LC-MS-based method for the qualitative and quantitative analysis of complex lipid mixturess⃞s⃞ The online version of this article (available at http://www.jlr.org) contains an additional figure. Published, JLR Papers in Press, January 28, 2006.

A simple and robust LC-MS-based methodology for the investigation of lipid mixtures is described, and its application to the analysis of human lipoprotein-associated lipids is demonstrated. After an optional initial fractionation on Silica 60, normal-phase HPLC-MS on a YMC PVA-Sil column is used first for class separation, followed by reversed-phase LC-MS or LC-tandem mass spectrometry using an Atlantis dC18 capillary column, and/or nanospray MS, to fully characterize the individual lipids. The methodology is applied here for the analysis of human apolipoprotein B-associated lipids. This approach allows for the determination of even low percentages of lipids of each molecular species and showed clear differences between lipids associated with apolipoprotein B-100-LDL isolated from a normal individual and those associated with a truncated version, apolipoprotein B-67-containing lipoproteins, isolated from a homozygote patient with familial hypobetalipoproteinemia. The methods described should be easily adaptable to most modern MS instrumentation.


Introduction
A large variety of methods have been published for the separation of lipids, either by thin layer chromatography (TLC) or by liquid chromatography (LC); the methods have usually been described for analysis of specific classes of compounds (www.cyberlipid.org). Mass spectrometric (MS) methods for the characterization of lipid mixtures have also been published in recent years, mostly centered on the use of MALDI-TOF (matrix-assisted laser desorption/ionization -time-of-flight) MS and ESI (electrospray ionization) MS (1).
Sophisticated methods like the characterization of complex glycolipids directly from TLCplates by vibrationally cooled MALDI FT-ICR (Fourier transform -ion cyclotron resonance) MS (2) require instrumentation that is not yet widely available. A variety of elegant nanospray MS methods have been described (3)(4)(5) that are generally a good choice for the characterization of lipids, but may not be fully capable of both qualitative and quantitative analysis of highly complex mixtures. LC-MS offers possibilities for a better determination of minor compounds whose signals might otherwise be suppressed. It also allows for an additional level of characterization of components based on their chromatographic behavior, as well as the MS results. Existing HPLC methods for the separation of lipids are limited, however, in that they either target only selected classes, or are not compatible with subsequent MS. Pulfer and Murphy suggested that, for a complete separation of lipids, normal-and reversed-phase chromatography should be combined (6).
Because of its high sensitivity and the additional information it provides, MS is widely recognized as a superior detection method, when compared to the classic methods of UV or light scattering. We demonstrate here a simple, robust and reproducible methodology for lipid analysis, which has been achieved by adapting to LC-MS several separation systems

