a, Mutations in the Vps29-interaction surface of Vps35 block assembly in vivo. Lysates from HeLa cells transfected with cDNAs encoding wild-type or mutant forms of HA-tagged Vps35, or lysates from untransfected cells (lanes 8 and 16, NT) were subjected to immunoprecipitation (IP) using mouse monoclonal antibody to the HA epitope. Lysates (2% of total, lanes 9−16) and immunoprecipitates (lanes 1−8) were analyzed by SDS-PAGE and immunoblotting with antibodies (Abs) to the HA tag, Vps26 and Vps29. b, Mutations in the Vps35-interaction surface of Vps29 abrogate assembly, with the exception of L101D. Lysates from HeLa cells transfected with cDNAs encoding wild-type or mutant Vps29-myc, or lysates from untransfected cells (lanes 10 and 20, NT) were subjected to IP using a mouse monoclonal antibody to the myc epitope. Lysates (5% of total, lanes 11−20) and immunoprecipitates (lanes 1−10) were analyzed by SDS-PAGE and immunoblotting with antibodies to Vps35, Vps26 and the myc tag. c, The cargo recognition complex is an equimolar complex of Vps26, Vps29, and Vps35. Gels were stained with Coomassie blue (Simply Blue™ SafeStain; Invitrogen). Image acquisition and analysis was done in an Epichemi Darkroom (UVP BioImaging Systems) using Labworks 4.5 software. The integrated area from each peak was normalized to the calculated molecular mass for each protein, and the value was determined relative to Vps29. d, Coomassie-stained gel showing that the N-terminus of Vps35 forms a stable subcomplex with the C-terminal lobe of Vps26. For each lane, GST-tagged Vps35 fragments and untagged Vps26 or its fragments were co-expressed in E. coli and purified by glutathione-Sepharose chromatography.

a, The Vps26:Vps29:Vps35 complex is predicted to align roughly parallel to the membrane (green line at bottom), such that its multiple SNX 4, 15 and cargo-binding sites 25 cooperatively interact. The arrows mark the central region about which Vps35 is proposed to flex so as to interact with cargo embedded in curved membranes. Binding sites that have been mapped to individual residues within crystallized components are shown in blue. Binding sites that have been mapped to regions of Vps35 or to as yet uncrystallized portions of Vps35 are marked by red bars aligned with the region of interest. Binding sites for yeast cargo proteins are not necessarily conserved in human Vps35, however the overall architecture of the yeast and other orthologous complexes is proposed to be very similar to the human complex. b, Schematic rendering of a speculative model for the retromer coat on a tubular vesicle, colored as above, with the SNX dimer in purple.

a, Vps29 is green, and Vps35 red. b, The surface of Vps35, with the residues blocking the metallophosphoesterase site of Vps29 in grey, and other residues that contact Vps29 in slate. c, The surface of Vps29, with residues surrounding the metallophosphoesterase site in light blue, and other Vps35 contacting residues in slate. d, Hydrophobic grooves on the outer surface of Vps35 are formed between even-numbered helices. The probability of the surface to participate in ligand binding was colored from lowest to highest in a blue to red gradient using the hotpatch server http://hotpatch.mbi.ucla.edu/. Structural figures were generated with pymol, http://www.pymol.org/.

a, Alignment of human Vps35 repeats. Columns of residues that are likely to form the hydrophobic core of the structure have yellow background. Apolar residues are in red. Cylinders show the α-helical regions. Magenta boxes within the α-helices indicate the positions of apolar residues in the consensus sequences. The C-terminal region of the crystal structure is in bold. Loop regions predicted based on the multiple sequence alignment of Vps35 proteins are shown by lower case letters. b, 1−5. Averaged images of the cargo recognition complex from negative stain electron microscopy (Fig. S6). The images were obtained by multivariate statistical analysis with reference-free alignment. The number of images per class 1 to 5 is respectively 288, 362, 367, 206, 327. c, 1−5. Corresponding projections, limited to 25 Å resolution, of the cargo recognition complex model shown in d (C-terminal crystallized region of Vps35, red; Vps29, green; Vps26, cyan; and N-terminal modeled region of Vps35, orange). Each image in c is oriented such that Vps35:Vps29 corresponds always to the top part of the image and Vps26 to the bottom part, as in d. The correlation coefficient between the EM class average and the corresponding model projection is for class 1 to 5 0.79, 0.70, 0.86, 0.75, 0.73 respectively. Scale bar in c-5 = 100 Å.