Structural and Functional Analyses of the Human Toll-like Receptor 3

Toll-like receptors (TLRs) play critical roles in bridging the innate and adaptive immune responses. The human TLR3 recognizes foreign-derived double-stranded RNA and endogenous necrotic cell RNA as ligands. Herein we characterized the contribution of glycosylation to TLR3 structure and function. Exogenous addition of purified extracellular domain of TLR3 (hTLR3 ECD) expressed in human embryonic kidney cells was found to inhibit TLR3-dependent signaling, thus providing a reagent for structural and functional characterization. Approximately 35% of the mass of the hTLR3 ECD was due to posttranslational modification, with N-linked glycosyl groups contributing substantially to the additional mass. Cells treated with tunicamycin, an inhibitor of glycosylation, prevented TLR3-induced NF-κB activation, confirming that N-linked glycosylation is required for bioactivity of this receptor. Further, mutations in two of these predicted glycosylation sites impaired TLR3 signaling without obviously affecting the expression of the protein. Single-particle structures reconstructed from electron microscopy images and two-dimensional crystallization revealed that hTLR3 ECD forms a horseshoe structure similar to the recently elucidated x-ray structure of the protein expressed in insect cells using baculovirus vectors (Choe, J., Kelker, M. S., and Wilson, I. A. (2005) Science 309, 581-585 and Bell, J. K., Botos, I., Hall, P. R., Askins, J., Shiloach, J., Segal, D. M., and Davies, D. R. (2005) Proc. Natl. Acad. Sci. U. S. A. 102, 10976-10980). There are, however, notable differences between the human cell-derived and insect cell-derived structures, including features attributable to glycosylation.

Toll-like receptors (TLRs) play critical roles in bridging the innate and adaptive immune responses. The human TLR3 recognizes foreign-derived double-stranded RNA and endogenous necrotic cell RNA as ligands. Herein we characterized the contribution of glycosylation to TLR3 structure and function. Exogenous addition of purified extracellular domain of TLR3 (hTLR3 ECD) expressed in human embryonic kidney cells was found to inhibit TLR3-dependent signaling, thus providing a reagent for structural and functional characterization. Approximately 35% of the mass of the hTLR3 ECD was due to posttranslational modification, with N-linked glycosyl groups contributing substantially to the additional mass. Cells treated with tunicamycin, an inhibitor of glycosylation, prevented TLR3-induced NF-B activation, confirming that N-linked glycosylation is required for bioactivity of this receptor. Further, mutations in two of these predicted glycosylation sites impaired TLR3 signaling without obviously affecting the expression of the protein. Single-particle structures reconstructed from electron microscopy images and two-dimensional crystallization revealed that hTLR3 ECD forms a horseshoe structure similar to the recently elucidated x-ray structure of the protein expressed in insect cells using baculovirus vectors ( Proteins that recognize pathogen-associated molecular patterns are key factors in the cascade of events from the detection to the elimination of an invading organism. This form of innate immunity is conserved in eukaryotes. For example, the Drosophila melanogaster Toll protein is responsible for resistance to fungal and bacterial infections (3,4), and plants can encode disease-resistance proteins that are important in determining the outcome of infection (5). The vertebrate pathogen-detecting proteins called Toll-like receptors (TLRs) 5 are key players in the activation of both the innate and adaptive arms of the immune system (6 -9).
The TLRs and related pathogen sensors contain leucine-rich repeat motifs that form docking sites for pathogen ligands or adaptors that bind pathogen ligands, the binding of which will activate signal transduction pathway(s) (10 -12). TLR3 recognizes double-stranded RNA and may be a part of a redundant sensor system to detect viral infections (13)(14)(15). Although specific features in the ligands required to interact with TLR3 remain to be identified, TLR3 is activated by polyinosinepolycytidylic acid (poly(I:C)) and has been reported to be activated by RNAs extracted from necrotic cells (16).
A number of issues concerning TLR3 structure and function remain to be elucidated. For example, TLR3 can apparently act both on the surface of the plasma membrane, as it does in fibroblasts, and by attaching to the membranes of intracellular vacuoles, where it is proposed to act in immature dendritic cells (17,18). The trafficking of TLRs should be influenced by glycosylation in general, and N-linked glycosylation of TLR2 and TLR4 has been shown to play essential roles in its localization (19,20). A significant recent advance in TLR3 was the elucidation of a 2.1 Å structure of the soluble ectodomain by Choe et al. (1). Bell et al. (2) independently elucidated a highly similar structure. In both studies, the crystallized ectodomains were produced in a baculovirus expression system and formed a horseshoe-shaped solenoid structure that was extensively decorated with glycosyl modifications, some of which were partially resolved in the structure. Whether the glycosylations are important in TLR3 localization and/or function were not directly addressed in these works (1,2). However, de Bouteiller et al. (21) showed that a change of Asn-247 to an arginine in TLR3 negatively affected TLR3 activity.
