REVIEW ARTICLE A structural overview of the PDI family of proteins Guennadi Kozlov, Pekka Ma ¨ a ¨ tta ¨ nen, David Y. Thomas and Kalle Gehring Department of Biochemistry, Groupe de Recherche Axe ´ sur la Structure des Prote ´ ines, McGill University, Montre ´ al, Que ´ bec, Canada Introduction Protein disulfide isomerase (PDI) was the first protein folding catalyst discovered [1]. Since its discovery more than 40 years ago, studies of this remarkable enzyme have shown that PDI acts as a dithiol–disulfide oxido- reductase that is capable of reducing, oxidizing and isomerizing disulfide bonds. Independently of its redox activity, PDI can also act as a chaperone both in vitro [2] and in vivo [3]. PDI is the founding member of a family of 20 related mammalian proteins that are chiefly located and function in the endoplasmic reticu- lum (ER) (Fig. 1). PDI family members are abundant and play a significant role in protein folding and qual- ity control in the calcium-rich oxidative environment of the ER [4]. The members vary in length and domain arrangement, but share the common structural feature of having at least one domain with a thioredoxin-like structural fold, babababba. Most PDI family members contain both catalytic and non-catalytic thioredoxin- like domains that are identified as either a or b based on the presence or absence of a catalytic motif, with use of the prime symbol to indicate their position in the protein. PDI has four such domains, a, b, b¢ and a¢ [5]. The a and a¢ domains functionally resemble thiore- doxin, and each contains catalytic Cys-x-x-Cys motifs that react with thiols of newly synthesized proteins to confer disulfide oxidoreductase activity. The b and b¢ domains, although structurally similar to thioredoxin, do not contain catalytically active cysteines. Instead, the b and b¢ domains appear to act as spacers, and are often responsible for substrate recruitment [6–9]. The non-catalytic domains have lower sequence identity than the catalytic domains across PDI family members, and show more structural variability. For instance, the b domain of ERp44 (ER protein 44 kDa) has an unorthodox arrangement of the secondary structure elements, bbabbba [10]. The family members most Keywords disulfide; endoplasmic reticulum; ERp44; ERp57; ERp72; PDI; protein folding; protein structure; thioredoxin-like; X-ray crystallography Correspondence K. Gehring, Department of Biochemistry, McGill University, 3649 Promenade Sir William Osler, Montre ´ al, Que ´ bec, H3G 0B1, Canada Fax: +1 514 398 2983 Tel: +1 514 398 7287 E-mail: kalle.gehring@mcgill.ca (Received 8 April 2010, revised 11 July 2010, accepted 27 July 2010) doi:10.1111/j.1742-4658.2010.07793.x Protein disulfide isomerases (PDIs) are enzymes that mediate oxidative pro- tein folding in the endoplasmic reticulum. Understanding of PDIs has historically been hampered by lack of structural information. Over the last several years, partial and full-length PDI structures have been solved at an increasing rate. Analysis of the structures reveals common features shared by several of the best known PDI family members, and also unique features related to substrate and partner binding sites. These exciting breakthroughs provide a deeper understanding of the mechanisms of oxida- tive protein folding in cells. Abbreviations CNX, calnexin; CRT, calreticulin; ER, endoplasmic reticulum; PDI, protein disulfide isomerase; SAXS, small-angle X-ray scattering. 3924 FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS similar to PDI have a short interdomain region between the b¢ and a¢ domains that is termed the x-linker [11]. Recently determined structures of several PDI family members have revealed their detailed architecture and led to mechanistic insights into their function. The most exciting breakthrough came when the full-length crystal structure was solved. Determination of the structure of yeast Pdi1p (yeast PDI) showed that its four domains form an overall ‘U’ shape, suggesting how substrates may be positioned relative to the two catalytic domains [12]. The structure of ERp44 showed that its three thioredoxin-like domains (abb¢) are arranged like a cloverleaf. A long C-terminal tail folds back and makes contacts with the a and b¢ domains [10]. This capping function of the C-terminal tail may have assisted the successful crystallization and struc- ture determination. Recently, the crystal structure of full-length human ERp57 (ER protein 57 kDa, also known as protein disulfide-isomerase A3 or 58 kDa glucose-regulated protein) in complex with tapasin was also solved [13]. The ERp57 structure provides the first structural insight into protein binding by the catalytic domains. Two other crystal structures of PDI-like pro- teins have been solved: human ERp29 (ER protein 29 kDa) [14] and yeast Mpd1p (member of the protein disulfide isomerase family 1) [15]. Additionally, the structures of the non-catalytic fragments of human ERp57 [16], human PDI [8,17] and rat ERp72 (ER protein 72 kDa, also known as protein disulfide-iso- merase A4) [18] have been determined. Here we review these structural studies, with special focus on mammalian PDIs and what can be learnt from their similarities and differences. Excellent reviews of the process of disulfide bond formation in the ER and the biology of PDIs are also available [19,20]. What constitutes a PDI family member? The PDI family contains both thiol-reactive and thiol non-reactive members, and this has led to some confusion. A thioredoxin-like domain has been loosely Fig. 1. Domain architectures of human disulfide isomerases (PDIs). Catalytic thiore- doxin-like domains (a and a¢) are colored pink, and non-catalytic domains (b and b¢) are blue. The first domain of PDILT, which does not contain active site cysteines, is hatched to indicate its strong similarity to the a domains of other PDIs. Yellow boxes correspond to the linker between the b¢ and a¢ domains (x-linker). The DnaJ domain of ERdj5 and the C-terminal