Lipidome-Wide Characterization of Phosphatidylinositols and Phosphatidylglycerols on C=C Location Level

Abstract

Phosphatidylglycerol (PG) and phosphatidylinositol (PI) are two essential classes of glycerophospholipids (GPs), playing versatile roles such as signalling messengers and lipid-protein interaction ligands in cell. Although a majority of PG and PI molecular species contain unsaturated fatty acyl chain(s), conventional tandem mass spectrometry (MS/MS) methods cannot discern isomers different in carbon-carbon double bond (C=C) locations. In this work, we paired phosphate methylation with acetone Paternò–Büchi (PB) reaction, aiming to provide a solution for sensitive and structurally informative analysis of these two important classes of GPs down to the location of C=C. A liquid chromatography-tandem mass spectrometry (LC-MS/MS) workflow was established. Offline methylated PG or PI mixtures were subjected to hydrophilic interaction chromatographic separation, online acetone PB reaction, and MS/MS via collision-induced dissociation (CID) for C=C location determination in positive ion mode. This method was sensitive, offering limit of identification at 5 nM for both PG and PI standards down to C=C locations. On molecular species level, 49 PI and 31 PG were identified from bovine liver, while 61 PIs were identified from human plasma. This workflow also enabled ratiometric comparisons of C=C location isomers (C18:1 Δ9 vs. Δ11) of a series of PIs from type 2 diabetes (T2D) plasma to that of normal plasma samples. PI 16:0_18:1 and PI 18:0_18:1 were found to exhibit significant changes in C=C isomeric ratios between T2D and normal plasma samples. The above results demonstrate that the developed LC-PB-MS/MS workflow is applicable to different classes of lipids and compatible with other established lipid derivatization methods to achieve comprehensive lipid analysis.

Graphical Abstract

1. INTRODUCTION

Besides being key structural components of lipid membrane, phosphatidylglycerol (PG) and phosphatidylinositol (PI), two classes of glycerophospholipids (GPs), also play important roles in cell signalling, division, and interactions with proteins [1–6]. PG and PI are considered as anionic lipids under physiological conditions due to the presence of negatively charged functional groups, viz. the phosphate in head group and phosphate modification on the inositol ring of PI (can be vaired from one to three). The negative charge is important for these lipids to engage in interactions with positive charge on membrane proteins and play signalling or regulating roles in cell metabolism [7, 8]. PG is generally of low abundance (< 1% mol% of total lipids) in human plasma and mammalian cells [9, 10], while PI accounts for around 5% of total lipids in mammalian cells and 1% in human plasma [9, 11].

PG and PI show large structural diversity in a given lipidome and they often consist of multiple structural isomers, including acyl chain compositional isomers, sn-isomers, carbon-carbon double bond (C=C) location isomers, and double bond cis/trans isomers, which present challenges for confident structural identification [12]. Currently, tandem mass spectrometry (MS/MS) employing collision-induced dissociation (CID) can readily identify GP for fatty acyl composition but the location of C=C within unsaturated acyl chain cannot be determined from commonly used lipidomic workflows [13]. The pitfall hampers understanding of lipid metabolism and function on a molecular level. Recently, alternative MS/MS methods have been developed for locating C=C in complex lipids, such as electron impact excitation of ions from organics (EIEIO) [14], ozone-induced dissociation (OzID) [15], ion/ion reactions [16] and ultraviolet photodissociation (UVPD) [17]. Alternatively, several C=C derivatization methods have shown success when coupled with MS/MS, including the Paternò-Büchi (PB) reaction and epoxidation reaction [18–23]. These derivatization methods have advantages of being MS instrument independent, allowing them to be incorporated onto shotgun [24], liquid chromatography-mass spectrometry (LC-MS) [25], and imaging mass spectrometry [26, 27] for lipid analysis.

