INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

The organization of the secretory pathway requires that membrane proteins be sorted: individual proteins have to be selectively incorporated into, or excluded from, the transport vesicles that move between organelles. This sorting is determined ultimately by the amino acid sequence, and the fate of many proteins is determined by motifs present in their cytoplasmic tails, which can interact directly with the various coat proteins that mold transport vesicles (Cosson and Letourneur, 1994; Ohno et al., 1995; Honing et al., 1998; Rapoport et al., 1998; Rodionov and Bakke, 1998).

A second form of sorting is dependent on transmembrane domains (TMDs). In the case of Golgi enzymes, the length of the TMD has a powerful influence on their retention in the Golgi system, and this has been suggested to be due to their selective partitioning into subdomains of the lipid bilayer that differ in composition and hence thickness (Bretscher and Munro, 1993; Munro, 1995). TMD-dependent protein sorting has been demonstrated in both the late Golgi and the early endosomes of yeast cells (Rayner and Pelham, 1997; Lewis et al., 2000). Some endoplasmic reticulum (ER) proteins, such as yeast Sec12p, are also retained by mechanisms that involve, at least in part, hydrophilic residues in their TMDs. A putative receptor, Rer1p, itself an integral membrane protein, is thought to be involved in recognizing the TMDs in the Golgi and mediating their retrograde transport to the ER (Sato et al., 1996, 1997; Boehm et al., 1997).

Recognition and sorting of TMDs may occur in part as a protective measure. Their sequences are usually hydrophobic, allowing them to mingle freely with lipids, but polar residues are sometimes present, and in the hydrophobic environment these have a strong tendency to form hydrogen bonds with each other (for discussion, see Bowie, 2000). This can lead to the stabilization of TMD interactions within multispanning proteins and between subunits of oligomers, but unpaired polar residues are potentially hazardous to the cell; like hydrophobic residues exposed to the aqueous environment, they may form promiscuous interactions that cause aggregation or interfere with the folding of multispanning proteins. It seems likely that cells have evolved mechanisms to "chaperone" individual polar TMDs, such as those in unassembled protein subunits.

A potential example is provided again by Rer1p, which in addition to sorting Sec12p has also been implicated in the retention in the ER of proteins with potentially charged residues in their TMDs (Letourneur and Cosson, 1998). ER retention of such proteins is well documented in animal cells, particularly among the individual subunits of the T cell receptor, which contain acidic and basic residues and are held in the ER until assembly masks them (Bonifacino et al., 1990, 1991).

In this paper, we address the role of TMDs in the sorting of proteins in late endosomes. In yeast, as in animal cells, multivesicular bodies (MVBs) form in the endocytic pathway by a process in which vesicles invaginate from the outer endosomal membrane and pinch off to form discrete structures within the lumen. Fusion of the outer endosomal membrane with the vacuole then delivers these internal vesicles to the vacuolar interior, where they are eventually degraded (Burd et al., 1998; Odorizzi et al., 1998). Proteins destined for degradation enter these internal vesicles, whereas proteins that function on the vacuolar surface avoid this fate. One functional membrane protein, the vacuolar protease Cps1p, is also known to enter the internal vesicles (Odorizzi et al., 1998).

The situation is complicated by the existence of at least two quite distinct transport routes to the vacuole for membrane proteins. Most proteins, whether they come from the Golgi complex or the plasma membrane, pass through late endosomes. Accordingly, they become trapped in an aberrant endosomal structure in so-called class E vacuolar protein sorting mutants such as vps4, which are defective in the late stages of this pathway (Raymond et al., 1992; Rieder et al., 1996; Babst et al., 1997; Hicke et al., 1997; Bryant et al., 1998; Conibear and Stevens, 1998; Odorizzi et al., 1998). However, a small number of proteins, including the vacuolar alkaline phosphatase (ALP) and the syntaxin Vam3p, have been shown to be transported directly from the Golgi to the vacuole, even in class E mutants, by a mechanism that depends on the AP-3 adaptor coat protein complex (Cowles et al., 1997a,b; Piper et al., 1997; Stepp et al., 1997; Darsow et al., 1998).

We have used SNARE proteins as models to test the features that determine whether proteins reach the vacuole surface or its interior. The latter process we refer to as internalization; to avoid confusion with endocytosis, we use the term only in this sense. In particular, we have studied the vacuolar proteins Nyv1p and Vam3p (Darsow et al., 1997; Nichols et al., 1997; Wada et al., 1997; Srivastava and Jones, 1998) and the endosomal syntaxin Pep12p (Becherer et al., 1996; Burd et al., 1997; Lewis et al., 2000). We show that Nyv1p, like Vam3p, uses the AP-3 pathway to reach the vacuole and that this is sufficient to prevent internalization, which evidently can occur only in late endosomes. Pep12p is present in late endosomes and can reach the vacuole from them, but it is not normally internalized. Strikingly, however, we find that the insertion of acidic residues in the cytoplasmic half of the TMD is sufficient to force internalization of Pep12p, and of Vam3p if this protein is routed through endosomes. Mutant TMDs that cause this internalization can also affect sorting at other locations. Depending on the protein to which they are appended, they can induce partial ER retention or transport of a protein from Golgi to endosomes, rather than to the cell surface. Testing of a variety of naturally occurring TMDs shows that internalization in late endosomes does not exclusively require acidic residues but is promoted by other polar residues, and that the features required are strikingly similar to those that, in other contexts, promote ER retention.

