Mature lipid droplets are accessible to ER luminal proteins

Lipid droplets serve to store metabolic energy in the form of neutral lipids, particularly triacylglycerols and steryl esters. They form globular intracellular structures composed of a core of neutral lipids that is covered by a phospholipid monolayer harboring a limited set of proteins, many of which are important for neutral lipid synthesis or breakdown. However, triacylglycerols and steryl esters are synthesized by endoplasmic reticulum (ER)-localized enzymes. This observation supports a model of lipid droplet biogenesis in which neutral lipids first accumulate between the two leaflets of the ER membrane. These lipid lenses then grow in size and mature in their protein composition to form nascent lipid droplets, which are then postulated to bud from the ER membrane to form cytosolic structures (Ohsaki et al., 2014; Pol et al., 2014; Thiam et al., 2013).

Lipid droplets are closely associated with different cellular organelles including mitochondria, vacuoles, peroxisomes and ER (Barbosa et al., 2015; Murphy et al., 2009). The association of lipid droplets with the ER has been extensively investigated. Live-cell microscopy has shown close association of lipid droplets with the ER membrane and concurrent movement of the two structures (Szymanski et al., 2007; Targett-Adams et al., 2003). Freeze fracture experiments have revealed that the ER membrane is continuous with the surface layer of lipid droplets or that it encloses lipid droplets tightly (Blanchette-Mackie et al., 1995; Robenek et al., 2006). Functional studies have indicated that the activation of fatty acids and their incorporation into triacylglycerol occurs at specific ER microdomains from where lipid droplets grow (Kassan et al., 2013; Xu et al., 2012b). The contact sites between the ER and lipid droplets have also been characterized and shown to be stabilized by the seipin complex, which acts as a diffusion barrier between the two compartments (Grippa et al., 2015).

Both proteins and lipids can be exchanged between the ER and lipid droplets and this process is dynamically regulated. Neutral lipid accumulation induces relocalization of integral membrane proteins from the ER to lipid droplets, and these proteins move again back into the ER when neutral lipid pools are degraded (Jacquier et al., 2011; Kassan et al., 2013; Wilfling et al., 2013; Zehmer et al., 2009). In addition, intermediates of neutral lipid synthesis and degradation are shuffled between the two compartments to allow growth and shrinkage of lipid droplets, depending on the metabolic needs of the cell (Markgraf et al., 2014). Moreover, lipid droplet size is controlled by the exchange of neutral lipids between lipid droplets, a process that is mediated by members of the CIDE family (Xu et al., 2012a).

The fact that one of the two triacylglycerol biosynthetic enzymes in yeast, Lro1, an integral membrane enzyme of the ER, has its active site positioned within the ER lumen indicates that neutral lipids can be formed in the inner leaflet of the ER membrane and reach nascent, or possibly even mature, lipid droplets (Choudhary et al., 2011). Moreover, lipid droplets accumulating triacylglycerol formed by Lro1 are seemingly indistinguishable from those having neutral lipids formed by the acyl-CoA-dependent acyltransferase Dga1 (Sandager et al., 2002; Sorger et al., 2004). These observations, together with the fact that lipoproteins of animal cells, which are structurally very similar to lipid droplets, mature and assemble within the ER lumen from where they are then secreted into the extracellular space, prompted us to investigate the topology of lipid droplet formation in more detail (Fisher and Ginsberg, 2002; Hussain et al., 2003). Although it is clear that lipid droplets are accessible to soluble cytosolic factors, such as the perilipins (PLINs), which are abundant structural proteins that are targeted to the lipid droplets through conserved amphipathic helices, it is unknown whether lipid droplets are also accessible from within the lumen of the ER (Rowe et al., 2016).

Here, we probe the topological relationship between lipid droplets and the ER by targeting cytosolic lipid droplet marker proteins into the lumen of the ER. Remarkably, these ER luminal probes are still capable of localizing to lipid droplets, indicating that lipid droplets are accessible from within the ER luminal compartment. These results suggest that lipid droplets represent a domain within the ER that is accessible from both the cytosolic and the luminal site. A model for the biogenesis of lipid droplets that is consistent with these observations is discussed.

To probe the topology of lipid droplet formation, we targeted lipid-droplet-resident proteins into the ER lumen and examined their subcellular distribution. Therefore, two yeast lipid droplet marker proteins, Erg6, an enzyme of the ergosterol biosynthetic pathway, and Tgl1, a steryl ester lipase, were fused to the signal sequence of Pry1 (amino acids 1 to 19), a secreted glycoprotein (Choudhary and Schneiter, 2012; Koffel et al., 2005; Leber et al., 1994). The localization of GFP-tagged versions of these proteins was then analyzed by confocal microscopy. Given that Tgl1 contains an N-terminal transmembrane domain, we fused a truncated version lacking the first 29 amino acids (Tgl1Δtm) to the signal sequence of Pry1 (Koffel et al., 2005). The native versions of Erg6 and the cytosolic version of Tgl1 lacking the transmembrane domain localized to punctate structures that co-stained with Nile Red, a lipophilic dye that stains lipid droplets, indicating that both reporter proteins localize to lipid droplets. Unexpectedly, the signal-sequence-containing versions of these proteins, denoted ss-Erg6 and ss-Tgl1(Δtm), also showed a punctate pattern and the GFP signal merged with Nile-Red-stained lipid droplets, indicating that the presence of the ER signal sequence allows proper targeting of these proteins to lipid droplets (Fig. 1A).

