Acquisition and erosion of toxin–antitoxin systems in bacterial chromosomes

Abstract

Toxin–antitoxin systems are widespread in bacterial genomes. Yet, their integration, persistence, and impact on chromosome dynamics remain unclear. Here, we identified 80 type II toxin–antitoxin systems in the single chromosome of Photorhabdus laumondii TT01, 50 of which were experimentally validated. Comparative analysis across the Photorhabdus genus revealed a highly heterogeneous distribution, with toxin–antitoxin systems frequently clustering within discrete genomic regions, either alone or associated with cointegrate-forming transposases and integrases. Toxin–antitoxin systems rarely clustered with other putative defense systems and are preferentially associated with different types of recombinases, suggesting distinct pathways of acquisition for the two types of functions. Functional analyses showed that most validated toxin–antitoxin systems display addictive properties and stabilize plasmids. These addictive toxin–antitoxin systems are preferentially located in genomic regions characterized by high gene turnover, consistent with recent acquisition events. Despite their plasmid-stabilizing capacity, toxin–antitoxin systems do not promote long-term conservation of their immediate chromosomal neighborhoods. Instead, we observed frequent toxin–antitoxin loss, either through complete deletion or toxin pseudogenization, indicating relaxed selection for their persistence in bacterial lineages. We propose a stepwise model for toxin–antitoxin evolution in bacterial chromosomes: initial acquisition mediated by mobile genetic elements, preferential integration into permissive genomic regions, subsequent genetic streamlining of linked loci, and progressive gene loss. The short-lasting linkage between toxin–antitoxin systems and their genomic neighborhoods is consistent with the view that toxin–antitoxin modules can behave as autonomous, selfish genetic elements.

Graphical Abstract

Graphical Abstract.

Introduction

Bacteria can rapidly adapt to challenges, largely because they have the capacity of acquiring genes from other bacteria that may already have the genetic solution for the specific challenge. This results in highly variable gene repertoires, i.e. large pangenomes, among strains of most bacterial species. Horizontal gene transfer most often requires the action of mobile genetic elements (MGEs), like conjugative elements or phages, that transfer large part of their own genomes and occasionally also bacterial DNA (Touchon et al. 2017; Garcillán-Barcia et al. 2025). This means that adaptive genes are embedded within a larger repertoire of acquired genes, of which few are thought to be adaptive, and many could be neutral or even deleterious. In this context, postsegregational killing (PSK) systems (or addictive systems) are particularly intriguing. PSK systems include toxin–antitoxin (TA) and restriction-modification (RM) modules (Kobayashi 2001; Mruk and Kobayashi 2014; Fraikin et al. 2020; Jurėnas et al. 2022; Fraikin and Van Melderen 2024). These systems encode genetically linked toxins and their cognate antidotes that require continuous expression; loss or silencing of the locus leads to antitoxin depletion, resulting in toxin activation and subsequent growth arrest or cell death. The evolutionary dynamics of PSK systems are particularly difficult to resolve because these systems are widely distributed across chromosomes and MGEs and are frequently transferred horizontally. In addition, their persistence can be driven by selection acting on their own function, by selection on linked genetic elements, or by their intrinsic addictive behavior. Short-time scale experiments in the lab have found that such systems can stabilize their whole linkage group, e.g. conjugative plasmids or integrons (Handa et al. 2001; Kobayashi 2001; Loftie-Eaton et al. 2016; Stalder et al. 2017; Richard et al. 2024). Yet, these systems can have other physiological functions beyond their addictive behavior, i.e. both types of systems can protect bacteria from phages (LeRoux and Laub 2022; Kelly et al. 2023; Anton et al. 2025).

TA systems are currently classified into eight different types based on the nature and mode of action of the antitoxin (Fraikin et al. 2020; Jurėnas et al. 2022). Type II TA systems, in which both the antitoxin and the toxin are proteins (hereafter TA systems), appear to be the most abundant. They are found in bacterial chromosomes and MGEs, such as plasmids, prophages, and ICEs (integrative and conjugative elements), and also associated with IS and transposons (Tn3 and Tn7 families) (Anantharaman and Aravind 2003; Makarova et al. 2009; Leplae et al. 2011; Yadav and Rathore 2018; Lima-Mendez et al. 2020; Benler et al. 2021; Ross et al. 2021; Bethke et al. 2024). In some instances, their addictive properties were experimentally demonstrated (e.g. Wozniak and Waldor 2009; Boampong et al. 2020; Gu et al. 2024; Hollingshead et al. 2025). The PSK mechanism was initially characterized for plasmid-encoded TA systems but can be expanded for TAs encoded by other MGEs (Mendoza-Guido and Rojas-Jimenez 2025). An extension of the PSK is the involvement of TAs in competition between MGEs: they were shown to play a role in plasmid-plasmid competition (Cooper and Heinemann 2000; Cooper et al. 2010; Lin et al. 2025) and are increasingly recognized as antiphage systems (LeRoux and Laub 2022; Kelly et al. 2023). Strikingly, most of the bacterial chromosomes contain TA systems, with some chromosomes containing several dozens of them, including homologous ones (Ramisetty and Santhosh 2016; Fraikin et al. 2020; Jurėnas et al. 2022). They are generally located on genomic islands or alone between two core genes, constituting small islets, without any other MGE signature genes (Hayes and Van Melderen 2011). In addition to acting as antiphage systems, some of the TAs were shown to function as anti-addictive modules, thereby safeguarding the cell from homologous systems carried by plasmids (Saavedra De Bast et al. 2008).

While there are indications that conjugative plasmids and ICEs are involved in the dissemination of TA systems between bacteria of the same or phylogenetically distant species (Wozniak and Waldor 2009; Boampong et al. 2020; Gu et al. 2024; Hollingshead et al. 2025), it remains unclear how TA systems integrate into bacterial chromosomes. Another long-standing question in the field is how these systems persist in chromosomes and what are the selective pressures underlying their evolutionary success. In this work, using comparative genomics and experimental validation, we analyzed the localization of TA systems, their relationships with their genomic environment, and their evolutionary trajectories within the Photorhabdus genus, which encompasses genomes exhibiting a wide diversity of TA repertoires and high plasticity (Gaudriault et al. 2008; Ogier et al. 2010).

Results

The Photorhabdus laumondii TT01 genome encodes 50 functional TA gene pairs

To gain insights into the diversity, genomic location, and evolution of type II TA (hereafter referred to as TA systems), we sought to select a bacterial genus comprising genetically tractable bacteria with genomes containing a wide range of TA numbers. We first identified TA systems in the chromosomes of 17 representative γ-proteobacteria using a custom procedure based on TA-Finder (see Materials and methods). Consistent with previous reports (Leplae et al. 2011; Ramisetty and Santhosh 2016; Horesh et al. 2020; Kandel et al. 2021; Singh et al. 2024), the number of predicted systems varied widely across genomes, ranging from 8 to 61 per genome. The ParE/RelE toxin family was found to be the most frequent in the majority of genomes (Fig. 1a). This was particularly evident in the Photorhabdus genus, which was represented by three genomes in this initial analysis. We thus further selected 39 Photorhabdus genomes and analyzed them in detail. This revealed between 35 and 64 predicted TAs, among which 17 to 44 toxins belong to the ParE/RelE family (Table S1). Of the 39 Photorhabdus genomes, that of Photorhabdus laumondii TT01 (hereafter TT01) contained one of the largest numbers of predicted TA systems (61 systems). Careful curation further revealed 19 more systems, resulting in a total of 80 predicted TAs in this single genome (Fig. 1b, Table S2). Around 60% of the predicted toxins belong to the ParE/RelE family.

