The shape of the transmembrane domain is a novel endocytosis signal for single-spanning membrane proteins

The protein composition of cell membranous organelles depends on the efficiency with which proteins are incorporated into transport intermediates. This is generally encoded in cargo proteins in the form of small linear motifs that interact with adaptor proteins, which, in turn, are recognised by the vesicle coats, resulting in cargo recruitment. These motifs localise to the cytosolic region of transmembrane cargo proteins or transmembrane adaptors (Traub and Bonifacino, 2013). However, transmembrane domain (TMD) features may also determine the sorting of transmembrane proteins in several steps of the secretory pathway (Reggiori et al., 2000; Rayner and Pelham, 1997). For instance, the length of TMDs is involved in Golgi versus plasma membrane (PM) localisation. It was proposed that long TMDs would accommodate better to the thicker bilayer of the PM (Bretscher and Munro, 1993; Dukhovny et al., 2009; Munro, 1995; Rayner and Pelham, 1997; Ronchi et al., 2008). More recently, bioinformatics analyses confirmed that TMDs have organelle-specific features (Quiroga et al., 2013; Sharpe et al., 2010). Work from our laboratory indicates that geometric features of TMDs, including both the length of the TMD and the volume of the amino acids that constitute their exoplasmic halves, could determine Golgi retention or transport to the PM for type-II membrane proteins, in yeast and mammalian cells. Using transmembrane SNARE proteins as models, we showed that short TMDs with high-volume exoplasmic halves are retained in the Golgi complex, short TMDs with low-volume exoplasmic halves are transported to the PM and long TMDs are transported to the PM regardless of the volume of their exoplasmic halves (Quiroga et al., 2013).

A role for TMD length was also postulated in the exclusion of mammalian surface proteins from clathrin-coated vesicles (Mercanti et al., 2010). Signals present in the cytoplasmic region of cargo proteins are recognised by adaptor proteins that recruit clathrin to the membrane and initiate endocytosis (Bonifacino and Traub, 2003). In Saccharomyces cerevisiae, endocytosis and recycling of transmembrane proteins result in a polarised distribution at the PM due to the slow diffusion of these proteins in this particular membrane. This is how the SNARE Snc1 attains its polarised localisation. The SNARE Sso1 is not actively endocytosed and thus presents a homogeneous distribution at the PM (Valdez-Taubas and Pelham, 2003). In yeast, endocytosis is almost exclusively clathrin mediated, although a minor clathrin-independent pathway is described (Prosser and Wendland, 2012; Weinberg and Drubin, 2012).

Here, we report that increasing the volume of the residues that constitute the exoplasmic hemi-TMD of Sso1 results in a polarised distribution of the mutant protein. Analysis of this mutant in different strains affected in either endocytosis or recycling revealed that this polarisation is due to endocytic cycling. Moreover, a bioinformatics analysis of the S. cerevisiae proteome detected several proteins with naturally occurring high-volume exoplasmic hemi-TMDs, some of which were reported to exhibit a polarised distribution. TMDs from these proteins can confer a polarised distribution to the Sso1 cytoplasmic domain. Our experiments indicate that a high-volume exoplasmic hemi-TMD is a novel endocytosis signal and polarity determinant in yeast. Expression of chimeric proteins with high-volume exoplasmic hemi-TMDs also results in endocytosis in a cultured mammalian cell line.

Sso1 is a plasma membrane SNARE that belongs to the syntaxin family (Aalto et al., 1993). It contains a single TMD at its C-terminus, predicted to be 22 amino acids long, and the residues that constitute its exoplasmic half are on average of small volume (Fig. 1A) (Quiroga et al., 2013). Sso1 displays a homogeneous distribution at the PM of vegetative cells and shmoos, the yeast mating intermediates. If an endocytosis signal such as NPF is appended to its N-terminus, it is now internalised and recycled, resulting in a polarised distribution in both types of cells (Valdez-Taubas and Pelham, 2003).

