Cryo-EM structures of membrane bound dynamin in a post-hydrolysis state primed for membrane fission

SUMMARY

Dynamin assembles as a helical polymer at the neck of budding endocytic vesicles, constricting the underlying membrane as it progresses through the GTPase cycle to sever vesicles from the plasma membrane. While atomic models of the dynamin helical polymer bound to GTP analogs define earlier stages of membrane constriction, there are no atomic models of the assembled state post GTP hydrolysis. Here we used cryo-EM methods to determine atomic structures of the dynamin helical polymer assembled on lipid tubules, akin to necks of budding endocytic vesicles, in GDP-bound, super-constricted state. In this state, dynamin is assembled as a 2-start helix with an inner lumen of 3.4 nm, primed for spontaneous fission. Additionally, by cryo-electron tomography we trapped dynamin helical assemblies within HeLa cells using the GTPase-defective dynamin K44A mutant and observed diverse dynamin helices, demonstrating dynamin can accommodate a range of assembled complexes in cells that likely precede membrane fission.

Graphical Abstract

eTOC

Dynamin severs endocytic vesicles. Jimah and Kundu et al. find dynamin in a post GTP-hydrolysis state organizes as a super-constricted 2-start helix on membranes, primed for spontaneous membrane fission and cryo-ET of dynamin assemblies in vivo show it accommodates a range of diameters that enable membrane constriction and fission.

INTRODUCTION

Membrane fission events are ubiquitous and critical for cellular function^1–3. Dynamin, a large GTPase, was the first identified membrane fission protein and is the founding member of a superfamily of structurally-related, GTP-driven, membrane-remodeling proteins that includes Drp1, mitofusin, Atlastin, OPA1, and GBP^4–6. Dynamin functions in endocytosis, synaptic membrane recycling, and the scission of vesicles from the Golgi^2,7. Mutations or dysregulation of dynamins in humans is associated with neuropathies, myopathies, and cancers^8–12. Additionally, viruses such as SARS-CoV-2 and HIV-1 hijack dynamin-dependent endocytic pathways for host cell invasion, a process that can be blocked by dynamin inhibitors^13–16. Dynamin’s key cellular roles necessitate understanding the structural basis of its function, which will in turn help illuminate the common processes shared among dynamin superfamily proteins.

The assembly and fission activity of dynamin at the necks of budding vesicles is coordinated by dynamin’s five domains (Figures 1A and 1B)^7,17. The C-terminal proline rich domain (PRD) interacts with proteins that direct dynamin to sites of clathrin-mediated endocytosis^18,19. The pleckstrin homology (PH) domain binds to negatively charged phospholipids, including phosphoinositide-4,5-bisphosphate (PIP₂)²⁰. The stalk domain has three interfaces that enable dynamin monomers to assemble as helical polymers on the necks of budding vesicles^17,21–23. The GTPase domain binds and hydrolyzes GTP to cause structural changes in the contiguous bundle-signaling element (BSE) domain, referred to as the powerstroke of dynamin²⁴. Based on crystal structures of the truncated GTPase-BSE constructs, the BSE is in an “up” position in the GTP-bound state and swings down relative to the GTPase domain upon GTP hydrolysis^24,25. These changes are believed to propagate to the stalk domain to cause severing of the encompassed membrane through further constriction and disassembly of the dynamin polymer^17,24–26.

Figure 1. Cryo-EM structures of dynamin helical assemblies on lipid tubules in the super-constricted state.

(A) Sketch of membrane fission during endocytosis, as dynamin progresses through the GTPase cycle. (B) Domain organization of dynamin showing the GTPase domain (green), Bundle Signaling Element (BSE, pink), Stalk (blue), Pleckstrin homology (PH) domain (orange), and Proline Rich Domain (PRD, grey). Atomic model of a dynamin monomer generated from our super-constricted cryo-EM structure, color-coded by domain. (C) Cryo-EM map of ΔPRD dynamin-K44A incubated with GTP and organized around lipid tubules in the two-start super-constricted state (^ΔPRD-K44Adyn^GDP). (D) Side and top views of ^ΔPRD-K44Adyn^GDP cryo-EM map. Domains of dynamin are color-coded as in (B). (E) Atomic model of ^ΔPRD-K44Adyn^GDP tetramer, with one dimer color-coded by domain and the other in grey, showing interfaces 1, 2, and 3 that drive oligomerization.

See also Figures S1–S4, and Table 1.

In the presence of liposomes in vitro, dynamin forms protein-decorated lipid tubules akin to the necks of budding endocytic vesicles^6,27–29. This behavior has allowed for structural determination by cryo-electron microscopy (cryo-EM), revealing molecular mechanisms of the distinct stages of the GTPase cycle in dynamin-mediated membrane fission. In the presence of non-hydrolyzable GTP analogs, dynamin helices constrict the opposing lipid bilayers to a ~7 nm inner luminal diameter. The high-resolution structure of this assembly, termed the “constricted state” has been published^17,30. With GTP itself, dynamin constricts further still to an inner luminal diameter of ~3–4 nm, referred to as the super-constricted state^17,28,31,32. However, to date, no high-resolution structure of dynamin in this state on lipid tubules has been solved.

After GTP hydrolysis, dynamin-decorated lipid tubes fragment and dynamin rapidly dissociates from the membrane^{24,27,29,31–37}, which has impeded previous attempts at high-resolution structural determination. Here, we took advantage of a hydrolysis-deficient point mutant, K44A, to allow us to determine the high-resolution structure of the dynamin helical polymer bound to lipid tubules in the super-constricted, post-hydrolysis state. Our high-resolution helical reconstructions of full-length and ΔPRD dynamin assembled on lipid tubules in the GDP-bound state provides molecular details of the 2-start super-constricted state at resolutions of 3.6 and 3.3 Å, respectively. Surprisingly the BSE in the GDP-bound state remains in an up position similar to the GTP-bound state.

Despite the detailed structural insight provided by studies of in vitro lipid tubules, it is unclear how these structures correspond to dynamin behavior in the cell. Studying endocytic structures in situ provides a deeper understanding of endocytosis and complement in vitro structural studies.^38,39. Therefore, in this study, we also applied cryo-electron tomography (cryo-ET) to observe the organization of dynamin helical polymers within actual cells. We found that dynamin forms flexible helical assemblies of different diameters, allowing for a dynamic range to constrict the necks of budding vesicles.

Altogether, this study presents mechanistic and structural details of how subtle changes in the dynamin monomer translates to large conformational changes of the dynamin polymer leading to significant membrane constriction and the penultimate state prior to membrane fission. We further confirm that although the PRD is crucial for dynamin’s cellular location, it does not play a role in dynamin’s mechanochemical properties. We also show that even in a crowded super-constricted state the PH domain remains highly flexible, accommodating a range of membrane curvatures, as evident from the diameter range of dynamin polymers in vitro and in situ.

