A diverse proteome is present and enzymatically active in metabolite extracts

The study addresses whether proteins persist in metabolite extracts and, if so, whether residual enzymatic activity post-extraction alters metabolite profiles, potentially confounding biological interpretation. Contextually, metabolomics relies on solvent-based extraction to precipitate proteins and other macromolecules, but extraction conditions (solvent composition, water content, pH, timing) vary widely and impact detection. The accepted view that protein precipitation is complete lacks broad empirical support; prior work reported 2–6% of serum proteins remain in the soluble metabolite fraction depending on solvent. The purpose here is to systematically evaluate extraction water content and common extraction modalities, quantify protein carryover, test for post-extraction enzymatic activity, and develop a workflow that improves polar metabolite coverage while preventing protein-mediated interconversions. This is important because metabolite-protein co-incubation during extraction and after resuspension could create an unrecognized observer effect, altering metabolite abundances and isotopologue distributions and jeopardizing accurate phenotype detection across diverse biological fields using metabolomics.

The authors highlight extensive protocol diversity in metabolomics extractions (e.g., 80% methanol; acetonitrile/methanol/water mixtures; Bligh–Dyer biphasic extractions) developed to balance metabolome coverage versus sample type constraints. Reporting and reproducibility remain challenging due to innumerable method variations. A prior report showed 2–6% of total serum proteins remain in the soluble fraction in a solvent-dependent manner, suggesting protein carryover may be more common than assumed. AMW (40% acetonitrile, 40% methanol, 20% water) and related protocols are widely used, but extraction water content varies in the literature and has not been systematically assessed at scale. These gaps motivate evaluating protein precipitation completeness, the impact of extraction water content on metabolite coverage, and whether residual proteins can drive post-extraction chemistry.

• Sample types: Mouse liver (primary model), brain, skeletal muscle, perigonadal adipose tissue, plasma, and human HEK293 (Phoenix-AMPHO) cells.
• Extraction modalities: AMW20 (40:40:20 acetonitrile:methanol:water); AMW with post-precipitation water titration to final 25–60% water (AMW25–AMW60); 80% methanol; Bligh–Dyer (aqueous and organic phases). Standard homogenization or vortex/sonication protocols were used with incubation on ice.
• Water titration experiment: Pooled cryopulverized mouse liver extracted with AMW20 followed by incremental water addition in 5% steps to evaluate effects on polar metabolite detection.
• Protein quantification and composition: BCA assays on extracts (with and without 3 kDa filtration). Quantitative proteomics by DIA LC–MS/MS (Orbitrap Eclipse with FAIMS); searched against mouse/human proteomes (Spectronaut). Gene set enrichment analysis (GSEA) via clusterProfiler using GO Biological Process; false discovery controlled at 1% or via FDR adjustments.
• Protein removal: Passive 3 kDa centrifugal filtration (Amicon Ultra-2 3K) applied to extract supernatants before drying for metabolomics.
• Metabolomics: Targeted LC–MS using three orthogonal methods: (1) Ion-paired reversed-phase in ESI negative mode (Orbitrap Exploris 240) with tributylamine; (2) HILIC (BEH Amide) in ESI positive; (3) reversed-phase (T3) in ESI positive (Orbitrap ID-X). Untargeted analyses by Compound Discoverer with mzCloud/NIST library matching and formula prediction; features filtered by blanks and quality metrics. Time-course injections (approx. every 4 h over 84 h) assessed time-dependent drift in AMW20/AMW50 ± filtration.
• Enzymatic activity assays: Post-resuspension stable isotope spike-ins: D5-glutamate; and [U-13C5,15N]-glutamate to trace transamination and futile cycling. Pan-transaminase inhibitor aminooxyacetic acid (AOA, 0.5 mM) added at resuspension for inhibition controls. Monitoring of isotopologues (e.g., D4-glutamate, 13C5-glutamate, 15N-glutamate, 13C5-αKG, 15N-aspartate) by high-resolution LC–MS; MS/MS to confirm deuterium/nitrogen loss locations. Quantitative calibration curves for glutamate, αKG, and aspartate used for absolute concentration estimates.
• Additional tests: Overnight -80 °C extract incubation versus standard 4 °C handling to assess effects on residual protein abundance/activity. Assessment of glutathione dynamics (GSH, GSSG, S-methyl-GSH) and discovery of glutathionylated derivatives via characteristic fragment ion (m/z 308.0912) in T3 ddMS2 data.
• Statistics: LIMMA eBayes for proteomics differential abundance; Benjamini–Hochberg FDR adjustments. MetaboAnalyst for metabolite statistics (ANOVA/t-test, FDR). Cheminformatics analysis of LogP versus recovery correlations using RDKit; Spearman correlations.

