LIPID CHARACTERIZATION OF INDIVIDUAL PORCINE OOCYTES BY DUAL MODE DESI-MS AND DATA FUSION

Abstract

The development of sensitive measurements to analyze individual cells is of relevance to elucidate specialized roles or metabolic functions of each cell under physiological and pathological conditions. Lipids play multiple and critical roles in cellular functions and the application of analytical methods in the lipidomics area is of increasing interest. In this work, in vitro maturation of porcine oocytes was studied. Two independent sources of chemical information (represented by mass spectra in the positive and negative ion modes) from single oocytes (immature oocytes, 24-hour and 44-hour in vitro matured oocytes) were acquired by using desorption electrospray ionization-mass spectrometry (DESI-MS). Low and mid-level data fusion strategies are presented with the aim of better exploring the large amount of chemical information contained in the two mass spectrometric lipid profiles. Data were explored by principal component analysis (PCA) within the two multi-block approaches to include information on free fatty acids, phospholipids, cholesterol-related molecules, di- and triacylglycerols. After data fusion, clearer differences among immature and in vitro matured porcine oocytes were observed, which provide novel information regarding lipid metabolism throughout oocyte maturation. In particular, changes in TAG composition, as well as increase in fatty acid metabolism and membrane complexity were evidenced during the in vitro maturation process. This information can assist the improvement of in vitro embryo production for porcine species.

Graphical abstract

1. Introduction

Lipids play multiple and critical roles in many biological processes, such as oxidative stress, signal transduction, gamete development, and epithelial morphogenesis. Together with environmental perturbations, pathological conditions and other external factors, variable expression of lipids and other cellular components (e.g., chromosomal material and metabolites) causes heterogeneity of biological systems at a cellular and even sub-cellular level [1–4]. Thus, the investigation of lipid content in single cells and their internal compartments may elucidate specific functionality of cell populations and provide new insights into complicated biological phenomena [5–7]. Several research efforts have been focused lately on mammalian reproductive biotechnologies [3, 8–11]. Indeed, these single cells are characterized by high complexity regarding the array of intracellular molecules [10]. The knowledge of the lipid content in individual oocytes and embryos is of interest in fundamental studies of embryology, in the attempts to understand cellular pluripotency, and in efforts to optimize in vitro culture conditions necessary for the improvements of assisted reproductive techniques [11]. Due to the relevant metabolic similarities between oocytes and embryos with cancer cells [12], the acquired knowledge on cell metabolism may be of interest for stem cell and cancer research as well.

The development of sensitive measurements for single cell analysis is a current need in analytical chemistry. Over time, traditional analytical protocols based on fluorescence microscopy, gas chromatography (GC) or liquid chromatography (LC) combined with mass spectrometry (MS) have been progressively substituted by direct MS systems, due to the necessity to scale-down analytical methods to a single cellular level [4, 6, 7, 10, 13]. Indeed, chemical analysis of lipids via traditional methods involves the pooling of large amount of cells that is unachievable for most studies and prevents analyzing individual variability [1, 8–9]. Moreover, such methods require complex clean-up and extraction procedures, as well as labeling or derivatization steps. On the contrary, direct MS techniques, such as matrix-assisted laser desorption ionization-mass spectrometry (MALDI-MS) [14], secondary ion mass spectrometry (SIMS) [15], and ambient MS techniques, including desorption electrospray ionization (DESI) [16], liquid extraction surface analysis (LESA) [1], and laser ablation electrospray ionization (LAESI) [17], offer rapid and sensitive fingerprinting approaches to obtain lipid profiles from intact cells [9]. Preliminary chemical manipulation – of different extent, depending on the technique – is usually required. On the other hand, these direct MS protocols are less useful in performing quantitation and identification of lipids with similar structures (e.g., same nominal masses) [18], even though the utilization of high mass accuracy m/z measurements or tandem mass spectrometry helps for identification purposes [8,18]. Many applications for MS lipid detection and single cell analysis have been reviewed recently and details can be found elsewhere [6,7]. Briefly, MALDI and SIMS are usually employed for imaging experiments, since they provide adequate spatial resolution [6, 7, 18,19,20]. In comparison to these vacuum systems, ambient techniques offer the advantage of operating in open environment [2] making them suitable for live cell analysis, but they are limited in spatial resolution and, hence, usually adopted for MS profiling. A first attempt to perform live single-cell MS analysis was reported by Mizuno et al, which adopted a nano-ESI tip as a way to trap single mammalian cells and a means of ionization to detect numerous compounds by high-resolution MS [21].

