Structural Study of Lipomannan and Lipoarabinomannan from Mycobacterium chelonae PRESENCE OF UNUSUAL COMPONENTS WITH (cid:1) 1,3-MANNOPYRANOSE SIDE CHAINS*

Lipomannan (LM) and lipoarabinomannan (LAM) are major glycolipids present in the mycobacterial cell wall that are able to modulate the host immune response. In this study, we have undertaken the structural determi-nation of these important modulins in Mycobacterium chelonae , a fast growing pathogenic mycobacterial species. One-dimensional and two-dimensional NMR spectra were used to demonstrate that LM and LAM from M. chelonae , designated CheLM and CheLAM, respectively, possess structures that differ from the ones reported earlier in other mycobacterial species. Analysis by gas chromatography/mass spectrometry of the phosphati-dyl- myo -inositol anchor, which is thought to play in RPMI 1640 supplemented with 10% fetal calf serum and glutamine. The purified molecules LAMs from M. Chelonae , M. tuberculosis Erdman, and M. smegmatis were added at a final concentration of 10 (cid:4) g/ml, and triplicates were performed in order to measure cytokine release. Culture supernatants were collected after 6 or 24 h for TNF- (cid:1) or IL-8 production, respectively. Specific enzyme-linked immunosorbent assays commercial kits were used according the manufacturer’s instructions. Human IL-8 and TNF- (cid:1) kits were purchased from Bender Med systems Diagnostic and R & DSystems, respectively. Cytokine production was quantified with a microtiter plate reader in comparison with a standard curve generated with recombinant human cytokines.

Mycobacterium species are responsible for important human diseases including tuberculosis and leprosy. Infection and im-munopathogenesis of these diseases widely implicate the mycobacterial cell wall (1) which is abundantly composed of mannoconjugates, notably polysaccharides and lipoglycans. The latter consist mainly of phosphatidyl-myo-inositol mannosides (PIMs), 1 lipomannan (LM), and the structurally related lipoarabinomannan (LAM). LAM is a major cell wall component and is considered as a modulin through its various immunoregulatory and anti-inflammatory effects, which favor the survival of the mycobacteria within the infected host. These effects include suppression of T lymphocyte proliferation through interference with antigen processing (2), inhibition of macrophage activation by interferon-␥ (3,4), and scavenging of oxygen-derived free radicals (5). LAM is not only a virulence factor responsible for macrophage deactivation, but it is also implicated in phagocytosis of mycobacteria into phagocytic cells (6). In addition, PIMs that are believed to be precursors of LM and LAM have recently been proposed to recruit NK T cells, which play a primary role in the granulomatous response (7,8).
The biosynthetic relationship of phosphatidylinositol (PI), PIMs, LM, and LAM has recently been supported by biochemical (9,10) and genetic studies (11,12), but the details of this pathway remain highly speculative. However, the structures of LAM from several species including Mycobacterium tuberculosis, Mycobacterium leprae, Mycobacterium bovis BCG, and Mycobacterium smegmatis have been extensively described during the last decade. LAM is a complex glycolipid composed of Dmannan and D-arabinan attached to a PI moiety that anchors the glycolipid in the mycobacterial cell wall (13). The biosynthesis of LAM involves the addition of mannopyranosyl (Manp) residues to PI to produce both the short PIMs (2-5 Man residues) and LM, which is further glycosylated with arabinan to form LAM (9,(13)(14)(15). In all the species described so far, Dmannan consists of a highly branched structure with an ␣1,6linked Manp backbone substituted at C-2 by single Manp units (16). The mannan size and the degree of branching can vary depending on the species. The arabinan consists of a linear ␣1,5-linked arabinofuranosyl (Araf) backbone punctuated by branching produced with 3,5-O-linked ␣-D-Araf residue. The lateral chains are organized either as linear tetra-arabinofuranosides ␤-D-Araf-(1,2)-␣-D-Araf-(1,5)-␣-D-Araf-(1, 5)-␣-D-Araf or as a biantennary hexa-arabinofuranosides [␤-D-Araf-(1,2)-␣-D-Araf-(1-] 2 -3 and 5)-␣-D-Araf-(1,5)-␣-D-Araf (17,18). Comparative analyses of LAMs from different mycobacterial species have shown that the non-reducing termini of the arabinosyl side chains are differentially modified. M. tuberculosis and M. leprae modify the termini with Manp residues, thereby yielding "ManLAM," whereas the rapidly growing species M. smegmatis uses inositol phosphate, generating "AraLAM" (19). It is thought that these modifications are responsible for the marked differences in the biological activities of ManLAM and AraLAM (15,19,20). However, as mentioned above, all available LAM structures are derived from a very limited panel of mycobacterial species, and whether these structures are invariably present in most species remains to be investigated. Awareness that subtle differences among LAMs may affect their biological properties prompted us to establish the structure of LM/LAM from Mycobacterium chelonae, a rapidly growing pathogenic mycobacterium that is found in soil and fresh water throughout the world (21). M. chelonae infection typically causes localized skin lesions, often following penetrating trauma or injections. Disseminated disease usually occurs in patients with a significant immune compromised state that is most commonly attributable to exogenous steroid use (22,23).
We report here the detailed structure of LM/LAM from M. chelonae and provide evidence for important differences such as the acylation composition of the PI, branching of the mannan core, and the absence of Manp and inositol phosphate caps. Because these structures were found to be unique among a panel of various mycobacterial species, we propose to designate these components as CheLM and CheLAM.

EXPERIMENTAL PROCEDURES
Strain and Culture Conditions-All mycobacterial species used were grown on plates containing Middlebrook 7H11 agar supplemented with 10% oleic acid-albumin-dextrose-catalase enrichment (Difco) or in liquid Sauton medium. Except for M. chelonae (ATCC 19536), which was grown under shaking in Sauton medium at 30°C for several days, all other species were grown at 37°C. Confirmation of the identity of the M. chelonae strain was done by analyzing its mycolic acid profile, which is rather unusual because it consists of 60% of ␣-mycolates and 40% of ␣Ј-mycolates (24,25).
