Novel ovine polymorphisms and adaptive evolution in mammalian TLR2 suggest existence of multiple pathogen binding regions

Toll-like receptors initiate in ﬂ ammatory responses following the recognition of a wide repertoire of pathogens includingbacteria,fungi,protozoaandviruses.Theyarecomposedofanextracellularleucine-richrepeatdomain responsible for detecting pathogen-associatedmolecular patterns,amembrane spanningregionand anintracel- lular Toll/Interleukin 1 receptor domain which invokes signal transduction. Toll-like receptor 2 is the most diverse of these receptors as it recognises infectious agents from a range of pathogenic groups. Over 1400 breeds of sheep exist worldwide that inhabit a diverse range of environments, which leads to the potential contact with a wide variety of pathogens likely detected by Toll-like receptor 2. In this study, we evaluated the extent of both long term evolutionary changes, across the mammalian phylogeny of the TLR2 gene, and recent divergence of this same gene in sheep breeds. Evolutionary analyses identi ﬁ ed positive selective pressure across the mammalian phylogeny, and differential selection pressure within the artiodactyl and primate lineage. Finally, we identi ﬁ ed localised positively-selected sites within two regions of the extracellular domain which suggest that multiple binding regions in TLR2 may be involved in pathogen detection. These results are consistent with the hypothesis that competition between host and pathogen is driving adaptation of Toll-like receptor 2 genes. 2014 The Authors.


Introduction
Competition between host and pathogen is a key process in the evolution of species. This was proposed as the Red Queen Hypothesis which states that pathogen interaction with a host results in both the host and the pathogen evolving new strategies to compete with each other (Van Valen, 1973). The competition pressure exerted from both sides results in an evolutionary arms race, where both species evolve in order to equip themselves with mechanisms in an attempt to out-compete the other. This tussle results in evolutionary pressure exerted on competing groups of co-evolving genes (Woolhouse et al., 2002). It is these genes that ultimately develop adaptations and counter adaptations through mutational events which are then fixed by natural selection. The mammalian immune system comprises two parts, the innate and the acquired immune response. Innate immunity is an ancient form of immune defence and provides an immediate response to infection (Murphy et al., 2008). Toll-like receptors (TLRs) are a major group of proteins that form part of innate immunity by recognising molecular components of infectious agents, known as pathogen associated molecular patterns (PAMPs) (Jann et al., 2008). Within pathogens, these molecular motifs are components that are integral to the survival of the organism, such as components of the cell wall and motility structures, thus resulting in them being largely conserved within pathogen genomes (Kimbrell and Beutler, 2001;Macnab, 1992).
To date, 13 TLRs have been described, although it has been shown that not all species contain this full complement of receptors. For example, TLRs 11, 12 and 13 are present in mice but not human whereas TLR10 is present in human and not mice (Hans and Hans, 2011). TLR2 is widely expressed across species and recognises the greatest number of PAMPs, detecting components from bacteria, viruses and fungi (Barbalat et al., 2009;Gay and Gangloff, 2007). TLR2 is predicted to have an expanded range of binding ligand spectrum through its unique ability to form heterodimers with TLR1 or TLR6 (Farhat et al., 2008). Single nucleotide polymorphisms (SNPs) have been identified in the coding sequence of TLR2 of different species (Bhide et al., 2009;Jann et al., 2008;Merx et al., 2007) with certain SNPs demonstrated to alter the immune response to PAMPs (Merx et al., 2007). Mutations in TLR2 have also been associated with increased susceptibility to infection such as Mycobacterium avium subsp. paratuberculosis (MAP) infection in sheep (Bhide et al., 2009), tuberculosis in humans (Ben-Ali et al., 2004) and those caused by the herpes simplex virus (O'Connell, 2007).
There are over 1400 breeds of sheep worldwide (Rischkowsky and Pilling, 2007) that inhabit a diverse range of environments. This leads to potential contact with a wide variety of pathogens, many of which will interact with TLR2 (Entrican et al., 2010;Mun et al., 2003). To investigate potential variation of TLR2 at different evolutionary scales, novel SNPs in sheep TLR2 genes were identified and signatures of positive selection acting across the TLR2 gene were inferred. The distribution of sheep SNPs and the evolutionary imprint across the mammalian phylogeny have identified mutations with the potential to impact upon the immune response to pathogenic invasion.