A common aspect to all these direct MS analyses is the acquisition of untargeted lipid profiles to include a broad range of species, which can be used as fingerprints of cells or cellular compartments. As with the adoption of different ionization sources, the use of different mass analyzers and data acquisition systems can provide deeper structural insights into lipids of biological interest, resulting in an information-rich dataset [22]. For example, in the case of ion mobility spectrometry (IMS), an additional dimension of information is available by using chemical characterization [22,23].

Multivariate data analysis is normally used for pattern recognition purposes, in order to efficiently compress high-dimensional data and extract informative chemical features [1, 2, 6, 7, 10]. A wide portfolio of methods is available for two-way and even multi-way analysis, thus following the needs of the acquired MS data structures.

Whenever MS analyses do not alter the morphology of the cell, multiple sources of information can be acquired by repeatedly analyzing the same cell with either different MS techniques or the same one operating under different conditions (e.g., acquisition in positive and negative ion mode mass spectra). Based on the assumption that a broader range of lipids is detected in this way (e.g., different ionization sources may enhance or suppress desorption / ablation / ionization of specific compounds), the exploration of the multi-block sources of information allows a deeper characterization of the cell. By analyzing the same cell, stronger correlations among lipid species detected by multi-block experiments can be derived, and this may help defining biological pathways and processes [1]. To our knowledge, just two recent studies on single cells describe multi-block experiments to widen the acquisition of several classes of lipids from the same bovine embryo. They involve the adoption of MALDI with the utilization of a binary matrix [9], and DESI [11] with a so-called “morphologically friendly” solvent for the spray system [24,25]. While the first study only lists the lipids detected in bovine embryos, in the second one, a data fusion (i.e., multi-block) analysis was adopted to chemically characterize them [11], comprehensively considering the lipids detected in positive and negative ion mode mass spectra. Indeed, data fusion is capable to merge different data sources and analyze them to obtain a comprehensive exploration of the whole information content and model the data in a more robust way [26–30]. Data fusion was previously and extensively applied to merge datasets acquired by MS coupled with LC, GC or ICP – for elemental as opposed to molecular analysis [27] – and NMR, UV-Vis or IR spectroscopy [26,27,31–36], whereas the DESI-MS study on bovine oocytes and embryos is the only report on direct MS analysis.

The strategy of merging two MS profiles (±), instead of tackling them separately, revealed chemical differences between bovine oocytes developed in vivo and in vitro, highlighting inefficiencies of the current in vitro embryo production systems, with possible deleterious effects on the health status of the offspring [11]. Therefore, in the present study, an analogous dual mode DESI(±)-MS analysis, followed by principal component analysis (PCA) within a data fusion strategy, was adopted to explore the lipid content of porcine oocytes during the in vitro maturation (over 44 hours). The oocyte quality, which may affect developmental capacity, can only be poorly appreciated by morphological examination under a stereomicroscope and the investigation of chemical, metabolic, and genomic fingerprints is necessary instead. By contrast, the visual inspection of the cell is the gold-standard technique used to check the stage of development with nearly 100% confidence. That is, the unbiased exploration (i.e., the exploration of the data not guided by a priori categorization of samples, as it would be the case for supervised techniques like PLS-DA) of the information content enclosed into MS spectra by PCA is the initial interest.

Porcine oocytes have never been investigated before for lipid profiling by ambient MS. However, they present the highest content of lipids among domestic species [37–39], and pig is considered a reliable model in comparison to mice and cattle to study human diseases and stem cells [40]. To date, porcine in vitro embryo production and cryopreservation present poor success compared to mouse and bovine [41], and the efficiency of porcine in vitro maturation remains low. Its improvement may positively influence biomedical technologies, such as the production of transgenic pigs [42].

This work represents a further step of our ongoing research on dynamic lipid characterization of female gametes in the field of reproductive biology.

2. Materials and Methods

2.1. Porcine oocytes retrieval

Porcine oocytes obtained from ovaries collected in a local slaughterhouse were matured in vitro, as described in detail elsewhere [43]. Briefly, cumulus-oocyte-complexes (COCs) were placed into TCM-199 medium (50 COCs per 500 μL drop) supplemented with 0.1 mg mL⁻¹ cysteine, 0.5 IU mL⁻¹ LH, 0.5 IU mL⁻¹ FSH and 10 ng mL⁻¹ epidermal growth factor at 39 °C in 5% CO₂. The complete maturation of the oocytes was performed for the 44-hour in vitro matured oocytes by visual inspection of the first polar body under a stereomicroscope, after removal of the cumulus cells by vortexing in 0.1% hyaluronidase.

