Search

Abstract Growth is a fundamental process of life. Growth requirements are well‐characterized experimentally for many microbes; however, we lack a unified model for cellular growth. Such a model must be predictive of events at the molecular scale and capable of explaining the high‐level behavior of the cell as a whole. Here, we construct an ME‐Model for Escherichia coli—a genome‐scale model that seamlessly integrates metabolic and gene product expression pathways. The model computes ∼80% of the functional proteome (by mass), which is used by the cell to support growth under a given condition. Metabolism and gene expression are interdependent processes that affect and constrain each other. We formalize these constraints and apply the principle of growth optimization to enable the accurate prediction of multi‐scale phenotypes, ranging from coarse‐grained (growth rate, nutrient uptake, by‐product secretion) to fine‐grained (metabolic fluxes, gene expression levels). Our results unify many existing principles developed to describe bacterial growth.

Abstract The genome‐scale model (GEM) of metabolism in the bacterium Escherichia coli K‐12 has been in development for over a decade and is now in wide use. GEM‐enabled studies of E. coli have been primarily focused on six applications: (1) metabolic engineering, (2) model‐driven discovery, (3) prediction of cellular phenotypes, (4) analysis of biological network properties, (5) studies of evolutionary processes, and (6) models of interspecies interactions. In this review, we provide an overview of these applications along with a critical assessment of their successes and limitations, and a perspective on likely future developments in the field. Taken together, the studies performed over the past decade have established a genome‐scale mechanistic understanding of genotype–phenotype relationships in E. coli metabolism that forms the basis for similar efforts for other microbial species. Future challenges include the expansion of GEMs by integrating additional cellular processes beyond metabolism, the identification of key constraints based on emerging data types, and the development of computational methods able to handle such large‐scale network models with sufficient accuracy.

Abstract The initial genome‐scale reconstruction of the metabolic network of Escherichia coli K‐12 MG1655 was assembled in 2000. It has been updated and periodically released since then based on new and curated genomic and biochemical knowledge. An update has now been built, named iJO1366, which accounts for 1366 genes, 2251 metabolic reactions, and 1136 unique metabolites. iJO1366 was (1) updated in part using a new experimental screen of 1075 gene knockout strains, illuminating cases where alternative pathways and isozymes are yet to be discovered, (2) compared with its predecessor and to experimental data sets to confirm that it continues to make accurate phenotypic predictions of growth on different substrates and for gene knockout strains, and (3) mapped to the genomes of all available sequenced E. coli strains, including pathogens, leading to the identification of hundreds of unannotated genes in these organisms. Like its predecessors, the iJO1366 reconstruction is expected to be widely deployed for studying the systems biology of E. coli and for metabolic engineering applications.

Abstract Metabolic network reconstruction encompasses existing knowledge about an organism's metabolism and genome annotation, providing a platform for omics data analysis and phenotype prediction. The model alga Chlamydomonas reinhardtii is employed to study diverse biological processes from photosynthesis to phototaxis. Recent heightened interest in this species results from an international movement to develop algal biofuels. Integrating biological and optical data, we reconstructed a genome‐scale metabolic network for this alga and devised a novel light‐modeling approach that enables quantitative growth prediction for a given light source, resolving wavelength and photon flux. We experimentally verified transcripts accounted for in the network and physiologically validated model function through simulation and generation of new experimental growth data, providing high confidence in network contents and predictive applications. The network offers insight into algal metabolism and potential for genetic engineering and efficient light source design, a pioneering resource for studying light‐driven metabolism and quantitative systems biology.

Abstract Laboratory evolution studies provide fundamental biological insight through direct observation of the evolution process. They not only enable testing of evolutionary theory and principles, but also have applications to metabolic engineering and human health. Genome‐scale tools are revolutionizing studies of laboratory evolution by providing complete determination of the genetic basis of adaptation and the changes in the organism's gene expression state. Here, we review studies centered on four central themes of laboratory evolution studies: (1) the genetic basis of adaptation; (2) the importance of mutations to genes that encode regulatory hubs; (3) the view of adaptive evolution as an optimization process; and (4) the dynamics with which laboratory populations evolve.

