MAGI-1, a Membrane-associated Guanylate Kinase with a Unique Arrangement of Protein-Protein Interaction Domains*

Membrane-associated guanylate kinase (MAGUK) proteins participate in the assembly of multiprotein complexes on the inner surface of the plasma membrane at regions of cell-cell contact. MAGUKs are characterized by three types of protein-protein interaction modules: the PDZ domain, the Src homology 3 (SH3) domain, and the guanylate kinase (GuK) domain. The arrangement of these domains is conserved in all previously known MAGUKs: either one or three PDZ domains in the NH2-terminal half, followed by the SH3 domain, followed by a COOH-terminal GuK domain. In this report, we describe the cDNA cloning and subcellular distribution of MAGI-1, a MAGUK with three unique structural features: 1) the GuK domain is at the NH2 terminus, 2) the SH3 domain is replaced by two WW domains, and 3) it contains five PDZ domains. MAGI-1 mRNA was detected in several adult mouse tissues. Sequence analysis of overlapping cDNAs revealed the existence of three splice variants that are predicted to encode MAGI-1 proteins with different COOH termini. The longest variant, MAGI-1c, contains three bipartite nuclear localization signals in its unique COOH-terminal sequence and was found predominantly in the nucleus of Madin-Darby canine kidney cells. A shorter form lacking these signals was found primarily in membrane and cytoplasmic fractions. This distribution, which is reminiscent of that seen for the tight junction protein ZO-1, suggests that MAGI-1 may participate in the transmission of regulatory signals from the cell surface to the nucleus.

The prototypical members of the membrane-associated guanylate kinase (MAGUK) 1 family are the Drosophila tumor suppressor protein DLG, the erythrocyte membrane protein p55, and the neuronal protein PSD-95/SAP90 (1,2). These proteins share a common modular structure that consists of either one or three PDZ domains, a single Src homology 3 (SH3) domain, and a single region of homology to Saccharomyces cerevisiae guanylate kinase (GuK), known as the GuK domain. The MAGUK family includes the epithelial tight junction proteins ZO-1 and ZO-2, the Caenorhabditis elegans vulval protein LIN-2A, and the neurexin-binding protein CASK (3)(4)(5). The latter two represent a subfamily of MAGUKs characterized by an additional domain at the NH 2 terminus that is similar to calmodulin kinase II. All MAGUKs studied to date localize to regions of cell-cell contact, such as tight junctions in epithelial cells and synaptic junctions in neurons, where they nucleate the assembly of multiprotein complexes via their protein-protein interaction domains (6,7). In addition, ZO-1 was found in the nucleus of cultured cells under certain growth conditions (8).
PSD-95/SAP90 is the prototype of a subfamily of neuronal MAGUKs that includes SAP97/hDLG, chapsyn-110/PSD-93, and SAP102 (9 -15). The first and second PDZ domains of PSD-95 proteins bind to the cytoplasmic COOH termini of the Shaker-type K ϩ channel and N-methyl-D-aspartate receptor 2 subunits, resulting in clustering of these molecules on the neuronal surface (11, 16 -18). DLG, the Drosophila homologue of the mammalian PSD-95 family, colocalizes with Shaker K ϩ channels in larval neuromuscular junctions, and these channels fail to cluster at synapses in Drosophila larvae that harbor mutant DLG alleles (19). In heterologous cells, the channel clustering activity of PSD-95 requires only the first PDZ domain and a region at the NH 2 terminus that is conserved among the other PSD-95 family members, suggesting that the second and third PDZ domains may recruit additional membrane or cytoplasmic proteins (20).
The name PDZ domain is derived from three members of the MAGUK family: PSD-95, DLG, and ZO-1 (21). However, PDZ domains are not restricted to MAGUKs. More than 50 proteins are known to contain PDZ domains; many of these lack the GuK and SH3 domains, and several contain domains that confer protein kinase, protein phosphatase, or other enzymatic activities (3,7). PDZ domains are repeats of 80 -100 amino acids in length that engage in either homotypic interactions (PDZ-PDZ dimers) or, more commonly, heterotypic interactions in which the ligand is the COOH terminus of a transmembrane or a cytoplasmic protein. The single PDZ domain in neuronal nitric-oxide synthase binds to PDZ domains in PSD-95, PSD-93, and ␣1-syntrophin (10,22); these are currently the only known examples of homotypic interactions mediated by PDZ domains. In contrast, several heterotypic ligands for PDZ domains have been identified. In addition to the PSD-95 ligands described above, these interactions include the following: 1) the COOH terminus of the FAS receptor binds to the third PDZ domain of the tyrosine phosphatase FAP-1 (23); 2) the COOH terminus of the APC tumor suppressor protein binds to the second PDZ domain of SAP97/hDLG (24); 3) the COOH terminus of the testis determining factor SRY binds to PDZ domains in the nuclear protein SIP-1 (25); and 4) the transient receptor potential protein, eye protein kinase C, and phospholipase C-␤ bind to the third, fourth, and fifth PDZ domains, respectively, of the Drosophila photoreceptor protein InaD (26). The latter finding supports the notion that multivalent PDZ domain proteins enable the formation of multiprotein complexes via interaction of their individual PDZ domains with distinct targets (3).
