Targeting of the Arf-GEF GBF1 to lipid droplets and Golgi membranes

Organelle identity is maintained dynamically by specific recruitment of proteins to membranes of a characteristic lipid composition. In many cases, an organelle is defined by a combination of a rare lipid and specific proteins, most commonly phosphoinositides and activated small G proteins (Behnia and Munro, 2005). This combination allows peripheral proteins such as protein coats, lipid modifying enzymes and actin cytoskeleton regulators to associate specifically with each compartment and contribute to their functioning and maturation. Although small G proteins such as Arf1 and Rab1 are known to function in recruiting proteins to membranes of the early secretory pathway (Gillingham and Munro, 2007), these membranes lack a characteristic phosphorylated phosphoinositide to confer compartment identity. Instead, other signature features of these compartments are used as targeting information, such as lipid packing defects, that are sensed by a specific class of amphipathic helices (Bigay and Antonny, 2012; Vamparys et al., 2013; Vanni et al., 2013).

Lipid droplets (LDs), the major energy storage depots of eukaryotic cells, are now recognized as bona fide organelles that interface with membrane trafficking pathways (Liu et al., 2004; Walther and Farese, 2012). They are distinct from other organelles in having a neutral lipid core surrounded by a phospholipid monolayer rather than a bilayer encompassing an aqueous interior. However, the composition of the phospholipids encircling LDs shares features with that of early secretory pathway membranes, notably its lack of a specific phosphorylated phosphoinositide as a compartment marker (Bartz et al., 2007a). In addition to triglycerides, which are the major energy storage molecules of eukaryotic cells, the neutral lipid core of LDs also contains esterified cholesterol. These two classes of neutral lipids serve as storage precursors of the two major lipid components of cellular membranes. In yeast, it has been shown that trafficking through the secretory pathway is coordinated with storage of neutral lipids in LDs, probably to provide the cell with the capacity to respond rapidly to demands for new membrane synthesis when growth is stimulated, and to channel newly synthesized membrane precursors from the ER to LDs for storage when growth is inhibited (Gaspar et al., 2011; Gaspar et al., 2008).

The metabolism of LDs has received considerable attention in recent years because of their central role in metabolic diseases, notably lipodystrophies and obesity (Walther and Farese, 2012; Zechner et al., 2012). Obesity is a particularly serious health concern because of its prevalence, and its predisposition to serious illnesses such as diabetes, fatty liver disease, heart disease and cancer (Bozza and Viola, 2010; Cohen et al., 2011; Greenberg et al., 2011). LDs are also required for the infectious cycle of pathogens such as hepatitis C virus, Dengue virus and Chlamydia, which target fat metabolic cells and/or lipid metabolic pathways for their propagation (Saka and Valdivia, 2012; Stehr et al., 2012).

Both the mechanisms linking LDs to the secretory pathway, and targeting of proteins to the LD surface are poorly understood. We and others have identified a crucial role of the Arf1 small G protein and its regulators in recruitment of LD-associated proteins to the LD surface, a function that is evolutionarily conserved from Drosophila to humans (Beller et al., 2008; Guo et al., 2008; Soni et al., 2009). However, the mechanisms by which Arf1 and its regulators themselves are recruited to LDs are not well understood. In the secretory pathway, Arf1 is involved in recruitment of coats to membranes to form transport vesicles (Bonifacino and Lippincott-Schwartz, 2003). At the cis-Golgi, the Arf1 activator GBF1, a guanine nucleotide exchange factor (GEF) that switches Arf1 from its inactive GDP-bound to active GTP-bound form, recruits the COPI coat to membranes, the first step in formation of COPI-coated vesicles (Beck et al., 2009). At the trans-Golgi, Arf1 is activated by members of a distinct family of GEFs, the BIG/Sec7 family. The GBF and BIG GEFs are large proteins that share a common domain structure (Bui et al., 2009; Gillingham and Munro, 2007). Among these large Arf GEFs, the mechanism of recruitment of the BIG/Sec7 proteins is the best characterized. However, two distinct localization mechanisms involving direct interactions with the BIG GEF proteins have been demonstrated (Christis and Munro, 2012; Richardson et al., 2012). The HDS1 domain downstream of the catalytic domain of yeast Sec7p is necessary for binding to membranes in an Arf1-dependent manner, as part of a positive feedback loop enhancing membrane binding (Richardson et al., 2012). It has also been shown that the N-terminus of the mammalian and Drosophila homologue of Sec7p, BIG1, interacts with the small G protein Arl1, a trans-Golgi-localized protein, which mediates the specific recruitment of BIG1 to trans-Golgi membranes (Christis and Munro, 2012). How these two localization mechanisms are coordinated is not known. For the GBF family of GEFs, the N-terminal region of mammalian GBF1 has been shown to interact directly with Rab1, which contributes to its localization to the Golgi (Monetta et al., 2007). Here we show that two regions of GBF1, both the HDS1 domain and the N-terminus, are required for its Golgi localization, whereas the HDS1 domain alone is required for localization to LDs. We show that the HDS1 domain is a lipid-binding domain, which can associate both with bilayer liposomes and LDs, in vitro and in cells, through an amphipathic helix within the domain.

