Insights into ligand binding by a viral tumor necrosis factor (TNF) decoy receptor yield a selective soluble human type 2 TNF receptor

Etanercept is a soluble form of the tumor necrosis factor receptor 2 (TNFR2) that inhibits pathological tumor necrosis factor (TNF) responses in rheumatoid arthritis and other inflammatory diseases. However, besides TNF, etanercept also blocks lymphotoxin-α (LTα), which has no clear therapeutic value and might aggravate some of the adverse effects associated with etanercept. Poxviruses encode soluble TNFR2 homologs, termed viral TNF decoy receptors (vTNFRs), that display unique specificity properties. For instance, cytokine response modifier D (CrmD) inhibits mouse and human TNF and mouse LTα, but it is inactive against human LTα. Here, we analyzed the molecular basis of these immunomodulatory activities in the ectromelia virus–encoded CrmD. We found that the overall molecular mechanism to bind TNF and LTα from mouse and human origin is fairly conserved in CrmD and dominated by a groove under its 50s loop. However, other ligand-specific binding determinants optimize CrmD for the inhibition of mouse ligands, especially mouse TNF. Moreover, we show that the inability of CrmD to inhibit human LTα is caused by a Glu-Phe-Glu motif in its 90s loop. Importantly, transfer of this motif to etanercept diminished its anti-LTα activity in >60-fold while weakening its TNF-inhibitory capacity in 3-fold. This new etanercept variant could potentially be used in the clinic as a safer alternative to conventional etanercept. This work is the most detailed study of the vTNFR–ligand interactions to date and illustrates that a better knowledge of vTNFRs can provide valuable information to improve current anti-TNF therapies.

The tumor necrosis factor superfamily (TNFSF) 4 comprises 19 cytokines and 29 cellular receptors involved in essential biological processes such as cell death, immunity, and organ development (1). However, a deregulated production of these cytokines can provoke autoimmunity and inflammatory disorders. For instance, Crohn's disease, psoriasis, ankylosing spondylitis, rheumatoid arthritis, and inflammatory bowel disease are often associated with the uncontrolled activity of tumor necrosis factor ␣ (TNF), the archetypical cytokine of the family (2). To date, there are five TNF inhibitors approved by the Food and Drug Administration for the treatment of these diseases: four monoclonal antibodies (infliximab, adalimumab, certolizumab, and golimumab) and a soluble decoy receptor, etanercept (3). Etanercept is a fusion protein of the ligand-binding domain of the cellular TNF receptor 2 (TNFR2) with the Fc portion of a human IgG1 (4). Importantly, despite its proven efficiency in ameliorating patient symptomatology, etanercept can also cause major adverse effects, including increased susceptibility to infections, lymphoma, and heart failure (5). These side effects might be explained, at least in part, by the fact that etanercept blocks not only TNF but also other TNFSF ligands essential for homeostasis and immunity like lymphotoxin ␣ (LT␣) (6). Redesign of this molecule appears to be desirable for its safer application in the clinic.
Poxviruses have evolved a similar strategy to inhibit TNFSF ligands in their hosts. Members of the Orthopoxvirus genus encode up to four different soluble viral TNF decoy receptors (vTNFRs), termed cytokine response modifier B (CrmB), CrmC, CrmD, and CrmE, that display differential ligand and species specificity profiles (7). For instance, CrmC and CrmE are specific mouse TNF (mTNF) and human TNF (hTNF) inhibitors, respectively, whereas CrmB and CrmD inhibit TNF, LT␣, and LT␤ (8 -12). Furthermore, although CrmD, the only active vTNFR encoded by the mouse-specific ectromelia virus (ECTV), is the vTNFR with the highest binding affinity for mouse LT␣ (mLT␣), it fails to neutralize human LT␣ (hLT␣) (12). In fact, CrmB, encoded by the human variola virus, is the only vTNFR that blocks hLT␣ (12). Thus, it appears that vTNFRs have evolved to fulfill the particular immunomodulatory needs of poxviruses according to their host species. This level of specialization to discriminate between mouse and human cytokine counterparts or between the highly related TNF and LT␣ is rare among cellular TNFSF receptors (TNFRs) (13). Therefore, understanding the molecular determinants of the vTNFR-ligand interactions could reveal new molecular strategies to improve the TNF specificity of etanercept and increase its clinical safety.
To date, the uncomplexed form of CrmE is the only vTNFR whose structure has been solved (14). This structure confirmed that vTNFRs mimic the three-dimensional folding of the ligand-binding moiety of cellular TNFRs, which is formed by a variable number of cysteine-rich domain (CRD) pseudorepeats. The folding of a typical CRD is maintained by three disulfide bonds established by six highly conserved cysteines (15). Cellular TNFRs may comprise up to five CRDs (16). 15 of the 29 different TNFRs contain at least three CRDs. In most of these, CRD2 and CRD3 constitute the principal ligand-binding sites (17). In particular, two loops located in these two CRDs and designated the 50s and the 90s loop, respectively, are known to act as the dominant ligand-binding determinants in several receptor-ligand complexes (18 -20). In contrast, the CRD1, although usually not directly involved in ligand binding, can mediate the self-association of some cellular TNFRs in a ligandindependent manner, which has been proposed to enhance the ligand-binding affinity and the signaling potency of the receptor (21)(22)(23). For this reason, the CRD1 was termed preligand assembly domain (PLAD).
However, little is known about how vTNFRs interact with their ligands. T2, a CrmB homolog encoded by the Leporipoxvirus myxoma virus (MYXV), is the only vTNFR whose ligandbinding site has been characterized to some extent. Analysis of T2 truncated mutants showed that, like in many cellular TNFRs, the CRD2 and CRD3 were essential for TNF binding (24). However, the precise molecular bases of the vTNFRligand interactions remain mostly unexplored. Interestingly, a PLAD-like function has been attributed to the CRD1 of MYXV T2. It was shown that T2 can interfere with TNF signaling in a ligand-independent manner by interacting with the CRD1 of TNFR1 or TNFR2 to generate unresponsive heterotrimers (25). Conversely, the structure of CrmE did not confirm the existence of a PLAD in other vTNFRs (14), and whether the CRD1 can induce self-oligomerization in vTNFRs is not completely understood. Furthermore, even the number of CRDs that constitute the TNF-binding domain of vTNFRs remains controversial, and precise allocation of the TNF-binding moiety in these viral decoy receptors is still missing. This is especially important in the case of CrmB and CrmD, whose TNF-binding domain precedes an extended C-terminal domain absent from CrmC and CrmE and named smallpox-encoded chemokine receptor (SECRET) domain that binds and inhibits a limited number of chemokines (26). We have recently demonstrated that CrmD is an essential virulence factor for ECTV, which causes mousepox, a smallpox-like disease in mouse. In addition, we showed that both CrmD activities, anti-TNF and antichemokine, are required for a successful evasion of the host immune response by ECTV (27).
Here, we precisely define the TNF-binding moiety of CrmD, discuss the existence of a PLAD in this vTNFR, and provide new insights into the molecular bases of its interaction with TNF and LT␣ of mouse and human origin. Importantly, we applied the information extracted from our CrmD analysis to severely curb the anti-LT␣ activity of etanercept without dismantling its TNF-inhibitory properties. This study offers molecular data that might help to understand the role of each CrmD-inhibitory activity in ECTV pathogenesis and provides proof of principle that structural characterization of vTNFRs may be of great value to improve current anti-TNF therapies.

