Host control and the evolution of cooperation in host microbiomes – Nature.com

Theory: the barriers to cooperation within the microbiome

We focus on a host and its symbiotic microbes - where both sides of the relationship can evolve to invest in traits that provide a fitness benefit to the other (Fig.1a, Methods,Table 1). For example, microbes could invest in production of a vitamin that benefits the host or simply evolve to be benign e.g. a strain that competes with pathogens and refrains itself from breaching the epithelial barrier, even though this restraint reduces its available nutrients. Hosts, meanwhile, might direct carbon towards the symbionts, such as the provision of glycosylated mucins.

a Cartoon of the model: Both hosts and microbiota can invest in cooperation. Host can also invest in host control that preferentially benefits more cooperative symbionts. Microbes migrate into the system at rate M from a fixed environmental pool of largely uncooperative microbes between host generations, and at rate m each symbiont generation within host generations(Methods, Table 1). b Example dynamics from the model. Cooperation evolves when the benefits of cooperation are high, symbiontrelatedness is high (i.e. within-species diversity is low) and the microbiome is short lived (the ratio of symbiont to host generations is 1). Increasing the number of symbiontgenerations within a single host generation (generation ratio)increases symbiont competition within the host and cooperation with the host collapses (unless stated, parameters are x=y=2, R=0.5, f=0.02, g=0.1, m=1106, M=0.05). c Effect of relatedness and benefit to cost ratio of the evolution of cooperation. Cooperation is only stable at high relatedness, high benefit to cost ratio and low generation ratio. Increasing the generation ratio leads to the collapse of cooperation across a wide parameter space.

Each host generation, microbes colonise new hosts from two sources. A proportion M comes from an environmental pool, which has not coevolved with the host and, therefore, has a low baseline level of cooperation. The rest of the microbes (1-M) come from the hosts of the previous generation, based upon their frequency there. If symbionts help their host, this will increase its fitness, and this effect can feedback as a benefit that increases the symbionts genotype in the next host generation (a between-host effect in the terminology of social evolution23,32). Intuitively, so long as the benefits are high and the costs are low, one might predict that cooperation will evolve under these circumstances. If the symbionts, for example, evolve some level of investment in the host, this can incentivise investment by the host in return, which in turn can favour further investment by the symbionts. However, there is a potential problem with this argument. The benefit to helping a host can be countered by competition between symbionts. This effect arises because genotypes that invest their energy in cooperation are expected to, all else being equal, have less energy for survival and reproduction than non-cooperative genotypes in the same host (a within-host effect).

Many microbiomes are relatively open and diverse, which means a focal strain will experience competition from diverse microbial genotypes10. The question of how genetic diversity among social partners influences cooperation is central to evolutionary biology23,33,34, and captured by relatedness, R (Methods)35. Distinct from phylogenetic relatedness, this term in microbes captures the extent to which the genotype of a focal cell predicts the genotypes of all cells in the species under study36. In a simple case, with one strain, the focal cell genotype will predict all cell genotypes and R=1. While, for ten randomly-selected strains, the genotype of any one cell will only predict one in ten of the cells genotypes and R=0.1.

Why is this measure important? Consider when cooperation first emerges as a new symbiont genotype, such that the allele for cooperation is rare. When R=1, if one cell cooperates with the host, all cells will as they are genetically identical, and all will share in the benefits, meaning that cooperation may readily evolve. By contrast, if R=0.1, if one cell cooperates with the host, only one in ten cells will cooperate and yet all will again benefit from the cooperation. The effect is that the other 9/10 cells all get the benefit of cooperation without themselves paying the cost. The cooperative genotype, therefore, is likely to be outcompeted by these other strains. In this case, natural selection may favour symbionts that do not invest in cooperation, but receive any benefits from the cooperation of other symbionts in the microbiota. Over time, this can drive down the cooperation provided by the microbiota so far that the host no longer benefits from investing in the microbiota, and so cooperation is lost on both sides of the relationship.

