Targeting CAL as a Negative Regulator of ΔF508-CFTR Cell-Surface Expression

PDZ domains are ubiquitous peptide-binding modules that mediate protein-protein interactions in a wide variety of intracellular trafficking and localization processes. These include the pathways that regulate the membrane trafficking and endocytic recycling of the cystic fibrosis transmembrane conductance regulator (CFTR), an epithelial chloride channel mutated in patients with cystic fibrosis. Correspondingly, a number of PDZ proteins have now been identified that directly or indirectly interact with the C terminus of CFTR. One of these is CAL, whose overexpression in heterologous cells directs the lysosomal degradation of WT-CFTR in a dose-dependent fashion and reduces the amount of CFTR found at the cell surface. Here, we show that RNA interference targeting endogenous CAL specifically increases cell-surface expression of the disease-associated ΔF508-CFTR mutant and thus enhances transepithelial chloride currents in a polarized human patient bronchial epithelial cell line. We have reconstituted the CAL-CFTR interaction in vitro from purified components, demonstrating for the first time that the binding is direct and allowing us to characterize its components biochemically and biophysically. To test the hypothesis that inhibition of the binding site could also reverse CAL-mediated suppression of CFTR, a three-dimensional homology model of the CAL·CFTR complex was constructed and used to generate a CAL mutant whose binding pocket is correctly folded but has lost its ability to bind CFTR. Although produced at the same levels as wild-type protein, the mutant does not affect CFTR expression levels. Taken together, our data establish CAL as a candidate therapeutic target for correction of post-maturational trafficking defects in cystic fibrosis.

Overexpression of CAL in heterologous cells leads to a dramatic decrease in the plasma-membrane levels of CFTR (21) and of several other membrane proteins that are known to interact with it, including Clc-3 chloride channels, the ␤ 1 adrenergic receptor, and the somatostatin receptor subtype 5 (22)(23)(24). In the case of CFTR, the effect is mediated by reductions in the rate of membrane insertion and in the half-life of the channels at the cell surface (21), and can be prevented by blocking endocytosis or lysosomal degradation (25). The negative effect of CAL overexpression on CFTR expression levels can also be reversed by the simultaneous overexpression of NHERF1, which competes for the C-terminal TRL binding motif (21), or by overexpression of TC10, a Rho GTPase whose constitutively active form redistributes CAL intracellularly toward the plasma membrane (26).
It thus appears that CAL plays an important role in the intracellular trafficking and localization of CFTR. Furthermore, because high levels of CAL reduce CFTR levels, it is possible that endogenous CAL acts as a negative regulator. If so, targeted modulation of the CAL-CFTR interaction could provide a mechanism for up-regulating CFTR trafficking in a therapeutic context, in analogy to the rescue of ⌬F508-CFTR seen upon overexpression of NHERF1 (27). However, previous studies have focused on the effects of CAL overexpression on WT-CFTR. As a result, no evidence has been available as to whether endogenous CAL is limiting for CFTR expression nor whether its effects apply to disease-associated mutants. Furthermore, analysis of the regulatory interactions has so far been confined to heterologous cells, even though trafficking pathways depend strongly on cellular context (28 -30).
In the experiments reported here, we test the hypothesis that suppression of endogenous CAL expression levels will increase the cell-surface expression of functional ⌬F508-CFTR and that it will do so in a polarized human bronchial epithelial cell line. In addition, we assess the ability of a localized mutational knock-out of the CAL PDZ binding pocket to abrogate CALmediated suppression of cell-surface CFTR, providing new insights into the mechanism of interaction. Taken together, our results establish the potential therapeutic relevance of pharmaceutical inhibition of the CAL PDZ binding domain.

EXPERIMENTAL PROCEDURES
siRNA-mediated Targeting of Endogenous CAL Expression-CFBE41o-cells (31,32) stably transduced with the ⌬F508-CFTR gene under control of a cytomegalovirus promoter ("CFBEϩ⌬F508" cells) (33) were a generous gift of Dr. J. P. Clancy (University of Alabama, Birmingham) and were maintained in the Dartmouth CF Core Facility. Monolayers of CFBEϩ⌬F508 cells were grown in 6-well plates and transfected with 160 nM CAL-specific siRNA (GOPC3; Qiagen) or nonspecific siRNA (control, non-silencing siRNA; Qiagen) or an equal volume of medium, using the transfection reagent Lipofectamine 2000 (Invitrogen). After 20 h, cells were provided with fresh medium. To measure cell-surface CFTR, 72 h after transfection, cells were washed with ice-cold phosphate-buffered saline (Invitrogen), incubated with EZ-Link Sulfo-NHS-LC-Biotin (Pierce; 1 mg/ml in phosphate-buffered saline with 1 mM MgCl 2 , 0.1 mM CaCl 2 , pH 8.2) for 1 h at 4°C, washed, lysed in lysis buffer (25 mM HEPES, pH 8.2, 1% (v/v) Triton X-100, 10% (v/v) glycerol, 1 Complete tablet/50 ml (Roche)), collected using a cell scraper (Sarstedt), and centrifuged. An aliquot of clarified whole-cell lysate (WCL) was subjected to SDS-PAGE and analyzed by Western blotting with CFTR-, CAL-, and ezrin-specific antibodies. The remaining clarified WCL was incubated with streptavidin beads overnight at 4°C, after which the beads were washed three times with lysis buffer. Proteins were eluted in Laemmli sample buffer/dithiothreitol (DTT) at 85°C for 5 min, and resolved by SDS-PAGE. Western blotting was performed with antibodies specific for CFTR, breast cancer resistance protein (BCRP), and the Na ϩ /K ϩ -ATPase ␣1 subunit. Horseradish perioxidase-conjugated secondary antibody (Bio-Rad) and Western Lightning Chemiluminescence Reagent Plus (PerkinElmer Life Sciences) were used for visualization.
