Evolutionary conservation of TWEAK–Fn14 interactions
Human and Xenopus Fn14 (referred to as hFn14 and xeFn14) share only 37% sequence identity overall and 40% in the extracellular domain , but share several common features (Fig. 1), including a single extracellular CRD with six identically spaced cysteine residues that are separated by short stretches of homologous amino acids. To test the role of the conserved residues in interacting with TWEAK, we measured the binding of human or Xenopus Fn14 to human TWEAK by fluorescence-activated cell sorting (FACS), enzyme-linked immunosorbent assay (ELISA) and surface plasmon resonance (SPR). Figure 2A shows a FACS analysis in which we transfected full-length hFn14 or xeFn14 into 293T cells and measured the binding of Fc-hTWEAK (a fusion protein between the Fc portion of human IgG1 and the TNF homology domain of hTWEAK). Similarly, Fig. 2B shows ELISA results for Fc-hTWEAK titrated on plates coated with human or Xenopus Fn14. We observed comparable binding to proteins of both species, with an EC50 of 0.3 nm for Fc-hTWEAK binding to each receptor. The values are in agreement with those recently reported for binding of a conjugated form of human TWEAK to hFn14 in human HT1080 and HT29 cells . The high affinity detected for the xeFn14/hTWEAK interaction prompted us to measure the true, monovalent affinity of human and Xenopus Fn14 for human TWEAK by SPR. Figure 2C,D shows the results of equilibrium binding of monomeric Xenopus or human Fn14 to hTWEAK measured by SPR. hFn14 binds to hTWEAK with a KD of 65 nm (Fig. 2C), while xeFn14 binds to hTWEAK with a KD of 1.3 μm (Fig. 2D); a 20-fold lower affinity than hFn14. This difference, which cannot be detected in avidity-prone, multivalent interaction assay formats , becomes apparent only when the single-site binding affinity is measured. Importantly, hTWEAK induced similarly robust nuclear factor κB (NF-κB) activation when co-transfected with either hFn14 or xeFn14 in 293T cells, indicating that the observed binding interaction with xeFn14 also results in functional intracellular signaling (Fig. 2E). By comparison, cells transfected with TWEAK alone show only a slight increase in NF-κB signaling, which was negligible compared to that obtained by co-transfection with Fn14 (1.5-fold increase, P = 0.01, which may be attributed to stimulation of endogenous Fn14). Taken together, these data strongly suggest that endogenous interactions between TWEAK and Fn14 are mostly highly conserved during evolution.
Figure 1. Sequence alignment of Fn14 and other small TNF receptors. (A) Sequence alignment of Fn14 from human (Homo sapiens), mouse (Mus musculus), frog (Xenopus laevis), fish (Gasterosteus aculeatus) and a protochordate (Branchiostoma floridae). Conserved residues in the extracellular domain are shaded. Numbering corresponds to the human sequence. TMD, transmembrane domain; TNF receptor associated factor (TRAF), consensus TRAF binding site. Exon boundaries, when known, are indicated. (B) Sequence alignment of the central portion of the extracellular domains of single-CRD TNF receptors and TNFR1. The elements of secondary structure (β-sheets in green, α-helices in red) are indicated, as well as the disulfide bonds. Residues with a purple background were experimentally shown to participate in ligand binding (for Fn14, BCMA, TACI and BAFFR) or to contact the ligand in the ligand–receptor complex (TNFR1). Yellow and orange backgrounds indicate the Cys pairs. The module composition of the CRD is shown on the right. dWengen, Drosophila Wengen.
