Quantitative structural multiclass lipidomics using differential mobility: electron impact excitation of ions from organics (EIEIO) mass spectrometry

Abstract

We report a method for comprehensive structural characterization of lipids in animal tissues using a combination of differential ion mobility spectrometry (DMS) with electron-impact excitation of ions from organics (EIEIO) mass spectrometry. Singly charged lipid ions in protonated or sodiated forms were dissociated by an electron beam having a kinetic energy of 10 eV in a branched radio-frequency ion trap. We established a comprehensive set of diagnostics to characterize the structures of glycerophospholipids, sphingolipids, and acylglycerols, including glycosylated, plasmalogen, and ester forms. This EIEIO mass spectrometer was combined with DMS as a separation tool to analyze complex lipid extracts. Deuterated quantitative standards, which were added during extraction, allowed for the quantitative analysis of the lipid molecular species in various lipid classes. We applied this technique to the total lipids extracted from porcine brain, and we structurally characterized over 300 lipids (with the exception of cis/trans double-bond isomerism in the acyl chains). The structural dataset of the lipidomes, whose regioisomers were distinguished, exhibit a uniquely defined distribution of acyl chains within each lipid class; that is, sn-1 and sn-2 in the cases of glycerophospholipids or sn-2 and (sn-1, sn-3) in the cases of triacylglycerols.

Lipids serve many physiological functions in biosystems, playing essential roles in energetics, cell structure, and cell signaling. They also are implicated as playing direct and causal roles in the pathophysiology of human diseases, including Parkinson’s disease, Alzheimer’s disease, depression, cardiovascular diseases, and various cancers (1–4). Much of lipids’ actions can be related to their structures, with small structural differences in isomeric lipid molecules having dramatic impact on their physiological functions [e.g., the diastereomers prostaglandin-D2 and prostaglandin-E2 (5, 6)]. Recently, a slight permutation in double-bond position in one of the acyl chains of a phospholipid was shown to be a marker of breast cancer (7), and the presence of a Δ7 vs. a Δ9 double bond in a specific molecular species of triacylglycerol (TG) was correlated with adverse cardiovascular outcomes (1). For these reasons, differentiation of lipids by complete structural characterization is “a real promise in achieving true molecular level lipidomics analysis generation” (8). Not only qualitative structure characterization of each lipid molecule, but also quantitative analysis across many lipid classes in tissues, will be the next target in lipidomics to elucidate cross-class lipid metabolism.

Conventionally, NMR is used as a tool for full characterization of lipid structures (9), but it needs a reasonably large amount (approximately milligrams) of purified samples and very long analysis times (approximately days), which results in low throughput. Mass spectrometry (MS) is used widely as a high-throughput method for versatile biomolecular identification (10). Conventional techniques for lipid analysis in MS are based on a combination of separation techniques (liquid chromatography, gas chromatography, ion mobility spectrometry, etc.) and collision-induced dissociation (CID) in the gas phase (1). This approach, however, does not provide full structural information of lipid isomers, which is often limited to characterization of the lipid class and brutto class, which represents the head group and the combined lengths of chains and numbers of double bonds among the acyl chains. To elucidate in-depth structure, including regioisomerism and double-bond locations and their cis/trans isoforms, arduous efforts are required using additional dissociation techniques, such as high-energy CID (11), ozone-induced dissociation (OzID) (12), a UV-induced Paternò-Büchi reaction (7, 13), and UV photo dissociation (UVPD) (14, 15), as well as complex chromatographic strategies.

Recently, we reported a new methodology for the structural characterization of complex lipids using electron-induced dissociation (EID) or, in historical terminology, electron impact excitation of ions from organics (EIEIO) (16) in a novel branched radio-frequency (RF) ion trap (17–21). An electron beam with a kinetic energy of 10 eV was applied to positively charged lipid ions so that the EIEIO spectra displayed informative fragment patterns, which identifies the lipid class, the number of double bonds and their relative positions within the chains, and regioisomerism [i.e., distinction of fatty acyl chain position on the glycerol backbone of glycerolipids (GLs) and glycerophospholipids (GPLs)]. We also demonstrated the ability to distinguish cis(Z)/trans(E) isomers at a double bond in the chains of some classes of intact complex lipids (21). We have applied this technique to single class lipidomes extracted from egg phosphatidylcholines (PCs) (18), brain sphingomyelins (SMs) (19) and TGs in edible oils (20).

