Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2013 Apr 27.
Published in final edited form as: Mol Cell. 2012 Mar 22;46(2):226–237. doi: 10.1016/j.molcel.2012.02.012

Genome-Wide Identification and Functional Annotation of Dual Specificity Protein- and Lipid-Binding Modules That Modulate Protein Interactions at the Membrane

Yong Chen 1,#, Ren Sheng 1,#, Morten Kallberg 2,#, Antonina Silkov 3,#, Moe P Tun 1, Nitin Bhardwaj 2, Svetlana Kurilova 1, Randy A Hall 4, Barry Honig 3, Hui Lu 2,*, Wonhwa Cho 1,5,*
PMCID: PMC3431187  NIHMSID: NIHMS366236  PMID: 22445486

Abstract

Emerging evidence indicates that membrane lipids regulate protein networking by directly interacting with protein-interaction domains (PIDs). As a pilot study to identify and functionally annodate lipid-binding PIDs on a genomic scale, we performed experimental and computational studies of PDZ domains. Characterization of 70 PDZ domains showed that 40% had submicromolar membrane affinity. Using a computational model built from these data, we predicted the membrane binding properties of 2000 PDZ domains from 20 species. The accuracy of the prediction was experimentally validated for 26 PDZ domains. We also subdivided lipid-binding PDZ domains into three classes based on the interplay between membrane and protein binding sites. For different classes of PDZ domains, lipid binding regulates their protein interactions by different mechanisms. Functional studies of a PDZ domain protein, rhophilin2 suggest that all classes of lipid binding PDZ domains serve as genuine dual-specificity modules regulating protein interactions at the membrane under physiological conditions.

INTRODUCTION

Regulation of cellular processes, such as cell signaling, involves myriad protein-protein interactions that are typically mediated by modular protein interaction domains (PIDs), such as SH2, SH3, PDZ, and WW domains (Bhattacharyya et al., 2006; Pawson, 2004; Pawson and Nash, 2003). Cellular membranes, including the plasma membrane (PM), offer unique local environments that facilitate protein-protein interactions and therefore serve as the main sites for protein complexes and networks (Bray, 1998; Cho, 2006). Accumulating evidence suggests that membrane lipids play a key role in protein complex formation or networking through direct interactions with signaling proteins, scaffold proteins in particular (Cho, 2006; Winters et al., 2005). Membrane recruitment of cellular proteins is mediated by lipid-binding domains or motifs that either recognize specific lipids or non-specifically interact with the anionic membrane surface (Cho and Stahelin, 2005; DiNitto et al., 2003; Lemmon, 2008). Thus, it has been generally thought that the protein networking and interactions at the membrane involve the coordinated action of separate lipid binding domains (or motifs) and PIDs in the same molecules (Di Paolo and De Camilli, 2006; Lemmon, 2008). Recent studies have shown, however, that PIDs, such as PDZ domain (Feng and Zhang, 2009; Zimmermann, 2006) and PTB domain (Ravichandran et al., 1997; Zhou et al., 1995), can directly interact membrane lipid(s) and thus mediate both protein-protein and protein-lipid interactions. It has also been reported that some lipid binding domains, such as the PH domain (Yao et al., 1994) and the PX domain (Lee et al., 2006), can interact with proteins as well as lipids. These findings suggest that PIDs and lipid binding domains may serve as dual specificity lipid- and protein-binding modules that play a crucial role in protein interactions and networking. To test this hypothesis, we have developed new experimental and bioinformatics tools for the identification and characterization of dual-specificity PIDs on a genomic scale and applied these tools to the study of PDZ domains.

The PDZ domain is a small (≈ 90 amino acids) modular PID that interacts with a short C-terminal sequence of its target protein(s) (Feng and Zhang, 2009; Sheng and Sala, 2001). The domain was originally identified in three unrelated proteins, Postsynaptic density 95 (PSD95), Disc large 1 (DLG1), and Zonular occludens 1 (ZO1), but has since been found in a large number of proteins, including postsynaptic proteins and cell junction proteins. A SMART search (Schultz et al., 1998) in genomic mode identifies 148 human proteins containing >500 different PDZ domains, making them one of the most ubiquitous PIDs in vertebrates. Most PDZ domain-containing proteins contain multiple copies of PDZ domains, serving as prototype scaffold proteins that reversibly interact with multiple binding partners and thereby dynamically coordinating signaling complex formation and protein networking (Feng and Zhang, 2009; Sheng and Sala, 2001). Recent studies have shown that PDZ domains can directly interact with anionic model membranes and that in some cases this PDZ-membrane interaction is important for the cellular function of their host proteins (Meerschaert et al., 2009; Pan et al., 2007; Wu et al., 2007; Zimmermann et al., 2002). However, it is still not known if lipid binding is a general property of PDZ domains, if they can serve as authentic dual-specificity modules under physiological conditions, and how their protein and lipid binding are inter-related. Thus, the PDZ domain is an excellent candidate for the pilot study for the genome-wide identification and characterization of dual specificity PIDs.

RESULTS

SPR Analysis of 70 Mammalian PDZ Domains

A recent study measured the binding of 74 PDZ domains to anionic vesicles by vesicle pelleting assay (Wu et al., 2007). Although the study revealed a high tendency of PDZ domains to bind lipids, the qualitative nature of the data limits their application to systematic analysis (or prediction) of membrane binding properties of PDZ domains and of the interplay between their membrane and protein interactions. It was therefore necessary to collect a robust quantitative database large enough for statistical and systematic analysis. All reported membrane-binding PDZ domains bind anionic membranes with low to no lipid headgroup specificity (Meerschaert et al., 2009; Pan et al., 2007; Wu et al., 2007; Zimmermann et al., 2002). Also, a majority of PDZ domain proteins interact with protein partners that are associated with the PM (Feng and Zhang, 2009; Sheng and Sala, 2001) whose inner (i.e., cytoplasmic) layer is highly anionic due to the presence of phosphatidylserine (PS) and phosphatidylinositiol-4,5-bisphosphate (PtdIns(4,5)P2) (Cho and Stahelin, 2005; McLaughlin and Murray, 2005). Thus, PDZ domains are most likely to interact with the inner layer of PM. For this reason we used the vesicles whose lipid composition recapitulates that of inner PM (i.e., PM-mimetic vesicles) (Cho and Stahelin, 2005) as a model membrane and rigorously determined the Kd values for 70 monomeric PDZ domains from 35 different mammalian proteins by the surface plasmon resonance (SPR) analysis (Cho et al., 2001). We mainly selected uncharacterized PDZ domains (i.e., 51) for this study but also reevaluated some of previously characterized PDZ domains (i.e., 19) (Wu et al., 2007) to directly compare the results from the two different analyses.

As shown in Table 1, 27 out of 70 tested PDZ domains (≈40%) have submicromolar Kd’s for the PM vesicles with the highest affinity in the range of 10−8 M, which is comparable to that of canonical lipid binding domains (Cho and Stahelin, 2005). Syntenin1 PDZ domains (Zimmermann et al., 2002) and the second PDZ domain of ZO1 (Meerschaert et al., 2009), which were reported to have physiologically significant membrane affinity, have 1–3 μM Kd’s under our experimental conditions (see Table 1). Our results thus indicate that membrane binding is a more general property of PDZ domains than previously thought and might play a role in their cellular localization (see Fig. S1) and/or function. Table 1 also shows a significant discrepancy between our data and the previous results (Wu et al., 2007). In particular, five (out of 17) PDZ domains that were previously classified as non-membrane binder turned out to bind PM-mimetic vesicles with Kd = 140 to 930 nM. For those PDZ domains with submicromolar affinity for the PM vesicles, we also measured the selectivity for phosphoinositides (PtdInsP) and most PDZ domains did not show appreciable PtdInsP selectivity. To validate our SPR data we also determined Kd values for selected PDZ domains by a fluorescence resonance energy transfer (FRET) assay (see Table S1 and Fig. S2). Results show that Kd values determined by the two different methods are comparable.

