Lipids play an integral part in human physiology and exhibit a high degree of specialization in specific cellular compartments, functioning as structural components of membranes, as a medium for energy storage, as an anchor for proteins, as intra- and inter-cellular signaling molecules or as cofactors in modulating protein activity. A multitude of nutritionally and metabolically regulated processes maintain lipid homeostasis in healthy conditions. Defects or alterations in the enzymatic metabolism of lipids may contribute to the pathogenesis of common diseases [1], like Alzheimer’s disease [2], atherosclerosis [3], insulin-resistant diabetes [4], cancer [5], or schizophrenia [6]. Thus, lipid biomarkers have potential applications in the understanding of mechanisms underlying disease pathogenesis, in the prediction of future disease risk and in monitoring the responses to therapies. Lipid researchers refer to the entire collection of chemically distinct lipid species in a cell, an organ, or a biological system as a “lipidome” [7].

With the development of “omics” strategies, the investigators in the lipid field recognized that they should investigate the metabolism of the entire lipidome in a systems biology approach [8]. Lipidomics, a branch of metabolomics, was first placed forward by Han and Gross [9] and is a lipid-targeted metabolomics approach aiming at the comprehensive analysis of lipids in biological systems as well as their interactions with other lipids, proteins and metabolites [10]. Recent technological advancements in mass spectrometry (MS) and chromatography have greatly enhanced the development and applications of metabolic profiling of diverse lipids in complex biological samples, helping to unravel their diversity and to disclose their specificity in biological fluids and tissues and leading to the discovery of new potential biomarkers of pathological disorders [11-13].

[enlarge]

Figure 1. A typical workflow of lipidomic analysis of biological samples.

Cellular lipids are highly complex and show remarkable structural diversities in terms of length of the aliphatic chain, degree of unsaturation, position of double bonds, potential branches, etc; there are tens to hundreds of thousands molecular species [14-16] that the Lipid MAPS consortium has classified into eight major categories: fatty acids (FAs) and acyls, glycerolipids, glycerophospholipids (GPLs), sphingolipids, sterol lipids, prenol lipids, saccharolipids, and polyketides [17]. This complexity and diversity result in a wide variety of physical properties, such as solubility and polarity, and usher in a strong effort in the development of analytical tools that can cover such a diverse range of molecular structures [18]. Integrated platforms, such as Sciex The Lipidyzer™, can quantify over 1100 lipids, and simplify the work, for example, quantitative cholesterol ester and phospholipid analysis for neurons [19].

Targeted analysis of lipids is the lipidomics approach that focuses on a limited number of predefined lipid-specific signals to establish precisely and accurately their relative abundances [20].

Untargeted lipid analysis, in contrast, aims to detect every lipid species present in a sample simultaneously and results in a huge number of compounds to interrogate. It must be coupled with chemometric methods to extract valuable information of relevant signals which are subsequently identified by database searching. It often requires very specialized and expensive software to illustrate the data, and it is usually semiquantitative, due to the impossibility to use internal standards for quantification. The advantage of the untargeted approach is that it has the potential to unravel interesting novel molecules, only limited by the sample preparation method and the analytical techniques used [21]. For this reason, hypothesis-generating studies usually include global lipidomics protocols.

However, specific targeted methods are required for lipid classes present at low concentrations in biological samples or characterized by instability or other physicochemical features that limit the analytical procedure (e.g., bile acids, steroid lipids, specific signaling lipids, and lipids involved in immune system regulation). The main drawbacks of untargeted lipidomics are the complexity of data processing and the necessity to identify the molecules discovered. In contrast, targeted lipidomics benefits of less complexity in data manipulation but prior knowledge of the lipids present in the sample is mandatory.

[enlarge]

Figure 2. Collection of biological samples for lipidomics analysis.

All analytical procedures used in lipidomics have a well-grounded workflow (Figure 1) that can be summarized below:

1. Sample collection
2. Sample storage
3. Sample extraction and analytical calibration
4. Analytical measurements
5. Data processing and identification of lipids
6. Possible applications

The first step in biomedical lipidomics studies involves the collection of biological samples. Tremendous progress has been made in sample preparation strategies, opening new avenues for the application of many different analytical techniques to a variety of biological samples which are subdivided into two categories:

• Classic (conventional): biological samples used in routine analyses - e.g., urine, excrement, blood and its derivatives (plasma/serum)
• Alternative (non-conventional): biological samples used for supplementary analyses - e.g., saliva, tissues, cerebrospinal fluid, tears, nails).

Sample integrity is affected by the temperature of storage, its duration and the number of freeze-thaw cycles (Figure 2).

While immediate analysis of the lipidome would be ideal for minimizing any compositional changes due to lipase activity; in reality, this is not practical; therefore correct storage and sample preservation are imperative. Care must be taken to minimize lipolysis and lipogenesis during storage, by freezing the samples immediately at -20°C or, much better, at -80°C. Studies on human milk have shown that while freezing the milk at -20°C for 3 months resulted in a significant loss of lipids (up to 20%), storage at -70°C or -80°C stops enzyme activity within the samples and sample lipid integrity is best preserved [22, 23].

Sample integrity is also affected by the duration of storage and by the number of freeze-thaw cycles [24].

With regards to blood, some studies reported that the total concentration of lipids in serum is higher than in EDTA or citrate plasma, particularly for phosphatidylcholines (PCs) and triglycerides (TGs) [25, 26]. In plasma, citrate addition lowers the levels of several PCs, phosphoethanolamines (PEs), sphingomyelin (SM) and TGs in comparison to EDTA plasma [27]. If possible, the preprocessing of the blood-based samples (clotting, centrifugation) should be done at 4°C to minimize any alteration of the lipid profiles.

While most studies state that the majority of lipids are relatively stable when kept at 4°C for a limited time (<24 h), several lysophosphatidylcholines (LPCs) in plasma increase when stored at 4°C for 24 h [28] or room temperature for 48 h [29]. Extended storage at room temperature (1–4 days) leads to increased lysophosphatidylethanolamine (LPE) and LPC while correspondingly several PEs and PCs levels decreased and the content of free FAs was also increased [27], indicating the degradation of phospholipids by the membrane enzyme phospholipase A2, which will hydrolyze phospholipids, and release LPCs and FAs [30] or degradation of plasmalogens [31].

However, certain lipid species show more resistance to degradation, as several GPLs and sphingolipids have been shown to be stable during a simulated transport with cold packs for at least 24 h [28] and serum samples stored at 4°C for 1 week have shown no significant changes in the lipid composition [32] in studies that have a limited coverage of the lipids. Thus, it is generally advisable to transport the samples to -80°C as fast as possible. Short storage (<6 months) at -80°C has not shown any effect on the levels of specific lipids [29], while extended storage time of 5 years leads to increased concentrations of LPCs [28].

As most of the samples are frozen upon collection, most of them undergo at least one freeze-thaw cycle before the actual analysis. Several studies have shown that most of the molecular lipids are sufficiently stable at least after one freeze-thaw cycle [29, 32, 33] but repeated freeze-thaw cycles significantly modified their concentrations [25, 26].

Following appropriate sampling and storage for lipid analysis, sample preparation is essential to ensure the accuracy and reproducibility of the results (Table 1). Interferences such as lipids and sugars are removed, and the sample is enriched in the lipids of interest using purification steps such as liquid-liquid extraction (LLE) and/or solid-phase extraction (SPE). Neutral lipids (TGs, waxes, pigments among others) are readily extracted from tissues with ethyl ether, chloroform, benzene or other organic solvents that do not permit lipid clustering driven by hydrophobic interactions. Membrane lipids are more successfully extracted by more polar organic solvents, such as ethanol or methanol, which diminish the hydrophobic interactions among lipid molecules while also weakening the hydrogen bonds and electrostatic interactions that bind membrane lipids to membrane proteins.

For untargeted lipidomics, LLE is the more appropriate method, while SPE gives better extraction efficiency and specificity in targeted lipidomics or class-specific fractionation after liquid extraction.

LLE techniques are used to separate lipids by their relative solubility in different immiscible liquids. Variations of the 1950s methods of Folch [34, 53, 54] and Bligh–Dyer [55], using chloroform, methanol, and water in ratios 8:4:3 and 1:2:0.8 respectively, are the most commonly used techniques. Other than the solvent ratio, the difference in these methods is that Bligh–Dyer uses smaller volumes of solvent and is a less time-consuming protocol [55]. While the Bligh–Dyer extraction was first developed on fish muscle, Folch extraction was developed on brain tissue; however, both of them are easily adaptable to other tissue types. When these solvents are added to the sample, the lipids are dissolved into the organic phase (chloroform) and are separated from the aqueous phase (methanol and water, containing carbohydrates and salts) by a layer of cell debris and protein.

While the use of these methods is well established, the drawbacks include the use of hazardous solvents, such as chloroform, and the collection of chloroform extract from the bottom layer, which may cause the carry-over of water-insoluble impurities or the loss of some lipid-containing phase. In addition, the time necessary to collect the phase manually makes it impossible to automate the process.

These methods have been modified to either replace the use of hazardous solvents, such as chloroform with dichloromethane or increase extraction efficiency with the introduction of centrifugation to enhance phase separation and the omission of water [56-58].

More recently a methyl-tertbutyl ether (MTBE) extraction method similar to the Folch and Bligh and Dyer method, that separates lipids using phase separation has been developed [45] and employed for cell, tissue, and plasma lipid extraction [19, 39]. However, using the MTBE method, the organic phase containing lipids instead forms the top layer, and the aqueous phase forms the lower layer. This method has made extraction of lipids simpler, minimizes the potential of cross-contamination and is more feasible to high throughput and automation. A drawback is that the MTBE phase contains a significant amount of aqueous component that may carry over water-soluble contaminants.

A method that compensates the above techniques with less water-soluble contaminants is the butanol/methanol method (BUME), but its disadvantage is the difficulty in evaporation of the butanol component in the organic phase. Butanol/methanol method can also follow the chloroform extraction. For example, Greenwood DJ et al extracted lipids from cultured macrophages with butanol/methanol (1:1, v/v) containing 5 µM ammonium formate after biphasic partitioning with chloroform, methanol (1:3:3, v/v/v) [47].

None of the extraction methods will allow quantitative recoveries of all lipid species. For example, acidic lipids are efficiently extracted only at acidic pH, but the use of acidic or alkaline conditions is at risk of inducing hydrolysis of specific lipids, resulting in the artificial generation of LPCs [59].

Recently, some other approaches are available for extraction of specific lipids in a sample, such as elevated temperature, microwave, and ultrasound Soxhlet extraction, which offers better recovery [43, 60], supercritical fluid extraction, which separates one component from another using supercritical fluids as the extracting solvent [61, 62], and pressurized fluid extraction, that extracts components near their supercritical region [48]. However, these enhanced extraction methods are at risk of degradation of unstable lipids.

SPE is an extraction method that can reduce degradation and be automated for a concurrent preparation of many samples, as it does not involve partition of the solvent/water mixture and it is rapid and efficient [63]. For example, Chopra S et al extracted lipids from mouse bone marrow–derived dendritic cells wtih a C18 solid phase extraction cartridge (StrataX C18) from Phenomenex for an LC-MS analysis [64]. This technique is suitable for qualitative analyses only; indeed, it has poor reproducibility for the amount and type of lipids absorbed by the cartridge.

Additional optional steps can be added before MS analysis, to achieve sufficient efficiency and unbiased recovery of individual lipids species from the biological samples [65]. The complexity of the extract can be simplified, especially for direct MS applications without previous separation or enrichment through chromatography; physical approaches (e.g., LLE or SPE to separate polar vs. nonpolar lipids) [66] or chemical approaches(e.g., base hydrolysis to enrich low abundance sphingolipids from complex lipid extracts containing high abundance phospholipids and/or glycerolipids) [67, 68] can be used. When lipids lack inherently charged moieties, thus cannot be efficiently ionized, or lack characteristic fragmentation patterns, the extracts can be derivatized by chemically tagging specific functional groups of lipids [69].

The use of quality control is essential to minimize influences, such as sample matrix effects and instrument variations, that could affect method accuracy and reproducibility, but it is often overlooked.

