Critical residues and motifs for homodimerization of the first transmembrane domain of the plasma membrane glycoprotein CD36

The plasma transmembrane (TM) glycoprotein CD36 is critically involved in many essential signaling processes, especially the binding/uptake of long-chain fatty acids and oxidized low-density lipoproteins. The association of CD36 potentially activates cytosolic protein tyrosine kinases that are thought to associate with the C-terminal cytoplasmic tail of CD36. To understand the mechanisms by which CD36 mediates ligand binding and signal transduction, we have characterized the homo-oligomeric interaction of CD36 TM domains in membrane environments and with molecular dynamics (MD) simulations. Analysis of pyrene- and coumarin-labeled TM1 peptides in SDS by FRET confirmed the homodimerization of the CD36 TM1 peptide. Homodimerization assays of CD36 TM domains with the TOXCAT technique showed that its first TM (TM1) domain, but not the second TM (TM2) domain, could homodimerize in a cell membrane. Small-residue, site-specific mutation scanning revealed that the CD36 TM1 dimerization is mediated by the conserved small residues Gly12, Gly16, Ala20, and Gly23. Furthermore, molecular dynamics (MD) simulation studies demonstrated that CD36 TM1 exhibited a switching dimerization with two right-handed packing modes driven by the 12GXXXGXXXA20 and 20AXXG23 motifs, and the mutational effect of G16I and G23I revealed these representative conformations of CD36 TM1. This packing switch pattern of CD36 TM1 homodimer was further examined and confirmed by FRET analysis of monobromobimane (mBBr)-labeled CD36 TM1 peptides. Overall, this work provides a structural basis for understanding the role of TM association in regulating signal transduction via CD36.

thrombosis, atherosclerosis, cellular adhesion, and lipid transport (1)(2)(3)(4)(5). Widely distributed in a variety of cells, CD36 expression is prominent on platelets and capillary epithelial cells where it is thought to take part in transmembrane signaling and in regulating vessel angiogenesis (6,7). As a scavenger receptor, CD36 is involved in the binding and uptake of oxidized lowdensity lipoprotein by macrophages and the formation of foam cells during arterial atherogenesis (8 -11). As the receptor for thrombospondin 1 (TSP-1) 3 on endothelial cells, CD36 is reported to mediate the anti-angiogenic effect of TSP-1 (12,13). CD36 also serves as a selective and non-redundant receptor or sensor of microbial diacylglycerides that signal via the TLR2/6 heterodimer (2).
CD36 is the canonical member of class B scavenger receptor family. In this protein family, all the members conform to a two-TM membrane topology, with the first and the second TM domains flanking a large extracellular loop that is highly N-glycosylated (see Fig. 1A). Similar to CD36, many members of the class B scavenger receptor family share a common functional feature in recognizing and binding hydrophobic molecules such as fatty acids and pheromones (14,15). The extracellular domain of CD36 is characterized by a hydrophobic stretch (residues 184 -204) that may interact with the plasma membrane (16). The binding sites of CD36 for fatty acids, modified LDL, the growth hormone releasing peptide, and hexarelin have been mapped to residues 155-183 (17), whereas that for TSP-1 was shown to be residues 93-120 (18). Both intracellular domains of CD36 are relatively short but important for its efficient expression in the plasma membrane (19). Moreover, there are two cysteine residues in each intracellular domain, and their palmitoylation is thought to be an important factor in positioning CD36 into caveolae and lipid rafts (20 -22).
CD36 forms homodimers or higher order oligomers in cultured cells and tissues, which may contribute to its ligandbinding and signaling functions (23). However, the underlying molecular mechanisms of signal transduction of CD36 have . The authors declare that they have no conflicts of interest with the contents of this article. This article contains supplemental text and Figs. S1-S4. 1 Both authors contributed equally to this work. remained elusive. A recent report showed that the first transmembrane domain (TM1) of scavenger receptor class B, type I (SR-BI), a homolog of CD36, is critically involved in its oligomerization and function (24). Small-residue motif 15 GXXGXXXAXXG 25 in TM1 of SR-BI was identified to be critical to receptor homo-oligomerization and ligand binding via mutagenesis analysis. Furthermore, 18 GXXXAXXG 25 in TM1 of SR-BI significantly contributes to the homo-oligomerization and lipid uptake activity of SR-BI but has no influence on the negative cooperativity of HDL binding. Like SR-BI, TM1 of CD36 contains several conserved small residues (see Fig. 1B).
Therefore, it addresses great importance to identify the oligomerization mechanism of CD36 in this paper. First, FRET assay was adopted as in vitro method to test whether CD36 TM1 and TM2 could form homo-oligomers and identified the oligomeric state; TOXCAT assay was then adopted as an in vivo method to perform small-residue scanning to reveal the potential oligomerization motif in CD36 TM1 sequence; MD simulations were subsequently used as computational method to investigate the assembly process of CD36 TM1 dimerization in detail; finally, FRET assay was used to confirm two packing modes of CD36 TM1 homodimer revealed in the MD simulations. Correlation of computational studies with experimental results on the CD36 TM1 domain may provide insights for understanding the role of the association of TM domains on the signal transduction via CD36.

