ATF-7, a Novel bZIP Protein, Interacts with the PRL-1 Protein-tyrosine Phosphatase*

We have identified a novel basic leucine zipper (bZIP) protein, designated ATF-7, that physically interacts with the PRL-1 protein-tyrosine phosphatase (PTPase). PRL-1 is a predominantly nuclear, farnesylated PTPase that has been linked to the control of cellular growth and differentiation. This interaction was initially found using the yeast two-hybrid system. ATF-7 is most closely related to members of the ATF/CREB family of bZIP proteins, with highest homology to ATF-4. ATF-7 homodimers can bind specifically to CRE elements. ATF-7 is expressed in a number of different tissues and is expressed in association with differentiation in the Caco-2 cell model of intestinal differentiation. We have confirmed the PRL-1·ATF-7 interaction and mapped the regions of ATF-7 and PRL-1 important for interaction to ATF-7's bZIP region and PRL-1's phosphatase domain. Finally, we have determined that PRL-1 is able to dephosphorylate ATF-7 in vitro. Further insight into ATF-7's precise cellular roles, transcriptional function, and downstream targets are likely be of importance in understanding the mechanisms underlying the complex processes of maintenance, differentiation, and turnover of epithelial tissues.

It is clear that many cellular processes are regulated through protein phosphorylation. This post-translational modification is responsible for the control of a wide variety of important processes, including the regulation of metabolism, cell proliferation, the cell cycle, gene expression, protein synthesis, and cellular transport (1,2). Because phosphorylation is a dynamic and reversible process, it follows that phosphatases are as important as kinases in its regulation (3,4). Phosphorylation of transcription factors and their associated proteins is one of the principal methods used to regulate gene expression (1,2,5,6). Alterations in protein phosphorylation states bring about these changes in a number of different ways, including the regulation of subcellular localization (7)(8)(9), changes in DNA binding (10,11), or alterations in transactivating ability. Classic examples of this latter phenomenon include the basic leucine zipper (bZIP) 1 proteins CREB and c-Jun, where phosphorylation of specific residues in the transactivating domain has been demonstrated to up-regulate transactivation, probably by allowing interaction of these proteins with transcriptional coactivators such as CREB-binding protein (12,13). Kinases or phosphatases may also, in some situations, bind transcription factors but influence transcription by acting on proteins other than the transcription factors themselves (7,14). An example of this phenomenon is the nuclear tyrosine kinase c-Abl, which binds to p53 and increases its transactivating ability without phosphorylating it. It is thought that Abl may be able to execute this function by phosphorylating the C-terminal domain of RNA polymerase II, which is known to be extensively phosphorylated on tyrosine (14,15).
The PRL-1 protein-tyrosine phosphatase (PTPase) was initially identified as an immediate-early response gene in regenerating liver and mitogen-stimulated fibroblasts (16). PRL-1 is a 20-kDa protein that contains the "signature" amino acid sequence for the active site of PTPases but otherwise does not contain regions of homology to any previously described protein (16). PRL-1 is primarily localized to the cell nucleus with a discrete, reproducible "speckled" pattern on immunofluorescence. Under certain circumstances, PRL-1 also is localized to extranuclear sites in the cell (17,18). PRL-1 is found in the insoluble cellular fraction, despite the fact that it is readily soluble when expressed in bacteria (16). This is likely a result of protein prenylation, because PRL-1 is a farnesylated protein (19). When PRL-1 was stably overexpressed in 3T3 fibroblasts, altered growth characteristics became apparent, including a faster doubling time, growth to a greater saturation density, altered morphology, and evidence of anchorage-independent growth manifested by the ability of these cells to grow in soft agar (16). Overexpression of PRL-1 in epithelial cells resulted in tumor formation in nude mice (19).
PRL-1 is also significantly expressed in intestinal epithelia, and in contrast to PRL-1's expression pattern in liver, its expression is associated with cellular differentiation in the intestine. Specifically, PRL-1 is expressed in villus but not crypt enterocytes, and in differentiated, but not proliferating, Caco-2 cells (20). Recently, PRL-1 protein was found to be expressed during development in a number of differentiating epithelial tissues (17). These results suggest that PRL-1 may have divergent roles in different tissues. It is an established feature of some growth response genes that they may play a role in terminal differentiation in some tissues (21)(22)(23)(24)(25). The apparently paradoxical dual roles may be explained by the availability of different substrates or cofactors in different cells, different kinetics of protein expression, or by the presence of scaffolding or anchoring proteins that may direct an enzyme to a different cellular location and different substrates (26,27).