The crystal structures of the third PDZ domain of SAP97/ hDLG and of the third PDZ domain of PSD-95 bound to a synthetic peptide ligand were recently determined (27,28). This information, together with in vitro biochemical studies (29), has begun to reveal the mechanism by which specificity is achieved in heterotypic PDZ domain interactions. One group of PDZ domains, including those in the PSD-95 family, bind COOH-terminal sequences that conform to the consensus (S/ T)XV (where X is any amino acid), such as those found in the Shaker K ϩ channel and N-methyl-D-aspartate receptor subunits. The hydroxyl group of the serine or threonine at the Ϫ2-position in this consensus forms a hydrogen bond with a histidine (His 372 in PSD-95) that is conserved in most PDZ domains (28). The single PDZ domains in p55 and LIN-2A contain a valine in place of the semiconserved histidine, and these domains prefer ligands in vitro that contain aromatic residues in the Ϫ2-position (29). The PDZ domain of neuronal nitric-oxide synthase represents a third variant in which the histidine is replaced by a tyrosine. Besides forming homotypic complexes with other PDZ domains, the neuronal nitric-oxide synthase PDZ domain binds to peptides in vitro that terminate in the sequence DXV (30). Thus, a single residue in each PDZ domain (corresponding to His 372 in PSD-95) may dictate which amino acid is found in the Ϫ2-position of its cognate ligand. Several other structural features of PDZ domains are predicted to contribute in a more subtle manner to the selection of a particular COOH-terminal tripeptide sequence as a ligand (27)(28)(29). Moreover, interactions with amino acids that are adjacent to the tripeptide consensus probably impart additional specificity.
GuK domains in MAGUKs were named because of their similarity to S. cerevisiae GuK, which uses ATP to phosphorylate GMP, producing guanosine diphosphate and adenosine diphosphate (31). Three classes of GuK domains have been found in MAGUKs: those that lack both the GMP-and ATPbinding residues (ZO-1 and ZO-2), those in which the GMPbinding residues are conserved but the ATP-binding motif is absent (PSD-95 family), and those that conserve both the GMPand ATP-binding residues (p55, LIN-2A, and CASK) (32). However, none of the GuK domains in MAGUK proteins are known to be catalytically active. A clue to the function of the GuK domain in MAGUK proteins emerged recently when two groups identified a synaptic protein, designated GKAP by one group and SAPAP by the other, which binds to the GuK domains of the PSD-95 proteins (33,34). GKAP was found in a complex with PSD-95 and Shaker K ϩ channels, suggesting that it may serve to anchor this complex in the postsynaptic density. Although it is not yet known if the interaction between GKAP and PSD-95 is regulated by guanine nucleotide binding, this finding suggests that the GuK domains in other MAGUK proteins may also function as sites for protein-protein interaction.
A protein-protein interaction module that has not previously been found in MAGUK proteins is the WW domain (35,36). The WW domain was named for two conserved tryptophan residues and was first identified as two repeats in mouse YAP65 (Yesassociated protein of 65 kDa) (37). WW domains have been identified by sequence similarity in more than 30 proteins from various species, and interactions mediated by WW domains have been implicated in human diseases such as Liddle's syndrome, muscular dystrophy, and Alzheimer's disease (38). WBP1 and WBP2 are candidate ligands for the single WW domain in human YAP65 (39). These proteins bind to the YAP65 WW domain via polyproline sequences that conform to the consensus sequence PPXY, designated the PY motif. However, the WW domains in formin-binding proteins bind to a proline-rich sequence in formins that does not match the PY motif, indicating that not all WW domains bind to this consensus (40). The same study also demonstrated overlap between the WW domain binding site and a SH3 domain binding site, raising the possibility that these two domains may sometimes compete with one another for the same ligand. Phosphorylation of the tyrosine in a synthetic PY motif abolished its interaction with the YAP65 WW domain, suggesting a potential regulatory mechanism for some WW domain interactions (41).
In this report, we describe a MAGUK with three features that distinguish it from all other known members of the family: 1) the GuK domain is at the NH 2 terminus rather than at the COOH terminus; 2) the SH3 domain is replaced by two WW domains; and 3) it contains five PDZ domains rather than the usual one or three. This protein has been designated MAGI-1, for membrane-associated guanylate kinase with an inverted arrangement of protein-protein interaction domains.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid System-The yeast strain L40, the plasmids pBTM116 and pVP16, and the mouse embryo cDNA library in pVP16 (42) were obtained from Dr. Michael White of the University of Texas Southwestern Medical Center at Dallas. Bait constructs were prepared by polymerase chain reaction amplification of the desired fragment containing 5Ј-BamHI and 3Ј-PstI restriction sites. The product of each reaction was subcloned into the corresponding sites in pBTM116, and each plasmid was verified by DNA sequencing. For ␤-galactosidase assays, a 2-ml culture of each yeast strain was grown for 16 h in ϪTrp/ϪLeu medium and then diluted 10-fold with the same medium and grown for an additional 3 h. Cells were harvested by centrifugation, resuspended in buffer Z (60 mM Na 2 HPO 4 , 40 mM NaH 2 PO 4 , 10 mM KCl, 1 mM MgSO 4 , pH 7.0), and lysed by two freeze/thaw cycles, followed by vortexing in the presence of glass beads. The lysate was centrifuged for 10 min at 1.5 ϫ 10 4 g, the supernatant was transferred to a new tube, and the protein concentration of each supernatant was determined with the BCA protein assay reagent (Pierce). Five g of protein from each lysate was adjusted to a total volume of 30 l with buffer Z and mixed with 200 l of reaction buffer from the Luminescent ␤-Galactosidase Detection Kit II (CLONTECH). After a 5-min incubation at room temperature, at which time each reaction was progressing in a linear manner, the relative light units emitted by each reaction were measured in a luminometer. Blank values were obtained by measuring the ␤-galactosidase activity in 5 g of protein from a lysate of nontransformed yeast.
