To compete against other strains, P. mirabilis strain BB2000 requires both the idrD gene and the T6SS, suggesting that the idrD-encoded protein functions as a T6SS-associated effector (Saak et al., 2017). The idrD gene contains an Rhs region within its N-terminus. Many Rhs-containing effectors often contain an enzymatic domain in the C-terminus (Koskiniemi et al., 2013; Ma et al., 2017). As a result, we investigated the function of the final 138 amino acids at IdrD’s C-terminus, now renamed ‘RdnE’ for recognition DNA nuclease effector. We measured bacterial growth using a strain derived from BB2000 that has disruptions in its native idrD and downstream genes (Wenren et al., 2013). This P. mirabilis culture had 1000 fewer cells per mL when engineered to overproduce RdnE in trans than the negative control containing the parent empty vector (Figure 1A). An equivalent growth pattern occurred in Escherichia coli cells under the same conditions (Figure 1—figure supplement 1). Thus, RdnE was lethal in vivo.

RdnE’s initial 86 amino acids contain a PD-(D/E)XK motif, which is suggestive of nucleotide degradation. The PD-(D/E)XK superfamily includes proteins with broad functions, including effectors that degrade DNA or RNA (Jana et al., 2019; Kosinski et al., 2005; Yadav et al., 2021). Three residues in the catalytic site—D, D/E, and K—are required for activity (Steczkiewicz et al., 2012). Therefore, we changed the corresponding residues in RdnE (D39, E53, and K55) to alanine, separately and together. P. mirabilis producing these mutant proteins showed growth equivalent to the negative control lacking RdnE (Figure 1A). We also saw that E. coli cells that were producing RdnE had morphologies that were indicative of DNA damage or stress, consistent with an SOS response (Friedberg et al., 2005; Kreuzer, 2013). These cells were elongated, and the DAPI-stained DNA was distributed irregularly within the cells (Figure 1—figure supplement 1). Cells producing a D39A mutant (RdnE_D39A) largely did not have this appearance, although a few elongated cells remained, suggesting that the D39A mutant retained partial activity (Figure 1—figure supplement 1). Therefore, the PD-(D/E)XK motif was essential for cell death.

The importance of the PD-(D/E)XK motif for activity suggested that RdnE was a nuclease, but defining its molecular target required direct analysis. Due to its lethality in P. mirabilis and E. coli cells, we synthesized RdnE with a C-terminus FLAG epitope tag using in vitro translation, which resulted in nanogram quantities (Figure 1—figure supplement 1). We added phage lambda DNA (methylated or unmethylated) to progressively higher RdnE protein concentrations and then performed agarose gel electrophoresis analysis. Degradation of lambda DNA occurred in the presence of RdnE, regardless of the DNA methylation state (Figure 1B). The RdnE_D39A construct caused a slight reduction in lambda DNA, while the negative control showed no DNA loss (Figure 1B). RdnE also caused a reduction in plasmid DNA, indicating it has endonuclease activity (Figure 1—figure supplement 1). These results revealed that RdnE caused DNA degradation in vitro in a PD-(D/E)XK-dependent manner.

RdnE appeared to have two different domains, as a region directly follows the PD-(D/E)XK motif. A two-domain architecture is similar to that described for DNases (Lowey et al., 2020; Schiltz et al., 2019). Yet, the PD-(D/E)XK domain could also be sufficient for DNase activity of some effectors such as PoNe-containing DNases (Hespanhol et al., 2022; Jana et al., 2019). Therefore, we examined whether RdnE’s PD-(D/E)XK motif was sufficient for DNA degradation or whether both domains were required for activity. We made independent deletions of each potential RdnE domain. One construct deleted the first alpha helix without disturbing the catalytic residues; the other deleted the region after the PD-(D/E)XK motif, which we termed ‘region 2’. The resulting proteins, produced via in vitro translation, were assayed for DNase activity as described above. The truncated proteins resulted in no loss of lambda DNA (Figure 1C), indicating that both domains were necessary for degradation activity.

We next asked whether RdnE homologs also act as DNA nucleases. A bioinformatics search revealed the closest RdnE homolog outside of Proteus was found in the Actinobacteria, Rothia dentocariosa C6B. Rothia species are inhabitants of the normal oral flora, dwelling in biofilms within the human oral cavity and pharynx (Wilbert et al., 2020). The two RdnE proteins (^ProteusRdnE and ^RothiaRdnE) share approximately 55% amino acid sequence identity, mostly within the PD-(D/E)XK domain; they have similar predicted secondary structures (Figure 1D). Given this, we hypothesized that ^RothiaRdnE also acted as a DNA nuclease.