We suggest that cells have multiple quality control mechanisms that recognize proteins with abnormally hydrophilic TMDs. These mechanisms ensure either the retention of such proteins in the ER or their delivery to the interior of the vacuole for degradation.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Yeast Strains

Yeast strains are described in Table 1. For gene disruption, the entire coding region was replaced with selectable markers with the use of PCR primers containing ~50 bases of identity to the regions flanking the ORF. APM3 was disrupted with the HIS5 gene of Schizosaccharomyces pombe flanked by coliphage loxP sites. A pho13 pho8 mutant was obtained by disrupting PHO13 with HIS5 flanked by loxP sites and subsequently disrupting PHO8 with TRP1 also flanked by coliphage loxP sites. The selectable markers were removed by excisive recombination at the loxP sites with the transient expression of Cre recombinase (Sauer, 1987).

Plasmids

Plasmids expressing SNARE chimeric constructs were all based on pRS406 or pRS416 vectors (Sikorski and Hieter, 1989). Unless indicated, PCR-generated derivatives were cloned as an EcoRI-BamHI fragment behind sequences expressing the mut2 green fluorescent protein (GFP) variant (Cormack et al., 1996) and the TPI1 promoter as described by Wooding and Pelham (1998). Constructs shown in Figures 1 and 4 were cloned in pRS406, whereas constructs shown in all other figures were cloned into pRS416.

Plasmids expressing Snc1p and Snc1p-Nyv1p (here called GSN) have been described previously (Lewis et al., 2000). GSV (SNC1-VTI1), GNS (NYV1-SNC1), and PSEC (PEP12-SEC12) and PCPS (PEP12-CPS1) constructs were obtained by PCR of SNC1 (Lewis et al., 2000), NYV1 (Nichols et al., 1997) and PEP12 (Rayner and Pelham, 1997), respectively, with the use of primers encoding the appropriate transmembrane domain. PEP12 deletions and point mutations were obtained by PCR with the use of appropriate primers. Point mutations in PSSO (PEP12-SSO1), VAM3, SSO1, and SSSO (SNC1-SSO1) were introduced also by PCR of the corresponding constructs described elsewhere (Nichols et al., 1997; Rayner and Pelham, 1997; Lewis et al., 2000). For PEP12-UFE1 chimeras, the cytosolic N terminus of Pep12p was cloned by PCR as an EcoRI-HindIII fragment. Previously described mutants of the UFE1 TMD (Rayner and Pelham, 1997) were inserted as HindIII-BamHI fragments. The HindIII site introduces the amino acids KL at the junction between the cytoplasmic domain and TMD. In all cases, changes were confirmed by sequencing.

A truncated version of the PHO8 gene (lacking the first 60 N-terminal amino acids) was cloned as a StuI-EcoRI fragment into the expression plasmid described above cut with HindIII, blunted with Klenow, and finally digested with EcoRI to eliminate GFP.

To check the ability of PEP12 and PSSO constructs to complement the pep12 phenotype, the TPI1 promoter and GFP were replaced with genomic sequence from the 820 base pairs (bp) immediately upstream of the PEP12 ORF. Genomic sequence from the 300 bp immediately downstream of the PEP12 ORF was also inserted as an BamHI-XbaI fragment.

Imaging of Live Cells

Cells were grown to early log phase, harvested by centrifugation, and washed with water to promote homotypic fusion of fragmented vacuoles. They were placed onto slides coated with concanavalin A, sealed under coverslips, and examined with the use of either a Zeiss (Thornwood, NY) Axioskop microscope equipped with a Micromax charge-coupled device camera (Princeton Instruments, Princeton, NJ) (Figure 1 only) or an MRC-600 confocal microscope (Bio-Rad, Richmond, CA). Fluorescence intensity profiles were generated with the use of NIH Image software.

To stain the endocytic pathway, a 1-ml culture of early log phase cells was centrifuged and cells were resuspended in 100 µl of medium containing 20 µM FM4-64 (Molecular Probes, Eugene, OR). They were incubated for 15 min, harvested by centrifugation, resuspended in 1 ml of fresh medium, and incubated for another 30 min. They were collected by centrifugation, washed with 1 ml of water, and resuspended in water for microscopy.

Random PCR Mutagenesis

Random PCR mutagenesis of PSSO was performed as described (Cadwell and Joyce, 1992) with the use of primers flanking the PEP12 ORF by 200 bp. PCR products were cotransformed with the expression plasmid described above that was linearized with EcoRI-BamHI. Yeast colonies were scored microscopically for their ability to internalize the GFP-PSSO construct. To map the region responsible for internalization of the mutant protein, sequences coding for the first 101 amino acids (EcoRI-NheI), the central 152 amino acids (NheI-KpnI), and the last 35 amino acids (KpnI-BamHI) were swapped with the unmutated version.

Protein Extraction and Immunoblotting

Early log phase cultures were lysed, and proteins were extracted as described previously (Reggiori et al., 1997). Western blot analysis was performed with the use of rabbit polyclonal antibody to Pep12p (Holthuis et al., 1998), to Vti1p (a gift from T. Stevens, University of Oregon, Eugene), or to GFP (Molecular Probes). All antibody incubations were carried out in PBS containing 3% dried milk. After incubation with anti-rabbit peroxidase-conjugated antibodies (Bio-Rad), detection was performed with the use of ECL reagents (Amersham, Arlington Heights, IL). Coimmunoprecipitation of Vam3p and Vti1p was carried out with anti-Vam3p antibody (Nichols et al., 1997) as described (Holthuis et al., 1998).

