Decoding the Oxylipin Chemical Space Using Ion Identity Molecular Networking

Abstract

Oxylipins are bioactive lipid mediators that play key roles in biological and pathological processes. Their remarkable chemical diversity makes their identification by untargeted LC-MS/MS analyses challenging. To date, effective solutions for their comprehensive characterization remain unavailable. Here, we present the first implementation of the recently refined Ion Identity Molecular Networking (IIMN) strategy to map the chemical space of oxylipins, together with a systematic evaluation of factors that hinder accurate annotation in MS/MS datasets. Building on recent mzmine software developments, we implemented a fully local strategy to perform the IIMN analysis without requiring web platforms or external tools. We established a high-quality MS/MS spectral library from 67 commercially available oxylipin standards using LC-MS data obtained in data-dependent acquisition mode. Integrating the detailed characterization of ion species generated during electrospray ionization into IIMN reduced network complexity. Across configurations, the modified cosine algorithm proved most effective for separating full-length from cyclized forms and for clustering oxylipins through structurally coherent relationships. Application of the IIMN workflow to mouse spleen extracts, in combination with our in-house and publicly available experimental MS/MS libraries, enabled the organization of oxylipins into molecular families, facilitating their structural characterization and the discovery of novel species. Although manual curation remained necessary for certain coeluting isomers and ambiguous fragments, the IIMN-based approach significantly improved network interpretability and understanding. Overall, this study establishes IIMN as a robust bioinformatic tool for decoding oxylipin diversity and provides a successful strategy for mapping their chemical space, characterizing them within samples, and discovering novel mediators in biological matrices. The new combined reference spectral library has been made publicly available and will serve as a valuable resource for future redox lipidomics research.

Introduction

Oxylipins are a family of bioactive lipid mediators found in almost all organisms. In humans, they are involved in key physiological processes such as the regulation of blood flow, smooth muscle contraction, and paracrine and autocrine signaling, and they play a central role in the modulation of inflammation processes. , Moreover, oxylipins have been implicated in various pathological conditions, including responses to ischemia–reperfusion injury and diseases such as cancer and metabolic and neurodegenerative disorders. These mediators originate from oxidative modifications of polyunsaturated fatty acids (PUFA)including linoleic acid (LA, C18:2n-6), α-linolenic acid (ALA, C18:3n-3), dihomo-γ-linolenic acid (DGLA, C20:3n-6), arachidonic acid (AA, C20:4n-6), eicosapentaenoic acid (EPA, C20:5n-3), and docosahexaenoic acid (DHA, C22:6n-3), among others. PUFA are highly susceptible to oxidation through enzymatic and non-enzymatic pathways. Enzymatic oxidation primarily involves lipoxygenases (LOX), cyclooxygenases (COX), cytochrome P450 monooxygenases (CYP), and soluble epoxide hydrolase (sEH). Non-enzymatic mechanisms, such as auto-oxidation, , produce non-enzymatic oxygenated PUFA (also referred to as NEO-PUFA) that are associated with oxidative stress and are recognized as valuable markers in this context, with isoprostanes being a known example.

Oxylipins exist in various forms, including free (non-esterified oxylipins) as well as esterified to complex lipids, such as glycerophospholipids, glycerolipids (e.g., triacylglycerols), and sterol lipids (e.g., cholesteryl esters). Both forms are controlled by a turnover pathway that regulates oxylipin bioavailability. , The wide array of oxylipins includes full-length and cyclized variants. Some of these can undergo rapid metabolism to produce short-chain (truncated) forms or can be conjugated with other molecules (e.g., glucuronic acid, methyl groups, or amino acids). Oxylipins include octadecanoids, eicosanoids, and docosanoids, which are named after the length of their carbon chains (18-carbon, 20-carbon, and 22-carbon, respectively). They generally exhibit high structural diversity and extensive chemical heterogeneity due to the presence of different oxygen-containing functional groups, including alcohols (−OH), hydroperoxides (−OOH), epoxides (−O−), aldehydes (−CHO), ketones (O), and carboxylic acids (−COOH). These functional groups also give rise to numerous isobaric and isomeric species. , These remarkable chemical complexities result in an extensive range of solubility and polarity, among other physicochemical properties.

The combination of all these factors dramatically complicates the analysis of oxylipins and constitutes a real pitfall for researchers who must take great care when isolating, analyzing, and identifying them from biological samples. The current scientific consensus is that these difficulties limit our understanding of the biology of oxylipins. Technical issues are mainly related to their low abundance (nM–pM range) in biological samples, their susceptibility to chemical and thermal degradation, and the lack of standardized protocols, which introduces variability across studies. ,,

Furthermore, reversed-phase (ultra)­high-performance liquid chromatography (RP-(U)­HPLC) coupled to electrospray ionization tandem mass spectrometry (ESI-MS/MS) introduces additional analytical considerations. Most LC-MS/MS methods targeting oxylipins rely on low-resolution MSe.g., triple quadrupole (QqQ)which provides high sensitivity and specificity, but is not well suited for comprehensive structural characterization. To address this, untargeted high-resolution tandem mass spectrometry (HR-MS/MS) platforms (e.g., Orbitrap or quadrupole time-of-flight (QTOF)) are employed to broaden coverage, generate high-quality lipidomics data across major oxylipin species in biological samples, and provide detailed structural information. ,, However, several limitations arise when it comes to identifying oxylipins. These include extensive coeluting isomers, MS¹ spectral complexity from adducts, in-source fragmentation (ISF), and clusters of non-volatile salts formed in the ESI source. , There is also frequent co-isolation of precursors in MS/MS experiments, leading to chimeric spectra and making structural elucidation difficult. ,

Additionally, researchers aiming to obtain a comprehensive view of oxylipin diversity in biological matrices face further challenges. These difficulties include the scarcity of MS/MS spectral libraries, limited availability of standards, inconsistent nomenclature, and incomplete knowledge of fragmentation patterns and mechanisms. Despite these limitations, significant progress has been made in recent years through the development of databases, libraries, and analytical pipelines.

In this regard, LIPID MAPS, the most sophisticated open lipid-focused database, has recently launched a library of experimental MS/MS spectra (including low- and high-resolution) for standard oxylipins. Moreover, specific supporting software tools that leverage rule-based fragmentation predictions and in silico MS/MS spectral libraries have emerged to simplify and automate the annotation process. , From these, the open-source software LPPtiger2 is the only platform designed specifically for oxidized lipids identification (i.e., esterified oxylipins). Other examples include Lipid Data Analyzer 2 (LDA), Lipostar2, MS-DIAL 5; however, these general-purpose suites are not optimized for oxidized lipids and typically include only a limited subset of oxylipins. Furthermore, while LOBSTAHS is specifically tailored for oxidized lipids, its identification relies solely on MS¹-level data, lacking MS/MS information. In addition, some analytical strategies for oxylipin annotation have combined specific chemical derivatization approaches with MS/MS fragmentations. ,

Despite these resources and contributions, full oxylipin identification is still a long way ahead. There is an urgent need for efficient structural annotation strategies that both complement existing approaches and integrate multiple layers of information, including not only MS¹ ion species profiles but also retention times (RT), MS/MS fragmentation patterns, physicochemical properties, and metabolic reactions.

