Introduction

Specificity between protein-protein interactions is key for many biological processes, such as metabolism, development, and intercellular signaling. Binding to an incorrect partner or disrupted binding of the specific (cognate) protein can cause a diseased state or cell death¹. For bacterial social behaviors, cognate protein-protein interactions between cells impact organism fitness and community structure, such as excluding foreign cells². This importance of specificity between cognate partners in social behaviors remains largely unexplored. However, in other contexts, flexible (“promiscuous”) binding allows protein partners to retain their interactions when undergoing rapid mutational changes, such as during immune recognition of viral particles^3,4. Unknown is whether flexible binding between noncognate proteins can occur during bacterial social behaviors and thereby impact microbial communities. An expanded protective function would reveal new bacterial behaviors that influence individual fitness and community structure in dynamic ecosystems.

Microbes often exist within dense, multi-phyla communities, like the human gut microbiome, where they communicate and compete with neighbors. Bacteria can use effector-immunity protein (EI) pairs in these environments to gain advantages^5,6. Unlike bacterial toxin-antitoxin (TA) systems in which a single cell produces both toxic and neutralizing proteins, bacteria inject cell-modifying proteins (called interbacterial “effectors”) directly into neighboring cells via several contact-dependent transport mechanisms, including the type VI secretion system (T6SS), type IV secretion system (T4SS), and contact-dependent inhibition (CDI)^7,8. Clonal siblings produce the matching immunity protein that modifies the effector’s activity. For lethal effectors, both clonal and non-clonal cells are negatively impacted when binding is disrupted or absent⁵. These interactions between EI pairs can shape community composition by changing bacterial fitness.

The interaction specificity between matching EI pairs has historically defined immunity protein protection, but recent studies raise doubts. Of note, EI pairs interact within a neighboring cell which creates unique restrictions for both their protection mechanisms and their evolution⁹. Currently, the predominant model is that T6SS-associated EI pairs act like a tumbler lock-and-key, where each effector protein has a single cognate partner¹⁰. Immunity proteins bind their cognate effectors, often at the active site, to neutralize effector activity^11,12. However, experiments with engineered proteins reveal that small amino acid sequence changes to an immunity protein can allow it to bind effectors other than its cognate partner^13,14. Also, the T6SS-associated effector and immunity proteins from Salmonella enterica subsp. enterica serovar Typhimurium and Enterobacter cloacae, which are phylogenetically close, bind each other in vitro and protect against the other in vivo¹⁵. Another example is Tde1 and Tdi1. Homologous Tdi1 immunity proteins lacking a cognate effector (i.e., “orphans”) bound and protected against the effector from a different organism¹⁶. These studies indicate that the widely used tumbler lock-and-key model does not account for the potential breadth of immunity protein protection.

We studied an EI pair in Proteus mirabilis to examine this prevailing model. This opportunistic pathogen resides in human and animal guts and can cause recurrent and persistent urinary tract infections¹⁷. P. mirabilis encodes two T6SS-dependent EI pairs (one lethal and one non-lethal) that impact collective motility and relative fitness^18,19. For the lethal EI pair, previously termed Idr¹⁹, the molecular functions remained unknown. Here, we characterized this EI pair and determined the critical residues for activity, leading to the identification of two protein families. We showed that proteins in the immunity protein family bind non-cognate effectors produced by bacteria from different phyla and result in altered population structures. Structure-function assays revealed that a conserved region within the C-terminus of the immunity proteins is necessary to neutralize the P. mirabilis effector protein. Further, we found that the flexible EI pairs from various phyla naturally co-occur in individual human microbiomes. These findings provide compelling evidence for cross-protection and support a critical revision of the model for EI pairs, particularly in consideration of ecological significance.

Results

RdnE is a DNA nuclease and seeds a PD-(D/E)XK subfamily

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 effector¹⁸. The idrD gene contains an Rhs region within its N-terminus. Many Rhs-containing effectors often contain an enzymatic domain in the C-terminus^20,21. 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 genes¹⁹. 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.