Figure 1.

Photorhabdus laumondii TT01 genome encodes 80 predicted type II TA systems. a) Predicted type II TA systems within a representative subset of gamma-proteobacteria genomes, including the P. laumondii TT01 genome (in red). TA systems were classified based on the toxin families and represented by different colors as shown in the inset. b) Localization and experimental validation of the TA systems of P. laumondii TT01. The P. laumondii TT01 is represented by a circle. The 80 predicted TA systems are represented according to the color code as in (a). Experimental validation results are represented by the following color code: black represents the systems validated by the kill-rescue and plasmid stabilization assays, gray represents the systems validated only by the kill-rescue assay, and white represents the systems that were not validated by the kill-rescue assay and therefore not tested for plasmid stabilization. TA systems separated by two genes or less are linked by bold brackets.

To test whether these predicted TA systems encode functional ORFs, classical kill-rescue assays were performed in Escherichia coli, as it is technically impossible to delete all predicted TA gene pairs in the TT01 chromosome. Briefly, putative toxin and antitoxin genes were cloned in compatible plasmids, under the control of differentially controlled promoters. An E. coli strain deleted for type II TA systems (Δ10) (Goeders et al. 2013) was co-transformed with toxin-expressing and cognate antitoxin-expressing plasmids. Kill-rescue data showed that 50 TA pairs out of the 80 predicted encoded functional ORFs, i.e. overexpression of the toxin inhibited E. coli growth and overexpression of the cognate antitoxin rescued the toxin-dependent growth defect (Fig. 1b, Figures S1 and S2, Table S3). Of the 30 systems predicted but not validated by the kill-rescue assay, the majority exhibited significant truncations in the toxin ORFs, with only two cases showing point mutations, indicating that a significant proportion of predicted TA systems are pseudogenes (Figure S1). Our analysis did not reveal any discernible bias in favor of a specific toxin family being more prone to be functional.

Type II TA systems cluster in small genomic spots

To study the distribution and chromosome localization of the TAs, we computed the pangenome of the 39 genomes of Photorhabdus genus. The core genome of 1,727 genes allowed us to define genome intervals, which are regions between two consecutive core genes. When intervals are homologous across genomes, i.e. they have the same flanking core genes, they make genome spots. The resulting 3,138 accessory genes of the TT01 genome were found in 481 genomic intervals (as illustrated in Figure S3) (Oliveira et al. 2017). They include all 80 TA systems, most of which (70/80) were identified in intervals with fewer than 50 genes, and 12 were the only genes in the interval. We then confirmed that TA-associated intervals are on average smaller than expected (P-value <0.001, Z-score = −3.88) (Fig. 2d, Table S5, simulation id 1). To test the generality of these results, we clustered the TAs of the genus by sequence similarity (>80% identity) to generate 280 non-redundant TA families, which include the 80 detected in the TT01 genome. TAs were indeed enriched in spots of smaller-than-expected size (Fig. 2d). Intriguingly, TAs also tend to cluster in the chromosome of TT01: we found 14 TA co-localized pairs (expected 2.67, χ² test, P = 1.7 × 10⁻¹²; Table S4, Fig. 1b, bold bracket). This trend was confirmed in the broader dataset including genomes of all the genus (Fig. 2d, Figure S4, Table S6, simulation id 1). Collectively, these findings indicate that TA systems frequently localize and co-localize in small intervals in the genomes of Photorhabdus.

Figure 2.

Type II TA and antiphage systems in Photorhabdus display distinct patterns of integration and cluster with different MGE signature genes. Illustrations of the methodology used a) to compare the tendency of an element (here a TA system) to be retrieved in spots of specific size to a random distribution and b) to compare the tendency of an element (here, an APS for antiphage system) to be retrieved in the same spot than another type of element (here a TA system, TA-assoc. spots) (Avg. #, average number). In-depth illustrations can be found in Figures S4 and S5. The observed distribution (O) of TA systems (blue, N = 280), antiphage systems (red, N = 158), transposases (TNP) (green, N = 1,937), and integrases (INT) (purple, N = 215) was compared to the expected distribution (e) generated by randomly placing them across the Photorhabdus genomic spots (10,000 simulations). A deviation score calculated as (O − E)/(O + E) was applied to assess the difference between the real distribution and the random expectation. c) Deviation scores indicating whether TA systems, antiphage systems, transposases, and integrases are preferentially located in spots that are smaller or larger than expected by random distribution, with spot size defined by the number of distinct genes detected. d) Deviation score indicating whether TA or antiphage systems co-occur in the same spots with another element type indicated in brackets i.e. TA, antiphage, transposase, and integrase, more or less frequently than expected by chance. From either the TA or antiphage perspective, the average number of co-occurring elements of another type was compared with expectations obtained by randomly redistributing the other element type. e) Same co-occurrence analysis was applied to transposase subclasses categorized by their described transposition mechanism: co-integrate forming (coint.) (N = 148), copy-and-paste (copy) (N = 216), cut/peel and paste (cut & peel) (N = 642), and those of unknown mechanism (N/A). All statistics (P-value, Z-score) associated with these analyses are detailed in Tables S5 and S6.

TA and antiphage loci are distributed differently

Several TA systems have been reported to sense and be activated by phage infection, resulting in the growth inhibition or death of infected cells (LeRoux and Laub 2022; Kelly et al. 2023). This process quoted as abortive infection prevents viral replication in infected cells and protects the rest of the population. The TT01 genome possesses one of the largest antiphage arsenals, top 1.5% among 21,739 genomes (Tesson et al. 2025). Since antiphage systems were shown to cluster in so-called defense islands (Makarova et al. 2011; Hochhauser et al. 2023), we tested whether TA systems were integrated in regions prone to harbor antiphage systems. Using Defense Finder (Tesson et al. 2022), we identified 158 distinct antiphage systems in the Photorhabdus genomic spots among which 25 in the TT01 genome. Note that MazEF TA systems were also identified as antiphage by Defense Finder; however, they were not classified as antiphage systems in this study, as they are considered TA systems, which are the focus of the study. As observed across all the bacteria domain (Tesson et al. 2025), RM systems represented the most prevalent antiphage systems in Photorhabdus with 25 distinct occurrences (4 in TT01). Contrary to TA systems, antiphage systems are more frequently found in large spots (Fig. 2c). We thus enquired if such systems are very frequently co-occurring with prophages, which might drive their transfer. Detection of prophages by Genomad revealed that 38% did occur in spots with prophages (Figure S5a). The same analyses applied to TAs revealed a slightly smaller fraction of co-occurrence (36%) (Figure S5b). Hence, prophages may drive the transfer of some systems, but in most cases, these are in spots where phages were not found to integrate.

