Overview of Untargeted Lipidomics

Untargeted Lipidomics, also termed Global Lipidome Profiling, represents a holistic analytical strategy for characterizing lipidomes. In contrast to hypothesis-driven targeted approaches, this methodology systematically identifies and quantifies the full spectrum of lipid species within biological specimens without a priori target restrictions. By enabling discovery of novel lipid entities, reconstruction of metabolic networks, and elucidation of lipid-mediated physiological/pathological mechanisms, it serves as a cornerstone for functional lipid research. This review comprehensively examines the methodological foundations, experimental workflows, computational strategies, and translational applications of untargeted lipidomics across biomedical disciplines.

Definition

Untargeted lipidomics represents a comprehensive analytical strategy employing high-resolution mass spectrometry (HRMS) coupled with chromatographic separation techniques (e.g., LC-MS, GC-MS) to systematically characterize molecular profiles of lipid species within biological matrices (e.g., cells, tissues, biofluids). This approach operates without prior hypotheses, enabling holistic profiling of lipid diversity, abundance variations, and dynamic alterations across metabolic pathways. Key objectives include:

• Elucidating lipid diversity: Cataloging structurally distinct lipid classes (e.g., glycerophospholipids, sphingolipids, sterols) and their numerous subclasses (over 10,000 identified variants) within metabolic networks.
• Investigating biological roles: Decoding functional contributions of lipids in fundamental processes such as membrane integrity, energy homeostasis, cellular signaling, and inflammatory responses.
• Uncovering novel entities: Identifying previously unrecognized lipid species (e.g., post-translationally modified lipids, microbe-derived lipids) through data-mining frameworks, followed by functional hypothesis generation.

If you want to know the difference between targeted and non-targeted, please see "Comparative Analysis of Untargeted and Targeted Lipidomics".

Core Characteristics

Comprehensive Analytical Scope

• Broad taxonomic coverage: Detects diverse lipid categories, including phospholipids (PC, PE, PS), sphingolipids (SM, Cer), neutral lipids (TG, DG), sterol derivatives (cholesteryl esters), lyso-lipids (LPC, LPE), glycolipids, and ether lipids (e.g., plasmalogens).
• Polarity adaptability: Addresses compounds spanning hydrophilic (e.g., phosphatidic acid) to highly hydrophobic species (e.g., triacylglycerols) via polarity-switching ionization modes (positive/negative).
• On-selective detection: Employs full-scan acquisition to capture all ionizable species without reliance on predefined databases, enhanced by high-mass-accuracy platforms (Orbitrap, Q-TOF) for precise mass measurements (<1 ppm error) and fragment-ion analysis, minimizing false annotations.
• Isomer differentiation: Utilizes ion mobility spectrometry (IM-MS) to resolve structural isomers (e.g., sn-1 vs. sn-2 fatty acid positional isomers in phospholipids).

High-Throughput Capabilities

• Scalable sample processing: Enables parallel analysis of hundreds to thousands of samples per batch (e.g., epidemiological cohorts) through integrated robotic liquid handling systems, reducing preprocessing time.
• Efficient data pipelines: Leverages software suites (XCMS, LipidSearch) for automated peak alignment, metabolite identification, and statistical normalization. Advanced computational approaches (cloud computing, deep learning) expedite interpretation of large datasets.

Exploratory Discovery

• Biomarker screening: Applies multivariate statistical models (PLS-DA, OPLS-DA) to identify discriminative lipid signatures associated with diseases (e.g., cancer subtypes, metabolic disorders).
• Pathway reconstruction: Integrates multi-omics datasets (transcriptomics, proteomics) to map lipid metabolic networks, elucidating connections between sphingolipid dysregulation and oxidative stress or inflammation.
• Dynamic profiling: Tracks temporal changes in lipidomes under varying physiological conditions (e.g., nutrient deprivation, hormonal stimuli), revealing spatiotemporal regulatory patterns.

Technical Adaptability

• Platform versatility: Supports platform-specific optimizations, such as LC-MS for polar lipids (phospholipids, cholesterol esters) and GC-MS for volatile species (free fatty acids, alcohols). Hybrid techniques (LC-MS/NMR) enhance structural elucidation.
• Matrix resilience: Addresses complex biological matrices (plasma, tissues) through tailored extraction protocols (MTBE/methanol co-solvent systems) and chromatographic adjustments (HILIC columns) to maximize sensitivity.

