TLR9-mediated dendritic cell activation uncovers mammalian ganglioside species with specific ceramide backbones that activate invariant natural killer T cells

CD1d-restricted invariant natural killer T (iNKT) cells represent a heterogeneous population of lipid-reactive T cells that are involved in many immune responses, mediated through T-cell receptor (TCR)–dependent and/or independent activation. Although numerous microbial lipid antigens (Ags) have been identified, several lines of evidence have suggested the existence of relevant Ags of endogenous origin. However, the identification of their precise nature as well as the molecular mechanisms involved in their generation are still highly controversial and ill defined. Here, we identified two mammalian gangliosides—namely monosialoganglioside GM3 and disialoganglioside GD3—as endogenous activators for mouse iNKT cells. These glycosphingolipids are found in Toll-like receptor-stimulated dendritic cells (DC) as several species varying in their N-acyl fatty chain composition. Interestingly, their ability to activate iNKT cells is highly dependent on the ceramide backbone structure. Thus, both synthetic GM3 and GD3 comprising a d18:1-C24:1 ceramide backbone were able to activate iNKT cells in a CD1d-dependent manner. GM3 and GD3 are not directly recognized by the iNKT TCR and required the Ag presenting cell intracellular machinery to reveal their antigenicity. We propose a new concept in which iNKT cells can rapidly respond to pre-existing self-molecules after stress-induced structural changes in CD1d-expressing cells. Moreover, these gangliosides conferred partial protection in the context of bacterial infection. Thus, this report identified new biologically relevant lipid self-Ags for iNKT cells.


Introduction
Type I or invariant natural killer T (iNKT) cells are a subset of "innate-like" αβ T lymphocytes that rapidly produce large amounts of cytokines and chemokines and orchestrate the ensuing innate and adaptive immune responses during infection, inflammatory disorders, and cancer [1]. iNKT cells recognize self and exogenous lipid antigens (Ags) presented by the quasimonomorphic CD1d molecule expressed by Ag-presenting cells [2]. In contrast to the highly diverse T-cell receptor (TCR) repertoire of peptide-specific/major histocompatibility complex (MHC)-restricted conventional αβ T cells, iNKT cells express a semiconserved TCR composed of a unique and invariant TCRα chain, using TCR TRAV11 to TRAJ18 rearrangement in mice (TRAV10 to TRAJ18 in humans), paired with a limited array of Vβ chains [1].
Accordingly, the antigenic diversity and specificity of iNKT cells was intuitively believed to be limited. Over the last decade, numerous studies have attempted to identify cognate Ags involved in iNKT cell thymic selection and/or in their activation during infection and stressful conditions in the periphery (for reviews, [2,3]). To date, several candidates have been proposed to act as self-Ags for iNKT cells, including β-linked glycosphingolipids (GSLs) (isoglobotrihexosylceramide [iGb3] [4,5], gangliosides [6,7] and β-glucosylceramide [β-GlcCer] [8]), and phospholipids [9]. However, the physiological activity of these compounds has either been called into question or demonstrated to have weak and/or subset-specific activity. For example, iGb3 activates some mouse and human iNKT [4] and has been proposed to be responsible for iNKT cell development [5]. However, mice deficient for the enzyme involved in direct iGb3 anabolism (A3galt2) present no iNKT cell defect [10]. In addition, iGb3 is virtually undetectable in mouse peripheral tissues [11], and the human gene encoding iGb3 synthase appears to be nonfunctional [12]. β-GlcCer was also proposed as a self-Ag for both human and mouse iNKT cells [8]. However, it was subsequently reported that this activity was restricted to mammalian α-linked monoglycosylceramides contaminants [13,14]. Initially thought to be absent in mammals, small quantities of α-linked monoglycosylceramides (e.g., α-galactosylceramide [α-GalCer]) that may act as iNKT cell self-Ags have been identified [14]. Although the origin of mammalian α-GalCer is currently uncertain, further evidence suggests that some bacterial communities from the gut microbiota, including Bacteroides fragilis, can produce this particular hexosylceramide [15,16]. In addition, diet could also provide a source of these Ags, as recently demonstrated by the presence of α-linked monohexosylceramides in bovine milk [17]. Thus, the repertoire of self-lipids for iNKT cells lacks consensus and is incompletely characterized.
The possibility that the precise structure of the lipid chain of iNKT self-Ags could be important for their reactivity to the TCR has only been partially investigated. Nevertheless, some GSLs with particular ceramide structures have been shown to carry variable antigenic capacity [8,14]. Ceramide metabolism is tightly regulated by multiple enzyme families, including ceramide synthases (de novo synthesis) [18] and ceramidases (degradation) [19]. The impact of inflammation on the regulation of these enzymes is currently unknown, as well as their potential involvement in iNKT cell biology.
We and others have demonstrated that Toll-like receptor (TLR) engagement on/in Ag-presenting cells resulted in lipid metabolism perturbations uncovering self-Ag(s) for iNKT cells [20,21]. Specifically, we observed that TLR7/9-mediated activation of dendritic cells (DCs) resulted in the accumulation of charged β-linked GSLs able to activate mouse iNKT cells in a CD1d-dependent manner [21], but the precise nature of these activating compounds remained unknown.
Here, we highlighted two particular ganglioside species found in TLR-stimulated DCs that act as iNKT cell activators. Importantly, reactivity of the iNKT cells to gangliosides is highly dependent on the specific structure of the ceramide backbone, particularly the N-acyl fatty acid chain length and its geometry. Both synthetic d18:1-C24:1 GM3 and d18:1-C24:1 GD3 gangliosides activate iNKT cells in vitro and in vivo in a dose-and CD1d-dependent manner. These synthetic gangliosides are unlikely to engage iNKT TCR in their native forms and require unidentified intracellular molecular changes to become antigenic. Thus, we propose a new concept in which iNKT cells can rapidly respond to pre-existing self-molecules after stress-induced structural changes in CD1d-expressing cells. Collectively, our data bring to light a class of self-lipid Ags that can activate iNKT cells and that may be relevant during inflammatory and/or pathological settings.

