doi:10.1016/j.jmb.2005.10.031 J. Mol. Biol. (2006) 355, 224–236

Structural Evidence for Adaptive Ligand Binding of Glycolipid Transfer Protein

Tomi T. Airenne†, Heidi Kidron†, Yvonne Nymalm, Matts Nylund Gun West, Peter Mattjus and Tiina A. Salminen*

Department of Biochemistry Glycolipids participate in many important cellular processes and they are and Pharmacy, A˚ bo Akademi bound and transferred with high specificity by glycolipid transfer protein University, Tykisto¨katu 6A (GLTP). We have solved three different X-ray structures of bovine GLTP at FIN-20520 Turku, Finland 1.4 A˚ , 1.6 A˚ and 1.8 A˚ resolution, all with a bound fatty acid or glycolipid. The 1.4 A˚ structure resembles the recently characterized apo-form of the human GLTP but the other two structures represent an intermediate conformation of the apo-GLTPs and the human lactosylceramide-bound GLTP structure. These novel structures give insight into the mechanism of lipid binding and how GLTP may conformationally adapt to different lipids. Furthermore, based on the structural comparison of the GLTP structures and the three-dimensional models of the related Podospora anserina HET-C2 and Arabidopsis thaliana accelerated cell death protein, ACD11, we give structural explanations for their specific lipid binding properties. q 2005 Elsevier Ltd. All rights reserved. Keywords: crystal structure; homology modeling; conformational change; *Corresponding author cavity; fluorescence

Introduction to the diverse roles of glycolipids in the cell, GLTP could potentially function as a modulator or sensor Glycolipid transfer protein (GLTP) is a 24 kDa of glycolipid levels. Most glycosphingolipids are basic cytosolic protein, which has been identified synthesized at the luminal side of Golgi but gluco- from a variety of organisms and cell types.1,2 GLTPs sylceramide, which is the precursor of other glyco- from mammals share high sequence identity sphingolipids and the simplest member of the (w90%) and they have in vitro been shown to glycosphingolipid family, is synthesized on the specifically transfer glycolipids that have glycosyl cytosolic surface of the Golgi membrane6 and units attached to the lipid hydrocarbon backbone another monohexosylceramide, galactosylcera- (either ceramide or diacylglycerol) by b-linkage.3 mide, is synthesized at the cytosolic side of the Glycolipids, and particularly glycosphingolipids, endoplasmic reticulum. It has been suggested that are known to function in many important cellular GLTP could participate in the transfer of glucosyl- processes like development, adhesion and cell–cell ceramide from the cytosolic surface of the Golgi recognition both in eukaryotes and prokaryotes. membrane to the inner leaflet of the plasma The glycosphingolipids can function as cell surface membrane through the cytosol.7,8 markers and modulators of membrane protein GLTP has no sequence homology with other functions, e.g. as binding sites for certain bacteria, lipid-binding proteins found in the animal kingdom toxins and viruses and as stimulators of cell growth, but it shares sequence identity with two lipid- in differentiation and DNA synthesis, and in sorting binding proteins found in fungi and plants; the and trafficking of proteins.4,5 Very little is known HET-C2 protein from Podospora anserina and the about the specific function of GLTP in vivo, but due accelerated cell death protein, ACD11, from Arabi- dopsis thaliana. The inactivation of the het-c gene, which encodes the HET-C2 protein, leads to † T.T.A. & H.K. contributed equally to this work. 9 Abbreviations used: GLTP, glycolipid transfer protein; abnormal ascospore formation in P. anserina, rGLTP, recombinant GLTP; MIR, multiple isomorphous while the inactivation of the acd11 gene in A. thaliana replacement; NEM, N-ethylmaleimide. causes activation of programmed cell death and 10 E-mail address of the corresponding author: expression of defence-related genes. HET-C2 has tiina.salminen@abo.fi been shown to specifically bind glycosphingolipids

0022-2836/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. Lipi Binding of Glycolipid Transfer Protein 225 in vitro11 and ACD11 has been proposed to be significance of disulfide bonds for the glycolipid homologous to GLTP, but it does not bind and transfer activity of bovine GLTP was analyzed. transfer glycolipids, instead it has in vitro transfer activity of the single chain lipid sphinghosine.10 We have recently reported the preliminary X-ray Results analysis of GLTP12 and now we present the X-ray structureofbovineGLTPat1.6A˚ resolution X-ray structures of bovine GLTP (intermediate-GLTP) using the multiple isomor- phous replacement (MIR) method. In addition, we Glycolipid transfer proteins recognize a wide have solved two structures with the molecular range of different glycolipids3 and have a high replacement method at 1.4 A˚ (apo-GLTP) and 1.8 A˚ ganglioside transfer activity.14 Mammalian GLTPs resolution (GM3-GLTP) using the MIR structure as share high sequence identity; the amino acid an initial model. The apo-GLTP structure reveals a sequences of bovine and porcine GLTP are identical conformation similar to the apo-conformation of and human and bovine GLTP differ only by five human GLTP13 while the other two structures amino acid residues, none of which are involved in represent an intermediate conformation of the ligand binding (Figure 2). In order to investigate the apo-GLTP and lactosylceramide-bound GLTP mechanism of lipid binding to GLTP we have (LacCer-GLTP) (Figure 1).13 The apo-GLTP and the solved three different high-resolution X-ray intermediate-GLTP structures are bound to a fatty structures of bovine GLTP. All the bovine GLTP acid while the GM3-GLTP structure is bound to a structures have a similar fold to the human GLTP;13 double-chain glycolipid, ganglioside GM3. Based a unique two-layer all-a-helical topology with a on the structural analysis of the novel bovine GLTP sugar moiety binding pocket and a channel where structures and the known human GLTP structures hydrophobic ligands can bind (Figures 1 and 4). we suggest a mechanism for lipid recognition and Furthermore, a large positively charged area binding by GLTP. In order to provide structural formed by four lysine residues (Lys87, Lys137, explanations for the specific and distinctive lipid Lys138 and Lys208) is located on the surface of binding of GLTP and the GLTP-related HET-C2 and GLTP in the vicinity of the sugar-binding pocket ACD11 proteins, three-dimensional models of HET- (Figure 3(a)). The intermediate-GLTP (PDB code C2 and ACD11 protein were constructed using the 1TFJ) and GM3-GLTP (PDB code 2BV7) structures GLTP X-ray structures as templates. In addition, the represent a different conformation than either of the

