Evidence for positive selection in the TLR9 gene of teleosts

Fish & Shellfish Immunology (2008) 24, 234e242

ava i lab le a t www.sc iencedi rec t .com

j ourna l homepage : www.e lsev ie r . com/ loca te / f s i

Evidence for positive selection in the TLR9gene of teleosts

Johnny Shou-Chung Chen a, Tzi-Yuan Wang b, Tzong-Der Tzeng c,Chun-Yi Wang a, Daryi Wang a,*

a Research Center for Biodiversity, Academia Sinica, Taipei 115, Taiwan, ROCb Genomics Research Center, Academia Sinica, Taipei 115, Taiwan, ROCc College of Liberal Education, Shu-Te University, Kaohsiung 824, Taiwan, ROC

Received 6 September 2007; revised 2 November 2007; accepted 8 November 2007Available online 21 November 2007

KEYWORDSToll-like receptor 9;Positive selection;Molecular evolution;Leucine-rich repeats

* Corresponding author. Tel.: þ8862789 9624.

E-mail address: [email protected]

1050-4648/$ - see front matter ª 200doi:10.1016/j.fsi.2007.11.005

Abstract Toll-like receptors (TLRs) have been identified as key sensors of invading microbesby identifying pathogen-associated molecular patterns and activating innate immuneresponses. Whereas purifying selection has been suggested in mammalian TLR9, evolutionaryfeatures of TLR9 in teleosts have not been investigated in detail. We therefore analysedTLR9 DNA sequences of eight teleost species, including zebrafish (Danio rerio), Japanese floun-der (Paralichthys olivaceus), pufferfish (Takifugu rubripes), and five seabreams. Eleven sitessubjected to positive selection were identified using the codon-substitution models of PAML3.15. Ten of these 11 sites were found to be associated with leucine-rich repeats (LRRs). Sevenof these 10 positively selected sites were associated with the convex surface of the LRR sole-noids, leading to variations of the structures of the LRRs possibly by the introduction of flex-ibility into the LRR solenoids. The positive selection of LRRs in TLR9 may indicate theadaptation of teleosts to different oligodeoxynucleotides present in different bacterialspecies.ª 2007 Elsevier Ltd. All rights reserved.

Introduction

Innate immunity is the body’s first line of defense, and itprovides the earliest detection and response against patho-gens. Unlike acquired immunity, which is found only in ver-tebrates, innate immunity is hardwired into the genome of

2 27893 0159; fax: þ886 2

a.edu.tw (D. Wang).

7 Elsevier Ltd. All rights reserved

all multicellular organisms [1]. Innate immunity recognizeshighly conserved structures such as lipopolysaccharides,polyunmethylated CpG dinucleotides, flagella, doublestranded RNA, and peptidylglycans. These constitutivelyand exclusively expressed structures from bacteria and vi-ruses are also known as pathogen-associated molecular pat-terns (PAMPs) [2e4]. The recognition of PAMPs by the innateimmune system is mediated by a class of pattern recognitionreceptors (PRRs) known as the toll-like receptors (TLRs).TLRs are characterised by a highly conserved intracellularToll-interleukin-like receptor (TIR) domain that activates

.

Positive selection in teleost TLR9 235

downstream pro-inflammatory cascades and a variable num-ber of extracellular leucine-rich repeats (LRR) that directlyrecognize PAMPs [5,6]. Some TLRs (TLR1, 2, 4, 5, and 6) canbe found across the plasma membrane of the cell, and areassociated with PAMPs on the surface of the pathogens,such as peptidylglycans, lipopolysaccharides, and flagella.Other TLRs (TLR3, 7, 8, and 9) can be found stretching acrossthe membranes of intracellular compartments. These TLRsrecognize PAMPs that are associated with nucleic acids,such as CpG polyunmethylated oligodeoxynucleotides anddouble stranded RNAs, requiring intracellular digestion ofthe pathogen prior to recognition [7].

TLRs were first recognized to play a major role in thedetermination of the dorsal and ventral axes in the earlydevelopment of Drosophila melanogaster embryos [8]. Itwas revealed that TLRs are also involved in the recognitionof peptidylglycans of Gram-positive bacteria and fungi[2,9,10]. Homologous genes of TLRs were later discoveredin vertebrates, including mammals, birds, and fish [11,12].More than 10 TLRs have been identified in vertebrates [2].However, although a comparison between the sequencesof Drosophila and vertebrate TLRs revealed striking se-quence conservation [1e3,11,13], only vertebrate TLRsmay recognize PAMPs directly. Insect TLRs, in contrast, re-quire a separate receptor such as Gram-negative bindingprotein (GNBP) and peptidoglycan receptor protein (PGRP)to bind to the PAMPs [2,4,11,14]. These findings indicateddifferent characters between vertebrate and invertebrateTLR systems and are also demonstrated by the separateclustering of TLR9 and related genes between vertebrateand invertebrate systems [15].

