Structure of PICK1 and other PDZ domains obtained with the help of self-binding C-terminal extensions



PDZ domains are protein–protein interaction modules that generally bind to the C termini of their target proteins. The C-terminal four amino acids of a prospective binding partner of a PDZ domain are typically the determinants of binding specificity. In an effort to determine the structures of a number of PDZ domains we have included appropriate four residue extensions on the C termini of PDZ domain truncation mutants, designed for self-binding. Multiple truncations of each PDZ domain were generated. The four residue extensions, which represent known specificity sequences of the target PDZ domains and cover both class I and II motifs, form intermolecular contacts in the expected manner for the interactions of PDZ domains with protein C termini for both classes. We present the structures of eight unique PDZ domains crystallized using this approach and focus on four which provide information on selectivity (PICK1 and the third PDZ domain of DLG2), binding site flexibility (the third PDZ domain of MPDZ), and peptide–domain interactions (MPDZ 12th PDZ domain). Analysis of our results shows a clear improvement in the chances of obtaining PDZ domain crystals by using this approach compared to similar truncations of the PDZ domains without the C-terminal four residue extensions.

PDZ domains are composed of ∼90 amino acids that act as scaffolds for promoting protein–protein interactions (Harris and Lim 2001; Sheng and Sala 2001; Hung and Sheng 2002; van Ham and Hendriks 2003). They consist of six β-strands (βA–βF) and 2 α-helices (αA and αB), which fold into a compact domain with a clearly defined groove for binding peptides (Doyle et al. 1996; Morais Cabral et al. 1996). In the most common form of interaction they bind the C terminus of the interaction partner. Three different C-terminal consensus sequences have been identified for binding to PDZ domains, generally referred to as class I, class II, and class III (Songyang et al. 1997; Stricker et al. 1997; Hung and Sheng 2002). These motifs are: class I, -(S/T)XΦ; class II, -ΦXΦ; and class III, -(D/E)XΦ where Φ represents any hydrophobic residue. In addition, a small number of examples of PDZ domains binding to non-C-terminal residues have been identified, such as the interaction between nNOS and syntrophin (Hillier et al. 1999). In the class I interaction the (S/T) at position −2 (where position 0 represents the C-terminal residue of the bound protein/peptide) forms a hydrogen bond with a histidine at αB1 (the first residue on helix αB) of the PDZ domain. In the class II interaction, the hydrophobic residue at position −2 interacts with hydrophobic residues on αB. In class III, the (D/E) interacts with an αB1 tyrosine side chain. These C-terminal consensus motifs consist of three residue sequences; however, it has been shown that residues up to position −8 can influence binding specificity (Songyang et al. 1997).

Many PDZ domain-containing proteins possess multiple PDZ domains (Fig. 1). To distinguish between differing PDZ domains within the same protein we have adopted the naming convention of “protein@number”; hence, MPDZ@7 is the seventh PDZ domain of MPDZ (counting from the N terminus). In this study the PDZ domains that were structurally characterized were from the proteins PICK1, DLG2, MPDZ, and NHERF-2, domains that are structurally diverse based on amino acid sequence and also represent potential therapeutic targets due to the proteins that they bind.

Figure Figure 1..

Domain organization of the targeted PDZ domain proteins.

PICK1 is composed of a PDZ domain and a BAR domain (Peter et al. 2004) (Fig. 1). PICK1 is so named because it binds protein kinase C, alpha (protein interacting with C-Kinase, PKCα), which it does through its PDZ domain (Staudinger et al. 1995, 1997). However, PICK1 is known to have many binding partners (Dev 2004; Kim and Sheng 2004; Madsen et al. 2005), of which the most studied is the GluR2 subunit of the AMPA receptor. This interaction between PICK1 and GluR2 is important for cerebellar long-term depression (Steinberg et al. 2006; Yawata et al. 2006). Furthermore, blocking the interaction between GluR2 and the PICK1 PDZ domain by use of a decapeptide prevented the insertion of AMPARs that lack GluR2 and the resultant changes in the AMPAR:NMDAR ratio caused by cocaine, factors that may be involved in this drug's long-term addiction (Bellone and Lüscher 2006). The interaction between PICK1 and AMPA is also part of the increasing evidence for a link between PICK1 and schizophrenia: The pick1 gene is itself located at a genetic locus with links to schizophrenia; two associations between SNPs in the PICK1 PDZ domain and schizophrenia have been reported (Hong et al. 2004; Fujii et al. 2006); and AMPA, kainate (both loci for schizophrenia), DAT, and serine racemase all interact with PICK1 and have strong links to schizophrenia (Dev and Henley 2006).