RESULTS

Dynamin K44A forms super-constricted helical polymers on lipid tubules

We sought to determine the structural details of dynamin organization on membranes just before fission (Figure 1A). Upon addition of GTP, dynamin-decorated lipid tubules immediately undergo rapid narrowing to the super-constricted state followed by disassembly, as observed by cryo-EM images of dynamin tubes frozen within ~150 ms after mixing with GTP (Figures S1A and S1B)^17,40. Therefore, to capture dynamin in a prefission state, we utilized a hydrolysis-deficient mutant (^K44Adyn) that hydrolyzes GTP more slowly than wild type dynamin (^WTdyn)⁴¹. ^K44Adyn produces dynamin-decorated lipid tubules with similar outer diameters to ^WTdyn under various nucleotide treatment (APO, GTP and GDP) (Figures S1C and S1D). For example, the APO and GDP-bound states of both ^K44Adyn and ^WTdyn have an outer of ~50 nm and in GTP treated tubes an outer diameter is ~38 nm. However, the number of super-constricted lipid tubes treated with GTP was much higher for ^K44Adyn GTP than ^WTdyn, due to the rapid disassembly and fission of ^WTdyn lipid tubes. The ^K44Adyn decorated lipid tubes incubated with GTP allowed us to calculate a high-resolution structure of dynamin in the super-constricted state, with opposing inner leaflets of the bilayer only 3.4 nm apart as determined by cryo-EM (Figures 1C, 1D, S1C–E and S2–S4). In addition, typical super-constricted assemblies were observed with other negatively charged lipids such as PIP₂ and as shown previously, the super-constricted state is a 2-start helix (Figure S1F–H)^17,28.

We determined structures of both the ΔPRD:K44A mutant (^ΔPRD-K44Adyn) (Figures 1C–1E, S4A, S4C and S4D) as well as full-length:K44A (^FL-K44Adyn) (Figures S4B and S4D). Both ^FL-K44Adyn and ^ΔPRD-K44Adyn structures are organized as 2-start helices on lipid tubules, suggesting that a 2-start helix is energetically favorable for super-constriction (Figure 1C). The helical organization for ^ΔPRD-K44Adyn and ^FL-K44Adyn are defined by a rise of 14.675 Å and 14.673 Å and a twist of 26.146° and 26.155°, respectively (Figure S2 and Table 1). Although we did not resolve the proline rich domain, the essentially identical structures (Figure S4D and S4E) suggest that the PRD does not cause different organization of the dynamin domains during membrane constriction.

Table 1.

Cryo-EM data collection, refinement, and validation, related to STAR Methods.

	ΔPRD-K44AdynGDP	ΔPRD-K44AdynGDP	ΔPRD-K44AdynGDP	ΔPRD-K44AdynGDP	ΔPRD-K44AdynGDP	FL-K44AdynGDP	FL-K44AdynGDP	FL-K44AdynGDP
	helical	z-clipped	stalk PH1	stalk PH2	tetramer	helical	z-clipped	tetramer
	EMD-40903	EMD-40861	EMD-40901	EMD-40902	EMD-40861	EMD-40946	EMD-40942	EMD-40942
	PDB:8SZ8	PDB: 8SXZ	PDB: 8SZ4	PDB:8SZ7	PDB:8TYN	PDB:8T0R	PDB:8T0K	PDB:8TYM
Data collection & processing
Magnification	130,000	130,000	130,000	130,000	130,000	130,000	130,000	130,000
Voltage (kV)	300	300	300	300	300	300	300	300
Electron exposure (e−/Å2)	60	60	60	60	60	60	60	60
Defocus range (μm)	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8	0.6 – 1.8
Pixel size K2 dataset (Å)	0.5371	0.5371	0.5371	0.5371	0.5371	0.5371	0.5371	0.5371
Symmetry imposed	helical	helical	helical	helical	helical	helical	helical	helical
Initial particle images (no.)	2,692,477	2,692,477	2,692,477	2,692,477	2,692,477	5,892,784	5,892,784	5,892,784
Final particle images (no.)	609,714	609,714	609,714	609,714	609,714	626,859	626,859	626,859
Map resolution (Å)	3.55	3.26	2.86	2.84	3.26	3.97	3.58	3.58
FSC threshold	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	6DLV
Model resolution (Å) at 0.143 FSC threshold	4.6	4.2	3.7	6	3.8	4.6	4.6	4.1
Model resolution (Å) at 0.5 FSC threshold	9	8.9	6.1	11.6	8.5	9	9	8.6
Map-model CC	0.43	0.37	0.26	0.2	0.35	0.52	0.44	0.41
Model composition
Non-hydrogen atoms	22,540	22,540	2,162	2,005	22,544	22,544	22,544	22,544
Protein residues	2,783	2,783	259	240	2,784	2,784	2,784	2,784
Ligands	4	4	0	0	4	4	4	4
R.M.S. deviations
Bond length (Å)	0.003	0.003	0.003	0.003	0.003	0.007	0.007	0.003
Bond angle (°)	0.541	0.541	0.589	0.478	0.644	0.755	0.755	0.678
Validation
MolProbity score	1.00	1.00	0.98	0.97	1.09	1.08	1.08	1.08
Clashscore	2.23	2.23	2.07	1.99	2.83	2.91	2.91	2.91
Poor rotamers (%)	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0
Ramachandran plot
Favored (%)	98.48	98.48	98.01	99.14	97.93	98.84	98.84	98.91
Allowed (%)	1.52	1.52	1.99	0.86	2.07	1.16	1.16	1.09
Disallowed (%)	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0

The super-constricted state has 13.8 units per turn in contrast with the 15.2 units per turn observed in the constricted, GMPPCP-bound state (^ΔPRDdyn^PCP)¹⁷ (Figure S4C). The three interfaces that enable dynamin oligomerization were largely similar in the super-constricted state compared to the previously reported GMPPCP-bound constricted state (Figures 1E, S5A and S5B). Therefore, the reduced diameter of the super-constricted state likely results from extension along the helical axis (i.e. a larger helical pitch) rather than large conformational changes within the building block of the dynamin polymers.

Dynamin-lipid tubes achieve super-constriction through GTP hydrolysis

The density associated with the nucleotide in the super-constricted dynamin lipid tubules accommodates only a GDP molecule, lacking density for a gamma phosphate, (Figures 2A, 2B and S3H). This finding, while initially surprising, was further supported by results of GTPase enzyme kinetics showing that ^K44Adyn is capable of GTP hydrolysis, though at a much slower rate compared to wild type dynamin (Figure 2C). Super-constriction was only observed in samples treated with GTP but not with GDP (Figures S1C and S1D), suggesting that GTP hydrolysis is important for super-constriction. To determine the state of the nucleotide after ^K44Adyn tubes are incubated with GTP, the sample was centrifuged to enrich for dynamin-decorated lipid tubules in the pellet and examined by HPLC, UV spectroscopy, and mass spectrometry. The percentage of protein in the pellet and supernatant are 85% and 15% respectively as shown by SDS-PAGE (Figure S1I). The sample contains GDP and GTP, and as expected, the ratio of GDP to GTP is significantly higher in the pellet (1.2:1), compared to the supernatant (1:6), further confirming hydrolysis occurs (Figure S1I). We will therefore refer to our structures as ^FL-K44Adyn^GDP and ^ΔPRD-K44Adyn^GDP. The bound GDP is in the South conformation, similar to previous structures of nucleotides bound to dynamin⁴². The strong nucleotide density also suggests a high occupancy of GDP in the dynamin helical polymer. These represent high-resolution maps of dynamin assembled on lipid tubules in a post-GTP-hydrolysis state. The presence of GDP suggests GTP hydrolysis drives super-constriction and agrees with a previous low-resolution super-constricted structure of dynamin following GTP hydrolysis^17,28 (Figures 1 and 2).