We expressed the extracellular domain (ECD) of the human TLR3 in human embryonic kidney cells (HEK 293T) and demonstrated that it was modified with N-linked glycosylations. Using the Glc-NAc-transferase inhibitor tunicamycin, a concentration-dependent inhibition of TLR3 activity was observed. Systematic mutational analysis of the predicted N-linked glycosylation sites identified two asparagine residues in leucine-rich repeats 8 and 15 that are important for TLR3 activity. The mutant proteins remain expressed at levels similar to wild type. In addition, because our ectodomain was produced in human cells as opposed to insect-derived protein, which differs in the presence of glycosylations, we independently established the structural framework for the hTLR3 ECD using electron microscopy of single particles and two-dimensional crystals in conjunction with image reconstruction.

MATERIALS AND METHODS
Expression and Purification of hTLR3 ECD-The full-length human TLR3 cDNA, identical to accession number U88879, was amplified from human dendritic cells and cloned into pcDNA3.1. The extracellular domain of TLR3, consisting of amino acids 1-703, was also cloned into pcDNA3.1 to generate the hTLR3-ECD antigen with a C-terminal hexahistidine tag. The protein was expressed and secreted by transiently transfected HEK 293E cells. During the secretion process, residues 1-26 were cleaved, generating a protein containing residues 27-703. The secreted protein was purified from the supernatant of 293E cells using a BD-TALON TM metal affinity resin (BD Biosciences) and further purified by gel filtration chromatography.
Cell-based Assays-HEK 293T cells were harvested from an actively growing culture and plated in CoStar White 96-well plates at 4.4 ϫ 10 4 /ml for transfection. When the cells were ϳ65-90% confluent, they were transfected with a mixture of the Lipofectamine 2000 reagent (Invitrogen) and plasmids pNF-〉-Luc (Stratagene Inc., La Jolla, CA), pFL-TLR3, and phRL-TK (Promega Corp., Madison, WI) that, respectively, code for the firefly luciferase reporter, full-length wild-type and/or TLR3s with mutations in the ectodomain, and the Renilla luciferase transfection control. The cells were allowed to incubate for 24 h to allow expression from the plasmids. Poly(I:C) (1 g/ml, unless stated otherwise) was then added to appropriate sets of transfected cells to induce TLR3-dependent NF-B activity. After another 6 h of incubation, the cells were harvested using the Dual Glo Luciferase Assay System reagents (Promega). Luminescence was quantified using the FLU-Ostar OPTIMA Plate Reader (BMG Labtech, Inc).
Fluorescence-activated Cell Sorting (FACS) Analysis-FACS analyses were performed with 293T cells grown in 6-well collagen-coated plates (BD Biosciences) at a concentration of 2 ϫ 10 6 cells/well. The cells were transfected with 1 g of the appropriate plasmids using Lipofectamine 2000 (Invitrogen Inc.). Twenty-four hours after transfection, the cells were harvested and washed twice with ice-cold FACS buffer (1ϫ phosphate-buffered saline, 10 mM phosphate, 150 mM NaCl, pH7.4, ϩ 3% fetal bovine serum ϩ 0.04% sodium azide) before suspension at ϳ2 ϫ 10 7 cells/ml in FACS buffer. The cells were stained for 30 min at 4°C with 1 g of phycoerthrin-labeled anti-human TLR3 mAb (clone TLR3.7; eBioscience, San Diego, CA) or a negative control mouse IgG1 control antibody. The antibodies were added to aliquots of the cells in 96-well plates and incubated for 30 min on ice in the dark. The cells were washed twice with FACS buffer to remove unbound antibody and then resuspended in FACS buffer. Viaprobe (BD Biosciences) was added to the cultures to exclude dead cells. The cells were transferred to the appropriate tubes and analyzed using a FACSCalibur machine (BD Biosciences).
