Report

T Cell Activation by Lipopeptide Antigens

See allHide authors and affiliations

Science  23 Jan 2004:
Vol. 303, Issue 5657, pp. 527-531
DOI: 10.1126/science.1089353

This article has a correction. Please see:

Abstract

Unlike major histocompatibility proteins, which bind peptides, CD1 proteins display lipid antigens to T cells. Here, we report that CD1a presents a family of previously unknown lipopeptides from Mycobacterium tuberculosis, named didehydroxymycobactins because of their structural relation to mycobactin siderophores. T cell activation was mediated by the αβ T cell receptors and was specific for structure of the acyl and peptidic components of these antigens. These studies identify a means of intracellular pathogen detection and identify lipopeptides as a biochemical class of antigens for T cells, which, like conventional peptides, have a potential for marked structural diversity.

Dendritic cells (DCs) play a central role in stimulating T lymphocyte responses (1). Myeloid DCs and Langerhans cells of the skin can be identified in tissues on the basis of their abundant and specific expression of the major histocompatability complex (MHC) class I–like protein CD1a, among other markers (2, 3). The function of CD1a involves the presentation of self and foreign antigens to T cells (46). The general molecular mechanism of antigen presentation by CD1 involves insertion of the alkyl chains of amphipathic lipids into a deep hydrophobic groove formed by the α1 and α2 domains of the CD1 heavy chain. This positions carbohydrate or other hydrophilic elements of the antigen on the outer surface of CD1, where they can directly contact T cell receptors (TCR) (79). CD1a-restricted T cells recognize mycobacteria-infected cells and have antibacterial effects, suggesting a possible role in host defense (5, 10). In addition, CD1a is normally expressed on DCs early in their maturation program, so CD1a-mediated T cell activation events that occur at the surface of infected DCs are well positioned to influence DC maturation (2, 11). However, the molecular structures of foreign antigens presented by CD1a have not been determined.

To identify candidate antigens presented by CD1a, TCR α and β chains from the CD1a-restricted and mycobacteria-specific T cell line CD8-2 were cloned and transfected into J.RT3-T3.5 T lymphoblastoid cells to generate a reporter cell line, J.RT3.CD8-2 (5). Along with untransfected J.RT3 cells, this cell line was used to screen for antigens in chromatographic fractions generated from complex mixtures of compounds found in M. tuberculosis cell walls. Antigenic factors capable of J.RT3.CD8-2 activation were efficiently extracted from whole mycobacteria by using chloroform: methanol (2:1), suggesting that the antigens were lipids, which were not covalently bound to the arabinogalactan complex of the mycobacterial cell wall (12). Elution of the stimulatory lipids from silica columns in polar solvents, such as methanol, indicated that the antigens displayed characteristics of polar lipids. Purification by high-performance liquid chromatography (HPLC) led to the isolation of a fraction that contained a set of structurally related compounds, which, by mass spectrometry analysis, yielded a prominent [M+H]+ ion at mass/charge ratio (m/z) 838.5. This antigenic compound was initially named 838 on the basis of its nominal mass.

C1R lymphoblastoid cells transfected with human CD1a, but not CD1b, CD1c, or CD1d, were able to present 838 to J.RT3.CD8-2 (Fig. 1A). T cell activation was not seen with other known CD1-presented lipid antigens such as mycolic acid, glucose monomycolate, or mannosyl phosphoisoprenoids (Fig. 1B). Also, 838 did not activate polyclonal T cells or J.RT3 transfectants expressing TCRs that are specific for lipid antigens presented by CD1b or CD1c (Fig. 1B) (1315). Together, these studies indicated that 838 did not function as a mitogen, but instead activated cells through CD1a-restricted recognition mediated by clonally variable regions of the TCR α and β chains.

Fig. 1.

