Intramolecular protein-protein and protein-lipid interactions control the conformation and subcellular targeting of neuronal Ykt6

Soluble N-ethylmaleimide-sensitive-factor attachment protein receptors (SNAREs) are central components of the intracellular membrane fusion machinery (for reviews, see Hay, 2001; Ungar and Hughson, 2003). Many mammalian SNAREs have yeast homologs with similar functions but others have evolved more specialized roles in specific tissues with specific membrane trafficking needs. For example, syntaxin 1A, SNAP-25 and VAMP2 are SNAREs involved in synaptic-vesicle exocytosis that lack direct counterparts in yeast. It is not yet clear, however, what the specialized functions of certain tissue-specific SNAREs are. For example, we do not know why muscle cells need VAMP5 on their plasma membranes (Zeng et al., 1998), what the function of syntaxin 17 is in the smooth endoplasmic reticulum (ER) of steroidogenic cells (Steegmaier et al., 2000) or what the apparently specialized role of syntaxin 11 is in the immune system (Prekeris et al., 2000). In addition, some SNAREs [e.g. syntaxin 16 (Dulubova et al., 2002)] are selectively highly enriched in brain but are probably not involved in regulated exocytosis.

Mammalian Ykt6 is another brain SNARE likely to have unexpected function(s). Yeast Ykt6p appears to be involved in multiple transport pathways between the Golgi and vacuole (Dilcher et al., 2001; Tsui and Banfield, 2000), and is required for homotypic vacuole fusion (Ungermann et al., 1999). Hence, although yeast YKT6 is an essential gene, it does not appear to have a single function in yeast. In fact, it appears to be an example of a multifunctional member of the R-SNARE family (Fasshaur et al., 1998), perhaps partially overlapping in function with other R-SNAREs such as Sec22p and Nyv1p. In support of this idea, Ykt6p expression was specifically upregulated in Sec22p-lacking strains and appeared to compensate partially for the Sec22p deletion by participating in ER-to-Golgi SNARE complexes normally containing Sec22p (Liu and Barlowe, 2002).

Mammalian Ykt6 might have even more diverse roles, because it has several novel and unexpected features. For example, rat Ykt6 is highly enriched in brain (Hasegawa et al., 2003). This finding was surprising, because a SNARE involved in constitutive biosynthetic transport to the lysosome (as yeast Ykt6p is) would probably be very widely expressed. Furthermore, Ykt6 was localized to punctate vesicular structures spread throughout the cytoplasm that did not colocalize with established markers of the endomembrane system. These findings implied that mammalian Ykt6 might function in a specialized transport pathway in neurons instead of or in addition to a constitutive role in transport to the lysosome. Biochemical experiments indicated that particulate Ykt6 in brain and neuroendocrine cells was not extractable with salt or extremes of pH, but was fully extracted with nonionic detergent. Likewise, particulate Ykt6 had a buoyant density similar to lysosomal membranes. These features of particulate Ykt6 suggested that the unique punctate staining pattern probably represented a membrane compartment (Hasegawa et al., 2003). Intriguingly, localization of mammalian Ykt6 to the particulate structures did not require prenylation. Instead, the N-terminal profilin-like domain of Ykt6 was necessary and sufficient for normal localization (Hasegawa et al., 2003).

Another striking feature of Ykt6 was its existence in both particulate and freely soluble pools (Hasegawa et al., 2003). Both yeast and mammalian Ykt6 lack transmembrane domains but are prenylated at their C-termini (McNew et al., 1997). Interestingly, the cytosolic pool of Ykt6 appears to be fully prenylated in both species (Hasegawa et al., 2003; McNew et al., 1997). This feature implied that Ykt6 must either have an escort protein to sequester the lipid group(s), or else must provide its own intramolecular lipid escort domain. One possibility is that the Ykt6 profilin-like N-terminal domain could provide the lipid escort function, in addition to mediating the targeting to the Ykt6 membrane structures. This domain might be involved in other regulatory features as well. Cytosolic Ykt6 was found to be refractory to promiscuous SNARE interactions, unlike bacterially expressed recombinant Ykt6, suggesting the existence of an inhibitory regulatory mechanism in the cytosol. Because Ykt6 behaved as a monomer in solution, an intrinsic autoinhibitory mechanism seemed more likely than regulation by other proteins. Consistent with an intrinsic autoinhibited state, cytosolic Ykt6 was present in a more compact, protease resistant conformation than the more promiscuous recombinant Ykt6. Because SNARE N-terminal domains are established, in the case of syntaxins, to fold back and pack against the SNARE motif and inhibit SNARE complex formation (Munson et al., 2000), one possibility was that the Ykt6 profilin-like domain did something similar in the case of cytosolic, but not recombinant Ykt6. In support of an autoregulatory role for this domain, the yeast Ykt6p N-terminal domain was found to engage in protein-protein interactions with the SNARE motif that caused a modest inhibition in the kinetics of Ykt6p SNARE complex assembly (Tochio et al., 2001).

