Pleckstrin homology and phosphotyrosine-binding domain-dependent membrane association and tyrosine phosphorylation of Dok-4, an inhibitory adapter molecule expressed in epithelial cells.

Dok-like adapter molecules represent an expanding family of pleckstrin homology (PH) and phosphotyrosine-binding (PTB) domain-containing tyrosine kinase substrates with negative regulatory functions in hematopoietic cell signaling. In a search for nonhematopoietic counterparts to Dok molecules, we identified and characterized Dok-4, a recently cloned member of the family. dok-4 mRNA was strongly expressed in nonhematopoietic organs, particularly the intestine, kidney, and lung, whereas both mRNA and protein were expressed at high levels in cells of epithelial origin. In Caco-2 human colon cancer cells, endogenous Dok-4 underwent tyrosine phosphorylation in response to pervanadate stimulation. In transfected COS cells, Dok-4 was a substrate for the cytosolic tyrosine kinases Src and Fyn as well as for Jak2. Dok-4 could also be phosphorylated by the receptor tyrosine kinase Ret but not by platelet-derived growth factor receptor-beta or IGF-IR. In both mammalian cells and yeast, Dok-4 was constitutively localized at the membrane in a manner that required both its PH and PTB domains. The PH and PTB domains of Dok-4 were also required for tyrosine phosphorylation of Dok-4 by Fyn and Ret. Finally, wild type Dok-4 strongly inhibited activation of Elk-1 induced by either Ret or Fyn. The attenuation of this inhibitory effect by deletion of the PH domain and its restoration by the addition of a myristoylation signal suggested an important role for constitutive membrane localization of Dok-4. In summary, Dok-4 is a constitutively membrane-localized adapter molecule that may function as an inhibitor of tyrosine kinase signaling in epithelial cells.

Increased tyrosine kinase activity is a common mechanism for initiation of intracellular signaling following stimulation of many cell surface receptors. The specificity of the information conveyed by such events is determined in large part by the amino acid context of tyrosine residues accessible for phospho-rylation within the cytoplasmic tail of the involved receptor(s) and by the availability of motif-specific Src homology 2 or phosphotyrosine-binding (PTB) 1 domain-containing partner molecules (1), including adapter molecules. While devoid of enzymatic activity, adapter molecules often contain additional sites for tyrosine phosphorylation and can generate signaling complexes through protein-protein and protein-lipid interactions in a phosphorylation-dependent or -independent manner, leading to signal amplification or diversification (2).
One example of a well characterized adapter molecule involved in tyrosine kinase signaling is the insulin receptor substrate (IRS)-1 molecule. IRS-1 belongs to a family of at least four members with overlapping yet distinct tissue expression patterns (3)(4)(5)(6). IRS molecules are characterized by an aminoterminal pleckstrin homology (PH) domain immediately followed by a PTB domain and a C-terminal region of variable length containing potential docking sites for Src homology 2 domains. IRS-1 interacts with both membrane phospholipids and with the phosphorylated insulin receptor through its PH and PTB domains, respectively (7). Following tyrosine phosphorylation by the insulin receptor, IRS-1 allows the recruitment and activation of phosphatidylinositol 3-kinase through multiple YXXM motifs contained in its sequence. In addition, perhaps because its association with the membrane and the insulin receptor is weak and transient, IRS-1 can translocate from the membrane to other cellular compartments where its signal is propagated (8).
Recently, the Dok family of adapter molecules (including Dok-1, -2, and -3) has emerged as an expanding group of related signaling molecules composed of an amino-terminal tandem of PH and PTB domains reminiscent of IRS molecules. Dok-1 was identified as a major 62-kDa RasGAP-associated phosphoprotein in v-Abl-transformed B cells and in Bcr-Abl-expressing leukemic cells (9,10). Subsequently, Dok-2 was identified as a partner for the interleukin-4 receptor (11) and the angiopoietin receptor Tie-2 (12) and as a substrate of the Src family tyrosine kinase Lyn (13). We and others cloned a third member, Dok-3, as a molecule associating with the tyrosine kinases Abl (14) and Csk (15).
The first three members of the Dok family (in particular Dok-2 and Dok-3) are primarily expressed in hematopoietic cells (11,14,15). dok-1 is also expressed at significant, although reduced levels in nonhematopoietic tissues (9,10,16). These three molecules can serve as substrates for cytosolic tyrosine kinases of the Abl and Src families (15), which allows them to recruit Src homology 2 domain-containing molecules such as RasGAP, Csk, and SHIP-1 (9,15). In addition, Dok-1 and/or Dok-2 may be phosphorylated by a variety of receptor tyrosine kinases (RTK) including macrophage colony-stimulating factor receptor (17), vascular endothelial growth factor receptor (18), insulin receptor (19), epidermal growth factor receptor (20), platelet-derived growth factor receptor (PDGFR) (21), Tie-2 (12), Ret (22), and c-Kit (23), although the last seems to also require Src kinase activity (24). Dok-1 can also be phosphorylated in response to cell adhesion (19), presumably because its PTB domain interacts with various ␤-integrin chains (25). It is believed that IRS and Dok proteins function as docking molecules with regulated membrane-targeting properties (3,21,23,34). Whereas inducible membrane localization may be a common feature of these molecules, the mechanisms through which this occurs may vary. For instance, phosphatidylinositol 3-kinase-generated phosphoinositides seem necessary for membrane targeting and phosphorylation of Dok-1 in response to PDGF stimulation (21), but they are not involved in IRS-1 recruitment to the insulin receptor. Moreover, among the different members of the IRS family, distinct patterns of intracellular distribution have been observed (6). The PH domain of Dok-1 is necessary and sufficient for phosphatidylinositol 3-kinase-mediated membrane translocation (21), and the PH domain of IRS-1 appears equally important in subcellular targeting and function (34,35).
PH domains form a family of highly divergent amino acid sequences with a remarkably conserved tridimensional structure (reviewed in Ref. 36). They are generally believed to bind phospholipids and as such serve as membrane-targeting domains. Some PH domains, such as that of Btk, display specificity for binding products of phosphatidylinositol 3-kinase metabolism (37), which allows their membrane localization to be regulated (38). However, unlike that of Btk, most PH domains studied so far bind phospholipids with low affinity and specificity (39), and the identity of their physiological ligands therefore remains controversial (40). Nevertheless, it has been postulated that either post-translational modification (such as serine/threonine or tyrosine phosphorylation) or oligomerization might enable low affinity PH domain-containing molecules to bind membranes with more avidity (40).
PTB domains are structurally related to PH domains (41,42). Like PH domains, some PTB domains may bind phospholipids (43,44), although they mainly serve as protein-protein interaction modules (1,45). Whereas the typical target peptide initially described for the PTB domains of Shc and IRS-1 is a phosphorylated NPXY sequence (46,47), it has more recently been recognized that many PTB domains can bind peptide sequences in a phosphorylation-independent (48) or in a tyrosine-independent manner (49 -52).
