Sensing bacterial flagellin by membrane and soluble orthologs of Toll-like receptor 5 in rainbow trout (Onchorhynchus mikiss).

Rainbow trout (Onchorhynchus mikiss) possess two genes encoding putative leucine-rich repeat (LRR)-containing proteins similar to human TLR5. Molecular cloning of these two LRR proteins suggested the presence of a TLR5-like membrane form (rtTLR5M) and a soluble form (rtTLR5S). Here we elucidated the primary structures and the unique combinational functions of these fish versions of TLR5. The LRR regions of rtTLR5S and rtTLR5M exhibited 81% homology and relatively high (35.6 and 33.7%) homology to the extracellular domains of human TLR5 (huTLR5). Thus, two distinct genes encode the TLR5 orthologs in fish, one of which has a consensus intracellular domain (TIR). In order to test their functions, we constructed fusion proteins with the LRR region of rtTLR5S (S-chimera) or that of rtTLR5M and the TIR of huTLR5 (M-chimera). The S- and M-chimeras expressed in HeLa or CHO cells signaled the presence of Vibrio anguillarum flagellin, resulting in NF-kappaB activation. rtTLR5M was ubiquitously expressed, whereas rtTLR5S was predominantly expressed in the liver. In the hepatoma cell lines of the rainbow trout RTH-149, stimulation of rtTLR5M with V. anguillarum or its flagellin allowed the up-regulation of rtTLR5S. Flagellin-mediated NF-kappaB activation was more significant in the presence of or simultaneous expression of rtTLR5S. Therefore, a two-step flagellin response occurred for host defense against bacterial infection in fish: (a) flagellin first induced basal activation of NF-kappaB via membrane TLR5, facilitating the production of soluble TLR5 and minimal acute phase proteins, and (b) the inducible soluble TLR5 amplifies membrane TLR5-mediated cellular responses in a positive feedback fashion.

Toll-like receptors (TLR) 1 recognize foreign material and alert for clearance and immune responses in a systemic fashion (1). The NF-B activation pathway and type I IFN-producing pathway are usually activated in response to microbial constituents, which is referred to as pathogen-associated molecular patterns (PAMPs) (2,3). TLRs are signaling receptors that engage in an innate immune recognition together with catch-up receptors (opsonin receptor, etc.). This precedes activation of the acquired immune system, including T/B lymphocytes. Although it has been believed that the innate immune system (including TLRs) is present in all vertebrates as well as invertebrates, no functional properties of the TLR have been reported in lower vertebrates. The functional feature of TLRs has been extensively investigated in human and mice. The results of these studies are generally compared with those of Drosophila, which have a Toll family of proteins with considerably different functional properties (4).
In humans, ten members of the TLR family (TLR1-10) have been identified (1)(2)(3)(4)(5). All TLRs have an extracellular domain containing leucine-rich repeats (LRRs) flanked by a C-terminal region and a cytoplasmic signaling domain, called the Toll/IL-1 receptor homology domain (TIR) (1)(2)(3)(4)(5). Each TLR recognizes a distinct ligand(s) and elicits different, sometimes overlapping, immune responses (2,3). A variety of immune responses are induced through TLRs in dendritic cells (DCs) (1,2), which are major antigen-presenting cells that first encounter foreign material to mount its antigens for presentation to lymphocytes. The activation of DCs is differentially regulated by TLRs and the adaptors that bind the TIR domain of TLRs, selecting appropriate signal pathways and resulting output (2,3).
According to the pufferfish Fugu (Fugu rubripes) genome project, the signature TIR domain has been conserved across evolution (6). Three major differences were identified in the family of the Fugu Toll-like receptors (fgTLR) when compared with human TLRs (6): 1) There is a soluble form of the TLR5 ortholog in the fish but not in humans. 2) No TLR4 ortholog has been identified in the fish, and 3) TLRs named TLR21 and TLR22 are novel and specific to fish. fgTLR2, -3, -5, -7, -8, and -9 structurally correspond to those of mammalian TLRs. A possible interpretation is that TLR1, -2, -3, -4, -5, -7, -8, -9, -21, and -22 existed in both fish and mammals, in a common ancestral genome. TLR4 was lost in the fish lineage whereas TLR21 and 22 were lost in the mammalian lineage. Solitary ascidian (Ciona intestinalis) has only three putative Toll-like proteins (7), which, like Caenorhabditis elegans Toll (8), represent primordial forms before expansion into the Toll family. Drosophila has nine Tolls, and their functional properties were mostly related to body patterning as well as host defense (9). Current evidence suggests that the development of TLR genes is earlier than that of fish, but not of the ascidian, separate from mammals. The results of various investigations including ours conclude that the mammalian innate system was established preceding or concomitant with the assembling of the acquired immunity. This reflects selection pressure exerted by pathogens under distinct environments.
