Toxic antiphage defense proteins inhibited by intragenic antitoxin proteins

Significance

Here, we document the function of small genes-within-genes, showing they encode antitoxin proteins that block the functions of the toxic DNA endonuclease proteins encoded by the longer rpn (Recombination-promoting nuclease) genes. Intriguingly, a sequence present in both long and short proteins shows extensive variation in the number of four amino acid repeats. Consistent with a strong selection for the variation, we provide evidence that the Rpn proteins represent a phage defense system.

Abstract

Recombination-promoting nuclease (Rpn) proteins are broadly distributed across bacterial phyla, yet their functions remain unclear. Here, we report that these proteins are toxin–antitoxin systems, comprised of genes-within-genes, that combat phage infection. We show the small, highly variable Rpn C-terminal domains (Rpn_S), which are translated separately from the full-length proteins (Rpn_L), directly block the activities of the toxic Rpn_L. The crystal structure of RpnA_S revealed a dimerization interface encompassing α helix that can have four amino acid repeats whose number varies widely among strains of the same species. Consistent with strong selection for the variation, we document that plasmid-encoded RpnP2_L protects Escherichia coli against certain phages. We propose that many more intragenic-encoded proteins that serve regulatory roles remain to be discovered in all organisms.

The annotation of bacterial genes generally assumes that each coding sequence directs the synthesis of one protein product. However, there are a few examples where two protein products are translated from the same gene, one product corresponding to the full open reading frame, and the second translated from an internal translation initiation start (iTIS) (1). The functions of most products of these genes-within-genes are not known, though for the few that have been characterized, the two products can have complementary or opposing functions. One example of a complementary function is provided by a Synechocystis type I-D CRISPR-Cas Cascade system, where the full-length Cas10d protein forms a complex with Cas11d, which is translated from an iTIS (2). The complex is required for specific DNA binding by the type I-D Cascade complex, and without Cas11d, the Cascade complex has little or no DNA binding activity. Recent experiments to examine transcriptome-wide ribosome binding (ribo-seq) in the presence of inhibitors that trap the ribosome on translation initiation sites suggested that there are more iTIS than initially considered (3, 4). The five rpn (recombination-promoting nuclease) genes of Escherichia coli each have an iTIS that could potentially direct the synthesis of small proteins corresponding to the variable C-terminal tail (Fig. 1A and SI Appendix, Fig. S1A).

Fig. 1.

Broadly-distributed rpn genes move by horizontal gene transfer. (A) Diagram of rpn genes adapted from ref. 5. Positions of catalytic residues for RpnA and internal promoter (P) and iTIS are indicated in red font. (B) Species distribution of rpn orthologs. Clades with more than 15 children are displayed. The pie charts represent the percentage of species with (red) and without (blue) a rpn ortholog. The size of the pie chart represents the number of species in that clade.

Rpn proteins, which are members of the diverse PD-(D/E)XK phosphodiesterase superfamily (6), have been proposed to be involved in horizontal gene transfer based on the report that the conserved N-terminal domain has homology to transposases (7). Previous studies showed that overexpression of E. coli Rpn proteins reduced cell viability in a recA⁻ background and induced the DNA damage (SOS) response in a recA⁺ host (5). Furthermore, the purified RpnA protein was shown to possess DNA endonuclease activity. These observations support the idea that Rpn proteins are DNA-mobilizing enzymes, but the physiological role of this DNA cleavage activity is unknown. Thus, we set out to characterize the long, full-length proteins (Rpn_L) as well as the small C-terminal proteins (Rpn_S), which we hypothesized are translated from the iTIS and could act in conjunction with the Rpn_L proteins.

Results

rpn Genes Move by Horizontal Gene Transfer but Do Not Encode Conventional Transposases.