Reversed-phase LC-MS:
Fractions obtained from the normal-phase column could be further characterized by reversed-phase LC-MS/MS using a (300 µm x 15 cm) Atlantis dC18 capillary column (Waters Corp.) with either the Quattro II-based system described above, or with a Waters CapLC system interfaced to an Applied Biosystems/Sciex QStar split to 5 µL/min, or 5 µL/min without a split; typical injections of standards were about 1 ng per compound class. For the separation of both the glycerophosphoethanolamines andcholines, the following solvents and gradient were used: Solvents were (G) 10 mM ammonium acetate in MeOH/IPA/water (90:5:5), and (H) 10 mM ammonium acetate in MeOH/IPA/water (94:5:1). The LC method consisted of holding solvent G for 10 min, a gradient over 15 min to solvent H, and holding H for 5 min. On the triple quadrupole mass spectrometer, spray voltage was typically set to -4 kV and collision voltage to -80 V in negative ion mode, and to 4 kV and 60-80 V in positive ion mode, respectively. LC-MS precursor ion scans for m/z 184 in positive ion mode were performed at a cone voltage of 80 V and a collision energy of -35 eV, and neutral loss scans (-141 u) in positive ion mode at 60 V (cone) and -20 eV (collision energy).
Nanospray MS: was performed on either type of mass spectrometer. Nanospray tips were prepared from borosilicate glass capillaries (World Precision Instruments, Inc., Sarasota, FL) using a Model P-97 tip puller (Sutter Instrument Co., Novato, CA). Samples were screened on the triple quadrupole instrument using MS and MS/MS in the product ion, precursor ion, or neutral loss scan mode, essentially as described by Brügger et al. (3). In addition, precursor ion scans were performed for acyl fragments in the negative ion mode, and for sugar fragments and cholesterol in the positive ion mode, as well as an additional neutral loss scan (-260 u) for PI in the positive ion mode. For selected samples, data was also obtained with the QStar QoTOF MS using MS and MS/MS in the product and precursor ion modes.
Quantitation: Quantitation was based on the separations on the PVA-Sil column. For the experiments with biological samples, data obtained for external standards run on the same day were used for estimation of the sample quantities. These standards contained a defined amount, typically 50 ng per compound class for each injection. To compensate for significant differences in sensitivity, amounts were adjusted for some classes, e.g., usually 50 % less standard was used for LPCs, and 2-4 times the amount was used for cholesterol and gangliosides. The reported values were based on signal height or signal area of the SIC 6 at the appropriate retention time. The results were not corrected for variation in ionization by on December 20, 2009 www.jlr.org Downloaded from efficiencies within a given compound class. Compounds evaluated for use as internal standards included dimyristoyl-sn-glycero-3-phosphocholine-d 54 and a rare diacylglycerophosphoethanolamine (C16:0, 17:1).

Results
Our goal in this study was to develop a robust methodology that can be used with simple as well as sophisticated instrumentation, and is especially suitable for the analysis of complex mixtures that are difficult to analyze by conventional nanospray MS. Additionally, for biological samples containing some known or potential non-lipid contamination, or in the case of very small amounts of polar lipids in the presence of large amounts of nonpolar lipids (or vice versa), we employ an initial separation on conventional Silica 60. For the primary separation step, we chose the route of separation by compound classes by normalphase LC-MS (which already delivers the information necessary for a more reliable quantitation of minor components, as well as a level of information about the molecular composition that might be sufficient to fulfill the analytical requirements in many cases).
Fractions collected from this normal-phase column can then be further characterized by reversed-phase LC-MS, and/or by nanospray MS.

Separation of polar and nonpolar lipids on Silica 60:
Lipid standards containing diverse nonpolar, phospho-and glycolipids have been reproducibly separated on the basis of polarity by elution from Silica 60 resin with MTBE and MeOH. Compounds such as CholE, TAG and glycerophospholipids were found to be well separated, but compounds of medium polarity were detected in different amounts in both fractions (MAGs, short cerebrosides). Use of the 6-mL SPE columns generated inconveniently large volumes of solvent that had to be evaporated, but self-plugged Pasteur pipettes did not always deliver reproducible results and therefore these were only used for samples containing small amounts (< 1 µg). Substitution of chloroform for MTBE, or addition of 10 mM ammonium acetate to the MeOH elution did not significantly alter the profile. Even the most polar compound tested, the trisialoganglioside GT1b, could be later, MTBE and ammonium acetate were selected for general use.