For experiments with polarized monolayers, CFBEϩ⌬F508 cells were seeded at low density. For biochemical experiments, 10 5 cells were seeded on 24-mm diameter Transwell filters (Corning) and allowed to grow for 3 days prior to transfection. For electrophysiological experiments 3.3 ϫ 10 4 cells were seeded on 12-mm diameter Snapwell filters (Corning), and allowed to grow for 4 days prior to transfection. In both cases, subconfluent monolayers were transfected overnight with 50 nM CAL-specific or nonspecific siRNA (GOPC3 or control, non-silencing siRNA, respectively; Qiagen), using HiPerFect transfection reagent (Qiagen) according to the manufacturer's protocol. Confluent monolayers were allowed to form, and cells were serum-starved for 24 h, and switched to 27°C for 24 -36 h prior to experimentation to increase signal intensity. Monolayers were apically biotinylated using EZ-Link Sulfo-NHS-LC-Biotin (Pierce), and WCL and surface-biotinylated samples were prepared and analyzed as described above for non-polarized cells.
Recombinant Protein Expression Vectors-Full-length human CAL (GenBank TM accession AF450008; TrEMBL accession number Q969U8) was subcloned into the pET16b expression vector (Novagen) on an NdeI/BamHI fragment generated by PCR to yield the vector pHCAL1. The 5Ј primer was designed to introduce a decahistidine purification tag at the N terminus of the construct. The CAL PDZ domain (amino acids 278 -362) was also PCR subcloned into pET16b as an NdeI/BamHI fragment to yield the vector pHCALP5. Its 5Ј primer was designed to introduce an N-terminal decahistidine tag followed by a TEV protease recognition sequence. CALbinding site mutants were prepared using the QuikChange and Multichange protocols (Stratagene) in the eukaryotic expression vector pECFP-CAL, containing full-length CAL inserted as an EcoRI/BamHI fragment into the pECFP-C1 backbone (Clontech): "CAL-D" ϭ S294D,T296E,K340D,K342E; "CAL-E" ϭ K299D,K340D,K342E; and "CAL-TϩL" ϭ L291E,G292E,I295E,H341F,L348N. Full-length and PDZ domain mutant constructs were subcloned into the bacterial expression vectors described above and into the mammalian expression vector encoding HA-tagged full-length CAL (25).
pGST-CFTRC, encoding C-terminal residues 1377-1480 of CFTR as a glutathione S-transferase (GST) fusion protein in the pGEX-4T-1 vector (GE Healthcare), was obtained from the Dartmouth Cystic Fibrosis Core Facility and was originally a generous gift of Drs. P. Devarajan and A. Swiatecka-Urban. pGST-CFTRC⌬TRL was PCR subcloned as a BamHI/SalI site fragment into pGEX-4T-1. All protein expression constructs were verified by DNA sequencing.
Protein Expression-pHCAL1-transformed BL21(DE3) RIL cells (Novagen) were grown at 37°C in LB medium to an A 600 of ϳ0.6. Protein expression was induced with 0.1 mM isopropyl ␤-D-thiogalactopyranoside and allowed to proceed for 16 h at 20°C. Cells were harvested, resuspended in lysis buffer T (50 mM Tris, pH 8.5, 150 mM NaCl, 10% (w/v) glycerol, 1 mM DTT, 0.1 mM ATP, 25 units/ml benzonase (EMD Biosciences), 2 mM MgCl 2 , supplemented with one EDTA-free Complete tablet per 50 ml) and lysed using a French press. pHCALP5-transformed BL21(DE3) RIL cells were grown at 37°C in 2ϫ YT medium to an A 600 of ϳ0.8. Induction, expression, and lysis conditions were identical to those for full-length CAL, except that the lysis buffer did not contain glycerol. Isotopically labeled CAL PDZ protein was expressed for NMR analysis in 15 N M9 minimal media (including 1ϫ BME vitamins (Sigma), 4 mg/liter thiamine HCl (Sigma), and 1% (w/v) glucose). 15 NH 4 Cl was obtained from Spectra Stable Isotopes or Cambridge Isotope Laboratories. Mutant CAL and CAL-PDZ proteins were expressed under the same conditions as wildtype proteins.
