Production of Recombinant Human Trimeric CD137L (4-1BBL)

The interaction between 4-1BB ligand (CD137L), a member of the tumor necrosis factor superfamily, and its receptor 4-1BB provides a co-stimulatory signal for T lymphocyte proliferation and survival. However, the structure of 4-1BBL has not been thoroughly investigated, and none of the human recombinant 4-1BBL molecules available have been described as capable of co-stimulating T cells. The present work provides a model of the three-dimensional structure of the tumor necrosis factor homology domain of 4-1BBL and describes the production of a recombinant human soluble 4-1BBL whose originality lies in that it contains the whole extracellular tail preceding the tumor necrosis factor homology domain and an AviTag peptide (AviTag-4-1BBL) allowing enzymatic biotinylation and multimerization via streptavidin. We provide evidence that this chimeric protein exists as a homotrimer, whereas commercial FLAG-tagged 4-1BBL does not. This resulted in a much higher affinity for 4-1BB (1.2 nm) as compared with FLAG-4-1BBL (55.2 nm). We demonstrate that the single extracellular cysteine residue in the tail (Cys-51) could form a disulfide bond, both in our recombinant protein and in physiologically expressed 4-1BBL. The mutation of this cysteine residue exerted no effect on trimerization but increased the dissociation rate of AviTag-4-1BBL from 4-1BB. In its soluble form, AviTag-4-1BBL did not stimulate purified T cells but dramatically inhibited proliferation of peripheral blood mononuclear cells stimulated with anti-CD3 mAb. In contrast, a very significant co-stimulatory effect was observed on purified T cells once AviTag-4-1BBL was immobilized onto streptavidin beads. In addition, we show that the cross-linking of two trimeric AviTag-4-1BBL molecules was the minimum step required to elicit significant costimulatory activity.

importance of 4-1BB/4-1BBL interactions in anti-tumor immunity by showing that the administration of agonistic anti-4-1BB to the mouse resulted in dramatic tumor regressions, even in weakly immunogenic tumors.
Considering the growing interest in studying 4-1BB stimulation in T cell expansion, activation, and survival, it is surprising that only two reports have described the use of recombinant 4-1BBL (9,10) and that it was mouse 4-1BBL that significantly differed from human 4-1BBL (only 36% homology at the protein level) in both cases. In fact, most of the in vitro and in vivo studies have been performed either with a stimulatory anti-4-1BB mAb (8,11,12) or with cells transfected with 4-1BBL cDNA (13)(14)(15)). In addition, despite different recombinant human 4-1BBLs being commercially available today (Alexis, Chemicon, Ancell), they are poorly characterized from a biochemical point of view, and none of them are described as being able to co-stimulate T lymphocytes. As a result, little evidence has been available to date in the literature concerning the structure of 4-1BBL and its organization necessary to achieve T cell co-stimulation. For this reason, we set out to produce a soluble form of human 4-1BBL to determine its structure, its binding characteristics toward 4-1BB and to evaluate its functional activity.

MATERIALS AND METHODS
Sequence Alignment and Molecular Modeling-Sequences were obtained from the Swiss-Prot data base. Sequence alignment of selected templates was firstly deduced from a pairwise structure alignment automatically generated by the CE (16) and FSSP (17) programs. The 4-1BBL sequence was then manually aligned with the template multi-alignment according to conserved amino acids and predicted ␤-strand structure. Homology modeling was performed with the Modeler module of InsightII (18) using the x-ray crystallographic structure of mouse TNF␣, 2 mouse RANKL, human CD40L, and human APO2L (Protein Data Bank codes 2TNF, 1IQA, 1ALY and 1DG6 respectively). Loop refinement of selected models was performed with the refined loop Modeler program. An evaluation of the generated models was performed using the Verify-3D (19), ProsaII (20), and Procheck (21) programs. Energy minimization was then carried out with the CHARMm module of Insight and CHARMm forcefield (22) in a four-step procedure (1, all atoms fixed except hydrogen; 2, backbone fixed; 3, backbone of strand structure fixed; 4, all atoms free) using the 100 steepest descent steps followed by conjugate gradient steps, until a convergence gradient of 0.001 was attained. The energy scores provided by ProsaII and the global compatibility score given by Verify-3D were in accordance with those of the templates. Furthermore, the stereochemical quality of the selected predicted model, as evaluated by Procheck, compares favorably with the template. The Ramachandran plot showed 97.7% residues in allowed regions. The protein molecular surface of the selected model was defined using a solvent probe radius of 1.4 Å and continuum elec-trostatic calculations were performed with the DELPHI package (23,24) under Insight. The formal charge set was used, with an ionic strength of 0, and the dielectric constants of the protein and the surrounding medium were, respectively, 2 and 80. In addition, the electrostatic potentials were mapped onto cubic grids with a 0.72-Å point spacing. The percentage grid fill was 50% for 4-1BBL (129 ϫ 129 ϫ 129 points/ slide), and the boundary potential was full coulombic. Following this, the linear Poisson-Boltzmann equation was solved iteratively, and the Engelman-Steitz scale was used to display protein surface hydrophobicity (25).