Results
Sequence alignment of CD36 TM domains from diverse species reveals that CD36 TM domains are conserved. The first TM domain of human CD36, denoted here as TM1, shares 96, 96, 100, 100, 100, 83, and 65% sequence identity with monkey, rat, mouse, cattle, pig, chicken, and tropicalis, respectively (Fig.  1B). In comparison, the sequence of the second TM domain, or TM2, is less conserved. Many conserved glycine residues were found in the TM1 domain of CD36, fitting the GXXXG motif, which is a common sequence motif that mediates association of TM helices (25,26). Separated by three residues in the primary sequence, the two Gly residues are spatially next to each other in an ␣-helix, permitting close approach of interacting helices and facilitating extensive packing interactions between pairs of surrounding residues (27). Many TM domains containing the GXXXG or GXXXG-like motif have been shown to form dimers (28 -30).

Characterization of the CD36 TM1 homodimer in SDS detergent
To test whether the transmembrane domain of CD36 forms an oligomer, unlabeled CD36 peptides TM1 and TM2 were synthesized and characterized (Table 1). Additional Lys residues were added to both ends to increase the solubility of these peptides. Using CD spectroscopy, we confirmed that SDS and dodecylphosphocholine (DPC) detergent micelles support the ␣-helical secondary structure of both unlabeled peptides, whereas n-decyl-␤-D-maltopyranoside (DM) does not (supplemental Fig. S1). SDS-PAGE was further applied in FRET experiments because these peptides have the most ␣-helical conformation in SDS. As shown in supplemental Fig. S2, two distinct bands, likely corresponding to the monomer and dimer bands, were observed for the unlabeled TM1. In contrast, only one band was visible for the unlabeled TM2. Unlabeled G16I, corresponding to the key residue Gly 18 for dimerization of TM1 of SR-BI, was examined to be monomer as well. These results indicated that TM1 of CD36, but not TM2, has the ability to form dimers in SDS, in which residue Gly 16 plays a critical role.
FRET is a powerful and sensitive tool to monitor the change in the distance between interacting proteins labeled with donor and acceptor groups (31,32). To characterize the kinetic and thermodynamic properties of oligomeric CD36 TM1 peptide, we adopted the FRET experiment with fluorophores pyrene and coumarin as the donor/acceptor pair. The relatively large Förster radius (R 0 ) (60 Ϯ 5 Å) of Cou/Pyr pair ensures that the FRET signal between them should be independent of the details of TM peptide complex geometry (33). All peptides were mixed with SDS based on the results of CD and SDS-PAGE of CD36 TM domains, either together or separately as needed, at a molar ratio of 1:1000. Fig. 2A showed the coumarin excitation spectra of 300 -450 nm, with emission measurement at 500 nm. Little emission was observed for 2 M Pyr-TM1, whereas standard emission with a maximum at 374 nm was observed for 2 M Cou-TM1. When 2 M Pyr-TM1 was mixed with 2 M Cou-TM1, Cou excitation at 374 nm did not change, but considerable additional Cou excitation was observed at 344 nm. The local maxima of 344 nm corresponded to that in Pyr absorption (34), indicating that the increased Cou excitation was due to FRET from Pyr-TM1. To characterize the oligomeric state of the TM1 complex, the ratio between Pyr-and Cou-TM1 was systematically varied, whereas the overall conjugated TM1 concentration was kept constant as described in other studies on membrane peptide complexes  (32,34,35). The efficiency of energy transfer in each mixture was measured and plotted against the mole fraction of Cou-TM1. As illustrated in Fig. 2B, the plot could be fitted to a straight line (R 2 ϭ 0.987), demonstrating that the TM1 complex in the SDS is a dimer.
Unlabeled TM peptides were added at equal molar concentration to the mixture of Pyr-and Cou-TM1 to confirm the specific association between them. Fig. 2C showed that the addition of an equal amount of unlabeled TM1 resulted in the loss of half of the pyrene-sensitized coumarin excitation signal at 344 nm, because unlabeled TM1 was expected to compete with labeled peptide to form the dimer and thus reduce the FRET signal. In comparison, the fact that the addition of equal molar TM2 to the mixture had no effect on the pyrenesensitized coumarin excitation revealed that TM2 neither interferes the homodimerization of TM1 nor interacts with TM1 to form heterodimer. Moreover, addition of equal molar unlabeled G16I to the mixture did not significantly change the Cou excitation spectrum (Fig. 2D), demonstrating that residue Gly 16 is crucial for homodimerization of the CD36 TM1 domain.