Fine mapping of the TNF-and chemokine-binding domains of CrmD
The TNF-binding domain of CrmD is proposed to be formed by four CRDs (10); however, the amino acids that mark the end of the TNF-binding moiety and the beginning of the antichemokine SECRET domain have never been experimentally identified. Therefore, we first wanted to precisely delimit the amino acid sequence that constitutes the TNF-binding domain of CrmD.
We addressed this question indirectly by identifying the beginning of the SECRET domain. Three chemokine-binding proteins with sequence and functional similarities to the SECRET domain have been characterized in poxviruses, named SECRET-containing protein 1 (SCP1), SCP2, and SCP3 (26). As shown in Fig. 1A, the amino acid sequences of these SCPs align with that of the SECRET domain of CrmB and CrmD, a domain missing from CrmC and CrmE. Interestingly, an FN motif (Phe 163 and Asn 164 ) conserved at the N terminus of all SCPs is also present in the putative CRD4 of CrmD and CrmB but absent in CrmC and CrmE (Fig. 1A). This observation suggested that Phe 163 and Asn 164 of CrmD could actually be important for chemokine binding. To evaluate this, we compared the chemokine-binding affinities of two CrmD truncated mutants for two different forms of the SECRET domain: SECRET163 (Phe 163 -Asp 320 ), containing the Phe 163 and Asn 164 conserved in the SCPs, and SECRET181 (Ser 181 -Asp 320 ), which includes the C-terminal sequence immediately after the putative CRD4 of CrmD (Fig. 1B). These two SECRET domain forms were expressed and purified by recombinant baculoviruses alongside the full-length CrmD and CRD-CrmD (Asp 21 -Ser 162 ), a truncated CrmD mutant containing the sequence immediately N-terminal to the sequence of the SECRET163 protein (Fig. 1, B and C).

An etanercept mutant lacking anti-hLT␣ activity
whereas full-length CrmD bound both mTNF and Ccl25, CRD-CrmD and SECRET163 interacted only with mTNF or Ccl25, respectively. Importantly, SECRET163, but not SECRET181, fully conserved the chemokine-binding affinities of the fulllength protein ( Fig. 2A). For instance, although CrmD bound Ccl25 and Ccl27 with a K D of 4.86 and 5.41 nM, the binding affinities of SECRET181 for these chemokines were 54.70 and 288.0 nM, a 1-log and about 2-log K D increase, respectively ( Fig.  2A). By contrast, SECRET163 maintained the binding affinities of CrmD for these two chemokines (K D , 5.99 and 7.91 nM, respectively; Fig. 2A). Complementarily, the binding affinities of CRD-CrmD (K D , 0.21-12.41 nM) for TNFSF ligands were similar to those of the full-length CrmD (K D , 0.10 -50.40 nM) in all cases ( Fig. 2A). These results indicated that the sequence from the putative CRD4 included in SECRET163 is dispensable for the binding of TNFSF ligands but required for high-affinity chemokine interactions.
We then analyzed whether CRD-CrmD and SECRET163 also conserved the capacity of the full-length protein to inhibit the bioactivity of TNF and chemokines, respectively. As shown in Fig. 2C, SECRET163 inhibited the Ccl25-induced chemotaxis of MOLT-4 cells to the same extent that CrmD did, reducing cell migration nearly 50% when incubated with the chemokine at a 2 M excess and almost neutralizing chemotaxis

An etanercept mutant lacking anti-hLT␣ activity
at a 10 M excess (Fig. 2C). As expected, CRD-CrmD did not inhibit the cell migration induced by Ccl25 (Fig. 2C). In contrast, CRD-CrmD, but not SECRET163, protected L929 cells from the cytotoxic effect of mTNF. A 10 M excess of both CrmD and CRD-CrmD protected at least 50% of the cells against mTNF (Fig. 2D). These results confirmed our conclusions based on the binding assays and indicated that the second half of the putative CRD4 is not involved in TNF binding but it is most likely part of the SECRET domain.