DNA samples and primer design
Genomic DNA was extracted from peripheral blood mononuclear cells (PBMC) of ten sheep breeds (n = 69). Seven breeds were of varied European origin and three of African origin (provided in the supplementary information). Published primers were used to amplify the coding sequence of TLR2 (Jann et al., 2008) (provided in the supplementary information).

DNA amplification and sequencing
PCR amplification was carried out using the proofreading KOD Hot Start DNA Polymerase (Novagen-TOYOBO, Darmstadt). PCR was performed with 160 ng of template total DNA for each 150 μl reaction with 0.2 mM of each dNTP, 1.5 mM MgSO 4 , 0.6 μM of each primer, 1× KOD buffer and 0.02 U/μl KOD polymerase. Each reaction was subjected to 30 cycles as per manufacturer's instructions using an annealing temperature of 60°C. PCR products were gel purified and DNA was extracted using NucleoSpin Extract II (Macherey-Nagel GmbH & Co.). Samples were then pooled in pairs, by breed, and sequenced using overlapping forward and reverse primers designed with Primer3 (Rozen and Skaletsky, 2000) (provided in the supplementary information).

Sequence analysis and SNP detection
Sequenced fragments were processed and assembled for each breed using the programs Pregap and Gap4 from the Staden sequence analysis package (Staden, 1996). Assembled sequences were trimmed to remove non-coding sequence. Mutations and SNPs were detected using the Contig Editor function and associated chromatograms. Bi-allelic peak heights were identified based on published peak heights for paired pooled samples (Ngamphiw et al., 2008). Only polymorphic sites detected in both forward and reverse reads with a Phred quality score of N 30 were included in this study to ensure high confidence of SNP base calls. A quality score of 30 has the probability of 1/1000 of the base call being incorrect and is defined as a base call of high accuracy (Ewing and Green, 1998;Ngamphiw et al., 2008). Where a polymorphism was detected in a single breed then a base change was only accepted if at least two pooled sequences contained the change to confirm that variation was due to a mutation and not a result of sequencing error.
The consensus sequences for each breed were generated using the Gap4 function. The complete set of breed consensus sequences were imported into Molecular Evolutionary Genetics Analysis software package Version 4 (MEGA4) (Tamura et al., 2007). Ambiguity codes, representing polymorphic sites, were incorporated into consensus sequence for relevant breed to produce a library of all detected genetic variability. Using MEGA4, nucleotide sequences were then converted and translated to amino acid sequences to determine the impact the mutation of SNP had on the corresponding codon i.e. non-synonymous or synonymous mutation and type of amino acid change resulting from the mutational event.

Protein structure modelling
Like other TLR proteins, TLR2 is composed of an extracellular region of leucine-rich repeats. The domain architecture for this region was determined by LRR finder (Offord et al., 2010). Other domains were identified by SMART6 (Letunic et al., 2009) and TMHMM (Krogh et al., 2001). Tertiary structure predictions for sheep TLR2 were generated using Swiss-Model (Arnold et al., 2006). The predicted protein structure of ovine TLR2 was based on the crystal structure of the extracellular domain (ECD) of mouse TLR2 (PDB reference number: 3a7b) and the crystal structure of the human TLR2 TIR signalling domain (PDB reference number: 1fywA). The protein models were then modified using the MolSoft ICM Browser (Abagyan et al., 2006) to incorporate sites under selection and identified mutations.

Tests to measure selective pressure
Genes encoding 44 mammalian TLR2 were obtained from Ensembl and GenBank (provided in the supplementary information). Translated protein sequences were aligned using Muscle (Edgar, 2004) and back translated to obtain a codon alignment. An unrooted phylogenetic tree of the 44 mammalian sequences was created based on an accepted mammalian phylogeny (Murphy et al., 2001). Individual branch lengths were then calculated by CodeML using the M0 codon model in PAML package version 4 (Yang, 2007). This tree was used as the fixed tree topology for all subsequent analysis. To detect positive selection at individual codons within a gene (sites analysis) two pairs of models were compared using codeML: M1a (neutral model) was compared with M2a (adaptive model) (Nielsen and Yang, 1998;Wong et al., 2004) and M7 (beta) was compared with M8 (beta plus omega) (Yang et al., 2000). Statistically-significant evidence of positive selection was inferred by a likelihood ratio test (LRT) comparing 2× the log likelihood difference of each set of nested models. To investigate positive selective pressure within a clade or a specific lineage the branch-sites test was implemented. Analysis was conducted as per previously published methods (Smith et al., 2012). All positions reported herein correspond to Ovine TLR2 (GenBank Accession No: NM_001048231.1).