While PICK1 binds both class I and class II motifs, as well as nonclassical PDZ-binding motifs, it has been shown to bind a prototypical class II sequence with 10-fold greater affinity than the class I sequence of PKCα (Madsen et al. 2005). Furthermore, the single-point PICK1 mutation K27E, located on the carboxylate-binding loop, abolished interaction with GluR2 but not with PKCα (Dev et al. 2004). Such specificities, together with the already successful use in vivo of peptide blockers of PICK1 (Bellone and Lüscher 2006) give hope that PICK1 may be a viable drug target for treatment of schizophrenia in the future (Dev and Henley 2006).

DLG2 (discs, large homolog 2, also known as PSD-93) and DLG3 (discs large homolog 3) are members of the membrane-associated guanylate kinase (MAGUK) family. They both contain three PDZ domains in addition to a guanylate kinase and an SH3 domain (Fig. 1). The third PDZ domain (in addition to the first two) of DLG2 interacts with the C terminus of the inwardly rectifying potassium channel Kir2.1 (Leyland and Dart 2004).

MPDZ (Multiple PDZ Domain Protein or MUPP1) contains 13 PDZ domains (Fig. 1) (Ullmer et al. 1998). It is concentrated at tight junctions in epithelial cells where it may function as a scaffold to help form the macromolecular assemblies at the tight junction (Hamazaki et al. 2002). The diverse peptide specificities of the various PDZ domains of MPDZ provide the basis for the formation of the macromolecular assembly from a variety of proteins. The human target for MPDZ@7 is as yet unknown; however, it is a target for the oncogenic Adenovirus E4-ORF1 protein (Lee et al. 2000). Two proteins that interact with MPDZ@13 are the coxsackievirus and adenovirus receptor, which mediates cell attachment and infection by coxsackie B viruses and by a number of adenoviruses (Coyne et al. 2004), and TAPP1, which binds specifically to phosphatidylinositol-3,4-diphosphate and may function as a recruitment point for other proteins to the plasma membrane (Kimber et al. 2002).

NHERF-2 (SLC9A3R2 or E3KARP), the Na+/H+ exchanger 3 (NHE3) kinase A regulatory protein contains two PDZ domains (Fig. 1). NHERF-2@2 has a nanomolar affinity for the C terminus of the cystic fibrosis transmembrane conductance regulator (CFTR) (Sun et al. 2000). Since NHERF-2 also binds ezrin, which binds protein kinase A (PKA), it was proposed that the function of these interactions may be to localize CFTR and PKA (Sun et al. 2000). This PDZ domain also interacts with the epithelial calcium channel TRPV5, and together with the serum and glucocorticoid inducible kinase SGK1 is involved in its regulation and therefore in cellular calcium homeostasis (Embark et al. 2004; Palmada et al. 2005). Other membrane proteins that bind NHERF-2@2 include the ATP gated channel P2Y1 (Fam et al. 2004), the sulphate transporter SLC26A3 (Down-Regulated in Adenoma, DRA) (Lamprecht et al. 2002), and the lysophosphatidic acid receptor LPA2 (Oh et al. 2004).

According to the SMART database (Schultz et al. 1998) there are ∼124 proteins containing PDZ domains in the human genome, with a total of ∼234 PDZ domains. There is now a growing interest in the possibility of using the interaction between PDZ domains and their binding partners as a drug target (Dev 2004; Wen et al. 2006), with the feasibility of this approach having been demonstrated for PICK1 (Bellone and Lüscher 2006). To design suitable inhibitors structural information on the PDZ domains and their interactions is advantageous. In an effort to determine the structures of several PDZ domains by X-ray crystallography we have taken advantage of their functional role as binders of protein C termini to design truncation mutants with appropriate C termini to form intermolecular interactions. This general approach has previously been applied to the first PDZ domain of the Na+/H+ Exchanger Regulatory Factor, NHERF, using the known C-terminal binding peptide from the cystic fibrosis transmembrane conductance regulator (Karthikeyan et al. 2001). Here we extend the procedure to use typical binding peptides that fall into the class I and II PDZ binding motifs, revealing new binding properties for specific PDZ domains and C-terminal peptides as well as providing clues to potential new PDZ binding partners for these domains.

