|Home | About | Journals | Submit | Contact Us | Français|
Multicellularity inherently involves a number of cooperative behaviors that are potentially susceptible to exploitation but can be protected by mechanisms such as kin discrimination. Discrimination of kin from non-kin has been observed in swarms of the bacterium Bacillus subtilis, but the underlying molecular mechanism has been unknown. We used genetic, transcriptomic, and bioinformatic analyses to uncover kin recognition factors in this organism. Our results identified many molecules involved in cell surface modification and antimicrobial production and response. These genes varied significantly in expression level and mutation phenotype among B. subtilis strains, suggesting interstrain variation in the exact kin discrimination mechanism used. Genome analyses revealed a substantial diversity of antimicrobial genes present in unique combinations in different strains, with many likely acquired by horizontal gene transfer. The dynamic combinatorial effect derived from this plethora of kin discrimination genes creates a tight relatedness cutoff for cooperation that has likely led to rapid diversification within the species. Our data suggest that genes likely originally selected for competitive purposes also generate preferential interactions among kin, thus stabilizing multicellular lifestyles.
The advent of multicellularity fundamentally changed the evolutionary trajectory of organisms. Single cell based selection pressures were supplemented with group-level selection, allowing for adaptive traits like division of labor and public goods production, behaviors that would not be maintained in a selfish population . Multicellularity is ubiquitous across all life, and results from physical attachment of cells and coordination of their behavior through intercellular communication, producing a unit with united evolutionary interests. In bacteria, examples of multicellularity include filaments, biofilms, and swarms .
Cooperation provides advantages to multicellular organisms, including efficient resource sharing, communal predation, and protective enclosure in an extracellular matrix. These altruistic acts are theoretically vulnerable to selfish entities that reap the benefits without contributing themselves. This can be prevented if cooperative acts are preferentially directed to other cooperators. One way to identify other cooperators is through “kin discrimination”: if a cooperative behavior is directed toward organisms of sufficient genetic relatedness (kin), those recipients are likely to also contain the genes for the cooperative act .
Cooperation is inherent to multicellularity, and kin discrimination has been reported in a variety of multicellular eukaryotes . Bacteria also exhibit kin discrimination-like behavior, in particular while engaged in swarming, a form of multicellularity in which cells hyperflagellate and group together to rapidly spread across surfaces . Myxococcus xanthus, Proteus mirabilis, and Bacillus subtilis each form visible boundaries between swarms of different strains [5–7]. In B. subtilis these swarm boundaries strongly correlate with relatedness, as strains with less than 99.4% identity at four housekeeping loci are never recognized as kin .
How can cells make distinctions between degrees of relatedness at the molecular level? Each organism seems to have its own unique system, but all fall into two categories. Polymorphic extracellular receptors are the most common mechanism found so far, with examples in Dictyostelium discoideum [8, 9], Saccharomyces cerevisiae , and M. xanthus . These systems use transmembrane proteins that only productively interact with the cognate protein of identical sequence on the neighboring cell. A different approach is taken in the bacterium P. mirabilis, which uses a “Type VI” secretion system (T6SS) to inject effector proteins into the contacted cell [13, 14]. In that case, the identification of kin arises from immunity to the effectors, such that non-kin lacking the proper immunities will be killed. T6SSs and other secretion systems mediate interspecies competition in a wide variety of bacteria , but their function as kin discriminators in multicellularity is much less studied. Similarly, contact-dependent inhibition (CDI) proteins with cognate immunity genes exist in many bacteria  and may be used in multicellular contexts to identify kin , but this has not been explicitly tested.
We sought to expand on our previous work that observed kin discrimination in B. subtilis  by determining the molecular mechanism underlying this behavior. Here we report that B. subtilis cells use a combinatorial kin identification system that includes CDI proteins, a multitude of secreted antimicrobials, cell-surface molecules, and mobile elements such as phages. These numerous factors combine to ensure that unrelated cells do not interact in multicellular contexts.
Our previous work demonstrated that swarms of closely-related B. subtilis strains merged when they met on an agar surface while less-related swarms formed a visible boundary . However, not every strain pair exhibited the same degree of boundary formation. For example, when the commonly used B. subtilis strain NCIB 3610 met the wild strain FENS 2–3–5 the boundary formed was wider and more opaque than the boundary between NCIB 3610 and another wild isolate, CO39 (Figure 1A). This boundary heterogeneity is similar to the variable phenotypes observed in P. mirabilis swarms, which have somewhat different boundary lines between strains .