cDNA Cloning-The 478-bp cDNA insert was excised from pVP16-K65 by digesting with NotI and radiolabeled with [␣-32 P]dCTP using the PrimeIt II random primer labeling kit (Stratagene). This probe was used to screen an oligo dT/random-primed mouse lung cDNA library in Lambda ZAP II (Stratagene), according to the manufacturer's instructions. Eleven MAGI-1 cDNAs were isolated from a total of ϳ5 ϫ 10 5 plaque-forming units. Additional MAGI-1 cDNAs were isolated in a similar manner from an oligo dT-primed mouse brain library. The size of each cDNA insert was determined by restriction mapping, and selected clones were analyzed by DNA sequencing using T3, T7, and specific internal primers. The entire 5,296-bp MAGI-1b cDNA was sequenced on both strands.
Blot Hybridization of RNA-A 678-bp XhoI restriction fragment, spanning nucleotides 759 -1437 of MAGI-1, was radiolabeled with [␣-32 P]dCTP and used to probe a mouse multiple tissue Northern blot (CLONTECH). The probe was used at a concentration of 2 ϫ 10 6 cpm/ml in Rapid-hyb buffer (Amersham Corp.), according to the manufacturer's instructions. After exposure to Kodak X-Omat AR film for the indicated time, the blot was stripped and rehybridized with a glyceraldehyde-3phosphate dehydrogenase (GAPDH) probe.
Antibodies-A 417-bp fragment that encodes amino acids 2-140 of MAGI-1 was amplified by polymerase chain reaction and subcloned into the vectors pRSETA (Invitrogen) and pGEX-2TK (Pharmacia). The resulting plasmids, designated pRSETA-MAGI 2-140 and pGEX-2TK-MAGI 2-140 , were transformed into BL21(DE3) bacteria (Novagen). Recombinant His 6 -MAGI 2-140 protein was expressed and purified by Ni 2ϩ -Sepharose affinity chromatography and used to immunize rabbits as described previously (43,44). GST-MAGI 2-140 was expressed and purified by glutathione-agarose affinity chromatography, and the glutathione S-transferase was cleaved with thrombin as described previously (43), yielding MAGI 2-140 . An affinity column was prepared by crosslinking 2 mg of MAGI 2-140 to AminoLink Coupling Gel (Pierce), and anti-MAGI-1 polyclonal IgG was affinity-purified as described previously (43). Preimmune IgG was isolated from crude serum obtained from the same rabbit by purification on protein A-agarose. A polyclonal antibody directed against lamin B was prepared by immunizing rabbits with a synthetic peptide corresponding to amino acids 541-554 of mouse lamin B that had been coupled to keyhole limpet hemocyanin, and anti-lamin B IgG was isolated from crude serum by purification on protein A-agarose. All polyclonal antibodies were used at a concentration of 2 g/ml; the anti-FT␣ monoclonal antibody, described previously (45), was used at a concentration of 5 g/ml.
Mammalian Expression Vectors, DNA Transfection, and Immunoblot Analysis-A 4,680-bp SmaI-EcoRI fragment of the MAGI-1b cDNA, containing the complete open reading frame, was subcloned into the mammalian expression vector pcDNA3 (Invitrogen), and the resulting plasmid was designated pcDNA3-MAGI-1b. Digestion of this plasmid with ApaI removed a 2,050-bp fragment from the 3Ј-end of the insert, which was replaced by the corresponding region from the MAGI-1c cDNA, yielding pcDNA3-MAGI-1c. Monolayers of human embryonic kidney 293 cells were transfected with 1 g of plasmid DNA/100-mm dish as described previously (43). After a 16-h incubation, whole-cell lysates were prepared in buffer A (0.05 M Tris-HCl, 0.1 M NaCl, 1 mM EDTA, 2 M urea, 1% (w/v) SDS, pH 7.5), and the protein concentration of each lysate was determined as described above. An equal amount of protein from each lysate was subjected to SDS-polyacrylamide gel electrophoresis and subjected to immunoblot analysis with the indicated antibody as described previously (46).