Download figure to PowerPoint
Figure 2. Cross-species binding of Fn14 to human TWEAK. (A) 293T cells were transiently transfected with the indicated amount of plasmids encoding full-length hFn14 or xeFn14, stained with Fc-hTWEAK at the indicated concentrations, and analyzed by FACS. MFI, mean fluorescence intensity. (B) ELISA assay. Immobilized hFn14 and xeFn14 binding to Fc-hTWEAK and detection with secondary reagents against human Fc. A four-parameter equation fitting yielded an apparent EC50 of 0.3 nm. (C,D) Binding of monomeric Fn14 (in solution) to immobilized trimeric His-tagged hTWEAK on a Biacore chip. The plot of the signal achieved at equilibrium (Req) as a function of the hFn14 (C) or xeFn14 (D) concentration was fitted to a single-site binding model as described in Experimental procedures. (E) Induction of NF-κB-luciferase reporter activity in 293T cells transfected with varying amounts of xeFn14 or hFn14 plasmids, in the presence or absence of a fixed amount of co-transfected TWEAK. The signal in cells expressing TWEAK in the absence of co-transfected Fn14 is represented by the white square and the white circle (hidden below the white square) at 0 ng per well of Fn14. NF-κB activity is expressed as a fold increase relative to cells transfected without Fn14.
Download figure to PowerPoint
Solution structure of hFn14 and xeFn14 extracellular domains
To understand the structural basis underlying the high level of cross-species conservation of TWEAK–Fn14 interactions, we determined the structure of the extracellular ligand-binding domain of both hFn14 and xeFn14 by NMR (Fig. 3A,B; Table 1). The solution structure obtained for hFn14 is in agreement with the published structure , with pairwise root-mean-square deviations (RMSDs) in the structured regions of 0.62 and 0.47 Å (family of 20 structures versus PDB code 2RPJ, structure 1) and the same tertiary arrangement. hFn14 and xeFn14 share strong similarity in terms of the NMR spectral pattern (chemical shifts and Nuclear Overhauser Effects (NOEs), Fig. S1), which translates to highly similar secondary and tertiary structures (RMSD 1.0 Å over secondary structure regions: residues 40–52 and 61–64, backbone atoms, Fig. 3B; numbering as in hFn14). The disulfide pattern of hFn14 was determined using tandem mass spectrometry , and was further confirmed by NMR for both hFn14 and xeFn14. Both proteins adopt a nested pattern for the two most C-terminal disulfide bonds, in contrast to the intercalated pattern observed in BCMA and TACI_d2 (second CRD of TACI), the two TNFR super-family members previously thought to be most closely related to Fn14 (Figs 1B and 3A,C). The structures of the highly ordered β-hairpin domains (RMSDs 0.24 and 0.20 Å, respectively, for each ensemble of 20 structures of hFn14 and xeFn14) converge to an RMSD of 0.56 Å. This area of the molecule has been previously reported to provide the majority of the interaction interface with the ligand for BCMA and BAFFR, and to be essential for binding in TACI and multi-domain TNF receptors (TNFR1 and DR5, death receptor 5) [18, 19].
Figure 3. NMR solution structure of human and Xenopus Fn14. (A) Superposition of 20 low-energy NMR-derived structures of hFn14 (backbone atoms 36–67). The three disulfide bridges are shown by yellow ball-and-stick models for one of the structures. The boundaries of the A1 and C2 structural modules are indicated by arrows. (B) Superposition of low-energy structures of hFn14 (gray) and xeFn14 (red), displayed as ribbons. Residues 36, 40–52 and 61–64 of hFn14 and corresponding residues of xeFn14 were used in the superposition. (C) Top view of hFn14 with cysteines highlighted by yellow ball-and-stick models. The arrows indicate the changes in disulfide connectivity that are required to swap from nested to intercalated disulfide bridges in the C2 module. (D) The hydrophobic core of hFn14 comprises Phe63 and Met50 (carbon atoms in green), and is surrounded by small amino acids (Ser41, Ser43, Ser54 and Ser61, carbon atoms in blue). Protons are colored white, oxygens are colored red, and sulfur is colored yellow.
