Over 250 PDZ (PSD95/Dlg/ZO-1) domain-containing proteins have been described in the human proteome. As many of these possess multiple PDZ domains, the potential combinations of associations with proteins that possess PBMs (PDZ-binding motifs) are vast. However, PDZ domain recognition is a highly specific process, and much less promiscuous than originally thought. Furthermore, a large number of PDZ domain-containing proteins have been linked directly to the control of processes whose loss, or inappropriate activation, contribute to the development of human malignancies. These regulate processes as diverse as cytoskeletal organization, cell polarity, cell proliferation and many signal transduction pathways. In the present review, we discuss how PBM–PDZ recognition and imbalances therein can perturb cellular homoeostasis and ultimately contribute to malignant progression.
The PDZ domain-containing family of proteins were first recognized in the early 1990s and were classified as such, by having stretches of amino acids in common with the first such members to be identified, PSD95 (postsynaptic density protein 95), Dlg (discs large) and ZO-1 (zonula occludens protein 1), hence the origin of the name . They are found from bacteria to vertebrates  and are involved in an extremely large number of activities within the cell, including control of cell migration and invasion, cell proliferation, cell polarity, cell attachment and cell–cell contact, apoptosis and immune cell recognition and signalling. PDZ domains themselves are sites of protein–protein interaction, and, as many of these proteins have multiple PDZ domains capable of multiple protein interactions, they are generally considered, somewhat simplistically, to act as scaffolding molecules, around which multi-protein signalling complexes are assembled.
Before discussing how PDZ domain-containing proteins might affect tumorigenesis, we need to consider the molecular basis for PDZ domain recognition. It is often impossible to separate the functions of PDZ domain proteins from those of their ligands that possess PBMs (PDZ-binding motifs). Therefore one must always remember that this is a two-way interaction where, for example, the PDZ domain-containing protein may exert tumour-suppressor activity by regulating the activity of a PBM-containing ligand, such as Scribble with β-Pix . Conversely, a PBM ligand may act directly on, and inactivate or sequester, a PDZ domain-containing partner, for example, the interaction between the HPV (human papillomavirus) E6 oncoprotein and Dlg .
STRUCTURAL BASIS OF PDZ–PBM RECOGNITION
PDZ domains are regions of usually 80–90 amino acid residues, all having certain defining structural elements: six β-sheets (βA–βF) and two α-helices, one short (αA) and one long (αB) [5–7]. These are shown as a cartoon in Figure 1(A), aligned with each of the three PDZ domains of human Dlg1. Although there is considerable variation in the length and structure of the linking sequences, these conserved elements form a highly conserved fold, which contains the carboxy-binding loop, between the βA and βB sheets [5,6]. This is the so-called GLGF motif, actually R/K-X-X-X-G-Φ-G-Φ where X is any residue and Φ is a hydrophobic residue; the side chains of these residues form the hydrophobic pocket that provides the binding activity of PDZ domains [5,8]. In fact, apart from the totally conserved second glycine residue, the exact residues forming the GLGF motif can vary quite significantly and thus contribute to binding specificity.
PDZ domain structure
The canonical PDZ-binding domains fold to bring the N- and C-termini of the domain close together at the ‘back’ of the structure, while the binding groove is formed by the βB sheet and αB helix, with the GLGF motif at the end, reminiscent of the contact point in an electric socket. The shape of a correctly folded PDZ domain is shown in Figure 1(B), based on the crystal structure of the Dlg PDZ domain 2 . It has been shown in the PDZ2 domain of PSD95, that the loop connecting the βA and βB sheets (which contains the GLGF motif), and the loop connecting the αB helix and βF sheet both retain flexibility in solution, whereas the remainder of the structure is relatively rigid . It is easy to envisage that the flexibility of the loops could allow the rigid elements to move relative to one another, facilitating the insertion of the binding ligand, whereas the rigidity of the basic structure would keep the domain in binding readiness, even in the absence of a ligand.
The discussion above refers to simple PDZ structures, but as PDZ domains are often found in multiples on the same protein, it is clear that they may not necessarily work in isolation. Indeed, some PDZ domains function in tandem, structurally stabilizing each other, as in the case of the GRIP1 (glutamate-receptor-interacting protein 1) tandem [10,11]. The PDZ1/PDZ2 domains of PSD95 and of Dlg may also function as tandems in order to bind certain dimeric ligands [11,12], but they have also been shown to function independently.
