Control of lipid droplet size in budding yeast requires the collaboration between Fld1 and Ldb16 Free

Lipid droplets (LDs) are found in virtually all eukaryotes. The neutral lipids triacylglycerol and sterol ester, stored within LDs, provide an energy source and membrane building blocks for cells to maintain lipid homeostasis (Daum et al., 1998; Leber et al., 1994; Walther and Farese, 2009; Walther and Farese, 2012). Most cells keep their LDs to a certain size, which suggests that cells control the number and size of their LDs to balance lipid usage and storage processes (Yang et al., 2012).

LDs are believed to originate from the outer leaflet of the endoplasmic reticulum (ER). However, an exact mechanism for LD biogenesis is still lacking. During LD growth, the synthesis and packaging of neutral lipids are coordinated with the expansion of the phospholipid monolayer on the surface of emerging LDs. Phosphatidylcholine (PtdCho) is a key regulator of the control of LD size. The PtdCho -synthesizing enzyme CTP:phosphocholine cytidylyltransferase (CCT) targets to the LD surface in Drosophila melanogaster S2 and mammalian cells (Krahmer et al., 2011). Compromised PtdCho synthesis results in supersized LDs in both S2 cells and in budding yeast (Fei et al., 2011b; Krahmer et al., 2011).

LD size could increase through other mechanisms. Human seipin and the yeast, Saccharomyces cerevisiae, ortholog ‘few lipid droplets protein 1’ (Fld1), are the evolutionarily conserved components for LD size maintenance. The recessive mutations of the human seipin gene (BSCL2, termed seipin hereafter) are linked to the most severe form of congenital generalized lipodystrophy type 2 (CGL2) (Payne et al., 2008; Simha and Garg, 2003). Two independent genetic screens for yeast mutants with aberrant LD morphology have both identified fld1Δ mutant strains that accumulate two distinct pools of LDs: small clustered and supersized (Fei et al., 2008b; Szymanski et al., 2007). Aberrant LD morphology is a general phenotype of seipin deficiency found in various species. Knockout of seipin in the salivary gland and midgut of Drosophila caused LDs to become larger and more numerous (Cui et al., 2011). The fibroblasts and lymphocytes derived from CGL patients accumulated small and numerous LDs (Boutet et al., 2009; Szymanski et al., 2007). Seipin knockout in mice also causes severe CGL and small unilocular LDs in the adipose tissues (Cui et al., 2012). Interestingly, human seipin, but not the lipodystrophy-linked seipin mutant A212P, complements the LD defects in fld1Δ yeast (Fei et al., 2008b; Szymanski et al., 2007), implying that lipodystrophy caused by mutations in seipin might link to its LD-related functions.

How seipin/Fld1 controls LD size remains obscure (Cartwright and Goodman, 2012). Because LDs in fld1Δ appear more fusogenic, deficiency of Fld1 might cause LD enlargement via LD fusion (Fei et al., 2008b). Deficiency of seipin in various species results in changes in neutral lipid and fatty acids to varying degrees (Boutet et al., 2009; Fei et al., 2011a; Fei et al., 2008b; Tian et al., 2011). In fld1Δ cells, genes associated with phospholipid biosynthesis pathways are upregulated and phosphatidic acid, but not PtdCho, levels are higher, implying that seipin/Fld1 deficiency might have a pleiotropic impact on phospholipid distribution (Fei et al., 2011b). Although seipin/Fld1 might modulate lipid metabolism, the protein contains no known domain or enzymatic fold. Thus, seipin/Fld1 might facilitate lipid distribution and LD maintenance through a unique mechanism. The finding that seipin/Fld1 is a transmembrane protein residing in the ER and is enriched at the junctions between the ER and LD (termed ER-LD contact sites) argues that the protein might be directly involved in coupling neutral lipid synthesis with LD assembly (Szymanski et al., 2007). Although it is not clear how Fld1 is organized in the membrane, Fld1 isolated from yeast exists as a discrete homo-oligomer consisting of approximately nine copies and appears as a toroid, as shown by negative staining electron microscopy (EM) (Binns et al., 2010). Thus, it has been proposed that seipin/Fld1 might have structural roles at the ER–LD contact site, such as forming a collar, to facilitate LD assembly processes (Binns et al., 2010).

Unraveling the function of seipin/Fld1 requires finding additional molecules that work together with, or function in the same pathway as, seipin/Fld1. Here, we report a novel Fld1-interacting protein termed low dye-binding protein 16 (Ldb16) and show that Fld1 and Ldb16 are mutually dependent for the maintenance of LDs. Fld1 is needed for the targeting of Ldb16 to the ER–LD contact site, a process involving the formation of a transmembrane Fld1–Ldb16 complex mediated by their hydrophobic helices. We suggest that the architecture of the transmembrane Fld1–Ldb16 complex at the ER–LD contact sites is crucial for the control of LD size in yeast.

