Engineering and Characterization of a Stabilized α1/α2 Module of the Class I Major Histocompatibility Complex Product Ld*

The major histocompatibility complex (MHC) is the most polymorphic locus known, with thousands of allelic variants. There is considerable interest in understanding the diversity of structures and peptide-binding features represented by this class of proteins. Although many MHC proteins have been crystallized, others have not been amenable to structural or biochemical studies due to problems with expression or stability. In the present study, yeast display was used to engineer stabilizing mutations into the class I MHC molecule, Ld. The approach was based on previous studies that showed surface levels of yeast-displayed fusion proteins are directly correlated with protein stability. To engineer a more stable Ld, we selected Ld mutants with increased surface expression from randomly mutated yeast display libraries using anti-Ld antibodies or high affinity, soluble T-cell receptors (TCRs). The most stable Ld mutant, Ld-m31, consisted of a single-chain MHC module containing only theα1 andα2 domains. The enhanced stability was in part due to a single mutation (Trp-97 → Arg), shown previously to be present in the allele Lq. Mutant Ld-m31 could bind to Ld peptides, and the specific peptide·Ld-m31 complex (QL9·Ld-m31) was recognized by alloreactive TCR 2C. A soluble form of the Ld-m31 protein was expressed in Escherichia coli and refolded from inclusion bodies at high yields. Surface plasmon resonance showed that TCRs bound to peptide·Ld-m31 complexes with affinities similar to those of native full-length Ld. The TCR and QL9·Ld-m31 formed complexes that could be resolved by native gel electrophoresis, suggesting that stabilized α1/α2 class I platforms may enable various structural studies.

The class I major histocompatibility complex (MHC) 4 encodes a cell surface glycoprotein that presents antigenic pep-tides to CD8 ϩ cytotoxic T lymphocytes. The heterotrimeric complex of peptide-MHC consists of a polymorphic heavy chain (ϳ45 kDa) with three domains (␣ 1 , ␣ 2 , and ␣ 3 ), an invariant light chain called ␤2-microglobulin (12 kDa), and an 8-to 10-amino acid peptide. The peptide binds within a cleft formed by the membrane distal ␣ 1 and ␣ 2 domains. The peptide, together with helices from the ␣ 1 and ␣ 2 domains, are recognized by the T-cell receptor (TCR). The MHC locus is the most highly polymorphic gene locus known, with hundreds of different MHC alleles at each of three class I MHC loci: in humans, HLA-A, HLA-B, and HLA-C and in mice, H-2K, H-2D, and H-2L (although H-2L is absent in many strains of mice) (1). This allelic variation has important immunological consequences, affecting peptide binding, thymic repertoire selection, and allograft rejection (2). Numerous studies have shown that some human MHC alleles are associated with disease susceptibility or resistance (3,4). Thus, there is considerable interest in understanding the diversity of structures and the peptide-binding features represented by this class of proteins.
Polymorphisms of class I MHC proteins are primarily localized to the peptide binding pocket, where the cleft is formed by two parallel ␣ helices and the floor is formed by a ␤-pleated sheet (5)(6)(7). Amino acid differences influence the shape of the pocket as well as the set of peptides that a particular MHC allele will bind. MHC class I genes have evolved many allelic variants to protect the population against pathogens which have developed mechanisms to evade the immune system (2). Collectively, MHC polymorphisms have yielded a diverse array of proteins with varying degrees of stability. Less stable MHC complexes have been difficult to study biochemically and structurally. Various efforts have been made to generate single-chain forms of class I complexes, with some success (reviewed in Ref. 8). For example, a single-chain trimer consisting of peptidelinker-␤2m-linker-heavy chain has been shown to be expressed as a stable complex on the surface of target cells (9).
Although many proteins have been amenable to engineering by in vitro directed evolution and peptide-MHC complexes have been displayed on phage (10 -12) or yeast (13), few studies have used directed evolution approaches to engineer more stable MHC molecules (14 -16). In this regard, H-2L d has been shown through both biochemical and structural studies to be less stable than many other MHC complexes. For example, L d is expressed on the surface of splenocytes at 3-to 4-fold lower levels than the class I products K d and D d (17,18). Consistent with this difference, Beck et al. (19) showed that L d complexes exhibited slower intracellular processing and weaker association with ␤2m compared with K d and D d . A chimera combining the N-terminal D d domain (␣1/␣2) and the C-terminal L d domain (␣3) exhibited the same rate of intracellular processing and level of association with ␤2m as D d . This result suggests that the ␣1 and/or ␣2 domains influence the stability of the L d complex (19). More recently, using multiple chimeras of L d and L q , a single amino acid at residue 97 of the L d complex was shown to be responsible for decreased cell surface expression of L d and its lower affinity for ␤2m (20).