Purification of CheLAM and CheLM-Extraction of CheLM and CheLAM was adapted from Nigou et al. (26) based on the Triton X-114 phase partitioning. Briefly, cells were harvested, washed in PBS (20 mM K 2 HPO 4 (pH 7.5), 0.15 M NaCl), and resuspended in lysis buffer (8% (v/v) Triton X-114 in PBS, 5 mM EDTA, 10 mM MgCl 2 ). Cells were then heat-inactivated, disrupted using a French pressure cell and stirred overnight at 4°C. Cellular debris were removed by centrifugation (27,000 ϫ g, 30 min, 4°C), and phase separation was induced at 37°C. Lipoglycans present in the lower phase were precipitated by adding 5 volumes of cold ethanol and collected by centrifugation (27,000 ϫ g, 30 min, 4°C). The pellet was dissolved in water, and proteinase K was added to a final concentration of 10 g/ml for 20 min at 55°C. Proteins were extracted twice by adding saturated phenol. Combined aqueous phases containing lipoglycans were dialyzed for 72 h against water, lyophilized, and resuspended in Tris deoxycholate buffer (10 mM Tris-HCl (pH 8.0), 10 mM EDTA, 0.2 M NaCl, 0.25% deoxycholate). CheLAM and CheLM were then separated by gel filtration on a Sephacryl S-200 (Amersham Biosciences) column (80 ϫ 1.5 cm) in the same buffer. The eluted fractions were monitored by 13% SDS-PAGE stained for carbohydrates according to Tsai and Frasch (27). The appropriate LAM and LM fractions were pooled and dialyzed for 48 h against 10 mM Tris-HCl (pH 8.0) and then for 48 h against water prior to lyophilization. The endotoxin content of all reagents was measured in a chromogenic Limulus lysate assay (BioWhittaker). The LAM preparations contained insignificant amounts of endotoxin (Ͻ25 pg/ml) NMR Analysis-Prior to NMR spectroscopic analysis, LM (15 mg) and LAM (5 mg) were repeatedly exchanged in 2  The one-dimensional proton 1 H spectrum was measured using 90°t ipping angle for the pulse and 1.5 s as a recycle delay between each of 32 acquisitions of 2.4 s. The spectral width of 4006 Hz was collected in 16,384 complex data points. The one-dimensional 13 C was recorded using a spectral width of 20,161 Hz, and 32,768 data points were collected to obtain a free induction decay resolution of 0.6 Hz per point. The 31 P spectra of both compounds were acquired with a spectral width of 16,233 Hz collected in 16,384 data points. Both experiments were recorded using a composite pulse decoupling during acquisition using globally optimized alternating phase sequence at the carbon or phosphorus frequency (28). An exponential transformation (line broadening factor ϭ 5 for 13 C and 3 Hz for 31 P) was applied prior to processing data points in the frequency domain.
Two-dimensional homonuclear ( 1 H-1 H) spectra (COSY-ROESY-TOCSY) were measured using standard Bruker pulse programs. ROESY spectra were acquired with various mixing times (50,100,200, and 400 ms) and acquired in States mode according to Bax and Davis (29), whereas both COSY and relayed COSY were acquired in the magnitude calculation mode. Moreover, the two-dimensional TOCSY spectrum was recorded using a MLEV-17 mixing sequence of 120 ms. The spin lock field strength corresponded to a 90°pulse width of 35 s. The spectral width was 4000 Hz in both dimensions. 512 spectra of 4096 data points with 32 scans per t1 increment were recorded giving a spectral resolution of 0.9 Hz/point in F2 and ϳ8 Hz/point in F1. Heteronuclear experiments ( 1 H-13 C and 1 H-31 P) were obtained with standard Bruker pulse sequences such as HMQC (inv4tp), HMQC-HOHAHA (inv4mltp), and HMBC (inv4lrnd). HMQC and HMQC-HOHAHA were acquired in the phase-sensitive increment time proportional phase increment (TPPI) method, whereas HMBC was recorded in magnitude mode calculation. All parameters (pulse widths, pulse powers, and delays) were optimized for each experiment. Acquisitions and processing conditions are expressed in the figure legends.
Gas Chromatography Techniques-Monosaccharides were analyzed as alditol-acetate derivatives. Lipoglycans were hydrolyzed in 4 N trifluoroacetic acid for 4 h at 100°C and reduced with NaBH 4 in 0.05 N NH 4 OH for 4 h. Reduction was stopped by dropwise addition of acetic acid until the pH reached 6, and borate salts were co-distilled by repetitive evaporation in dry methanol. Per-acetylation was performed in acetic anhydride at 100°C for 2 h, and derivatives were analyzed in GC on a BPX70 12 m ϫ 0.22 mm ID column (Chrompak).
For exact quantification of glycerol and myo-inositol, 100 g of lipoglycans along with 1 g of scyllo-inositol, as an internal standard, were hydrolyzed with 6 N hydrochloric acid constant boiling (Pierce), at 110°C for 24 h, and analyzed as TMS derivatives (30) using GC on a DB-1 60 m ϫ0.25 mm inner diameter by on-column injection.
Linkage analyses of monosaccharides was achieved by two steps of per-methylation according to Ciucanu and Kerek (31) and followed by derivatization with acetyl groups with acetic anhydride.
Acylglycerols were analyzed according to Nigou et al. (26) after cleavage of the phosphodiester bond by acetolysis. Briefly, 100 g of lipoglycans were treated with 400 l of anhydrous acetic acid/acetic anhydride, 3:2 (v/v), at 110°C for 12 h. Acetylated acylglycerols were extracted by cyclohexane and analyzed by GC/MS on a WCOT fused silica 30 m ϫ 0.25 mm inner diameter column (Chrompak).
Fatty acids were analyzed from intact lipoglycans as well as from extracted acetylated acylglycerol as pyrrolidine derivatives. They were released by methyl esterification with 0.5 M HCl in anhydrous methanol at 80°C for 20 h, extracted with heptane, and derivatized with 200 l of pyrrolidine/acetic acid, 9:1 (v/v), at 80°C for 2 h. Pyrrolidine-derivatized fatty acids were repetitively extracted by CHCl 3 /H 3 O, 1:1 (v/v), and analyzed by GC/MS on a WCOT fused silica 30 m ϫ 0.25 mm inner diameter column (Chrompak).
