Identification of a Potent Microbial Lipid Antigen for Diverse NKT Cells

This article is featured in In This Issue, p.2505

Natural killer T cells are a subset of αβ TCR⁺ T cells that recognize lipids presented by the MHC class I–like molecule CD1d (1). These cells are divided into two categories based upon TCR usage: semi-invariant/type I NKT (iNKT) cells and diverse/type II NKT (dNKT) cells. iNKT cells primarily express an invariant TCRα-chain (Vα24-Jα18 in human, Vα14-Jα18 in mice) complexed with a limited repertoire of TCRβ-chains, whereas dNKT cells typically express diverse TCRα- and TCRβ-chain sequences (1). For the past two decades, much of the work in the field has focused on iNKT cells because of the ability of α-galactosylceramide (α-GalCer)-loaded CD1d tetramers to specifically identify these cells (2).

iNKT cells and dNKT cells are physiologically distinct cell populations. Not only do these two cell populations recognize different lipids bound within CD1d molecules, even the topology of how their TCRs recognize the CD1d–lipid Ag complex can be clearly different (3). For iNKT cells, the orientation between the iNKT TCR and the CD1d–α-GalCer complex is parallel and focused over the F′ pocket of CD1d, biasing the majority of the TCR–CD1d interaction toward the invariant TCRα, with CDR1α and CDR3α accounting for all interactions with the α-GalCer Ag headgroup (4, 5). In contrast, two recent studies described the crystal structures of dNKT (clone XV19)-derived TCRs in ternary complexes with the glycolipids sulfatide or lysosulfatide bound to CD1d (6, 7). They revealed that these TCRs bound in a manner more analogous to MHC-restricted TCRs, with an orthogonal orientation in which both TCRα’s and TCRβ’s CDR1 and CDR2 loops bind, perched over the A′ pocket, to CD1d, and the CDR3β loop provided the major contact with the bound sulfatide headgroup. Whether this is typical of all dNKT TCR–CD1d–Ag interactions remains to be determined, although recent crystallographic studies of a human γδ TCR, and a hybrid δαβ TCR, interacting with lipid Ags α-GalCer and sulfatide, presented by CD1d, also showed orthogonal docking over the A′ pocket of CD1d (7–9). The fact that dNKT TCRs use diverse TCRα- and TCRβ-chains, and that the XV19 CD1d–dNKT TCR structural studies revealed that the variable CDR3 loops can dominate in lipid Ag recognition, suggests that dNKT cells may possess the capacity to recognize a great range of self and foreign lipid Ags.

One of the key distinguishing features of dNKT cells is that unlike iNKT cells, they do not respond to α-GalCer and, therefore, are not identified by CD1d–α-GalCer tetramers. With the finding that dNKT cells may be present in humans at higher levels than iNKT cells, there is great interest in identifying physiologically relevant lipid Ags for dNKT cells (6, 10). Many of the identified dNKT cell lipid Ags were identified or confirmed by screening a panel of dNKT cell hybridomas. Using these T–T hybridomas, several endogenous mammalian lipid Ags (e.g., sulfatide, phosphatidylglycerol, lysophosphatidylcholine, lysophosphatidylethanolamine, and diphosphatidylglycerol) were confirmed as dNKT cell Ags (11–18).

With the notable exceptions of sulfatide-reactive and Gaucher lipid–reactive dNKT cells (12, 19), no other dNKT cell population has been directly identified in vivo because of the failure of tetramers to bind. Instead, the role of dNKT cells was inferred indirectly by comparing mice lacking iNKT cells (Jα18-knockout [KO] mice) with mice lacking both dNKT and iNKT cells (CD1d-KO mice) (20, 21). Studies with these KO mice demonstrated a protective role for dNKT cells in a variety of pathogenic states, including type 1 diabetes, Con A–induced hepatitis, and murine infection with Schistosoma mansoni or Listeria monocytogenes (1, 10). However, these studies were confounded by the fact that Jα18-KO mice have additional TCR Jα defects, resulting in a limited TCRα repertoire that has only recently been appreciated (22).

Previously, we identified the first microbial dNKT cell lipid Ags (13). Using dNKT cell hybridomas, we found that phosphatidylglycerol and diphosphatidylglycerol (DPG) derived from Mycobacterium tuberculosis, the related Corynebacterium glutamicum, and mammalian phosphatidylglycerol are Ags for a subset of dNKT cells. The microbe-derived phosphatidylglycerols and DPGs contained the same headgroup as their mammalian counterparts but differed in their dominant fatty acid tail structure. The Corynebacterium-derived phosphatidylglycerol/DPG variants were weak Ags and equivalent to the similarly weak mammalian phosphatidylglycerol/DPG with regard to their ability to activate dNKT cells. We reasoned that microbial lipid Ags for dNKTs that are distinctly more active than mammalian phosphatidylglycerol might exist. Therefore, we designed an independent and unbiased search for potential dNKT lipid Ags from other microbes, such as L. monocytogenes.

Listeria is a Gram-positive facultative intracellular bacterium that infects and resides within the cytosol of macrophages, dendritic cells, hepatocytes, and epithelial cells (23–25). This pathogen is a common food contaminant and causes significant mortality in immunocompromised individuals and spontaneous abortions in pregnant women (26). There are three key reasons for investigating Listeria for lipid Ags. First, as an intracellular pathogen the Listeria-derived lipids would be likely to access the intracellular CD1d Ag-presentation system in vivo. Second, data from the comparison of bacterial burdens in Jα18-KO mice and CD1d-KO mice suggest a role for dNKT cells in clearing this pathogen (27, 28). Third, Listeria has been subjected to lipidomics analysis, whereby Fischer and Leopold (29) identified many unique Listeria lipids that do not have mammalian homologs and, thus, might be recognized as foreign. Furthermore, our analysis of the lipidomics data revealed that a number of these lipids are capable of binding to CD1d, making them potential dNKT lipid Ags.