T cell activation by antigen 838 is CD1a and TCR dependent. (A) C1R B lymphoblastoid cells transfected with cDNA constructs encoding the indicated human CD1 isoform or an empty vector (mock) were coincubated overnight with M. tuberculosis lipid fractions containing the 838 antigen. After adding J.RT3.CD8-2 cells at a 1:1 ratio and subjecting cells to a 1 min centrifugation (spin) to initiate cell contact, we measured calcium flux by flow cytometry of J.RT3 T cells labeled with calcium-sensitive fluorescent dyes Fura-red and Fluo 4 (32). (B) J.RT3 T lymphoblastoid cells were transfected with the TCR α and β chains cloned from the human T cell lines CD8-2, LDN5 (CD1b-restricted and glucose monomycolate specific), or CD8-1 (CD1c-restricted and mannosyl phosphoisoprenoid-specific) (14, 15, 33). Using monocyte-derived dendritic cells as antigen-presenting cells, we added monoclonal antibody to CD3 (OKT3) (30 ng/ml) or lipid antigens at or above concentrations that were previously known to be activating for T cells. These antigens were 838 (1 μM), mycolic acid (MA) (30 μM), glucose monomycolate (GMM) (30 μM), or mannosyl phosphoisoprenoid (MPI) (one-tenth dilution of fraction purified by 2D thin-layer chromatography). Interleukin-2 (IL-2) release was measured with the HT-2 bioassay (32).

Further insights into the molecular composition of 838 were obtained through high-resolution mass measurements. Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) demonstrated the [M+H]+ to be m/z 838.5684, matching the value calculated for C47H76N5O8 (838.5688) (table S1). Tandem mass spectrometry (MS/MS) yielded product ions at m/z 642 and m/z 197, which were each 16 u (the mass of oxygen) smaller than the previously described mycobactic acid and cobactin fragments of mycobactin, a known mycobacterial lipopeptide with iron-scavenging properties (16, 17). Therefore, we named the antigen didehydroxymycobactin (DDM-838). Identification of MS/MS products at m/z 727 and 84 indicated that the hydroxylysines found in mycobactin were substituted by lysines in the proposed structure for DDM (17). The presence of lysine in DDM was confirmed by gas chromatography–mass spectrometry (GC-MS) of acid hydrolysis products. Unexpectedly, GC-MS detected an uncommon amino acid, α-methyl serine, instead of serine and threonine, which are present in most mycobacterial mycobactins (12, 16, 18). The identity of α-methyl serine as a component of DDM was confirmed by the nuclear magnetic resonance (NMR) spectrum of DDM (Fig. 2C), as well as by differences in the MS/MS spectra of M. tuberculosis DDM compared with synthetic DDM homologs that contained threonine (DDM-thr) (19). These studies establish the structure of the CD1a-presented antigens produced by M. tuberculosis and show that they conform to a general structure in Fig. 2B, which is composed of salicylic acid, α-methyl serine, acylated lysine, hydroxybutyrate, and cyclized lysine (Fig. 2B).

Fig. 2.

Structure of the CD1a-presented antigen, DDM-838. (A and B) The DDM structure was deduced from the MS/MS spectrum of [M+H]+ m/z 838. Major complementary ions derived from DDM-838 were present at m/z 624 and m/z 215, which differed by 16 u from mycobactic acid and cobactin fragments of mycobactin (inset) (16, 19). Simple one-bond fragmentations are indicated by dashed lines and assume hydrogen transfer to nitrogen and oxygen during amide and ester cleavage. Multistep fragmentations explain the ions at m/z 376, m/z 350, m/z 333, m/z 332, and m/z 84. (C) Downfield regions of one-dimensional proton NMR spectra of DMM-838 (upper) and DDM-840 (lower) show resonances that are assigned as aromatic (HΦ) or other protons (Ha-n), as denoted in (B). The olefinic protons Hd and Hf are present in DDM-838, but not in DDM 840, and are coupled to each other with a J constant of 12.1 Hz. This is consistent with a double bond, which is likely present at C2–3, because the doublet corresponding to Hf is directly coupled to only one proton, Hd. Hd is directly coupled to the two HI and is shifted downfield of its olefinic partner, Hf, as expected for a CEmbedded ImageC-CEmbedded ImageO system. Further supporting this conclusion, Hk protons were also shifted slightly for DDM-838 relative to DDM-840 as a result of the influence of the olefin, which provides evidence for conjugation with the carboxyl group. The identification of α-methyl serine was based in part on the observations that Hh protons are only split by each other, suggesting that they are geminal protons in an oxazoline ring adjacent to a quaternary carbon. The acyl chain gave a typical terminal methyl triplet at 0.880 parts per million (ppm) and a large signal centered at 1.257 ppm, which obscures the signal from the methyl protons of the methyl serine associated with the oxazoline ring. Two obvious impurity peaks (6.97 and 3.775 ppm, upper panel), as well as spinning sidebands from the CHCl3, have been deleted for clarity. Unassigned resonances, probably impurities, are denoted with an asterisk. Two-dimensional 1H-1H data were also obtained and corroborate all implied adjacent proton relations (not shown). (D) This biosynthetic scheme is based on the previously proposed pathway of nonribosomal peptide synthesis (16, 2022) and on the discovery of DDM as a didehydroxy form of mycobactin, which does not avidly bind iron. The DDM structure suggests that lysines are incorporated in the peptide, which is terminated by cyclization of the terminal lysine, acylated and hydroxylated on lysine residues as depicted (black arrows). This process would be an interesting variation of previously described siderophore biosyntheses in which amino hydroxylation occurs after acylation, and it remains possible that mycobactin could also be converted to DDM. The addition of hydroxamate moieties to the lysine residues adds two sites that participate in high-affinity interactions with ferric iron (Fe3+), which facilitates the capture of iron from the host and subsequent transfer to the bacterium (16).