Expression and purification of CrmD mutants
To study the molecular bases of the CrmD interaction with TNFSF ligands, we performed alanine site-directed mutagenesis of its TNF-binding domain, which we had just delimited to the sequence Asp 21 -Ser 162 . We generated 14 different fulllength CrmD mutants affecting residues throughout this sequence, including some related to three structural features known to be important for ligand binding in most TNFRs: the PLAD and the 50s and 90s loops.
The CRD1 or PLAD mediates the oligomerization of TNFRs prior to the interaction with the ligand (21). A similar activity was reported for the CRD1 of MYXV T2 (25), suggesting that the PLAD could be an activity extendable to the CRD1 of vTNFRs. However, beyond the six canonical Cys residues, there is no discernible conserved PLAD sequence across the CRD1 of vTNFRs and TNFRs (Fig. 3A). TNFR1 K19A/Y20A and K32A mutants are known to lose the self-association and ligand-binding abilities of the wildtype (WT) receptor (28). These residues are somewhat conserved in CrmD (Fig. 3A). Therefore, to evaluate whether they play a similar role in vTNFRs, we generated the CrmD mutants D35A/Y36A and K44A (Fig. 3A). Using size exclusion chromatography and native gel electrophoresis, we found that the oligomeric state of the protein was unaltered in the CrmD D35A/Y36A mutant (Fig. S1), which argued against a central role of these residues in the potential oligomerization of CrmD.
The 50s loop in the CRD2 and the 90s loop in the CRD3 are known to be essential ligand-binding determinants in most TNFRs (18, 20, 29 -31). Initial binding models proposed that conserved hydrophobic interactions in the 50s loop, critical for the binding affinity, orient the ligand to favor polar interactions by the 90s loop, which defines the ligand specificity (18). How-

An etanercept mutant lacking anti-hLT␣ activity
ever, a significant number of other TNFSF ligand-receptor complexes are known to diverge from this general model (19,20,30,32,33). Therefore, the precise ligand-binding determinants of TNFRs can be very variable and difficult to predict solely from structural models. In fact, even the application of receptor-based molecular replacement has proved to be problematic for the structural determination of some TNFR-ligand complexes (34). In this situation, we selected our CrmD alanine mutants (targeted residues highlighted in red in Fig. 3A) based on its alignment with other vTNFRs and its more highly related cellular counterparts, TNFR1 and TNFR2 (Fig. 3A). We mutated residues unique to CrmD as well as conserved amino acids either within the 50s and 90s loops or in other regions of the CRD2 and CRD3 (Fig. 3A). To guarantee that at least one of our mutants lost all ligand-binding capabilities, besides its corresponding alanine mutation, we generated CrmD N77F, equivalent to a reported TNFR1 N66F mutant known to be unable to interact with TNF (21).
All 14 CrmD mutants were expressed by recombinant baculoviruses and purified from supernatants of infected insect cells (Fig. 3B). To rule out the possibility that the introduced mutations drastically affected the overall CrmD three-dimensional folding, we tested the ability of the mutants to inhibit Ccl25-induced chemotaxis. As shown in Fig. 3C, the chemokine-inhibitory activity of the WT protein was preserved in all mutants.

Conserved and ligand-specific molecular bases drive the interaction of CrmD with TNF and LT␣ of mouse and human origin
To define the implication of the mutated residues in the CrmD ligand-binding mechanism, we determined the binding affinity of all mutants for four different CrmD ligands, mTNF, hTNF, mLT␣, and hLT␣, by SPR (Table 1). Fig. 4A shows some examples of the sensorgram fittings performed to calculate the binding affinity and kinetic constants presented in Table 1. Those cases in which no binding was detected or the binding was so weak that it did not allow a reliable determination of the affinity constants within the system detection limits are indicated as nbd for "no binding detected" (Table 1 and Fig. 4B). Fig.  4B summarizes the -fold change caused by each mutation on the corresponding ligand binding K D of the WT protein. Fig. 4C shows the position in the surface of a CrmD structural model of those residues that when mutated resulted in a significant modification of the protein K D for each ligand. CrmD N77F was not able to interact with any cytokine (Table 1 and Fig. 4B); how-

