Research Article

Crystal Structure of Human Toll-Like Receptor 3 (TLR3) Ectodomain

See allHide authors and affiliations

Science  22 Jul 2005:
Vol. 309, Issue 5734, pp. 581-585
DOI: 10.1126/science.1115253


Toll-like receptors (TLRs) play key roles in activating immune responses during infection. The human TLR3 ectodomain structure at 2.1 angstroms reveals a large horseshoe-shaped solenoid assembled from 23 leucine-rich repeats (LRRs). Asparagines conserved in the 24-residue LRR motif contribute extensive hydrogen-bonding networks for solenoid stabilization. TLR3 is largely masked by carbohydrate, but one face is glycosylation-free, which suggests its potential role in ligand binding and oligomerization. Highly conserved surface residues and a TLR3-specific LRR insertion form a homodimer interface in the crystal, whereas two patches of positively charged residues and a second insertion would provide an appropriate binding site for double-stranded RNA.

Innate immunity is based on an ancient and ubiquitous system of cells and molecules that defend the host against infection. This system can recognize virtually all microbes, using a limited repertoire of germ-line–encoded receptors that recognize broadly conserved components of bacterial and fungal cell walls or genetic material, such as double-stranded viral RNA (1, 2).

The Toll-like receptors (TLRs) are among the most important sensors of the innate immune system (3). The ten known human Toll-like receptors recognize pathogen-associated molecules, such as lipoteichoic acid (recognized by TLR2), lipopolysaccharide (TLR4), flagellin (TLR5), and unmethylated CpG DNA motifs (TLR9). Binding of these ligands to TLRs initiates a series of signaling processes that stimulate and orchestrate the innate and adaptive immune responses (4, 5). Human TLRs are implicated in a number of diseases and, hence, constitute potential therapeutic targets (6, 7).

TLRs are integral membrane proteins located either on the cell surface or in intracellular compartments. Their extracellular or ectodomains (ECDs) are responsible for ligand binding and contain 19 to 25 leucine-rich repeat (LRR) motifs that are also found in a number of other proteins with diverse cellular functions (8).

The intracellular domain, known as the Toll/interleukin-1 receptor homology (TIR) domain, recruits adaptor molecules, such as MyD88, TRIF, and TIRAP, to initiate the signaling process (4, 9).

Human TLR3 is activated by double-stranded RNA (dsRNA) associated with viral infection (10), endogeneous cellular mRNA (11), and sequence-independent small interfering RNAs (12). TLR3 is distinct from other TLRs in that it is not dependent on MyD88 but rather on TRIF for signaling (13). Other key features of TLR3 signaling include a requirement for phosphorylation of tyrosine residues in the TIR domain (14) and the involvement of phosphatidylinositol-3 kinase (15). In turn, TLR3 activates genes for secreted antiviral cytokines, such as interferon (IFN-β), and those that encode intracellular, viral, stress-inducible proteins (16).

Overall structure. The complete ectodomain of human TLR3 without the N-terminal signal sequence (residues 27 to 700) was expressed using a baculovirus system and purified by Ni-nitrilotriacetic acid (NTA) affinity, ion-exchange, and size-exclusion chromatography (17). TLR3 native crystals diffracted to 2.1 Å, and a 3.1 Å multiwavelength anomalous diffraction (MAD) data set from a K2OsO4 derivative was used for the initial phase calculation (18). The three-dimensional structure of human TLR3 (ECD) reveals a large, horseshoe-shaped assembly consisting of 23 LRRs that adopt a right-handed solenoid structure (Fig. 1). Human TLR3 (ECD) contains 23 canonical and two irregular LRRs, which is the largest number yet observed among LRR-containing proteins with known structures. The inner and the outer radii of the TLR3 horseshoe are ∼26 Å and 40 Å, respectively, which is substantially larger than that of ribonuclease inhibitor (inner radius, 18 Å) (fig. S3) (19). The concave (inner) surface is formed from 25 parallel β strands, 23 from LRRs and one each from the N- and C-terminal cap regions, that assemble into a highly curved, continuous β sheet that spans 270° of arc. The convex (outer) surface contains more diverse secondary structure elements, including a variety of different length of loops, four α helices, seven 310 helices, and two β strands. The superposition of 23 LRRs shows that the residues that compose the inner β strands overlap extraordinarily closely with each other, with root mean square deviations (RMSDs) ranging only from 0.3 to 0.8 Å (Fig. 2A). In contrast, the rest of the motif is more diverse, with RMSDs of 1.0 to 2.5 Å when equivalent Cα atoms are used in the superposition. One disulfide bond was observed between Cys95 and Cys122, linking LRR2 and LRR3 (Fig. 1).

