Empowering drug off-target discovery with metabolic and structural analysis.

Elucidating intracellular drug targets is a difficult problem. While machine learning analysis of omics data has been a promising approach, going from large-scale trends to specific targets remains a challenge. Here, we develop a hierarchic workflow to focus on specific targets based on analysis of metabolomics data and growth rescue experiments. We deploy this framework to understand the intracellular molecular interactions of the multi-valent dihydrofolate reductase-targeting antibiotic compound CD15-3. We analyse global metabolomics data utilizing machine learning, metabolic modelling, and protein structural similarity to prioritize candidate drug targets. Overexpression and in vitro activity assays confirm one of the predicted candidates, HPPK (folK), as a CD15-3 off-target. This study demonstrates how established machine learning methods can be combined with mechanistic analyses to improve the resolution of drug target finding workflows for discovering off-targets of a metabolic inhibitor.

Blood donor exposome and impact of common drugs on red blood cell metabolism.

Computational models based on recent maps of the RBC proteome suggest that mature erythrocytes may harbor targets for common drugs. This prediction is relevant to RBC storage in the blood bank, in which the impact of small molecule drugs or other xenometabolites deriving from dietary, iatrogenic, or environmental exposures ("exposome") may alter erythrocyte energy and redox metabolism and, in so doing, affect red cell storage quality and posttransfusion efficacy. To test this prediction, here we provide a comprehensive characterization of the blood donor exposome, including the detection of common prescription and over-the-counter drugs in blood units donated by 250 healthy volunteers in the Recipient Epidemiology and Donor Evaluation Study III Red Blood Cell-Omics (REDS-III RBC-Omics) Study. Based on high-throughput drug screenings of 1366 FDA-approved drugs, we report that approximately 65% of the tested drugs had an impact on erythrocyte metabolism. Machine learning models built using metabolites as predictors were able to accurately predict drugs for several drug classes/targets (bisphosphonates, anticholinergics, calcium channel blockers, adrenergics, proton pump inhibitors, antimetabolites, selective serotonin reuptake inhibitors, and mTOR), suggesting that these drugs have a direct, conserved, and substantial impact on erythrocyte metabolism. As a proof of principle, here we show that the antacid ranitidine - though rarely detected in the blood donor population - has a strong effect on RBC markers of storage quality in vitro. We thus show that supplementation of blood units stored in bags with ranitidine could - through mechanisms involving sphingosine 1-phosphate-dependent modulation of erythrocyte glycolysis and/or direct binding to hemoglobin - improve erythrocyte metabolism and storage quality.

Microbial Synthesis of Human-Hormone Melatonin at Gram Scales.

Melatonin is a commercially attractive tryptophan-derived hormone. Here we describe a bioprocess for the production of melatonin using Escherichia coli to high titers. The first engineered strain produced 0.13 g/L of melatonin from tryptophan under fed-batch fermentation conditions. A 4-fold improvement on melatonin titer was further achieved by (1) protein engineering of rate-limiting tryptophan hydroxylase to improve 5-hydroxytryptophan biosynthesis and (2) chromosomal integration of aromatic-amino-acid decarboxylase to limit byproduct formation and to minimize gene toxicity to the host cell. Fermentation optimization improved melatonin titer by an additional 2-fold. Deletion of yddG, a tryptophan exporter, exhibited an additive beneficial effect. The final engineered strain produced ∼2.0 g/L of melatonin with tryptophan supplemented externally and ∼1.0 g/L with glucose as the sole carbon source for tryptophan supply. This study lays the foundation for further developing a commercial melatonin-producing E. coli strain.

A White-Box Machine Learning Approach for Revealing Antibiotic Mechanisms of Action.

Current machine learning techniques enable robust association of biological signals with measured phenotypes, but these approaches are incapable of identifying causal relationships. Here, we develop an integrated "white-box" biochemical screening, network modeling, and machine learning approach for revealing causal mechanisms and apply this approach to understanding antibiotic efficacy. We counter-screen diverse metabolites against bactericidal antibiotics in Escherichia coli and simulate their corresponding metabolic states using a genome-scale metabolic network model. Regression of the measured screening data on model simulations reveals that purine biosynthesis participates in antibiotic lethality, which we validate experimentally. We show that antibiotic-induced adenine limitation increases ATP demand, which elevates central carbon metabolism activity and oxygen consumption, enhancing the killing effects of antibiotics. This work demonstrates how prospective network modeling can couple with machine learning to identify complex causal mechanisms underlying drug efficacy.

Description and interpretation of adaptive evolution of Escherichia coli K-12 MG1655 by using a genome-scale in silico metabolic model.

Genome-scale in silico metabolic networks of Escherichia coli have been reconstructed. By using a constraint-based in silico model of a reconstructed network, the range of phenotypes exhibited by E. coli under different growth conditions can be computed, and optimal growth phenotypes can be predicted. We hypothesized that the end point of adaptive evolution of E. coli could be accurately described a priori by our in silico model since adaptive evolution should lead to an optimal phenotype. Adaptive evolution of E. coli during prolonged exponential growth was performed with M9 minimal medium supplemented with 2 g of alpha-ketoglutarate per liter, 2 g of lactate per liter, or 2 g of pyruvate per liter at both 30 and 37 degrees C, which produced seven distinct strains. The growth rates, substrate uptake rates, oxygen uptake rates, by-product secretion patterns, and growth rates on alternative substrates were measured for each strain as a function of evolutionary time. Three major conclusions were drawn from the experimental results. First, adaptive evolution leads to a phenotype characterized by maximized growth rates that may not correspond to the highest biomass yield. Second, metabolic phenotypes resulting from adaptive evolution can be described and predicted computationally. Third, adaptive evolution on a single substrate leads to changes in growth characteristics on other substrates that could signify parallel or opposing growth objectives. Together, the results show that genome-scale in silico metabolic models can describe the end point of adaptive evolution a priori and can be used to gain insight into the adaptive evolutionary process for E. coli.