Abstract Metabolic coupling of Mycobacterium tuberculosis to its host is foundational to its pathogenesis. Computational genome‐scale metabolic models have shown utility in integrating ‐omic as well as physiologic data for systemic, mechanistic analysis of metabolism. To date, integrative analysis of host–pathogen interactions using in silico mass‐balanced, genome‐scale models has not been performed. We, therefore, constructed a cell‐specific alveolar macrophage model, iAB‐AMØ‐1410, from the global human metabolic reconstruction, Recon 1. The model successfully predicted experimentally verified ATP and nitric oxide production rates in macrophages. This model was then integrated with an M. tuberculosis H37Rv model, iNJ661, to build an integrated host–pathogen genome‐scale reconstruction, iAB‐AMØ‐1410‐Mt‐661. The integrated host–pathogen network enables simulation of the metabolic changes during infection. The resulting reaction activity and gene essentiality targets of the integrated model represent an altered infectious state. High‐throughput data from infected macrophages were mapped onto the host–pathogen network and were able to describe three distinct pathological states. Integrated host–pathogen reconstructions thus form a foundation upon which understanding the biology and pathophysiology of infections can be developed.

Abstract After hundreds of generations of adaptive evolution at exponential growth, Escherichia coli grows as predicted using flux balance analysis (FBA) on genome‐scale metabolic models (GEMs). However, it is not known whether the predicted pathway usage in FBA solutions is consistent with gene and protein expression in the wild‐type and evolved strains. Here, we report that >98% of active reactions from FBA optimal growth solutions are supported by transcriptomic and proteomic data. Moreover, when E. coli adapts to growth rate selective pressure, the evolved strains upregulate genes within the optimal growth predictions, and downregulate genes outside of the optimal growth solutions. In addition, bottlenecks from dosage limitations of computationally predicted essential genes are overcome in the evolved strains. We also identify regulatory processes that may contribute to the development of the optimal growth phenotype in the evolved strains, such as the downregulation of known regulons and stringent response suppression. Thus, differential gene and protein expression from wild‐type and adaptively evolved strains supports observed growth phenotype changes, and is consistent with GEM‐computed optimal growth states.

Abstract The availability and utility of genome‐scale metabolic reconstructions have exploded since the first genome‐scale reconstruction was published a decade ago. Reconstructions have now been built for a wide variety of organisms, and have been used toward five major ends: (1) contextualization of high‐throughput data, (2) guidance of metabolic engineering, (3) directing hypothesis‐driven discovery, (4) interrogation of multi‐species relationships, and (5) network property discovery. In this review, we examine the many uses and future directions of genome‐scale metabolic reconstructions, and we highlight trends and opportunities in the field that will make the greatest impact on many fields of biology.

Abstract The biological community is now awash in high‐throughput data sets and is grappling with the challenge of integrating disparate data sets. Such integration has taken the form of statistical analysis of large data sets, or through the bottom–up reconstruction of reaction networks. While progress has been made with statistical and structural methods, large‐scale systems have remained refractory to dynamic model building by traditional approaches. The availability of annotated genomes enabled the reconstruction of genome‐scale networks, and now the availability of high‐throughput metabolomic and fluxomic data along with thermodynamic information opens the possibility to build genome‐scale kinetic models. We describe here a framework for building and analyzing such models. The mathematical analysis challenges are reflected in four foundational properties, (i) the decomposition of the Jacobian matrix into chemical, kinetic and thermodynamic information, (ii) the structural similarity between the stoichiometric matrix and the transpose of the gradient matrix, (iii) the duality transformations enabling either fluxes or concentrations to serve as the independent variables and (iv) the timescale hierarchy in biological networks. Recognition and appreciation of these properties highlight notable and challenging new in silico analysis issues.

Rhizobiaceas are bacteria that fix nitrogen during symbiosis with plants. This symbiotic relationship is crucial for the nitrogen cycle, and understanding symbiotic mechanisms is a scientific challenge with direct applications in agronomy and plant development. Rhizobium etli is a bacteria which provides legumes with ammonia (among other chemical compounds), thereby stimulating plant growth. A genome-scale approach, integrating the biochemical information available for R. etli, constitutes an important step toward understanding the symbiotic relationship and its possible improvement. In this work we present a genome-scale metabolic reconstruction (iOR363) for R. etli CFN42, which includes 387 metabolic and transport reactions across 26 metabolic pathways. This model was used to analyze the physiological capabilities of R. etli during stages of nitrogen fixation. To study the physiological capacities in silico, an objective function was formulated to simulate symbiotic nitrogen fixation. Flux balance analysis (FBA) was performed, and the predicted active metabolic pathways agreed qualitatively with experimental observations. In addition, predictions for the effects of gene deletions during nitrogen fixation in Rhizobia in silico also agreed with reported experimental data. Overall, we present some evidence supporting that FBA of the reconstructed metabolic network for R. etli provides results that are in agreement with physiological observations. Thus, as for other organisms, the reconstructed genome-scale metabolic network provides an important framework which allows us to compare model predictions with experimental measurements and eventually generate hypotheses on ways to improve nitrogen fixation.