Cell Culture and Subcellular Fractionation-MDCK cells were maintained in Dulbecco's modified Eagle's medium containing 100 units/ml penicillin, 100 g/ml streptomycin, and 10% (v/v) fetal bovine serum. Cells were plated on day 0 at a density of 5 ϫ 10 5 cells/100-mm dish and harvested on day 3 by scraping in phosphate-buffered saline. Cell pellets were resuspended in ice-cold buffer B (10 mM Tris-HCl, 0.1 M sucrose, 1 mM EDTA, 0.5 mM phenymethylsulfonyl fluoride, 1% (v/v) aprotinin, 10 g/ml leupeptin, 10 g/ml pepstatin, pH 7.5) and lysed by 10 strokes in a Dounce homogenizer on ice. The homogenate was centrifuged at 4°C for 10 min at 10 3 ϫ g, and the supernatant was transferred to a fresh tube and designated the postnuclear supernatant. The 10 3 ϫ g pellet was resuspended in buffer B, subjected to an additional 10 strokes in a Dounce homogenizer, and centrifuged as described above. The supernatant from this step was combined with the postnuclear supernatant and then centrifuged at 4°C for 1 h at 10 5 ϫ g to yield 10 5 ϫ g supernatant (S100) and pellet (P100) fractions. The 10 3 ϫ g pellet was resuspended in buffer B, layered onto a cushion of buffer C (0.05 M Tris-HCl, 0.1 M NaCl, 1 M sucrose, pH 7.5), and centrifuged for 30 min at 10 4 ϫ g. The supernatant from this step was discarded, and the pellet was resuspended in buffer B and designated the nuclear fraction. The protein concentration of each fraction was determined, and an equal percentage of each fraction was subjected to SDS-polyacrylamide gel electrophoresis followed by immunoblot analysis with the indicated antibodies.

RESULTS
Localization of K-RasB to the plasma membrane requires a COOH-terminal farnesylated cysteine plus a polybasic domain, a stretch of six contiguous lysines (amino acids 175-180) that are separated from the farnesylated cysteine by four amino acids (47). The molecular mechanism by which these structural features target K-RasB to the plasma membrane is not known. We used the yeast two-hybrid system to search for proteins that interact with the COOH terminus of K-RasB. Preliminary studies revealed that full-length K-RasB failed to interact with Raf in the yeast two-hybrid system, whereas a mutant that was truncated after cysteine 185, thus preventing its farnesylation, interacted efficiently with Raf (data not shown). Farnesylated K-RasB is apparently unable to be transported into the nucleus, where it must reside to function properly in the yeast two-hybrid system. Thus, the bait used in subsequent cDNA library screens, designated K-RasB 165-185 , contained a segment from the COOH terminus of K-RasB that was truncated after cysteine 185.
A 478-bp cDNA, designated K65, was isolated from a mouse embryo library. K65 interacted with K-RasB 165-185 and with an essentially full-length bait, K-RasB 1-185 . K65, like Raf, failed to interact with a farnesylated K-RasB bait (data not shown). However, K65 also failed to interact with nonfarnesylated versions of H-Ras and Rac1, a Ras-related protein that also has a polybasic domain, suggesting that its interaction with K-RasB 1-185 was specific (data not shown). A comparison of the deduced amino acid sequence of K65 with sequences in Gen-Bank TM revealed a region of ϳ100 amino acids that represented a novel PDZ domain, which was designated MAGI 1 (Fig.  1). MAGI 1 retained the ability to interact with K-RasB 1-185 when removed from the flanking sequences present in K65, suggesting that K-RasB was behaving as a heterotypic ligand for this novel PDZ domain (Table I, line a).
To investigate the structural features of K-RasB 1-185 responsible for its interaction with MAGI 1 , various mutant K-RasB 1-185 baits were prepared, and the ␤-galactosidase activity resulting from their interaction with MAGI 1 was measured in a quantitative assay using a chemiluminescent substrate (Table I). Deletion of the COOH-terminal cysteine in K-RasB 1-185 caused a 45-fold decrease in ␤-galactosidase activity (Table I, line b). The activity decreased by similar magnitudes when the COOH-terminal cysteine was replaced by serine, methionine, or alanine (Table I,   Many heterotypic PDZ domain ligands contain a threonine or serine three amino acids from the COOH terminus (referred to as the Ϫ2-position), which forms a hydrogen bond with a histidine in their cognate PDZ domain (28), corresponding to histidine 372 in the third PDZ domain of PSD-95 (PSD-95 3 , Fig. 1). ␤-Galactosidase activity decreased 140-fold when threonine 183 in K-RasB 1-185 was replaced with alanine (Table I, (Table I, line a). Most known PDZ domains prefer ligands in which the COOH-terminal amino acid is a valine (29). The ␤-galactosidase activity increased by 22-fold when cysteine 185 in K-RasB 1-185 was replaced by valine (Table I, line l).
Insertion of three alanines between lysine 180 and serine 181 in K-RasB 1-185 caused a 34-fold decrease in ␤-galactosidase activity, suggesting that one or more lysines in the polybasic domain might participate in the interaction with MAGI 1 ( Table  I, line g). Insertion of three alanines resulted in a similar reduction in activity when cysteine 185 was replaced by valine (Table I; compare lines l and m), indicating that the spacing between the polybasic domain and the COOH terminus was important even in the context of a COOH-terminal tripeptide that conformed to the (S/T)XV consensus preferred by most PDZ domains. Substitution of lysines 175 and 176 in the polybasic domain with glutamine caused a 4-fold reduction in ␤galactosidase activity (Table I,  Raf, indicating that those that failed to interact with MAGI 1 were capable of functioning in the yeast two-hybrid system (data not shown).