Innovative examples supporting complex annotation include molecular networking (MN), machine-learning tools, MS search/query tools (e.g., MassQL, MASST), RT prediction models (QSPR), and metabolic network analysis platforms. So far, only a subset of these strategies has been applied specifically to oxylipins, with MN-based workflows representing some of the first reported examples in this area. In parallel, pathway-mapping tools have also been applied to interpret oxylipin profiles (e.g., LimeMap, WikiPathways, and PathVisio).

Among these strategies, an increasing number of researchers are directing their attention to MN. This promising tooltraditionally applied within the context of natural productsuniquely enables exploration of the chemical space by visualizing MS/MS data as networks of structurally related compounds, revealing relationships based on fragmentation patterns. It relies on the assumption that structurally similar molecules yield similar MS/MS spectra. Spectral similarity is typically computed using algorithms such as the modified cosine, DreaMS, and MS2DeepScore. The resulting network comprises nodes (MS/MS spectra) and edges (spectral similarities), organizing compounds into clusters or molecular families.

Over time, three main types of MN tools have emerged. The first was the classical molecular networking (CLMN), which connects MS/MS spectra solely by spectral similarity. Feature-based molecular networking (FBMN) subsequently integrated chromatographic features such as mass-to-charge ratio (m/z), RT, and isotopic patterns to define each node, improving isomer resolution. Most recently, in 2021, Schmid et al. introduced Ion Identity Molecular Networking (IIMN) as an extension of FBMN that further improves MN by grouping the ion species of the same molecule generated under the ESI source (i.e., ion identities), based on RT match and peak-shape correlation. This approach succeeded in reducing the size and complexity of the network and addresses a key limitation of earlier MN methods, in which adducts and/or clusters appeared as separate, disconnected nodes due to differing fragmentation behaviors.

Although MN originated in natural products, its use has gradually expanded to other areas, including lipidomics, although with few examples. − Specifically, for oxylipins, however, MN applications remain nascent. To date, only two recent studies have been reported in the literature. Watrous et al. employed a CLMN-style workflow incorporating RT prior to FBMN construction with an MS/MS library of 225 commercial oxidized standards. In contrast, Elloumi et al. built a library from commercial standards and in-house oxidized mixtures generated by in vitro PUFA auto-oxidation and generated FBMN. Both efforts enabled the tentative structural characterization of novel oxylipins in human plasma and serum samples. In these cases, the generation of the MN was carried out within the Global Natural Products Social Molecular Networking (GNPS) web platform.

In this study, we present the first implementation of IIMN to oxylipins analysis, aimed at improving their characterization in untargeted LC-MS/MS datasets. We provide a fully detailed, step-by-step workflow designed to guide researchers through the practical implementation of IIMN. Because this is a novel application to oxylipins, we placed particular emphasis on understanding the tool in depth and evaluating its impact on annotation accuracy and network interpretability. We provide both a practical protocol and the rationale behind it to facilitate broader adoption in biological samples.

To this end, we constructed a customized HR, accurate mass MS/MS spectral library that also captures ion identity specific fragmentation behaviors. For the first time in oxylipin analysis, we generated IIMN using recent mzmine developments, specifically its native IIMN module introduced in 2022. We further present the first ion identity profile for oxylipins under our UHPLC-ESI-QTOF-MS conditions, which increases confidence in structural assignments and minimizes network redundancy. Although developed under our specific setup, the characterization framework and step-by-step mzmine workflow are readily transferable to other laboratories and platforms.

Building on these resources, we then leveraged the full information from the MS spectral data acquired under ESI(−) mode to interpret IIMN results. We conducted a comprehensive evaluation of the key aspects of the workflow, including the occurrence and fragmentation patterns of adducts, clusters, and signals from ISF; the identification of isobaric and isomeric interferences; and the selection of the most suitable algorithm for IIMN construction by assessing the three options available in mzmine.

Furthermore, to showcase the utility and robustness of the workflow, we have performed a practical application to mouse spleen samples, integrating our in-house MS/MS spectral library with publicly available resources (i.e., LIPID MAPS Oxylipin Standard Database) to comprehensively map oxylipins and organize them into molecular families. Given the scarcity of dedicated bioinformatics tools for oxylipin annotation, this integration provides validated, practical resources that streamline structural characterization and enable reliable analysis of real samples. Our results demonstrate that IIMN accurately expands the detectable chemical space of structurally related oxylipins in complex matrices and that combining ion identity information with MS/MS similarity improves structural elucidation and clarifies relationships among oxylipins.

Finally, we provide a new HR-MS/MS reference library with a broad coverage of oxylipin ion species available in repository, built by integrating public repositories and databases with our experimental acquisition. It will be of great help for researchers interested in oxylipin discovery.

Experimental Section

Oxylipin Compound Library

The oxylipin library comprises 67 commercially available standards of free oxylipins derived from PUFA, representing a broad range of oxylipin subclasses. Specifically, it includes 9 oxylipins derived from linoleic acid (LA, C18:2n-6), 2 from α-linolenic acid (ALA, C18:3n-3), 34 from arachidonic acid (AA, C20:4n-6), 12 from eicosapentaenoic acid (EPA, C20:5n-3), 4 from dihomo-γ-linolenic acid (DGLA, C20:3n-6), and 6 from docosahexaenoic acid (DHA, C22:6n-3). The first two groups represent C18-PUFA-derived oxylipins (octadecanoids), the following three groups correspond to C20-PUFA-derived oxylipins (eicosanoids), and the final group consists of C22-PUFA-derived oxylipins (docosanoids). The oxylipin library encompasses a total of 56 full-length and 11 cyclized oxylipins, designed to maximize the coverage of chemical diversity within the oxylipin family. An oxylipin standard mixture was prepared and analyzed (Table S1). Detailed information regarding reagents and solutions, is provided in the Supporting Information.

Biological Sample Preparation

Due to previous experience in our group showing that spleen tissue contains detectable levels of free oxylipins, these samples were selected as the biological sample. Pooled spleen samples from healthy mice were processed for the extraction of free oxylipins using solid-phase extraction (SPE), with the aim of validating the proposed workflow in a biological matrix. The animal study was reviewed and approved by CBMSO Ethics Committee for Animal Experimentation in Madrid, the Consejo Superior de Investigaciones Científicas (CSIC) ethics committee, and the Regulatory Unit for Animal Experimental Procedures from the Comunidad de Madrid. Detailed information about the SPE protocol for free oxylipin extraction is provided in the Supporting Information.

RP-UHPLC-ESI­(−)-QTOF-MS Data Acquisition

Reversed-phase ultra-high-performance liquid chromatography (RP-UHPLC) coupled with quadrupole time-of-flight mass spectrometry (QTOF-MS) equipped with a dual Agilent Jet Stream (AJS) electrospray (ESI) source was used for data acquisition (Agilent 1290 Infinity II coupled to Agilent 6546 QTOF), following the oxylipin method described by Camunas-Alberca et al. Detailed experimental parameters for RP-UHPLC-ESI(−)-QTOF-MS analysis are provided in the Supporting Information.

High-Resolution Tandem Mass Spectrometry (HR-MS/MS) - Spectral MS/MS Dataset Generation

To guarantee maximum coverage, an HR-MS/MS spectral dataset in ESI(−) was generated using the automated iterative exclusion (IE) data-dependent acquisition (DDA) strategy (IE-ddMS²-top3). Standard mixture (comprising 67 oxylipin standards) and samples were injected three times. With this strategy, precursors selected for MS/MS fragmentation in one run were excluded from selection in subsequent runs. Exclusion lists for iterative MS/MS fragmentation were generated from signals detected in blank samples.

