The N-terminal region of acyl-CoA synthetase 3 is essential for both the localization on lipid droplets and the function in fatty acid uptake.

Cytosolic lipid droplets (LDs) are storage organelles for neutral lipids derived from endogenous metabolism. Acyl-CoA synthetase family proteins are essential enzymes in this biosynthetic pathway, contributing activated fatty acids. Fluorescence microscopy showed that ACSL3 is localized to the endoplasmic reticulum (ER) and LDs, with the distribution dependent on the cell type and the supply of fatty acids. The N-terminus of ACSL3 was necessary and sufficient for targeting reporter proteins correctly, as demonstrated by subcellular fractionation and confocal microscopy. The N-terminal region of ACSL3 was also found to be functionally required for the enzyme activity. Selective permeabilization and in silico analysis suggest that ACSL3 assumes a hairpin membrane topology, with the N-terminal hydrophobic amino acids forming an amphipathic helix restricted to the cytosolic leaflet of the ER membrane. ACSL3 was effectively translocated from the ER to nascent LDs when neutral lipid synthesis was stimulated by the external addition of fatty acids. Cellular fatty acid uptake was increased by overexpression and reduced by RNA interference of ACSL3. In conclusion, the structural organization of ACSL3 allows the fast and efficient movement from the ER to emerging LDs. ACSL3 not only esterifies fatty acids with CoA but is also involved in the cellular uptake of fatty acids, presumably indirectly by metabolic trapping. The unique localization of the acyl-CoA synthetase ACSL3 on LDs suggests a function in the local synthesis of lipids.

cloned but did express only very poorly. A3 Nt -GFP contains the amino acids 1-135 of human ACSL3 followed by GFP. The cDNA of Caco-2 cells was used as a template for PCR with primers s-H3-ACSL3 (5 ′ -GAATTCAAGCTTACCATGAATAACCACGTGTC-TTC-3 ′ ) and a-ACSL3GFP (5 ′ -ACGTACCGGTGGATCCGCAAGC CAATTATACTGTCC-3 ′ ). The PCR product was digested with Hin -dIII and Age I and ligated into pEGFP-N1 (Clontech). A3 Nt -RFP is comprised of amino acids 1-135 of human ACSL3 followed by mRFP. For this, the plasmid A3 Nt -GFP was digested with Hin dIII and BamHI and ligated into Murr1-RFP.pcDNA3 ( 25 ); this replaced the Murr1 cDNA and yielded a cDNA containing monomeric RFP in frame ( 26 ). Transcription of the ACSL3 RNAi plasmid generates a small hairpin RNA. The target sequence of ACSL3 comprised nucleotides 1540-1558 of the human cDNA. Oligos sA3J1 (5 ′ -GATCTCCGGTGGATACTTTAATACTGTTCAAGA-GACAGTATTAAAGTATCCACCTTTTTTGGAAC-3 ′ ) and aA3J1 (5 ′ -TCGAGTTCCAAAAAAGGTGGATACTTTAATACTGTCTCT TGAACAGTATTAAAGTATCCACCGGA-3 ′ ) were annealed and ligated into pSuper digested with Bgl II and Xho I ( 27 ). Subcloning was into the retroviral plasmid pRVH1-puro ( 28 ). The cDNA of RFP-FATP4 is an N-terminal mRFP followed by full-length mouse FATP4. Plasmid FATP4.pcDNA3 ( 29 ) was digested with Apa I and Hin dIII and ligated into RFP-ER containing N-terminal mRFP, thereby replacing the cDNA coding for Sec61 ␤ . The dominantnegative cav1 DGI -GFP has a deletion of amino acids 1-81 of canine caveolin-1. This construct was modeled after cav3 DGV and has an intact scaffolding domain ( 30 ). PCR of canine caveolin1-GFP ( 31 ) with primers s-DGI (5 ′ -ACGTAGATCTCGAGACCATG GATGGCATCTGGAAGGCC-3 ′ ) and a-CV-GFP (5 ′ -CCATGG TGG-C GACCGG-3 ′ ) was followed by digestion with Xho I and Age I and reinsertion into the caveolin-GFP plasmid. All cDNAs derived from PCR products were fully sequenced. For transient expression, cells grown to 80% confl uency in a 6-well plate (10 cm 2 /well) were incubated for 4 h with the transfection mix consisting of 2.0 µg plasmid, 10 µl FUGENE HD (Roche, Mannheim, Germany), and 100 µl Opti-MEM (Invitrogen) in 2 ml of standard medium. After 4 h, cells were washed and incubated for a further 20 h in 2 ml of standard medium without antibiotics before analysis. Control cells for biochemical experiments were transfected with pcDNA3.
Stable depletion of ACSL3 by RNAi was achieved by retroviral integration of the ACSL3 RNAi plasmid into A431 cells. The generation of replication-defi cient retrovirus pseudotyped with VSV-G from phoenix-gp cells, transduction, and antibiotic selection (2 days, 4.0 µg/ml puromycin for A431 cells) were as described earlier ( 28 ). Control cells were transduced with the plasmid pRVH1-puro containing no shRNA sequence.
␤ -oxidation (13) ] and fatty acid uptake driven by vectorial acylation at the plasma membrane ( 14 ). Furthermore, the ACSVL proteins have also been put forward as fatty acid transporters (FATPs) at the cell surface ( 15 ).
ACSL3 was cloned initially from rat brain ( 16 ) but is ubiquitously expressed ( 17 ) and has a substrate preference for polyunsaturated fatty acids ( 16,18 ). ACSL3 was found to be an abundant protein on lipid droplets prepared from Huh7 human hepatoma cells ( 19 ). Interestingly, the distribution of ACSL3 between LDs and heavier cell membrane fractions correlated with the amount of intracellular lipid droplets ( 20 ).
Here, we present evidence that the N-terminus of ACSL3 is necessary and suffi cient for the localization to lipid droplets. Our data suggest that N-terminal anchoring of ACSL3 to the cytosolic leafl et of the endoplasmic reticulum (ER) by an amphipathic helix allows effi cient translocation to emerging lipid droplets. Depletion of ACSL3 by RNAi caused a signifi cant reduction in fatty acid uptake, suggesting that the activation of fatty acids serves not only to provide reactive molecules for lipid metabolism but is also a driving force for the transport of fatty acids into the cell.

