Impact of Natural Transformation on the Acquisition of Novel Genes in Bacteria

Abstract

Natural transformation is the only process of gene exchange under the exclusive control of the recipient bacteria. It has often been considered as a source of novel genes, but quantitative assessments of this claim are lacking. To investigate the potential role of natural transformation in gene acquisition, we analyzed a large collection of genomes of Acinetobacter baumannii (Ab) and Legionella pneumophila (Lp) for which transformation rates were experimentally determined. Natural transformation rates are weakly correlated with genome size. But they are negatively associated with gene turnover in both species. This might result from a negative balance between the transformation's ability to cure the chromosome from mobile genetic elements (MGEs), resulting in gene loss, and its facilitation of gene acquisition. By comparing gene gains by transformation and MGEs, we found that transformation was associated with the acquisition of small sets of genes per event, which were also spread more evenly in the chromosome. We estimated the contribution of natural transformation to gene gains by comparing recombination-driven gene acquisition rates between transformable and non-transformable strains, finding that it facilitated the acquisition of ca. 6.4% (Ab) and 1.1% (Lp) of the novel genes. This moderate contribution of natural transformation to gene acquisition implies that most novel genes are acquired by other means. Yet, 15% of the recently acquired antibiotic resistance genes in A. baumannii may have been acquired by transformation. Hence, natural transformation may drive the acquisition of relatively few novel genes, but these may have a high fitness impact.

Introduction

Horizontal gene transfer (HGT) accelerates the evolution of bacterial genomes by three main mechanisms: DNA conjugation, DNA transfer in viral particles (phages), and natural transformation. Among these three, only natural transformation is under the direct control of the recipient bacteria (Huang et al. 2021). During transformation, bacteria take up exogenous DNA from their environment and integrate it into their chromosome by homologous recombination if it exists a high sequence similarity between exogenous and native DNA (Johnston et al. 2014). Recombination can result in a simple allelic exchange between homologous regions. It has been shown that naturally transformable bacteria tend to have higher rates of recombination, leading to such exchanges in core genes (Oliveira et al. 2017). Yet, if there are two distinct regions of homology in the DNA sequence, recombination may result in the acquisition of novel genes from the exogenous DNA and cause gene deletions in the native chromosomal region. The exact outcome will then depend on the intervening sequence between the two homologous regions in the chromosome and in the incoming DNA (Johnston et al. 2014). This process may be frequent, since it was previously shown that core genes next to recently acquired DNA show higher rates of recombination than the others (Oliveira et al. 2017; Torrance et al. 2025). Furthermore, the potential of transformation to result in the acquisition of novel genes is known for almost a century. Natural transformation was discovered in 1928 by Griffith in Streptoccocus pneumoniae while observing conversions between virulent and non-virulent strains (Griffith 1928). The reported transforming principle was later demonstrated to be DNA (Avery et al. 1944), encoding the complete capsule locus (Griffith 1928; Blokesch 2016). Since then, the core components of the machinery necessary for DNA uptake by natural transformation were uncovered (Johnston et al. 2014; Dubnau and Blokesch 2019; Zuke and Burton 2024). A wide variety of bacterial species are competent and naturally transformable (Lorenz and Wackernagel 1994), among which are numerous pathogens.

Natural transformation was proposed to provide several benefits. The uptaken DNA might be used as a source of nutrients for the recipient bacteria (Redfield 1993b), as a substrate for repair of damaged DNA (Redfield 1993a), or as a source of genetic variation by allelic recombination (Ambur et al. 2016). Allelic recombination would enhance the action of natural selection by reducing linkage disequilibrium (Mazzamurro et al. 2024), purging deleterious alleles (Treangen et al. 2008), helping the fixation of adaptive mutants (Cavassim et al. 2021), and curing chromosomes of costly mobile genetic elements (MGEs) (Croucher et al. 2016). The latter process is favored by the relatively small average size of exogenous DNA, which leads to an excess of gene deletions over acquisitions of non-homologous DNA (Croucher et al. 2016; Apagyi et al. 2018; Tuffet et al. 2024). This results in intragenomic conflict between the bacterium and the MGEs as the latter strive to defend themselves against deletion by blocking transformation (Croucher et al. 2016). The average size of incoming DNA is crucial to the chromosome-curing model (Croucher et al. 2016). Transfer by natural transformation of genomic segments of different sizes (up to 150 kb) and carrying various functions has been observed in a wide variety of bacterial species in the laboratory (Blokesch 2017; Matthey et al. 2019; Godeux et al. 2022). This process can result in the acquisition of small transposons (Domingues et al. 2012; Kloos et al. 2021), plasmids (Kothari et al. 2019), integrons (Domingues et al. 2012), and genomic islands (Hülter and Wackernagel 2008; Domingues et al. 2012; Godeux et al. 2022; Maree et al. 2022). This process of acquisition of MGEs is different from the usual one, since integration in the chromosome takes place by homologous recombination instead of site-specific recombination. Even though the role of transformation in the acquisition of novel genes has often been described (Hülter and Wackernagel 2008; Domingues et al. 2012; Kothari et al. 2019; Maree et al. 2022), its impact in bacterial populations is difficult to assess because the process only leaves recombination tracts, which are not easy to distinguish from the outcome of HGT by high frequency recombination (HFR)-like conjugation (Lloyd and Buckman 1995), phage transduction (Makky et al. 2021), or other less-understood processes like outer-membrane vesicles (Dell’Annunziata et al. 2021). Hence, the impact of natural transformation in gene acquisition in natural isolates has not been sufficiently studied.