Figure 2. GTP hydrolysis to GDP leads to super-constriction of dynamin-decorated lipid tubules.

(A) Model of GTPase domain dimer of super-constricted ^ΔPRD-K44Adyn^GDP with trans-stabilizing residues shown in boxed regions 1 and 3, and nucleotide-interacting residues in box 2. The two GTPase domains are colored green and purple, respectively and the cryo-EM density around the nucleotide is shown as a black mesh. (B) Atomic model of GDP fitted in cryo-EM map of ^ΔPRD-K44Adyn^GDP with the cryo-EM density in mesh. (C) Michaelis-Menten plot of wild type ΔPRD dynamin (black) and ΔPRD dynamin-K44A (blue) GTPase activity, in the presence of PS liposomes. Velocity measurement from three experiments shown as mean +/− SEM. (D) Overlay of the GG BSE dimer of dynamin in the GDP-bound super-constricted state (^ΔPRD-K44Adyn^GDP); (GTPase domain in green and BSE in pink) compared to the GMPPCP-bound constricted state (Δ^PRDdyn^PCP, grey). (E) The three-helix bundle or BSE resides in an up position in the super-constricted (GDP) state of assembled dynamin in contrast to the crystal structures of dynamin in the GDP state with the BSE in the down position (5D3Q).

See also Figures S3H, Video S1.

The GTPase domain dimerization interface and nucleotide binding residues are similar when compared to the constricted, GMPPCP-bound model (^ΔPRDdyn^PCP) (Figures 2A and 2D). Surprisingly, we found that the BSE remains in the “up” position, more similar to the crystal structure bound to GTP than GDP (Figure 2E). This suggests the powerstroke is hindered in the ^ΔPRD-K44Adyn helical polymer. However, while not undergoing a powerstroke, the BSE does make a 4.7 Å upward swing away from the lipid bilayer compared to ^ΔPRDdyn^PCP (Figure 2D and Video S1). This change in the BSE orientation demonstrates that small changes in the GTPase domain upon GTP hydrolysis are propagated and amplified via the BSE to other dynamin domains in the assembled polymer to further constrict the membrane.

As with the constricted helical assembly of dynamin¹⁷, there is asymmetry in the dimer, with the BSE kinked in one monomer (Figure 2D). The asymmetry extends to the PH domains, where the PH domain associated with the kinked monomer is closer to the stalk (10 Å), better resolved, and likely more stable (Figure S3C). Conversely, the other PH monomer is further away from the stalk (31 Å), less resolved (Figure S3D), and more elastic.

PH domains of super-constricted dynamin move inward due to a swing in the stalk and BSE domains

Consistent with the hypothesis that progression through the GTPase cycle directs dynamin-mediated membrane constriction and fission, the GMPPCP-bound, constricted state and post-GTP-hydrolysis, super-constricted states have inner luminal diameters of 7.4 nm and 3.4 nm, respectively (Figure 3A)^2,17,28. When the GTPase dimer interface of the ^ΔPRD-K44Adyn^GDP is overlayed with ^ΔPRDdyn^PCP, we observe that the stalk and PH domains swing towards the membrane in the super-constricted state (Figure 3B). Overall, these GTP-induced conformational changes increase the helical pitch and allow for a 2-start helix, resulting in tubular constriction from an inner luminal diameter of 7.4 to 3.4 nm.

Figure 3. The stalk and PH domains swing toward the encompassed lipid tubule to help drive super-constriction.

(A) Comparison of dynamin cryo-EM helical structures in the constricted and super-constricted states, highlighting the decrease in outer diameter from 40 nm to 36 nm and inner lumen from 7.4 nm to 3.4 nm. (B) Overlay of super-constricted ^ΔPRD-K44Adyn^GDP (domains colored as in Figure 1) and constricted ^ΔPRDdyn^PCP dimers (white) about the GTPase dimer interface. The stalk moves toward the membrane upon GTP hydrolysis; 6.3 Å for the kinked, stable monomer, (indicated by asterisk) and 9.6 Å for the flexible monomer. The variable loops of the PH domain are color-coded as shown in C. (C) Cryo-EM map (outlined in white) and model of ^ΔPRD-K44Adyn^GDP partial stalk (blue) and PH domain (orange) obtained after local refinement of ^ΔPRD-K44Adyn^GDP. PH domain variable loops 1–4 are colored red, purple, green, and blue respectively in B and C. (D) Surface map showing the electrostatic surface potential. Blue is positive and red is negative.

See also Figures S2 and S3.

Compared to previous dynamin helical structures bound to lipid^17,28, the resolution of the PH domain region bound to lipid is significantly improved due to the stabilizing presence of membrane and local refinement strategies (Figures 3C, S2, and S3). It is clear that lipid binding stabilizes the PH domain since the PH domain was completely unresolved in the 2-start GMPPCP-bound structure of the dynamin helix in the absence of lipid membranes⁴³. The density around the PH domain allowed us to accurately dock it into the cryo-EM map (Figure 3C). The RMSD values of the PH domains in our map (PH1 and PH2) compared to that in a previous crystal structure (PDB 3SNH²¹) are 1.009 and 1.081, respectively (Figures S3C and S3D). Variable loops 1, 2, 3, and 4, known PH domain features, are clearly observed, with mostly loop 1 (red, amino acid residues 531 – 537) and to some extent loop 4 (blue, amino acid residue 576–583) interacting with the lipid membrane. The lipid interaction is enhanced by a positively charged electrostatic surface potential of the PH domain on the membrane-facing surface, further stabilizing the interaction of dynamin to negatively charged lipid (Figure 3D).

Residues proximal to interface 3 facilitate super-constriction

Though the interfaces that enable dynamin oligomerization are almost identical, an amino acid interaction is present in the super-constricted state that is not found in the constricted state (Figure 4A). This interaction occurs between residues L402 and D406 around interface 3 (Figure 4A). A D406A mutation, designed to disrupt the L402/D406 interaction proximal to interface 3, impaired endocytosis (Figures 4B and S5C). Transferrin uptake, the measure for endocytosis, in cells expressing dynamin K406A was intermediate compared to the high levels of uptake in cells expressing wild type dynamin and the low levels of uptake in cells expressing dynamin K44A. Interestingly, L402 is conserved in human, bovine, rat and mouse dynamin, as well as human Drp1. D406 is conserved in the aforementioned dynamin proteins, except it is a similar residue, E406, in Drp1. This suggests the importance of the L402 and D406 interface residues in dynamin, likely in forming a critical hydrogen bond. This finding suggests that specific and subtle changes in interface 3 allows for dynamin super-constriction and subsequent vesicle fission necessary for endocytosis.