The Paternò-Büchi (PB) reaction is a [2+2] cycloaddition reaction between exited carbonyl and C=C in alkene, forming four-membered oxetane ring products [28]. MS/MS of the PB products via CID or UVPD initiates retro-PB reactions and produces fragment ions carrying the initial C=C location information [19, 29]. Sensitive identification of C=C location as well as isomer quantitation, however, relies on the formation of C=C diagnostic ions, which is affected by the ion polarity upon PB-MS/MS. That is, the C=C diagnostic ions are preferably formed from CID of positively charged PB products than those of the opposite polarity. For GPs which are readily to be protonated, i.e. phosphatidylcholine (PC) and phosphatidylethylamine (PE), PB-MS/MS produces abundant C=C diagnostic ions with limit of identification (LOI) on C=C level achieved at 5 nM for PC standards [30]. PB-MS² CID of GPs in negative ion mode, such as deprotonated PE, PS, PG, PI, or anion adduct of PC, cannot produce C=C diagnostic ions directly. By performing MS³ CID of the PB modified fatty acyl anions formed from MS² CID, C=C diagnostic ions can be detected. Such an MS³ approach provides definitive information on C=C location of a specific fatty acyl chain, which is highly useful when the lipid analyte consists of two different unsaturated fatty acyl chains or contains a group of fatty acyl composition isomers or C=C location isomers [19, 31]. One drawback of this approach is that formation of C=C diagnostic ions is largely suppressed in negative ion mode and it becomes negligible for polyunsaturated fatty acyl (PUFA) anions due to competition from facile loss of CO₂ [32]. This aspect makes PB-MS³ CID in negative ion mode not sensitive in analyzing unsaturated PG and PI although they show excellent ionization efficiency as deprotonated ions by ESI. On the other hand, direct PB-MS² CID of protonated PG and PI in positive ion mode was not sensitive either due to their low ionization efficiency in positive ion mode [33]. Given the above reasons, profiling of PG or PI with C=C location information has not been achieved on a lipidome scale.

Methylation of phosphate groups has been demonstrated for enhancing the ionization efficiency of anionic GPs in positive ion mode through neutralization of negative charges (Scheme 1). TrEnDi method uses diazomethane, a toxic gas, to effect fast and efficient methylation, which has been demonstrated for various GPs [34, 35]. Clark et al. reported the use of trimethylsilyldiazomethane (TMSD) as a stable substitute of diazomethane to achieve methylation of polyphosphoinositides (PPI) [36]. Wang et al. reported that TMSD methylation allowed distinguishing isomeric PG and bis(monoacylglycero)phosphate (BMP) from biological samples [37]. TMSD methylation has also been developed to quantify a variety of PPI on fatty acyl composition level and achieved limit of detection at 0.5 nM [38]. These studies suggest that TMSD methylation is promising with regard to improving analysis sensitivity for PG and PI in positive ion mode.

Scheme 1.

Schematics of phosphate head group methylation via TMSD for PG and PI.

In this report, we combined offline TMSD methylation with online acetone PB reaction for sensitive and structural informative analysis down to C=C location. An LC-PB-MS/MS workflow was developed based on this strategy, in which abundant C=C diagnostic ion pairs were obtained, with LOI of 5 nM for both PG and PI standards. The analytical performance of the workflow was demonstrated by profiling unsaturated PG and PI down to fatty acyl composition and C=C location in complex lipid extracts. This method was also applied to analyze alternations of C=C isomeric ratios in human plasma samples due to type 2 diabetes (T2D). Two groups of unsaturated PI species were found to exhibit significant changes in C=C isomeric ratios (Δ9 vs. Δ11 in C18:1) between T2D and normal plasma samples with <20% interpersonal variation.

2. EXPERIMENTAL SECTION

2.1. Nomenclature

Annotation of lipids followed the classification system recommended by LIPIDMAPS (http://www.lipidmaps.org) [39]. For example, 2-(9Z-oleoyl)-1-palmitoyl-sn-glycero-3-phosphoglycerol is denoted as PG 16:0/18:1 (9Z). Briefly, lipid class was represented by the first two letters, viz, PG. Fatty acyl chain information was indicated as number of C-atoms: number of double bonds (16:0 or 18:1). Fatty acyls linked to the glycerol were written as sn-1/sn-2 with known sn-locations, or separated by an underscore without differentiating sn-position. For unsaturated fatty acyls, e.g. 18:1, Δ-nomenclature was used for annotating the position of C=C double bond from the carboxyl end, followed by an indication of C=C geometry (Z or E) if known.

2.2. Materials and Reagents

Detailed information on lipid standards, solvents (LC-grade), and other reagents are provided in the Supporting Information. Pooled normal human plasma with anticoagulant lithium heparin was purchased from Innovative Research Inc. (Novi, MI, USA). Normal human plasma samples and T2D patient’s plasma samples were supplied by the specimen bank of Dongfeng Hospital of Hubei University of Medicine, compliant with relevant ethical regulations set by the Ethical Review Board of Tsinghua University (IRB No. 2017007).