Another structural variation is dimerization, which can be either interdomain or intradomain. In interdomain dimerization, the PDZ domain of one protein molecule binds the PDZ domain of another; usually ‘back-to-back’ and antiparallel, allowing two ligands to be bound simultaneously and in close proximity. Such dimers have been shown for MAGI (membrane-associated guanylate kinase with inverted domain structure)-1 PDZ1  and Shank1 PDZ . In intradomain dimers, two protein molecules intertwine forming two PDZ domains, with each molecule providing part of one domain and part of the other. However, these have only been described in the ZO-1 and ZO-2 proteins [14,15], where this type of PDZ domain structure appears to modulate the PBM-containing ligand, such as the polymerization of claudins through binding ZO-2 and the regulation of connexin43 through binding ZO-1 [15,16].
A note of caution should be added at this stage. As will be seen below, many PDZ domains and PBMs are reported to have multiple protein partners and it is often difficult to determine which are biologically relevant. This is often due to the context in which the interactions have been analysed: at low stringency, a class I PBM could bind any class I PDZ, and obviously this is not the situation in vivo. This is perfectly illustrated with the example of E6 and Dlg, where in vivo recognition is known to occur through PDZ domain 2, yet under in vitro conditions, including crystallization, E6 can also associate with PDZ domain 3 .
As the PDZ domains have a variety of structural characteristics to give specificity to their interactions, so too do the PBMs of their ligands.
PDZ domains have traditionally been classified by the sequence of the PBM that binds them, the exact number of classes depending on the criteria used. PBMs are short amino acid motifs, usually found at the extreme C-termini of PDZ-binding proteins , although internal PBMs exist [18,19]. In general, the ligand inserts the C-terminal PBM into the binding fold of the PDZ domain to contact the GLGF motif. In Class I PDZ domains, the PBM binds the fold as a β-strand antiparallel to the βB strand and parallel to the αB helix of the PDZ domain; baseline binding being provided by the ligand's C-terminal -COOH contact with the GLGF loop; and by the serine/threonine residue contacting a conserved histidine residue in the αB helix ; this is shown in cartoon form in Figure 1(C).
Traditionally, PDZ domains were classified into classes I–V, depending on the consensus sequences of the PBMs. However, studies showing that five, six  or even seven  amino acid residues of the PBM can be involved in binding prompted an analysis of 72 PDZ domains that gave a classification having 16 subclasses . This classification is shown in an abbreviated form in Figure 2, where it can be seen that even this system leaves considerable flexibility. Thus the robustness of basic binding combined with the variations in specificities induced by variation of either the PDZ domain or the ligand combine to provide a strong evolutionary pressure towards refining binding specificity.
PDZ domain classes are defined by the consensus sequence of the PBM
Control of PDZ binding
Since many PDZ–PBM interactions involve signalling molecules, it must be assumed that, in addition to having exquisite binding specificity, there must also be a means of control. Many PBMs have PKA (protein kinase A)-recognition sequences embedded within the motif and subsequent PKA phosphorylation can block the PBM–PDZ recognition, as in the case of Kir2.3 with PSD95 , and HPV E6 with Dlg . Interestingly, phosphorylation of the PTEN (phosphatase and tensin homologue deleted on chromosome 10) PBM also blocks Dlg and MAGI binding, but may stimulate binding to other targets , whereas PKA phosphorylation of Class III ligands can enhance binding . Phosphorylation of the PDZ domain can also regulate PBM recognition, with phosphorylation either inhibiting or increasing the interaction, depending on the particular PDZ–PBM combination [26,27]. In short, it is clear that modifications to the PDZ-binding activity can have major downstream effects, emphasizing the importance of control of PDZ interactions in cellular homoeostasis, many of which may well be disrupted in malignancy.
ONCOGENIC VIRUSES AND PDZ DOMAIN-CONTAINING PROTEINS
The demonstration that certain human tumour virus oncoproteins could target PDZ domain-containing substrates provided an exciting early indication of their relevance in the development of human tumours.
One of the first suggestions that PBMs might confer oncogenic potential and, by implication, that their PDZ domain-containing substrates might have tumour-suppressor potential, came from studies in a mouse model of mammary tumorigenesis. In this system, adenovirus 9 E4ORF1 protein exhibited potent transforming potential, which was dependent on an intact PBM . The precise contributions of specific PDZ domain-containing substrates have remained elusive, although Dlg and MAGI-1 appear to be good candidates as tumour suppressors in this system . However, studies suggest that, rather than simply inhibiting Dlg function, the adenovirus 9 E4ORF1 interaction may actually alter Dlg function such that it becomes pro-oncogenic .
