PREPARATION AND CHARACTERIZATION OF SOLUBLE RECOMBINANT HETEROTRIMERIC COMPLEXES OF HUMAN LYMPHOTOXINS ALPHA AND BETA

Abstract The lymphotoxin (LT) protein complex is a heteromer of α (LT-α, also called tumor necrosis factor (TNF)-β) and β (LT-β) chains anchored to the membrane surface by the transmembrane domain of the LT-β portion. Both proteins belong to the TNF family of ligands and receptors that regulate aspects of the immune and inflammatory systems. The LT complex is found on activated lymphocytes and binds to the lymphotoxin-β receptor, which is generally present on nonlymphoid cells. The signaling function of this receptor-ligand pair is not precisely known but is believed to be involved in the development of the peripheral lymphoid organs. To analyze the properties of this complex, a soluble, biologically active form of the surface complex was desired. The LT-β molecule was engineered into a secreted form and co-expressed with LT-α using baculovirus/insect cell technology. By exploiting receptor affinity columns, the LT-α3, LT-α2/β1, and LT-α1/β2 forms were purified. All three molecules were trimers, and their biochemical properties are described. The level of LT-α3-like components in the LT-α1/β2 preparation was found to be 0.02% by following the activity of the preparation in a WEHI 164 cytotoxicity assay. LT-α3 with an asparagine 50 mutation (D50N) cannot bind the TNF receptors. Heteromeric LT complexes were prepared with this mutant LT-α form, allowing a precise delineation of the extent of biological activity mediated by the TNF receptors. A LT-α3 based cytotoxic activity was used to show that the LT-α1/β2 form cannot readily scramble into a mixture of forms following various treatments and storage periods. This biochemical characterization of the LT heteromeric ligands and the demonstration of their stability provides a solid foundation for both biological studies and an analysis of the specificity of the LT-β and TNF receptors for the various LT forms.

Recently, a family of receptors and ligands having structural homology to TNF 1 and its receptors was elucidated at the molecular level (1,2). This family includes the TNF, LT-␣, LT-␤, Fas, CD27, CD30, CD40, OX-40, 4 -1BB, and NGF systems. TNF and LT-␣ were the first members of this family to be defined, and because they exhibited similar biological activity and receptor binding properties, LT-␣ was originally believed to be functionally identical to TNF. Surprisingly, biochemical characterization of a surface form of LT-␣ showed that it was a complex of two proteins (3,4). Surface LT-␣ is distinguished from surface TNF because LT-␣ does not retain the transmembrane region but rather is tethered to the surface via complexation with a related gene called LT-␤ (5,6). In addition to differences in expression patterns, this observation was one of the first indications that a separate biological role for LT may exist (7). A separate receptor specifically recognizing surface LT was identified validating the hypothesis that the LT system was not simply a redundant aspect of TNF function (8). Furthermore, disruption of the LT-␣ gene in mice leads to loss of lymph node development further reinforcing the concept of unique biological functions for the LT pathway distinct from those of TNF (9,10).
Each of the ligands in the TNF family has now been shown to possess a corresponding unique receptor. The TNF system is complex because there are two receptors, i.e. the 55-60-and 75-80-kDa forms (referred to here as the TNF-R55 and TNF-R75 but also called TNF-R1 and TNF-R2, respectively) recognizing both LT-␣ and TNF (11). The interactions of these receptors has been studied and both TNF-Rs can bind to a minor form of surface LT that has an apparent LT-␣2/␤1 composition (8). The primary surface LT form, which is expected to be LT-␣1/␤2, binds to a new receptor termed the LT-␤ receptor (LT-␤-R) (8). The LT-␤-R is structurally related to both the TNF-R55 and TNF-R75 yet does not recognize either soluble LT-␣ or TNF. Signalling in this family is believed to occur when the trivalent ligand is able to bring together two or more receptors. The oligomerization of receptor extracellular domains presumably alters the arrangement of the intracellular domains, which is in turn translated into some further signal (12). The major question that becomes apparent upon consideration of the LT system is how does receptor signal transduction occur? In the case of LT-␣1/␤2 on a cell surface, the three receptor binding clefts are not equivalent and therefore the delineation of which receptors bind to which clefts will help to explain how signaling can occur. A precise characterization of the LT ligand forms would form the basis for such studies.
Recently, the size of TNF family of receptors and ligands has increased dramatically leading to an enhanced appreciation of the complexity of the control mechanisms governing the immune system. The types of communication occurring between cells in this system have been reasonably defined in the TNF, CD40, and Fas signaling systems. Understanding of the function of CD40 and Fas was aided by the linkage of these mole-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
cules to the profound phenotypes accompanying the lpr mouse in the Fas case (13) and the hyper IgM globulinemia associated with CD40 ligand defects in humans (14). TNF, on the other hand, benefitted from its natural existence as a soluble mediator and the ready availability of a recombinant form. We have taken the approach of converting the surface LT heteromeric complexes into a TNF-like secreted form. The generation of the surface LT complexes in soluble forms has allowed us to examine several key facets of this unusual structure. First, the LT-␣ and LT-␤ stoichiometry in the soluble complexes could be unambiguously determined. Next we were able to assess the stability of the trimers and ascertain whether scrambling could occur. Such interconversions would vastly complicate any biological studies performed using the soluble molecules as mimics of the surface forms. The receptor binding properties of these ligands will be analyzed in a subsequent paper. 2

Baculovirus Constructs
The baculovirus constructs encoding for LT-␣ and a truncated version of LT-␤ with a 10-amino acid myc tag have been described (8). An altered form of LT-␤ that retained the extracellular domain and the VCAM-1 leader sequence yet lacked the myc tag was prepared by ligating the VCAM-1 signal sequence to a polymerase chain reaction product of LT-␤. A NotI blunt end fragment of the signal sequence was obtained as described (8). The LT-␤ fragment was produced by polymerase chain reaction amplification off the cDNA with the primers 5Ј-GACCCCGGGGCACAGGCCCAG-3Ј and 5Ј-CAGTGCGGCCGCT-CACGCACTCGCACCAC-3Ј using Pfu DNA polymerase (Stratagene, La Jolla, CA) in 20 mM Tris-HCl, pH 8.75, 10 mM KCl, 10 mM (NH 4 ) 2 SO 4 , 2 mM MgSO 4 , 0.1% Triton X-100, 0.1 mg/ml bovine serum albumin, 10% Me 2 SO, and 0.2 mM dNTPs for 25 cycles at 94°C for 30 s, 63°C for 1 min, and 72°C for 2 min with a 5-min extension at 72°C. The polymerase chain reaction product was gel purified, cut with NotI, and ligated to the VCAM-1 signal sequence and NotI-linearized Blue-BacIII. The NotI mutant of BlueBacIII (Invitrogen, San Diego, CA) was a gift from E. Garber (Biogen). Recombinant virus were produced and purified as described previously (8).