Whereas PH and PTB domains co-exist in IRS and Dok family proteins, a related adapter molecule, FRS2, possesses a myristoylation signal instead of a PH domain (53), presumably bypassing the requirement for PtdIns binding in membrane targeting. In contrast to IRS family members and other RTKassociated adapter molecules, which are generally involved in signal amplification, Dok family molecules have been shown to function primarily as inhibitors of tyrosine kinase signaling (14, 15, 26 -29).
Since Dok molecules had been found to be prominent Src kinase substrates in hematopoietic cells, we hypothesized that additional nonhematopoietic family members might also exist to serve as substrates downstream of these ubiquitously expressed kinases. In the current study, we present evidence that a novel Dok family member, Dok-4, is expressed prominently in epithelial cells and can be phosphorylated by a number of tyrosine kinases, including Src family members. The structural basis of Dok-4 tyrosine phosphorylation was studied as well as its relationship to intracellular localization. Surprisingly, we found that Dok-4 was constitutively membrane-associated in mammalian cells as well as in yeast, through a mechanism that required both its PH and PTB domains. Tyrosine phosphorylation of Dok-4 also required an intact PH and PTB domain. Finally, Dok-4 inhibited the tyrosine kinase-induced activation of the transcription factor Elk-1, suggesting that inhibitory signaling is a general property of Dok family molecules.

MATERIALS AND METHODS
Ribonuclease Protection Assay-RNase protection assays were performed essentially as described previously (15,54) with minor modifications. Radiolabeled antisense RNA probes were prepared by in vitro transcription (Promega kit) of appropriate cDNA templates in the presence of [␣-32 P]UTP (PerkinElmer Life Sciences). The templates for probe synthesis were as follows. For dok-4, pBluescriptSK-D4RPA was prepared by subcloning the EcoRI/XmnI fragment from expressed sequence tag (EST) W85217 (a putative splice variant of dok-4 containing a 62-bp deletion compared with wild-type dok-4) into the EcoRI/SmaI sites of the pBluescriptSK vector. After linearization with EcoRI and transcription with T3 RNA polymerase, this resulted in a probe of ϳ325 nucleotides when undigested. Because of the internal position of its 62-bp deletion (presumably allowing loop formation in the sense RNA transcripts with full protection of some antisense probes), this probe resulted in RNase-protected fragments of 255 and 211 nucleotides when hybridized to full-length dok-4 (data not shown). For dok-5, a template containing nucleotides Ϫ9 to 293 of murine dok-5 was derived from plasmid pPCR-Script-Dok-5 (see below) by BglII/EcoRI digestion and vector religation. After linearization with NotI and transcription with T7 RNA polymerase, the resulting probe had ϳ395 nucleotides when undigested and 302 nucleotides when RNase-protected. A probe for the L32 riboprotein transcript (template from BD Pharmingen; 141 nucleotides undigested, 112 nucleotides protected) was included with each dok-4 and dok-5 hybridization reaction to ensure equal loading. Each sample contained 25 g of total RNA obtained from organs of adult CD-1 mice using the TRIzol reagent (Invitrogen).
Cloning of dok-4 and dok-5-The full-length murine dok-4 cDNA was obtained from EST AA111459 and later cloned independently by reverse transcription (RT)-PCR from mouse kidney RNA to confirm the sequence. The oligonucleotides used to clone dok-4 were AACCATGGC-GACCAATTTCA (4-3) and CAGCTCCTCGAGGCTGTC . The fulllength murine dok-5 cDNA was obtained by RT-PCR cloning using oligonucleotides derived from the consensus sequence of ESTs AK012430 and AU051646 as follows. First, mouse brain total RNA was reverse transcribed using oligonucleotide GAAAATCACACAAATC-CACA . An initial PCR was performed using oligonucleotides AAAGTGGCTGCTGGGCG (6-7) and 6-4 (see above), followed by a second nested PCR with oligonucleotides CTGTCTGGGATGGCTTC-CAA (6-5) and 6-4. All PCR products were blunt-ended and cloned in the SrfI site of the pPCR-Script vector (Stratagene). All constructs were fully sequenced (Sheldon Biotechnology Centre, McGill University).
Cells-COS-1 and 293 cells as well as primary mouse mesangial cells (from Dr. D. Baran) were cultured in high glucose Dulbecco's modified Eagle's medium containing pyridoxine-HCl and sodium pyruvate (Invitrogen) with 10% fetal bovine serum (Invitrogen). Caco-2 human colon cancer cells (from Dr. C. Stanners) were cultured in ␣-minimal essential medium containing ribonucleosides and deoxyribonucleosides (Invitrogen) and 10% fetal bovine serum. Differentiated Caco-2 cells were obtained by keeping fully confluent monolayers in culture for 14 days with medium change every 2-3 days. LLC-PK1 cells (from Dr. J. Orlowski) were grown in RPMI 1640 with 6% fetal bovine serum. MDCK cells (from ATCC) were grown in minimal essential medium containing Earle's salts, NaHCO 3 , nonessential amino acids, and sodium pyruvate. Parental and cytosolic phospholipase A2-transfected rat glomerular epithelial cells (GEC and GEC-cPLA 2 , from Dr. A. Cybulsky) were grown in K1 medium as described previously (55).
Mammalian Expression Constructs for Dok Proteins-The dok-4 cDNA originally cloned in pPCR-Script vector (see above) was subcloned in the pCDNA3.1 mammalian expression vector (Invitrogen). All sequence modifications were performed by PCR using the proofreading Pwo DNA polymerase (Roche Applied Science). Dok-1 and Dok-4 were tagged with a Myc epitope by replacing their stop codon with an inframe Hind III site and cloning in the pCDNA3.1(Ϫ)Myc-His A vector (Invitrogen). A Myc-tagged Dok-5 construct was also created with the same vector after replacing the stop codon with an in-frame KpnI site. The antisense oligonucleotides used for these three constructs were, respectively, as follows: TTAAGCTTGGTGGAACCCTCAGACTT (Dok-1 Myc), TTAAGCTTCTGGGCAGGGGTCTTGG (Dok-4 Myc), and ATATAGGTACCGTGCTCAGACCGGTAGGT (Dok-5 Myc) (restriction sites underlined). The Dok-4 Myc ⌬CT mutant (deletion of amino acids 234 -325) was created with the antisense oligonucleotide CTGTATAA-GCTTCTGCTCGGCGATGGCCAG (D4⌬CT).
Dok-4 ⌬PH (deletion of amino acids 1-99) was generated with the sense oligonucleotide CAACCATGGCGACCAATTTCAACGCAGAAGA-GTGGTACAAG (D4⌬PH). The PCR product was cloned in pPCR-Script with the initiation codon toward the EcoRI site. The EcoRI/PpuMI fragment resulting from this was used to replace the corresponding sequence in pCDNA3.1-Dok-4 Myc. Dok-4 Myc Myr-⌬PH (substitution of amino acids 1-99 of Dok-4 by amino acids 1-6 of FRS2 (53)) was generated with the sense oligonucleotide TATGAATTCACCATGGGCT-CATGTTGTTCGGCAGAAGAGTGGTACAA (EcoRI site underlined), followed by EcoRI/PpuMI exchange as above. Dok-4 Myc ⌬PTB (deletion of amino acids 100 -233) was generated by overlap PCR using the sense oligonucleotide TCAGAGCTGGAGCACAAGCGGGTCCTGCTG and the antisense oligonucleotide GACCCGCTTGTGCTCCAGCTCT-GAGTCACA. The BGHR antisense and T7 sense oligonucleotides (Invitrogen) derived from pCDNA3.1 were used to complete the reactions. The final PCR product was used for an EcoRI/PpuMI exchange as above.
