Expression, Purification, and Functional Analysis of Murine Ectodomain Fragments of CD8αα and CD8αβ Dimers*

Soluble mouse CD8αα and CD8αβ dimers corresponding to the paired ectodomains (CD8f) or their respective component Ig-like domains (CD8) were expressed in Chinese hamster ovary cells or the glycosylation variant Lec3.2.8.1 cells as secreted proteins using a leucine zipper strategy. The affinity of CD8ααf for H-2Kb as measured by BIAcore revealed a ∼65 μm K d , similar to that of CD8αβf. Consistent with this result, CD8ααf as well as CD8αβf blocked the effector function of N15 T cell receptor transgenic cytolytic T cells in a comparable, dose-dependent fashion. Furthermore, both Lec3.2.8.1-produced and Chinese hamster ovary-produced CD8 homodimers and heterodimers were active in the inhibition assay. These results suggest that the Ig-like domains of CD8 molecules are themselves sufficient to block the requisite transmembrane CD8-pMHC interaction between cytolytic T lymphocytes and target cells. Moreover, given the similarities in co-receptor affinities for pMHC, the findings suggest that the greater efficiency of CD8αβ versus CD8αα co-receptor function on T cells is linked to differences within their membrane-bound stalk regions and/or intracellular segments. As recently shown for sCD8αα, the yield, purity and homogeneity of the deglycosylated protein resulting from this expression system is sufficient for crystallization and x-ray diffraction at atomic resolution.

CD8 has been shown to function in mediating signal transduction and adhesion on a subset of cells within the T cell compartment and is critically involved in the development of T cells expressing MHC class I-restricted T cell receptor (TCR) 1 (1). CD8 is encoded by two distinct genes, termed CD8␣ (or Lyt 2) and CD8␤ (or Lyt 3) and expressed on the cell surface as a mixture of disulfide-linked CD8␣␣ homodimers and CD8␣␤ heterodimers (2)(3)(4)(5)(6)(7). CD8␣ is a 34 -37-kDa transmembrane glycoprotein whose extracellular segment contains a compact 122amino acid (aa) N-terminal Ig-like domain and an extended 48-residue stalk region. A 28-aa cytoplasmic domain, including a cysteine motif responsible for interaction with p56 lck , follows a canonical hydrophobic transmembrane anchor. CD8␤ is a 32-kDa glycoprotein sharing a similar architecture as CD8␣ but with Ͻ20% sequence identity (4,5). The stalk regions of the CD8␣ and ␤ chains are quite different in length, with the CD8␤ stalk being 10 -13 residues shorter than that of CD8␣. Interestingly, the sialic acid content of O-linked glycans adducted to CD8␤ selectively decreases on thymocytes and activated T cells compared with that found on resting T cells. For its cell surface expression, CD8␤ requires association with the CD8␣ subunit, forming a CD8␣␤ heterodimer (6,8). Moreover, CD8 genes are selectively expressed. Although CD8␣␤ heterodimers are predominantly found on the surface of TCR␣␤ T cells and thymocytes, CD8␣␣ homodimers are additionally expressed on a subset of ␥␦ T cells, intestinal intraepithelial lymphocytes and natural killer cells (9,10). Hence, it is safe to conclude that the two sets of CD8 co-receptors subserve distinct functions.
The importance of CD8␣ in p56 lck -linked T cell activation and signaling has been defined by multiple studies (11,12). Although unable to bind p56 lck directly, other findings emphasize the contribution of CD8␤ to the efficacy of T cell recognition and its ability to broaden the range of antigen recognition (13,14). Several lines of evidence including our own show that CD8␣␤ is a more effective co-receptor than CD8␣␣ in enhancing the sensitivity to peptide antigens as well as alloantigens recognized by TCRs (15,16). A role for the cytoplasmic portion of the CD8␤ chain in enhancing Lck kinase activity and promoting T cell development has been suggested (17,18). In addition, at least for certain TCRs, the contribution of CD8␣␤ as a co-receptor may be due, in large part, to its extracellular components (16).
Monoclonal antibody blocking studies, cellular adhesion assays, and direct molecular interaction studies as well as the recent studies of the crystal structures of the Ig domain of the human CD8␣␣ (hCD8␣␣) dimer in complex with HLA-A2 and murine CD8␣␣ (mCD8␣␣) in complex with H-2K b have shown that the natural ligand of CD8␣ is the MHC class I molecule (19,20). These structures show unequivocally that one CD8␣␣ dimer binds to one pMHC complex. Moreover, these findings are consistent with mutational analyses indicating that the ␣3 loop of the MHC class I (MHCI) molecule is the major CD8␣ binding component. Although the CD8␣␤ heterodimer is thought to bind to the MHCI ␣3 region as well, currently little is known about the specific molecular details.