Phylogenetic analysis revealed that members of the rpn orthologous gene cluster (COG5464) are widely distributed and are found in 34 bacterial and two archaeal phyla (Fig. 1B). Their distribution is strongly nonuniform indicating that rpn genes are prone to frequent horizontal gene transfer. Additionally, the number of rpn genes can vary between different strains of the same species. For example, no rpn genes were found in the genome of Bacillus thuringiensis serovar kurstaki str. T03a001 (assembly accession: GCF_000161575.1) while 26 copies are present in Bacillus thuringiensis serovar yunnanensis (assembly accession: GCF_002147825.1).

rpn genes are commonly annotated as encoding a “putative transposase” and genomic synteny among strains differing in rpn genes suggests that the rpn genes were recently acquired or lost as a single gene cluster, a pattern similar to insertion sequences (8). However, unlike typical insertion sequences, the coding regions of rpn genes are not flanked by identifiable inverted terminal repeats or target-site duplications. Furthermore, comparative genomic analysis of closely related genomes with and without rpn genes shows that the ends of the genomic region containing the rpn gene vary, contrasting with insertion sequences where the ends are generally defined. In genomes where rpn gene arrays were found, the arrays appear to be generated by tandem direct duplications of individual genomic regions rather than by transposition of different rpn genes into the same genomic locus given the absence of target-site duplications and the presence of different types of duplications (SI Appendix, Fig. S1B). Although the TnsA protein of the heteromeric Tn7 transposase has a PD-(D/E)XK catalytic domain, there are no known autonomous transposases with this protein fold (9, 10), and we find no evidence that rpn genes are capable of autonomous transposition. We therefore think that it is highly unlikely that Rpn proteins function as transposases.

Small Proteins Are Translated from Within rpn Genes.

We next wanted to determine the role of the iTIS (Fig. 2A) and ribosome profiling signal detected for all five E. coli rpn genes after treatment with the translation inhibitors Onc112 or retapamulin (3, 4) (Fig. 2 B, C, D, and E and SI Appendix, Fig. S2A). To see whether small proteins corresponding to the C-terminal tails are synthesized, we introduced a sequential peptide affinity (SPA) tag on the chromosome upstream of each rpn stop codon, permitting detection via immunoblot analysis. In cells grown to exponential (“exp”) or stationary (“stat”) phase, all five small proteins corresponding to the Rpn_S were observed at different levels (Fig. 2 F, G, and H and SI Appendix, Fig. S2B). Among the Rpn_L, only RpnB_L, RpnC_L, and RpnE_L were detected under the growth conditions examined (Fig. 2 G and H). The higher levels of the Rpn_S proteins are consistent with previous differential RNA-seq (dRNA-seq) data (11), which shows a transcription start site within all rpn genes for a short transcript expressed at higher levels than the full-length mRNA. The introduction of mutations in the ribosome binding site (RBS) and predicted iTIS with minimal change to the rpnA coding sequence (RpnA_{K240R+E241D+M244L}, hereafter denoted RpnA_L*) on the E. coli chromosome (Fig. 2A) abolished RpnA_S expression (Fig. 2F), confirming that the RpnA_S protein is translated from the predicted iTIS.

Fig. 2.

The rpn genes encode smaller proteins (Rpn_S). (A) Sequence of RpnA_S RBS, and mutations to eliminate rpnA_S ribosome binding and start codon. Browser images of ribosome profiling data for rpnA (B), rpnB (C), rpnC (D), rpnE-ypaA/rpnE_Ss (E). Ribosome density for an untreated control (gray) and cells treated with Onc112 (blue) (4) or retapamulin (red) (3) are shown. Immunoblot analysis of the levels of SPA-tagged RpnA_S (F) and RpnB_S, and RpnC_S (G) RpnE_S and YpaA/RpnE_Ss (H). E. coli MG1655 strains were grown to exponential (exp) and stationary (stat) phase in LB. The SPA tag was detected with monoclonal anti-FLAG M2-peroxidase (HRP) antibody. Ponceau S staining of membranes served as loading controls. (I) Diagram showing an example of insertion of a rpnC_Ss gene (light green) and a rpnE_Ss gene (light mustard) downstream of the rpnC and rpnE genes, respectively, among strains of E. coli.