To create a mutated version of LT-␣ in which Asp 50 was replaced by Asn, a NotI fragment containing the entire human LT-␣ gene (5) was subcloned into a pUC8 derivative, pNN11. Unique site elimination mutagenesis (Pharmacia Biotech Inc.) was used to alter the Asp 50 codon, GAC, to AAC (Asn). Plasmids containing the mutation were screened for the loss of a RsrII site and then confirmed by DNA sequence analysis. The entire NotI fragment was sequenced to ensure that extraneous mutations were not introduced during the new strand synthesis reaction. For expression of the mutant protein in insect cells, the gel purified NotI fragment was ligated into NotI linearized, dephosphorylated BlueBacIII. Recombinant virus were produced and plaque purified as described by Invitrogen. For homo-or heterotrimer production, the LT␣ D50N virus was used to infect High Five (Invitrogen) cells alone or in combination with the VCAM/c-myc-tagged wild type LT-␤ virus.

Preparation of Receptor Affinity Resins
LT-␤-R (LT␤-R-Fc) and TNF-R55 (TNF-R55-Fc) immunoglobulin chimeras were constructed as described previously (15). Chinese hamster ovary cells expressing either LT␤-R-Fc or TNF-R55-Fc were grown in suspension for 10 -14 days in a Dulbecco's modified Eagle's medium/ Ham's F-12-based medium (JRH Biosciences, Lenexa, KS) supplemented with 10% fetal bovine serum, 10 g/ml bovine insulin, 10 g/ml bovine transferrin, and 0.1% F-68 shear protectant (Life Technologies, Inc.). At harvest, sodium azide was added to a final concentration of 0.05%, and the conditioned medium was clarified by sequential dead end filtration using a 5-micron followed by a 0.3-micron Polygard filter cartridges (Millipore, Bedford, MA). The filtrated medium was concentrated by ultrafiltration using a S10Y30 spiral cartridge system (Amicon, Danvers, MA) and stored at Ϫ20°C. The receptor-Fc chimeras were purified by Protein A affinity chromatography (POROS A column, PerSeptive Biosystems, Framingham, MA). To prepare LT-␤-R-Fc and TNF-R55-Fc affinity resins, both preparations were immobilized on cyanogen bromide-activated Sepharose 4B (Pharmacia) overnight in 0.1

Generation of LT-␣/␤ Containing Conditioned Medium from Baculovirus-infected Insect Cells
To prepare LT-␣/␤ heteromers, High Five (Invitrogen) insect cells were grown to late log phase in a bioreactor controlled for oxygen containing 10 liters of sf900II medium (Life Technologies, Inc.) at 28°C. The cells were infected at a density of 2.2 ϫ 10 6 cells/ml with two different recombinant baculovirus preparations coding for LT-␣ and LT-␤ at a multiplicity of infection of 2 and 7, respectively. Fresh sf900II medium (1.5 liters) was added at the time of infection, and the culture was harvested 49 h post infection. Cells and cell debris were removed by centrifugation. A protease inhibitor mixture containing phenylmethylsulfonyl fluoride (Sigma) and EDTA were added to the supernatant at a final concentration of 0.15 mM and 1 mM, respectively, prior to filtration using a 10-inch 5-micron followed by a 10-inch 0.3-micron spiral Polygard cartridge filter (Millipore). The clarified insect cell supernatant containing LT-␣/␤ was concentrated 10-fold by ultrafiltration using three S1Y10 spiral membrane cartridges (Amicon) connected in series. The concentrate was stored frozen at Ϫ70°C. To prepare LT-␣3, sf9 insect cells (Life Technologies, Inc.) were grown in the bioreactor to late log phase in 9 liters sf900II culture medium. The cells were infected with baculovirus coding for LT-␣ at a multiplicity of infection of approximately 5. The culture was harvested 48 h post infection and processed as described for the LT-␣/␤ co-infected culture.