Self-derived simple sialylated ganglioside(s) activate(s) iNKT cells
We previously reported that the activating iNKT cell Ag(s) that are present in DCs following TLR9 triggering was/were charged β-GlcCer derivative(s) [21]. Amongst the four biosynthetic pathways in vertebrates that stem from β-GlcCer, the ganglioside family contains most of the charged GSLs [22] (Fig 1A). In order to determine whether this class of GSLs was responsible for iNKT cell activation in our system, we used DCs generated from mice deficient in GA2/ GM2/GD2/GT2 synthase (B4galnt1) or GM3 synthase (St3gal5), two key enzymes involved in ganglioside anabolism (Fig 1A). Cytosine-phosphate-guanine oligodinucleotide (CpG ODN)stimulated DCs were used as a source of lipids for in vitro testing. Remarkably, while the lipid fraction from B4galnt1 −/− DCs, which lack all complex gangliosides, retained its capacity to activate iNKT cells, the lipid extract from St3gal5 −/− DCs failed to do so ( Fig 1B). Normalphase high-pressure liquid chromatography (NP-HPLC) analysis revealed that the activating Base-treated charged lipids extracted from unstimulated or CpG ODN-stimulated DCs derived from WT, B4galnt1 −/− , or St3gal5 −/− mice were exposed to DCs (1/500 in DMSO; 100,000 DCs) for 16 h and, after extensive washes, were cocultured with iNKT cell-enriched liver MNCs (500,000 cells) for 24 h. In this setting, IFNγ secretion is measured as a read-out for iNKT cell activity, although we cannot exclude that it can also be produced by other innate lymphocytes (e.g., NK cells) through bystander effect. Data represent the mean ± SEM of three independent experiments performed in duplicates. (C) Glycosphingolipids were extracted and isolated from CpG ODN-stimulated B4galnt1 −/− and St3gal5 −/− DCs, as described in the methods. Charged species were then separated from neutral species using DEAE ion-exchange chromatography prior to NP-HPLC. (D) Sialidase Atreated or untreated base-treated charged lipids extracted from CpG ODN-stimulated DCs were tested for iNKT cell antigenicity as in Fig  1B. Data represent the mean ± SEM of two independent experiments performed in triplicates. �� P < 0.01; � P < 0.05. Underlying data used in the generation of this figure can be found in S1 Data. CpG, cytosine-phosphate-guanine; DC, dendritic cell; DEAE, diethylaminoethyl; fraction (CpG ODN-stimulated B4galnt1 −/− DCs) contains only two detectable species corresponding to the gangliosides GM3 (95%) and GD3 (5%), which are absent from the fraction isolated from CpG ODN-stimulated St3gal5 −/− DCs (Fig 1C). This suggested that the charged activating lipid(s) in CpG ODN-stimulated DCs is/are simple ganglioside(s). Since gangliosides are charged through incorporation of N-acetyl-neuraminic acid, we next tested the impact of sialidase treatment on the activity of the lipid(s). Interestingly, this treatment led to a significant reduction of its biological activity ( Fig 1D). Altogether, these results indicate that the activating GSL(s) in TLR9-stimulated DCs include(s) simple sialylated ganglioside(s).

TLR9 stimulation in DCs modulates ceramide moiety of simple ganglioside species
In line with earlier studies [23], we previously suggested that TLR9 triggering in DCs resulted in metabolic changes in the general GSL pathway [21]. For instance, TLR9 engagement in DCs induced a modulation of gene expression in the GSL biosynthetic pathway, especially in key enzymes involved in the anabolism of simple gangliosides [21]. Accordingly, NP-HPLC analysis indicated that CpG ODN treatment led to a subtle increase of GM3 and GD3 in DCs (Fig  2A). Since the precise structure of the ceramide moiety is important for the antigenic properties of GSLs [8,14,24], we evaluated the composition of the lipid backbone of GM3 and GD3 contained in resting versus TLR9-stimulated wild-type (WT) DCs. As ceramide anabolism and catabolism is dependent, among others, on ceramide synthases (CerS) and ceramidases, respectively [19,25], we assessed the expression of these enzymes in DCs at the transcriptional level. The transcripts of two ceramidases (Asah1 and Naaa) were detected in resting DCs. Of note, CpG ODN stimulation did not affect Asah1 transcript expression, whereas it led to a significant reduction for Naaa expression (Fig 2B). In addition, we detected the mRNA expression of three ceramide synthases (Cers2/5/6) in resting DCs (Fig 2B) specialized in the synthesis of C16 (Cers5 and Cers6) and C22/C24 (Cers2) ceramides [25]. Importantly, TLR9 triggering in DCs led to Cers2 and Cers6 up-regulation but Cers5 down-regulation ( Fig 2B). In line with this gene expression profile, identification of the N-acyl chain heterogeneity of GM3 and GD3 by mass spectrometry in resting DCs revealed the presence of various species differing in length and degree of unsaturation ( Fig 2C). We detected the presence of at least six (assuming d18:1 sphingosine: C16:0, C18:0, C20:0, C22:0, C24:0, and C24:1) and three (C18:0, C24:0, and C24:1) species for GM3 and GD3, respectively, with variable abundance (Fig 2C). In line with our transcriptional analysis, TLR9 triggering led to a modulation in the proportion of specific ceramide forms. Specifically, CpG ODN stimulation preferentially favored C24:1 and C16:0 acyl chains in GM3, while reducing C22:0 and C24:0 species (Fig 2C). Meanwhile, CpG ODN led to an increased proportion of C24:1 acyl chain in GD3, paralleled with a reduction in C16:0 and C24:0 species (Fig 2C). Importantly, commercial gangliosides from bovine buttermilk, which have been widely used to test GSL antigenicity to iNKT cells, differ greatly in their N-acyl chain length composition compared to DC-derived GSLs (Fig 2D and S1  Table). For instance, C16:0 and C24:1 that are the main ganglioside species in mouse DCs are barely found in bovine buttermilk. On the other hand, C22:0, C23:0, and C24:0 gangliosides were abundant in bovine buttermilk, while they were minimally represented or undetectable in mouse DCs. Taken together, these results show that TLR9 stimulation induces not only increased production of GM3 and GD3 in DCs but also leads to an exquisite tuning of their ceramide backbone composition, both of which may have important biological consequences.