Production of AviTag-4-1BBL in the Insect Cell Line S2-To allow secretion of AviTag-4-1BBL protein by S2 Drosophila cells, the cDNA coding for the IL6 signal peptide was added to the 5Ј-end of the bacterial construct, and the whole construct was subcloned in the pRmHa-3 expression vector which carries a copper-inducible promoter (a kind gift from Dr. Goldstein). S2 cells were co-transfected with the pRmHa-3 and pCoblast, which carries blasticidin resistance using the CaCl 2 method and selected with blasticidin over a period of 3 weeks (25 g/ml, Invitrogen). The bulk culture of blasticidin-resistant cells was cloned by limiting dilution using non-transfected irradiated S2 Drosophila cells as a feeder. Clones were assayed for AviTag-4-1BBL production by intracytoplasmic staining with an anti-4-1BBL monoclonal Ab (BD Biosciences). Clone 1 was grown to a concentration of 5.10 6 /ml in a serumfree culture medium (Invitrogen), and protein production was induced by the addition of 0.75 mM CuSO 4 for 5 days at 27°C. The supernatant was concentrated on a VivaFlow concentrator cell, separated on a DEAE column, and biotinylated as described above. The protein was finally purified by anion exchange chromatography on a MonoQ HR 10/10 column (Amersham Pharmacia Biotech). FLAG-4-1BBL was purchased from Alexis.
In certain cases, proteins were cross-linked prior to Western blotting using the non-reversible, homobifunctional, and water-soluble crosslinking agent BS 3 (Pierce). BS 3 (10 mM in water) was then added to 40 pmol of protein, to give the required final concentration (125-500 M) and cross-linking was allowed to take place for 1 h at room temperature. The reaction was quenched with 1 M Tris, pH 7.5.
For the analysis of native 4-1BBL, membrane proteins were extracted from T2 cells using Triton X-114 as previously described (26). Briefly, 60ϫ10 6 T2 cells were lysed with 1 ml of lysis buffer (10 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-114 (Sigma) and a protease inhibitor mixture (Complete, Boehringer) for 30 min on ice. The lysate was centrifuged at 900 ϫ g for 10 min to remove cell debris and nuclei, and overlaid on an ice-cold sucrose cushion (6% w/v sucrose, 10 mM Tris, pH 7.4, 150 mM NaCl, 0.06% Triton X-114). A clouding of the solution occurred after a 3-min incubation at 30°C. The tube was then centrifuged for 3 min at 300 ϫ g at room temperature to recover the detergent phase as an oily droplet from the bottom of the tube. The detergent phase enriched in membrane proteins was precipitated by MEOH/CHCl 3 4:1 (v/v) before loading onto a 12% SDS-PAGE and immunoblotting as described above.
Surface Plasmon Resonance Analyses-Binding experiments of the different recombinant 4-1BBL proteins to a recombinant human 4-1BB-Fc chimera (R&D Systems) were performed with a BIAcore 2000 optical biosensor (BIAcore AB, Uppsala, Sweden).The human 4-1BB/Fc protein was covalently coupled to a carboxymethyl dextran flow cell (CM5 BIAcore) as recommended by the manufacturer. The level of immobilization was set at 500 resonance units. The binding of purified mutant and native 4-1BBL was assayed at 25°C with concentrations ranging from 0.725 to 72.46 nM at a flow rate of 40 l/min in HBS-EP buffer (0.01 M HEPES, pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% surfactant P20). Association was monitored for 5 min before initiating the dissociation phase for another 10 min with HBS-EP buffer. The flow cells were regenerated by a 1-min pulse with 10 mM glycine-HCl at pH 1.8. The resulting sensorgrams were analyzed using the BIA Evaluation software 3.2 (BIAcore AB).
