Evidence for an extended structure of the T-cell co-receptor CD8 alpha as deduced from the hydrodynamic properties of soluble forms of the extracellular region.

We expressed soluble forms of the human T-cell coreceptor CD8 alpha extracellular region, CD8 alpha 161, and the amino-terminal immunoglobulin-like domain, CD8 alpha 114, in Chinese hamster ovary cells and Escherichia coli, respectively. Both molecules were readily purified to homogeneity in milligram amounts and were recognized by a large panel of monoclonal antibodies. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis analysis indicated that approximately 70% of CD8 alpha 161 was secreted as a disulfide-linked dimer, but CD8 alpha 114 was not disulfide-linked. To investigate the structural features of CD8 alpha 161 and CD8 alpha 114 under native conditions, we performed gel filtration and sucrose gradient sedimentation analysis. In spite of being partially or totally noncovalently bound, both recombinant molecules were stably associated homodimers, as no monomers could be detected at a fairly low protein concentration (approximately 1 microM). This suggests that the CD8 alpha amino-terminal domain alone strongly contributes to chain association. Determination of the Stokes radius (Rs) and sedimentation coefficient (s20,w) gave results consistent with CD8 alpha 114 having a globular shape and CD8 alpha 161 being an asymmetric molecule. Taking into account the contribution of hydration to the frictional coefficient, we obtained for CD8 alpha 161 an axial ratio of approximately 5, when modeled as a prolate ellipsoid. These results indicate that the elongated structure of CD8 alpha 161 is essentially contributed by the hinge region and help to explain how the CD8 alpha is able to bridge the distance between the T-cell surface and its binding site in the alpha 3 domain of major histocompatibility complex class I molecules on the target cell.

70% of CDSa161 was secreted as a disulfide-linked dimer, but CD8a114 was not disulfide-linked. To investigate the structural features of CD8a161 and CD8a114 under native conditions, we performed gel filtration and sucrose gradient sedimentation analysis. I n spite of being partially or totally noncovalently bound, both recombinant molecules were stably associated homodimers, as no monomers could be detected a t a fairly low protein concentration (-1 PM). This suggests that the CD8a amino-terminal domain alone strongly contributes to chain association. Determination of the Stokes radius (Rs) and sedimentation coefficient (sz0J gave results consistent with CD8a114 having a globular shape and CDSal61 being an asymmetric molecule. Taking into account the contribution of hydration to the frictional coefficient, we obtained for CDSal61 an axial ratio of approximately 5 , when modeled as a prolate ellipsoid. These results indicate that the elongated structure of CDSa161 is essentially contributed by the hinge region and help to explain how the CD8a is able to bridge the distance between the T-cell surface and its binding site in the a3 domain of major histocompatibility complex class I molecules on the target cell. CD8 is a cell surface transmembrane glycoprotein expressed on a subset of T-cells that recognize peptide antigen presented by major histocompatibility complex class I molecules (1,2). Antibody-blocking studies on cytotoxic T-cells (3) and gene * This work was supported in part by grants from the Institute Pasteur, from the Conseil National de la Recherche Scientifique, and from the Fondation pour la Recherche Medicale. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked transfer experiments in T-cell hybridomas (4,5) have shown that CD8 is required for efficient antigen recognition and cellular activation. Several lines of evidence indicate that CD8 binds to class I molecules (6)(7)(8) and, in particular, to a nonpolymorphic site in the membrane-proximal a3 domain (9). Mutations in a 7-residue loop exposed to solvent in this domain have been found to affect CD8 binding (10,11). Furthermore, studies with these mutants have shown that the simultaneous binding of CD8 and T-cell receptor to the same class I molecule on the target cell is essential for T-cell recognition and activation (10, l l ) , indicating that CD8 acts as a co-receptor. These data suggest that a ternary complex is formed in which the T-cell receptor interacts with the peptide antigen:al/az domains of class I, while CD8 is engaged in the recognition of a more distal site in the a3 domain.
CD8 is found on the cell surface as a disulfide-linked a/.
homodimer and as a disulfide-linked a l p heterodimer (2). The CD8a chain (34-37 kDa) (12)(13)(14) is composed of an extracellular region containing an amino-terminal Ig-like domain of approximately 114 amino acids and a stretch of -48 residues (hinge region) particularly rich in proline, threonine, and serine and heavily glycosylated by 0-linked oligosaccharides (15,16). This hinge region is followed by a hydrophobic transmembrane anchor and a short (28 residues) cytoplasmic tail. The latter contains a site of attachment for the srcrelated tyrosine kinase, p56Ick (17,18), thought to regulate signal transduction following engagement of CD8 with class I (19). X-ray crystallographic analysis of the CD8a aminoterminal domain indicates that it folds similar to an Ig variable domain and forms a homodimer (16). Mutation analysis of CD8a has provided initial evidence that this domain contains a binding site for class I (20). The CD80 chain (30 kDa) (21-23) which is expressed on the cell surface only if paired to the a chain, shows a similar overall structural organization, with the Ig-like domain containing, in addition, a J-like sequence (24). The CD80 chain does not appear to associate with p56Ick (25).
