Direct analysis of sialylated or sulfated glycosphingolipids and other polar and neutral lipids using TLC-MS interfaces.

Gangliosides and sulfatides (STs) are acidic glycosphingolipids (GSLs) that have one or more sialic acids or sulfate substituents, in addition to neutral sugars, attached to the C-1 hydroxyl group of the ceramide long chain base. TLC is a widely employed and convenient technique for separation and characterization of GSLs. When TLC is directly coupled to MS, it provides both the molecular mass and structural information without further purification. Here, after development of the TLC plates, the structural analyses of acidic GSLs, including gangliosides and STs, were investigated using the liquid extraction surface analysis (LESA™) and CAMAG TLC-MS interfaces coupled to an ESI QSTAR Pulsar i quadrupole orthogonal TOF mass spectrometer. Coupling TLC with ESI-MS allowed the acquisition of high resolution mass spectra of the acidic GSLs with high sensitivity and mass accuracy, without the loss of sialic acid residues that frequently occurs during low-pressure MALDI MS. These systems were then applied to the analysis of total lipid extracts from bovine brain. This allowed profiling of many different lipid classes, not only gangliosides and STs, but also SMs, neutral GSLs, and phospholipids.

ESI is softer than MALDI and is thus a good alternative for the analysis of gangliosides (30)(31)(32). The method was employed by the Peter-Katalinić group for brain ganglioside research in 2001 ( 33 ). By a combination of the MS/ MS information provided by both negative-ion analysis of native gangliosides and positive-ion mode analysis of permethylated derivatives, these investigators achieved identifi cation and characterization of isomeric gangliosides.
To avoid ion suppression by phospholipids, chromatography of lipid mixtures prior to MS has usually been considered helpful. In the past, MS analysis had followed the scraping of bands from TLC plates and extraction of the lipid components from the bands ( 34,35 ); this approach has been called "indirect sampling TLC-MS" ( 36 ). There are several disadvantages to this "off-line" TLC-MS, including the low recovery rate from scraping the silica gel, the necessity for additional processing with a column, and the contamination of the analyte by residual silica gel. "Direct sampling TLC-MS" that interfaces TLC with MS has been developed for in situ use and permits the determination of both molecular mass and structural information on GSLs directly from the bands, immediately following chromatographic separation and without requiring any further purifi cation. The challenge of the TLC-MS coupling, which is different from the other chromatography-MS approaches, arises between the extraction from the plate and introduction into the mass spectrometer, because the sample is embedded within the stationary phase. Therefore, most common methods developed for direct sampling TLC-MS are today based on surface or laser desorption/ionization techniques ( 37,38 ). However, because of the high risk for elimination of sialic acid ( 39 ), unless special devices are employed, such as the vibrationally cooled MALDI-FTMS system mentioned above ( 40 ) or a MALDI-Q-o-TOF MS that can be operated at relatively high ion source pressure ( 29 ), gangliosides cannot be individually analyzed. In addition, the introduction of an energy-absorbing matrix that is necessary in the MALDI method adds more complexity into the analysis system. character of GSLs increases the challenge in the isolation, separation, and detection of this group of lipids has long been well-known by lipid researchers. TLC is used commonly and conveniently to analyze and resolve individual GSLs in complex mixtures ( 23 ). When compared with HPLC, TLC seems to be a rather old technology, but it offers several advantages for lipid analysis: simplicity, easy visualization of analytes, high resolution, high throughput (many samples can be loaded on a TLC plate and analyzed simultaneously), and freedom from carryover. However, because GSLs constitute a category of compounds with high heterogeneity, and TLC is a method, in principle, for separation and comparison (but not de novo identifi cation), it is very necessary to supplement TLC with other methods to achieve more informative analysis results.
