A Membrane Coat Complex Essential for Endosome-to-Golgi Retrograde Transport in Yeast

Unless otherwise stated, reagents were obtained from Boehringer Mannheim Corp. (Indianapolis, IN), Fisher Scientific Co. (Fairlawn, NJ), or from Sigma Chemical Co. (St. Louis, MO). All yeast strains were grown in yeast extract, peptone, dextrose medium (YPD), or yeast nitrogen base medium (YNB; Sherman et al., 1979). Yeast transformations were performed using alkali cation treatment (Elble, 1992), and Escherichia coli transformations were performed according to Hanahan (1983). Yeast strains used in this study are listed in Table I.

Samples solubilized in SDS-PAGE buffer (6 M urea, 50 mM Tris-HCl, pH 6.8, 1% SDS) were loaded and electrophoresed (constant current of 20 mA per gel) on either 8% or 10% polyacrylamide gels. After electrophoresis, the gel was briefly immersed in transfer buffer (5% methanol, 50 mM Tris, 380 mM glycine) and then placed flat against a sheet of nitrocellulose (Schleicher & Schuell, Inc., Keene, NH) and assembled into an apparatus (Bio-Rad Laboratories, Hercules, CA) for electrophoretic transfer of the proteins onto the nitrocellulose. Transfer was accomplished at constant current (150 mA) for 10–12 h. The blot was then blocked with blocking buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 0.25% gelatin, 0.1% Nonidet P40, 0.1% sodium azide) for an hour before incubation with antibodies. Most antiserum was used at a dilution of 1:1,000. Antibody incubations were generally for 2 h at room temperature on a rocking table. Washes were for 30–60 min with several changes of blocking buffer. [¹²⁵I]protein A (Amersham Corp., Arlington Heights, IL) was used to detect the first antibodies, and this was used at a dilution of 1:1,000. After further blocking, the blots was exposed to x-ray film (Eastman Kodak Co., Rochester, NY).

CPY sorting was performed as described previously in Seaman et al. (1997). Similarly, cell fractionation experiments using differential centrifugation to separate P100 membranes from P13 membranes and cytosol (S100) were performed essentially as described previously in Seaman et al. (1997). Sucrose density gradient fractionation of P100 membranes was performed as in Seaman et al. (1997) with the exception that a P100 membrane fraction was loaded onto the gradient instead of a supernatant from a 13,000 g spin. The P100 generated for this experiment was from the lysis of ∼15 ODs of cells, and was resuspended in 0.5 ml of lysis buffer and briefly dounced in a 1-ml dounce homogenizer before loading onto the sucrose density gradient.

Fractionation of P100 membranes by sephacryl S1000 column chromatography was performed as follows. Cells grown in YPD media were harvested at an OD of 2–3 (generally 6,000–7,000 ODs of cells were used). The cells were spun down using an RC5B centrifuge (Sorvall, Newtown, CT) in a GS3 rotor (10 min at 5,000 rpm). The cell pellet was washed with water and spun again. The cells were then resuspended into 500 ml of spheroplasting buffer (0.8 M sorbitol, 10 mM Tris-HCl, pH 7.5, 10 mM CaCl₂, 2 mM DTT, and 10 μg/OD zymolyase) and incubated at 30°C with gentle shaking (90 rpm). Spheroplasting was monitored by measuring the OD 600 nm of an aliquot of cells placed in 1 ml of water. When spheroplasting was 80–90% complete (∼60 min), the cells were spun out again (this time in a GSA rotor) for 10 min at 4°C and 1,500 rpm. The cell pellet was washed with wash buffer (0.8 M sorbitol, 20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA) and then pelleted again as before. The cells were then lysed in chilled lysis buffer (0.2 M sorbitol, 20 mM Hepes-KOH pH 7.0, 50 mM potassium acetate, 2 mM EDTA) plus protease inhibitors (PMSF, 100μg/ml, pepstatin 1 μg/ml, leupeptin 1 μg/ml, antipain 5 μg/ml and α2-macroglobulin 10 μg/ml) with the aid of a 15-ml dounce homogenizer (∼15 strokes). The lysate was spun at 13,000 g for 10 min in a Sorval SS34 rotor (10,500 rpm) at 4°C. The resulting supernatant was centrifuged for 1 h at 100,000 g in a 60 ti rotor (32,000 rpm; Beckman Instruments, Inc., Fullerton, CA) at 4°C. The resulting P100 pellet was resuspended in ∼1.5 ml of lysis buffer with the aid of a 1-ml dounce homogenizer, was briefly spun at 13,000 g to pellet any aggregated material, and was then loaded onto a sephacryl S1000 (Pharmacia, Piscataway, NJ) column (118 cm × 1.5 cm) that had been equilibrated with lysis buffer. The column was run at 4°C, the samples were eluted using a flow rate of ∼250 μl per min, and 2.5-ml fractions were collected. Fractions were analyzed for various proteins by Western blotting, and were also analyzed by EM. To strip the P100 membranes, lysis buffer containing 250 mM NaCl was used to resuspend the membranes after the 100,000-g spin. The membranes were dounced in a 1-ml dounce as described previously, and were then centrifuged again at 100,000 g to pellet the stripped membranes. The supernatant was then fractionated on the sephacryl S300 column as described below. Alternatively, the supernatant was dialyzed against ∼1-liter lysis buffer (0.2 M sorbitol, 20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA) overnight at 4°C. After dialysis, the supernatant was spun briefly at 13,000 g, and was then fractionated on a sephacryl S300 column.

Cells grown in YNB media were first spheroplasted and then labeled for 15 min with [³⁵S]methionine/cysteine. Excess cold methionine and cysteine were added, and the cells were chased for 45 min. The cells were then spun out and resuspended in lysis buffer (20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA, 0.2 M sorbitol) at ∼7 ODs/ml. A lysate was produced by douncing the cells in a dounce homogenizer using 15 strokes of the pestle. The lysate was then transferred to eppendorf tubes and spun at 300 g for 5 min at 4°C to remove unbroken cells and large organelles such as nuclei. In some cases the resulting cleared lysate was then treated with the cross-linker 3,3′-dithiobis sulfosuccinimidyl-proprionate (DTSSP; Pierce Chemical Co., Rockford, IL), which was added to a final concentration of 2 mM. In other experiments, the cleared lysate was first spun at 13,000 g for 5 min at 4°C to remove vacuolar membranes, mitochondria, plasma membranes, and endoplasmic reticulum. The supernatant from this spin was centrifuged further by spinning at 100,000 g for 1 h to generate a P100. This supernatant was then resuspended into 100 μl of lysis buffer with the aid of a plastic pestle, and cross-linker was added to a final concentration of 2 mM. Cross-linking was allowed to proceed for 30 min at room temperature. The reaction was stopped by adding TCA to a final concentration of 10%, and the tubes were placed on ice for 20 min to allow the proteins to precipitate. The precipitated proteins were pelleted by centrifugation at 13,000 g for 5 min, and were then washed twice with ice-cold acetone to remove the TCA. The samples were lyophilized and then resuspended into 100 μl of urea buffer (6 M urea, 50 mM Tris-HCl, pH 7.5, 1 mM EDTA, 1% SDS). Protein pellets were resuspended with the aid of a sonicating waterbath. 1 ml of immunoprecipitation buffer was added (25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% Tween-20, 0.1 mM EDTA), the samples were spun for 10 min at 13,000 g to remove insoluble debris, and then antisera was added. Immunoprecipitations were carried out as described previously (Seaman et al., 1997).