Gradient for polar lipids on a normal-phase column:
A mixture of phospho-and glycolipid standards was applied to the PVA-Sil column. As shown in Figure 1, the phospholipid groups could be well separated from one another. The larger amounts of accompanying nonpolar compounds elute first from the column, before the lipids of interest. The first class of polar lipids to elute are PEs (at about 10 min). We observed some separation within a single class, especially for sphingomyelins, due to the minor hydrophobic interaction properties of this column, but even the SMs still separated from diacyl-and monoacylglycerophosphocholines that eluted before and after them, respectively. Cerebrosides and gangliosides appear at distinctive positions over the same gradient ( Figures 1, 2). Although the cerebroside shown here, N-stearoyl-DLdihydrogalactocerebroside, clearly eluted earlier than the phospholipid standards, Nstearoyl-DL-dihydrolactocerebroside, which is one hexose larger and therefore more polar, runs slightly slower than the PE standards (not shown). As expected, the palmitoyl species are not separated from their respective stearoyl homologs. Figure 2 also demonstrates the usefulness of this gradient for the determination of another important lipid class, the gangliosides.
In earlier experiments, ammonium acetate was evaluated as the stabilizing salt. While its separation and ionization efficiency were found to be similar to ammonium formate, some of the lipid species formed double peaks, likely due to incomplete protonation (not shown).

Gradient for nonpolar lipids on a normal-phase column:
similar chromatographic results; heptane was chosen for convenience. Additions of salts or buffers to the heptane reservoir in amounts sufficient to allow for proper ionization were found to interfere with the quality of the separation, so a 2.5-µL/min feed of 10 mM ammonium acetate in IPA/MeOH/water 60:40:1 delivered from a syringe pump was added, after the postcolumn split, to the flow (about 10 µL/min., or 10 %) directed towards the mass spectrometer. This addition could be expected to slightly degrade the appearance of the chromatogram, but should not significantly affect the separation itself or the fraction collection. A mixture of nonpolar lipids and a PC was applied to the PVA-Sil column. As shown in Figure 3, all species were retained on the column to a certain degree. Fatty acid methyl esters and cholesteryl esters were barely resolved from one another, and diacyglycerols eluted together with cholesterol and free fatty acids. The acids are only visible in negative ion mode, and therefore do not appear in the positive-ion spectrum shown in Figure 3. The related groups (TAG, DAG, and MAG; cholesterol and its esters) were each well separated. It must be noted that normal column washes were not sufficient for re-equilibration; the system had to be purged for about 3 min at 1 mL/min without the column prior to re-equilibration of the column between runs. This problem might be specific for this LC system or assembly.

Gradients for PE and PC lipids on a reversed-phase column:
Separation of the molecular species of diacyl-PE and -PC standards could be achieved on a Definition of the molecular composition of the lower abundance components requires LC-MS runs performed on one of the more sensitive, automated instruments, as described below.

Analysis of B100-LDL and B67-lipoproteins:
The data obtained for a normal LDL lipid sample was compared to that acquired for three different samples (from the same individual) of lipids extracted from apolipoprotein B67containing lipoprotein fractions. Preliminary nanospray MS survey spectra already indicated distinct differences: most obviously, a lower amount of cholesterol and cholesteryl esters, and a higher amount of triacylglycerols. Figure 5 shows the mass spectra, summed over the elution window 2-24 min, of the gradient for polar lipids on the PVA-Sil column, for the B100-LDL sample and for the B67 lipid samples. Even though the appearance is different from the nanospray spectrum, its information content is the same. However, the possibility for signal suppression of the polar lipids by the more abundant cholesterols and TAG can now be excluded. More information was gained by summing up the ion chromatograms for both PVA-Sil gradients, fraction by fraction.
Distinct pattern differences between lipids derived from human B67-LDL and -IDL and those from normal B100-LDL could be seen, especially for cholesteryl esters ( Although the interpretation is not so straightforward, because there are two acyl residues per molecule, here also a pattern with a lower amount of palmitoleic acid and arachidonic acid can be assumed for the B67-LDL sample. Analysis of these fractions on the reversedphase column, using the triple quadrupole instrument operated at high cone voltage, allowed matching of the major fatty acyl components, as shown in Figure 8. The results suggest a narrower distribution of fatty acids in the PCs from B67-LDL; they are more centered around the 18:2 species (likely linoleic acid) than are those from normal LDL.  (9). No clear differences between the PE species in B67-LDL and B100-LDL could be found.