The profilin-like domain of Ykt6 has been recognized to be part of a larger family of domains with virtually identical protein folds despite little or no sequence similarity. These domains have been termed `longin' domains, because they account for certain R-SNAREs being longer than others (Filippini et al., 2001). In mammals, three R-SNAREs contain longin domains: Ykt6, Sec22b and VAMP7. The atomic structure of the yeast Ykt6p (Tochio et al., 2001) and mammalian Sec22b (Gonzalez et al., 2001) longin domains have been solved and are virtually superimposable; the VAMP7 longin domain is assumed to have a similar profilin-like fold and has also been implicated in protein localization (Martinez-Arca et al., 2003). Interestingly, similar domains appear to reside on two non-SNARE isoforms of Sec22, called rSec22a (Hay et al., 1996) and hSec22c (Tang et al., 1998). Likewise, another non-SNARE protein, sedlin, whose structure was recently determined to be superimposable with the SNARE longin domains, appears to consist entirely of a longin domain and nothing else (Jang et al., 2002). Although obviously not a SNARE, sedlin is probably involved in ER-to-Golgi transport, because it is the mammalian homolog of Trs20p, a subunit of the TRAPP ER-Golgi tethering complex (Sacher and Ferro-Novick, 2001). Point mutations in human sedlin cause the disease spondyloepiphyseal dysplasia tarda (Gedeon et al., 1999). Longin domains appear to be versatile structures that might regulate SNARE activity as well as perform other essential roles in transport reactions.

We have taken advantage of the unique subcellular localization of Ykt6 and the known structure of the Ykt6p longin domain to define the structural features that control Ykt6 targeting. We discovered that the Ykt6 longin domain controls Ykt6 localization not only by direct targeting to the Ykt6 subcellular particle but also by acting as a chaperone for the Ykt6 SNARE motif and lipid group(s). In so doing, the Ykt6 longin domain simultaneously keeps the fully lipidated SNARE soluble, masks spurious targeting signals that would otherwise cause abnormal localization and directs the molecule to its specialized subcellular location.

All the cDNA sequences used in this study have been reported previously: rat Ykt6 (GenBank accession number AA956066); yeast Ykt6p (GenBank accession number NC001143); human sedlin (SEDL) (GenBank accession number BC008889); mouse VAMP7 (GenBank accession number BC003764); rat Sec22a (GenBank accession number NM057147); mouse Sec22b (GenBank accession number U91538); rat GOS-28 (GenBank accession number U49099). Various point mutants used in the targeting studies were generated by polymerase chain reaction (PCR) based site-directed mutagenesis using cDNAs that encode yeast Ykt6p longin domain (residues 1-139), full-length yeast Ykt6p, rat Ykt6 longin domain (residues 1-137) or full-length rat Ykt6 as templates. Truncation mutants for rSec22a, mSec22b and mVAMP7 were created by introducing a stop codon after residues 157, 134 and 125, respectively. All the chimeric proteins used in this study were produced by gene splicing by overlap extension (Horton et al., 1989). For the longin-domain-replacement study, cDNAs that correspond to residues 1-157 of rSec22a, residues 1-134 of mSec22b, residues 1-125 of mVAMP7 or residues 1-140 of hSEDL were directly spliced to cDNA that encoded residues 137-198 of rYkt6. For the transmembrane-domain-replacement study, cDNA corresponding to residues 1-193 of rYkt6 was directly joined to cDNAs corresponding to residues 196-215 of mSec22b, residues 187-220 of mVAMP7 or residues 231-250 of GOS-28. Rat Ykt6 SNARE motifs used in the binding assay were made by PCR amplification of cDNAs corresponding to residues 137-198 for lipidated SNARE motif and residues 137-193 for non-lipidated SNARE motif. For N-terminal Myc-tagging, all the DNA sequences described above, except for SEDL-related sequences, were cloned into pCMV-Myc (Hay et al., 1997) using SacII/XbaI sites. Full-length hSEDL was cloned into pCMV-Tag3B (Stratagene) using BamHI/XhoI sites and the chimeric cDNA encoding hSEDL(1-140)/rYkt6(137-198) was cloned into BamHI/EcoRV sites of pCMV-Tag3B after 3′-end blunting with XbaI. Expression plasmids coding hemagglutinin (HA) tagged yeast full-length Ykt6p and yeast Ykt6p longin domain (residues 1-139) were created by inserting corresponding PCR products into SacII/EcoRV-digested pCruzHA (Santa Cruz Biotechnology). For each construct, the sequence was verified at the University of Michigan Sequencing Core Facility by reading both strands.

PC12 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 4.5 g/l glucose, 5% equine serum and 5% iron-supplemented calf serum (Hyclone, Logan, UT) without antibiotics in a humidified 5% CO₂ incubator at 37°C. PC12 cells plated on poly-lysine-coated glass coverslips in standard six-well culture dishes were transfected using Lipofectamine 2000 (Invitrogen). Except where methanol fixation was used, transfected cells were fixed in 0.1 M sodium phosphate, pH 7.2, 2% paraformaldehyde and 0.05% Triton X-100, for 30 minutes at room temperature 24-36 hours after transfection. Fixed cells were then quenched twice in PBS containing 0.1 M glycine for 10 minutes each and incubated in permeabilization buffer [0.4% saponin, 1% bovine serum albumin (BSA), 2% goat serum in PBS] for 15 minutes at room temperature. This was followed by incubation in primary antibodies for 1 hour in permeabilization buffer. After four consecutive washes (5 minutes each) in permeabilization buffer, cells were incubated with secondary antibodies in permeabilization buffer for another 1 hour. Primary antibodies used in this study were chicken anti-rat Ykt6 (Hasegawa et al., 2003), mouse anti-Myc monoclonal antibody (hybridoma 9E10), rabbit polyclonal anti-Myc (Covance), mouse anti-HA monoclonal antibody (hybridoma 16B12), rabbit anti-GM130 (gift from M. Lowe, University of Manchester, UK), mouse monoclonal anti-syntaxin 1 (hybridoma HPC-1) and anti-syntaxin 5 (Hay et al., 1997). FITC- and Texas-Red-conjugated secondary antibodies were purchased from Jackson ImmunoResearch. After the secondary antibody labeling, cells were washed in permeabilization buffer four more times and coverslips were mounted using Vectashield mounting medium (Vector Laboratories), then sealed with nail polish. LysoTracker™ Red DND-99 (Molecular Probes) was used to label acidic compartments of PC12 cells. To inhibit in vivo palmitoylation, PC12 cells were incubated in 150 μM 2-bromopalmitate (Aldrich), starting 5 hours after transfection until cells were fixed at 30 hours after transfection. These conditions are similar to those used in previous studies (Webb et al., 2000).

