Human homologue of the Drosophila discs large tumor suppressor protein forms an oligomer in solution. Identification of the self-association site.

The human homologue of the Drosophila discs large tumor suppressor protein (hDlg), a member of the membrane-associated guanylate kinase (MAGUK) superfamily, interacts with K(+) channels, N-methyl-d-aspartate receptors, calcium ATPase, adenomatous polyposis coli, and PTEN tumor suppressor proteins, and several viral oncoproteins through its PDZ domains. MAGUKs play pivotal roles in the clustering and aggregation of receptors, ion channels, and cell adhesion molecules at the synapses. To investigate the physiological basis of hDlg interactions, we examined the self-association state of full-length hDlg as well as defined segments of hDlg expressed as recombinant proteins in bacteria and insect Sf9 cells. Gel permeation chromatography of full-length hDlg revealed that the purified protein migrates as a large particle of size >440 kDa. Similar measurements of defined domains of hDlg indicated that the anomalous mobility of hDlg originated from its amino-terminal domain. Ultrastructural analysis of hDlg by low angle rotary shadow electron microscopy revealed that the full-length hDlg protein as well as its amino-terminal domain exhibits a highly flexible irregular shape. Further evaluation of the self-association state of hDlg using sedimentation equilibrium centrifugation, matrix-assisted laser desorption/ionization mass spectrometry, and chemical cross-linking techniques confirmed that the oligomerization site of hDlg is contained within its amino-terminal domain. This unique amino-terminal domain mediates multimerization of hDlg into dimeric and tetrameric species in solution. Sedimentation velocity experiments demonstrated that the oligomerization domain exists as an elongated tetramer in solution. In vitro mutagenesis was used to demonstrate that a single cysteine residue present in the oligomerization domain of hDlg is not required for its self-association. Understanding the oligomerization status of hDlg may help to explicate the mechanism of hDlg association with multimeric K(+) channels and dimeric adenomatous polyposis coli tumor suppressor protein. Our findings, therefore, begin to rationalize the role of hDlg in the clustering of membrane channels and formation of multiprotein complexes necessary for signaling and cell proliferation pathways.

Membrane-associated guanylate kinase homologues (MAGUKs) 1 are believed to play important roles in the assembly of signal transduction complexes at the interface of the membrane-cytoskeleton, and may couple extracellular signals to the intracellular signaling pathways (1)(2)(3)(4)(5). The core primary structure of MAGUKs is composed of either one or three PDZ domains, a SH3 motif, and a carboxyl terminus domain homologous to the mammalian guanylate kinase (1). MAGUKs have been localized at the pre-synaptic and post-synaptic density and at the regions of cell to cell contact where they are thought to play an important role in the regulation of synaptic transmission and cell proliferation (6 -9). We have previously shown that MAGUKs associate with the cytoskeletal protein 4.1 (9 -12). This interaction is mediated by the binding of the HOOK domain of MAGUKs with the amino-terminal FERM domain of protein 4.1, thus providing a paradigm for localizing MAGUKs at the interface of membrane-cytoskeleton and at the sites of cell to cell contacts in epithelial cells (9,12,13).
Unlike other MAGUKs, hDlg contains a unique amino-terminal domain of unknown function (9). As a result of alternative splicing-in of a 33-amino acid insertion-1, hDlg contains a proline-rich sequence near its amino terminus that has been shown to interact with the p56 Lck tyrosine kinase (14). The characterization of the PDZ domains of hDlg has established the ϳ100-amino acid modules as sites of protein-protein interactions (15,16). They either participate in homotypic association with other PDZ domains or recognize a consensus sequence T/SXV located at the carboxyl terminus of target protein (15,16). The determination of their three-dimensional structure as well as elucidation of peptide sequences that bind to various PDZ domains has revealed the molecular basis of the specificity of PDZ domain interactions (17)(18)(19). The third PDZ domain of PSD-95 binds a novel protein termed CRIPT that may anchor PSD-95 MAGUK to the tubulin-based cytoskeleton in excitatory synapses (20). The first and second PDZ domains of hDlg form a conformationally stable module that binds to the carboxyl termini of several transmembrane and cytoplasmic proteins (5,21). The PDZ domain-mediated interaction of MAGUKs with ion channels, receptors, and cell adhesion proteins plays a critical role in the localization, aggregation, and clustering of these proteins at the synapses and neuromuscular junctions (22). Indeed, the co-expression of PSD-95/ hDlg with ion channels and receptors in heterologous cells results in the clustering of ion channels and receptors at the cell surface, whereas these channels/receptors fail to assemble into clusters when expressed singly in heterologous cells (23). Further evidence for a direct role of MAGUKs in the clustering of ion channels and receptors comes from in vivo studies of the Drosophila discs large tumor suppressor (Dlg) protein (24). The Dlg colocalizes with the Shaker-type K ϩ channels at the neuromuscular junction, and induces clustering of these ion channels in vivo (24). Moreover, the clustering of these channels is abolished in the Dlg loss-of-function mutants (24). The Dlg has also been implicated in the clustering and stabilization of the transmembrane protein fasciclin II at the neuromuscular junctions (25,26). Similarly, Dlg has been localized at the mammalian neuromuscular junctions and may function in the maintenance and organization of protein complexes at these junctions (27).
The function of the centrally located SH3 domain of MAGUKs is poorly understood. These domains are predicted to recognize the proline-rich consensus sequence of proteins that presumably contribute to the formation of multiprotein complexes at the plasma membrane (28,29). The enzymatically inactive guanylate kinase-like domains of PSD-95 and hDlg interact with a family of closely related proteins termed GKAP/ DAP-1␤, SAPAP-1/DAP-1␣, and SAPAP-2/DAP-2 (30 -32). The function of these proteins is not yet known. The observation that the multimeric APC tumor suppressor protein binds to hDlg raised an intriguing possibility that hDlg may self-associate in vivo (33). The APC protein regulates cell proliferation by blocking the transition of cells from the G o /G 1 phase to S phase (33,34). To investigate the self-association state of hDlg, we have examined the hydrodynamic properties of recombinant hDlg protein using a combination of biochemical and biophysical techniques. Here we report that the hDlg protein selfassociates into dimeric and tetrameric species in solution, and that the self-association site is located within its unique aminoterminal domain.

