Retromer oligomerization drives SNX-BAR coat assembly and membrane constriction

Abstract

Proteins exit from endosomes through tubular carriers coated by retromer, a complex that impacts cellular signaling, lysosomal biogenesis and numerous diseases. The coat must overcome membrane tension to form tubules. We explored the dynamics and driving force of this process by reconstituting coat formation with yeast retromer and the BAR-domain sorting nexins Vps5 and Vps17 on oriented synthetic lipid tubules. This coat oligomerizes bidirectionally, forming a static tubular structure that does not exchange subunits. High concentrations of sorting nexins alone constrict membrane tubes to an invariant radius of 19 nm. At lower concentrations, oligomers of retromer must bind and interconnect the sorting nexins to drive constriction. Constricting less curved membranes into tubes, which requires more energy, coincides with an increased surface density of retromer on the sorting nexin layer. Retromer-mediated crosslinking of sorting nexins at variable densities may thus tune the energy that the coat can generate to deform the membrane. In line with this, genetic ablation of retromer oligomerization impairs endosomal protein exit in yeast and human cells.

Keywords: endosomes; lysosomes; membrane traffic; retromer; yeast.

Figure 1. Assembly of retromer coats on supported membrane tubes

1. Dynamics of scaffold formation. 25 nM SNX‐BARs and 25 nM retromer^mClover in PBS were added to SMTs and imaged by confocal microscopy at a frame rate of 1 Hz for 5 min. Scale bar: 2 μm.

2. Kymograph of the tubule shown in (A). See Movie EV1.

3. SNX^GFP colocalizes with retromer^mRuby on SMTs. SMTs containing 1 mol % of the fluorescent lipid Cy5.5‐PE were incubated with 25 nM of SNX^GFP and retromer^mRuby for 2 min. Then, the tubes were imaged by confocal microscopy. Scale bar: 2 μm.

4. Line scan analysis along the boxed tubule from (C).

5. Scaffold formation at low SNX‐BARs concentration is facilitated by retromer. SMTs were incubated as in (A), using 10 nM SNX‐BAR^GFP complex in combination with either 50 nM retromer or only control buffer. Scale bar: 2 μm.

6. Scaffold formation by elevated concentrations of SNX‐BARs alone. 100 nM SNX‐BARs‐GFP was added to SMTs and imaged by confocal microscopy at a rate of 0.5 Hz for 5 min. Scale bar: 2 μm.

7. Kymograph of the tubule highlighted in (F).

8. Line scan analysis of the tubule highlighted in (F). This experiment is also shown in Movie EV2.

Figure 2. Constriction of membrane tubes by SNX‐BARs and SNX‐BARs/retromer

1. SMTs labeled with Texas‐Red DHPE were incubated with non‐tagged proteins at 25°C for 3–5 min and analyzed by fluorescence microscopy. Proteins were used at the following concentrations: 100 nM SNX‐BARs; 25 nM SNX‐BARs/25 nM retromer; 50 nM dynamin. Scale bar: 2 μm.

2. Line scan analysis along the tubules from (A). The lower boundaries of fluorescence are indicated by horizontal lines in the respective colors.

3. Distribution of Texas‐Red DHPE fluorescence in constricted domains for tubules coated by SNX‐BARs (n = 16), SNX‐BARs plus retromer (n = 18), or dynamin (n = 15). Error bars represent the standard deviation from the mean. P‐values were calculated by Welch's t‐test. ****P < 0.0001.

4. Constricted domain radius as a function of starting (non‐constricted) tube radius. Radii of constricted and non‐constricted regions of a variety of lipid tubes were determined using the known diameter of a dynamin‐coated tube as a reference. Experiments were performed as in (A), using 25 nM SNXs (n = 16) or 25 nM SNX‐BARs plus 25 nM retromer (n = 18). Error bars represent the standard error from the mean.

5. Binding of retromer^mClover to constricted SNX‐BARs domains. SMTs were first incubated with 100 nM SNX‐BARs for 2 min at 25°C until constriction zones were visible through reduced lipid fluorescence. Then, 50 nM retromer^mClover was added under continuous acquisition at 0.5 Hz. Scale bar: 2 μm.

6. Kymograph of a tubule from (E).

7. Quantification of Texas‐Red‐DHPE fluorescence under the constriction zones before and after retromer^mClover addition (n = 16 tubes). Error bars represent the standard error from the mean. P‐values were calculated by Welch's t‐test. n.s.: not significant (P = 0.235).

