|Home | About | Journals | Submit | Contact Us | Français|
TNF-like 1A (TL1A) is a newly described member of the TNF superfamily that is directly implicated in the pathogenesis of autoimmune diseases, including inflammatory bowel disease, atherosclerosis and rheumatoid arthritis. We report the crystal structure of the human TL1A extracellular domain at a resolution of 2.5 Å, which reveals a jelly-roll fold typical of the TNF superfamily. This structural information, in combination with complementary mutagenesis and biochemical characterization, provides insights into the binding interface and the specificity of the interactions between TL1A and the DcR3 and DR3 receptors. These studies suggest that the mode of interaction between TL1A and DcR3 differs from other characterized TNF ligand/receptor complexes. In addition, we have generated functional TL1A mutants with altered disulfide bonding capability that exhibit enhanced solution properties, which will facilitate the production of materials for future cell-based and whole animal studies. In summary, these studies provide insights into the structure and function of TL1A and provide the basis for the rational manipulation of its interactions with cognate receptors.
TNF-like 1A (TL1A), also known as vascular endothelial growth inhibitor (VEGI), is a newly described member of the TNF superfamily involved in a wide range of human pathologies, including autoimmunity and tumor progression (1). The functional receptor of TL1A is death receptor 3 (DR3), a TNF receptor family member that contains a cytoplasmic death domain. TL1A is expressed by endothelial and dendritic cells, and expression in human umbilical vein endothelial cells (HUVEC) is significantly enhanced by treatments with TNF-α or IL-1, but reduced by IFN-γ (1). DR3 is predominantly expressed by activated T-cells and to a lesser extent by endothelial cells (2, 3). Engagement of DR3 by TL1A expressed on dendritic cells triggers a costimulatory signal in T cells that induces the secretion of IFN-γ via NF-κB-associated pathways (1, 4). The treatment of DR3-expressing tumor cell lines with recombinant soluble TL1A results in caspase-dependent apoptosis when proliferative signals are blocked by the protein synthesis inhibitor cycloheximide (1). Similarly, overexpression of secreted TL1A protein from vascular endothelial cells results in autocrine apoptosis, which may function to inhibit tumor neovascularization and progression (2, 3). Notably, mice deficient in either TL1A or DR3 exhibit normal immune system development; however, both strains show significant reductions in the development of autoimmune diseases, highlighting an important role for the TL1A-DR3 interaction in autoimmune pathology (5, 6).
The TL1A-DR3 interaction can be inhibited by the DcR3 decoy receptor (1), a soluble TNFR family member that is expressed by a wide range of normal human tissues, particularly adult spleen, colon and lung (7, 8). In addition to TL1A, DcR3 also neutralizes the TNF family members LIGHT (9) and FasL (8). Although the gene coding for DcR3 is absent from the murine genome, transgenic mice ubiquitously expressing human DcR3 exhibit systematically attenuated Th1 responses (10). In vitro studies demonstrated that the treatment of dendritic cells with DcR3 polarizes T cell development towards a Th2 response (11, 12). In addition to the modulation of immune function, DcR3 can induce angiogenesis and neovascularization by inhibiting the autocrine effect of TL1A (3). Based on these results and the observation that DcR3 overexpression is associated with the development of a variety of malignancies (7, 8, 13), it has been suggested that tumor cells might secret DcR3 as an immune evasion mechanism that inhibits T cell function through neutralization of TNF ligands and may further promote neovascularization (3, 8). Additionally, transgenic NOD mice that express DcR3 specifically in pancreatic β-islets showed localized suppression of Th1 cells and protection from autoimmune diabetes (14). Interpretation of these results is complicated by the multiple interactions involving DcR3, which results in a complex inter-relationship between multiple signaling pathways. Detailed knowledge of these interactions, and their contributions to signaling, will aid in the development of therapeutic strategies for both cancer and autoimmune diseases.
Based on sequence homology, TL1A belongs to the conventional TNF ligand family, which currently includes eight other members: FasL, LIGHT, TNFα, LTα, LTβ, TRAIL, RANKL and CD40L (Figure 1) (15). Members of the TNF superfamily are typically type-II transmembrane proteins that form noncovalent homotrimers composed of β-sandwich ‘jelly-roll’ protomers. DR3 and DcR3 belong to the TNFR superfamily that is composed of type-I transmembrane proteins or secreted proteins characterized by pseudorepeats of between one to six extracellular cysteine-rich domains (CRDs). The three reported structures of TNF ligand/receptor complexes (LTα/TNFR1; TRAIL/DR5; OX40L/OX40) exhibit a similar overall organization, in which the trimeric ligands recruit three receptor molecules to form a three-fold symmetric assembly with a 3:3 receptor-to-ligand stoichiometry (16–18). In these complexes, the receptor binding surfaces are located at the interfaces formed between two ligand protomers and there are no interactions between receptor molecules.