Yeast cells grown in YPD media were harvested at an OD 600 nm of ∼1.0. Approximately 200 ODs of cells were spun down in a bench-top clinical centrifuge. The pellet was washed with water and then resuspended in 50 ml of spheroplasting buffer (0.8 M sorbitol, 10 mM Tris-HCl, pH 7.5, 10 mM CaCl₂). DTT and zymolyase (Seikagaku, Tokyo, Japan) were added to final concentrations of 2 mM and 10 μg/OD, respectively. Spheroplasting was monitored by observing the drop in OD 600 nm when an aliquot of the spheroplasting cells was placed in water. When spheroplasting was 80– 90% complete, the cells were spun out as before and washed with chilled wash buffer (0.8 M sorbitol, 20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA). After recovering the cells by centrifugation, the cells were resuspended into lysis buffer at ∼100 ODs per ml. The cells were lysed with the aid of a dounce homogenizer. The lysate was spun at 13,000 g at 4°C to remove unbroken cells and larger membranes. The supernatant was then spun for 1 h at 4°C and 100,000 g to pellet the remaining membranes. 1 ml of the supernatant (yeast cytosol) was loaded onto a sephacryl S300 (Pharmacia Biotech, Inc., Piscataway, NJ) column that had been equilibrated with lysis buffer, and was then fractionated with a flow rate of 0.4 ml per min using lysis buffer to elute the proteins. 1.25-ml fractions were collected after an initial wait period of 80 min. The fractions were precipitated with 10% TCA, and the pellets were washed with acetone to remove the TCA as described before. The proteins were resuspended into 100 μl of urea sample buffer (6 M urea, 50 mM Tris-HCl, pH 6.8, 1% SDS, 5% β-mercaptoethanol, 0.1% bromophenol blue). 30 μl of the sample was loaded per lane onto an 8% SDS polyacrylamide gel. After electrophoresis, the proteins were transferred to nitrocellulose for Western blotting.

Standard molecular biology techniques and protocols for DNA manipulation were used throughout this study (Sambrook et al., 1989). To generate an antisera against Vps29p, a glutathione S-transferase (GST) fusion (Smith and Johnson, 1988) with Vps29p was created. Using an oligonucleotide primer incorporating a BamH1 site at the 5′ end and a primer corresponding to the T3 promoter, the region of VPS29 downstream of the intron (Seaman et al., 1997) was amplified by PCR from pVPS29 -1(Seaman et al., 1997). The ∼1.6-kb fragment was cloned using the TA cloning vector according to manufacturer's instructions (Invitrogen Corp., Carlsbad, CA). The fragment was excised by BamH1 digestion, and was subcloned into pGEX 3X (Smith and Johnson, 1988) that had been digested with BamH1 to generate pGST-VPS29. E. coli transformed with pGST-VPS29 was grown in liquid culture and induced to express by adding isopropyl β-d-thiogalactopyranoside (IPTG) to a final concentration of 0.1 mM. The protein produced was isolated by glutathione affinity chromatography (Smith and Johnson, 1988) and by following the manufacturer's instructions, and on SDS gels was found to migrate as a protein with an apparent size of ∼120 kD on SDS polyacrylamide gels. This was much larger than the predicted size of the fusion (∼50 kD), and may have been due to the very acidic nature of the Vps29 protein, which has a pI of ∼4.3. Purified fusion protein was used to immunize New Zealand white rabbits following a standard immunization protocol (Harlow and Lane, 1988). After four boosts, the animal was bled out, and 5 ml of antisera was affinity-purified on GST-Vps29p that had been coupled to CNBr activated sepharose (Pharmacia Biotech., Inc.) following the manufacturer's instructions.

The vps35dn allele was generated and screened for in an identical manner to the vps35ts allele (Seaman et al., 1997) with the exception that the gapped plasmid was a multicopy plasmid (pRS424) containing VPS35. VPS26 was cloned by amplifying the gene using PCR and genomic DNA as template. The primers used for the PCR reaction were designed to flank the open reading frame, annealing ∼200 bp on either side of the VPS26 gene. A single product of ∼1.7 kb was obtained from the PCR, and was cloned using the TA cloning kit to create pTA26. VPS26 was then subcloned into Not1, and Kpn1 digested pRS414 and pRS416 after excising VPS26 from pTA26 by Not1 and Kpn1 digestion. The resulting plasmids (pVPS26-414 and pVPS26-416) were transformed into SEY26-1, a yeast strain carrying a mutation in VPS26. Both plasmids were able to compliment the CPY sorting defect in SEY26-1 (data not shown). A deletion of VPS26 was generated by removing ∼80% of the coding region of VPS26 by digesting pVPS26-416 with HindIII and Xba1. A HindIII–XbaI fragment containing the LEU2 gene excised from pBlueScript (Stratagene, La Jolla, CA) containing LEU2 was then subcloned into the cut pVPS26-416 and ligated. The resulting plasmid pVPS26-LEU2 was used as a template to amplify the disrupted gene by PCR. The same primers were used as before. The PCR product was used to transform wild-type yeast (SEY6210). Transformants were selected by growth on -leu plates. Deletion of the gene was confirmed by PCR on genomic DNA prepared from leu+ transformants. Transformation of the vps26Δ strain (MSY2600) with pVPS26-414 or pVPS26-416 was able to compliment the CPY sorting defect observed in strain MSY2600 (data not shown). An epitope-tagged version of VPS26 that compliments a vps26 mutant has been reported previously (Bachhawat et al., 1994), and was a gift from E. Jones (Carnegie-Mellon University, Pittsburgh, PA). However, it was necessary to subclone a HindIII fragment from the integrating plasmid p8HA22 (Bachhawat et al., 1994) into HindIII-cut pVPS26-416 to generate an expression plasmid for the hemagglutinin (HA)-tagged VPS26. This subcloning created pVPS26-HA, which was transformed into MSY2600 and was found to compliment the CPY sorting defect fully.

A yeast strain with a myc-tagged VPS10 gene integrated into the genome was generated in the following way. The plasmid containing VPS10-myc, pEMY10-21 (Cereghino et al., 1995), was cut with BglII to excise the VPS10-myc construct. The linear DNA was gel-purified, and was used to transform the strain EMY3 (vps10Δ::HIS3; Marcusson et al., 1994) along with pCYI50 (Bankaitis et al., 1986) that encodes a CPY-Invertase fusion. Transformants were selected for growth on -ura plates. Replicas were screened for the ability to sort CPY-Invertase to the vacuole using a plate assay (Paravicini et al., 1992). Cells in which the VPS10-myc had integrated into the genome scored as white colonies capable of sorting CPY-Invertase to the vacuole. Four independent colonies were found to be VPS+. These colonies also contained the VPS10-myc gene as determined by the presence of the ∼200-kD band in immunoprecipitates from ³⁵S- labeled cells using the anti-myc antibody 9E10 (data not shown). Three of the four were also found to now be his−, indicating that the VPS10-myc gene had integrated into the vps10::HIS3 locus, displacing the HIS3 gene in the process. One of the his− VPS+ colonies was designated MSY10-21, and was used in these studies.

