Teleost TLR22 Recognizes RNA Duplex to Induce IFN and Protect Cells from Birnaviruses¹

The type I IFN system is a host defense against microbial pathogens in animals (1, 2). In acute viral infections, IFN induces Mx and oligoadenylate synthetase to suppress viral replication (3). In late-phase infection it orchestrates cellular immunity including T and NK cells and protects hosts from persistent or repetitive viral infections (2, 4).

Earlier reports suggested that in mammalian fibroblasts, dsRNA (or its analog poly(I:C)) acts as an inducer for type I IFN, but the receptors for triggering IFN induction had not been identified until recently. Currently, in mammals, TLR3 on the endosomal membrane and retinoic acid-inducible gene I (RIG-I)⁵ and melanoma differentiation-associated gene 5 (MDA5) in the cytoplasm are identified as sensors for dsRNA (1, 5). When viral genome RNA replicates in the cytoplasm, RIG-I and MDA5 sense it and assemble the adaptor mitochondrial antiviral signaling protein (MAVS; also called Cardif, IPS-1, orVISA) on the mitochondrial membrane (5). RIG-I preferentially recognizes 5′-phosphates of RNA (6, 7), whereas MDA5 recognizes the signature of dsRNA (8). They are distributed ubiquitously in cells/tissues and trigger IFN regulatory factor 3 (IRF-3) activation followed by type I IFN induction through the MAVS signal cascade (1, 2, 5). This intrinsic pathway appears to link main protective responses against RNA virus infection in mammals.

On the other hand, human TLR3 signals the presence of extrinsic dsRNA, recruits the adaptor Toll-IL-1R homology domain-containing adaptor protein 1 (TICAM-1), and induces IRF-3 activation followed by IFN-β promoter activation (1, 2, 9). Human TLR3 resides limitedly in myeloid dendritic cells, fibroblasts, and epithelial cells (10). TICAM-1 recruits TNF receptor-associated factor (TRAF) and TNF receptor-associated NF-κB kinase (TANK) family proteins for IFN-inducing signaling (11, 12 ; M. Sasai, H. Oshiumi, and T. Seya, unpublished data). NAK-associated protein 1 (NAP1), like other TANK family subunits (13), assembles two kinases, IKKε and TBK1, which activate the transcription factor IRF-3 (14). We call this extrinsic pathway the TICAM-1 pathway.

The TICAM-1 pathway and the cytoplasmic MAVS pathway converge on NAP1 to activate IRF-3 in human cells (15). Gene-disrupted mouse analyses show that TICAM-1 is involved in induction of the anti-mCMV immune response for host protection (16). The TICAM-1 pathway also appears to be involved in other DNA virus infections (16, 17). No clear involvement of TLR3 and TICAM-1 in defense against RNA virus infection has been offered using gene-disrupted mice, although many studies have anticipated that the TICAM-1 pathway has antiviral function against RNA viruses.

In fish studies, teleost IFN was recently discovered in the zebrafish (Danio rerio) (18), and since has been found in many fish species (19). Although the predicted protein sequence of the fish IFN has low (<20%) similarity to mammalian and avian type I IFNs, IFN-α and IFN-β, it up-regulates Mx and IFN-stimulated regulatory element (ISRE) promoter activation (18). Hence, fish possess IFN-inducing machinery. Since fish are exposed to viruses and RNA in water, they must have IFN-inducing receptors. However, no receptors for IFN induction have been identified yet in fish.

The existence of the TLR family in teleosts has been predicted from the genome database (20, 21, 22, 23). According to the database of Takifugu rubripes (puffer fish), this teleost species possesses an ortholog of human TLR3 and other orthologs of human TLR members (20). Additionally, this teleost has a gene encoding a fish-specific TLR (hereafter called TLR22), whose functions are unknown. Fish may have a part of the gene of MDA5-like product, but its functional features are also unidentified. What happens in the IFN response during viral infection accordingly remains to be addressed in fish.

We found TLR22 in many aquatic vertebrate species (19, 20, 24), but not in birds and land animals (24). In this study, we demonstrate that TLR22 is a dsRNA-recognizing pattern receptor that recruits TICAM-1 to induce IFN and exerts a protective role in fish cells against dsRNA virus infection.

Accession numbers for all genes used in this study are listed below: T. rubripes (fg)TLR22 (AB197916), rainbow trout (rt)TLR22-1 (AJ628348), rtTLR22-2 (AJ878915), rtTLR3 (DQ459470), rtIFN (AJ582754), and infectious pancreatic necrosis virus (IPNV) (NC001915). Rainbow trout has two orthologs of fgTLR22, rtTLR22-1 and rtTLR22-2, which were 93.0% homologous to each other (data not shown). Appropriate primers for detection of mRNAs of these genes are listed in Table I.

