TRACE

TRACE is an integrated experimental–computational framework for untargeted LC–MS metabolomics. It uses complete ¹³C/¹⁵N labeling to validate biological metabolites, then maps how competing unlabeled nutrients enter metabolism by isotopic dilution rather than label enrichment.

The work is published in Analytical Chemistry (doi:10.1021/acs.analchem.6c02292). Software documentation is at drruili.github.io/TRACE. TRACE runs on MSdev objects and xcms peak tables.

Why TRACE

Untargeted LC–MS can survey microbial metabolism at scale, but two problems remain tightly coupled:

1. Which peaks are biological? Modern instruments report tens of thousands of features. Many are adducts, in-source fragments, or chemical contaminants rather than metabolites made by the cell.
2. Which nutrients are used when several are present? Classical isotope tracing follows one labeled substrate at a time. It does not ask how cells choose among competing carbon and nitrogen sources.

Existing tools, including our earlier single-peak method PAVE, validate features in isolation. Fixed m/z and retention-time tolerances then generate large numbers of false connections as sensitivity increases, and they struggle to group the many MS forms of one metabolite.

TRACE addresses both problems with one design: fully label cells with ¹³C-glucose and ¹⁵N-ammonium, treat isotope-confirmed peaks as network seeds, calibrate error tolerances from those seeds, then map unlabeled nutrient fate by isotopic dilution.

Experimental design

Saccharomyces cerevisiae CEN.PK was grown for more than ten generations in minimal medium with [U-¹³C₆]-glucose and [U-¹⁵N₂]-(NH₄)₂SO₄ as the sole carbon and nitrogen sources, so de novo metabolites carry predictable CN mass shifts.

Four tracer groups define the seeds:

For nutrient competition, unlabeled supplements (leucine, threonine, tryptophan, adenine, uracil, acetate, or a 14-compound mixture) were added on this labeled background. Incorporation of an unlabeled nutrient dilutes the ¹³C/¹⁵N labeling of downstream metabolites.

From seeds to a calibrated network

TRACE first keeps coeluting peak pairs whose m/z differences match theoretical ¹³C/¹⁵N shifts, then requires a complete unlabeled / ¹⁵N / ¹³C / dual-label CN bundle whose intensities match the expected 1:1:1:1 pattern (Pearson ρ > 0.75).

In a representative positive-mode data set this collapsed ~25,000 features and hundreds of millions of pairs to 1,098 high-confidence bundles. True CN links cluster tightly around zero m/z error and RT shift. TRACE fits that cluster as a Gaussian signal against a nonparametric background and uses the central 99% of the signal as dynamic tolerances, then propagates adduct, fragment, and isotope edges only inside those bounds.

Global assignment and annotation

Edges are typed as isotope, fragment, or adduct and scored as chemical transformations (for example [M+H]⁺ → [M+Na]⁺ is H₋₁Na₁). TRACE then partitions seeds into compatible subnetworks so that one metabolite and its MS forms share a seed, instead of resolving conflicting pairwise assignments.

Against an authentic-standard library, TRACE removed PAVE hits with large m/z or RT error, corrected adduct mis-assignments caused by pairwise conflicts, and recovered features that PAVE had missed. About 70% of TRACE-annotated peaks matched HMDB, YMDB, KEGG, or an in-house database; about 27% of candidates matched in-house RT standards.

More sensitivity is not more biology

The same labeled yeast extract was analyzed on four Orbitrap platforms (Q Exactive Plus, Exploris 480, Astral, Excedion Pro) and at several resolutions. Total features rose by 331% from QE Plus to Excedion Pro, while TRACE-validated metabolites rose only 117%. The share of high-confidence biological annotations fell from 2.94% to 1.48%. Extra peaks are mostly non-biological unless annotation is specific.

Nutrient fate under competition

With formulas and CN networks fixed from the four tracer groups, TRACE maps nutrient conditions onto that network. For each metabolite, the labeling fraction is the abundance-weighted ¹³C or ¹⁵N atom count; the absorption fraction is the drop in labeling relative to the fully labeled control.

The complete mixture contributed carbon and nitrogen across sulfur, purine, and amino-acid metabolism. Individual supplements were pathway-selective: tryptophan into tryptophan metabolism, adenine into purine metabolism, threonine into serine/glycine metabolism, acetate as C₂/C₄/C₆ units.

Two molecules test the method at opposite extremes of pathway architecture:

• Glutathione (GSH) is a linear tripeptide. Its isotopologues tracked glutamate, glycine, and cysteine, including acetate-derived carbon in the glutamate moiety and threonine-derived C₂N₁ in glycine.
• NAD⁺ is a hub (adenine + two riboses + nicotinamide). Unlabeled adenine produced both salvage (C₅N₀) and ongoing de novo (C₁₀N₅) forms; unlabeled tryptophan diluted nicotinamide via kynurenine; a C₀N₁ nicotinamide form indicated glutamine nitrogen entering independently of the carbon backbone.

Software

A typical analysis loads a pre-processed MSdev object (xcms peaks with sample.type set to the four isotope groups), runs the pipeline, and exports tables:

library(MSdev)
library(TRACE)

obj <- TRACE_workflow(
  obj,
  rt.tol = 10,
  ppm = 5,
  cpdb = "path/to/trace.cp.db.xlsx"
)

TRACE_export(obj, file = "TRACE_results.xlsx")

Install the article snapshot with devtools::install_github("DrRuiLi/TRACE", ref = "v1.0.0"). See Get started for the full workflow.

Scope

TRACE is built for systems in which complete isotopic labeling (the 1:1:1:1 CN bundle) is achievable, as in yeast grown for many generations on labeled glucose and ammonium. It is not a drop-in seed detector for mammalian cells. High-confidence TRACE annotations can still be transferred by running mammalian samples in the same batch as fully labeled microbial references and aligning features.