The main objectives of this work were to prove the efficiency of this method for determining structures of PDZ domains, especially where the physiological binding partners are not yet known, and also to provide significant structural characterization of PDZ domains with a view to their future use as targets for drug design. In this respect, a structure with any bound peptide ligand is valuable. We show that this construct design method should be applicable to most PDZ domains, and significantly we show by use of this method an approximate threefold improvement in our ability to obtain PDZ domain crystals, compared to use of similar constructs without C termini designed for self-binding.

Results and Discussion

Construct design and protein purification

In an effort to determine the structures of a selection of PDZ domains we used a multiple construct design approach. For each PDZ domain we chose an appropriate four amino acid C-terminal extension, designed for self-binding to the PDZ domain binding groove, based on known literature interactions of other PDZ domains and sequence similarity. If no information was available on the selectivity of a PDZ domain then we tried both class I and II extensions. The C-terminal extensions incorporated into the constructs were identified using PDZBase (Beuming et al. 2005) and were (1) DSWV (class I from ARVCF, the catenin gene family member; gi:4502247), (2) ETSV (class I from the plasma membrane calcium-transporting ATPase 4; gi: 14,286,105), (3) STRL (class I from the dehydrogenase/reductase SDR family member 2; gi: 3,915,733) and (4) YYKV (class II from ephrin B1; gi:1783361). For the target PDZ domain constructs that provided structures the extensions represented a class II sequence in the case of PICK1 and class I sequences in the case of the other PDZ domains. We made a number of truncation constructs for each PDZ domain, with a variable length between the PDZ domain boundary and the extension. The total number of novel PDZ domains for which we designed constructs was 18, the average number of constructs per PDZ domain was 5.2, and the average number of different extensions per PDZ domain was 1.6. We purified protein from these constructs and set up crystallization trials. The truncation constructs that eventually yielded the structures are presented here (Table 1), although in some cases there were additional constructs of the same domain that also yielded crystals.

Table Table 1.. Characterization of PDZ domains that crystallized using C-terminal extensions
original image

The PDZ domain fragments were overexpressed in Escherichia coli and purified by immobilized metal affinity chromatography using an N-terminal hexahistidine tag, and most were then further purified by gel filtration. For five of the proteins the N-terminal hexahistidine tag was removed by treatment with TEV protease (Table 1).

Structures and crystal packing analysis

As individual PDZ domains display varying degrees of peptide–ligand specificity (Appleton et al. 2006; Zhang et al. 2006), the use of a generic PDZ binding motif in this study would not be expected to change the overall structure of the PDZ domains, which is what we observe. In all of the structures obtained using this approach, a standard PDZ domain topology was observed, with the C-terminal PDZ-binding motif bound in an extended conformation between strand βB and helix βA (the binding groove), as expected (Fig. 2). In the case of PICK1, DLG2@3, and the tetragonal crystal form of MPDZ@7, this C-terminal extension is involved in creating a dimeric arrangement of protein molecules (Fig. 2). For NHERF-2@2, MPDZ@3, and the orthorhombic crystal form of MPDZ@7 the single molecule in the asymmetric unit forms an extended head-to-tail arrangement with molecules aligned along a crystallographic axis (Fig. 3). For the MPDZ@13 there are three molecules in the asymmetric unit, but a similar head-to-tail arrangement with molecules from neighboring asymmetric units is observed, again aligned with a crystallographic axis. For DLG3@1, the molecules pack in a closed tetrameric head-to-tail arrangement with twofold crystallographic symmetry (Fig. 3). For four of the eight PDZ domains there were a number of interesting and localized structural changes, each of which is examined in greater detail below.

Figure Figure 2..

Ribbon diagrams of each of the 10 crystal forms (eight unique PDZ domains). The C-terminal four residues representing the PDZ recognition motif are colored green. Two molecules from each structure are shown, except for DLG3@1 where four molecules are shown; some diagrams represent molecules related by crystallographic symmetry.

Figure Figure 3..

Space filling diagrams of the five crystal forms where the molecules are not arranged as a dimer. The C-terminal four residues representing the PDZ recognition motif are colored purple. In (AD) of the crystal forms the molecules form an extended head-to-tail arrangement, while for (E) DLG3@1, the molecules form a closed tetrameric arrangement.