Western Blots-293T cells were transiently transfected with wild type, mutant TLR3, or with control pcDNA as described above. 36 h posttransfection, cells were lysed using passive lysis buffer (Promega Inc.) and sonicated to degrade chromosomal DNA. Equal amounts of proteins from each sample were separated on NuPAGE 4 -12% bis-tris gel (Invitrogen) and blotted onto polyvinylidene difluoride membrane. The anti-TLR3 antibody IMG315A (Imgenex Inc.) was used in Western blots because the mAb used in the FACS analysis was unable to detect denatured TLR3. The blots were developed with peroxidase-conjugated secondary antibodies and the ECL-Plus Western blotting detection sys-tem (Amersham Biosciences). When the effects of tunicamycin or swainsonine were examined, the cells were transfected for 6 h before addition of the drugs.
Mutational Analysis of TLR3-N-linked glycosylation sites in hTLR3 ECD were predicted by the computer program Motif (NCBI). Site-directed mutagenesis was performed using oligonucleotides annealed to the target sequence and the QuikChange kit (Stratagene, Inc.). Several clones that resulted from the mutational analysis were sequenced to confirm that no unintended mutations were made. Mutant clones with affected activity were sequenced in their entirety.
Mass Spectrometry-Samples were concentrated by use of a NanoSep centrifugation concentrator (Pall Inc.), desalted using C-18 Zip Tips (Millipore Corp., Billerica, MA), and eluted with 50% acetonitrile/0.1% trifluoroacetic acid. One microliter of the sample was co-crystallized with the matrix 2,5-dihydroxybenzoic acid in acetonitrile/water (50:50) containing 0.1% trifluoroacetic acid. MALDI experiments were recorded on an ABI Voyager-DE STR mass spectrometer. To deglycosylate hTLR3 ECD, 1 g of the protein was treated with a mixture of N-glycanase, O-glycanase, and sialidase for 24 h at 37°C. The products were analyzed by MALDI-TOF spectroscopy as described above. Also, the proteins were subjected to electrophoresis on a 4 -12% denaturing protein gel and then stained with the glycosylation-specific dye, SYPRO Ruby Protein gel stain (Invitrogen Inc.).
Electron Microscopy and Image Processing-For single particle imaging, 2.5 l of a 50-g/ml solution of TLR3 ECD in Tris-buffered saline (pH 7.4) was either applied to freshly glow-discharged carbon-coated copper grids (G400) or to holey carbon films. The samples were washed with distilled water and stained for 1 min with an aqueous solution of uranyl acetate (1% w/v, pH 4.25) for the G400 grids or stained with 5% (w/v) ammonium molybdate (pH 6.0) containing 0.1% (w/v) trehalose for the holey carbon film according to Harris and Scheffler (22). The two-dimensional crystals were grown using the lipid monolayer approach (23). Briefly, 0.2 l of nickel-nitrilotriacetic acid-1,2-dioleoylglycerol-3[phospho-L-serine] (100 g/ml) and egg phosphatidylcholine (400 g/ml) in chloroform/hexane (1:1) were layered onto 15 l of 10 g/ml of hTLR3ECD in 50 mM Tris (pH 7.5), 100 mM Li 2 SO 4 , and 0.6 M sodium formate and incubated overnight in a humid chamber at 4°C. Crystals attached to the lipid monolayer were transferred to transmission electron microscopy grid by placing a hydrophobic carbon-coated grid onto the solution surface for 1 min, followed by washing and staining as described above. Specimens were observed in a JEOL 1200EX transmission electron microscope operated at an acceleration voltage of 100 kV. Electron micrographs were recorded at a calibrated magnification of ϫ39,000 on Kodak 4489 electron image sheet film and digitized with a Leafscan 45 using a scan step corresponding to 0.514 nm/pixel at the specimen level.
Electron micrographs of TLR3 two-dimensional crystals were processed using the CRISP2 software package (24). Pixel areas of 512 ϫ 512 were screened for the best order, indexed, and refined, and Fourier projection maps were calculated in p2 or p22 1 2 1 , depending on the crystal type. The p2 symmetry was established based on the presence of local two-folds. For the second crystal type, p22 1 2 1 was selected based on the phase residual criterion, unit cell parameters, and systematic absences in the Fourier transforms of the type h00 with h representing odd and 0k0 with k representing odd.
The three-dimensional image reconstruction of single particles was performed using the EMAN software package (25). Approximately 2000 particles were selected with the program BOXER. After filtering and centering, STARTNRCLASSES was used to classify the raw particles into 50 classes; each class was averaged to improve signal:noise ratio.
Eight classes were selected to generate a first three-dimensional reconstruction with C1 symmetry using STARTANY. This first three-dimensional reconstruction served as a starting point for iterative refinement until the reconstruction was stable as judged by Fourier shell correlation. The three-dimensional volume of the final reconstruction was displayed with the program Chimera (26).