MATERIALS AND METHODS
cDNA Constructs of hDlg-The following cDNA constructs encoding defined segments of hDlg were engineered ( Fig. 1) (9). (i) Full-length hDlg  (amino acids 1-926). (ii) Truncated hDlg termed hDlg 201-904 (amino acids 201-904). This segment includes the three PDZ domains, the SH3 motif, the insertion-2, and the guanylate kinase-like domain. (iii) hDlg segment termed hDlg 1-229 (amino acids 1-229) includes the unique amino-terminal domain and the proline-rich insertion-1 sequence. (iv) hDlg 1-161 (Cys-66) (amino acids 1-161) segment contains the unique amino-terminal domain of hDlg. The cDNA for this segment was amplified using the primer pair Nf (sense, BamHI) 5Ј-GCTGGATCCAT-GCCGGTCCGGAAG; and Nr (antisense, EcoRI) 5Ј-GGCGAATTCCTT-TATTGGTGAAAT. (v) Mutant hDlg 1-161(C66S) segment where a single cysteine residue at position 66 was changed to serine by substituting guanine for cytosine using the QuickChange site-directed mutagenesis kit (Stratagene Inc.). (vi) hDlg 1-624 (amino acids 1-624) segment encodes a truncated form of hDlg lacking the carboxyl-terminal domain. This construct contained a substitution of cytosine to thymidine at position 2061 thus introducing a stop codon at amino acid 625 located within the SH3 domain. All cDNA constructs were cloned into pGEX-2T vector and verified by DNA sequencing. Recombinant fusion proteins carrying glutathione S-transferase (GST) at the amino terminus were expressed in Escherichia coli DH5␣ cells as described previously (35). In addition, the cDNA constructs encoding hDlg 1-161(Cys-66) and hDlg 1-161(C66S) were subcloned into pFastBacHTb vector for expression in the insect Sf9 cells using the BAC-to-BAC Baculovirus Expression System (Life Technologies Inc.). The recombinant proteins expressed in Sf9 cells contained a histidine-tag attached to the amino terminus.
Purification of Recombinant hDlg Fusion Proteins-Recombinant GST-hDlg fusion proteins were expressed in bacteria and purified by affinity chromatography using glutathione-Sepharose beads (35). The GST portion of the fusion proteins was cleaved using thrombin (35). Thrombin does not cleave recombinant proteins internally as has been demonstrated previously (5). The recombinant His-tagged proteins, designated as BV hDlg  and BV hDlg 1-161(C66S) , were expressed in the insect Sf9 cells and affinity purified using Ni-NTA-agarose beads according to the manufacturer's instructions (Life Technologies Inc.). The recombinant fusion proteins were eluted from the beads using 100 mM imidazole. The amino-terminal His-tag sequence of the fusion proteins was then cleaved using the rTEV protease (Life Technologies Inc.). No internal cleavage site was predicted for rTEV protease, and incubation of cleaved proteins with rTEV protease for extended periods did not result in any further cleavage of the proteins. The cleaved recombinant hDlg proteins were dialyzed against buffer A (20 mM Tris-HCl, pH 7.6, 20 mM NaCl, 1.0 mM EGTA, 0.5 mM DTT, and 2.0 mM MgCl 2 ) and purified on a Mono Q anion exchange column (fast protein liquid chromatography) using a linear gradient of NaCl (20 -400 mM). The peak fractions containing pure protein were pooled, dialyzed, and concentrated in buffer B (20 mM sodium phosphate, pH 7.6, 150 mM NaCl, 1.0 mM EGTA, 2.4 mM dithiothreitol). The dialyzed proteins were analyzed by gel permeation chromatography, electron microscopy, sedimentation velocity, and sedimentation equilibrium techniques.
Gel Permeation Chromatography-Purified hDlg proteins (30 -60 g) were analyzed by gel filtration chromatography on a Superose 12 column (HR 10/30; Amersham Pharmacia Biotech) equilibrated with buffer B at a flow rate of 15 ml/h. Fractions of 0.5 ml were collected and analyzed by SDS-PAGE (36). Molecular mass standards were analyzed under the same conditions to generate a calibration curve.
Electron Microscopy-Peak fractions containing purified hDlg proteins were obtained from the Superose 12 column (fast protein liquid chromatography). Samples were analyzed using low angle rotary shadow electron microscopy (37). Protein samples were reconstituted in 70% glycerol (phosphate-buffered saline), sprayed onto freshly cleaved mica, and rotary shadowed using platinum-carbon and carbon anodes. The replicas were floated and mounted on 400-mesh copper grids for electron microscopy (37).
Analytical Ultracentrifugation-Experiments were performed at 4 or 5°C using a Beckman Optima XL-A analytical ultracentrifuge equipped with scanning absorbance optics. For sedimentation equilibrium runs samples were loaded into 12-mm path length Yphantis-type 6-channel centerpieces. The solution column height in each case was 1.5 mm. Rotor speeds were as follows: 10,000, 12,000 rpm (full-length hDlg 1-926 ); 5,000 rpm (segment hDlg 1-624 ); 7,000, 9,000, 11,000, and 15,000 rpm (segment hDlg  ); 20,000 and 25,000 rpm (segment hDlg  in the absence of DTT); 15,000 and 25,000 rpm (segment hDlg 1-161(C66S) in the presence of DTT; segment BV hDlg  in the presence of DTT; segment BV hDlg 1-161(C66S) in the absence of DTT). In all cases the rotor (An 60-Ti) was over-speeded to 40,000 or 47,000 rpm in order to deplete the solution column of macromolecular solute so that a true optical baseline could be obtained. The equilibrium solute distributions were recorded at either 280 or 277 nm, depending on the protein being studied. Equilibrium at each rotor speed was ascertained by the overlaying of scans acquired 3 h apart. The apparent whole cell weight average molecular mass (M w,app ) was obtained by fitting the data with the equation describing the equilibrium distribution of a single thermodynamically ideal macromolecular solute (A r ϭ A 0 exp[H⅐M(r 2 Ϫ r 0 2 ] ϩ E) incorporated in the Beckman XL-A analysis software. This model is a subset of a broader model (Equation 1 (38)) which was used initially to analyze self-interaction in solute systems where the elevated value of M w,app above the known monomer molecular mass was indicative of some level of oligomeric behavior. Equation 1 describes a self-associating system defined by up to three association constants (K a 2 , K a 3 , K a 4 in units of (absorbance) Ϫ(n-1) for oligomers composed of n 2 , n 3 , and n 4 monomers, where A r is the absorbance at radial position r and A 0 the absorbance at a reference position r 0 (cm); H is the constant (1 Ϫ v ) 2 , v is the partial specific volume of the macromolecule (ml/g); is the solvent density (g/ml); is the rotor speed (radians/s); R is the gas constant (8.314 ϫ 10 Ϫ7 erg K Ϫ1 mol Ϫ1 ); T is the temperature (K); M is the molecular weight of the solute (g/mol); E is the optical baseline offset (obtained by over-speeding of the rotor). A system in which the elevation in M w,app is due solely to