Figure 1.

Figure 1 - Figure supplement 1.

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^22–24. Three residues in the catalytic site—D, D/E, and K—are required for activity²⁵. 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^26,27. 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^28,29. Yet, the PD-(D/E)XK domain could also be sufficient for DNase activity of some effectors such as PoNe-containing DNases^22,30. 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 pharynx³¹. 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-CT³²) form their own sub-clade within other PD-(D/E)XK-containing nucleases³⁰.

RdnI binds and neutralizes RdnE

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.

Figure 2.

Figure 2 - Figure supplement 1.

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 structures¹⁹. 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^33,34. 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^35,36. 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.

Expansion of the RdnE and RdnI protein families revealed similar gene architecture and secondary structures.

Gene neighborhood analysis can guide homology inference and protein comparisons. We conducted consecutive searches with BLAST³⁷ and HMMER³⁸ 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.

Figure 3.

Figure 3 - Figure supplement 1.

Figure 3 - Figure supplement 2.

Table 1:

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), though 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 pairs⁷.

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^39,40 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)²⁵. 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 families⁴¹. 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.

Binding flexibility in RdnI allows for cross-species protection

We deployed a structure-function approach to determine the conserved motif’s role in ^ProteusRdnI’s activity. Analysis using AlphaFold2^39,40 and Consurf⁴² 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 (Figures 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.

Figure 4.

Figure 4 - Figure supplement 1.

Figure 4 - Figure supplement 2.

Figure 4 - Figure supplement 3.

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 to 305, which contained an intact conserved motif, or (3) amino acids 235 to 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 pairs⁴. 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.

RdnE and RdnI proteins from diverse phyla are present in individual human microbiomes

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). 2,296 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.

Figure 5.

Figure 5 - Figure supplement 1.

Figure 5 - Figure supplement 2.

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^16,43,44. 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.

Discussion

Using these results as a foundation, we propose an extension to the prevailing model of selective, cognate EI partners⁴⁵ to incorporate “EI skeleton keys.” In this revised model, flexible (“promiscuous”) binding between non-cognate effector and immunity proteins enables broader protection in mixed-species communities (Figure 5E). Here, we demonstrated that RdnE-RdnI binding is necessary but not sufficient to neutralize RdnE, which differs from many previously described T6SS immunity proteins. We showed that full-length RdnE, containing both its PD-(D/E)XK domain and variable C-terminal region, is a DNA-degrading endonuclease. Likewise, RdnI requires its entire protein to bind and neutralize RdnE, including a newly identified conserved C-terminal motif. Our findings point to a possible two-step mechanism for how the RdnI immunity protein works: the N-terminal variable-sequence domain mediates binding to an effector, while the C-terminal conserved domain contributes to neutralization in a not-yet-determined mechanism (Figure 5E). These findings have potential impacts on our understanding of immunity protein evolution, molecular functions, and microbial community structure.

The domain architectures of RdnE and RdnI suggest possible evolutionary trajectories for these EI pairs. While both require the full-length protein, RdnE and RdnI can function with residue changes within the domains and even retain activity in cross-phyla hybrid proteins. Tri1 immunity proteins also contain two domains, but these domains are associated with distinct functions. Their conserved region corresponds with the broad-acting enzymatic domain, which is independent of their cognate-specific binding domain,⁴⁶ suggesting that conserved enzymatic activity can be maintained alongside potential coevolution necessary for strict cognate pair binding. Therefore, individual domains in EI proteins may evolve independently rather than the entire protein experiencing coevolution. As such, RdnE’s PD-(D/E)XK motif and RdnI’s conserved motif might be maintained for activity, while the variable domains may diversify in sequence independently. Depending on the selective pressures, the variable regions could reinforce specificity between cognate EI pairs as they coevolve. Additional evolutionary analysis would reveal how the balance between specificity and flexibility evolves in EI pairs, both within domains and across the entire protein.