As part of a systematic analysis of the effect of TMD shape in protein sorting, we made a mutant version of Sso1 in which the exoplasmic half of the TMD was replaced by an extended version of that of the Golgi SNARE Sft1, resulting in the chimeric protein Sso1₁₁–Sft1₁₁ (Fig. 1A) (Quiroga et al., 2013). This chimeric protein, which contains high-volume amino acids in the exoplasmic half of its TMD, localises to the PM where it is highly polarised to daughter cells (buds) and to the tips of shmoos (Fig. 1B). This chimeric protein also localises to some intracellular structures, which mostly correspond to TGN, as shown by co-localisation with the Sec7 marker (72.3%). Some structures correspond to late endosomes as shown by co-localisation with Vps8 (26.2%) (Fig. S1). In an sla1Δ strain defective for clathrin-mediated endocytosis (Warren et al., 2002), Sso1₁₁–Sft1₁₁ is observed at the PM, but polarity is lost. In a ric1Δ strain, which allows endocytosis but blocks recycling (Siniossoglou et al., 2000), the mutant protein is almost exclusively intracellular, whereas Sso1 remains at the PM (Fig. 1B). In order to obtain a quantitative measure of polarity, we determined the fluorescence intensity at the bud tip and at the centre and rear of the mother cell, and the values were normalised to the fluorescence at the bud tip for each cell. Fig. 1C shows a quantification of polarity for Sso1 and Sso1₁₁–Sft1₁₁, which confirms polarised distribution of the mutant protein in a WT strain and loss of polarity in the sla1Δ strain. Both constructs are expressed at similar levels in all the strains analysed (Fig. S2A). These experiments indicate that polarised localisation of Sso1₁₁–Sft1₁₁ is achieved by endocytic cycling and that this high-volume exoplasmic hemi-TMD can act as an endocytic signal.

To confirm that Sso1₁₁–Sft1₁₁ endocytosis is due to the volume of the exoplasmic hemi-TMD and not to a specific amino acid sequence, the order of the amino acids in the exoplasmic hemi-TMD was randomised, resulting in an Sso1₁₁–Sft1_11(scr) chimera (Fig. 1A). This mutant is also polarised in an endocytosis- and recycling-dependent manner (Fig. 1B), confirming that the endocytic signal does not depend on a particular amino acid sequence but rather on the average volume of the TMD exoplasmic half.

It is possible that the Sso1 TMD allows for interactions that preclude endocytosis so that it is the absence of its low-volume hemi-TMD, rather than the presence of a high-volume hemi-TMD in the chimera, that results in endocytosis. To analyse this possibility, we generated a chimera in which the Sso1 exoplasmic hemi-TMD was replaced by the low-volume exoplasmic hemi-TMD from the cell wall integrity sensor Mid2 (Ketela et al., 1999; Philip and Levin, 2001; Rajavel et al., 1999) (Fig. 2A). We chose this strategy instead of scrambling the Sso1 exoplasmic half, since it is mostly composed by valine residues, and scrambling it would result in a very similar sequence (see Fig. 1A). Fig. 2B shows that a substantial amount of Sso1₁₁–Mid2₁₁ reaches the PM, where it displays a homogeneous distribution in WT cells, and it is not affected by sla1Δ and ric1Δ mutations, similarly to Sso1.

The proteome of S. cerevisiae was analysed in silico (see Materials and Methods). We obtained a list of 755 proteins predicted to contain a single TMD; 37 of these proteins are annotated as being localised to the PM. Fig. 3A shows a plot of the average volume of amino acids in the exoplasmic half of the TMD versus TMD length for all S. cerevisiae proteins predicted to have a single TMD (PM proteins are red). The two PM proteins that have the highest average amino acid volume in the exoplasmic half of their TMDs are Hkr1 and Kre6. These proteins are known to display a polarised distribution on the PM, so we selected them for experimental analysis.

Hkr1 is a type I membrane protein involved in osmosensing (Tatebayashi et al., 2007). To determine whether its TMD bears an endocytic signal, we generated a chimera between the Sso1 cytoplasmic domain and the TMD of Hkr1 (Fig. 3B). Since Hkr1 is a type I membrane protein, the sequence of the TMD was inverted to maintain the topological relationships. Fig. 3C shows that the TMD of Hkr1 can confer polarised distribution to the cytoplasmic domain of Sso1 both in vegetative cells and shmoos, and this is dependent on endocytic cycling because polarity is lost in sla1Δ cells and the protein accumulates in intracellular structures in ric1Δ cells. The expression levels of Sso1 and Sso1(Hkr1) were analysed by western blot and are similar in all the strains we analysed (Fig. S2B).