self-association should be describable by a single system of association constants. While the sedimentation equilibrium data for the hDlg proteins could be globally fitted with single self-association models the fits were improved if the K a values were allowed to vary for each data set. The need to use separate K a values can be indicative of the presence of incompetent monomer (i.e. monomer that cannot self-associate) or contaminant solute (or aggregate) that takes no part in the association but that nonetheless adopts an equilibrium solute distribution. The program PCNONLIN (39) which uses a more generalized form of Equation 1 was used to this end. For each hDlg segment a range of models was tested, starting with the simplest possible model that could conceivably result in the observed trend in reduced molecular weight data (i.e. M w,app ). In each case data acquired for a range of loading concentrations and rotor speeds were fitted simultaneously with the chosen model. In most cases the association constants were then fitted separately in order to improve the fit. Fits obtained with this procedure were optimized by minimizing the variance and judged by the randomness and magnitude of the resultant residuals. Solvent densities were obtained from data tables (40); partial specific volumes were calculated from the amino acid compositions of the protein solutes and data for the v of these constituent amino acids (41). Sedimentation velocity data were acquired for three concentrations of segment hDlg 1-161(C66S) in the presence of DTT, segment BV hDlg  in the presence of DTT, segment BV hDlg 1-161(C66S) in the absence of DTT. Samples were loaded into double sector centerpieces. The rotor speed was 30,000 rpm for the highest concentration samples and 35,000 rpm for the lower two concentrations. Sixty scans were acquired in continuous scanning mode using 280 nm absorbance at 10-min intervals so that the data could be analyzed using the program DCDT (42). The best results were obtained by analysis of scans 21-40 alone. This yielded the unnormalized distribution function ĝ (s*) which is defined as follows.
From this the weight average sedimentation, coefficient, s w , was derived, where c is concentration (in absorbance units), t is the equivalent time of sedimentation (s), r m is the radial position of the meniscus (cm), r is the radius (cm), c p is the concentration in the plateau region of the sedimentation velocity trace. The subscript corr indicates that the value of Ѩ c /Ѩ t has been corrected as described by Stafford (43). Chemical Cross-linking Studies-Copper phenanthroline catalyzed reversible cross-linking of hDlg proteins was carried out as described before (44). The recombinant hDlg protein (30 g) dialyzed in triethanolamine buffer (50 mM triethanolamine-HCl, pH 8.0, 100 mM KCl) was incubated with 1.7 mM copper sulfate and 3.4 mM O-phenanthroline either in the presence or absence of 24 mM DTT. After 2 h of incubation at room temperature, the oxidation reaction was stopped by the addition of 3.2 mM EDTA. The protein samples were incubated with SDS sample buffer in the absence of reducing agent for 20 min at 37°C and analyzed by SDS-PAGE. Amine cross-linking of the hDlg proteins was carried out using the NHS-esters, DSS (disuccinimidyl suberate) and BS 3 (bis(sulfosuccinimidyl)suberate) (Pierce) as described by the manufacturer. Briefly, purified proteins were dialyzed against phosphate buffer (20 mM sodium phosphate, pH 7.5, 150 mM NaCl, 1.0 mM EGTA), and incubated with an increasing concentration (up to 10-fold molar excess) of the cross-linker at room temperature for 30 min. The crosslinking reaction was stopped by the addition of 2.0 M Tris-HCl, pH 7.5, to a final concentration of 35.0 mM. The reaction mixture was incubated for 15 min at room temperature followed by the addition of SDS sample buffer containing ␤-mercaptoethanol. The cross-linked polypeptides were analyzed by 12% SDS-PAGE.

RESULTS
Chromatography of hDlg Proteins-The full-length hDlg 1-926 protein eluted as a considerably larger polypeptide (Ͼ440 kDa) from the Superose 12 column in contrast to its calculated molecular mass of 104 kDa (Table I). Based on this observation, the oligomerization state and molecular shape of the hDlg protein were investigated. A series of hDlg proteins were engineered and expressed as recombinant GST fusion proteins in bacteria (Fig. 1). The GST portion was cleaved by thrombin, and recombinant proteins were purified by chromatography on the Mono Q anion-exchange column using the fast protein liquid chromatography. Purified hDlg proteins were then analyzed on a pre-calibrated gel filtration (Superose 12) column ( Fig. 2, Table I). The full-length hDlg 1-926 protein eluted immediately after the void volume peak corresponding to an estimated size of ϳ1400 kDa as calculated from the extrapolated calibration curve shown in Fig. 2. A similar elution behavior was observed for the hDlg 1-624 segment lacking a portion of the SH3 motif and the carboxyl-terminal guanylate kinase-like domain. The hDlg 1-624 protein with a calculated molecular mass of 69 kDa eluted as a ϳ1200-kDa species (Fig. 2). To further investigate the role of other segments of hDlg, the mobility of mutually exclusive segments hDlg 1-229 and hDlg 201-904 was measured. Interestingly, the relatively small hDlg 1-229 segment with a calculated molecular mass of 27 kDa eluted immediately after the void volume peak (size ϳ1200 kDa) whereas the hDlg 201-904 segment with a calculated molecular mass of 79 kDa migrated as a ϳ190-kDa species (Fig. 2). These results suggests that the full-length hDlg protein as well as hDlg 1-624 and hDlg 1-229 segments either self-associate or adopt a non-globular shape or both in solution.
A shared feature of the full-length hDlg  , and the segments hDlg 1-624 and hDlg 1-229 is the presence of the aminoterminal domain and a proline-rich insertion-1 sequence (Fig.  1). The amino-terminal proline-rich domain that is unique to hDlg is absent from the hDlg 201-904 segment implying that this domain may contribute to the anomalous mobility of hDlg on the Superose 12 column. Since the proline-rich insertion-1 is encoded by an alternatively spliced exon, a cDNA construct was designed that encoded only the first 161 amino acids of hDlg thus lacking the proline-rich insertion-1 (Fig. 1). Interestingly, the hDlg 1-161 segment with a calculated molecular mass of 19 kDa eluted as a ϳ650-kDa species compared with the hDlg 1-229 segment which eluted as a ϳ1200-kDa species (Fig. 2). Taken together, these data indicate that the unique amino-terminal domain as well as the proline-rich insertion-1 contribute to the anomalous mobility of hDlg protein as measured by gel filtration chromatography.