The deletion of seipin homolog FLD1 in yeast results in aberrant LDs (Fei et al., 2008b; Szymanski et al., 2007). To understand how LD assembly is controlled in budding yeast, we performed genome-wide screening and found that ldb16Δ had LD phenotypes (unpublished results). Ldb16 was first identified as a mutant that shows low affinity to dye, which suggests a glycosylation defect (Corbacho et al., 2005). Although a reduced LD number has been found in ldb16Δ yeast (Fei et al., 2008a), nothing is known about the molecular function of Ldb16 in yeast or its homologs in other species. By staining LDs with BODIPY, or direct visualization by the differential interference contrast (DIC), we found that ldb16Δ, like fld1Δ, accumulated two types of LDs: supersized and small clustered (Fig. 1A). Quantitative results showed that the average size of LDs in fld1Δ was slightly larger than that in ldb16Δ (Fig. 1B). The small clustered LDs appeared more frequently in ldb16Δ than fld1Δ cells. Further examination by thin-sectioned EM revealed that the small clustered LDs in ldb16Δ, like fld1Δ, often congregated with ER-like structures (Fig. 1C). Importantly, neither of the supersized nor small clustered LDs was seen in the wild type (Fig. 1A-C).

Phospholipid metabolism is important for the control of LD size. In yeast, several supersized LD (SLD) mutants were found to be defective in the phosphatidylethanolamine N-methyltransferase (PEMT) pathway, which involves a series of methylation reactions that convert phosphatidylethanolamine to PtdCho (Fei et al., 2011b). Consistent with the previous findings (Fei et al., 2011b), supersized LDs formed when budding yeast cells lacking functional PEMT pathway proteins, such as the methyltransferases Cho2 and Opi3, and transcriptional activators Ino2 and Ino4, were cultured in minimal [synthetic complete (SC)] media (Fig. 2A). The SLD mutants, such as opi3Δ, did not accumulate small clustered LDs as in fld1Δ and ldb16Δ cells (Fig. 1C). Addition of the phospholipid precursors choline or inositol, but not ethanolamine, restored normal LD morphology in the SLD mutants (Fig. 2A). Intriguingly, adding inositol, but not choline or ethanolamine, reduced supersized LDs in fld1Δ and ldb16Δ cells; however, the LD size of wild-type cells was not affected by the addition of these phospholipid precursors (Fig. 2A). Moreover, in fld1Δ and ldb16Δ cells, inositol addition increased the fraction of small clustered LDs, rather than converting supersized LDs to normal ones (Fig. 2A). Thus, we conclude that phospholipid metabolism might be altered similarly in ldb16Δ and fld1Δ. Analyses of neutral lipid and phospholipid profiles in ldb16Δ and fld1Δ showed only subtle differences when compared with that in the wild type (supplementary material Fig. S1). However, microarray data showed an elevated expression of multiple enzymes involved in phospholipid metabolism in ldb16Δ cells (supplementary material Fig. S1), similar to the previous findings for fld1Δ (Fei et al., 2011b).

Furthermore, we found that fld1Δ and ldb16Δ cells showed higher sensitivity to terbinafine, a drug that perturbs ergosterol biosynthesis by inhibiting the function of squaline expoxidase [also known as squalene monooxygenase (Erg1)], whereas sensitivity to terbinafine of all SLD mutants was similar to that of the wild type (Fig. 2B). Moreover, immunoblotting results indicated that protein levels of Fld1 and Ldb16 were not affected in the SLD mutants, suggesting that the Fld1 and Ldb16 machinery in the SLD mutants probably remains intact (Fig. 2C). Interestingly, the level of Ldb16 was reduced in fld1Δ cells, supporting a close relationship between Fld1 and Ldb16 (Fig. 2C). In summary, the fld1Δ and ldb16Δ mutants, but not SLD mutants, share similar phenotypes.

Based on the similar phenotypes of ldb16Δ and fld1Δ, we tested whether Ldb16 is a partner of Fld1. We show that Ldb16 and Fld1 were fractionated into the ER-enriched P13 fractions and were fully stripped from the P13 pellets only when detergent was added (Fig. 3A,B). The ER lumenal protein Kar2 served as a control for the integrity of the ER. Thus, Ldb16, like Fld1, is likely a transmembrane protein. In addition, using Fld1 as bait in tandem affinity purification (Fld1-TAP) pulled down Ldb16–13×Myc and vice versa, whereas neither Fld1-TAP nor Ldb16-TAP interacted with the ER–LD protein Erg1 or the ER protein sterol O-acyltransferase 2 (Are2) (Fig. 3C). Yeast two-hybrid assay further confirmed the Fld1–Ldb16 interaction and Fld1 self-interaction (Fig. 3D). Thus, Fld1 and Ldb16 form a complex.