In cells under normal growth conditions, GBF1 localizes primarily to membranes of the early secretory pathway, including the cis-Golgi and the ER–Golgi intermediate compartment (ERGIC), cycling dynamically between membranes and cytosol (García-Mata et al., 2003; Niu et al., 2005; Zhao et al., 2006). We made the surprising observation that the HDS1 and HDS2 domains of GBF1 (shown schematically in Fig. 1A) localize to LDs, and not to the Golgi, when expressed on their own in cells treated with fatty acids (Ellong et al., 2011) (see also Fig. 3C). However, this result is consistent with the known role of GBF1 in LD metabolism, including localization of the full-length protein to LDs in fatty-acid-treated cells (Soni et al., 2009), and leads to two hypotheses: either HDS1 and HDS2 localize only to LDs, and other domains of GBF1 specifically target the protein to secretory pathway membranes, or HDS1 and HDS2 mediate localization to both sites in a regulated manner. To distinguish between these possibilities, we tested localization of GBF1 and deletions of HDS1, HDS2 or the two domains together (Fig. 1). Treatment of cells with the Arf1-GEF inhibitor brefeldin A (BFA) for 10–20 minutes leads to an increase in the amount of GBF1 on ERGIC elements and the Golgi, as determined by both fluorescence microscopy and fractionation (Niu et al., 2005; Szul et al., 2005; Zhao et al., 2006). As is the case for certain other peripherally associated Golgi proteins that cycle dynamically between membranes and cytosol, release from Golgi membranes occurs during the fractionation procedure (Panic et al., 2003), but for GBF1, treatment of cells with BFA prior to fractionation appears to protect this release from membranes, at least partially, resulting in ∼60% of the protein remaining associated with the membrane fraction (supplementary material Fig. S1). In contrast, deletion of either the HDS1 or the HDS2 domain, or both together, resulted in a primarily cytosolic localization of the protein as assayed by fluorescence imaging (Fig. 1B–D) or subcellular fractionation (Fig. 1E). Quantification of fluorescence images demonstrated that the HDS1 and HDS2 deletion mutants were significantly delocalized from the perinuclear Golgi region, both in the absence and presence of BFA (Fig. 1B,C). Hence the HDS1 and HDS2 domains are required for GBF1 localization to ERGIC and Golgi membranes. Deletion of the entire C-terminal portion of GBF1 downstream of the catalytic Sec7 domain abolished the association of the protein with Golgi and ERGIC membranes, in the presence or absence of BFA, as monitored both by fluorescence microscopy and subcellular fractionation (supplementary material Fig. S2). However, truncations removing sequences downstream of the HDS1 and HDS2 domains all resulted in localization to membranes in both assays (supplementary material Fig. S2), indicating that no other regions of the GBF1 C-terminal region are required for membrane targeting. Hence the HDS1 and HDS2 domains of GBF1 are required both for localization to secretory pathway membranes and to LDs.