CD36 TM1 dimerized in a cell membrane
TOXCAT assay is widely used to quantitatively monitor selfassociation of TM helices in a cell membrane environment (36).
To assess the dimerization of CD36 TM domains in a natural cell membrane, we have inserted the CD36 TM1 and TM2 domains separately into the TOXCAT construct ( Table 1). The TM domains of glycophorin A (GpA-WT) and its mutant GpA-G83I that disrupts GpA dimerization were included in the experiment to serve as positive and negative controls, respectively. Comparison of chloramphenicol acetyltransferase (CAT) activities indicated that the CD36 TM1 could dimerize in the cell membrane, to the extent ϳ70% of that of the strongly dimerizing GpA-WT (Fig. 3A). In contrast, the CAT activity of CD36 TM2-containing construct was similar to that of the negative control GpA-G83I, suggesting that the CD36 TM2 does not dimerize significantly. To rule out the affecting variables, Western blots of all chimeras were applied to show the similar expression levels (Fig. 3A), and MalE complementary assay was used to confirm the correct transmembrane topology because they were all able to grow on a minimal medium with maltose as the only carbon source (Fig. 3B).
Gaidukov et al. (24) recently identified a conserved smallresidue motif 18 GXXXAXXG 25 , which significantly contributes to the homodimerization of SR-BI. This motif corresponds to 16 GXXXAXXG 23 sequence in CD36 TM1. To assess the role of small residues in the TM1 domain in mediating its dimerization, site-specific mutation scanning was introduced to the chi- Table 1 Sequences of CD36 TM1, TM2, and their mutants Two packing models switch in CD36 TM1 homodimer mera containing CD36 TM1. As shown in Fig. 3A, mutating residues Gly 12 , Gly 16 , and Ala 20 to Ile all significantly reduced the CAT activity of TM1, whereas G16I mutations produced the largest disruptive effect. Furthermore, mutating Gly 16 to Ala had the same disruptive effect as the G16I mutation, demonstrating the absolute requirement of this crucial glycine residue in TM1 homodimerization. The CAT activity showed that G23I in CD36 TM1 corresponding to Gly 25 in SR-BI moderately disrupted the dimerization, whereas Gly 24 in CD36 TM1 corresponding to Gly 26 in SR-BI did not affect the dimerization. Overall, these results revealed that the small residues Gly 12 , Gly 16 , Ala 20 , and Gly 23 is critical to CD36 TM1 homodimerization.