Download figure to PowerPoint
Table 1. Restraints and statistics for the ensemble of 20 structures of hFn14, xeFn14 and BCMA
|Medium range 1 < |i − j | < 5||28||31||45|
|Long range |i − j | ≥ 5||34||50||56|
|RMSD from experiment|
|NOE constraints (Å)||0.015||0.035||0.060|
|RMSD from idealized geometry|
|Most favored regions (%)||56.3 ± 6.7||61.2 ± 5.0||56.0 ± 4.5|
|Additionally allowed regions (%)||34.2 ± 5.7||34.0 ± 5.9||30.8 ± 4.5|
|Generously allowed regions (%)||5.8 ± 3.4||3.1 ± 2.3||8.9 ± 2.4|
|Disallowed regions (%)||0.2 ± 0.7||0.1 ± 0.4||0.2 ± 0.6|
|Average pairwise RMSDs (Å)|
|Ser40/xPro31 to Cys67/xCys58||0.747||0.760|| |
The relative arrangement of the C-terminal lobe of Fn14 proteins (helix 1, residues 52–55, loop, and helix 2, residues 62–69) and the N-terminal β-hairpin is well defined by a number of long-range NOEs radiating from the central aromatic residue Phe63/xPhe54 (for simplicity, the first residue refers to hFn14 and the residue prefixed ‘x’ is the corresponding residue in xeFn14; NOEs to Ser41/xAla32, Trp42/xTyr33, Ser43/xSer34, Met50/xMet41). In the case of xeFn14, long-range contacts between the two acidic residues xAsp53 and xAsp36 may also be clearly observed. To quantitatively evaluate NOEs and their impact on the final structures, we utilized the program queen . The average information, Iave, for long-range restraints connecting residues in the loop (Phe63, Gly66, Cys67; xPhe54, xAsn57 and xCys58) to residues in the N-terminal region (Ser41, Ser43 and xArg31–xSer34) is 32% and 35%, respectively, of the total structural information, indicating a dominant role in the structure. NOEs in the helix 2 region have a higher contribution to the dataset for hFn14 than for xeFn14 (i.e. a better chemical shift dispersion is observed). Correspondingly, helix 2 in the human protein is better defined and folds into a more compact coil (Fig. 3B).
The observed structures closely resemble that of TACI_d2 (RMSD 0.52 Å between residues 40 and 50 of hFn14, and residues 77 and 87 of TACI_d2, lowest NOE violation structure, PDB code 1XUT) (Fig. 4A,C).
Figure 4. Comparison of the structural features of hFn14 with BCMA, TACI_d2 and BAFFR. Selected residues are displayed by ball-and-stick models: green, residues in the hydrophobic core; dark blue, residues relevant for binding in the β-hairpin and helix 2. The conserved aromatic residue is shown in light blue and disulfide bridges are shown in yellow. (A) NMR structure of hFn14. (B) NMR structure of human BCMA determined in this study. The Tyr13 that fulfils a similar structural role to Phe63 of Fn14 is shown. (C) NMR structure of hTACI_d2 (PDB code 1XUT). (D) Crystal structure of hBAFFR (PDB code 1OQE). The structures shown in (A–D) were first superimposed on their β-hairpins. Note that the N- and C-terminal domains adopt different relative orientations in Fn14 than in BCMA and TACI_d2, which may allow engagement of different epitopes in ligand binding.
Download figure to PowerPoint
Solution structure of BCMA
The relative orientation of the N- and C-terminal lobes of Fn14 determined by NMR is strikingly different from those of TACI_d2 and especially BCMA as determined by crystallography (Fig. 4) [12, 13]. We confirmed that the solution structure of BCMA is highly similar to the crystal structures of BCMA determined previously in complex with BAFF (B-cell activation factor of the TNF family) or APRIL (a proliferation-inducing ligand) [12, 21]; the backbone RMSD over the ordered regions is 1.06 Å (lowest NOE violation structure versus PDB code 1OQD). A similar analysis of the NMR restraints (queen) indicates that the relative orientation of the N- and C-terminal lobes is well characterized by long-range NOEs, which constitute about 19% of the total structural information (sum of Iave), involving residues homologous to those observed for Fn14 (Pro34, Thr36, Tyr40, Cys41 to Asn11, Tyr13 and Phe14).