Two crystal structures of L d , complexed with an endogenous peptide p29 (YPNVNIHNF) (21) or with a mixture of peptides (22) have been solved, providing insight into the weaker association of ␤2m and the generally low affinity of peptides for L d . Balendiran et al. (21) suggested that the structural instability of the L d -␤2m complex is due to the unique orientation of the ␣ 1 /␣ 2 domains relative to ␤2m and that this orientation causes a loss of productive contacts between ␣ 1 /␣ 2 and ␤2m. For example, a comparison of the intermolecular interactions, van der Waals, and hydrogen bonds between ␣ 1 /␣ 2 and ␤2m for three different MHC complexes, K b , D b , and L d revealed 17, 21, and 6 contacts, respectively. Structural features of the peptide binding cleft influence peptide selection and peptide binding affinity, and these properties can also influence the stability of the peptide-MHC complex. The L d -␤2m protein binds octamer or nonomer peptides with anchor residues of proline at peptide position 2 (P2) and leucine, methionine, or phenylalanine at the C-terminal position (23). However, peptides that lack proline at P2 have also been identified as L d -binding peptides (e.g. tum-, TQNHRALDL and p2CA, LSPFPFDL, a peptide recognized by the 2C TCR used in the present studies) (24,25). The peptide binding pocket for proline exhibits less shape complementarity than other MHC alleles with a proline anchor residue at P2 (26). Also, the peptide binding pocket of L d is composed of more hydrophobic residues than other MHC complexes, limiting the possible number of hydrogen bonds that can form between the peptide and ␣ 1 /␣ 2 residues. Finally, a central bulge from the floor of the peptide binding cleft in L d prevents peptides from having a central, stabilizing anchor residue as has been observed in K b and other class I molecules (22). These characteristics most likely contribute to the lower affinities of peptides for L d . It has been suggested that these unique features of the L d -␤2m complex provide an alternative mechanism to the classic antigen presentation pathway (26,27). Notwithstanding the physiological function, the instability of peptide⅐L d -␤2m complexes has been problematic in attempts to examine structural features of this system. 5 In the present study, L d mutants were engineered to overcome stability problems by a process of directed evolution. Randomly mutated L d -␤2m products were displayed as singlechain fusions on the surface of yeast. Stabilized mutants of L d , complexed with the L d -binding peptide QL9 (QLSPFPFDL) (28), were selected by flow cytometric sorting using anti-L d antibodies and a soluble QL9⅐L d -specific high affinity TCR (29). This approach yielded several mutants, including a stabilized, single-chain MHC module consisting of only the ␣ 1 /␣ 2 peptide binding domains. The ␣ 1 /␣ 2 module was capable of binding QL9, and QL9⅐L d -specific TCRs bound to the QL9-␣ 1 /␣ 2 complex with the same affinities as wild-type QL9⅐L d complex on the cell surface. Bacterial expression and characterization of soluble complexes of the QL9-␣ 1 /␣ 2 -L d module indicated that they could be bound by soluble TCRs with affinities similar to full-length peptide⅐L d complexes. The results demonstrate that class I MHC can be engineered to generate more stable complexes and that individual peptide-binding domains can be produced for biochemical and structural studies.
Yeast Display of the Single-chain L d /␤2m-The mouse MHC L d ␣ chain gene was fused to the mouse ␤2m gene using PCR overlap extension (30). The L d -␤2m PCR product consisted of the L d heavy-chain gene (␣ 1 , ␣ 2 , and ␣ 3 ) covalently linked to the mouse ␤2m gene with a 45-bp Gly-Ser (Gly 3 Ser) linker followed by a c-myc epitope tag (Fig. 1A). The L d -␤2m PCR product and the yeast display vector pCT302 were digested with restriction enzymes NheI/XhoI, ligated, and transformed into Escherichia coli strain DH10B by electroporation. E. coli colonies were grown on LB/ampicillin (50 g/ml) plates and then inoculated into LB/ampicillin for 24 h at 37°C. Plasmids were isolated using Qiagen mini-prep kits (Valencia, CA), diagnostically analyzed with restriction enzymes, and sequenced. L d -␤2m-encoded plasmids were transformed by the lithium acetate transformation method (31) into Saccharomyces cerevisiae strain EBY100 (32). Transformed yeast colonies were grown on nutrient-selective media that lacked tryptophan (SD-CAA, glucose 2 wt%, yeast nitrogen base 0.67 wt%, casamino acids 0.5 wt%).
Soluble T-cell Receptors-Selection of yeast displayed L d libraries was performed with a soluble high affinity derivative of the 2C TCR called m6 (2C-m6), which was expressed as an ␣␤ heterodimer in insect cells as previously described (33,34). Culture supernatants from these transfectants were used directly as a source of soluble high affinity TCR for staining cells (34). For some experiments, TCR was purified from culture supernatants by affinity chromatography with nickel-nitrilotriacetic acid-agarose (Qiagen) and size exclusion chromatography (Sephacryl S-200 HR, Amersham Biosciences) (34,35).