In this work, we report a comprehensive quantitative and structural lipidomics workflow that generates a “true” molecular-level lipidomics analysis (8). To move comprehensive diagnostics beyond that previously reported for PCs, SMs, and TGs, we applied EIEIO to standards of multiple lipid classes in GPLs, GLs, and sphingolipids (SLs), where monoglycosylated and diglycosylated GLs and SLs as well as ether-linked and plasmalogen chains were included. For quantitative analysis, a mixture of deuterated lipid standards across multiple lipid classes was added to the specimens, where the absolute concentrations were verified. This comprehensive method was applied on total lipids in the brain and other tissues, with a major focus on the brain sample because it contains many lipid classes with various fatty acyl chains (22–24). We did not include fatty acids, cardiolipins, and gangliosides in this work, because these classes require another approach different from the described methods in this work or further modification of the instruments.

MATERIALS AND METHODS

Samples

Standard samples used in this work are phosphatidylethanolamine (PE): PE 16:0/16:0; phosphatidylserine (PS): PS 16:0/16:0; phosphatidic acid (PA): PA 16:0/16:0; phosphatidylglycerol (PG): PG12:0/13:0; phosphatidylinositol (PI): PI 16:0/16:0; phosphatidylcholine-plasmalogen: PC 18:0p/18:1(n-9Z), phosphatidylcholine-diether: PC 18:0e/18:0e; phosphatidylethanolamine-plasmalogen: PE 18:0p/18:1(n-9Z); ceramide: Cer d18:1/12:0; phytoceramides: Cer t18:0/24:0; glucosyl diacylglycerol: Glu-DG16:0/18:1; galactosyl diacylglycerol: Gal-DG18:3/16:3; glucosyl ceramide: Glu-Cer d18:1/18:1(n-9Z); galactocyl ceramide: Gal-Cer d18:1/18:1(n-9Z); and di-glycosylated ceramides Lactocyl-Cer d18:1/18:0. All these standards were purchased from Avanti Polar Lipids (Alabaster, AL). Natural lipid samples analyzed in this work are the total brain lipid extracts (porcine), egg PAs (chicken), egg PGs (chicken), and liver-PIs (bovine). These complex lipid extracts were also purchased from Avanti Polar Lipids.

Quantitative lipid standards for the Lipidyzer^TM Platform (Sciex, Framingham, MA) were used, which are a mixture of 54 deuterated lipid standards. Other lipid classes not included in this kit were purchased from Avanti Polar Lipids, i.e., quantitative analysis standards (nondeuterated), PI 12:0/13:0, PS 12:0/13:0, and PA 12:0/13:0. Nominal concentrations of each standard are listed in supplemental Table S1.

We used two types of solvents in this work; one contained ammonium acetate at a concentration of 0.5 mM in a mixture of methanol and dichloromethane (50:50, vol:vol), and another contained sodium acetate at a concentration of 0.5 mM in a mix of methanol and dichloromethane (50:50, vol:vol). The HPLC-grade methanol and dichloromethane and two acetate salts were purchased from Caledon Laboratory Chemicals (Georgetown, Canada) and Sigma-Aldrich (Oakville, Canada), respectively. The ammonium acetate solvent produced protonated precursor ions for PCs, PEs, SMs, Cers, and glycosylated lipids, and the sodium acetate solvent produced sodiated precursor ions of all lipid classes. Both working solvents were used to characterize the brain lipidome. The concentration of the brain extract analyzed was 100 µg/ml in both cases.

Instrument

A shotgun approach using infused ESI was used in this work. Turbo V^TM ESI source (Sciex) was operated at spray voltage at +5,000 V. Working sample solutions were infused at a flow rate of 0.3 ml/h by using a syringe pump. SelexION^® (Sciex) differential mobility spectrometry (DMS) was intentionally used with chemical modifier of 2-propanol to eliminate overlap of isobaric species, where the separation voltage was fixed. Not only for separation of lipid classes (25), but also the flow rate of nitrogen counter gas (DR) and the compensation voltage (COV) were tuned finely for each target precursor ion to eliminate isobaric contamination from ¹³C isotopomers and oxidized species. Such DMS optimization strategy was described in our previous work (20). The mass analyzer used in this work was reported previously (18–20), which is a quadrupole TOF mass spectrometer with a branched RF ion trap EIEIO device placed between Q1 mass filter and Q2 collision dissociation device (17). Lipid ions were first isolated by using Q1 as a mass filter (m/z 0.7 units wide) to subject a small subset of the lipid mixtures to EIEIO at one time. The EIEIO device was operated in simultaneous trapping mode (17) for a150-ms duration with electron kinetic energy of 10 eV, with each EIEIO spectrum requiring 1 min of total acquisition time.

RESULTS AND DISCUSSION

Comprehensive structure diagnostics of complex lipids

EIEIO spectra of isolated precursor ions of the standards in protonated and sodiated forms were obtained (supplemental Figs. S1–S3). We generalized the diagnostic rules shown in our previous works (18–20) to universal cases in GPLs, SMs, and GLs (Tables 1, 2). For GPL, the experimentally verified head group diagnostic rule in PCs (18) was simply expanded to universal GPLs by replacing the molecular mass of the head group. New rules were introduced for ceramides (Cers), phytoceramides, and glycosylated lipids by referring the experimental spectra.