In this study, we used a panel of iNKT and dNKT hybridomas to screen fractionated Listeria lipids for CD1d-restricted Ags. Interestingly, this unbiased screen revealed reactivity in some of the same chemical classes of lipids identified previously, such as phosphatidylglycerol, but with critically important differences. Listeria-derived phosphatidylglycerol and DPG differed significantly in their fatty acid architecture compared with the mammalian/Corynebacterium variants. Importantly, Listeria-derived phosphatidylglycerol is a strikingly more potent Ag for dNKT cells. By performing lipid Ag CD1d-binding assays and tetramer staining, we provide insights into the structural basis for the high potency of microbial compared with mammalian lipid Ags that share identical lipid headgroups.

Wild-type L. monocytogenes (strain 10403S; a gift from H. Shen, University of Pennsylvania) was grown in brain-heart infusion broth (BD Biosciences) supplemented with 200 μg/ml streptomycin (Sigma-Aldrich) overnight to stationary phase at 37°C and 225 rpm. The following day, flasks containing prewarmed brain-heart infusion broth (BD Biosciences) supplemented with 200 μg/ml streptomycin were inoculated at ∼1:420 v/v and grown until mid-log phase (OD₆₀₀ ≥ 0.4). Once at mid-log phase, bacteria were pelleted by centrifugation, washed with PBS, and lyophilized. After up to 48 h of lyophilization, Listeria pellets were processed for extraction of crude polar lipids, as previously described (30). Once isolated, lipids were weighed, resuspended in 2:1 v/v chloroform:methanol (C:M), and stored in 15-ml glass tubes at −20°C until used.

The following mouse NKT hybridomas were tested for reactivity against Listeria lipids: 24.9, DN32, 14S.6, 14S.10, 14S.15, 431.A11, TBA7, VII68, VIII24.1.D, and XV19.2 (31–34). The hybridomas not generated in the Brenner laboratory were kindly provided by S. Cardell (Göteborgs Universitet) and A. Bendelac (University of Chicago). Hybridoma cells were maintained in NKT growth media (RPMI 1640 [Life Technologies] supplemented with 10% v/v FBS [Gemini], 10 mM HEPES [Life Technologies], 2 mM l-glutamine [Life Technologies], 100 U/ml penicillin [Life Technologies], 100 μg/ml streptomycin [Life Technologies], and 55 nM 2-ME [Life Technologies]). RAW 264.7 (RAW) and RAW cells stably transfected with mouse CD1d (RAW-CD1d) were maintained in Complete DMEM (DMEM supplemented with 10% v/v FBS [Gemini], 2 mM l-glutamine [Life Technologies], 100 U/ml penicillin [Life Technologies], 100 μg/ml streptomycin [Life Technologies]). When hybridomas were cocultured with RAW or RAW-CD1d cells or with plate-bound rCD1d, cells were incubated overnight in Complete RPMI (RPMI 1640 [Life Technologies] supplemented with 10% v/v FBS [Gemini], 2 mM l-glutamine [Life Technologies], 100 U/ml penicillin [Life Technologies], 100 μg/ml streptomycin [Life Technologies]). For tetramer and dextramer experiments, the TCR-deficient T cell hybridoma BW58 (35) was stably transfected with CD3, TCRα, and TCRβ from one of four hybridomas (Vβ8.2 [an iNKT cell hybridoma], TBA7 [TBA7^high], 14S.6, or XV19). BW58- and TCR-transfected clones were maintained in DMEM-NKT media (DMEM [Life Technologies] supplemented with 10% v/v FBS [Gemini], 15 mM HEPES [Life Technologies], 2 mM l-glutamine [Life Technologies], 100 U/ml penicillin [Life Technologies], 100 μg/ml streptomycin [Life Technologies], 55 nM 2-ME [Life Technologies], 1 mM sodium pyruvate [Life Technologies], and 1× nonessential amino acids [Life Technologies]). All cells were maintained in 37°C incubators at 5% (RPMI media) or 10% (DMEM media) CO₂.

Preparative HPLC experiments were carried out using a custom Waters (Milford, MA) Autopurification HPLC system comprising a Waters 2767 one-bed injection-collection sample manager, a 2545 quaternary gradient module that can pump up to 150 ml/min, a system fluidic organizer coupled to a single-quadrupole Waters 3100 Mass Detector equipped with Z Spray API ion source, a Waters 2424 evaporative light scattering detector (ELSD), and a Waters 2998 photodiode array detector. In addition, during preparative mode, the system is coupled to two Waters 515 HPLC pumps used for make-up liquid delivery, as well as a 1000:1 splitter that can tolerate a flow rate of 8–30 ml/min. During fractionation, 99.9% of the lipid is sent to a fraction collector for use in future bioassays. A small percentage (0.1%) of the polar lipid is sent to the detectors to identify the lipid-containing fractions to help with fractionation. The entire system was controlled by MassLynx 4.1 software. Chromatographic analyses and separation of crude polar extracts were performed based on the eluent method on a chemically bonded polyvinyl alcohol-silica stationary phase (36). The total liquid chromatography–mass spectrometry (LC-MS) run was pooled into 19 fractions using time and an ELSD to designate fractions. Pooled fractions were weighed, resuspended in 2:1 C:M, and stored at −20°C until ready for use.

