Getting membrane proteins on and off the shuttle bus between the endoplasmic reticulum and the Golgi complex

Twenty to thirty years ago, a handful of seminal studies (Wieland et al., 1987; Mizuno and Singer, 1993; Nishimura and Balch, 1997; Martínez-Menárguez et al., 1999) set the stage for the still ongoing debate on the role of bulk flow versus receptor-mediated transport of cargo molecules through the secretory pathway of eukaryotic cells. Bulk flow is the process by which cargo passively distributes between the donor compartment and the transport vesicles it generates, resulting in equal cargo concentration within these two compartments. By contrast, signal-mediated transport results in concentration of cargo within transport carriers; here, discrete export signals on the cargo are recognized and captured by specific receptors that are concentrated at sites of vesicle budding.

Cargo capture by receptors is, of course, undisputed (reviewed in Lee et al., 2004; Dancourt and Barlowe, 2010); the issue of debate is the relative importance of this process, that is, the proportion of receptor-dependent cargoes and the extent of their dependency (i.e. the extent of receptor-mediated acceleration of the transport of receptor-dependent cargoes). Given the rather small number of cargo receptors that have been identified (Herzig et al., 2012) in comparison to the thousands of cargo molecules that must be transported and the lack of any identified export signals on many of them, it appears likely that many cargo proteins reach their destination without the help of specific receptors (Thor et al., 2009).

In addition to bulk flow and cargo capture, a third transport mechanism, which is less often considered but is relevant to cargo membrane proteins, is that involving partitioning within the lipid bilayer (Lippincott-Schwartz and Phair, 2010; Hanulová and Weiss, 2012). This process depends on the general physico-chemical features of the cargo membrane protein and on the interactions of these features with the collective properties of the bilayer, instead of the one-to-one intermolecular interactions that exist between discrete signals and their receptors. Transport by partitioning is distinct from both bulk flow and receptor-mediated transport, because differently from bulk flow, cargo can be concentrated in transport carriers, yet the concentrating process does not depend on specific cargo receptors (Fig. 1 and Box 1).

Partitioning of membrane proteins is thought to play an important role in their sorting between the Golgi complex and post-Golgi compartments (Sharpe et al., 2010; Holthuis and Menon, 2014). Whether a similar phenomenon operates at the interface between the ER and the Golgi has been less clear and will be discussed in the first part of this Commentary.

Transport through the secretory pathway is bidirectional, with retrograde transport carriers required to maintain the size of the different compartments in the face of forward flux, as well as to recycle components of the transport machinery and donor-compartment residents that have escaped (Spang, 2013). As for export from the ER, the relative importance of cargo capture, bulk flow and partitioning must also be considered for retrograde transport. This raises the question of whether forward-directed cargoes escape from retrograde flux, and if so, how do they achieve this? This will be the subject of the second part of the article.

The Golgi complex is an organelle at the boundary between two distinct membrane territories of the eukaryotic cell; a region of thin, cholesterol- and sphingolipid-poor bilayers, with neutral cytoplasmic surface charge and loosely packed phospholipids, and a region of thick, cholesterol- and sphingolipid-rich, bilayers with tight lipid packing and negative cytoplasmic surface charge (Bigay and Antonny, 2012; Holthuis and Menon, 2014). These two territories comprise the early and distal secretory pathways, respectively. A systematic analysis of bitopic (single-spanning) membrane proteins of the secretory pathway has revealed that the transmembrane domains (TMDs) of proteins that localize to the early secretory pathway (ER–Golgi) are, on average, shorter than those of proteins located in post-Golgi (distal secretory pathway) compartments (∼20 vs ∼24.5 residues, respectively, in vertebrates), indicating that TMD length is optimized to match the thickness of the bilayer of their residence (Sharpe et al., 2010). Artificially lengthening the TMD of Golgi-localized glycosyl transferases or shortening of the TMD of a plasma-membrane-resident protein leads to a relocation of these proteins to the plasma membrane and Golgi complex, respectively, in line with the idea that partitioning of TMDs into bilayer domains with matching thickness governs the sorting of membrane proteins at the Golgi–post-Golgi interface (Munro, 1995).