MALDI-Time-of-Flight Mass Spectrometry-The molecular mass of the lipoglycans was measured by matrix-assisted laser desorption ionization on a Vision 2000 time-of-flight instrument (Finnigan Mat) equipped with a 337-nm UV laser. Samples were dissolved in water at a concentration of 100 pmol/l. One l of the solution was mixed with an equal volume of 2,5-dihydroxybenzoic acid (10 mg/ml; dissolved in water/methanol, 4:1 (v/v)) matrix solution on the target and then allowed to crystallize at room temperature.
Western Blot Analysis-Western blot analyses were conducted either on purified LMs and LAMs or on crude mycobacterial lysates. For purified LM and LAM samples, the sugar content was estimated by GC, and the equivalent of 0.5 g of mannose for each sample was analyzed by 13% SDS-PAGE. For crude lysates, mycobacterial cells were harvested, resuspended in 0.8 ml of PBS, and disrupted for 10 min with a Branson Sonifier 450. Protein concentrations were determined using the BCA Protein Assay Reagent kit (Pierce). Equal amounts of proteins (30 g) were then separated by 13% SDS-PAGE and then transferred onto a Hybond-C Extra membrane (Amersham Biosciences). Membranes were then saturated with 5% bovine serum albumin in PBS, 0.1% Tween 20, and probed overnight with either ConA-DIG or GNA-DIG (Roche Molecular Biochemicals, dilution 1:1000). After washing, membranes were subsequently incubated with anti-DIG antibodies conjugated to alkaline phosphatase (Roche Molecular Biochemicals, 1:1000 dilution).
Cytokine Production-The human promonocytic THP-1 cell line was grown in RPMI 1640 (Invitrogen) supplemented with 10% fetal calf serum, 2 mM L-glutamine, and 2 10 Ϫ5 M ␤-mercaptoethanol in an atmosphere of 5% CO 2 at 37°C. THP-1 cells were induced to express CD14 by treatment with 50 nM 1,25-dihydroxyvitamin D 3 (Calbiochem) for 48 h. Cells were then washed twice with RPMI 1640 and cultured in 96-well plastic culture plates at a density of 2 ϫ 10 5 cells/well in RPMI 1640 supplemented with 10% fetal calf serum and glutamine. The purified molecules LAMs from M. Chelonae, M. tuberculosis Erdman, and M. smegmatis were added at a final concentration of 10 g/ml, and triplicates were performed in order to measure cytokine release. Culture supernatants were collected after 6 or 24 h for TNF-␣ or IL-8 production, respectively. Specific enzyme-linked immunosorbent assays commercial kits were used according the manufacturer's instructions. Human IL-8 and TNF-␣ kits were purchased from Bender Med systems Diagnostic and R & D Systems, respectively. Cytokine production was quantified with a microtiter plate reader in comparison with a standard curve generated with recombinant human cytokines.

Purification of CheLAM and CheLM
The experimental protocols used to extract LM and LAM from M. chelonae are based on successive detergent and phenol extractions, leading to the recovery of nucleic acid-, protein-, and lipid-free materials. Purity of the preparation was assessed by GC/MS and SDS-PAGE. CheLAM and CheLM were finally resolved on a gel permeation S200 column. Fig. 1a represents the elution profile of each compound and shows that CheLM was present in higher amounts than CheLAM. This was confirmed by SDS-PAGE analysis of the total Triton X-114 extract (Fig. 1b). The CheLM/CheLAM ratio (w/w) was also determined by routine monosaccharide analysis and revealed that CheLM was three times as abundant as CheLAM. This result was found for several independent extractions (data not shown). It is noteworthy that the relative abundance of these two compounds largely differs from the ones reported previously for other mycobacterial species, in which LAM represents the major component. For instance, an approximate LM/LAM ratio (w/w) of 1:2.5 was found in M. bovis BCG (26) compared with the 3:1 ratio observed in M. chelonae.

Structural Analysis
Because LM is thought to be a biosynthetic precursor of LAM, we first undertook the structural elucidation of LM, and we subsequently determined the structure of the more complex LAM molecule.
Structure of CheLM-The molecular weight of CheLM was investigated by MALDI analysis. It showed a broad unresolved peak centered at 6900 Da (data not shown). Quantitative analysis of the alditol-acetate derivatives of CheLM led to an average composition of 35 mannose, 1 arabinose, and 1 myo-inositol residues. The exact quantification of TMS derivatives by GC using on-column injection demonstrated the presence of 1.07 unit of glycerol per molecule of myo-inositol, which establishes the ratio of mannose/arabinose/inositol/glycerol as 35:1:1:1.
Analysis of partially methylated alditol-acetate derivatives revealed the presence of three major components (Table I) identified on the basis of their retention times and fragmentation patterns as t-Manp, 6-Manp, and 3,6-Manp. Another product, different from a mannitol-acetate, was characterized by (M ϩ NH 4 ) ϩ at m/z 338, indicative of a di-acetylated, tetramethylated inositol. EI/MS fragments at m/z 200, 191, and 75 identified this compound as a 2,6-Ac 2 -1,3,4,5-Me 4 -Ins, which is in agreement with an earlier published report (16). An (M ϩ NH 4 ) ϩ ion at m/z 366 corresponding to a tri-acetylated, trimethylated inositol was detected as a minor component. However, because its fragmentation pattern was unclear, the detailed structure of this product was not analyzed further.
In order to establish the main features of both the polysaccharide and the putative GPI anchor of LM from M. chelonae, an exhaustive NMR-based study was conducted. NMR experiments were recorded successively in D 2 O and in Me 2 SO. As shown previously (26), an improved resolution was obtained in Me 2 SO, which was used for most experiments.
Polysaccharide Moiety-1 H and 13 C NMR parameters of the polysaccharide moiety from CheLM were assigned using onedimensional 1 H and 13 C experiments as well as two-dimensional 1 H-1 H homonuclear and two-dimensional 1 H- 13  NMR parameters from CheLM are summarized in Table II.
The anomeric proton region is dominated by two signals at ␦ 4.94 and ␦ 4.71 ppm (Fig. 2a). The former signal correlates in the COSY 90 experiment with a single H-2 signal at ␦ 3.84 ppm, whereas the other correlates with two largely distinct H-2 signals at ␦ 3.94 ppm and ␦ 3.68 ppm (Fig. 2d), suggesting the presence of two anomeric protons of distinct origins at ␦ 4.71 ppm. The configuration of these three spin systems was unambiguously attributed to Manp in accordance with TOCSY and NOESY experiments. Moreover, the magnitude of the 1 J H1,C1 coupling constant, 168 -170 Hz, provided evidence for the ␣-anomeric configuration of these mannose units. This observation was confirmed by the presence of an intra-residual H-1/ H-2 NOE contact (data not shown).
