LOF Metric Reduces 79–86% of LC–MS/MS Features by Linking Degenerate Peaks via Shape Consistency

Progress in liquid chromatography-tandem mass spectrometry (LC–MS/MS) instrumentation and chemometric tools has accelerated untargeted metabolomics, yet the resulting data sets remain daunting: thousands of signals appear, but only a handful can be named. The surplus consists largely of degenerate features—multiple peaks spawned by one metabolite through adducts, in-source fragments, and natural isotopes. To untangle this redundancy, researchers from the University of Michigan (Ann Arbor, Michigan) have introduced lack-of-fit (LOF), a peak-shape congruence measure that gauges the residual mismatch between candidate signals within a narrow retention window; pairs scoring below 20% are deemed degenerate and grouped together.

LCGC International spoke to Caitlin Cain, lead author of the paper presenting this work,¹ about her group’s research.

In untargeted liquid chromatography-mass spectrometry (LC–MS) metabolomics, how do chromatographic resolution and mass spectral resolving power jointly influence the ability to distinguish unique metabolites from degenerate features such as adducts and in-source fragments?

In an ideal untargeted metabolomics experiment, LC–MS would separate all metabolites in a sample into fully resolved peaks. Each metabolite would then be detected as a single ion species, which can be either its protonated ([M+H]⁺) or deprotonated ([M-H]^-) form. In practice, however, metabolites frequently generate additional ion species, including salt adducts, in-source fragments, and isotopologues (e.g., degenerate features). Poor chromatographic resolution or insufficiently mass spectral resolving power exacerbates this issue by increasing coelution and ion overlap, causing signals from multiple metabolites and their associated forms to appear simultaneously.

What are the limitations of correlation-based feature clustering methods (CAMERA or MS-CleanR) when analytes coelute at low chromatographic resolution (Rs ≤ 0.3) and how does this impact metabolite annotation confidence?

Conventional correlation-based clustering approaches often fail at low chromatographic resolutions (R_s ≤ 0.3) because they primarily rely on signal co-variation. Hence, overlapped metabolites will have high co-variation since they share similar retention times. As a result, features from different metabolites could erroneously be grouped together.

How does the lack-of-fit (LOF) metric leverage chromatographic peak shape and MS signal behavior differently from Pearson correlation coefficients when identifying degenerate LC–MS features?

The LOF metric addresses the co-variation limitation of these correlation-based approaches by directly evaluating chromatographic peak shape fidelity. Here, two features are determined to come from the same metabolite if they quantitatively share a similar elution profile (for example, retention times, peak widths, peak symmetry, etc.).

Why is data-dependent acquisition (DDA) inherently biased toward high-abundance features in liquid chromatography-tandem mass spectrometry (LC–MS/MS) and how do chromatographic separation strategies or iterative DDA workflows help mitigate this bias?

Data-dependent acquisition (DDA) prioritizes precursor ion selection based on the MS1 intensity, inherently favoring high-abundance ions while undersampling low-abundance metabolites. Coelution exacerbates this bias, as dominant species repeatedly trigger MS/MS scans. Improved chromatographic resolution reduces precursor competition by spreading analytes across the time axis. Iterative DDA is an alternative approach that utilizes repeated injections and exclusions lists to progressively target unsampled features, increasing MS/MS coverage without altering the original chromatographic separation.

From a chromatographic standpoint, how do separation time and sample preconcentration affect feature degeneracy, ion suppression, and the depth of metabolome coverage in LC–MS analyses?

Longer separation times act as our first approach to increase peak capacity, improving chromatographic resolution and decreasing ion suppression. In turn, these improvements should translate to deepening our metabolome coverage. Sample preconcentration can also enhance our view of the metabolome by increasing the sensitivity for natively low-abundance metabolites. However, longer separation times and preconcentrated samples can also increase the presence of degeneracies.

How do in-source fragmentation and adduct formation complicate MS1 feature lists, and what chromatographic or ionization conditions can exacerbate or reduce these degeneracies?

Both in-source fragmentation and adduct formation can inflate feature lists and obscure true metabolite counts in an untargeted experiment. In-source fragmentation is commonly seen with high MS source voltages and elevated temperatures, whereas high salt content in the sample and poor chromatographic resolution can cause adduct formation. Therefore, a focus should be placed on using softer MS ionization conditions, desalting samples, and increasing chromatographic resolutions to reduce these degeneracies.

Compare the use of chemometric decomposition methods such as multivariate curve resolution-alternating least squares (MCR-ALS) with LOF clustering for resolving overlapping chromatographic peaks. What are the strengths and weaknesses of each approach in LC–MS metabolomics?

MCR-ALS is a chemometric decomposition technique to mathematically resolve overlapped chromatographic signals. MCR-ALS excels when different metabolites share nearly identical retention times and peak shapes but different in their mass spectral signatures. However, MCR-ALS can also produce multiple solutions (i.e., modeling ambiguity), which are worsened at low chromatographic resolutions, and is time intensive to scale across large untargeted data sets. LOF clustering is computationally efficient and directly leverages the chromatographic data without using complicated modeling efforts to detect and resolve overlapped metabolites.

Why is the absence of predefined adduct lists a significant advantage for LOF-based clustering in untargeted metabolomics, particularly when analyzing complex biological matrices like brain dialysate?

LOF clustering does not rely on predefined adduct or fragment lists, which are often incomplete and matrix-dependent. This is particularly advantageous in complex biological samples like brain dialysate, where we observed several unique in-source fragments and dimers with background ions. If we were relying on predefined lists, we could have labeled those degenerate features as unique chemical signals, artificially inflating our true metabolite count.

How does applying a degeneracy-reduction strategy at the MS1 level improve downstream MS/MS spectral matching and database-based metabolite identification?

Reducing degeneracy at the MS1 level improves precursor selection by ensuring that MS/MS spectra are triggered on unique chemical entities (ideally those [M+H]⁺ or [M-H]^- ions) rather than redundant adducts/fragments. In turn, this leads to cleaner spectra and more accurate database annotations.

In what ways could LOF clustering be adapted or validated for data-independent acquisition (DIA) workflows, given the increased spectral complexity compared to DDA in LC–MS/MS?

In DIA, multiple precursors are fragmented simultaneously, increasing spectral complexity and fragment interference. LOF clustering could be adapted to evaluate whether MS/MS traces for a fragment ion conform to the same peak shape as observed in the MS1-level data. However, more work into this topic is needed to validate our idea.

Reference

1. Cain, C. N.; Ma, C. H.; Popov, P. et al. Improving the Detection of Analyte Degeneracies in Untargeted Liquid Chromatography-Tandem Mass Spectrometry Data. Anal. Chem. 2025, 97 (48), 26439-26448. DOI: 10.1021/acs.analchem.5c03704