Kre6 is a type II membrane protein involved in glucan biosynthesis and displays a polarised distribution in the PM (Kurita et al., 2011; Roemer et al., 1993). This polarity is maintained by endocytic cycling, since it is lost in a sla1Δ strain, and the fluorescence is internal in a ric1Δ strain (Fig. S3). The predicted TMD for Kre6 is 21 amino acids long and has a high-volume exoplasmic hemi-TMD (Table S1, see Fig. 4A for a scheme). We generated a chimera to the cytoplasmic domain of Sso1 and expressed it in yeast cells. Microscopy analyses indicate that this chimera localises to the vacuole lumen (Fig. 4B). Vacuole localisation was established by labelling the vacuole membranes with the FM4-64 lipophilic dye (Vida and Emr, 1995). This localisation is indicative of protein degradation, which was confirmed by western blot analysis (Fig. 4C). However, when the chimera was expressed in a sla1Δ strain, some fluorescence accumulated at the PM in a non-polarised fashion since it cannot be endocytosed. This suggests that this chimera reaches the PM in the WT strain, but it is actively removed from this location and degraded (Fig. 4B). It could be that the chimeric protein is recognised by the Golgi-localised quality control mechanism for transmembrane proteins, which involves the ubiquitin ligase Tul1 (Reggiori and Pelham, 2002) and the Bsd2 protein (Hettema et al., 2004). Expression of Sso1(Kre6) in a bsd2Δtul1Δ mutant, results in localisation of the chimera at the PM, with a polarised distribution (Fig. 4B). This distribution in the bsd2Δtul1Δ strain confirms that the protein undergoes endocytosis when it reaches the PM. In a WT strain, the protein may again be recognised by the quality control machinery when it cycles through the Golgi, resulting in complete vacuolar localisation. Most of the protein in the sla1Δ mutant reaches the vacuole, presumably directly from the Golgi. Taken together, these results indicate that the Kre6 TMD is able to dictate endocytosis.

Analysis of the TMDs from the chimeric proteins Sso1₁₁–Sft1₁₁, Sso1(Kre6) and Sso1(Hkr1), showed that, besides having increased volumes in their exoplasmic hemi-TMDs compared with Sso1, they also have increased hydrophobicity. Hydrophobicity was calculated using the GES hydrophobicity scale, which specifically addresses hydrophobicity of amino acids in α-helices such as TMDs (Engelman et al., 1986) (Fig. 5A). To discriminate if it is the hydrophobicity, the volume or both properties of the exoplasmic hemi-TMDs, acting as an endocytic signal, we generated the artificial construct Sso1₁₁–hvlh₁₁ (where hvlh is high volume, low hydrophobicity), in which the three phenlyalanines of the exoplasmic hemi-TMD of Sso1₁₁–Sft1₁₁ have been replaced by two tryptophans and a leucine residue, which are less hydrophobic according to the GES scale. This results in a hemi-TMD with high volume, but with a hydrophobicity similar to that of the Sso1 hemi-TMD (Fig. 5A,B). The construct was transformed into yeast and analysed by fluorescence microscopy. Fig. 5C shows that this protein is polarised in the PM of WT yeast cells, the polarity is lost in the sla1Δ strain and it is internal in the ric1Δ strain. These data suggest that the endocytosis determinant is the volume and not the hydrophobicity of the hemi-TMDs.

To understand how these high-volume hemi-TMDs are endocytosed, we looked at the polarity of Sso1₁₁–Sft1₁₁ in all early endocytic mutants as defined by Brach et al. (2014) and found no effect (not shown). We also tested Yap1801 and Yap1802, the specific adaptors for Snc1 (Burston et al., 2009). Accordingly, the double mutant strongly affected the polarity of Snc1 but not that of Sso1₁₁–Sft1₁₁ (Fig. S4).