Multimerization of the Different Recombinant 4-1BBLs-AviTag-4-1BBL (wild type (WT) and C51S) was coated onto M280 streptavidin magnetic beads by incubating decreasing concentrations of biotinylated proteins (ranging from 10 g/ml to 5 ng/ml) with 67.10 3 M280 magnetic microbeads (Dynal Biotech) for 2 h at room temperature under continuous agitation. The FLAG-4-1BBL was coated in the same conditions using M280 tosylactivated magnetic beads covalently linked to an anti-FLAG mAb (clone M2, Sigma). The saturation of the beads was checked by flow cytometry using either a mouse monoclonal PE-conjugated anti-4-1BBL (clone C65-485, BD Biosciences) or a goat anti-4-1BBL polyclonal Ab followed by an adsorbed fluorescein isothiocyanate-conjugated donkey anti-goat Ab as a secondary reagent (Serotec). Saturation, estimated by the fluorescence plateau, was obtained with 0.5 pg/bead of AviTag-4-1BBL and 2 pg/bead of FLAG-4-1BBL.
Tetramerization with streptavidin is usually attained by mixing streptavidin (Sigma) and AviTag-4-1BBL at a molecular ratio of 1:4 for 1 h at room temperature. To obtain a mixture of the different multimers (tetramers, trimers, dimers, and monomers) AviTag-4-1BBL was incubated with a molar excess of streptavidin, and the different forms were then separated by gel filtration.
Cell Isolation and Functional Assays-Human peripheral blood mononuclear cells (PBMC) from healthy donors were isolated by Ficoll-Hypaque gradient centrifugation and resuspended in RPMI medium (Sigma) supplemented with 1% L-glutamine and 8% human serum. Purified T cells were obtained from fresh blood by negative selection using a RosetteSep isolation kit (StemCell Technologies). The purity of the T lymphocyte preparation was checked by flow cytometry with a fluorescein isothiocyanate-conjugated anti-CD3 mAb (BD Biosciences) and was always above 95%. T lymphocytes (10 5 /well, flat bottom 96-well plates, Falcon) were stimulated for 72 h with biotinylated anti-CD3 mAb (clone UCHT1, BD Biosciences) immobilized on M280 streptavidin magnetic beads. OKT3 (Orthoclone) immobilized in flat bottom 96-well plates was used to stimulate PBMC (10 5 /well) overnight at 4°C in phosphate-buffered saline instead of anti-CD3-coated beads, because macrophages tended to engulf the beads. All 4-1BB ligands were added after 6 h of CD3 stimulation to allow the induction of 4-1BB expression on T lymphocytes. Proliferation was estimated by measuring the incorporation of tritiated thymidine in quadruplicate samples during the last 16 h of culture.
Statistical Analysis-Results are expressed as mean Ϯ S.E. Results were analyzed by analysis of variance followed by either Tukey-Kramer or Dunnett post-tests. Post-tests were only performed when the analysis of variance test showed a significant difference (p Ͻ 0.05) between groups.