Table 1.

The membrane binding affinity of the 70 experimentally tested PDZ domains used for training the classification model

Gene Species Residue number Kd (nM) for PMa PtdInsP selectivityc Results from the previous studyd
NHERF1/EBP50-PDZ1 Human 10 – 98 24 ± 1 low
Dvl2-PDZ Human 261 – 353 33 ± 3 PtdIns(4,5)P2
Dvl1-PDZ Human 245 – 337 45 ± 6 ND
Dvl3-PDZ Human 243 – 335 50 ± 5 PtdIns(4,5)P2
Tamalin-PDZ Mouse 100 – 186 90 ± 8 low
SAP102-PDZ3 Rat 404 – 482 140 ± 5 PtdIns(4,5)P2
PtdIns(3,4,5)P3
no binding
LNX1-PDZ4 Mouse 638 – 721 180 ± 40 low
PDZK2-PDZ3 Mouse 263 – 343 280 ± 50 low
MAGI1-PDZ5 Human 998 – 1091 290 ± 10 low
PDZ-GEF-PDZ Human 385 – 470 290 ± 32 low
β2-syntrophin-PDZ1 Human 115 – 195 320 ± 80 low
PDZK2-PDZ2 Mouse 151 – 255 320 ± 32 low
nNOS-PDZ Human 17 – 96 340 ± 10 low
PSD95-PDZ3 Rat 313 – 391 390 ± 30 low no binding
INADL-PDZ6 Human 1068 – 1160 480 ± 190 low
Chapsyn110-PDZ3 Rat 421 – 499 510 ± 50 low
γ2-syntrophin-PDZ Mouse 73 – 153 530 ± 140 low
Harmonin-PDZ1 Mouse 87 – 165 600 ± 70 low
MAGI3-PDZ5 Human 1021 – 1100 610 ± 190 low no binding
SAP97-PDZ3 Rat 465 – 543 620 ± 70 PtdIns(3,4)P2 no binding
LNX2-PDZ1 Mouse 232 – 314 670 ± 100 low
MAGI-2-PDZ3 Human 605 – 683 750 ± 170 low
α-syntrophin-PDZ1 Mouse 81 – 161 860 ± 70 low binding
MAGI-2 –PDZ5 Human 920 – 1007 900 ± 170 low
PSD95-PDZ2 Rat 160 – 244 930 ± 120 low no binding
PDZ-PhoGEF-PDZ1 Human 47 – 120 950 ± 110 low
LNX1-PDZ1 Mouse 278 – 360 960 ± 120 low
ZO-1 PDZ-2 Mouse 186–261 980 ± 200 PtdIns(3,4)P2
PtdIns(3,4,5)P3
PtdIns(4,5)P2
binding
INADL-PDZ5 Human 686 – 772 1070 ± 110
β1-syntrophin-PDZ1 Human 538 – 613 1440 ± 180
Rhophilin-1-PDZ1 Mouse 111 – 191 1440 ± 160
Syntenin1-PDZ1 Human 100 – 195 2200 ± 250 binding
SAP102-PDZ1 Rat 149 – 233 4980 ± 870
PSD95-PDZ1 Rat 65 – 149 NMb no binding
MAGI-2-PDZ2 Human 426 – 492 NM
MAGI-2-PDZ4 Human 778 – 859 NM
SAP97-PDZ1 Rat 224 – 308 NM no binding
Spinophilin-PDZ1 Rat 496 – 581 NM
Neurabin-PDZ1 Rat 505 – 590 NM no binding
NHERF-1-PDZ2 Human 150 – 235 NM
NHERF-2-PDZ2 Human 150 – 230 NM no binding
NHERF-2-PDZ1 Human 11 – 88 NM
SAP97-PDZ2 Rat 318 – 402 NM no binding
CAL-PDZ1 Human 288 – 368 NM
PDZK1-PDZ3 Mouse 243 – 320 NM
PDZK1-PDZ1 Mouse 9 – 87 NM
MAGI-1-PDZ1 Human 295 – 401 NM
MAGI-1-PDZ3 Human 643 – 720 NM
MALS-1-PDZ1 Human 108 – 187 NM
E6TP1-PDZ1 Human 953 – 1025 NM
LNX1-PDZ3 Mouse 508 – 591 NM
LNX1-PDZ2 Mouse 385 – 465 NM
Densin-180-PDZ1 Rat 1403 – 1493 NM
MAGI-2-PDZ1 Human 17 – 98 NM
MALS-3-PDZ1 Mouse 93 – 172 NM
LNX2-PDZ2 Mouse 338 – 418 NM
MAGI-1-PDZ2 Human 472 – 554 NM
PDZK2-PDZ4 Mouse 394 – 472 NM
Harmonin-PDZ2 Mouse 211 – 289 NM
MAGI-3-PDZ1 Human 410 – 476 NM
MAGI-3-PDZ3 Human 726 – 807 NM no binding
MAGI-3-PDZ4 Human 851 – 935 NM no binding
PDZK1-PDZ2 Mouse 128 – 215 NM
MUPP1-PDZ6 Mouse 996 –1077 NM
MUPP1-PDZ7 Mouse 1139 – 1231 NM no binding
MUPP1-PDZ8 Mouse 1338 – 1421 NM no binding
MUPP1-PDZ12 Mouse 1847 – 1933 NM no binding
MUPP1-PDZ13 Mouse 1972 – 2055 NM no binding
MAGI-3-PDZ2 Human 578 – 641 NM
a

Mean ± S.D. values determined by SPR analysis using PM-mimetic vesicles.

b

Not measurable by SPR analysis with up to 10 μM protein (i.e, Kd>10 μM).

c

Determined by measuring the relative affinity for POPC/PtdInsP (97:3) by SPR analysis.

d

Taken from (Wu et al., 2007)

Classification Model for Predicting Membrane Binding Affinity of PDZ Domains

Significant percentage of membrane-binding PDZ domains in our dataset allowed us to build a high-accuracy prediction model for other PDZ domains. The PDZ domains in general have a high degree of sequence similarity (Feng and Zhang, 2009; Sheng and Sala, 2001). However, our data show that sequence similarity between any two PDZ domains does not translate into similar membrane binding properties, making it a poor indicator for classification and prediction purposes. This is in contrast to other lipid binding domains, such as the FYVE domain, for which a good correlation between sequence similarity and relative membrane affinity was observed (Blatner et al., 2004). Therefore, it was necessary to build a more sophisticated model based on quantification of physical and chemical characteristics of the domains.