Translational Relevance

• Clinical diagnostics: Facilitates disease stratification (e.g., distinguishing triple-negative vs. luminal breast cancer) and therapeutic monitoring (e.g., statin-induced lipidome modulation).
• Systems physiology: Constructs "lipid-phenotype" interaction networks, exemplified by correlations between hepatic triglyceride accumulation and inflammatory cytokine profiles in obesity.
• Nutritional science: Evaluates bioactive lipid components (e.g., ω-3 polyunsaturated fatty acids) in functional foods (oils, nutraceuticals) for preventive healthcare.

Technical Challenges

• Structural complexity: Structural isomers (e.g., glycosylated lipids) and post-translational modifications (phosphorylation) complicate unambiguous identification.
• Dynamic range limitations: Signal suppression occurs when highly abundant lipids (triacylglycerols) overshadow trace analytes (e.g., signaling lipids like DAG).
• Database gaps: Current repositories (LIPID MAPS, HMDB) cover ~20% of known lipid diversity, necessitating continuous curation and algorithmic innovation for novel molecule prediction.

Methodological Framework of Untargeted Lipidomics

Untargeted lipidomics employs advanced analytical technologies to comprehensively characterize lipidomes, primarily through the integration of mass spectrometry (MS) with chromatographic separation techniques.

Untargeted lipidomics workflow and data processing (Xiong L et al., 2024).

Mass Spectrometry Platforms

Ionization Methods

• Electrospray Ionization (ESI): Optimized for polar lipids such as phospholipids and sphingolipids.
• Matrix-Assisted Laser Desorption Ionization (MALDI): Enables spatial lipid imaging and analysis of high-molecular-weight species.
• Atmospheric Pressure Chemical Ionization (APCI): Preferred for nonpolar lipids including triglycerides and sterols.

Mass Analyzers

• Quadrupole: Facilitates rapid lipid screening and precursor ion selection.
• Time-of-Flight (TOF): Delivers high mass accuracy for lipid identification and imaging applications.
• Orbitrap: Provides ultra-high resolution for precise molecular weight determination and structural elucidation.
• Ion Trap: Enables multi-stage fragmentation (MSⁿ) for detailed lipid characterization.

Data Acquisition Strategies

• Full Scan Mode: Enables broad-spectrum lipid detection across predefined mass ranges.
• Data-Dependent Acquisition (DDA): Automatically selects abundant ions for sequential fragmentation.
• Data-Independent Acquisition (DIA): Performs parallel fragmentation of mass windows for high-throughput quantification.

Chromatographic Separation Techniques

• Liquid Chromatography (LC): Effective for resolving polar and semi-polar lipids using reversed-phase or hydrophilic interaction columns.
• Gas Chromatography (GC): Ideal for volatile nonpolar lipids post-derivatization (e.g., fatty acid methyl esters).

Technical Synergy

• The combination of MS and chromatography enhances lipid coverage, resolution, and identification confidence. ESI/LC-MS excels in polar lipid analysis, while APCI/GC-MS targets nonpolar species. MALDI-TOF complements these with spatial resolution in tissue imaging.

Workflow of Untargeted Lipidomics

Untargeted lipidomics investigations typically follow a structured workflow comprising four key phases:

Workflow of the untargeted lipidomics approach (Ciccarelli M et al., 2022).

1. Specimen Collection and Stabilization

• Sample Acquisition: Tailor collection protocols to biological matrices (e.g., biofluids, tissues, cells).
• Cryogenic Preservation: Store specimens at ultra-low temperatures (−80°C) to inhibit enzymatic degradation and lipid oxidation, minimizing freeze-thaw cycles.

2. Sample Preparation

• Lipid Extraction: Isolate lipid species using organic solvent systems (e.g., chloroform-methanol).
• Matrix Purification: Remove non-lipid contaminants via solid-phase extraction (SPE) or liquid-liquid partitioning (LLP).
• Derivatization: Chemical derivatization (e.g., fatty acid methylation) enhances MS detectability for specific lipid classes.

3. Analytical Instrumentation

Ionization Method Selection

• Electrospray ionization (ESI) for polar lipids (phospholipids, sphingolipids).
• Atmospheric pressure chemical ionization (APCI) for neutral lipids (triglycerides, sterols).