Although only a qualitative result, we have observed a significantly higher success rate in obtaining protein crystals from PDZ domains with a C-terminal PDZ binding motif than from similar truncation mutants lacking the C-terminal PDZ binding motif. From 81 purified proteins (22 separate PDZ domains) without C-terminal binding motifs put into crystallization trials we obtained three structures (the PDZ domain of RGS3, PDB ID 2F5Y; the second PDZ domain of DLG3, PDB ID 2FE5; and the second PDZ domain of DLG2; PDB ID 2BYG) and poorly diffracting crystals of an additional 2 PDZ domains (MPDZ@7 and MPDZ@12). From 90 purified proteins of mostly the same set of 18 PDZ domains but with C-terminal binding motifs we have obtained the eight structures presented here (10 crystal forms) as well as diffracting crystals of an additional four PDZ domains for which structure determination is in progress. Our statistics therefore indicate that there is at least a three times better chance of obtaining diffracting PDZ domain crystals if using a C-terminal PDZ binding motif, all other factors being equal. Furthermore, the success rates of 44% of PDZ domains and 9% of protein preparations from which we have produced structures represents a point in time and are therefore underestimates of the true efficacy of this approach.

All of these PDZ domain fragments are monomers in solution (according to gel filtration analysis) and so the PDZ binding interactions are apparently strong enough to influence crystallization without being strong enough to extensively alter the solution state properties in the buffer compositions we used in purification. The addition of the C-terminal PDZ binding motifs has therefore had exactly the effect we had hoped for: improved crystallization behavior and observation of the PDZ domain binding interactions.

Flexibility of the C-terminal linker region

The constructs we created had a variable number of residues (typically two, four, or six) forming the linker region between the predicted end of the last β-strand of the PDZ domain (βF) and the four residue C-terminal PDZ binding motifs. Analysis of the truncation constructs from which these crystal structures derive shows that for five out of the eight proteins there are two residues between the predicted end of βF according to GenThreader (McGuffin and Jones 2003) and the binding motif. The exceptions are MPDZ@13 where the linker comprises three residues, MPDZ@12 and DLG3@1 with five, and DLG2@3, which is a special case as there is an additional α-helix C-terminal to βF. For DLG2@3 there are three residues between the end of this α-helix and the binding motif. It therefore appears that five or less residues after the PDZ domain in addition to the four residues of the binding motif provide a suitable length tag to allow the PDZ domain to bind to the binding site of a neighboring molecule without being so long as to create an excessively flexible arrangement less favorable for crystallization. The different crystal packing arrangements of proteins with similar folds and similar length extensions with which to bind to neighboring molecules demonstrates that, although short, the extensions do still provide sufficient flexibility to allow for the different crystal packing arrangements necessary to crystallize different PDZ domains.

In the case of MPDZ@7 and MPDZ@12, we obtained two different crystal forms from each protein. For both pairs of structures the terminal four residues comprising the designed binding motif are bound in the same relative position to the PDZ domain while the remainder of the linker region adopts different relative positions. As well as emphasizing the flexibility of the C-terminal linker it also shows that in these cases the interaction between the C-terminal extension and the binding site is a significant part of the total protein–protein interactions in respect of forming crystal–crystal contacts.

While residues up to position −8 can influence binding specificity (Songyang et al. 1997) all of our structures were obtained with C-terminal extensions shorter than nine residues from the end of the PDZ domain, and in some cases the C-terminal sequences are not of known binding partners. This shows that interactions between the PDZ domain and positions −6 to −8 of a free protein chain are not necessary for forming interactions strong enough to influence crystallization behavior, as was also seen in the previous structure of the first PDZ domain of NHERF-1 (Karthikeyan et al. 2001).

Structure of PICK1

The structure of PICK1 reveals how this PDZ domain can bind proteins with both class I and class II motifs. We crystallized PICK1 with a C-terminal extension containing the class II sequence motif —YYKV. Overall, the structure superimposes well with other PDZ domains. In particular, helix αB, which contains many of the residues involved in binding, shows no significant displacement from its expected position, and since the peptide backbone of the C-terminal class II motif also shows no significant deviation from a canonical position the differences in specificity must be mostly explained by the sequence differences between PICK1 and PDZ domains that bind class I sequences. Only six residues of the C-terminal binding motif make contact with PICK1 (positions 0 to −5) (Fig. 4A).

