ExcludonFinder: mapping transcriptional overlaps between neighboring genes

Abstract

Bacteria regulate neighboring genes via overlapping transcription in untranslated regions (UTRs), forming excludons. This overlap leads to transcriptional interference and RNase III-mediated mRNA degradation, resulting in mutually exclusive gene expression, where the activation of one gene suppresses its neighbor. Although individual examples of excludons have been described in various bacterial species, a comprehensive excludon map of a bacterial genome has yet to be established. In this study, we constructed the excludon map of Escherichia coli and Staphylococcus aureus using publicly available RNA-seq data and a newly developed computational tool, ExcludonFinder (https://excludonfinder-unavarra.com). Our analysis identified 16 divergent and 165 convergent excludons in E. coli, as well as 10 divergent and 28 convergent excludons in S. aureus. To validate these findings, we used four independent datasets: detection of double-stranded RNA capture via the Tombusvirus p19 protein, accumulation of short RNAs from RNase III activity, overlap with predicted transcriptional terminators, and single-cell expression analysis. As a proof of concept, we examined transcriptional changes in E. coli under antibiotic stress, revealing that the relBE-ydfV excludon exhibits opposing expression patterns in response to multiple antibiotics. Our findings reveal the widespread presence of excludons and their broad relevance in bacterial gene regulation.

Graphical Abstract

Graphical Abstract.

Introduction

In bacteria, genes encoding proteins involved in the same biological process are often organized into multigene clusters that are transcribed together from a single promoter. These clusters are called operons [1]. This genetic arrangement ensures the coordinated expression of genes whose products are required for a specific biological process and prevents the process from starting if one of the genes is not expressed [2–4]. High-throughput transcriptomic (RNA-seq) approaches have further revealed that bacteria can coordinate the expression of neighboring genes by a mechanism involving overlapping transcription [5, 6]. The overlapping process can occur in two different topologies depending on the transcriptional orientation of the genes: convergent overlap (→←), when the transcripts overlap in the 3′ untranslated regions (UTR); and divergent overlap (←→), when the transcripts overlap in the 5′ UTR. Regardless of the region involved, the overlapping process coordinates the expression of neighboring genes in two ways: through transcriptional interference, where the RNA polymerase transcribing one gene can interfere with the machinery of the other [7, 8], and through the degradation of overlapping RNAs by RNase III [9–12]. The result is that the gene with the higher transcription rate suppresses the expression of the neighboring gene. This regulatory strategy, wherein adjacent genes coordinate their expression through overlapping transcription in their UTRs was termed an excludon [13–16]. RNA-seq analyses across various bacteria have identified excludons in both Gram-negative and Gram-positive bacteria [17]. For instance, Alvarez-Escribano et al. [18] demonstrated that in Nostocales cyanobacteria, the expression of glutamine synthetase (glnA) is regulated through 3′ UTR transcriptional overlap with its neighboring gene gifA. Similarly, recent studies have demonstrated that intracellular Edwardsiella piscicida adopts a mitochondria-like energy program, marked by suppressed glycolysis and enhanced TCA cycle activity. Interestingly, genes predicted to be part of excludons are significantly enriched among those with altered sense and antisense transcription, suggesting a broad but critical role for excludons in intracellular acclimatization [19]. In Listeria monocytogenes, the transcriptional repressor MogR forms an excludon with genes encoding components of the flagellar export apparatus (FliP and FliQ). Induction of the secretion apparatus genes suppresses the expression of MogR, while expression of MogR, in turn, inhibits the secretion apparatus both at the transcriptional and post-transcriptional levels [20]. Beyond conventional excludons, where overlap occurs in either the 5′ or 3′ UTRs, a transcriptional architecture has recently been described in which overlap occurs simultaneously at both UTRs. In this configuration, genes belonging to the same operon are separated by divergently transcribed genes. This organization, termed noncontiguous operons, represents a variation of the classical operon model, combining simultaneous transcription under the control of a shared promoter with overlapping transcription to coordinate gene expression [21, 22]. A striking example of noncontiguous operon architecture is observed in Staphylococcus aureus and Escherichia coli phages [23, 24]. These phages encode a family of proteins known as Tha, which confer immunity to superinfection by interacting with the tail proteins of competing phages. To avoid disrupting the assembly of their own tail structures, the phages must tightly repress tha gene expression during replication. This is accomplished by organizing the tha-1 gene and its inhibitor, ith-1, within a noncontiguous operon. Upon replication initiation, ith-1 is expressed and represses tha-1 through overlapping transcription, preventing a tha-1-dependent autoimmune-like response.

In the same way that operons coordinate the expression of functionally-related genes, the identification of excludons anticipates the existence of functional relationships between the products of the genes [25–27]. In the case of operons, their transcriptional organization ensures that the products of all the genes involved are present in the bacterium at the same time. In contrast, excludons are organized to prevent the concurrent expression of neighboring genes, implying that their products are involved in processes that should not occur simultaneously. Currently, there is no tool to systematically identify the excludons present in a bacterial genome. In this study, we developed ExcludonFinder, a web-based tool for the systematic detection of overlapping transcriptional signals between neighboring genes using transcriptome datasets, enabling optimized and automated identification of potentially novel regulatory interactions. We applied ExcludonFinder to transcriptomic data available in databases and determine the map of excludons in the genomes of E. coli and S. aureus. Our analysis revealed that E. coli possesses significantly more excludons than S. aureus and that convergent excludons (overlapping in the 3′ UTR) are more prevalent than divergent ones (overlapping in the 5′ UTR). Moreover, these excludon maps provide insights beyond those of conventional transcriptome analyses, which tipically focus on identifying up- or downregulated genes. By revealing neighboring gene pairs that exhibit coordinated expression reversal, excludon maps uncover an additional layer of regulation, deepening our understanding of how bacteria dynamically adapt to environmental changes.