To generate significant amounts of Vps5p in bacteria for in vitro studies, a GST-VPS5 fusion was generated. This was achieved by subcloning an Xba1-Xho1 fragment of VPS5 into Xba1-Xho1 digested pGEX KG (Pharmacia Biotech., Inc.). The resulting plasmid was termed pGST-VPS5. Digestion at the Xba1 site in Vps5p resulted in the loss of six amino terminal residues of Vps5p and Xho1 cut downstream of the coding region. This procedure formed an in-frame fusion of VPS5 with GST, which expressed almost full-length VPS5. An identical fusion was made between VPS5 and GST in pEG KT, a GST expression system for yeast (Mitchell et al., 1993). This plasmid was designated pMSGST-5, and when expressed in vps5Δ, cells were able to compliment the CPY sorting defect fully (data not shown). Bacteria harboring pGST-VPS5 were induced to express by adding IPTG. The fusion protein was soluble, although poorly expressed. A protein of ∼120 kD was observed on SDS polyacrylamide gels, which corresponded to the full-length fusion. Some proteolysis of the GST-Vps5p occurred in the bacteria, resulting in breakdown products ranging in size from 60 to 30 kD. GST-Vps5p that had been eluted from the glutathione-sepharose was dialyzed against ∼1 liter of lysis buffer (0.2 M sorbitol, 20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA) overnight at 4°C. This buffer was loaded onto the sephacryl S300 column and fractionated in a manner identical to that for yeast cytosol.

Wild-type and vps5Δ cells harboring a plasmid to express GST-Vps5p in yeast (pMSGST-5) were grown in -ura, -leu minimal media to an OD 600 nm of ∼1 ODU/mL. 1,000 OD 600-nm equivalents of cells were harvested and converted to spheroplasts by zymolyase digestion. The spheroplasted cells were lysed at 100 OD 600-nm equivalents per ml in lysis buffer (0.2 M sorbitol, 20 mM Hepes-KOH, pH 7.0, 50 mM potassium acetate, 2 mM EDTA) containing 0.5% triton with the aid of a dounce homogenizer. The lysate was spun at 20,000 g in a sorvall SS34 rotor for 10 min at 4°C to remove insoluble material. The supernatant was collected and incubated with 150 μl of glutathione sepharose slurry for 15 min at room temperature on a rocking table. After several washes of the sepharose with 10-ml aliquots of lysis buffer, the bound proteins were eluted using 10 mM of reduced glutathione in 50 mM Tris, pH 8.0. The proteins were precipitated by adding TCA to 10%, and after acetone washes were resuspended into SDS-PAGE sample buffer.

EM on frozen yeast cells or frozen P100 membrane fractions to localize Vps5p was performed essentially as described previously (Rieder et al., 1996; McCaffery and Farquhar, 1995). vps4 cells (SEY 4-1) transformed with the plasmid pAH31 to express an HA-tagged Vps5p (Nothwehr and Hindes, 1997) were grown to midlog (∼OD 600 nm 0.6) before being spun down and fixed for 18–24 h in 3% formaldehyde dissolved in PBS, pH 7.4. The cells were then dispersed in 1% agarose, which was trimmed into mm³ blocks and then cryoprotected in 2.3 M sucrose containing 25% polyvinylpyrollidone (10,000 D) and frozen onto cryopins in liquid nitrogen. Ultrathin cryosections were cut on a Reichert Ultracut E (Leica, Deerfield, IL) equipped with an FC4 cryostage, and were collected onto 300 mesh, formvar/carbon-coated grids and immunolabeled as described previously (Reider et al., 1996). Sections were adsorption stained/embedded with 3.2% polyvinyl alcohol, 0.2% methyl cellulose (400 centiposes), and 0.1% uranyl acetate contained in H₂O. Monoclonal anti-HA antibodies (Berkeley Antibody Co., Inc., Richmond, CA) were used to label the HA-tagged Vps5p, and these were visualized using anti-mouse coupled to 5 nm colloidal-gold.

For size-fractionated Vps5p complexes, the complexes were adsorbed to 300 mesh, formvar/carbon-coated grids (freshly ionized) fixed for 10 min in 3% formaldehyde dissolved in PBS, pH 7.4, and were subsequently immunolabeled (or not) as described previously (McCaffery and Farquhar, 1995). The complexes were then negatively stained in 2% uranyl acetate and observed on a JEOL 1200 EX II TEM in paraformaldehyde as in Rieder et al. (1996).

Previous studies have identified several genes whose products are required for the correct trafficking of the vacuolar hydrolase receptor, Vps10p. Strains carrying mutations in the VPS35, VPS29, or VPS30 genes have phenotypes that are virtually indistinguishable, and cause Vps10p to become mislocalized to the vacuolar membrane (Seaman et al., 1997). Detailed analysis of the vps35 mutant strongly suggested that a defect in the retrieval of Vps10p from the prevacuolar endosome to the late-Golgi was responsible for the vacuolar protein-sorting defect (Seaman et al., 1997). The very similar phenotypes of the vps35, vps29, and vps30 mutants suggested that the gene products may all function at a common step in the retrieval pathway. To test this directly and to gain a better understanding of the roles that the VPS35, VPS29, and VPS30 genes play in the retrieval of Vps10p, a dominant negative mutant of VPS35 (vps35dn) was generated. When introduced at high copy number into wild-type cells, this allele caused a significant defect in the trafficking of CPY to the vacuole, resulting in 50% of the CPY being secreted from the cell as the Golgi-modified p2 precursor form (Fig. 1,A, lanes 1 and 2). Effects of dominant negative alleles are often due to titration of other factors by the dominant negative protein. Candidates for these other factors in this case would clearly include Vps29p and/or Vps30p. Therefore, the strain carrying the vps35dn allele was transformed with high-copy plasmids to overexpress either the VPS29, VPS30, or VPS35 gene products. Only cells overexpressing VPS29 were able to suppress the CPY sorting defect caused by the vps35dn allele (Fig. 1 A, lanes 3 and 4). These data suggest an interaction between Vps35p and Vps29p.

To address whether this genetic interaction corresponded to a physical interaction between Vps35p and Vps29p, we first investigated the subcellular localization of Vps29p. If Vps35p and Vps29p physically interact, then it is logical to expect the proteins to colocalize. To investigate the subcellular localization of Vps29p, a polyclonal antiserum was raised against a recombinant GST-Vps29p fusion protein (Materials and Methods). This antiserum recognized a protein with an apparent molecular mass of 45 that was absent in immunoprecipitations from vps29Δ cells, and was greatly elevated in cells containing a multicopy VPS29 plasmid (Fig. 1,B and data not shown). The apparent molecular mass of Vps29p is significantly larger than its predicted molecular mass of ∼30. This discrepancy may be due to some form of posttranslational modification, or could be caused by the acidic nature of Vps29p, which has a pI of ∼4.3. Some cross-reactivity of the antiserum with a protein(s) slightly larger than Vps29p was observed, and is seen in Fig. 1,B, panel 2 when a lysate from vps29Δ cells was treated with the Vps29p antiserum. When the subcellular distribution of Vps29p was investigated (Fig. 1,B, panel 1) it was found to be a predominantly soluble protein, but ∼20% of Vps29p was associated with the 100,000 g particulate fraction (P100) that contains Vps35p (Paravicini et al., 1992) and the Golgi membrane proteins Kex2p and Vps10p (Marcusson et al., 1994). No Vps29p was detected in the P13 fraction that contains vacuolar membranes, ER, mitochondria, and plasma membranes (Marcusson et al., 1994). Deletion of VPS35 resulted in Vps29p becoming completely soluble (Fig. 1,B, panel 4), demonstrating that Vps35p is required for membrane association of Vps29p. In contrast, there was no change in the localization of Vps29p in cells in which VPS30 was deleted (Fig. 1 B, panel 3).