It has been reported that Fld1 localizes to the ER–LD contact sites (Szymanski et al., 2007). We found that Ldb16–3×GFP (green fluorescent protein) also formed puncta that were in the ER and largely associated with LDs (supplementary material Fig. S2). The quantified data indicated that ∼50% of the Fld1 and Ldb16 puncta colocalized, whereas the remaining puncta contained either one of these proteins alone (Fig. 3E). A triple-color imaging technique was used to further confirm the relationship of Fld1 and Ldb16 puncta with ER and LDs. We show that Ldb16–Venus was enriched in the junctions between the ER and LDs, labeled by Elo3 tagged with cyan fluorescent protein (CFP) and Erg6–mCherry, respectively, and that Fld1–mCherry and Ldb16–Venus colocalized at one side of LD marked by Pet10–CFP (Fig. 3F). Quantitative data further revealed that ∼87% of the colocalized Fld1–Ldb16 puncta were associated with the neck of LDs, indicative of a strong correlation with the ER–LD junctions. Thus, the Fld1–Ldb16 complex probably regulates LD size at the same ER–LD contact sites. Moreover, the colocalized Fld1–Ldb16 puncta, for most of the time, are stably associated with the neck of LDs. As we routinely pulled down ∼50% of Ldb16 and Fld1 together (Fig. 3C), the interaction between Fld1 and Ldb16 is stable. Interestingly, the overexpression of Fld1 or Ldb16 failed to suppress the defects caused by the absence of the other, although small clustered LDs were more evident when Fld1 was overexpressed in ldb16Δ cells (supplementary material Fig. S3). Thus, we conclude that Ldb16 and Fld1 work in a complex; however, each carries out an essential and independent function for LD maintenance.

We next studied the topology of Ldb16 in the ER using a protease protection assay. When 2×Myc tags were fused to Ldb16 at amino (N)- or carboxyl (C)-termini, the fusion proteins were susceptible to treatment with proteases (Fig. 4A), indicating that both termini of Ldb16 face the cytoplasm. The fact that Ldb16 was efficiently degraded in the lysate with addition of detergent, a condition under which Kar2 remained stable (Fig. 4A), suggests that Ldb16 is very unstable in the absence of association with the membrane.

To understand further the importance of Ldb16 and Fld1 in the Fld1–Ldb16 complex, we mapped their interactions and functional domains using both yeast two-hybrid and pulldown assays. We found that the C-terminal cytosolic domain (residues 101–256) of Ldb16 was not required for interaction with Fld1 by either yeast two-hybrid or pulldown assays (Fig. 4B,C). Moreover, functional analyses support the notion that the C-terminal region is dispensable, as the Ldb16 fragment (residue 1–100) showed normal LDs and terbinafine sensitivity (Fig. 4D,E). The minimal functional domain of Ldb16 consists of residues 11–100, with two putative transmembrane helices (Fig. 4D,E). Further deletion from the N-terminal of Ldb16 resulted in abnormal LDs, although only a small portion (residues 31–100) of Ldb16 was sufficient for interaction with Fld1 using the two-hybrid assay (Fig. 4B). Thus, we conclude that the putative transmembrane domain of Ldb16 is important for interacting with Fld1 and for controlling LD size (Fig. 4F).

Fld1/seipin is known to adopt a topology on the ER with both termini facing the cytoplasm (Lundin et al., 2006; Melén et al., 2003) (Fig. 5A). When a similar deletion approach was applied to Fld1, using pulldown assays, we found that Fld1 self-interaction, and interaction with Ldb16, diminished when either of the putative transmembrane segments was disrupted (residues 1–265 and 21–285) (Fig. 5C). In addition, all mutants harboring these interaction defects showed abnormal LDs, increased terbinafine sensitivity and reduced Ldb16 levels (Fig. 5D,E and data not shown). Therefore, the Fld1 transmembrane segments are also crucial for Ldb16 interaction and LD maintenance. All Fld1 deletion mutants that were unable to interact with Fld1 also failed to interact with Ldb16 by yeast two-hybrid assay (Fig. 5B), suggesting that Fld1 self-interaction is required for the interaction with Ldb16. Deletion of the first and last ten residues of Fld1 had no effect on interactions with Fld1 or Ldb16 using a pulldown assay; however, the deletions caused reduced number and increased size of LDs (Fig. 5C,D). Thus, both the N- and C-terminal small cytosolic regions of Fld1 are essential for Fld1 function (Fig. 5G). Moreover, the affinity of Fld1–Fld1 reduced in ldb16Δ (Fig. 5F), implying that Ldb16 also contributes to Fld1 assembly. Overall, these data establish that the interaction between Fld1 and Ldb16 involves their putative transmembrane helices and that the interaction is crucial for maintenance of LDs.

To further understand the nature of the Fld1–Ldb16 complex, we solubilized Ldb16 and Fld1 from the ER-enriched P13 fraction and analyzed their organization using a velocity sedimentation gradient. We found that Fld1 and Ldb16 co-fractionated into the middle of the gradient (Fig. 6A), suggesting that their physical interaction was further organized into higher-order protein complexes.