The QC measures, also referred to as Internal Standards (IS), are generally determined by several factors including the target lipid class of the study, the availability, and cost of the standards and researcher preference.

IS can be added in known concentrations, by normalization to total protein, wet/dry tissue weight, or fluid volume for lipid quantitation, to the sample during its preparation, before extraction, to assess the variability that occurs in sample storage and extraction recovery [70], or after extraction, to monitor instrument performance and variability. For optimal results, more than one IS should be used. For example, Chopra S et al added 10 ng each of prostaglandin E₁-d4, resolvin D₁-d5, leukotriene B⁴-d4, 15-HETE-d8, and arachidonic acid-d8, and 100 ng each of cholesteryl heptadecanoate and triheptadecanoyl glycerol to homogenates from 5x10⁶ mouse bone marrow–derived dendritic cells as internal standards for an LC-MS lipidomics analysis [64]. Xie SZ et al chose cellular inorganic phosphate level to normalize the measurements of sphingolipidome [71].

The choice of the internal standards is essential: these compounds should represent the physical properties of the entire lipid class of interest as closely as possible, and they should be absent or present in very minimal amounts in the lipid extracts from biological samples. Labeled compounds that are, or behave as, the compound/s of interest should be used or, in the absence of commercially available adequate labeled standards, unlabeled lipids which are presumed not to be present in the sample should be preferred [70].

Pooled QC or commercial QC should be analyzed periodically within an experiment to monitor instrument efficiency. They should always be matrix-matched to account for biological matrix effects.

The separation-based techniques are sophisticated analytical tools for the analysis of complex biological samples that can be applied in modern lipidomics analysis. There exist a lot of commonly used classical chromatographic methods, such as thin-layer chromatography (TLC), gas chromatography (GC), liquid chromatography (LC), high-performance liquid chromatography (HPLC), supercritical fluid chromatography (SFC), capillary electrophoresis (CE), and also two-dimensional (2D) techniques, such as comprehensive 2D high-performance liquid chromatography (LCxLC or 2D-LC) and comprehensive 2D gas chromatography (GCxGC or 2D-GC) (Table 2).

TLC is a commonly used chromatography method in analytical chemistry because it is a simple, convenient, fast, inexpensive and quantitative method if reliable standards are available [93]. It is carried out by using a silica plate and non-polar solvent for lipid class separation; the classes can then be collected and analyzed using platforms such as GC or LC. A refined TLC version, high-performance thin-layer chromatography (HPTLC), is also an indispensable tool, characterized by increased resolution due to the reduction of the particle size and the thickness of the layer [94]. Among these, 2D-TLC, by separating samples in two orthogonal directions, is a popular method for providing peak capacity and higher resolution compared to TLC [95].

In classical TLC, lipid fractions are visualized by binding to a dye which may preserve or not the lipid structure and be more or less specific.

Exposure of the developed TLC plate to iodine vapors is a very used approach. Iodine forms a brown non-covalent complex with lipids, The drawback of this method is that completely saturated lipids can hardly be stained, while it is difficult to completely remove iodine from highly unsaturated lipids because it binds double bonds [96]. Alternatively, other dyes such as 2,7-dichlorofluorescein, which provides yellow spots if the TLC plate is illuminated with a UV light, or rhodamine, that gives pink spots instead can be used. The dyes mentioned above can be easily removed when the polarity of the solvent is changed, or the lipid (with the bound dye) is passed over a short column [97].

A very popular method is spraying the complete TLC plate with a corrosive reagent and charring the plate to render the lipids visible. Sulfuric acid (50%) in methanol or water [98], or potassium dichromate (5%) in sulfuric acid (40%), or a solution of cupric acetate (3–6%) in phosphoric acid (8–10%) can be used. Saturated and unsaturated lipids require different times to be completely reduced to carbon.

Different reagents also exist that react selectively with a specific lipid species generating colored products. The most popular method for the most sensitive stain is to immerse the TLC plate in a 0.2% amido black 10B in 1M NaCl. This way the lipids can be detected with very high sensitivity [98].

GC is not a very widely used method in lipidomics because it requires complex derivatization before separation and it is time-consuming. Furthermore, most of the lipids due to the reduced volatility cannot be analyzed by GC.

However, it is considered a reliable method for separation, identification, and determination of complex mixtures of fatty acids, as first described in 1952 [99] and in another pioneering work, where it was used to separate fatty acid methyl esters (FAME).

Now the characterization of fatty acids’ composition by esterification to FAME and subsequent determination by GC has become one of the most widely performed analysis in lipidomics laboratories; GC coupled with flame ionization detection (GC-FID) is used for quantification whereas detection with MS has been used for qualitative analysis of FAME [100]. GC-FID is inadequate for identification due to the lack of any selectivity and the misidentification of FAME in the presence of contaminants, artifacts or co-eluting compounds, making it not useful for the analysis of complex extracts and FAME of relatively low abundance [101]. However a possible application of GC-FID is for the separation of regioisomers of FAs, showing good accuracy in the quantification of trans-monounsaturated FAs (MUFAs) and polyunsaturated FAs (PUFAs) in hydrogenated fats [86], however longer columns (up to 100 m) are used for separation of dietary FAME isomers and this leads to extended time for method preparation and run. The advantages of GC- FID compared to GC-MS are that it is considerably cheaper to purchase and maintain compared to MS detectors, it allows the analysis of large numbers of samples before the need for any maintenance and does not have the same requirements and issues as MS (such as ionization source cleaning and ionization issues).

The use of chromatography limits LC throughput as compared with infusion/flow injection-based methods, however, when hyphenated with MS it is the more commonly used technique, together with shotgun MS, because some lipid metabolites have structural isomers that are resolved only by chromatography. For example, prostaglandin D2 and E2 have similar MS/MS spectra, and MS fails to differentiate these mediators, while they are baseline resolved on an LC column. Chopra S et al analyzed the changes in different classes of prostaglandins and other lipids in mouse bone marrow–derived dendritic cells, human monocyte–derived dendritic cells with LPS treatments through an LC-MS analysis [64].

HPLC is a technique still widely used in lipidomics due to its reproducibility, high resolving power, high efficiency, and selectivity. HPLC is also used in connection with secondary chromatographic systems and mass spectrometers for enhanced lipid detection and analysis performance [102]. It is a quite versatile technique as it can be applied for separating many kinds of different lipids, thanks to the number of available columns, and allows quantitative analysis even if its application in a routine analysis is currently often ousted by LC-MS due to the lower costs. It is an easily automated and quantitative method and, since it is isolated from the environment, reduces the contact of the samples with air, preventing sample degradation and autooxidation of lipids [103].

Various HPLC approaches have been used in lipidomics. Normal-phase liquid chromatography (NPLC) was conceived for separation of phospholipids into classes [104], but now it is replaced by hydrophilic interaction chromatography (HILIC, see later). The stationary phase is polar and composed of pure silica, which can be deriva¬tized by polar groups, such as diol, cyano-propyl, amino-propyl or poly-vinyl-alcohol [105], it is impregnated with AgNO3 and is used for separation of groups of compounds having the same number and configuration of double bonds. The mobile phase is typically composed of chloro¬form or hexane and the polarity increased by the addition, in gradient mode, of methanol or 2-propanol. The separation is based on absorption and retention times and the different polarity of the phospholipid parts.

HILIC is the more often used for phospholipid analysis respect to NPLC [106]. It allows the separation of substances highly hydrophilic and amphiphilic, usually poorly retained on the stationary phase in reversed-phase liquid chromatography (RPLC), while too strongly retained in non-aqueous mobile phases in NPLC. The advantages of this technique over NPLC and RPLC are that it is suitable for the analysis of compounds present in complex matrices [107] and does not require the use of extreme pH, preventing hydrolyzation of the analyte.

RPLC is widely used in lipidomics analysis. The stationary phase is hydrophobic and based on silica bonded with alkyl chains, mainly C8 and C18. This technique carries out lipid separation based on lipophilicity, which depends upon the chain length and number of double bonds. The more hydrophobic lipids are more strongly retained on the stationary phase and are eluted later, compared with the more polar lipids. Retention time decreases with shorter chain length and increases with the growing number of double bonds [108, 109]. Better separation reproducibility and electrospray ionization (ESI) efficiency are achieved in RPLC compared with NPLC.

SFC is a high-resolution technique with a detection sensitivity even higher than that of LC and more efficient in separating lipid isomers compared to HPLC [110]. SFC is characterized by better and faster analysis compared to HPLC, due to the higher diffusion coefficients and lower viscosities of supercritical fluids, that are formed when dense compressed gas is subjected to a specific pressure and temperature, compared to regular liquids. CO2 is the most commonly used supercritical solvent, and its non-polar properties make it ideal for separating non-polar lipids like TGs [111]. These features make SFC very appropriate for the simultaneous analysis of different lipids’ classes with a wide range of polarities. There are two kinds of SFC, open tubular column (OT-SFC), which allows high-resolution separation also in very complex samples, and packed column (PC-SFC), faster and with large sample capacity for the analysis of minor components and for preparative isolation [112-114]. Advantages of SFC include no requirement for derivatization, and the ability for SFC to be coupled with all detector types, such as FID or MS, as well as its low cost and waste output relative to LC, the use of less organic solvents than LC, and faster separation and higher resolution than LC and GC in metabolomics analyses [115]. The drawbacks of the technique are the difficult recovery of analytes after depressurization when CO2 turns into a gas and aerosolizes the lipids and the necessity to maintain constant high pressure because supercritical fluids are highly compressible.

However, SFC is well suited to the analysis of multiple lipid classes in samples characterized by a wide range of polarities.

CE is a technique that requires minimal amounts of sample and solvents, and it can easily interface with other techniques such as MS [116]. A typical CE instrument consists of a high‐voltage power supply, fused silica capillary externally coated with polyimide (to allow for flexibility) with an internal diameter ranging from 20 to 200 µm, two buffer reservoirs that house the capillary ends, two electrodes connected to the power supply, and a detector (usually ultraviolet). In CE, the capillary is filled with a suitable buffer, and after injecting analytes from the anode side (under normal polarity conditions), a high voltage is applied at its both ends. The velocity of movement of analytes depends on their charge (positive or negative): the different electrophoretic mobility allows their separation. However, neutral molecules, since they do not bear any charge, move with the solvent front and elute as a single band and thus cannot be separated. To solve this problem, charged surfactants above their critical micelle concentration (CMC), that is the concentration at which the solution properties of the surfactant abruptly change at a given temperature, are added in the CE running buffer, which allows separation of uncharged molecules along with the charged ones. Surfactants in general, comprised of a hydrophobic portion, usually a long alkyl chain, attached to hydrophilic or water-soluble functional groups.

Capillary zone electrophoresis (CZE) has limited application in lipidomics, due to the low UV absorbance and poor aqueous solubility of lipids, that make this technique complete inadequate to resolve extremely hydrophobic lipids]. However it has been used for separation of both unsaturated [117] and saturated fatty acids, based on their differences in charge-to-mass ratios; since UV detections are problematic especially for saturated fatty acids, indirect UV is preferred. Non-aqueous CE (NACE) with indirect UV detection is useful for determination of lipids including GPLs [117-119] and FAs [120]. Also, a hyphenated methodology with spectrophotometric detection can be used successfully [121-123].

CZE has been useful also for the separation of gangliosides, that are difficult for lipidomics analysis because chromophores enabling high selectivity are absent, and gangliosides form micelles in aqueous media; thus an organic solvent is needed. Acetonitrile or cyclodextrins have been used successfully to disperse the micelles in a borate buffer [124].

Capillary electrochromatography (CEC) is a liquid chromatography technique where the liquid is moved through the column by the electro-osmotic flow [125] thus allowing high velocity. This method provides superior analyte selectivity, high efficiency, high resolution, minimal solvent consumption and is ideally suitable for the determination of thermally labile compounds. It is more convenient than GC because GC requires the application of thermally stable columns and derivatization of a sample before analysis. Various stationary phases have been used for CEC, ranging from typical HPLC ones to polyacrylamide gels and silica derived compound also with open-tubular columns with the inner surface coated with different molecules and monolithic columns made by monomers that produce gel [126, 127].