ALP Liquid Assay

This was derived from the procedure described by Toh-e et al. (1976). Early log phase cells (2.5 OD₆₀₀) were harvested by centrifugation, washed with 1 ml of ice-cold water, and resuspended in 100 µl of lysis buffer (50 mM Tris-HCl, pH 7.4, 1% Triton X-100, 5% DMSO, 1 µg/ml aprotinin, 1 µg/ml leupeptin, 5 µg/ml pepstatin, 1 µg/ml antipain, 1 mM benzamidine, 200 µg/ml PMSF). Glass beads were added, and cells were disrupted by agitation at 4°C for 1 min. After pelleting of cell debris, 20 µl of lysate was mixed with 480 µl of a p-nitrophenyl phosphate solution (Sigma Fast p-nitrophenyl phosphate tablet sets, Sigma Chemical, St. Louis, MO) prewarmed at 37°C. Incubation was carried out at 37°C for 20 min and stopped by adding 500 µl of 10% trichloroacetic acid. Samples were neutralized by adding 1 ml of sodium sulfite-saturated water solution and centrifuged at 13,000 rpm for 10 min. One milliliter of supernatant was taken to measure the absorbance at either 420 or 550 nm. ALP activity was expressed in Miller units (units of activity per minute per OD₆₀₀): 1000 × (A₄₂₀  1.06A₅₅₀)/(0.993 × min × vol × OD₆₀₀), where OD₆₀₀ = 25 and vol = 0.02 ml.

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

The Cytoplasmic Domains of SNAREs Determine Their Route to the Vacuole

We set out to investigate the features of SNAREs, as model membrane proteins, that determine whether they remain on the outer membrane or are incorporated into internal vesicles when delivered to the vacuole. Specifically, we sought to determine whether sorting signals could be identified in their cytoplasmic domains, transmembrane domains, or both. The main focus of this study is on the membrane-spanning segments, but it was first necessary to determine the contributions of the cytoplasmic domains.

Initially, we compared two related SNAREs: Nyv1p, which is normally present on the outer vacuolar membrane, and Snc1p, an exocytic SNARE that can be redirected to the vacuole interior. The locations of these and other SNARE proteins in living cells can be monitored readily with the use of GFP chimeras. Figure 1A shows typical distributions of GFP-Snc1p, which is found on the plasma membrane and also on early endosome and Golgi compartments through which it recycles (Lewis et al., 2000), and of GFP-Nyv1p. Previous studies have shown that when the transmembrane domain of Snc1p is altered, the protein fails to recycle from early endosomes to the Golgi and instead is mislocalized to the vacuole. The altered Snc1p is internalized in transit and, unlike Nyv1p, ends up in the vacuolar lumen (Lewis et al., 2000). To determine whether the difference between the fates of Nyv1p and Snc1p was due to their cytoplasmic or transmembrane domains, we exchanged their TMDs. As shown in Figure 1B, the addition of the Nyv1p TMD to Snc1p (construct GSN) did not prevent its internalization. A similar result was obtained when the Snc1p TMD was replaced with that of Vti1p (GSV), another SNARE that is found on the vacuolar membrane (Ungermann et al., 1999). Conversely, replacement of the Nyv1p TMD with that of Snc1p did not affect its localization (GNS). Hence, the fates of these SNAREs seem to be determined largely by their cytoplasmic domains.

View larger version (54K): [in this window] [in a new window]   Figure 1.   Nyv1p travels to the vacuole via the AP-3 pathway. (A) GFP-tagged versions of Snc1p and Nyv1p were expressed in wild-type cells (wt), and Nyv1p was expressed in apm3 and vps4 mutants. The structures visible with GFP-Nyv1p in wild-type and vps4 cells are vacuoles. (B) GFP-tagged versions of Snc1p with the TMD from Nyv1p (GSN) or Vti1p (GSV), or GFP-tagged Nyv1p with the TMD from Snc1p (GNS), were expressed in the indicated strains. The vps4 strains were also stained with FM4-64 to reveal the PVC and vacuolar membranes; arrowheads indicate equivalent positions in the FM4-64 and GFP images.

A possible explanation is that the cytoplasmic domain of Nyv1p (like that of Vam3p) specifies transport via the AP-3 pathway, bypassing the site of internalization in late endosomes. In agreement with this, delivery of Nyv1p to the vacuole was unaffected in vps4 cells (Figure 1A). It also reached the vacuole in a vps45 mutant (our unpublished observations) in which delivery to late endosomes is blocked (reviewed by Conibear and Stevens, 1998). In apm3 cells, it still reached the vacuole, but rather than being restricted to the outer vacuolar membrane, GFP fluorescence was visible within the vacuoles. This finding suggests that Nyv1p is rerouted to late endosomes under these conditions, as are other substrates of the AP-3 pathway (Cowles et al., 1997a; Stepp et al., 1997), and becomes partially internalized there.

As expected, sorting to the AP-3 pathway was determined by the cytoplasmic domain of Nyv1p rather than its TMD. Thus, Nyv1p with the Snc1p TMD (construct GNS) reached the vacuole in vps4 cells but accumulated on the plasma membrane in apm3 cells (Figure 1B), reflecting the fact that GNS lacks both an endocytic signal and the Golgi-endosome targeting information provided by the Nyv1p TMD (Lewis et al., 2000). Conversely, Snc1p derivatives bearing the Nyv1p (GSN) or Vti1p (GSV) TMDs were delivered to the vacuolar lumen in apm3 cells, but in a vps4 mutant they accumulated in the abnormal prevacuolar compartments (PVCs) that accumulate in this mutant and can be visualized with the use of the endocytic tracer dye FM4-64 (Vida and Emr, 1995).

In the case of Vam3p and ALP, an acidic dileucine motif has been identified as an AP-3 pathway transport signal (Darsow et al., 1998; Vowels and Payne, 1998). Inspection of the Nyv1p sequence revealed at least three potential motifs of this form (residues 97-102, 117-122, and 192-197). Preliminary deletion analysis suggested that at least two of these sequences contribute to its sorting (our unpublished observations), but we did not investigate these in detail.

We conclude that Nyv1p normally travels via the AP-3 pathway and that this ability to bypass the site at which MVBs form may account in part for the presence of Nyv1p on the vacuolar surface. When forced to travel instead via endosomes, GFP-Nyv1p appeared to be at least partially internalized.