The data described above were obtained from hDlg segments that were expressed as recombinant proteins in bacteria. Therefore, a possibility exists that the observed anomalous mobility of hDlg proteins on the gel filtration column may be a consequence of artifactual aggregation resulting from the bacterial expression system. To rule out this possibility, the hDlg 1-161 segment was expressed as a recombinant His-tag fusion protein in insect Sf9 cells using the Baculovirus Expression System. We designate this segment as BV hDlg 1-161 . The amino-terminal His-tag was cleaved of by the rTEV protease and the recombinant BV hDlg 1-161 protein was purified by chromatography on the Mono Q anion exchange column. Purified protein was then analyzed by gel filtration chromatography using a Superose 12 column. The BV hDlg 1-161 segment with a calculated molecular mass of 20 kDa eluted as a ϳ650-kDa species similar to the elution profile of the hDlg 1-161 segment expressed in bacteria (Fig. 2, Table I). This result indicates that the observed anomalous mobility of hDlg protein is an intrinsic property of the molecule and is not influenced by in vitro expression systems.
Ultrastructural Visualization of hDlg Proteins-To investigate whether the abnormal mobility of hDlg may be explained by its non-globular shape, an analysis of its ultrastructure was performed using low angle rotary shadow electron microscopy. Full-length hDlg 1-926 appeared as an ill-defined flexible molecule with an irregularly shaped contour (Fig. 3, arrowheads). The hDlg 1-624 and hDlg 1-229 segments also exhibit similar asymmetrical shapes albeit with a relatively smaller particle size (Fig. 3, Table II). In contrast, the hDlg 201-904 segment displays a well defined bilobed structure presumably reflecting the contribution of PDZ1ϩ2ϩ3 and guanylate kinase-like domains (5). It is noteworthy that the hDlg 1-229 segment with a calculated molecular mass that is one-third the mass of hDlg 201-904 segment and exhibits a particle size that is about two times larger than the size of hDlg 201-904 segment (Fig. 3, Tables I and II). These results suggest that the presence of the amino-terminal domain and the proline-rich insertion-1 sequence may bestow the characteristic molecular shape and size of hDlg, thus providing an explanation for the anomalous behavior of hDlg proteins by gel permeation chromatography. To further investigate the basis of these observations, we examined the association state and molecular shape of hDlg proteins by analytical ultracentrifugation.
Sedimentation Equilibrium and Sedimentation Velocity Centrifugation Analysis of hDlg-The hDlg proteins were analyzed by analytical ultracentrifugation to assess their degree of oligomerization and elongation. The apparent whole cell weight average molecular mass (M w,app ) obtained from fitting with the model for a single, thermodynamically ideal solute (see "Materials and Methods") to the sedimentation equilibrium data for the full-length hDlg 1-926 , segment hDlg 1-624 , segment hDlg 1-229 , segment hDlg  in the absence of DTT, segment hDlg 1-161(C66S) in the presence of DTT, segment BV h-Dlg 1-161(Cys-66) in the presence of DTT, and segment BV hDlg 1-161(C66S) in the absence of DTT is plotted as a function of solute concentration in Fig. 4A (a-e). The goodness of fit with this model was variable from protein to protein. Specifically, the fits for the full-length hDlg 1-926 were poor (reflected in the comparably large error bars in Fig. 4A (a)). Instead the raw data were better fitted as a self-associating system using the procedure described under "Materials and Methods." The reduced data of Fig. 4A (a) imply that the full-length hDlg exists either in monomer-dimer equilibrium (with a very low K d ) or perhaps as a dimer in equilibrium with a higher order oligomer, given that M w,app exceeds dimer mass in some instances. Indeed fits with a monomer-dimer model were very poor (and impossible when M w,app exceeded dimer mass). Similarly, because some values of M w,app below dimer mass were obtained it was unsurprising that the system could not be described in terms of a dimertetramer equilibrium. Instead it was possible to globally fit the 8 data sets with models for monomer-trimer, monomer-tetramer, and monomer-dimer-tetramer equilibria. The residuals for these fits are presented in Fig. 4B (a) from which it is apparent that the best model of the three is the monomerdimer-tetramer. The values for K d1-2 and K d1-4 obtained for this fit are depicted in Fig. 4B (b). The magnitude of K d1-2 remains fairly constant as a function of loading concentration and rotor velocity (Ϸ10 -30 M) but K d1-4 has to increase to adequately describe the data (10 -100 M in the concentration range examined). An increase in K d with concentration is symptomatic of thermodynamic non-ideality which can arise either from excess unshielded net molecular surface charge or from molecular elongation or asymmetry. Given that hDlg does not carry significant net charge in the buffer used for these studies it seems likely that the tetramer species is elongated or highly asymmetric.
The reduced sedimentation equilibrium data for segment hDlg 1-624 are presented in Fig. 4A (b). Thus the whole cell weight average molecular mass is in excess of a dimer mass and, at infinite dilution, approaches that of trimer. The decrease in M w,app with increasing concentration is indicative of thermodynamic non-ideality. The raw data were fitted with a number of models. Those that gave fits were monomer-trimer, monomer-tetramer, and dimer-tetramer. The residuals of these fits are shown in Fig. 4B (c). Marginal improvement in these residuals is attained by the dimer-tetramer fit but in reality there is little to choose between them. Certainly all the fits show residuals indicative of non-ideality echoing the trend observed in Fig. 4A (b) for the reduced data. However, the fits were not improved when a finite value was given to the second virial coefficient in the fitting equation. It is likely that in a self-association of already elongated molecules a single second virial coefficient will be insufficient to describe the non-ideality presented by the various oligomers. The trend in dissociation constants with loading concentration shown in Fig. 4B (d) reveals deficiencies in all models. Therefore all that can be concluded for certain is that self-association is occurring, that the species involved are likely to be elongated or highly charged, and that trimer or tetramer is the largest true oligomer in the system.