Detection of L d -␤2m on Yeast Cell Surface-Transformed yeast colonies were inoculated into SD-CAA medium and grown at 30°C. After 18 -24 h, cells were harvested by centrifugation and incubated in SG-CAA (2 wt% galactose replacing glucose in SD-CAA) at 20°C. Where indicated, L d -binding peptide QL9 (QLSPFPFDL) was added in excess (1 M) to SG-CAA medium to stabilize L d -␤2m complexes expressed on the yeast surface. After 48 h, yeast cultures were harvested by centrifugation, washed with PBS (10 mM NaPO 4 , 150 mM NaCl, pH 7.3) containing 0.5% bovine serum albumin (BSA), and incubated on ice with anti-L d antibodies (28.14.8 or 30-5-7) or high affinity 2C-m6 TCR. After a 30-min incubation, cells were washed with PBS/BSA and incubated on ice with either biotinylated goat anti-mouse IgG followed by streptavidin-phycoerythrin (SA-PE) or fluorescein isothiocyanate F(abЈ) 2 goat anti-mouse IgG. Cells incubated with 2C-m6 TCR were washed with PBS/BSA and incubated with biotinylated anti-TCR antibodies (F23.1 or H57-597), then washed again and incubated with SA-PE. Labeled yeast cells were analyzed on a Coulter Epics XL flow cytometer at the Flow Cytometry Center of the University of Illinois at Urbana-Champaign (UIUC) Biotechnology Center.

Engineering MHC Proteins
and pCT302 vector (150 ng) were digested with NheI/XhoI, combined, and transformed into EBY100 yeast by electroporation in multiple reactions (20 -40 electroporations). Library sizes, determined by plating an aliquot of the pooled transformations, averaged ϳ10 6 independent variants. Error-prone PCR and homologous recombination were used to generate a second random error-prone library, using a template from a clone identified in the first error-prone L d -␤2m library called L d -m8.
Flow Cytometric Sorting of Yeast Libraries-The first generation L d -␤2m error-prone yeast library was sorted in four sequential rounds with anti-L d antibodies. Briefly, the yeast library was grown in SD-CAA media for ϳ24 h at 30°C followed by induction of protein expression in SG-CAA media containing 1 M QL9 peptide (QLSPFPFDL) for ϳ48 h at 20°C. Cells were harvested, washed with PBS/BSA, and incubated on ice with anti-L d (␣ 3 ) antibody (28.14.8) followed by incubation with fluorescein isothiocyanate F(abЈ) 2 goat anti-mouse IgG. The Cytomation MoFlo cell sorter (UIUC Biotechnology Center) analyzed cells at a rate of 30,000 cells/s, and the 1% of yeast cells exhibiting the highest fluorescence was collected. These cells were immediately cultured in SD-CAA for 24 h at 30°C and induced for ϳ48 h at 20°C. The second, third, and fourth sorts used anti-L d (␣ 2 ) antibody (30-5-7), but in these sorts the 0.25-0.5% of yeast cells that exhibited the highest fluorescence were collected. Following four rounds of sorting, cells were plated on SD-CAA agar plates and individual clones were analyzed. Plasmids were rescued using Zymoprep, a yeast plasmid mini-prep kit from Zymogen Research (Orange, CA) and transformed into E. coli by electroporation. Plasmids were isolated from E. coli using the Qiagen mini-prep kit and sequenced at the DNA Sequencing Facility at the UIUC Biotechnology Center.
The second-generation library, using mutant L d -m8 as a template, was screened through four rounds of sorting using two different strategies. The strategies were: one round with anti-L d (␣ 2 ) antibody 30-5-7 followed by three rounds with 2C-m6 TCR (Strategy 1) or three rounds with anti-L d (␣ 2 ) antibody 30-5-7 followed by one round with 2C-m6 TCR (Strategy 2). To obtain more thermostable L d mutant clones, libraries were incubated at a higher temperature (40°C) prior to each sort (37,40). Yeast libraries were induced for 48 h at 20°C, harvested, and incubated at 40°C for 30 min. Cells were then incubated with either 30-5-7 followed by fluorescein isothiocyanate goat anti-mouse IgG or 2C-m6 TCR followed by biotinylated anti-TCR (F23.1) and SA:PE. The 0.5-1% of yeast cells with the highest fluorescence was collected. Following the fourth sort, cells were plated on SD-CAA agar, and individual clones were analyzed.
Site-directed Mutagenesis-Single-site mutations were generated using QuikChange site-directed mutagenesis (Stratagene, La Jolla, CA) with overlapping forward and reverse primers that contained the mutations. Reactions included Pfu polymerase (2.5 units/l), reaction buffer, L d -␤2m pCT302 template (10 ng), dNTPs (25 M), and primers (125 ng). PCR cycles were as follows: 1 cycle of 95°C for 30 s; 16 cycles of 95°C for 30 s, 55°C for 1 min, and 68°C for 16 min. The restriction enzyme DpnI (Invitrogen) was added to PCR reactions for 60 min at 37°C to degrade methylated template L d -␤2m pCT302 plasmid. Plasmid DNA was precipitated with ethanol and transformed into electrocompetent DH10B E. coli cells.