Micellar electrokinetic electrophoresis (MECK) relies upon the differential partitioning of an analyte between a biphasic system (aqueous and micellar); it was initially conceived to separate mainly neutral compounds and is useful for resolving linear saturated C12-C24 free fatty acids mixtures with the addition of neutral cyclodextrins in the place of hydroorganic media. Currently, due to the increased availability of a variety of surfactants it is the most useful among the CE techniques because the separation of the analytes can be obtained due to the difference in electrophoretic mobilities, as well as the differences in solute partitioning in the micelle. When an analyte is injected into the micellar solution, a fraction of it is incorporated into the micelle and migrates at the velocity of the micelle. The remaining fraction of the analyte remains free from the micelle and migrates either with the electroosmotic velocity or with its electrophoretic mobility. The higher the percentage of analyte that is distributed into the micelle, the slower it migrates. MEKC is very useful for separation of GPLs, due to the high hydrophobicity that renders them very soluble in the micellar phase [128].

Microchip CE is based on the marriage of chemical analysis and microfabrication techniques, where micro-channels are fabricated in microchips by micro-molding or photolithography to form channels for sample injection and CE separation. The first report of a microfabricated liquid chromatography system appeared in 1990. CE proved to be an excellent match for microchip technologies because it easily manipulates volumes at the nanoliter scale, requires no moving parts, and provides fast, high-resolution separations. Although the initial primary focus in the field was on DNA analysis [129], microchip CE has been rapidly adapted to many biological, environmental, and industrial applications [130]. Standard analysis protocols for microchip CE involve an injection, separation, and detection. There are two common forms of injection in microchip CE, gated injection and cross injection. In gated injection, a flow boundary is established between two solutions, the mobile phase, and the sample solution, at the intersection of several channels. No mixing of solutions occurs at this intersection because the interaction time is small, minimizing diffusion and the flow regime is laminar. A gated injection is simple but results in the biasing of sample injection toward cations over anions owing to their higher mobility under normal flow conditions. Cross injection is more complicated but does not give a biased injection. In cross injection, the sample solution is flowed across the separation channel using two side channels. After a fixed period, the voltage is switched to direct flow down the separation channel. Microchip CE separations occur in much the same way as conventional CE separations. Several modes of detection have been demonstrated with microchip CE with the most common being laser-induced fluorescence (LIF). LIF is very sensitive for the detection of the small mass quantities present in microchip separations [130].

Separation using a single separation mode is often inadequate when high-abundance lipids mask large numbers of low-abundance lipids of interest. It is possible to apply 2D-LC to characterize complex lipids, since LC techniques can separate lipids according to many independent molecular properties, such as electrostatic forces related to the compound polarity, hydrophobic character of the molecule, size exclusion, ion exchange, etc. [131, 132].

In the 2D-LC approach, lipids can be separated according to three different modes: online, off-line and stop-flow. The online approach is automated, the analysis is faster and reproducible, but it is difficult to use. On the other hand, the off-line 2D-LC transfers the fractions from the first dimension to the second dimension, not in real time. The fractions are collected manually or by a fraction collector, and then re-injected for further analysis, typically with a different column or mobile phase. It is more time-consuming, laborious, difficult to automate, can lead to the formation of artifacts, loss of samples and the analysis are not very reproducible [133, 134]. The online system may be divided into comprehensive and heart-cutting [135]. The comprehensive mode allows a full separation; all peaks from the first dimension are analyzed in the second dimension. Heart-cutting mode resolves components within a selected retention time window. The selected fraction (one or a few peaks from the first dimension) is injected and analyzed in a second dimension. Stop-flow mode is a combination of off-line and online approach. It is based on stopping the mobile phase elution in the column of the first dimension until complete elution of the analytes from the second column [136]. The stop-flow 2D-LC uses a trap column that collects eluent from the first dimension, before the injection into a second dimension. However, this lengthens the time of analysis [137]. Usually, in the 2D-LC approach, the first dimension to fractionate different lipid classes is made by NPLC or HILIC and the second dimension is followed by an analysis of the collected fractions by RPLC. Hence, the time of separation is meaningfully increased when using off-line 2D-LC (about 1–2 hours per single analysis).

The 2D-LC approach allows sensitive and global analysis of lipids, however, its setup is complex and the second dimension suffers from sample dilution effect. The first application of this method for lipidomics was in 2009 when a lab-made non-polar column was applied to the first column and a commercial normal phase column to the second. The two columns were eluted with methanol: chloroform 50:50 (v/v). Since that, this method has been widely applied in research of lipids on different samples [131, 134, 135, 138-140], due to its high resolution and high sensitivity for lipids present in low amounts.

In the last few years, the 2D-GC has been applied successfully in lipidomics’ studies. In this methodology, the detectability of analytes can be increased due to the cryofocusing operated by the modulator, the heart of a 2D-GC system, which traps the compounds eluting from the first dimension through a cold spot, focuses these small portions of eluate and introduces them into a second column for further separation. The main advantage of 2D-GC, where two columns with different polarities are coupled together, is the improvement of the chromatographic resolution due to the increase in the peak capacity. Presently, the multidimensional GC system is becoming more and more widely used for analyzing complex biological samples, like fatty acids’ profiles in many different biological materials, e.g., human serum [141], different cells or other biological materials.

Shotgun lipidomics has become one of the more widely used approaches in lipidomics, particularly for high-throughput analysis of lipids [142, 143]. It uses the direct infusion of the sample through a syringe pump, and higher flow rates guarantee a more stable flow. The pros are that it is efficient and avoids alterations in concentration, chromatographic anomalies, and ion-pairing alterations [144-147]. The cons are the difficulty to achieve automation, risk of clogging of the capillary, the large volume of sample necessary to reach a relatively high flow rate required for stabilizing the ion current [69] and the inherent complexity of shotgun datasets that might consist of several hundreds of MS and MS/MS spectra and comprise > 100,000 of unique peaks, of which only a few hundred is eventually attributed to lipids [148]. The drawbacks cited above are eliminated with robotic nanoflow ion sources (e.g., NanoMate device) which automate the direct infusion [149, 150], thereby guaranteeing high reproducibility, sequential runs of a series of requested mass spectra, and accurate quantitation; unfortunately this devices come at considerable cost and precautions to prevent solvent evaporation of the small volume liquid samples during lengthy automated analysis are necessary [149], i.e. inclusion of less volatile solvents such as isopropanol [151] or sealing of the sample plates with thin aluminum foil [149].

An essential requirement for shotgun lipidomics is the maintenance of a constant concentration of the delivered lipid solution. This condition is mandatory to keep the interactions between lipid species and their contribution to the ion current in an electrospray ionization source constant, thereby leading to a constant ratio of ion peak intensities between lipid species of a class, despite changes of ionization conditions, MS instruments, laboratories [186], provided that the solution is analyzed in the low concentration region to avoid occurrence of lipid aggregation, which is another big concern for lipid quantification [187]. Under a constant concentration condition also ion suppression between each other within a lipid class or between lipid classes is constant. Furthermore this allows the researchers to have unlimited time to set the best experimental conditions (i.e., improving mass spectral signal/noise ratio, mapping with multiple fragmentation techniques including precursor-ion scanning (PIS) and neutral loss scanning (NLS)), to conduct multi-stage MS/MS analyses, and to ramp different instrumental variables). Of course, this is impossible in “on the fly” analysis during chromatographic elution due to the concentration changing and time constraints. Direct infusion of individual fractions collected after LC separation (including those from an SPE column) also falls to the category of shotgun lipidomics due to its maintenance of a constant concentration of the delivered lipid solution.

Another significant feature of shotgun lipidomics is that a full mass spectrum can be acquired to display the molecular ions of all the species of a lipid class of interest. Hence, identification of individual molecular species underlying each ion peak could be done by product-ion mass spectral analysis of the ion peaks or by identifying individual ion peaks in each of the mass spectra through the determination of the building blocks and direct comparison with their internal standards [168]. This can be achieved by PIS of the particular fragment ions and/or NLS of the interested neutrally lost fragments on the individual molecular species of a lipid class or a category of lipid classes leading to their identification and quantitation without time constraints. Then the majority of lipid species that are characterized by a combination of a handful of the identified building blocks is recognized [16]. Additional information can be obtained by changing a series of instrumental conditions [168] such as fragmentation protocol (e.g., collision gas pressure, collision energy, and collision gas) and ionization parameters (e.g., source temperature and spray voltage) [142, 168].

A mass spectrometer is a “molecule smasher” that measures molecular and atomic masses of whole molecules, molecular fragments and atoms by generation and detection of the corresponding gas-phase ions, separated according to their mass-to-charge ratio (m/z), where m is the molecular or atomic mass, and z is the electrostatic charge unit [188]. A Tandem mass spectrometer is a single instrument using two (or more) mass analyzers. The simplest form consists of two quadrupole mass spectrometers (MS/MS) in series connected by a chamber known as “collision cell.” The sample to be examined is essentially sorted and weighed in the first mass spectrometer which separates the mixture of ions allowing only the user-specified certain ions (“precursor” or “parent” ions) to pass to the collision cell. The precursor ions are then bombarded with an inert gas (Xe, Ar, etc.) in the collision cell and are further broken down into different charged and mass ions called “product” or “daughter” ions. The “product” ions are then run through an additional spectrometer further to separate the ions, which is set to monitor specific ion fragments. This process can be repeated several times to get highly particular readings [189].

Because individual lipid species of a polar lipid class possess a common head group, one or more characteristic fragment ion(s) yielded from the head group are usually detected from the class of lipid species after collision-induced dissociation (CID). The key to being successful with this approach indeed is the specificity of the characteristic fragment.

This approach of shotgun lipidomics has been widely employed for lipidomics profiling of biological samples because of its simplicity and robustness [155-159, 190].

Besides the ion source, the mass analyzer in a mass spectrometer is a vital part. There are many kinds of mass analyzers, including the sector magnetic analyzer, the quadrupole (Q) analyzer, the ion trap (IT) analyzer, the time of flight (TOF) analyzer, the Fourier-transform ion cyclotron resonance (FTICR) analyzer and the Orbitrap analyzer. Commercially-available hybrid type mass spectrometers (e.g., quadrupole-time of flight (Q-TOF) or Q-Exactive (i.e., quadrupole-Orbitrap)) possess not only an improved duty cycle that increases the detection sensitivity but also very high mass resolution and mass accuracy [191, 192]. These instruments are able to sensitively acquire full mass spectra of lipid samples of interest and also to rapidly conduct product-ion MS analysis of lipid species in a small mass window step-by-step to determine all the fragments in an entire mass region of interest [151, 193-195], providing accurate measurement of the masses of individual molecular ions as well as fragment ions. This way, the risk of any possible false identification is excluded. Lately, this strategy has been evolved to acquire full product-ion mass spectra of lipid extracts in the negative- and positive-ion modes by using either Q-Exactive or Fusion instruments [196-198] and identifying lipid species by comparing ion intensities of the individual lipid species of interest and of their corresponding internal standard(s).

The high-resolution mass analyzers, including FTICR and Orbitrap, have significantly influenced the research of lipidomics, especially facilitating direct infusion ESI-MS for the simultaneous analysis of multiple lipid classes without the need for prior separation [199-201]. QTOF-MS is usually used for untargeted lipid analysis, like in this work by Choi and colleagues where plasmatic lipid changes were detected after Rosuvastatin treatment helping to elucidate the side effects of the drug [202], but it has also been employed in targeted lipids analysis. For example, Flaherty SE et al evaluated the levels of multiple lipid species in bone marrow-derived macrophages with spiked internal standards using an Agilent 1200 HPLC system coupled with an Applied Biosystem Triple Quadrupole/Ion Trap mass spectrometer (3200Qtrap) [203]. There are several detection modes in the triple quadrupole mass spectrometer, including full scan mode, single ion monitor (SIM), selected reaction monitor (SRM), multi-reaction monitor (MRM), PIS, NLS and daughter ion scan (DIS). Using a single detection mode, the triple quadrupole mass spectrometer can selectively detect one or one kind of lipid with high sensitivity and accuracy [204]. For instance, by the MRM detection mode, LPCs and leukotriene levels have been identified as potential diagnostic markers for colorectal and lung cancer, respectively [205, 206]. Saito T et al performed multiple reaction monitoring of mouse liver samples after liquid chromatography using a Xevo TQ-S micro triple-quadrupole mass spectrometry system (Waters) equipped with an electrospray ionization source and identified a total of 524 individual lipid species [207]. It is foreseen that the development of highly sensitive mass analyzers and their further applications to lipid analysis will lead to the discovery of more functional lipids and consequently to the identification of new biomarkers and therapeutic targets.