Plasmids
Human adipose differentiation-related protein (ADRP/ perilipin 2) followed by mRFP (23) was used as a lipid droplet marker protein. RFP-ER contains mRFP fused to human Sec61 ␤ ( 24 ), which is a widely used ER marker protein. CaBP1/PDI A6 is a rat liver protein disulfi de isomerase ( 22 ) and is a soluble lumenal ER protein. ACSL3 HA is full-length human ACSL3, tagged at the C terminus with the hemagglutinin epitope. The full-length cDNA for human ACSL3 (clone MGC 48741) was obtained from the RZPD (Deutsches Ressourcenzentrum fuer Genomforschung; now imaGenes, Germany), and the cDNA was verifi ed. The sequence was identical to the reference sequence (gb: NM_004457.3). PCR with primers sA3kpn (5 ′ -ACGTGGTACCAC CATGAATAACCACGTGTCTTC-3 ′ ) and aA3 (5 ′ -A CGTCTCGA GTCAAGCGTAATCTGGAACATCGTATGGGTATTTTCTTCC-ATACATTCGCTCAATGTCC-3 ′ ) yielded a cDNA coding for Cterminally HA (hemagglutinin) epitope tagged ACSL3. This was digested with Kpn I and Xho I and ligated into pcDNA3 (Invitrogen). GFP-ACSL3 HA codes for GFP, and HA tagged full-length human ACSL3. The cDNA of ACSL3 HA was moved with KpnI and ApaI into pEGFP-C1 (Clontech, Mountain View, CA). GFP-⌬ Nt-ACSL3 HA lacks the cDNA part, which codes for amino acids 1-68 of human ACSL3. GFP-ACSL3 HA was digested with Bgl II, the released fragment was removed by gel electrophoresis, and the remaining vector was religated. ⌬ Nt-ACSL3 HA was also Lipid extraction and thin layer chromatography (TLC) were performed as described ( 29 ), except that the running solvent was changed to chloroform:methanol:acetic acid:water 75:43:3:1 ( 33 ). The TLC was exposed to phosphorimaging plates, which were scanned by the BAS-1500 system (Fuji, Tokyo, Japan). Quantifi cation was done with ImageJ.

Quantitative real-time PCR
Measurement of mRNA expression was by calibrator-normalized effi ciency corrected relative quantifi cation using the Light-Cycler system (Roche, Mannheim, Germany) as described ( 25 ). The reference gene was human ␤ -actin (primer, 5 ′ -3 ′ orientation: AGGATGCAGAAGGAGATCACT; GGGTGTAACGCAACTAAGT-CATAG). Calibration plasmids FATP1, FATP2, and FATP4 were provided by Paul A. Watkins, Kennedy Krieger Institute, Baltimore, MD and FATP3 by Johannes Berger, Medical University Vienna, Austria. For human ACSL1 the PCR product obtained with the RT-PCR primers was cloned into the pGEM-T vector (Promega).
Images were arranged with Adobe Photoshop and labeled with Adobe Illustrator (Adobe Systems Inc., Mountain View, CA).