TABLE 1.

Lipid class diagnostics

Lipid Class	HG	C Type	O Type	N Type	Additional Diagnostic Peaks
PA-Na	120.966	160.9972	162.9764
PA-K	136.941	176.9719	178.9511
PE-H	142.027	182.0583	184.0375		−141.018
PA-2Na	142.948	182.9793	184.9585
PA-NaK	158.923	198.954	200.9332
PE-Na	164.009	204.0403	206.0195		−141.018, −163.000
PA-3Na	164.929	204.9605	206.9397
PE-K	179.983	220.0143	221.9935
PC-H	184.074	224.1053	226.0845
SM-H	184.074			225.101
PE-2Na	185.990	226.0215	228.0007
PS-H	186.017	226.0483	228.0275		−46.005, −185.008
PA-2NaK	189.890	229.9211	231.9004
MGalDG-Na	201.037	243.0839	245.0632		156.039, −134.058, −180.063
MGIuDG-Na	201.037	243.0839	245.0632		−134.058, −180.063
PE-NaK	201.965	241.9962	243.9754
PC-Na	206.056	246.0873	248.0665		−59.073 −85.088, −183.065
SM-Na	206.056			247.083	−59.073, -85.088, −183.065
PS-Na	207.999	248.0303	250.0095		−46.005, −87.031, −185.008, −206.990
PG-K	210.977	251.0082	252.9874
PG-2Na	216.985	257.0163	258.9955		−74.037
MGDG-K	217.011	259.0579	261.0372
PC-K	222.030	262.0613	264.0405
SM-K	222.030			263.056
PS-K	223.973	264.0043	265.9835
PS-2Na	229.980	270.0115	271.9907
PS-NaK	245.955	285.9862	287.9654
Pl-Na	283.019	323.0505	325.0297		−162.053, no cross-ring cleavage
PHex-Na	283.019	323.0505	325.0297		−60.074, −81.987, −162.053
Pl-K	298.993	339.0243	341.0035		No cross ring cleavage
Cer-Na	−31.018				SL backbone diagnostics
Cer-H	−48.021				SL backbone diagnostics
MGCer-H	−180.063			222.097	−162.053 (no Gal/Glu difference)
MGalCer-Na	−180.063			224.079	156.039, −134.058
MGIuCer-Na	−180.063			224.079	−134.058
DGCer-H	−342.116			384.15	−134.058, −180.063, −296.111
DGCer-Na	−342.116			406.132	−134.058, −180.063, −296.111

Head group (HG) diagnostic peaks, glycerol backbone diagnostic peaks (C types and O types for glycerol backbone defined in Fig. 1), and sphingosine backbone diagnostic (N type defined in Fig. 1) are listed. Additional diagnostic peaks help to validate the head groups. Negative values: neutral loss from the precursor m/z.

TABLE 2.

Regioisomerism and chain diagnostics

sn-1	C2-C1	C1-O	O-(C1 = O)	C2-C3
sn-l acyl diagnostics
14:0	−241.217	−227.201 (0, ±H)	−212.214 (±H)
14:1	−239.201	−225.185 (0, ±H)	−210.198 (±H)
16:0	−269.248	−255.232 (0, ±H)	−240.245 (±H)
16:1	−267.232	−253.217 (0, ±H)	−238.230 (±H)
16:2	−265.217	−251.201 (0, ±H)	−236.214 (±H)
18:0	−297.279	−283.264 (0, ±H)	−268.276 (±H)
18:1	−295.264	−281.248 (0, ±H)	−266.261 (±H)
18:2	−293.248	−279.232 (0, ±H)	−264.245 (±H)
18:3	−291.232	−277.217 (0, ±H)	−262.230 (±H)
20:0	−325.310	−311.295 (0, ±H)	−296.308 (±H)
20:1	−323.295	−309.279 (0, ±H)	−294.292 (±H)
20:2	−321.279	−307.264 (0, ±H)	−292.276 (±H)
20:3	−319.264	−305.248 (0, ±H)	−290.261 (±H)
20:4	−317.248	−303.232 (0, ±H)	−288.245 (±H)
20:5	−315.232	−301.217 (0, ±H)	−286.230 (±H)
22:0	−353.342	−339.326 (0, ±H)	−324.339 (±H)
22:1	−351.326	−337.310 (0, ±H)	−322.323 (±H)
22:2	−349.310	−335.295 (0, ±H)	−320.308 (±H)
22:3	−347 295	−333.279 (0, ±H)	−318.292 (±H)
22:4	−345.279	−331.264 (0, ±H)	−316.276 (±H)
22:5	−343.264	−329.248 (0, ±H)	−314.261 (±H)
22:6	−341.248	−327.232 (0, ±H)	−312.245 (±H)
sn-1 plasmalogen diagnostics
C14p	−225.222	−211.206 (0, −H)
C16p	−253.253	−239.237 (0, −H)
C18p	−281.284	−267.269 (0, −H)
C20p	−309.316	−295.300 (0, −H)
sn-1 ether diagnostics
C14e	−227.237	−213.222 (0, −2H)
C16e	−255.269	−241.253 (0, −2H)
C18e	−283.300	−269.284 (0, −2H)
C20e	−311.331	−297.316 (0, −2H)
SL: backbone diagnostics
d16:1				−212.219
d16:2				−210.199
d18:1				−240.250
d18:2				−238.230
d20:1				−268.270
d20:2				−266.254
d22:1				−296.301
d22:2				−294.286
d24:1				−324.333
d24:2				−322.317