EI and CI have been applied to analyze many species of lipids since their introduction in the late 1940s [208] and late 1960s, respectively.

EI is widely used in mass spectrometry, especially for the analysis of gases and volatile organic molecules, in which high energetic electrons interact with gas-phase atoms or molecules to produce ions. EI-MS has also been used in the determination of nonvolatile compounds, like sterol [209], cholesterol [210] and fatty acids [211], but derivation (e.g., esterification for FAs [211, 212] ) is necessary for them. The drawback of EI-MS is that molecular ion signal is usually weak due to the high energy collision. To compensate for this, CI was developed, to generate an easily identifiable intact molecular ion species, in which ions are produced through the collision of the analyte with ions of a reagent gas with lower energy, which are present in the ion source [213].

However, the fact that derivative steps are unpleasant and time-consuming and the sensitivity of this technique is low have restrained its further application in lipid analysis.

FAB has been widely used to identify the structure of nonvolatile lipids, including FAs [214], monoacylglycerols [215], GPLs [216] and sphingolipids [217, 218]. However, FAB-MS is so far not appropriate for the analysis of complex lipid due also to the inconvenience of FAB-MS association.

Utilization of HPLC followed by FAB-MS provided abundant information on the specialized lipid compositions present in diverse cell types, subcellular compartments and many organ systems including the human heart [163]. The use of HPLC followed by direct ionization of intact non-volatile lipids and analysis by mass spectrometry heralded the beginning of a new era in understanding the pleiotropic roles of lipids in cellular function.

ESI is the primary ionization method in MS for lipid analysis from the body fluid, cell, bacteria, virus, and tissue. In the first report for shotgun lipidomics, by Han and Gross in 2003 [9], ESI-MS was used for the direct analysis of lipids without pre-separation by LC.

ESI is one of the softest ionization techniques; it uses an electrospray produced by applying a strong electric field to liquid passing through a capillary tube to create a fine aerosol from which ions are formed by desolvation or CID. In this technique, tuning the pH value (e.g., neutral pH in negative ion detection mode, or adding some specific ionization reagents in solution, like LiOH, in positive ion detection mode) allows selective lipids detection [9, 219]. Therefore, only (quasi)molecular ions of lipids are displayed in the spectrum when ESI-MS is used for lipid analysis and, since the ionization is very soft, some complex dimers and solvent adducts can be detected [144, 220]. Also, in this case, it is important to perform the analysis in a properly low concentration region of lipids to avoid lipid aggregation [187]. ESI-MS offers multiple advantages: 1) its ion source can act as a separation device to selectively ionize a certain category of lipid molecular species based on the charge property of lipid classes with high efficiency without prior chromatographic separation [142, 168, 186] ; 2) the ionization efficiency of lipids in ESI-MS is incomparably higher than other traditional MS ion sources [221] and will continue to improve as the instruments become more sensitive; 3) the instrument response factor of individual molecular species of a polar lipid class detected by a full MS analysis is comparable to experimental errors after 13C de-isotoping, always in case the analysis is performed in a properly low concentration region of lipids. The minimal source fragmentation and selective ionization largely contribute to the same response factor for the species of a polar lipid class [9, 219, 222]. The quantification of individual molecular species of a polar lipid class is possible through direct comparison of ion peak intensities to that of a selected internal standard, or through the peak-area measurement in the case of LC-MS; 4) a nearly linear relationship between an ion peak intensity (or area) of a polar lipid molecular species and the concentration of the compound is present over a wide dynamic range, in the appropriate concentration region of total lipids [149, 150]. 5) modification of the charge properties of the less-ionizable lipid classes can be conducted through derivatization to enhance their ionization [223, 224]. 6) high reproducibility with a direct infusion of a prepared lipid sample in the presence of an internal standard can be readily achieved [150, 219, 225, 226] independent of variables like the laboratory, analyst, and instrument, allowing accuracy and reduction of the required number of samples for replication. Accordingly, due to these advantages, the extraordinary sensitivity, and the advantage that enzyme digestion and derivatization procedures are not necessary for most lipids. ESI-MS-based lipid analysis has become an essential tool for measuring cellular lipidomes [9, 164, 168, 219, 227-229].

Multiple ESI-MS techniques have been established and widely used for the analyses of various classes, subclasses, and individual molecular species of lipids from biological sources [219, 229-231].

• Lipid analysis in the positive-ion mode. In general, lipids are apt to form small cation adducts in the positive-ion mode, due to the soft ionization process and affinity of the cations with the dipole that is present in certain lipid species, of course depending on the availability of the small cations in the matrix. For example, sodium adducts of lipids are often the more abundant ions observed in the mass spectra of ESI-MS in the positive-ion mode [144, 221] because the sodium ion is the most common one in the natural source, and because it possesses high affinity with polar lipids and even non-polar lipids (e.g., TGs). However, lipids can also be adducted with other cations when modifiers like organic acids or ammonium or lithium salts are employed before the sample is infused into ESI-MS [142].
• Lipid analysis in the negative-ion mode. In negative ion ESI-MS spectra, lipid species in the deprotonated form or with an anionic adduct are displayed depending on whether the lipid molecule species carry an ionizable bond, like acidic lipids and thus, are detected as deprotonated ions [144, 219, 232], or they lack any ionizable bond, like polar lipids or lipids belonging to the strong zwitterionic lipid class, all of which can form their anionic adducts with small anion(s) depending on the concentrations present in the matrix and their affinities with these lipid species [146, 221, 233].
• ESI-MS analysis of lipids in the product-ion mode. As said before, ESI full scan mass spectra display the (quasi)molecular ions of lipids. So structural determination of individual molecular species of lipid classes can be achieved by product-ion analysis with low-energy CID detail [227, 228, 234]. Hsu and Turk elucidated the rules for fragmentation of different lipid classes, and a set of studies focusing on the structural characterization of each lipid class was conducted in a few articles [235-238].
  • Product-ion analysis of lipids in the positive-ion mode. Product-ion mass spectra of protonated phospholipid species display little informative ions upon structural characterization, and usually they are straightforward, but they can be integrated through in silica tandem MS-database searches based on the product-ion mass spectra and/or on the product-ion MS spectra of alkaline adducts [235, 239].
  • Product-ion analysis of lipids in the negative-ion mode. Charge-driven fragmentation processes are ubiquitous for lipid classes with ESI-MS in the negative-ion mode; the determinant in the negative-ion mode is the loss of a fatty acyl substituent that results from the gas-phase basicity of the precursor ions [240]. Different lipid classes possess different head groups from which different fragment ions derive due to the different gas-phase basicity. These distinct product-ion mass spectra can be used to distinguish these lipid classes. Moreover, the charge-driven fragmentation processes are also related to the degree of unsaturation in the fatty acyl substituents [221].
• ESI-MS/MS shotgun lipidomics. Many applications of ESI-MS/MS-based shotgun lipidomics were studied to quantitate the lipids in biological samples [155, 164, 241]. First, at least two molecular species of a lipid class of interest should be added as internal standards into a lipid extract during the extraction of a biological sample, largely considering the differential fragmentation patterns of individual lipid species of a lipid class after CID. Next, a unique tandem MS analysis of a building block that is specific to the lipid class of interest, through either NLS or PIS, is performed. The concentration of each molecular species of the class of interest can be calculated from its ion-peak intensity from the acquired tandem MS spectrum by comparing it to the ion-peak intensities of the internal standards. Many experimental factors that may introduce experimental errors are essentially eliminated by utilizing these internal standards.

Since the tandem MS-based shotgun lipidomics is straightforward and has many advantages, this methodology has become very important to analyze many specific lipid classes [242-244] despite the necessity to choose an appropriate internal standard to avoid inaccurate quantification results and inadequate sensitivity of the tandem MS for some lipid classes [219].

The phenomenon of mutual conversion and ion suppression among different lipids may lead to a systemic error when detecting complex lipid extracts by direct analysis of ESI-MS, for example ionization may easily determine the loss of the choline group of LPC that becomes artificial lysophosphatidic acid (LPA) that interferes with the measurement of LPA present in the extract [245, 246].

ESI can be easily hyphenated to many separation techniques to overcome these problems:

• LC coupled to ESI was established specifically for the accurate measurement of LPA [246]. ESI permits the direct analysis of lipids as they are separated by LC, minimizing the ion suppression effect and introducing the retention time in the LC column as an additional parameter for the identification of a compound other than the MS signal.
• 2D-HPLC coupled to ESI was also developed to study the lipid metabolism disorder in many diseases, including obesity, hypertension, diabetes and liver cancer. Xu et al used a novel online stop-flow 2D LC method coupled with QTOF-MS to analyze complex lipids in a plasma sample [137].
• CE [118] and a microfluidic system [247] coupled to ESI represents a highly efficient technique for comprehensive lipidomic research, thanks to the capability of the microfluidic technique to integrate different functions on one single chip, such as the lysis of cells, the capture of lipids and the elution of captured lipids from a solid phase for the microscale purification of lipids.

Recently atmosphere pressure chemical ionization (APCI), atmosphere pressure photoionization (APPI) and desorption electrospray ionization (DESI) were also developed for lipid analysis.

APCI is a soft ionization technique that utilizes gas-phase ion-molecule reactions at atmospheric pressure. Ionization occurs along an electrode where the relative proton affinities of the reactant gas ions and the gaseous analyte molecules allow either proton transfer/abstraction or adduct formation to produce the molecular ions.

In APCI-MS the solvent acts as the CI reagent gas to ionize the samples. APCI-MS can ionize many classes of lipids, and its spectra are relatively simple for analysis [230]. However, the APCI-MS is not soft enough as an ionization technique compared to ESI.

APPI-MS is another useful ionization technique for lipid analysis, including neutral lipids and phospholipids and compounds that ionize poorly by ESI and APCI. This technique uses a vacuum-ultraviolet lamp designed for photoionization detection in gas chromatography as a source of 10-eV photons. The mixture of samples and solvent, after fully evaporated, is introduced into the photoionization region where the photoions react to completion with solvent and analyte molecules because the ion source is at atmospheric pressure and the collision rate is high.

It displays higher sensitivity and lower detection limit in comparison to APCI-MS [248-250]. Compared with ESI, which only uses electrical fields, APCI and APPI are particularly suitable for the analysis of nonpolar lipids, which cannot form charged droplets in solutions [118, 251] and are less susceptible to the effect of ionization suppression and salt buffer effects [247, 252-254]. While ESI is very sensitive for detecting glycerophosphocholines, glycerophosphoethanolamines, acylcarnitines, bile acids, sulfate, etc., APCI is suitable for analyzing cyclic alcohols, fatty acids, and linoleic acids. APPI is proven to be appropriate in detecting steroids, sphingolipids, some amino acids, nucleosides, and purines in plasma [255].

Cooks first introduced DESI in 2005 [256] ; it is an ambient ionization technique that is a combination of ESI and desorption ionization methods. Ionization occurs by pneumatically directing a charged electrospray mist to the sample surface where subsequent splashed droplet scary desorbed, ionized analytes travel into the atmospheric pressure interface of the mass spectrometer

DESI offers greater advantages concerning clinical applications, as it can be performed under ambient conditions with minimal sample preparation, making it suitable for direct tissue analysis [257].

Although ESI/APCI/APPI MS or MS/MS are mighty, it is still a big challenge to identify all the lipids, and there remains a need for simple and reliable methods to determine the double-bond positions with high accuracy and capacity for complex lipids with multi double-bonds [258].

ESI tandem MS (MS/MS) was employed for locating the double-bond position in lipids. After derivatization by ozone [259, 260], pyrrolidines [261], trimethylsilyloxy [262] or dimethyl disulfide, the derivatives yield easily recognizable key fragments, which allow for a determination of the position of the double bond. Recently, methods based on olefin cross-metathesis [258] and charge-remote fragmentation [263, 264] were also proposed for the determination of double-bond positions. Some lipids with multi-phosphate groups, like phosphoinositide, should be derived firstly to improve the sensitivity of detection. Some lipids’ structures, like saccharolipids, are so complicated, that it is still a difficult task to analyze them [265]. The discrimination of isomers of lipids, like cardiolipins, is always a challenge for any MS method. In addition, the reproducibility needs to be considered for quantitative lipids analysis.