TLRs of vertebrates were known to be characterised bythe highly conserved nature of PAMPs [2,4,16,17]. Roachet al. [16] studied the evolutionary patterns of all verte-brate TLRs and confirmed the notion that purifying selec-tion, which is characterised by higher rates of synonymousthan non-synonymous substitutions, plays a major role inthe evolution of TLRs. Modlin [18] suggested that mamma-lian TLRs were also affected by purifying selection in a re-view of all available mammalian TLRs. In addition to TLRs,the TLR pathways were found to be highly conserved inteleost fish due to various mechanisms [16,17]. More specif-ically, in investigations of TLR5 and its ligand, it has beenshown that the recognition site on TLR5 and the recognitionsurface on the bacterial flagella are both highly conserved[19e21]. However, recent investigations suggested positiveselection of at least two of the TLRs in bovine and aviansystems [22,23]. In this case, the non-synonymous substitu-tion rate in TLR is higher than that of the synonymous sub-stitution, suggesting a fast evolving process. Furthermore,CpG polyunmethylated oligodeoxynucleotides may not beas invariant as once thought due to species-specific varia-tion in unmethylated CpG oligodeoxynucleotide sequences[24e28]. Neujahr et al. [28] tested and discovered markeddifferences in the immunostimulatory effect of CpG oligo-deoxynucleotides from various bacterial species to murinespleen cells. A recent study by Dalpke et al. [29] suggestedthat different bacteria that have varied sequences of CpGoligodeoxynucleotide repeats have varying degrees of ca-pacity to activate TLR9, the receptors of CpG oligodeoxynu-cleotides [25,30]. These observations have raised questionsregarding the conservation of TLRs [22,23,29].

Although there was a renewed interest in studying teleostimmunity, most research in the field has been functional innature, lacking evolutionary aspects of the TLR proteinfamily [16,31e33]. This is important because teleosts havea less well developed acquired immunity [33e35]. Teleostorthologues of mammalian TLR9 have been identified fromthe genomes of the Japanese pufferfish, Takifugu rubripes(Temminck and Schlegel), and the zebrafish, Danio rerio(Hamilton) [16,36,37]. TLR9s of a few other teleost species,such as the Japanese flounder Paralichthys olivaceus (Tem-minck and Schlegel), and the seabream (Sparus aurata L.)are also available in GenBank at the National Center for Bio-technology Information (NCBI) [26,27,38]. In order to under-stand the evolutionary patterns of TLR9s in teleosts, wesequenced the TLR9 gene from four additional seabreams,including Acanthopagrus berda (Forsskal), Acanthopagrusschlegelii (Bleeker), Dentex tumifrons (Temminck andSchlegel), and Pagrus major (Temminck and Schlegel),from primers based on the TLR9 sequences of S. aurata [38].

These new data were added to the currently availablesequences of TLR9, and a comparison was made of thepatterns and significance of evolution among all eight of thepiscine TLR9s. The patterns of selection at the individualamino acid level were analysed by utilising the codon-substitution models of PAML [39], and evidence of positiveselection was found.

Materials and methods

DNA extraction, PCR amplification and sequencing

The experimental fish A. berda, A. schlegelii, D. tumifrons,and P. major were obtained from local markets in Taipei, Tai-wan. Genomic DNA was extracted from 2 ml of blood using thePuregene Genomic DNA Isolation Kit (Gentra Systems, Minne-apolis, MN), following instructions from the manufacturers.

Primers were designed based on the genomic sequencesof S. aurata (accession no. AY751798; Table 1). Partialgenes were amplified by each pair of primers (Table 1).DNAs were amplified using Ampliqon III PCR 2X PolymeraseKit (Ampliqon III, Bie and Bernsen A-S, Roedovre, Denmark)in 50 ml reactions following the manufacturer’s recommen-ded reaction concentrations with 2 mg of genomic DNA astemplate. Each thermocycling reaction was run for 39cycles of 95 �C for 40 s, 56 �C for 30 s, and 72 �C for 3 minusing a MJ Research PTC 200 Peltier Thermal Cycler (MJResearch, Waltham, MA). PCR products were either cleanedor gel-extracted using the Puregene DNA Purification Kit(Biokit, Miaoli, Taiwan) for sequencing. Sequencing wasperformed with the Sequenase PCR Product SequencingKit (United States Biochemical Corp., Cleveland, OH). Com-mercial sequence kits (BigDye� Terminator Cycle Sequenc-ing Ready Reaction Kits of Applied Biosystems, Foster City,CA) and sequence data were generated using ABI model 377automated DNA sequencers. Approximately 4 kb of partialsequence encompassing all of the codons was sequenced.

