Structural Basis for Lipid-Antigen Recognition in Avian Immunity

CD1 proteins are expressed on the surface of APCs and are known to bind and present lipid Ags to T cells (1). The CD1 lipid Ag presentation system thus represents a unique variation of the MHC-based paradigm established for peptide Ag recognition by T cells. Based on sequence similarity, CD1 proteins are thought to have evolved from the classical MHC class I (MHC I) family, but probably well after the emergence of the MHC I genes (2). The CD1 proteins thus fall into a broad family referred to as nonclassical MHC I. The MHC I and CD1 proteins share common structural features. Both proteins exist as cell-surface heterodimers. The H chain consists of an N-terminal ectodomain (the α1, α2, and α3 extracellular domains) followed by a transmembrane domain and a short cytoplasmic tail sequence at the C terminus. The α1 and α2 domains are tightly connected, each contributing four β-strands and a single α-helix to form the Ag binding site. The α3 Ig domain noncovalently interacts with the structural light chain β₂-microglobulin (β₂m) for stability of the complex (3). Once on the cell surface, the short C-terminal cytoplasmic tail allows some CD1 isoforms to bind adaptor proteins, leading to reinternalization and recycling through intracellular compartments (4, 5). During trafficking through various intracellular compartments, CD1 encounters various lipids that can be loaded into the binding pocket and subsequently presented at the cell surface for recognition by TCRs. The basic function of CD1, therefore, is to survey and display the lipid contents of APCs at their cell surface to T cells for immune recognition.

To enable binding of various ligands, the α1-α2 superdomain is the most diverse structural feature among the different CD1 isoforms. This allows the formation of a variety of unique Ag-binding pockets. However, there is little or no sequence polymorphism in the CD1 isoforms between individuals in the population (6, 7). This is in stark contrast to the highly polymorphic MHC proteins. Classical MHC I proteins posses a shallow and mostly polar binding site capable of binding a vast array of short peptide sequences. Hundreds of allelic variants of MHC I within the human population provide additional capacity for binding unique peptides. In contrast, there are only few CD1 proteins, which have a deep and narrow hydrophobic binding groove to permit anchoring of lipid acyl chains (8). The lack of polymorphism in CD1 may be the result of the low binding specificity of CD1 binding pockets. Binding of lipids to CD1 is largely promiscuous and mainly depends on the number and length of the lipid-acyl chains, whereas the polar head groups present on most Ags generally protrude out of the binding groove to the surface for recognition by TCRs (8–10). The various but fixed shapes of the CD1 pockets thus represent multiple evolutionary solutions for binding and anchoring lipids.

The human CD1 family is composed of five nonpolymorphic genes (CD1a, CD1b, CD1c, CD1d, and CD1e), whereas only CD1d exists in mice (11). Other mammals have variable numbers of CD1 genes reflecting duplication and deletion of particular isoforms within a particular species (12, 13). The chicken has two CD1 isoforms designated chCD1-1 and chCD1-2, but these do not exhibit high sequence similarity or expression patterns common to any particular prototypical mammalian CD1 isoform (14, 15). Current structural data on human CD1a, -b, and -d and mouse CD1d (17–19) reveal a basic dual-pocket architecture for the mammalian CD1 proteins in which lipid Ags possessing two alkyl chains are bound (8). In contrast, the first chicken CD1 structure to be solved, chCD1-2, revealed an apparently primordial single-pocket binding groove (20) that would only accommodate fatty acids or single alkyl chain lipids of up to 16 carbons in length. These data suggested that lipid Ags recognized by the avian CD1 proteins could be quite distinct from those previously described for mammals.

To address the questions raised by the unusual structure of chCD1-2, we solved the structure of chCD1-1 from the chicken (Gallus gallus). This is the second and final CD1 structure in chicken and also the first time in which the structures for all extant CD1 isoforms in a single species has been determined. Interestingly, analysis of the unique groove geometry of chCD1-1 in conjunction with in vitro binding experiments support a structure similar to mammalian CD1 in terms of overall volume and geometry, as well as the capacity to bind large dual chain lipids. These data provide new insight for the mechanisms of lipid recognition by CD1 proteins and have important implications for the potential function of these molecules in birds.