We can see this effect as we decrease relatedness in the modelequivalent to increasing the number of different strains competing within the hostwith a decrease in the region where cooperation is favoured (Fig.1). Another key factor is the benefit to cost ratio: how much a recipient gains from cooperation relative to the costs of being cooperative. As relatedness is reduced, cooperation only evolves for a relatively high benefit to cost ratio (Fig.1). Relatedness in the model captures the effects of competition between strains i.e. strains within the same niche in a host. However, a system like the human microbiome contains many such niches and many species that fill them. Here, a requirement for a high benefit to cost ratio may present a significant barrier to cooperation. With many species in a host, each symbiont strain is relatively rare and, all else being equal, less able to provide strong benefits for the host. This effect suggests that, in addition to the impact of low relatedness and competition within a given niche (Fig.1), between-species diversity may also limit the evolution of cooperation in microbiomes.

A standard model of cooperation between species, therefore, suggests that systems like the human microbiome may have limited scope for cooperative evolution. However, missing from such models is the potential for there to be many symbiont generations per host generation. For example, one human generation can take ~30y in contrast to symbiotic bacteria estimated to replicate on a timescale of hours37. This means that competition between strains is prolonged and chronic. Introducing this prolonged competition into the model (Methods) causes further problems for the evolution of cooperation (Fig.1). Cooperating symbionts perform particularly poorly under these conditions, because their investment in the host makes them grow more slowly than symbionts that do not cooperate. The effect is to further decrease the likelihood of symbiont cooperation (i.e., at high generation ratios in Fig.1, Supplementary Fig1). This, in turn, disincentivises the host from investing in the symbionts, which leads to a collapse of cooperation between host and microbiota.

This prediction is robust to changes in parameters and modelling assumptions. High generation ratios lead to the collapse of cooperation across broad parameter sweeps of both relatedness and the cost-to-benefit ratio of cooperation (Fig.1c). The shape of the relationship between the investment in cooperation and its benefit can be important in some contexts38,39. We compared a range of functional forms relating symbiont cooperation to host benefit, and found consistently that cooperation collapses at high generation ratios (Supplementary Fig1). Increasing symbiont immigration from the environment (M) to very high levels does generate cooperation. However, this only occurs because we assume a baseline level of cooperation in these immigrants, and this forcing effect on cooperation is again not robust to high generation ratios (Supplementary Fig2).

Where does the human microbiome fit within these parameter sweeps? The available estimates for average symbiont relatedness is relatively high40 but, critically, the generation ratio is extremely high due to human life span being so long relative to that of microbes. These parameters again, therefore, lead to the prediction that cooperation will collapse due to competition within hosts (Supplementary Fig3a).

Our findings fit well with another recent model of host-microbiota evolution, which also concluded that the conditions for cooperation were very limited in systems like the mammalian microbiota26. However, we have so far overlooked the expectation that a host is under strong selection to promote symbiont cooperation10,11,30. Hosts can promote cooperation in a variety of ways, including selective feeding, influencing adhesion to the mucosa, and, of course, via the immune system28,29,30. Animal immune systems, for example, use toll-like receptors (TLRs) to detect conserved microbial features known as microbial associated molecular patterns (MAMPs), such as lipopolysaccharide and flagella. The presence of MAMPs can drive inflammation or other responses that targets and suppresses microbes41. Many of these mechanisms are of course already well known to counter specific pathogens42,43,44. Here, we are interested in their role more broadly in the evolution of a cooperative microbiota.