Refer to Hasegawa et al. (Hasegawa et al., 2003) for detailed description of epifluorescence microscopic image capturing and image deconvolution. Quantification of colocalization in immunofluorescence images was carried out using ImageJ software (US National Institutes of Health) and calculated as the number of doubly labeled pixels (positive for both Myc and endogenous Ykt6) divided by the total number of Myc-positive pixels.

Wild-type rat Ykt6 longin domain (residues 1-137) was PCR amplified and cloned into pGEX-KG (Guan and Dixon, 1991) using XbaI/SacI restriction sites to make an N-terminal glutathione-S-transferase (GST) fusion protein. Using this wild-type rat Ykt6 longin domain as template, mutant Ykt6 longin domains (V8D, F39E/F42E, R50E/R56E or V59E) were created by using the same PCR-based site-directed mutagenesis used above. The sequence of all the mutants in pGEX-KG plasmid were verified at the University of Michigan Sequencing Core Facility by reading both strands. Expression plasmids coding GST-rYkt6 longin wild-type and mutants were transformed into Escherichia coli strain JM109. Transformants were cultured in LB medium at 37°C. When the optical density at 600 nm of the E. coli culture reached 0.5, cultures were shifted to 15°C and protein production was induced by adding 0.1 mM IPTG for 6 hours. Bacterial cultures was centrifuged at 6000 g and the pellet was resuspended in French press buffer (50 mM Tris-HCl, pH 8.0, 0.1 M NaCl, 1 mM EDTA, 0.05% Tween 20) supplemented with protease inhibitor cocktail (Roche). After cell lysis by French press, the lysate was first centrifuged at 20,000 g for 30 minutes and then at 100,000 g for 1 hour. The final supernatant was stored in aliquot at –80°C until used in the binding assay.

For bead preparation, bacterial lysates containing wild-type or mutant GST-rYkt6 longin domain were mixed with 10 μl bed volume of glutathione/Sepharose-4B (Amersham Biosciences) in 0.5 ml microcentrifuge tubes to immobilize GST fusion proteins. After an overnight incubation at 4°C with constant rotation, the beads were washed five times in binding buffer (20 mM Hepes, pH 7.2, 150 mM KCl, 2 mM EDTA, 5% glycerol, 0.1% BSA, 0.1% Triton X-100). Bound proteins were then analysed by sodium-dodecyl-sulfate polyacrylamide-gel electrophoresis (SDS-PAGE) and Coomassie-blue staining. Equivalent loading of six different proteins was achieved by adjusting the input lysate.

To prepare the lipidated rYkt6 SNARE motif and non-lipidated motif, confluent PC12 cell cultures in 10 cm dishes were transfected with Myc-rYkt6 (residues 137-198) and Myc-rYkt6 (residues 137-193), respectively. At 24 hours after transfection, the cells were detached from the plates using Hanks' balanced salt solution containing 1 mM EDTA and collected by centrifugation. The cell pellets were resuspended in lysis buffer [20 mM Hepes, pH 7.2, 250 mM KCl, 2 mM EDTA, 5% glycerol, 0.3% Triton X-100 (v/v), protease inhibitor cocktail]. After a 30 minute incubation on ice, lysates were centrifuged at 100,000 g for 1 hour and supernatant was used as the ligand-containing fraction for the binding assay. To determine the relative concentrations of the two different Myc constructs, both lysates were subjected to SDS-PAGE (3-20% gradient gel) and analysed by western blotting; the lipidated and non-lipidated recombinant SNARE motifs were detected by anti-Myc (9E10) and anti-rat-Ykt6 antibodies as a band of the expected size (∼8 kDa). For the bimolecular binding assay, a 10 μl bed volume of protein-loaded glutathione/Sepharose-4B was mixed with 50-100 μl PC12 cell lysate and binding buffer in a total assay volume of 150 μl. The assay mixture was incubated at 4°C for 1 hour with constant rotation. The beads were washed five times, then the beads were boiled in 40 μl SDS sample buffer. For SDS-PAGE, 20 μl of each sample was loaded on a 3-20% gradient gel. The proteins were electrotransferred to nitrocellulose membranes, followed by immunoblotting with anti-Ykt6 antisera.