Tertiary fold and binding epitopes for single-CRD TNFRs
Overall, the fold of Fn14 is similar to those of BCMA and TACI_d2 and partially similar to that of BAFFR, with the most striking similarity in the β-hairpin. The conserved DXL motif, located at the tip of the β-hairpin and essential for ligand binding [11, 12], is replaced in Fn14 by an SXDL sequence that contributes to ligand binding. Whereas the hydrophobic core in h/xeFn14 is very compact (Phe63/xPhe54 and Met50/xMet41, surrounded by a ring of small amino acids: Ser41/xAla32, Ser43/xSer34, Ser54/xVal45 and Ser61/xSer52, Fig. 3D), the hydrophobic core of TACI_d2 involves a number of bulkier hydrophobic residues, including Phe78, Ile87, Ile92 and Pro97  (Fig. 4C), causing a tertiary arrangement that is progressively more open going from Fn14 to TACI_d2 to BCMA (the N- and C-terminal lobes arrange in an ‘open clamp’ versus ‘closed clamp’ in Fn14, Fig. 4).
As a functional consequence of the tertiary structure, BCMA and BAFFR engage BAFF (and APRIL) mostly through the β-hairpin interface [13, 21], with no fundamental contribution from residues in the C-terminal helix-loop-helix region. The intermediate conformation in TACI_d2 involves a wider binding surface to APRIL, encompassing the concave face of the molecule (DXL hairpin plus residues in the helix-loop-helix motif) . In the ‘closed clamp’ conformation seen in both human and xeFn14, the β−hairpin and the helix-loop-helix motifs are in close proximity and define the binding epitope presented to TWEAK (Fig. 4).
A discontinuous binding epitope: mutagenesis and molecular structure
Previous studies have concluded that minimal ligand-induced structural changes occur when BAFFR, BCMA or TACI_d2 bind their respective ligands [11-13], an observation confirmed by the present solution structure of BCMA. A model of hFn14 bound to TWEAK was constructed based on the assumptions that (a) the structure of free Fn14 and TWEAK-bound Fn14 is similar, and (b) the region contacted by Fn14 in TWEAK corresponds to that contacted by BAFFR, BCMA and TACI_d2 in BAFF and APRIL. We based our homology model on the available structure of the APRIL–TACI complex (PDB code 1XU1) . Fn14 contacts TWEAK along a relatively narrow but elongated surface comprising, on one hand, the C-terminal end of the CD loop plus β-strand E of TWEAK, and, on the other hand, the β−hairpin and part of the C-terminal portion of Fn14.
The molecular model of the TWEAK–Fn14 complex, which takes into account the experimentally determined structure of Fn14, clearly highlights that residues belonging to the two lobes of Fn14 (the β−hairpin and the helix-loop-helix domain) make determinant contributions to the binding interface. Based on the complex structure, selected point mutations were introduced in Fn14 and TWEAK in order to probe potential residues involved in the receptor–ligand interactions. Full-length forms of Fn14 were expressed in 293T cells and stained with FLAG-tagged TWEAK in FACS analysis, or with a monoclonal antibody against Fn14 (ITEM-4) to control expression levels (Fig. 5A). Leu46 was deemed important for binding, and its mutation (L46A) reduced the binding affinity to TWEAK; all other Fn14 mutants tested bound FLAG-tagged TWEAK as efficiently as wild-type Fn14 did (Fig. 5B). Consequently, we introduced a series of double mutations in the hFn14 sequence that combine a residue from the β-hairpin's tip and a residue from the helix-loop-helix, expressed them in 293T cells, and tested them for binding to hTWEAK. Several double mutants resulted in partial or complete loss of binding to TWEAK: L46A/P59A, L46A/S61A, L46A/D62A and L46A/F63A (Fig. 5C). To confirm the importance of the SXDL motif, a number of double mutants with two mutated residues within the β-hairpin were also examined. Mutants S43A/L46A, D45A/L46A, D47A/L46A and K48A/L46A almost completely lost the ability to bind to TWEAK (Fig. 5C).