We analyzed ^RothiaRdnE for PD-(D/E)XK-dependent DNA nuclease activity by producing it and a predicted null mutant, ^RothiaRdnE_D39A, using in vitro translation. Samples containing the ^RothiaRdnE_D39A protein or a negative control had similar DNA levels (Figure 1E). By contrast, samples with the wild-type ^RothiaRdnE protein showed a loss of lambda DNA regardless of methylation state, indicating that ^RothiaRdnE also had DNA nuclease activity (Figure 1E). Given that region 2 was necessary for activity in ^ProteusRdnE but has greater amino acid sequence diversity than the PD-(D/E)XK domain between the two proteins, we queried whether domains from foreign organisms could complement one another. We exchanged region 2 between the ^ProteusRdnE and ^RothiaRdnE sequences and assayed for nuclease activity. The hybrid proteins degraded lambda DNA, unlike the negative control (Figure 1F), demonstrating the cross-phyla protein domains could complement one another. Altogether, these findings demonstrated that these RdnE proteins form a PD-(D/E)XK-containing DNA nuclease subfamily. This conclusion is also consistent with recent literature showing that RdnE-containing proteins (formerly IdrD-CT [Sirias et al., 2020]) form their own sub-clade within other PD-(D/E)XK-containing nucleases (Hespanhol et al., 2022).

As effectors have cognate immunity proteins that are often located adjacently on the chromosome, we hypothesized that rdnI (formerly “idrE”), which is the gene directly downstream of rdnE in P. mirabilis (Figure 2A), encodes the cognate immunity protein. RdnI did not have defined domains, and its function was unknown. We assessed RdnI’s activity using microscopic and cell growth analysis. Swarming P. mirabilis cells are normally elongated with DAPI-stained DNA found along the cell body (Figure 2B). By contrast, swarming cells producing RdnE in trans did not elongate, had a reduced DAPI signal, and had an accumulation of misshapen cells (Figure 2B). Cell shape and DNA-associated fluorescence levels returned to normal when cells concurrently produced the RdnE and RdnI proteins (Figure 2B). RdnI production also rescued cell growth in E. coli cells producing RdnE (Figure 2—figure supplement 1). These data suggested that RdnI inhibits RdnE’s lethality.

We next evaluated whether RdnI provided protection against injected RdnE within mixed communities similar to native ecosystems. We used well-established swarm competition assays, which combine one-to-one mixtures of P. mirabilis strains to measure dominance in two-dimensional population structures (Wenren et al., 2013). The control strain was wild-type strain BB2000 (herein called ‘BB2000’), which naturally produces RdnE and RdnI. The other was strain ATCC29906, which does not naturally produce RdnE and RdnI. These two strains formed a visible boundary between swarming monoculture colonies (Figure 2C). The mixed-strain colony merged with BB2000 in one-to-one competitions, demonstrating BB2000’s dominance in the two-dimensional population structure (Figure 2C). A similar outcome was seen when ATCC29906 produced a vector-encoded Green Fluorescent Protein (GFPmut2) under the fla promoter, which results in constitutive gene expression in swarming P. mirabilis cells (Belas et al., 1991; Jansen et al., 2003). However, BB2000 did not outcompete ATCC29906 engineered to produce vector-encoded RdnI with a C-terminal Strep-tag II epitope tag (‘RdnI-StrepII’) under the fla promoter; this is visible in the merging of the mixed-strain colony with ATCC29906 (Figure 2C). Thus, RdnI protected cells against injected RdnE within mixed communities.

Based on the prevailing EI model, we predicted that a cognate EI pair should bind to one another, which we evaluated in vivo and in vitro. We used the attenuated mutant (RdnE_D39A-FLAG) for these assays because producing the wild-type RdnE protein kills cells. For in vivo analysis, we used bacterial two-hybrid assays (BACTH) in which the reconstitution of the T18 and T25 fragments of adenylate cyclase results in the colorimetric change to blue in the presence of the substrate, X-gal (Battesti and Bouveret, 2012; Karimova et al., 1998). Constructed vectors contained genes for RdnE_D39A-FLAG, RdnI-StrepII, or GFPmut2 on the C-termini of the T18 or the T25 fragment. When the reporter strain produced RdnE_D39A-FLAG or RdnI-StrepII with GFPmut2, the resultant yellow color was equivalent to when X-gal was absent (Figure 2D). There was also minimal color change when an individual protein was produced on both fragments (Figure 2D). However, the reporter strains made blue colonies when X-gal was present, and the cells concurrently produced RdnE_D39A-FLAG and RdnI-StrepII. These results indicated that RdnE and RdnI bind to each other in vivo.

