The FATP1–DGAT2 complex facilitates lipid droplet expansion at the ER–lipid droplet interface

Lipid droplets (LDs) are key regulatory sites for storage and mobilization of neutral lipids in response to changes in cell growth and energy demand (Farese and Walther, 2009; Goodman, 2009; Mak, 2012). Elevation of cellular fat level can be accommodated by an increase in LD size (Kuerschner et al., 2008). The sequestration of fat in the form of neutral lipids, such as triglycerides into LDs, is critical for cellular defense against lipotoxicity (Listenberger et al., 2003; Unger and Scherer, 2010). Excess free fatty acids that cannot be accommodated in LDs may cause cellular stress and insulin resistance in obese subjects. Although substantial progress has been made with regard to fat mobilization from LDs (Grönke et al., 2005; Zechner et al., 2009), the molecular mechanisms that facilitate fat deposition into LDs are poorly understood. Juxtaposition of ER and LDs has been observed in yeast and mammalian cells (Blanchette-Mackie et al., 1995; Szymanski et al., 2007; Robenek et al., 2009; Skinner et al., 2009), and it is plausible that such association may be key for coordinating lipid synthesis and cellular fat storage. However, it is unclear how ER–LD interaction is maintained in a dynamic cellular environment.

Mammalian FATP1 was originally identified as a putative fatty acid transporter (Schaffer and Lodish, 1994). Nevertheless, it possesses acyl-CoA synthetase activity (Coe et al., 1999; Hall et al., 2003) and belongs to a large family of acyl-CoA synthetases based on sequence homology. Studies in mammalian cells suggest that FATP1 may localize to the cell surface or the ER (Coe et al., 1999; Lewis et al., 2001; Stahl et al., 2002).

LD expansion may require de novo synthesis of neutral lipids, such as triglycerides, which is dependent on two DAG acyltransferases (DGATs), DGAT1 and DGAT2, in mammals (Yen et al., 2008; Harris et al., 2011). These enzymes catalyze the conjugation of a fatty acyl-CoA to DAG. Although DGAT1 and DGAT2 were originally assigned as microsomal enzymes (Cases et al., 1998; Stone et al., 2006), recent evidence suggests that DGAT2 may also localize to the LDs upon lipid loading in mammalian cells (Kuerschner et al., 2008; Stone et al., 2009; McFie et al., 2011). Furthermore, a budding yeast orthologue of DGAT2, Dga1p, is localized to LDs (Sorger and Daum, 2002).

In this paper, we identify the Caenorhabditis elegans orthologues of FATP1 and DGAT2 as genetic regulators of LD expansion. Biochemical, imaging, and functional analysis revealed that FATP1 and DGAT2 form a conserved protein complex at the ER–LD interface to promote LD expansion. Our results suggest one mechanism for triacylglycerol (TAG) synthesis and channeling that links ER and LD through a macromolecular complex.

To identify genes that are required for LD expansion in an unbiased manner, we performed a genetic suppressor screen in a daf-22 mutant background. We have previously shown that animals lacking DAF-22, the terminal thiolase for peroxisomal fatty acid β oxidation in C. elegans, accumulate excess triglycerides in expanded LDs that are >3 µm in diameter (Zhang et al., 2010a). We mutagenized daf-22(ok693) animals with ethyl methanesulfonate and recovered >15 independent mutant alleles that suppressed the LD expansion phenotype. Here, we report the cloning of acs-22 and dgat-2. We isolated four recessive and one semidominant alleles of acs-22 and determined that the hj25 and hj26 alleles are molecular nulls (Fig. S1). The acs-22 gene encodes a very long chain fatty acyl-CoA synthetase, which is orthologous to mammalian FATP1 and FATP4 (Schaffer and Lodish, 1994; Stahl et al., 1999). It has been reported that acs-22 is required for cuticle integrity in C. elegans (Kage-Nakadai et al., 2010). The loss of acs-22 function suppressed LD expansion in daf-22 mutant animals and reduced their TAG content (Fig. 1, D and G; and Fig. 2, B and G, bar 2). In comparison with wild-type animals, the TAG levels of daf-22 and daf-22; acs-22 animals were elevated by 50 and 24%, respectively (Fig. 1 G). ACS-22 is specifically required for LD expansion in daf-22 animals because individual knockdown of 20 other acs (acyl-CoA synthetase) gene family members by RNAi did not block LD expansion in daf-22 animals (Table S1). However, ACS-22 was not required for LD biogenesis or triglyceride synthesis in otherwise wild-type animals (Fig. 1, C and G). Our results suggest that other acyl-CoA synthetases may act redundantly to fulfill these functions because individual knockdown of other acs genes did not detectably impair animal growth and development (Table S1). We detected GFP expression in the intestine, pharynx, excretory cell, vulval muscle, and anal muscle in animals carrying a rescuing bicistronic transgene that directs GFP expression under the control of the acs-22 5′ regulatory region and its open reading frame (Fig. 1 I). Genetic mosaic and tissue-specific rescue experiments indicated that ACS-22 acts cell autonomously in the intestine to facilitate LD expansion (Fig. S2).