Materials and methods

Transcriptomic datasets

To identify the excludon map for E. coli, all RNA-seq samples for the strain MG1655 available since 2023 were retrieved from the Sequence Read Archive (SRA). For S. aureus, the S. aureus subsp. aureus USA300 strain was used. However, due to the limited amount of data, all samples available from 2021 onwards were included. We filtered the datasets to retain only stranded RNA-seq samples using Salmon [28], and further required that at least 80% of aligned reads mapped to coding sequences (CDS) using the featureCounts function from the Rsubread [29] package. This filtering ensured high-quality, strand-specific RNA-seq data relevant to the strains of interest, with validated alignment accuracy for robust transcriptional annotation. To assess the consistency of our results across different sequencing platforms, we conducted a comparative analysis using Oxford Nanopore long-read libraries from Bioprojects PRJNA1052951 [30] and PRJNA1132232 [31]. Validation in S. aureus was performed using data from PRJNA80817 [10], which includes Illumina libraries for RNA fractions shorter than 50 nucleotides (nt). For E. coli, validation was performed using transcriptomic data from PRJNA512059 [32], which contained genomic ranges of RNase III targets overlapping protein-coding genes. For the single-cell level analysis, we used scRNA-seq data from GSE46915, generated using the PETRI-seq method [33], in a study examining transcription-replication interactions in bacterial genome regulation [34]. Finally, to investigate the modulation of excludon expression under stress conditions, we analyzed transcriptomic data from GSE220559 [35], which encompasses E. coli responses to nine representative classes of antibiotics: tetracycline, mitomycin C, imipenem, ceftazidime, kanamycin, ciprofloxacin, polymyxin E, erythromycin, and chloramphenicol.

ExcludonFinder tool

ExcludonFinder is a web-based and command-line tool specifically developed to identify sense/antisense mRNA transcriptional overlaps between neighboring genes. It is available as a user-friendly web server (https://excludonfinder-unavarra.com) and as a command-line application via the Bioconda repository and GitHub (https://github.com/Alvarosmb/ExcludonFinder) and Zenodo (https://doi.org/10.5281/zenodo.14755373). Comprehensive documentation is provided to guide users through installation and execution. ExcludonFinder integrates RNA-seq data from both Illumina and Oxford Nanopore platforms and is compatible with both local and high-performance computing environments, supporting parallel processing on computing clusters. The tool implements an end-to-end pipeline that includes RNA-seq alignment, nucleotide-level coverage calculation, and transcription unit (TU) annotation to enable genome-wide identification of excludons. For RNA-seq alignment, BWA-mem2 [36] is used with strand-specific Illumina data, while minimap2 [37] is applied for Oxford Nanopore direct RNA sequencing data. These aligners are chosen for their ability to preserve strand specificity, which is crucial for accurately detecting transcriptional overlaps. Strand-specific genomic coverage is calculated using the Samtools package [38], allowing precise signal discrimination between the sense and antisense strands.

Excludome analysis

Intergenic regions and excludon organization were analyzed using genome annotation files (.GFF files). An in-house R script parsed these files, calculate intergenic distances, and identify overlapping transcription units. Statistical comparisons between excludon and non-excludon gene pairs were performed using Wilcoxon rank-sum tests.

Validation of excludon map results

Prediction of terminators in excludome gene pairs

To map predicted transcriptional terminators in E. coli and S. aureus, we used Transterm HP [39], with a modified version of the software capable of processing locus tags instead of gene symbols.

Abundance of symmetrically distributed short RNA reads in excludons in S. aureus

Short RNA-seq reads were aligned to the S. aureus NCTC8325 reference genome using STAR [40], following the methodology described by Lasa et al. [10]. Standard RNA-seq reads were aligned with the ExcludonFinder tool. The number of short reads mapping to CDSs and intergenic regions was quantified using the featureCounts function from the Rsubread package [29]. To assess the correlation in read abundance, Spearman's rank correlation coefficient was calculated.

Abundance of dsRNAs in E. coli excludons

Excludons were identified using our tool with the E. coli K-12 MG1655 genome (U00096.2) as the reference, following the methodology described in [32]. A custom script was used to calculate the percentage of overlap with RNase III targets, based on the genomic coordinates provided by the original study. Only RNase III targets involving 3′ and 5′ overlapping types were included in the analysis.

Analysis of expression ratio using scRNA-seq

To analyze the excludons at single cell level, filtered count matrices were processed using the Seurat package [41]. Cells with low-quality RNA-seq data were excluded. The E. coli dataset included 467,914 cells, and the S. aureus dataset comprised 901,959 cells. For each excludon, the expression ratio, defined as the expression level of one gene relative to its neighboring gene, was calculated. This was compared to a control set of convergent or divergent gene pairs not classified as excludons. To ensure that this latter set did not include possible excludons, both genes in these pairs had to exhibit expression levels above the median, with their transcription start site (TSS) or termination site (TTS) being more than 50 bp apart. To estimate statistical significance in the expression ratio differences Wilcoxon rank sum test with continuity correction was used.

