Design of natural killer T cell activators:Structure and function of a microbialglycosphingolipid bound to mouse CD1dDouglass Wu*†, Dirk M. Zajonc†‡, Masakazu Fujio*, Barbara A. Sullivan§, Yuki Kinjo§, Mitchell Kronenberg§,Ian A. Wilson‡¶, and Chi-Huey Wong*¶

Departments of *Chemistry and ‡Molecular Biology and The Skaggs Institute for Chemical Biology, The Scripps Research Institute, 10550 North Torrey PinesRoad, La Jolla, CA 92037; and §Division of Developmental Immunology, La Jolla Institute for Allergy and Immunology, 10355 Science Center Drive,San Diego, CA 92121

Contributed by Chi-Huey Wong, January 11, 2006

Natural killer T (NKT) cells provide an innate-type immune responseupon T cell receptor interaction with CD1d-presented antigens. Wedemonstrate through equilibrium tetramer binding and antigenpresentation assays with V�14i-positive NKT cell hybridomas thatthe Sphingomonas glycolipid �-galacturonosyl ceramide (GalA-GSL) is a NKT cell agonist that is significantly weaker than �-galac-tosylceramide (�-GalCer), the most potent known NKT agonist. ForGalA-GSL, a shorter fatty acyl chain, an absence of the 4-OH on thesphingosine tail and a 6�-COOH group on the galactose moietyaccount for its observed antigenic potency. We further determinedthe crystal structure of mCD1d in complex with GalA-GSL at 1.8-Åresolution. The overall binding mode of GalA-GSL to mCD1d issimilar to that of the short-chain �-GalCer ligand PBS-25, but itssphinganine chain is more deeply inserted into the F� pocket dueto alternate hydrogen-bonding interactions between the sphinga-nine 3-OH with Asp-80. Subsequently, a slight lateral shift (>1 Å)of the galacturonosyl head group is noted at the CD1 surfacecompared with the galactose of �-GalCer. Because the relativelyshort C14 fatty acid of GalA-GSL does not fully occupy the A� pocket,a spacer lipid is found that stabilizes this pocket. The lipid spacerwas identified by GC�MS as a mixture of saturated and monoun-saturated palmitic acid (C16). Comparison of available crystal struc-tures of �-anomeric glycosphingolipids now sheds light on thestructural basis of their differential antigenic potency and has ledto the design and synthesis of NKT cell agonists with enhancedcell-based stimulatory activities compared with �-GalCer.

adjuvant � glycolipid

The CD1 family of antigen-presenting glycoproteins mediatesT cell responses through the presentation of self and foreign

lipids, glycolipids, lipopeptides, or amphipathic small moleculesto T cell receptors (TCR) (1–10). In humans, the various CD1isoforms are categorized into group I (CD1a, CD1b, CD1c, andCD1e) and group II (CD1d) based on sequence similarity (11).Through the binding and presentation of endogenous and ex-ogenous lipid antigens to TCRs, the CD1 pathway is reminiscentof peptide presentation by MHC class I and class II molecules.

The group II isotype, CD1d, presents antigens to a uniquepopulation of T lymphocytes termed natural killer T (NKT) cells(4), which express an invariant V�14i TCR in mice or V�24iTCR in humans, in addition to NK 1.1 and other receptorstypical of NK cells (12). These cells are sometimes called V�14ior V�24i NKT cells to distinguish them from other T lympho-cytes that can express NK1.1. The first defined and most potentV�14i NKT cell agonist, �-galactosylceramide (�-GalCer), wasoriginally isolated from a marine sponge and subsequentlyoptimized by medicinal chemistry (13). When bound to CD1d,�-GalCer strongly activates V�14i NKT cells, causing a rapidrelease of T helper type 1 and 2 cytokines. Although �-GalCerhas been of great value and use in exploration of NKT cell

biology, its unusual origin left unresolved questions about therole of NKT cells in host defense against microbial infections.

The recent discovery of �-glycuronosylceramides (14–17),including �-galacturonosyl and �-glucuronosylceramide, inSphingomonas bacteria as CD1d-presented V�14i NKT cellantigens, established the role of V�14i NKT cells in antimicro-bial defense. Sphingomonas are Gram-negative bacteria that lackLPS, and recognition of these �-glycuronosylceramides couldserve as alternatives to the LPS response in an innate-typereaction that activates dendritic cells and other cell types.Additionally, mycobacterial phosphatidylinositol mannosides (3)and the self-glycolipid ganglioside GD3 can activate a smallsubpopulation of NKT cells, and the self-lipid isoglobotrihexo-sylceramide (iGB3) (8) is apparently required for V�14i NKTdevelopment in mice. The potency of all of these antigens,however, is significantly lower than that of �-GalCer. Thestructural explanation for the high potency of �-GalCer is stillnot resolved nor are the means by which diverse antigens arerecognized by TCRs with an invariant �-chain and a restrictedV� usage.