Based on the complete assignment of its 1 H and 13 C NMR parameters and according to the published literature (14,32), the mannose unit deriving from the anomeric proton at ␦ 4.94 was typified as a terminal ␣-Manp(IV). The signal at ␦ 4.71 ppm correlated with a single C-1 signal at ␦ 100.7 ppm and was tentatively assigned to the anomeric protons of the 6-O-substituted and 3,6-O-substituted mannose units previously observed through methylation analysis. Because of the 1 H-13 C HMQC (Fig. 2b), HMQC-HOHAHA (data not shown), COSY and HMBC experiments (Fig. 2a), all the carbon parameters of these two units could be assigned. In particular, two distinct C-3 signals were identified, one at ␦ 72 ppm corresponding to an unsubstituted C-3, and one deshielded at ␦ 80.3 ppm indicative of a substituted C-3. On the other hand, a single deshielded C-6 resonance at ␦ 66.6 ppm, and a single slightly shielded C-5 resonance at ␦ 72.42 ppm could also be observed. These parameters confirm that the two mannosyl units are 6-O-substituted and 3,6-O-substituted. Altogether, these NMR data support the fact that the major 1 H signal at ␦ 4.71 corresponds to the anomeric proton of both 6-and 3,6-substituted ␣-Manp residues, labeled VI and VIII, respectively. Through 3 J H-C signal assignments, HMBC experiment ( Fig.  2a) allowed us to define intra-and inter-residual correlations originating from the anomeric proton of t-Man (IV), 6-Man (VI), and 3,6-Man (VIII) residues. For t-Man, intra-residual correlations H-1/C-3 and H-1/C-5 were observed at ␦ 71.4 and ␦ 74.26 ppm, respectively, whereas an extra-residual 1-IV/3-VIII correlation was observed at ␦ 80.8 ppm. This unambiguously confirmed the attachment of t-Man residues to the C-3 of 3,6-Man residues. As observed above, the C-3 chemical shift is the only parameter significantly differing between 6-Man and 3,6-Man residues. Thus, the 3 J H-C signals starting from anomeric proton at ␦ 3.67 ppm were attributed as follows: the signal at ␦ 72 ppm was attributed to the intra-residue connectivities 1-VI/3-VI; signals at ␦ 80.31 and ␦ 72.42 ppm to intra-residue connectivities 1-VIII/3-VIII and 1-VIII/5-VIII, respectively; the signal at 66.56 ppm was attributed to inter-residue correlations between the anomeric protons of VI/VIII units and C-6s of VI/VIII units.
These signals demonstrate that VI and VIII residues are connected to each other through 6-O-substitutions. In addition, an intense signal at ␦ 70.5 ppm was also attributed to the 1-VI/ 2-VI correlation. Altogether, these data deriving from methylation and NMR analyses are consistent with the occurrence of a linear backbone of 6-substituted ␣-D-Manp residues partially substituted in C-3 by single ␣-D-Manp residues.
GPI Anchor-The structure of the GPI anchor of CheLM was first investigated through the use of 1 H-31 P HMQC and 1 H-31 P HMQC-HOHAHA experiments (Fig. 3). This method has been successfully applied to study the GPI anchors of LM and LAM isolated from M. bovis BCG by Nigou et al. (33). Coupled to HOHAHA experiments, they allow assigning glycerol, inositol, and mannose residues that constitute the GPI anchor.
The 31 P spectrum of LM in Me 2 SO showed only two sharp peaks at ␦ 1.65 and ␦ 1.85 ppm, indicative of the presence of two distinct GPI anchors. Integration of these signals gave us a P1/P3 ratio of 73:27. Assuming that the phosphorus atom of the GPI anchor is linked to both glycerol and myo-inositol, we used the 1 H-31 P HMQC experiment to assign Gro H-3 and myo-Ins H-1 (Fig. 3b). As expected, each phosphorus atom correlated with two slightly different sets of signals that were identified by analogy to the literature as myo-Ins H-1 (at ␦ 4.03 and ␦ 3.93 ppm, respectively) and Gro H3/H3Ј (at ␦ 3.80 and ␦ 3.85 ppm, respectively). Then 1 H-31 P HMQC-HOHAHA experiments (Fig.  3a) permitted to assign the proton spin systems of myo-Ins and Gro linked to each phosphorus atom. The assignment of these signals was confirmed by the use of sequential multirelayed 1 H-1 H COSY and HOHAHA (data not shown) sequences, using a mixing time of 120 ms. It appeared that the glycerol protons derived from both phosphorus atoms presented very similar parameters: H-1 at ␦ 4.32-4.35 ppm; H-1Ј at ␦ 4.09 -4.11 ppm; H-2 at ␦ 5.07-5.09 ppm; and H3/H3Ј at ␦ 3.80 -3.85 ppm (Fig.  3d). Their most relevant feature was the presence of a very deshielded H-2 proton at ␦ 5.07-5.09 ppm, characteristic of a 1,2-di-acylated glycerol (34). On the other hand, the two myoinositols exhibited distinct 1 H NMR parameters; in particular, the H-3 signals differed by about 1.4 ppm at ␦ 4.57 and ␦ 3.20 ppm. Such a deshielding is in agreement with an acylated position at C-3 occurring on one of the two myo-inositol types (Fig. 3c). Sets of parameters of Gro and myo-Ins matched perfectly with some of the previously analyzed GPI anchors (33,35). By using the nomenclature proposed by these authors, we could typify the two phosphorus atoms at ␦ 1.65 ppm as P1 and at ␦ 1.85 ppm as P3, where P1 containing GPI is characterized by the presence of a diacylglycerol and a C-3-acylated myoinositol, whereas P3 containing GPI is characterized by the presence of a diacylglycerol and non-acylated myo-inositol.