Rainbow trout has two TLR5 isoforms, membrane (TLR5M) and soluble (TLR5S) forms (10) similar to fgTLR5S (6). These two forms of TLR5 recognized Vibrio anguillarum flagellin. Here, we studied the TLR5-mediated immune response in the rainbow trout. A possible role for this unique flagellin recognition system of fish is discussed.
Molecular Cloning of a Soluble Form of rtTLR5 (rtTLR5S)-The EST clone that contains the leucine-rich repeat (AF281346) from rainbow trout exposed to V. anguillarum was cloned in our laboratory (12). Full-length rtTLR5S cDNA was cloned using the SMART RACE cDNA amplification kit (BD Biosciences). For this, PCR was performed with liver cDNA templates and rtTLR5S gene-specific internal primers (for 5Ј-RACE: S-GSP/F and S-NGSP/F, for 3Ј-RACE: S-GSP/R) (Table I) paired with either the Primer 1 (P1)-or the P2-specific end primers (BD Biosciences). The PCR products were ligated directly into the pGEM-T Easy vector (Promega, Madison, WI), and several independent transformants were subjected to DNA sequencing using ABI 3100 sequencer (PE Applied Biosystems, Foster City, CA) for assessing sequence accuracy. Full-length rtTLR5S cDNA was obtained from the liver cDNA library by RT-PCR using primers SF and SR (Table I). The cDNA encoding rtTLR5S without signal sequences was placed between the HindIII-SalI sites of the pFLAG-CMV plasmid.
Molecular Cloning of a Membrane Form of rtTLR5 (rtTLR5M)-A partial rtTLR5M fragment was obtained by PCR with degenerate primers based on the sequences of the TIR domains of huTLR5 (13), moTLR5 (14), and fgTLR5 (6). The sequences were aligned using ClustalW (clustalw.genome.ad.jp/) (Fig. 1A). First-strand cDNAs were reversetranscribed from random-primed RNA templates prepared from rainbow trout liver using H-MLV(Ϫ) reverse transcriptase (Promega). Second, an rtTLR5M full-length sequence was obtained using the Marathon RACE cDNA amplification kit (BD Biosciences). Liver cDNA templates and rtTLR5M gene-specific internal primers (for 5Ј-RACE: MGSPF and MNGSPF, for 3Ј-RACE: MNGSPR and MNGSPRF) ( Table  I) paired with either the adaptor primer 1 (AP1)-or the AP2-specific end primers (BD Biosciences) were used. The PCR products were ligated directly into the pGEM-T Easy Vector, and several independent transformants were subjected to DNA sequencing as described above. Finally, full-length rtTLR5M cDNA was obtained from liver cDNA by RT-PCR using primers MF and MR (Table I). The cDNA encoding rtTLR5M with deleted signal sequences was placed between the SalI-NotI sites of the pFLAG-CMV plasmid.
Southern Blotting-10 g of genomic DNA extracted from rainbow trout liver was digested with various restriction endonucleases and separated on a 1.0% agarose gel by a standard electrophoretic separation method (15). DNA was transferred from the agarose gel onto nylon membrane (Hybond Nϩ, Amersham Biosciences) according to the manufacturer's instructions. The PCR-amplified probes S and M ( Fig. 2A) were labeled with [␣-32 P]dCTP using a MegaPrime DNA labeling system (Amersham Biosciences). Hybridization was performed in Ex-pressHyb Solution (BD Biosciences) for 1 h. After hybridization, blots were washed three times in 2ϫ SSC containing 0.05% SDS at 24°C for 20 min followed by two washes in 0.1ϫ SSC containing 0.1% SDS at 50°C for 20 min. The blots were subjected to an imaging plate and analyzed using the FLA imaging analyzer (Fuji Film, Tokyo, Japan).