L d -m31 Protein Expression and Purification-For recombinant L d -m31 expression, the platform MHC (L d -m31) was shortened by seven additional amino acids, truncating it between the ␣ 2 and ␣ 3 domains. The resulting construct refolded as efficiently as the original yeast mutant and was used for all further studies. L d -m31 was cloned into pET28a as two different constructs, one containing a GGS spacer and a C-terminal biotin ligation domain (GLNDIFEAQKIEWHE) (41) and one without this domain into NcoI/XhoI sites. Expression of L d -m31 was done in BL21(DE3) Codon Plus E. coli (Stratagene). Cultures were grown in LB/kanamycin (30 g/ml) to an A 600 of 1.0 and induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside for 3.5 h. Cells were passed through a microfluid-izer, and inclusion bodies were solubilized in 8 M urea, 50 mM MES, pH 6.5 at room temperature overnight. Biotinylated L d -m31 was expressed using an in vivo biotinylation method in which L d -m31-biotin pET28a was co-transformed into BL21(DE3) E. coli (Stratagene) with a plasmid containing an arabinose-inducible gene for E. coli BirA ligase (provided by John Cronan at UIUC) (42). Expression was carried out as described above, with the exception that 10 g/ml chloramphenicol was added to all media for maintenance of the BirA plasmid, and BirA ligase expression was induced with 2 mM L-arabinose in the presence of 50 M d-biotin when A 600 ϳ 0.6.
For the refolding reaction, 8 mg of purified QL9 peptide was added to 400 ml of refolding buffer (100 mM Tris-HCl, 400 mM L-arginine, 0.5 mM phenylmethylsulfonyl fluoride, 0.5 mM oxidized/5 mM reduced glutathione, pH 8.0). Over a 24-h period, 200 mg of urea-solubilized L d -m31 inclusion bodies were added in three equal additions. The refold mixture was filtered, dialyzed against 20 mM Tris-HCl, 150 mM NaCl, pH 8.0, concentrated in a stirred cell concentrator (Amicon), and subjected to size-exclusion chromatography (Superdex 75, Amersham Biosciences). Purity of select fractions was determined by SDS-PAGE, and fractions were pooled and concentrated in a YM10 Centriprep (Amicon). The yield of refolded protein from a 400-ml refold was 2-4 mg. Excess QL9 peptide was added to the purified protein to prevent unfolding of L d -m31 due to dissociation of peptide.
TCR⅐QL9⅐L d Affinity Measurements-To examine if L d mutants displayed on the surface of yeast exhibited structures similar to normal L d on T2-L d cells, the affinity of QL9⅐L d complexes for the 2C-m6 TCR was measured. Purified 2C-m6 TCR was used in an equilibrium binding flow cytometry assay. Yeast cells or human T2 cells transfected with L d (44), loaded with exogenous QL9 peptide, were washed with PBS/0.5% BSA, and incubated with varying concentrations of 2C-m6 TCR at 4°C for 60 min. Flow cytometry experiments for detection of bound 2C-m6 TCR were carried out as described above.
Binding of 2C scTCRs to immobilized biotinylated QL9⅐L d -m31 was analyzed by surface plasmon resonance (SPR) on a BIAcore 3000 instrument. Streptavidin sensor chips (BIAcore) were conditioned with 1 M NaCl, 50 mM NaOH. Biotinylated QL9⅐L d -m31 was immobilized to a level of 450 -500 response units, and the remaining surface was blocked with 1 M d-biotin. 2C scTCRs were diluted to the desired concentrations (0.0008 -11.4 M, depending on the scTCR being tested) in HBS buffer containing 0.005% Tween-20 and 1 M QL9 peptide and injected at 25°C at a flow rate of 30 l/min. Binding to a blank biotin-blocked control surface was subtracted from all measurements. Kinetic data were analyzed using simultaneous k a /k d determination with a 1:1 Langmuir binding model on BIAevaluation 3.2 software (BIAcore). Equilibrium affinities were determined by linear regression of response units at equilibrium versus response units/concentration (40).
Structural Models-Models were generated using Swiss-Pdb Viewer (GlaxoSmithKline) and are based on the p29-L d /␤2m structure (21). Swiss-Pdb Viewer was used to insert mutations and to energy minimize the model using the GROMAS96 algorithm with 100 -200 steps of steepest decent and 100 -200 steps of conjugate gradient.