Significant insight into PRL-1's specific cellular functions and the reasons for its apparently varied expression pattern in different tissues may be derived from identification of PRL-1's substrates and other cellular partners. To that end, we performed a yeast two-hybrid screen using PRL-1 as bait. We have identified a novel protein that interacts with PRL-1. This protein, which we have designated ATF-7, is a novel bZIP protein most closely related to members of the ATF/CREB family. We have functionally confirmed that ATF-7 is a bZIP protein by showing that its homodimers specifically bind to cyclic AMP response (CRE) elements. We have confirmed the interaction of PRL-1 and ATF-7 using GST binding and coimmunoprecipitation assays, and we have mapped the sites of interaction to include PRL-1's phosphatase domain and ATF-7's bZIP domain. ATF-7 is expressed in a number of different tissues, and it is expressed in association with differentiation in the Caco-2 cell model of intestinal differentiation. Finally, we have determined that PRL-1 is able to dephosphorylate ATF-7 in vitro. It is likely that further insight into ATF-7's precise cellular roles, transcriptional function, and downstream targets will be of importance in understanding the mechanisms underlying the complex processes of maintenance, differentiation, and turnover of epithelial tissues.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid System and ␤-Galactosidase Assays-The N-terminal 132 amino acids of PRL-1 fused to the C terminus of the GAL4 DNA binding domain in the yeast expression vector pGBT9 (CLON-TECH) was constructed from the full-length PRL-1 cDNA. This construct contains most of the full-length PRL-1 cDNA except for the C-terminal basic region and CCIQ farnesylation domain. The active site cysteine (Cys-104) was mutated to serine as previously described (C104S) (16). Use of active site cysteine-serine mutant PTPases to demonstrate binding of PTPases to other proteins is a well-established and validated method (28 -30). A 3T3-L1 adipocyte library was synthesized from fully differentiated adipocytes with a cDNA synthesis kit (Stratagene) and constructed in the pGAD-10 GAL4 vector (CLON-TECH) (gift of Dr. Alan Saltiel) (31). The yeast strain HF7c was cotransformed with the GAL4-PRL-1 construct and with the 3T3-L1 adipocyte library. The resulting transformants were plated on selection medium lacking tryptophan, leucine, and histidine and were incubated at 30°C for 4 -5 days. 5 ϫ 10 6 clones were analyzed. Colonies positive for growth on selective media lacking histidine were blotted on filter paper (Whatman number 5), permeabilized in liquid nitrogen, and placed on another filter soaked in Z buffer (60 mM Na 2 HPO 4 , 40 mM NaH 2 PO 4 , 10 mM KCl, 1 mM MgSO 4 , 37.5 mM ␤-mercaptoethanol) containing 1 mM 5-bromo-4-chloro-3-indolyl-␤-D-galactoside. Yeast colonies were scored as positive when a bright color developed within 3 h. Library-derived plasmids were rescued from positive clones and then transformed into HB101 Escherichia coli by electroporation. False positives were eliminated by transforming the rescued plasmids back into yeast along with either the PRL-1 bait construct, empty vector, or a control p53 bait construct. True positives were identified by their requirement for the PRL-1 bait construct to activate the reporter genes, and these were then sequenced. Sequence analysis was performed using GenAlign software, and FASTA and BLAST searches were performed against the SwissProt and GenBank data libraries.
Northern Blots-RNA preparation, Northern blot analyses, and labeling of recombinant plasmids have been described elsewhere (16,32). Caco-2 cells were grown and harvested with respect to proliferating and differentiated phenotypes as previously described (20). Total RNA was extracted from cells and from mouse tissues using the techniques previously described (32,33). Hybridization buffer consisted of 10% dextran sulfate, 40% formamide, 0.6 M NaCl, 0.06 M sodium citrate, 7 mM Tris (pH 7.6), 0.8ϫ Denhardt's solution, and 0.002% heat-denatured, sonicated salmon sperm DNA. Northern blots were hybridized at 42°C for 16 h and washed for 30 min twice at 60°C in 0.015 M NaCl-0.0015 M sodium citrate-0.1% SDS prior to exposure to film (33).
