Lipidomics by ultrahigh performance liquid chromatography-high resolution mass spectrometry and its application to complex biological samples

Abstract

An improved approach for selective and sensitive identification and quantitation of lipid molecular species using reversed phase chromatography coupled to high resolution mass spectrometry was developed. The method is applicable to a wide variety of biological matrices using a simple liquid-liquid extraction procedure.

Together, this approach combines three selectivity criteria: Reversed phase chromatography separates lipids according to their acyl chain length and degree of unsaturation and is capable of resolving positional isomers of lysophospholipids, as well as structural isomers of diacyl phospholipids and glycerolipids. Orbitrap mass spectrometry delivers the elemental composition of both positive and negative ions with high mass accuracy. Finally, automatically generated tandem mass spectra provide structural insight into numerous glycerolipids, phospholipids, and sphingolipids within a single run.

Method validation resulted in a linearity range of more than four orders of magnitude, good values for accuracy and precision at biologically relevant concentration levels, and limits of quantitation of a few femtomoles on column.

Hundreds of lipid molecular species were detected and quantified in three different biological matrices, which cover well the wide variety and complexity of various model organisms in lipidomic research. Together with a reliable software package, this method is a prime choice for global lipidomic analysis of even the most complex biological samples.

1. Introduction

The general term “lipids” is used to describe a plethora of biomolecules that differ in structure as well as in function, ranging from simple fatty acids to complex saccharolipids and from use in energy storage to signal transduction and as vitamins [1,2]. Lipidomics is the general study of lipids and their contributions to both physiological processes and pathological conditions, such as cancer and diabetes [3].

The two approaches in lipidomic analyses are shotgun techniques and separation techniques [4]. Shotgun techniques rely on stable infusion of sample into the ion source without prior separation and subsequent detection by either high-resolution mass spectrometers [5–12] or a combination of tandem mass spectrometry (MS/MS) scans [13–15] to identify lipids by their mass and/or characteristic head group and chain fragments. Similar techniques are intrasource separation [16], where addition of modifiers causes selective ionization of lipid classes, and ion-mobility mass spectrometry, where the physicochemical properties of ions can be used to separate lipid classes and structural or even positional isomers [17–23].

While shotgun methods are an excellent tool for high-throughput profiling, ion suppression effects may limit their ability to detect low-abundance components, and the lack of separation prior to mass analysis may diminish the informative value of MS/MS spectra. Adducts of lipids belonging to different classes, or their isotopes, may have the same nominal mass which cannot be resolved during MS/MS isolation, leading to an MS/MS spectrum with signals from multiple precursor lipids [24]. Also, some labile lipids, such as phosphatidylserine [25–27] or cerebrosides (unpublished observation), may undergo spontaneous in-source fragmentation, artificially forming ions of phosphatidic acid (PA) or ceramides, respectively. Chromatography causes these lipids to elute at different retention times so that the artificially generated ions can be clearly discriminated from the authentic sample components.

Furthermore, separation of lipids has the advantage of minimizing the negative influence of interfering compounds in terms of ion suppression and mixed MS/MS spectra. Ultrahigh performance liquid chromatography (UHPLC), the evolution of conventional high performance liquid chromatography (HPLC) brought by the advent of sub-2 µ particle columns, is the most widespread separation method. Additionally, core-shell columns deliver performance comparable to UHPLC at HPLC pressures [28]. Supercritical fluid chromatography (SFC) is a method with promising potential for lipidomic analyses providing shorter analysis times and highest chromatographic resolution [29,30].

The method previously developed in our lab [31], while at the time able to generate unprecedented data quality, still suffered from some drawbacks, which collectively made it unsuitable for analysis of highly complex lipidomes. Firstly, use of different HPLC mobile phases for positive and negative polarity caused lipids to elute at different retention times in positive and negative mode, complicating data analysis. Secondly, the chromatographic resolution and separation power proved to be insufficient for more sophisticated lipidomic samples. Additionally, the hybrid ion trap-Fourier transform ion cyclotron resonance (FTICR) instrument used had limited scan speed, causing long duty cycles. Furthermore, the sensitivity of both the ion trap and the FT analyzer (both are technologies from more than a decade ago) are quite limited when compared to more timely mass analyzers.

While most lipidomic research is focused on study of mammalian organisms containing almost exclusively fatty acyls with an even number of carbon atoms, lower organisms such as Caenorhabditis elegans have a high amount of odd-carbon fatty acyls in their lipids [6,32,33]. This drastically increases the complexity of the lipidome as well as the number of potentially interfering compounds for each analyte and requires an appropriately powerful method for correct analysis.