As the ER and the cis-Golgi (and presumably the ER–Golgi intermediate compartment, ERGIC) are thought to be part of the same membrane territory, it is more difficult to envisage a role of partitioning in sorting at the ER–Golgi interface. Although a role for TMD length in sorting between the ER and the Golgi is now widely recognized, the mechanism underlying this TMD-dependent sorting is less clear. Below, I will first summarize published examples of TMD-dependent sorting at the ER–Golgi interface and then discuss possible mechanisms underlying this phenomenon.

The initial step of export from the ER is mediated by coat protein complex II (COPII)-coated vesicles. The COPII coat minimally comprises the small GTPase Sar1 and the two protein complexes Sec23–Sec24 and Sec13–Sec31. Sar1, activated by GTP, recruits the Sec23–Sec24 complex to specialized sites of the ER, called ER exit sites (ERES); Sec23–Sec24 in turn recruits the Sec13–Sec31 complex, which, by polymerizing to form a cage-like oligomeric lattice (Stagg et al., 2008), bends the membrane of the emerging vesicle (reviewed in Springer et al., 1999; Lee et al., 2004; Sato and Nakano, 2007). The Sec24 subunit of the coat recognizes export signals located in the cytosolic tail of cargo membrane proteins or cargo receptors, thereby coupling cargo sorting to vesicle budding (reviewed in Dancourt and Barlowe, 2010). The question is whether additional mechanisms, other than the interaction with COPII, underlie the loading of cargo into these transport vesicles.

In 1993, Lankford and co-workers demonstrated that shortening the TMD of the CD3-ε polypeptide (encoded by CD3E) conferred the capacity to retain chimeric constructs in the ER and, conversely, that a mutated form of the Tac antigen (also known as interleukin-2 receptor), which is retained in the ER, could be exported upon lengthening of its TMD (Lankford et al., 1993). More recently, similar observations have been made for the secretory membrane protein vesicular stomatitis virus glycoprotein (VSVG) (Dukhovny et al., 2009).

Does the short TMD of bona fide ER-resident membrane proteins play a role in preventing their export? That this is indeed the case was first demonstrated for cytochrome b5 (Pedrazzini et al., 1996; Bulbarelli et al., 2002), and subsequently for ubiquitin-conjugating enzyme UBC6 (Yang et al., 1997), as well as for the yeast ER-resident target (t)-SNARE Ufe1p (Rayner and Pelham, 1997). These three proteins are tail-anchored, defined by anchorage to the bilayer through a C-terminal TMD and by the absence of a lumenal domain (reviewed in Borgese and Fasana, 2011). None of these proteins possess any known COPII-interacting motifs. Lengthening the tail anchor by four or five hydrophobic residues results in export of the modified proteins from the ER.

Subsequent studies have demonstrated a role for the TMD, often in combination with other features, in the retention of ER-resident membrane proteins with a variety of different topologies (see Box 2). General physico-chemical features of the TMD of these proteins, such as length or hydrophobicity, rather than specific sequences are involved in the sorting process. In addition, post-translational modifications, such as palmitoylation, may affect the way in which the TMD interacts with the bilayer, and consequently its trafficking (Lam et al., 2006; Morozova and Weiss, 2010; Abrami et al., 2008; Chum et al., 2016).

As ER residency is based on two distinct processes – the exclusion of resident proteins from ERES and the retrieval through retrograde transport of those residents that have escaped from the ER (Spang, 2013; Geva and Schuldiner, 2014) – either one of these processes could, in principle, be influenced by TMD length. Indeed, Rer1p, a cargo receptor involved in retrograde transport, recognizes the TMD of escaped membrane proteins of the ER (Sato et al., 2003), and the artificially lengthened TMD of cytochrome P450 2C1 (see Box 2) interferes with retrieval of the altered protein from the Golgi (Szczesna-Skorupa and Kemper, 2000). However, for a number of the proteins listed in Box 2, the short TMD prevents their entry into ERES, a phenomenon known as static retention; by contrast, longer TMDs are recruited to sites of vesicle budding (Ronchi et al., 2008; Dukhovny et al., 2009; Vander Heyden et al., 2011). Below, I discuss some possible mechanisms underlying this TMD-dependent sorting within the ER.