Abstract An updated genome‐scale reconstruction of the metabolic network in Escherichia coli K‐12 MG1655 is presented. This updated metabolic reconstruction includes: (1) an alignment with the latest genome annotation and the metabolic content of EcoCyc leading to the inclusion of the activities of 1260 ORFs, (2) characterization and quantification of the biomass components and maintenance requirements associated with growth of E. coli and (3) thermodynamic information for the included chemical reactions. The conversion of this metabolic network reconstruction into an in silico model is detailed. A new step in the metabolic reconstruction process, termed thermodynamic consistency analysis, is introduced, in which reactions were checked for consistency with thermodynamic reversibility estimates. Applications demonstrating the capabilities of the genome‐scale metabolic model to predict high‐throughput experimental growth and gene deletion phenotypic screens are presented. The increased scope and computational capability using this new reconstruction is expected to broaden the spectrum of both basic biology and applied systems biology studies of E. coli metabolism.

Genome-wide transcription start site (TSS) profiles of the enterobacteria Escherichia coli and Klebsiella pneumoniae were experimentally determined through modified 5' RACE followed by deep sequencing of intact primary mRNA. This identified 3,746 and 3,143 TSSs for E. coli and K. pneumoniae, respectively. Experimentally determined TSSs were then used to define promoter regions and 5' UTRs upstream of coding genes. Comparative analysis of these regulatory elements revealed the use of multiple TSSs, identical sequence motifs of promoter and Shine-Dalgarno sequence, reflecting conserved gene expression apparatuses between the two species. In both species, over 70% of primary transcripts were expressed from operons having orthologous genes during exponential growth. However, expressed orthologous genes in E. coli and K. pneumoniae showed a strikingly different organization of upstream regulatory regions with only 20% identical promoters with TSSs in both species. Over 40% of promoters had TSSs identified in only one species, despite conserved promoter sequences existing in the other species. 662 conserved promoters having TSSs in both species resulted in the same number of comparable 5' UTR pairs, and that regulatory element was found to be the most variant region in sequence among promoter, 5' UTR, and ORF. In K. pneumoniae, 48 sRNAs were predicted and 36 of them were expressed during exponential growth. Among them, 34 orthologous sRNAs between two species were analyzed in depth, and the analysis showed that many sRNAs of K. pneumoniae, including pleiotropic sRNAs such as rprA, arcZ, and sgrS, may work in the same way as in E. coli. These results reveal a new dimension of comparative genomics such that a comparison of two genomes needs to be comprehensive over all levels of genome organization.

Adaptive laboratory evolution (ALE) under controlled conditions has become a valuable approach for the study of the genetic and biochemical basis for microbial adaptation under a given selection pressure. Conventionally, the timescale in ALE experiments has been set in terms of number of generations. As mutations are believed to occur primarily during cell division in growing cultures, the cumulative number of cell divisions (CCD) would be an alternative way to set the timescale for ALE. Here we show that in short-term ALE (up to 40-50 days), Escherichia coli, under growth rate selection pressure, was found to undergo approximately 10(11.2) total cumulative cell divisions in the population to produce a new stable growth phenotype that results from 2 to 8 mutations. Continuous exposure to a low level of the mutagen N-methyl-N'-nitro-N-nitrosoguanidine was found to accelerate this timescale and led to a superior growth rate phenotype with a much larger number of mutations as determined with whole-genome sequencing. These results would be useful for the fundamental kinetics of the ALE process in designing ALE experiments and provide a basis for its quantitative description.

Bacterial survival requires adaptation to different environmental perturbations such as exposure to antibiotics, changes in temperature or oxygen levels, DNA damage, and alternative nutrient sources. During adaptation, bacteria often develop beneficial mutations that confer increased fitness in the new environment. Adaptation to the loss of a major non-essential gene product that cripples growth, however, has not been studied at the whole-genome level. We investigated the ability of Escherichia coli K-12 MG1655 to overcome the loss of phosphoglucose isomerase (pgi) by adaptively evolving ten replicates of E. coli lacking pgi for 50 days in glucose M9 minimal medium and by characterizing endpoint clones through whole-genome re-sequencing and phenotype profiling. We found that 1) the growth rates for all ten endpoint clones increased approximately 3-fold over the 50-day period; 2) two to five mutations arose during adaptation, most frequently in the NADH/NADPH transhydrogenases udhA and pntAB and in the stress-associated sigma factor rpoS; and 3) despite similar growth rates, at least three distinct endpoint phenotypes developed as defined by different rates of acetate and formate secretion. These results demonstrate that E. coli can adapt to the loss of a major metabolic gene product with only a handful of mutations and that adaptation can result in multiple, alternative phenotypes.

