Lipid profile migration during the tilapia muscle steaming process revealed by a transactional analysis between MS data and lipidomics data

Lipids are fundamental food constituents affecting nutrition and flavor and are categorized into several classes including fatty acids, glycerolipids, glycerophospholipids, and others. LC-MS-based lipidomics, particularly UHPLC-QE Orbitrap MS, offers high sensitivity and resolution for lipid profiling and has been applied across various food matrices. Tilapia, rich in unsaturated fatty acids and with relatively low total lipid content, provides a suitable model to investigate lipid profile changes during processing. Prior studies explored lipid profiles across tilapia tissues and under different thermal processes, but none examined lipid migration from muscle to the juice produced during steaming. This study aims to develop a transactional analysis integrating UHPLC-QE Orbitrap MS data with lipidomics outputs to reduce instrument workload and remove false positives, and to characterize lipid profile migration from tilapia muscle to juice during steaming at 0, 10, 30, and 60 minutes. The work seeks to understand how lipid classes and individual species redistribute or transform, informing nutritional assessment of steamed tilapia and its juice.

The paper reviews LC-MS lipidomics advancements and the advantages of UHPLC-QE Orbitrap MS for high-resolution, accurate mass lipid identification and quantification. Previous applications include profiling of dairy milks, bee pollen, and marine phospholipids, and analyzing non-esterified fatty acids without derivatization. In tilapia, lipidomics distinguished lipid profiles among muscle, head, and viscera, and showed that different thermal processes (steaming, boiling, roasting) yield distinct muscle lipid profiles. Effects of salting on long-chain free fatty acids in tilapia have also been studied. Steaming has been recognized as a cooking method that preserves nutrients and sensory qualities, with documented impacts on flavor, metabolites, and contaminants in various foods. However, prior works relied on standard lipidomics outputs that provide a single value per lipid from parallel LC-MS runs, leading to high LC-MS workloads, and they generally did not address false positives from software-based identifications. No prior study investigated substance (lipid) migration from tilapia muscle to the steaming-generated juice.

Study design: Tilapia fillets were steamed for 0 (raw), 10, 30, and 60 minutes. Lipids from muscle and the corresponding juice were extracted and analyzed by UHPLC-QE Orbitrap MS. A transactional analysis workflow integrated lipidomics (LipidSearch) and MS (TraceFinder) data to correct identifications, normalize across matrices, and enable statistical comparisons. - Samples: Frozen tilapia fillets (180–200 g each) were thawed, halved, and steamed in bowls over boiling water (induction cooker). After bringing water to boil at 2100 kW, steaming was conducted at 1600 kW for 10, 30, or 60 min; raw (0 min) served as control. Each condition was prepared in hexaplicates. - Lipid extraction: For muscle, 1.0 g paste was extracted with chloroform/methanol (2:1, v/v) following Folch/Bligh-Dyer with modifications, including vortexing, shaking (200 rpm, 10 min), and centrifugation (1274 × g, 15 min, 4 °C; last step 15,285 × g). Organic phases were combined, concentrated at 35 °C to ~300 µL, then freeze-dried. For juice, 1.0 mL was extracted similarly. - UHPLC-QE Orbitrap MS analysis: Heated ESI in positive and negative modes; column ACQUITY UPLC HSS T3 (100 × 2.1 mm, 1.8 µm); binary mobile phases (A: acetonitrile/water 60:40 with 10 mM ammonium acetate; B: isopropanol/acetonitrile 90:10 with 10 mM ammonium acetate); flow 300 µL/min; gradient from 37% to 98% B over 20 min, hold, re-equilibrate. Instrument settings: spray 3000 V, heater 300 °C, capillary 320 °C, sheath gas 35 Arb, aux gas 10 Arb, column 45 °C, tray 10 °C. Data-dependent MS/MS: full scan res 70,000; MS2 res 17,500; m/z 240–2000 (positive), 200–2000 (negative). QC samples (pooled) monitored retention time, RSD of peak area, and mass accuracy. - Transactional analysis workflow: 1) Acquire raw MS data and process in TraceFinder 4.1. 2) Import into LipidSearch 4.1.3 to identify compound lipids; obtain lipidomics tables for both ion modes. 3) Screen lipidomics data by thresholds: MainS/N ≥ 5, IDNum ≥ 3, MainMScore ≥ 5, MainArea ≥ 1×10^5; deduplicate to one row per lipid; merge positive/negative, keeping higher MainArea. 4) Normalize screened lipidomics data to initial muscle amount to enable muscle–juice comparisons using provided equations dividing peak areas by initial muscle mass considering sample mass/volume at each time point. 5) Screen MS data by removing lipids not present in normalized lipidomics; merge ion modes; apply MS screening (m-Score ≥ 5, MainArea ≥ 1×10^5); deduplicate across ionizations and quantified ions (sum MainAreas if needed); require IDNum ≥ 3. 6) Normalize screened MS data using the same normalization approach as in step 4; compute total normalized peak areas across muscle and juice. 7) Correct lipidomics data by removing lipids absent from normalized MS data (delete false positives) to obtain corrected lipidomics tables. 8) Statistical analysis: Import normalized MS data (lipid classes and individual lipids) into MetaboAnalyst 5.0 for PLS-DA, VIP scoring (VIP > 1 for discriminatory variables), heat maps, and one-way ANOVA (adjusted p < 0.05; Fisher’s LSD post hoc). - Data outputs: Corrected lipidomics for individual lipids and classes (muscle, juice, total system) and normalized MS data for multivariate statistics.