RT-PCR-Total RNAs were isolated from muscle, testis, gill, stomach, intestine, kidney, heart, spleen, brain, and liver of rainbow trout using the TRIzol reagent. The purified RNA samples were used as templates for reverse transcription reactions. PCR was performed with primers specific for each type of rtTLR5, rtIL-1␤, rtTNF-␣, and rt-␤actin under the following conditions: initial denaturation at 94°C for 2 min followed by 25ϳ 40 cycles at 94°C for 30 s, 55°C for 30 s, and 72°C for 1 min. The primer sequences used for RT-PCR are shown in Table II. The PCR products were separated in a 1.5% (w/v) agarose gel and stained with ethidium bromide.
Constructs for Reporter Assay-The S-chimera was constructed by fusing cDNAs encoding the FLAG-tagged extracellular domain of rtTLR5S (amino acids 29 -597) to the C-terminal flanking region, transmembrane, and cytoplasmic domains of huTLR5 (amino acids 574 -858). The construct was cloned into the pEFBOS expression vector (17). The M-chimera was constructed to fuse the FLAG-tagged extracellular domain of rtTLR5M (amino acids 21-596) to the transmembrane and cytoplasmic domains of huTLR5 (amino acids 574 -858). Full-length huTLR5 was obtained from human monocyte cDNA by RT-PCR using primers hu5F and hu5R (Table I). The cDNA of huTLR5 was placed in the pEFBOS plasmid. The promoter region of human E-selectin (ELAM) (Ϫ241 to Ϫ54) was ligated between KpnI-HindIII sites of pGV-E2 (Toyo Inc., Tokyo, Japan). This plasmid (designated pELAMluc) was used as a reporter for NF-B activation. phRL-TK vector (Promega) was used as an internal control.
Reporter Gene Assay for NF-B Promoter-HeLa or CHO cells were plated in 24-well plates (1 ϫ 10 5 cell per well), and incubated for 12 h. Then they were transiently transfected with pEFBOS for expression of S-chimera, M-chimera, huTLR5 (100 ng), or vector alone (control), together with a pELAM-luc reporter gene (100 ng) (10) using Lipo-fectAMINE Plus reagent (Invitrogen). The total amount of transfected DNA (400 ng) was adjusted by adding empty vector. 0.1 ng of phRL-TK was used as an internal control. The cells were cultured for 36 h after transfection. Where indicated, the effect of exogenously added rtTLR5S on NF-B promoter activity in the M-chimera-expressing cells was determined. That is, 24 h after transfection with M-chimera, culture media in the M-chimera-expressing cells were replaced with the medium from CHO or HeLa cells, which were transiently transfected with pEFBOS/rtTLR5S cDNA (1, 10, 100 ng) and cultured for 12 h. Then the cells were treated with rFlaA (0.01, 0.1, 1 g/ml), LPS (0.1 g/ml), PGN (10 g/ml), poly(I⅐C) (2 g/ml), or CpG-ODN (2 M) at 37°C for 5 h. In some experiments, recombinant rtTLR5S (1, 10, 100 ng/ml) were directly added to the medium of M-chimera-expressing cells 12 h before rFlaA stimulation. The cells were stimulated with rFlaA for 5 h. The cells were lysed with lysis buffer (Promega), and luciferase activity was measured with the reagents and protocols from the dual-luciferase reporter gene assay kit (Promega) using a luminometer (BLR-201, Aloka, Tokyo, Japan). Specific activity was calculated from light intensity measurements with a Renilla luciferase internal control. Values were expressed as mean relative stimulation with S.D. from triplicate values from a minimum of three separate experiments.
Recombinant rtTLR5S by Baculovirus-Recombinant rtTLR5S was produced with Bac-To-Bac baculovirus expression system (Invitrogen). In brief, the full-length rtTLR5S with the His 6 tag at the C-terminal end was constructed using PCR. The cDNA was placed between the SalI and NotI sites of the pFastBac1 plasmid, which was designated pFast-Bac1/rtTLR5S. To generate the Bacmid DNA encoding rtTLR5S with the His 6 tag, E. coli DH10Bac cells were transformed with pFastBac1/ rtTLR5S using transposition activity in the cell. The bacmid DNA was obtained by the standard method from cells and transferred to Sf21 cells by UniFECTOR (B-Bridge International, San Jose, CA) to generate the recombinant viruses. The recombinant viruses were used to infect a monolayer of Sf21 cells in serum-free medium at a multiplicity of 5. After a 3-day culture, the rtTLR5S protein was purified from the medium using nickel-nitrilotriacetic acid beads (Qiagen) and Mono Q ion exchange columns (Amersham Biosciences). Proteins were identified by immunoblotting using anti-His 6 tag antibody. With this procedure, 5 g of rtTLR5S were obtained from 500 ml of culture medium.