helical domain of ERp29 are shown in green. White boxes indicate transmembrane domains. The sec- ond column lists available structures of mammalian PDIs. Among notable non-mam- malian structures are two structures of yeast PDI crystallized at different tempera- tures (PDB accession numbers 2b5e and 3boa) and yeast Mpd1p (PDB accession number 3ed3). The structure of full-length ERp57 (PDB accession number 3f8u) was determined in complex with tapasin. G. Kozlov et al. PDI family of proteins FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS 3925 defined as anything with a predicted thioredoxin-like structure (based on sequence). However, some authors have used a more specific definition of a thioredoxin- like domain as one capable of reacting with cysteines [21]. This definition appears to make the most sense, as members of a disulfide isomerase family should be capable of reacting with cysteine side chains. However, the founding member of this family, PDI, also exhibits non-specific polypeptide-binding chaperone activity [22]. Thus, a more inclusive definition of PDI family members would comprise proteins that contain non-thiol-reactive thioredoxin-like domains with chap- erone-like activities for ER folding and secretion of proteins. This includes ERp29 [23] and ERp27 (ER protein 27 kDa) [24], two PDI family members that do not contain thiol-reactive active sites. The proteins PDILT (protein disulfide isomerase-like protein of the testis) and TMX2 (thioredoxin-related transmembrane protein 2) have catalytic motifs, Ser-x-x-Cys, that lack the N-terminal cysteine required for full oxidoreduc- tase activity. However, because they contain C-termi- nal cysteines, and PDILT has been shown to form mixed disulfides with partners and substrates in vivo [25], we categorize these as thiol-reactive. Their contri- butions to oxidative protein folding in cells remain unclear, and because of their inability to act as oxido- reductases, their chaperone functions are probably more important [26]. Similar considerations apply to ERp44, AGR2 (anterior gradient protein 2 homolog), AGR3 (anterior gradient protein 3 homolog) and TMX5 (thioredoxin-related transmembrane protein 5), which c ontain Cys-x-x-Ser motifs that lack the C-terminal cysteine. PDIs are a diverse family PDI family members have functions as diverse as their sequences and domain arrangements. In Fig. 1, we dis- tinguish between thiol-reactive and non-reactive PDI family members. Most PDIs contain more than one active site, and usually contain a combination of active and inactive thioredoxin-like domains. The inactive domains perform functions such as substrate or part- ner recruitment. Importantly, although in vitro activi- ties have been demonstrated for most PDIs, their function in vivo is more difficult to determine, and may be intrinsically more complex, involving other partners or specific conditions. For example, while PDI gener- ally promotes protein folding, it can act as an unfol- dase, favoring ER exit of cholera toxin [27]. The specific function of ERp57 is unclear, but its gene knockout is embryo-lethal at day 13.5 [28] for reasons that may relate to its modulation of STAT3 (signal transducer and activator of transcription 3) signaling [29]. A conditional B-cell knockout has adverse effects on folding of glycosylated influenza virus hemaggluti- nin, but little effect on folding and secretion of the Semliki Forest virus coat spike proteins p62 and E1. ERp57 knockout did not change ER morphology or function drastically, and ER stress levels were not affected, suggesting more functional overlap between PDIs than previously appreciated. Remarkably, treat- ment with castanospermine rescued the folding of viral hemagglutinin in ERp57 ) ⁄ ) mouse fibroblasts [30], pre- sumably by preventing its entry into the calnexin cycle, and thereby allowing other disulfide isomerases to act on it. These results highlight the need for in vivo stud- ies to clarify the functions of the various PDIs, and the difficulty in assigning functions to PDIs based on their in vitro activities or structures alone. With nota- ble exceptions such as ERp57, ERdj5 [31], and AGR2, which is involved in the production of intestinal mucin [32,33], use of knockouts to address the in vivo functions of mammalian PDIs has not been reported. Catalytic sites In thioredoxin-like domains, the conserved catalytic Cys-x-x-Cys motif is found at the N-terminus of a long helix, a2 (Fig. 2). Within the catalytic motif, the two cysteines play distinct roles. The N-terminal cysteine forms a mixed heterodimer with a protein substrate, while the C-terminal cysteine is involved in substrate release [34]. Several other residues in the catalytic site contribute to the reaction mechanism. A conserved glutamate positioned below the C-terminal cysteine functions in proton transfer during substrate release [35]. A neighboring arginine modulates the pK a of this catalytic cysteine by its placement in the active site [36,37]. The recent determination of the structure of the ERp72 catalytic domains allowed a glimpse into the effects of local rearrangements of the N-terminal part of helix a2 on positioning of the conserved argi- nine residue [38]. In the a 0 domain, the arginine side chain is surface-accessible, but in the a domain, the equivalent arginine points towards the catalytic site and forms a salt bridge with the glutamate residue, Glu200, that is implicated in substrate release (Fig. 2). This suggests that coordinated arginine–glutamate interactions may serve to modulate the catalytic activ- ity of protein disulfide isomerases. Substrate binding sites One major area of interest has been how PDI family members recognize and bind substrate molecules. In PDI family of proteins G. Kozlov et al. 3926 FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS pioneering studies, the substrate binding site of mammalian PDI was identified by cross-linking with radiolabeled model hydrophobic peptides