Human HEK293 or HeLa cells were cultured as described previously (10). A fibroblast-like cell line (OLHd-rRe3) of Japanese medaka fish, Oryzias latipes, was cultured in L-15 medium containing 20% heat-inactivated FCS and 10 mM HEPES at 33°C. RTG-2 cells derived from rainbow trout were cultured in a medium containing 10% heat-inactivated FCS and antibiotics (100 U/ml penicillin and 100 μg/ml streptomycin) at 20°C. LPS was purchased from BD Biosciences. Peptidoglycan (PGN) was purified from Staphylococcus aureus (Fluka). All nucleotide primers and oligodeoxynucleotides (ODN) containing CpG motifs (CpG-ODN) were purchased from Sigma-Genosys. Poly(I:C), polycytidylic acid (polyC), polyuridylic acid (polyU), and poly(dI:dC) were purchased from Amersham Biosciences. Variable-sized dsRNAs were transcribed from a cDNA template of measles virus with MEGAscript (Ambion) in vitro.

First-strand cDNAs were reverse-transcribed from random-primed RNA templates extracted from T. rubripes kidney or eye tissues using M-MLV(−) reverse transcriptase (Promega). FgTLR3, fgTLR22, and fgTICAM-1 full-length cDNA fragments were amplified by PCR using primers shown in Table I. The PCR products were cloned using pCR-Blunt vector (Invitrogen). Several independent clones were subjected to DNA sequencing using ABI 3100 sequencer (PE Applied Biosystems) for assessing sequence accuracy. IFN of T. rubripes was searched by homology to zebrafish IFN using the basic local alignment search tool (BLAST) server. The 5′ promoter region identified from T. rubripes genomic DNA as −1 to −777 was subcloned into pEFBOS vector. FgMyD88 Toll-IL-1 receptor (TIR) domain was found in the T. rubripes database by BLAST search with reference to the human MyD88 protein sequence, and amplified by PCR using the primers shown in Table I.

The cDNAs encoding fgTLR3 and fgTLR22 were placed between the XhoI and NotI sites of the pFEBOS expression vector. Flag-, hemagglutinin (HA)-, and Myc-tags were attached to the C- or N-terminus of the proteins as described previously (15, 25). Constitutively active CD4/TLR22 was constructed by fusing cDNAs encoding the extracellular domain of huCD4 (from 1 to 391 aa region) to the transmembrane and cytoplasmic domains of fgTLR22 (from 565 to 935 aa region). The obtained chimera CD4/TLR22 construct was placed between the XhoI and NotI sites of pEFBOS. Flag-tag was inserted just before the stop codon of fgTLR22 and placed between the XhoI and NotI sites of pEFBOS. To make the fgTLR22-myc expression vector, myc-tag was added to the C-terminal end of the signal peptide sequence of fgTLR22 and subcloned into XhoI and NotI sites of the pEFBOS expression vector. FgTLR3-yellow fluorscent protein (YFP) expression vector was constructed by inserting the full-length fgTLR3 into the pEYFP-N1 vector at the XhoI and KpnI sites. The constructs of the dominant-negative human TICAM-1 (TICAM TIR (P434H)) and MyD88 (TIR) were described previously (25). Dominant-negative forms of fgTICAM-1 and fgMyD88 were constructed as human dominant-negative counterparts. Precisely, cDNA encoding the 341–479 aa region of fgTICAM was inserted into the pEFBOS XhoI-NotI sites and substituted proline at 382 with histidine. The fgMyD88 TIR domain-encoding region (152–288 aa region) was subcloned into the XhoI-NotI sites of pEF-BOS.

HEK293 cells (1 × 10⁶ cells/well) were transiently transfected in 24-well plates using Lipofectamine 2000 reagent (Invitrogen) with pEFBOS human TLR2, human TLR3, human TLR4, human TLR9, fgTLR3, fgTLR22 (200 ng), dominant-negative human TICAM-1 (P434H), human MyD88 (TIR) (50, 200 ng) or empty vector, together with a luciferase-linked IFN-β promoter gene (Stratagene, 100 ng). RTG-2 cells (1 × 10⁶ cell/well) were transiently transfected in 24-well plates using FuGene HD (Roche) with fgTLR22, fgTLR3 (200 ng), dominant-negative fgTICAM-1 (50 or 200 ng), or empty vector together with a luciferase-linked fgIFN promoter gene. In some experiments, stable clones with this gene (named RTG (Luc-fgIFN) cells) were used instead of human IFN reporter. phRL-TK vector (1 ng) (Promega) was used as an internal control. Twenty-four hours after transfection, cells were stimulated with LPS (100 ng/ml), PGN (10 μg/ml), CpG (2 mM), or poly(I:C) (10 or 50 μg/ml), polyC (10 μg/ml), polyU (10 μg/ml), poly(dI:dC) (10 μg/ml), and in vitro-transcribed dsRNA (typically 10 μg/ml) for 6 h. The cells were lysed with lysis buffer (Promega) and luciferase activity was measured using a dual-luciferase reporter assay kit (Promega) with luminometer (Berthold Technologies). Specific activity was calculated from light intensity measurements with a Renilla luciferase internal control. Values were expressed as mean relative stimulation with SD from triplicate values from a minimum of three separate experiments.