Figure 5. Mutagenesis of Fn14 and TWEAK. (A) The indicated full-length forms of wild-type or mutant Fn14 were expressed in 293T cells and stained by FACS with FLAG-tagged TWEAK at the indicated concentrations (1 μg·mL−1 corresponds to 50 nm), or with a monoclonal antibody against Fn14 (ITEM-4) as an expression control. (B) Binding of wild-type TWEAK to wild-type and mutant Fn14 was monitored as shown in (A). The mean fluorescence intensity of TWEAK binding on cells expressing 102–103 fluorescence units of EGFP [rectangle shown in the first scattergram of (A)] was normalized to the staining intensity obtained with ITEM-4, and plotted as a function of TWEAK concentration. (C) As for (B), except that the Fn14 mutations were combined with the L46A mutation. (D) 293T cells transfected with full-length wild-type Fn14 were stained with the indicated concentration of FLAG-tagged TWEAK with the indicated mutations. Only TWEAK mutation Y176A abolished binding to Fn14. (E) As for (D), but with the Fn14 L46A mutant. The TWEAK mutants K178A, R190A and L192E all failed to interact with the Fn14 L46A mutant, which displays a weaker affinity for TWEAK.
Download figure to PowerPoint
Similarly, we mapped the TWEAK binding interface, revealing the importance of Tyr176 (Y176A completely abolishes binding; Fig. 5D), and the contribution of Lys178, Arg190 and Leu198 (K178A, R190A and L192E all failed to interact with Fn14 L46A, the weakened receptor for TWEAK; Fig. 5E). The residues thus discovered all appear to be involved in specific receptor–ligand interactions, and are highlighted in Fig. 6. For example, Leu46 on Fn14 interacts with Leu192 of TWEAK, Asp47 of Fn14 interacts with Lys235 of TWEAK, and Ser61/Asp62 of Fn14 interact with Lys178 of TWEAK. Pro59, which has a moderate effect on binding affinity in double mutant Fn14, may play a structural role or additionally directly participate in binding, similar to the corresponding residue Pro97 of TACI_d2 (part of the hydrophobic surface that interacts with APRIL). We also attribute a structural role to Met50: as part of the hydrophobic core of Fn14 (only 20% of its surface is accessible), it is not predicted to contact TWEAK directly but rather to play a major role in determining the relative orientation of the N- and C-terminal lobes. A similar effect was reported previously for the equivalent position in TACI_d2 (Ile87)  (Fig. 4C). Several residues showed no contribution to binding (Fig. 5C): Arg56, Arg58 and Leu65 point away from the TWEAK interface in the model and make no or little contact with TWEAK, while His60 points toward Glu194 and Phe195 of TWEAK.
Figure 6. Molecular model of the TWEAK–Fn14 complex. (A) Residues of hFn14 that were tested in the mutagenesis study. Residues shown in red were critical for ligand binding (in combination with mutation L46A), the residue shown in orange decreased binding, and residues shown in green had no or little effect on ligand binding when mutated to alanine. Residues marked with an asterisk are within 4 Å of TWEAK in the model. Note that residues that are important for binding are located in both the A1 and C2 modules. (B) Model of hFn14 binding to hTWEAK. Residues of both Fn14 and TWEAK that are likely to participate directly in the interaction and whose mutation affects the binding are highlighted (green, hydrophobic; red, acidic; blue, basic). The TWEAK trimer is represented as ribbons in various shades of yellow. Fn14 is represented as a gray ribbon. Fn14 in (A) and (B) is rotated ~ 180°, to allow optimal viewing of the residues described.
