Structure of a CGI-58 Motif Provides the Molecular Basis of Lipid Droplet Anchoring*

Background: CGI-58 activates the key intracellular lipase ATGL. Results: Solution structure of the N-terminal lipid droplet (LD)-binding motif of CGI-58 bound to dodecylphosphocholine micelles. Conclusion: The LD-binding motif acts independently to anchor proteins to LDs and consists of two LD-binding arms. Significance: The structure of the peptide LD anchor sheds light on the interaction of CGI-58 with LDs.

Triacylglycerols (TGs) stored in lipid droplets (LDs) are hydrolyzed in a highly regulated metabolic process called lipolysis to free fatty acids that serve as energy substrates for ␤-oxidation, precursors for membrane lipids and signaling molecules. Comparative gene identification-58 (CGI-58) stimulates the enzymatic activity of adipose triglyceride lipase (ATGL), which catalyzes the hydrolysis of TGs to diacylglycerols and free fatty acids. In adipose tissue, protein-protein interactions between CGI-58 and the LD coating protein perilipin 1 restrain the ability of CGI-58 to activate ATGL under basal conditions. Phosphorylation of perilipin 1 disrupts these interactions and mobilizes CGI-58 for the activation of ATGL. We have previously demonstrated that the removal of a peptide at the N terminus (residues 10 -31) of CGI-58 abrogates CGI-58 localization to LDs and CGI-58-mediated activation of ATGL. Here, we show that this tryptophan-rich N-terminal peptide serves as an independent LD anchor, with its three tryptophans serving as focal points of the left (harboring Trp 21 and Trp 25 ) and right (harboring Trp 29 ) anchor arms. The solution state NMR structure of a peptide comprising the LD anchor bound to dodecylphosphocholine micelles as LD mimic reveals that the left arm forms a concise hydrophobic core comprising tryptophans Trp 21 and Trp 25 and two adjacent leucines. Trp 29 serves as the core of a functionally independent anchor arm. Consequently, simultaneous tryptophan alanine permutations in both arms abolish localization and activity of CGI-58 as opposed to tryptophan substitutions that occur in only one arm.
Triacylglycerols (TGs) 2 are stored in lipid droplets (LDs) comprising a core of neutral lipids (TGs and sterol esters) surrounded by a monolayer of phospholipids (1). The protein "comparative gene identification 58 ," also known as ␣/␤-hydrolase domain 5 (ABHD5), is an important stimulatory protein of the first step in intracellular lipolysis (2,3). In this catabolic process, adipose triglyceride lipase (ATGL) catalyzes the hydrolysis of TGs stored in LDs to diacylglycerols and free fatty acids (FFAs). Hormone-sensitive lipase and monoacylglycerol lipase subsequently hydrolyze diacylglycerols and monoacylglycerols, respectively, to generate FFAs and glycerol molecules (4).
Mutations in the human gene encoding CGI-58 lead to neutral lipid storage disease with a severe skin defect termed ichthyosis (NLSD-I) (5). Although the ATGL stimulating function of CGI-58 appears causative for the neutral lipid storage phenotype in affected patients, the frequently observed symptoms of hepatomegaly, hepatic steatosis, and ichthyosis are indicative of an ATGL-independent function of CGI-58 (6 -8).
The rate of intracellular lipolysis on the surface of LDs depends on post-translational modification events, multiple protein-protein interactions and lipase-ligand interactions at the lipid-water interphase (2, 9 -23). Under basal conditions, CGI-58 binds to the LD coating protein perilipin 1 in 3T3-L1 adipocytes (20). In this state, CGI-58 does not interact with ATGL and ATGL activity remains low (24,25). Phosphorylation of perilipin 1 and CGI-58 by protein kinase A (PKA) leads to rapid release of CGI-58 from perilipin 1-coated LDs and subsequent activation of ATGL (18,26).