We recently developed a machine learning-based prediction method for membrane binding domains that uses a numerical vector representation obtained from primary and tertiary structures of proteins as input features and various machine learning algorithms as classifiers (Bhardwaj et al., 2006). To apply this method to our current task of discriminating membrane binding properties among highly homologous PDZ domains, we incorporated residue-specific features derived from the domain sequence data in addition to the previously used protein-level features from structural data. Protein-level features enabling a domain to interact with membranes include nonspecific electrostatic attraction between anionic membranes and basic protein residues (Mulgrew-Nesbitt et al., 2006), association of hydrophobic protein residues with the membrane hydrocarbon core (Stahelin et al., 2003), and hydrogen bonds between key protein residues and lipid headgroups (Cho and Stahelin, 2005). To incorporate residue-specific features, we determined the score of each residue and the cumulative score for a segment around it (Park et al., 2008) by calculating the recursive functional classification (RFC) matrix. This statistical scoring approach helps identify residues that are more likely to be observed at certain positions in membrane binding PDZ domains than in non-binding PDZ domains. Fig. 1 depicts an example of this scoring procedure, displaying the score of each residue and the cumulative score for the neighboring segment of the rat PSD95-PDZ3 domain. It is evident that certain residues (i.e., R1, H5, T9, E19, D37, L38, S39, E40, and Q72) have strong influence on membrane binding of the domain. Interestingly, these residues are not exclusively located in the electrostatically positive region, which is generally involved in binding to anionic membranes, but in both electro-positive and -negative regions. This pattern is also seen with several other membrane-binding PDZ domains. Among those residues found in the electronegative region, some residues (e.g., E and D) may form specific hydrogen bonds with lipids, as seen with PH domains (DiNitto and Lambright, 2006) while others may play indirect roles, such as guiding the membrane-binding orientation of the domain. It should be noted that the identity and the relative contribution of membrane binding esidues vary significantly among similar PDZ domains. This is demonstrated in residue and cumulative scoring plots for three different PDZ domains, SAP102-PDZ3, rhophilin 2-PDZ, and tamalin-PDZ (see Fig. S3). A few high scoring residues make predominant contribution for SAP102-PDZ3 whereas many residues contribute relatively evenly to membrane binding for tamalin and rhopholin2 PDZ domains.

Fig 1.

Fig 1

Quantification of residue-specific features. (A) Residue and cumulative score obtained from the RFC matrix for the third PDZ domain of PSD95. (B) The structure of the domain along with electrostatic iso-surfaces. Key membrane binding residues identified by the scoring system are highlighted and labeled.

We also optimized the classification method for the prediction of membrane-binding PDZ domains. To develop a binary classification method, one needs to define positive and negative cases. Since PDZ domains have a wide range of continuous Kd values, it was necessary to choose a specific Kd value as a threshold for physiologically significant membrane binding. In general, it is not straightforward to predict the cellular membrane binding of a particular protein from its Kd value for a model membrane. It is because membrane binding of a protein is different from chemical binding of two species with well-defined binding sites (White et al., 1998) and because it is technically challenging to accurately determine the cellular lipid concentrations (Yoon et al., 2011). We have thus taken a combinatorial approach of determining the relative membrane affinity (i.e., in terms of relative Kd) of a family of proteins by the SPR analysis and then measuring their cellular membrane binding properties to estimate the threshold Kd value for their cellular membrane binding (Blatner et al., 2004; Cho and Stahelin, 2005). Since syntenin1-PDZ and ZO1-PDZ2 whose membrane affinity is physiologically significant (Meerschaert et al., 2009; Zimmermann et al., 2002) have 1–3 μM Kd’s, we set the threshold Kd of PDZ domains to 1 μM. This threshold value divided our SPR-tested PDZ domains into 27 binding cases and 41 non-binding cases. Lowering the cutoff Kd value to 0.5 μM would reduce the positive cases to 15. For evaluation of prediction, we tested two machine learning algorithms that have proven successful in diverse classification applications (Bhardwaj et al., 2005; Bhardwaj et al., 2006), i.e., the kernel-based support vector machine (SVM) methodology and the decision tree algorithm C4.5 combined with the boosting algorithm AdaBoost (referred to as ABC4.5). Table S2 summarizes the results form these algorithms with 10-fold cross validations and with different feature sets. The prediction was more accurate when structural and sequence features are used in combination than independently. Between the two algorithms, the SVM did well on both the 0.5 and 1 μM Kd-cutoff datasets (see also Fig. S4). Also, SVM algorithm achieved better accuracy (94%) with balanced sensitivity and selectivity with 1 μM Kd-cutoff. We thus decided to use SVM with all features and Kd = 1 μM as a threshold for the genome-wide prediction of membrane binding activity of PDZ domains.

Predictions for 2000 PDZ Domains from 20 Different Species

Using our optimized protocol, we predicted the membrane binding properties of all 2000 PDZ domains found in 20 different species. Since we used both structural and sequence features, domains included in our prediction are from all sequences for which reliable homology models could be generated (Supplemental Methods). As seen in Fig. 2, ≈30% of PDZ domains are predicted to have submicromolar membrane binding affinity although some degree of variation is found among species. It thus seems evident again that membrane binding is a common property among PDZ domains. The complete collection of the PDZ domains annotated in this study can be found in our online resource MeTaDoR (Membrane Targeting Domains Resource) (http://metador.bioengr.uic.edu/) (Bhardwaj et al., 2007). Several options for searching the collection is given, among them are host protein name, organism, and binding annotation. There is also an option to classify the domains with variable threshold Kd values. For each domain the host protein and the domain location in the host protein is given along with relevant links to public databases.

Fig 2.

Fig 2

Membrane binding statistics for 2000 PDZ domains found in 20 species. Predicted percentage of membrane-binding PDZ domains is shown for each species. SVM classifier was used for prediction with all features included and Kd= 1 μM as a threshold.

Experimental Validation of Prediction

To further validate our prediction model, we selected 25 PDZ domains from the list of 2000 predictions and measured their membrane binding by SPR analysis. As with the initial screening of PDZ domains, we mainly selected uncharacterized PDZ domains for validation with the addition of a few PDZ domains previously characterized (Wu et al., 2007). Table 2 compares the experimental results with our prediction values obtained using 1 μM Kd cutoff. All the binding cases were classified correctly while three non-binding cases were classified as binding ones. Thus the overall accuracy on the test set is ≈90%, which is similar to the cross-validation accuracy. Even the three misclassified cases for non-binding PDZ domains (i.e., ZO-2-PDZ2, PAR3-PDZ1, and ZO-3-PDZ2) were borderline, low-confidence cases with prediction values ≤0.1, with 0 being the cut-off score separating binding and non-binding domains. In particular, ZO-2-PDZ2 was predicted to be a membrane binder while the experimental Kd value (i.e., 1.2 ± 0.4 μM) is only slightly above the 1-μM Kd threshold. Collectively, this evaluation demonstrates the accuracy and reliability of our prediction. The selection of 70 domains used for initial database and 25 domains used for evaluation did not bias the outcome of our prediction: i.e., when PDZ domains in the two groups were interchanged, essentially the same results were obtained in terms of classification and prediction accuracy.

Table 2. Experimental evaluation of our prediction for membrane binding of PDZ domains.

The prediction values were calculated using the 1 μM Kd-SVM model. Positive values indicate membrane binding whereas negative values indicate non-binding. The further away from zero, the more confident the prediction is. Prediction values for three mis-classified cases were shown in bold italic

Domain Species Residue number Kd (nM) for PMa Prediction
C2PA-PDZ1 Mouse 185–271 NMb −0.20
Chapsyn110-PDZ1 Rat 98–182 NM −1.70
Chapsyn110-PDZ2 Rat 193–277 510 ± 50 0.17
GRIP-PDZ3 Rat 252–333 NM −0.30
GRIP-PDZ4 Rat 471–557 NM −0.30
GRIP-PDZ5 Rat 572–654 NM 0.00
GRIP-PDZ6 Rat 672–751 NM −0.70
InaD-PDZ1 Fruit Fly 17–103 NM −0.30
MUPP1-PDZ10 Mouse 1614–1697 NM −0.2
PAPIN-PDZ1 Rat 85–177 NM −0.30
PAR3-PDZ1 Rat 271–359 NM 0.05
PAR3-PDZ3 Rat 590–681 NM −0.06
PTPN3-PDZ1 Human 510–595 450 ± 90 0.60
PTPN13-PDZ1 Mouse 1084–1167 NM −0.10
Rhophilin-2-PDZ Human 515 – 593 500 ± 30 0.37
SAP102-PDZ2 Rat 244–328 NM −0.50
Shank1-PDZ1 Rat 663–754 NM −0.40
ZO-1 PDZ-1 Mouse 23–107 NM −0.70
ZO-1 PDZ-3 Mouse 421–502 NM −0.70
ZO-2 PDZ-1 Mouse 10–94 NM −0.90
ZO-2 PDZ-2 Mouse 287–365 1200 ± 400 0.10
ZO-2 PDZ-3 Mouse 489–570 NM −0.40
ZO-3 PDZ-1 Mouse 11–90 NM −0.20
ZO-3 PDZ-2 Mouse 187–261 NM 0.10
ZO-3 PDZ-3 Mouse 370–448 NM −0.20
a

Mean ± S.D. values determined by SPR analysis using PM-mimetic vesicles (i.e, Kd>10 μM).

b

Not measurable by SPR analysis (i.e, Kd>10 μM).