Antiphage systems are known to cluster together in many genomes (Makarova et al. 2011; Hochhauser et al. 2023). Indeed, 80% of the antiphage systems co-localize with other systems in spots, the latter encoding an average of 7.5 other antiphage systems (Fig. 2d, Figure S4, Table S6, simulation id 10). RM systems exemplify this trend since most (76%) are in spots containing other antiphage systems. We then tested if TA and antiphage also co-occur. We found that their co-occurrence is not significantly different from the random expectation (Fig. 2d, Figure S4, Table S6, simulation ids 2 and 9). Most TA systems did not co-localize with antiphage systems in the genus (197/280) or the TT01 genome (54/80, Figure S6). We then compared the functions of genes co-localizing with TA or antiphage genes. These revealed some commonalties, such as high frequency of unknown function genes and low frequency of some metabolic genes. The major difference related to functions implicated in replication, recombination, and repair that were over-represented in spots of antiphage functions and under-represented in spots with TAs (Figure S7). These findings suggest key differences in the genomic localization of TA and antiphage systems.

TA and antiphage systems are associated with distinct MGEs

Differences in the distribution of loci encoding TA and antiphage systems may reflect distinct modes of mobility and integration. To enquire into this matter, which may help to understand how TAs are acquired, we identified the spots with integrase and transposase genes. The analysis of integrases (215 occurrences) revealed they are predominantly located in large spots, with an average of 272 genes (Fig. 2c). Most integrase genes (63%) were in spots encoding at least one antiphage system, a frequency of co-occurrence that is higher than expected (Fig. 2d, Figures S4 and S6, Table S6, simulation id 16). While TA systems also often co-localize with integrase genes, the effect is low (Z-score: 2.6 versus 5.61 for antiphage functions, Fig. 2d, simulation id 8). These findings suggest that MGEs with integrases are more important contributors to the acquisition of antiphage than TA systems.

The analysis of transposases revealed a more complex picture. Transposase genes are far more abundant than integrase genes in the Photorhabdus genus (1,937 distinct occurrences) and they preferentially localize in smaller spots, a pattern reminiscent of the distribution of TA systems (Fig. 2c). Yet, TA co-occurrence with transposase genes did not differ from random expectation (Fig. 2d, simulation id 3), whereas antiphage systems tend to co-occur with transposase genes (Fig. 2e, simulation id 11). These intriguing findings could be reconciled with the others if different types of transposases co-localize with TAs and antiphage systems. We classified all transposase genes into four functional groups based on their mode of transposition: cointegrate-forming, copy-and-paste, non-replicative (cut or peel-and-paste), and those of unknown mechanisms (Ross et al. 2021). Antiphage systems often co-occur with copy-and-paste transposases, and rarely with cointegrate-forming transposases (Fig. 2e, Figure S8, simulation ids 12 and 13). TA systems often co-occur with cointegrate-forming transposases and, to a lesser extent, with copy-and-paste transposases (simulation ids 4 and 5), but rarely with non-replicative (cut or peel-and-paste) transposases (P-value < 0.001, Z-score = −4.21) (Fig. 2e, simulation id 6). These findings show that TA and antiphage systems preferentially co-localize with different types of recombinases, suggesting that their integration is driven by different mechanisms.

Most functional TA systems of P. laumondii TT01 are addictive

Once a TA is acquired, it may be stabilized in the genome if it provides a fitness advantage to the bacterium or if it's addictive. Addiction is very well documented for plasmid-encoded TA systems; however, little is known about the capacity of chromosomally encoded TAs to mediate stabilization of an unstable replicon. We therefore tested the capacity of the 50 TT01 TA systems validated as functional in the kill-rescue assay (Fig. 1b, Figures S1 and S2) to stabilize an otherwise unstable plasmid in E. coli. This assay is closer to natural physiological conditions than the kill-rescue assay, since the toxin and antitoxin genes are expressed under their native promoter, from a low-copy number plasmid. However, it may be affected by differences of promoter activity in Photorhabdus versus E. coli. The 50 TA pairs, including their native promoter regions, were cloned into an unstable plasmid encoding a constitutively expressed GFP reporter, and plasmid retention was measured by flow cytometry. We confirmed that the control plasmid is lost at a rate of 4% per generation in the absence of any selective pressure, as previously described for this setup (Fraikin and Van Melderen 2024). Thirty-four out of the 50 TA systems (68%) tested elicited a statistically significant increase in plasmid retention as compared to the control plasmid (Fig. 3a, Table S3). Altogether, these data indicate that most of the functional TA systems located on the TT01 chromosome are addictive, in a way reminiscent to those encoded by bacterial plasmids.

Figure 3.

Thirty-four of the 80 Photorhabdus laumondii TT01 TA systems are addictive. a) The ability of the 50 TA systems (color-coded as in Fig. 1a) validated by the kill-rescue assay to stabilize the pNF06 unstable plasmid was tested by flow cytometry. Plasmid retention after 8 to 10 generations was compared to the control (empty pNF06) using the Wilcoxon test (ns: nonsignificant, ***P-value < 0.001). The addiction score was obtained by normalizing each value to that of the control. TA systems that showed statistically significant scores were considered addictive.

Presence of TA systems does not result in the conservation of genetic neighborhoods

Given that 68% of the functional TT01 TA systems provide stability to an unstable replicon, we tested the hypothesis that TAs contribute to the stabilization of their immediate genomic environment in the Photorhabdus genomes. We considered the five accessory genes on each side of a TA system in the TT01 genome as its genomic neighborhood (hereby termed gene neighbors, GN). We then computed the frequency at which these genes have homologs in the corresponding intervals of the other genomes (Fig. 4a and b). The co-occurrence of TAs and their respective gene neighborhoods in the other genomes’ intervals, defined as R_TA/GN, fades quickly as the genomes are increasingly distant from P. laumondii TT01 (vertical gradient of blue in Fig. 4c). This was expected since distant genomes are separated by a larger number of events of gene gain and loss. What was more surprising was that the presence of homologs of the neighborhood genes was nearly independent of the presence of homologs of the TA system (Fig. 4c, rightmost columns as compared to the leftmost columns). For example, in P. laumondii sp. clarkei BOJ-47 genome, we found homologs of around half of the TT01 TA systems (37 out of 80, last column in Fig. 4c) in the corresponding syntenic intervals and their presence had little association with the frequency of homologs of the neighboring genes in the focal genome (ranging from 33% to 58% for the five categories) (Fig. 4c). In the Photorhabdus bodei LJ24-63 genome, the trend is opposite to the expected one, with GN being most conserved when TA systems are the least conserved. Finally, in quantitative terms, the correlation between the percentage of GN conservation and the percentage of TA conservation was not statistically significant in 36 out of the 39 genomes (Spearman, P > 0.05).