Many yeast PM proteins are modified with ubiquitin, resulting in their endocytosis (reviewed in Dupré et al., 2004). This modification is carried out by the ubiquitin ligase Rsp5. Since Rsp5 is essential for viability, we tested a strain carrying a hypomorphic rsp5 allele (Nikko et al., 2008). Fig. 6A shows that in this mutant, the polarity of Sso1₁₁–Sft1₁₁ is diminished, whereas Snc1 is fully polarised. Quantification of fluorescence images indicates that the amount of Sso1₁₁-Sft1₁₁ fluorescence at the mother cell (rear and centre) almost doubles for the rsp5 mutant strain compared with the WT (Fig. 6B). In the rsp5 mutant, partial loss of polarity is also observed for Sso1₁₁–Sft1_11(scr) and the Sso1(Hkr1) chimera, whereas Sso1(Kre6) can be detected at the PM, which is indicative of diminished endocytosis (Fig. 6A). Cells were also stained with FM4-64 to label vacuoles. All Sso1 chimeras show fluorescence in the vacuole lumen to different degrees in the WT strain. This fluorescence accumulates in the vacuole membrane in the rsp5 mutant (Fig. 6A), since this protein is also required for the vacuolar degradation of numerous transmembrane proteins (reviewed in Belgareh-Touzé et al., 2008).

To test if endocytosis of Sso1₁₁–Sft1₁₁ is due to Rsp5-mediated ubiquitylation, we generated mutant versions of Sso1 and Sso1₁₁–Sft1₁₁ that are devoid of lysine residues, and therefore cannot be ubiquitylated (Δlys). Immunofluorescence followed by confocal microscopy observations of a WT strain transformed with these mutants, indicates that Sso1₁₁–Sft1₁₁Δlys is fully polarised whereas Sso1Δlys is still homogeneously distributed in the PM (Fig. 6C), indicating that the effect of Rsp5 on the endocytosis of the high-volume exoplasmic hemi-TMD proteins is indirect. It is possible that a protein that interacts with or is modified by Rsp5, is responsible for Sso1₁₁–Sft1₁₁ endocytosis. Members of the yeast family of arrestins control the levels of membrane transporters by linking Rsp5 to its PM substrates (Nikko and Pelham, 2009; Nikko et al., 2008). Analysis of a strain lacking nine arrestin-like adaptors [art(1-8)Δ,art10Δ] showed that the polarity of Sso1₁₁–Sft1₁₁ is unaffected (data not shown).

To test whether TMDs with high-volume exoplasmic halves could act as endocytic signals in mammalian cells, we transfected CHO-K1 cells with mammalian vectors expressing yeast Sso1 and Sso1₁₁–Sft1₁₁ fused to mCherry, and assessed the steady state localisation of these constructs by confocal microscopy. Yeast Sso1 has been shown to localise almost exclusively to the PM of CHO-K1 cells (Quiroga et al., 2013) (Fig. 7A). Sso1₁₁–Sft1₁₁ does reach the PM but is also localised to intracellular structures. Co-localisation experiments with Rab11–GFP (recycling endosome marker) and Rab5–GFP (early endosome marker) indicate that the intracellular staining from Sso1₁₁–Sft1₁₁ corresponds to both the recycling endosome and early endosomes (Fig. 7A). Recycling endosome localisation was also confirmed by fluorescent transferrin-uptake experiments (not shown). Co-localisation experiments were quantified and Pearson correlation coefficients were calculated (Fig. 7B), confirming that a substantial amount of Sso1₁₁–Sft1₁₁ is present in endosomal structures, which is suggestive of the protein being endocytosed.

To confirm that Sso1₁₁–Sft1₁₁ is actively endocytosed from the plasma membrane, we carried out dynamic fluorescence loss in photobleaching (FLIP) experiments. Cells expressing either mCherry–Sso1 or mCherry–(Sso1₁₁–Sft1₁₁) were repeatedly photobleached at the recycling endosome region (labelled with fluorescent transferrin) and imaged for up to 60 min in the presence of cycloheximide. Quantification of the fluorescence intensity at the plasma membrane indicates that up to 70% of Sso1₁₁–Sft1₁₁ is lost from this organelle during the experiment, whereas Sso1 remains stable (Fig. 7C).