A particularly exciting development was the realization that oncogenic HPVs, the causative agents of cervical cancer, also encode an oncoprotein possessing a PBM [28,31]. Intriguingly, only those HPV types that cause cervical cancer, so called ‘high-risk’ types, such as HPV-16 or HPV-18, possess an E6 oncoprotein with PDZ-binding potential. Low-risk viruses, which cause benign lesions, such as HPV-6 and HPV-11, do not have PBMs on their E6 proteins. In addition, an intact PBM on E6 is essential for many of E6's associated activities, including the normal viral life cycle, induction of EMT (epithelial–mesenchymal transition) and induction of malignancy in transgenic mouse models . Since continued E6 expression is required for maintenance of the malignant phenotype in cervical tumours , this suggests that E6 PBM–PDZ interactions play critical roles during the viral life cycle and at different stages of malignant progression. Another human tumour virus HTLV-1 (human T-cell leukaemia virus type 1) also encodes an oncogene, Tax, which has a PBM, through which it interacts with Dlg . Unlike E6, Tax does not seem to be involved in the later stages of disease, but an intact PBM does seem essential for the capacity of Tax to transform cells , suggesting that PBM–PDZ interactions might contribute to tumour initiation by HTLV-1.
A great deal of attention has been paid to identifying relevant PDZ domain-containing proteins for HPV-induced carcinogenesis. Currently over ten such substrates of HPV-16 and HPV-18 E6 have been described . Which of these are critical players in HPV-induced transformation remains to be elucidated. This is complicated by the fact that different HPV E6 oncoproteins target different PDZ domain-containing substrates with different efficiencies. Because of single amino acid differences in the PBMs, HPV-16 E6 interacts with Scribble, whereas HPV-18 E6 prefers Dlg . Both are essential regulators of the cell polarity control module, with perturbation of either being expected to disturb the function of the whole complex . However, the precise outcome for viral-induced tumorigenesis is likely to be different, depending on the choice of PDZ domain-containing substrate by the different HPV E6 oncoproteins.
In human cells, the function of Dlg is still open to question, although it has been linked with two known tumour suppressors, APC (adenomatous polyposis coli) and PTEN (see below) and two proto-oncogenes, PI3K (phosphoinositide 3-kinase) and Net1 (see below). Significantly more is known about Scribble. It has been shown to be a potent regulator of the Ras/MAPK (mitogen-activated protein kinase) signalling pathway, with mutations in or loss of human Scribble co-operating with activated Ras-induced carcinogenesis; similar results have also been found with the c-Myc oncogene [37,38]. In both cases, Scribble appears to function partly by down-regulating the ERK (extracellular-signal-regulated kinase) signalling pathway, possibly via direct interaction with ERK . Scribble also directly regulates the JNK (c-Jun N-terminal kinase) signalling pathway and this contributes to oncogene-induced transformation, and affects the susceptibility of cells to enter apoptosis . Thus Scribble can be considered to be a true tumour suppressor in these models of tumorigenesis. What about in human tumours? Certainly in many cancers of cervix, breast and colon [41,42], Scribble protein levels decrease dramatically as the tumours progress, and, by late-stage metastatic cancer, the levels are invariably very low. However, in pre-malignant cervical cancer, and in some other tumours , Scribble is often overexpressed and mislocalized, suggesting that aberrant patterns of expression might also contribute to tumorigenesis, depending on the particular context.
Another intriguing PDZ target of HPV E6 was found to be MAGI-1 . As noted below, this is implicated in many pathways, including the potential regulation of the PTEN tumour suppressor. However, in the context of HPV-induced malignancy, no alterations in PTEN signalling have been reported. However, like most multi-PDZ domain-containing proteins, MAGI-1 is multifunctional, and also has a role in the regulation of TJs (tight junctions): the capacity of E6 to bind and degrade MAGI-1 correlates perfectly with the capacity of the virus to perturb TJ assembly .
RhPV (rhesus papillomavirus) induces cervical cancer in the rhesus macaque, and interestingly was found to have a PBM on the E7 oncoprotein instead of E6. A proteomic screen of interacting partners identified Par3 (partitioning defective 3) (see below) as the PDZ domain-containing partner . This was an exciting observation since Par3 functions in the same pathway of polarity control as Scribble and Dlg. While not wishing to overinterpret a single study, it is nonetheless intriguing that multiple studies in human cancers have suggested a role for the Par complex in tumour development , and the demonstration that cancer-causing RhPV E7 targets this complex supports this.