Figure 4. A unique amino-acid interaction observed in the super-constriction state is crucial for basal levels of endocytosis.

(A) Interactions between D406 and L402 observed in interface 3 of the GDP-bound super-constricted state, ^ΔPRD-K44Adyn^GDP. Top, ^ΔPRD-K44Adyn^GDP tetramer highlighting interface 3. Bottom, H-bond between D406 and L402 shown in model without (below) and with density in white (above). (B) Transferrin uptake assays quantifying endocytosis in cells transfected with WT, K44A, and D406A dynamin. Black bars, cells expressing dynamin plasmids (WT, K44A, D406A). Blue bars, cells not expressing dynamin plasmids. In each experiment, samples were run in triplicate (shown as dots). The error bars are standard deviation. The experiment was performed twice with comparable results. Pair Sample T-test is comparing intra-experiment triplicates, significance marked with *p=0.0007, **p=0.003.

See also Figure S5C.

Dynamin assemblies in vivo are organized in an assortment of helical polymers

To visualize the dynamin assembly within cellular contexts we transfected HeLa cells with a GFP-tagged version of the K44A mutant and examined the cells by cryo-ET. The transfected cells were grown on cryo-EM grids and unroofed to retain the plasma membrane on the grid³⁹. From these cells, we obtained 7 cryo-electron tomograms of regions with dynamin assemblies (Videos S2–S5). One of the tomograms was especially well-positioned, showing the dynamin helical polymer adjacent to a clathrin-coated vesicle and numerous actin filaments (Figure 5 and Videos S2–S3). Though we were not able to collect data on enough dynamin assemblies for subtomogram averaging, we were able to measure the outer diameter, inner luminal diameter, and inter-rung distance of several individual dynamin assemblies in situ. These assemblies exhibited a diversity of outer diameters, ranging from ~35 to 80 nm, inner luminal diameters, ranging from ~4 to 40 nm, and inter-rung distances, ranging from ~8 to 20 nm, consistent with parameters observed in vitro in the apo-, GMPPCP-, and GDP bound-states (Figure 5D). Thus, dynamin K44A can accommodate a range of conformations in the native environment of the cell. This work provides the foundation for examining molecular machines composed of dynamin assemblies and partner proteins in vivo by subtomogram averaging methods in the future.

Figure 5. Dynamin helical polymer observed within a HeLa cell.

(A) Slice through a tomogram revealing a dynamin-decorated membrane tube with a clathrin coat at the tip. (B) Segmented tomogram with dynamin, clathrin, vesicle, and actin colored in pink, yellow, grey, and green, respectively. (C) Model of a dynamin helical polymer (rainbow-colored) fit into tomogram. Scale bar, 50 nm. (D) Measurements of outer and inner luminal diameters and the inter-rung distance of dynamin helical structures in cells. The lengths of these measurements are shown as a scatter plot and mean ±S.D.

See also Videos S2–5

DISCUSSION

Although dynamin is the prototypical membrane fission protein, the conformational changes that drive the culminating steps of fission remain poorly understood². A model for dynamin-membrane fission postulates that dynamin assembles on lipid tubules and constricts in two stages: once upon GTP-binding—as observed in structures of the constricted, GMPPCP-bound state—and then once again upon GTP hydrolysis (Figure 6 and Video S6). Here, we determined two distinct structures of dynamin in the lipid-bound, post-hydrolysis state: full-length and ΔPRD dynamin-K44A on PS lipid tubules. Further, we explored the in vivo applicability of this model using cryo-electron tomograms of dynamin assemblies in HeLa cells.

Figure 6. Model for dynamin organization on lipid tubules in the 1-start GTP-bound constricted state and the 2-start GDP-bound super-constricted state.

(A) Top view highlighting dynamin-mediated constriction of the underlying membrane (yellow) from an inner lumen of 7.4 nm in the ^ΔPRDdyn^PCP model to 3.4 nm in the ^ΔPRD-K44Adyn^GDP model. (B) Side view illustrating an increase in pitch upon GTP hydrolysis that allows for a second rung of dynamin to assemble. GTPase domains dimerizes (G dimer) between rungs of the helix.

See also Video S6

The high-resolution cryo-EM maps of full-length and ΔPRD dynamin were similar, with no changes in domain organization. While the PRD consists of 121 amino acids and is flexible and disordered, it did not affect dynamin domain organization in the post GTP-hydrolysis state and both polymers are organized as 2-start helical polymers, supporting previous work showing that this helical organization enables super-constriction^17,28,43. While a recent structure of the dynamin helical polymer bound to GMPPCP and in the absence of lipid membranes was also organized as a 2-start helix, it had an outer diameter intermediate between the lipid-bound GMPPCP and GDP-bound models, suggesting that the presence of a membrane resists constriction of the dynamin helical polymer in the GTP-bound state (Figure S6). Our super-constricted structures likely represent the penultimate stage of dynamin-mediated fission. With an inner luminal diameter of 3.4 nm, membranes will spontaneously fission upon disassembly of the stabilizing protein coat⁴⁴.

At the level of domain organization, progression from constriction to super-constriction relied on the BSE domains swinging away from the lipid membrane, while the GTPase, stalk, and PH domains move towards the membrane. At the level of amino acid residues, the interfaces required for dynamin oligomerization are similar between the constricted and super-constricted states. However, we identified an interaction in the super-constricted state between L402 and D406, around interface 3, that impairs endocytosis when mutated in living cells.

Our structure of ^ΔPRD-K44Adyn^GDP in the super-constricted state supports the model that fission occurs after hydrolysis. Curiously, the positioning of the BSE domain in super-constricted dynamin-K44A GDP-bound structures (^ΔPRD-K44Adyn^GDP and ^FL-K44Adyn^GDP) adopts a conformation similar to the GTP-bound state observed in the ^ΔPRDdyn^PCP helical polymer¹⁷. In crystal structures, the BSE extends out from the GTPase domain in the GTP-bound state and swings down closer to the GTPase domain in the GDP-bound and transition states^24,25. The fact that the BSE remains extended in our assembled GDP-bound state suggests potential energy is stored in this state which may be released when the BSE swings down upon further rounds GTP hydrolysis. The positioning of the BSE in the down position would lead to the disorder of the dynamin polymer. Fission may then occur as the protein scaffold disassembles allowing for the un-constrained lipid to undergo spontaneous fission⁴⁴.