Chemical synthesis of the C16:0 and C24:1 ganglioside species
Based on these observations, we synthesized GM3 and GD3 with either a d18:1-C16:0 or a d18:1-C24:1 ceramide backbone (Fig 3A). The structures and synthetic schemes of the four gangliosides are illustrated in Supporting information (S1 Fig and S11-S14 Figs). With respect to the stereo-controlled construction of GM3 and GD3, we first produced a stereo-controlled β-GlcCer (Fig 3B and S15 Fig), which was then used as a building block for gangliosides assembly. β-GlcCer from commercial sources usually comes with α-contaminants (13,14). On the contrary, NP-HPLC and hydrophilic interaction liquid chromatography-tandem mass spectrometry (HILIC-MS 2 ) analyses indicate that there are no such contaminants in our stereocontrolled β-GlcCer compound (Fig 3C and S2 Fig), as compared to a synthetic mixture containing α anomers (S2 Fig). To demonstrate further the absence of α-hexosylceramide contaminants in our products, we probed, in a cell-free system, the structure of CD1d/synthetic GSL complexes using the L363 monoclonal antibody (mAb) that specifically recognizes CD1d/αglycosylceramide interactions [14]. While it allowed detection of CD1d/α-GalCer and CD1d/ commercial β-GlcCer complexes, L363 failed to bind to CD1d/ganglioside and CD1d/stereocontrolled β-GlcCer complexes ( Fig 3D). Thus, we demonstrated the synthesis of β-linked ganglioside species for further iNKT antigenicity assessment. Plate-bound CD1d dimers loaded with various glycolipids were incubated with the PE-conjugated L363 mAb. Fluorescence at 575 nm was measured using a microplate reader. One representative experiment out of two performed in triplicates is shown. Underlying data used in the generation of this figure can be found in S1 Data. β-GlcCer, β-glucosylceramide; CD1d, cluster of differentiation 1d; HPLC-ELSD, high-performance liquid chromatography-evaporative light scattering detector; mAb, monoclonal antibody; PE, phycoerythrin. https://doi.org/10.1371/journal.pbio.3000169.g003

C24:1 gangliosides induce CD1d-dependent iNKT cell activation in vitro
To assess ganglioside antigenicity, we cocultured DCs and iNKT cell hybridomas in the presence of synthetic GM3 and GD3. Both C24:1 GM3 and GD3 activated iNKT hybridomas in a CD1d-and dose-dependent manner, as measured by interleukin 2 (IL-2) production ( Fig 5A  and S4 Fig). In line with our affinity measurements, gangliosides bearing a C16:0 ceramide backbone showed little to no biological activity ( Fig 5A). Consistent with a previous report [14], C24:1 β-GlcCer failed to activate iNKT cell hybridomas. Of note, iNKT cell hybridomas failed to respond to commercial gangliosides extracted from bovine buttermilk (S5 Fig). In addition, C24:1 gangliosides induced the proliferation of splenic iNKT cells ( Fig 5B)

but not conventional T cells in vitro (S6 Fig).
Since C24:1 ceramide biosynthesis is largely dependent on the activity of CerS2 [28], we tested the ability of CpG ODN to uncover iNKT cell endogenous ligands in Cers2-deficient DCs. Interestingly, while CpG ODN-stimulated Cers2-competent DCs activated the iNKT cell hybridoma DN32.D3, Cers2-deficient DCs failed to do so ( Fig 5C). As previously reported [29], it is noteworthy to mention that DCs from CerS2-null mice expressed reduced amounts of surface CD1d (approximately 50%) (S7A Fig). However, this only minimally affected their ability to present exogenous α-GalCer and C24:1 gangliosides to iNKT cell hybridomas (Fig Collectively, these results indicate that C24:1 GM3 and C24:1 GD3 gangliosides are activating molecules for mouse iNKT cells in a CD1d-dependent manner.