SNP detection in ovine TLR2
A total of 18 polymorphic sites were detected in sheep TLR2 (Table 1). One SNP, T2008C (F670L) has previously been reported (Bhide et al., 2009) and the remaining are novel. Of the eight non-synonymous (ns) SNPs, four were located in the ECD and four in the TIR signalling domain ( Fig. 1).
When mapped to the protein structure model of TLR2 ECD, three nsSNPs were positioned in the central region, within the known ligand binding and dimerization region ( Fig. 1). SNPs in codon D225A were identified in five of the ten sheep breeds analysed, whilst two SNPs altering residues R315W and V388F were detected in single breeds, Dorper and Suffolk, respectively (Table 1).
The protein structure of TLR2 TIR signalling domain revealed two SNPs around the BB-loop region (Fig. 1). The SNP affecting residue R697C is positioned in the BB loop of the TIR domain and was detected in two sheep breeds, Djallonke and Red Maasai breeds. The previously reported SNP which alters residue F670L is adjacent to the BB loop and was the most common polymorphic site in this data set being detected in eight of the 10 breeds analysed (Table 1 and Fig. 1).

Adaptive evolution in mammalian TLR2
Using a maximum likelihood approach, evidence of positive selection was detected in TLR2 since the divergence of mammals. Statistically significant evidence (posterior probability Pb N 0.95) of positive selection was found for five codons (E88, Q185, R297, L355, Q454) across the mammalian phylogeny. When using the combined branch-site test (see Methods) to compare each clade independently, significant evidence (Pb N 0.95) of positive selection was obtained for the artiodactyl clade (H99, T172), which contains the sheep lineage, but not in the Laurasiatheria clade (which includes the artiodactyls) (Fig. 2). Positive selection was not found to be acting on Euarchontoglires but within the component primate clade two sites were identified as under selection (D38, Y276). These results are summarised in Table 2.
All nine positively-selected sites were positioned within the extracellular domain of the ovine TLR2 (Table 2). Positively-selected sites were not detected in the transmembrane region or cytoplasmic TIR domain of TLR2 (Table 2, Fig. 3). Three sites (Y276, R297, L355) were located around the central region of the ECD involved in bacterial lipopeptide ligand detection and protein-protein heterodimerization interactions with TLR1 and TLR6 (Fig. 3). Five sites (D38, E88, H99, T172, Q185) formed another cluster towards the N terminal region of the protein between LRRNT and LRR6 with one site (Q454) identified within the C-terminal region of the extracellular domain (Fig. 3).