Over the last decades, MS has become a rapid, highly accurate, and sensitive method for identifi cation and structural characterization of GSLs. Antibodies cannot distinguish among GSLs with different fatty acyl chain lengths on the sphingoid backbone moiety. MS is not only able to accomplish that, but it can also detect the number of double bonds or other modifi cations on the lipids. Imaging mass spectrometry using a MALDI ion source has recently become a powerful tool to visualize the distribution of GSLs from tissues in situ (24)(25)(26); it requires few sample processing steps, and gives direct well-resolved positional information on GSLs in tissue. However, the imaging MALDI data should be interpreted with caution because sialic acid is quite labile under high vacuum laser desorption/ionization conditions ( 27 ). The Boston University School of Medicine vibrational cooling MALDI-Fourier Transform Mass Spectrometry (FTMS) system was developed in 2002 to address this concern by introducing a short pulse of gas to cause gentle collisions and thereby diffuse the excess energy imparted during irradiation, thereby stabilizing the labile glycosidic bond of sialic acid and preventing its cleavage during laser irradiation ( 28 ). Our further experiments showed that a MALDI-Q-o-TOF MS source could be operated at a pressure suffi cient to provide vibrational cooling ( 29 ). , respectively. The fragmentation pattern for the d18:1/C18:0 GD1a homolog is shown in (B). pipette tip, forming a liquid-surface junction that facilitated liquid extraction of the analytes. The liquid junction was held in place for 1 s and 1.5 l of the solution was aspirated back into the tip. Collision-induced dissociation (CID) was performed with nitrogen gas to achieve the fragmentation information of the compounds of interest. The collision energy was set at a value between Ϫ 50 and Ϫ 90 eV, depending on the lipid structures. Usually, singly charged gangliosides and STs were analyzed using an energy (approximately Ϫ 90 eV) higher than that needed to dissociate multiply charged gangliosides or singly charged phospholipids (approximately Ϫ 50 eV). The GD1a ganglioside standard was employed to calibrate the TOF mass scale in the MS/MS mode. After calibration, this mass spectrometer was capable of achieving better than 15 ppm mass accuracy in both MS and MS/MS modes.
The CAMAG TLC-MS interface was coupled to the turbo-ESI ion source of the QSTAR Pulsar i mass spectrometer in the negative-ion mode. Solvent system D [isopropyl alcohol/methanol/ water (5:3:2, v/v/v) with 0.1% of ammonia] was continuously delivered by the System Gold 125 HPLC pump (Beckman Coulter Inc., Fullerton, CA) at a constant fl ow rate of 0.07 ml/min. The ion source temperature was 400°C and the capillary voltage was maintained at Ϫ 3.5 kV. The CID conditions were the same as those used in the LESA TLC-MS experiments described above.

RESULTS AND DISCUSSION
Several variables infl uence the behavior of GSLs on HPTLC plates using TLC-MS interfaces, including the type of TLC plate, the solvent systems, and the extraction time for recovery of the GSLs from the TLC plates. The instrumental parameters must be considered individually for each TLC-MS interface. First, the proper type of TLC plates should be chosen for the separation of acidic GSLs. Silica gel HPTLC plates on both glass and aluminum can be used for GSL analysis with the CAMAG TLC-MS interface, because the CAMAG TLC-MS interface provides a direct extraction of analytes from the silica gel. (Though analytes have been embedded within the silica gel, the extracting solvent is allowed to fl ush through the band to elute the analytes.) On the other hand, we found that the LESA TLC-MS interface can only utilize silica gel TLC plates with glass backing. (Aluminum foil-based TLC plates were not able to hold the extracting solvent in place, and this caused solvent spread.) Second, the time for direct extraction should be considered. Because longer extraction times were found to result in carry-over in the CAMAG TLC-MS interface, the time appropriate for extraction from the TLC plates needed to be determined for the various sample classes. In contrast, the LESA TLC-MS interface did not exhibit