2.3. Lipid extraction

A methyl tert-butyl ether (MTBE) protocol was employed to extract total lipids from plasma [40]. Briefly, 200 μL human plasma, 1.5 mL MeOH, and 5 mL MTBE were mixed. After an addition of 1.25 mL water, the mixture was centrifuged at 10,000 rpm for 10 min for phase separation. The upper phase was collected. The lower phase was re-extracted with 2 mL of upper phase of system MTBE/MeOH/H₂O (10:3:2.5, v/v/v). Finally, the combined organic phases were collected and dried under N₂ flow for further use.

2.4. TMSD Methylation

A modified TMSD methylation method was employed [41]. Briefly, lipid standards or lipid extracts were re-dissolved in 682.5 μL of MTBE/MeOH/HCl (2 M) (500 μL/150 μL/32.5 μL). Phase separation was induced by mixing 250 μL HCl (0.1 M) and centrifugated at 6,500 rpm for 2 min. The upper organic phase was collected and washed with 500 μL of the lower phase of system MTBE/MeOH/HCl (0.01 M) (65:20:15, v/v/v). TMSD (2.5 M, 40 μL) was added to the mixture and the solution was incubated at room temperature for 20 min. The reaction was terminated by adding 6 μL glacial acetic acid. The solution was washed twice with 500 μL of the lower phase of system MTBE/MeOH/H₂O (65:20:15, v/v/v). The upper organic phase was collected and dried in N₂ flow and re-dissolved in 1,000 μL MTBE/MeOH (75:25, v/v).

2.5. HPLC-MS

HPLC-MS system consisted of an ExionLC AC system (Sciex, Toronto, ON, CA), a 4500 QTRAP mass spectrometer (Sciex, Toronto, ON, CA). Separation was performed on Accucore-150-amide-hydrophilic interaction chromatography (HILIC) column (150 × 3 mm I.D.; 2.6 μm) (Sigma-Aldrich, St Louis, MO, USA). Mobile phase A consisted of acetone/acetonitrile (ACN)/isopropanol (IPA) (75/22.5/2.5, v/v/v) and B consisted of 10 mM ammonium acetate aqueous solution. Solutions of lipid standard (2 μL), lipid extract of bovine liver (5 μL), and human plasma (5 μL) were injected for analysed, respectively. The flow rate of the mobile phase was 200 μL min⁻¹. Gradient elution was applied for separation: A started from 97%, kept for 3 min, decreased to 95% at 5 min and kept till 9 min, decreased to 85% at 13 min and kept at 85% and kept till 15 min, decreased to 75% at 23 min and kept till 32 min, then increased back to 97% at 33 min.

MS analysis was performed on a 4500 QTRAP mass spectrometer equipped with an electrospray ionization (ESI) source. Precursor ion scan (PIS), neutral loss scan (NLS), and enhanced product ion (EPI) were used to collect data. Parameter of MS was set as following: ESI voltage (+4,000 V in positive mode or −4,000 V in negative mode), curtain gas (20 psi), interface heater temperature (250 °C), nebulizing gas (GS1, 30 psi), nebulizing gas (GS2, 20 psi), and declustering potential (35 V).

2.6. Flow microreactor for online PB Reaction.

A home-built flow microreactor as described previously [25] was installed between the HPLC column and ionization source (Scheme S1). FEP tubing (0.01-in. i.d., 1/16-in. o.d., Zeus, Orangeburg, SC, USA) was employed as the flow path, which was concentrically coiled around a scaffold with a diameter of 2.5 cm. A low-pressure mercury lamp with emission centred around 254 nm (Model No. 80-1057-01, BHK Inc., CA, USA) was inserted perpendicularly through the centre. The apparatus was wrapped by aluminium foil to prevent leak of UV light. An additional stream of 10 mM ammonium acetate buffer solution was introduced by a syringe pump to the reactor (80 μL/min), merged with the LC stream in a Y-type mixing tee. The solution which underwent the PB reaction typically contained approximately water/acetone/ACN/IPA (30/52/16/2, v/v/v/v).