For fusion of Dok molecules to enhanced green fluorescent protein (EGFP), the NheI/HindIII inserts from pCDNA3.1Myc constructs were cloned in-frame in the NheI/HindIII sites of pEGFP-N1 vector (Clontech). Dok-4 PH-EGFP (fusion of amino acids 1-108 of Dok-4 to EGFP) was obtained by PCR amplification with the antisense oligonucleotide TATAAAGCTTGGACAGTGTCTTGATCCA (HindIII site underlined) and cloning of the resulting NheI/HindIII in pEGFP-N1.
COS Cell Tranfections-One day prior to transfection, COS-1 cells were trypsinized and plated on 100-mm culture dishes at a density of ϳ25%. Transfection was performed with LipofectAMINE 2000 reagent (Invitrogen) in a ratio of 3 l per g of plasmid. After overnight incubation, 1 volume of Dulbecco's modified Eagle's medium with 20% fetal bovine serum and 2ϫ penicillin/streptomycin was added. The next morning, cells were washed in PBS, collected by scraping, and lysed in ice-cold immunoprecipitation (IP) buffer. Where applicable, stimulation with IGF-IR was performed with the addition of human IGF-IR (PeproTech) (100 ng/ml) prior to lysis.
Immunoprecipitation and Immunoblotting-Cells were lysed on ice in buffer consisting of 50 mM Tris-HCl (pH 8.0), 2 mM EDTA, 1% Nonidet P-40, 100 mM sodium fluoride, 10 mM sodium pyrosphosphate, 1 mM sodium orthovanadate, 0.1 mM phenylmethylsulfonyl fluoride, 2 g/ml leupeptin, 1 g/ml aprotinin, and 0.7 g/ml pepstatin A. IP was performed by the addition of either 1 or 2 g of affinity-purified antibody or 5 l of Dok-4 antiserum followed by 8 l of protein A-agarose beads (Santa Cruz Biotechnology). For monoclonal antibodies, protein A-agarose was precoupled to 5 g of rabbit anti-mouse IgG (Jackson Immunoresearch). After three washes, immunoprecipitates were eluted in 1ϫ Laemmli buffer, boiled, and resolved on an SDS-polyacrylamide gel. Proteins were transferred to polyvinylidene difluoride membranes (Hybond P; Amersham Biosciences), and immunodetection was performed according to a standard protocol using horseradish peroxidaselabeled secondary antibodies (Amersham Biosciences) and ECL or ECL Plus detection solutions (Amersham Biosciences).
Fluorescence Microscopy-COS-1 cells plated on poly-L-lysine-coated coverslips were transfected with EGFP constructs using Lipo-fectAMINE 2000 and coverslips were processed 1 day after transfection. Alternatively, 293 cells expressing inducible Dok-1-EGFP or Dok-4-EGFP were plated on coated coverslips and processed 24 h after induction with ponasterone A (10 M). Coverslips were washed in PBS, fixed in 4% paraformaldehyde, mounted in PBS/glycerol (1:9 ratio), sealed with nail polish, and immediately viewed under a fluorescence microscope (Nikon) or a confocal microscope (Zeiss).
Membrane Localization Assay in Yeast-The Cdc25␣ yeast strain containing a temperature-sensitive mutation of Cdc25 was obtained from Stratagene. Expression of human Sos fusion proteins in yeast was achieved by cloning the relevant Dok cDNA fragments in the pSOS vector (Stratagene). Since both dok-1 and dok-4 contain a unique NcoI site overlapping their start codon (ccATGg), this could be done by subcloning the relevant NcoI/NotI fragments from pCDNA3.1 directly to the pSOS vector. For pSOS-Dok-4 PH, the NcoI/SalI fragment from pEGFP-Dok-4 PH was subcloned in pSOS. Yeasts were transformed by the standard lithium acetate/heat shock method, selected on leucinedeficient (Yc ϪL) plates, and frozen without delay. Expression of the constructs was confirmed by blotting with anti-SOS, anti-Dok-1, and anti-Dok-4 immunoblotting. To test for membrane localization, an inoculum from a stock of recently thawed yeast was simultaneously spread onto two separate Yc ϪL plates, one of which was incubated at 25°C and the other at 37°C. At least three clones representing each pSOS construct were tested in this manner.
GST Fusion Proteins and in Vitro Binding-To create GST Dok-4 PTB, the sequence coding for amino acids 100 -233 of Dok-4 was amplified by PCR and cloned in the BamHI/EcoRI sites of pGEX2T (Amersham Biosciences) using oligonucleotides TATGGATCCGCAGAAGA-GTGGTACAAG (sense; BamHI site underlined) and TATGGATCCGC-GACCAATTTCAACGAC (antisense; EcoRI site underlined). To create GST Dok-1 PTB, the sequence coding for amino acids 107-260 of Dok-1 was amplified by PCR and cloned in the SmaI site of pGEX2T using oligonucleotides ATACCCGGGTAGCACCGCCTGGGTGCAG (sense) and AATCCCGGGTCACTGGCCCACCTTTCCTTG (antisense; SmaI sites underlined). The fusion proteins were induced in BL21 Escherichia coli by the addition of 1 mM isopropyl-1-thio-␤-D-galactopyranoside for 2 h. After lysis and sonication in IP buffer containing protease and phosphatase inhibitors, GST or GST fusion protein was coupled to glutathione-agarose beads (Sigma) in the presence of 5 mM dithiothreitol. Protein concentration was estimated by resolving aliquots on SDSpolyacrylamide gels and staining with Coomassie Blue. An in vitro binding assay was performed by incubating lysates of transiently transfected COS cells with 25 l of GST-or GST fusion protein-coupled beads for 1.5 h at 4°C in the presence of 5 mM dithiothreitol. After three washes, proteins were eluted in 1ϫ Laemmli buffer, boiled, and loaded onto SDS-polyacrylamide gels. Immunoblots were carried out as described above.