Figure 1. A stereo view of X-ray structures of bovine GLTP. The superimposed bovine apo (red), intermediate (grey) and GM3-GLTP (blue) structures are shown as cartoons. Secondary structure elements (a-helices) are numbered according to the apo-GLTP structure. 226 Lipi Binding of Glycolipid Transfer Protein

Figure 2. A sequence alignment of 11 GLTP-like sequences. The amino acid numbering follows the bovine GLTP sequence. The a-helices defined by the bovine intermediate-GLTP structure are shown as cylinders. The residues conserved in all sequences are shaded with dark gray, while the residues that are conserved in GLTPs but not in ACD11 are shaded with light gray. Residues of particular interest are indicated in the alignment: lysine residues forming the membrane contact area in bovine GLTP (:), amino acids in the sugar-binding site (C) and amino acids forming the hydrophobic channel (B). The organisms and sequence accession numbers are referred to as follows: Bovine (P17403); Human (Q9NZD2); Mouse (Q9JL62); Drosophila (Drosopila melanogaster, Q9VXV1); P. anserina (Podospora anserina, Q01494); P. involutus (Paxillus involutus, Q5V8K7); FAPP2_Human (Q80W71); FAPP2_Mouse (Q80W71); A. thaliana (Arabidopsis thaliana, Q6NLQ3); C. elegans (Caenorhabditis elegans, Q9BKS2); ACD11 (Arabidopsis thaliana, O64587).

Figure 3. Surface properties of GM3-GLTP,HET-C2 and ACD11. The electrostatic potentials of (a) GM3-GLTP,(b) HET-C2 and (c) ACD11 are mapped to the surfaces. The four lysine residues (K87, K137, K138 and K208) forming a locally charged area on the surface of GM3-GLTP are indicated. Only one of these residues is conserved in HET-C2 (K123) and none in ACD11. Lipi Binding of Glycolipid Transfer Protein 227 recently reported human apo or the LacCer-GLTP the hydrophobic cavity. In the bovine GM3-GLTP, structures,13 whereas the apo-GLTP (PDB code the sphingosine chain does not fit into the narrower 1WBE) structure resembles the human apo-GLTP. channel but lies on the surface of the protein. Two of the solved bovine GLTP structures, the apo In all known GLTP structures the acyl chain and intermediate-GLTP, have a bound fatty acid cavity, which is located between the a1, a7 and a8 (Figure 4(a) and (b)), which is likely to be a decanoic helices, is structurally conserved. Small differences acid according to mass spectrometry analysis (data in the acyl chain cavities of human and bovine not shown) and originate from the bacterial GLTP structures are, however, seen probably due to expression of the protein used for crystallization. the crystal packing or different N termini of the Moreover, the intermediate-GLTP structure con- crystallized proteins. All crystallized bovine GLTP tained uninterpretable electron density near the structures reported here contained an N-terminal bound fatty acid within the hydrophobic cavity 6xHis-tag, whereas in the recently reported human (Figure 4(b)). The third bovine GLTP structure, GLTP structures13 the His-tag was proteolytically GM3-GLTP, contains a bound GM3 ligand cleaved off. In the bovine GLTP structures and the (Figure 4(c), the structure of GM3 is shown in human apo-GLTP structure the acyl chain cavity is Supplementary Figure 1), which was introduced open and hence accessible for solvent, while in the during the crystallization as ganglioside micelles human LacCer-GLTP structure the acyl chain cavity and co-crystallized with the protein. In our GM3- is partially blocked by the side-chain of the GLTP structure, electron density is observed only N-terminal Leu4. Interestingly, the acyl chain cavity for the first sugar unit (glucose), whereas the other of the bovine intermediate-GLTP extends through sugar units are not visible in the electron density the GLTP fold forming a tunnel through the maps, probably due to high thermal motion. The structure (Figure 4(b)). A similar extension of the sugar unit and the acyl chain of GM3-GLTP are at acyl chain cavity is also seen in the bovine GM3- the same positions and stabilized by similar GLTP structure (Figure 4(c)) but in the other hydrogen bonds and hydrophobic interactions as currently known GLTP structures the acyl chain in the human LacCer-GLTP complex structure extension is blocked (in the bovine apo-GLTP by (Figure 5(a)).13 The acyl chain fits very well into Phe42, in the human apo-GLTP by Phe42 and the hydrophobic channel, but electron density is not Leu152, and in the human LacCer-GLTP by observed for all of its carbon atoms (Figure 5(c)). Phe148) (Figures 4(a) and (d) and 5(a)). The eight first carbon atoms of the sphingosine Altogether, the differences in the cavity archi- chain, which could be traced, are located on the tecture of the GLTP structures are mainly achieved surface of GM3-GLTP structure unlike in the human by shifting the a2 and a6 helices and by changing LacCer-GLTP structure where the sphingosine the side-chain conformations of residues Phe42, chain is buried inside the protein (Figure 4(c), (d) Ile45, Phe148 and Leu152 that may act as switches and (e)). critical in controlling the ligand binding process of GLTPs. These structural changes are likely to occur Lipid binding to GLTP step by step, gradually widening the connecting channel between the acyl chain cavity and the A hydrophobic channel is located between the sugar-binding pocket during ligand binding. The two layers of helices that are characteristic for the intermediate-GLTP and the GM3-GLTP structures GLTP fold. The major difference between the known seem to represent transitional stages in the lipid GLTP structures is seen in the shape and size of this binding process of GLTP. They share structural channel, particularly at the location where the sugar features with both the human apo and LacCer- binding pocket is linked to the acyl chain binding structures: (1) the conformation of the loop between end of the channel (the acyl chain cavity) upon lipid helices a1 and a2 (loop a1-a2) in the intermediate- binding (Figures 4 and 5). In the apo-GLTP GLTP and GM3-GLTP structures resembles the structure as well as in the intermediate-GLTP human LacCer-GLTP conformation. (2) The N structure the connection between the acyl chain terminus of the a2 helix of both the intermediate- cavity and the sugar-binding pocket is blocked by GLTP and GM3-GLTP structures is bent compared to the side-chain of Ile45 from the a2 helix and Phe148 the apo-GLTP conformation but it is not in the same from the a6 helix. In GM3-GLTP, the side-chain of orientation as in the human LacCer-GLTP structure. Ile45 is shifted in such a way that a connecting (3) The a6 helix, which in the human LacCer-GLTP channel between the acyl chain cavity and the structure has shifted outwards with respect to the sugar-binding pocket is formed. The channel is a2 helix, is in the apo-GLTP conformation in the clearly narrower in the GM3-GLTP structure than in intermediate-GLTP and GM3-GLTP structures. the human LacCer-GLTP structure, which is mainly caused by the side-chain of Phe148. In GM3-GLTP, Structural models of HET-C2 and ACD11 the side-chain of Phe148, which is in the same position as in the apo-GLTPs (Figure 5(a)), partially Proteins that are homologous to the mammalian blocks the opening between the sugar-binding GLTPs can be found in many different species of pocket and the acyl chain cavity, while in the eukaryotes. Based on the multiple sequence align- human LacCer-GLTP structure Phe148 has turned ment, the known mammalian GLTPs are highly aside to accommodate the sphingosine chain within similar to each other, having sequence identities of 228 Lipi Binding of Glycolipid Transfer Protein