For analytical thin layer chromatography (TLC), lipids were spotted onto aluminum-backed silica TLC plates (EMD Chemicals) and dried under low pressure. Next, lipids were resolved in a solvent system designed for the optimal separation of phospholipids (chloroform:acetic acid:methanol:water at a ratio of 40:25:3:6 v/v/v/v). Plates were dried under low pressure, cut (if appropriate), and sprayed with one of four TLC stains: Dittmer-Lester reagent (phosphate stain), α-naphthol (sugar stain), molybdophosphoric acid (MPA; a general stain), or ninhydrin (an amino group stain) (29, 30). Charring of sugar-stained, MPA-stained, and amino group–stained plates was used to develop those plates. Preparative TLCs were performed by first spotting the lipid across the origin of a plastic-backed TLC plate (EMD Chemicals). The plates were then dried under low pressure, resolved in the 40:25:3:6 system, and dried as above. Next, a small segment of the TLC plate was cut off for staining with one of the above TLC stains, which was used to mark the bands of lipid in the unstained section of the plate. The marked silica regions for each lipid band were scraped off the plastic and moved to 15-ml glass tubes, and the lipid was extracted from the silica with three sequential washes of 2:1 C:M. After drying, lipids were weighed, resuspended in 2:1 C:M, and stored at −20°C until used. It is critical to note that the amount of TLC purified lipid in the tube cannot be determined solely by measuring weight, because the measured weight includes some transferred silica. When the identity of the purified lipid was unknown (Fig. 3B), lipids were resuspended in 2:1 C:M based on the total weight of the tube and an estimated expected yield. Because the weights for resuspended lipids were approximated, the amount loaded is shown as the fold-dilution of the stock lipid. To determine the relative concentration for identified lipids, concentrations were measured by spotting both the TLC-purified lipid and a relevant standard (Corynebacterium phosphatidylglycerol, DPG, or Streptococcus digalactosyldiacylglycerol [DGDG]) onto an analytical TLC plate at various concentrations to generate a standard curve. The plates were then dried, resolved with the 40:25:3:6 system, redried, stained with MPA, and charred to develop. The plates were scanned at ≥ 600 dpi, densitometry was performed with ImageJ software (National Institutes of Health), and quantification was determined by comparison with known standards.

RAW or RAW-CD1d cells were resuspended in Complete RPMI, and 5 × 10⁴ cells were plated onto 96-well round-bottom plates (Falcon). Lipids were transferred to 10-ml conical glass tubes (Kimble Chase), dried in a Genevac EZ-2 personal evaporator, resuspended in Complete RPMI, serially diluted, and added to each well as appropriate. No-lipid control wells were given Complete RPMI from 10-ml glass tubes dried with an equivalent volume of the lipid vehicle (2:1 C:M). After a preincubation at 37°C and 5% CO₂ for ≥30 min, 5 × 10⁴ hybridoma cells were added to each well and cultured for 16–18 h. Supernatants were removed and analyzed for the presence of IL-2 by sandwich ELISA with matched anti-mouse IL-2 Ab pairs (BD Pharmingen). IL-2 protein for ELISA standards was from R&D Systems.

Lipids or 2:1 C:M vehicle was dried in a Genevac EZ-2 personal evaporator and resuspended in 50 mM (pH 6) citrate buffer supplemented with 0.25% CHAPS (Sigma-Aldrich). The lipid was then mixed with biotinylated CD1d (provided by the National Institutes of Health Tetramer Facility) that also was diluted into 50 mM (pH 6) citrate buffer supplemented with 0.25% CHAPS at a 20:1 w/w ratio in glass HPLC insert tubes (Supelco). Lipids were bound to CD1d for 24–48 h, after which the pH was adjusted to 7.4 with 1 M Tris (pH 9). Finally, PBS was added to double the initial loading volume before storage at 4°C until use. For binding CD1d onto plates, 0.05, 0.2, or 0.25 μg CD1d was added per well of a streptavidin-coated plate (Thermo Scientific Pierce) for 1 h at room temperature. Plates were washed extensively with sterile PBS before the addition of 5 × 10⁴ hybridoma cells/well in Complete RPMI. After 16–18 h at 37°C and 5% CO₂, supernatants were collected, and IL-2 was analyzed by ELISA.

Biotinylated mouse CD1d (provided by the National Institutes of Health Tetramer Facility) was diluted into TBS (pH 8) to 0.4 mg/ml, and lipids were dried as above and then resuspended in TBS (pH 8) supplemented with 0.05% tyloxapol (TBS-Tyl; Acros Organics) at 1 mg/ml. Lipids or TBS-Tyl vehicle (mock) was mixed with CD1d in HPLC insert tubes at a molar ratio of 35:1 lipid/CD1d molecules. After a 20–24-h incubation at 37°C, CD1d was formed into tetramers by incubation with streptavidin-allophycocyanin (Invitrogen) or diluted with PBS to 0.1 mg/ml CD1d and stored at 4°C for future loading into dextramers. Dextramer-allophycocyanin backbone (a gift from Immudex, Copenhagen, Denmark) was mixed with CD1d at a ratio of 4 CD1d molecules:1 streptavidin binding site on the Dextramer backbone ≥30 min before use.

