Lipidomic analyses of Mycobacterium tuberculosis based on accurate mass measurements and the novel "Mtb LipidDB".

The cellular envelope of Mycobacterium tuberculosis is highly distinctive and harbors a wealth of unique lipids possessing diverse structural and biological properties. However, the ability to conduct global analyses on the full complement of M. tuberculosis lipids has been missing from the repertoire of tools applied to the study of this important pathogen. We have established methods to detect and identify lipids from all major M. tuberculosis lipid classes through LC/MS lipid profiling. This methodology is based on efficient chromatographic separation and automated ion identification through accurate mass determination and searching of a newly created database (Mtb LipidDB) that contains 2,512 lipid entities. We demonstrate the sensitive detection of molecules representing all known classes of M. tuberculosis lipids from a single crude extract. We also demonstrate the ability of this methodology to identify changes in lipid content in response to cellular growth phases. This work provides a customizable framework and resource to facilitate future studies on mycobacterial lipid biosynthesis and metabolism.

oleic acid-dextrose catalase (Difco) and 0.05% Tween 80 (Sigma-Aldrich; St. Louis, MO) were inoculated with 2 ml of a logarithmic culture to achieve a theoretical optical density at 580 nm (OD 580 ) of 0.004. The fl asks were capped tightly and incubated at 37°C under constant stirring at 150 rpm. The OD 580 of each culture was measured at regular intervals for 11 days, and aliquots (10 ml) were taken at 73, 108, and 265 h, corresponding to logarithmic, transitionary, and stationary growth phases, respectively. The cell pellets were harvested and washed three times with PBS, pH 7.4. Crude lipid extracts were prepared as described above and randomized for LC/MS analyses.

LC
An Agilent 1200 HPLC (Agilent Technologies; Palo Alto, CA) with a 2.1 inner diameter (ID) × 150 mm, 3.5 µm XBridge C18 column (Waters Corp.; Milford, MA) heated to 45°C was used with a binary solvent system and a fl ow rate of 320 µl/min. A 2.1 ID × 10 mm, 3.5 µm XBridge C18 guard column (Waters) was placed in series in front of the analytical column. The system was equilibrated with 100% solvent A [5 mM ammonium acetate in methanol-water (99:1; v/v)], and an aliquot of the lipid extract (5 µl = 20 µg dried extract) was applied to the column. Solvent A was maintained at 100% for 2.0 min, followed by a 30.0 min linear gradient to 100% solvent B [5 mM ammonium acetate in n -propanol-hexane-water (79:20:1; v/v/v)], and held at 100% solvent B for 3.0 min. All solvents and chemicals purchased were MS or HPLC grade.

MS
An Agilent 6220 time-of-fl ight (TOF) mass spectrometer equipped with an Agilent ESI/atmospheric-pressure chemical ionization (APCI) multimode source was used for accurate mass analysis of the LC eluent. Positive-(+) and negative-( Ϫ ) ion data were generated by operation of the mass spectrometer in a mixed ESI/APCI mode with a capillary voltage of 3000 V, nebulizer of 45 psig, drying gas of 8 l/min, gas temperature of 300°C, vaporizer temperature of 200°C, corona of 2 µA, fragmentor of 120 V, charging voltage of 2000 V, skimmer of 60 V, and octopole radio frequency voltage of 250 V. Mass spectra were acquired in 4 GHz high-resolution mode at a rate of 1.02 spectra/s and 9,700 transients/spectrum, and data were collected as profi led spectra over a mass range of 250 to 3,200 Da. Mass calibration was performed with an Agilent tune mix from 100 to 2,700 Da, and an external reference sprayer introduced mass ions of m/z 922.009798 (+ ion) and m/z 980.016375 ( Ϫ ion) to enable accurate mass determinations. Data were collected with the Agilent MassHunter WorkStation Data Acquisition software, version B.02.00.
An Agilent 6520 qTOF was used for MS/MS analyses of the triacylglycerol (TG) lipid-group [M+NH 4 ] + ions. The instrument setup was the same as described above, except that the OCT RF was set at 750 V. Positive-ion mass spectra were acquired in Auto MS/MS mode, and collision energies with slope of 6.5 V/100 Da and offset 2.0V were used for fragmentation.