Given the importance of understanding the molecular interactions between the CD8 and MHC molecules, recombinant CD8 ectodomain fragments have been produced in a variety of systems (21-28) but without successful secretion of homogeneous products. To overcome this limitation, we expressed the Ig-like domain of mCD8␣␣ or the larger ectodomain fragment utilizing an engineered leucine zipper (LZ) system (29), which we previously applied to the expression of heterodimeric solu-ble TCR ␣ and ␤ chains. The CD8␣ Ig-like domain protein purified from Lec3.2.8.1 cells could be readily crystallized (20) in complex with the murine MHC class I H-2K b molecule loaded with VSV8 octapeptide (30). The same strategy was used to produce the Ig-like domain of soluble recombinant murine CD8␣␤ protein as well as the entire extracellular segment. The binding of proteins to pMHC was examined by surface plasmon resonance as well as by functional inhibition studies of killing activity using N15 TCR (29) transgenic (tg) cytolytic T cells. As detailed below, we offer an explanation for the paradox as to why the transmembrane CD8␣␣ versus CD8␣␤ co-receptors have such large differences in efficiency of facilitating TCR recognition (15,16) and yet bind to class I MHC with comparable affinities.

EXPERIMENTAL PROCEDURES
Molecular Design and Engineering of Secreted CD8 Co-receptors Using a Heterodimeric Coiled Coil Sequence-To promote the secretion of recombinant dimeric proteins in eukaryotic cells, the N-terminal extracellular segment of the mCD8␣ chain (residues 1-122) corresponding to the predicted Ig-like domain was fused via a flexible linker (aa residues 123-137) and a thrombin cleavage site (aa residues 129 -132) to either Acid-p1 or Base-p1 leucine zipper fragments (31) forming CD8␣-A and CD8␣-B, respectively (Fig. 1A). For this purpose, a 5Ј-primer encoding aa 1-5 of CD8␣ and a 3Ј-primer encoding aa 115-122 plus 10 amino acids of flexible linker were used for polymerase chain reaction of the CD8␣ Ig-like domain from the pHbAPr-1 neo/Lyt2 cDNA clone (3). The resulting DNA fragment was cloned into the pCR2 vector and sequenced. Subsequently, the error-free CD8␣-Ig-like DNA fragment was then digested with SpeI and BamHI and ligated with a BamHI-SpeI DNA fragment encoding Acid-p1 or Base-p1 peptides to form the CD8␣-A and CD8␣-B, respectively. The DNA fragments were then subcloned into the XbaI site of the pEE14 vector (CellTech Ltd, Berkshire, UK) to form plasmids, pEE14 CD8␣-A, and pEE14 CD8␣-B, respectively. The CD8␤ construct pEE14 CD8␤-A was generated in a similar fashion using a 5Ј-primer encoding residues 1-6 of murine CD8␤, a 3Ј-primer encoding CD8␤ aa 109 -115 and 10 amino acids of flexible linker and the CD8␤ Ig-like domain from pHbAPr-1 neo/Lyt3 cDNA clone (8) as a template. The error-free CD8␤ Ig-like DNA fragment then was digested with XbaI and BamHI and ligated with the BamHI-EcoRI DNA fragment encoding Acid-p1 to form the CD8␤-A. Subsequently, the DNA fragment was subcloned into the XbaI and EcoRI site of pEE14 vector to form plasmid, pEE14 CD8␤-A. After sequence verification, the pEE14 CD8␣-A plus pEE14 CD8␣-B cDNAs or pEE14 CD8␣-B plus pEE14 CD8␤-A cDNAs were pairwise co-transfected into Lec3.2.8.1 cells (32) to produce murine CD8␣␣-LZ or CD8␣␤-LZ proteins.
