Title: The Escherichia coli Transcriptome Consists of Independently Regulated Modules

Underlying cellular responses is a transcriptional regulatory network (TRN) that modulates gene expression. A useful description of the TRN would decompose the transcriptome into targeted effects of individual transcriptional regulators. Here, we applied unsupervised learning to a compendium of high-quality Escherichia coli RNA-seq datasets to identify 70 statistically independent signals that modulate the expression of specific gene sets. We show that 50 of these transcriptomic signals represent the effects of currently characterized transcriptional regulators. Condition-specific activation of signals was validated by exposure of E. coli to new environmental conditions. The resulting decomposition of the transcriptome provided: (1) a mechanistic, systems-level, network-based explanation of responses to environmental and genetic perturbations, (2) a guide to gene and regulator function discovery, and (3) a basis for characterizing transcriptomic differences in multiple strains. Taken together, our results show that signal summation forms an underlying principle that describes the composition of a model prokaryotic transcriptome. Main: The transcriptional regulatory network (TRN) senses and integrates complex environmental and intracellular information to coordinate gene expression of a cell. Reverse engineering the TRN informs how an organism responds to diverse stresses and unfamiliar environments. A fully characterized TRN would enable the prediction and mechanistic explanation of an organism’s dynamic adaptation to environmental or genetic perturbations. Reconstruction of a genome-scale TRN requires a substantial number of experiments to integrate the binding sites for each regulator and characterize their activities. Unlike eukaryotic TRNs, which contain highly-connected co-associations, prokaryotic TRNs exhibit a simpler structure; over 75% of genes in the model bacteria Escherichia coli are known targets of two or fewer TFs (Fig. S1a). The TRN structure is encoded in the genome as regulator binding sites and is invariant to environmental dynamics. However, environmental and genetic perturbations alter the activity states of transcriptional regulators to change their DNA binding affinity, which in turn modulates the transcriptome in a condition-specific manner. Thus, a measured expression profile reflects a combination of the activity of all transcriptional regulators under the examined condition, posing a fundamental deconvolution challenge. Compendia of expression profiles have been leveraged to infer TRNs by identifying shared patterns across gene expression profiles, rather than using direct DNA-TF binding information. Many inference methods define groups of genes, or modules, with similar expression profiles that are often functionally related or co-expressed. A recent review showed that independent component analysis (ICA), a signal deconvolution algorithm, outperformed most other module detection algorithms in identifying groups of coregulated genes. ICA is a blind source separation algorithm used to deconvolute mixed signals into their individual sources and determine their relative strengths. Prior application of ICA to microarray expression data has identified co-expressed, functionally-related gene sets that often map to metabolic pathways. The overall expression levels, or activities, of the gene sets have been leveraged to classify tumor samples and connect transcriptional modules to disease states. A current challenge for analyzing transcriptional regulation is to separate the conditioninvariant network structure from its condition-dependent expression state on a genome scale. Here, we overcome this limitation for the E. coli TRN by simultaneously extracting its structure and regulator activities from a transcriptomics compendium. This approach relies on: (1) the availability of high-quality, self-consistent, and condition-rich expression profiling datasets; (2) the use of ICA to concurrently identify regulator targets and activities; and (3) validation through the association of inferred regulator targets with observed molecular interactions. The elucidated TRN structure deconvolutes transcriptomic responses of E. coli into a summation of conditionspecific effects of individual transcriptional regulators.