pGEX-4T-1-, pGST-CFTRC-, and pGST-CFTRC⌬TRLtransformed BL21(DE3) cells were grown at 37°C in LB medium to an A 600 of Ն0.6. Protein expression was induced by addition of 0.5 mM isopropyl ␤-D-thiogalactopyranoside and allowed to proceed overnight at 20°C. Cells were harvested, resuspended in lysis buffer ((phosphate-buffered saline (PBS): 137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , pH 7.3), 1 mg/ml lysozyme, 10 g/ml DNase I (Roche), 5 mM DTT, 5 mM MgSO 4 , supplemented with 1 Complete tablet in 50 ml). After incubation for 30 min on ice, the cells were lysed using a French press. Protein Purification-All lysates were clarified by centrifugation at 40,000 rpm in a Ti45 rotor for 1 h at 4°C. Imidazole was added to the CAL-PDZ supernatants to a final concentration of 10 mM before application to a nickel-nitrilotriacetic acid Superflow (Qiagen) column (bed volume 10 ml), which had been pre-equilibrated with 5 column volumes (CV) of TBS-CAL (50 mM Tris, pH 8.5, 150 mM NaCl, 1 mM DTT, 0.1 mM ATP), containing 10 mM imidazole. Following sample application, the column was washed with 10 CV of TBS-CAL containing 10 mM imidazole, and protein was eluted in TBS-CAL with a linear gradient of 10 -400 mM imidazole over 20 CV. Eluates were collected in tubes containing Chelex 100 Molecular Biology grade resin (Bio-Rad). CAL was purified using a similar protocol, except that TBS-CAL was supplemented for metal affinity chromatography with 10% (w/v) glycerol and 0.1% (w/v) Triton X-100. CAL-or CAL-PDZ-containing fractions were pooled, centrifuged at 3700 ϫ g for 10 min, and filtered through a 0.45-m polyvinylidene difluoride filter (Millipore) to remove any residual Chelex resin.
GST-CFTRC fusion proteins were purified by affinity chromatography using glutathione-Sepharose 4 Fast Flow beads (Sigma) (bed volume 12 ml). The column was equilibrated with 3 CV of PBS containing 0.05% (v/v) Tween 20 (ICN; PBS/ Tween). Following sample application, the column was washed with 5 CV of PBS/Tween and the fusion protein eluted with 4 CV of PBS containing 25 mM glutathione.
Pull-down Binding Assay-Pull-down experiments were carried out by directly mixing the two proteins under a given interaction condition described below. 500 l of glutathione-Sepharose bead slurry (Sigma) was aliquoted into an Eppendorf tube. After a brief centrifugation (1,000 ϫ g; 1 min), the liquid above the beads was carefully aspirated. The beads were equilibrated twice with 1 ml each of PBS/Tween. An aliquot containing 200 g of GST or GST fusion protein was added (after removal of residual glutathione using a PD10 desalting column; GE Healthcare), and the volume adjusted to 1 ml with the same buffer. The mixture was incubated on ice for 1 h with shaking every 10 min to permit GST capture. After centrifuging the tubes for 5 min at 1000 ϫ g, unbound material was discarded, and the beads were washed thoroughly. An aliquot containing 200 g of CAL or CAL-PDZ proteins was added to the captured GST or GST fusion protein, and the volume adjusted to 1 ml.
The interaction was allowed to proceed for 1 h on ice with shaking every 10 min. After complex formation was completed, to remove unbound protein, the beads were repeatedly washed until the supernatant contained no protein as detected using Bradford reagent. The washed beads were resuspended with an equal volume of SDS-PAGE loading buffer, boiled for 3 min at 95°C, and bound proteins visualized by SDS-PAGE followed by Coomassie Brilliant Blue R-250 staining.
Homology Modeling-Homology modeling was performed via the web-based SWISS-MODEL server (38). In one case, the program selected the templates automatically (Protein Data Bank entries 1QAV and 2PDZ, both syntrophin; 1UEZ, KIAA1526 PDZ1; 1UF1, KIAA1526 PDZ2; 1BE9, PSD-95 PDZ3). In a second case, the program was provided with a userdefined template (Protein Data Bank entry 1I92) corresponding to the NHERF1-PDZ1-CFTRC crystal structure (39). The latter model was combined with the CFTR C-terminal peptide contributed by a symmetry-related molecule in the crystal lattice to generate the model of the CAL⅐CFTR complex shown in Fig. 4. Least-squares superpositions were performed using the program LSQKAB (40). Structural representations were prepared using the program MOLSCRIPT (41).
Mass Spectrometry and NMR Analysis of CAL Protein and Mutants-Following SEC purification, wild-type CAL, CAL-D, and CAL-E mutants and the corresponding PDZ domain proteins were subjected to MALDI-TOF analysis in the Dartmouth Molecular Biology & Proteomics Core Facility. The CAL-PDZ-D mutant domain was also subjected to 1 H, 15 N-heteronuclear single quantum correlation spectroscopy (HSQC) analysis, as described (42).

Assays of CFTR Expression in the Presence of CAL-binding Site
Mutants-A GFP-CFTR fusion protein was expressed in African green monkey kidney (COS-7) cells in the presence or absence of wildtype and mutant HA-CAL. Both proteins were detected by Western blotting as previously described (25).

RESULTS
Endogenous CAL Down-regulates ⌬F508-CFTR Cell-Surface Expression-Our previous studies had shown that overexpression of CAL in heterologous cell lines reduces the levels of recombinant WT-CFTR found in whole cell lysates and at the cell surface. This effect could be blocked by the overexpression of NHERF1 together with CAL (21). Recently, overexpression of NHERF1 has been shown to rescue the cell-surface expression of ⌬F508-CFTR in a human bronchial epithelial cell line (27). Given the apparent antagonism of CAL and NHERF1, we suspected that reduction of endogenous CAL expression could provide an alternative mechanism for increasing cell-surface levels of ⌬F508-CFTR.