The synthesis of neutral lipids in yeast is under the control of four enzymes, two of which, Dga1 and Lro1, produce triacylglycerols, and the other two of which, Are1 and Are2, generate steryl esters (Czabany et al., 2007). Cells lacking all four neutral-lipid-synthesizing enzymes are devoid of lipid droplets (Sandager et al., 2002). In the absence of lipid droplets, many of the otherwise lipid-droplet-localized proteins relocalize to the ER membrane (Jacquier et al., 2011; Kassan et al., 2013; Wilfling et al., 2013; Zehmer et al., 2009). To test whether the signal-sequence-containing reporter proteins would also localize to the ER membrane in cells lacking lipid droplets, their subcellular localization was analyzed in an inducible strain in which three of the genes required for neutral lipid synthesis (ARE1, ARE2 and DGA1) were deleted and the fourth gene was placed under transcriptional control of a glucose repressible promoter (GAL-LRO1). Under conditions where LRO1 is repressed, Erg6 and Tgl1(Δtm) and the signal-sequence-containing reporters ss-Erg6 and ss-Tgl1(Δtm), displayed a circular staining that merged with the localization of the ER luminal marker protein ss-mCherry-HDEL, indicating that signal-sequence-containing proteins relocalize to the ER in cells that have no lipid droplets (Rogers et al., 2014) (Fig. 1B). Thus, signal-sequence-containing lipid droplet proteins behave similar to the non signal-sequence-containing counterparts in that they localized to lipid droplets in cells that made neutral lipids, whereas in the absence of lipid droplets the proteins localized to the ER.

To examine the localization of ss-Erg6 on lipid droplets that are labeled by a lipid-droplet-resident protein instead of Nile Red, we generated a strain that expressed a GFP-tagged version of ss-Erg6 and an mCherry-tagged version of wild-type Erg6. Cells were cultivated in medium containing oleate to increase the size of lipid droplet. Both proteins displayed circular rim labeling on these large lipid droplets. The fluorescence intensity along a line crossing one of these large lipid droplets reveals extensive overlap between the localization of ss-Erg6 and Erg6, and more than 99% of the lipid droplets were co-stained with both ss-Erg6–GFP and Erg6–mCherry (Fig. 1C). Similarly, cytosolic and signal-sequence-containing versions of Erg6 and Tgl1(Δtm) displayed a high degree (>86%) of colocalization with lipid droplets marked by Erg6–mCherry in cells grown in rich medium (Fig. 1A).

Erg6 and Tgl1 are both membrane-associated proteins; thus, we wondered whether a truly cytosolic lipid droplet protein could also be targeted to lipid droplets from within the ER luminal compartment. Therefore, we tested whether addition of a signal sequence from the secreted glycoprotein Pry1 (amino acids 1 to 19) to the human lipid-droplet-scaffolding proteins PLIN1–PLIN3, would also result in targeting of the proteins to lipid droplets. PLINs constitute a family of related proteins, sharing a common PAT domain, originally identified in the founding members of that family, PLIN1, PLIN2 (also known as adipophilin, ADRP) and PLIN3 (also known as TIP47) (Bickel et al., 2009; Miura et al., 2002). PLINs are properly targeted to lipid droplets when expressed in yeast, and they are soluble proteins in cells that lack the capacity to form lipid droplets (Jacquier et al., 2013). When fused to the signal sequence of Pry1, all three PLINs localized to Nile-Red-positive punctate structures, indicating that even soluble proteins when targeted into the ER lumen localize to lipid droplets (Fig. 2A). Both cytosolic and signal-sequence-containing versions of these PLINs (denoted ss-PLIN1, ss-PLIN2 and ss-PLIN3) also displayed a high degree (>84%) of colocalization with lipid droplets marked by Erg6–mCherry (Fig. 2A).

To confirm that the appended signal sequence effectively targeted the PLINs into the ER lumen, we analyzed their subcellular distribution in cells lacking lipid droplets. Expression of PLINs in cells lacking neutral lipids, profoundly affected their subcellular localization. Instead of showing a punctate lipid droplet localization as observed in wild-type cells, the cytosolic PLINs showed a diffuse cytosolic distribution (Jacquier et al., 2013) (Fig. 2B). In marked contrast, ss-PLIN1 and ss-PLIN3 displayed ring-like staining and colocalized with the ER luminal marker ss-mCherry-HDEL (Fig. 2B). ss-PLIN2 was hardly detectable when expressed in cells lacking lipid droplets, possibly because the protein was rapidly degraded. To visualize its localization in cell lacking lipid droplets, we deleted Hrd1, a ubiquitin protein ligase required for ER-associated degradation (ERAD) of misfolded ER luminal proteins (Carvalho et al., 2006). Whereas cytosolic PLIN2 displayed cytosolic localization in cells lacking lipid droplets, ss-PLIN2 displayed ER staining in a hrd1Δ mutant background (Fig. 2B). These results indicate that ss-PLINs are localized to the lumen of the ER in cells that have no lipid droplets and that these ER luminal proteins can reach and localize to lipid droplets when cells have lipid droplets.

