Beyond High-Accuracy Mass Spectrometry: Why Chromatographic Retention Time Must Reclaim Its Role in Analyte Identification

Metabolomics (metabolic phenotyping, or metabotyping) aims to catalogue and interpret the small molecule content of a biological system and study metabolite trends in concentrations that reflect the biochemical state of the studied cells, organs, or organisms. In practice, this means discovering small molecule metabolites, including lipids, and their patterns that discriminate between samples or groups of samples, to reveal differences between, for example, physiological and pathological states. Ultimately, these findings help scientists to generate new knowledge and understand the underlying or perturbed biochemical mechanisms of the topic of their investigation. If the topic of study is disease, this can lead to improved diagnosis or prognosis; if the study is on sports biochemistry, we can obtain a new understanding of the mechanism of energy depletion during physical exercise and the body’s mechanisms for replenishment, the mechanism of oxidative stress, and so on. For plants, we are interested in (among others), plant protection mechanisms, and the processes of plant and product growth. The true value of any untargeted metabolomics experiment depends not on the number of features detected during sample analysis, but on how the detected features can be translated into reliable metabolite identities and, further, how many of the metabolites are related to biochemical phenomena. Obviously, without trustworthy identification, any mechanistic interpretation becomes speculative, and the claimed biomarkers become little more than statistical artefacts.

Metabolite identification is straightforward to define. For a compound to be claimed as “identified”, its signal should match, within predefined tolerances, data from an authentic standard measured under the same conditions. Community frameworks and guidelines, such as those from the Metabolomics Standards Initiative (MSI),¹ or the Lipidomics Standards Initiative (LSI),² formalize this with confidence levels.

Level 1 identifications are reserved for compounds that have been confirmed by direct comparison to standards: chromatographic retention time, mass-to-charge ratio (m/z), and tandem mass spectrometry (MS/MS) spectra. Lower levels of confidence correspond to putative structures or class-level annotations (a schematic is depicted in Figure 1). Annotation is the process of assigning putative chemical identities to detected signals (for example, peaks, features) without reaching definitive confirmation. Schemes and publications defining these processes are widely cited and have been available for more than a decade.³

Despite this apparent clarity, practice often diverges sharply from principle. A growing body of evidence, including systematic reviews of the literature, has shown that metabolite identification in untargeted liquid chromatography mass spectrometry (LC–MS) studies is frequently reported with inadequate supporting data and with a worrying tendency to overstate confidence.⁴ It is still common to see compounds described as “identified” when, in reality, they are the result of a single database match on accurate mass alone; sometimes this is supported by comparison of MS/MS spectra, yet with no chromatographic information and no comparison with data from the analysis of standards. In effect, annotation and identification are treated as synonyms, even though existing guidelines clearly distinguish them. This blurring in the use of terms erodes transparency, complicates reproducibility, and may mislead readers who may assume that reported metabolites are securely characterized when they are not.

We argue that the underutilized power of liquid chromatographic retention time (t_R) and related chromatographic observables can offer critical pieces of orthogonal evidence that can, and should, be used alongside mass spectrometric data.

Current Situation

Despite significant investment by instrument manufacturers, software developers, database curators and the metabolomics community at large, the identification step remains the major bottleneck and a persistent source of error. A significant portion of the published studies provides inadequate evidence in support of their assigned markers, while there are several papers with misannotations.⁵ What is more concerning is the fact that such errors often pass through peer review, especially in journals that do not primarily focus on metabolomics and may lack specialized editors and reviewers.⁶ As a result, incorrect metabolite identities are enshrined in the public record and are subsequently cited as fact and then can be further used to support newer studies or to populate databases. The current landscape is further complicated by the rapid growth of computational and AI-based tools. Automated workflows for spectral matching, in silico fragmentation and pathway mapping are now widely available and are extremely attractive, particularly for non-specialists, because they appear to offer rapid identification at scale. However, these tools are only as reliable as the data and assumptions that underpin them. When errors in metabolite identification enter the literature or databases, AI-driven analyses may reproduce and amplify those errors.

The scientific consequences of this situation are far from trivial. Poorly supported identifications can distort pathway analyses, drive the development of fragile mechanistic narratives and mislead biomarker discovery efforts. Misidentified metabolites deposited into databases can become a source of future misannotations. In the era of large-scale text and data mining, including AI-based literature analysis, such errors can propagate even faster. If we allow such a “fake reality” of metabolomic knowledge to become entrenched, the credibility of biomarker discovery metabolomics is at risk.

In recent perspective articles, we provide examples of failed annotations⁶; a good number of these could be avoided by simply checking the retention time and the physicochemical properties of the proposed failed annotation.