Functional Classification of Membrane Binding PDZ Domains

To systematically analyze the location of membrane binding sites and the interplay between membrane and protein-binding sites, we calculated the surface electrostatic potential for all mammalian PDZ domains using either known structures or best homology model structures (see Supplemental Methods). This analysis revealed that most of these PDZ domains have at least one prominent surface cationic patch that may serve as an anionic lipid binding site. Depending on the location of the cationic patch in relation to the canonical peptide binding site they could be subdivided into two classes. Class A PDZ domains have a main cationic patch (or two largest patches in the case that two or more patches are found) with no topological overlap with the peptide binding site (see Fig. S5) whereas Class B PDZ domains contain the cationic patch proximal to the peptide pocket (Fig. S6). We reasoned that this structural classification has functional implications because the prominent cationic patch in each PDZ domain is likely to represent its lipid binding site. To test this hypothesis, we experimentally determined the location of lipid binding sites and the interplay between their lipid and protein binding for selected members of each class.

Among Class A PDZ domains, we selected the SAP102-PDZ3 that has a prominent cationic patch (R449, R459, and R484) in the opposite side of the peptide binding pocket (Fig. 3A). Interestingly, this cationic patch also forms a groove, suggesting that it may specifically bind a lipid headgroup. Our SPR analysis confirmed that this PDZ domain has definite selectivity for PtdIns(4,5)P2 and PtdIns(3,4,5)P3 over other PtdInsPs (Fig. S7A), shows PtdIns(4,5)P2 dependency in membrane binding (Fig. S7B), and binds soluble inositol 1,4,5-trisphosphate (Ins(1,4,5)P3) (Fig. S8). Also, mutations of cationic residues in the groove (e.g., R449E) greatly reduced affinity for PtdIns(4,5)P2-containing vesicles (Fig. S7C). Furthermore, none of these mutations decreased binding to the C-terminal peptide of stargazin, an interaction partner of SAP-102 (Fig. S7D). Thus, this PDZ domain has relatively a well-formed binding site for PtdIns(4,5)P2 and PtdIns(3,4,5)P3 that is distant from the peptide binding pocket and it can simultaneously bind a PtdIns(4,5)P2 (or PtdIns(3,4,5)P3) and a protein molecule (see Fig 3A). The list of Class A PDZ domains with a well-formed cationic groove and thus with potential lipid headgroup specificity is shown in Table S3. Based on our results, we postulate that Class A PDZ domains have topologically distinct and functionally orthogonal lipid and protein binding sites. This notion is supported by functionally independent lipid and peptide binding sites observed for two additional members of Class A family, PICK1-PDZ (Pan et al., 2007) and NHERF1-PDZ1 (Sheng et. al., submitted).

Fig 3.

Fig 3

Functional classification of membrane binding PDZ domains. (A) The energy-minimized model structure of the SAP102-PDZ3-peptide(RTTPV)-Ins(1,4,5)P3 ternary complex. Electrostatic surface potential (left) and the ribbon diagram (right) show the separation of the peptide- and the lipid-binding sites. In the ribbon diagram, R449, R459, and R484 that form the cationic groove are shown in space-filling representation and labeled. (B) The energy-minimized model structure of the rhophilin2-PDZ-peptide(EYLGLDVPV)-Ins(1,4,5)P3 ternary complex. Electrostatic surface potential (left) and the ribbon diagram (right) show the proximity of the peptide- and the lipid-binding sites. In the ribbon diagram, K576 and K579 in the α2A helix that are involved in lipid binding are shown in space-filling representation and labeled. (C) The energy-minimized model structure of tamalin-PDZ-peptide(IRDYTQSSSSL) binary complex. Because of severe steric clash, an Ins(1,4,5)P3 molecule could not be docked on the PDZ-peptide complex. In the ribbon diagram, R166, H167, and R168 in the α2A helix that constitute the lipid binding site are shown in space-filling representation and labeled. Electrostatic calculations were performed using GRASP2 (Petrey and Honig, 2003). Blue and red colors indicate positive and negative electrostatic potential respectively. See Supplementary Methods for molecular docking procedures.

To directly determine the interplay between lipid and peptide binding of Class A PDZ domains, we quantified the binding between SAP102-PDZ3 and the N-fluorescein-labeled stargazin peptide in the presence and absence of PM vesicles (notice that they contain 1% PtdIns(4,5)P2 for which SAP102-PDZ3 shows selectivity) by fluorescence anisotropy measurements. As shown in Fig. 4A, the presence of PM vesicles had little effect on the peptide binding of SAP102-PDZ3. Since a significant portion of the PDZ domain was vesicle-bound under these experimental conditions, the result verifies the notion that SAP102-PDZ3 can simultaneously bind the membrane and a protein molecule. Also, PM vesicles did not affect the binding of SAP102-PDZ3 to other peptides (see Table S4), indicating that lipid binding of Class A PDZ domain does not directly modulate its protein specificity.

Fig 4.

Fig 4

Effects of lipid binding of each class of the PDZ domain on its peptide binding. (A) Binding of Class A SAP102-PDZ3 to F-Ahx-RTTPV in the absence (filled symbols) and presence (open symbols) of 150 μM PM-mimetic vesicles. (B) Binding of Class B1 rhophilin2-PDZ to F-Ahx-EYLGLDVPV in the absence (filled symbols) and presence (open symbols) of 150 μM PM-mimetic vesicles. (C) Binding of Class B2 tamalin-PDZ to F-Ahx-IRDYTQSSSSL in the absence (filled symbols) and presence (open symbols) of 150 μM PM-mimetic vesicles. The peptide concentration was 5 nM. Notice that for the Class A and B1 PDZ domains, vesicles have a modest to no effect on peptide binding whereas for the Class B2 PDZ domain, vesicles greatly interfere with the peptide binding. See Table S4 for Kd values and Methods for experimental details.

