Restriction-modification systems have shaped the evolution and distribution of plasmids across bacteria

Abstract

Many novel traits such as antibiotic resistance are spread by plasmids between species. Yet plasmids have different host ranges. Restriction-modification systems (R-M systems) are by far the most abundant bacterial defense system and therefore represent one of the key barriers to plasmid spread. However, their effect on plasmid evolution and host range has been neglected. Here we analyse the avoidance of targets of the most abundant R-M systems (Type II) for complete genomes and plasmids across bacterial diversity. For the most common target length (6 bp) we show that target avoidance is strongly correlated with the taxonomic distribution of R-M systems and is greater in plasmid genes than core genes. We find stronger avoidance of R-M targets in plasmids which are smaller and have a broader host range. Our results suggest two different evolutionary strategies for plasmids: small plasmids primarily adapt to R-M systems by tuning their sequence composition, and large plasmids primarily adapt through the carriage of additional genes protecting from restriction. Our work provides systematic evidence that R-M systems are important barriers to plasmid transfer and have left their mark on plasmids over long evolutionary time.

Graphical Abstract

Figure 1.

Avoidance of short palindromes (k = 6) is stronger but more variable in plasmids. (A) Significantly greater avoidance of 6-bp palindromes in plasmid genes compared to core and non-core chromosomal genes (P < 0.001, two-sided Wilcoxon paired test). (B) Mean avoidance is strongly structured by species, with a strong correlation between avoidance in core and plasmid genes (Spearman's ρ = 0.55, P < 0.001). (C) Relative palindrome avoidance for species for core vs. plasmid genes (>0 denotes greater avoidance in plasmid genes). Points are mean, error bars show standard error. The modelled effect was computed using a phylogenetically-controlled GLMM (see Materials and Methods). Data shown are mean avoidance scores of 6-bp palindromes (4³= 64) calculated with R’MES after pangenome construction then subsampling each per-isolate pangenome compoment to 50kbp i.e. only genomes with at least 50kbp are included (3912 isolate genomes across 44 species). The inset panel shows within-species variation in mean palindrome avoidance score for each pangenome component. Only species with at least 3 genomes meeting these criteria are shown. For 4-bp palindromes, there was no significant difference between plasmid and core genes (Supplementary Figure S2 within Supplementary Data 1) and mean avoidance was uncorrelated with 6-bp palindrome avoidance (Spearman's ρ = 0.005, Supplementary Figure S3). Notably, in a rare previous stud, Wilkins et al. (44) found that 4-bp palindromes were not strongly avoided in the IncP-1 backbone and suggested that R-M systems with 6-bp targets were a stronger selective pressure, in line with our findings here.

Figure 2.

The taxonomic distribution of R-M systems correlates with avoidance of their targets. (A–C) Methodological approach to connect Type II R-M system distribution to target avoidance: (A) We search for Type II R-M systems in n = 8552 genomes from 72 species, detecting complete systems with confident prediction of targets them in 2740 genomes (Table 1). From these hits, we created a taxonomic hierarchy of their targets across a set of species. (B) We construct a pangenome for each species in our dataset, then separate each individual isolate into genes in three pangenome components: core, non-core and plasmid. (C) We subsample pangenome components to a fixed size and use R’MES to calculate exceptionality scores for fixed-length k-mers for k= 4, 5, 6 for each species, using the taxonomic hierarchy of R-M targets to correlate exceptionality scores with R-M distribution. (D, E) Exceptionality scores for 6-mers by pangenome component as a function of the taxonomic hierarchy of R-M targets: (D) averaged over all species and (E) for individual species. Subsampling is to 50kbp for each within-isolate pangenome component. Other subsampling lengths show the same pattern (see github repository).

Figure 3.

Larger plasmids have a higher density of the targets of within-species R-M systems. (A, B) Results for the best-sampled species in our genomic dataset, Escherichia coli, for the mean density of within-species R-M targets of length (A) k= 5 (4 targets) and (B) k= 6 (33 targets). Each point is the mean density of targets within a single plasmid (no deduplication), black lines show median for each category. (C–E) Results for at a per-species level for different values of k. Species without R-M systems with targets of length k are omitted. Each point represents the median of the mean densities of within-species R-M targets for plasmids in that species, including only size/species combinations with >5 plasmids. Dashed lines shows the expected) density of a random k-mer in a random sequence (4^−k). Comparisons between the largest (>100 kb) and smallest (<10 kb) plasmid categories are significant (P < 0.05) for k= 5 and 6 but not for k= 4.

Figure 4.

PTU host range is associated with greater avoidance of 6-bp palindromes. Avoidance of 6-bp palindromes in PTUs >10 kb correlates with PTU host range (excluding unassigned plasmids, Spearman's ρ = –0.26, P = 0.003). Each point is one PTU (mean exceptionality score) apart from unassigned plasmids (those not classified into a PTU) and lines show median within host ranges. There is no correlation for avoidance of 4-bp palindromes (Supplementary Figure S11).

Figure 5.

Small and broad host range PTUs have stronger avoidance of R-M targets. (A–C) Coefficients in linear models (mean estimates with standard error shown by errorbars) for the exceptionality score of R-M targets. A different model was run for each possible of level of R-M targets within the taxonomic hierarchy, from R-M targets of R-M systems within-species to within-kingdom, with three variables for each PTU: host range, median length, and number of plasmids. (A) PTU host range, converted to a numeric variable for modelling where larger values denote broader host range, is negatively associated with exceptionality score of R-M targets, i.e. broader host range PTUs have stronger avoidance. (B) Median length of plasmids within PTU (log₁₀ for modelling) is positively associated with exceptionality score of R-M targets, i.e. larger plasmids have weaker avoidance. (C) Number of plasmids within the PTU has no significant effect. (D) Total variance explained by each model, with colours denoting the three different variables (red: host range, blue: length, green: number of plasmids).

Figure 6.

Large plasmids with a broad host range are more likely to carry MTases. Numbers show the number of plasmids in that category with at least one MTase out of the total number of plasmids.