To avoid cysteine mispairing and incorrect disulfide bond formation, a construct encoding the soluble full-length extracellular domain of CD8␣␣ was terminated prior to cysteine residue 166. Moreover, cysteine 151 in the stalk region was mutated to a serine residue. A 5Ј oligonucleotide primer corresponded to aa 1-5 of CD8␣ and a 3Ј-primer encoding aa 148 -165 and 10 amino acids of flexible linker were used to polymerase chain reaction the CD8␣ f from the pHbAPr-1 neo/Lyt2 cDNA clone template. The DNA fragments generated were cloned into the pCR2 vector, and sequenced. The error-free CD8␣ f DNA fragment then was digested with SpeI and BamHI and ligated to the BamHI-SpeI DNA fragment encoding Acid-p1 or Base-p1 segments to form CD8␣ f -A and CD8␣ f -B, respectively. The DNA fragments were then subcloned into the XbaI site of pEE14 vector to generate the pEE14 CD8␣ f -A and pEE14 CD8␣ f -B plasmids. The construction of pEE14 CD8␤ f -A was generated in a similar fashion to pEE14 CD8␣ f -A, terminating immediately prior to cysteine 150. As with CD8␣, the cysteine at aa 137 in the murine CD8␤ stalk region was mutated to serine. The DNA fragment of the CD8␤ full-length extracellular domain was generated by polymerase chain reaction from the pHbAPr-1 neo/Lyt3 cDNA clone as a template using the 5Ј-primer encoding aa 1-6 of CD8␤ and the 3Ј-primer encoding aa 135-149 as well as 10 aa of flexible linker. The error-free CD8␤ f -A DNA fragment then was digested with XbaI and BamHI and ligated to the BamHI-EcoRI DNA fragment encoding the Acid-p1 to form the CD8␤ f -A. Subsequently, the DNAs were subcloned into the XbaI and EcoRI sites of pEE14 vector to form the plasmid pEE14 CD8␤ f -A. After sequence verification, the pEE14 CD8␣ f -A plus pEE14 CD8␣ f -B cDNAs or pEE14 CD8␣ f -B plus pEE14 CD8␤ f -A cDNAs were pairwise co-transfected into Lec3.2.8.1 cells to produce soluble murine CD8␣␣ f -LZ or CD8␣␤ f -LZ protein.
Expression of CD8 Ectodomain Fragments Using a Glutamine Synthetase Vector-To produce large quantities of soluble recombinant murine CD8␣␣-LZ protein in Lec3.2.8.1 cells, a method similar to that described in detail for soluble recombinant TCRs (33) was employed. 20 g of SalI linearized plasmid DNA pEE14 CD8␣-A and pEE14 CD8␣-B were used for transfection by a calcium phosphate precipitation method using a Transfection MBS kit (Stratagene) following the manufacturer's protocol. 48 h after transfection, the cells were trypsinized, resuspended into 10 ml of Glasgow minimal essential medium-supplemented containing 25 M methionine sulfoximine and cultured onto 96-well plates. Three to four weeks later, the growing clones were assayed for secretion of soluble CD8␣␣ homodimer by ELISA. In brief, 10 g/ml of the anti-leucine zipper antibody, 2H11, was coated onto Immulon plate (Dynatech) at room temperature for 2 h, and then the plates were blocked with 1% bovine serum albumin in borate-buffered saline at room temperature for 2 h. 50-l culture supernatants of individual clone were plated overnight at 4°C, mixed with 5 g/ml of biotinylated anti-CD8␣ mAb 53.6.72 (34) for 2 h, and developed with horseradish peroxidase-conjugated streptavidin (Sigma). The positive clones were picked and transferred to 24-well plates. Subsequently, the highest secretors were ranked by rescreening the supernatants using an indirect capture with the anti-velcro mAb 2H11 (29) on BIAcore (Pharmacia Biosensor) (33). The identical ELISA was used for detection of CD8␣␣ f -LZ proteins. The productions of murine CD8␣␤-LZ and CD8␣␤ f -LZ were generated similarly using two pairs of SalI linearized plasmid DNAs, pEE14 CD8␤-A/pEE14 CD8␣-B and pEE14 CD8␤ f -A/ pEE14 CD8␣ f -B, respectively, for transfection into Lec 3.2.8.1 cells. The stable clones producing CD8␣␤-LZ or CD8␣␤ f -LZ were identified by the ELISA method described above except using 5 g/ml of biotinylated anti-CD8␤ mAb YTS156 (35) as the detecting antibody. Transfections of pEE14 CD8␣ f -A/pEE14 CD8␣ f -B and pEE14 CD8␣ f -A/pEE14 CD8␤ f -B were performed in CHO cells as well.