Plasmids and Antibodies-The cDNA insert of a full-length ATF-7 clone isolated in the yeast two-hybrid screen and the cDNA insert of full-length C104S-PRL-1 were cloned into the pSPUTK (Stratagene) both with and without an Myc epitope fused to the N-terminal end. This vector provides a strong Kozak (34) consensus sequence and methionine for translation. C104S-PRL-1 and wild type PRL-1 were cloned into the pGEX GST fusion vector (Amersham Pharmacia Biotech). The appropriate restriction sites and the indicated epitope tags were added by performing the polymerase chain reaction with primers containing these sequences. Plasmid constructs were confirmed by sequencing and by protein expression. For interaction site mapping studies, the following constructs were made in pSPUTK: pSPUTK-ATF-7-amino (amino acids (aa) 1-139) and pSPUTK-ATF-7-bZIP (aa 140 -217). The following constructs were made in pGEX: PGEX-PRL-1-aa-1-96, PGEX-PRL-1aa-1-132, PGEX-PRL-1-aa-60 -118, PGEX-PRL-1-aa-97-173, PGEX-PRL-1-aa-97-132, and PGEX-PRL-1-aa-118 -173. All deletion and truncation constructs were made using polymerase chain reaction with appropriate primers and restriction sites and were sequenced completely prior to use in experiments. The in vitro translated proteins (pSPUTK) or bacterially expressed proteins (pGEX) were all of the predicted size. Rabbit polyclonal anti-ATF-7 antibodies were prepared by Cocalico Biologicals (Reamstown, PA) against bacterially expressed purified denatured ATF-7 protein expressed as a fusion protein linked to six histidines (35). The anti-Myc epitope tag monoclonal antibody (9E10) was obtained commercially (Babco, Richmond, CA).
In Vitro Transcription/Translation-In vitro transcription/translation was performed using the TnT-coupled lysate system (Promega) according to the manufacturer's instructions. The reaction was incubated for 2 h at 30°C in the presence of [ 35 S]methionine, and the products were analyzed directly by SDS-PAGE or subjected to immunoprecipitation or binding assays prior to SDS-PAGE as described in the text.
GST Binding Assays-Radiolabeled in vitro translated proteins (5 l) were incubated with GST or GST-PRL-1 C104S fusion proteins (1 g), or with the indicated PRL-1 protein fragments attached to glutathione-Sepharose beads in 500 l of binding buffer (20 mM HEPES, pH 7.5, 150 mM NaCl) for 1 h at 4°C with gentle rotation. The beads were washed five times with binding buffer and resuspended in Laemmli sample buffer, and the sample was analyzed by SDS-PAGE followed by autoradiography.
Coimmunoprecipitations-The two proteins or protein fragments being tested were in vitro translated simultaneously as described above. One protein used was fused to the Myc-tag epitope. 7.5 l of the in vitro translated protein was incubated with to 5 l of the anti-Myc-tag antibody or control sera in 500 l of IP buffer (50 mM Tris, pH 7.4, 1 mM EDTA, 150 mM NaCl, 1% Triton) for 1 h at 4°C. Immunocomplexes were bound to protein A-agarose beads and, after washing four times in IP buffer, were resolved by SDS-PAGE and visualized by autoradiography.
Electromobility Shift Assays-Preannealed, gel-purified, doublestranded oligonucleotides were radiolabeled and incubated with 5 l of in vitro translated proteins or 10 g of mouse liver nuclear extract for 15 min at room temperature in binding buffer (10 mM Tris, pH 7.5/50 mM NaCl/1 mM EDTA/1 mM dithiothreitol/5 mM MgCl 2 /10% (v/v) glycerol). 1-2 g of poly(dI-dC) was used as a nonspecific competitor in each reaction. Nuclear extracts from liver were prepared according to the method of Hattori (36), with modifications (37). The mixtures were electrophoresed on a nondenaturing polyacrylamide gel in 1ϫ TBE buffer (88 mM Tris, 88 mM boric acid, 2 mM EDTA). Gels were dried and exposed to x-ray film. Supershifts were performed by incubating 1-1.5 l of primary antibody with in vitro translated proteins in binding buffer for 45 min at 4°C, prior to addition of labeled oligonucleotide. Cold competitions were performed by incubating unlabeled oligonucleotide in 100-fold excess with in vitro translated proteins in binding buffer for 45 min at 4°C prior to the addition of labeled oligonucleotide. The double-stranded oligonucleotides used were: CRE: TCATGGTAAAAA-TGACGTCATGGTAATTA, C/EBP: GATCCGGTTGCCAAACATTGCG-CAATCT, and AP1: TATCGATAAGCTATGACTCATCCGGGGGA. Oligonucleotides were end-labeled with [␥-32 P]ATP using polynucleotide kinase.