In vivo, we observe that the dynamin helical polymer is quite flexible and organizes in a variety of helical symmetries, consistent with those observed in in vitro reconstitutions (Figures S1 and Videos S2–S5). The flexibility of the PH domains likely contributes to dynamin’s ability to bind different lipid curvatures. This would facilitate a progression through the series of constriction intermediates required to cut lipid tubules. The transition from apo to super-constricted also appears to involve a state of disorder on the membrane as shown by the cryo-EM images frozen within milliseconds. In this case, dynamin may disassemble from the membrane and rebind, or remain loosely attached to the membrane in small oligomers that rearrange into a tighter super-constricted state. While we were unable to collect enough data for full reconstructions, this strategy is promising for future cryo-ET and subtomogram averaging studies of dynamin in situ.

This study provides structural, biochemical, and mechanistic details of dynamin membrane fission pathways. These findings have implications for numerous cellular processes that require dynamin such as constitutive endocytosis, synaptic vesicle recycling, and vesiculation from the Golgi. These high-resolution structures will enable investigations into the structural basis of mutations in dynamin linked to diseases. Furthermore, the better-resolved PH domain will provide insights into the mechanism of other PH-containing proteins and how they interact with membranes building on previous studies of PH-interaction with specific phospholipids²⁰. The unique context of a PH domain in contact with PS-containing membranes will inform future computational modeling and molecular dynamics simulations of PH domain-membrane interactions⁴⁵. For example, our experimentally determined model supports a role for the variable loops 1 and 4 of the PH domain for interactions with the membrane as predicted in all-atom simulations of PH domain interacting with membranes⁴⁵. Our cryo-EM structure depicts that variable loop 4 is in close proximity with membranes and likely functions in membrane binding, and dynamin’s dissociation from the membrane. Our assertion on the role of variable loop 4 is supported by recent experimental evidence, including membrane binding, that implicate variable loop 4 in interacting with membranes⁴⁶.

Additional questions remain about the mechanism of dynamin beyond the scope of this study. What is the structural basis of the dynamin hemifission intermediate?⁴⁷ How are dynamin helical polymers organized in native cellular membranes with the numerous binding partners of dynamin? As we gather more “snapshots” of this process in vitro and in situ, we will be able to construct a full model of dynamin-mediated membrane fission from start to finish.

Limitations of the Study

Our study is limited in several ways. First, for our cryo-EM studies of dynamin assemblies on membranes used liposomes composed of only phosphatidylserine (PS) to determine the atomic models of ^FL-K44Adyn^GDP and ^ΔPR-K44Adyn^GDP. For future cryo-EM studies, it would be informative to use other liposome compositions as models of the plasma membranes. Second, while this work presents cryo-ET models of dynamin assemblies within actual cells, we only used the HeLa cell line for our investigations. Our findings show that dynamin helices in situ were of similar size ranges as those observed in in vitro reconstituted systems with protein and liposomes. It would be informative to investigate, by cryo-ET, dynamin assemblies in a variety of cell types beyond the HeLa cells used in this study.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Additional information and request for reagents and resources should be directed to the lead contact, Dr. Jenny E. Hinshaw (jenny.hinshaw@nih.gov).

Materials availability

Plasmids generated in this study are available upon request to the lead contact.

Data and code availability

Structural data supporting this study have been deposited to the Protein Data Bank (PDB) and the Electron Microscopy Data Bank (EMDB). The access codes for the atomic models and cryo-EM maps have been provided for ^ΔPRD-K44Adyn^GDP z-clipped (EMD-40861, PDB: 8SXZ), ^ΔPRD-K44Adyn^GDP stalk PH1 (EMD-40901, PDB: 8SZ4), ^ΔPRD-K44Adyn^GDP stalk PH2 (EMD-40902, PDB:8SZ7), ^ΔPRD-K44Adyn^GDP helical (EMD-40903, PDB:8SZ8), ^ΔPRD-K44Adyn^GDP tetramer (EMD-40861, PDB:8TYN), ^FL-K44Adyn^GDP z-clipped (EMD-40942, PDB:8T0K),^FL-K44Adyn^GDP helical (EMD-40946, PDB:8T0R), and ^FL-K44Adyn^GDP tetramer (EMD-40942, PDB:8TYM).

• Any additional information required to reanalyze the data reported in this work paper is available from the Lead Contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

The E. coli strains BL21 and DH5 alpha were used. Also, FreeStyle HEK 293F cells were used and HeLa cells. See Key resource table for information on these bacterial and human cell lines.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
E. coli BL21 (DE3) Competent Cells	Novagen	Cat. #69450-3
E. coli DH5 alpha competent cells	New England Biolabs	Cat. #C2987H
Chemicals, peptides, and recombinant proteins
10 nm BSA gold tracer solution	Electron Microscopy Sciences	Cat. # 25486
DMEM	Life Technologies	Cat. # 31053-036
IPTG (isopropylthio-β-galactoside)	ThermoFisher Scientific	Cat. # 15529019
HEPES	Millipore Sigma	Cat. # 3375
Imidazole	Millipore Sigma	Cat. # I5513
2-Mercaptoethanol (BME)	Millipore Sigma	Cat. # M6250
Potassium chloride (KCl)	Millipore Sigma	Cat. # P3911
Protease inhibitor cocktail	Millipore Sigma	Cat. # 11836170001
Deoxyribonuclease I (DNase I)	Millipore Sigma	Cat. # 11284932001
Lysozyme	Millipore Sigma	Cat. # 1.05281
EGTA	Millipore Sigma	Cat. # 324626
FreeStyle™ 293 Expression Medium	ThermoFisher Scientific	Cat. # 12338026
PEI transfection reagent	Sigma-Aldrich	Cat. #408727100ML
293Fectin™ Transfection Reagent (1ml)	Invitrogen	12347-019
1,2-dioleoyl-sn-glycero-3-phospho-L-serine in chloroform	Avanti	Cat. # 840035
Fibronectin	Millipore Sigma	Cat. # F1141
Alexa-Fluor 647 bound transferrin	Invitrogen	Cat. # T23366
Deposited data
ΔPRDdynGDP z-clipped	This paper	EMD-40861, PDB: 8SXZ
ΔPRDdynGDP stalk PH1	This paper	EMD-40901, PDB: 8SZ4
ΔPRDdynGDP stalk PH2	This paper	EMD-40902, PDB:8SZ7
ΔPRDdynGDP helical	This paper	EMD-40903, PDB:8SZ8
ΔPRDdynGDP z-tetramer	This paper	EMD-40861, PDB: 8TYN
FLdynGDP z-clipped	This paper	EMD-40942, PDB:8T0K
FLdynGDP helical	This paper	EMD-40946, PDB:8T0R
FLdynGDP tetramer	This paper	EMD-40942, PDB:8TYM
Experimental models: Cell lines
HeLa	ATCC	Cat. # CCL-2
FreeStyle HEK 293-F	ThermoFisher Scientific	Cat. # R79007
Recombinant DNA
pET-47b (+) DNA	Millipore Sigma	Cat. # 71461-3
dynamin 1 pcDNA3.1	Addgene	Cat. # 34682
Software and algorithms
ImageJ	Schindelin et al.48	Fiji; RRID:SCR_002285
SerialEM	Mastronarde49	SerialEM; RRID: SCR_017293
IMOD	Kremer et al.50	IMOD; RRID: SCR_003297
GraphPad Prism 8.0.2	https://www.graphpad.com/	RRID:SCR_002798
FACSDiva Software	BD Biosciences	RRID:SCR_001456
UCSF Chimera	Pettersen et al., 200451	https://www.cgl.ucsf.edu/chimera/
ChimeraX	Goddard et al., 201852	https://www.cgl.ucsf.edu/chimerax/
Coot	Emsley et al.53	https://www2.mrclmb.cam.ac.uk/personal/pemsley/coot
cryoSPARC	Punjani et al., 201754	https://cryosparc.com/
PHENIX	Adams et al., 201055	https://phenixonline.org/documentation/index.html
The PyMOL Molecular Graphics System	Schrödinger, LLC	https://pymol.org/2/
PDBePISA	Krissinel and Henrick, 200756	https://www.ebi.ac.uk/pdbe/pisa/
MolProbity	Williams et al.,57	http://molprobity.biochem.duke.edu/
Isolde	Croll58	https://isolde.cimr.cam.ac.uk
Rosetta	Wang et al.59	https://www.rosettacommons.org/software
HI3D program	Sun et al.60	https://jiang.bio.purdue.edu/hi3d/
Other
Nickel NTA Agarose Beads	GoldBio	Cat. #H-350-5
Enrich SEC 650 column	Bio-Rad	Cat. # 7801650
300-mesh Quantifoil R2/1 grids	Electron Microscopy Sciences	Cat. # Q325AR1
C-FLAT l .2UM HOLE 400 MESH GOLD	Electron Microscopy Sciences	Cat# CF413-100-AU