BIAcore Studies-All binding studies were performed with PBS/ Tween 20 (0.005%) on a BIA1000 surface plasmon resonance biosensor (BIAcore Inc.). To study the CD8-H-2K b interaction, we took advantage of the CD8 leucine zipper constructs by capturing the CD8-leucine zipper molecules with an anti-leucine zipper mAb coupled to the sensor chip by standard N-hydroxysuccinimide/N-ethyl-(dimethylaminopropyl) carbodiimide chemistry following standard procedures (BIAcore). This approach aligns all CD8 molecules in a similarly ordered manner, making them accessible for the H-2K b interaction. Because 2H11 has a relatively fast dissociation rate, we employed an additional mAb, termed 13A12, generated in our laboratory (data not shown). 20 l of 13A12 at 100 g/ml in 10 mM NaAc, pH 4.5, were immobilized on a CM5 sensor chip at a flow rate of 5 l/min resulting in ϳ5000 RU coupled. CD8␣␣ f -LZ and H-2K b samples were buffer exchanged into PBS, which is equivalent to the buffer in the reservoir. 20 l of CD8␣␣ f -LZ (1 M) were injected onto the 13A12 surface at a flow rate of 10 l/min. After a 5-min dissociation period, 5 l/min of H-2K b -VSV8 (64 M) were injected onto the 13A12-CD8␣␣ f -LZ surface. This procedure was repeated using H-2K b -VSV8 concentrations of 1-32 M. The specificity of the binding was assured by H-2K b injections (1-64 M) directly onto the 13A12 surface (data not shown). The specific RU for H-2K b -CD8␣␣ f -LZ binding at equilibrium were determined by subtracting background RU (H-2K b on the 13A12 surface) from the total RU. The K d was derived from a Scatchard plot RU/concentration versus RU and linear regression analysis (Kaleidagraph software). The experiment was repeated three times with CD8␣␣-LZ and CD8␣␣ f -LZ, and only once with CD8␣␤-LZ, CD8␣␤ f -LZ, CHO-CD8␣␣ f -LZ, and CHO-CD8␣␤ f -LZ. For all BIAcore studies, the H-2K b ␣-chain (residues 1-274) and the murine-␤2microglobulin (residues 1-99) were expressed as separate inclusion bodies in Escherichia coli and refolded in the presence of VSV8 peptide as described earlier (30).
Cytotoxicity Assays-To obtain N15 CTL, splenocytes were isolated from Rag2 Ϫ/Ϫ mice expressing a transgenic N15 TCR (36) and stimulated to obtain a short term cytotoxic T cell line as described previously (37). For stimulation, the splenocytes were co-cultured in 24-well plates (5 ϫ 10 6 cells/well) with irradiated N1 cells (5 ϫ 10 5 cells/well). N1 cells constitutively present the cognate VSV8 peptide as an expressed minigene. After 5-7 days of culture with rat-derived concanavalin A supernatants, N15TCR ϩ CTLs were harvested and used as effector cells in the cytotoxicity assay. 1 ϫ 10 6 EL-4 target cells were labeled with 80 Ci of sodium [ 51 Cr]chromate for 1 h at 37°C. After four washings with RPMI complete, the K b -expressing EL-4 cells were resuspended at 50,000 cells/ml in complete RPMI and transferred into 96-well U bottom plates (100 l/well). To determine a suboptimal dose of peptide for inhibitory experiments, peptides were added at concentrations between 10 Ϫ6 and 10 Ϫ12 M in a final volume of 100 l of complete RPMI and incubated for 1 h. Finally, CTLs (100 l) were added at an E:T ratio of 10:1 or 5:1, and 51 Cr release was determined using standard conditions (36). In the actual blocking experiment, EL-4 cells were pulsed with VSV8 at 2 ϫ 10 Ϫ10 M. As a negative control either SEV9 (irrelevant peptide) or only PBS was added to the wells. Anti-CD8␣ 53.6.72 mAb was used as positive control, whereas a two-domain recombinant CD4 (rCD4) (38) was used as a negative control for blocking experiments. 1 M 53.6.72 mAb was added to CTL cells for 10 min at room temperature. CD8␣␣-LZ, CD8␣␤-LZ, CD8␣␣ f -LZ, CD8␣␤ f -LZ, CHO-CD8␣␣ f -LZ, and CHO-CD8␣␤ f -LZ were concentrated using a Centricon-10 (Amicon), and 27.5-220 M (final concentration) of CD8 protein or 27.5-220 M of rCD4 were added to the antigen presenting cells 10 min prior to addition of 5 ϫ 10 4 N15 CTL (in 100 l)/well. To determine background release, 100 l of RPMI completed medium was added to 51 Cr-labeled targets. The maximal 51 Cr release was quantitated by adding 100 l of 1% Triton X to 51 Cr-labeled target cells. After 4 h at 37°C, the plates were centrifuged, and 50% of the well solution was removed and counted in a ␥-counter. To this end, a set of CD8 co-receptor constructs was produced as depicted schematically in Fig. 1. For example, the N-terminal extracellular segment of the mCD8␣ chain (residues 1-122) corresponding to the predicted Ig-like domain was fused via a flexible linker (residues 123-137) and a thrombin cleavage site to either an Acid-p1 or Base-p1 30-aa-long leucine zipper fragment to form CD8␣-Acid (CD8␣-A) and CD8␣-Base (CD8␣-B), respectively (Fig. 1A). Subsequently, cDNAs encod- ing CD8␣-A and CD8␣-B were transfected into Lec3.2.8.1 cells, and clones secreting the CD8␣␣ ectodomain were screened by ELISA. mCD8␣␣-LZ protein was then affinity purified from the producer clone CD8-22-1 using the anti-leucine zipper mAb 2H11. As shown in Fig. 2A, under either nonreducing conditions (NR, lane 1) or reducing conditions (R, lane 2), the recombinant mCD8␣␣-LZ protein runs in the Coomassie-stained SDS-PAGE gel as two bands corresponding to CD8␣-A and CD8␣-B monomer components with apparent molecular masses of 29 and 25 kDa, respectively. The difference in mobility of the CD8␣ chains is a consequence of the divergent charges within the appended acid or basic leucine zipper sequences. By this analysis, the affinity purified mCD8␣␣-LZ material is ϳ90% pure with a 1:1 ratio of CD8␣-A and CD8␣-B. Although not shown, gel filtration chromatography on Sephadex 75 demonstrated that mCD8␣␣-LZ was dimeric and readily removed the trace amount of higher molecular mass contaminants and/or aggregates. Moreover, the expected Nterminal sequence of mCD8␣␣-LZ was verified by amino acid sequencing (not shown).