We next examined Fld1 and Ldb16 in the purified ER and LD fractions to investigate how the proteins might be organized at the ER–LD contact sites. The ER-resident proteins Sec61 and Lro1 were largely recovered in the isolated ER, whereas the cytosolic protein Pgk1 was reduced in the fraction (Fig. 6B). Similarly, isolated LDs were enriched in the LD-resident proteins Erg1 and Erg6, but not Pgk1 and the ER proteins Kar2 and Lro1 (Fig. 6C). Pet10 and Ubx2, which populate ER and LDs, were found in both fractions. In fld1Δ and ldb16Δ, the main LD protein, Erg1, showed a higher level in total cell lysate, ER and LD fractions in comparison with that in wild-type controls (Fig. 6B,C). However, the level of other LD proteins, including Pet10, Ubx2 and Erg6, was reduced in the LD fraction (Fig. 6C,D). Intriguingly, in wild-type cells, Ldb16 was found only in the isolated ER, whereas Fld1 was enriched in both ER and LDs (Fig. 6B,C). Quantification of the data further indicated that the level of Fld1 was largely reduced in the LDs isolated from ldb16Δ cells (Fig. 6D), even though the steady-state level of the protein in the lysate was comparable to that of the wild type. Thus, Ldb16 might contribute to partitioning of Fld1 from ER to LDs at the ER–LD contact sites, suggesting that the Fld1–Ldb16 complex is important for LD protein composition.

To understand the organization of the Fld1–Ldb16 complex at the ER–LD contact site in vivo, we performed a colocalization experiment in the fld1Δ and ldb16Δ cells. Fld1–GFP was observed to maintain its normal localization pattern and was associated with one side of the supersized LDs labeled with Erg6–mCherry in ldb16Δ cells, indicative of localizing to the ER–LD contact site (Fig. 6E). Because Fld1 alone failed to control normal LD size in ldb16Δ, Ldb16 is probably instrumental for Fld1 function at the ER–LD contact site. By contrast, Ldb16–GFP signal was not detectable in fld1Δ cells (Fig. 6E), indicating that Fld1 is crucial for localization of Ldb16 to the ER–LD contact site.

We next asked what might cause the reduced Ldb16 level in fld1Δ mutants (Fig. 2C). Microarray and quantitative PCR results revealed that LDB16 mRNA was not significantly reduced in fld1Δ cells (unpublished results). Results from a cycloheximide chase experiment indicated that Ldb16 was a stable protein in the wild-type control but became unstable in fld1Δ (Fig. 7A). As Ldb16 is an ER-localized protein, we considered the possibility that ERAD machinery, used for protein quality control in ER (Vembar and Brodsky, 2008), might contribute to the degradation of Ldb16 by proteasomes in fld1Δ. We found that the level of Ldb16 indeed increased in fld1Δ cells treated with the proteasome inhibitor MG132, but not in controls treated with DMSO (Fig. 7B). In addition, Ldb16 levels increased when npl4-1 fld1Δ and ufd1-2 fld1Δ mutants were shifted to the nonpermissive temperature, but did not change at the permissive temperature (Fig. 7C). As Npl4 and Ufd1 are adaptors for the ubiquitin-selective chaperone Cdc48 to segregate ERAD substrates from ER for proteasomal degradation (Braun et al., 2002; Rape et al., 2001), we conclude that Ldb16 is targeted for ERAD in fld1Δ cells.

Unlike Ldb16, which is degraded quickly in fld1Δ cells, Fld1 maintained its normal level in ldb16Δ (Fig. 7D). Ldb16 levels were restored in fld1Δ cells lacking three well-established components of the ERAD-C pathway (Fig. 7D): the E3 ligase Ssm4 (also known as Doa10), the E2 enzyme Ubc7, and the Ubc7 receptor Cue1 (Biederer et al., 1997; Carvalho et al., 2006). However, Ldb16 levels were not restored in fld1Δ cells lacking the ERAD-L and ERAD-M components Hrd1 and Der1 (Fig. 7D). In addition, deletion of the ERAD-C components did not increase the level of Ldb16 in the wild type (Fig. 7D), implying that Fld1 availability is key to the regulation of Ldb16 breakdown. Thus, Ldb16 is recognized for degradation by the ERAD-C mechanism, unless it is bound to Fld1.

We next tested whether overproducing Ldb16 might be detrimental for LD morphology. Indeed, wild-type cells accumulated larger but fewer LDs when expressing additional LDB16 through the strong GPD (TDH3) promoter, whereas overexpression of Fld1 by the GPD promoter had only little effect (Fig. 7E). In addition, Ldb16 overexpression gave rise to species of higher molecular weight (Fig. 7E) that can be recognized by an antibody against ubiquitin (Fig. 7F), implying that excess Ldb16 is polyubiquitinylated and that Fld1 availability is the limiting factor for Ldb16 stability. Both Ldb16 and Fld1 also contain monoubiquitinylated species of unknown function (Fig. 7F). Overall, we conclude that Fld1 controls the stability of Ldb16 and that the ERAD-C mechanism mediates degradation of Ldb16 by proteasomes in the absence of Fld1.