The role of perilipin 1 in the recruitment of CGI-58 to LDs is not well understood. CGI-58 has been shown to activate ATGL also on artificial LD substrates lacking perilipins (12,15). Currently, it remains unknown whether CGI-58-mediated activa-tion of lipolysis occurs due to increased access to the substrate, conformational changes induced in ATGL, increased product release (e.g. channeling the produced FFA away from the reaction site), or increased lipolysis due to interaction with fatty acid-binding proteins (16). Interestingly, the selectivity of ATGL hydrolysis at the sn-2 position of the glycerol backbone broadens to the sn-1 position upon interaction with CGI-58 (27). The lack of high-resolution structures of CGI-58 and ATGL or the protein-protein complex in the presence of a LD surface represents a major bottleneck in understanding the function of these proteins and their interaction surfaces.
The interaction of CGI-58 with ATGL occurs within the N-terminal patatin domain-related region of ATGL (12,28). A homology model of CGI-58 reveals a core ␣/␤-hydrolase structure consisting of eight mostly parallel ␤-strands surrounded by ␣-helices and loops, a cap region comprising ␣-helices, and a short mostly unstructured N-terminal tryptophan-rich stretch (15). Almost the entire CGI-58 protein is required to activate ATGL, because CGI-58 variants with major deletions from the N or C terminus are not capable of activating ATGL (15). The N-terminal Trp-rich region serves an essential role in the localization of CGI-58 to LDs, which remains a strict requirement for ATGL activation (15).
To better understand the mechanism of CGI-58 LD binding, we solved the structure of the N-terminal fragment of CGI-58 (peptide Val 10 to Lys 43 , CGI-58_V10-K43) bound to dodecylphosphocholine (DPC) micelles, which serve as excellent mimics of the LD surface. The structure reveals that the region Ser 19 -Cys 30 constitutes a LD anchor motif with Trp 21 and Trp 25 forming a hydrophobic core along with the hydrophobic residues Leu 22 and Leu 26 . This hydrophobic core constitutes the left arm of the anchor. The more isolated Trp 29 is flanked by two prolines (Pro 27 and Pro 31 ) and serves as the functionally independent right arm of the anchor.
A fusion protein containing just the CGI-58 LD anchor motif (amino acids 19 -35) fused to yellow fluorescent protein (YFP) localizes to LDs supporting the concept of the LD-anchor as an independent functional motif. Selective permutations converting single tryptophans of the LD anchor to alanines do not alleviate the ability of CGI-58 to localize to LDs or activate ATGL. However, substitutions in both arms of the LD anchor (W21A and W29A) abolish the ability of CGI-58 to localize to LDs and to activate ATGL.

Experimental Procedures
Generation of Trp Variants of CGI-58 -Wild type (WT) and all mutants of CGI-58 were cloned into the plasmid pEYFP-N1 (BD-Biosciences Clontech) coding for a C-terminal EYFP tag. Generation of wild type-CGI-58 (WT-CGI-58) and the pointmutated variants W21A, W29A, and W21A/W25A was described earlier (15). The single point mutant W25A and the double mutant W21A/W29A were generated by site-directed mutagenesis of a vector encoding for WT-CGI-58 with a C-terminal EYFP tag. The N-terminal peptide containing just the LD anchor mCGI_19 -35 was generated upon amplification of the sequence using the forward primer mCGI_19 YFP N1 forward, 5Ј-GTGATGACCTCGAGATGTCAGGATGGCTG-3Ј; and the reverse primer mCGI_35 YFP N1 reverse, 5Ј-GGAATAG-GATCCGCTGATGTAGATGTGGGACACC-3Ј followed by ligation into XhoI and BamHI sites of the vector. The correctness of all sequences was verified by DNA sequencing (LGC Genomics, Berlin, Germany).
Cellular Localization of mCGI-58 Variants-For localization studies, monkey embryonic kidney cells (COS-7, ATCC CRL-1651) were transfected with expression vectors (pEYFP-N1) encoding WT full-length and point mutants of mouse CGI-58 (mCGI-58) with a C-terminal fusion of YFP. COS-7 cells were maintained in Dulbecco's modified Eagle's medium (DMEM, Life Technologies) containing 4.5 g/liter of glucose, 10% fetal calf serum (FCS), and penicillin/streptomycin under a humidified atmosphere, 37°C, and 5% CO 2 . COS-7 cells were seeded on glass coverslips in 6-well plates (1.2 ϫ 10 5 cells/well) and transfected with YFP-tagged full-length or mutated mCGI-58 variants. 24 h after transfection, cells were incubated for 20 h in DMEM containing FCS, and supplemented with oleic acid (400 M) complexed to fatty acid-free BSA (Sigma) in a ratio of 3:1 to increase LD formation. LDs were stained with HCS LipidTOX Red Neutral Lipid stain (Life Technologies) and incubated for 10 min at 37°C. Microscopy was performed using a Leica TCS SP5 confocal microscope (Leica Microsystems GmbH) with a HCX PL APO CS 63ϫ 1.2 water objective. YFP fluorescence was excited at 514 nm and detected at 522-558 nm. LipidTOX was excited at 633 nm and detected at 650 -669 nm. Transmission images of cultured cells were also recorded. All presented experiments were repeated independently at least three times.