The majority of class I-restricted T-cells express mostly the a l p heterodimer (21, 22) while a / a homodimers alone have been found on some gut intraepithelial T-cells (26), on some T-cell receptor y/6 T-cells (27) and on natural killer cells (28). Several experiments have shown that the CD8a homodimer is sufficient to ensure binding to class I (6-8) and signal transduction (4,5,19), although recent evidence suggests that the CD8 a//3 heterodimer confers a more efficient antigenspecific response (29).
In major histocompatibility complex class 11-restricted Tcells the CD4 molecule is the functional homologue of CD8 (2). CD4 binds to class I1 molecules (30), and recent work has shown that CD4 interacts with the Pz domain of class I1 (31,32) at a membrane-proximal site bearing structural homology 2013 to the CD8 binding site of class I. In contrast to CD8, however, CD4 is a monomer, and its extracellular region shows a different structural organization since it is formed of an array of four Ig-like domains. X-ray crystallographic studies of the first two Ig-like domains (33, 34), which bear the binding site for class I1 (35,36), and additional biochemical and physicochemical studies (37, 38) strongly suggest that the extracellular region is an extended structure. This is likely to yield a molecule roughly equal in length to the sum of the T-cell receptor and part of class I1 (the a1 and domains) interacting at the cell suface. Thus, the length of CD4 would be compatible with the first two Ig-like domains reaching the site of interaction with the p2 domain of class 11.
How then is CD8, whose unique Ig-like domain is connected to the transmembrane segment through -48 residues only, able to bridge a distance similar to the CD4 which contains four Ig-like domains? Although it has been proposed that the hinge region assumes an extended conformation (16,39), and some support to this hypothesis comes from the observation that this region appears unstructured in the x-ray crystallographic analysis (16), no direct evidence has been provided to date. Thus, to investigate the structural basis for CD8 function, we expressed two secreted soluble forms of the CD8a chain, one comprising its entire extracellular region and the other, the Ig-like domain only, using CHO' cells and Escherichia coli, respectively. We examined the hydrodynamic properties of these soluble forms of CD8a by gel filtration and sedimentation analyses. We show that both recombinant molecules are produced as stable homodimers, suggesting that the interaction between CD8a Ig-like domains provides an important contribution to chain association. Moreover, our data are entirely consistent with extracellular domain of the CD8a homodimer being an elongated structure, and this feature is essentially conferred by the hinge region. The successful expression of large quantities of the CD8 Ig-like domain should facilitate future structural studies.

MATERIALS AND METHODS
Monoclonal Antibodies-The following anti-human CD8 mAbs were used 21Thy-2D3 (IgG,) ( Antibodies were used either purified or as ascities fluid. CD8a161 and CD8a114 Constructs-To obtain CD8a161, a 1kilobase BamHI-FspI cDNA fragment encoding the human CD8a chain (13) was subcloned into M13mp19 and subjected to two sequential rounds of site-directed mutagenesis as described (45). Oligonucleotides 5'-TCCTGGGGAGGGATCCATGGCCTTA-3' and 5'-GACTTCGCCAGTGATTAATACATCTGG-3' introduced a RamHI site immediately prior to the translation initiation codon and changed the codon for Ile-162' (ATC) to a stop codon, respectively. T h e latter oligonucleotide also changed the codon for Cys-160 (TGT) t o a Ser codon (AGT A parallel amplification was carried out on CD8a161 DNA using the 5' primer oligonucleotide 5'-GCCCAACCAGCGATGGCCAGCCAGTTCCGG-GTGTCG-3' (primer 3) carrying the last six codons of pectate lyase B and hybridizing to the sequence corresponding to the first six codons of the mature CD8a and the 3' primer oligonucleotide, (primer 4) hybridizing to the sequence corresponding to amino acids 108-114 of CD8a followed by a stop codon and a BglII site. The two amplified fragments were purified on agarose gel, mixed together, and subjected to a third PCR using the M13RP oligonucleotide and primer 4 for generating the chimeric construct pectate lyase B-CD8a114. The amplified DNA fragment was digested with Hind111 and BglII, inserted into the vector pUC19, under the control of the inducible lac2 promoter, and introduced into the BMH 71-18 strain of E. coli. Each construct of CD8a was sequenced in its entirety by the dideoxy method (47).