Our model predicts that allowing host control mechanisms to evolve will often rescue the evolution of cooperation (Fig.2, Supplementary Fig1)25. This prediction fits with a growing body of theory and data in social evolution supporting the importance of control (or enforcement) mechanisms for the evolution of cooperation, including a model of the plant microbiome27,45, When is host control most important for the evolution of cooperation? At low generation ratios, we find that control will only evolve under conditions where relatedness is relatively low. This result fits with classic evolutionary theory46 and occurs because host control is less effective and useful when relatedness is high. At higher generation ratios, the effects of relatedness are weakened by extended competition and evolution within the symbionts, and host control evolves across the whole range of relatedness (Fig.2c).

a Schematic of the model: Both hosts and microbiota can invest in cooperationand, in addition, hosts can invest in control mechanisms that favourmore cooperative symbionts over less cooperative ones. Hosts control alsonegatively effects allsymbionts at cost (f) and hosts pay a direct cost for control (g). b Within-host evolution ofsymbiontcooperation (shown here for the first host generation, as an illustration). Increasing symbiont generations perhost generation (generation ratio)promotes symbiont cooperationwhen there is host control, but hinders cooperationwhen there is not. c Effect of relatedness and benefit to cost ratio on the evolution of cooperation.Cooperation evolves across broad parameter ranges with host control, where increasing the symbiont to host generation ratio only increases the range of conditions where cooperation is stable. The regions where cooperation evolves for hosts and symbiont overlap perfectly and so we show only a single plot for cooperation. d Cooperation collapses when symbionts can evolve cooperation independently of the trait that is thetarget of host control. Mutualism is stable while the trait and cooperation are fixed (original model) but when symbionts are allowed to evolve the trait-cooperation link, cooperation and control are quickly lost. Reinstating the relationship again renders host control effective and restores cooperation. Unless stated, parameters are: x=y=2, f=0.02, g=0.1, m=1106, M=0.05.

At high generation ratios, host control also becomes more effective, because the selection imposed by hosts now acts across many symbiont generations and has a greater impact on genotype frequencies (Fig.2b). Interestingly, this implies that the same property that can undermine cooperation in the microbiota of long-lived hosts (Fig.1b, c) can help to rescue cooperation if there is host control (Fig.2, Supplementary Fig1). Consistent with this, when we again use parameters motivated by the human microbiome, our model predicts that host control can robustly rescue cooperation (Supplementary Fig3b). We also provide parameter sweeps of the costs of host control (Supplementary Fig4), the strength of host control (Supplementary Fig5), and symbiont immigration rates from the environment (Supplementary Fig6). As expected, higher costs of control result in hosts investing less in control at equilibrium. Nevertheless, across all parameter sweeps, the evolution of host control is widely predicted whenever there are a high number of symbiont generations per host generation. The same conclusion is reached when we consider the range of alternative relationships between symbiont cooperation and the benefit to the host (Supplementary Fig1).

An exception to these conclusions occurs when there is no immigration of environmental symbionts, because here host control can collapse. This effect is well-known from previous models of enforcement25,47,48. Without immigration, host control drives all symbiont genotypes to be cooperative. This lack of symbiont variability means host control no longer has a benefit and is lost and with it, cooperation. In reality, there are many sources of symbiont variability, whether it is immigration or mutation, which means that host control is expected to be evolutionarily stable25. For example, in addition to general immigration of environmental genotypes (M in our model), an important source of such variability is the potential for pathogens. To account for this possibility, we developed an individual-based version of our model where we can follow a subset of immigrating genotypes that are especially costly for the host. As expected, including the potential for pathogens only increases natural selection for host control (Supplementary Fig7b). This result underlines the potential for host control mechanisms, and indeed cooperation in the microbiome, to be shaped by pathogens that represent a particularly high risk to a host.

A final consideration is the potential for members of the microbiota to escape from mechanisms of host control. Specifically, natural selection is expected to favour symbionts that reduce their investment in cooperation, while keeping whatever trait the host targets to exert its control. We, therefore, asked what happens if symbiont evolution can alter the link between the trait under host control and their cooperation. Figure2d shows the impacts of this change on evolutionary dynamics. When symbionts are constrained, cooperation and control both rapidly evolve. Indeed, host investment in control is greatest early on because this is when it is most needed to select cooperative symbionts. As symbiont cooperation increases, and symbiont variability decreases, host investment in control drops but to a stable level, which is set by the costs of control (above, Supplementary Fig4).