HeLa, OLHd-rRe3, and RTG-2 cells were plated onto coverglass in a 24-well plate. In the following day, cells were transfected with indicated plasmids using FuGene HD. The amount of DNA was kept constant by adding empty vector. After 24 h, cells were stimulated with poly(I:C) and fixed with 3% of formaldehyde in PBS as indicated periods, and then washed four times with PBS. Cells were permeabilized with PBS containing 0.2% Triton X-100 for 15 min. Permeabilized cells were blocked with PBS containing 1% BSA, and were labeled with anti-Flag mAb (Sigma-Aldrich) or anti-HA pAb (Sigma-Aldrich) in 1% BSA/PBS for 1 h at room temperature. The cells were then washed with 1% BSA/PBS and treated for 30 min at room temperature with Alexa-conjugated Abs (Molecular Probes). For organelle marker staining, cells were treated with calnexin (StressGen Biotechnologies) and then FITC-labeled goat anti-mouse IgG secondary Ab. In some cases cells were prestained with anti-TLR Abs. Thereafter, micro-coverglasses were mounted onto a slide glass using PBS containing 2.3% 1,4-diazabicyclo(2.2.2)octane (DABCO) and 50% of glycerol. In some experiments, we used YFP-labeled fgTLR3 instead of tagged fgTLR3, since the background by the secondary Ab (against tag) was increased for unknown reasons in the case of fgTLR3. The stained cells were visualized at ×60 magnification under a FluoView (Olympus).

IPNV was propagated with RTG-2 cells. RTG-2 cells cultured in a 75-cm² T flask were infected with IPNV. After 4–7 days incubation, viruses were harvested from infected cell lysates by freeze-thaw cycles. Cell debris was removed by centrifugation at 10 krpm for 10 min, and then virus particles were concentrated by ultracentrifuge at 23 krpm for 1 h. Viral RNA was extracted from 15 ml of these lysates with 500 μl of TRIzol (Invitrogen) according to the manufacturer’s instructions.

Total RNA of RTG-2 cells were extracted using TRIzol reagents, and cDNA was made by using MV-reverse transcriptase with random primers. iQ SYBER Green Supermix was used for PCR reactions and analyzed with iCycler iQ real-time PCR analyzing system (Bio-Rad). Primers for quantitative PCR are shown in Table I. Relative rtIFN mRNA levels were calculated by dividing the relative amounts of mRNA of rtIFN by those of the rainbow trout β-actin mRNA.

HEK293FT cells were transiently transfected with expression vectors in a 6-well plate using Lipofectamine 2000 reagent (Invitrogen) and incubated for 24 h. Cells were lysed with lysis buffer (25 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% of Nonidet P-40, 2 mM PMSF, 25 mM iodoacetaminde, 10 mM EDTA), and proteins were immunoprecipitated with anti-Flag mAb and then washed four times with lysis buffer. Obtained samples were analyzed by SDS-PAGE (7.5 or 10% gel) and Western blotting using anti-HA pAb (Sigma-Aldrich).

RTG-2 cells (1 × 10⁶ cells/well) were infected with IPNV (typical multiplicity of infection (moi) of 0.1). After 24 h, supernatants were serially 10-fold diluted and incubated with RTG-2 cells placed on a flat 96-well plate, and 100 μl medium with 8× dilution cycles was added to determine the 50% tissue culture-infective dose (TCID₅₀). Cells were incubated at 15°C for 7 days. On day 7 of incubation, the plates were examined for the presence of viral cytopathic effect (CPE) under the microscope.

We used the predicted protein sequence of fgTLR22 for a BLASTP search on the National Center for Biotechnology Information (NCBI) BLAST server for a nonredundant database, and found that teleosts and frogs (Xenopus tropicalis) have TLR22 orthologs by phylogenetic analyses (20, 24) (Fig. 1 A). However, no expressed sequence tag or genome region encoding TLR22 was found in mammals and birds. Thus, these homology search analyses indicate that the TLR22 gene is conserved across aquatic vertebrates. To further confirm the absence of the TLR22 gene in the chicken genome, we conducted a TBLASTN search with fgTLR22 against the Gallus gallus whole genome. The best hits sequence in the chicken genome was the region encoding LOC415623, but the protein was an ortholog of TLR21 (data not shown). In the mouse genome, the most similar sequence to fgTLR22 was mouse TLR13. These analyses confirmed that TLR22 is conserved in vertebrates living in water or wet conditions, but not in animals living on land.