Transfection and Amplification in CHO Cells-Dihydrofolate reductase-negative CHO cell line DG44 (48), a gift of L. Chasin (Columbia University, New York), was grown in Ham's F-12 medium (GIBCO) supplemented with 10% heat-inactivated FCS, 2 mM glutamine, 50 IU/ml penicillin, and 50 pg/ml streptomycin a t 37 "C, in 5% CO,. For transfection, DG44 cells were seeded at 5 X 105/100-mm tissue culture plate (Nunc). The next day, cells were washed with HeBs buffer (20 mM Hepes (sodium salt), 137 mM NaC1, 5 mM KC1, 0.8 mM NaHP04, 5.5 mM D-glUCOSe, pH 7.0) and transfected with 15 pg of plasmid pTG1563-CD8a161 using the calcium phosphate method (49). Four hours later, cells were treated for 1 min with 15% glycerol in HeBs. After washing with HeBs, cells were incubated for additional 48 h in 10 ml of complete Ham's F-12 medium and then placed in selective medium (nucleoside/deoxynucleoside-free Alpha medium (GIBCO) supplemented with 10% dialyzed FCS). Colonies began to appear after 10-15 days and were tested for soluble CD8a expression by metabolic labeling and immunoprecipitation. Amplification of CD8a-dihydrofolate reductase genes was carried out by subjecting the clone expressing the highest levels of CD8a161 to several rounds of selection in 5, 10, 15, and 50 nM of MTX (Sigma). Further increase of MTX concentration (up to 3 p M ) did not result in higher CDSal61 production.
Metabolic Labeling of CHO Cells and E. coli and Immunoprecipitation-Four x lo5 CHO cells/well were seeded in 24-well plates in nucleoside/deoxynucleoside-free Alpha medium with 10% FCS. The next day, cells were incubated for 1 h in cysteine-free Dulbecco's modified Eagle's medium (GIBCO) with 10% dialyzed FCS and then in the same medium (0.3 ml) containing 100 pCi (1 mCi = 37 MBq) of ~-[~~S ] c y s t e i n e (Amersham Corp.) for an additional 6 h. Supernatants were centrifuged to eliminate detached cells, supplied with l mM EDTA, 1 mM EGTA, and protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 pg/ml each of antipain, leupeptin, pepstatin, aprotinin, and trypsin inhibitor), and immunoprecipitation was carried out with OKT8 mAb adsorbed to protein A-Sepharose as described previously (50). Beads were boiled in Laemmli reducing sample buffer (51) and the eluted proteins analyzed by electrophoresis on a 10% acrylamide-SDS gel. Gels were fixed, treated with Amplify (Amersham), dried, and autoradiographed. For metabolic labeling of 5"CGACACCCGGAACTGGCTGGCCATCGCTGGTTGGGC-3',

5"TGGCGTAGATCTGGGCTACGCTGGCAGGAAGACCGG-3'
A. Pavirani, unpublished data. bacteria, an overnight culture at 37 "C in 5 ml of 2YT medium containing 1% glucose and 200 pg/ml ampicillin was washed, resuspended in 5 ml of the same medium (without glucose) containing 1 mM isopropyl 1-thio-0-D-galactopyranoside, 100 pg/ml ampicillin, and 60 pCi/ml ~-[~~S ] c y s t e i n e a n d incubated for an additional 3-4 h at 25 "C. Periplasmic proteins were extracted as described (52), immunoprecipitated with anti-CD8 mAbs, and analyzed on a 14% acrylamide SDS gel. A panel of anti-CD8 mAbs was used to immunoprecipitate purified recombinant CD8a-soluble forms as above and the CD8a left in the supernatant was measured using the T8 cell-free enzyme-linked immunosorbent assay kit (T-cell Science).
Large Scale Culture of CHO Cells and Purification of CDBa161-Roller bottles of 850 cm2 were seeded with 2-3 X lo7 CHO cells expressing CD8a161 in 100 ml of complete Alpha medium with 4% FCS containing 50 nM of MTX and cells grown to confluence. This culture could be maintained for 20 days by replacing the medium every 2 days. Cell supernatants were supplied with a mixture of protease inhibitors (see above), pooled, concentrated 5-fold on a tangential ultrafiltration Amicon system, and applied to a 20-ml Affi-Gel 10 column (Bio-Rad) coupled with 80 mg of OKT8 mAb. The column was washed sequentially with 200 ml of 10 mM phosphate buffer, pH 7.5, containing 0.5 M NaC1, 1 mM EDTA, 1 mM EGTA, and 1 mM phenylmethylsulfonyl fluoride and 200 ml of the same buffer plus 150 mM NaC1. Bound CD8a161 was eluted with 0.1 M NaC1, 50 mM ethanolamine buffer, pH 11.5, and the fractions were immediately neutralized with 0.5 M phosphate buffer, pH 4.5, pooled, and concentrated using a microconcentration unit (Centricon 10, Amicon). CD8a161 was further purified by gel filtration on a Sephadex G-150 or on a Superose 6 column (Pharmacia LKB Biotechnology Inc.) equilibrated with 0.1 M NaC1, 50 mM Tris-HC1 buffer, pH 7.4. Protein concentration was determined by the Bio-Rad protein assay, using human IgG as a standard, and purity was analyzed by SDS-PAGE.