To test this hypothesis, we investigated a number of commercially available CAL-specific siRNA constructs for their ability to reduce CAL protein levels. Because recent studies have shown that CFTR endocytic and endocytic recycling processes depend strongly on cell type (28 -30), we performed these experiments in an epithelial cell line derived from human airway. Furthermore, because rescue of cell-surface expression of CFTR is therapeutically relevant only for disease-associated mutants, we selected the CFBEϩ⌬F508 cell line. These cells were originally derived from a cystic fibrosis patient homozygous for the ⌬F508-CFTR mutation and have been stably transduced to express increased levels of ⌬F508-CFTR (30,32,33). Previous studies have demonstrated that mature, glycosylated ⌬F508-CFTR is expressed at the apical plasma membrane both in parental CFBE41o-cells and in CFBEϩ⌬F508 cells at 37°C, and shown that significant amounts of ⌬F508-CFTR are rescued by incubation at 27°C in the transduced cell line (30,35). Other studies have shown that the CFBEϩ⌬F508 cells express functional ⌬F508-CFTR chloride channels (34).
A CAL-specific siRNA construct was identified that reproducibly reduced CAL protein expression levels to Ͻ40% of those seen in mock-transfected CFBEϩ⌬F508 cells or in cells transfected with a nonspecific control siRNA (Fig. 1). The CALspecific siRNA had no detectable effect on the expression of ezrin (Fig. 1A), which was used to normalize all protein determinations. CAL knock-down by 65% led to an increase of more than 80% in the amount of cell-surface ⌬F508-CFTR detected FIGURE 1. Targeting of endogenous CAL increases ⌬F508-CFTR expression at the cell surface of a human bronchial epithelial cell line. Monolayers of CFBE41o-cells stably transduced with ⌬F508-CFTR were transfected with CAL-specific siRNA (siCAL), non-silencing control siRNA (siNeg), or transfection reagent only (mock). After 72 h, cells were labeled with biotin and lysed. A, whole cell lysates (CAL, ezrin) or biotinylated proteins (⌬F508-CFTR, BCRP, Na ϩ /K ϩ -ATPase ␣1 subunit) were resolved by SDS-PAGE and visualized by Western blotting. CAL-siRNA treatment decreased CAL expression and increased cell-surface ⌬F508-CFTR expression compared with untreated cells and cells treated with a nonspecific control. Ezrin expression levels were not significantly affected. B, protein expression levels of CAL, biotinylated ⌬F508-CFTR (⌬F508-BT), total ⌬F508-CFTR (⌬F508-WCL), biotinylated BCRP (BCRP-BT), and biotinylated Na ϩ /K ϩ -ATPase ␣1 subunit (Na ϩ /K ϩ -BT) were quantified following Western blotting and normalized with respect to ezrin for mock-transfected cells (white) and for cells transfected with control siRNA (gray) or CALspecific siRNA (black). Mean values are shown Ϯ S.E. (n ϭ 8 for CAL and ⌬F508-CFTR, n ϭ 4 for BCRP and Na ϩ /K ϩ -ATPase), normalized with respect to mock transfected cells for each protein. Knock-down of endogenous CAL expression was accompanied by a statistically highly significant increase in cell-surface levels of ⌬F508-CFTR (**, p Ͻ 0.01), but no statistically significant changes in overall ⌬F508-CFTR or in the cell-surface levels of BCRP or the Na ϩ /K ϩ -ATPase ␣1 subunit. by biotinylation of CFBEϩ⌬F508 cells following siRNA treatment (Fig. 1B). The effect was statistically significant (p Ͻ 0.01). No corresponding change was observed in the total amount of ⌬F508-CFTR found in whole cell lysates (Fig. 1B), presumably reflecting the fact that only a small fraction of total ⌬F508-CFTR is normally expressed at the cell surface.
To test the possibility that the siRNA treatment might have caused a nonspecific increase in membrane trafficking, cellsurface expression of two unrelated proteins was also quantified following biotinylation. Neither BCRP, an ABC transporter up-regulated in some tumors (43), nor the Na ϩ /K ϩ -ATPase ␣1 subunit (44) showed an increase in cell-surface expression following CAL-specific versus mock or nonspecific siRNA treatment ( Fig. 1), indicating that the effect seen for ⌬F508-CFTR was not due to a generalized up-regulation of membrane protein levels.
To determine whether cell-surface ⌬F508-CFTR rescued by CAL-specific siRNA is functional, we transfected subconfluent CFBEϩ⌬F508 cells grown on permeable supports and allowed them to form monolayers. Monolayers were transferred to 27°C prior to analysis to increase release of ⌬F508-CFTR from the endoplasmic reticulum. Biochemical analysis showed 89% knock-down of CAL expression compared with cells treated with a nonspecific control siRNA. This knock-down is greater than that seen in unpolarized cells, (Fig. 2A), and was associated with a correspondingly larger effect on ⌬F508-CFTR. Cell-surface expression of ⌬F508-CFTR was 4.4-fold greater than in control monolayers ( Fig. 2A), compared with 1.8-fold in unpolarized cells (Fig. 1B). The increase in cell-surface protein was also accompanied by a smaller, 2.7-fold increase in total cellular levels of ⌬F508-CFTR ( Fig. 2A), again consistent with the idea that only a fraction of cellular ⌬F508-CFTR is normally present at the cell surface.