ER/Golgi SNAREs (syntaxin 5, membrin and rBet1) were expressed in bacteria as GST fusion proteins, and cleaved and purified using our standard methods (Xu et al., 2000). Myc-rYkt6 (residues 137-193) and Myc-rYkt6 (residues 137-198) were prepared as above. For details of the SNARE assembly assay, refer to Hasegawa et al. (Hasegawa et al., 2003). Briefly, PC12 extract containing Myc-rYkt6 137-193 was incubated overnight on ice in the presence or absence of 2 μM each of the ER/Golgi SNAREs in a total volume of 300 μl of binding buffer The incubations were then gel filtered on a 24 ml Superdex-200 column, and the gel filtration fractions analysed by immunoblotting using the anti-Ykt6 antibody.

PC12 cells were cultured in growth medium with or without 150 μM 2-bromopalmitate for 24 hours under normal culture conditions. Cells were then harvested in Hank's/EDTA buffer and collected by centrifugation. Cell pellets were resuspended and homogenized in homo buffer (20 mM HEPES, pH 7.0, 0.25 M sucrose, 2 mM EGTA, 2 mM EDTA) supplemented with protease inhibitor cocktail (Roche, Complete), followed by centrifugation for 10 minutes at 1000 g to obtain a post-nuclear supernatant (PNS). A portion of the PNS fraction was reserved for analysis and the remainder was centrifuged at 100,000 g for 60 minutes to separate the membrane and soluble fractions. All fractions were subjected to acetone precipitation and the precipitated pellets were boiled in SDS sample buffer. Equivalent amounts of each fraction were then subjected to SDS-PAGE and immunoblotting with anti-GAP-43 antibody (hybridoma 9-1E12).

Primary hippocampal cultures containing both neurons and glial cells were prepared from the CA1-CA3 regions (excluding the dentate gyrus) of 1-4-day-old rats (Wistar; Taconic M&B, Ry, Denmark). The hippocampi were dissected out and incubated in 1% trypsin (Sigma, T1005) for 4-5 minutes at room temperature before they were dissociated by triturating with silicone-coated Pasteur pipettes in the presence of 0.05% DNase (Sigma, D5025). The cells were counted and plated on Matrigel (Becton Dickinson, USA) coated small coverslips (Assistent, Germany) at a density of ∼92,000 cm^–2. The cells were allowed to attach by incubating them in a single drop for 15 minutes before more cell medium was added [Gibco's modified Eagle's medium (MEM) with the addition of 0.3 mg ml^–1 glutamine, 2.5 μg ml^–1 insulin (Sigma, I-5500), 5-10% fetal calf serum (FCS), 2 μl ml^–1 B-27 (Gibco) and 0.02-0.10 μl ml^–1 ARA-C (Sigma, C6645)]. The cultures were maintained in a 5% CO₂, 95% air incubator at 37°C. Their medium was changed at 1 day, 4 days, 8 days and 12 days in vitro (DIV). The medium at 1 DIV contained ARA-C to inhibit cellular proliferation; 10% FCS was used at day 0 but, later, the added medium contained 5% FCS. The cultures were used for experiments at around 15 DIV.

Cultures at 15 DIV were fixed in 2% paraformaldehyde by the following procedure. Freshly prepared fixative [2% paraformaldehyde in 0.1 M sodium phosphate buffer (NaPi) pH 7.4] was heated to 37°C before adding it to the culture medium (equal volumes). After 30 minutes, this mixture was substituted with fixative alone. After 1 hour, the fixative was substituted by 0.1 M NaPi containing 0.2% paraformaldehyde. The cells were quenched twice with 0.1 M glycine in 0.1 M NaPi for 10 minutes, and subsequently permeabilized for 15 minutes in PS (0.4% saponin, 1% BSA and 2% normal sheep serum in 0.1 M NaPi). The cultures were incubated in primary antibody in PS for 1 hour, washed three times in PS alone before incubation in secondary antibody in PS for 30 minutes, followed by one wash in PS and three washes in 0.1 M NaPi. The coverslips were then mounted with glycerol and subjected to investigation with a Leica TCS NT confocal laser-scanning microscope. The chicken anti-Ykt6 antibody was used at 1:10 and the anti-Sec6 antibody (9H5, hybridoma supernatant) was undiluted. Secondary antibodies were Rhodamine-Red™-X-conjugated anti-chicken and Cy2-conjugated anti-mouse (Jackson ImmunoResearch, Baltimore, USA) used at 1:100.

A previous report demonstrated that sequence-specific localization determinants reside within the mammalian Ykt6 longin domain and target the SNARE to an unidentified, apparently membranous structure in PC12 pheochromacytoma cells (Hasegawa et al., 2003). Because the yeast Ykt6p longin domain can precisely target to the same vesicular structures in PC12 cells, it seemed likely that conserved surface features of the Ykt6 longin domain were responsible for targeting. We examined the solved nuclear magnetic resonance (NMR) structure (Tochio et al., 2001) of the yeast Ykt6p longin domain for surface residues that were conserved between yeast and mammals. Fig. 1 shows a space-filling model of the yeast Ykt6p longin domain, highlighting the 31 conserved surface residues, which are labeled and color coded as hydrophobic (yellow), polar (green), acidic (red) or basic (blue). Mutations in 28 of these residues were introduced, singly or in pairs, into the isolated, Myc-tagged yeast longin domain and transfected into PC12 cells. As shown in Fig. 2 and summarized in Table 1, none of the surface mutations affected Ykt6p-longin-domain targeting to the cytoplasmic Ykt6 puncta. Owing to space limitations, only six mutants are shown in Fig. 2. Because the yeast longin domain does not cross-react with our chicken anti-rat-Ykt6 antibody (Hasegawa et al., 2003), we were able to detect both the Myc-tagged longin domains and the endogenous rat Ykt6 to confirm that, in each case, the longin mutants displayed a similar localization to rat Ykt6. Several representative colocalizations are shown in Fig. 2. The longin mutants often colocalized with endogenous Ykt6, producing a total staining overlap of 12-18% (Fig. 2). We do not know why many puncta were positive for the longin constructs and negative for endogenous Ykt6, and vice versa. One possibility is inefficient labeling with both antibodies caused by epitope masking on the Ykt6 structures. The mutant longin constructs colocalized with endogenous Ykt6 to the same degree as full-length yeast Ykt6p constructs (Hasegawa et al., 2003) [supplementary Fig. S1B]. This experiment indicated that conserved sequence-specific surface features were probably not responsible for the ability of yeast and rat Ykt6 to target to the same vesicular structures.