An alignment of the MAGI 1 sequence with the sequence of PSD-95 3 revealed the presence of four acidic amino acids in a segment of MAGI 1 that corresponds to the loop between the second and third ␤-strands in PSD-95 3 (Fig. 1). These residues are predicted to lie in close proximity to amino acids in the Ϫ5 and Ϫ6-positions of a bound ligand (28). Substitution of these four acidic residues with neutral amino acids (Fig. 1, Mut. 1) abolished the interaction of MAGI 1 with K-RasB 1-185 (Table I,  The 478-bp K65 cDNA was used as a probe to isolate several longer cDNAs from mouse brain and lung libraries. The longest cDNA isolated to date is 5,296 bp in length, consisting of a 924-bp 5Ј-untranslated region and a 3,513-bp open reading frame followed by a termination codon and a 3Ј-untranslated region of 856 bp (Fig. 2A). The putative initiating methionine lies within a favorable context for translation initiation according to Kozak's consensus and is preceded by termination codons in all three reading frames. This cDNA was isolated from a random-and oligo dT-primed library and appears to have been the product of a random priming event, since it lacks a poly(A) tail. An overlapping cDNA was isolated, which extended the 3Ј-untranslated region by 75 bp and ended in a poly(A) tail ( Fig.  2A). The combined length of the two overlapping cDNAs is 5,371 bp, which is consistent with a transcript of ϳ5.4 kb that was detected on a Northern blot (see Fig. 4). This sequence is predicted to encode a protein of 1,171 amino acids, designated MAGI-1b, with a molecular mass of 128 kDa.
A comparison of the predicted MAGI-1b amino acid sequence with sequences in the GenBank TM data base revealed the presence of a GuK domain near the NH 2 terminus, followed by two WW domains and five PDZ domains (Fig. 2A). The partial MAGI-1 cDNA isolated from the yeast two-hybrid library corresponded to nucleotides 2,080 -2,558 of MAGI-1b, spanning the first PDZ domain. As shown schematically in Fig. 3, the domain structure of MAGI-1 is unique when compared with

in the yeast two-hybrid system
The last 10 -14 amino acids of various K-RasB 1-185 baits are shown in the left column (Bait), and the relative ␤-galactosidase activities resulting from their interaction with wild-type (WT) and mutant (Mut. 1 and Mut. 2) MAGI 1 are depicted in the columns on the right. The amino acid substitutions in mutants 1 and 2 are shown in Fig. 1. ␤-Galactosidase activities, measured as described under "Experimental Procedures," are expressed as a percentage of the activity resulting from the interaction of K-RasB 1-185 with wild-type MAGI 1 (711,300 relative light units, line a), which was assigned a value of 100%. Amino acid substitutions and insertions in K-RasB 1-  other members of the MAGUK family. Amino acids 152-348 of MAGI-1b, a segment that begins within the GuK domain and ends just before the second WW domain, are 93% identical to the predicted amino acid sequence of a partial cDNA that was isolated from a human cDNA expression library via its interaction with synthetic peptides that contained PY motifs (48).
Two additional classes of MAGI-1 cDNAs were isolated whose sequences diverged from MAGI-1b downstream of an apparent mRNA splice site at nucleotide 4,293 (triangle in Fig.  2, A-C), just beyond the fifth PDZ domain. These cDNAs are predicted to encode proteins of 1,139 amino acids (124 kDa; Fig.  2B) and 1,374 amino acids (152 kDa; Fig. 2C) and were designated MAGI-1a and MAGI-1c, respectively. The longest MAGI-1a cDNA isolated to date contains 3,770-bp and was found in a brain library. The sequence of this cDNA begins within the GuK domain and ends with a poly(A) tail. If its 5Ј-end is assumed to be identical to MAGI-1b, it would be 5,212 bp in length, which may correspond to a transcript of ϳ5 kb detected in brain mRNA on a Northern blot (see Fig. 4). The longest MAGI-1c cDNA is 3,038 bp in length; its sequence begins between the second and third PDZ domains and ends in the 3Ј-untranslated region. Since it lacks a poly(A) tail, it is not yet possible to predict the total length of the transcript from which it originated, but it may correspond to either the ϳ6.9or the ϳ7.8-kb transcript detected on a Northern blot (see Fig. 4). The COOH-terminal 145 amino acids of MAGI-1c consist of 37% lysine plus arginine and 19% aspartate plus glutamate. This region contains three bipartite nuclear localization signals (49), the second and third of which are overlapping (boxes in Fig. 2C).
The tissue distribution of MAGI-1 mRNA was examined by Northern blot analysis of mRNA from various mouse tissues. A probe derived from near the 5Ј-end of the MAGI-1b cDNA hybridized to transcripts of ϳ5.4, ϳ6.9, and ϳ7.8 kb in mRNA from kidney, liver, and lung (Fig. 4, top, lanes 2, 4, and 5). The 5.4-and 7.8-kb transcripts were also detected in testis and heart (lanes 1 and 8); prolonged exposures of the blot to film revealed the 5.4-and 6.9-kb transcripts in skeletal muscle (lane 3) and transcripts of ϳ5.0 and ϳ5.6 kb in brain (lane 7), indicating that expression of MAGI-1 in adult tissues is widespread.