The chCD1-1 cDNA was derived from the cell line BM2. Total RNA was generated from growing cells using RNAeasy kit according to the manufacturer’s instructions (Qiagen, Hilden, Germany). The cDNA from BM2 was generated using Superscript first strand synthesis system for RT-PCR (Invitrogen, Carlsbad, CA). Primers were designed based on genomic sequence data for the chicken (21).

The forward primer was 5′-TTTTGAATTCTCAGATGTGCCCCCTCC-CCTCAT-3′ and the reverse primer was 5′-AAAAGGATCCGACACCA-TGCAGCCCCATGCCA-3′. These were used to amplify the chCD1-1 for direct cloning into pSEQ4 cloning vector (Invitrogen). The sequence was confirmed and matched with previously published data for the chCD1-1 open reading frame (14, 22). The ectodomain of chCD1-1 from Gallus gallus and the full-length chicken β₂m were cloned into the dual promoter baculovirus transfer vector pBACpHp10, with a similar strategy used for mouse CD1d (23).

Expression and purification was carried out in a manner similar to chCD1-2 (20). Briefly, Spodoptera frugiperda 9 cells were infected with a high titer (1–2 × 10⁸ PFU/ml) of chCD1-1–bearing baculovirus at a multiplicity of infection of three and kept at 27.5°C on a shaking platform (145 rpm) for 3 to 4 d. S. frugiperda 9 cells were spun down at 1000 × g for 10 min at 4°C, and the cell culture supernatant including secreted chCD1-1 protein was exchanged against PBS and concentrated to 300–400 ml by tangential flow-through filtration (Millipore, Bedford, MA). A total of 10 mM imidazole (pH 8.0) and 5 ml settled volume Ni-NTA beads (Novagen, Madison, WI) were added, and the solution was stirred for at least 4 h or overnight at 4°C. The Ni-NTA beads were collected using a Buchner funnel and were washed briefly with PBS. The Ni-NTA beads were transferred into an Econo column (Bio-Rad, Hercules, CA) and were washed with 50 ml of a solution containing 50 mM Tris-HCl (pH 8.0), 300 mM NaCl, and 10 mM imidazole. The protein was eluted with 30–50 ml of a solution containing 50 mM Tris-HCl (pH 8.0) and 250 mM imidazole. Eluted chCD1-1 was dialyzed overnight against 10 mM Tris-HCl (pH 8.0) and was purified by ion-exchange chromatography on MonoQ resin, using an AKTA Purifier (GE Healthcare, Piscataway, NJ). Eluted fractions were pooled, concentrated using Amicon Ultra filtration devices (Millipore), and purified to homogeneity by size-exclusion chromatography on a Superdex S200 16/60 column (GE Healthcare) followed by dilution to 20 mM HEPES (pH 7.4) and 50 mM NaCl and concentration of ∼5 mg/ml.

The basic protocol to visualize lipid loading onto CD1 molecules has been described (24). Briefly, chCD1-1 (in 20 mM HEPES [pH 7.4], 50 mM NaCl) was mixed with lipid (dissolved in DMSO) in ∼1:10 molar ratio and incubated overnight at 22°C. Typically, 4 μg protein was applied to the native isoelectric focusing (IEF) gel. Electrophoresis was performed using a PhastGel system (GE Healthcare) with IEF gels at a pH range of 5–8. Proteins in IEF gel were detected by staining with Coomassie R-350 (Sigma-Aldrich, St. Louis, MO). The lipids we have used include brain porcine sulfatides extract (Avanti Polar Lipids, Alabaster, AL), brain porcine extract of total gangliosides (Avanti Polar Lipids), ganglioside GT1b from bovine brain (Sigma-Aldrich), mycolic acid extract from human strain of Mycobacterium tuberculosis (Sigma-Aldrich), and a synthetic C20:2 α-galactosylceramide, as a control for an uncharged lipid.