To examine the localization of ss-PLIN2 on lipid droplets that are labeled by a lipid-droplet-resident protein instead of Nile Red, we generated a strain that expressed a GFP-tagged version of ss-PLIN2 and an mCherry-tagged version of cytosolic PLIN2. Cells were again cultivated in medium containing oleate to increase the size of lipid droplet. Both proteins displayed circular rim staining on these large lipid droplets. The fluorescence intensity along a line crossing two of these large lipid droplets revealed extensive overlap between the localization of cytosolic PLIN2 and the ER luminal ss-PLIN2, and more than 91% of the lipid droplets were co-stained with both markers (Fig. 2C).

Next, targeting of the signal-sequence-containing marker proteins into the ER lumen was assessed by western blotting. Whereas the cytosolic version of Erg6–GFP essentially ran as a discrete band at its expected molecular mass (∼70 kDa), ss-Erg6–GFP migrated as two bands of higher molecular mass, irrespective of whether the protein was expressed in wild-type cells or in cells lacking lipid droplets (Fig. 3A). Analysis of the primary sequence of Erg6 revealed the presence of two predicted N-glycosylation sites, suggesting that the high-molecular-mass bands observed in ss-Erg6 are due to N-linked glycosylation of the protein, a posttranslational modification that is confined to the luminal compartment of the ER. PLIN2 and PLIN3, by contrast, had no predicted N-glycosylation sites, but their signal-sequence-containing versions still ran as multiple bands with lower electrophoretic mobility compared to their cytosolic versions, suggesting that these ss-PLINs are subject to O-linked glycosylation. Accordingly, treatment of cell lysates with endoglycosidase H, which removes N-linked glycans, resulted in a downshift of ss-Erg6 and of that of a glycosylated ER protein, Wbp1, a subunit of the oligosaccharyl transferase complex, which served as a control for the endoglycosidase treatment (Fig. 3B). The migration of ss-PLIN2 and ss-PLIN3, however, was unaffected (Fig. 3B). By contrast, treatment with trifluoromethanesulfonic acid (TFMS), which cleaves O-glycosidic bonds, resulted in the conversion of the slow-migrating forms of ss-PLIN2 and ss-PLIN3 into faster migrating forms of these proteins (Fig. 3C). TFMS treatment also converted the high-molecular-mass form of Gas1 (130 kDa), a glycosylphosphatidylinositol (GPI)-linked cell surface protein, into its deglycosylated form (∼105 kDa). Tgl1(Δtm) and PLIN1 were not included into these studies, because the cytosolic and signal-sequence-containing variants of these proteins migrated as a single band upon separation by SDS-PAGE, indicating that the translocated versions of these proteins were not subject to posttranslational modifications (data not shown). Taken together, these results show that ss-Erg6, ss-PLIN2 and ss-PLIN3 are subject to glycosylation, and hence are translocated into the luminal compartment of the ER.

To further confirm that the signal-sequence-containing reporter proteins are present in the ER lumen, we isolated microsomes and tested whether the proteins would be accessible to degradation by proteinase K. Consistent with a cytosolic accessibility of Erg6, Tgl1(Δtm) and PLIN1–PLIN3, these proteins were rapidly degraded upon incubation of microsomes with proteinase K (Fig. 3D). The signal-sequence-containing versions of these proteins, however, were protected from proteinase K degradation and they became protease sensitive only upon addition of detergents, demonstrating that these proteins are in a luminal compartment. The ER luminal chaperone Kar2 was degraded only after detergent addition, showing that microsomes were intact during the experiments. Proteinase K accessibility of ss-Erg6 and ss-PLIN3 was independent of whether the cells contained lipid droplets or not. This indicates that the circular ER labeling observed in Figs 1B and 2B in cells lacking lipid droplets is indeed due to an ER luminal localization rather than an association of these proteins with the membrane periphery (Fig. S1A).

To test whether the signal-sequence-containing lipid-droplet-localized markers were also enriched on lipid droplets, we performed subcellular fractionation and isolated lipid droplets by flotation on Ficoll density gradients. ss-Erg6 and ss-Tgl1(Δtm) were enriched on isolated lipid droplets to a similar degree, as were the corresponding non signal-sequence-containing proteins Erg6, Tgl1(Δtm) and Ayr1, a marker protein for lipid droplets (Athenstaedt and Daum, 2000) (Fig. 3E). We have previously shown that the cytosolic PLINs are greatly enriched on isolated lipid droplets when expressed in yeast (Jacquier et al., 2013). A similar result was found for the signal-sequence-containing PLIN1–PLIN3 proteins (Fig. 3E).

