Recognition of Unique Carboxyl-Terminal Motifs by Distinct PDZ Domains

See allHide authors and affiliations

Science  03 Jan 1997:
Vol. 275, Issue 5296, pp. 73-77
DOI: 10.1126/science.275.5296.73


The oriented peptide library technique was used to investigate the peptide-binding specificities of nine PDZ domains. Each PDZ domain selected peptides with hydrophobic residues at the carboxyl terminus. Individual PDZ domains selected unique optimal motifs defined primarily by the carboxyl terminal three to seven residues of the peptides. One family of PDZ domains, including those of the Discs Large protein, selected peptides with the consensus motif Glu-(Ser/Thr)-Xxx-(Val/Ile) (where Xxx represents any amino acid) at the carboxyl terminus. In contrast, another family of PDZ domains, including those of LIN-2, p55, and Tiam-1, selected peptides with hydrophobic or aromatic side chains at the carboxyl terminal three residues. On the basis of crystal structures of the PSD-95-3 PDZ domain, the specificities observed with the peptide library can be rationalized.

Many cytosolic signaling proteins and cytoskeletal proteins are composed of modular units of small protein-protein interaction domains that allow reversible and regulated assembly into larger protein complexes. Examples are SRC homology 2 (SH2) and SH3 domains and phosphotyrosine-binding (PTB) domains (1). PDZ domains have been observed in more than 40 cytosolic proteins, many of which are located at specific regions of cell-cell contact, such as tight junctions, septate junctions, and synaptic junctions. The name PDZ derives from three proteins that contain repeats of this domain: mammalian postsynaptic density protein, PSD-95; Drosophila disc large tumor suppressor, Dlg; and the mammalian tight junction protein, ZO1 (25).

Certain PDZ domain-containing proteins bind directly to the last several (COOH-terminal) residues of transmembrane proteins. For example, the second PDZ domain of PSD-95 binds the N-methyl-D-aspartate receptor through interaction with the COOH-terminal Ser/Thr-Xxx-Val sequence (6). PDZ domains of PSD-95 and Dlg bind similar COOH-terminal sequences on Shaker-type K+ channels, and PDZ domains may be necessary for the clustering of these channels on the cell surface (7).

These results have raised several interesting questions about PDZ domains: (i) Do these domains (like SH2 and SH3 domains) recognize internal sequences on proteins, or can they bind only to the free COOH-terminus of the target protein? (ii) Do individual members of the more than 80 PDZ domains defined to date recognize unique linear sequences? (iii) What is the structural basis for protein or peptide binding to PDZ domains?

The crystal structures of the third PDZ domains of hDlg (hDlg-3) alone (8) and PSD-95 (PSD-95-3) bound to a peptide (9) provide a starting point from which to answer these questions. The side chain of the COOH-terminal Val of the associated peptide is buried in a deep hydrophobic pocket. The free carboxylate of the valine interacts with amide nitrogens from a loop between two β structures in a sequence (Gly-Leu-Gly-Phe) that is highly conserved in most PDZ domains. However, the COOH-terminus is not deeply buried, raising the possibility that a free COOH-terminus may not be necessary for binding to all PDZ domains.

To address these questions, we used an oriented peptide library approach (10, 11). A soluble mixture of peptides of the same length and some common internal fixed residue or residues was passed over a column containing the domain of interest, and the subgroup of peptides retained by the column was sequenced to obtain a consensus motif (12). A library of peptides in which the COOH-terminal eight positions had degenerate amino acids (11) was used to investigate the binding specificities of nine PDZ domains (Table 1). In general, the PDZ domains bound preferentially to peptides that terminated in a hydrophobic amino acid (usually Val or Ile). In addition, most (but not all) of the PDZ domains selected for peptides with either Ser, Thr, or Tyr located two residues from the COOH-terminus (−2 position). To increase the fraction of peptides with high affinity, we constructed a library with the degeneracy at the −2 position restricted to Ser, Thr, or Tyr. This method increased our ability to determine selectivities at other positions (Table 1). Results with this library were consistent with those obtained with the fully degenerate library. Additional selection specificity was observed out to the −8 position for some domains (Table 1).

Table 1.

PDZ domain specificity deduced through the use of oriented peptide libraries (10) and known binding sites of PDZ domains. Either a peptide library with the sequence KNXXXXXXXX-COOH or a library with the sequence KNXXXXXX(S/T/Y)XXCOOH was screened with the indicated GST-PDZ domain. X indicates all amino acids except Cys or Trp. Selectivities greater than 1.5 are shown. Values larger than 2.0 are indicated in bold. On the basis of peptide library studies on other protein modules, a value greater than 1.5 indicates significant selection, and a value greater than 2.0 indicates strong selection. Underlined residues indicate the position where the degeneracy is limited to Ser, Thr, and Tyr. ND, not determined. Abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

View this table:

To determine whether an internal sequence could be recognized by PDZ domains, we used a library with a fixed internal Tyr that was nine residues from the COOH-terminus (MAXXXXYXXXXAKKK-NH2, where X indicates positions with degeneracy of all amino acids except Cys, Trp, Tyr, Ser, and Thr; total degeneracy ∼2.5 billion). The fixed COOH-terminus of this library (AKKK-NH2) is predicted not to bind to PDZ domains on the basis of the results with the other libraries (Table 1). Consistent with the idea that PDZ domains do not bind to internal sequences, none of the PDZ domains investigated retained specific peptides from this third library (13).