• Substantial protein carryover in metabolite extracts:
  • BCA suggested 11.3 and 15.7 µg protein per mg liver in AMW20 and AMW50 extracts, respectively, vs 153.8 µg/mg in whole liver lysates. After accounting for <3 kDa species via filtration, proteins >3 kDa measured 2.7 µg/mg (AMW20) and 4.1 µg/mg (AMW50).
  • DIA proteomics identified 1939 (AMW20) and 1760 (AMW50) proteins in extracts versus 5177 in whole liver. 1241 proteins overlapped across AMW20/AMW35/AMW50; 161, 187, and 191 were unique to AMW20, AMW35, AMW50 respectively. GSEA showed strong enrichment for metabolic processes (e.g., small molecule metabolic process).
  • Protein presence was ubiquitous across methods (AMW, 80% MeOH, Bligh–Dyer aqueous and organic phases) and across matrices (liver, adipose, brain, muscle, plasma, HEK cells). Even the Bligh–Dyer organic layer contained >1800 proteins enriched for metabolic functions.
• Extraction water content strongly alters metabolomics readouts:
  • Incremental water addition increased detection of polar metabolites, especially nucleotides; nucleotide triphosphates increased 5–8-fold between AMW20 and AMW30. Hydrophobic species (e.g., long-chain acyl-carnitines) decreased with higher water.
  • A significant inverse correlation between LogP and water-induced fold-change emerged at AMW30, but weakened at higher water, indicating extrinsic factors (e.g., proteins) affect recovery beyond intrinsic hydrophobicity.
• Direct evidence of post-extraction enzymatic activity:
  • D5-glutamate (added at resuspension) decreased with a concomitant rise in D4-glutamate in a water-dependent manner; D8-tryptophan was unaffected; extract pH unchanged. Proteomics detected 11 transaminases (e.g., GOT1), and pyridoxal-5P increased with water.
  • Both 3 kDa filtration and AOA (transaminase inhibitor) prevented D5→D4 glutamate conversion and preserved D5-glutamate over 24 h post-resuspension.
  • [U-13C5,15N]-glutamate tracing in AMW50 showed extensive transaminase-mediated futile cycling: an 80–90% decrease from the 82 µM spike with equimolar distribution to 13C5-glutamate and 15N products, including ~28 µM 15N-glutamate and ~37 µM 15N-aspartate; minor 13C3-αKG (~1 µM). Filtration or AOA prevented formation of labeled products; filtration (but not AOA) prevented 13C3-αKG accumulation. Mass spectrometry fragmentation supported specific α-carbon deuterium loss.
  • Overnight -80 °C incubation did not reduce protein abundance or abrogate transaminase activity compared to standard 4 °C handling.
• Untargeted time-course reveals broad protein-mediated drift:
  • Of 330 features passing QC, 298 exhibited significant time-dependent changes in at least one group; 47 features were unique to AMW50. PCA showed largest injection-order drift in unfiltered AMW50 (PC1 56.4%); accumulated Euclidean drift ~13.1 vs 5.0 (AMW20), 3.8 (AMW20F), 4.8 (AMW50F).
  • De novo formation of γ-glutamyl-glutamate (Glu–Glu) observed in AMW50 over time; deuterium patterns indicated only one glutamate moiety sourced from the free pool, implicating enzymatic/protein sources and abolished by filtration.
  • Glutathione dynamics: time-dependent GSH depletion and complex GSSG/Me-GSH changes suggested enzymatic turnover. Proteomics detected 25 glutathione-metabolizing enzymes including GSR and multiple GSTs. T3 ddMS2 revealed 14 precursor ions giving 67 chromatographic peaks with the glutathione fragment ion in AMW50 but absent in AMW50F, indicating a protein-mediated glutathione sink.
• Workflow solution: High-water AMW with 3 kDa filtration (AMW50F)
  • Improves polar metabolite recovery (e.g., nucleotides and other charged metabolites up 4–25-fold vs AMW20) while preventing enzyme-mediated conversions.
  • Filtration reduced long-chain polar lipids (likely due to membrane binding) and prevented formation of non-endogenous products (e.g., Glu–Glu) and protein-mediated depletion (e.g., nucleosides/nucleobases, pyridoxal→pyridoxal-5P). NADPH detection was >200-fold greater in AMW50F vs AMW50, indicating protein-mediated depletion in unfiltered extracts.
• Across matrices, ATP signals increased with water addition, and GOT1/GOT2 proteins were detected in most sample types (except plasma); skeletal muscle showed in-extract transaminase activity by D1/D5 glutamate readouts.