that mimic unfolded proteins [6]. The major binding site for unfolded proteins was shown to be the b¢ domain, and this was confirmed by structural studies that used NMR chemical shifts to map the binding site of unfolded RNase A and peptide ligands onto the struc- ture of the bb¢ fragment of PDI [8,9]. The binding site consists of a large hydrophobic cavity between helices a1 and a3, comprising Phe223, Ala228, Phe232, Ile284, Phe287, Phe288, Leu303 and Met307 side chains (Fig. 3A). Determination of the crystal structure of the b¢x domain of human PDI provided a more detailed insight into substrate binding, as the x-linker is folded back and mimics how hydrophobic stretches bind to the b¢ domain [17]. Specifically, the side chains of Leu343 and Trp347, which are part of the linker between the b¢ and a¢ domains, are inserted into the hydrophobic cavity of the b¢ domain. The binding sur- face appears to be conserved across species, as the same pocket is fully accessible and well positioned for protein substrate binding in the structure of yeast PDI [12]. In human PDI, the b¢a¢ fragment is required for efficient binding of non-native protein substrates [6]. A recent study of Humicola insolens PDI also demon- strated that the a¢ domain assists in substrate binding, as the b¢a¢ fragment shows extensive contacts with the hydrophobic peptide mastoparan [39]. ERp44 contains three thioredoxin-like domains, a, b and b¢, in addition to a C-terminal regulatory domain [10]. There are obvious structural similarities between the b¢ domains of ERp44 and PDI, and most of the substrate binding PDI residues are conserved in ERp44 (Fig. 3). The b¢ domain of ERp44 also has a hydrophobic pocket. As observed with the PDI b¢x fragment, a hydrophobic stretch C-terminal to the b¢ domain of ERp44 folds back and binds to the a1–a3 cavity as a short helical segment using the side chains of Phe358 and Leu361 (Fig. 3B). The C-terminal tail also partly shields a hydrophobic patch of the a domain, and its removal increases the in vitro activities of ERp44 as an oxidase, reductase, isomerase and chaperone [10]. Strikingly, tail-less ERp44 formed mixed disulfides with endogenous proteins in several cell types [10], suggesting that the C-terminal cap of the substrate binding domains contributes to the speci- ficity of ERp44. How the action of the C-terminal tail of ERp44 is regulated in cells is an intriguing question. Structural studies of two other major PDIs, ERp57 and ERp72, showed that they do not contain hydro- phobic pockets [16,18]. Structurally, ERp57 lacks a protein substrate binding site in its b¢ domain. Instead, the a1– a3 surface is mostly negatively charged. The corresponding surface of ERp72 is likewise polar and is unable to bind hydrophobic peptides. The residues Arg398 and Glu459 of ERp72 form a salt bridge to occlude a potential substrate binding cavity (Fig. 3C). This interaction is stabilized by a hydrogen bond between Tyr416 and Glu459. These two positions are characteristic of b¢ domains of protein disulfide isome- rases that do not bind directly to hydrophobic stretches. In ERp57, Gln256 forms a salt bridge with the corresponding Glu310 that is stabilized by a hydro- gen bond with Tyr264 (Fig. 3D). The charged nature of the a1–a3 surface explains the inability of ERp57 and ERp72 to bind model peptides in vitro [40,41]. In contrast, several other PDIs are predicted to bind to substrates via hydrophobic pockets in their b¢ domains. PDIp (pancreas-specific protein disulfide iso- merase, also known as PDIA2), PDILT and ERp27 have mostly identical or similar hydrophobic residues A B Fig. 2. Role of arginines in the ERp72 catalytic thioredoxin-like domains in modulating disulfide isomerization activity. (A) Phe97 and Pro138 of the a 0 domain restrict access of Arg155 to the Cys-x-x-Cys active site. (B) A conformational change in the a domain a2 helix allows Arg270 to enter the hydrophobic core and form a salt bridge with the conserved buried residue Glu200. G. Kozlov et al. PDI family of proteins FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS 3927 as PDI in their hydrophobic pockets (Fig. 3E). PDIp can bind the amphipathic peptide D-somatostatin [42], as can ERp27 [24]. The redox-inactive bb¢ domains of PDIp exhibit chaperone activity in vitro and in vivo [43]. Binding of liver- and testis-specific PDILT to D-somatostatin and unfolded bovine pancreatic trypsin inhibitor has been demonstrated [26], and PDILT has been shown to form mixed disulfides with substrates in HeLa cells via its unusual Ser-x-x-Cys motif [25]. The lack of a hydrophobic substrate binding site in the b and b¢ domains of ERp57 and ERp72 indicates that these PDIs either require protein partners that assist in substrate recognition, or that they directly interact with substrates through the active sites of their cata- lytic domains. ERp29 utilizes an alternative site for substrate recog- nition. The putative peptide binding site of its single thioredoxin-like domain is located in the b2–a2 and a3–b4 loop area of its N-terminal thioredoxin-like domain [14]. This is opposite to the a1–a3 surface used by PDI. ERp29 forms a tight dimer, and its b domain is sufficient for peptide and substrate binding. It binds peptides with two or more aromatic residues, and favors peptides with basic character [14]. The structure of ERp18 revealed that this protein also adopts a thioredoxin-like fold and has a conserved Pro113 that results in an unusually bent a2 helix when ERp18 is in its oxidized form [44]. This conserved pro- line might be important for ERp18 function, although a specific requirement for ERp18 function has yet to be determined. ERp18 shows specificity for a compo- nent of the complement cascade, the pentraxin-related protein PTX3 [45], and has been implicated in the reduction of gonadotropin-releasing hormone [46]. The recently determined ERp57–tapasin structure provided the first structural insights into protein bind- ing by the catalytic