Pep12p Avoids Internalization in Late Endosomes

To allow further investigation of the sorting signals active during the formation of MVBs, we sought an example of a SNARE that clearly avoids internalization in these structures. A good candidate is Pep12p, which is present in late endosomes and sometimes can be observed on vacuoles. Indeed, overexpression (by approximately eightfold) of a GFP-tagged version of Pep12p resulted in substantial vacuolar fluorescence, the protein clearly being present on the outer vacuolar membrane (Figure 2A). The different fates of Pep12p and Snc1p upon delivery to the vacuole are determined by their cytoplasmic domains: replacing the Pep12p TMD with that of the plasma membrane syntaxin Sso1p (Figure 2A, PSSO) (Aalto et al., 1993) or of other SNAREs (Vam3p, Nyv1p, Vti1p, Tlg1p, or Snc1p) did not affect its location on the vacuolar membrane (Table 2), whereas when Snc1p was joined to the same set of TMDs it was entirely internalized in each case (Figure 1B) (Lewis et al., 2000; our unpublished observations).

View larger version (38K): [in this window] [in a new window]   Figure 2.   Pep12p with the Sso1p TMD (PSSO) reaches the outer vacuolar membrane via late endosomes. (A) Wild-type cells (wt) or indicated mutants expressing GFP-Pep12p or PSSO. GFP-Pep12p labels vacuoles, as shown by the FM4-64 staining; arrowheads indicate equivalent positions in these double-label images. PSSO labels vacuoles in wild-type and apm3 cells but labels perivacuolar PVC in vps4 cells, which also labels with FM4-64 (arrowheads). (B) Deletion mutants of PSSO, lacking the indicated residues of Pep12p sequence, are also found on the outer vacuolar membrane. The junction with Sso1p occurs at residue 261.

The different behavior of the Pep12p and Snc1p cytoplasmic domains implied either an internalization signal in Snc1p or an anti-internalization signal in Pep12p. To search for possible signals in Pep12p while excluding any possible influence of its normal TMD, we mutated the GFP-Pep12p construct bearing the Sso1p TMD (PSSO). Controls confirmed that this construct traveled to the vacuole via endosomes, because its passage was blocked in vps4 cells but unaffected in apm3 cells (Figure 2A). We found that removal of the first 129, 186, or 243 residues of the Pep12p sequence, or the last 69 residues (193-261, leaving the Sso1p TMD), did not prevent transport to the vacuole, and in each case transport was still sensitive to vps4 (our unpublished observations). However, none of these truncated proteins was internalized (Figure 2B). Next, we introduced random mutations into the Pep12p sequence by error-prone PCR and screened for mutants that were efficiently internalized into the vacuole. Although we obtained multiple simultaneous amino acid changes by this procedure (see below), no cytoplasmic mutations that led to internalization were identified.

Thus, Pep12p appears not to contain a cytoplasmic signal that prevents internalization. Instead, the small SNAREs Snc1p and Nyv1p may be specifically internalized; it is possible that these proteins, which in contrast to the syntaxins are likely to be largely unstructured (Hazzard et al., 1999), are recognized by the cell as "unfolded" and hence suitable for degradation. In any case, it appears that Pep12p provides a robust model protein in which to test the potential effects of TMDs on internalization.

Acidic Residues in the TMD Cause Internalization

An initial clue that TMD sequences are indeed important came from the random mutagenesis of the Pep12-Sso1 construct. Three mutants were obtained that, when cells picked directly from yeast colonies were examined, showed internal vacuolar fluorescence. The mutations responsible were mapped to the TMD and adjacent sequences. DNA sequencing revealed two or three changes in this region, and in each case a negative charge had been introduced into the TMD (Figure 3A). Not all of the mutants had an equally strong phenotype: when grown exponentially in liquid culture, two of them had a much reduced phenotype but one, with an acidic residue at position 5 of the TMD, continued to show internal vacuolar fluorescence (Figure 3B). Thus, an internalization signal can be generated by specific changes to the TMD.

View larger version (43K): [in this window] [in a new window]   Figure 3.   Mutants of PSSO that reach the interior of vacuoles. PCR-generated mutants were transformed into yeast, and cells from individual colonies were screened by fluorescence microscopy. (A) Sequence changes in the mutants. (B) Confocal images of cells grown to exponential phase in liquid culture (only vacuoles are visible). Note that under these conditions, only mutant p2.1-26 retains a strong internalization phenotype.

In the light of these results, we systematically introduced aspartic acid residues at alternative positions throughout the Sso1p TMD, which was appended to the Pep12p cytoplasmic domain. Figure 4 shows that single Asp residues were indeed sufficient to cause internalization of the chimera, but their position was crucial. At positions 3 and 5 of the TMD (counting from the cytoplasmic side), the effect was very obvious, but more C-terminal positions had a much weaker effect. This was confirmed by profiles of fluorescence intensity across single vacuoles (lower panel in Figure 4). Whereas the normal TMD resulted in clear peaks of fluorescence at the edges of the vacuole, with a dark interior, the 3D and 5D mutations resulted in maximum intensity at the center of the vacuole, with no detectable peaks at the edges. Other mutants typically gave a mixture of these two patterns, with membrane staining clearly detectable. Changing the Arg residue at position 1 to an Asp had a modest effect, but little change in distribution resulted from changing the Lys at 3 or from insertion of an Arg residue at position 3 or 5. All of the mutant proteins still passed through late endosomes, as demonstrated by their accumulation in PVCs in vps4 cells; this is illustrated for one of the strongest alleles (PSSO3D) in Figure 4. The small proportion of the protein that reached the vacuole in these cells was not internalized, suggesting that functional endosomes are required for this effect.