In addition, a second spectral MS/MS dataset was acquired following a predefined dynamic list of target ions (m/z) and RT to enhance the analysis of known adducts, clusters, and ISF for each oxylipin from our compound library.

The MS/MS parameters were set as follows: scan rate of 3 scans/s, a MS¹ mass window of m/z 250–450, an MS/MS scan range 50–450 Da, a narrow MS/MS isolation width (∼1.3 amu), and a 1,000-count threshold. For iterative analyses (i.e., IE-ddMS²-top3), 3 acquisitions were performed at a collision energy (CE) of 20 eV, charge state set at 1, and common isotope mode. The same CEs were used for targeted MS/MS analyses. The selection of these parameters was based on prior experience and relevant publications. ,, Acquisition of MS raw data is also described in detailed in the Supporting Information.

Ion Identity Molecular Networking (IIMN) in mzmine Software

QTOF-MS raw data files (in “.d” format) were converted to .mzML files using MSConvert (Proteo Wizard, version 3.0.23048). MS¹ and MS/MS data from standard mixtures and biological samples were processed using mzmine (version 4.7.8), , following a streamlined workflow optimized for each dataset. Feature deconvolution, feature grouping, blank subtraction, and IIMN were applied to explore adduct relationships and structural connectivity using modified cosine, DreaMS, and MS2DeepScore algorithms. For the biological sample, the workflow was adapted incorporating additional smoothing and duplicate filtering steps. Followed workflows are described in detail in the Supporting Information.

MS/MS Spectral Libraries Generation

Three MS/MS spectral libraries, in .mgf, .json, and .csv formats, were generated and used for oxylipin annotation in biological samples. Two of the databases corresponded to experimental datasets, and one to the public LIPID MAPS database, as described below:

• i. CEMBIO_standard_mix_20eV, generated from a mixture of 67 oxylipin standards fragmented at a CE of 20 eV for all deprotonated adducts and other ion identities, using the mzmine module for library generation (313 spectra).
• ii. CEMBIO_standard_individual_20eV, built from individual injections of 75 oxylipin standards as deprotonated adducts fragmented at 20 eV CE (75 spectra). This library includes all compounds present in Table S1.
• iii. LIPID_MAPS_filtered, derived from the LIPID MAPS Oxylipin Standard Database, includes experimental HR-MS/MS spectra of deprotonated adducts for 290 oxylipin standards (494 spectra). The library consists of commercial pure standards together with in-house in vitro generated oxidized mixtures (including data reported by Elloumi et al.) acquired on HR instruments (i.e., Thermo Fisher Orbitrap ID-X Tribrid, Q Exactive HF, Orbitrap Exploris 240, and Q Exactive Focus). The file preserves full metadata for each spectrum (e.g., oxylipin name, formula, exact mass, and CE, among others).

All three MS/MS spectral libraries were merged to generate a final file (merged_library), hereafter referred to as the oxylipin reference MS/MS spectral library. Libraries include associated metadata such as compound name, exact mass, adduct, SMILES, instrument, and sourceindicating the originating library for each MS/MS spectrumamong other fields. The pipeline used to generate the spectral libraries is described in detail in the Supporting Information.

Workflow Overview

As illustrated in Figure , we established an LC-ESI(−)-MS workflow that integrates MS¹ and MS/MS information to build an IIMN resource and a reference MS/MS library for oxylipins.

1.

Workflow developed for Ion Identity Molecular Networking (IIMN) using LC-MS-based oxylipin analysis. ^†LIPID MAPS Oxylipin Standard Database combines pure chemical standards and characterized oxidized mixtures generated in vitro. Data processing was performed using mzmine software version 4.7.8. DDA, data-dependent acquisition; ESI, electrospray; ISF, in-source fragmentation; MS, mass spectrometry; MS/MS, tandem mass spectrometry; QTOF, quadrupole time-of-flight; RP, reversed-phase; UHPLC, ultra-high-performance liquid chromatography.

First, an oxylipin standard mixture was analyzed by RP-UHPLC-ESI(−)-QTOF-MS to acquire HR-MS¹ data, which were used to build an MS¹ database containing accurate m/z and reproducible RT values. Using the same standards, a thorough ion identity profileincluding adducts, clusters, and signals derived from ISFwas developed and appended to the MS¹ database. MS/MS data were then collected in negative ESI mode using DDA with iterative exclusion (IE- ddMS²-top3) and targeted DDA. Raw data files were processed in mzmine (version 4.7.8) to generate an MS/MS spectral library with associated metadata.

Subsequently, Ion Identity Networks (IIN) were constructed by linking features that coeluted within a shared RT window, exhibited high peak-shape correlation, and matched the expected mass differences for defined ion identities. These IIN were integrated into a molecular network based on MS/MS spectral similarity, yielding a robust IIMN. Workflow parameterssuch as feature detection, grouping, correlation, and blank subtractionwere optimized using the oxylipin standards. Finally, we combined the in-house spectral collections with the LIPID MAPS Oxylipin Standard Database to assemble a reference MS/MS library. We applied the IIMN workflow, together with this library, to biological samples to map oxylipins and facilitate their structural characterization and discovery in complex matrices.

Data Availability

All study files are deposited in Zenodo (doi: 10.5281/zenodo.17176661). The record includes raw LC-MS/MS data in .mzML format; mzmine batch files (.mzbatch) with the exact processing parameters; and the curated MS/MS spectral libraries in .mgf, .json, and .csv formats. A detailed README describes the folder structure and file naming. The Supporting Information provides step-by-step instructions for loading the materials into mzmine (version 4.7.8 or later) to enable one-click reproduction of the workflow.

Lipid Nomenclature

The lipid nomenclature followed the classification system established by LIPID MAPS (https://www.lipidmaps.org/lipid_nomenclature) and the shorthand nomenclature outlined in their guidelines (https://www.lipidmaps.org/lipid_nomenclature).

Results and Discussion

Evaluation of Oxylipins under Untargeted LC-ESI(−)-MS Conditions

Comprehensive detection of oxylipins in complex contexts requires the identification of their unique chemical characteristics. This called for a thorough and meticulous examination of our oxylipin compound library within the context of our experimental LC-ESI(−)-MS conditions.

Therefore, we first evaluated general information, related to the reverse-phase chromatographic behavior observed through RT, as well as the formation of ion species under ESI(−). Second, to determine whether oxylipins may be further experimentally distinguished, we also inspected MS¹ data for extensive isobaric/isomer overlaps and signals derived from ISF. Finally, since IIMN is based on MS/MS fragmentation pattern matching, experimental MS/MS data acquisition is crucial for the generation of realistic and reliable MS/MS spectral libraries. Thus, the evaluation of the acquisition mode, and the parameters and CE that influence the outcome, was carefully examined.

Reversed-Phase Chromatographic Separation: Elution Profile of Oxylipins

To achieve optimal resolution of oxylipins, it is critical to perform the correct chromatographic separation. Therefore, the 67 oxylipins present in the commercial standard mixture were separated using RP-UHPLC, with 11 cyclized oxylipins eluting at shorter RT than their 56 linear counterparts (Figure S1). This separation was due to the higher polarity of cyclized oxylipins relative to the less polar linear species. LC-MS analysis revealed that oxylipins eluted between 2.0 and 16.0 min on the LC gradient, exhibiting consistent peak shapes.