We used batch in vitro co-immunoprecipitation assays to confirm the in vivo binding result. Separate E. coli strains produced either RdnE_D39A-FLAG or had a negative control, exogenous FLAG-BAP (E. coli bacterial alkaline phosphatase with a FLAG epitope tag) added to cell lysate. An anti-FLAG western blot showed both FLAG-BAP (~50 kDa) and RdnE_D39A-FLAG (~17 kDa) in the soluble and elution fractions. RdnI-StrepII eluted with RdnE_D39A-FLAG but not the negative control (Figure 2E). The western blot results corresponded with the Coomassie blue-stained gels (Figure 2—figure supplement 1). Overall, our data showed that Proteus RdnE and RdnI form a cognate EI pair with impacts on population structure. Questions about their prevalence among bacteria and their ecological relevance remained.

Gene neighborhood analysis can guide homology inference and protein comparisons. We conducted consecutive searches with BLAST (Altschul et al., 1990) and HMMER (Eddy, 2009) to identify sequences that encoded proteins with high similarity to RdnE and RdnI (Figure 3—figure supplement 1). The final list contained 21 EI pairs from a variety of phyla that are located adjacently in their respective genomes (Table 1). Although the genes surrounding these putative EI pairs differed, many shared mobile-associated elements, such as Rhs sequences or other similar peptide-repeat sequences (Figure 3A). Several gene neighborhoods had secretion-associated genes, such as the T6SS-associated vgrG/tssI gene and the CDI-associated cdiB gene. A few also included putative immunity proteins from other reported families, like immunity protein 44 (Pfam15571) in Taylorella asinigenitalis MCE3 and immunity protein 51 (Pfam15595) in Chryseobacterium populi CF314. Notably, these organisms varied widely in origin and residence. Some were from the soil rhizosphere (Pseudomonas ogarae and C. populi) and others from the human microbiome (P. mirabilis, R. dentocariosa, and Prevotella jejuni; Figure 3A). The prevalence of these genes across the phylogenetic tree (Figure 3B) and the presence of secretion-associated loci in the gene neighborhoods suggested a role in cell-cell interactions and potentially community structure.

Given the diversity in species, we next examined the relationship between the various RdnE- and RdnI-like proteins and whether there was syntony between these proteins given that they are encoded adjacently on each identified genome. We constructed maximum likelihood trees to examine the relationship between the identified RdnE- and RdnI-like proteins. The RdnE and RdnI trees diverged from the species tree (Figure 3B). However, overall, the arrangement of RdnE- and RdnI-like proteins within the maximum likelihood trees was similar and showed syntony (Figure 3—figure supplement 1), although small differences were present (Figure 3C). For example, P. mirabilis and R. dentocariosa proteins shared more similarities than to those from more closely related genera. These results are consistent with the potential horizontal gene transfer reported for other EI pairs (Ruhe et al., 2020).

Given these results, we reasoned that the domain architectures and amino acid diversity could reveal functions for the two families. When we examined the predicted secondary structures of the RdnE-like proteins, they were conserved despite differences in the amino acid sequences (Figure 3D, Figure 3—figure supplement 1). The RdnE proteins showed two distinct domains, as with the Proteus and Rothia results (Figure 1): the PD-(D/E)XK region followed by a sequence variable region (region 2). Further, the AlphaFold2-generated (Jumper et al., 2021; Mirdita et al., 2022) predicted tertiary structures were consistent with PD-(D/E)XK folds (three β-sheets flanked by two α-helices, α/β/α) found in other proteins (Figure 3E, Figure 3—figure supplement 2; Steczkiewicz et al., 2012). These findings suggested that domains in RdnE are conserved across diverse phyla and further confirm that the sequences seed a distinct PD-(D/E)XK subfamily.

While immunity proteins within a family have diverse overall amino acid sequences, conserved secondary structures and some conserved residues are common in some immunity protein families. Indeed, they are often used to characterize these families (Zhang et al., 2012). We found that while the RdnI proteins shared minimal primary amino acid sequence identity, they were predicted to contain several α-helices (Figure 3D, Figure 3—figure supplement 1) and had similar AlphaFold2-predicted tertiary structures (Figure 3E, Figure 3—figure supplement 2). We also discovered a region with three alpha-helices and several conserved residues, which we named the ‘conserved motif’ (Figure 3D). The RdnI conserved motif might be a key domain for seeding this novel immunity protein family.