Negative values: neutral loss from the precursor m/z.

The diagnostic process contains three consecutive steps: 1) identification of lipid class and brutto class, 2) regioisomerism, and 3) chain structural identifications (i.e., double-bond locations). Cis/trans identification of the carbon–carbon double bond is possible by using advanced EIEIO (21), but we did not apply this strategy in this study.

Lipid class and brutto class identification.

The lipid classes of complex lipids are identified by the backbone and the head group fragment ions. Glycerol and sphingosine are possible backbones in common complex lipids. The head group gives many varieties in complex lipids. PC, PE, PS, PI, PG, PA, Glu (glucose), Gal (galactose), H (hydrogen), and fatty acyls are common head groups considered in this work. In the cases the head group position is occupied by a hydrogen atom, the lipids are diacylglycerols (DAGs) or Cers. TGs is equivalent to a GL in that its head group position is occupied by a fatty acyl group.

The head group diagnostic peak in EIEIO is often produced as the strongest fragment at the C₃(glycerol backbone)-O bond in the cases of GPLs and GLs or the C₁(sphingosine backbone)-O bond in the cases of SLs (Fig. 1A, C). This rule was generalized to any phospholipid (Table 1) and verified experimentally by using the lipid standards (supplemental Fig. S1). In the cases that a glycan was the head group, the direction of the hydrogen transfer for nonradical production was inverted from GPLs [i.e., the head group lost a hydrogen atom (m/z 202.047) (Fig. 1B; supplemental Fig. S2)]. Frequently, glycolipids contain either Gal or Glu, which EIEIO can distinguish by a cross-ring cleavage fragment; for example, a fragment ion of m/z 156.04 appears in the cases of galactosylated lipids, but we did not observe in the cases of glucosylated ones (supplemental Fig. S2). This cross-ring fragmentation was observed only in the cases of sodiated precursors, but not in protonated cases. This suggested that a sodium ion can be localized on Gal but not on Glu.

Fig. 1.

Dual chain fragmentations of GPLs (A), glycosylated GLs (B), SMs (C), and glycosylated Cers (D).

The two major backbone types, glycerol or sphingosine, were clearly distinguished by EIEIO with a following strong rule caused by a dual chain loss process. In the cases of GPLs, strong two peaks, “C type” and “O type,” appeared in EIEIO spectra as a peak pair with m/z +40.031 (C type) and +42.011 (O type) from the head group diagnostic peak (Fig. 1, supplemental Figs. S1, S2). In the cases of SMs and glycosylated Cers, on the other hand, a single “N-type” peak appeared at m/z +41.027 from the head group diagnostic peak (Fig. 1C, D) (supplemental Fig. S3) (19). This type of dual chain loss fragmentation is allowed to EIEIO, but not to CID.

Regioisomer identification in GLs.

We have reported a glycerol backbone cleavage at C₁-C₂ bond by EIEIO (18, 20), which is a single radical peak. This cleavage has not been observed by CID, and only infrequently in high-energy EID (26) and metastable atom activation (27). This peak solves permutation of two chains at sn-1 and sn-2 (Table 2). Once this peak was identified, chain length and number of double bonds in each chain at sn-1 and sn-2 are calculable using the precursor mass and the head group mass (Table 2) (18). Modified rules for plasmalogens and ether-linked species are also shown in Table 2. In cases of SLs, backbone chain length was easily obtained by finding an intense radical d18:1 peak (precursor m/z −240) in many SLs. In cases that there was no such peak, other types of sphingosine backbone were suggested (d18:2, d:20:1 and others) (Table 2). Once the backbone is identified, length and number of double bonds in the amido chain are also calculable.

