After the metadata preparation, the sub-tomogram particles were made in RELION4 (Zivanov et al, 2022 ) with the binning factor of six. Local 3D refinement with a mask covering one 96-nm MTD was performed. After refinement, the features pf 24-nm repeat of outer dynein arm was observed (Fig EV6D). A smaller mask only covering the inner dynein arm and radial spoke was created for the focused classification without alignment (Fig EV6E). After classification, equivalent 96-nm MTD repeat classes were separated (Fig EV6E). One of the classes was selected for the future process. New sub-tomogram particles with binning factor three and corresponding 10.2 Å of pixel size were generated. Another local refinement was performed with a 96-nm MTD mask. Resolution was estimated using the Fourier shell correlation (FSC) at a cut-off of 0.143 in RELION4 (Zivanov et al., 2022 ). Details of acquisition parameters and particle numbers are summarized in Table EV3.