Rapid quantification of fatty acids in plant oils and biological samples by LC-MS

Graphical abstract Analysis of fatty acids (FA) in food and biological samples such as blood is indispensable in modern life sciences. We developed a rapid, sensitive and comprehensive method for the quantification of 41 saturated and unsaturated fatty acids by means of LC-MS. Optimized chromatographic separation of isobaric analytes was carried out on a C8 reversed phase analytical column (100 × 2.1 mm, 2.6 μm core–shell particle) with a total run time of 15 min with back pressure lower than 300 bar. On an old triple quadrupole instrument (3200, AB Sciex), pseudo selected reaction monitoring mode was used for quantification of the poorly fragmenting FA, yielding limits of detection of 5–100 nM. Sample preparation was carried out by removal of phospholipids and triglycerides by solid-phase extraction (non-esterified fatty acids in oils) or saponification in iso-propanol (fatty acyls). This is not only a rapid strategy for quantification of fatty acyls, but allows the direct combination with the LC-MS-based analysis of fatty acid oxidation products (eicosanoids and other oxylipins) from the same sample. The concentrations of fatty acyls determined by means of LC-MS were consistent with those from GC-FID analysis demonstrating the accuracy of the developed method. Moreover, the method shows high precisions with a low intra-day (≤ 10% for almost all fatty acids in plasma and ≤ 15% in oils) and inter-day as well as inter-operator variability (< 20%). The method was successfully applied on human plasma and edible oils. The possibility to quantify non-esterified fatty acids in samples containing an excess of triacylglycerols and phospholipids is a major strength of the described approach allowing to gain new insights in the composition of biological samples. Supplementary Information The online version contains supplementary material available at 10.1007/s00216-021-03525-y.


Introduction
Fatty acids play a fundamental role in the biology of living organisms, e.g., by influencing properties of biomembranes, storing and providing energy, or being involved in cell signaling [1][2][3]. Especially long-chain polyunsaturated fatty acids (PUFA) such as arachidonic acid (ARA), eicosapentaenoic acid (EPA) or docosahexaenoic acid (DHA) are involved in many (patho)physiological processes, i.a., through their oxidation products. These eicosanoids and other oxylipins are highly potent lipid mediators regulating for example inflammation, vasoconstriction or pain [4,5]. Dietary supplementation of n3-PUFAs or reducing the intake of n6-PUFA is a promising way to modulate endogenous fatty acid distribution which has been demonstrated in numerous intervention studies [6][7][8].
The basic structure of fatty acids is a linear hydrocarbon chain with a varying number of double bonds. A large number of structurally similar molecules, e.g., n3-vs. n6-PUFA, leads to challenges in analytics. Fatty acyls are often analyzed by means of gas chromatography coupled to flame ionization detection (GC-FID) or mass spectrometry (GC-MS) due to the high separation efficiency and good sensitivity of GC [9][10][11]. However, transesterification or derivatization is necessary for this analytical procedure, which is on the one hand laborious and time-consuming and on the other hand can lead to discrimination of analytes [12]. Similarly, the need for derivatization makes quantifying non-esterified fatty acids (NEFA) in biological samples by GC challenging due to the high amount of fatty acyls occurring in different lipid classes such as triacylglycerols or phospholipids. Fractionation of FAME C20:4 n3; Ann Arbor, MI, USA). The used edible oils (refined and virgin sunflower oil as well as virgin flaxseed oil) samples were bought in local supermarkets in Wuppertal, Germany. Pooled human EDTA plasma was generated from healthy individuals as described [26] in accordance with the guidelines of the Declaration of Helsinki and approved by the ethics committee of the University of Wuppertal. Acetonitrile (ACN) and ethanol (EtOH) were obtained from VWR (Darmstadt, Germany) and methanol (MeOH), iso-propanol, as well as acetic acid (HAc) from Fisher Scientific (Schwerte, Germany). Ultra-pure water was generated using the Barnstead Genpure Pro system from Thermo Fisher Scientific (Langenselbold, Germany). All other chemicals were from Merck (Darmstadt, Germany).

Calibration and quantification
For calibration, the peak area ratios (analyte/IS) were plotted against the concentration ratio (analyte/IS). Calibration curves were calculated using linear or quadratic least square regression (weighting: 1/x 2 , Table 1). The limit of detection (LOD) was determined by a signal-to-noise ratio of ≥ 3 and the lower limit of quantification (LLOQ) by signal-to-noise ratio of ≥ 5 and accuracy of ± 20% within the calibration curve. For fatty acids which are ubiquitously detectable in blank injection, i.e., injection of EtOH, the LLOQ was set to the concentration yielding a peak height of at least twofold of the peak height in blank injections and accuracy within the calibration curve of ± 20%.