As discussed above, there do not seem to be any large differences between the physical characteristics of the bilayers of the ER and cis–medial Golgi compartments; hence, the idea of a cargo receptor that recognizes the long TMDs of proteins that are destined to the TGN and to post-Golgi compartments is appealing. One single receptor could recognize a multitude of secretory membrane proteins, limiting the number of receptors needed to recruit the numerous and diversified post-Golgi-directed cargoes to ERES. The conserved Erv14-cornichon family of proteins (Erv14, cornichon and CNIH proteins in yeast, Drosophila and mammals, respectively) might represent such receptors. Clients of these receptors are exclusively transmembrane proteins (usually polytopic) that reside in the late secretory pathway and have long TMDs (Castillon et al., 2009; Herzig et al., 2012; Pagant et al., 2015 and references therein). Detailed analysis of one client supported the idea that Erv14 is a general receptor for long-TMD-bearing transmembrane proteins (Herzig et al., 2012). Interestingly, some Erv14 targets, such as Saccharomyces cerevisiae Yor1 (Pagant et al., 2015) and VSVG (Simpson et al., 2007), also bear discrete COPII-interacting signals, suggesting that the dual recognition by Erv14 and Sec24 underlies their efficient recruitment to ERES (Pagant et al., 2015).

Notwithstanding the interesting results concerning the Erv14-cornichon family, these cargo receptors cannot account for the export of all the many membrane proteins that lack COPII-interacting signals. Indeed, only about one-third of the yeast plasma membrane proteome is Erv14-dependent, and thus far, no other cargo receptors that could be responsible for the export of Erv-independent proteins have been identified. In particular, the TMD-dependent sorting of tail-anchored proteins at the ER–Golgi interface is independent of Erv14 (Herzig et al., 2012; Pagant et al., 2015). Furthermore, deletion of Erv14, either alone or in combination with that of its paralog Erv15, does not abolish export of Erv14 clients completely, pointing to other transport mechanisms.

Although it is possible that additional cargo receptors that are sensitive to TMD length will be discovered in the future, the failure to identify these thus far suggests that alternative receptor-independent mechanisms have an important role in TMD-dependent sorting at the ER–Golgi interface. Furthermore, the involvement of TMD-dependent cargo receptors in the export of long TMDs does not explain the exclusion of short TMDs from ERES, a process that presumably does not depend on specific ligand–receptor interactions (Ronchi et al., 2008). Although lumenal proteins could be retained in the ER through inclusion into a gel phase that would be excluded from budding transport vesicles (Pfeffer and Rothman, 1987; Tatu and Helenius, 1997), such a mechanism does not apply to a number of retained ER-resident membrane proteins, which are freely mobile and diffuse at rates that are incompatible with their inclusion in a matrix or in large oligomeric complexes (Szczesna-Skorupa et al., 1998; Snapp et al., 2003; Ronchi et al., 2008; Vander Heyden et al., 2011).