A database search identified Ldb16 homologs in various fungi, and the most conserved regions reside within their N-terminal regions, in agreement with the minimal functional domain we mapped (Fig. 4). However, we found no obvious Ldb16 homologs in higher eukaryotes. To understand the relationship of human seipin to Fld1 and Ldb16, we integrated the wild-type and pathogenic seipin missense alleles into the yeast genome (Fig. 8A) controlled by the strong GPD promoter. We found that the wild-type and neuronal seipinopathy mutation, S90L, expressed in fld1Δ gave normal LDs (Fig. 8B), consistent with previous observations (Cartwright and Goodman, 2012; Fei et al., 2008b; Szymanski et al., 2007). Another neuronal seipinopathy mutation, N88S, gave normal LDs previously (Fei et al., 2008b) but larger LDs in our analyses, which might reflect a difference in expression level or quantification. Importantly, all lipodystrophy missense mutants, except T78A, resulted in aberrant LDs, ranging from supersized, such as A212P and Y187C, to very large-sized, such as A91P (Fig. 8B). Thus, the results were consistent with the notion that lipodystrophy might be attributed to the defects of seipin in LD maintenance. Interestingly, all seipin missense mutants that complemented the defect in fld1Δ were enriched at the ER–LD contact sites in yeast (supplementary material Fig. S4). By contrast, those that failed to complement the defect in fld1Δ were mislocalized. Thus, localization of seipin to the ER–LD contact site is likely to be important for its function.

We next tested whether seipin interacts with Ldb16 in yeast by pulldown assay using Ldb16-TAP. We found that seipin, unlike Fld1, failed to interact with Ldb16-TAP, even when overexpressed to a higher level than the endogenous Fld1 (Fig. 8C). Interestingly, the association of the lipodystrophy mutant G225P Fld1 (allelic to seipin A212P) with Ldb16-TAP was reduced by ∼60% compared with wild-type Fld1. Thus, the Fld1 lumenal domain mutation might affect the interaction with Ldb16 (Fig. 8C). Intriguingly, human seipin rescued LD size but did not restore Ldb16 levels in fld1Δ cells (Fig. 8D), implying that either a small amount of Ldb16 was sufficient, or human seipin rescued the LD defects in fld1Δ independently of Ldb16 in yeast.

Remarkably, overexpression of human seipin fully restored LD size in ldb16Δ and fld1Δ ldb16Δ cells, even though Ldb16 and seipin share no similarity in their sequences or domain organization (Fig. 8E). By contrast, Fld1 overexpression failed to complement LD size defects in ldb16Δ (Fig. 8F). The seipin core sequence (residues 1–280) that complemented fld1Δ (Fei et al., 2008b) also complemented ldb16Δ (Fig. 8F), indicating that the complementation is not mediated by seipin C-terminal extension, which is absent in Fld1. Furthermore, all lipodystrophy-associated seipin mutations, including the missense mutations A212P, Y187C and L91P and the truncated R275X, that were unable to complement fld1Δ also failed to complement ldb16Δ (Fig. 8F and data not shown). Thus, the role of both Ldb16 and Fld1 in LD maintenance correlates with seipin etiology during lipodystrophy. Based on these data, we propose that the functions of the Fld1–Ldb16 complex in LD maintenance converge on seipin through evolution.

The seipin recessive mutations in humans are associated with the most severe form of lipodystrophy CGL2, whereas its dominant mutations cause neuronal seipinopathies (Cartwright and Goodman, 2012). Seipin/Fld1 localizes to ER and is enriched at the ER–LD contact sites (Fei et al., 2011a; Szymanski et al., 2007). In most cell types, the protein is crucial for LD maintenance, particularly LD size and distribution. However, genetic screen and biochemical studies in yeast and other organisms have yet to identify any factor functioning together with seipin/Fld1.

In this study, we reveal that Ldb16 is important for the control of LD size in yeast (Fig. 1). Similar to fld1Δ cells, ldb16Δ cells accumulate both supersized and small clustered LDs. The LD number and size vary under different growth conditions, perhaps reflecting why ldb16Δ has not been identified in previous genome-wide screens in yeast. Supersized LDs are linked to compromised PtdCho synthesis (Krahmer et al., 2011), and indeed a recent study has identified several yeast SLD mutants defective in enzymes that convert phosphatidylethanolamine to PtdCho (Fei et al., 2011b). We provide several lines of evidence to show that supersized LDs in fld1Δ and ldb16Δ are likely caused by mechanisms distinct from those in the SLD mutants. First, the SLD mutants did not accumulate small clustered LDs, as in fld1Δ and ldb16Δ (Fig. 1C). Second, supersized LDs in the SLD mutants could be converted to wild-type LDs when supplemented with choline and inositol, whereas the supersized LDs in fld1Δ and ldb16Δ cells were converted to small clustered LDs upon supplementation with inositol (Fig. 2). Third, the SLD mutants, unlike fld1Δ and ldb16Δ, accumulated more neutral lipids and exhibited a distorted phospholipid profile when compared with that in the wild type (supplementary material Fig. S1). It has been reported that the phospholipid profile is not greatly changed in fld1Δ – however, INO1 expression is largely induced (Fei et al., 2011b). As INO1 expression is regulated by the INO1 suppressor Opi1 in ER through interaction with Scs2 and phosphatidic acid (Henry et al., 2012; Young et al., 2010), it has been proposed that the distribution of ER-localized phosphatidic acid is altered in fld1Δ cells. Consistent with the finding, our data also revealed that most of the phospholipid biosynthesis genes, including INO1, were upregulated in ldb16Δ cells (supplementary material Fig. S1). Overall, we conclude that fld1Δ and ldb16Δ define a unique subset of LD mutants in yeast.