Purified Recombinant hTLR3 ECD Can Affect Signaling by Full-length
TLR3-We expressed and purified amino acids 27-704 of the extracellular domain of human TLR3 in 293T cells to initiate characterization of its structure and function. The recombinant protein contained six additional histidine residues at the C terminus to facilitate purification and is henceforth named hTLR3 ECD. In SDS-PAGE, purified hTLR3 ECD migrated as a mass of ϳ110 kDa, much larger than the predicted mass of ϳ75 kDa (Fig. 1A). The difference is likely to be because of posttranslational modifications. To confirm that this is the case, we translated residues 1-704 of TLR3 in rabbit reticulocyte lysates and found that the resultant protein migrated at the expected mass (Fig. 1B). These results indicate that hTLR3 ECD is heavily modified, likely by glycosylation, a feature that we will examine below.
To examine the function of the hTLR3 ECD and how mutations affected TLR3 activity, we used a cell-based assay that measures TLR3dependent activation of gene expression. Briefly, the assay was performed by transiently transfecting 293T cells with a mixture of three plasmids: one expressing wild-type TLR3, a second reporter plasmid in which firefly luciferase sequence was expressed from a promoter containing two NF-〉 binding sites, and a third control plasmid that con-stitutively expresses the Renilla luciferase under the control of the Herpes Simplex Virus thymidine kinase promoter. The assay thus detects TLR3 function by the amount of reporter firefly luciferase activity produced through activation of the NF-kB transcription factor. The Renilla luciferase activity was used to detect any potentially toxic effects of our manipulations. HEK 293T cells are suitable for these assays because they do not produce detectable TLR3 activity (21) and do not contain detectable TLR3 in Western blots (Fig. 1C). Furthermore, this reporter assay is strictly dependent on the transient expression of TLR3 (data not shown) and specifically dependent on the addition of poly(I:C) (Fig. 1D). A typical assay will increase the ratio of firefly luciferase to Renilla luciferase by 4-to 7-fold (Fig. 1D).
When hTLR3 ECD was added to the reporter assay to a final concentration of 12.5-100 g/ml, we observed a concentration-dependent inhibition of wild-type TLR3-induced NF-B activity (Fig. 1D), demonstrating that hTLR3 ECD can effectively compete with wild-type human TLR3 for poly(I:C)-induced NF-B activation. These results indicate that hTLR3 ECD is suitable for additional biochemical analyses.
hTLR3 ECD Is Glycosylated-Mass spectrometry was used to examine the mass of hTLR3 ECD with greater precision. The purified hTLR3 ECD ionized as a large number of peaks that ranged from 105 to 115 kDa, indicating that it exists in a highly heterogeneous state (data not shown). A similar proportion of the extra mass in the ectodomain of TLR2 and TLR4 is due to glycosylation (19).
To visualize the glycosyl groups on hTLR3 ECD directly, we separated hTLR3 ECD by SDS-PAGE and stained the gel with a glycosylation-specific fluorescent dye, SYPRO Ruby ( Fig. 2A). Treatment with NANase or O-glycosidase, which releases ␣ (2, 3)-linked sialic acids and unsubstituted Ser-or Thr-linked GalGalNac, respectively, did not alter the electrophoretic mobility of hTLR3 ECD or abolish binding by SYPRO Ruby dye. PNGlycosidase F, which releases the N-linked glycans, not only decreased the molecular mass of hTLR3 ECD from ϳ115 to ϳ90 kDa ( Fig. 2A) but also reduced the amount of bound dye. An ϳ90-kDa band was detected in amounts similar to the input when stained with Coomassie Blue (data not shown) These results confirm those from mass spectrometry that hTLR3 ECD contains a significant amount of glycosylation and further demonstrate that the majority of the removable glycosyl groups are N-linked. The fact that the mass of the protein was not reduced to the unmodified 75 kDa, the residual staining with Ruby Red, and the heterogeneous peaks in mass spectroscopy suggest that additional modifications other than N-linked glycosylations are present.