Modeling of the Three-dimensional Structure of 4-1BBL-4-1BBL
belongs to the TNF superfamily and, as such, is presumed to be trimeric like the other members of the family (27), although no information on its three-dimensional structure has yet been provided. We thus decided to generate a model of 4-1BBL structure. We firstly performed a sequence alignment of the TNF-homology domain (THD) of human 4-1BBL with that of mouse TNF, mouse RANKL, human CD40L and human APO2L (Fig. 1A). The THD of 4-1BBL displayed a 17.6, 21.6, 17.8, and 19.5% sequence homology with TNF, RANKL, CD40L, and APO2L, respectively. With this alignment and based on x-ray structures of mouse TNF, mouse RANKL, human CD40L, and human APO2L, a model of the THD of human 4-1BBL was generated (see "Materials and Methods"). A ribbon representation of the final model is given in Fig. 1B, showing two stacked ␤-pleated sheets each made up of five anti-parallel ␤-strands that adopt the classical "jelly-roll" topology found in proteins belonging to the TNF family (27). The inner sheet (strands A, AЈ, H, C, and F) may be involved in trimer contacts (Fig. 1B, left) and the outer sheet (strands B, BЈ, D, E, and G) may be exposed to the solvent (27). An analysis of hydrophobicity revealed a large hydrophobic area containing residues constituting the sheet potentially involved in trimer formation: Phe-7 and Val-11 (strand A), Val-149, Phe-153, Val-155, and Ile-159 (strand H), Val-55 and Phe-59 (strand C), Phe-112, Phe-114, Leu-118, and Leu-119 (strand F) (Fig. 1B, middle). In addition, charged residues such as Arg-65, Glu-63, and Asp-99 located close to the hydrophobic area may also contribute to trimer formation (Fig. 1B, right). A previous crystallographic analysis of the LT␣/TNFR1 complex showed that loops of the trimeric ligand are particularly involved in contact with the receptor, especially loops AAЈ, DE, EF, and CD (28). In our model, the low number of conserved residues found in those loops of 4-1BBL and the high flexibility of these regions support their implication in defining the specificity of ligand-receptor interaction.
The Extracellular Domain of 4-1BBL Forms a Trimer Containing a Disulfide Bond-To produce recombinant soluble 4-1BBL, we engineered a cDNA construct coding for the whole extracellular domain of human 4-1BBL coupled to a biotin tag, the AviTag. This construct included the 43-amino-acid long tail spanning from the membrane to the THD and thus differed from a commercially available FLAG-tagged 4-1BBL (Alexis) (Fig. 1C). It should be noted that this tail contains a cysteine residue close to the membrane.
We first produced the AviTag-4-1BBL protein with a theoretical mass of 23 kDa as inclusion bodies in E. coli. The purity of inclusion bodies exceeded 95%, as evaluated by SDS-PAGE and Coomassie staining ( Fig. 2A). After refolding and concentration, gel filtration analysis revealed two major peaks, Peaks 1 and 2, with a molecular mass of around 140 and 70 kDa, respectively (Fig. 2B). Peak 2 was compatible with a trimeric form of AviTag-4-1BBL. SDS-PAGE electrophoresis followed by Western blotting of Peak 2 showed that the trimeric form dissociated into a monomer and a dimer. The homodimer contained a disulfide bond as demonstrated by its disappearance under reducing conditions (Fig. 3A). An analysis of Peak 1 by Western blot under reducing and non-reducing conditions showed an identical profile (not shown), suggesting that it corresponded to a multimeric complex of AviTag-4-1BBL linked by disulfide bonds. This form was discarded, as it was considered to be wrongly folded, because it turned out to be totally inactive in the functional assays described later. As expected, FLAG-4-1BBL appeared only as a monomer on SDS-PAGE because it lacks the cysteine-containing region (Fig. 3A).
We next sought to determine whether the disulfide bond present in the trimeric form of AviTag-4-1BBL could have been artificially generated during the refolding of bacterial inclusion bodies. For this purpose, soluble AviTag-4-1BBL was produced in the Drosophila S2 expression system. In this system, the expressed protein is naturally processed inside the cell and then excreted into the culture medium. The analysis of AviTag-4-1BBL produced in S2 supernatants revealed a similar profile as that obtained with refolded inclusion bodies (Fig. 3B). This suggested that the disulfide bond was naturally formed in eucaryotic cells during AviTag-4-1BBL processing and export. To ascertain whether the disulfide bond was actually present in physiologically expressed 4-1BBL, Western blot analyses were performed on Triton X-114 extracts from T2 hybridoma cells that constitutively express 4-1BBL (Fig. 3C). A similar profile was obtained, with a dimeric form that could be reduced by 2-mercaptoethanol (Fig. 3D), thus pointing to the physiological nature of the disulfide bond.