Here we present a massively improved approach for lipidomic profiling of even the most complex biological samples with the potential to structurally analyze and quantify hundreds of glycero-, phospho- and sphingolipids in a single run. Highest selectivity is ensured by combining information from chromatography, high-resolution full scans and tandem mass spectra in conjunction with automated data analysis.

2. Material and Methods

2.1. Chemicals

Acetonitrile, isopropanol, methanol, tert-methyl-butyl ether (MTBE) (all Chromasolv grade) and ammonium formate (LC/MS grade) were purchased from Sigma-Aldrich (St. Louis, MO, USA); chloroform and formic acid were purchased from Merck (Darmstadt, Germany). Deionized water was obtained from an in-house MilliQ Gradient A10 system (Millipore, Billerica, MA, USA).

2.2. Lipid standards

LIPID MAPS quantitative lipid standards (1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine, 1-heptadecanoyl-2-(9Z-tetradecenoyl)-sn-glycero-3-phosphocholine, 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-icosatetraenoyl)-sn-glycero-3-phosphocholine, 1-henicosanoyl-2-(4Z,7Z,10Z,13Z,16Z,19Z-docosahexaenoyl)-sn-glycero-3-phosphocholine, 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphoethanolamine, 1-heptadecanoyl-2-(9Z-tetradecenoyl)-sn-glycero-3-phosphoethanolamine, 1-heptadecanoyl, 2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phosphoethanolamine, 1-heneicosanoyl-2-(4Z,7Z,10Z,13Z,16Z,19Z-docosahexaenoyl)-sn-glycero-3-phosphoethanolamine, 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-[phospho-rac-(1-glycerol)], 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1'-rac-glycerol), 1-heneicosanoyl-2-(4Z,7Z,10Z,13Z,16Z,19Z-docosahexaenoyl)-sn-glycero-3-phospho-(1'-rac-glycerol) (ammonium salt), 1-heptadecanoyl-2-(9Z-tetradecenoyl)-sn-glycero-3-phospho-(1'-rac-glycerol) (ammonium salt), 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphoserine (ammonium salt), 1-heptadecanoyl-2-(9Z-tetradecenoyl)-sn-glycero-3-phosphoserine (ammonium salt), 1-heptadecanoyl, 2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phosphoserine (ammonium salt), 1-heneicosanoyl-2-(4Z,7Z,10Z,13Z,16Z,19Z-docosahexaenoyl)-sn-glycero-3-phosphoserine (ammonium salt), 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phospho-(1’-myo-inositol)(ammonium salt), 1-heptadecanoyl-2-(9Z-tetradecenoyl)-sn-glycero-3-phospho-(1'-myo-inositol)(ammonium salt), 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1’-myo-inositol)(ammonium salt), 1-heneicosanoyl-2-(4Z,7Z,10Z,13Z,16Z,19Z-docosahexaenoyl)-sn-glycero-3-phospho-(1’-myo-inositol)(ammonium salt), 1-(10Z-heptadecenoyl)-2-hydroxy-sn-glycero-3-phosphocholine, d5-TG internal standard mixture I, d5-DG internal standard mixture I, ceramide/sphingoid internal standard mixture I, and cardiolipin internal standard mixture I), 1,2-dilauroyl-sn-glycero-3-phosphocholine, 1,2-dilauroyl-sn-glycero-3-phosphoethanolamine, 1,2-dilauroyl-sn-glycero-3-phospho-L-serine (sodium salt), 1,2-dilauroyl-sn-glycero-3-phospho-(1'-rac-glycerol) (sodium salt), 1-pentadecanoyl-2-hydroxy-sn-glycero-3-phosphocholine, 1-(10Z-heptadecenoyl)-2-hydroxy-sn-glycero-3-[phospho-L-serine] (sodium salt), 1-heptadecenoyl-2-hydroxy-sn-glycero-3-phosphoethanolamine, N-heptadecanoyl-D-erythro-sphingosine, N-heptadecanoyl-D-erythro-sphingosylphosphorylcholine, 1,2-dilauroyl-sn-glycerol, 1-(1Z-octadecenyl)-2-oleoyl-sn-glycero-3-phosphocholine, 1,2-distearoyl-sn-glycero-3-phosphocholine, 1-stearoyl-2-oleoyl-sn-glycero-3-phosphocholine, 1-(1Z-octadecenyl)-2-oleoyl-sn-glycero-3-phosphoethanolamine, 1,2-distearoyl-sn-glycero-3-phosphoethanolamine and 1-stearoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine were all purchased from Avanti Polar Lipids (Alabaster, AL, USA) and triheptadecanoin from Larodan Fine Chemicals AB (Malmö, Sweden).