Download figure to PowerPoint
Within the series of single-mutant hFn14 constructs, the finding that each mutation alone did not disrupt ligand binding (or only reduced it, e.g. L46A) indicates that none of these mutations resulted in gross unfolding or intracellular retention of hFn14. It is noteworthy that our results are in contrast with a previous report indicating that the D62A mutation totally prevented the secretion of Fn14 protein, and several single point mutations disrupted binding . Secretion of the soluble D62A mutant may be affected by the use of full-length human Fn14 in this study, versus use of a murine sequence of the extracellular domain of Fn14 fused to murine Fc and a Myc tag . We also tested mutations in neighboring amino acids, S61A and F63A, which yielded Fn14 mutants that are able to bind TWEAK. In addition, the binding assay modalities were largely different: binding affinities in this study were compared by FACS analysis on cells expressing the described Fn14 mutants with increasing concentrations of FLAG-tagged TWEAK (> 10 μg·mL−1), whereas the previous study  measured binding affinity to TWEAK immobilized on a solid surface at two concentrations of purified receptors, and the weakest detected binding was reported with a KD of 71 nm. To confirm correct folding of the mutant proteins, a representative double mutant, L46A/R58A hFn14, was expressed and purified, and examined by 1H,15N-HSQC NMR. The protein spectrum (Fig. S2) closely tracks that of wild-type hFn14, with chemical shift perturbations corresponding only to the sites adjacent to the mutated residues as expected, indicating that the wild-type and mutant proteins are similarly folded.
Taken together, our data indicate that the Fn14–TWEAK interaction involves elements from the β-hairpin as well as from the helix-loop-helix motif. The observation that single point mutations are well tolerated appears to indicate that the multiple sites of this non-contiguous interaction interface may contribute substantially to the energy of binding, which should be taken into account for development of low-molecular-weight molecules intended to modulate the TWEAK–Fn14 interaction.
Fn14: the evolutionary perspective
The TNF super-families of ligands and receptors have expanded considerably during evolution, from a single pair in Drosophila (Eiger–Wengen) to more than 20 in human, as higher organisms came to rely more and more on these ligand–receptor interactions for controlling cell fate decisions. However, the evolutionary origin and diversification of these pairs is incompletely understood. In the present study, we provide experimental evidence indicating that the interface between the TNF ligand TWEAK and its cognate receptor Fn14 has been greatly conserved during evolution.
Although it may seem remarkable that Xenopus Fn14 cross-reacts with human TWEAK, inter-species cross-reactivity is not uncommon in the TNF family: 72% of TNF ligands and receptors cross-react between human and mouse , human BAFF binds to chicken BAFFR , chicken BAFF binds to human BCMA , and mammalian ectodysplasin-A (EDA) is even active in fishes . The strong structural similarity observed between human and Xenopus Fn14 despite divergent primary sequences is probably due to the presence of a structural core stabilized by disulfide bonds, as shown previously for other protein families [26, 27].
Most TNF receptor family members, such as TNFR1, TNFR2 and Fas, contain three or four cysteine-rich domains, usually composed of A and B modules (see  for classification), suggesting that these receptors may have arisen by exon duplication of an ancestral cysteine-rich domain such as that found in Wengen in Drosophila  (Fig. 1B). The relatively recent identification of single-CRD receptors in mammals raised the intriguing possibility that they may represent ancestral molecules from which multi-CRD mammalian TNF receptors evolved . The TWEAK–Fn14 pathway may be one of the most ‘ancient’ from the perspective of evolution. The other two single-CRD TNF receptors BAFFR and BCMA regulate B-cell function in vertebrates only, whereas an Fn14-like sequence with perfectly conserved inter-cysteine spacing and exon–intron boundaries is found in the protochordate Branchiostoma floridae (Fig. 1A). This suggests that the origin of the TWEAK–Fn14 pathway preceded vertebrate evolution and the development of adaptive immunity.
The structure of Fn14 allows formulation of a hypothesis regarding the evolutionary relationship between modules that differ in their patterns of disulfide bridges. Although the A1–C2 cysteine-bridging pattern described in both human and Xenopus Fn14 (the two C2 disulfides are nested) contrasts with the A1–D2 pattern of BCMA and TACI (intercalated disulfides), the three-dimensional structure of Fn14 is able to accommodate the A1–D2 disulfide pattern with only minor structural adjustments (Fig. 3C), suggesting that a certain level of ‘ambiguity’ in disulfide linkage formation may have existed in the primordial single-CRD TNF receptor, and that this flexibility may account for the emergence of various modules found in mammalian TNF receptors. Similarly, swapping of disulfide pairing has been reported to produce bioactive disulfide bond isomers of epidermal growth factor-like domains .