Figure Figure 4..

Crystal structure of PICK1 in two different orientations. The C-terminal residues of the interacting molecule are colored green.

The most notable sequence change is that PICK1 has a Lys residue in the αB1 position (the first residue on helix αB, Lys83). Primary sequence alignments suggest that PICK1 is the only human PDZ domain to have a Lys at this position. Typically, for a class I interaction the PDZ domain would have a histidine at this position which forms a hydrogen bond with the Ser/Thr at position −2. In PICK1 the aliphatic part of the Lys83 side chain forms a hydrophobic interaction with the Tyr at position −2 of the ligand (closest distance 3.9 Å). Other residues interacting with the Tyr are Val84 and Ala87. Madsen et al. (2005) have confirmed by mutation and binding studies that replacement of this residue with histidine reverts the preference of PICK1 to that for a class I motif. In their molecular model (with the peptide TLRHWLKV, corresponding to the class II C-terminal PDZ binding motif of the dopamine transporter, DAT [SLC6A3]) they predicted the hydrophobic interaction with the aliphatic part of Lys83, and that Lys83 would not interact with other residues in the PDZ domain. Interestingly however, and contrary to prediction, the side chain amine of Lys83 forms hydrogen bonds with the backbone carbonyl of position −4 of the ligand and also with the carbonyl group of Gly39 from the PDZ domain (Fig. 4B).

Thus, compared to the histidine–threonine interaction in canonical class I binding, this single residue change allows PICK1 to maintain an equivalent hydrogen bond to the ligand while adding a hydrophobic interaction to bind efficiently a class II motif. Although in our structure the Tyr at position −2 represents a large hydrophobic ligand, the structure shows that these interactions could be maintained with a smaller hydrophobic residue, and a smaller side chain at position −2 would also allow Val84 to adopt its most common conformation. Ile37, which also forms part of the pocket for position −2 is in two conformations in the structure with a closest distance of 4.3 Å to the Tyr Cβ (Fig. 4). It is likely that a smaller hydrophobic residue such as valine or leucine at position −2 would fit tighter in the pocket, packing closer to Ile37 and fixing its conformation to one with maximal distance to position −2. The freedom given to the side chain of Ile37 by the surrounding protein environment may be a factor in the promiscuity of PICK1 with respect to the −2 position.

Another factor in this promiscuity is the Ala residue on the second turn of helix αB (Ala87) (Fig. 4B). This is a fairly uncommon residue at this position for a PDZ domain, and its small size may assist PICK1 in binding to diverse sequence motifs. Mutation of this residue to Leu eliminated binding to peptides representing the C termini of DAT and PKCα (Madsen et al. 2005), which have Leu or Ser at position −2, respectively. Interestingly, of PDZ domains that have a hydrophobic residue at this position few have a Leu or Ile, and it may be that a Val would still allow binding to class I sequences.

The Val at the C terminus (position 0) easily fits into the hydrophobic binding pocket. The distance of closest approach by the methyl groups of the valine side chain is 3.8 Å to Ile37 and 4.1 Å to Ala87. Based on the likely conformation of a Leu at this position (for instance, as in our structures of MPDZ@7 and NHERF-2@2), a Leu may force the side chain of Ile37 into a conformation with maximal distance to position 0. This explains the observed preference of PICK1 for position 0 residues of Val > Ile > Leu (Madsen et al. 2005), but also suggests that for PICK1 the residue at position 0 influences the specificity at position −2, and vice versa, since a larger residue at position 0 would reduce the conformational flexibility of Ile37 and therefore the variability of the binding pocket for position −2. This is in agreement with the results of Madsen et al. (2005), which showed that the effects of substitution at position 0 were not independent of the sequence of the rest of the peptide. An Ile at position 0 may be more favorable than a Leu due to adoption of a conformation that does not impact on Ile37 attaining its most favorable conformation. Since an Ile residue is observed at a position equivalent to Ile37 in a significant number of PDZ domains, conformational variation at this position may be one of the general mechanisms for controlling specificity of binding.