Upon ligand binding, depending on the composition of the intracellular domain and cellular contexts, TNFRs recruit either TNF receptor associated factors (TRAF) or death domain adapter proteins. Generally, the recruitment of TRAFs induces the activation of NF-κB mediated pathways, which initiate transcriptional programs for proinflammatory cytokine production and cellular proliferation; in contrast, the recruitment of death domain adapter proteins activates caspases and leads to apoptosis (19). These pathways represent major targets for therapeutic intervention. For example, Enbrel, a soluble TNF receptor (TNFR)-Ig fusion protein, and Remicade, a monoclonal antibody against TNF-α, both sequester TNF-α and block activation of TNFR-associated inflammatory responses. These reagents represent leading treatments for inflammatory diseases, including rheumatoid arthritis and inflammatory bowel disease.
TL1A-associated pathways also represent targets of potential therapeutic importance, as they are implicated in the pathogenesis of several inflammatory diseases, including inflammatory bowel disease (4), atherosclerosis (20) and rheumatoid arthritis (21). In particular, the involvement of TL1A in Crohn’s disease has been suggested on the basis of mouse models (4) and the correlation of disease with single nucleotide polymorphisms in the TL1A gene (22). As a consequence, TL1A-associated signaling pathways are promising therapeutic targets for Crohn’s disease (4, 23). In this light, it will be of considerable clinical importance to develop TL1A-directed reagents, such as soluble DR3 receptor, soluble DcR3 or blocking antibodies directed against TL1A (4, 23). Here, we report the crystallographic and biochemical characterization of the extracellular domains of TL1A that suggests a mode of receptor recognition distinct from other TNF family members. In addition, we describe a series of fully functional cysteine mutants with enhanced solution behavior. These studies provide insights into the function of this important cytokine and facilitate the manipulation of its interactions with DR3 and DcR3.
TL1A cDNA was obtained from a human kidney cDNA library (BioChain, Hayward, CA) by PCR. The extracellular domain of human TL1A (L72-L251) was cloned into pET28a vector and expressed with an N-terminal His-tag in E. coli strain BL21-AI (Invitrogen, Carlsbad, CA). In this strain, expression of a chromosomal copy of the T7 RNA polymerase gene is under the tight control of the arabinose-inducible araBAD promoter. The cells were grown in LB medium supplemented with kanamycin (50 μg/ml) at 37°C to an OD600 of 0.5 and protein expression was induced by addition of 0.2% arabinose and 1mM IPTG. Cells were harvested by centrifugation after an additional 20 hours of growth (250 RPM, 20°C), resuspended in lysis buffer (50mM NaH2PO4 pH8, 300mM NaCl, 20 mM imidazole, 5μg/ml DNaseI) and lysed by French press. The lysate was clarified by centrifugation at 48,384 g for 30 minutes and the supernantant applied to Qiagen Ni-NTA agarose resin (Qiagen, Valencia, CA). After washing with 50ml of wash buffer (50mM NaH2PO4 pH 8.0, 300mM NaCl, 50 mM imidazole), fractions that were eluted with elution buffer (50mM NaH2PO4 pH 8.0, 300mM NaCl, 250mM imidazole) were pooled and dialyzed against 50mM MES (pH 6.5), supplemented with 20mM NaCl and 4mM βME. The His-tag was cleaved with thrombin and the TL1A protein was purified to homogeneity by cation exchange chromatography (FPLC) using a Mono S HR 10/10 column (Pharmacia). For crystallization, the purified protein was dialyzed into 10mM Tris (pH 8.0), supplemented with 100 mM NaCl, 1mM βME and 1mM cysteamine.
TL1A point mutations were generated with the QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA), and the mutant proteins were purified as described for the wild type.
Diffraction quality crystals of TL1A were obtained by sitting drop vapor diffusion at 17°C by mixing 1 μl of protein (10mg/ml) with 1 μl reservoir solution (0.1M Tris-HCl pH 8.5, 20% methanol and 0.01M CaCl2) and allowing for equilibration over 1 ml of reservoir. C95S and C95S/C135S mutant proteins were crystallized under the same conditions, but the optimal protein to reservoir solution ratio was 1:2 and 2:1 for C95S and C95S/135S mutants, respectively. Cubic shaped crystals typically appeared after one week. Prior to X-ray data collection, crystals of wild type TL1A were flash cooled in liquid nitrogen after cryoprotection with mother liquor supplemented with 30% glycerol. Mutant TL1A crystals were cryoprotected in 20% PEG 400, 50% iso-proponal, 0.1M Tris pH 8.5 and 0.01M CaCl2 prior to flash cooling. All TL1A crystals exhibited diffraction consistent with the cubic space group P4132 (unit cell: a=b=c=120.4), with 1 monomer in the asymmetric unit. Data were collected at the X29 beam line of the National Synchrotron Light Source (Brookhaven National Laboratory, Upton, NY), using a ADSC Quantum-315 CCD detector and 1.0809 Å radiation. The wild type, C95S mutant and C95S/C135S mutant crystals diffracted to resolutions of 2.5, 2.3 and 2.05 Å, respectively.