Because P. mirabilis swarm boundaries are a result of cell killing [13, 19], we tested whether B. subtilis swarms also have a higher number of dead cells where non-kin strains meet. Cells were collected from the boundary region between non-kin swarms and stained with a membrane-impermeable dye (propidium iodide, PI) to identify cells with a compromised membrane. There was a clear increase in the amount of PI-stained dead cells where non-kin swarms met than where kin swarms met (Figure 1B), suggesting that B. subtilis strains may also use an attack and defense strategy to distinguish kin.
To determine how B. subtilis cells identify kin, we undertook a transposon-mediated genetic screen in the NCIB 3610 strain. We plated mutant cells directly onto swarm-inducing medium at a low density to allow each mutant to form a small swarm before encountering the surrounding swarms (Figure 2A). Plates were examined for the creation of a boundary around a single swarm, indicating identification as non-kin (Figure 2A, lower enlarged region). It is therefore a screen for loss of the ability to recognize or be recognized as kin.
We uncovered mutants with transposon insertions in six genes (Figure 2B), four of which were recovered multiple times. The strongest phenotype was observed in the mutant with the transposon inserted into wapA, a gene encoding a CDI protein . The downstream immunity gene, wapI, likely was also disrupted since deletion of wapA alone did not create a boundary with wild type and eliminated the boundary with wapAIΩTn10 (Figure S1A). wapAI mutants have reduced viability when co-cultured with wild-type strains, and the toxin domain of WapA varies among B. subtilis strains , making it an attractive candidate for a kin recognition system. Indeed, swapping the wapAI allele from NCIB 3610 into CO39 resulted in a prominent boundary with wild-type CO39 (Figure S1A). However, there must be additional factors contributing to kin discrimination because some strains have identical wapAI alleles yet still form boundaries between their swarms (such as NCIB 3610, AUSI98, and PS-216; see below), and the CO39 strain with wapAI from NCIB 3610 still formed a boundary with NCIB 3610 (Figure S1A).
Transposon insertions were also found in the tuaD gene, which is in the operon responsible for the biosynthesis of teichuronic acids, an alternative wall teichoic acid lacking phosphate . These molecules on the outer surface of the cell could be directly recognizing neighboring cells, or alternatively could act as a barrier preventing entry of harmful molecules. Indeed, teichoic and teichuronic acids affect the response to antimicrobials and autolysins [22, 23] and are the binding sites for many phages . Thus the tuaD mutant might be connected to the cell death observed in swarm boundaries through the lack of defense against non-kin attack. Changes to the chemical properties of the cell surface could affect the way swarms interact in purely physical ways too.
The last four mutants from our screen contained insertions in regulatory genes: the kinase-phosphatase pair ptkA and ptpZ, and the phosphate starvation response two-component sensor kinase phoR and the phosphate transporter pstBA. Neither the phoRΩTn10 – pstBAΩTn10 nor ptkAΩTn10 – ptpZΩTn10 swarms formed boundaries with each other (Figure S1B), likely indicating each pair is acting through the same pathway. PtkA and PtpZ are known to post-translationally modify TuaD and Ugd, which are involved in the synthesis of cell-surface polymers , and both mutants have biofilm defects . Both ptkAΩTn10 and ptpZΩTn10 swarms still form boundaries with the tuaDΩTn10 swarm however (Figure S1B), indicating their effects are not through TuaD or are pleiotropic. PhoPR also regulates cell-surface genes like tuaD  and a number of antimicrobial molecules  possibly including wapA . However, phoRΩTn10 swarms do not merge with tuaDΩTn10 or wapAΩTn10 swarms (Figure S1B), so the phoR mutant likely acts through many of its target genes.
Because the medium we used to observe the meeting of swarms has low phosphate, we tested whether phoR and pstBA mutants lost their boundary phenotypes on plates with high phosphate. Indeed, phoRΩTn10 and pstBAΩTn10 swarms merged with wild-type swarms in high-phosphate conditions (Figure 2C), whereas boundaries persisted around the other transposon mutants and between different strains (Figure S1C). Thus, physiological changes induced by phosphate starvation can contribute to kin recognition, but are not the sole determinants.
