To elucidate the function of cell polarity proteins in mammalian epithelial cells, we searched biochemically for proteins interacting with hDlg1 (human Dlg1/SAP97 (synapse-associated protein 97, A). hDlg1 protein-containing complexes were immuno-affinity-purified from MCF7 breast epithelial tumor cell extracts using a specific antibody (Supplementary Figure S1A in the Supplementary Data) and fractionated by gel electrophoresis. Two coprecipitating proteins of 110 and 65 kDa were visible on gels, in addition to a triplet of bands at the size expected for hDlg1 splice variants (B, lane 1). Mass spectrometric analyses confirmed the identity of the ~140-kDa bands as hDlg1 splice variants and revealed that the other two proteins were CASK and MPP7, respectively (B, lane 1, and Supplementary Figure S2). CASK and MPP7 were not present when immunoprecipitation was performed with a control antibody (B, lane 2). The identity of these proteins was confirmed by coimmunoprecipitation and Western blot analysis with specific antibodies (C, lane 1, and Supplementary Figure S1A).
hDlg1, CASK, and MPP7 are all members of the MAGUK family of proteins, the defining features of which are at least one PDZ domain, followed by an SH3 domain and a catalytically inactive guanylate kinase domain (Funke et al., 2005
). CASK contains an additional amino-terminal region homologous to calcium/calmodulin-dependent protein kinase II and has previously been shown to interact with hDlg1 (Nix et al., 2000
; Lee et al., 2002
). MPP7 is a so far uncharacterized member of the p55 subfamily of MAGUKs (Funke et al., 2005
). Importantly, we could recover both proteins associated with hDlg1 in immunoprecipitates from a wide range of epithelial cell lines derived from different tissues (C, lanes 2–6) independently of whether the cells were polarized or not (data not shown).
To characterize further the hDlg1 complexes, we first examined whether the interaction between hDlg1, MPP7, and CASK represented one or more distinct complexes. Size exclusion chromatographic fractionation of MCF7 lysates followed by Western blotting revealed that cellular hDlg1 elutes at a size of ~500–600 kDa, suggesting one or more multimolecular complexes (D, top). A fraction of MPP7 and CASK also eluted at a similar position to hDlg1 (D, middle and bottom), though the majority of the cellular pool of these two proteins eluted elsewhere. Several MPP7 splice variants eluted mainly at around 200 kDa, whereas the bulk of CASK eluted as a 300-kDa complex (D, middle and bottom). The same comigration pattern of hDlg1, MPP7, and CASK was obtained using cell extracts of unpolarized (i.e., in low calcium) MCF7 cells (data not shown). Interestingly, the elution profile of CASK and MPP7 in the hDlg1 region was slightly different, suggesting the existence of two distinct hDlg1 complexes.
To examine this further, myc-tagged MPP7 or VSVG-tagged CASK were ectopically expressed in HEK 293T cells. Immunoprecipitation with an anti-myc antibody followed by Western blot analysis revealed an interaction between myc-MPP7 and endogenous hDlg1, but not between myc-MPP7 and endogenous CASK (E, lane 3). In contrast, anti-VSVG antibodies coprecipitated VSVG-CASK and endogenous hDlg1, but not endogenous MPP7 (E, lane 8). Moreover, in the presence of myc-MPP7, the amount of endogenous CASK associated with endogenous hDlg1 was reduced, whereas in the presence of VSVG-CASK significantly less endogenous MPP7 could be detected in an hDlg1 immunoprecipitate (E, lanes 2, 7, and 11). We conclude that CASK and MPP7 compete for the same or overlapping binding sites on hDlg1.
A previous report demonstrated that the interaction between hDlg1 and CASK is mediated through L27 domains present on both proteins (Lee et al., 2002
). To map the interaction sites of MPP7 with hDlg1, myc-tagged MPP7 deletion constructs were expressed in HEK 293T cells, and anti-myc immunoprecipitations probed on Western blots for endogenous hDlg1. Full-length myc-MPP7 could readily be coprecipitated with endogenous hDlg1 and vice versa (F, lane 1, and data not shown); however, removal of the L27N domain of MPP7 abolished the interaction (F, lane 2), indicating that this protein motif is required for binding to hDlg1. In agreement with this, a fragment comprising the L27N and L27C domains only of MPP7 was sufficient to specifically associate with endogenous hDlg1 (F, lane 8). These results point to an interaction between the L27 domain of hDlg1 and the L27 domains of MPP7. To test this, key hydrophobic residues were substituted with a polar hydrophilic residue within the L27N (L38S) or L27C (L95S) domains of MPP7. myc-MPP7 and myc-MPP7/L95S coprecipitated with endogenous hDlg1 when expressed in HEK 293T cells using either myc or hDlg1 antibodies (G, lanes 1 and 3), whereas myc-MPP7/L38S did not (G, lane 2). We conclude that hDlg1 exists in two discrete complexes with either MPP7 or CASK in epithelial cells and that both proteins use L27 domains to interact with hDlg1.