3. RESULTS AND DISCUSSION

3.1. Coupling methylation of PI and PG with online acetone PB reaction

Phosphate methylation via TMSD was tested using PG 16:0/18:1(9Z) and PI 16:0/18:1(9Z) as model compounds, respectively (Scheme 1). Almost quantitative conversion was achieved within 20 min, consistent with literature reports [42, 43]. Ammonium acetate buffer (10 mM) was added to LC mobile phase to facilitate the formation of ammonium adducts for the detection of methylated PG (PG^Me) and PI (PI^Me) by ESI in positive ion mode. Methylation of phosphate group led to increased hydrophobicity of PG and PI and thus decreased their retention on a HILIC column as shown in the extracted ion chromatograms (EICs) in Fig. 1a and b [44]. The ESI-MS signals were increased by 6.4 ± 0.1 folds for PG 16:0/18:1 and 3.7 ± 0.2 folds for PI 16:0/18:1 after phosphate methylation (Fig. 1c and d). MS² CID of [PG^Me + NH₄]⁺ and [PI^Me + NH₄]⁺ produced dominant fragment peak due to neutral loss of the methylated head group, i.e. 203 Da for PG^Me and 291 Da for PI^Me (Fig. S1). NLS scans were thus developed for profiling of PG and PI after methylation in positive ion mode.

Fig. 1.

EICs of (a) PG 16:0/18:1 and (b) PI 16:0/18:1 before (blue trace) and after (red trace) methylation. MS¹ spectra of (c) PG 16:0/18:1 and (d) PI 16:0/18:1 before (blue trace) and after (red trace) methylation ([PG + NH₄]⁺, m/z 766.5; [PG^Me + NH₄]⁺, m/z 780.5; [PI + NH₄]⁺, m/z 854.5; [PI^Me + NH₄]⁺, m/z 868.5)

The PB reactions of methylated PG and PI standards were performed in a flow microreactor, online connected with LC-MS. Acetone was added into the mobile phase, which also served as the PB reagent. A high proportion of organic solvent (> 90%) was found beneficial to the separation of PG^Me and PI^Me on a HILIC column. However, this solvent condition caused severe side reactions (Norrish Type I cleavage) (Fig. S2) [45]. To obtain both reasonable LC separation and good PB reaction, a tee-union was used to introduce 10 mM ammonium acetate (NH₄Ac) aqueous solution after LC separation right before the solution went through the flow microreactor for the PB reaction. This led to a reaction solvent of acetone/ACN/IPA/H₂O (NH₄Ac, 10 mM) (52/16/2/30, v/v/v/v), by which 20–30% conversion of the PB reaction was obtained for PG^Me 16:0/18:1 or PI^Me 16:0/18:1 in 30 s (Fig. S3). Fig. 2a shows a typical MS¹ spectrum from online PB reaction of PG^Me 16:0/18:1 using such a solvent condition. Notably, there was little interference from side reactions. MS² CID of the PB product (Fig. 2b), [^PBPG^Me + NH₄] (m/z 838.5), generated a pair of C=C diagnostic ions specific to the Δ9 C=C in C18:1; that is, ⁹F_A (m/z 467.5) and ⁹F_O (m/z 493.5). The subscript “A” or “O” in the notation of the fragmention ion indicates an aldehyde or an olefin moiety at the cleavage site in the fragment, while the superscript, 9, denotes the location of the C=C following Δ-nomenclature. The mass difference between F_A and F_O formed from the same C=C location is 26 Da, characteristic for acetone PB-MS/MS. Comparing the C=C diagnostic ions generated from methylated PB products (Fig. 2b) to those without methylation (Fig. 2c), their relative ion abundances were increased by ~3 times after methylation. Similarly, phosphate methylation improved the formation of C=C diagnostic ions for PI 16:0/18:1 (Fig. S4). LOI on C=C location level (S/N >3 for the C=C diagnostics ions) was achieved at 5 nM for PG^Me 16:0/18:1(9Z) and PI^Me 16:0/18:1(9Z) from PB-MS² CID, respectively, which was 10 times lower than those obtained without methylation (Fig. S5). In summary, phosphate methylation improved ionization efficiency in MS¹ and formation of the C=C diagnostic ions in MS² for PI and PG. The collective effect led to enhanced identification of unsaturated PI and PG at C=C location level.

Fig. 2.

(a) MS¹ spectrum of the PB reaction of PG^Me 16:0/18:1 (10 μM) performed using an online flow microreactor under optimized reaction solvent condition. Comparison of MS² CID spectra of the PB products of (b) methylated PG ([^PBPG^Me + NH₄] ⁺, m/z 838.6) and (c) PG without methylation ([^PBPG + NH₄] ⁺, m/z 824.5).