This all changes when we remove the constraint on symbiont evolution. Now, symbionts rapidly evolve to maintain the trait under host control while reducing investment in cooperation. Host control becomes ineffective because it cannot select for the more cooperative symbionts, and is no longer favoured by natural selection leading to the collapse of cooperation (Fig.2d). Another prediction of the model, therefore, is that cooperation rests upon the evolution of control mechanisms that cannot easily be escaped via counter evolution in the symbionts. This prediction is similar to the idea that the immune system needs to find conserved targets for pathogen recognition44, but here we are considering host control over the microbiota as a whole. As for our earlier results, parameter sweeps confirm that this prediction is robust to changes in relatedness and cost-to-benefit ratios (Supplementary Fig8).

Our model predicts that host control mechanisms have been central to the evolution and maintenance of cooperation within diverse long-lived microbiomes, such as the human microbiome. The potential for host control is clear from the wide variety of mechanisms that can influence the microbiota, including the innate and adaptive immune systems of animals10. However, it is not known whether these mechanisms have been generally important for the evolution of host-associated microbiomes. A challenge for such a broad assessment is that the microbial traits associated with cooperation will typically differ among different host and symbiont species. We, therefore, sought a microbial trait that (i) is widely found and easily identified in genomic data (ii) influences whether symbionts benefit or harm the host and (iii) is subject to strong host control. These criteria led us to bacterial flagella.

Many bacteria possess flagella, which are used to swim and move between microenvironments. Flagella can confer strong benefits to bacteria in a host. Swimming has been shown to help bacteria persist in the mammalian gut49 and, similarly, to escape peristalsis and ejection from the zebra fish gut50. For many pathogens, flagella are also essential for reaching the epithelial layer51,52,53. Due to this latter effect, flagella are important for cooperation and whether bacteria are likely to be beneficial to a host. Specifically, possession of flagella is often associated with harm to the host as a mechanism that allows bacteria to breach the epithelial barrier50,51,52,53,54 In E. coli, for example, only some strains appear to express flagella in the host, and these strains are associated with inflammation and disease54. Consistent with the importance for the host, the key structural component ofbacterial flagella (flagellin)is amongst the most immunogenic of all microbial factors55, with a dedicated receptor in vertebrates (TLR5)56. Mice that lack this receptor have an increase in detectable flagellin in their microbiome57. Conversely, inducing the production of anti-flagellin IgA in mice decreases flagellin levels and limits the encroachment of the microbiota at the epithelial barrier58. Importantly, these experimental studies suggest that host control can limit flagellated bacteria and help in maintaining a cooperative relationship by preventing epithelial encroachment56. However, they leave open the question of how important these processes have been for the evolution of host microbiomes.

We therefore sought evidenceacross animalsthat host control mechanisms have served to suppress flagellated bacteria in spite of the documented benefits of swimming in the host50,51,52,53. We estimated both the frequency of flagellated species and the rate of flagella loss in environmental and host-associated bacteria using a database of 3833 sequenced bacterial strains (1262 host-associated and 2571 environmental)59 (see Materials and Methods) (Fig.3a). Using the software BayesTraits, we assessed transitions between flagellated/non-flagellated and host/environmental bacteria, and fit the data to a simple model where the two traits are independent, and a complex model where rate of change in flagella status was dependant on host association status and vice-versa (Fig.3b). Comparing the likelihood of both models, we can robustly reject the simple model in favour of a complex model where the two traits are dependant (Log Bayes Factor (LogBF)=47.24). We tested for implicit biases in the dataset by performing 100 replicates with random label switching, which produced no significant results (LogBF=42.73).

a 16S phylogeny for strains in the PATRIC representative dataset. We only show Firmicutes here as an example because the full phylogeny is too large to show effectively. Host association was determined using metadata from the PATRIC and BacDive databases. Flagella status was determined by identified conserved motifs of flagellin genes. b Transitions between the four states in the data set, with and the posterior distributions of the transition rates calculated using Bayestraits111. c Posterior distribution of flagella loss rates for host-associated and environmental bacteria. Our model provides evidence for a significant difference in the rate of flagella loss between host-associated bacteria and environmental bacteria. Source data are provided as a Source Data file.