An etanercept mutant lacking anti-hLT␣ activity
ever, only the alanine mutation of this residue was considered for the construction of Fig. 4C.
Some similarities were found in the molecular bases that drive the interaction of CrmD with mTNF, hTNF, mLT␣, and hLT␣. In particular, certain residues in the CRD1 (Asp 35 and Tyr 36 ), CRD2 (Phe 72 , Thr 73 , Ser 74 , Gln 103 , and Asp 104 ), and CRD3 (Asp 146 and Ile 148 ), although to a different extent in some cases, were consistently important for the CrmD interaction with all four ligands (Fig. 4, B and C). The most evident of these common binding determinants was a groove in the CRD2 (Fig.  4C). Mutation of Phe 72 , Thr 73 , Ser 74 , Gln 103 , and Asp 104 , located in a CRD2 groove formed under the 50s loop, abolished all ligand-binding abilities of CrmD (Fig. 4, B and C). Similarly, mutation of two residues of the distal CRD3, Asp 146 and Ile 148 , consistently reduced the binding affinity for all ligands (8.9-, 3.6-, 3.3-, and 5.4-fold K D increases for mTNF, hTNF, mLT␣, and hLT␣, respectively) (Fig. 4B). In addition, mutation of three other amino acids in the C-terminal region of CRD3, Tyr 141 , Ser 142 , and Ser 143 , also decreased the CrmD-binding affinity for mTNF (8.6-fold), hTNF (5.4-fold), and mLT␣ (11.9-fold) but not for hLT␣ (Fig. 4B). These results indicated that this C-terminal region of the CRD3 is important for the ligand-binding properties of CrmD. Finally, we found that mutation of Asp 35 and Tyr 36 in the CRD1 significantly increased the K D for all ligands, with a more markedly deleterious effect on mTNF (17.3-fold K D increase) than on hTNF or mLT␣ binding (5.3and 3.7-fold K D increase, respectively) (Fig. 4B). Surprisingly, we did not detect hLT␣ binding by this D35A/Y36A mutant (Table 1 and Fig. 4, A and B); however, it is important to note that due to cytokine supply limitations, 1 M was the highest ligand concentration included in our experiments. In these conditions, given the lower binding affinity of CrmD for hLT␣ (K D , 50.40 nM), 20-fold was the highest K D increase detectable for this ligand. The same applies to the mutant Q93A/E95A, which did not bind hLT␣ but conserved the hTNF-and mLT␣binding affinity of the WT protein and displayed a 9-fold increase in the mTNF K D (Fig. 4B).
We also identified several ligand-specific differences among our CrmD mutants. First, interestingly, the K D for mTNF was increased in many more mutants than for any other cytokine ( Fig. 4B). Besides the conserved binding determinants mentioned above, mutation of amino acid residues located in the 50s loop (His 78 and Ser 79 ), the 90s loop (Glu 116 , Phe 117 , and Glu 118 ), and the connecting region between these loops (Arg 86 , Lys 88 , Ser 90 , Ser 91 , Gln 93 , and Glu 95 ) reduced the mTNF-binding affinity without significantly damaging the binding of other ligands (Fig. 4B). This suggests that CrmD might establish a closer contact to mTNF with more residues involved in the interaction. Interestingly, many of these mTNF-specific binding determinants appeared to obstruct the hLT␣ binding by CrmD. For instance, mutation of His 78 and Ser 79 or Arg 86 and Lys 88 increased the hLT␣-binding affinity 4.3-and a 3.7-fold, respectively (Fig. 4B). More strikingly, an E116A/F117A/E118A mutant displayed a remarkable 22.7-fold surge in the hLT␣binding affinity (Fig. 4, A-C, and Table 1).
We concluded that the core of the molecular mechanism of CrmD binding to mTNF, hTNF, mLT␣, and hLT␣ is formed by essentially the same group of amino acids. However, consistent with the fact that CrmD is expressed by ECTV, a strictly mouse pathogen, additional residues appear to have specialized CrmD for the interaction with mTNF at the cost of weakening its interaction with hLT␣, a biologically irrelevant cytokine for a virus that does not infect humans.

Inhibition of TNFSF cytokines by CrmD mutants
Although vTNFRs usually neutralize their ligands, there are examples of noninhibitory vTNFR-ligand high-affinity interactions. For instance, CrmE binds both mTNF and hTNF with a binding affinity below the nanomolar range but only blocks hTNF (12). This suggests that the ligand-binding and -neutralizing determinants may differ in vTNFRs. Therefore, to complete our understanding of the CrmD biochemistry, we analyzed the ability of each mutant to inhibit the cytotoxic activity of mTNF, hTNF, mLT␣, and hLT␣ on L929 cells.
As shown in Fig. 5, in the absence of inhibitors, mTNF, hTNF, mLT␣, and hLT␣ were cytotoxic and killed the cells (Fig.  5, Cytokine). However, in the presence of recombinant protein (Fig. 5, CrmD WT or mutant), added at the indicated cytokine: protein molar ratios, different cell survival rates were recorded. As expected, those mutants with no ligand-binding capacity,

An etanercept mutant lacking anti-hLT␣ activity
the N77F control mutant and the CRD2 groove mutants F72A/ T73A/S74A and Q103A/D104A/R105A, failed to protect the cells from the activity of any of the four cytokines tested (Fig. 5). Therefore, we concluded that the CRD2 groove is critical not only for the binding but also for the inhibitory properties of CrmD.
An evident ligand-specific effect was not detected among the other CrmD mutants. Mutations that impaired to some extent the capacity of CrmD to inhibit mTNF also affected to different degrees the anti-hTNF and anti-mLT␣ activity of the protein (Fig. 5). This suggests that CrmD deploys a similar molecular mechanism to inhibit these three cytokines. However, interestingly, the anti-hTNF effect of CrmD was more markedly susceptible to mutagenesis than its ability to block the mouse cytokines, which may suggest a more robust design to target murine ligands. For instance, low doses of CrmD WT (1:20) were sufficient to cause nearly 100% cell survival in both mTNF-and hTNF-mediated cytotoxicity assays (Fig. 5). However, at this same dose, the mutants D35A/Y36A, N77A, Q93A/E95A, and D146A/I148A were inactive against hTNF, whereas they pro-  Table 1. The corresponding analyte and CrmD mutant are indicated above each graph column and at the top right corner of each graph, respectively. The top graph row includes sensorgrams for CrmD mutants whose binding affinity for the indicated cytokine was comparable with that of WT CrmD. The middle row contains sensorgrams for mutants with significantly reduced (Y141A/S142A/S143A:mTNF, D35A/Y36A:hTNF, and N77A:mLT␣) or enhanced (E116A/F117A/E118A:hLT␣) binding affinity, and the bottom row shows examples where no binding was detected (nbd) or it was too weak to be analyzed accurately. B, -fold change (K D(mutant) / K D(WT) ) caused by each CrmD mutant on the K D of CrmD WT for mTNF, hTNF, mLT␣, and hLT␣. The K D -fold change range that was considered nonsignificant (0.3-3) is shaded in gray. nbd, no binding detected. In C, residues whose alanine mutation resulted in a significant increase (Ͼ3-fold change) or decrease (Ͻ0.3-fold change) in the K D of CrmD WT for each ligand are located, colored, and labeled on a model of the CrmD surface (gray) generated by I-TASSER (62). The corresponding ligands are shown above the CrmD models. The legend indicates the colors assigned to the different K D -fold change ranges. The maximum detectable K D -fold increase (red) varied among the different analytes (Ͼ20 for hLT␣, Ͼ200 for mLT␣, and Ͼ2,000 for mTNF and hTNF).