Diffraction data were integrated and scaled with HKL2000 software package (24). An initial structure of wild type TL1A was determined by molecular replacement with the program PHASER 1.1 (25), using TNF-α monomer (PDB ID: 2TNF) as the search model. The remainder of the model was built with ARP/WARP (26) and refined with REFMAC5 (27). The mutant TL1A structures were solved by difference Fourier methods. Data collection and refinement statistics are presented in Table 1. Atomic coordinates and structure factors of the wild type protein and two mutant proteins, C95S and C95S/C135S, have been deposited to the PDB and are available under the accession code 2QE3, 2RJK and 2RJL, respectively. Subsequent to the release of our wild-type TL1A structure, another human wild-type TL1A structure was reported (PDB 2RE9 and 2O0O) (28). These structures are similar with RMSDs of about ~0.9 Å for 139 Cα atoms and 0.33 Å for 125 Cα atoms excluding disordered regions (SI Figure 1).
The oligomeric state of the wild type TL1A ectodomain was examined by sedimentation velocity with a Beckman XL-I analytical ultracentrifuge. Proteins at three concentrations (10.7μM, 26.7μM and 47.3μM) were prepared in 10mM Tris buffer (pH 8.0) containing 100mM NaCl and 1mM TCEP. The sedimentation velocity experiments were performed at 45,000 rpm with Ti-60 rotor and double-sector cells at 20°C. The sedimentation boundaries were monitored by the absorption at 280 nm and analyzed using DCDT+ software (29). The partial specific volume, buffer densities and viscosities were calculated using program SEDNTERP (30) and were used to correct the resulting sedimentation values to standard conditions, S20,w.
TL1A ectodomain and DcR3 cysteine rich domain (CRD) were also examined by size exclusion chromatography on a Superdex 200 column. Purified TL1A and DcR3-CRD proteins (see supplementary information for expression and purification of DcR3-CRD) were mixed in a 1:1 ratio at a final total concentration of 1 mg/ml and incubated at room temperature for 2 hours. TL1A alone (5mg/ml) and the incubated TL1A/DcR3 mix were applied to a Superdex-200 column equilibrated with 10 mM Tris (pH 8.0) and 100 mM NaCl and eluted using the same buffer at a flow rate of 0.5ml/min. The column was calibrated with cytochrome C (12.4 kDa), carbonic anhydrase (29 kDa), albumin (66 kDa) and alcohol dehydrogenase (150 kDa) using the protocol described above.
Samples for non-reducing SDS-PAGE were prepared by mixing freshly purified protein and Laemmli sample buffer (Bio-Rad, Hercules, CA) with no addition of reducing agents. After boiling for 2 minutes, the samples were loaded onto precast 4–20% gradient gels (Bio-Rad, Hercules, CA) and electrophoresis was performed at 200 V for 38 minutes, using chamber buffer composed of 25 mM Tris (pH 8.0), 192 mM glycine, 1% SDS. Gels were stained with Coomassie blue. To prepare reducing SDS-PAGE, β-mercaptoethanol was included in the sample buffer at a final concentration of 5%.
SPR binding assays were performed with a BIAcore 3000 optical biosensor at 25 °C. Recombinant human DcR3-Ig fusion protein (R&D Systems, Minneapolis, MN) was immobilized on CM5 sensor chips by free amine coupling. TL1A mutant and wild type proteins were injected over the sensor chip at a concentration of 10 nM. The maximum association response of each mutant is divided by that of the wild type TL1A and represented as percentage of wild type binding response. All mutants were generated in the C95S/C135S background, as this double mutant exhibited superior solution behavior compared to the wild type protein while maintaining the same DcR3 binding affinity (see below and SI Figure 3).