Crystals were grown by the sitting-drop vapor diffusion method after mixing 0.5 μl protein with equal volume of the reservoir solution (20% PEG 4000, 0.2 M NH₄F). A single crystal was grown from 1–6 mo at 4°C and flash-cooled at 100 K in a mother liquor solution containing 20% glycerol for cryo-protection. X-ray diffraction data were collected on beamline 7-1 at the Stanford Synchrotron Radiation Laboratory (Menlo Park, CA). The data were indexed and integrated using MOSFLM (25) and merged and scaled with SCALA (26) from the CCP4 program package (27). Diffraction data statistics are presented in Table I.

The structure was solved by molecular replacement in PHASER (28) searching the data (46.0–2.2 Å) independently with two homology models constructed with SWISS-MODEL (29); one based on the structure of the H chain of chCD1-2 (Protein Data Bank [PDB] ID 3BDX) and the second based on the human β₂m also present in the 3BDX structure. Consistent with the intensities for the axial reflections (0,h,0), (0,k,0), and (0,l,0) being systematically absent, respectively, when h, k, and l = 2n + 1, the molecular replacement solution confirmed the space group P2₁2₁2₁ and located two CD1/β₂m complexes in the asymmetric unit (ASU) with an initial R-factor of 47.7%. Several cycles of restrained refinement using REFMAC5.5 (30) maintaining medium noncrystallographic symmetry restraints between the H chains and L chains along with manual model editing using COOT (31) resulted in R = 25.4% and Free-R = 29.1%. At this point, very clear density was visible in the lipid-binding grooves of both CD1 molecules of the ASU in both Fo-Fc and 2Fo-Fc Fourier differences maps. Fitting of one 45-carbon and another 16-carbon alkyl chain in each CD1 molecule reduced the Free-R by 0.6% and modeling of sugars and additional water molecules yielded R = 22.8% and Free-R = 27.5%. TLS refinement (two groups for each CD1 molecule, residues 6–179 and residues 180–279, while keeping each β₂m as a single moving group) reduced the Free-R by an additional 1.6%. After a few cycles of refinement and manual model editing, the R and Free R factors converged to 21.6% and 25.1%, respectively (Table I). Ramachandran plot analysis using PROCHECK (31) shows that 92.9% of the residues are in the core, 7.1% in the allowed, and 0.0% and 0.0% in the generously allowed and disallowed regions.

All structural images were produced with PyMol (33), whereas the cavities surface calculations were performed with HOLLOW (34) for presentation with PyMol. Pocket volume calculations were done with GRASP (35).

The structure of the chCD1-1 was determined at 2.2 Å resolution (Fig. 1, Table I). The ASU contains two virtually identical CD1-β₂m complexes, with a Cα root mean square deviation (RMSD) of 0.27 Å for the CD1 H chains (A and B), and an RMSD of 0.10 Å for the light β₂m chains (C and D). Therefore, we will refer to one complex throughout the manuscript. There are 328 refined water molecules and 6 N-acetylglucosamine molecules (three per CD1 molecule). Similar to chCD1-2 (19), the overall backbone structure of chCD1-1 is similar to the mammalian CD1 structures and has been reviewed elsewhere (10, 35). Structural alignment, using SSM (37), of the Ag-binding domain (α1-α2) of chCD1-1 with the corresponding domains of CD1 structures yielded the following RMSD values (for the Cα backbone atoms): 1.31 Å with chCD1-2 (3DBX), 1.45 Å with hCD1d (1ZT4), 1.51 Å with mCD1d (2FIK), 1.67 Å with hCD1a (1ONQ), and 1.51 Å with hCD1b (1UQS). Considering the errors on the coordinates of some of these structures, the observed differences in the Cα RMSD values may not be significant. In contrast, more significant differences are observed in the primary structure of these domains in which chCD1-1 is found slightly more similar to the mammalian CD1, with the highest identity (27.5%) with CD1b and highest similarity (50.0%) with CD1d (Fig. 2A) (38). Moreover, the corresponding phylogenetic analysis suggests that chCD1-2 is the most distant sequence (Fig. 2B). Interestingly, a conserved (Fig. 2A) disulfide bond anchoring the α2 helix to the base of the β-sheet platform exists in chCD1-1 (Cys98–Cys163) and is missing in chCD1-2, which lacks a cysteine residue on its α2 at F170 (Supplemental Fig. 1). This is likely a derived feature of chCD1-2 because virtually all CD1 (with the exception of the mouse CD1d2 isoform) and MHC I have conserved cysteine at this position.