Functional annotation of excludomes

To functionally annotate and characterize the excludomes of E. coli and S. aureus, we used EggNOG-mapper [42], which enabled the assignment of functional categories to genes, including some of those previously annotated as hypothetical, by identifying homologs with known functions across bacterial species (Supplementary Table S3). To detect overrepresented functional categories, we conducted enrichment analysis using the topGO package in R [43]. Furthermore, to evaluate the degree of functional relatedness between genes within each excludon, we calculated semantic similarity scores between Gene Ontology (GO) terms using the following formula:

Here, Sim(t_i, t⍰) represents the semantic similarity between GO terms ti and tj, while Information Content (IC) quantifies term specificity. GO terms that are less frequently represented in the database have higher IC values, allowing us to distinguish between broadly defined functions from more specific, biologically informative ones. To complement the GO-based analysis, we also examined whether gene pairs within excludons shared annotations in KEGG pathways. This approach enabled the identification of metabolic and signaling networks potentially coordinated through excludon architecture, and highlighted functional modules that may be unique to each bacterial strain.

Conservation of excludons across phylogeny

To assess the conservation of excludons across phylogeny, we performed a synteny analysis by first identifying homologous gene pairs in multiple bacterial species using BLAST [44] with an E-value threshold of 1e-5. We used a custom database harboring all reference genomes and annotation files from NCBI RefSeq [45] for Staphylococcaceae family, in the case of S. aureus, and Enterobacteriaceae in the case of E. coli. For these databases, we only kept genomes annotated by RefSeq and with a complete assembly level. Genomic context was examined using GFF annotation files, and synteny was considered conserved when homologous gene pairs were adjacent and maintained the same relative orientation. The analysis included both known excludon pairs and control sets of divergent and convergent non-excludon gene pairs. We calculated the percentage of species in which homologous gene pairs preserved their genomic neighborhood and transcriptional orientation (i.e. maintained opposite-direction transcription). Statistical comparison was performed using the Wilcoxon rank-sum test, with the alternative hypothesis that excludon pairs exhibit a higher degree of synteny conservation than non-excludon controls.

Differential expression analysis

To investigate how excludon expression is modulated under different environmental conditions, we applied our tool to identify excludons in RNA-seq datasets corresponding to nine distinct antibiotic treatments. Following excludon identification, aligned reads were quantified using FeatureCounts [29]. Differential expression analysis was then conducted with the DESeq2 v1.34.0 [46] package. Genes were considered to be differentially expressed if they exhibited an adjusted P < 0.05.

Results

Identification of excludons based on RNA-seq coverage

The first step in identifying adjacent genes with overlapping transcription is to define the transcript boundaries based on RNA-seq read alignments. Following alignment to the reference genome, as detailed in the “Materials and Methods” section, the algorithm calculates nucleotide-level read coverage across the genome and systematically evaluates how coverage changes at each nucleotide flanking every CDS (Fig. 1). Transcript boundaries are defined as regions where coverage, averaged over a three-nucleotide window, drops below a user-defined threshold, by default, 50% of the average coverage for the corresponding gene. Adjacent genes whose transcript boundaries overlap under this criterion are classified as excludons. The choice of threshold is a critical parameter, as it directly influences the balance between sensitivity and specificity. Lower thresholds increase sensitivity, potentially detecting more true overlaps but also introducing false positives. In contrast, higher thresholds enhance specificity but may overlook genuine excludons. Based on recent studies, we selected a 50% cut-off as a conservative yet effective compromise between these trade-offs [47].

Figure 1.

Excludon Finder workflow. The ExcludonFinder workflow begins with RNA-seq data alignment against a reference genome (1), followed by calculating nucleotide-level coverage (2). Convergent (→ ←) and divergent (← →) gene pairs are identified, and their median coverage is calculated (3). Transcriptional units (TUs) are defined based on RNA-seq coverage patterns (4), with TSS and TTS determined when coverage falls below a specified threshold. Excludons (5) are identified as regions where TUs of convergent or divergent gene pairs exhibit overlapping transcription..

To demonstrate the utility of our algorithm, we analyzed the transcriptomes of E. coli MG1655 (4.6 Mb) and S. aureus USA300 (2.8 Mb) from publicly available databases to identify transcriptional overlap between adjacent genes. Since the accurate identification of RNA ends depends on transcript quality, only transcriptomes that passed quality control were included for further analysis (see Material and Methods). After filtering, 58 transcriptomic samples for E. coli strain MG1655 and 7 samples for S. aureus USA300 were included in the analysis (Supplementary Table S1).

To define the core set of excludons, hereafter referred to as the excludome, for each genome, we included only excludons present in at least 50% of the analyzed transcriptomes. The E. coli MG1655 excludome consists of 165 convergent and 16 divergent excludons (Table 1), while the S. aureus USA300 excludome includes 28 convergent and 10 divergent excludons (Table 2). To further validate our findings, we applied ExcludonFinder to recent direct RNA sequencing datasets generated using Oxford Nanopore technology [30, 31]. The results demonstrated strong concordance with the excludons identified in Illumina-sequenced transcriptomes. In E. coli, 141 out of 165 convergent excludons and 10 out of 16 divergent excludons were confirmed. In S. aureus, 16 out of 28 convergent excludons and 4 out of 10 divergent excludons were validated. These excludons are distributed across the genome, with no specific regions showing enrichment of this transcriptional organization (Fig. 2A). Given that intergenic distance (IR) may influence the likelihood of transcriptional overlap between neighboring genes, we analyzed the average IR for excludons. The results showed that IRs had a mean size of 95 nt in E. coli and 118 nt in S. aureus. In both species, gene pairs forming excludons had significantly shorter intergenic distances compared to non-overlapping neighboring genes (Fig. 2B). We also assessed the size of the transcriptional overlap region. E. coli excludons exhibited larger overlap regions than those in S. aureus, and in both species, convergent excludons showed greater overlap than divergent ones (Fig. 2C). Overall, these findings suggest that excludon-forming gene pairs are generally separated by shorter intergenic regions, that convergent excludons tend to have larger overlaps than divergent ones, and that the number of excludons, at least in E. coli and S. aureus, correlates with genome size. The algorithm developed to identify neighboring genes with overlapping transcription has been implemented in a computational tool named ExcludonFinder.