We determined that ACS-22 is a functional orthologue of mammalian FATP1 because transgenic expression of mouse FATP1 restores LD expansion in daf-22; acs-22 mutant animals (Fig. S3 E). Using the ¹³C isotope labeling technique (Perez and Van Gilst, 2008), we found that C16:0 fatty acid absorption in wild-type and acs-22(hj26)–null animals differed by 6% (Fig. 1 H). Therefore, ACS-22 is not essential for fatty acid uptake in C. elegans. Our genetic evidence indicates that the acyl-CoA synthetase activity is necessary for ACS-22 to facilitate LD expansion. The semidominant hj36 allele encodes a D496N substitution, which did not affect the expression level of ACS-22 (Fig. S1). The D496 residue is conserved in mammalian FATP1, and it is required for the catalytic activity of acyl-CoA synthetases for its binding to the AMP moiety of fatty acyl-AMP, an intermediate in the conversion of fatty acid to fatty acyl-CoA (Hisanaga et al., 2004). Overexpression of ACS-22 (D496N) interfered with wild-type ACS-22 function and inhibited LD expansion in daf-22 mutant animals (Fig. 2 G, bar 5; and Fig. S3 A). Because FATP1 can form oligomers (Richards et al., 2003), it is conceivable that mutant ACS-22 binds to wild-type ACS-22 to form defective oligomers that fail to yield acyl-CoA precursors for triglyceride synthesis. An unbiased forward genetic screen followed by targeted sequencing of the acs-22 coding region revealed 13 additional conserved residues that are critical for ACS-22 stability or function (Fig. S1, A and C). Notably, mutation at the ATP-binding P loop (G253S) or the highly conserved L motif for intramolecular interaction (D510N) did not affect protein stability (Fig. S1 C) or localization (Fig. S3, G and H) but abolished the ability of mutant ACS-22 to support LD expansion in daf-22(−) animals (Fig. S3, I and J). These results provide further support that the acyl-CoA synthetase activity of ACS-22 is necessary for LD expansion.

We hypothesized that ACS-22/FATP1 might act cooperatively with a triglyceride synthesis enzyme to support LD expansion in daf-22 mutant animals. Sequence analysis using mammalian DGAT2 (Cases et al., 2001), a key enzyme for triglyceride synthesis, identified four putative C. elegans orthologues. To test whether any of these proteins were regulators of LD size, we knocked down W01A11.2, K07B1.4, Y53G8B.2, and F59A1.10 individually by RNAi and found that F59A1.10 was uniquely required for LD expansion in daf-22 mutant animals. Accordingly, we isolated a single missense allele of F59A1.10 from a daf-22 suppressor screen and renamed the gene dgat-2. Similar to ACS-22, loss of dgat-2 function prevented LD expansion in daf-22 mutant animals and reduced their triglyceride content to wild-type levels (Fig. 1, F and G; and Fig. 2, C and G, bar 7). We detected GFP expression in the intestine in animals carrying a bicistronic transgene that directs GFP expression under the control of the dgat-2 5′ regulatory region and its open reading frame (Fig. 1 J). Overexpression of DGAT-2 from this transgene not only restored LD expansion in daf-22; dgat-2 mutant animals, it was sufficient to drive LD expansion in wild-type animals (Fig. 1 J). In contrast, overexpression of ACS-22 failed to promote LD expansion in wild-type animals. Our results suggest that dgat-2 is limiting in wild-type animals, and its expression level is tightly regulated. Transgenic expression of mouse DGAT2 restored LD expansion in daf-22; dgat-2 mutant animals, indicating that DGAT-2 is indeed the functional orthologue of mammalian DGAT2 (Fig. S3 F). We noted that the basal triglyceride level was not significantly affected in dgat-2 mutant animals (Fig. 1 G), suggesting that at least one other DGAT-related enzyme substitutes for DGAT-2 in triglyceride synthesis. Indeed, the C. elegans genome also encodes a putative orthologue of mammalian DGAT1 (H19N07.4; Cases et al., 1998).

We propose that ACS-22 acts as an acyl-CoA synthetase to generate acyl-CoA, which is then conjugated to DAG for triglyceride synthesis by DGAT-2. Our model predicts that ACS-22 and DGAT-2 may form a stable complex to allow the transfer of acyl-CoA from ACS-22 to DGAT-2. We tested this hypothesis by first asking whether ACS-22 and DGAT-2 could physically associate with each other when they were coexpressed in mammalian cells. We found that ACS-22 coimmunoprecipitated with DGAT-2 but not the YFP Venus (Fig. 3 A). Next, we tested whether mouse FATP1 and DGAT2 could also associate with each other. Although DGAT2 could only be expressed at a low level (undetectable in Fig. 3 C [lane 2] as 10% input and weakly detectable in Fig. 3 C [lane 6] after enrichment by immunoprecipitation), we found that FATP1 strongly associated with DGAT2 (Fig. 3 C, lane 6). In contrast, little FATP1 was recovered by coimmunoprecipitation with the highly expressed DGAT1, which served as a control for nonspecific binding to membrane-associated proteins (Fig. 3 C, lane 5). In addition, the YFP Venus failed to associate with FATP1 (Fig. 3 C, lane 7), whereas FATP1 was able to associate with itself as previously reported (Fig. 3 C, lane 8; Richards et al., 2003). Therefore, FATP1–DGAT2 interaction was specific under our experimental conditions. We next asked whether the FATP1–DGAT2 interaction was dependent on the catalytic activity of FATP1. To this end, we generated mutant forms of FATP1 that had no ATPase activity (S250A; Stuhlsatz-Krouper et al., 1998) or carried amino acid substitutions at conserved residues that were mutated in acs-22(hj23) and acs-22(hj36) mutant worms (T596I and D492N, respectively). Coimmunoprecipitation experiments indicated that FATP1 function was not necessary for its association with DGAT2 (Fig. 3 D).