METHOD DETAILS

Dynamin ΔPRD expression and purification

Recombinant dynamin 1 ΔPRD and dynamin 1 K44A ΔPRD were cloned into a pET47b vector with an N-terminal 6xHis tag upstream of an HRV 3C proteolytic cleavage site and expressed in E. coli BL21. Culture was grown to OD₆₀₀= 0.6, cold shocked on ice for 30 minutes, and protein expression induced using 0.1 mM IPTG. Cells were harvested after 20 hours of growth at 18°C with shaking at 250 RPM and cell pellets stored at −80°C. Pellets were thawed in 50 mL lysis buffer (50 mM HEPES, pH 8.0, 150 mM KCl, 10 mM imidazole, 5 mM BME) with protease inhibitors (Roche), 10 μg/mL DNase I, and 1 mg/mL lysozyme. Resuspended cells were incubated at 4°C for 30 min and lysed using a One Shot French press. Lysate was centrifuged for 1 hour at 230,000 × g and the supernatant was purified by affinity chromatography using Ni-NTA resin. Protein was evaluated by SDS-PAGE with Coomassie staining, and fractions containing dynamin were pooled, combined with 10xHis-tagged HRV 3C protease, and dialyzed overnight at 4°C into modified HCB150 buffer (20 mM HEPES, pH 7.3, 150 mM KCl, 2 mM EGTA, 1 mM MgCl₂, 5 mM BME, 5.7 mM imidazole). Dialyzed protein was purified again with Ni-NTA resin to remove the protease and uncleaved dynamin. Fractions containing cleaved dynamin were pooled and concentrated to 500 μL using an Amicon centrifugal filter unit. The sample was further purified by an S650 size exclusion column (BioRad) with HCB150 (20 mM HEPES, pH 7.3, 150 mM KCl, 2 mM EGTA, 1 mM MgCl₂, 1 mM DTT). Fractions judged to contain dynamin at >95% purity by SDS-PAGE and Coomassie staining were pooled, flash frozen in liquid nitrogen, and stored at −80°C. For rapid freezing experiments with Spotiton⁴⁰, recombinant dynamin 1 ΔPRD was generated in Sf9 insect cells using Bac-to-Bac Baculovirus Expression System (ThermoFisher Scientific). Briefly, post inoculating with recombinant baculovirus (cell density of 1.6 × 10⁶ with 1/100 volume of virus/final volume of medium), the cells were grown for 72 h at 27°C, followed by centrifugation at 1000 × g, 5 min, 4°C. The pellet was resuspended in 30 ml of HSB150 buffer (50 mM HEPES, 150 mM KCl, 5 mM beta-mercaptoethanol,10 mM imidazole, [pH 8.0]) with protease inhibitor cocktail (Millipore Sigma) followed by cell lysis using sonication (5 s pulse-on and 15 s pulse-off; total time 8 min). Post lysis, cells were centrifuged (20,000 × g, 15 min), supernatant was collected, and the expressed protein was purified using Ni-NTA affinity chromatography⁶¹.

Full length (wild type) dynamin expression and purification

Full length dynamin 1 and dynamin 1 K44A were expressed with an N-terminal 6xHis tag in FreeStyle^™ HEK 293-F cells (ThermoFisher Scientific) using a pcDNA3.1 vector⁶². Cells were grown in suspension in Freestyle^™ 293 Expression Medium (ThermoFisher Scientific) to a density of ~1.2 × 10⁶ cells/mL. The expression vector was diluted to 10 μg/mL in expression media containing 1.5% polyethylenimine (Sigma Aldrich) and incubated at room temperature for 10 minutes. 7 mL of this mixture were then added to 70 mL of cell culture. Cells were grown at 37°C, 125 RPM, and 8% CO₂. After 48 hours, cells were harvested, and pellets were stored at −80°C. Pellets were thawed under 50 mL lysis buffer with protease inhibitors (Roche) and lysed using a One Shot French press. Lysate was centrifuged for 1 hour at 230,000 × g and the supernatant was purified by affinity chromatography using Ni-NTA resin. Fractions containing dynamin were pooled and concentrated to 500 μL using an Amicon centrifugal filter unit. The sample was further purified by an S650 size exclusion column (BioRad) with HCB150. Fractions judged to contain dynamin at >95% purity by SDS-PAGE and Coomassie staining were pooled, flash frozen in liquid nitrogen, and stored at −80°C.

Liposome preparation

Phosphatidylserine (PS) (50 μl of 10 mg/ml of 1,2-dioleoyl-sn-glycero-3-phospho-L-serine in chloroform, from Avanti) was dried under Argon gas. The lipid was dried overnight in a desiccator. Liposomes were produced by resuspending dried PS lipid film in HCB0 150 mM KCl buffer (20 mM HEPES pH 7.2, 1 mM MgCl₂, 2 mM EGTA, 1 mM DTT, 150 mM KCl), in warm water bath and vertexing intermittently, to produce a 2 mg/ml PS liposome stock. The liposomes were extruded ~19 times through a 0.8 μm pore-size polycarbonate membrane. For rapid freezing experiments with Spotiton⁴⁰, PS liposomes were generated as above except the lipid resuspension buffer was (50 mM HEPES, 150 mM KCl, 2 mM EGTA, 1 mM MgCl₂, 1 mM TCEP) and vesicles were extruded 21 times through 0.4 μm pore-size polycarbonate membrane.

