The XCMS-METLIN Story

The origins of The Analytical Scientist’s 2023 Innovation Award-winning XCMS-METLIN platform trace back over three decades. The story begins in the early '90s with a bold idea from Richard Lerner, then president of Scripps. Richard wanted to explore the cerebrospinal fluid of animals in a sleep-deprived state, looking for endogenous metabolites that might induce sleep. My task was to identify these molecules. We initially used GC-MS for the analyses, but this didn’t give us the comprehensive data we needed.

That led to our first LC-MS-based metabolomic and lipidomic experiments, which culminated in several key publications (1,2). We discovered molecules correlating with the sleep-wake cycle, including one that induced a sleep-like state. But those early experiments were plagued by challenges. Data alignment and identification, in particular, posed significant barriers to progress. For example, retention time variability from run to run made it difficult to align data and discern real signals from noise. The complexity of molecular identification was another major hurdle, especially when manual methods were our only option.

The birth of XCMS – and nonlinear retention time alignment
 

The breakthrough came in the early 2000s when I challenged Colin Smith, a talented staff scientist, to improve our data analysis methods. The result? XCMS (3). XCMS introduced a novel concept: nonlinear alignment of LC/MS data. This allowed us to adjust for variations across experiments and vastly improved our ability to distinguish real signals from noise.

The first XCMS nonlinear correction plot (Figure 1) and algorithm, which addressed the challenge of LC/MS drift, has since become a widely emulated approach across the field. And given the large population of XCMS users, we are constantly listening to their thoughts on how XCMS can be improved. For example, pairwise analysis was always standard, however, the addition of XCMS single sample analysis and XCMS multi-group analysis came from user input.

Enter METLIN – the gold standard for molecular identification
 

Even with XCMS’s alignment solutions, we still faced the challenge of identifying the myriad peaks from LC/MS data. Accurate mass measurements alone proved unreliable because isomers and isobaric compounds – like glucose, lactose, and fructose – share identical molecular weights. We needed something more.

That "something" was tandem mass spectrometry (MS/MS), which provides an additional level of molecular characterization. This realization led us to create METLIN – a comprehensive MS/MS database.

Initially, METLIN was built by collecting known endogenous metabolite and lipid standards, generating MS/MS data at multiple collision energies (0, 10, 20, and 40 eV). Over the first decade, METLIN cataloged over 10,000 molecules, growing to almost 20,000 in the second decade. Today, METLIN hosts MS/MS data on over 935,000 molecular standards from over 350 classes of molecules (Figure 2), this represents exponential growth made possible by solving three key challenges:

• Acquiring molecular standards: We gathered a vast range of molecules from over 350 chemical classes to populate METLIN, the acquisitions occurred from individual labs, chemical companies, and pharmaceutical firms. (Special thanks to Avanti Polar Lipids for the vast store of lipids they provided (Figure 3).)
• Automating data acquisition and maintaining data quality: High-throughput analysis capabilities emerged after a major lab flood in 2017 destroyed several of our instruments. Ironically, this disaster enabled us to rebuild with even better equipment and higher efficiency, translating into quality data generation during high throughput analyses (Figure 4).
• Informatics integration: Aries Aisporna, a key team member, deserves much credit for creating informatics solutions that allowed us to simultaneously process molecular information, guide the analyses, and integrate everything into a user-friendly platform.

Winnie Uritboonthai also played a critical role, optimizing our mass spectrometry systems and processing over a million molecules with a success rate of around 80 percent. Thanks to her tireless efforts, METLIN has become the gold standard for MS/MS data, with experimental data for over 935,000 molecules at various collision energies.

XCMS-METLIN’s impact – science and beyond
 

If I had to distill the real value of XCMS and METLIN, it would come down to their impact on science. These platforms have catalyzed key breakthroughs:

• XCMS: Nonlinear alignment of LC/MS data (3)
• METLIN: Streamlined molecular identification (4,5,6)
• Activity Metabolomics: Enabled by XCMS and METLIN (7,8,9,10)
• Phantom Metabolites Unveiled: An intriguing discovery from METLIN (Figure 5) is that much of the LC/MS data we once treated as meaningful is, in fact, noise – caused by in-source fragmentation (ISF). Through METLIN’s unique acquisition of data at 0 eV, we have been able to distinguish true molecular ions from these fragments, simplifying our understanding of the metabolome and lipidome (11). 

Beyond its scientific contributions, XCMS-METLIN has had significant commercial impact. Early cloud-based versions of XCMS have raised concerns about data privacy, especially for industry users. In response, our local version of the new XCMS-METLIN platform (Figure 6), allows companies and institutes to process their data in-house with the major added advantage of streamlined molecular identification with an unrivaled database. 

But if there’s one thing I’ve learned throughout the development of XCMS-METLIN, it’s that no innovation happens in isolation. I’ve simply learned to listen and value the ideas of others –  especially from brilliant scientists like Colin, Aries, and Winnie. XCMS-METLIN is the culmination of a tremendous team effort.

^{Image credits: Supplied by Author}

Techniques & Tools Technology

The Analytical Scientist Innovation Awards 2023

12/12/2023 | 14 min read

Which analytical advances are set to open doors in 2023 – and beyond?