While the length of TMDs has long been recognised as a protein sorting signal, the volume of exoplasmic hemi-TMDs has only recently been incorporated as a novel determinant in membrane protein localisation, more specifically in Golgi retention (Quiroga et al., 2013; Sharpe et al., 2010). We previously showed that single-spanning membrane proteins with long TMDs are transported to the PM, regardless of their geometry (Quiroga et al., 2013). Here, we show that the geometry of these long TMDs can also function as a sorting determinant at the PM, in this case, high-volume exoplasmic hemi-TMDs act as an endocytosis signal, both in yeast and in mammalian cells. In yeast, this signal also results in a polarised distribution. These findings were obtained using a chimeric TMD (Sso1₁–Sft1₁₁) and were validated with endogenous TMDs (Hkr1 and Kre1). Interestingly, Mercanti et al. (2010) showed that proteins with long TMDs were effectively excluded from clathrin-coated pits. We analysed the TMDs of the proteins used in those experiments, and they all correspond to low-volume exoplasmic hemi TMDs (not shown). A comprehensive analysis of type II and type I membrane proteins (Quiroga et al., 2013) shows that there is a tendency for plasma membrane proteins to have low-volume exoplasmic hemi-TMDs; however, the distribution is broad and a significant number of PM proteins with high-volume exoplasmic hemi-TMDs exists (Quiroga et al., 2013). In this work, we have experimentally demonstrated that the sorting of type II membrane proteins may be determined by the volume of their exoplasmic hemi-TMDs. The relevance of this sorting signal for type I membrane proteins remains to be assessed.

Increasing the volume of the residues that form an exoplasmic hemi-TMD may also result in an increase in its hydrophobicity. Using an artificial TMD with high-volume residues, but a relatively low hydrophobicity (comparable to that of low-volume Sso1 exoplasmic hemi-TMD), we were able to establish that it is the volume of this domain and not the hydrophobicity that dictates endocytosis.

To get an insight into the mechanism by which these high-volume TMDs are endocytosed, we tested a variety of known endocytosis-adaptor mutants, but we were unable to find any adaptors that were specifically required for the endocytosis of this type of cargo. The only protein that showed a moderate but specific effect on the internalisation of high-volume exoplasmic hemi-TMDs, was the ubiquitin ligase Rsp5. In the presence of a hypomorphic allele of this gene, the polarity of our reporters is reduced by ∼50%. We showed that the effect of Rsp5 in the endocytosis of these constructs is indirect, since a version of Sso1₁₁–Sft1₁₁ that is devoid of lysine and therefore not able to be ubiquitylated, is still polarised. How then does Rsp5 affect the polarity of our constructs? General effects of Rsp5 on endosomal traffic, which could alter the speed at which Sso1₁₁–Sft1₁₁ is recycled to the PM are unlikely, since Snc1 is fully polarised in the rsp5 strain. A general decrease in the speed of recycling would lead to internal staining of Snc1 and a less polarised localisation. This also indicates that Snc1 is efficiently endocytosed in the rsp5 mutant. Additionally, the Sso1(Kre6) construct, which in a WT strain is vacuolar and completely absent from the PM, can be observed at the PM in the hypomorphic rsp5 strain, as in a sla1Δ strain, suggesting that it is endocytosis of these particular constructs that is diminished in these conditions. A general role for Rsp5 activity in endocytosis that is independent of the ubiquitylation status of the cargo has been proposed (Dunn and Hicke, 2001) but it does not appear to affect Snc1 endocytosis. Finally, we cannot rule out the possibility that endocytosis of Sso1₁₁–Sft1₁₁ is to some extent independent of Rsp5. Since the experiments were carried out in conditions in which Rsp5 activity is reduced but not eliminated, we cannot discriminate whether the remaining Rsp5 activity is sufficient to confer a polarised localisation to Sso1₁₁–Sft1₁₁ or if the remaining polarity is attained through an Rsp5-independent mechanism.

How these high-volume exoplasmic hemi-TMDs are recognised by the endocytic machinery is interesting since most known endocytic adaptors are cytosolic or recognise cytosolic signals. It is possible that a receptor that specifically recognises this type of TMDs exists and leads Sso1₁₁–Sft1₁₁ into the endocytic pathway in an Rsp5-dependent manner. A receptor that recognises features of TMDs has been implicated in ER exit (Herzig et al., 2012).

An interesting possibility is that high-volume exoplasmic hemi-TMDs mediate endocytosis by determining the partition of proteins into a specific plasma membrane sub-domain that is active in endocytosis. The endogenous Hkr1 and Kre6 TMDs would naturally partition into this domain. The existence of subdomains in the yeast PM that are either protected or active in endocytosis has been postulated, but it is not devoid of controversy. The membrane compartment of Can1 (MCC) subdomain was proposed to be inactive in endocytosis (Grossmann et al., 2008). Brach et al. (2011) confirmed that this compartment is indeed inactive in both endocytic and exocytic traffic, but they postulated that localisation of proteins in this domain would not protect them from endocytosis since the proteins are able to diffuse out of the MCC. Recently, numerous co-existing domains in the yeast plasma membrane were identified (Spira et al., 2012) whose nature and functions are not well understood.