One hypothesis for the mechanism of exclusion of ER-resident proteins from ERES invokes the notion of crowding; here, budding vesicles are considered to be so full of captured cargo as to leave no room for proteins lacking export signals. Although proteins with bulky lumenal domains might be excluded from crowded transport vesicles, as reported for glycosylphosphatidylinositol-anchored proteins (GPI-APs; Bhagatji et al., 2009), this possibility is less likely for small proteins that lack a lumenal domain. A positively curved budding vesicle has a larger surface area on the cytosolic side of the bilayer than on the lumenal side, so it is difficult to envisage how crowding could inhibit the entry of a small tail-anchored protein such as cytochrome b5 into ERES (Ronchi et al., 2008). In a highly crowded transport vesicle – the synaptic vesicle – quantitative proteomics reveals that about 20% of the bilayer area is occupied by TMDs (Takamori et al., 2006), suggesting there would be room for a small tail-anchored protein even in such a protein-rich trafficking organelle. The idea of crowding-based exclusion (Pentcheva et al., 2002) also appears to be irreconcilable with recognized bulk flow phenomena (Thor et al., 2009) and with the observation that removal of identified export signals (Fiedler et al., 1996; Nishimura and Balch, 1997; Iodice et al., 2001) or deletion of cargo receptors (Herzig et al., 2012; Pagant et al., 2015) usually only results in a slowdown of cargo export, but not its complete inhibition.

The above considerations leave the partitioning of proteins with different TMD lengths into different bilayer domains as the most attractive hypothesis to explain their sorting behavior. Partitioning could explain both the exclusion of short TMDs from ERES and the recruitment of long TMDs into transport vesicles.

Below, I consider possible scenarios that could account for such a partitioning-based sorting, keeping in mind that (i) the TMDs of plasma-membrane-directed proteins are positively mismatched with respect to the width of the ER membrane [by∼6 Å in vertebrates (Sharpe et al., 2010; Killian, 1998; Jensen and Mouritsen, 2004; see Box 1)] so that, if possible, they will partition out of the thin ER bilayer into thicker better matched bilayer domains; and (ii) that the ER and the cis- and medial-Golgi belong to the same lipid territory (Sharpe et al., 2010; Holthuis and Menon, 2014), raising the question as to how partitioning could underlie sorting between two organelles that are endowed with similar bilayer properties.

One way to diminish the energetic cost of mismatched TMDs is to reduce the contact area between the protein and lipids by generating clusters of the mismatched protein (Ren et al., 1999; Sparr et al., 2005; Schmidt et al., 2008). Oligomers might preferentially partition into curved membrane domains, as found in budding transport vesicles. In agreement with this idea, a recent study has demonstrated that oligomerization facilitates packaging of a membrane protein into COPII vesicles in the absence of any known export signal (Springer et al., 2014).

In the case of lumenally oriented oligomers, the preference for curved membrane domains might be explained by the compressive pressure acting at the lumenal leaflet of the bilayer when the membrane bends to generate a bud (discussed in Stachowiak et al., 2013; Springer et al., 2014). However, it is more difficult to envisage how the oligomerization of cytosolically oriented mismatched proteins could result in a similar partitioning phenomenon. Ronchi et al. (2008) observed that a newly synthesized mismatched tail-anchored protein was distributed both to ERES and to the tubular portions of the ER, whereas it was excluded from sheets, suggesting that also in this case partitioning to curved membranes might play a role in the sorting process. However, using a model system, Fossati et al. (2014b) did not observe any preferential recruitment of the same mismatched tail-anchored protein to tubules extracted from giant unilamellar vesicles of defined composition.

The ER is the site of synthesis of essentially all secretory pathway lipids or of their precursors. Phospholipids and cholesterol are generated in the ER, where cholesterol concentrations are kept low, as most of this sterol is exported to the post-Golgi membrane territory (van Meer et al., 2008). Likewise, the simplest sphingolipid, ceramide, is assembled in the ER, before it is transported to the Golgi, where it serves as precursor to sphingomyelin and glycolipids (Futerman and Riezman, 2005).