Figure 4. (Legend opposite) Lipi Binding of Glycolipid Transfer Protein 229 over 90% (Figure 2). On the other hand, the GLTPs correspond to Val42 and Phe134 in HET-C2 and to from Drosophila melanogaster, P. anserina, Paxillus Ala47 and Val141 in ACD11, respectively (Figure 5). involutus, A. thaliana and Caenorhabditis elegans The connecting channel is open in the model of apo- have low sequence identities (20–30%) to the ACD11, while in the apo-HET-C2 model (data not mammalian GLTPs. In mammalia, there is also a shown) the side-chains of Val42 and Phe134 block phosphatidylinositol-4-phosphate adaptor protein-2 the opening between the hydrophobic cavity and (FAPP2), which contains a C-terminal domain the sugar-binding pocket, similarly to apo-GLTP. with 30% sequence identity to GLTP, but its The positively charged surface area, formed by the biological function is unknown. These proteins lysine residues 87, 137, 138 and 208 in the bovine could be grouped as a glycolipid sensor protein GLTP, is not totally conserved in the other family, since they all likely recognize glyco- sequences (Figure 2). Three of the lysine residues lipids.7,11,15 Another protein, ACD11 from are conserved in mammalian GLTPs but only one A. thaliana, has been proposed to be homologous (Lys137) is conserved in HET-C2 and none in to GLTP,10 even though it has less than 20% ACD11. Consequently, there is no large positively sequence identity to the mammalian GLTPs. The charged area at the same location on the surface of glycolipid transfer activity has been tested for three the LacCer-HET-C2 and the ACD11 models as in the non-mammalian proteins; GLTP from P. anserina bovine GLTP structures (Figure 3). (HET-C2),11 and GLTP from A. thaliana (sequence accession number Q6NLQ3, data not shown) have Disulfide bonds affect transfer activity been shown to transfer glycolipids in vitro, while ACD11 has been reported to transfer sphingosine There are three conserved cysteine residues in instead of glycolipids.10 mammalian GLTPs. Two of these are buried inside Three-dimensional models of HET-C2 with a the protein and can make an intra-subunit disulfide bound lactosylceramide (LacCer-HET-C2) and of bond, while the third one is located on the surface ACD11 without a ligand were constructed in order and can form an inter-subunit disulfide bond, thus to compare their lipid-binding properties to GLTP. forming a GLTP dimer.16,17 In order to analyze the The conserved amino acid residues of the GLTPs are biological significance of the conserved cysteine mainly located in the sugar-binding pocket, not in residues in GLTPs, the effects of copper sulfate the hydrophobic channel. The hydrophobic nature and NEM (sulfhydryl-specific reagent N-ethyl- of the channel is, however, conserved. The polar maleimide) on disulfide formation of bovine amino acids Asp48, Asn52, Trp96, and His140 are in recombinant GLTP (rGLTP) were analyzed by the sugar-binding pocket of the mammalian GLTPs non-reducing SDS-PAGE and glycolipid transfer (Figure 5(a)). Based on our sequence alignment and activity measurements. Copper sulfate is known to structural models, the sugar-binding pocket is promote both intra- and inter-subunit disulfide much more charged in ACD11 than in HET-C2 bond formation, whereas NEM is believed to and bovine GLTP (Figure 3). This is mainly caused block surface cysteine residues only. by the substitution of residues Asn52 and Trp96 in Two different forms of bovine rGLTP, which mammalian GLTP with a lysine and an arginine, correspond to monomers with and without an respectively, in ACD11 while in HET-C2 the two internal disulfide bridge, are detected on non- amino acids are conserved. Even though the overall reducing SDS-PAGE (Figure 6(a), lane 1). The sequence identity between the human GLTP and treatment of rGLTP with 1 mM copper sulfate for HET-C2 is only 28%, as many as 15 of the 22 60 min decreases the amount of the GLTP monomer residues (68%) at 4 A˚ distance from the lactosylcera- without the internal disulfide bond and results in mide in the human LacCer-GLTP structure are the formation of a GLTP dimer too (Figure 6(a), identical between HET-C2 and human GLTP lane 2). After the treatment of rGLTP with 0.45 mM (Supplementary Table 1). However, in ACD11 only (or higher concentrations) of NEM for 30 min, only seven of these 22 residues (31%) are identical to the monomeric rGLTP without the disulfide bond human GLTP. The amino acids Ile45 and Phe148, is detected on SDS-PAGE, while at concentrations controlling the shape and size of the connecting lower than 0.05 mM NEM both monomers are channel between the acyl chain cavity and the visible (Figure 6(a), lane 3 and 4). The addition of sugar-binding pocket of the mammalian GLTPs, 1 mM copper sulfate to the NEM-treated (0.45 mM)