To investigate further the potential association of Vps35p with Vps29p, and to identify any other proteins that may interact with the Vps35p/Vps29p complex, a series of cross-linking experiments was undertaken. Lysates from wild-type cells that had been cleared of unbroken cells were treated with the homobifunctional reducible cross-linker DTSSP. The reaction was stopped by adding TCA to precipitate the proteins, and then the sample was denatured in a buffer containing SDS and urea to break noncovalent interactions. Vps35p or Vps29p was immunoprecipitated from resulting lysates under nonreducing conditions. As seen in Fig. 2 A, when no cross-linker is added, the antisera against Vps35p only immunoprecipitated Vps35p (lane 1). Similarly, the Vps29p antisera predominately recognized Vps29p (lane 2). When cross-linker was added and either Vps35p (lane 3) or Vps29p (lane 4) was immunoprecipitated, a number of other proteins were detected. Vps29p cross-linked to a protein with identical mobility to Vps35p, and both Vps29p and Vps35p cross-linked to three other proteins—p95, p70, and p50—based upon their respective migrations on the SDS polyacrylamide gel. A faint band corresponding to Vps29p was also present in the Vps35p immunoprecipitates. The relatively low Vps29p signal in the Vps35p immunoprecipitates is due to two factors: first, the cross-linking reaction is not completely efficient, which means that not all Vps29p associated with Vps35p will be cross-linked; and second, Vps29p is a small protein with relatively few methionines and cysteines, and hence will not give a strong signal compared with larger proteins, such as Vps35p.

The interaction between Vps35p and Vps29p indeed was confirmed in an experiment in which an enriched source of Vps35p was used as a source of material for cross-linking. 80% of the total Vps35p is present in an endosome/Golgi/vesicle-enriched fraction that sediments at 100,000 g (P100). Therefore, P100 membranes isolated by centrifugation from [³⁵S]methionine-labeled wild-type cells were resuspended in lysis buffer and then treated with cross-linker. Vps35p or Vps29p was immunoprecipitated from the lysate using specific antisera. In the presence of cross-linker (Fig. 2 B, lane 1), Vps35p coimmunoprecipitates with Vps29p and the three other proteins: p95, p70, and p50. The identity of Vps29p was confirmed by treating cross-linked immunoprecipitated Vps35p with reducing agent to break the cross-linker. Vps29p is then immunoprecipitated from the resulting lysate (lane 2). In lane 3, antibodies against Vps29p immunoprecipitated both Vps29p and Vps35p, along with the three other proteins that coimmunoprecipitate with Vps35p (compare lanes 1 and 3). The association of Vps29p with Vps35p was cross-linker–dependent; when no cross-linker was added to the P100 membranes, Vps29p did not coimmunoprecipitate with Vps35p (lane 4). As a control for efficient cross-linking, Vps5p was immunoprecipitated from the extracts used for lane 1 (lane 5). Vps5p has been shown previously to be stably associated with Vps17p (Horazdovsky et al., 1997), and hence this interaction was deemed to be a suitable positive control for cross-linking. It was noticed that Vps5p and Vps17p exhibited mobilities very similar if not identical to two of the proteins (p95 and p70) that coimmunoprecipitated with Vps35p and Vps29p. Therefore, to test directly the possibility that Vps5p and Vps17p are interacting with Vps35p/Vps29p, P100 membranes were cross-linked as before, and Vps35p was immunoprecipitated from the resulting lysates. After treating the immunoprecipitates with reducing agent to break the cross-linker, antibodies against Vps30p, Vps5p, or Vps17p were added. It was not possible to detect Vps30p after reducing the Vps35p immunoprecipitate (lane 6); however, antiserum against Vps5p did immunoprecipitate a protein with the mobility of Vps5p from the reduced lysate (lane 7), and Vps17p could also be detected, although to a lesser extent (data not shown).

Previously published data have demonstrated that the P100 membrane–associated Vps35p can be floated out of 60% sucrose using a simple two-step sucrose gradient (Paravicini et al., 1992). Vps5p has been found to behave in a similar fashion (Horazdovsky et al., 1997) and therefore the cross-linked complex described here is unlikely to represent a very large sedimentable particle, but rather a peripheral membrane-associated complex.

As an independent test of the interactions discovered through the use of cross-linkers, a fusion protein between GST and Vps5p was expressed in yeast, and was used to confirm the presence of the multimeric complex containing Vps35p, Vps29p, Vps17p, and Vps5p. The GST-Vps5p was functional, as expression in a vps5Δ strain was able to complement the CPY sorting defect fully (data not shown). GST-Vps5p isolated from either wild-type (data not shown) or vps5Δ (Fig. 2,C) yeast lysates using glutathione-sepharose was found to be bound to Vps17p, Vps35p, and Vps29p (Fig. 2,C, lane 2), but none of these proteins was detected when a control lysate from a strain expressing just GST was treated with glutathione-sepharose (Fig. 2 C, lane 1).

To identify the p50 component of the multimeric complex, a database search of the cloned VPS and PEP genes was performed to identify candidate 50-kD proteins. The best candidate was VPS26/PEP8. This gene encodes an ∼40-kD protein that migrates on SDS gels with a molecular mass of ∼50 (Bachhawat et al., 1994). The phenotype of vps26 mutants is intermediate between that of the class A mutants (vps35, vps29, and vps30) and the class B mutants (vps5 and vps17) (Raymond et al., 1992). Therefore, a strain in which VPS26 was deleted was generated as described in Materials and Methods, and was used in a cross-linking experiment to determine if Vps26p is p50. In Fig. 3 A, lysates were treated with cross-linker as before, and Vps35p was immunoprecipitated. When no cross-linker was added, only Vps35p was present (lane 1), while in lane 2, cross-linker was added to a lysate, and the pattern of bands contained Vps5p, Vps17p, p50, and Vps29p. However, when an identical experiment was performed using lysate from cells in which VPS26 was deleted, the p50 band was missing (compare lanes 2 and 3). It is interesting to note that Vps35p continues to coimmunoprecipitate with Vps5p, Vps17p, and Vps29p, suggesting that Vps26p is not essential for formation of the complex. More recently, cross-linking experiments using a strain expressing an epitope-tagged Vps26p have confirmed that Vps26p does indeed interact with Vps35p, Vps29p, Vps5p, and Vps17p in a large complex (data not shown).

To address the question of which components of the complex are essential for forming the Vps35p/Vps29p/Vps26p/ Vps5p/Vps17p complex, lysates from various [³⁵S]methionine-labeled yeast strains were treated with cross-linker, and either Vps35p or Vps29p was immunoprecipitated. Because the localization of the proteins is affected in the various strains tested (e.g., Vps35p is localized to the P13, vacuole-enriched fraction in a vps29Δ strain), a crude lysate cleared of unbroken cells was used for these cross-linking experiments. When lysates from wild-type cells treated with cross-linker were immunoprecipitated with antibodies against Vps35p and Vps29p (Fig. 3,B, lanes 1 and 2, respectively), the same pattern of bands was observed as seen in Fig. 2,A. When a lysate of the vps35Δ strain was cross-linked and Vps29p was immunoprecipitated, Vps29p was not cross-linked to Vps5p/Vps17p (Fig. 3,B, lane 3). Similarly, when the lysate from a vps29Δ strain was cross-linked and Vps35p was immunoprecipitated, Vps35p no longer cross-linked to Vps5p/Vps17p or Vps26p (Fig. 3,B, lane 6). In contrast, however, deletion of VPS30 (Fig. 3,B, lanes 4 and 5) had no effect upon the ability of the complex to assemble. Deletion of either VPS17 (Fig. 3,B, lanes 7 and 8) or VPS5 (Fig. 3,B, lane 9) results in the loss of the interaction between Vps35p and Vps5p/Vps17p (Fig. 3,B, lanes 7 and 9). Likewise, Vps29p no longer interacts with Vps5p/ Vps17p (Fig. 3,B, lane 8). Interestingly, it appears that the interaction between Vps35p and Vps29p is unaffected in either the vps17 or vps5 mutants (Fig. 3 B, lanes 7–9), and Vps35p/Vps29p can still interact with Vps26p. These results hint at the possibility that two distinct subcomplexes assemble together to form a multimeric complex. These two subcomplexes appear to comprise a Vps35p/Vps29p/ Vps26p complex and a Vps5p/Vps17p complex.