At different times during in vitro maturation (0, 24 and 44 hours), the COCs were removed from the maturation medium, the surrounding cumulus cells removed by vortexing in 0.1% hyaluronidase and the denuded oocytes were washed three times in PBS supplemented with 0.1% polyvinyl alcohol (PVA) and in methanol:ultra pure water (1:1 v/v). The oocytes were individually placed onto the surface of a glass slide (Gold Seal, Portsmouth, NH, USA). The glass slides were frozen at –80°C until DESI-MS analysis, for which they were allowed to come to room temperature for 15 min and dried.

2.2. DESI-MS

The DESI protocol allows a straightforward analysis of intact individual oocytes and does not require any chemical manipulation, only the manual deposition of the cell onto glass slides is needed. DESI-MS spectra were acquired in positive and then negative ion modes from the same samples. For data acquisition, the background (glass slide) spectrum and then data for up to 10 samples were recorded (~30 s for each oocyte) in one chronogram [11]. The MS data for a total number of 96 immature oocytes (n = 31), 24-hour in vitro matured oocytes (n = 25) and 44-hour in vitro matured oocytes (n = 40) were collected in one working day, randomizing the order of analysis but never changing the order of acquisition between positive and negative ions.

An LTQ linear ion trap mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) was used for the experiments. DESI-MS profiles were acquired using a lab-built stage with the DESI spray positioned at ~2 mm from the glass slide containing the oocytes at an incident angle of 50° [11]. In positive ion mode, the DESI spray was nebulized by applying 5 kV to the stainless steel needle syringe and using a nitrogen gas pressure of 180 psi. Other instrument parameters were: injection time = 100 ms, 2 microscans, m/z range = 600–1250, capillary voltage = 0 V and tube lens voltage = 55 V. Acetonitrile (ACN) doped with 3.0 μg mL⁻¹ silver nitrate at 5.0 μL min⁻¹ flow rate was used as the DESI spray system. The use of AgNO₃ in the DESI spray allows obtaining lipid silver adducts, which are easily recognized by the characteristic 1:1 abundance ratio of the ¹⁰⁷Ag:¹⁰⁹Ag isotopes [44]. In the negative mode, the voltage applied to the stainless steel needle syringe was –5 kV. Other operating conditions were: nitrogen gas pressure = 180 psi, injection time = 500 ms, 2 microscans, m/z range = 250–1000, capillary voltage = −27 V and tube lens voltage = −10 V. The solvent combination ACN and DMF (1:1 v/v) was used at 1.0 μL min⁻¹ flow rate, as already reported elsewhere [10, 24].

Both acetonitrile doped with AgNO₃ and the combination ACN-DFM (1:1 v/v) are “morphologically friendly” solvent systems. The term refers to operating with spray solvents that do not cause damage to the tissue. The use of morphologically compatible solvents is mandatory to acquire DESI mass spectra of individual oocytes because they allow samples to be sprayed for enough so that both positive and negative mass spectra can be acquired, without causing damage. The mechanism behind DESI-MS is a spot-by-spot micro-extraction of molecules from the sample. The extraction process occurring with morphologically friendly solvents is comparable to the fixative procedures commonly used in histology for morphological examination: lipids are removed whereas the intracellular and extracellular proteins stained in the histochemical treatment are preserved. Further details can be found elsewhere [24,25].

The attribution of lipid species was made using high mass resolution DESI-MS analysis performed with a Thermo Scientific Exactive (San Jose, CA, USA) mass spectrometer, and collision-induced dissociation (CID) tandem MS experiments. Molecular formula matching and error calculations were performed using the instrument software Xcalibur v.1.0.1.03 (Thermo Fisher Scientific, San Jose, CA, USA) and online search for lipids was carried out in the LIPID MAPS [45] and METLIN [46] databases.

2.2. Principal component analysis (PCA) within a multi-block approach

For each sample, a list of m/z values and ion abundances from the averaged mass spectrum (~30 s of data acquisition) was imported into Matlab (The MathWorks, Inc., Natick, USA) and multivariate data processing was performed using in-house Matlab routines. All MS spectra were normalized by the standard normal variate (SNV) transform, so as to correct for both baseline shifts and global intensity variations [47,48].