A packing switch pattern of CD36 TM1 homodimer in a DPPC bilayer revealed by coarse-grained (CG) simulations and all-atom (AT) simulations
CG modeling is extensively used as a powerful tool to investigate the structure and dynamics of various membrane proteins or their complexes (37)(38)(39)(40). To better understand the homodimerization of CD36 TM1 at molecular level, we adopted our well-established DPPC bilayer system (41)(42)(43) to perform the self-assembly process of CD36 TM1 and its mutations by CG-MD. The initial structures of two CD36 TM1 helices were positioned parallelly into a performed DPPC bilayer with a distance of 55 Å. This long-distance separation between the TM helices makes sure that the initial TMs form homodimer only because of the helix-helix interaction instead of artificial dimer because of the starting close contact. After the simulation began, two CD36 TM1 helices diffused randomly in this DPPC bilayer until they encountered one another and interacted to form a stable homodimer (supplemental Fig. S3). Five independent samples were carried out with the same initial configuration but different seeds of random velocities. As shown in Fig. 4A, the trajectories of the backbone distance of CD36 TM1 of all five samples were distinct. CD36 TM1 selfassembled rapidly to homodimerize in DPPC bilayer, and the resulting dimer was long-lasting throughout the rest of the simulation. Crossing angle distribution and helix-helix contacts were further examined in more depth. Homodimer of CD36 TM1 exhibited two major right-handed packing models with the corresponding crossing angles of Ϫ23°(RH1) and Ϫ35°( RH2); the representative CG structures for RH1 and RH2 are shown as well (Fig. 4B). Furthermore, the residue contacts of these two alternative conformations were explored with a cutoff of 1 nm. The critical contacts for both conformations were circled out. Residues Gly 12 , Gly 16 , and Ala 20 , which located near the N-terminal region, were mostly accounted for by the RH1 packing model (Fig. 4D), whereas Ala 20 and Gly 23 accounted for the RH2 packing model (Fig. 4C).
To improve these two packing models of CD36 TM1 homodimer and check on the stability, the CG structures of the RH1 and RH2 models were converted to full atomistic models followed by a fragment-based procedure (44). AT simulations of RH1 and RH2 were each performed for 50 ns with three independent samples as previous described (41)(42)(43). C␣ RMSD analysis showed that RH1 and RH2 models were both relatively stable because of the mild drift (Fig. 5, A and B), whereas the RH1 model was more stable than RH2 with a lower RMSD. Furthermore, the representative structures of RH1 and RH2 were distinct in the N-terminal distance as shown in Fig. 5C. Therefore, we proposed that two packing models, RH1 and RH2 of the CD36 TM1 homodimer, were dominated by 12 GXXXGXXXA 20 motif and 20 AXXG 23 motif, respectively.
To obtain a better understanding of the intermolecular mechanisms of CD36 TM1 homodimer, CG simulations of mutations (TM1-G12I, TM1-G16I, TM1-A20I, and TM1-G23I) were subsequently performed as described for CD36 TM1. Backbone distance analysis showed that these mutants have disruptive effects on CD36 TM1 homodimer because the dimer disassembled during the simulations (supplemental Fig.  S4), especially for TM1-G12I, TM1-G16I, and TM1-A20I. These results were significantly consistent with the previous TOXCAT data, which demonstrated that Gly 12 , Gly 16 , and Ala 20 residues have the most disruptive effect on dimerization of CD36 TM1. Interestingly, mutants TM1-G16I and TM1-G23I resulted in a significant shift in crossing angle distribution (Fig. 6). We noted that mutants TM1-G16I and TM1-G23I have a dominant preference for RH2 and RH1 packing, respectively. Therefore, mutants TM1-G16I and TM1-G23I were subsequently investigated as the representative RH2 and RH1 models to further compare these two packing patterns of CD36 TM1 homodimer.