Preparation of Cell Extracts for Triglyceride Hydrolase Assay-COS-7 cells were transiently transfected with the different CGI-58 clones and pcDNA4/HisMax coding for Histagged ATGL (28) with Metafectene TM (Biontex GmbH) as described earlier (29). The cells were disrupted by sonication and resuspended in lysis buffer (0.25 M sucrose, 1 mM dithiothreitol, 1 mM EDTA, 20 g/ml of leupeptine, 2 g/ml of antipain, 1 g/ml of pepstatin, pH 7.0). Then, nuclei and unbroken cells were removed by centrifugation at 1000 ϫ g at 4°C for 5 min, and the supernatants were used for triglyceride hydrolase activity assays.
Assay for Triglyceride Hydrolase Activity-The substrate for the triglyceride hydrolase activity assay was prepared as described previously with minor modifications (29). Briefly, triolein and [9,10-3 H]triolein (10 Ci/ml) were emulsified in the presence of phosphatidylcholine/phosphatidylinositol using a sonicator (Virsonic 475, Virtis, Gardiner, NJ) and adjusted to 2.5% BSA (FFA free). The final substrate concentration was 0.3 mol/ml of triolein and 0.15 mg/ml of phosphatidylcholine/ phosphatidylinositol (3:1). The reaction mixture was prepared of lysates containing overexpressed HisMax-mATGL (30 g total protein) and the lysates expressing the different variants of CGI-58 (30 g of total protein). Activity assays were performed using 0.1 ml of cell lysate mixture and 0.1 ml of substrate in a water bath at 37°C for 60 min. The reaction was terminated by adding 3.25 ml of methanol/chloroform/heptane (10:9:7) and 1 ml of 0.1 M potassium carbonate, 0.1 M boric acid, pH 10.5. After centrifugation at 800 ϫ g for 20 min, the radioactivity in 0.2 ml of the upper phase was determined by liquid scintillation counting.

The Lipid Droplet Anchoring Peptide of CGI-58
Statistical Analysis-TG hydrolase activity measurements were performed in triplicates. Measured activities are represented as mean Ϯ S.D. Statistical significance was determined by the Student's unpaired two-tailed t test. Groups were considered to be significantly different for p Ͻ 0.05 (*), p Ͻ 0.01 (**), and p Ͻ 0.001 (***).
Preparation of the Synthetic Peptide "G18-E39"-A 22-residue peptide containing the amino acid sequence GSGWLTG-WLPTWCPTSTSHLKE corresponding to residues Gly 18 to Glu 39 of mCGI-58 (referred to as peptide "G18-E39") was purchased from a commercial supplier (Peptide Special Laboratories, Heidelberg). For NMR experiments, it was dissolved in buffer 7 containing 100 mM DPC-d 38 at a concentration of 1 mM and measured upon addition of 5% D 2 O.
NMR Experiments with Peptides V10-K43 and G18-E39 -All experiments on peptides V10-K43 and G18-E39 were recorded in the presence of DPC micelles. Due to the low solubility of the peptides in aqueous solvents, assignment of CGI-58 peptides V10-K43 and G18-E39 in the absence of detergent was not feasible.