Positive ions covered the mass range m/z 600–1250 (7801 m/z variables), whereas for the negative ion mode, each mass spectrum was composed by 6301 m/z variables over the mass range m/z 425–950. The acquisition step of m/z values was of m/z 0.0833 in both cases. The mass range of the negative ions was reduced from m/z 250–1000 to m/z 425–950 to exclude uninformative or confounding regions of the spectra in which signal from the background (lower masses) or only random noise (higher masses) was collected. Therefore, two data matrices were built: 96 rows (samples: immature oocytes, n = 31; 24-hour in vitro matured oocytes, n = 25; 44-hour in vitro matured oocytes, n = 40) and 7801 columns (m/z values) for positive ions, 96 rows and 6301 columns for negative ions.

PCA is commonly used for exploratory investigations of the complex information contained in full mass spectral datasets, to allow consideration of all the spectral variables and their inter-correlations simultaneously [49,50,51]. By means of PCA, the data information content of the original multivariate collinear m/z variables is reorganized and compacted in few principal components (PCs): that is, PCA can be used as an unsupervised data compression technique. When dealing with high-dimensionality data, the compression of the relevant information is a priority step in order to efficiently manage the data and extract useful information.

PCA was performed at two levels of data fusion to visualize relationships among samples and variables, taking into account the whole information content and inter-correlations among all variables. A graphical scheme of the different PCA strategies is shown in Fig. 1. Briefly, two different approaches of increasing complexity were adopted for data fusion, usually referred to as low and mid-level fusion, respectively [26–30]. In low-level fusion, raw signals obtained from different data sources are merely concatenated and then processed as a unique fingerprint of the samples (Fig. 1b). Usually, this type of approach requires suitable scaling or weighting of data, especially when the different data blocks are dimensionally unbalanced [28]. In the so-called mid-level fusion, informative features extracted from the raw signals of each analytical source – by means of variable selection procedures or latent-variable compression techniques (e.g., PCA itself) – are combined into a new (reduced) single set and then processed by further multivariate analysis (Fig. 1c).

Fig. 1.

(a) DESI-MS lipid profiling of single oocytes. (b) Schematics of two matrices enclosing positive (P) and negative (N) ions for individual porcine oocytes. Positive and negative mass spectra can be visualized for the same sample by selecting the same row R_i in both of the matrices (represented as dark violet and dark green rows in P and N, respectively). (c) PCA within a low-level data fusion strategy: concatenation of two matrices with R rows (96 samples) and V variables (positive ions: 7801; negative ions: 6301) - after block-scaling to equal variance - and then computation of scores (S), loadings (L), and residuals (E). (d) PCA within a mid-level data fusion strategy: PCA was performed individually on the original datasets with R rows and V variables (positive ions: 7801; negative ions: 6301), then the same number of PCs (5) was selected as for each original dataset, and the scores of the selected PCs were merged into a new data matrix, which underwent PCA again, after column autoscaling.

For this study, in the low-level fusion the two matrices of dimensions 96×7801 (positive ions) and 96×6301 (negative ions) were concatenated column-wise. Before concatenation, the two blocks were scaled to the equal variance (equal to 1) [52]. After concatenation, the matrix was column-centered to allow PC rotation around the global centroid of the data (Fig. 1b). In the mid-level fusion, PCA was applied separately to the two matrices. From both, the first PCs that cumulated about 90% of the total variance were selected, merged together in a new dataset, and used to apply a further PCA, after column autoscaling of the new data (Fig. 1c). Lastly, PCA was also performed on positive and negative ions separately to confirm the improved characterization achieved by data fusion in comparison to analyzing the datasets separately.

3. Results and discussion

In this work, a dual mode DESI(±)-MS protocol was used to analyze in a few minutes single porcine oocytes (about 150 μm in diameter), which contain about 156 nanograms of total lipids [53]. PCA was applied within two multi-block approaches (low and mid-level) on the DESI-MS spectra (±) obtained from single cells represented by individual porcine oocytes. The focus is on the porcine oocyte maturation dynamics, which involves a narrow time-window of about 2 days.