Large Scale Production and Purification of CDBall4-Bacteria (200-400 ml) grown overnight a t 37 "C in 2YT medium containing 200 pg/ml ampicillin and 1% glucose were pelleted and resuspended in 10 times the original volume in the same medium without glucose and incubated a t 25 "C until reaching an Aew of 1.5. Expression was induced by addition of 1 mM isopropyl 1-thio-0-D-galactopyranoside, and the culture was continued for 4-h at 25 "C. Bacteria were pelleted, resuspended in 1/15-1/20 of the original volume in ice-cold 0.2 M Tris-HC1, pH 7.5, containing 0.5 M sucrose, mixed with an equal volume of ice-cold 1 mM EDTA, 0.1 mg/ml lysozyme, and incubated for 30 min on ice. After centrifugation for 30 min at 12,000 X g the supernatant (periplasmic extract) was recovered, and protease inhibitors added and passed through an immunoaffinity column of 20 ml, containing 80 mg of the anti-CD8 mAb lODll coupled to Affi-Gel 10 beads. Washing and elution conditions were identical to those used for CD8a161. Eluted fractions containing CD8a114 were pooled and dialyzed against 20 mM sodium acetate buffer, pH 5.5, and applied to a CM-cellulose Memsep Cartridges (Millipore). After extensive washing with the same buffer, CD8a114 was eluted with 30 mM NaC1,20 mM sodium acetate buffer, pH 5.5. The pH of the fractions was adjusted to 7.0 with 1 M Tris-HC1, pH 8, pooled, and concentrated using a Centricon 3 unit.
Amino Acid Sequencing-Purified CD8a161 and CD8a114 were electrophoresed under nonreducing conditions in a 10 and 14% acrylamide gel, respectively, and blotted onto Polybrene-coated glassfiber sheets (53). Amino acid amino-terminal sequence determination was carried out in an Applied Biosystem 470 gas-phase sequenator.
Determination of Soluble CDBa Stokes Radius-Samples of 0.1 ml containing purified CD8a161 and CD8a114 at the indicated concentrations were injected onto a Superose 6 HR 10/30 column (Pharmacia) connected to a fast protein liquid chromatography apparatus (Pharmacia) equilihrated with 100 mM NaCl, 20 mM Tris-HC1 buffer, p H 7.5. Proteins were monitored a t 280 nm and fractions analyzed by SDS-PAGE. Tyrosine was used as an internal control in each run. Stokes radii (Rs) of CD8a161 and CD8a114 were obtained using a linear relationship between V,-V, ( Vt, total volume, V,, elution volume) and Rs of the following proteins (used a t -1 mg/ml) (54): catalase (5.22 nm), aldolase (4.81 nm), tryptophan synthase p2 (4 nm), horseradish peroxidase (3.02 nm), chymotrypsinogen 01 (2.09 nm), and ribonuclease A (1.64 nm). Purified Fv D1.3 (55), used a t a concentration of 2 mg/ml, was a gift of G. Boulot (Pasteur Institute).
Sedimentation Analysis-The sedimentation coefficient at 20 "C in water (sZoW) of CD8a161 was determined by using a preformed 5-20% sucrose gradient in 100 mM NaC1, 20 mM Tris-HC1 buffer, pH 7.5. Samples of 0.15 ml (at 0.5-6 mg/ml) were centrifugated in parallel with protein markers in an SW 41 rotor (Beckman) a t 200,000 X g, for 20 h a t 4 "C. After centrifugation, the gradients were pumped through a Uvicord I1 (LKB) detector and proteins monitored at 280 nm. Fractions of 0.5 ml were collected and the percentage of sucrose measured with a refractometer (Zeiss). Protein peaks were analyzed by SDS-PAGE. The szo,W of CD8a161 was determined by interpolation in a linear standard curve established with the following protein markers of known s20,w (54, 56): catalase (11.15 S), aldolase (7.6 SI, chymotrypsinogen a (2.5 S), and ribonuclease A (2 S). To determine the partial specific volume (5)

RESULTS
Expression in CHO Cells of the Extracellular Region of CDBa in a Soluble Form-To express a soluble form of the human CD8a chain consisting of the extracellular region, CD8a161 (Fig. l), the corresponding cDNA was subjected to site-directed mutagenesis which introduced a termination codon at Ile-162, predicted to be the first amino acid of the transmembrane region (13). Cys-160 of the hinge region was also mu-L I g -l i k e H TM Cyto.  (13). Negatiue numbers designate the beginning of the signal peptides. The position of the cysteine residues (c) is shown. CD8a161 was obtained by changing, by oligonucleotide-directed mutagenesis, the codons of Cys-160 to a stop codon and Ile-162 to a Ser (s) codon. CD8a114 was constructed by inserting a stop codon at Lys-115 and by replacing the sequence corresponding to the CD8a signal peptide with that of the pectate lyase signal peptide (pelB L) by PCR Methods." amplification using the procedure described under "Materials and of the H u m a n CD8a Homodimer tated to Ser to avoid potential formation of illegitimate intrachain bonds with Cys-142 and to reduce the possibility of obtaining covalently linked multimers which have been observed for membrane-bound CD8 (15).