Quantitation:
Standard mixtures containing about 50 ng lipid per class and injection were run under identical conditions on the PVA-Sil column before and after the biological samples, and a blank run was performed when the switch was made from standard to unknown samples.
Calculations of the quantities reported here were based on either peak heights or peak areas for single ion current signals at the expected retention times. Based on the reproducibility of standards run on the same day including tests of linearity, the error margin is currently about 20% for individual polar lipids, and up to 50% (for cholesterol) for nonpolar lipids, while the error margins for relative amounts during the same run are at about half these amounts.
Initially, we explored internal quantitation using a deuteriated GPC species, dimyristoylsn-glycero-3-phosphocholine-d 54 , but this lipid exhibited incomplete deuteriation. This feature made the interpretation unnecessarily complex, and increased the likelihood of interference with compounds of interest during the runs.

Discussion:
The characterization of many lipid samples can be accomplished with ESI MS (5-6).
However, samples are often encountered that have insufficient material, or are too complex to be reliably analyzed with standard nanospray methodology. As outlined below, LC and LC-MS methods have been developed for a variety of compound classes, but a robust LC-MS methodology to characterize a broad range of lipids has so far been lacking. Recent sophisticated advances in ESI-MS (5) address these issues, but the LC-MS methodology presented here has the advantage that it is based on relatively simple instrumentation that should be available in all MS laboratories. The primary goal of these studies was to devise an efficient and reliable methodology for routine use in our own laboratory; it should be easily reproducible elsewhere and should provide a basis for developments involving further automation and more advanced instrumentation.
The initial separation of lipids on Silica 60 is, in different variations, a well-established procedure. A laborious but useful separation of most classes is theoretically possible even by this step alone (10), and it has been employed, for example, for the enrichment of one or more phospholipid classes prior to reversed-phase HPLC (11)(12). We utilize it for a simple two-step elution protocol such that nonpolar lipids are eluted with MTBE, and polar lipids with MeOH. Addition of 0.2 % acetic acid to MTBE can improve the elution of certain lipids (10), but might be problematic when very sensitive compounds are being analyzed. This procedure is especially useful for analysis of contaminated samples, especially those that may contain particulate matter (in order to protect the following column and LC system), for samples containing very small amounts of polar lipids together with abundant nonpolar lipids or vice versa, or when, in the following step, pure fractions need to be collected. As described for the following steps, this initial separation does not always have to precede methods for the quantitation and characterization of compounds.
The gradients used on both columns were adapted from established procedures, not employing MS detection, published for separation of lipids by thin layer chromatography and Urwin (14). Also useful was information on the separation of different groups of compounds on normal-phase (15)(16)(17) or reversed-phase (18-21) resins, with subsequent MS detection. A review of the published methodology is available at www.cyberlipid.org. Our selection of solvent systems has been based, not only on their chromatographic properties, but also on their compatibility with mass spectrometry (no nonvolatile salts, no ion pairing reagents, etc.) and, where possible, on their lesser toxicity.
We chose the PVA-Sil column, with polymeric vinyl alcohol bonded silica as the resin, because of the robustness of the material. It can be washed with water, as well as hexane, and functions stably over a wide pH range. The column assembly has been used intensively for more than a year on a broad variety of samples of that differed widely in purity, and so far we have not seen any decrease in the quality of the analytical results.
Ammonium formate is added to the solvents for two reasons. Ammonium adduct ions, [M+NH 4 ] + , are known to be more stable than [M+H] + ions in the positive on mode, and are easier to fragment than the metal-cationized species, for example, the sodium adduct ions, The most frequently used solvent, chloroform, was replaced by MTBE, the solvent that had been reported to be chromatographically superior to chloroform by Hamilton and Comai (13). MBTE is slightly less polar, but is better miscible with small amounts of water in a gradient, and is also slightly less toxic (MSDS data sheets).