It should be realized that the localization of tagged Ykt6p constructs and the degree of colocalization between these constructs and endogenous Ykt6 were independent of the tag used or the fixation method. Supplementary Fig. S1A demonstrates that both HA/yeast-Ykt6p and HA/yeast-Ykt6p-longin-domain displayed 17-18% colocalization with endogenous Ykt6, and Fig. S1B demonstrates that Myc/yeast-Ykt6p displayed 24% colocalization with endogenous Ykt6 in methanol-fixed PC12 cells.

Because mutation of surface-exposed residues did not disrupt normal targeting, we wondered whether the overall tertiary structure of the longin domain was responsible for the targeting of the isolated longin domain to Ykt6 puncta. The longin domain is constructed of a profilin-like protein fold that is virtually superimposable between different proteins having the domain, despite limited or no sequence similarity (Dietrich et al., 2003). We expressed the isolated Myc-tagged longin domains from the SNAREs VAMP7 and mSec22b, the non-SNARE Sec22 homolog rSec22a, and sedlin [a protein composed of only a longin domain and that has been implicated in vesicle tethering reactions and the genetic disease spondyloepiphyseal dysplasia tarda (Jang et al., 2002)]. As shown in Fig. 3A, all of these Myc-tagged longin domains targeted the cytoplasmic puncta in PC12 cells and overlapped with endogenous Ykt6 to a similar degree as bona fide Ykt6p constructs (Fig. 3). This implies that it was actually the profilin-like structural scaffold rather than sequence-specific surface features of Ykt6 that are responsible for targeting the isolated domain. It is important to realize that VAMP7 and Sec22b isoforms do not normally colocalize with Ykt6 in PC12 cells (not shown), presumably because the full-length proteins possess other targeting determinants. How can the experiment in Fig. 3A, implicating the overall longin structure and not Ykt6-specific features, be reconciled with the previous observation that normal Ykt6 targeting required sequence-specific features of the Ykt6 longin domain (Hasegawa et al., 2003)? As shown in Fig. 3B, when chimeras containing each of the above longin domains spliced onto the SNARE motif and C-terminus of Ykt6 were expressed in PC12 cells, all of them displayed generalized mislocalization to the Golgi area and plasma membrane. Thus, although any longin domain on its own can target to the Ykt6 punctate structures, only the Ykt6 longin domain is sufficient to target the whole Ykt6 molecule to these cytoplasmic structures. This suggested that the Ykt6 longin domain might have sequence-specific surface features that suppress otherwise-dominant `mistargeting' determinants present on the SNARE motif and/or prenylation motif.

To examine the possibility of longin-domain surface features that suppress otherwise-dominant targeting determinants on the Ykt6 SNARE motif and prenylation site, we introduced the same panel of surface mutations examined above into full-length Myc-tagged yeast Ykt6p and expressed them in PC12 cells. As shown in Fig. 4 and summarized in Table 1, several surface mutations resulted in striking mislocalization phenotypes. In all cases, double-labeling indicated that the localization of endogenous Ykt6 was unaffected. The surface residues that affected Ykt6 localization fell into three groups that might define distinct surfaces important in the control of Ykt6 targeting. The first surface is composed of a conserved hydrophobic patch including F39, F42, V36 and the flanking residue Q38. Of these residues, F39 and F42 had been previously shown to participate in intramolecular interactions between the Ykt6p longin domain and the SNARE motif. These intramolecular interactions appeared to exert a weak inhibitory effect on SNARE-SNARE interactions using purified, nonprenylated, recombinant SNAREs (Tochio et al., 2001). A second apparent surface was composed of a mixture of charged residues and exposed hydrophobic residues, including R50, R56 and I59, on the same face of the domain as the above hydrophobic patch. These residues have not been previously implicated in intramolecular interactions. A third surface consisted of a deep, conserved hydrophobic crevice on the opposite face of the longin domain and included V8 and Y101, and the flanking charged residue E100. Again, these residues were not observed to interact with the SNARE motif in the previous NMR studies with purified, nonprenylated Ykt6p. All of the aforementioned mutations on the first and second surfaces resulted in mislocalizations that included Golgi area as well as apparent plasma membrane staining, with a few cytoplasmic vesicles that did not discernibly colocalize with endogenous Ykt6 (Fig. 4, F39E, F42E, I59E, R50E/R56E). Interestingly, mutations in the hydrophobic crevice (surface 3) resulted in predominantly plasma membrane staining and overlapped by >45% with syntaxin-1 staining (Fig. 4, V8N, E100K/Y101N). None of the surface mutations examined in Fig. 4 are likely to have caused major disruptions of the tertiary structure of the longin domain because, when present in the isolated longin domain, they did not detectably affect localization or stability of the transfected proteins (Fig. 2). The results of Fig. 4 indicate that intramolecular interactions might be important for control of the normal localization of Ykt6 in PC12 cells.