RESULTS
Expression of Single-chain L d -␤2m on Yeast-To engineer more stable L d -␤2m complexes, a single-chain gene was fused

Engineering MHC Proteins
to yeast mating factor Aga-2, which is expressed on the yeast cell wall (Fig. 1A) (32). The conformation and expression levels of the L d construct were assessed by flow cytometry with two different anti-L d antibodies and a soluble high affinity TCR. The two antibodies, 30-5-7 and 28.14.8, are conformation-specific probes for the ␣ 2 and ␣ 3 domains, respectively (43). TCR 2C-m6, a high affinity derivative of the 2C TCR, is a specific probe for the conformation of the peptide-␣ 1 /␣ 2 domains of the QL9⅐L d -␤2m complex (29). Thus, collectively the antibodies and TCR are useful probes for the proper folding of different domains of the L d -␤2m proteins.
L d -␤2m expressing yeast were induced in the presence of the L d -binding peptide QL9 (QLSPFPFDL) to stabilize L d on the yeast cell surface. The published sequence of L d contains a leucine at position 126, whereas the fusion amplified from the template L d gene contained a proline at position 126. Two L d -␤2m fusions were examined, one with the leucine (L d -␤2m Leu-126 ) and one with proline (L d -␤2m Pro-126 ) at position 126. The L d -␤2m Pro-126 fusion was positive by flow cytometry with the anti-L d (␣ 3 ) antibody but negative with the anti-L d (␣ 2 ) antibody or the 2C-m6 TCR (Fig. 1B). The L d -␤2m Leu-126 fusion was positive with both anti-L d antibodies and weakly positive with the 2C-m6 TCR (Fig. 1C). Not surprisingly, an initial error-prone PCR derived library of mutants, generated using the L d -␤2m Pro-126 fusion as a template, yielded mutants that all contained the leucine at position 126 (see below). Thus, the selection process was capable of identifying single-site mutations that confer increased yeast cell surface levels.
Selection of L d -␤2m Variants from Error-prone PCR Libraries-Previous studies have shown that it is possible to identify mutations that increase stability of a fusion protein by selection for increased yeast display (36,37,40,44). As described in the methods, an error-prone PCR library of ϳ10 6 independent transformants was generated from the L d -␤2m Pro-126 mutant. The yeast library was incubated with L d -binding peptide QLSPFPFDL (QL9) and selected through four rounds of sorting, growth, and induction. Fluorescence-activated cell sorting was performed with the anti-L d (␣ 3 ) antibody in the first round and the anti-L d (␣ 2 ) antibody in subsequent rounds. Following the fourth round of sorting, yeast clones were screened for binding to the anti-L d (␣ 2 ) antibody. Multiple clones with the same sequence were identified. Flow cytometry results for one of these clones, L d -m8, incubated with and without QL9 peptide are shown in Fig. 2 (A and B). Addition of QL9 to the yeast cells enabled binding of the soluble 2C-m6 TCR. This suggests that cell surface L d was capable of binding the QL9 peptide and presenting it in a conformation recognized by the 2C TCR. Sequencing revealed that L d -m8 had three mutations: Pro-126 3 Leu (CCG to CTG), Lys-196 3 Asn (AAA to AAT), and a deletion in the c-myc tag at the C terminus that extended the open reading frame another 23 residues to a stop codon, resulting in a hydrophilic tail (CNSSRSDNNS-VDVTKSTLKPLYF). Expression of the Lys-196 3 Asn mutation alone in L d -␤2m Pro-126 did not yield increased display as measured by anti-L d (␣ 2 ) antibody binding, suggesting that the Lys-196 3 Asn mutation does not contribute to the stability of this protein (data not shown). Thus, it is very likely that the Pro-126 3 Leu reversion accounted for the higher surface levels of the L d -m8 protein.
It has been shown previously that successive mutagenesis and screening for increased surface levels at higher temperatures could yield TCR variants with synergistic, stabilizing mutations (37,40). To generate such variants in L d , a new errorprone library was generated using the L d -m8 gene as the template for error-prone PCR mutagenesis. A library of ϳ10 6 trans- . Flow cytometry histograms of representative L d -␤2m mutants. Yeast cells were induced to express L d -␤2m variants in the presence of exogenous QL9 peptide. Cells were stained with anti-L d antibodies or 2C-m6 TCR as described in Fig. 1 (shaded histograms). Unshaded histograms represent background staining from cells incubated with secondary antibodies and SA-PE only. SEPTEMBER 1, 2006 • VOLUME 281 • NUMBER 35 formants was induced in the presence of the QL9 peptide, incubated at 40°C, and selected using two different sorting strategies. In one strategy the library was first sorted with anti-L d (␣ 2 ) antibody followed by three sorts with soluble 2C-m6 TCR. In the second strategy, the library was sorted three times with anti-L d (␣ 2 ) antibody and the fourth time with soluble 2C-m6 TCR. For both of the sorting strategies, 0.25-1% of the cells with the highest fluorescence were collected. Increased surface levels of L d , as judged by staining with anti-L d (␣ 2 ) antibody and 2C-m6 TCR, were observed by the fourth sort using both strategies (data not shown).