To determine the mechanism by which the HDS1 domain localizes to membranes, we purified HDS1 as a GST fusion from Escherichia coli, and tested binding to liposomes in vitro using a sucrose gradient flotation assay. We tested liposomes of different compositions, including a phospholipid mixture derived from soybeans called azolectin [∼55% phosphatidylcholine (PC), 20% phosphatidylethanolamine (PE), and other less abundant phospholipid species], a mixture of phospholipids of composition similar to Golgi membranes [‘Golgi mix’: 50% egg PC, 19% egg PE, 10% liver phosphatidylinositol (PI), 5% brain phosphatidylserine (PS), and 16% cholesterol, where the percentage given is a molar ratio] (Bigay et al., 2005) or pure egg PC. We found that the HDS1 domain showed a high level of binding to azolectin liposomes, to an extent approaching that of myristoylated Arf1–GTP (Fig. 2A,B). HDS1 also showed high levels of binding to Golgi mix liposomes, and to pure PC liposomes (Fig. 2B). A single point mutation of a highly conserved tryptophan residue (HDS1-W1028S, see below) severely inhibited liposome binding (Fig. 2C). These results indicate that HDS1 is a direct lipid-binding domain. Interestingly, the HDS1 domain of GBF1 is able to bind to liposomes containing only neutral lipids, and therefore does not have a requirement in vitro for charged or rare lipids (e.g. phosphorylated PIs). Furthermore, the GBF1 lipid-binding HDS1 domain does not require another protein, such as Arf1, to efficiently bind to liposomes, a property that differs from yeast Sec7p, which does require Arf1–GTP for lipid binding in vitro (Richardson et al., 2012). Hence, although GBF1 shares a similar domain structure with that of members of the BIG/Sec7 family of Arf GEFs (its HDS1 domain shares 23% identity and 45% similarity with that of yeast Sec7p), the two GEFs appear to differ in their requirement for Arf1 for lipid binding.

We sought to test the ability of HDS1 to associate with artificial LDs in vitro, using a recently developed in vitro method to produce artificial droplets and assay protein binding by flotation analysis (see Materials and Methods). Artificial LDs can be formed by combining triglycerides and phospholipids, and are stable when the level of PC, which acts as a surfactant, is sufficiently high (Krahmer et al., 2011). Below this concentration, an emulsion of small LDs (less than a micrometer in diameter) in aqueous buffer undergo spontaneous fusion to produce very large droplets. We tested two phospholipid compositions, either PC alone or Golgi mix liposomes containing 50 mol % PC, above the minimum concentration to stably maintain artificial LDs. Droplets of a relatively uniform size were produced by extrusion. Binding of proteins to artificial LDs was assayed by flotation, and to validate the method we first used different forms of myristoylated Arf1 (myr-Arf1) as a control. Arf1 in its GDP-bound form (myr-Arf1–GDP) is primarily soluble, whereas its GTP-bound form (myr-Arf1–GTP) is tightly membrane bound when binding is assayed on liposomes of a wide range of compositions, including PC alone (Antonny et al., 1997). Proteomics studies have indicated that Arf1–GTP is present on LDs purified from cells induced to form LDs by oleate treatment (Bartz et al., 2007b). When assayed on either pure PC or Golgi mix artificial droplets, myr-Arf1–GDP failed to bind, whereas myr-Arf1–GTP bound as efficiently to these artificial LDs as to liposomes (Fig. 2D). We then tested binding of GST–HDS1, and found that it associated efficiently with artificial droplets, in contrast to GST alone, which was unable to bind (Fig. 2E). There was no significant difference in binding to pure PC or Golgi mix droplets. These results demonstrate that HDS1 is capable of binding to a phospholipid monolayer surface in vitro, and that this domain can bind efficiently to artificial LDs containing only neutral phospholipids.

The C-terminal region of HDS1 is predicted to form an extended amphipathic helix (AH) with the characteristics of a lipid-binding AH (supplementary material Fig. S3A). Within this region is a particularly highly conserved tryptophan residue (Bui et al., 2009), on the hydrophobic side of the predicted helix, within a smaller 21-amino-acid region with the best potential to form an AH (Fig. 3A). To determine whether this region affects the lipid-binding properties of HDS1 in vitro, we introduced the W1028S mutation and deletion of the 21-amino-acid AH of HDS1 into GST–HDS1 (region shown in Fig. 3A, ΔAH) and purified the proteins from E. coli. The HDS1-W1028S mutant failed to bind liposomes, as describe above (Fig. 2C). Both the HDS1-W1028S and ΔAH mutations abolished binding of HDS1 to artificial droplets (Fig. 3B). Hence the C-terminal amphipathic region of HDS is essential for binding to artificial droplets in vitro.

To test the importance of the amphipathic C-terminal region (Fig. 3A) in binding of HDS1 to LDs in cells (Fig. 3C), we first developed an algorithm to quantify the level of an mCherry-tagged protein associated with BODIPY-labeled LDs in fluorescence images. Fluorescence intensity was determined along a vector from a point within the LD interior at a fixed distance from the edge (to allow quantification of LDs of different sizes), and extending perpendicularly towards the cytosol (Fig. 3C, inset). If the protein is bound to LDs, maximum fluorescence will be found in a narrow region encircling the LD, and the relative fluorescence will decrease with increasing distance from the LD. If the protein is cytosolic, this value will be relatively constant at all points from the LD circumference. Normalized fluorescence values, averaged for all LDs in a cell, were plotted as a function of distance from the LD starting point. Results for cells expressing mCherry–HDS1 or mCherry alone are plotted in Fig. 3D. To simplify comparison of different HDS1 mutants for their capacity to associate with LDs, levels of protein in the cytosol were quantified. Specifically, the values of each plot at the plateau reached at a distance of 18 pixels (corresponding to ∼2 µm) from the starting point were determined, and normalized against the value obtained for wild-type HDS1. The results obtained for cytosolic mCherry alone compared with wild-type mCherry-HDS1 are shown in Fig. 3E.