Reporter Gene Assays-293 and Caco-2 cells were transiently transfected in triplicates on 24-well plates using Fugene 6 and Lipo-fectAMINE 2000 reagents, respectively. In addition to Dok-4 and/or kinase expression constructs, the pFR-Luc vector (Gal-4/luciferase) was transfected in combination with either pFA2-Elk-1 (Elk-1/Gal-4 DNAbinding domain) or pFA2-c-Jun (c-Jun/Gal-4 dbd) (all from Stratagene). Transfection with pFR-Luc and pFC-dbd (Gal4 dbd) was used to determine background transactivation. The pRL-TK Renilla luciferase vector (Promega) was included for normalization. Empty vector was added where necessary to equalize the amounts of transfected DNA. Unless otherwise specified, stimulation of Ret with its ligand was performed by the addition of glial cell line-derived neurotrophic factor (GDNF) (100 ng/ml, from PeproTech) and GFR␣-1-Fc (800 ng/ml, from R & D Systems) 24 h after transfection. Cells were lysed 48 h after transfection, and luciferase activity was determined using the Dual-Luciferase Reporter system (Promega) on a Lumat LB 9507 luminometer (Berthold). Reporter gene activity was expressed as the ratio of firefly to Renilla luciferase activity Ϯ S.E.

Dok-4 Is a Ubiquitous Member of the Dok Family Prominently Expressed in Epithelial Cells and Tissues-To identify
novel Dok family members expressed in nonhematopoietic cells, we searched the EST data base for new molecules with homology to the PTB domain of Dok-1. We found several ESTs representing at least two novel Dok-like molecules in both humans and mice (see "Materials and Methods") (data not shown). Two of these novel molecules were recently re-ported by Grimm et al. and given the names Dok-4 and Dok-5 (56). Using available ESTs as templates for radiolabeled probe synthesis, we examined the tissue distribution pattern of these two novel dok family members by RNase protection assay. As shown in Fig. 1A, dok-4 was expressed almost ubiquitously in mouse organs, with especially high levels in the small intestine, kidney, heart, and lung; lower expression in muscle and brain; but virtually undetectable levels in liver and spleen. The appearance of two distinct dok-4 bands resulted from our use of a variant EST as probe template (see "Materials and Methods"). Whereas this EST was initially thought to represent a splice variant, closer examination suggested a likely cloning artifact. Indeed, hybridization with dok-4 alone could produce both the larger and smaller RNase protected bands (data not shown). In contrast to dok-4, dok-5 was expressed almost exclusively in the central nervous system with faint expression detected in heart and muscle as well (Fig. 1B).
In order to clone the dok-4 and dok-5 cDNAs, we designed oligonucleotides based on the consensus sequence of all published murine ESTs and used them for RT-PCR as described under "Materials and Methods." Full-length dok-4 and dok-5 cDNAs were isolated from mouse kidney and brain RNA, respectively. In addition, an identical full-length dok-4 cDNA was unexpectedly obtained from a commercially available EST clone (accession number AA111459). Like other Dok family members, Dok-4 and Dok-5 are composed of tandem aminoterminal PH and PTB domains and a unique C-terminal region with no significant homology to known molecules. During the course of our studies, both molecules were identified by Grimm et al. (56) as partners and substrates for the Ret tyrosine kinase, the receptor for GDNF. Among all five members of the Dok family, the amino acid sequence of the PTB domain is highly conserved (56). However, the PH domains of Dok-4 and Dok-5 are in fact slightly closer in primary sequence to IRS-1 and SKAP55 than to Dok-1 (Fig. 2). Grimm et al. (56) demonstrated by in situ hybridization that there was very little overlap of ret and dok-4 expression in mouse embryos. In contrast, they reported that the developing vasculature was a prominent site of dok-4 mRNA expression and that the endothelial cell-specific receptor Tie-2 could associate with Dok-4 in transfected cells. On this basis, it was proposed that Dok-4 is primarily an endothelial molecule, which might account for its expression in multiple organs. However, results of our studies in adult mice (Fig. 1A) suggested an equally important expression of dok-4 in epithelial cells. This interpretation was strengthened by the observation that the great majority of published ESTs originated from other epithelial tissues including the skin, mammary gland, and colon (data not shown).
To further examine this possibility, we performed gene expression studies on kidney-derived epithelial cell lines. Three cell lines representing different segments of the nephron were chosen, including LLC-PK1 (derived from pig proximal tubule), MDCK (derived from dog distal tubule), and glomerular epithelial cells (GEC; derived from rat glomerular visceral epithelial cells). Because RNase protection assays require strict identity between probe and target mRNA, our murine dok-4 probe could not be used with RNA derived from different species. Therefore, we used an RT-PCR approach for these experiments. Although the cDNA sequences of dok-4 from rat and pig were not available, we selected a pair of mouse-derived primers that differed from the human dok-4 cDNA sequence by only one nucleotide and used nonstringent PCR conditions (see "Materials and Methods"), minimizing the risk of false negative results. As shown in Fig. 3A, a strong and specific dok-4 signal was detected in both MDCK cells and GEC but was essentially absent from LLC-PK1 cells, although we could not exclude the existence of a larger form of dok-4 in these cells. As expected, no signal was detected by PCR in the absence of prior reverse transcription (Fig. 3A, lane 2). The identity of the dok-4 RT-PCR product was further confirmed by digestion at an expected PstI restriction site conserved in mouse and human dok-4 sequences (Fig. 3B) and by cloning and sequencing of the PCR products obtained in MDCK cells and GEC (data not shown).
To directly compare dok-4 mRNA expression in epithelial and endothelial cells, we performed RT-PCR on RNA obtained from MDCK cells, GEC, and human umbilical vein endothelial cells. To ensure comparable loading for all of these samples, we amplified ␤-actin as a housekeeping control and performed both reactions under nonsaturating conditions. As shown in Fig. 3C, both MDCK cells and rat GEC appeared to contain similar levels of ␤-actin mRNA but higher levels of dok-4 mRNA compared with human umbilical vein endothelial cells.
In separate experiments, we also detected some dok-4 expression in two additional examples of renal cells, namely cultured mouse mesangial cells (the smooth muscle-like cells of the glomerulus) and isolated rat glomeruli (composed of mesangial, endothelial, and epithelial cells) (Fig. 3D).  8 and 10 and data not shown). In addition to the expected 37-kDa band, an ϳ45-kDa band was observed clearly in Caco-2 cells (Fig. 4A, double arrowhead). The presence of this additional band was also observed in two other human cell lines, T47D and HeLa, but not in other cells (data not shown), suggesting a human-specific splice variant. 2 Notably, as shown in Fig. 4B, the expression of Dok-4 in Caco-2 cells increased markedly after growth arrest-induced differentiation (57), whereas expression of a control protein, the tyrosine phosphatase Shp-2, remained essentially unchanged. When quantified with the NIH ImageJ software and normalized for Shp-2 expression, the increase in Dok-4 expression in lysates was 2.4-fold. Therefore, epithelial cells derived from multiple organs displayed prominent expression of Dok-4 at both the mRNA and protein level, and Dok-4 expression appeared to correlate with epithelial differentiation.

Dok-4 Protein Is Expressed and Phosphorylated in Epithelial
To determine whether endogenous Dok-4 could undergo tyrosine phosphorylation, Caco-2 cells were subjected to stimulation with pervanadate followed by anti-Dok-4 IP and antiphosphotyrosine immunoblotting. As shown in Fig. 4C, pervanadate treatment resulted in tyrosine phosphorylation of two bands that corresponded to 37-and 45-kDa Dok-4.