The results from our transposon mutagenesis pointed toward a kin discrimination system that depends on antimicrobial attack and defense, consistent with the increased cell death in swarm boundaries (Figure 1B). Each mutant had a connection to this hypothesis, yet when we tested the mutants against each other to establish complementation groups, the only swarms that merged were the aforementioned phoR – pstBA and ptkA – ptpZ mutants (Figure S1B). This may signify that multiple factors contribute to kin recognition, all of which are necessary to be considered kin.
To determine if our classification and interpretation of the transposon mutants was valid, we tested deletions of candidate genes involved in related processes. Deletion of the sdpABC and sdpIR operons, which encode “cannibalism” toxins and immunities , created a clear boundary with wild type (Figure 3A). Likewise, removal of genes that modify the chemical nature of the cell surface, lytC and dltA, also affected the ability to be recognized as kin by the wild-type parent. These genes’ protein products alter peptidoglycan linkages and attach D-alanine to teichoic acids, respectively [30, 31], which impact susceptibility to various antimicrobials [22, 23].
Lastly, three regulatory mutants involved in antibiotic responses, ΔlytST, ΔyvrHb, and ΔsigW, produced discernible swarm boundaries (Figure 3A). The two-component system LytST is activated by loss of cell wall integrity and regulates expression of antiholin and autolysin genes, among others . YvrHb is a two-component system that falls into both the “attack” and “defense” categories we have proposed, as it regulates expression of the antimicrobials wapA and sunA (sublancin) plus the cell surface modifiers dltA-E and lytABC . SigW is a sigma factor that responds to cell wall stress, and targets genes providing resistance to many antimicrobials, including SdpC .
While these mutations all had clear effects on kin recognition, many other antimicrobial, cell surface, and regulatory gene deletions had no phenotype (see Table S1 for the full list of mutants tested). Because not all antagonistic genes had an effect, this argues that the identified genes have a specific role in kin discrimination, at least in NCIB 3610.
We next used a fluorescent transcriptional reporter of SigW activation (PsigW-yfp) to visually assess whether non-kin activate this stress response pathway. YFP fluorescence was detected throughout each swarm, indicating basal expression of the SigW regulon in the normal course of swarming (Figure 3B). Upon encountering each other, the FENS 2–3–5 strain increased its PsigW transcription substantially more than NCIB 3610, which did not change at all (Figures 3B left panel, S2A, S2B left and right panels). Furthermore, neither strain increased PsigW expression when encountering a kin swarm (Figure S2B top and bottom panels).
Since one of the triggers of SigW is SdpC , we examined PsigW-yfp expression when encountering strains lacking both cannibalism toxins. FENS 2–3–5 ΔsdpA-R ΔskfA-H still increased its PsigW-yfp expression (though much less) when it met NCIB 3610 ΔsdpA-R ΔskfA-H (Figures 3B right panel, S2A). So while these toxins are a large contributor, there must be additional molecules activating SigW. Moreover, NCIB 3610 ΔsdpA-R ΔskfA-H had a YFP signal when it met wild-type NCIB 3610 (Figures 3B top panel, S2A), suggesting its boundary is due to cell wall stress from lack of SdpC detoxification. When FENS 2–3–5 ΔsdpA-R ΔskfA-H met wild-type FENS 2–3–5, however, PsigW-yfp expression did not increase and no boundary formed (Figures 3B bottom panel, S2A). This suggests that FENS 2–3–5 does not produce cannibalism toxins as highly as NCIB 3610, possibly contributing to their designation as non-kin.
After seeing SigW activation at swarm boundaries, we examined the effect of ΔsigW on non-kin recognition. While the boundary formed between NCIB 3610 ΔsigW and wild-type FENS 2–3–5 was unchanged, the FENS 2–3–5 ΔsigW boundary with wild-type NCIB 3610 was noticeably more pronounced (Figure 3C), and further amplified by deleting sigW in NCIB 3610. However, introducing ΔsigW into CO39 did not affect boundary formation (Figure 3C), likely because CO39 does not have the appropriate defense genes for SigW to activate (see gene expression data below). The ability to respond to antimicrobials produced by encountered strains might thus be a key determinant in B. subtilis boundary formation: strains that can mount an appropriate response are considered kin, while those that cannot are killed.