Table 1.

Excludome of E. coli MG1655. Distribution of convergent and divergent excludons, along with noncontiguous operons identified in ≥50% of analyzed RNA-seq samples. Gene names are listed when available; otherwise, locus tags are shown. Genes in bold were also detected using Oxford Nanopore long-read RNA-seq

Type of Excludon	N° of Excludons	Excludons
Convergent	165	insL1-mokC, setA-leuD, yadE-panD, rpnC-panC, dkgB-yafC, yafS-rnhA, yafJ-dpaA, crl-insB9, yagJ-yagK, ykgL-ykgO, yahM-yahN, yaiW-yaiY, yaiZ-ddlA, malZ-acpH, ybaY-ybaZ, ybaA-pdeB, hemH-aes, cueR-ybbJ, ybcW-ylcI, entS-fepB, ybdL-ybdM, ybdR-yldA, nei-abrB, pnuC-zitB, modC-ybhA, ybhQ-ybhR, rlmF-ybiO, fsaA-moeB, yliI-gstB, mdfA-ybjH, amiD-ybjS, ybjD-ybjX, ycaM-ycaN, ycbJ-elyC, zapC-ycbX, sxy-yccS, yccX-tusE, torT-torR, phoH-pgaD, clsC-opgC, yceK-msyB, murJ-flgN, ymfI-ymfJ, tfaP-tfaE, emtA-ycgR, ychH-dauA, ychO-narL, ycjG-mpaA, smrA-dgcM, sieB-ydaF, trg-ydcI, rimL-ydcK, rlhA-yncJ, patD-yncL, ydcD-yncO, pptA-yddH, tam-yneE, marB-eamA, speG-ynfC, clcB-bidA, tqsA-pntB, folM-ydgC, tus-fumC, ydhK-sodC, purR-punR, ydhS-ydhT, ydiE-selO, cho-ves, ynjE-ynjF, nudG-ynjH, yeaK-yoaI, yeaL-nimR, yeaO-yoaF, pdeD-yoaE, yebW-pphA, exoX-ptrB, uspC-otsA, yedP-dgcQ, yedA-vsr, dgcE-alkA, yegD-yegI, yegV-yegW, setB-yeiW, yejF-yejG, arnF-pmrD, yfcH-rpnB, flk-yfcJ, smrB-yfcO, ypdI-yfdY, insL3-pdeA, yfeH-ypeB, xseA-yfgJ, csiE-hcaT, mltF-tadA, nadB-yfiC, yfiM-kgtP, yfjT-yfjU, glaR-kbp, gutQ-norR, galR-lysA, lysR-ygeA, scpC-ygfI, glcC-yghO, plsY-ttdR, sstT-ygjV, yhaL-cyuA, yhaV-agaR, yhbO-yhbP, yhbQ-yhbS, argG-yhbX, yrbL-mtgA, aaeR-tldD, yrdA-yrdB, mscL-arfA, rpnA-bioH, yhhA-ugpQ, yhhL-yhhM, dcrB-yhhS, yhjV-dppF, yiaC-bisC, yiaU-yiaV, mtlR-yibT, waaL-yibX, yicG-ligB, yicL-yicU, yidQ-yidR, cbrA-dgoT, yidX-yidA, asnA-viaA, rbsR-hsrA, bioP-metR, rhaR-rhaT, uspD-fpr, frwD-yijO, oxyR-sthA, yjaH-zraP, zraR-purD, aceK-arpA, yjbJ-zur, pdeC-soxS, fxsA-yjeH, gdx-blc, yjfP-ulaR, idnK-ahr, kptA-yjiJ, yjjJ-lplA, ytjC-rob, tatD-rfaH, ydbA-insD2, ybdD-hcxA, ydbJ-hslJ, yohO-yehW, yicS-nepI, yaiT-ytiB, yoeA-insD3, insO-insI3, yjeV-queG, ykgS-insB3, pmrR-basS, ymgM-cvrA, yneP-ydeM, ymiD-pdeR, yncP-hokB, ysaE-hokA, ythB-nanS
Divergent	16	yaaW-mbiA, gloB-ykfO, acrA-acrR, tesA-ybbA, gfcA-insA4, rne-yceQ, pliG-ymgN, tpx-ycjG, rlmD-barA, yghE-yqhJ, cyaY-yzcX, epmB-efp, queG-nnr, yciZ-ymiD, yoaM-nrdB, yqgG-yqgC
Noncontiguous Operons	3	mpaA-ycjG- tpx, yjeV-queG-nnr, pdeR-ymiD-yciZ

Table 2.