The N-Terminus of ACSL3 contains sorting information for lipid droplets and the endoplasmic reticulum
We observed before that the N-terminal domains of the mammalian acyl-CoA synthetases ACSL1 and FATP4/ ACSVL4 were suffi cient to target reporter proteins to specifi c subcellular compartments and that their localization
For staining of endogenous ACSL3, fi xed cells were permeabilized with 0.1% saponin and blocked with 0.5% gelatin and 0.5% BSA. Indirect immunofl uorescence was performed as described ( 29 ).
Selective permeabilization of transiently transfected COS cells was achieved by using 10 µM digitonin for 10 min at room temperature. To stain the lumenal ER antigen CaBP1/PDI A6, fi xed cells were treated for 2 min with methanol prechilled at Ϫ 20°C.
For following the same cell over time, 12 mm cover slips were fi tted into a metal slide holder. After addition of control or oleic acid containing medium, cells were imaged using a 60× oil immersion objective mounted on an Olympus BX41 equipped with a F-view II CCD camera controlled by the cell^D software (Olympus, Hamburg, Germany). This microscope was also used for Figs. 3A, 4B, and 4C . Figs. 1, 2B-E, 4B-D, 5B, and 6A were obtained with the confocal microscope Leica TCS SP2 as described ( 32 ).

Floatation of lipid droplets
Transiently transfected COS cells (A3Nt-GFP, GFP-ACSL3HA, and GFP-⌬ Nt-ACSL3 HA ) from one 55 cm 2 dish and nontransfected COS cells from two 152 cm 2 dishes were incubated for 39 h with 600 µM oleic acid bound to fatty acid free BSA in a molar ratio of 6:1. Cells were then collected with PBS, mixed, and centrifuged for 5 min at 500 g . Separation of lipid droplets from other cellular components was essentially done as described ( 23 ). Briefl y, the cell pellet was resuspended in sucrose buffer A (50 mM Tris-HCl [pH 7.4], 20 mM sucrose) and homogenized by passing 15 times through a syringe fi tted to a 22 gauge needle followed by 20 strokes in a glass Dounce homogenizer. Homogenization was confi rmed by inspection under a microscope. The homogenate was centrifuged at 1,000 g for 10 min at 4°C, yielding a postnuclear supernatant, which was mixed with an equal amount of high-sucrose buffer (50 mM Tris-HCl [pH 7.4], 2 M sucrose) and overlaid with buffer A. The gradients were centrifuged in a SW40 rotor at 100,000 g for 3 h at 4°C. Four fractions with equal volume were collected: a top fraction containing lipid droplets (#1), an intermediate fraction (#2), and two bottom fractions (#3/4) containing cytosol and membranes. Proteins were concentrated by methanol-chloroform precipitation, separated by 8% SDS-PAGE, transferred by Western blotting, and detected by ECL.

Fatty acid uptake and TLC
COS-7 or A431 cells were preincubated for 1 h in DMEM supplemented with glutamine but without FCS. For short-term uptake, cells from one 10 cm 2 well were collected with PBS, centrifuged at 200 g , and resuspended into 400 µl PBS. To three 100 µl aliquots, 100 µl of labeling mix (340 µM [3H]oleate 8 Ci/ mol, 340 µM BSA in PBS) were added for 60 s. Uptake was terminated by adding an excess of ice cold 0.5% w/v BSA in PBS. Cells were washed four times and solubilized in 1 M NaOH. Aliquots were measured in a ␤ -counter (LS 6500; Beckman Coulter, Brea, CA) for oleate content and with Bradford solution (Biorad, Hercules, CA) for protein concentration. Long-term uptake was analyzed with adherent cells grown in 12-well plates. Eight hundred microliters of labeling mix (200 µM [3H]oleate/[3H]arachidonate 0.5 Ci/ mol, 100 µM BSA in DMEM) replaced the standard medium for 3 h. Preincubation, washing, and analysis was as for short-term uptake. Three aliquots of each cell population were analyzed for every experiment. Statistical analysis by Student's t -test was based on at least three independent experiments (SPSS 16.0; SPSS Inc., Chicago, IL). Errors bars correspond to SEM in Fig. 5 and to SD in Fig. 6 . Radioactive fatty acids were from PerkinElmer (Waltham, MA).