Taken together, the overall fold is quite similar for all the CD1 structures, and the differences in the side chains account for the various binding-groove topologies that determine the ligand specificity. Indeed, the chCD1-1 protein, which has a slightly higher sequence similarity to CD1b and CD1d, shows a deep hydrophobic dual-pocket (A’ and F’) Ag-binding groove (Fig. 3).

The structure of chCD1-1 shows clear and continuous electron density in the binding groove formed between the α1 and α2 domains: a long tube of density in the A’ pocket and a shorter one in the F’ pocket (Fig. 3A, Supplemental Fig. 2). Because no exogenous ligand was added during protein purification and crystallization, it must have been acquired from the S. frugiperda 9 insect cells during expression. Based on their length and the hydrophobic nature of their surroundings, we have refined two alkyl chains (45 and 16 carbons long) per CD1 heterodimer. Although each refined ligand is accommodated in a separate deep pocket, they emerge from a very close position at the top of the groove between the A’ and F’ portals (Fig. 3B, 3C), suggesting they are part of a single dual-chain lipid. Although very clear electron density is present for the hydrocarbon chains inside the A’ and F’ pockets, no significant density exists for a connecting head group, thus precluding structural identification of this presumably dual-chain lipid. Regardless of the exact nature of the ligand, we can learn from the length and the position of the refined alkyl chains about potential lipid classes that can be bound by chCD1-1. For example, we can clearly see that the ligand in the A’ pocket is ∼3 times longer than the single-chain ligand (C16) observed in chCD1-2 structure.

chCD1-1 has an A’ pocket similar to mammalian CD1b and CD1d but tilted inside the binding groove. However, unlike other CD1 molecules, chCD1-1 has a wide entrance to the A’ pocket, which is connected to a hydrophobic cleft referred to in this paper as the A’ cleft (Fig. 3C). The refined alkyl chain in the A’ pocket exits the A’ portal, and its seven terminal carbons are stabilized in the A’ cleft by nonpolar van-der Waals interactions. Next to the F’ portal, we found an additional unique hydrophobic cleft (F’ cleft, Fig. 3). The F’ cleft also shows clear electron density for a bound ligand, similar to that seen in the A’ cleft. However, we have not modeled it, because it is not connected to the density in the F’ or A’ pockets and is also too distant from the projected branching point of both alkyl chains to be the result of a bound lipid head group (Fig. 3A). The lipid head group likely protrudes from the center of the chCD1-1 binding groove, where residues along the α1-helix (Ser68, Ser71, and Asp75) are ideally positioned to form polar interactions (Fig. 3C). In mammalian CD1, Arg79 (homologous to Arg 78 of chCD1-1) is conserved and important in mediating interaction with the various lipid head groups and/or the cognate TCR (9, 39, 40). However, in chCD1-1, both Arg78 and Arg74, which is in fact closer to the putative head group position (Fig. 3C), point away from the binding groove and are likely not involved in head group binding. An alternative position for a lipid head group, which would be unusual but one that we cannot exclude, could be mediated by two aspartic residues (D61 and D165; Fig. 3C) that point upward from the A’ cleft.