Fig. 1.

Overall architecture of TLR3 ECD in a ribbon representation. The N-terminal cap region is colored blue; the 23 canonical LRRs are in green; and the C-terminal region is in pink. N-linked sugars (N-acetylglucosamines) that are observed in the electron density maps are shown in ball-and-stick representation, attached to their respective Asn residues. The disulfide bond linking LRRs 2 and 3 is drawn in orange, adjacent to the glycosylation site. (A) Side view of TLR3 with the convex face pointing outwards, the concave face inwards, and the heavily glycosylated side face pointing toward the viewer. (B) View rotated 45° from (A) that highlights the continuous β sheet that forms the concave surface. The position of the large insertion in LRR12 that extends toward the glycosylation-free face is marked with an asterisk.

Fig. 2.

Structural features of the LRR motifs in TLR3. (A) The 23 canonical LRR motifs of TLR3 are superimposed with the Cα atoms of the eight residues that form the concave surface of the repeating β strands. The positions of the two major insertions in LRR12 and LRR20 can readily be visualized, as well as the loops and helices that connect each LRR motif. (B) The conformation of the LRR motif (LRR7) highlights the tight packing of the conserved hydrophobic residues with their van der Waals' radii outlined in dots. The residues are numbered by their position in the 24-residue consensus motif (fig. S1). (C) A portion of the continuous β sheet that highlights additional main-chain bifurcated hydrogen bonds formed at both ends of the β strands. Hydrogen bond interactions with distances less than 3.5 Å are indicated. (D) The intricate hydrogen bond network formed by the conserved asparagines (position 10) that stabilize the assembly of repeating LRR motifs.

Leucine-rich repeats. The LRRs of TLR3 (ECD) follow the typical consensus motif of a 24-residue repeat, consisting of xL2xxL5xL7xxN10xL12xxL15xxxxF20xxL23x, where L represents obligate hydrophobic residues, which for TLR3 include leucine (most prevalent), isoleucine, valine, methionine, and phenylalanine; F is a conserved phenylalanine; and N is a conserved asparagine (20) (fig. S1). The seven conserved hydrophobic residues of the LRRs (at positions 2, 5, 7, 12, 15, 20, and 23) (fig. S1) point inward into the solenoid and form a tightly packed hydrophobic core (Fig. 2, A and B) that provides lateral stability to the repeating LRR motifs. The parallel β sheet that forms the inner concave surface is fairly typical for LRR proteins, except that, at either end, the carbonyl groups form additional bifurcated hydrogen bonds with main-chain amides (Fig. 2C). The side chains of the conserved asparagines (position 10) at the end of type VI β-turn regions also make extensive bridging interactions, including three hydrogen bonds to the amide and carbonyl groups from the previous LRR motif and one to the carbonyl group of residue 7 from the same LRR repeat, that further stabilize the solenoid structure (Fig. 2D). Seventeen of the 23 human TLR3 LRRs have the canonical 24-residue motif, and only LRR12 and LRR20 have insertions longer than 5 residues (fig. S1). LRR12 is composed of 33 residues with a 10-residue insertion after position 11 that points toward the glycosylation-free side face (marked with an asterisk in Fig. 1B) (21); however, six of these inserted residues are disordered and are not included in the refined structure. LRR20 has 32 residues with an 11-residue insertion after position 13, which points toward the outer convex surface.

