Multiple regions contribute to membrane targeting of Rab GTPases

Small GTPases of the Rab family are key regulators of membrane trafficking. Each Rab shows a characteristic subcellular distribution, and may serve as an important determinant of organelle identity. The molecular mechanisms responsible for targeting Rabs to specific intracellular compartments, however, remain poorly understood. The divergent C-terminal hypervariable region was postulated to contain Rab targeting information. We generated a series of hybrid Rab proteins by exchanging the hypervariable domains of Rab1a, Rab2a, Rab5a, Rab7 and Rab27a, and analysed their subcellular localisations. We found that the various hybrid proteins retained their targeting to the parent organelle and were functionally active. We conclude that the hypervariable region does not contain a general Rab targeting signal. Furthermore, we identified other regions within the RabF and RabSF motifs that are required for specific targeting of Rab27a to secretory granules or melanosomes, and Rab5a to endosomes. We observed only partial overlap between targeting-determining regions in the Rab proteins examined, suggesting that Rab recruitment may be complex and at least partially Rab-specific. Mutations in these targeting-determining regions induced localisation to the ER, an observation that further strengthens our previous finding that ER/Golgi membranes serve as the default location for Rabs that have lost targeting information.

localisation of the hybrid Rab to Rab7-positive late endosomal structures (Chavrier et al., 1991). Conversely, the C terminus of Rab5 targeted a chimeric Rab6/Rab5 (Stenmark et al., 1994) to early endosomes. In Saccharomyces cerevisiae, the hypervariable domain of Ypt1p reportedly re-localised Sec4p, and vice versa (Brennwald and Novick, 1993). However, complementation studies showed that the hybrid Ypt1p/Sec4p proteins retained their original function indicating that at least a proportion of the protein reached its original compartment (Brennwald and Novick, 1993;Dunn et al., 1993).
Targeting of Rab proteins via the hypervariable region is a widely accepted model, but the underlying molecular mechanisms have not been elucidated further. In this study we analysed the structural requirements for targeting of Rab GTPases, with a focus on the specific determinants present in the C-terminal hypervariable domain. Surprisingly, we found that the hypervariable region did not represent a general targeting domain. We identified other sequence elements within Rabs that appear to be necessary for specific localisation, suggesting a model where multiple regions contribute to subcellular targeting.

Plasmid constructs
Rab hybrid proteins were initially generated by mutating the conserved ETSAK motif to introduce an Eco47III site and silencing the internal Eco47III site present within the pEGFP vector using the Stratagene Quickchange site-directed mutagenesis system (SDM), as described previously (Gomes et al., 2003). The C termini of selected Rabs were exchanged by digesting the respective pEGFP-Rab plasmid with Eco47III and NdeI, then ligating the desired Eco47III-NdeI fragment from a different Rab. The length of the hypervariable domain on the hybrid proteins was adjusted by SDM where desired. Trypanosoma brucei cDNAs encoding Rab5a orthologues (Pal et al., 2002) were a gift from Mark Field (Imperial College, London, UK) and subcloned into pEGFP-C2. S. cerevisiae cDNAs encoding Ypt51, Ypt52 and Ypt53 in pEGFP-C3 were a gift from Ruth Collins (Cornell University, NY). Full details for each plasmid constructed are described in the supplementary material.
Mice and melanocyte cell line C57BL6/J mice homozygous for ashen (Rab27a ash ) (Barral et al., 2002) were crossed with C57BL6/J mice carrying an Ink4a-Arf exon 2 deletion (Serrano et al., 1996), at Texas A&M University. Litters homozygous for both mutations were obtained at the third generation, and trunk skins from these neonatal mice were used at St. George's Hospital Medical School for preparation of melanocyte cultures, as described previously (Sviderskaya et al., 2002).
