METLIN: A Tandem Mass Spectral Library of Standards

Abstract

Untargeted mass spectrometry metabolomics studies rely on accurate databases for the identification of metabolic features. Leveraging unique fragmentation patterns as well as characteristic dissociation routes, allows structural information to be gained for specific metabolites and molecular classes, respectively. Here we describe the evolution of METLIN as a resource for small molecule analysis as well as the tools (e.g. Fragment Similarity Search and Neutral Loss Search) used to query the database and their workflows for the identification of molecular entities. Additionally, we will discuss the functionalities of isoMETLIN, a database of isotopic metabolites, and the latest addition to the METLIN family, METLIN-MRM, which facilitates the analysis of quantitative mass spectrometry data generated with triple quadrupole instrumentation.

1. Introduction

As the interest of the scientific community set eyes on endogenous metabolites almost 20 years ago due to their implications in diagnostics and biomarker discovery, technical challenges had to be overcome in order to use mass spectrometry or metabolomics for these biomedical purposes. Particularly, chromatographic data processing and metabolite characterization/identification of untargeted metabolomic experiments represented the biggest obstacles for researchers worldwide[1]. While the development of chromatographic data processing tools is covered in this book, this chapter will focus on the metabolite database METLIN and its associated tools, from inception with a few hundred metabolites to the most current version with over 600,000 compounds (November 2019).

METLIN was developed to assist metabolomics research across a broad range of disciplines, especially to facilitate the identification of metabolites in LC-MS based metabolomics and bridge the gap to other “omics” sciences such as genomics and proteomics[1]. The molecular or macromolecular identification process to assign names and annotations to genes and proteins from genomics and proteomics experiments has been made possible not only due to great efforts in their respective communities but also due to the predicative sequence of nucleosides and amino acids. On the other hand, the large chemical diversity found in the metabolome as well as the vast number of different molecular entities, limits the prediction of fragmentation patterns (MS/MS) in MS-based experiments thereby limiting the ability to make identifications[2].

METLIN was first developed in 2003 and contained a few hundred metabolites and MS/MS spectra[1]. However, its growth in terms of the number of metabolites and MS/MS spectra but also the incorporation of new tools has been continuous. For example, after it was made publicly available in 2005, METLIN grew from a few hundred to more than 10,000 metabolites with their respective MS/MS spectra by 2012[3]. Additionally, tools to facilitate and automate the identification of known and unknown metabolites have been integrated [4–6].

Concomitantly, other academic, public and private entities, have developed metabolite/small molecule databases, which could be classified into two categories: 1) pathway-centric and 2) compound-centric[7]. METLIN belongs to the latter as it contains chemical structures, spectral profiles and its tools are designed to facilitate metabolite identification.

Similar databases containing MS/MS spectra include: MassBank, HMDB, GNPS, MoNA, LIPID MAPS, NIST 14 and mzCloud[8–12]. For a more detailed comparison between these databases we direct the reader to the work of Vinaixa, M. et. al[2]. Briefly some of the main differences are data collection and curation processes, instrumentation used to generate data, standard reference materials, types of molecules, source and origin of molecule and accessibility. In METLIN, MS/MS spectra have been acquired at the Scripps Center for Metabolomics following strict protocols and only using standard reference material that is either commercially available or has been synthesized. Early on in the development process, it was decided against the use of complex biological samples for reference fragmentation spectra in order to avoid interfering molecules and provide the best spectral quality. This is unlike other databases where MS/MS spectra have been acquired in different laboratories under different conditions and instrumentation. Over the years, METLIN has also incorporated metabolites from a wide range of molecular classes and biological or chemical origins. Therefore, the broad spectrum of small molecules available in METLIN include endogenous metabolites such as lipids, amino acids, nucleotides and carbohydrates; toxicants have recently been included to assist exposome research[13]. Further, drugs and drug secondary metabolites can be found in METLIN as well, as the use of untargeted metabolomics in drug development has gained traction in recent years[14].

2. Tools for the identification of known and unknown metabolites

2.1. Basic Search Engines

In a traditional untargeted metabolomics workflow, dysregulated features (i.e. specific m/z values with their respective retention time values) will be searched against a database for putative metabolite identifications based on accurate mass. METLIN provides three search options: 1) Simple, 2) Batch and 3) Advanced Search (Fig. 1). Simple Search allows the search of both m/z values and neutral masses within a selected mass tolerance. Several adducts in both positive and negative polarities can be selected. Batch Search has the same functions and capabilities as Simple Search, however, several m/z values can be searched simultaneously, facilitating the annotation of adducts and common losses that stem from the same metabolite. Similarly, ions with a different molecular origin can be easily distinguished and linked to other putative m/z values with this search feature. Advanced Search, on the other hand, allows the user to search based on other metabolite information such as molecular formula, metabolite names, SMILES and KEGG, CAS and MID numbers. These searches provide a list of metabolite names (or molecular formulas) that could potentially correspond to the dysregulated feature of interest. However, given the low elemental diversity in biomolecules (C, H, N, O, P and S), this list can contain tens to hundreds of putative metabolite identifications.