Crystal structures of human (h)CD1a (18, 19), hCD1b (20, 21),hCD1d (22), and mouse CD1d (mCD1d) (23–26) have revealedhow differences in the topology of their respective bindinggrooves enable them to have some degree of ligand specificitywhile maintaining their ability to present a diverse set ofantigenic lipids (27). The initial structure of mCD1d revealed anoverall fold similar to MHC class I proteins (23). The antigenbinding groove had two deep pockets, designated A� and F�,which were lined with hydrophobic residues optimal for bindinglong hydrophobic chains, such as lipid tails. Structures for hCD1aand mCD1d complexed with glycosphingolipids and phosphati-dylcholine more clearly elucidated the structural requirementsfor binding to CD1d (22–26).

Here, we compared the antigenic potency and equilibriumbinding of CD1d tetramers loaded with bacterial glycosphingo-lipid �-galacturonosylceramide (GalA-GSL) with those loadedwith a synthetic analog, galactose glycosphingolipid (Gal-GSL),which contains a galactose coupled to the bacterial glycosphin-golipid and with those loaded with �-GalCer, which allowed usto assess how differences in the lipid or carbohydrate moieties ofthis closely related group of glycosphingolipids influence avidity

Conflict of interest statement: No conflicts declared.

Abbreviations: NKT, natural killer T; GalA-GSL, galacturonosyl glycosphingolipid; Gal-GSL,galactose glycosphingolipid; �-GalCer, �-galactosylceramide; TCR, T cell receptor.

Data deposition: The atomic coordinates have been deposited in the Protein Data Bank,www.pdb.org (PDB ID code 2FIK).

†D.W. and D.M.Z. contributed equally to this work.

¶To whom correspondence may be addressed. E-mail: wilson@scripps.edu or wong@scripps.edu.

© 2006 by The National Academy of Sciences of the USA

3972–3977 � PNAS � March 14, 2006 � vol. 103 � no. 11 www.pnas.org�cgi�doi�10.1073�pnas.0600285103

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020

and antigenic potency. We report the crystal structure of mCD1din complex with GalA-GSL at 1.8 Å resolution that representsthe first structure of a natural, microbial antigen bound to CD1d.This crystal structure sheds light into the molecular basis for itsantigenic potency and provides structural information on therole of the ‘‘spacer’’ lipid in the A� pocket, which is likely to bepresent when shorter chain lipid antigens are recognized. Thisstudy has subsequently led to the design and synthesis ofglycolipids with biological activities that surpass �-GalCer asmeasured by the production of IFN-� and IL-4 in a cell-basedassay.

ResultsV�14i NKT Cell Activation Correlates with TCR Avidity for Glycosphin-golipid Compounds. The antigenic potency of the natural bacterialantigen GalA-GSL was compared with the well studied V�14iNKT cell agonist, �-GalCer (Fig. 1). GalA-GSL has severaldifferences with �-GalCer, including a galacturonic head group,in which the 6�-OH is oxidized to a carboxylate. The lengths ofthe fatty acyl moieties also differ; �-GalCer has a C26 andGalA-GSL a C14 tail. Finally, the 4-OH of the sphingosine isabsent so that GalA-GSL has a sphinganine base rather than thephytosphingosine in �-GalCer. In addition, we also analyzedGal-GSL, a biosynthetic precursor to the Sphingomonas glyco-lipid, in which the galacturonic sugar is replaced by galactose(Fig. 2C). Comparison of the V�14i NKT cell response toGal-GSL with that of �-GalCer can assess the effect of these lipidmodifications. Similarly, comparison of the V�14i NKT cellresponse to GalA-GSL with that of Gal-GSL addresses anyeffect of the 6�-COOH modification of the sugar.

The in vitro response of spleen cells from V�14i TCR trans-genic mice to dendritic cells pulsed with these three compoundswas analyzed. Secretion of IFN-� and IL-4 was greatest in

response to �-GalCer and was substantially reduced with Gal-GSL. A higher dosage (approximately two orders of magnitude)was required for cytokine responses that were equivalent to�-GalCer. The GalA-GSL response was further decreased com-pared with Gal-GSL, especially at the lower concentrations ofantigen (Fig. 1 Upper).

Equilibrium binding of CD1d tetramers separately loadedwith each of the three different glycosphingolipids was consis-tent with the functional studies. Tetramers loaded with �-GalCerbound to the 2C12 V�14i NKT cell hybridoma with a Kdapproximately �100-fold better than tetramers loaded withGal-GSL. The combined data indicate that modifications to theceramide backbone in the bacterial antigen have a substantialinfluence on antigenic potency and overall avidity of the trimo-lecular interaction of TCR, glycolipid, and CD1d. Previously, weshowed that absence of the 4-OH alone had a relatively smallinfluence on antigenic potency and a �5-fold effect on equilib-rium binding of a soluble TCR or binding of CD1d tetramersloaded with this variant glycolipid to live cells (28). The shorterC8 acyl chain of PBS-25 and a C14 analog of �-GalCer also didnot cause a great decrease in antigenic potency (29, 30). Basedon these findings, we conclude that the combined differences ofthe shortened acyl chain and the absence of the sphingosine4-OH have a synergistic effect that is greater than the sum of theindividual differences. A further decrease in both antigenicpotency and avidity was measured by equilibrium tetramerstaining when GalA-GSL is compared with Gal-GSL, indicatingthat the carboxylate modification of the galactose also is impor-tant for the overall trimolecular interaction. In conclusion, eachof these substitutions distinguishes the natural bacterial antigen(GalA-GSL) from �-GalCer and results in a significant differ-ence in antigenic potency that is correlated with reduced TCRaffinity.