Results and discussion
A sensitive and selective quantification of fatty acids by means of LC-MS was developed: mass spectrometric detection was optimized and a rapid, efficient chromatographic separation was developed. The method performance was characterized and the results were compared with those from GC-FID analysis. Finally, the method was applied to the analysis of fatty acyls as well as NEFA in plasma and in plant oils.  Table 1). Monitoring of collision-induced dissociation (CID) fragment spectra revealed no fragmentation for fatty acids with ≤ 3 double bonds, while for PUFA with ≥ 4 double Table 1 LC-ESI(−)-MS/MS parameters and performance for the quantification of fatty acids. Shown are all fatty acids covered by the method, their mass transitions for quantification in scheduled SRM mode, specific electronic MS parameters (declustering potential (DP), collision energy (CE)), their internal standards (IS), retention time (t R ), full peak width at half maximum (FWHM), the calibration range, the limit of detection (LOD) and the lower limit of quantification (LLOQ). Scheduled selected reaction monitoring mode (SRM) using nitrogen as collision gas (12 psi) with a detection window of ± 35 s around the expected retention time was used for analyte detection

Analyte
Mass transition Electronic parameters Internal standard t R a FWHM b Calibration range LOD LLOQ  (Fig. 1). The lack of detection of fragments of the linear hydrocarbon backbone is consistent with earlier reports [20]. Interestingly, for n6-PUFA, the intensity of the fragment resulted from CO 2 loss was higher compared to n3-PUFA which might be due to the spatial proximity of the double bonds to the carboxyl group in n6-PUFA. This may lead to a higher fragmentation rate through stabilization of the fragment ion charge in the unsaturated carbon chain. Pseudo scheduled selected reaction monitoring mode (pseudo-SRM), i.e., isolating the m/z of [M-H] − ions in Q1 and Q3, was used for quantification of fatty acids which do not show fragment ions in CID. Using the highest collision energy which did not lead to a decrease of the [M-H] − ion in pseudo-SRM mode, co-eluting isobaric matrix is likely to be fragmented which increases the specificity of the detection. PUFA bearing ≥ 4 double bonds were detected in pseudo-SRM and additionally in regular SRM mode by using transition of the loss of CO 2 . However, because pseudo-SRM and decarboxylation are rather unspecific, chromatographic separation is crucial for isomeric fatty acids. The following critical isobaric separation pairs were identified among the biological occurring PUFA: C18:3 n3 (ALA) and n6 (GLA), C20:3 n6 (DGLA) and n9, C20:4 n3 (n3-ARA) and n6 (ARA), as well as C22:5 n3 (n3-DPA) and n6 (n6-DPA).