Our experiments suggest that EIEIO also distinguishes the sn-2 lyso form from the regular sn-1 lyso form in GPLs. Such lyso PCs obeyed the head group and the backbone diagnostic rules, except that the C-type peaks appear much weaker than the O-type peak (supplemental Fig. S1H, I). In the case of sn-1 lyso, the regioisomer diagnostic rule provides the same product of O type because a nonradical product was generated after the sn-1 chain dissociated from the intact lipid ion. In the case of sn-2 lyso, however, the rule provided a C₂OH loss from the sn-1 position. We found such event in the brain in both sodiated and protonated forms (supplemental Fig. S1I).

Characterizing double-bond positions.

We have reported the identification of carbon-carbon double-bond locations by EIEIO in our publications on PCs, SMs, and TGs (18–20). The rule of thumb for the double-bond diagnostics is that carbon-carbon single bonds were cleaved well, but carbon-carbon double bonds were not cleaved by EIEIO at electron kinetic energy of the 10 eV range. This rule was satisfied by conjugate double bonds, too (20). This rule was confirmed in any complex lipid classes in this work.

These three steps can be applied on each EIEIO spectra manually referring to Tables 1 and 2, but simple offline analysis software was coded for this purpose, which was reported previously for PCs, SMs, and TGs (18–20).

Sensitivity calibration

The sensitivities of each lipid class in the DMS-EIEIO-MS workflow were obtained by the following procedures using the quantitative standards. First, we analyzed brain total lipid extract having a concentration of 100 ng/µL prepared for ESI in either the ammonium acetate type and sodium acetate type solvent systems. Into these solutions, we spiked the deuterated quantitative analysis standards: 10 µL each of deuterated PC, TG mixture, SM mixture, lyso-phosphatidylcholine, lyso-phosphatidylethanolamine, Cer, hexosylceramide, DG mixture, and 20 µL of PE mixture from LipidyzerTM platform, and 10 µL each of PI12:0/13:0, PS 12:0/13:0, and PA 12:0/13:0 from the Avanti’s standards. Once a stable ESI spray was established, COV scan on each solution was replicated three times. The ion counts for each quantitative analysis standard were integrated over the COV at which the samples were transmitted. These ion counts were plotted as a function of the concentration for each lipid class. Linear fitting was applied to the plots, and linear coefficients of three repetitions for each class were averaged. Averaged coefficients, i.e., calibration constants (ions/µM) and their standard deviations, are listed in supplemental Table S2.

Measurement of total brain extract

Our methodology was applied to the total brain extract. The same separation and measurement strategy described in our previous paper was used in this work (20): step 1, survey COV scans collected using different DMS-resolving gases (DRs); step 2, making precursor ion list using the COV scan data at DR = 0 (highest transmission but lowest DMS separation); step 3, finding a COV and DR condition for each precursor ions not to be contaminated by other lipid classes, oxidized species, and ¹³C isotopomers (allowed contamination level was 20% in this work); and step 4, Targeted EIEIO spectrum accumulation at the predetermined COV and DR condition, where accumulation time was from 1 to 5 min, decided by the precursor ion intensity. The working solution was infused at a flow rate of 0.3 ml/h for each type of the solvents. The sodium acetate solvent produced sodiated precursor ions with single to multiple sodium acetate adducts, which we removed from selection for EIEIO analysis based upon the characteristic mass differences between the sodiated or ammoniated lipids and these clusters. Supplemental Fig. S4 shows the most intense EIEIO spectrum and a weak EIEIO spectra to indicate the quality of the spectra by accumulating a preset duration (1 and 4 min), as well to show that this dissociation method covered a wide range of precursor intensity (at least two order of magnitude). Long accumulation helps to improve signal-to-noise ratio.

The experimental intensities of each precursor ion were obtained from the COV scan data by integrating over COV. The obtained ion intensities were converted to weight, mg per 1 g of the brain extract, by using the obtained sensitivity using the quantitative standards (supplemental Table S2).

The EIEIO spectra we obtained were analyzed by the offline analysis software, and all identified lipids were characterized except four spectra shown in supplemental Fig. S5. Fig. 2A lists the number of identified complex lipids in each lipid class and its charge reagent. The column “quant” represents numbers of events that were analyzed quantitatively. Fig. 2B shows the intensity profile of complex lipids, which are reordered in their intensity ranking. In the cases that double-bond locations were not able to be allocated in each chain, they were listed as single mixed events in the table (e.g., PE C18:1p/18:1 and PC 18:1/18:1). The dynamic range of the sensitivity in this crude sample analysis was approximately four orders of magnitude (i.e., from 25 mg/g to 1 µg/g), as shown in Fig. 2B. Potassiated precursors were often observed in both the ammonium acetate and sodiated acetate experiments; the origin of potassium was apparently from the tissue itself. For structure analysis of the potassiated species, the same diagnostic rules were able to be applied except changing the mass of sodium (22.989 amu) to potassium (38.964 amu).