All of the preceding work was done using yeast Ykt6p and the yeast Ykt6p longin domain. To determine whether similar intramolecular interactions control the localization of the mammalian Ykt6, we repeated most of the analysis using the bona fide mammalian protein that is enriched in brain neurons and PC12 cells. As shown in Fig. 5 and Table 2, all three of the surfaces identified in the yeast protein similarly affected the localization of the mammalian Ykt6, although there were a few residues in mammalian Ykt6 that had less severe mislocalization phenotypes. Similar to what we found for the yeast protein, none of the localization-disrupting mutations had any effect on the localization or expression level of the isolated longin domain, implying that it was probably surface features and intramolecular interactions rather than overall structural disruptions that resulted in the mislocalization phenotypes with the full-length constructs.

We wondered which structural features of Ykt6 were responsible for the abnormal targeting of the protein observed in the various mutants described above. As shown in Fig. 6 and summarized in Tables 1 and 2, deletion of the C-terminal CCAIM sequence (CCIIM in yeast), which is the prenylation motif, rescued the abnormal localization phenotypes of all yeast and mammalian mislocalization mutants, restoring cytoplasmic vesicular punctate staining that displayed normal amounts of overlap with endogenous Ykt6. For reasons of space, Fig. 6 shows only a handful of examples. Formally, hydrophobic modifications at the C-terminus of Ykt6 present dominant mistargeting information that is normally masked by sequence-specific surface features of the longin domain.

We wondered whether any of the longin domain surfaces were specifically involved in suppressing otherwise-dominant determinants associated with one or the other of the pair of cysteines at the Ykt6 C-terminus. The precise structure of the fully modified Ykt6 terminus is not known in yeast or mammals. The sequence would predict farnesylation (a C15 isoprenoid), although geranylgeranylation is also a possibility. Probably, the second cysteine (C195) is linked to a farnesyl lipid. The upstream cysteine (C194) could be unmodified or be farnesylated or palmitoylated. Other farnesylated proteins with a double cysteine motif (e.g. Ras) are palmitoylated on the upstream cysteine (Hancock et al., 1989). In these cases, palmitoylation of the first cysteine requires prior farnesylation of the second cysteine. Hence, based upon Ras, it would not be feasible to delete the second cysteine to create a palmitoylated but not farnesylated mutant. However, it would be possible, by mutation of the upstream cysteine, to create a farnesylated protein that does not get palmitoylated. As shown in Fig. 6B and Table 2, mutation of C195 to an alanine restored the collection of mammalian mislocalization mutants to normal distribution, just like deletion of the entire motif had done, consistent with a lack of modification of either cysteine. However, mutation of C194 to alanine had various effects on the mislocalized constructs. For mutations of the large hydrophobic patch (surface 1, including F39 and F42), which had previously been implicated in protein-protein interactions with the SNARE motif, and the mixed hydrophobic/charged surface on the same face (surface 2, including V59), the substitution restored some normal, spotty cytoplasmic staining, but there remained significant abnormal staining in the secretory pathway and plasma membrane areas. This is consistent with the hypothesis that Ykt6 is indeed modified on both cysteines and with these two longin-domain surfaces being partially responsible for suppression of targeting information from the upstream cysteine. However, targeting of the V8D mislocalization mutant was seemingly completely restored to normal Ykt6 staining by substitution of C194 only. This implies that this hydrophobic crevice might be specifically involved in masking the hydrophobic moiety attached to C194.

Interestingly, unlike other mutants, mislocalization mutants in the V8 crevice, especially of the yeast Yk6p (Fig. 4), were predominantly mislocalized to the plasma membrane only, implying release of only a portion of the targeting determinants masked by the longin domain. Because palmitoylation is known to target proteins specifically to the plasma membrane and palmitoylation is generally found on the upstream cysteine in farnesylated proteins with double cysteines, our results imply that the crevice around V8 might be a binding pocket for a palmitoyl group attached to C194. In support of this hypothesis, as shown in Fig. 7A, incubation of cells overnight with 2-bromopalmitate, a competitive inhibitor of protein palmitoylation, restored at least the normal level of colocalization to two distinct mutations affecting the V8 hydrophobic surface on yeast Ykt6p (V8N, K100E/Y101N) while having no discernible effect on mutants of surface 1 (F42E or F39E) or surface 2 (not shown). The 2-bromopalmitate conditions used did, in fact, abrogate protein palmitoylation, because GAP-43 (a plasma membrane palmitoyl protein present in neurons and PC12 cells) underwent a drastic reduction in membrane association upon the drug treatment (Fig. 7B).