Two packing patterns of CD36 TM1 homodimer confirmed by FRET
As observed in both CG and AT simulations of CD36 TM1, RH1 and RH2 models were dramatically different in N-terminal distance. Therefore, the N-terminal distances of RH1 and RH2 were plotted as a function of time for the last 1 s in one random sample in Fig. 7A. The plotting trends were relative stable, and the N-terminal distance of RH1 was ϳ7 Å closer than that of RH2. Then the trajectories for the last 1 s of all five samples were concatenated and analyzed by frequency count. The frequency distribution of N-terminal distance of RH1 model was concentrated on 0.95 nm, which is much closer than that of 1.65 nm for RH2 (Fig. 7B).
To further confirm the difference between these two packing patterns we achieved in MD simulations, we applied a trypto-  A, the evolution of backbone distance between CD36 TM1 helices for all five samples. B, crossing angle histogram of CD36 TM1 homodimer, and the representative structures of RH1 and RH2 were shown. C and D, interhelical contact matrices of two packing models of CD36 TM1 homodimer calculated using a 1-nm distance cutoff. Contact distance ranged from 0 nm (blue) to 1 nm (red). The key interhelical contacts were circled out.

Two packing models switch in CD36 TM1 homodimer
phan-induced quenching FRET method (45)(46)(47) to probe the N-terminal distances of both RH1(TM1-G23I) and RH2 (TM1-G16I) models. This method provides a straightforward and sensitive way to address questions of conformational dynamics in proteins by mBBr labeling. The integrated fluorescence intensities with (F w ) and without (F 0 ) the presence of the tryptophan in the peptides were used to calculate the degree of fluorescence quenching. The F 0 /F w ratios represent C␣-C␣ distances of the proteins.
Four peptides (mBBr-G16I, W-G16I, mBBr-G23I, and W-G23I) were synthesized as shown in Table 1. In Fig. 7C, integrated fluorescence intensities (F 0 and F w ) of the mutant G16I and G23I were shown respectively. For the RH2 model, F 0 /F w equaled 1 after adding W-G16I in mBBr-G16I, no fluorescence quenching occurs, whereas for the RH1 model, F 0 /F w equaled 1.3 after adding W-G23I in mBBr-G23I, indicating that the C␣-C␣ distance is ϳ11 Å by calculation (46). Because the applicable C␣-C␣ distance range is 5-15 Å for tryptophan/mBBr FRET, these results were significantly consistent with the results of CG simulations (Fig. 7B). Thus, we have confirmed the RH1 and RH2 packing models of CD36 TM1 homodimer, which were revealed by our MD studies. We propose that there are two packing models coexisting in homodimerization of CD36 TM1, and the RH1 model with a 12 GXXXGXXXA 20 motif is more favorable than the RH2 model with a 20 AXXG 23 motif.

Discussion
In this paper, we have characterized the oligomerization of CD36 TM domains. FRET analysis of CD36 TM1 revealed that it forms an oligomer in SDS (Fig. 2A). The oligomeric state of the CD36 TM1 complex was determined by the efficiency of energy transfer. It was shown to be fitted to a straight line (R 2 ϭ 0.987), which indicated that the CD36 TM1 oligomer in the SDS is a dimer (Fig. 2B). We also demonstrated that CD36 TM2 did not interfere the homodimerization of CD36 TM1 by forming heterodimer (Fig. 2C). Moreover, mutation G16I of the TM1 domain significantly abolished its ability to self-associate in the FRET assay as well (Fig. 2D). The TOXCAT assay showed that the TM1 domain, but not the TM2 domain, could selfassociate in a cell membrane (Fig. 3A). The CAT activity induced by TM1 dimerization was ϳ70% of that induced by  GpA-WT. Consistent with this result, unlabeled TM1 peptide, but not unlabeled TM2 peptide, can form two bands in SDS-PAGE, one of which corresponds to the molecular weight of CD36 TM1 homodimer (supplemental Fig. S2). Site-specific mutation scanning of small residues in CD36 TM1 has provided direct evidence of its homodimerization, which was mediated by Gly 12 , Gly 16 , Ala 20 , and Gly 23 residues.
By applying CG simulations, we have revealed two distinct packing models (RH1 and RH2) of CD36 TM1 homodimer in a DPPC bilayer (Fig. 4). The packing mechanism of CD36 TM1 homodimerization was further investigated by mutation studies. Combined with AT simulations and TOXCAT data, we found that RH1 model was more stable than RH2 model (Fig. 5). Results of crossing angle distributions, backbone distances, and residue contacts all showed that the Gly 12 , Gly 16 , Ala 20 , and Gly 23 residues in CD36 TM1 were accounted for by the homodimerization (Fig. 6 and supplemental Fig. S4), and two packing models RH1 and RH2 of CD36 TM1 homodimer were dominated by 12 GXXXGXXXA 20 motif and 20 AXXG 23 motif, respectively. We have further confirmed the major difference between RH1 and RH2 packing models by mBBR/Tryptophan FRET (Fig. 7), indicating that two packing models coexist in homodimerization of CD36 TM1, and the RH1 model is more favorable than the RH2 model.
Our results are consistent with the recent study that identified a glycine-containing motif in the TM1 domain of SR-BI as important in its homo-oligomerization and lipid uptake. In this article, we showed that Gly 12 , Gly 16 , Ala 20 , and Gly 23 residues in CD36 are critical to CD36 TM1 homodimerization. The CD36 TM2 domain does contain a conserved polar residue instead of a GXXXG sequence (Fig. 1B). Although polar residues can mediate oligomerization of TM helices (48 -51), the CD36 TM2 domain failed to induce significant CAT activity in the TOXCAT assay or form a dimeric band in SDS, suggesting that it may not oligomerize. However, the SR-BI TM2 domain, which similarly contains a conserved polar residue, and its surrounding sequences have been implicated in mediating oligomerization of SR-BI in membranes of cultured cells (14). Responsible residues in the region have not been identified by mutagenesis studies. The functional role of the conserved polar residue in the CD36 TM2 domain still remains to be determined.