Immunofluorescence and confocal microscopy Cells were washed with PBS, fixed with 3% paraformaldehyde for 15-30 minutes, washed three times with PBS and once with 50 mM NH 4 Cl in PBS, then permeabilised in PBS containing 0.05% saponin and 1% bovine serum albumin. Permeabilised cells were incubated with primary antibody followed by Alexa568-conjugated secondary antibody (Molecular Probes). All incubation and washing steps were carried out in PBS/BSA/saponin. Coverslips were mounted in ImmunoFluor medium (ICN, Basingstoke, Hants, UK) and fluorescence was visualised using a Leica DM-IRBE confocal microscope (Leica, Wetzlar, Germany). Images were processed with TCS-NT software (Leica) and Adobe Photoshop 5.5 software. All images presented are single section in the z-plane and representative of at least 80% of the transfected cells.

Results
Targeting of a tissue-specific Rab protein in a specialised cell type In order to study the targeting determinants present in the Cterminal hypervariable domain, we produced a series of Journal of Cell Science 117 (26) Fig. 1. Schematic representation of wild-type and C-terminal hybrid Rab proteins used in this study. The numbers above each bar are residue numbers indicating the total length of each wild-type Rab, or the point at which the hypervariable region from another Rab was fused to generate a hybrid protein. On the right, the subcellular localisations of the proteins used are summarised: SG, secretory granule; EE, early endosome; G, Golgi; *denotes partial mislocalisation.
chimeric Rabs containing exchanged C termini (Fig. 1). One of the model Rabs used in this study was Rab27a, which is expressed in a tissue-specific manner, and is localised to melanosomes in pigmented cells (Bahadoran et al., 2001;Hume et al., 2001;Wu et al., 2001) and dense-core secretory granules in neuroendocrine cells (Desnos et al., 2003;Tolmachova et al., 2003;Zhao et al., 2002). In AtT20 anterior pituitary-derived cells, EGFP-Rab27a co-localised with the secretory granule marker ACTH, particularly at the tips of extended cell processes, which are enriched in mature secretory granules ( Fig. 2A-C). Little Rab27a was detected in a perinuclear ACTH-positive compartment representing hormone precursor in transit through the secretory pathway (Wasmeier and Hutton, 2001). For the construction of Rab27 hybrid proteins (Fig. 1), the N-terminal 187 residues of Rab27 had to be retained to allow membrane association of the proteins, perhaps reflecting the fact that Rab27a is naturally an inefficient substrate for prenylation and is thus more sensitive to structural rearrangements (Larijani et al., 2003). The Cterminal 35 amino acids were proposed to be the minimal sequence necessary and sufficient for targeting of Rab5 and Rab7 (Chavrier et al., 1991), therefore we fused the hypervariable regions to a residue 35 amino acids away from the C terminus of Rab27, located just downstream of helix α5. The Rab27/Rab5 hybrids (Rab27 187 /Rab5 35 and Rab5QL 182 / Rab27 39 ) and Rab27/Rab1 hybrids (Rab27 187 /Rab1 40 and Rab1 172 /Rab27 40 ) were expressed in AtT20 cells. Neither the Rab5a hypervariable region (Fig. 2M-R) nor the Rab1a hypervariable region (Fig. 3J-O) resulted in the re-targeting of Rab27a from secretory granules to either the early endosome AtT20 cells expressing the dominant active form of Rab5a, Rab5aQ79L, showed characteristically enlarged endosomes, positive for the endosomal marker transferrin receptor (Tfn-R). Similarly, enlarged endosomes were observed in cells expressing Rab5QL 182 /Rab27 39 , suggesting that the hybrid protein retained functional activity. To test whether the Rab27a C terminus was important for function, we used melanocytes derived from the ashen (Rab27a null) mouse, which characteristically accumulate pigment granules in the perinuclear region (

Rab5 orthologues are targeted to endosomes in mammalian cells
To define conserved determinants for endosomal targeting of the ubiquitously expressed Rab5 protein, we used the Clustal W sequence alignment algorithm to analyse the C-terminal hypervariable domains of several Rab5 orthologues in lower eukaryotes. Apart from the cysteine residues in the prenylation motif at the extreme C terminus and a phenylalanine residue in helix α5, there was no significant sequence conservation at the amino acid level between mammalian and lower eukaryote Rab5 sequences (see Fig. S1 in supplementary material). We expressed EGFP-tagged Rab(Ypt) proteins representing Rab5 homologues from the lower eukaryotes Saccharomyces cerevisiae (Ypt51, Ypt52 and Ypt53) and Trypanosoma brucei (TbRab5A and TbRab5B) in HeLa cells. Using confocal Journal of Cell Science 117 (26) microscopy, we observed localisation of all five Rab5-like proteins to dispersed punctate structures positive for the Tfn-R, an endosomal marker (Fig. 5), suggesting that Rab5 orthologues are correctly targeted to early endosomes in HeLa cells. In addition, some Ypt52p, TbRab5A and TbRab5B were also observed in the perinuclear region, consistent with a recycling endosome and/or Golgi localisation (Fig. 5). Despite their divergent hypervariable domains, it appears that all Rab5 proteins examined contained the necessary targeting information for localisation to early endosomes.