3.1. DESI-MS mass spectra obtained by direct analysis of oocytes

High-throughput protocols often compromise the quality of the chemical information that can be obtained from the samples. For example, high-throughput UHPLC-MS/MS methods often deal with scarce quantitative performances, partially sacrificed to expand the UHPLC-MS/MS multi-analyte capability to the utmost. On the other hand, when the high-throughput “morphologically compatible” DESI(±)-MS protocol is adopted to detect lipids from individual microscopic oocytes, a huge amount of complex chemical information can be acquired. Indeed, DESI is capable of softly ionizing lipids of several classes, including intact complex species (such as PLs, in the negative ion mode, and TAG, in positive ion mode). Figure 2 shows positive and negative ion mode mass spectra of three samples selected based on the outcomes of PCA performed within a mid-level data fusion scheme. Selection was made for the objects that were most distant (i.e., different) from each other and from the global centroid (i.e., the vector of the mean value for each variable) to show maximum differentiation. In the negative ion mode, free fatty acids were detected over the mass range of m/z 200–400, fatty acid dimers (resulting from the combination of free fatty acids in gas phase) were detected over the mass range of m/z 500700, and complex phospholipids (phosphatidylethanolamine, PE; phosphoatidylserine, PS; phosphatidylglycerol, PG; and phosphatidylinositol, PI) over the mass range of m/z 800–1000. Using similar experimental conditions, some lipid species have been previously detected in mouse [10] and bovine [11] oocytes and preimplantation embryos, proving that the chemical information detected by DESI-MS is consistent through oocytes and embryos of various animal species. However, differences can be revealed and are expected, since the lipid abundance in oocytes and embryos seems to be species-specific and time-dependent [54]. Lipid attributions based on high-resolution mass measurements and CID MS/MS experiments are listed in Table 1.

Fig. 2.

DESI mass spectra of porcine oocytes selected to show maximum differentiation. Negative ion mode mass spectra: (a) immature oocytes, (b) 24-hour in vitro matured oocytes, (c) 44-hour in vitro matured oocytes. Positive ion mode mass spectra: (d) immature oocytes, (e) 24-hour in vitro matured oocytes, (f) 44-hour in vitro matured oocytes. See Table 1 for tentative lipid class assignments of the major peaks.

Table 1.

Values of m/z with high resolution mass accuracy, predicted molecular formula, ion description, error in ppm (Delta ppm) and tentative attribution of lipids detected in negative and positive ion modes by DESI-MS.

Ions	Predicted Molecular formula	Ion description	Delta ppma	Attributionb
253.21610	C16H29O2	[M-H]−	−1.203	16:1c (palmitoleic acid)
255.23173	C16H31O2		−1.224	16:0 (palmitic acid)
279.23163	C18H31O2		−1.324	18:2 (linoleic acid)
281.24728	C18H33O2		−1.324	18:1 (oleic acid)
283.26290	C18H35O2		−1.354	18:0 (stearic acid)
303.23150	C20H31O2		−1.454	20:4 (arachidonic acid)
311.29407	C20H39O2		−1.484	20:0 (eicosanoic acid)
329.24708	C22H33O2		−1.524	22:5 (docosapentaenoic acid)
339.32558	C22H43O2		−1.274	22:0 (docosanoic acid)
465.30219	C27H45O4S		−2.214	Cholesterol sulphate
511.47073	C32H63O4		−2.454	FA dimer
537.48644	C34H65O4		−2.394	FA dimer
539.50207	C34H67O4		−2.414	FA dimer
559.47058	C36H63O4		−2.604	FA dimer
585.50738	C35H69O6		−2.583	FA dimer
609.48489	C40H65O4		−3.944	FA dimer
788.54144	C42H79O10NP		−3.267	PS (36:1)
747.51486	C40H76O10P		−2.201	PG (34:1)
804.57281	C43H83O10NP		−3.197	PS (37:0)
835.53082	C43H80O13P		−3.382	PI (34:1)
857.51494	C45H78O13P		−3.612	PI (36:4)
885.54613	C47H82O13P		−3.722	PI (38:4)
887.56518	C47H84O13P		−0.322	PI (38:3)
686.18886	C30H50O3NAg2	[M+Ag2NO3]+	0.223 (for m/z	Squalene
			686.18803)d
725.42684	C39H70O5Ag	[M+Ag]+	−0.041	DAG (36:2)c
751.44284	C41H72O5Ag		0.426	DAG (38:4)
909.60999	C51H94O6Ag		0.0445	TAG (48:2)
911.62492	C51H96O6Ag		−0.346	TAG (48:1)
935.62574	C53H96O6Ag		0.540	TAG (50:3)
937.64121	C53H98O6Ag		0.346	TAG (50:2)
939.65570	C53H100O6Ag		−0.889	TAG (50:1)
961.64088	C55H98O6Ag		−0.005	TAG (52:4)
963.65635	C55H100O6Ag		−0.192	TAG (52:3)
965.65848	C55H102O6Ag		1.606	TAG (52:2)
987.65692	C57H100O6Ag		0.390	TAG (54:5)
989.67258	C57H102O6Ag		0.399	TAG (54:4)
991.69092	C57H104O6Ag		3.111	TAG (54:3)
1013.67296	C59H102O6Ag		0.764	TAG (56:4)
1015.68910	C59H104O6Ag		1.245	TAG (56:3)
1037.67308	C61H102O6Ag		0.862	TAG (58:8)
1039.68994	C61H104O6Ag		2.024	TAG (58:7)
1041.70459	C61H106O6Ag		1.060	TAG (58:6)
1140.48205	C59H90O7NAg2	[M+Ag2NO3]+	−0.612 (for m/z 1138.48184)d	Ubiquinone (Coenzyme Q10)

^aDelta ppm column displays the difference between the specified mass and the calculated mass in ppm units.