We cloned the fgTLR22 cDNA from the kidney and eye of T. rubripes to test its function. The cDNA sequence of the isolated fgTLR22 was identical to the sequence we previously predicted, and its open reading frame was encoded by three exons (20). The fgTLR22 protein had 15 leucine-rich repeats (LRRs) and one C-terminal LRR domain at the extracellular region and a TIR domain in the cytoplasmic region, suggesting that fgTLR22 possesses a typical structure of TLR but differs in the primary structure from TLR3 (Fig. 1 B).

Vertebrate IFN promoter and NF-κB activation has been determined successfully using human cells that have the human TLR signal system (26, 27). Thus, we first tested the functional ability of the TIR domain of fgTLR22 in human cell line HEK293. To estimate the output of fgTLR22 signaling, we made a chimera protein in which the Ig-like domain of CD4 was ligated with the transmembrane and intracellular region of fgTLR22. The human TLR4-CD4 chimera construct is known to activate the downstream signal of TLR4 by dimerization of the extracellular CD4 region (28). We examined whether the CD4-fgTLR22 chimera protein activates the human IFN-β promoter or NF-κB. The chimera protein was coexpressed in human HEK293 cells with an IFN-β promoter or NF-κB reporter plasmid. We observed minimal activation of NF-κB and about a 4-fold significant activation of the IFN-β promoter by expression of the chimera protein (Fig. 2 A), suggesting that the TIR domain of fgTLR22 retains the capacity to activate the human IFN-β promoter and, to a lesser extent, the transcription factor NF-κB in human cells.

The functional ability of fgTLR22 in HEK293 cells encouraged us to look for the ligand of fgTLR22. We subcloned the fgTLR22 full-length cDNA into the expression vector pEFBOS and transfected it into human HEK293 cells with reporter plasmids. FgTLR22 or mock-transfected cells were stimulated with PGN, LPS, CpG, or poly(I:C) for 6 h, and cell lysates were then prepared to measure reporter activation. PGN, LPS, and CpG activated neither NF-κB nor the IFN-β promoter via fgTLR22, although they activated the reporter via stimulation of the relevant human TLRs. In contrast, only poly(I:C) significantly activated the IFN-β promoter in fgTLR22-expressing cells (Fig. 2,B). Thus, poly(I:C) is a ligand for fgTLR22. Because T. rubripes possesses an ortholog of TLR3 (20), we cloned fgTLR3 and measured the IFN-β promoter activation induced by poly(I:C) in fgTLR3-expressing cells. fgTLR3 also conferred the responsiveness to poly(I:C) on HEK293 cells in a way similar to fgTLR22 (Fig. 2 C). Hence, the T. rubripes possesses two types of TLRs that recognize poly(I:C).

We next compared the response of fgTLR3 and fgTLR22 to polyU, polyC, poly(dI:dC), poly(I:C), and dsRNA using HEK transfectants. Ultimately, poly(I:C) and in vitro-transcribed dsRNA, but not polyU, polyC, or poly(dI:dC), activated the reporter in fgTLR22- or fgTLR3-expressing cells (data not shown), suggesting that both fgTLR22 and TLR3 signal the presence of dsRNA in human cells.

Next, we used various sizes of in vitro-transcribed dsRNA (29). Human TLR3 activated reporter genes in response to variable-sized dsRNA to similar extents irrespective of the length of dsRNA. In contrast, fgTLR22 and fgTLR3 showed different properties on responsiveness to variable-sized dsRNAs. In fgTLR22-expressing cells, 1000-bp dsRNA most strongly activated the reporter gene, but in fgTLR3-expressing cells, 200-bp dsRNA induced preferential activation of the reporter (Fig. 2 D). Therefore, fgTLR3, fgTLR22, and human TLR3 exhibit differences in preference to the size of dsRNA, although the three TLRs essentially confer response to poly(I:C) or dsRNA on HEK293 cells. These observations using the human cell indicate a unique role of fgTLR22 in poly(I:C) or dsRNA recognition.