2.3. Lipid extraction

The MTBE based lipid extraction protocol employed was a modified version of the original extraction protocol published by Matyash et al. [32]. Methanol (1.5 mL) and MTBE (5 mL) were added to the samples (approx. 100 mg of C. elegans embryos, 50 mg murine liver or 10⁶ A549 adenocarcinomic human alveolar basal epithelial cells, respectively) in 12 mL glass tubes with teflon lined caps and the mixture was incubated for 10 minutes in an overhead shaker at room temperature. After addition of 1.25 mL deionized water and 10 minutes of additional shaking, the mixture was centrifuged for 5 min at 1 350 x g and the upper phase was transferred to a new glass tube. The lower phase was re-extracted with 2 mL of the upper phase of MTBE/methanol/deionized water (10:3:2.5, v/v/v) and again the upper phase was collected, combined with the upper phase from the first extraction, evaporated in a vacuum centrifuge (Thermo Fisher Scientific, Waltham, MA, USA) and dissolved in chloroform/methanol (1:1, v/v) for storage at -20 °C. The liver extract was dissolved in a volume of 30 mL, the C. elegans extract in 500 µL and the A549 cell extract in 1 mL to account for the varying lipid content. Prior to analysis, the internal standards were added (see section 2.6) and the storage solvent was replaced with the injection solvent isopropanol/chloroform/methanol (90:5:5, v/v/v).

2.4. LC method

Chromatographic separation was performed on a Waters (Waters, Milford, MA, USA) BEH C8 column (100 x 1 mm, 1.7 µm), thermostatted to 50 °C in a Dionex Ultimate 3000 RS UHPLC system. Mobile phase A was deionized water containing 1 vol% of 1 M aqueous ammonium formate and 0.1 vol% of formic acid as additives. Mobile Phase B was a mixture of acetonitrile/isopropanol 5:2 (v/v) with the same additives. Gradient elution started at 50 % mobile phase B, rising to 100 % B over 40 minutes; 100 % B were held for 10 minutes and the column was re-equilibrated with 50 % B for 8 minutes before the next injection. The flow rate was 150 µL/min, the samples were kept at 8 °C and the injection volume was 2 µL.

2.5. MS method

The Orbitrap Velos Pro hybrid mass spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) was operated in Data Dependent Acquisition mode using a HESI II ion source. Prior to experiments, lens settings were tuned and source parameters were optimized to the signal of PE 16:0/18:1 using the protonated and the deprotonated molecule. Every sample was measured once in positive polarity and once in negative polarity. Ion source parameters for positive polarity were as follows: Source Voltage: 4.5 kV; Source Temperature: 275 °C; Sheath Gas: 25 arbitrary units; Aux Gas: 9 arbitrary units; Sweep Gas: 0 arbitrary units; Capillary Temperature: 300 °C. Ion source parameters for negative ion mode were: Source Voltage: 3.8 kV; Source Temperature: 325 °C; Sheath Gas: 30 arbitrary units; Aux Gas: 10 arbitrary units; Sweep Gas: 0 arbitrary units; Capillary Temperature: 300 °C. Full scan profile spectra from m/z 400-1200 for positive ion mode and from 400-1600 in negative ion mode were acquired in the Orbitrap mass analyzer at a resolution setting of 100 000 at m/z 400. For MS/MS experiments, the 10 most abundant ions of the full scan spectrum were sequentially fragmented in the ion trap using He as collision gas (Normalized Collision Energy: 50; Isolation width: 1.5; Activation Q: 0.2; Activation Time: 10) and centroided product spectra were collected. The exclusion time was set to 10 s.

2.6. Data processing and quantitation

LC/MS data were processed using Lipid Data Analyzer (LDA), as previously described [34]. Briefly, the algorithm identifies lipids with a 3D algorithm, using the three dimensions m/z, retention time, and intensity to correctly integrate peaks, while also taking into account the isotopic distribution. MS/MS spectra (displayed, but not automatically interpreted by the program) were manually inspected to obtain information about characteristic head group fragments and fatty acyl composition, as described previously [35–37]. Lipid species contributing less than 1 % to the total amount of the respective lipid class were omitted from the analysis.