We deployed a structure-function approach to determine the conserved motif’s role in ^ProteusRdnI’s activity. Analysis using AlphaFold2 (Jumper et al., 2021; Mirdita et al., 2022) and Consurf (Ashkenazy et al., 2016) revealed seven highly conserved residues within this region; four of these (Y197, H221, P244, E246) clustered together within the AlphaFold2 structure and are identical between sequences (Figure 4A). In a sequence-optimized (SO) RdnI, we independently changed each of these four residues to alanine and discovered that each alanine-substituted variant behaved like the wildtype and inhibited RdnE activity (Figure 4—figure supplement 1). We then replaced all seven residues (Y197, S235, K258, and the original four) with alanine (^ProteusRdnI_7mut-StrepII) and found that, unlike wild-type ^ProteusRdnI-StrepII, this construct was not protective in swarm competition assays (Figure 4B). However, the ^ProteusRdnI_7mut-StrepII mutant still bound RdnE_D39A-FLAG in bacterial two-hybrid assays (Figure 4C). Therefore, the seven residues in the conserved motif are critical for RdnI’s neutralizing function but dispensable for binding RdnE.

Given that the conserved motif and nearby regions are likely involved in protective activity, we queried for potential functions in the remainder of the RdnI protein. We engineered variants that were either (1) the first 85 amino acids, (2) amino acids 150–305, which contained an intact conserved motif, or (3) amino acids 235–305, which contained the last alpha helix of the conserved motif (Figure 3—figure supplement 1). None of these constructs protected against RdnE’s lethality in vivo during the swarm competition assay (Figure 4D), demonstrating that the entire protein is likely essential for function. However, the variant containing the first 85 amino acids of ^ProteusRdnI was the only construct to bind ^ProteusRdnE, indicating that the N-terminal region is sufficient for binding between this P. mirabilis EI pair (Figure 4E). Thus, our data suggests that binding is necessary but not sufficient for neutralization. Also, the inhibitory activity might reside within the second half of RdnI. As the prevailing model defines cognate-specificity by binding activity, our structure-function results for RdnE (Figure 1F) and RdnI (Figure 4E) suggest that this model does not fully explain the complex interactions between effectors and immunity proteins.

Therefore, we explored the relationship between non-cognate RdnE and RdnI proteins from various phyla. We first asked whether non-cognate RdnI immunity proteins could protect against injected ^ProteusRdnE. Using the swarm competition assays, we competed BB2000 against ATCC29906 engineered to produce vector-encoded RdnI homologs from P. mirabilis, R. dentocariosa, P. jejuni, or P. ogarae (^ProteusRdnI-StrepII, ^RothiaRdnI-StrepII, ^PrevotellaRdnI-StrepII, and ^PseudomonasRdnI-StrepII, respectively) under the fla promoter (Figure 2). BB2000 dominated the swarm when ATCC29906 produced GFPmut2, ^PrevotellaRdnI-StrepII, or ^PseudomonasRdnI-StrepII (Figure 4F). However, ATCC29906 outcompeted BB2000 when making ^ProteusRdnI-StrepII or ^RothiaRdnI-StrepII (Figure 4F). Expression levels of the transgenic RdnI proteins in ATCC29906 were similar (Figure 4—figure supplement 2). Further, the RdnI immunity proteins from Proteus and Rothia consistently bound ^ProteusRdnE_D39A in the in vivo and in vitro assays; the Prevotella and Pseudomonas variants did not (Figure 4G, Figure 4—figure supplement 3). The binding to and protection of ^RothiaRdnI against ^ProteusRdnE demonstrated that cross-protection between non-cognate EI pairs from different phyla is possible, provides a fitness benefit during competition, and influences community structure.