The necessity of a postcolumn feed for highly nonpolar solvents, as well as the technical difficulties involved in serving all lipid classes well over a single gradient, suggested the 14 use of different solvent gradients for nonpolar and polar lipids. In general, it was by on December 20, 2009 www.jlr.org Downloaded from unnecessary to separate the compound classes in advance, since polar lipids can be simply washed from the column after the first gradient, while the flowthrough containing the bulk of nonpolar lipids elutes several minutes before the first of the phospholipids tested, so long as the column is not overloaded. The flowthrough fraction in the polar lipid gradient can also be used as the starting material for the nonpolar lipid gradient (12).
Monoacylglycerols can be determined with either gradient, or likely with both, if polar and nonpolar lipids have been separated in advance. The chromatogram for the polar lipid gradient features well-resolved, sharp signals and is quite reproducible. In the chromatogram for the nonpolar lipids, however, the peaks for the acylglycerols and cholesteryl esters exhibited significant tailing, and the retention times sometimes shifted, especially for triacylglycerols. The use of another column, e.g., a diol column, could possibly solve the first problem; the second could be addressed by automating the washes between runs, if this is allowed by the chromatographic system being used. In any case, the gradient described above separated the species well enough to distinguish among them and, therefore, allowed for quantitation and further characterization.
Reversed-phase separation of lipids has been typically published for one or a few compound classes at once. We used it in a way that enabled us to further characterize the PVA-Sil fractions, developing the method by using the readily available triple quadrupole instrument (QQQ). The goal was to use the same LC method later with a more sensitive quadrupole orthogonal time-of-flight (QoTOF) instrument with automated MS/MS data acquisition. After consideration of the components likely to be encountered, the Atlantis dC18 column was chosen for its complete silanol endcapping, in order to avoid unwanted interactions, and for its diameter of 300 µm: its optimal flow rate of 5 µl/min can be directly sprayed towards either the Quattro II QQQ or the QStar QoTOF mass spectrometer. The solvents are compatible with those used for peptide analysis, so that no completely separate system is required for the analysis of lipids in a laboratory usually set up for peptide sequencing by LC-MS/MS. As shown here for PEs and PCs, the mass spectra obtained during the reversed-phase LC-MS experiments using high cone voltage in the ESI source (Figures 4, 8) gave sufficient information about the major molecular species in a mix to enable, in positive ion mode, assignment to the PE or PC group, and, in negative by on December 20, 2009 www.jlr.org Downloaded from ion mode, to determine which acyloxy fragments are derived from it. For PCs, a high cone voltage in the negative ion mode also helps to avoid generating heterogeneity in the signals due to variation in acetate addition and/or methyl group loss. The appropriate precursor or neutral loss scans could be applied for better selectivity when only the ions of interest were to be detected. The isolated PC fractions from the biological samples were also characterized by this method. Nevertheless, nanospray MS/MS, as described elsewhere (3)(4)(5), performed on the QStar QoTOF mass spectrometer is clearly a more reliable method than LC/MS on the Quattro II QQQ mass spectrometer, due to the higher sensitivity this mass spectrometer allows and the definite correlation of MS signals to their fragments. Alternative LC methods have been published for certain compound classes, like silver ion chromatography (23), or the addition of other metal ions for ionization and fragmentation (1,6). These likely should only be tried on dedicated columns; regarding either choice, we were reluctant to risk contaminating our system that is used for many other purposes, and therefore did not have the opportunity to obtain information about the long-term influence of metal ions on the mass spectrometer performance.
Apolipoprotein B67 is a truncated version of apolipoprotein B100 that corresponds to its Nterminal 67%. It has been found in some cases of familial hypobetalipoproteinemia characterized by low levels VLDL and LDL and high HDL cholesterol (7). The lipids analyzed in the study reported herein were obtained from a B67 homozygote. Although this variant does not seem to be harmful, detailed characterization of the lipids associated with this mutated species may nevertheless afford some insights on the interaction of apoB with certain lipid classes. In this study, one sample of B100-LDL and three B67-containing lipid fractions from the same patient were analyzed and the results were compared to published data for B100-LDL. A larger number of samples will need to be processed 16 before broad medical conclusions can be made. While the absolute amounts were by on December 20, 2009 www.jlr.org Downloaded from expectedly variable, especially for the nonpolar lipid groups in the B67 samples, the molecular pattern within each group was surprisingly consistent, and the distribution in the B67 lipids was clearly different from that of the corresponding B100 lipids.