For quantitation purposes, phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylserine (PS), lysophosphatidylcholine (LPC), lysophosphatidylethanolamine (LPE) and sphingomyelin (SM) were analysed as [M+H]⁺ ions, ceramides (Cer) and hexosylceramides (HexCer) as [M+H–H₂O]⁺, and diacylglycerols (DG) and triacylglycerols (TG) as [M+NH₄]⁺ ions in positive polarity, while cardiolipin (CL), phosphatidylinositol (PI) and phosphatidylglycerol (PG) were quantified in negative ion mode as [M–H]^– ions.

Lipids were quantified using LIPID MAPS internal standards containing odd-chain fatty acyls for phospholipids or sphingolipids, or deuterated glycerolipids. Quantitation of naturally occurring lipids was performed using a one-point calibration. To 60 µL of the liver extract and the A549 cell extract, all LIPID MAPS internal standards listed in chapter 2.2 were added to a concentration of 1.5 µM prior to measurement. To 60 µL of the C. elegans sample, only PC 25:0, PE 25:0, PS 25:0, PG 25:0, PI 25:0, the deuterated TG and DG mix, and ceramide/sphingoid internal standard mixture were added at a concentration of 5 µM. Lysophospholipids, for which there is no quantitative LIPID MAPS standard available, were quantified using a 25:0 diacyl species.

Lipid shorthand nomenclature is used according to Liebisch et al. [38]. Furthermore, we adhere to the recommendations by D.A. Volmer concerning nomenclature in mass spectrometry publications [39]

2.7. Method validation

Method validation was performed similar to [31]. A liver lipid extract was spiked with synthetic standards (PC 12:0/12:0, PE 12:0/12:0, PS 12:0/12:0, PG 12:0/12:0, LPC 15:0/0:0, Cer d18:1/17:0, SM d18:1/17:0, TG 17:0/17:0/17:0 and DG 12:0/12:0/0:0) to 13 concentration levels ranging from 1 nM to 100 µM. Each concentration level, along with an unspiked lipid extract, was analyzed in triplicate as described in sections 2.4-2.6. Lipids used for validation purposes were analyzed in both polarities (see also Table 1). While arguably a better choice as validation standards, the LIPID MAPS internal standards used for quantitation are supplied only at relatively low concentrations (4-25 µmol/L); thus the lipid species listed above were chosen as validation standards.

Table 1.

Validation data (unweighted) for lipids in positive and negative mode, determined for one molecular species per lipid class. Values for lower limit of quantitation (LLOQ) were conservatively determined because of the minimal but detectable natural abundance of the measured analytes. n=5; n.a.: not applicable (not within linear range)

					Accuracy (%)			Precision (RSD %)
Lipid molecular species	Type of ion	LLOQ (fmol on column)	Linear range (μM)	R2	Low (0.05 μM)	Medium (1 μM)	High (50 μM)	Low (0.05 μM)	Medium (1 μM)	High (50 μM)
TG 51:0	[M+NH4]+	100	0.05–7.5	0.9874	176	95	n.a.	7.1	1.9	1.6
DG 24:0	[M+NH4]+	100	0.05–10	0.9918	0	1	n.a.	27.1	5.2	4.9
PC 24:0	[M+H]+	10	0.005–25	0.9981	144	115	n.a.	2.0	1.7	1.9
	[M+HCOO]−	10	0.005–25	0.9886	91	113	n.a.	5.4	0.9	2.1
PE 24:0	[M+H]+	10	0.005–100	0.9957	97	91	59	1.5	2.1	3.2
	[M-H]−	10	0.005–10	0.9955	109	121	n.a.	4.1	3.6	2.9
PS 24:0	[M+H]+	10	0.005–100	0.9981	54	120	82	5.8	1.7	2.1
	[M-H]−	10	0.005–100	0.9976	57	126	83	6.8	1.6	1.1
PG 24:0	[M+H]+	100	0.05–100	0.9982	28	90	102	6.3	1.6	2.5
	[M-H]−	10	0.005–100	0.9945	117	122	67	2.6	0.8	1.8
LPC 15:0	[M+H]+	10	0.005–25	0.9954	124	119	n.a.	2.3	1.6	2.4
	[M+HCOO]−	20	0.01–100	0.9939	51	102	71	3.1	1.0	2.0
Cer 17:0	[M+H-H2O]+	10	0.005–10	0.9867	78	112	n.a.	4.5	2.8	3.2
	[M-H]−	100	0.05–10	0.9940	100	80	n.a.	3.0	2.1	2.1
SM 17:0	[M+H]+	20	0.01–50	0.9968	135	121	95	1.4	2.4	3.6
	[M+HCOO]−	100	0.05–10	0.9940	76	90	n.a.	3.7	1.8	2.6

The matrix effect was evaluated by comparing the response of the LIPID MAPS internal standards analyzed both alone and spiked into a liver lipid extract at the same concentration.