Thus, although the list of PDZ domain-containing targets of HPV E6 is probably not complete, this viral protein has nonetheless directed us towards a small number of cellular PDZ domain-containing proteins that might be expected to play roles in the development of human cancers. As we will see from the following discussion, this seems to be the case for many of these substrates.
PDZ DOMAINS AND WNT SIGNALLING
Many PDZ-containing proteins are involved in signalling, and this is particularly striking in the case of Wnt signalling. This facilitates communication between neighbouring cells through the secretion of Wnt proteins, which interact with Frizzled receptors, thereby triggering a signalling cascade in the stimulated cells. Normally active during embryonic development, the unscheduled reactivation of Wnt signalling in adult tissues is a hallmark of tumorigenesis .
Central to the Wnt signalling cascade is Dvl (Dishevelled), a single PDZ domain-containing protein that transduces the signal from activated Frizzled receptors to the downstream effectors of the pathway (Figure 3). Dvls bind directly to Frizzled receptors, the majority of which also have PBMs, in a PDZ-dependent manner, through an internal PBM located in the intracellular segment of Frizzled .
PDZ domain interactions and the regulation of EMT and Wnt signalling
Depending on the particular combination of Wnt protein and Frizzled receptor, Dvl can activate the canonical or the non-canonical Wnt signalling pathways . The former involves the disassembly of the β-catenin destruction complex, resulting in β-catenin stimulation of cell proliferation through the transcriptional activation of TCF (T-cell factor)/LEF (lymphoid enhancer factor) target genes . The non-canonical or PCP (planar cell polarity) pathway, promotes cytoskeleton remodelling and cell migration [52,53]. This involves Dvl activation of Daam1 (Dvl-associated activator of morphogenesis 1), in a DEP (Dvl, Egl-10 and pleckstrin)- and PDZ domain-dependent manner. Active Daam1 and Dvl form a ternary complex with WGEF [weak-similarity GEF (guanine-nucleotide-exchange factor)], which in turn leads to the activation of the Rho/ROCK (Rhoassociated kinase) signalling and cytoskeleton remodelling [53,54]. Both the canonical and non-canonical Wnt pathways are implicated in tumorigenesis, with the former enhancing proliferation and in loco tumour formation and the latter contributing to invasion and cancer progression [55,56].
PDZ regulation of the tumour suppressor APC
APC plays a fundamental role in the Wnt pathway by promoting the degradation of β-catenin. Germline mutations in the APC gene are associated with familial adenomatous polyposis , characterized by hyperplasia of colon epithelial cells, leading to the formation of polyps, which invariably progress to carcinoma. APC has multiple protein interaction sites, one of which includes a C-terminal class 1 PBM [6,62]. The PBM-dependent interaction with Dlg appears to regulate the correct localization of APC, with important consequences for the regulation of cell-cycle progression and cell adhesion [63,64]. The APC PBM also binds to the PDZ domain of the FAP-1 (Fas-associated phosphatase 1) tyrosine phosphatase (see below), through which components of the APC–β-catenin destruction complex can be regulated , and overexpression of FAP-1 inhibits the proliferation of Wnt3a-stimulated cells in a phosphatase-dependent manner .
The APC PBM appears to be critical for some of its tumour-suppressor functions. Although many cancer-causing mutations occur in the N-terminal half of APC, truncating mutations that abolish only the PBM have been found in cancers [67,68]. Furthermore, there is compelling experimental evidence linking the PDZ-binding potential of APC with its ability to influence cell spreading and regulate cell substratum adhesion, both of which appear to require association with Dlg .
PDZ DOMAINS AND THE REGULATION OF EMT
Although Wnt signalling is a critical regulator of EMT, a number of other cell signalling pathways are also involved. As can be seen from Figure 3, many of the critical regulatory steps in these pathways are subject to regulation by PDZ–PBM combinations. For example, β-catenin has a PBM through which it can interact with the PDZ domain-containing proteins MAGI-1 and NHERF-1 (Na+/H+-exchanger regulatory factor 1). Likewise, the PTEN phosphatase, a known tumour suppressor , has a PBM which confers interaction with the PDZ domains of Par3, MAGI-1 and NHERF-1, which incidentally are also targets of the HPV E6 oncoproteins. Of these, Par3 and NHERF-1 have been closely linked to cancer development, and MAGI-1 somewhat less so.