Excludome of S. aureus: Distribution of convergent and divergent excludons, as well as noncontiguous operons identified in ≥ 50% of analyzed samples. Gene names are shown when available; otherwise, locus tags are listed. Genes in bold were also detected using Oxford Nanopore long-read RNA-seq

Type of Excludon	N° of Excludons	Excludons
Convergent	28	SAUSA300_RS00140-SAUSA300_RS00145, SAUSA300_RS00175-SAUSA300_RS00180, SAUSA300_RS00470-SAUSA300_RS00475, SAUSA300_RS01000-ptsG, SAUSA300_RS03000-vraX, SAUSA300_RS03330-SAUSA300_RS03335, SAUSA300_RS04220-SAUSA300_RS04225, SAUSA300_RS05495-SAUSA300_RS05500, SAUSA300_RS05895-SAUSA300_RS05900, SAUSA300_RS07815-SAUSA300_RS15875, SAUSA300_RS07845-SAUSA300_RS07850, SAUSA300_RS07855-SAUSA300_RS15950, SAUSA300_RS09120-SAUSA300_RS09125, SAUSA300_RS09150-SAUSA300_RS09155, SAUSA300_RS09410-SAUSA300_RS09415, yidD-menC, SAUSA300_RS09565-SAUSA300_RS15405, SAUSA300_RS10380-SAUSA300_RS10385, SAUSA300_RS10420-SAUSA300_RS15465, SAUSA300_RS10555-pepG1, SAUSA300_RS10780-SAUSA300_RS15890, SAUSA300_RS10790-SAUSA300_RS10795, SAUSA300_RS11945-SAUSA300_RS11950, SAUSA300_RS12220-SAUSA300_RS12225, SAUSA300_RS13585-SAUSA300_RS13590, copZ-SAUSA300_RS13865, SAUSA300_RS15610-SAUSA300_RS13430, SAUSA300_RS16020-SAUSA300_RS11570
Divergent	10	SAUSA300_RS00405-SAUSA300_RS00410, SAUSA300_RS01055-SAUSA300_RS01060, SAUSA300_RS02895-tadA, SAUSA300_RS07850-SAUSA300_RS07855, ytkD-yidD, SAUSA300_RS10785-SAUSA300_RS10790,SAUSA300_RS10805-SAUSA300_RS10810, glp-moaC, SAUSA300_RS14975-SAUSA300_RS00305, SAUSA300_RS15965-SAUSA300_RS01655
Noncontiguous Operons	3	SAUSA300_RS07845-SAUSA300_RS07850-SAUSA300_RS07855, menC -yidD-ytkD, SAUSA300_RS10795- SAUSA300_RS10790- SAUSA300_RS10785

Figure 2.

Characteristics of excludons in the genomes of E. coli MG1655 and S. aureus USA300. (A) Genomic distribution of excludons. Inner bars represent divergent excludons, while outer bars represent convergent excludons in the genomes of E. coli (left) and S. aureus (right). Selected excludons are labeled with gene names when available; otherwise, locus tags are shown. (B) Distribution of intergenic distances. Boxplots display the distribution of intergenic distances for gene pairs forming convergent and divergent excludons in E. coli (left) and S. aureus (right), compared with neighboring gene pairs that do not form excludons. Boxes indicate the interquartile range; the line within each box represents the median. The x-axis is in nucleotides. (C) Overlap region lengths. Boxplots show the lengths of transcriptional overlap regions for convergent and divergent excludons in E. coli (left) and S. aureus (right). Boxes represent the interquartile range, and the line within each box indicates the median. The x-axis is in nucleotides.

Validation of the excludon map

To validate that the excludons identified by the ExcludonFinder tool exhibit the expected behavior of overlapping transcription between genes, we conducted several types of analyses:

Prediction of terminators in excludome gene pairs.

Overlap between the 3′ UTR regions of convergent genes can arise either because a terminator is located within the 3′ UTR of a neighboring gene or due to leaky termination, where transcription extends beyond the designated TTS [48, 49]. While overlap caused by leaky terminators may have been evolutionarily tolerated or even selected for, its biological significance remains debated, as it is often regarded as transcriptional noise. In contrast, overlapping terminators provide stronger evidence of functional relevance. This scenario implies that the terminator sequence lies within the transcription unit of the adjacent gene, thereby ensuring that overlap consistently occurs whenever either gene is transcribed. To evaluate the proportion of convergent excludons with overlapping terminators, we predicted the genomic locations of the terminators for our set of convergent excludons using Transterm HP algorithm [39]. Our analysis revealed that in E. coli, 146 out of 165 convergent excludons had overlapping terminators, whereas in S. aureus, 6 out of 28 convergent excludons showed this feature (Fig. 3A). These findings support the conclusion that a substantial proportion of convergent excludons result from terminators embedded within the 3′ UTRs of neighboring genes, reinforcing their classification as bona fide excludons.

Figure 3.

Validation of identified excludons. (A) Overlap between convergent excludons identified by ExcludonFinder using RNA-seq coverage data and those predicted based on overlapping terminator positions annotated by Transterm HP. (B) Read mapping to representative excludon regions in S. aureus. Standard and short Illumina RNA-seq reads from dataset PRJNA80817 (10) were mapped to representative excludon loci. The top panel shows a convergent excludon (SAOUHSC_00693–SAOUHSC_00 694), while the bottom panel shows a divergent excludon (SAOUHSC_00663–SAOUHSC_00 664). In both cases, transcript overlap regions display an accumulation of short RNAs on both strands, consistent with RNase III-mediated cleavage of overlapping RNAs. Read alignments were visualized using IGV software. (C) Short RNA Read Abundance: Comparison of the abundance of unambiguously mapped short RNA reads between the antisense and sense strands. Data are shown for S. aureus excludons (left) and intergenic regions of adjacent genes that do not form excludons (right). (D) Correlation with dsRNA regions: Overlap between excludon regions identified by ExcludonFinder in E. coli and double-stranded RNA (dsRNA) regions captured by p19 protein pull-down assays.