Engineering MHC Proteins
Characterization and Structural Analysis of Yeast-displayed L d Mutants-Clones from each sorting strategy were examined by flow cytometry for binding to anti-L d (␣2) and anti-L d (␣3) antibodies and the 2C-m6 TCR (Fig. 3 and data not shown). Most of the clones showed improved surface levels by anti-L d (␣ 2 ), staining compared with L d -m8 (Fig. 3A and data not  shown). Although the anti-L d (␣ 2 ) antibody bound all of the clones, many of the clones now lacked binding to the anti-L d (␣ 3 ) antibody ( Fig. 3B and data not shown). Clones L d -m31, L d -m40, and L d -m61 exhibited the highest levels of L d as detected with the 2C-m6 TCR (Fig. 3C). Flow cytometry profiles of two clones (L d -m31 and L d -m37) that represent the two different classes of mutants (i.e. reactive or non-reactive with anti-L d ␣ 3 antibody) are shown in Fig. 4. Mutant L d -m31 expressed surface levels that were ϳ10-fold higher than L d -m8, as detected with both the anti-L d (␣2) antibody and the 2C-m6 TCR. Mutant L d -m37 expressed surface levels that were 3-to 5-fold higher than L d -m8, as detected with all three reagents.
Plasmids from several clones were rescued and sequenced. Identical sequences were found for two clones, L d -m31 and L d -m40, which were negative for anti-L d (␣3) antibody binding. Mutant L d -m31 contained six mutations, Asn- 30  shares only the Trp-97 3 Arg mutation with L d -m31. This mutation has been shown to stabilize peptide binding and ␤2m association (20). Because Trp-97 3 Arg is the only mutation shared between these two mutants, we examined the influence of the mutation when cloned into the L d -␤2m construct. As shown in Fig. 6, the Trp-97 3 Arg mutation enhanced the surface levels of L d , as detected with all three reagents. However, the surface levels of the L d -m37 mutant were severalfold higher than the L d Trp-97 3 Arg mutant, suggesting that the ␤2m mutations in L d -m37 may contribute additional stability. Among the five mutations in ␤2m, none of them seem to be included in the interface with L d .
Relative Affinities of a TCR for QL9⅐L d , QL9⅐L d -m31, and QL9⅐L d -m37-It is possible that mutations affecting either the stability of L d or the conformation of the peptide could also influence the binding affinity of the TCR. To assess this possibility, the relative binding affinities of the TCR for the L d variants on yeast were compared with the affinity of the TCR for normal, wild-type L d expressed on a mammalian cell line (the human T2 cell line, transfected with L d ) (45). Yeast cells bearing L d -m31 or L d -m37 and T2-L d cells were incubated with QL9 peptide and then analyzed by flow cytometry with varying concentrations of purified soluble 2C-m6 TCR (Fig. 7). The results indicated that all three QL9⅐L d molecules FIGURE 5. Energy-minimized models of QL9⅐L d -␤2m mutants. Models of L d -␤2m variants L d -m31 (A) and L d -m37 (B). Mutated side chains (yellow) were modeled into the p29-L d -␤2m structure (Balendiran et al. (21)) and energy-minimized using GROMOS96 in the Swiss-Pdb viewer. The structural figures were produced with PyMOL. L d heavy chain (␣1, ␣2, and ␣3 domains) is shown in cyan, ␤2m in gray, and QL9 peptide in magenta. The full C terminus of L d -m31 is shown up to residue 186. This C-terminal extension was truncated by seven amino acids for recombinant expression based on the structure.

Engineering MHC Proteins
have similar affinities for TCR 2C-m6, suggesting that the mutations in L d -m31 and L d -m37 do not affect the conformation of the region recognized by the 2C-m6 TCR. Another high affinity TCR mutant that binds to QL9⅐L d called 2C-m13 was also able to bind to yeast cells bearing L d -m31 and L d -m37 (data not shown), further verifying that these mutants have retained the normal structure of L d . Thus, these proteins should be useful in further studies to explore the structural and biochemical features of L d .
Expression, Purification, and Analysis of Soluble QL9⅐L d -m31 Complexes-In previous studies, we have shown that singlechain TCRs (scTCRs) with stabilizing mutations identified by yeast display could be expressed in high yields from E. coli (40,46). To determine if the stabilized mutant L d -m31 could be expressed as a soluble ␣1/␣2 module, the L d -m31 gene was cloned into the pET28a vector. Inclusion bodies were arginine refolded in the presence of excess QL9 peptide, and the concentrated peptide⅐MHC complexes were subjected to size-exclusion chromatography. As shown in Fig. 8A, a major component migrating at the size expected of monomers (ϳ21 kDa) was observed, and the protein showed size homogeneity by SDS-PAGE (Fig. 8A, inset).