domains of a mammalian PDI [13]. In the structure, the a domain of full-length ERp57 is linked to tapasin by a disulfide bond. Tapasin is a chaperone associated with editing the peptide cargo of the major histocompatibility complex class I. ERp57 specificity for tapasin appears to be determined pri- marily by the catalytic domains [47]. It has been sug- gested that the interdomain distance between the a and AB C E D Fig. 3. A cavity on the a1–a3 surface of the b¢ domain defines the ability of the PDI to bind hydrophobic stretches of protein sub- strates. The stretches after the b¢ domain of PDI (A) and ERp44 (B) interact with the hydrophobic surface in the corresponding crystal structures. The residues that bind to the hydrophobic groove are labeled. The cor- responding surfaces in ERp72 (C) and ERp57 (D) are occluded by polar interactions involving conserved glutamates that also form hydrogen bonds with tyrosine side chains. Hydrogen bonds are shown as black dashed lines, and the residues involved are labeled. (E) Structure-based sequence align- ment of the b¢ domains from human PDIs and rat ERp72. The glutamates that contrib- ute to the inability of the ERp57 and ERp72 b¢ domains to bind hydrophobic stretches are highlighted in pink, and their polar equiv- alents are shown in turquoise. The residues potentially involved in substrate binding are highlighted in gray. The alignment includes the N-terminal part of the x-linker that appears to be an integral part of the b¢ domain. The consensus secondary structure is shown above the alignment. PDI family of proteins G. Kozlov et al. 3928 FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS a¢ domains makes ERp57 particularly suited for bind- ing to tapasin; however, the mobility of the a and a¢ domains relative to the bb¢ base for each of these pro- teins (discussed further below) suggests that the a to a¢ interdomain distance is not strictly maintained (Fig. 4A). The suggestion that the interaction is stabi- lized by cumulative complementary protein–protein interactions seems more likely, with the a and a¢ domains together providing an avidity effect that secures binding to tapasin. Although most ERp57 resi- dues within 4.5 A ˚ of tapasin in the complex are con- served in ERp57, ERp72 and PDI, three (K366, C92 and R448) are unique to ERp57. Further studies are required to understand ERp57 specificity for tapasin, but these results illustrate the importance of PDI–sub- strate complementarity within catalytic domains. Protein partner binding sites PDI is highly abundant in the ER, and is a member of distinct protein complexes with specific functions. PDI is the b-subunit of the prolyl-4-hydroxylase complex that is important for hydroxylating proline residues of collagen [48], and also is a subunit of the microsomal triglyceride transfer protein [49]. Binding of PDI to the prolyl-4-hydroxylase complex minimally requires intact b¢ and a¢ domains of PDI, but the assembly and activ- ity of the complex is further enhanced by the addition of a and b domains [50]. Mutagenesis of individual residues in PDI confirmed the importance of the a and a¢ domains in the assembly of an active complex but, surprisingly, mutations in the hydrophobic substrate- binding site in the b¢ domain had no effect [51], for assembly of the prolyl-4-hydroxylase complex. Rather than directly binding to substrates, ERp57 requires a protein partner to assist in substrate protein folding [41]. The partner protein, either calnexin (CNX) or calreticulin (CRT), recruits glycoprotein substrates through a lectin domain. NMR titrations and mutagenesis studies mapped CNX ⁄ CRT binding to a site centered on the N-terminal half of helix a2of the b¢ domain of ERp57 [16]. This area is abundant in positively charged residues and displays charge com- plementarity to the negatively charged tip of the CNX ⁄ CRT P-domain, which is responsible for ERp57 binding [52]. In particular, mutations R282A and K214A in ERp57 abrogate or greatly decrease CNX P-domain binding in vitro [16] and inhibit substrate interactions in vivo [31]. This region of ERp57 has also been shown to mediate binding to the PDI ERp27 [24]. ERp27 has a hydrophobic peptide binding site on its second thioredoxin-like domain and may recruit substrates to ERp57. ERp27 has no catalytic cysteines of its own. A yeast homolog of CNX, Cne1p, interacts with the oxidoreductase Mpd1p. Mpd1p and ERp57 present very different overall architectures. Mpd1p contains only two domains: an N-terminal catalytic domain and a C-terminal non-catalytic domain. The recently deter- mined crystal structure of Mpd1p has a positively charged surface at the beginning of the second thiore- doxin-like domain that has been suggested to be a potential Cne1p binding site [15]. ERp72 also does not possess a hydrophobic sub- strate binding site in its b and b¢ domains. Compared to PDI and ERp57, ERp72 contains an additional cat- alytic a 0 domain at its N-terminus. The recently deter- mined high-resolution crystal structure of the bb¢ fragment of ERp72 reveals strong structural similarity to ERp57 in terms of both the individual domains and relative domain orientation [18]. The ERp72 surface corresponding to the CNX binding site of ERp57 is AB a′ a′ b′ b′ Fig. 4. Involvement of the catalytic domains in protein binding. (A) Representation of the ERp57–tapasin structure showing that only the cat- alytic a and a¢ domains of ERp57 (green) interact with tapasin (turquoise). The ERp57 residues that interact with CNX are shown in blue [16]. (B) Structural model of ERp72 based on structures of the a 0 a (pink), bb¢ (turquoise) and a¢ (yellow) domain fragments overlaid upon full- length ERp57. The catalytic cysteines and adjacent hydrophobic residues in ERp57 and ERp72 are shown in orange and gray, respectively. G. Kozlov et al. PDI family of proteins FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS 3929 negatively charged, and consequently does not bind the CNX P-domain, but