Evaluation of Flagellin-rtTLR5S Association-The binding efficacy of His 6 -labeled rtTLR5S to GST-FlaA was determined by GST pulldown assay. Briefly, 20 ng of rtTLR5S and indicated amounts (0 -5 g) of GST or GST-FlaA in 300 l of PBS (pH 8.0) were incubated at 4°C for 2 h and incubated with 10 l of glutathione-Sepharose 4B for 2 h with gentle shaking. Glutathione-Sepharose was harvested by centrifugation and washed four times with PBS. The protein was eluted from Sepharose with 10 mM glutathione in PBS. The eluate was subjected to SDS-PAGE under reducing conditions followed by immunoblotting using anti-His 6 antibody (0.2 g/ml). Proteins were developed with horseradish peroxidase-labeled goat anti-rabbit IgG second antibody (BIO-SOURCE, Camarillo, CA) and an ECL kit (Amersham Biosciences).

Identification of TLR5-like Sequences from the Rainbow
Trout-Based on an EST sequence obtained from rainbow trout infected with V. anguillarum (AF281346), we cloned a cDNA by 3Ј-and 5Ј-RACE that predicted the existence of LRR-containing proteins. A motif search, Kyte and Doolittle plot, and ClustalW comparison suggested that the amino acid sequence corresponds to the extracellular LRR domains of human TLR5 and surprisingly contained no transmembrane region or TIR domain (Fig. 1B). SMART analysis defined ten LRRs in this protein (Fig. 1C). Based on these results, we designated this putative protein of rainbow trout soluble TLR5 (rtTLR5S).
To identify the membrane form of TLR5 in rainbow trout, degenerate primers were designed according to the conserved sequences among the TIR domains of human, mouse, and fgTLR5 (Fig. 1A). PCR using degenerate primers (Fig. 1A) followed by the RACE method allowed us to clone additional TLR5-like cDNA. Homology search analysis suggested that this is an ortholog of human TLR5, which contained putative signal peptides, 11 LRRs (including the LRR C-terminal flanking region), and the TIR domain (Fig. 1, B and C). The complete sequences of these two rainbow trout (rt) TLR5 are shown in Fig. 1B.
We have cloned seven independent RT-PCR products of rtTLR5S using mRNA and SF and SR primers ( Table I). The collective sequence consisted of a 1,992-bp open reading frame (ORF) and a 1,168-bp 3Ј-UTR. Three polyadenylation signals were found in the 3Ј-UTR, the last of which was followed by the poly(A) tail. The predicted amino acid sequence of this cDNA is 664 amino acids (Fig. 1). rtTLR5S was 35.6 and 37.1% homologous to the extracellular domains of human and mouse TLR5, respectively (10). rtTLR5S showed Ͻ25% homology to that of other human or mouse TLRs. Nine of the eleven cysteines in the extracellular domains of mo/huTLR5 were conserved in rtTLR5S (Fig. 1B).
We have cloned four independent RT-PCR products for rtTLR5M using MF and MR primers ( Table I). The collective deduced sequence consisted of a 2,637-bp ORF and 544-bp 3Ј-UTR. The polyadenylation site in the 3Ј-UTR was followed by the poly(A) tail. The predicted amino acid sequence of this cDNA is 879 amino acids (Fig. 1B). rtTLR5M was 40.1, 40.5, and 48.5% homologous to hu/mo/fgTLR5, respectively. rtTLR5M is Ͻ25% homologous to that of other hu/mo/fgTLRs. Ten cysteines found in the extracellular domains of mouse and human TLR5 were conserved in rtTLR5M (Fig. 1B).
The LRR regions of rtTLR5M and rtTLR5S exhibited 81.0% homology. Several tissues were tested for rtTLR5M-and rtTLR5S-specific messages using the specific primer sets shown in Table I (Fig. 1D). rtTLR5M was ubiquitously expressed in all tissues whereas rtTLR5S was predominantly present in liver. The presence of the soluble form of TLR5 is surprising since no such TLR5 has been reported either in human and mouse. Gene and Southern Analysis of rtTLR5s-Sequence analysis of the rtTLR5S gene revealed a 757-bp intron near the signal peptides (20 bp from the ATG start codon) ( Fig. 2A). The gene of rtTLR5M had no introns. Southern blot analysis was carried out using two distinct probes covering the TIR region of rtTLR5M (probeM: rtTLR5M 1867-2707 nt, 841 bp) or the LRR region largely common to rtTLR5M and rtTLR5S (probeS: rtTLR5S 354 -1314 nt, 961 bp) (Fig. 2). Based on the nucleotide sequences of the ORFs, the intron, and 3Ј-UTs of the DNAs, there was only one site each for EcoRI, HindIII, and PstI downstream of the probe S in rtTLR5S. Thus, based on the restriction fragment lengths, the size of the fragments must be larger than 2.4 kbp. Actual fragments obtained on Southern blots were Ͻ2.4 kbp in addition to the expected two fragments in each lane (Fig. 2B) suggesting the existence of an additional gene with a similar but different sequence. ϳ1.5-kbp fragments were detected with probe S when digested with PstI and EcoRI, further confirming this point.