Figure 4.

TA systems and their neighboring genes are not conserved across the Photorhabdus phylogeny. a) Methodology to assess the conservation of TA systems and their genomic neighborhoods in the Photorhabdus genomes. Genomic neighborhood associated with a given TA system (GN) was defined as the five upstream and five downstream genes in P. laumondii TT01, if not interrupted by a core gene. The presence of the TA system in the syntenic intervals across the Photorhabdus genomes was determined using TblastN. In case the TA system was detected, the neighboring genes in the genome of interests were compared to the GN established in TT01 (using blastP). Otherwise, all genes within the interval were screened for potential matches with the GN genes. b) Schematic representation of the co-occurrence or the absence of co-occurrence between the P. laumondii TT01 TA systems and their GN across the Photorhabdus genomes. c) For each genome, each GN was classified into one of five categories depending on its overall conservation (in terms of number of genes compared to the original GN), from divergent (0% to 20%) to highly conserved (81% to 100%). For each category and in each genome, the R_TA/GN ratio was computed by determining the proportion of GN retrieved with their associated TA system. Cases where the TA system constitutes the only genes present in the interval were excluded from the analysis. The number of TT01 TA systems retrieved in each genome is indicated in red in the right column. d) Detailed data of (c) represented as a heatmap. The red color indicates that the TA system is conserved; yellow indicates that it is absent. The intensity of the color represents the degree of conservation of the GN (in % of genes present in the TT01 GN). In case the TA system constitutes the only genes within the TT01 interval, only the TA was searched in the corresponding GN in other genomes. The presence/absence of the TA system is represented in purple/grey.

To detail this analysis at the level of each TA system, the co-occurrence of each of the 80 TA systems and GN of TT01 in the Photorhabdus genus was represented as a heat map (Fig. 4d). TA systems are found both when homologs of neighboring genes in the focal genome are also present and when they are absent. For example, the genes neighboring the ParE/RelE4 system in TT01 (indicated by an arrow in Fig. 4d) are retrieved in their corresponding intervals in several Photorhabdus genomes, while the TA system is not. On the contrary, the ParE/RelE16 system is present in its associated interval in distant genomes, but the neighboring genes are not (indicated by an arrow in Fig. 4d). These findings suggest that TA systems have little, if any, influence on the conservation of their flanking genes in Photorhabdus.

Addictive TA systems are preferentially located in spots with high turnover

To gain insight into the impact of TAs on the evolution of the gene repertoires of spots, we measured their genetic diversity with the Sørensen dissimilarity index (β_SOR) (Baselga 2010). This index is the sum of 2 components, the turnover (β_SIM) and the nestedness (β_NES). In this context, turnover measures gene replacement, while nestedness measures changes produced by gene loss or gain (Fig. 5a) (Baselga 2010; Oliveira et al. 2017). To accurately compute these parameters, β_SOR was quantified only among the six fully assembled Photorhabdus genomes and in the spots unaffected by chromosomal rearrangements (see Materials and Methods). Of the 80 TA systems in TT01, 40 are present in spots meeting these criteria (Table S7), including 17 addictive (Fig. 1b, Table S3). We found that spots containing addictive TA systems are more diverse than the others (Fig. 5b, higher β_SOR, P-value = 0.024, Mann–Whitney–Wilcoxon test). The most important contributor to the higher diversity values was gene turnover, since their β_SIM value is much higher (Fig. 5c). These results indicate that spots containing addictive TA systems are more diverse in terms of gene repertoires, which is caused by high gene turnover in the spot.

Figure 5.

Photorhabdus spots encoding addictive TA systems exhibit higher rates of genetic changes compared to those encoding nonaddictive TA systems. a) Schematic representation of the turnover (β_SIM) and nestedness (β_NES) in a given genomic spot. Turnover measures the replacement of a gene family by another, while nestedness estimates the difference in gene numbers and represents gene loss/accretion. b) β-Diversity (β_SOR) distribution for addictive TA systems (N = 17, pink) and nonaddictive systems (including those tested as nontoxic) (N = 23, blue). Only fully assembled genomes were used to accurately compute these values, and spots including breakpoint intervals were excluded from this analysis. c) Partition of the β-diversity in turnover and nestedness of the genomic spots containing either addictive TA systems or nonaddictive TA systems. Dashed circle indicates TA systems for which the turnover is equal to zero (N = 4, 2 overlapping dots).

Independent acquisitions and pseudogenization of TA systems

Our results suggest that TAs can be lost even when the neighborhood is conserved. To understand these processes, we searched for pseudogenized copies of the 50 functional TT01 TAs (Fig. 1b) (see Materials and Methods). We detected pseudogenes for 41 out of the 50 TA systems. Eight of the nine remaining systems showing no evidence of pseudogenization were able to stabilize the E. coli plasmid, although to various extent, except for HicA3, which was inactive (Fig. 3 and 6b). Some are present in a few genomes, and the observed lack of pseudogenes may just be a consequence of their recent acquisition (e.g. RelE/ParE11 and 12). Yet, others are found in other species (e.g. ParE/RelE31 and 42), and our observation further suggests that TA systems exhibit addictive properties upon acquisition.

Figure 6.

Most TA systems from the Photorhabdus laumondii TT01 genome present pseudogenized copies across the Photorhabdus phylogeny. The 50 validated TT01 toxin or antitoxin sequences were searched within their syntenic intervals in 39 Photorhabdus genomes using TblastN (e-value < 0.1; identity ≥ 50%). The best hit was selected by prioritizing ORF-encoded sequences with the lowest e-value; hits not associated with an ORF were considered DNA remnants of toxin and/or antitoxin genes. TA systems were classified into two categories based on the presence of at least one putatively defective copy across the Photorhabdus phylogeny. a) Heatmap showing TA systems with at least one pseudogenized copy. b) Heatmap showing TA systems without detectable pseudogenes. For each toxin (T) or antitoxin (a), colors indicate the status of the best hit: black, intact ORF (≥80% identity and ≥80% coverage); dark gray, truncated ORF; light gray, not encoded within an ORF (no start and stop codons); white, absence in the genome. The HicA3 TA system was classified as non-pseudogenized due to uncertainty in distinguishing true from false positives. c) Global TA repertoire per genome. Blue indicates TA systems with intact toxin and antitoxin genes; red, pseudogenized version of the toxin only; green, pseudogenized version of the antitoxin only; and purple, pseudogenized version of both.