The different effects of ΔsdpA-R ΔskfA-H and ΔsigW in FENS 2–3–5 and NCIB 3610 led us to test other kin recognition mutants in the FENS 2–3–5 background. The wapAI transposon insertion, which had the strongest effect in NCIB 3610, also produced a marked boundary in FENS 2–3–5 (Figure 4A). Of the 11 other mutants, however, only ptpZΩTn10, ΔlytST, and ΔlytC displayed minor boundaries; all others had no phenotype (Figures 4A, S3A). Transposon insertions also had variable effects in the PS-216 strain background (Figure S3B), further highlighting the differences among strains.
We next tested whether any mutations that change the identification of kin to non-kin could also cause the opposite: inappropriate recognition of non-kin as kin, causing swarms of different strains to merge. We used the FENS 2–3–5 strain since its boundary with NCIB 3610 is less stark (see Figure 1A), and thus is likely closer to being recognized as kin. Our attempts using single mutants yielded mostly negative results, with only FENS 2–3–5 phoRΩTn10 somewhat diminishing the boundary with NCIB 3610 (Figure S3C). Even quadruple mutations in wapAI, sdpA-R, skfA-H, and a peptide antibiotic (sunA or bacA) did not affect the boundary (Figure 4B), even though these genes are some of the most variable both in sequence and expression (see transcriptomic and bioinformatic experiments below) [20, 29]. Thus the genes we have identified are necessary to properly recognize kin but not sufficient to remove a “non-kin” designation. This, together with the different phenotypes observed in different strain backgrounds, may indicate that each strain uses multiple mechanisms to distinguish non-kin strains.
To find these alternative mechanisms, we repeated the genetic screen in a different strain background: PS-216, a strain with high genome similarity to NCIB 3610 [34, 35], yet the two form a partial boundary (see Figure 7B below). We recovered one kin recognition mutant that contained an insertion in the epsI gene (Figure 4C), part of the biosynthetic operon that produces the polysaccharide component of the extracellular matrix . epsI contributes to the cell surface and the matrix is known to provide resistance to antibiotics , consistent with our hypothesis that boundary formation is due to antimicrobial susceptibility. As noted with other cell surface genes, though, physicochemical differences in the matrix may provide additional means of strain segregation.
We next tested whether extracellular matrix mutants would have the same effect in other strains. We used ΔepsA-O, ΔtapA-sipW-tasA, and ΔbslA, which make the polysaccharide and the two main protein components of the matrix . In NCIB 3610, these mutants formed only minor and inconsistent boundaries with wild type (Figure 4D) and had no effect on non-kin boundaries (Figure S3D). In CO39, however, deletion of matrix components greatly enlarged their boundaries with NCIB 3610 (Figure 4E), yet only formed moderate boundaries with wild-type CO39 (Figure S3E). Together these results imply that different B. subtilis strains use many different molecules with unique emphases to separate kin from non-kin, yet all tend to fall into the same general categories of antimicrobial attack and defense.
Because our genetic analyses indicated a kin discrimination system with myriad components, we performed a broad gene expression analysis on meeting swarms. We used nanoString nCounter, a probe hybridization-based assay, to quantify the transcript levels of 284 B. subtilis genes. These represent approximately 10% of the operons in the genome, chosen to represent distinct regulons and components of the multicellular lifecycle of B. subtilis, plus the attack and defense categories identified above.
Though the nCounter technique does not survey the entire transcriptome, it requires much less sample than more global techniques. This allowed us to directly sample swarms at either a non-kin boundary (Figure 5A: b, d, f, h) or where two kin swarms met (Figure 5A: a, c, e, g). Ratios between these samples reflect either inherent transcriptome differences between strains (a/c, e/g, c/g ratios) or the changes elicited by non-kin (a/b, c/d, e/f, g/h). For example, to compare the basal transcriptome of NCIB 3610 to FENS 2–3–5, the a/c ratio was taken, while the a/b ratio represents the change in NCIB 3610 when it met FENS 2–3–5. We identified 111 genes with at least two-fold difference in expression between samples (Figure 5B, Data S1). The greatest differences were between strains rather than elicited by non-kin, indicating each strain has a unique transcriptional fingerprint. This may explain why some gene deletions affected strains differently (Figures 3, ,4,4, S3).