Class B PDZ domains typically have cationic residues clustered around the α2A helix that forms a wall of the peptide binding pocket (Fig. S6). Some PDZ domains belonging to this group have been reported to have partially overlapping (Zimmermann et al., 2002) or mutually exclusive (Meerschaert et al., 2009) lipid and peptide binding sites. Since the peptide and lipid binding modes of PDZ domains can vary significantly, however, it is difficult to predict the degree of functional overlap between the two sites based solely on structural examination. We therefore selected two members (rhophilin 2-PDZ and tamalin-PDZ) of this family and determined the location of their lipid binding sites and the degree of their overlap with respective peptide binding sites. Rhophilin 2-PDZ interacts non-specifically with anionic lipids, including PtdInsPs. It has two cationic residues (K576 and K579) on the same side of the α2A near its C-terminal end (Fig. 3B and Fig S6) that may be involved in anionic lipid binding. As shown in Fig. 5A, double-site mutations of K576 and K579 (i.e., K576A/K579A or K576E/K579E) greatly reduced the affinity of rhophilin-2-PDZ for PM vesicles, suggesting their direct involvement in binding to anionic membranes. Interestingly, these mutants bind the C-terminal peptide of ErbB2 as well as the wild type (WT), suggesting that the lipid binding site does not overlap with the peptide binding site (Fig. 5B). We identified this peptide as the best binding partner for the rhophilin-2-PDZ through screening a small library of PDZ domain-binding peptides because its physiological binding partners have not been reported. Our molecular modeling also supports that rhophilin-2-PDZ can simultaneously interact with an anionic lipid headgroup and a peptide (Fig. 3B). To rigorously test the functional independence of lipid and peptide binding sites of rhophilin 2-PDZ, we measured the binding between rhophilin 2-PDZ and the N-fluorescein-labeled peptides in the presence and absence of PM vesicles by fluorescence anisotropy analysis (Fig. 4B and Table S4). Interestingly, the presence of PM vesicles caused a 2-fold increase in the affinity of rhophilin 2-PDZ for the ErbB2 peptide while modestly (i.e., <1.8-fold) decreasing the affinity for other peptides. Thus, neighboring lipid and peptide binding sites of rhophilin 2-PDZ can interact with their binding partners simultaneously but unlike the case of Class A PDZ domains, its lipid binding may enhance the specificity of protein binding. A similar pattern of the interplay between neighboring lipid and peptide binding sites was also seen in Dvl2-PDZ (Sheng et. al., submitted). We thus designated this subgroup of Class B PDZ domains as Class B1.

Fig 5.

Fig 5

Effects of lipid binding of rhophilin 2-PDZ on the cellular localization and function of rhophilin 2. (A) Membrane binding of rhophilin 2-PDZ wild type (WT) and mutants. Binding of PDZ domains to PM-mimetic vesicles was measured by SPR analysis. The two mutants show significantly reduced membrane affinity. (B) GST pull-down assay for rhophilin 2-PDZ WT and K576E/K579E. 5 μg of GST or GST-EYLGLDVPV was incubated with 0–5 μg of rhophilin 2 proteins and the GST-bound proteins were analyzed by SDS-PAGE and immunoblotting by a rhophilin 2 antibody. (C) Confocal microscopic images of F-actin (green) and rhophilin 2 (red) in fixed HeLa cells transiently transfected with mRFP-tagged rhophilin 2 WT. Notice that only the rhophilin 2-expressing cell shows distinct morphology and dramatically reduced F-actin stress fibers. (D–E) Confocal images for K576A/K579A and K576E/K579E mutants, respectively. These mutants show distinctly different subcellular localization patterns from WT and have little effect on F-actin. (F). The control images taken using HeLa cells transfected with the empty expression vector. White bars indicate 10 μm.

The crystal structure of tamalin-PDZ showed that two phosphate ions are bound to the N-terminal end of the α2A, suggesting that three cationic residues, R166, H167, and R168, be involved in anionic lipid binding (Sugi et al., 2008). We found that tamalin-PDZ also lacks definite PtdInsP specificity (Fig. S9A). Mutation of any of the cationic residues to A or E reduced its binding to the PM-mimetic (or PtdInsP-containing) membranes, supporting that these residues are involved in non-specific anionic lipid binding (Fig. S9B). Unlike the case of rhophilin2, however, these mutants showed greatly reduced binding to the C-terminal peptide of mGluR5, an interaction partner of tamalin, indicating a significant degree of overlap between the two binding sites (Fig. S9C), which is consistent with molecular modeling (see Fig. 3C). The functional overlap between the two binding sites is further verified by the finding that the presence of PM vesicles significantly reduced the binding of tamalin-PDZ to the N-terminal-fluorescein-labeled mGluR5 peptide (Fig. 4C) and other peptides (Table S4). To distinguish these PDZ domains from rhophilin-like Class B1 PDZ domains, we designate them Class B2 PDZ domains. Together, these results show that our systematic structural analysis can predict the location of the lipid binding site of PDZ domains with high accuracy. For Class A PDZ domains, lipid and peptide binding sites are topologically distinct and functionally orthogonal but for those Class B PDZ domains with neighboring lipid and peptide binding sites, functional analysis is required to determine the interplay of lipid and peptide binding. Our analysis can also identify those PDZ domains (mostly Class A) with a relatively well-defined lipid binding pocket and hence with lipid specificity.

Physiological Significance of Membrane Binding of PDZ Domains

Physiological roles of the lipid binding activity of Class A (Pan et al., 2007) and Class B2 (Meerschaert et al., 2009; Zimmermann et al., 2002) PDZ domains have been reported. However, the physiological significance of lipid binding activity of Class B1 PDZ domains has not been established presumably due to the subtle nature of the interplay between lipid and protein binding for these PDZ domains. We thus selected rhophilin2-PDZ, a prototypical Class B1 PDZ domain with modest membrane affinity (i.e. Kd = 500 nM), and performed cell studies to determine how lipid binding controls its cellular activity. Rhophilin-2 is a single PDZ domain-containing RhoA-binding protein that inhibits the RhoA’s activity to induce F-actin stress fibers (Peck et al., 2002; Watanabe et al., 1996). It was reported that overexpression of rhophilin-2 in HeLa cells caused disassembly of F-actin stress fibers and that this activity required the presence of its PDZ domain (Peck et al., 2002). When the full-length rhophilin-2 (or its PDZ domain) was expressed as a monomeric red fluorescence protein (mRFP)-fusion protein in HeLa cells, it showed membrane localization with predominant distribution at the perinuclear region and the PM (Fig. 5C). Most important, the cells expressing rhophilin-2 exhibited distinct morphology with dramatically reduced stress fibers, confirming the reported function of rhophilin-2. As illustrated in Fig. 5D and 5E, K576A/K579A and K576E/K579E with reduced membrane affinity show primarily cytosolic distribution with little membrane localization although they have intact protein binding activity (Fig. 5B). Furthermore, both mutations abrogated the activity of rhophilin-2 to disassemble stress fibers as all mutant-expressing cells contain as many stress fibers as control cells (Fig. 5F). Thus, it is evident that lipid binding of rhophilin2-PDZ is important for the physiological function of rhophilin2. These results in conjunction with previous reports on Class A and Class B2 PDZ domains suggest that lipid binding activity of all three classes of lipid binding PDZ domains is important for physiological function and regulation of their host proteins.

DISCUSSION

This study describes genome-wide identification, characterization, and classification of membrane-binding PDZ domains. Experimental characterization of 95 PDZ domains confirms that membrane binding is a common property of PDZ domains. Our quantitative data reveal that PDZ domains have a wide continuous range (i.e., 20 nM to >10 μM) of affinity for PM vesicles, making it difficult to arbitrarily distinguish membrane-binding domains from non-binding ones. This also underscores the risk of identifying membrane binding PDZ domains by a qualitative assay. Taking this factor into account, we developed a flexible and robust binary classification strategy in which a threshold or cut-off Kd value is arbitrarily set and the domains are then divided into those with higher affinity (binding) and those with lower affinity (non-binding). Our on-line resource provides an option to set the threshold Kd value as a variable and one can thus predict the membrane binding activity of PDZ domains with flexibility. Our new classification and prediction protocols represent a major advancement in bioinformatics computation because it allows accurate prediction of membrane-binding proteins from a group of proteins with high sequence and structural similarity. The same methodology can be applied to the prediction of any other membrane binding PIDs that might act as dual-specificity protein- and lipid-binding modules.