Protein Purification and Structural Determination. Soluble het-erodimeric complexes of mouse CD1d-�2 microglobulin (resi-dues 1–279 heavy chain and 1–99 �2 microglobulin) were se-creted by Spodoptera frugiperda (SF9) cells upon infection withrecombinant baculovirus and purified to homogeneity by affinityand size exclusion chromatography. GalA-GSL was loaded byincubating mCD1d with a 6-fold molar excess of glycolipid in theabsence of any detergents. The GalA-GSL�mCD1d complex wasfurther purified by size exclusion chromatography. Using thesitting drop vapor diffusion method, the mCD1d�glycolipidcomplex was crystallized, and its three-dimensional structurewas determined by molecular replacement with the proteincoordinates of the CD1d-sulfatide complex (Protein Data BankID code 2AKR) as the search model (Fig. 2 and Table 1, whichis published as supporting information on the PNAS web site)(26) with the sulfatide atoms omitted. The crystal structure wasrefined to 1.8-Å resolution with Rcryst and Rfree of 20.8% and24.3%, respectively.

The mCD1d�GalA-GSL Complex. The overall architecture is similarto other CD1d complex structures previously described (22, 25,26). mCD1d is a heterodimer composed of the �1, �2, and �3domains of CD1 noncovalently associated with �2 microglobulin.The �1- and �2-helices sit atop a six-stranded �-sheet platformthat together form a deep and hydrophobic binding groovesuitable for binding lipids (Fig. 2 A and B).

Like �-GalCer, the bacterial GalA-GSL is an �-linked glyco-sphingolipid (Fig. 2C). The fatty acid and sphinganine tails ofGalA-GSL extend into the A� and F� pockets, respectively,whereas the galacturonic head group is presented in the centerof the binding groove at the CD1 surface for recognition by anincoming TCR (Fig. 2D). Several nonpolar, van der Waals’interactions between the GalA-GSL acyl chains and mCD1dstabilize the lipid backbone. The A� pocket is larger than the F�

Fig. 1. Activation and equilibrium tetramer binding of GalA-GSL and Gal-GSL to V�14i NKT cells. (Upper) IL-4 (Left) and IFN-� (Right) production bylymphocytes stimulated with glycolipid pulsed dendritic cells. Data aremean � SD of triplicate wells. One representative set of data from threeexperiments is shown. (Lower) Equilibrium tetramer binding is shown forCD1d tetramers loaded with �-GalCer, GalA-GSL, Gal-GSL, and vehicle to 2C12hybridoma cells. Tetramer mean fluorescence intensity (MFI) is plotted againsttetramer concentration. One representative of two experiments is shown.

Wu et al. PNAS � March 14, 2006 � vol. 103 � no. 11 � 3973

IMM

UN

OLO

GY

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020

pocket and can accommodate longer lipid tails. Extending intothe A� pocket, the C14 fatty acid tail initially follows a pathperpendicular to the �-sheet floor. The lipid tail then transitionsinto a circular path that runs parallel to the six-stranded �-sheetand centers around Cys-12. The last four carbons of the C14 fattyacid tail wrap around the central pole in a counterclockwiseorientation that is reminiscent of the circular paths around thecentral pole (Phe-70 and Val-12) in the CD1a and CD1bstructures.

An extensive hydrogen bonding network is formed betweenthe polar functional groups of GalA-GSL and mCD1d. Asp-80stabilizes the 3-OH of the sphinganine base while Thr-156hydrogen bonds with the anomeric oxygen. As with the short-chain �-GalCer analog PBS-25, Asp-153 hydrogen bonds to the2�-OH and 3�-OH groups of the sugar moiety. No specifichydrogen bonding partners are observed for the galacturonic4�-OH or 6�-COOH groups that accounts for its tolerancetoward modifications at these positions (4�-OH axial to equa-

torial and 6�-OH to 6�-COOH) for binding to mCD1d. However,the importance of these two positions for TCR contact is stronglysuggested by their effects on antigenic potency. For example,�-glucosylceramide, which differs from �-GalCer only withregard to the orientation of the 4�-OH is a less potent antigenwith a 10-fold reduced affinity for the TCR when bound to CD1d(28). Similarly, the 6�-COOH reduces the antigenic potency andTCR avidity compared with galactose, as detailed above.

Comparison of the GalA-GSL and PBS-25 Complexes. Both GalA-GSLand PBS-25 are �-linked glycosphingolipids, but the potency ofPBS-25 is more akin to �-GalCer. Although the two ligands arebound in a similar orientation by mCD1d, their structuraldifferences can explain the dissimilarity in their biological ac-tivity (Fig. 3). The rms deviation between the GalA-GSL�mCD1d and PBS-25�mCD1d structures is 0.43 Å for all C�

atoms. Thus, both proteins are structurally very similar. In termsof the hydrogen bonding network, differences arise mainly from