Mammalian expression vectors were prepared that contained the complete open reading frames of MAGI-1b and MAGI-1c, and these plasmids were transiently transfected into human embryonic kidney 293 cells. Lysates of these cells were analyzed by immunoblotting with anti-MAGI-1, a polyclonal antibody directed against amino acids 2-140 of MAGI-1. This antibody detected proteins of ϳ122 and ϳ142 kDa in cells transfected with MAGI-1b and MAGI-1c, respectively (Fig. 5A,  top panel, lanes 2 and 4). These proteins were not detected in lysates of cells that were transfected with expression vector alone (lane 1), nor were they detected with preimmune IgG (Fig. 5A, bottom panel). Proteins of slightly slower mobilities (ϳ137 and ϳ148 kDa) were detected in a whole-cell lysate of MDCK cells with anti-MAGI-1 IgG (top panel, lane 3) but not with preimmune IgG (bottom panel). Proteins with molecular masses identical to those seen in MDCK cells were also detected in lysates of mouse lung and liver (Fig. 5B, top panel,  lanes 7 and 8), suggesting that they represent posttranslationally modified versions of MAGI-1b and MAGI-1c, respectively, rather than mRNA splice variants unique to MDCK cells. Although transfected MAGI-1a has not yet been analyzed on SDS gels, its mobility is expected to be similar to that of MAGI-1b, since their predicted molecular masses differ by only 3.5 kDa. Thus, the 137-kDa protein will be tentatively referred to as MAGI-1a/b and the 148-kDa protein as MAGI-1c. MAGI-1a/b was the predominant version expressed in mouse brain, whereas liver expressed primarily MAGI-1c (Fig. 5B, compare  lanes 6 and 8); the two forms were present in approximately equal levels in mouse lung (lane 7).  2 and 4,  respectively). Fifty g of protein from the MDCK lysate and 10 g from each 293 lysate were resolved by SDS-polyacrylamide gel electrophoresis on replicate 8% gels and analyzed by immunoblotting with affinitypurified anti-MAGI-1 polyclonal IgG (top) or with preimmune IgG (bottom). The migration of prestained molecular weight standards is indicated on the right. The blots were exposed to film for 10 min. B, 50 g of protein from MDCK whole-cell lysate (lanes 5 and 9) and 150 g of protein from mouse brain, lung, and liver lysates (lanes 6 -8) were analyzed by immunoblotting with either anti-MAGI-1 polyclonal IgG (top) or with preimmune IgG (bottom) as described for panel A. The blots were exposed to film for 40 min. C, 100,000 ϫ g supernatant (S100, lane 11), 10 5 ϫ g pellet (P100, lane 12), and nuclear (Nuclei, lane 13) fractions were prepared from MDCK cells as described under "Experimental Procedures." An equal percentage of protein from each fraction (50 g of S100, 75 g of P100, 50 g of nuclei) plus 10 g of protein from whole-cell lysates of 293 cells that had been transiently transfected with MAGI-1b (lane 10) or MAGI-1c (lane 14) were analyzed as described for panel A with the following antibodies: anti-MAGI-1 polyclonal IgG (top), anti-lamin B polyclonal IgG (middle), and anti-FT␣ monoclonal IgG (bottom). The migration of prestained molecular weight standards is indicated on the right. The anti-MAGI-1 and anti-FT␣ blots were exposed to film for 10 min; the anti-lamin B blot was exposed for 3 min.
To determine the subcellular distribution of MAGI-1 proteins in MDCK cells, S100, P100, and nuclear fractions were prepared, and an equal percentage of each fraction was subjected to immunoblot analysis with anti-MAGI-1 IgG (Fig. 5C,  top). MAGI-1a/b was found primarily in the P100 and S100 fractions (lanes 12 and 11, respectively), with a small amount in the nuclear fraction (lane 13). In contrast, MAGI-1c was found almost entirely in the nuclear fraction (lane 13); a small amount was detected in the P100 fraction (lane 12), but it was completely absent from the S100 fraction (lane 11). These same fractions were analyzed on replicate blots with antibodies against the nuclear envelope protein lamin B (Fig. 5C, middle panel) and against the ␣-subunit of farnesyltransferase (FT␣), a cytoplasmic enzyme (Fig. 5C, bottom). Lamin B was detected almost entirely in the nuclear fraction (lane 13) and was also detected in the whole-cell lysates of transiently transfected 293 cells (lanes 10 and 14). In contrast, FT␣ was detected only in the S100 fraction (lane 11).
An amino acid sequence alignment revealed that the GuK domain of MAGI-1 is 27% similar to S. cerevisiae GuK over a 104-amino acid region (Fig. 6A). The similarity increased to 42% if the alignment was restricted to the region corresponding to amino acids 34 -99 of S. cerevisiae GuK (amino acids 142-205 of MAGI-1). Although the MAGI-1 GuK domain lacks an ATP-binding motif, 7 of 11 amino acids required for GMP binding in yeast GuK (31) are conserved, and two of the four mismatches represent conservative substitutions. Thus, the MAGI-1 GuK domain is similar to the GuK domains in PSD-95 and DLG, which also conserve the GMP-binding residues but lack an ATP-binding motif (32).