To examine whether it is indeed the glycosylated forms of these proteins that are enriched on lipid droplets, we compared the migration of the cytosolic versions of Erg6, PLIN2 and PLIN3 to their ER luminal versions isolated from lipid droplets. ss-Erg6, ss-PLIN2 and ss-PLIN3 from isolated lipid droplets showed bands of higher molecular mass compared to their cytosolic variants, indicating that the glycosylated versions of these ER luminal proteins are enriched on lipid droplets (Fig. 3F).

To address the possibility that the signal-sequence-containing proteins are inefficiently translocated into the ER lumen or that they get retro-translocated back into the cytosol resulting in an lipid droplet labeling due to trace amounts of these proteins in the cytosol, we fused them to an in vivo dual topology reporter system composed of a fusion between the hemagglutinin (HA) epitope, invertase (SUC2) and HIS4C (Kim et al., 2003b). HIS4C encodes histidinol dehydrogenase, which catalyzes the conversion of histidinol into histidine and thereby allows his4 mutant cells to grow on medium containing histidinol. Under standard conditions, this conversion takes place in the cytosol. Therefore, when His4 is targeted to the ER lumen or when it is fused to an ER luminal protein, histidine is not produced. This growth readout thus provides a sensitive assay for the efficiency of translocation and has been used to isolate mutants in the translocon (Deshaies and Schekman, 1987). his4Δ cells expressing the dual topology reporter fused C-terminally to the lipid droplet marker proteins containing or lacking the signal sequence were then assayed for growth on histidinol-containing plates. Whereas all the strains expressing the cytosolic versions of the lipid droplet marker proteins were capable of growing on this medium, those expressing the signal-sequence-containing marker proteins and those bearing only an empty plasmid failed to grow even though they all expressed the fusion constructs as monitored by western blotting (Fig. S2A,B).

To further probe the accessibility of the signal-sequence-containing PLINs to cytosolic factors, we introduced a biotinylation site into these reporters. Biotinylation occurs by a cytosolic biotin-CoA ligase and thus inefficient translocation of these reporters into the ER lumen or their retro-translocation into the cytosol, would allow access of the ligase to these reporters resulting in their biotinylation. Detection of biotinylated proteins by western blotting with avidin, revealed that the cytosolic version of PLIN3 is efficiently biotinylated, whereas the signal-sequence-containing PLIN3 was not biotinylated. ss-PLIN3, however, became a substrate for biotinylation when the cell extract was incubated with detergents (1% Triton X-100), an ATP-regenerating system and free biotin for 2 h at 30°C, indicating that the ER luminal version of PLIN3 is a substrate for biotinylation in vitro but not in vivo, most likely because ss-PLIN3 is shielded from the ligating enzyme by a membrane barrier (Fig. S1B).

To test whether PLINs can penetrate membranes, we targeted PLIN3 to mitochondria by adding the mitochondrial presequence of subunit 9 of the F0 ATPase of Neurospora crassa to the N-terminus of GFP–PLIN3. The resulting mito-GFP–PLIN3 fusion colocalized with the mitochondrial marker mito-RFP but not with Nile-Red-stained lipid droplets, indicating that PLIN3 can be targeted and confined to a membrane-enclosed compartment (Fig. S1C).

By using an inducible strain in which turning on expression of one of the neutral lipid biosynthetic enzymes, LRO1, in a background where the remaining three genes for neutral lipid synthesis are deleted, results in lipid droplet formation, we aimed to assess whether a luminal lipid-droplet-binding protein, such as ss-PLIN3, could be recruited to growing lipid droplets. Therefore, the GAL-LRO1 strain expressing either cytosolic PLIN3 or ss-PLIN3 was shifted from glucose to galactose medium and the time-dependent formation of lipid droplets was monitored after staining cells with Nile Red. Nile-Red-positive punctate structures were first observed ∼30 min after galactose induction, and the structures then grew in size and number as expression of LRO1 continued (Fig. 4). Cytosolic PLIN3 was first observed to concentrate over the Nile-Red-positive puncta after 4 h of induction. The ER luminal ss-PLIN3, by contrast, already displayed punctate localization at Nile-Red-positive structures 30 min after induction of LRO1, showing that growing lipid droplets are accessible from within the lumen of the ER. In addition, the data also indicate that the ER luminal protein can recognize and concentrate over growing lipid droplets long before these structures become decorated by the corresponding cytosolic lipid-droplet-binding protein, possibly because the concentration of these proteins in the ER lumen is higher than in the cytosol.