The PDZ domains in Table 1 can be divided into two major groups on the basis of the amino acid selected at the −2 position. One group, including the third and fifth PDZ domains of the phosphotyrosine phosphatase PTPbas/FAP-1 (PTPbas-3 and PTPbas-5) (14) and the three PDZ domains of murine hDlg (mDlg) (12), selected peptides with amino acids containing hydroxyl groups (Ser, Thr, or Tyr) at position −2 (Table 1). However, another group, including the PDZ domains of p55 (15), human LIN-2 (16), Tiam-1 (17), and AF-6 (18), selected peptides with hydrophobic amino acids at −2 (Table 1). Most members of this latter group preferred Phe at −2, although Tyr was also selected at this position. This result was surprising because previous studies led to speculation that a hydroxyl group at position −2 is important for binding to PDZ domains (7, 9).

The peptides from the library that appeared to be optimal for binding to the mDlg, PTPbas-3, and Tiam-1 PDZ domains were synthesized and investigated for their abilities to bind to each of these domains (Fig. 1). As measured by BIAcore, the estimated dissociation constants (Kd's) for mDlg-2 and PTPbas-3 PDZ domains to their optimal peptides were 42 nM (Kon = 2.4 × 104 M−1 s−1 and Koff = 1.0 × 10−3 s−1) and 154 nM (Kon = 1.3 × 104 M−1 s−1 and Koff = 2.1 × 10−3 s−1), respectively. These affinities are in the same range as those observed for binding of optimal peptides of these sizes to SH2 and SH3 domains (19, 20). Moreover, the mDlg-2 PDZ domain (at 0.5 μM) failed to bind to the PTPbas peptide, whereas the PTPbas-3 PDZ bound only weakly to the Dlg peptide (Kd ∼ 1.5 μM) (Fig. 1, A and B). None of the three PDZ domains, when added at 0.5 μM, bound significantly to the optimal Tiam-1 peptide on the basis of the BIAcore technique (21). One reason for this result could be either that the cross-linking procedure interfered with binding or that the affinity of this peptide is out of the sensitivity range of BIAcore analysis. Nevertheless, the specificity of the Tiam-1 peptide could be demonstrated by competitive binding experiments in which a PDZ domain precipitation approach was used (Fig. 1C). Whereas 50% maximal displacement of Tiam-1 PDZ domain binding occurred at 10 μM of the Tiam-1 optimal peptide, no significant competition for binding was observed at 50 μM of the peptides designed for the other two PDZ domains. These results indicate that the peptides predicted by the library approach have relatively high affinity for their respective PDZ domains and that distinct PDZ domains bind to distinct optimal sequences.

Fig. 1.

Specific binding of predicted optimal peptides to PDZ domains of Dlg, PTPbas, and Tiam-1. Optimal peptide for mDlg-1/2 (Dlg-Pep: KKKKETDV-COOH) or PTPbas-3 (PTP-Pep: KDDQESNV-COOH) was coupled to BIAcore sensor CM5 chips (Biosenser). The binding of PDZ domains to these peptides was monitored with the use of BIAcore 2000. Time 0 to 120 s indicates the association phase where GST-PDZ domain fusion proteins were injected. The arrows indicate injection of 20 μM competitor peptides. To determine Kon and Koff rates, we injected various concentrations of GST-PDZ fusion proteins and analyzed data with the BIAevaluation 2.1 software (Biosensor). (A) Relative response (binding) when GST-mDlg-2 PDZ domain (0.5 μM) was passed through surfaces coated with Dlg-Pep or PTP-Pep. (B) Relative response when GST-PTPbas-3 PDZ domain (0.5 μM) was passed through surfaces coated with Dlg-Pep or PTP-Pep. (C) Tiam-1 optimal peptide (Tia-Pep: SSRKEYYA-COOH) was coupled to cyanogen bromide-activated Sepharose beads (Sigma). The beads were then incubated with GST or GST PDZ domain fusion proteins (4 μg/ml) with various concentrations of peptides (0 to 100 μM) in TSN buffer containing BSA (1 mg/ml) and DTT (1 mM). Bound proteins were washed three times with TSN buffer, separated by SDS-polyacrylamide gel electrophoresis, and visualized by protein immunoblotting with antibody to GST (Transduction Lab). Relative binding was quantified with the use of an AGFA Scanner.