For flow cytometry, 5 × 10⁴ hybridoma cells/well were preincubated with Brilliant Violet 421–labeled PBS-57 (an α-GalCer analog) CD1d tetramers (provided by the NIH Tetramer Facility) and then stained with APC-labeled mock or loaded tetramers (0.8 μg CD1d/well) or dextramers (3 μl Dextramer-allophycocyanin backbone plus 1.15 μg CD1d/well). Finally, cells were stained with PE-labeled anti–TCR-β (clone H57-597; BioLegend) and analyzed on a BD FACSCanto II flow cytometer. Data were analyzed with FlowJo (TreeStar). For analysis, samples were gated first by forward scatter (FSC)-A and side scatter-A to identify live cells, followed by singlet gating (FSC-A by FSC-H). Next, cells were gated for TCR expression. Tetramer or dextramer (mean fluorescence intensity [MFI] of allophycocyanin-A) levels were determined.

Both high-resolution (R = 100,000 at m/z 400) and low-energy collision activated dissociation tandem mass spectrometry (MS/MS) were performed as previously described, with the exceptions that samples were dissolved into methanol instead of 2:1 C:M, the automatic gain control of the ion trap was set to 5 × 10⁴, and the electrospray needle was set to 4.0 kV (13). For structural analysis of fatty acids, TLC-purified lipids were treated via alkaline hydrolysis to liberate fatty acids, which were then isolated and derivatized with N-(4-aminomethylphenyl) pyridinium (AMPP) and subjected to mass spectrometry, as described (37).

Synthesis of mammalian (18:1/16:0) or Listeria (15:0/17:0) phosphatidylglycerols was performed in a stepwise fashion, starting with a glycerol backbone, as described in Supplemental Fig. 3.

Mouse CD1d/β2m expression vector with a BirA and 6-histidine tag (construct originally provided by Mitchell Kronenberg, La Jolla Institute of Allergy and Immunology) was expressed, purified, and biotinylated in-house, as previously described (38). Soluble mouse TCR production was achieved using chimeric mouse-variable–human-constant domains, as previously described (38). Individual TCR chains were cloned into pET-30 (Novagen) vectors for the TCRβ-chain or pET-28 (Novagen) for the TCRα-chain and expressed in BL-21 Escherichia coli (DE3)pLysS. Inclusion body protein preparations were isolated and refolded as previously described (39), with the exception of the addition of 5 M urea into the refold buffer. TCRs were purified by anion-exchange chromatography, immobilized metal-affinity chromatography, and gel filtration. TCR purity was assessed by gel electrophoresis, and predicted mass was confirmed by time-of-flight mass spectrometry. TCR refolding was confirmed by ELISA using an Ab reactive against a conformational epitope for the TCR constant domain (clone 12H8; produced in-house) and anti-Vβ8.1/8.2 (clone KJ16-133; eBioscience).

Disialoganglioside GD3 (GD3; Matreya #1504) was suspended in TBS-Tyl at 1 mg/ml and mixed at a 3:1 molar ratio with in-house–generated biotinylated mouse CD1d (CD1d-bio) at 1 mg/ml in TBS (pH 8) for 20 h at room temperature. GD3-loaded CD1d-bio was purified by MonoQ anion-exchange chromatography. Listeria phosphatidylglycerol or Corynebacterium phosphatidylglycerol, resuspended at 1 mg/ml in TBS-Tyl, was incubated at a 30:1 molar ratio with purified GD3-loaded CD1d-bio. Phosphatidylglycerol-loaded fractions were purified using MonoQ anion-exchange chromatography. These fractions were used in affinity measures with TBA7 by surface plasmon resonance (SPR). Excess lipid and detergent were removed prior to each chromatography run using a PD10 desalting column (Amersham Biosciences).

SPR experiments were performed at 25°C on a Biacore 3000 instrument and conducted in HEPES buffered saline (10 mM HEPES [pH 7.4] 150 mM NaCl); 1% BSA was added to prevent any nonspecific binding. Four thousand response units (RU) CD1d-bio loaded with the lipid Ag Corynebacterium phosphatidylglycerol or Listeria phosphatidylglycerol–LC-MS fraction E were coupled onto the streptavidin sensor chip. An HLA class I molecule was immobilized on one flow cell for reference subtraction. Biotin was injected to block the free streptavidin sites. Nine serial dilutions of TBA7 from 200 to 0.78 μM were passed through as analyte. BIAevaluation software was used for data analysis.

All IL-2 ELISA graphs, fold-change graphs, and percentage displacement graphs were generated using GraphPad Prism 5.0b. All statistical analyses (one-way ANOVA) were performed with GraphPad Prism 5.0b.

We isolated polar Listeria lipids from mid-log phase bacteria cultures and cocultured the crude polar lipid mixture with two iNKT cell hybridomas (Fig. 1A, 1B) and eight dNKT cell hybridomas (Fig. 1C–J). To be considered activated, the following requirements were necessary: a dose response between lipid Ag concentration and hybridoma IL-2 production; IL-2 production when cocultured with lipid Ag, and RAW-CD1d is significantly higher than when cocultured with no exogenously added lipid; and little to no response when cocultured with lipid Ag and untransfected RAW cells. By these requirements, both iNKT cell hybridomas (Fig. 1A, 1B) failed to be activated by crude Listeria polar lipids, as also noted by other investigators (31–34, 36, 40). In contrast, six of the eight dNKT cell hybridomas were positively activated by Listeria polar lipids in a CD1d-dependent manner. Based on these results, we chose two dNKT hybridomas, 14S.6 and TBA7, which gave strong responses to the polar lipid extract for further study to identify the relevant Listeria lipid Ags present in the crude extract.