Generation of Adducts, Clusters, and In-Source Fragmentation in ESI(−): Distribution of Patterns Based on the Oxylipin Subclass

During LC-ESI(−)-MS analyses, the deprotonation process is not the only ionization event that occurs in the ESI source. Background ionssuch as metals, impurities, and coeluting speciescan react with oxylipins to form adducts, clusters, and dimers, which cause data redundancy. However, this redundancy, as well as other ions generated by ISF, can be exploited to help make sense of the IIMN data. This is mainly based on their relative mass differences (see Figure S2 for an example). The chemical structure of specific oxylipin subclasses, along with the applied LC-MS method, imposes intrinsic and method-dependent constraints on adduct, cluster, and ISF formation. Although ISF can complicate MN interpretation, it is useful when carefully considered for each compound class, experimental setup, and sample matrix, as exemplified in this study.

Therefore, the preliminary characterization of ion species profiles was crucial in order to link ions from the same molecule and improve network clarity using IIMN. This characterization involved verifying the accurate mass (m/z) of ion species in MS¹ spectra within 10 ppm of the theoretical values for potential ion identities, ensuring consistent RT values, and manually assessing chromatographic peak shape for symmetry and integrity.

Ion identity profiles from our library of 67 oxylipins revealed that they produce abundant [M–H]⁻ ions (up to 85%) via the deprotonation of the carboxylic functional group, as illustrated in Figure . Then, cluster ions attributed to non-volatile salts from the mobile phases were frequently observed as repeating units (n = 1–2). These included significant abundances of sodium acetate clusters with the general structure of [M–H+(CH₃COONa) _n ]⁻ (up to 24% for n = 1 and 7% for n = 2) and sodium nitrate clusters with the general structure of [M–H+(NaNO₃)]⁻ (up to 4%). To our knowledge, the latter cluster has not been previously reported, making it particularly noteworthy. The formation of low-abundance chloride and nitrate adduct ionsrepresented as [M–Cl]⁻, [M+NO₃]⁻–, and anionic oxylipin-sodium ion complexesreferred to as [M–2H+Na]⁻were detected for both full-length and cyclized oxylipins. These oxylipin-metal cation complexes are formed though deprotonation of both the carboxylic acid moiety and deprotonation of a bis-allylic position between double bonds of the acyl chain, yielding a stabilized carbanion (Figure ). Notably, their formation was particularly abundant (up to 24%) in oxylipins containing a second acidic group, which is preferentially deprotonated (e.g., LTE₄ and 20-COOH-LTB₄).

2.

Ion identity profiles for (A) full-length and (B) cyclized oxylipins from our oxylipin compound library, including in-source fragmentation (i.e., neutral losses of water molecules), adducts, and clusters. Oxylipins are grouped by the substitution pattern of their oxygenated functional groups and, for cyclized species, by their ring substitution pattern. Abbreviations are listed in the Glossary of this article due to their length.

In certain cyclized oxylipins having hydroxyl groups on the cyclopentane ringsuch as PGD₁, PGD₂, PGD₃, PGE₁, PGE₂, and PGE₃water molecules can be eliminated yielding the corresponding water-loss fragment ions [M–H–H₂O]⁻ and [M–H–(H₂O)₂]⁻ (up to 16 and 8%, respectively). The neutral loss of water is favored by the stabilization of the resulting double bond through conjugation with a carbonyl group (CO) within the ring structure (Figure B). In the case of PGJ₂, which has a hydroxyl on its side chain instead of the ring, the [M–H–H₂O]⁻ ion was observed, resulting from side-chain dehydration and formation of a resonance-stabilized product, eventually leading to 15-deoxy-Δ^12,14-PGJ₂, which no longer undergoes ISF water loss. In contrast, PGF₂ contains two hydroxyl groups on the ring but not a carbonyl group; this structural feature makes the dehydration mechanism unfavorable.

Full details for detected ion identities for each oxylipin are compiled in Table S2 of the Supplementary MS1 Database.

Isobaric and Isomeric Overlaps of Oxylipins during Manual MS Data Inspection

Despite the high resolving power and mass accuracy of modern HR-MS instruments, accurate characterization of oxylipins remains challenging due to isobaric and isomeric interferencesfurther aggravated by coelutionand ISF.

Oxylipins typically fall within a narrow m/z range of ∼250–450 Da, reflecting their high structural similarity as derivatives of PUFA, with differences mainly arising from their degree of oxidation and unsaturation. This narrow mass range favors the occurrence of type II isobaric interferences, where the M+2-isotopologue of a highly unsaturated species overlaps with the M+0 ion of a structurally similar molecule bearing one fewer double bond (e.g., 9-HETE and 15-HETrE; see Figure S3.A).

In addition to isobaric species, oxylipin data sets often contain a high number of structural isomers (see Figure S4), which exhibit identical exact masses and isotopic patterns. The high degree of structural similarity among oxylipins frequently leads to the overlap of closely related isomeric species, further complicating confident annotation (see also Figure S3A).

Although an optimized chromatographic method was used to achieve broad coverage of oxylipins, coelution remains a common pitfall of structural variants. In some cases, isomers with distinct structural features (e.g., cyclized versus full-length forms) can be differentiated based on RT (e.g., PGD₂ and LXA₄, with identical molecular formula and [M–H]⁻ m/z 351.2177, eluting at RT 7.88 and 8.31 min, respectively) (Figure S4). However, other structural isomers such as 8-HETE and 12-HETE coelute and exhibit nearly identical MS/MS spectra, making them indistinguishable by chromatography and MS/MS alone (Figure S3A).

Furthermore, ISF poses an additional source of misannotation if not properly accounted for. ISF can result in signals that mimic authentic isomeric or isobaric species, particularly when different ion forms are present in the MS¹ spectrum. Common ISF products include dehydrated ions such as [M–H–H₂O]⁻ and [M–H–(H₂O)₂]⁻, which coelute with their parent compounds and display identical fragmentation patterns. In contrast, genuine oxylipins that differ by one or two water losses typically elute at distinct RT due to polarity differences, despite sharing similar MS/MS spectra. This is exemplified by PGD₂ (m/z 351.2178, RT 7.86 min), PGJ₂ (m/z 333.2071, RT 9.36), and 15-deoxy-Δ^12,14-PGJ₂ (m/z 315.1965, RT 12.32 min). In this case, PGD₂ undergoes ISF to yield PGJ₂ and 15-deoxy-Δ^12,14-PGJ₂ as single and double dehydrated forms (Figure S5).

In summary, a thorough understanding of oxylipin structural diversity is essential for accurate identification, particularly when dealing with coeluting isomers and ISF artifacts. While IIMN can aid in distinguishing certain type II isobaric overlaps and differentiate between various isomeric species, it cannot discriminate coeluting isomeric compounds, underscoring the continued importance of expert manual curation in oxylipin analysis.

Data-Dependent Acquisition (DDA) MS/MS Parameters

Several DDA strategies were tested to optimize fragmentation coverage of the oxylipin compound library, ensuring inclusion of all oxylipins and their relevant ion species. DDA methods can be broadly classified into intensity-dependent (e.g., ddMS²-topN), hybrid (e.g., iterative exclusion with dynamic inclusion lists), and list-dependent approaches (e.g., targeted DDA). ,

This classification highlights the balance between untargeted coverage and targeted specificity in MS/MS acquisition strategies. To evaluate their suitability for oxylipin profiling, all approaches were tested with the aim of maximizing coverage of both compounds and their relevant adducts. The most effective strategy for building the spectral library combined three elements: (i) automated iterative exclusion DDA (IE-ddMS²-top3) with a static exclusion list derived from blanks to avoid background signals, (ii) targeted DDA focused specifically on deprotonated adducts to ensure acquisition of MS/MS spectra for the most abundant ion species, and (iii) targeted DDA informed by previously characterized ion identities. This combined approach provided broad precursor coverageincluding various ion identitieswhile ensuring acquisition of the predominant ion ([M–H]⁻) for each compound.