Fig. 2.

A: The number of identified lipids in the porcine brain using ammonium acetate and sodium acetate solvent. This table suggests the applicability of EIEIO on each lipid classes as well as selection of charging reagents, ammonium acetate, or sodium acetate. Quant., analyzed quantitatively. B: Shown are the intensity profiles of identified lipids.

Table 3 lists the most intense 70 lipids, and supplemental Table S3 lists all the identified lipids in the brain observed in this measurement. The strongest species was a cerebroside, GalCer d18:1/24:1(n-9), the second: SM d18:1/18:0, then a PC with an arachidonic chain in sn-2 position followed. In total, captured lipids in this measurement were 466 mg per 1,000 mg of total brain extract, including ¹³C isotopomers. Missed species should be lipids that are outside of our targeted m/z range, such as fatty acids whose m/z would be less than 500 and cardiolipins whose m/z would be over 1,200. Very weak complex lipid species could also be included in this list.

TABLE 3.

Observed complex lipids in the porcine brain (top 70)

Concentration, mg/g	Found Lipids
26.753 (1.3272)	GalCer d18:1/24:1(n-9)
23.239 (0.4935)	SM d18:1/18:0
22.836 (4.8058)	PC 16:0/20:4(n-6,n-9,n-12.n-15)
20.298 (3.8052)	PC16:0/18:1(n-9)
18.579 (2.1825)	PC16:0/16:0
17.870 (1.9347)	GalCer d18:1/24:0 hydroxy
17.704 (3.3133)	PC18:0/18:1(n-9)
17.592 (1.8222)	GalCer d18:1/24:0
11.461 (1.9487)	GalCer d18:1/24:1(n-9)hydroxy
9.602 (0.8445)	SM dl 8:1/24:1(n-9)
9.106 (0.1234)	PE C18:1p/18:1 // (n-6)(31%),(n-9)(69%)
8.340 (1.5634)	PC16:0/18:1(n-6)
7.683 (1.5133)	PC 18:0/20:4(n-6,n-9,n-12,n-15)
7.131 (0.5844)	PE18:0/20:4(n-6,n-9,n-12,n-15)
6.648 (1.2463)	PC16:0/18:1(n-12)
6.267 (1.1768)	PS18:0/18:1(n-9)
5.876 (0.7032)	GalCer d18:1/24:1(n-7)hydroxy
5.633 (0.6354)	PC18:0/20:1(n-9)
5.510 (0.1857)	GalCer d18:1/18:0
5.364 (1.0056)	PC18:1/16:0(n-9)
5.306 (0.1378)	PE18:0/22:6(n-3,n-6,n-9,n-12,n-15,n-18)
5.186 (0.9367)	PC18:0/18:1(n-15)
5.178 (0.5065)	PE C18p/22:6(n-3,n-6,n-9,n-12,n-15,n-18)
4.485 (0.8393)	PC18:0/18:1(n-6)
4.472 (0.7396)	PC18:1/18:1 // (n-3)(14%),(n-6)(34%),(n-9)(52%)
3.594 (0.4785)	PC16:0/22:6(n-3,n-6,n-9,n-12,n-15,n-18)
3.566 (0.0903)	PE C18p/18:1(n-9)
3.437 (0.5382)	PE17:2(n-6,n-9)/20:4(n-6,n-9,n-12,n-15)
3.347 (0.6046)	PC18:0/18:1(n-3)
3.326 (0.1797)	PE C18p/22:4(n-6,n-9,n-12,n-15)
3.233 (0.1124)	PE C18:1p/22:4(n-6,n-9,n-12,n-15)
3.164 (0.2661)	PE C16p/18:1(n-9)
3.065 (0.4135)	DAG_20:4(n-6,n-9,n-12,n-15)_18:0
2.975 (0.0970)	SM d 18:1/22:0
2.696 (0.0748)	SM d20:1/18:0
2.680 (0.2582)	GalCer d18:1/21:1(n-9)hydroxy
2.510 (0.2206)	SM d18:1/24:1(n-3)
2.391 (0.2698)	PC18:0,20:1(n-6)
2.340 (0.0867)	PE18:1/18:1(n-9)
2.334 (0.1347)	PE C18:1p/20:1 // (n-6)(5%),(n-7)(18%),(n-9)(77%)
2.314 (0.5030)	PC C16p/18:0
2.286 (0.0749)	PE 18:0/22:4(n-9,n-12,n-15,n-18)
2.237 (0.2173)	Cer d18:1/18:0
2.230 (0.0696)	Ga1Cer d18:1/23:0
2.170 (0.2764)	PC 18;1(n-7)/20:4(n-3,n-6,n-9,n-12)
2.162 (0.1824)	GalCer d18:1/21:2 hydroxy
2.098 (0.2517)	PC18:0/22:6(n-3,n-6,n-9,n-12,n-15,n-18))
2.051 (0.1309)	PE C16p/20:4(n-6,n-9,n-12,n-15)
1.877 (0.2764)	PC18:1 (n-3)/20:4(n-3,n-6,n-9,n-12)
1.839 (0.2879)	PE 17:2 (n-9,n-12)/20:4(n-6,n-9,n-12,n-15)
1.802 (0.0498)	SM d18:1/20:0
1.802 (0.3268)	PI18:0/20:4(n-6, n-9,n-12, n-15)
1.721 (0.1607)	GalCer d18:1/26:1(n-7)hydroxy
1.715 (0.1647)	GalCer d18:1/21:1(n-6)hydroxy
1.714 (0.1443)	PE C18:1p/16:0(n-9)
1.550 (0.0678)	GalCer d18:1/20:0
1.497 (0.0501)	GalCer d18:1/22:1(n-7)hydroxy
1.490 (0.0230)	GalCer d18:1/24:2(n-6,n-9)
1.424 (0.0986)	GalCer d18:1/26:1(n-9)hydroxy
1.404 (0.0973)	GalCer d18:1/26:1 hydroxy
1.394 (0.1670)	GalCer d18:1/24:1 (n-3)hydroxy
1.349 (0.2784)	PS18:1/18:1 // (n-6)(35%),(n-9)(64%),(n-12)(1%)
1.297 (0.1463)	PC18:0/20:1(n-3)
1.275 (0.0676)	GalCer d18:1/25:1 (n-9)
1.275 (0.1518)	PC C18:1p/18:1 // (n-3)(46%),(n-9)(54%)
1.262 (0.0708)	PE C16p/22:6(n-3,n-6,n-9,n-12,n-15,n-18)
1.213 (0.0306)	PE C16p/20:1(n-9)
1.192 (0.1226)	PC16:0/16:1(n-12)
1.178 (0.1211)	PC16:0/16:1(n-9)
1.173 (0.2764)	PC18:1(n-8)/20:4(n-3,n-6,n-9,n-12)