For each analysis, the TICs of the normal-phase LC-MS runs were summed up fraction-byfraction; this process allows for a direct comparison by compound classes. For most lipid classes, in spite of different total amounts, differences in the molecular pattern between the B67 and the B100 lipids were small, but they were more obvious in the case of cholesteryl esters and glycerophosphocholines. The most dominant species in the two cases were the same; they distinguished, for example, LDL lipids from HDL lipids. In a comparison not shown here, the oleic acid containing compounds were found to dominate in HDL.
Differences between B67 and B100 lipids existed in the distribution of less-abundant components, and were very similar among all three B67 samples. The relative abundances of minor species containing acyl fragments, e.g., 16:1, 20:4,  Plasmalogens are known to be constituents of LDL PE as well as LDL PC (24) and are often found to be present in tissue PE at considerable levels (1). One possible approach to confirm these structures would be mild acid hydrolysis of the vinyl ether linkage (9); this step will likely be part of a more detailed analysis of these LDL extracts.
Overall, it appears that the B67 mutant has a somewhat higher binding selectivity for certain acyl groups on lipids than does normal B100. Whether these differences are due to the apoB forms or whether they are diet-related remains to be determined, as the present dataset is too small to draw further conclusions.
Quantitation is generally difficult for complex lipid mixtures. While the major factor that determines ionization efficiencies in positive or on negative ion modes is the compound class (for polar lipids, the nature of the head group), the length and degree of saturation of the acyl chains also play a role. An example of the challenge has been provided by Brügger et al. (3) who used four different, well characterized internal standards in order to correct for acyl group chain length variations in a precursor ion scan for PCs. Relative quantitation of aminophospholipids can be achieved by isotope-tagging (26). While the accuracy of relative abundance determinations is sufficient for the characterizations and comparisons described in this work, the estimations were performed only semi-quantitatively, in that the dataset used for the evaluation and the sample characterization is small, only external standards were used, and no correction for the bound fatty acids was performed. According to Han and Gross (5), in this context, the unequal response is a minor issue. Although internal standards were evaluated, they were not used in the final analysis since the deuteriated compound was not sufficiently homogeneous, and the PE species we had considered as a candidate for the internal standard might have interfered with compounds to be identified. For general use, carrying out comparative runs with and without internal standards would likely be too time consuming. If so, internal standards that may be employed to achieve more quantitative results include MAGs, which can be used in both 18 gradients on the normal-phase column, or other, non-natural PEs, which are likewise efficiently ionized in both ion modes. The quantitation as outlined here needs to be optimized in order to compare its potential against existing methods, especially also by the use of more automated instrumentation.
As an additional note, we observed that the effect of varying the cone voltage depends on the specific sample cone supplied by the manufacturer of the mass spectrometer: different results were obtained for sample cones that could not be distinguished from one another by visual examination. In some cases, similar in-source fragmentation could be observed with cone voltages 20 V lower (absolute value) than reported here.
The LC-MS methodology presented provides a fairly robust and technically simple method for the investigation of complex lipid mixtures. The method used here for quantitation via external standards allows for a semiquantitative characterization of the total composition and can, when necessary, be improved by an appropriate choice of internal standards. The methodology should be easily adaptable for higher throughput and sensitivity using more sophisticated LC and MS instrumentation than we used for this initial approach.     Correlating ion intensities and retention times therefore allows for the determination of the fatty acid composition in the major compounds. All of the injected PC standard mix (100 ng total) was directed towards the mass spectrometer.