Recent work has revealed that lipid transport from the ER occurs both through transport vesicles and through cytoplasmic lipid-transfer proteins, many of which operate at membrane contact sites between the ER and downstream compartments of the secretory pathway (Holthuis and Menon, 2014). The importance of the latter mechanism does not, however, preclude the possibility of the selective enrichment of specific lipids into budding transport vesicles, as has been directly demonstrated in the case of COPI-coated vesicles generated from Golgi cisternae (Brügger et al., 2000; Contreras et al., 2012). A simple idea, then, is that transport vesicles that originate from the ER might be enriched in lipids destined to the thick membranes of the post-Golgi territory, thereby creating an environment that is suitable for long TMDs. There are indeed some observations that suggest that this could be the case. For example, acidic phospholipids have key roles in the formation of COPII-coated carriers (Pathre et al., 2003; Blumental-Perry et al., 2006). Moreover, concentrative sorting of phosphatidylserine with long saturated acyl chains into transport vesicles generated in vitro from plant microsomes was reported some years ago (Sturbois-Balcerzak et al., 1999), and depletion of cholesterol from the ER inhibits COPII-dependent transport (Ridsdale et al., 2006; Runz et al., 2006).

An extensively investigated example of the involvement of a specific lipid class in protein transport from the ER is the role of ceramide in the export of GPI-APs. GPI-APs constitute a particular category of secretory membrane cargo that contain an exoplasmic domain attached to the external leaflet of the plasma membrane through a glycolipid anchor (Orlean and Menon, 2007; Pittet and Conzelmann, 2007). To be recruited into COPII vesicles, these proteins need a cargo receptor, which is supplied by members of the p24 family (reviewed in Kinoshita et al., 2013). In addition, yeast GPI-APs require ongoing ceramide synthesis to exit the ER (Horvath et al., 1994; Sutterlin et al., 1997; Watanabe et al., 2002). Furthermore, newly synthesized, remodeled GPI-APs are segregated from other membrane cargoes and can be recovered in a cold-detergent-insoluble fraction (Bagnat et al., 2000; Castillon et al., 2009).

In yeast, like in mammals, ceramide is transported from the ER to the Golgi both by non-vesicular mechanisms and by vesicular COPII-dependent mechanisms (Funato and Riezman, 2001; Hanada et al., 2003; Kajiwara et al., 2014). The extensive work in the yeast system suggests that the COPII-dependent pathway of ceramide export is coupled to GPI-AP biogenesis, as correct sorting of these two classes of molecules appears to be mutually interdependent (Kajiwara et al., 2008; Loizides-Mangold et al., 2012). These findings illustrate how a cargo membrane protein can affect the lipid composition of the vesicles that carry it.

The possibility that different interactions with lipids could underlie TMD-dependent sorting of tail-anchored proteins was tested by Ceppi et al. in a model system comprising liposomes composed of a matrix of an abundant ER lipid (palmitoyl-oleyl phosphatidyl choline) supplemented with a higher-melting-temperature lipid (hypothesized to be exported in transport vesicles – ceramide or phosphatidylserine) and with either of two forms of cytochrome b5 – the ER-resident wild type or a mutant with a lengthened TMD (Ceppi et al., 2005). Differential scanning calorimetry and fluorescence measurements revealed that the lengthened but not the wild-type TMD preferentially partitioned into a lipid phase that was enriched in the higher-melting-temperature lipid (ceramide or phosphatidylserine).

These results are in line with the idea that protein–lipid interactions at the ER–Golgi interface could be the basis for TMD-dependent partitioning. It is worth mentioning that the recruitment of ‘non-ER’ lipids into COPII vesicles would not necessarily alter the composition of the acceptor compartment (ERGIC or cis-Golgi) because these lipids could be rapidly moved to downstream compartments through lipid exchange proteins or vesicular transport. How the recruitment of non-ER lipids at ERES could be achieved is discussed in the next section.

As summarized above, a sizeable portion of secretory membrane cargo proteins carry export signals that are recognized by the COPII Sec24 subunit (Ma and Jan, 2002; Dancourt and Barlowe, 2010; Geva and Schuldiner, 2014) or by cargo receptors such as Erv14 (Bökel et al., 2006; Herzig et al., 2012; Pagant et al., 2015). I suggest here that the signal-mediated capture of membrane protein cargoes not only serves the purpose of efficient export of the captured cargoes themselves but that it also plays a much more general role, in that it creates a microdomain in the bilayer into which membrane proteins that are devoid of export signals partition according to the length of their TMD (Fig. 2).