shows high sequence conser- vation among ERp72 proteins from various species, suggesting functional importance. Given that ERp72 does not interact with CNX, and lacks a hydrophobic substrate binding site, how might it interact with substrates? Perhaps ERp72 relies on its three catalytic domains for specificity toward sub- strates or partners. The relative contributions of each of the catalytic Cys-x-x-Cys motifs of ERp72 to reduc- tase activity on insulin in vitro suggest unequal contri- butions to binding and catalysis [53]. Cysteine-to-serine mutations in the N-terminal a 0 domain affect the k cat for insulin reduction, but the K m is unaffected. The same mutations in the a domain have intermediate effects on both k cat and K m , while loss of the catalytic motif of the a¢ domain primarily affects K m [53]. These results were interpreted as indicating that the a 0 domain is primarily involved in catalysis, the a domain has intermediate roles in catalysis and binding, and the a¢ domain functions primarily to bind substrates. How- ever, the unequal contributions to binding and cataly- sis of the catalytic domains could also indicate that other structural features are important for the activi- ties. These kinetic studies on ERp72 must be inter- preted carefully, because PDI-catalyzed oxidoreductase reactions do not necessarily follow simple Michaelis– Menten rules. Compared to the other abundant PDIs, relatively few endogenous substrates were identified for ERp72 in HT1080 human fibroblasts [45]. However, due to the use of Cys-x-x-Ala substrate-trapping mutants that form mixed disulfides during reduction or isomerization reactions, substrates oxidized by ERp72 may have been missed. ERp72 may play a more spe- cialized role in protein oxidation, or act on specific substrates not readily detected by the methods used. Further work is required to understand ERp72–sub- strate and ERp72–partner interactions. Recently, the transmembrane PDI TMX4 was char- acterized and found to interact with ERp57 and caln- exin [54]. Unlike ERdj5, over-expression of TMX4 does not accelerate ER-associated decay of the NHK (null-Hong Kong) variant of a1-antitrypsin. Interac- tion of TMX4 with ERp57 was dependent on its cata- lytic active site, suggesting that it may reduce ERp57 in the ER. On the other hand, the interaction of TMX4 with CNX did not require an intact Cys-x-x- Cys motif. Further work is necessary to determine how TMX4 might interact with substrates, although sub- strate recruitment by CNX in a fashion analogous to the ERp57-CNX complex is a possibility. While TMX4 lacks a b¢-like domain that can interact with the P-domain of CNX, other mechanisms are possible. TMX4 does not isomerize scrambled RNase A in vitro, suggesting that it requires a co-factor ⁄ partner for sub- strate recruitment. Interdomain mobility The question of interdomain flexibility is relevant for PDIs that comprise multiple thioredoxin-like domains. Recent crystallographic and small-angle X-ray scatter- ing (SAXS) studies provide support for interdomain mobility. This may be the main reason why PDI was resistant to crystallization efforts for a long time. Only two-four-domain PDIs (ERp57 and yeast PDI) have been crystallized, and both represent partner- or pseudo-substrate-bound structures. ERp57 was crystal- lized as a heterodimer with tapasin, while yeast PDI was crystallized with another yeast PDI molecule mim- icking a bound substrate. Crystallization of the three- domain ERp44 structure was potentially favored by reduced mobility due to binding of the protein C-ter- minus to the thioredoxin-like domains. Apart from these examples, only the two-domain proteins ERp29 and Mpd1p, and protein fragments of ERp57 and ERp72 (bb¢ for ERp57 and a 0 a and bb¢ for ERp72) have been crystallized. The best strategy for crystalliz- ing PDIs with domains connected by flexible linkers appears to be immobilization of their domains via sub- strate or partner binding, or focusing on smaller frag- ments instead of the whole protein. The clearest evidence of interdomain flexibility comes from the two crystal structures of yeast PDI (Fig. 5A) [12,55]. When the structures are superposed, the catalytic domains clearly adopt different positions. The ability of human PDI to adopt open and closed conformations was demonstrated by sedimentation equilibrium and SAXS experiments [56,57]. Although only one crystal form of ERp57 is available, compari- son of a SAXS reconstruction of the protein free in solution with the ERp57 ⁄ tapasin crystal structure sug- gests mobility of the catalytic a and a¢ domains [13,16]. Among PDIs with x-linkers, mobility between the b¢ and a¢ domains is likely to be much more pronounced than between the a and b domains, which have a short interdomain linker. A recent study of domain mobility in human PDI showed that the sites of greatest prote- ase sensitivity are located between the b¢ and a¢ domains [58]. Another striking example of domain flexibility was provided by recent SAXS, crystallographic and NMR studies of ERp72 [18,38]. The crystal structures of the a 0 a and bb¢ fragments allow modeling of the full-length protein by overlaying the a and bb¢ domains onto the ERp57 structure (Fig. 4B). Interestingly, this generates PDI family of proteins G. Kozlov et al. 3930 FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS a potential substrate binding site comprising the cata- lytic a 0 , a and a¢ domains. The ERp72 model presents one possible orientation of the domains, but does not illustrate the full range of possible interdomain confor- mations. SAXS measurements of full-length ERp72 revealed multiple relative orientations of the domains, with the greatest mobility between the a 0 and a domains and the b¢ and a¢ domains [18]. Limited prote- olysis revealed that the a¢ domain is cleaved most rapidly from ERp72, probably due to flexibility of the x-linker (P.M., unpublished data). The recently deter- mined a 0 a structure illustrates