Standard backbone experiments (30) (HNCA, HN(CA)CO, HNCACB) of the peptide V10-K43 were recorded on a 600 MHz Bruker spectrometer equipped with an Avance I console and a cryogenically cooled TCI 5-mm probe. The side chain experiments HCCH-TOCSY and (H)C(C-CO)NH-TOCSY were recorded on a 500 MHz Varian spectrometer equipped with a Unity Inova console and an HCN cryoprobe. An H(CC-CO)NH-TOCSY was recorded on a 750 MHz Bruker spectrometer equipped with a cryogenically cooled TCI 5-mm probe and an Avance III console. 15 N-and 13 C-dispersed three-dimensional NOESY experiments were recorded on a 700 MHz Varian spectrometer equipped with an Agilent dd2 console and HCN salt-tolerant cryoprobe (150 ms mixing time) and a 900 MHz Bruker spectrometer with Avance II console and TCI cryoprobe (80 ms mixing time), respectively. All backbone and side chain experiments, with the exception of the (H)C(C-CO)NH-TOCSY, were recorded using non-uniform sampling where 15-20% of the indirect dimension grid was sampled using Poisson Gap Sampling (31). Heteronuclear three-dimensional NMR experiments were recorded at 310 K to minimize transversal relaxation times on residues immersed in the micelles. Concomitantly, this temperature reduced the dynamic range of the sample, as residues exposed to the solvent exchange more rapidly with water at elevated temperatures and therefore lost some of their otherwise high signal intensity. Homonuclear two-dimensional TOCSY and two-dimensional NOESY spectra of peptide G18-E39 were recorded on a 900 MHz Avance II Bruker spectrometer equipped with a cryogenically cooled probe at 303 K using 90 and 200 ms mixing times, respectively. Uniformly collected NMR spectra were processed with NMRpipe (32). Non-uniformly sampled spectra were processed with hmsIST in combination with NMRpipe (31,33). All NMR spectra were visualized and analyzed with CcpNmr (34).
Paramagnetic relaxation enhancement (PRE) values were extracted and converted to distance restraints according to published protocols (37,38). We calculated the hydrodynamic radius of the DPC micelles to be ϳ30 Å (r ϭ 30 Å) based on the translational diffusion coefficient measured by dynamic light scattering (described below). The constants g (7.98 Å) and k (253 mM Ϫ1 Å 3 ) were used according to the literature (38). Gd(DTPA-BMA) was purchased as Gadodiamide from (Toronto Research Chemicals, Toronto, Canada) and added from a 60 mM stock in H 2 O. PRE-derived distance restraints were weighted at 30% with respect to NOEs, upper and lower boundaries of Ϯ2 Å were used.
NOEs from 15 N-and 13 C-dispersed NOESY-HSQCs of V10-K43 and a homonuclear NOESY of peptide G18-E39 were picked, assigned, integrated, and converted to distance restraints in CcpNmr (34). Restraints for torsion angles were prepared with TALOSϩ (39) and PRE-derived distance restraints were calculated as described above. 100 structures were calculated with a simulated annealing protocol using CYANA (40) and the 20 structures with the lowest energy target functions were chosen for deposition. Structures were visualized with PyMOL (41) and the quality of the structures was assessed with PSVS (42) and iCING (43).
Circular Dichroism (CD) Spectroscopy-For CD spectroscopy, a sample of peptide V10-K43 was prepared at 0.76 mg/ml in buffer 7 and DPC as described above for the preparation of NMR samples. A corresponding baseline sample was prepared without peptide. Data were measured with a Jasco J-715 spectropolarimeter at 0.01-cm path length between 190 and 260 nm with 0.1-nm steps, 1-nm bandwidth, and 1-s averaging time at 50 nm min Ϫ1 scanning speed. 10 spectra were recorded, averaged, and baseline corrected.
Dynamic Light Scattering-The hydrodynamic radius of micelles in the presence of peptide was measured to be 30 Å with dynamic light scattering (Protein Solutions DynaPro MS/X instrument, Protein Solutions Inc., Lakewood, NJ). The dynamic light scattering micelles with the peptides were mea-sured at 5-s acquisition time, 30% laser power, and 20 acquisitions. As a reference, 100 mM DPC was measured in H 2 O. The measured radius of 21 Å for the free micelle is in good agreement with the literature (38,44,45). The difference in micelle size is presumably due to higher salt and DPC concentration, which is a direct result of the peptide preparation process using spin concentrators.