While overall the EI pairs showed syntony with each other (Figure 3—figure supplement 1), amino acid changes can have critical impacts on whether there are specific or flexible interactions between non-cognate protein pairs (Schreiber and Keating, 2011). Therefore, we next evaluated which region(s) of RdnI contributes to cross-protection. We first moved the conserved motif of the three foreign RdnI homologs into ^ProteusRdnI-StrepII and measured neutralizing activity using swarm competition assays and binding activity using BACTH. The conserved motifs from Rothia and Prevotella were sufficient to preserve ^ProteusRdnI’s neutralizing (Figure 4H) and binding functions (Figure 4I). However, the conserved motif from Pseudomonas was not sufficient to neutralize ^ProteusRdnE (Figure 4H), even though the construct still bound ^ProteusRdnE_D39A (Figure 4I). We then moved the Proteus conserved motif into the RdnI variants from Prevotella and Pseudomonas. These Prevotella and Pseudomonas hybrid proteins did not protect against ^ProteusRdnE in the swarm competition assay (Figure 4H) or bind to it in the BACTH assay (Figure 4I), indicating that the conserved motif is not required for binding but, alone, is insufficient to confer protection. Thus, RdnI-like immunity proteins containing this conserved motif can protect against non-cognate effector proteins if binding has been established.

Our findings revealed that immunity proteins such as RdnI could provide a broader protective umbrella for a cell beyond inhibiting the effector proteins of their siblings. If so, one would expect to find evidence of RdnE and RdnI homologs from different phyla in the same environment or microbial community. We tested this hypothesis by analyzing around 500,000 publicly available microbiomes (metagenomes) for the specific rdnE and rdnI gene sequences examined in Figure 4 (Figure 5A). 2296 human and terrestrial metagenomes contained reads matching with over 90% identity to these rdnE sequences (Figure 5B). We used this cutoff to ensure that each nucleotide sequence queried in the metagenomes closely matched experimentally characterized reference sequences. As a control, we applied a 70% identity threshold, which would retain related but more divergent sequences. We saw similar patterns with a total ~2% change in the number of genomes per category (Figure 5—figure supplement 1). The reads mapped to the expected niche for each organism, underscoring the presence of the genes encoding these specific effector proteins in naturally occurring human-associated microbiomes.

The rdnE and rdnI genes from various human-associated bacteria occurred concurrently in individual human oral and, to a lesser extent, gut metagenomes. The rdnE and rdnI genes from Rothia and Prevotella co-occurred in approximately 5% of the metagenomes analyzed (Figure 5C, Figure 5—figure supplement 2). Stringent detection parameters were utilized, so the true number could be higher. We then compared the abundance of rdnI to rdnE reads, since metagenomic coverage (i.e. the number of short reads that map to a gene) approximates the underlying gene’s abundance in the sampled community. In most gut samples, rdnI recruited more reads than rdnE, although there was substantial variance (Figure 5D). These data could indicate the presence of orphan rdnI genes, which is consistent with published T6SS orphan immunity alleles (Bosch et al., 2023; Kirchberger et al., 2017; Koskiniemi et al., 2014). These metagenomic patterns suggest that a single community can produce multiple RdnE and RdnI proteins from different phyla, thereby providing a potential for them to interact in a host environment.

All strains are described in Table 2. Strains for bacterial two-hybrid assays were transformed the day before. Overnight cultures were grown aerobically at 37 °C in LB (Lennox) broth (Belas et al., 1991). E. coli strains were plated on LB (Lennox) agar surfaces (1.5% Bacto agar) and P. mirabilis strains were plated on LSW^– agar (Belas et al., 1991) for single-colony growth or 25 mL CM55 media (Thermo Fisher Scientific Cat# CM0055B) for swarms. When necessary 35 μg/mL kanamycin or 100 μg/mL carbenicillin was included in the media.

Plasmids were constructed according to Table 3. Primers and gBlocks were ordered from Integrated DNA Technologies (IDT), Coralville, IA. P_idrA-RdnE was constructed using Polymerase Chain Reaction (PCR) to amplify the last 416 bp of the idrD gene from BB2000 and clone it into the SacI and AgeI sites of the pBBR1-NheI vector, resulting in plasmid pAS1054. RdnE is the final 138 amino acids of IdrD (out of its total of 1581). P_idrA-rdnE-rdnI was constructed by PCR amplifying the last 416 bp of the idrD gene through the end of the rdnI gene from BB2000, resulting in the plasmid pAS1059. The gBlock and primer sequences are archived on an OSF website (https://osf.io/scb7z/).

We used several standard protocols for vector construction. Seamless ligation cloning extract (SliCE) was adapted from Zhang et al., 2014. Restriction-digest reactions were based on manufacturer’s protocols. Overlap extension (SOE) PCR Amplification was adapted from Heckman and Pease, 2007. Plasmids were transformed into OmniMax E. coli and confirmed using Sanger Sequencing (UC Berkeley DNA Sequencing Facility and Genewiz, South Plainfield NJ).