Fig. 2. Overview of the structure and conformation of GalA-GSL in the mCD1d binding groove. (A) Front view of mCD1d (�1–�3 and �2 microglobulin domains)heterodimer (gray) with bound GalA-GSL (green) and palmitic acid (yellow) in stick representation. N-linked carbohydrates are depicted as gray stick models.Atom colors for all structural representations are green, gray, or yellow (carbon); red (oxygen); and blue (nitrogen). (B) Top view looking down into the CD1dbinding groove. The GalA-GSL fatty acid is bound in the larger A� pocket and the sphinganine backbone in the F� pocket. The palmitic acid spacer lipid sits inthe otherwise unoccupied A� pocket. N42 and N165 represent two ordered N-linked glycosylation sites. (C) Chemical structure of the glycolipids used in this studyfor structural comparison. The length of the individual alkyl chains is given as the number of carbon atoms (C8–C26). The differences between GalA-GSL and theother glycolipids are highlighted in red. (D) After refinement, a 2Fo � Fc electron density map was calculated and contoured at 1� as a blue mesh around GalA-GSL(green) and the palmitic acid (yellow). Several important contact residues that are involved in ligand binding are depicted and labeled. The refinement of Cys-12showed two possible conformations, both of which are depicted. (E) View looking down into the CD1d binding groove (TCR view).

3974 � www.pnas.org�cgi�doi�10.1073�pnas.0600285103 Wu et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020

GalA-GSL containing a sphinganine (3-OH) tail, whereasPBS-25 has a phytosphingosine (3-OH and 4-OH) tail. In thePBS-25 structure, the phytosphingosine backbone is stabilizedthrough interactions of the 3-OH and 4-OH groups with Asp-80.Arg-79 also is oriented so that it may participate in hydrogenbonding with the same 3-OH. When GalA-GSL is bound, theabsence of a 4-OH causes the sphinganine backbone to orientitself slightly lower in the pocket to make optimal hydrogenbonds with the terminal oxygens of Asp-80. Thus, GalA-GSL sitsslightly deeper in the groove, and the hydrogen bonding distancebetween the 3-OH and Asp-80 decreases slightly to 2.6 Å versus2.8 Å in the PBS-25 structure. As a result, the hydrogen bondbetween Arg-79 and the sphinganine backbone is lost and Arg-79is now oriented differently, possibly affecting its interactionbetween CD1d and the TCR. Despite these differences, thecontacts between the �2-helix and GalA-GSL remain the sameas seen with PBS-25. Asp-153 and Thr-156 still maintain exten-sive contacts with the sugar head group, anomeric oxygen, andfatty acid amide.

The electron density for the galacturonic head group is not aswell defined as for the lipid backbone. This observation, alongwith the higher B values, suggests that the sugar head group isnot as tightly bound. However, attempts to model an alternativeconfiguration were unsuccessful because of limitations of sterichindrance with the protein and the lack of any other obvioushydrogen bonding partners. Overall, the B values for the ligandare almost twice as high as the surrounding protein residues,which could also indicate that the binding groove is not fullyoccupied with GalA-GSL. Consistent with this hypothesis, re-finement of the structure with 50% occupancy of the ligandresulted in comparable B values for ligand and protein. This lackof rigidity in the head group binding may directly influence thepotency of GalA-GSL as a V�14i NKT cell agonist and explainswhy �-GalCer, with its tightly bound galactose head group, is amuch more potent NKT cell antigen.

Identification of the Spacer Lipid. Well resolved electron density fora linear hydrophobic molecule occupies most of the circular groovein the A� pocket. This previously unidentified spacer lipid wassuspected to be a C16 fatty acid (25). To identify the chemical natureof the molecule, a sample of mCD1d loaded with GalA-GSL wasrefluxed in a solution of methanol�HCl (29:3) to break fatty acidamide linkages and to methylate the free fatty acids. This methodfacilitated the extraction of all hydrophobic molecules into hexanes.The sample was then subjected to analysis via GC�MS. Theanalytical GC�MS data were in agreement with the observed

electron density for a fatty acid. A library search of the GC�MSfragmentation data identified fatty acid methyl esters myristic(C14:0) and palmitic (C16:0 and C16:1) acids. The presence of myristicacid was expected because it is released from GalA-GSL, and itsdetection then serves as an internal positive control for the exper-iment. No myristic acid was detected in protein samples, which werenot loaded with GalA-GSL (data not shown). Palmitic acid (C16:0or C16:1) could then be built with confidence into the electrondensity for the spacer lipid.

The location and orientation of the carboxylic functionalgroup could be established by the bifurcated electron density atonly one terminus of the spacer lipid. Further computationalanalysis supported this conclusion because of favorable electro-static interactions of the carboxyl with the backbone nitrogens ofVal-29 and Trp-40, which can partially neutralize any negativecharge on the ligand and contribute to its binding (Fig. 4A).Additionally, because of the high resolution of the data, Cys-12was observed to exist in two alternate conformations (Fig. 4B).One of these conformation locates the free thiol closer to thefatty acid carboxylate and, thus, provides additional specificityfor binding by the formation of an additional hydrogen bond.