Finally, electrophysiological analysis of filter-grown, temperature-rescued CFBEϩ⌬F508 monolayers showed that cells treated with CAL-specific siRNA had more than triple the ⌬F508-CFTR-mediated chloride current compared with monolayers treated with a control siRNA (Fig. 2B). This confirms that CAL knock-down is effective in polarized cells and that the rescued ⌬F508-CFTR channels are functional. The effect of CAL-specific siRNA is seen in addition to the effects of temperature rescue, suggesting that CAL inhibitors may complement therapies aimed at correcting ⌬F508-CFTR biogenesis.
In Vitro Reconstitution Reveals a Direct CAL-CFTR Binding Interaction-Interference with the CAL-CFTR binding interaction could provide a potential alternative to CAL-specific RNA interference as a strategy for the stabilization of ⌬F508-CFTR at the cell surface. However, such an approach requires knowledge of the biochemistry of the interaction, and in particular, whether it involves direct binding of the CAL and CFTR proteins or is mediated by additional proteins acting as adaptors. Previous investigations of the CAL-CFTR interaction have involved studies of co-localization in cells and co-immunoprecipitation from cell extracts (21,25), and thus could not distinguish between these alternatives.
We therefore reconstituted the interaction in vitro using bacterially expressed, purified components, allowing us to charac-terize the interaction under rigorously defined conditions and in the absence of other proteins. Both full-length CAL and its PDZ domain were expressed with N-terminal polyhistidine tags, whereas the CFTR C terminus was expressed as a GST fusion protein (Fig. 3A). All constructs were isolated from bacterial lysates at high purity (Fig. 3B). Pull-down experiments using glutathione-Sepharose beads clearly demonstrated a specific interaction between CAL and the CFTR C terminus in the absence of any other cellular components (Fig. 3C, left-hand  panel, G-CF lane). The protein concentrations used for the pull-down analysis were in the low micromolar range, consistent with the affinities typically observed for PDZ-peptide interactions (13). This provides clear evidence that the CAL and CFTR can interact directly.
To ascertain whether the reconstituted binding interaction reproduces the essential features of the functional interaction observed in cell extracts (21,25), we also assessed the domain/motif requirements of CAL:CFTR binding in vitro. It has been shown in vivo that the CAL PDZ domain is sufficient to mediate the interaction, and that loss of the CFTR C-terminal tripeptide "TRL" dramatically reduces it (21). We therefore FIGURE 2. CAL-specific siRNA increases ⌬F508-CFTR-mediated chloride currents in polarized monolayers. Subconfluent monolayers of CFBEϩ⌬F508 cells were transfected with CAL-specific (siCAL, black) or nonsilencing control (siNeg, white) siRNA. After 96 h, cells had formed confluent monolayers and were analyzed for functional expression of ⌬F508-CFTR. Monolayers were serum starved for 24 h and incubated at 27°C for 24 -36 h prior to analysis. A, cells were labeled with biotin and lysed. Protein levels were visualized, quantified, and normalized in whole cell lysates (CAL, ⌬F508-CFTR, and ezrin) and in the surface-biotinylated fraction (⌬F508-CFTR) as described in the legend to Fig. 1B. Mean values Ϯ S.E. are shown for CAL and for total and surface-biotinylated ⌬F508-CFTR, compared with control cells (n ϭ 3). Knock-down of CAL expression by 89% (*, p Ͻ 0.05) was associated with a statistically very highly significant (***, p Ͻ 0.001) change in the levels of both surface-biotinylated (⌬F508-BT; 4.4-fold control levels) and total (⌬F508-WCL; 2.7-fold) ⌬F508-CFTR. B, filters were placed in an Ussing chamber, and short-circuit currents (I sc ) were determined under conditions of low apical chloride, following application of 50 M amiloride and then 50 M genistein. 5 M CFTR inh -172 was then applied, and the difference versus genistein-stimulated I sc determined. Monolayers treated with CAL-specific siRNA showed a statistically highly significant (**, p Ͻ 0.01) 3.3-fold larger CFTR inh 172-sensitive chloride current (n ϭ 12), than did cells treated with control siRNA (n ϭ 6). expressed and purified a GST-CFTR C-terminal fusion protein, in which the C-terminal class I PDZ binding motif TRL had been deleted (Fig. 3, A and B). As expected, this ⌬TRL construct was unable to pull down purified CAL (Fig. 3C, left-hand panel,  G-⌬T lane). Finally, to confirm that the interaction is mediated by the CAL PDZ domain, the pull-down experiments were repeated with the purified CAL PDZ domain, which exhibited the same specificity for the intact CFTR C terminus as did fulllength protein (Fig. 3C, right-hand panel).