N-terminal and C-terminal caps. Most proteins with LRR motifs contain flanking regions at their N- and C-termini that cap the exposed edges of the hydrophobic core of the solenoid structure formed by the canonical LRRs. In TLR3, the N-terminal capping region consists of residues 27 to 51, in which Val34 to Ser36 initiate a modified LRR repeat that extends the continuous β sheet. Residues Ala35, Leu40, Leu42, Val45, and Leu49 point toward the hydrophobic core of LRR1, whereas the remaining residues are mainly hydrophilic and form a lid over one end of the hydrophobic core (fig. S2). The C-terminal region of human TLR3 spans from Asn636 to Ala700, but a large part of this domain is disordered in the crystal and only Asn636 to Thr664 are included in the refined structure. As with the N-terminal cap, a modified LRR is found where Glu639 to Leu641 form a β strand that extends the continuous β sheet (22), and Leu637, Leu640, Met642, and Phe647 contact the other edge of the hydrophobic core at LRR23 (fig. S2). Although the C-terminal region contains the characteristic four cysteine motif (CxCxnCxnC), no disulfide bonds were observed due to disorder in the C-terminal residues.

Glycosylation. Fifteen potential N-linked glycosylation sites are located in the extracellular domain of human TLR3 (23). Electron density for one or two N-acetylglucosamines of each carbohydrate moiety was observed at eight of these locations, corresponding to Asn124, Asn252, Asn265, Asn275, Asn291, Asn398, Asn413, and Asn507 (Fig. 1). When fully glycosylated sugars, which in insect cells are mainly oligomannans such as Man9GlcNAc2, are modeled onto the 15 predicted N-glycosylation sites, most of the TLR3 surface is masked by carbohydrate, including the inner concave and outer convex surfaces and one of the two major faces of the horseshoe-shaped structure (Fig. 3). Importantly, one face is completely devoid of any glycosylation (Fig. 3). When the locations of the predicted N-glycosylation sites of the other nine human TLRs are examined, the concave surface, which is formed from LRR residues 2 to 9 (fig. S1), usually contains glycosylation sites and suggests that glycosylation of the inner concave surface is not unique to TLR3. This result is unexpected because the concave surface was predicted to contain the ligand-binding site of TLRs, including that for dsRNA in TLR3 (20).

Fig. 3.

N-linked glycosylation of TLR3. (A) Oligomannose-type sugars, as found predominantly in insect cells (up to nine mannoses and two N-acetylglucosamines), are modeled onto the 15 predicted N-linked glycosylation sites to give a representation of the potential extent of the TLR3 surface masked by carbohydrate (same view as Fig. 1A). The TLR3 surface is represented in green and the sugars in yellow. An end view of an A-form dsRNA is also shown in pink for comparison. The exact nature of the carbohydrate, whether complex or high mannose, is not the issue here, only that glycosylation covers a large part of the TLR3 surface. (B) View rotated 180° from (A) showing the glycosylation-free face.

Electrostatic surface potential. When the surface electrostatic potential of TLR3 was calculated with GRASP (24), the concave surface was largely negatively charged (Fig. 4) and again inconsistent with its predicted interaction with dsRNA. Considering that dsRNA is acidic, we then searched for positively charged regions on the TLR3 surface. Such a patch was found on one sector of the glycosylation-free face and is formed from Lys331, Lys335, Arg394, Lys416, Lys418, Lys467, Arg489, and Lys493 (Fig. 4C). The convex surface close to the N terminus contains another patch of positively charged residues, including Arg65, Lys89, Lys117, Lys137, Lys139, Lys145, Lys147, and Lys163. Thus, either of these two regions could correspond to potential RNA binding sites, but the former is also adjacent to a TLR3-specific insertion that would further suggest some functional significance.

Fig. 4.

Surface representation of TLR3, with the surface colored according to the electrostatic potential calculated by GRASP. Red surfaces indicate negative potential and blue positive (±15 kT/e). (A) The glycosylated side face with some of the positively charged residues labeled. (B) The concave surface showing its predominantly negative electrostatic potential. (C) The glycosylation-free face that highlights two clusters of positively charged residues (upper left and lower right) that might interact with the ligand dsRNA. The position of the large insertion in LRR12 is close to the major positive patch (upper left) and is marked with an asterisk; it is also consistent with the proposed location of the ligand binding site. Although these patches are not absolutely conserved in all known TLR3 sequences, they do partially overlap with conserved patches when homologous amino acid substitutions are allowed. K, lysine; R, arginine.