Figure 4. Adaptive ligand-binding of GLTP. The bovine apo-GLTP (a), intermediate-GLTP (b) and GM3-GLTP (c) structures as well as the human LacCer-GLTP (d) structure are shown as cartoons. The bound fatty acid (a and b), GM3 (c) and LacCer (d) are shown as spheres. An arrow indicates the point where the sphingosine and acyl chain branches in the GM3-GLTP and LacCerGLTP structures (c and d). The uninterpretable electron density near the bound fatty acid in intermediate-GLTP is shown in blue (b). The a-helices are numbered. The hydrophobic channel is shown as a transparent grey surface. The sugar-binding sites are labelled (S) and the three different parts of the hydrophopic channel are numbered: (I) acyl chain binding cavity, (II) connecting channel and (III) extension of the acyl chain cavity. (e) A stereoview of the GM3 and LacCer ligands of the superimposed bovine GM3-GLTP and human LacCer-GLTP structures. The carbon atoms of GM3 and LacCer ligands are coloured yellow and blue, respectively. The three different parts of the hydrophopic channel are numbered as above. The arrow points to the solvent accessible sphingosine part of the GM3 ligand. Figure 5. Glycolipid-binding site of GLTPs. (a) Comparison of the superimposed bovine apo (red), intermediate (white) and GM3-GLTP (blue) as well as human apo (cyan) and LacCer-GLTP (magenta) structures. The side-chains of D48, N52, W96 and Y207 at the sugar-binding site and the side-chains F42, I45, and L152 that are important for regulating the size and shape of GLTP cavities are labeled. Two alternative rotamers are shown for F148 of bovine apo-GLTP. The GM3 ligand in GM3-GLTP is shown in green and the hydrophobic cavity in GM3-GLTP is shown as a transparent surface. (b) Comparison of the superimposed X-ray structures of bovine apo-GLTP (red) and human LacCer-GLTP (magenta) together with the homology models of LacCer-HET-C2 (yellow) and ACD11 (cyan). The same side-chains of GLTP structures as in (b) and the corresponding residues in the HET-C2-LacCer and ACD11 structures are shown. The labels for residues of HET-C2 and ACD11 are indicated by an asterisk (*) and a double asterisk (**), respectively. The acyl chain binding cavity (I), connection channel (II) and extension to the acyl chain cavity (III) are indicated. The LacCer ligand of LacCer-HET- K C2 model is shown in green and the ligand-binding cavity of LacCer-GLTP is shown as a transparent surface. (c) The weighted 2Fo Fc electron density map (blue) is drawn with a 2 A˚ radius around the atoms of the GM3 ligand (green Ca atoms) and the residues D48, N52 H140 and Y207 (white Ca atoms) of bovine GM3-GLTP structure. Contours are shown at 1.0s. Lipi Binding of Glycolipid Transfer Protein 231

the 0.45 mM NEM-treated rGLTP or about 50% of that of the control rGLTP. Combined with the gel electrophoresis results, this suggests that the NEM treatment blocks the internal cysteine residues as well as the surface cysteine, thus preventing the copper sulfate-aided formation of the internal disulfide bond.

