Experimental Evolution Reveals Unifying Systems-Level Adaptations but Diversity in Driving Genotypes.

Genotype-fitness maps of evolution have been well characterized for biological components, such as RNA and proteins, but remain less clear for systems-level properties, such as those of metabolic and transcriptional regulatory networks. Here, we take multi-omics measurements of 6 different E. coli strains throughout adaptive laboratory evolution (ALE) to maximal growth fitness. The results show the following: (i) convergence in most overall phenotypic measures across all strains, with the notable exception of divergence in NADPH production mechanisms; (ii) conserved transcriptomic adaptations, describing increased expression of growth promoting genes but decreased expression of stress response and structural components; (iii) four groups of regulatory trade-offs underlying the adjustment of transcriptome composition; and (iv) correlates that link causal mutations to systems-level adaptations, including mutation-pathway flux correlates and mutation-transcriptome composition correlates. We thus show that fitness landscapes for ALE can be described with two layers of causation: one based on system-level properties (continuous variables) and the other based on mutations (discrete variables). IMPORTANCE Understanding the mechanisms of microbial adaptation will help combat the evolution of drug-resistant microbes and enable predictive genome design. Although experimental evolution allows us to identify the causal mutations underlying microbial adaptation, it remains unclear how causal mutations enable increased fitness and is often explained in terms of individual components (i.e., enzyme rate) as opposed to biological systems (i.e., pathways). Here, we find that causal mutations in E. coli are linked to systems-level changes in NADPH balance and expression of stress response genes. These systems-level adaptation patterns are conserved across diverse E. coli strains and thus identify cofactor balance and proteome reallocation as dominant constraints governing microbial adaptation.

Adaptive laboratory evolution for improved tolerance of isobutyl acetate in Escherichia coli.

Previously, Escherichia coli was engineered to produce isobutyl acetate (IBA). Titers greater than the toxicity threshold (3 g/L) were achieved by using layer-assisted production. To avoid this costly and complex method, adaptive laboratory evolution (ALE) was applied to E. coli for improved IBA tolerance. Over 37 rounds of selective pressure, 22 IBA-tolerant mutants were isolated. Remarkably, these mutants not only tolerate high IBA concentrations, they also produce higher IBA titers. Using whole-genome sequencing followed by CRISPR/Cas9 mediated genome editing, the mutations (SNPs in metH, rho and deletion of arcA) that confer improved tolerance and higher titers were elucidated. The improved IBA titers in the evolved mutants were a result of an increased supply of acetyl-CoA and altered transcriptional machinery. Without the use of phase separation, a strain capable of 3.2-fold greater IBA production than the parent strain was constructed by combing select beneficial mutations. These results highlight the impact improved tolerance has on the production capability of a biosynthetic system.

A biochemically-interpretable machine learning classifier for microbial GWAS.

Current machine learning classifiers have successfully been applied to whole-genome sequencing data to identify genetic determinants of antimicrobial resistance (AMR), but they lack causal interpretation. Here we present a metabolic model-based machine learning classifier, named Metabolic Allele Classifier (MAC), that uses flux balance analysis to estimate the biochemical effects of alleles. We apply the MAC to a dataset of 1595 drug-tested Mycobacterium tuberculosis strains and show that MACs predict AMR phenotypes with accuracy on par with mechanism-agnostic machine learning models (isoniazid AUC = 0.93) while enabling a biochemical interpretation of the genotype-phenotype map. Interpretation of MACs for three antibiotics (pyrazinamide, para-aminosalicylic acid, and isoniazid) recapitulates known AMR mechanisms and suggest a biochemical basis for how the identified alleles cause AMR. Extending flux balance analysis to identify accurate sequence classifiers thus contributes mechanistic insights to GWAS, a field thus far dominated by mechanism-agnostic results.

Machine learning with random subspace ensembles identifies antimicrobial resistance determinants from pan-genomes of three pathogens.

The evolution of antimicrobial resistance (AMR) poses a persistent threat to global public health. Sequencing efforts have already yielded genome sequences for thousands of resistant microbial isolates and require robust computational tools to systematically elucidate the genetic basis for AMR. Here, we present a generalizable machine learning workflow for identifying genetic features driving AMR based on constructing reference strain-agnostic pan-genomes and training random subspace ensembles (RSEs). This workflow was applied to the resistance profiles of 14 antimicrobials across three urgent threat pathogens encompassing 288 Staphylococcus aureus, 456 Pseudomonas aeruginosa, and 1588 Escherichia coli genomes. We find that feature selection by RSE detects known AMR associations more reliably than common statistical tests and previous ensemble approaches, identifying a total of 45 known AMR-conferring genes and alleles across the three organisms, as well as 25 candidate associations backed by domain-level annotations. Furthermore, we find that results from the RSE approach are consistent with existing understanding of fluoroquinolone (FQ) resistance due to mutations in the main drug targets, gyrA and parC, in all three organisms, and suggest the mutational landscape of those genes with respect to FQ resistance is simple. As larger datasets become available, we expect this approach to more reliably predict AMR determinants for a wider range of microbial pathogens.