Mass Analyzer Selection

• Quadrupole for rapid screening.
• Orbitrap/TOF for high-resolution mass measurement.
• Ion traps for multi-stage fragmentation (MSⁿ).

Data Acquisition Strategy

• Full-scan mode for global lipid detection.
• Data-dependent (DDA) or data-independent (DIA) acquisition for targeted fragmentation.

4. Computational Lipidomics

• Preprocessing: Perform peak detection, noise reduction, retention time alignment, and intensity normalization.
• Molecular Annotation: Cross-reference experimental spectra with databases (LIPID MAPS, HMDB) and apply fragmentation pattern analysis.
• Advanced Analytics:
  • Multivariate statistics (PCA, PLS-DA) to identify discriminatory lipids.
  • Pathway enrichment tools (KEGG, Reactome) to map lipid metabolic networks.

Analytical Framework for Untargeted Lipidomics Data

The interpretation of untargeted lipidomics data bridges experimental outputs with biological insights, necessitating rigorous quality control and robust analytical pipelines. Below is a structured overview of key analytical phases:

1. Data Preprocessing: From Raw Signals to Structured Data

Signal Optimization and Peak Extraction

• Peak Detection: Utilize algorithms such as continuous wavelet transform (CWT) or adaptive threshold segmentation (ATS) to identify lipid-specific features (e.g., m/z, retention time) in high-dimensional datasets.
• Dynamic Range Optimization: Apply logarithmic transformation or quantile normalization to harmonize detection sensitivity across lipid abundance levels, addressing disparities between high-concentration species (e.g., triglycerides) and low-abundance lipids (e.g., ceramides).

Noise Reduction and Baseline Correction

• Noise Filtering: Integrate Gaussian smoothing and wavelet denoising to isolate true lipid signals from instrumental noise.
• Baseline Stabilization: Implement sliding-window averaging (e.g., 5-scan window) to mitigate baseline fluctuations and enhance signal-to-noise ratios.

Alignment and Retention Time Calibration

• Cross-Sample Alignment: Anchor retention times using internal standards (e.g., PC 15:0/18:1) and correct drift via local weighted regression (LOESS) or piecewise linear models.
• Isotope Exclusion: Filter isotopic clusters (e.g., ¹³C artifacts) by mass tolerance thresholds (±0.001 Da) and charge-state filtering.

Normalization Strategies

• Global Methods: Total ion current (TIC) or peak area normalization to account for technical variability.
• Advanced Approaches: Machine learning-assisted normalization (e.g., random forest models) or endogenous reference metabolites (e.g., creatinine) to correct systemic biases.

2. Lipid Annotation: From Spectral Features to Molecular Identity

Hierarchical Annotation Framework

• Level 1: Match accurate mass (m/z error<1 ppm) and retention time to databases (e.g., LIPID MAPS, HMDB) for preliminary identification.
• Level 2: Validate using tandem MS (MS/MS) fragmentation patterns (e.g., acyl chain length, double bond positions).

Cross-Database Integration

• LipidBlast: Expand coverage of non-canonical lipids (e.g., microbial species) via spectral similarity scoring.
• Custom Libraries: Incorporate study-specific lipids (e.g., drug metabolites) to enhance relevance.

Fragmentation Pattern Analysis

• Neutral Loss Scanning: Detect structural motifs (e.g., glycerophosphate loss in phosphatidylcholines).
• Multistage MS (MSⁿ): Resolve complex lipid architectures (e.g., plasmalogens).

Machine Learning Enhancements

• Deep Learning Models: Train neural networks (e.g., LipidNet) to predict fragmentation patterns and annotate low-abundance species.
• Confidence Scoring: Assign reliability grades (1–5 stars) based on mass accuracy, fragment matches, and isotope distributions.

3. Statistical Modeling and Biological Interpretation

Multivariate Analytical Approaches

• Unsupervised Learning: Principal component analysis (PCA) or UMAP for outlier detection and subgroup clustering.
• Supervised Learning: PLS-DA or random forests to identify disease biomarkers (e.g., elevated LysoPC 18:0 in hepatocellular carcinoma).

Pathway and Network Analysis

• Metabolic Mapping: Link dysregulated lipids to pathways (e.g., sphingolipid metabolism) via KEGG or Reactome.
• Interaction Networks: Model lipid-protein interactions (e.g., PPARγ binding) using tools like AutoDock Vina.