Discussion

The X-ray structures of bovine apo, intermediate and GM3-GLTP were refined to 1.8 A˚ or better resolution and, thus, they provide reliable infor- mation for studying the structure–function relation- ship of GLTPs. The bovine apo-GLTP structure is similar to the human apo-GLTP structure,13 whereas the intermediate and GM3-GLTP struc- tures represent novel intermediate conformations sharing features of both the apo-GLTPs and the human LacCer-GLTP.13 The structural differences in the GLTP structures are mainly seen in the a2 and a6 helices and in the loop a1-a2(Figure 1). The acyl chain cavity of the bovine apo and intermediate- GLTP structures contains a fatty acid, which probably originates from the expression host. Similarly, the human apo-GLTP structure has been Figure 6. Non-reducing SDS-PAGE analysis and reported to have an undefined hydrocarbon in the glycolipid transfer activity of the recombinant bovine 13 GLTP. (a) Panel I shows silver stained SDS-PAGE and acyl chain cavity. Since the acyl chain cavity in panel II immunoblotting results using an anti-GLTP these structures is solvent accessible the presence of polyclonal antibody. Arrows indicate the rGLTP dimer, a ligand is probably favorable for the stability of the and the two monomeric forms: monomer 1 without an GLTP structure. The GM3-GLTP structure has a internal disulfide bond and monomer 2 with an internal glycolipid ligand, GM3, which was added at the disulfide bond. Lane 1, untreated rGLTP; lane 2, CuSO4 crystallization setup. Only one of the sugar units, (1 mM)-treated rGLTP; lane 3, NEM (0.45 mM)-treated the glucose that is b-linked to the ceramide, is seen rGLTP; lane 4, NEM (0.05 mM)-treated rGLTP; lane 5, in the electron density maps of the sugar-binding NEM (0.45 mM) and CuSO4 (1 mM)-treated rGLTP. pocket. The other hydrophilic sugar units most (b) The in vitro transfer rate of fluorescence-labeled AV-GalCer for the samples treated in the same way as likely extend into the solvent and are highly for the SDS-PAGE analysis. The transfer data were flexible. The sphingosine chain of GM3 lies on the normalized to the transfer rate of untreated GLTP. surface of GLTP and only the acyl chain is inserted into the hydrophobic channel. Therefore, GM3 binds to GLTP in a different way than lactosylcera- mide in the human LacCer–GLTP complex13 where the sphingosine and acyl chains of lactosylceramide are both inside the channel. rGLTP leads to only minor increase in the amount of The lipid binding to GLTP induces changes in the the GLTP monomer with the internal disulfide bond shape and size of the hydrophobic channel (Figure 6(a), lane 5). (Figure 4). The conformational changes during The glycolipid transfer activity of the bovine lipid binding do not seem to be simultaneous, rGLTP is also clearly affected by the copper sulfate since the rearrangement of the loop a1-a2 and the and the NEM treatment. The treatment with bending of the N terminus of a2 helix are 0.05 mM NEM results in a slight reduction (10%) independent of the movement of the a6 helix. The of the glycolipid transfer activity, but treating with a helices can move several A˚ ngstro¨ms with respect to higher concentration of NEM diminishes more than each other, which makes the hydrophobic channel half of the transfer activity compared to the very flexible and allows adaptive binding of lipids. untreated rGLTP (Figure 6(b)). These results indi- The flexibility is enhanced by the conformational cate that the internal disulfide bond is important for changes of a few key residues (Phe42, Ile45 and glycolipid transfer activity. The copper sulfate Phe148). Ile45 from the a2 helix functions as a gate; treatment of rGLTP reduces the transfer activity it may open the connecting channel between the dramatically (70%), which can be explained by the acyl chain cavity and the sugar-binding site by formation of inactive GLTP dimers. Interestingly, rotating its side-chain. Phe148, which is located the transfer activity of rGLTP treated with both near Ile45, has a similar role; it can accommodate a NEM and copper sulfate is roughly the same as for second lipid chain in the hydrophobic channel by 232 Lipi Binding of Glycolipid Transfer Protein altering its side-chain conformation. In the human surface area could make favorable contacts with the LacCer–GLTP complex structure Phe148 blocks the negatively charged donor membrane and thus extension of the acyl chain cavity, which is observed anchor GLTP to the membrane. GLTP has been in the bovine intermediate and GM3-GLTP suggested to transfer newly synthesized glycosyl- structures (Figure 5(a)), whereas it is blocked by ceramide from the cytosolic face of the Golgi to the Phe42 in the bovine apo-GLTP structure and by plasma membrane.7,8 The positively charged sur- Phe42 and Leu152 in the human apo-GLTP face area of GLTP might interact with the negatively structures. Interestingly, the glycolipid transfer charged inner leaflet of the plasma membrane in the activity of GLTP was reduced when the hydro- similar way as with the negatively charged donor phobic gate residues, Ile45 and Phe148, were vesicles. Interestingly, the negatively charged donor mutated to the polar residues aspargine and serine, vesicles do not affect the glycolipid transfer rate of respectively.13 HET-C2 in the same way as the transfer rate of The sugar-binding pocket of GLTP has several GLTP.11 This is in agreement with our homology conserved residues of which Asp48, Asn52, His140, model of HET-C2, which lacks the positively and, especially Trp96, are known to be important charged patch on its surface (Figure 3(b)). for ligand recognition and binding.13,15 A similar The formation of an intra-molecular disulfide aspartate, asparagine and tryptophan environment bond has been described for GLTP purified from has been observed in the sugar-binding site of other pig brain16 and shown to increase the lipid-binding proteins like in the Escherichia coli galactose affinity and the turnover rate of lipid transfer.17 In chemoreceptor protein18 (PDB code 2GBP). Inter- agreement with the earlier study with the native estingly, the residues in the sugar-binding pocket GLTP, our results show that rGLTP monomers and the hydrophobic channel of GLTP are well produced in E. coli have decreased lipid transfer conserved in HET-C2, which suggests that the lipid activity when the pre-existing intra-molecular binding of GLTP and HET-C2 are similar. However, disulfide bond is disrupted. Therefore, it seems in ACD11 neither the sugar-binding nor the gate that GLTP can form an internal disulfide bond, residues are well conserved and, in addition, the which has an effect on its function. However, the surface potentials of the ACD11 model are native GLTP has not been reported to form dimers16 markedly different from those of GLTP and HET- and we did not detect the dimeric form in the C2 (Figure 3). Thus, our study provides structural untreated rGLTP produced in E. coli, indicating that explanation for why ACD11 is unable to transfer the inter-molecular disulfide bond is not formed glycolipids even though it is homologous to GLTP.10 in vivo. ACD11 has been shown to transfer sphingosine No intra or inter-molecular disulfide bonds were in vitro but the binding mode is still unclear.10 detected in our three crystal structures of bovine Sphingosine is a component of both GM3 and GLTP. The three cysteine residues in GLTP are lactosylceramide, but the binding mode of sphin- located on the surface of the protein (Cys36), in the gosine to ACD11 might be different than it is in the a4 helix (Cys112) and in the N terminus of the a8 bovine GM3–GLTP or human LacCer–GLTP helix (Cys176). Cys36 can probably form a disulfide complex. The sphingosine could bind to the bond with the corresponding cysteine in another putative acyl chain cavity of ACD11 either in a GLTP molecule and thereby form a dimer, while similar manner to the acyl chain in the GM3-GLTP Cys112 and Cys176 that are at 9.5 A˚ distance from structure or as the fatty acid in the bovine apo and each other (measured from the Ca-atoms) could intermediate-GLTP structures. Based on these form an internal disulfide bond. The formation of results we suggest that ACD11 should not be the intra-molecular bond between the cysteine classified as a GLTP protein, even though it most residues 112 and 176 would, however, require at likely shares the two-layer all-a-helical fold with least one of the helices to shift. Noteworthy, the GLTP. Moreover, three putative GLTP sequences of putative movement of the a8 helix to bind the two A. thaliana (sequence accession numbers: Q6NLQ3, cysteine residues would affect the lipid binding, O22797 and Q9LU33) have a higher sequence since the a8 helix of GLTP connects the solvent identity (over 20%) to mammalian GLTPs than accessible end of the acyl chain cavity to the C ACD11 (18%) and, furthermore, the sugar-binding terminus of the protein, which is in the vicinity of and gate residues lining the hydrophobic channel of the sugar-binding pocket. Therefore, it seems GLTP are conserved in them unlike in ACD11. possible that the internal disulfide bond regulates Actually, one of the putative A. thaliana GLTPs the lipid transfer activity of GLTP through the (Q6NLQ3; Figure 2) has recently been expressed movement of the a8 helix and might affect both in E. coli and tested to have glycolipid transfer the recognition of the sugar unit and the binding of activity in vitro (unpublished results). the acyl chain. Previously, Mattjus et al.19 have shown that Taken together, the results presented here suggest negatively charged donor lipid membranes the following mechanism for glycolipid binding to decrease the glycolipid transfer rate of GLTP. GLTP. Firstly, the sugar units of a glycolipid are Electrostatic potential mapping of our structures recognized and bind to the sugar-binding pocket. shows that a large positively charged surface area, The conserved amino acid residues Asp48, Asn52, formed mainly by four lysine residues, is located Trp96 and His140 are likely to play an important close to the sugar-binding site (Figure 3(a)). This role in this step. The internal disulfide bond might Lipi Binding of Glycolipid Transfer Protein 233 affect the sugar binding by signaling through the a8 using 2–5 mM heavy-atom concentrations. Three out of helix. Secondly, the loop a1-a2 rearranges, the N five heavy-atom derivatives were found to bind to the terminus of the a2 helix bends and the side-chain of crystals (see Supplementary Table 1). In the case of the Ile45 rotates to open the connecting channel apo-GLTP data set, the crystal was soaked with 12 nM 8 between the sugar-binding pocket and the acyl GM3 for 30 min at 25 C before the diffraction data collection, but the soaked ligand was not seen in the final chain cavity. This permits the acyl chain of the structure. The GM3-GLTP crystals were produced by glycolipid ligand to displace the bound fatty acid or co-crystallizing GLTP with GM3 in a molar ratio of 1:1. hydrocarbon chain. Thirdly, in order to enable the sphingosine chain to enter the hydrophobic Data collection and structure determination channel, the a6 helix shifts outwards and the side- chain of Phe148 in the connecting channel rotates Synchrotron radiation was used to collect diffraction into the extension of the acyl chain cavity. data from the intermediate,12 apo and GM3-GLTP crystals as well as from the heavy-atom derivative crystals either at beam line X13, EMBL/DESY Hamburg or at beam lines Materials and Methods ID14.1 and ID14.2 at ESRF, Grenoble (see Table 1 and Supplementary Table 2). For the data collection, all crystals were cryoprotected with 20% (v/v) glycerol, Crystallization and heavy atom soaking except the GM3-GLTP crystal that was dipped in 2 M LiSO4 and 20% glycerol, and flash-frozen either in a 100 K The bovine GLTP protein was expressed, purified nitrogen stream or using liquid nitrogen. The data were and crystallized as described recently.12 For the processed and scaled with the program suite XDS.20 The purposes of MIR method, five different GLTP crystals intermediate-GLTP data set (PDB code 1TFJ) was were heavy-atom derivatized with (1) potassium tetra- collected first and hence used in the MIR method. No chloroplatinate (II) (Cl4K2Pt), (2) potassium dicyano- other X-ray structures of proteins similar to bovine GLTP 21 aurate (I) (KAu(CN)2), (3) mercury (II) chloride (HgCl2), were available at that time. The program SOLVE was (4) 4-(chloromercuri)benzensulfonic acid sodium salt used to find and refine the heavy-atom positions as well (C6H4ClHgO3SNa) and (5) potassium tetranitroplatinate as to calculate the initial phases and electron density ˚ 22,23 (II) (K2Pt(NO2)4), which correspond to JBScreen Heavy maps at 2.5 A resolution. The program RESOLVE was (Jena Bioscience) tubes 1, 2, 4, 5 and 6, respectively. used to build automatically an initial GLTP model. After Heavy-atom soakings were carried out at 25 8C for 1–48 h density modification and phase extension to 1.61 A˚