For better visualization of the CD1 groove, the binding pockets were extracted from the chCD1-1 structure and illustrated as a molecular surface (Fig. 4). The bulky A’ pocket of chCD1-1 is formed by many hydrophobic amino acid side chains to accommodate the long 45-carbon alkyl chain (Fig. 4A). As seen for the mammalian CD1a, -b, and -d, the A’ pocket has a doughnut shape, but it is more prominent in chCD1-1, allowing the ligand to complete a whole circle around the central A’ pole formed by residues I69 and L11. This chain originates from the opening between the A’ and F’ pockets (a likely position for a putative head group) and reaches the bottom of the A’ pocket and then loops back out and over to the A’ cleft, formed primarily by the surface residues L57, L169, and I166. The wide A’ portal permits long alkyl chains to enter the A’ pocket very close to their exit site. This exit port, in combination with the A’ cleft of chCD1-1, represents a novel approach to accommodating large alkyl chains that would exceed the size of a typical A’ pocket. This is in contrast to the A’ pockets of mammalian CD1a and CD1d, which are restricted in size and have a defined terminus (8).

Although we have modeled only a simple alkyl chain to fit the loop-shaped electron density, the shape of the Ag-binding groove around that loop suggests it could accommodate lipid Ags with substitutions on their long acyl chain that would fit in three observed empty cavities inside the A’ pocket (indicated as S1, S2, and S3 in Fig. 4B). These cavities appear equally spaced from each other along the ligand, namely seven to eight carbons between S1 and S2 and seven to eight carbons between S2 and S3. In the current structure, we see no electron density in these cavities, but their shape and size suggest they could accommodate small substitutions such as methyl, methoxyl, cyclopropyl, or keto groups, which are, for example, commonly found at various positions along the long acyl chain of mycolic acids (41).

Comparison of the lipid-binding grooves of chCD1-1 and chCD1-2 shows a clear difference in their size and shape (Fig. 5A). The binding groove of chCD1-1 consists of two distinct pockets (A’ and F’) whose positions flank that of the single-chain binding groove of chCD1-2. It is three times the volume of chCD1-2 (1440 versus 470 ų) because it is more expanded toward the N-terminus of the α1 helix (large A’ pocket). The individual A’ and F’ pockets of chCD1-1, separated by amino acids Y72, I96, and F110 (Fig. 4A), are collapsed in the chCD1-2 structures (Fig. 5A). In fact, both the shape and the volume of chCD1-1 appear more similar to those of the mammalian CD1 proteins (2200 ų for hCD1b; 1280 ų for hCD1a; 1650 ų for hCD1d; Fig. 5B), as also suggested by its slightly higher sequence similarity with human CD1b and CD1d (Fig. 2). The overlay of the groove of chCD1-1 and the mammalian CD1a, -b, and -d shows that it is very similar in shape to CD1d with distinct A’ and F’ pockets. However, the opening to the groove from the top is wider for chCD1-1 because its A’ pocket and A’ cleft are united (Fig. 5B), thereby allowing long hydrophobic ligands to protrude out of the deep A’ pocket to the surface of the protein (Fig. 3). Moreover, it can be seen that the maximum width of the groove opening for chCD1-1 is much larger (∼31 Å) than those measured for all other CD1 proteins, with CD1d having the closest width of ∼24 Å, and the most narrow opening of ∼12 Å exists for chCD1-2 (Fig. 5, Supplemental Fig. 3). However, accessibility of the hCD1a groove is the lowest among the mammalian structures, due to narrow neck of its F’ portal (9 Å) (Fig. 5B).