The systems with pseudogenes were present in many different species, indicating that pseudogenization is frequent (Fig. 6a and b), but tends to be more frequent for genomes distant from TT01. This is exemplified by the CcdB/MazE2 and RelE/ParE2 systems (Fig. 6a and c). In contrast, systems like HipA3, RES2 and RelE/ParE19, although showing signs of inactivation in some genomes, are found in distant groups of genomes, suggesting they were acquired independently multiple times. In most cases, the toxin alone (N = 185 of all copies in Photorhabdus) or both the toxin and the antitoxin are pseudogenized (N = 186) (Fig. 6a and c). However, in a few cases, the toxin gene appears intact despite the antitoxin gene absent or defective (N = 8), as exemplified for ParE/RelE27 in two genomes (Fig. 6a). Only nine systems out of 50 showed no detectable pseudogenization (Fig. 6b). Five of these systems are only present in genomes that are very closely related to the TT01 genome (Fig. 6b and c). The remaining four are clustered around a few sets of closely related genomes of Photorhabdus. Closer inspection of the genomic environment of these systems revealed the presence of MGEs around the ParE/RelE12 system (two identical IS110) and around the ParE/RelE11 system (a large ICE). These results indicate that TA systems can be independently acquired multiple times at the same genomic location, likely through repeated integration of MGEs encoding them.

Discussion

TA systems are widespread across MGEs and bacterial chromosomes, yet their distribution varies markedly both among species and between isolates of the same species, suggesting frequent horizontal gene transfer and loss (Makarova et al. 2009; Ramisetty and Santhosh 2016; Jurėnas et al. 2022). The mechanisms driving these processes have remained poorly characterized. Our analysis revealed the presence of 80 type II TA systems in the P. laumondii TT01 chromosome, among which 50 encode bona fide TA gene pairs.

Although TA systems are increasingly recognized for their role in antiphage defense (LeRoux and Laub 2022; Kelly et al. 2023), comparisons with other antiphage systems reveal clear and fundamental differences between the two. Antiphage systems are more often found in larger intervals, a pattern consistent with the concept of defense islands, co-localizing with other antiphage systems and with integrase genes (Makarova et al. 2011; Karamycheva et al. 2026). Antiphage systems have often been reported in large MGEs, such as phages and ICEs, which use integrases to insert into the chromosome (Bondy-Denomy et al. 2016; LeGault et al. 2021; Getz and Maxwell 2024; Brenes and Laub 2025; Darracq et al. 2025; Kieffer et al. 2025). Differences in terms of the distribution of TA and antiphage systems are further supported by their distinct associations with transposable genes, suggesting that they are predominantly mobilized by different classes of MGEs, without excluding that at least some to the TA systems could have antiphage properties.

TA systems tend to be in small spots, cluster together, and co-localize with cointegrate-forming transposases. This aligns with previous findings showing that different TA systems are associated with members of the Tn3 transposon family (Loftie-Eaton et al. 2016; Stalder et al. 2017; Lima-Mendez et al. 2020). For some instances, we could trace how these associations have evolved across Photorhabdus genomes. The ParE/RelE 26 system in TT01 is flanked by two nearly identical palindromic sequences closely resembling the resolution site (res-I) of a Tn3 transposon that carries a highly similar ParE/RelE system (>75% nucleotide identity across the entire operon, including the promoter) on a Klebsiella plasmid (Fig. 7a). Notably, this system is absent in the closely related sp10-54 genome, which contains only a single copy of the res-I site, suggesting that this locus may serve as a hotspot for integration or excision of the TA system. Likewise, the ParE/RelE 4 and ParE/RelE 23 systems in TT01 are nearly identical (99% nucleotide identity) and likely originated from a Tn3-associated ParE/RelE system located at a different chromosomal position in the Photorhabdus temperata Meg1 genome (Fig. 7b), despite the absence of recognizable mobility genes at these loci in the TT01 chromosome. In the closely related Photorhabdus clarkei BOJ-47 genome, both systems are absent, and a ∼28-nucleotide palindromic sequence is present at the ParE/RelE 23 locus, although it could not be associated with any known MGE. Collectively, these observations support that Tn3 transposons might play a key role in the dissemination of TAs across Photorhabdus genomes (Lima-Mendez et al. 2020). All these three ParE/RelE systems identified in TT01 display addictive properties, indicating an active role of these systems either in the transposition mechanism or by forcing transposition as proposed previously (Loftie-Eaton et al. 2016; Stalder et al. 2017; Lima-Mendez et al. 2020).

Figure 7.

Genomic context and putative mobility of ParE/RelE TA systems in P. laumondii TT01. Schematic representation of the chromosomal localization of ParE/RelE 26 (a) and ParE/RelE 4/ParE/RelE 23 (b) TA systems in P. laumondii TT01. Genomic intervals encompassing these loci were compared with syntenic regions in other Photorhabdus genomes lacking the corresponding TA systems to identify DNA signatures potentially associated with TA integration or excision. TA operons, including promoter regions, were further screened across Photorhabdus genomes to detect shared sequence features suggestive of mobility. Candidate sequences were queried against the NCBI plasmid database to identify potential mobile genetic element origins. a) The ParE/RelE 26 TA system is absent in the closely related P. laumondii sp10-54 genome, while the surrounding genomic context is conserved. At the corresponding locus, sp10-54 harbors a 24-bp palindromic sequence that is duplicated upstream and downstream of the TA operon in the P. laumondii TT01 genome. This palindrome shares 90% nucleotide identity with the resolution site I (res-I) of a Tn3 transposon located on the Klebsiella plasmid pRHBSTW-00138-6, which encodes a ParE/RelE 26 homolog displaying >75% nucleotide identity across the full operon, including the promoter region. b) ParE/RelE 4 and ParE/RelE 23 are the systems in the P. laumondii TT01 genome that are nearly identical at the nucleotide level (>95% identity), consistent with a recent duplication event. Both systems are absent in the closely related P. subsp. clarkei BOJ47 genome. At the ParE/RelE 23 locus, a palindromic sequence replaces the TA system, although it could not be assigned to a known mobile genetic element. In the P. temperata subsp. temperata Meg1 genome, a homologous TA system is located at a distinct genomic locus and is flanked by putative Tn3 transposases and a putative resolvase.

Although we observed that most of the active TT01 TA systems are addictive and able to stabilize an otherwise unstable plasmid under laboratory conditions, our comparative analysis across the Photorhabdus phylogeny indicates that TA systems do not confer long-term stability to their immediate genomic vicinity. This conclusion is supported by the lack of correlation between the conservation of TAs and that of adjacent genes, as well as by the observation that spots harboring addictive or nonaddictive TA systems are similarly prone to gene loss, as measured by β_NES. This apparent discrepancy most likely stems from the experimental setups commonly used to assess the contribution of TA systems to plasmid stability (Gerdes et al. 1986; Van Melderen et al. 1994; Loftie-Eaton et al. 2016; Stalder et al. 2017; Fraikin and Van Melderen 2024). Such assays typically rely on highly unstable replicons, in our case the mini-F replicon, that encode only the minimal machinery required for replication and its control and lack partition systems that are usually present on large natural plasmids. While this reductionist assay is ideal to evaluate TA-mediated addiction and therefore their selfish character, it may not further inform us on the role of TA systems at longer time scales. Of note, the contribution of TA systems to the stability of large natural plasmids, which often encode multiple TA modules in addition to dedicated partition and maintenance systems, has, to our knowledge, remained largely unexplored.