Membrane targeting of the related GTPase K-Ras is dependent on basic residues within the C-terminal region (Hancock et al., 1991;Hancock et al., 1990). We replaced the six basic residues in the hypervariable domain of Rab5a with non-charged residues (Rab5∆BR). These mutations were introduced into the dominant active form of Rab5a, Rab5aQ79L (Bucci et al., 1992). Transiently expressed EGFP-Rab5QL∆BR associated with enlarged structures that also contained the endosomal marker EEA1, suggesting that the mutant protein was both correctly targeted to early endosomes and functionally active (our unpublished results).
The C-terminal hypervariable domains of Rab1, Rab2 and Rab7 did not alter the endosomal localisation of Rab5 To extend our analysis of the role of the hypervariable region in Rab targeting, we chose other well-characterised Rabs with distinctive cellular distributions in mammalian cells. We generated a series of hybrid Rab proteins bearing the Nterminal 174 residues of Rab5a attached to the corresponding C terminus of another Rab (Fig. 1). All hybrid Rabs described in this study, except where noted, were prenylation-competent as evidenced by three criteria. Firstly, they partitioned into the detergent phase in Triton X-114 extraction experiments (see Fig. S2 in supplementary material); secondly, they did not accumulate within the nucleus of transfected cells, unlike soluble EGFP-tagged Rabs lacking a prenylation motif; thirdly, they resisted cell permeabilisation prior to fixation. Prenylation competence of a hybrid or mutant Rab protein is strong evidence that the protein has retained its overall Rab structure. This is because of the high stringency and specificity of the REP/Rab geranylgeranyl transferase system, such that only correctly folded Rab proteins undergo geranylgeranylation (Seabra, 2000).
EGFP-tagged Rab5a hybrid proteins were transiently expressed in HeLa cells, and their subcellular localisation was analysed by confocal light microscopy. Wild-type Rab5a decorates Tfn-R-positive early endosomes (Fig. 5A-C) and regulates early endosome dynamics (Zerial and McBride, 2001). The constitutively active mutant, Rab5aQ79L, induces the formation of enlarged endosomes which tend to cluster in the perinuclear region ( Fig. 2 and data not shown) (Zerial and McBride, 2001). The hypervariable regions of Rab1a, Rab2a or Rab7 did not disrupt the targeting of Rab5a to early endosomes, as shown by co-localisation of the hybrid proteins with myc-tagged wild-type Rab5a (Fig. 6A-C, G-I and M-O, respectively) or Tfn-R (not shown). Whilst most of these hybrids were only observed on endosomes labelled by the co-transfected marker myc-Rab5, a proportion of the Rab5 174 /Rab7 41 hybrid showed a perinuclear pattern, which could represent either ER/Golgi and/or late endosomal localisation (Fig. 6M). Furthermore, constitutively active forms (Q79L) of the Rab5a hybrids formed enlarged endosomes that co-localised with the Rab5 effector EEA1 (Fig. 6D-F,J-L,P-R) and Tfn-R (data not shown), suggesting that the hybrid proteins retained Rab5a activity. These results suggest that the hypervariable regions of Rab1a and Rab2a are not sufficient to target Rab5a to their cognate compartments. Furthermore, the Rab5a hypervariable domain itself is not necessary for targeting to early endosomes, or function as measured by endosome fusion.
