Three binding sites in protein-disulfide isomerase cooperate in collagen prolyl 4-hydroxylase tetramer assembly.

Protein-disulfide isomerase (PDI) is a modular polypeptide consisting of four domains, a, b, b', and a'. It is a ubiquitous protein folding catalyst that in addition functions as the beta-subunit in vertebrate collagen prolyl 4-hydroxylase (C-P4H) alpha(2)beta(2) tetramers. We report here that point mutations in the primary peptide substrate binding site in the b' domain of PDI did not inhibit C-P4H assembly. Based on sequence conservation, additional putative binding sites were identified in the a and a' domains. Mutations in these sites significantly reduced C-P4H tetramer assembly, with the a domain mutations generally having the greater effect. When the a or a' domain mutations were combined with the b' domain mutation I272W tetramer assembly was further reduced, and more than 95% of the assembly was abolished when mutations in the three domains were combined. The data indicate that binding sites in three PDI domains, a, b', and a', contribute to efficient C-P4H tetramer assembly. The relative contributions of these sites were found to differ between Caenorhabditis elegans C-P4H alphabeta dimer and human alpha(2)beta(2) tetramer formation.

The PDI polypeptide comprises four domains, a, b, bЈ, and aЈ, of which the first and last contain the -CGHC-catalytic site required for thiol-disulfide exchange activity (12) (see Fig. 1). The a and b domains have a thioredoxin fold (13,14), and the bЈ and aЈ domains are also thought on the basis of homology to have the same fold. In addition, there is a short linker region x between the bЈ and aЈ domains (15), and the C terminus of the polypeptide contains a highly acidic 29-amino acid extension c, which is not critical for any of the major functions of the protein except ER retrieval (16). The principle substrate binding site of PDI is located in the bЈ domain (17), which by itself is capable of binding short peptide substrates. We have recently identified a putative hydrophobic pocket in the bЈ domain by structural homology modeling and sequence alignments and shown by site-directed mutagenesis that this pocket contains several residues that are important for the binding of short peptide substrates (15). The most important residue in this respect was Ile 272 , since its replacement by all of the amino acids studied greatly inhibited substrate binding with no observable effect on the structure of the bЈ domain or the full-length PDI polypeptide (15). Substitutions of the residues Leu 242 , Leu 244 , and Phe 258 also markedly reduced binding (15). The bЈ domain is essential but not sufficient for the binding of longer polypeptides, since binding required addition of the ab or aЈ domain (17). The importance of the bЈ domain residue Ile 272 for substrate binding was further emphasized by the reduced binding affinity of the I272W mutant full-length PDI to the nonnative protein "scrambled" RNase (15).
We have shown previously that the minimum requirement for assembly with a C-P4H ␣ subunit to form an active ␣ 2 ␤ 2 tetramer is fulfilled by domains bЈ and aЈ of PDI but that the presence of the a and b domains greatly enhances assembly (18). The a and b domains can in part be replaced in C-P4H assembly by the corresponding domains of ERp57 (18), a structural and functional homologue of PDI involved in the folding of newly synthesized glycoproteins (19,20) but which cannot in itself replace PDI in C-P4H tetramer assembly (21). The active site cysteines in the catalytic a and aЈ domains of PDI are not required for the assembly or activity of the C-P4H tetramer (7).
The present work set out to study the role of the primary substrate binding site in the bЈ domain of PDI in C-P4H tetramer assembly and activity. Surprisingly, no mutations in any of the bЈ domain residues previously shown to be important in peptide binding prevented C-P4H tetramer assembly or reduced the enzyme activity. We therefore used sequence conservation among the thioredoxin family to identify residues that might form substrate binding sites in the a and aЈ domains. Efficient assembly of an active C-P4H tetramer was found to be dependent on interaction sites in three PDI domains, a, bЈ, and aЈ, and the same sites were also found to play a role in the assembly of a Caenorhabditis elegans ␣ subunit-human PDI C-P4H dimer. The relative importance of each site for PDIprotein interactions was greatly dependent on the nature of the interaction partner, and there was clear evidence of cooperativity between the three binding sites in PDI.

Generation of Mutant PDI Expression Vectors-Plasmids pLWRP64,
an Escherichia coli expression vector encoding mature human PDI with an N-terminal His tag in-frame with the cloned gene; pVL␤, an insect expression vector encoding full-length PDI; and a vector encoding ER-paPDIbbЈaЈc in an insect expression vector had been generated previously (6,18,22). PDI mutants were generated on these vectors using site-directed mutagenesis performed using the QuikChange TM site-directed mutagenesis kit (Stratagene) as recommended by the manufacturer. All of the plasmids generated were sequenced to ensure that there were no errors in the cloned genes.