Mtb LipidDB development
To enable accurate mass MS-based searching of lipid ions, the novel database " Mtb LipidDB" was created with Microsoft Excel 2007 Pro (Microsoft; Redmond, WA). The database was populated with molecular formulas, structures, and exact masses for the known lipids of Mtb . This information was obtained from a survey of the extensive Mtb biochemical literature describing lipid structures purifi ed from Mtb H37Rv and Mycobacterium bovis Bacillus Calmette-Guérin (BCG) strains (see references in supplementary material). The collective data from these principal publications were used to calculate every theoretically possible fatty acyl combination for each lipid subclass and level 4 class. The reported structural variations were used to generate a representative molecular structure for resonance mass data obtained from direct infusions of complex Mtb lipid extracts. The algorithm isolates ions from spectra, and assigns identities from a "user-defi ned" lipid library based on exact mass. The methodology facilitated rapid comparisons of highly complex spectra and yielded signifi cant fi ndings on the metabolic control of virulence lipids ( 15 ). Shui et al. ( 24 ) took an alternative strategy and demonstrated the utility of C18 reversedphase HPLC in combination with ESI-MS to effi ciently separate and detect complex lipids of lower abundance. Several anionic Mtb lipid classes were characterized, and lipid profi ling was able to identify mycolic acid profi le shifts that occurred in response to different physiological growth conditions. However, these previous MS-based analyses were not supported by a complete and portable Mtb lipid database, and did not fully integrate database interrogation with the ability to resolve individual lipids by HPLC and ion data extraction techniques. Building upon the previous success of MS-based lipidomic approaches for Mtb and to provide tools for comprehensive lipid profi ling, we generated a novel database of Mtb lipids, " Mtb LipidDB" that allows for lipid identifi cation using accurate mass measurements obtained in either negative-or positive-ion modes. When coupled to molecular feature (MF) detection from LC/MS spectra, the Mtb LipidDB allows for automated exact mass-based lipid identifi cation. Applying this approach to crude Mtb lipid extracts, we achieved identifi cation of the majority of known extractable lipids from this bacterium. Moreover, we demonstrated the utility of this methodology to rapidly identify quantitative changes that occur in the lipid profi le of Mtb in response to growth conditions.

Bacterial growth and lipid extractions
Mtb strain H37Rv was propagated in glycerol alanine salts medium ( 25 ) at 37°C for 14 days, as previously described ( 26 ). Cells were harvested by centrifugation (3,000 g for 10 min), washed three times with PBS, pH 7.4, and inactivated by ␥ irradiation ( 27 ). To prepare crude lipid extracts, a modifi ed Bligh and Dyer ( 28 ) method was used. Briefl y, 200 mg of wet cells were lyophilized in a 13 × 100 mm silanized borosilicate glass tube with a Tefl on screw cap. The dried cells were extracted with 6 ml of chloroform-methanol-water (10:10:3; v/v/v) overnight, with constant stirring at room temperature. The sample was centrifuged at 3,000 g for 10 min to remove the delipidated cells. The monophasic lipid extract was transferred to a 13 × 100 mm glass tube, dried under a gentle stream of nitrogen, and stored at Ϫ 20°C until use. Prior to LC/MS analysis, the dried lipid extract was dissolved in chloroform-methanol-water (10:10:3; v/v/v) to approximate a 4 µg/µl concentration, centrifuged at 3,000 g for 10 min, and transferred to an autosampler vial. To prepare the dilution series, cells were homogenized by bead-beating and serially diluted 2-fold with water in glass tubes. The diluted homogenates were lyophilized, and lipids were extracted as described above. Based on the starting wet weight of the cell pellet, the lipid dry weights were estimated from routinely observed lipid extraction yields.
For Mtb H37Rv growth phase comparisons, three 500 ml Nephelo-sidearm fl asks containing 50 mm stirbars and 200 ml Middlebrook 7H9 broth (Difco; Detroit, MI) supplemented with Santa Clara, CA). MFs were extracted from the raw data using the MF extraction (MFE) algorithm. This algorithm locates related covariant ions (isotopes and charge states) from accurate mass LC/MS data, and combines these ions into a single feature. The MFE parameters used were: extraction algorithm, small molecule; peak fi lters, у 500 counts; ion species, +H and Ϫ H only; peak spacing tolerance, 0.0025 m/z plus 7.0 ppm; isotope model, common organic molecules; charge state, 1-2; compound fi lters, none; mass fi lters, none; and mass defect, none. The resulting MFs were then identifi ed with MassHunter by searching the +MH_MtbLipid. csv (+ ion data) or the Ϫ MH_MtbLipid.csv ( Ϫ ion data) database fi le with the following search parameters: values to match, mass only; match tolerance, 5 ppm; charge carriers, +H and Ϫ H; and charge state range, 1-2. The lists of both identifi ed and unidentifi ed features were exported as an analysis report in Microsoft Excel (Microsoft) format. Further data comparisons were accomplished with Microsoft Excel. For quantitative purposes, lipid groups were compared using the most-dominant molecular ion volumes observed for each lipid subclass as follows: TG,