Finally, we performed cross-linking experiments with BS 3 on refolded AviTag-4-1BBL and FLAG-4-1BBL to visualize trimers by SDS-PAGE. Cross-linking of AviTag-4-1BBL with BS 3 resulted in the appearance of bands of higher molecular mass, of which one could correspond to a trimer (around 70 kDa) and another (around 140 kDa) may represent dimers of trimers (Fig. 3E). This observation further demonstrated that AviTag-4-1BBL refolded as a trimer. In marked contrast, cross-linking of FLAG-4-1BBL revealed no trimeric forms but only a band compatible with dimers and a strong band corresponding to large aggregates that did not penetrate into the gel (Fig. 3F). Considering that the dimeric forms probably originated from the dissociation by SDS of pre-existing and partially cross-linked trimers, our data suggest that only a very small fraction of FLAG-4-1BBL was in trimeric form (and thus undetectable as such after BS 3 cross-linking), whereas most of it was monomeric.
The Disulfide Bond Stabilizes Trimeric AviTag-4-1BBL-To investigate the role of the disulfide bond in 4-1BBL structure and function, we produced a mutated AviTag-4-1BBL where cysteine was replaced by serine (C51S). Because the cysteine residue is located well outside the THD of 4-1BBL, it was anticipated that the absence of the disulfide bond would not affect trimerization but may result in a less stable trimer. As a matter of fact, C51S AviTag-4-1BBL behaved as a trimer under gel filtration after refolding and the higher molecular weight form (Peak 1) Middle and right, the Connolly surface of modeled 4-1BBL is colored with either the hydrophobicity index of the exposed residues (hydrophobic in red, hydrophilic in yellow) (middle) or with the electrostatic potential (right). C, schematic representation of the different human 4-1BBL proteins studied. The AQL sequence corresponds to the beginning of the TNF homology domain. In the AviTag-4-1BBL protein, the entire extracellular domain of human 4-1BBL (amino acids 49 -254) was coupled to a biotin tag (AviTag), whereas in FLAG-4-1BBL, the FLAG peptide (8 amino acids) was coupled to the THD through an undefined 11-amino-acid linker. DECEMBER 16, 2005 • VOLUME 280 • NUMBER 50 previously seen with the WT AviTag-4-1BBL could not be detected (Fig.  4A). Western blot analysis after SDS-PAGE showed that under nonreducing conditions, the trimer dissociated into monomers only with no dimers being present, as expected (Fig. 4B). Cross-linking with BS 3 on the other hand revealed both dimeric and trimeric forms of C51S Avi-Tag-4-1BBL (Fig. 4C). It was therefore concluded that the disulfide bond is not required for AviTag-4-1BBL refolding as a trimer.

Soluble Human 4-1BBL Is Co-stimulatory Only When Cross-linked
Using surface plasmon resonance, we proceeded to determine the kinetic constants of binding of native or mutated AviTag-4-1BBL and FLAG-4-1BBL to immobilized 4-1BB-Fc. The sensorgrams obtained with a range of concentrations (0.1-5 g/ml) of each of the three proteins are shown in Fig. 5. The first striking feature observed was that FLAG-4-1BBL displayed a much lower binding affinity to 4-1BB than native or mutated AviTag-4-1BBL (K d of 55.2 nM for FLAG-4-1BBL versus 1.2 nM for WT AviTag-4-1BBL and 2.3 nM for its C51S variant). This lower affinity mainly resulted from a 30-fold lower association rate (k on ) for FLAG-4-1BBL as compared with WT AviTag-4-1BBL (6.7 ϫ 10 3 versus 2.1 ϫ 10 5 M Ϫ1 s Ϫ1 , respectively), whereas the dissociation AviTag-4-1BBL appeared as a major band around 23 kDa, the expected molecular mass of the monomer. B, gel filtration analysis of 30 g of refolded AviTag-4-1BBL on a S200 column. The first peak corresponded to excluded aggregates. K av of the peaks are shown, and the calibration curve is shown in the inset. K av 2 corresponded to a molecular mass around 70 kDa, compatible with trimeric 4-1BBL, whereas K av 1, estimated at around 140 kDa, may represent dimers of trimers. constants (k off ) were comparable (2.6 ϫ 10 Ϫ4 s Ϫ1 for WT AviTag-4-1BBL versus 3.7 ϫ 10 Ϫ4 s Ϫ1 for FLAG-4-1BBL). Given that unlike the dissociation constant, the association rate depends on the concentration of ligand, the most likely interpretation of the data is that only a small fraction of FLAG-4-1BBL had the proper trimeric conformation.