Fig. 1.

METLIN search functions for small molecule identification. (A) Simple and Batch Search allow users to search small molecules against a database of 1 million compounds based on both m/z values and neutral masses within a selected mass tolerance. Advanced Search allows searches based on metabolite information such as molecular formula, metabolite names, SMILES and KEGG, CAS and MID numbers. (B) With the MS/MS Spectrum Match Search, experimental and library MS/MS spectra can be searched, matched, and scored in an automatic way. (C) Fragment Similarity Search and Neutral Loss Search aid the identification of metabolites or chemical structures by searching m/z values of the fragments or neutral losses, respectively, regardless of the precursor mass. Analytical Chemistry 2018, 90, 3156–3164. Figure 1, with permission from ACS and RightsLink.

2.2. Identification of Known Metabolites: MS/MS Spectrum Match Search

In order to reduce the list of putative identifications obtained from m/z value searching, experimental MS/MS spectra are compared against spectral libraries in terms of fragmentation patterns (m/z of fragments and their intensities) (Fig. 1). MS/MS Spectrum Match Search is a tool that allows the autonomous identification of metabolites. Here, users upload the fragmentation profile as a table of m/z values and intensities, enter the mass of the precursor with a specific tolerance, collision energy and polarity. This tool then searches, compares and scores the similarity of the experimental spectra with the reference spectra in the library for all metabolites within the selected mass tolerance, relying in a modified X-Rank similarity algorithm [15]. Without a doubt the 600,000 molecular standards with MS/MS spectra are METLIN’s most valuable contribution to metabolomics research. The fragmentation spectra for these ~300,000 molecules have been acquired at four collision energies (0, 10, 20 and 40 V) in both positive- and negative- ion mode. While low collision energy spectra were previously not extensively used in the identification process, they have been recently used as a second layer in metabolite annotation algorithms and data deconvolution[16]. The combination of METLIN’s large spectral library and annotation tools, significantly simplifies data analysis and improves the confidence of putative identifications.

2.3. Identification of Unknown Metabolites

Interestingly, spectral libraries serve a dual purpose in the identification/characterization of metabolic features. As mentioned above, the main use of these libraries is to compare experimental spectra with reference spectra for the purpose of identification (up to level 2 according to the Metabolomics Standards Initiative)[2,17]. However, given the large number of metabolites and the broad range of chemistries, no library is complete despite considerable efforts dedicated to increasing their populations. Furthermore, the number of metabolites in nature still is a subject of debate but estimates in the million range are not hyperbolized. Such numbers dwarf the ~20,000 genes and proteins (without taking post translational modifications into account). Thus, the second purpose of spectral libraries is to aid the identification/characterization of known metabolites without MS/MS spectra available and unknown metabolites, whose structures have not been previously described in any library or resource. For this particular purpose, two tools, Fragment Similarity Search and Neutral Loss Search, have been developed and continually refined over the last 11 years[4]. These tools leverage the large number of spectral information and the dissociation routes similarities between compounds with related structures and chemical moieties.

2.3.1. Fragment Similarity and Neutral Loss Search

The Fragment Similarity Search algorithm was originally implemented into METLIN to find chemical similarities between the desired unknown features and the known molecules available in in the library based on the experimental MS/MS acquired by the user and the over 2.4 million high resolution MS/MS spectra in the METLIN library. Specifically, the Fragment Similarity Search algorithm relies on a shared peak count method and facilitates the identification of the molecule of interest or molecular class by prioritizing molecules with a larger chemical fragment overlap[4].

The Neutral Loss Search algorithm was designed as a complementary tool to Fragment Similarity Search. While in Fragment Similarity Search shared fragments (i.e. ions with the same m/z value) provide structural information of the ion of interest, similar structures with different molecular formulas, ergo different masses and m/z values, would generate fragments with different masses and no similarity would be determined. However, in Neutral Loss Search, mass differences between precursor ions and their fragments ions can provide structural information based on common “leaving groups” in their respective dissociation routes. Both these tools leverage the vast number of carefully curated MS/MS spectra from a wide range of small molecular entities in the METLIN library to facilitate the identification of not only known molecules with no available MS/MS spectra (discussed in more detail in chapter 6), but also of unknown molecules which have not been previously described in any form.