- Transactional analysis performance: Deletion of false positives in lipidomics data ranged from 22.4% to 36.7%. In muscle samples, 16/59 (0 min), 22/72 (10 min), 29/79 (30 min), and 11/30 (60 min) lipids were false positives relative to corrected sets. In juices, 62/237 (10 min), 48/214 (30 min), and 57/235 (60 min) lipids were false positives. The workflow reduced UHPLC-QE Orbitrap MS workloads by eliminating the need for parallel runs solely to generate replicates for lipidomics software. - Migration patterns: Individual lipid changes observed included disappearance (e.g., AcCa(10:0), Cer(d16:0_22:0), Cer(d16:1_23:0)); full migration to juice (e.g., LPC(18:2), TG(18:0_16:0_20:0)); and appearance in juice (e.g., BisMePA(32:1_18:1/18:2/18:3), DG(16:0_18:1/18:3/20:4/22:4)). At the class level, six change types were identified: disappearance (AcCa, WE); full migration to juice (DG, PG); appearance in juice (BisMePA, CmE, PEt, SPH); appearance in muscle (LPE, LPG, dMePE, MG); appearance in both muscle and juice (FA, PI, PS, SM); retention in muscle (CL, LPI, LPS). - Abundance trends (corrected lipidomics, Table 2): In raw muscle, TG was dominant (~E+14). After steaming, FA (~E+13), LPC (~E+13), and TG (~E+14) were highest in muscles; TG also dominated in juice (~E+14). Total normalized peak areas across muscle+juice remained high for TG (e.g., 5.20E+14 at 10 min) and increased presence of FA and LPC in muscle with steaming time. - Multivariate statistics (normalized MS data): • Muscles: Lipid classes at 0/10/30/60 min clearly separated (PLS-DA cumulative variance 55.1%). Nine discriminating classes (VIP>1): LPE, LPI, LPC, FA, LPG, TG, PS, SM, CL. Individual lipids also separated (variance 40.6%); 52 individual lipids had VIP>1. • Juices: Lipid classes were not clearly separated by time; five classes had VIP>1 (FA, DG, PG, CmE, SM). Individual lipids separated (variance 27.0%); 116 individual lipids had VIP>1; heatmaps showed clear time-dependent patterns for 10/30/60 min. • Total muscle+juice system: Lipid classes clearly separated (variance 57.8%); ten discriminating classes (VIP>1): SPH, LPC, FA, LPI, BisMePA, LPG, LPE, PEt, PC, WE. Individual lipids separated (variance 42.5%); 178 individual lipids had VIP>1. - Summary metric: Compared to individual lipids, fewer lipid class variables sufficed for discrimination: muscles 9 vs 52; juices 5 vs 116; total system 10 vs 178, indicating strong time-dependent migration signatures at both class and species levels.

The study directly addresses the research question of whether and how lipids migrate from tilapia muscle to juice during steaming by integrating high-resolution MS with a transactional data correction approach. The workflow eliminated a substantial fraction of false positives common in software-based lipidomics, enhancing confidence in lipid identifications and quantitation without requiring increased LC-MS runs. Clear temporal separations for muscles and the combined muscle–juice system, and identification of discriminatory lipid classes and species, demonstrate dynamic lipid redistribution and transformation under heat. Amphipathic classes (Cer, LPC, PC, PE, TG) present in both matrices across times suggest potential liposome formation in aqueous juice, consistent with literature on liposome behavior. The six observed change patterns (disappearance, full migration, appearance in juice or muscle, appearance in both, and retention) reflect heat-induced hydrolysis, remodeling, and partitioning processes (e.g., DG/TG hydrolysis to FA; PE/PG to LPE/LPG; SM to SPH), providing mechanistic context. While juice lipid classes were less separable by PLS-DA than muscles, individual lipid signatures in juice were strongly time-dependent, indicating complex, species-specific migration and formation events. Overall, the findings validate the transactional analysis as a robust strategy for lipidomics data curation and reveal significant lipid migration from muscle to juice during steaming.

This work introduces a transactional analysis procedure integrating UHPLC-QE Orbitrap MS and lipidomics outputs to reduce LC-MS workload and remove 22–37% false positives, improving reliability of food lipidomics. Applying this approach to steamed tilapia revealed substantial lipid migration from muscle to juice, with six characteristic change patterns and strong time-dependent differentiation at both class and species levels. Nutritionally, TG remained dominant, with notable increases in FA and LPC in steamed muscles; amphipathic lipids likely form liposomal structures in juice. Future research should: (1) employ targeted lipidomics to resolve specific precursor–product relationships; (2) incorporate appropriate internal standards (ideally per lipid class) to enable absolute quantification; (3) conduct comprehensive free fatty acid profiling optimized for FA; and (4) extend the approach to other food matrices and processing methods to generalize migration phenomena.

- No internal standards were used to report absolute amounts (e.g., mg/g), potentially limiting quantitative accuracy; this choice avoided interference with lipid measurement but is a primary shortcoming. - The untargeted lipidomics approach cannot unambiguously assign origins/products of specific lipid transformations during heating. - Fatty acid data derived from compound lipid settings may be incomplete or less accurate; a dedicated FA method is needed. - Lipid classes in juices were not clearly separable by PLS-DA, suggesting matrix complexity and potential variability affecting class-level resolution.