Expression of the Full-length sCD8␣␣ and sCD8␣␤ Ectodomains-The stalk region that connects the CD8␣ and CD8␤ Ig domain to the cell membrane contains ϳ44 aa residues in the ␣ chain and ϳ35 aa residues in the ␤ chain (depending on species). To assess the functional contribution of O-linked glycans in the stalk region of CD8 to MHC class I binding, we expressed the soluble full-length sCD8␣␣ f and sCD8␣␤ f in wt CHO cells or the glycosylation mutant Lec3.2.8.1. The former expression system synthesizes proteins with full-length glycan adducts. In contrast, Lec3.2.8.1 cells produce glycoproteins with all N-linked carbohydrates truncated to the Man 5 form and Olinked carbohydrate truncated to a single GalNAc (32).
Several reports indicated that the formation of disulfidelinked homodimers is extremely inefficient when the entire native ectodomain of either hCD8␣␣ or rat CD8␣␣ was expressed. The resulting products contained a mixture of disulfide-and nondisulfide-linked CD8 homodimers, monomers, and aggregates (21-24, 26, 41). To avoid this complexity, the extracellular segment of the mCD8␣ chain (residues 1-165) N-terminal to the last extracellular cysteine was fused to either an Acid-p1 or Base-p1 to form CD8␣ f -A and CD8␣ f -B, respectively (Fig. 1C). Furthermore, the cysteine residue at aa 151 of CD8␣ was converted to serine by polymerase chain reaction mutagenesis to obviate disulfide scrambling. Both cDNAs were engineered into the pEE14 vector system and expressed in CHO or Lec3.2.8.1 cells. To express the full-length mCD8␣␤ f -LZ, the CD8␤ cysteine at position 137 was similarly mutated to serine, and the extracellular segment of the mCD8␤ chain (residues 1-149) was fused to an Acid-p1 forming CD8␤ f -A. The pEE14 CD8␣ f -B and pEE14 CD8␤ f -A plasmids were co-transfected to generate stable cell lines, producing mCD8␣␤ f -LZ as described under "Experimental Procedures." Lec3.2.8.1 or wt CHO secreted mCD8␣␣ f -LZ and mCD8␣␤ f -LZ were affinity purified from culture supernatants using the 2H11 mAb.
As shown in Fig. 3, under nonreducing conditions, the wt CHO-produced recombinant mCD8␣␣ f -LZ runs as one broad band of apparent molecular mass of 30-37 kDa (Fig. 3, lane 1). By contrast, the Lec3.2.8.1-produced recombinant mCD8␣␣ f -LZ protein runs as a discrete band with a molecular mass of 32 kDa (lane 3). The wt CHO-produced recombinant mCD8␣␤ f -LZ runs as two rather broader bands representing CD8␣ f -B and CD8␤ f -A subunits at molecular masses 35 and 30 kDa, respectively (Fig. 3, lane 2), whereas Lec3.2.8.1-produced recombinant mCD8␣␤ f -LZ runs as discrete bands of molecular masses 32 and 29 kDa (lane 4). The difference in the molecular masses of these products reflects the different nature of the glycosylated adducts. The affinity purified mCD8␣␤ f -LZ material is quite pure and shows a 1:1 ratio of CD8␣ f -B and CD8␤ f -A in Coomassie gel (Fig. 3). The affinity puri- fied mCD8 molecules bear native epitopes as measured by BIAcore analysis using multiple mAbs reactive with CD8 co-receptors found on T lymphocytes (see below).