Probe M was designed to hybridize with the downstream regions that encode TIR of rtTLR5M (Fig. 2C). ϳ1-kbp fragments were detected in the EcoRI-and PstI-digested lanes, which reflected the presence of the rtTLR5M gene. A 3.5-kbp fragment obtained upon HindIII digestion may correspond to the rtTLR5M gene. Additional upper bands, although faint, suggest the presence of additional sequences containing similar TIR sequences. In addition, we have also sequenced a pseudogene (third gene) with a number of stop codons in the putative coding region encoding a peptide similar to rtTLR5M (data not shown).
The genes of rtTLR5M and rtTLR5S contained two Sau3AI, AluI, and PvuII sites in their ORFs. Fragments obtained in Southern analysis are in line with these restriction sites (Fig.  2B). However, the three-band profile observed in PvuII digestion (Fig. 2C) cannot be explained. Because no message for the putative third gene of TLR5 was identified, a pseudogene similar to rtTLR5M must be present in addition to the two identified TLR5 genes.
rtTLR5S Is Induced in Response to V. anguillarum-Next we examined whether rtTLR5S is up-regulated by infection with V. anguillarum (Fig. 3). Fish cell lines of various origins were tested for this purpose. Of these, rainbow trout hepatoma cell line RTH-149 cells were used for this investigation, because rtTLR5S message induction was significant in this cell line in response to dead V. anguillarum (Fig. 3). The rtTLR5S mRNA was barely detectable in resting RTH-149 cells. The rtTLR5S mRNA was induced 4 h after bacterial stimulation. The level of mRNA became maximal at 6 h, then gradually decreased. No message was detected 48 h post-stimulation. On the other hand, the mRNA of rtTLR5M was expressed constitutively in this cell line regardless of stimulation (Fig. 3). Under the same conditions, rtIL-1␤, an acute phase cytokine, was found to be up-regulated within 1 h and peaked around 3-10 h. Thus, rtTLR5S is another acute phase protein responding to stimulation. Similar induction profile of rtTLR5S was observed in a genital gland cell line RTG-2. These results were confirmed by quantitative PCR (data not shown). Hence, the acute phase response of rtIL-1␤ and rtTLR5S largely results from bacterial stimuli, presumably through a membrane form of rtTLR5. To test whether flagellin and rtTLR5M participate in this early phase event, we expressed two major forms of V. anguillarum flaA and flaC in E. coli.
Purification of Recombinant V. anguillarum Flagellin-Recombinant GST-FlaA and C were produced in E. coli, and the proteins were purified using glutathione-Sepharose (Fig. 4A). Any LPS contamination is virtually less than the detection limit (Ͻ0.02 pM) of the LPS determination kit in the flagellin preparations. The purified material was further treated with polymixin B to completely eliminate any possible effect of LPS. Purified rFlaA and rFlaC are shown by SDS-PAGE in Fig. 4. Fig. 4A shows the GST-flagellin fusion protein, and Fig. 4B is the flagellin protein from which GST was liberated by thrombin. The purified GST fusion proteins were 60 kDa, which match the molecular masses predicted from the flagellin sequence plus GST tag.