The optimal peptides predicted by the peptide library approach can be rationalized on the basis of the recent crystal structure of the PSD-95-3 PDZ domain bound to a high-affinity peptide (9). The PDZ domain of PSD-95 is composed of two α helices (designated αA and αB), six β strands (designated βA to βF), and several loop regions, with the loop between βA and βB interacting with the carboxylate group of the peptide's COOH-terminal residue. In Fig. 2, we have highlighted the residues of the PSD-95-3 PDZ domain that are predicted to interact with the side chains of the associated peptide. Residues forming each pocket on PSD-95-3 are also listed in Table 2. Corresponding residues on the other PDZ domains were deduced by alignment of their primary sequences with PSD-95-3 and labeled according to their positions in the crystal structures of PSD-95-3 (for example, the first residue in helix αB is named αB1). To emphasize the nature of the binding pockets in Fig. 2 and Table 2, we show basic, acidic, hydrophobic, and hydrophilic (uncharged) residues in different colors. In particular, we have focused on His-αB1 (His372 of PSD95-3), which coordinates the hydroxyl group of the Thr at postion −2; Glu-BC2, which is near the −4 (Lys) side chain; and Ser-βC4, which coordinates the Gln at position −3. Phe-βC5, also labeled, does not directly contact residues on the associated peptide but is at a position that could affect binding of residues with large side chains at the −1 or −3 positions, or both. These residues vary among the PDZ domains that we have studied (Table 2).

Fig. 2.

COOH-terminal peptide recognition by PSD95-3 PDZ domain. The peptide binding site of PSD95-3 PDZ domain was presented with the PDZ domain backbone shown in ribbon and peptide KQTSV-COOH shown in sphere structures (9). Residues that form various binding pockets are indicated in color tubes. Color designation: basic residues (blue), acidic residues (red), hydrophobic residues (black), and hydrophilic (uncharged) residues (green).

Table 2.

Position of residues of the PDZ domain peptide-binding pockets that are predicted to interact with the side chains of the associated peptide. Color designation: basic residues (blue), acidic residues (red), hydrophobic residues (black), and hydrophilic (uncharged) residues (green). Ec-Htra (26) and Ss-Ctpa (27) are bacterial and plant proteases, respectively.

View this table:

The first residue in the αB helix (αB1) correlates strongly with peptide selectivity at position −2. We have therefore grouped PDZ domains into subgroups (Table 2). The domains in group 1A with His at the αB1 position selected peptides with Ser, Thr, or Tyr at position −2, in agreement with His-αB1 coordinating the hydroxyl group (9). Consistent with this prediction, LIN-7 binds directly to LET-23, which has the sequence SQKETCL at the COOH-terminal (22). Those PDZ domains that lack a basic residue at αB1 (group II), however, selected hydrophobic or aromatic amino acids at position −2. The Tiam-1 PDZ domain has an Asp at αB1, and although it still selects Tyr, it is less selective, with Phe being the second best. It is possible that the Asp carboxylate can also weakly coordinate the hydroxyl group of Tyr. In contrast, the PDZ domains of p55 (Val-αB1), LIN-2 (Val-αB1), and AF-6 (Gln-αB1) selected for Phe over Tyr at position −2. Thus, the residue at αB1 is a good indicator of the selectivity at the −2 position and therefore determines the selectivity subgroups in Table 1.

Specificity at the other positions can also be rationalized. The residues that make up the P0 pocket are hydrophobic and relatively highly conserved (9). All the domains studied selected hydrophobic residues at the 0 position, and in most instances, Val was optimal. Small differences in the residues of the pocket are likely to explain the selectivities for Phe or Ala of Tiam-1 and p55 at this position.

Considerable variability in selectivity at the −1 position was observed, although in no instance was the selectivity at this position very strong (compared to the 0 and −2 positions). This selectivity might be explained by residue βC5. Although this residue does not directly contact the −1 Ser in the PSD-95-3 structure (Fig. 2), the ε amino group of a Lys at βC5 could be close enough to the −1 position to explain the selection for Asp at this location by mDlg-1/2. In contrast, the PTPbas-5 PDZ domain has an Asp at βC5 and selects weakly for Lys at position −1. Finally, basic residues at βC5 and βC4 might also help explain the selectivity of several PDZ domains for peptides with Glu at the −3 position.

The optimal motifs predicted by the peptide library are in good agreement with known binding sites of PDZ domains (Table 1). For mDlg-2 and PTPbas-3 PDZ domains, their known binding sites on channel proteins and Fas antigen are consistent with the optimal binding motifs. In addition, p55 and LIN-2 were demonstrated to bind glycophorin C and neurexin, respectively (Table 1) (23, 24). In agreement with our peptide library prediction, the last three residues of these proteins are hydrophobic and contain aromatic residues at both the −2 and −1 positions (Table 1).

One possible function of PDZ domains is their role in cytoskeletal and membrane organization. Many PDZ domain-containing molecules (for example, LIN-7, Dlg, p55, and PTP-meg) not only associate with the membrane, but also carry sequence motifs for cytoskeleton localization (5, 22). The clustering function of PDZ domains can be achieved through network binding of PDZ domains, because many proteins carry multiple copies of PDZ domains. In addition to their ability to bind COOH-terminal sequences, some PDZ domains can also dimerize with each other (25). The three PDZ domains of Dlg were shown to recognize similar motifs, whereas PDZ domains of PTPbas recognize different motifs (Table 1). Thus, PDZ domains may cooperate to enhance the binding to their common targets, or in the case of PTPbas, they may help to bind simultaneously to multiple, different targets.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
View Abstract

Navigate This Article