BIAcore Affinity Measurements-The interactions between soluble CD8 and MHC class I H-2K b (complexed to VSV8) were examined using an SPR biosensor, which allows direct measurement of kinetic interactions between immobilized and solution-phase molecules (42). The affinity for H-2K b -VSV8 was measured using all six different CD8 protein derivatives  Table I summarizes the affinities of the different CD8 proteins for pMHCI. As shown, CD8␣␣ f -LZ has an affinity of 35-75 M for H-2K b -VSV8, whereas that of CD8␣␤ f -LZ is comparable at 67 M. Moreover, CD8 constructs lacking the C-terminal stalk regions (CD8␣␣-LZ and CD8␣␤-LZ) have comparable affinity to the full-length CD8␣␣ f and CD8␣␤ f , implying that the Ig domain is necessary and sufficient for MHC binding. Moreover, CD8␣␣-LZ and CD8␣␤-LZ affinities of 57-86 and 30 M, respectively, are quite similar.
CD8 on the T cell surface is known to be heavily O-linked glycosylated in the stalk region and to possess a sialylated Ig domain. Thus, it was of interest to investigate whether the fully glycosylated CHO-produced full-length CD8 proteins differ in their affinities for pMHC relative to Lec3.2.8.1-derived proteins. However, CHO-CD8␣␣ f -LZ showed an affinity of 64 M, and CHO-CD8␣␤ f -LZ showed an affinity of 22 M. Based on these results in comparison to the Lec 3.2.8.1. derivatives, there appears to be no difference in affinity (within a factor of Ͻ3) for K b resulting from the distinct glycosylation pattern of the CD8 protein.
sCD8␣␣ and sCD8␣␤ Are Able to Block the Cytotoxic Activity of N15 CTL-Next we tested the ability of sCD8 molecules to functionally inhibit the cytolytic activity of class I MHC-dependent N15 CTL derived from N15 tg Rag2 Ϫ/Ϫ H-2 b mice. In these experiments, CTLs were added at an E:T ratio of 5:1 to 51 Cr-labeled EL-4 cells pulsed with 2 ϫ 10 Ϫ10 M VSV8 peptide, and 51 Cr release was determined in the presence or absence of CHO-CD8␣␣ f -LZ or CHO-CD8␣␤ f -LZ proteins. Anti-CD8␣ 53.6.72 (1 M) was used as positive control, and a two-domain recombinant CD4 (rCD4) (38) was used as a negative control for the blocking experiments. Fig. 5A shows the results of one representive experiment when the CD8 ectodomain dimers were added to EL-4 cells for 10 min prior to the cytotoxicity assay. Both CHO-CD8␣␣ f -LZ and CHO-CD8␣␤ f -LZ proteins are able to block the cytotoxic activity of N15 CTL in a dosedependent fashion, inhibiting 81.1 and 73.7%, respectively, at a 220 M concentration. The concentration of soluble CD8 necessary for 50% functional inhibition (55-110 M) correlates with the affinities of the CD8 dimers for class I MHC as measured by BIAcore. At 110 M soluble CHO-CD8␣␤ f -LZ inhibited slightly better than CHO-CD8␣␣ f -LZ at the same molar concentration. Note that anti-CD8␣ antibody 53.6.72 inhibited killing by about 95%, whereas rCD4 showed a negligible percentage of inhibition.
To address whether the Ig-like domains of the CD8 dimer alone were able to mediate functional CTL inhibition, we compared the effects of CD8␣␣-LZ and CD8␣␤-LZ with those of CHO-CD8␣␣ f -LZ and CHO-CD8␣␤ f -LZ. Fig. 5B shows results for all four CD8 ectodomain dimer fragments at 220 M. The CD8 Ig-like domain per se, in absence of the stalk region and complex glycans, is able to block cytolytic function. Compared with the CHO derivatives, Lec 3.2.8.1 CD8␣␣-LZ only inhibited 36.5%, whereas Lec 3.2.8.1 CD8␣␤-LZ inhibited 48.5%. Thus, the level of CTL inhibition with CD8␣␣-LZ and CD8␣␤-LZ is somewhat lower than that observed with the CHO-produced full-length ectodomain CD8␣␣ f -LZ and CD8␣␤ f -LZ protein.