any carry-over, because a new tip and nozzle were used for each extraction. Third, the extraction solvent system can be a variable. Because this study focused on the analysis of acidic GSLs, organic solvents were required to extract the lipid analytes. However, silica gel HPTLC plates have absorbent surfaces, and thus an eluting liquid containing only organic solvents cannot be utilized for TLC plates because of the low surface tension of organic solvents. Therefore, different solvent systems were evaluated to achieve optimum extraction of GSLs. A Therefore, we have now developed methods for coupling TLC and MS with an ESI interface to study acidic GSLs such as gangliosides and STs. Two approaches were developed based on two different interfaces, both commercially available, the TLC-MS interface from CAMAG and the liquid extraction surface analysis (LESA™) TLC-MS interface from Advion BioSciences. These two interfaces provide universal hands-free connection between the TLC and MS, without requiring any other modifi cation. With its depth profi le, the CAMAG TLC-MS interface extracts the complete volume of the band containing the substance of interest, while the LESA TLC-MS interface recovers only components that lie near the band surface. Both systems have been applied for analysis of metabolites, peptides, dried blood spots, thin tissue sections, etc. (41)(42)(43)(44)(45). To our knowledge, this is the fi rst report of their application to GSLs, and the fi rst direct comparison of these two interfaces in terms of sensitivity, analysis time, and potential features.

Materials
Silica gel 60 high-performance (HP)TLC plates with aluminum or glass backing were purchased from EMD Chemicals Inc. (Gibbstown, NJ). The acidic GSL standards, GM1 and ST (d18:1/ C24:1), and the bovine brain total lipid extract were obtained from Avanti Polar Lipids, Inc. (Alabaster, AL). The ganglioside standard, GD1a, and the detection agents, orcinol and primuline, were from Sigma Chemical Company (St. Louis, MO). All the HPLC grade solvents were purchased from Merck (Darmstadt, Germany).

HPTLC
Acidic GSLs and the bovine brain total lipid extracts were dissolved in chloroform/methanol (1:1, v/v) and applied by means of a microliter syringe (Hamilton Co., Reno, NV) as 3 mm spots on silica gel-coated plates. Plates for the analysis of acidic GSLs were developed with solvent system A [chloroform/methanol/0.2% calcium chloride in water (55:45:10, v/v/v)]. The bovine brain total lipid extract was developed by solvent system B [chloroform/methanol/water/acetic acid (90:50:5:2, v/v/v/v)] ( 46 ) for subsequent lipid detection. Each GSL was deposited twice on one silica-TLC plate and the plate was cut into two pieces after development. One piece was stained as a reference for the identifi cation of positions of the lipids of interest, and the other was used for TLC-MS analysis. Two staining reagents, orcinol and primuline, were applied for the detection of lipids.

Direct sampling TLC-MS
The TriVersa Nanomate LESA™ device (Advion BioSciences, Inc. Ithaca, NY) was coupled to the nanoESI source of a QSTAR Pulsar i quadrupole-orthogonal TOF mass spectrometer (AB Sciex, Foster City, CA) and the mass spectrometer was operated in the negative-ion mode. Typical LESA-MS experimental conditions were: spray voltage of Ϫ 1.5 kV, nitrogen delivery gas at a pressure of 0.8 psi, and solvent system C [isopropyl alcohol/ methanol/water (9:1:1, v/v/v)] as extracting solvent. A total volume of 8 l of extracting solvent C was used for each extraction with 6 l being dispensed 0.8 mm above the surface by the analytes, the signal could be detected for у 10 pmol. This detection limit takes into account the extraction effi ciency for samples located near the top surface of the TLC plate. Detection limits, mass accuracy, and resolution do, of course, vary with the type of mass spectrometer. Because of demands on instrument time, the full set of comparisons reported here was carried out with the Applied Biosystems QStar Q-o-TOF MS instrument; however, we also obtained data for these classes of compounds with the Thermo-Fisher LTQ-Orbitrap MS and the Bruker SolariX 12-T FTMS and observed results consistent with those reported herein.