We also wanted to determine whether an Rpn_S protein is expressed from rpn genes from other organisms and thus cloned an rpn gene from the Gram⁺ bacterium Clostridioides difficile into a plasmid that we introduced into E. coli. For this construct, we also observed expression of an Rpn_S protein, which again was eliminated by the introduction of mutations of the predicted RBS and iTIS (SI Appendix, Fig. S2 C and D). This observation suggests coexpression of Rpn_L and Rpn_S proteins is broadly conserved.

Varied Expression of Small Proteins.

In an analysis of the rpn genes present in diverse E. coli strains (SI Appendix, Fig. S3 A and B), we observed that while most of the genes are encoded on the chromosome, some are encoded on plasmids. Intriguingly, on both the chromosome and plasmids, the Rpn_S protein can be expressed from either within an rpn gene or as a separate gene, annotated as encoding proteins with the DUF4351 domain of unknown function (PF14261) in some species (Dataset S1). For example, we noted that a second copy of RpnE_S was encoded by the ypaA gene (denoted rpnE_Ss here) downstream of the rpnE gene. RpnE_Ss was detected upon SPA tagging indicating the protein is synthesized (Fig. 2 H and I). A second copy of rpnC_S is similarly found downstream of rpnC in some E. coli strains (Fig. 2I). Phylogenetic analysis, as demonstrated in SI Appendix, Fig. S3C, revealed that the downstream small protein homologs RpnC_Ss and RpnE_Ss cluster together with their respective upstream small protein homologs, RpnC_S and RpnE_S, rather than clustering with each other. This observation suggests that the origin of the downstream small proteins occurred independently, with RpnC_Ss originating from RpnC and RpnE_Ss originating from RpnE. The rpn_Ss gene also can be lost through the deletion of the encoding region or fusion of the rpn N-terminal domain and rpn_Ss encoding region.

For further analysis of Rpn protein function, we expressed both RpnA and RpnB proteins from a rhamnose-inducible promoter as done previously (here denoted as expressing RpnA_LS or RpnB_LS) (5) or from their own promoters on a low-copy plasmid. We also expressed plasmid-encoded RpnP2_LS from its own native promoter on a low-copy plasmid. To first compare the relative levels of the proteins expressed from these constructs, we again integrated an SPA tag upstream of the stop codon. Intriguingly, the relative levels of Rpn_L and Rpn_S varied widely with higher levels of Rpn_L than Rpn_S from the rhamnose-inducible promoter than the native promoter, except for RpnP2_L (SI Appendix, Fig. S2 E and F). The reasons for the differential expression are not known but suggest additional regulation.

Rpn_S Proteins Block Rpn_L-Dependent Growth Inhibition.

Although previous studies showed that Rpn proteins expressed from the rhamnose promoter reduce cell viability when overexpressed (5), Rpn_S likely was unknowingly coexpressed from these constructs, complicating the interpretation of these results. To deconvolute the effects of the two proteins, we constructed plasmids carrying rpnA or rpnB with mutated iTIS and RBS sequences (RpnA_L* or RpnB_L*) to fully abolish the expression of RpnA_S or RpnB_S. When assessed by cell density measurements in liquid culture, there was no effect on E. coli ER2170 growth (Fig. 3A) with the empty plasmid (black) or RpnA_LS overexpression (green), while a defect was observed when expressing RpnA_L* (red). For the RpnB constructs (Fig. 3B), we observed some growth defect for cells expressing RpnB_LS (green), but the effect was even stronger for overexpression of RpnB_L* alone (red). For RpnP2 (Fig. 3C), all constructs for RpnP2_L* expressed from its native promoter had additional inactivating site mutations indicating strong selection against expression of RpnP2_L alone. We were only able to obtain the intact RpnP2_L* construct when the rpnP2 gene was cloned behind the P_BAD promoter, which is repressed by glucose and activated by arabinose. In the presence of glucose, the RpnP2_L* cells showed a slight growth defect (top panel, red). In contrast, growth was strongly inhibited by RpnP2_L* upon the addition of arabinose (bottom panel, red), while RpnP2_LS cells (green), which also expressed the RpnP2_S protein, grew like vector control cells (black) under both conditions. Thus, Rpn proteins have the properties of toxin–antitoxin systems (12) with Rpn_L proteins inhibiting growth and Rpn_S proteins serving as the cognate antitoxin.