Protein Data Bank (PDB) and Biological Magnetic Resonance Bank (BMRB) Accession Numbers-Coordinates and NMR resonance assignments have been deposited in the Protein Data Bank (PDB code 5A4H) (46) and Biological Magnetic Resonance Data Bank (BMRB code 25684) (47).

Results
The LD-binding Motif of CGI-58 Tolerates the Loss of Any Single Tryptophan Residue, but Not the Simultaneous Loss of Trp 21 and Trp 29 -Full-length mammalian CGI-58 localizes to LDs in differentiated 3T3-L1 adipocytes and COS-7 cells (Fig.  1A). This interaction involves the tryptophan-rich N terminus of CGI-58 (Trp 21 , Trp 25 , and Trp 29 ) in LD binding (15,20,48,49). CGI-58 lacking the first 31 residues or harboring changes in the three N-terminal Trp residues failed to co-localize to LDs (15), whereas conversion of Trp 21 to alanine (W21A) alone did not prevent the localization of CGI-58 to LDs or the activation of ATGL (Fig. 1C). Similarly, CGI-58 variants W25A and W29A retained their ability to localize to LDs; although somewhat reduced ATGL stimulation was observed for the variants W21A and W25A (Fig. 1, B and C).
Next, we generated variants with double amino acid exchanges, W21A/W25A, and W21A/W29A. Although the W21A/ W25A variant localized to LDs and concomitantly activated ATGL with undiminished capacity the W21A/W29A variant failed to localize to LDs and to activate ATGL (Fig. 1, B and D). This strengthens the functional relevance of proper CGI-58 localization observed previously (15) and supports the proposed prominent role for the N-terminal region of CGI-58.
To investigate whether the N-terminal region self-sufficiently localizes to LDs, we expressed a YFP-tagged peptide ranging from Ser 19 to Ser 35 (CGI_19 -35) in COS-7 cells and monitored its intracellular localization. The peptide localized to LDs in a manner reminiscent of wild type CGI-58 (Fig. 1, A  and E). However, when we tested the peptide CGI_19 -35 for its ability to activate the triacylglycerol (TG) hydrolase activity of ATGL, we did not observe any stimulating effect (Fig. 1F).
Resonance Assignments of the CGI-58 Peptides G18-E39 and V10-K43-To characterize the three-dimensional structure of the N-terminal LD binding region, we determined the solution structure using NMR spectroscopy. 96% of backbone and 74% of side chain resonances of the peptide V10-K43 bound to DPC micelles were assigned (Fig. 2A). The heteronuclear 15 N-and 13 C-dispersed NOESY-HSQC spectra of the peptide V10-K43 did not contain a sufficient number of cross-peaks for structure calculation, which is likely attributed to dynamics of the sample at 310 K. Therefore, we recorded homonuclear TOCSY and NOESY experiments on the synthetic 22-residue peptide mCGI-58_G18-E39 (G18-E39) at a lower temperature of 303 K. The resonance assignments from the longer peptide V10-K43 could be transferred and consequently, we assigned all NH and H␣ resonances of the unlabeled 22-residue peptide G18-E39 with the exceptions of the His 36 -H␣ proton and resonances corresponding to Gly 18 at the N terminus. 82% of non-water exchangeable side chain protons were also assigned. Trp 21

JOURNAL OF BIOLOGICAL CHEMISTRY 26365
The Trp 29 H⑀-1 resonance and additional Trp aromatic side chain resonances were assigned from the homonuclear spectra (Fig. 2B). Assignments were deposited in the BMRB accession number 25684.

