Extensive structural rearrangement of intraflagellar transport trains underpins bidirectional cargo transport

Summary Bidirectional transport in cilia is carried out by polymers of the IFTA and IFTB protein complexes, called anterograde and retrograde intraflagellar transport (IFT) trains. Anterograde trains deliver cargoes from the cell to the cilium tip, then convert into retrograde trains for cargo export. We set out to understand how the IFT complexes can perform these two directly opposing roles before and after conversion. We use cryoelectron tomography and in situ cross-linking mass spectrometry to determine the structure of retrograde IFT trains and compare it with the known structure of anterograde trains. The retrograde train is a 2-fold symmetric polymer organized around a central thread of IFTA complexes. We conclude that anterograde-to-retrograde remodeling involves global rearrangements of the IFTA/B complexes and requires complete disassembly of the anterograde train. Finally, we describe how conformational changes to cargo-binding sites facilitate unidirectional cargo transport in a bidirectional system.

A -Particles from each train were picked manually.Intra-train averages without alignments were then generated and used to manually estimate angles to bootstrap local refinements.In Relion, local refinement was first performed on the entire structure at bin4.A bin2 alignment was then performed with a tighter central mask, and 3D classification was performed with an original wider mask.This separated good from bad particles by removing particles that had no signal outside of the refinement mask in the previous step.Further local refinements were performed on the different regions of the map to generate the components of the final composite map.B -Representation of angular distribution of subtomograms contributing to the final IFTA2 map from an end-on view (as if looking down the microtubule).More particles are indicated by increasingly red bars.The majority of our subtomograms were side views of the filament, resulting in anisotropic resolution.C -Side view of C, as if looking along the microtubule.IFT57  IFT38  IFT88 IFT52 IFT70 IFT57 (377-end) IFT38 (198-end) IFT140 IFT122 (718-end) IFT139 IFT121 Starting structures for model building An overview of all the input structures we used to model the retrograde IFT train.We started with the structure of the entire anterograde IFT train (center), and split it into its four constituent subcomplexes (IFTA1, IFTA2, IFTB1, IFTB2).These models were supplemented with Alphafold2 multimer predictions ("AF prediction"); pLDDT and PAE plots for each prediction are provided in subsequent figures.B -Structures of IFT139 (pink) in complex with IFT121 (yellow) in IFTA2.Left, single particle structure of Human IFTA (PDB 8FGW), focussed on the IFTA2 region.(IFT139, pink, IFT121, yellow, remaining subunits, grey).Center left, Alphafold2 prediction of C. reinhardtii IFT139/IFT121  , showing high structural similarity to the human single particle structure. This wasused to model IFT139, so that we did not need to cut the protein into two rigid bodies manually.Center right, prediction coloured by pLDDT (red is higher confidence).Right, pairwise pAE plot (green is higher confidence).C -Structures of the bridge interaction between IFT144 (red) and IFT121 (yellow).Left, single particle structure of Human IFTA (PDB 8FGW25), focussed on the bridge region.Center left, Alphafold2 prediction of C. reinhardtii IFT144 1064-1367 /IFT121  , showing high structural similarity to the single particle structure.This was used to model the bridge interaction.Center right, prediction coloured by pLDDT (red is higher confidence).Right, pairwise pAE plot (green is higher confidence).B -Alphafold2 prediction of IFT172 1-600 /IFT57 1-250 .These regions were selected based on the previously identified interaction between IFT172-WD and IFT57-CH.This prediction was used to model the WD-domain of IFT172 and its interaction with IFT57-CH domain.Top left, prediction coloured by subunit.Top right, prediction coloured by pLDDT (red is higher confidence).Bottom, pairwise pAE plot (green is higher confidence).C -Novel Alphafold2 prediction of IFT172 1-600 /IFT57 1-200 /IFT140 1050-1374 /IFT172  .This was used to build and validate the IFTB2 polymeric interface.Top left, prediction coloured by subunit.Bottom left, prediction coloured by pLDDT (red is higher confidence).Center, pairwise pAE plot (green is higher confidence).Top right, the prediction docked into the retrograde density for the IFTB2 polymeric interface. Bottom right, thesame view as above, showing two repeating units of the final retrograde model.Right, pairwise pAE plot (green is higher confidence).Star indicates flexible hinge also seen in overlapping region in B B -Alphafold2 prediction of IFT81 340-590 /IFT74340-590/IFT22.This was use to model the IFT81/74/22 complex at the IFTB2 polymeric interface.Left, Crystal structure of equivalent region from Tetrahymena.Center, Alpha-fold2 prediction coloured by subunit and by pLDDT (red is higher confidence).The crystal structure and Alpha-fold2 model are identical in this region, and the Alphafold2 model was used since it uses the Chlamydomonas sequence.Right, pairwise pAE plot (green is higher confidence).C -Alphafold2 prediction of IFT81340-390/IFT74340-390/IFT1391-472.This was used to model the most N-terminal segment of IFT81/74 identified in our density. Lef, prediction coloured by subunit. Cente, prediction coloured by pLDDT (red is higher confidence).Right, pairwise pAE plot (green is higher confidence).The only region of the IFT complex to contain tandem WD domains is IFTA1.We fit IFTA1 from our anterograde structure (left) and from single particle structures (center, 8BBF shown).They showed good overall agreement with the density, and required only minor structural modifications to fit into the density in our final retrograde model (right).
Step 3: Expand model into repeating density Step 4: Additional WD domain identified as IFT121 in IFTA2 Since IFTA1 is linked to IFTA2 through IFT122, it followed that IFTA2 would reside next to our newly placed IFTA1.Indeed, we identified an additional WD domain in the adjoining density (yellow), which contained the characteristic WD/TPR domain angles of IFT121.
We fit in IFTA2 from our anterograde structure (left) and from single particle structures (center, 8BBE shown) into our density.They showed good agreement with the density, but we identified a conformational change to fit IFT139 into the density.We used the IFT139/121 Alphafold2 prediction 0, since it best represented the straight conformation of IFT139 in the density (final refined retrograde model, right).
IFTA used for modelling A -Structures of IFTA1 used to model the retrograde train.Left, Single particle structure of Human IFTA1 (PDB 8BBF, IFT144 (red), 140 (orange) and the C-terminus of IFT122(red-orange)). Center left, Alphafold2 prediction of C. reinhardtii IFT144 363-898 /IFT140 380-941 /IFT122 718-1239 coloured by subunit, showing high structural similarity to the single particle structure.This model was used to fill in the C-terminus of IFT122, in complex with IFT144/140, since it was left partially unmodelled in our anterograde structure.Center right, prediction coloured by pLDDT (predicted local different test) score.Higher confidence residues are coloured red, lower confidence white to blue.Right, pairwise pAE (predicted aligned error) plot, indicating the confidence of the location of each residue in the structure compared to each other residue.Lower pAE score (green) is higher confidence.