Sequence analysis

Open reading frames (ORF) from the genomic DNA sequenceswere extracted from the alignment using the known cDNA

Table 1 Primers used in PCR reaction for the TLR9 gene

Primername

Sequence 50e30 Primerstart sitea

FTLR-1F AATGTTTCCTCTCCCGACCT 8FTLR-1R GCCCTTATGTCCTGCCCAAA 3312FTLR-2F CTTGAGCCCATTGTGGATTT 44FTLR-2R GCCAGGAGGTTGTCAGAGAG 1601FTLR-3F GCAAATCTGTGGGGAGCACT 574FTLR-3R GGATGTTTTGGGAGAGGTCA 1912FTLR-4F CATCCTTTGCCAGCTTCTTC 706FTLR-4R GGAATCCAGTCCCTCTCCTC 3476FTLR-5F GGCAACTTTTTCCACATCGT 1769FTLR-6F TGTAGGCATGGAAAATGCTG 2218FTLR-6R GCAGGGTGAGTTTCTGWAGG 3016FTLR-7F TTCCATGGCAAAACATCTCA 2751FTLR-7R GGAATCCAGTCCCTCTCCTC 3482FTLR-8F GGCTCTCCTATTTGCCTCCT 2811FTLR-8R TGGAATGACAACATTTTGGA 4293FTLR-9F CTGTCCTGCCCCTACTGAAG 3246FTLR-9R CCCATGTAGCTGTCATTGTGA 4055TLR-10F CTCTACGGCTGGGATCTTTGGTA 3269TLR-10R GCCGCTGCTGGACCATGAAGAA 3620

a The genomic sequences of S. aurata TLR9 was used asa reference.

236 J.S.-C. Chen et al.

sequences of S. aurata TLR9 (accession no. AY751797). TheTLR9 sequences of three other teleosts, including the Japa-nese flounder, P. olivaceus (accession no. AB234024; [26]),zebrafish, D. rerio (accession no. XM_685911; [36]), andthe pufferfish, T. rubripes (accession no. AC156439; [37]),were downloaded from the GenBank. To construct the phy-logenic tree, the ORF sequences were aligned using ClustalWin MEGA 3.1 [40,41]. Modeltest v. 3.7 [42] was used to selectthe substitution model that best describes the data. Phylo-genetic analysis for the TLR9 gene was performed usingthe neighbor-joining (NJ) method [43] with the GTRþ G(G Z 0.9362) model as implemented in MEGA 3.1. The ro-bustness of statistical support for the tree branch was deter-mined by 1000 bootstrap replicates. The protein domains ofthe amino acid sequences were characterised using theSMART program; (http://smart.embl-heidelberg.de/; [44]).Secondary structures of the proteins were predicted usingSSPro [45] using default settings.

In order to estimate the selective constraints on theTLR9 gene, the ratio of non-synonymous and synonymoussubstitutions (dN/dS), also known as u, was calculated usingthe NeieGojobori method with JukeseCantor correction[46]. A u> 1 is indicative of positive selection whilea u< 1 is indicative of purifying selection. The codeml

Figure 1 Amino acid alignment of TLR9 genes of D. rerio, P. olivaceand S. aurata. Sequence homologies are represented by dots. LRRthick underlines. Signal peptides are indicated by long dashed unare indicated by double underlines and the canonical CXXC motif ofacid insertions between the two CXXC motifs are indicated by waveamino acid of the LRR. Residues that are dot-dashed underlined areboxed. Shaded residues correspond to positively selected sites.

program of the PAML package [39] was utilised to calculatethe site to site variation in u. As recommended by Yanget al. [47], the following codon-substitution models wereutilised: M0 (one ratio), as one ratio is assumed in all sites[48] and M3 (discrete) has discrete classes of sites for eachbranch in the phylogeny with different u ratios. M7 (beta)and M8 (beta & u), assuming beta distribution, were usedto study variation among sites. In this case, M7 assumed u

to vary between zero and one while M8 has an additionalclass of sites so that u> 1. M1 (neutral) analysis was not in-cluded due to its tendency to fail to account for the sub-stantial proportions of sites with 0< u< 1. M2 (positiveselection) analysis was not included due to its high fre-quency of failure to detect positive selection identified byother models [47]. All sequences were manually adjustedto remove sequence gaps and regions of misalignment priorto PAML analysis. Bayesian approaches were used to esti-mate the significance of dN/dS differences [49]. Likelihoodratio tests were also implemented to verify the positive se-lection of each model by comparing M0 v. M3 and M7 v. M8[47,48].

Results

Sequences of the seabream TLR9 gene

The nucleotide sequences of the gene from each of thefour newly sequenced TLR9s, including the three exonsand the two introns, have been deposited into theGenBank. The accession numbers include A. berda(EU256332), A. schlegelii (EU256333), D. tumifrons(EU256335), and P. major (EU256334). The amino acid se-quences of the TLR9 are presented (Fig. 1). All five of theseabream TLR9s have the same ORF size at 3192 bp codingfor 1063 amino acid residues. When translated, A. berdaand A. schlegelii have 15 and 16 LRRs, respectively, whileD. tumifrons and P. major have 17 LRRs. All five seabreamTLR9s have their leucine-rich repeat C-terminus at aminoacid positions 791e842 (Fig. 1). As reported by Franchet al. [38], all five of the seabream TLR9s have the first25 amino acid residues as the signal peptide and aminoacid residues at positions 844e866 as the transmembranedomain (Fig. 1). The Toll-interleukin receptor (TIR), how-ever, is located at different positions than S. aurata[38], although all four of the sequenced TLR9s have theTIR in the same position (Fig. 1). The sequences in thepredicted binding pocket between the two CXXC motifsare highly conserved, with P. olivaceus sharing the exactsame sequences as the five seabreams while T. rubripesand D. rerio differ by only one and two amino acids, re-spectively (Fig. 1).