The fluorescence-monitored thermal denaturation of wild type and mutant TL1A was performed using the iQ5 real-time PCR detection system (Biorad). Briefly, 20 μl of each protein at 10 μM concentration was mixed with 0.5 μl 200× Sypro orange solution and pipetted into separate wells of a 96-well PCR plate. After centrifugation to remove air bubbles, the plate is loaded into the PCR machine, and the temperature ramped from 20 °C to 99 °C, in 1 °C increments with a dwell time of 6 seconds. The negative first derivative of the fluorescence change (−dRFU/dT) for each protein is plotted against temperature and the melting temperature is defined as the minimum in the −dRFU/dT curve.
The crystal structure of the human TL1A extracellular domain was determined by molecular replacement and refined to a resolution of 2.5 Å (Table 1). TL1A adopts the jelly-roll fold typical of the TNF family, with inner and outer β sheets composed of the A′ AHCF and B′ BGDE strands, respectively. Structural alignments with DALI (31) demonstrate that the TL1A monomer is most similar to the TNF family members CD40L (PDB code, 1ALY) and TRAIL (PDB code, 1D4V) with RMSDs of 1.7 Å (136 Cαs) and 1.4 Å (135 Cαs), respectively. The β-sheet cores of these molecules superimpose well (Figure 2, SI Figure 2), with the greatest structural differences residing in the AA′, CD and EF loop segments. The AA′ and CD loops of TL1A are significantly longer than those present in most TNF ligands, whereas its EF loop is shorter (Figure 1, SI Figure 2). Electron density for most of the CD loop (residues M91-N107) and the tip region of the AA′ (residues T40-Q46) loop is weak or absent, and as a result, these residues are not included in the final model of wild type TL1A.
The asymmetric unit of the TL1A crystals contains a single protomer, which is orientated relative to the crystallographic three-fold rotation axis to produce a tightly packed trimeric assembly that is typical of the TNF superfamily (Figure 2A). The solvent accessible surface area of each protomer buried in the trimeric assembly is 1977 Å 2, comparable to that observed in other stable trimeric TNF ligands (TNFα, 2412 Å2; TRAIL, 2261 Å2; CD40L, 2091 Å2). Similar to other conventional TNF ligands, the subunit interface of TL1A is formed by interactions between the edges of the β-sandwich in one protomer (E and F strands) and the inner sheet of the neighboring protomer (A, H, C and F strands). The central region of this interface is composed predominantly of hydrophobic residues (Figure 3), with F81, Y146, F182 and L184 from each protomer positioned to contribute to the hydrophobic core of the trimer. The corresponding residues at these four positions in other conventional TNF ligands are generally conserved and form the most prevalent interdomain hydrophobic contacts (32). In addition, M150 from one protomer (F strand) and L58 and F142 from the adjacent protomer (AA′ loop and the F strand) contribute to the interface. The bottom of the TL1A subunit interface is lined with polar residues and is stabilized by 6 salt bridges and 3 hydrogen bonds. As a consequence of the short EF loop, the top of the TL1A trimer has relatively few inter-subunit interactions compared to the significant number of predominantly polar interactions present in this region in TNFα (32). The structure of Jin et al (28) showed 3 backbone hydrogen bonds at the top of TL1A subunit interface, which are absent in our structure due to disorder in the CD loop.
Size exclusion chromatography of TL1A yielded an apparent molecular weight of approximately 66 kDa (Figure 4A). With the exception of the E53A-E55A mutant, all mutants examined, including the R99A-R103A-D108A triple mutant and the Y121F single mutant, exhibited size exclusion chromatographic behavior similar to the native trimer, strongly indicating that these constructs exist as stable trimers in solution with interprotomer interactions similar to the wild type (SI Figure 5C). The observation of trimeric species is strong evidence that these mutant proteins adopt correct tertiary and quaternary structures, and are thus capable of interacting with DcR3 in a productive manner. In contrast, E53A-E55A exhibits a monomer-oligomer equilibrium, suggesting these residues may be important for maintaining the oligomeric state (SI Figure 5D)
Velocity sedimentation experiments with the wild type TL1A yielded an S°20,w of 4.227 ± 0.013 S (Figure 4B), which corresponds to an apparent molecular weight (S/D) of 63.05 ± 1.63 kDa. These values are consistent with the calculated molecular weight of the purified TL1A trimer (62.67 kDa). Increasing the TL1A concentration (10.7μM to 47.3μM) results in a slight decrease of S20,w (Figure 4B), indicating that there is no significant self-association of the TL1A trimer at the concentrations examined. Sedimentation experiments with the TL1A cysteine mutants (i.e., C95S, C135S, C95S/C135S) at a concentration of 26.7μM yielded S20,w values indistinguishable from the wild type protein (data not shown), confirming that the cysteine connectivity does not contribute to the formation or maintenance of the oligomeric state.