The N-terminal Peptide of CGI-58 Reveals a Mostly
Unstructured Anchor-The observed chemical shifts of a protein or peptide are sensitive indicators of ␣-helix and ␤-sheet elements when compared with average random coil shifts. Thus, the assignments of the N-terminal peptides of CGI-58 reveal initial per-residue information on secondary structure elements. In particular, downfield shifts of 13 C␣ and 13 CO and upfield shifts of 1 H␣ resonances with averaged changes of 2.6, 1.7, and 0.38 ppm, respectively, would indicate an ␣-helix. Upfield shifts with averaged changes of 1.4 ppm for 13 C␣ and 13 CO, and downfield shifts of 0.38 ppm for 1 H␣, would indicate ␤-sheets (50 -52). Examination of 13 C␣, 1 H␣, and 13 CO shifts of the CGI-58 peptide V10-K43 in DPC micelles did not provide an indication of ␣-helix or ␤-sheet elements, although a propensity for helix formation might be inferred for the region Gly 20 -Gly 24 (Fig. 3,  A-C). Additionally, a circular dichroism (CD) spectrum of the peptide V10-K43 showed characteristics of a mostly unfolded peptide with a minimum around 200 nm (Fig. 3D). These experimental results are in good agreement with structure predictions. Secondary structure predictions indicate peptide V10-K43 to be partially unstructured, with a propensity to form ␣-helices (PSIPRED, Jpred (53,54)). A homology model of CGI-58 based on the structure of the Aspergillus niger epoxide hydrolase also indicates helical and unstructured parts in the N terminus of CGI-58 (15).
Relaxation The relaxation experiments on the CGI-58 peptide V10-K43 revealed a clear separation of three distinct regions, namely flexible N-and C-terminal regions and a more rigid central region (Fig. 4). T 1 times increase markedly between Trp 21 and Ser 33 (Fig. 4A), indicating reduced mobility. This is in agreement with the T 2 values (Fig. 4B), which were less than 100 ms for residues between Trp 21 and Ser 35 . T 2 times less than 100 ms in the LD binding region correspond to motion dynamics of a large (Ͼ20 kDa) protein. This indicates that this region comprises the LD anchor and is embedded in the LD mimicking micelle. Moreover, the rapid dynamics observed for the terminal regions indicate that these regions move independently of the LD anchor. This is further substantiated with the 15 N[ 1 H] NOE experiment. Residues experiencing rapid internal motion flank a considerably more rigid core between Gly 20 and Thr 32 (Fig. 4C).
Paramagnetic Relaxation Enhancements Reveal the Immersion Depth of the Peptide Anchor-To elucidate the orientation of the CGI-58 peptide V10-K43 in DPC micelles and to determine the boundaries of the LD-binding motif, we recorded longitudinal relaxation experiments in the presence of various Gd(DTPA-BMA) concentrations. PREs collected in the presence of this highly water soluble and inert compound correlate with the insertion depths of peptide residues and can be con-   verted to distance restraints during structure calculation (37,38). Based on dynamic light scattering experiments, the hydrodynamic radius of the DPC micelles was calculated to be 30 Å in the presence of the CGI-58 peptide. For protons more than 28 Å from the micelle center, only lower distance restraints were generated. All protons calculated to be within 28 Å of the micelle center correspond to residues ranging from Trp 21 to Trp 29 and thus reaffirm the depth of immersion of the LD anchor motif ( Table 1). The paramagnetic restraints are deposited with the accession number 5a4h at the Protein Data Bank.
Solution Structure of the N-terminal LD Anchor of CGI-58 -To assess the differences in motion dynamics and establish the mode of binding of the peptide V10-K43, we calculated its solution structure immersed in DPC micelles. 533 NOE distance restraints were used for structure calculation, along with 36 backbone angular restraints ( and ) and 66 PRE-derived distance restraints. The complete summary of quality statistics and experimental restraints is provided in Table 2. As expected, the backbone dihedral angles and predominantly occupy coil regions of the Ramachandran plot (56). The flexible N-and C-terminal regions corresponding to residues Val 10 -Ala 15 and Ser 33 -Lys 43 , respectively, are poorly defined due to the inherent dynamics in this region, as evidenced by the NMR relaxation experiments (Fig. 4). The central region ranging from Gly 16 to Thr 32 accounted for almost two-thirds (329 NOEs) of the NOE restraints and converged during structure calculation, revealing key aspects of LD binding. The predominantly hydrophobic and aromatic residues Ser 19 -Cys 30 are immersed in the LD mimicking DPC micelles and constitute the LD anchor (Fig. 5). The left arm of this anchor comprises tryptophans 21 and 25 along with the leucines 22 and 26. These hydrophobic residues form a compact core along with a short helix between Gly 20 and Gly 24 (Fig. 5, B and C). A network of NOEs between tryptophan NH⑀-1 and NH␦-1 protons and the Leu␦-protons exemplifies these interactions (Fig. 6). Pro 27 isolates Trp 29 from the other tryptophans and together with Pro 31 prevents the formation of a longer and more stable helix. The residue pairs Gly 18 /Ser 19 and Cys 30 /Pro 31 mark the interface between the DPC micelles and the solvent (Fig. 5C). A representation of the electrostatic potential on the solvent accessible surface of the peptide reveals the highly polar nature of the terminal segments and a predominantly hydrophobic LD anchor (Fig. 5, B and D).