Fig. 3.

Rpn_S proteins function as antitoxins. RpnA_L (A) and RpnB_L (B) inhibit E. coli growth. E. coli ER2170 cells harboring indicated plasmids were grown in LB. (C) RpnP2_L inhibits E. coli growth. E. coli MG1655 cells harboring indicated plasmids were grown in LB with added glucose or arabinose. For A–C, three biological replicates are shown. (D) RpnA_S, but not RpnB_S, blocks RpnA_L induction, and (E) RpnB_S, but not RpnA_S, blocks RpnB_L induction of the dinD-lacZ reporter of the SOS DNA damage response. RpnA_LS and RpnB_LS are overexpressed from the rhamnose-inducible P_rhaB promoter. RpnA_S and RpnB_S are overexpressed from the arabinose-inducible P_BAD promoter. For D and E, average of three independent biological repeats is given with SD. For A–E, the ER2170 or MG1655 strain backgrounds were WT for the rpn genes. (F) Immunoblot analysis of intein-tagged RpnA_LS, RpnA_L*, and RpnA_S overexpression. Ponceau S staining of membrane served as loading control. (G) Coomassie-stained Tris-glycine SDS-PAGE gel of purified RpnA_LS, RpnA_L*, and RpnA_S. (H) Purified RpnA_S blocks dsDNA endonuclease activity of RpnA_L*. Indicated purified proteins were incubated with pUC19. (I) RpnA_L activity is pH-dependent. Purified RpnA_L* (lanes 1 to 3) or RpnA_LS (lanes 4 to 6) were incubated with pUC19 at pH 7.0, pH 8.0, or pH 9.0. Images in H and I are inverted from the original. Each of the nuclease assays was repeated at least twice.

Rpn_S Proteins Block Rpn_L-Dependent SOS Induction.

To test whether the Rpn_S proteins can block the DNA damage caused by Rpn_LS overexpression reported previously (5), we employed a two-plasmid system to separately overexpress Rpn_S. We observed that, when the proteins are overexpressed from the rhamnose-inducible P_rhaB promoter, RpnA_LS and RpnB_LS both induce the SOS response (SI Appendix, Fig. S4A), as monitored by the chromosomally encoded, DNA damage-inducible P_dinD-lacZ reporter (5). For these experiments, induction of RpnA_LS and RpnB_LS was titrated to allow for overexpression without significantly impacting growth. Coexpression of RpnA_S or RpnB_S reduced SOS induction in the corresponding Rpn_LS strain (Fig. 3 D and E). This block is no longer observed when a stop codon was introduced in the RpnA_S (at E25) or RpnB_S (at D29) coding sequence indicating that the repressive effect is due to the protein and not due to overexpression of the RNA. The repressive effect of each small protein is specific to the Rpn_L protein encoded by the same sequence as overexpression of RpnA_S did not block RpnB_LS-dependent dinD-lacZ induction while RpnB_S did not block RpnA_LS-dependent induction. These observations further support the conclusion that RpnA_S and RpnB_S specifically block the activities of the corresponding larger protein.

RpnA_S Blocks RpnA_L-Dependent DNA Cleavage In Vitro.