VPS26/PEP8 has been shown previously to be important for sorting CPY to the vacuole (Bachhawat et al., 1994). The data presented here suggested that Vps26p is part of a multimeric membrane–associated complex that also contains Vps35p, Vps29p, Vps5p, and Vps17p. If Vps26p is indeed interacting with Vps35p/Vps29p and/or Vps5p/Vps17p, then it is reasonable to expect a vps26 mutant to have a similar phenotype(s) to either a vps35 or vps5 mutant. Therefore, cell fractionation experiments were performed in which localization of Vps10p was investigated in both wild-type and vps26Δ strains. It has been observed that Vps10p is mislocalized in both vps5 and vps35 mutants (Horazdovsky et al., 1997; Seaman et al., 1997), although to a lesser extent in vps5 mutants. In Fig. 4 the distribution of Vps10p was examined using a differential centrifugation method that separates the smaller Golgi, endosomal, and vesicular membranes away from larger vacuolar (and also ER, mitochondrial, and plasma) membranes (Marcusson, et al., 1994). Wild-type cells exhibited a normal distribution of Vps10p with ∼90% of the Vps10p residing in a P100 membrane fraction that contains Golgi, endosomal, and vesicular membranes. In vps26Δ cells ∼90% of Vps10p is found in the vacuole-enriched P13. This means that vps26Δ cells, with respect to the localization of Vps10p, have a phenotype very similar to the vps29 or vps35 mutants (Seaman et al., 1997). Interestingly, a significant proportion of Vps35p becomes soluble in vps26Δ cells (data not shown), which makes vps26 mutants somewhat unique, as other mutants (e.g., vps29 or vps5) do not affect the membrane association of Vps35p, but rather cause Vps35p to shift to a different population of membranes (Seaman et al., 1997 and data not shown).

Previous studies have shown that the distribution of Vps35p on sucrose gradients can be resolved into two distinct populations of membranes (Seaman et al., 1997), while Vps5p has been shown to cofractionate with membranes of a density similar to those of small vesicles such as COPI-coated vesicles (Horazdovsky et al., 1997). The cross-linking data suggest that Vps35p and Vps5p should be present on the same population of membranes. To test this hypothesis, a P100 membrane fraction prepared from [³⁵S]methionine-labeled cells was spun to equilibrium on a sucrose density gradient. Fractions were collected, and various proteins were immunoprecipitated to determine their location in the gradient. Consistent with previously published data, Vps35p is found distributed between two populations of membranes (Fig. 5). Furthermore, there is clear colocalization between Vps35p and Vps5p in the denser fractions (fractions 8 and 9). Based upon the fractionation profiles of the late-Golgi protein Kex2p and the endosomal t-SNARE Pep12p (Becherer et al., 1996), it is possible to suggest that the lighter population of membranes with which Vps35p associates (fractions 5 and 6) is most likely an endosomal compartment. The fractions in which Vps35p and Vps5p colocalize (fractions 8 and 9) appear to be more dense than the fractions that contain Kex2p, and could correspond to a population of coated vesicles or tubules.

The data presented here and previously (Horazdovsky et al., 1997; Seaman et al., 1997) suggest that Vps5p and Vps35p are associated with a dense population of membranes whose buoyant density is similar to those of known coated vesicles. As most vesicles also have a characteristic size in the range of 40–80 nm (Schekman and Orci, 1996), we have investigated the size of the membranes on which Vps35p and Vps5p are associated by gel filtration chromatography. A yeast strain expressing an epitope-tagged Vps10p was fractionated to generate a P100 membrane fraction, which was then loaded onto a sephacryl S1000 column. The fractions collected were assayed for the presence of various proteins by Western blotting. We observed that Vps10p was most abundant in fractions 60–66 (Fig. 6 A), where the late-Golgi protein Kex2p (fractions 60–62) was also detected (data not shown). The bulk of Vps35p and Vps5p coeluted slightly later in fractions 66–72. These fractions also contained Vps29p (data not shown), indicating that Vps35p and Vps5p are present on membranes somewhat smaller than the membranes that contain the bulk of Vps10p (i.e., Golgi and endosomal membranes). As a marker for a known coated vesicle, the distribution of COPI/coatomer was determined. COPI-coated vesicles have been shown to have a size of 40–60 nm (Ostermann et al., 1993; Serafini et al., 1991), and eluted from the sizing column in fractions 66–76, although the bulk of the signal is detected in the later fractions 72–76. Thus, it appeared that Vps35p and Vps5p were present on vesicular membranes slightly larger than COPI/coatomer-coated vesicles. The fractionation of Vps35p and Vps5p in fractions smaller than Golgi or endosomes was not the result of fragmentation of membranes during homogenization, as P100 membranes prepared from osmotically lysed cells fractionated in an identical fashion (data not shown).

Fractions that contained the bulk of the Vps35p/Vps5p signal were pooled and processed for morphological analysis by EM, and were found to contain numerous small vesicles (data not shown). Attempts to label these vesicles with antisera against Vps35p, Vps5p, Vps17p, or Vps29p were not successful; however, as an alternative approach, a P100 membrane fraction from cells expressing an epitope-tagged VPS5 gene was subjected to cryoelectron microscopy. Frozen-thin sections were labeled with anti-HA antibodies and 5 nm anti-mouse colloidal gold. The gold particles were found to decorate vesicular and tubulovesicular structures of 80–120 nm (Fig. 6 B).

The distribution of Vps5p in vivo was investigated by using an epitope-tagged VPS5 gene. This has been shown to complement the vps5Δ mutant, and when expressed in a class E vps mutant is observed by immunofluorescence to associate with the exaggerated prevacuolar endosome present in class E vps mutants (Nothwehr and Hindes, 1997). Therefore, cells expressing the tagged Vps5p were fixed and prepared for cryoelectron microscopy. Frozenthin sections were labeled with a monoclonal anti-HA antibody, which was followed by 5-nm colloidal gold coupled to anti-mouse antisera. In Fig. 7, an electron micrograph of a vps4 cell expressing the HA-tagged Vps5p is shown. The prevacuolar endosome is decorated with 5-nm gold particles denoting the presence of Vps5p (Fig. 7,A). Strikingly, the labeling is localized to specific regions in the prevacuolar endosomal membrane, and appears most concentrated at the swollen rims of the cisternae that comprise the prevacuolar endosome in a class E mutant (Fig. 7 B, inset) (Babst et al., 1997; Rieder et al., 1996). This localized distribution of Vps5p is in contrast to the localization of the vacuolar ATPase, which is found evenly distributed throughout the membranes in a class E mutant (Rieder et al., 1996).

How do the proteins (Vps35p, Vps29p, Vps26p, Vps17p, and Vps5p) assemble onto the membrane, and what is the nature of the membrane-associated complex? To address this question, we have attempted to strip the complex off membranes and then separate the components of the complex by gel filtration. Previous studies have shown that Vps35p and Vps5p can be stripped off the membrane by treatment with 1M NaCl (Horazdovsky et al., 1997; Paravicini et al., 1992), and in our preliminary experiments we found that as little as 250 mM NaCl would also strip Vps35p and Vps5p from the membrane (data not shown). Therefore, a P100 membrane fraction was treated with a buffer containing 250 mM NaCl to strip the membranes. The soluble proteins were then separated away from the stripped membranes by centrifugation, and the supernatant was loaded onto a sephacryl S300 column that had been equilibrated in buffer containing 250 mM NaCl. Fractions from the column were assayed for components of the multimeric complex by Western blotting.