Four minor deshielded anomeric protons were clearly observed in the 1 H spectrum at ␦ 5.15, ␦ 5.13, ␦ 5.19, and ␦ 5.11 ppm (Fig. 4). On the basis of their C-1 parameters at ␦ 98.7, ␦ 98.7, ␦ 100.96, and ␦ 101.5 ppm, respectively, as observed on the 1 H-13 C HMQC spectrum (Fig. 5), and their H-2 parameters at ␦ 3.61, ␦ 3.58, ␦ 3.79, and ␦ 3.78 ppm as observed on the HOHAHA spectrum (Fig. 4c), these four residues were tentatively identified as Manp residues. On the ROESY and NOESY spectra (Fig. 4, a and b), all four anomers showed intra-residue correlations with their respective H-2, which substantiate their attribution as ␣-Manp residues. Moreover, starting from H-1 signals at ␦ 5.19 ppm, NOESY and ROESY spectra showed intense NOE effect with H-2 myo-Ins P1 at ␦ 4.22 ppm, whereas starting from signal at ␦ 5.11, NOESY and ROESY showed NOE effect with H-2 myo-Ins P3 at ␦ 4.19 ppm. A weaker correlation signal was also observed on the NOESY spectrum between anomeric proton at ␦ 5.19 ppm and H-3 Ins P1 at ␦ 4.57 ppm. These data clearly demonstrate the glycosylation of both myo-inositol types P1 and P3 on C-2 position by a mannose residue, labeled Man-2 (P1) and Man-2 (P3), respectively. Similarly, anomeric protons at ␦ 5.15 and ␦ 5.13 ppm correlated with their respective H-2 signals and with myo-Ins P1 H-6 at ␦ 3.66 ppm and myo-Ins P3 H-6 at ␦ 3.62 ppm, respectively. This confirmed the substitution of P1 and P3 myo-Ins at C-6 by mannose residues labeled Man-1(P1) and Man-1(P3). These data are in total agreement with previous studies (9, 16, 26) on GPI anchors of PIMs, LM, and LAM where inositol was invariably substituted at the C-2 by a single ␣-D-Manp residue and at C-6 by the mannan core. However, our experiments did not enable us to distinguish between a terminal mannose residue and substituted mannose residues. Considering the relative proximity of the acylation position on the myo-Ins residue with the two glycosylation positions, it is noteworthy that the parameters of Man-2 H-1 protons are highly affected by the acylation of the myo-Ins, whereas the Man-1 H-1 protons are not.
The acylation state of the glycerol moiety was subsequently studied by GC/MS by liberating the intact acylglycerol by acetolysis as described previously by Nigou et al. (26). To confirm the exact nature of substituting fatty acids, these were cleaved from the acetylated acylglycerol by methanolysis and analyzed as pyrrolidine derivatives by GC/MS. The occurrence of hydroxylated fatty acids was specifically investigated by derivation with heptafluorobutyric anhydride (36). Acetolysis products appeared as being predominantly constituted of diacylglycerols. Quantification of each form showed that monoacylglycerols represented less than 1% of the total acylglycerols analyzed. They included hexadecenoic acid (C 16:1 ), octadecenoic acid (C 18:1 ), and nonadecanoic acid (C 19 )-substituted glycerol. This is in agreement with the 1 H-31 P HMQC NMR experiments that only showed the presence of a diacylglycerol unit in CheLM through the observation of P1-and P3-type phosphorus atoms. The region of the EI total ion current chromatogram profile corresponding to the diacylated products exhibited 12 peaks that were all assigned by EI and CI/MS (Fig. 6a). The three most abundant species were assigned as 1/2-tetradecanoyl-1/2-nonadecanoyl-3-acetylglycerol, 1/2-hexadecanoyl-1/2octodecenoyl-3-acetylglycerol, and 1/2-hexadecanoyl-1/2-nonadecanoyl-3-acetylglycerol, which represent 45, 34 and 13%, respectively, of the total acylglycerols isolated from CheLM. They were easily characterized because of the (M ϩ NH 4 ) ϩ ions at m/z 642, 654, and 670, respectively, and to the fragment ions resulting from the loss of each acyl group (m/z 327 and 397 for C 14 -C 19 Gro, m/z 355 and 381 for C 16 -C 18:1 Gro, and m/z 355 and 397 for C 16 -C 19 Gro). Each of these compounds was attributed to two peaks showing a constant ratio of 1:5. These were tentatively attributed to the two possible positions of fatty acids, but the ions resulting from the fragmentation between C-1 and C-2 of the glycerol were not intense enough to unequivocally localize the acyl group on each carbon and thus to determine the predominant form. Similarly, the remaining peaks were attributed to a very heterogeneous family of diacylglycerols, representing less than 10% of the total products. They include C 14 -C 18:1 , C 16 -C 16:1 , C 16 -C 16 , C 16 -C 17:1 , C 15 -C 19 , C 16 -C 18 , C 16 -C 19:1 , and C 17 -C 18:1 di-substituted glycerols. Further fatty acid analysis based on acetolysis products indicated that the octadecenoic acid was present as a mixture of C 18:1 ⌬ 8 and C 18:1 ⌬ 9 and permitted the identification of nonadecanoic acid as tuberculostearic acid (10-methyloctadecanoic acid).