In the MDMS-based shotgun lipidomics approach, a two-step procedure has been developed to quantify lipid molecular species [168, 233, 266]. First, ratiometric comparisons to a pre-selected internal standard of the class after 13C de-isotoping are applied for individual molecular species of the class, which are non-overlapping and abundant in the lipid extract. Next, the contents of other low abundance or overlapping lipid molecular species are determined by using all of the already determined molecular species of the class, including the pre-selected internal standard as standards via one or more tandem mass traces (each of which represents a specific building block of the class of interest). This second step is similar to the tandem MS-based shotgun lipidomics approach. The difference is the use of standards to quantify lipid molecular species in the second step, which come endogenously and are not all added before the lipid extraction. The advantage is that the endogenous standards can represent much more comprehensive physical properties of structural similarity. The linear dynamic range of lipid quantitation can be extended by filtering the overlapping molecular species and by eliminating background noise through the quantitation process from this second step [219]. A limitation of this methodology is that the accuracy of the lipid molecular species in the second step is not as good as those quantitated in the first step. Therefore, it is essential to minimize any potential experimental error which can be propagated from the first step to the second step by baseline correction and/or by using an appropriate amount of added internal standard to produce comparable ion peak intensities between the internal standard and the lipid species of a class [267, 268].

MALDI-MS is a soft ionization technique used in mass spectrometry that allows the analysis of large and/or labile molecules (e.g., peptides, proteins, lipids, and polymers) and is particularly useful for MS imaging of tissue or cell samples. This technique involves embedding analytes in a matrix that absorbs energy at the wavelength of the laser. The pulsed laser irradiates the analytes, triggering ablation and desorption of the analytes and matrix material to facilitate the ionization of the analyte molecules in the hot plume of ablated gases.

MALDI-MS has many advantages for lipid analysis, for example, great sensitivity and selectivity and the short time necessary for analysis and sample preparation [269-271]. However, the lipid identification using MALDI-MS has been limited, because it is critical to choose the appropriate matrix to obtain the high quality and high-intensity mass spectral data of the analytes.

Photoreactions, such as trimerizations occurring upon laser irradiation, as well as incomplete matrix cluster decomposition and adduct formation, may generate a multitude of matrix peaks at higher m/z values (100–500 Da), which suppress or obscure the lipid signals with a molecular weight lower than 500 Da. In addition, the lipid extracts from a biological sample are usually a complex system, where the interferences and discriminations of different molecules make it more difficult to analyze. The choice of a matrix is the most important issue for a successful MALDI-MS analysis. Among all of the matrixes, 2,5-dihydroxybenzoic acid (DHB) in acetone is predominantly used as a matrix in lipid studies [272]. In addition, LPA [273], trihydroxyacetophenone [274], p-nitroaniline [274], 9-aminoacridine hemihydrates (9-AA) [275] or ionic liquid matrices [276], 9-AA134 and 1,8-bis(dimethylamino)- naphthalene [277] demonstrated high sensitivity in the analysis of some specific lipids. An inorganic zinc complex ((1,3-bis(bis(pyridine-2-ylmethyl)- amino)propan-2-olato)dizinc(II)) was useful for Sphingosine-1-phosphate detection [278], and metal oxide was used for the analysis of lipid extracts from bacterial and algal sources [279], to avoid the interference of the traditional organic matrix. An aqueous suspension of citrate-capped gold nanoparticles (AuNPs) as a matrix could selectively detect TGs under high PCs conditions [280], showing the feasibility of developing a new matrix for the selective determination of lipids.

Regardless MALDI-MS offers many advantages for the analysis of lipid species such as speed, convenience, high sensitivity, repeatability, and high throughput, limitations are also obvious. 1) the lipid analysis in the low mass-to-charge region is complicated with the general presence of severe matrix background. 2) the presence of multiple adducts and/or ion forms of each lipid molecular species as a common phenomenon not only complicates the analysis of individual molecular species of a lipid mixture for both identification and quantitation but also reduces the sensitivity of the analysis. 3) the analytes are distributed in the sample spot heterogeneously with the majority of matrices, due to the presence of lipid-lipid interactions and lipid aggregation during crystallization. 4) although the post-source decay in MALDI-MS is useful for structural elucidation, it is problematic for quantitation due to the differential fragmentational kinetics. 5) capability of quantification by MALDI-MS still needs to be improved because it is critical to developing a new matrix for the selective determination of lipids with a low content and the stable-isotope labeling of molecules as an internal standard would enable accurate quantification, but it is impossible to synthesize all required stable-isotope labeled compounds [281, 282]. The rather poor reproducibility of MALDI-MS is mainly due to the heterogeneity of the matrix-analyte crystals. A uniform matrix-analyte cocrystal avoids the variability of signal intensity across different locations on the target surface, due to the heterogeneous crystals, and dramatically improves spot-to-spot reproducibility [283]. MALDI-MS can be hyphenated with pre-chromatographic separation techniques like TLC, HPLC, and LC for lipids analysis [284, 285].

Therefore, MALDI-MS should be used more like a qualitative tool to rapidly screen the lipid profile of a biological sample than as a quantitative instrument suitable for lipidomics.

MALDI-IMS allows the direct analysis of tissue slices [286] and has been successfully applied to imaging lipids, peptides, proteins, drugs, and drug metabolites to determine their distribution and relative concentration [287-298]. The type of matrices used for MALDI-IMS is similar to that used for MALDI-MS as described above. For example, 9-AA was very suitable to analyze the phospholipids and sulfatides in rat brain tissue sections [299]. For analysis of PC and SM, it was demonstrated that a mixture of dihydroxyacetophenone, heptafluorobutyric acid, and ammonium sulfate as the matrix could provide better results for rat brain tissue [300]. Solvent-free matrix methods can be used for lipid analysis by MALDI-IMS [301]. Recently, the MALDI-IMS technique has been extended to lipid profiling of single cells [302], cells from lung [303], lung tissues [304], single zooplankter individuals [305], muscle tissues [306, 307], brain tissues [308-310] and even entire bodies [311]. Depending on the matrices and the other reagents used in MALDI-IMS, either sodium adducts [305] or protonated molecular species [305] of PC is displayed abundantly in these spectra of MALDI-IMS. Moreover, many lipid species have been measured from adult mouse brain tissue sections by MALDI-IMS in the negative-ion mode [312]. MALDI-IMS is becoming an established technique and can be another excellent choice for analysis of lipids.

High-resolution MALDI-IMS is an emerging application for the comprehensive and detailed analysis of the spatial distribution of ionized molecules in situ on tissue slides [313]. Furthermore, MALDI-IMS can be even employed in single-cell lipid imaging. Römpp et al combined high spatial resolution, high mass accuracy and high mass resolution MS for imaging a single Hela cell, and numerous compounds, including small metabolites, such as adenine, guanine and cholesterol, as well as different lipid classes, such as phosphatidylcholine, sphingomyelin, diglycerides and triglycerides were imaged in an individual spot of 7 μm in diameter [172].

Recently, ‘ion-mobility’ technology is attracting extensive attention as another separation axis in addition to chromatographic retention time and m/z [176]. Since ion-mobility reflects the shape and the volume of ions, isobaric ions might be resolved depending on their structural difference. Application of this new technology on the separation of isobaric leukotrienes and protectins has been reported [314], suggesting its potentials on the separation of various lipid metabolites. This technique combines an ion mobility apparatus with a mass spectrometer. In the method, samples are ionized by all kinds of ion sources. For example, a radioactive ionization source and the secondary electrospray ionization technique can be used for the vapor sample [315, 316] the ESI source can be used well for the liquid sample [317, 318] and a MALDI source is commonly used for the solid sample [319]. Ionized molecules in a carrier buffer gas under the electric fields of a drift tube were separated based on the different ion mobilities of the ions. Therefore, IM-MS can provide an additional separation and characterization of lipids to the other existing methods for the analysis of the complex lipid mixtures of biological samples. Moreover, any chromatographically co-eluting chemical noise can be resolved by IM-MS, leading to an enhanced signal to noise ratio. IM-MS was also efficient in the separation and identification of isomeric and isobaric species of lipids [317, 318]. It was shown that the head group, the length of the fatty acyl chain, the degree of unsaturation, and the cationization of individual species of phospholipids were important factors leading to the changes in the drift time of phospholipids.

TIMS is an avdancement of IM-MS; it is a gas-phase separation method coupled to quadrupole orthogonal acceleration time-of-flight mass spectrometry. In the TIMS analyzer ions are held stationary in the first segment of the TIMS analyzer, using an axial electric field that counteracts the drag force of a moving gas (while in classic IM-MS the gas is stationary) and a quadropolar field that confines the ions also radially, ensuring higher ion trwansmission and sensitivity. Then ion packages are eluted according to their mobility, by decreasing progressively the magnitude of the axial electric field.

This versatile technique is fast, can be efficienty integrated with MS analyzers, is flexible and adjustable to different analytes and is able to separate species with even small differences in their ion mobiity features [179, 320]. Being a ionization technique ion suppression effects and interfering compounds can affect the analysis and the high sensitivity implies a more laborious data anaysis [179].

Vibrational spectroscopy consists of two complementary techniques: infrared (IR) and Raman Spectroscopy (RS). The first one is based on the phenomenon of absorption, whereas the other is based on non-elastic light scattering. The subject of research for both of them are the transitions between the vibrational levels of a molecule that are associated with the chemical bonds in a sample. The benefits of IR spectroscopy are that it is fast, uses fewer reagents, and is non-destructive and non-invasive and provides a fingerprint of the sample [321, 322].

Each molecule has a unique vibrational spectrum, like a fingerprint; only complete vibrational spectra (infrared and the Raman spectrum) will contain full information about the molecular composition. Both vibrational spectroscopic techniques are, by far, more frequently used only for qualitative analysis, even if efforts have also been made to apply them for quantitative analysis, with difficulties due to the complex chemical composition of biological materials. A RS and Fourier Transform IR (FT-IR) spectrum provides information about all components of the material at the same time, which makes it extremely difficult to assign the band to a specific compound strictly and, indeed, even to a particular group of substances. Moreover, the intensity of the signals depends in a complex way on several factors. RS and FT-IR spectroscopies hold, however, different advantages, including mostly the possibility for a direct measurement of lipids within the cellular structure of interest and relatively quick determination of lipids, as well as detailed information about the structural composition of these lipids, e.g., the degree of unsaturation, length and branching of the chain, etc.

Both spectroscopic techniques allow for the detection of a wide range of compositional/physical structure parameters of lipids and a precise assignment to a particular structure in cell/ tissue. Indeed they make it possible to determine structural factors such as: (1) the cis/trans ratio, calculated by examining cis and trans bands, which differ for their position or the occurrence of both bands in case of comorbidity of the two conformations (2) the degree of unsaturation, useful to characterize the fatty acid composition of adipose tissue [323, 324] and important factor in disease development [325, 326] (3) molar unsaturation (the ratio of C = C bonds per molecule), (4) the content of conjugated double bonds, (5) chain length and (6) branching by calculating the ratio of, for example, the band corresponding to CH3 to the band assigned to CH2 (branching) or the ratio of the band corresponding to CH2 to the bands representing the overall lipid signals (CH2 C CH3). Moreover, they detect changes of substances soluble in lipids [327] or they differentiate between free and esterified cholesterol present in tissues [328].

Quantitative analysis in terms of the spectroscopic method usually requires a reference method providing quantitative results that can be correlated with a spectral assessment. The procedure of a quantitative determination based on Raman or FT-IR spectra involves obtaining spectra for the samples (along with quantitative results from a different method) and building a model of correlation with the use of a proper chemometric tool. For Raman and FT-IR based studies, the most popular tool is Partial Least Square Regression (PLS). The reference method can be, for example, GC-MS.