Dok-4 Is a Substrate for Cytosolic Tyrosine Kinases of the Src and Jak Family-Collectively, our gene expression studies confirmed the results of Grimm et al. (56) but also suggested that dok-4 mRNA was expressed in a wide range of cell lineages with prominence in several epithelial cell types. Intriguingly, neither Ret nor Tie-2 is known to exist in epithelial cells. Indeed, we were unable to detect Ret or Tie-2 expression in Caco-2 cells by IP (data not shown). Ret expression is mostly confined to specific tissues of the developing embryo and of the mature central nervous system, whereas Tie-2 is exclusively expressed in endothelial cells. This would suggest either that the function of Dok-4 in epithelial cells is independent of tyrosine phosphorylation or that Dok-4 functions downstream of kinases other than Ret or Tie-2. Likely candidates for Dok-4 kinases would be members of the Src family, since these are probably involved in phosphorylation of other Dok family members in hematopoietic cells (13,15). The three ubiquitous members of the Src family, Src, Fyn, and Yes, are most relevant to epithelial cell function because of their essential role in integrin signaling (58) and possibly in other pathways. For instance, it has been recently recognized that Fyn plays a key role in the highly specialized glomerular epithelial cells of the kidney (so-called podocytes) by maintaining the integrity of the glomerular filtration barrier (59).
To confirm the potential relevance of Src kinases in relation to Dok-4 function, we examined the expression of Fyn and Src in various epithelial cell lines where Dok-4 had been detected. As shown in Fig. 5A, one or both kinases were strongly expressed in all cell lines examined, including Caco-2 (colon), T47D (breast), MDCK (kidney tubule), GEC (kidney glomerulus), HeLa (uterine cervix), and TT (thyroid).
To determine whether Fyn could phosphorylate Dok-4, we transiently co-transfected COS cells with Dok-4 and either Fyn, Ret, or PDGFR-␤. Ret had already been shown to phosphorylate Dok-4 under similar conditions (56). Lysates obtained from these cells were subjected to IP with a Dok-4specific antiserum followed by anti-phosphotyrosine immunoblotting. As shown in Fig. 5B, Dok-4 expressed alone was not phosphorylated (lane 2), but co-transfection with either Fyn or Ret resulted in strong tyrosine phosphorylation (lanes 3 and 4).
In sharp contrast, co-expression with PDGFR-␤ resulted in no detectable phosphorylation of Dok-4 (lane 5) although PDGFR-␤ was clearly autophosphorylated under these conditions (data not shown). Similarly, overexpressed IGF-IR was able to phosphorylate itself as well as endogenous COS cell proteins but was unable to phosphorylate Dok-4, even after the addition of its ligand IGF-I (Fig. 5C).
The inability of PDGFR-␤ to phosphorylate Dok-4 was surprising, because Dok-1 had previously been shown to undergo 2 C. Baldwin and S. Lemay, unpublished data. tyrosine phosphorylation in Dok-1-transfected PDGF-treated fibroblasts (21). To confirm this apparent difference, we directly compared the capacity of Ret and PDGFR-␤ to phosphorylate Dok-1, Dok-4, and Dok-5 in COS cells. As shown in Fig.  5D, Ret could phosphorylate all three Dok molecules tested, whereas PDGFR-␤ could phosphorylate only Dok-1. In subsequent experiments, we tested the capacity of additional cytosolic tyrosine kinases to phosphorylate Dok-4. As shown in This configuration is thought to play a key role in membrane translocation through cooperative interactions with membrane lipids and tyrosine-phosphorylated transmembrane proteins, respectively. Because of post-translational lipid modification, cytosolic kinases of the Src family are associated with the inner leaflet of the plasma membrane (60). Therefore, intracellular distribution and the capacity to associate with the plasma membrane compartments also determine the suitability of various molecules as substrates of either RTKs or Src family kinases.
To determine the structural requirements for phosphorylation of Dok-4 by Fyn and Ret, we generated deletion mutants of Dok-4 that lacked the PH domain (⌬PH), the PTB domain (⌬PTB), or the C-terminal region (⌬CT). In addition, we generated a construct where the PH domain was replaced by the first 6 amino acids of FRS2, including the myristoylation signal (MGSCCS) (Myr-⌬PH). These mutant Dok-4 molecules were all tagged with a Myc epitope and were co-transfected in COS cells with Ret, Fyn, or empty vector. Cell lysates were subjected to anti-Myc IP followed by anti-phosphotyrosine immunoblotting. As shown in Fig. 6, deletion of the PH domain greatly reduced the ability of Dok-4 to undergo phosphorylation by either Ret or Fyn. Similarly, deletion of the PTB domain abolished Dok-4 phosphorylation. In contrast, deletion of the C-terminal portion of Dok-4 caused no detectable reduction in Dok-4 phosphorylation. The addition of a myristoylation signal to the ⌬PH mutant of Dok-4 completely restored its capacity to undergo tyrosine phosphorylation by both Fyn and Ret. To insure that mouse immunoglobulin light chain was not obscuring a phosphotyrosine signal from the similarly sized Dok-4 ⌬PTB, additional experiments were performed using a rabbit polyclonal anti-Myc antibody for IP (Fig. 6B). This confirmed the absence of any Ret-induced tyrosine phosphorylation in this mutant (Fig. 6B, lane 4) compared with wild-type Dok-4 (lane 2). These results suggested that the major sites of tyrosine phosphorylation were located within the PH and/or PTB domains or that either or both of these domains were necessary for proper localization or interaction of Dok-4 with respect to its kinase. The fact that a membrane-targeting signal was sufficient to restore phosphorylation after deletion of the PH domain suggested that the PH domain might serve in membrane localization of the native Dok-4 molecule.

Dok-4 Adapter in Epithelial Cells
calization of Dok-4, we fused various Dok-4 molecules to EGFP and studied their distribution in transfected cells using fluorescence microscopy. Transient expression of EGFP alone in COS cells resulted in a diffuse pattern of distribution within cells (Fig. 7A). Fusion of EGFP to wild-type Dok-4 (Dok-4-EGFP; Fig. 7B) resulted in a distinct pattern of distribution that included discontinuous areas of the plasma membrane. In addition, a prominent punctate cytoplasmic distribution pattern was observed, the nature of which remains under investigation. In contrast to wild-type Dok-4, deletion of the PH domain (Dok-4 ⌬PH-EGFP) (Fig. 7C) resulted in a diffuse cytosolic distribution similar to that of EGFP alone. Interestingly, the addition of a myristoylation signal to the PH domain deletion mutant (Dok-4 Myr-⌬PH-EGFP) (Fig. 7D) resulted in a linear pattern of membrane localization distinct from the wild-type Dok-4 construct, suggesting that the PH domain of Dok-4 targeted specific membrane subdomains. Surprisingly, deletion of the PTB domain (Dok-4 ⌬PTB-EGFP) (Fig. 7E) also abolished membrane localization of Dok-4, whereas deletion of the C-terminal region (Dok-4 ⌬CT-EGFP) (Fig. 7F) had no detectable impact on cellular localization. It therefore appeared as if membrane localization of Dok-4 occurred through the cooperative action of both the PH and PTB domains. Indeed, the PH domain alone, even when more broadly defined (residues 1-108 instead of 1-99; Dok-4 PH-EGFP) (Fig. 7G) was unable to localize at the membrane.