Dynamin helical polymer formation on lipid tubules

Dynamin helical tubes were generated by addition 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (Avanti, PS) liposomes to 80 μg/mL to 1 mg/mL ^FL-K44Adynamin and ^ΔPRD-K44Adynamin in HCB150 and incubated for 1 hour at room temperature¹⁷. ^K44Adynamin-lipid tubes were incubated with 2 mM GTP or GDP for 30 minutes. ^WTdynamin was incubated 30 minutes with GDP and 5 sec with 2 mM GTP, due to rapid fission of ^WTdynamin tubes with GTP.

Cryo-EM sample preparation and imaging

3.5 μL of sample was applied to C-flat grids (EMS Cat. # CF413–100-AU) that had been glow discharged for 90 seconds. Sample was blotted with filter paper for 3 seconds and then plunged into liquid ethane using a Leica EM Grid Plunger set to 90 % humidity. Samples were stored in liquid nitrogen until imaged by cryo-EM. Imaging was performed on a Titan Krios microscope (ThermoFisher Scientific) at 300 kV with a K2 Summit camera (Gatan).

For rapid freezing experiments with Spotiton⁴⁰, dynamin decorated tubes were generated by incubating 3 μl of PS liposomes with 40 μl of protein (0.8 mg/ml, in 10 mM Tris, 10 mM KCl, 1 mM MgCl₂, [pH 7.4]) for 2 hours. Using the first piezo dispensing tip of the Spotiton, ~50 pL droplet of pre-formed dynamin decorated tubes was sprayed on the nanowire (self-wicking) grid. After 10 ms of the first droplet spayed, second droplet of GTP (2 mM, 4 mM) or buffer (control) is sprayed using the second piezo dispensing tip. The samples from both the spay gets mixed on the grid and the grid is vitrified under ~150 ms. Images were collected on F20/K2 camera and F20/TVIPS detector. The diameters of the samples were measured in Fiji⁴⁸.

Cryo-EM image processing

cryoSPARC was used for cryo-EM image processing. Pre-processing of micrographs involved patch motion correction and patch CTF estimation. The cryoSPARC filament picker was used to find dynamin-decorated lipid tubules in the micrographs, followed by particle extraction. 2D classification was used to remove bad particles. Cryosparc helical symmetry tools was used to determine the rise and twist of the dynamin-decorated lipid-tubules, which was agreement with previous low-resolution models of the super-constricted state. Next, cryoSPARC helical refinement was used to determine the cryo-EM maps shown in Table 1. Local refinement of 25% z was done to obtain high-resolution maps for model building.

Model building and refinement

Model building was done with Coot, Rosetta, and Isolde. Refinement of models was completed with Phenix. Specifically, an initial model for dynamin (PDB 6DLV) was docked into the cryo-EM maps. Manual building and corrections to the model was done in Coot and Isolde followed by refinement in Rosetta and real-space refinement in Phenix. The GDP ligand was docked into the cryo-EM map using Coot, and was included in the overall model refinements. The cryo-EM data collection, processing, and refinement statistics are shown in Table 1. Structural analysis, including comparison of models, measurements and generation of figures was done using ChimeraX.

Radial Profile

Radial profiles of dynamin helical structures were generated using the HI3D program from the Jiang Lab at Purdue University⁶⁰.

CryoET sample preparation, imaging, and image processing

300-mesh Quantifoil R2/1 grids (Electron Microscopy Sciences Q325AR1) were UV-treated for 20 minutes and coated with fibronectin (1:40 dilution of Sigma F1141–1mg in water) for 20 minutes. HeLa cells, transfected with Dynamin1 (K44A)-GFP were seeded onto the grids and grown overnight. Some samples were sorted using a BD FACS Aria II Cell Sorter prior to seeding. These sorted cells produced the samples imaged to generate the tomograms in videos S3 and S4. The following morning, these cells were unroofed by syringe spray of 0.5% paraformaldehyde in stabilization buffer (30 mM HEPES at pH 7.4, 70 mM KCl, 5 mM MgCl₂, 3 mM EGTA). They were then rinsed briefly with stabilization buffer then blotted from the side immediately prior to adding 4 μL of 1:1 stabilization buffer and 10 nm BSA coated gold beads (Electron Microscopy Sciences #25486) to the top face of the grid. These grids were plunge frozen into liquid ethane using a Leica EM GP automatic plunge freezer with 3 seconds back blotting. One grid (for tubules 1–3 in video S3) was imaged on a Thermo Fisher TF20 with a Gatan K2 direct detection camera using SerialEM with the following settings: (bidirectional tilt series from −50 to 50 degrees, imaging every 2 degrees. Mag= 5000x, pixel size 7.22 A, target defocus of −5 μm, 16 frames over 4 seconds at each tilt angle, targeting a total 120 electrons/A² for the tilt series). The other two grids (that generated the tomograms in videos S4 and S5) were clipped (Thermo Fisher C-clip #1036171 and C-clip ring #1036173)⁴⁹, imaged on a Leica cryoCLEM fluorescence microscope, then imaged in fluorescence positive regions using a Thermo Fisher Titan Krios with a Gatan K2 camera using the following settings (bidirectional tilt series from −50 to 50 degrees, imaging every 2 degrees. Mag= 19500x, pixel size 4.52 A, target defocus of −10 μm, 8 frames over 2 seconds at each tilt angle, targeting a total 120 electrons/A² for the tilt series). Tomograms were generated with IMOD 4.11.5 and segmented with EMAN 2 semi-automated segmentation^50,63. For analysis of the tomograms, since each dynamin polymer varied along it’s length, each was divided into five segments, and measurements of the outer diameter, inner luminal diameter, and inter-rung distance determined using IMOD⁵³.

Cell Culture

HeLa cells (ATCC CCL-2) were maintained in phenol red-free DMEM growth media (DMEM, Life Technologies 31053–036; 10% fetal bovine serum (FBS), Life Technologies 26140–079: 1 mM sodium pyruvate, Sigma S8636–100ML; 2mM Glutamax, Life Technologies 35050–061) with 5% CO2 at 37° C. HeLa cells were early passaged stocks directly obtained from ATCC which have been tested and shown to be mycoplasma free. The wild type and mutants of human dynamin1-GFP used in this work have been fully sequenced and are being submitted to the Addgene repository database.