Long term storage of recombinant TL1A requires reducing agents (for instance 1mM DTT) to prevent aggregation. Non-reducing SDS-PAGE analysis of the wild type TL1A protein reveals three distinct species: a band corresponding to a dimer (~42 kDa) and two distinct lower molecular weight bands around the predicted monomer position (~21 kDa) (Figure 5A, lane A). All three bands coalesce to a single monomeric species on reducing SDS-PAGE (data not shown), suggesting that the heterogeneity observed in non-reducing conditions are the result of interchain and intrachain disulfide bonds. Mutation of either or both of the cysteines in TL1A (i.e., C95S, C135S and C95S/C135S) results in proteins resistant to aggregation in the absence of reducing agents. Non-reducing SDS-PAGE gel shows that the C135S protein is completely homogeneous (Figure 5A, lane C). However, the C95S mutant still possesses a dimer population (Figure 5A, lane B). These observations suggest that the lower-monomer band associated with the wild type protein might result from an intra-chain disulfide bond between the CD and EF loops (Figure 5C) as observed in the structure of Jin et al. (28), while the dimer band might be the consequence of a disulfide bond formed by two C135 residues located in the EF loops of adjacent protomers in the homotrimer assembly (Figure 5B). The simultaneous presence of monomer and covalently linked dimer populations in native C95S mutant protein (demonstrated in Figure 5A, lane B) were confirmed by thermo Electron LTQ mass spectroscopy (data not shown). Non-reducing SDS-PAGE analysis of the C95S mutant TL1A crystals shows the same pattern observed in solution (data not shown), demonstrating that a particular species did not preferentially crystallize and that the crystals contains disulfide-linked protomers as well as reduced protomers.
Non-reducing SDS-PAGE gels of the wild type TL1A protein indicate that the dimer population is about 20% of the total protein (FIGRE 5). In the structure of wild type TL1A, C135 from the three protomers of the same homotrimer are in close proximity, with a distance of about 3.9 Å between adjacent sulfur atoms (7.6 Å between Cαs). Due to this mixture of disulfide-linked and reduced protomers, as well as inherent dynamics, the electron density of the EF loop containing C135 is weak and is compatible with the presence of a 20% population of linked protomers, which would be below the level of detection.
The C95S and C95S/C135S mutant proteins crystallized under similar conditions as the wild type protein and are isomorphous with the wild type crystals (Table 1). However, the mutant protein crystals are significantly larger in size and exhibited higher quality diffraction (Table 1). All three structures are similar with RMSD between coordinates of their Cα atoms about 0.6 Å (SI Figure 4). The biggest differences reside in the structurally dynamic EF loops. In all structures, the CD loops and the tip segment of the AA′ loops were disordered. Sedimentation velocity measurements demonstrate that all cysteine mutants (C95S, C135S and C95S/C135S) are tight non-dissociating trimers in solution (data not shown). The TL1A structure reported by Jin et at. shows the C95S-C135S disulfide bond with the complete CD loop being present in the model (28). Therefore, the disulfide bond heterogeneity in our crystals may lead to disorder in the CD loop, and to flexibility of the EF loops. However, sedimentation velocity experiments show that this disulfide bond heterogeneity does not alter the quaternary structure of TL1A.
The interactions of wild type human TL1A with the human neutralizing receptor hDcR3 were evaluated using surface plasmon resonance (SPR) (SI Figure 3). The analysis of interaction between hDcR3 and wild type TL1A yielded an equilibrium dissociation constant (Kd) of 1.1 nM, which is consistent with the previously reported value of 1.8 nM (1). In gel filtration studies, a 1:1 mixture of DcR3 cysteine rich domain (CRD) and TL1A C95S/C135S double mutant results in a single species with significantly higher apparent molecular weight compared to TL1A alone (Figure 4A), reflecting tight binding between the two proteins and putative 3:3 binding stoichiometry. SPR analyses for the three TL1A cysteine mutants (C95S, C135S single mutants and C95S/C135S double mutant) resulted in binding affinities comparable to the wild type TL1A, with Kds ranging from 0.6 nM to 1.8 nM (SI Figure 3). These findings indicate that the absence of the C95-C135 disulfide link and the possible presence of interprotomer disulfide bonds do not effect the interaction between human TL1A and DcR3.