us, T. rubripes, A. berda, A. schlegelii, D. tumifrons, P. major,s are underlined. Leucine-rich repeat C-termini are shown by

derlines. Flanking regions homologous to LRR8 of human TLR9that LRR is indicated by short dashed underlines. The six amino

d underlines. Strikethroughs denote other insertions at the 12thin the intermembrane region. The TIR region of each species is

Positive selection in teleost TLR9 237

Table 2 Likelihood scores and parameter estimates for various hypotheses

Model ln L dN/dS

(average u)Parameters Df 2Dl Significance

M0 (one ratio) �12 371.672 0.225 u Z 0.225M3 (discrete) �12 067.335 0.342 p1 Z 0.495, p2 Z 0.41, p3 Z 0.0951,

u1 Z 0.024, u2 Z 0.420, u3 Z 1.69

4 608.674 p< 0.0001

M7 (beta) �12 073.354 0.301 p Z 0.286, q Z 0.664M8 (beta & u) �12 069.137 0.342 p Z 0.32, q Z 0.977, p0 Z 0.948,

p1 Z 0.0525,u Z 1.953

2 8.433 p< 0.05

2Dl: Log likelihood difference between models using the c2-test; u: ratio of dN/dS; p: number of free parameters for the u ratios.Parameters indicating positive selection are in bold.

238 J.S.-C. Chen et al.

PAML analysis

The results of the PAML analysis and the subsequentlikelihood ratio tests suggested the presence of positivelyselected sites for M3 and M8 (Table 2). The presence of sitessubjected to positive selection was suggested by the signif-icantly better fit of M3 than M0 to the dataset. Eleven siteswith posterior possibility of at least 0.90 were detectedusing the native empirical Bayes approach in M3 becauseBayes empirical Bayes (BEB) is not implemented in thismodel (Table 3; [49]). Of these 11 sites, five have a posteriorpossibility of greater than 0.95 (Table 3). Although modelM8, with u Z 1.953, fits the data significantly better thanneutral expectations, no site was found to be under signif-icant positive selection (Table 2).

In the case of sites subjected to positive selection, thePAML analysis revealed that 10 of 11 sites (91%) areassociated with LRRs (Fig. 1; Table 3), important active sitesthat interact with the CpG unmethylated DNA oligodeoxynu-cleotides [25]. All five positively selected sites that havea posterior possibility of >0.95 are located in LRRs. In allthe examined fishes, the remaining one of the 11 positivelyselected sites is in regions of uncharacterised domain(Fig. 1).

To further document the positive selection of thesesites, a quantileequantile plot (qeq plot) was constructedby comparing the distribution of posterior probability

Table 3 Significant sites in across species test using M3 of the

D. rerio P. olivaceus T. rubripes S. aurata A. schleg

125* 117* 108* 119 119*272* 265* 255* 266* 266*380* 373* 363 374* 374384* 377* 367* 378* 378

492 500 482 498* 498*593 601 583 599 599686* 694 676 692* 692*722* 732* 714* 730* 730*732* 742* 724* 740* 740*742* 752* 734* 750* 750*745* 755* 737* 753* 753*

Asterisks mark the sites located within LRRs. Regular type: p> 0.9; b

values of all of the codons against the normal distributionpredicted from these values (Fig. 2; [50]). If the sites wereunder purifying selection, then the distribution would berandom and be no different than a normal distribution.However, if the sites were positively selected, then thedistribution would be significantly different from a normaldistribution. The results from the ShapiroeWilk test(W Z 0.534, p< 0.0001) suggested that the distribution ofthe posterior possibility values is not random, suggestingpositively selected sites.

The positively selected sites on LRRs

Various methods have been proposed for the LRRs torecognize their ligands [26,30,51e53]. The recognitionmay be based on insertions in positions 12 and 17 of theLRRs, as well as insertions between two CXXC motifs[26,30]. Alternatively, it may be based either on the varia-tions of the amino acids in the solvent-exposed beta sheets[52,53] or on the variations of the structure [51,54]. Wefound that all of the positively selected sites are on theflanking sequences (Supplementary Fig. 1; Fig. 3). Becauseit is known that the extracellular domain (ECD) of TLRsforms a horseshoe-like solenoid structure [52,55] and ithas been predicted that the convex side of the solenoid isimportant for ligand binding [30], the relative locations ofeach of the LRR-associated positively selected sites were

substitution models implemented in PAML 3.15

elii A. berda D. tumifrons P. major Posteriorprobability

119* 119* 119* 0.968

266* 266* 266* 0.902374 374 374 0.92378 378 378 0.963

498* 498* 498* 0.966

599 599 599 0.906692* 692* 692* 0.914730* 730* 730* 0.949740* 740* 740* 0.969

750* 750* 750* 0.903753* 753* 753* 0.961

old type: p> 0.95.

Figure 2 Quantileequantile (qeq) plot of the posteriorpossibility values of all of the codons. The Y-axis is the normaldistribution based on the posterior possibility values and the X-axis is the observed distribution of the posterior possibilityvalue. The dashed diagonal line is the 1:1 equivalence line be-tween the normal distribution and the posterior possibilityvalues.