RdnE proteins were produced using the New England Biolabs PURExpress In Vitro Protein Synthesis Kit (New England BioLabs Inc, Ipswich MA). Template DNA contained the rdnE gene and required elements specified by the PURExpress kit. We adapted this protocol from prior in vitro DNA-degradation assays (Hughes and Cidlowski, 1997). Reactions were performed with 250 ng of template DNA and incubated at 37 °C for 2 hr (no template DNA added to negative control reaction). The protein amount was determined using an anti-FLAG western blot with a known gradient of FLAG-BAP (2.5, 5, 10, and 20 ng). Synthesized protein (2.5, 5, and 10 ng) was added to 0.5 µg of lambda DNA (methylated and unmethylated), 5 µL of New England Biolabs Buffer 3.1, and up to a final volume of 25 µL. For plasmid DNase assays, 10 ng of synthesized protein was added to 250 ng of circular or linear plasmid DNA (pids_BB [Gibbs et al., 2008]). This reaction was incubated for 1 hr at 37 °C, then Proteinase K (New England Biolabs Inc, Ipswich MA) was added and incubated for an additional 15 min at 37 °C. The reaction was then run on a 1% agarose gel for analysis.

Overnight cultures were grown at 37 °C in a shaking incubator in LB broth with appropriate antibiotics. Cultures were normalized to an optical density at 595 nm (OD₅₉₅) of 1 and diluted 1:100 into LB broth with 35 μg/mL kanamycin, with and without 200 nM anhydrotetracycline (aTc). Samples were analyzed for OD₅₉₅ every thirty minutes for 16 hr in a 96-well plate using a TECAN. Other samples were incubated at 37 °C for 6 hr while rocking. At indicated time points, 100 μL of sample was removed, diluted, and then plated on fresh LB agar plates to measure colony- forming units per mL (CFU) after overnight growth at 37 °C using standard protocols.

We performed microscopy on P. mirabilis strain idrD::Tn5 (Cm^R) (also called, idrD*), which has a transposon insertion to disrupt rdnE and rdnI expression (Wenren et al., 2013), carrying either vector pBBR1-NheI or pDS0002 (producing RdnE) and on E. coli carrying either pBBR1-NheI, pDS0002, or pDS0048 (producing RdnE_D39A). P. mirabilis cells were normalized to OD₅₉₅ of 0.1 after overnight growth in LB broth supplemented with kanamycin. Cells were inoculated onto CM55 swarm pads containing 10 µg/mL DAPI and 10 nM aTc and grown in humidified chambers at 37 °C. Images were taken at five and six hours after growth. From overnight cultures, E. coli cells were grown in LB broth plus kanamycin until mid-logarithmic phase and then mounted directly onto glass slides. Glass coverslips were sealed with nail polish. For all microscopy, we captured phase contrast and DAPI (150ms exposure) images using a Leica DM5500B microscope (Leica Microsystems, Buffalo Grove IL) and CoolSnap HQ CCD camera (Photometrics, Tucson AZ) cooled to –20 °C. MetaMorph version 7.8.0.0 (Molecular Devices, Sunnyvale CA) was used for image acquisition.

The P. mirabilis rdnI-StrepII sequence was difficult to genetically engineer due to its low GC% content (23%). As such, we engineered the sequence to have a higher GC%, called ‘Sequence optimized (SO) ^ProteusRdnI-StrepII’ without changes to its amino acid sequence. The change to the nucleotide sequence did not affect the construct’s ability to offer protection to a vulnerable strain (Figure 4B).

The swarm competition (territoriality) assay was adapted from Wenren et al., 2013. 5 mL cultures were grown in LB broth with appropriate antibiotics overnight in a 37 °C rocking incubator. Overnight cultures were normalized to an OD₅₉₅ of 1. For the competition samples, the strains were mixed 1:1. 2 μL of each sample was inoculated onto CM55 agar with the appropriate antibiotic. Plates were incubated at 37 °C for 22 hr and then photographed and assessed for boundary formation. All RdnI-producing strains contained a low-copy vector with the rdnI gene under the fla constitutive promoter; see Figure 4—figure supplement 2 for relative protein production.