The optimization of DDA parameters is essential for ensuring high-quality MS/MS spectral coverage for accurate oxylipin annotation. Thus, after defining a threshold of 1,000 counts, key parameters were carefully optimized and a static exclusion list was established, comprising a number of m/z values from blank solvent samples that should never be selected for MS/MS. Further details are explained in the Supporting Information.

First, the mass range was restricted to 250–450 m/z, which encompasses the majority of oxylipins according to the LIPID MAPS Structure Database (LMSD) and excludes higher-mass species such as dimers. A scan rate of 3 spectra/s was applied to increase MS/MS event density across chromatographic peaks.

A narrow isolation window (∼1.3 Da) was used to minimize co-isolation of adjacent ions, improving fragmentation specificity and overall spectral clarity. Because many oxylipin species coelute, they can still lead to chimeric MS/MS spectra (i.e., spectra containing fragment ions from multiple precursor ions) during untargeted fragmentation experiments (Figure S3B,C). In the case of type II isobaric interferences, coelution is expected to produce chimeric MS/MS spectra. However, under our experimental conditions, the limited contribution of interfering isotopologues resulted in spectra that appeared visually clean. In contrast, when isomers coelute within the same chromatographic peak, the resulting MS/MS spectra are fully chimeric, substantially complicating accurate annotation.

Finally, the CE for optimal MS/MS fragmentation was adjusted. Upon collision-induced dissociation (CID) in ESI(−), oxylipins typically generate fragment ions, including abundant neutral losses (e.g., CO₂, H₂O, and −CH₂CH₂) that are common and specific to the structural oxylipin family. On the other hand, ions arising from acyl chain cleavages are of low abundance and are characteristic to the modification type and position along the hydrocarbon chain. For example, in oxylipins bearing hydroxyl groups, such as 15-HETE, the most abundant fragments typically result from the cleavage of the carbon-carbon bond adjacent to the hydroxyl group ,− : m/z 219.1390, 175.1491, and 113.0971 for 15-HETE (Figure S6).

As both fragmentation efficiency and spectral information depend heavily on the selected CE, it is important to optimize this parameter. A low CE favors the generation of common neutral losses, whereas an excessively high CE leads to overfragmentation, which reduces the intensity of both neutral losses and diagnostic ions. We evaluated three CE settings: at 10 eV, neutral losses were prominent but position-specific fragments were weak; at 40 eV, the overall spectral quality declined due to non-specific fragmentation. A CE of 20 eV provided the optimal balance, yielding high-intensity signals for both types of fragments and was thus selected for MN of oxylipin standards and a biological sample.

Fragmentation Behavior of Adducts, Clusters, and In-Source Fragmentation

During ESI, oxylipins often produce multiple ion species or ion identitiesincluding adducts, clusters, and ISFthat frequently exhibit distinct MS/MS fragmentation behaviors. The specific voltage and temperature conditions of the ESI modulate the internal energy and coordination geometry of ion-ligand complexes, resulting in significant and often unpredictable variations in fragmentation patterns and fragment intensities. Figure illustrates this phenomenon for a representative full-length oxylipin, directly linking ion identity formation in the source with their subsequent fragmentation. First, Figure A depicts the ion identity profile detected in MS¹, showing the specific mass offsets relative to the neutral molecule. These characteristic mass shifts serve as the key parameters for grouping features in the IIMN workflow. Subsequently, Figure B displays the distinct MS/MS yielded by these ion identities. If not properly addressed through IIN, then such spectral heterogeneity can artificially separate or misgroup structurally related molecules in molecular networks.

3.

MS¹ and MS/MS behavior of 15-HETE ion identities. (A) MS¹ precursor ion species and their specific mass offsets relative to the neutral mass [M] of 15-HETE. These values represent the adducts, clusters, and signals from in-source fragmentation (ISF) used for Ion Identity Networking. Mass differences are positive for adducts and clusters and negative for deprotonated ions and ISF. (B) MS/MS spectra of 15-HETE ion species generated under negative electrospray ionization ESI(−). Spectra on the left represent those ion identities that retained structural information at a collision energy (CE) of 20 eV, using the [M–H]⁻ ion as a reference. Spectra on the right represent those that did not retain structural information at a CE of 20 eV, also using [M–H]⁻ as a reference. Common MS/MS neutral losses (e.g., H₂O and CO₂) shared across the HETE oxylipin family are highlighted in gray, while MS/MS position-specific fragments unique to 15-HETE are shown in blue. HETE, hydroxy-eicosatetraenoic acid.

Using 15-HETE as a representative case study, we compared CID-based MS/MS spectra across its major ion identities (Figure B). While the deprotonated ion [M–H]⁻ and [M–H+(CH₃COONa)]⁻ exhibited similar fragmentation patterns, [M–H–H₂O]⁻ and [M–Cl]⁻ shared only a few common neutral losses and lacked position-specific fragments. One of the dissociation pathways from the [M–Cl]⁻ and [M+NO₃]⁻ adducts leads to the loss of a chloride ion (Cl^–) or a nitrate ion (NO₃ ^–) respectively, and no particular information related to structure is obtained.

In contrast, for [M+NO₃]⁻ and [M–H+(NaNO₃)]⁻, as well as [M–2H+Na]⁻ ions, no overlap with the reference spectrum from [M–H]⁻ is observed.

Notably, NO₃ ^– loss from nitrate species consistently produced a dominant fragment at m/z 61.9890, while all other adducts, including [M–H]⁻, displayed high intensity at the m/z 59.0133 fragment from CH₃COO^– loss. These intense, non-specific fragments were recurrent across oxylipins and inflated similarity scores between unrelated ion speciesparticularly between nitrate speciesdue to shared fragmentation artifacts rather than true structural similarity. For instance, two [M+NO₃]⁻ from unrelated full-length 12,13-DiHOME and the cyclized prostaglandin PGE₃ showed a similarity score of 0.876 in the molecular network when using the modified cosine algorithm. However, their direct pairwise MS/MS comparison yielded a true score of only 0.281, with just two matched fragments, clearly illustrating the impact of shared non-specific fragments on artificially inflated similarity scores (Figure S7).

To overcome this, we excluded MS/MS fragment ions between m/z 50.0000 and 62.0000 in our IIMN workflow (described in the Supporting Information). This filtering step allowed us to focus on more informative MS/MS patterns based on diagnostic neutral losses and position-specific fragments.

The CID spectrum of [M–2H+Na]⁻ ion complexes has been extensively studied for PUFA. In general, it provides rich structural information about double bonds, generating product ions via charge-driven fragmentation processes at the site of the resonance-stabilized carbanion. However, the CID spectrum of the [M–2H+Na]⁻ observed with 15-HETE differing significantly from the CID spectrum of the corresponding [M–H]⁻ ion. Notably, no structural information related to the position of the oxidative modification was obtained.