To test the idea that the C-terminal amphipathic region of HDS1 forms an AH, we mutated residues that would alter its amphipathic properties, and assayed the localization of the mutant HDS1 domains fused to mCherry. Positions of the residues within this region that were mutated are shown in Fig. 3A. Mutations in hydrophobic residues in the predicted non-polar face of the HDS1 AH (W1028S, L1024A) led to decreased LD association and an increase in the cytosolic pool of the domain (Fig. 3F,G). Interestingly, the L1024A mutation in a leucine residue predicted to be at the center of the proposed amphipathic α-helix (Fig. 3A), had a severe effect on localization of HDS1, whereas mutation of the adjacent leucines to alanines had little effect (Fig. 3G). The triple mutation of three hydrophobic residues on the non-polar face (L1024A, I1031A, L1038A) had the same effect on localization as the single L1024A mutation (Fig. 3G). Mutations in the charged residues of the polar face (R1025 and R1040) of the HDS1 AH also resulted in loss of LD association (Fig. 3G). A double mutant of both the conserved Trp on the hydrophobic side and a Lys residue on the polar side (W1028S, K1029S) had the strongest effect on localization of HDS1 (Fig. 3G). However, mutation of the isoleucine predicted to be at the interfacial region of the AH to a lysine (I1023K) had no significant effect on localization (Fig. 3G). Deletion of a major portion of the predicted amphipathic helix (residues 1021–1041) also reduced LD targeting of HDS1 in cells (supplementary material Fig. S3B). These results support the conclusion that the amphipathic property of this predicted helical region plays a crucial role in LD binding in cells.

Our results have shown that, both in vivo and in vitro, the C-terminal amphipathic region of the GBF1 HDS1is required for its binding to LDs. To determine whether this region is sufficient on its own to target LDs, we expressed amino acids 1021–1048 fused to mCherry in cells treated with oleic acid. This predicted amphipathic helix was able, on its own, to localize to LDs (Fig. 4A). Compared with the full-length HDS1 domain, there was a higher level of the C-terminal AH in the cytosol (Fig. 4B), suggesting that it targets less efficiently or is less stably associated with LDs. In vitro, the C-terminal amphipathic helical region of HDS1 (amino acids 1005–1048 fused to GST) exhibited a high level of binding both to liposomes of different composition, and to artificial droplets (Fig. 4C). Hence the C-terminal AH of HDS1 is a lipid-binding motif, which can target both a monolayer droplet surface and a lipid bilayer, in vitro and in cells.

Our results have shown that the HDS1 domain expressed in oleic-acid-treated cells localizes very efficiently to LDs, with little if any localization to other organelles detectable at steady state. To further understand this localization mechanism, we determined how stably mCherry–HDS1 is associated with LDs by fluorescence recovery after photobleaching (FRAP) analysis. Little if any significant fluorescence recovery was detected up to 6 minutes after photobleaching of an entire LD, indicating a slow replenishment of the majority of mCherry–HDS1 onto the LD surface (Fig. 5A, quantification shown in Fig. 5B). Hence LD-associated mCherry–HDS1 does not rapidly exchange with other pools within the cell. In cells not treated with oleic acid, HDS1 was found associated both with the few LDs present, and with structures in close proximity to the ER (Fig. 6A and data not shown). Subcellular fractionation indicated that the majority of HDS1 is present in the membrane fraction under these conditions (Fig. 6C). Hence HDS1 is targeted to membrane structures when not associated with LDs.