Structure of CheLAM-The structure of CheLAM was elucidated by the same experimental protocols used for CheLM. Analysis of CheLAM by MALDI showed a broad peak centered on m/z at 17,000 Da and ranging from 9000 to 23,000 Da. It is noteworthy that in identical experimental conditions CheLAM gives a much more intense response in MALDI analysis than CheLM. Composition analysis of this compound showed a mannose/arabinose/inositol/glycerol ratio of 31:80:1:1. Analysis of partially methylated and acetylated monosaccharides residues indicated the presence of t-Man, 6-Man, 3,6-Man, t-Ara, 5-Ara, 2-Ara, and 3,5-Ara (Table I). These experiments led to a first insight into the nature of the capping of LAM isolated from M. chelonae. Indeed, both CheLM and CheLAM showed a sim-  (14,15). This observation was reinforced by the fact that proportions of t-Man, 6-Man, and 3,6-Man were similar in both CheLM and CheLAM and that no other type of Man unit was present in CheLAM (Table I). Calculation of the degree of capping based on the ratio of 3,5-Araf minus t-Araf to 3,5-Araf, providing that the capping motif is not alkali labile, gives a very low capping percentage of about 5%. This result is also in accordance with the absence of an oligomannosyl cap. The use of another calculation method, based on the ratio of 2-Araf minus t-Araf to 2-Araf (19), gives a capping percentage below zero due the ratio of 2-Araf to t-Araf which is inferior to 1. The significantly lower relative abundance of 2-Araf compared with t-Araf may be due either to the fact that some of the t-Araf residues may not be linked to 2-Araf residues or to an underestimation of the relative abundance of 2-Araf in our experimental conditions. However, such a discrepancy between the two methods has already been observed (37). Similarly, the presence of a single myo-inositol residue per molecule of LAM suggests that no other myo-inositol residue than the one incorporated in the GPI anchor is present in the molecule, and consequently that arabinan chains are not terminated by phosphomyo-inositol groups as observed in fast growing Mycobacterium sp. (19) including M. smegmatis (38). Both observations were confirmed by multiple NMR experiments, as shown below. Attributed 1 H and 13 C NMR parameters from CheLAM are summarized in Table III. Comparison of the anomeric region of 1 H-13 C HMQC spectra from CheLM and CheLAM (Fig. 5) allowed us to identify signals from the mannan core of CheLAM. On this basis, the intense 13 C resonance at ␦ 103,1 ppm was tentatively identified as the t-␣-Manp anomeric carbon (IV), whereas the 13 C resonance at ␦ 100.55 was identified as the 6-␣-Manp and 3,6-␣-Manp anomeric carbons (VI and VIII). Two-dimensional 1 H-1 H homonuclear experiment confirmed that the 1 H/ 13 C signal at ␦ 4.92/103.1 ppm correlated with a single spin system characteristic of a t-Manp residue, whereas the 1 H/ 13 C signal at ␦ 4.70/100.55 ppm correlated with two distinct spin systems characteristic of 6-Manp and 3,6-Manp units (data not shown). Assignment of their respective 13 C resonances owing to 13 C-1 H heteronuclear experiment confirmed this observation. Furthermore, as observed in CheLM, an HMBC experiment showed inter-residual correlations between IV H-1 and VI C-3, VI H-1 and VI/VIII C-6, and between VIII H-1 and VI/VIII C-6 (data not shown). These data confirm that the mannan cores of CheLM and CheLAM are similar. Fig. 5 shows that four minor signals previously assigned to anomeric carbons of ␣-Man-1P1 and P3 and ␣-Man-2P1 and P3 in CheLM were also clearly observed in the CheLAM spectrum. The remaining anomeric signals of the spectrum were attributed by comparison with previous spectral NMR data of LAM from M. tuberculosis (32) and M. smegmatis (38). Spin systems deriving from each anomer was identified owing to 1 H-1 H homonuclear and 13 C-1 H heteronuclear experiments. This way the remaining signals in the anomeric region of the spectra were all attributed to the anomeric signals of the arabinan core of CheLAM, in agreement with the methylation analysis: 3,5-␣-Araf (I), 5-␣-Araf (II 1 to II 4 ), 2-␣-Araf (III 1 , III 2 , and III 3 ), and t-␤-Araf (V 1 and V 2 ). No additional mannose residues could be identified. Most types of Ara units showed multiple anomeric signals presenting identical spin systems. This multiplicity of signals was previously attributed to the different positions that each unit may take within the arabinan core (14). HMBC spectrum showed a complex pattern of 3 J H-C correlations that was entirely resolved, except for the region between ␦ 4,95 and 5 ppm where all intra-residue connectivities from III 1 , III 2 , II 3 , and II 4 residues could not be unambiguously attributed because of the very close chemical shift of their anomeric protons (not shown). Nevertheless, this shed some light onto the general sequence of the arabinan core. In particular all the anomeric protons of 3,5-␣-Araf residues (I) and 5-␣-Araf residues (II 1 to II 4 ) showed intense 3 J H-C correlation signals with a substituted C-5 at ␦ 67.9 ppm, unambiguously attributed as inter-residue connectivities with C-5 I and C-5 II. Similarly, both H-1 of t-␤-Araf residues (V 1 and V 2 ) showed inter-residue connectivities with a very deshielded carbon at ␦ 88.29 ppm attributed to the C-2 of 2-␣-Araf, confirming the attachment of all t-␤-Araf at C-2 of 2-␣-Araf residues. On the other hand, it is noteworthy that no other connectivity with the C-2 of 2-␣-Araf residues could be observed suggesting that no arabinose residues other than t-␤-Araf residues are linked to 2-␣-Araf units. HMBC experiment also enabled us to distinguish III 3 residues from III 1 and III 2 residues on the basis of their respective 3 J H-C connectivities. Indeed, whereas H-1 III 1 /III 2 showed an intense inter-residue correlation with C-5 I/IIs at 67.9 ppm, H-1 III 3 did not show any correlation at ␦ 67.9 ppm but an intense and broad signal around ␦ 83.5 ppm attributed to C-4-III 3 and to C-3-I. These data establish that III 1 and III 2 residues are linked to C-5 of 3,5-or 5-␣-Araf residues, whereas III 3 is linked to C-3 of 3,5-␣-Araf. Therefore, NMR data arising from the arabinan chain of CheLAM show very little discrepancy with published NMR data of LAM from M. tuberculosis (32) and M. smegmatis (38), strongly suggesting that a similar arabinan core is shared by the three mycobacterial species. On the other hand, methylation and composition analyses have previously suggested that M. chelonae differs in its capping status from M. tuberculosis and M smegmatis. Indeed, as already mentioned, homonuclear 1 H-1 H and heteronuclear 1 H- 13 C NMR experiments showed no evidence of the presence of other mannose residues than the one found in the mannan core, suggesting the absence of oligomannosyl caps on the arabinan side chains as observed in M. tuberculosis and M. bovis. Furthermore, it was observed on the HMBC spectra  F2 ( 1 H). The original data matrix were expanded to 4096 ϫ 512 real matrix moreover for processing a sine-bell window shifted by ⌸/2 was applied in both dimensions. c, expanded region (␦ 1 H, 5.20 -3.00, and 4.60 -4.54); d, expanded region (␦ 1 H, 5.20 -3.00, and 4.40 -4.28) of homonuclear two-dimensional clean TOCSY spectrum. This experiment was recorded with a spin-lock time of 120 ms. The data matrix was 2048 ϫ 512 (states) points with 32 scans per t1 increment and was expanded to 4096 ϫ 1024. that no additional 3 J H-C correlation occurred between any anomeric protons of mannose residues (IV, V, and VIII) and any carbon of arabinose residues. The possibility of the occurrence of phospho-myo-inositol type capping motifs is discussed below.
