing our model, we have decided to err - if anything – on the side of weak binding. More specifically, the attractive potential responsible for binding the scaffold to the membrane was chosen to be only strong enough so that short segments of the scaffold tend to bind to a pre-curved membrane rather than remaining in solution. This is essentially the weakest we can make the binding between the scaffold while ensuring that binding does occur. Notice that this is visible in the simulations in which we cut the filament into dimer-sized pieces but maintain the adhesion between these pieces and the lipid substrate (Video 3): as the neck region widens, or dimers diffuse into the less strongly curved lobes, these dimers have a tendency to unbind from the membrane. While this does of course not constitute a full free energy calculation, we believe it documents the most relevant point that our coarse-grained binding parameters have not been chosen unrealistically strong.

We had previously explained this choice in the protein modeling part of our Methods section, but we have now expanded the Discussion section to make the rationale more clear.

4) The relationship with the results of other simulation models has to be thoroughly discussed. Specifically:

The conclusion that the scission of the stalk/micelle, may not be physiological relevant sounds problematic. Other simulation works show that a stalk formed between fusing vesicles is highly long living even for symmetric, membrane forming lipids such as POPC (Risselada, 2014). Those systems essentially do not differ from the ones simulated here after dynamin disassembly. In support of those results, it was recently found by one of the reviewers (using a string method) that a 'dimple' stalk formed between POPC membranes is thermodynamically stable and that scission would require a ~20 kBT barrier, quite in contrast to the here-envisaged fast rupture after Dynamin assembly. The question is whether the Crooks model underestimates the stalk stability and thereby falsely advocates the conclusion that the coined "second barrier" may not be physiological relevant. In flat membranes POPC stalks can become metastable and even stable under membrane dehydration (stabilization of the stalk is hydration repulsion driven in that case). The solvent free Crooks model cannot reveal such a behavior since it has a strong inherent tendency to form lamellar structures. Furthermore, the stalk may equally well expand after Dynanim disassembly, i.e., progression of fusion, since progression of fission and fusion are competitive pathways at this stage (they share the shame intermediate). The statement that small hemifused vesicles – which is essentially the structure one obtains after Dynamin disassembly – are poised to undergo fission is the exact opposite of the widely accepted observation that highly curved vesicles are fusogenic (even when being protein-free). In fact, there is evidence that completion of fission relies on feedback mechanisms and may involve several constriction cycles. This could suggest that scission attempts may fail, and that the mechanism perhaps relies on a dynamically imposed stress.

The reviewer points out an important issue that is indeed worth discussing more (as we now do in our manuscript), even though we will not be able to conclusively settle the question at this point. The challenge is that this is a highly complex problem that suffers from at least four independent difficulties:

1) Boundary conditions. The fate of a cylindrical lipid micelle connecting two vesicles, and the question whether the vesicles ultimately fuse or split, depends critically on the applied boundary conditions as well as other constraints imposed during a simulation. For instance, in the work by Zhang and Müller, (2017) the micelle is simply held at a given length and cannot shrink; something similar holds for the work by Risselada mentioned by the reviewer: we suspect that the boundary conditions applied in this work, which are different from ours, might prevent or at least delay the rearrangements needed to see the system progress to either fission or fusion. We don’t claim that our results contradict these studies; our point is that it is very difficult to compare them considering the important differences in constraints: when our vesicles separate, it’s neither a cylindrical micelle that’s breaking, nor a stalk. It’s a configuration in which a cylindrical micelle has essentially shrunk to zero length and now has the opportunity to annihilate two point defects against one another by splitting - a situation that is topologically different from the question whether a stalk (a single point defect) would split and complete towards fission. These two states may or may not be on different sides of an important free energy barrier, and in the absence of a detailed free energy study, which goes beyond the scope of our work, no premature conclusion about stability or the lack thereof can be drawn based on the outcomes of our simulations. Finally, notice that even in a case of free vesicles and no constraint on their separation, the situation would still be different from our simulation if one explicitly accounts for the solvent, because the associated constraint on volume translates into a constraint on tension via the Young-Laplace law, and then the tension relaxation will look quantitatively different.

