TL;DR
This paper introduces a Bayesian inference framework to integrate multiomics data (proteomics, metabolomics) with metabolic models for optimizing microbial production of itaconate in Yarrowia lipolytica. By quantifying uncertainties in flux control coefficients (FCCs), they identify key enzymes like phosphoglycerate mutase and acetyl-CoA synthetase as promising targets for strain engineering to boost itaconate yields.
What Problem Does This Paper Address?
Economically viable production of bio-based chemicals and biofuels relies heavily on optimizing microbial cell factories. A major bottleneck in this process is precisely tuning enzyme expression to achieve desired metabolic pathway fluxes. This is challenging because data from different "omics" techniques (like proteomics and metabolomics) are often disparate and difficult to integrate with existing biological knowledge and metabolic models. Traditional methods for identifying rate-limiting enzymes and predicting the impact of genetic modifications are often point-estimation based and struggle with incomplete experimental data. This paper aims to bridge this gap by providing a robust, probabilistic approach to integrate multiomics data with metabolic models, enabling more accurate identification of key enzymes for strain engineering.
Methods
The core of this work is the application of Bayesian Metabolic Control Analysis (BMCA). BMCA is an extension of traditional Metabolic Control Analysis (MCA) that leverages modern probabilistic programming techniques to handle uncertainty in parameter estimation.
Key methodological aspects include:
- Linlog Kinetics: The authors use linlog kinetics to model reaction rates, which approximates Michaelis-Menten kinetics near a reference state and allows for more computationally feasible analysis compared to full ODE models.
- Probabilistic Framework: BMCA uses Bayesian inference to estimate parameters, providing posterior distributions rather than single point estimates. This allows for quantifying the uncertainty associated with these parameters.
- Integration of Multiomics Data: Proteomics data (enzyme expression levels) and metabolomics data (metabolite concentrations) are integrated into the BMCA framework.
- Flux Control Coefficients (FCCs): The primary output of interest is the FCC matrix, which quantifies how changes in enzyme expression affect the flux of specific metabolic reactions.
- Automatic Differentiation Variational Inference (ADVI): For computational efficiency, ADVI is used to approximate the posterior distributions of the model parameters.
Datasets
The study utilizes two multiomics datasets from Yarrowia lipolytica strains engineered for itaconate production:
- 9-sample dataset: This dataset consists of data from a laboratory IA-producing strain and a UV mutant strain, serving as an initial validation of the BMCA methodology.
- 24-sample dataset: This larger dataset includes data from four different strains (including knockouts) grown on two different carbon sources (glucose and malate), providing a more comprehensive evaluation.
The paper does not introduce new benchmark datasets but rather uses existing experimental data to validate their computational approach.
Key Insights and Ideas
- Uncertainty Quantification is Crucial: BMCA's ability to provide posterior distributions for FCCs is a significant advantage over traditional MCA. This allows researchers to identify targets with high confidence (narrow HPDs not overlapping zero) and avoid making decisions based on uncertain predictions.
- Central Metabolism is Key for Itaconate: The analysis highlights that key enzymes involved in central metabolism, rather than just the direct itaconate pathway enzymes, are critical for optimizing itaconate production. This suggests a holistic approach to strain engineering.
- Correlation vs. Causation (Mechanistic Insight): While correlation analysis (Table 2) can suggest relationships between enzyme expression and itaconate production, FCCs derived from BMCA offer a more mechanistic understanding of how enzyme modulation directly impacts flux. The paper emphasizes the value of FCCs for guiding strain design.
- Prior Knowledge vs. Data-Driven Insights: The comparison between prior (gray) and posterior (red) FCCs (Figure 3a) demonstrates how the data-driven BMCA approach refines initial assumptions and identifies novel targets.
Key Results
- Identification of Key Enzymes: The study identified several enzymes with significantly non-zero FCCs, indicating their potential as targets for increasing itaconate production. The top candidates from the 24-sample dataset include acetyl-CoA synthetase (ACSm), carbonic anhydrase (HCO3E), phosphoglycerate mutase (PGM), pyrophosphatase (PPAm), and homoserine dehydrogenase (HSDxi).
- Model Validation: Parity plots (Figure 2) showed good agreement between predicted and measured metabolite concentrations and protein expression levels, particularly for the 24-sample dataset, validating the BMCA model's predictive accuracy.
- Correlation Insights: Significant positive correlations were found between several enzymes and the itaconate boundary flux, with PGM showing both a high correlation and a significant FCC.
Limitations and Future Work
- Reduced Model Scope: The current study focused on a reduced central metabolic model of Y. lipolytica. While this allowed for a proof-of-concept, the authors acknowledge that a full genome-scale model is needed for broader analysis and more comprehensive predictions.
- Limited Experimental Flux Data: The model did not incorporate experimental flux data, relying on estimated flux values for the reference strain. The authors suggest that incorporating actual fluxomic data would improve model accuracy.
- Iterative Approach: The authors propose that their methodology is best used iteratively, especially when IA yield/titer measurements or more detailed fluxomic data become available for engineered strains.
- Causal Inference Engine: A future direction is the development of a causal inference engine to identify FCCs of engineered pathways using multiomics data and causal knowledge, enabling a deeper understanding of mediation in pathways.
- Simultaneous Modulation: The authors plan to focus on the simultaneous modulation of multiple enzymes identified by FCC analysis to further optimize bioproduct titer, rate, and yield.
Summary updated