Arachidonic fatty acyl (C20:4) is enriched in PIs in comparison with other classes of GPs in mammalian cells and tissues [46]. We tested if the online acetone PB reaction solvent system was optimal for PI containing PUFAs. Fig. 3a demonstrates the PB reaction MS spectrum of an equimolar mixture (5 μM each) of PI^Me 16:0/18:1 (9Z) and PI^Me 18:0/20:4 (5Z, 8Z, 11Z, 14Z) using a relatively short reaction time, viz. 15 s of UV exposure. Sequential PB reaction products on PUFA (m/z 1034.8) only contributed to a small fraction (< 5%), while the PB conversion of monounsaturated fatty acyl (MUFA) chains (m/z 926.8) was still reasonable (10 – 20 %). Although longer UV exposure (30 s) provided higher PB conversion, it also led to higher degree of sequential PB reactions of PUFA (Fig. S6), which was undesirable due to increased chemical interferences. It is worth noting that PB-MS³ CID of PG and PI in negative ion mode generated low abundances of C=C diagnostic ions, making it difficult for C=C identification [20]. As a comparison, PB-MS² CID of PI^Me 18:0/20:4 in positive ion mode produced diagnostic ions for confident assignment of each C=C (m/z 439/465, 479/505, 519/545, 559/585 in Fig. 3b). Relative abundances of these diagnostic ions were about two times higher than those of PI 18:0/20:4 without methylation (Fig. 3c). Therefore, the above reaction condition was determined suitable for methylated PIs containing different degrees of unsaturation.

Fig. 3.

(a) MS¹ spectrum of the acetone PB reaction of an equimolar mixture (5 μM each) of PI^Me 16:0/18:1 and PI^Me 18:0/20:4 performed using an online flow microreactor under optimized reaction solvent condition. Comparison of MS² CID spectra of the PB products of (b) methylated PI 18:0/20:4 ([^PBPI^Me + NH₄]⁺, m/z 976.8) and (c) PI 18:0/20:4 without methylation ([^PBPI + NH₄]⁺, m/z 962.8).

3.2. Analysis of PG and PI from bovine liver polar extract

Analysis of PG and PI on C=C location level from complex mixtures was investigated by coupling offline methylation with LC-PB-MS/MS. Commercially obtained polar lipid extract of bovine liver was used for demonstration. We divided the lipid extract into two equal aliquots (0.2 mg mL⁻¹, 200 μL per aliquot). One was subjected to TMSD methylation while the other remained as it is. The use of amide-HILIC allowed good separation of different classes of methylated phospholipids with an order of elution: PG^Me, PI^Me, unmodified PC, and PC^Me/ PE^Me (Fig. 4a). PG and PI were observed at much lower content than PC and PE in bovine liver polar extract. Phosphate methylation for positively charged lipids, such as PC and sphingomyelin (SM), was at lower efficiency [43]. Accordingly, intact PC (20.2–22.3 min) and methylated PC (22.6–25.2 min) were both observed after derivatization.

Fig. 4.

Analysis of PGs and PIs in polar lipid extracts from bovine liver. (a) HILIC separation of methylated GPs in bovine liver lipid extracts. (b) Profile of PI^Me resulting from NLS of 291 Da from (RT: 8.4 –10.0 min); (c) Profile of PG^Me resulting from NLS of 203 Da (RT: 3.5 – 4.2 min). (d) LC-MS² CID of PI 35:2 ([PI-H]⁻, m/z 847.6). (e) LC-PB-MS/MS of methylated PI 35:2 ([^PBPI^Me +NH₄]⁺, m/z 880.8) to obtain information on C=C location. (f) LC-MS² CID of PG 34:1 ([PG-H]⁻, m/z 747.6) to obtain fatty acyl composition. (g) LC-PB-MS/MS of methylated PG 34:1 ([^PBPG^Me +NH₄]⁺, m/z 780.6) to obtain C=C location. (h) Rel.% of the Δ9 and Δ11 C=C isomers for PG and PI containing C18:1 acyl chain.