Generation of Baculoviruses for PDI Mutants and Expression of the Recombinant Proteins in Insect Cells-
The recombinant PDI mutant plasmids generated were cotransfected into Spodoptera frugiperda Sf9 insect cells with a modified Autographa californica nuclear polyhedrosis virus DNA (BaculoGold; PharMingen) by calcium phosphate transfection (23). The resultant viral pools were collected 4 days later, amplified twice, and used for recombinant protein production. Other recombinant baculoviruses used in this work coded for the human C-P4H ␣(I) and ␣(II) subunits, the human PDI/␤ subunit (6,9), and the C. elegans ␣ subunit PHY-1 (32). Sf9 cells were cultured in TNM-FH medium (Sigma) supplemented with 10% fetal bovine serum (BioClear) at 27°C as monolayers. To produce recombinant proteins, insect cells seeded at a density of 0.7-1.5 ϫ 10 6 /ml were infected with the recombinant viruses at a multiplicity of 5. The cells were harvested 72 h after infection; washed with a solution of 0.15 M NaCl and 0.02 M phosphate, pH 7.4; homogenized in a 0.1 M glycine, 0.1 M NaCl, 10 M dithiothreitol, and 0.01 M Tris buffer, pH 7.8, supplemented with 0.1% Triton X-100, using 1/20 of the original culture volume; and centrifuged at 10,000 ϫ g for 20 min.
Analysis of Recombinant PDI Mutants and C-P4H Tetramers Expressed in Insect Cells-Aliquots of the Triton X-100-soluble insect cell supernatants were analyzed by denaturing 8% SDS-PAGE under reducing conditions, nondenaturing 8% PAGE, and Western blotting using an anti-PDI antibody (32). The amount of wild-type and mutant C-P4H tetramers were compared by densitometry of the Coomassiestained nondenaturing PAGE using a GS-710 calibrated imaging densitometer (Bio-Rad). Triton X-100-soluble samples corresponding to PDI and five mutants were transferred from SDS-PAGE to a ProBlott™ membrane (Applied Biosystems) by electroblotting and stained with Coomassie Blue. The samples were excised from the blot, and their N-terminal sequence was analyzed in a ProciseTM 492 sequencer (Applied Biosystems).
Enzyme Activity Assay-The P4H activity of the Triton X-100-soluble insect cell fractions was assayed by a method based on the hydroxylation-coupled decarboxylation of 2-oxo[1-14 C]glutarate (24), using a synthetic peptide (PPG) 10 (Peptide Institute) as the substrate. K m values for (PPG) 10 were determined by varying its concentration while the concentrations of the other reaction components were kept constant.
Protein Expression in E. coli and Purification-Protein production was carried out in E. coli strain BL21 (DE3) pLysS grown in LB medium at 37°C and induced at an A 600 of 0.3 for 3 h with 1 mM isopropyl ␤-D-thiogalactoside. Cell lysis was carried out by freeze-thawing, and the proteins were purified using BD TALON TM single step columns (BD Biosciences). All of the proteins were Ͼ95% pure in SDS-PAGE analysis, with the minor impurities being at a low level of degradation as seen in E. coli. The concentration of each protein was determined spectrophotometrically using a calculated absorption coefficient of PDI (45,040 M Ϫ1 cm Ϫ1 ; M r 56,386) at 280 nm.
Cross-linking and Protease Sensitivity Assay-Cell extracts from E. coli were prepared by freeze-thawing. Bolton-Hunter 125 I labeling of ⌬-somatostatin (AGSKNFFWKTFSS) was performed as recommended by the manufacturer (Amersham Biosciences), and cross-linking was performed using the homobifunctional cross-linking reagent disuccinimidyl glutarate (Sigma), as described for ⌬-somatostatin or "scrambled" RNase (17). The protease sensitivity assay was carried out by adding proteinase K (0 -50 g/ml) or V8 (0 -100 g/ml) to 2 l of lysates in a total volume of 10 l and then incubated on ice for 20 min. After that, 0.5 l of 20 mM phenylmethylsulfonyl fluoride was added, and the samples were run on a denaturing 12.5% SDS-PAGE under reducing conditions and Western blotted using an anti-PDI antibody (Stressgen). The stringent condition used for screening all of the mutants was 20 g/ml proteinase K.