Fig. 1.
The N-terminus of ACSL3 confers localization to the ER and to lipid droplets. A: Domain organization of ACSL3 and overview of the constructs used. Wild-type human ACSL3 contains 720 amino acids, corresponding to 80.4 kDa (http://www.uniprot.org/uniprot/ O95573). The AMP-binding domain spanning S136 to A612 (as defi ned by Pfam at http://smart.embl-heidelberg.de/) is colored in remaining ACSL and ACSVL/FATP family proteins, but none of these was found on LDs ( 37 ). A striking example of the differential localization of acyl-CoA synthetases is given by RFP-FATP4 located at the ER and A3 Nt -GFP on LDs after fatty acid supplementation ( Fig. 2B ). This hints at compartmentalized metabolism of acyl-CoA synthetases even though it is commonly assumed that they are all oriented toward the cytosolic side of cellular membranes ( 38 ).

Emerging lipid droplets
Overnight incubation with fatty acids leads to the appearance of numerous clustered A3 Nt -GFP labeled LDs with an apparent size range from 1.5 to 2.0 µm ( Figs. 1C, D and was identical to the full-length wild-type proteins ( 29 ). Analysis of human ACSL3 by the Pfam protein family database ( 35 ) showed a large AMP-binding domain fl anked by shorter N-terminal and C-terminal regions ( Fig. 1A ). We designed a cDNA coding for a fusion protein containing the N-terminal amino acids of ACSL3 followed by GFP and expressed this plasmid in mammalian COS-7 cells (A3 Nt -GFP) ( Fig. 1B-E ). Immunofl uorescence microscopy showed a fi ne reticular staining in the cytoplasm, which colocalized with a red fl uorescent marker protein for the endoplasmic reticulum (ER) ( Fig. 1B ). Several circular objects (typically 2 µm in diameter) did not correspond to the ER but overlapped with the lipid droplet (LD) marker adipose differentiation-related protein (ADRP/perilipin 2) ( Fig. 1C ), which is consistent with the localization of endogenous ACSL3 to LDs in human hepatoma cells ( 19,20 ). The localization of A3 Nt -GFP to LDs was especially striking after overnight treatment of cells with oleic acid, which stimulated the formation of large LD clusters in the cytoplasm ( Fig. 1C, bottom). A3 Nt -GFP showed the same staining pattern as epitope-tagged, full-length ACSL3 ( Fig. 1D ) and endogenous ACSL3 ( Fig. 1E ). Floatation of LDs on sucrose density gradients confi rmed the presence of A3 Nt -GFP on LDs and demonstrated the cofractionation with endogenous ACSL3 ( Fig. 1D ). LD fractions contained TIP47/perilipin-3, a long established marker protein ( 36 ). This suggests that the essential targeting information for localization to LDs and the ER is contained within the N-terminus of ACSL3.
The dual localization of ACSL3 was further investigated in two different cell lines. Kidney-derived epithelial Vero cells showed predominantly a cytoplasmic reticular ER staining of A3 Nt -GFP, with only a few scattered LDs ( Fig. 2A , top panel). In HepG2 hepatoma cells, A3 Nt -GFP marked predominantly clusters of LDs colocalizing with ADRP/perilipin 2 ( Fig. 2A, bottom panel). This indicates that the distribution of ACSL3 is cell type specifi c, in addition to the dependence on fatty acid supply.
The localization of A3 Nt -GFP to LDs was unique among mammalian acyl-CoA synthetases. We also analyzed nine other GFP fusion proteins containing the N-termini of the purple, and the hydrophobic domain (HD) is indicated in yellow. A3 Nt -GFP contains the N-terminus of ACSL3 (M1-L135) fused to GFP (green). HA (light blue) is the hemagglutinin epitope tag. GFP-⌬ Nt-ACSL3 HA lacks the N-terminus of ACSL3 (M1-Y68). B: Partial localization of A3 Nt -GFP to the endoplasmic reticulum (ER). Transient coexpression of A3 Nt -GFP and the red fl uorescent membrane marker protein for the ER (RFP-ER) in COS-7 cells. Cells were fi xed 24 h post transfection and analyzed by confocal laser scanning microscopy. A single representative section is shown. The colocalization is especially evident at the nuclear envelope, which is continuous with the ER. Bar, 10 µm. C: Partial localization of A3 Nt -GFP to lipid droplets (LD). Transient expression of A3 Nt -GFP and the LD marker protein (ADRP/ perilipin 2) in COS-7 cells. The colocalization between A3 Nt -GFP and ADRP-RFP is even more prominent if the culture medium (DMEM) is supplemented with oleic acid overnight (lower panel; DMEM + OA). Confocal sections; bar, 10 µm. D: Comparison of A3 Nt -GFP to full length wild-type ACSL3. COS-7 cells were fi xed and permeabilized 24 h after cotransfection with A3 Nt -GFP and epitope-tagged human ACSL3. Indirect immunofl uorescence staining of ACSL3 was achieved by sequential incubation with mouse anti-HA and donkey antimouse coupled to Cy3 (red). The overlap between both proteins is apparent under standard conditions (DMEM) as well as after oleic acid supplementation (DMEM + OA). Confocal sections; bar, 10 µm. E: Comparison of A3 Nt -GFP to endogenous ACSL3. Left: Indirect immunofl uorescence microscopy. The affi nity purifi ed antibody against ACSL3 (red) stains the same structures as A3 Nt -GFP expressed transiently in COS-7 cells. Confocal sections; bar, 10 µm. Right: Subcellular fractionation. Postnuclear supernatants of homogenized COS-7 cells were applied to sucrose density gradients. Membrane (M) and cytosolic (C) proteins remain at the bottom of the gradient (fractions 3 and 4), but lipid droplet associated proteins fl oat into the top fraction (fraction 1). Endogenous ACSL3 was detected by the antibody raised against the C-terminal half of recombinant human ACSL3. A3 Nt -GFP was recognized by the GFP antiserum, and TIP47 served as a lipid droplet marker protein. 2B ). To determine how the formation of these lipid droplets was initiated, we used Ptk2 cells, which feature a large and fl at cell body, allowing the easy discrimination of subcellular structures by light microscopy. Already 2 min after the addition of oleic acid, A3 Nt -GFP appeared in numerous tiny punctate entities that increased in fl uorescence intensity over time ( Fig. 3A ). A dominant-negative caveolin mutant had been suggested as an expressable marker to follow LD biogenesis ( 4,39 ). We compared cav1 DGI -GFP with a red fl uorescent A3 Nt variant and found an exceptionally high degree of colocalization after the addition of fatty acid to the medium ( Fig. 3B ). To further verify that the tiny puncta are indeed emerging LDs, we demonstrated that A3 Nt -GFP showed the same behavior in COS-7 cells ( Fig. 3C ). These cells also allowed the labeling of endogenous TIP47/perilipin-3 by indirect immunofl uorescence, confi rming that A3 Nt -GFP is present on early lipid droplets ( Fig. 3D ). The localization shift of A3 Nt -GFP from the ER to newly emerging LDs was due to the translocation of already existing molecules because inhibition of protein synthesis with cycloheximide did not change the immunofl uorescence pattern (data not shown).