the difficulty in drawing conclusions about interdomain mobility based on crys- tal structures alone [38]. Although the a 0 a crystal structure suggests a single conformation, NMR chemi- cal shifts in the a 0 a fragment and a 0 domains suggest that the two domains do not form a rigid pair in solu- tion (G.K., unpublished data). The bb ¢ domains form a rigid spacer between catalytic domains Although future studies will better define the structural diversity of PDIs, an obvious feature is the limited mobility of the bb¢ fragment in the four- and five- domain PDIs. A structural overlay of this region from human PDI, ERp57, ERp72 and yeast PDI showed striking similarity in the domain orientations (Fig. 5). As there are two crystal structures available for yeast PDI (at 4 °C and room temperature) and for ERp57 (full-length and the bb¢ fragment), more conclusive comparisons can be made concerning the rigidity of these bb¢ pairs. The bb¢ domains of the two yeast PDI structures superpose with an rmsd of 1.7 A ˚ over 210 Ca atoms. Much greater variability is observed in the positions of the catalytic a and a¢ domains (Fig. 5A). The bb¢ domain orientation is also similar in human and yeast PDI (Fig. 5B). Likewise, overlay of the bb¢ structures from ERp57 and ERp72 results in an rmsd of 1.7 A ˚ for backbone atoms (Fig. 5C). The difference results from a small (10°) rotation at the interdomain interface. A similar comparison between the two full- length ERp57 molecules in the crystal structure with tapasin shows the bb¢ domains overlay with an rmsd of 1.1 A ˚ [13]. These examples indicate that the bb¢ tan- dem forms a relatively rigid base, providing a spacer for the attachment of more mobile active site domains that can access substrates from opposite sides simulta- neously. The bb¢ domains may also jointly contribute to functional substrate or protein binding sites. As one example, the CNX binding site of ERp57 includes a contribution from Lys214 in the b domain in addition to the residues in the b¢ domain, which form the majority of the binding site [16]. In contrast to the bb¢ base, the catalytic domains in the available structures show a much larger degree of mobility, which may be important for recognition of protein substrates of vari- able sizes as well as adjustment to conformational changes in substrate during folding and disulfide rear- rangement. In the proteins most closely related to PDI, the a–b linker is generally very short, while the b¢–a¢ linker (x-linker) is significantly longer. It is very likely that other PDI-like proteins such as PDIp and PDILT will display a similar structural arrangement of their domains. The recently determined structure of yeast Mpd1p shows a very different orientation of its two thioredox- A BC DE Fig. 5. The non-catalytic bb¢ fragment provides a relatively rigid base in PDIs containing four or five thioredoxin-like domains, while allowing greater mobility of the catalytic domains. (A) An overlay of the yeast PDI structures crystallized at 4 °C (pink) and room tem- perature (yellow) shows high similarity in the orientation of the bb¢ domains and significant differences in orientation of their a and a¢ domains. The catalytic cysteines (orange) of the 4 °C yeast PDI structure face each other. (B) The b and b¢ domains (gray) of human PDI are oriented similarly to those from yeast PDI crystal- lized at room temperature (pink). (C) An overlay of the bb¢ struc- tures of ERp57 (green) and ERp72 (turquoise) shows a very similar domain orientation. (D) An overlay of the bb¢ structures of ERp57 (green) and ERp44 (brown) shows roughly similar domain orienta- tion. (E) Representation of the Mpd1p structure with the N-terminal a domain shown in similar orientation to the b domains in (A)–(D). The structure, colored blue to red from the N- to C-terminus, reveals the C-terminal helix contacts the a domain. G. Kozlov et al. PDI family of proteins FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS 3931 in-like domains, which are locked in a rigid orientation by numerous interdomain contacts (Fig. 5E) [15]. This led to suggestions that other PDIs cannot be reliably modeled using yeast PDI and ERp57 structures. Nev- ertheless, there are several reasons why this argument may not apply to most mammalian PDIs. Mpd1p does not have a close structural homolog in the mammalian ER. Only its N-terminal catalytic domain has signifi- cant sequence homology to human PDIs P5 (39% identity) and ERdj5 (34% identity), while the C-termi- nal non-catalytic domain has no detectable sequence similarity to human proteins. In contrast, mammalian PDIs have significant (30–40%) sequence identity (Table 1) that translates into structural similarity. Although the non-catalytic domains have low sequence similarity, there is strong sequence similarity between the human PDI bb¢ domains (residues 136–354) and PDIp (39% identity), ERp27 (32%), PDILT (27%) and ERp72 (24%). The sequence identity of 28% between the bb¢ domains of ERp57 and ERp72 results in a strikingly similar domain orientation, and the bb¢ domains of PDI, ERp57 and ERp72 adopt similar rel- atively rigid conformations. Based on this, the non-cat- alytic domains of ERp27 are expected to adopt a very similar structure. Although the bb¢ domains of ERp44 show a somewhat different interdomain angle, the ori- entation is still largely similar to other known struc- tures of mammalian PDIs (Fig. 5D), and could reflect the effect of the protein C-terminus. In contrast, the a–b interfaces of PDIs are more dif- ficult to compare to one another. Indeed, even within the same protein, the a–b or b¢–a¢ interfaces assume different orientations (Fig. 5A). As Mpd1p has only two domains, it may adopt a unique orientation to provide the catalytic, substrate binding and partner binding sites afforded by the four domains in ERp57. These observations suggest that Mpd1p is a structural outlier when compared to mammalian PDIs. Table 1. Sequence identity (%) between human