Our data indicate that addiction does not influence chromosomal gene loss, which is typically gradual, occurring through point mutations, truncations, or small deletions. Furthermore, we frequently observe pseudogenization of TA systems across the Photorhabdus phylogeny, in most of the cases, with patterns indicative of a two-step decay process in which toxin inactivation precedes antitoxin loss. We observed some cases where the antitoxin appears defective and the toxin functional. One possible explanation for these unexpected findings is that toxins may be defective due to nonsynonymous mutations or not be expressed due to polar effects. Antitoxin loss could also be compensated by the presence of other similar TA systems in the genome. From a broader perspective, loss of TA systems could have a variety of selective reasons (Mine et al. 2009). It could be due to relaxed selective pressure, e.g. because those having antiphage or anti-addiction function are no longer useful against phages or plasmids that have evolved resistance, or due to the loss of TA addictive potential, e.g. toxin pseudogenization or complementation by another system(s). Finally, successful TA loss might be infrequent because of its addictive potential, but these rare cases might be strongly selected for when the system is costly to the bacterium.

Altogether, our data suggest a plausible scenario for the acquisition, conservation, and loss of TA systems. We suggest that TA systems are often carried on transposable elements, consistent with the associations with Tn3-family transposons and IS110 in this work, and with ISPa17 and the Tn7 element in past works (Benler et al. 2021; Ross et al. 2021). These elements are frequently carried by conjugative plasmids, facilitating their horizontal transfer and potentially benefiting from TA-mediated stabilization. Following transfer to a new host, TA-encoding transposons may transpose into the chromosome. TAs associated with integrative MGE elements integrate together with their carrier, accounting for the significant albeit weak correlation observed between integrases and TA systems. Both events will preferentially occur in permissive regions containing most MGEs (spots), which explain the association between TAs and high gene turnover (as measured by β_SIM). These newly acquired TAs are addictive and may confer selective advantages, either by functioning as anti-addiction and/or antiphage modules (Saavedra De Bast et al. 2008; LeRoux and Laub 2022; Kelly et al. 2023). Meanwhile, neighboring genes, including transposases, can be lost even if the TA maintains addictive properties, either by translocation or gene deletion. This may result in very small genomic islands that contain only TA systems. Such compact loci can easily be further spread in bacterial populations by transduction, HFr-like conjugation, or natural transformation, since they can be integrated into the chromosome by homologous recombination at the flanking core genes (Mazzamurro et al. 2025), explaining the frequent lack of MGEs in the TA islets. Over time, as selective pressures diminish and toxins become defective, TA systems eventually get lost by deletion or progressive pseudogenization. Collectively, our findings underscore the selfish nature of TA systems.

Materials and methods

Data

All the genomic sequences used in this work were retrieved from the NCBI dataset (last accessed in 2022 Feb) and their accession numbers are indicated in Tables S8 and S9. Detection of TA systems was performed on a representative subset of gamma-proteobacteria genomes (17 genomes rooted with an alpha-proteobacteria genome). Unless specified otherwise, 39 genomes of Photorhabdus isolates, with a maximum of 250 contigs, were considered in this analysis. Phylogenetic trees were reconstructed using the PanACoTA software 1.4.0 (Perrin and Rocha 2021) with default settings, unless otherwise indicated, which (i) predicts all ORFs using Prodigal v2.6.3 (Hyatt et al. 2010), (ii) clusters them based on their amino acid sequence identity using MMseqs2 v13.45111 (−cluster-mode 1, –min-seq-id 0.8 for the Photorhabdus genus or –min-seq-id 0.5 for the gamma-proteobacteria dataset) (Steinegger and Söding 2017) to compute the pangenome, (iii) establishes the core genome as the set of genes present in only one copy in all the genomes under analysis, (iv) produces a protein-guided multiple alignment of every gene family of the core genome using MAFFT v7.490 (Katoh et al. 2002), (v) uses these alignments to generate the phylogenetical tree with the maximum likelihood algorithm using IQtree v2.0.3 (GTR model, -b 10,000) (Nguyen et al. 2015). All tree representations were created using the ITOL website, and their bootstraps are available in Figures S9 and S10.

Definition of intervals and spots

Intervals and spots were defined according to Oliveira et al. (2017) and are illustrated in Figure S3. Briefly, intervals were delimited in the P. laumondii TT01 genome by considering the genomic region between two contiguous core genes as determined by PanACoTA. The content of an interval therefore refers to the set of accessory genes within this region. In P. laumondii TT01, a total of 1,727 genomic intervals were defined, among which 481 contained all the accessory genes. The remaining 72% of intervals lacked any gene (empty intervals) and were excluded from most analyses. Intervals flanked by the same core gene family pair in other genomes were defined as syntenic intervals. Finally, a genomic spot was defined as the set of syntenic intervals, i.e. those intervals in different genomes that are delimited by the same families of core genes. In non-fully assembled genomes, the assignment of intervals to the same spot was done as follows. First, if other unrelated core genes were retrieved in the studied contigs, the syntenic interval was defined as the region between the core gene of interest and the border of the contig opposite to the unrelated core gene. Otherwise, every potential syntenic interval was generated and compared to the reference interval genes using MMseqs2 (easy-cluster algorithm, –min-seq-id 0.8, -c 0.8 on protein sequences), and the combination with the highest number of conserved genes was kept. Note that all syntenic interval interrupted by a non-related core gene were classified as breakpoint interval (BI, illustrated in Figure S3) and excluded from the analysis (N_BI = 518 out of 18,488 non-empty intervals).

Detection of TA systems in the Gammaproteobacteria and Photorhabdus genomes

Type II TA systems were identified using our own custom implementation of TAfinder that supports non-fully assembled genomes. Open reading frames (ORFs) were first predicted using Prodigal, and the resulting protein sequences were screened with hmmsearch v3.3.2 (Potter et al. 2018) to detect type II toxins and antitoxins. For each toxin hit, potential antitoxin partners were searched within the same contig and on the same DNA strand, with a maximum distance of 350 bp between the two genes. Canonical type II TA systems were then classified according to the toxin domain family (ParE/RelE, AtaT, CcdB/MazF, HicA, VapC, HipA, RES, YafO, and FIC). The TAfinder implementation used in this study is available at the following GitHub repository: https://github.com/LudovicBelot/TA_defense_search. To complement this automated approach, the P. laumondii TT01 genome was manually inspected to identify potential TA systems that failed to be detected automatically. All predicted protein sequences from the MAGE Genoscope annotation platform (Vallenet et al. 2006) were submitted to CD-Search from the NCBI website and HHpred (Söding et al. 2005). Candidates showing similarity to either toxin or antitoxin proteins and located within 50 bp of another candidate were retained as part of the P. laumondii TT01 AT repertoire.