Table 1. Data collection and refinement statistics

Intermediate-GLTP Apo-GLTP GM3-GLTP Data collectiona Space group P21 P21 P21 Cell dimensions a, b, c (A˚ ) 55.4, 34.9, 58.2 55.7, 35.4, 58.4 55.5, 34.6, 58.3 a, b, g (deg.) 90, 116, 90 90, 117, 90 90, 116.5, 90 Resolution (A˚ ) 20–1.61(1.70–1.61) 25–1.36(1.45–1.36) 25–1.8(1.9–1.8) b Rmerge (%) 6.7(41.4) 3.9(49.7) 8.1(52.1) I/sI 12.2(3.3) 21.65(2.55) 16.6(4.2) Completeness (%) 97.1(95.1) 94.7(73.2) 99.5(99.7) Redundancy 4.1(3.8) 6.1(3.6) 6.1(6.1) Refinement Resolution (A˚ ) 20–1.6 25–1.4 25–1.8 No. reflections 105,133 256,245 115,280 c c,d Rwork /Rfree 18.2/22.1 18.4/21.9 18.9/25.7 No. atoms Protein 1698 1774 1639 Ligand/ion 19 12 51 SO4 ––2 Fatty acid/glycolipid 12 12 43 Glycerol 6 6 – Water 151 292 119 B-factorse Protein 26.8 18.2 21.9 Ligand/ion 44.4 31.3 39.0 Water 35.3 32.4 28.7 r.m.s. deviations Bond lengths (A˚ ) 0.018 0.018 0.012 Bond angles (deg.) 1.7 1.7 1.3

a One crystalP wasP used for each dataset.P P Values for the highest resolution shell are shown in parenthesis. b Z K = Rmerge hkl i jIiðhklÞ IðhklÞj hkl i IiðhklÞ. c Z K = Rwork=free SjFobs Fcalcj jFobsj. d Performed on 5% of the reflections. e Average value for all atoms. 234 Lipi Binding of Glycolipid Transfer Protein resolution using the program DM24 of the CCP4 suite,25 program.38 Cavities were calculated with the program the initial model was used as an input for automatic Surfnet39 using 1.4 A˚ and 4.0 A˚ radii for minimum and model building with ARP/wARP.26 The model was maximum gap spheres, respectively. Figure 2 was refined with Refmac527 (CCP4 suite) and modified and generated with the program ALSCRIPT.40 Average rebuilt with O.28 Solvent atoms were added to the model B-factors over all atoms were calculated using with the automatic procedure of ARP/wARP.29 The MOLEMAN.41 apo-GLTP structure (PDB code 1WBE) was built with ARP/wARP using the intermediate-GLTP structure as an Protein Data Bank accession code existing model, rebuilt manually with O and refined with Refmac5. The DKA ligand was built with the ARP/wARP 30 The structures have been deposited in the RCSB Protein LigandBuild program and solvent atoms were built as Data Bank with accession codes 1TFJ, 1WBE and 2BV7. above. The GM3-GLTP structure was solved with MOLREP of the CCP4 suite using the intermediate- GLTP structure as the search model, rebuilt manually with O and refined with Refmac5. All structures were analyzed with the programs PROCHECK31 and 32 Acknowledgements WHATIF. Professor Mark Johnson is acknowledged for the Homology modeling excellent facilities at the Structural Bioinformatics Laboratory at the Department of Biochemistry and The sequences of 11 GLTP-like proteins from seven Pharmacy, A˚ bo Akademi University. Professor organisms were aligned (Figure 2) with the program J. Peter Slotte is acknowledged for access to the MALIGN33 within the Bodil visualization and modeling package.34 Homology models of HET-C2 and ACD11 fluorescence facility. This project was supported were made using the program MODELLER35 based on by the Academy of Finland, Sigrid Juse´lius the pair-wise alignment derived from the multiple Foundation, Magnus Ehrnrooth Foundation, Oscar alignment and the crystal structures of human LacCer- O¨ flund Foundation, Svenska Kulturfonden, GLTP13 and bovine apo-GLTP, respectively. Medicinska Understo¨dsfo¨reningen Liv och Ha¨lsa, the National Graduate School of Informational and ˚ SDS-PAGE analysis and glycolipid transfer activity Structural Biology (ISB), Abo Akademi and the measurements European Community—Access to Research Infrastructure Action of the Improving Human