Evidence of adaptive evolution in TLR2 mammalian phylogeny
TLR2 has previously been shown to be subject to adaptive evolution (Areal et al., 2011;Jann et al., 2008;Takaki et al., 2012;Wlasiuk and Nachman, 2010). This current study has taken a maximum likelihood approach to further explore these initial findings by comparison of evolutionary rates between specific clades of an extended mammalian phylogeny.
Specifically, five codons (E88, Q185, R297, L355, Q454) have been discovered as undergoing positive selective pressure since the divergence of the mammals (Table 2). Two specific lineages, the primates and the artiodactyls, display differential patterns of adaptive evolution compared to the rest of the mammals, at codons D38, Y276 and H99, T172, respectively. Whilst our findings support previous studies, in that positive selection is acting across the phylogeny, specific sites under selection differ.
Both Wlasiuk and Nachman (2010) and Takaki et al. (2012) concentrated their investigation on the evolution of TLR2 in the primate phylogeny. Takaki and colleagues focused on site specific divergence and adaptation in TLR2 between two closely related species of macaques. Our analyses involved lineage specific adaptation of the primate phylogeny, using 11 representative species, and reports on the sitespecific adaptive-divergence of primates (D38, Y276) compared to the rest of the mammalian phylogeny. The different approaches taken between phylogenies compared to specific close range species, are likely to account for different positively selected sites within TLR2 being identified across these two studies.
The Wlasiuk and Nachman (2010) study also used 11 representative species and PAML4 to investigate selection of primate TLR2 gene. Their results initially revealed no specific sites under selection using PAML4 evolutionary analysis tool. The contrast in adaptive sites to our results is likely due to the inclusion of our extended mammalian phylogeny which provides the benefit of allowing evolutionary pressure within the primate lineage to be compared to other mammalian lineages. Wlasiuk and Nachman (2010) did however identify specific sites under selection when relaxing the inclusion criteria, posterior probability (Pb) N 90% in PAML4, compared to our analysis which identified sites under selection with Pb N95%. Additionally, they employed the use of the molecular analysis tool, Datamonkey webserver, and also reported on sites under selection when at least two models reported on the same sites. The variation in approaches used can explain specific differences in results between our studies and also results generated in the Areal et al. (2011) study.
The study by Areal et al. (2011), investigated positive selection across the mammalian phylogeny of TLR2, and adopted a similar approach to Wlasiuk and Nachman (2010), in that both PAML and Datamonkey webserver were employed to independently analyse their data sets. PAML4 analysis identified two specific sites under selection across the mammalian phylogeny (296 & 453) (Areal et al., 2011). Since the species reference sequence is not detailed we have made the assumption that this corresponds to (R297 & Q454) in our analyses. The additional sites we identified as under positive selection across the mammalian phylogeny (E88, Q185, L355) are again likely to be due to increased power obtained by the inclusion of an additional 21 species compared to Areal et al. Increasing the number of quality sequences increases the power to detect accurate signatures of positive selection. Use of Datamonkey webserver revealed additional sites of interest. The approach of this package differs to PAML4. For example, the use of the REL tool makes the most assumptions of all the Datamonkey Table 1 SNPs detected in ovine TLR2. 615  674  943  1162  1401 1688  1731 1737  1749  1949 2008  2037 2055 2089  2168 2232  2235 2238   Amino acid position  205  225  315  388  467  563  577  579  583  650  670  679  685  697  723  744  745  746   Domain   Codon wildtype  CTG  GAT  CGG  GTC  AAT  CGC  GAT  CGC  TCT  CGG  TTT  TTT  ATT  CGC  CGT  GCC  GTT  CCC   Codon mutant  CTT  GCT  TGG  TTC  AAC  CAC  GAC  CGG  TCG  CAG  CTT  TTC  ATC  TGC  CAT  webserver models, and is susceptible to high rates of false positives in extreme cases (Pond and Frost, 2005). Jann et al. (2008) analyses used models in PAML4 to determine where variation exists across the mammalian phylogeny but did not determine site-specific adaptation. To allow direct comparison of results, we employed a stringent approach comparable to our previous study of TLR5 genes (Smith et al., 2012).

Comparison of adaptive evolution between mammalian TLR2 and TLR5 genes
We have previously examined evolution of mammalian TLR5 genes (Smith et al., 2012). When compared with the results of this study, a contrasting pattern of selection emerges. In particular, the artiodactyl   (Murphy et al., 2001). Mammals from Africa or an African origin form the clade Afrotheria. Euarchontoglires form the clade of mammals that include rodents, lagomorphs and primates. Laurasiatheria forms a clade of placental mammals including carnivorans, the majority of hoofed mammals, bats, whales, shrews and hedgehogs. Artiodactyls are a clade including even-toed ungulates (information provided by the Tree of Life Project).    (Jin et al., 2007). Positive selected sites under selection are mapped to leucine-rich repeats (LRR). Regions of the protein known to be involved in function are indicated by dashed lines. 1 = Region required to respond to peptidoglycan (PGN) (Mitsuzawa et al., 2001). 2 = Region implicated in response to PGN, Lipoteichoic Acid and Zymosin (Meng et al., 2003). 3 = Region involved in synthetic ligand binding; diacylated lipopeptide ligand, Pam2CSK4 and triacylated lipopeptide ligand, Pam3CSK4, and heterodimerization (Jin et al., 2007;Kang et al., 2009) lineage for TLR5 has evidence of extensive positive selection compared with TLR2. Both receptor phylogenies contain domestic livestock species. However, TLR2 does not exhibit adaptive evolution of genus specific (porcine, ovine and bovine) lineages as was seen for TLR5. This difference may be a reflection of receptor function. Both TLR2 and TLR5 detect components of bacteria but whilst TLR5 is currently thought to act only as a homodimer and specifically detect flagellin, TLR2 shows more versatility. TLR2 is known to form dimers with TLR1 and TLR6 and recognises the greatest number of known ligands of all TLRs. This range of TLR2 increases possible combinations of interacting proteins and hence interacting ligands. Our results indicate that porcine, ovine and bovine lineages are not undergoing species specific adaptive change within the TLR2 gene but possibly utilise the strategy maintaining diversity by use of multiple TLR dimerization and ligand (and hence pathogenic species) recognition. Most proteins have integral regions that are subject to functional constraint and non-integral regions where a relaxation of constraint is observed (Zhang et al., 2011). It is plausible that there is a wider range of essential residues within the TLR2 gene that are involved in dimerization and ligand interactions, compared to TLR5. Should this be the case, then it is expected that there would be a greater number of residues in the ECD of TLR5 that are free to undergo adaptive change, in order to adapt or enhance function, without impacting the current ligand specificity of the protein. The noticeable difference in dimerization and ligand interaction may account for the difference in selection pressure observed between the artiodactyl lineage and these two receptors.