Mass spectrometry is central to metabolomics and lipidomics because it provides sensitivity, dynamic range and the ability to generate structural information via high-resolution measurements and MS/MS fragmentation. However, MS alone is rarely sufficient for unambiguous identification. Isobaric and isomeric species are abundant in the small-molecule space; many compounds share the same nominal mass or even identical elemental composition. In such cases, accurate mass and a plausible MS/MS spectrum may still leave multiple candidate structures, and there are obvious examples in the numbers of isomeric fatty acids or saccharides. In addition, experimentally compiled LC–MS spectral libraries are often incomplete, heterogeneous in quality, and often acquired under diverse conditions, while a big part of the online databases contains in silico spectra. To conclude, blind reliance on single MS¹ acquisition-based database matching increases the risk of errors.

This is where orthogonal data and, in particular, chromatographic retention time can help filter the database hits. The utility of retention time in metabolite identification manifests at several levels. At the most direct level, matching the retention time of a feature in a sample to that of an authentic standard measured in the same system provides strong evidence of identity. When combined with matching MS and MS/MS data, this forms the backbone of MSI/LSI Level 1 identification.² Similarly, regulatory guidance from the U.S Food and Drug Administration (FDA) and the European Union (EU) explicitly mandates the combined use of retention time and MS spectral information in a numeric scoring system for the identification of targeted analytes (Decision EC 657/2002, Regulation (EU) 808/2021).^7,8 These guidelines are very specific on the matching criteria to be applied and advise that the retention time of an analyte in a sample extract must match the retention time of the corresponding calibration standard (or matrix-matched/matrix-fortified standard) within ± 0.1-minute tolerance or 5% relative deviation, and mass deviation for “diagnostic ions” shall be < 5 ppm using a high-resolution mass spectrometry (HRMS) instrument (Figure 2).

Unfortunately, retention times are barely exploited beyond the peak detection and alignment steps, and identification workflows typically exploit only the precursor ion m/z values within automated search engines and MS databases. The truth is that retention time is not an arbitrary value; it reflects the physicochemical properties of a molecule, its hydrophobicity, polarity, and its interactions with the stationary and mobile phases, under a defined set of chromatographic conditions. Retention time offers a powerful (orthogonal to MS) descriptor that can be used to support, refine, or refute candidate identifications derived from mass spectrometry database searching.

Systematic use of retention time can help address some of the broader problems in metabolomics. Even when reference standards are not available, retention order can provide useful information on the molecular properties of the analytes. This has been common knowledge for decades and simply needs to be applied in practice. Within homologous series or lipid classes, retention times provide an internal logic that can be used to assess the plausibility of annotations. For example, in reversed-phase LC (RPLC) one expects longer acyl chains or more hydrophobic species to elute later; increasing unsaturation generally shifts retention earlier. Deviations from these patterns may indicate misannotations or unexpected chemistry and should prompt further scrutiny. For example, concerns arise when the expected polarity or logP of a candidate is inconsistent with its reported retention time, or when its elution behavior does not align with that of structurally related compounds.

In a more formal and structured way these correlations can be elaborated via retention indices or by leveraging retention prediction models that estimate tR from molecular structure.⁹ While such models are imperfect and method-specific, they can provide an additional layer of evidence. For example, a proposed molecular structure whose predicted retention time is grossly incompatible with the observed value is unlikely to be correct. Such approaches do not require identical retention times across laboratories, which is impractical. Rather, what matters is consistency in retention order, supported by the use of appropriate calibrants and quality control samples.

Chromatographic data are not a panacea and will not, on their own, provide definitive evidence for metabolite identification. Retention times may drift, and co-elution remains a persistent challenge in highly complex matrices. Nevertheless, such limitations can be effectively managed through good analytical practice, including the use of internal standards, quality control samples, improved retention alignment strategies, retention indexing, and thorough documentation of methods. Hence, reporting chromatographic conditions, retention times for key standards, and representative chromatograms should be seen not as optional extras, but as essential components of a transparent metabolomics study that represents a significant investment in time and resources. Simple measures, such as using a defined set of calibrant compounds to anchor retention behaviour, or reporting normalised retention indices instead of retention time, could improve comparability between laboratories and instruments.

Another benefit would be the possibility for improved reporting of recurrent unknowns.

Integration of retention time into identification workflows can offer improvements for specific reasons. For lipids and other homologous series, trend plots of retention versus chain length or degree of unsaturation can be generated as a diagnostic tool. Recently the METLIN small molecule retention time (SMRT) dataset was introduced, providing the largest data set of experimentally acquired retention times, covering more than 80,000 molecules.¹⁰ In addition, several retention time databases have been developed for specific applications, with focus on natural products and metabolomics, including the Fiehn Lab Retention Time Libraries and the RIKEN MSn and RT libraries (PRIMe).^11,12 Such databases offer a useful resource to compare retention times and elution orders. We would clarify that such approaches may not reach perfection or accurate calculation of the experimental retention time. Rather, they can reflect a mindset in which chromatographic data are treated as a useful source of information. Integrating retention data into automated pipelines as a filter, or within a scoring term or an input into machine learning models, can make such models more robust and less prone to propagate obviously implausible identifications.