Par3 belongs to the Par complex of proteins [Par3, Par6, Cdc42 (cell division cycle 42) and aPKC (atypical protein kinase C)] which it recruits to TJs in polarized epithelium in part via a homotypic PDZ domain interaction with Par6. This in turn recruits Cdc42 and aPKC . Spatial restriction of this complex is essential for correct apico-basal polarity, and mislocalization or overexpression of Par6 and aPKC are common to many cancers . Par3 also interacts with the Tiam1 (T-cell lymphoma invasion and metastasis 1) (see below) Rac-GEF ; this is essential for TJ assembly and contributes to the control of polarized cell migration and of EMT by restricting Rac activity. In contrast, a pro-oncogenic activity has been indicated for Par6, with overexpression being linked to increased proliferation and also the induction of TJ breakdown. Furthermore, TGFβ (transforming growth factor β)-induced phosphorylation of Par6 increases its interaction with the ubiquitin ligase Smurf1 (Smad ubiquitylation-regulatory factor 1), thereby resulting in RhoA degradation and the induction of a more mesenchymal phenotype . Thus the PDZ domain-containing components of the Par complex can function as promoters and inhibitors of EMT, depending on the balance of Par3/Par6 activities .
The function of the MAGI family of proteins is still poorly understood, but tumour-suppressor potential was indicated by their targeting by viral oncoproteins , and MAGI-1 expression is down-regulated in acute lymphoblastic leukaemia . The MAGI PDZ2 domain binds PTEN, protecting it from proteasome-mediated degradation , while MAGI PDZ domain 5 can associate with β-catenin , potentially creating a trimeric β-catenin–MAGI–PTEN complex . This membrane-bound PTEN down-regulates the PI3K signalling pathway , thus inhibiting several processes related to tumour formation and cancer progression, including cell growth, survival and migration [69,77].
Although PTEN is a tumour suppressor , the contribution of its PBM and PDZ domain-containing substrates to this activity are still unclear, and only one study has described a PTEN mutant with a defective PBM in human cancers . In addition, although MAGI-1 is a target of the HPV E6 oncoproteins, there is scant indication that HPV E6 directly perturbs PTEN function.
A further intriguing link between the regulation of Wnt signalling and PTEN and PDZ domain interactions is provided by NHERF-1. This protein has two PDZ domains, has been reported to be down-regulated during the development of colon cancer  and mutations have been reported in its PDZ domain in a number of breast cancers . A recent report also suggests a direct interaction of NHERF-1 with Frizzled receptors in a PBM–PDZ-dependent manner, resulting in down-regulation of β-catenin . It had also been shown that NHERF-1 could potentially associate with PTEN and thereby regulate receptor signalling and PI3K/Akt activity . Furthermore, NHERF-1 has recently been found to be a target of HPV-16 E6 [81a], an activity also linked to increased PI3K signalling and Akt activity. Thus, although the roles of the PDZ interactions of NHERF-1, particularly with PTEN, are the subject of some debate, it seems highly likely that some of these activities directly link growth factor receptor signalling to control of the Wnt pathway and again suggests a potential tumour-suppressor role in the context of some cancers.
PDZ DOMAIN-CONTAINING PTPs (PROTEIN TYROSINE PHOSPHATASES)
Three PDZ domain-containing PTPs, FAP-1, PTPH1 and PTP-MEG1 , have been described, with PTPH1 and FAP-1 closely linked to tumorigenesis, although their roles are still controversial.
Early studies suggested a possible tumour-suppressor potential for PTPH1 (PTPN3), since mutations were reported in some colorectal cancers : however, more recent studies are indicative of pro-oncogenic activity. K-Ras was shown to increase the expression levels of PTPH1 and p38γ , with PTPH1 and p38γ interacting in a PDZ–PBM-dependent manner. This resulted in dephosphorylation of phospho-p38γ and inhibition of the phospho-p38γ-mediated down-regulation of Ras signalling [84,85], generating a positive-feedback loop. PTPH1 was also found to be overexpressed in a number of breast cancers, resulting in a perturbation of vitamin D receptor localization and stimulation of cell proliferation . Thus, although PTPH1 has also been reported to be a PDZ target of the HPV E6 oncoprotein, suggesting tumour-suppressive potential in some contexts , in others a pro-oncogenic activity seems most likely.