GD1a The ganglioside GD1a was deposited on a silica TLC plate having an aluminum backing, and the plate was subjected to solvent development. Because the sampling spot for the LESA TLC-MS interface is rather small, analysis of multiple spots on one TLC band is possible and often necessary. The signal intensity was measured to investigate whether different spots on the same band yielded different signal intensities ( Fig. 3 , left side). Spots 1 and 2 showed similar signal intensities. Although GD1a was detectable in spots 3 and 4, they showed signals that were an order of magnitude lower than those generated from spots 1 and 2. Therefore, it became clear that signal strength is variable across the TLC bands and it is very important to perform sampling from the "right" location by LESA TLC-MS interface to obtain the best signal. This result also indicates that quantifi cation with this system will require the use of internal standards.
To determine the effi ciency of extraction in LESA TLC-MS, GD1a was applied, the plate subjected to solvent development, and spots were analyzed in the negative-ion mode ( Fig. 3 , right side). One spot from the GD1a band was chosen for analysis and was analyzed repeatedly (four times) using LESA TLC-MS. There was no change in the signal intensities from the fi rst to the second extraction, but an order of magnitude drop in signal intensities was observed for the third and fourth extraction. This result demonstrates that the analyte on the TLC plate is 9/1/1 mixture of isopropyl alcohol, methanol, and water produced the best extraction effi ciency (data not shown). Fourth, instrumental parameters had to be optimized for the GSL analysis. The CAMAG TLC-MS interface was coupled to an HPLC pump and the turbo-ESI equipped QSTAR Q-o-TOF MS, whereas the LESA TLC-MS interface was mounted on an Advion NanoMate that was connected to nano-ESI QSTAR Q-o-TOF MS (see Experimental section).
In the negative-ion mode, the detection limit for direct LESA TLC-MS was determined by analysis of multiple dilution series of GD1a and ST (d18:1/C24:1) ( Fig. 2 ). Different amounts of analytes (100 pmol, 50 pmol, 10 pmol, and 1 pmol) were applied onto the silica-based TLC plates and the plates were developed in a closed chamber. For both corresponded to the head group, e.g., m/z 96.7 (sulfate) and m/z 241.0 (sulfated hexose), but also refl ected the composition of the ceramide chains, e.g., m/z 390.4 (indicating the C24:1 fatty acyl chain) and m/z 522.3 (corresponds to the head group and the deacylated sphingoid backbone).
The CAMAG TLC-MS interface was also evaluated for use in the direct ESI-MS analysis of acidic GSLs on TLC plates. This interface is a semi-automatic instrument involving an automatic piston movement to create a pressure seal over the appropriate TLC/HPTLC zone, on either glass plates or aluminum foils, followed by deposition and recovery of a suitable solvent delivered by the HPLC pump and then taken back up for injection into the mass spectrometer. The CAMAG TLC-MS interface is thus a micro-fl ow system that can be coupled to a mass spectrometer that has a turbo-spray ion source. This device scrapes/extracts the whole band of silica gel particles from selected spots on the TLC plate. MS and MS/MS spectra of 10 pmol of GD1a and 20 pmol of ST (d18:1/C24:1) acquired by the CAMAG TLC-MS system after analyte separation on the silica TLC plate and represented the detection limits ( Fig. 4C, D and Fig. 5C, D , respectively). Even when slightly different extraction solvent systems were used (see Experimental section for the details), the tested standards showed similar spectra for MS and MS/ MS. Spectra obtained by the CAMAG TLC-MS system exhibited higher signal intensities and lower background noise, probably due to the fact that the extraction area used by the CAMAG TLC interface is about 10 times larger reextractable and reanalyzable with the LESA TLC interface, but a given spot is eventually depleted. The CID spectrum of GD1a showed sugar fragments containing the sialic acid moiety: m/z 290.1 (B 1 ␣ , B 1 ␤ ), 470.1 (B 2 ␣ ), and 673.2 (B 3 ␣ ). The CID spectrum of GD1a also included Y-series ions that consist of increasing numbers of sugar residues attached to the ceramide moiety at m/z 592.6 (Y 0 ), 754.6 (Y 1 ), 1,410.8 (Y 3 ␣ ), and 1,572.9 (Y 4 ␣ and Y 2 ␤ ). Although no peak corresponding to Y 2 ␣ was observed at its calculated value ( m/z 1,207.8), its secondary fragment was seen at m/z 916.7 (Y 2 ␣ /Y 2 ␤ ). Elimination of the sialic acid residue from the Y 3 ␣ fragment resulted in the peak observed at m/z 1,119.7 (Y 3 ␣ /Y 2 ␤ ). The peak, due to the loss of both sialic acid residues from the molecular ion (Y 4 ␣ /Y 2 ␤ ), appears at m/z 1,281.8.