Dynamic and Dose-Response Modeling

• Metabolic Flux Analysis: Quantify lipid turnover using stable isotopes (e.g., ¹³C-palmitate).
• Dose-Effect Curves: Assess drug impacts (e.g., statins on cholesterol synthesis) through concentration-response modeling.

4. Emerging Technologies and Future Directions

AI-Driven Automation

• AutoML Integration: Optimize preprocessing and modeling via tools like TPOT.
• Federated Learning: Train global models on decentralized datasets while preserving privacy.

Spatially Resolved Lipidomics

• Mass Spectrometry Imaging (MSI): Map lipid heterogeneity in tissues (e.g., tumor margins) using MALDI-TOF or DESI-MSI.

Single-Cell Innovations

• Subpopulation Identification: Leverage dimensionality reduction (e.g., t-SNE variants) to classify lipid-defined cell clusters (e.g., pro-inflammatory macrophages).

Applications of UnTargeted Lipidomics

Untargeted lipidomics serves as a versatile analytical tool in biological and medical research, offering insights across multiple domains:

(1) Disease Studies

• Biomarker Identification: Comprehensive analysis of lipid profiles under healthy and pathological conditions enables the discovery of lipid-based biomarkers linked to specific diseases. e.g., This investigation employed high-resolution untargeted lipidomics to analyze 219 hepatocellular carcinoma (HCC) cases and matched controls from the ATBC cohort. Lipidome remodeling was strongly associated with HCC development, with 34.2% of annotated lipid species (158/462) demonstrating significant correlations to disease risk (FDR<0.05). Notably, monounsaturated fatty acids (MUFA) and MUFA-enriched triglycerides exhibited pronounced associations. Mechanistically, the rate-limiting enzymes SCD1 (stearoyl-CoA desaturase-1) and ceramidase emerged as key regulators, potentially shaping a tumor-permissive microenvironment through lipid compositional alterations that drive cellular proliferation and apoptotic resistance. These findings identify novel lipidomic biomarkers and therapeutic targets for early-stage HCC linked to metabolic dysregulation (Barupal DK et al., 2024).
• Mechanistic Exploration: This approach investigates the involvement of disrupted lipid metabolism in disease progressio. e.g., In this study, the characteristics of abnormal lipid metabolism in breast tissue of patients with non-puerperal mastitis (NPM) were revealed by liquid chromatography-mass spectrometry (LC/MS) untargeted lipid proteomics analysis. Comparing 20 NPM patients with 10 controls, it was found that triglyceride (TGs), phosphatidylethanolamine (PEs) and cardiolipin (CLs) were the main differential lipids, and 35 potential biomarkers were screened out, and the arachidonic acid metabolic pathway was identified as the key regulatory network. Weighted gene co-expression analysis (WGCNA) further correlated lipid group with clinical phenotype, indicating that lipid metabolism disorder may participate in the pathogenesis of NPM through arachidonic acid pathway, which provided a new basis for molecular typing and therapeutic target exploration of the disease (Chen X et al., 2023).

(2) Pharmaceutical Development

• Target Elucidation: By assessing drug-induced alterations in lipid pathways, researchers can elucidate pharmacological mechanisms and potential therapeutic targets. e.g., In this study, the molecular mechanism of salvianolic acid B(Sal B) in improving coronary heart disease (CHD) in rats was revealed through untargeted lipidomics: the CHD model was established by vitamin D3 combined with high-fat diet. After Sal B intervention, myocardial injury markers (LDH, CK-MB, cTn1) and atherosclerotic lipids (LDL-c, ApoB) were significantly reduced, HDL-c and ApoA1 levels were increased, and the myocardium was relieved. A total of 26 kinds of CHD-related differential lipids (such as phosphatidylcholine, lysophosphatidylethanolamine, sphingosine, etc.) were identified in lipidomics, and 22 of them tended to normalize after Sal B intervention. Enrichment analysis showed that they inhibited oxidative stress and lipid peroxidation by regulating glycerophospholipid metabolism, sphingolipid metabolism and arachidonic acid metabolism pathway, thus playing a protective role in CHD, providing theoretical basis for the treatment of cardiovascular diseases by targeting lipid metabolism with natural products.
• Metabolic Profiling: Monitoring lipid variations in drug metabolites aids in evaluating pharmacokinetic properties, optimizing drug efficacy, and minimizing adverse effects. e.g., This research examined the protective role of cyanidin-3-O-glucoside (C3G) against bisphenol A (BPA)-induced hepatic lipid dysregulation. Leveraging lipidomic profiling, results demonstrated that C3G supplementation markedly alleviated dyslipidemia—characterized by elevated low-density lipoprotein cholesterol and triglycerides—and normalized serum levels of liver enzymes (ALT, AST) in BPA-exposed mice. Mechanistic insights revealed that C3G counteracted BPA-driven hepatic lipid imbalances by restoring metabolic pathways of triglycerides (TGs), phosphatidylethanolamine, and phosphatidylcholine, while modulating lipid synthesis-associated gene expression. These findings underscore C3G's capacity to recalibrate lipid homeostasis disrupted by environmental toxins, offering a scientific foundation for employing natural compounds to mitigate endocrine-disrupting chemical toxicity.