Prior to determining C=C locations, subclass information as well as fatty acyl chain composition of lipids need to be characterized. PIS of m/z 241 (inositol phosphate ion) and PIS of m/z 153 (1,2-cyclic phosphodiester of glycerol ion) in negative ion mode are commonly used scans for profiling of intact PI and PG in mixtures, respectively (Fig. S7a and S7b) [13]. It is worth noting that ions at m/z 153 can be produced from deprotonated phospholipids of different classes; thus, it is non-specific to PG detection. Furthermore, m/z 153 ion is of low abundance in MS² CID of [PG-H]⁻, reducing the sensitivity for PG analysis [47]. Thus, PGs of low abundance were barely detected above the noise threshold from PIS m/z 153 in bovine liver lipid extract. On the other hand, head group specific fragment ions were formed abundantly in positive ion mode after phosphate methylation, viz. neutral loss of 203 Da for PG^Me and neutral loss of 291 Da for PI^Me. Consequently, NLS of 203 Da provided higher confidence in the identification of low-abundance PG species when compared to PIS of m/z 153 (Fig. 4c and S7b). By applying NLS of 291 Da to PI and 203 Da to PG, sensitive and specific profiling of PI and PG from bovine liver lipid extract was thus obtained (Fig. 4b and 4c). Quantitative or semi-quantitative information of PI and PG could be obtained from these NLS profiles with a use of proper internal standard(s) [37, 48]. PI 38:4 and PI 38:3 were the most abundant species, consistent with previous reports on PI profiling of mammalian tissues [49, 50]. PG consisted of a higher content of MUFA chains, with PG 36:2 as the most abundant species .

Methylation of phosphate reduced ionization efficiency of PI and PG in negative ion mode, hindering identification of fatty acyl composition. For instance, equimolar amounts (10 μM) of PG^Me and PG revealed an order of magnitude difference in ionization response in negative ion mode (Fig. S8). Therefore, a separated LC-MS/MS run was performed on lipid mixture which was not subjected to methylation. Different classes of intact phospholipids, such as PG, PI, PE and PC, were well separated on the amide-HILIC column (Fig. S9). Targeted MS² CID on deprotonated ions was conducted on PG and PI according to the molecular information obtained from NLS of the methylated lipids.

PI^Me 35:2 (m/z 880.7) was detected as a low abundance lipid, relative ion abundance of which accountted 2% when normalized to PI^Me 38:4 in the data of NLS 291 Da (Fig. 4b). Deprotonated ions of the corresponding peak, [PI - H]⁻ at m/z 847.6, was selected for beam-type CID (collisional energy = 40 eV). According to the detection of fatty acyl anions at m/z 269 (C17:0) and m/z 279 (C18:2), we deduced PI 35:2 as PI 17:0_18:2 (Fig. 4d). It should be noted that ions specific to the inositol head group were observed at m/z 315, 297, 259, 241, and 223 in the same spectrum [51]. Efforts were not extended to identify the sn-position of acyl chains; therefore, the “_” sign notation was adopted for unknown lipids identified in mixtures.

To determine C=C locations in unsaturated fatty acyls, LC-PB-MS² CID was applied to the methylated lipids. Targeted MS² CID was applied to a list of precursor ions ([^PBPI^Me + NH₄]⁺) corresponding to the PB product of methylated PI. PB-MS² CID of PI^Me 17:0_18:2 (m/z 938.8), generated two pairs of C=C diagnostic ions at m/z 481/507 and 521/547, which suggested the existence of Δ9 and Δ12 C=Cs in the C18:2 chain (Fig. 4e). Combining the class, fatty acyl, and C=C location information together, this PI species was identified as PI 17:0_18:2 (Δ9,12).

Similarly, [PG 34:1-H]⁻ at m/z 747.6 was selected for beam type CID (collisional energy = 40 V). We deduced it as PG 16:0_18:1 according to the detection of fatty acyl anions C16:0 (m/z 255) and C18:1 (m/z 281) in the MS² spectrum (Fig. 4f). Targeted LC-PB-MS² CID was then applied to the ammonium adduct of ^PBPG^Me 16:0_18:1 (m/z 838.6), generating two pairs of PB diagnostic ions (Fig. 4g). Peaks at m/z 467 and 493 indicated Δ9 C=C, while peaks m/z 495 and 521 suggested Δ11 C=C for C18:1 chain. Relative quantification of the two isomers were achieved from calculating the peak area ratio of the corresponding C=C diagnostic ions, viz. (I_467+493 / I_495+521).