The third spectroscopic technique is NMR spectroscopy. This method differs from the previously discussed vibrational spectroscopy techniques. The basis of this technique is the excitation of nuclear spins situated in an external magnetic field and the registration of the electromagnetic radiation arising as a result of the phenomena of relaxation (the return of the nuclear spins to the thermodynamic equilibrium). It is a potent technique, which makes it possible to study the molecular structure, dynamics, reaction state, etc. It is the spectroscopic technique most commonly used in biology and medicine due to the number of advantages, like a wide range of applicability and the possibility to obtain metabolic [329], physiological [330] and anatomical data via NMR [331] both in in vivo and in vitro settings. In addition, this technique does not require the use of any ionizing radiation and is non-invasive and non-destructive.

The NMR technique is ideally suited for the study of the structure of naturally occurring lipids and their derivatives [332, 333] and interactions and it is widely used in a wide range of contexts. Here are reported only some examples of the enormous potential of this technique in biological studies. NMR technique has been employed to study lipid membranes, both in terms of lipid structure as well as their role in membrane processes [182-184, 334-338] and interaction with proteins [339, 340] ]. This technique has also been used to characterized hepatic fat in humans [341], the structure of gangliosides [185].

NMR is also applied to the study of interactions between drugs and antibiotics [342-344].

A significant advantage of the NMR spectroscopy is that it allows not only for the determination of the content (and differences in it between samples) but also a possibility to follow the chemical changes of individual components. This can be done with the use of fraction labeling, which can be easily adapted by ¹³C-NMR spectroscopy. As such, it allows for the tracking of fragments of molecules, which were used or incorporated into the newly synthesized molecules. Labeling, combined with a high-resolution dynamic nuclear polarization, which significantly improves the signal-to-noise ratio, has already been used for a detailed investigation of the biosynthesis pathways of membrane lipids [345, 346].

The NMR technique also includes the magnetic resonance spectroscopy (MRS), which is mostly performed and known in the form of magnetic resonance imaging (MRI). It can be used in vivo to detect any changes connected with disturbances in lipid metabolism [347-351]. Despite the great help provided by the NMR technique, the most challenging task remains to link the observed differences in lipidic profiles and metabolic changes leading to certain diseases.

[enlarge]

Figure 3. A typical workflow of bioinformatic data processing in lipidomics.

The study of lipidomics, especially non-targeted lipid analysis, has generated overwhelming amounts of data, which need bioinformatics technology to aid in data processing for acquiring meaningful biology information (Figure 3). Data processing usually includes three parts:

• principal component analysis, clustering methods, discriminant analysis methods or machine-learning based methods to search for differential lipids;
• database retrieval combined with MS/MS spectra for the identification of differential lipids;
• data interpretation for acquiring meaningful biology information.

There exist numerous kinds of data-processing software, for example, Progenesis from Waters, Clinpro from Bruker, etc. Some organizations, like the lipid metabolite and pathways strategy (LIPID MAPS, ), the human metabolome database (HMDB), ChemSpider, etc, which provide free access to their database. In addition, some websites, like www.metaboanalyst.ca, were established for aiding in data processing, free of charge. Under the assistance of analytical software, more of the functional lipidome will be identified, which will greatly enhance the understanding of the mechanisms of disease and push forward the development of lipidomics.

Statistical analysis of lipidomics data is necessary to identify differences between the abundance of lipids under different conditions. The most used tool in the treatment of lipidomic data is the analysis of the hypothesis testing the difference between means of two groups of samples, distribution of the population or the correlation between variables. First, the family of t-tests is used to compare the means between two populations, depending on the assumptions fulfilled by the different populations regarding its distribution and variance equality. Student t-test is applied in case of equal variance, while Welch’s t-test is used for a non-equal variance. Nonparametric t-tests are employed when no assumption about the probability distribution of the data can be accepted [352].

If the groups are more than two, analysis of variance (ANOVA) can be used as a statistical test for the equality of means. Various ANOVA approaches can be applied to the analysis, as one-way ANOVA for a single measured variable and a factor or two-way ANOVA for a single measured variable and more than one factors, allowing extraction of information about significance but also interactions between the factors. The nonparametric alternative to ANOVA is the Kruskal-Wallis test that can be used when there is

no assumption about the probability distribution of the variables [352].

Multivariate ANOVA (MANOVA) can be used when more than a single variable is measured (e.g., multiple lipid concentrations at a given time).

Post-hoc tests like the Bonferroni correction, which has the drawback of being too much conservative and reducing both the false and true positives, or Tukey’s test, Dunnett’s test or Scheffé’s method are required to correct the obtained results to control the incorrect rejection of true null hypothesis, because with a high number of comparisons the number of false positives that can occur by chance increases [353].

Principal component analysis (PCA) is the most used explorative multivariate method. It is a dimension-reducing technique that aims to represent the originally measured data with a small number of “artificial” derived and underlying variables (defined ”principal components”) which retain most of the relevant information from the original data. These principal components are linear combinations of the original variables that hierarchically describe the directions of maximum variation in the data without repeating information which is guaranteed by the orthogonality between them [354].

Mathematically, PCA decomposes the experimental data matrix, D, in the product of two matrices, T and PT, according to the following equation:

D=TP^t+E

where T is the matrix of scores containing information on samples, P^t is the matrix of loadings providing information on the lipid concentrations or spectral/chromatographic channels used as variables and E is the unexplained variation in data matrix D. The most common graphic representations of the PCA analysis are representations of scores and loadings in the space defined by PC1 e PC2; ideally the experimental groups should appear far enough to be completely distinguished, and correlations between two variables can be established considering the cosine of the angle between the vectors from the origin of coordinates to the variable position in the plot (positive correlations show values close to +1 and inverse correlations show values close to -1). It is preferred working with components that explain a high amount of variance as the conclusions extracted from the interpretation of the plots are more reliable.

Cluster analysis is mainly used to organize samples characterized by a set of variables in groups characterized by maximum similarity within each group and high external heterogeneity. The similarity between samples is measured through distance (e.g., Euclidean or Mahalanobis distances) [352].

Clustering analysis can be hierarchical or non-hierarchical. In hierarchical clustering grouping of samples is characterized by a tree structure, a dendrogram, where each level reflects a splitting or joining of groups of neighborhood levels, often showed together with heat maps to provide information about the features causing the clustering of the samples. In non-hierarchical clustering the partition of the samples into a predetermined number of groups is done using an iterative algorithm that optimizes a selected criterion [352].

Discriminant methods try to differentiate the classes by finding the boundaries between them.

LDA analysis reduces the dimensionality of the data by simultaneously minimizing the within-class distance and maximizing the between-class distance achieving maximal class separation. The decision boundaries used to achieve the separation of the classes in the multidimensional space of the variables are linear surfaces (hyperplanes). The direction is known as the discriminant function and is used to set the classification rule. While LDA identifies directions that maximize the separation between the classes, PCA maximizes the direction with maximum explained variance [352]. If multiple classes are considered, LDA assumes that the variable pace can be partitioned into different disjointed regions, on the bases of the highest probability that a sample belongs to a given class compared to the other classes.

PLS-DA is a popular method to classify the samples that overcome the limitations of the LDA method when dealing with a large number of variables.

As a regression method, PLS-DA has two data matrices, X and Y, where X corresponds to the experimental data matrix, and Y corresponds to a vector (two-class discrimination, PLS1-DA models) or matrix (multiple classes, PLS2-DA models) with class information [352]. The purpose of the PLS regression method is to link X and Y using a small number of relevant factors (latent variables) that represent the maximum covariance between X and Y. So the regression constructs a set of loading weights W which give the relationship between X and Y during the regression process [352]. From a PLS-DA model, different information can be extracted providing information similar to those obtained in PCA.

Machine learning methods have also been used to analyze lipidomics data.

RF is a technique proposed by Breiman in 2001 based on classification trees used to predict membership of samples in the classes of a categorical depending variable considering the measured variables. It consists of an ensemble of simple tree predictors (decision trees), each one constructed via a tree classification algorithm and is capable of producing a class membership from a set of independently predicted values with one of the categories present in the dependent variable [352].

SVM uses a set of data to generate a hyperplane with minimization of empirical classification errors and maximization of the geometric margin and allows separation between the classes of samples. It can also be used for nonlinear data associated with kernels.

The LipidXplorer software [148, 355] is designed for shotgun lipidomics profiling of samples of any lipid composition using any mass spectrometry platform. First, it organizes acquired MS and MS/MS spectra into a flat-file database (MasterScan). Then, the molecular fragmentation query language (MFQL) formalizes fragmentation pathways specific for analyzed lipid classes and implements them into customized spectra interpretation methods for probing the MasterScan, identifying lipid species and reporting the abundances of user-defined ions for subsequent lipid quantification [356]. For identifying lipids, LipidXplorer uses de novo spectra interpretation for lipid identification; thus the identification confidence is not compromised if several isobaric species of different lipid classes are present in the sample [356]. The identified lipids are then reported for quantification. LipidXplorer can also identify novel lipids that bear some structural similarity to known lipids [357]. Even if the software supports many features, it lacks ease of use and automation.

LipidSearch is a software tool (made available by Thermo Fisher Scientific) developed jointly by Prof. Ryo Taguchi and MKI (Tokyo, Japan) [358] (https://www.thermofisher.com/order/catalog/product/IQLAAEGABSFAPCMBFK). It is a powerful new tool for automatic identification and relative quantification of cellular lipid species from large amounts of mass spectrometric data obtained from both LC-MS and shotgun lipidomics approaches. For example, Fanning S et al quantified lipids from yeast, rat and human neurons with LipidSearch [359]. It contains over 1.5 million lipid ions and their predicted fragment ions and supports a variety of instruments and acquisition modes, including PIS, NLS, and product ion analysis. First, it identifies lipids based on the polar head groups or fatty acids using a combination of PIS and NLS from lipids mixtures. Then, it discriminates each lipid by matching the predicted fragmentation pattern stored in the database [69].

The identified lipids are quantified by detecting their precursor ions from the full MS scans or integrating extracted ion areas from LC-MS chromatograms. Broader validation is still needed to demonstrate its power for identification and quantification of lipid species [69]. The software package can be purchased from Thermo Fisher Scientific Co.

LipidHome is a database initially populated with lipids generated in silica based on chemical and structural properties and has proposed the standardized nomenclature way to fill the “resolution gap” between existing lipid structure databases and the current identification accuracy of modern MS approaches [360]. Lipids are organized hierarchically, and the database is specifically tailored to handle high throughput mass spectrometry-based approaches.

The LIPID MAPS consortium has developed a suite of structure-drawing tools that greatly increase the efficiency of data entry into lipid structure databases and permit custom structure generation in conjunction with a variety of MS prediction tools. The format that represents lipid structures [17] enables a more consistent, error-free approach to drawing lipid structures and has been used extensively in populating the LIPID MAPS structure database (LMSD), which currently contains over 10 000 molecules [361] (www.lipidmaps.org/tools/index.html).

The concept of lipid homology was recently established to provide a rigorous metric comparable to genetic analyses based on sequences [362]. Recent studies on lipidomes of biological model organisms and clinically relevant human samples emphasize that lipidomes are fundamental to give insight into health status and provide molecular fingerprints similar to expression profiles [363-371]. The “Lipidome jUXtaposition (LUX) score” is a homology metric that can quantify systematic differences in the composition of a lipidome. The structural similarity between all lipids is determined by conversion of all identified lipids into Simplified Molecular Line Entry Specification (SMILES) (http://www.daylight.com/smiles/index.html) [372]. It takes into account all possible isomeric structures that are not distinguishable by the chosen mass spectrometric approach [373] and the limited approaches still available to identify the position of double bonds [259, 374].

LipidBlast is a library containing tandem mass spectra in the product ion mode created in silica, validated to a great extent, and maintained by the Fiehn laboratory at University of California-Davis [375]. This library is freely available for commercial and non-commercial use at http://fiehnlab.ucdavis.edu/projects/LipidBlast/.