The C-terminal hypervariable region of Rab5a did not re-target other Rab proteins to early endosomes To extend the above findings, we generated the converse hybrid proteins by fusing the C-terminal hypervariable domain of Rab5a to the N-terminal region of Rab1a or Rab2a (Fig. 7). Wild-type Rab1a and Rab2a localised to the Golgi apparatus, as previously reported ( Fig. 7A-C,G-I) (Chavrier et al., 1990;Goud et al., 1990). The Rab1 156 /Rab5 50 and Rab2 160 /Rab5 41 hybrids associated with perinuclear membranes where they co-localised with Golgin-97, a Golgi marker (Fig. 7D-F and J-L, respectively). Neither of these hybrids was present in Tfn-R-positive endosomes ( Fig. 7M-O; our unpublished results).
All Rab chimeras described in the above studies had been fused to EGFP. There is evidence in the literature suggesting that EGFP fusion does not alter localisation or indeed function (Tolmachova et al., 2003). Furthermore, some EGFP-Rab5 chimeras were found to co-localise with co-expressed myc-Rab5 (Fig. 6). Nevertheless, to exclude the possibility that targeting or function of the hybrid Rabs could have been altered by the presence of the EGFP tag in our studies, we replaced it with the much smaller myc-tag in a number of our constructs. Myc-tagged Rab27 187 /Rab5 35 was localised to secretory granules in AtT20 cells (see Fig. S3D-F in supplementary material), similar to the EGFP-tagged version (Fig. 2M-R). Myc-Rab5QL 182 /Rab27 39 , however, localised to Tfn-R-positive enlarged endosomes (see determinants N-terminal of the hypervariable domain. In recent studies, we performed large-scale bioinformatics analyses on Rab sequences from a range of species where we identified Rab family-specific (RabF) and Rab subfamily-specific (RabSF) sequences (Pereira-Leal and Seabra, 2000;. Five short sequence motifs that distinguish mammalian Rab proteins from other classes of small GTPases were defined as RabF regions, while four larger noncontiguous segments specifically conserved between subfamily members were designated RabSF regions. To test the role of RabF and RabSF regions in membrane targeting, we created hybrid proteins based on either Rab5a or Rab27a by exchanging residues within the RabF and RabSF regions between the two proteins. In most cases, targeting was not affected (our unpublished results). One example is the introduction of the Rab27a SF1 (residues 17-21) plus SF4 (residues 183-202) into Rab5a (Fig.  8M-O). Substitution of the RabSF2 region resulted in a mutant protein unable to associate with membranes ( Fig. 8P-R), most likely because of defects in prenylation, given the lack of hydrophobicity of the transfected protein (see Fig. S2 in supplementary material). Unfortunately, the cytosolic distribution of this fusion protein did not allow us to assess its role in targeting. However, certain changes resulted in dramatic effects on targeting of membrane-associated Rabs. Introducing the RabSF3 region of Rab27a (residues 109-132) into Rab5a led to a redistribution of the mutant Rab5QL/27 SF3 protein, resulting in the partial co-localisation with the ER marker PDI and the Golgi marker Golgin-97 ( Fig. 8A-I). Substitution of the RabF4 region of Rab5a (residues 86-93; Rab5QL/Rab27 F4 ) with that of Rab27a had a similar effect (Fig. 8J-L).
Mutations were also introduced into Rab27a and analysed in AtT20 cells. Substituting part of the RabSF2 region of Rab27a (amino acid residues 35-40) with equivalent residues present in Rab5a led to a loss of secretory granule targeting and the redistribution of the hybrid Rab27/5 SF2 to other cellular membranes, most prominently in the perinuclear region where it co-localised with the Golgi marker Golgi-58K (Fig. 9A-F). Similarly Rab27/5 SF3 (substitution of residues 110-116) which contained part of the Rab5 SF3 region, was observed on ER and Golgi membranes instead of secretory granules (Fig. 9G-L). Unlike Rab5QL/Rab27 F4 , the substitution of the RabF4 region in Rab27a did not affect the localisation of Rab27/Rab5 F4 , which was seen exclusively on ACTH-positive granules (Fig. 9M-O). Altogether, our studies suggest that selected regions other than the hypervariable region are required for membrane targeting, including RabF4 and RabSF3 in Rab5a, and RabSF2 and RabSF3 in Rab27a.