The data in Fig. 4A (c) for segment hDlg 1-229 show a small amount of variability with rotor speed but in fact the main source of the scatter is the variable reliability of the fits with the single ideal species form of Equation 1. Again, M w,app exceeds that for monomer, dimer, and trimer and increases with increasing cell loading concentration suggesting that the system may be in a reversible equilibrium. Accordingly the raw data were fitted with numerous models (monomer-tetramer, dimer-tetramer, dimer-octamer, and dimer-tetramer-octamer). The best and worst of these fits are shown in Fig. 4B (e). There is no difference between the monomer-tetramer and dimertetramer fits, both are the poorest of the fits obtained (Fig. 4B  (e and i)). The fits were significantly improved by the introduction of an octameric component. Fitting with a straight dimeroctamer equilibrium yielded the most satisfactory results while the more complex dimer-tetramer-octamer model yielded meaningless equilibrium constants for the dimer-tetramer equilibrium. The dimer-octamer fit for data acquired at rotor speeds of 7,000, 9,000, and 11,000 rpm is shown in Fig. 4B (e  and ii), a fit of similar quality was obtained for the global fitting of data acquired at rotor speeds of 9,000, 11,000, and 15,000 rpm. In these fitting procedures the fit is optimized by allowing the separation of the equilibrium constants, to allow for nonideality or for the presence of incompetent monomer (or other protein which plays no part in the equilibrium). The resultant equilibrium constants are plotted in Fig. 4B(f).
The next in the series of reduced length segments (segment hDlg 1-161(Cys-66) ) maintains the self-associative behavior of its larger counterparts. The reduced data for this segment obtained in the absence of DTT are shown in Fig. 4A (d). M w,app is in excess of the value for a tetrameric species. The raw data were thus fitted with models for monomer-octamer and monomer-tetramer-octamer but the fits were very poor. An improvement was obtained by setting the lowest mass species to dimer and fitting for dimer-tetramer-octamer. But the best fits were obtained for a straightforward tetramer-octamer model (Fig.  4B (f)) with the equilibrium constant allowed to vary in order to account for the imperfect equilibrium within the cell. The resultant dissociation constants are shown in Fig. 4B (g). Even allowing the equilibrium constant to vary, the fluctuations in its magnitude are comparably small.
Finally, the reduced data for sedimentation equilibrium analysis of segment hDlg 1-161(C66S) in the presence of DTT, segment BV hDlg 1-161(Cys-66) in the presence of DTT, and segment BV hDlg 1-161(C66S) in the absence of DTT are shown in Fig.  4A (e). Again M w,app exceeds the value for monomer, dimer, and, at 15,000 rpm, trimer. The first striking aspect of this plot  is the significant drop in M w,app with increased rotor speed for all three species. Additionally and unsurprisingly it was impossible to fit the data sets acquired at the two rotor speeds with one self-association model. It appears thus that all three of these segment samples contain a significant amount of aggregate that remains in the bulk of the solution column at 15,000 rpm but is mostly removed at 25,000 rpm. Certainly, any fits to the data acquired at 15,000 rpm generated residuals highly indicative of the presence of aggregate, even with the inclusion of an octameric species. Therefore only the 25,000 rpm data were further interpreted. For BV hDlg 1-161(Cys-66) in the presence of DTT the best fits were obtained with a mono-mer-tetramer or dimer-tetramer model for which no significant improvement was obtained by allowing the equilibrium constants to vary. The resultant residuals for the monomer-tetramer model are shown in Fig. 4B (h) for which K d14 ϭ 3.09 M 3 . A similar situation exists for segment BV hDlg 1-161(C66S) in the absence of DTT although the data for this segment were undeniably best fitted with the monomer-tetramer model. The residuals for this are shown in Fig. 4B (i) and K d14 ϭ 6.23 M 3 . Unfortunately, owing to experimental difficulties, the third in this series of fragments (hDlg 1-161(C66S) in the presence of DTT) was only studied at one loading concentration (15.4 M) and these data were also best fitted with a monomer-tetramer (i) monomer-trimer; (ii) monomer-tetramer; (iii) monomer-dimer-tetramer. b, dissociation constants used to fit sedimentation equilibrium data for hDlg 1-926 with a monomer-dimer-tetramer model plotted as a function of monomer loading concentration. K d1-2 , 10,000 rpm (OE); K d1-2 , 12,000 rpm (f); K d1-4 , 10,000 rpm (‚); K d1-4 , 12,000 rpm (Ⅺ). c, residuals of fitting 9 sets of sedimentation equilibrium data (loading concentrations 0.47-5.69 M) with self-association models plotted as a function of local concentration (expressed in terms of monomer) in the centrifuge cell for hDlg  . (i) monomer-trimer; (ii) monomer-tetramer; (iii) dimer-tetramer. d, dissociation constants used to fit sedimentation equilibrium data for segment hDlg 1-624 with a monomer-trimer model (E), monomer-tetramer model (q), and dimer-tetramer model (‚) plotted as a function of monomer loading concentration. e, residuals of fitting 15 sets of sedimentation equilibrium data (five loading concentrations (1.18 -5.91 M) equilibrated at three rotor speeds (7,000, 9,000, and 11,000 rpm)) with self-association models plotted as a function of local concentration (expressed in terms of monomer) in the centrifuge cell for hDlg  . (i) monomer-tetramer; (ii) dimer-octamer. f, dissociation constants used to fit sedimentation equilibrium data for segment hDlg 1-229 with a dimer-octamer model at rotor speeds 7,000 rpm (q), 9,000 rpm (‚), and 11,000 rpm (؉) plotted as a function of monomer loading concentration. g, residuals of fitting 7 sets of sedimentation equilibrium data (loading concentrations 5.9 -20.2 M) equilibrated at two rotor speeds (20,000 and 25,000 rpm)) with a self-association model for a tetramer-octamer equilibrium plotted as a function of local concentration (expressed in terms of monomer) for segment hDlg  in the absence of DTT. h, dissociation constants used to fit sedimentation equilibrium data for segment hDlg  in the absence of DTT with a tetramer-octamer model at rotor speeds 20,000 rpm (E) and 25,000 rpm (؉) plotted as a function of monomer loading concentration. (i) Residuals of fitting 3 sets of sedimentation equilibrium data (loading concentrations 9.0 -15.5 M) equilibrated at 25,000 rpm with a self-association model for a monomer-tetramer equilibrium plotted as a function of local concentration (expressed in terms of monomer) for BV hDlg  in the presence of DTT. j, residuals of fitting 3 sets of sedimentation equilibrium data (loading concentrations 7.4 -12.9 M) equilibrated at 25,000 rpm with a self-association model for a monomer-tetramer equilibrium plotted as a function of local concentration (expressed in terms of monomer) for BV hDlg 1-161(C66S) in the absence of DTT. C, weight-average sedimentation coefficient from analysis with DCDT plotted as a function of monomer solute loading concentration for segment BV hDlg  in the presence of DTT (q); segment BV hDlg 1-161(C66S) in the absence of DTT (p); segment hDlg 1-161(C66S) in the presence of DTT (Ⅺ). model for which K d14 ϭ 2.66 M 3 .