A recent study indicated that at least 80% of mouse PDZ domains have protein (or peptide) binding activity (Stiffler et al., 2007). Given that ≈30% of mouse PDZ domains have submicromolar affinity for the PM membrane, the probability that a mouse PDZ domain is a lipid- and protein-binding dual-specificity module is >24%. Also, if a PDZ domain is found to bind membranes the probability that it can also bind proteins is >90%. These are conservative estimates and actual numbers might well be higher for PDZ domains from mouse and other species. Thus, it is safe to state that almost all lipid-binding PDZ domains are dual-specificity modules. To gain insight into the evolution of lipid- and protein-binding activities of PDZ domains, we constructed a dendrogram depicting the evolutionary relationship of a collection of PDZ domains (Fig. S10). The dendrogram shows that the binding to specific protein classes is well preserved in the tree, whereas membrane-binding properties vary even for evolutionary closely related PDZ domains (e.g., Dvl and SAP97/PSD95/Chapsyn110 clusters). Also, most PDZ domains have peptide binding activity and the location of their peptide binding pockets is essentially invariable (Sheng and Sala, 2001), while they show a wide range of membrane binding activity and the location of their lipid binding sites are highly variable. These findings all suggest that dual specificity lipid-and protein-binding PDZ domains evolved from protein-binding ancestor PDZ domains through convergent evolution. Newly acquired lipid binding activity should confer additional functionality to PDZ domains and also allow extra layer of regulation on the critical cellular functions of their host proteins.

Systematic and comprehensive electrostatic potential calculation of membrane-binding PDZ domains reveals two distinct patterns of cationic patch distribution, allowing for their classification into two groups. Mutational and functional analysis confirmed that Class A PDZ domains have topologically distinct and functionally orthogonal lipid and protein binding sites. Thus, these PDZ domains can serve as dual specificity lipid- and protein-binding modules that mediate membrane-associated protein networking through coincident binding. Also, many of Class A PDZ domains have a relatively well-formed cationic groove (Fig. S5 and Table S3), suggesting that they have definite lipid headgroup selectivity (Table 1) in contrast to most Class B PDZ domains with low lipid selectivity. Our results show that lipid binding of Class A PDZ domains affects neither affinity nor specificity per se for peptide binding. Under physiological conditions, however, their lipid binding should enhance affinity and specificity for their protein partners, many of which are PM-associated proteins, due to reduction in dimensionality (Cho, 2006; McCloskey and Poo, 1986). Thus, their dual-specificity should be pivotal for the formation and regulation of membrane-associated protein networking.

Essentially all Class B PDZ domains have their prominent cationic patches in or near the α2A helix that forms a wall of the peptide binding pocket. Interestingly, Class B1 PDZ domains have cationic patches confined near the C-terminal end of the α2A helix whereas Class B2 PDZ domains have cationic patches in the N-terminal end of the helix or scattered over the helix (Fig. S7). Since a protein typically enters the pocket from the N-terminal end of the α2A helix to place its C-terminus near the carboxylate-binding loop, it is possible that lipid binding at the N-terminal end of the α2A helix interferes with the protein binding more than that at the C-terminal end of the helix. Further characterization of Class B PDZ domains would undoubtedly expand the list of Class B1 PDZ domains and also reveal if the distribution of cationic residues in the α2A is the main determinant for Class B1 PDZ domains. Class B1 PDZ domains are similar to Class A PDZ domains in that both act as dual-specificity modules that mediate membrane-associated protein networking through coincident binding. Unlike the case of Class A PDZ domains, however, lipid binding may enhance the protein binding specificity of Class B1 PDZ domains, presumably through a local conformational change of the peptide binding pocket. Undoubtedly, further studies are necessary to test this potentially important notion. Since lipid and peptide binding are mutually exclusive and compete with each other for Class B2 PDZ domains, lipid binding should negatively control their protein binding; i.e., lipids may act as molecular switch that regulates the accessibility of the protein binding pocket.

The main regulatory role of lipids is modulation of the localization and the activity of proteins. Thus, lipid binding of PDZ domains would primarily control the membrane localization and/or the activity of their host proteins. For most reported Class A and Class B2 PDZ domains, their lipid binding activity appears to be important for the cellular localization of their host proteins. Our results from rhophilin2 containing a Class B1 PDZ domain would seem to indicate that this is the common feature of all dual-specificity PDZ domains. However, the correlation between membrane affinity (i.e., PM affinity) and cellular membrane localization may not be straightforward because their membrane localization may also depend on interactions with membrane proteins or membrane-associated proteins. In the case of the yeast scaffold protein Ste5, lipid binding was reported to modulate the dynamics and function of the protein at the PM rather than PM localization per se (Winters et al., 2005). Likewise, lipid binding of some PDZ domains may exerts more effects on the dynamics and function of their host proteins at the membrane than on their membrane localization. Also, some PDZ domains are found associated with intracellular organelles (Meerschaert et al., 2009; Zimmermann et al., 2002), suggesting that PDZ domains may interact with other cellular membranes. In any case, however, our experimental and predicted PM affinity should still serve as a reliable indicator of the likelihood of a PDZ domain to interact with any cell membrane because the electrostatic interaction between the PDZ domain and anionic lipids is the main driving force for its binding to all intracellular cell membranes. Lastly, it should be noted that some PDZ domains may hetero- or homo-dimerize (or oligomerize) under physiological conditions (Feng and Zhang, 2009; Sheng and Sala, 2001), which may modulate their effective membrane affinity through avidity or conformational effect.

Taken all together, our results strongly support the hypothesis that many PIDs involved in protein networking at the membrane serve as dual specificity lipid- and protein-interaction modules. Also, lipid binding of different classes of PDZ domains regulates the cellular function and regulation of their host proteins by different mechanisms. Thus, it is becoming increasingly evident that the interpretation of complex data on the regulation of cellular protein interactions and networking entails elucidation of the membrane binding properties of PIDs. Our experimental and computational data on PDZ domains should serve as a valuable resource not only for those investigators working on PDZ domain proteins but also for other proteins that mediate cellular protein interactions and networking. Also, our structure-based functional classification approach should provide framework for further functional characterization of all dual-specificity PDZ domains and other PIDs.

METHODS

Protein Expression and Purification

All PDZ domains were expressed as His6-tagged proteins in Escherichia coli BL21 (DE3) pLysS (Novagen). For details, see Supplemental Methods.

Lipid Vesicles Preparation and SPR Analysis

PM-mimetic vesicles were prepared by mixing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoserine, cholesterol, liver phosphoinositol, and 1,2-dipalmitoyl derivatives of phosphatidylinositol-(4,5)-bisphosphate in a molar ratio of 12:35:22:22:8:1. Large unilamellar vesicles were prepared using a Liposofast (Avestin) microextruder with a 100 nm polycarbonate filter. All SPR measurements were performed at 23 °C using a lipid-coated L1 chip in the BIACORE X system as described (Stahelin and Cho, 2001). 20 mM Tris-HCl, pH 7.4, containing 0.16 M NaCl was used as the running buffer while PM-mimetic vesicles and POPC vesicles were coated on the active surface and the control surface, respectively. For details, see Supplemental Methods.

Mammalian Cell Assay

HeLa cells were maintained in minimum essential medium eagle’s medium (MEME) supplemented with 10% fetal bovine serum at 37°C in 10% CO2. Cells were transiently transfected with 0.5 μg of appropriate plasmid DNA using Lipofectamine Plus (Invitrogen). After 18 hours, the cells were fixed using 4% formaldehyde solution, and the F-actin was stained using Oregon Green 488-conjugated phalloidin (Invitrogen). Images were taken using a Zeiss LSM510 confocal microscope.