In Vitro Dephosphorylation of Tyrosine-phosphorylated ATF-7 by PRL-1-GST-ATF-7 was expressed in bacteria and purified using the methods outlined above. The protein, attached to glutathione beads, was tyrosine-phosphorylated with 32 P using c-Src kinase (Upstate Biotechnology) according to the manufacturer's directions. After the kinase reaction was stopped, the beads were washed four times in 1 ml of phosphatase buffer (50 mM HEPES, pH 7.5, 0.1% ␤-mercaptoetha-nol), resuspended, and split into three equal aliquots. PRL-1 active phosphatase and C104S-PRL-1 inactive mutant phosphatase were prepared as described previously (16). Equal amounts of active or mutant PRL-1, or buffer (negative control), were added to each tube, which were then incubated for 60 min at 37°C. The phosphatase reaction was terminated by the addition of equal volumes of 2ϫ Laemmli buffer, the products were then boiled, and run on an SDS-PAGE gel, which was then dried and exposed to x-ray film (Kodak). The results were quantified by densitometry.

Isolation of ATF-7 as a PRL-1 Binding Protein with the Yeast
Two-hybrid System-To identify PRL-1-interacting proteins, a truncated PRL-1 (amino acids 1-132) fused to the DNA binding domain of GAL4 was used as bait to screen a 3T3-L1 adipocyte cDNA expression library fused to the GAL4 transcriptional activation domain in the yeast two-hybrid interaction system. The bait construct used contains most of the full-length PRL-1 protein except for the C-terminal basic and farnesylation domains. Full-length PRL-1 bait did not yield any positives (true or false), although we later determined that full-length PRL-1 can in fact interact with ATF-7. We were able to document that the full-length fusion protein was able to be expressed in the yeast (data not shown). Because PRL-1 is a farnesylated protein residing in the insoluble cellular fraction (despite being readily soluble itself when expressed in bacteria) (16), we surmise that the C-terminal basic and farnesylation domains caused misdirection of the protein to a cellular site in the yeast where it was unable to interact with potential prey proteins. Of 5 ϫ 10 6 total transformants, 38 colonies positive for ␤-galactosidase activity were isolated from histidine-minus plates. When the library-derived plasmids were recovered from these 38 colonies, we determined that 31 were true positives. 16 of these clones encoded either full-length or near full-length ATF-7. The remaining 15 clones all encoded another novel protein that has no homology to ATF-7 and is not a transcription factor. 2 This protein will be the subject of a separate report. As summarized in Fig. 1, each of the GAL4-AD/ATF-7 clones induced histidine-minus growth and ␤-galactosidase activity only when they were coexpressed with PRL-1-derived/ GAL4-BD fusion protein and not with an unrelated GAL4-BD fusion protein containing p53 or the GAL4 DNA binding domain alone (empty vector).
cDNA Sequence of ATF-7-Sequence analysis of the clone with the longest insert (1.6 kb) revealed a single open reading frame of 651 bp fused in-frame to the GAL4 activation domain. The sequence is shown in Fig. 2A. An ATG initiation codon was present near the 5Ј-end, which was a good match for the canonical Kozak eukaryotic translation initiation consensus (34).
The sequence encodes a protein of 217 amino acid residues, with a predicted molecular mass of 24 kDa and a pI of 5. 46. In vitro translation of the ATF cDNA yielded an ϳ30-kDa protein (Fig. 4), probably due to post-translational modification of the protein.
Comparison with data bases revealed that the sequence is novel but that it has several characteristics of a bZIP transcription factor. The extreme C-terminal end of the predicted ATF-7 protein contains three leucines and three valines, each separated by six other amino acids, suggesting a leucine zipper structure (38). This hybrid leucine-valine zipper is unique to ATF-7 among the previously described bZIP proteins. Immediately upstream of this leucine-valine zipper sequence is an arginine-lysine-rich basic domain, thought to be necessary for sequence-specific DNA binding by bZIP proteins (39,40). The N-terminal end of the predicted protein is negatively charged and proline-rich, reminiscent of the acidic activation domains of bZIP transcription factors. The bZIP family can be divided into three groups on the basis of binding site preference (41): (i) the C/EBPs, (ii) the AP1 group of transcription factors, and (iii) the CREB/ATF family, which contains the ATFs, the original CREBs, and the CRE modulators. Distinctions within the bZIP family are also based upon differences in transactivating ability, patterns of tissue expression, and phosphorylation by specific kinases (41). As shown in Fig. 2B and summarized in Fig.  2C, ATF-7 is most closely related to ATF-4 (also known as CREB-2, C/ATF, and TAXREB67 (42)(43)(44)(45)). In a number of cases, especially near the C terminus, ATF-7 and ATF-4 contain identical amino acids that diverge from the consensus deduced from the other bZIP proteins (Tyr-227, Asp-230, Glu-234, Val-235, Lys-237, Arg-239, and Gln-241).