3. Results and Discussion

3.1. Chromatography

We evaluated multiple columns in terms of chromatographic resolution, peak shape, reproducibility and practicability and found the BEH C8 column to suit our needs best. Supplementary Figures 1 and 2 show example chromatograms obtained from five different RP-HPLC columns. The BEH C8 column provides the necessary chromatographic performance to separate lipids spanning a broad range of polarity: The very polar species LPC 13:0 and LPC 14:0 are chromatographically separated and both elute well after the solvent peak; and there is over 3 mins of elution time left after the retention time of TG 60:1 to allow for elution of even longer triacylglycerols.

Using the same acidic mobile phase for positive and negative ionization we did not observe drastically reduced ionization in negative ion mode but significantly gained selectivity. Every lipid elutes at the same retention time both in negative and in positive ion mode, allowing for alignment of chromatograms to additionally confirm correct species identification (Supplementary Figures 3 and 4). Almost all lipids form sodium adducts in positive ion mode and for some lipids (CL and PS), sodium adducts even occur in negative ion mode as [M 2H+Na]^– [40]. Sodium adducts can be particularly misleading, since they have only a marginal m/z difference from the protonated form of another lipid of the same class. For example, [PC 38:4+Na]⁺ and [PC 40:7+H]⁺ have a difference in molecular mass of only 2.4 mDa which cannot be baseline separated at the mass resolution used (Supplementary Figure 4).

Retention time is an essential parameter in lipidomic LC/MS data analysis. As previously shown and well-described by the equivalent carbon number model [31,41–43], retention of lipids in reversed phase chromatography increases with total number of carbon atoms and decreases with the number of double bonds. The construction of homologous series for each lipid class can be used to discard false positive results from structural isomers belonging to different lipid classes, or nominally isobaric interferences such as the aforementioned sodium adducts (Figure 1). Odd-carbon PE and PC species, in use as quantitative internal standards, are structural isomers of even-carbon PC and PE, respectively (e.g. internal standard PC 31:1 and analyte PE 34:1), but are clearly separated by retention time.

Figure 1.

Construction of homologous series to aid in identification of lipid molecular species. Reversed phase retention behavior of phospholipids is based on number of carbon atoms and degree of unsaturation. Extracted ion chromatograms of protonated PE 37:4 shows numerous interferences from (a) isobaric PC 34:4, (b) the second isotope of PE 37:5, and (c) sodiated PE 35:1. Correct peaks identified by Lipid Data Analyzer are filled red, green background denotes that tandem mass spectra of the corresponding analytes are available. Note that most of the lipid molecular species depicted show multiple chromatographic peaks belonging to structural isomers differing in fatty acid composition.

The method presented delivers good chromatographic performance for more than 10 phospho-, sphingo- and glycerolipid classes, but analysis of classes such as phosphatidic acid (PA) and phosphatidylinositolphosphate (PIP) is hindered by poor chromatographic peak shapes. The strong peak tailing of PA and PIP is attributed to interactions of the terminal phosphate group with stainless steel surfaces [44,45], an effect avoided by using phosphate buffers as HPLC solvent [41], chemical derivatization [46] or hydrophilic interaction liquid chromatography [27,47].

3.2. Lipid detection and identification

The Orbitrap instrument used is able to provide exceptional spectral quality, and routinely reaches values for resolution (FWHM) of over 110 000 for lysophospholipids, over 90 000 for phospholipids, over 80 000 for triacylglycerols and over 70 000 for cardiolipins. Mass accuracy is usually well below 2 ppm, and below 5 ppm even for low-abundance analytes. The ability to obtain MS/MS spectra in the ion trap in parallel to full scanning in the Orbitrap is an enormous benefit. Using this technique, we achieve cycle times of 1.7-2 s (for one Orbitrap full scan at full resolution and 10 ion trap MS/MS scans), which is sufficient for description of almost all analytes with more than 10 data points per mass chromatographic peak. MS/MS coverage is also very good, even without an inclusion list; e.g. for one representative liver sample, tandem mass spectra were available for 127 of the 195 quantified lipid molecular species, and additionally for all internal standards.