To evaluate if the protein could form complexes with soluble TCRs, the purified QL9⅐L d -m31 protein was incubated with soluble scTCR 2C-m6 and the wild type 2C scTCR (T7). The mixtures were electrophoresed in gradient native polyacrylamide gels (Fig. 8B). A unique band, indicating complex formation, was observed in samples that contained the high affinity scTCR 2C-m6 and the QL9⅐L d -m31 protein, when compared with each component analyzed separately. Furthermore, this complex was only observed when QL9 and not a control peptide called MCMV, was used in the refolding of the L d -m31 protein (data not shown). The lower affinity scTCR 2C-T7 has been shown to bind to QL9⅐L d -m31 (see below) but does not show binding by native gel.
Surface plasmon resonance was used to measure scTCR⅐QL9⅐L d -m31 binding affinities. The L d -m31 gene was cloned with a C-terminal tag for enzymatic addition of biotin. L d -m31-biotin inclusion bodies were purified from E. coli cells that overexpressed the biotin ligase and refolded. QL9⅐L d -m31-biotin was immobilized on streptavidin-coated chips, and both the high affinity scTCR 2C-m6 and wild type scTCR 2C-T7 were analyzed (Fig. 8C). The K d values measured for the 2C-T7 scTCR⅐QL9⅐L d -m31 interaction was 1.6 M (measured using steady-state analysis). The K d measured for the 2C-m6 scTCR⅐QL9⅐L d -m31 interaction was 8 nM (kinetic) with an on-rate of 4.5 ϫ 10 5 /Ms and an off-rate of 3.4 ϫ 10 Ϫ3 /s. These values are similar to those previously reported for both the 2C TCR (33) and the m6 TCR (35). These results confirm in two cases that our engineered platform MHC is biochemically equivalent to L d that includes ␣ 3 domain and ␤2m.

DISCUSSION
Various studies have sought to engineer peptide⅐class I/␤2m complexes to facilitate expression on the mammalian cell surface or to improve yields of properly assembled soluble forms of the complexes. In many cases, efforts have focused on the generation of covalent linkages between these molecules (reviewed in Ref. 8). Early studies explored the linkage of ␤2m to the class I heavy chain (47) or of the peptide to class I heavy chain (48,49). It has been argued that single-chain strategies with peptides may be of variable success, due to the need for anchoring peptide termini into class I pockets. However, Hansen and colleagues recently generated mammalian cell surface-expressed forms of peptide-linker-␤2m-linker-heavy chain constructs (called a single-chain trimer) (9). The single-chain trimers exhibited improvements in stability and ability to stimulate T cells (9), and they could be expressed and refolded from E. coli in native form (50). In addition, a Tyr-84 3 Ala mutation that would potentially allow the C-terminal extension of the peptide to exit the MHC groove more favorably was also shown to be properly assembled.
Although these engineering approaches have been guided to some extent by the structures of peptide⅐class I⅐␤2m complexes, few studies have attempted to use a process of directed evolution to identify improvements in soluble peptide⅐class I⅐␤2m complexes. We, and others, have used yeast display to select for mutated class II products that are expressed on the surface of yeast (14,15). Based on our studies with single-chain TCRs (36,37,40,44,51), we have predicted that such displayed proteins are likely to exhibit higher levels of expression and folding as soluble constructs. Our previous study also showed that a single-chain K b /␤2m construct could be displayed on yeast (13), but efforts to engineer the construct by directed evolution were not attempted. In this report, we used a process of random mutagenesis and selection using yeast display to identify mutants of the class I molecule L d that were expressed at high levels on the surface of yeast. One of these mutants contained only the N-terminal ␣1 and ␣2 domains, and in the presence of peptide the protein refolded from E. coli with high yields. Such ␣1/␣2 platforms may be of general use for studying class I MHC products. In previous studies, the structures of peptide⅐class II complexes were used to design a single-chain class II molecule that consisted of only the ␤1/␣1 peptide-binding domains (52,53), suggesting that stabilized forms of these class II platforms might also be amenable to yeast display and directed evolution.
Although the crystal structures of two L d complexes have been reported (21,22), crystallization of additional complexes has been problematic. 6 For example, our efforts to crystallize QL9⅐L d , the ligand for the 2C TCR used in the present studies, have been unsuccessful. These problems may be associated with several features of L d that lead to reduced stability compared with other mouse class I proteins: 1) unusual orientation  Refolded QL9⅐L d -m31 and scTCRs incubated for 5 min at room temperature run on a native 8 -25% gradient polyacrylamide gel (PhastSystem, Pharmacia Biotech). Relative stoichiometries of proteins in each lane are indicated above each lane. Position of the scTCR⅐QL9⅐L d -m31 complexes are shown with an asterisk. 2C-m6 refers to the high affinity scTCR, and 2C-T7 refers to the wildtype, stabilized 2C scTCR. C, SPR sensograms of immobilized QL9⅐L d -m31 binding soluble scTCRs 2C-T7 and 2C-m6. Refolded biotinylated QL9⅐L d -m31 was immobilized on a Streptavidin sensor chip (Biacore) at 450 -500 response units (RU). Soluble scTCR 2C-T7 was flowed over the sensor chip surface at 11.4, 5.7, 2.8, 1.4, 0.7, 0.18, and 0.09 M. Soluble scTCR 2C-m6 was flowed over the sensor chip surface at 0.11, 0.053, 0.026, 0.013, 0.007, 0.003, 0.002, and 0.0008 M. Binding of scTCRs to a blank control surface was subtracted from binding to the QL9⅐L d -m31 surface to account for nonspecific binding to the sensor chip surface. K d values measured for 2C-T7 were 1.6 M (steady-state affinity), whereas the on-rate and off-rate where too fast to be determined with SPR. K d values measured for 2C-m6 was 8 nM measured for simultaneous fitting of kinetic parameters, and the on-rate was 4.5 ϫ 10 5 /Ms, whereas the off-rate was 3.4 ϫ 10 Ϫ3 /s.