Circular Dichroism Spectrum Analyses-Far UV circular dichroism spectra were recorded on a Jasco J600 spectrophotometer. All scans were collected at 25°C as averages of eight scans, using a cell with a path length of 0.1 cm, a scan speed of 20 nm/min, a spectral bandwidth of 1.0 nm, and a time constant of 1 s. The maximal high tension voltage was 750 V. All spectra were corrected for blanks run with no protein added.

Point Mutations in the Substrate Binding Site in the bЈ Domain of PDI Do Not by Themselves Impair C-P4H Tetramer
Assembly-We have recently identified several residues in the bЈ domain that are important for the binding of peptide substrates (15). Replacement of Ile 272 by alanine, tryptophan, asparagine, glutamine, or leucine in the individual bЈ domain and full-length PDI greatly impaired binding of the small peptides ⌬-somatostatin and mastoparan, whereas replacement of the residues Leu 242 , Leu 244 , and Phe 258 by various amino acids also reduced peptide binding, but to a lesser extent (15). The I272W mutation had the greatest effect on binding affinity, and in addition to the short peptide substrates it also reduced the affinity of the full-length PDI to "scrambled" RNase (15). We therefore tested whether the mutation of any of these four residues or the spatially adjacent residue Ser 256 in the fulllength PDI polypeptide would impair the assembly of a C-P4H tetramer.
Recombinant full-length PDI polypeptides containing the single point mutation L242W, L244W, F258A, F258W, I272A, I272W, or S256D in the bЈ domain ( Fig. 1) were expressed together with the human C-P4H ␣(I) or ␣(II) subunits in insect cells. The cells were harvested 72 h after infection, homogenized in a buffer containing Triton X-100, and centrifuged. Assembly of a recombinant human type I [␣(I)] 2 ␤ 2 or type II [␣(II)] 2 ␤ 2 C-P4H tetramer was then analyzed by nondenaturing PAGE of the Triton X-100-soluble proteins (Fig. 2), and P4H activity was assayed by a method based on the hydroxylation-coupled decarboxylation of 2-oxo-[1-14 C]glutarate (Table  I). Surprisingly, none of the bЈ domain mutations tested was found to inhibit C-P4H tetramer assembly (e.g. see Fig. 2, lanes  1-3), and none caused a marked reduction in P4H activity  (Table I). In fact, the L244W, F258A, I272A, and S256D mutations significantly increased the amount of activity generated (Table I), coinciding with an increased amount of the corresponding enzyme tetramers (e.g. see Fig. 2, lane 3). The higher amounts of C-P4H tetramers were not due to increased expression levels of the mutant PDI polypeptides, since no differences were seen relative to those of the mutant and wild-type PDI polypeptides in SDS-PAGE analysis followed by Western blotting with an anti-PDI antibody (e.g. see Fig. 3A, lanes 1-4). In addition, the K m values of the type I and type II S256D mutant C-P4H tetramers for the (PPG) 10 peptide-substrate were comparable with those of the wild-type enzymes (Table I); hence, their higher P4H activities were not due to increased affinity for the substrate. The data thus show that none of the bЈ domain residues previously shown to be critical for peptide binding by PDI is by itself essential for C-P4H tetramer assembly, and surprisingly, that certain mutations in these residues lead to more efficient assembly in insect cells.
To study further the effect of the PDI I272W mutation on the assembly properties of C-P4H, we expressed a PDIbbЈaЈc domain construct and the corresponding I272W mutant together with the ␣(I) or ␣(II) subunits in insect cells. Consistent with previous observations (18), no activity was generated when either the wild-type or the mutant domain construct was coexpressed with the ␣(I) subunit (Table I). The amount of activity obtained when the bbЈaЈc construct was co-expressed with the ␣(II) subunit was 35% of that obtained with the full-length PDI, which was again consistent with previous observations (18), whereas the activity generated with the I272W mutant bbЈaЈc construct was only 3% (Table I). The wild-type and mutant bbЈaЈc constructs were expressed at equal levels ( Fig.  3A, lanes 10 and 11), and a faint band corresponding to an enzyme tetramer could be detected in nondenaturing PAGE by Western blotting (data not shown). The data therefore indicate that when the PDI a domain is missing, the bЈ domain residue Ile 272 becomes critical for C-P4H tetramer assembly.