2) Model resolution. Evidently, the question we are pondering here pertains to physics at the nanometer level, which is the scale at which we coarse grain. We have provided evidence that an important set of physical properties of the Cooke model work remarkably well down to that scale, and this is the reason why we even use it for the dynamin fission problem. However, there is evidently no guarantee that all properties, including even dynamical ones, come out right, and for that reason we would never rely on a model as coarse grained as this to proclaim that the local details of a fission or fusion process need to be rethought. All we do at this point is advance a hypothesis, and we now clearly mark it as such.

3) Topology. Bringing two vesicles together and fusing them creates an almost instant topological fusion barrier that depends on the value of the Gaussian curvature modulus, as one of us has recently pointed out (see the contribution by Deserno in the 2018 biomembrane curvature and remodeling roadmap Bassereau et al., (2018)). Since at present we know close to nothing about this modulus for any real system, it is extremely difficult to predict what kind of lipid will have what kind of topological contribution to a fusion process.

4) Idealizations. Real systems are of course even more complicated: even if we forget about the host of accessory proteins, we know that membranes consist of a mixture of lipids, and extremely high curvature regions can significantly sort them by, say, preferred curvature. This is, in fact, a phenomenon that has been observed and quantified with the Cooke model as one of its first applications: in Cooke and Deserno, (2006), Cooke and Deserno showed that while lipid spontaneous curvature is not enough to significantly sort lipids across curvature differences on the order of transport vesicle size (a claim subsequently experimentally confirmed by Tian and Baumgart, (2009)), they showed that the highly curved regions in budding necks can noticeably sort lipids.

Our comments that the second free energy barrier may not necessarily be relevant are largely independent of our specific simulation. We simply point out that the micelle need not break in the middle in order for fission to commence, since it also has two point defects to work with. In our discussions with many scientists since submission of this manuscript, we have not encountered anyone who would not consider that as a perfectly plausible alternative (Zhang and Müller in fact mention this in their paper). Our simulations indeed suggest that this is a possibility, even though we would never claim that we’ve “proved” that it works like that, considering the resolution of the Cooke model. We have now emphasized this caveat even more.

As far as the question goes whether the Cooke model is not able to show stable stalks because “it has a strong inherent tendency to form lamellar structures”, we are not sure where the reviewer takes the information from to make such a claim. We do not expect the model to be any more or less lamellar than other models, and the estimated spontaneous curvature of our lipids, the tilt fluctuations and their modulus, as well as the pivotal plane distance suggest that the model is no different in lamellarity than other models. Notice also that it correctly shows stable cylindrical micelles that refuse to break.

We thank the reviewer for reminding us of the common wisdom in the community that highly curved vesicles are fusogenic, even if protein-free. We have heard the same claims, but we think they are not as well established as their prevalence suggests. For instance, many protein-free spontaneous fusion assays use calcium as a means to get the membranes stick together (and help with dehydration). More importantly, in a very recent paper François-Martin, Rothman and Pincet, (2017) actually strive to quantify the free energy barrier towards protein-free fusion of small vesicles. They find about 30kT, which they say is at the lower end of estimates that have been proposed, and as such conclude that initiating fusion is easier than previously imagined. But looking at their experimental data, one can also see the following: in their assay, which contained thousands of highly curved vesicles at high concentration, about 2% of the vesicles had fused after a waiting time of 30 minutes. We would not want to describe this as highly fusogenic, and neither do these authors. Instead, they write:

“Finally, 30 kBT is an ideal value to enable facile membrane fusion as directed on demand in living cells: It will not happen spontaneously between bare membranes, yet as soon as specific fusion machinery is in place, it will be easily triggered.”

In other words, a fusion machinery is needed after all.

A related issue is the lack of movement of the (centers of mass of) daughter vesicles in Video 5 and Video 6 (comparable in size "protein pieces" move)? Is there a constraint? If so, does it affect the hemifission stability? Breakage of the cylindrical micelle upon shortening (Video 6) looks puzzling, one would rather expect formation of a stable stalk-like structure.