DiscussionGalA-GSL was the first bacterial antigen described that canactivate the majority of V�14i and V�24i NKT cells. Its discovery

Fig. 3. Hydrogen bonding network between GalA-GSL and mCD1d and comparison to the mCD1d�PBS-25 structure. A close-up view of the hydrogen bondingnetwork formed between ligand and protein. The glycolipid is oriented such that the alkyl chains point straight down into the hydrophobic pockets. (A) mCD1dis depicted in gray ribbon backbone with selected side chains shown involved in hydrogen bonding. GalA-GSL is in green. Hydrogen bonds are represented asred dashes with distances in angstroms. (B) In the same view, superimposition (C� atoms) of the previously determined mCD1d�PBS-25 structure (Protein DataBank ID code 1Z5L) (25) and the GalA-GSL structure (rms deviation � 0.43 Å). mCD1d�PBS-25 is depicted in orange, and the coloring scheme for GalA-GSL isconsistent with A. Hydrogen bonds and distances between PBS-25 and mCD1d are depicted in black. Note that the GalA-GSL ligand is tilted slightly such thatthe sphinganine chain is inserted deeper into the F� pocket.

Fig. 4. Detailed view of the fatty acid in the A� pocket. Palmitic acid (inyellow) can be observed in the A� pocket. (A) Electrostatic surface calculation(red, electronegative, and blue, electropositive; �25 to �25 kT per electron)of the A� pocket. An electropositive patch formed by nitrogen backboneatoms of Val-29 and Trp-40 is available for interaction with the carboxylate ofthe fatty acid. (B) A 2Fo � Fc electron density map contoured at 1� as a bluemesh around palmitic acid and the fatty acid of GalA-GSL. Cys-12 is calculatedto adopt two conformations, one of which is able to interact with the terminalcarboxylate of palmitic acid. The hydrogen bond is labeled with the distancein angstroms.

Wu et al. PNAS � March 14, 2006 � vol. 103 � no. 11 � 3975

IMM

UN

OLO

GY

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020

firmly established that NKT cells are able to provide an innateimmune response to microbial antigens that is important forbacterial clearance and complements the other roles that thesedistinctive T cells play in autoimmunity (31, 32) and cancer (33).Thus, NKT cells are an attractive target for immunotherapy onthe basis of their ability to rapidly secrete cytokines and mod-ulate immune responses.

As seen from the functional data, �-GalCer is the optimalV�14i NKT cell agonist. Its crystal structure illustrates theimportance of the 3-OH and 4-OH groups of the sphingosinebase for orienting the sugar head group for TCR recognition (22,25). However, in GalA-GSL, the absence of the 4-OH causes anoverall shift of the lipid backbone into the CD1d groove so thatthe 3-OH retains the hydrogen bond with Asp-80, but loses theinteraction with Arg-79. These result in an obvious change in thelateral disposition of the sugar head group in the binding groove.In addition, the lack of well-defined electron density for thesugar suggests that GalA-GSL has more flexibility in the bindingpocket than �-GalCer. Because the 6�-OH makes no directhydrogen bonds with mCD1d residues, it is unlikely that thechange from a galactose to galacturonic acid head group affectsbinding to CD1d. However, because the sugar 6�-OH is locatedon the face where an incoming TCR would likely make contact,the oxidation from an alcohol to a carboxylic acid changes thecharacter of the recognition surface, consistent with the reducedantigenic potency of GalA-GSL compared with Gal-GSL. Fur-thermore, the changed orientation of the sphinganine in thegroove and the alteration in the position of Arg-79, which is likelyto contact the TCR (34), also most likely work in concert withthe 6�-COOH substitution to affect the overall TCR avidity andantigenic potency of the Sphingomonas-related glycolipids.

The fatty acid spacer lipid in the A� pocket seems to be aconserved element when short chain lipids are bound to mCD1d(25). It may function in providing stability to the A� pocket whenthe hydrophobic groove is unoccupied or until it is exchanged,presumably with the assistance of lipid-transfer proteins (35).The charge of the carboxylate is masked through electrostaticinteractions with backbone nitrogens and possibly through hy-drogen bonding with Cys-12.

GC�MS analysis showed the presence of palmitic acids (C16:0 andC16:1) that could be assigned to extra density in the A� pocket. Thesefatty acids make up �40% (wt�wt) of the total fatty acid compo-sition in Sf9 cells used for CD1d expression (36). Whether theloading of a fatty acid in the A� pocket is a consequence of the insectexpression system used to produce the protein or whether it occursnaturally remains unclear. F� pocket spacer lipids, although possiblein unloaded CD1d molecules, are unlikely to exist when the pocketis partially or fully occupied by a lipid antigen. The length of the F�pocket is optimal for a C18 chain, such as the sphingosine tail of�-GalCer. For situations in which the F� pocket is occupied by lipidtails of shorter length, such as the C9 sphingosine of synthetic analogOCH, the spacer lipid would have to be approximately C9 or less inlength. Because fatty acids of such lengths comprise only a minorpercentage of the total fatty acid composition of Sf9 cells, they areless likely to be acquired during protein folding in the endoplasmicreticulum.

For V�14i NKT cell stimulation to occur, the TCR of a V�14iNKT cell must first recognize a bimolecular complex of antigenand mCD1d. The crystal structures of GalA-GSL and PBS-25 intheir complex with mCD1d show how the optimal syntheticantigen uniquely assumes a rigid orientation of the sugar medi-ated by hydrogen bonds. The lateral shift of GalA-GSL causedby the absence of the sphingosine 4-OH in combination with thesugar 6�-COOH drastically impacts the overall potency of themolecule. Moreover, the binding of GalA-GSL alters the con-formation of �1-helix residues likely to interact with the TCR.