C24:1 ganglioside antigenicity requires intracellular machinery and is partially blocked by the L363 mAb
As some GSLs require endosomal lipid transfer factors to be efficiently loaded onto CD1d [30,31], the importance of the intracellular pathway in antigenic capacity of C24:1 gangliosides was investigated. Glutaraldehyde fixation of DCs prior to culture severely impaired the capacity of gangliosides to activate DN32.D3, whereas α-GalCer partially retained its stimulating properties ( Fig 6A). The response to α-Gal(α1-2)GalCer, a glycolipid that requires intracellular processing to reveal its antigenicity [24], was completely abolished by DC fixation (Fig 6A). In addition, the preincubation time of DCs with gangliosides strongly influenced their activating properties (S8 Fig). In the same vein, pretreatment of DCs with bafilomicyn A or concanamycin, two lysosomal acidification inhibitors, prior to ganglioside addition also abrogated iNKT cell activation in response to gangliosides (Fig 6B).
To test whether C24:1 gangliosides could be recognized in their native form by the iNKT TCR, DN32.D3 were cultured in presence of immobilized CD1d:ganglioside complexes. In this setting, we failed to detect any iNKT cell activation, as judged by cytokine release (Fig 6C). Moreover, mouse CD1d tetramers loaded with either GM3 or GD3 did not bind to iNKT cell hybridomas (Fig 6D), suggesting that the native compounds are not directly recognized by the iNKT TCR.
To assess whether intracellular processes such as hydrolysis and/or putative anomerization could convert β-linked gangliosides to α-linked GSLs, resulting in iNKT cell activation, we used L363 mAb in our assay. L363 mAb dose-dependently abolished DN32.D3 reactivity toward α-GalCer but only partly inhibited activity of gangliosides (approximately 50%-60% with the highest dose) (Fig 6E). Albeit at lower levels, it is noteworthy that iGb3, a self-GSL that activates iNKT cells when presented by CD1d in a β-conformation [32], was also significantly inhibited (approximately 20%) in the presence of high concentrations of L363 mAb ( Fig  6E). Moreover, L363 binds to approximately 7% of resting DCs, suggesting the presence of small amounts of α-hexosylceramide:CD1d complexes at cell surface ( Fig 6F). While α-GalCer loading increased L363 binding to DCs, gangliosides did not (Fig 6F). In conclusion, we suggest that gangliosides do not activate iNKT cells in their native forms and require the intracellular machinery to be presented into their antigenic conformation to the iNKT TCR.

C24:1 gangliosides activate lung iNKT cells and partially protect mice against lethal pneumococcal infection
Activation of lung iNKT cells by intranasal inhalation of α-GalCer was shown to promote host defense mechanisms against respiratory bacterial (including pneumococcal) infections [33]. To evaluate ganglioside biological activity in vivo, we administered C24:1 GM3 and C24:1 GD3 intranasally. Interestingly, both lipids dose-dependently activated lung iNKT cells to produce interferon gamma (IFNγ) and IL-17A (Fig 7A). Unlike α-GalCer, ganglioside administration did not result in a strong T-helper (Th1)-biased response by iNKT cells, as assessed by IFNγ/IL-17A ratio (Fig 7A). Similar to α-GalCer, iNKT cell activation was accompanied by the transactivation of gamma delta (γδ)T cells and NK cells as well as the recruitment of neutrophils (Fig 7B-7D). In line with in vitro experiments, C16:0 ganglioside species' inhalation failed to activate lung iNKT cells (S9 Fig). Importantly, CD1d blockade abrogated lung iNKT cell activation in response to C24:1 gangliosides (S10 Fig). Analysis of cytokine release in lung homogenates from WT and Cd1d1 −/− mice indicated that C24:1 ganglioside-induced immune response was dependent on the CD1d molecule ( Fig 7E). (E) DCs were cultured with DN32.D3 in presence of purified glycolipids (23 pM for α-GalCer, 1.5 μM for iGb3, and 6.5 μM for gangliosides) or vehicle for 24 h with increasing concentrations of L363 mAb or isotype control. NKT cell activity was evaluated as in Fig 5A. The dose-dependent L363 inhibition normalized to isotype control for each glycolipid is represented. Means ± SEM of two experiments performed in triplicates are shown. (F) Unloaded or glycolipid-loaded DCs were subjected to L363 mAb or Ig-control staining. Dot plots are representative of one experiment out of two for each condition. �� P < 0.01; � P < 0.05. Underlying data used in the generation of this figure can be found in S1 Data. α-GalCer, α-galactosylceramide; CD1d, cluster of differentiation 1d; DC, dendritic cell; Ig, immunoglobulin; iGb3, isoglobotrihexosylceramide; iNKT, invariant natural killer T; mAb, monoclonal antibody; NKT, type I natural killer T cell; ns, not significant; PE, phycoerythrin; TCR, T-cell receptor. https://doi.org/10.1371/journal.pbio.3000169.g006 Finally, we investigated the effects of ganglioside inoculation on the host protective response against Streptococcus pneumoniae. As depicted in Fig 7F, all vehicle-treated mice died within 5-6 days following pneumococcal infection. In agreement with our previous findings [33], mice pretreated with α-GalCer did not present any observable clinical signs of disease and survived the infection. Interestingly, inoculation of synthetic GM3 or GD3 partially protected mice (around 50%) against lethal S. pneumoniae-induced pneumonia (Fig 7F). In sum, exogenous administration of C24:1 GM3 and C24:1 GD3 activates iNKT cells in vivo and protects against lethal respiratory bacterial infection.