Both fld1Δ and ldb16Δ are sensitive to terbinafine, which interferes with the squaline expoxidase Erg1 (Fig. 2B). Unlike the dga1Δ lro1Δ are1Δ are2Δ mutant, which exhibits complete loss of neutral lipids, shows terbinafine sensitivity, and a decreased level of Erg1 (Sorger et al., 2004), fld1Δ and ldb16Δ cells exhibit higher levels of Erg1 than the wild type (Fig. 6B–D). The increase in Erg1 levels in the mutants was not due to upregulation of ERG1 mRNA, as determined by microarray and quantitative PCR (unpublished results). Erg1 is the rate-limiting enzyme in sterol homeostasis. Recent evidence has revealed that cellular sterol, most likely lanosterol, regulates the degradation of Erg1 by means of the ubiquitin ligase Ssm4 branch of ERAD (also known as ERAD-C) (Foresti et al., 2013). Although the level of ergosterol in fld1Δ and ldb16Δ appeared normal, sterol ester levels in the two mutants were higher (supplementary material Fig. S1). The greater sensitivity of the two mutants to fluconazole, an inhibitor of lanosterol demethylase (Erg11), implies that sterol intermediate metabolites in fld1Δ and ldb16Δ might be altered (unpublished results). Thus, we suspect that the higher level of Erg1 in fld1Δ and ldb16Δ might reflect an alteration of sterol biosynthetic intermediates, such as reduced lanosterol level, and thus causing Erg1 accumulation by feedback control. Alternatively, Fld1 and Ldb16 deficiency might impair Ssm4 function and thus cause Erg1 accumulation. These possibilities await future investigation.

Similar to Fld1, Ldb16 is an integral membrane protein and localizes to ER–LD contact sites (Fig. 3). By using colocalization, pulldown and yeast two-hybrid analyses, we showed that Fld1 and Ldb16 form a protein complex (Fig. 3C,D) and that their putative transmembrane helices are particularly important for their interaction and functions at the contact sites (Figs 4, 5). Moreover, we demonstrated that prevention of ERAD-C-mediated degradation of Ldb16 requires association of Ldb16 with Fld1 (Fig. 7). Thus, we propose that Fld1 could contribute to Ldb16 folding in the ER membrane, which might further coordinate the targeting of Ldb16 to the ER–LD contact sites. Our functional analyses clearly demonstrated that only the hydrophobic domain of Ldb16 is used for its function in LD size control and for interaction with Fld1 (Fig. 4). The C-terminal cytosolic domain of Ldb16, consisting of about two-thirds of the protein, is not needed for LD size maintenance (Fig. 4). The C-terminal domain might regulate Ldb16 stability as its deletion led to stabilization of the protein (unpublished results). We found that both Ldb16 and Fld1 are ubiquitinylated proteins (Fig. 7F). In mammals, seipin is polyubiquitinylated and several human seipin mutations enhance the level of ubiquitinylation (Ito and Suzuki, 2007). However, the importance of seipin/Fld1 ubiquitinylation remains to be studied.

Because Fld1 and Ldb16 proteins harbor no obvious functional motif, domain or enzymatic activities, they might not play a direct role in phospholipid metabolism. The fld1Δ and ldb16Δ mutants accumulate two distinct pools of LDs. Although the small clustered LDs might be a transition state to supersized LD formation, sorting of LD proteins, including Erg1, Erg6 and Pet10, into the two LD populations might be different (unpublished results). In addition, their relative abundance in the LDs isolated from fld1Δ and ldb16Δ mutants are different when compared with the wild type (Fig. 6B–D). Thus, the Fld1–Ldb16 complex at the ER–LD contact site might provide a structure barrier for sorting of membrane and/or proteins into LDs during LD assembly. Alternatively, the complex might sense the content or physical property of the emerging LDs to control lipid filling into LDs. A similar model has been proposed by J. Goodman, whose recent findings suggest that the purified Fld1 complex is organized into a toroid (Binns et al., 2010); however, the dimension of the toroid might not fit the contact sites. Future EM and photoactivation and/or photobleaching studies might provide further insight into the Fld1–Ldb16 structure at the ER–LD contact sites.

We found that the Fld1–Ldb16 complex is further organized into a large protein complex, although it remains unclear whether the complex contains only Fld1 and Ldb16. Recently, seipin has been shown to interact with the phosphatidic acid phosphatase lipin (Sim et al., 2013). It would be interesting to test whether the lipin ortholog, phosphatidic acid phosphohydrolase 1 (Pah1), in yeast is associated with the Fld1-Ldb16 complex and whether the Fld1-Ldb16 complex regulates Pah1 function. Analyses on purified ER and LD reveal that Fld1 is localized to both ER and LDs, whereas Ldb16 is found only in ER (Fig. 6B,C). Ldb16 might regulate Fld1 partitioning to LDs, because the Fld1 level in LDs was much reduced in ldb16Δ cells (Fig. 6C,D). Self-interaction of Fld1 is a prerequisite for its interaction with Ldb16, and the presence of Ldb16 also seems to strengthen Fld1 self-interaction (Fig. 5). How these interactions are organized into the architecture of the Fld1-Ldb16 complex at the ER–LD contact site is a central question and will be important for further dissecting the molecular functions of the Fld1–Ldb16 complex.