Even if the components required for natural transformation are encoded in many bacterial genomes (Denise et al. 2019), not all these bacteria could be shown to be transformable in the laboratory (Johnston et al. 2014). Variations in the regulation of natural transformation could explain part of these negative results. Indeed, large variations in transformation rates have been observed between and within bacterial species (Carlson et al. 1983; Sikorski et al. 2002; Maughan and Redfield 2009; Evans and Rozen 2013; Godeux et al. 2018; Durieux et al. 2019). These variations could contribute to understanding the impact of natural transformation on the acquisition of novel genes. For this, we previously obtained a large collection of genomes and transformation rates of two bacterial species Legionella pneumophila (Lp) and Acinetobacter baumannii (Ab) to study the genetic basis of such variations. These 2 bacterial species are phylogenetically distant Gammaproteobacteria with very different lifestyles. Their genomes are highly dynamic and prone to extensive events of recombination (Gomez-Valero et al. 2011; Snitkin et al. 2011; David et al. 2017; Costa et al. 2018). This makes them particularly relevant to explore whether, and to what extent, natural transformation affects genome evolution. Their propensity to HGT also renders them particularly interesting to evaluate the contribution of natural transformation to gene exchanges, and notably gene acquisitions, among other HGT mechanisms. We have quantified transformation rates (TF^log) in these species using a luminescence-based transformation assay corresponding to a gene gain (Mazzamurro et al. 2024). We set a threshold based on the transformation rates of a known non-transformable strain to distinguish qualitatively transformable from non-transformable isolates (TF^bin) (Mazzamurro et al. 2024). 64% of Ab isolates and 52% of Lp isolates were thus classified as transformable and the rest as non-transformable (Mazzamurro et al. 2024). Our analyses showed that the sequence divergence between the recipient chromosome and the homology arms of their transforming DNA did not affect the patterns of transformation rates. It also showed independence of transformation rates and phylogenetic distances between strains, possibly because divergence was always low (<1%) (Mazzamurro et al. 2024). The large number of sequence types (STs) prevented us from finding any association between STs and transformation rates. However, transformation rates were affected by the origin of the isolate, with environmental isolates exhibiting higher transformation rates than clinical ones in Ab. We observed large variations in transformation rates, over 6 orders of magnitude in Ab and 4 in Lp, as well as frequent recent losses of transformability in both species, and that transformation loss was counter-selected (Mazzamurro et al. 2024). This counterintuitive result could be explained by intragenomic conflicts between MGEs and natural transformation, in accordance with the view that the latter may cure the genome from the former (Croucher et al. 2016). Our previous work focused on the chromosome-curing role of natural transformation, and therefore, gene losses. Here, we shift our focus to the role of natural transformation in diversification through the acquisition of novel genes. For this, we disentangle the impact of natural transformation and other processes on the acquisition of novel genes by HGT.

Our hypothesis is that if natural transformation is important for the acquisition of novel genes, then the comparison between transformable and non-transformable strains should reveal it. For an accurate analysis, because transformation may result in the acquisition or loss of genes (Blokesch 2017), this requires studying the patterns of recombination and transformation in light of the chromosome organization and the types of genes that are acquired. We thus used the existing data on transformation rates in Ab and Lp to assess the impact of this process in the acquisition of novel genes. We evaluated the effect of transformation on gene gains and on the balance between gains and losses. We then assessed gene gains coinciding with homologous recombination tracts, which potentially result from natural transformation. By leveraging the existence of subpopulations of isolates with stable transformation phenotypes in the recent past, we could estimate the effective contribution of natural transformation to the acquisition of novel genes. We characterized these events in terms of the number of genes they encompass, their frequency, and compared them with those caused by integration of MGEs.

Results

Impact of Natural Transformation on Gene Turnover and Genome Size

We analyzed our previously published dataset of genomes and experimental determination of transformation phenotypes in 496 Ab isolates and 786 Lp isolates (supplementary table S1, Supplementary Material online). Transformation was measured using a bioluminescence assay and varied over 6 orders of magnitude in Ab and 4 in Lp (Mazzamurro et al. 2024). Based on the experimental detection limits, we previously classified 64% of Ab and 52% of Lp isolates as transformable and the rest as non-transformable (Fig. 1, supplementary table S2, Supplementary Material online). Transformation rates provide a quantitative (TF^log) measure of transformation, while this classification provides a qualitative (TF^bin) one. In parallel, we identified the pangenomes of both species, which contain 31,103 gene families in Ab and 11,932 in Lp. The persistent gene families were defined as those present in more than 95% of the genomes of each species. We identified 2,629 (Ab) and 2,325 (Lp) persistent genes that were used to compute recombination-free phylogenetic trees, as described before (Mazzamurro et al. 2024), which are used to infer ancestral states, identify terminal branches, and whose phylogenies are taken into account in statistical tests. Across this study, we focus exclusively on terminal branches because the inference of ancestral states of transformation rates and identification of recombination tracts are less accurate for deeper regions of the tree (Mazzamurro et al. 2024). Importantly, the average lengths of terminal branches are not significantly different between transformable and non-transformable strains (Ab: Wilcoxon, P > 0.05; Lp: Wilcoxon, P > 0.05).

Fig. 1.

Recombination-free phylogenetic trees of the core genomes of A. baumannii (top) and L. pneumophila (bottom). The phylogenetic tree was retrieved from (Mazzamurro et al. 2024). The arcs around the trees represent (from inside to the outside). (a) the transformation phenotype of the isolates whose transformation phenotype remained the same along the terminal branch (NT(ancestral) → NT(current), T(ancestral) → T(current)) (b) the transformation phenotype of the isolates whose transformation phenotype changed along the terminal branch (NT(ancestral) → T(current), T(ancestral) → NT(current)) (c) net gene turnover in terminal branch. (d) Number of gene gains in terminal branches. (e) Number of protein-coding genes in the genome.

We first tested the hypothesis that transformation is associated with differences in genome size, as measured by the total number of protein-coding genes. In Ab, the number of genes per genome was not significantly different between non-transformable and transformable isolates when using the qualitative classification of transformation (TF^bin) and was positively associated with transformation rates when using the quantitative phenotype (TF^log) (supplementary table S3, Supplementary Material online: T1, T7). Yet, the latter effect was very low (R² between 0.0061 and 0.0041) and at the edge of statistical significance (P = 0.045). In Lp, transformable genomes were significantly smaller in all the variants of the analysis (P between 0.0015 and 10⁻⁸), but the trait also explained a very small fraction of the variance (R² between 0.01 and 0.038).

We then used the pangenomes and the phylogenetic trees to infer past gains and losses of genes (Fig. 1). We used this information to compute gene turnover as the number of gains minus the number of losses, divided by their sum. Gene turnover was significantly more negative in transformable isolates than in non-transformable ones in all cases for both species (all statistical tests are in supplementary table S3, Supplementary Material online: see T4, T10), even though effects were low (R² between 0.008 and 0.026 in Ab and between 0.045 and 0.028 in Lp). Finally, we analyzed the specific association between gene gains and the transformation phenotype (supplementary table S3, Supplementary Material online). We found no significant association in Ab (supplementary table S3, Supplementary Material online: T2, T8, P > 0.05) and a significant but weak negative association in Lp (supplementary table S3, Supplementary Material online: T2, T8, P < 0.0001, R² ≤ 0.03). In summary, transformation is weakly associated with a net loss of genes in genomes. In Lp, this is accompanied by a lower rate of gene gain, whereas in Ab, there is only a significantly higher rate of loss.

To gain a better understanding of the impact of transformability on gene turnover, we controlled more tightly for the effect of change in transformability (using the binary classification of transformability). We compared gene turnover between isolates predicted to have a non-transformable phenotype in the recent past (first parental node in the tree) and those that have changed to become transformable (NT → T) or remained unchanged (NT → NT). In both species, the isolates that became recently transformable had a more negative gene turnover than those that remained non-transformable (supplementary table S3, Supplementary Material online: T24), the effect being small and only statistically significant for Ab (P < 0.005). Altogether, these results suggest that transformation has a small negative impact on the net rate of acquisition of novel genes.