First, an in silica library has been created, by importing them from Lipid MAPS or creating new structures with Lipid MAPS tools [361]. Additional compounds were generated by combinatorial chemistry algorithms provided by ChemAxon Reactor11 (JChem v.5.5, 2011; http://www.chemaxon.com/) to cover many bacterial and plant lipids absent in Lipid MAPS. Next, MS/MS spectra were experimentally acquired on different platforms, and structural class-specific fragmentations and rearrangements were interpreted. When pure standards were unavailable, MS/MS spectra from several publications were selected. Characteristic fragmentations and heuristic modeling of ion abundances were also generated for possibly detectable adduct ions of lipid species of individual lipid class. The in silica generated MS/MS mass spectra were rigorously validated. LipidBlast is suitable to analyze MS/MS data from many different types of mass spectrometers.

Multiple programs and/or software packages were developed based on the principles of shotgun lipidomics, including LIMSA [376] available through the website www.helsinki.fi/science/lipids/software.html, that processes data from individual and tandem MS spectra, LipidProfiler [194] and LipidInspector [195] dealing with the multiple PIS and NLS data, AMDMS-SL program, developed to identify and quantify individual lipid species from the data obtained from the MDMS-SL approach [16], whereas ALEX [377] processes the data sets acquired from high mass accuracy/high mass resolution instruments (e.g., Orbitrap).

Besides this, other identification software packages, including LipidQA [378], FAAT [379], lipid [380] and Greazy [381] that provides the solution to estimate the false discovery rate (FDR) for the objective identification of phospholipids by means of a kernel density estimation method to distinguish true and false lipid assignments, can be applied for lipid identification. On the other hand, HMDB [382] has provided the existing or biologically expected lipid structures in human, which can reduce false positive identifications in searching lipid structures by precursor m/z queries.

On the other hand, LipidFrag [383] provides Bayesian classifiers based on the distribution of MetFrag [384] in silica MS/MS assignment’s scores per lipid class to improve the reliability of lipid annotations.

A challenging field of bioinformatics in lipidomics remains the identification of new lipids, and strong efforts are put in predicting molecular structures from unknown MS/MS spectra. Using a machine-learning approach, programs like the CFM-ID program and the CSI:FingerID program [385] allow to yield the predicted MS/MS spectrum from structure inputs. The MS-FINDER program [386] applies theoretical hydrogen rearrangement rules and predicts a reasonable compound structure from compound candidates. In addition to structure elucidation tools, the Metabolic in silica Network Expansion (MINE) database [387] also tackles the identification of new metabolites. The goal of the integration of the computational tools and structures is discovering novel bioactive compounds, and the use of these tools will help researchers to interpret the results obtained by comprehensive lipidome analyses.

Following lipidomics data processing and identification, data analysis usually includes exploration of data as well as of their putative biological meaning. An important application of bioinformatics in metabolomics data mining is to facilitate pathway analysis [14, 376]. Development of bioinformatics and systems biology approaches to link the changes of cellular lipidome to alterations in the biological functions, including the enzymes involved in lipid biosynthesis, remains challenging. Since the lipids of the same class may be in part regulated by the same enzymes, a high degree of within-class co-regulation is to be expected. Correlation network analysis has proved to be a valuable tool for exploring and visualizing co-regulations in metabolomics data [388-390]. Publically available databases such as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and the LipidMAPS databases may help place quantitated lipidomics results into a meaningful biological context. To gain insight into the molecular mechanisms underlying the observed regulatory effects, the clustered lipids need to be mapped into the known pathways. KEGG lipid pathway representation is generally limited to generic lipid classes, while the level of information from MS studies is a specific instance of the subclass. Thereby, it is necessary to convert generic enzymatic and pathway information from KEGG database to a specific instance under study; conversion of KEGG generic names into LIPID MAPS common subclass names and in turn to specific instance names has been used [14] allowing mapping of identified lipids into pathways directly from MS-based studies with other levels of information. However, manual curating of data is inherently time-consuming, requires considerable knowledge of relevant pathways and principles and is precluded in case of extremely large datasets. Alternatively, bioinformatics software platforms such as VANTED and MAVEN [391, 392] may provide assistance with lipid pathway analysis, identifying affected pathways and analytes.

Metabolic syndrome is a severe health condition which is becoming increasingly common, characterized by obesity, hypertension, dyslipidemia, and hyperglycemia that is associated with an increased chance for metabolically related diseases such as cardiovascular diseases, type 2 diabetes, atherosclerosis, stroke, and non-alcoholic fatty liver disease. Lipotoxicity is tightly associated with metabolic syndrome [4, 393-395]. Thereby lipidomics can play a crucial role in mechanistic studies, risk prediction, and therapeutic monitoring for metabolic syndrome-related diseases [13]. Lipidomics of plasma and lipoprotein fractions has provided insights into the complexities of the high-density lipoprotein (HDL) lipidome, unraveled the controversies surrounding HDL-based therapies to attenuate vascular disease [396, 397], and identified phospholipids as a major bioactive component of HDL [396, 398]. Lipidomics has been useful also to characterize the lipid content of the liver of ob/ob mice [399] and to elucidate the circadian changes in lipid abundance of GPLs, TGs, and many other lipid classes in mouse liver dissecting its clock/feeding dependency [400]. The comprehensive and systematic quantitative analysis of multiple lipid classes including oxidized lipids [401] has been fundamental in vascular health and ischemic heart research, allowing to study the pathogenesis, to identify biomarkers and to monitor the therapy outcome.

The brain contains the highest amount of lipids. Naturally, neurological disorders are associated with lipid signaling, metabolism, trafficking, and homeostasis. Lipidomics can be used to investigate these aspects and to develop biomarkers for early diagnosis and prognosis of these disorders. Lipidomics has been employed to study brain complications since its development [402]. Here, examples of neurological diseases in which lipidomics has been successfully applied are reported, with a particular focus on Alzheimer’s disease, a brain pathology where the role of lipid metabolism is especially essential.

AD, defined as a type of degenerative dementia, was first described in 1907 by Alois Alzheimer; it is an irreversible neurodegenerative disorder and is the most common cause of dementia in adults older than 65 years of age; notably, it presents a challenging task for health care in developed countries, since it will dramatically increase in the future, due to increased life expectancy [403-405]. The clinical symptoms of AD comprise a progressive loss of cognitive function [406], typically memory; AD is pathologically distinguished by an extensive loss of synapses and neurons, as well as by the presence of neuritic plaques enriched with amyloid-β (Aβ) peptides in brain [407] and of neurofibrillary tangles composed of hyperphosphorylated tau proteins [408, 409].

• Lipid changes and their correlation with the amyloidogenic pathway

  The Aβ peptide is formed by proteolysis of the precursor β-amyloid pro¬tein (APP), which is a transmembrane protein [410]. Proteolysis of APP can follow two pathways: the amyloidogenic one, pathogenic, or the nonamyloidogenic pathway, which is not pathogenic.

  The amyloidogenic pathway is initiated with the cleavage by β-secretase, also called BACE1, of APP that leads to the release of a soluble peptide, sAPP β. The cleavage by ϒ-secretase of the remaining APP domain anchored in the membrane results in the formation of Aβ peptides [411]. The theory that an imbalance between the production and clearance of Aβ pep¬tide is responsible for triggering senile plaque formation, is known as the “amyloid cascade” [412].

  Unfortunately, currently, available treatment options aim mainly to reduce symptoms, rather than to cure the dis¬ease. Progress in the understanding of the pathophysiological mechanisms has encour¬aged research to find new drugs to alter the course of the disease. Both histological signatures of AD are related to protein disorders. However, in 1907, Alois Alzheimer mentioned the existence of a third lesion named “fat inclusion” or “lipoid granules.” Analytical tools for lipids were absent at the time; thus this interest in lipids remained dormant for many decades until the advent of modern lipidomics, which elucidated changes in brain lipid profiles providing novel insights on the pathogenesis of AD and unveiling potential markers. This is obvious if we think that the brain, which represents the primary site of pathology in AD, is the most lipid-enriched organ in the human body, as said before. The primary lipid compositions of the brain are cholesterol, phospholipids, and sphingolipids, as well as the lipid derivatives such as 4-hydroxy-2-trans-nonenal (HNE). Furthermore, APP, BACE1, and components of the γ-secretase complex are all transmembrane proteins that are expected to be substantially affected by changes in membrane lipid bilayer composition and organization [413]. Cholesteryl esters may play a role in pTau and Aβ peptide pathologies [19].

  Accordingly, alterations in lipid levels are known to modulate the generation of Aβ [19, 405, 414, 415]. The advent of lipidomics leads to the identification of novel lipid biomarkers. Studies show decreased levels of phosphatidylethanolamine plasmalogen and sulfatide (ST), a class of myelin-specific glycosphingolipids, but increases in ceramide, SM and diacyl-glycerol, in the AD-brain or CSF [405, 416-419], as well as few putative lipid signatures from AD-plasma [420]. The loss of plasmalogen content is likely associated with oxidative stress [402], which is common in AD [421], while the increased content of ceramide may be related to inflammation and neuronal cell death, since ceramide mediates many cell stress responses, including cell death [422, 423] and aging [424]. The decrease in ST may be linked to the white matter lesions present in AD [425, 426]. A mechanistic study revealed that ST metabolism is tightly associated with ApoE metabolism, is dependent on its expression levels, turnover rates, and isoforms [417, 427-430] and its metabolic pathway connects the majority of the risk factors for AD [426, 431-437]. Also reported are higher levels of low-density lipoprotein (LDL) cholesterol, which is closely linked with AD [438]. Cholesterol and cholesteryl ester are reduced in the CSF of AD patients [439] despite their elevated levels in the brain tissue and high β- and γ-secretase activities [440, 441], which were confirmed also in AD animal models [442, 443], probably because cholesterol transport from the brain tissue into CSF is impaired in the pathology.

  Reducing the membrane cholesterol level has been shown to inhibit the production and secretion of Aβ by modulation of the enzymatic activities of β- and γ-secretases [444, 445]. Thus, enhanced cholesterol level may promote the enzyme activities of β- and γ-secretases, thereby accelerating the amyloidogenic processing of APP, resulting in the increased accumulation and deposition of Aβ in AD patients, as confirmed by in vitro observations [446].

  Increased levels of 27-hydroxycholesterol (27-OHC), an oxidative product of cholesterol, were shown to increase Aβ level through elevating the protein levels of APP and BACE1 in human neuroblastoma SH-SY5Y cells [447]. Furthermore, 27-OHC was also found to induce Aβ aggregation by increasing ApoE levels, that is known to induce pathological conformation changes in Aβ peptides [448, 449]. Despite a myriad of different studies indicating that accumulation of specific sphingolipid classes contributes to AD pathogenesis by promoting Aβ accumulation, attenuation of the entire sphingolipid biosynthetic pathway leads to increased production of Aβ42 [450], probably due to complex compensatory mechanisms that occur when specific sphingolipid classes are modulated. In this case, a systematic and comprehensive lipidomic analysis may come on stage, encompassing all sub-members of the sphingolipid and glycosphingolipid families in patient postmortem tissues or animal AD models to identify the actual sphingolipid metabolite governing the central mechanisms underlying amyloid plaque formation.

• Lipid changes in hyperphosphorylated Tau-induced AD pathology

  Tau is a cytoplasmic protein that interacts with and stabilizes microtubules. In a pathological state, however, Tau proteins become hyperphosphorylated and detach from microtubules, resulting in the subsequent disintegration of the latter, and undergo alterations in a configuration that promotes the formation of insoluble neurofibrillary tangles (NFT). Lipid compounds were reported to enhance Tau transformation promoting NFT formation, such as cholesterol and its oxidative product 27-OHC [451, 452] and the lipid oxidation product 4-HNE [453] ; accordingly, the levels of 5-lipoxygenase were also found to correlate with Tau pathology [454, 455].

• Lipid changes in relation to neuronal and synaptic loss

  Sulfatides in the nervous system are concentrated in the oligodendrocytes and Schwann cells and wrap around axons representing a structural component of myelin, besides playing a role in immature cells development [456-458]. In AD mouse models [429, 442] and AD patients [417, 440] the level of sulfatides was decreased compared to the control mice, and this effect was found to be mediated by ApoE.

  While the roles and consequences of lipids in AD have been described, detailed profiles of blood lipids with AD, especially those of different lipoproteins, have not yet been thoroughly investigated.