�-Linked ceramide-based glycolipids may have immunother-apeutic potential. The development of a more potent V�24i

NKT cell agonist relies on understanding the structural featuresthat make �-GalCer unique in its potency. From this structure,it is evident that the sphingosine 3-OH and 4-OH are vital foroptimal glycolipid orientation in the hydrophobic groove. Ga-lactose still remains unrivaled as the prototypical head group tomaximize NKT cell activation. In contrast, the presence of a fattyacid spacer lipid stabilizing the A� pocket suggests that the entireC26 N-fatty acyl tail of �-GalCer may not be fully necessary tomaximize activity. Previously, it has been reasoned that trun-cating the �-GalCer fatty acyl tail diminished its potency (4, 30).However, based on this study, we have designed and synthesized�-GalCer analogs replacing the C26 N-fatty acyl tail with a C6–10N-fatty acyl tail and terminal benzene that exhibit higher po-tency than the parent compound (our unpublished results). It isthought that the interactions formed between the aromaticsubstituents of the glycolipid and aromatic CD1d side-chainresidues contribute additional stability to the complex (Fig. 5).In summary, the studies here provide a structural basis forunderstanding the very different biological properties of syn-thetic and natural �-linked glycosphingolipids and will aid in thefuture design of glycolipid immunomodulators.

MethodsBiological Assays, Protein Expression, Purification, Crystallization, andSpacer Lipid Identification. Materials and methods in addition todata collection and refinement statistics for the CD1d-GSLcomplex are reported separately as Supporting Materials andMethods and Table 1, which are published as supporting infor-mation on the PNAS web site.

Structure Determination. Crystals were flash-cooled at a temper-ature of 100 K in mother liquor containing 25% glycerol.Diffraction data from a single crystal was collected at Beamline9.2 of the Stanford Synchrotron Radiation Laboratory, pro-cessed to 1.8 Å with MOSFLM (37) in spacegroup P222 (unit celldimensions: a � 42.3 Å; b � 107.7 Å; c � 110.7 Å) and scaledwith SCALA, as part of the CCP4 suite (38). The asymmetric unit

Fig. 5. Modeling of an �-GalCer analog in the hCD1d binding groove. Thepredicted conformation of a potent �-GalCer analog, in purple, wherethe N-hexacosanoyl tail has been replaced by a N-(8-phenyloctanoyl) tail. Theautomated docking (47) result is superimposed on the crystal structure ofhCD1d (gray ribbon) and �-GalCer (yellow) (22). Selected side-chains arelabeled for orientation and to show possible interactions with the terminalphenyl moiety. Note how the benzene ring sits among a cluster of CD1daromatic residues. (Inset) Structure of the N-(8-phenyloctanoyl) �-GalCeranalog.

3976 � www.pnas.org�cgi�doi�10.1073�pnas.0600285103 Wu et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020

contains one CD1d–lipid complex with an estimated solventcontent of 52.8%, based on a Matthews’ coefficient of 2.6 Å3�Da.Molecular replacement identified the space group as P212121 andwas carried out in CCP4 with the program MOLREP (39), with themCD1d–sulfatide structure minus the ligand (Protein Data BankID code 2AKR) as the search model, and resulted in a Rcryst of46.1% and a correlation coefficient of 0.53. Subsequent rigid-body refinement in REFMAC 5.2 resulted in an Rcryst of 39.1%. Theinitial refinement included several rounds of restrained refine-ment against the maximum likelihood target in REFMAC 5.2. Ata later stage of refinement, carbohydrates were built at two of thethree N-linked glycosylation sites. The refinement progress wasjudged by monitoring the Rfree for cross-validation (40). Themodel was rebuilt into �-weighted 2Fo � Fc and Fo � Fcdifference electron density maps with the program O (41). Watermolecules were assigned during refinement in REFMAC by usingthe water ARP module for �3� in a Fo � Fc map and retainedif they satisfied hydrogen-bonding criteria and returned 2Fo � Fcdensity �1� after refinement. Starting coordinates for GalA-

GSL ligand were obtained from the CD1d-short chain �-GalCer(PBS-25) structure (Protein Data Bank ID code 1Z5L) andmodified accordingly by using the molecular modeling systemINSIGHT II (Accelrys). The ligand library for REFMAC (42) wascreated with the Dundee PRODRG2 server (43). The CD1d–GSLstructure has a final Rcryst � 20.8% and Rfree � 24.3%. Thequality of the model (Table 1) was assessed with the programMOLPROBITY (44). The program PYMOL (45) was used to prepareFigs. 2–5. The program APBS (46) was used to calculate theelectrostatic surface potential in Fig. 4A.

We thank the staff of the Stanford Synchrotron Radiation LaboratoryBeamline 9.2 for support with data collection and Julie Vanhnasy andArchana Khurana for technical assistance. This study was supported byNational Institutes of Health Grants GM62116 (to I.A.W. and M.K.),CA58896 (to I.A.W.), AI45053 (to M.K.), CA52511 (to M.K.), andAI062015 (to B.A.S.); The Skaggs Institute for Chemical Biology(I.A.W., C.-H.W., D.W., and D.M.Z.); and the Cancer Research Institute(Y.K.).