To test whether FATP1 and DGAT2 could interact directly, we translated both proteins in vitro using reticulocyte lysates and subjected them to coimmunoprecipitation assays. FATP1 associated with DGAT2 but not the negative control Venus YFP (Fig. 3 E, lanes 4 and 5). We also generated a mutant DGAT2 that was catalytically dead (H163A; Stone et al., 2006) and noted that the H163A mutation did not appreciably affect its association with FATP1 in vitro (Fig. 3 E, lane 6). To further define the regions in FATP1 and DGAT2 that are necessary for their interaction, we generated a series of deletion mutants. By coimmunoprecipitation of in vitro translated proteins, we determined that the central region of DGAT2 (amino acids 116–228) but not its N terminus is sufficient for interaction with FATP1 (Fig. 3 F, compare lane 8 with lane 6). We also found that FATP1 (amino acids 191–313) was sufficient for direct interaction with DGAT2 (Fig. 3 G, lane 10). An overlapping region of FATP1 (amino acids 101–257) served as the negative control (Fig. 3 G, land 9). Our results indicate that FATP1 and DGAT2 form a protein complex in vitro and in vivo.

In addition to physical interaction, genetic evidence lends further support that acs-22 and dgat-2 act in the same pathway. Loss of either acs-22 or dgat-2 function suppressed LD expansion in daf-22(−) animals, and loss of dgat-2 function in daf-22; acs-22 mutant animals did not alter the degree of suppression or cause synthetic phenotypes (Fig. 2, D and G, bar 10). Furthermore, overexpression of ACS-22 in daf-22 mutant animals strongly promoted LD expansion that was strictly dependent on DGAT-2 function (Fig. 2, E–G, bars 4 and 9). Collectively, our results suggest that FATP1 and DGAT2 are physically and functionally coupled as part of a triglyceride synthesis complex that facilitates LD expansion.

Mammalian FATP1 was reported to localize to the ER or cell surface (Coe et al., 1999; Lewis et al., 2001; Stahl et al., 2002), whereas DGAT2 was shown to be an ER protein that could relocate to the proximity of LDs upon lipid loading in fixed mammalian cells (Kuerschner et al., 2008; Stone et al., 2009; McFie et al., 2011). No live-imaging experiments have been reported to date. To determine the subcellular localization and dynamics of ACS-22/FATP1 and DGAT-2 in live animals, we expressed low levels of functional ACS-22::GFP, GFP::DGAT-2, and mRuby::DGAT-2 fusion proteins from rescuing single-copy transgenes (Fig. 2 G, bars 3 and 8; and Fig. S3, C and D). We also generated transgenic animals that expressed an ER resident signal peptidase fused to GFP or mKO2 (GFP::SP12 or mKO2::SP12; Rolls et al., 2002). In wild-type animals, mKO2::SP12 and GFP::SP12 illuminate the nuclear envelope and a reticular tubular network that extends to the cell cortex of intestinal cells (Fig. 4 B and Fig. S4 B), which closely resembles the ER architecture observed in yeast and mammalian cells (Voeltz et al., 2002). We found that the subcellular localization of ACS-22::GFP fusion protein was indistinguishable from mKO2::SP12 (Fig. 4, A–C). In FRAP experiments, the lifetime of ACS-22::GFP and GFP::SP12 was comparable (Fig. 5 H and Fig. S4), strongly suggesting that ACS-22::GFP is a resident ER protein in C. elegans.

In contrast to ACS-22::GFP, the GFP::DGAT-2 fusion protein was localized to spherical structures <2 µm in diameter in wild-type intestinal cells (Fig. 4 D). Such localization was independent of the fluorescent protein fusion partner because mRuby::DGAT-2 also localized to similar structures (Fig. 5 B). These structures may represent LDs or lysosome-related organelles (LROs) that contain autofluorescent material (Schroeder et al., 2007; Zhang et al., 2010b). We performed live imaging in λ mode followed by linear unmixing, which clearly indicated that GFP::DGAT-2 was absent from the surface of autofluorescent LROs (Fig. 4, D–G). In FRAP experiments, GFP::DGAT-2 failed to recover within 200 s after photobleaching, in contrast to the dynamic movement of ACS-22::GFP within the ER in wild-type animals (Fig. S4). Therefore, we conclude that GFP::DGAT-2 is a LD resident protein. Our results also imply that the LD targeting of GFP::DGAT-2 cannot be achieved by rapid lateral diffusion from membrane structures, such as the ER, suggesting that the ER and LDs are unlikely to be contiguous in C. elegans intestinal cells.