Here, we provide an example using these tools to identify unknown metabolites detected in a murine macrophage cell line (RAW264.7). The analysis was performed using an I-class UPLC system coupled to a Synapt G2-Si mass spectrometer (Waters Corp. Milford, MA). After the unsuccessful identification of a metabolic feature using the using accurate mass and MS/MS spectra matching procedures described in section 2.1 and 2.2 respectively, Fragment Similarity Search and Neutral Loss Search tools were employed to gain structural information and therefore clues to molecular identity in the following steps:

1. Selected fragments from the murine macrophage metabolite MS/MS spectrum were searched against the METLIN MS/MS spectral database using Fragment Similarity Search tool. High intensity fragments are more likely to provide structural information of the metabolic feature, and in some cases, these can be very specific to the metabolite of interest. On the other hand, lower mass fragments are commonly shared with many molecules, making it harder to gain useful structural information from their analysis. Additionally, larger errors in the mass accuracy are inherent for low mass fragments given the mass accuracy definition[18]. In this case, we selected the fragments 149.02, 102.06, 74.02 and 56.05 from the unknown feature of interest for analysis, which resulted in many hits to molecules in METLIN. However, Fragment Similarity Search orders the results based on the number of matching fragments. Methionine sulfoxide (Met sulfoxide) matched 4 of the 4 fragments searched, hinting at a metabolite containing such a chemical structure or moieties (Fig. 2).

2. In order to gain more structural information, other fragments (higher mass) were searched as described above. The higher mass fragment searches did not yield any results that matched a particular compound or molecular class.

3. Assuming that Met sulfoxide with a monoisotopic neutral mass of 165.05 is part of the molecule of interest and we can observe a prominent fragment with the mass 166.05, the next step is to use the mass of the molecule to identify what chemical structures could constitute the rest of the molecule. After calculating the mass difference between the precursor ion and the potential methionine sulfoxide fragment (295.10 – 166.05 = 129.04), we use the Neutral Loss Search tool and search for 129.04 within a selected ppm window. Most of the Neutral Loss Search results consist of molecules that contain glutamic acid (Glu), suggesting Glu is a second piece of the unknown metabolite.

4. After using Fragment Similarity Search and Neutral Loss Search, we have gathered some information about the possible chemical structures of the metabolite of interest. We then deduced how the two pieces might be connected. A plausible link between methionine sulfoxide and glutamic acid is an amide bond, which would form the dipeptide Glu Met sulfoxide.

5. Advanced Search can be used to search for such a peptide in the METLIN database by name and molecular formula. Unfortunately, this peptide is not available in the library. However, a similar dipeptide containing Glu and methionine (Met) is available with MS/MS spectra. While the only difference between the structures is the oxidation of the sulfur atom in methionine, such a small modification can have considerable differences in the dissociation routes and those differences should be considered.

6. The MS/MS spectra between the metabolite of interest (putatively Glu Met sulfoxide) and of the dipeptide Glu Met are compared next to try to further characterize the metabolite identity. Several fragments with a mass difference of an oxygen atom (15.99) are observed and were attributed to the oxidation of the sulfur atom in methionine. An additional feature of METLIN is the MetFrag algorithm developed by S. Neumann and co-workers, which indicates putative structures for the different fragment ions in the MS/MS library [19]. Based on MetFrag’s analysis, we saw that several of the predicted structures that have a mass difference of 15.99 actually correspond to fragments that contain the thiol moiety. This further increases the confidence in the putative identification of the unknown feature as Glu Met sulfoxide (Fig. 3).

Fig. 2.

Fragment Similarity Search facilitates the identification of unknown metabolites where no MS/MS spectral data are available. The fragments of an unknown metabolite were searched against METLIN and all of the four fragments were found to match with methionine sulfoxide. The comparison between experimental and library MS/MS spectra implies high structural similarities.

Fig. 3.

Neutral loss search aides the identification of unknown metabolites based on mass differences between precursor ions and their fragments ions (i.e. common “leaving groups”). A Neutral Loss Search of 129.04 (295.10 – 166.05= 129.04) yielded 168 results, where ~70% contain a glutamic acid moiety. Based on the masses of Met sulfoxide and Glu, a dipeptide is a likely option. Such peptide is not available in the library, however, a similar compound Glu Met contains MS/MS spectra for comparison. Several fragments, which contain the thiol moiety, have a mass difference of 15.99 (monoisotopic mass of oxygen atom) corresponding to the oxidation of sulfur in methionine.