Residues in the a and aЈ Domains of PDI Are Important for Efficient C-P4H Tetramer Assembly-Since mutations of the critical residues identified at the primary substrate binding site in the bЈ domain of the full-length PDI had no inhibitory effect on C-P4H tetramer assembly, there must be additional binding sites in PDI that contribute to assembly. We have shown previously that the bЈ and aЈ domains of PDI fulfill the minimum requirement for functioning as the C-P4H ␤ subunit in combination with the ␣(II) subunit but that the addition of the a and b domains enhances tetramer formation and is es- . The samples were analyzed by 8% nondenaturing PAGE followed by Coomassie staining. The migration position of the enzyme tetramers is indicated by an arrow, and that of the free PDIs is shown by a bracket. Lanes containing enzyme tetramers with single PDI point mutations are indicated with a solid line above, whereas those containing double and triple mutations are indicated by a dotted and a dashed line, respectively. D, densitometry values of the mutant tetramers compared with the type II wild-type collagen prolyl 4-hydroxylase tetramer obtained using a calibrated imaging densitometer. ND, not determined for the tetramers not visible by Coomassie staining to be measured. There is a good correlation between the effects of the mutations on tetramer assembly and on P4H activity (see Tables I-IV). Values are given in dpm/100 g of extractable cell protein (mean Ϯ S.D. for at least three experiments) and as percentages of the value obtained with the wild-type PDI. The enzyme activity generated by expressing the ␣(I) subunit alone, 930 Ϯ 120 dpm/100 g, or ␣(II) alone, 1160 Ϯ 40 dpm/100 g, was subtracted from all of the values. Significances of the differences relative to the wild type are indicated by footnotes sential for PDI to form a tetramer with the ␣(I) subunit (18). These results, combined with the observations that the isolated a and aЈ domains are able to function as efficient catalysts of thiol-disulfide exchange reactions in peptides and proteins (25), suggest that both the a and aЈ domains contain substrate binding sites.
To identify such potential sites, we used sequence homology conservation among the members of the thioredoxin family to identify putative protein-protein interaction sites in the two catalytic domains a and aЈ. The crystal structures of the periplasmic thiol-disulfide oxidase DsbA from a variety of bacterial species reveal a hydrophobic binding groove leading to the active site (26 -28). This groove includes a conserved cisproline lying under the active site -CXXC-sequence, which is implicated not only in substrate binding by DsbA but also by other superfamily members including thioredoxin and glutaredoxin (29,30). Multiple sequence alignment of the thioredoxinlike domain of the DsbAs and the catalytic domains of the human PDI family allowed the identification of the residues in PDI equivalent to those that form the hydrophobic binding groove in DsbA. Based on these analyses, Leu 42 , Pro 83 , Ile 85 , Tyr 99 , and Trp 111 were identified in the a domain of PDI, and Leu 386 , Pro 424 , Leu 426 , Tyr 440 , and Phe 452 in the corresponding positions in the aЈ domain (Fig. 1). The Pro 83 and Pro 424 sites in PDI were not investigated in this study, since mutations in cis-proline residues will most probably result in structural changes in the protein. Mutations were performed in the remaining positions to try to partially occupy these putative substrate binding sites (Fig. 1), as had been done previously for the binding site in the bЈ domain (15).
When the L42W and W111I mutant PDI polypeptides were co-expressed with either the human ␣(I) or ␣(II) subunit, the amount of C-P4H activity generated was markedly reduced relative to that obtained with the wild-type PDI (Table II), and a concomitant decrease in the amounts of type I and type II enzyme tetramers was seen (e.g. see Fig. 2, lane 5). The I85W mutation had a less marked effect than L42W or W111I when the mutant PDI was co-expressed with the ␣(I) subunit, whereas no significant reduction was observed when co-expressed with the ␣(II) subunit (Table II). In contrast, the Y99W mutation led to a 1.5-1.8-fold increase in C-P4H activity and tetramer assembly with both ␣ subunits (Table II; Fig. 2, lane  4). Western blot analysis of all four mutant PDI polypeptides showed that their expression levels were comparable with that of the wild-type PDI (e.g. see Fig. 3A, lanes 5-7).