Discussion
We present evidence suggesting that the Rab hypervariable region is not a general targeting domain for Rab GTPases. Instead, we identified other regions in the RabF and RabSF motifs that are necessary for specific targeting. We observed only partial overlap between regions determining the targeting of the different Rab proteins examined, indicating that Rab recruitment may be complex and partially Rab-specific. Our studies further suggest that the ER/Golgi membranes serve as the default location for Rabs that have lost targeting information.

The C-terminal hypervariable region is not a general targeting determinant
Early studies suggested that the highly variable C-terminal end of Rab proteins was the main structural determinant for their specific targeting to distinct cellular membranes (Chavrier et al., 1991;Stenmark et al., 1994). Surprisingly, our present studies do not support this model. Reciprocal exchanges of the hypervariable domains of Rab1a, Rab2a, Rab5a, Rab7 and Rab27a failed to re-direct the hybrid proteins away from their original compartment to the new compartment designated by the hypervariable region. Furthermore, lower eukaryote (S. cerevisiae and T. brucei) homologues of Rab5 fused to EGFP are correctly targeted to early endosomes in mammalian cells, as previously described for S. cerevisiae Rab5 orthologues (Singer-Kruger et al., 1995), despite little conservation of the hypervariable domain at the amino acid level. We also show here that a hybrid Rab27a with a Rab5a hypervariable domain retained Rab27 function, as determined by its ability to rescue the melanosome distribution defect observed in melanocytes derived from the ashen (Rab27a null) mice (Fig. 4). Our present findings are consistent with a previous study which suggested that the Ras hypervariable region in the context of Rab6 did not affect targeting of the fusion protein to the Golgi apparatus, the location of wild-type Rab6 (Beranger et al., 1994). In yeast, reciprocal exchanges of the hypervariable domains between Ypt1 and Sec4 had no effect on their function. Ypt1p containing the hypervariable domain of Sec4p fully complemented a deletion of YPT1 in yeast (Brennwald and Novick, 1993;Dunn et al., 1993) and similarly Sec4p with the hypervariable domain of Ypt1p complemented the SEC4 deletion (Brennwald and Novick, 1993). These complementation studies suggest that a significant proportion of the hybrid proteins were correctly targeted to their original site of action and that they were indeed functional.
The difference between our conclusions and earlier results proposing a key role for the hypervariable C-terminal domain in targeting remains puzzling. However, it should be noted that while the current study examines Rab proteins acting in very distinct cellular locations and pathways (such as melanosomes, endosomes and Golgi), the main focus in previous work had been on Rabs functioning within the endosomal system. Given that targeting appears to involve multiple and partially Rabspecific regions, it is possible that Rabs with an 'endosomal character' share some localisation determinants, but possess Journal of Cell Science 117 (26) additional targeting information in their hypervariable domains, which contributes to their accurate localisation within the endosomal system. It may be significant that the only construct in our studies that was similar to one used in the early studies, the Rab5/Rab7 hybrid, partially exhibited an aberrant localisation. Whilst it is impossible to rule out late endosomal localisation, we found this aberrant localisation more consistent with ER mistargeting (Fig. 6). Furthermore, it remains possible that the GFP and myc-tags may have influenced our results, although the overwhelming evidence suggests that the tags do not introduce such artifacts. Finally, the expression systems may have influenced the results as the vaccinia system used previously allowed observation at earlier time points after transfection.