Two packing models switch in CD36 TM1 homodimer
In summary, our study has demonstrated that the CD36 TM1 domain can form a homodimer in detergent micelles, cell membranes, and lipid bilayers. The 12 GXXXGXXXAXXG 23 sequence plays a critical role in mediating the homodimerization of CD36 TM1 domain, and two distinct packing models (RH1 and RH2) for such dimerizing capability may contribute to CD36 clustering in the membrane microdomains, its interaction with cytoskeleton, and CD36-mediated signaling transduction across the cell membrane.
There are two crucial cysteine residues (Cys 3 and Cys 7 ) in the intracellular domain close to the N-terminal region of CD36 TM1 (20 -22), and their palmitoylation is considered to be an important factor in positioning CD36 in caveolae and lipid rafts (2,52,53). We have proposed a schemed representation to interpret how CD36 TM1 homodimerization may be involved in lipid raft (Fig. 8). RH1 model is dominated by the 12 GXXXGXXXA 20 motif, and the N-terminal region of CD36 TM1 homodimer was closed, leaving the intracellular cysteines in close distances, whereas the RH2 model is dominated by the 20 AXXG 23 motif, and the N-terminal region of CD36 TM1 homodimer was open, leaving the intracellular cysteines in far distances. The conformational change of CD36 TM1 homodimer thus may affect the subsequent recruitment in lipid raft, signal transduction, and LDL uptake of CD36.

Solid-phase peptide synthesis and purification
All peptides were synthesized via Fmoc (N-(9-fluorenyl)methoxycarbonyl) solid-phase peptide synthesis ( Table 1). The coupling conditions with the HATU/DIEA activation pair was used with a 4-fold excess amino acid. Double couplings were applied, if necessary, on hydrophobic residues. Cleavage reaction was carried out in 82.5% (v/v) TFA, 5% water, 5% thioani-sole, 5% phenol, and 2.5% dimercaptoethane. After cleavage, the peptide was precipitated and washed with 30 ml of cold diethyl ether four times, then dried with nitrogen gas and a vacuum dryer, and finally purified by reverse-phase HPLC. The identities of the purified peptides were confirmed by MALDI-TOF. Removal of TFA and other organic solvents was performed by a rotary evaporator, and the dry powder of these peptides were obtained by a vacuum freeze dryer.

Fluorescence labeling
To label the peptide with fluorophore during the synthesis, the N-terminal amino group of the resin-bound peptide was activated by HATU and hydroxybenzotriazole (HOBt) at a 10-fold excess of either Cou or Pyr under basic conditions. The reaction was allowed to proceed overnight in the dark. mBBr label was conjugated to peptides with an added cysteine in the N terminus and then treated with a 10-fold molar excess of monobromobimane in methanol at 25°C overnight in the dark. The cleavage was carried out as described above, and the labeled peptides were purified by reverse-phase HPLC as well.