Recent advances in structural bioinformatics have enabled the prediction of protein-drug off-targets based on their ligand binding sites. Concurrent developments in systems biology allow for prediction of the functional effects of system perturbations using large-scale network models. Integration of these two capabilities provides a framework for evaluating metabolic drug response phenotypes in silico. This combined approach was applied to investigate the hypertensive side effect of the cholesteryl ester transfer protein inhibitor torcetrapib in the context of human renal function. A metabolic kidney model was generated in which to simulate drug treatment. Causal drug off-targets were predicted that have previously been observed to impact renal function in gene-deficient patients and may play a role in the adverse side effects observed in clinical trials. Genetic risk factors for drug treatment were also predicted that correspond to both characterized and unknown renal metabolic disorders as well as cryptic genetic deficiencies that are not expected to exhibit a renal disorder phenotype except under drug treatment. This study represents a novel integration of structural and systems biology and a first step towards computational systems medicine. The methodology introduced herein has important implications for drug development and personalized medicine.

State-of-the-art DNA sequencing technologies are transforming the life sciences due to their ability to generate nucleotide sequence information with a speed and quantity that is unapproachable with traditional Sanger sequencing. Genome sequencing is a principal application of this technology, where the ultimate goal is the full and complete sequence of the organism of interest. Due to the nature of the raw data produced by these technologies, a full genomic sequence attained without the aid of Sanger sequencing has yet to be demonstrated.We have successfully developed a four-phase strategy for using only next-generation sequencing technologies (Illumina and 454) to assemble a complete microbial genome de novo. We applied this approach to completely assemble the 3.7 Mb genome of a rare Geobacter variant (KN400) that is capable of unprecedented current production at an electrode. Two key components of our strategy enabled us to achieve this result. First, we integrated the two data types early in the process to maximally leverage their complementary characteristics. And second, we used the output of different short read assembly programs in such a way so as to leverage the complementary nature of their different underlying algorithms or of their different implementations of the same underlying algorithm.The significance of our result is that it demonstrates a general approach for maximizing the efficiency and success of genome assembly projects as new sequencing technologies and new assembly algorithms are introduced. The general approach is a meta strategy, wherein sequencing data are integrated as early as possible and in particular ways and wherein multiple assembly algorithms are judiciously applied such that the deficiencies in one are complemented by another.

A transcriptional regulatory network (TRN) constitutes the collection of regulatory rules that link environmental cues to the transcription state of a cell's genome. We recently proposed a matrix formalism that quantitatively represents a system of such rules (a transcriptional regulatory system [TRS]) and allows systemic characterization of TRS properties. The matrix formalism not only allows the computation of the transcription state of the genome but also the fundamental characterization of the input-output mapping that it represents. Furthermore, a key advantage of this "pseudo-stoichiometric" matrix formalism is its ability to easily integrate with existing stoichiometric matrix representations of signaling and metabolic networks. Here we demonstrate for the first time how this matrix formalism is extendable to large-scale systems by applying it to the genome-scale Escherichia coli TRS. We analyze the fundamental subspaces of the regulatory network matrix (R) to describe intrinsic properties of the TRS. We further use Monte Carlo sampling to evaluate the E. coli transcription state across a subset of all possible environments, comparing our results to published gene expression data as validation. Finally, we present novel in silico findings for the E. coli TRS, including (1) a gene expression correlation matrix delineating functional motifs; (2) sets of gene ontologies for which regulatory rules governing gene transcription are poorly understood and which may direct further experimental characterization; and (3) the appearance of a distributed TRN structure, which is in stark contrast to the more hierarchical organization of metabolic networks.