Screening of Rhophilin2-PDZ-Binding Peptides

Since physiological binding partners for the Rhophilin-2 PDZ domain are unknown, we constructed a small library of PDZ domain-binding peptides and screened them against the Rhophilin-2-PDZ. The library contains several representative C-terminal nonapeptides for each class of PDZ domains with differential peptide specificity (Sheng and Sala, 2001); i.e., neuroligin (PSD95-PDZ3-specific), CFTR (NHERF1-PDZ1-specific) and Frizzled7 (Dvl2-PDZ-specific) for Class I PDZ domains; GluR2 (GRIP-PDZ5-specific) and ErbB2 (erbin-PDZ-specific) for Class II PDZ domains; melatonin receptor (nNOS-PDZ-specific) and merlin (syntenin PDZ1-specific) for Class III PDZ domains. The GST pull-down and immunoblotting assay showed that rhophilin-2-PDZ had high selectivity for the C-terminal peptide of ErbB2, EYLGLDVPV.

PDZ-peptide binding assay by fluorescence anisotropy

Fluorescein-6-aminohexanoyl (F-Ahx)-labeled peptides, F-Ahx-RTTPV (for SAP102-PDZ binding), F-Ahx-EYLGLDVPV (for rhophilin2-PDZ binding), and F-Ahx-IRDYTQSSSSL (for tamalin-PDZ binding), were dissolved in 20 mM Tris buffer, pH 7.9, containing 160 mM NaCl, 300 mM imidazole and 5% dimethylsulfoxide. To each well of a 96-flat bottom black polystyrol plate was added 100 μl solution containing each peptide (5 nM) and PDZ domain (100 nM to 1 mM) with or without 150 μM PM vesicles. After 30-min incubation, the plate was inserted into Tecan Genios Pro spectrofluorometer and the fluorescence anisotropy (r) was measured with excitation and emission wavelengths set at 485 and 535 nm, respectively. Since PoPepo under our conditions, the Kd for the PDZ domain-peptide binding was determined by the nonlinear least-squares analysis of the binding isotherm using the equation:

Pepbound/Pepo=Δr/Δrmax=11+Kd/Po

where Pepbound, Pepo, and Po indicate the concentration of bound peptide, total peptide and total PDZ domain, respectively, and Δr and Δrmax are the anisotropy change for each Po and the maximal Δr, respectively.

Bioinformatics Methods, Molecular Docking and Calculation of Electrostatic Potential

Detailed descriptions for feature calculations, classifiers, classifier evaluations, homology modeling, and electrostatic potential calculation with GRASP2 (Petrey and Honig, 2003) are included in Supplemental Information

Supplementary Material

01

HIGHLIGHTS.

  • Membrane-binding PDZ domains are identified and classified on a genomic scale.

  • Many PDZ domains serve as a dual-specificity lipid- and protein-binding module.

  • Lipid binding of PDZ domains regulates their function by different mechanisms.