Discussion
Despite strong evidence suggesting the existence of self-lipid Ags for TCR-dependent iNKT cell selection and activation during inflammation, their precise nature remains unclear. Here, we have identified two particular species of the ganglioside family, which can be detected in TLR9-stimulated DCs, as molecules endowed with "indirect" antigenic capacities for iNKT cells. Synthetic d18:1-C24:1 β-linked GM3 and GD3 gangliosides can activate mouse iNKT cells in vitro and in vivo in a CD1d-dependent fashion. However, our data indicate that these molecules are not recognized in their native forms but require further intracellular changes to engage iNKT TCRs.
The contribution of the lipid moiety for iNKT cell antigenicity has mainly been explored with synthetic α-linked glycolipids. Nevertheless, some mammalian GSLs with particular ceramide structures present variable antigenic properties. For instance, mammalian α-GalCer, GlcCer, and sulfatide with a ceramide d18:1-C24:1 backbone have greater antigenic properties compared to other subspecies [8,14]. Similarly, despite sharing a similar polar head group, we observed that d18:1-C24:1 gangliosides activate iNKT cells, while their d18:1-C16:0 counterparts failed to do so, an effect which may be due to their low capacity to bind to mouse CD1d molecule. Our data also highlighted the importance of sialic acid residues for binding to CD1d. While both C24:1 GM3 and GD3 interact with CD1d with a K D of μM range, the neutral C24:1 GlcCer presents a 10-fold lower affinity. Thus, the negative charge conferred by the sialic acids is likely to facilitate the loading into CD1d. At this stage, it remains unclear whether the sialic acid residue(s) participate(s) directly in the antigenicity of the compounds; however, we mainly observed in vitro and in vivo that GD3 had a higher activity on iNKT cells compared to GM3. While the use of B4galnt1-and St3gal5-deficient mice allowed us to suggest the activating capacity of GM3 and GD3, the existence of multiple charged ganglioside subspecies that may act as iNKT activators such as GM1, GM2, or GD1a cannot be definitely excluded.
It has been shown that some cancer cells contain gangliosides (GD3 and N-glycolyl-GM3) that can modulate iNKT cell activity [6,7,34]. In this context, the tumor Ag N-glycolyl-GM3 has been shown to interact with CD1d to induce moderate human iNKT cell proliferation [6]. GD3 antigenicity toward iNKT cells was only observed in preimmunized mice and was subset specific [7]. However, other studies have demonstrated that GD3 had inhibitory effects on TCR-dependent iNKT cell activation although confirming its binding to CD1d [35,36]. In these studies, authors have used bulk gangliosides (either purified or commercial), which may contain contaminants as well as various ceramide species, resulting in a potential competition between inhibitory and activating molecules. Moreover, we observed that C24:1 gangliosides are almost absent in commercial bovine buttermilk-derived GM3.
Amongst the high diversity of molecules bound to the CD1d molecules under steady state, it is interesting to note that GM3 could be eluted from both human and mouse CD1d, whereas GD3 was virtually undetectable [37,38]. However, the authors were only able to determine the structure for up to 25% of the total pool [37], which may explain the absence of GD3.
In line with previous studies [23,38], we observed that TLR triggering on DCs influences GSL abundance and diversity, including gangliosides. However, TLR-signaling ability to generate CD1d-restricted endogenous Ags is unlikely to be solely attributed to the neosynthesis of activating ligands. In fact, the time needed to generate new glycosphingolipids does not fit with the rapid response of iNKT cells. Here, we propose a concept in which TLR9 signaling affects ceramide metabolism to favor a pool of preexisting iNKT cell Ag precursors. Deciphering the respective contribution of C24:1 ceramide synthesis (de novo or salvage) versus catabolism of other species that may represent competitors for CD1d binding (C20:0-C24:0) in TLR9-dependent mechanism will be of the utmost importance. In addition, whether or not the triggering of other TLR members points to a similar mechanism will also need further investigation.
Our data also indicate a critical contribution for the intracellular machinery to efficiently present ganglioside to iNKT cell TCRs. Surprisingly, despite the lack of detectable α-GSL contaminants in our synthetic lipids, we observed that when used at high concentrations (μg/ml), the L363 mAb partially inhibited the antigenicity of both GM3 and GD3, suggesting a requirement for DC-derived α-glycosylceramides in these settings. Therefore, we can envision several nonmutually exclusive scenarios, including i) the involvement of putative specific endogenous anomerases, ii) the synergistic activity of endogenous α-glycosylceramides [14] with the C24:1 β-gangliosides, iii) the catabolism of gangliosides to simpler structure(s) with direct antigenic properties [39], or iv) that the iNKT TCRs reshape the primary β-linked sugar of the gangliosides into a new conformation that mimics α-linked glycolipids, as previously described for iGb3 [32]. The presence of traces of α-contaminants (below the threshold of our analytic procedures) in our synthetic lipids could also explain some of our findings, i.e., the inhibitory effect of L363 on ganglioside activity in our biological assay. However, while L363 is strongly active on α-GalCer at concentrations as little as 10 ng/ml, its effect on gangliosides is only observable with 100-fold higher concentrations. In addition, iGb3 activity is also partially inhibited with highest doses of L363 mAb. Interestingly, this observation can only be made in the coculture setting since L363 was neither able to bind immobilized CD1d:ganglioside complexes or ganglioside-pulsed DCs. Moreover, loading of CD1d tetramers with gangliosides did not allow detection of iNKT cells. Based on this, we believe that this finding is unlikely due to contamination but rather to a certain degree of nonspecificity for the L363 mAb against β-GSLs or to a yet-to-be-defined biological phenomenon, as stated above.
Can GM3 and/or GD3 represent self-Ags during iNKT cell development? Porubsky and colleagues demonstrated that St3gal5 −/− and St8sia1 −/− mice present a normal iNKT cell compartment [40]. Thus, simple gangliosides do not appear to be critical in iNKT cell positive selection or maturation/differentiation. However, in line with its involvement in iNKT cell antigenicity, we recently demonstrated the critical involvement of CerS2, an enzyme involved in the generation of very long-chain sphingolipids, including d18:1-C24:1 GSLs in iNKT cell development and survival [29]. Globally, ceramide synthases and ceramidases have been shown to display variable substrate specificity. In this context, a tight regulation of ceramide metabolism could be critical in either favoring or hindering self-iNKT cell Ag formation. Thus, a better characterization of the role of enzymes of the ceramide metabolism including other families such as β-Glucocerebrosidase, sphingomyelinases, and hexosaminidases in iNKT cell development and/or activation is warranted.
Since the level of expression of endogenous Ags must be low and tightly controlled under homeostatic conditions to prevent iNKT cell autoreactivity and development of autoinflammatory processes, the uncovering of gangliosides as iNKT cell activators makes sense. Indeed, while native gangliosides do not present direct antigenic properties on iNKT cells, this preexisting pool could be rapidly converted into bioactive Ags, resulting in rapid response of iNKT cells upon stressful conditions. Furthermore, gangliosides do not have a strictly host origin. Lipid-enriched diet might represent an exogenous source of gangliosides. Given the role of iNKT cells in diet-induced obesity and insulin resistance [41,42], the contribution of dietary gangliosides in these diseases should be considered. In line with this, GM3 has been proposed to participate in insulin resistance in genetic models of diabetes [43]. Gut microbiota has been proposed to negatively regulate iNKT cell number and functions [44], questioning the putative involvement of gangliosides in this mechanism. Since gangliosides have never been reported to be synthesized/produced by commensals, they are unlikely to directly contribute to this phenomenon. It is also worth noting that gangliosides are overexpressed in human milk with GM3/GD3 as major species in colostrum and GM3 in mature milk [45]. Since human iNKT cells developing during fetal life can mature in the small intestine [46], encounters with milkderived gangliosides shortly after birth may provide positive signals required in this process. A recent study elegantly demonstrated the presence of active α-linked monohexosylceramides Ags in bovine milk that could also play a part in these mechanisms [17]. Hence, the putative presence of active α-linked gangliosides would also be worth investigating. Finally, since we and others [45] observed that bovine and human milk have strong differences in ganglioside species, a comparative analysis of the iNKT cell compartment in breast-fed versus bottle-fed infants would be worthy of study.
In conclusion, our study reinforces the current concept that "innate-like" T lymphocytes are able to sense fine modulation in cellular metabolism to become activated. We demonstrate how an innate signal can be detected by iNKT cells through subtle alterations of the ceramide and/or ganglioside metabolism and the generation of specific antigenic self-lipids.

