Even in the best-studied model bacterium, Escherichia coli, the cis-regulatory architecture — the base-pair positions of transcription factor binding sites and the identity of the factors that bind them — remained unknown for roughly 65% of promoters when Reg-Seq was introduced. Reg-Seq couples a massively parallel reporter assay (mutagenized 160 bp promoter windows linked to barcodes, read out by RNA sequencing across growth conditions) with DNA-affinity chromatography and mass spectrometry to move promoters from complete regulatory ignorance to identified architectures, but it was demonstrated on only about one hundred promoters. The source paper frames an explicit open problem: what is required to scale this strategy to the entire E. coli regulatory genome. Later literature (2020-2026) resolves parts of the surrounding methodology — genome-wide promoter-activity maps, deep-learning binding-site callers, simulation theory for overlapping sites, and CRISPR-interference-based (CRISPRi, targeted repression of chosen genes) transcription-factor perturbation screens — but a 2025 successor study still covers only about one hundred genes across 39 environments, still uses the 160 bp window that misses distal regulatory elements, and identifies transcription factors only where identifiable. The complete genome-wide, base-pair-resolution, condition-dependent regulatory dissection of E. coli, with transcription-factor identity, therefore remains open.

Background

A bacterial promoter is the DNA region where RNA polymerase (RNAP) initiates transcription, and its activity is controlled by transcription factors (TFs) — sequence-specific DNA-binding proteins — that activate or repress transcription from binding sites within or near the promoter. The cis-regulatory architecture of a promoter specifies the base-pair positions of these binding sites and which TF binds each site; it is the information needed to predict expression from sequence and to design synthetic regulatory circuits. Massively parallel reporter assays (MPRAs) measure the expression driven by thousands to millions of sequence variants of a regulatory region in one pooled experiment. In the bacterial MPRA Sort-Seq, a mutagenized promoter library drives expression of a fluorescence reporter, cells are sorted by fluorescence, and sequencing of the sorted bins yields, for every mutated base pair, its effect on expression. Reg-Seq, introduced in the source paper (Ireland et al., eLife 2020), replaces cell sorting with barcoded RNA sequencing so that many promoters can be interrogated in parallel: each promoter's 160 bp region (115 bp upstream to 45 bp downstream of the transcription start site, TSS — the DNA position where transcription begins) is mutagenized, and mutual-information 'information footprints' reveal which base pairs affect expression, i.e., where binding sites lie. For each footprint-detected site, an energy matrix — a position-specific model assigning each base a TF-DNA binding energy — can be inferred, and the TF itself is identified by DNA-affinity chromatography (fishing with the site as bait) followed by mass spectrometry, with confirmation from knockout strains. Reg-Seq recapitulated known architectures and discovered regulation for more than 80 previously uncharacterized promoters in 12 growth conditions, yet in E. coli the architecture of over 65% of promoters was (and largely remains) unknown. The source paper identifies shortcomings that must be overcome to reach a complete genome-wide dissection — stated chiefly in its Discussion, with the inability to resolve overlapping binding sites appearing in its automated site-identification method and appendix — and this framing was later taken up directly by follow-up work: (1) putative binding sites were found by manual curation of information footprints, and automated statistical methods with better power, accuracy, and scalability are needed; (2) TF identification by DNA-affinity chromatography with mass spectrometry plus knockout confirmation is low-throughput (only 13 of the TF targets in that study were identified by mass spectrometry), and the paper proposes multiplexed mass spectrometry and multiplexed Reg-Seq on libraries of gene knockouts as routes forward; (3) the 160 bp contiguous mutagenized window around the TSS cannot capture regulatory action at a distance, such as repressor binding sites that mediate DNA looping (bending of the DNA so that a distant site contacts RNAP) or enhancer-like elements (distal DNA sequences that activate a target promoter from afar), so non-contiguous genomic regions must become interrogable; (4) the site-identification algorithm cannot resolve two binding sites that are very close together or overlapping (two TFs, or a TF over an RNAP site), and — as later simulation work showed — such overlaps distort MPRA summary statistics and inferred energy matrices, so strategies to deconvolve overlaps remain an ongoing challenge; (5) the method can fail when the mutation window misses the most active TSS, when strong sites obscure weak ones, or when the tested growth conditions do not elicit the regulation. These five gaps jointly block one objective: a base-pair-resolution map of the regulatory logic of the whole E. coli genome.