Gene Gains Associated With Transformability

During natural transformation, the exogenous DNA integrates the chromosome by homologous recombination. To investigate gene gain potentially caused by natural transformation, we started by identifying the genes recently acquired through homologous recombination, since some of these acquisitions could result from natural transformation. We identified the tracts in the genomes that resulted from recent events of homologous recombination (in terminal branches). We previously showed that transformation is associated with elevated rates of recombination (Mazzamurro et al. 2024). To compare regions across the genomes, which are highly variable in terms of gene repertoires, we used a method we have previously developed (Oliveira et al. 2017). We define intervals as the regions between two consecutive persistent genes in a genome. Hence, a gene gain can be placed in a specific interval of its genome. A set of intervals flanked by the same persistent gene families makes a spot (Fig. 2a). Our past results showed that most spots are devoid of accessory genes because HGT is concentrated in a few ones (hotspots) (Oliveira et al. 2017). Accordingly, we identified 2,612 spots in Ab, among which only 588 had accessory genes. Around 80% of the latter had at least one recent gene gain. In Lp, we got 2,315 spots, among which only 449 had accessory genes and 50% had at least one gene gain. We then analyzed the recent gene gains in the spots in relation to the presence of homologous recombination tracts in the flanking persistent genes (Fig. 2b). This analysis reveals that there are tracts of homologous recombination across the chromosome, even if a few loci reveal higher rates. Intervals flanked by persistent genes covered by at least one recombination tract are positively associated with the presence of recent gene gains in both species and using both transformation phenotypes (Lp: χ², odd ratio_{gain/no gain} = 14 in T and NT, P < 2.2 × 10⁻¹⁶; Ab: χ², odd ratio_{gain/no gain} = 5 in T and NT, P < 2.2 × 10⁻¹⁶, see Materials and Methods). This positive association between recombination and recent gene gain suggests recombination in flanking persistent genes favors the acquisition of novel genes in both species.

Fig. 2.

Analysis of gene gains in spots in relation to recombination in flanking accessory genes. a) Scheme describing the definition of a spot of persistent genes in a collection of genomes and its characterization by the presence of gene gains and/or recombination tracts. b) Distribution of the recombination and gene gain rates at spots in relation to the transformation phenotype. T: transformable, NT: non-transformable. c) Distribution of the accessory genes in the different recombination contexts. Statistical tests T1 and T2 are χ² association tests. Odds ratios were calculated as the ratio of the fractions of the number of accessory genes in each genomic context that are gains over the ones that are not (see Materials and Methods). ****: P-value <2.2 × 10⁻¹⁶.

We then detailed the association between homologous recombination and gene gain while taking into account the existence of breaks in the genome assembly. For each accessory gene, we identified if it was gained in the terminal branch and assessed its recombination context (Fig. 2 and supplementary fig. S1, Supplementary Material online): flanked by two (A1), one (A2; B1) or zero (A3; B2) recombining persistent genes. Genes in context A1, A2, and B1 were considered to potentially belong to the recombination tract of their neighboring recombining persistent genes. While this may seem less obvious for A2 and B1, these account for cases where one of the edges of the recombination tract is in another contig (B1) or is not identifiable (A2). Also, events of illegitimate recombination in the non-homologous DNA (accessory genes) have been described and could result in patterns like A2 and B1 (Hülter and Wackernagel 2008). Genes in contigs lacking persistent genes (C1 context) (Lp: 13% of the total; Ab:14% of the total) were excluded. The proportions of accessory genes identified as gains in A1 (Lp: 9.2%; Ab: 27%) and in A2 context (Lp: 9.4%; Ab: 19%) were not significantly different in Lp (χ², P = 0.87) and significantly lower in A2 compared to A1 in Ab (χ², P < 2.2 × 10⁻¹⁶), in agreement with the view that A2 may also account for recombination events. More importantly, both A1 and A2 revealed higher rates of gene gains than A3 in both species (Lp: χ², P < 2.2 × 10⁻¹⁶; Ab: χ², P < 2.2 × 10⁻¹⁶). When comparing the different contexts (A1&A2 vs. A3 or B1 vs. B2), we found more gene gains in the presence of recent recombination tracts (Fig. 2c).

We then restricted our analysis to the gene gains associated with recombining persistent genes (A1, A2, and B1) and weighed them against all the gene gains that were inferred regardless of the recombination context. By doing so, and assuming that all these gains were caused by the identified recombination event, we got an upper limit estimate of the contribution of recombination to gene gains: 24.8% and 7.7% of all the gene gains in Ab and Lp, respectively. Again, this suggests more acquisition of novel genes by recombination in Ab than in Lp. Interestingly, this also indicates that the majority of gene gains in Ab and Lp are not acquired through homologous recombination, and it is likely that MGEs using site-specific integration are responsible for these acquisitions.

As mentioned above, incoming DNA recombining with the chromosome may have been acquired by transformation or by other processes such as conjugation or transduction. To disentangle the contribution of transformation, we analyzed the isolates showing a stable transformation phenotype along the terminal branch leading to them, i.e. that had a binary classification of transformability similar to that of their most recent parental node in the species phylogenetic tree. In case of the isolates that remained non-transformable, the observed recent gene gain caused by recombination, i.e. gains in A1, A2, and B1 contexts, should not be the result of natural transformation. It can thus be taken as an indication of the basal contribution of recombination by the other processes. The fraction of gene gains associated to recombination in these stably non-transformable isolates is 17.3% in Ab and 5.9% in Lp. This can be compared with the rates of gene gains associated with recombination in stably transformable isolates, which are 23.7% in Ab and 7.0% in Lp (supplementary fig. S2, Supplementary Material online). If one assumes that the other sources of DNA are unchanged for isolates capable of natural transformation, then one can estimate the share of natural transformation in the acquisition of novel genes in transformable isolates by making the difference with the average values for the isolates that remained non-transformable (see Discussion, supplementary fig. S2, Supplementary Material online, supplementary table S5, Supplementary Material online). This would suggest that gene gains due to natural transformation are 6.4% [bootstrap interval of confidence: (0.68; 12.5)] of the total in Ab and 1.0% [bootstrap interval of confidence: (−5.1;6.6)] in Lp, which would represent 25.8% (Ab) and 13.0% (Lp) of the novel genes acquired by recombination in these species.