Machine learning and structural analysis of Mycobacterium tuberculosis pan-genome identifies genetic signatures of antibiotic resistance.

Mycobacterium tuberculosis is a serious human pathogen threat exhibiting complex evolution of antimicrobial resistance (AMR). Accordingly, the many publicly available datasets describing its AMR characteristics demand disparate data-type analyses. Here, we develop a reference strain-agnostic computational platform that uses machine learning approaches, complemented by both genetic interaction analysis and 3D structural mutation-mapping, to identify signatures of AMR evolution to 13 antibiotics. This platform is applied to 1595 sequenced strains to yield four key results. First, a pan-genome analysis shows that M. tuberculosis is highly conserved with sequenced variation concentrated in PE/PPE/PGRS genes. Second, the platform corroborates 33 genes known to confer resistance and identifies 24 new genetic signatures of AMR. Third, 97 epistatic interactions across 10 resistance classes are revealed. Fourth, detailed structural analysis of these genes yields mechanistic bases for their selection. The platform can be used to study other human pathogens.

Updated and standardized genome-scale reconstruction of Mycobacterium tuberculosis H37Rv, iEK1011, simulates flux states indicative of physiological conditions.

BACKGROUND: The efficacy of antibiotics against M. tuberculosis has been shown to be influenced by experimental media conditions. Investigations of M. tuberculosis growth in physiological conditions have described an environment that is different from common in vitro media. Thus, elucidating the interplay between available nutrient sources and antibiotic efficacy has clear medical relevance. While genome-scale reconstructions of M. tuberculosis have enabled the ability to interrogate media differences for the past 10 years, recent reconstructions have diverged from each other without standardization. A unified reconstruction of M. tuberculosis H37Rv would elucidate the impact of different nutrient conditions on antibiotic efficacy and provide new insights for therapeutic intervention. RESULTS: We present a new genome-scale model of M. tuberculosis H37Rv, named iEK1011, that unifies and updates previous M. tuberculosis H37Rv genome-scale reconstructions. We functionally assess iEK1011 against previous models and show that the model increases correct gene essentiality predictions on two different experimental datasets by 6% (53% to 60%) and 18% (60% to 71%), respectively. We compared simulations between in vitro and approximated in vivo media conditions to examine the predictive capabilities of iEK1011. The simulated differences recapitulated literature defined characteristics in the rewiring of TCA metabolism including succinate secretion, gluconeogenesis, and activation of both the glyoxylate shunt and the methylcitrate cycle. To assist efforts to elucidate mechanisms of antibiotic resistance development, we curated 16 metabolic genes related to antimicrobial resistance and approximated evolutionary drivers of resistance. Comparing simulations of these antibiotic resistance features between in vivo and in vitro media highlighted condition-dependent differences that may influence the efficacy of antibiotics. CONCLUSIONS: iEK1011 provides a computational knowledge base for exploring the impact of different environmental conditions on the metabolic state of M. tuberculosis H37Rv. As more experimental data and knowledge of M. tuberculosis H37Rv become available, a unified and standardized M. tuberculosis model will prove to be a valuable resource to the research community studying the systems biology of M. tuberculosis.

Evaluation of rate law approximations in bottom-up kinetic models of metabolism.

BACKGROUND: The mechanistic description of enzyme kinetics in a dynamic model of metabolism requires specifying the numerical values of a large number of kinetic parameters. The parameterization challenge is often addressed through the use of simplifying approximations to form reaction rate laws with reduced numbers of parameters. Whether such simplified models can reproduce dynamic characteristics of the full system is an important question. RESULTS: In this work, we compared the local transient response properties of dynamic models constructed using rate laws with varying levels of approximation. These approximate rate laws were: 1) a Michaelis-Menten rate law with measured enzyme parameters, 2) a Michaelis-Menten rate law with approximated parameters, using the convenience kinetics convention, 3) a thermodynamic rate law resulting from a metabolite saturation assumption, and 4) a pure chemical reaction mass action rate law that removes the role of the enzyme from the reaction kinetics. We utilized in vivo data for the human red blood cell to compare the effect of rate law choices against the backdrop of physiological flux and concentration differences. We found that the Michaelis-Menten rate law with measured enzyme parameters yields an excellent approximation of the full system dynamics, while other assumptions cause greater discrepancies in system dynamic behavior. However, iteratively replacing mechanistic rate laws with approximations resulted in a model that retains a high correlation with the true model behavior. Investigating this consistency, we determined that the order of magnitude differences among fluxes and concentrations in the network were greatly influential on the network dynamics. We further identified reaction features such as thermodynamic reversibility, high substrate concentration, and lack of allosteric regulation, which make certain reactions more suitable for rate law approximations. CONCLUSIONS: Overall, our work generally supports the use of approximate rate laws when building large scale kinetic models, due to the key role that physiologically meaningful flux and concentration ranges play in determining network dynamics. However, we also showed that detailed mechanistic models show a clear benefit in prediction accuracy when data is available. The work here should help to provide guidance to future kinetic modeling efforts on the choice of rate law and parameterization approaches.