Development of the novel Mtb LipidDB
The Mtb LipidDB (supplementary data and summarized in Table 1) was organized in a manner that adhered as closely as possible to the classifi cation hierarchy and structural nomenclature set forth by the LIPID MAPS consortium (LIPID Metabolites and Pathways Strategy; http:// www.lipidmaps.org) ( 30,31 ). Six of the eight lipid categories defi ned by LIPID MAPS are represented in the Mtb LipidDB. The Mtb LipidDB contains 15 lipid main classes, 46 lipid subclasses, and 16 level 4 lipid classes. In instances where Mtb lipids did not fi t into the LIPID MAPs classifi cation system, 30 novel lipid subclasses (e.g., diacyltrehaloses) and 16 level 4 classes [e.g., ␣ -mycolic acids (Alpha-MA)] were created. These novel subclasses and level 4 classes were developed for Mtb LipidDB organization purposes only, and have not been accepted by LIPID MAPS as part of their classifi cation hierarchy. A key organizational difference between the Mtb LipidDB and LIPID MAPs occurs at the species level of classifi cation ( Fig. 1 ). In lieu of "lipid species," the Mtb LipidDB uses "lipid groups," in which each Mtb lipid subclass or level 4 class using ChemBioDraw Ultra 11.0 software (Cambridgesoft; Cambridge, MA). Lipids considered to be biosynthetic intermediates, such as phosphatidic acid, hydroxymycolates, and sugar-linked decaprenyl phosphates, were not included in the Mtb LipidDB. Simple free FAs, including tuberculostearic acid, as well as lipid subclasses lacking suffi cient detailed structural defi nition (i.e., defi nition of the FA types), such as the triacylated trehaloses ( 29 ), were also not included in the database. Although absent in the laboratory strain Mtb H37Rv used in this study, the glycosylated phthiodiolone dimycocerosates (DIM Bs) and glycosylated phthiocerol dimycocerosates (DIM As) (phenolic glycolipid) were included in the database because of their biological importance in clinical strains.
In addition to the parent Mtb LipidDB, two searchable database fi les named "+MH_MtbLipid.csv" and " Ϫ MH_MtbLipid.csv" were developed to interface with the Agilent MassHunter software and allow for searching of MS data against the Mtb LipidDB. The +MH_ MtbLipid.csv and Ϫ MH_MtbLipid.csv fi les contained entries for each molecular ion that corresponded to a single lipid group in positive-and negative-ion modes, respectively. The MassHunter software provides the option to identify and collapse multiple molecular ions and adducts into a single feature. However, when only a single molecular ion was detected, the software default was to subtract or add the mass of a proton (1.007276 Da) to the experimentally derived molecular ion before interrogation of the +MH_Mtb_Lipid.csv and Ϫ MH_MtbLipid.csv database fi les, respectively. Thus, to allow for the matching of lipid groups that ionized as a single adducted molecular ion other than H + or H Ϫ , (e.g., [M+Na] + and [M+Ac] Ϫ ions), each +MH_Mtb_Lipid.csv and Ϫ MH_MtbLipid.csv database fi le entry contained a mass value equal to the appropriate adduct but with 1.007276 Da subtracted from or added to the calculated molecular ion, respectively. The selection of molecular ions included in each searchable fi le was an iterative process based on empirically observed ionization properties of each lipid subclass. For instance, in positive-ion mode [