An etanercept mutant lacking anti-hLT␣ activity
tected at least 50% of the cells from mTNF (Fig. 5). Furthermore, the mutant Y141A/S142A/S143A showed no signs of hTNF-inhibitory capacity, whereas it protected 25 and 75% of the cells from mTNF and mLT␣ at the low and high protein dose, respectively. Most of these mutations that impaired the CrmD-inhibitory properties also displayed a reduced binding affinity for the corresponding cytokine. However, this was not always the case. For instance, the anti-mTNF activity of H78A/ S79A, R86A/K88A, and S90A/S91A, despite their significant loss of mTNF-binding affinity (Table 1 and Fig. 4B), was indistinguishable from that of CrmD WT. Therefore, these residues contribute to the binding but are dispensable for the neutralization of mTNF by CrmD.
Consistent with the lower binding affinity of CrmD for hLT␣ (K D , 50.40 nM; Table 1) and as reported previously (12), CrmD failed to inhibit this cytokine. Even in the presence of a 100 M excess of CrmD WT, hLT␣ killed 100% of the L929 cells (Fig. 5). This hindered our ability in this assay to study the functional effects of those mutations that reduced the CrmD-binding affinity for hLT␣. However, we found that the CrmD mutant E116A/F117A/E118A displayed a remarkable gain of anti-hLT␣ activity, reaching full cellular protection at the higher dose tested (Fig. 5). This was consistent with the high hLT␣binding affinity of this mutant (K D , 2.22 nM; Table 1). This result indicates that CrmD is perfectly equipped to inhibit hLT␣, but an EFE motif in its 90s loop, although innocuous for the capacity of the protein to block mTNF, hTNF, and mLT␣, keeps it inactive against hLT␣.

Transfer of the EFE motif of CrmD into the 90s loop of etanercept specifically impairs its anti-hLT␣ activity
The inflammatory diseases currently treated with etanercept are predominantly TNF-driven. Therefore, the anti-hLT␣ activity of etanercept not only appears to be clinically unnecessary, but it could also pose a source of unwanted complications. Having shown that a 90s loop EFE motif in CrmD specifically hinders its anti-hLT␣ activity, we hypothesized that, by transferring the EFE motif of CrmD into the 90s loop of etanercept, we could disrupt its anti-hLT␣ activity while keeping it active against hTNF.
The amino acid sequence alignment in Fig. 3A shows that the EFE motif (Glu 116 -Phe 117 -Glu 118 ) in the CRD3 of CrmD aligns with an ALS motif (Ala 127 -Leu 128 -Ser 129 ) in the 90s loop of human TNFR2, which corresponds to the TNF-binding moiety of etanercept. Furthermore, we observed that in the crystal structure of the TNFR2-TNF complex (Protein Data Bank (PDB) code 3ALQ) the TNFR2 Ser 129 was not facing the ligand (Fig. 6A). Similarly, the structural superimposition of a CrmD model with the structure of TNFR2 suggested that the third amino acid of this motif in CrmD, Glu 118 , would also be far from the ligand interface. Thus, to introduce the lowest number of modifications into the original sequence of etanercept, we mutated only the TNFR2 Ala 127 and Leu 128 to their equivalent amino acids in the 90s loop of CrmD (Glu and Phe). We generated recombinant baculoviruses for the expression of the double mutant A127E/L128F and the two single mutants A127E and L128F. Unfortunately, for unknown reasons, we were not able to successfully express the A127E protein in this system. SPR analysis revealed that the hTNF-and hLT␣-binding affinities of etanercept were not significantly affected in the mutants (Fig. 6B). However, as we have discussed above, binding affinity does not always correlate with neutralizing potency. Therefore, we compared the capacity of L128F, A127E/L128F, and WT etanercept to block the cytotoxic activity of hTNF and hLT␣ on L929 cells. As shown in Fig. 6C, 20 nM etanercept WT was enough to fully neutralize both hTNF and hLT␣ (Fig. 6C), reaching 50% cell viability at 10 -20 and 5 nM (EC 50 ), respec-

An etanercept mutant lacking anti-hLT␣ activity
tively. The L128F mutant was functionally indistinguishable from WT etanercept (Fig. 6C). In contrast, the A127E/L128F etanercept protected 50% of the cells from hTNF at 30 -60 nM and reached full protection at 90 nM (Fig. 6C). However, this double mutant showed a very low anti-hLT␣ activity and required a high 300 nM dose to protect only 35% of the cells from this cytokine (Fig. 6C), positioning its anti-hLT␣ EC 50 at even higher concentrations. Therefore, although the anti-hTNF activity of etanercept was reduced by 3ϫ in the A127E/ L128F mutant, this variant was at least a 60ϫ weaker hLT␣ inhibitor. These results confirmed our observations in CrmD and indicated that the 90s loop of viral and cellular TNFRs may contain important molecular determinants for the LT␣ species specificity.