The fact that once associated with LDs, HDS1 is not readily exchanged, suggests that it might be important for cells to regulate this tight association. Indeed, in contrast to HDS1 alone, only a small fraction of full-length GBF1 is found at LDs in cells treated with oleic acid (Soni et al., 2009), suggesting regulation of its localization. To determine whether other domains of GBF1 are involved in membrane association, we expressed larger fragments of the protein. Strikingly, we found that the Sec7 domain–HDS1 tandem expressed in cells had a predominantly cytosolic localization, both by fluorescence imaging (Fig. 6A,B) and fractionation (Fig. 6C). In addition, the purified Sec7–HDS1 tandem was unable to associate with liposomes in vitro, in stark contrast to HDS1 alone (Fig. 6D,E). Hence the Sec7 domain modulates the lipid-binding capacity of HDS1, inhibiting its ability to bind membranes both in vitro and in cells. Because the N-terminal region of yeast Sec7p, including both the Sec7 and HDS1 domains, binds to membranes only in the presence of Arf1 (Richardson et al., 2012), we tested the idea that the presence of the Sec7 domain might now confer a requirement for Arf1 in liposome binding in vitro. The Sec7–HDS1 tandem was therefore incubated with liposomes in the presence of myr-Arf1. However, neither the GDP-bound nor the GTP-bound forms of myr-Arf1 increased the level of binding of Sec7–HDS1 to liposomes (Fig. 6E).

Given that the Sec7–HDS1 tandem does not localize to membranes in cells or to liposomes in vitro, we were interested in determining which additional domains of GBF1 are required for membrane targeting. The C-terminal regions of GBF1 downstream of HDS2, although they may play a role, are not required for localization to ERGIC–Golgi membranes in cells (supplementary material Fig. S2). Hence all or part of the N-terminal region, containing two homology domains called DCB and HUS, is likely to be important for Golgi localization. Indeed, we found that N-terminal truncations of GBF1 that lacked the DCB domain failed to localize to ERGIC or Golgi structures (supplementary material Fig. S2). We then expressed a construct extending from the HUS domain to the end of the HDS2 domain, encompassing both the catalytic Sec7 and HDS1 domains. This portion of GBF1 was associated with membranes in cells, as determined by fluorescence and fractionation analysis (Fig. 7A,B). However, this construct, like the N-terminal truncations, did not accumulate in the Golgi region, even when cells were treated with BFA (data not shown), although it did fractionate with membranes under these conditions (Fig. 7B). In cells treated with oleic acid, this HUS-to-HDS2 construct localized to LDs, similar to the HDS1 domain alone (Fig. 7A, bottom panels), although with a higher level in the cytosol (Fig. 7A, compare left and right bottom panels). Hence the HUS domain is not sufficient to mediate targeting of the downstream Sec7–HDS1–HDS2 domains to the Golgi, but requires in addition the DCB domain for Golgi localization.

We demonstrate here that the peripherally associated regulator GBF1 localizes to lipid droplets (LDs) and to Golgi membranes through the same lipid-binding domain. GBF1, an activator of the Arf1 small G protein, is the first Arf1-GEF to function in the secretory pathway, and also plays an important role in LD metabolism (Donaldson and Jackson, 2011; Walther and Farese, 2012). Two domains of GBF1, HDS1 and HDS2, localize on their own to LDs in cells, whereas the full-length GBF1 protein localizes primarily to membranes of the secretory pathway, even in cells induced to produce LDs by treatment with fatty acids (Ellong et al., 2011; Soni et al., 2009). In this study, we focused on the HDS1 domain, located directly downstream of the catalytic Sec7 domain of GBF1, showing that it is capable of binding both bilayer liposomes and monolayer droplets in vitro. We also show that this potent LD-binding capacity of HDS1 is attenuated by the Sec7 domain, because the Sec7 domain–HDS1 tandem is soluble, both in vitro and in cells. Further addition of the upstream HUS domain results in a construct that can bind to membranes in cells, but does not accumulate on Golgi or ERGIC membranes at steady state. The full N-terminal region is required for GBF1 to maintain a persistent association with Golgi membranes.

During Golgi maturation, there is sequential activation of Arf1, first by a GBF family member at the cis-Golgi, then by a BIG family member at the trans-Golgi (Gillingham and Munro, 2007). Fromme and colleagues have proposed that the Arf1-dependent interaction of the Sec7p HDS1 domain is important to restrict binding of this late-acting Arf-GEF to membranes on which Arf1 has already been activated (Richardson et al., 2012). In this context, it is significant that the GBF1 HDS1 domain does not require Arf1 for binding, nor does activated Arf1 enhance binding of the HDS1 domain to liposomes in vitro. The Arf1-independent binding of GBF HDS1 and the Arf1-dependent lipid binding of the equivalent domain in BIG/Sec7 will ensure that GBF1 Arf-GEFs are recruited prior to BIG/Sec7 proteins, thus providing directionality in trafficking through the Golgi.