FAP-1 (PTPN13) has also been reported to possess both oncogenic and tumour-suppressive activity, depending on the particular cellular context. Originally identified as a Fas-associated protein, FAP-1 was shown to suppress the Fas-induced apoptosis in a PDZ domain-dependent manner [88,89]. FAP-1 was implicated in the development of ESFT (Ewing's sarcoma family of tumours), being a direct transcriptional target for EWS-FLI1 (Ewing sarcoma breakpoint region 1/Friend leukaemia virus integration 1) , and being progressively up-regulated as tumours progressed towards malignancy. However, in other settings, FAP-1 has all the hallmarks of a tumour suppressor. As noted above, high levels of FAP-1 can down-regulate Wnt signalling, and mutations in FAP-1 have been reported in colorectal cancers . It can act to down-regulate Src kinase (Figure 3 and see below), an activity that is dependent on the PDZ–PBM interaction between the Src regulator RIL (reversion-induced LIM protein) and FAP-1, thereby recruiting FAP-1 to the Src–RIL complex to dephosphorylate Src [91,92]. RIL expression is also lost in a number of cancer-derived cells, including those from colon, liver and breast cancers , providing further evidence of an important role for the FAP-1/RIL module in tumorigenesis. Interestingly, FAP-1 appears to also act downstream of Src, through its PDZ2 domain-dependent interaction with the Src effector protein TRIP6 (thyroid receptor-interacting protein 6), which it dephosphorylates and inhibits downstream targeting of the focal adhesion-associated proteins Crk and p130cas [93,94]. Clinically, FAP-1 expression is a prognostic marker for survival in breast cancer , and expression is lost in many tumour types [96,97]. Furthermore, mutation within the PDZ2 domain has been reported in a number of different cancers, resulting in a loss of Fas- and TRIP6-binding activity . Taken together with previous studies showing that FAP-1 is a target of the HPV-16 E6 oncoprotein , this underlines further the role of FAP-1 as a PDZ-dependent tumour suppressor.
PDZ interactions of Src
Discussion of PDZ-containing PTPs would be incomplete without further mention of the PBM-containing Src tyrosine kinase , mutations in which have been associated with a number of cancers . Src is activated by dephosphorylation of the C-terminal Tyr530, followed by autophosphorylation at Tyr419. Dephosphorylation of Tyr530, which is close to the PBM, allows Src to bind a number of PDZ proteins, including the E3 ubiquitin ligase LNX1 (Ligand of Numb protein X1), which has been reported to be down-regulated in some cancers. Src phosphorylates LNX1, which in turn ubiquitylates Src, promoting its degradation , thus restricting the levels of activated Src. Src dephosphorylated at Thr530 also binds the PDZ domain-containing junctional adhesion protein AF-6 (ALL1-fused gene from chromosome 6) (see below), which connects the cytoskeleton to transmembrane proteins at cell junctions and also binds and regulates Ras and Rap1 GTPases. Binding AF-6 prevents Src autophosphorylation, holding Src in a semi-activated state at specific regions in the cell periphery, thus confining its access to a subset of potential substrates .
Mutation of the Src PBM in breast epithelial cells promotes Src activation and the development of invasive characteristics . These findings, combined with the fact that PBMs are found only in those Src proteins expressed in adherent cells, indicates the importance of the PDZ-mediated control of Src activation in epithelia. Although activating mutations in Src are rare, the Gln531* truncation, which lacks the PBM, is found in colon cancer (Sanger Catalogue of Somatic mutations in Cancer, COSMIC http://www.sanger.ac.uk/perl/genetics/CGP/cosmic?action=bygene&ln=SRC&start=1&end=5378coords=AA:AA), suggesting that loss of PDZ-mediated control of Src can contribute to cancer development, implicating LNX1 and AF-6 as being critical PDZ domain-containing regulators of Src.
PDZ DOMAINS AND CYTOSKELETAL REMODELLING
Cell motility is a fundamental feature of eukaryotic cells during embryogenesis and in the development and progression of cancer metastasis . Migration is tightly controlled by the organization of the actin cytoskeleton and arguably the most important regulators of this are the family of Rho GTPases . Rho proteins shuttle between an inactive GDP-bound and an active GTP-bound conformational state. This switch is controlled by different regulatory proteins: positive regulators of which are GEFs that stimulate the exchange of GDP for GTP, resulting in the active GTPase. More than 70 different Rho-GEFs have now been described and 37% of these possess PBMs at their C-termini , others have PDZ domains, and some have both PBMs and PDZ domains (Figure 4) implicating PDZ–PBM interactions as critical regulators of Rho-GEF function.