Discussion
Multiple efforts have been made to characterize the immunomodulatory capacities of vTNFRs in vivo and in vitro; however, the molecular mechanisms of vTNFR-ligand interactions remain poorly analyzed. This work contributes to this gap of knowledge by delineating the molecular bases of the CrmD interaction with TNF and LT␣. In addition, we present the first empirical evidence for the CrmD TNF-binding domain boundaries. More importantly, our work proves that structural analyses of vTNFRs can be of great value to improve anti-TNF therapies based on soluble decoy receptors.
There is no consensus yet about the number of CRDs that constitute the TNF-binding domain of vTNFRs. For instance, CrmC and CrmE have been indistinctly represented in the literature with either four or three CRDs (9,11,14,26,35,36). It is assumed that CrmD and CrmB consist of an N-terminal four-CRD TNF-binding domain and a C-terminal SECRET domain (10,26). However, we have demonstrated here that the second half of the CRD4, beginning at an FN motif conserved in all SECRET-containing proteins, although dispensable for the CrmD anti-TNF activity, is essential for the highaffinity chemokine interactions by the SECRET domain. Consistent with this, the C-terminal region of the CRD4 of CrmD and CrmB contains two amino acids, Asp 167 and Glu 169 (Fig. 1A), that were identified as key chemokine-binding determinants in the structure of the SECRET-CX3CL1 complex (37). Therefore, we concluded that Phe 163 and the sequence at its C terminus belong to the SECRET domain. In which domain the first half of the CRD4 should be allocated remains to be elucidated. A previous analysis of T2 truncated mutants showed that the CRD4 was dispensable for the anti-TNF activity of this CrmB homolog (38). Therefore, the vestigial CRD4 of orthopoxvirus-encoded vTNFRs is unlikely to contribute to TNF binding. The function of the CRD4 in cellular TNFR1 and TNFR2 remains unclear. Its proximity to the cell membrane suggests that the CRD4 may facilitate the surface exposition of the ligand-binding domain and the transition to the intracellular signaling domain. This would explain why a complete CRD4 is unnecessary in soluble vTNFRs.