Hence the sedimentation equilibrium data indicate that hDlg self-associates with high affinity into dimers and tetramers in solution and that the unique amino-terminal domain contains the oligomerization site. The data further indicate that under nonreducing conditions, segment hDlg 1-161(Cys-66) tends to form higher order oligomers and exists in a tetrameroctamer equilibrium. Under reducing conditions, the segments BV hDlg 1-161(Cys-66) and hDlg 1-161(C66S) form predominantly a tetramer and exist in dimer-tetramer or monomer-tetramer equilibrium suggesting a role for cysteine at position 66 in the formation of higher order oligomers.
Sedimentation velocity data were analyzed with DCDT (42). The weight-average sedimentation coefficients obtained were corrected for the effects of solvent viscosity and density and for temperature and thus standardized to s 20,w , the sedimen- tation coefficient at 20°C in water. These reduced data for segment BV hDlg  in the presence of DTT, segment BV hDlg 1-161(C66S) in the absence of DTT, and segment BVhDlg 1-161(C66S) in the presence of DTT are plotted as a function of solute loading concentration in Fig. 4C. In each case the raw data were well described by a single Gaussian, indicating that the sample contains predominantly one molecular species. On the basis of comparatively few data points in Fig.  4C it is hard to extrapolate formally to infinite dilution to obtain the normally quoted s 20,w 0 but it is clear that there is little difference between the three samples and that all have s 20,w 0 close to 3.0 S. Accepting that, from sedimentation equilibrium data (above), these three segments are in a monomertetramer equilibrium, the K d for these equilibria are: 3.09 M 3 ( BV hDlg  in the presence of DTT), 6.23 M 3 ( BV hDlg 1-161(C66S) in the absence of DTT), and 2.66 M 3 (hDlg 1-161(C66S) in the presence of DTT). Thus, at the concentrations used in the sedimentation velocity studies, the samples will consist largely of tetramer. A spherical tetramer would have s 20,w 0 ϭ 6.22 S if it were hydrated to a typical protein level (0.4 g of water/g of protein). In order for the tetramer to have a sedimentation coefficient half this value it must either be very highly charged and/or elongated. Segment hDlg 1-161 does not have a particularly high charge density therefore its tetramer must be elongated. Thus the sedimentation velocity analysis results indicate that the amino-terminal domain of hDlg exists as an elongated tetramer in solution.
Based on the sedimentation equilibrium and sedimentation velocity data it is now possible to rationalize the anomalous behavior of hDlg proteins observed with gel permeation chromatography (Fig. 2). Furthermore, the observed EM images of hDlg proteins as highly flexible asymmetrical structures are consistent with these results, and reflect elongated tetramers that may curl around during the rotary shadowing procedure (Fig. 3).
Chemical Cross-linking of hDlg Proteins-The amino-terminal domain of hDlg contains a single cysteine residue at position 66 that may mediate oligomerization of hDlg through disulfide linkage as indicated from the sedimentation equilibrium data. To investigate the role of cysteine 66, the hDlg 1-161(Cys-66) segment was subjected to chemical cross-linking using copper/ phenanthroline in the presence or absence of a reducing agent. The copper/phenanthroline induces reversible oxidation of the sulfydryl groups forming disulfide bonds between near neighbors in solution. This approach has been previously used to investigate the oligomeric state of the red cell protein dematin (44). Similarly, nonreducing SDS-PAGE can be used to demonstrate the presence of disulfide-linked dimers (45,46). When separated by SDS-PAGE under nonreducing conditions, the disulfide bonds of dimers are preserved and can be detected as protein complexes migrating at approximately twice the size of the corresponding monomeric form. The cross-linked product of the hDlg 1-161(Cys-66) segment migrated as a dimer in the absence of DTT (Fig. 5A, lane 1), whereas the same product  (lanes 1-4) and hDlg 1-161(C66S) (lanes 5-8). Proteins were incubated with an increasing concentration of DSS as described under "Materials and Methods." The cross-linker to protein molar ratio was 1 (lanes 1 and 5), 3 (lanes  2 and 6), 5 (lanes 3 and 7), and 10 (lanes 4 and 8). Arrows identify the position of monomer, dimer, tetramer, and octamer bands. As the concentration of DSS increases, the hDlg proteins are cross-linked into higher order oligomers. An identical pattern of cross-linked products was observed for both hDlg 1-161(Cys-66) and hDlg 1-161(C66S) segments. C, amine cross-linking of BV hDlg 1-161(Cys-66) (lanes 1-3), BV hDlg 1-161(C66S) (lanes 4 -6), and hDlg 201-904 (lanes 7-9). Proteins were incubated with increasing concentrations of BS 3 . The cross-linker to protein molar ratio was 3 (lanes 1, 4, and 7), 5 (lanes 2, 5, and 8), and 10 (lanes 3, 6, and 9). The arrowhead and asterisk identify the position of intact and degraded products of hDlg 201-904 segment, respectively. The hDlg 201-904 segment was included as a negative control. D, separation of hDlg 1-161(Cys-66) and hDlg 1-161(C66S) segments on native polyacrylamide gels. Linear 4 -15% gradient nondenaturing polyacrylamide gels were used without DTT. Cys-66 and Ser-66 designations refer to hDlg segments hDlg 1-161(Cys-66) and hDlg 1-161(C66S) , respectively. Albumin (pI 4.7-4.9; molecular mass ϳ66,430 Da) and ovalbumin (pI 4.5-4.9; molecular mass ϳ43,000 Da) were used as markers to roughly estimate the position of dimeric species of hDlg 1-161(Cys-66) segment (pI ϳ 4.93). migrated as a monomer in the presence of DTT (Fig. 5A, lane  2). The dimeric band was then excised and analyzed by SDS-PAGE in the presence of ␤-mercaptoethanol. As shown in Fig.  5A (lane 3), the hDlg 1-161(Cys-66) dimer dissociated and migrated as a monomeric species. A similar result was obtained with the hDlg 1-229 segment (data not shown). These results indicate that the hDlg 1-161 and hDlg 1-229 segments can be cross-linked into dimers via the oxidation of their sulfydryl group.