The ST (d18:1/C24:1), another type of acidic GSL, was also investigated by TLC/ESI-MS/MS using both systems. The region, including m/z 560-960 in these mass spectra, is shown in Fig. 5A, C ; for each system, the CID spectrum of the precursor ion [M -H] Ϫ m/z 888.6 was evaluated and the results are shown in Fig. 5B, D . The product ions To extend our work to other classes of lipids, the commercial bovine brain total lipid extract was also separated using the development solvent system B. The plate was stained with primuline ( Fig. 6C ) and analyzed by both TLC-MS systems. The spatial resolution for sampling with LESA TLC-MS was higher, and thus 62 spots were taken to perform experiments from the seven primuline-positive areas, as compared with 20 spots that could be extracted by the CAMAG TLC-MS system. We recorded MS spectra in the range m/z 600-2,000, and performed MS/MS on the precursor ions with intensities that were higher than two counts. Compounds identifi ed by the LESA and CAMAG TLC-MS systems are listed in supplementary Table I. Seven classes of compounds were identifi ed by the LESA TLC-MS and eight classes were found with the CAMAG TLC-MS. than that accessed by the LESA TLC interface. However, although the analysis using the CAMAG TLC interface allowed spray duration of 1 min, no reextraction was possible. As a result, multiple loading of the analytes was required to perform different experiments. In other words, in comparison to the CAMAG-based system, the LESA TLC-MS delivers a smaller portion of the sample from the TLC plate and sprays longer (one analysis allows both MS and MS/MS experiments). Acidic GSLs on TLC plates can be reanalyzed and reextracted by the LESA TLC-MS system. There is no carry-over, but it is critical to pick the right position for analysis with either TLC-MS interface. The LESA interface was expected to provide a better signal because the tip is changed for every spray; however, the high ratio of organic solvent used in the lipid extraction was found to dissolve material inside the tips, thus increasing the background signal.