Optimization of chromatographic separation
Liquid chromatographic separation of low to moderately polar substances, including fatty acids, is commonly achieved using a C18 column [16,20,21,27]. Using a state-of-the-art sub-2 μm particle filled C18 reversed phase column (column 1; Zorbax Eclipse Plus, 150 × 2.1 mm, Agilent, Waldbronn, Germany) and a linear H 2 O/MeOH/ACN gradient (Table 2), retention times of fatty acids were unacceptably long. Even using an optimized gradient with a long isocratic step (11 min) at 100% B, long-chain fatty acids eluted at late retention times (C24:0 20.28 min). Moreover, ALA and GLA were not separated (R = 0.87). Different columns with lower hydrophobicity were tested aiming to increase the selectivity to separate isomers. ALA and GLA were the most critical separation pair; thus, a gradient was chosen yielding an optimal retention factor (k) and an isocratic step at their retention time ( Table 2): In order to adjust the appropriate elution power (percentage B) for the isocratic step, it was calculated from the linear starting gradient at which %B ALA elutes and the isocratic step was set to this calculated percentage B minus 5%. With a biphenyl stationary phase (column 2; Nucleoshell, 150 × 2.0 mm, 2.7 μm particle size (core-shell), Macherey-Nagel, Düren, Germany) sufficient separation of ALA and GLA (R = 0.88) could not be achieved. It seems that the π-π interactions between the isolated double bonds of the fatty acids and the aromatic double bonds of the biphenyl phase do not provide sufficient selectivity. This is supported by the results from a second biphenyl phase (Raptor Biphenyl, 100 × 2.1 mm, 2.7 μm core-shell particle), Restek, Bad Homburg vor der Höhe, Germany). On this column with a considerably lower hydrophobicity, a separation of ALA and GLA was not possible (R = 0.95), despite the shorter length comparable to that of column 2. Modification of the aromatic ring structure to pentafluorophenyl residues (column 3; PFP Kinetex, 100 × 2.1 mm, 2.6 μm core-shell particle) or linkage of a phenyl moiety to the silica gel via an ether bridge (column 4; Synergi Polar-RP, 100 × 2.0 mm, 2.5 μm particle size, both  A baseline separation of ALA and GLA (R > 1.5) could be achieved on a C8 reversed phase column with dimensions of 100 × 2.1 mm, 2.6 μm core-shell particle (Kinetex, Phenomenex, Aschaffenburg, Germany). Increasing the hydrophobicity by ethyl-bridged hybrid silica (column 5, Triart, 100 × 2.0 mm, 1.9 μm particle size, YMC, Dinslaken, Germany) failed to further improve the separation (R ALA/ GLA = 1.25).
With the optimized chromatographic conditions on the Kinetex C8 column, 41 fatty acids and 11 internal standards could be separated within 13.5 min (Fig. 2, ESM Fig. S2). The saturated fatty acids eluted over the entire run time, while the retention times for unsaturated fatty acids depended on the number of double bonds (Fig. 2). Hu et al. and Bromke et al. also described a pronounced relationship between retention time, number of carbon atoms, and number of double bonds [20,21]. C24:0 eluted last with a retention time of 12.96 min. In order to remove potentially retained non-polar matrix the isocratic step at 100% B was held for one void volume (0.24 ml, 0.8 min). Including re-equilibration the final run time was 15 min with highly stable retention times showing a variation (relative standard deviation, RSD) of < 0.20% or < 0.02 min for intra-batch (n = 24) and < 0.75% or < 0.07 min for inter-batch (three batches, n = 30; Table 1).
Regarding the start conditions of the gradient, it turned out that a pre-concentration step [28] with low elution power (20% B) is required for a good separation of early eluting fatty acids. The lipophilic nature of fatty acids makes the use of the more non-polar injections solvent EtOH necessary; otherwise, long-chain fatty acids are discriminated (ESM Fig. S3). Without the pre-concentration step, the strong elution power of the injection solvent deteriorates the peak shape of the early eluting analytes (Fig. 3). With a retention factor k > 1, the analytes are well separated from void volume (k = 2.36 for C6; Table 2) using 20% B for the initial step. ALA and GLA are separated by a long isocratic step at 66% B (6.5 min, R = 1.88), other critical separation pairs such as n3-and n6-DPA were separated within a linear solvent gradient. Interestingly, while C20:3 n9 and n6 were easily separable (R = 2.89), C20:3 n6 and n3 as well as C18:1 n9 and n7 could not be separated on any of the tested columns. It is consistent with literature that RP-LC does not allow to separate C20:3 n6 and n3 [16,22]. However, in our experience, C20:3 n3 and C18:1 n7 do not occur or only occur at low concentrations in biological samples [7,10,29]. Thus, the quantification of C20:3 n6 and C18:1 n9 in sum with the  respective isomer does not seem problematic for a correct determination of the quantitative fatty acid pattern in cells, blood, tissues and the most edible oils. The method described herein is superior compared to previous LC-MS approaches. It requires no derivatization as used by several other groups [14,15,23,24], which shortens sample preparation considerably. The total run time of 15 min allows rapid analysis. Even though a run time of around 15 min is also achieved in other methods, these methods quantify a considerably lower number of analytes (23 fatty acids [25], 14 fatty acids [14], 30 fatty acids [22]).