Finally, another recurring issue is the reporting of chemically implausible compounds, such as complex synthetic drugs, laboratory detergents, or industrial dyes reported as differential metabolites in contexts where their presence is extremely unlikely. Many of these annotations would not withstand even a basic plausibility assessment, namely whether the proposed compound could be expected in the analyzed specimen. Examples of such cases documented in the literature are discussed in our recent perspective article.⁶

Conclusion

Improved data curation is central for the future development of metabolomics. Such data curation includes manual checking of individual identifications, systematic use of all available experimental evidence, including chromatographic, spectrometric, and, where available, ion mobility information. Curated reference datasets that include retention times, collision cross section (CCS values), MS and MS/MS spectra for a broad panel of metabolites and lipids in specific matrices are particularly valuable. They provide a concrete, experimentally verified backdrop against which new annotations can be evaluated. When such resources are developed under rigorous quality assurance protocols (considering adduct patterns, m/z tolerance, retention behaviour), they become powerful tools for improving the reliability of metabolite identification. Overall, to promote and consolidate the field, authors need to engage with published guidelines and recommendations when designing experiments, developing identification strategies, reporting their experiments, and writing manuscripts. Reviewers and editors need to insist that evidence levels are clearly stated and that claims of identification are supported by appropriate real data, such as chromatographic and spectrometric data. Journals can help by aligning their author instructions with community standards and by encouraging, or even requiring, explicit statements of identification confidence and the provision of raw or processed chromatographic data for key metabolites.

References

1. Fiehn, O.; et al. The Metabolomics Standards Initiative (MSI). Metabolomics 2007, 3, 175–178. DOI: 10.1007/s11306-007-0070-6
2. Liebisch, G.; Vizcaíno, J. A.; Köfeler, H.; et al. Shorthand Notation for Lipid Structures Derived from Mass Spectrometry. J Lipid Res 2013, 54, 1523–1530. DOI: 10.1194/jlr.M033506
3. Sumner, L. W.; Amberg, A.; Barrett, D.; et al. Proposed Minimum Reporting Standards for Chemical Analysis: Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI). Metabolomics 2007, 3, 211–221. DOI: 10.1007/s11306-007-0082-2
4. Kodra, D.; Pousinis, P.; Vorkas, P. A.; et al. Is Current Practice Adhering to Guidelines Proposed for Metabolite Identification in LC–MS Untargeted Metabolomics? A Meta-Analysis of the Literature. J Proteome Res 2022, 21, 590–598. DOI: 10.1021/acs.jproteome.1c00841
5. Theodoridis, G.; Gika, H.; Goodacre, R.; et al. Ensuring Fact-Based Metabolite Identification in LC–MS-Based Metabolomics. Anal Chem 2023, 95, 3909–3916. DOI: 10.1021/acs.analchem.2c04609
6. Theodoridis, G.; Fiehn, O.; et al. What’s in a Name: Compound Annotations in Metabolomics Studies. Metabolomics 2026, 22, 22. DOI: 10.1007/s11306-025-02387-0.
7. European Commission. Commission Decision 2002/657/EC of 12 August 2002 Implementing Council Directive 96/23/EC Concerning the Performance of Analytical Methods and the Interpretation of Results. Off. J. Eur. Communities 2002, L221, 8–36.
8. European Commission. Commission Implementing Regulation (EU) 2021/808 of 22 March 2021 on the Performance of Analytical Methods for Residues of Pharmacologically Active Substances Used in Food-Producing Animals and on the Interpretation of Results as Well as on the Methods to Be Used for Sampling and Repealing Decisions 2002/657/EC and 98/179/EC. Off. J. Eur. Union 2021, L180, 84–109.
9. Stanstrup, J.; et al. PredRet: Prediction of Retention Time by Direct Mapping between Multiple Chromatographic Systems. Anal. Chem. 2015, 87, 9421–9428. DOI: 10.1021/acs.analchem.5b02287
10. Domingo-Almenara, X.; Guijas, C.; Billings, E.; et al. The METLIN Small Molecule Dataset for Machine Learning-Based Retention Time Prediction. Nat Commun 2019, 10, 5811. DOI: 10.1038/s41467-019-13867-3
11. Kind, T.; Wohlgemuth, G.; Lee, D. Y.; Lu, Y.; Palazoglu, M.; Shahbaz, S.; Fiehn, O. FiehnLib: Mass Spectral and Retention Index Libraries for Metabolomics Based on Quadrupole and Time-of-Flight Gas Chromatography/Mass Spectrometry. Anal Chem 2009, 81, 10038–10048. DOI: 10.1021/ac9019522
12. Sawada, Y.; Nakabayashi, R.; Yamada, Y.; et al. RIKEN Tandem Mass Spectral Database (ReSpect) for Phytochemicals: A Plant-Specific MS/MS-Based Data Resource and Database. Phytochemistry 2012, 82, 38–45. DOI: 10.1016/j.phytochem.2012.07.007