Transformation Favors Fewer and Shorter Gene Acquisition Events

Gene gain by transformation is expected to be associated with events implicating relatively few genes, when compared with the impact of phages and conjugative elements. This is because the entry of the transforming DNA into the cell is accompanied by the action of endonucleases that create relatively small fragments (Provvedi et al. 2001; Nielsen et al. 2007). In contrast, viral particles transduce DNA fragments of the size of the phage genome (often around 50 kb for temperate phages), and conjugation can transfer entire chromosomes. To understand if transformation might be associated with the acquisition of small numbers of genes, we identified the individual events of gene gain and loss. We clustered into one event the arrays of genes gained in the same terminal branch and present between the two same persistent genes (A context) or between the same persistent gene and the end of a contig (B context), when they have few or no other gene in between them (see Materials and methods, Fig. 3a). The allowance of some gene families classed as not recently acquired is to account for the transposition of insertion sequences or gene families at high copy number. Yet, these interruptions remained exceptional since most events have none in Ab (85%) and in Lp (91%) (supplementary fig. S3, Supplementary Material online). We counted the number of acquisition events per genome and compared them between the 2 phenotypes (T and NT) while taking into account the phylogeny. In Ab, the number of acquisition events was significantly and positively associated with transformable isolates when using both the quantitative (supplementary table S3, Supplementary Material online: T11, P = 0.037) and the qualitative phenotype (supplementary table S3, Supplementary Material online: T5, P = 0.035) even though the transformation trait explained little of the variance (R² = 0.028). In Lp, no significant association was found between the transformation phenotype of an isolate and the number of acquisition events we observed in its genome (supplementary table S3, Supplementary Material online: T5, T11).

Fig. 3.

Characterization of the gene acquisition events by recombination. a) Definition of gene acquisition events between two recombining persistent genes. b) Distribution of the lengths of gene acquisition events per genome depending on the transformation phenotype in A. baumannii and L. pneumophila. c) Average proportion of gene acquisition events of different lengths depending on the transformation phenotype in A. baumannii and L. pneumophila and their association with transformability.

The acquisition events had between 1 and 107 genes in Ab and between 1 and 41 genes in Lp (Fig. 3b). The average length of acquisition events was longer in Ab compared to Lp (Ab: 4.4 genes; Lp: 2.8 genes). To detail these observations and explore the possible relation between event length and transformation phenotype, we defined three categories of acquisition events: short (≤5 genes), intermediate (5 to 10 genes), and long (≥10 genes), and compared their proportions among the events observed in each genome (Fig. 4c). Short events were positively associated with transformability in Ab and Lp (supplementary table S3, Supplementary Material online: T25, T27), even if for Lp this was only significant for the quantitative transformation phenotype and Ab for the qualitative phenotype (Fig. 3c). Expectedly, intermediate-size and long gene acquisition events followed the inverse trends. In Ab, both were associated with non-transformability, whether we used the qualitative or the quantitative transformation phenotype (Fig. 3c, supplementary table S3, Supplementary Material online: T26, T27, T29, and T30). In Lp, the intermediate-size events were also significantly associated with non-transformability using the quantitative phenotype (Fig. 3c, supplementary table S3, Supplementary Material online: T29), but we found no statistical signal for long events, possibly because these are quite rare and the statistical test may lack power. Hence, transformable isolates tend to have an excess of short events and a lack of long ones. These results are thus in agreement with the hypothesis that transformation favors events of acquisition of a small number of genes.

Fig. 4.

Characterization of hotspots of gene gains. a) Distribution of recombination rates in persistent genes flanking hotspots and coldspots depending on the transformation phenotype. b) Comparison of the proportion of intervals with at least one gene gain in hotspots. The data is stratified in terms of the evidence for recombination in neighboring persistent genes in terms of the transformation phenotype. c) Proportion of intervals with recently acquired prophage in Ab and with recently acquired ICE in Lp within hotspots. The data are stratified in terms of the evidence for recombination in neighboring persistent genes of the intervals considered and in terms of the transformation phenotype. Dots stand for the average proportions, and bars for the standard error. Statistical tests are Wilcoxon tests with ***: P < 0.001, **: P < 0.01.

Exchanges in Hotspots of Gains

Spots with many gene gains (hotspots) can result from the integration of MGEs by site-specific recombination or gene acquisition by homologous recombination. Genes acquired through homologous recombination, and potentially via natural transformation, might be responsible for a difference in the localization of gains between transformable and non-transformable strains. We separated transformable and non-transformable isolates and defined the sets of unique gene families that were gained in each spot. We defined hotspots as the minimal set of spots necessary to accumulate 90% of the gene families gained across all spots (see Materials and methods). The other spots were deemed cold spots. Hotspots represented only 3.48% of all the spots in Ab and 6.05% in Lp. They are scattered along the chromosomes of both species (supplementary fig. S4, Supplementary Material online). We then identified the hotspots common to the sets of transformable and non-transformable isolates in each species. Two-thirds of the hotspots were similar in the two sets of Ab, while only around 30% were similar in Lp (supplementary fig. S4, Supplementary Material online). The specificity of some hotspots for one phenotype could be explained by a small number of genomes having this spot and harboring gene gains, thus decreasing the likelihood of having it represented in both phenotypes. This is exactly what we observed: in both species, the hotspots specific to transformable and non-transformable isolates had, on average 3 times fewer genomes with gene gains than the hotspots common to both phenotypes. All the next results are qualitatively similar when performing the analyses with the hotspots defined on each group or on all genomes regardless of the transformation phenotype (Fig. 4 and supplementary fig. S5, Supplementary Material online). In both species and for both transformation phenotypes, hotspots of gains had a higher recombination rate at the flanking persistent genes than coldspots (Lp:Wilcoxon, P = 4.5 × 10⁻⁵; Ab: Wilcoxon, P = 3.9 × 10⁻⁷) (Fig. 4a). Within hotspots, we collected the intervals with gene gains and then separated them into two groups: those with and those without evidence of homologous recombination. The majority of intervals in hotspots with recent gains lacked flanking recombining genes (Fig. 4b) in both transformable and non-transformable isolates. Hence, large integration events are more frequently flanked with recombining persistent genes, but still most integration events in these loci are not associated with homologous recombination.

These results are consistent with the idea that most large events of gene acquisition are of MGEs encoding their own integrases. To confirm this hypothesis, we characterized the large MGEs in both species, especially integrative conjugative elements (ICEs) and prophages. Conjugative systems were more frequent in Lp (811 conjugative systems in 699 genomes out of 786) than in Ab (13 conjugative systems in 13 genomes out of 496) (supplementary fig. S6, Supplementary Material online). Prophages were absent from Lp, as previously described (Mazzamurro et al. 2024), and were frequent in Ab (1,215 elements), where 88% of the isolates carried at least one element (supplementary fig. S7, Supplementary Material online). On average, Ab had 2 to 3 (2.6) prophages, which is consistent with previous studies that also observed polylysogeny with 2 to 3 prophages per genome on average (Tenorio-Carnalla et al. 2024). As expected, intervals with gene gains which had no evidence of homologous recombination in flanking persistent genes were more often the sites of integration of large MGEs than the ones with homologous recombination at the flanking persistent genes (Fig. 4c). This was true for hotspots in transformable isolates (Lp:Wilcoxon, P = 0.015; Ab:Wilcoxon, P = 0.0062) and in non-transformable isolates (Lp:Wilcoxon, P = 0.014; Ab:Wilcoxon, P = 0.0038). This confirms the hypothesis that integration of the large MGEs tends to be regrouped in a few loci in the chromosome, the hotspots, by site-specific recombination rather than homologous recombination. As expected, these elements are not usually acquired by natural transformation.