The modified CD8a cDNA was then subcloned into the pTG1563 dicistronic expression vector upstream of the dihydrofolate reductase gene. Expression of both CD8a161 and dihydrofolate reductase is driven by the adenovirus major late promoter to give a dicistronic mRNA. This results in a lower translation efficiency of dihydrofolate reductase compared with the upstream mRNA (59) and should allow a more efficient selection of clones expressing higher levels of CD8a161 a t early stages of the gene amplification process.
This construct was transfected into dihydrofolate reductase-negative DG44 cells by the calcium phosphate method and clones selected initially in medium deprived of nucleosides and deoxynucleosides. A number of growing clones (21 clones) were screened for secreted CD8a protein by immunoprecipitation with anti-CD8 mAbs from the supernatant of L -[~~S ] cysteine-labeled transfectants and SDS-PAGE analysis. Secreted CD8a was detected as a 25-kDa species under reducing conditions (data not shown), the expected size for CD8a161. One clone secreting the highest levels of CD8a161 was subjected to amplification a t progressively higher concentrations of MTX. Several rounds of selection (up to 50 nM MTX) allowed an overall improvement of approximately 70-fold in the level of secreted CD8a161 compared with unamplified cell line as assessed by quantitative immunoprecipitation and densitometric scanning of gel autoradiography (data not shown).
Biochemical Characterization of Purified CD8a161 "Supernatants of the cell population selected a t 50 nM MTX were concentrated by ultrafiltration, and CD8a161 was purified by affinity chromatography on an OKT8 mAb column followed by gel filtration on Sephadex G-150 or Superose 6 columns. The gel filtration step was necessary to remove a small fraction (about 10% of the affinity purified material) of aggregated CD8a161 and minor impurities which were eluted with the void volume. Typically, we obtained 2 mg of purified CD8a161 per liter of cell supernatants.
Analysis by SDS-PAGE, under nonreducing conditions of purified CD8a161 (Fig. 2 A ) shows a major species (approximately 70% of the material stained with Coomassie Blue) of a n apparent molecular mass of 47 to 50 kDa and a minor one of 25 kDa. When reduced, the upper band migrates as a 25-kDa polypeptide chain. This indicates that a minor fraction of the secreted CD8a is not disulfide-linked. It is unlikely that the presence of the latter species is the result of proteolytic cleavage of part of the hinge region, including Cys-142, since its apparent molecular mass under nonreducing conditions is similar to that of the reduced form. The broad appearance of CD8a161 under reducing and nonreducing conditions is most likely because of the presence of differently glycosylated molecules. Amino acid sequence analysis of the 50-47-kDa and 25-kDa species, purified from nonreducing gels, detected the amino-terminal sequence (SQFVR) predicted from the CD8a cDNA (13) (data not shown).
Gel filtration on Superose 6 shows that CD8a161 is eluted as a single homogeneous peak (Fig. 3A), indicating that no apparent dissociation of noncovalently bound molecule takes place under the conditions used. This was also true at dilutions of CD8a161 up to -20 pg/ml (= 0.5 p~) (data not shown), indicating that non-disulfide-linked CD8a161 molecules are stably associated.  This is further reinforced by the fact that most of these mAbs did not detect CD80161 in Western blot (data not shown).

The Hydrodynamic Properties of CD8a161 Indicate an Overall Elongated Shape-We determined the Stokes radius (Rs)
of CD8a161 by gel filtration on a Superose 6 column, using a standard curve established with proteins of known Rs. CD8a161 was found to have a Rs of 4.0 ? 0.1 nm (Fig. 3B), which is close to that of the marker tryptophan synthase p2, a globular protein dimer of 88 kDa (60), almost twice the molecular mass of CD8a161 homodimers estimated by SDS-PAGE. One possible explanation for this result is that the overall shape of the extracellular domain of CD8a homodimer is rather asymmetric.