Amino acid sequences corresponding to the five PDZ domains in MAGI-1, designated MAGI 1 -MAGI 5 , were aligned with the sequences of the third PDZ domains in PSD-95 (PSD-95 3 ) and hDLG/SAP97 (hDLG 3 ; Fig. 6B). MAGI 1 -MAGI 5 are 34, 21, 32, 18, and 36% similar, respectively, to PSD-95 3 . The highest similarity among the PDZ domains within MAGI-1 is 36% between MAGI 1 and MAGI 5 , suggesting that each domain may interact with distinct ligands. Only MAGI 1 contains four acidic residues in the region that is predicted to form the loop between the second and third ␤ strands, consistent with the observation that K-RasB 1-185 failed to interact with the other four PDZ domains in MAGI-1. 2 The most divergent MAGI-1 PDZ domain is MAGI 4 , showing only 18% similarity to PSD-95 3 . However, 7 of the 12 amino acids that contacted a peptide ligand in the crystal structure of PSD-95 3 (28) are conserved in MAGI 4 , and three of the five mismatches are conservative substitutions. Leucine 323 and leucine 379 in PSD-95 3 are on opposite sides of a hydrophobic pocket that binds the COOH-terminal amino acid in heterotypic ligands. All five MAGI-1 PDZ domains contain a phenylalanine in the position corresponding to leucine 323, and only MAGI 2 contains a leucine at the position corresponding to leucine 379 in PSD-95 3 (Fig. 6B). This position is occupied by phenylalanine in MAGI 1 and MAGI 4 , methionine in MAGI 3 , and isoleucine in MAGI 5 . Whether these subtle substitutions in the MAGI-1 PDZ domains result in unique binding properties will await identification of ligands for each domain. Four of the five MAGI-1 PDZ domains contain a histidine at the position corresponding to histidine 372 in PSD-95 3 , and the other (MAGI 4 ) contains a lysine at this position, suggesting that each of these domains bind to ligands that contain serine, threonine, or tyrosine at the Ϫ2-position (28,29).
The amino acid sequences corresponding to the WW domains in MAGI-1, designated MAGI WW1 and MAGI WW2 , were aligned with the WW domain in human YAP65 (hYAP), both WW domains in mouse YAP65 (mYAP 1 and mYAP 2 ), and the first WW domain in NEDD4 (NEDD4 1 , Fig. 7). MAGI WW1 and MAGI WW2 are 60 and 36% similar, respectively, to hYAP. The solution structure of hYAP revealed a three-stranded ␤-sheet that adopts a slightly curved conformation (50). Prolines 174 and 202 interact with tryptophan 177 on the convex side of the sheet, stabilizing the domain. These three positions are conserved in both MAGI-1 WW domains (Fig. 7). The side chains of 2 I. Dobrosotskaya and G. L. James, unpublished data.  (Fig. 7), and these domains both bind to a PY motif in vitro. 3 However, the side chains corresponding to amino acids 194 -198 of hYAP are also predicted to contribute to the specificity of a given WW domain-ligand interaction (50), and several of these positions in MAGI WW1 and MAGI WW2 differ from the corresponding positions in hYAP. Thus, the ligand binding specificity of each MAGI-1 WW domain remains to be determined. DISCUSSION The first PDZ domain of MAGI-1 was isolated on the basis of its interaction with the COOH terminus of K-RasB in the yeast two-hybrid system. This interaction required lysine 180, threonine 183, and cysteine 185 of K-RasB (Table I). These structural features are similar to those required for localization of K-RasB to the plasma membrane (47), consistent with a possible role for MAGI-1 in this process. However, K-RasB in mammalian cells is covalently modified by farnesyl and methyl groups on its COOH-terminal cysteine, modifications that apparently preclude the ability of K-RasB to function in the yeast two-hybrid system. While it is possible that these modifications increase the affinity of the interaction between MAGI 1 and K-RasB, we have been unable to demonstrate an interaction between K-RasB and full-length MAGI-1b in lysates of transfected mammalian cells, in which K-RasB contains its full complement of posttranslational modifications. Moreover, threonine 183 of K-RasB is not required for plasma membrane localization or transformation of Rat-1 fibroblasts by oncogenic K-RasB, 2 suggesting that if K-RasB does interact with MAGI-1 in cultured cells, the interaction is not required for either of these processes. Thus, despite the specificity of the interaction between MAGI 1 and K-RasB in the yeast two-hybrid system, at this time we have no evidence that the interaction occurs in mammalian cells.