Together with these previous studies, our results support the idea of a general mechanism for binding of the GBF and BIG Arf-GEFs to membranes. For GBF1, both the HDS1 and HDS2 domains are required for GBF1 localization in cells, and the N-terminal region, which has been shown to bind to Rab1 (Monetta et al., 2007), is necessary to target GBF1 to Golgi membranes. Localization of BIG1, a mammalian and Drosophila homologue of yeast Sec7p, to late Golgi membranes requires binding of the BIG1 N-terminus to activated Arl1, a marker of the late Golgi (Christis and Munro, 2012). Hence, as for GBF1, both interaction of the HDS1 domain directly with lipids, and interaction of the N-terminus with another small G protein are involved in specific targeting of the BIG/Sec7 Arf-GEFs.

Our in vitro data indicate that HDS1, through its C-terminal amphipathic region, is able to efficiently bind to liposomes or artificial droplets containing PC alone. Hence a specific phosphorylated phosphoinositide, other charged lipids, or cholesterol are not required for lipid binding by HDS1, in keeping with the composition of both early Golgi membranes and LDs. Interestingly, a recent study has shown that a larger region of GBF1, including HDS1 and HDS2, as well as a 90-amino-acid region downstream, can target the leading edge of migrating neutrophils through specific recognition of phosphatidylinositol 3-phosphate and other products of phosphoinositide 3-kinase gamma (Mazaki et al., 2012). HDS1 and HDS2 on their own do not bind to phosphatidylinositol 3-phosphate, nor target the leading edge, but require the 90-amino-acid proline-rich region immediately downstream of HDS2 for this targeting. The fascinating picture that emerges is that the HDS1 domain (and probably HDS2 as well) are lipid-binding domains that have the capacity to bind to different membranes in cells, with other regions of the GBF1 protein conferring specific information that targets the protein either to membranes of the early secretory pathway, LDs of oleic-acid-treated cells or the leading edge of migrating cells. For Golgi targeting of GBF1, protein–protein interactions are probably involved in addition to the protein–lipid interaction we reveal here. For LD localization of either GBF1 or the HDS1 domain alone, we have not ruled out the possibility that protein–protein interactions are involved in targeting. Although the HDS1 domain can localize to liposomes in vitro in the absence of additional proteins, we have evidence of interactions with LD-localized proteins that may possibly contribute to targeting (Ellong et al., 2011); coincidence detection mechanisms involving both may operate. Further studies are required to determine whether protein–lipid interactions may be solely responsible for targeting HDS1 to LDs, and by what mechanism.

The C-terminal region of the lipid-binding HDS1 domain of GBF1 is an amphipathic region predicted to form an α-helix on membranes. We have shown here that this region on its own is able to bind to LDs in cells, as well as to droplets in vitro, indicating that it is an LD targeting motif. Our mutagenesis analysis of this region supports the conclusion that it forms an amphipathic helix in vitro and in cells, and that this property is important for its targeting to LDs. Interestingly, other proteins, including perilipins (Bulankina et al., 2009), hepatitis C virus (HCV) core protein (Boulant et al., 2006), HCV NS5A protein, the antiviral peptide viperin (Hinson and Cresswell, 2009) and CTP:phosphocholine cytidylyltransferase α (CCTα) (Krahmer et al., 2011), all have amphipathic helices that are necessary and sufficient for targeting to LDs. Because the LD surface is a phospholipid monolayer, amphipathic helices are ideal localization devices, as they enter the interfacial region of only a single phospholipid monolayer. Although HDS1 can bind both to bilayer liposomes and droplets surrounded by a phospholipid monolayer in vitro, it localizes primarily to LDs in cells, in a stable manner. Other LD-associated proteins that are localized through AHs, including specific perilipins (Wang et al., 2009) and CCTα (Krahmer et al., 2011) are also stably associated with the LD surface. This could be due to either interactions of these AH-containing proteins with other LD resident proteins, or the biophysical properties of the LD phospholipid monolayer surface that lead to a stable association of an AH once it is bound. Interestingly, in addition to AH-containing LD proteins, the enzyme catalyzing the first step in triglyceride biosynthesis, glycerol-3-phosphate acyltransferase 4 (GPAT4), has been shown to localize both to the ER and to LDs, and once on the LD surface, it does not return to the ER (Wilfling et al., 2013). This property may be a general principle of LD-binding proteins.