Figure EV1. Procedure for quantifying constrictions along tubules

A series of line scans were performed along the entire length of an SMT, perpendicular to the tube. For each line scan, a Gaussian curve was fitted, and the maximum height of the coat signal (e.g., GFP) and the lipid signal (e.g., Texas Red DHPE) were extracted. The maximum height was then plotted against the intensity of the lipid signal along the same scan. This was performed for every line scan along the tube, yielding two populations. The number of dots in each population is a measure for the length of constricted and non‐constricted regions. For each tube, the zones corresponding to the constricted and non‐constricted areas were determined manually and their mean fluorescence value was used to calculate the tube diameter by comparison with the signals obtained with dynamin‐coated tubes. The mean fluorescence intensity of the tube under the protein scaffold was equated to the tube radius by I = K × R where I is the tube fluorescence under the scaffold and R is the tube radius. Using Dynamin (R = 11.2 nm) we can then calculate the calibration constant K.

Figure 3. Bidirectional elongation of the coat

Supported membrane tubes labeled with Cy5.5‐PE were first incubated with 50 nM SNX‐BARs and 50 nM retromer^mRuby at 25°C until coat formation was initiated (~ 90 s). Then, non‐bound SNX‐BARs and retromer^mRuby were washed out and the tubes were subjected to a second incubation with 50 nM SNX‐BARs and 50 nM retromer^mClover (3 min). Elongation (mClover, green) of the pre‐existing coat (mRuby, red) can thus be observed.

1. Scheme of the experiment.

2. Representative tube of the experiment performed as described in (A). Tubes were imaged by confocal microscopy at a framerate of 0.5 Hz. Scale bar: 2 μm.

3. Kymograph of the tubule shown in (B). Arrowheads show the origin of coat polymerization.

Figure 4. Stabilization of pulled membrane tubes by retromer/SNX‐BARs

1. Coat formation on a pulled membrane tubule. Confocal pictures of a GUV labeled with Rhodamine‐DHPE (red). A membrane tubule has been pulled from the GUV through a small bead and optical tweezers. The GUV is shown before and at several time points after ejection of SNX‐BARs/retromer^mClover (green) from a pipette in the vicinity of the GUV. The images shown are representative of a total of five tubes analyzed, which all showed similar behavior.

2. Measurement of the force exerted on the bead as a function of time after protein ejection, taken from the experiment shown in (A).

3. Repetitive pulling and stabilization. Confocal pictures of a GUV labeled with Rhodamine‐DHPE (red). A tubule has been pulled as in (A) and SNX‐BARs/retromer^mClover (green) was added. The GUV is shown before and after protein ejection, and at several stages of subsequent re‐pulling and stabilization through additional coat recruitment. Protein quickly populates new tube regions generated by pulling back the GUV. The images are representative of a total of five tubes analyzed.

4. Measurement of the force exerted on the bead as a function of time for the experiment shown in (C). Arrowheads mark the timepoints when the GUV has been pulled back.

5. Kymograph of the portion of the tubule boxed in (D), showing growth of retromer coat into a newly pulled portion of the tubule.

Figure EV2. Stabilization of pulled membrane tubes by SNX‐BARs

1. Repetitive pulling and stabilization. Confocal pictures of a GUV labeled with Rhodamine‐DHPE (red). A tubule has been pulled as in (A) and SNX‐BAR‐GFP (green) was added. The GUV is shown at several stages of subsequent re‐pulling and stabilization through additional coat recruitment. Protein quickly populates new tube regions generated by pulling back the GUV. The images show a representative experiment. Equivalent behavior was observed with another, independently prepared tube.

2. Measurement of the force exerted on the bead as a function of time for the experiment shown in (A). Arrowheads mark the timepoints when the GUV has been pulled back.

Figure 5. Variable saturation of the SNX‐BARs layer with retromer

1. Recruitment of additional subunits to pre‐formed coats using differentially labeled retromer and SNX‐BARs. Scheme of the experiment shown in (B) and (C).

2. Coats were formed on SMTs using 25 nM SNX‐BARs and 25 nm retromer^mRuby. Excess protein was washed out with buffer, and 50 nM SNX^GFP or 50 nM retromer^mClover was added. SMTs were imaged after SNX‐BARs/retromer^mRuby coat formation and 2 min after addition of retromer^mClover or SNX^GFP. Scale bar: 2 μm. A magnification showing constriction zones for each panel is shown.

3. Ratio of retromer^mRuby signals in the constricted areas before and after addition of retromer^mClover. Quantification of the experiment in (B). 36 coats from 10 different tubes were analyzed. Means and SEM are shown.