Positive selection in teleost TLR9 239

investigated. We found that three of the positively selectedsites are located on the concave beta sheet of the solenoidwhile the remaining seven sites are found on the convexsurface (Supplementary Fig. 1; Fig. 3). With regard to theLRR homologous to the LRR8 of human TLR9, none of theinserted amino acids was subjected to positive selection(Figs. 1, 3) and the positively selected site is also on the

Figure 3 Alignment of LRR homologous to LRR8 of the humanTLR9. On the LRR consensus motif, L denotes the amino acidleucine. F denotes phenylalanine, X denotes any amino acid,F indicates any hydrophobic residue, double underlines indi-cate amino acids on the concave side of the solenoid, thick un-derlines denote amino acids on the convex surface of thesolenoid, and dashed underlines indicate insertions. On theLRRs of the TLR9, lower case letters indicate positively se-lected sites. Underlined letters indicate coils. Boxed letters in-dicate helix. Shaded letters indicate amino acids that make upbeta sheets.

flanking convex surface of the solenoid and in the coils(Fig. 3). In summary, none of the positively selected sitesis located on the solvent-exposed beta sheet. The positivelyselected sites are rather on the structural components ofthe LRRs, a majority of which are on the coils.

Discussion

The conservation of TLR9 sequences

The four newly sequenced TLR9s from A. berda, A. schlege-lii, D. tumifrons, and P. major share similarities in structurewith the previously established TLR9 of S. aurata. Thesefour TLR9s have their signal peptides on the same locationsas the S. aurata. However, these four seabream TLR9s havefewer LRRs than S. aurata, although all five seabream TLR9spossess the homologous region of LRR 8 of the human TLR9.All of the fish TLR9s also share the leucine-rich repeatC-terminus (LRRCT). All eight TLR9s share the transmem-brane region. The intracellular TIR component of TLRs isknown to be highly conserved and this is reflected in thepiscine TLR9s here (Fig. 1; [56,57]). The highly conservedTIR component is presumably used for binding of MyD88 toactivate the subsequent inflammatory pathway [16,38,56].

Selection patterns of teleost TLR9

TLRs are generally perceived to undergo purifying selectiondue to the highly conserved nature of the PAMPs [2]. How-ever, the findings of this paper form one of the earliestpieces of evidence for the positive selection of this genefamily [11,22,38,56,58,59]. PAML model M3 revealed 11positively selected sites, 10 of which are associated withLRRs, the active sites of the TLRs [52]. These results suggestpositive selection of LRRs in the teleosts, which is consis-tent with the findings of Yilmaz et al. [22] who found posi-tive selection along the length of LRRs in the TLR1 ofchickens. Whereas TLR1 may have benefited from formingheterodimers with TLR2 and TLR6 for recognizing a widervariety of pathogens, leading to positive selection [22],TLR9 does not form heterodimers as they interact withthe CpG oligodeoxynucleotides directly [5,60]. One possi-bility for the positive selection of piscine TLR9s is thatthe LRRs of TLR9 may have higher rates in non-synonymousthan in synonymous nucleotide substitutions, in order tocompensate for the fast evolving rates that have beenfound in bacterial CpG oligodeoxynucleotides [25]. Higherrates of dN/dS have been found in the LRRs of other pro-teins, including the leucine-rich repeat receptor-like ki-nases of Arabidopsis thaliana [61], and rice [62], both ofwhich are also involved in pathogen recognition.

Bauer et al. [25] demonstrated that CpG DNAs ex-tracted from different bacteria differ in their potentialto activate TLR9. Whereas TLR9 in humans recognizesGTCGTT optimally, the optimal murine recognition se-quence is GACGTT. The differential sensitivity to CpG oli-godeoxynucleotide sequences was also demonstrated inJapanese flounders, as their TLR9 was sensitive to CpG-oli-godeoxynucleotide 1668 instead of CpG-oligodeoxynucleo-tide 1720 [26]. Our discovery of positive selection in

Supplementary material

Supplementary material associated with this article canbe found, in the online version, at doi:10.1016/j.fsi.2007.11.005.

240 J.S.-C. Chen et al.

piscine TLR9 may therefore suggest that the differentTLR9 in the respective fish species may be responding todifferent viral and bacterial pathogens in their environ-ment that have different unmethylated CpG oligodeoxynu-cleotide sequences.

Patterns of positive selection on theleucine-rich repeats

The crystallized structures of LRRs of TLR have been de-scribed as horseshoe-like solenoids [55]. The solenoid-likestructure of LRRs of TLR is similar to the LRRs of many othercrystallized proteins [52]. The LRR solenoids contain a con-cave b-face and a convex surface [30,52]. Our results sug-gest that the LRRs of TLR9 bear positively selected sitesin the flanking sequences, with a majority (seven of 10)on the convex surface of the TLR solenoid. The LRR thatcontains the proposed binding pocket is indicated by the in-sertion between the two CXXC motifs [26,30]. This bindingpocket has a positively selected site in all of the teleostsinvestigated. The positively selected site is not found inthe six amino acid insertions between the two canonicalCXXC motifs, but on the convex surface of the LRR. Thefinding here agrees with the investigation of Bell et al.[30] who suggested that such variations may be importantto cause flexibilities of the LRR solenoid, leading to se-quence-specific recognition of CpG oligodeoxynucleotides[51]. Together, our observations suggest that these posi-tively selected sites in the convex surface may possiblyinfluence the flexibility of the LRR solenoid. However, be-cause none of the positively selected sites was found onthe interstitial codons of the solvent-exposed beta sheet,it is contrary to the investigations of Kobe and Deisenhofer[53,63e66].