Pattern of positively selected sites across ECD of TLR2
In the ECD, positively-selected sites were mapped to the protein structure of TLR2 (Fig. 3). Sites identified may highlight possible regions of the ECD that are likely to be of biological importance. It is well documented that positively-selected sites within proteins can correlate to ligand binding regions (Ballingall et al., 2010;Furlong and Yang, 2008;Smith et al., 2012). Indeed, our recent publication on mammalian TLR5 identified a cluster of positively-selected sites within the putative ligand binding region of the ECD (Smith et al., 2012) which has been subsequently confirmed, by X-ray crystallography resolution, as a region involved in binding flagellin (Yoon et al., 2012). The first cluster of positively selected sites in TLR2 is positioned within the known region of ligand binding and dimerization with TLR1 and TLR6 (Jin et al., 2007;Kang et al., 2009). The second cluster is positioned towards the N-terminal region. Due to the extensive ligand binding repertoire of TLR2, it has been proposed that multiple binding sites may exist across the ECD (Mitsuzawa et al., 2001). TLR2 LRR-deletion studies implicated the N-terminal region, from LRRNT and up to LRR7, as integral for ligand detection for a variety of ligands including lipoteichoic acid, peptidoglycan and zymosin (Meng et al., 2003;Mitsuzawa et al., 2001). Furthermore, TLR2 LRR-deleted constructs, LRRNT-LRR7, induced an NF-κB response to the synthetic ligands Pam 2 CSK 4 and Pam 3 CSK 4 , indicating that this region is not required to invoke a functional response to these ligands (Meng et al., 2003). To be functional as a signalling molecule TLR2 needs to be relatively conserved, but to function as a protein interacting with multiple ligands there is a need for variability within the ligand binding domain. We propose that this is achieved in TLR2 as selective pressures exerted on different parts of the ECD are driving adaptation through the competition between the host and the pathogens detected by different binding regions. As such, functionally relevant mutations in one ligand binding region may alter ligand specificity, but not necessarily impact upon pathogen recognition of the other region.

Distribution and potential impact of SNPs in ovine TLR2
Previous studies have demonstrated that not all mutations necessarily impact upon the structure or the function of a protein (Bromham and Penny, 2003;Douville et al., 2010;Kim et al., 2007). As such, it is important to identify which sites are likely to affect function.
In this study we correlated mutations to regions of known function and to sites undergoing adaptive change. Of the mutations identified in the ovine data set, five are good candidates with the potential to impact upon the biology of TLR2. Three mutations are situated in the central region involved in PAMP recognition, in close proximity to sites under positive selection. As such, these mutations have the potential to impact ligand detection. Two SNPs (D225A, R315W) result in radical amino acid replacements and one SNP (V388F) is an un-favoured amino acid replacement in extracellular regions of proteins (Betts and Russell, 2003). D225A is detected in five of the ten sheep breeds analysed whilst the remaining two SNPs were breed specific, at least in this data set. R315W was detected in the Djallonke breed and V388F was detected in the Suffolk breed and may convey breed specificity.
Two remaining SNPs of interest are positioned within the TIR signalling domain. Functional constraint was found to be acting on the TIR domain. This was expected due to its role in signalling (Werling et al., 2009). F670L, is a common variant, at least in this data set, and has previously been associated with increased susceptibility to M. avium subsp. paratuberculosis infection in Tsigai sheep and demonstrated to result in decreased cytokine (IL-4, IL-8, IL-10, IL-12 and IFN-y) expression in dendritic cells compared with the wild type form (Bhide et al., 2009). This leads to the expectation that this SNP has the ability to alter the immune response of a wider selection of breeds and to impact a variety of PAMPs detected by ovine TLR2. Finally, R697C was located in two sheep breeds in this study, Djallonke and Red Maasai breeds. This SNP is positioned within the BB loop of the TIR domain, the region involved in dimerization and the interaction with adaptor proteins (Nyman et al., 2008). The importance of the BB loop has been demonstrated in mouse TLR4 where a SNP was found to inhibit the response to lipopolysaccharide (LPS) (Jiang et al., 2006;Poltorak et al., 1998) indicating the potential for the SNP in sheep TLR2 to affect signal transduction. Should this SNP be beneficial to these breeds in host defence then we might surmise that natural selection would act to maintain the mutation and it would spread throughout populations. However, natural selection would also act to remove deleterious mutations from the population, and their removal would require a number of generations. The random incorporation of a deleterious mutation to the host gene could inadvertently offer a competitive advantage to pathogens by altering the immune response of the host to infection.