DNA Binding and Tissue Expression Pattern of ATF-7-Because bZIP proteins bind specific DNA elements, we next sought to confirm that ATF-7 is indeed a DNA-binding protein and identify which DNA sites ATF-7 it can bind. We used electromobility shift assays to test the ability of in vitro translated ATF-7 to bind different DNA sequences known to bound by bZIP proteins. As shown in Fig. 2D shown). Because ATF-4 has been reported to bind to C/EBP sites (45)(46)(47), we also tested ATF-7's ability to bind to a C/EBP site oligo. These data are also shown in Fig. 2D. We found that ATF-7 could not bind this element as a homodimer. The single band that appears in the ATF-7 (second) lane, is also present when reticulocyte lysate alone is used (first lane). This band is not supershifted by the addition of anti-ATF-7 antibody (third lane) and is not eliminated by the addition of excess cold competitor oligo (fourth lane). The fifth lane shows results with liver nuclear extract. As expected, there is prominent binding evident, indicating that the failure of ATF-7 to bind is not due to a problem with the oligo or binding conditions. We have also Shaded areas indicate identical amino acids; boxed areas indicate homologous amino acids. In a number of cases, especially near the C terminus, ATF-7 and ATF-4 contain identical amino acids that diverge from the consensus deduced from the other bZIP proteins (see text for details). C, table summarizing the homology among the bZIP proteins shown in B. In D: left panel), CRE oligonucleotide was radiolabeled and incubated with in vitro translated ATF-7 (first lane) as described under "Experimental Procedures." The mixtures were electrophoresed on a nondenaturing polyacrylamide gel, which was then dried and exposed to film. Supershift (second lane) and competition assays (third and fourth lanes) were performed as described in the text. Specific ATF-7 and nonspecific bands are indicated by arrows. Right panel, C/EBP oligonucleotide was radiolabeled and incubated with reticulocyte lysate alone (first lane), in vitro translated ATF-7 (second, third, and fourth lanes), or mouse liver nuclear extract (fifth and sixth lanes) as described under "Experimental Procedures." The mixtures were processed as indicated for supershift and competition experiments in the same manner as was done for the CRE experiments. The supershift and cold competition data confirm that the band seen is a nonspecific band that is not due to ATF-7 binding. not found evidence that ATF-7 homodimers can bind to AP1 sites (data not shown).
Northern blot analysis was performed to determine the tissue distribution of ATF-7, and, as shown in Fig. 3A, it was found to be expressed ubiquitously. The highest levels of expression appear to be in the liver, lung, adipose tissue, heart, and skeletal muscle. We also examined the expression pattern of ATF-7 in situations where PRL-1 has a distinctive pattern of expression. We did not find variation in the level of ATF-7 expression during liver regeneration (data not shown). We then examined the expression of ATF-7 in Caco-2 cells, a human colonic adenocarcinoma cell line, which exhibits spontaneous functional differentiation when the cells have grown to confluence. This differentiation is characterized by the development of an apical brush border, expression of high levels of intestinespecific enzymes such as lactase and sucrase, and the formation of a polarized cell layer with domes (48,49). As shown in Fig. 3B, ATF-7 is expressed to a significantly greater degree in the post-confluent, differentiated cells than it is in the preconfluent undifferentiated cells. Interestingly, ATF-7's pattern of expression in these cells is reminiscent of that of PRL-1 (20).
Confirmation and Mapping of the PRL-1⅐ATF-7 Interaction-Coimmunoprecipitation assays were performed to verify the interaction of ATF-7 with PRL-1 in vitro. As shown in Fig.  4A, anti-Myc-tag antibody (9E10) was able to coimmunoprecipitate in vitro translated ATF-7 along with Myc-tagged C104S-PRL-1 (lane 5), whereas control antisera did not (lane 4). The anti-Myc-tag antibody could not coimmunoprecipitate luciferase, a control in vitro translated protein along with Myctagged C104S-PRL-1 (lane 6), indicating that the coimmunoprecipitation of ATF-7 is specific.