To provide support for this hypothesis, we carried out centrifugation analysis in isokinetic sucrose gradients to determine the sedimentation coefficient ( s~~,~) of CD8a161. As shown in Fig. 4A, a standard curve constructed with protein markers of known gave 2.5 f 0.15 S for CD8a161. This is a rather low value for a globular protein of 50-47 kDa. Indeed, CD8a161 sedimented similarly to chymotrypsinogen a (25 kDa). Thus, together, gel filtration and sedimentation analyses are strongly suggestive of a molecule of pronounced asymmetric shape.
We next determined the partial specific volume ( V ) of CD8a161 according to the method described by Meunier et al. (57), which takes advantage of the differential migration of a macromolecule from the meniscus of sucrose gradients established in H20 and D20 (see "Materials and Methods" for details). By using this method (Fig. 4B)
the Svedberg equation, to calculate a molecular mass of approximately 51 k 4 kDa for CD8a161. This value is in close agreement with the estimation made on SDS-PAGE and the size of two CD8a161 polypeptide chains deduced from the primary structure plus their carbohydrate content (16). In addition, this result excludes that the elution volume observed for CD8a161 in gel filtration could be attributed to the recombinant molecule forming tetramers (e.g. two stably associated homodimers of CD8a161 making 90-100 kDa) composed of dimers having an overall globular shape. From the above data we could calculate a frictional ratio f / f o of 1.6 for CD8a161 (see "Materials and Methods"), a value indicative of an extended structure.
Expression of the Ig-like Domain of CDBa, CDBal 14, in E. coli: Purification, Gel Filtration, and Sedimentation Analyses-The amino-terminal domain of CD8a chain has been shown to fold as an Ig variable domain. It was therefore interesting to compare its hydrodynamic properties with those of the entire extracellular region of CD8a. This should allow us to understand which one of the CD8a161 regions is responsible for the observed hydrodynamic properties and to further validate the experiments reported above.
For the purpose of the present study and, at the same time, to obtain a recombinant molecule which could be advantageous for further structural studies, we attempted expression of the amino-terminal domain of CD8a in CHO cells. The CD8a cDNA was modified by introducing a termination codon a t Lys-115 using PCR amplification. This should result in expression of only the predicted Ig-like domain of CD8a, CD8a114. However, expression in CHO cells using the pTG1563 vector did not give detectable secreted protein.
Expression in E. coli was therefore tried by replacing the signal sequence present in CD8a114 with that of the bacterial protein pectate lyase (Fig. 1) as described under "Materials and Methods." This strategy has been used previously for obtaining secretion of eukaryotic proteins, including Fv frag- Determination of sedimentation coefficient (szo,,) and partial specific volume (V) of CD8a161 by sucrose gradient centrifugation. Panel A , samples in 0.15 ml containing purified CD8a161 at 0.5 mg/ml or protein markers, used at the same concentration, were loaded on top of a preformed 5-20% sucrose gradient (10 ml) in a 100 mM NaC1, 20 mM Tris-HC1 buffer, pH 7.5, and sedimented in a Beckman SW 41 rotor at 200,000 X g and 4 "C for 20 h. Protein absorbance was recorded at 280 nm. The top of the gradient corresponds to 0 ml. The s~, , ,~ of CD8a161 (open circle) was determined from a standard curve obtained with the following protein markers (closed circles) indicated in the order of increasing sz0, , , , : ribonuclease A (2 S ) , chymotrypsinogen a (2.5 S ) , aldolase (7.6 S ) , and catalase (11.15 S ) . Standard deviation of the curve was ? 0.15 S. Panel B , sucrose gradients and centrifugations were performed as in panel A . CD8a161 was loaded at a concentration of 0.5 mg/ml in a volume of 0.15 ml. The distance r of CD8a161 in DzO (rd = 1.23 ml) and in HzO (rh = 1.8 ml) was used for calculating the partial specific volume (5)  Large scale purification of CD8a114 by affinity chromatography on an anti-CD8 mAb (10D11.5) column followed by cation exchange chromatography gave 300-500 pg of purified protein/liter of bacteria. Monoclonal antibody 10D11.5 was used instead of OKT8 since the latter did not react with CD8a114 (see Table I).
The purified CD8a114 molecule migrates in SDS-PAGE as a 12-kDa species in both reducing and nonreducing conditions ( Fig. 2B), in close agreement with the size of the polypeptide chain predicted from the DNA sequence. Amino-terminal amino acid sequencing indicated that the pectate lyase B signal sequence had been cleaved at the predicted amino terminus of CD8a (data not shown).