In COS cells transfected with either Dok-4-EGFP or Dok-4 Myr-⌬PH-EGFP, a single globular area of dense fluorescence could be observed that did not overlap with the nucleus under phase-contrast microscopy. This may have represented artifac-tual protein aggregation, as is known to occur with some GFP fusion proteins. Therefore, in order to create more physiological and transfection-independent conditions for intracellular localization studies, we generated stable clones of 293 cells that could express either Dok-4-EGFP or Dok-1-EGFP under the control of an ecdysone promoter. In the basal state, these cells contained no detectable fusion protein expression as determined by both immunoblotting and fluorescence microscopy (not shown), but after treatment with ponasterone A for 24 h, expression of the EGFP fusion proteins was strongly induced. As shown in Fig. 8A, the localization of Dok-4-EGFP in these cells was similar to the discontinuous membrane distribution observed in transiently transfected cells, whereas the dense cytosolic aggregates observed in transiently transfected cells were absent. In contrast to Dok-4-EGFP, Dok-1-EGFP was expressed in a diffuse cytosolic pattern (Fig. 8B). Membrane localization of Dok-4-EGFP and cytosolic localization of Dok-1 EGFP were also confirmed in transiently transfected cells using confocal microscopy (Fig. 8, C-F). Collectively, these results suggested that, unlike its homolog Dok-1, Dok-4 was constitutively associated with the plasma membrane through a PH and PTB domain-dependent mechanism.
To more precisely define the mechanism of Dok-4 membrane localization, we used a modification of a yeast-based assay previously described by Isakoff et al. (61). In this assay, the membrane-targeting property of molecule X is revealed by the capacity of a fusion protein of X and human Sos (or activated Ras) to restore growth at 37°C in the temperature-sensitive Cdc25 mutant yeast strain. Assuming that endogenous yeast membrane proteins do not themselves recruit the molecule of  6 and 7), the C-terminal region (⌬CT, lanes 8 and 9) or with substitution of the PH domain by a myristoylation signal (Myr-⌬PH, lanes 10 and 11). IP anti-Myc was followed by antiphosphotyrosine immunoblotting (top panel) and anti-Myc reprobing (second panel). Control immunoblots were performed for Fyn and Ret (bottom two panels). B, Dok-4 wild type (lanes 1 and 2) or Dok-4 ⌬PTB (lanes 3 and 4) were co-transfected with either empty vector (lanes 1 and 3) or Ret (lanes 2 and 4). IP was performed with a rabbit polyclonal anti-Myc antibody and was followed by anti-phosphotyrosine immunoblotting (top panel) and anti-Dok-4 reprobing (middle panel). Expression of Ret in cell lysates was confirmed by anti-Ret immunoblotting (bottom panel).

Dok-4 Adapter in Epithelial Cells
interest, membrane localization in this assay implies a direct interaction with membrane lipids. Indeed, this assay has correctly predicted the binding of PH domains to PtdIns 4,5bisphosphate (under basal conditions) or PtdIns 3,4,5-trisphosphate (in the presence of phosphatidylinositol 3-kinase expression) (61,62). Sos fusion protein constructs were generated and transformed into Cdc25 yeast, and their expression was confirmed by anti-Sos immunoblotting (data not shown). As shown in Fig. 9, all yeast clones grew normally at 25°C. However, expression of human Sos was insufficient to permit growth of Cdc25 yeast at the nonpermissive temperature of 37°C. In contrast, fusion of Dok-4 to Sos allowed the yeast to grow at 37°C, whereas deletion of either the PH or PTB domain or substitution of Dok-4 by Dok-1 abolished this effect. Moreover, the PH domain alone (pSOS-Dok-4 PH domain; containing amino acids 1-108 of Dok-4) was also unable to permit growth at 37°C. Identical results were obtained with two additional clones of each pSOS construct studied separately (data not shown). Thus, the structural requirements for membrane localization of Dok-4 appeared to be the same in yeast as in mammalian cells and probably involved direct binding of the PH and PTB domains to lipid components of the normal eukaryotic cell membrane such as PtdIns(4,5)P2, which represents a major membrane phosphatidylinositide in resting cells (63).
The PTB Domain of Dok-4 Binds Poorly to Dok-1 PTB Ligands-The PTB domain of Dok-4 appeared necessary for membrane targeting properties in both yeast and mammalian cells, suggesting its involvement in direct phospholipid binding. In contrast, the identification of Dok-4 as a Ret-associated molecule through yeast two-hybrid screening (56) suggested a possible role of the PTB domain in binding RTKs, as had already been shown for the PTB domains of Dok-2 (12,20) and IRS-1 (64). Indeed, our own results showed that the PTB domain was required for phosphorylation of Dok-4 by Ret or Fyn (Fig. 6). To determine whether the PTB domain of Dok-4 could be implicated in binding Ret, we performed pull-down assays using a fusion protein of GST and the PTB domain of either Dok-1 or Dok-4. Dok-1 was known to bind to the PTB-binding region of Ret (22). Whereas the GST-Dok-1 PTB fusion protein bound strongly to Ret present in the lysates of transfected COS cells, the Dok-4 PTB domain retained only slightly more Ret protein than did GST alone (Fig. 10A). In other pull-down experiments, we found that neither the Dok-1 nor the Dok-4 PTB domains were able to bind Fyn (not shown).
A degenerate peptide screening approach had recently revealed that one potential target for the Dok-1 PTB domain was phosphorylated tyrosine 146 of Dok-1 itself (65), suggesting that Dok-1 may undergo homodimerization. This also raised the possibility of Dok-1 recruiting the PTB domain of other Dok family members, resulting in Dok heterodimers. Oligomerization of PH domain-containing proteins has been proposed as a potential mechanism of regulated membrane recruitment through an additive increment in avidity for phospholipids (40). To determine whether the PTB domain of Dok-4 could mediate binding to Dok-1, we performed pull-down assays on lysates of COS cells transfected with either wild-type Dok-1 or Dok-1 Y146F in the presence or absence of Fyn or its kinasenegative mutant. As shown in Fig. 10B, Fyn caused robust phosphorylation of Dok-1 on tyrosine, which allowed GST- Dok-1 PTB to bind Dok-1 in a Tyr 146 -dependent manner. However, GST-Dok-4 PTB bound to Dok-1 only weakly and in a phosphotyrosine-independent manner. In other pull-down experiments, we found that GST-Dok-4 PTB was unable to bind full-length Dok-4, even under conditions of tyrosine phosphorylation (data not shown). Thus, the binding specificity of the Dok-4 PTB domain was clearly distinct from that of Dok-1. In contrast to Dok-1, the PTB domain of Dok-4 is not likely to be involved in homodimerization.

