Four-dimensional trapped ion mobility spectrometry lipidomics for high throughput clinical profiling of human blood samples

Lipids play essential roles in metabolic and physiological processes and in disease, but few lipid measures beyond conventional clinical lipids are used in practice. Recent evidence and advanced MS-based lipidomics have enabled clinical risk scores (e.g., CERT1/2) based on ceramides and phosphatidylcholines. To further clinical translation, standardized and validated quantitative methods, stringent pre-analytical and analytical criteria, cross-laboratory validation, and consensus concentration levels are needed. Improvements in MS technologies (high-resolution Orbitrap/TOF; ion mobility spectrometry) add an orthogonal separation based on collisional cross section (CCS). Trapped ion mobility spectrometry (TIMS) with parallel accumulation serial fragmentation (PASEF) increases speed and sensitivity, enabling four-dimensional (4D) lipidomics integrating m/z, retention time (RT), CCS, and MS/MS spectra. However, high-throughput confident annotation requires strict matching parameters (RT, accurate mass, CCS, isotopic pattern, MS/MS libraries), robust curation protocols, and expanded 4D reference libraries; otherwise, false discovery rates can be high (~50%). Here, the authors develop a high-throughput 4D TIMS lipidomics platform combining automated lipid extraction with microflow UHPLC-TIMS-MS/PASEF, an integrated 4D annotation strategy, and TOF MS survey-based quantification using level-3 internal standards. They benchmark on NIST plasma/serum SRMs, cross-validate with MRM and inter-lab data, and demonstrate applicability by profiling intra-individual, multidien lipidomes across plasma, serum, whole blood, and dried blood spots.

Background work emphasizes: (1) clinical relevance of lipidomics and emergence of ceramide/phosphatidylcholine-based risk scores (CERT1/2) for cardiovascular risk; (2) need for method standardization, validation, and cross-laboratory harmonization in lipidomics; (3) advances in high-resolution MS and ion mobility spectrometry providing an added CCS dimension; (4) development of TIMS and PASEF improving acquisition speed, sensitivity, and multi-precursor fragmentation; (5) prior reports highlight risks of high false discovery rates without stringent annotation criteria, and the utility of RT and CCS for confident annotation; (6) availability and limitations of existing spectral/CCS databases and predictors (e.g., LipidIMMS, CCS predict). These works motivate building comprehensive 4D libraries and robust feature selection criteria to reduce misannotations in high-throughput workflows.

Overview: The workflow integrates automated high-throughput MTBE-based lipid extraction on a robotic platform with microflow UHPLC-ESI-TIMS-TOF MS using PASEF, a curated in-house 4D spectral/RT/CCS library (ClinLip Analyte List), stringent feature selection and annotation criteria, and quantitative analysis via survey TOF MS using level-3 class-specific internal standards (ISTDs) with multi-point calibration; cross-validation includes MRM and inter/intra-day reproducibility and comparisons to inter-laboratory studies.

Sample preparation and automated extraction: MTBE-based liquid-liquid extraction adapted to a robotic platform was optimized for organic phase removal (pipetting volume, depth, speed) to minimize carryover of polar phase. NIST SRM 1950 plasma and serum were used for development and benchmarking. Recovery and matrix effects were evaluated by spiking deuterated ISTDs pre- and post-extraction (automated vs manual vs no extraction). Automated extraction achieved comparable recoveries/matrix effects to manual methods with low CVs; processing time per 96-well plate ~3 h; many analytes showed CV <10% across 32 replicates; all classes CV <20%.

Liquid chromatography: Microflow UHPLC on a C18 Luna Omega column (100 × 2.1 mm, 1.6 µm, 45 °C), 20 min gradient, flow 0.2 mL/min. Mobile phases: negative mode A MeOH/H2O (1:1), B MeOH/IPA (2:8), both with 0.1% formic acid, 7.5 mM ammonium formate, and 0.1% triethylamine; positive mode same without TEA. Autosampler at 4 °C. Injection: 20 µL (neg; 1 µL plasma on-column) and 10 µL (pos; 0.5 µL plasma on-column). Carryover was absent with wash protocol.

MS acquisition (TIMS-TOF, PASEF): TIMS-TOF Pro (negative) and TIMS-TOF flex (positive), ESI; source: end plate 500 V; capillary 3600 V (neg), 4500 V (pos); dry gas N2 6 L/min at 200 °C; nebulizer 2.5 bar; threshold 100 counts. Scan range m/z 100–1350; acquisition cycle 0.1 s; mobility 1/K0 range ~0.55–1.86/1.87 Vs cm−2; collision energy 45 eV (neg), 35 eV (pos). Weekly external calibration (Agilent ESI tune mix) and online recalibration each run using tune mix + 1 mM sodium formate (reduced from 10 mM to limit sodium adducts). Three segments: 0–0.05 min equilibration, 0.05–0.3 min calibrant infusion, 0.3–20 min sample data.