Trimeric 4-1BBL Co-stimulates T Lymphocytes Only When
Cross-linked-To investigate the functional properties of AviTag and FLAG-4-1BBL, we firstly checked their ability to recognize 4-1BB on CD3-stimulated T lymphocytes. Flow cytometry analyses revealed that biotinylated AviTag-4-1BBL stained a similar percentage of 4-1BB positive activated T cells (60%) as the anti-4-1BB mAb used as a positive control (64%), despite having a lower mean fluorescence intensity (Fig.  6). In contrast, staining with FLAG-4-1BBL was too low to conclusively identify all positive cells.
We then tested the ability of the two proteins in soluble form to co-stimulate the proliferation of either PBMC or purified T cells activated through immobilized anti-CD3 mAb. The major difference between these two experimental designs is that immobilized anti-CD3 mAb alone is sufficient to elicit a strong proliferation of PBMC because   DECEMBER 16, 2005 • VOLUME 280 • NUMBER 50 of the presence of other cell types, such as macrophages and B cells, that can provide co-stimulation signals to T cells, whereas it is much less efficient at stimulating purified T lymphocytes proliferation. Soluble AviTag-4-1BBL dose dependently inhibited the proliferation of CD3stimulated PBMC with a maximal inhibition of ϳ57% at 10 g/ml (Fig.  7A) but had no effect on CD3-stimulated T lymphocytes (Fig. 7B). Considered together, these data strongly suggested that the binding of soluble AviTag-4-1BBL to 4-1BB on T lymphocytes did not activate the co-stimulatory pathways but instead efficiently prevented 4-1BB/4-1BBL interactions between T lymphocytes and APCs. Likewise, soluble C51S AviTag-4-1BBL inhibited CD3-induced PBMC proliferation, although less efficiently than WT AviTag-4-1BBL (33.0 Ϯ 9.6% inhibition for C51S versus 57 Ϯ 5.8% inhibition for WT at 10 g/ml, p Ͻ 0.05) (Fig. 7A). This finding further documented that the presence of the disulfide bond increased the stability of the trimeric form in solution. Finally, FLAG-4-1BBL was the least efficient to inhibit proliferation in accordance with its poor binding affinity for 4-1BB (Fig. 7A).

Soluble Human 4-1BBL Is Co-stimulatory Only When Cross-linked
Because AviTag-4-1BBL showed no co-stimulatory activity when used in soluble form, it was decided to test its effect on T cell proliferation following immobilization on a matrix. For this purpose, streptavidin beads were coated with biotinylated AviTag-4-1BBL (either WT or C51S) or anti-FLAG mAb-coupled beads with FLAG-4-1BBL, and saturation of the beads was controlled by immunofluorescence with an anti-4-1BBL mAb. Interestingly, similar intensities of staining could be observed with beads coated with WT or C51S AviTag-4-1BBL, but no staining was observed with beads coated with FLAG-4-1BBL (Fig. 8A). The use of a polyclonal anti-4-1BBL antibody allowed us to ascertain whether FLAG-4-1BBL was indeed coated onto the beads. A test was then performed on these beads coated with different forms of 4-1BBL to assess their ability to co-stimulate proliferation of purified T cells. As shown in Fig. 8B, the addition of beads coated with WT or C51S Avi-Tag-4-1BBL resulted in a very significant increase in proliferation of CD3-stimulated T cells (4793 Ϯ 426 cpm for CD3 alone versus 14335 Ϯ 1239 cpm for CD3 ϩ WT AviTag-4-1BBL, n ϭ 10, p Ͻ 0.001), whereas no stimulatory effect was observed with FLAG-4-1BBL-coated beads. Similar results were obtained with beads coated with WT AviTag-4-1BBL produced in S2 insect cells (data not shown). This demonstrated that, once AviTag-4-1BBL was cross-linked on beads, it could stimulate T cell proliferation, which was not the case with FLAG-4-1BBL that remained non-stimulatory after cross-linking. In addition, it should be noted that no significant difference in stimulatory activities were seen between beads coated with WT or C51S AviTag-4-1BBL. This suggest either that cross-linking stabilized trimeric C51S AviTag-4-1BBL or that the differences in activity were overwhelmed by the amount of stimulatory signals.