In a similar manner, as shown in Fig. 4B, the anti-Myc-tag antibody was able to coimmunoprecipitate in vitro translated C104S PRL-1 along with Myc-tagged ATF-7 (lane 5), whereas control antisera did not (lane 4). The specificity of this experiment was confirmed by demonstrating that the luciferase control protein could not be coimmunoprecipitated along with Myc-ATF-7 (lane 6). Taken together, these results confirm the interaction of PRL-1 and ATF-7 in vitro. To further confirm this interaction, a glutathione S-transferase (GST)-C104S-PRL-1 fusion protein was used in an in vitro binding assay with full-length in vitro translated ATF-7. As shown in Fig. 4C, the GST-C104S-PRL-1 protein bound to glutathione Sepharose beads interacts with full-length ATF-7 (lane 2), but ATF-7 did not interact with the control GST protein (lane 1). We also performed the same experiment using wild type GST-PRL-1 protein and obtained similar results (data not shown).
To determine the regions of PRL-1 that are important for the interaction with ATF-7, six truncated GST-PRL-1 constructs were synthesized and tested for their ability to bind in vitro translated ATF-7. The constructs made spanned different regions of the 189-amino acid full-length PRL-1. As shown in Fig.  5A, GST fused to PRL-1 amino acids 1-96, 60 -118, or 118 -173 are not able to bind to in vitro translated ATF-7, whereas GST constructs fused to PRL-1 amino acids 1-132 and 97-173 could bind to ATF-7. The results with the construct containing amino acids 1-132 are not surprising, because this construct corresponds to the "bait" construct used in the two-hybrid screen. Analysis of these data (see diagram in Fig. 5B) generated the hypothesis that the region comprising amino acids 97-132, which corresponds to PRL-1's PTPase domain, was critical in mediating PRL-1's ability to interact with ATF-7. Accordingly, we synthesized a GST fusion protein containing only amino acids 97-132, and determined, as shown in lane 1 of Fig. 5A, that it was able to bind in vitro translated ATF-7, albeit less efficiently than the aa 97-173 construct (lane 3 of Fig. 5A) or full-length PRL-1 (Fig. 4C). The region contained in the 97-132 construct spans the PRL-1's phosphatase domain and the 15 amino acids that follow it. Neither the phosphatase domain alone nor the 15-amino acid region alone is sufficient for ATF-7 binding, because neither the 60 -118 nor 118 -173 constructs, which contain the complete phosphatase domain or the 15amino acid region, respectively, was able to bind ATF-7. In summary, these results indicate that the PRL-1 PTPase domain, combined with an adjacent small amino acid region, is necessary and sufficient for ATF-7 binding,. In addition, there may be regions in the C-terminal of PRL-1 that contribute to ATF-7 binding, although they are not absolutely required.
To determine the regions of ATF-7 that are important for mediating the interaction with PRL-1, we employed GST bind-

FIG. 3. ATF-7 is expressed in a number of different mouse tissues and is expressed in association with differentiation in Caco-2 cells.
A, total RNA was extracted from mouse tissues using the techniques previously described (32,33), Northern blots were prepared and probed for ATF-7 as described under "Experimental Procedures." The bottom panel shows the ethidium stain of the gel used to prepare the Northern blot. B, Caco-2 cells were grown in Dulbecco's modified Eagle's medium, 10% fetal calf serum, and proliferating cells (Pre) were harvested before they reached confluence. For the differentiated phenotype (Post), cells were allowed to reach confluence, and media was changed every 2 days until 7 days post-confluence when the cell were harvested (73).
ing assays. Full-length (C104S) PRL-1 fused to GST was used in these assays along with two in vitro translated ATF-7 fragments, ATF-7 amino acids 1-138 and ATF-7 amino acids 139 -218. The latter construct consists of only the bZIP region of ATF-7. As shown in Fig. 5C, the ATF-7-139-218 (bZIP domainonly) protein was able to bind to GST-C104S-PRL-1 (lane 6), whereas the large region upstream of the bZIP region by itself was not able to bind GST-C104S-PRL-1 (lane 5). As expected, neither fragment was able to bind to GST alone (lanes 3 and 4). These data indicate that the bZIP region is necessary and sufficient for PRL-1 binding. Taken together, it can be concluded that the PRL-1⅐ATF-7 interaction is mediated by interaction of PRL-1's PTPase domain and ATF-7's bZIP domain.