Regardless of whether a membrane protein destined to the plasma membrane carries an export signal or not, its TMD length is, on average, longer than those of ER-resident proteins and is thus expected to be positively mismatched to the ER bilayer (Sharpe et al., 2010). When COPII-interacting membrane cargoes are concentrated at ERES, it is likely that they alter the environment of the bilayer to overcome the energy cost of the mismatch. As summarized in Box 1, studies in model systems have shown that in phase-separated bilayers of heterogeneous composition, TMDs partition into the phase that best matches their length. Additionally, in non-phase-separated bilayers of heterogeneous composition, an integrated TMD can trigger lateral segregation of lipids by recruiting those species that are best matched to its length (Sperotto and Mouritsen, 1993; Tocanne et al., 1994; Killian, 1998; Kaiser et al., 2011). Importantly, even in bilayers of homogeneous composition, positively mismatched TMDs locally increase bilayer thickness by stretching the phospholipid fatty acyl chains. This behavior, initially predicted on theoretical grounds in the well-known mattress model of lipid-protein interactions (Mouritsen and Bloom, 1984), has since been confirmed in simulations (Jensen and Mouritsen, 2004; Venturoli et al., 2005; Schmidt et al., 2008) and experimental studies (Watnick et al., 1990; de Planque et al., 1998; Nezil and Bloom, 1992; Killian, 1998).

The above-described protein-driven lipid sorting, by creating membrane domains in COPII vesicles that are matched to the length of the TMDs they transport, would explain both the recruitment of secretory membrane proteins devoid of signals and the exclusion of ER-resident proteins; these would preferentially partition into the thin bilayer of the bulk ER. Thus, the presence of export signals on a portion of secretory membrane proteins could be of fundamental importance in the sorting of all membrane proteins at the ER–Golgi interface.

Measurements of the rates of secretion of a bulk flow marker indicate that half of the lumenal volume of the ER is drained out by transport vesicles in about 40 min (Thor et al., 2009). Based on the estimated volumes of the ER and of COPII vesicles, the authors calculated that in Chinese hamster ovary (CHO) cells, 155 COPII-coated vesicles bud and carry away ∼0.02% of the ER volume every second. Considering the surface-to-volume ratio of COPII vesicles and of the bulk ER (Griffiths et al., 1984), the proportion of ER membrane being exported is expected to be considerably higher. To maintain organelle homeostasis, this removed material must be replaced by retrograde membrane trafficking from the Golgi back to the ER (reviewed in Lee et al., 2004; Spang, 2013). In addition to ensuring compartment size homeostasis, retrograde transport also guarantees that the vesicular transport machinery, such as vesicle (v)-SNAREs and cargo receptors, as well as any escaped ER-resident proteins, are retrieved to the ER.

Although only one pathway, which is mediated by COPII vesicles, operates in anterograde ER-to-Golgi transport, more than one mechanism is involved in the return trip from the Golgi. The best-characterized pathway is dependent on vesicles coated with the coat protein complex COPI, whereas the other (or others) involve COPI-independent transport carriers.

COPI-dependent retrograde transport has been investigated in great detail, as reviewed recently (Faini et al., 2013; Spang, 2013; Jackson, 2014). As in the case of COPII vesicles, the COPI heptamer is recruited to the Golgi membrane by a small GTPase (Arf1), and has the dual function of bending the membrane and of binding to cargo proteins or receptors through recognition of discrete sorting signals on their cytosolic tails (Jackson et al., 1990; Cosson et al., 1998; reviewed in Dancourt and Barlowe, 2010). In the case of COPI vesicles, inclusion of cargo is essential for the stabilization of the coat (Bremser et al., 1999); thus, retrograde vesicles are expected to form in response to the volume of incoming traffic from the ER.