Transferrin uptake assay

This assay was performed as previously reported¹⁷ and described briefly here¹⁷. HeLa cells were plated onto six-well plates (Fisher Scientific 08–772-1B) and transfected overnight with the GFP protein in question using lipofectamine 2000 (Invitrogen 11668019). The next morning, cells were serum starved for 45 minutes in growth medium lacking FBS. The cells were then rinsed in ice cold PBS4+ (PBS containing 1 mM CaCl₂, 1 mM MgCl₂, 0.2% bovine serum albumin, 5 mM glucose) and then placed in ice cold PBS4+ containing 5 μg/mL Alexa-Fluor 647 bound transferrin (Invitrogen T23366) for 5 minutes. These cells were then placed back in their growth incubator at 37° C for 20 minutes. The transferrin solution was then removed, rinsed with ice cold PBS, then incubated on ice for 10 minutes in 1 mL of 2 mg/mL pronase (Sigma 10165921001) in PBS. Most of the cells were no longer adherent. Any that were still adherent were scraped and pipetted gently prior to adding 0.25 mL of 16 % paraformaldehyde (Electron Microscopy Sciences 15710) for a 20-minute fixation. The cells were removed from ice during this incubation. They were spun down and resuspended in 300 μL PBS for flow cytometry (BD LSR II flow cytometer running BD FACSDiva Software version 8.0.1). The final samples included 1.7 ng/mL DAPI. Single cells were gated using forward, side scattering, and DAPI signal. In each experiment, every experimental condition was performed in triplicate. The gated cells were plotted with Alexa Fluor 647 transferrin vs. GFP fluorescence. Gating for GFP +/− and Alexa Fluor 647 transferrin uptake were chosen based on untransfected and no uptake incubation controls. Two separate experiments were performed and gave similar results.

GTPase assays

Turnover number (k_cat) and Michaelis constant (K_m) were determined using a colorimetric assay as described in (Leonard et al. 2005)⁶⁴. In brief, dynamin was thawed, centrifuged at 13,000 xg at 4°C for 15 min to remove aggregate, and quantified by Bradford assay according to the manufacturer’s instructions. Protein and GTP were diluted in HCB150, preincubated separately at 37°C for 10 min, and then combined in equal volumes and incubated at 37°C for the duration of the time course. At each time point, 20 μL of the reaction mix were removed and combined with 5 μL of 0.5 M EDTA in a 96-well plate to halt the reaction. For the basal activity determination, protein was used at a final concentration of 0.5 μM and GTP was used at a range of final concentrations from 0.025–2 mM. Time points were collected from 5–60 min. For the lipid-stimulated activity determination, protein was used at a final concentration of 0.1 μM and was incubated with 100% PS liposomes for 30 minutes at room temperature before being moved to 37°C. To measure free phosphate concentrations, 150 μL of room temperature malachite green reagent (1 mM Malachite Green and 10 mM ammonium molybdate in 1N HCl) was added to each well and absorbance at 650nm was measured on a Victor 2 plate reader. Free phosphate concentrations were determined by subtracting a blank of GTP only from each time point and fitting the resultant absorbance to a standard curve of 10–100 μM P_i measured in parallel. All data presented are the result of at least 3 independent experiments. The k_cat and K_m were determined in GraphPad Prism by plotting v₀ (initial reaction velocity as determined by the slope of [P_i] vs. time for each GTP concentration) vs. GTP concentration and fitting the data to the curve defined by the equation v₀ = k_cat*[dyn]*[GTP] / (K_m + [GTP]).

Nucleotide identification with HPLC, UV spectroscopy and mass spectrometry

To determine the identity of nucleotides after incubating the sample comprising ^K44Adynamin with PS liposomes, we performed a sedimentation assay followed by analysis with HPLC, UV spectroscopy, and mass spectrometry. The samples were prepared similar to the Cryo-EM sample preparation method. Briefly, PS liposomes were added to 1 mg/mL dynamin K44A and incubated for 1 hour at room temperature¹⁷. 2 mM of GTP was added to reaction mixture, incubated for 30 minutes followed by immediate high-speed ultracentrifugation (100,000xg, 15 min at 4 °C, Beckman Coulter, TLA 100 rotor). The supernatant and pellet fractions were analyzed by a modified HPLC method originally described by Giuseppe, et al⁶⁵. The 15 ml samples were analyzed in 0.9 ml/min flow for 48 minutes using the Agilent 1100 HPLC (Agilent Technologies, USA) equipped with a reverse phase column, SUPERCOSIL LC-18-T, 150 × 4.6 mm, 3 mm (SUPELCO Analytical, USA) with two buffers, Buffer A, 10 mM KH₂PO₄ pH 7.01, 10 mM tetrabutylammonium hydroxide, 1 % methanol, and Buffer B, 100 mM KH₂PO₄ pH 5.47, 2.8 mM tetrabutylammonium hydroxide, 30 % methanol. The gradient started with 0 % of buffer B and changed % of B as follows: 8 min at 0 %, 12 min at up to 40 %, 22 min at up to 44 %, and 35 min at up to 100 %. 100% B was held for 5 min and then back to the initial setting within 1 min, and the column was re-equilibrated for 7 min at 0 %. GDP and GTP were detected at 252 nm, and all data were analyzed using Agilent software, ChemStation B. 04.03 (Agilent Technologies, USA). GDP and GTP peaks were collected from the HPLC and their masses were analyzed by Agilent 6530C QTOF-LC/MS equipped with a Poroshell 120 HILIC-Z column (2.1 × 50 mm, 2.7 mm, Agilent Technologies, USA). The analytes were eluted with two buffers, Buffer A, 20 mM ammonium formate pH3.30, and Buffer B, 20 mM ammonium formate pH3.30 in acetonitrile. Positive electrospray ionization MS spectra were obtained in the mass range of 200 ~ 1200 m/z. The drying gas temperature was 300°C with a drying gas flow of 8 L / min and a pressure of 30 psi. The capillary and fragmentor voltages were 3500 V and 120 V, respectively. Mass spectra were analyzed using Agilent software, MassHunter Qualitative Analysis version B.07 (Agilent Technologies, USA).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were determined using the GraphPad Prism software (v 8.0.2). In the figure legends, the details of the number of experiments, the center and dispersion and precision measures (e.g. mean, SD and SEM) are provided. Significant differences are defined by p<0.05.

Supplementary Material

2

Video S1, related to Figure 2: Morphing of the GTPase domain between constricted (^ΔPRDdyn^PCP) and super-constricted (^ΔPRD-K44Adyn^GDP) dynamin helical polymers on lipid tubules.

3

Video S2, related to Figure 5. Modeling of a dynamin polymer assembled on the neck of budding clathrin-coated vesicle in the cell.

4

Video S3, related to Figure 5. Tomogram of dynamin in HeLa cells showing regions 1, 2, and 3.

5

Video S4, related to Figure 5. Tomogram of dynamin in HeLa cells showing region 4.

6

Video S5, related to Figure 5. Tomogram of dynamin in HeLa cells showing regions 5, 6, and 7.

7

Video S6, related to Figure 6. Animation (morph) of dynamin organization on lipid tubules in the 1-start GTP-bound constricted state and transition to the 2-start GDP-bound super-constricted state.

Highlights.

• Dynamin is organized as a 2-start helix on membranes primed for fission.

• Dynamin-decorated lipid tubules achieve super-constriction through GTP hydrolysis.

• Cryo-EM captures PH domain variable loops 1 and 4 interacting with membranes.

• Cryo-ET shows dynamin assemblies in vivo as an assortment of helical polymers.