Sensitivity
The limit of detection (LOD) and lower limit of quantification (LLOQ) was determined according to the Guideline on Bioanalytical Method Validation of the European Medicines Agency (EMA) [30]. The LOD was set to the lowest injected standard yielding a signal-to-noise ratio (S/N) of ≥ 3 and the LLOQ was set to the lowest standard yielding a signal-tonoise ratio of ≥ 5 and an accuracy of 80-120% within the calibration curve. The LODs of the fatty acids detected by pseudo-SRM ranged mainly between 5 and 25 nM (0.01-0.08 ng on column; Table 1), whereas the LODs for the transition based on decarboxylation were higher (25-50 nM; 0.08-0.2 ng on column) due to the low intensity of the [M-H-44] − fragment. Therefore, fatty acids bearing ≥ 4 double bounds were also quantified in pseudo-SRM mode. The sensitivity is consistent with earlier reported LODs, e.g., 0.02-0.1 μM (0.05-0.32 ng on column) [25] and 0.03-0.3 μM (0.1-1 ng on column) [22]. It should be noted that we used a 20-year-old middle class mass spectrometer. With state-ofthe-art high-resolution MS instruments such as LTQ Orbitrap Elite [16] or highly sensitive QqQ MS QTRAP 5500 [20], LODs of 1-2 orders of magnitude lower can be achieved.
The group of Hu et al. found a low LOD also for C16:0 and C18:0 (0.05 ng/ml; 0.20 nM and 0.18 nM, respectively) [20]. In our hands, blank injections and even LC-MS measurements without injection also showed peaks for C16:0 and C18:0 (ESM Figs. S4 and S5) which could not be completely reduced by using pure solvents, glassware instead of plastic and methanol as well as iso-propanol for washing the injection system between runs. High background signals of these fatty acids-used ubiquitously in consumer products such as plastic ware-were also described by other groups [18,23]. For these analytes, we set the LLOQ to the concentration yielding a peak height of at least twofold of the peak height in blank injections and an accuracy within the calibration curve of 80-120% (Table 1) which was 0.5 μM for C16:0 and 1 μM for C18:0. The ULOQ was set to 20 μM. By using a deuterated internal standard for each of these compounds (C16:0-d4 and C18:0-d5) ion suppression occurring at this high concentration could be compensated allowing a linear regression. (ESM Figs. S4 and S6). In order to enable simultaneous quantification of C18:1 n9, which is a main FA in biological samples, quadratic least square regression (weighting: 1/x 2 ) was used. Only low carry-over was observed in the preceding injection of a high-concentrated standard (ESM Fig. S5). For quantification of all other fatty acids, linear calibration up to 2.5 μM was used. This strategy using fatty acid-specific concentration ranges of the calibration series (Table 1)

Accuracy and precision
The accuracy of the developed analytical LC-MS method was assessed by comparing the fatty acyl concentrations in plasma and plant oils with those obtained by a validated GC-FID analysis (Fig. 4) which can be considered the gold standard of fatty acid analysis. The use of GC-FID provides an orthogonal quantification which is not dependent on standard concentrations due to the mass-sensitive detector allowing to deduce absolute concentration based on one reference compound. Sample preparation for gas chromatographic determination included lipid extraction with methanol/methyl tertbutyl ether (MTBE) and transesterification to FAME [12], whereas for LC analysis, the samples were diluted with isopropanol and the supernatant after centrifugation was directly hydrolyzed (Fig. 4a) [31]. The major fatty acyls quantified in the plasma were C16:0, C18:2 n6, and C18:1 n9 + n7, followed by C18:0 and C20:4 n6 as previously described for healthy subjects [7,32]. The main n3-PUFA were ALA, EPA, and DHA having concentrations of around 150 μM (Fig. 4b).
Given the difference in sample preparation, both methods showed an excellent match of the determined concentrations (agreement 80-120%, Fig. 4b). Only for C18:0, slightly higher concentrations are obtained by means of LC-MS, presumably because of its high background signal. Differently diluted hydrolyzed plasma samples show that the matrix leads only to low or no ion suppression allowing robust quantification of the fatty acyls in biological samples (ESM Fig. S7).
In plant oils, similar levels of fatty acyls were found following quantification by means of GC-FID or LC-MS (Fig.  4c, d). For the main fatty acyls in flaxseed oil or sunflower oil, the concentration agreement was also good (70-130%, Fig.  4c, d). Therefore, we conclude that the simple sample preparation by dilution with iso-propanol and saponification is suitable for the quantification of fatty acyls in protein-rich matrices as well as in fatty matrices. It should be noted that more fatty acids could be quantified by LC-MS than by GC-FID due to the higher sensitivity of the LC-MS method. The LLOQ of the LC-MS measurement is ≤ 75 nM for almost all fatty acids, whereas it is more than one order of magnitude higher for the GC-FID analysis. This made it possible to quantify for example C22:5 n6 in plasma or C22:1 n9 in flaxseed oil allowing to gain a more comprehensive picture of the fatty  Fig. 3 Effect of pre-concentration step in the gradient on peak shapes of the first eluting fatty acids. Shown are injections (10 μl) of a fatty acid standard (0.5 μM) in ethanol at different initial gradient conditions. The initial conditions were held for 1 min, then the % B was increased to 66% B in 0.5 min acid pattern of biological samples and plant oils. It should be noted that GC-(MS) following transesterification on the one hand allows the simultaneous detection of a more comprehensive set of fatty acids compared to LC-MS and on the other hand electron ionization MS provides more structural information based on fragments. For example, Lisa et al. found 81 fatty acyls in animal fats including also branched and transisomers [33]. However, if only major fatty acids are of interests as it is in numerous studies of modern life science, LC-MS is in our hands the method of choice because of the rapid sample preparation and analysis. In order to evaluate the precision of the analytical procedure, human plasma samples as well as two edible oils with different fatty acid pattern were analyzed on three separate days (inter-day variance Fig. 4, n = 3). The intra-day variability was assessed by calculating the RSD on each single day (n = 3; ESM Table S1). Both parameters were lower than 15% for almost all fatty acyls in plasma and plant oils and thus meet the criteria required by the EMA guideline [30] demonstrating a high precision of the developed method. Only long-chain saturated and monounsaturated fatty acids such as C20:0 and C20:1 n9 showed in part higher variations. Interestingly, the results using transitions based on decarboxylation resulted in higher RSD, e.g., 15% for C22:5 n3 vs. 5% in pseudo-SRM, which might be explained by low intensity of the [M-H-44] − fragment and thus low peak heights. Therefore, quantification should be carried out by pseudo-SRM mode and the second transition can additionally be used for confirmation.
The determination of fatty acyls by LC-MS offers a much faster and more sensitive method than GC-FID analysis. We could show that the method leads to consistent and precise results. The easy and rapid sample preparation via direct saponification in iso-propanol is not only fast but allows the simultaneous total oxylipin determination (ESM Table S2) [31]. Thus, our approach makes it possible to analyze both oxidized fatty acyls and their precursors from a single sample preparation.