RdnI’s potential two-step mechanism adds to a growing number of ways in which immunity proteins neutralize effector proteins. However, RdnI’s neutralization mechanism remains unknown. Many crystalized structures of EI complexes show that immunity proteins can bind and occlude an effector’s active site, effectively neutralizing the effector’s function^12,47. However, some immunity proteins allosterically inhibit their effector without blocking the active site^48,49. Recent studies revealed more neutralization mechanisms. The Tri1 immunity protein has a conserved enzymatic function that neutralizes its effector’s activity, allowing for protection against foreign effectors in addition to selective cognate-binding activity,⁴⁶ while the Tdi1 immunity protein conformationally disrupts its effector, Tde1’s, binding and active sites to prevent DNA nuclease activity¹⁶. One will need to experimentally determine the structure of the RdnE-RdnI complex is necessary to define how it neutralizes RdnI and how this molecular mechanism compares to other immunity proteins.

Regardless, our results indicate that RdnI’s conserved domain is essential for protective activity and that a combination of seven highly conserved residues mediates that protection. There are several possibilities for how this domain aids neutralization, such as an ion-binding pocket, structural stability, or protein partner binding. Another possibility is that RdnI’s conserved region reinforces binding to the effector, aiding non-cognate interactions or co-evolving pairs. For example, while our assays indicate the primary binding domain for RdnI is in the N-terminus, the conserved domain could reinforce an initial, transient binding interaction. Indeed, multiple binding domains have been recorded for TA systems and likely protect the cell during co-evolution⁵⁰. Binding affinities between effector and immunity proteins are not well-documented; those reported vary. For DNA nuclease colicins, non-cognate interactions have affinities in the nanomolar range whereas cognate interactions have picomolar affinities⁵¹. However, the T6SS-associated EI pair, Tde1 and Tdi1, have similar nanomolar affinities for both cognate and non-cognate orphan EI pairs¹⁶. In addition, it is unclear what equilibrium dissociation constant (K_D) for EI binding would confer protection in native systems. This K_D may be especially important in the case of highly motile bacteria, such as P. mirabilis, that only need to survive long enough to escape. Additionally, interbacterial effectors act within a neighboring cell, which may make determining the native ratios of effector to immunity proteins challenging. It will be interesting to see how binding affinities between other EI pairs compare, both cognate and non-cognate interactions, and how these affinities relate to protection in mixed communities.

When considering the impacts on community structure, the broadened activity of RdnI proteins against RdnE effectors from multiple phyla likely increases bacterial fitness, which is advantageous in dense environments. Our analysis measured protection during a two-dimensional, swarm-structured competition, where RdnI offered a susceptible strain protection against trans-cellular RdnE delivered natively. As such, we can conclude that RdnI production increased individual cells’ fitness and modified the community structure; it enabled vulnerable bacteria to inhabit previously restricted spaces. Supporting this experimental data, both gut and oral metagenomes showed evidence of multiple rdnE-rdnI pairs within individual samples, particularly between Rothia and Prevotella. Interestingly, the oral microbiome had roughly equivalent abundance between the effector and immunity genes, which might reflect that bacteria occupy distinct spatiotemporal niches within oral microbiomes, e.g. R. dentocariosa is predominantly on tooth surfaces⁵². By contrast, rdnI genes had greater abundance compared to rdnE in the gut microbiomes, which may reflect the greater diversity in member species and community structures found in the gut⁵³. Orphan immunity genes are indeed a known phenomenon in T6SS EI literature but are usually documented through single isolate sequencing, with notable exceptions such as the Ross et al. 2019 and Bosch et al. 2023 papers. This community-level assessment affirms the presence of rdnI orphan genes on a population scale and points to relatively widespread immunity genes in hundreds or thousands of samples.

