Model-driven discovery of underground metabolic functions in Escherichia coli Gabriela I. Guzmána, José Utrillaa, Sergey Nurkb, Elizabeth Brunkc, Jonathan M. Monkd, Ali Ebrahima, Bernhard O. Palssona,e,f, and Adam M. Feista,e,1 Departments of aBioengineering, dNanoEngineering, and fPediatrics, University of California, San Diego, La Jolla, CA 92093; bAlgorithmic Biology Laboratory, St. Petersburg Academic University, Russian Academy of Sciences, St. Petersburg, Russia; cJoint BioEnergy Institute, Lawrence Berkeley National Laboratory, Emeryville, CA 94608; and eNovo Nordisk Foundation Center for Biosustainability, Technical University of Denmark, Lyngby, Denmark Edited by Marc W. Kirschner, Harvard Medical School, Boston, MA, and approved December 5, 2014 (received for review August 5, 2014) Enzyme promiscuity toward substrates has been discussed in of other proteins based on sequence homology or enzyme familial evolutionary terms as providing the flexibility to adapt to novel relationships. In the present study, this task is approached from environments. In the present work, we describe an approach to- a different perspective by taking advantage of in vivo experimental ward exploring such enzyme promiscuity in the space of a meta- techniques to gain insight into activities that are more physiologi- bolic network. This approach leverages genome-scale models, cally relevant. In this way, as has been demonstrated in other in which have been widely used for predicting growth phenotypes vivo studies, many of the challenges associated with removing in various environments or following a genetic perturbation; how- enzymes from their native environment are circumvented (8). ever, these predictions occasionally fail. Failed predictions of gene Specifically, the present study focuses on examination of the reg- essentiality offer an opportunity for targeting biological discovery, ulatory and evolutionary capacity of a cell in vivo. Theories re- suggesting the presence of unknown underground pathways garding genome duplications have suggested that an enzyme with stemming from enzymatic cross-reactivity. We demonstrate a a side activity that is selected for may be enhanced via gene du- workflow that couples constraint-based modeling and bioinfor- plication followed by mutation accumulation (11). Thus, laboratory matic tools with KO strain analysis and adaptive laboratory evo- evolutions may provide insight into these evolutionary mechanisms lution for the purpose of predicting promiscuity at the genome involving enzyme promiscuity. Furthermore, exploration of an scale. Three cases of genes that are incorrectly predicted as essen- underground metabolic network that takes advantage of enzyme SYSTEMS BIOLOGY tial in Escherichia coli—aspC, argD, and gltA—are examined, and cross-reactivity through native regulatory adaptations is best ex- isozyme functions are uncovered for each to a different extent. amined in the context of a whole cell (6, 8). Seven isozyme functions based on genetic and transcriptional ev- A top-down, model-driven approach coupled with in vivo ex- idence are suggested between the genes aspC and tyrB, argD and perimentation to explore enzyme promiscuity could provide new astC, gabT and puuE, and gltA and prpC. This study demonstrates insights into the physiological role of underground metabolism how a targeted model-driven approach to discovery can system- and complement the current approaches to enzyme research. atically fill knowledge gaps, characterize underground metabo- Computational predictions of gene essentiality are a commonly lism, and elucidate regulatory mechanisms of adaptation in used application of genome-scale models and constraint-based response to gene KO perturbations. modeling (12, 13). When these models fail to predict gene es- sentiality, it signifies a missing link in our knowledge of metab- underground metabolism | substrate promiscuity | systems biology | olism and provides targets for further exploration (14). Various isozyme discovery | genome-scale modeling computational algorithms, including SMILEY and GrowMatch, have been published with the intent of reconciling such knowledge he notion that enzymes are highly specialized to carry out gaps (15, 16). The following is a proof-of-principle study that Ta single function is often untrue. It has been demonstrated that many enzymes exhibit flexibility, or promiscuity, in regard to Significance what substrates their catalytic pockets recognize. This lack of substrate specificity can lead to accuracy-rate tradeoffs that may Organisms have evolved to take advantage of their environment. affect evolutionary trajectories (1). How has enzyme promiscuity Enzymes drive this adaptability by displaying flexibility in terms of shaped the