Membrane topology of ACSL3
The endoplasmic reticulum has an aqueous lumen segregated from the cytosol by a classical membrane bilayer, but lipid droplets feature a hydrophobic core surrounded by a phospholipid monolayer. These fundamental differences led us to the question of which topological organization would allow ACSL3 to be tightly associated with both membrane systems.
As assumed in general for acyl-CoA synthetases, the enzyme domain of ACSL3 is expected to face the cytosol. In silico predictions suggested that amino acids Ile21-Phe43 could form an integral transmembrane domain ( Fig. 4A ), which would put the fi rst 20 amino acids into the lumen of the ER. We tested this by selective permeabilization using an ACSL3 variant containing two antibody epitopes, one  of fatty acids with CoA ( 13 ). We therefore analyzed lipid extracts of A431 RNAi cells by TLC after labeling with radioactive fatty acids. We observed a relative decrease of phosphatidylethanolamine ( Ϫ 11%**) and a relative increase of phosphatidylcholine (PC) (+9.3%*) when oleic acid was used for labeling. With arachidonic acid, these changes were less pronounced and did not reach statistical significance. However, phosphatidylserine showed a relative decrease ( Ϫ 14%*).

Functional requirement for the N-terminus of ACSL3
The N-terminus of ACSL3 is suffi cient for the targeting of GFP to lipid droplets and the ER ( Figs. 1, 2 ), but it was not clear if ACSL3 would contain additional sorting information. Therefore, we constructed GFP-⌬ Nt-ACSL3 HA lacking the fi rst 68 amino acids, including the hydrophobic domain. The corresponding protein localized to the cytoplasm but was excluded from LDs ( Fig. 6A ). Subcellular fractionation confi rmed that GFP-⌬ Nt-ACSL3 HA was not associated with LDs, in contrast to GFP-ACSL3 HA containing the full length ACSL3 ( Fig. 6B ). Cells expressing GFP-⌬ Nt-ACSL3 HA did not show increased acyl-CoA synthetase activity ( Fig. 6C ), suggesting that the proper membrane anchoring by the N-terminus is functionally required. In line with this, there was no effect on the cellular uptake of fatty acids when GFP-⌬ Nt-ACSL3 HA was overexpressed in COS-7 cells ( Fig. 6D ).