In order to study the GPI anchor of CheLAM isolated from M. chelonae, we used an experimental approach used above for CheLM.  that GPI of CheLAM and CheLM shared identical acylation states, characterized by a mixture of P1 and P3 phosphates, in similar ratios. Detailed analysis of the acetolysis products of CheLAM by GC/MS showed again that diacylglycerols of CheLM and CheLAM present the same heterogeneity, reinforcing the hypothesis that CheLM and CheLAM share strictly identical GPI anchors.
It is noteworthy that the 1 H-31 P HMQC spectrum did not reveal any other type of phosphorus than those attributed to the GPI anchors of the lipopolysaccharide. In particular, no evidence for the presence of an additional phospho-myo-inositol group substituting the arabinan side chains was found, as observed in LAM from M. smegmatis (38). Altogether, with the results of the composition analysis showing a myo-inositol/ glycerol ratio of 1:1, the NMR data confirmed that CheLAM does not possess a phospho-myo-inositol capping motif.

Mapping with Lectins
Occurrence of a new type of mannan core in lipoglycans isolated from M. chelonae was investigated by screening purified LM and LAM from various mycobacterial species with several mannose-recognizing lectins. The most relevant results were obtained with lectins isolated from Canavalia ensiformis (ConA) and from Galanthus nivalis (GNA). Although ConA is relatively unspecific toward mannose-containing glycoconjugates, GNA is known to be highly specific for non-reducing terminal ␣1,3-linked mannose residues (39). As expected, ConA interacted with purified LM from M. chelonae, M. bovis BCG, and M. smegmatis (Fig. 7a). This suggests that ConA does not differentiate mannan cores branched at C-2 (M. bovis BCG and M. smegmatis) from mannan cores branched at C-3 (M. chelo-nae). Surprisingly, this lectin only recognized purified LAM from M. tuberculosis (Fig. 7a), but not LAM from M. chelonae or LAM from M. smegmatis, although LM and LAM share common mannan cores. The absence of reactivity between ConA and LAM from M. chelonae or M. smegmatis may be explained by steric hindrance of the mannan core by the arabinan polymer. Specific interaction observed between ConA and LAM from M. tuberculosis would originate from the presence of an oligomannosyl-type capping only in this species. These results are in agreement with findings from Prinzis et al. (40) who reported that ConA recognizes purified LAM from M. bovis BCG and from M. tuberculosis Erdman but not from a rapidly growing strain.
GNA exhibited a much more restricted specificity than ConA. As observed in Fig. 7b, GNA recognizes exclusively CheLM. Considering the described specificity of GNA toward a Man(␣1-3) determinant, this result is in total agreement with direct structural studies in this report that distinguish the mannan core of CheLM from those of other mycobacterial species on the basis of their respective branching positions. As observed for ConA, GNA did not bind to CheLAM, again suggesting that arabinan masks the ␣1,3-Manp side chains of the mannan core.
Given the specificity of GNA for the Man(␣1,3)-substituted mannan core of CheLM, we addressed the question whether this lectin may be particularly useful to rapidly detect the presence of similar structures in various other mycobacterial species, including rapid, slow growing, pathogenic, and non-pathogenic species Mycobacterium xenopi, Mycobacterium gastri, and Mycobacterium malmoense were transferred to a membrane, incubated with GNA-DIG, and revealed with anti-DIG antibodies. Western blot analysis shows that Man(␣1,3)-substituted mannan core was only present in M. chelonae (Fig. 7c). Interestingly, GNA did not interact with LM from M. fortuitum, known to be a M. chelonae related species. This indicates that GNA constitutes a useful tool for a rapid screening of LM and LAM structures with ␣1,3-linked mannose residues without the need of a purification step. It also suggests that among the nine mycobacterial strains analyzed, the structure of M. chelonae appears to be unique.

TNF-␣ and IL-8 Secretion by Various LAM Preparations
To investigate the consequences of the structural differences established between CheLAM, ManLAM, and AraLAM, we compared TNF-␣ and IL-8 secretion by differentiated human macrophages stimulated with LAMs isolated from M. tuberculosis, M. smegmatis, and M. chelonae. As shown in Fig. 9, incubation THP-1 cells with LAM isolated from M. smegmatis resulted in a sharp increase of both TNF-␣ and IL-8 secretion. In contrast, LAMs from M. chelonae and M. tuberculosis were not able to induce cytokine production. Previous studies (38,41,42) have been reported that mannose-capped LAM from M. tuberculosis was inactive toward TNF-␣ and IL-8 secretion, whereas phosphoinositol-capped LAMs from a fast growing Mycobacterium species and from M. smegmatis were strong cytokine inducers. Considering the fact that CheLAM is neither capped by oligomannosyl nor by phosphoinositol groups, our results strongly suggest that TNF-␣ and IL-8 secretion by LAM was very likely modulated by the presence of the phosphoinositol-type capping. This also suggests that the mannose type capping does not influence cytokine induction.  (44) are mannosecapped (ManLAM) to varying degrees. Studies based on a fast growing Mycobacterium sp. and on M. smegmatis demonstrated that the arabinan termini are uncapped (AraLAM), whereas a minor portion terminates with inositol phosphate caps (19,38,40).
So far, LAM structures from only a restricted panel of mycobacterial species have been determined. The relatively low structural diversity encountered in these molecules may limit the possibilities of studying relationships between structure and biological activities. The aim of the present work was to decipher new sources of structural variability in LM and LAM in order to generate new models for such studies. We therefore used M. chelonae as a model. M. chelonae is a nontuberculous mycobacterium that is usually encountered in water sources and soil. Although it has originally been proposed to be Friedmann's turtle tubercle bacillus (45), it sometimes causes disease in humans. M. chelonae presents also the advantage of being a rapidly growing species. Furthermore, we have recently shown that it can be transformed, thus making it attractive for future genetic studies (24). Considering the intrinsic polydispersity of mycobacterial lipoglycans, structural studies were undertaken, consisting of an independent comparison of each structural feature of these molecules with their homologs that have been studied in other species. The first observation was that LAM from M. chelonae shares a similar overall structure with other LAMs, composed of an arabinan chain and a mannan chain linked to a GPI anchor. However, it shows unique features that unequivocally distinguishes it from other LAMs. Significant differences observed in M. chelonae are illustrated in Fig. 8. Considering the growing inter-specificity heterogeneity observed in LAMs and, as proposed by Khoo et al. (19), we referred to lipoglycans isolated from M. chelonae CheLM and CheLAM, according to their origin, rather than to their structural features.