The large lipid-binding groove described above for chCD1-1 encouraged us to test its capacity to bind natural and synthetic lipids recognized by other CD1 proteins (8). For this purpose, we used a loading assay in which recombinant chCD1-1 protein was incubated with lipids of various length and charge values. We then monitored the change in mobility due to the additional negative charge of the lipid-protein complex by native IEF. The lipids we used posses varying negative charge values at pH 7.0 (Fig. 6A): sulfatides (sulfoglycosphingolipid, single charge), GM₃ (monosialoganglioside, single charge), GT_1b (trisialoganglioside, thee charges), total gangliosides extract (varying charge values), C20:2 α-galactosylceramide (a synthetic monoglycosylceramide, uncharged), and mycolic acid (single charge). Although the purified chCD1-1 migrates as two bands on the IEF gel (Fig. 6B, left lane), similar to what has been also observed for mouse CD1d (24), the results demonstrate partial loading with sulfatides and GM₃, but complete loading with the trisialoganglioside GT1b, as indicated by the changes of the protein mobility toward the positive anode (Fig. 6B). On the other hand, the mixture with the neutral lipid α-galactosylceramide shows a minor and more diffuse change in mobility, likely due to a subtle change in the isoelectric point but not as a result of a discrete change in charge. These results suggest that the endogenously bound ligand/lipid is likely not charged under the experimental conditions, analogous to the endogenous ligands found in both CD1d (23, 24) and CD1b (42) when expressed in insect cells. Moreover, the observed chCD1-1 double band is a result of protein heterogeneity rather than a result of binding lipids with different charges. Intriguingly, even mycolic acid (with 60–90 carbons), though inefficient, shows a noticeable mobility shift indicative of loading onto chCD1-1 at neutral pH. Loading was abrogated at more acidic pH values comparable to what would be encountered in lysosomes (Fig. 6C). These data support the capacity of the chCD1-1 to bind dual-chain lipids comparable to those identified for mammalian CD1.

Although the structural data for chCD1-1 does not reveal electron density for a putative head group for connecting the two refined alkyl chains, we can make certain predictions about the size and nature of possible ligands based on the size and shape of the individual pockets and the bound endogenous ligands. The clear presence of continuous electron density for the endogenous ligands, 45 carbons in length in the A’ pocket and 16 carbons in the F’ pocket, and the detectable in vitro loading of mycolic acid (Fig. 6C) stimulated us to model a mycolate lipid inside the chCD1-1 groove. Such long lipids are α-branched, β-hydroxy fatty acids, commonly found in mycobacteria and well-defined ligands for human CD1b (18, 43). CD1b is hitherto the only CD1 isoform known to bind mycolic acid and its derivative glucose monomycolate (GMM) due to its large binding groove (2200 ų) and unique pocket architecture (A’, C’, and F’ pockets and T’ tunnel). The crystal structure of CD1b in complex with GMM shows how the long meromycolate chain fits inside its large hydrophobic groove (44). We have used the coordinates for GMM from the CD1b structure (PDB ID 1UQS) as a starting model and fitted the 17-carbon α-branched alkyl chain in the F’ pocket and a longer 50-carbon β-hydroxy (meromycolate) chain in the elongated A’ pocket of chCD1-1. To improve the fit of this model with the observed electron density in the binding groove of chCD1-1, we refined it against the x-ray data. The resultant electron density (not shown) supported the positions of the acyl chain of GMM very well, but as expected, no electron density was observed for the polar carbohydrate head-group of GMM. Therefore, the R-free value increased by 0.1% compared with refinement with the endogenously visible ligands (Fig. 3).

Although GMM is not likely to exist in current structure of chCD1-1, the above modeling can be used to speculate about the putative binding of mycolic acids in chCD1-1 groove compared with that of CD1b. Like the path the meromycolate chain assumes in the CD1b structure (occupying the A’ pocket, the connecting T’ tunnel, and part of the F’ pocket), it occupies only the lengthy A’ pocket of chCD1-1, protrudes out of the wide A’ portal, and binds at the hydrophobic A’ cleft (Fig. 7). Meromycolate chains with up to 40 carbons could be accommodated inside the A’ pocket, whereas 50-carbon chains could be fitted in the A’ pocket and A’ cleft together. Thus, the A’ cleft could act as an escape hatch for ligands exceeding the maximum size of the A’ pocket, similar to the C’ portal in human CD1b (18, 44).