4. Occupation of SNX‐BARs domains with retromer as a function of the starting radius of the tube (naked tube radius). Arrays of SMTs were incubated with 25 nM SNX‐BARs and 25 nM retromer^mClover. The density of retromer^mClover in constricted domains was traced through its fluorescence signal. The starting radius of the tube was estimated through Texas Red‐DHPE fluorescence in non‐constricted regions and calibration with dynamin. This radius is indicated for each tube. Scale bar: 2 μm.

5. The density of retromer^mClover in SNX‐BARs/retromer^mClover coats from (D) was plotted as a function of the radius of the non‐constricted tube. 36 tubes were analyzed. Error bars represent the standard error of the mean.

6. Experiment as in (D) using 25 nM SNX^GFP and 25 nM retromer.

7. The fluorescence signals of SNX^GFP in SNX‐BARs/retromer coats from (F) were quantified plotted as a function of the radius of the non‐constricted tube as in (E). 26 tubes were analyzed. Error bars represent the standard error of the mean.

Figure 6. Substitutions destabilizing the Vps35‐Vps35 interface

1. Structure of the pentameric retromer complex (Kovtun et al, ; Leneva et al, 2021). The boxes highlight the Vps35 dimerization interface. Residues substituted in vps35 ^PISA and vps35 ^AAA3KE are shown red and green, respectively, in the structure from (C). thermophilum and in a model of the Saccharomyces cerevisiae complex derived from the Chaetomium thermophilum structure (PDB 7BLR) using the online modeling tool Swiss‐model (https://swissmodel.expasy.org).

2. Sequence alignment of the Vps35 dimerization domains from different species. Amino acids substituted in vps35 ^PISA and vps35 ^AAA3KE are shown in red and green, respectively. One residue (in red‐green) is shared between the two.

3. Coomassie‐stained SDS–PAGE gel of purified retromer^mClover complexes containing the indicated Vps35 variants.

4. Blue native PAGE gel showing the formation of higher‐order assemblies for retromer^mClover complexes containing Vps35 variants and their tentative assignment as monomers, dimers and tetramers. The samples loaded stem from the same protein preparation as in (C). The bands for monomer, dimers and tetramer (red boxes) are linked to retromer because they show altered migration upon addition of the mClover tag (monomer and dimer), or are destabilized by the PISA and AAA3KE substitutions (tetramer). See also Appendix Fig S3.

Figure 7. Effect of Vps35 dimerization on coat constriction

1. Experimental setup: supported membrane tubes (SMTs) labeled with Texas Red DHPE were preincubated with 25 nM SNX‐BARs for 3 min to load them with SNX‐BARs but not allow formation of constrictions. After a wash with protein‐free buffer, 50 nM of retromer^mClover carrying the indicated Vps35 variants was added. The tubes were imaged by confocal microscopy at a framerate of 0.5 Hz.

2. Images of three timepoints after addition of retromer^mClover variants. Scale bar: 2 μm.

3. Kymographs of the entire reactions. Experiments are shown in Movies [Link], [Link].

4. Quantification of retromer^mClover recruitment over time. SMTs were incubated as in (A). mClover fluorescence appearing along the entire length of the tubes during the second incubation phase was quantified over time. n = 10 tubes per variant. Curves represent the mean and shaded areas around the curves represent the SEM.

Figure EV3. Lack of constricted domain formation after co‐incubating SNX‐BARs with retromer containing vps35 ^PISA or vps35 ^AAA3KE

1. A, B

   SMT assay with the dimerization mutants. 50 nM SNX‐BARs and 50 nM of retromer^mClover carrying the indicated Vps35 variants were incubated with SMTs and imaged by confocal microscopy. Note that for retromer^wt already 25 nM reliably induces constrictions. The higher concentration chosen for the substituted retromer complexes is safely above that critical threshold and provides rapid formation of constrictions. Scale bars: 2 μm.

2. C

   Binding of retromer variants to pre‐formed SNX‐BARs domains. SMTs were incubated with high SNX‐BARs concentration (100 nM) for 2 min to pre‐form constrictions zones. Unbound SNX‐BARs were washed away and 50 nM retromer^mClover containing the indicated Vps35 variants was added. Panels show tubes before SNX‐BARs addition (0 s), after the formation of SNX‐BARs only coats (192 s) and after incubation with retromer^mClover variants (370 s). Tubes were imaged by spinning disc confocal microscopy at 0.5 Hz throughout the experiment. Scale bars: 2 μm.

3. D

   Kymographs of the tubes shown in (C).