Ethics statement
All experiments were conducted on C57BL/6J genetic background mice, and the performance was in compliance with current national and institutional regulations and ethical guidelines (B59-350009) and approved by the Comité d'Ethique en Experimentation Animale Nord-Pas de Calais (C2EA-75) under the protocol number 2015121722376405. All efforts were taken to minimize mouse usage to maximize necessary results; provide the best veterinary care; and minimize discomfort, distress, and surgery with anesthetic procedures and euthanasia. Euthanasia was performed using a lethal injection of pentobarbital. In survival experiments, mice were euthanized when reaching one of these endpoints: dehydration, loss of ability to ambulate, labored respiration, or weight loss (>20%).
Lipids were extracted from resting or CpG ODN-stimulated DCs, as previously described [21]. To remove all terminal sialic acid residues, lipids were dried under nitrogen and resuspended in 50 mM sodium acetate, 1 mg/ml sodium taurodeoxycholate, pH 5.5. N-acetylneuraminate glycohydrolase (sialidase, EC 3.2.1.18, Arthrobacter ureafaciens recombinant expressed in E. coli), 50 mU, was added for 48 h in a total volume of 20 μl. All lipid fractions were desalted over C18 cartridges (pre-equilibrated with 2 x 1 ml methanol and 2 x 1 ml water). The volume of digest was made up to 100 μl with water and then the sample was applied to the cartridge; the sample tube was washed in 2 x 100 μl chloroform: methanol: water 1:2.2:1 (v/v/v) and applied to the cartridge. The cartridge was then washed with 3 x 1 ml water and the lipids were eluted with 1 x 1 ml chloroform: methanol 98:2 (v/v), 2 x 1 ml chloroform: methanol 1:3 (v/v), and 1 x 1 ml methanol and mixed thoroughly. Purified lipid fractions were dried under nitrogen for in vitro testing after validating the digestion by subjecting 0.5% of the lipid fraction to ceramide glycanase digestion, anthranilic acid labeling, and NP-HPLC analysis, as previously described [48].

RNA extraction, cDNA synthesis, and real-time PCR
Total RNA from resting or CpG ODN-treated DCs were isolated with the Nucleospin RNA Plus extraction kit (Macherey-Nagel, Hoerdt, France), and cDNA were synthesized from 1 mg of total RNA with random hexamer primers and Superscript III (Invitrogen, Cergy Pontoise, France) according to standard procedures. cDNAs were used as templates for PCR amplification with the SYBR Green PCR Master Mix (Molecular Probes, Leiden, the Netherlands) and the ABI PRISM 7700 Sequence Detector (Applied Biosystems, Foster City, CA). Primers, which are listed in S2 Table, were designed by the Primer Express Program (Applied Biosystems) and used for amplification in triplicate assays. PCR amplification of Gapdh was performed to control for sample loading and to allow normalization between samples. ΔCt values were obtained by deducting the raw cycle threshold (Ct values) obtained for Gapdh mRNA, the internal standard, from the Ct values obtained for investigated genes. For graphical representation, data are expressed as relative expression of mRNA levels.

