3D reconstruction

An initial model was created by back-projecting a portion of a single straight T5 tail (total of ~6 consecutive trimeric stacks) and by applying threefold symmetry using Spider ⁴⁸. This initial model was refined using a classical projection matching method as described⁴⁹. Briefly, the contrast transfer function (CTF) corrected tail images were boxed into 128 × 128 pixel windows with an overlap of ¾ of the box. The generated stacked images (1000 images) were aligned against the projections of the model and averaged into classes. The classes were then back projected to obtain a new threefold symmetrised model. After 10 cycles of refinement, we obtained a map at ~10 Å resolution in which the crystal structure of pb6 could be fitted without ambiguity. In order to further improve the resolution, a larger data set was processed as described below.

Images were first motion corrected either with Digital Micrograph (Gatan Inc., USA) for T5 tails or with MotionCorr⁵⁰ for tails incubated with FhuA and pb6 tubes. The latter were also binned two times by Fourier cropping before further processing (final pixel size of 1.64 Å per pixel). The same image analysis procedure was followed for all samples. Tubes were first boxed as overlapping segments with boxer⁵¹. The interbox distance was set to one axial rise (z = 40 Å) for all samples. The coordinates were then imported into Relion 1.4 ⁵². Ctffind4 ⁵³ was used to determine the CTF parameters. 2D classification was then performed in Relion to remove outliers. The low-resolution 3D initial model (see above) or a featureless cylinder were used as the initial reference for a first 3D refinement imposing C3 symmetry. As nearly all possible helical segments were picked by choosing an interbox distance equal to the axial rise, no further helical averaging was needed. This was confirmed later by two different refinement strategies, both enforcing helical symmetry, using IHRSR⁵⁴ and Relion 2 ⁵⁵. Both approaches gave 3D density maps nearly identical to the one obtained without helical symmetry empty and full tails were then separated by successive 3D classifications. 3D classes displaying central density were grouped together while those having an empty channel formed a second group. A last 3D refinement was then done on the 22704 empty (T5 tail incubated with FhuA) and the 24243 full (T5 tails) particles, which yielded 3D reconstructions at 5.8 and 6.2 Å, respectively (Fig. 2g, j, h, k and Supplementary Table ²) after post processing in Relion with a threshold mask. For the pb6 tubes, no 3D classification was needed as the tubes were all empty, and the final 3D reconstruction had 8.8 Å resolution and included 12,244 particles (Fig. 2i, l and Supplementary Table ²). Map visualisation, segmentation and fitting were done with Chimera ³⁰.

As the 3D reconstructions obtained for full or empty T5 tails are virtually identical at the pb6 level, the two data sets were combined by scaling the T5 tails acquired at the nominal magnification of 15,500 from a pixel size of 1.24–1.64 Å per pixel. The combined data set was processed as described above but with Relion 2, which implements helical symmetry during image processing. For consistency reasons, the data sets for full and empty T5 tails were re-processed with Relion 2. As expected, the 3D reconstructions for both samples were nearly identical to the one obtained with Relion 1.4. The final 3D reconstruction for the combined data set included 55,447 particles and reached 6 Å resolution at FSC = 0.143, but with enhanced interpretability, notably at the subunit interfaces (Supplementary Fig. ⁴). This map was used to develop a pseudo-atomic model of the pb6 tube.