CD8a114 was recognized by 10 of 11 anti-CD8 mAbs ( Table  I), suggesting that its conformation is close to the native one and that the epitope recognized by the unreactive mAb OKT8 is within or very near to the hinge region. Fig. 3C shows that CD8a114 was eluted from a Superose 6 column as a single homogeneous peak and had a Rs of 2.3 f 0.2 nm (Fig. 30). CD8a114 is eluted almost at the same position as chymotrypsinogen a (25 kDa) and as an Ig Fv ( VH + VL) dimer (-25 kDa, Rs: 2.2 f 0.2 nm). When analyzed by sedimentation on a 5-20% sucrose gradient at low protein concentration (10-20 Fg/ml), CD8a114 was detected as a single homogeneous peak with a s~, , ,~ of 2.3 f 0.15 S, almost identical to the value observed for the Ig Fv, 2.4 f 0.15 S (data not shown), further confirming that they share similar hydrodynamic properties. By assuming a t of 0.73 cm3/g (a typical value for most unglycosylated proteins) (56) a molecular mass of 22 kDa can be deduced for CD8a114, fully compatible with it being an homodimer. Thus, similar to the Ig Fv, CD8a114 appears to be a stable dimer since no monomer could be detected at fairly low protein concentrations (20 pg/ml -1 FM). This observation suggests therefore that the Ig-like domain may be sufficient per se to ensure a strong interchain interaction in the dimer formed by the entire extracellular region.
In addition, the above data indicate that the CD8a hinge region confers unique hydrodynamic properties consistent with an elongated structure of the CD8a homodimer.

DISCUSSION
To provide further insight into the structure and function of the CD8a co-receptor, we investigated its hydrodynamic properties and chain association requirements. This was made possible by the successful expression of soluble forms of the CD8a chain extracellular region, CD8a161, in CHO cells and of the Ig-like domain, CD8a114, in E. coli. Both recombinant molecules were readily purified in milligram amounts. Analysis by SDS-PAGE, gel filtration, and sedimentation on sucrose gradients indicated that they were both secreted as homodimers, the former being in large part (270%) disulfidelinked, whereas the latter, as expected (16), was not.
Several lines of evidence suggest that the structure of recombinant CD8a161 and CD8a114 is close to the native one. First, 10 mAbs tested by immunoprecipitation recognized the recombinant molecules. OKT8 mAb reacted only with CD8a161 and is presumably directed at an epitope in the hinge region. Second, the pattern of proteolysis of CD8a161 with different proteases7 closely resembled the pattern observed for the membrane-bound form (62). Third, the purified recombinant molecules consistently gave homogeneous peaks in both gel filtration and sedimentation analysis, and no spontaneous aggregates were detected.
The presence of a fraction of CD8a161 dimers non-disulfide-linked could not be attributed to proteolytic removal of part of the hinge (at least 20 residues) including Cys-142, during expression or purification. Non-disulfide-linked CD8a161 dimers might be caused by the absence of Cys-160 in our construct implying perhaps that membrane-bound CD8a may exist in different isoforms which use one or the other (or both) cysteines to form interchain disulfide bridges.
However, others have reported that a soluble extracellular region of human CD8a (CD8a162) expressed in CHO cells and containing both cysteines, paradoxically, did not form interchain disulfide bonds (16). A likely explanation for these observations is that, as we anticipated when designing our constructs, illegitimate intrachain disulfide bonds are formed which, surprisingly, may be favored in the soluble molecule. Similar to our finding, it was found that a soluble human CD8a, truncated at residue 146 (including only one cysteine of the hinge), contained a fraction of non-disulfide-linked homodimers (16). Together, these observations suggest that the hinge region possesses a high degree of flexibility. The presence of the membrane anchor in the intact molecule may, however, facilitate chain proximity, thus favoring correct and quantitative formation of disulfide bonds.
Covalent bonding of CD8a161 dimers may not be essential t o maintain association since monomers, which could be derived from the non-disulfide-linked molecules, were not detected by gel filtration and sedimentation analysis even a t micromolar protein concentrations. Likewise, CD8all4 dimers, at similar protein concentrations, were found to be stably associated. Although the behavior in solution of CD8a114 is in agreement with the crystallographic model indicating that the two Ig-like domains of CD8a form an Fvlike dimer via multiple contacts at the interfaces (16), its stability in the absence of the hinge region, observed here, suggests that the latter may only marginally contribute to dimer association and/or stabilization.