1. Moody, D. B., Reinhold, B. B., Guy, M. R., Beckman, E. M., Frederique, D. E.,Furlong, S. T., Ye, S., Reinhold, V. N., Sieling, P. A., Modlin, R. L., et al. (1997)Science 278, 283–286.

2. Beckman, E. M., Porcelli, S. A., Morita, C. T., Behar, S. M., Furlong, S. T. &Brenner, M. B. (1994) Nature 372, 691–694.

3. Fischer, K., Scotet, E., Niemeyer, M., Koebernick, H., Zerrahn, J., Maillet, S.,Hurwitz, R., Kursar, M., Bonneville, M., Kaufmann, S. H. E. & Schaible, U. E.(2004) Proc. Natl. Acad. Sci. USA 101, 10685–10690.

4. Kawano, T., Cui, J., Koezuka, Y., Toura, I., Kaneko, Y., Motoki, K., Ueno, H.,Nakagawa, R., Sato, H., Kondo, E., Koseki, H. & Taniguchi, M. (1997) Science278, 1626–1629.

5. Moody, D. B., Young, D. C., Cheng, T.-Y., Rosat, J.-P., Roura-mir, C.,O’Connor, P. B., Zajonc, D. M., Walz, A., Miller, M. J., Levery, S. B., et al.(2004) Science 303, 527–531.

6. Shamshiev, A., Gober, H.-J., Donda, A., Mazorra, Z., Mori, L. & De Libero,G. (2002) J. Exp. Med. 195, 1013–1021.

7. Van Rhijn, I., Young, D. C., Im, J. S., Levery, S. B., Illarionov, P. A., Besra,G. S., Porcelli, S. A., Gumperz, J., Cheng, T.-Y. & Moody, D. B. (2004) Proc.Natl. Acad. Sci. USA 101, 13578–13583.

8. Zhou, D., Mattner, J., Cantu, C., III, Schrantz, N., Yin, N., Gao, Y., Sagiv, Y.,Hudspeth, K., Wu, Y.-P., Yamashita, T., Teneberg, S., et al. (2004) Science 306,1786–1789.

9. Wu, D. Y., Segal, N. H., Sidobre, S., Kronenberg, M. & Chapman, P. B. (2003)J. Exp. Med. 198, 173–181.

10. Moody, D. B., Ulrichs, T., Muhlecker, W., Young, D. C., Gurcha, S. S., Grant,E., Rosat, J.-P., Brenner, M. B., Costello, C. E., Besra, G. S. & Porcelli, S. A.(2000) Nature 404, 884–888.

11. Calabi, F., Jarvis, J. M., Martin, L. & Milstein, C. (1989) Eur. J. Immunol. 19,285–292.

12. Kronenberg, M. (2005) Annu. Rev. Immunol. 23, 877–900.13. Morita, M., Motoki, K., Akimoto, K., Natori, T., Sakai, T., Sawa, E., Yamaji,

K., Koezuka, Y., Kobayashi, E. & Fukushima, H. (1995) J. Med. Chem. 38,2176–2187.

14. Wu, D., Xing, G.-W., Poles, M. A., Horowitz, A., Kinjo, Y., Sullivan, B.,Bodmer-Narkevitch, V., Plettenburg, O., Kronenberg, M., Tsuji, M., Ho, D. D.& Wong, C.-H. (2005) Proc. Natl. Acad. Sci. USA 102, 1351–1356.

15. Kinjo, Y., Wu, D., Kim, G., Xing, G.-W., Poles, M. A., Ho, D. D., Tsuji, M.,Kawahara, K., Wong, C.-H. & Kronenberg, M. (2005) Nature 434, 520–525.

16. Mattner, J., DeBord, K. L., Ismail, N., Goff, R. D., Cantu, C., Zhou, D.,Saint-Mezard, P., Wang, V., Gao, Y., Yin, N., Hoebe, K., et al. (2005) Nature434, 525–529.

17. Sriram, V., Du, W., Gervay-Hague, J. & Brutkiewicz, R. (2005) Eur. J. Im-munol. 35, 1692–1701.

18. Zajonc, D. M., Elsliger, M. A., Teyton, L. & Wilson, I. A. (2003) Nat. Immunol.4, 808–815.

19. Zajonc, D. M., Crispin, M. D., Bowden, T. A., Young, D. C., Cheng, T. Y., Hu,J. D., Costello, C. E., Rudd, P. M., Dwek, R. A., Miller, M. J., et al. (2005)Immunity 22, 209–219.

20. Gadola, S. D., Zaccai, N. R., Harlos, K., Shepherd, D., Castro-Palomino, J. C.,Ritter, G., Schmidt, R. R., Jones, E. Y. & Cerundolo, V. (2002) Nat. Immunol.3, 721–726.

21. Batuwangala, T., Shepherd, D., Gadola, S. D., Gibson, K. J. C., Zaccai, N. R.,Fersht, A. R., Besra, G. S., Cerundolo, V. & Jones, E. Y. (2004) J. Immunol.172, 2382–2388.