Detection of antiphage systems and MGE-associated genes in the Photorhabdus genus

Antiphage systems were identified using DefenseFinder (v2.0.0; models v2.0.2, updated 2025 Apr 14) with default parameters, applied to all Photorhabdus predicted proteomes in gembase format (Tesson et al. 2022). Transposase genes associated with insertion sequences were detected by downloading the complete HMM profile database from ISEScan (last accessed 2023 Nov 16) (Xie and Tang 2017) and running hmmsearch with them (default parameters). Integrase genes were detected using hmmsearch with the PFAM profiles PF00589, PF07508, and PF00239. Transposase genes were further categorized into subclasses depending on their reported mechanism of transposition. Each HMM from ISEScan that matched at least one gene in our dataset was (i) submitted to the EMBL-EBI hmmsearch service to retrieve a representative sequence, (ii) the latter was queried against ISfinder to determine the associated IS family (Siguier 2006), and (iii) annotated for transposition mechanism according to Tnpedia (Ross et al. 2021). The classification resulting from this operation is presented in Table S10. Prophages were identified by submitting all studied genomes to the Genomad web version hosted on the Galaxy server (default parameters) (Camargo et al. 2024).

Genomic spot size and diversity of TA systems, antiphage systems, and MGE-associated genes

The size of each Photorhabdus genomic spot, expressed as the number of distinct genes acquired across the genus in a region delimited by the same pair of core gene, was estimated through the following steps: (i) accessory genes from all Photorhabdus genomes were assigned to genomic spots using their chromosomal coordinates relative to the core genome determined by PanACoTA (as described above), (ii) genes located within the same spot were clustered into protein families using MMseqs2 (easy-cluster algorithm, –min-seq-id 0.8), (iii) for each gene family in a spot, we retained the maximum number of occurrences found within a single interval/genome, to account for possible multiple acquisitions of a same gene family while reducing gene redundancy. The diversity of integrase and transposase genes was evaluated using the same strategy but applied exclusively to genes annotated as integrases or transposases. For TA and antiphage systems, which may consist of more than a single gene, redundancy was determined at the system level. Two systems were then considered redundant if at least one of their component genes were clustered together by MMseqs2. Prophage redundancy was established by clustering their complete nucleotide sequence (same parameters).

Tendency of each type of elements to cluster in an interval or spot of specific size

The tendency of each element type (TA systems, antiphage systems, transposases, integrases) to preferentially integrate into genomic spots of a specific size was evaluated by comparing the observed average spot size associated with each element to the values obtained from random simulations (illustrated in Figure S11). For each simulation, every element from one type was assigned to a random spot, with the probability of selection proportional to the size of that spot (expressed in the number of distinct genes acquired in such a region). If the spot selected was smaller than the element itself (expressed as the number of genes it comprises), a new spot was randomly selected until a suitable one was obtained. Once an element was assigned to a spot, its size was subtracted from the remaining spot size so that the probability of selecting the same region again decreased proportionally. The simulation continued until all studied elements were placed, and the average spot size associated with the elements was computed. A total of 10,000 simulations were performed for each element type. Unilateral P-values were calculated by counting the number of simulations yielding a mean spot size strictly below or above the observed value. Z-scores were computed as:

$$Z\text{-}\mathrm{score}=\frac{\mathrm{real}\mathrm{mean}\mathrm{value}-\mathrm{average}\mathrm{of}\mathrm{the}\mathrm{simulations}\mathrm{value}}{\mathrm{standard}\mathrm{deviation}\mathrm{of}\mathrm{the}\mathrm{simulations}}$$ (1)

In addition, a deviation score was calculated as:

$$\mathrm{Deviation}\mathrm{score}=\frac{\begin{array}{c}\mathrm{Observed}\mathrm{average}\mathrm{spot}\mathrm{size}-\mathrm{Expected}\mathrm{mean}\\ \mathrm{of}\mathrm{the}\mathrm{average}\mathrm{spot}\mathrm{size}\mathrm{values}\end{array}}{\begin{array}{c}\mathrm{Observed}\mathrm{average}\mathrm{spot}\mathrm{size}+\mathrm{Expected}\mathrm{mean}\\ \mathrm{of}\mathrm{the}\mathrm{average}\mathrm{spot}\mathrm{size}\mathrm{values}\end{array}}$$ (2)

A similar procedure was applied to evaluate whether TA systems preferentially integrate into accessory regions of a specific size in the P. laumondii TT01 chromosome. In this case, the analysis was restricted to this single genome, and the size of the genomic intervals identified in P. laumondii TT01 was used instead of considering the spot size.

Propensity of TA and antiphage systems to cluster with different types of MGEs

The propensity of P. laumondii TT01 TA systems to cluster along the chromosome was assessed as described in (Makarova et al. 2009). Briefly, inter-distances between TA systems (expressed by the number of genes) were compared with expectations for a random placement. We applied χ² tests using distance thresholds corresponding to the 25th, 33rd, and 50th percentiles of the observed distribution (Table S4). To evaluate co-occurrence patterns at the genus level, we quantified the tendency of TA or antiphage systems to reside in genomic spots containing other element types (TA, antiphage, transposase, or integrase genes). For each TA or antiphage system, we counted how many elements of the second class were located in the same spot and calculated the average co-occurrence. This observed mean was compared to the expected one generated by randomly distributing the elements of the second type in genomic spots (as illustrated in Figure S12). Note that, when assessing the clustering within the same class (e.g. TA with TA systems or antiphage with antiphage systems), the average expected value was computed by taking each randomly placed element as the perspective point and counting the number of other randomly distributed elements. Simulations were performed 10,000 times, and the statistics were calculated as above.

Kill-rescue assay

Toxin genes were cloned into the pBAD33 vector (Guzman et al. 1995) under the control of the arabinose pBAD promoter and transformed into the E. coli MG1655 Δ10 strain (Goeders et al. 2013). Overnight cultures (37 °C) grown in LB medium supplemented with chloramphenicol (20 µg/mL) and glucose (1%) were diluted in fresh medium to an OD_600nm of 0.01. When the cultures reached an OD_600nm of 0.3, 1 mL was centrifuged at 5,000 RPM for 5 min, resuspended in LB medium without glucose, diluted, and spotted on LB chloramphenicol agar plates containing arabinose (0.2%) or glucose (1%). Toxicity was assessed after overnight incubation of the plates at 37 °C. Antitoxin genes were cloned into the compatible pKK-iq plasmid (Brosius and Holy 1984) under the control of the ptac promoter. The recombinant plasmids were co-transformed with the pBAD33 plasmids encoding the cognate toxin gene. Bacteria containing both plasmids were grown as above in LB medium supplemented with chloramphenicol (20 µg/mL), ampicillin (100 µg/mL), and glucose (1%) and plated on LB plates containing the same antibiotics with or without IPTG (isopropyl β-D-thiogalactoside) (0.1 mM) to induce the expression of the antitoxin genes.

Plasmid stabilization assay

For each of the 50 TA systems validated by the kill-rescue assay, the operons and their promoter regions (∼300 pb upstream of the first ORF) were cloned into the pNF06 plasmid, an unstable derivative of the mini-F plasmid encoding the GFP gene (Fraikin and Van Melderen 2024). The recombinant plasmids were transformed into the MG1655 Δ10 strain (Goeders et al. 2013). Overnight cultures grown in MOPS medium supplemented with kanamycin (25 µg/mL) and glucose (0.4%) were diluted to an OD_600nm of 0.003 in the same medium without kanamycin. After 8 to 10 generations, the fluorescence of 10,000 individual cells was quantified by flux cytometry. The same experiment was performed with an empty pNF06 plasmid to establish the fluorescence baseline. The addictive score was calculated by normalizing the fluorescence to that of the control (empty pNF06 plasmid).