Didehydroxymycobactin most likely functions as an intermediate in mycobactin synthesis. The mycobactin locus in M. tuberculosis encodes mycobactin synthase genes, MbtA to MbtJ, which function as a nonribosomal peptide synthesis pathway (20). Previously proposed schemes of mycobactin biosynthesis have emphasized a likely role of MbtA, MbtB, MbtC, MbtD, MbtE, and MbtF in activating the salicyl moiety and synthesizing the peptide (Fig. 2D) (19, 21, 22). The final steps in mycobactin synthesis were thought to involve peptide termination by intramolecular attack of the lysine side chain on its carboxy terminus and by acylation and hydroxylation of the two lysine residues (Fig. 2D). Although the precise order in which these terminal steps occur is not known, the discovery of DDM clarifies the order by suggesting that the lysines are incorporated into the peptide and subsequently hydroxylated to yield mycobactins (Fig. 2D). The two hydroxyl moieties, which are present in mycobactin but absent in DDM, form two sites, which mediate high-affinity (∼10–26 M) binding to iron (16). Consistent with the predicted roles of these hydroxyl groups in iron binding, we found that mycobactin was detected in the iron-bound form solely as [M+Fe-2H]+ at m/z 923.5, whereas DDM was detected solely in the unbound form as a proton adduct of m/z 838.6.

During infection, bacteria must obtain iron from host stores to support a variety of reduction-oxidation reactions necessary for normal bacterial metabolism. Bacteria scavenge iron by producing siderophores, which bind iron with high affinity at or near the bacteria-host interface and deliver iron to the bacterium (23). Mycobacteria produce mycobactin and related siderophores, whose synthesis is triggered by derepression of mycobactin synthase genes during growth in low-iron conditions (24). This process normally occurs during growth in host cells and is known to be necessary for M. tuberculosis survival within human macrophages (25). These considerations suggested that DDM might be synthesized as an intermediate in mycobactin production during intracellular infection. We found that DDM-specific T cell activation occurs efficiently in cells infected with five bacteria per cell of M. tuberculosis, whereas M. tuberculosis grown in complete medium requires the equivalent of 1011 bacteria to produce detectable levels of DDM, and in some cases does not produce DDM at all (Fig. 3A). Because CD1a-restricted T cells are able to kill mycobacteria-infected cells (26), CD1a presentation of DDM may represent an early warning system for intracellular pathogen recognition, whereby bacterial metabolites, which are necessary for adapting to intracellular growth, result in T cell activation.

Fig. 3.

The T cell response is specific for the structures of the lipid and peptide moieties of DDM-838. (A) IL-2 secretion by CD8-2/J.RT-3 above background levels found in resting JRT-3 cells was measured after coculture with monocyte-derived DCs that were infected with five bacteria per cell of M. tuberculosis H37Rv for 2 days. (B) After polar lipids from M. tuberculosis lipids were initially purified by precipitation in cold acetone and normal phase silica chromatography, an antigenic fraction containing DDM-838 was further resolved by HPLC on a C18 column. Individual lipid species were detected by ultraviolet absorbance (solid line) and mass spectrometry (not shown). Fractions were collected at 15-s intervals, dried, resuspended in media, and tested for their ability to stimulate IL-2 release by J.RT3.CD8-2. (C) Lipids corresponding to peaks A, B, E, and F were subjected to further HPLC until resolved to homogeneity and examined by MS/MS. Compounds were named DDM-X, where X is the nominal mass of the [M+H]+ ion. MS/MS analysis demonstrated that material in peaks A, B, E, and F all yielded a product ion at m/z 546 (Fig. 2B), which indicates that the mass of the acyl chain, not the peptide, differed among these compounds, as shown in red. These compounds and mycobactin were tested for their ability to stimulate IL-2 release by J.RT3.CD8-2. (D) We prepared a molecular model of the CD1a-DDM complex using the programs MolScript and Raster3D after superimposing DDM-838 on the coordinates of CD1a (8, 34). The molecular surface of the CD1a binding pocket (bottom panel) was calculated in GRASP and is shown as a transparent gray surface that is predicted to interact with the carbon (yellow), nitrogen (blue), and oxygen (red) atoms of DDM (35). The volume and curvature of the hydrophobic CD1a groove correlate well with the size and shape of the optimal C20:1 (cis) acyl chain found in DDM-838, predicting that the peptidic portion of the antigen lies at the junction of the A′ and F′ pockets, near the α-helical surface of CD1a.