The simplest interpretation of the data in Fig. 6B and Fig. 7 is that Ykt6 is palmitoylated on C194 and the palmitoyl group is normally sequestered in the longin-domain hydrophobic crevice around V8. However, no direct demonstration of Ykt6 palmitoylation has been reported, and other models are possible. For example, the hydrophobic crevice could interact with palmitoyl groups on other proteins that are affected by bromopalmitate treatment. Yeast Ykt6p has recently been demonstrated to participate in a specific protein-palmitoylation event (Dietrich, 2004).

The evidence presented so far suggests that intramolecular protein-protein and protein-lipid interactions might control the targeting of Ykt6 in neuroendocrine cells. One way in which this might be accomplished would be the formation of a compact conformation in which the longin domain doubles back and packs against the SNARE motif, and the lipid moieties protrude back along the SNARE motif to be buried in binding sites on the longin-domain surface. To test whether the localization phenotypes we observed are indeed due to the disruption of such a conformation of Ykt6, we developed a bimolecular binding assay between purified, recombinant rat Ykt6 longin domain on beads and soluble longin-domain-deleted Ykt6 constructs present in detergent extracts of PC12 cells. As shown in Fig. 8A, a specific interaction was observed between the GST/longin-domain and the longin-deleted Ykt6 construct with an intact CCAIM motif. This binding phenomenon was not complicated or affected by the inclusion of the construct in Triton X-100 micelles, because the same result was obtained using octyl-glucopyranoside extracts diluted well below the critical micellar concentration [supplementary Fig. S2]. Strikingly, every mutation we tested that had disrupted the localization of full-length Ykt6, from each of the three surfaces, completely abrogated binding between the respective GST/longin-domain and the soluble longin-deleted Ykt6 construct containing CCAIM. This implies that, as we hypothesized, the three highlighted surface features are indeed important for intramolecular interactions, resulting in a Ykt6 conformation that sequesters spurious targeting signals on the SNARE motif and/or modified C-terminus. Although surface residues in the first of the three highlighted surfaces had been previously implicated in protein-protein interactions between the longin domain and SNARE motif, we wondered whether any of the identified intramolecular interactions were in fact of a protein-lipid nature. Deletion of the CCAIM modification site at the C-terminus of the longin-deleted construct completely eliminated all detectable binding with GST-longin protein (Fig. 8A, lower blot). The Myc-rYkt6 137-193 construct was in fact functional/accessible for protein interactions in the extract, because incubation of the cell extract with purified ER/Golgi SNAREs resulted in quantitative incorporation of the construct into high-molecular-weight SNARE complex(es) detectable by gel filtration (Fig. 8D). In addition, purified rYkt6 137-198 produced in bacteria (and thus not lipidated) failed to interact detectably with GST-longin beads in the same binding assay format (not shown). The results of Fig. 8 imply that, indeed, longin-domain/lipid interactions contribute to the normal targeting of the molecule to cytoplasmic structures, probably by formation of a compact conformation that sequesters the SNARE motif and C-terminal lipids.

If Ykt6 indeed functions as a SNARE on the vesicular structures, one might wonder why this SNARE would rely upon a soluble intermediate for targeting to this site of function. Prenyl proteins are thought to be initially inserted into the endoplasmic reticulum during the modification process (Gelb et al., 1998), so it would be conceivable for Ykt6 to travel to its destination compartment using the endomembrane system after insertion into the ER, like other SNAREs. To test whether the Ykt6 longin domain could provide the targeting information necessary to target the SNARE to the Ykt6 structures through the endomembrane system, we removed the lipid modification site from the full-length wild-type Myc-tagged construct and replaced it with each of several proteinaceous transmembrane domains (TM) taken from the late endosomal SNARE VAMP7, the ER/Golgi SNARE Sec22b and the medial/trans-Golgi SNARE GOS-28. As shown in Fig. 9, the Ykt6 longin domain was completely unable to target the transmembrane Ykt6 constructs to its normal vesicular structures. In fact, in each case, the construct localized similarly to the SNARE from which the transmembrane domain was derived; Myc-Ykt6/Sec22bTM localized to the Golgi area and colocalized with syntaxin 5, Myc-Ykt6/VAMP7TM localized to the limiting membrane of abnormally swollen, acidic lysosome-like structures and Myc-Ykt6/GOS28TM localized to the Golgi area but did not colocalize well with the cis-Golgi marker GM130, consistent with the construct being localized to the late Golgi. We conclude that either the normal punctate Ykt6 structure is unavailable from the mainstream endomembrane system or the targeting determinants in the Ykt6 longin domain are subordinate to the signals present in the transmembrane domains used. In either case, it emphasizes the importance of the intramolecular interactions described above for the `proper' subcellular targeting of Ykt6 in cultured neuroendocrine cells.

It was interesting that the Myc-Ykt6/VAMP7TM construct caused aberrant lysosome-like structures (Fig. 9, middle). This implies that Ykt6 might have a role in lysosome fusion, consistent with the function of Ykt6p in yeast. However, it does not clarify the relationship between the unique Ykt6 punctate structures and lysosomes and/or Ykt6 function.