The changes in transcription when meeting non-kin were less dramatic but yielded important insights. There was an asymmetry in transcriptional changes at both boundaries examined: one strain drastically altered its expression profile while the other changed far less (compare a/b to c/d, and e/f to g/h in Figure 5B). When NCIB 3610 and FENS 2–3–5 met, NCIB 3610 slightly lowered three genes while FENS 2–3–5 changed expression of 67 genes. These results suggest that (at least in the cases of these non-kin) one strain tends to have a greater effect on the other, provoking many physiological changes but not changing much itself. The strain changing more was different in the two cases, with NCIB 3610 affecting FENS 2–3–5 more but not CO39.
We next looked specifically at “attack and defense” genes involved in antimicrobial production, cell surface modification, and stress response (Figure 5C). All three strains had distinctive antimicrobial expression patterns, including some likely absent in FENS 2–3–5 or CO39 (e.g. sdpC in CO39). Likewise, genes that affect cell surface composition also showed remarkable expression variation between strains and in response to non-kin. The mRNA abundance values obtained for the stress response factor sigW were consistent with our transcriptional reporter (Figure 3B) and knock-out data (Figure 3C): FENS 2–3–5 upregulated sigW upon encountering NCIB 3610 (Figure 5C c/d) but not vice versa (a/b ratio), and CO39—which had no ΔsigW phenotype—did not alter sigW expression at the boundary with NCIB 3610 (g/h ratio). Overall, these data reveal remarkable interstrain transcriptome variation and support our hypothesis that each strain has its own combination of attack and defense factors that discriminate kin. Mutation of many of these highly variable genes gave a kin recognition phenotype (Figures 2, ,3,3, ,4),4), but not all did (Table S1) and none eliminated boundaries between strains (Figures 4, S3).
To further investigate broad genetic differences, we performed a whole-genome alignment of two non-kin strains: 168 (a domesticated version of NCIB 3610 ) and RO-NN-1 (an isolate from the Mojave Desert ) (Figure 6). The genome of 168 was used instead of NCIB 3610 since it is better annotated and they only differ by 41 base pairs and the absence of the pBS32 plasmid in 168 . Additionally, RO-NN-1 and NCIB 3610 form a very clear swarm boundary (Figure S4A).
The two genomes are 84.4% identical overall (3,732,604 nucleotides aligned out of 4,423,862 total) with 97.97% average nucleotide identity (ANI), and show very tight synteny (Figure S4B). However, there are many gaps of non-conservation (Figure 6), which often correspond to insertion of mobile DNA elements like prophages and transposons, as previously observed in comparative genomic hybridization experiments  and other B. subtilis genome alignments . This is consistent with our transcriptional data that detected no mRNA from some prophage genes, including sunA and bsrH, in FENS 2–3–5 and CO39 (Figure 5C, a/c and e/g). Nearly all of the gaps in our alignment contained genes that could contribute to the formation of boundaries between swarms, such as antimicrobial production (“toxin”) and resistance genes, hinting at the genomic nature of non-kin distinction, in particular the role of horizontal gene transfer (HGT).
After seeing the lack of conservation of many antimicrobials, we performed a more directed survey of such genes. We searched for homologs of 23 confirmed antimicrobial genes from NCIB 3610 in all 52 fully sequenced B. subtilis genomes (Figure 7A, see Figure S5 for full table and phylogenetic tree). Genes with sufficient sequence identity (>75% identity across >90% of the gene) were considered true homologs, otherwise they were marked as absent in that strain.
Fewer antimicrobial genes are conserved as strains become more distant (more white boxes on the right side of Figure 7A). Gene type and genomic location have a significant impact on conservation: large biosynthetic operons like bacA-F and srfAA in the core genome tend to be better conserved than toxin-antitoxin systems in mobile elements, for example. Most prophage genes showed poor conservation, especially in the SPβ prophage, which are absent even within the NCIB 3610 clade (Figure S5B). This agrees with the different phage content in the 168 – RO-NN-1 genome alignment (Figure 6), suggesting HGT is a major contributor to antimicrobial gain/loss and thus the evolutionary dynamics of kin groups.