Non-esterified fatty acids in plasma and edible oils
Concentrations of NEFA were determined in plasma (ESM Table S3) and virgin, cold-pressed sunflower oil as well as refined sunflower oil (Fig. 5)   dilution of the sample with iso-propanol. The obtained concentrations (ESM Table S3) were in good agreement with the results described by other groups [16,17]. The accuracy of the measurement was supported by the standard addition procedure using ARA resulting in a concentration of 2.0 μM in plasma, compared to 2.1 ± 0.1 μM by external calibration (ESM Fig. S8). This underlines the robustness of the method against matrix effects and demonstrates that the method allows the reliable quantification of three parameters from a single sample preparation: quantification of fatty acyls, NEFA and total oxylipins. For quantification of NEFA in plant oils, the excess of triacylglycerols was reduced by SPE using aminopropyl cartridges as described [13,34]. As expected, the concentrations of NEFA were considerably lower in refined sunflower oil than in virgin sunflower oil, since the NEFA are removed in the deacidification step of the refining process [35]. C18:2 n6 was the fatty acid with the highest concentration, followed by C18:1 n9 + n7, C16:0, and C18:0. Thus, the concentrations of NEFA represent the total fatty acid distribution (Fig. 4c). Because of the sensitivity of the method, we could detect low-concentrated fatty acids such as C22:1 n9, which often could not be reported [36,37]. Due to the direct analysis of the NEFA by LC-MS, our method does not require derivatization for gas chromatographic analysis compared to previous methods, where different derivatization strategies such as esterification [38], silylation [36] or dimethylamidation [39] are used. The targeted approach allows quantification of NEFA even in the presence of triacylglycerols that may not have been completely removed by the SPE. Thus, the presented LC-MS method herein allows the reliable quantification of NEFA in biological samples, which is of pivotal importance for the characterization of both edible oils [35,40] as well as biological samples such as plasma [2,41].

Conclusion
A new LC-MS method for the quantification of fatty acids in biological samples was developed. Using an optimized C8 reversed phase column, 41 fatty acids and 11 isotopically labeled fatty acids as internal standards could be separated within a total run time of only 15 min. Despite using a rather old, middle class QqQ MS, the method is sensitive with a LLOQ of 10-75 nM for most fatty acids. The low inter-day and inter-operator variability of < 20% indicates a high precision of the method. The concentrations of fatty acyls determined by LC-MS in plasma and plant oils are consistent with those of a gas chromatographic analysis ensuring accurate and comparable results by the developed method. A major strength of the approach is the rapid sample preparation by hydrolysis and dilution allowing high sample throughput. Moreover, the analysis can be combined with the analysis of PUFA oxidation products (eicosanoids and other oxylipins) [31]. Finally, LC-MS analysis allows to quantify NEFA in presence of triacylglycerols which is of pivotal importance for the analysis of biological samples such as plant oils.

Declarations
Ethics approval The collection of blood for plasma generation was approved by the Ethics Committee of the University of Wuppertal and an informed consent was obtained from all human subjects. Fig. 5 Concentration of nonesterified fatty acids in virgin and refined sunflower oil determined by LC-MS; 10 mg sunflower oil were dissolved in chloroform/isopropanol (2/1, v/v) and triacylglycerols were removed by solidphase extraction on aminopropyl cartridges [13]. Refined and virgin oils were obtained from a local supermarket and analyzed in triplicate (mean ± SD)