To map the DcR3-binding sites on TL1A, a series of residues on the surface of the TL1A C95S/C135S mutant were altered on the basis of sequence conservation and structural position. The stability of these mutants was examined by thermal melting experiment. All the mutants, except G57D and E53A-E55A, exhibited melting temperatures that were equivalent to or higher than that exhibited by the C95S-C135S mutant (Tm=77°). The G57D single mutant and E53A-E55A double mutant, both of which are located at C-terminal base of AA′ loop near the interprotomer interface, exhibited significantly lower melting temperatures (Tm=47–49°). (SI Figure 5E, F)
The ability of TL1A mutants to bind DcR3 was evaluated by SPR at a ligand concentration of 10 nM. As shown in Figure 6, the E53A-E55A double mutant and the G57D single mutant in the C-terminal base of the AA′ loop, the Y121F mutant in the DE loop and the E174A single and K173A-K176A double mutants in GH loop all showed significantly impaired interactions with the DcR3 decoy receptor. Several other mutants, including the L56A mutant in the AA′ loop, the S120A-E123A double mutant in the DE loop, and the K173A and D175A single mutants in the GH loop showed very modest decreases in DcR3 binding.
SPR kinetics analysis of the interaction between DcR3 and the R99A-R103A-D108A triple mutant in the CD loop yielded a Kd of 0.52 nM, essentially indistinguishable from that of the C95S-C135S TL1A mutant (Kd of 0.66 nM) (SI Figure 5). By contrast, the Kd of the Y121F single mutant is 6.75 nM, about 10 fold increase (SI Figure 5), and is consistent with the results depicted in Figure 6. These quantitative results serve to validate the use of maximal Biacore responses to provide qualitative data that is mechanistically informative.
The thermal dentauration studies suggest that the mutations at the C-terminal base of AA′ loop (E53-E55 and G57) may indirectly influence DcR3 binding through effects in protein stability and/or oligomeric state. By contrast, the Y121F, E174A and K173A-K176 mutants in DE loop and GH loop have equivalent or even higher melting temperatures compared with wild type protein, suggesting that these residues are directly involved in DcR3 binding. Overall, these results indicate that the DE loop, GH loop of TL1A are involved in the binding of DcR3 receptor, and the C-termial base of AA′ loop is important for structural stability and oligomeric state of the protein.
The current work demonstrates that the TL1A ectodomain adopts a typical TNF family “jelly-roll” fold and exists as a stable trimer in both solution and the crystalline state. As depicted in Figures 2 and and3,3, the TL1A interprotomer interface is formed by two layers of hydrophobic and hydrophilic interactions located at the middle and the bottom of the molecule, respectively. The top of the TL1A trimer is mainly composed of hydrophilic residues, but there are fewer interprotomer contacts in this region. This organization contrasts other conventional TNF family members which utilize three layers of alternating hydrophilic, hydrophobic and hydrophilic interactions (represented by TNFα, CD40L and TRAIL in Figure 3). More generally, the physicochemical properties of the interprotomer interfaces are distinct in each TNF family member, providing a mechanism that prevents inappropriate heterotrimer formation and enforces homotrimer formation.
The current structure allows for the generation of a structure-based sequence alignment, which highlights a number of residues that are invariant among the conventional TNF ligands (shown as red positions in Figure 1). W52, L72, Y80, L153, and F179 (TL1A numbering) are invariant, likely due to their contribution to the hydrophobic core of each protomer (SI Figure 6). The invariant G78 residue is located at the base of BC loop and adopts a dihedral angle that is unfavorable for other amino acids, but is required to support the connection between the BC loop and C strand (SI Figure 6). Y83, G148 and G180 are also invariant due to their contributions to the interprotomer interface. In silico mutation of G148 to alanine results in a steric clash with the side chain of F81 from the same protomer, which is part of the hydrophobic core at the trimer interface. The side chain of Y83 extends towards the neighboring β sheet, occupying the available space above G180, and its phenolic hydroxyl group forms hydrogen bonds with the side chain of the nearly invariant H31 from the same protomer and the backbone amide of M150 from the adjacent protomer (SI Figure 7). These inter/intra protomer interactions involving Y83 are conserved among all the conventional TNF ligands with known structures (TNFα, LTα, TRAIL and CD40L).
The two cysteines present in the CD and EF loops of TL1A are conserved among most conventional TNF ligands (Figure 1). Because TNFα and CD40L both possess intramolecular disulfide bonds connecting these two loops (32), LIGHT, FasL and TL1A were all predicted to have similar disulfide connectivity (33). As described above, recombinant wild type TL1A protein produced in E. coli appears to be heterogeneous in terms of disulfide bond content, with both an intra-chain disulfide bond between the CD loop and EF loop and an inter-chain disulfide bond between the EF loops from adjacent protomers in the same homotrimer assembly. Sedimentation velocity measurements of the cysteine mutants reveal that they all have the same quaternary structure as the wild type protein. Based on the nearly identical binding of wild type and C95S/C135S mutant TL1A to DcR3 (Figure 4A, SI Figure 3), these cysteines and the associated disulfide linkages are not critical for structural integrity or DcR3 binding. Interestingly, a small fraction of naturally secreted human TL1A also has inter-chain disulfide bonds (34), reflecting the in vivo occurrence of TL1A disulfide bond heterogeneity. However, since the intra-chain disulfide bond between the CD loop and EF loop is present in the TL1A structure reported by Jin et al. (28), the disulfide bond heterogeneity observed in our structure may in part be the consequence of our recombinant expression protocol in E. coli. Importantly, since all three TL1A cysteine mutants described above are more resistant to aggregation than the wild type protein expressed in E. coli, they provide superior and more economical alternatives for future mechanistic and clinical studies.