We observed that Vps5p coeluted with Vps17p (Fig. 8), suggesting that they remain associated. Some proteolysis of the Vps5 protein occurs during this procedure, resulting in two breakdown products that migrate a little below the intact Vps5p. Vps5p and Vps17p coeluted with a predicted size of ∼400–500 kD, which is significantly larger than a simple dimer of Vps5p/Vps17p, which has a combined molecular mass of ∼140 kD. Similarly, Vps35p coeluted with Vps29p, consistent with the data from the cross-linking experiments (Figs. 2 and 3 B). It appears that Vps35p and Vps29p also form a complex (∼230 kD) larger than the sum of their two respective molecular masses (110 kD + 30 kD). It is possible that Vps26p is also part of the Vps35p/Vps29p complex, although the lack of antiserum against Vps26p precludes direct testing of this hypothesis. Adding 40 kD of Vps26p to the Vps35p/Vps29p complex would bring the size of the complex closer to 230 kD. Furthermore, in double mutants of vps29Δ/vps26Δ there was no detectable Vps35p by Western blotting, and the protein appeared to be very unstable, with a half-life of ∼30 min (data not shown). Together with data from the cross-linking experiments, these findings support the notion that two subcomplexes of Vps35p/Vps29p/Vps26p and Vps5p/ Vps17p, respectively, assemble together on an endosomal membrane.

So far this report has focused primarily on the membrane-associated pool of the proteins being described. It must be noted, however, that a significant proportion of each of the proteins is also found in the cytoplasm (Table II). For example, ∼40–60% of the cellular Vps5p is soluble (Horazdovsky et al., 1997; Nothwehr and Hindes, 1997). The only exception is Vps35p, of which only 5% of the total Vps35p is present in the cytoplasm (Paravicini et al., 1992; Seaman et al., 1997). It is reasonable to suppose that part of the function of these proteins involves cycling on and off the membrane. We may therefore gain some insight into how these proteins assemble on the membrane by studying their properties in the cytosol. Thus, cytosol was prepared from wild-type cells, and was subjected to sephacryl S300 column chromatography. Fractions collected were assayed for various proteins by Western blotting. In Fig. 9,A the graph depicts the elution profile of Vps5p, which eluted in the void volume and appeared to be part of a very large (>10⁶ D) complex. This profile contrasts with the data presented in Fig. 8, in which Vps5p that had been stripped off the membrane with NaCl was found to fractionate with a predicted size of ∼400–500 kD. However, that fractionation was performed in the presence of 250 mM NaCl, which may have disrupted protein– protein interactions. Therefore, to see if Vps5p will assemble into the >10⁶ D complex in the absence of NaCl, a P100 membrane fraction was stripped with NaCl as before. The membranes were spun out, and the soluble proteins were dialyzed against Hepes lysis buffer to remove the salt gradually. This dialyzed supernatant was then fractionated on the sephacryl S300 column. In Fig. 9 B, dialyzed Vps5p was found in the void volume fraction, indicating that it can assemble into a >10⁶-D complex in the absence of salt. Interestingly, it was found that fractionation of cytosolic Vps5p was unchanged in any mutant tested, including a vps17Δ mutant (data not shown), hinting that Vps5p alone may possess this self-assembly activity.

When the fractionation of Vps35p and Vps29p was investigated, it was found that they too coeluted in the void volume fraction, and thus could be part of a large cytosolic complex (Fig. 9, C and D, solid lines). It must be noted, however, that the bulk of Vps29p was found in fractions consistent with its predicted molecular weight. In vps29Δ cells, the bulk of the cytosolic Vps35p was no longer in the void volume fraction, but was shifted to a later fraction (Fig. 9,C, broken line). Similarly, in the absence of Vps35p, cytosolic Vps29p no longer eluted in the void volume fraction (Fig. 9 D, broken line). Thus, Vps35p and Vps29p associate with each other in the cytoplasm in a complex that is >10⁶ D.

The experiments performed on both cytosolic Vps5p and Vps5p that had been stripped from membranes and then allowed to reassemble into a large complex (Fig. 9, A and B) suggested that Vps5p may possess intrinsic self-assembly activity. This was most strongly suggested when cytosol from vps17Δ cells was fractionated. Even though Vps5p stably associates with Vps17p (Horazdovsky et al., 1997), cytosolic Vps5p from vps17Δ cells continued to elute in the void volume fraction (data not shown), suggesting a self-assembly property intrinsic to Vps5p. This hypothesis was tested directly by expression of Vps5p as a recombinant fusion protein. A GST-Vps5p fusion expressed in yeast has previously been shown to be functional (Fig. 2,C), and hence an identical fusion was expressed in bacteria. The GST-Vps5p was purified by glutathione affinity chromatography, dialyzed against Hepes lysis buffer, and fractionated on a sephacryl S300 column. In Fig. 10,A, the elution profile from the column is shown. GST-Vps5p (solid line) eluted in the void volume fraction, while GST alone (broken line) eluted in fractions consistent with its molecular weight. Fractions from the gel filtration column were analyzed by SDS-PAGE (Fig. 10 B). Lane M contains molecular weight markers, and lane L contains a sample of the material loaded onto the column. It is apparent that GST-Vps5p is very susceptible to degradtion, and that most of the protein loaded onto the column is in fact a breakdown product of ∼55 kD. However, the majority of the protein that elutes in the void volume fraction is full-length GST-Vps5p with a molecular mass of ∼120 kD. As the breakdown products of the GST-Vps5p were not enriched in the void volume fraction, it seems unlikely that the assembly of GST-Vps5p into the high molecular weight complex was not the result of nonspecific aggregation.

Furthermore, analysis of the void volume fraction by EM revealed that the GST-Vps5p had assembled into a homogeneous population of 15–20-nm spherical structures. In Fig. 10 C, two panels of four electron micrographs showing the void volume fraction bound to the electron microscope grid are shown. In the top four micrographs, the void volume fraction was bound to the grid, and was negatively stained with uranyl acetate, while in the lower four micrographs the bound GST-Vps5p was labeled with anti-Vps5p antisera followed by anti-rabbit coupled to 5-nm colloidal gold before negative staining. Generally one or two gold particles per structure were observed, reflecting a probable lack of access of the antibody to the antigen. The regular homogeneous appearance and the GST-Vps5p particles along with the discrete antibody labeling seen strongly argues for specific assembly of the GST-Vps5p, and not for nonspecific aggregation. Recombinant Vps5p therefore assembles into a high molecular weight complex similar to the endogenous Vps5p.

Previous studies have identified a number of vps mutants that affect the localization of the vacuolar hydrolase receptor Vps10p, and have been proposed to act in the endosome-to-Golgi retrieval of not only Vps10p, but other late-Golgi resident proteins such as Kex2p and DPAP A (Horazdovsky et al., 1997; Nothwehr and Hindes, 1997; Seaman et al., 1997). These mutants could be divided into two distinct groups based upon phenotypic analysis. The first group comprises vps35, vps29, and vps30, all of which cause a severe defect in trafficking of CPY to the vacuole, even though another vacuolar hydrolase, proteinase A (PrA), reaches the vacuole and is matured with wild-type kinetics (Seaman, et al., 1997). The defect in the trafficking of CPY to the vacuole is the result of complete mislocalization of Vps10p to the vacuole membrane in the vps35, vps29, and vps30 mutants. These mutants therefore appear to affect localization of a subset of proteins, one of which is Vps10p, and have been proposed to act in a cargo-selective manner in the endosome-to-Golgi retrieval pathway (Seaman et al., 1997). The second group comprises vps5 and vps17 mutants. Superficially, the vps5 and vps17 mutants have quite different phenotypes from the first group. In vps5 and vps17 mutants there is a significant defect in PrA maturation, and the vacuoles appear fragmented. Vps10p is mislocalized to the vacuolar membrane in vps5 and vps17 mutants; however, this mislocalization is less dramatic than that observed in the vps35, vps29, and vps30 mutants. Thus, it was suggested that Vps5p and Vps17p are part of a more general mechanism for endosome-to-Golgi retrieval such that the vps5 and vps17 mutants cause a complete block in retrieval, ultimately leading to a partial defect in forward transport from the Golgi to the endosome due to lack of retrieval of factors required for forward transport (Horazdovsky et al., 1997).