View larger version (47K): [in this window] [in a new window]   Figure 4.   Single point mutations cause internalization of PSSO. Mutants are designated by the position of the change (numbering as indicated on the sequence) and the residue at that position (D or R). Confocal images are shown for constructs expressed in wild-type cells (only vacuoles are visible), and in the case of PSSO3D, also in vps4 cells. The bottom panel shows profiles of fluorescence intensity taken across individual representative vacuoles for the Asp mutant series. The peaks visible in the wild-type (wt) profile correspond to the membrane at each side of the vacuole; intensity between these peaks indicates internal material.

The effects of the Asp insertions were not limited to the Pep12-Sso1p chimera, because introduction of an Asp at position 3 or 5 of the normal Pep12p TMD was also sufficient to cause internalization (Figure 5A). Remarkably, however, this did not prevent Pep12p from providing its function, as demonstrated by the restoration of growth at 37°C in a pep12 null strain. The mutants showed activity comparable to that of the wild-type protein (Figure 5B), presumably functioning transiently as they passed through endosomes. The retention of activity suggests that the Asp insertions did not cause major misfolding or aggregation of Pep12p.

View larger version (72K): [in this window] [in a new window]   Figure 5.   Mutations in the Pep12p TMD affect sorting but do not abolish function. (A) Aspartic acid mutations in Pep12p (as indicated) cause internalization. (B) Mutations do not block function. Drops containing serial 10-fold dilutions of cells were grown on agar at 24 or 37°C. The pep12 mutation was complemented by a CEN plasmid expressing wild-type Pep12p from the TPI promoter. Other plasmids contained the PEP12 promoter, driving PEP12 itself, or the mutant forms with Asp residues at the 3 or 5 position (PEP3D, PEP5D).

As a further test for charge-dependent internalization, we sought an alternative approach that does not depend on the presence of GFP. First, we established that the cytoplasmic domains of proteins targeted to the interior of the vacuole do eventually become exposed to vacuolar proteases, even though they should initially be protected by the membranes of the internal vesicles. Immunoblot analysis of a GFP-Snc1p chimera bearing the Nyv1p TMD, which is targeted to the vacuole interior (GSN, Figure 1B), revealed that 60% of it was proteolytically cleaved in wild-type cells (Figure 6A). This cleavage was abolished in cells lacking the vacuolar protease Pep4p, other minor cleavages then being detected. The total amount of GFP was similar in wild-type and pep4 cells, indicating that GFP is substantially resistant to vacuolar proteases and thus is a reasonable marker for the vacuole delivery pathway. As a control, Figure 6A shows that 90% of a GFP-Nyv1p chimera, present on the outer membrane of the vacuoles (Figure 1A), remained intact even in wild-type cells. Thus, Pep4p cleavage can be used to measure the delivery of membrane proteins to the vacuolar lumen.

View larger version (23K): [in this window] [in a new window]   Figure 6.   Proteolytic processing of chimeras in the vacuole. (A) GFP-tagged Snc1p with the Nyv1p TMD (GSN; apparent size 45 kDa) and GFP-Nyv1p (NYV; apparent size 60 kDa) were expressed in wild-type (WT) or pep4 cells and detected by immunoblotting with anti-GFP. The position of free GFP generated by proteolysis of GSN is indicated. (B) pho8 pho13 cells were transformed with vector alone (pRS416) or with plasmids expressing ALP-Pep12p-Sso1p chimeras without (APSSO416) or with (APSSO3D416) an aspartic acid at position 3 of the TMD. ALP activity was assayed in cell lysates.

To avoid GFP altogether, we made use of ALP. ALP is synthesized as a precursor that is proteolytically activated by Pep4p upon arrival at the vacuole (Klionsky and Emr, 1989). The precursor is a membrane protein that travels via the AP-3 pathway, but the activity resides in the lumenal domain, which is devoid of vacuolar targeting signals (Cowles et al., 1997b; Vowels and Payne, 1998). We fused this domain to the Pep12p-Sso1p chimera and found that only low levels of phosphatase activity were produced (Figure 6B). This fits with the predicted location of the ALP sequences on the cytoplasmic side of the vacuolar membrane. Insertion of an Asp residue at position 3 of the TMD caused a substantial increase in phosphatase activity (Figure 6B). This finding confirms that the mutation induced internalization of the construct and its eventual exposure to vacuolar proteases.

We conclude that there is a specific mechanism that responds to the presence of acidic residues in the TMD, particularly the part of the TMD that spans the cytoplasmic leaflet of the membrane bilayer, and that directs the protein into the internal vesicles of MVBs.

Acidic TMDs Are Internalized Only in Endosomes

Having identified a strong sorting signal that functioned in endosomes, we investigated more specifically whether it could be recognized in vacuoles themselves. To do this, we introduced an Asp residue at position 3 of the TMD of a GFP-Vam3p chimera. Vam3p itself normally reaches the vacuole via the AP-3 pathway, as shown by its efficient delivery in a vps4 mutant (Figure 7A). However, when this pathway is disrupted by the apm3 mutation, Vam3p still reaches the vacuole and is found on its surface. Thus, like Pep12p, Vam3p is evidently able to pass through endosomes without being internalized. In apm3 cells, it presumably uses a default pathway to endosomes, as it does when the AP-3 targeting signal is removed from it (Darsow et al., 1998).

View larger version (54K): [in this window] [in a new window]   Figure 7.   Aspartic acid residues cause Vam3p to be internalized only on passage through endosomes and do not affect its function. (A) GFP-tagged Vam3p with (VAM3D) or without (Vam3) an aspartic acid at position 3 of the TMD was expressed in wild-type (wt), vps4, or apm3 cells. Confocal images of the vacuoles are shown. Insets show fluorescence profiles along a horizontal axis through the largest vacuole in each image. (B) vam3 cells, and cells of the same strain expressing either GFP-Vam3p or GFP-VAM3D from a plasmid, were labeled with FM4-64. Note the restoration of vacuole morphology by both proteins. The bottom panel shows immunoblots of extracts of the same cells probed for Vam3p and anti-Vam3p immunoprecipitates of these extracts probed for Vti1p.