The COPI-independent route has been characterized to a much lesser extent; it has been implicated in the constitutive recycling of Golgi-resident enzymes between the Golgi and the ER, and in the retrograde transport of Shiga and Shiga-like toxins (Girod et al., 1999; Sengupta et al., 2015), as well as in Golgi-to-ER recycling of at least some plasma-membrane-directed membrane proteins (see next section). This pathway is regulated by Rab6A, a small GTPase of the large Rab family (Martinez and Goud, 1998; Pfeffer, 2013), and is thought to involve tubular rather than vesicular carriers (White et al., 1999). [There are two alternatively spliced forms – Rab6A and Rab6A′ – that might play different roles in trafficking in and out of the Golgi (Echard et al., 2000); hereafter, Rab6 refers to either the first of the isoforms or to the combination of the two.] In addition to its role in Golgi-to-ER retrograde transport, Rab6 has been involved in a number of other trafficking pathways to and from the Golgi (Grigoriev et al., 2007; Storrie et al., 2012; Mallard et al., 2002), and a large number of Rab6-interacting proteins have been identified, (Fernandes et al., 2009; Miserey-Lenkei et al., 2010; Lee et al., 2015; Kano et al., 2009). An exact definition of the different interactions involved in the multiple functions of Rab6 at the Golgi complex has not been attained yet (reviewed in Heffernan and Simpson, 2014).

Why is more than one transport pathway required for Golgi–ER retrograde transport? One possibility is that the cargo-regulated and concentrative COPI pathway, although it returns trafficking machinery and escaped residents to the ER, might not be sufficient to replace the fluid and membrane lipids drained out of the ER by anterograde transport. Fluid could, in principle, also be replaced from the cytosol, but the large surface area of lipid bilayer exported from ERES must be returned by transport carriers. The high surface-to-lumen ratio of tubules would be well suited to take on this task. The roles of tubular and vesicular transport in the retrograde pathway are schematically illustrated in Fig. 3.

If correctly folded soluble and membrane cargoes can leave the ER by bulk flow and partitioning, respectively, what keeps them from entering retrograde transporters once they arrive at the Golgi? Recruitment of anterograde cargoes into retrograde carriers would trap them in a futile ER–Golgi–ER cycle, thus delaying their transit through the Golgi. Does this occur, and if it doesn't, how is it prevented?

At least some soluble secretory proteins, for example, the pro-enzymes of the exocrine pancreas, exit the ER by bulk flow and are concentrated at the level of the ERGIC, implying that they are somehow excluded from retrograde transport (Martínez-Menárguez et al., 1999). As for the exclusion of resident ER proteins from COPII vesicles, it has been proposed that the ERGIC and cis-Golgi environment favors coalescence of secretory cargo into large assemblies, which would be excluded from COPI vesicles (Warren and Mellman, 1999). This mechanism, however, is unlikely to hold for artificial soluble cargoes, such as GFP, horseradish peroxidase or a small domain of a viral capsid protein, which have not evolved to transit through the secretory pathway and are nonetheless secreted very rapidly (Thor et al., 2009). Furthermore, when cells are incubated at 20°C, a temperature that blocks anterograde transport from the TGN, artificial secretory cargo clears the ER and accumulates in the TGN (e.g. von Blume et al., 2009), suggesting that such cargo is excluded from Golgi-to-ER traffic, which does occur at this temperature (Fossati et al., 2014a). These results indicate that a ‘valve’-like system operates at the ER–Golgi interface that prevents backflow of bulk-flow cargo. I suggest that the basis of such a valve system is in the relative contributions of COPI-dependent and -independent retrograde transport. As argued in the previous section, the amount of fluid transported by Golgi-to-ER-directed COPI vesicles may be inferior to that arriving from the ER; furthermore, tubular Rab6-dependent carriers would be poor fluid carriers. These characteristics of the retrograde pathway could be sufficient to account for the concentration of soluble cargo proteins in the TGN (see Fig. 3).