Plasmids used in this study are listed in supplementary material Table S1. All constructs were confirmed by DNA sequencing. HDS1 was cloned into pGEX-4T1 using EcoRI and SalI sites; Sec7 and Sec7-HDS1 using BamHI and XhoI sites. GBF1 and truncated forms were cloned into XhoI and KpnI sites of pVenus-C1 and pmCherry-C1 (generous gifts from George Patterson, NIH, USA). GBF1-ΔHDS1, -ΔHDS2 and -ΔHDS1-2 were derived from pEGFP-GBF1-887Xho (which has a XhoI site after amino acid 887 of GBF1; a kind gift from Y. Mazaki and H. Sabe, Kumamoto, Japan) by PCR amplification of appropriate C-terminal regions of GBF1 and cloning into XhoI and NotI sites. The following antibodies were used: mouse anti-GST (Sigma-Aldrich, St Louis, MO, USA), mouse anti-GFP (Roche Diagnostics, Indianapolis, IN, USA), mouse anti-GBF1 (BD Biosciences, Franklin Lakes, NJ, USA), rabbit anti-ERGIC53/p58 (Sigma-Aldrich), mouse anti-β-actin (Abcam, Cambridge, MA, USA), rabbit anti-ATGL (Cell Signaling Technology, Danvers, MA, USA), guinea pig anti-ADRP (Progen Biotechnik, Heidelberg, Germany), mouse anti-GM130 (BD Biosciences).

GST was expressed and purified as described previously (Ellong et al., 2011). Myr-Arf1 was purified from BL21Gold(DE3) cells transformed with pET11d-Arf1 and pBB(NMT) (generous gifts of Bruno Antonny, Sophia Antipolis, France) as described (Franco et al., 1995). GST-HDS1 (wild type, W1028S and ΔAH) were induced in E. coli strain C41(DE3) with isopropyl-β-D thiogalactopyranoside (IPTG, Euromedex, Strasbourg, France) at 30°C for 4 hours. GST-Sec7-GBF1 and GST-Sec7-HDS1-GBF1 were induced in E. coli strain Rosetta(DE3)pLys with IPTG at 20°C overnight (for GST-Sec7) or at 37°C for 2 hours (for GST-Sec7-HDS1). For HDS1-containing proteins, lysis was carried out in TEX buffer (50 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 1 mM DTT) in the presence of Complete Mini Protease Inhibitor cocktail (Roche Diagnostics), 2 mM phenylmethylsulfonyl fluoride (PMSF, USB), 0.5% Nonidet P40 substitute (NP40, Sigma-Aldrich) and 0.5% Triton X-100 (Sigma-Aldrich). Proteins were purified on a glutathione–Sepharose 4B resin (GE Healthcare, Little Chalfont, Buckinghamshire, UK) at 4°C and eluted in 15 ml TEX supplemented with 50 mM reduced L-glutathione, 0.5% NP40 and 0.5% Triton X-100, then dialyzed against 10 mM Tris-HCl pH 8.0, 50 mM NaCl, 1 mM DTT buffer and concentrated with a Vivaspin system (Sartorus Stedim, Palaiseau, France). Concentrations were determined with BCA Protein Assay kit (Thermo Scientifique/Pierce Brebières, France).

For GST–Sec7–GBF1, cells from a 1 liter culture were sonicated in 20 ml of buffer A (100 mM sodium phosphate pH 8.0, 300 mM NaCl, 2 mM DTT) supplemented with 0.5% sodium deoxycholate and Complete Mini Protease Inhibitor cocktail. Cells were centrifuged and the supernatant was loaded on a glutathione–Sepharose 4B column. The resin was washed with buffer A, then with buffer A containing increasing concentration of reduced L-glutathione: 10 mM, 30 mM then 50 mM. Protein eluted in the final two washing steps was dialyzed against 10 mM Tris-HCl pH 8.0, 100 mM NaCl, 2 mM DTT buffer, and concentrated with a Vivaspin system.

Azolectin (L-phosphatidylcholine type IV-S) was from Sigma-Aldrich; other lipids were from Avanti Polar Lipids/Coger (Paris, France). Golgi-mix phospholipids contained (mol%) egg PC (50), egg PE (19), brain PS (5), liver PI (10), cholesterol (16), NBD-PE (0.2). For liposome preparation, a dried film was prepared in a glass tube by evaporation of a mixture of the indicated lipids in chloroform (azolectin, Golgi mix or PC only) and resuspended in HK buffer (50 mM HEPES-KOH pH 7.2 and 120 mM potassium acetate). After five freeze–thaw cycles, the liposome suspension was extruded 19 times through polycarbonate filters (pore size of 0.4 µm for defined composition liposomes or 0.1 µm for azolectin liposomes) using a hand extruder (Avanti Polar Lipids) at a final lipid concentration of 20 mg/ml for azolectin liposomes and 1 mM for defined composition liposomes. Liposomes were stored at room temperature and used within 2 days after preparation. Artificial droplets were prepared as described previously (Thiam et al., 2013). Briefly, 70 µl triolein (Sigma-Aldrich) were mixed in a glass tube with 0.5 µmol phospholipids and 1 pmol NBD-PE. Chloroform was evaporated under an argon flow, then under vacuum. The oil was dispersed in 1 ml of HK buffer by vigorous vortexing, then extruded nine times through a 1 µm pore size nitrocellulose membrane. Artificial droplets were stored at room temperature and used the day of the preparation.