Discussion
In this study, we describe the three-dimensional solution structure of the N-terminal LD anchor of CGI-58. Three anchor points (Trp 21 , Trp 25 , and Trp 29 ) act synergistically to tether CGI-58 stably to LDs. The peptide sequence immersed    30 . We demonstrate that a slightly longer CGI-58 sequence stretching from Ser 19 -Ser 35 also recruits the otherwise cytosolic yellow fluorescent protein to LDs. However, this LD anchor lacks the ability to activate ATGL, indicating that other regions of CGI-58 are necessary for ATGL activation. The data presented here also corroborate earlier studies that LD binding of CGI-58 is a strict requirement for ATGL activation (15). Single amino acid mutagenesis of any of the three tryptophans of the CGI-58 LD anchor and a variant lacking two tryptophan side chains (W21A/W25A) had no effect on CGI-58 LD co-localization. The ability of the CGI-58 variants to activate ATGL was reduced at most by one-third. In contrast, when we tested a W21A/W29A variant, the ability of CGI-58 to localize to LDs and to activate ATGL was completely abrogated. The unstructured nature of the LD anchor enables conformational flexibility and functional promiscuity. Therefore, it is conceivable that the CGI-58 LD anchor undergoes a conformational change upon CGI-58 binding to ATGL. The LD anchor might be necessary for the correct orientation of CGI-58 on LDs, which provides the platform for interaction with ATGL, or for positioning TGs favorably with respect to CGI-58 bound ATGL. Alternatively, CGI-58 might serve as mediator to transfer released FFAs from LD-bound ATGL to the water-soluble and cytosolic fatty acid-binding proteins, yet the LD anchor motif of CGI-58 is not required for binding to fatty acid-binding proteins or ATGL (15,16). Again, the correct positioning of CGI-58 with respect to all interaction partners could be real-ized via the LD-anchor. Obviously, additional structural data of the involved binary and ternary complexes are required to learn more about this complex network. Hydrophobic residues anchoring the peptide in the LD mimicking DPC micelle form a hydrophobic arc along the interface with the micelle. They are depicted as sticks and colored by atom. C, V10-K43 is shown immersed into the DPC micelle. Ser 19 and Cys 30 delineate the interface between the micelle and the solvent. Three tryptophan and two leucine side chains fix the peptide in the micelle. D, representation of electrostatic potential on the surface of the peptide V10-K43 positioned in the DPC micelle. Solvent exposed stretches carry partial charges. The surface inside the micelle is increasingly hydrophobic as it approaches the micelle center. The structures of membrane proteins have become a prominent and rapidly expanding field of structural biology, due to novel experimental and methodological breakthroughs. Structural studies of proteins acting at the water-lipid or membrane interface pose a tremendous experimental challenge. Consequently, the interaction of proteins with LDs remains largely uncharted territory. CGI-58 was initially demonstrated to bind to LDs and perilipin 1 simultaneously in adipocytes. Upon activation of lipolysis, CGI-58 dissociates from perilipin 1 and forms a LD bound complex with ATGL (57). Perilipin 1 independent binding of CGI-58 to LDs was demonstrated in COS-7 cells (15,58). Cell types that do not express perilipin 1 often express other members of the perilipin family (perilipins 2-5). Perilipin 5 interacts with CGI-58 and ATGL in a mutually exclusive manner (59) and the interactions of CGI-58 with perilipins 2 and 3 appear to be functionally less significant (60).