Flagellin Stimulation Induces rtTLR5S in RTH-149 Hepatoma Cells-To test whether rtTLR5M on RTH-149 hepatoma cells recognizes flagellin, RTH-149 cells were stimulated with purified rFlaA. The rFlaA stimulation led to the induction of rtTLR5S as well as rtIL-1␤, as in the case of V. anguillarum stimulation (Fig. 4C). The mRNA level of rtTNF-␣ was also increased in RTH-149 cells. mRNAs of rtTLR5S and rtIL-1␤ were induced by flagellin with profiles similar to those that occurred by bacterial infection. It is interesting to note that no induction of rtTLR5M was again observed with purified rFlaA. Similar induction profiles of the cytokines and rtTLR5S were observed in RTG-2 cells in response to rFlaA (Fig. 4C). The results were further confirmed with quantitative PCR (Fig. 4D). Similar results were obtained with rFlaC (not shown). Thus, in fish, flagellin recognition by the hepatic cells may be the major cause of V. anguillarum-mediated up-regulation of rtTLR5S. Because rtTLR5M is constitutively expressed in these cell lines, rtTLR5M may participate in the first recognition of flagellin in the cell lines. In fact, the retention time for cytokine induction by flagellin in the RTH-149 cells is consistent with that observed in human dendritic cells (18). Thus, the infection-mediated induction of rtTLR5S can be mimicked by exogenously added flagellin. We next examined the NF-B activation by luciferase reporter assay.
Recognition of Flagellin by a Chimera rtTLR5S-TIR of Human TLR5-Because flagellin is a ligand for human and mouse TLR5, we tested whether rtTLR5S recognizes flagellin. For this, we made a fusion receptor cDNA consisting of rtTLR5S, transmembrane, and TIR domains of human TLR5 (S-chimera). The NF-B reporter assay was performed in HeLa cells. Native HeLa cells neither expressed TLR5 nor responded to flagellin (not shown). The cDNA of the chimera was transfected into HeLa cells. Its expression was confirmed using anti-FLAG antibody by immunoblotting (data not shown). If rtTLR5S has the ability to sense flagellin, this chimera receptor could signal the presence of flagellin by the ectodomain and activate NF-B via TIR in HeLa cells. A representative result using the chimera-expressing HeLa cells is shown in Fig. 5A. This chimera molecule activated the NF-B reporter gene in HeLa cells to a degree similar to human TLR5 at similar expression levels (Fig. 5A). rFlaA and C had identical S-chimera stimulation potency (data not shown). To test the specificity, the S-chimera was stimulated with PGN (a ligand of TLR2), poly(I⅐C) (a ligand of TLR3), LPS (a ligand of TLR4), non-methylated CpG-ODN (a ligand of TLR9), and rFlaA (a ligand of TLR5) (Fig. 5B). The appropriate doses of these ligands were determined with human TLRs in our laboratory as described previously (19). The chimera molecule as well as human TLR5 exclusively recognized rFlaA to activate the NF-B reporter gene (Fig. 5B). Thus, it is apparent that a functional interaction between rF-laA and rtTLR5S exists.
A Chimera rtTLR5M-TIR of Human TLR5 Recognizes Flagellin in Combination with rtTLR5S-The function of rtTLR5M was first tested by the expression of full-length rtTLR5M in HeLa and CHO cells, but no NF-B response was detected, albeit its expression was confirmed (data not shown). NF-B activation was next determined with a chimera consisting of the LRRs of rtTLR5M and the TM and TIR of huTLR5 (M-chimera) (Fig. 6). After several trials, we selected two cell lines for transfection of M-chimera. They are CHO, with a relatively high transfection efficiency, and HeLa, with a low transfection efficiency. The response of M-chimera to rFlaA (1 g/ml) (Fig. 6A) and rFlaC (not shown) was detected as NF-B activation in the CHO system, although the degree of the response was low. Flagellin-mediated NF-B activation was marginally detected in M-chimera-expressing HeLa cells, probably due to its low transfection efficiency (Fig. 6A). Activation of NF-B was specific to rFlaA in M-chimera-expressing cells, since TLR ligands, LPS (0.1 g/ml), PGN (10 g/ml), poly(I:C) (2 g/ml), and CpG-ODN (2 M) induced activation of NF-B in these systems even at their optimal concentrations (Fig. 6B). We inferred from these that incompatibility of the TIR of rtTLR5M to mammalian adapter molecules resulted in no NF-B response.
Next the combination effect of rtTLR5M and rtTLR5S was studied. The experiment was designed so that the conditioned medium of rtTLR5S-expressing CHO cells that contained rtTLR5S was added to rtTLR5M-expressing CHO cells, and the cells were treated with rFlaA (Fig. 6C). This resulted in NF-B activation in proportion to the dose of the transfected cDNA of rtTLR5S that was present in the CHO cell medium. Addition of the supernatant of M-chimera-expressing cells did not result in such up-regulation of NF-B activation in the same system (data not shown). In M-chimera-expressing HeLa cells, flagellin-mediated NF-B activation was much more augmented in the presence of the conditioned medium of rtTLR5S-expressing HeLa cells (Fig. 6C).
