In silico method for modeling metabolism and gene product expression at genome scale

1. In silico method for modeling metabolism and gene product expression at genome scale Lerman, Joshua A., Palsson, Bernhard O. Nat Commun 2012/07/03

2. So far – Metabolic Models (M-models) • Predict reaction flux • Genes are either ON or OFF • Special ‘tricks’ to incorporate GE (iMAT) • ‘tricks’ are imprecise, more tricks needed (MTA) • Objective function debatable • Usually very large solution space • Flux loops are possible leading to unrealistic solutions. • No regulation incorporated

3. New – Metabolism and Expression(ME-models) • Add transcription and translation • Account for RNA generation and degradation • Account for peptide creation and degradation • Gene expression and gene products explicitly modeled and predicted • All M-model features included • GE and proteomic data easily incorporated • No regulation incorporated.

4. ME-model: the details

5. The Creature • Model of the hyperthermophilicThermotogamaritime (55-90 °C) • Compact 1.8-Mb genome • Lots of proteome data • Few transcription factors • Few regulatory states…

6. Adding Transcription and Translation to Model

7. Modeling Transcription(Decay and Dilution of m/t/r-RNA) • Flux creating mRNA: (GE) • Fluxes deleting mRNA: • (mRNA transferred to daughter cell) • (NTPNMP) • Controlled by two coupling constants: • (mRNA half life, from lab measurements) • (lab measured or sampling) • Fluxes are coupled: • Means 1 mRNA must be removed for every times it is degraded • Cell spends energy in rebuilding NMPNTP

8. Modeling Translation:mRNAEnzymes • Flux creating peptides: • Translation limited by , upper bound on rate of single mRNA translation, estimated from protein length, ribosome translation-frame and tRNA linking rate (global) • Fluxes are coupled: • Means 1 mRNA must be degraded every times it is translated

9. Modeling Reaction Catalysis • (Michaelis-Mentenkinetics) • is turnover number • is complex concentration • is substrate concentration • is substrate-catalyst affinity • Assume • Means one complex must be removed for every times it catalyzes • Whole proteome synthesized for doubling • Fast catalysis faster doubling (dilution)

10. Building the Optimization Framework

11. M-model - reminder Total Biomass Reaction: • Experimentally measure lipid, nucleotide, AA, growth and maintenance ATP • Integrate with organism to define reaction approximating dilution during cell formation • Cellular composition known to vary with • Cellular composition known to vary with media • LP used to find max growth subject to (measured) uptake rates

12. ME-model Structural Biomass Reaction: • Account only for “constant” cell structure • Cofactors like Coenzyme A • DNA like dCTP, dGTP • Cell wall lipids • Energy necessary to create and maintain them • Model approximates a cell whose composition is a function of environment and growth rate • Cellular composition (mRNA, tRNA, ribosomes) taken into account as dynamic reactions • LP used to identify the minimum ribosome production rate required to support an experimentally determined growth rate

14. Validation

15. RNA-to-Protein Mass Ratio • RNA-to-protein mass ratio (r) observed to increase as a function of growth rate (μ) • Emulate range of growths in minimal medium • UseFBAwith LP to identify minimum ribosome production rate required to support a given μ • Assumption: expect a successful organism to produce the minimal amount of ribosomes required to support expression of the proteome • Consistent with experimental observations, ME-Model simulated increase in r with increasing μ

16. Comparison to M-model • max biomass on minimal media, many solutions • Sample and approx. Gaussian, chance of finding solution as efficient as ME-model. • Can be found by minimizing total flux (many solutions stem from internal flux loops).

17. Optimal pathways in ME-model • Produces small metabolites as by-products of GE • Accounts for material and energy turnover costs • Includes recyclingS-adenosylhomocysteine, (by-product of rRNA and tRNAmethylation) and guanine, (by-product of tRNAmodification) • Frugal with central metabolic reactions, proposes glycolytic pathway during efficient growth • M-Model indicates that alternate pathways are as efficient

18. Blue – ME-model paths, Gray – M-model alternate paths

19. System Level Molecular Phenotypes • Constrain model to μ during log-phase growth in maltose minimal medium at 80 °C • Compare model predictions to substrate consumption, product secretion, AA composition, transcriptome and proteome measurements. • Model accurately predicted maltose consumption and acetate and H2secretion • Predicted AA incorporation was linearly correlated (significantly) with measured AA composition 

20. Driving Discovery • Compute GE profiles for growth on medium: L-Arabinose/cellobioseas sole carbon source • Identify conditionally expressed (CE) genes - essential for growth with each carbon source • In-vivo measurements corroborate genes found in simulation – evidence of tanscript. regulation • CE genes may be regulated by the same TF • Scan promoter and upstream regions of CE genes to identify potential TF-binding motifs • Found high-scoring motif for L-ArabCE genes and a high-scoring motif for cellobiose CE genes • L-Arab motif similar to Bacillus subtilisAraR motif  

21. Summary

22. Advantages • Because ME-Models explicitly represent GE, directly investigating omics data in the context of the whole is now feasible • For example, a set of genes highly expressed in silico but not expressed in vivo may indicate the presence of transcriptional regulation • Discovery of new TF highlights how ME-Model simulations can guide discovery of new regulons

23. Downsides • ME-model is more intricate then M-model, more room for unknown/incomplete knowledge • May keep ME-model simulations far from reality on most organisms • Lack of specific translation efficacy for each protein • Lack of specific degradation rates for each mRNA • lack of signaling • Lack of regulatory circuitry

24. Thank You