Affinity Purification of LT Forms
To purify the LT-␣/␤ heteromers contained in the insect cell conditioned medium, the concentrate was loaded onto a TNF-R55-Fc affinity column at a flow rate of 0.5 ml/min. The flow-through material was stored at 4°C, and the column was sequentially washed with 5 column volumes of PBS, 5 column volumes of PBS supplemented with 500 mM NaCl, and again with 5 column volumes of PBS. The bound material was eluted with elution buffer into fractions containing a 5% final volume of 500 mM sodium phosphate, pH 8.6 (neutralization buffer). To ensure that all the free LT-␣ was removed, the flow-through material from the first TNF-R55-Fc column was reloaded onto a second TNF-R55-Fc column, and the flow-through was again collected. The elution fractions from the two TNF-R55-Fc column runs that contained LT-␣3 and LT-␣2/␤1 were identified by UV absorbance and pooled. To obtain pure LT-␣2/␤1, the combined TNF-R55-Fc elution pools were immediately loaded onto a LT-␤-R-Fc affinity column equilibrated in PBS at a flow rate of 0.5 ml/min. The column was washed and eluted as described for the TNF-R55-Fc affinity column. To obtain pure LT-␣1/␤2, the flow-through from the second TNF-R55-Fc column was loaded onto a LT-␤-R-Fc affinity column at a flow rate of approximately 0.5 ml/min. The column was washed as described for the TNF-R55-Fc column and eluted with the elution buffer at pH 3.5 instead of pH 2.8. To obtain pure LT-␣3 trimers, the concentrate from the insect cell culture infected with the LT-␣ virus only was purified using a TNF-R55-Fc affinity column. The bound proteins were washed as described earlier, and the column was eluted with elution buffer at pH 3.5 into tubes containing 5% neutralization buffer. For all affinity purifications, the fractions containing protein were identified by UV absorbance, pooled, and analyzed on 10 -20% gradient SDS-PAGE gels (Daiichi, distributed by Integrated Separation Systems, MA). The purified LT preparations were stored in portions at Ϫ70°C.

Purification of LT-␣ and LT-␣/␤ Multimers Containing the D50N Mutated LT-␣ Form
Human D50N mutated LT-␣1/␤2 and LT-␣2/␤1 were purified using LT-␤-R-Fc affinity followed by cation exchange chromatography. The two heterotrimers were isolated from clarified culture supernatant on a LT-␤-R-Fc Sepharose column essentially as described for the wild type LT trimers and were eluted with elution buffer at pH 2.8 and immediately neutralized. To separate D50N mutated LT-␣1/␤2 from LT-␣2/␤1 contained in the LT-␤-R-Fc Sepharose elution pool, the sample was diluted with two volumes of buffer A (see below) and chromatographed on a carboxymethyl column as described below under "Analytical Methods." 2 J. L. Browning, manuscript in preparation.

Analytical Methods
Cation Exchange Chromatography-The affinity purified LT trimers were diluted 1:2 with 16.66 M HEPES, 16.66 M MES, 16.66 M sodium acetate, pH 6.5 (buffer A), and loaded onto a carboxymethyl column (POROS, 4.6 ϫ 100 mm, PerSeptive Biosystems, Framingham, MA) equilibrated with buffer A. The bound protein in each of the runs was washed with buffer A and eluted with a linear 20 column volume gradient from 0 to 1.0 M NaCl in buffer A at a flow rate of 5.0 ml/min. The effluent was monitored for absorbance at 280 nm, and the elution profiles for all LT preparations were compared.
Gel Filtration Chromatography-The LT preparations were loaded onto a 7.8 ϫ 300-mm TSK G3000SWXL column (TosoHaas, Montgomeryville, PA) equilibrated in PBS at a flow rate of 0.5 ml/min with PBS. The column effluent was monitored for absorbance at 280 nm. To calibrate the column, gel filtration standard (Bio-Rad) was also run using identical conditions. Gel filtration of the crude culture supernatant from insect cells infected with baculovirus coding for LT-␤ only was carried out on a fast protein liquid chromatography Superose-6 column (Pharmacia). The column was developed in PBS at a flow rate of 20 ml/h at ambient temperature.
Reversed Phase Chromatography-The samples were diluted with an equal volume of 8 M guanidine HCl, 0.1% trifluoroacetic acid and loaded onto a C4 reversed phase HPLC column (Vydac 2.1 ϫ 250 mm, Hesperia, CA) equilibrated with 0.1% trifluoroacetic acid. Bound proteins were washed with 0.1% trifluoroacetic acid and eluted with a linear gradient of 0 -75% acetonitrile in 0.1% trifluoroacetic acid over 30 min at a flow rate of 2.0 ml/min. The effluent was monitored for absorbance at 280 nm. The eluted fractions were collected in SDS to obtain a final concentration of 0.1% SDS in the fraction and dried under vacuum in a speedvac. Selected fractions corresponding to the absorbance peaks were resuspended in 2% SDS, 5% 2-mercapto ethanol, 0.2% bromphenol blue, 10% glycerol, 80 mM Tris-HCl, pH 6.8, and characterized by 10 -20% gradient SDS-PAGE stained with Coomassie Brilliant Blue.
LT-␤ Dot Blot/Western Analysis-Portions (50 l) of crude culture supernatant or selected gel filtration fractions were applied under vacuum to a nitrocellulose filter (0.45 m, Schleicher & Schuell) mounted in a 96-well dot blot chamber. The loaded filter was removed from the dot blot chamber, blocked with 0.3% Tween 20 in Tris-buffered saline (block buffer) for 45 min at ambient temperature. To detect bound myc-tagged LT-␤, the filter was incubated for 1 h at ambient temperature in a solution containing the c-myc-specific mAb 9E10 diluted 1:2000 in block buffer. After washing with block buffer, the bound 9E10 mAb was reacted with goat anti-mouse conjugated to horseradish peroxidase (Bio-Rad) diluted 1:40,000 in block buffer. The filter was washed again with block buffer and developed with the ECL chemiluminescence kit (Amersham Corp.) following the manufacturer's instructions.