To directly examine the effect of the Rpn_L and Rpn_S proteins on DNA cleavage, RpnA derivatives, which were easier to express given that they were the least toxic, were overexpressed as C-terminally tagged intein fusion proteins and purified. When the WT rpnA gene was cloned into an overexpression vector, the C-terminally tagged RpnA_L and RpnA_S (61 kDa and 33 kDa, respectively) were both detected (Fig. 3F) and purified together (Fig. 3G). Thus, we also overexpressed and purified the RpnA_L* and RpnA_S proteins separately (Fig. 3 F and G). In assays for DNA cleavage activity using both double-stranded (Fig. 3H) and single-stranded (SI Appendix, Fig. S4B) DNA substrates, the RpnA_L* protein had strong DNA cleavage activity (lane 2). Consistent with the conclusion that RpnA_S blocks RpnA_L activity, we found that RpnA_LS has significantly less endonuclease activity (lane 3), and the addition of RpnA_S to RpnA_L* completely blocks the cleavage (lanes 5 and 6). Interestingly, the DNA cleavage activities of both RpnA_L* and RpnA_LS are higher at a more alkaline pH (Fig. 3I) as was previously observed for RpnA_LS (5), suggestive of a role for the cleavage activity at higher pH or other specific growth condition with similar consequences.

RpnA_S Forms a Complex with RpnA_L.

To examine whether RpnA_S is inhibiting RpnA_L through a direct interaction, the oligomerization states of the purified proteins were assessed by size exclusion chromatography (SEC) (Fig. 4A). The individually purified RpnA_L* (33.3 kDa) and RpnA_S (5.4 kDa) proteins eluted as distinct peaks. However, when RpnA_L* and RpnA_S were mixed, a stable RpnA_L*–RpnA_S complex was observed that eluted at the same volume (arrow) as the native RpnA_LS complex. SDS/PAGE analysis of the peak fractions confirmed that both RpnA_L* and RpnA_S are in the corresponding fractions (SI Appendix, Fig. S5A). In a multiangle light scattering coupled with SEC (SEC-MALS) experiment, the molecular weight of RpnA_S was determined to be 10.2 ± 0.2 kDa, consistent with a dimer. The molecular weight of the RpnA_LS complex was determined to be 74.1 ± 0.5 kDa by SEC-MALS, consistent with a tetramer of two RpnA_S subunits and two RpnA_L subunits. Intriguingly, the RpnA_LS complex is stable at pH 7.0 and 8.0, yet the subunits dissociate at pH 9.0 and 10.0 (SI Appendix, Fig. S5B), paralleling the increase in DNA cleavage activity observed for RpnA_LS (Fig. 3I).

Fig. 4.

RpnA_L and RpnA_S form a complex. (A) Size exclusion chromatograms of indicated proteins on a Superdex 200 column. Although the RpnA_LS peak migrates similarly to the 158 kDa marker peak, SEC-MALS analysis, which is less influenced by protein shape, indicates that the complex has a molecular weight of 74.1 ± 0.5 kDa. The RpnA_L* protein migrates as a large molecular weight complex indicative of an aggregate, but we do not observe precipitates of this protein. (B) Sequence alignments for E. coli RpnA_S and RpnC_S as well as Clostridioides difficile Rpn_S. The sequence logo was built based on all the sequences for each group though only representative individual sequences are shown. The four amino acid repeat that varies between strains is colored in the alignment, with a bracket indicating the number of repeats. The alignments were manually curated. (C) Crystal structure of RpnA_S. Pink asterisks represents approximate position where four amino acid insertions occur (H261 in RpnA_L). (D) Only RpnC_S with same number of four amino acid repeats blocks RpnC_L induction of dinD-lacZ. RpnC_LS and RpnC_LS+REPEAT are expressed from the rhamnose-inducible P_rhaB promoter, while RpnC_S and RpnC_S+REPEAT are expressed from the arabinose-inducible P_BAD promoter. For the REPEAT derivatives, a four amino acid repeat “KGIE” is inserted in the same position. Average of four independent biological repeats is given with SD.

Rpn_S Vary by Four Amino Acid Repeats and Comprise an Oligomerization Domain.