We demonstrated that Fld1 and Ldb16 are mutually dependent for LD functions in yeast. Previous findings have shown that human seipin, but not the lipodystrophy mutant A212P, restores normal LD morphology in fld1Δ (Fei et al., 2008b; Szymanski et al., 2007). Here, we extended the analyses to additional CGL2 mutations and reveal that most of them, indeed, are unable to complement the LD defects in fld1Δ (Fig. 8B). Seipin shares similarity in sequence and domain organization with Fld1, but not with Ldb16. Intriguingly, the sequence homologs of LDB16 can only be found in fungi. We find that seipin can neither interact with Ldb16, nor restore the level of Ldb16 in fld1Δ cells (Fig. 8C,D). Strikingly, seipin can functionally complement the LD defect in fld1Δ, ldb16Δ, and fld1Δ ldb16Δ strains (Fig. 8E), whereas most of the CGL2-linked seipin mutants cannot (Fig. 8F). Given that the lipodystrophy mutant Fld1 G225P, possessing a mutation in the Fld1 lumenal domain, reduced the interaction with Ldb16 by 60% and resulted in supersized LDs (Fig. 8C), it seems plausible that the lipodystrophy mutation is linked to compromised functions of Ldb16 in yeast. Given these data, we propose that Fld1 and Ldb16 in yeast might have evolved into a single molecule, seipin, that alone is sufficient for the function of the Fld1–Ldb16 complex.

We suspect that the architecture of seipin at the ER–LD contact sites might mimic the yeast Fld1–Ldb16 complex and, thus, is capable of coordinating LD assembly and size maintenance. Future structural analyses of the Fld1-Ldb16 complex and seipin will be crucial for understanding how Fld1 and Ldb16 evolved into seipin and how seipin mutations disturb LD homeostasis and cause lipodystrophy.

All yeast strains and plasmids used in this study are listed in supplementary material Table S1. We used either a pRS306-based or a PCR-based transformation method (Longtine et al., 1998) for generating the yeast strains. Yeast cells were grown in yeast extract peptone dextrose (YPD; 2% peptone, 1% yeast extract and 2% glucose) or synthetic complete media (SC; 0.67% yeast nitrogen base, amino acids and 2% glucose) at 30°C unless otherwise mentioned.

Yeast LDs were stained with 0.1 µg/ml BODIPY 494/503 nm (Invitrogen). Cells were imaged by taking 10 optical section images spaced at 0.5 µm captured by the Olympus IX81 fluorescence microscope using the green fluorescent protein (GFP) filter (excitation HQ470/40 nm, emission HQ525/50 nm) with a 100× objective lens with numerical aperture (NA) 1.4 and a CoolSNAP^TM HQ² CCD camera (Photometrics). We used the analysis LS professional software (Olympus) for image acquisition and processing, and Metamorph (Molecular Devices) for LD quantification, as described previously (Wang and Lee, 2012). The dual- or tri-color images were acquired by a DeltaVision system (Applied Precision) with 100× objective lens (NA = 1.4) and a CoolSNAP^TM HQ² charge-coupled device (CCD) camera (Photometrics) controlled by softWorx Suite (Applied Precision). The filters used with the DeltaVision system were GFP (525/50 nm), mCherry (632/60 nm), yellow fluorescent protein (YFP) (559/38 nm) and cyan fluorescent protein (CFP) (470/24 nm).

We used a freeze-substitution method, as described previously, for sample preparation (Wang and Lee, 2012). Ultrathin sections of 70–90 nm were placed on copper grids and stained with uranyl acetate and lead citrate. Images were taken using a Philips CM 100 transmission electron microscope equipped with a Gatan Orius CCD camera.

Antisera against Fld1, Ldb16, Are2, or Lro1 were raised in rabbits using glutathione S-transferase (GST)–Fld1, GST–Ldb16 (121–256), maltose-binding protein (MBP) –Are2 (1–200) and MBP–Lro1 (301–661) fusion proteins as antigens, respectively. Antibodies against Erg1, Erg6, Pet10 and Ubx2 have been described previously (Wang and Lee, 2012). Antibodies against Kar2, CPY, Sec61 and Pgk1 were gifted by the Schekman lab (UC Berkeley, Berkeley, CA), an antibody against protein A was obtained from Jackson ImmunoResearch, an antibody against Myc was obtained from Covance and antibodies against actin or ubiquitin were obtained from Millipore.