Analysis of the structure of GM3 and GD3 ceramide tail
Aliquots corresponding to 20,000 BMDCs/μL were mixed with internal lipid standards for analysis by LC MS/MS using an Aquity I-class UPLC and a Xevo TQ-S "triple-quadrupole" instrument, both from Waters. Using a CORTECS HILIC column (2.1 mm x 100 mm; 1.7 μm, Waters), gangliosides were measured in negative mode with a gradient between 80% solvent A (90% acetonitrile) and 100% solvent B (50% acetonitrile), both containing 10 mM ammonium formate as additive.
A stereo-controlled synthetic route was adopted for the production of GD3 and GM3 (S13 and S14 Figs) in pure β form according to the report by Shunichi Hashimoto [51]. This strategy features coupling of trisaccharide (22) or disaccharide (27) with a stereocontrolled building block (17), which excludes possibility of α isomer contamination. In this regard, a disarmed donor (14) was chosen in the glycosylation reaction to form β-glycoside in terms of neighboring participation effect, and β-glycosylceramide was thus prepared (S15 Fig). Removal of chloroacetyl protecting group of 16 gave acceptor (17), which was used as generic acceptor for the assembly of GD3 and GM3 subsequently. Compound 17 was further subject to saponification and Birch reduction to afford pure β GlcCer. Assembly of stereo-controlled GD3 tetrasaccharide was implemented by coupling of (17) with (22), followed by global deprotection and reduction. The stereo-controlled synthesis of GM3 was completed in the same fashion as stereo-controlled GD3. The stereo-controlled GD3 and GM3 matched up with GD3 and GM3 synthesized with general method by 1 HNMR spectra comparison (S16 and S17 Figs). Comparison of controlled GD3 or GM3 with GD3 or GM3 made with conventional method in biological assays using type I natural killer T cell (NKT) hybridomas indicated no differences in their antigenic capacities. Separation and identification of α-and β-GlcCers was conducted with the recently published HILIC-MS 2 method [15]. 1 H NMR, HPLCs, and LC-MS profiles of synthesized GD3 and GM3 are included in (S18-S21 Figs). Synthetic chemistry of gangliosides will be published separately.

Microscale thermophoresis
Quantitative analysis of the interaction between mCD1d and glycolipids was performed by Microscale Thermophoresis (MST) using a Monolith NT115 instrument (NanoTemper Technology GmbH Munich, Germany). Recombinant mCD1d protein (MW = 33.7 kDa) was labeled with a reactive RED dye (NT-647) by N-hydroxysuccinimide (NHS) coupling (Nano-Temper red-NHS kit) following the manufacturer's protocol. Briefly, mCD1d (15 μM) was mixed with the dye (45 μM) and incubated for 30 min in the labeling buffer (130 mM NaHCO 3 , 50 mM NaCl, pH 8.2) at room temperature. Unincorporated dye was removed by gel filtration with a Sephadex G-25 column, and mCD1d was finally collected in PBS 25 mM, 0.05% Tyloxapol to a final concentration of 2.5 μM. The thermodynamic affinity constant characterizing the molecular interaction of mCD1d with each analyzed glycolipid was determined by performing a titration of the corresponding nonfluorescent glycolipids against a constant concentration of NT-647-labeled mCD1d (125 nM). Briefly, C16:0 GM3, C24:1 GM3, C16:0 GD3, C24:1 GD3, and C24:1 β-GlcCer were solubilized in DMSO. A 10-fold factor dilution of these stock solutions was performed in PBS 25 mM, 0.05% Tyloxapol to reach a final maximal proportion of DMSO equal to 5%. Thus, titration of each glycolipid was performed by serial dilutions of these solutions in PBS 25 mM, 0.05% Tyloxapol to which an equal volume of NT-647-labeled mCD1d was added. Then, reaction mixture was loaded into Premium capillaries and subsequently analyzed by MST using 60% MST power with a laser-on time of 30 sec per samples and an intensity of the light-emitting diode (LED) of 20%. Fluorescence time trace for each glycolipid concentration was recorded for each interaction. All analysis was performed in triplicate by using the NanoTemper MO.affinity analysis software version 2.1 and thermodynamic dissociation constant (K D ) characterizing the molecular interaction between mCD1d, and each glycolipid was determined by plotting the temperaturedependent change of the normalized fluorescence (ΔFnorm = F Hot /F cold , with F cold = fluorescence intensity before the IR-Laser is on [area marked in blue], and F hot = fluorescence intensity 1 sec before the laser is off [area marked in red]) with the corresponding concentration of each unlabeled glycolipid. The resulting binding curves were fitted using a 1:1 binding model to determine the average K D values.

iNKT cell activation assays
To investigate iNKT cell reactivity, 1 x 10 5 BMDCs were cultured with 1 x 10 5 mouse iNKT hybridomas DN32.D3 and/or 2C12 in presence of CpG ODN, purified glycolipids or vehicle in complete RPMI media supplemented with 5% FCS for 24 h. In some cases, neutralizing or control Abs were added during the coculture. To fix DCs, cells were exposed to glutaraldehyde (0.05% in PBS) for 3 min and then extensively washed. To investigate the activity of lipids extracted from DCs, 1 x 10 5 DCs were exposed to lipid fractions (1/500 of total extracts from 5 x 10 7 BMDCs) or vehicle alone for 16 h, washed, and then cocultured with liver MNCs (5 x 10 5 ) in the presence or absence of recombinant mIFNβ (1000 U/ml). Coculture supernatants were collected, and cytokine production was measured by ELISA (R&D Systems).