Fluorescence spectroscopy
Fluorescence spectra were obtained using a Hitachi F-4500 fluorescence spectrometer. As to Cou/Pyr FRET, the concentrations of the labeled peptides were 2 M as determined by UV absorption using extinction coefficients of 32,000 (Ϯ2600) M Ϫ1 cm Ϫ1 at 344 nm for Pyr and 16,500 (Ϯ3200) M Ϫ1 cm Ϫ1 at 387 nm for Cou. Mixing with SDS at a molar ratio of 1:1000, the peptide/SDS and dye were prepared in 50 mM sodium phosphate, 10 mM NaCl, pH 7.5. The 1-cm cuvette with a minimal volume of 600 l was used. The excitation wavelength was set to 500 nm, and the emission spectra were collected as averages of three scans. All spectra were baseline-corrected by using Two packing models switch in CD36 TM1 homodimer control samples without peptide. The efficiency of energy transfer (E) is defined with respect to r and R 0 , the characteristic Forster radius, by E ϭ 1/[1 ϩ (r/R 0 )6]. As to mBBR/tryptophan FRET, the peptides concentrations were 2 M. Mixing with SDS at a molar ratio of 1:2000, the peptide/SDS and dye were prepared in 50 mM sodium phosphate, 10 mM NaCl, pH 7.5. The fluorescence emission spectra were measured from 395 to 600 nm (1-nm slits) while being excited at 381 nm (2-nm slits).

TOXCAT assay
Gene fragments encoding CD36 TM domain residues were amplified by PCR from human CD36 cDNA and inserted into the pccKAN plasmid (kindly provided by Dr. D. M. Engelman). The resulting pccKAN-based plasmids containing CD36 TMs and mutants were transformed into MM39 cells. The proper membraneinsertionandorientationoftheToxR-TM-MBPconstructs was checked with the maltose complementation tests. The expression levels of the fusion proteins were confirmed by Western blot against MBP. The activity of CAT was measured as previous described (36).

CG simulations
All CG-MD simulations were carried out by GROMACS version 4.5.3 (54,55). The MARTINI force field was applied for proteins, lipids, and water (56,57). The initial models of the CD36 TM1 and its mutants were constructed by PyMOL (58,59). Two identical helix structures were parallelly inserted into a pre-equilibrated DPPC lipid bilayer with a distance of 55 Å. The simulation box is ϳ8 ϫ 8 ϫ 12 nm, containing 186 DPPC molecules and 4000 CG waters. The N and C termini of the peptide were not acetylated or amidated. Lennard-Jones interactions were shifted to 0 between 9 and 12 Å, and electrostatics were shifted to 0 between 0 and 12 Å, with a relative dielectric constant of 15. The non-bonded neighbor list was updated every 10 steps with a cutoff of 14 Å. A Berendsen thermostat was used for temperature (323 K, coupling constant of 1 ps) and pressure (1 bar, coupling constant 5 ps, compressibility 4.5 ϫ 10 Ϫ5 bar Ϫ1 , semiisotropic coupling) (60). The integration step was 20 fs. These parameters followed from the recommendations of Marrink et al. (61) and Winger et al. (62). The energy of each system was minimized again and then position-restrained in a 3-ns simulation to allow for better packing of the lipid molecules around the TM helices. For each system, five replicas with different initial velocities were performed for 3 s.

AT simulations
AT models of CD36 TM1 were derived and converted from the representative CG models of RH1 and RH2 packing. Then the AT models were inserted into a pre-equilibrated atomistic lipid bilayer, containing 128 DPPC molecules. The Inflate GRO35 method was then adopted to set up the feasible system (63). AT model simulations were performed by the GRO-MOS96 53a6 force field (64). Water molecules defined as the SPC model were added. The detailed simulation parameters were adopted as previous described (41,42). For each system, three replicas with different initial velocities were performed for 50 ns. All trajectories were analyzed by GROMACS tools.
Visualization and graphics were generated with VMD software (65).