Here we show that five peripheral membrane proteins— Vps35p, Vps29p, Vps26p, Vps5p, and Vps17p—assemble together to form a multimeric complex on a prevacuolar membrane. Vps5p has been localized to the prevacuolar membrane, and was found to be clustered at specific sites that appear to be nascent budding vesicles. Furthermore, Vps5p can assemble in vitro in a large macromolecular structure. These data, along with previous studies and phenotypic characterization of the various mutants, point to a role for these proteins as a membrane coat complex that directs retrieval of Vps10p and other molecules from the endosome to the Golgi. To denote its role in retrograde transport from the endosome to the Golgi, we term this complex the retromer.

The behavior and properties of the individual components of the retromer complex have been examined to understand the role that each component plays within the complex. A genetic interaction between VPS35 and VPS29 was identified through the use of a dominant negative allele of VPS35. Only overexpression of VPS29 was able to suppress the CPY sorting defect seen in cells carrying the vps35dn allele. This result could be interpreted two ways: either (a) the dominant negative vps35 titrated Vps29p out of the system into a nonfunctional complex; the resulting loss of Vps29p would then cause the defect in CPY sorting; or (b) Vps29p acts downstream of Vps35p, and hence overexpression can bypass the defect observed in cells overexpressing the vps35dn allele. We favor the former, as no suppression of the previously described (Seaman et al., 1997) temperature conditional allele of vps35 has been observed in cells overexpressing Vps29p (Seaman and Emr, unpublished data), which would be expected if Vps29p acts downstream of Vps35p. Furthermore, cell fractionation experiments clearly show that the proportion of Vps29p that associates with a P100 (Golgi/endosome/vesicle) membrane fraction requires the presence of Vps35p for that membrane association.

Cross-linking of both a crude lysate and the P100 membrane fraction demonstrated a direct interaction between Vps35p and Vps29p, and also revealed the presence of three other proteins in a complex with Vps35p/Vps29p. These proteins were subsequently identified as Vps26p, Vps5p, and Vps17p, and are all peripheral membrane proteins that are found in both the cytoplasm and on P100 membranes (Table II). Cross-linking studies performed on extracts of mutant cells indicated that the retromer complex in fact comprises two distinct subcomplexes. When lysates from vps5 or vps17 mutant cells were cross-linked, it was found that Vps35p remained associated with both Vps29p and Vps26p. As previous studies had reported that Vps5p and Vps17p form a stable complex with each other (Horazdovsky et al., 1997), the two retromer subcomplexes would then comprise Vps35p/Vps29p/Vps26p and Vps5p/Vps17p. These data were supported by an experiment in which P100 membranes were stripped with salt to dissociate the peripheral membrane proteins. When the solubilized proteins were subjected to gel filtration to size the individual components, it was observed that Vps5p remained associated with Vps17p in a complex with a predicted size of ∼400–500 kD. This value is significantly larger than the sum of the molecular masses of the individual components (∼140 kD), and requires further investigation to establish the nature of this complex. One possibility, however, is that Vps5p/Vps17p can assemble into structures containing three or possibly four copies of a simple Vps5p/Vps17p dimer, producing a multimer of ∼400–500 kD.

Although larger than its predicted size, the Vps35p/ Vps29p complex (∼230 kD) may also contain Vps26p (∼40 kD). Association of Vps26p with the Vps35p/Vps29p complex remains to be proven, however, and awaits production of antisera against endogenous Vps26p. Mutants of vps26 cause Vps10p to become localized to a vacuolar membrane fraction in a manner indistinguishable from that of vps35 or vps29 mutants, consistent with Vps26p interacting with Vps35p and/or Vps29p. Furthermore, the acute instability of Vps35p in a double vps29Δ/vps26Δ mutant supports the notion that Vps35p forms a subcomplex with both Vps29p and Vps26p. Although vps30 mutants have phenotypes almost identical to those of vps29 and vps35 mutants, no Vps30p was detected in the retromer complex, and the cross-linking data indicated that Vps30p was not required for complex formation. Although untested, it seems likely that Vps30p may somehow act to regulate complex formation, or alternatively Vps30p may function downstream of the retromer complex.

It is worth noting that the components of this cross-linkable membrane-associated retromer complex display varying degrees of membrane association. Vps35p is 80% membrane-associated, while only 20% of Vps29p is localized to the membrane, and Vps5p/Vps17p and Vps26p fall in between these two values (Table II). If all five proteins were stably associated with each other all of the time, it would be reasonable to expect them all to be equally membrane-associated or cytosolic, which is clearly not the case. Hence, it seems likely that formation and disassembly of the complex is a dynamic process involving cycling on and off the membrane.

As all the proteins that comprise the retromer complex are peripheral membrane proteins, and therefore presumably cycle between the cytoplasm and the membrane, cytosol was size-fractionated to investigate what sort of complexes exist in the cytoplasm. Strikingly, it was found that cytosolic Vps5p existed in a very large complex that eluted in the void volume fraction with a predicted size >10⁶ D. As this fractionation was performed in a buffer that did not contain high salt, it seemed possible that in the absence of the salt, Vps5p could assemble (presumably with other proteins; for example, Vps17p) into a large complex. This possibility was tested directly by allowing the retromer complex components that had been salt-stripped from membranes to assemble in vitro by dialyzing away the salt. It was found that Vps5p could indeed assemble into a large complex that was >10⁶ D. Even though Vps5p forms a stable complex with Vps17p and requires Vps17p for its membrane association (Horazdovsky et al., 1997), Vps17p was not required for Vps5p to assemble into the >10⁶-D complex. Interestingly, however, we have found that Vps17p in vps5Δ cells is very unstable, and is rapidly degraded with a half-life of ∼30 min (Seaman and Emr, unpublished data). These data hint that in vivo, Vps5p and Vps17p play different roles within their subcomplex. Vps17p is required for membrane association, while Vps5p appears to have intrinsic assembly activity. To test the ability of Vps5p to assemble in vitro, a fusion protein expressed in bacteria was purified by affinity chromatography. Consistent with the behavior of cytosolic Vps5p, the recombinant GST-Vps5p was found to assemble into a complex that elutes in the void volume from a gel filtration column, and forms homogenous particles 15–20 nm in diameter. These structures, however, probably do not perfectly mirror the structures that endogenous Vps5p forms in vivo, as Vps17p would be expected to play a role in determining how and where Vps5p assembles in the cell.

It appeared from the gel filtration analysis of cytosol that Vps35p could also assemble into a large structure. This activity was dependent upon Vps29p, and may reflect an oligomerization property of Vps35p. As almost all of the Vps35p in the cell is membrane-associated (Paravicini et al., 1992), this in vitro assembly of cytosolic Vps35p may reflect an in vivo property of the membrane-associated Vps35p. We can imagine that if Vps35p acts to recruit specific cargo (e.g., Vps10p) into vesicles for retrieval, a clustering of Vps35p through some sort of oligomerization would facilitate the concentration of receptors in a specific membrane domain. Vps29p would then serve a catalytic role, activating Vps35p and promoting its oligomerization. Thus, in a vps29 mutant, Vps35p may remain associated with the membrane, and may still associate with the cytoplasmic tails of receptors such as Vps10p, but will no longer cluster to concentrate the receptors into a vesicle. This hypothesis is consistent with the observed behavior of Vps35p in a vps29 mutant, in which Vps35p follows Vps10p to a vacuolar membrane fraction (Seaman et al., 1997).