The interaction between MAGI 1 and K-RasB in the yeast two-hybrid system required four acidic residues in the PDZ domain (Asp 470 , Glu 471 , Asp 473 , and Glu 474 of MAGI-1b) and lysine 180 in K-RasB, which is 5 positions removed from the COOH-terminal cysteine (Table I). These sequence requirements were maintained when cysteine 185 of K-RasB was replaced by valine. The clustered acidic residues in MAGI 1 are in a region that is predicted to lie in close proximity to amino acids in the Ϫ5 and Ϫ6-positions in a bound ligand (27,28). Consistent with this observation, we recently isolated a candidate ligand for MAGI 1 whose sequence terminates in a TXV consensus and contains lysines in the Ϫ5 and Ϫ6-positions. 2 This novel protein, designated M 1 BP-L, binds to MAGI 1 but not to the other four PDZ domains in MAGI-1, 2 all of which lack the four clustered acidic residues found in MAGI 1 (Fig. 6B). MAGI 1 is currently the only known PDZ domain that contains four acidic residues in this region, but several PDZ domains contain two adjacent acidic residues in this loop (27,28). Although no previous studies have directly demonstrated a requirement for these residues in a PDZ domain-ligand interaction, the third PDZ domain in mouse DLG binds preferentially to a synthetic peptide that contains five lysines upstream of a TXV consensus (29). Thus, amino acids in the loop between ␤B and ␤C may be required in other PDZ domain interactions as well.
MAGI-1 mRNA undergoes alternative splicing to produce three transcripts that encode proteins with unique COOH termini (Fig. 2). The COOH termini of MAGI-1a and MAGI-1b contain 16 and 48 amino acids of unique sequence, respectively, and their molecular masses differ by only 3.5 kDa. A protein that corresponds to MAGI-1a and/or MAGI-1b was found primarily in the S100 and P100 fractions of MDCK cells, with a small amount in the nucleus (Fig. 5C). The COOH terminus of MAGI-1c extends 251 amino acids beyond the splice site and is highly charged. The last 145 amino acids consist of 37% lysine plus arginine, 19% aspartate plus glutamate, and 15% serine plus threonine, which could contribute additional negative charge via phosphorylation. This region contains three segments of 17 amino acids each (residues 1251-1267, 1293-1309, and 1299 -1315) that conform to the consensus sequence for a bipartite nuclear localization signal (Fig. 2C). This consensus sequence consists of two adjacent basic residues, followed by a spacer of 10 amino acids, followed by a second cluster of basic residues in which three of the next five residues are basic (49). Consistent with the presence of nuclear localization signals in the COOH terminus of MAGI-1c, it was found primarily in the nucleus of MDCK cells (Fig. 5C). Seven repeats of the sequence SPX(R/K) (where X is glutamate, serine, or threonine), which matches the consensus sequence for phosphorylation by p34 cdc2 (51), are clustered in and around the three nuclear localization signals, suggesting a potential regulatory mechanism for nuclear localization of MAGI-1c.
ZO-1, a MAGUK that was originally described as a component of the tight junction, was recently found in the nucleus of subconfluent, but not confluent, MDCK cells (8). ZO-1 also undergoes alternative splicing, resulting in two proteins that are distinguished by the presence or absence of an 80-amino acid segment, designated the ␣ domain (1). The two isoforms, ZO-1␣ ϩ and ZO-1␣ Ϫ , are expressed in different cell types in a pattern that correlates with junctional plasticity, but both variants were able to localize to the nucleus of subconfluent cultured cells (8). In contrast, at least two of the three MAGI-1 isoforms were co-expressed in MDCK cells and in mouse lung, and only MAGI-1c was found preferentially in the nucleus (Fig.  5). Confluent MDCK cells were used in this experiment, suggesting that MAGI-1c is present in the nucleus at a time when ZO-1 is excluded from the nucleus. Whether the subcellular distribution of MAGI-1c is influenced by changes in cell density is not yet known.
In contrast to lung, mouse brain expressed almost entirely MAGI-1a/b, whereas liver expressed primarily MAGI-1c (Fig.  5B). This differential expression of MAGI-1c in mouse tissues, together with its nuclear localization, implies that its function is distinct from that of MAGI-1a/b. The short COOH-terminal sequences unique to MAGI-1a and MAGI-1b do not contain any features that would suggest a functional difference between the two. Antibodies that recognize the unique COOH termini of these two MAGI-1 proteins will reveal whether they are coexpressed in various tissues and cell types and, if so, whether 3 B. Firulli and G. L. James, unpublished data. FIG. 7. Sequence alignment of WW domains. Amino acid sequences of the WW domain from hYAP, the two WW domains of mouse YAP65 (mYAP 1 and mYAP 2 ), the first WW domain of human NEDD4 (NEDD4 1 ), and the two WW domains of mouse MAGI-1 (MAGI WW1 and MAGI WW2 ) were aligned as described in Fig. 1B. Amino acids that are identical to the corresponding position in human YAP65 are shaded black; conservative substitutions are shaded gray. The numbers correspond to the sequence of human YAP65. Two highly conserved tryptophans are indicated by filled circles. they have distinct subcellular distributions.
MAGI-1 is currently the only known protein in which a GuK domain is combined with five PDZ domains and two WW domains (Figs. 2 and 3). Based on sequence comparisons with similar domains from other proteins, the PDZ and WW domains in MAGI-1 are each predicted to interact with distinct proteins. If the MAGI-1 GuK domain also functions as a site for protein-protein interaction, as it does in the PSD-95 proteins (33,34), MAGI-1 would have the potential to interact with up to eight different proteins simultaneously. Although the role of ZO-1 and MAGI-1 in the nucleus is not yet known, their presence there suggests that they may participate in the transfer of information directly from the cell surface to the nucleus. Future studies will seek to identify ligands for the various proteinprotein interaction domains in MAGI-1, which will provide clues to the function of this newest member of the MAGUK family.