CD8a161 was found to have a Rs of 4.0 f 0.1 nm and a szow of 2.5 f 0.15 S, both values highly inconsistent with a globular protein having the molecular mass of the dimer estimated in SDS gel. By using these parameters and the experimentally determined partial specific volume of CD8a161, the molecular mass was found to be 51 kDa, in close agreement with the size of two polypeptide chains of 17.5 kDa with 14 0-linked carbohydrates (16), allowing an average of 1 kDa/carbohydrate chain. Furthermore, the hinge region appears to largely contribute to the hydrodynamic properties of CD8a since CD8a114 shows a Rs and s20,w similar to those of an Ig Fv dimer, whose three-dimensional structure is known (61).
A frictional ratio f/fo of 1.6 was calculated for CD8a161. However, this value must be corrected for the hydration coefficient (6), which may be high for a molecule of elongated structure and containing, at the same time, carbohydrate chains which tend to be highly hydrated. For CD8a161 the carbohydrate content was estimated to be approximately -30% of its total mass. Typical 6 values for most proteins can vary between 0.1 and 0.6 g of HsO/g of protein, but more extreme values of 0.8 have been reported (56). Thus, when a 6 between 0.6 and 0.8 was used, to take into account high carbohydrate content, we calculated an axial ratio of 6 to 5 for CD8a161, assuming a prolate ellipsoid (58). In this respect, a meaningful comparison can be made with the extracellular region of CD4 whose f / f o was also found to be 1.6, and the axial ratio was roughly 6, after correction for hydration (37). From these studies and from x-ray crystallographic data (33, 34, 37), the length of CD4 extracellular domain has been predicted to be -14 nm.
We could calculate for CD8al61 a maximal theoretical length of 18-20 nm for a prolate ellipsoid model (58), sufficient for the Ig-like domain to reach the ag domain of class I. This implies that, after subtracting the length of the Ig-like domain (2.5-3 nm), the hinge region can be approximately 15-17 nm long. Although these values must be regarded as indicative because of the uncertainties about the molecular model (prolate ellipsoid) and the hydration coefficient as-sumed, the primary structure of the hinge region and additional considerations may be help to explain these results. The presence of 10 prolines scattered throughout this region can be expected to disrupt formation of a-helices and/or pstrands often and suggests that the corresponding polypeptide chain may be, partially, a random coil. Indeed, circular dychroism spectra of CD8a161 (data not shown) tend to support this hypothesis. Furthermore, the hinge region contains several 0-linked sialylated carbohydrate chains, on threonine and serine residues, a feature found in a number of proteins of extended structure (63-66). At least in the case of mucins, which have been well studied, it was found that 0-linked oligosaccarides were required for maintaining an extended conformation of the polypeptide chain (63), and it was predicted that the segments of the molecule bearing carbohydrate chains should have an average length of -0.25 nm/residue. If part of the CD8a hinge region (48 residues) will have a length/ residue between this value and the theoretical value of 0.36 nm/residue, for a fully extended peptide backbone configuration, it would be possible for the entire extracellular domain to have a longitudinal dimension close to the calculated one.
Finally, the absence of electron density for the first 27 residues of the hinge region present in the crystallized molecule (16) indicates that they possess a high degree of mobility. Thus, it can be speculated that although the Ig-like domain provides stable association of the dimer, as also suggested by our observations, loose or no interchain contacts within a large section of the hinge region may favor a high conformational freedom of this region. This may facilitate CD8 flexibility and therefore the capacity to reach the ligand on the target cell.
As also reported by others for human (16) and rat (39) CD8a, we were unable to detect secreted CD8a114 in CHO cells, suggesting that the hinge region plays a role in facilitating the intracellular transport of CD8a. However, CD8a114 expressed with the bacterial pectate lyase B leader sequence or with its authentic leader sequence was secreted in the periplasmic space of E. coli or in supernatants of SF9 insect cells using recombinant baculovirus,8 respectively. The reason for the different fate of CD8a114 in CHO cells versus E. coli and insect cells is unclear, but it may be related to the phylogenically distant origin of the host cells whose secretory machinery may exert more or less stringent controls (e.g. transport, retention, degradation) on foreign proteins. That this may, as a consequence, result in the production of differently folded CD8a in different cellular environments is not supported by the fact that truncated versions of CD8a expressed in different hosts were equally reactive with a large panel of specific mAbs.
The expression in E. coli of the CD8a114 dimer in large quantities has applications for structural studies. This recombinant molecule needs no further treatment for deglycosylation and protease digestion as has been the case for a truncated CD8a146 expressed in CHO cells utilized for x-ray crystallographic analysis (16) and may prove useful for future attempts to co-crystallize CD8 with class I. Furthermore, this approach may be applied to the expression of the aminoterminal domain of CD8 a/@ dimer, the most abundant and perhaps more physiological form of CD8 expressed on T-cells. These recombinant molecules are currently under construction. of the Human CD8a Homodimer