• Lipidomics and the discovery of biomarkers for AD

  Preclinical diagnosis of AD nowadays mainly relies on magnetic resonance imaging (MRI) scan of structural changes in the brain, positron emission tomography (PET) for detection of neutral-related molecular changes, as well as quantitation of the levels of Aβ42 and Tau (absolute amount and phosphorylated amount) in the CSF [459]. Lipidomics has provided a previously unexplored angle of AD pathogenesis by moving the focus away from a single protein- and gene-centric view in the past decade and providing panels of lipid biomarkers and lipid pathways relevant to the disease. Longitudinal studies have also been performed where different species of lipids in the plasma and serum have been found modulated by the pathology, giving insights into the pathogenesis of AD. For instance, decrease total PC [66] and specific PC species [460-462], as well as total SM [461], have been previously observed using various techniques including EI-MS, HPLC-MS, and UPLC-EI-QTOF-MS. Reduced SM levels [66] and elevated ceramide level has also been independently observed by several research groups [66, 463, 464] in the serum and plasma of AD patients indicating that SM and ceramide levels may be good pre-clinical predictors of memory impairment [464].

  Still, a pressing need exists to identify novel biomarkers to delineate the early stage of the disease, which would extend the therapeutic window for treatment; thus considerable efforts have been dedicated to unveiling lipid-related biomarkers in the plasma/serum of AD patients. Also, imaging mass spectrometry will be of great help to unravel the spatial-specific lipid aberrations in the different regions of the brain.

  While the plethora of lipidomic studies have unveiled a myriad of candidate lipid biomarkers for AD diagnosis and treatment intervention, unifying evidence for the disease across the many different studies still lacks. Given the complex nature of AD pathogenesis, a comprehensive method of lipidomic analysis that includes most classes of lipids relevant to neurobiology and provides a clear view of global changes in the lipidome during AD pathogenesis also in incipient AD is imperative to unify the various lipid candidates identified across the different studies, thus providing both drug targets and early biomarkers for diagnosis.

Major depressive disorder (MDD) is a severely debilitating mental disease that leads to a high rate of suicide and economic burden. Accumulating data suggest that low serum total cholesterol levels are related to an increased risk of depression [465, 466], due to decreased neuromembrane cholesterol content and consequent failure of presynaptic serotonin reuptake [467, 468]. Furthermore, an imbalance in the lipid content in the plasma is also associated with MDD [469, 470]. Also the lipid composition in the specific brain areas shows association with the pathology: low levels of Docosahexaenoic acid (DHA), the major FA component of brain essential for brain development [471] were found in post-mortem orbitofrontal cortex of patients with MDD [472] and also the DHA levels in erythrocytes negatively correlate with the scores of depression of MDD patients [473-475].

In conclusion, MDD is associated with peripheral and central deficits long-chain PUFAs, particularly the omega-3 DHA [476, 477] and it represents a pathology where lipidomics analysis can be particularly useful for the discovery of new biomarkers and monitoring of therapy efficacy.

Bipolar disorder (BD) is characterized by transient depressive and manic episodes [478] and is associated with a 5-17 fold increase in suicide rate compared to the general population [479, 480]. Despite strong evidence that BD is associated with neurobiological changes, the molecular mechanisms underlying its pathophysiology remain largely undetermined [481] ; however, there is evidence indicating that arachidonic acid metabolism is altered in BD [478, 482-484]. Arachidonic acid (AA) is one of the most abundant FAs in the brain, present in similar quantities to DHA [485] which derivatives are associated with neuroinflammatory processes. Accordingly, postmortem studies indicate that brain AA metabolism is upregulated in the BD brain in association with neuroinflammation and excitotoxicity [478, 484]. Drugs used for BD treatment downregulated AA turnover and related enzymes in brain phospholipids, suggesting that targeting the AA cascade might be useful for evaluating new medicines for BD [486-489].

A high-throughput mass spectrometry approach found significant changes in the levels of FFAs and PC in the gray and white matters and red blood cells (RBCs) of BD subjects [490]. A 1H-MRS study evidenced increased signal in the orbitofrontal cortex and hippocampus of BD patients for choline-containing compounds suggesting increased membrane turnover and the consequent release of membrane-bound choline compounds [491], as other neurodegenerative disorders such as AD and Huntington's disease [492, 493]. These and other alterations [494] also observed in post-mortem brains [495] suggest that lipid abnormalities may be an intrinsic feature of BD that is reflected by significant changes in the central nervous system as well as in peripheral tissues. Lipidomics will have a pivotal role in future studies of the specific signaling pathways and lipid mediators linking n-3 and n-6 PUFAs to BD pathogenesis, leading to the development of targeted dietary and medication strategies.

Schizophrenia (SCZ) is a devastating neuropsychiatric disorder affecting 1% of the general population, and is characterized by symptoms such as delusions, hallucinations, and blunted affect. It has been frequently shown that in both central and peripheral tissue, SCZ patients demonstrate altered FAs levels, particularly in arachidonic acid [496] and several omega-3 and omega-6 PUFAs are changed in the plasma of SCZ patients [54, 497] Schwarz and colleagues examined lipid alternations in post-mortem brain samples from SCZ and found significant alterations in levels of FFAs and PC in gray and white matters of SCZ samples compared to controls [498]. Ceramides have significantly increased in the white matter of SCZ. Reductions of PC levels have previously been reported for different regions of the SCZ brain [499, 500] and have been linked to an increase in SM turnover, as PC is the choline donor to SM in neurons and oligodendrocytes [500]. Lipidomics analyses can be complicated by the high incidence of metabolic syndrome in SCZ and its induction/worsening by treatment with antipsychotics [501]. However, in the case of plasmalogen, several alterations have been reported both in plasma, platelets, and brain [502-506].

Lipidomics techniques will be fundamental for future studies of plasmalogens to explore their role in the pathogenesis of SCZ and to unravel whether their restoration to normal levels mediates the therapeutic effects of antipsychotic drugs.

Lipids play many key roles in all of the basic processes essential for rapidly proliferating cancer cells, as building blocks of cellular membranes and as energy storage depot in tumor development [507]. In addition, bioactive lipids, such as lysophospholipids and hydrolysis products of phosphatidylinositol and its phosphorylated derivatives, play important roles in signaling, functioning as second messengers and as hormones in cancer cells to promote cell proliferation, survival, and migration [508, 509]. Notably, cancer cells undergo profound changes in lipid metabolism and homeostasis, thus offering new diagnostic and therapeutic opportunities that could be unraveled by lipidomics. Lipidomics on body fluids can be useful for identifying biomarkers and monitoring the efficacy and toxicity of anticancer therapy.

Lipidomics has provided important insights into the stability of tear film and biomarkers for diagnosis, prognosis, and treatment of ocular surface diseases [510, 511] and enabled effective lipid-based therapy for the ocular disease [512].

Lipidomics can be effectively used in nutrition research to follow-up the metabolic responses of specific dietary changes, their relationship with health and disease [513] and their inter-individual variability [514]. In addition, lipidomics can evaluate the dietary intake in a more standardized and precise way for monitoring the acute, medium term, and chronic effects of dietary components [515] and has provided valuable insights into the molecular mechanisms underlying the health benefits of dietary omega-3 PUFAs and the regulatory role of omega-3 and -6 FAs in the inflammatory response [516-518] and the effect of probiotics [519] and cholesterol-lowering food intake [520] on lipidomic profiles.

Lipidomics may play a pivotal role when implemented in pharmacological research, especially in the search for new drug targets and in the long process that accompanies the development of new drugs (molecule screening, evaluation of toxicity, preclinical tests, etc.). For example, a major interest in de novo lipogenesis inhibitors is their proapoptotic effects on cancerous cells. Lipidomics can be used to screen a large variety of candidate anticancer drugs for those that inhibit de novo lipid synthesis [521] and to identify novel drug efficacy biomarkers [522, 523]. Another example is baicalin, a flavonoid compound used for the treatment of idiopathic pulmonary fibrosis (IPF), that has been demonstrated to correct IPF-induced hepatic lipid changes thanks to lipidomic techniques [524].

There is increasing demand for new lipidomic workflows that ensure rapid high throughput screening of the lipid composition of biological samples, without compromising sensitivity, especially when the available sample amount is a limiting factor.

One clear example of this scenario is mouse-based research; in time course studies, you can get only a few microliters of blood from each mouse, in order to keep them alive for the subsequent time points of the experiment. This means a few microliters of plasma, that usually needs to be employed for multiple downstream applications. Thus technologies that employ a reduced amount of plasma are highly advocated, better if in the nanoliter range. The analysis of small cell subpopulations obtained by cell sorting or of small biopsies would benefit enormously of very sensitive techniques, as well.

Vasilopoulou and colleagues recently published a paper describing a workflow ot lipidomics that meets such needs: they conjugated nanoflow chromatography with trapped-ion mobility spectrometry (TIMS), a technique that allows to separate lipids according to the position of double bonds and geometries in a very sensitive manner, and employed a MS scan mode termed PASEF (parallel accumulation serial fragmentation) that increases tha scan rate without losing sensitivity [180].

More in detail, the research group choose a fast extraction method that can be easily automated and scales well to 1 ul of plasma, the MTBE lipid extraction protocol. They injected the extract into a nanofow-LC system coupled to a high-resolution TIMS quadropole TOF mass spectrometer.

Then they separated ions with a dual TIMS analyzer, where ions are positioned in an electrical field by the drag of a gas flow, they are accumulated in the first TIMS analyzer and then separated by ion mobility in the second one in an inverse order, with low mobility-high mass ions released first, followed by high mobility-low mass ones [180].

With PASEF, multiple ions in the ion path produced by the TIMS analyzer, are fragmented in each TIMS ramp by rapidly switching the quadropole mass position within 1 ms to capture as many pecursors as possible. This increases the acquisition rate without loss of sensitivity. Fragment ions maintain the same ion mobility position as their precursors, giving useful information to connect the PASEF MS/MS spectra to their corresponding MS features extracted from the four dimensional (retention time, m/z, ion mobility and intensity) data space. Finally, a software assignes lipids to the spectra.

With this method, a 30 min analysis of 1 µl of human plasma provided 15 fold more MS/MS spectra than without PASEF and exceeded 3 to 4-fold the number of glycerolipids and glycerophospholipids identified by the lipid MAPS consortium [525] and another study on human plasma lipidome [526]. Also the percentage of overlap between the identiifed lipids in the different studies was high. Since they njected 1/20th of the lipids obtained from 1 µl of plasma, the sensititvity of the entire workflow is in the attomol range.The method was tested also with 1 mg of mouse liver and with 0,5 x 106 HeLa cells; overall PASEF increased the number of identified lipids on average 3,6-fold with a very high MS/MS coverage of lipidomics samples. The identified lipid specied covered all major lipid classes with high sensitivity and quantitative accuracy [180].

Another challenge for lipidomics studies is resolving lipid species with similar masses in a narrow mass window. Li and colleagues [320] recently published a protocol that employs HILIC and IM and enables separation of lipids based on headgrooup polarity and gas-phase size and shape. Briefly they used a Folch extraction protocol followed by HILIC-IM-MS. This protocol achieved high lipid resolution in a neuroblastoma cell line enabling to identify the effect of the pharmacological treatment aalysed in the study with high efficiency [320].

Lipidomics has its strengths and weaknesses in the wide variety of existing techniques: on one side, they allow to obtain always more comprehensive analysis of lipids in different conditions and from a variety of biological samples and lead to the identification of an increasing number of new molecules and biomarkers; on the other side, this may be an obstacle to unify the different studies to get a comprehensive view of the global lipidome in a given condition. However, an accurate selection of the cohort of samples may make this weakness into a strength, as the different techniques may allow the detection of different lipid structures providing a clear global view of lipid composition in a sample.

The fields of study where lipidomics may be useful are countless, from biomedicine to environmental, nutritional and toxicological studies. The progress made in the development of techniques with increased sensitivity and higher performances is overcoming the obstacles still present, like for example the necessity of complex and time-consuming derivatization steps, the difficulty to identify some lipid features, such as the number of double bonds, and to detect lower abundance species. Techniques to evaluate lipid turnover kinetics are still missing, thus the challenge to develop more high sensitivity, high throughput, and broad coverage dynamic techniques in the future needs to be taken on. Also, the development of new bioinformatic tools able to integrate lipidomics with other omics is challenging but necessary, as it would significantly enhance the power of lipid analysis.