IFT172
Full length AF prediction 1-1104 in anterograde model 1104-1755 AF model retrograde model IFT172-WD(1-600) / IFT57-CH(1-250) IFT172-WD/57 : IFT140 : IFT172-CTD Alphafold multimer prediction Polymeric interface modelling aided by Alphafold prediction IFTB2 used for modelling A -Alphafold2 prediction of IFT172 full-length.This was used to model the C-terminus of IFT172 that was not present in the anterograde model.Top left, prediction coloured by region present (green) or absent (white) in the anterograde model.Top right, prediction coloured by pLDDT (red is higher confidence).Bottom, pairwise pAE plot (green is higher confidence).
IFTB1 used for modellingA -Alphafold2 prediction of IFT81 460-640 /IFT74 460-640 /IFT27/IFT25/IFT52 370-454 /IFT46 . These rgions were selected based on previously identified interactions.This was modelled into the density at the end of IFTB1inner, corresponding to the biochemically identified interaction first established by Taschner et al.Left, prediction coloured by subunit.Center, prediction coloured by pLDDT (red is higher confidence).
tandem WD domains at a C2 symmetry axis (yellow region).The presence of this symmetry pre-determined the stoichiometry of the model in this region.
model into the adjacent repeats, thus narrowing the possible regions that other domains could fill.
the high-resolution experimental single structure of human IFTA in which the bridge is present (PDB 8FGW) fits into our density (left), and corresponds well with our refined IFTA model (right) Fitting the anterograde model of IFTB2 into the remaining density provided strong matches (green region, and next step).Density covered by the refined IFTA model colored yellow.zWe again expanded the built region of the map into the adjacent repeats.density and IFTB2 models.First row, orthogonal views of the anterograde IFTB2 structure (PDB 8BD7) docked into the IFTB2 outer position.The model fit into this region strongly as a rigid body.Second row, orthogonal views of the refined IFTB2 outer model, showing the introduction of the IFT172 WD domains into the continuous density.Third row, docking of anterograde IFTB2 into the IFTB2 inner position shows a strong match with the remaining density, with a mismatch in the TPR domains and in the N-terminal WD domain (arrow into density).Bottom row, the refined retrograde model, showing the introduction of the TPR domain into the continuous density, and the movement of the IFT172-WD/IFT57-CH domain into the density .