Yeast were converted into spheroplasts and lysed osmotically in HKMS buffer (20 mM HEPES, pH 7.4, 150 mM KCl, 250 mM sorbital, 1 mM MgCl₂) containing 1× Complete^TM EDTA-free protease inhibitor cocktail (Roche). Unbroken cells were pelleted by centrifugation of 800 g for 1 min. The total cell lysate (T) was subjected to centrifugation at 13,000 g for 10 min at 4°C, resulting in the soluble (S13) and pellet (P13) fractions. The S13 was used for another centrifugation step at 100,000 g for 30 min at 4°C, resulting in the soluble (S100) and pellet (P100) fractions. For membrane biochemistry assays, the P13 fractions were resuspended in HKMS alone or HKMS containing 1 M KCl, 0.1 M Na₂CO₃, 3 M urea and 1% Triton X-100. After incubation on ice for 10 min, another centrifugation step at 13,000 g for 10 min at 4°C was performed, resulting in the soluble (S) and pellet (P) fractions. All fractions were precipitated immediately with 10% trichloroacetic acid (TCA) on ice, washed with acetone and then resuspended in MURB (50 mM sodium phosphate, 25 mM 2-(N-morpholino)ethanesulfonic acid (MES), pH 7.0, 1% sodium dodecyl sulfate (SDS), 3 M urea and 5% β-mercaptoethanol). For protease protection assay, the P13 pellet was resuspended in lysis buffer with or without proteinase K (100 µg/ml) and 1% Triton X-100. After incubation on ice for 30 min, samples were resuspended in MURB and heated to 55°C for 10 min before being subjected to SDS-PAGE and immunoblot analysis.

Yeast cells were converted to spheroplasts and Dounce homogenized in lysis buffer (20 mM HEPES, pH 7.4, 150 mM NaCl, 1 mM MgCl₂, 1 mM dithiothreitol) containing protease inhibitors. The P13 fraction was solubilized in lysis buffer containing 1% Triton X-100 for 15 min on ice. After centrifugation at 13,000 g for 5 min, the resulting supernatant fraction was layered on top of a gradient consisting of linear 10–35% glycerol in lysis buffer containing 0.5% Triton X-100 and protease inhibitors. The gradient was subjected to centrifugation at 35,000 r.p.m. for 19 h at 4°C in a SW41 rotor. Samples were collected from the top of the gradients into 15 fractions and were resuspended in MURB followed by SDS-PAGE and immunoblot analysis.

Yeast cells were subjected to glass beads lysis in the lysis buffer (20 mM HEPES, pH 7.4, 150 mM NaCl, 1 mM MgCl₂, 1% Triton X-100, 1 mM dithiothreitol) containing protease inhibitors. The lysates were first cleared by centrifugation at 13,000 g for 10 min and the resulting supernatant was incubated with IgG sepharose^TM 6 Fast Flow (GE Healthcare) at 4°C for 4 h. The beads were then washed four times in the lysis buffer containing protease inhibitors. Bound proteins were resuspended in MURB, boiled for 5 min and subjected to SDS-PAGE, followed by immunoblot analysis.

To examine proteins in isolated ER and LDs, 1 liter of yeast cells grown in SC to OD₆₀₀ = 2 was used according to procedures described previously for ER (Wuestehube and Schekman, 1992) and LDs (Leber et al., 1994). The yields of isolated ER and LDs were quantified by measuring OD₆₀₀, and the same amount of each fraction was resuspended in MURB for SDS-PAGE and immunoblot analysis. We loaded 0.0125 OD₆₀₀ of the isolated ER and 0.03 OD₆₀₀ of the isolated LDs and quantified the relative abundance of selected proteins on the same immunoblot using UVP ChemiDoc-It^TM imager and software.

For the cycloheximide chase experiment, cells were grown in YPD to OD₆₀₀ = 0.8 and then treated with 1 mg/ml cycloheximide or an equal amount of dimethyl sulfoxide (DMSO). For the MG132 experiment, MG132 was added to a final concentration of 100 µM, or an equal volume of DMSO was added. At the indicated time-point, 1 ml of cells was precipitated with 10% TCA, washed with acetone and resuspended in MURB (100 µl per OD₆₀₀), followed by SDS-PAGE and immunoblot analysis.

PJ69-4a (James et al., 1996) cells transformed with pGAD- and pGBD-containing plasmids were grown on SC-Leu-Trp dropout plates. At least three independent transformants were tested for their two-hybrid interactions on SC-Leu-Trp-His+1 mM or 5 mM 3-aminotriazole (3-AT) and SC-Leu-Trp-His-Ade dropout plates. The relative strength for the two-hybrid interactions was determined by serial-dilution of cells to compare their growth on the plates. The plates were incubated at 30°C for 3–5 days and photographed by the UVP ChemiDoc-It^TM imager and software.

All experiments were repeated at least three times, and most of the data are presented as mean±s.d. as indicated in the figure legends. In some LD quantification experiments, yeast LD phenotypes were quantified by grouping and we counted n>100 cells for each set of experimental data. We repeated the experiments three times; however, only one result is shown.

We thank Rey-Huei Chen (IMB, Academia Sinica) for sharing lab resources and critical reading of our manuscript, Pauline Yang for editing our manuscript, Stefan Jentsch (Max Planck Institute, Germany) and Randy Schekman (UC Berkeley, CA) for strains and antibodies, Yu-Ching Wu (IPMB, Academia Sinica) for UPLC/MS analyses, Su-Ping Lee (IMB, Academia Sinica) for transmission EM, Shu-Yun Tung for microarray and qPCR (IMB, Academia Sinica), Mei-Jane Fang for imaging (IPMB, Academia Sinica) and Chih-Wei Wang for executing some initial experiments.