It has been shown for mammalian SNARE proteins, that the hydrophobic mismatch between the length of TMDs and the thickness of the bilayer can induce clustering and segregation of proteins to subdomains in model membranes (Milovanovic et al., 2015). In our experiments, the TMD length of Sso1 and its chimeras was maintained, so the effects are only due to the volume of the exoplasmic hemi-TMD. In vivo, hydrophobic mismatch probably also plays a part, since shorter TMDs are prone to endocytosis (Mercanti et al., 2010).

Finally, we showed that a high-volume exoplasmic hemi-TMD could function as an endocytosis signal in mammalian cells, suggesting that this may be a conserved endocytosis signal. Bioinformatics analyses of mammalian single-spanning membrane proteins that are annotated as residents of the plasma membrane (from the dataset generated by Quiroga et al., 2013) indicate that up to 14% have long transmembrane domains (over 20 residues) with high-volume exoplasmic hemi-TMDs (over 145 ų) (not shown).

The identification of this novel endocytosis determinant in the shape of the exoplasmic hemi-TMDs shows a new way in which endocytic cargo can be recognised that relies on a different fundamental principle from previously known endocytic signals, and thus adds novel insights into the mechanism of endocytosis. Finally, the idea that both the shape and length of a TMD determine protein sorting not only at the Golgi but also at the level of the PM, suggests that this is a general sorting mechanism that operates in several transport steps in the cell.

sla1Δ, ric1Δ and deletions of early endocytic adaptors are from the EUROSCARF consortium. The bsd2Δtul1Δ strain was described in Hettema et al. (2004); yap1801Δ yap1802Δ strain was a gift from Dr Elizabeth Conibear (Centre for Molecular Medicine and Therapeutics, Child and Family Research Institute, Vancouver, University of British Columbia, Vancouver, Canada); rsp5 hypomorph (EN44) and art(1-8)Δart10Δ (EN60) strains were gifts from Dr Hugh Pelham (MRC Lab. of Molecular Biology, Cambridge, UK.).

Sso1₁₁–Sft1₁₁ was described in Quiroga et al. (2013). To generate Sso1₁₁–Sft1_11(scr), Sso1(Hkr1), Sso1(Kre6), Sso1₁₁–Mid2₁₁ and Sso1₁₁–HVlv₁₁, synthetic DNA fragments were ordered from Genscript (NJ, USA) and introduced by homologous recombination in plasmid pJV572. All these constructs are tagged at the N-terminus with GFP. The TMDs predicted for Kre6 differ slightly according to the software used for the prediction, indicating that the borders of this TMD are not easy to determine, but all of them produce a high-volume exoplasmic hemi-TMD. Therefore, the chimera was generated with a few amino acids to the sides, to allow insertion in the native configuration. To generate Sso1 and Sso1₁₁–Sft1₁₁ devoid of lysines, synthetic DNA fragments were ordered from Genscript (NJ, USA) and introduced into yeast vectors with 5×His tags at the N-terminus. All constructs are in centromeric plasmids and driven by either TPI1 or GAL1 promoters.

For mammalian expression of Sso1 and Sso1₁₁–Sft1₁₁, the coding regions of these proteins were amplified by PCR and cloned in frame into pMCherry-C1 (Clontech Laboratories). GFP–Rab11 and GFP–Rab5 expression vectors were gifts from Dr José Luis Daniotti, Centro de Investigaciones en Química Biológica de Córdoba (CIQUIBIC), CONICET, Universidad Nacional de Córdoba, Córdoba, Argentina.

For western blotting, the following antibodies were used: anti-GFP ROCHE (Life Sciences, 11814460001; 1:1500) and anti-Pgk1 (Molecular Probes; 1:2000). Secondary antibodies were: anti-mouse IRDye 800 (Li-Cor, 51- 926-32210; 1:20,000). For immunofluorescence, antibodies used were: anti-His (Life Technologies, 37-2900; 1:500), anti-HA (Sigma-Aldrich, H9658; 1:400). Secondary antibodies were: anti-mouse IgG Alexa Fluor 488 (Molecular Probes, A11029; 1:1500) and anti-mouse IgG Alexa Fluor 546 (Invitrogen, A21045; 1:1500).