Recent divergence within a breed diverse species
We combined our analyses of adaptive evolution across the mammalian phylogeny to examine the long term evolutionary impact across the TLR2 gene, compared to the recent, close range divergence occurring within a species by using domestic sheep as an example. At least five haplotypes have been identified in sheep, indicating multiple domestication events (Guo et al., 2005;Hiendleder et al., 2002;Meadows et al., 2007;Pedrosa et al., 2005). The multiple haplotypes within sheep provide a unique opportunity to assess the extent of recent divergence and ongoing adaptation within a species.
It is possible that certain SNPs detected in our data set (G615T, C943T, G1162T, T1401C, G1688A, G2168A) may be the result of a random mutation within a specific breed. If this is the case then the selective breeding process may still result in the increased risk of fixation or removal of a mutation from fragmented or small populations. Should, due to association with a beneficial trait, a mutation with a deleterious impact on infection protection become fixed in a population then this could have adverse effects as it spreads through the wider breed population, conversely beneficial alleles may be lost through random genetic drift. The differential distribution of SNPs may also be the result of the domestication events that gave rise to domestic sheep. Sheep breeds that arose from different domestication events may have acquired varied mutations to other founding haplotypes. Multiple domestication events could account for the sporadic distribution of SNPs identified as breeds that may have acquired different mutations with some SNPs specific to haplotypes. There is a lack in consensus surrounding the origins of domestic sheep (Hiendleder et al., 2002). The rapid diversification of domestic sheep (Ovis aries) breeds is thought to contribute to difficulties in defining taxonomic relationships and the uncertainty in ancestral relationships further complicates the phylogeny (Guo et al., 2005). The genus Ovis includes both wild and domesticated sheep and there is debate surrounding the true ancestors of the genus Ovis (Piper and Ruvinsky, 1997). Three groups of wild sheep are often put forward as candidate ancestors of domestic sheep; mouflon, urial and argali (Hiendleder et al., 2002), thus further complicating the construction of a breed phylogeny. The multiple domestication events in sheep may account for their high genomic diversity compared to other domestic species, such as dogs and cattle (Kijas et al., 2012). Genetic variation detected in the sheep data set indicates that on-going adaptation within the TLR2 gene may be occurring locally, enabling breed/region specific responses to evolve in response to pathogens. Should this be the case then based on our evolutionary analysis of TLR2 it appears that beneficial alleles have not yet sufficiently spread throughout the species to be considered statistically significant adaptive sites. It is worth noting that with over 1400 sheep breeds (Rischkowsky and Pilling, 2007) the data set utilised for this study may not be large enough to provide a representative quantification of TLR2 SNPs across the species.

Conclusion
Of TLRs identified, most detect a narrow range of PAMPs of limited microbial origin. TLR2 is so far unique in that it is known to detect a wide repertoire of PAMPs from a diverse range of pathogens and as such plays a prominent role in innate immunity (Andersen-Nissen et al., 2007;Jin and Lee, 2008;Jin et al., 2007). There is clear evidence that mammalian TLR2 is undergoing adaptation at key regions of the ECD that are involved in pathogen detection. We suggest that multiple binding sites have evolved in TLR2, enabling it to respond to a wide range of pathogenic threats. However, there is a price to pay for this versatility, in that the number of sites able to undergo adaptation may be limited in order to maintain detection of the diverse array of ligands. The cooperation with other innate immune receptors is likely to compensate this cost as they recognise distinct components from the same pathogenic type. Collectively, sets of innate immune genes collaborate to gain a selective advantage in the evolutionary arms race.