Glycosylation of TLR3 Is Required for TLR3 Function-Inhibitors of protein glycosylation were used to examine whether glycosylation is required for TLR3 function in 293T cells (Fig. 2, B-D). Tunicamycin inhibits the addition of N-acetylglucosamine, the first sugar of N-linked glycosylation, whereas swainsonine affects the processing of the terminal mannoses by lysosomal mannosidase II that could lead to more complex glycosylation structures (27,28). Both compounds were added between 0.2 and 5 g/ml. These concentrations do not exhibit toxicity as determined by the normal expression of Renilla luciferase expressed from the constitutive thymidine kinase promoter (data not shown). Tunicamycin inhibited TLR3-dependent induction of the NF-B reporter in a concentration-dependent manner (Fig. 2B). However, swainsonine did not apparently affect TLR3 signaling even at 5 g/ml, a concentration that should be sufficient to inhibit glycosylation (Fig. 2C) (29). Another mannosidase II inhibitor, deoxymannojirimycin, also did not affect TLR3 activity in our assay, 6 suggesting that TLR3 activity is 6 C. Kao, data not shown.

TLR3 Structure and Function
not affected by modifications that require mannosidase II. The differential effects of tunicamycin and swainsonine have been reported for a number of glycoproteins (30,31).
To determine whether treatment with tunicamycin or swainsonine affected the level of TLR3 expression, transfected HEK297 cells were analyzed by Western blot. Tunicamycin addition to the cultures significantly decreased the amount of full-length TLR3, whereas the addition of swainsonine to 5 g/ml did not have an obvious effect on TLR3 levels (Fig. 2D). The decrease in TLR3 levels was consistently observed in three independent transfection experiments and suggests that proper N-linked glycosylation of TLR3 is necessary for both the expression and function of TLR3.
We also examined the level of TLR3 expression upon tunicamycin and swainsonine treatment using FACS with a monoclonal antibody to the ECD of hTLR3, TLR3.7. Human 293T cells transfected with the empty vector control had a high background, with a mean fluorescence intensity (MFI) of 51.6 (Fig. 2D). Cells transfected with wild-type TLR3 had an MFI of 79.8. When the transfected cells were treated with tunicamycin, the MFI of TLR3-positive cells was reduced to 60.1, suggesting that tunicamycin treatment does affect TLR3 expression. Again, treatment with swainsonine had no significant effects on TLR3 levels as detected by FACS analysis.
Mutation of Potential N-linked Glycosylation Sites-The hTLR3 ECD has nine sites that perfectly matched the N-linked glycosylation motif of Asn-X-Ser/Thr (Fig. 3A). Of these, five (Asn-57, -196, -247, -275, and -413) were conserved in all of the mammalian species we examined, whereas the rest varied in the mammalian orthologs (Fig. 3A). We individually mutated the conserved asparagines within these nine putative N-linked glycosylation sites to alanines. An additional four asparagines that do not match the conserved motifs (Asn-70, -124, -252, and -388)

. Recombinant hTLR3 ECD is glycosylated, and glycosylation is required for TLR3 function in 293T cells.
A, effects of deglycosidases on the mobility of the hTLR3 ECD in a denaturing protein gel electrophoresis. The enzymes used to treat 300 ng of hTLR3 ECD are listed above the corresponding sample in the gel. M, molecular mass markers (Candycane markers) purchased from Invitrogen; ⌽, untreated control; N, treated with NANase; O, treated with O-glycosidase; P, treated with PNGase F. Where all three deglycosidases were used, the corresponding three symbols were noted above the appropriate lane (NOP). Deglycosidases were purchased from Sigma. All samples were treated for 24 h. The denaturing protein gel was subsequently stained with the glycosylation-specific dye SYPRO Ruby and visualized on a transluminator. B, the effects of tunicamycin treatment on TLR3-dependent signaling of a reporter construct driven from the NF-B promoter. Each symbol represents the ratio of the firefly luciferase (indicating TLR3 signaling) to Renilla luciferase (transfection control). The horizontal bar indicates the mean value of all of the assays. For every condition, usually ten independent cultures of transfected 293T cells were assayed. The presence of the tunicamycin (in g/ml) is shown below the graph. The presence or absence of poly(I:C) (PIC) at 10 g/ml is denoted by "ϩ" or "Ϫ", respectively. C, effects of swainsonine on TLR3 activity in the cell-based assay. D, Western blot and FACS analysis of TLR3 expression in transiently transfected 293T cells, as affected by tunicamycin or swainsonine. TLR3.7 was used for the FACS analysis but not the Western blots because it does not recognize denatured TLR3 (17). Western blots were treated with the monoclonal antibody IMG315A from Imgenix Inc. Tunicamycin or swainsonine was added to a final concentration of 5 g/ml. were also mutated. For these four residues, Choe et al. (1) observed electron densities associated with N-glucosamines for residues Asn-124 and Asn-252 and Bell et al. (2) observed glycosylations associated with Asn-70 and Asn-252, but not Asn-124. We do not have evidence that Asn-388 is associated with glycosylations. Plasmids containing the mutant constructs were individually transfected into HEK 293T cells, and the effects on the NF-B reporter activity were assessed. Changes in the four asparagines that do not match the consensus N-linked glycosylation sites did not significantly affect TLR3-dependent activation of the reporter assay, although the mutation at Asn-252 decreased NF-B activity slightly in this assay and had modest effects in other assays ( Fig.  3B and data not shown). However, changes in two of the five conserved potential N-linked glycosylation sites, N247A and N413A, consistently reduced TLR3-mediated NF-B activation to half or less of the wildtype level (Fig. 3C). We also tested both the N413A and N265A mutations at a range of plasmid concentrations from 5 to 20 ng/transfection and found that the defect relative to wild type was observed at all concentrations tested (data not shown). The change at Asn-247 is in agreement with the observations of de Bouteiller et al. (21), who characterized a change of this residue to an arginine and found the resultant protein to be nonfunctional. A version of TLR3 containing both the N247A and the N413A mutations reduced TLR3 activity to background (Fig. 3C).