The a domain of ERp57 can in part compensate for the loss of the corresponding PDI domain in C-P4H tetramer assembly (18). The positions in ERp57 that correspond to the PDI residues Leu 42 , Ile 85 , and Trp 111 are occupied by two leucines and a histidine, respectively (21). Since residue Trp 111 was found to be the most important one among the PDI a domain residues studied for C-P4H tetramer assembly, we examined whether the compensatory ability of the ERp57 a domain in tetramer assembly could be increased by mutation of the corresponding histidine, His 103 , to tryptophan. A H103W point mutation generated in the ERpaPDIbbЈaЈc polypeptide, which was found to be expressed in insect cells at a level corresponding to the nonmutant polypeptide (data not shown) was then co-expressed with the human ␣(I) and ␣(II) subunits in insect cells. The H103W mutation did not increase C-P4H activity in the case of the ␣(I) subunit (Table II), but in the case of the ␣(II) subunit, this activity increased from 37 to 48% (i.e. by 25%) (Table II). This suggests that the difference between the a domains of ERp57 and PDI in C-P4H assembly can in part be attributed to the Trp 111 position of PDI.
None of the mutations generated in the aЈ domain of PDI had as marked an effect on C-P4H tetramer assembly and activity as those in the a domain (Table II), but the L386W mutation did reduce the amount of activity to 44 and 81% of that obtained with the wild-type PDI when co-expressed with the ␣(I) and ␣(II) subunits, respectively (Table II). The F452W mutation also lowered the amount of enzyme activity, whereas F452I had the opposite effect (Table II). Corresponding changes were observed in the amounts of the mutant enzyme tetramers (Fig. 2, lanes 6 -8). The mutations L426W and Y440W did not cause marked changes in C-P4H assembly and activity (Table  II), and none of the aЈ domain mutations affected the amount of recombinant PDI polypeptide expressed in insect cells (e.g. see Fig. 3A, lanes 8 -9).
Binding Sites in Three PDI Domains Contribute to C-P4H Tetramer Assembly-None of the single point mutations introduced here into the a, bЈ, or aЈ domains of PDI fully inhibited C-P4H tetramer assembly or enzyme activity. Although the I272W bЈ domain mutation alone had no effect on tetramer assembly or activity, it almost completely inhibited the assembly of active C-P4H in the absence of the a domain (Table I), suggesting that this site in the bЈ domain is most probably involved in tetramer assembly. We therefore studied the effect of combining the a or aЈ domain mutations with the I272W bЈ domain mutation.
When the I272W mutation was combined with any of the a or aЈ domain mutations, the C-P4H activities obtained were markedly reduced, ranging from 3 to 74% of those of the wildtype C-P4Hs (Table III). Furthermore, in all cases, the activities obtained were reduced by at least 35% relative to those generated with PDI polypeptides harboring only single a or aЈ domain mutations (compare Tables II and III). The reductions in enzyme activity were accompanied by corresponding decreases in the amount of the C-P4H tetramers (e.g. see Fig. 2,  lanes 9 -12), whereas none of the double mutations affected the expression level of the PDI polypeptides (e.g. see Fig. 3B, lanes  1-5). Of the double mutations in the a and bЈ domains, W111I/ I272W resulted in the lowest activities, 7 and 11% of those generated by the wild-type PDI with the ␣(I) and ␣(II) sub-  (lanes 1, 4, 6 and 9) in B (lane 6), and in C (lane 3) were N-terminally sequenced, and in all cases, the sequence of the first seven amino acids was found to be that of mature recombinant human PDI. units, respectively (Table III), and only about one-fifth of those obtained with the W111I mutant PDI (compare Tables II and  III), whereas the most effective double mutation in the bЈ and aЈ domains was I272W/L386W, which reduced the activities to 3% of those in the wild-type cells (Table III).
In contrast with the additive effect seen by combining the mutation in the bЈ domain with one of those in the a or aЈ domains, the combination of a and aЈ domain mutations did not lead to any further decrease in C-P4H activity (compare Tables  II and III) or assembly (e.g. see Fig. 2, lane 13). None of the combined a and aЈ domain mutations affected the expression levels of the PDI polypeptides (e.g. see Fig. 3B, lanes 6 and 7).