Data processing and analyses
LC/MS data fi les were processed with the MassHunter Qualitative Analysis Software, version B.02.00 (Agilent Technologies;

Important observations to consider in MF assignment
As expected, it was possible to observe multiple MFs with nearly identical retention times that matched to different each lipid group possessed a unique chemical formula corresponding to a unique exact mass. Unlike lipid species, an individual lipid group was not distinguished by stereochemistry, unsaturated bond position, or the length and position of individual fatty acyl substituents. However, each lipid group allowed assignment of the head group composition and the sum composition (total fatty acyl carbon number and total degree of fatty acyl unsaturation), and in many cases, represents multiple unique lipid species. The lipid group classifi cation level was designed for the express purpose of simplifying exact mass-based MS identifi cation, in which a successful database query returns a single lipid group that could represent multiple isobaric lipid species. An additional justifi cation for this classifi cation level was the general paucity of suffi cient detailed structural information at the lipid species level for many of the mycobacterial lipids reported in the current body of literature. In total, the Mtb LipidDB contains 2,512 lipid groups, and when all potential ion adducts were incorporated, the searchable database fi les contain 14,489 mass entries.

LC/MS and automated ion identifi cation allow lipidomic analysis of Mtb
A crude lipid extract from Mtb strain H37Rv grown in standard liquid medium was subjected to C18 reversedphase chromatography, and the effl uent was directly analyzed by accurate-mass (ESI/APCI)-MS. Data collected in the positive-and negative-ion modes were processed using the MFE algorithm of the MassHunter software package, resulting in 1,916 and 744 MFs for the positive-and negative-ion spectra, respectively. The MF lists were searched against the +MH_MtbLipid.csv and Ϫ MH_MtbLipid.csv database fi les with a mass error tolerance of ±5 ppm, yielding How ever, the MassHunter software provided a scoring algorithm for each database match based on a combination of ppm accuracy and isotopic abundance pattern fi t. This scoring mechanism generally appeared to select the correct lipid group, inasmuch as the top candidate displayed a retention time in agreement with the respective parent lipid subclass.

Quantitative and detection limit capabilities
In addition to providing mass and retention time information, the MassHunter MFE algorithm provided a measurement of abundance, termed "volume" for each MF, where volume = [retention time window] × [isotope ion peak heights]. To assess the linearity of this quantitative feature for identifi ed lipid groups, a dilution series of Mtb cells was extracted and the total lipids were analyzed by LC/MS as described above. The MF volumes for representative lipid groups yielded a linear relationship to the quantity of lipid over a dynamic range of approximately three logs. Further, the most abundant lipid subclasses (DIM A, DIM B, PI, TG) could be detected in extracts that were equivalent to <10 6 Mtb cells ( Fig. 4 ).

Assessment of Mtb lipid profi les from different growth phases
Triplicate liquid cultures of Mtb strain H37Rv were sampled at three time points corresponding to logarithmic molecular ion entries from the same lipid group, owing to the formation of multiple ion adducts. + ions for PEs. However, we also observed instances of multiple MFs with different chromatographic behaviors that matched to the same database entry (i.e., the same molecular ion). Specifi cally, there were 109 (+ ion mode) and 24 ( Ϫ ion mode) MFs assigned to the same database entry as that of at least one other MF. This probably resulted from the detection of multiple isobaric lipid species that shared the same sum chemical composition but disparate elution times caused by slightly different fatty acyl repertoires. Thus, subtracting instances in which multiple MFs matched to the same lipid group, a total of 415 unique lipid groups out of the 2,512 present in the database were identifi ed in the Mtb lipid extract. Positive-ion and negative-ion data identifi ed 314 and 193 unique lipid groups, respectively, with an overlap of 92 lipid groups.
The exact mass search window (in this case ±5 ppm) could also impact MF assignment. Specifi cally, in the Mtb LipidDB, there were 29 instances in which a lipid group mass was within ±5 ppm of another lipid group mass. For example, the mass of the sulfolipid III (SL-III) (C90) lipid group is 0.8 ppm less than the mass of CL (81:1). The issue of mass window overlap was further complicated by the inclusion of multiple molecular ions for each lipid group. For instance, the database mass entry for the diacylated diacylglycerophosphoinositoldimannoside (Ac 2 PIM 2 ) (70:0) [M+Na Ϫ 2H] Ϫ molecular ion is 1.4 ppm less than the database mass entry for the Ac 2 PIM 2 (72:

3) [M Ϫ H]
Ϫ molecular ion. In total, ions were observed that were consistent, with C16:1 and C18:1 being the primary FAs of the major logarithmic TG (52:3) group, and C16:0, C18:0, and C26:0 FAs esterifi ed to the major stationary phase-specifi c TG (60:0) group. Growth phase-dependent changes in glycerophospholipid (GP) sum compositions were also observed (see supplementary Fig. IVA-C). For instance, PE (34:1) was the most-abundant PE lipid group in the logarithmic phase, whereas the PE (34:0) lipid group dominated in the stationary phase. Interestingly, the sum composition of the PI and PE lipid groups differed signifi cantly from each other, even in the same growth phase, indicating that FA distributions were specifi c to each GP subclass. Regardless of GP subclass, however, each of the PE, PI, and CL subclasses showed a signifi cant increase in the saturated bond content of their fatty acyl moieties as the cultures aged ( Fig. 6 ).
Finally, notable changes were observed within the wax diester subclass. Comparisons of the relative abundances of DIM A lipid groups revealed signifi cant increases in higher-mass DIM A forms as the cultures entered stationary phase ( Fig. 7 ). The average DIM A mass was 1,367.2 (±0.6 SD) atomic mass units (amu) in logarithmic phase, 1,377.4 (±1.5 SD) amu in transitionary phase, and 1,385.9 (±1.8 SD) amu in stationary phase. Comparable levels of mass increases were also observed for DIM B lipid groups (data not shown). In addition to the observed mass increases, the ratio of total DIM A to total DIM B lipid group ion volumes increased as the cultures aged. The total DIM A/B content was 1.31 (±0.08 SD) in logarithmic phase, 1.85 (±0.16 SD) in transitionary phase, and 3.43 (±0.40 SD) in stationary phase. phase, stationary phase, and an intermediate growth phase between these two termed the "transitionary" phase (see supplementary Fig. II). Crude lipid extracts were prepared from the collected cells and subjected to LC/MS analyses. MF volumes were used to establish relative quantities for each identifi ed lipid group, and comparisons revealed growth phase-dependent differences in lipid class abundance and composition.
Striking differences in both total TG abundance and TG fatty acyl composition were observed across each of the three growth phases. When expressed as a percentage of all identifi ed lipid groups (+ ion data), TG lipid groups accounted for 35.8%, 4.1%, and 77.3% of the ion volumes in logarithmic, transitionary, and stationary phases, respectively, indicating signifi cant changes in cellular concentrations of TG as the cultures aged ( Fig. 5A , inset). Large differences in TG group abundances were also observed from the LC/MS chromatograms (see supplementary Fig.  II). The TG FA composition also varied between growth phases, with a shift from predominantly lower mass TG lipid groups in logarithmic phase to higher mass TG groups in stationary phase ( Fig. 5A ; see supplementary Fig. III). The sum compositions indicated that the smaller TGs were probably esterifi ed with C16, C18, and C19 FAs, whereas longer C22, C24, and C26 acyl functions probably esterifi ed the larger, stationary phase-specifi c TGs. The latter observation fi ts well with previous structural reports of mycobacterial TG ( 13,32,33 ). To further support this structural information, the most-abundant logarithmic and stationary phase TG group [M+NH 4 ] + ions were targeted for collisioninduced dissociation MS/MS ( Fig. 5B ). Major diacyl product combinations of the FAs reported in the literature for a specifi c lipid. This approach resulted in database inclusion of lipid structures containing fatty acyl combinations not previously reported, but that are possible based on the biochemical literature. This expansion of the database was most noted with the mycobacterial lipids commonly referred to as phosphatidylinositolmannosides (PIMs), where 1,037 of the 2,512 lipid structures in the Mtb LipidDB were assigned to a PIM structure. This refl ects the variable composition of the PIM structures with one to six mannose residues and one to four acyl moieties ( 35 ). Of the 1,037 potential PIM structures included in the Mtb LipidDB, 77 were identifi ed in our analyses. This can be directly compared with other MS-based studies of PIMs isolated from M. bovis BCG ( 36,37 ), where 73 of the possible 1,037 PIM groups included in our database were identifi ed. The inclusion of theoretical lipid structures based on sound biochemical evidence, and more-general descriptions in the literature of acyl composition, are not uncommon. In fact in eukaryotes, it is estimated that there are at least 25,000 TG structures, including those with ether-linked FAs ( 38 ).
Beyond the ability to identify individual lipid groups or species, MS offers the potential to perform relative quanti-