Homology Modeling of the CAL-CFTR Interaction-With a reconstituted binding system in hand, we wished to test the hypothesis that stereochemical interference with the CAL binding pocket could provide an alternative to CAL protein knock-down as a mechanism for increasing the amount of CFTR at the cell surface. As a result, we designed an atomic model of the CAL-CFTR interaction as a basis for site-directed mutagenesis. The crystal structure of the NHERF1 PDZ1 domain had already been determined in complex with a peptide corresponding to the CFTR C terminus (PDB entry 1I92, Ref. 39). We used this as a manually selected template for SWISS-MODEL (38), which first performs a sequence alignment (Fig.  4A), and then uses it to map the target CAL sequence onto the known NHERF1 template structure. The resulting three-dimensional model of CAL-PDZ is shown in Fig. 4, B and C. Because the CAL sequence was mapped onto a ligand-bound structural template, in silico modeling effectively folded it into position around the CFTR ligand. The homology model shown in Fig. 4 provided the first three-dimensional, structural insights into the CAL-CFTR interaction.
In general, homology modeling is challenging at the level of sequence identity between the CAL PDZ domain and the NHERF1 PDZ1 domain (26%), which is close to the threshold for the technique (38). To assess its validity, we considered two main criteria. One involved the stereochemical complementarity of the computationally apposed binding interfaces. The model preserves key features of the C-terminal CFTR binding site in the PDZ domain, including (i) the "GLGF" (in CAL: GLGI) motif that forms a binding site for the ligand carboxylate (purple ribbon in Fig. 4B); (ii) a hydrophobic pocket for the aliphatic C-terminal side chain in the class I PDZ motif (green side chains in Fig. 4C); and (iii) a conserved His side chain that interacts with the Ser/Thr Ϫ2 side chain in the motif (green side chain His 341 in Fig. 4C). In addition, two polar residues Ser 294 and Thr 296 are positioned to interact with the Asp Ϫ3 side chain (blue side chains in Fig. 4C), and a cluster of lysines (red and purple side chains in Fig. 4C) is poised to interact with upstream acidic elements in the CFTR sequence (shown schematically as "EEE" at the lower end of the pocket in Fig. 4B, corresponding to 1472 EEE 1474 ). Overall, the modeled CAL binding site thus provides an excellent stereochemical fit to the CFTR C terminus.
A second test of the likely accuracy of the model was obtained by performing a separate homology modeling procedure, in this case allowing SWISS-MODEL to select templates automatically. The program selected five PDZ domain structures with sequence identities to CAL ranging between 37 and 49%, sufficient to support robust modeling calculations (Fig. 4A). We then compared the resulting model of the CAL domain with that generated from the NHERF1 template. Both agree very well, with a 2.2-Å root mean square difference in C␣ positions. On the basis of these assessments, we proceeded with the design and testing of binding site mutants, as described below.
Design and Biochemical Characterization of CAL Binding-site Mutants-The CAL-CFTR model provided a basis for investigating the role of side chains in binding, using structure-based mutagenesis. If mutations can be found that abrogate the affinity of CAL for CFTR without gross disruption of the protein fold, the physiological effects of blocking the interaction can also be probed. Working from the model shown in Fig. 4, we generated three sets of mutations designed to interfere with the binding interaction. One mutation (CAL-TϩL) targeted the hydrophobic pocket that accommodates the C-terminal Leu side chain in the CFTR ligand, together with the conserved His side chain that interacts with the Thr Ϫ2 side chain (green in Fig. 4C). A second set of mutations (CAL-D) targeted CAL residues thought to interact with Asp Ϫ3 (blue and purple in Fig. 4C). A third set of mutations (CAL-E) targeted a cluster of lysines (red and purple in Fig. 4C) that could be responsible for binding interactions upstream of the canonical binding site, involving, e.g. the triple-glutamate motif shown in Fig. 4B.
All three CAL mutants were generated both as full-length and PDZ domain constructs. All were expressed and purified using metalaffinity chromatography and SEC. During SEC purification, the CAL-TϩL construct eluted in the void volume and exhibited elevated proteolytic susceptibility, suggesting misfolding and aggregation. It was not analyzed further. In contrast, throughout purification, the hydrodynamic behavior and proteolytic stability of CAL-D and CAL-E mutants was similar to that of wildtype protein, and final purity was also comparable (Fig. 5B). They did show somewhat different mobilities in SDS-PAGE gels (Fig. 5B), but FIGURE 4. Homology modeling of the CAL-CFTR binding interaction and design of site-directed binding mutants. A, sequence alignments used to generate homology models. The CAL sequence is shown at the top. Homologous sequences include the syntrophin PDZ1 domain (PDB entries 1QAV and 2PDZ; 49% identity), the KIAA1526 PDZ1 (1UEZ), and PDZ2 (1UF1) domains (40% identity, in each case), the PSD-95 PDZ3 domain (1BE9; 37% identity), and the NHERF1 PDZ1 domain (1I92; 26% identity). Secondary structure elements are shown above the alignment, based on the structure of the NHERF1 domain (39). Strictly conserved residues are highlighted in red. Similar residues are shown in blue. B, a three-dimensional structural model of the CAL PDZ domain is shown in ribbon representation, as determined by modeling based on homology to the NHERF1 PDZ1 domain (39). The "GLGI" motif is shown in purple. The CFTR C-terminal peptide (stick figure colored by atom type) was positioned in its bound conformation relative to the NHERF1 template, revealing the binding pocket interactions shown in C. The CFTR EEE motif (residues 1472-1474) is shown schematically. C, candidate amino acids predicted to interact with Thr Ϫ2 and Leu MALDI-TOF analysis of the purified proteins revealed molar mass differences associated with the mutations that were within 100 Da of the expected values, and we attribute the change in electrophoretic mobility to net charge differences (⌬z ϭ Ϫ6) associated with the mutations themselves.