The vectors are described in Battesti and Bouveret, 2012 with an added linker region between the T25 or T18 fragments and multiple cloning sites. BTH101 cultures were grown at 30 °C overnight in LB broth with kanamycin and carbenicillin. 10 μL of the overnight culture were inoculated onto LB agar with kanamycin, carbenicillin, 1 mM IPTG, and 0 or 40 μg/mL of X-gal (Thermo Fisher, Waltham MA), and grown at 30 °C for 24 hr. Color was amplified by an additional 24 hr at 4 °C, and then samples were imaged.

The protocol was adapted from Cardarelli et al., 2015. E. coli cells were harvested from LB broth, grown for either 3 hr after induction with 200 nM aTc at 37 °C or 16–20 hr at 16 °C after induction with 1 mM IPTG. Cells were spun down into pellets using centrifugation and then flash frozen in liquid nitrogen. RdnE-containing samples were lysed in 50 mM Tris pH 7.4, 150 mM NaCl, and 1 x Protease Inhibitor Cocktail (Selleck Chemicals LLC, Houston TX), via bead bashing for 20 min at 4 °C. RdnI-containing samples were lysed in 100 mM Tris-HCl pH 8, 180 mM NaCl, and 1 x Protease Inhibitor Cocktail via 10x10 s sonication pulses. The soluble fraction for both samples was obtained after centrifugation at 15,000 rpm for 15 min. FLAG epitope-containing samples were incubated with prepared resin for 2 hr at 4 °C. The resin was then washed twice (50 mM Tris pH 7.4, 150 mM NaCl, and 1% Tween-20), incubated with approximately 1 mL of the soluble fraction of the RdnI-StrepII-containing samples for another 2 hr, and washed thrice more. The protein was finally eluted with 50 μL of 300 ng/μL 3 x FLAG peptide (Sigma-Aldrich, St. Louis, MO) for 45 min at 4 °C. Sample buffer (63 mM Tris pH 6.8, 2% Sodium Dodecyl Sulfate, 10% glycerol, 5% 2-Mercaptoethanol) was added to samples, boiled at 95 °C for 10 min, and frozen at –80 °C.

The protocol was adapted from Cardarelli et al., 2015. P. mirabilis cells were harvested from CM55 swarm plates, grown overnight at 37 °C. Cells were washed twice in LB broth, spun down by centrifugation, and flash frozen in liquid nitrogen. Cells were lysed in 100 mM Tris-HCl pH 8, 180 mM NaCl, and 1 x Protease Inhibitor Cocktail via 10 x via bead bashing for 20 min at 4 °C. The whole cell extract was obtained after centrifugation at 6000 × g for 15 min. The soluble fraction for both samples was obtained after subsequent centrifugation at 15,000 rpm for 15 min. Sample buffer (63 mM Tris pH 6.8, 2% Sodium Dodecyl Sulfate, 10% glycerol, 5% 2-Mercaptoethanol) was added to samples, boiled at 95 °C for 10 min, and then immediately used for SDS-PAGE and western blotting.

The protocol was adapted from Cardarelli et al., 2015. Protein samples were separated by gel electrophoresis using 13% Tris-Tricine polyacrylamide gels and either transferred to a 0.45 μm nitrocellulose membrane (Bio-Rad Laboratories, Hercules CA) or stained with Coomassie blue (Bio-Rad Laboratories, Hercules CA). Western blot membranes were probed with primary antibody, either 1:4000 rabbit anti-FLAG (Sigma-Aldrich Cat# F3165, RRID:AB_259529) or 1:4000 or 1:2000 mouse anti-StrepII (GenScript Cat# A01732, RRID:AB_2622218) for 1 hr at room temperature or overnight at 4 °C and with secondary antibody either 1:5000 goat anti-rabbit or anti-mouse respectively conjugated to horseradish peroxidase (HRP) (SeraCare KPL Cat# 5220–0395, RRID:AB_3698113 and SeraCare KPL Cat# 074–1506, RRID:AB_2721169, respectively) for 30 min for co-IPs or 1 hour for P. mirabilis protein expression at room temperature. Samples were finally visualized using Immun-Star HRP substrate kit (Bio-Rad Laboratories, Hercules, CA) and the ChemiDoc XRS system (Bio-Rad Laboratories, Hercules, CA). TIFF files were analyzed on Fiji (ImageJ, Madison, WI).