Metabolic network reconstructions represent valuable scaffolds for '-omics' data integration and are used to computationally interrogate network properties. However, they do not explicitly account for the synthesis of macromolecules (i.e., proteins and RNA). Here, we present the first genome-scale, fine-grained reconstruction of Escherichia coli's transcriptional and translational machinery, which produces 423 functional gene products in a sequence-specific manner and accounts for all necessary chemical transformations. Legacy data from over 500 publications and three databases were reviewed, and many pathways were considered, including stable RNA maturation and modification, protein complex formation, and iron-sulfur cluster biogenesis. This reconstruction represents the most comprehensive knowledge base for these important cellular functions in E. coli and is unique in its scope. Furthermore, it was converted into a mathematical model and used to: (1) quantitatively integrate gene expression data as reaction constraints and (2) compute functional network states, which were compared to reported experimental data. For example, the model predicted accurately the ribosome production, without any parameterization. Also, in silico rRNA operon deletion suggested that a high RNA polymerase density on the remaining rRNA operons is needed to reproduce the reported experimental ribosome numbers. Moreover, functional protein modules were determined, and many were found to contain gene products from multiple subsystems, highlighting the functional interaction of these proteins. This genome-scale reconstruction of E. coli's transcriptional and translational machinery presents a milestone in systems biology because it will enable quantitative integration of '-omics' datasets and thus the study of the mechanistic principles underlying the genotype-phenotype relationship.

Recent advances in reconstruction and analytical methods for signaling networks have spurred the development of large-scale models that incorporate fully functional and biologically relevant features. An extended reconstruction of the human Toll-like receptor signaling network is presented herein. This reconstruction contains an extensive complement of kinases, phosphatases, and other associated proteins that mediate the signaling cascade along with a delineation of their associated chemical reactions. A computational framework based on the methods of large-scale convex analysis was developed and applied to this network to characterize input-output relationships. The input-output relationships enabled significant modularization of the network into ten pathways. The analysis identified potential candidates for inhibitory mediation of TLR signaling with respect to their specificity and potency. Subsequently, we were able to identify eight novel inhibition targets through constraint-based modeling methods. The results of this study are expected to yield meaningful avenues for further research in the task of mediating the Toll-like receptor signaling network and its effects.

The study of dynamic functions of large-scale biological networks has intensified in recent years. A critical component in developing an understanding of such dynamics involves the study of their hierarchical organization. We investigate the temporal hierarchy in biochemical reaction networks focusing on: (1) the elucidation of the existence of "pools" (i.e., aggregate variables) formed from component concentrations and (2) the determination of their composition and interactions over different time scales. To date the identification of such pools without prior knowledge of their composition has been a challenge. A new approach is developed for the algorithmic identification of pool formation using correlations between elements of the modal matrix that correspond to a pair of concentrations and how such correlations form over the hierarchy of time scales. The analysis elucidates a temporal hierarchy of events that range from chemical equilibration events to the formation of physiologically meaningful pools, culminating in a network-scale (dynamic) structure-(physiological) function relationship. This method is validated on a model of human red blood cell metabolism and further applied to kinetic models of yeast glycolysis and human folate metabolism, enabling the simplification of these models. The understanding of temporal hierarchy and the formation of dynamic aggregates on different time scales is foundational to the study of network dynamics and has relevance in multiple areas ranging from bacterial strain design and metabolic engineering to the understanding of disease processes in humans.

Genome-scale metabolic network models can be reconstructed for well-characterized organisms using genomic annotation and literature information. However, there are many instances in which model predictions of metabolic fluxes are not entirely consistent with experimental data, indicating that the reactions in the model do not match the active reactions in the in vivo system. We introduce a method for determining the active reactions in a genome-scale metabolic network based on a limited number of experimentally measured fluxes. This method, called optimal metabolic network identification (OMNI), allows efficient identification of the set of reactions that results in the best agreement between in silico predicted and experimentally measured flux distributions. We applied the method to intracellular flux data for evolved Escherichia coli mutant strains with lower than predicted growth rates in order to identify reactions that act as flux bottlenecks in these strains. The expression of the genes corresponding to these bottleneck reactions was often found to be downregulated in the evolved strains relative to the wild-type strain. We also demonstrate the ability of the OMNI method to diagnose problems in E. coli strains engineered for metabolite overproduction that have not reached their predicted production potential. The OMNI method applied to flux data for evolved strains can be used to provide insights into mechanisms that limit the ability of microbial strains to evolve towards their predicted optimal growth phenotypes. When applied to industrial production strains, the OMNI method can also be used to suggest metabolic engineering strategies to improve byproduct secretion. In addition to these applications, the method should prove to be useful in general for reconstructing metabolic networks of ill-characterized microbial organisms based on limited amounts of experimental data.