Ion Identity Molecular Networking (IIMN) Evaluation

Ion Identity Networking (IIN)

IIN was essential to accurately group related ions and enhance the interpretability of the molecular networks generated through the IIMN workflow. The IIN results confirmed that the most abundant ion species observed in our data set (Figure S8) exhibited strong chromatographic correlation, supporting the robustness of the ion identity characterization strategy.

Figure S9 illustrates two representative examples15-HETE (a full-length oxylipin) and PGE₂ (a cyclized oxylipin)showing high correlation (≥75% in feature shape and ≥60% in feature height) between ion identities using the metaCorrelate algorithm, with nodes colored by their relative abundance.

Additionally, IIN enabled the detection of compound-specific ISF products, thereby supporting more accurate annotation of oxylipins (Figure ). For example, PGD₂ exhibited signals derived from ISF corresponding to [M–H–H₂O]⁻ and [M–H–(H₂O)₂]⁻, consistent with sequential dehydration events. PGJ₂ showed a single dehydration product, while no such fragments were detected for 15-deoxy-Δ^12,14-PGJ₂, in line with its structure lacking hydroxyl groups.

4.

Ion Identity Networks (IIN) of prostaglandins (PG) illustrating how in-source fragmentation aids compound characterization. (A) Structures of PGD₂, PGJ₂, and 15-deoxy-Δ^12,14-PGJ₂ along with their corresponding m/z values. An m/z difference of 18.0105 Da distinguishes these species, consistent with the loss of one water molecule. (B) Ion identity networks of PGD₂, PGJ₂, and 15-deoxy-Δ^12,14-PGJ₂. PGD₂ exhibits two dehydration products ([M–H–H₂O]⁻ and [M–H–(H₂O)₂]⁻), while PGJ₂ shows a single dehydration event. No dehydrated species are detected for 15-deoxy-Δ^12,14-PGJ₂, in line with the absence of hydroxyl groups in its structure. The networks show the ion identities and their relative abundances (log₁₀ scale), with color indicating intensity.

In this context, IIN proved particularly valuable for oxylipin analysis, where the presence of isomers, adducts, and ISF can complicate annotation. By grouping related ion species and revealing compound-specific patterns, IIN significantly improved structural insight and network clarity.

Ion Identity Molecular Networking (IIMN) with Different Algorithms

To identify the most reliable strategy for preserving structural relationships among oxylipins, we compared MN generated within the IIMN framework using three algorithms available in mzmine: the modified cosine, DreaMS, and MS2DeepScore. These approaches differ in how they assess MS/MS similarity and cluster related features, allowing us to evaluate their performance in accurately grouping structurally similar oxylipins.

Among them, the modified cosine algorithm proved most effective at grouping oxylipins based on structural similarity. It aligns shared MS/MS fragment ions and neutral losses, while accounting for differences in precursor masses (Δm/z shifts). This allows to link related species that differ only by specific chemical modifications. This algorithm was able to clearly separate full-length oxylipins from their cyclized counterparts within the same network and to capture fine-grained relationships between isomers and molecules differing only in the position or type of oxidative modification, among [M–H]⁻ adducts with structural information (Figure ). A representative example is the subnetwork formed by 5-HETE, 5-HEPE, 5-oxo-ETE, 20-OH-LTB₄, 5,15-DiHETE, and RvE₁. These are oxylipins derived from different PUFA precursors, with distinct molecular formulas and oxidative modifications, yet grouped due to position-specific fragments, reflecting a shared oxidative modification at C5. These connections highlight the algorithm’s ability to capture meaningful structural relationships among oxylipins.

5.

Ion Identity Molecular Network (IIMN) generated using the modified cosine algorithm in mzmine software. (A) Network showing clear separation between full-length and cyclized oxylipins, with a representative subnetwork illustrating structural similarity based on position of the oxidative modification. (B) Structures and m/z values of the oxylipins in the subnetwork, highlighting oxidation at position C5.

In contrast, the DreaMS algorithm partially separated full-length from cyclized oxylipins, but grouped them into distinct regions rather than forming a coherent network. DreaMS is a transformer-based neural network pretrained in a self-supervised manner to generate rich molecular representations directly from large-scale experimental MS/MS data, only in ESI­(+). Oxylipin clustering was mainly driven by spectral similarity among isomers like 5-HETE, 15-HETE, and 20-HETE, based on common neutral losses (Figure S10). Some links were observed between different adducts, such as the [M+Cl]⁻ ions of 5-HETE and 15-HETrE. These connections likely reflect limited or inconclusive structural information, as the fragmentation of these adducts does not retain position-specific diagnostic ions, leading to ambiguous groupings within the network.

The MS2DeepScore algorithm was the least effective at capturing structural detail within the oxylipins. MS2DeepScore is a Siamese neural network model that predicts chemical similarity between mass spectra, across ionization modes. It failed to separate full-length from cyclized species and generated sparse subnetworks with limited connectivity (Figure S11). While some groupings reflected shared structural motifssuch as hydroxyl groups in equivalent positions relative to the terminal carbon (e.g., 15-HETE, 14,15-DiHETrE, PGD₁, and PGF_2α, all containing an −OH group in the n-6 position)many connections were instead driven by less-informative adducts such as [M–H+(CH₃COONa)]⁻ and [M–H–H₂O]⁻. These likely reflect common neutral losses rather than meaningful structural similarity, reducing the interpretability of the resulting networks.

The limited performance of MS2DeepScore and DreaMS may be attributed to the structural constraints of oxylipins, which derive from a small set of PUFA precursors and are produced by a limited number of oxidative enzymes. As a result, they share a conserved carbon backbone and fall within a narrow mass range (250–450 Da), leading to low mass and spectral variability. Their diversity arises mainly from oxygenation patterns or cyclization, producing subtle fragmentation differences often missed by algorithms trained on chemically diverse datasets. In contrast, the fragment-level comparison of the modified cosine algorithm better captured these fine structural variations, highlighting the importance of algorithm selection for structurally conserved lipid classes.

Decoding Oxylipin Connections Using the Modified Cosine Algorithm

Given its ability to capture structural relationships between oxylipins, we focused on networks generated with the modified cosine algorithm, which reflects MS/MS spectral similarity through shared position-specific fragments and neutral losses. This approach allowed us to examine how structural attributes, such as the type and position of oxidative modifications or the conservation of ring systems influenced clustering, providing insight into the structural characterization of novel oxylipins in biological samples.

Deep Insights into Full-Length Oxylipins

Oxidative Modifications at the Same Position and Same Functional Group

A clear pattern emerged among oxylipins sharing oxidative modifications at the same carbon position. For example, 15-HEPE and 15-HETE (score 0.626) both carry a hydroxyl group at C15. Their spectral similarity is largely driven by position-specific fragment ions (e.g., m/z 175.1291), which result from α-elimination adjacent to the hydroxyl group. In this case, the number of carbon atoms and double bonds downstream of the hydroxyl group is conserved, leading to structurally similar fragments. Such shared features enable strong spectral connections despite differences in the precursor PUFA structure (Figure S12).

Oxidative Modifications at the Same Position and Different Functional Groups

Beyond identical modifications, the algorithm also revealed connections between oxylipins with different oxidative groups at the same carbon. For example, 11-HETE, 11,12-DiHETrE, and 11(12)-EpETrE all carry modifications at C11hydroxy, dihydroxy, and epoxy, respectivelyand are linked by scores of 0.534–0.699. Their spectral similarity is driven by a shared intense fragment at m/z 167.1081, resulting from α-elimination adjacent to the hydroxyl group at C11. Notably, epoxides can undergo ring opening during fragmentation to form hydroxy-like structures. Neutral losses were also matched, especially for 11-HETE and 11(12)-EpETrE, which are functional group isomers. 11,12-DiHETrE displayed similar neutral losses, shifted by 18.0115 Da (H₂O) due to the presence of an additional −OH and the loss of a double bond. Fragmentation mechanisms for HETE, EpETrE, and DiHETrE oxylipins have been described previously. These features highlight that fragmentation centered on a specific carbon can override differences in functional group chemistry, reinforcing network connectivity (Figures S13–S15).