Ability of PRL-1 to Selectively Dephosphorylate ATF-7 in Vitro-Because our data indicated that the PRL-1 phosphatase domain was necessary for interaction with ATF-7, we sought to determine whether PRL-1 is capable of dephosphorylating ATF-7 in vitro. We used c-Src kinase to phospholabel GST-ATF-7 on tyrosine. This kinase was not able to phosphorylate GST alone (data not shown). We split the products of the kinase reaction into equal aliquots and then used each in a phosphatase assay using either PRL-1, C104S inactive PRL-1 (MUT), or buffer. Because all three phosphatase reactions derive from the same common kinase reaction, the ratio of labeled ATF-7 to labeled c-Abl must be constant among the three tubes before the addition of the phosphatase or control. This ratio would not be affected by uneven division of the kinase reaction or uneven loading of the gel, because there would be more or less of both proteins in the same proportion. No change in this ratio would be seen if PRL-1 nonspecifically dephosphorylated both c-Src and ATF-7, because the phosphorylation of both would decrease. The degree to which this ratio decreases thus reflects selective dephosphorylation of ATF-7 by PRL-1. A representative result is shown in Fig. 6A. We found that significantly less ATF-7 remained phosphorylated relative to c-Abl after treatment with PRL-1 than after treatment with the C104S-PRL-1 or buffer controls. This indicates that PRL-1 was able to selectively partially dephosphorylate the tyrosine-phosphorylated ATF-7. The results of four separate experiments were quantified by densitometry and are shown in Fig. 6B. PRL-1 was significantly (p Ͻ 0.01) more able to dephosphorylate the labeled ATF-7 than either C104S-PRL-1 or buffer alone control. DISCUSSION Using the yeast two-hybrid system, we have identified ATF-7 as a novel bZIP protein that interacts with the PRL-1 nuclear PTPase. The interaction of PRL-1 and ATF-7 has been confirmed using GST binding and coimmunoprecipitation assays, and the sites of interaction have been mapped to include PRL-1's phosphatase domain and ATF-7's bZIP domain.
An important issue is the role of ATF-7 and its partner PRL-1 in cellular differentiation. There are a number of items that support the existence of such a role. We have previously shown that PRL-1 expression is associated with differentiation (20). Recently, we have also found that PRL-1 is expressed both in the adult and during development in a number of differentiating epithelial tissues, including intestine, stomach, kidney, and lung (17). We have also found that PRL-1 is expressed in 3T3-L1 adipocytes in association with differentiation, 2 and we identified the ATF-7⅐PRL-1 interaction by two-hybrid screening of a 3T3-L1 adipocyte library. We have further shown here that ATF-7 is expressed to a much greater degree in postconfluent, differentiated Caco-2 cells than in preconfluent undifferentiated cells, a pattern reminiscent of that of PRL-1. All of these findings suggest that ATF-7 may play an important role in the development and maintenance of differentiating epithelial tissues.
Ultimately, experiments involving ectopic overexpression or ablation of ATF-7 in specific cells and tissues will be necessary to determine whether it has a direct role in modulating cellular differentiation. A potential mechanism might involve the ZIPK/ DLK kinase (50). It is interesting to note that the highly homologous ATF-4 protein has been shown to interact with this protein, which in turn has been linked to both differentiation and apoptosis (51,52). Concomitant roles in differentiation and apoptosis are plausible in the intestine, where enterocytes sequentially pass through proliferation, differentiation, and apoptosis phases during their life cycle (53). Agents that induce differentiation in several tissue models have also been shown capable of promoting apoptosis. One example is the ability of butyrate to sequentially induce these two processes in intestinal cells (54 -56).
We show here that tyrosine-phosphorylated ATF-7 can be selectively dephosphorylated by PRL-1 in vitro. The control of transcription through the regulation of transcription factor phosphorylation is a well-established concept (1,2,5,6). However, our results must be interpreted with caution. Although they are consistent with the possibility that PRL-1's cellular role may involve dephosphorylating ATF-7, much more work will need to be done before this can be established. We have observed only partial dephosphorylation of the labeled ATF-7 in these experiments, a result that could indicate that proper reaction condi- tions are not present, or that ATF-7 is phosphorylated at multiple sites, only some of which are dephosphorylated by PRL-1. However, it is also possible that the basis for the PRL-1⅐ATF-7 interaction is not that of a phosphatase⅐substrate interaction. Future studies will be geared toward analyzing whether this phenomenon occurs in vivo, mapping the specific residue(s) that are affected, and determining the transcriptional consequences of such a reaction. If ATF-7 itself is not a target of PRL-1, it may serve to bring PRL-1 into proximity with its true substrates. In this manner, PRL-1 could influence transcription by acting on transcriptional cofactors or elements of the basal transcription machinery. In some situations, kinases or phosphatases may bind transcription factors but influence transcription by acting on proteins other than the transcription factors themselves. A classic example of this phenomenon is the regulation of the NFB transcription factor by phosphorylation of its sequestering inhibitor IB, which leads to IB's degradation (7,57). Another example is the nuclear tyrosine kinase c-Abl, which binds to p53 and increases its transactivating ability without phosphorylating it.