DISCUSSION
Lipid droplets have become an intensely studied intracellular organelle in recent years because of their high relevance for lipid homeostasis and associated metabolic diseases. Here, we have characterized ACSL3, an enzyme of the acyl-CoA synthetase family, which is localized to lipid droplets.

Targeting of ACSL3 to LDs
We show that the N-terminal region of ACSL3 is sufficient to confer localization of fl uorescent reporter proteins to emerging and mature lipid droplets. In addition, ACSL3 is localized to ER membranes, which is especially apparent when cells contain only few LDs. Some LD-associated proteins feature more than one targeting region (e.g., perilipin A [ 43 ]). We therefore analyzed the localization of a mutant ACSL3 lacking the N-terminus, demonstrating that this region is essential for sorting to LDs. We found that the enzyme activity of ACSL3 is also dependent on the N-terminal region. This suggests that the membrane association and proper orientation of this enzyme at LD and ER membranes is functionally required.
Proteomics of purifi ed lipid droplets has greatly contributed to our understanding of this organelle ( 7 ), but some reservations remain ( 9,44 ). In addition to ACSL3, ACSL1 and ACSL4 have been identifi ed repeatedly in these studies. However, we found ACSL1 on mitochondria by immunofl uorescence microscopy ( 29 ), and this localization did not change after induction of LDs (B. Rudolph, R. Großmann, and J. Füllekrug, unpublished observations). ACSL4 cloned at each end of the protein (GFP-ACSL3 HA ). The detergent digitonin permeabilizes the plasma membrane effi ciently, allowing access of antibodies to cytoplasmically oriented epitopes. The ER membrane, however, is not suffi ciently permeabilized, which we verifi ed by antibodies against a luminal ER protein ( Fig. 4C ). HA antibodies stained GFP-ACSL3 HA as expected. GFP antibodies also gave a distinct pattern with digitonin, regardless whether GFP-ACSL3 HA was localized to the ER or to LDs ( Fig. 4B ). This suggests that ACSL3 has both the N-and the C-terminus oriented toward the cytosol. The hydrophobic domain at the N-terminus likely forms an ␣ -helical secondary structure ( Fig. 4A ), and the corresponding projection showed a striking segregation between hydrophobic and polar side chains ( Fig. 4D ). This amphipathic helix would allow equal anchoring in the cytoplasmic leafl et of the ER and the phospholipid monolayer of lipid droplets.

Overexpression and knockdown of ACSL3 correlate with fatty acid uptake
Overexpression of ACSL3 increases the cellular acyl-CoA synthetase activity ( 18,40 ) ( see Fig. 6C ). Building on the general concept that metabolism contributes to fatty acid uptake ( 41 ) and our previous observations regarding ACSVL4/FATP4 and ACSL1 ( 29 ), we analyzed the connection between ACSL3 and the cellular incorporation of fatty acids. Overexpression of ACSL3 in COS-7 cells resulted in an increased uptake of oleic acid when measured after 60 s or 3 h ( Fig. 5A ). Because ACSL3 is localized intracellularly to the ER and LDs, this effect is indirect and is best explained by assuming that fatty acids are metabolically trapped as CoA derivatives. COS-7 cells allow high transfection rates but are not of human origin, which complicates DNA sequence-specifi c experimental approaches like RNAi and qRT-PCR. ACSL3 is the second most abundant protein on LDs prepared from human A431 cells ( 42 ). Therefore, we chose this cell model to analyze the functional consequences of a reduced amount of ACSL3. We confi rmed that A3 Nt -GFP localized to LDs and the ER in A431 cells ( Fig. 5B ), similar to what we observed for COS-7 cells. Stable depletion of ACSL3 by RNA interference was achieved by retroviral integration of an shRNA plasmid. Quantifi cation of transcription by qRT-PCR gave a knockdown effi ciency of 81%, and the expression of the ACSL3 protein was diminished correspondingly ( Fig. 5C ). The uptake of arachidonic acid was reduced by 19%; the reduction for oleate was 10% ( Fig. 5D ). The stronger reduction for arachidonic acid coincides with the substrate preferences reported for ACSL3 ( 16,18 ). The expression of ACS enzymes is redundant: The A431 RNAi cells do not only contain the remaining ACSL3 (relative expression level by qRT-PCR: 1.90) but also ACSL1 (14.19), ACSL4 (6.58), and several FATP acyl-CoA synthetases (ACSVL4/FATP4 0.88, FATP3 0.61, and FATP1 0.26). Therefore, even the effi cient depletion of ACSL3 leaves a considerable amount of enzyme activity with the A431 RNAi cells.
Many observations suggest that mammalian acyl-CoA synthetases infl uence lipid metabolism beyond the esterifi cation