Pull-down experiments with the mutant CAL proteins (Fig.  5A, left-hand panel) revealed that the CAL-E mutant retained affinity for the CFTR C terminus. In contrast, the CAL-D mutant could no longer be efficiently captured by the immobilized CFTR C terminus (Fig. 5A). The same pattern was observed with the correspondingly mutated CAL-PDZ domains (Fig. 5A, right-hand panel). The contrasting affinities of the CAL-D and CAL-E mutants provide an opportunity for testing whether CAL-mediated down-regulation of CFTR levels correlates with the affinity of the PDZ binding pocket.
The Specific Role of the CAL PDZ Binding Pocket in Reducing CFTR Expression-To exclude the possibility that the CAL-D mutant had lost affinity for CFTR due to a global disruption of the protein fold, we used NMR spectroscopy to confirm that the CAL-D PDZ domain retained its native structure. Using 15 Nlabeled CAL-D protein, a 1 H-15 N HSQC spectrum was obtained, which exhibits a wide distribution of resonances across the spectral field, confirming that the protein adopts a stable fold (Fig. 5C, blue). The pattern of resonances closely resembles that of wild-type CAL (Fig. 5B, red) (42), as well as those of other PDZ domains (45). Peaks in the wild-type spectrum that have shifted in the mutant spectrum (e.g. asterisks in Fig. 5C) are associated with residues in the vicinity of the mutated side chains, as would be expected in the absence of a global conformational disruption. The NMR data confirm that the loss of CFTR binding affinity for the CAL-D mutant is due to the selective and localized disruption of the binding site, whereas the native fold of the protein is preserved. Any functional differences observed are thus attributable to the affinity of the PDZ binding pocket, making CAL-D an excellent probe of the specific role of the PDZ binding interaction in CAL function. This is particularly important given the promiscuous nature of PDZ domain scaffolding interactions, because PDZ adaptor proteins can mediate their effects either by binding to the target protein of interest directly, or by displacing interactions of other adaptor molecules with scaffolding proteins.
To test the hypothesis that the effect of CAL on CFTR protein levels requires direct CAL-CFTR binding, we took advantage of the assay initially used to characterize the interaction. HA-tagged versions of CAL and the CAL-D and CAL-E mutants were individually co-expressed in COS-7 cells transfected with a GFP-CFTR fusion protein (25). The expression of increasing amounts of wild-type CAL significantly reduced the level of mature GFP-CFTR in whole cell lysates in a dose-dependent fashion (Fig. 6A, top panels, p Ͻ 0.05), consistent with previous reports (25). In contrast, the expression of equivalent levels of the CAL-D mutant had no significant effect on GFP-CFTR expression levels (Fig. 6A, middle panel). As a control, we tested the effect of the CAL-E mutant, which shares two of the  15 N-HSQC NMR spectra of the CAL wild-type (red) and CAL-D (blue) PDZ domains show dispersed resonance peaks characteristic of PDZ domains. Peaks in the WT spectrum that lack a corresponding peak in the mutant spectrum (e.g. asterisks) can be assigned to the mutated residues themselves and a few residues in contact with them, as expected due to side chain substitutions. All other residues in the mutant spectrum could be assigned and confirmed an essentially native structure for the mutant domain.
four side chain substitutions present in CAL-D but retains binding affinity for CFTR (Fig. 5A). CAL-E is able to suppress CFTR expression essentially as well as WT-CAL (Fig. 6A, bottom panel, p Ͻ 0.05). CAL WT and mutant expression levels are shown in Fig. 6B as a function of the amount of transfected DNA, confirming a similar dose dependence of expression for all three constructs. Mean CFTR expression levels are shown in Fig. 6C. It is clear that CAL and CAL-E can efficiently suppress CFTR expression, whereas CAL-D cannot. The ability of CAL to negatively regulate CFTR expression thus correlates closely with its in vitro binding affinity.

DISCUSSION
The observation that CAL knock-down can boost the amount of functional ⌬F508-CFTR at the cell surface in a patient-derived bronchial epithelial cell line provides additional evidence for the antagonistic roles of CAL and NHERF1 in regulating CFTR levels. Earlier studies had shown that cooverexpression of NHERF1 could reverse the effects of CAL overexpression on WT-CFTR in a heterologous cell line (21). More recently, it was shown that NHERF1 overexpression could increase cell-surface levels of ⌬F508-CFTR in airway epithelial cell lines (27). If CAL and NHERF1 compete for CFTR binding and mediate opposing effects on its cell-surface expres-sion, we hypothesized that a similar result could be obtained by reducing CAL expression. This prediction was confirmed, and the magnitude of the CAL knock-down effect on cell-surface ⌬F508-CFTR levels in polarized cells (4.4-fold) was even greater than that seen with NHERF1 overexpression (ϳ40% increase) (27).