We fractionated the Listeria polar lipid extract by preparative HPLC. This preparative LC-MS system involves the use of three isocratic solvent mixtures on a polyvinyl chloride silica column to facilitate separation of lipids based on headgroup structure. Four detectors are used to reveal lipids, including a UV detector, a single quadrupole mass spectrometer, an ELSD, and a photodiode array detector. We used ELSD (measured in light scattering units [LSU]) and retention time as the criteria for separating the total LC-MS run into 19 fractions, which were tested for biological activity. Although fractions encompassing the whole run were tested for activity, the major antigenic fractions correlated with the presence of detectable LSU signals (Fig. 2A, data not shown). The two major LSU peaks appeared at ∼10 min and between 38 and 48 min elution time. The peak at 10 min (Peak 1) was identified by TLC to consist of free fatty acids and roughly equal amounts of 1,2- and 1,3-diacylglycerol (data not shown). The stimulatory Peak 1 fraction was weaker than those collected between 38 and 48 min and was not characterized further. The six LC-MS fractions collected between 38 and 48 min activated the diverse hybridomas 14S.6 and TBA7 in a CD1d-dependent manner (Fig. 2B–E).

The fact that all of these fractions are eluting from the LC-MS column around the same time suggested that they may share common polar headgroups. Further, we noted previously that, in our HPLC system, LSU peaks eluting between 38 and 48 min typically indicate the presence of phospholipids. By analyzing migration on TLC plates in a solvent system designed for resolving phospholipids, we found that fractions A–F included the presence of phosphate-containing lipids that migrated to similar heights, suggesting that they all shared similar structures in different ratios (Supplemental Fig. 1). At the concentrations tested, LC-MS fraction C showed a clear dose response across concentrations and was available in sufficient quantities for further analysis (Fig. 2D). After further separation by TLC, we subjected LC-MS fraction C to a variety of stains (Fig. 3A) (13). Dittmer-Lester (phosphate) stains phosphate-containing molecules a blue color on a white background. α-Naphthol (sugar) stains lipids containing carbohydrate groups a dark purple color. MPA (general) is a general lipid stain that is thought to stain fatty acid tails. Finally, LC-MS fraction C also was stained with ninhydrin (amino), which stains amino groups reddish-pink but also can nonspecifically mark some lipids with a brown color.

Based on these TLC stains, we identified 12 lipid bands from LC-MS fraction C and isolated each by preparative TLC. These bands were tested for activity with hybridoma 14S.6 (Fig. 3B). The top six bands (TLC 1–6), which contained all of the phosphate- or sugar-positive lipid bands, activated hybridoma 14S.6, but the bottom six bands did not. MS/MS on the six active TLC bands identified these as lipid species with glycerol backbones, and each lipid was dominated by fatty acid tails of length 17:0/15:0 in the sn1 and sn2 positions, respectively. By MS/MS analysis of the TLC-purified lipids, TLC bands 1–3 consisted of DPG (Fig. 3C), TLC band 4 was identified as phosphatidylglycerol (Fig. 3D), and TLC band 5 was DGDG (Supplemental Fig. 2C). However, further TLC-based analysis of TLC-purified DGDG from Listeria determined that this activity was likely due to a comigrating UV light⁺ molecule. DGDG from either Streptococcus pneumoniae or LC-MS fraction B (that did not contain the comigrating UV⁺ band) was inactive (Supplemental Fig. 2B). TLC band 6 was not visible by any of the four TLC stains, and MS/MS data were inconclusive. We next generated AMPP derivatives of the fatty acid tails for further analysis of the fatty acid structure by gas chromatography-MS (GC-MS) (37). GC-MS on AMPP derivatives from DPG, phosphatidylglycerol, and DGDG revealed that the fatty acids are predominately anteiso methyl-branched fatty acids (Fig. 3E, 3F), a structure that is not found in mammals (35, 41).

LC-MS fractions A–F contain phospholipids that resolve on TLC plates with similar mobilities, suggesting that these fractions contain lipids with the same headgroups (Supplemental Fig. 1). Next, we used TLC to isolate putative DPG and phosphatidylglycerol bands from the other LC-MS fractions that contained enough materials for further study (fractions B–E) and confirmed by MS/MS that these were DPG or phosphatidylglycerol and that the major fatty acid tails were anteiso isomers by GC-MS on AMPP derivatives of the fatty acid tails. In total, we isolated DPGs from LC-MS fractions B–E and phosphatidylglycerols from LC-MS fractions C–E. Although the same lipid was found throughout multiple fractions, we identified differences in the fatty acid substituent compositions. These differences included the presence or absence of plasmenyl phosphatidylglycerol (fatty acids with an sn1 ether linker and an sn2 ester linker) (37) and different ratios of phosphatidylglycerol species (e.g., the ratio of 15:0/15:0 to 17:0/16:0 and 17:0/17:0) (Fig. 3D, data not shown).