The data so far indicate that ER luminal proteins with a high affinity to lipid droplets can target to lipid droplets but they do not allow to discriminate whether this targeting occurs only at the early stages of lipid droplet biogenesis or whether targeting from within the ER lumen is also possible on lipid droplets that have already been formed. To address this question, we developed a mating-based assay in which one of the mating partners contributes mature lipid droplets that are marked by Erg6–mCherry whereas the other mating partner lacks lipid droplets (lro1Δ dga1Δ are1Δ are2Δ) but expresses an ER luminal GFP-tagged ss-PLIN3. At the pre-fusion stage, these two marker proteins do not colocalize, as they are expressed in two separate cells. However, upon cell fusion and subsequent karyogamy, the ER luminal ss-PLIN3 rapidly colocalizes with lipid droplets as evidenced by the line scan at the 15-min time point (Fig. 5A,D). Merging of the ER luminal ss-GFP–PLIN3 onto the pre-existing lipid droplets appeared to occur with a constant rate onto the majority of lipid droplets, indicating that there are no subpopulations of lipid droplets that are shielded from access by the ER luminal probe (Fig. 5B,C). Interestingly, the only outliers shown in the box plot of Fig. 5C are lipid droplets with a high ratio of ss-GFP-PLIN3 to Erg6-mCherry. Visual inspection of these outliers indicate that they are likely due to the formation of new lipid droplets within the part of the ER that originates from the mating partner who originally lacked the respective biosynthetic enzymes, suggesting that these lipid droplets are newly formed during the mating event, probably by movement of the biosynthetic enzymes into the ER of the ss-GFP–PLIN3-expressing mating partner (Fig. S3). Taken together, these data indicate that lipid droplets remain accessible to ER luminal probes even once they have fully matured. The data also excludes the presence of individual cytosolic lipid droplets, as they would be shielded from rapid access by the ER luminal probe.

To examine whether expression of the ER luminal PLINs affects the morphology, membrane structure, or ER association of lipid droplets, we fixed cells expressing either cytosolic or ER luminal versions of PLIN1 or PLIN3 by high-pressure freezing, embedded cells in Spurr's resin, and analyzed their morphology by transmission electron microscopy. Lipid droplets from cells expressing either cytosolic or ER luminal versions of these PLINs appeared as electron translucent structures, characteristic for lipid droplets. Their morphology, however, was indistinguishable from lipid droplets of wild-type cells (Fig. 6; Fig. S4). In all cell types, lipid droplets were frequently observed in close association with the nuclear envelope or with the ER membrane, which appear as electron translucent, ribosome free ribbons, as indicated by the yellow arrows (ER) and arrowheads (nuclear envelope) in Fig. 6 and Fig. S4. On some lipid droplets, ribosome-free zones that appeared to partially or even fully encircle the lipid droplet were observed, indicating that at these places lipid droplets were separated from the bulk of the cytosol (red arrows in Fig. 6, Fig. S4). Whether these ribosome free zones are due to the presence of an additional membrane layer, however, cannot unambiguously be resolved from these two-dimensional sections of a possibly complex three-dimension structure. These data thus indicate that the expression of either cytosolic or ER luminal PLINs does not substantially affect the morphology of lipid droplets or their association with the ER membrane, but they cannot conclusively resolve the question as to whether lipid droplets are located inside or only at the ER membrane. The data can also not exclude that expression of ER luminal PLINs could have subtle effects on the topology of lipid droplets that would escape direct visualization by electron microscopy.

So far we have shown that lipid droplets are accessible to ER luminal lipid-droplet-binding proteins in yeast. To examine whether this is a unique property of yeast lipid droplets, we examined lipid droplet targeting of a signal-sequence-containing PLIN2 protein in mammalian cells. Therefore, we transfected Schwann and HEK293T cells with plasmids expressing cytosolic PLIN2 or with PLIN2 containing an N-terminal signal sequence from BiP (amino acids 1–19; also known as HSPA5), an ER luminal chaperone. GFP–PLIN2 labeled punctate structures and colocalized with mCherry–PLIN2, a construct that has been previously shown to label lipid droplets (Eyre et al., 2014) (Fig. 7A). The signal obtained from the mammalian signal-sequence-containing PLIN2 (mss-GFP–PLIN2) also revealed staining of circular structures; however, the fluorescence signal was weak, suggesting that the protein might be subject to rapid degradation, possibly through ERAD (Xu et al., 2005). Treating cells with the proteasome inhibitor MG132 resulted in improved signal intensity and revealed extensive colocalization of the ER luminal protein with the cytosolic mCherry–PLIN2 over lipid droplets, demonstrating that targeting of ER luminal proteins to lipid droplets is not a unique feature of yeast cells but is also observed in mammalian cells. To confirm that mss-PLIN2 is indeed translocated into the ER lumen we again isolated microsomes and tested accessibility of both cytosolic PLIN2 and mss-PLIN2 to degradation by proteinase K. Whereas the cytosolic version of PLIN2 was readily degraded in the presence of proteinase K, mss-PLIN2 was protected from the protease and degraded only when microsomes were treated with detergents (Fig. 7B). Thus mss-PLIN2 behaved the same as the ER luminal chaperone calreticulin, demonstrating that the protein indeed translocated into the ER lumen. Cytosolic and ER luminal forms of PLIN2 both run as a discrete band with the same electrophoretic mobility, indicating that the ER luminal version of PLIN2 is not subject to glycosylation as is the case in yeast. This difference is likely explained by the fact that O-glycosylation in yeast starts in the ER lumen whereas in higher eukaryotes this reaction is generally confined to the Golgi (Lommel and Strahl, 2009). These data thus show that lipid droplets of yeast, as well as those of mammalian cells, are accessible from within the lumen of the ER.