Abundance of symmetrically distributed short RNA reads in S. aureus excludons

A characteristic feature of overlapping transcription is that simultaneous expression of both transcripts can lead to the formation of double-stranded RNA (dsRNA). This dsRNA may either promote target RNA degradation via RNase III or enhance transcript stability by shielding it from endo- and exoribonucleases [50]. In a previous RNA sequencing study, we observed numerous 22-nucleotide short RNAs generated by RNase III digestion of overlapping transcripts, with reads mapping in approximately equal abundance to both DNA strands [10]. Based on this observation, we hypothesized that excludons, by virtue of overlapping transcription, would generate symmetrically distributed short RNA reads, indicative of dsRNA formation. To test this, we analyzed S. aureus transcriptomic datasets from Lasa et al. [10], which included both standard and short RNA fractions (<50 nucleotides) (Fig. 3B). From the standard RNA libraries, we annotated 59 convergent and 24 divergent excludons. We then calculated the sense-to-antisense ratio of short RNA reads within overlapping regions by dividing read counts from the less expressed strand by those of the more highly expressed strand. As a control, we computed the same metric for neighboring intergenic regions not associated with excludons.

Our results revealed that excludon regions exhibited a mean sense-to-antisense ratio of 0.68, whereas non-excludon intergenic regions showed a significantly lower ratio of 0.28 (Fig. 3C). These findings support the hypothesis that overlapping excludon gene pairs generate dsRNA, which is subsequently processed by RNase III, a mechanism not observed in adjacent non-overlapping gene regions.

Correlation between double-stranded RNA detected with p19 and excludons identified by ExcludonFinder in E. coli

Due to the lack of available short RNA-seq libraries for E. coli, we employed an alternative strategy to validate overlapping transcription in excludons identified by ExcludonFinder. The plant Tombusvirus p19 protein, which binds to dsRNAs generated by RNaseIII digestion, was employed to characterize the overlapping RNA fraction in the E. coli genome [33]. Our hypothesis was that excludon gene pairs would be overrepresented among the dsRNAs captured by p19 when overexpressed in E. coli. To test this, we analyzed transcriptomic data from Bioproject (PRJNA512059) to identify the set of excludons and determined how many of these excludons were included in the set of dsRNAs captured by p19. The analysis revealed a clear enrichment of p19 clusters in genes forming divergent excludons compared to divergent non-excludon pairs. Notably, 40% of divergent excludons identified by ExcludonFinder overlapped with dsRNA fragments captured by p19, supporting their biological relevance. However, as the p19 pull-down method has limited sensitivity compared to transcriptome-based identification, excludons with low expression levels were likely underrepresented (Fig. 3D). For convergent excludons, validation was not feasible because the p19 protein captured very few dsRNAs in the 3′ UTR regions. This limitation is likely due to sequence-specific biases in p19 binding efficiency, particularly related to differences in GC content.

Analyzing excludons at single cell level

A defining feature of excludon gene pairs is that the expression of one gene typically excludes the expression of its counterpart. While conventional transcriptomic analyses, conducted on bulk bacterial populations, can capture global gene expression patterns, they are unable to reveal whether the expression of one gene truly prevents the other's expression at the single-cell level. To address this, we analyzed gene expression at the single-cell level. Specifically, we compared the expression ratio of genes within excludons to that of neighboring non-excludon gene pairs. When only one gene in an excludon is expressed, the expression ratio will be 0, whereas if both genes are expressed simultaneously, the ratio will be greater than 0. We hypothesized that the likelihood of simultaneous expression would be significantly lower for overlapping gene pairs (i.e. excludons) than for non-overlapping pairs. Consistent with this hypothesis, the results showed that the frequency of two adjacent genes being expressed simultaneously was significantly lower when they were part of an excludon than when they were not (Fig. 4). Despite the limited coverage inherent to single-cell RNA sequencing, which often results in undetected expression for lowly expressed genes, these findings provide strong evidence that overlapping excludon gene pairs are rarely co-expressed within the same cell.

Figure 4.

Expression analysis of excludon-forming and non-excludon-forming gene pairs in E. coli and S. aureus. Density plots show the distribution of expression ratios for adjacent gene pairs with transcriptional overlap (excludons) versus neighboring non-excludon pairs, based on single-cell RNA sequencing data. For excludon pairs, expression ratios near 0 indicate mutually exclusive expression, when one gene is active, the other is typically repressed. In contrast, non-excludon gene pairs exhibit ratios deviating from 0, consistent with co-expression and the absence of transcriptional interference.

Synteny and functional analysis of the excludon gene pairs

The conservation of genetic order, or synteny, can serve as an indicator of coordinated gene expression. Based on this rationale, we hypothesized that gene pairs forming excludons would exhibit higher synteny than adjacent gene pairs not involved in excludons. To test this, we examined the conservation of gene order for both convergent and divergent excludons and compared it to that of neighboring non-excludon gene pairs. Among the 165 convergent excludons identified in E. coli, BLASTN analysis revealed that in five cases, at least one of the two genes forming the excludon lacked a detectable homolog in the other genomes analyzed. These five excludons were therefore excluded, and the synteny analysis was performed on the remaining 160. Similarly, in S. aureus, six of the 28 identified convergent excludons were excluded due to the absence of one or both genes in the comparison genomes, leaving 22 excludons for analysis. Gene conservation was assessed genome-wide using ortholog searches across the relevant species. With the exception of divergent excludons in S. aureus, which were too few to support robust conclusions, excludon-forming gene pairs showed significantly higher synteny than adjacent non-excludon gene pairs in both E. coli and S. aureus (Fig. 5A).

Figure 5.