HGT drives the acquisition of antimicrobial resistance (AMR) genes, raising the hypothesis that natural transformation is involved in this process (von Wintersdorff et al. 2016; Winter et al. 2021). We identified AMR genes in both species and estimated their proportion brought by recombination. We focused on the AMR genes, which were not persistent and thus susceptible to having been brought by any HGT process. In Ab, 1,515 accessory genes were marked as AMR. These correspond to 69 gene families of the pangenome and 13 different AMR classes. In Lp, we identified 1,781 genes corresponding to only 3 gene families and 2 AMR classes. We then searched for these genes among the events of gene acquisition by homologous recombination identified above, that is to say, in context A1, A2, and B1. In Lp, no AMR gene was recently acquired in the terminal branch, and the analysis couldn’t be done. In Ab, 9.6% of the AMR genes were gained in the terminal branch. 14% of them were found within the acquisition events we identified above and thus acquired by homologous recombination (supplementary table S6, Supplementary Material online). None of the acquisition events carrying these AMR genes were found in phages or integrative conjugative elements. They did not belong to plasmids either since we only considered acquisition events bordered by persistent genes, which would not be present on plasmids. Regarding the nature of these AMR genes, we verified whether they belonged to the 38 most frequent antimicrobial resistance genes (ARG) observed in A. baumannii (Hernández-González et al. 2022) that are often affected by HGT or homologous recombination. Among the 21 acquisition events, 4 events result in the gain of a gene belonging to the 38 most frequent ARGs: abaF (n = 1), adeC (n = 2) and amvA (n = 1). The others ARGs acquired by homologous recombination are bla (n = 1), fosA (n = 12) and ant(3″)-IIA (n = 4). fosA, responsible for the second mechanism of resistance against fosfomycin in Ab after abaF (Kyriakidis et al. 2021), was described as intermittently found and shown to be present in 2% of a collection of 1,915 A. baumannii genomes (Ito et al. 2017). According to the Comprehensive Antibiotic Resistance Database (CARD) catalogue, ant(3″)-IIA has been observed in 0.35% of NCBI chromosome, 0.05% of NCBI plasmid, and 1.5% NCBI Whole Genome Sequencing (WGS) of Ab (https://card.mcmaster.ca/ontology/41195). Around 86% of all the recently acquired AMR genes through homologous recombination were in events of gene gain with fewer than 5 genes. Some could hence result from transformation events. Many AMR in A. baumannii are known to accumulate in the AbaR island (Bi et al. 2019). Unfortunately, this locus cannot be analyzed with the current dataset, because the large number of transposable elements breaks contig assemblies complicating the analysis of the spots (context C in supplementary fig. S1, Supplementary Material online).

Discussion

Throughout this study, we aimed at quantifying the effect of natural transformation on the acquisition of novel genes. For this effect, we identified the imprint of recombination in bacterial genomes, inferred the recombination events resulting in gene gains, and then assigned these gains to transformation. All these tasks are challenging. Recombination is intrinsically difficult to measure because recombination tracts are difficult to pinpoint when exchanges take place between very similar sequences and because multiple overlapping events of recombination produce complex tracts. We used the information on recombination at flanking persistent genes to assess the likelihood that in between these persistent genes, accessory gene gains were caused by recombination. We assumed that all recent gene gains flanked by recombination tracts were caused by homologous recombination. This should be seen as an overestimate since recombination and gene gain may occasionally have occurred independently. This overestimate is expected to be larger when only one of the flanking persistent genes is identified as recombining, and indeed, cases A2 show a slightly smaller effect than A1 in Fig. 2c, even if both are significantly higher than cases A3. As datasets with completely assembled genomes become bigger, one will be able to better detail the chronologies of gene gains and recombination to confirm how frequently the latter results in the former. As mentioned above, gene gains by recombination may arise from other processes such as conjugation and transduction. Unfortunately, there is no specific hallmark allowing for distinguishing between these different sources of homologous recombination, and the concentration of accessory genes in hotspots of MGEs makes disentangling the different contributions particularly hard. Still, by comparing transformable and non-transformable bacteria, we were able to leverage the differences in recombination and gene gain to propose a first estimate of the impact of transformation in the acquisition of novel genes. This estimate is relatively low (6% in Ab and 1% in Lp), suggesting that transformation is responsible for the acquisition of a minor fraction of the genes in the genome.

The effect of natural transformation on the bacterial gene repertoires might be reflected in the genome size of bacteria. Despite the potential changes in gene repertoires caused by natural transformation, the gene repertoire sizes in Ab and Lp were similar between transformable and non-transformable isolates. Contrary to Aggregatibacter actinomycetemcomitans (Jorth and Whiteley 2012) and Streptococcus pneumoniae (Croucher et al. 2016) whose non-competent strains had smaller genomes than the competent ones, genome size was not much affected by bacterial transformability in our much larger datasets of Ab and Lp isolates. The lack of a strong effect of transformation on genome size could be explained by the balance between gene gain by natural transformation and MGE losses by chromosome curing. Transformable bacteria acquire some genes by transformation, but might effectively acquire fewer genes by MGE (because the latter are quickly deleted by transformation). In contrast, non-transformable bacteria will not get novel genes by transformation, but the increased stability of the acquired MGEs compensates for this absence. In both species, gene gains and gene losses were not exactly balanced, and transformable bacteria showed a net loss of genes in genomes. In isolates that recently became transformable, this net loss of genes could be related to the loss of transformation-inhibiting MGEs. This complex interplay between two sources of gene gain, transformation and MGE-mediated, might lead to a small net negative effect of transformation on gene turnover. But this net loss of genes was small enough not to significantly affect the genome size.

Gene acquisition events by homologous recombination differ between transformable and non-transformable isolates in number and in length. In Ab, transformable isolates had more acquisition events than the non-transformable ones. The difference was not significant for Lp. Ab also has a higher number of acquisition events than Lp. Because terminal branches are shorter in Lp phylogeny compared to Ab phylogeny, Lp cannot accumulate as many acquisition events as Ab. This could explain the lack of significance observed in Lp. In both species, however, transformable isolates were associated with short acquisition events while their non-transformable counterparts were associated with longer ones. Previous experimental studies in S. pneumoniae (Croucher et al. 2012), Helicobacter pylori (Lin et al. 2009), and Haemophilus influenzae (Maughan and Redfield 2009) showed that most recombination events observed during transformation are short. In a parallel evolution experiment in Bacillus subtilis, transferred segments between two B. subtilis lineages contained an average of 5.1 genes and had an exponential length distribution (Power et al. 2021). The overrepresentation of short-length fragments in transformation events may be due to the action of endonucleases during the DNA entry in the cell (Provvedi et al. 2001). After its uptake during replication, the transforming DNA, especially the heterologous region flanked by homologous sequences, can also be exposed to digestion by restriction endonucleases from restriction-modification systems (Johnston et al. 2013b). This overrepresentation of short-length fragments could be reinforced by the quality of DNA available to the transformable bacteria, since exogenous DNA may often be fragmented in the environment, thus providing short DNA fragments rather than very large ones. The experiment in B. subtilis also showed that the transferred segments were distributed over the entire core genome of the recipient (Power et al. 2021). In agreement with this observation, we found that long events of gene acquisition are concentrated in a few regions and related to the presence of large MGEs such as phages in Ab and ICEs in Lp, whereas gene gains by homologous recombination tend to be more scattered in the chromosome. Hence, transformation may result in more homogeneous recombination rates across the chromosome.