The results presented here reveal that lipid droplets are accessible from both sides of the ER membrane, that is, from the luminal compartment and from the cytosol. The data thus indicate that lipid droplets form a domain within the ER that is recognized by proteins that have a high affinity for lipid droplets regardless of whether these proteins are present in the cytosol or within the ER lumen. Although the precise signal that is recognized by lipid-droplet-localized proteins is not yet well defined, it has been speculated that some of these proteins, such as PLIN3, recognize lipid-packaging defects (Bulankina et al., 2009). This is supported by recent results indicating that amphipathic helices present in a number of lipid-droplet-localized proteins, including PLINs, are sufficient for lipid droplet targeting (Grippa et al., 2015; Rowe et al., 2016). Thus, similar to proteins containing lipid-binding amphipathic helices, as present in proteins with an ALPs motif or those of the N-BAR family, which recognize lipid packaging defects induced by membrane curvature, PLIN3 and possibly other lipid-droplet-targeted proteins harboring amphipathic helices, might recognize altered spacing of phospholipid headgroups, possibly induced by the presence of cone-shaped or neutral lipids within a flat bilayer membrane (Bulankina et al., 2009; Drin and Antonny, 2010).

Neutral lipids are soluble to few molar percent within a phospholipid bilayer (Hamilton et al., 1983). They are highly concentrated in the lipid droplet core from where they likely diffuse into the adjacent lipid bilayer and thereby alter the spacing of phospholipid headgroups within the bilayer. The resulting local increase in phospholipid headgroup spacing in turn could be recognized by lipid-droplet-binding proteins containing amphipathic helices. In this scenario, lipid-droplet-binding proteins would thus assemble onto a membrane bilayer enriched in neutral lipids rather than directly onto the monolayer that wraps the hydrophobic core of the lipid droplet.

To account for the observed colocalization of cytosolic and ER luminal lipid droplet proteins, one might postulate that lipid droplets that emanate from the cytosolic leaflet of the ER membrane become encapsulated by an additional layer of ER membrane, resulting in an ‘egg cup’ arrangement in which the lipid droplets is held in place by the ER membrane similar to a cup holding an egg (Robenek et al., 2006) (Fig. 8A). In this ‘egg cup’ model, the limiting phospholipid monolayer of the lipid droplet is directly decorated by cytosolic PLINs (red ovals in Fig. 8A), whereas the ER luminal PLINs cover the structure from within the ER extensions that hold the lipid droplet in place (green ovals in Fig. 8A).

Alternatively, if the lens of neutral lipids that is thought to form in the hydrophobic core of the ER bilayer early during lipid droplet biogenesis bends towards the ER lumen rather than towards the cytosol, as is frequently postulated, it could be enwrapped by the ER to yield a structure that would morphologically resemble a lipid droplet in close proximity to the ER, facing the cytosol, as is typically observed by electron microscopy (Fig. 6; Fig. S4; Fig. 8B). Topologically, however, this type of lipid droplet would position its neutral lipid core into the ER lumen, thus allowing direct binding of ER luminal PLINs onto the limiting lipid monolayer (green ovals in Fig. 8B). In this model, cytosolic PLINs could still assemble onto the cytosolic leaflet of the ER membrane, again by recognizing altered lipid spacing induced by the diffusion of triacylglycerol into the confining membrane (red ovals in Fig. 8B). Thus, in this model, lipid droplet biogenesis is topologically identical to the biogenesis of lipoprotein particles, which is initiated by the translocation of a lipoprotein into the ER lumen. This nascent apo-lipoprotein then becomes lipidated by the action of a microsomal triglyceride transfer protein (MTP), an ER luminal protein that forms a complex with protein disulfide isomerase (PDI), and thereby matures into a lipoprotein particle which is secreted into the circulation (Fisher and Ginsberg, 2002; Hussain et al., 2003). In support of such an ER luminal localization of lipid droplets, lipid droplet formation and lipoprotein processing in hepatocytes are two closely linked processes. Apo-lipoprotein B-100 tightly associates with nascent lipid droplets when its degradation is inhibited. This association requires MTP activity and thus occurs between an ER luminal protein and lipid droplets (Ohsaki et al., 2008). Remarkably, immunogold labeling of such arrested apo-lipoproteins reveals that the lipoproteins are separated only by a membrane monolayer from the lipid droplet, the lipid droplet surface is thus directly accessible to an ER luminal protein (Ohsaki et al., 2008).