Expression and Localization of Mutant TLR3 Proteins-The lack of TLR3 activity with mutations N247A and N413A could be either because of an effect on protein expression, protein localization, or on a specific activity such as ligand binding. To examine the expression of the mutant TLR3 proteins, we first performed a Western blot with mAb IMG315A. The results from three independent experiments show that the levels of all three mutant proteins, N247A, N413A, and the double mutant, were comparable with the WT TLR3 in 293T cells and to TLR3 with a mutation at Asn-196 ( Fig. 4A and data not shown), suggesting that the level of protein expression was not responsible for the decrease in activity in the mutants. Notably, we consistently observed that each of the mutant proteins tended to migrate at a slightly lower electrophoretic position relative to the wild-type TLR3, suggesting that the mutations did affect posttranslational modifications of the proteins.
FACS analysis was used to determine whether the mutant proteins are expressed on the surface or within transfected 293T cells. Staining for intracellular protein was also performed with detergent-permeabilized cells. For cell surface expression, the background MFI in cells transfected with pcDNA was at 11, whereas cells transfected with wild-type TLR3 showed MFI levels over 27. All mutants, including N247A and N413A, had MFI Ͼ18, indicating that all of the proteins were expressed on the cell surface at levels comparable with the WT TLR3 (Fig. 4B). The background level for permeabilized cells was higher, likely because of the monoclonal antibody recognizing other proteins nonspecifically (see Fig. 1C and Ref. 17). However, cells transfected to express the mutant proteins had significantly higher MFI compared with the control (Fig. 4B), confirming the results of the Western blots that these mutant proteins are expressed. The effects of the mutations thus cannot be attributed simply to a defect in the lack of expression.
Electron Microscopy of hTLR3 ECD-Because hTLR3 ECD protein was produced in human cells, it is reasonable to expect that it would be more extensively glycosylated in comparison with the protein crystallized by Choe et al. (1) and Bell et al. (2). Furthermore, most of the glycosyl groups were not well resolved in the x-ray structures. We used electron microscopy to investigate possible molecular interactions and the quaternary structure of hTLR3 ECD in solution and on a lipid surface. To this end, two-dimensional crystals were grown on lipid monolayers. Several different types of two-dimensional crystals were observed and imaged. Fourier transformations of selected crystalline areas displayed readily discernible reflections to ϳ3 nm resolution. The major two-dimensional crystal type is shown in Fig. 5, A and B. This type of crystal displayed a p22 1 2 1 symmetry (a ϭ 26.5 nm, b ϭ 12.5 nm, ␥ ϭ 90 o ) and featured one subunit of hTLR3 ECD interacting through a terminus to the central portion of a second subunit. The other crystal type (Fig. 5, C and D) displayed p2 symmetry (a ϭ 9.7 nm, b ϭ 18.2 nm, ␥ ϭ 109.9°), with each asymmetric unit having two horseshoe-shaped molecules related by a 2-fold axis. Furthermore, there is a notable deficit in the middle of the horseshoe-shaped molecule that was not observed in the structure of Choe et al. (1). This central deficit could not be seen in the other structure because of the overlap between the two subunits.