Treatment of 0.4 mg/ml GLTP with NEM and CuSO4 Potential Programme to the EMBL Hamburg was done under stirring for 30 min and 60 min, respec- Outstation, contract number: HPRI-CT-1999-00017. tively. The GLTP samples were incubated in Laemmli We acknowledge the European Synchrotron buffer without b-mercaptoethanol at 95 8C before the non- Radiation Facility for provision of synchrotron reducing SDS-PAGE analysis. The SDS-PAGE gel was radiation facilities and we thank the staff for stained with the silver staining procedure kit (Pierce, assistance in using the ID14 beamlines. The staff at Rockford, IL, USA). Commercially raised rabbit poly- clonal anti-GLTP antibody (MedPro, Oslo, Norway) was EMBL X13 beamline at the DORIS storage ring, used in Western blot analysis of GLTP. The resonance DESY, Hamburg are acknowledged too. Johan energy transfer assay for measuring anthrylvinyl- Edqvist is acknowledged for the work on A. thaliana galactosylceramide (AV-GalCer, N-[(11E)-12-(9-anthryl)- GLTP and Elina Palonen is acknowledged for 11-dodecenoyl]-1-O-b-galactosylsphingosine) transfer technical assistance. Veli-Matti Leppa¨nen and was presented recently36 and is based on the method Esko Oksanen are thanked for assistance with data described.37 Briefly, AV-GalCer transfer from probe collection at ESRF. The X-ray facilities of the Turku sonicated donor vesicles consisting of 1% AV-GalCer, Centre for Biotechnology are also acknowledged. 3% dihexadecyloxacarbocyanine perchlorate and 96% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) to probe sonicated pure POPC acceptor vesicles was started by GLTP injection. The assay as well as the Supplementary Data vesicle preparation was done in a sodium phosphate buffer containing 10 mM sodium dihydrogen phosphate, Supplementary data associated with this article 1 mM dithithreitol and 1 mM EDTA at pH 7.4. can be found in the online version at doi 10.1016/j. jmb.2005.10.031 Miscellaneous methods

Figures 1 and 3–5 were created with the PyMOL References Molecular Graphics System† and/or Corel Draw 11 software. Electrostatic potentials were calculated using 1. Abe, A. (1990). Primary structure of glycolipid the APBS software package‡ plugin in PyMOL. Figure 3 transfer protein from pig brain. J. Biol. Chem. 265, was rendered using POV-Ray 3.6§. The secondary 9634–9637. structure elements were assigned with the DSSP 2. Lin, X., Mattjus, P., Pike, H. M., Windebank, A. J. & Brown, R. E. (2000). Cloning and expression of glycolipid transfer protein from bovine and porcine † http://www.pymol.org brain. J. Biol. Chem. 275, 5104–5110. ‡ http://agave.wustl.edu/apbs 3. Yamada, K., Abe, A. & Sasaki, T. (1986). Glycolipid § http://www.povray.org/ transfer protein from pig brain transfers glycolipids Lipi Binding of Glycolipid Transfer Protein 235