Recently, we reported that phosphatidylglycerol and DPG were dNKT cell Ags derived from mammals or C. glutamicum (13). However, the two species had very similar lipid structures; notably the same dominant fatty acid tails (16:0 and 18:1) were present but opposite in sn1 and sn2 orientation (Fig. 3G, 3H). Importantly, when Corynebacterium phosphatidylglycerol or DPG was compared with mammalian phosphatidylglycerol or DPG, there was no difference in their potency with regard to the activation of dNKT cells (13). Unlike the previously described phosphatidylglycerol and DPG lipids from mammals, Listeria phosphatidylglycerol and DPG have a distinct fatty acid architecture, prompting us to ask whether Listeria phosphatidylglycerol and DPG are more or less stimulatory than the corresponding mammalian (or Corynebacterium) sources. When comparing the ability to activate hybridomas 14S.6 and TBA7, Listeria DPG was an equally potent Ag to Corynebacterium DPG (Supplemental Fig. 2A). However, we found that Listeria-derived phosphatidylglycerols were strikingly more potent Ags than Corynebacterium-derived phosphatidylglycerol, as measured by their ability to activate the dNKT hybridomas at lower lipid concentrations (Fig. 4A). Next, we calculated the Listeria phosphatidylglycerol concentration needed to obtain an equivalent level of IL-2 production as the first Corynebacterium concentration to be clearly above background levels (Table I). When we calculated the fold change in concentration needed to get the same level of activity, we found that the Listeria phosphatidylglycerols were 10–100-fold more potent than Corynebacterium-derived phosphatidylglycerol, which is similar in potency to phosphatidylglycerol from mammals (Fig. 4B, Table II).

To confirm these findings, we synthesized the dominant Listeria phosphatidylglycerol variant (a17:0/a15:0) and compared its ability to activate TBA7 cells with that of synthetic mammalian phosphatidylglycerol (16:0/18:1). These results supported our previous observations that Listeria phosphatidylglycerol was a more potent Ag than mammalian phosphatidylglycerol and had a similar fold difference in activity (∼13-fold) to Listeria LC-MS fractions C and D phosphatidylglycerols (Fig. 4C).

To determine whether cellular processing of Listeria phosphatidylglycerol was required for presentation to dNKT cells, we tested activity using an APC-free system. We loaded the most active Listeria phosphatidylglycerol (phosphatidylglycerol–LC-MS fraction E) or the prototypical iNKT cell Ag α-GalCer onto biotinylated CD1d and then bound 0.2 μg of CD1d/well to streptavidin-coated plates. Different NKT hybridomas were incubated with the plate-bound CD1d (Fig. 5A). As expected, α-GalCer–loaded CD1d activated the iNKT DN32 hybridoma but did not activate the dNKT TBA7 hybridoma. Importantly, phosphatidylglycerol–LC-MS fraction E–loaded CD1d activated the dNKT TBA7 hybridoma but not the iNKT DN32 hybridoma. Mock-loaded CD1d did not activate DN32 cells, but it weakly activated TBA7 cells, reflecting the known CD1d autoreactivity seen with many dNKT hybridomas. Next, we determined whether the different antigenic properties of the various phosphatidylglycerols would be reflected in this APC-free system. Indeed, when we loaded Corynebacterium phosphatidylglycerol or the various TLC-purified Listeria phosphatidylglycerols into plate-bound CD1d, we measured a dose-dependent increase in IL-2 production (Fig. 5B, compare left and right panels). The amount of IL-2 produced at 0.25 μg CD1d/well (Fig. 5B, right panel) closely mimicked the fold difference in activity seen in the system using live RAW-CD1d APCs (Fig. 4B, right panel), suggesting that the difference in activity is not due to processing of the LC-MS fractions within RAW-CD1d cells.

These results prompted us to determine whether Listeria phosphatidylglycerol was an Ag for the other six dNKT hybridomas. We performed APC-containing (RAW-CD1d cells) and APC-free (plate-bound CD1d) experiments to determine reactivity to Listeria phosphatidylglycerol–LC-MS fraction E (Table III). In the APC-free system, three of the hybridomas (14S.6, 14S.10, and TBA7) were reactive to Listeria phosphatidylglycerol. These results match what we (13) found for Corynebacterium phosphatidylglycerol. In the APC-containing system, an additional two dNKT hybridomas (VII68 and VIII24.1.D) were activated by Listeria phosphatidylglycerol. Because we generally found that the plate-bound CD1d system is less sensitive for weak Ags than the APC-containing system, this likely reflects a low-affinity reactivity for Listeria phosphatidylglycerol by these two hybridomas.

The high activity of Listeria phosphatidylglycerol in the CD1d plate-bound assay compared with other established phosphatidylglycerol sources prompted us to determine whether Listeria phosphatidylglycerol-loaded CD1d tetramers would bind to TBA7 cells. In the CD1d plate-bound assay, excess detergent and lipid are washed away before adding hybridoma cells to the wells. To minimize excess detergent and lipid in the tetramer preparations, we first optimized the system by modifying our lipid-loading protocol. In addition, we generated hybridoma cell lines that expressed high levels of TCR. The TCR⁻ immortalized T cell line BW58 was transfected with CD3 and the TBA7 TCR (TBA7^high) or a typical iNKT cell TCR (Vβ8.2) (35). As expected, CD1d tetramers loaded with PBS-57 (a synthetic α-GalCer analog) stained Vβ8.2 cells but not TBA7^high cells (data not shown). Further, Listeria phosphatidylglycerol-loaded CD1d tetramers did not bind to Vβ8.2 cells (Fig. 5C), the parent BW58 hybridoma cells (data not shown), or BW58 cells transfected with the phosphatidylglycerol-nonreactive dNKT XV19 hybridoma cell TCR (data not shown). In contrast, we found that tetramers loaded with Listeria phosphatidylglycerol–LC-MS fraction E stained TBA7^high cells, whereas mock-loaded CD1d tetramer or tetramers made of CD1d loaded with the irrelevant lipid DGDG from S. pneumoniae (DGDG-Sp) bound only at background levels (Fig. 5C, left panels). Together, these functional and tetramer-staining experiments demonstrate that Listeria phosphatidylglycerol-loaded CD1d is a cognate Ag for the dNKT TBA7 TCR.