In contrast to techniques without precursor ion selection such as MS^E, All Ion Fragmentation, All Ions MS/MS or MS/MS ALL, our approach relies on generation of fragment spectra of only one m/z value. Assignment of fatty acyl composition and position is facilitated when all fragments displayed ideally belong to only one precursor ion. As is the case with all ion trap scans, our MS/MS spectra suffer from low mass cutoff. While the major diagnostic head group fragment (phosphocholine, m/z 184) of long-chain protonated PC is lost by low mass cutoff, MS/MS spectra of the corresponding sodium adducts may be interpreted. These show characteristic head group neutral losses of 59 Da (trimethylamine) and 183 Da (phosphocholine) which are not affected by low mass cutoff. For almost all lipid classes investigated, interpretation of both positive and negative MS/MS spectra delivers specific head group fragments as well as fragments indicative of both identities and positions of the fatty acyls. Exceptions are glycerolipids (TGs and DGs) of which the positions of the fatty acyls cannot easily be deduced, since the fatty acyls at the sn-1 and sn-3 position behave identically upon collisional activation.

Using the approach presented, hundreds of lipids can be identified in a single lipidomic sample. It is necessary to limit oneself concerning the number of lipid species for several reasons: Firstly, the pareto principle also applies to lipidomics, and a small number of lipid species usually constitutes the bulk of the lipid biomass. An in-depth analysis of TGs in murine liver shows that out of a total of 375 identified molecular species, 18 species already represent over 90 %, and 128 species represent over 99.9 % of the total class amount (Supplementary Figure 5). This effect is even more pronounced for the class of PI, where only 5 molecular species represent over 95 % of the total class amount. Thus it is necessary to introduce a cut-off simply to reduce the number of molecular species which contribute only negligibly to the whole lipid class. Another reason why we do not advocate analyzing extremely low-abundance species with global methods is that these species tend to have high values of standard deviation, increasing the chances of finding no difference where in fact there is one. We instead argue that extremely low-abundance compounds should be quantified with specialized methods and not with global approaches. For these reasons, a class-specific cut-off value of 1 % was chosen, which means that for each lipid class, all lipid species which together contribute less than 1 % to the total class amount were omitted from the analysis. We argue that this is a good compromise between comprehensiveness and data analysis effort for a global untargeted approach.

In total, 195 lipid species were identified in murine liver (58 glycerophospholipids, 106 glycerolipids, and 31 sphingolipids), while a total of around 350 lipid species were identified in C. elegans (see Figure 2 for overview and Supplementary Tables 2-4 for complete results). The difference in the number of species found is simply due to the fact that while liver contains in its lipids predominantly fatty acyls with an even number of carbon atoms, lower organisms such as C. elegans contain a large amount of fatty acyls with an odd number of carbon atoms, thereby drastically increasing the number of lipid species.

Figure 2.

Number of glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (SP) identified in murine liver, A549 cells and C. elegans with a 1 % cutoff for each lipid class. Glycerolipids were not analysed in the A549 sample.

3.3. Separation of structural and positional isomers

The chromatographic approach used is able to separate a wide variety of structural isomers, which can then be identified based on their retention behavior or tandem mass spectra. The simplest cases are PC or PE odd-carbon internal standards, which are structural isomers of even-carbon PE or PC analytes, but can easily be differentiated. However, many other diacyl phospholipids or triacylglycerols show more than one peak on the lipid species/bond type level (Figure 3). Inspection of the tandem mass spectra reveals that these peaks almost always correspond to structural isomers with different fatty acyl compositions which are completely or at least partially chromatographically separated.

Figure 3.

Separation of PE structural isomers from a C. elegans lipid extract. The chromatogram shows extracted ion chromatograms of deprotonated PE 34:2. Manual inspection of tandem mass spectra reveals three structural isomers, differing in fatty acyl composition separated by reversed phase chromatography. The positions of the fatty acyls can be determined by comparing the relative intensities of the carboxylate anions.

Many lysophospholipids also show double peaks, which are most likely positional isomers (Supplementary Figure 6). Our observations are consistent with published data [48–51] stating that the regioisomers with the acyl group at the sn 2 position elute earlier in reversed phase HPLC and upon collisional activation of the protonated molecule show a higher relative intensity of the head group-derived fragment (m/z 184 and neutral loss of 141 for LPC and LPE, respectively) as compared to the neutral loss of water. However, these results have not been confirmed with regioisomerically pure standards and it is possible that intra-molecular acyl migration, at least to some extent, may occur artificially, during sample preparation or storage, depending on the exact circumstances [52].