In all cases, except the F452I and Y99W/L426W mutations, the effect on assembly with the ␣(I) or ␣(II) subunits showed the same trends, but the effects on ␣(I) assembly were greater. The significance of the two apparent exceptions is unclear, but it should be noted that the F452I mutation may have a minor effect on the structure of PDI (see below), and the L426W mutation in isolation had no effect on ␣(II) assembly. The data obtained above suggest that residues in three domains of PDI, a, bЈ, and aЈ, contribute to the assembly of an active C-P4H tetramer. We studied this further by generating triple-mutant recombinant PDI polypeptides that contain a mutation in all three domains, the one in the bЈ domain always being I272W. The C-P4H activities generated were 3-28% of those of the wild-type C-P4Hs, the I85W/I272W/L426W, I85W/   Tables III and IV). The triple mutants including W111I or L386W did not cause any further decrease in the already very low levels of tetramer assembly (Fig. 2, lanes 14 and 16) and activity obtained with the double mutants that contained either one of these mutations (compare Tables III and IV). All of the triple mutant PDI polypeptides were expressed at a level comparable with that of the wild-type PDI when analyzed by Western blotting (e.g. see Fig. 3C, lanes  1-4), and thus the decreased C-P4H assembly and activity cannot be explained by lower expression levels of the mutant PDI polypeptides.
Our results suggest that C-P4H tetramer assembly results from interaction of the ␣ subunit with at least three distinct sites in PDI, in the a, bЈ, and aЈ domains. The interaction sites found in the a and aЈ domains appear to act independently from each other but cooperate with that in the bЈ domain, which is identical to the site identified earlier as being involved in the binding of short peptide substrates (15).

The I272W Mutation in the bЈ Domain of Human PDI Completely Inhibits Its Assembly with a C. elegans ␣ Subunit to
Form an Active C-P4H Dimer-The major C-P4H form in the nematode C. elegans is a mixed tetramer consisting of two different ␣ subunits, PHY-1 and PHY-2, and two molecules of PDI (31). In the absence of one of the ␣ subunits, the remaining one assembles into an active dimer with PDI, which can either fully or partially compensate for the lack of the mixed tetramer (31). Since recombinant C. elegans PHY-1 forms an active C-P4H dimer very efficiently with human PDI when co-expressed in insect cells (32), we next studied the effects of some of the single and double mutations in human PDI on the assembly and activity of the PHY-1/human PDI dimer.
In contrast to the human C-P4H tetramers, assembly of the PHY-1/human PDI dimer was significantly reduced by the mutations F258W and F258A in the bЈ domain and completely abolished by the I272W mutation (Table V). Interestingly, the mutation I272A, which also significantly reduced ⌬-somatostatin binding (15), did not significantly reduce the amount of P4H activity generated (Table V). The S256D mutation resulted in a higher amount of C-P4H activity (Table V), as in the case of the human C-P4H tetramers ( Table I).
The effects of the single a and aЈ domain mutations on the assembly of an active PHY-1/human PDI dimer were very similar to those observed with the type I human C-P4H tetramer (compare Tables II and V). The only exceptions were the Y99W mutant, which resulted in no change, and the L426W mutant, which reduced PHY-1/human PDI dimer assembly and activity significantly (Table V). As expected, no C-P4H activity was generated when the PHY-1 polypeptide was co-expressed with double mutant PDI polypeptides that contained the bЈ domain mutation I272W in addition to a mutation in the a or aЈ domain, as the former alone abolished dimer assembly (Table  V). Most combinations of the a and aЈ domain point mutations appeared to have an additive effect in reducing the assembly and activity of the PHY-1/human PDI dimer (Table V) by contrast with their effect on the assembly of human C-P4H tetramers (Table II). a Values are given in dpm/100 g of extractable cell protein (mean Ϯ S.D. for at least three experiments) and as percentages of the value obtained with the wild-type PDI. The enzyme activity generated by expressing the ␣(I) and ␣(II) subunits alone was subtracted from all of the values. p Ͻ 0.001 (significance of the difference relative to the wild type). a Values are given in dpm/100 g of extractable cell protein (mean Ϯ S.D. for at least three experiments) and as percentages of the value obtained with the wild-type PDI. The enzyme activity generated by expressing PHY-1 alone, about 0.3% of that for PHY-1 with PDI, was subtracted from all of the values. Significances of the differences relative to wild-type are indicated by footnotes b-d. b p Ͻ 0.001. c p Ͻ 0.01. d p Ͻ 0.05.

Mutant PDI Polypeptides Do Not Show a Significant Alteration in Their
Structure-To ensure that the effects on C-P4H assembly observed here were not due to gross structural effects of the PDI mutations, wild-type and mutant polypeptides were compared by means of a variety of biophysical analyses.