Despite the difficulty in disentangling the effect of transformation from all the other HGT processes that may act in recombination regions, we propose a first estimate of the contribution of natural transformation to the acquisition of novel genes. Our method relies on the assumption that the basal contribution of other mechanisms to recombination, e.g. conjugation and transduction, is the same in transformable and non-transformable isolates. One might think that this results in an under-estimate of the effect of transformation because transformable isolates have fewer MGEs. Yet, the MGEs driving conjugation and transduction are not in the recipient cell; they are in the donor. If incoming DNA comes from a random isolate of the species, then there is no effect of the recipient transformability on the probability of acquisition of DNA from the donor by these mechanisms. If the incoming DNA comes preferably from the same strains, then we may have underestimated the contribution of transformation since similar strains will tend to have fewer MGEs and thus less MGE-mediated transfer. Yet, similar strains have similar genomes, and gene gains by recombination are unlikely in such cases. Hence, the effect of chromosome curing of MGEs in recipient cells may have a very minor effect on our estimate of gene gains by transformation. Both species are known to recombine at significant rates (Snitkin et al. 2011; Sánchez-Busó et al. 2014; David et al. 2016), but the impact of recombination in gene gain seems higher in Ab. Future work will be needed to assess how this is affected by sampling strategies, which were very different for both species (Mazzamurro et al. 2024). Interestingly, among genes acquired by recombination, the estimated contribution of transformation is similar between the species. Our estimates suggest that transformation does play a significant role in the acquisition of novel genes, albeit smaller compared to MGE-mediated HGT processes.

The percentage of genes acquired by transformation does not suffice to evaluate the evolutionary importance of this process. Transformation may result in the acquisition of adaptive functions and will less often lead to the transfer of selfish DNA. The acquisition of a novel capsule operon by transformation in Griffith's experience, leading to a change in virulence, is a remarkable example of that (Johnston et al. 2013a). Similar serogroup conversions by transformation were observed in Vibrio cholerae (Blokesch and Schoolnik 2007). Other functions, such as enzymes involved in metabolism, were found in V. cholerae (Miller et al. 2007) and Campylobacter coli (Vorwerk et al. 2015). Our results suggest that almost 15% of the recently acquired AMR genes arrived by recombination independently of MGEs. This supports the idea that natural transformation plays a role in the dissemination of antimicrobial resistance (von Wintersdorff et al. 2016). Transfer of clinically relevant AMR genes by natural transformation was observed in Enterococcus faecalis (Lu et al. 2020), H. influenzae, Haemophilus parainfluenzae (Leidy et al. 1956), S. pneumoniae, Streptococcus oralis, Streptococcus mitis (Janoir et al. 1999), Neisseria meningitidis (Bowler et al. 1994), Acinetobacter baylyi (Sezmis et al. 2023), and A. baumannii (Traglia et al. 2019; Godeux et al. 2022; Perez and Stiefel 2022). These transfers may occur between species of the same genus; e.g. the large antibiotic resistance islands AbaR can be transferred by transformation across pathogenic Acinetobacter species with down to 90% sequence identity (Godeux et al. 2022). Despite the experimental data showing that transformation can spread AMR in the laboratory, evidence in clinical isolates has been missing. This is partly due to the difficulty in assigning the source of recombination-driven gene transfer to a specific HGT mechanism, including transformation. Further work is required to systematically disentangle the impact of the different mechanisms of transfer on the observed patterns of gene gain by recombination.

Materials and Methods

Collection of Acinetobacter baumannii and Legionella pneumophila Isolates

We studied a collection of 496 environmental (n = 375) and clinical (n = 121) isolates of Ab and another collection of 786 clinical isolates of Lp (supplementary table S1, Supplementary Material online) (Mazzamurro et al. 2024). Here, we provide a short summary of these methods, which can be found in our previous publication (Mazzamurro et al. 2024). We analyzed the genome assemblies to define the sequence types (STs) (Lp:n = 146; Ab:n = 164), pangenomes, phylogenies, and recombination tracts of these isolates in both collections. Draft assemblies were annotated with prokka (Seemann 2014) called from PanACoTA v.1.2.0 (Perrin and Rocha 2021). Both pangenomes were built using single-likage clustering so as to form pangenome families with PanACoTA. Pangenome families are sets of proteins sharing at least 80% identity. We defined a pangenome family as a persistent gene family when at least 95% of the genomes had a unique member of the family. Recombination-free phylogenies were built on the concatenated alignment of persistent genes ordered according to a reference genome with IQTree v.1.6.12 modelfinder (best-fit model: TVM + F + I + G4; 1,000 ultrafast bootstraps) and were rooted with an outgroup species. The recombination tracts were identified from the alignments of persistent genes using Gubbins v.2.4.1 (Croucher et al. 2015). While Gubbins is originally intended to analyze only closely related genomes, a recent study showed it was equivalent to other state-of-the-art approaches in identifying recombination, which don't scale to datasets of this size, when one focuses on recent events (as is the case in this work) (Mostowy et al. 2017). A discussion on these issues can be found in our previous publication (Mazzamurro et al. 2024). Transformation rates of the isolates were measured with a transformation luminescence assay, and the isolates were categorized as transformable or non-transformable based on the transformation rate of known non-transformable strains (supplementary table S1, Supplementary Material online). The transformation phenotype across the phylogenetic tree was previously reconstructed using PastML v.1.9.34 (Ishikawa et al. 2019), but this reconstruction was, here, only of interest in the terminal branches. We also thoroughly characterized the diverse mobile genetic elements present in the genomes, notably integrative conjugative elements (ICEs) reduced to conjugative systems present on contigs that were not of plasmid origin (Macsyfinder 20221213.dev; Abby et al. 2014; Neron et al. 2023), prophages (VirSorter v.2.2.3; Guo et al. 2021, CheckV v.0.7.0; Nayfach et al. 2021), and insertion sequences (ISEScan v.1.7.2.3; Xie and Tang 2017). As described by Mazzamurro et al. (2024), the contigs that were of plasmid origin were identified using their weighted gene repertoire relatedness (wGRR) with plasmids, allowing us to distinguish conjugative systems of ICEs from those of plasmids.