We assessed the function of fgTLR22 in a teleost fibroblastic cell line, RTG-2. Teleosts also have an ortholog of human IFN, but previous phylogenetic analyses suggest that fish IFN forms a clade distinct from that of mammalian type I IFNs (18). The zebrafish (zf) IFN can be induced by overexpression of zfTICAM-1 (30), and it exerts antiviral effects through the induction of antiviral protein like Mx (31). We cloned the promoter region of the T. rubripes IFN gene that corresponds to the zfIFN and fused it to the luciferase open reading frame. fgTLR22 is widely expressed in various tissues and cell lines (20). Of the cell lines tested, the rainbow trout cell line, RTG-2, expressed only minute levels of the rtTLR22/rtTLR3 messages, which allowed us to employ RTG-2 for the fgTLR3/22 reporter analyses.

The RTG-2 cells were transfectable, but the transfection efficacy was ∼10% by lipofection. fgTLR22 protein showed 80.8% similarity to rtTLR22 (rtTLR22-1 and rtTLR22-2), which was barely expressed in both unstimulated and poly(I:C)-stimulated cells (see Fig. 3,B). The TIR sequences of fgTLR22 and rtTLR22 were 93.8% similar (data not shown), suggesting that they are functionally compatible. We transfected fgTLR22, fgTLR3, or mock expression vectors into RTG-2 cells together with the fgIFN promoter reporter plasmid (Fig. 3,A). At 24 h after transfection, cells were stimulated with poly(I:C). Cell lysate was prepared 6 h after poly(I:C) stimulation, and luciferase activities were measured. Poly(I:C) barely activated the fgIFN reporter in control cells, but significantly activated it in fgTLR22- or fgTLR3-transfected cells (Fig. 3 A), as observed in the human reporter assay system.

We next examined whether fgTLR22 transmits the signal to the endogenous IFN promoter (Fig. 3,B). The NCBI DNA database has two cDNA sequences of the IFN gene in rainbow trout (AJ582754 and AM235738). We examined the poly(I:C)-inducible IFN (AJ582754) expression through fgTLR22 stimulation with poly(I:C) by RT-PCR analysis. RTG-2 cells transfected with fgTLR22 or empty vectors were incubated with medium containing poly(I:C), and expression of rtIFN was examined by RT-PCR and quantitative PCR. The IFN mRNA expression was more strongly increased 2 h after poly(I:C) stimulation in fgTLR22-expressing cells than in control cells, and then gradually decreased in both types of cells (Fig. 3 B).

Because human TLR3 expression is induced by poly(I:C) (32), we investigated whether teleost TLR22 or TLR3 transcriptions are activated by poly(I:C) stimulation. Endogenous rtTLR22 or rtTLR3 expression was scarcely observed without stimulation, and rtTLR22 expression was induced from 12 h after stimulation. rtTLR3 was also up-regulated but more mildly than fgTLR22 in response to poly(I:C) stimulation (Fig. 3 C). These data are consistent with the notion that TLR22 is involved in the dsRNA recognition pathway.

Why are there two TLRs that respond to poly(I:C) or dsRNA in teleosts? The answer may lie in the differences between the two TLRs. fgTLR3 and fgTLR22 preferentially recognize different sizes of dsRNA, as suggested by the reporter analyses. An additional difference is that fgTL22 localizes to the surface of the cell membrane while fgTLR3 localizes inside the cell. TLRs are type I transmembrane proteins, and their subcellular localizations are determined based on their primary structures and coupling proteins (33, 34). Human TLR1, 2, 4, 5, and 6 are expressed on the cell surface, but TLR3, 7, and 9 are mainly localized at intracellular compartments, endoplasmic reticulum (ER), and early endosomes (10, 35). In the case of human TLR3, its linker region between the transmembrane and TIR domain is a critical determinant for its localization (33). Human TLR3 and fgTLR3 share similar linker regions, but fgTLR22 does not have a similar linker sequence (Fig. 1 B). Thus, we expected that fgTLR22 and fgTLR3 are distinctly localized.

To test this premise, we transfected HEK293 cells with myc-tagged fgTLR22 or Flag-tagged fgTLR3 and examined their expression using FACS analysis. fgTLR22 protein was observed partially on the cell surface (data not shown). On the cell surface, however, no fgTLR3 expression was observed, consistent with the case of dendritic cell human TLR3, which resides in the cytoplasmic compartments. Next, we overexpressed fgTLR22 or fgTLR3 in HeLa cells and visualized their localizations using a confocal microscope. YFP-labeled fgTLR3 was used in this study to avoid artificial deposition of fluorescence-labeled secondary Abs onto the cells. YFP labeling and HA or Flag labeling gave a similar imaging profile to fgTLR3 (data not shown). fgTLR3 was localized in the cytoplasm (Fig. 4). FgTLR3 was largely merged with calnexin (an integral protein localized in ER), suggesting that fgTLR3 localizes in the ER. In contrast, fgTLR22 was largely situated in close proximity to the cell surface, with only a minor population sitting in the cytoplasm (Fig. 4). These results confirm the predominant localization of fgTLR22 on the cell surface.