In this study, we demonstrate for the first time that a virulent mycobacterium species exhibits a LAM structure that is devoid of mannose and inositol phosphate capping motifs (Fig.  8). Therefore, both CheLM and CheLAM represent suitable molecules to use along with ManLAMs or AraLAMs to evaluate the role of mannose and inositol phosphate capping in relation to various biological functions attributed to LAM. As an example, it has been observed for a long time that LAMs extracted from avirulent mycobacterial species were more potent inducers of TNF-␣ than those extracted from virulent species (46). A correlation was rapidly made between the absence of an oligomannosyl capping and the potency to trigger TNF-␣ release. It was also reported that the activity of LAM from the avirulent strain M. smegmatis was diminished after alkali treatment (38), suggesting the importance of alkali-labile phosphoinositol groups in eliciting cytokine production. The present study confirms this observation by showing that a LAM naturally lacking mannose and phosphoinositol caps does not induce cytokine secretion.
In addition, our results demonstrate that both CheLM and CheLAM possess an unusual mannan core characterized by the presence of ␣1,3-Manp side chains, rather than the ␣1,2-Manp substitution as found in all mycobacterial LAM structures reported so far (Fig. 8). Analysis by Western blotting using GNA, an ␣1,3-Manp-specific lectin, revealed that ␣1,3-Manp side chains containing LM were only present in M. chelonae but not in other mycobacterial species. Whether these substitutions are important for biological activities of CheLM or CheLAM  M. smegmatis (c). The emphasis was put on the parts of the molecules showing inter-specific heterogeneity, in particular the nature of capping, the substitution of the mannan core, and the acylation state of the GPI anchor. Despite subtle structural differences between LAM from M. tuberculosis and M. bovis BCG, both species do not show important discrepancies concerning the sus-mentioned points and were therefore associated. When known, nature of the fatty acids that substitute glycerol was indicated. C-X represents 12-O-(methoxypropanoyl)-12-hydroxystearic acid. Only certified acylation positions are indicated. remains to be investigated. Another substantial difference arising from the present study concerns the LM/LAM ratio. It is generally assumed that LM is a precursor of LAM and is usually present in smaller amounts than LAM (26). We observed that M. chelonae contains a LM/LAM ratio of 3:1 (w/w), corresponding to ϳ10 times more molecules of CheLM than CheLAM. The presence and the possible physiological impact of larger amounts of CheLM than CheLAM in M. chelonae is presently unknown. This difference in the CheLM/CheLAM balance may arise from a deficiency in arabinosyltransferase(s) expression in M. chelonae compared with M. tuberculosis and M. bovis BCG. On the other hand, arabinosyltransferase(s) involved in arabinan polymerization may be less active in M. chelonae than in M. tuberculosis. At present, this remains very difficult to investigate because no arabinostransferases involved in LAM biosynthesis have been clearly identified.
Another unexpected difference between CheLAM and Man-LAMs concerns the lipid part of the PI anchors. Nigou et al. (26) have demonstrated that cellular ManLAM anchors from M. bovis BCG mainly contain 1-or 2-palmitoylglycerol, 1-or 2-tuberculostearoylglycerol, 1,2-dipalmitoylglycerol, and 1-tuberculostearoyl-2-palmitoylglycerol. Surprisingly, the same authors found that the PI anchor of parietal ManLAM from M. bovis BCG is composed of a single type of acylglycerol consisting of 1-[12-O-(methoxypropanoyl)-12-hydroxystearoyl]glycerol. In addition, parietal ManLAM was found to be a better inducer of TNF-␣ and IL-8 by human dendritic cells than cellular LAM, thus suggesting that the unusual 1-[12-O-(methoxypropanoyl)-12-hydroxystearoyl]-glycerol part is the major cytokine-regulating component of the ManLAMs (26). Moreover, the fact that biological activities of ManLAMs are abrogated after deacylation and that ManAMs devoid of the PI anchor are unable to stimulate cytokine production are consistent with the assumption that the lipid part of the PI anchor is essential for the immunological properties of the ManLAMs. The present study shows that the lipid part of the PI anchor of CheLAM is fundamentally different from that of previously described PI anchors because it presents a very large heterogeneity, due to a combination of fatty acids ranging from myristic acid to tuberculostearic acid with eventual unsaturations (Fig. 8). This heterogeneity generates a large pool of molecules with various PI anchors, which may influence the biological activities of the whole molecule. Whether this heterogeneity is due to numerous acylglycerol acyltransferases with different specificities in M. chelonae remains an open question. It is also noteworthy that CheLM and CheLAM show perfectly identical mixtures of diacylglycerols. This observation substantiates the hypothesis that LM is a biosynthetic precursor of LAM.
Mycobacteria represent a major source of microbial antigens that are presented by CD1 molecules. Most of the lipids that can be presented by CD1 molecules have two hydrophobic tails and a polar head group. This led to the suggestion that each hydrophobic tail fits into a pocket of the CD1 groove (47). Binding studies also revealed that the lipid portion of the antigen is required for CD1 binding (48,49). Sieling et al. (50) have shown that ManLAMs from M. tuberculosis Erdman and M. leprae can be presented in the context of CD1 molecules and are able to stimulate CD4/CD8 double negative ␣␤ T cells. How different CD1 molecules bind, in particular LM/LAM, is not known. Available data suggest that human CD1b and CD1d molecules require a minimal hydrocarbon chain length but that they are not as selective with regard to the type of lipid they will bind (48,49). Therefore, it would be interesting to study the PI anchors of CheLM or CheLAM in relation to CD1 binding for presentation of these glycolipids to T cells. Given their unusual structures, CheLM and CheLAM constitute new tools to compare the biological activities of LMs and LAMs such as for antigen presentation via CD1 molecules, cell adhesion, or T cell activation and should help to clarify their roles in mycobacterial virulence.