The ER and Golgi as a default localisation for Rab proteins that have lost targeting information
We had previously observed that the ER and Golgi are preferential sites of membrane translocation for mistargeted Rabs, i.e. Rabs that have lost targeting information. Rabs with mutations in the prenylation motif that resulted in singly rather than doubly geranylgeranylated proteins showed mistargeting to the ER and loss of function (Gomes et al., 2003). Furthermore, the expression of Rab27a in undifferentiated cultured cell types such as HeLa or HEK-293 cells led to ER localisation (Hannah et al., 2003). Thus, the absence of specialised compartments such as secretory granules and/or lysosome-related organelles in these cells appears to result in the association of Rab27a with the ER by default (Hannah et , 2003;Hume et al., 2001;Tolmachova et al., 2003;Yi et al., 2002). Here we show several examples of ER/Golgi targeting upon domain swapping and we suggest that this localisation reveals mistargeting.

Regions necessary for targeting
Mistargeting to the ER was thus used as an assay to search for regions necessary for Rab targeting. We focused our mutagenesis studies on the previously characterised RabF and RabSF regions (Pereira-Leal and Seabra, 2000;. The RabF and surrounding sequences form the switch regions, which change conformation upon GTPbinding, and the RabSF regions are areas of greater conservation between subfamily members, such as the Rab5/21/22 subfamily and the Rab27/3 subfamily. Thus, targeting information could potentially reside in these regions. We observed clear mistargeting and possibly loss of function when swapping certain internal elements between Rab5a and Rab27a including RabSF2, RabF4 and RabSF3. The RabF4 region, typically YYRGA, is generally highly conserved between Rabs but Rab27a possesses a significantly different F4 region, FFRDA. This may have caused local structural changes resulting in loss of targeting information. The RabSF2 region, more precisely loop 2, had previously been implicated in Rab6 targeting to the Golgi (Beranger et al., 1994) and our results are consistent with this observation as they suggest that Rab27 SF2 (loop 2) is necessary for targeting to secretory granules. The RabSF2 region may also be involved in targeting other members of the Rab family, but in some cases (such as Rab5) mutagenesis in this region resulted in soluble proteins, making it difficult to analyse its contribution. Finally, our results suggest that the RabSF3 region may play an important role in targeting.
It is noteworthy that two of the candidate targeting regions reported here partially overlap with the switch regions (RabSF2 with switch I and RabF4 with switch II), and are expected to participate in effector binding. This raises the possibility that the interaction with effectors may promote Rab targeting. Taken together, our data suggests that Rab targeting Journal of Cell Science 117 (26) is determined by multiple sequence motifs within Rab proteins that are not necessarily localised in the same vicinity.

Models for Rab membrane targeting
The molecular mechanisms regulating Rab membrane delivery by REP/GDI proteins remain unclear (Pfeffer, 2001;Seabra and Wasmeier, 2004). At least three activities appear to be necessary: recognition of the correct intracellular target membrane, dissociation of the Rab:REP/GDI complex through a GDI displacement factor (GDF) and nucleotide exchange to prevent membrane extraction. Recently, Pfeffer and co-workers demonstrated that PRA1/Yip3 possesses GDF activity (but not exchange activity) towards endosomal Rabs (Sivars et al., 2003). Yip proteins, a family of Rab-binding integral membrane proteins, could aid in Rab targeting but may not represent the main targeting determinant since they display broad specificity towards different Rabs (Sivars et al., 2003). Another interesting model suggested that a Rab protein may regulate the activation of another Rab acting in a downstream transport step by recruiting its nucleotide exchange factor (GEF), as proposed for the recruitment by Ypt32 of Sec2p, a GEF for Sec4p (Ortiz et al., 2002). Certain Rab effectors, once recruited, may also serve to expand the Rab-specific membrane domain through generation of a positive feed-back loop (Zerial and McBride, 2001). This would argue for a direct link between Rab targeting and Rab function, as discussed above. Taken together, it seems likely that multiple factors contribute to the specificity and regulation of Rab recruitment.
The present studies identifying targeting determinants within Rabs represent the first step towards testing these models. Future experiments will be aimed at elucidating the role of these targeting-determining regions in Rab localisation. This should provide further mechanistic insights into the regulated delivery of Rab GTPases to specific intracellular membranes and into the mechanisms regulating the organisation of intracellular compartments.