DISCUSSION
The global characterization of the Mtb lipid composition presents a signifi cant technical challenge, owing to the structural diversity within this group of macromolecules and the number of complex structures that are unique to Mtb and related species. However, based on the efforts of earlier investigators to fully elucidate the structure of a large majority of these products, it was possible to derive a database of Mtb lipids that possessed 2,512 structures and that could be interrogated with MS data based on the 14,489 mass entries that accounted for multiple ion adducts of both positive-and negative-ion data. The interrogation of this database with MS data of whole-lipid extracts from Mtb allowed for the automated detection and identifi cation of 415 lipid structures encompassing all 15 of the known classes of Mtb lipids. This work represents the largest single survey of Mtb lipid structures from a single study and provides the groundwork for lipidomics studies that are on par with those now being performed with eukaryotic organisms ( 34 ).
The Mtb LipidDB was developed to allow lipidomic investigations in an unbiased manner. Thus, lipid group structures and masses were obtained by calculating all potential study, the stationary phase was the point at which the total bacterial growth rate ceased as measured by optical density. Such tightly capped batch cultures grown to high cellular densities have been suggested to be depleted of oxygen ( 40 ); thus the observed accumulation of TG in the stationary phase would agree with previous reports demonstrating TG production increases under hypoxic and/or stressful tative analyses based on the total ion abundance of individual MFs ( 39 ). This approach was used to monitor lipid changes over different phases of Mtb growth in batch culture. TG abundance waned and waxed dramatically as the cultures aged, and TG FA sum compositions also shifted markedly, with the appearance of higher-molecular-weight TG groups in stationary phase. For the purposes of this + molecular ions for representative TG groups. The major logarithmic-phase TG (52:3) ion produced two major diacyl product ions at m/z 575.5010 and m/z 603.5308 that resulted from neutral losses of C18:1 and C16:1 FAs, respectively. The major stationary phase-specifi c TG (60:0) ion produced three major diacyl product ions at m/z 579.5345, m/z 691.6605, and m/z 719.6682, which indicated neutral losses of C26:0, C18:0, and C16:0 FAs, respectively. ( 15,20 ). These mass increases were shown to result from addition of methylene units to the esterifi ed mycocerosic acids. These data build on evidence that suggest that Mtb actively modulates its lipid composition in response to the changing microenvironment. Thus, this lipidomic approach provides a working platform for the analyses of more-refi ned in vitro growth culture models that incorporate defi ned stress factors ( 19,(41)(42)(43)(44)(45)(46)(47)(48), as well as comparative analyses of lipid profi les from Mtb cells isolated from infected tissues. Such in vivo studies would provide valuable insight into the metabolism and physiology of Mtb . Additionally, combining the two-dimensional 1 H- 13 C heteronuclear single quantum coherence NMR approach applied by Mahrous,Lee,and Lee ( 22 ) would potentially provide synergistic methods to produce complementary levels of information. A current limitation to the quantitative lipid profi ling is that the true stoichiometric relationships between lipid classes cannot be determined. Even differences between lipid groups belonging to the same lipid subclass could not be used to determine true abundance, considering that acyl chain length and degree of unsaturation are known to infl uence ionization responses ( 49 ). Absolute quantifi cation could be achieved with the incorporation of multiple internal lipid standards to which individual lipid ionization could be normalized. Such approaches are becoming more commonplace for eukaryotic-based lipidomics studies ( 34 ). However, appropriate isotopically labeled standards specifi c for many of the unique Mtb lipids are lacking, and require development to achieve this level of analysis.