A BLAST (Altschul et al., 1990) search of the P. mirabilis RdnE protein sequence revealed seven RdnE homologs from a variety of species. The downstream genes of these RdnE homologs were identified using the DOE Joint Genome Institute (JGI) Integrated Microbial Genomes and Microbiomes (IMG/M) (Chen et al., 2021; Mukherjee et al., 2021). The seven RdnE and RdnI amino acid sequences were separately aligned with MUSCLE using Jalview (Edgar, 2004; Waterhouse et al., 2009). These alignments were then used as seeds for a second homology search using HMMER search (HmmerWeb version 2.41.2; Finn et al., 2015; Finn et al., 2011) and the Ensembl Database (Cunningham et al., 2022). The two data sets were then compared for genomes that contained both rdnE and rdnI genes next to one another within their respective genomes. Any EI pairs that contained disrupted PD-(D/E)XK motifs within their RdnE sequence were removed.

Gene neighborhoods were obtained using JGI’s IMG/M Neighborhood viewer and then redrawn using Adobe Illustrator (Adobe Inc, 2022). Locations of predicted functions are approximate primarily based on the Pfam domain calling by IMG/M. The gene neighborhoods, relevance, and niche were also accessed from IMG/M.

The final 21 RdnE and RdnI sequences were aligned with MUSCLE using Jalview (Edgar, 2004; Waterhouse et al., 2009). The conserved residues were identified using Jalview, and the cartoons were created using Adobe Illustrator (Adobe Inc, 2022). The sequence logo for the RdnI conserved motif was generated with WebLogo (Gabler et al., 2020) and constrained to only visualize the conserved motif. Trees for unrooted maximum likelihood trees of the RdnE and RdnI were created with RaxML (Kozlov et al., 2019). The phylogenetic tree is based on NCBI taxonomy. A tanglegram (Scornavacca et al., 2011) was made from the RdnE and RdnI protein families from the 21 sequences.

Secondary structure predictions of the MUSCLE aligned sequences were determined with Ali2D from the MPI Bioinformatics toolkit (Edgar, 2004; Gabler et al., 2020; Zimmermann et al., 2018). The resulting predictions were made into cartoons manually using Adobe Illustrator (Adobe Inc, 2022). Tertiary structure predictions were done with AlphaFold2 (Jumper et al., 2021) using Mmseqs2 on Google Colab (Mirdita et al., 2022). Query protein sequences were inputted into the program and then run, producing 5 models ranked 1–5. Rank 1 models are shown. pIDDT scores indicate confidence levels for each amino acid position. Structures were analyzed in PyMOL (The PyMOL Molecular Graphics System, Version 2.2.3 Schrödinger, LLC.). The pIDDT graphs are in the supplemental data.

A sourmash-based approach was used to screen approximately 500,000 public metagenomes stored on NCBI’s SRA (https://github.com/sourmash-bio/2022-search-sra-with-mastiff; sourmash, 2023) for the presence of the 10 genomes shown in Figure 3A. Hits with a containment score greater than 0.2 were downloaded for further analysis, representing 9137 metagenomes. Each metagenome was then mapped with bbmap (Bushnell, 2014) against a reference database the 10 rdnE and rdnI gene sequence pairs with a stringency of 90% (minid = 0.9), along with quality filtering (trim1=20, minaveragequality = 10). 70% stringency was also included and resulted in similar results (Figure 5—figure supplement 1). After mapping, metagenomes were retained if they had (1) a mean coverage greater than 2 X, (2) at least one base covered greater than 5 X, and (3) more than half of the bases on reference rdnE-rdnI sequence receiving coverage. Coverage of other domains, if any, upstream of the C-terminal domain was not considered for subsequent analysis. 2857 metagenomes met these criteria, of which 2296 contained P. mirabilis, R. dentocariosa, P. jejuni, or P. ogarae sequences and could be confidently assigned to samples obtained from the human gut or oral microbiome or from terrestrial sources. Gene-level coverage in a sample was then summarized as each gene’s average nucleotide coverage. The ratio of rdnI to rdnE coverage was then calculated for each sample and log10-transformed, and the distribution of ratios was summarized with Python’s seaborn kdeplot using a bandwidth of 0.4 (Waskom, 2021).

The sequence files and associated data, including sequence datasets and protein modeling files, are archived on an OSF website (https://osf.io/scb7z/) and were made publicly available upon publication. Primer and gBlock sequences are available at the OSF link above. Also included are any newly generated custom and reused computer code. All plasmids and strains are available by contacting the corresponding author and will be shipped in accordance with University of California, Berkeley and relevant authorities expediently upon written request.

All biological experiments were performed a minimum of three independent times, each time with independent isolates. The data describes biological replicates. The authors have no conflicts of interest to report. No human or animal subjects were used in this study.