Both models of lipid droplet biogenesis outlined in Fig. 8 could account for the observed colocalization of cytosolic and ER luminal PLINs. It is interesting to note, however, that lipid droplets have recently been shown to bud into the ER lumen in cells lacking the fat-storage-inducing transmembrane (FIT) proteins (Choudhary et al., 2015). FIT proteins constitute an evolutionary conserved class of ER proteins that bind triacylglycerol in vitro and they are important for neutral lipid storage in vivo (Gross et al., 2011; Kadereit et al., 2008). Deficiency of FIT2 in the adipose tissue of mice results in progressive lipodystrophy of white adipose depots and metabolic dysfunction (Miranda et al., 2014). In yeast, mouse fibroblasts and C. elegans, lack of FIT proteins results in budding of lipid droplets into the ER lumen (Choudhary et al., 2015). This observation has been taken to suggest that FIT proteins are required for budding of lipid droplets from the ER towards the cytosol. An alternative interpretation, however, is that the absence of FIT proteins reveals a process that is morphologically not discernible in their presence. For example, if FIT proteins were important to connect or coordinate the expansion of the lipid droplet monolayer with that of the adjacent ER bilayer, their absence could result in the accumulation of intraluminal lipid droplets, similar to those labeled as ‘nascent stage’ in Fig. 8B.

Both models of lipid droplet biogenesis shown in Fig. 8 postulate stable membrane domains with high curvatures. Stabilizing such highly curved membrane domains is likely to require specialized proteins, such as the hairpin-anchored membrane proteins that are frequently found to localize to lipid droplets and non-bilayer forming lipids. The requirement for such non-bilayer forming lipids might explain why phospholipids from isolated lipid droplets have a distinct fatty acid composition from those of the ER and might also account for the observation that lipid droplet formation in yeast is dependent on the formation of diacylglycerol, a lipid that rapidly flips between membrane leaflets (Adeyo et al., 2011; Schneiter et al., 1999; Tauchi-Sato et al., 2002). The observations reported here that lipid droplets in both yeast and mammalian cells are accessible to ER luminal proteins further support the notion that lipid droplets are closely associated with the ER, and they indicate that this association does not only occur early during lipid droplet biogenesis but that it is likely maintained throughout its lifetime.

Yeast strains and their genotypes are listed in Table S1. Double- and triple-mutant strains were generated by crossing of single mutants and by gene disruption, using PCR deletion cassettes and a marker rescue strategy (Gueldener et al., 2002).

The signal-sequence-containing reporters were expressed from a pGREG506-based plasmid in which the GAL promoter was replaced by an ADH promoter and the proteins were fused to the PRY1 ER signal sequence, amino acids 1–19 (Choudhary and Schneiter, 2012). Cloning and expression of PLINs in yeast was performed as previously described (Jacquier et al., 2013). For mammalian expression, we used pLENTI6-mCherry-Plin2 (Eyre et al., 2014) and GFP–PLIN2 and mss-GFP–PLIN2 that were cloned into pSiCoR-GFP-EF1α (Jacob et al., 2014). All constructs were verified by sequencing. Crude lipid droplets were prepared as described previously (Leber et al., 1994).

Erg6, Tgl1(Δtm) and the PLINs were detected using a monoclonal antibody against GFP (#11814460001, Roche Diagnostics, Rotkreuz, Switzerland, dilution 1:2000). Primary antibodies against Wpb1 (Markus Aebi, ETH Zurich, Switzerland, dilution 1:10,000), Gas1 (Howard Riezman, University of Geneva, Switzerland, dilution 1:2000), Kar2 (Randy Schekman, University of California at Berkeley, CA, dilution 1:5000), Ayr1 (Günther Daum, TU-Graz, Austria, dilution 1:10,000) and calreticulin (Beat Schwaller, University of Fribourg, Switzerland, dilution 1:3000) were detected by using horseradish peroxidase (HRP)-conjugated secondary antibodies (Santa Cruz Biotechnology, Dallas, TX, #sc-2030 and #sc-2302, dilution 1:10,000). Western blots and fractionation experiments were repeated at least two times with essentially similar results.

Proteinase K and endo H treatment were performed as previously described (Tatzer et al., 2002; Choudhary and Schneiter, 2012).

Localization of GFP- and mCherry-tagged proteins was performed by fluorescence microscopy of fixed mammalian cells using a Leica TCS SP-II confocal microscope. For living yeast cell imaging, either a Leica TCS SP5 confocal microscope with LAS AF software, a Visitron VisiScope CSU-W1 (Visitron Systems, Puchheim, Germany) or a DeltaVision Elite imaging system (GE Healthcare, Pittsburgh, PA) was used. The Delta Vision Elite imaging system consisted of an Olympus 1X71 inverted microscope equipped with a CCD camera (CoolSNAP HQ², Photometrics, Tucson, AZ). Images were acquired with a U PLAN S-APO 100×1.42 NA oil immersion objective (Olympus) and a GFP or mCherry filter set. Six to eight 0.2 μm optical sections were deconvolved using the iterative constrained deconvolution program in softWoRx (Applied Precision). Single sections are displayed. Confocal images produced with the Leica TCS SP5 microscope were recorded with an APO 40×1.3 NA or an APO 63×1.3 NA oil immersion objectives (Leica) with a zoom of 6 or 3, respectively. The Visitron spinning disk CSU-W1 consisted of a Nikon Ti-E inverted microscope, equipped with a CSU-W1 spinning disk head with a 50-µm pinhole disk (Yokogawa, Tokyo, Japan), an Evolve 512 (Photometrics) EM-CCD camera, and a PLAN APO 100× NA 1.3 oil objective (Nikon). Live cells were stained with Nile Red (10 mg/ml, Sigma-Aldrich, St Louis, MO) for 1 min at room temperature and washed twice with PBS. Quantification of the GFP and mCherry signals was performed using Fiji software (Schindelin et al., 2012). Colocalization was assessed manually by counting the number of lipid droplets per cells that were stained by the GFP- and mCherry-tagged marker proteins. The rate of fluorescent protein transfer onto lipid droplets upon mating was calculated from maximal intensity projections of seven confocal sections spaced by 0.6 µm. Data were analyzed using R (version 3.2.5). All microscopic experiments were performed at least two times with essentially similar results.