The existing structures of TNF:TNFR family complexes suggest a common mechanism for receptor recognition and provide considerable insights into the receptor binding surfaces on the ligands. In particular, the receptors interact with solvent accessible segments that form the groove between two adjacent ligand protomers. In TRAIL and LTα, as depicted in Figures 7A and 7B, the receptor binding surfaces can be divided into two regions: the lower level composed of the DE and AA′ loops, and the upper level mainly formed by the CD loop (Figure 7) (17, 35). This conservation in the receptor binding interface allows for the prediction of the TL1A surfaces involved in the recognition of the functional DR3 receptor and the neutralizing DcR3 receptor.
In the lower region of the receptor binding groove of TL1A, Y121 located at the tip of the DE loop (Figures 7C and 7D) is one of the most conserved residues among conventional TNF ligands (Figure 1). Structural alignments reveal that these conserved tyrosines are located very close in space with similar conformations (SI Figure 8). In the structures of the LTα/TNFR1 and TRAIL/DR5 complexes, the phenolic hydroxyl groups from these conserved tyrosines participate in hydrogen bonding interactions with the backbone atoms of their cognate receptors (16, 17), and mutagenesis studies demonstrate that alteration of this tyrosine residue in TNFα, LTα, FasL, LIGHT and TRAIL results in dramatic decreases in receptor binding affinity (36–40). Consistent with a conserved role of this tyrosine in TNF family function, Y121 in TL1A is situated at the bottom of the putative binding groove (Figure 7) with considerable solvent accessible surface area (141 Å2), and the Y121F mutation reduces TL1A binding to DcR3 by 90% (Figure 6). Interestingly, two other conventional TNF ligands, RANKL and CD40L, possess Ile and Arg instead of Tyr in this position, respectively (Figure 1), suggesting some variability in the detailed mechanisms of receptor recognition.
In contrast to the significant impact of Y121F mutation in TL1A, the substitutions of other conserved residues in the DE loop (E123A single mutant and S120A-E123A double mutant) do not significantly affect TL1A binding to DcR3 (Figure 6). In the structure of the TRAIL/DR5 complex, S215 and D218 of TRAIL (counterparts of TL1A S120 and E123, respectively) both form polar interactions with side chains of H106 and D109 on the DR5 receptor (17). However, the analogous region of DcR3 (residues F80 to Y83) is composed mainly of hydrophobic residues (SI Figure 9), precluding polar interactions with the side chains of S120 and E123. It is thus likely that Y121 is the predominant contributor of the TL1A DE loop to DcR3 binding. Notably, the analogous segment in DR3 contains several hydrophilic residues (E82, H84 and H85) that can potentially participate in polar interactions with the side chains of S120 and E123 in TL1A, suggesting detailed differences in the mechanisms used to recognize the two receptors.
The C-terminal end of AA′ loop in conventional TNF ligands is positioned at the same relative height within the trimer as the DE loop (Figure 7 and SI Figure 2) and represents another critical region in the lower level of the receptor binding interface (40). Residues corresponding to TL1A residues E53 and G57 are among the most conserved in the receptor binding groove of the TNF family ligands. In TL1A, the E53A-E55A double mutant and the G57D single mutant both exhibited substantial losses of binding affinity for DcR3 (Figure 6). Similarly, mutation of the G57 analog in LTα (D50) dramatically decreased the affinity of receptor binding (38). The equivalent of TL1A E53 in TRAIL is E155, and its side chain forms a salt bridge with the R115 guanidine group of the DR5 receptor (17). This Arg is conserved in DR3, but is substituted by Tyr in DcR3 (Y89 in SI Figure 9). Thus, compared with the TRAIL/DR5 structure, E53-E55 in TL1A might interact with DcR3 in a different fashion. Notably, the gel filtration and thermal melting experiments suggest that G57D and E53A-E55A mutants may contribute to DcR3 binding through destabilizing the protein structure or altering the oligomeric state (SI Figure 5D, 5E, 5F).