(3) Nutritional Science

• Dietary Impact Assessment: Investigations into how dietary lipids modulate metabolic pathways provide critical insights for designing nutritionally balanced interventions. e.g., This study revealed the differential effects of different sources of n-3 polyunsaturated fatty acids (PUFA) (fish oil, krill oil, flaxseed oil) on the lipid composition of Portunus trituberculatus hepatopancreas through 8 weeks of dietary intervention and untargeted lipid proteomics analysis: krill oil significantly increased the enrichment of 20:5n-3 and 22:6n-3 in sn-2 positions of phosphatidylcholine and phosphatidylethanolamine, while fish oil promoted the full chain length (. The metabolic response mechanism of crustaceans to dietary n-3 PUFA was studied, which provided a theoretical basis for lipid nutrition optimization in aquaculture (Yuan Y et al., 2021).
• Health Correlation Analysis: Research on lipid metabolism dynamics enhances understanding of its association with metabolic syndromes, including diabetes and obesity. e.g., This study revealed the difference of lipid metabolism between metabolically healthy morbid obesity (MO) and type 2 diabetes mellitus (T2DM) through untargeted proteomics analysis: ceramide, sphingomyelin and triglyceride increased significantly, while acyl carnitine and bile acid decreased in morbid obesity group; In MO combined with T2DM group, there were further characteristic lipid disorders such as deoxycholic acid, phosphatidylcholine (PC 34:3) and lysophosphatidylinositol (LPI 16:0). Studies have confirmed that specific lipid profiles (such as PC and PE O-38:4) can be used as potential metabolic markers for the progression of MO to T2DM, providing a new basis for the stratification of obesity-related metabolic risks (Bertran L et al., 2024).

(4) Plant and Microbial Lipidomics

• Plant Lipid Functions: Exploration of plant lipid roles in developmental processes and stress adaptation supports advancements in agricultural biotechnology. e.g., This study revealed the regulatory mechanism of Oxathiapiprolin(FRAC action mode group F9), a untargeted lipid proteomic analysis, on lipid metabolism of Phytophthora sojae: after treatment, the subcategories of glycerolipid (30.10%), glycerophospholipid (50.59%) and sphingolipid (7.28%) changed significantly, among which lysophosphatidylcholine (LPC) and phosphatidylcholine (PC). At the specific molecular level, 542 lipid molecules have changed significantly, including 212 glycerides (such as TG and DG) and 167 glycerophospholipids (such as PC and LPC). Orthogonal partial least squares discriminant analysis showed that the metabolic disorder of Cer, TG and some glycerophosphates was the key feature of OXATIAPRIL, which provided an important basis for analyzing its mechanism of targeted inhibition of lipid homeostasis of oomycetes (Liu X et al., 2023).
• Microbial Lipid Dynamics: Studies on microbial lipid interactions during host-pathogen communication contribute to strategies for combating infectious diseases and modulating immune responses. e.g., This study revealed the dynamic remodeling of host lipid metabolism during vesicular stomatitis virus (VSV) infection by untargeted HILIC-IM-MS proteomics: glycerophospholipid and sphingolipid were significantly modified, in which the level of lysophosphatidylcholine (LPC) decreased, ceramide and sphingolipid increased, and were enriched in virus particles. Ontology analysis shows that the change of lipid spectrum reshapes the characteristics of host membrane by regulating membrane bending (LPC depletion) and signal transduction (sphingolipid accumulation), which provides key lipid support for virus germination and replication, and confirms that sphingolipid metabolic reprogramming is the core regulation mechanism of VSV life cycle.