Receptor ELISA-Microtiter plates (96 well, Immulon-II, Costar) coated with 50 l/well of anti-human Fc antibody (Jackson ImmunoResearch) at 1 g/ml were blocked for 1 h with PBS containing 1% bovine serum albumin, 0.05% Tween (blocking buffer). An aliquot (50 l) of TNF-R55-Fc or LT-␤-R-Fc at 1 g/ml in blocking buffer was added to each well and left to incubate for 2 h at ambient temperature. The wells were washed 6 times with 300 l of PBS containing 0.05% Tween 20 (PBS/T-20). Samples containing LT-␣3, LT-␣2/␤1, and LT-␣1/␤2 were diluted to cover a range of six logs (10 pg/ml to 10 g/ml) in blocking buffer, added to the wells, and incubated for 60 min at ambient temperature. The wells were washed 6 times with 300 l of PBS/T-20. 50 l of the mouse anti-human LT-␣ antibody AG9 (15) at 2.3 g/ml were added to each well, and the plates were incubated for 1 h at ambient temperature. The wells were washed with blocking buffer and then treated with 50 l of horseradish peroxidase-labeled goat antimouse heavy and light chain (Jackson ImmunoResearch, West Grove, PA) diluted 1: 50,000 in blocking buffer. Peroxidase activity was determined by incubation with the chromogenic substrate tetramethylbenzidine (420 M final, IC Biochemicals, Costa Mesa, CA) followed by absorbance measurement at 450 nm.

Biological Assays
LT-␣3-like activity was measured with either a 3-day anti-proliferative assay or a 1-day cytotoxicity assay in the presence of 10 g/ml cycloheximide using the TNF-sensitive murine WEHI 164 (clone 13) cell line. Cells were grown in RPMI 1640 medium with 10% Hyclone defined fetal bovine serum (Hyclone Laboratories, Logan, UT), glutamine, penicillin/streptomycin, and 10 mM HEPES buffer, pH 7.5. In the antiproliferative assay, 3000 cells were plated in each well of a 96-well plate along with various dilutions of cytokines as described (16). After 3 days, growth was assessed by the addition of 3-(4,5-dimethylthiazol-2-yl)2,5diphenyltetrazolium bromide (MTT), the formazan generated was solubilized by the addition of 0.1 ml of 10% SDS in 0.01 N HCl, and the absorbance was determined after 24 h at 37°C. In the cytotoxicity assay, cells were preplated 1 day prior to the assay at 10,000 cells/well in 0.1 ml. The next day 0.05 ml was removed and replaced with the cytokine plus 10 g/ml cycloheximide. One day later, the viable cell mass was quantitated with MTT .

Expression of Soluble LT-␣/␤ Heteromers in Insect Cells-To
prepare a soluble form of the heteromeric LT-␣/LT-␤ complex, the small intracellular N-terminal domain and the transmembrane spanning region of human LT-␤ were removed and replaced with the leader sequence of VCAM. Cleavage of the VCAM leader was expected to result in secretion of the LT-␤ extracellular domain. To facilitate detection, a 10-amino acid peptide tag, which is recognized by the mAb 9E10 was also added between the VCAM leader and the beginning of the extracellular receptor binding domain (Fig. 1). A second form of LT-␤ was prepared with the same leader; however, the myc peptide tag was removed and substituted with an additional 10 amino acids from LT-␤. This form is referred to as soluble myc-less LT-␤.
To express soluble LT-␣/␤ complexes, High Five insect cells were infected in late log phase with a mixture of two baculovirus stocks coding for LT-␣ and LT-␤, respectively. Kinetic analysis of insect cell supernatant by Western blotting revealed maximal expression of both LT forms at about 60 h post infec- tion. However, degradation products of LT-␤ in the culture medium could be detected as early as 25 h post infection. In contrast, LT-␣ appeared to be stable in the culture supernatant well after peak expression was reached (data not shown). To avoid excessive degradation of LT-␤ during the production phase, all subsequent cultures were harvested between 40 and 50 h post infection. The cells were immediately removed by centrifugation, phenylmethylsulfonyl fluoride and EDTA were added as protease inhibitors prior to ultrafiltration, and the concentrate was stored frozen at Ϫ70°C.
Purification of LT-␣/␤ Heteromers Using Receptor Affinity Chromatography-Although LT-␣ expressed in bacteria or Chinese hamster ovary cells has been well-characterized, we wished to prepare LT-␣ from insect cells for comparison to the insect cell-derived heteromeric LT forms. Pure LT-␣ was readily obtained by TNF-R55-Fc affinity column purification in high yields as described (17). Likewise, we wished to obtain pure LT-␤3 via a similar route. LT-␤ prepared by insect cell expression and purification over an anti-LT-␤ B27 mAb affinity column (15) was found to be highly aggregated, and the majority of the preparation did not enter a nonreducing SDS-PAGE gel. To determine if the LT-␤ was aggregated in the cell supernatant prior to purification, the conditioned medium was sized by gel exclusion chromatography, and the fractions were analyzed by a dot blot/Western analysis (Fig. 2). Essentially all of the LT-␤ eluted in a size range greater than 200 kDa, and it was concluded that LT-␤ was highly aggregated. Attempts to obtain pure LT-␤ by elution from a LT-␤-R-Fc affinity column were unsuccessful. Therefore, LT-␤ is expressed at reasonable levels in insect cells, and it appears to immediately form large aggregates that either lack good receptor binding sites or cannot be eluted from an affinity column.