Because hDlg1 forms two separate protein complexes with MPP7 and CASK, we made use of Caco-2 cells to examine their subcellular localization. Unlike MCF7 cells, Caco-2 cells are capable of establishing a highly polarized morphology in tissue culture conditions. We raised an antibody to MPP7 for immunofluorescence studies and found that endogenous MPP7 associates primarily with lateral membranes and is largely excluded from the basal site in polarized Caco-2 cells (, A and B). hDlg1 is both lateral and basal and therefore colocalizes with MPP7 along the lateral side (A). In contrast, the basolateral protein CASK only partially overlaps with MPP7 at the lateral site (B), but colocalizes with hDlg1 along the basal membrane (data not shown, Lee et al., 2002
). Interestingly, MPP7 is found slightly more apical along the lateral membrane than hDlg1 (A, right panel, inset, arrowhead). In addition, both MPP7 and hDlg1 overlap with markers for adherens junctions (, C and D) and tight junctions (, E and F). MPP7 staining is specific, because almost no signal is detectable in MPP7 siRNA treated cells (Supplementary Figure S1, B and C).
Figure 2. MPP7 and hDlg1 colocalize at the lateral surface in Caco-2 cells. (A and B) Confocal microscopic X-Y sections near the cell junctions of Caco-2 cells grown to confluence for 14 d, fixed, and stained with antibodies to (A) hDlg1 (green) and MPP7 (red) (more ...)
To examine the potential interdependency for plasma membrane localization of MPP7 and hDlg1, we returned to a more easily transfectable epithelial cell line, MCF7. First, it has previously been reported that the association of hDlg1 with the plasma membrane is dependent on the cortical actin cytoskeleton (Reuver and Garner, 1998
). Treatment of MCF7 cells with cytochalasin D for 1 h to disassemble the actin cytoskeleton induced a cytoplasmic relocalization of both endogenous hDlg1 and MPP7 (G). Next, a series of myc-tagged MPP7 deletion constructs were introduced into MCF7 cells and analyzed by immunofluorescence microscopy with anti-myc antibodies (A and Supplementary Figure S3). Full-length myc-MPP7 localized to the plasma membrane similarly to endogenous protein. However, deletion of the amino-terminal, hDlg1-interacting L27N domain abolished membrane localization of MPP7 (A and Supplementary Figure S3). Surprisingly, further analysis of deletion constructs revealed a second domain required for membrane localization of MPP7. A mutant lacking the SH3-HOOK domain also did not localize to the plasma membrane despite containing the L27N domain (A and Supplementary Figure S3). We conclude that both the hDlg1-interacting L27N domain and the SH3-HOOK regions are required for targeting of MPP7 to the plasma membrane.
Figure 3. hDlg1 is required for plasma membrane localization of MPP7. (A) Schematic representation of the different myc-MPP7 constructs used to examine their ability to bind to hDlg1 in HEK 293T cells (first column), to target to the plasma membrane in MCF7 cells (more ...)
To examine whether the interaction with hDlg1 is required for membrane localization, gene silencing by siRNA was used in MCF7 cells. Immunofluorescence microscopy and immunoblotting demonstrated efficient and uniform hDlg1 depletion (, B and C). In the absence of hDlg1, endogenous MPP7 no longer localizes to the plasma membrane (B, middle row), although strong membrane staining can be seen in control cells and in untransfected cells (B, top row). Components of adherens junctions, such as E-cadherin, and tight junction markers, such as occludin, appeared to localize normally at the plasma membrane in the absence of hDlg1 (D; data not shown). In contrast, depletion of MPP7 by siRNA did not interfere with the membrane localization of hDlg1 (B, bottom). Thus, the recruitment of MPP7 to the plasma membrane is dependent on hDlg1.
Our observation that the SH3-HOOK domain in MPP7 functions as an additional membrane-targeting domain prompted us to investigate the underlying mechanism. We hypothesized that proteins other than hDlg1 might function through the SH3-HOOK domain. In fact, although myc-MPP7 similarly localized to the plasma membrane of several cell types, we noticed that it did not localize to the membrane when expressed in MCF10A mammary breast epithelial cells, despite high level expression and normal plasma membrane localization of endogenous hDlg1 in these cells (A). A recent report showed that MCF10A fail to form tight junctions, because they express very low levels of the apical polarity transmembrane determinant, Crb3 (Fogg et al., 2005
). To examine whether Crb3 might also affect MPP7 localization, FLAG-tagged Crb3 together with myc-MPP7 were cotransfected into MCF10A cells and analyzed by immunofluorescence microscopy. In Crb3-expressing cells, myc-MPP7 was recruited to the plasma membrane (C, first column), whereas it remained cytosolic when expressed with a Crb3 construct lacking the C-terminal PDZ-binding motif (, B and C, second column). Importantly, the localization of endogenous hDlg1 remained unaffected in MCF10A cells expressing FLAG-Crb3-ΔERLI, illustrating a requirement of Crb3 in plasma membrane localization of MPP7 (D). myc-MPP3, the closest homolog of MPP7 in the p55 MAGUK family, was not recruited to the plasma membrane in a Crb3-dependent manner in MCF10A cells, demonstrating a specific effect of Crb3 on MPP7 (E, compare first and second columns). As shown previously, both the L27N and the SH3-HOOK domains are required for membrane targeting of MPP7 (A, second column). In agreement, transfection of deletion constructs of MPP7 (A, third column) with full-length Crb3 revealed that indeed the SH3-HOOK domain of MPP7 is required for Crb3-mediated membrane targeting (data not shown). In summary, these data suggest that plasma membrane localization of MPP7 is dependent on its L27N domain-mediated binding with hDlg1 and on a Crb3-dependent recruitment via the SH3-HOOK domain.