It has previously been shown that mutations that destabilize the aЈ domain of PDI (e.g. F449R) result in a decrease in ⌬-somatostatin binding by the bЈ domain (33). Accordingly, all 15 single and double a and aЈ domain mutants of PDI generated here were screened for ⌬-somatostatin binding. They all showed binding with levels approaching those of the wild-type polypeptide, except for those bearing the W111I mutation, which showed very significant decreases in the levels of binding (Fig. 4, lane 5). Control experiments showed that peptide binding by the bЈ domain was significantly decreased in all of the polypeptides that included the I272W mutation (data not shown), which is consistent with the reported effects of this mutation on the substrate binding site in the bЈ domain (15).
The full-length PDI mutants expressed in E. coli were screened for protease stability. Proteinase K digestions under stringent conditions revealed that most mutations had no effect on protease resistance but that four mutations reduced the stability of the protein, the magnitudes of the effects being W111I Ͼ F452I Ͼ L386W Ͼ F452W (data not shown). The destabilizing effects appeared to be additive when these mutations were combined. Even in the case of the most effective destabilizing mutation, W111I, however, some undigested protein was left under these stringent conditions. To examine further the possible degree of destabilization, titration curves of proteinase K concentration versus digestion were created for the W111I and I272W/F452I mutants. The results indicated that the I272W/F452I mutant was only marginally less stable than the wild-type polypeptide and the W111I mutant was only slightly less stable (Fig. 5). Further protease resistance tests performed using V8 protease indicated that only the W111I mutation led to decreased stability and that this effect was marginal relative to the wild-type (data not shown).
The full-length wild-type PDI and three mutants, W111I/ I272W/F452W, L42W/I272W/L386W, and Y99W/I272W/ Y440W, were then purified from recombinant expression in E. coli by immobilized metal affinity chromatography, exploiting N-terminal His tags (data not shown). CD spectra in the far UV region of the L42W/I272W/L386W and the Y99W/I272W/ Y440W were essentially identical to the wild-type protein, whereas the W111I/I272W/F452W showed a slight difference in spectra, suggesting a small change in secondary structure (Fig. 6).
Taken together, these analyses indicate that all of the mutant PDI polypeptides generated, excluding those containing W111I, have essentially the same structure as the wild type and suggest that the structures of the W111I mutants are only marginally different from that of the wild-type protein, as supported by the C-P4H assembly and enzyme activity results (Fig. 2, lane 5; Tables II and III). DISCUSSION It has been shown previously that the bЈ domain of PDI is sufficient for the binding of short peptide substrates and essential for the binding of larger nonnative protein substrates but that the a and aЈ domains also contribute to the binding of larger substrates (17). The data suggest that, of the three binding sites, the one in the bЈ domain has the highest affinity. This is consistent with data on the kinetics of thiol-disulfide oxidation by the isolated a domain (34), which implied that substrate binding by the a domain was of low affinity, and with data showing that the bЈ domain of PDI is essential for the catalysis of thiol-disulfide isomerization reactions (25). The substrate binding site in the bЈ domain of PDI has recently been localized by mutagenesis studies, which demonstrated that the single mutation I272W reduced peptide binding very substantially with no discernable effect on the structure of the PDI (15). A reduction in nonnative protein binding by the I272W mutant has been reported, but significant binding was still observed, presumably via the influence of the substrate binding sites in the a and aЈ domains. Conservation of the protein/peptide-binding site within the bЈ domain has been reported for ERp57 (35) and Wind, the Drosophila homologue of ERp28 (36).
The transient nature of the interaction between PDI and its substrate makes it difficult to investigate the substrate binding sites of PDI. In this work, we exploited the fact that the interaction of PDI with a human C-P4H ␣ subunit to form a stable tetrameric complex can be seen to mimic the interaction of PDI with protein substrates, since the role of PDI in the C-P4H tetramer is to keep the ␣ subunit soluble. We were able to show that mutations in the substrate binding site in the bЈ domain of PDI did not reduce C-P4H tetramer assembly or activity. We then combined information previously published for the putative substrate binding sites of members of the thioredoxin superfamily to identify presumed substrate binding sites in the a and aЈ domains of PDI. Single point mutations at either of these sites resulted in a reduction in C-P4H activity and assembly, but a more significant reduction was seen when such mutations were combined with the I272W mutation in the bЈ domain. A further additive effect was seen when mutations at multiple putative substrate binding sites in the a, bЈ, and aЈ domains were combined, since these reduced P4H activity by more than 95% and reduced tetramer assembly to nondetectable levels. A combination of studies on peptide binding, protease stability, and spectroscopic analysis of protein structure revealed no discernable effect on PDI structure with the exception of a minor effect of the W111I and F452I mutations. These results, in combination with the observation that expression levels of all of the mutant PDI polypeptides in insect cells were comparable with that of the wild-type protein, indicate that the present observations were due to a direct effect of the mutations on human C-P4H tetramer assembly and not to indirect effects on the structure of PDI.