We finally performed cross-linking of WT AviTag-4-1BBL with a molar excess of streptavidin to obtain all multimeric forms, from tetramers to monomers (Fig. 9A). Despite no clear cut separation of the peaks by gel filtration, narrow fractions corresponding to each peak were tested on CD3-stimulated T lymphocytes. It was observed that high complexes (tetramers and trimers) co-stimulated proliferation very efficiently and that dimers were slightly less active, whereas monomers were inactive (Fig. 9B). This confirmed the critical requirement of cross-linking to obtain a co-stimulatory effect of AviTag-4-1BBL on T cell proliferation.

DISCUSSION
We report the production, and biochemical and functional properties of a soluble form of the human 4-1BBL extracellular domain coupled to a biotinylation peptide, the AviTag. This protein was designed to include not only the THD of 4-1BBL but also the 43-amino-acid long tail. A model of the three-dimensional structure of the THD was generated that is compatible with a trimeric organization of 4-1BBL. We provide evidence by gel filtration analysis and cross-linking experiments that our chimeric protein AviTag-4-1BBL, whether produced in bacteria or in insect cells, could indeed refold as a trimer. In contrast, our data suggest that the commercial FLAG-4-1BBL, produced in HEK293 cells, was mainly present as a monomer in solution because no trimeric form and only a weak dimeric form could be detected after cross-linking with BS 3 . The hypothesis that FLAG-4-1BBL has a different conformation to AviTag-4-1BBL is further supported by several findings. First, our binding experiments showed that FLAG-4-1BBL displayed a much lower affinity for 4-1BB than AviTag-4-1BBL. Second, our staining experiments demonstrated that a monoclonal antibody against cell surfaceexpressed 4-1BBL could recognize AviTag-4-1BBL-coated beads but not FLAG-4-1BBL-coated beads (although we cannot rule out the possibility that this antibody recognized an epitope in the tail of 4-1BBL, a tail that is absent in FLAG-4-1BBL). Finally, we observed that monomeric FLAG-4-1BBL was soluble in aqueous solution, whereas monomeric AviTag-4-1BBL was not. An unsuccessful attempt was made to purify soluble monomeric AviTag-4-1BBL, in particular by denaturing trimers with heat or mild detergents, but no monomers were ever recovered due to aggregation (not shown). We surmise that the FLAG peptide, which is very hydrophilic, may actually hinder the trimerization of FLAG-4-1BBL while contributing to the solubility of the monomeric FLAG-4-1BBL. In support of such a supposition, a recent publication reported that short tags can have a major impact on protein conformation and crystallization (29). In our construct, the AviTag was placed at the end of the 43-amino-acid long tail, a position that apparently prevented the tag from interfering with the trimerization process.
In addition, we report that cysteine 51 located in the extracellular tail  . A, S00 gel filtration analysis of oligomers of WT AviTag-4-1BBL. Trimeric AviTag-4-1BBL was mixed with a molar excess of streptavidin to obtain all degrees of multimerization. Estimated molecular weight based on elution volume are indicated over each peak. The theoretical molecular weight of the different oligomers are mentioned in the inset. B, multimeric complexes (tetrameric, trimeric, and dimeric) efficiently co-stimulated purified T lymphocyte proliferation. A narrow fraction of each peak represented in A was added to purified T lymphocytes activated for 6 h with biotinylated anti-CD3 mAb coated beads. Results are expressed as mean Ϯ S.E. of three experiments. **, p Ͻ 0.01 compared with CD3 alone. DECEMBER 16, 2005 • VOLUME 280 • NUMBER 50 region could form an interchain disulfide bond, and we provide evidence that this disulfide bond was naturally present in 4-1BBL expressed on T2 cells. Using BIAcore analysis it is demonstrated that the abrogation of the disulfide bond by mutation of the cysteine resulted in a 2-fold higher dissociation rate of the AviTag-4-1BBL⅐4-1BB complex. The absence of the disulfide bond also resulted in a diminished ability of the mutant AviTag-4-1BBL to inhibit 4-1BBL⅐4-1BB interactions during CD3-induced proliferation of PBMC. This suggests that the disulfide bond is important for the stabilization of the trimeric form of Avi-Tag-4-1BBL and that it may also play a role in stabilizing physiologically expressed 4-1BBL. Moreover, the presence of a single disulfide bond within trimeric 4-1BBL leaves one cysteine free to establish an additional disulfide bond with an adjacent 4-1BBL trimer, leading to a higher degree of oligomerization of 4-1BBL at the cell surface. We did detect higher molecular weight forms after refolding of 4-1BBL (Fig. 2) and BS 3 cross-linking (Fig. 3E) that may represent dimers of trimers, but their presence at the cell surface needs to be confirmed. In our experience these higher molecular weight forms were non-stimulatory (not shown) but this may have been because of an incorrect conformation of these forms in solution. In line with these observations, it would be useful to investigate whether an interchain disulfide bond is present in other TNF ligand family members such as CD40L, Apo-2L, or RANKL that also have cysteine residues in their tail region.