Co-occurrence of P. laumondii TT01 TA systems with their genomic neighborhood across the Photorhabdus phylogeny

Each TA system was associated with an interval based on its genomic coordinates in the P. laumondii TT01 chromosome. The genomic neighborhood associated to the TA system (GN) was defined as the five upstream and five downstream accessory genes if not interrupted by a core gene. In the latter case, the neighborhood is limited to the accessory genes between the core gene and the TA system. The presence of both the toxin and antitoxin genes in the corresponding syntenic intervals of 39 Photorhabdus genomes was determined by tblastN. The TA system was considered present when both components were retrieved with at least 80% identity and coverage and located on the same DNA strand and separated by a maximum of 350 bp. If the TA system was detected in a different genome, the five upstream and five downstream genes were compared to the reference GN using blastP with a threshold of 80% of identity and coverage. Otherwise, each homolog of each protein from the GN was identified within the syntenic interval by using blastP (same parameters). The presence of TA systems and of homologs of their GN (expressed in percentage of genes present) in each Photorhabdus genome was represented as a heatmap. To ascertain whether TA and their GN co-occurred across the genomic spots, each GN was categorized in relation to the presence of homologs in each genome. The ratio of co-occurrence of each category in each genome was calculated as:

$$R_{\mathrm{TA}/\mathrm{GN}}[\mathrm{GN}\mathrm{conservation}\mathrm{category}]=\frac{\mathrm{Number}\mathrm{of}\mathrm{GN}\mathrm{conserved}\mathrm{with}\mathrm{their}\mathrm{associated}\mathrm{TA}}{\mathrm{Total}\mathrm{number}\mathrm{of}\mathrm{GN}\mathrm{conserved}}$$ (3)

Results were concatenated into a heatmap in which the color intensity corresponds to the ratio of GN that co-occurred with their associated TA system.

Analysis of diversity of TA systems’ spots

Using six fully assembled Photorhabdus genomes, the genetic diversity of each spot was established by generating a matrix with the count of each gene family in each genome (computed from Photorhabdus genomic spots). From this matrix, the beta-diversity was calculated using a revisited version of the multiple-site version of the Sørensen dissimilarity index (β_SOR) (Baselga 2010; Oliveira et al. 2017):

$${\beta }_{\mathrm{SOR}}=\frac{{\mathrm{\Sigma }}_{i<j}\min (b_{i,j},b_{j,i})+{\mathrm{\Sigma }}_{i<j}\max (b_{i,j},b_{j,i})}{2\cdot ({\mathrm{\Sigma }}_i{S}_i-S_T)+{\mathrm{\Sigma }}_{i<j}\min (b_{i,j},b_{j,i})+{\mathrm{\Sigma }}_{i<j}\max (b_{i,j},b_{j,i})}$$ (4)

where S_i is the total number of accessory genes within the spot of interest in the genome i, S_T stands for the number of gene families in all genomes, and b_i,j/b_j,i represents the number of accessory genes presents in the genome i but not in j and vice versa. Spot-turnover and nestedness were computed by partitioning the β_SOR index into two additive terms (β_SOR = β_SIM + β_NES). Gene content diversity across the different intervals constituting the spot was measured using the Simpson dissimilarity index (β_SIM):

$${\beta }_{\mathrm{SIM}}=\frac{{\mathrm{\Sigma }}_{i<j}\min (b_{i,j},b_{j,i})}{{\mathrm{\Sigma }}_i{S}_i-S_T+{\mathrm{\Sigma }}_{i<j}\min (b_{i,j},b_{j,i})}$$ (5)

The nestedness, reporting the difference in length between the different sites constituting the spot, results from the following subtraction:

$${\beta }_{\mathrm{NES}}={\beta }_{\mathrm{SOR}}-{\beta }_{\mathrm{SIM}}$$ (6)

P. laumondii TT01 TA systems were attributed to a spot based on their coordinates in the reference genome and classified into two categories depending on their addictive properties. Note that only spots displaying no breakpoint intervals were retained.

Identification of eroded TA systems

We searched for homologs of each TA component in the genomic spots in the 39 Photorhabdus genomes using tblastN with a permissive threshold (e-value ≤ 0.1) to capture gene remnants. It should be noted that the 30 TA systems that were not validated by the kill-rescue assay were considered as pseudogenes and were not included in this analysis. We assessed whether the hits were parts of ORFs by searching for start/stop codons within the same frame. Start/stop codon search was restricted to a maximal range of 33% of the length of the studied gene before and after the coordinates of the hits. The hits with the lowest e-value and comprised, at least partially, within an ORF were kept as a priority; otherwise, the best hit not encoded within an ORF was kept. For each strain, the best hits were translated and compared to the reference protein sequences, and these results were summarized in a heatmap. A TA copy was considered putatively defective when either (i) only one of the components was detected, (ii) the toxin or the antitoxin genes presented important truncations, or (iii) only DNA remnants remained for at least one of the components.

For each TA system categorized as putatively defective, at least one of its copies was manually scrutinized and validated before classifying it as such (Table S11). Additionally, for each genome, we quantified the number of fully conserved TA systems, defined as those in which both toxin and antitoxin sequences matched the corresponding reference with ≥80% identity. All remaining loci were classified as putatively defective and categorized depending on which component was putatively defective (either toxin, antitoxin, or both components).

Author contributions

Conceptualization: L.V.M., A.P., L.B., and E.P.C.R.

Data curation: L.B.

Formal analysis: L.B., E.P.C.R., and L.V.M.

Funding acquisition: L.V.M.

Investigation: L.B. and A.P.

Supervision: L.V.M. and E.P.C.R.

Visualization: L.B., D.J., E.P.C.R., and L.V.M.

Writing—original draft: L.B., D.J., E.P.C.R., and L.V.M.

Writing—review & editing: L.B., A.P., D.J., E.P.C.R., and L.V.M.

Supplementary material

Supplementary material is available at Molecular Biology and Evolution online.

Funding

Work in the L.V.M. lab is supported by the FRFS-WELBIO (ADV X.1537.26F), the International Brachet Stiftung, the Université Libre de Bruxelles (Actions Blanches), and the FNRS-FRS (RICOTTAS, T.0209.22 PDR). L.B. was supported by a fellowship from “Les Amis de l’Institut Pasteur à Bruxelles” (2020 to 2021), an FNRS-FRIA fellowship (2021 to 2024), and a Van Buuren fellowship (2024 to 2025). A.P. has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No 801505.

Data availability

The code used for the analyses in this study is available on GitLab (https://github.com/LudovicBelot/TA_genomic_analysis). The corresponding data is available at Zenodo (https://doi.org/10.5281/zenodo.22145373).