Oxidative Modifications at Different Positions and Same Functional Groups

Isomeric oxylipins that share the same functional group but differ in the position of modificationsuch as the epoxides 9(10)-EpOME and 12(13)-EpOMEexhibited a similarity score of 0.727 (Figure S16). This connection was primarily driven by common neutral losses rather than position-specific fragments, illustrating the algorithm’s sensitivity to similar fragmentation behavior among positional isomers. Notably, this pattern of spectral similarity extended beyond isomers to oxylipins with the same type of oxidative modification but located at different positions and derived from distinct precursor PUFA. For instance, 12,13-DiHOME and 14,15-DiHETrEtwo dihydroxy oxylipins with different carbon chain lengthswere connected with a score of 0.710. In this specific case, the observed similarity is driven by a shared fragment at m/z 129.0924, resulting from α-elimination between C11 and C12 in DiHOME and between C13 and C14 in DiHETrE (Figure S17). This fragmentation yields a dihydroxylated moiety downstream of the n-8 position with identical m/z in both compounds. In addition, the modified cosine score captures their most intense diagnostic fragments: m/z 183.1393 for 12,13-DiHOME and m/z 207.1392 for 14,15-DiHETrE. These ions result from cleavage at homologous positions and differ by exactly 24.0000 Da, reflecting the presence of two extra carbon atoms and two extra double bonds in DiHETrE. This example shows that position-specific fragments and mass shifts due to chain length or unsaturation can still yield strong MS/MS similarity. This pattern extends to other dihydroxylated oxylipins, where key fragments are captured either by direct or modified cosine matching, revealing structural relationships beyond isomerism (as observed in Figure ).

Deep Insights into Cyclized Oxylipins

Conserved Ring Substitution and Identical Side Chains

Cyclized oxylipins formed a distinct and well-connected subnetwork in the molecular network, largely reflecting their conserved cyclopentane ring system and similar MS/MS spectra (for ring substitution patterns see Figure S18). Isomeric PG such as PGE₂ and PGD₂ showed high spectral similarity (score 0.787), driven by both similar substitution patterns on the ring and identical side chains (Figure S19). These structural similarities result in overlapping fragmentation profiles, reinforcing the strong influence of both the cyclopentane backbone and side-chain composition on network connectivity.

Conserved or Similar Ring Substitution with Variable Side Chains

PG with identical or similar substitution patterns on the cyclopentane ring but differences in their side chains also showed notable spectral similarity. For instance, PGE₃ and PGE₂ (score 0.867) share identical ring substitution and R₁ side chain but differ in the number of double bonds in the R₂ chain (Figure S20). Similarly, PGE₂ and PGD₁ (score 0.689) have comparable ring substitutions and the same R₂ chain, differing in R₁ in the number of double bonds (Figure S21). A particular case is the connection between PGJ₂ and 15-deoxy-Δ^12,14-PGJ₂ (score 0.503), which share the same substitution pattern on the ring but differ by the loss of a hydroxyl group in the R₂ side chain of 15-deoxy-Δ^12,14-PGJ₂ (Figure S22). This lower similarity score emphasizes the role of specific functional groupssuch as hydroxyl moietiesin driving fragmentation similarity. Notably, this structural difference impacts spectral similarity more than the number or position of double bonds (as observed in aforementioned examples), likely due to the absence of diagnostic fragments associated with the loss of that functional group.

Divergent Ring Substitution with Conserved Side Chains

Notable spectral connections were also observed between PG with different substitution patterns on the ring but conserved side chains. For instance, PGE₂ and PGJ₂ (score 0.843) share the same side chain yet differ in the functional groups present on the cyclopentane ring (Figure S23). This connection indicates that side-chain conservation can drive spectral similarity, even when ring substituents are modified.

Overall, both the type and position of oxidative modifications, as well as conserved ring and side-chain structures, strongly influenced spectral similarity and network. The modified cosine algorithm effectively detects these chemical features, enabling detailed characterization of known oxylipins and the potential discovery of novel compounds.

Practical Application for the Characterization of Oxylipins in Biological Matrices

We applied IIMN using the modified cosine algorithm to a pooled sample of healthy mouse spleen extracts, integrating three MS/MS spectral libraries: (i) our in-house mixture of standards library, (ii) our in-house library of individually acquired oxylipin standards, and (iii) the curated, HR, negative-mode LIPID MAPS Oxylipin Standard Database. Using a cosine score of >0.500 followed by manual verification, we obtained confident MS/MS matches for 28 oxylipins (Table S3).

The resulting IIMN exhibited clear chemical organization, separating full-length oxylipins from cyclized derivatives, consistent with the behavior in the standard mixture. IIMN effectively grouped features into chemically coherent subnetworks based on shared fragmentation patterns, partitioning by core architecture: (i) non-oxidized fatty acids (annotated via the mzmine lipid annotation module), (ii) full-length oxylipins, (iii) prostanoids (prostaglandins and isoprostanes with a five-membered ring), and (iv) more polar prostanoids, likely reflecting their downstream catabolic pathways. Additional subnetworks (e.g., lyso-glycerophospholipids) also appeared under our LC-MS conditions (Figure S24).

Within the linear branch, subnetworks were divided by oxidation pattern along the acyl chainepoxy-, hydroxy-, dihydroxy-, and trihydroxy-oxylipin speciesdriven by common neutral losses and fragmentation mechanisms (e.g., 15-HETE and 12-HHTrE). Position-specific similarity was also evident within subnetworks, as nodes bearing a C12-hydroxy group clustered together (e.g., 12-HEPE, 12-HETE, 5,12-DiHETE, and 6-trans-LTB₄).

The prostanoid subnetwork was denser and more self-contained, as the cyclopentane ring and side-chain similarities yielded higher cosine scores. Both enzymatically PG and non-enzymatic isoprostanes were annotated; although their MS/MS spectra are indistinguishable, they were assigned based on their respective entries in the reference library. This network was further organized into keto-, hydroxy-, and dihydro seriestypical phase-I transformations that increase polarity and reduce RT. , Connections within this subnetwork directly mirror plausible pathway steps (e.g., 8-isoPGF_2α links to 13,14-dihydro-15-keto-isoPGF_2α within the subnetwork). Overall, IIMN grouped oxylipins in complex matrices into coherent molecular families, aiding structural elucidation.

Accurate characterization still required manual curation due to overlapping fragmentation profiles among oxylipins and ISF. For instance, PGD₂ clustered with PGJ₂ and 15-deoxy-Δ^12,14-PGJ₂ within the same ion identity network, which corresponded to [M–H–H₂O]⁻ and [M–H–(H₂O)₂]⁻ signals from ISF, respectively. Evaluation of ion identity relationships confirmed PGD₂ as the parent ion (Figure S25A), underscoring the need to disentangle isomers from ISF during oxylipin annotation.