^bAttribution of lipids is based on Lipid Maps (www.lipidmaps.org) and Metlin (http://metlin.scripps.edu) searches of the predicted molecular formulae and possible adducts.

^c(C:U) represents the number of carbon atoms (C) and the number of unsaturations (U) of the fatty acyl chains.

Abbreviations used: DAG – diacylglycerol; FA, fatty acid; PI- phosphatidylinositol; PS - phosphatidylserine; PG - Phosphatidylglycerol, TAG – triacylglycerol. The ‘o-’ suffix is used to indicate the presence of an alkyl ether substituent, whereas the ‘p-’ suffix is used for the 1Z-alkenyl ether (plasmalogen) substituent.

^dDue to the fact the squalene and ubiquinone were adducts presenting Ag₂NO₃, the exact mass has been calculated based on the monoisotopic mass as reported in PLoS One. 2013 Sep 20;8(9):e74981.

In the positive ion mode, the most abundant ions in the three experimental groups (immature oocytes, 24-hour and 44-hour in vitro matured oocytes) were triacylglicerol (TAG) lipids, which have 48 to 58 carbon atoms in the fatty acyl residues and one to five units of unsaturation (see Fig. 2 and Table 1). Indeed, TAGs have been reported to be the major intracellular lipids in porcine oocytes by conventional lipid extraction and chromatographic analysis of thousands of pooled porcine oocytes [55]. Other lipids detected in the DESI-MS profile were diacylglycerols (DAG), squalene and ubiquinone (see Fig. 2 and Table 1). Table 1 lists the major DAG and TAG species detected by DESI-MS in terms of the total number of carbon atoms and the unsaturation present in the fatty acyl residues (C:U). The unique identity of the fatty acyl residues in the DAG and TAG species can be specifically defined by MS/MS experiments.

3.2. PCA within a multi-block approach

Multivariate analysis was performed to extract useful information from the porcine oocytes among the large amount of data embedded within the DESI-MS profiles. As graphically represented in Fig. 1, both low- and mid-level data fusion strategies were tried to compute PCA. For the two of them, Fig. 3 shows the score and loading plots of the lowest-order PCs along which the major separation among the objects (i.e., samples) is visible. In more detail, the score plots from the low and mid-level approaches, show separation among the color-labeled objects, which represent immature oocytes (green circles), 24-hour in vitro matured oocytes (red diamond) and 44-hour in vitro matured oocytes (blue triangles). Notably, the separation for the mid-level fusion is visible within the PC1 vs. PC2 score space (Fig. 3c) and appears clearer both in respect to the low-level fusion (Fig. 3a) and to the PCA performed on positive and negative ions separately (Fig. 4a and 4c). Moreover, in the latter cases the exploration of higher-order PCs is necessary to see groupings among the target categories (Fig 3a, Fig. 4a and 4c). This means that the inter-correlations among positive and negative blocks are informative and that the compression of the whole information content through a mid-level strategy is so efficient that the separation of interest shows up as the dominant fraction of variability into the MS data.

Fig. 3. PCA of the fused datasets from positive and negative ion mode mass spectra in a low-level approach:

(a) PC2 vs. PC3 score plot. (b) PC2 vs. PC3 original loading plots labeled in terms of m/z ratio. PCA of the fused datasets from positive and negative ion mode mass spectra in a mid-level approach: (c) PC1 vs. PC2 score plot. (d) PC1 vs. PC2 original loading plots labeled in terms of m/z ratio. Immature oocytes: green circles (n = 31), 24-hour in vitro matured oocytes: red diamonds (n = 25), 44-hour in vitro matured oocytes: blue triangles (n = 40). Negative ions: green; positive ions: violet.