Figure 8. In vivo effect of Vps35 dimerization mutants on Vps10 in yeast

1. Vps10^yEGFP localization. Yeast cells carrying Vps10^yEGFP and expressing the indicated vps35 alleles as the sole source of Vps35 were logarithmically grown in SC medium. Their vacuoles were labeled with FM4‐64. Cells were harvested by brief centrifugation and immediately imaged by confocal microscopy. Single confocal planes are shown. A brightfield image was used to outline the cell boundaries (shown in the merged images). Scale bar: 5 μm.

2. Co‐localization of Vps10^yEGFP and FM4‐64 in cells from (A) was measured using Pearson's coefficient. 15 Confocal planes with 20–30 cells each from three independent experiments were analyzed. Means and standard error of the mean are shown. P‐values were calculated by Welch's t‐test. ****P < 0.0001.

3. CLEM analysis of Vps10^yEGFP localization in vps35^PISA mutant cells. Logarithmically growing cells were high‐pressure frozen and processed by freeze substitution and embedding. Scale bar: 5 μm.

Figure 9. Effects of Vps35 dimerization mutants in human kidney (HK2) cells

1. GLUT1 at the plasma membrane. HK2 cells were treated with siRNA targeting VPS35 or with mock siRNA. Cells were fixed and stained with antibody to GLUT1 (red) and with DAPI (blue). Cells were not detergent permeabilized to preferentially show GLUT1 at the cell surface. Maximum projections of image stacks (step size in z of 300 nm) are shown. Scale bars: 10 μm.

2. Influence of VPS35 variants on GLUT1. HK2 cells silenced for VPS35 were transfected with a plasmid carrying siRNA‐resistant wildtype or mutant forms of GFP‐VPS35. GLUT1 was detected by fixation and immunofluorescence staining as in (A). Maximum projections of image stacks (step size in z of 300 nm) are shown. Scale bars: 10 μm.

3. Quantification of GLUT1 immunofluorescence in cells from (B). Regions of interest (ROIs) corresponding to cells expressing the indicated VPS35 variants, and some regions outside the cells (background), were manually defined using ImageJ software. Total cell fluorescence was integrated and corrected for background fluorescence. 105 cells per condition stemming from three independent experiments were analyzed. Mean and SEM are indicated. P‐values were calculated by Welch's t‐test. The analysis was performed with 99% confidence: ***P < 0.001.

4. Expression of Vps35 variants in cells from (B) was analyzed by SDS–PAGE and Western blot against Vps35. Tubulin served as loading control.

5. Quantification of the Vps35/tubulin ratio in cells from (B). Data stem from three independent experiments and show the mean and SEM. P‐values were calculated by Welch's t‐test. NS: not significant (P > 0.01).

Figure EV4. Dominant negative effect of hVPS35 variants on GLUT1 sorting in HK2 cells

1. HK2 cells were transfected with a plasmid carrying the indicated variants of hVPS35^GFP, or with an empty plasmid, and cultivated for 24 h. They were fixed, stained with antibody to GLUT1 (red) and with DAPI (blue) and analyzed by confocal microscopy. The cells were not detergent permeabilized. Scale bar: 10 μm.

2. Quantification of GLUT1 fluorescence in the cells from (A) was performed as in Fig 9C.

3. Expression of Vps35 variants in cells from (A) was analyzed by SDS–PAGE and Western blot against Vps35. Tubulin served as loading control.

4. Quantification of the Vps35/tubulin ratio in cells from (A). Data stem from three independent experiments and show the mean. NS: not significant according to a two‐tailed t‐test (P > 0.01).

Figure EV5. Effect of VPS35 variants on the size of LAMP1 compartments

1. Effect of VPS35 knockdown. HK2 cells were treated with siRNA targeting VPS35 or with mock siRNA. The cells were fixed and immuno‐stained with anti‐LAMP1 antibody and DAPI and imaged by confocal microscopy. Representative images are shown. Scale bar: 10 μm. Arrowheads point to examples of enlarged LAMP1‐compartments. Scale bar: 10 μm.

2. Effect of VPS35 variants. Vps35 knock‐down cells were transfected with a plasmid expressing siRNA‐resistant wildtype or mutant forms of VPS35^GFP, or with an empty plasmid. Cells were fixed, immuno‐stained with anti‐LAMP1 and DAPI and imaged by confocal microscopy. Scale bar: 10 μm. Arrowheads point to examples of enlarged LAMP1‐compartments.

3. Western blot assessing the siRNA Vps35 knockdown efficiency. Tubulin was used as a loading control.

4. Quantification of the siRNA knockdown efficiency of Vps35 for tree‐independent experiments.