The supported model contains a number of transitions between states that could influence a link between flagella status and host status. To confirm that host association is driving the evolution of flagella loss, we examined the key transition rate from flagellated to non-flagellated bacteria. This analysis revealed that the data support a model where host association is predictive of flagella loss rate (LogBF>2). Moreover, in line with the predicted effect of hosts control, flagella loss rates are higher in host-associated bacteria than in environmental strains (Fig.3c).

The use of flagella by bacteria is associated with breaches of the epithelial barrier and inflammation50,51,52,53,54 and limiting flagella has the potential to improve the cooperativity of the microbiota58. However, in this case, cooperation is the absence of a trait, rather than the presence of a trait that provides benefits to the host, which is a more typical example in the literature. We, therefore, sought a second independent test of the importance of host control, involving a beneficial microbial trait. In the mammalian gut, anaerobic bacteria produce short chain fatty acids, including butyrate, which is considered central to the host-microbiota relationship. Butyrate is a major source of nutrition for the colonic epithelium and is monitored by the immune system (Fig.4a). Butyrate binds to G-protein coupled receptors in host cells, which influences the levels of regulatory T-cells and lowers intestinal inflammation60,61. In addition, butyrate is made by obligate anaerobes and so the maintenance of an anaerobic gut by a mammalian host62 is a second mechanism likely to favour butyrate production.

a Cartoon of butyrate biology: the short chain fatty acid is produced by members of the mammalian microbiome and is a key energy source for the host colonocytes. The anaerobic environment of the gut is favourable to butyrate producing bacteria and is reinforced by metabolism of butyrate by colonocytes, which lowers the oxygen potential in the gut. In addition, butyrate can reduce inflammation via effects on regulatory T cells by binding to G-protein couples receptors (GPCR)60,61, (b) Evolutionary loss rate of a pyruvate to butyrate operon based upon the genomes of the PATRIC database (Methods). c Posterior distribution for butyrate loss rates for symbionts associated with vertebrate hosts against environmental or invertebrate associated hosts. Source data are provided as a Source Data file.

If host control is important, the prediction is that butyrate production will be better maintained (lost less often) in the mammalian microbiome relative to other microbiomes. To test this, we searched the same dataset as above for operons associated with butyrate production63, to study the loss rate of butyrate productionacross bacteria that live in different hosts and environments. Butyrate production may also be importantfor host physiology in vertebratesother than mammals64, and so we first compared loss rates in all vertebrate microbiotas (including mammals) versus all other microbiotas(Fig. 4). We also performed the more stringent test of mammalmicrobiotas versus all others.In both cases, the data support a model where host association and butyrate production are non-independent (LogBF=58.37 for vertebrate analysis, LogBF=45.77 for mammal analysis). Moreover, the loss rate is lower where we predict i.e. lower in vertebrate microbiotas than all others (LogBF=36.17) (Fig.4)and lower in mammalian microbiotas than all others (LogBF=33.42).

The data for both flagella and butyrate metabolism, therefore, are consistent with the prediction that host controlincluding immunological responses to bacterial traitshas influenced microbiome evolution and cooperation. Importantly, both tests could refute our hypothesis and yet both were consistent with our modelling predictions, and the published experimental work showing that the immune system can modulate bacterial traits in the microbiome57,58. However, both tests are also very broad, spanning a wide range of hosts (all animals) and symbionts (all bacteria). As a result, we cannot exclude the possibility that other factors are important in the patterns we observe. We, therefore, sought additional tests of our modelling predictions.