In addition to the C-terminal end of AA′ loop, the TRAIL/DR5 structure shows that the middle region of the extended TRAIL AA′ loop also plays an important role in receptor binding (17). This interaction is precluded in the LTα/TNFR1 structure because of a shorter AA′ loop (17). Similar to TRAIL, TL1A also possesses a relatively long AA′ loop (Figure 1). However mutations in the middle region of this loop (i.e., F43A-K44A double mutant) did not change the DcR3 binding affinity, suggesting that the C-terminal end of AA′ loop in TL1A provides the most important binding determinants in this loop.
The upper level of the receptor binding interfaces in both TRAIL and LTα is formed predominately by contributions from the CD loops (Figure 7). Compared with most TNF ligands, the CD loop in TL1A is significantly longer (Figure 1, SI Figure 2). The R89A single mutation and R99A-R103A-D108A triple mutation in the TL1A CD loop have no effect on the interaction between TL1A and DcR3 (Figure 6). These results again indicate differences in receptor recognition between TL1A and TRAIL/LTα, and suggest that the CD loop of TL1A may not serve as part of the receptor binding interface. Importantly, due to the limitations of mutagenesis approaches, at this point, we cannot exclude interactions contributed by the backbone atoms of TL1A to the binding interface.
In the “center” of the receptor binding groove of TL1A, K173, E174, D175 and K176 contributed from the tip of the GH loop form a region with considerable ionic character (Figure 7). In most TNF ligands this region is composed of charged or hydrophilic residues (Figure 1); however, neither LTα nor TRAIL utilize this region for receptor binding(16, 17). In TL1A, the E174A single mutant and K173A-K176A double mutant both significantly reduced the DcR3 binding affinity (Figure 6), suggesting additional differences in TL1A-DcR3 recognition compared to other family members.
The structural and mutagenesis studies map the DcR3-binding surface of TL1A, and show both common and unique features compared with known TNF ligand-receptor complexes. The DcR3-binding interface appears to involve the surface of the lower half of the groove formed between two TL1A protomers, with additional contributions from the central region, but few interactions involving the upper half of the TL1A molecule (Figure 7C). These features suggest an asymmetric interaction pattern for DcR3 along the interprotomer groove in TL1A. By contrast, other TNF ligands interact with their cognate receptors in more symmetrical fashion, such that the bound receptor runs parallel to the ligand three-fold symmetry axis with the interaction surfaces evenly distributed along the lower and upper parts of the ligand. Notably, potential interactions between the side chain atoms of the TL1A DE loop and DR3 suggest possible differences in TL1A recognition by the functional receptor DR3 and decoy receptor DcR3. Finally, and most intriguingly, the apparent asymmetric interaction between DcR3 and TL1A may be an important element for the unique ability of DcR3 to recognize and neutralize multiple TNF ligands (i.e., TL1A, LGIHT and FasL).
In summary, our structural and biochemical data confirms the stable trimeric assembly of the TL1A cytokine, maps the DcR3 binding sites on TL1A, predicts interactions that might contribute to binding specificity and highlights unique features of the TL1A-DcR3 interaction that may be important for biological function. We also characterized a series of TL1A cysteine mutants with enhanced solution properties that are structurally and functionally similar to the wild type protein. Together these findings provide important tools and insights for the further mechanistic dissection of TNF/TNFR function.
We gratefully acknowledge the staff of the X29 beamline at the National Synchrotron Light Source for their assistance in data collection and Dr. Hui Xiao at Laboratory for Macromolecular Analysis & Proteomics, Albert Einstein College of Medicine, for assistance with mass spectroscopy. This work was supported by National Institutes of Health Grant AI07289 (to S.G.N. and S.C.A.), the Albert Einstein Cancer Center Grant (National Cancer Institute, P30CA13330) and the Albert Einstein College of Medicine Macromolecular Therapeutics Development Facility.
†This work was supported by NIH grants AI07289 (to S.G.N. and S.C.A.), the Albert Einstein Cancer Center (P30CA13330) and the Albert Einstein Macromolecular Therapeutics Development Facility.
¶The structure of the wild type TL1A protein and two mutant proteins, C95S and C95S/C135S, have been deposited to the PDB as entries 2QE3, 2RJK and 2RJL, respectively.
Supporting Information Available
Protocols for DcR3 expression and purification, and structural comparison. Illustrations for structural superimposition of TL1A with conventional TNF ligands, highly conserved residues in TL1A and TNF receptor sequence alignment. Kinetics analysis of the interaction between immobilized DcR3-Ig fusion protein and TL1A mutants. Thermal stability of TL1A mutants. This information is available free of charge via the Internet at http://pubs.acs.org.