Figure 4. Crb3 recruits MPP7 to the plasma membrane. (A) MCF7 and MCF10A cells were transfected with myc-MPP7, fixed 48 h after transfection, and stained with anti-myc antibodies (top panels). Note the absence of plasma membrane staining in MCF10A cells. MCF7 and (more ...)
To provide further evidence for an interaction between MPP7 and Crb3, HEK 293T cells were transfected with myc-tagged MPP7 and FLAG-tagged wild-type and mutant forms of Crb3. Myc-MPP7 was not detected in immune precipitates with the PDZ-binding-site mutant of Crb3, whereas it readily coprecipitated with the wild-type, the FERM-, and the RP-motif mutants (A). Previous studies have shown that the PDZ-binding motif of Crb3 binds to PALS1/MPP5, another member of the p55 subfamily of MAGUK proteins (Bachmann et al., 2001
; Hong et al., 2001
; Roh et al., 2002b
). To examine whether the interaction between MPP7 and Crb3 is directly or indirectly mediated by PALS1 (MPP5), wild-type and deletion constructs of MPP7 were examined for their ability to interact with MPP5. HEK 293T cells were cotransfected with myc-tagged MPP7 and FLAG-tagged MPP5, and, as shown in B (lane 3), myc-MPP7 can be coprecipitated with FLAG-MPP5 and vice versa. Importantly, MPP3, another member of the p55 subfamily of MAGUKs, did not interact with myc-MPP7, suggesting a specific interaction between MPP7 and PALS1 (MPP5; B, lane 2). Moreover, PALS1 (MPP5) associated much more efficiently with Crb3 than MPP7 (C, compare lanes 5 and 7), supporting the idea that MPP7 binding to Crb3 is indirectly mediated via its interaction with PALS1 (MPP5). Importantly, the interaction with PALS1 (MPP5) required the SH3-HOOK domain, but not the PDZ domain of MPP7 (A, fourth column, and 5D, compare lanes 4, 6, and 7), excluding a potential unspecific interaction with the PDZ-binding motif of Crb3.
Figure 5. MPP7 binding to Crb3 is indirectly mediated via PALS1 (MPP5). (A) Immunoprecipitations were performed using FLAG antibodies from HEK 293T cells cotransfected with myc-MPP7 and FLAG-tagged Crb3 constructs as indicated. Samples of the lysates and FLAG IPs (more ...)
Because hDlg1 and MPP7 show overlapping localization to tight junctions, we directly investigated whether these proteins are required for the functional assembly of epithelial tight junctions. To examine this, pools of Caco-2 cells stably depleted of either hDlg1 or MPP7 by infection with retroviral-mediated shRNA were established. Western blot analysis confirmed that protein expression of the two proteins was attenuated (A). To assess tight junction function, we measured transepithelial electrical resistance (TER) as a quantitative readout for tight junction integrity in combination with a calcium-switch assay. hDlg1 and MPP7-depleted Caco-2 cells were plated on collagen-coated filters and grown to confluence over 5 d to establish polarity. The medium was then replaced with a low-calcium medium, and the cells were left overnight to disrupt cell–cell contacts. After this time, normal growth medium was readded and TER was measured over a 3-d time course. Control cells rapidly reformed tight junctions within 10–15 h after returning to normal growth medium (Supplementary Figure S4, A and B), whereas there was a delay in the formation of tight junctions in the hDlg1-depleted cell lines (B and Supplementary Figure S4A). shRNA-mediated elimination of MPP7 led to a significant retardation of the development of TER and the extent of TER inhibition correlated with the degree of MPP7 suppression (, A and C, and Supplementary Figure S4B). Moreover, the start and end point TER-values remained significantly lower than of control cells, suggesting that MPP7 may also be involved in tight junction maintenance. To examine whether the defects observed in Caco-2 cells lacking hDlg1 or MPP7 correlated with improper localization of cell junction proteins, cells were fixed and stained after the calcium-switch assays. In low-calcium medium, cell–cell junctions were completely disrupted as shown in D (t = 0 h). However, we found no obvious mislocalization of tight junction markers, such as occludin and ZO-1 (D and data not shown), or adherens junction markers, such as E-cadherin and β-catenin (data not shown).
Figure 6. hDlg1 and MPP7 are necessary for tight junction integrity. (A) Lysates from Caco-2 control-shRNA and two independent hDlg1-shRNA (hDlg1-A and -B) and two independent MPP7-shRNA (MPP7-A and -B) expressing pools of cells were analyzed by Western blotting (more ...)