To investigate further the role of cysteine 66 in the oligomerization of hDlg 1-161 segment, the cysteine 66 residue was changed to serine 66 by in vitro mutagenesis. As expected, the hDlg 1-161(C66S) segment did not cross-link into a dimer by copper/phenanthroline oxidation (Fig. 5A, lanes 4 and 5). The hDlg 1-161(Cys-66) and hDlg 1-161(C66S) segments were then crosslinked using the amine cross-linkers, DSS and BS 3 , under nonreducing conditions. The irreversibly cross-linked products were generated at cross-linker to protein molar ratio of 1, 3, 5, and 10 ( Fig. 5B). Both the hDlg 1-161(Cys-66) and hDlg 1-161(C66S) segments showed an identical pattern of cross-linked products by SDS-PAGE under reducing conditions (Fig. 5B). At the lowest ratio of cross-linker to protein, both segments are crosslinked into dimer, tetramer, and a small amount of octamer (Fig. 5B, lanes 1 and 5). The dimeric species is the most predominant form followed by the tetramer. Predictably, the hDlg segments cross-linked to higher order oligomers as the concentration of the cross-linker increased (Fig. 5B, lanes 4 and 8).
Identical results were obtained when the baculovirus-expressed segments BV hDlg 1-161(Cys-66) (Fig. 5C, lanes 1-3) and BV hDlg 1-161(C66S) (Fig. 5C, lanes 4 -6) were cross-linked using increasing molar concentration of the amine cross-linker BS 3 , under nonreducing conditions. It should be noted that the mutually exclusive segment hDlg 201-904 did not cross-link into higher order oligomers when exposed to increasing concentration of the cross-linker BS 3 demonstrating the specificity of the cross-linking reaction (Fig. 5C, lanes 7-9). The amine crosslinking experiments further support the sedimentation equilibrium data that the unique amino-terminal domain of hDlg contains the oligomerization site. The hDlg 1-161 segment can self-associate into dimer and tetramer independent of the presence of a single cysteine 66 residue.
The self-association state of the amino-terminal domain of hDlg was examined by native gel electrophoresis in the absence of DTT (Fig. 5D). The hDlg 1-161(Cys-66) segment exists in a dimer-tetramer-octamer equilibrium thus confirming the results obtained by the sedimentation equilibrium centrifugation assay (Fig. 5D, lane 1). In contrast, the hDlg 1-161(C66S) segment migrated predominantly as a tetramer on the native polyacrylamide gel (Fig. 5D, lane 2). Taken together with the chemical cross-linking data, these observations suggest that the single cysteine 66 residue although not essential for the self-association of hDlg 1-161 segment can contribute to the formation of higher order oligomers under appropriate oxidizing conditions.
Mass Spectrometric Analysis of hDlg Proteins-Finally, the self-association state of the amino-terminal domain of hDlg was analyzed by MALDI mass spectrometry. As shown in Fig.  6, two distinct peaks at 20,156 Da and 39,759 Da are detected which correspond to the monomeric and dimeric species of BV hDlg 1-161(Cys-66) segment. Similar results were obtained for the hDlg 1-161(Cys-66) segment expressed in bacteria. It should be noted that it is not feasible to detect higher order oligomers of hDlg 1-161 by MALDI mass spectrometer although relatively weak signals at 59,714 and 79,438 Da representing the trimeric and tetrameric are detected. These results indicate that the unique amino-terminal domain of hDlg self-associates at least into dimeric and tetrameric species in solution. DISCUSSION In this study, we have analyzed hydrodynamic and ultrastructural properties of recombinant hDlg protein and its defined segments that were expressed either in bacteria or in the insect Sf9 cells. Our data indicate that the recombinant hDlg self-associates in solution, and exists in a monomer-dimertetramer equilibrium. The anomalous mobility of full-length hDlg 1-926 on the Superose 12 column prompted us to identify the domain(s) responsible for its elution properties from the gel filtration column. The aberrant migration of hDlg results from a combination of anomalies in its molecular shape as well as the self-association of monomers into higher oligomeric species. Low angle rotary shadow electron microscopy shows that the full-length hDlg 1-926 and its amino-terminal segments hDlg 1-624 and hDlg 1-229 exhibit a highly flexible and an ill defined molecular shape (Fig. 3). Sedimentation velocity data indicate that the amino-terminal segment hDlg 1-161 exists as an elongated tetramer, a finding that may be extended to full-length hDlg  and to other segments of hDlg containing the amino-terminal domain. Interestingly, both the hDlg 1-624 and hDlg 1-229 segments migrate as a large aggregate of similar size as compared with the hDlg 201-904 segment that exhibits a well defined bilobed structure. Moreover, the mobility of the hDlg 1-229 segment on the gel filtration column is similar to the mobility of the hDlg 1-624 segment (Fig. 2 I). These similarities between the hDlg 1-229 and hDlg 1-624 segments, with respect to their shape and size, are rationalized on the basis of sedimentation equilibrium measurements. The sedimentation equilibrium results indicate that the hDlg 1-624 segment exists in a monomer-dimer-tetramer equilibrium, whereas the hDlg 1-229 segment forms higher order equilibrium of dimer-tetramer-octamer. The difference between the degree of oligomerization of hDlg 1-624 and hDlg 1-229 segments could result from the spatial restrictions imposed by the three PDZ domains and the SH3 motif present in the hDlg 1-624 segment. Furthermore, a comparison of the hydrodynamic properties of hDlg 1-229 and hDlg 1-161 segments underscores the importance of the proline-rich insertion-1 sequence in contributing to the overall shape and mobility of full-length hDlg  .