Turning now to membrane protein cargo, it has become clear that at least some newly synthesized cell surface transmembrane proteins do partition into Golgi-to-ER retrograde carriers in a seemingly futile recycling phenomenon that delays their transport to the cell surface. Using fluorescence-recovery after photobleaching (FRAP) techniques, Fossati et al. (2014a) investigated the Golgi-to-ER trafficking of different newly synthesized plasma-membrane-integral proteins in mammalian cultured cells. They found that a tail-anchored construct (GFP anchored to the bilayer through a long TMD) and the EGF receptor underwent recycling at the ER–Golgi interface, resulting in slow and poorly synchronized transport of these cargoes to the cell surface; in contrast, recycling of two other cargoes, VSVG and synaptobrevin-2, was not detected. These observations raised the question as to which factors are responsible for the different behaviors of the investigated cargoes at the ER–Golgi interface.

Partitioning into lipid microdomains might explain in part the exclusion of some cargoes from the retrograde pathway and their subsequent rapid progression through the Golgi (Lippincott-Schwartz and Phair, 2010). However, protein–protein interactions also appear to be involved in this sorting phenomenon. Indeed, Fossati et al. found that a mutant form of VSVG, in which the COPII-interacting export sequence (DxE, where ‘x’ represents any amino acid) had been deleted, was caught in the Golgi-to-ER recycling pathway (Fossati et al., 2014a). This observation suggests that the DxE motif in the cytosolic tail, in addition to being responsible for VSVG recruitment to ERES (Nishimura and Balch, 1997; Sevier et al., 2000), also plays a role in driving the progression of VSVG through the Golgi. Recently, five basic residues upstream of the DxE motif have been shown to favor the transport of VSVG through the Golgi through a COPI–CDC42-regulated pathway (Park et al., 2015).

Fossati et al. also investigated the pathway used for retrograde transport of recycling cargoes and show that it depends on Rab6 but not on COPI (Fossati et al., 2014a). This is in line with the idea that Rab6-dependent tubules, which act as vehicles for the constitutive return of bilayer to the ER, might capture any membrane protein that does not have access to specific escape pathways (Fig. 3).

In summary, in addition to folding and assembly within the ER, and to the efficiency of recruitment to ERES, an important rate-limiting factor in cargo progression through the secretory pathway is Golgi-to-ER recycling. The extent of escape of different cargoes from this pathway could contribute to the widely differing rates of transport of plasma membrane proteins from their site of synthesis in the ER to their final destination.

According to a simplified view of the secretory pathway, the principal rate-limiting factor for protein transport is exit from the ER, determined by the rate of folding and by recruitment to ERES, the latter process mediated by protein–protein interactions. Instead, progression through the Golgi based on cisternal progression (Nakano and Luini, 2010) should not require any particular signal and should thus occur at similar rates for different cargoes. In this Commentary, I have discussed possible alternative scenarios – membrane cargoes devoid of discrete export signals may be recruited into ERES by partitioning, whereas escape from partitioning into retrograde tubules requires, at least in some cases, discrete signals that favor progression through the Golgi through protein–protein interactions, as illustrated by the recent work on VSVG.

The recent findings on the early secretory pathway discussed here raise a number of new questions that will hopefully be addressed during the next years. Among these are: the role of membrane proteins with export signals on the modulation of the lipid composition of COPII vesicles and on the partitioning of secretory membrane proteins; the generality of the seemingly futile recycling of anterograde cargo between the Golgi and the ER (Fossati et al., 2014a), and the possibility of the existence of additional discrete signals that determine cargo exclusion from the retrograde pathway and that drive transport through the Golgi; the full characterization of the Rab6-dependent retrograde pathway; and the roles of COPI in retrograde and anterograde transport. The remarkable recent technological advances in proteomics, lipidomics, imaging, biophysics and gene editing should make it possible to obtain answers to these and many other questions regarding the regulation of protein export from the ER.

I am grateful to Sara Colombo, Matteo Fossati and Paolo Ronchi, who generated the results from my group discussed in this review, and who, with their ideas and enthusiasm, created an exceptionally stimulating and pleasant atmosphere in the laboratory during the past years. I am indebted to Pietro De Camilli for his constant encouragement and for his critical and constructive review of the manuscript.