Genome organization can be studied through analysis of chromosome position-dependent patterns in sequence-derived parameters. A comprehensive analysis of such patterns in prokaryotic sequences and genome-scale functional data has yet to be performed. We detected spatial patterns in sequence-derived parameters for 163 chromosomes occurring in 135 bacterial and 16 archaeal organisms using wavelet analysis. Pattern strength was found to correlate with organism-specific features such as genome size, overall GC content, and the occurrence of known motility and chromosomal binding proteins. Given additional functional data for Escherichia coli, we found significant correlations among chromosome position dependent patterns in numerous properties, some of which are consistent with previously experimentally identified chromosome macrodomains. These results demonstrate that the large-scale organization of most sequenced genomes is significantly nonrandom, and, moreover, that this organization is likely linked to genome size, nucleotide composition, and information transfer processes. Constraints on genome evolution and design are thus not solely dependent upon information content, but also upon an intricate multi-parameter, multi-length-scale organization of the chromosome.

Public gene expression databases are a rapidly expanding resource of organism responses to diverse perturbations, presenting both an opportunity and a challenge for bioinformatics workflows to extract actionable knowledge of transcription regulatory network function. Here, we introduce a five-step computational pipeline, called iModulonMiner, to compile, process, curate, analyze, and characterize the totality of RNA-seq data for a given organism or cell type. This workflow is centered around the data-driven computation of co-regulated gene sets using Independent Component Analysis, called iModulons, which have been shown to have broad applications. As a demonstration, we applied this workflow to generate the iModulon structure of Bacillus subtilis using all high-quality, publicly-available RNA-seq data. Using this structure, we predicted regulatory interactions for multiple transcription factors, identified groups of co-expressed genes that are putatively regulated by undiscovered transcription factors, and predicted properties of a recently discovered single-subunit phage RNA polymerase. We also present a Python package, PyModulon, with functions to characterize, visualize, and explore computed iModulons. The pipeline, available at https://github.com/SBRG/iModulonMiner, can be readily applied to diverse organisms to gain a rapid understanding of their transcriptional regulatory network structure and condition-specific activity.

Generalist microbes have adapted to a multitude of environmental stresses through their integrated stress response system. Individual stress responses have been quantified by E. coli metabolism and expression (ME) models under thermal, oxidative and acid stress, respectively. However, the systematic quantification of cross-stress & cross-talk among these stress responses remains lacking. Here, we present StressME: the unified stress response model of E. coli combining thermal (FoldME), oxidative (OxidizeME) and acid (AcidifyME) stress responses. StressME is the most up to date ME model for E. coli and it reproduces all published single-stress ME models. Additionally, it includes refined rate constants to improve prediction accuracy for wild-type and stress-evolved strains. StressME revealed certain optimal proteome allocation strategies associated with cross-stress and cross-talk responses. These stress-optimal proteomes were shaped by trade-offs between protective vs. metabolic enzymes; cytoplasmic vs. periplasmic chaperones; and expression of stress-specific proteins. As StressME is tuned to compute metabolic and gene expression responses under mild acid, oxidative, and thermal stresses, it is useful for engineering and health applications. The modular design of our open-source package also facilitates model expansion (e.g., to new stress mechanisms) by the computational biology community.

The transcriptional regulatory network (TRN) of E. coli consists of thousands of interactions between regulators and DNA sequences. Regulons are typically determined either from resource-intensive experimental measurement of functional binding sites, or inferred from analysis of high-throughput gene expression datasets. Recently, independent component analysis (ICA) of RNA-seq compendia has shown to be a powerful method for inferring bacterial regulons. However, it remains unclear to what extent regulons predicted by ICA structure have a biochemical basis in promoter sequences. Here, we address this question by developing machine learning models that predict inferred regulon structures in E. coli based on promoter sequence features. Models were constructed successfully (cross-validation AUROC > = 0.8) for 85% (40/47) of ICA-inferred E. coli regulons. We found that: 1) The presence of a high scoring regulator motif in the promoter region was sufficient to specify regulatory activity in 40% (19/47) of the regulons, 2) Additional features, such as DNA shape and extended motifs that can account for regulator multimeric binding, helped to specify regulon structure for the remaining 60% of regulons (28/47); 3) investigating regulons where initial machine learning models failed revealed new regulator-specific sequence features that improved model accuracy. Finally, we found that strong regulatory binding sequences underlie both the genes shared between ICA-inferred and experimental regulons as well as genes in the E. coli core pan-regulon of Fur. This work demonstrates that the structure of ICA-inferred regulons largely can be understood through the strength of regulator binding sites in promoter regions, reinforcing the utility of top-down inference for regulon discovery.