Identification of Antimicrobial Resistance Genes

Antimicrobial resistance genes were searched in the genomes of Lp and Ab with AMRFinderPlus v.3.12.8 (Feldgarden et al. 2021) (default parameters).

Ancestral Reconstruction of Pangenome Families Presence/Absence

We represented the pangenome as a binary matrix of the presence or absence of each gene family in each genome. We reconstructed the ancestral states of this pangenome in terms of presence–absence in each node of the recombination-free rooted phylogenetic tree. This was done using the Marginal Posterior Probabilities Approximation (MPPA) prediction method and the F81 evolutionary model of PastML v.1.9.34 (Ishikawa et al. 2019). Since the MPPA algorithm can keep more than one ancestral state per node if they have similar and high probabilities, we only kept the events where both ancestral and descendant nodes had one-single distinct state, either presence or absence. Using this method, on average, per node, more than 99% (Lp:99.8%; Ab:99.7%) of all the pangenome families had a confident presence/absence state in both species.

Measure of Gene Gains, Gene Losses and Gene Turnover in Terminal Branches

We focused on recent gains and losses of pangenome gene families, i.e. on events that were inferred to have occurred in the terminal branches of the species trees. A gene gain is defined as a gene family that is present in the leaf but inferred to be missing in the parental node. A gene loss is defined as a gene family that was inferred to be present in the parental node but missing in the leaf. We defined gene turnover of an isolate as the balance between gene gains and losses:

$$\mathrm{gene}\mathrm{turnover}={\displaystyle \frac{\mathrm{\#}\mathrm{gene}\mathrm{gains}-\mathrm{\#}\mathrm{gene}\mathrm{losses}}{\mathrm{\#}\mathrm{gene}\mathrm{gains}+\mathrm{\#}\mathrm{gene}\mathrm{losses}}}$$

This ratio provides a symmetric measure of gene turnover bounded between −1 and 1. The division by the sum of gains and losses allows for normalizing the gene turnover across branches that have different sizes and thus naturally different numbers of events. We observed a significant positive relationship between gene turnover and branch lengths (Ab: Estimate = 27.7, P < 2e−16, adjusted-R² = 0.156; Lp: Estimate = 379.5, P = 4.1 × 10⁻⁴; adjusted-R² = 0.015) and between absolute gene turnover, i.e. gene turnover not normalized by the number of genetic events, and branch lengths (Ab: Estimate = 9813.024, P < 2e−16, adjusted-R² = 0.1678; Lp: Estimate = 86,366.442, P < 2e−16; adjusted-R² = 0.1355). But since the distribution of terminal branch lengths is not significantly different between leaves with a transformable phenotype and leaves with a non-transformable phenotype (Ab: Wilcoxon, P > 0.05; Lp: Wilcoxon, P > 0.05), this correlation does not bias our comparison of gene turnover between transformable and non-transformable isolates. If the gene turnover is negative, then gene losses are more prominent than gains. If it is positive, then bacteria favor gene gains over losses. Close to 0, gene gains and losses are approximately equal in similar numbers.

Definition of a Spot

We use persistent genes to identify the location of gene acquisitions. Genomes are split into a collection of intervals, which are defined as a succession of pairs of persistent genes. Hence, each interval in each genome is a region comprised between two consecutive persistent genes within the same contig. Since persistent genes are present in most genomes, one can group intervals from different genomes based on these persistent genes. We define a spot as a collection of intervals (typically one per genome) that are flanked by the same two families of persistent genes. We only consider spots with intervals present in at least 50% of the genomes of the collection. In fact, in our collection of genomes, all spots included at least 50% of all genomes in Ab and in Lp, but the median of the proportion of genomes included in spots was much higher: 99.6% in Ab and 100% in Lp.

The propensity of bacterial genomes to undergo rearrangements could lead to the loss of a set of intervals (because the two originally flanking persistent genes are not consecutive in the list of persistent genes anymore). To quantify this problem, we analyzed RefSeq data on these species because it only includes completely assembled genomes and therefore we can be sure of the genome organization (downloaded in May 2023). We studied and defined spots on these complete genomes of Lp (n = 112) and Ab (n = 515). Since spots are a contiguous pair of persistent genes on the circular bacterial chromosome, we expect to have as many spots as we have persistent genes. In complete genomes, this is the case, which leads to 2,280 an 2,441 spots in Lp and Ab, respectively. When recreating the graph of these spots by linking together spots that shared a persistent gene, we observed a single circular graph. This indicates that all complete genomes shared the same organization of persistent genes and shows that genomes in each species are collinear.

However, since we are working with draft assemblies, another problem arises: some spots may be lost if there are many contig breaks at the same locations. We quantified these cases and found them to affect a small fraction of the spots. Given the number of persistent genes in our draft assemblies, which should equal the number of spots, we estimate to we have lost 17 spots in Ab, i.e. 0.6% of all spots, and 10 in Lp, i.e. 0.4% of all spots. This shows that our stringent approach still leads to the loss of little information.

Measure of the Recombination and Gene Gain Rates of a Spot

We consider the interval as recombining in an isolate when at least one of the two flanking persistent genes is part of a recombination tract. This simplification comes with limitations. In the case of the two flanking persistent genes being part of the recombination tract, it is likely that the genes comprised in the interval were also part of the recombination tract. However, when only 1 of the 2 flanking persistent genes belongs to a recombination tract, it is harder to precisely delimit the recombination tract. This could lead to overestimation of the number of gains obtained by recombination. An interval harboring one or several gene gains in an isolate was considered to have at least one acquisition event (see Identification of Gene Acquisition Events and Fig. 3). For every interval in each isolate, we consider two binary variables: presence/absence of an acquisition event (G) and presence/absence of a recombination event (R) in the flanking genes. We then summarized this information on all isolates that had the spot. We were thus able to calculate a recombination rate and a gene gain rate for every spot (Fig. 2a). We defined them as the number of times an interval of the spot was observed as having a recombination or acquisition event divided by the number of genomes present in this spot, respectively. The odds ratio of an interval with recombination evidence having an acquisition event was calculated as follows:

$$\mathrm{Odds}\mathrm{rati}{\mathrm{o}}_{\mathrm{gain}/\mathrm{nogain}}={\displaystyle \frac{I_{\mathrm{gain}+\mathrm{R}}/(I_{\mathrm{gain}}-I_{\mathrm{gain}+\mathrm{R}})}{I_{\mathrm{nogain}+\mathrm{R}}/(I_{\mathrm{nogain}}-I_{\mathrm{nogain}+\mathrm{R}})}}$$

with I_gain the total number of intervals with at least one acquisition event, I_{gain  +  R} the number of intervals with at least one acquisition event and with evidence of recombination, I_nogain the total number of intervals without any acquisition event, and I_{nogain  +  R} the number of intervals without any acquisition event but with evidence of recombination.

Of note: since our analysis of recombination and gene acquisitions is made only on terminal branches, each event is only counted once (no phylogenetic dependency).