Our data provide the first direct evidence that endogenous CAL acts to limit cell-surface levels of ⌬F508-CFTR in human airway epithelial cells. As a result, CAL may reinforce the pathophysiology of cystic fibrosis and could hinder therapeutic efforts to restore ⌬F508-CFTR cellsurface expression. In terms of its negative effect on both WT and mutant CFTR cell-surface expression, CAL stands in contrast to many PDZ proteins, which, like NHERF1 (27), tend to favor the trafficking, localization, and clustering of their binding partners in the plasma membrane (13). However, because it is pharmacologically easier to block a deleterious interaction than to stabilize a beneficial one, the unfavorable influence of CAL on CFTR may actually make it a more attractive therapeutic candidate than other, more benign PDZ counterparts. To understand the mechanism of the CAL-CFTR interaction better, we reconstituted it in vitro, and showed that it involves the binding of the two proteins to each other. To probe the stereochemical basis of the interaction, we pursued a structurebased mutagenesis approach. The fact that the CAL-E mutant retained its ability to bind the CFTR C terminus suggests that the cluster of three lysines (red and purple side chains in Fig. 4C) at the N-terminal end of the binding site is not essential for the CAL-CFTR interaction, although we cannot exclude a small modulatory effect on its affinity. Because the two lysines mutated in CAL-D (purple side chains in Fig. 4C) are also mutated in CAL-E, which retains CFTR binding affinity, the key difference appears to be the result of changes to Ser 294 and Thr 296 (blue in Fig. 4C), both of which are predicted to interact with Asp Ϫ3 . Although not part of the canonical class I motif (13,14), this side chain has been seen to play an important role in some PDZ interactions (46). Furthermore, the mutations were designed not only to abrogate potentially favorable contacts in the wild-type binding site, but also to introduce a charge incompatibility with peptides containing a negatively charged side chain at the Ϫ3 position. Regardless of the relative contributions of these two effects, the overall loss of binding affinity associated with the CAL-D mutation is clear, and in stark contrast to the retention of binding function by the CAL-E mutant.
Because the CAL-D mutant domain retains its native threedimensional conformation, its loss of affinity for the CFTR C terminus allowed us to establish the functional importance of the PDZ binding pocket, independent of any gross changes in the tertiary structure of the protein, such as those associated with truncation mutants (21). The highly localized and specific disruption of the CAL-CFTR binding interaction preserves CFTR expression levels in the presence of CAL protein (Fig. 6), presumably by suppressing the degradation of mature CFTR protein (25). This mutagenetic uncoupling mimics that which would be achieved by the design of small-molecule competitive inhibitors specifically tailored to block the CAL PDZ binding site.
Our success in designing a PDZ binding mutant of CAL also serves to validate the homology model. Even though structural templates with higher sequence identity were available in the data base, the model was developed using a particular template with only borderline (26%) identity (38), because that template had been crystallized in the presence of our ligand of interest (39). This allowed us to model the interaction, rather than just the structure of the isolated binding pocket. As the focus of structural biology shifts increasingly from individual proteins to protein-ligand and protein-protein complexes, this dilemma is likely to recur, in which the template structure with the most relevant binding partners may not be the same as that with the highest homology and thus the greatest likelihood of accuracy. Our approach was to generate models using either the most biologically relevant or the most structually plausible templates and to compare them. In the case of CAL, both models were very similar, increasing confidence in the predicted interaction. However, in cases where the models diverge, an alternative strategy could involve least-squares superposition of the highhomology model onto the biologically relevant template. In either situation, perhaps the most important information about plausibility is provided by the stereochemical compatibility of the modeled binding interaction, which was clearly satisfied by the CAL model.
Whereas mutagenesis and functional characterization experiments were underway, we also pursued the three-dimensional structure determination of the CAL PDZ domain by NMR spectroscopy (42). NMR data retrospectively confirmed our homology model, as shown in Fig. 4D, yielding a strong DALI similarity score to the model (z ϭ 10.5, Ref. 47) and an overall 1.9-Å root mean square difference in C␣ positions, excluding two peripheral loops (marked by asterisks, Fig. 4D) whose conformations are flexible. Chemical shift data obtained in the presence of a peptide corresponding to the C terminus of CFTR also confirmed our identification of CAL side chains that interact with the ligand (Fig. 4C and Ref. 42).
These studies provide a detailed molecular basis for future screening and design approaches to identify CAL-specific small molecule inhibitors. CAL inhibitors should prove useful in dissecting the multiple potential trafficking pathways involved in CFTR regulation (21,25,26). Given that CAL acts as a negative regulator of ⌬F508-CFTR cell-surface levels, such compounds could also help to stabilize mutant CFTR expression levels at the apical membranes of lung epithelia. This could provide an important alternative or complementary approach to current efforts aimed at correcting folding defects (48,49), especially because ⌬F508-CFTR trafficking defects include not only inefficient maturation (8), but also a reduced biochemical half-life (30, 50 -53). The potential for such complementarity is underscored by the ability of CAL knock-down to enhance functional, cell-surface expression of ⌬F508-CFTR beyond the levels induced by low temperature rescue alone.