We noted that the interactions between the two monomers in both crystal forms are different from the dimer proposed by Choe et al. (1) and Bell et al. (2). In addition, the fact that at least two different ways of crystal packing were observed suggests that the interactions between

TLR3 Structure and Function
TLR3 ECD molecules can be variable. In the p22 1 2 1 crystal, two modes of interaction between the molecules exist even within one-unit cells : (i) molecules are packed in parallel "filaments" of alternating polarity consisting of horseshoe-shaped TLR3 molecules in a flip-flop arrangement where one terminal domain of a horseshoe interacts with the central domain of another molecule, and (ii) two molecules interact via their corresponding terminal domains, thereby forming centers of 2-fold symmetry. A more complete understanding of the native quaternary structure of TLR3 must await the analysis of full-length protein in the lipid bilayer.
We also performed a three-dimensional reconstruction of hTLR3 ECD using a single-particle analysis. Two different sample preparation methods were used. First, we used glow-discharged continuous carbon films to absorb hTLR3 ECD, followed by staining of the molecules with 1% uranyl acetate (Fig. 6, A and B). The second method suspends the sample over holey carbon film followed by embedding and negative staining with trehalose/ammonium molybdate (Fig. 6B). Although the contrast is lower, the latter method better preserves the proteins in a state closer to their native conformation (22).
Well separated single particles of ϳ9-nm diameter were readily discernible with both sample preparation methods (Fig. 6, A and B). The observed structures appeared to be monomeric in solution and have a characteristic horseshoe shape. Classification and average of ϳ2000 particles by multistatic analysis and multireference alignment with IMAGIC-5 software revealed heterogeneity of hTLR3 ECD in terms of the sizes and shapes of hTLR3 ECD (Fig. 6B, inset). For example, some molecules exhibit a more opened horseshoe structure, whereas in others the termini portions appear to be almost in contact with each other (Fig. 6B). Although some of these variations may be because of subtle differences in the orientations of the molecules relative to the electron beam, others may be caused by different conformation and perhaps differences in the patterns of glycosylation between molecules. This degree of conformational flexibility would not be difficult to detect when the protein is in a more rigid crystal lattice. In addition, it should be noted that some peanut-shaped particles with two distinct density maxima were occasionally observed (Fig. 6A, inset). These projections likely represent side-on views.
The three-dimensional reconstruction using the data obtained from single particles negatively stained with uranyl acetate and C1 symmetry resulted in a structure with ϳ2.5-nm resolution (Fig. 6, C and D). The outer diameter of the final structure is 9 -9.5 nm, and the inner boundary of the horseshoe measures ϳ3.5-4 nm. The extra densities in our hTLR3 ECD are likely contributed from glycosylation. However, higher resolution structures would be needed to accurately locate the sites of glycosylation. Lectin-hTLR3 ECD Interaction-We used electron microscopy and single-particle analysis to directly examine the extent of glycosylation in hTLR3 ECD. This approach required that we identify a lectin that will bind hTLR3 ECD. Seven commercially available lectins (Vector Laboratories; Burlingame, CA) were tested for the ability to detect decreasing amounts of hTLR3 ECD spotted on membranes. The Ricinus commonalis lectin (120 kDa) recognize hTLR3 ECD better than others and was selected for this analysis. It was mixed at a 2:1 ratio with the hTLR3 ECD for 1 h and stained with uranyl acetate. Particles that have additional density from the hTLR3 ECD were collected, and class averages were obtained. The results show that the R. commonalis lectin was able to bind at multiple positions on the outer perimeter of the horseshoe structure (Fig. 7A), providing experimental evidence confirming that glycosyl groups exist at multiple sites of the hTLR3 ectodomain.
Modeling of the hTLR3 ECD-As shown above, our mutagenesis data have implicated that mutations at Asn-247 and Asn-413 significantly affected TLR3 function and that a third residue, Asn-252, may have a modest effect on TLR3 function. The locations of these residues were modeled based on the structure of the TLR3 ECD solved by Bell et al. (2). Asn-252 is pointed into the concave surface of TLR3 ECD (Fig. 7B). Asn-247 were found to lie in the outer convex surface, and Asn-413 was closer to the inner concave surface of the molecule.
Lastly, given that the structure of the hTLR3 ECD derived from our studies appears to be somewhat larger than the molecules reported by Bell et al. (2) and Choe et al. (1), we superimposed the structures derived from x-ray crystallography with our lower resolution structures (Fig.  7C). We observed that the electron microscopy structure had the largest difference from those from x-ray crystallography in the diameter of the solenoid, especially near the middle of the protein. We speculate that this difference may be because of differences in glycosylation between the protein expressed in insect and human cells.