Identification of α-monohexosylceramide: CD1d complexes
To probe the presence of α-monohexosylceramides in our synthetic compounds, glycolipids were loaded on CD1d dimers for 16 h (at a 200 molar excess of lipids except for α-GalCer that was loaded at a 20 molar excess) in presence of 0.5% of Tyloxapol. Then, CD1d:glycolipid complexes were coated on a flat bottom 96-well plate for 4 h in presence of PE-labelled L363 mAb or appropriate Ig control. Fluorescence at 578 nM was measured on a microplate reader TECAN infinite (Männedorf, Switzerland). To test the presence of α-monohexosylceramide: CD1d complexes in ganglioside-loaded DCs, WT-derived BMDCs were loaded for 16 h with the various synthetic gangliosides (6.5 μM), extensively washed, and labeled with the L363 mAb or Ig control. α-GalCer (0.2 nM) was used as a positive control.

Loading of CD1d tetramers with synthetic glycolipids
Unloaded CD1d tetramers were loaded for 16 h with a 5 (α-GalCer) or 50 (gangliosides) molar excess of glycolipids in presence of 0.5% of Tyloxapol. Unloaded or loaded CD1d tetramers were then probed against NKT hybridomas.

Preparation of lung leukocytes and flow cytometry
Mice were IN injected with α-GalCer (500 ng/mouse) or ganglioside species (5 or 25 μg/ mouse) in 50 μl of saline. In some cases, mice were IP pretreated (4 h) with an anti-CD1d mAb (19G11; 500 μg/mouse) or its isotype control (LTF-2). Twelve hours later, lungs were harvested, and leukocyte suspensions were prepared by classical procedures. Briefly, perfused lungs were harvested and finely minced in a Petri dish. Then, lung pieces were enzymatically digested (20 min at 37˚C) in saline containing 1 mg/ml of collagenase type VIII (Sigma-Aldrich) and 1 μg/ml of DNase type I (Roche). After washes, pellets were resuspended in a 20% Percoll gradient and centrifuged (2,000 rpm at RT for 15 min). Cells in pellet were collected and washed in PBS containing 2% FCS. Erythrocytes were removed using a red blood cell lysis buffer (Sigma-Aldrich). Lung mononuclear cells were incubated in complete RPMI 1,640 medium in presence of Golgi Plug/Golgi Stop (BD Biosciences) (2 h at 37˚C). After washes, cells were labelled with the appropriate dilutions of the various mAbs (30 min at 4˚C) in PBS containing 2% FCS. Cells were then washed, and fixed using an intracellular fixation buffer (eBioscience, CliniSciences, Montrouge, France). Next, fixed lung mononuclear cells were treated with a permeabilization buffer (eBioscience) according to the manufacturer's instructions. Cells were stained with anti-IFNγ and anti-IL-17A mAbs or corresponding isotype controls and acquired on a Fortessa (Becton Dickinson, Rungis, France) cytometer. Analyses were performed using the FlowJo software (Treestar, OR, USA).

Inoculation of glycolipids and infection with S. pneumoniae
S. pneumoniae serotype 1 clinical isolate E1586 sequence type ST304 has been described elsewhere [33]. Mice were anesthetized and administered IN with 50 μl PBS containing live bacteria (1 x 10 6 cfu). Twelve hours prior infection, mice were IN injected with 500 ng of α-GalCer (0.58 nmol) or 25 μg of gangliosides (20 nmol for GM3 and 16 nmol for GD3) in 40 μl of saline. Then, mice were monitored daily for illness and mortality for a period of 9 days.

Statistical analysis
Results are expressed as the mean ± SEM. All statistical analysis was performed using Graph-Pad Prism software (San Diego, CA, USA). The statistical significance was evaluated using nonparametric (paired or unpaired) Mann-Whitney U or Kruskal-Wallis (followed by a Dunn's post-test) tests to compare the means of biological replicates in each experimental group. Survival rates after S. pneumoniae challenge were analyzed using a log-rank test. Results with a P value of less than 0.05 were considered significant. ns, not significant; � P < 0.05; �� P < 0.01; ��� P < 0.001. Lung iNKT cells from vehicle-or C16:0 GD3-treated mice (12 h) were screened for intracellular IFNγ and IL-17A production. Individual values and means ± SEM from two independent experiments are shown (5 mice/group) for panel. Underlying data used in the generation of this figure can be found in S1 Data. IFNγ, interferon gamma; IL-17A, interleukin-17A; iNKT, invariant natural killer T. (TIF) S10 Fig. CD1d blockade abrogates lung iNKT cell activation upon C24:1 ganglioside instillation. Lung iNKT cells from vehicle-, α-GalCer-or C24:1 GM3-treated mice (12 h) were screened for intracellular IFNγ production. Mice were intraperitoneally pretreated (4 h prior glycolipid instillation) with either an isotype control mAb (LTF2, 500 μg/mouse) or an anti-CD1d mAb (19G11, 500 μg/mouse). Means ± SEM are shown (n = 3/group). α-GalCer, αgalactosylceramide; IFNγ, interferon gamma; iNKT, invariant natural killer T.