By analogy with other members of the TNF family, LT-␣/␤ heteromeric forms were expected to be trimers. Therefore, LT-␣ and LT-␤ co-infection of insect cells could theoretically produce LT-␣3, LT-␣2/␤1, LT-␣1/␤2, and LT-␤3 forms, and the separation of these related LT complexes presented an interesting challenge. Initially, we explored affinity chromatography using immobilized mAbs specific for LT-␤ and LT-␣. However, the affinity of the LT-␤ specific mAbs was high, and it was difficult to eluate bound material from the resin without destroying the column. Attempts to take advantage of the "FACS dull" and "FACS bright" groups of mAbs specific for LT-␣ (15) as an affinity purification strategy for LT-␣1/␤2 were equally unsuccessful. Thus, we abandoned mAb-based affinity columns for purification purposes.
Receptor-based affinity columns provided an alternate method that exploited the expected specificity of the TNF-R55 for LT forms containing a LT-␣/LT-␣ cleft, i.e. LT-␣3 and LT-␣2/␤1 and the specificity of the LT-␤-R for LT forms containing LT-␤/LT-␤ or LT-␣/LT-␤ clefts, i.e. LT-␣1/␤2 and LT-␣2/␤1. Shown in Fig. 3A is a schematic diagram of the purification strategy that yielded essentially pure LT-␣1/␤2 and LT-␣2/␤1 (Fig. 3B). Conditioned medium from dually infected insect cells was passed several times over a TNF-R55-Fc affinity column to remove both LT-␣3 and LT-␣2/␤1. The depleted flow-through from this step was loaded onto a LT-␤-R affinity column, and the eluate contained pure LT-␣1/␤2. The eluate from the original p55 TNF-R column was loaded onto a LT-␤-R affinity column, and LT-␣2/␤1 was eluted. The preparation obtained from the TNF-R55-Fc flow-through LT-␤-R elution arm is referred to as LT-␣1/␤2; the material prepared from the TNF-R55-Fc eluate that was further fractionated on the LT-␤-R affinity column is referred to as LT-␣2/␤1. These preparations are used in the analytical evaluations. Switching the order of the receptor affinity columns to using the LT-␤-R column first and the TNF-R55-Fc column second also results in the successful separation of the two heteromers where LT-␣1/␤2 is contained in the flow-through and LT-␣2/␤1 in the elution pool of the TNF-R55-Fc column.
A receptor-based ELISA was developed to aid in the purification and provided our first insight into the receptor specificities of the various ligand forms. The LT-␣1/␤2 form was found to bind selectively to the LT-␤-R and poorly to the TNF-R55, whereas the LT-␣2/␤1 form bound well to both receptors. As expected, LT-␣3 only bound to the TNF-R55, and no appreciable binding could be detected to LT-␤-R (Table I). The ELISA was especially useful for optimizing the affinity column elution pH, the buffer compositions, and storage temperatures used during the purification. The conditions described under "Materials and Methods" are optimized for ease of elution, stability, and purity of the eluted material.
Biochemical Characterization of the LT Heteromers-Direct sequencing of the purified LT-␣ and LT-␤ forms expressed from baculovirus-infected insect cells showed that the N terminus of secreted LT-␣ was as expected based on previous literature reports, and there was no evidence for substantial amounts of the des-20 LT-␣ form described earlier (18). The N terminus of both the myc-tagged and myc-less secreted LT-␤ forms was as indicated in Fig. 1, i.e. the VCAM leader was processed as expected. LT-␤-derived degradation products were abundant in the crude culture media but were much reduced in the receptor affinity purified products, suggesting that only essentially intact LT-␤ polypeptides can productively assemble with LT-␣ into receptor binding multimers. Interestingly, the levels of LT-␤ degradation products observed in the myc-less preparations appeared to be more extensive than was observed in the material containing the myc tag. Whether the presence of the myc tag provides increased protection against degradation or the myc-less LT-␤ virus stock contained higher levels of protease activities than the myc-tagged virus stock used for infection was not determined. The purity of the preparations in terms of LT-␣ and LT-␤ content was high based on the SDS-PAGE analysis (Fig. 3B) and C4 reversed phase HPLC (Fig. 4). SDS-PAGE and reversed phase HPLC analysis of the purified fractions were carried out not only to assess purity but also to quantify the ratio of LT-␣ to LT-␤. Densitometry of the Coomassie Blue-stained SDS-PAGE containing the purified LT-␣2/␤1 and LT-␣1/␤2 preparations revealed LT-␣ to LT-␤ ratios of approximately 2:1 and 1:2, respectively. Reversed phase HPLC further supported these stoichiometries. Because LT-␣ and LT-␤ both possess two tryptophan residues and similar numbers of tyrosine residues, the relative absorbance at 280 nm will reflect the actual stoichiometric ratio of LT-␣ to LT-␤. The peak heights shown in Fig. 4 indicate that the preparations designated LT-␣1/␤2 and LT-␣2/␤1 possess the expected ratios of LT-␣ to LT-␤.
To determine which of the various heteromeric forms were present in the preparations derived from the affinity columns, cation exchange chromatography was found to resolve the LT-␣3, LT-␣2/␤1, and LT-␣1/␤2 forms (Fig. 5). Judged on this basis, the purity of these preparations was greater than 95%. It was difficult using biochemical methods to ascertain more accurately the levels of contamination by other forms. The ion exchange data suggested that the LT-␣1/␤2 preparation was contaminated with about 5% LT-␣2/␤1, whereas the LT-␣2/␤1 preparation lacked obvious LT-␣3 or LT-␣1/␤2 contaminants. Based on the receptor binding data obtained in the ELISA (Table I) and preliminary BIAcore data, 2 it is clear that the LT-␣1/␤2 is unable to bind with reasonable affinity to the TNF-R55 and that LT-␣3 cannot bind to the LT-␤-R. On this basis one would expect a high level of purity in the preparations obtained using combinations of TNF-R55-Fc and LT-␤-R-Fc affinity columns. It is possible that the preparations contain different glycoforms of the two LT-␣/␤ or the LT-␣3 forms with isoelectric points similar to the bulk LT-␣1/␤2 or LT-␣2/ ␤1. The occurrence of minor forms with different oligosaccharides may result in contaminating species and would underlie the parent LT-␣1/␤2 and LT-␣2/␤1 peaks in the cation exchange chromatographic analysis and lead to a potential overestimation of the purity of the preparations. However, it is likely that such cross-contamination is only present at low levels if at all because the LT-␣ to LT-␤ ratios determined by SDS-PAGE and C4 reversed phase analysis suggest a precise stoichiometry of 2:1.