Schematic representation of some of the PDZ–PBM-dependent interactions of GEFs and their downstream consequences
One of the first examples of this in cancer development is Bcr (breakpoint cluster region), which is rearranged in human leukaemia and is fused with the tyrosine kinase Abl, forming the Bcr–Abl fusion gene. The Bcr–Abl chimaera has constitutively activated Abl kinase, which is thought to be responsible for most of its oncogenic properties . However, Bcr also has a C-terminal PBM  which allows complex formation with the AF-6 PDZ domain protein in epithelial cells . Normally, the ternary complex of Bcr, AF-6 and Ras at cell junctions down-regulates Ras-mediated signalling and cell proliferation; however, in the Bcr–Abl fusion gene product, this PDZ-binding capacity is lost, thereby contributing to increased Ras signalling and malignant potential .
The PDZ-containing RhoA-specific GEF LARG (leukaemia-associated Rho-GEF) is another example of mutational inactivation of a GEF during tumorigenesis. LARG is fused to the mixed-lineage leukaemia gene in primary acute myeloid leukaemia , resulting in loss of the LARG N-terminus and its PDZ domain, suggesting that, in this context, loss of PDZ-binding capacity might contribute to tumorigenesis. However, the LARG PDZ domain has been reported to interact with multiple protein partners, including the IGF-1 (insulin-like growth factor 1) receptor, where it is required for transmission of extracellular IGF-1 to the generation of cytoskeletal rearrangements by activation of Rho/ROCK . The PDZ domain also interacts with CD44, which it links to EGFR (epidermal growth factor receptor) signalling and Rho/Ras activation in the development of head and neck squamous cell carcinomas . Furthermore, LARG is overexpressed in Shwachman–Diamond syndrome patients, who have a high propensity to develop acute myeloid leukaemia . However, in breast and colorectal cancers, LARG has tumour-suppressor characteristics, with frequent gene deletion and underexpression of the protein in many of the tumours . Thus it seems likely that LARG may have both tumour-suppressor and oncogenic potential, depending on the precise cellular context.
Another PDZ domain-containing GEF implicated in tumour development is the Rac-GEF P-Rex1, a critical regulator of ErbB2 signalling in breast cancer [113,114] and possibly having a role in prostate cancer metastasis . In both cases, P-Rex1 is overexpressed, resulting in increased activation of Rac signalling and cell invasion. How the PDZ domain contributes to this activity is not clear, as little is known about its PBM-containing interacting partners. However, S1P (sphingosine 1-phosphate) receptors are one such target. S1P is also involved in regulating cell migration and it is possible that the interaction between P-Rex1 and S1P1 (S1P receptor 1) can regulate the correct localization of S1P1 at the cell membrane, with an increase in P-Rex1 increasing S1P1 membrane localization, thereby contributing to increased cell migratory potential . Regardless of the precise mechanism of action, overexpression of this particular PDZ domain-containing GEF is strongly associated with cancer development in a variety of different contexts.
Tiam1 is a ubiquitously expressed Rac-specific GEF and contains a single PDZ domain and potential class I PBM, and thus can connect directly to other PDZ domain-containing proteins. Although Tiam1 itself has not been reported to be lost/overexpressed during tumorigenesis, a reported PBM-containing partner CADM1 (cell adhesion molecule 1)/TSLC1 (tumour suppressor in lung cancer 1) has. This glycoprotein belongs to the Ig-like family of cell adhesion molecules, loss of which is common to many different tumour types [117,118]. Although the role of PDZ binding has not been shown in tumour development, the CADM1 PBM was found to be essential for correct cellular localization and for suppression of EMT in vitro. Intriguingly, this correlated with high levels of Rac activity , suggesting that the interaction between Tiam1 and CADM1 may contribute to the regulation of this pathway. However, whereas CADM1 may be a tumour suppressor in most contexts, in ATL (adult T-cell leukaemia/lymphoma), CADM1 binding to Tiam1 and induction of Rac activation and actin reorganization is believed to contribute to the cell's invasive characteristics , once again underlining the context dependence by which PDZ–PBM interactions need to be evaluated. As noted above, Tiam1 interacts with Par3 , and this interaction also restricts Tiam1 Rac activity. Since Tiam1 contains a class I PBM, one would have expected that the association would have been via this motif; however, this does not seem to be the case .
A key feature of many GEFs is their ability to control the Rho-GTPase activity both spatially and temporally, and this often occurs through its interaction via PBM–PDZ recognition. β-Pix is a PBM-containing GEF which plays a critical role in the regulation of polarized cell migration through activation of Rac and Cdc42. However, the correct localization of β-Pix is determined by Scribble . As noted above, Scribble is a target of HPV E6 and, as with many multi-PDZ domain-containing proteins, Scribble can be considered to be multifunctional. In the case of β-Pix, the interaction with Scribble is PDZ domain-dependent and is essential for the correct cellular localization of β-Pix and localized activation of Cdc42 and Rac at the leading edge of migrating cells. In contrast, loss of Scribble does not destroy a cell's migratory capacity, but directionality is lost . What does this mean from a tumour perspective? Again, as noted above, there are major changes in the levels and patterns of Scribble expression during the development of many cancers, and these in turn could be expected to perturb the normal positioning of the PBM-containing partners such as β-Pix.