A PICK1 double mutation, K27A and D28A, has been shown to prevent binding to PKCα (Staudinger et al. 1997) and GluR2 (Xia et al. 1999), while a K27E mutant retained the interaction with PKCα but not GluR2 (Dev et al. 2004). Typically the conserved Lys/Arg equivalent to Lys27 interacts with the COOH terminus of the ligand via a highly ordered water molecule (Doyle et al. 1996). The loss of this interaction will reduce the binding affinity for position 0. These results suggest that this loss is more significant for GluR2 than for PKCα, and that therefore the binding of position 0 is more important for a class II motif than for a class I motif, in agreement with the observation that a substitution to Ala at position 0 was less important for the class I motif of PKCα than for the class II motif of DAT (Madsen et al. 2005). This may be because in class II binding the hydrophobic interactions of position 0 are part of an overall hydrophobic interaction running from position 0 to position −2, and a weakening of this interaction would be more significant than for class I binding where the Ser/Thr residue at position 2 gains additional stabilization from a hydrogen bond to the Lys residue at αB1.

Differences in the carboxylate binding loop of MPDZ@12

The flexibility of loop between βA and βB is shown in a comparison of the two space groups of MPDZ@12. The C-terminal extension in these two structures approaches the PDZ domain binding groove from very different angles (Fig. 2). Only the residues at positions 0 to −4 of the C-terminal extension occupy the same position and conformation, and the βA–βB loop moves substantially to accommodate these different arrangements.

The 12th PDZ domain of MPDZ has an SLGI sequence (residues 1844–1847) in the place of the “GLGF” motif on the carboxylate binding loop. Interestingly, the side chain of Ser1844 forms a hydrogen bond with the backbone NH of Gly1846 of this motif (distances vary from 3.0 Å to 3.3 Å), and also with the COOH terminus of the ligand (distances from 2.8 Å to 3.3 Å) (Fig. 5A). A Ser residue is found in this position in the sequence in >25 human PDZ domains, and there are seven NMR structures of these in the PDB (1VJ6, 1D5G, 1OZI, 1GM1, 3PDZ, and 1Q7X). Our structures of MPDZ@7 (2FCF, 2IWQ) also have a Ser in this position (distances to the equivalent backbone NH are 3.5 Å and 3.4 Å, respectively), as does the first PDZ domain of the related drosophila protein InaD (1IHJ) with a similar distance to the backbone NH. None of these other structures has the Ser involved in hydrogen bonding with the protein backbone as tightly as in this PDZ domain, although not all of those structures have a ligand bound, which may affect the conformation of the carboxylate binding loop.

Figure Figure 5..

(A) Crystal structure of MPDZ@12. Showing the weak electron density for the C-terminal residue of the bound protein C terminus (in gray). The hydrogen bonds affecting the C-terminal carboxylate are shown in dashed red lines. (B) Crystal structure of DLG2@3. The interactions between the C-terminal helix αC (in dark red) and the loop between sheets βB and βC are shown, as is the interaction between the βB–βC loop and the bound protein C terminus (in green).

The two structures of MPDZ@12 both have two molecules in the asymmetric unit. In each of the four protein chains the position 0 residue is slightly disordered, in particular with significantly weaker electron density around the C-terminal carboxylate (Fig. 5A). Generally these atoms would be expected to be well ordered, as seen in all of the other structures presented here. It is interesting to speculate that the shorter distances between the Ser side chain and the backbone NH that bring these atoms within hydrogen bonding distance may have a destabilizing effect on the binding at position 0, either to reduce the overall binding affinity for the ligand of the PDZ domain, or to increase the relative importance of residues other than position 0 as a method of controlling the binding of a ligand.

Influence on peptide binding due to the extra C-terminal helix in DLG2@3

There is an additional α-helix at the C terminus of DLG2@3 (Fig. 5B), which is also predicted for the third PDZ domain of the related MAGUK proteins DLG1 and DLG4. These have a similar domain organization to DLG2 and strong sequence conservation over the C-terminal regions of their third PDZ domains (Figure S-1; see Supplemental Material). This helix αC packs against the top of βC and βB, and the loop between them. It is stabilized in its position by a hydrophobic patch formed by Tyr505 and Phe508 on the helix and Phe445 on βC and Val436 on βB. Phe445 on βC is very well conserved as a hydrophobic residue among PDZ domains. The position of the helix is further stabilized through three salt bridges with amino acids from the βB–βC loop: Glu442–Arg507, Glu442–Lys511, and Asp440–Lys551 (Fig. 5B). Glu439 on the βB–βC loop interacts with the position −4 His of the bound peptide. As the conformation of the βB–βC loop is dictated by the interactions with the αC helix the residues presented toward the binding peptide are likely to have an influence on the specificity with respect to positions −4 onward. Although these positions are not critical for forming crystal contacts, as discussed above, it is interesting to note that in four of our structures (PICK1, NHERF-2@2, DLG3@1, and the C2221 crystal form of MPDZ@12) the residues at positions −5 to −8 pass through the position occupied by the C-terminal helix in this structure.