To analyze more precisely the role of DDM structure in mediating T cell activation, we used HPLC to isolate several natural lipopeptides from M. tuberculosis. In addition to DDM-838 (Fig. 3B, peak E), M. tuberculosis produced a series of structurally related compounds (peaks A to D and F). Although the abundance, as determined by ultraviolet absorbance, correlated well with the measured T cell response for some compounds (Fig. 3B, peaks A, C, D, and E), others showed weak (peak F) or no (peak B) ability to activate T cells. Compounds in peaks A, B, E, and F were purified and found to have nominal masses of 810, 812, 838, and 840, respectively. MS/MS demonstrated that DDM-838 bore a C20:1 fatty acid, whereas DDM-840, DDM-812, and DDM-810 contained peptides of equal masses but were substituted with differing fatty acids, C20:0, C18:0, and C18:1, respectively (Figs. 2B and 3C). A comparison of NMR spectra for DDM-838 and DDM-840 indicated that the unsaturation of the fatty acyl component of DDM-838 was at the C2–3 position, likely in a cis conformation (Fig. 2C). Whereas DDM-838 gave the most potent T cell response, homologous lipopetides that had shorter or saturated fatty acids were substantially less stimulatory (Fig. 3C). Natural mycobactins were not recognized, suggesting that the hydroxylation of the lysine residues prevents T cell recognition (Fig. 3C). Also, lipid fractions containing mycobactic acid, which corresponds to a truncated lipopeptide lacking the butyric acid–lysine moiety (Fig. 2B, m/z 642 fragment), were recognized weakly or not at all (12). Thus, the T cell response was specific for the structure of the peptide and the length and saturation state of the fatty acyl chain.

These studies identify lipopeptides as a biochemical class of antigens for T cells, which share structural features of MHC-presented peptides and CD1-presented glycolipids. The crystal structure of CD1a shows that the antigen-binding groove is composed of an F′ pocket, which is wide, largely exposed, and broadly contiguous with the predicted TCR contact surface (8). The A′ pocket is largely hydrophobic, with no obvious polar groups that could hydrogen bond with the peptidic portion of DDM in the way that MHC molecules bind to peptides. Also, the A′ pocket is narrow and terminates deep within the CD1a structure, so that it may act as a ruler to select out acyl chains of a particular length. A molecular model shows that the C20:1 fatty acyl moiety could fit well within the A′ pocket, positioning the peptide backbone at the broad junction of the A′ and F′ pockets, so that it would be available for contact with the TCR (Fig. 3D). Although the orientation of the peptidic moiety cannot be predicted precisely, the only polar residues in the binding groove are located at the A′-F′ junction, so it seems plausible that the same residues that are known to bind the sulfogalactosyl moiety of sulfatide also hydrogen bond with the peptidic portion of DDM (8).

This model, along with the demonstrated role of TCR variable regions in antigen recognition (Figs. 1B and 3B), predict that the polypeptide backbone of the antigen serves as the contact for variable regions of the TCR, as is the case for conventional peptide antigens presented by MHC proteins (27, 28). Mammalian ribosomes translate DNA-encoded peptide sequences, which are post-translationally modified by acylation so that their structures resemble mycobacterial DDM but are highly varied in their peptide sequences (29). The autoreactivity of T cells for CD1a proteins suggests that mammalian autoantigens bind to CD1a, and peptidases have been shown to regulate the recognition of CD1d presented antigens (4, 11, 30, 31). These observations raise the possibility that CD1 might also function to present structurally diverse lipopeptides encoded in the DNA of bacteria, viruses, or mammalian cells.

Supporting Online Material

www.sciencemag.org/cgi/content/full/303/5657/527/DC1

Materials and Methods

Table S1

References and Notes

View Abstract

Navigate This Article