Gel exclusion chromatography of the purified fractions showed that each of the three forms eluted in the size range of 40 -70 kDa with the LT-␤ containing forms being larger (Fig.  6). Trimeric TNF migrates as a compact 40-kDa molecule that is smaller than its actual molecular mass of about 51 kDa. Because these structures are also expected to be relatively compact, we interpret the LT sizing results to indicate that all three forms are trimers. If the sizing results were representative of dimers, it would be impossible to reconcile the SDS-PAGE and reversed phase HPLC ratio analyses. Taking together, the ratio analyses and the sizing results show that the LT-␣1/␤2 and LT-␣2/␤1 heteromers form trimers. There was some concern that the myc tag attached to the N terminus of LT-␤ may affect the properties of the complexes. A soluble LT-␤ construct essentially identical to the myc-tagged version was prepared without the myc tag. Heteromeric complexes prepared by the same route described above yet lacking the myctagged LT-␤ had identical biochemical properties.
Biological Activity of LT Heteromeric Forms-TNF/LT-␣3 biological activity was measured using a conventional WEHI 164 cytotoxicity assay. LT-␣3 is very active in this assay with 50% cell death typically occurring in the presence of 0.5-5 pg/ml LT-␣ (Fig. 7, top). The specific activities of LT-␣ prepa-rations generated from Chinese hamster ovary or insect cells were similar. The specific activity of LT-␣2/␤1 was about 100fold less than pure LT␣3 in this assay (Fig. 7, middle). LT-␣2/␤1 has a single high affinity TNF-R binding site that can complex with TNF-R and prevent further receptor cross-linking and as a consequence, LT-␣2/␤1 can actually block the biological activity of TNF or LT-␣3 in some assays. 3 For this reason the flattened titration curve shown in Fig. 7 and the lack of complete killing at high concentrations was expected. LT-␣1/␤2 was about 5000-fold less active than pure LT␣3 in this assay. Assuming that this activity results solely from contaminating LT␣3-like molecules, one can calculate that there is 0.02% LT-␣3 in the LT-␣1/␤2 preparation. The activity was blocked by anti-LT-␣ neutralizing mAbs or soluble TNF-R55-Fc consistent with the hypothesis that the activity was contaminating LT-␣3. 3 P. Hochman, manuscript in preparation. To more carefully address the issue of contaminating LT-␣3 versus direct signaling by LT-␣1/␤2 forms on LT-␤-R, we prepared a mutated form of LT-␣. Mutation of Asp 50 to Asn 50 was shown to disrupt the activity of human LT-␣ in the mouse assay system (19). Preparation of this mutant and expression in the baculovirus/insect cell system confirmed a 100,000-fold reduction in cytotoxic activity both in the mouse WEHI 164 (Fig. 7) and in a human TNF-R55-based cytotoxic system using WiDr cells (data not shown). LT-␣1/␤2 and LT-␣2/␤1 heteromeric forms containing D50N LT-␣ were prepared by first selecting all forms capable of binding to the LT-␤-R affinity resin, i.e. LT-␣1/␤2 and LT-␣2/␤1. These two forms were then resolved by ion exchange chromatography. All three mutated forms, LT-␣3, LT-␣2/␤1, and LT-␣1/␤2 were subjected to the same biochemical tests described for wild type LT preparations, which confirmed that the D50N mutation in LT-␣ allowed trimer formation with the expected LT-␣:LT-␤ ratios. All WEHI 164 activity in the heteromeric forms was eliminated by the D50N mutation, strongly indicating that the residual WEHI 164 activity in the heteromeric forms could be accounted for by minor amounts of LT-␣3 contaminates. It is possible that other receptor components are involved in LT-␣1/␤2 signaling, e.g. a putative LT-␣/LT-␤ cleft specific receptor, and that the D50N mutation also affects this interaction resulting in the loss of activity. At this point, however, the presence of low level contaminating LT-␣3 represents the simplest explanation of the data. Based on the ability to purify the D50N mutated form of LT-␣1/␤2 by LT-␤-R-Fc affinity chromatography, the mutation does not grossly affect binding to the LT-␤-R. Human LT-␣1/␤2 is cytotoxic to the human adenocarcinoma cell line HT-29, and this event is mediated by the LT-␤-R (20). In this assay, the D50N LT-␣1/␤2 form retained essentially full specific activity. 4 LT-␣1/␤2 lacking the myc tag was also fully active in the HT-29 assay.