Sucrose density gradient analysis of the P100 membrane population with which Vps35p and Vps5p are associated suggested that Vps35p is associated with two distinct pools of membranes; a lighter endosomal fraction and a denser membrane fraction. Vps5p is predominantly associated with only the denser pool of membranes, which has previously been shown to have a buoyant density very similar to known coat vesicles such as COPI and COPI (Horazdovsky et al., 1997). These observations are consistent with the hypothesis that Vps35p/Vps29p (and also Vps26p) are involved in the recruitment of specific cargo into vesicles. A proportion of Vps35p and Vps29p would logically be expected to be associated with endosomal membranes, and not present in vesicles. Vps5p/Vps17p may then assemble onto the membrane where the receptors are clustered, and drive the budding of a vesicle. Size fractionation of membranes by sephacryl S1000 column chromatography shows that Vps5p and Vps35p cofractionate on a population of membranes smaller than Golgi membranes where Vps10p and Kex2p reside, but slightly larger than the COPI-coated vesicles.

When fractions from the sephacryl S1000 column were analyzed by EM, they were found to contain numerous small 40–100-nm vesicles. However, attempts to label these fractions with antibodies against retromer components were unsuccessful, possibly because the membrane association of the retromer proteins is somewhat labile, and during the course of fractionation or possibly after elution from the column, the proteins dissociated from the membrane. Therefore, a cruder P100 membrane fraction was labeled with antibodies against an epitope-tagged VPS5, and was found to be predominately localized to vesicular and tubulovesicular structures. In vivo localization of Vps5p was achieved using a class E mutant (vps4) that accumulates an exaggerated endosomal compartment (Babst et al., 1997; Rieder et al., 1996). In this instance, Vps5p was found to be localized to the swollen rims of the cisternal membranes that are characteristic of the class E endosome. These structures are not dissimilar to the membranes with which the epitope-tagged Vps5p is associated when a P100 membrane fraction is analyzed by cryoimmunogold EM. The key observation is that Vps5p is clustered on specific regions of the prevacuolar compartment, not evenly distributed throughout as has been shown for the vacuolar ATPase that also accumulates in the prevacuolar endosome in a class E mutant (Rieder et al., 1996). That Vps5p appears localized to the swollen rims of the cisternae is consistent with its proposed role in vesicle formation.

Mammalian homologues of the VPS35, VPS29, VPS26, and VPS5 genes have been identified either as ESTs, or as full-length cDNA clones. VPS35 is homologous to the mouse gene Mem3 (Hwang et al., 1996), VPS29 has EST homologues in both mouse and humans (Seaman, et al., 1997), and the mammalian homologue of VPS26/PEP8 (Bachhawat et al., 1994) is termed Hβ58, and is essential for embryogenesis in the mouse (Lee et al., 1992). Interestingly, Vps26p shares modest sequence homology (27% identity over a stretch of 130 aas) with the Sec21 protein, a component of the yeast COPI vesicle coat. The best-characterized mammalian homologue of the retromer components is SNX1, which is homologous to VPS5 (Horazdovsky et al., 1997; Nothwehr and Hindes, 1997). SNX1 has been shown to interact with the cytoplasmic tail of the EGF receptor, and overexpression of SNX1 causes an increase in the turnover and degradation of the EGF receptor (Kurten et al., 1996). It has been suggested that SNX1 is required for endosome-to-lysosome trafficking, and may play a role in the downregulation of activated receptor tyrosine kinases (Kurten et al., 1996). SNX1 and Vps5p share most sequence homology in a conserved domain that has been termed the PX domain (Ponting, 1996). The precise role that this conserved domain plays in the function of Vps5p and SNX1 remains to be determined, but it is interesting to note that Vps17p also contains a PX domain, as does Grd19p (Voos and Stevens, 1998) and Mvp1p (Ekena and Stevens, 1995), two proteins both believed to be important for endosome-to-Golgi retrieval in yeast. The PX domain may therefore confer a function common to all these proteins, perhaps targeting to an endosomal membrane by binding a specific lipid. This mechanism, however, remains to be tested.

The data presented in this report are consistent with the complex containing Vps35p, Vps29p, Vps26p, Vps5p, and Vps17p acting in formation of vesicles for endosome-to-Golgi retrieval, but do not exclude the possibility that other proteins also function in this pathway (e.g., Grd19p, Mvp1p, Vps30p, and Vps1p). One of these candidates (Vps1p, a dynamin-like GTPase) has been shown to be required for proper localization of Kex2p (Wilsbach and Payne, 1993a). Furthermore, vps1 and vps26 mutants have been classified in the same phenotypic group, implying an interaction or action of both proteins at the same step (Raymond et al., 1992).

Clathrin has also been proposed to act in retrieval from the endosome to the Golgi, and clathrin mutants also affect the localization of Kex2p (Payne and Schekman, 1989; Seeger and Payne, 1992b; Wilsbach and Payne, 1993b). Interestingly, however, the effect of clathrin inactivation upon the localization of Kex2p occurs independently of the putative late-Golgi localization signal in the cytoplasmic tail of Kex2p (Redding et al., 1996). Hence, it seems more likely that clathrin functions at the Golgi, and not in endosome-to-Golgi retrieval.

Thus, we propose that Vps35p acts with Vps29p and Vps26p to select cargo for retrieval from the endosome to the Golgi by interactions (either directly or indirectly via accessory factors) with the cytoplasmic tails of proteins such as Vps10p and Kex2p (Fig. 10). Aromatic amino acid– containing motifs in the tails of Vps10p and Kex2p have been shown to be important for their retrieval to the Golgi (Wilcox et al., 1992; Cereghino et al., 1995; Cooper and Stevens, 1996), and these motifs are similar to a diaromatic motif present in the tail of the M6P receptor that is important for its retrieval to the Golgi (Schweizer et al., 1997). Vps5p/Vps17p assemble onto the clustered receptors through interactions with Vps35p/Vps29p/Vps26p and/or the membrane. The self-assembly activity of Vps5p may drive the budding of the vesicle, although this function may also be provided by a different component of the retromer complex or by an as yet unidentified protein. The retromer complex therefore appears to display the two key activities associated with a vesicle coat complex–cargo recogniton and self-assembly. Further studies are required to elucidate the precise role the retromer components play in endosome-to-Golgi retrieval, and the regulatory mechanisms such as a docking protein and/or a small GTPase that govern vesicle coat assembly on endosomal membranes have also yet to be revealed. However, these proteins provide the best candidate for a vesicle coat complex functioning in endosome-to-Golgi retrieval, and a similar complex may function in mammalian cells at an analogous point. Cloning and characterization of the mammalian homologues of the yeast retromer components will aid the molecular dissection of endosome-to-Golgi retrieval.

This work was supported by National Institutes of Health grant CA 58689 (to S.D. Emr). M.N.J. Seaman was supported by a postdoctoral research fellowship from the European Molecular Biology Organization and the Howard Hughes Medical Institute. S.D. Emr is supported as an Investigator of the Howard Hughes Medical Institute.

CPY

carboxypeptidase Y

DPAP A

dipeptidyl amino peptidase A

DTSSP

3,3′-dithiobis sulfosuccinimidyl-proprionate

GST

glutathione S-transferase

HA

hemagglutinin

M6P

mannose-6-phosphate