Fig. 4. PCA of negative ion mode mass spectra:

(a) PC2 vs. PC5 score plot. (b) PC2 vs. PC5 original loading plots labeled in terms of m/z ratio. PCA of positive ion mode mass spectra: (c) PC2 vs. PC3 score plot. (d) PC2 vs. PC3 original loading plots labeled in terms of m/z ratio. Immature oocytes (green circles, n = 31), 24-hour in vitro matured oocytes (red diamonds, n = 25), 44-hour in vitro matured oocytes (blue triangles, n = 40).

The mid-level fusion also appears to be more informative in the attempts to extract the relevant chemical features from the loading plot with a balance among the information provided by both positive and negative ions (Fig 3d). Note that the original MS data are described by a higher number of variables for the positive ion mode and two orders of magnitude of difference for the absolute ion intensities (positive ions are more intense than negative ions). By contrast, in the low-level fusion, the loadings associated to the positive ion mode are compressed to the centroid and appear to be unbalanced in comparison to those of negative ions in computing the PCs (Fig 3b), although the two blocks of data have been block-scaled to equal variance [53]. Note that the loadings (L) originally coming from the dataset of positive ions were color-labeled in violet, while those of the dataset of negative ions were color-labeled in green, and computed through PCA from the relation:

$$\mathbf{X}=\mathbf{S}\cdot {\mathbf{L}}^{\mathbf{T}}+\mathbf{E}$$

where X is the concatenated mass dataset of positive and negative ions, S are the scores, and E the residuals (Fig. 1b).

In Fig. 3d, the PC1 vs. PC2 loading plot shows the contribution of the original variables in determining the final PCs, after mid-level data fusion. Loadings of the combined PCA and related to the original variables (O) have been computed by:

$$\mathbf{O}={\mathbf{L}}^{\mathbf{T}}\cdot \mathbf{F}$$

where F is obtained by merging column-wise the individual loading matrices obtained in the separated PCA (L_pos^T and L_neg^T), and L^T is the loading matrix obtained from the PCA performed onto the merged score matrices (Fig. 1c). In more detail, F is a J_tot × V_tot matrix, where J_tot is the total number of principal components from the individual PCA and V_tot is the number of original variables; L^T has dimensions K × J_tot, where K is the number of PCs retained in the final PCA. Therefore, O has dimension K × V_tot.

The visualization of the original variable contribution from the two independent MS sources in determining the final PCs allows a straightforward interpretation of the differences in the chemical features among sample categories. From the loading plot, it is possible to extract information on the inter-correlation among lipids (based on the relative position of different ions in the loading plot) and to highlight ions that seem to contribute the most to PC computation and, thus, to the separation among sample categories that is visible along the same PCs in the score plot. In interpreting the loadings for this specific study, it is important to remark that, although SNV has the peculiarity to potentially shifting information across the signal range, so that the interpretation of the results referring to the original signals may be deceiving [47,48]. In the outcomes of the present study, the changes in the ion relative abundances highlighted by PCA are in accordance with the information shown by the averaged mass spectra of each sample category (see Fig. S1 of the Supplementary Material).

In more detail, in the loading plot of PC1 vs. PC2, the positive ions (violet m/z values in the loading plot) of m/z 725.5 (DAG 36:2), 751.5 (DAG 38:4) appear to be more abundant in the immature oocytes (green circles). Diacylglycerol is well characterized as a second messenger signaling lipid originating from the hydrolysis of the phospholipid phosphatidylinositol 4,5bisphosphate (PIP₂) by the enzyme phospholipase C (PLC). In the same reaction, inositol trisphosphate (IP₃) is produced. Although IP₃ diffuses into the cytosol, diacylglycerol is hydrophobic and remains within the plasma membrane where it is believed to activate protein kinase C, thereby regulating calcium influx [56]. The increase in DAG in immature oocytes may be related to the active signaling processes needed to resume meiosis and promote maturation. By using MS, it is possible to obtain more details on the structure (number of carbon in the fatty acyl chains and unsaturation) of the DAG present in the oocytes and, if needed, MS/MS can be used to provide the identity of the fatty acyl residues. Diacylglycerols tend to be minor components of most tissues, in quantitative terms, but they are very important in animal tissues, because they function as second messengers in many cellular processes, modulating vital biochemical mechanisms. Indeed, mouse oocytes with higher rate of meiosis resumption presented increased content of DAG [57]. Also, alteration in lipid metabolism involving DAG species has been reported in adipose tissue of Bscl2-deficient mice [58].