In daf-22 mutant animals, LDs undergo continual expansion throughout development (Zhang et al., 2010a). In these animals, ACS-22::GFP was observed in reticular structures and appeared to envelop expanded LDs (Fig. 5 D). This is consistent with ultrastructural evidence that the ER closely associates with LDs, especially under lipid loading conditions (Blanchette-Mackie et al., 1995; Robenek et al., 2009). Several lines of evidence indicate preferential enrichment of ACS-22::GFP at the ER–LD interface during LD expansion in daf-22(−) animals. First, ACS-22::GFP showed significantly higher degree of colocalization with mRuby::DGAT-2 than another ER resident protein, GFP::SP12 (Fig. 5, F and G). Second, ACS-22::GFP at the ER–LD interface recovered significantly less than GFP::SP12 after photobleaching in FRAP experiments (P < 0.0001 at 33 s after bleach, t test; Fig. 6, A–G). This observation indicates that ACS-22::GFP is less mobile than GFP::SP12 and suggests that ACS-22 is engaged in a stable protein complex. Third, we found ACS-22::GFP and GFP::DGAT-2 fusion proteins in discrete foci at or near the LD surface in daf-22 mutant animals by immunoelectron microscopy (Fig. 6, I and J). Collectively, our results are consistent with a model that ACS-22/FATP1 and DGAT-2 form a complex at the ER–LD interface to support LD expansion.

Although mammalian DGAT2 had been found on LD surface in cultured cells under lipid loading conditions (Kuerschner et al., 2008; Stone et al., 2009), it was unclear whether this was necessary for DGAT2 to support LD expansion. We directly tested this hypothesis by appending the ER-anchoring region of mouse cytochrome b5 (residues 100–134; Honsho et al., 1998) to the C terminus of C. elegans DGAT-2 and asked whether the fusion protein could substitute for wild-type DGAT-2 function. The cb5-anchoring region had been shown to retain heterologous proteins to the ER (Zhu et al., 1996), and the GFP::DGAT-2::cb5 fusion protein expressed from a single-copy transgene was indeed localized to the ER (Fig. 7 A). More importantly, expression of GFP::DGAT-2::cb5 failed to restore LD expansion in daf-22; dgat-2 mutant animals (Fig. 7, B and E, bar 3). To demonstrate that the cb5 anchor did not interfere with DGAT-2 function per se, we introduced four charged residues in the hydrophobic transmembrane domain of cb5 (Fig. 7 C). Disruption of ER anchoring in GFP::DGAT-2::cb5_mut fusion protein partially restored LD localization and its ability to support LD expansion (Fig. 7, D and E, bar 4). To measure the catalytic activity of various DGAT2 mutant proteins, we expressed them in a yeast strain (Δ4) that could not synthesize TAG (Sandager et al., 2002). The DGAT activity of cell extracts from yeast expressing mRuby::DGAT2, mRuby::DGAT2::cb5, or mRuby::DGAT2::cb5_mut proteins was comparable (Fig. 7 F, lanes 3–5). This result confirmed that the cb5-anchoring region did not interfere with the catalytic activity of DGAT2. Collectively, our results suggest that LD localization of DGAT-2, which allows direct deposition of its TAG product into LDs, is critical for LD expansion.

Because ACS-22/FATP1 and DGAT-2 are highly conserved in mammals, our model predicts that FATP1 and DGAT2 should cooperate to promote LD expansion in mammalian cells. We transiently expressed FATP1::Venus, Venus::DGAT2, or mRuby::DGAT2 fusion proteins in COS7 cells and monitored their localization after oleic acid loading. We found FATP1::Venus at the ER (Fig. 8 A) and Venus::DGAT2 and mRuby::DGAT2 at LDs (Fig. 8 B and Fig. S5 B) consistent with previous studies (Coe et al., 1999; Kuerschner et al., 2008; Stone et al., 2009; McFie et al., 2011). Furthermore, mRuby::DGAT2 colocalized with Perilipin2::GFP on the LD surface (Fig. S5). Coexpression of FATP1::Venus and mRuby::DGAT2 under the same conditions strongly promoted LD expansion (Fig. 8, C and F). In addition to ER localization, FATP1::Venus was also enriched around LDs that were marked by mRuby::DGAT2 (Fig. 8 C). The functional synergy between FATP1 and DGAT2 in promoting LD expansion was dependent on their respective catalytic activities (Fig. 8 F). Interestingly, the catalytically dead mRuby::DGAT2 (H163A) mutant protein failed to localize to the LD surface (Fig. 8 E). Extensive colocalization of DGAT2 (H163A) with FATP1 in the ER did not significantly affect their affinity to each other (Fig. 3 E). Therefore, preferential association of wild-type FATP1 and DGAT2 is highly specific (Fig. 3, C and E) and may indeed stabilize the docking of LDs to the ER.

We used a split GFP approach to further probe the spatial relationship between FATP1 and DGAT2 in mammalian cells. The GFP1–10 and GFP11 fragments can associate to reconstitute full-length GFP when they come into contact with each other, which can be facilitated by their fusion partners (Cabantous et al., 2005). In oleic acid–loaded COS7 cells, no signals were detected when GFP11 was coexpressed with GFP1–10 or GFP1–10::DGAT2 (Fig. 9, A and B). Coexpression of FATP1::GFP11 with GFP1–10 yielded signals in the ER (Fig. 9 C). In contrast, reconstituted GFP gave strong fluorescence signals around LDs when FATP1::GFP11 and GFP1–10::DGAT2 were coexpressed (Fig. 9 D). Our results indicate that FATP1 and DGAT2 are in close proximity of each other at the ER–LD interface. We conclude that FATP1 and DGAT2 act in concert to promote LD expansion at the ER–LD interface in an evolutionarily conserved manner.