22. Koch, M., Stronge, V. S., Shepherd, D., Gadola, S. D., Mathew, B., Ritter, G.,Fersht, A. R., Besra, G. S., Schmidt, R. R., Jones, E. Y. & Cerundolo, V. (2005)Nat. Immunol. 6, 819–826.

23. Zeng, Z.-H., Castano, A. R., Segelke, B. W., Stura, E. A., Peterson, P. A. &Wilson, I. A. (1997) Science 277, 339–345.

24. Giabbai, B., Sidobre, S., Crispin, M. D., Sanchez-Ruiz, Y., Bachi, A., Kronen-berg, M., Wilson, I. A. & Degano, M. (2005) J. Immunol. 175, 977–984.

25. Zajonc, D. M., Cantu, C., Mattner, J., Zhou, D., Savage, P. B., Bendelac, A.,Wilson, I. A. & Teyton, L. (2005) Nat. Immunol. 6, 810–818.

26. Zajonc, D. M., Maricic, I., Wu, D., Halder, R., Roy, K., Wong, C., Kumar, V.& Wilson, I. A. (2005) J. Exp. Med. 202, 1517–1526.

27. Moody, D. B., Zajonc, D. M. & Wilson, I. A. (2005) Nat. Rev. Immunol. 5,387–399.

28. Sidobre, S., Hammond, K. J. L., Benazet-Sidobre, L., Maltsev, S. D., Richard-son, S. K., Ndonye, R. M., Howell, A. R., Sakai, T., Besra, G. S., Porcelli, S. A.& Kronenberg, M. (2004) Proc. Natl. Acad. Sci. USA 101, 12254–12259.

29. Goff, R. D., Gao, Y., Mattner, J., Zhou, D., Yin, N., Cantu, C., Teyton, L.,Bendelac, A. & Savage, P. B. (2004) J. Am. Chem. Soc. 126, 13602–13603.

30. Brossay, L., Naidenko, O., Burdin, N., Matsuda, J., Sakai, T. & Kronenberg,M. (1998) J. Immunol. 161, 5124–5128.

31. Hong, S., Wilson, M. T., Serizawa, I., Wu, L., Singh, N., Naidenko, O. V., Miura,T., Haba, T., Scherer, D. C., Wei, J., et al. (2001) Nat. Med. 7, 1052–1056.

32. Singh, A. K., Wilson, M. T., Hong, S., Olivares-Villagomez, D., Du, C., Stanic,A. K., Joyce, S., Sriram, S., Koezuka, Y. & Van Kaer, L. (2001) J. Exp. Med.194, 1801–1811.

33. Kawano, T., Cui, J., Koezuka, Y., Toura, I., Kaneko, Y., Sato, H., Kondo, E.,Harada, M., Koseki, H., Nakayama, T., Tanaka, Y. & Taniguchi, M. (1998)Proc. Natl. Acad. Sci. USA 95, 5690–5693.

34. Burdin, N., Brossay, L., Degano, M., Iijima, H., Gui, M., Wilson, I. A. &Kronenberg, M. (2000) Proc. Natl. Acad. Sci. USA 97, 10156–10161.

35. Zhou, D., Cantu, C., III, Sagiv, Y., Schrantz, N., Kulkarni, A. B., Qi, X.,Mahuran, D. J., Morales, C. R., Grabowski, G. A., Benlagha, K., et al. (2004)Science 303, 523–527.

36. Marheineke, K., Grunewald, S., Christie, W. & Reilander, H. (1998) FEBS Lett.441, 49–52.

37. Otwinowski, Z. & Minor, W. (1997) Methods Enzymol. 326, 307–326.38. Collaborative Computational Project (1994) Acta Crystallogr. D 50, 760–763.39. Vagin, A. & Teplyakov, A. (1997) J. Appl. Cryst. 30, 1022–1025.40. Brunger, A. T. (1992) Nature 355, 472–475.41. Jones, T. A., Zou, J.-Y., Cowan, S. W. & Kjeldgaard, M. (1991) Acta Crystallogr.

A 47, 110–119.42. Murshudov, G. N., Vagin, A. A. & Dodson, E. J. (1997) Acta Crystallogr. D 53,

240–255.43. Schuttelkopf, A. W. & van Aalten, D. M. F. (2004) Acta Crystallogr. D 60,

1355–1363.44. Lovell, S. C., Davis, I. W., Arendall, W. B., III, de Bakker, P. I., Word, J. M.,

Prisant, M. G., Richardson, J. S. & Richardson, D. C. (2003) Proteins 50,437–450.

45. De Lano, W. (2002) PYMOL, A Molecular Graphics System (DeLano Scientific,San Carlos, CA).

46. Baker, N. A., Sept, D., Joseph, S., Holst, M. J. & McCammon, J. A. (2001) Proc.Natl. Acad. Sci. USA 98, 10037–10041.

47. Morris, G. M., Goodsell, D. S., Halliday, R. S., Huey, R., Hart, W. E., Belew,R. K. & Olson, A. J. (1998) J. Comput. Chem. 19, 1639–1662.

Wu et al. PNAS � March 14, 2006 � vol. 103 � no. 11 � 3977

IMM

UN

OLO

GY

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

Aug

ust 2

7, 2

020