In a previous study of TLRs in vertebrates, Roach et al.[12] did not find any evidence of positive selection in thethree representative TLR families (TLR3, TLR5, andTLR11) using the codeml program in the PAML package[39]. Contrary to their results, we were able to detect pos-itive selection in TLR9, a member of the TLR gene family, inteleosts. The absence of positive selection in the study byRoach et al. [16] could be attributed to the difficulty insequence alignment resulting from sequence gaps in align-ment with distantly related organisms [67,68]. Partly due tothe advantage of utilising closely related species, we wereable to document positive selection of the piscine TLR9gene. In conclusion, patterns of selection suggest that pos-itive selection acts on the structural components, not theinteractive sites of the TLR9. This observation is consistentand serves as an explanation to the findings that variationsin structure are important to recognize different CpG meth-ylation patterns in bacteria [25,26,51]. The positive selec-tion of LRRs in TLR9 may ultimately aid the differentspecies of fish to adapt to different CpG oligodeoxynucleo-tides present in different bacterial species.

Acknowledgements

We thank Dr. Yun-huei Tzeng for his valuable comments. Thiswork was supported by research grants from the National

Science Council, Taiwan (96-2621-B-001-008-MY3) and Re-search Center for Biodiversity, Academia Sinica to D.W.

References

[1] Janeway Jr CA, Medzhitov R. Innate immune recognition.Annu Rev Immunol 2002;20:197e216.

[2] Horng T, Barton GM, Medzhitov R. TIRAP: an adapter moleculein the Toll signaling pathway. Nat Immunol 2001;2:835e41.

[3] Medzhitov R, Janeway Jr CA. Innate immunity: impact on theadaptive immune response. Curr Opin Immunol 1997;9:4e9.

[4] Medzhitov R, Janeway Jr CA. Innate immunity: the virtues ofa nonclonal system of recognition. Cell 1997;91:295e8.

[5] Rutz M, Metzger J, Gellert T, Luppa P, Lipford GB, Wagner H,et al. Toll-like receptor 9 binds single-stranded CpG-DNA ina sequence- and pH-dependent manner. Eur J Immunol2004;34:2541e50.

[6] Takeda K, Akira S. Toll receptors and pathogen resistance.Cell Microbiol 2003;5:143e53.

[7] Akira S, Uematsu S, Takeuchi O. Pathogen recognition andinnate immunity. Cell 2006;124:783e801.

[8] Hashimoto C, Hudson KL, Anderson KV. The Toll gene ofDrosophila, required for dorsal-ventral embryonic polarity,appears to encode a transmembrane protein. Cell 1988;52:269e79.

[9] Belvin MP, Anderson KV. A conserved signaling pathway: theDrosophila toll-dorsal pathway. Annu Rev Cell Dev Biol 1996;12:393e416.

[10] Lemaitre B, Nicolas E, Michaut L, Reichhart JM, Hoffmann JA.The dorsoventral regulatory gene cassette spatzle/Toll/cactus controls the potent antifungal response in Drosophilaadults. Cell 1996;86:973e83.

[11] Medzhitov R, Preston-Hurlburt P, Janeway Jr CA. A humanhomologue of the Drosophila Toll protein signals activationof adaptive immunity. Nature 1997;388:394e7.

[12] Roach JC, Glusman G, Rowen L, Kaur A, Purcell MK, Smith KD,et al. The evolution of vertebrate Toll-like receptors. ProcNatl Acad Sci U S A 2005;102:9577e82.

[13] Mushegian A, Medzhitov R. Evolutionary perspective on innateimmune recognition. J Cell Biol 2001;155:705e10.

[14] Alexopoulou L, Holt AC, Medzhitov R, Flavell RA. Recognitionof double-stranded RNA and activation of NF-kappaB by Toll-like receptor 3. Nature 2001;413:732e8.

[15] Luo C, Zheng L. Independent evolution of Toll and relatedgenes in insects and mammals. Immunogenetics 2000;51:92e8.

[16] Purcell MK, Smith KD, Hood L, Winton JR, Roach JC. Conserva-tion of Toll-like receptor signaling pathways in teleost fish.Comp Biochem Physiol Part D Genomics Proteomics 2006;1:77e88.

[17] Kruithof EK, Satta N, Liu JW, Dunoyer-Geindre S, Fish RJ. Geneconversion limits divergence of mammalian TLR1 and TLR6.BMC Evol Biol 2007;7:148.

[18] Modlin RL. Mammalian toll-like receptors. Ann Allergy AsthmaImmunol 2002;88:543e7. quiz 8-50, 83.

