Structure of a Bacterial Dynamin-like Protein Lipid Tube Provides a Mechanism For Assembly and Membrane Curving

Summary Proteins of the dynamin superfamily mediate membrane fission, fusion, and restructuring events by polymerizing upon lipid bilayers and forcing regions of high curvature. In this work, we show the electron cryomicroscopy reconstruction of a bacterial dynamin-like protein (BDLP) helical filament decorating a lipid tube at ∼11 Å resolution. We fitted the BDLP crystal structure and produced a molecular model for the entire filament. The BDLP GTPase domain dimerizes and forms the tube surface, the GTPase effector domain (GED) mediates self-assembly, and the paddle region contacts the lipids and promotes curvature. Association of BDLP with GMPPNP and lipid induces radical, large-scale conformational changes affecting polymerization. Nucleotide hydrolysis seems therefore to be coupled to polymer disassembly and dissociation from lipid, rather than membrane restructuring. Observed structural similarities with rat dynamin 1 suggest that our results have broad implication for other dynamin family members.

the refined structure. For this, a total of four grid searches using 5 x 5 different combinations of helical rise and rotation were performed. Each time, the grid around the highest cross-correlation peak was subdivided into a smaller grid consisting of 5 x 5 symmetry combinations, thus refining the symmetry to a helical rise of 3.1926 Å and a helical rotation of 63.815˚ ( Supplementary Fig. S2, top). For each combination, four cycles of reconstructions were computed. The resulting structure was projected, convoluted with an average CTF of the segments, Fourier transformed and subsequently compared with the sum of the experimental tube segments, which had been compensated for the decay of amplitudes to enhance the visibility of higher resolution layer lines to 11.5 Å. Side-by-side display of these powerspectra derived from the maximum cross-correlation shows good visual agreement in the position of the layer lines ( Supplementary Fig. S2E). During the final grid refinement ( Supplementary Fig. S2H), a mask was applied to the powerspectra to include only those intensities between 40 and 9 Å for cross-correlation analysis, which made the comparison more sensitive to small changes in the positions of high-resolution layer lines.
Using these refined symmetry parameters, the processing was continued as described previously (Sachse et al., 2007) (Supplementary Fig. S2J). Final projection matching was performed by sampling around the helical axis in 0.5º intervals and out of plane tilt was taken into account by ±14º in 0.5º increments. Using these alignment parameters, each image was inserted 23 times (23 x 3.19 Å = 74.4 Å stepsize of segmentation) into the reconstruction corresponding to the helical symmetry. In addition, the existence of a local rotational two-fold axis allowed further signal averaging yielding a total of 350,000 asymmetric units that could be included in the final image reconstruction. The Fourier shell correlation between the half data sets indicates a resolution of 11.0 Å and 8.5 Å according to the 0.5 and 0.143 cutoff criterion, respectively (Supplementary Fig. S2I, J). In order to avoid overestimation of the resolution, we applied a mask that was dilated by an additional 20 Å around the structure. The map was sharpened by applying a B-factor of -400 Å 2 according to the decay of amplitudes while applying a figure-of-merit weighting scheme (Rosenthal and Henderson, 2003).
A similar strategy was applied for the labelled BDLP tube reconstruction data set. However, the quality of the reconstruction was not as high as for native BDLP due to distortion and disorder in lattice packing induced by the label. The Fourier shell correlation between the half data sets indicates a resolution of 16.9 Å and 12.7 Å according to the 0.5 and 0.143 cutoff criterion, respectively (Supplementary Fig. S2I and J).
The BDLP crystal structure was fitted as three separate parts (GTPase, neck and trunk domains, see text). A new GTPase dimer was created by superimposing the BDLP domain onto the hGBP1 dimer (Ghosh et al., 2006) (see Fig. 4F). All three parts were fitted manually into the density map.
The reconstructed density of native BDLP lipid tubes has been deposited in the EMDB data bank with accession code EMD-1589 and the fitted BDLP coordinates have been deposited in the PDB protein data bank with accession code 2W6D.

Figure S1. Gallery of cryo TEM images of BDLP.
A, End-on views of tube 'stubs'. Most tubes show the two lines at radii 5 nm and 7 nm that may correspond to a compressed lipid bilayer. Another view is shown as an example in Fig. 3C2. B, Un-coated vesicle. The bilayer is clearly resolved. C, Large vesicle in the process of being tubulated. BDLP coats both the membrane of the vesicle with a large radius as well as the narrow tubes that are the subject of this study. It seems that the membrane is sucked into the tubes by adding more protein to the tube end. D-G, Various structures formed by BDLP and membranes. Note the clearly resolved bilayer in E. All scale bars are 50 nm.

Figure S2. Single-particle based helical reconstructions.
A quantitative cross-correlation analysis of powerspectra of the experimental in-plane rotated segments with the simulated powerspectra of the computed reconstruction was carried out. A grid search of 5 x 5 fixed combinations of helical rise and helical rotations was performed and evaluated. A, Four side-by-side displays of experimental and simulated powerspectra, which do not show satisfactory visual agreement at the corner grid positions of B. Colour-coded cross correlation grid of 5 x 5 different combinations of evaluated helical symmetry parameters. C, Four real-space reconstructions show poor helically averaged densities, which correspond to the corner grid positions of B at a cross correlation below 0.1. D, F, H, Three refined 5 x 5 grid searches around maximum cross-correlation peak of the previous grid yields refined helical symmetry parameters at a helical rise of 3.1926 Å and a helical rotation of 63.815˚. E, Side-by-side display of these refined symmetry parameters shows good visual agreement in the layer line positions. Both displays have been visually enhanced by dividing by the rotational average of the experimental powerspectrum (left) to compensate for high-resolution falloff of the data. G, Threedimensional reconstruction of E reveals subunit-like features in real space. Please note that since the simulated power spectrum has been visually enhanced by dividing through the rotational average to compensate for the decay of amplitudes and because it originates from an ideal helically symmetric structure, layer lines can be seen far beyond the true resolution limit because of symmetrised noise. This is true for data on the meridian and far off the meridian. It should be noted that these data are not included in the cross-correlation measurements because a mask between 9 and 40 Å resolution was applied. I, Resolution estimate for BDLP native and labelled tube reconstructions. Fourier shell correlation between two half data sets of native BDLP indicates a resolution of 11.0 Å at the 0.5 cutoff criterion. For the labelled BDLP tube reconstruction, resolution was estimated at 16.9 Å. J, Data collection and reconstruction data summary table.