Using the workflow described above, we identified 31 PG and 49 PI down to C=C locations in bovine liver polar extract (Table S1). Among them, 92% were unsaturated lipids. For PGs and PIs containing fatty acyl C16:1, both Δ7 (minor) and Δ9 (major) isomers were identified. For C18:2, only C18:2(Δ9, 12) was identified, consistent with previous report which identified PCs and PEs in bovine liver with C=C location information [25]. It has been shown that fatty acyl C18:1 in mammalian lipidome typically consist of the Δ9 C=C location isomer as the major component while the Δ11 isomer as the minor one [15, 17, 20, 25, 52, 53]. Interestingly, we found that the Δ11 isomer was the main isomer of PG 16:0_18:1, with Rel.% of 65%. Fig. 4h compares the Rel.% of C=C Δ9 and Δ11, calculated based on ratios of peak areas of C=C diagnostic ions for PI and PG containing C18:1 acyl chain. Due to lack of synthetic standards for these unsaturated lipids, conversion of the Rel.% of C=C diagnostic ions to mol% of the isomers was not possible. However, Rel.% could be used as a qualitative estimation of the composition of the two isomers. It was evident that the fraction of C18:1 Δ11 isomer in PG was higher than that of PI, implying that fatty acyl C18:1(Δ11) was preferably selected during the synthesis or remodeling processes for PG. The above set of experiments proved that offline methylation coupled with LC-PB-MS/MS was successful for achieving detailed structural analysis of unsaturated PG and PI from complex mixtures.

3.3. Analysis of PI from human plasma samples

PGs are of low abundance in human plasma, usually at a few nanomolar and picomolar [54]. Consequently, our efforts in determining C=C location in PG were not successful. The total content of PI is higher than that of PG, accounting for ~1% (mol%) of plasma lipids [9]. Fig. 5a shows the profile of PI^Me obtained via NLS of 291 Da, leading to detection of 26 PI species of distinct molecular masses. Among those, 21 were unsaturated lipids (Fig. 5b). The number of identification was about two times more than those identified from multi-lab comparisons on human plasma lipidome, although different pooled human plasma samples were used [9, 54]. In addition, all PI species reported in the multi-lab study were detected in our results, with similar distributions; that is, PI 38:4 was the most abundant species (100%), followed by PI 36:2 (41%). PIs were further analyzed for fatty acyl chain composition and C=C location. Four low abundance unsaturated PIs (relative ion abundances < 0.5 %) were identified for fatty acyl composition; however, identification for C=C location was not successful (Fig. 5b). The rest of unsaturated PIs were successfully analyzed for both fatty acyl composition and C=C locations. A totle of 52 unsaturated PI molecular structures were identified in human plasma (Fig. 5b, Table S2). Among them, C18:2 in PIs was identified as C18:2 (Δ9, 12). Fatty acyl C18:1 contained Δ9 (major) and Δ11 (minor) isomer pair (Fig. S10a and 10b). FA 18:3 and cholesterol ester (CE) 18:3 were reported as mixtures of ω−3 (C=C at Δ9, 12, 15) and ω−6 (C=C at Δ6, 9,12) isomers in human plasma, while the ω−3 isomer was found more abundant than ω−6 isomer [52, 53, 55, 56]. Interestingly, we did not observe any ω−3 isomer present in C18:3 in PI 18:0_18:3 (Figure S11a). However, it is possible that minor ω−3 isomer present in PI 18:0_18:3 was below our LOI (5 nM). Similarly, fatty acyl C20:3, C20:4 and C22:4 of PIs in human plasma lipid extract were identified to be ω−6 PUFA, namely, C20:3 (Δ8,11,14), C20:4 (Δ5,8,11,14) and C22:4 (Δ7,10,13,16) (Fig. S11a–S11c), while fatty acyl C22:5 was identified as the ω−3 PUFA, viz. C22:5 (Δ7,10,13,16,19) (Fig. S11d). This type of information is a new addition to the knowledge base of lipid C=C isomers in human plasma.

Fig. 5.

(a) Profile of PIs from human plasma from NLS 291Da of methylated lipids. (b) PIs identified at different structural levels in pooled human plasma. (c) Relative isomer ratio (I_Δ9/I_Δ11) of PI 16:0_18:1 and PI 18:0_18:1 in normal and T2D plasma samples. (d) Relative quantitation (I/I_IS) of PI 16:0_18:1 and PI 18:0_18:1 in normal and T2D plasma samples (PI 16:0/16:0 used as the internal standard). Differences between the two groups were evaluated using the two-tailed student’s t test where *P < 0.05, ** P < 0.01, and *** P < 0.001. Error bars represent ± s.d. (N = 5).