The endogenous punctate Ykt6 structures do not colocalize significantly with markers of the endomembrane system (Hasegawa et al., 2003), so we considered the possibility that they were not a natural structure but an aberrant structure present only in transformed cell lines like PC12. We thus investigated the subcellular localization of endogenous and Myc-tagged recombinant Ykt6 expressed in primary cultures of hippocampal neurons. As shown in Fig. 10A, endogenous Ykt6 staining (red) was punctate in nature, similar to that in PC12 cells and ng108 cell lines. Punctate staining was heavy throughout the entire cell body of neurons, as well as major dendrites and axons. The Ykt6 structures did not appear to be enriched in any particular part of the cell but rather accumulated wherever there was a significant volume of cytoplasm. As shown in Fig. 10B-E, Ykt6 puncta were even observed within individual synaptic terminals, where the presynaptic plasma membrane was indicated with an antibody to the exocyst component rSec6. This work demonstrates that the unusually specialized localization of Ykt6 in cultured neuroendocrine cells is similar to that observed in bona fide brain neurons, where Ykt6 is most abundant (Hasegawa et al., 2003).

Many features of mammalian Ykt6 remain a mystery. Perhaps most importantly, we do not yet know the molecular identity of the punctate structures to which the protein targets or whether Ykt6 functions there as a SNARE. Markers of the endomembrane system did not significantly colocalize with Ykt6, but particulate Ykt6 behaved as a membrane-associated protein with respect to detergent extractability and buoyant density. By contrast, cytosolic Ykt6 behaved as a purely monomeric protein, without a hint of aggregation or higher-order organization (Hasegawa et al., 2003). The best interpretation of the data so far is that cytosolic Ykt6 is an intermediate in the localization to the specialized compartment (i.e. that cytosolic Ykt6 targets to the punctate structures by binding to them directly from the cytosol). Whether the punctate structures represent traditional membrane-bilayer-enclosed organelles is still an active question.

We have now thoroughly defined the structural features of Ykt6 that allow targeting to its unique membrane location. The overall conserved longin tertiary structure, rather than sequence-specific or surface structure, contains the signal for direct localization to the Ykt6 compartment. However, the Ykt6 longin domain indirectly controls the localization of the protein by regulating the availability of targeting signals in the SNARE motif and C-terminal lipid moieties. This regulation involves sequence-specific intramolecular interactions between the longin domain and the C-terminal lipids and SNARE motif. We identified three distinct surfaces on the Ykt6 longin domain that are important for these intramolecular interactions. One included residues previously identified by Banfield and colleagues to participate in protein-protein interaction with the SNARE motif (Tochio et al., 2001). Another surface on the same face of the domain appears to involve a mixture of hydrophobic and charged residues. Finally, a third surface represents a deep hydrophobic crevice on the opposite face of the domain. Indirect evidence we present suggests that this crevice might represent a palmitoyl-group binding pocket. Our interpretation of the role of the intramolecular interactions in targeting is that they allow the SNARE to adopt a compact, closed conformation that sequesters the SNARE motif and lipid(s), thus preventing otherwise promiscuous targeting interactions.

Interestingly, significant bimolecular interactions were not observed between the longin domain and SNARE motif of Ykt6 in the absence of lipid modification (Fig. 8A). At first glance, this seems at odds with Banfield and colleagues, who reported that intramolecular protein-protein interactions between the longin domain and SNARE motif were detectable by NMR in the purified, recombinant, non-hydrophobically modified Ykt6 (Tochio et al., 2001). However, those purely protein-protein interactions might have been weak or transient, as indicated by their very modest inhibitory effect on SNARE complex formation. Our interpretation of the results is that the protein-protein interactions by themselves are not strong enough to be detected in the bimolecular binding assay we used. However, they contributed significantly, because mutation of the residues (e.g. F42) that Banfield and colleagues found to be important for protein-protein interactions (Tochio et al., 2001) inhibited binding in our assay. Thus, we propose that both protein-protein and protein-lipid interactions are required to produce a stable intramolecular interaction and tightly closed conformation.

Our present and previous (Hasegawa et al., 2003) results suggest several unique conformational properties of Ykt6 in neuroendocrine cells, which are summarized in the model of Fig. 11. As depicted, fully prenylated, perhaps palmitoylated Ykt6, exists as a freely soluble monomer before targeting to its specialized vesicular structure. It maintains solubility by having the longin domain act as a lipid chaperone that sequesters these groups in hydrophobic binding pockets. Partly as a consequence of the intramolecular protein-lipid interactions, the longin domain is locked in place tightly against the SNARE motif, which also interacts with the longin domain. This compact cytosolic conformation sequesters would-be promiscuous interactions of the SNARE motif and lipid group(s), thus preventing mistargeting. The longin domain of cytosolic Ykt6 is, however, free to interact with an unknown targeting component on the particulate Ykt6 compartment. This targeting interaction requires only the tertiary structure of the longin domain and is not directly affected by any of the surface mutations explored in this study. Once targeted to the Ykt6 particle, the lipid group(s) might partition into a fully lipidic environment, strongly anchoring Ykt6 there in a detergent-extractable but not chaotrope-extractable condition. The retraction of the lipid group(s) from their longin-domain-binding sites would dramatically weaken the longin-SNARE intramolecular interactions, opening up the molecule and making the SNARE motif available for potential participation in SNARE interactions. The existence of intramolecular protein-protein as well as protein-lipid interactions allows the longin domain to regulate Ykt6 conformation over a wide dynamic range, such that it virtually shuts off all interactions when soluble, whereas only mildly modulating SNARE availability when embedded in membranes. Hopefully, future work will elucidate the nature of the particulate structure that neuronal Ykt6 goes to such lengths to arrive at.

This work was supported by NIH grant MH 68524 to J.C.H. S.D. and L.O. received support from EU grant QLG3-CT-2001-02089.