In-house library and annotation (ClinLip Analyte List): 200 lipid standards (150 positive, 158 negative; 16 classes) individually analyzed (triplicates) to record RT, CCS, m/z, and MS/MS spectra; dominant adducts characterized (e.g., [M−H] in negative for most classes except LPC/PC/SM [M+HCOO]; positive [M+H] general; [M+NH4] for PG, CE, TAG). Expanded library by manually curated lipid species from NIST plasma SRM using spectral matching (MS-DIAL), rule-based fragmentation, RT and CCS differences, and CCS predict to infer isomer trends. Final library: 801 entries (391 negative, 424 positive). Annotation scoring combined precursor mass deviation (1–3 ppm), RT deviation (0.1–0.5 min), CCS deviation (0.2–1.5 Å2), isotopic fit, and MS/MS score (cutoffs: good ≥900, moderate ≥500). Feature detection used MetaboScape with intensity threshold 200, recursive feature extraction across ≥17/32 analyses, inclusion only if present in 32/32, and background subtraction.

Feature selection and unknowns: From 15,899 initial features (negative mode across 32 plasma extracts), filters reduced to 3,013 (1/32), then 1,445 (≥17/32), 1,436 after background subtraction, and 698 overlapping with dilution experiment (2 to 0.0625 µL plasma on-column, triplicates). 470 features exhibited dilution response (Pearson ≥0.9; SD from mean >0.1) and were retained as stable unknowns for future annotation.

Quantification strategies: Level-3 class-specific ISTDs (set 1 exogenous lipids uncommon in plasma; set 2 deuterated ISTDs) used. Multi-point calibration: 7-point external standards per class spiked with ISTDs; class-specific linear ranges determined; calibration via (AES/AIS)/(CES/CIS) with regression parameters to compute concentrations; acceptance ±20% accuracy per point. One-point calibration: single-point parameter m = (AES/AIS)/(CES/CIS) used to quantify; evaluated against multi-point. Survey TOF MS peak areas exported from MetaboScape; recursive algorithm peak areas were excluded due to inconsistencies; calculations performed in Excel; results normalized to nmol/mL. LLOD and LLOQ computed per class by ANOVA with LLOD = 3.3σ/s and LLOQ = 10σ/s.

Cross-validation: Quantified selected species by LC-MRM (AB Sciex 5500 QTrap) with polarity switching and established MRM transitions to compare with TIMS-TOF results. Compared concentrations to inter-laboratory consensus (NIST SRM 1950; Bowden et al.) and HILIC-UHPLC-MS data (Wolrab et al.).

LBlooD study (clinical applicability): Four volunteers; five matrices: plasma, serum, whole blood (EDTA), venous DBS, finger-prick DBS; three time points (day 0, 1 week, 1 month); total 180 samples per polarity (45 per individual). Standardized sampling/storage; extraction by MTBE/MeOH; multi-point quantification with set 1 ISTDs (SM d18:1_12:0 for manually extracted SM). Statistical analyses: Wilcoxon signed-rank tests (matrix comparisons), Friedman tests (time-point variation) with Benjamini–Hochberg correction, PCA, and random forest classification with 3×4 cross-validation (AUROC reported).

• Automated MTBE-based robotic extraction achieved comparable recoveries and matrix effects to manual extraction and low variability: most analytes CV <10% over 32 plasma samples; all classes <20% CV; overall extraction time ~3 h per 96-well plate; full 96-sample workflow in <3 days including calibration/QC and processing.
• High inter-assay 4D reproducibility: median precursor mass error 0.58 ppm; median RT CV 0.19%; median CCS CV 0.11%; average per-class inter-day metrics (negative mode): mass error 0.25–0.99 ppm; RT CV 0.12–0.33%; CCS CV 0.09–0.18% (Table 2).
• 4D library and stringent feature selection improved confident annotation and reduced false positives. In 32 NIST plasma SRM extracts, 370 lipids annotated with high confidence (55 unique to negative mode, 191 unique to positive); 359 qualified for quantification. In NIST serum SRM, 364 lipids identified using the same criteria (>90% of plasma coverage). 470 stable unknown features with dilution response were prioritized for future structural elucidation from 15,899 initial features (negative mode).
• Chromatographic and TIMS separation enabled partial isomer resolution (e.g., cis/trans PC 18:1/18:1 and PG 18:1/18:1); reversed-phase LC offered superior isomer resolution to TIMS under 20 min method; ultra-high mobility resolution can resolve select TG isomers.
• Quantification performance: Class-specific linear ranges spanned ≥64-fold; LLOD/LLOQ established per class in both polarities. Multi-point quantification showed inter-day and inter-plate reproducibility with most species CV <20% and overall <30% across 64 analyses. One-point calibration agreed well with multi-point for most classes (CV <13%), but showed larger deviations for ceramides, underscoring advantage of multi-point curves.
• ISTD evaluation: Set 1 vs set 2 ISTDs yielded generally consistent concentrations (average between-set CV ≤25% for most classes in negative mode; higher CVs for Cer, FAHFA, HexCer). Some ISTDs had endogenous presence (e.g., PC 17:0_14:1, SM d18:1_12:0) requiring corrections or alternatives depending on polarity. TG and CE exhibited decay during long autosampler residence; decays partially compensated by corresponding ISTDs; operational controls recommended.
• Cross-validation: TIMS-TOF quantification in both polarities agreed with targeted LC-MRM for selected species (average CV <20%). Concentrations aligned with inter-lab NIST SRM reports (Bowden et al.) and HILIC-UHPLC-MS data (Wolrab et al.) for majority of lipids (≈60–70% with CV <30%), with deeper coverage here.
• Concordance across ion modes: 125 species quantified in both modes showed strong agreement; 55 negative-mode unique and 191 positive-mode unique species also quantified.
• Plasma vs serum: About 90% of NIST plasma lipids detected in serum with similar quantitative patterns; 354 lipids quantified in serum.
• Clinical phenotyping (LBlooD): Distinct lipidome phenotypes among matrices; plasma and serum similar; venous and finger-prick DBS similar; both distinct from whole blood. Random forest multiclass AUROC = 0.92; pairwise AUROCs near random only for plasma vs serum and venous vs finger DBS; others showed near-perfect separability. Friedman test showed no significant multidien changes across 3 time points; intra-individual lipidomes relatively stable; TGs showed higher multidien variability (≥20%). Method reproducibly phenotyped individuals across matrices/time. CERT2 scores derived from this platform indicated low cardiovascular risk in NIST SRM and in individuals, consistently across time points.