Stability of Heteromeric Complexes-Given the complex nature of these heteromeric LT forms, it was considered important to ascertain their stability under typical biological conditions such as in tissue culture medium. The sensitivity of the WEHI 164 assay was exploited to detect increased levels of LT␣-like activity, which may arise following various treatments. Fig. 8 shows the results of one such experiment wherein LT-␣1/␤2 was stored in tissue culture medium with 10% fetal bovine serum either at 4 or 37°C. Storage at 37°C over 7 days in a CO 2 incubator did not lead to the appearance of increased levels of LT-␣ like activity. In contrast, storage at 4°C without CO 2 buffering slowly led to generation of LT-␣-like activity and within 7 days about 3% of the LT-␣1/␤2 had converted into LT-␣3. In other experiments, storage at either 4 or 37°C in PBS supplemented with 10 mg/ml bovine serum albumin (Table II) did not lead to such pronounced lability. Storage of the pure LT heteromers in PBS without carrier protein showed conversion of only 0.3% into LT-␣3-like activity after 4 months at 4°C. For this reason, it appears that the preparations are not intrinsically cold labile, but either (i) the protein is less stable at alkaline pH, i.e. in the alkaline pH conditions reached in tissue culture medium left without CO 2 buffering, or (ii) enhanced proteolysis is occurring at the alkaline pH levels. Storage of pure protein in a pH 8 Tris buffer led to more rapid increases in LT-␣3-like activity than storage in neutral or pH 6 buffers, supporting the concept of lability at alkaline pH. Concerning proteolysis, we have noticed that the des-20 form of human LT-␣, when present in the original insect cell supernatant, does not associate with LT-␤, therefore mild proteolysis of the N-terminal tail of LT-␣ or for that matter LT-␤ may lead to destabilization of the complex and accelerated scrambling into a LT-␣ form. The des-20 LT-␣3 form retains full specific activity (18), which would be consistent with this hypothesis.
During these storage experiments, LT-␣1/␤2 activity was also analyzed in another cytotoxicity assay. LT-␣1/␤2 is cytotoxic to the HT-29 cell line via a mechanism that involves the LT-␤ receptor (20). Solutions of LT-␣1/␤2 that had undergone high levels of conversion such as shown for the sample stored for 7 days (Fig. 8) did not have altered activity in the HT-29 assay. Therefore, even preparations that contained 1-3% of LT-␣3-like activity as determined by the WEHI 164 cytotoxicity assay showed no substantial change in the properties of the bulk LT-␣1/␤2 as assessed with the HT-29 cytotoxicity assay. Lastly we attempted to force scrambling by repeated freezethawing and even following multiple rounds; there was no increase in LT-␣3-like activity whether measured by the cytotoxicity assay or biochemically by ion exchange chromatography. Taken together, these experiments show that the LT-␣1/␤2 form is stable and does not readily scramble. DISCUSSION To study the function of surface LT heteromers we chose to express and characterize soluble recombinant versions of the LT-␣/␤ complexes. Recombinant LT-␣ has been expressed in bacteria (21)(22)(23) or secreted from Chinese hamster ovary cells (16,24,25) and insect cells (17,26). In view of the efficient expression observed in the baculovirus system of LT-␣, we concentrated on the simultaneous expression of LT-␣ and LT-␤ in that system. The co-infection of insect cells with baculoviruses encoding the LT-␣ and a re-engineered secreted LT-␤ 4 F. MacKay and C. Ambrose, unpublished data.

FIG. 8. WEHI 164 cytotoxicity assay used to follow the generation of LT-␣3-like forms in solutions of LT-␣1/␤2.
A 400 ng/ml solution of LT-␣1/␤2 was stored for 0 (f), 2 (E), or 7 (Ⅺ) days in minimum essential medium with 10% fetal bovine serum at either 4°C in air or at 37°C in a CO 2 incubator. Samples were assayed in a 24-h cytotoxicity assay in the presence of 10 g/ml cycloheximide. A fresh dilution of insect cell derived recombinant LT-␣3 is included for comparison (q). Plotted is the absorbance of reduced MTT.
protein led to secretion of complexes of LT-␣/␤, and the LT-␣2/␤1 and LT-␣1/␤2 forms could be selectively purified using receptor based affinity resins. We wished to address the following questions: (i) Was there sufficient discrimination by the TNF-R55-Fc and LT-␤-R-Fc affinity columns to purify the expected forms? (ii) What stoichiometric ratios of LT-␣ and LT-␤ were present in the heteromers? (iii) Were the heteromers trimers? (iv) Could the defined heteromeric forms eventually scramble into a mixture of all four conceivable forms?
The data presented here show that soluble LT-␤ was highly aggregated when expressed by itself and only by co-expression with LT-␣ did proper trimeric forms result. Presumably, LT-␤ is incapable of packing into a stable homomeric trimers and inclusion of LT-␣ compensates for the incorrect geometry. We were unable to purify a pure LT-␤ form using the LT-␤-R-Fc or TNF-R55-Fc affinity columns even when LT-␤ was expressed in the absence of LT-␣. In contrast, the purification scheme described here led to pure preparations of LT-␣1/␤2 and LT-␣2/ ␤1. The combination of an approximately trimeric size on gel exclusion chromatography coupled with the 2:1 ratios observed by SDS-PAGE and in the reversed phase HPLC analyses definitively demonstrates the trimeric nature of these soluble forms. The dramatic conversion of the LT-␤ aggregates into LT-␣/␤ trimers when LT-␤ is co-expressed with LT-␣ is interpreted as strong biochemical evidence favoring the heteromeric complex as the physiological LT ligand. Previous studies on cell surface LT were also consistent with a complex that was at least as large as the trimeric LT␣1␤2 structure (4,7). Human TNF and LT-␣ have been shown to exist as trimers by a number of techniques (7). Likewise, recent crystallographic analysis of the CD40 ligand also revealed it to be a trimer (27). The structural homology among the members of the family occurs in the regions of the molecule known to be involved in intersubunit contacts, and therefore, one would expect that the trimer packing would be conserved as a fundamental feature of the entire family. The delineation of the soluble LT forms as trimers strengthens the hypothesis that the cell surface forms are similar trimeric forms. Moreover, the ability to separate the two LT-␣/␤ complexes rather than obtaining heterogeneic mixed aggregates indicates that assembly into trimeric structures is definitely preferred. Therefore, the interactions between LT-␣ and LT-␤ subunits must lend considerable stability to the trimeric structure.