The ions of m/z 1013.6 (TAG 56:4), 1015.6 (TAG 54:3), 939.6 (TAG 50:1) are more abundant in 24-hour in vitro matured oocytes (red diamonds), whereas the ions of m/z 886.6, 987.6 (TAG 54:5), 989.6 (TAG 54:4), 963.6 (TAG 52:3) are more abundant in the 44-hour in vitro matured oocytes (blue triangles). This outcome suggests that TAG metabolism occurring during oocyte maturation involves changes in the abundance of some TAG species and these changes can be monitored by DESI-MS. As a matter of fact, a decrease in TAG content reflects the massive uptake of TAG as energy source during oocyte maturation [59].

In contrast to bovine immature and in vitro matured oocytes [11], squalene (m/z 688.6) does not contribute to differentiate the samples according to the maturation state. On the other hand, ubiquinone (m/z 1140.4) is of low abundance in porcine oocytes in comparison to bovine oocytes [11]. This result may suggest that differences in cholesterol metabolism and mitochondrial activity during maturation exist between bovine and porcine oocytes.

In the negative ion mode (green m/z values in the loading plot), the relative abundance of the ions of m/z 788.5 (PS 36:1) is higher after 24 hours of maturation (red diamonds). Dimers of free fatty acids (over the mass range m/z 500–600) appear to be more abundant in the 44-hour in vitro matured oocytes (blue triangles), as well as in the ions of m/z 804.4 (PS 37:0), 857.5 (PI 36:4), 885.5 PI (38:4). The relative abundances of all these lipids appear to be lower in the immature oocytes (green circles). Therefore, as indicated by DESI-MS, the increase of free fatty acid concentration (probably by TAG metabolism) as well as membrane specialization for signaling and apoptosis processes should occur during porcine oocyte maturation. Higher abundance of heavier and more complex lipids has been previously characterized by DESI-MS lipid profiles of mouse and bovine oocytes after maturation [10, 11]. The wide range of PLs in in vitro matured oocytes compared with immature oocytes was related to a preparation for fertilization and reflects the acquisition of functional and structural specialization by cellular membrane [10].

It should be mentioned that limitations of direct lipid MS analysis in single cells exist and are mainly related to the measurement of the relative amount of lipids present in the samples instead of to absolute quantification, even though good agreement between relative signals in DESI and LC-MS/MS analyses has been reported [60]. This is due to the impossibility of adding internal standards and to the absence of sample extraction processing. Therefore, even though direct analysis in a complex mixture allows for sensitive and broad lipid characterization, it may introduce bias in the data. Nevertheless, the simplified information extracted by PCA, displayed in score and loading plots, highlights relevant chemical features and may operate as a screening step. Indeed, further MS analyses for quantification can be focused on the lipids that appear to be characterizing.

4. Conclusions

The ambient MS technique DESI was utilized to acquire two independent sources of chemical information – positive and negative ion mode mass spectra – from porcine oocytes through the in vitro maturation process. Data of both ion modes were acquired from the same microscopic samples by switching polarity and optimizing the DESI spray solvent system. Using a data fusion protocol, PCA was applied to extract and visualize the information content. The mid-level approach showed improvements in data analysis in comparison to working up the data separately. Through this methodology, it was possible to observe a dynamic change in TAG composition as well as an increase in fatty acid metabolism and membrane complexity during porcine oocyte maturation in vitro. These findings are coherent with previous knowledge about the mammalian oocyte maturation process. Remarkably, this dual mode DESI(±)-MS analysis followed by a multi-block approach for data analysis proved sensitive to even minor changes in the lipid content, and thus feasible to reveal dynamic difference in single cells, besides differences between oocytes and embryos at different developmental stages and between in vivo and in vitro counterparts [10,11].

With the assumption that multiple sources of chemical information may be joined to improve data analysis and the resulting biological interpretation, data fusion is feasible for single cell analysis using independent combinations of MS techniques; for example, DESI followed by MALDI [61], to combine lipid and protein profiles. Prospectively, these molecular fingerprints can be used jointly to model one or more classes of embryos and chemically compare them with counterparts developed under different conditions.

The expansion of this multi-block methodology to process MS images and multi-way datasets acquired on the same sample is also possible. For imaging MS, ambient techniques may be of preference, since neither spatial dislocation nor chemical modifications of endogenous compounds occur on the sample surface [62].

Highlights.

• Repeated analysis by DESI(±)-MS of intact single oocytes for lipid characterization

• Deployment of a data fusion strategy to merge positive and negative ion mode data

• Enhanced interpretation of metabolic changes by more efficient analysis of spectral data

• Discovery of increased fatty acid metabolism and membrane complexity during maturation

• Assistance in the improvement of in vitro embryo production for porcine species