Listeria phosphatidylglycerol–LC-MS fraction E–loaded tetramers gave a signal that was ∼2-fold higher than vehicle-loaded tetramers (Fig. 5C, right panels). We next attempted to increase the positive–negative signal separation using dextramer technology. Dextramers are dextran backbones containing multiple fluorophore and streptavidin binding sites/molecule (42). Because they contain 12–24 Ag-presenting molecules/dextramer backbone, they are useful for identifying rare primary T cell populations because of their increased TCR avidity and fluorescence compared with tetramers. Indeed, Kasmar et al. (43) successfully used CD1a dextramers to identify dideoxymycobactin-restricted T cells ex vivo from human PBMCs. We generated mock-loaded CD1d dextramers along with Listeria phosphatidylglycerol–LC-MS fraction E–, phosphatidylglycerol from C. glutamicum (phosphatidylglycerol-Cg), or DGDG-Sp–loaded CD1d dextramers and used them to stain dNKT TBA7^high BW58 cells transfected with the 14S.6 TCR (14S.6^high), as well as iNKT Vβ8.2 cells (Fig. 5D, left panel, data not shown). We found that phosphatidylglycerol LC-MS fraction E–loaded CD1d dextramers specifically bound to TBA7^high cells and gave a signal that was ∼8-fold higher than for vehicle-loaded CD1d dextramers (Fig. 5D, right panel). The phosphatidylglycerol LC-MS fraction E–loaded CD1d dextramers did not bind to 14S.6^high cells (data not shown), suggesting that the 14S.6 TCR affinity for this complex is very low. Finally, we also tested the ability of the less active phosphatidylglycerol from Corynebacterium to identify TBA7 TCR-expressing T cells with this newly optimized system. Phosphatidylglycerol-Cg–loaded CD1d dextramers also consistently stained the TBA7^high cells, albeit to a much lower degree than did the Listeria phosphatidylglycerol–LC-MS fraction E–loaded dextramers; however, the difference did not reach statistical significance (∼1.5-fold higher MFI than mock loaded).

Previous studies on iNKT cell Ags demonstrated that alterations in the fatty acid tails of lipids can dramatically alter their activity, either by modulating their binding to CD1d or by indirectly affecting iNKT TCR recognition (44–46). Listeria phosphatidylglycerols are dominated by short chain lengths and fully saturated anteiso methyl-branched fatty acids. In contrast, the less antigenic mammalian or Corynebacterium phosphatidylglycerol have longer acyl chains, are unsaturated, and lack anteiso branches. These differences in fatty acid tail composition could alter how well the lipid binds into CD1d, alter the orientation of the phosphatidylglycerol headgroup into a more favorable position for TCR binding, or alter CD1d conformation that subsequently impacts on TCR binding. To test whether the more potent Listeria phosphatidylglycerol–LC-MS fraction E loads into CD1d more efficiently than Corynebacterium phosphatidylglycerol, we performed a GD3 lipid-displacement assay. GD3 is a negatively charged lipid with a large sugar-based headgroup that can be displaced from CD1d by other lipids. We loaded CD1d with GD3, purified the GD3-loaded CD1d complexes by MonoQ anion-exchange chromatography, and measured the ability of Listeria or Corynebacterium phosphatidylglycerol to displace GD3, which results in earlier elution relative to the CD1d–GD3 complex. We found that Listeria phosphatidylglycerol–LC-MS fraction E displaced ∼2-fold more GD3 than did Corynebacterium phosphatidylglycerol under the same conditions (35% versus 19% loaded, Fig. 6B).

We next used SPR to determine whether the TBA7 TCR affinity for Listeria phosphatidylglycerol–LC-MS fraction E differed from that of Corynebacterium phosphatidylglycerol. CD1d was loaded with Listeria phosphatidylglycerol or Corynebacterium phosphatidylglycerol using the GD3-displacement approach described above to ensure optimal loading. CD1d with endogenous lipid Ag was used as a control. These preparations were immobilized to the streptavidin SPR chips via their biotin tags. Purified and soluble TBA7 TCR was then passed over CD1d-Ag, and the binding affinity was measured (in RU). The TCR affinity for Listeria phosphatidylglycerol–LC-MS fraction E–loaded CD1d (K_D = 71 μM) was similar to CD1d loaded with Corynebacterium phosphatidylglycerol (K_D = 94 μM) (Fig. 6C, 6D). For comparison, these affinities are lower than the previously published TCR affinities for the dNKT hybridoma XV19 TCR binding to sulfatide + CD1d (K_D = 24 μM) or the nanomolar iNKT TCR affinity observed for α-GalCer–loaded CD1d (K_D = 0.07 μM) (6, 38). Accordingly, our data suggest that the increased potency of Listeria phosphatidylglycerol is probably not due to higher-affinity interactions with the TCRs; rather, it may be attributable to improved loading and/or binding to CD1d.

There is a growing appreciation that many T cells do not recognize peptides in the context of MHC class I or MHC class II molecules (47). Such non-MHC–restricted T cells, which can recognize lipid Ags (48) or riboflavin metabolites (49), include NKT cells (CD1d restricted), mucosal-associated invariant T cells (MR1 restricted), CD1a/CD1b/CD1c T cells, and TCR γδ T cells. Collectively, these cells can represent between 10 and 50% of circulating T lymphocytes, depending on the human donor (10, 50–53). By using CD1 or MR1 multimers loaded with specific Ags, investigators have started to interrogate these non-MHC–restricted T cell populations in vivo (43, 52, 54).