Concentration in this table represents the weight in milligrams in 1 g of the total lipid extract. The full list appears in supplemental Table S3.

Figure 3 shows lipid class separation by DMS in the total brain extract, where peak centers of COV scan profile in each lipid species are plotted. Lipid classes were separated roughly by DMS as previously reported in GPLs (25), but other lipid classes are also included. The general trend in the DMS separation was that a precursor with a higher m/z requires higher COV in both sodiated and protonated cases. In some cases, data points with similar m/z were aligned vertically, but a heavier species has a larger COV value. This subtle shift may be caused by the lipids’ minor structural differences caused by double-bond isomerization, (i.e., a higher unsaturated species gave a lower COV). By comparing protonated and sodiated profiles, DMS separation for protonated precursors was better than sodiated precursors for PCs, PEs, SMs, and Cers.

Fig. 3.

MS and DMS separation of brain total lipids protonated precursors using the ammonium acetate solvent (A) and sodiated precursors using the sodium acetate solvent (B).

show the distribution of the various acyl chains present in lipids from different classes, where the regioisomer and double-bond locations were distinguished. This type of display was introduced in our previous publication on TG characterization (20). Besides aligning the acyl chains with the various lipid classes, the color codes on each data point reveal the specific intensities (and therefore distribution profiles) of specific sn-1 and sn-2 chains. In the cases of PCs (Fig. 4A), acyl groups at sn-1 position are less variable than at sn-2 position, i.e., palmitin, stearin, olein, and some plasmalogen groups appeared, but no long chains were observed. In contrast, the sn-2 position of PCs displayed more variety of chains, including long chains. Interestingly, long chains were not present in the sn-2 position of PCs when the sn-1 position was occupied by olein. In the cases of PEs (Fig. 4B), sn-1 characteristics were similar to PC, but plasmalogens were much relevant than PCs. sn-2 had long chains, such as arachidonyl and docosahexaenyl acyls, because such phospholipids are reservoirs of arachidonic fatty acids for eicosanoid production and polyunsaturated fatty acids (28).

Fig. 4.

GPL profiles of brain lipids. A: PCs. B: PEs.

For SLs (Fig. 5), d18:1 was dominant as the backbones as expected, but other lengths of backbone were also observed in Cers and SMs. The distribution of amide chain lengths was shifted to the lighter side than glycerophospholipids, i.e., 18:0 was dominant. Cers and SMs had similar acyl chain profiles (Fig. 5A, B), but the galactosylceramides had a different profile (Fig. 5C), which was very rich hydroxylated species with the longer chains (29, 30), otherwise known as cerebrosides. Phytoceramides were not observed in our sample at this time, although it should be contained in the brain (31).