protein disulfide isomerases. Domains PDI PDIp PDILT ERp57 ERp72 PDI (residues 26–471) abb¢a –49 32 34 35 PDIp (residues 44–492) abb¢a¢ 49 – 33 30 31 PDILT (residues 45–490) abb¢a¢ 32 33 – 23 25 ERp57 (residues 27–478) abb¢a¢ 34 30 23 – 42 ERp72 (residues 179–632) abb¢a¢ 35 31 25 42 – A BC DE Fig. 6. Structural organization of the x-linker. (A) Representation of the full-length ERp57 structure with the region recognized as the x-linker colored in red. (B) The N-terminal part of the x-linker folds against the b¢ domain in the structures of the ERp57 bb¢ fragment (blue), full-length ERp57 with tapasin (green) and the ERp72 bb¢ fragment (turquoise). The side chain of Leu361 in the ERp57 x-lin- ker inserts into a cavity formed by three aromatic side chains. This interaction is also observed in ERp72, involving residues Leu508 of the x-linker and residues Phe484, Phe499 and Phe503 of the b¢ domain. Towards the middle of the x-linker, Tyr364 of ERp57 (and Val511 of ERp72) fit into a small hydrophobic pocket below the C-terminus of helix a3 of the b¢ domain. For clarity, ERp72 residues are not labeled and side chains of the ERp57–tapasin structure are not shown. (C) The structures of the b¢ domains of human PDI (gray) and the two crystal forms of yeast PDI (yellow and pink) show similar interactions. Ile334 of the x-linker is inserted into the pocket formed by the three aromatic side chains Phe325, Phe329 and Tyr310 of human PDI (gray). In yeast PDI, the corresponding residues are Ala361 from the x-linker and Tyr325, Leu352 and Phe356 of the b¢ domain. (D) An overlay of the a¢ domain from full- length ERp57 (green) and the isolated domain (brown) of ERp57 shows a similar structure for the C-terminal part of the x-linker between residues L365 and P377. The N-terminal part of the x-lin- ker is disordered in the NMR structure of the isolated a¢ domain. Only one of many possible conformations is shown. (E) An overlay of the a¢ domains of full-length yeast PDI crystallized in various crystal forms at 4 °C (yellow) and room temperature (pink) shows very similar conformations of the C-terminal region of the x-linker between Ser367 and Ser377. PDI family of proteins G. Kozlov et al. 3932 FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS The x-linker consists of distinct structural regions The x-linker is a conserved stretch of approximately 20 residues connecting the b¢ and a¢ domains [5]. Recent progress in the structural characterization of a number of PDIs has provided new insights into the structural role of this region. The structures of the full-length proteins ERp57 and yeast PDI reveal that the linker consists of two distinct regions (Fig. 6A) [12,13]. The N-terminal part (approximately seven residues) folds onto the b¢ domain and runs perpendicular to the strand b5. The interactions are mostly mediated by the hydrophobic residues (Leu361 and Tyr364 in ERp57) that contact the hydrophobic surface at the edge of the b sheet (Fig. 6B). In ERp57, the side chain of Leu361 occupies a hydrophobic pocket formed by the aromatic rings of Phe336, Phe352 and Tyr356 (Fig. 6B). These interactions are conserved in the structures of ERp72 and PDI, suggesting that the N-terminal region of the linker is an integral part of the b¢ domain (Fig. 6B,C). In agreement with this conclusion, removal of the x-linker decreases the midpoint for denaturation of the PDI b¢ domain from 2.32 m guanidinium chloride to 1.65 m [7]. In contrast, the C-terminal portion of the x-linker takes on strikingly different conformations in the structures of the isolated b¢ domain and full-length proteins. In the crystal structure of the b¢x fragment of human PDI, the C-terminal part of the x-linker turns back to contact the b¢ hydrophobic surface between helices a1 and a3 [17]. In the full-length proteins ERp57 (Fig. 6D) and yeast PDI (Fig. 6E), the C-termi- nal half of the x-linker is structurally associated with the a¢ domain, and displays an irregular conformation without conserved salt bridges or hydrophobic interac- tions with the a¢ domain. Overlay of structures from full-length ERp57 and the isolated ERp57 a¢ domain (PDB accession numbers 3f8u and 2dmm) shows that the linker structure is preserved even in the absence of the preceding domain (Fig. 6D). At least three NMR solution structures of a¢ domains from human PDI (PDB accession number 1x5c), human ERp72 (PDB accession number 2dj3) and H. insolens PDI (PDB accession number 2djj) have been obtained without residues comprising the x-linker. The a¢ domain from rat ERp72 is also well-folded on its own [38]. This sup- ports the idea that the C-terminal part of the x-linker loosely interacts with the a¢ domain and does not con- tribute to its structural integrity. As discussed previ- ously, studies with both PDI and ERp72 have shown significant interdomain mobility at the b¢a¢ domain interface. N- and C-terminal extensions A number of PDIs contain N- or C-terminal tails out- side the thioredoxin-like domains (Fig. 1). The C-ter- minal tail of yeast PDI forms a protruding a helix [12]. The C-terminus of ERp44 forms short helical turns while folding back and interacting with the b¢ and a domains [10]. Likewise, the C-terminus of Mpd1p binds to the N-terminal domain as an a helix [15]. Despite these examples, the tails of many mammalian PDIs are unlikely to be structured due to their low sequence complexity and highly charged nature. In particular, the C-terminal tail of PDI and N-terminal tail of ERp72 are very acidic, while the C-terminus of ERp57 is positively charged. NMR spectra suggest that the above-mentioned acidic stretches are unstruc- tured in solution (unpublished data). They could become more structured during interaction with ligands or protein partners. These acidic extensions have been previously implicated in calcium binding, and recent circular dichroism measurements showed that a similar extension of the ER luminal chaperone CRT becomes structured upon binding calcium [59]. Although not generally critical for the disulfide isomer- ase activity, the N- and C-termini of PDIs may also be important for mediating protein–protein interactions with other ER chaperones [60]. Concluding remarks The mammalian PDI family currently consists of 20 proteins with diverse functions in oxidative folding of protein substrates in the ER. As the availability of structures for the PDI family grows, the functions of its members are becoming clearer. Most PDIs con- sist of multiple thioredoxin-like domains with a similar organization: central non-catalytic domains that often form a rigid scaffold for binding substrate or partner chaperones, surrounded by more mobile catalytic domains with active site cysteines. The mobility of cat- alytic domains may be beneficial when acting upon incorrectly disulfide-bonded proteins and substrates of various sizes. This model of PDI function is consistent with sequence analysis showing that the non-catalytic domains have the greatest sequence diversity while the catalytic domains are more highly conserved, especially in regions flanking the Cys-x-x-Cys catalytic sites. It is currently unclear why most PDIs have multiple catalytic domains. One possibility is that these domains are involved in oxidative folding of complex substrates with many disulfide bonds. Another possi- bility is that multiple catalytic domains enhance the avidity for substrates, or provide modules that G. Kozlov et al. PDI family of proteins FEBS Journal 277 (2010) 3924–3936 ª 2010 The Authors Journal compilation ª 2010 FEBS 3933 [...]... Karala AR, Lappi AK & Ruddock LW (2010) Modulation of an active-site cysteine pKa allows PDI to act as a catalyst of both disulfide bond formation and isomerization J Mol Biol 396, 883–892 Kozlov G, Azeroual S, Rosenauer A, Maattanen P, Denisov AY, Thomas DY & Gehring K (2010) Structure of the catalytic a0 a fragment of the protein disulfide isomerase ERp72 J Mol Biol 401, 618–625 Serve O, Kamiya Y, Maeno... Helenius A, Wuthrich K & Ellgaard L (2002) TROSY-NMR reveals interac- 3936 53 54 55 56 57 58 59 60 tion between ERp57 and the tip of the calreticulin P-domain Proc Natl Acad Sci USA 99, 1954–1959 Satoh M, Shimada A, Keino H, Kashiwai A, Nagai N, Saga S & Hosokawa M (2005) Functional characterization of 3 thioredoxin homology domains of ERp72 Cell Stress Chaperones 10, 278–284 Sugiura Y, Araki K, Iemura S,... isomerase is both an enzyme and a chaperone FASEB J 7, 1515–1517 23 Mkrtchian S & Sandalova T (2006) ERp29, an unusual redox-inactive member of the thioredoxin family Antioxid Redox Signal 8, 325–337 24 Alanen HI, Williamson RA, Howard MJ, Hatahet FS, Salo KE, Kauppila A, Kellokumpu S & Ruddock LW (2006) ERp27, a new non-catalytic endoplasmic reticulum-located human protein disulfide isomerase family. .. clues as to how interactions with substrates are mediated, whether directly or through partner proteins Although many PDI structures remain unsolved, the recent surge of structural studies of PDIs has led to clear advances in understanding how PDIs function in cells 8 9 10 Acknowledgements K.G and D.Y.T acknowledge the financial support received from the Canadian Institutes of Health Research K.G is a Chercheur.. .PDI family of proteins G Kozlov et al determine substrate or partner specificity Structural studies of PDI substrate complexes using trapping mutants may provide further insight into the roles of the multiple catalytic domains PDIs are the workhorses of oxidative protein folding in cells Future research should better define the pool of substrates for each particular PDI Structures provide valuable... Schrag JD, Pollock S, Cygler M, Nagar B, Thomas DY & Gehring K (2006) Crystal structure of the bb’ domains of the protein disulfide isomerase ERp57 Structure 14, 1331–1339 Nguyen VD, Wallis K, Howard MJ, Haapalainen AM, Salo KE, Saaranen MJ, Sidhu A, Wierenga RK, Freedman RB, Ruddock LW et al (2008) Alternative conformations of the x region of human protein disulphideisomerase modulate exposure of the. .. H, Anelli T, Degano M, Ronzoni R, Sitia R, Sun F & Wang CC (2008) Crystal structure of human ERp44 shows a dynamic functional modulation by its carboxy-terminal tail EMBO Rep 9, 642–647 Freedman RB, Gane PJ, Hawkins HC, Hlodan R, McLaughlin SH & Parry JW (1998) Experimental and theoretical analyses of the domain architecture of mammalian protein disulphide-isomerase Biol Chem 379, 321–328 Tian G, Xiang... Maeno A, Nakano M, Murakami C, Sasakawa H, Yamaguchi Y, Harada T, Kurimoto E, Yagi-Utsumi M et al (2010) Redox-dependent domain rearrangement of protein disulfide isomerase coupled with exposure of its substrate-binding hydrophobic surface J Mol Biol 396, 361–374 Kramer B, Ferrari DM, Klappa P, Pohlmann N & Soling HD (2001) Functional roles and efficiencies of the thioredoxin boxes of calcium-binding proteins. .. Holmgren A (1997) Effects of buried charged groups on cysteine thiol ionization and reactivity in Escherichia coli thioredoxin: structural and functional characterization of mutants of Asp26 and Lys57 Biochemistry 36, 2622–2636 Lappi AK, Lensink MF, Alanen HI, Salo KE, Lobell M, Juffer AH & Ruddock LW (2004) A conserved arginine plays a role in the catalytic cycle of the protein disulphide isomerases J... contribution of ERdj5/JPDI to endoplasmic reticulum protein quality control in the salivary gland Biochem J 425, 117–125 32 Park SW, Zhen G, Verhaeghe C, Nakagami Y, Nguyenvu LT, Barczak AJ, Killeen N & Erle DJ (2009) The protein disulfide isomerase AGR2 is essential for PDI family of proteins 33 34 35 36 37 38 39 40 41 42 43 44 production of intestinal mucus Proc Natl Acad Sci USA 106, 6950–6955 Zhao F, Edwards . REVIEW ARTICLE A structural overview of the PDI family of proteins Guennadi Kozlov, Pekka Ma ¨ a ¨ tta ¨ nen, David Y. Thomas and Kalle Gehring Department of. of the process of disulfide bond formation in the ER and the biology of PDIs are also available [19,20]. What constitutes a PDI family member? The PDI family