Problem Statement

How can the Reg-Seq strategy be scaled from roughly one hundred promoters to a complete regulatory dissection of the Escherichia coli genome — a base-pair-resolution map, for essentially all promoters, of the binding-site positions, the identity of the transcription factors that bind them, and the growth conditions under which they act? Following the shortcomings identified in the source paper and taken up by follow-up work, this requires overcoming five linked bottlenecks: (i) automated statistical analysis of information footprints to identify putative binding sites, replacing manual curation, with sufficient power and accuracy for genome-wide operation; (ii) high-throughput identification of the transcription factors occupying discovered sites, replacing the low-throughput DNA-affinity chromatography plus mass spectrometry plus knockout-confirmation workflow (for example via multiplexed mass spectrometry or multiplexed Reg-Seq on libraries of gene knockouts); (iii) extension beyond the contiguous 160 bp mutagenized window around the transcription start site so that regulatory action at a distance — DNA looping and enhancer-like elements — is captured, including non-contiguous genomic regions; (iv) strategies to deconvolve overlapping binding sites (multiple transcription factors, or transcription factor sites overlapping RNAP sites) so that inferred energy matrices reflect true DNA-protein interaction energies; and (v) designs that include the active transcription start site and the growth conditions under which each promoter is actually regulated. A complete answer delivers the methods that remove these bottlenecks together with their demonstration at genome scale, yielding the full cis-regulatory map of E. coli as the completion boundary; advances on individual bottlenecks are partial progress toward that single objective.

References

  1. Ireland, William T.; Beeler, Suzannah M.; Flores-Bautista, Emanuel; McCarty, Nicholas S.; Röschinger, Tom; Belliveau, Nathan M.; Sweredoski, Michael J.; Moradian, Annie; Kinney, Justin B.; Phillips, Rob (2020). Deciphering the regulatory genome of Escherichia coli, one hundred promoters at a time. eLife 9:e55308. DOI: 10.7554/eLife.55308. https://doi.org/10.7554/eLife.55308
  2. Urtecho, Guillaume; Insigne, Kimberly D.; Tripp, Arielle D.; Brinck, Marcia S.; Lubock, Nathan B.; Acree, Christopher; Kim, Hwangbeom; Chan, Tracey; Kosuri, Sriram (2023). Genome-wide Functional Characterization of Escherichia coli Promoters and Sequence Elements Encoding Their Regulation. eLife, 2023. DOI: 10.7554/eLife.92558. https://doi.org/10.7554/eLife.92558 (preprint 2020, bioRxiv 10.1101/2020.01.04.894907)
  3. Pan, Rosalind Wenshan; Röschinger, Tom; Faizi, Kian; Garcia, Hernan G.; Phillips, Rob (2024). Deciphering regulatory architectures of bacterial promoters from synthetic expression patterns. PLoS Computational Biology 20: e1012697. DOI: 10.1371/journal.pcbi.1012697. https://doi.org/10.1371/journal.pcbi.1012697
  4. Han, Yichao; Li, Wanji; Filko, Alden; Li, Jingyao; Zhang, Fuzhong (2023). Genome-wide promoter responses to CRISPR perturbations of regulators reveal regulatory networks in Escherichia coli. Nature Communications 14: 5757. DOI: 10.1038/s41467-023-41572-4. https://doi.org/10.1038/s41467-023-41572-4
  5. Röschinger, T.; Lee, H. J.; Pan, R. W.; Solini, G.; Faizi, K.; Quan, B.; Chou, T. F.; Mani, M.; Quake, S.; Phillips, R. (2025). Illuminating the uncharacterized regulatory genome of E. coli with massively parallel reporters (v1 title: The Environment-Dependent Regulatory Landscape of the E. coli Genome). bioRxiv preprint. DOI: 10.1101/2025.05.13.653802. https://doi.org/10.1101/2025.05.13.653802
  6. Karshenas, Arman; Röschinger, Tom; Garcia, Hernan G. (2026). Predictive modeling of gene expression and localization of DNA binding site using deep convolutional neural networks. PLoS Computational Biology 22: e1014092. DOI: 10.1371/journal.pcbi.1014092. https://doi.org/10.1371/journal.pcbi.1014092
  7. Li, Gaoyuan; Burrows, Joshua T.; Lou, Xuwen A.; Palsson, Bernhard O.; Zielinski, Daniel C. (2026). Disentangling factors affecting bacterial transcriptional regulatory network inference. iScience 29: 116834. DOI: 10.1016/j.isci.2026.116834. https://doi.org/10.1016/j.isci.2026.116834