Ion identity profiles in biological matrices were also consistent with standards, although some differences were observed. In spleen tissue, the relative abundance of [M–H+(CH₃COONa)]⁻ clusters decreased, while [M+NO₃]⁻ ions became dominant after [M–H]⁻ (e.g., for PGD₂) (Figure S25B). Matrix-dependent variations in adduct distribution influenced IIMN topology, underscoring the importance of considering ionization behavior across different sample types.

Our workflow also enabled confident assignment of trihydroxy-oxylipins. Within their subnetwork, only 9,12,13-TriHOME was annotated with an MS/MS library match classical cosine score of 0.667 and was confirmed manually. In contrast, 9,10,13-TriHOMEdespite being present in our MS/MS spectral librarydid not obtain any match (with cosine score of >0.500), likely due to its lower abundance and noise, but was confidently manually characterized through position-specific fragments. Crucially, a putative novel oxylipin, 9,10,11-TriHODE, which lacked a reference spectrum was annotated by propagation within this molecular subnetwork. The node was tightly connected to 9,10,13-TriHOME, sharing position-specific fragments and neutral losses. This network context, together with manual inspection of its fragmentation pattern, supported its assignment as 9,10,11-TriHODE (Figure S25B,C). This case illustrate how the nodes of known compounds can be used to annotate novel species through structural propagation.

These examples highlight how IIMN bridges gaps in developing spectral libraries and, when combined with expert curation, strengthens structural elucidation in complex matrices.

Conclusions

This study pioneers the application of IIMN to the analysis of oxylipins, establishing a workflow that significantly advances structural characterization beyond conventional MN strategies. Unlike traditional approaches, which often yield Molecular Networks populated by redundant ion species, IIMN integrates MS¹ peak-shape correlation to overcome this limitation. By unifying related features into coherent molecular families, this approach drastically reduces network redundancy and enhances interpretability, enabling the clear distinction between multiple ion identities and genuine structural relationships.

A key finding of this work is the systematic evaluation of MN algorithms. We demonstrate that the modified cosine scoring algorithm is particularly effective for oxylipins, as it successfully links structurally related species by capturing position-specific fragmentation patterns. This contrasts with other strategies that, while powerful for diverse metabolomics datasets, may lack the specificity required for the subtle structural variations of isomeric oxylipins.

Crucially, the practical utility of this approach was validated through its application to biological samples, rich in these bioactive lipid mediators. The workflow demonstrated remarkable robustness in a complex matrix, accurately grouping biologically relevant oxylipins into coherent molecular families while effectively distinguishing genuine signals from ISF and adducts. This validates IIMN as a robust tool for untargeted redox lipidomics studies.

Finally, we addressed the annotation bottleneck in the field. By implementing the workflow using the native IIMN module within mzmine and releasing a dedicated oxylipin reference MS/MS spectral library, we offer the scientific community a transferable, open-source strategy for oxylipin annotation and discovery. This resource promotes data reproducibility and facilitates deeper structural insights into the role of oxylipins in health and disease.

Glossary

Glossary

AA

arachidonic acid (C20:4n-6)

ALA

α-linolenic acid (C18:3n-3)

CE

collision energy

CID

collision-induced dissociation

CLMN

classical molecular networking

DDA

data-dependent acquisition

DGLA

dihomo-γ-linolenic acid (C20:3n-6)

DHA

docosahexaenoic acid (C22:6n-3)

DiHETE

dihydroxy-eicosatetraenoic acid

DiHETrE

dihydroxy-eicosatrienoic acid

DiHOME

dihydroxy-octadecenoic acid

DreaMS

Deep epresentations empowering the annotation of mass spectra

EIC

extracted ion chromatogram.

EPA

eicosapentaenoic acid (C20:5n-3)

EpDPE

epoxy-docosapentaenoic acid

EpETE

epoxy-eicosatetraenoic acid

EpETrE

epoxy-eicosatrienoic acid

EpOME

epoxy-octadecenoic acid

ESI

electrospray ionization

FBMN

feature-based molecular networking

HEPE

hydroxy-eicosapentaenoic acid

HETE

hydroxy-eicosatetraenoic acid

HETrE

hydroxy-eicosatrienoic acid

HR-MS

high-resolution mass spectrometry

HR-MS/MS

high-resolution tandem mass spectrometry

HODE

hydroxy-octadecadienoic acid

HOTrE

hydroxy-octadecatrienoic acid

HpOTrE

hydroperoxy-octadecatrienoic acid

IIMN

ion identity molecular networking

IIN

ion identity networking

ISF

in-source fragmentation

ISTD

internal standard

LA

linoleic acid (C18:2n-6)

LC

liquid chromatography

LC-MS/MS

liquid chromatography coupled to tandem mass spectrometry

LT

leukotriene

LX

lipoxin

MassQL

Mass spectrometry query language

MASST

Mass spectrometry search tool

MN

molecular networking

MS/MS

tandem mass spectrometry

MS

mass spectrometry

MS¹

mass spectrometry full-scan mode

m/z

mass-to-charge ratio

NEO-PUFA

nonenzymatically oxygenated oxylipins

oxo-ETE

oxo-eicosatetraenoic acid

oxo-ODE

oxo-octadecadienoic acid

PG

prostaglandin

PUFA

polyunsaturated fatty acid

QSPR

quantitative structure–property relationship

QTOF

quadrupole time-of-flight

RT

retention time

TriHODE

tridhydroxy-octadecadienoic acid

TriHOME

trihydroxy-octadecenoic acid

TX

thromboxane

UHPLC

ultra-high-performance liquid chromatography

Additional data are available in the Zenodo repository (doi: 10.5281/zenodo.17176661).

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.5c06084.

• Experimental section with detailed information regarding reagents and solutions, free oxylipin extraction from spleen samples, Quality Assurance (QA) and Quality Control (QC) analysis, RP-UHPLC-ESI(−)-QTOF-MS data acquisition, high-resolution tandem mass spectrometry (HR-MS/MS) spectral MS/MS dataset generation, Ion Identity Molecular Networking (IIMN) in mzmine software, and MS/MS spectral libraries generation; Figures S1–S25 with chromatograms, MS/MS spectra, and network visualizations to support and complement the main text; and Tables S1 and S3 with LC-MS descriptors for oxylipin standards and spectral matches in biological samples, respectively (PDF)

• Table S2 with the MS¹ database with chemical formulas, exact masses, SMILES, retention times, and experimental ion identity profiles (adducts, clusters, and in-source fragmentation) for the analyzed oxylipin standards (XLSX)

S.M.C-A.: conceptualization, formal analysis, methodology, data analysis, data curation, investigation, writingoriginal draft, and writingreview and editing; F.B.: data analysis, data curation, and investigation; H.C.: data analysis and data curation; A.S-.I.: conceptualization, investigation, and data analysis; D.P-G.: conceptualization and investigation; P.C-M.: data analysis; A.G-F.: data analysis, and writingreview and editing; J-M.G.: conceptualization, investigation, and writingreview and editing; T.D.: conceptualization, investigation, and writingreview and editing; A.G.: conceptualization, supervision, data analysis, data curation, investigation, writingoriginal draft, writingreview and editing; and C.B.: conceptualization, supervision, resources, writingreview and editing, project administration, and funding acquisition.

This work was funded by grants from the Spanish Ministry of Science and Innovation (PID2020-112830RB-I00 and PID2023-149557OB-I00) (AEI/FEDER,UE) and “la Caixa” Foundation (HR24-00686  FUNPATH).

The authors declare no competing financial interest.