3. METLIN family

3.1. isoMETLIN

In parallel to the development of METLIN, other related databases have been released relying in the growing number of analytical standards available in METLIN to reach new goals in the field of metabolomics. isoMETLIN was implemented in 2014 as a database for isotope-based metabolomics[20]. With an analogous interface to METLIN, isoMETLIN provides accurate mass of all computed isotopologues accrued in METLIN, compounds with a different number of isotope-labeled atoms and consequently, different m/z values. isoMETLIN Search includes the most commonly used stable isotopes in labelling experiments, such as ¹³C, ¹⁵N, ²H and ¹⁸O. Even though isotopologues can be discerned by accurate mass measurements, the further analysis of their MS/MS spectra is necessary to determine the position of the isotopic label within the same isotopologues, a pivotal feature for the investigation of metabolic pathways[21,22]. To accomplish this, isoMETLIN also incorporates the MS/MS spectra for hundreds of isotopomers (same isotopologue with different location of labeled atoms) to help in tracing the isotopic label, providing a vast amount of information about the de novo synthesis of metabolites in certain pathways (Fig. 4). Although the principal application of isoMETLIN’s capabilities is the analysis of metabolic fluxes, which constitutes an emerging “omics” discipline by itself, other uses of isoMETLIN include isotope dilution quantitative metabolomics and identification of compounds, as is shown below.

Fig. 4.

isoMETLIN Simple Search menu allows searching the m/z of isotopologues within a selected mass tolerance. Type of isotopic labeling, ion charge and ion adducts can also be selected. This search renders a list of all possible metabolites taking all possible isotopic combinations into account. Additionally, MS/MS spectra can be accessed when available.

3.1.1. Untargeted generation of MS/MS data using isotope-labeled microorganisms

Similar to METLIN, isoMETLIN fragmentation spectra was initially acquired on qToF instruments at different collision energies, using authentic isotope-labeled standards. However, a major shortcoming in populating a spectral database with MS/MS spectra of metabolite isotopologues is that the number of isotopomers increases with molecular weight (number of atoms) and that most isotopomers are not commercially available. To address this limitation, a novel approach using uniformly-labeled microorganisms was recently developed[23]. Briefly, two metabolite extracts were generated by growing Pichia pastoris in ¹²C-glucose- and ¹³C-glucose-containing media (Fig 5A). Before the acquisition of the fragmentation spectra, the labeling efficiency of the yeast metabolites was verified to be above 99%. These two unlabeled- and uniformly-labeled metabolite extracts were analyzed by different LC-MS platforms in an untargeted way. The unlabeled metabolites were grouped with all their isotopologues containing the isotopic trace and the MS/MS spectra of hundreds of fully labeled metabolites was generated (Fig. 5). It is worth mentioning that the reduced cost and time efforts are crucial advantages of this systematic workflow of generating MS/MS spectra for several hundreds of isotopic-labeled.

Fig. 5.

Systematic generation of MS/MS spectra of uniformly-labeled metabolites in isoMETLIN. Pichia pastoris was grown in unlabeled and ¹³C-labeled glucose, producing uniformly-labeled metabolites after several generations. Unlabeled and labeled metabolite extracts were analyzed by high-resolution untargeted metabolomics to generate pairs of unlabeled and labeled putative metabolites. Finally, the MS/MS spectra of identified pairs was incorporated into isoMETLIN after a careful curation.

3.1.2. Identification of metabolites using isotopes

An additional application of fully labeled MS/MS spectra is the gain of structural information for metabolite identification. By leveraging the mass differences between analogous fragments of isotopologues, the number of C atoms in each fragment can be determined. This information is of great interest for the identification of metabolites, whose MS/MS is not available, as it is the case in the identification of pseudouridine (Fig. 6A). Furthermore, two additional examples of identified metabolites using uniformly-labeled Pichia pastoris can be found in the work of Guijas, C. et. al[23].

Fig. 6.

(A) The MS/MS of an unknown metabolite is matched with the MS/MS of the Pichia pastoris extract unlabeled compound. To consider the Pichia pastoris pair of compounds for identification, the retention time of the unknown metabolite should match under the same analytical conditions. The number of carbons of each fragment can be used to determine the structures of both parent and fragment ions. In this example, pseudouridine was identified. (B) Proposed parallel analysis of microorganisms labeled with other stable isotopes. MS/MS spectra of fully-labeled known metabolites can be incorporated into isoMETLIN, whereas MS/MS spectra of unknown metabolites can be used for their identification as described in (A).