Close association of LDs with the ER has been observed (Blanchette-Mackie et al., 1995; Robenek et al., 2009), but the molecular mechanisms that enable their physical and functional coupling are not known. In this paper, we combine genetic, biochemical, and imaging approaches to demonstrate that a conserved FATP1–DGAT2 complex acts at the ER–LD interface to promote TAG synthesis and LD expansion.

FATP1 belongs to a large family of acyl-CoA synthetases that conjugates CoA to free fatty acid of various chain lengths, and it has been proposed that functional specificity of acyl-CoA synthetases may derive from their subcellular localization and interacting downstream partners (Ellis et al., 2010). Here, we provide a clear example of how this is achieved for FATP1 by identifying DGAT2 as its physical and functional partner in facilitating LD expansion through TAG synthesis. The physical interaction between FATP1 and DGAT2 and the distinct subcellular localization of their C. elegans orthologues prompt us to propose a model in which FATP1 and DGAT2 form a triglyceride synthesis protein complex that links LDs to the ER (Fig. 9 E). This complex fulfills two important functions: (1) coupling of acyl-CoA synthesis in the ER to TAG synthesis on the LD surface; (2) deposition of newly synthesized TAG into the LD core. We further demonstrated that functional coupling of FATP1 and DGAT2 is evolutionarily conserved in mammalian cells.

The FATP1–DGAT2 complex defines a molecular link between the ER and LDs for the first time. The ER appears to use distinct macromolecular complexes to physically interact with other intracellular organelles. For example, the Mitofusin complex and the ER–mitochondria encounter structure complex have been shown to tether the ER to mitochondria (de Brito and Scorrano, 2008; Kornmann et al., 2009). Loss of Mitofusin 2 or components of the ER–mitochondria encounter structure complex disrupts the ER and/or mitochondrial morphology (de Brito and Scorrano, 2008; Kornmann et al., 2009). However, the ER morphology in mutant worms that lack ACS-22/FATP1 or DGAT-2 is normal, which suggests that the FATP1–DGAT2 complex is not required to maintain the structural integrity of the ER.

Although DGAT2 was originally assigned as a microsomal (ER) enzyme through biochemical fractionation experiments (Cases et al., 1998; Stone et al., 2006), it can also localize to mitochondrial-associated membranes and LDs upon lipid loading in mammalian cells (Kuerschner et al., 2008; Stone et al., 2009; McFie et al., 2011). Association with phospholipid bilayers by mouse DGAT2 has been attributed to its two predicted transmembrane segments, whereas discrete elements at its C terminus appeared to be necessary but not sufficient for LD association (McFie et al., 2011). The budding yeast orthologue of DGAT2, Dga1p, is strongly enriched in LDs in biochemical experiments (Sorger and Daum, 2002). It is therefore intriguing that mammalian DGAT2 has not been found broadly in LD proteomes. Perhaps the purification procedures were not amenable to preserving DGAT2 association with LDs. This may be caused in part by the interaction of DGAT2 with multiple ER resident proteins (see next paragraph). Nevertheless, our imaging data on DGAT2 in C. elegans and mammalian cells strongly suggest that DGAT2 can stably reside on the LD surface in vivo (Fig. 4 D, Fig. 8 B, and Fig. S5). More importantly, ER-tethered DGAT2 cannot substitute for LD-associated DGAT2 in promoting LD expansion (Fig. 7). Our results clearly demonstrate the physiological significance of LD-associated DGAT2.

A multisubunit protein complex for TAG synthesis had been isolated from rat intestinal villus cells, which contained acyl-CoA synthetase, acyl-CoA acyltransferase, monoacylglycerol acyltransferase, and DGAT activities (Manganaro and Kuksis, 1985; Lehner and Kuksis, 1995). It is plausible that FATP1 is responsible for the acyl-CoA synthetase activity in this complex. Additional proteins that are required for synthesis and modification of fatty acid and other lipid precursors of TAG may also be present in the FATP1–DGAT2 complex. For example, the steroyl-CoA desaturase SCD1, which catalyzes the desaturation of saturated fatty acids, was reported to associate with DGAT2 (Man et al., 2006).

TAG synthesis involves the conjugation of fatty acyl-CoA to DAG. DAG can be synthesized in the ER by phosphatidic acid phosphatases, such as Pah1p in budding yeast and Lipin-1 in metazoans (Reue and Brindley, 2008; Karanasios et al., 2010; Han et al., 2012). DAG has also been detected in purified LDs (Kuerschner et al., 2008), possibly the result of lipases such as adipose triglyceride lipase, and it is conceivable that DGAT2 on the LD surface may have direct access to such a pool of DAG. Furthermore, specific isoforms of Lipin-1 have been found at LDs (Valdearcos et al., 2011; Wang et al., 2011). Our model does not address the spatial distribution of the immediate precursors of DAG that are synthesized by enzymes in the Kennedy pathway or the monoacylglycerol pathway, which are classically regarded as microsomal (Coleman and Lee, 2004; Yen et al., 2008). It is plausible that dedicated proteins are responsible for transporting DAG and its precursors to the ER–LD interface. Alternatively, such precursors may be channeled through membrane connections between the ER and LDs.