A new mode of peptide binding for MPDZ@3

MPDZ@3 has an Asn (Asn441) at αB1, where for a typical class I binding PDZ domain there would be a His (Fig. 6A). This suggested that this domain might show different specificity to a typical class I binding PDZ domain. Regardless, we were able to crystallize this domain with a typical class I motif C-terminal extension of -ETSV. The structure reveals that this domain utilizes a new mode of binding, distinct from other PDZ domains of known structure. Asn441, like Lys83 of PICK1, forms two hydrogen bonds with backbone groups: in this case with the backbone carbonyl group of Tyr394, the side chain of Ser403 and the backbone amide group of Gly404 from the βB–βC loop, and is not involved with binding the ligand protein C terminus (Fig. 6B). The side chain of the Ser at position −1 forms a hydrogen bond with Thr390 on βB.

Figure Figure 6..

Crystal structure of MPDZ@3. (A) Superimposition of the structures of MPDZ@3 and MPDZ@7 showing the opposite displacements of the carboxylate binding loop and helix αB of MPDZ@3 compared to canonical positions represented by MPDZ@7. MPDZ@3 is shown in blue, with the bound protein C terminus in green, while MPDZ@7 and its bound protein C terminus are in yellow. (B) Shows the binding interactions of MPDZ@3. The bound protein C terminus is shown in green, Asn441 is shown in pink, and polar interactions are indicated by dashed blue lines.

The C-terminal extension has a Thr at position −2, to which a His at αB1 would be expected to bind. However, in this structure, the helix αB has moved along its axis and twisted slightly, with an overall displacement of about 1.5 Å toward the N-terminal end of the helix by comparison with other PDZ domains as shown in a superimposition with MPDZ@7 (Fig. 6A). The carboxylate binding loop and βB are displaced ∼1 Å in the opposite direction, and since positions 0 to −3 form the usual antiparallel β-sheet interactions with βB, they too are displaced by 1 Å. These two movements combined with a greater kink in the ligand at position −1 have the overall effect of a dramatic alteration of binding, with the position −2 Thr side chain now closer to being above the third turn of helix αB, whereas typically the position −2 Thr is located above the second turn. Unusually, the primary interaction of the Thr with the PDZ domain is not through its hydroxyl group but through its methyl group, which forms a hydrophobic interaction with the position 0 Val (closest distance 3.8. Å), Val445 (3.7 Å), and the aliphatic part of Arg449 (3.9 Å). Notably, the αB helix displacement is not caused by a bulky residue in the central section of αB since a Val corresponding to Val445 is commonly observed in class I binding PDZ domains. The displacement allows the guanidino group of the position −5 Arg to form a planar stacking interaction with the guanidino group of Arg449. Both of these Arg residues form salt bridges with Glu446 (Fig. 6B).

Among the eight PDZ domain structures presented here, these movements are the most substantial. However, the movements do not result in gross structural changes to the PDZ domain and remain targeted to regions (βB and the carboxylate binding loop) that have been shown previously to move depending on the specific bound ligand (Grembecka et al. 2006). This structure shows a new binding arrangement for a standard class I motif and illustrates the flexibility of this PDZ domain, and potentially PDZ domains in general, to vary their binding modes for different C-terminal sequences. The mode of binding for MPDZ@3 also implies that it could be selective even among class I peptides as it may select, based on hydrophobicity, for peptides with a Thr rather than a Ser side chain.


