**Background:** Approximately 20% of Saccharomyces cerevisiae proteins remain poorly characterized, many with human homologs. The diauxic shift—the metabolic transition from glucose fermentation to ethanol respiration—requires substantial metabolic network reconfiguration and is poorly understood despite its relevance to cancer metabolism (Warburg effect). Previous work using dynamical flux balance analysis (dFBA) with semi-automated gene regulatory models identified ten regulatory genes with the largest predicted growth rate discrepancies between models (M1 and M1Smart). This study investigates whether untargeted intracellular metabolomics can generate functional hypotheses for these genes and validate model revisions.
**Methods:** Ten haploid deletion mutants (BY4741 background) were selected from the EUROSCARF collection based on the highest absolute post-shift growth rate differences between M1 and M1Smart models (range: 0.00251 h⁻¹ for YGR067C to 0.000427 h⁻¹ for DLD3). Strains were cultured in YNB medium with 1.25 g/L glucose in 384-well plates using the automated laboratory cell 'Eve'. Samples were collected pre-shift (0.5 OD₅₆₀, fermentative phase) and post-shift (1.0 OD₅₆₀, respiratory phase). Metabolite extraction used 75% ethanol at 95°C. Untargeted LC-qTOF-MS analysis was performed on an Agilent UHPLC-qTOF-MS system with reverse-phase chromatography (Waters Acquity UPLC HSS T3 column) and positive-mode ESI. Peak processing used MS-Dial (v4.7), with identification by manual comparison to the Riken library. After QC-RLSC drift correction and quantile normalization, 431 distinct metabolite features remained. Discriminatory analysis used oPLS-DA and LIMMA linear modeling via MetaboAnalyst (v5.0.0). Pathway enrichment used FELLA (v1.14.0) with diffusion-based methods (p-score < 0.05). For model validation, metabolite flux-sums (Φ) were calculated from dFBA simulations and compared to measured accumulations using balanced accuracy.
**Key Results:** The diauxic shift produced clearly separable metabolic profiles (oPLS-DA classification), with 215 of 431 metabolic features significantly affected (raw p-value ≤ 0.1, linear model). FELLA enrichment (p-score < 0.05) identified significant pathway changes in central carbon metabolism, glutathione metabolism, proline metabolism, glycerophospholipid metabolism, and tryptophan metabolism. Key accumulated metabolites included L-glutamate, spermidine, glutathione, 5-methylthioadenosine, and 5-oxoproline in the glutathione-proline metabolism intersection. Choline, phosphocholine, and 3-sn-glycerophosphocholine showed significant changes in glycerophospholipid metabolism. Kynurenic acid and indole-3-acetate were altered in tryptophan metabolism while tryptophan itself remained unperturbed.
STRAIN-SPECIFIC RESULTS
YGR067C and TDA1 showed large metabolic effects with low correlation to WT, affecting sphingolipid and nucleotide metabolism in both phases. TDA1 deletion significantly altered sphinganine and phytosphingosine levels. YGR067C additionally perturbed central carbon metabolism (citric acid cycle, C5-branched dibasic acid metabolism) post-shift. RTS3 affected valine, leucine, isoleucine, and tryptophan metabolism. MEK1 deletion impacted stress response pathways. FAA1 deletion affected cell-membrane maintenance and TCA cycle in the glucose phase. OCA1 and PCL1 deletions shared similarity in purine and tryptophan pathways. RME1 deletion perturbed TCA cycle, valine, and isoleucine synthesis post-shift, and glutathione metabolism pre-shift. GAL11 showed high correlation with WT in both phases, indicating minimal impact on diauxic shift metabolism.
MODEL VALIDATION
M1Smart strictly outperformed M1 on 6 of 10 selected metabolites. Balanced accuracy improved for L-Glutamate (0.265 to 0.478), L-Tryptophan (0.250 to 0.413), Adenine (0.000 to 0.469), Guanine (0.406 to 0.544), L-Proline (0.300 to 0.325), and L-Histidine (0.289 to 0.316). Neither model predicted L-Leucine or Uracil activity (balanced accuracy 0.000 for both). M1 outperformed M1Smart for Adenosine (0.644 vs 0.544) and PEP (0.507 vs 0.382).
**Clinical Implications:** While this is a basic yeast biology study, the findings have translational relevance. The diauxic shift parallels the Warburg effect in cancer cells, where metabolic reprogramming from oxidative to glycolytic metabolism occurs. Understanding regulatory genes controlling metabolic transitions could inform cancer metabolism research. The human homologs of FAA1 (fatty acid activation) and DLD3 (dehydrogenase) are relevant to metabolic disorders. The demonstration that untargeted metabolomics can be deployed in high-throughput, automated workflows supports its potential for accelerating functional gene discovery in human-relevant systems. The study also validates semi-autonomous model revision approaches that could be applied to human metabolic network models.