Soluble Human 4-1BBL Is Co-stimulatory Only When Cross-linked
Most of the TNF ligand family members are synthesized as membrane-bound proteins, but soluble forms can be generated by limited proteolysis. Although some of the proteases involved in this process have been identified, such as ADAM proteases for TNF and RANKL (30,31), matrilysin for FasL (32) or furins for BAFF, APRIL, TWEAK, and EDA (33,34), others still remain unknown. Whether the soluble forms thus generated are biologically active depends on which ligand is considered and is still a matter of debate in some instances. For example, TNF␣ and lymphotoxin-␣ are undeniably active in soluble forms (35,36), whereas despite earlier reports (37,38) it is now agreed that FasL loses most of its apoptotic-inducing capacity when released in soluble form (39,40). Similarly, soluble CD40L was initially reported to be as active as the membrane-bound form in stimulating B cell proliferation (41), but a recent study provided conclusive evidence that limited oligomerization is necessary for both FasL and CD40L activity (42). Accordingly, the B cell-stimulating activity of the commercial FLAG-CD40L (Alexis) is described by the manufacturer as greatly enhanced by cross-linking with an anti-FLAG mAb. Our own experience with FLAG-CD40L, used to induce dendritic cell maturation, confirmed that cross-linking with anti-FLAG mAb was necessary for optimal activity (43). The data that we present in this report about 4-1BBL are in agreement with those concerning CD40L and FasL. In fact, we demonstrated that soluble trimeric AviTag-4-1BBL could bind recombinant 4-1BB with a high affinity (1.2 nM for WT and 2.3 nM for C51S AviTag-4-1BBL) but that this binding has no co-stimulating effect on CD3-stimulated T lymphocytes. Nonetheless, this binding efficiently prevented 4-1BBL⅐4-1BB interactions among CD3-stimulated PBMC and thus significantly inhibited proliferation. These observations are at odds with a previous publication reporting that soluble 4-1BBL released by certain B cell lines such as the Raji cell line could co-stimulate CD3-activated T lymphocytes (44). However, in this latter report, the authors used concentrated crude supernatants from Raji cells as a source of soluble 4-1BBL, and it is therefore difficult to rule out the presence of aggregates and/or membrane fragments or vesicles that may have achieved sufficient 4-1BBL cross-linking to activate T cells. In fact, we demonstrated that AviTag-4-1BBL trimers became highly efficient in co-stimulating T lymphocyte proliferation once they were cross-linked either on beads or by streptavidin. The requirement for cross-linking to obtain T cell co-stimulating activity of recombinant soluble 4-1BBL has already been suggested in the mouse (9). Our data also showed that two cross-linked AviTag-4-1BBL trimers were sufficient to trigger co-stimulatory signals, although a higher degree of oligomerization led to enhanced co-stimulation (Fig.  9). These data tally with those previously reported with FasL and CD40L showing that two trimeric FasL or two trimeric CD40L were sufficient to trigger apoptosis or B cell proliferation, respectively, whereas a higher degree of multimerization further enhanced CD40L stimulation (but not FasL) (42).
In conclusion, we have produced a soluble human 4-1BBL that should be a versatile and useful tool to study 4-1BBL⅐4-1BB interactions in T cell activation, because it contains the proper trimeric conformation, has a high affinity for 4-1BB, and can be used either to mimic or block these interactions.