Acknowledgments

This work was in part supported by the Chicago Biomedical Consortium with support from The Searl Funds at the Chicago Community Trust and the National Institutes of Health (GM68849). The work was also supported by the World Class University program R31-2008-000-10105-0 (W.C.) through the National Research Foundation of Korea funded by the Ministry of Education, Science and Technology. M.K. thanks support from FMC Technologies Fund Fellowship.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Bhardwaj N, Langlois RE, Zhao G, Lu H. Kernel-based machine learning protocol for predicting DNA-binding proteins. Nucleic Acids Res. 2005;33:6486–6493. doi: 10.1093/nar/gki949. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bhardwaj N, Stahelin RV, Langlois RE, Cho W, Lu H. Structural Bioinformatics Prediction of Membrane-binding Proteins. J Mol Biol. 2006;359:486–495. doi: 10.1016/j.jmb.2006.03.039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bhardwaj N, Stahelin RV, Zhao G, Cho W, Lu H. MeTaDoR: a comprehensive resource for membrane targeting domains and their host proteins. Bioinformatics. 2007;23:3110–3112. doi: 10.1093/bioinformatics/btm395. [DOI] [PubMed] [Google Scholar]
  4. Bhattacharyya RP, Remenyi A, Yeh BJ, Lim WA. Domains, motifs, and scaffolds: the role of modular interactions in the evolution and wiring of cell signaling circuits. Annu Rev Biochem. 2006;75:655–680. doi: 10.1146/annurev.biochem.75.103004.142710. [DOI] [PubMed] [Google Scholar]
  5. Blatner NR, Stahelin RV, Diraviyam K, Hawkins PT, Hong W, Murray D, Cho W. The molecular basis of the differential subcellular localization of FYVE domains. J Biol Chem. 2004;279:53818–53827. doi: 10.1074/jbc.M408408200. [DOI] [PubMed] [Google Scholar]
  6. Bray D. Signaling complexes: biophysical constraints on intracellular communication. Annu Rev Biophys Biomol Struct. 1998;27:59–75. doi: 10.1146/annurev.biophys.27.1.59. [DOI] [PubMed] [Google Scholar]
  7. Cho W. Building signaling complexes at the membrane. Sci STKE. 2006:pe7. doi: 10.1126/stke.3212006pe7. [DOI] [PubMed] [Google Scholar]
  8. Cho W, Bittova L, Stahelin RV. Membrane binding assays for peripheral proteins. Anal Biochem. 2001;296:153–161. doi: 10.1006/abio.2001.5225. [DOI] [PubMed] [Google Scholar]
  9. Cho W, Stahelin RV. Membrane-protein interactions in cell signaling and membrane trafficking. Annu Rev Biophys Biomol Struct. 2005;34:119–151. doi: 10.1146/annurev.biophys.33.110502.133337. [DOI] [PubMed] [Google Scholar]
  10. Di Paolo G, De Camilli P. Phosphoinositides in cell regulation and membrane dynamics. Nature. 2006;443:651–657. doi: 10.1038/nature05185. [DOI] [PubMed] [Google Scholar]
  11. DiNitto JP, Cronin TC, Lambright DG. Membrane recognition and targeting by lipid-binding domains. Sci STKE. 2003:re16. doi: 10.1126/stke.2132003re16. [DOI] [PubMed] [Google Scholar]
  12. DiNitto JP, Lambright DG. Membrane and juxtamembrane targeting by PH and PTB domains. Biochim Biophys Acta. 2006;1761:850–867. doi: 10.1016/j.bbalip.2006.04.008. [DOI] [PubMed] [Google Scholar]
  13. Feng W, Zhang M. Organization and dynamics of PDZ-domain-related supramodules in the postsynaptic density. Nat Rev Neurosci. 2009;10:87–99. doi: 10.1038/nrn2540. [DOI] [PubMed] [Google Scholar]
  14. Lee CS, Kim IS, Park JB, Lee MN, Lee HY, Suh PG, Ryu SH. The phox homology domain of phospholipase D activates dynamin GTPase activity and accelerates EGFR endocytosis. Nat Cell Biol. 2006;8:477–484. doi: 10.1038/ncb1401. [DOI] [PubMed] [Google Scholar]
  15. Lemmon MA. Membrane recognition by phospholipid-binding domains. Nat Rev Mol Cell Biol. 2008;9:99–111. doi: 10.1038/nrm2328. [DOI] [PubMed] [Google Scholar]
  16. McCloskey MA, Poo MM. Rates of membrane-associated reactions: reduction of dimensionality revisited. J Cell Biol. 1986;102:88–96. doi: 10.1083/jcb.102.1.88. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. McLaughlin S, Murray D. Plasma membrane phosphoinositide organization by protein electrostatics. Nature. 2005;438:605–611. doi: 10.1038/nature04398. [DOI] [PubMed] [Google Scholar]
  18. Meerschaert K, Tun MP, Remue E, De Ganck A, Boucherie C, Vanloo B, Degeest G, Vandekerckhove J, Zimmermann P, Bhardwaj N, et al. The PDZ2 domain of zonula occludens-1 and -2 is a phosphoinositide binding domain. Cell Mol Life Sci. 2009;66:3951–3966. doi: 10.1007/s00018-009-0156-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Mulgrew-Nesbitt A, Diraviyam K, Wang J, Singh S, Murray P, Li Z, Rogers L, Mirkovic N, Murray D. The role of electrostatics in protein-membrane interactions. Biochim Biophys Acta. 2006;1761:812–826. doi: 10.1016/j.bbalip.2006.07.002. [DOI] [PubMed] [Google Scholar]
  20. Pan L, Wu H, Shen C, Shi Y, Jin W, Xia J, Zhang M. Clustering and synaptic targeting of PICK1 requires direct interaction between the PDZ domain and lipid membranes. EMBO J. 2007;26:4576–4587. doi: 10.1038/sj.emboj.7601860. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Park WS, Heo WD, Whalen JH, O’Rourke NA, Bryan HM, Meyer T, Teruel MN. Comprehensive identification of PIP3-regulated PH domains from C. elegans to H. sapiens by model prediction and live imaging. Mol Cell. 2008;30:381–392. doi: 10.1016/j.molcel.2008.04.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Pawson T. Specificity in signal transduction: from phosphotyrosine-SH2 domain interactions to complex cellular systems. Cell. 2004;116:191–203. doi: 10.1016/s0092-8674(03)01077-8. [DOI] [PubMed] [Google Scholar]
  23. Pawson T, Nash P. Assembly of cell regulatory systems through protein interaction domains. Science. 2003;300:445–452. doi: 10.1126/science.1083653. [DOI] [PubMed] [Google Scholar]
  24. Peck JW, Oberst M, Bouker KB, Bowden E, Burbelo PD. The RhoA-binding protein, rhophilin-2, regulates actin cytoskeleton organization. J Biol Chem. 2002;277:43924–43932. doi: 10.1074/jbc.M203569200. [DOI] [PubMed] [Google Scholar]
  25. Petrey D, Honig B. GRASP2: visualization, surface properties, and electrostatics of macromolecular structures and sequences. Methods in enzymology. 2003;374:492–509. doi: 10.1016/S0076-6879(03)74021-X. [DOI] [PubMed] [Google Scholar]
  26. Ravichandran KS, Zhou MM, Pratt JC, Harlan JE, Walk SF, Fesik SW, Burakoff SJ. Evidence for a requirement for both phospholipid and phosphotyrosine binding via the Shc phosphotyrosine-binding domain in vivo. Mol Cell Biol. 1997;17:5540–5549. doi: 10.1128/mcb.17.9.5540. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Schultz J, Milpetz F, Bork P, Ponting CP. SMART, a simple modular architecture research tool: identification of signaling domains. Proc Natl Acad Sci U S A. 1998;95:5857–5864. doi: 10.1073/pnas.95.11.5857. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Sheng M, Sala C. PDZ domains and the organization of supramolecular complexes. Annu Rev Neurosci. 2001;24:1–29. doi: 10.1146/annurev.neuro.24.1.1. [DOI] [PubMed] [Google Scholar]
  29. Stahelin RV, Cho W. Differential roles of ionic, aliphatic, and aromatic residues in membrane-protein interactions: a surface plasmon resonance study on phospholipases A2. Biochemistry. 2001;40:4672–4678. doi: 10.1021/bi0020325. [DOI] [PubMed] [Google Scholar]
  30. Stahelin RV, Long F, Peter BJ, Murray D, De Camilli P, McMahon HT, Cho W. Contrasting membrane interaction mechanisms of AP180 N-terminal homology (ANTH) and epsin N-terminal homology (ENTH) domains. J Biol Chem. 2003;278:28993–28999. doi: 10.1074/jbc.M302865200. [DOI] [PubMed] [Google Scholar]
  31. Stiffler MA, Chen JR, Grantcharova VP, Lei Y, Fuchs D, Allen JE, Zaslavskaia LA, MacBeath G. PDZ domain binding selectivity is optimized across the mouse proteome. Science. 2007;317:364–369. doi: 10.1126/science.1144592. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Sugi T, Oyama T, Morikawa K, Jingami H. Structural insights into the PIP2 recognition by syntenin-1 PDZ domain. Biochem Biophys Res Commun. 2008;366:373–378. doi: 10.1016/j.bbrc.2007.11.138. [DOI] [PubMed] [Google Scholar]
  33. Watanabe G, Saito Y, Madaule P, Ishizaki T, Fujisawa K, Morii N, Mukai H, Ono Y, Kakizuka A, Narumiya S. Protein kinase N (PKN) and PKN-related protein rhophilin as targets of small GTPase Rho. Science. 1996;271:645–648. doi: 10.1126/science.271.5249.645. [DOI] [PubMed] [Google Scholar]
  34. White SH, Wimley WC, Ladokhin AS, Hristova K. Protein folding in membranes: determining energetics of peptide-bilayer interactions. Methods Enzymol. 1998;295:62–87. doi: 10.1016/s0076-6879(98)95035-2. [DOI] [PubMed] [Google Scholar]
  35. Winters MJ, Lamson RE, Nakanishi H, Neiman AM, Pryciak PM. A membrane binding domain in the ste5 scaffold synergizes with gbetagamma binding to control localization and signaling in pheromone response. Mol Cell. 2005;20:21–32. doi: 10.1016/j.molcel.2005.08.020. [DOI] [PubMed] [Google Scholar]
  36. Wu H, Feng W, Chen J, Chan LN, Huang S, Zhang M. PDZ domains of Par-3 as potential phosphoinositide signaling integrators. Mol Cell. 2007;28:886–898. doi: 10.1016/j.molcel.2007.10.028. [DOI] [PubMed] [Google Scholar]
  37. Yao L, Kawakami Y, Kawakami T. The pleckstrin homology domain of Bruton tyrosine kinase interacts with protein kinase C. Proceedings of the National Academy of Sciences of the United States of America. 1994;91:9175–9179. doi: 10.1073/pnas.91.19.9175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Yoon Y, Lee PJ, Kurilova S, Cho W. In situ quantitative imaging of cellular lipids using molecular sensors. Nat Chem. 2011;3:868–874. doi: 10.1038/nchem.1163. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Zhou MM, Ravichandran KS, Olejniczak EF, Petros AM, Meadows RP, Sattler M, Harlan JE, Wade WS, Burakoff SJ, Fesik SW. Structure and ligand recognition of the phosphotyrosine binding domain of Shc. Nature. 1995;378:584–592. doi: 10.1038/378584a0. [DOI] [PubMed] [Google Scholar]
  40. Zimmermann P. The prevalence and significance of PDZ domain-phosphoinositide interactions. Biochim Biophys Acta. 2006;1761:947–956. doi: 10.1016/j.bbalip.2006.04.003. [DOI] [PubMed] [Google Scholar]
  41. Zimmermann P, Meerschaert K, Reekmans G, Leenaerts I, Small JV, Vandekerckhove J, David G, Gettemans J. PIP(2)-PDZ domain binding controls the association of syntenin with the plasma membrane. Mol Cell. 2002;9:1215–1225. doi: 10.1016/s1097-2765(02)00549-x. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

01

RESOURCES