[19] Smith KD, Andersen-Nissen E, Hayashi F, Strobe K, Bergman MA,Barrett SL, et al. Toll-like receptor 5 recognizes a conserved

Positive selection in teleost TLR9 241

site on flagellin required for protofilament formation and bac-terial motility. Nat Immunol 2003;4:1247e53.

[20] Andersen-Nissen E, Smith KD, Bonneau R, Strong RK,Aderem A. A conserved surface on Toll-like receptor 5 recog-nizes bacterial flagellin. J Exp Med 2007;204:393e403.

[21] Andersen-Nissen E, Smith KD, Strobe KL, Barrett SL,Cookson BT, Logan SM, et al. Evasion of Toll-like receptor 5by flagellated bacteria. Proc Natl Acad Sci U S A 2005;102:9247e52.

[22] Yilmaz A, Shen S, Adelson DL, Xavier S, Zhu JJ. Identificationand sequence analysis of chicken Toll-like receptors. Immuno-genetics 2005;56:743e53.

[23] Dhara A, Saini M, Das DK, Swarup D, Sharma B, Kumar S, et al.Molecular characterization of coding sequences and analysisof Toll-like receptor 3 mRNA expression in water buffalo(Bubalus bubalis) and nilgai (Boselaphus tragocamelus).Immunogenetics 2007;59:69e76.

[24] Ahmad-Nejad P, Hacker H, Rutz M, Bauer S, Vabulas RM,Wagner H. Bacterial CpG-DNA and lipopolysaccharides acti-vate Toll-like receptors at distinct cellular compartments.Eur J Immunol 2002;32:1958e68.

[25] Bauer S, Kirschning CJ, Hacker H, Redecke V, Hausmann S,Akira S, et al. Human TLR9 confers responsiveness to bacterialDNA via species-specific CpG motif recognition. Proc Natl AcadSci U S A 2001;98:9237e42.

[26] Takano T, Kondo H, Hirono I, Endo M, Saito-Taki T, Aoki T. Mo-lecular cloning and characterization of Toll-like receptor 9 inJapanese flounder, Paralichthys olivaceus. Mol Immunol 2007;44:1845e53.

[27] Hirono I, Takami M, Miyata M, Miyazaki T, Han HJ, Takano T,et al. Characterization of gene structure and expression oftwo toll-like receptors from Japanese flounder, Paralichthysolivaceus. Immunogenetics 2004;56:38e46.

[28] Neujahr DC, Reich CF, Pisetsky DS. Immunostimulatory proper-ties of genomic DNA from different bacterial species. Immu-nobiology 1999;200:106e19.

[29] Dalpke A, Frank J, Peter M, Heeg K. Activation of toll-likereceptor 9 by DNA from different bacterial species. InfectImmun 2006;74:940e6.

[30] Bell JK, Mullen GE, Leifer CA, Mazzoni A, Davies DR, Segal DM.Leucine-rich repeats and pathogen recognition in Toll-likereceptors. Trends Immunol 2003;24:528e33.

[31] Boyd AP, Ross PJ, Conroy H, Mahon N, Lavelle EC, Mills KH.Bordetella pertussis adenylate cyclase toxin modulates innateand adaptive immune responses: distinct roles for acylationand enzymatic activity in immunomodulation and cell death.J Immunol 2005;175:730e8.

[32] Philbin VJ, Iqbal M, Boyd Y, Goodchild MJ, Beal RK, Bumstead N,et al. Identification and characterization of a functional,alternatively spliced Toll-like receptor 7 (TLR7) and genomicdisruption of TLR8 in chickens. Immunology 2005;114:507e21.

[33] Magnadottir B. Innate immunity of fish (overview). Fish Shell-fish Immunol 2006;20:137e51.

[34] Plouffe DA, Hanington PC, Walsh JG, Wilson EC, Belosevic M.Comparison of select innate immune mechanisms of fish andmammals. Xenotransplantation 2005;12:266e77.

[35] Du Pasquier L, Wilson M, Greenberg AS, Flajnik MF.Somatic mutation in ectothermic vertebrates: musings on se-lection and origins. Curr Top Microbiol Immunol 1998;229:199e216.

[36] Jault C, Pichon L, Chluba J. Toll-like receptor gene family andTIR-domain adapters in Danio rerio. Mol Immunol 2004;40:759e71.

[37] Oshiumi H, Tsujita T, Shida K, Matsumoto M, Ikeo K, Seya T.Prediction of the prototype of the human Toll-like receptorgene family from the pufferfish, Fugu rubripes, genome.Immunogenetics 2003;54:791e800.

[38] Franch R, Cardazzo B, Antonello J, Castagnaro M,Patarnello T, Bargelloni L. Full-length sequence and expres-sion analysis of Toll-like receptor 9 in the gilthead seabream(Sparus aurata L.). Gene 2006;378:42e51.

[39] Yang Z. PAML: a program package for phylogenetic analysis bymaximum likelihood. CABIOS 1997;13:555e6.

[40] Kumar S, Tamura K, Nei M. MEGA: Molecular Evolutionary Ge-netics Analysis software for microcomputers. Comput Appl Bi-osci 1994;10:189e91.