People Also Ask

What are the techniques used in lipidomics analysis?

The most important separation techniques used in lipidomics are RPLC (reverse phased liquid chromatography), NPLC (normal phase liquid chromatography) and HILIC (hydrophilic interaction liquid chromatography).

What are the advantages of untargeted metabolomics?

One primary advantage of untargeted metabolomics is its ability to uncover novel metabolites and pathways without bias.

How to do untargeted metabolomics?

Untargeted metabolomic analyses predominantly employ LC-MS due to its capacity to detect a broad spectrum of chemically diverse metabolites. This approach facilitates the simultaneous identification and quantification of small molecules across varying polarities and molecular weights.

References

1. Barupal DK, Ramos ML, Florio AA, Wheeler WA, Weinstein SJ, Albanes D, Fiehn O, Graubard BI, Petrick JL, McGlynn KA. "Identification of pre-diagnostic lipid sets associated with liver cancer risk using untargeted lipidomics and chemical set analysis: A nested case-control study within the ATBC cohort." Int J Cancer. 2024 Feb 1;154(3):454-464. doi: 10.1002/ijc.34726
2. Ciccarelli M, Merciai F, Carrizzo A, Sommella E, Di Pietro P, Caponigro V, Salviati E, Musella S, Sarno VD, Rusciano M, Toni AL, Iesu P, Izzo C, Schettino G, Conti V, Venturini E, Vitale C, Scarpati G, Bonadies D, Rispoli A, Polverino B, Poto S, Pagliano P, Piazza O, Licastro D, Vecchione C, Campiglia P. "Untargeted lipidomics reveals specific lipid profiles in COVID-19 patients with different severity from Campania region (Italy)." J Pharm Biomed Anal. 2022 Aug 5;217:114827. doi: 10.1016/j.jpba.2022.114827
3. Chen X, Shao S, Wu X, Feng J, Qu W, Gao Q, Sun J, Wan H. "LC/MS-based untargeted lipidomics reveals lipid signatures of nonpuerperal mastitis." Lipids Health Dis. 2023 Aug 8;22(1):122. doi: 10.1186/s12944-023-01887-z
4. Liu R, Jin Y, Liu B, Zhang Q, Li X, Cai D, Tian L, Jiang X, Zhang W, Sun J, Bai W. "Untargeted Lipidomics Revealed the Protective Effects of Cyanidin-3-O-glucoside on Bisphenol A-Induced Liver Lipid Metabolism Disorder in Rats." J Agric Food Chem. 2023 Jan 18;71(2):1077-1090. doi: 10.1021/acs.jafc.2c06849
5. Bertran L, Capellades J, Abelló S, Aguilar C, Auguet T, Richart C. "Untargeted lipidomics analysis in women with morbid obesity and type 2 diabetes mellitus: A comprehensive study." PLoS One. 2024 May 14;19(5):e0303569. doi: 10.1371/journal.pone.0303569
6. Li YP, Wang CY, Shang HT, Hu RR, Fu H, Xiao XF. "A high-throughput and untargeted lipidomics approach reveals new mechanistic insight and the effects of salvianolic acid B on the metabolic profiles in coronary heart disease rats using ultra-performance liquid chromatography with mass spectrometry." RSC Adv. 2020 May 1;10(29):17101-17113. doi: 10.1039/d0ra00049c
7. Yuan Y, Xu F, Jin M, Wang X, Hu X, Zhao M, Cheng X, Luo J, Jiao L, Betancor MB, Tocher DR, Zhou Q. "Untargeted lipidomics reveals metabolic responses to different dietary n-3 PUFA in juvenile swimming crab (Portunus trituberculatus)." Food Chem. 2021 Aug 30;354:129570. doi: 10.1016/j.foodchem.2021.129570
8. Liu X, Li C, Chen Y, Xue Z, Miao J, Liu X. "Untargeted lipidomics reveals lipid metabolism disorders induced by oxathiapiprolin in Phytophthora sojae." Pest Manag Sci. 2023 Apr;79(4):1593-1603. doi: 10.1002/ps.7334