Model-driven discovery of underground metabolic functions in Escherichia coli

Abstract

Enzyme promiscuity toward substrates has been discussed in evolutionary terms as providing the flexibility to adapt to novel environments. In the present work, we describe an approach toward exploring such enzyme promiscuity in the space of a metabolic network. This approach leverages genome-scale models, which have been widely used for predicting growth phenotypes in various environments or following a genetic perturbation; however, these predictions occasionally fail. Failed predictions of gene essentiality offer an opportunity for targeting biological discovery, suggesting the presence of unknown underground pathways stemming from enzymatic cross-reactivity. We demonstrate a workflow that couples constraint-based modeling and bioinformatic tools with KO strain analysis and adaptive laboratory evolution for the purpose of predicting promiscuity at the genome scale. Three cases of genes that are incorrectly predicted as essential in Escherichia coli--aspC, argD, and gltA--are examined, and isozyme functions are uncovered for each to a different extent. Seven isozyme functions based on genetic and transcriptional evidence are suggested between the genes aspC and tyrB, argD and astC, gabT and puuE, and gltA and prpC. This study demonstrates how a targeted model-driven approach to discovery can systematically fill knowledge gaps, characterize underground metabolism, and elucidate regulatory mechanisms of adaptation in response to gene KO perturbations.

Keywords: genome-scale modeling; isozyme discovery; substrate promiscuity; systems biology; underground metabolism.

Fig. 1.

Schematic of the general workflow used for isozyme discovery involving both in vivo and in silico experiments. Starting from the topmost box, false-negative model predictions and isozyme candidates were identified using FBA and BLASTp. The workflow was then followed vertically downward, examining KO strain growth, expression levels of candidate isozyme genes, and multi-KO strain phenotypes. Deviations from the schematic occurred when growth discrepancies were encountered. The workflow was terminated once a synthetic lethal interaction of a false-negative gene and isozyme candidate(s) was identified. The output was enzymatic activities characterized and added to the current genome-scale model reconstruction of E. coli.

Fig. 2.

Workflow-guided results used to discover isozymes of aspC. (A) Bar chart of the qPCR results in terms of normalized expression of the tyrB isozyme candidate in the ΔaspC and WT strains (SE ratio was calculated: P < 0.05; n = 1, two biological duplicates and six technical replicates). A fold increase of 4.8 is observed in the ΔaspC strain compared with WT. (B) Growth data on glucose minimal medium in terms of cellular density is reported for ΔaspC and ΔaspCΔtyrB strains. The ΔaspCΔtyrB strain shows no growth for >150 h. DW, dry weight.

Fig. 3.

Workflow-guided results used to discover isozymes for argD. (A) Bar chart of the relative expression, compared with WT, of candidate isozyme genes in the SKO, DKO, and TKO strains shows up-regulation that guided the multi-KO strain construction (SE ratio was calculated: P < 0.05; n = 1, two biological duplicates and six technical replicates). Note that puuE was not up-regulated until the construction of the ΔargDΔastCΔgabT TKO strain. (B) Growth data on glucose minimal medium in terms of cellular density are reported for the four strains iteratively constructed as guided by the workflow. The last strain constructed, ΔargDΔastCΔgabTΔpuuE, continued to show no growth after 700 h of incubation.

Fig. 4.

Workflow-guided results used to discover isozymes of gltA. (A) Cellular density results from the ALE of ΔgltA on glucose minimal medium are illustrated. A vertical drop in cellular density corresponds to manual passaging of a fraction of the cell culture for a fresh batch of medium. The independent ALE experiments reached a different apparent final density. (B) Bar chart shows qPCR results as a fold increase in expression of the prpC isozyme candidate in four ALE end-point clones in relation to a WT strain (SE ratio was calculated: P < 0.05; n = 1, two biological duplicates and six technical replicates).

Fig. 5.

Summary of GPR associations to be added to the E. coli metabolic network reconstruction iJO1366 based on findings from three cases: gltA, aspC, and argD. Proposed associations are highlighted in red. (A) GPR additions for tyrB and ilvE. (B) GPR additions for the quadruple synthetic lethal interaction set. (C) GPR for gltA and prpC highlights the requirement of evolution or mutagenesis for the suggested association. Reaction abbreviations are from the iJO1366 reconstruction.