The study demonstrates that integrating TIMS-PASEF with a curated 4D library and strict feature selection enables confident, routine annotation and reproducible quantification suitable for high-throughput clinical lipidomics. The added CCS dimension, combined with accurate RT and MS/MS, reduces misannotations and supports cross-laboratory portability of annotations. Automated MTBE extraction and microflow UHPLC-TIMS-TOF provide high precision and throughput, addressing key requirements for bioanalytical assays. Quantification using survey TOF MS with level-3 ISTDs and multi-point calibration achieved broad dynamic range and inter-day reproducibility, with cross-validation against MRM and inter-lab data supporting accuracy. The workflow successfully profiled intra-individual lipidome phenotypes across diverse blood matrices and showed multidien stability, important for clinical marker reliability. The selection of 470 stable unknown features underscores discovery potential for expanding lipidome coverage. While some isomeric separations remain challenging with short gradients and standard mobility resolution, targeted ultra-high mobility modes and enhanced data processing can further improve structural resolution. Overall, the platform advances the clinical translation of lipidomics by harmonizing annotation criteria, improving reproducibility, and enabling robust phenotyping and marker quantification (e.g., CERT2) in high-throughput settings.

This work delivers a validated high-throughput 4D lipidomics platform that combines automated extraction, microflow UHPLC-TIMS-PASEF-MS, and a curated 4D library (ClinLip Analyte List) to achieve confident annotation and reproducible quantification of hundreds of circulating lipids. In NIST plasma/serum SRMs, 370/364 lipids were confidently annotated and 359 quantified; cross-validation with MRM and inter-lab datasets supports accuracy. The approach enables reliable intra-individual lipidome phenotyping across multiple blood matrices and shows multidien stability, with consistent CERT2 risk scores, highlighting clinical applicability. Future work should expand 4D libraries (standards and spectra), improve mobilogram-based data processing for isomer resolution, optimize ISTD panels across lipid classes and conditions, refine operational procedures (e.g., autosampler residence times for hydrophobic lipids), and drive harmonization efforts to facilitate cross-platform and cross-laboratory comparability and broader clinical translation.

• Isomer resolution limitations under short reversed-phase gradients and standard TIMS resolution: certain TG isomers and phospholipid isobars were co-eluting/partially resolved; lyso sn1/sn2 isomers not baseline-separated by mobility (CCS differences <1 Å2).
• Annotation software constraints: limited delineation of partially separated or co-eluting isomers; required manual inspection for some TG isomers.
• Quantification dependencies on ISTD selection: some class-specific ISTDs show endogenous presence (e.g., PC 17:0_14:1, SM d18:1_12:0) or batch-dependent issues, affecting accuracy without corrections; between-set differences notable for Cer, FAHFA, HexCer; one-point calibration less reliable for ceramides.
• Autosampler-induced decay for highly hydrophobic lipids (TGs, CEs) over long batch times can bias quantification if not controlled; decay compensation by ISTDs may be incomplete for low-abundance species.
• TIMS alone did not increase annotation coverage versus LC; CCS primarily augments confidence rather than separation depth under these conditions.
• Differences versus inter-lab consensus partly reflect higher structural resolution; direct numerical comparability at varying structural levels (species vs molecular species) may be limited.