evolution and divergence of organisms? The “patch- substrate specificity and catalytic promiscuity. This enzyme pro- work” model theorizes that primitive enzymes possessed a high miscuity has been observed in a limited number of laboratory degree of substrate promiscuity because it conferred a greater de- experiments; however, a larger underground network of reactions gree of catalytic versatility when the pool of available enzymes was may occur within a cell below the level of detection. It is not until limited (2–5). The existence of promiscuous proteins further serves acell’s metabolic capabilities are probed that these novel functions as a starting point for evolving new functions, allowing for novel come to light. In this study, a workflow is presented for probing adaptations. Thus, organisms may exhibit latent, underground promiscuous activity at the genome scale. This workflow combines metabolic pathways that form the basis of their capacity to adapt to genome-scale reconstructions of metabolic networks with gene KOs changing environments (6–8). Substrate promiscuity, also referred and adaptive laboratory evolution. Such tools become increasingly to as “moonlighting activity” and “cross-reactivity,” has thus been important when designing drugs targeting pathogenic bacteria or studied in terms of evolution, and ties have been made between engineering enzymes and bacteria for biotechnology applications. enzymes and their superfamilies (9). How novel enzyme functions Author contributions: G.I.G., J.U., B.O.P., and A.M.F. designed research; G.I.G., J.U., S.N., arise within superfamilies is thus examined, and provides a basis for E.B., and J.M.M. performed research; G.I.G., S.N., E.B., J.M.M., and A.E. contributed new predicting promiscuous behavior among these protein families. reagents/analytic tools; G.I.G., J.U., S.N., E.B., and A.M.F. analyzed data; and G.I.G. and However, defining targets for studies of promiscuity outside of A.M.F. wrote the paper. these families and on a larger scale can become quite challenging. The authors declare no conflict of interest. Enzyme promiscuity has become widely accepted and exam- This article is a PNAS Direct Submission. ined on the enzyme level from a biochemical standpoint (10). 1To whom correspondence should be addressed. Email: [email protected]. These detailed biochemical studies provide an in vitro view of This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. enzyme promiscuity and may be extended to reflect the promiscuity 1073/pnas.1414218112/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1414218112 PNAS Early Edition | 1of6 Downloaded by guest on September 25, 2021 demonstrates the advantages of a workflow for examining pro- Isozyme Discovery Workflow miscuity at the genome scale that also encompasses an adaptive laboratory evolution (ALE) framework. Three cases are ex- in vivo Experiment Identify False plored to illustrate the capabilities of such a targeted, top-down Negative (FN) approach to uncover the underground, latent activities of in silico Experiment Predictions & Isozyme enzymes that reconcile gaps in our knowledge of metabolism. Candidate(s) Results and Discussion Developing a Model-Driven Workflow for Isozyme Discovery. The Supplement & Does KO Identify a No results from this study demonstrated that a top-down systems Wean Strain Substrate for Strain Grow in approach could be used to drive the discovery of enzyme sub- to Growth Growth Rescue Minimal Condition Media? strate promiscuity by using three genes, aspC, argD, and gltA, that using ALE were incorrectly identified to be essential as inputs. The isozyme Yes discovery workflow presented in this study is a prime example of RT-qPCR targeted analysis based on systems-level insights (Fig. 1). of Candidate The first step in the isozyme discovery workflow was to identify Isozyme Gene(s) the targets for exploration. These targets come from performing flux balance analysis (FBA) gene essentiality simulations in Escherichia coli using the iJO1366 metabolic reconstruction Identify (17, 18). When discussing computational gene essentiality pre- Up-Regulated dictions, the term “false-negative prediction” refers to a situation in Candidate(s) which a gene is predicted to be essential but is experimentally ob- served to be nonessential. This type of prediction failure can stem from lack of knowledge of an alternate pathway or isozyme (14). All Construct Multi genes associated with false-negative predictions in iJO1366 were KO Strain identified, and those genes with high-confidence candidate iso- zymes, based on sequence homology, were used as examples for this study. To identify potential isozymes based on sequence homology, Does KO the National Center for Biotechnology Information’sBLASTpal-