Comparison with other LRR proteins. The concave surface of human TLR3 (ECD) corresponds to the β sheet of the solenoid and is composed of 25 β strands, 23 from LRRs and one each from the N- and C-terminal capping regions. The radius of the concave surface defined by the inner β strand is ∼26 Å, and the overall rotation angle of the 25 β strands is ∼270°, which corresponds to 11.3° per repeating LRR unit. Comparison with other LRR proteins, including pRI (25), hGPIbα (26), ypYopM (27), hNogo (28), and mCD14 (29), indicates that the overall architectural characteristics of the repeating LRR motifs of human TLR3 are most similar to the hNogo receptor, which also has typical 24-residue LRR motifs (fig. S3). Human TLR3 exhibits the largest difference in curvature with pRI, because the outer surface of pRI is composed of repeating α helices rather than loops, as for human TLR3 (fig. S3) (30). The overall shape and dimensions of a tail-to-tail CD14 dimer observed in its crystal structure (29) roughly matches the overall dimensions of TLR3 (fig. S3). Although it is not clear that this CD14 dimer has any biological relevance, it provides some clues as to possible heterodimeric interactions with TLR4, which is predicted to have 21 LRRs.

dsRNA binding site and mode of signaling. One of the unexpected findings from the TLR3 (ECD) structure was the location of four potential glycosylation sites in the inner concave surface. Electron density for two N-acetylglucosamines was observed at Asn252 and Asn413. Although the inner diameter of the concave surface is ∼52 Å, which is large enough to hold a double-stranded A-form RNA with a diameter of ∼26 Å, fully glycosylated sugars (31) would severely hamper binding of dsRNA (Fig. 3). Considering that human TLR3 (ECD) appears to reside in the endosome (32), these glycosylation sites would be expected to be fully substituted, although the exact nature of glycoforms that are present on TLR3 in mammalian cells has not yet been determined. Notwithstanding, whether high mannose or complex sugars, these potential glycosylation sites of TLR3 (ECD) are not evenly distributed on the TLR3 surface and map to the concave (four sugars) and convex (four) surfaces, as well as to one of the two major side faces (seven). The other major side face is completely glycosylation-free, which leaves it accessible for interaction with ligands, including any potential accessory molecules, or for formation of self- or other oligomeric assemblies.

The combined presence of glycosylation sites and the predominant negative charge of concave surface make it an unlikely site for binding double-stranded RNA. This conclusion is contrary to the assumed binding mechanism for LRR-containing proteins that use the concave surface for interacting with their binding partners (20). However, analysis of several LRR proteins indicates that other regions can also act as binding sites. For example, ribonuclease inhibitor binds ribonuclease A on its concave surface, as well as on the side face proximal to the C terminus (33). Glycoprotein Ibα also binds its ligand, the von Willebrand factor A1 domain, on the concave surface, but with major contributions from a β finger from the N terminus and a β switch region from the C terminus (34). Other examples include RanGAP, which binds Ran on the side of the LRR domain (35), and the lipopolysaccharide-binding protein CD14, where the binding site is predicted to be located in a surface pocket close to the N terminus (29).

Examination of the symmetry-related interactions of TLR3 molecules in the crystal shows a dimer that could be of biological relevance (Fig. 5A). The homodimer interface is close to the C terminus on the glycosylation-free face (Fig. 5C). Although its buried surface area is relatively small (640 Å2), this region contains the densest cluster of conserved, surface-exposed residues, when five different TLR3 sequences were aligned (Fig. 5B and fig. S4). Interactions at this dimer interface were mostly hydrophilic, including ionic interactions (Glu442–Lys467 and Lys547–Asp575), hydrogen bonds between side chains, and water-mediated hydrogen bonds (fig. S5). A limited number of hydrophobic interactions were observed. Three residues involved in this hydrophobic interaction (Lys547, Ala549, and Pro551) come from the TLR3-specific insertion in LRR20 and significantly increase the contact area in the dimer interface. Although the human TLR3 ECD exists as a monomer in solution according to our size-exclusion chromatographic results, TLR3 is an integral membrane protein and could readily form a stable or transient dimer when embedded in the membrane and subject only to lateral diffusion. In fact, certain TLRs are known to form functional oligomers in vivo. For example, TLR2 makes functional heterodimers with TLR1 and TLR6 (36), and TLR4 assembles into a homodimer for signaling (37). However, the physiological significance of the dimer observed in the crystal is still speculative and awaits further confirmation by mutagenesis and other biophysical methods. Clearly, if not involved in dimerization, this surface would also be a strong candidate for a ligand binding site.