A central element to our MS-based lipidomics strategy was automated product identifi cation based on accurate mass matching. This approach has become more commonplace for metabolomics-based experiments that utilize high-resolution mass spectrometers ( 50,51 ). For global lipid characterization, however, this methodology can be complicated by differential acylation of the same lipid, resulting in isobaric lipid species. Thus, the Mtb LipidDB was designed such that the output returned only FA sum composition, thus limiting potential mass overlap between conditions ( 12-14, 16, 33 ). Our structural analyses suggested that C26:0 FAs were major components of these TG groups, also in agreement with previous studies on hypoxiainduced TG molecules ( 13,14,33 ). Alterations in FA composition were not limited to the TGs, inasmuch as the PE, PI, and CL lipid classes also demonstrated similar changes that were class specifi c. However, common to each class was an observed increase in overall saturation levels as the cultures aged. It should be noted that the presence of polysorbate (Tween 80) in the media may have infl uenced fatty acyl compositions, considering that this detergent supplies oleic acid esters for growth. The average size of DIM A and DIM B lipids was also shown to increase as the cultures aged. DIM size increases were previously observed in Mtb cells cultivated on odd-chain carbon sources and cholesterol, and also in Mtb cells from infected mice lungs  database entries. In this study, we demonstrated a relatively small number of instances in which mass overlap (±5 ppm) occurs between lipid groups in the Mtb LipidDB. Indeed, it may be advantageous that Mtb does not produce appreciable levels of phosphatidic acid, phosphatidylserine, or phosphatidylcholine, because these would overlap in mass with other Mtb phospholipids, as observed in eukaryotes ( 52 ). Furthermore, we note that the unique large, apolar molecular characteristics of many Mtb lipids provide distinguishing mass features for these molecules in a eukaryotic lipid background (unpublished observations). We also view the automated identifi cation process described in this study as a screening method, and believe that database identifi cations based on accurate mass alone should be regarded as tentative. Lipid identifi cations of biological interest should be subjected to MS/MS characterization and structural confi rmation, as was done with select TG groups in this study. Interpretation of the resulting data can be aided by well-documented Mtb lipid collision-induced dissociation fragmentation patterns. Indeed we have obtained confi rmatory MS/MS product ion spectra for representatives of many of the lipid subclasses identifi ed in this study (data not shown). Identifi cation confi dence has also been shown to be enhanced with the incorporation of retention time criteria into the database searching process in the presence of internal standards that allow for normalization of retention times (53)(54)(55)(56). Thus, performing MS/MS analyses in a spot-check fashion would allow for further retention time curation of the searchable database fi les, and this could easily be developed through an iterative process within a laboratory. Finally, the accuracy of the database identifi cations will also improve with the increasing resolution of newly developed mass spectrometers.
The limitations of the current Mtb LipidDB also need to be recognized. First, the database is constrained by the reported lists of FAs considered for the various lipid group repertoires. For example, only the FAs observed by Walker,Barakat,and Hung ( 32 ) were used to calculate GP lipid group masses, yet other atypical FAs may be esterified to mycobacterial GPs and would thus escape identification from the database search. The Mtb LipidDB does not take into consideration mass shifts due to oxidation or degradation of lipids, which may be keys to a complete understanding of Mtb lipid metabolism. Only well-defi ned lipids were included in the database; a novel and recently described wax ester lipid identifi ed from Mtb cultivated under iron-limiting conditions was not included because of its partial structural elucidation ( 48 ). Biosynthetic lipid intermediates, such as phosphatidic acid, acyl-CoA, and sugar-linked decaprenol phosphates, were also not included in the database. Finally, the database in general is restricted to the lipid composition of the Mtb H37Rv strain used in this study, and does not contain lipids found in other mycobacterial species. Keeping these limitations in mind, the fl exible data format of the database should allow users to easily amend and add lipid entries according to their specifi c research interests.