3.4. Quantitation

Classical quantitative analysis, as is customary in pharmacological LC analysis, relies on calibration curves for each analyte, ideally in an analyte-free matrix. In LC/MS lipidomics however, this is not possible due to the lack of analyte-free matrices and the sheer number of analytes investigated. Stable isotope marked internal standards, the gold standard of absolute quantitation, are hardly available and their use is rarely economical. While absolute amounts are often required, they should be treated with care and knowledge of the limitations of their accuracy. For these reasons, absolute quantitation at the highest level of quality is very difficult to achieve. We believe that absolute quantification in the field of LC/MS lipidomics is best considered semi-quantitative and its value and significance lies in comparing multiple groups (physiological/developmental states, wild-type vs. knockout, treatment vs. control etc.) and focusing on fold changes rather than on absolute amounts, similar to pure profiling methods.

The selection of suitable internal standards for lipidomic analysis is challenging. While quantitative standards containing odd-chain fatty acyls are used for lipidomic analysis of higher organisms, they are unsuitable when the organism of interest – such as C. elegans – contains high amounts of odd chain fatty acyls. A compromise would be the use of internal standards which merely correspond to low-abundance endogenous lipid species and spiking them at relatively high concentrations (e.g. 25:0 phospholipid species, as performed here), thereby superimposing the endogenous concentrations. The C. elegans lipid extract was measured with and without addition of the internal standards and the baseline abundance of the analytes corresponding to the internal standards was found to be negligible – either not detected or at least 90-fold lower than when internal standards were added. 1 (1Z alkenyl)-2-acyl-PC and -PE, also termed plasmalogens, for which there are no LIPID MAPS quantitative internal standards available, may be quantified with diacyl–PC and -PE internal standards, as the responses of PC P 36:1 and PE P 36:1 are within 20 % of the respective diacyl species, investigated in triplicate measurements at concentrations of 0.5 and 5 µM for both positive and negative polarities. We also argue that lysophospholipids may be quantified using a short-chain diacyl species if no other suitable standard is available.

Another point to be considered is that the LIPID MAPS diacyl phospholipids internal standards may contain small amounts of lysophospholipids, probably derived from chemical degradation during storage etc. This results in the corresponding lysophospholipid molecular species being present in an injection containing only LIPID MAPS diacyl phospholipid internal standards. Although their relative abundance may seem negligible (in the case of LPE, 0.5-2 % of the area of the respective diacyl PE species), it should be noted that lysophospholipids usually represent only a fraction of the respective diacyl phospholipid class and therefore, degradation of only a minor amount of diacyl phospholipids could already result in an overestimation of the corresponding lysophospholipid species. However, this problem concerns only few molecular species, but their results should be critically evaluated. Source fragmentation in contrast, which would also lead to an artificial increase of lysophospholipids, plays no role here, as the authentic lysophospholipids are easily distinguished by retention time from those generated in the ion source.

Although almost all lipid classes are detectable in both polarities, quantitation was only performed for one adduct ion per lipid class. To increase selectivity, tandem mass spectra of another adduct in the same polarity, or in the other polarity, can be inspected for further confirmation. E.g., while PE was quantified in positive polarity (due to the selectivity offered by the neutral loss of the head group), negative polarity is far better suited for identification of the identities and positions of the fatty acyls.

3.5. Method validation and evaluation of matrix effect

The chromatographic setup used is capable of reproducible performance. Retention time of PC 34:1 remained constant over the course of 50 consecutive injections of a liver extract (relative standard deviation 0.1 %). Relative standard deviation of the peak area of PC 34:1 in five replicate injections of liver, C. elegans and A549 cell lipid extracts was 6.8 %, 5.3 % and 2.2 %, respectively and retention time was unaffected by different kinds of sample matrix (overall relative standard deviation 0.4 %).

It is well established that the electrospray ionization process has a narrow linear dynamic range and the upper concentration limit is frequently cited as 10^-5 M [53–55], although this value is strongly expected to depend on the analytes and exact circumstances. The linearity of the presented method was evaluated for 9 different lipid classes by spiking a liver extract with a non-endogenous or at least low-abundance lipid species in the range of 1 nM to 100 µM. While some lipids were found to exhibit almost perfect linearity over the whole concentration range investigated (e.g. PG 24:0 [M+H]⁺, R²=0.9989 and [M–H]^– R²=0.9945), others show a distinct saturation-like response at higher concentrations (e.g. PC 24:0 [M+H]⁺)(Supplementary Figure 7). It is noteworthy that some lipid classes may have a broader linear range in one polarity than the other, and also adducts of the same lipid class deviate in their linear range (Table 1).