The very slow movement of the daughter vesicles is a direct consequence of the Langevin thermostat we employ to set the temperature. The presence of a friction force –ξv on every bead in the vesicle (as well as independent Gaussian noise) means that even cooperative motions, in which all beads move together, are still penalized by friction. (The diffusivity reduction of the dynamin fragments is less pronounced than that of the daughter vesicles, because they contain a smaller number of CG beads.) This would be different in a DPD thermostat, in which only relative velocities enter. This, incidentally, is a vivid example how the dynamics of two simulation setups can differ even if they, by construction, represent exactly the same equilibrium thermodynamics, and it is the reason why we are very cautious when we refer to dynamical processes. That being said, the delay in the response of the daughters to the applied friction force is not physically unreasonable, because in the real situation there will be hydrodynamic friction, which is especially high if solvent needs to be expelled from narrow crevices. Even though our Langevin friction has a different origin, we hence feel that this particular way of thermalizing the system is still more natural than a DPD thermostat would be.

As far as the question is concerned whether in movie 6 the contracting cylindrical micelle should or should not form a stalk, we would like to refer the reviewer to our answer from the previous point. We do not think that the situation is as intuitively obvious as suggested. The cylindrical micelle state contains two point defect, which in a plane that contains the micelle look a bit like half defects in nematic liquid crystals. In contrast, a stalk is a single defect, which in the same plane would look like a -1 defect. (The analogy with topological defects is incomplete, but we do not wish to get sidetracked here.) It is absolutely not obvious that the merger of two -half defects into a -1 defect is downhill in elastic energy. In fact, in liquid crystals defects of the same sign repel. Here, they are drawn together by the tendency of the micelle to shrink an unfavorable energy per unit length, at the expense of drawing two singularities closer. What the free energy landscape for this looks like would be a highly fascinating project, for which we think the outcome is neither obvious nor uninteresting, but it would lead us far beyond the scope of the present manuscript to open up this question.

https://doi.org/10.7554/eLife.39441.034

Article and author information

Author details

  1. Martina Pannuzzo

    Department of Physics, Carnegie Mellon University, Pittsburgh, United States
    Present address
    Istituto Italiano di Tecnologia, Genova, Italy
    Contribution
    Conceptualization, Software, Formal analysis, Supervision, Funding acquisition, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8629-0173
  2. Zachary A McDargh

    Department of Physics, Carnegie Mellon University, Pittsburgh, United States
    Present address
    Department of Chemical Engineering, Columbia University, New York, United States
    Contribution
    Conceptualization, Software, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9022-5593
  3. Markus Deserno

    Department of Physics, Carnegie Mellon University, Pittsburgh, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Investigation, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    deserno@andrew.cmu.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5692-1595

Funding

National Science Foundation (NSF CHE #1464926)

  • Markus Deserno

Carnegie Mellon University (Center of Excellence funding)

  • Markus Deserno

Horizon 2020 - Research and Innovation Framework Programme (Marie Sklodowska-Curie grant agreement no. 754490)

  • Martina Pannuzzo

National Science Foundation (NSF CHE #1764257)

  • Markus Deserno

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Marijn Ford, Aurelian Roux, Joshua Zimmerberg, Fred Lanni, Tina Lee, Adam Linstedt, Mathias Lösche, Martin Müller, and Antonio Raudino for discussions and critical comments. This work was supported by NSF Grants CHE #1464926, CHE #1764257, a Carnegie Mellon MCS/CIT Postdoctoral Fellowship, and the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement no. 754490.

Senior Editor

  1. Vivek Malhotra, The Barcelona Institute of Science and Technology, Spain

Reviewing Editor

  1. Michael M Kozlov, Tel Aviv University, Israel

Publication history

  1. Received: June 21, 2018
  2. Accepted: December 13, 2018
  3. Accepted Manuscript published: December 18, 2018 (version 1)
  4. Version of Record published: January 31, 2019 (version 2)

Copyright

© 2018, Pannuzzo et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,395
    Page views
  • 218
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)