Conservation and Functional Enrichment of Excludon Gene Pairs. (A) Conservation of gene pair proximity in excludons versus non-excludons across bacterial genomes. The distribution of synteny percentages was compared between excludon-forming and adjacent non-excludon gene pairs in E. coli and S. aureus. This analysis assessed gene pair conservation across species, distinguishing between neighboring genes that form excludons and those that do not. Gene pairs lacking homologs in the compared genomes were excluded. Boxplots show the interquartile range, with the median indicated by the line within each box. (B) GO term enrichment analysis of the E. coli excludome. Bars represent –log₁₀ (p-value) for significantly enriched GO annotations. Numbers on the bars indicate the number of genes associated with each annotation; numbers in parentheses show the proportion of excludon genes relative to the total number of genes with that annotation.

To investigate the functional relationships between gene pairs forming excludons, we performed a functional enrichment analysis of E. coli excludon-associated genes. A similar analysis was not feasible for S. aureus due to the smaller number of identified excludons and more limited functional annotation. In E. coli, excludon genes were significantly enriched in pathways related to stress response, transport systems, and core metabolic processes, suggesting that excludons may contribute to adaptive regulation in response to environmental stimuli (Fig. 5B). Further investigation of specific gene pairs proved strong evidence of functional interdependence. For example, uspD and fpr are both linked to oxidative stress response while entS and fepB are involved in enterobactin and siderophore transport, processes critical for iron acquisition. Additional excludon pairs such as yadE–panD and arnF–pmrD, participate in the same metabolic pathways, reinforcing their functional relationship (Supplementary Table S4).

Modulation of excludon expression based on environmental conditions: a case study

Transcriptome analysis is a widely used approach for examining how gene expression changes under various conditions, such as environmental stressors or the deletion of specific genes. However, conventional analyses typically overlook the presence of excludons and when analysing the transcriptome output do not take into account that neighboring genes may exhibit inverse expression patterns. To highlight the utility of integrating excludon mapping into comparative transcriptomics, we incorporated excludon information into RNA-seq data from E. coli exposed to nine different antibiotics [36]. By focusing on gene pairs forming excludons, we identified 20 excludons that displayed inverse expression patterns across treatment conditions (Supplementary Table S2). A notable example is the ydfV-relBE excludon, in which relBE was consistently repressed while ydfV was upregulated in response to four antibiotic treatments: erythromycin, ciprofloxacin, chloramphenicol, and ceftazidime (Fig. 6B). In most excludons, only one gene showed a significant change in expression, while the other remained unaltered, indicating selective transcriptional modulation within the excludon pair. These results demonstrate that incorporating excludon maps into transcriptomic analysis provides an additional layer of resolution, enabling the detection of coordinated but inverse regulation between neighboring genes.

Figure 6.

Integration of excludon mapping and transcriptomic profiling of E. coli in response to antibiotics. (A) Volcano plots illustrating differentially expressed genes in E. coli under different antibiotic treatments. Gene pairs forming excludons with opposite expression patterns across conditions are highlighted. The ydfV–relBE excludon displays significant inverse expression across four antibiotics, indicating its central role in the antibiotic response. (B) Model of dual-level regulation of E. coli growth by the ydfV–relBE excludon. Under normal conditions (left), expression of the relBE operon represses ydfV, and a balanced RelB (antitoxin)/RelE (toxin) ratio supports cell growth. Under stress conditions (right), ydfV transcription is upregulated, inhibiting relBE expression. Simultaneously, Lon protease degrades the labile RelB protein, allowing RelE accumulation and growth arrest.

Discussion

Although overlapping transcription between neighboring genes in bacterial genomes is well documented, a complete excludon map within a bacterial genome has never been defined. In this study, we mapped the excludons in E. coli and S. aureus using a computational tool that analyzes transcriptomic data to determine transcript boundaries and identify overlapping transcription between neighboring genes. The number of excludons identified varied slightly between the different RNA-seq datasets, probably due to differences in the efficiency of library construction or differences in the conditions under which the bacteria were grown for RNA purification. Regardless of these variations, a set of excludons was consistently detected in 100% of the transcriptomes. To construct the excludon map, we conservatively included only excludons identified in at least 50% of the transcriptomes, indicating that the actual number of excludons in E. coli and S. aureus is likely higher than reported.

The defining feature of an excludon gene pair is the overlap in their transcription. To validate the presence of transcriptional overlap using methods complementary to RNA-seq analysis, we conducted three types of analyses. First, we examined whether the ratio of short RNAs (20 nt) between sense and antisense strands in excludons differs from that observed in intergenic regions between genes that do not form excludons in S. aureus. The rationale behind this approach was that if the short RNAs (approximately 22 nucleotides) were generated by RNase III digestion of complementary mRNAs, the amount of short RNAs mapping to the sense strand should closely match the amount mapping to the antisense strand. The analysis revealed that the ratio of short RNAs on sense/antisense strands is significatively closer to 1 (0.68) in excludons, compared to a much lower ratio (0.28) in the intergenic regions of non-excludon genes. As short RNA libraries for E. coli are not available in databases, transcriptional overlap in excludons was evaluated by comparing the excludon map with the double-stranded RNA sequences captured by the Tombusvirus p19 protein. This analysis showed that 40% of the excludons predicted by ExcludonFinder were captured as dsRNAs by the p19 protein. Notably, most excludons undetected by p19 corresponded to those with lower expression levels, suggesting that the sensitivity of p19 immunoprecipitation is lower than that of RNA-seq analysis.