Except for the closest relatives, virtually every strain has a different combination of antimicrobial genes. Since we only looked for genes found in NCIB 3610 and thus excluded those only present in other strains, it is very likely that every non-kin strain has a unique complement of such genes. The differing complements of antimicrobial genes seen here, paired with differences in expression (Figure 5), would provide ample variation for an effective kin discrimination system. Indeed, a previous B. subtilis genomic study looking at intraspecies diversity found that the gene categories we consider “attack and defense” (e.g. cell wall, detoxification, antibiotic production) were significantly diverged among strains . Other categories identified as divergent seem less likely to contribute to kin recognition (e.g. germination, RNA regulation, protein folding).
To test the predictions of these data, we chose several strains of varying relatedness and tested their interaction with NCIB 3610 swarms (Figure 7B). As hypothesized, boundary clarity increased as strain relatedness decreased. The ATCC 6051 strain, which is a descendent of the same parent Marburg strain as NCIB 3610 and only differs by a few nucleotides , did not form a boundary. However, PS-216, which contains all the NCIB 3610 antimicrobials except those in the SPβ prophage, formed a boundary with NCIB 3610 (Figure 7B), though it was blurry and not always apparent. The NCIB 3610 boundary with AUSI98 is clearer than with PS-216 despite the fact that AUSI98 is also only missing the NCIB 3610 antimicrobials found in SPβ, possibly indicating additional AUSI98 genes not present in NCIB 3610. A clear step up is seen in the RO-NN-1 boundary, which only has 11 of the 23 NCIB 3610 antimicrobial genes (Figure 7A) plus a different wapAI allele . The number of shared antimicrobials thus seems a major determinant of the qualitative type of swarm boundary formed. The progression of increasing boundary clarity from ATCC 6051 to RO-NN-1 takes place over a remarkably small phylogenetic span, not even extending out of the subtilis subspecies (Figure S5B). This is consistent with our previous finding that kin/non-kin distinctions are made at an extremely close relatedness cutoff: few kin pairs share less than 99.9% housekeeping gene identity and none have less than 99.5% identity .
Armed with this new information we revisited our previous attempts to eliminate non-kin boundaries, which we hypothesized were unsuccessful because of the many differences between unrelated strains. Here we tested a much more closely related pair: NCIB 3610 and PS-216, whose major difference is the lack of SPβ in the latter  (Figure 7A), and in particular the major lantibiotic sublancin , encoded by the sunA gene. Deletion of sunA in NCIB 3610 virtually eliminated the boundary with PS-216 (Figure 7B), indicating sublancin is the primary factor distinguishing these strains as non-kin. Removing sunA had no visible effect on the stronger boundaries with RO-NN-1 and DV1-B-1, likely because they have many more differences in their antimicrobial makeup. The effect of ΔsunA on AUSI98 was mixed—the boundary became less strong, but was wider due to the NCIB 3610 ΔsunA swarm fading from the border. This could be consistent with our above suggestion that AUSI98 has its own antimicrobial(s) not found in NCIB 3610, and thus removing sublancin is not sufficient to abolish the boundary. These data overall suggest that the antimicrobial milieu produced by each strain is the prime contributor to kin discrimination in multicellular B. subtilis swarms.
Our investigations into the molecular nature of kin recognition in Bacillus subtilis lead us to conclude that this organism uses a multifactorial approach to establish relatedness. Instead of a single polymorphic locus identifying kin groups, B. subtilis genomes contain a multitude of kin discrimination loci (KDLs) that contribute in a combinatorial fashion to specify kin. This hypothesis explains many observations: I) strains form qualitatively different swarm boundaries, presumably due to the number of KDLs in common; II) except for ptkA-ptpZ and phoR-pstBA all KDL mutants fell into different complementation groups, suggestive of many separate genetic pathways; III) no combination of mutations was sufficient to eliminate significant boundaries between strains; IV) KDL mutations had different effects in different strain backgrounds; V) each strain tested had different KDL expression profiles and responses to non-kin, likely due to the particular attack and defense molecules in the encountered strain. A combinatorial system that demanded all KDLs to be identical to productively interact would also explain our previous finding that the kin/non-kin distinction occurs at a very close relatedness cutoff, requiring near-identity (of housekeeping gene sequences) to merge swarms .