A striking feature of the amino acid sequence present in both the Rpn_L and Rpn_S proteins, which is observed from the alignment of homologs from different strains of the same species, is variation by four amino acid repeats generally composed of one hydrophobic amino acid and three charged residues or glycine (Fig. 4B and SI Appendix, Fig. S6). This is seen for all the E. coli Rpn_S proteins including Rpn_Ss proteins as well as those found in distantly related bacteria such as Bacillus cereus, Bacteroides fragilis, Clostridioides difficile, and Leptospira interrogans.

Given the interaction between RpnA_L and RpnA_S, we hypothesized that this hypervariable four amino acid repeat region present in both forms of the Rpn proteins might comprise an oligomerization domain. To gain insights into the domain, we solved the structure of RpnA_S at a resolution of 1.9 Å using X-ray crystallography (Fig. 4C and SI Appendix, Table S1). Consistent with the SEC-MALS data, RpnA_S is a dimer in the crystals in which the two monomers each fold as a compact three α-helix bundle. The long first helix of each monomer interacts with the other, burying a total surface area of 976 Ų. In all Rpn_S proteins where they have been inserted, the four amino acid repeats begin 8 to 20 residues after the initiating methionine (SI Appendix, Fig. S6), which places the insertion point for the hypervariable region within the first α-helix of Rpn_S. Based on three-dimensional structure predictions by AlphaFold2 (13), the inserted repeats most likely increase the length of the α-helix rather than disrupt it (SI Appendix, Fig. S7).

To examine the consequences of changing the number of four amino acid repeats, we generated derivatives of the RpnC_LS and RpnC_S proteins, which have the highest number of repeats among E. coli strains, by adding one repeat. Interestingly, when assessed in the SOS response assay, the RpnC_S+REPEAT can no longer counteract RpnC_LS (Fig. 4D). Conversely, RpnC_S cannot counteract RpnC_LS+REPEAT, but repression is restored when the two proteins with the extra repeat are expressed together. These observations are consistent with the repeat regions affecting the binding between the two Rpn_S monomers and providing specificity for the interaction between the Rpn_L and Rpn_S proteins.

Rpn Proteins Contribute to Antiphage Defense.

The rpn genes share characteristics with known antiphage defense systems such as restriction–modification enzymes and toxin–antitoxin complexes, including the ability to quickly diversify and extensive horizontal mobility. The genes also show a tendency to cluster with other defense genes (Fig. 5A), including systems with dCas12a or Type I-E Cascade (14). Like most antiphage defense systems (15, 16), the rpn genes are not autonomously mobile but likely rely on other mechanisms such as recombination or “hitchhiking” with mobile elements to move. Given these similarities and the rapid change in the numbers of four amino acid repeats among Rpn_S proteins, which we surmised must be due to strong selective pressure acting on the rpn genes to diversify the C-terminal domain, we hypothesized that Rpn proteins might defend against phage infection.

Fig. 5.

Rpn proteins block phage infection. (A) Enrichment of phage defense genes in the vicinity of rpn genes compared to random genes. Box plots (Left) and dot plots (Right) display the percentage and distribution of defense-associated genes in proximity to these genes, with P-values indicating the significance levels determined by the Mann–Whitney U Test. (B) Efficiency of plaquing (EOP) for the indicated phages infecting cells producing RpnP2_LS grown at 30 °C on LB medium buffered to pH 8.5 and 18 °C on M9 glucose medium pH 7.1. (C) EOP for the indicated phages infecting cells producing RpnP2_LS grown in LB and M9 glucose, pH 8.5. For B and C, “s” that denotes smaller plaques were observed. (D) Images of BASEL phage #40 and #39 plaques for MG1655 cells carrying vector control or producing RpnP2_LS (Left) or ECOR13 or ECOR13 ∆rpnP2_LS cells (Right), from SI Appendix, Fig. S8. The back wedge denotes the 10-fold dilution series. (E) Growth of vector control or RpnP2_LS expressing cells infected with BASEL phage #40 (at indicated MOIs) in LB pH 8.5. Three biological replicates are presented. The partial regrowth of the strains infected at a MOI of 10 could be due to appearance of suppressors. For A–E, the MG1655 strain background was WT for the rpn genes. (F) Plaque-forming units (PFU) per mL of BASEL phage #40 (MOI = 0.005) used to infect vector control or RpnP2_LS expressing cells at 0, 10, 20, 30, 40, and 60 min postinfection. Individual data points and average of two biological replicates are shown. (G) Model for functions of Rpn_L and Rpn_S proteins. RpnA_L dimer (orange and red, Right side) and RpnA_LS tetramer (Left side, toxin-antitoxin complex) structures were predicted with AlphaFold-Multimer.