The very fact that the receptor-based purification scheme presented here yielded trimeric forms composed of LT-␣2/␤1 and LT-␣1/␤2 provides some information about the nature of the receptor interactions. The crystal structure of LT-␣ complexed with the TNF-R55 showed the receptor nested in the cleft between two LT-␣ subunits with productive interactions being made to two separate subunits (28). If this binding mechanism is generally valid for the family, which is a reasonable expectation, then only trimers containing a LT␣/LT␣ cleft should bind TNF-R55. The binding of LT-␣3 and LT-␣2/␤1 to the TNF-R55-Fc affinity column confirmed this expectation. Likewise the binding of LT␣1␤2 to the LT-␤-R-Fc column confirmed that a molecule with a LT-␤/LT-␤ cleft will bind the receptor, which was expected based on the weak interactions of the LT-␤-R with pure LT-␤ described earlier (8). The binding of LT-␣2/␤1 to the LT-␤-R indicated that a LT-␣/LT-␤ cleft is capable of interacting with the receptor. Further details of the receptor binding interactions will be discussed elsewhere. 2 The primary goal of this work was to prepare a soluble form of the LT-␣1/␤2 ligand that could be used in biological studies. Therefore, it was important to ascertain the composition and purity of the ligand. The final preparations are greater than 95% pure based on ion exchange resolution. Of paramount importance was the quantitation of the levels of LT-␣3-like contaminants in the LT-␣1/␤2 preparation because LT-␣3 is biologically active in many systems and its presence would confuse the interpretation of any study. The murine WEHI 164 cell dies in the presence of LT-␣3, providing a very sensitive indicator of LT-␣3 contaminating the LT-␣1/␤2 preparation. In such assays, the LT-␣1/␤2 preparation was found to contain approximately 0.02% LT-␣3-like activity. This activity could be blocked by soluble TNF-R55-Fc or anti-LT-␣ monoclonal antibodies and therefore is expected to represent signaling by a LT-␣3-like form. The ion exchange analysis suggested that there was about 2-5% LT-␣2/␤1 in the LT-␣1/␤2 preparation and a BIAcore analysis 2 was consistent with such a contaminant. Therefore, in our best assessment, this preparation contains greater than 95% LT-␣1/␤2, less than 5% LT-␣2/␤1, and less than 0.02% LT-␣3. The purity of the LT-␣2/␤1 preparation was more difficult to quantitate using the cytotoxicity assay; however, by ion exchange analysis it is greater than 95% pure. We were unable to exactly quantitate LT-␣3-like activity in this preparation because the LT-␣2/␤1 is an inhibitor 3 and obscures possible LT-␣3 components. BIAcore analysis of the binding of these preparations showed essentially no binding of LT-␣1/␤2 to the TNF-R55-Fc and no binding of LT-␣3 to the LT-␤-R-Fc. Therefore one would expect that affinity purifications of both heteromers based on a combination of these columns to be very pure, and that assumption is supported by these analyses.
In addition to the actual initial purity of these heteromers, we were concerned that these complexes would scramble into a mix of the various forms because there are no covalent interactions involved in trimerization. Some of the initial work on the quaternary structure of TNF suggested that there was an equilibrium between monomers, dimers, and trimers. In the original characterization of the surface form of LT, we added unlabeled soluble LT-␣ to cells displaying biosynthetically labeled LT-␣/␤ (3). The loss of immunoprecipitable LT-␣ label that would indicate exchange of LT-␣ on and off the surface complex was not observed, suggesting a relatively stable struc- a A 400 ng/ml solution was stored in either PBS with 1 mg/ml bovine serum albumin or minimum essential medium supplemented with 10% fetal bovine serum for the indicated times and then assayed in a 24 cytotoxicity assay with cycloheximide. Specific activity of baculovirus derived LT-␣3 in this assay was 1.6 ϫ 10 8 units/mg, whereby 50% of the cells in a 1 unit/ml solution will die. ture. Here we have shown that the LT-␣1/␤2 heteromer is indeed quite stable when stored in neutral pH buffers, and moreover, attempts to force scrambling by repeated freezethawing of the preparation did not readily lead to increased amounts of LT-␣-like activity. For these reasons, the heteromeric structures are concluded to be relatively stable and therefore useful for biological work. Outside the demonstration that stable LT-␣/␤ complexes exist on the lymphocyte surface, there is little information on the quaternary state of these TNF family ligands on cell surfaces. The immunoprecipitation of the surface TNF form showed the presence of 17 kDa of processed TNF in addition to the 26-kDa form retaining the transmembrane form (29). This result was interpreted to mean that some processed TNF chains must be complexed with unprocessed chains, i.e. oligomeric complexes are present. Because the soluble LT complexes form stabile trimers, we conclude that the same structures are present on the actual membrane surface.
The characterization of these soluble trimeric LT forms further indicates that the previously proposed model for surface LT structure is correct, i.e. a surface LT-␣1/␤2 complex. Additionally, the soluble forms are functionally active (20) and therefore suitable for immunological studies. The availability of these ligands now opens the way to an analysis of their binding to the LT-␤ receptor and elucidation of the signaling mechanism. A separate paper will describe the binding characteristics of these proteins to the various receptors in the TNF family. 2