We have shown on a significant sample size that a multiple construct based approach utilizing C-terminal extensions containing predicted PDZ-binding motifs is an effective method to obtain diffracting crystals of PDZ domains. The extensions we used represent general motifs that are known to bind to PDZ domains, rather than sequences of known binding partners, demonstrating that this technique is applicable even when the binding partners of a PDZ domain are unknown. Using this approach we have revealed a new mode of PDZ/peptide binding for MPDZ@3 involving a displaced αB helix, potential extra selectivity determinants through an additional helix for DLG3@1, a mechanism for reduced binding affinity at the carboxylate group for MPDZ@12, and how PICK1 uses the hydrophobic acyl section of the Lys83 side chain to form part of the binding site for a class II peptide.

Materials and Methods

Cloning, expression, and protein purification

DNA for each of the proteins was amplified by PCR from template DNA purchased from Origene except in the case of PICK1 where the template DNA was obtained from the Mammalian Gene Collection (IMAGE Consortium Clone ID 4026028). The four amino acid C-terminal extensions were created by inclusion of 12 extra bases in the reverse primer. The PCR products were incorporated into a home-made vector containing an N-terminal hexahistidine tag and TEV protease tag cleavage site (sequence MHHHHHHSSGVDLGTENLYFQSM), by ligation-independent cloning (Table 1). The resulting plasmids were transformed into E. coli BL21 (DE3) cells containing the pRARE2 plasmid from commercial Rosetta II (DE3) cells. Cultures were grown in shaker flasks containing 1L of terrific broth medium with the number of flasks per clone being dependent on the level of protein expression (previously determined). The cultures were grown at 37°C until an OD600 of 1–3 was reached. The temperature was then reduced to 18°C or 25°C before induction of protein expression with 1 mM isopropyl-β-D-thiogalactopyranoside. Cells were grown for an additional 4 h (25°C; DLG2, DLG3, PICK1) or overnight (20°C; NHERF-2, MPDZ) before harvesting by centrifugation. The cells were resuspended in lysis buffer (typically such as 20–50 mM Tris·HCl pH 8.0, 200–500 mM NaCl, 5% glycerol, 10 mM imidazole, 0.5 mM TCEP, and protease inhibtors) and lysed by high pressure homogenization. Polyethyleneimine was added to a concentration of 0.15%, and the insoluble debris removed by centrifugation.

The target protein was purified from the clarified cell extract by immobilized metal ion chromatography: The cell extract was passed over Ni2+ resin, and the resin was then washed with lysis buffer containing 25–40 mM imidazole. Protein was eluted with lysis buffer containing 250 mM imidazole and the eluted fractions were further purified on an S200 gel filtration column (buffers shown in Table 1). Where applicable, removal of the hexahistidine tag was accomplished using TEV protease at 4°C overnight followed by passing the solution over Ni2+ resin. Purified proteins were concentrated to the concentrations shown in Table 1 and stored at −80°C before crystallization.

Crystallization, data collection, and structure solution

Crystals were grown at 20°C using the sitting drop vapor diffusion method. Protein solution was mixed with crystallization buffer and equilibrated against a reservoir of crystallization buffer. Crystallization conditions and mixing ratios of protein to crystallization buffer are given in Table 2. Crystals were cryocooled by plunging into liquid nitrogen and X-ray data were collected at 100 K using a nitrogen stream. Some of the crystals were cryoprotected with ethylene glycol before cryocooling (Table 2). All X-ray data was collected on beamline X10A of the Swiss Light Source (Table 3).

Table Table 2.. Crystallization conditions
original image
Table Table 3.. Data collection and refinement statistics
original image

The data was processed with MOSFLM (Leslie 1999) or HKL2000 (Otwinowski and Minor 1997), and the CCP4 suite (Collaborative Computational Project Number 4 1994). Structures were solved by molecular replacement using PHASER (Storoni et al. 2004) and homologous models obtained from the PDB. Crystallographic models were rebuilt using O (Jones et al. 1991) or COOT (Emsley and Cowtan 2004), and refinement was performed using REFMAC5 (Murshudov et al. 1999).


We thank the staff of the Swiss Light Source for assistance with X-ray data collection. The Structural Genomics Consortium is a registered charity (number 1097737) funded by the Wellcome Trust, GlaxoSmithKline, Genome Canada, the Canadian Institutes of Health Research, the Ontario Innovation Trust, the Ontario Research and Development Challenge Fund, the Canadian Foundation for Innovation, VINNOVA, the Knut and Alice Wallenberg Foundation, the Swedish Foundation for Strategic Research and Karolinska Institutet.