HEK293T cells (ATCC) were grown in DMEM (Gibco) containing 10% fetal bovine serum (Biochrom) and 0.2% penicillin-streptomycin (Gibco). Purified primary rat Schwann cell cultures were obtained and grown as described previously (Jacob et al., 2008). For immunofluorescence, Schwann cells were treated 48 h after transfection with 10 µM MG132 for 6 h, washed with PBS, fixed with 4% paraformaldehyde for 15 min at room temperature, washed again in PBS, incubated for 30 min in 0.3% Triton X-100 in PBS and 5 min in DAPI in PBS, and mounted in Citifluor (Agar Scientific). Microsomes were prepared from HEK293T cells and protease protection assays were performed as described previously (Kaznacheyeva et al., 1998).

The respective genes were amplified either with or without the signal sequence from Pry1 (amino acids 1–19) and cloned into pJK90 by homologous recombination in strain STY50. Plasmid pJK90 contains the OST4 gene, three HA tags, part of the SUC2 gene and HIS4C (Kim et al., 2003a). Growth assays and western blots of dual topology reporters were repeated two times with similar results.

To target GFP–PLIN3 into mitochondria, PLIN3 was cloned into pVT100U, which contains the first 69 amino acids of subunit 9 of the F0 ATPase of Neurospora crassa as a mitochondrial presequence (Westermann and Neupert, 2000).

The biotinylatable versions of cytosolic PLIN3 were generated by amplifying the biotin acceptor peptide from YEpURA-CUP1-HBTubiquitin (Tagwerker et al., 2006) and cloning into SpeI-linearized pGREG576ADH-GFP-PLIN3 (Jacquier et al., 2013). The biotinylatable version of the ER luminal ss-GFP–PLIN3 was generated by PCR ligation of the biotin acceptor site from YEpURA-CUP1-HBTubiquitin and GFP–PLIN3 from pGREG576-ADH-GFP-PLIN3. All plasmids were verified by sequencing.

For in vitro biotinylation, cells were disrupted with glass beads in lysis buffer (100 mM NaPO₄, pH 7.5, 0.2 M sorbitol, 5 mM MgCl₂, 2 mM PMSF and 1% Triton X-100). 50 µg of cell extracts was incubated for 2 h at 30°C in the presence of biotin (100 µM), CoA (500 µM) and an ATP-regenerating system (75 mM phosphocreatine, 5 mM ADP, 10 mM ATP, 10 mM MgCl₂ and 25 units creatine phosphokinase). The biotinylation assay was repeated two times with similar results.

For cryofixation, mid-log phase cells were transferred into lecithin-treated 0.1 mm deep Leica membrane carriers and frozen by high-pressure freezing (Studer et al., 2001). The frozen specimens were freeze-substituted in acetone containing 2% osmium tetroxide (EMS, Hatfield, PA), 0.1% uranyl acetate (EMS), 1% methanol and 3% water (AFS, Leica-Microsystems) (Walther and Ziegler, 2002). The substitution protocol was as follows: −90°C for 27 h, −60°C for 8 h and −30°C for another 8 h. The samples were then washed four times for 30 min in 100% acetone, warmed up to room temperature and embedded in Spurr's resin (through a series of 30%, 50%, 70% and 2×100%). Blocks were polymerized at 60°C for 5 days. Ultrathin sections (75 nm) were collected on formvar-coated grids, post stained with uranyl acetate and lead citrate, visualized with a FEI Tecnai Spirit TEM (Hillsboro, OR), and images were recorded on a FEI Eagle CCD camera and processed with Photoshop.

We thank members of the laboratory for their support, suggestions and comments on the manuscript, A. Conzelmann for discussions and comments on the manuscript, A. Puoti for localizing ss-PLINs in nematodes, O. Duman for help with the mammalian cell culture, and M. Beard, A. Conzelmann, J. Nunnari, M. Rose and B. Schwaller, for reagents. Electron microscopy sample preparation and imaging were performed with devices supported by the Microscopy Imaging Center (MIC) of the University of Bern.