When the acidic residue was present in its TMD, the fate of Vam3p was strongly influenced by the route it took. In vps4 cells, it was distributed evenly around the vacuolar membrane, as normal. In apm3 cells, on the other hand, it was almost entirely internal to the vacuole (see the intensity profiles of fluorescence in the optical sections shown in Figure 7A). This indicates that the acidic residue is recognized as a sorting signal only when the protein passes through endosomes. In wild-type cells, there was a mixture of surface and internal fluorescence, suggesting that the chimera traveled partially by the AP-3 pathway and partially via endosomes.

It has been reported that some proteins, notably the SNARE Vti1p, can be transported out of the vacuole to late endosomes; as a consequence, they become trapped in PVCs in class E vps mutants (Bryant et al., 1998; Gary et al., 1998). In the case of Vam3p, the bulk of the protein remained on the vacuolar membrane in vps4 cells, and the same was true for the VAM3D mutant (Figure 7A). We infer that Vam3p does not normally cycle out of the vacuole, and the presence of a charge in its TMD does not induce its transport to PVCs in vps4 cells. It appears that an acidic TMD has little consequence once the protein has reached the vacuolar membrane.

Although polar residues in TMDs could potentially cause inappropriate interactions resulting in aggregation or misfolding, we observed no effect on the functional properties of Vam3p. Thus, the VAM3D mutant was as effective as the wild-type protein at restoring normal vacuolar morphology to vam3 cells (Figure 7B). Vam3p is normally found in a SNARE complex that includes Vti1p (Ungermann et al., 1999), and we found that identical amounts of Vti1p coprecipitated with the mutant and wild-type Vam3p (Figure 7B). The effects of the acidic residue on sorting, therefore, are not due to gross alterations to the physical or functional state of Vam3p.

Acidic TMD Residues Affect Sorting at Multiple Steps

It has been reported that acidic charges within a TMD can cause retention of a membrane protein in the ER due to RER1-dependent retrieval from the Golgi (Letourneur and Cosson, 1998), although we did not observe this effect with the Pep12p chimeras. Furthermore, in studies of Snc1p sorting in the late Golgi, we observed that mutations that introduced acidic residues into the TMD caused missorting directly to the endosomal system (M. Lewis and M. Black, unpublished observations). To examine more thoroughly such additional effects on sorting, we tested the set of Asp insertions in the Sso1p TMD in two further contexts: in Sso1p itself, and appended to Snc1p.

With Sso1p, several distinct effects could be observed, which are illustrated by the examples shown in Figure 8. When the constructs were expressed in an end4 mutant to eliminate the effects of endocytosis, it was clear that the 3D and 5D mutants were partially missorted directly from the Golgi to the endosomal pathway, ending up within the vacuole (Figure 8). Some missorting of the 7D mutant was also observed, but substitution of Asp residues at TMD position 1, 9, 11, 13, 15, 17, or 19, or of Arg residues at position 3 or 5, did not affect transport of Sso1p to the plasma membrane (see Figure 8 for examples).

View larger version (98K): [in this window] [in a new window]   Figure 8.   Effects of TMD mutations on Sso1p location. Confocal images are shown of cells expressing GFP-tagged Sso1p with the indicated representative mutations (D or R, at positions indicated by numbers). The distribution of each construct is shown in wild-type (wt) and end4 cells. Note that 3D and 5D are the only mutant proteins that are not fully delivered to the plasma membrane in end4 cells.

In wild-type cells, an additional effect of the mutations was apparent: the GFP-Sso1p was cleared from the plasma membrane and delivered to the vacuole. It is likely that Sso1p is normally endocytosed at a modest rate (see the internal fluorescence of GFP-Sso1p in Figure 8) and, like Snc1p, retrieved from early endosomes. In the case of Snc1p, retrieval is dependent on the TMD sequence (Lewis et al., 2000). Thus, alterations to the Sso1p TMD might be expected to prevent its retrieval; it is also possible that they can stimulate its endocytosis. Whatever the precise mechanism, removal of Sso1p from the cell surface was clearly a distinct phenomenon from the missorting in the late Golgi and the internalization in late endosomes. Unlike these two phenomena, transfer of Sso1p from the plasma membrane to the vacuole could be induced by Asp residues throughout the TMD and also by Arg residues. The difference is well illustrated by the 15D and 17D mutants (Figure 8), which are endocytosed to the vacuole yet inefficiently internalized. Vacuolar delivery was also observed with the 7D, 9D, 11D, 13D, 19D, 3R, and 5R mutants.

An Snc1p chimera bearing the Sso1p TMD normally passes to the plasma membrane, is endocytosed, and then, because it fails to be recycled to the Golgi, is delivered to the vacuole (Lewis et al., 2000) (SSSO, Figure 9). However, insertion of an Asp residue into the TMD of this chimera at position 3, 5, or 7 had an additional effect: a substantial proportion of the protein was found in the ER (Figure 9). Some did reach the vacuole and did so without passing via the plasma membrane (as judged by expression in end4 cells; our unpublished observations). Asp residues at other positions (9, 11, 13, 15, 17, 19, or 21) did not have this effect on ER retention (see Figure 9 for examples). Those mutants that did accumulate in the ER did so even in a rer1 deletion mutant, indicating that Rer1p was not solely responsible for their retention (Figure 9).

View larger version (63K): [in this window] [in a new window]   Figure 9.   Effects of mutant Sso1p TMD when appended to Snc1p. GFP-tagged constructs consisting of Snc1p with the Sso1p TMD (SSSO), with Asp residues at position 3, 5, 9, or 13 as indicated, were expressed in wild-type (wt) or rer1 mutant cells, and confocal images were obtained. Much of the fluorescence is in vacuoles, but arrowheads indicate nuclear envelope staining with the 3D and 5D mutants (nuclei being identified by DAPI staining of the cells). Fainter fluorescence corresponding to peripheral ER was also visible in these cells.