Biomarker discovery in plasma is desirable for disease diagnosis owing to its relatively low invasiveness of sampling. In previous studies from plasma samples of T2D patients, seven isomer pairs (C18:1 Δ9/Δ11 in PCs and PEs) have been observed with significant changes in relative isomeric ratios between T2D and normal plasma samples [25]. Here, relative isomeric ratios of Δ9/Δ11 C=C isomers from PIs containing C18:1 were measured from plasma samples from healthy human subjects (N=5) as controls and T2D patients (N=5). Three technical repeats were acquired for each sample. PI 16:0_18:1 and PI 18:0_18:1 were found to exhibit significant changes in Δ9/Δ11 composition between T2D and control (P = 3 × 10⁻² and P = 6 ×10⁻³, respectively, Fig. 5c). The relative ratio of the isomers (I_Δ9/I_Δ11) in PI 16:0_18:1 increased from 2.2 ± 0.3 (RSD = 14%) in control to 2.8 ± 0.4 (RSD = 16%) ) in T2D plasma samples. The relative isomer ratio of PI 18:0_18:1 decreased from 4.1 ± 0.5 (RSD = 12% ) in control to 3.0 ± 0.5 (RSD = 17%) in T2D plasma samples (Fig. 5c). Individual-to-individual variation was tremendous at lipid subclass level, with RSD of 33% (normal control ) and 70% (T2D) for relative quantitation of PI 18:0_18:1; therefore, no significant changes could be inferred from subclass quantitation (Fig. 5d). As a comparison, individual-to-individual variation was decreased to less than 20% for the ratios of C=C isomers. This result was consistent with the previous findings on PC and PE in T2D plasma samples [25].

4. CONCLUSION

Sensitive and high-throughput analysis of PG and PI down to C=C locations was achieved via pairing phosphate methylation with LC-PB-MS/MS. Methylation of phosphate head group enhanced both ionization efficiency of PG and PI and formation of C=C diagnostic ions in positive ion mode PB-MS/MS, allowing sensitive identification of C=C location of fatty acyls containing one or multiple degrees of unsaturation. We also established a reaction solvent system which allowed reasonable PB reaction conversion, minimum side reaction, and good HILIC separation for different classes of phospholipids. The analytical performance of coupling phosphate methylation with LC-PB-MS/MS was demonstrated by analyzing bovine liver polar lipid extract, where a total of 49 PIs and 31 PGs were identified on C=C locations level. We also discovered that fatty acyl C18:1(Δ11) was enriched in PG as compared to that of PI in bovine liver, although the biology behind it remained to be further examined. In addition, 61 PI molecular species in pooled human plasma, including 52 unsaturated PI were identified on C=C level, a first-time record on structural diversity of PIs in human plasma. By analyzing human plasma samples from T2D patients, we found that the ratios of C=C isomers were much less affected by interpersonal variations (< 20%) as compared to lipid subclass level (30%–70%), corroborating earlier findings that C=C isomer ratios could be used for the discovery of lipid biomarkers.

TMSD methylation was applied to phosphatidic acid (PA) and phosphatidylserine (PS), which also suffered low ionization efficiency via ESI in positive ion mode. Unfortunately, PS underwent fast hydrolysis and was converted to PA using current methylation condition (Scheme S2) [48]. Due to this reason, the workflow developed in this study would not be able to provide accurate identification for PS and PA from complex mixtures. Development of different strategy to improve ionization of PA and PS in positive mode therefore is highly desirable.

HIGHLIGHTS.

• Offline phosphate methylation of PG and PI was paired with online LC-PB-MS/MS, which allowed identification of unsaturated PI and PG on C=C location level at 5 nM.

• High coverage of PI and PG from complex mixtures was achieved by applying this workflow to lipid extracts from bovine liver and human plasma.

• Location isomers (C18 :1 Δ9 vs. Δ11) of PI 16:0_18:1 and PI 18:0_18:1 showed significant changes in plasma samples of T2D patients to those of normal controls.

ABBREVIATIONS:

PG

phosphatidylglycerol

PI

phosphatidylinositol

PC

phosphatidylcholine

PE

phosphatidylethylamine

PS

phosphatidylserine

PA

phosphatidic acid

CE

cholesterol ester

BMP

bis(monoacylglycero)phosphate

GPs

glycerophospholipids

PG^Me

methylated PG

PI^Me

methylated PI

PB

Paternò–Büchi

LC-MS/MS

liquid chromatography-tandem mass spectrometry

HILIC

hydrophilic interaction chromatography

CID

collision-induced dissociation

NLS

neutral loss scan

PIS

precursor ion scan

T2D

type 2 diabetes

LOI

limit of identification

PUFA

polyunsaturated fatty acid

TMSD

trimethylsilyldiazomethane

MTBE

methyl tert-butyl ether

ACN

acetonitrile

IPA

isopropanol