The combination of three distinct substrate binding sites within PDI resolves a potential dichotomy regarding the functions it has to perform. To act as a catalyst of protein folding, PDI must bind each folding intermediate with relatively low affinity to allow for a bind-release cycle. In contrast, to act as the ␤ subunit of C-P4H and the microsomal triglyceride transfer protein, PDI must bind with very high affinity to a specific protein (i.e. the ␣ subunits of these enzymes). As a consequence of the multiplicative effect of combining the affinities of multiple binding sites, three sites of relatively low affinity can result in very high affinity binding. One implication to be drawn from these contrasting interactions of PDI is that sites of interaction of the C-P4H ␣ subunit cannot act as a nonfolded protein. This is supported by the facts that the addition of the disulfide reductant dithiothreitol to C-P4H results in disassembly of the tetramer (37) although the cysteine residues of PDI are not required for tetramer formation (7), and site-directed mutagenesis studies have shown that two intrachain disulfide bonds in the human ␣(I) subunit are required for assembly (38,39). These data imply that PDI binds a structured ␣ subunit with higher affinity than an unstructured one.
In contrast to human C-P4H tetramer assembly, PHY-1/ human PDI dimerization was shown to be completely dependent on the substrate binding site in the bЈ domain of PDI, with only minor contributions from the sites in the a and aЈ domains. The bЈ domain is also essential for ⌬-somatostatin or mastoparan peptide binding by PDI, with no direct contributions from the sites in a and aЈ, whereas the binding sites of all three domains contribute to the binding of larger peptides and nonnative proteins (17). The picture that emerges is that PDIsubstrate binding is a complex process. The previous assumption that, of the three binding sites, the one in the bЈ domain has the highest affinity seems to be an oversimplification. Rather, it is likely that for each specific substrate the binding sites in a, bЈ, and aЈ contribute to different extents. This combination not only resolves the potential dichotomy in PDI function, but if each site has a different specificity, it also potentially gives PDI the breadth of binding specificity it needs to act as a folding catalyst for a wide range of protein substrates. In addition, this combination of three substrate binding sites provides an explanation for why ERp57 is still able to act as a thiol-disulfide oxidoreductase on a range of peptide and protein substrates (40 -42) although its substrate binding site in the bЈ domain has become specialized for interacting with the lectins calreticulin and calnexin (35).
PDI contains a fourth domain, the b domain, which has not been considered here. Previous studies have shown that the b domain does not contribute to the binding of peptides or nonnative proteins (17) or to any of the thiol-disulfide oxidoreduc-tase activities of PDI (25). To date, all domains that have a thioredoxin-like fold and interact with proteins and whose structure has been resolved include a cis-proline at the beginning of strand ␤ 4 . This cis-proline has been implicated in substrate binding for several thioredoxin superfamily members, including DsbA, thioredoxin, glutaredoxin, and glutathione Stransferase (27)(28)(29)(30)43). A proline can be found in a similar position in the model structure for the bЈ domain of PDI (16) and is conserved in the a, bЈ, and aЈ domains of all of the catalytically active human PDI family members with the exception of the bЈ domain of ERp72 (see Refs. 44 and 45). No modification of these proline residues was undertaken here or in any previous studies on PDI-substrate interactions, since mutation of a cis-proline residue is very likely to have a direct effect on the structure of the protein. The b domain is the only domain in PDI that does not have a proline at this position, and the b domains of the other members of the human PDI family also lack a proline residue at this position (see Ref. 22). Therefore, the b domain may not be involved in protein-substrate interactions. Since the b domain also lacks a catalytic site, it is unclear what its role in PDI function may be, unless it plays a structural role (e.g. ensuring the correct orientation of the substrate binding sites and catalytic sites in the a, bЈ, and aЈ domains with respect to each other. Only the three-dimensional structure of a full-length catalytically active PDI family member will eventually enable us to resolve the role of the b domain, and since these have been impervious to crystallization attempts for over 30 years, it is likely that PDI will retain a few of its mysteries for some time yet.