For proteins purified in the presence of detergent, Pierce Detergent Removal Spin Columns (Thermo Scientifique/Pierce) were used according to manufacturer's instructions. Proteins were incubated with liposomes or artificial droplets in HK buffer containing 1 mM MgCl₂ and 1 mM DTT (HKM buffer) at 37°C for 30 minutes in a total volume of 58 µl. Nucleotide (100 µM of GTP or GDP) was loaded onto Myr-Arf1 in the presence of lipids (liposomes or droplets) and 2 mM EDTA for 30 minutes at 37°C, then 2 mM MgCl₂ was added. The suspension was adjusted to 31.5% sucrose by adding 42 µl of a 75% w/v sucrose solution in HKM buffer, then overlaid with 75 µl HKM buffer containing 25% w/v sucrose, and finally 10 µl HKM buffer. The sample was centrifuged at 55,000 rpm (259,000 g) in a Beckman rotor TLS 55 for 1 hour at 20°C. The bottom (95 µl), middle (75 µl) and top (20 µl) fractions were manually collected from the bottom using a syringe (Hamilton Company, Bonaduz, Switzerland) and analyzed by SDS–PAGE using Blue Silver Staining (Candiano et al., 2004). Gels were scanned and quantifications performed using ImageJ (NIH, Bethesda, MD, USA). Except for azolectin liposomes, the recovery of the liposomes/artificial droplets after fraction collection was checked using NBD-PE fluorescence with a FUJI LAS-3000 fluorescence imaging system.

HeLa and Cos7 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 4.5 g/l glucose, 4 mM L-glutamine, 1 mM sodium pyruvate (GE Healthcare) and 10% fetal bovine serum (Invitrogen, Carlsbad, CA, USA). Retinal pigment epithelial (RPE-1) cells were grown in DMEM/Ham's F12 medium (1∶1; PAA, Les Mureaux, France) supplemented with 10% FBS. Subconfluent cells were transfected with plasmids using Lipofectamine 2000 (Invitrogen) or FuGENE-6 (Roche Diagnostics). Oleic acid (OA) complexed with BSA was prepared as described previously (Soni et al., 2009). For overnight OA treatment, at 8 hour post-transfection cells were treated with 100–400 µM OA and incubated for an additional 16–20 hours before preparation for imaging. BFA (Sigma-Aldrich) was added to cells at 5–10 µg/ml for 5–10 minutes before preparation for imaging.

Image processing algorithms were developed to quantify the intensity of the signal as a function of the distance to LD volume using Imaris 7.6 (Bitplane) and ImageJ. All images were processed with the same parameters. LD size and position were determined first with the Imaris Spots detection tool using the region growing option on the green channel which correspond to the LD core stained with BODIPY 493/503. An image of isolated LD positions was then generated and exported to ImageJ in order to identify pixels which were inside the cell and outside LDs in the red channel (pmCherry fusion proteins). These pixels were obtained using filtering and mathematical morphology in the following way: an exact Euclidean distance transformation in 3D was first performed from the image of isolated LD positions to create a new image where the values of the intensities of pixels were equal to the distance from the center of LD previously obtained. Successive thresholds of this distance map were performed in order to create a mask for each distance from LDs. These masks were used to select pixels and were quantified in the red channel. Another mask corresponding to borders of the cell was obtained by filtering (median filter) and thresholding with a low level in the red channel. Finally, the average signal in the red channel of the pixels included in the mask of the cell was calculated for each distance to the center of LDs. The average signal was normalized against the maximum value and plotted against the distance to LDs to generate an intensity profile of the red channel signal along the cross section of LDs.

We thank Ting Niu, George Patterson, Yuichi Mazaki, Hisataka Sabe and Bruno Antonny for plasmids. We are grateful to Bruno Antonny for communication of the artificial droplet assay, and to Ting Niu for preliminary GBF1 localization results.