Definition of the Recombination Context of any Accessory Gene

An accessory gene can be in 3 different types of genomic contexts regarding persistent genes (supplementary fig. S1, Supplementary Material online): in an interval delimited by two persistent genes (A), in a genomic segment delimited by only one persistent gene and the end of a contig (B) or in a contig lacking persistent genes and thus outside of any interval (C). Depending on the recombination state of the neighboring persistent genes, we can distinguish several subcontexts. In context A, the 2 flanking persistent genes are part of a recombination tract (A1), only 1 of the 2 flanking persistent genes is (A2), or none of them is (A3). We can do the same for context B: the neighboring persistent gene belongs to a recombination tract (B1) or not (B2). All the counts of accessory genes in each recombination context are to be found in supplementary table S4, Supplementary Material online. The odds ratios of an accessory gene being a gain in certain genomic contexts, shown in Fig. 2, were calculated as follows:

$$\mathrm{Odds}\mathrm{rati}{\mathrm{o}}_{i/j}={\displaystyle \frac{G_i/(T_i-G_i)}{G_j/(T_j-G_j)}}$$

With G_i and G_j being the number of gene gains in genomic contexts i and j, respectively, and T_i and T_j the total number of accessory genes in genomic contexts i and j, respectively.

Estimation of Transformation Contribution to Gains

Among all the accessory genes that were gained (G), the ones in context A1, A2, and B1 were considered to be acquired by recombination (G^REC). Indeed, the recombination tract covering the persistent gene in context A2 and B1 could extend further than the bordering persistent gene due to events of illegitimate recombination in the heterologous part of DNA and thus encompass these accessory genes. We focused on stable non-transformable isolates (NT > NT), i.e. isolates that remained non-transformable from the parent node to the leaf in the terminal branch. In these isolates, gene gains should not be the result of a transformation event. Hence, the proportion of gene gains acquired by recombination in stable NT isolates corresponds to the basal contribution of recombination to the acquisition of novel genes in the absence of transformation (presumably occurring by transduction or conjugation). One can then estimate the contribution of transformation to gene gain (G^{transformation}) by subtracting this basal contribution of recombination to the proportion of recombination-acquired gains in the transformable isolates that remained transformable from the parent node to the leaf in the terminal branch (supplementary fig. S2, Supplementary Material online):

$$G^{\mathrm{transformation}}=G_{\mathrm{T}>\mathrm{T}}^{\mathrm{REC}}-G_{\mathrm{NT}>\mathrm{NT}}^{\mathrm{REC}}$$

To obtain a confidence interval of the contribution of transformation to gene gain, we performed non-parametric bootstrapping. We considered the subpopulation of isolates with a stable phenotype (T > T and NT > NT): 395 isolates in Ab and 625 in Lp. To bootstrap, we picked 395 and 625 isolates randomly and with repetition in Ab and Lp subpopulations, respectively, and calculated the proportion of genes acquired via transformation in this new subpopulation. We then reiterated this procedure 1,000 times. We calculated the 95% percentile confidence interval of the distribution of the proportion of gains obtained through transformation of these 1,000 iterations. Distributions can be found in supplementary fig. S8, Supplementary Material online, and supplementary table S5, Supplementary Material online.

Identification of Gene Acquisition Events

The set of genes gained in an interval, i.e. between 2 consecutive persistent genes (context A1 and A2) or between one persistent gene and the end of a contig (context B1), may have been acquired in one or multiple events. To cluster gene gains in acquisition events, we had to choose a maximal number of genes separating two successive gene gains that belong to a single acquisition event. We made the choice of 3 genes based on several arguments. (i) By separating acquisition events by a number of accessory genes less than the average gap length (3.5 genes, supplementary fig. S9, Supplementary Material online) between two successive gene gains, we avoided underestimating the number of acquisition events by clustering too many gene gains together while they were coming from different acquisition events. (ii) The maximum number of genes separating two gene gains from the same acquisition event accounts for the disruption caused by insertion sequences (IS). Most IS (Lp: 99.8%, Ab:95.6%) that interrupts acquisition events are of a length less than or equal to 3 genes (supplementary fig. S10, Supplementary Material online). Transposable elements often have identical homologs in the genome, and in that case, they are not marked as gains. Our procedure allows us to avoid multiplying unduly the number of events because of these elements. In every interval with gene gains in the terminal branch, we then clustered successive gene gains separated by less than 3 accessory genes into single gene acquisition events (Fig. 3). In most intervals, only one acquisition event was observed (supplementary fig. S11, Supplementary Material online). The length of each acquisition event was measured as the number of genes (gains and genes in between) that are part of it. We classified the acquisition events in three different categories: short for when including less than 5 genes, intermediate for those having between 5 and 10 genes, and long for those with more than 10 genes.

Estimation of the Coincidence Between Recombination and Hotspots of Gene Gains

For each spot, we listed the set of unique pangenome families that were gained at least once across the genomes considered. We then ranked the spots from the biggest set size to the smallest. The bigger the set of a spots is, the higher the diversity of the acquired gene families is. We browsed the spots in the previously defined order, and at each spot, we counted the number of unique pangenome families that had been gained in this spot but not in the spot before. We then cumulated these numbers according to the previous spot ranking (supplementary fig. S12, Supplementary Material online). Spots with the biggest set size and whose cumulated set of unique gains constitute 90% of the total number of unique pangenome families gained across all spots were considered as hotspots of gene gains (supplementary fig. S12, Supplementary Material online). The gene content of these hotspots is more dynamic and encline to mobility as shown by the high diversity in gene gains they harbor. All the spots that were not categorized as hotspots were classified as coldspots. In each species, we applied this method by first considering all genomes and then each subcollection of transformable and non-transformable genomes. These hotspots were then analyzed in relation to recombination. We verified that working on all genomes or on each subcollection gave consistent results. We only present the results of each subcollection in the main results. The results obtained on all genomes can be found in the supplementary materials (supplementary fig. S5, Supplementary Material online).

Isolate Specificities Affected by Transformation Rates

To take into account the phylogenetic structure of the bacterial population in our statistical analyses of the impact of the transformation trait on other traits across the tree, we used phylogenetic linear models (phylolm function from phylolm R package, default parameters) (Tung Ho and Ané 2014). We expressed different traits such as the number of genes, the gene turnover, or the number of acquisition events as a function of the transformation phenotype. The latter was used either as a qualitative (TF^bin) or a quantitative (TF^log) phenotype. All the statistical tests were listed in supplementary table S3, Supplementary Material online. All graphs were produced with R v.4.1.0.

Supplementary Material

Supplementary material is available at Molecular Biology and Evolution online.

Data Availability

All major data, or identifiers to public repositories with the data, are incorporated into the article and its online Supplementary material. Other data, e.g. intermediary results files, underlying this article will be shared on reasonable request to the corresponding author. Custom scripts are available in figshare at https://doi.org/10.6084/m9.figshare.29136161.v1.