The flagella data set provided such an opportunity. Flagella are targeted by the invertebrate and vertebrate immune systems, but vertebrates show an elaboration of anti-flagella mechanisms. With vertebrates, there was the evolution of TLR5: a dedicated anti-flagellin receptor that mounts both innate and adaptive immune responses56, where the latter responses are absent in invertebrates that lack an adaptive immune system. The evolution of vertebrates is also associated with longer life and so a higher number of symbiont generations per host generation. Our model predicts that both of these effectsstronger host control and increased symbiont generations in a hostwill promote flagella loss (Fig.2, Supplementary Fig5). We compared patterns of flagella loss evolution in vertebrate symbionts relative to invertebrates but this analysis lacked power using our original dataset (PATRIC59). While the trends looked encouraging, there were too few invertebrate species to resolve patterns. We were then fortunate that a new larger dataset was published: the Genomes of Earths Microbiomes, which is a collection of genomes assembled from metagenomic sequences from environmental samples and from a variety of hosts65.

We first used this new data set of 13757 taxa to confirm our original flagella analyses (shown in Fig.3)65. This replicated the results of the PATRIC dataset in both the association of flagella and host-association traits (LogBF=33.61) and even stronger evidence of a difference in the rate of flagella loss between host-associated and environmental bacteria (LogBF=15.02). We next compared patterns in vertebrate vs invertebrate associated bacteria (3333 taxa in total). As predicted, we found a significantly higher flagella loss rate in vertebrate symbionts than invertebrate symbionts (LogBF=6.14) (Supplementary Fig9). This analysis, therefore, is again supportive of the predicted role of host control mechanisms in microbiome evolution.

Whenever a host is able to drive bacteria to lose their flagella, this is likely to be an effective way to promote cooperation because it will limit their ability to reach host tissue50,51,52,53,54. However, there is the possibility that symbionts might evade the immune system without losing their flagella, via modifications that prevent the flagella being detected. Our models predict the need for constraints on such counter evolution in symbionts for host control, and cooperation, to be stable (Fig.2d). We, therefore, explored the potential for counter evolution within the microbiome, as a final test of our modelling predictions. Here, we turned to the key mediator of flagella recognition in vertebrates, TLR5, which binds to flagellin, the main structural component of flagella. Consistent with ongoing host evolution, previous work found evidence that TLR5 is under positive natural selection66,67,68,69,70. For example, there is evidence that a core set of sites in TLR5 are under positive selection across all mammals69, with further residues that are positively selected within particular lineages or species66,68,69. Furthermore, differences in TLR5 are associated with host-specific phenotypes, with different host species responding to flagellins of different bacterial species with varying sensitivity71,72,73.

We looked for evidence that TLR5 evolution has driven comparable changes in the D1 domain of flagellin, which is the key region for TLR5 binding74. We studied the flagellin genes of six symbionts that are typically not pathogenic (Butyrivibrio fibrisolvens, Citrobacter freundii, Clostridium butyricum, Enterobacter cloacae, Escherichia coli, Roseburia intestinalis) and six major pathogens (Burkholderia pseudomallei, Helicobacter pylori, Proteus mirabilis, Pseudomonas aeruginosa, S. typhimurium, and Vibrio cholerae), all found in the human gastrointestinal tract. We included pathogens as we reasoned that evidence of counter evolution is most likely to be found there, and indeed might exclusively occur there, given the evolutionary pressures that hosts exert on pathogens75,76.