In the next experiment, we transfected OLHd-rRe3 fibroblastic cells derived from O. latipes (36) with fgTLR22 or fgTLR3 and tested their localizations. Although the transfection efficacy was very low in OLHd-rRe3 cells, we checked localization of TLRs by confocal analysis. As in the HeLa cells, cytoplasmic localization of fgTLR3 and cell surface-dominant localization of fgTLR22 were confirmed with OLHd-rRe3 cells (Fig. 4). Double staining of both fgTLR22 and fgTLR3 clearly supported their differential localizations. We also observed these localization differences in RTG-2 cells (data not shown). Taken together, fgTLR22 mainly localizes on the cell-surface membranes distinct from fgTLR3 in fish cells.

Human TLR family proteins use four adaptor proteins, MyD88, Mal/Toll-IL-1R domain-containing adaptor protein (TIRAP), TICAM-1/TRIF, and TICAM-2/TRAM, to induce cytokine production (1, 2). Several teleost genomes encode three adaptor proteins, MyD88, TIRAP, and TICAM-1 (37). Teleost TICAM-2 has not been found and may have been lost in the teleost lineage during evolution (30). At first we used HEK293 cells to determine which human adaptor protein is compatible with fgTLR22 by reporter assay. FgTLR22 and a dominant-negative form of human MyD88 or TICAM-1 were transfected into HEK293 cells with the human IFN-β promoter reporter, and the cells were stimulated with poly(I:C). The dominant-negative form of human TICAM-1 (P434H) inhibited IFN-β promoter activation in fgTLR22-expressing cells, but the dominant-negative human MyD88 did not (Fig. 5,A). Additionally, P434H inhibited fgTLR3-mediated IFN-β promoter activation in fgTLR3-expressing HEK293 cells (Fig. 5 A). Thus, human TICAM-1 can act as an adaptor for fgTLRs.

Next, we made the dominant-negative form of fgTICAM-1, which has a mutation similar to that of the human dominant-negative form. We transfected fgTLR22, fgTLR3, and type I IFN reporter into RTG-2 cells with or without the dominant-negative form of fgTICAM-1 and then examined reporter activation. As in human cells, dominant-negative fgTICAM-1 inhibited the reporter activation in fgTLR22- and fgTLR3-transfected cells (Fig. 5 B). Hence, fgTLR22 and fgTLR3 used fgTICAM-1 as the adaptor.

To further confirm the utilization of fgTICAM-1 as the adaptor by fgTLR22, we performed immunoprecipitation analyses. Full-length Flag-tagged fgTLR22 or fgTLR3 was transfected into HEK293 cells together with HA-tagged full-length fgTICAM-1. We prepared cell lysates 24 h after transfection, and fgTLR22 and fgTLR3 were immunoprecipitated with anti-Flag Ab. Precipitates were analyzed by SDS-PAGE and stained with anti-Flag or HA Abs. Co-immunoprecipitation was confirmed with fgTICAM-1 and fgTLR22 or fgTLR3 (Fig. 5 C), suggesting that fgTICAM-1 binds fgTLR22 and fgTLR3 in HEK cells.

On cytological analyses with OLHd-rRe3 cells, overexpressed fgTICAM-1 was uniformly distributed over the cytoplasm without stimulation. The localization profile of fgTICAM-1was not much changed after poly(I:C) stimulation in later time points (Fig. 6). Likewise, a major population of fgTLR22 barely changed its localization even when stimulated with poly(I:C) (Fig. 6,B). Only a small population appeared inside the cells in response to poly(I:C) and it partially merged with fgTICAM-1 (Fig. 6,B). FgTLR3 was distributed in the cytoplasm in resting cells, but 2 h after the stimulation the fgTLR3 gathered to form “speckles” in the cytoplasm (Fig. 6,A). Although some fgTLR22 and fgTLR3 colocalized with fgTICAM-1 on poly(I:C) stimulation, they gradually dissociated (Fig. 6). These data show that both fgTLR22 and fgTLR3 use TICAM-1 as the shuttling adaptor for IFN induction.

Type I IFN is crucial for the antiviral response in mammals. Because fgTLR22 induced IFN in fish cells, we expected it to be required for cell protection against virus infection. To test this hypothesis, we prepared dsRNA of the IPNV (38, 39). IPNV genomic dsRNA was extracted from IPNV-infected RTG-2 cells. Isolated dsRNA of IPNV was poured over RTG-2 cells expressing fgTLR22. Like synthetic dsRNA or rotavirus-derived dsRNA (SA11), dsRNA of IPNV-induced endogenous IFN gene expression in fgTLR22-expressing cells compared with control vector-transfected cells (no induction of IFN) 2 h after stimulation (Fig. 7).