To test whether Rpn proteins constitute an antiphage defense, we examined the ability of various phages to form plaques on E. coli MG1655 strains expressing plasmid-derived RpnP2_LS from its native promoter on a low-copy plasmid (Fig. 5 B and C and SI Appendix, Fig. S8). Considering the high levels of the RpnP2_S protein for cells grown in LB (Luria broth) pH 7 (SI Appendix, Fig. S2F) could inhibit RpnP2_L, we wanted to assay cells grown under conditions where the RpnP2_S–RpnP2_L interaction might change. Given that RpnA_L* has higher activity at high pH and RpnA_LS complex tends to dissociate at high pH, we first grew cells in LB pH 8.5. Under these conditions, cells expressing RpnP2_LS showed 10-fold reduced plaquing by phage T2. We observed smaller T2 plaques and 10-fold reduced plaquing by T4 for RpnP2_LS expressing cells grown in minimal glucose medium pH 7.1 at low temperature (Fig. 5B and SI Appendix, Fig. S8B). We also examined the ability of other T-even phages in the BASEL phage collection (17) to plaque on this strain grown in both LB and minimal glucose media pH 8.5 (Fig. 5C and SI Appendix, Fig. S8 C and D). RpnP2_LS protected against many of the T-even phages though resistance against BASEL phage #40 was strongest (Fig. 5D and SI Appendix, Fig. S8D). This protection was mostly eliminated by active site mutations. Additionally, we deleted rpnP2 in the ECOR13 strain. This deletion strain showed reduced protection against BASEL phage #40 compared to the ECOR13 wt strain (Fig. 5D and SI Appendix, Fig. S8E).

To further test for a phage defense function, we challenged RpnP2_LS expressing cells with BASEL #40 in liquid media at multiplicity of infections (MOIs) of either 10 or 0.005. Although both the vector control and RpnP2_LS cells were equally affected at high MOI, RpnP2_LS protects the cells at low MOI (Fig. 5E). A one-step growth curve also showed RpnP2_LS reduced the BASEL #40 burst size following a single round of infection (Fig. 5F). Under these BASEL #40 infection conditions, the levels of the RpnP2_S protein are reduced slightly at later time points (SI Appendix, Fig. S9). Together, our data suggest Rpn_LS toxin–antitoxin systems can provide protection against specific phage.

Discussion

Our data revealed that Rpn_S proteins are expressed together with Rpn_L proteins in E. coli as well as in C. difficile and serve as antitoxins for the toxic Rpn_L proteins (Fig. 5G). It is possible that other toxin–antitoxin systems are composed of additional components translated from iTIS as, for example, a clear intragenic ribosome profiling signal can be detected for prlF of the yhaV-prlF toxin–antitoxin system of E. coli (3, 4). Based on ribosome profiling data, genes-within-genes likely are far more prevalent than previously appreciated and not restricted to toxin–antitoxin genes.

Although the physiological roles of many toxin–antitoxin systems remain under debate, increasing numbers have been shown to constitute antiphage defenses (12, 18–24). Our bioinformatics data showing rpn gene enrichment in defense islands as well as the variation in the four amino acid repeats in Rpn_S across different bacteria species were the initial hints that Rpn proteins also might protect against phage infection. Indeed, we demonstrated that a representative system, plasmid-encoded RpnP2_LS, reduced plaque formation and infection by specific phages (Fig. 5 B–F). However, it is possible that some Rpn proteins have housekeeping roles in addition to antiphage defense.