These results indicate that introduction of an Asp residue into the Sso1p TMD can affect protein sorting at several different sites, depending on the protein to which the TMD is appended. Remarkably, the position of the Asp residue has a similar effect on sorting at the ER/Golgi interface, the late Golgi, and late endosomes, maximal effects being achieved at positions 3 and 5. This suggests that similar physical properties of the TMD are recognized at each of these sites. In contrast, the sorting of Sso1p at the early endosome and/or plasma membrane, which is also affected by charges in the TMD, has a clearly distinct specificity.

Naturally Occurring Polar TMDs Are Sorted to the Vacuolar Interior

Acidic residues are quite rare in TMDs, but other hydrophilic residues are more common. To determine whether natural TMD sequences could also affect the fate of Pep12p, we prepared additional chimeras. First, we tested the TMD of the only protein known naturally to follow the internalization route into vacuoles, carboxypeptidase S (Cps1p). As shown in Figure 10, the Cps1p TMD targeted Pep12p to the vacuole interior. Given the similarities between ER retention and vacuole sorting, we also tested the TMD of Sec12p, which has an Asn residue at position 4 of the TMD. Although this is a substrate for the Rer1p-dependent retrieval system, when appended to Pep12p it did not cause ER retention but rather internalization into the vacuole, regardless of whether Rer1p was present. Similarly, the TMD of the ER syntaxin Ufe1p, which contains an Rer1p-independent ER retention signal, failed to keep Pep12p in the ER but targeted it to the vacuole interior (Figure 10).

View larger version (91K): [in this window] [in a new window]   Figure 10.   TMDs that cause internalization. Images are shown of vacuoles in cells expressing GFP-Pep12p with the indicated TMDs. See Table 2 for details of the sequences and Ufe1 mutations.

An earlier study examined the effects of various mutations on the ability of the Ufe1p TMD to serve as an ER retention signal (Rayner and Pelham, 1997), and we tested these mutations also in the context of Pep12p. With regard to ER retention, replacement of pairs of polar residues in the first half of the TMD with leucines had little effect, but when all of them were altered some of the Pep12p could be found on the vacuole surface (Table 2; Figure 10). Lengthening the TMD from 16 to 20 residues proved a more effective means of saving Pep12p from internalization (Figure 10). In general, the features required for ER retention of Ufe1p and internalization of Pep12p were remarkably similar, although not identical (Table 2).

These experiments show that targeting to the vacuole lumen can readily be achieved by naturally occurring TMDs. The common features of these TMDs are quite difficult to discern, but most have a relatively short length and some hydrophilic residues, especially in their cytoplasmic half. Similar features are evidently recognized by the sorting machinery, which retains proteins in the ER and directs them to the vacuole lumen.

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Internalization of Membrane Proteins in Endosomes

The yeast vacuole has multiple functions, one of which is to degrade proteins. For membrane proteins, this can be achieved by their internalization into late endosomes in small vesicles that are delivered to the vacuole lumen and ultimately destroyed (Burd et al., 1998; Odorizzi et al., 1998). Clearly, proteins that function on the outer vacuolar membrane have to avoid this fate; therefore, internalization must be specific. Selective sequestration into the internal membranes of MVBs is also important in animal cells for inactivation of endocytosed cell surface receptors (reviewed by Sorkin, 1998).

One way for a vacuolar protein to avoid internalization is to bypass the site at which it occurs by following the AP-3-dependent pathway directly from the Golgi to the vacuole. This strategy is followed by Vam3p (Cowles et al., 1997a; Darsow et al., 1998), and we now show that it is also used by Nyv1p. Our data support the view that internalization is restricted to endosomes: if there is direct invagination of vacuolar membrane, it is either a minor pathway under normal growth conditions or uses different sorting criteria from the endosomal event. Specifically, a Vam3p mutant that gets internalized does so only if it is forced to pass through the endosomal system. Furthermore, disruption of late endosomal function in vps4 cells resulted in no detectable internalization of any construct tested.

Some proteins, e.g., Pep12p and Vam3p, can clearly pass through late endosomes and remain on the outer membrane. Our analysis of Pep12p revealed no specific cytoplasmic signal that might be required for this, but clearly the cytoplasmic domains of other SNAREs, such as Snc1p and Nyv1p, have features that cause them to be internalized (although neither would normally pass through late endosomes). We have been unable to identify any short internalization signal in these proteins, and we suspect that a global property, such as lack of tight structure, may be responsible (see below).

Polar Transmembrane Domains as Sorting Signals

Our most important conclusion is that sequences within the TMD can control the sorting of proteins in late endosomes, as they can at other points in the secretory and endocytic pathways. Specifically, a single aspartic acid residue in an otherwise hydrophobic TMD is sufficient to cause internalization, as are several natural TMDs that contain polar residues. These include that of Cps1p, which normally enters vacuoles in this way. This implies that the segregation mechanism is intimately connected to events within the lipid bilayer.

A rational explanation is that there is a quality control system that actively seeks membrane-spanning regions that contain inappropriately exposed residues and targets them for destruction. Polar residues exposed to the lipid bilayer could arise by mutation, misfolding, or incomplete assembly of membrane proteins. Such a system would be analogous to the proposed role of Vps10p as a receptor that can bind misfolded proteins in the Golgi lumen and carry them toward the vacuole (Hong et al., 1996; Li et al., 1999).