To test the resistance of RTG-2 cells to IPNV infection, we propagated a large-scale IPNV preparation amplified with RTG-2 cells. The preparation usually contained naked RNA of the IPNV genome. Thus, we prepared two infection sources for the assay: an IPNV preparation containing dsRNA, and an IPNV preparation depleted of free dsRNA. The latter sample was prepared by treating the IPNV preparation with RNase III (Fig. 8,A). The IPNV preparations depleted of dsRNA infected RTG-2 cells and induced CPE followed by apoptosis (Fig. 8,A). Infected RTG-2 cells were rescued from CPE if the cells expressed fgTLR22 beforehand (see the wells of moi of 0.01) (Fig. 8 A).

We then quantitatively determined fgTLR22-mediated inhibition of IPNV infection. IPNV (moi = 0.1) (10 μg/ml) with (Fig. 8,B) or without (Fig. 8,C) naked dsRNA was added to TLR22-expressing RTG-2 cells. We measured the virus titer (TCID₅₀/ml) in the supernatants of control and fgTLR22-expressing cells. Twenty-four hours after infection, the IPNV titer was greatly decreased in fgTLR22-expressing cells compared with untransfected cells (Fig. 8, B and C). The IFN message was efficiently induced in cells with fgTLR22 concomitant with IPNV infection. Control cells, however, barely raise the IFN message in response to IPNV (Fig. 8 D), suggesting that fgTLR22 has a critical role in IFN-mediated antiviral defense in fish cells. In conclusion, fgTLR22 governs antiviral response to a dsRNA virus IPNV via induction of fish IFN.

Herein, we demonstrated that TLR22 is a dsRNA recognition receptor. Since fish possess both TLR3 and TLR22, they have a dual dsRNA recognition system. TLR3 and TLR22 recruit a common adaptor, TICAM-1. TLR22 preferentially recognizes long-sized dsRNA, localizes to the cell surface, and is widely distributed to tissue/organs (20). This is the first study to reveal that the TICAM-1 pathway serves as a key alert for an RNA virus sensor on the vertebrates.

According to bootstrap probability analysis, TLR22 does not belong to the TLR3 family; instead, it is proximal to mouse TLR13, which has not been characterized as a dsRNA-recognizing TLR. Thus, two arms of the TICAM-1 pathway have evolved as dsRNA receptors in fish, and only one (TLR3) has been preserved in mammals. Development of TLR22 instead of TLR3 may have some merit for protection against viruses with dsRNAs by augmenting the susceptibility of the local IFN response to long RNA duplexes in the vertebrates.

We wanted to clarify why teleosts need a cell surface RNA recognition system. Fish live in water and are exposed to many kinds of negative-stranded RNA viruses belonging to the Rhabdoviridae and to dsRNA viruses (40, 41). Bacteria such as Rhodovulum sulfidophilum and perhaps other species extracellularly liberate ribosomal and transfer RNAs (42). Thus, the sea may contain RNA viruses and RNA products of microbial origin. The sea is home to a unique and mysterious microbial environment. During evolution, vertebrates in water may have been protected from these pathogens by developing the set of RNA-sensing TLRs and the IFN system, which are distinct from those expressed in land animals. Our results indicate that RNA-sensing by TLRs protects fish from spreading or exacerbating infection. Viral RNAs often form a stem-loop or duplex signature (43) and are released from infected individual fish into the sea. TLR22 may sense such floating RNA as an infection threat.

Overexpressed teleost TLR22 protects host cells from infection with IPNV, which is a naked bisegmented dsRNA virus belonging to the family Birnaviridae (44). Birnaviruses have a single T = 13 icosahedral shell composed of 120 subunits and they lack the characteristic inner capsid. Aquatic birnaviruses are distributed worldwide, can infect a range of fish and shellfish species (44, 45), and are viral pathogens that cause problems in fry and young fish. Although teleosts have the gene encoding a putative ortholog of the cytoplasmic RNA sensor MDA5 (24), IPNV efficiently infect teleost cells unless TLR22 is expressed in some population of cells. Thus, fish MDA5 is insufficient for protection against this type of dsRNA virus. Although not all cells express TLR22, IFN seems to be sufficiently induced by TLR22-expressing cells (Fig. 8 D) to provide antiviral environment in surrounding cells, resulting in host cell protection. However, how TLR22 detects the IPNV infection remains to be clarified. The necessity of TLR22 and the mode of its dsRNA recognition in fish are of interest for further investigation.