The TAG content of LDs can be modulated by de novo TAG synthesis and reesterification of free fatty acids from lipolysis at the LD surface (Brooks et al., 1982). Although we favor a role for the FATP1–DGAT2 complex in channeling acyl-CoA from the ER and synthesizing TAG at the ER–LD interface, we cannot rule out its involvement in recycling hydrolyzed free fatty acids. It will be interesting to determine whether the assembly or activity of the FATP1–DGAT2 complex is responsive to lipolysis in future studies.

It has been proposed that LDs emerge from the ER through accumulation of neutral lipids between the phospholipid bilayer of the ER and subsequent pinching of the bulged ER with the neutral lipid core (Martin and Parton, 2006; Ploegh, 2007). Our study adds insight into how LDs expand and grow. We present evidence that preexisting LDs can expand by TAG synthesis and deposition through reengagement with the ER via the FATP1–DGAT2 complex in C. elegans and mammalian cells. In budding yeast, LDs appear to remain connected with the ER (Jacquier et al., 2011). Therefore, TAG synthesized in the ER can be partitioned to the LD core without invoking specific protein transporters. It is probable that LD expansion is facilitated by multiple parallel mechanisms. The relative dominance of each mechanism may be dictated by substrate availability and cellular context.

The wild-type strain was Bristol N2. All animals were raised at 20°C. The following alleles and transgenes were used: LGII, daf-22(ok693); LGV, dgat-2(hj44); and LGX, acs-22(hj23), acs-22(hj24), acs-22(hj25), acs-22(hj26), acs-22(hj36), hjIs14[vha-6p::GFP::C34B2.10(SP12)::unc-54 3′UTR] (generated by microparticle bombardment and outcrossed four times with N2), hjSi29[vha-6p::acs-22 cDNA::GFP-TEV-3×FLAG::let-858 3′ UTR] IV, hjSi56[vha-6p::3×FLAG-TEV-GFP::dgat-2::let-858 3′UTR] IV, hjSi61[vha-6p::mKO2:: C34B2.10(SP12)::let-858 3′UTR] II, hjSi62[vha-6p::3×FLAG-TEV-mRuby::dgat-2::let-858 3′UTR] II, hjSi76[vha-6p::3×FLAG-TEV-GFP::dgat-2::cb5 C terminus::let-858 3′UTR] IV, and hjSi87[vha-6p::3×FLAG-TEV-GFP::dgat-2::cb5 C terminus_mut::let-858 3′UTR] IV.

Single-copy transgenes were generated as previously described (Frøkjaer-Jensen et al., 2008) using the direct injection protocol. The targeting plasmid (50 ng/µl) was injected with 50 ng/µl pJL43.1 (Pglh-2::MosTase::glh-2 3′ UTR) and coinjection markers into EG4322 or EG5003 for transgene insertion into LGII or LGIV, respectively. Transgenic strains were outcrossed at least two times with N2.

The five acs-22 alleles were isolated in a genetic screen for suppressors of the expanded LD phenotype of daf-22(ok693) mutants. Using a single nucleotide polymorphism–based mapping strategy with the Hawaiian C. elegans isolate CB4856 (Davis et al., 2005), we mapped hj23 to LGX between haw105601 and snp_Y81B9A[4]. Based on the annotation of genes in the 800-kb region, we chose to sequence all exons and exon/intron junction of acs-22 and identified a single C to T mutation that introduced a T603I substitution in animals carrying the hj23 allele. Sequencing of four other alleles in the same complementation group, hj24, hj25, hj26, and hj36, revealed additional mutations in acs-22. The daf-22(ok693); acs-22(hj26) mutant could be rescued by an acs-22p::acs-22::SL2-gfp transgene.

From a genetic screen for suppressors of the expanded LD phenotype of daf-22(ok693); hjSi29[acs-22::gfp]; acs-22(hj26) mutants, we isolated mutant alleles in the acs-22 coding region of the single-copy transgene as well as others that did not map to the transgene. We had previously determined that F59A1.10 RNAi could suppress LD expansion in daf-22(ok693) animals. Therefore, we sought alleles in F59A1.10 in the mutant allele collection by directly sequencing the exons and introns of F59A1.10 and recovered hj44. We renamed F59A1.10 as dgat-2 for its sequence and functional homology to mammalian DGAT2. The hj44 allele encodes a G296E substitution in C. elegans DGAT-2. The daf-22(ok693); dgat-2(hj44) mutant could be rescued by a dgat-2p::dgat-2::SL2-gfp transgene.

Lipid analyses involving total lipid extraction, TLC, and gas chromatography–mass spectrometry were performed to compare relative changes of major lipid species, quantify TAG levels, and profile fatty acid compositions. Procedures were essentially the same as previously described (Zhang et al., 2010a). For TAG quantification, the number of L4 animals for each replicate sample was scaled down from 24,000 to 12,000. Quantitation of fatty acid uptake using ¹³C isotope labeling was performed as previously described using the mass spectroscopy–selective ion monitoring mode (Perez and Van Gilst, 2008). Escherichia coli OP50 was grown in uracil-supplemented Spectra 9-U medium or Spectra 9-C medium (¹³C, 98%) obtained from Cambridge Isotope Laboratories. Lipid was extracted from 30,000 synchronized L4 larval animals for each sample.