The results presented in this article show that the aminoterminal domain of hDlg contains the oligomerization site. This unique domain of hDlg, encoded by the first 161 amino acids, is also conserved (83% identity) in SAP97, the rat homologue of hDlg. The amino-terminal 65 residues (amino acids 1-65) of SAP97 were recently shown to be required for its subcellular localization and attachment to the epithelial lateral membranes (47). The unique amino-terminal domain of hDlg is not found in other members of the MAGUK family. For example, the erythroid p55 protein contains a non-homologous aminoterminal domain encoded by the amino acids 1-71 preceding the PDZ domain (48). Indeed, the recombinant p55 protein exists as a monomer in solution (data not shown). Similarly, the PSD-95 MAGUK contains an amino-terminal domain encoded by the first 64 amino acids (49). The two cysteines (Cys-3 and Cys-5) located within the amino-terminal domain of PSD-95 were demonstrated to participate in the intermolecular disulfide linkage. The disulfide-mediated multimerization of PSD-95 was shown to play an important role in the clustering of K ϩ channels and NMDA receptor (49). However, recent work shows that both cysteines located within the amino-teminal domain of PSD-95 are palmitoylated and palmitoylation plays a critical role in the membrane localization, channel binding, and clustering of channels by PSD-95 (50,51). This view is challenged by the recent demonstration that the mutation of both Cys-3 and Cys-5 does not affect PSD-95 binding to Kv1.4 channel but prevents its self-association and channel clustering properties (52). Although the basis of these controversial findings remains unresolved, it is clear that the amino-terminal domain of PSD-95 is critical for its channel clustering function in heterologous cells. In contrast to PSD-95, the amino-terminal domain of hDlg contains only a single cysteine residue at position 66 and the amine cross-linking studies and sedimentation equilibrium data demonstrate that cysteine 66 is not required for the self-association of hDlg (Fig.  5). The detection of an octamer in hDlg 1-161(Cys-66) but not in hDlg 1-161(C66S) , by sedimentation equilibrium studies and by native gel electrophoresis under nonreducing conditions, indicates that cysteine 66 may facilitate formation of higher order oligomers under an appropriate redox potential. At this stage, it is not known whether cysteine 66 of hDlg undergoes palmitoylation in vivo.
Like PSD-95 MAGUK, the hDlg been implicated in the clustering and aggregation of K ϩ channels and NMDA receptors. The K ϩ channel has an intrinsic property to form a tetramer (53). An oligomer of hDlg may facilitate aggregation of individual potassium channel subunits into a tetrameric channel or their clustering into larger aggregates (Fig. 7). Although our results have not defined the core oligomerization site within the amino-terminal domain of hDlg, we speculate that a hydrophobic stretch of 25 amino acids (amino acids 37-61) located within the oligomerization domain of hDlg may be critical for the oligomerization process. This hydrophobic sequence of hDlg shows 96% amino acid identity with its rat homologue SAP97 implying that SAP97 may also self-associate by a similar mechanism. Similarly, a stretch of 25 residues (encoded by amino acids 14 -38 of hDlg/SAP97) within the oligomerization domain of hDlg contains a putative leucine zipper sequence (54). The leucine zipper sequences are known to mediate protein dimerization raising the possibility that this region of hDlg may be involved in the dimerization of hDlg/SAP97 (54). A comparison of the oligomerization domain of hDlg with the amino-terminal domain of PSD-95 reveals a stretch of 15 residues (encoded by amino acids 106 -120 of hDlg) that shows 87% amino acid identity. Whether this charged motif plays a role in the oligomerization of hDlg and PSD-95 remains to be determined. A systematic in vitro mutagenesis screen may answer some of these queries.
A number of recent findings indicate that the localization and multimerization mechanisms of PSD-95/Chapsyn-110 are distinct from hDlg/SAP97 MAGUKs. These observations include: (i) PSD-95 forms homodimers as well as heterodimers with Chapsyn-110 but not with SAP97/hDlg (55); (ii) hDlg/ SAP97 is associated with the presynaptic density whereas the PSD-95 and Chapsyn-110 MAGUKs are associated primarily with the postsynaptic density (7,56,57); (iii) hDlg/SAP97 forms large intracellular co-aggregates when co-expressed with Kv1.4 channels in heterologous COS-7 cells (58). In contrast, flat co-clusters are observed on the cell surface when the Kv1.4 channels are co-expressed with Chapsyn-110 and PSD-95 (58). Together, these observations suggest that the hDlg/SAP97 may have evolved a novel mechanism of self-association that is governed by the specificity of its targeting and clustering functions in vivo.
Our results showing the self-association of hDlg are consistent with the multimerization of other MAGUKs. The tight junction associated MAGUKs, ZO-1 and ZO-2, form heterodimers in epithelial cells (59). This association is mediated by the SH3 domain of ZO-1/ZO-2 and the proline-rich sequence of ZO-2/ZO-1, respectively. The Drosophila Dlg tumor suppressor protein localizes to the septate and neuromuscular junctions (6,24). At these junctions, the Dlg protein interacts with Shaker-type K ϩ channels, NMDA receptors, and Fasciclin II and has been implicated in the aggregation and clustering of these transmembrane proteins (24 -26). Since the hDlg/SAP97 can functionally substitute Dlg and rescues the dlg mutant phenotype, it is speculated that the physiologically active Drosophila Dlg may require an oligomerization step to function in vivo (60). Indeed, genetic studies indicate that the Dlg functions as a multimer and that the SH3 domain of Dlg may contribute to its multimerization (61).
The dimer-tetramer equilibrium of hDlg may be relevant to the localization and function of the APC colon tumor suppressor protein. APC exists as a dimer and interacts with hDlg at the sites of cell to cell contact and neuronal synapses thus modulating signaling pathways affecting neuronal function, cell proliferation, and cell motility pathways (33,34). The carboxyl-terminal 15 amino acids of APC interact with the PDZ2 domain of hDlg (33). The PDZ2 domain of hDlg is also the binding site for K ϩ channels and NMDA receptors (5,23,62). Indeed, the APC, K ϩ channels, and NMDA receptors localize in the terminal plexus of the basket cell fibers suggesting that APC, K ϩ channel, and NMDA receptors compete for binding to hDlg at these sites (63). The oligomerization of hDlg may allow simultaneous interaction with APC, K ϩ channels, and NMDA receptors. In principle, the tetrameric hDlg contributing four PDZ1ϩ2 modules can anchor a combination of PDZ domainbinding proteins. This networking arrangement may allow modulation of the clustering of channels and receptors, and synaptic transmission between basket cell fibers and Purkinje cells. The oligomerization of hDlg may also facilitate clustering of APC along the axis of microtubule extensions and maintenance of a larger protein complex essential for neurite outgrowth and cell motility (64).
Previously, we have shown that the PDZ3 domain of hDlg crystallizes as a dimer and in solution the PDZ3 domain can self-associate with millimolar affinity (18). Consistent with this observation is the fact that the hDlg 201-904 segment migrates as a dimer on the Superose 12 column. Whether domains other than the oligomerization domain identified in this study play a role in the multimerization of hDlg is currently unknown. The possibility that the multimerization of hDlg is regulated in vivo remains to be investigated. Detailed structural analyses may help to define the precise role of various domains in the multimerization of hDlg. In conclusion, the results presented in this paper suggest that each individual hDlg molecule in an oligomer can bind to a distinct set of structural and signaling molecules thus resulting in a large signal transduction complex at the interface of membrane-cytoskeleton. The formation of this macromolecular complex may be central to the function of hDlg in vivo.