Engineering MHC Proteins
of the ␣ heavy chain relative to ␤2m, 2) a reduced number of contacts between ␣ 1 /␣ 2 domains and ␤2m, and 3) lower affinity of the ␤2m for the ␣ chain (21). Interestingly, one substitution, Trp-97 3 Arg, in the structurally related L q allele has been shown to confer additional stability on L q (20). Remarkably, this mutation was also identified in the present study using a directed evolution approach. This observation further supports the notion that the efficiency of yeast display (i.e. cell surface levels) can be used to select for stabilizing mutations in the fused protein.
While Trp-97 3 Arg was the only mutation present in both the ␣1/␣2 mutant L d -m31 and in the full-length mutant L d -m37, several additional mutations were identified in each (Fig.  5). It is unclear if these mutations contribute to the enhanced surface display of the L d mutants. In the L d -m31 mutant, the four additional mutations are conservative substitutions. In the L d -m37 mutant, six of the seven additional mutations are located in ␣3 (five mutations) and ␤2m (one mutation) and thus could be involved in stabilizing these two domains. While subsequent studies in the present report focused on the ␣1/␣2 mutant, the full-length L d mutant may be of some interest in the future.
We also show here that E. coli expression of the ␣1/␣2 L d -m31 platform yielded a stable protein that could be refolded with L d -binding peptides. Our previous studies using singlechain TCRs or individual V␤ domains have shown that mutants with enhanced yeast surface display also showed improvements in secretion levels and thermostability (37,40,44,51). Thus, we anticipated that the yeast displayed L d mutants might also show similar improvements in E. coli expression and folding. In fact, the yields and solubility of purified peptide/L d -m31 complexes have been excellent, allowing their use in both binding studies and crystallization trials.
Although the selection process using the high affinity TCR m6 would assure that the QL9⅐L d mutant complexes were recognized by the specific TCR, it was formally possible that the affinity of the TCR⅐QL9⅐L d interaction could be influenced by the mutations. However, two independent approaches, including SPR, showed that TCRs bound to QL9⅐L d -m31 with the same affinities as reported for binding to the full-length QL9⅐L d complexes. The single-chain version of both the wild-type 2C TCR and the m6 TCR recognized soluble QL9⅐L d -m31 with affinities corresponding to previous measurements using fulllength L d (29,33,35). These results suggest that neither the Trp-97 3 Arg mutation nor other mutations influence the conformation of the peptide-MHC surface that is recognized by the TCR. Furthermore, it shows that at least in the case of this TCR system, the ␣3/␤2m domains do not appreciably influence the binding of TCRs. Interestingly, it was shown over a decade ago that a human class I MHC (HLA-Aw68), which underwent proteolytic cleavage of the ␣3 domain, exhibited a structure similar to the full-length peptide⅐HLA-Aw68⅐␤2m complex. However, ␤2m was retained in the complex, and based on many studies, including early efforts to generate multimers of class I, it has generally been assumed that ␤2m is essential for stabilization of the entire peptide⅐class I⅐␤2m complex (54,55).
Finally, it is reasonable to predict that it will be possible to engineer other class I molecules as ␣1/␣2 platforms, either through a process of directed evolution as described here or by rational design of the molecules (e.g. with substitutions of key hydrophobic residues that would be exposed upon removal of the ␣3 and ␤2m domains). The engineered proteins could be very useful in structural studies (e.g. NMR), in the development of tetramer reagents, or as modulating agents for T-cell responses. The elimination of the ␣3 domain, and thus the CD8-binding epitopes, on a class I complex could reduce problems with background binding of the class I tetramers to CD8 on T cells (56,57). In addition, class I molecules that lack ␣3/␤2m might be of some use in regulating the activity of T cells by either stimulating the highest avidity, CD8-independent T cells or by eliciting inhibitory signals through the TCR complex in the absence of CD8/lck signaling (58). In this regard, a recent study compared QL9⅐L d monomers and multimers in the 2C system to examine activation requirements (59). It will be of interest to examine how the QL9⅐L d -m31 ligand, which lacks the CD8-binding epitope, will function in eliciting T-cell activity.