Primary dNKT cells are poorly understood because of the lack of tools to identify them in vivo (55). dNKT cells restricted against the mammalian lipid sulfatide is the only primary dNKT cell population that has been carefully studied ex vivo with CD1d tetramers (12, 56). However, because of the wide range of TCR Vα- and Vβ-chains used by dNKT cells, it is unlikely that the function(s) of sulfatide-restricted dNKT cells can be generalized to all dNKT cells. Indeed, the dNKT XV19 hybridoma, which recognizes sulfatide and was the basis for testing sulfatide-loaded CD1d tetramers, does not recognize the Listeria phosphatidylglycerol Ags used in this study (Table III, data not shown). This finding highlights the need to identify more high-potency dNKT cell Ags for further interrogation of dNKT cells in vivo.

Microbial lipid Ags are ideal targets for dNKT cells. Although mammalian self-lipids may be responsible for dNKT cell selection in the thymus, like self-peptides for MHC-restricted T cells, the most potent Ags recognized in the periphery may be of microbial origin. Because Listeria is an intracellular microbial pathogen, it was an attractive model organism for identifying Ags for dNKT cells (29, 57). By performing an unbiased search for Listeria lipid Ags, we identified the microbial versions of two known dNKT cell phospholipid Ags, phosphatidylglycerol and DPG, as dNKT cell Ags. By measuring the concentration of Listeria phosphatidylglycerol needed to activate 14S.6 and TBA7 hybridoma cells as compared to phosphatidylglycerol-Cg (Fig. 4B), we found that Listeria phosphatidylglycerol is a 10–100-fold more potent Ag than the previously published mammalian or the structurally related Corynebacterium phosphatidylglycerol.

These results prompted us to consider why Listeria phosphatidylglycerol is a more potent Ag than the structurally similar Corynebacterium/mammalian phosphatidylglycerol. Because these two lipids share identical headgroups, it was unsurprising to find that the dNKT TBA7 TCR bound to CD1d loaded with either Listeria or Corynebacterium phosphatidylglycerol with similar affinities. However, we found that Listeria phosphatidylglycerol loaded into CD1d (displacing the charged lipid GD3) ∼2-fold more efficiently than did Corynebacterium phosphatidylglycerol. Because TCR activation leads to signaling cascades that can exponentially amplify the original signal, we attribute the higher potency of Listeria phosphatidylglycerol to its increased CD1d loading efficiency over Corynebacterium phosphatidylglycerol. However, we cannot discount the possibility that Listeria phosphatidylglycerol is more stably bound within CD1d over time as compared to Corynebacterium phosphatidylglycerol.

There are three differences between the fatty acid tails found in Listeria that could be modulating its ability to load or stay within CD1d more efficiently than mammalian phosphatidylglycerol. These differences include the presence of anteiso methyl groups, lack of a double bond, and shorter tail lengths. One hypothesis is that the anteiso methyl fatty acid branches (found only in some microbes) act as a hook within the CD1d hydrophobic channels and increase the stability of the lipid–CD1d complex. Alternatively, it is possible that other aspects of Listeria phosphatidylglycerol, such as its shorter tails or lack of a double bond, make it easier to load or remain bound within CD1d molecules.

Our previous attempts to generate Corynebacterium or mammalian phosphatidylglycerol CD1d tetramers were not successful. However, identification of the more potent Listeria phosphatidylglycerol variant prompted us to determine whether CD1d tetramers loaded with Listeria phosphatidylglycerol could bind with sufficient avidity to TBA7 cells to stain in flow cytometry. Indeed, Listeria phosphatidylglycerol–LC-MS fraction E–loaded CD1d tetramers specifically bound to TBA7 TCR-transduced cells but not to the same cells transduced with other (or no) TCRs (Fig. 5C, 5D, data not shown). By using dextramers that contain ∼12 CD1d molecules and multiple fluorophores/molecule, we were able to further increase the signal/noise ratio so that the majority of the Listeria phosphatidylglycerol dextramer-stained cells could be separated from the mock- or irrelevant lipid–loaded dextramer-stained cells. This increase in signal may be critical, because it provides a new reagent that is suitable to identify and interrogate dNKT cells in vivo. We are now optimizing this technology to stain primary human and mouse dNKT cells, including taking advantage of the recent insight that unlabeled anti-fluorophore Abs can stabilize TCR–multimer complexes with affinities similar to those found in this study (58).

In summary, we identified a new lipid Ag for dNKT cells with a distinctively microbial signature: short, fully saturated anteiso lipid tails. Notably, this microbial version of phosphatidylglycerol is much more active than the previously known phosphatidylglycerol Ags from mammals or Corynebacterium, which have structurally related lipid tails. Importantly, by identifying high-potency microbial dNKT cell Ags for different dNKT cell populations, we can begin to dissect the poorly understood nature of dNKT cells in vivo.

We thank the National Institutes of Health Tetramer Core Facility (contract HHSN272201300006C) for providing us with unloaded biotinylated CD1d monomers and the PBS-57 tetramer. We also thank Xavier Michelet, Sook Kyung Chang, Patrick Brennan, Daniel Pellicci, and other members of the Brenner, Godfrey, and Rossjohn laboratories for helpful discussions, support, and advice. Finally, we are grateful to Søren Jakobsen at Immudex for supplying us with dextramer backbones.

The authors have no financial conflicts of interest.