Given the ability of immunity genes to protect against non-cognate effectors, the presence of diverse orphan rdnI genes hints at the ecological complexity surrounding RdnE and RdnI. This community of immunity proteins is reminiscent of the model for shared immunity proteins within an ecosystem, called a “hyper-immunity state,” which was seen among colicins in wild field mice⁵⁴. In this hyper-immunity state, a set of immunity proteins shared among a community could offer an advantage against pathogens. Invading bacteria would be unable to defend themselves from certain effectors, while the community would be protected as they share the collective immunity proteins. Flexible binding like RdnI could contribute to such a “hyper-immunity state” to help a bacterial community maintain its niche.

Indeed, bacteria have a diverse set of protective measures to ward off foreign effectors in addition to flexible immunity proteins. Recent work has identified non-specific mechanisms of protection including stress-response, physical barriers, and a stronger offense¹⁰. Orphan immunity genes also exist throughout many bacterial genomes and may be a part of this system,⁵⁵ for example, orphan immunity genes offer a fitness advantage in vitro^12,16 and in mouse microbiomes⁵⁶. Flexible EI pairs are also not limited to secretion systems but are also seen among TA pairs⁵⁷ and bacteriocins^51,58. Our data extends the current repertoire of protection mechanisms by adding another tool: a flexible immunity protein collection, where each immunity protein acts as a skeleton key against a wider class of effectors. This flexibility is seen among orphan immunity proteins^12,16 and for immunity proteins with cognate effectors as in this study. Flexible binding could be a general property of T6SS immunity proteins that could be useful in dense, diverse communities, like human and soil microbiomes, where contact-dependent competition using EI pairs is critical to maintain one’s population. Indeed, the physical interactions between, and evolution of, effector and immunity proteins remain a rich area for new explorations.

Acknowledgements

We thank Caroline Boyd, Neils Bradshaw, Emma Keteku, Alecia Septer, Nora Sullivan, Adnan Syed, and Larissa Wenren for contributing experimental materials to this project. Rachelle Gaudet, Colleen Cavanaugh, and members of the Gibbs Lab provided valuable advice on the manuscript. The David and Lucile Packard Foundation, the George W. Merck Fund, Harvard University, and the University of California, Berkeley, and funded this research. A.K., D.S., D.U., and K.A.G designed and performed research as well as analyzed data. A.K., D.S., D.U., and K.A.G wrote the paper. We have no competing interests to declare.

Materials and Methods

Bacterial strains and media

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) broth³³. E. coli strains were plated on LB (Lennox) agar surfaces (1.5% Bacto agar); P. mirabilis strains were plated on LSW^- agar³³ for single-colony growth or 25 mL CM55 media (blood agar base agar [Oxoid, Basingstoke, England]) for swarms. When necessary 35 μg/mL kanamycin or 100 μg/mL carbenicillin was included in the media.

Table 2:

Plasmid construction

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/) and were made publicly available upon publication.

Table 3:

We used several standard protocols for vector construction. Seamless ligation cloning extract (SliCE) was adapted from Zhang et al⁵⁹. Restriction-digest reactions were based on manufacturer’s protocols. Overlap extension (SOE) PCR Amplification was adapted from Heckman 2007⁶⁰. Plasmids were transformed into OmniMax E. coli and confirmed using Sanger Sequencing (UC Berkeley DNA Sequencing Facility and Genewiz, South Plainfield NJ).

In vitro DNase assay

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 assays⁶¹. Reactions were performed with 250 ng of template DNA and incubated at 37°C for two hours (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_BB⁶²). This reaction was incubated for one hour at 37°C, then Proteinase K (New England Biolabs Inc., Ipswich MA) was added and incubated for an additional 15 minutes at 37°C. The reaction was then run on a 1% agarose gel for analysis.

E. coli liquid growth and viability assays

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 hours in a 96-well plate using a TECAN. Other samples were incubated at 37°C for 6 hours 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.