A notable difference between fgTLR22 and surface-expressed human TLR3 is that fgTLR22 is ubiquitous whereas surface TLR3 is expressed exclusively together with endosomal/ER TLR3 in TLR3-positive cells. In humans, cell-surface TLR3 is therefore distributed limitedly to the fibroblasts and epithelial cells. We previously found human TLR3 on the cell surface of fibroblasts, which therefore binds anti-TLR3 mAb, TLR3.7 (46). TLR3 on the fibroblasts recognizes exogenously added poly(I:C) to confer IFN-β induction (46). The IFN-inducing properties of human TLR3 by fibroblasts are blocked by this mAb. The result is a unique and exceptional TLR3 feature that reflects a differential TLR3 function (10). Ultimately, in humans TLR3 is expressed in the cytoplasmic compartments and on the surface of fibroblasts (33). Other reports also found that human bronchial, bile duct, and intestinal epithelial cells express TLR3 on the cell-surface membrane (47, 48, 49, 50). In this view, surface-expressed human TLR3 is a functional remnant of fish TLR22: TLR3 functions in the mucosal region wherein body fluids are continuously in contact with the flora. Because the cell surface-associated dsRNA recognition is important even in humans, TLR3 is expressed on human fibroblasts and epithelial cells (50, 51, 52).

In this study, we found that fgTLR3 and fgTLR22 recognize its ligand dsRNA at the different sites and recruit fgTICAM-1. Although TLR22 expression is limited to part of the cell surface, a colocalization study with cholera toxin did not support its expression being restricted to lipid rafts (data not shown). Unlike the case of human TLR3, the chloroquin treatment of HEK293 cells expressing TLR22 did not abrogate the activation of the IFN-β reporter gene by poly(I:C) (data not shown). Thus, it is unlikely that TLR22 captures poly(I:C) at the endosome, as does TLR3. Human and mouse TLR3 in myeloid dendritic cells have a role in driving the cells to a maturation stage sufficiently activating NK and T cells (9, 53). It is intriguing whether TLR22 possesses this function in dendritic cells.

In this context, the question is how TLR22 assembles TICAM-1 to transmit the dsRNA-recognition signal. Possible answers may lie in the fish-specific TLR22 pathway and in the functional difference between mammalian and teleost TICAM-1. A recent study of zfTICAM-1 suggested that overexpression of zfTICAM-1 activates the zfIFN promoter but that TICAM-1 does not interact with zfTRAF6 (30). The zfTICAM-1 N-terminal region does not contain a TRAF6-binding motif, and the C-terminal region of zfTICAM-1 sufficiently activates the IFN promoter, suggesting the involvement of RIP1-mediated NF-κB activation in zfIFN promoter activation (30). In OLHd-rRe3 cells, a zebrafish minimal Mx promoter that contains two ISRE barely responds to overexpressed fgTICAM-1 or its N-terminal-deleted form, but it enhances activation of the zfIFN promoter, as shown by reporter assay. In human TICAM-1, the N-terminal region is mandatory for IRF-3-mediated type I IFN induction in the human system (14, 54). Ultimately, fish TICAM-1 behaves like human TICAM-1, although fish TICAM-1 does not mediate IRF-3 for activation of the IFN-β promoter (30). It takes a longer period for fish cells (compared with human cells; see Refs. 54 , 55) to evoke TLR-TICAM-1 interaction in response to poly(I:C), which may be attributable to the temperature (∼20°C) where the fish cells are grown, rather than to the difference of signaling mode. Hence, the TICAM-1 pathway is conserved across fish and humans, but its molecular bases for IFN induction are different between them. Our speculation is that fish cells have an IFN output similar to that of human cells (54, 55) but that the signal cascade for IFN production is modally different. Teleost TICAM-1, which is partly dissimilar to human TICAM-1 in its structure, might explain the differential selection of their signal pathways.

An alternative possibility is that the IFN-inducing capacity due to the recruitment of TICAM-1 lies within the molecular feature of TLR22. Supporting evidence is that surface-expressed TLR22 retains the capacity to confer the responsiveness to poly(I:C) not only in fish RTG-2 cells but also in human HEK293 cells. FgTLR22 and even only the TIR domain of fgTLR22 can couple with human TICAM-1 as well as fish TICAM-1 to activate the human and fish ISRE promoter. Either way, TLR3 is not the only partner of TICAM-1 in the water vertebrates.

We are grateful to Drs. A. Ishii, M. Shingai, M. Sasai, K. Funami, and T. Ebihara in our laboratory for their critical discussions. Thanks are also due to Drs. T. Fujita (Kyoto Univiversity, Kyoto), T. Taniguchi (Tokyo Univiversity, Tokyo), and T. Maniatis (Harvard University, Boston, MA) for providing their plasmids.

The authors have no financial conflicts of interest.