[41] Kumar S, Tamura N, Nei M. MEGA3: integrated software forMolecular Evolutionary Genetics Analysis and sequence align-ment. Brief Bioinf 2004;5:150e63.

[42] Posada D, Crandall KA. MODELTEST: testing the model of DNAsubstitution. Bioinformatics 1998;14:817e8.

[43] SaitouN, NeiM.Theneighbor-joiningmethod:a newmethod forreconstructing phylogenetic trees. Mol Biol Evol 1987;4:406e25.

[44] Letunic I, Copley RR, Pils B, Pinkert S, Schultz J, Bork P. SMART5: domains in the context of genomes and networks. NucleicAcids Res 2006;34:D257e60.

[45] Baldi P, Brunak S, Frasconi P, Soda G, Pollastri G. Exploitingthe past and the future in protein secondary structure predic-tion. Bioinformatics 1999;15:937e46.

[46] Nei M, Gojobori T. Simple methods for estimating the numbersof synonymous and nonsynonymous nucleotide substitutions.Mol Biol Evol 1986;3:418e26.

[47] Yang Z, Nielsen R, Goldman N, Pedersen AM. Codon-substitu-tion models for heterogeneous selection pressure at aminoacid sites. Genetics 2000;155:431e49.

[48] Yang Z. Likelihood ratio tests for detecting positive selectionand application to primate lysozyme evolution. Mol Biol Evol1998;15:568e73.

[49] Yang Z, Wong WS, Nielsen R. Bayes empirical Bayes inferenceof amino acid sites under positive selection. Mol Biol Evol2005;22:1107e18.

[50] Tang K, Thornton KR, Stoneking M. A new approach for usinggenome scans to detect recent positive selection in the humangenome. PLoS Biol 2007;5:e171.

[51] Griebel PJ, Brownlie R, Manuja A, Nichani A, Mookherjee N,Popowych Y, et al. Bovine toll-like receptor 9: a comparativeanalysis of molecular structure, function and expression. VetImmunol Immunopathol 2005;108:11e6.

[52] Kobe B, Kajava AV. The leucine-rich repeat as a protein recog-nition motif. Curr Opin Struct Biol 2001;11:725e32.

[53] Kobe B, Deisenhofer J. A structural basis of the interactionsbetween leucine-rich repeats and protein ligands. Nature1995;374:183e6.

[54] Babiuk S, Mookherjee N, Pontarollo R, Griebel P, van DrunenLittel-van den Hurk S, Hecker R, et al. TLR9�/� and TLR9þ/þ

mice display similar immune responses to a DNA vaccine. Im-munology 2004;113:114e20.

[55] Choe J, Kelker MS, Wilson IA. Crystal structure of human toll-like receptor 3 (TLR3) ectodomain. Science 2005;309:581e5.

[56] Beutler B, Rehli M. Evolution of the TIR, tolls and TLRs: func-tional inferences from computational biology. Curr Top Micro-biol Immunol 2002;270:1e21.

[57] Chuang TH, Lee J, Kline L, Mathison JC, Ulevitch RJ. Toll-likereceptor 9 mediates CpG-DNA signaling. J Leukoc Biol 2002;71:538e44.

[58] Kimbrell DA, Beutler B. The evolution and genetics of innateimmunity. Nat Rev Genet 2001;2:256e67.

[59] Lamphier MS, Sirois CM, Verma A, Golenbock DT, Latz E. TLR9and the recognition of self and non-self nucleic acids. Ann N YAcad Sci 2006;1082:31e43.

[60] Takeda K, Kaisho T, Akira S. Toll-like receptors. Annu Rev Im-munol 2003;21:335e76.

[61] Zhang XS, Choi JH, Heinz J, Chetty CS. Domain-specific posi-tive selection contributes to the evolution of Arabidopsis

242 J.S.-C. Chen et al.

leucine-rich repeat receptor-like kinase (LRR RLK) genes. JMol Evol 2006;63:612e21.

[62] Sun X, Cao Y, Wang S. Point mutations with positive selectionwere a major force during the evolution of a receptor-kinaseresistance gene family of rice. Plant Physiol 2006;140:998e1008.

[63] Kobe B, Deisenhofer J. Proteins with leucine-rich repeats.Curr Opin Struct Biol 1995;5:409e16.

[64] Kobe B, Deisenhofer J. The leucine-rich repeat: a versatilebinding motif. Trends Biochem Sci 1994;19:415e21.

[65] Kobe B, Deisenhofer J. Crystal structure of porcine ribonucle-ase inhibitor, a protein with leucine-rich repeats. Nature1993;366:751e6.

[66] Kobe B, Deisenhofer J. Mechanism of ribonuclease inhibition byribonuclease inhibitor protein based on the crystal structure ofits complex with ribonuclease A. J Mol Biol 1996;264:1028e43.

[67] Felsenstein J. Inferring phylogenies. Sunderland, MA: SinaurAssociates; 2004.

[68] Graur D, Li W-H. Fundamentals of molecular evolution. Sun-derland, MA: Sinauer Associates; 2000.