The non-spiked liver matrix already contains detectable although minute amounts of the lipid standards added for validation purposes (e.g. naturally occurring 12:0/12:0 phospholipid species). These standards were nevertheless chosen for validation because their natural abundance is insignificant compared to the vast excess of other sample components and linearity is not negatively affected. While carryover of the injection system can be ruled out, lower limits of quantitation cannot be classically determined. The limit of quantitation was therefore conservatively determined as the lowest concentration point which differed substantially from the non-spiked sample (for an example, see Supplementary Figure 8). LLOQs vary for different lipid classes, but are generally in the range of a few fmol on column (Table 1).

While using multiple internal standards for one lipid class is beneficial, they may differ in their response (i.e. peak area), and additionally suffer from matrix effects, a well-known phenomenon in LC/MS [56,57]. For example, the four PS internal standards differ in their responses by as much as 44 % when analyzed at a concentration of 1.5 µM with and without sample matrix, whereas most other lipid classes analyzed do not exhibit grave matrix effects (Supplementary Table 1). Additionally, instrument response is known to depend on factors such as lipid class, acyl chain length and degree of unsaturation, even within the same lipid class [58]. Thus, internal standards of differing molecular species and eluting at different retention times may experience different degrees of ion suppression, or even ion enhancement, caused by coeluting sample components. Also, ionization efficiency (i.e. number of ions formed in the electrospray process) may be different, as solvent composition is not constant during gradient elution. This effect is observable to at least some degree in most lipid classes for which multiple internal standards are available (Supplementary Figure 9). It is particularly grave for the class of DG, where the internal standards differ enormously with chain length and number of double bonds. For example, under the conditions employed, the signal of IS_d5DG 28:0 is only 2 % of the signal of IS_d5DG 40:8 even within the same sample. On the other hand, most other lipid classes do not show grave differences in instrument response between species differing in chain length and degree of unsaturation.

To minimize these problems, the quantitation software employed uses a sophisticated algorithm, explained in greater detail in the original publication [34], which selects the best internal standard, if multiple internal standards are available – as is the case for most glycerophospholipids. Despite the previously mentioned differences in response, this approach results in very good values for accuracy for almost all lipid classes investigated while precision is excellent throughout (Table 1), resulting in a mean relative standard deviation of 7.1 % for all analytes quantified.

4. Conclusion

A sensitive and highly reliable method for detection and quantitation of molecular species of multiple glycerolipid, glycerophospholipid and sphingolipid classes was developed. The method is applicable to a wide variety of biological matrices with a versatile but simple MTBE extraction procedure. Intact lipid species are detected by a high mass resolution Orbitrap Velos Pro hybrid mass spectrometer in both positive and negative polarity, while tandem mass spectra acquired in the linear ion trap provide deep insight into the molecular structure. The reversed phase chromatography approach used is capable of robust performance, offers important additional selectivity, and is able to resolve structural and positional isomers. Finally, lipids are identified and quantified by ample use of quantitative internal standards together with an automated software package.

Method validation indicates a linear range of four orders of magnitude for most lipid classes investigated, while limits of detection are in the low femtomolar range. The method was applied to murine liver, adenocarcinomic cells, and C. elegans, resulting in quantitative and structural information about hundreds of lipid species in a manner applicable to large-scale lipidomic studies.

Abbreviations

Cer

ceramide

CL

cardiolipin

DG

diacylglycerol

EIC

extracted ion chromatogram

FTICR

Fourier transform ion cyclotron resonance

FWHM

full width half maximum

HexCer

hexosyl ceramide

HPLC

high performance liquid chromatography

LC/MS

liquid chromatography mass spectrometry

LLOQ

lower limit of quantitation

LPC

lysophosphatidylcholine

LPE

lysophosphatidylethanolamine

MS/MS

tandem mass spectrometry

MTBE

methyl tert butyl ether

PA

phosphatidic acid

PC

phosphatidylcholine

PE

phosphatidylethanolamine

PG

phosphatidylglycerol

PI

phosphatidylinositol

PIP

phosphatidylinositolphosphate

PS

phosphatidylserine

SFC

supercritical fluid chromatography

SM

sphingomyelin

TG

triacylglycerol

UHPLC

ultrahigh performance liquid chromatography

ULOQ

upper limit of quantitation