Currently, these spectra are being incorporated into isoMETLIN to aid researchers in the process of identifying of metabolites and the incorporation of analogous data generated by the analysis of microorganisms uniformly labeled with other isotopes, such as ¹⁵N and ³⁴S is a future goal. This will not only allow the addition of new MS/MS spectra of uniformly-labeled metabolites containing these two atoms, but it will also generate new layers of MS/MS spectra that can be complementary to the information provided by the ¹³C-labeled metabolites for the identification of metabolites (Fig. 6B).

3.2. METLIN-MRM

The METLIN-MRM library consists of a small-molecule transitions for multiple-reaction monitoring (MRM) for more than 15,500 unique small molecules. It was developed to streamline absolute quantitation, which typically is accomplished with triple-quadrupole (QqQ) mass spectrometers configured to monitor a particular set of precursor–product ion transitions. However, in order to determine the transitions for the different molecules of interest, each target molecule must be optimized with pure standard materials. In the library, three different types of transitions are available: 1) traditional experimentally optimized transitions, 2) computationally optimized experimental transitions and 3) public repository transitions (Fig. 7). Experimentally optimized transitions were acquired for more than 1,000 molecules in both positive and negative mode by following established protocols [24]. These small molecule transitions were optimized for the highest intensity to achieve low limits of detection.

Fig. 7.

METLIN-MRM ensembles three types of transitions: (1) transitions experimentally optimized by standard materials with QqQ via an established protocol, (2) transitions computationally (comp.) optimized by using METLIN’s MS/MS spectra and (3) and public repository transitions. Experimentally and computationally optimized transitions were optimized for sensitivity and selectivity, respectively. METLIN-MRM also serves as a public repository (PR), in which the community can populate the library with new transitions.

In addition to experimentally acquired data, transitions for more than 14,000 and 4,700 molecules in positive and negative mode, respectively, were computationally optimized by using the METLIN spectral library[23] (acquired at different CE on a qToF instrument) by ranking empirical MS/MS fragments according to their selectivity (uniqueness of a product fragment for a given molecule). The developed ranking algorithm compares the MS/MS spectra of compounds with precursors within a ±0.7 Da window and selects fragments with the best selectivity without compromising sensitivity. This strategy enables high-throughput quantitation analysis as transitions are no longer required to be optimized with standard reference materials and, by the same token, minimizes errors caused by interfering molecules as transitions less likely to be masked are selected. For more detailed information about the algorithms and the selection process, we refer the readers to the main published work[25].

Lastly, METLIN-MRM also serves as a public repository, in which community members can upload transitions. Submitted lists of transitions are assigned with a unique accession number that can be used as a reference for publications. This process ultimately facilitates the deposition of transitions used in experiments and scientific literature into a standardized and searchable database, thereby increasing the traceability and reproducibility of experiments and the reuse/sharing of optimized transitions. Currently, 3,300 transitions for more than 1,500 small molecules are available from peer-reviewed publications with their corresponding original source (doi).

4. METLIN population

METLIN contains experimental tandem mass spectrometry data on over 600,000 molecular standards and also has structures on over a million small molecules corresponding to a wide range of molecular classes. Over the years, small molecular entities have been incorporated into the library without bias for a particular class or type of compounds. Endogenous metabolites spanning the three-domain system (archaea, bacteria and eukarya), modified metabolites, synthetic drugs and toxicants. The growth of the MS/MS database has been exponential in the last year, growing from 14,000 in 2017 to over 600,000 in November 2019 in both positive- and negative-ion mode at multiple collision energies (Fig. 8)[26]. A growth curve we anticipate will continue into the future.

Fig. 8.

Experimental electrospray ionization tandem mass spectrometry data has been generated on over ~300 000 molecular standards and incorporated into METLIN. This data was generated in both positive and negative ionization modes and at multiple collision energies (0, 10, 20, and 40 V). The rapid increase in the rate of METLIN growth is the culmination of multiple analytical and informatic challenges being overcome in 2018. Analytical Chemistry 2018, 90, 13128–13129. Figure 1, with permission from ACS and RightsLink.

5. Conclusion

In this chapter, METLIN’s evolution since its beginnings, its tools for metabolite identification and future developments have been discussed. It has adapted and developed its tools to not only facilitate the identification of known compounds (with or without MS/MS spectra) but also the discovery of unknown compounds. In the coming years, hundreds of thousands of small molecules will be characterized using the same strict MS/MS protocols and data curation. Further, given the ubiquitous coupling of pre-ionization techniques and mass spectrometry instrumentation for untargeted metabolomics, tools for retention time prediction are currently being developed. Ultimately, it is safe to say that METLIN’s goal from its inception of facilitating metabolomics research has not changed, although its library and tools have continuously evolve to meet the changing needs of modern metabolomics research as well as other chemical entities.