Similar functions have also been reported for Dlg, which also interacts with several PBM-containing GEFs, such as Net1, where Dlg has been reported to regulate Net1 localization and activity. Conversely, oncogenic mutants of Net1 are believed to function through inhibition of Dlg function . Alterations in Dlg expression during cancer development are common, with high levels of expression, and loss of junctional localization in early-stage cancer, and complete loss of expression in late-stage metastases . Taken together, these results suggest that mislocalization of the PDZ domain-containing protein is a key factor in oncogenic progression. It also raises the possibility that, at certain stages of cancer development, the overexpression of such proteins might be oncogenic, something that has been implied for Dlg . Thus, again, in context-dependent settings, these PDZ domain-containing proteins, by interacting with a variety of GEFs, may possess oncogenic or tumour-suppressive activity.
In pancreatic cancers, the single PDZ domain-containing protein GIPC1 [GAIP (Gα-interacting protein)-interacting protein C-terminus 1]/synectin is often overexpressed . Furthermore, GIPC1 has also been reported to be a PDZ domain-containing target of the HPV E6 oncoprotein, suggesting further a role in tumour development . In a mouse model of pancreatic tumour invasion, it was also shown that blocking synectin's PDZ-binding potential inhibits tumour growth , providing compelling evidence that the PDZ–PBM interactions of synectin are important in the development of certain malignancies. Although many PBM-containing targets of synectin have been identified, including the IGF-1 receptor , which is a strong candidate as a clinically relevant PBM-containing target of synectin , a particularly intriguing PBM-containing partner is MyoGEF (myosin-interacting GEF). This GEF has been shown in vitro to be capable of regulating breast cancer cell polarization and invasion through its ability to activate RhoA and RhoC signalling . Therefore the temporal and spatial control provided by synectin in regulating RhoA and RhoC signalling through controlling the localization of MyoGEF is an additional possible route by which overexpressed synectin might contribute to tumorigenesis.
PDZ domain-containing proteins play essential roles in most aspects of cellular homoeostasis and, not surprisingly, are implicated in diverse aspects of tumour growth, development and metastasis. These can have both tumour-suppressive and oncogenic potentials, and the underlying characteristics of many of these proteins are their particular tissue-specific activities. The relevant PDZ–PBM interactions often differ from one tissue to another, and, in many cases, contradictory effects of either loss or overexpression of a given PDZ domain-containing protein or its PBM-containing partner have been reported; invariably this is due to differences in the particular experimental model or tissues that are being analysed. Thus the specificity of PDZ–PBM interactions is highly dependent on cellular context . This underlines their essential roles as nodes in multiple signal transduction pathways, critical aspects of which are reflected in very minor changes in the levels of protein expression or in its precise cellular location. Although acquiring a global understanding of their role in cancer development is still a long way off, the amount of information now available about the details and regulation of PBM–PDZ interactions offers unique potentials for the development of novel cancer therapeutics, tailored to specifically target a given PBM–PDZ interaction. The coming years will undoubtedly see great strides in the targeting of these PBM–PDZ nodes in the treatment of cancer.
ALL1-fused gene from chromosome 6
adenomatous polyposis coli
atypical protein kinase C
breakpoint cluster region
cell adhesion molecule 1
cell division cycle 42
Dishevelled-associated activator of morphogenesis 1
Fas-associated phosphatase 1
GAIP (Gα-interacting protein)-interacting protein C-terminus 1
human T-cell leukaemia virus type 1
insulin-like growth factor 1
Ligand of Numb protein X1
membrane-associated guanylate kinase with inverted domain structure
Na+/H+-exchanger regulatory factor 1
protein kinase A
postsynaptic density protein 95
phosphatase and tensin homologue deleted on chromosome 10
protein tyrosine phosphatase
reversion-induced LIM protein
S1P receptor 1
Smad ubiquitylation-regulatory factor 1
T-cell lymphoma invasion and metastasis 1
thyroid receptor-interacting protein 6
zonula occludens protein
L.B. gratefully acknowledges research support provided by the Associazione Italiana per la Ricerca sul Cancro, the Association for International Cancer Research, Telethon and the Wellcome Trust.
These authors contributed equally to this review.