Whether the CD8 stalk region in the CD8␣␣ f -LZ and CD8␣␤ f -LZ derivatives contributes to the functional inhibition by affecting additional components of the CTL or target cell surface is unknown. It is also possible that the greater size of the CHO-CD8␣␣ f -LZ and CHO-CD8␣␤ f -LZ protein relative to the Lec3.2.8.1 CD8␣␣-LZ and CD8␣␤-LZ proteins creates an additional steric inhibition that facilitates the observed functional blockade. DISCUSSION Various CD8 co-receptor ectodomain dimers were produced in soluble form to investigate the basis for differences in the functions of ␣␣ and ␣␤ dimers and the contribution of their respective Ig-like domain and stalk region segments. The yield of immunoaffinity purified protein derived from each construct   (20). The high level expression of mCD8␣␤ should allow a detailed structural analysis of CD8␣␤/K b alone or in complex with a class I MHC-restricted TCR as well.
When the CD8␣␣ dimer containing both cysteine residues in the stalk region was expressed in early studies (22-24, 26, 27, 41), the majority of the recombinant protein was monomeric, caused by intrachain pairing of the two cysteine residues in the stalk region (21). In contrast, the results of SDS-PAGE (Figs. 2 and 3) and gel filtration here indicate that the subunit products of CD8␣␣ f -LZ and CD8␣␤ f -LZ are in a molar ratio of 1:1 with minor or no aggregates. This result suggests that the mutation to serine of the cysteine residue at aa 151 of CD8␣ or aa 137 of CD8␤ in the stalk region avoided the disulfide bond scrambling completely. The yield of immunoaffinity purified full-length ectodomain using the anti-leucine zipper mAb, 2H11, was 8 -15 mg/liter for CD8␣␣ f -LZ and CD8␣␤ f -LZ expressed in CHO or Lec3.2.8.1 cells. Moreover, the affinity purified mCD8 molecules are in a native configuration as measured by BIAcore using five anti-CD8␣ mAbs and four anti-CD8␤ antibodies.
The stalk region connecting the Ig-like domain of mCD8 to the membrane consists of ϳ44 aa in the ␣ chain and ϳ35 aa in the ␤ chain (Fig. 1A). Each connecting peptide must be extended because it is rich in proline residues and contains a number of O-linked glycans. The biological function of these stalks is unclear as is the precise orientation of the TCR chains relative to the co-receptor subunits. However, the sialic acid content of CD8␤ O-linked glycans decreases significantly on thymocytes and activated T cells compared with the levels found on resting T cells (43)(44)(45), a phenomenon not observed for the CD8␣ chain. The results imply that the ability of CD8␤ to vary its overall charge and glycan size may have important consequences for CD8-TCR interaction. To examine the functional contribution of the O-linked glycan in the stalk region of CD8, we expressed the soluble full-length sCD8␣␣ and sCD8␣␤ in wt CHO cells or in the glycosylation mutant Lec3.2.8.1, allowing a direct comparison of the importance of the complex sugars to the MHC binding function of CD8. A cytotoxicity assay utilizing the N15 CTL derived from N15 tg Rag2 Ϫ/Ϫ H-2 b mice then assessed the ability of soluble CD8 molecules to inhibit the cytolytic activity of class I MHC-dependent N15 CTL. The results of two independent experiments show that CHO-produced CD8␣␣ f -LZ and CHO-CD8␣␤ f -LZ proteins are able to block the cytotoxic activity of N15 CTL in a dose-dependent fashion, with no significant difference between the CHO-CD8␣␣ f -LZ and CHO-CD8␣␤ f -LZ proteins. This result implies that the binding of soluble CD8␣␣ or CD8␣␤ to MHC class I molecules on target cells must be equivalent. These results are consistent with the BIAcore analysis showing similar K b binding for CD8␣␣ and CD8␣␤ dimers (Table I). Moreover, both the Lec3.2.8.1-produced CD8␣␣ and CD8␣␤ Ig-like domains, without the stalk region, are able to block cytolytic function.