This idea is appealing, because our data indicate that there are at least three different sites in the endomembrane system at which similar features of a TMD are used to minimize the possibility of surface expression. The first example involves the well-known tendency for abnormal or unassembled proteins to be retained in the ER, a location where refolding is possible and from which degradation can be initiated. Polar residues within a TMD clearly accentuate retention, albeit in a manner that depends also on the nature of the rest of the protein. The second decision point occurs at the exit from the late Golgi: acidic residues within the TMD are sufficient to inhibit transport to the cell surface and instead direct proteins toward the endocytic pathway. Finally, in late endosomes, these same features dictate internalization and eventual destruction of the membrane protein. Although these mechanisms may have a primary role in protecting cells from damage, some normal cellular proteins evidently exploit them for their own localization, either to the ER (Sec12p, Ufe1p) or to the vacuole (Cps1p).

The similar specificity at each of the sorting points is striking. The strong effect of an acidic rather than a basic residue, within the cytoplasmic rather than the lumenal half of the bilayer, is one example. The common features of the Ufe1p TMD required for ER retention and internalization in endosomes is another. The specificity is not universal to all TMD-dependent sorting events, however; the recycling of Sso1p to the plasma membrane is also disrupted by charged residues within the TMD, but this is restricted neither to charges at the cytoplasmic end nor to negative residues. This sorting event is also different in nature: it retrieves proteins for reuse rather than targeting them for destruction.

The specificity suggests that polar TMDs are recognized by a protein receptor. Some other simple possibilities can be ruled out. For example, the TMD is not simply shortened by the insertion of a charge; if this were the only requirement, one would expect that both basic and acidic residues would suffice and that they could be placed near either end. Indeed, the Sso1p TMD is significantly longer than the Pep12p TMD, and introduction of an Asp at position 3 leaves a hydrophobic stretch of 19 residues, longer than the 18-residue Pep12p TMD. It seems more likely that the key feature is an acidic residue buried in the membrane, one to two helical turns from a positively charged amino acid that, by interacting with the acidic cytoplasmic face of the lipid bilayer, defines the start of the TMD. A single acidic residue, of course, is only one form of signal. The Sec12p TMD has an Asn at position 4, and both the Ufe1p and Cps1p TMDs have more dispersed polar features.

A likely reason for the cell to destroy proteins with abnormal TMDs is to prevent inappropriate aggregation or misfolding, and this raises the question of whether it is such gross aberrations that are recognized rather than specific TMD features. This seems unlikely in our experiments for several reasons. First, the TMDs of proteins such as Cps1p, Sec12p, and Ufe1p occur naturally in the cell, and these proteins have not been reported to form aggregates even when overexpressed. Second, there is no reason why such effects would have the observed specificity with regard to the location of acidic residues. Third, Vam3p containing an Asp in its TMD shows no sign of aggregation on the vacuole membrane when delivered there by the AP-3 pathway, is functional, and forms a complex containing Vti1p with an efficiency indistinguishable from that of the wild-type protein. Similarly, Pep12p mutants with acidic TMDs retain function, and the Sso1p mutants, which partially reach the plasma membrane, also show no sign of large-scale aggregation. It is difficult to see how more subtle effects, such as partial or transient dimerization, could lead to such efficient recognition by the internalization machinery. It seems more likely, and more of an advantage, that the cell recognizes a potentially disruptive TMD directly rather than the damage induced by it.

The effects of polar TMDs, however, are not independent of the proteins to which they are attached. Evidently, other properties of these proteins contribute to their overall fate, particularly in determining exit from the ER. Interestingly, the v-SNARE Snc1p seemed particularly prone to ER retention, and we observed a similar phenomenon when charges were introduced into the TMD of its relative Nyv1p (our unpublished observations). Previous work has also shown that the Golgi v-SNARE Sft1p is easily relocated to the ER by the Ufe1p TMD (Rayner and Pelham, 1997). In contrast, the three syntaxins we tested, Sso1p, Pep12p, and Vam3p, showed no sign of ER retention, even when acidic residues were introduced into their TMDs. The v-SNAREs were also subject to destruction in the vacuole regardless of the TMD, whereas the syntaxins were not. This may reflect in part the fact that Snc1p and its relatives have very little structure when not part of a SNARE complex and thus are effectively unfolded proteins, whereas syntaxins can assume a stable four-helix structure (Hazzard et al., 1999; Misura et al., 2000). Unfolded proteins are likely to be subject to additional quality control mechanisms at various points in the cell. Thus, the fate of any individual protein will be determined by a complex interplay of chaperones and sorting receptors, sometimes acting in concert and sometimes having opposite effects.

How might the mutant TMDs be sorted? By analogy with the ubiquitin system, the proteins might be marked for destruction by some modification, although none has been detected to date. Alternatively, they may be sorted directly by a protein receptor. If so, our data would suggest that related receptors patrol the bilayer at diverse locations. Alternatively, a single protein could act at more than one point, e.g., a "suicide" receptor might recognize a charged TMD in the Golgi and travel with it to endosomes and then into the vacuole. Such a receptor might itself segregate according to bilayer composition, not only in the Golgi but perhaps also in late endosomes. The internal membranes of MVBs in animal cells are reported to have a unique composition, being enriched in the unusual lipid lyso-bis-phosphatidic acid (Kobayashi et al., 1998). In yeast, the formation of such structures requires Fab1p, a kinase that converts phosphatidylinositol 3-phosphate into phosphatidylinositol 3,5-bisphosphate, although whether the product is sorted selectively to internal membranes is not yet known (Odorizzi et al., 1998).

The only receptor that has been implicated in the sorting of polar TMDs is Rer1p (Sato et al. 1996, 1997; Letourneur and Cosson, 1998). Although it is possible that Rer1p contributes to the sorting of the constructs we have used, it cannot be the sole receptor because it is not essential for internalization in endosomes, or for the retention of Snc1p TMD mutants in the ER, or for the TMD-dependent retention of Ufe1p (Lewis and Pelham, 1996). Although there are no close relatives of Rer1p in yeast, many membrane proteins of unknown function share its tetraspanning topology. It is possible that one or more of them shares its properties.