The chicken genome has two genes that code for clear homologues of CD1. Examination of the primary sequence data revealed some motifs common to mammal CD1 proteins (14). However, fine tertiary structure details cannot be predicted from sequence alone, particularly for the Ag-binding pockets that are essential for CD1 function. We have now determined the crystal structure of the chCD1-1 protein from chicken, which, together with our previous data on the structure of chCD1-2, provides a complete set of CD1 structures from a single species. This represents the first such CD1 data set and provides a unique opportunity for comparative analysis.

Our previous data on the chCD1-2 revealed a structure of a single-pocket binding-groove with the maximum capacity for a single C16 alkyl chain (20). This suggested that the avian CD1 may have evolved to accommodate a distinct set of Ags, because most of the CD1 ligands from mammals are too large to fit in the observed binding groove. The structure of chCD1-1 allows us to address this question.

Although unique in its geometry, the chCD1-1 Ag-binding groove is clearly more similar to mammalian CD1 proteins, with a dual-pocket architecture formed by distinct A’ and F’ pockets and additional novel A’ and F’ clefts. The continuous circular shape of the A’ pocket can accommodate longer alkyl chains than those seen so far inside the A’ pockets by allowing a potential ligand to coil around the interior cavity and uniquely extend back outward to the surface A’ binding cleft. The electron density in the F’ cleft suggests that it too could potentially participate in the coordination of extended alkyl chains. Because these hydrophobic clefts are located externally close to the binding-groove opening, it would be interesting to study their possible role in providing additional specificity by allowing lipid alkyl chains or other hydrophobic groups to interact with potential TCRs. The wide opening to the A’ pocket serves both as an entry and exit point for the A’ ligand in the chCD1-1 structure. However, it could serve equally well as a double entry port for putative triacyl lipids having one acyl chain in the F’ pocket and two bound inside the A’ pocket.

The contrast in the size and shape of the Ag-binding pockets of chCD1-1 and chCD1-2 raises further questions as to the evolutionary scenario leading to the two isoforms. The two most likely scenarios are that the chCD1-2 with a smaller groove represents something closer to a primordial state for the avian CD1 from which chCD1-1 was duplicated and expanded. Conversely, chCD1-1 may be closer to the ancestral form, and chCD1-2 represents a degenerate/specialized isoform that lost the expanded A’ pocket and the entire F’ pocket as well. At this stage, we cannot distinguish between these two possible scenarios. Examination of sequence data from other avian species and more basal reptilian species may reveal and provide additional insight into the history of these genes.

Thus far, functional data on CD1 in birds has been difficult to obtain. This is due primarily to the inability to obtain long-term cultures of T cells in vitro for chickens or other birds. However, the structural data we have obtained may help focus these efforts. The ability to model GMM into the chCD1-1 structure together with the binding data suggest the possibility of using this Ag as an in vivo stimulator of CD1 restricted T cells. Mycobacterium avium bacteria is a well-known pathogen in chickens that causes a lethal disseminated granulomatous disease reminiscent of tuberculosis in mammals. Previous data has shown that M. avium and M. tuberculosis are able to produce similar or identical lipid Ags capable of presentation by human CD1 to specific T cells (43). Clearly, many open questions remain regarding the role of CD1-mediated lipid recognition in the avian immune system, but the current structural picture of a lipid Ag-presenting system in birds could pave the way for further studies in that direction. Do CD1 and lipid-reactive T cells exist in the bird, and, if so, what is their function? Ags can now be selected based on the structural compatibility with the chCD1 binding grooves and further analyzed in functional studies.

We thank the Stanford Synchrotron Radiation Laboratory staff for their support during x-ray data collection. We also thank Gurdyal S. Besra and Steven Porcelli for providing the C20:2 α-galactosylceramide and Enrico Girardi for critically reading our manuscript.

Disclosures The authors have no financial conflicts of interest.