An etanercept mutant lacking anti-hLT␣ activity
Like the CRD4, the CRD1 is rarely directly involved in ligand binding. However, it has been shown that the CRD1 or PLAD mediates the ligand-independent oligomerization of some TNFRs, including TNFR1, TNFR2, CD40, Fas, and TRAIL receptors (21, 22, 39 -41). Whether the CRD1 may also induce self-assembly in vTNFRs remains unclear. It has been reported that the mutations K19A/Y20A and K32A disrupt the PLAD activity of TNFR1, resulting in mutant receptors unable to oligomerize and interact with TNF (21). In contrast, here we have shown that equivalent mutations in CrmD, D35A/Y36A and K44A, were only partially detrimental or completely innocuous, respectively, for the activity of this vTNFR. In addition, we confirmed that the protein oligomeric state was not altered in the D35A/Y36A mutant. This suggests either that a PLAD is missing from CrmD or that the molecular bases of a potential PLAD in CrmD differ from those of the TNFR1 PLAD. Of note, the CrmE crystallographic structure did not confirm the presence of a PLAD in this vTNFR (14). Interestingly, the authors showed that CrmE Tyr 28 , equivalent to CrmD Tyr 36 , established hydrophobic interactions with the 50s loop that might contribute to the correct folding of this important ligand-binding site. This is consistent with molecular dynamics analysis of TNFR1 that proposed that the CRD1 is essential to stabilize the CRD2 conformation (42), suggesting that the ligand-binding defects observed in TNFR PLAD mutants might be caused by a subtle destabilization of the CRD2. The same might explain the partial loss of activity in the CrmD D35A/Y36A mutant; however, a direct implication of these residues in ligand binding cannot be excluded, especially because equivalent amino acids were involved in the formation of the OX40 -OX40L complex (32). It is important to mention that none of this invalidates the possibility that vTNFRs may interfere with cellular TNFR signaling by engaging receptor monomers through their CRD1 as demonstrated for MYXV T2 (25). It would be interesting to study whether CrmD can also impede the formation of active TNFR oligomers and whether the CrmD D35A/Y36A mutant lacks this activity.
An efficient soluble decoy receptor must interact with the target cytokine but, more importantly, mask those residues of the ligand involved in the binding to the cognate cellular receptor. This might explain the lack of correlation that we observed between the binding and neutralizing activity of some mutants. CrmD residues contacting ligand amino acids not required for the interaction with the cellular receptor may contribute to the binding affinity but be irrelevant for the neutralizing activity of CrmD. Conversely, mutations of CrmD residues blocking key receptor-binding determinants of the ligand should be expected to be the most detrimental. In this regard, we have shown that mutations in a CRD2 groove under the 50s loop stripped CrmD of all its anti-TNFSF properties. This is consistent with the critical role assigned to this region in other TNFRs (18,20,22,29) where it is known to constitute a critical binding site for a hydrophobic patch of the ligand where a highly conserved tyrosine residue is the major binding determinant (16). Mutation of this tyrosine is sufficient to abolish or seriously damage the receptor-binding capacity of many TNFSF ligands, including TNF, LT␣, FasL, TRAIL, TL1A, and LIGHT (43)(44)(45)(46)(47)(48). Therefore, this tyrosine should be a main target for any inhibi-tor of these cytokines. The severe deleterious effect of mutations in the CrmD CRD2 groove and the fact that all four ligands analyzed here conserve this important tyrosine (mTNF, Tyr 165 ; hTNF, Tyr 163 ; mLT␣, Tyr 139 ; hLT␣, Tyr 142 ) suggest that this is the case for CrmD. This may also explain the incapacity of CrmD N77F to bind and neutralize these cytokines. The addition of a large aromatic residue at the rim of the CRD2 groove may constitute a steric hindrance that prevents a competent fitting of the ligand's tyrosine. It is important to note that at present we cannot prove whether these inactive CrmD mutants were properly folded. However, their high expression yield and efficient secretion and the facts that their antichemokine activity was intact and that the localization of these mutations agrees with that of important ligand-binding determinants in many other receptors suggest that this is likely to be the case.
Despite the common binding mechanism deployed by the CRD2 of many TNFRs, these receptors are very ligand-specific. The determinants for this specificity appear to reside mostly in the CRD3, which binds to more highly variable secondary structural features of the ligand (31). The CRD3-mediated interactions are essential for the stability of the receptor-ligand complex. For instance, a Fas chimera carrying the CRD3 of TNFR1 failed to interact with FasL (49), and a T2 truncated mutant containing only the first two CRDs was unable to bind TNF (38). Here, we have shown that several residues at the C terminus of the CRD3 of CrmD significantly contribute to the binding and inhibitory activity of this vTNFR. Of note, these amino acids are located outside of the 90s loop at the C terminus of the CRD3 of CrmD, a region rarely implicated in ligand binding in cellular TNFRs, which might hint at a different binding mechanism by vTNFRs. More importantly, we proved that an EFE motif at the 90s loop of the CRD3 prevents the anti-hLT␣ activity of CrmD, whereas it does not affect its interaction with mLT␣. In contrast, the 90s loop of the effective hLT␣ inhibitors CrmB and etanercept contains nonaromatic and less negatively charged motifs (LLK and ALS, respectively). To our knowledge, this is the first evidence that the CRD3 may contain not only ligand but also species specificity determinants.
Etanercept is the only Food and Drug Administrationapproved antirheumatic drug that blocks not only TNF but also LT␣. Although it has been proposed that this double inhibitory activity of etanercept could be beneficial in some instances (50), the role of LT␣ in inflammatory joint diseases is not well defined. Furthermore, a rheumatoid arthritis clinical trial concluded that an LT␣-neutralizing antibody, pateclizumab, was not clinically different from placebo (51). On the contrary, several reports have shown that LT␣ is important for host defense against Mycobacterium infections (52)(53)(54), which constitutes one of the major infectious concerns in patients under anti-TNF therapy (55). Importantly, the LT␣ activity is more markedly decisive in these infections in the absence of TNF (56, 57), which simulates the undergoing immunological condition in these patients. Therefore, although the systemic blockade of TNF, a central cytokine of the immune response, is probably largely responsible for the infection vulnerability observed in etanercept patients, the anti-LT␣ activity of this drug may only contribute to aggravate the infections. A more specific anti-

An etanercept mutant lacking anti-hLT␣ activity
TNF treatment can be achieved with any of the four approved TNF monoclonal antibodies. However, it has been reported that the risk of infection associated with these antibodies can be even higher than with etanercept (58). This might be explained by the potent cytotoxic activity, complement-dependent cytotoxicity or antibody-dependent cellular cytotoxicity, mediated by the Fc region of these antibodies, which may result in the destruction of TNF-bearing immune cells (6,59). This cytotoxicity is diminished or missing in etanercept due to a structural defect in its Fc region (6). Furthermore, etanercept is known to be minimally immunogenic, whereas patients frequently develop neutralizing antibodies against infliximab and adalimumab (60). All this together suggests that a TNF-specific etanercept would be the ideal antirheumatic drug. Here, we demonstrate that the anti-hLT␣ activity of etanercept can be vastly hampered by making its 90s loop look more like that of CrmD in a A127E/L128F etanercept mutant. The slight defect observed in the anti-hTNF activity of this mutant could potentially be overcome in the clinic by a small dose increase without compromising hLT␣-mediated immune functions. Therefore, this A127E/L128F variant could set the foundation for a safer second generation of etanercept featuring the benefits of a soluble decoy receptor and the high TNF specificity of the antibody therapy.
In conclusion, with the few exceptions discussed above, our results indicate that CrmD uses a similar molecular mechanism to bind and block TNF and LT␣. This complicates the generation of recombinant ECTV expressing CrmD mutants active only against one of its TNFSF ligands, which would be invaluable tools to dissect the functions of these cytokines in the antipoxvirus response. More detailed structural analyses of CrmD-TNF and CrmD-LT␣ complexes may reveal ligand-specific binding determinants not detected here. However, we have already proved this mutagenesis study useful for the generation of a single point ECTV mutant incapable of blocking mTNF and mLT␣ (27). More importantly, we demonstrated here that vTNFRs can provide a new perspective on TNFSF ligand-binding mechanisms and uncover novel specificity and binding determinants that would be overlooked in studies limited to cellular TNFRs. Given the strikingly different ligand-binding and specificity profiles across vTNFRs, structural studies on each of these viral decoy receptors have the potential to reveal new molecular determinants for a more potent and specific anti-TNF activity, which would refine our ability to neutralize harmful TNF responses in the clinic.