Lipid droplet biogenesis
A3 Nt -GFP rapidly shifts from the ER to emerging lipid droplets when triglyceride synthesis is boosted by the external addition of fatty acids. This movement from one membrane system to another is different from the behavior from HepG2 cells also did not colocalize with LDs (R. Großmann and J. Füllekrug, unpublished observations). We are therefore confi dent that ACSL3 is the only long-chain fatty acyl-CoA synthetase signifi cantly associated with lipid droplets.  uptake is determined by extracellular supply as well as intracellular metabolic demand, with plasma membrane proteins facilitating and regulating the transport process ( 41,60,61 ). The role of ACSL3 and other acyl-CoA synthetases would not only be to provide activated fatty acids for lipid metabolism, but at the same time could prevent the loss of fatty acids to the outside. The intracellular metabolic trapping (or vectorial acylation [ 14] ) by esterifi cation with CoA might be the essential driving force for uptake of free fatty acids from the extracellular space. In line with this, exogenous expression of mammalian ACSL1, -4, and -6 rescued the fatty acid transport activity of yeast cells ( 62 ). Overexpression or depletion of acyl-CoA synthetases from the ACSVL (FATP) family also leads to corresponding changes in fatty acid uptake, but the interpretation of these fi ndings is complicated by the assertion from some laboratories that these proteins are fatty acid transporters localized to the plasma membrane ( 15,61 ). However, FATP4/ACSVL4 was shown by us to be resident in the endoplasmic reticulum, and this localization was suffi cient to drive the uptake of fatty acids in an enzymedependent manner ( 29 ). The intracellular localization of ACSL3 is not contentious, and the effect on fatty acid transport therefore necessarily indirect.
Key regulators of lipid metabolism modify the expression of ACSL3: The LXR nuclear receptors increased ACSL3 levels in placental trophoblasts (which corresponded to fatty acid uptake [ 63] ), and PPAR ␦ regulated the expression of ACSL3 in hepatoma HepG2 cells ( 64 ). RNAi of ACSL3 in primary hepatocytes decreased the activity of PPAR ␥ and other lipogenic transcription factors ( 65 ). Although the incorporation of external fatty acids was not investigated, this offers an alternative explanation for the reduced uptake observed by us: Less active PPAR ␥ could decrease the overall capacity for cellular lipid metabolism and, consequently, fatty acid uptake. However, in the same study, other acyl-CoA synthetases (ACSL1, -4, -5 and FATP2, -4, -5) did not change the activity of PPAR ␥ , suggesting that the impact of these enzymes on fatty acid uptake (summarized in References 13 and 66 ) is likely independent of transcriptional regulation.

Functional relevance of ACSL3 on lipid droplets
Activated fatty acids are rapidly metabolized, and excessive acyl-CoAs are used for the synthesis of triglycerides, which are stored in lipid droplets. It is tempting to speculate that fatty acids taken up with the help of ACSL3 are used preferentially for this storage pathway as opposed to membrane synthesis or ␤ -oxidation. The double immunofl uorescence localization of cytoplasmic ER-bound ACSVL4/FATP4 and LD-localized ACSL3 is highly reminiscent of compartmentalized metabolism ( Fig. 2B ). Although differing in kinetics and substrate affi nities, both enzymes catalyze the same reaction, the esterifi cation of long chain fatty acids (e.g., oleate) with CoA. The enzyme part of these membrane-bound proteins, however, is not lumenal but faces the cytosolic side of their respective subcellular organelles. It is not easy to see how the different localizations would have functional relevance because it is of the constitutive LD proteins of the perilipin family ( 5,6 ). For instance, ADRP/perilipin 2 and TIP47/perilipin 3 rapidly exchange between the cytosol and the LD surface ( 45,46 ). How membrane proteins like ACSL3 translocate from the ER to LDs is not obvious because these organelles are organized in a strikingly dissimilar fashion: Whereas the ER features an aqueous lumen segregated from the cytoplasm by a phospholipid bilayer, cytosolic LDs have a hydrophobic core that is surrounded by a lipid monolayer only.
ACSL3 displays a hairpin topology with both the N-and the C-terminus facing the cytosol, and this orientation is shared by DGAT2 ( 47 ) and caveolins ( 48 ), which are also localized partially to LDs. However, for these proteins, the stretch of hydrophobic amino acids is much longer than for ACSL3 and may give rise to two ␣ -helical transmembrane domains ( 8,49 ). The hydrophobic region of ACSL3 could in principle allow the formation of an integral transmembrane domain, but our data are more consistent with an amphipathic helix embedded only in the cytosolic leafl et of the ER double membrane. Similar to ACSL3, CTP:phosphocholine cytidylyltransferase, viperin, and two viral proteins (hepatitis C core protein and NS5A) were recently suggested to be bound to LDs by an amphipathic helix (50)(51)(52)(53). Assuming that the outer leafl et of the ER is at least initially continuous with the lipid monolayer of the emerging lipid droplet ( 10,54 ), the proposed topology would allow the traffi cking of ACSL3 molecules from the ER to nascent LDs by simple diffusion.
Regarding how ACSL3 is translocated so effi ciently to emerging lipid droplets, we think it unlikely that there is a proteinaceous ACSL3 receptor because this putative protein would need to be directed to nascent LDs even earlier than ACSL3. Because the surface lipid composition of LDs differs substantially from the ER membrane ( 58 ), it is conceivable that ACSL3 simply has a higher affi nity for the lipid monolayer of LDs. Lipid-mediated sorting has been primarily associated with the plasma membrane and the trans-Golgi network ( 59 ), but it seems certain that other organelles would also take advantage of this organizing principle.