To confirm this synergistic effect of rtTLR5S on NF-B activation, the rtTLR5S protein was produced in the baculovirus system. Its physical binding to GST-FlaA (Fig. 7A) and functional properties (Fig. 7B) were examined. Physical binding of rtTLR5S to flagellin was assessed by GST pull-down assay. The recombinant His-tagged rtTLR5S (20 ng) was mixed with GST-FlaA or GST. Glutathione-Sepharose was added to the mixture and eluted with 10 mM glutathione in PBS. After extensive washing in PBS, the amounts of bound rtTLR5S were checked by immunoblotting using anti-His antibody (Fig. 7A). The amounts of bound proteins were compared with that of lamprey complement 3 (AY359861) (20), which do not bind GST-FlaA. The binding assay suggested that rtTLR5S has sufficient affinity to hold flagellin. FIG. 3. RTH-149 cells response to V. anguillarum. mRNAs of rtTLR5M, rtTLR5S, and rtIL-1␤ were monitored by RT-PCR after stimulation of cells with dead V. anguillarum. rt-␤-Actin was used for the control. PCR products were analyzed by gel electrophoresis (1.5% TAE agarose) and visualized with ethidium bromide (1 g/ml). Three individual experiments were performed, and a representative one is shown.
Functional properties of rtTLR5S were tested on M-chimeraexpressing cells. Soluble rtTLR5S protein (1-100 ng/ml) rendered the M-chimera-expressing CHO and HeLa cells more sensitive to rFlaA in NF-B reporter assay (Fig. 7B). In particular, HeLa cells with rtTLR5M markedly responded to rtTLR5S. Thus, our hypothesis is that the initial recognition of flagellin by rtTLR5M induces rtTLR5S as well as inflammatory cytokines to a basal level in liver: Secondly, the combination of rtTLR5M and induced circulatory rtTLR5S systemically pro-voke robust activation of NF-B, which leads to full response to flagellin in the whole body.

DISCUSSION
In this investigation we presented evidence for the existence of a soluble and membrane form of TLR5 in rainbow trout. The membrane form of TLR5, rtTLR5M (M stands for membrane form), was similar to human and mouse TLR5 in its overall structure. However, no soluble TLR5 has been reported in mammals. Since the soluble TLR5-like gene is homologous to the gene encoding soluble TLR5 of F. rubripes (6), we presumed this to be an ortholog of the Fugu soluble form of TLR5, and named it as rtTLR5S (S stands for soluble form). When the LRR regions of these two proteins were compared with those of huTLRs, they showed the highest similarity to huTLR5 (10). A BLAST search also showed that they are most similar to TLR5. In fish, soluble TLR5 is differentially regulated from the putative membrane form in liver to augment TLR5-mediated inflammatory responses to bacterial flagellin.
We demonstrated that rtTLR5M and -S share similar ligandrecognition properties with the human and mouse TLR5 (21,22). Hence, the flagellin recognition system is conserved across the human and fish, and in the fish lineage, this system developed into a more sophisticated one. First, bacterial flagellin stimulates membrane TLR5 in fish and NF-B activation is increased proportionally, secondary to the induction of soluble TLR5 in the liver. Second, the induced soluble TLR5 efficiently catches-up bacterial flagellin in the circulation and brings it to membrane TLR5, which acts as a signaling receptor. In fish, robust activation of NF-B then occurs through the combination of the two forms of TLR5. The advantage of this recognition system for flagellin is comparable to the scenario of LPS recognition by TLR4 (23,24). Soluble LBP and CD14 facilitate LPS recognition by membrane TLR4 on the immune cell surface.
It is a matter of interest to clarify how flagellin provokes the host immune system through activation of the TLR5 signaling pathway. Flagellins have conserved N-and C-terminal regions that are important in its function (25). TLR5 signals the presence of the conserved regions of flagellin to activate NF-B and chemokine secretion (26). In fact, flagellins of various bacterial origin share the conserved regions (27,28) and induce TLR5- mediated NF-B activation in our chimera expression system (10). Flagellin activates IRAK via TLR5 (29). Recently, the TICAM-1 pathway has been shown to be responsible for IFN-␤ induction in TLR3 (30,31) and TLR4 (32)(33)(34). An adapter complex, TICAM-2 and TICAM-1, plays a key role in induction of IFN-␤ in TLR4 signaling in humans (32,33). Furthermore, in mammals, TLR5 together with TLR4 induces type I IFNs probably via TICAM-2 and TICAM-1 (35). Thus, what happens in TLR5 signaling in fish is an issue to be resolved.