This method of kin discrimination has advantages over systems used in other multicellular organisms. The ability to discriminate via secreted molecules instead of depending on physical contact may better prevent random intermixing of non-kin by creating a gradient of exclusion at a distance. Furthermore, a combinatorial system is less likely to misidentify non-kin as kin because it better represents overall relatedness. Single-gene kin identifiers are less stable over time because they are more likely to be invaded by cheater cells that have acquired the KDL but not the cooperative traits it protects . Many multicellular microbes use more than one KDL [8, 9, 12–14, 17], but so far none come close to the plethora used by B. subtilis, which could be an extreme case creating an especially high barrier to cooperation that might even preclude potentially positive interactions.
Although the use of antimicrobials to distinguish kin is not unique to B. subtilis [13, 18], killing non-kin rather than merely excluding them from the multicellular structure has many advantages. We previously observed that sharp antagonism between strains results in an almost complete lack of co-existence on plant roots . This is likely a stronger protection against cheater cells because it is an active suppression of non-kin instead of the passive lack of interaction seen in some organisms [8, 10, 11]. It is estimated that 4–5% of the B. subtilis genome is dedicated to antimicrobial production , so there is a large diverse stock to draw on. Moreover, these genes have additional selective advantages in classical intercellular competition, helping alleviate a potential problem in the maintenance of cooperative traits: the high cost of maintaining KDL diversity can outweigh benefits gained from the cooperative behavior . When KDLs are co-opted from other uses that benefit from genetic variation, however, their selection requirements are reduced. Thus, intense competition among bacteria may have helped facilitate the evolution of cooperation.
The source of B. subtilis KDLs is also of interest: many antimicrobials are inside mobile genetic elements (Figure 7A), and a lot of the differences between strains are due to HGT (Figure 6) . Previous works have noted the importance of HGT in cooperative systems via transfer of the cooperative genes themselves [45, 46], but we suggest transfer of KDLs may be just as impactful. Using HGT as a source of KDLs would create dynamic kin groups that continually update the definition of kin, likely aiding cheater prevention. Periodically purging the old kin group and resetting who is considered kin makes it less likely for a cheater to keep up with the combinatorial kin identifiers before it has a chance to take over the population . Rapid divergence of kin groups was recently seen in experimentally evolved M. xanthus strains , suggesting this might be a previously unappreciated cheater defense utilized by many microbes.
Our results highlight the duality of antibiotic molecules. Microbial natural products are traditionally viewed solely from a competitive perspective. While this is often true it is an incomplete picture, as many are known to serve signaling purposes as well . Even within their role as killing agents, antimicrobials can drive and maintain diversity in a population . We have found that many molecules classified as hostile in nature could also enable cooperation by preventing mixture of less-related strains, ensuring the stability of the multicellular lifecycle over evolutionary timescales. Our data put in a new light the observation that antimicrobials are surprisingly diverse and mostly act on members of the same species , and in some cases depend sharply on intraspecies relatedness . Therefore, what was previously seen strictly as microbial warfare actually promotes a more complex and diverse ecosystem.
A list of strains and extended methods can be found in the Supplemental Experimental Procedures. Swarming assays were done on 0.7% agar B medium at 37° C. Kin recognition screens were performed using transposons on plasmids with temperature-sensitive origins. NanoString nCounter gene expression analysis was done on cells scraped from border regions of three replicate swarm plates, averaging the mRNA abundances between the replicates. Sample ratios greater than two-fold were kept, except in cases of very low transcript abundance or high t-test (Data S1). Genome alignments and synteny were made using MAUVE and LASTZ software. Conservation of antimicrobial genes was assessed by BLASTn searches of B. subtilis genomes.
We would like to thank H. Vlamakis, M. Traxler, P. Beauregard, and J. van Gestel for technical assistance and helpful discussions, A. Earl for isolation of bacterial strains, D. Kearns for a generous gift of TnKRM lysates, and J. Planinc for technical help. This work was supported by the John Templeton Foundational Questions in Evolutionary Biology Program and NIH grant GM58218 (R.K.), the Helen Hay Whitney Foundation (N.A.L.), Slovenian Research Agency Program Grant JP4-116 and ARRS SLO-USA collaboration grant (I.M.M.).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
AUTHOR CONTRIBUTIONSN.A.L., B.K., P.S., I.M.M., and R.K. conceived, performed, and analyzed all experiments. N.A.L. wrote the manuscript with revisions from B.K., P.S., I.M.M., and R.K.