OptRAM: In-silico strain design via integrative regulatory-metabolic network modeling

Fangzhou Shen, Renliang Sun, Jie Yao, Jian Li, Qian Liu, Nathan D. Price, Chenguang Liu, Zhuo WangPLOS Computational Biology 2019 · DOI

TL;DR

This paper introduces OptRAM, a novel computational tool for designing microbial strains for industrial production. By integrating gene regulatory networks with metabolic models, OptRAM can predict complex genetic modifications, including gene overexpression, knockdown, and knockout, to optimize the production of desired chemicals while maintaining cell growth.

What Problem Does This Paper Address?

Metabolic engineering aims to create microbial "cell factories" for producing valuable compounds efficiently. While computational tools based on genome-scale metabolic models (GEMs) have been instrumental, they often overlook the crucial role of transcriptional regulation. Existing methods that do incorporate regulation typically rely on pre-existing, manually curated regulatory networks, limiting their applicability to well-studied organisms and often treating gene regulation in a binary (on/off) manner. This gap hinders the design of truly optimized strains, as gene expression plays a vital role in fine-tuning cellular metabolism under specific conditions. The authors address this by developing a method that infers regulatory networks and then uses this integrated information to propose more comprehensive strain designs.

Methods

OptRAM is built upon the IDREAM framework, which integrates inferred regulatory networks with metabolic models. The core of OptRAM is a simulated annealing algorithm that explores a vast combinatorial space of genetic modifications. What's novel here is:

  • Integrative Network: It combines an inferred transcriptional regulatory network (TRN) with a genome-scale metabolic model (GEM). The TRN is built using gene expression data and statistical methods (FDR cutoff) to infer TF-gene interactions.
  • Comprehensive Mutations: OptRAM can propose overexpression, knockdown, and knockout of both metabolic genes and transcription factors (TFs).
  • Novel Objective Function: A new objective function is introduced to ensure a favorable coupling between desired chemical production and cell growth. This function aims to maximize product yield while minimizing flux variability and ensuring robust production.
  • Systematic Evaluation: OptRAM includes a multi-faceted evaluation of candidate solutions, considering essential genes (to avoid lethal mutations), implementation cost (based on network path analysis), and overall genetic adjustment from the wild type (using Cosine similarity of flux vectors).

Datasets

The study utilizes:

  • Yeast Metabolic Network: The Yeast 7.6 metabolic reconstruction, a standard model for Saccharomyces cerevisiae.
  • Gene Expression Data: 2929 microarray datasets for S. cerevisiae were used to infer the transcriptional regulatory network.
  • Essential Gene Databases: SGD (Saccharomyces Genome Database) and DEG (Database of Essential Genes) were used to filter out lethal mutations.
  • LASER Database: This database of reported successful modifications was used to validate predicted gene targets.

Key Insights and Ideas

  • Regulation is Key: The paper strongly emphasizes that ignoring transcriptional regulation is a significant limitation in current strain design, as it dictates how metabolic potential is realized under different conditions.
  • Inferred Networks are Powerful: OptRAM's ability to infer regulatory networks from gene expression data makes it broadly applicable to organisms lacking pre-existing detailed TRNs.
  • Coupling Growth and Production: The novel objective function is crucial for ensuring that engineered strains not only produce more of the target compound but also maintain sufficient growth, a critical factor for industrial feasibility.
  • Holistic Evaluation: Moving beyond just maximizing yield, OptRAM's evaluation of essential genes, implementation cost, and genetic distance provides a more practical and experimentally relevant set of design recommendations.

Key Results

OptRAM was applied to three case studies in yeast: succinate, 2,3-butanediol, and ethanol production.

  • Improved Production: In all cases, OptRAM predicted strain designs that significantly increased theoretical production rates compared to wild-type and, in some instances, outperformed other computational methods like OptFlux and literature-based designs. For example, succinate exchange flux increased about 66-fold theoretically.
  • Validation: Many of the predicted gene and TF targets were found to be consistent with experimentally validated modifications in the LASER database.
  • Experimental Validation: The authors experimentally validated their ethanol production strain design, showing increased ethanol production with slightly reduced growth, confirming the tool's predictive power.
  • Comparison: OptRAM demonstrated superior performance in coupling target production with biomass growth, especially under strict biomass constraints, compared to other methods.

Limitations and Future Work

  • Kinetic Parameters: The authors acknowledge that incorporating kinetic parameters into the models could further refine predictions, allowing for more precise flux ranges.
  • Allosteric Regulation: Including other forms of regulation, such as allosteric regulation, could enhance the model's comprehensiveness.
  • Experimental Validation: While some experimental validation was performed, the authors stress the need for more extensive experimental validation across diverse scenarios to iteratively improve the method.
  • Computational Cost: While comparable to other advanced methods, the simulated annealing process can still be computationally intensive.

Summary updated

BibTeX
@article{shen2019optram,
  author = {Fangzhou Shen and Renliang Sun and Jie Yao and Jian Li and Qian Liu and Nathan D. Price and Chenguang Liu and Zhuo Wang},
  title = {{OptRAM: In-silico strain design via integrative regulatory-metabolic network modeling}},
  journal = {PLOS Computational Biology},
  year = {2019},
  doi = {10.1371/journal.pcbi.1006835},
}
Figure