The protein sequences of S. cerevisiae were retrieved from UniProt (UP000002311), and fed to Phobius (Käll et al., 2004) for transmembrane topology prediction. Single-spanning membrane proteins were selected. The transmembrane region of proteins predicted to be type II was inverted to match those of type I proteins topologically. The average volume of the amino acids in the exoplasmic half of each TMD was calculated using the amino acid volume scale TSAJ990102 (Tsai et al., 1999). When the predicted TMD had an odd number of amino acids, the central amino acid of the TMD was excluded from the average. The ‘cellular component’ fields were retrieved from UniProt, to assign a subcellular localisation for each protein. For PM proteins, the list was manually curated from the literature (Table S1).

For GFP- or mCherry-tagged proteins under the control of GAL1 promoter, cells were grown overnight in lactate as the sole carbon source, diluted and induced for 3 h in medium containing galactose. Cells were imaged live with an Olympus FV300 confocal microscope equipped with PLANAPO N 60× oil 1.4 numerical aperture objective or with a FV 1200 confocal microscope equipped with PLAPON 60× oil SC2 1.4 numerical aperture objective. For 6× His tagged constructs, immunofluorescence microscopy was carried out as described in Quiroga et al. (2013). Vacuolar staining was obtained by incubating cells with 20 µM FM4-64 (Sigma-Aldrich) for 15 min at 30°C; the cells were washed twice with YPD medium and incubated for 30 min in the same medium before visualisation. Quantification of polarity was performed by measuring fluorescence intensities at the bud tip, and centre and rear of the mother cell using the same ROI. The intensities were quantified using Fiji ImageJ software (NIH) from background-subtracted images. Values were normalised by the intensity at the bud tip for each cell. For fully polarised cells, we visualised the PM of the mother cells by co-expressing mCherry–Sso1. Quantifications show the mean of at least 20 cells for each experimental condition.

For co-localisation of the intracellular structures marked by GFP–(Ssso₁₁-Sft1₁₁), at least 150 GFP-positive dot-like structures were analysed and the number that was also positive for the different organelle markers, Sec7–MARS, Vps8–mCherry or Mnn9–MARS, was quantified.

CHO-K1 cells (ATCC, Manassas, VA, USA) were grown and maintained at 37°C in 5% CO₂ in DMEM (Dulbecco's modified Eagle's medium; Invitrogen) supplemented with 10% (v/v) FBS and antibiotics. Transfections were carried out with 1μg plasmid DNA per 35-mm-diameter dish using PEI (polyethyleneimine) (Sigma-Aldrich). Cells were analysed 16 h post transfection.

Cells grown on coverslips and transfected with fluorescent constructs were washed with PBS and fixed in 4% (w/v) paraformaldehyde for 10 min at room temperature. The cells were washed with PBS and mounted in FluorSave Reagent (Calbiochem/EMD Bioscience, MO, USA). Cells were imaged with an Olympus FV1200 confocal microscope equipped with PLAPON 60× oil SC2 1.4 numerical aperture. Pearson's correlation coefficients were calculated using the Fluorescent Correlation Analysis ImageJ plugin (NIH) with at least 10 cells for each experimental condition using background-subtracted images.

For FLIP experiments, CHO-K1 cells were transfected with fluorescent constructs and incubated for 16 h to allow for protein expression. Afterwards, 50 µg/ml of cycloheximide (Sigma-Aldrich) were added. Finally, cells were transferred to an Olympus FluoView FV1000 confocal microscope. Pre- and post-bleaching images were acquired every 6 min using a 63× PLAPON 1.4 numerical aperture objective for 1 h. Bleaching was carried out on the recycling endosome region using a 5-μm-diameter circular region. The average fluorescence intensities of the PM, pre-and post-bleaching were measured with ImageJ software. Values were normalised by the initial intensity at the PM.

Results are presented as means±s.e.m. Statistical analyses were performed using Student's t-test or ANOVA with GraphPad Prism 5.00 software. Significance was attributed at the 95% level of confidence (P<0.05).

We thank Juan José Nicola for writing the scripts for automated bioinformatics analysis and Dr Pablo Aguilar for critically reading the manuscript. We also like to thank the reviewers for their comments, which substantially improved this article.