with beta-linked sugars but not with alpha-linked impede the protein-mediated transfer of glycosphin- sugars at the sugar–lipid linkage. Biochim. Biophys. golipids between phospholipid bilayers. Biochemistry, Acta, 879, 345–349. 39, 1067–1075. 4. van Meer, G., Wolthoorn, J. & Degroote, S. (2003). The 20. Kabsch, W. (1993). Automatic processing of rotation fate and function of glycosphingolipid glucosylcera- diffraction data from crystals of initially unknown mide. Phil. Trans. Roy. Soc. Lond. ser. B, 358, 869–873. symmetry and cell constants. J. Appl. Crystallog. 26, 5. Degroote, S., Wolthoorn, J. & van Meer, G. (2004). The 795–800. cell biology of glycosphingolipids. Semin. Cell Dev. 21. Terwilliger, T. C. & Berendzen, J. (1999). Automated Biol. 15, 375–387. MAD and MIR structure solution. Acta Crystallog. sect. 6. Jeckel, D., Karrenbauer, A., Burger, K. N., van Meer, G. D, 55, 849–861. & Wieland, F. (1992). Glucosylceramide is synthesized 22. Terwilliger, T. C. (2000). Maximum-likelihood density at the cytosolic surface of various Golgi subfractions. modification. Acta Crystallog. sect. D, 56, 965–972. J. Cell Biol. 117, 259–267. 23. Terwilliger, T. C. (2003). Automated main-chain 7. Sasaki, T. (1990). Glycolipid transfer protein and model building by template matching and iterative intracellular traffic of glucosylceramide. Experientia, fragment extension. Acta Crystallog. sect. D, 59, 46, 611–616. 38–44. 8. Warnock, D. E., Lutz, M. S., Blackburn, W. A., Young, 24. Cowtan, K. (1994). “dm”: an automated procedure for W. W., Jr & Baenziger, J. U. (1994). Transport of phase improvement by density modification. Joint newly synthesized glucosylceramide to the plasma CCP4 ESF-EACBM Newsletter Protein Crystallog. 31, membrane by a non-Golgi pathway. Proc. Natl Acad. 34–38. Sci. USA, 91, 2708–2712. 25. Collaborative Computational Project Number 4. 9. Saupe, S., Descamps, C., Turcq, B. & Begueret, J. (1994). The CCP4 suite: programs for protein (1994). Inactivation of the Podospora anserina crystallography. Acta Crystallog. sect. D, 50, 760–763. vegetative incompatibility locus het-c, whose product 26. Perrakis, A., Morris, R. & Lamzin, V. S. (1999). resembles a glycolipid transfer protein, drastically Automated protein model building combined with impairs ascospore production. Proc. Natl Acad. Sci. iterative structure refinement. Nature Struct. Biol. 6, USA, 91, 5927–5931. 458–463. 10. Brodersen, P., Petersen, M., Pike, H. M., Olszak, B., 27. Murshudov, G. N., Vagin, A. A. & Dodson, E. J. (1997). Skov, S. & Odum, N. (2002). Knockout of Arabidopsis Refinement of macromolecular structures by the accelerated-cell-death11 encoding a sphingosine maximum-likelihood method. Acta Crystallog. sect. transfer protein causes activation of programmed D, 53, 240–255. cell death and defense. Genes Dev. 16, 490–502. 28. Jones, T. A., Zou, J. Y., Cowan, S. W. & Kjeldgaard, M. 11. Mattjus, P., Turcq, B., Pike, H. M., Molotkovsky, J. G. & (1991). Improved methods for building protein models Brown, R. E. (2003). Glycolipid intermembrane in electron density maps and the location of errors in transfer is accelerated by HET-C2, a filamentous these models. Acta Crystallog. sect. A, 47, 110–119. fungus gene product involved in the cell–cell 29. Lamzin, V. S. (1993). Automated refinement of protein incompatibility response. Biochemistry, 42, 535–542. models. Acta Crystallog. sect. D, 49, 129–147. 12. West, G., Nymalm, Y., Airenne, T. T., Kidron, H., 30. Zwart, P. H. & Lamzin, V. S. (2004). Modelling bound Mattjus, P. & Salminen, T. T. (2004). Crystallization ligands in protein crystal structures. Acta Crystallog. and X-ray analysis of bovine glycolipid transfer 60 protein. Acta Crystallog. sect. D, 60, 703–705. sect. D, , 2230–2239. 13. Malinina, L., Malakhova, M. L., Teplov, A., Brown, R. E. 31. Laskowski, R. A., Macarthur, M. W., Moss, D. S. & & Patel, D. J. (2004). Structural basis for glyco- Thornton, J. M. (1993). Procheck—a program to check sphingolipid transfer specificity. Nature, 430, the stereochemical quality of protein structures. 1048–1053. J. Appl. Crystallog. 26, 283–291. 14. Brown, R. E., Stephenson, F. A., Markello, T., 32. Vriend, G. (1990). WHAT IF: a molecular modeling Barenholz, Y. & Thompson, T. E. (1985). Properties and drug design program. J. Mol. Graph. 8, 52–56. of a specific glycolipid transfer protein from bovine 33. Johnson, M. S. & Overington, J. P. (1993). A structural brain. Chem. Phys. Lipids, 38, 79–93. basis for sequence comparisons. An evaluation of 15. Malakhova, M. L., Malinina, L., Pike, H. M., Kanack, A. T., scoring methodologies. J. Mol. Biol. 233, 716–738. Patel, D. J. & Brown, R. E. (2005). Point mutational 34. Lehtonen, J. V., Still, D. J., Rantanen, V. V., Ekholm, J., analysis of the liganding site in human glycolipid Bjorklund, D., Iftikhar, Z. et al. (2004). BODIL: a transfer protein: functionality of the complex. J. Biol. molecular modeling environment for structure–func- Chem. 280, 26312–26320. tion analysis and drug design. J. Comput. Aided Mol. 16. Abe, A. & Sasaki, T. (1989). Sulfhydryl groups Des. 18, 401–419. in glycolipid transfer protein: formation of an 35. Sali, A. & Blundell, T. L. (1993). Comparative protein intramolecular disulfide bond and oligomers by modelling by satisfaction of spatial restraints. J. Mol. Cu2C-catalyzed oxidation. Biochim. Biophys. Acta, Biol. 234, 779–815. 985, 38–44. 36. Nylund, M. & Mattjus, P. (2005). Protein mediated 17. Abe, A. & Sasaki, T. (1989). Formation of an glycolipid transfer is inhibited from sphingomyelin intramolecular disulfide bond of glycolipid transfer membranes but enhanced to sphingomyelin protein. Biochim. Biophys. Acta, 985, 45–50. containing raft like membranes. Biochim. Biophys. 18. Vyas, N. K., Vyas, M. N. & Quiocho, F. A. (1988). Sugar Acta, 1669, 87–94. and signal-transducer binding sites of the Escherichia 37. Mattjus, P., Molotkovsky, J. G., Smaby, J. M. & coli galactose chemoreceptor protein. Science, 242, Brown, R. E. (1999). A fluorescence resonance energy 1290–1295. transfer approach for monitoring protein-mediated 19. Mattjus, P., Pike, H. M., Molotkovsky, J. G. & glycolipid transfer between vesicle membranes. Anal. Brown, R. E. (2000). Charged membrane surfaces Biochem. 268, 297–304. 236 Lipi Binding of Glycolipid Transfer Protein

38. Kabsch, W. & Sander, C. (1983). Dictionary of protein 40. Barton, G. J. (1993). ALSCRIPT: a tool to format secondary structure: pattern recognition of hydrogen- multiple sequence alignments. Protein Eng. 6, 37–40. bonded and geometrical features. Biopolymers, 22, 41. Kleywegt, G. J., Zou, J. Y., Kjeldgaard, M. & 2577–2637. Jones, T. A. (2001). Around O. In Crystallography of 39. Laskowski, R. A. (1995). SURFNET: a program Biological Macromolecules (Rossmann, M. G. & Arnold, for visualizing molecular surfaces, cavities, and E., eds) International Tables for Crystallography, vol. F, intermolecular interactions. J. Mol. Graph. 323–330, pp. 353–356, Kluwer Academic Publishers, 307–308. Dordrecht, The Netherlands.

Edited by R. Huber

(Received 1 August 2005; received in revised form 6 October 2005; accepted 11 October 2005) Available online 8 November 2005