The affinity of CD8 to H-2K b -VSV8 has been measured using all six different CD8 constructs produced (Table I). CD8␣␣ f -LZ has an affinity of ϳ62 M (n ϭ 4) for H-2K b -VSV8, and CD8␣␤ f -LZ show a slightly higher affinity of ϳ44 M (n ϭ 2). CD8 constructs lacking the C-terminal stalk regions, CD8␣␣-LZ and CD8␣␤-LZ, do not appear to have lower affinity (ϳ67 M (n ϭ 3) and 30 M (n ϭ 1)) than their full-length protein counterparts (Table I). These affinities are comparable with those determined previously by Garcia et al. (46). In that report, the full-length ectodomains of mCD8␣␣ and mCD8␣␤ expressed with a histidine tag in a Drosophila system reveal a moderate affinity (ϳ39 and ϳ11 M, respectively) to MHC class I molecules loaded with different antigenic peptides, with the CD8␣␤ heterodimer having a slightly better affinity. However, the results are substantially different from the affinity measured by Wyer et al. (28). Utilizing the E. coli-expressed recombinant Ig-like domain of hCD8␣␣, the affinity of hCD8␣␣ to pMHCI was determined to be greater than 200 M (28). Species differences, allelic differences, the absence of the mucin-like stalk region for sCD8, and/or the presence of aggregated materials might explain these discrepancies. In the present expression system, Lec3.2.8.1-produced CD8 protein shows very little difference from the CHO-produced CD8. These findings imply no substantial difference in affinity because of the glycosylation pattern of the stalk region of the CD8␣␣ or CD8␣␤. Furthermore, they show that greater affinity is not the basis by which the CD8␣␤ heterodimer has a much more significant effect relative to the CD8␣␣ homodimer in influencing TCR/ pMHC binding on the T cell surface (14 -16, 47-50).
The overall structure of the mCD8␣␣/VSV8/H-2K b complex FIG. 5. Soluble recombinant CD8␣␣ and CD8␣␤ dimer fragments functionally block the cytotoxic activity of N15tg CTL to equivalent degrees. A, indicated concentrations of wt CHO-produced mCD8␣␣ f -LZ and mCD8␣␤ f -LZ were added to a cytotoxic assay using N15 CTL effectors generated from splenocytes of N15 TCR transgenic Rag2 Ϫ/Ϫ mice and EL4 cells pulsed with VSV8 peptide (2 ϫ 10 Ϫ10 M) as target cells at an E:T ratio of 5:1. The ability of anti-CD8␣ mAb 53.6.72 (positive control) and a two-domain rCD4 (negative control) to inhibit lysis is indicated by the solid arrow and open arrow, respectively. B, 220 M concentrations of CHO-CD8␣␣ f -LZ, CHO-CD8␣␤ f -LZ, CD8␣␣-LZ, and CD8␣␤-LZ as well as two-domain recombinant CD4 were added into the above cytotoxicity. determined crystallographically reveals a single VSV8/H-2K b molecule interacting with one mCD8␣␣ homodimer (20). The asymmetric arrangement of the ␣3 and ␤ 2 M domains of the H-2K b molecule creates a concave shape underneath the H-2K b ␣2 domain, suitable for accommodating a mCD8␣␣ dimer. This CD8␣␣ dimer binds in an antibody-like manner to the protruding H-2K b ␣3 CD loop, using all six CDR-like loops for the interaction. The binding to H-2K b is asymmetric; in particular, the N terminus of CD8␣1 (the "top" subunit) is buried between the ␤ 2 M and the ␣3 domain of the MHC molecule, in contrast to its counterpart in CD8␣2 (the "bottom" subunit), which is exposed to solvent as well as involved in crystal contacts. A similar picture was reported for the human CD8␣␣ interaction with HLA-A2 (19).
Currently, the details of molecular interaction of CD8␣␤ with MHC class I is unknown. It was proposed by Gao et al. (19), based on scoring favorable electrostatic interactions, that the hCD8␤ subunit of hCD8␣␤ heterodimer may replace the bottom subunit of the hCD8␣␣ homodimer. Recently, the finding that coexpression of the mutant R4A of hCD8␣ and wt hCD8␤ on CD8␣␤ heterodimers on COS cells failed to facilitate MHC class I molecular binding (25) leads to the conclusion that CD8␣, not CD8␤, provides the majority of contacts with MHC class I. However, replacing either of the mCD8␣ subunits in the mCD8␣␣ homodimer structure with a mCD8␤ model by molecular graphic modeling and visual inspection of these models did not allow us to infer the localization of CD8␤. Nevertheless, the mCD8␣ stalk consists of 29 residues including seven potential O-glycosylation sites prior to the first interchain disulfide bond. The corresponding N-terminal motif in the CD8␤ stalk has only 22 residues, containing two prolines and five potential O-glycosylation sites. Given these differences, it is reasonable to suggest that the longer CD8␣ subunit stalk rather than that of CD8␤ prefers to extend a greater distance so that CD8␣ lies in a position analogous to CD8␣2, whereas CD8␤ replaces the top CD8␣1 subunit position. In this proposed orientation, the CD8␣␤ co-receptor would be thermodynamically more efficient. The CD8␤ stalk is shorter and the CD8␣ is longer, thus offering a presumably preconfigured co-receptor for interaction with MHC class I. This preconfiguration may be largely responsible for CD8␣␤ co-receptor efficiency. Experiments aimed at testing this possibility are in progress.