Postfix Oil red O staining was performed essentially as previously described (O’Rourke et al., 2009). Animals were fixed in 1% formaldehyde/PBS for 10 min at 22°C in 15-ml glass conical tubes. Samples were frozen on dry ice and thawed for three cycles. After extensive washing with PBS, animals were stained with Oil red O in 60% isopropanol for 30 min at 22°C and washed and destained for 15 min in PBS at the same temperature. Images were acquired in bright field with the same settings on a microscope (Axiovert; Carl Zeiss) equipped with a 40×, NA 0.6 air objective (LD Achroplan; Carl Zeiss) and a camera (AxioCam HR; Carl Zeiss) controlled by the AxioVision software (Carl Zeiss).

Confocal imaging and size measurement of red BODIPY-stained LD on late-stage L4 animals were conducted essentially the same as previously described (Zhang et al., 2010a). Images were acquired on a confocal microscope (LSM 510; Carl Zeiss) using a 40×, NA 1.2 water C-Apochromat objective. Optical sections were taken at 0.45-µm intervals, and z stacks of 9.0 µm were exported to Imaris 6.2.1 (Bitplane) for identification, annotation, and measurement of diameter of all BODIPY-labeled structures. Spectrally unmixed images were first acquired on a confocal microscope (LSM 710; Carl Zeiss) in λ mode using a 40×, NA 1.2 water C-Apochromat objective and GFP and autofluorescence unmixed using the ZEN software (Carl Zeiss).

Images were taken on larval stage L4 worms using a confocal microscope (LSM 710) with a 40×, NA 1.2 water C-Apochromat objective. For each animal, GFP and mRuby fluorescence images were collected in multitrack mode to prevent leakage. A Pearson’s coefficient was calculated for each image based on pixels of interest (Manders et al., 1992). A threshold in the mRuby channel of the upper 5% of all pixels was sufficient for masking only relevant pixels across worms. To prevent underestimation of the colocalization coefficients, images with pronounced autofluorescent signals appearing as filled spherical structures were excluded (see Fig. 4 for a comparison of autofluorescent structures and LDs).

Photobleaching experiments were performed on larval stage L4 animals on a confocal microscope (LSM 710) using a 40×, NA 1.2 water C-Apochromat objective and controlled by the ZEN software. For each animal, one or more areas of 2 µm in diameter were selected. For Fig. 5 H, a time series of the entire field (2,500–3,600 µm²) was taken, and after the fifth frame, 10 cycles of photobleaching was performed using a 488-nm laser at maximum power. The time series continued until structures returned visibly with respect to neighboring unbleached regions or after 200 s of imaging. The fluorescence recovery of the bleached region was fit to an exponential after adjusting for the slow decay of fluorescence caused by imaging using areas of the image distant from the bleached region. At least eight photobleaching events were recorded for each strain. For Fig. 6 (A–F), a time series of the entire field (∼4,900 µm²) was taken, and 200 cycles of photobleaching was performed after the fifth frame using a 488-nm laser at maximum power. The time series was collected for another 30 s. Photobleaching curves that showed 65% or more absolute loss of fluorescence postbleach were normalized and collected into Fig. 6 G.

For gold immunolabeling of GFP, dissected C. elegans intestines were fixed with 2% formaldehyde and permeabilized with PBS + 0.1% Triton X-100. Samples were blocked in PBS + 1% BSA and incubated with anti-GFP antibody followed by goat anti–mouse super small gold (Electron Microscopy Sciences). Samples were then fixed with 2.5% glutaraldehyde in PBS (10 min) and washed with distilled water, and gold staining signal was enhanced with a silver enhancement kit (Aurion). After washing with distilled water, samples were postfixed in 1% OsO₄ containing 1.5% potassium ferrocyanide in PBS for 30 min at room temperature and washed with distilled water. Samples were dehydrated in ethanol, embedded in epoxy resin (room temperature for 2 d), and polymerized (60°C for 2 d). Sections (50–70 nm thick) were cut and transferred to Formvar-coated grids and stained with aqueous uranyl acetate and lead citrate. Sections were analyzed on an electron microscope (80 kV; Tecnai Spirit; FEI).

Human epithelial HeLa cells and human embryonic kidney 293 cells were cultured in 6-well plates and transfected with 4 µg per well of expression plasmids that were based on pcDNA5 (Invitrogen) using Lipofectamine 2000 (Invitrogen). Cells were harvested 24 h after transfection, and whole-cell lysates were prepared using MAPK buffer (10 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM EDTA, 1% NP-40, and protease inhibitors [Complete; Roche]). Immunoprecipitations with agarose beads conjugated with anti-FLAG were performed at 4°C for 2.5 h. In vitro translation was performed using a coupled reticulocyte lysate system (TnT; Promega). The lysates containing in vitro translated proteins were diluted by 1 vol MAPK buffer before immunoprecipitation. Protein complexes were eluted by boiling in 3× SDS loading buffer (mammalian cell lysates coimmunoprecipitation) or with 0.2 mg/ml 3× FLAG peptide at 4°C for 1 h (in vitro translated proteins coimmunoprecipitation) before SDS-PAGE.