The intriguing variation in four amino acid repeats in Rpn_L and Rpn_S proteins is observed across a wide range of bacterial species. Changes in four amino acid repeats have been observed in the context of the EcoR124I and EcoR124II endonucleases (25). EcoR124I, with two repeats of the sequence “TAEL”, is specific for a sequence with a 6 bp gap (GAAN₆RTCG), while EcoR124II, with three TAEL repeats, is specific for a sequence with a 7 bp gap (GAAN₇RTCG). Our structural studies showed that RpnA_S is composed of a three α-helix bundle. Additionally, HHpred (26) predicts relatedness to helix–turn–helix (HTH) motifs. In this context, it is noteworthy that many antitoxins have HTH motifs and adopt a strongly helical structure. In some examples, the toxin binding site overlaps these motifs (27, 28). The affinity between Rpn_S and Rpn_L is clearly tuned by the number of four amino acid repeats located in the long α-helix (Fig. 4D).

The overlapping rpn_L-rpn_S gene arrangement means there is obligatory coevolution of the two proteins. The mechanism of the expansion and possibly contraction of these dodecamer repeats warrant further study. Given the extensive variation within a species, we suspect the changes ultimately are driven by a phage factor. It is possible that the Rpn_L proteins themselves are involved in increasing the number of four amino acid repeats as well as in initiating the recombination events that allow the generation of the small protein homologs encoded downstream of some homologs and in generating the duplicated rpn genes.

How Rpn_L activity is released, a key unresolved issue for other toxin-antitoxin systems as well, is an important direction for future work. The dissociation of RpnA_L and RpnA_S and increased activity of RpnA_L at high pH led us to wonder whether Rpn proteins are activated at high pH or another specific physiological condition with similar effects. We found that RpnP2_LS provided more protection against different phages at alkaline pH. Since RpnA_S is stable at high pH, it is likely Rpn_S is released rather than degraded. In this context, we were intrigued to observe very different Rpn_L:Rpn_S ratios depending on how these genes were expressed suggesting regulated expression is another factor in modulating Rpn_L activity. Other important questions are whether Rpn_L proteins have preferred substrates, whether the Rpn_L nucleases specifically recognize and cleave phage DNA, and what domains comprise the DNA-recognition determinants of the proteins. We expect that further mechanistic understanding of the actions of the Rpn_L and Rpn_S proteins will allow the exploitation of these unique toxin–antitoxin systems.

Materials and Methods

Strains, Plasmids, and Oligonucleotides.

Strains, plasmids, and oligonucleotides used in this study are listed in Dataset S2. Details about strain and plasmid construction are provided in SI Appendix.

Bacterial Growth.

Bacterial growth in LB-rich medium or M9 minimal medium was carried out at indicated pH, indicated temperature, and indicated supplements as described in detail in SI Appendix.

Immunoblot Analysis.

Specific tagged proteins were detected by immunoblot analysis as described in detail in SI Appendix.

β-Galactosidase Activity Assays.

Induction of a dinD-lacZ reporter was assayed as described in detail in SI Appendix.

Biochemical Characterization.

RpnA protein purification and characterization by in vitro DNA cleavage assays, SEC analysis, and SEC-MALS analysis were carried out as described in detail in SI Appendix.

Structure Determination and Prediction.

The structure of RpnA_S was determined as described in detail in SI Appendix. All Rpn proteins structures were predicted with AlphaFold 2.2.0 (13, 29) using NIH's Biowulf cluster.

Bioinformatic Analysis.

Species distribution, phylogenetic analysis, and gene context analysis were carried out as described in detail in SI Appendix.

Bacteriophage Assays.

Plaque assays and efficiency of plaquing measurements, growth curves following phage infection, and one-step growth curves to measure burst size were carried out as described in detail in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Data, Materials, and Software Availability

The data are available in the manuscript, or protein structure data have been deposited in the Protein Data Bank (www.wwpdb.org) (PDB 7TH0) (30).