Bacteria have numerous distinctive groups of phage-plasmids with conserved phage and variable plasmid gene repertoires

Abstract

Plasmids and temperate phages are key contributors to bacterial evolution. They are usually regarded as very distinct. However, some elements, termed phage-plasmids, are known to be both plasmids and phages, e.g. P1, N15 or SSU5. The number, distribution, relatedness and characteristics of these phage-plasmids are poorly known. Here, we screened for these elements among ca. 2500 phages and 12000 plasmids and identified 780 phage-plasmids across very diverse bacterial phyla. We grouped 92% of them by similarity of gene repertoires to eight defined groups and 18 other broader communities of elements. The existence of these large groups suggests that phage-plasmids are ancient. Their gene repertoires are large, the average element is larger than an average phage or plasmid, and they include slightly more homologs to phages than to plasmids. We analyzed the pangenomes and the genetic organization of each group of phage-plasmids and found the key phage genes to be conserved and co-localized within distinct groups, whereas genes with homologs in plasmids are much more variable and include most accessory genes. Phage-plasmids are a sizeable fraction of the sequenced plasmids (∼7%) and phages (∼5%), and could have key roles in bridging the genetic divide between phages and other mobile genetic elements.

Figure 1.

Methods to screen databases for P–Ps. (A) 2383 phages were annotated using protein profiles specific of plasmids (conjugation, replication and partition), resulting in 122 putative P–Ps (Supplementary Table S4). (B) Carefully selected phage protein profiles were classified into distinct phage-specific functions such as structure components, packaging/maturation, lysis, etc. Their hits were used as features to train a machine learning method to distinguish plasmids lacking any kind of prophage from phages. We then used 10 random forest models to screen a plasmid database of 8901 plasmids yielding in 566 putative P–Ps (phage probability score (PSC) > 0.5) (Supplementary Table S4). (C) The screen for P–Ps was complemented by searching the literature for other potential P–Ps.

Figure 2.

Host and size distribution of P–Ps. (A) The frequency in bacterial genera of phages and plasmids (left) and P–Ps (center), for genera with at least three P–Ps (for the complete host distribution see Supplementary Table S4). The right panel shows the frequency of P–Ps per host genus normalized to the sizes of the databases of phages and plasmids. (B) Density plots (mean normalized counts) of replicon sizes from P–Ps (grey), phages (blue) and plasmids (orange). The number of phages and plasmids represent the number of all elements in the databases w/o the 780 P–Ps. The shaded boxes indicate the ranges of the first (Q1, 25%) and the third (Q3, 75%) quantiles representing 50% of the replicons.

Figure 3.

Sequence similarity network and detected communities. The communities are separated by gaps for better visibility. They were extracted, ordered in the figure by hierarchical clustering, and named after a representative P–P or a bacterial clade (in red). In the one-sided heatmap (below main diagonal), each row represents a P–P (n = 721). The 59 P–Ps not in communities were excluded (see Supplementary Figure S3). The range of the wGRR is given by the grey scale bar (from white to black). The first column on the left of the heatmap shows the phage score (PSC, given by the random forest models) and the second column indicates the database where the P–P was identified. The graph of the wGRR matrix is displayed on the right side of the heatmap. Communities that were curated into well-defined groups are shown above.

Figure 4.

Classification of P–Ps relative to phages and plasmids. (A and B) Distribution of P–Ps in terms of virus taxonomy (families) and of incompatibility types. NA: non-curated communities. (C) Boxplots of the genomic phage–plasmid quotients (gPPQs) for P–Ps (n = 677, grey) (Supplementary Table S4), phages (n = 458, blue) and plasmids (n = 1121, orange). A few P–Ps contained only a few genes homologous to phage or plasmid genomes. To increase the accuracy of the analysis, only elements with more than 10 genes with a PPQ were considered (see Materials and Methods). (D) Same as C for defined P–P groups (AB to N15) or communities (the rest) with at least 10 elements.

Figure 5.

Comparative genome analysis of the well-defined P1 and N15 groups. (A) Pangenome graphs of the N15 and P1 groups (the latter is split into two subgroups). The nodes represent genes of the persistent or shell genomes (see Supplementary Figure S8B and Supplementary Figure S9B for the entire pangenomes). The node colours indicate the phage–plasmid quotient (PPQ) scores (red for phage- and green for plasmid-association) that are computed from the average number of matches of the gene family with phage and plasmid genomes. The edges indicate contiguity between two genes in the P–P and their thickness indicates the frequency of this contiguity. For clarity, we removed the edges when the neighborhood was rare: for N15 < 25%, for P1 < 15%. (B) Comparisons between selected replicons plotted using genoplotR (93). Similarity between co-oriented bi-directional best hits (BBH) is shown in red and between anti-oriented ones in blue. Colour intensity reflects the degree of gene similarity. The values of wGRR are shown between the pairs of elements.

Figure 6.

Pangenome analysis of the AB group. (A) Pangenome graphs of the two AB subgroups. (B) Comparisons between selected replicons. For details, see legend of Figure 5.

Figure 7.

Conserved patterns in genomes of the SSU5 supergroup. (A) Pangenome graph of the SSU5 supergroup. (B) Comparisons between selected replicons. For details, see legend of Figure 5. The pCAV group was excluded from the analysis of the pangenome because it's not included in the SSU5 supergroup (see main text).

Figure 8.

Similarity analysis of the SSU5 supergroup and the less-related pCAV group. Pangenome graphs of the single SSU5_pHCM2, pKpn, pMT1, pSLy3 and pCAV groups and the entire SSU5 supergroup were colored in function of the values of PPQ (larger graphs) and similarity to the pangenome of the SSU5 supergroup (smaller graphs next to the arrows). Nodes and edges are as in Figure 5. The average number of homologs of a gene family with phage and/ or plasmid genomes is given in the PPQ graphs. Genes that are specific to one group are shown in blue in the SSU5 similarity graphs. Otherwise, genes and their orthologues (BBH) found in at least two P–P groups are indicated in orange/yellow/light yellow nodes (depending on their average identity) (see Methods). An example: The pMT1 pangenome (top left) is highly related to the one of the SSU5 supergroup (center), since the two similarity pangenome graphs next to the arrows show many similarities (colored in light yellow, orange to yellow). However, some co-localized gene families are only found in the pMT1 group (they are indicated in blue).