Physiological significance of the soluble forms of TLR5 is a matter of interest. The basolateral expression of TLR5 serves as the flagellin receptor in epithelial cells. Indeed, TLR5 expressed on the basolateral surface of intestinal epithelia detects the invasion of a large variety of microbes (36) and recruits inflammatory cells responding to the invading bacteria. The soluble TLR5 could be required in the fish to systemically amplify the inflammatory response generated by membrane TLR5. If this is the case, mammals had lost this soluble TLR5 function because the need for TLR5-mediated immune response is limited to a local environment where microbes invade, hence preventing systemic inflammatory responses induced by the bacteria (37). It has been known that fish are highly sensitive to flagellin, which induces endotoxin-like re-sponse in fish (38). Fish are highly resistant to LPS compared with mammals (39,40). Probably because of the conserved function of TLR5 but not TLR4, fish sense flagellin rather than LPS as a major endotoxin.
The current concept is that a prototype of TLR arose for host defense before plants and animals diverged. In mammals, TLRs are pattern recognition receptors that directly recognize molecular patterns specific to microbes. In contrast, in Drosophila, Toll appears to function as "cytokine" receptors, some of which associate with development (9). The genome projects of Ciona and C. elegans suggested that Toll proteins are not involved in major part of host defense (7,8). Thus, Drosophila Toll and mammalian TLR families must have independently evolved (9). Although proteins with LRR motifs such as TLRs and CARD proteins (41) induce similar signal responses and functional outputs across species, the modes of microbial recognition followed by host defense response are variable among species and individual LRR protein families. These findings are interpreted to mean that the structural signature LRR does not always confer the specificity on protein recognition: the functional features of LRR proteins are multifarious. Flagellin is known to stimulate host defense in a variety of organisms, including plants, insects, and mammals (42), presumably via their flagellin recognition systems. This study points to the fact that at least fish and human share a similar TLR system for the recognition of flagellin.
Our previous analysis revealed that F. rubripes has almost all the orthologs of the human TLRs (6). These findings suggest that TLR1, -2, -3, -5, -7, -8, -9 appeared before fish diverged from a mammalian ancestor, more than 400 million years ago (43). Our study adds the notion that the TLR system is conserved along with the NF-B system (44) for cytokine induction across mammals and fish in not only structural but also functional features. Taken together, it is not surprising that other fish TLRs essentially conserve their functions as in TLR5. TLRs are main PAMP recognition molecules. Thus, further functional analysis of fish TLRs may enable us to show that the current mechanisms of the PAMP recognition system were already established in the human and fish common ancestor.
Recently, two reports (45,46) mentioned the TLR family of zebrafish, which was deduced from a draft of the genome project (47). We found several inconsistent points between our results on Fugu TLRs and that of the zebrafish TLRs, one of which mentioned the absence of soluble TLR5 in zebrafish. We found soluble TLR5 in rainbow trout and pufferfish. At this immature stage of the genome project of zebrafish, however, one cannot conclude the absence of the soluble TLR5 in zebrafish. Once the fish system for analyzing PAMP actions is established, we will be able to clarify the signaling pathways of each TLR and determine the effect of PAMP on antibody production, CTL induction, and potentiation of NK activity. Zebrafish have been used for forward genetic screening and in in vivo experiments to find new functional aspects of genes (47). Therefore, zebrafish will become a powerful tool for the analyses of TLR functions.
In summary, the prototype of the mammalian type Toll family is likely to be conserved across fish and human. Fish TLR5 signals the presence of flagellin to activate NF-B, which en-ables us to interpret that human and fish share a similar flagellin recognition system and signaling pathway. The differences in the flagellin recognition system between fish and human are represented by the function of soluble TLR5 described here. Important factors for divergence between the fish and human TLR5 system would be selection pressure exerted by pathogens in distinct environments such as sea and land and the secondary adaptation of the Toll family genes to different sets of pathogens. Various amounts of the purified rtTLR5S (1, 10, and 100 ng/ml) were added to the cells expressing M-chimera 12 h before stimulation. At specified time intervals after rFlaA stimulation (5 h), cells were harvested, and NF-B activity in cell lysates was determined by luciferase assay. Experiments were performed in triplicate, and the results are expressed as means Ϯ S.D.