DGAT activity was measured as described previously (Liu et al., 2010) with minor modifications. The assay was conducted in a buffer containing 150 mM Hepes-KOH, pH 7.0, 2.5 mM MgCl₂, 0.1 mg/ml BSA, 308 µM dioleoyl glycerol, and 15 µM [¹⁴C]oleoyl-CoA (18 µCi/µmol) with 100 µg microsomal protein in a final volume of 200 µl. This assay mixture was incubated at 30°C for 10 min and quenched by addition of 4 ml chloroform/methanol (2:1). Nonlipid molecules were extracted by addition of 0.75 ml H₂O with thorough mixing. The aqueous phase was removed, and the organic phase was dried down under nitrogen. Lipids were reconstituted in 30 µl chloroform and resolved by TLC using a hexane/ether/acetic acid (80:20:1) mobile phase. The plate was analyzed using a phosphor storage screen, and the triglyceride band intensity was quantitated using ImageQuant Array software (GE Healthcare). Data presented are the mean of two independent microsome preps performed in duplicate and normalized to the Δ4 strain expressing the empty vector.

Wild-type [MATa ADE2] and Δ4 [MATα are1-Δ::HIS3 are2-Δ::LEU2 dga1-Δ::KanMX4 lro1-Δ::TRP1 ADE2] strains are previously described (Sandager et al., 2002). cDNAs encoding mRuby fluorescent protein and murine DGAT2 were cloned in frame into p426-GPD mRuby. CB5 and CB5 mutant tags were added to the C terminus of DGAT2 with no intervening sequence.

For Western blotting, monoclonal antibodies against the FLAG epitope (clone M2; Sigma-Aldrich; 1:1,000) and HA epitope (clone 3F10; Roche; 1:1,000), rabbit polyclonal antibodies against ACS-22 (amino acids 216–231; 1:300; YZ677; YenZym) and DGAT2 (NBP1-71700; Novus Biologicals; 1:500) were used. For immunoprecipitation, anti-FLAG M2 affinity gel (Sigma-Aldrich) was used.

COS7 cells were transfected with pcDNA3.1 (Invitrogen)-based expression plasmids using Lipofectamine (Invitrogen) in poly-d-lysine 4-well glass slides (CultureSlides; BD). Oleic acid (Nu-Chek Prep) conjugated with BSA was added to cells 8 h after transfection at a final concentration of 400 µM. Cells were incubated for 16 h before fixation with 4% formaldehyde (22°C for 15 min). Fixed cells were washed three times with PBS, once with 0.1% Triton X-100 in PBS (22°C for 2 min), and then two times with PBS. LDs were visualized by adding 100 ng/ml Nile red in PBS to the cells (22°C for 5 min). Cells were washed once with PBS before mounting with VECTASHIELD with DAPI (Vector Laboratories). Images were acquired on a confocal microscope (LSM 710) in channel mode using a 40×, NA 1.2 water C-Apochromat objective and 3 or 8× zoom and controlled by the ZEN software. GFP was excited with a 488-nm laser (4%), and mRuby and Nile red were excited with a 561-nm laser (5%).

Fig. S1 shows molecular lesions in the acs-22 gene. Fig. S2 shows that ACS-22 acts cell autonomously to promote LD expansion. Fig. S3 shows LD expansion in mutant animals carrying acs-22 or dgat-2 transgenes. Fig. S4 shows ACS-22 and DGAT-2 are ER and LD resident proteins, respectively, by FRAP analysis. Fig. S5 shows colocalization of DGAT2 and Perilipin2. Table S1 shows results of knockdown of acyl-CoA synthetases by RNAi.

We thank Yan Hao for generating hjSi62, Marc Van Gilst and Carissa Perez Olsen for advice on ¹³C isotope labeling experiments, Perry Bickel for protocols, and Robb Krumlauf for critical reading of the manuscript.

This work was supported by the Stowers Institute for Medical Research (to H.Y. Mak) and National Institutes of Health grant RO1 2R56DK56084 (to R.V. Farese Jr.).

Author contributions: N. Xu was responsible for Fig. 1 (I and J), Fig. 2 (D and G), Fig. 3, Fig. 4, Fig. 5 (A–F), Fig. 6 (A–F and H–J), Fig. 7 (A–E), Fig. 8, Fig. 9, Fig. S1, Fig. S2, Fig. S3 (C–J), Fig. S5, and Table S1. S.O. Zhang was responsible for Fig. 1 G, Fig. 2 (A–C and E–G), and Fig. S3 (A and B). R.A. Cole was responsible for Fig. 1 (A–F and H) and Table S1. S.A. McKinney was responsible for Fig. 5 (G and H), Fig. 6 G, and Fig. S4. F. Guo was responsible for Fig. 6 (H–J). J.T. Haas and R.V. Farese Jr. were responsible for Fig. 7 F. H.Y. Mak, N. Xu, S.O. Zhang, and R.A. Cole conducted genetic screens. N. Xu and R.A. Cole performed genetic mapping and molecular cloning of mutants. S. Bobba contributed analysis tools for Fig. 1 H. All authors participated in data analysis. H.Y. Mak conceived the project and wrote the manuscript.