Cells and reagents
L929 (ATCC, Manassas, VA) were grown in Dulbecco's modified Eagle's medium supplemented with 10% FCS. MOLT-4 cells (ATCC) were cultured in RPMI 1640 medium supplemented with 10% FCS. Recombinant baculoviruses were generated and amplified in adherent Hi5 insect cells cultured in TC-100 medium supplemented with 10% FCS and 1ϫ nonessential amino acids. Suspension Hi5 cells maintained in Express Five (Life Technologies) medium supplemented with 8 mM L-glutamine were used for the expression of recombinant pro-tein. Recombinant cytokines were purchased from R&D Systems (Minneapolis, MN) and reconstituted and stored following the manufacturer's recommendations.
The plasmids described above were used to generate recombinant baculoviruses using the Bac-to-Bac system (Life Technologies) following the manufacturer's instructions. Subsequently, viral stocks were amplified by infecting adherent Hi5 cells at low multiplicity of infection (0.1-0.01 pfu/cell).

Protein expression and purification
Hi5 suspension cells were infected with the corresponding recombinant baculovirus at high multiplicity of infection (2 pfu/cell). Supernatants were harvested 3 days after infection, clarified at 6,000 ϫ g for 40 min, and then concentrated to 2.5 ml in a Stirred Ultrafiltration Cell 8200 (Millipore). The concentrate was desalted, and buffer was exchanged to 0.1 M phosphate buffer containing 300 mM NaCl and 10 mM imidazole using PD-10 desalting columns (GE Healthcare). His-tagged CrmD proteins were purified by metal chelate affinity chroma-An etanercept mutant lacking anti-hLT␣ activity tography using nickel-nitrilotriacetic acid columns (Qiagen, Germantown, MD). Etanercept proteins were purified using protein A-coupled Sepharose columns (Sigma). Protein-containing fractions were pooled, concentrated, and dialyzed in PBS. Final protein concentration was calculated by gel densitometry.

Cytotoxicity assays
The ability of CrmD, etanercept, and their mutants to inhibit TNFSF ligands was tested by cytotoxicity assays on L929 cells as described previously (12). Briefly, 20 ng/ml hTNF, mTNF, and hLT␣ were incubated for 1 h at 37°C in the presence of increasing amounts of recombinant protein. Due to its low specific activity, mLT␣ was used at 780 ng/ml. Subsequently, the cytokine-protein mixtures were added to L929 cells seeded at 12,000 cells/well in 96-well plates in the presence of 4 g/ml actinomycin D (Sigma). Cell viability was assessed after 18 h using a Cell Titer Aqueous One Solution kit (Promega, Madison,WI)followingthemanufacturer'sinstructions,andtheabsorbance at 490 nm (A 490 ) was determined in a Sunrise microplate reader (Tecan, Mannedorf, Switzerland). The A 490 of all samples was normalized with the A 490 of cells incubated only with the cytokine (0% viability). Cell viability for each sample was calculated in reference to the A 490 obtained in wells where cells were incubated without cytokine ("media"; 100% viability).

Chemotaxis assay
MOLT-4 cells chemotaxis assays were performed using 96-well ChemoTx plates of 3-m pore size (Neuro Probe, Gaithersburg, MD). Before the experiment, cells were washed and resuspended at 10 ϫ 10 6 cells/ml in 0.1% FCS in RPMI 1640 medium. Mouse Ccl25 (R&D Systems) at 70 nM was incubated with increasing molar ratios of recombinant protein at 37°C for 30 min in the bottom wells. Subsequently, 2.5 ϫ 10 5 cells were placed on the top filter, and cell migration was allowed for 4 h at 37°C. Afterward, the filter was flushed with PBS to remove nonmigrated cells, and migration to the lower wells was determined as the A 490 using a Cell Titter Aqueous One Solution kit (Promega). Cell migration was calculated in reference to the maximal migration obtained when the cells were incubated with the chemokine alone (100% cell migration).

SPR assays
The ligand-binding properties of recombinant proteins were characterized by SPR using a Biacore X biosensor (GE Healthcare). For binding assays, recombinant proteins were immobilized through their amine groups on CM4 chips at high density (1,000 -2,000 response units). One flow cell of the chip was treated with buffer alone during immobilization to be used as a reference. TNFSF cytokines and chemokines were injected at 100 nM in HBS-EP buffer (0.01 M HEPES pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% surfactant P20) (GE Healthcare) at 10 l/min for 3 min followed by a 2-min dissociation. The chip surface was regenerated between injections with 10 mM glycine HCl, pH 2.0.
For determination of kinetic affinity constants, recombinant proteins were immobilized on CM4 chips at low density (Ϸ500 response units). Increasing concentrations of TNFSF cytokines and chemokines were injected in HBS-EP buffer at 30 l/min for 2 min, and a 5-min dissociation was recorded. A 0.1-1000 nM concentration range of analyte was used. Between analyte injections, the chip surface was regenerated with 10 mM glycine HCl, pH 2.0. Kinetic data were globally fitted to a 1:1 Langmuir model using BIAevaluation 3.2 software. Bulk refractive index changes were removed by subtracting the responses recorded in the reference flow cell, and the response of a buffer injection was subtracted from all sensorgrams to remove systematic artifacts. To compare the binding affinity constants of CrmD truncated mutants, mean Ϯ S.D. K D of 10 fittings from two independent experiments was calculated. To determine the kinetic affinity constants of CrmD point mutants, the average K D of 10 fittings containing sensorgrams for at least six different analyte concentrations from at least two independent experiments was calculated. The fitting providing the closest K D to the average K D was chosen to represent the kinetic affinity constants of each interaction.