We examined flagellins in 1761 strains across our 12 species. In all 11 species which are expected to be recognised by TLR5, the four key residues shown to be important for TLR5 binding (by alanine-scanning mutagenesis74) were extremely highly conserved. Specifically, at these four residues, there was only one change from the consensus sequences (E115 to K115) in one E. coli strain out of a total of 1535 strains across the 11 species, which suggests little or no evolutionary escape from TLR5 recognition (Fig.5a). Across species, one of the four residues (I112 in E. coli) is variable, but only between two similar hydrophobic amino acids (leucine and isoleucine) that are both known to allow TLR5 binding74. The exception that helps prove the rule is H. pylori flagellin which is not recognised by TLR5 and differs from the other species at three of the four key residues77.

a Alignment of the domain of flagellin which TLR5 recognises in symbionts and pathogens. Red bars indicate residues predicted to be in the interface between flagellin and TLR574. Red residues have been identified as important for TLR5 binding by alanine scanning mutagenesis74. As a member of the -proteobacteria, Helicobacter pylori has managed to escape TLR5 recognition and maintain motility by a serious of compensatory mutations74. b Schematic of flagellin alignments for the 12 species tested. Numbers indicate the total number of sequences in the alignment (and the number of unique sequences). Red domains indicate the TLR5 binding region as shown in the above alignment, yellow domains are a second site that also interacts with TLR5(aC-terminal region that also forms part ofthe D1 domain when the protein folds). Episodic positive selection was determined as any site with an LRT>2 and p<0.05 (calculated by MEME, and pervasive positive selection an >1 and p<0.05 calculated by FEL and are represented by +). Lines indicate pervasive negative selection at residues predicted by FEL to have a value of <0.05. For C. freundii, E. cloacae and E. coli variable domains made aligning the full flagellin sequence inaccurate, therefore we focused only on the N-terminalD1 domain, which is the primary binding site for TLR5.

Moreover, in contrast to host evolution in TLR5, we found few examples of positive selection in the TLR5 binding site for two measures of natural selection, across both the commensals and the pathogens (Fig.5). The first measure (FEL)78 assesses pervasive selection i.e. natural selection that is consistent and relatively constant at a given site within the gene of interest. Here, the majority of sites identifiedwere under strong pervasive negative (purifying) selection, which acts to limit evolutionary change. Moreover, all cases of positive selection were outside of the TLR5-binding D1 domain. The second measure (MEME) evaluates evidence for episodic site-specific selection where some alleles experience strong selection while others may not experience any79. This measure identified cases of positive selection across the species, which confirms there is statistical power to detect these sites. However, only three residues were in the D1 domain (two in E. cloacae and one in R. intestinalis) and then always on the very edge of the domain. In summary, we find that the key residues for TLR5 binding are highly conserved, and there is very limited evidence for positive selection in the D1 domain.

The data suggest distinct evolutionary patterns in the host and the microbiota. While host TLR5 appears free to evolve and tune its response to different bacterial flagella, the target of TLR5 in bacteria appears constrained. What drives this constraint? Part of it may be TLR5 itself, if this limits the sequences that bacteria use to those that are not highly immunogenic. However, a key cause is clearly structural. There is a highly conserved molecular interaction between the D1 and D0 domains of flagellin, which is critical to the polymerisation that builds the flagella. The importance of this region for flagella functioning was shown by detailed studies that mutated all residues in the D1 domain80,81. The great majority of residues are required for normal motility, suggesting that bacteria cannot easily change the D1 domain without affecting flagella functioning.

Our modelling predicts that for host control to be evolutionarily stable, it must target constrained bacterial traits that have limited potential for counter evolution, because otherwise bacteria are predicted to evolve to evade control (Fig.2d). In support of this prediction, we find little evidence for functional evolutionary change in the region of flagellin that is targeted by TLR5. As discussed above for the case of H. pylori, the only flagellin where escape from TLR5 detection is documented is that of the - and -Proteobacteria. These groups have a heavily altered TLR5 recognition region that does not illicit a TLR5 mediated immune response77,82. Importantly, to swim, these strains have also accumulated a series of compensatory mutations that maintain the flagella polymerisation and function77. This exception, therefore, is again consistent with there being a significant functional barrier to changes in the D1 region.

Link:

Host control and the evolution of cooperation in host microbiomes - Nature.com