Conventional transcriptomic analyses typically involve purifying RNA from millions of bacteria, meaning that the presence of transcripts from two adjacent genes does not necessarily indicate that both genes are being transcribed simultaneously in the same cell. In excludon pairs, one gene may be expressed in a subset of cells and the other in a different subset, with no co-expression at the single-cell level. As a result, bulk RNA-seq methods are limited in their ability to detect mutually exclusive gene regulation. Single-cell transcriptome analysis overcomes this limitation by providing gene expression data at the resolution of individual cells. When we examined the co-expression of neighboring gene pairs within single cells, we found that co-expression was significantly more common among non-excludon gene pairs than among excludons. It is important to note that a limitation of this analysis is the relatively low coverage of single-cell RNA-seq, which often results in undetectable expression for many gene pairs. As a consequence, the number of neighboring gene pairs available for analysis is somewhat limited. These findings support the hypothesis that transcriptional overlap within excludons enables the expression of one gene to exclude the expression of the other, suggesting a tightly regulated, mutually exclusive transcriptional mechanism.

A global analysis of S. aureus and E. coli excludomes reveals that overlaps in the 3′ UTR are significantly more frequent than those in the 5′ UTR. One possible explanation is that the region immediately upstream of the transcriptional start site includes the promoter and regulatory protein recognition sequences, while the 5′ UTR itself may also be subject to post-transcriptional regulation such as attenuation. As a result, for an overlap to occur in the 5′ UTR, at least part of the promoter sequence would have to be included in the 5′ UTR of the adjacent gene—an arrangement that naturally imposes sequence constraints. In contrast, overlaps involving the 3′ UTR typically encompass the transcriptional terminator, a region that is generally under fewer regulatory constraints. Moreover, 3′ UTR overlaps can also arise from leaky termination, where transcription extends beyond the intended TTS into the downstream gene. Importantly, such an extension cannot occur in the 5′ UTR due to the unidirectional nature of transcription.

Similar to the evolutionary conservation observed in operon gene clustering, our findings show that excludon gene pairs exhibit significantly closer genomic proximity than neighboring non-excludon genes. This spatial organization suggests that coordinated expression of excludon genes may improve metabolic efficiency and confer a selective advantage. The observed co-regulation also implies functional relatedness between excludon partners. Functional analysis supports this, revealing that excludon-associated genes participate in diverse biological processes, including ester bond hydrolysis and responses to oxidative stress. Broader insights are expected as excludon maps are generated for more bacterial species. An important application of excludon mapping is the functional annotation of previously uncharacterized genes. When one gene in an excludon has a known function, it can provide clues about its partner, uncovering associations that might otherwise go unnoticed. Thus, excludon maps could serve as valuable complements to functional databases such as STRING by adding regulatory context to gene pairs involved in shared biological pathways [51].

To illustrate how incorporating the excludon map enhances the interpretation of phenotypes associated with specific environmental conditions, we analyzed E. coli transcriptomes generated under various antibiotic treatments. Across all conditions, most excludon gene pairs displayed either inverted expression levels or significant changes in expression in only one member of the pair. In particular, the relBE-ydfV and rspT-yaaY excludons showed consistent expression modulation across different treatments, highlighting their potential role in the bacterial response to antibiotics. The relBE-ydfV excludon is particularly interesting as relB is a well-characterized component of the toxin-antitoxin (TA) system implicated in maintaining a small fraction of the population in a dormant state (known as persisters) that survives even at high antibiotic concentrations [52]. Overexpression of the relBE causes a reversible inhibition of cell growth resembling the dormant state characteristic of persister cells [53]. Growth resumes when RelE is neutralized by overexpression of its binding partner, the RelB anti-toxin. Under rich conditions, RelB is produced in sufficient amount to supress RelE activity, but under stress, Lon protease degrades the less stable RelB [54]. Our findings reveal that, beyond this known post-translational regulation, relB expression is also modulated at the transcriptional level through a mechanism involving its neighboring gene ydfV, adding an additional layer of regulatory control (Fig. 6B). In four different antibiotic treatments, we consistently observed that ydfV upregulation was accompanied by relB downregulation. Recent studies support the functional relevance of ydfV in stress adaptation. Yang et al. observed mutations in the ydfV coding region, overlapping with the relBE promoter, emerging in E. coli populations exposed to cyclic antibiotic treatments, with ydfV abundance increasing across successive cycles [55]. These mutations likely impact both ydfV expression and relBE transcription, promoting antibiotic tolerance through transcriptional overlap. Similarly, Tamura et al. reported that mutations caused by the insertion of the Tn10 transposon into the promoter region of the ydfV gene restored the growth capacity of RNaseE-deficient E. coli strains [56, 57]. Interestingly, reversal of the effect of the transposon insertion was achieved by overexpression of the downstream relB gene rather than by overexpression of ydfV. Taken together, these findings demonstrate that the integration of transcriptomic analysis with the excludon map reveals how E. coli modulates the relBE–ydfV excludon in response to antibiotics, positioning ydfV as a functional component of the TA system. This excludon likely contributes to persistence and tolerance via reciprocal transcriptional regulation. Moreover, the identification of additional excludons with similar expression inversions underscores the broader utility of excludon mapping in elucidating bacterial adaptive strategies under environmental stress.

Notes

Present address: Molecular and Celullar Gerontology Group, IMDEA Alimentación, Carr. de Canto Blanco, 8, Fuencarral-El Pardo, 28049, Madrid, Spain

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

Ministerio de Ciencia y Tecnología (Grant/Award Number: PID2020-113494RB-I00); Gobierno de Navarra (MEPERTROBE, Grant/award Number: PC098-099); Instituto de Salud Carlos III (Grant/Award Number: CP20/00154). Source of Open Access funding: Funding to pay the Open Access publication charges for this article was provided by Grants Funding.

Data availability

ExcludonFinder is available as a user-friendly web server (https://excludonfinder-unavarra.com) and as a command-line application via the Bioconda repository, GitHub (https://github.com/Alvarosmb/ExcludonFinder) and Zenodo (https://doi.org/10.5281/zenodo.14755373).