Dok-4 Inhibits Signaling Downstream of Ret and Fyn-
To determine the role of the Dok-4 PH and PTB domains on signaling events downstream of Ret and Fyn tyrosine kinases, we performed reporter gene assays using an Elk-1 reporter system. Grimm et al. (56) had previously reported that Dok-4 was involved in activation of Erk1/2 and the downstream transcription factor Elk-1. In 293 cells, either transfection of wildtype Ret followed by treatment with recombinant GDNF and GFR␣-1-Fc or transfection of activated Ret (C634R) resulted in strong activation of Elk-1 (Fig. 11). Unexpectedly, co-transfection with increasing amounts of Dok-4 expression plasmid did not enhance Ret-induced Elk-1 activity. In fact, co-transfection with high amounts of Dok-4 plasmid (200 g) was inhibitory in some experiments (Fig. 11A). Similarly, varying the concentration of ligand (Fig. 11B) or the amount of activated Ret plasmid (Fig. 11C) failed to reveal an impact of Dok-4 on Ret-induced Elk-1 activation. In contrast to 293 cells where the inhibitory effect of Dok-4 was inconsistent, similar experiments con-ducted in Caco-2 cells revealed a reproducible inhibition of Ret-induced Elk-1 activation by Dok-4 in a dose-dependent manner (Fig. 11D). Surprisingly, PDGF-induced Elk-1 activation was also inhibited by Dok-4. Although higher amounts of plasmids were required to overcome the reduced transfection efficiency of Caco-2 cells compared with 293 cells, the relative amounts of each plasmid used remained similar. To confirm the specificity of the inhibitory effect observed in Caco-2 cells, identical experiments were performed in parallel with either c-Jun/Gal4 or Elk-1/Gal4 as the trans-reporter. Dok-4 inhibited Ret-induced Elk-1 but was either neutral or slightly stimulatory on the basal c-Jun luciferase activity detected in these cells (Fig. 11E). The inhibitory effect of Dok-4 did not result from saturation of the transfection conditions, since Dok-4 did not inhibit Elk-1 or c-Jun luciferase activity induced by direct activation of the MAP kinase cascade by MEKK1 (Fig. 11F).
To examine the impact of Dok-4 on Src kinase signaling, we also used Caco-2 cells. In these cells, transfection of activated Fyn (Y528F) resulted in an intense activation of Elk-1 luciferase activity (Fig. 12A). When increasing amounts of Dok-4 were co-transfected with Fyn, the level of Elk-1 activation was consistently reduced. However, this inhibitory effect was blunted when Dok-4Myc wild type was replaced with Dok-4Myc ⌬PH and partially restored with the addition of a myristoylation signal (Dok-4Myc Myr-⌬PH). To confirm the specificity of this inhibitory effect, the impact of Fyn and Dok-4 and both Elk-1 and c-Jun was compared, and the impact of Dok-4 on MEKK1 signaling was determined. In contrast to its effect on Elk-1, Fyn Y528F alone did not activate c-Jun (Fig. 12B). Co-expression of Dok-4 was not inhibitory but instead resulted in a modest activation of c-Jun (Fig. 12B). Finally, Dok-4 failed to inhibit Elk-1 activation and tended to enhance c-Jun activation induced by activated MEKK1 (Fig. 12C). DISCUSSION In this paper, we report the cloning and initial characterization of Dok-4, a novel member of the Dok family of adapter proteins expressed in nonhematopoietic tissues. Consistent with a prior report by Grimm et al. (56), we found that dok-4 was expressed rather ubiquitously, whereas the related molecule dok-5 was expressed almost exclusively in the central nervous system (Fig. 1). Based on in situ hybridization studies of dok-4 and the capacity of the endothelium-specific RTK Tie-2 to associate with Dok-4 in 293 cells, it had been proposed that Dok-4 was primarily an endothelial molecule, at least during embryonic development. In contrast, our studies on tissues and cell lines derived from adult tissues using RNase protection assay, RT-PCR, and immunoblotting suggest that Dok-4 is expressed in a wide range of cell lineages with prominence in both endothelial and epithelial cells (Figs. 1, 3, and 4). Furthermore, Dok-4 expression increased following differentiation of Caco-2 cells (Fig. 4B). Endogenous Dok-4 underwent tyrosine phosphorylation following treatment of Caco-2 intestinal epi-

Dok-4 Adapter in Epithelial Cells
thelial cells with the phosphatase inhibitor pervanadate (Fig.  4C). Using transiently transfected COS-1 cells, we confirmed that both Dok-4 and Dok-5 could undergo phosphorylation by the RTK Ret, but we found that, unlike their homologue Dok-1, they were unable to undergo phosphorylation by PDGFR-␤ (Fig. 5, B and D). The RTK IGF-IR was also unable to phosphorylate Dok-4 (Fig. 5C). We also found that the ubiquitous cytosolic tyrosine kinases Fyn, Src, and Jak-2 could phosphorylate Dok-4 (Fig. 5E). Furthermore, we discovered that Dok-4 was localized at the plasma membrane in both yeast and mammalian cells, in contrast to Dok-1, which was cytosolic (Figs. [7][8][9]. Membrane localization of Dok-4 was dependent on both its PH and PTB domains. Dok-4 mutants unable to localize at the cell membrane were also defective in their capacity to undergo tyrosine phosphorylation by Fyn and Ret (Fig. 6). Whereas the Dok-1 PTB domain had affinity for both Ret and phosphorylated Tyr 146 of Dok-1 itself, Dok-4 PTB bound poorly to both molecules (Fig. 10). Contrary to what had been predicted from a previous report (56), we found no evidence that Dok-4 could enhance Elk-1 signaling downstream of Ret. Instead, we found that in Caco-2 cells Dok-4 inhibited Elk-1 activation induced by Ret, PDGFR-␤, and Fyn (Figs. 11 and 12).
Prior to this report, Dok-4 protein expression had not been reported. The finding that Dok-4 was prominently expressed in epithelial cells is significant, because neither of the two RTK previously identified as putative upstream kinases for Dok-4 (Ret and Tie-2 (56)) are usually expressed in these cells. This suggests that other kinases may phosphorylate Dok-4 under physiological conditions. Our study points to a potential role for the more ubiquitously expressed Src and Jak family of tyrosine kinases in Dok-4 phosphorylation. Cai et al. recently reported Dok-4 and Dok-5 as insulin receptor and IGF-I receptor substrates (66). However, we have been unable to detect Dok-4 phosphorylation in epithelial cells or in Dok-4-overexpressing 293 cells after treatment with IGF-I. Also, IGF-I failed to increase pervanadate-induced Dok-4 phosphorylation in Caco-2 cells (data not shown). Most convincingly, overexpressed IGF-IR was unable to phosphorylate Dok-4 in transiently transfected COS-1 cells (Fig. 5C).
The subcellular distribution of Dok-4 was another important and surprising finding of this study. Cell fractionation and fluorescence microscopy experiments have recently suggested that Dok-1 can translocate to the cell membrane in a phosphatidylinositol 3-kinase-dependent manner (21,23). In contrast, Noguchi et al. (19) detected constitutive (but adhesiondependent) membrane association of Dok-1 in serum-starved CHO cells overexpressing Dok-1 protein. In the current study, we used fluorescence microscopy of EGFP fusion protein-expressing mammalian cells as well as a yeast-based assay to show that Dok-4 associated with the membranes of unstimulated eukaryotic cells and that this property was dependent on the PH and PTB domains.