Fig. 5.

Possible mode of TLR3 dimerization and a model for dsRNA binding and mode of signaling. (A) A homodimer observed in the crystal is shown with one monomer in green and the other in pink. One monomer points away from and the other toward the viewer. No other interactions between monomers in the crystal have sufficient buried surface area to warrant consideration as physiologically relevant oligomers. (B) Surface representation of TLR3 (green) highlights sequence conservation on the TLR3 surface (red). The largest conserved patch is on the glycosylation-free face (same view as Fig. 3B). The five TLR3 sequences used in the alignment are Homo sapiens (BC059372), Pan troglodytes (XP_526756), Bos taurus (NP_001008664), Rattus norvegicus (NM_198791), and Takifugu rubripes (AC156436). (C) The dimer interface of TLR3 also maps to the glycosylation-free face. The residues involved in the dimerization are colored in red. The large patch of conserved surface residues of TLR3 (panel B) map to this dimer interface region. (D) Potential functional sites on the TLR3 surface. The positions of positive patches and one of the two large insertions in the LRR motifs that are implicated for ligand binding on the glycosylation-free surface are indicated. Similarly, the conserved buried surface and the other large TLR3 LRR insertion maps to this dimerization interface region. Oligomannose-type sugars are drawn in yellow. (E) Binding of a dsRNA, either itself or with an accessory protein, could either induce dimerization or changes in a pre-existing dimeric assembly and trigger signal transduction in mechanism similar to those seen for other cell-surface receptors, such as cytokine receptors.

We propose then that the putative ligand binding site for dsRNA is also located on the glycosylation-free face of TLR3 because it contains the densest cluster of basic residues on the entire TLR3 surface (Fig. 5D). Examination of other dsRNA binding proteins indicates that a strong positively charged region, though sometimes present, is not a necessary condition for RNA binding (3840). Importantly, the 10-residue insertion in LRR12 is also on this same face and only ∼15 Å from the positive patch. Although 7 out of 10 residues in this insertion are disordered and not seen in our current structure, they could potentially interact with the major groove of dsRNA through their six hydrophilic residues in the loop. Furthermore, as 36 residues at the C terminus are also disordered, the possibility of this region forming an extended binding site, as in glycoprotein Ibα, cannot be excluded.

If direct binding of dsRNA to TLR3 actually triggers signaling, binding to only one face of the receptor would appear to be a reasonable solution, because binding to the concave surface, like a ring on a finger, may induce uncontrolled oligomerization of TLR3 depending on the length of dsRNA. By binding to dsRNA using only one face of TLR3, two TLR3 monomers could potentially sandwich dsRNA in a ternary complex of two TLR3s and one dsRNA. This possible mode of dimerization would position the positive patches and the 10-residue insertion in LRR12 on either side of the dsRNA (fig. S6). Furthermore, binding of dsRNA in this fashion may either strengthen the dimerization or induce conformational changes in a pre-existing dimer that could trigger changes in the disposition of the intracellular C-terminal TIR domains to bring them closer for signal transduction (Fig. 5E). Thus, a ligand-induced change in a pre-existing dimer assembly, as observed in the erythropoietin receptor (41), or a ligand-induced dimerization, as for other cytokine receptors (42), would assist formation of the active complex for signaling.

However, although abundant experimental data have been accumulated that strongly support the proposition that dsRNA or RNA analogs induce immune activation by signaling through TLR3 (1012, 16, 43), no direct experimental evidence has yet been provided for the physical association of dsRNA with TLR3. Nevertheless, in our gel-shift assay, poly(I:C), a ligand widely used to activate TLR3 in vivo (10), binds to TLR3 ECD and changes its migration pattern (fig. S7) (44). Considering that some TLRs recognize their ligands in cooperation with co-receptor proteins (4547), it is still possible that TLR3 could recognize dsRNA in conjunction with a not-yet-identified accessory protein(s). However, the TLR3 structure would still suggest that any such ligand interaction would be made on the glycosylation-free face and induce dimer assembly or reorganization. The TLR3 structure now provides a framework to design experiments to characterize TLR's ligand specificity and define its mode of signaling.

Supporting Online Material

Figs. S1 to S7

Table S1

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article