Microscopy

We performed microscopy on P. mirabilis strain idrD::Tn5 (Cm^R) (also called, idrD*), which has a transposon insertion to disrupt rdnE and rdnI expression¹⁹, 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 (150 ms 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.

Sequence Optimized RdnI

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).

Swarm competition assay

The swarm competition (territoriality) assay was adapted from Wenren et. al. 2013¹⁹. 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 were inoculated onto CM55 agar with the appropriate antibiotic. Plates were incubated at 37°C for 22 hours 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.

BACTH assay

The vectors are described in Battesti and Bouveret 2012³⁵ 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 hours. Color was amplified by an additional 24 hours at 4°C, and then samples were imaged.

FLAG co-immunoprecipitation assays

The protocol was adapted from Cardarelli et. al. 2015². E. coli cells were harvested from LB broth, grown for either 3 hours after induction with 200 nM aTc at 37°C or 16 to 20 hours 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 50mM Tris pH 7.4, 150mM NaCl, and 1x Protease Inhibitor Cocktail (Selleck Chemicals LLC, Houston TX), via bead bashing for 20 minutes at 4°C. RdnI-containing samples were lysed in 100 mM Tris-HCl pH 8, 180 mM NaCl, and 1x Protease Inhibitor Cocktail via 10× 10 second sonication pulses. The soluble fraction for both samples was obtained after centrifugation at 15,000 rpm for 15 minutes. FLAG epitope-containing samples were incubated with prepared resin for two hours 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 hours, and washed thrice more. The protein was finally eluted with 50 μL of 300 ng/μL 3x FLAG peptide (Sigma-Aldrich, St. Louis, MO) for 45 minutes 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 minutes, and frozen at -80°C.

P. mirabilis swarm cell protein expression

The protocol was adapted from Cardarelli et. al. 2015². 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 1x Protease Inhibitor Cocktail via 10x via bead bashing for 20 minutes at 4°C. The whole cell extract was obtained after centrifugation at 6,000 g for 15 minutes. The soluble fraction for both samples was obtained after subsequent centrifugation at 15,000 rpm for 15 minutes. 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 minutes, and then immediately used for SDS-Page and Western blotting.

SDS-Page and Western blotting

The protocol was adapted from Cardarelli et. al. 2015². 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, St. Louis MO] or 1:4000 or 1:2000 mouse anti-StrepII [Genscript, Piscataway NJ]) for 1 hour 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) [KPL, Inc., Gaithersburg MD]) for 30 minutes 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).

Bioinformatics search for RdnE and RdnI homologs

A BLAST³⁷ 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). The seven RdnE and RdnI amino acid sequences were separately aligned with MUSCLE using Jalview⁶³. These alignments were then used as seeds for a second homology search using HMMER search (HmmerWeb version 2.41.2) and the Ensembl Database⁶⁴. 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 neighborhood and primary conservation analyses

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 final 21 RdnE and RdnI sequences were aligned with MUSCLE using Jalview⁶³. 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 WebLogo⁶⁵ and constrained to only visualize the conserved motif.

Secondary and tertiary structure predictions

Secondary structure predictions of the MUSCLE aligned sequences were determined with Ali2D from the MPI Bioinformatics toolkit^65,66. The resulting predictions were made into cartoons manually using Adobe Illustrator (Adobe Inc., 2022). Tertiary structure predictions were done with AlphaFold2³⁹ using Mmseqs2 on Google Colab⁴⁰. 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.

Metagenomic analyses

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) for the presence of the ten genomes shown in Figure 3A. Hits with a containment score greater than 0.2 were downloaded for further analysis, representing 9,137 metagenomes. Each metagenome was then mapped with bbmap⁶⁷ 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 2X, (2) at least one base covered greater than 5X, 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 were not considered for subsequent analysis. 2,857 metagenomes met these criteria, of which 2,296 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’s 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 0.4⁶⁸.

Open Science statement

The sequence files and associated data are archived on an OSF website (https://osf.io/scb7z/) and were made publicly available upon publication.