Intriguingly, our studies with mutant forms of Dok-4 clearly showed that, whereas the myristoylation signal of FRS2 could restore membrane localization and enhance tyrosine phosphorylation of Dok-4 ⌬PH, it resulted in a linear distribution pattern rather than the discontinuous pattern observed with wild-type Dok-4 (Fig. 7, B and D). This suggests that the PH domain enables binding to more specific and spatially limited targets within the membrane subdomains, a property that is likely to have significant biological repercussions.
Given its inability to rescue Cdc25 mutant yeast or to localize at the membrane in mammalian cells (Fig. 9), the PH domain of Dok-4 probably belongs to the large group of low affinity PH domains. Interestingly, there is evidence that some PH domains with low intrinsic affinity for phosphoinositide binding require the cooperative action of another domain for membrane targeting (30,67). The dual requirement for the PH and PTB domain of Dok-4 in membrane localization suggests that the PTB domain of Dok-4 could serve such a cooperative role with the adjacent PH domain.
Several mechanisms could explain the cooperative action of PH-PTB tandems in membrane targeting. First, the PTB domain could bind membrane proteins or lipids with an affinity too low to generate stable membrane targeting on its own but sufficient to complement that of the PH domain. The identical structural requirements for Dok-4 membrane localization that we observed in yeast and in mammalian cells would tend to support phospholipids as likely targets of PH and PTB domain binding, because yeast membranes contain phosphoinositides (such as PtdIns 4,5-bisphosphate) but are much less likely to share the same membrane protein composition as mammalian cells. Furthermore, PTB domains do not exist in yeast (31), and their putative partner proteins are therefore probably absent as well. The PTB domains of Shc and Dab1 have been previously shown to bind phospholipids in addition to peptide sequences (43,44). Another possible mechanism for cooperativity would be that the PTB domain of Dok-4 is required for an intramolecular interaction that enhances the affinity of the PH domain for membrane phospholipids. Direct binding assays may help clarify the role of the Dok-4 PTB domain in membrane targeting in the future. Whatever the mechanism of PH-PTB cooperativity in Dok-4, it is interesting to note a possible analogy to IRS-1, where the affinity of the isolated PH domain for PtdIns 3,4,5-trisphosphate was reported to be significantly lower than that of the tandem PH-PTB domains (32), although it was unclear if this resulted from direct phospholipid binding by the PTB domain or from a favorable intramolecular interaction (32,33).
Whether PTB domains bind phospholipids physiologically or not, their primary role as protein-protein interaction modules is well established (1,45). In this respect, the current results showed that the PTB domain of Dok-4 has a target specificity clearly distinct from that of the Dok-1 PTB domain. Specifically, its capacity to associate with Ret in pull-down assays was surprisingly low, and it was unable to bind tyrosine 146 of Dok-1 in a phosphorylation-dependent manner. Moreover, GST-Dok-4 PTB failed to bind any detectable phosphoproteins under conditions of robust tyrosine phosphorylation such as Fyn-overexpressing 293 cells (data not shown). We also showed that Dok-4 was a substrate for Ret but not for PDGFR-␤, whereas Dok-1 was an equally good substrate for both RTKs. Whether this latter finding resulted from distinct peptide-binding specificity of the two PTB domains remains to be determined, since we have so far been unable to show binding of either PTB domain to full-length PDGFR␤ in pull-down experiments (data not shown). Interestingly, despite the relatively weak interaction of the Dok-4 PTB domain with Ret, there was comparable phosphorylation of Dok-1 and Dok-4 by Ret, perhaps because the constitutive membrane localization of Dok-4 compensated for the weak receptor-substrate interaction. Finally, it should be recognized that, beyond a limited capacity for protein-protein (and possibly protein-lipid) interaction, the Dok-4 PTB domain may also play an important role in tyrosine phosphorylation as an actual substrate. Indeed, the PTB sequence contains six tyrosine residues, at least some of which become phosphorylated in the presence of Ret or Fyn. 3 In addition to its capacity for constitutive membrane localization demonstrated herein, a striking difference between Dok-4 and Dok-1 is that Dok-4 lacks obvious consensus binding sites for known Src homology 2 domains. Indeed, it is still unclear if phosphorylation of the Dok-4 C-terminal region occurs in vivo (see Fig. 6, lanes 10 and 11), let alone if it is necessary for Dok-4 function (see Fig. 11D). In this context, the intriguing observation that fusion of the Dok-4 C-terminal region to the intracellular portion of Ret restores activation of the Erk/Elk-1 pathway by a mutant epidermal growth factor receptor/Ret chimeric receptor in PC-12 and Neuro 2A cells (56) remains unexplained. In the current study, we were unable to demonstrate a similar enhancement of Ret signaling by Dok-4. Instead, we found that Dok-4 inhibited Ret-induced activation of Elk-1 in Caco-2 (Fig. 11D) and less consistently in 293 cells (Fig. 11A). This suggests that neuronal cell-specific factors and/or properties unique to the epidermal growth factor receptor/Ret/Dok-4 chimera were responsible for rescuing Erk/Elk-1 activation in the experiments reported by Grimm et al. (56). Covalent association of Dok-4 and Ret within a chimeric molecule is a highly unphysiological condition given the apparent weakness of the Dok-4/Ret interaction (Fig. 10A) (data not shown) and other PTB domain-mediated interactions. In any case, since Dok-4 probably works through recruitment of downstream effector molecules and since it might compete with Shc and other adapter molecules for binding to tyrosine 1062 of Ret, its ultimate effect on Erk/Elk-1 activation may vary depending on the cellular context. Obviously, more studies will be required to precisely define the basis for these apparently opposite effects of Dok-4.
In addition to its inhibitory effect on Ret signaling, we found that Dok-4 also inhibited Fyn-induced Elk-1 activation in Caco-2 cells and that this effect was partly dependent on the PH domain and on membrane localization of Dok-4 (Fig. 12A). In contrast, tyrosine phosphorylation of Dok-4 seemed dispensable for this inhibition of Elk-1, because Elk-1 activation induced by both phosphorylation-competent (Ret) and incompetent (PDGFR-␤) receptors was equally sensitive to inhibition by Dok-4 (Fig. 11D). Additional phosphorylation-dependent functions of Dok-4 may exist, but they will require further investigation.
The finding that Dok-4 has an inhibitory effect on tyrosine kinase signaling is reminiscent of the inhibitory effects reported for other more distant members of the Dok family, including Dok-1, -2, and -3 (14, 15, 26 -29) and suggests an evolutionarily conserved pattern. It remains to be determined whether this inhibitory effect of Dok-4 results from a direct impact on Ret or Fyn or on a specific downstream effector. Efforts to identify effectors of Dok-4 and their possible regulation by tyrosine phosphorylation should provide important insights into epithelial cell function.