The findings demonstrate that metabolite extraction does not fully remove proteins; instead, extracts contain a diverse proteome enriched in metabolic enzymes that can remain or become catalytically active post-extraction and after resuspension. This resolves the observed discrepancies between compound hydrophobicity and recovery and explains time-dependent signal drift as an enzymatic/protein-binding phenomenon rather than solely intrinsic compound instability. Demonstrating transaminase-mediated futile cycling and glutathione metabolism directly links residual enzymes to post-extraction metabolite interconversions, with clear implications for both absolute quantification and isotopologue distributions in stable isotope tracing experiments. The work underscores the risk of observer effects in metabolomics, whereby extraction conditions and residual proteins can create artifactual metabolite patterns unrelated to in vivo biology. Incorporating a simple 3 kDa filtration step into high-water AMW extraction effectively removes proteins, stabilizes metabolite signals, expands coverage of polar metabolites (notably nucleotides and redox cofactors), and prevents protein-mediated confounding. This approach supports more faithful biological phenotype detection and may reduce the need for extensive post hoc drift correction. The study also highlights potential analytical impacts of proteins on LC systems (e.g., retention, ion suppression), reinforcing the value of protein removal for method robustness.

This study reveals that metabolite extracts contain a broad, metabolically enriched proteome that can catalyze post-extraction reactions, altering metabolite abundances and isotopologue patterns. The authors provide mechanistic evidence (e.g., transaminase futile cycling, glutathione turnover) and show that a practical workflow—AMW extraction with increased water content combined with 3 kDa filtration—improves polar metabolite coverage while removing proteins and preventing enzymatic interconversions. These insights identify a previously unrecognized observer effect in metabolomics and offer a straightforward mitigation strategy. Future work should optimize filtration-based workflows for polar lipid analysis, systematically investigate protein–metabolite binding versus catalysis in extracts, and evaluate the impacts in additional matrices, extraction chemistries, and resuspension solvents, particularly for stable isotope tracing and redox metabolite quantification.

• The 3 kDa filtration step, while effective for protein removal, introduces an additional handling step that may add variability and some sample loss; long-chain polar lipids (e.g., acyl-carnitines, lysophospholipids) appear depleted, likely via non-specific binding to filter membranes, limiting lipid coverage without further optimization.
• Enzymatic activity was demonstrated primarily in extracts resuspended in water for reversed-phase analysis; behavior in other resuspension solvents (e.g., high-organic for HILIC/amide) was not directly examined here.
• The extent to which time-dependent changes arise from enzymatic catalysis versus protein–metabolite binding was not fully delineated; proteases/nucleases present could liberate additional metabolites, further complicating attribution.
• Effects are extraction-modality and matrix dependent; generalization across all protocols and sample types will require broader validation.
• Most experiments used pooled tissues with n=3 technical replicates per group; biological variability across individual subjects was not the focus.