Fig. 5.

SL profiles of brains. Cer (A), SMs (B), and monoglycosylated Cers (C).

In the case of TGs (Fig. 6), we can identify regioisomerism between at the center port (sn-2) and outer ports (sn-1 and sn-3), but the assignment of acyls at sn-1 or sn-3, i.e., chirality at C₂(glycerol), was not accomplished at this time (32). Although TGs were not rich in the brain extract, some “asymmetric” regioisomer distribution was observed. Fig. 6A shows chain distribution of the sum of outer sn-1 and sn-3 positions over the center sn-2 position. At the sn-2 position, palmitin (P) and olein (O) acyl groups were dominantly observed. The outer sn-1 and sn-3 have a little more variations, including linolein (L), than sn-2. In 3D profile (Fig. 6B, C), OPL (or LPO), OOO, and OOL (or LOO) were observed, but neither OPO nor LOL was observed. Interestingly, only one L at outside was allowed in brain TGs. Such two-dimensional plots in GPLs and SLs and 3D plots in TGs should be useful to display abnormal lipid profile in diseases as well as investigation of adulteration in edible oils, which were demonstrated previously (20).

Fig. 6.

TG profile of brains. A: Inner (sn-2) vs. outer (sn-1+sn-3). B: sn-1 vs. sn-3 (sn-2 = 16:0). C: sn-1 vs. sn-3 [sn-2 = 18:1(n-9)].

Overall, this demonstration suggested experimental tips for selection of the working solvents. Ammonium acetate should be the first choice when PCs, PEs, and SMs are mainly targeting because intense protonated precursors are produced, head group neutral losses are avoided, and DMS separates them well. For analysis of neutral and acidic lipids, sodium acetate should be selected because ammonium acetate does not produce positively charged precursor ions. Because the same method using the same instrument detected many species in PI (liver), PA (egg), and PG (egg) extracts, our method was verified for detecting such acidic GPLs. These results suggested that separated samples in each class are more sensitive than crude samples as expected. Liquid chromatography may be also useful to reduce the complexity of molecular species in ionization as well as the concentration of low abundant lipid species by column.

Iodinated SMs were not observed in the brain total extract, which was reported in our previous publication (19). Ionization efficiency for the weak lipid species may not be good in bulk ionization of crude samples. A small number of PIs and PAs and no PGs were identified in this crude sample analysis. In addition, another limitation of this method is that EIEIO is applicable only on positively charged precursor ions. We applied this technique on negatively charged lipid ions, but we had no evidence that negatively charged ions were dissociated by an electron beam with 0–20 eV range, although resonant electron capture was reported in negatively charged fatty acids (33).

Acyl groups with 18:1(n-7) were recently observed in PC 18:1/16:0 and PC 16:0/18:1 in porcine brain lipids by using OzID and UVPD techniques (12, 15). Although we found this acyl group 18:1(n-7) in other lipids (supplemental Table S3), we did not observe this double-bond isomer in PC 18:1/16:0 and PC 16:0/18:1 during this analysis. In OzID, UVPD, and Paternò-Büchi reaction (7, 13), the double-bond location diagnostic peaks appeared in a vacant m/z region because single bonds do not produce fragments that interfere the diagnostic peak. This is an advantage of these techniques when observing rare double-bond locations. However, EIEIO produced signals from both double and single bonds, so that deconvolution was required to separate multiple contributions from different double bond positions. In the porcine brain case, 18:1(n-9) was dominant, and n-7 type was not recognized as a significant contribution over a few percent of the total isomer composition of those lipids.

The spectrum accumulation time in this report was 1 min or longer because we tried to obtain spectra with peak intensities and signal-to-noise ratios that would provide good statistical reproducibility. Such long accumulation times are amenable for an infusion-based sample delivery to an ion source coupled with DMS. Background chemical noise is removed by DMS, in addition to lipid-class-specific selection. However, for an LC-EIEIO approach, this accumulation time represents a challenge and would need to be shorter, typically less than 1 s. A possible solution in the future would be combining a sensitivity increase technique, Zeno pulsing (34), with a EIEIO-TOF mass spectrometer.

CONCLUSIONS

We have developed a comprehensive structural diagnostic methodology to analyze mixtures of complex lipids (GPLs, SLs, and GLs) using EIEIO MS. Specifically, we have demonstrated a quantitative, comprehensive, structure-based lipidomics analysis method using EIEIO coupled with a preseparation of complex lipid extracts using DMS. This instrument opens a new window for real structural lipidomics (9). The structural lipidome plots by the high-throughput EIEIO-MS may display lipid disorders in many diseases in future medical studies, such as Krabbe’s disease (3), mental disorders, and health contribution of omega-3 fatty acids (4).