We analyzed a commercially available bovine brain total lipid extract with the LESA TLC-MS system. The total lipid extract was separated using the development solvent system A and compared with the TLC plate stained with orcinol ( Fig. 6A ). Gangliosides such as GD1a ( m/z 917.5 and 931.5), GD1b ( m/z 917.5 and 931.5), and GM1 ( m/z 1,544.9 and 1,572.9) were detected on the resolved TLC plate. In Although we have discovered that the fast oxidative reaction on double bonds is a common phenomenon on lipids that have been dried naturally on surfaces and exposed to light under ambient air (Y. Zhou, H. Park, and C. E. Costello, unpublished observations), we previously viewed the oxidized ST results for standards deposited on TLC plates. However, in the biological samples such as the bovine brain lipid extract, the oxidation reaction was not as fast as the reaction of the pure standards after TLC separation. There was no oxidative ST product for 3 days and small amounts of oxidative products appeared slowly only after 3 days (data not shown). In contrast to STs, phospholipids such as PS and PE-NMe 2 were easily detected in the bovine brain lipid by TLC-ESI-MS coupled to TLC-MS interfaces. We proposed that this might occur because of the presence of antioxidant components, such as lipid-soluble vitamin E, that exist in the brain tissue. The US Department of Agriculture National Nutrient Database for Standard Reference, Release 26 indicates that raw beef brain contains 0.99 mg of vitamin E in every 100 g ( 47 ) . Vitamin E is the most effi cient lipid-soluble antioxidant vitamin ( 48 ). It is also very hydrophobic, so that, during TLC development, it moves very close to the solvent front, in the region where the STs travel, and may protect the STs from the oxidation. However, after the TLC separation, the phospholipids such as PS and PE-NMe 2 were exposed to the ambient air without similar With both TLC-MS systems, not only acidic GSLs, but also phospholipids such as phosphatidylethonalamine (PE), phosphatidylinositol (PI), phosphatidylserine (PS), N , N -dimethyl phosphatidylethanolamine (PE-NMe 2 ), and SM were detected in the bovine brain lipids. It is signifi cant to note that we found more classes of compounds with the CAMAG TLC-MS system, but identifi ed more specifi c compounds within individual lipid classes with the LESA TLC-MS system. Hexosylceramides and some gangliosides were identifi ed only with the CAMAG device. As discussed above, the extracting solvent system that was used for CAMAG TLC-MS was different from that used with the LESA system, in addition, the CAMAG TLC-MS allowed solvent to thoroughly penetrate the silica gel. These factors helped the CAMAG system to fl ush out components over a wider range of polarities, even when the solvent system was not optimum for some of these. Scanning along smeared bands revealed compounds with different chain lengths. Shorter chain lipids from a given class, being less hydrophobic, were found at lower positions on the plate, and vice versa. All data in this report on acidic glycolipids were acquired in the negative-ion mode. Analysis of complex mixtures of neutral glycolipids will require optimizing the parameters for this class, including changing the polarity of the MS instrument, and modifying the extraction protocols, e.g., solvent systems (compositions, time, etc.) and TLC separation conditions. have developed for direct analysis of polar lipids through coupling of an HPLC with a Q-q-TOF MS through two different types of TLC-MS interface allows high resolution, high sensitivity, and accurate mass assignments for acidic GSLs in MS and MS/MS. Whereas the CAMAG TLC-MS interface was found to yield higher signal intensities and the spectra had higher signal-to-noise ratios, the LESA TLC-MS interface provided a longer analysis time that allowed for the summing of multiple MS and MS/MS analyses and this feature can be exploited to gain high sensitivity. Although TLC-MS has previously been performed with fast atom bombardment, liquid secondary ion, MALDI MS, and desorption electrospray ionization MS, we now recommend exploration of the coupling with the recently available TLC ESI-MS interfaces . These devices require only simple sample preparations and minimize loss of analytes, and the more fragile molecules, such as gangliosides and STs, have higher stability under the usual experimental conditions because ESI is a more gentle ionization method. The usefulness of the methods for complex biological mixtures has been demonstrated here with the analysis of a bovine brain total lipid extract. The methods seem very likely to fi nd broad application for "on-line" TLC-ESI MS of lipids.
protection by an antioxidant, and they were found to be oxidized. More evidence that this "position-related" protection theory may be correct was the PE class. PEs were the only class of phospholipids that traveled with the STs near the solvent front and, like the STs, no oxidative product was detected from the PEs. Thus, although the structures of PEs are more similar to other phospholipids than to the STs, these lipids remain protected from oxidation, like the STs. More details regarding lipid oxidation are discussed elsewhere (Y. Zhou, H. Park and, C. E. Costello, unpublished observations).

CONCLUSIONS
TLC-ESI MS coupling using either commercial or homebuilt TLC-MS interfaces to investigate gangliosides and STs has not been reported previously. The method we Ϫ peaks of the two homologs were isolated in Q1 and subjected to CID MS/MS. *The signature fragments that differentiate the glycan structures of GD1a and GD1b.