Only few structures of proteins that interact with LDs have been solved. Previously, Dunne and colleagues (61) showed the NMR structure of two CTP:phosphocholine cytidylyltransferase peptides in atomic detail on two overlapping 33-and 22-residue peptides, which span most of the amphipathic predominantly ␣-helical membrane-binding domain of rat CTP: phosphocholine cytidylyltransferase. The structural work on these peptides was performed in a membrane-binding context, yet Drosophila orthologues of CTP:phosphocholine cytidylyltransferase have been shown to localize to LDs as well (62,63). Additionally, structures of a few soluble orthologues or domains of LD-binding proteins have been characterized (64,65). The soluble C-terminal domain of the patatin family member TIP47 (perilipin 3) was solved nearly a decade ago (66), but the structure of the LD binding domain remains elusive. Intriguingly, the crystal structure of human monoacylglycerol lipase has been determined (67-69) and recently Nasr and colleagues (70) demonstrated that the cap region of human monoacylglycerol lipase interacts with nanodisc phospholipid bilayers, anchoring human monoacylglycerol lipase in a membrane-associated conformation. This leads to significantly increased V max and decreased K m values for the substrates arachidonoyl 7-hydroxy-6-methoxy-4-methylcoumarin ester and 2-arachidonoylglycerol (70). In a recent study, a combination of deuterium exchange experiments and molecular dynamics simulations suggests the membrane interface to act as allosteric activator of phospholipases, inducing conformational changes from an inactive to an active conformation (71). Clearly, unveiling the structure-function relationship of proteins associated directly with LDs and membranes will provide important mechanistic insight into protein-lipid interactions essential for different physiological processes including membrane remodeling, lipid signaling, and intracellular lipolysis.
Peptides that anchor proteins to membranes are often transmembrane ␣-helices or ␣-helices, which orient perpendicularly to the membrane. The mechanism by which peptides anchor proteins to LDs remains unknown, but it is clear that a trans-LD ␣-helix is not feasible, as LDs assume diameters of 0.1 to 100 m (72). The CGI-58 LD anchor observed here is not solely an ␣-helix that is perpendicular to the membrane. Within residues Ser 19 to Leu 26 of the LD-binding motif, Gly 20 to Gly 24 form a helix and the hydrophobic residues Trp 21 , Leu 22 , Trp 25 , and Leu 26 establish a compact hydrophobic core, representing the left arm of the LD anchor. The conserved CGI-58 prolines Pro 27 and Pro 31 impede the formation of a longer continuous helix. The third tryptophan resides between these two prolines and establishes the structurally independent right arm of the LD anchor. As a consequence, the substitution of tryptophans Trp 21 and Trp 25 with alanines does not abrogate CGI-58 function, because the right anchor arm comprising Trp 29 remains intact. On the other hand, replacement of Trp 21 and Trp 29 abrogates the function of CGI-58, because both anchor arms are affected by the substitutions. The importance of correct LD anchoring of CGI-58 is highlighted by conservation of the sequence of almost the entire LD anchor among vertebrates (3). The presence of two independently acting arms of the LD anchor could prevent defective anchoring as a result of malfunction in one LD anchor arm.
Selective inhibition of LD anchoring by CGI-58, or CGI-58 interactions with fatty acid-binding proteins, ATGL or perilipins, presents an opportunity for selective therapeutic targeting of lipid metabolism and peroxisome proliferator-activated receptor regulated gene expression. An essentially identical sequence to the LD anchor motif of CGI-58 can be found at the N terminus of a close relative of CGI-58, ␣/␤ hydrolase 4 (ABHD4) (55% sequence identity). Unlike CGI-58, ABHD4 is an active serine hydrolase and has been shown to hydrolyze N-acyl phosphatidylethanolamines (NAPEs) and lyso-N-acyl phosphatidylethanolamines (73). Moreover, ABHD4 was recently demonstrated to be a regulator of multiple classes of N-acyl phospholipids in the mammalian nervous system (74). From a drug design perspective the highly similar LD anchor of CGI-58 and ABHD4 can be both an opportunity and a limitation. Molecules that hinder CGI-58 LD anchoring may potentially affect ABHD4 activity as well. This would limit the specificity of an approach to target the LD anchor of CGI-58, but might also present a chance to trigger synergistic effects by influencing the activities of both proteins.
In summary, we show the structure of a LD anchor and describe the mechanism by which it binds to LDs using DPC micelles as LD mimics. The utilization of two independent LD binding arms reveals an intriguing strategy to protect CGI-58 against the loss of LD binding activity and highlights the importance of proper CGI-58 binding to LDs.