Role of ACSL3 in fatty acid uptake and metabolism
Depletion of ACSL3 reduced fatty acid uptake, whereas overexpression strongly increased incorporation of fatty acids. Although this experimental range of expression levels is unlikely to be realized under physiological circumstances, the results are nevertheless suggesting that the acyl-CoA synthetase activity of ACSL3 is a critical factor for the cellular uptake of fatty acids. In general, the extent of in common tacitly assumed that fatty acids and acyl-CoAs are equilibrated throughout the cytoplasm. The differences in subcellular localization combined with the apparently redundant expression of ACSL isoforms within the same cell has led to the hypothesis that channeling by acyl-Co synthetases may determine the metabolic fate of fatty acids ( 38,67 ). Over the years, considerable evidence has accumulated (reviewed in References 13, 68, and 69 ), but the molecular mechanism remains unclear. We observed a phospholipid class switch from phosphatidylethanolamine to PC in A431 RNAi cells stably depleted for ACSL3. A different study reported that transient knockdown of ACSL3 in Huh7 hepatoma cells led to a specifi c decrease of fatty acid incorporation into PC ( 70 ), suggesting that, even when the same acyl-CoA synthetase is depleted, there are still cell type-specifi c effects.
Local synthesis of triglycerides at the LD surface has been demonstrated ( 20,23 ), although it is not clear if this process would be effi cient enough to account for a sustained growth of lipid droplets independent from the endoplasmic reticulum. The acyl-CoA needed for the TG synthesis is very likely provided mostly by ACSL3 ( 20 ). Even more intriguing is the idea that the local synthesis of phosphatidylcholine could enable the rapid growth of the phospholipid monolayer to keep up with an expanding hydrophobic core when TG synthesis rate is high. The role of ACSL3 would be to provide activated fatty acids for the synthesis of PC by lyso-PC acyltransferases, which have recently been shown to be partially localized to LDs ( 42 ). When cellular ACSL activity is inhibited pharmacologically, LDs decrease in size and number ( 20 ). This suggests another possible function of LD-localized ACSL3 in counteracting basal lipolysis by promoting reesterifi cation of fatty acids into triacylglycerol, preventing the effl ux of fatty acids, which are constantly liberated even under unstimulated conditions.
In conclusion, the N-terminal region of ACSL3 displays a high affi nity for LDs and enables the smooth and effi cient movement from the ER to just-emerging lipid droplets. At the same time, ACSL3 activates fatty acids for lipid metabolism and contributes indirectly to cellular fatty acid uptake.
The machinery for the formation of LDs presumably has a long evolutionary history. It remains for future studies to determine if the biogenesis of lipid droplets is in essence a simple self-organizing process, relying on the segregation of triglyceride-enriched lipid membranes from the parental ER organelle. This primordial LD membrane would accumulate diffusing molecules with an inherent affi nity (like ACSL3), and these will in turn sustain and regulate the maturation to lipid droplets. There is little doubt that the relevance of LDs for lipid homeostasis and overall metabolic regulation is still underestimated, with many surprises ahead.
We thank Simone Staffer for help with fatty acid uptake assays, Sabine Tuma for assistance with tissue culture, Julia Wunsch for cloning RFP-FATP4, and Wolfgang Stremmel for continuous support. Paul A. Watkins (Kennedy Krieger Institute, Baltimore, MD) and Johannes Berger (Medical University Vienna, Austria) kindly provided calibration plasmids for qRT-PCR.