It is thought that Abl may be able to execute this function by phosphorylating the C-terminal domain of RNA polymerase II, which is known to be extensively phosphorylated on tyrosine (14,15). Alternatively, ATF-7 may have extra-transcriptional roles involving sequestration of specific proteins (e.g. ZIPK) with roles in the regulation of apoptosis or differentiation, as has been proposed for the highly homologous ATF-4 protein, (51,52,58). PRL-1 might impact upon these processes by dephosphorylating ATF-7 itself, or by acting on the target proteins bound by ATF-7.
We have determined that ATF-7 homodimers bind to CRE sites, and we have not found evidence that ATF-7 homodimers can bind to AP1 or C/EBP sites. Interestingly however, the only other published data about ATF-7 is the description of a partial clone, comprising only ATF-7's bZIP region, that was identified in a Far Western screen as interacting with the C/EBP␥ transcription factor (59). This suggests that ATF-7 can interact with C/EBP family members, which are known to play important roles in the differentiation and development of a number of tissues, including liver and intestine (60 -63). One manner in which this FIG. 6. Tyrosine-phosphorylated ATF-7 can be specifically dephosphorylated in vitro by PRL-1. A, GST-ATF-7 was expressed in bacteria, purified, and tyrosine-phosphorylated with 32 P using c-Src as outlined under "Experimental Procedures." After the kinase reaction was stopped, the beads were washed four times and split into three equal aliquots. Equal amounts of active or mutant PRL-1, or buffer (negative control), were added to each tube, which were then incubated for 60 min at 37°C. The phosphatase reaction was terminated by the addition of equal volumes of 2ϫ Laemmli buffer. The products were then boiled and run on an SDS-PAGE gel, which was then dried and exposed to x-ray film (Kodak). Because all three phosphatase reactions derive from the same common kinase reaction, the ratio of labeled ATF-7 to labeled c-Src must be constant among the three tubes before the addition of phosphatase or control. (This ratio would not be affected by uneven division of the kinase reaction or uneven loading of the gel, because there would be more or less of both proteins in the same proportion.) No change in this ratio would be seen if PRL-1 nonspecifically dephosphorylated both c-Src and ATF-7, because the phosphorylation of both would decrease. The degree to which this ratio decreases thus reflects selective dephosphorylation of ATF-7 by PRL-1. A representative result is shown in A. We found that significantly less ATF-7 remained phosphorylated relative to c-Src after treatment with PRL-1 than after treatment with the C104S-PRL-1 or buffer controls. This indicates that PRL-1 was able to selectively partially dephosphorylate the tyrosine-phosphorylated ATF-7. The results of four separate experiments were quantified by densitometry and are shown in B. PRL-1 was significantly (p Ͻ 0.01) more able to dephosphorylate the labeled ATF-7 than either C104S-PRL-1 or buffer alone control. could occur is through the formation of ATF-7-C/EBP heterodimers. It is a hallmark of bZIP proteins that they extensively heterodimerize with each other, both within and outside of individual families. These interactions are not completely promiscuous but are instead restricted by a specific and complex "dimerization code" (64,65). In some cases, cross-family heterodimers bind to the preferred site of one of the dimer members (66,67), whereas in other cases, the heterodimers bind to novel composite sites (45-47, 68 -70). In some cases, dimerization occurs only with members of other bZIP families. In this regard, it is interesting to note that ATF-4, the bZIP protein most similar to ATF-7, can form heterodimers with members of the C/EBP family, including C/EBP␤ and C/EBP␥ (45,46,68,71), but has not been reported to heterodimerize with any other member of the ATF/CREB family. In these situations, the heterodimeric complexes bind to either C/EBP sites or to ATF-C/EBP composite sites (45)(46)(47)70), although ATF-4 homodimers do not bind to either of these sites. In addition, ATF-4 has also been reported to form heterodimers with the AP1 proteins Fos and Jun (72). It is possible that ATF-7 may play an important role in transcriptional regulation in a similar manner as ATF-4 through interaction with C/EBP and/or AP1 family members. Future experiments will be designed to address this issue.
In summary, the identification of the ATF-7⅐PRL-1 interaction not only provides information about how PRL-1 may bring about the phenotypic states with which it is associated but also may have important implications for our understanding of the transcriptional regulation of target genes that modulate differentiation.