DISCUSSION
We have analyzed the structure and function of the human TLR3 ectodomain and determined that the protein can be expressed in human cells in a highly glycosylated form, with N-linked glycosylations contributing significantly to additional mass of the hTLR3 ECD. Additionally, treatment with tunicamycin, which inhibits N-linked glycosylation, abolished reporter signaling by a process stimulated by poly(I:C), likely by reducing the amount of TLR3 expressed. Furthermore, mutations in two of the thirteen potential glycosylation sites in hTLR3 ECD prevented signaling but did not apparently affect expression or cell surface localization. Lastly, we used negatively stained electron micrographs and molecular modeling to visualize hTLR3 ECD.
The structure for the hTLR3 ECD we observed is a solenoidshaped horseshoe, in agreement with the structures of Choe et al. (1) and Bell et al. (2). There are some differences that may be attributable to our preparation of the protein from human cells. One feature is a prominent shallow notch that is approximately at the middle of the structure (Fig. 5C). A second feature is that our structure has more variable conformation (Fig. 6D). This was missed in the structure of Choe et al. (1) because portions of the termini, especially the C termini, were not resolved in their structure. However, Bell et al.
(2) also observed a slightly larger C-terminal portion of their structure. A third feature is that our structure appears to be thicker around the solenoid, perhaps because of the shape of the protein in solution or a higher amount of glycosylation. This claim is consistent with our observation that the R. commonalis lectin, which recognizes N-linked glycosylations, associates with a number of sites in the outer perimeter of hTLR3 ECD. Lastly, the most prominent differ-ence is that we observed quite variable packing in two-dimensional crystals and a higher degree of flexibility in the molecule (Fig. 5). It is likely that hTLR3 ECD may be more flexible in solution in lipid surface two-dimensional crystals than in the three-dimensional crystals. The variable packing that we observed and the predominant monomeric structures of the hTLR3 ECD in single-particle analysis suggest that it is premature to assign a specific dimeric structure to the ectodomain of TLR3. We also noted that the ectodomains of TLR2 and TLR4 are expressed in monomeric forms in baculovirus vectors in Sf9 insect cells (19,20). It remains possible that ligand binding or the presence of the intact molecule could impact TLR3 oligomerization.
Of the eleven human TLRs, TLR2 and TLR4 are the best characterized for the effects of glycosylation on cell surface expression. The ectodomains of TLR2 and TLR4 expressed in insect and human cells (19) were found to migrate at the positions of ϳ110-kDa bands in SDS-PAGE. This is comparable with the situation we observed with the hTLR3 ECD. Furthermore, N-linked glycosylation is the primary form of glycosylation in TLR2, 3, and 4. Also, whereas the ectodomains of TLR2 and TLR4 have four glycosylation sites that are all important for secretion of the proteins in transfected HEK cells, single amino acid changes in all of the potential glycosylation sites we tested in hTLR3 ECD, including the key ones at Asn-247 and Asn-413, retained some expression at the surface of HEK cells (Fig. 4B). It is possible that mutations in a combination of the N-linked glycosylation sites would reveal a defect in TLR3 localization and that multiple glycosylation sites may function in concert, because we observed that a construct containing mutations at both Asn-247 and Asn-413 had markedly lower activity in the reporter assay than the single mutants.
TLR3 has been proposed to function at the plasma membrane in fibroblasts and intracellularly in dendritic cells (17). Because N247A and N431A are expressed on the cell surface and at levels similar to that of the wild-type TLR3, a potential explanation for reduced TLR3 function in the cell-based assay is that these glycosylation sites may directly or indirectly contribute to ligand binding. Currently, a number of functional ligands of TLR3 have been reported, including poly(I:C) (13), mRNAs, RNAs from necrotic cells, and potentially small interfering RNAs (14,16,32). In all of these reports, however, the interactions between TLR3 and the inducing RNAs have been examined through effects on reporter expression, and hence a direct interaction between TLR3 and the inducing RNAs cannot be assumed. The interaction between poly(I:C) and hTLR3 ECD can be observed in vitro using gel electrophoretic mobility change assays, changes in the intrinsic properties of hTLR3 ECD (1, 2). We have observed binding of poly(I:C) to hTLR3 ECD by UV cross-linking. 7 However, the affinity of this interaction is poor, raising the question of whether this interaction is biologically relevant. Therefore, the role of an adaptor molecule to mediate the interaction between TLR3 and double-stranded RNA ligand remains a viable possibility. We note that the structure of the ectodomain of the human TLR3 contains sufficient space within the concave surface of TLR3 ECD to accommodate possible adaptors and/or adaptor and ligand complexes.