Structure and expression of the ATFa gene.

The human ATFa proteins belong to the ATF/CREB family of transcription factors. We have previously shown that they mediate the transcriptional activation by the largest E1a protein and can heterodimerize with members of the Jun/Fos family. ATFa proteins have also been found tightly associated with JNK2, a stress-activated kinase. We now report on the structure of the ATFa gene, which mapped to chromosome 12 (band 12q13). Sequence analysis revealed that ATFa isoforms are generated by alternative splice donor site usage. A minimal promoter region of ∼200 base pairs was identified that retained nearly full transcriptional activity. Binding sites for potential transcription factors were delineated within a GC-rich segment by DNase I footprinting. Expression studies revealed that ATFa accumulates in the nuclei of transfected cells, and the nuclear localization signal was defined next to the leucine zipper domain. As revealed by hybridization with mouse ATFa sequences, low levels of ATFa mRNAs were ubiquitously distributed in fetal or adult mice, with enhanced expression in particular tissues, like squamous epithelia and specific brain cell layers. The possible significance of coexpression of ATFa, ATF-2, and Jun at similar sites in the brain is discussed.

Gene expression in eukaryotic cells is regulated by sequencespecific binding of transcription factors to cis-acting DNA elements. One example of such regulatory elements is the ATF 1 / CRE motif (TGACGTCA), which was originally identified in several adenovirus promoters as the ATF-binding site and in some cellular promoters as the CRE. Many cDNAs encoding ATF/CRE-binding proteins have now been cloned (for reviews, see Refs. [1][2][3]; the binding and trans-activation properties of all these proteins have been extensively studied. They belong to the family of "b-Zip" (basic-leucine zipper) proteins, which have a basic region in their C termini, rich in positively charged amino acids, flanked by a leucine zipper motif, and responsible both for sequence-specific DNA binding and protein dimerization (4,5). Many of these ATF/CREB factors have been found heterodimerized with members of the AP1 family (6 -10), giving new combinations of factors with different binding capacities, which may increase their response pattern. The ATF/CREbinding proteins include factors implicated in the cAMP response pathway (CREB, CREM, and ATF-1) (3) as well as closely related factors that mediate promoter induction by the largest product of the adenovirus E1a oncogene, CRE-BP1 (also called ATF-2) (11)(12)(13), and the ATFa proteins (14,15).
Four variants of ATFa (ATFa0, ATFa1, ATFa2, and ATFa3) have been characterized that have indistinguishable properties and differ only by short peptide motifs (15,16). Although these ATFa proteins had no detectable intrinsic trans-activation function, a transcriptional activation domain was identified after removal of C-terminal sequences. This cryptic activation domain comprises an essential N-terminal zinc-binding element, which, together with sequences located within the 170 C-terminal residues of the ATFa proteins, also contributes to the interaction of ATFa with the E1a oncoprotein. As expected, both N-and C-terminal ATFa elements are instrumental in mediating E1a responsiveness of target promoters (15). In vitro and in vivo experiments have demonstrated that the ATFa proteins can associate with c-Jun or c-Fos proteins and that they bind TPA response element sequences only when heterodimerized with members of the Jun family (10,17).
We also found that ATFa proteins were tightly associated with protein kinase activities and that this binding occurred both in vitro and in a cellular environment. We have shown that these kinase activities have properties related to specific members of the mitogen-activated protein kinase family (18). One of these kinases is the c-Jun N-terminus-associated kinase 2 (JNK2), a kinase that binds to N-terminal regions of ATF-2 and ATFa (18 -21). It has been suggested that ATF-2 plays important roles in the early events of signaling pathways, as phosphorylation of this factor by the c-Jun N-terminus-associated kinases increases its trans-activating properties and permits the early activation of the c-jun promoter (22). Despite their structural similarity to ATF-2, it is presently unclear whether the ATFa proteins serve similar functions in the cell, particularly because they display no transcriptional activity on their own. As a step toward understanding the cellular functions of the ATFa proteins and their connections with the ATF-2 and c-Jun factors, we undertook the cloning of the genomic ATFa sequences and the study of their expression. The structure and chromosomal location of the ATFa gene and promoter were determined. Our results indicate that the different ATFa isoforms arise by alternative splicing through differential splice donor usage and that these proteins are targeted to the cell nucleus by a bipartite nuclear localization sequence that is part of the basic region of the protein. While low levels of ATFa mRNAs appeared to be expressed rather ubiquitously in the adult mouse, enhanced expression was observed in particular tissues during embryonic development.

EXPERIMENTAL PROCEDURES
Human Gene Mapping-In situ hybridization was carried out on chromosome preparations obtained from phytohemagglutinin-stimulated human lymphocytes. A recombinant pBluescript plasmid containing an ATFa cDNA fragment (1449 bp) was tritium-labeled by nick translation to a specific activity of ϳ1.6 ϫ 10 8 dpm/g and was used as probe. The hybridization procedure was carried out essentially as described previously (23).
Genomic Cloning-A human placental genomic DNA library (constructed by J.-M. Garnier in GEM 12) was screened by hybridization with a random-primed 32 P-labeled ATFa3 cDNA probe using conventional cloning methods (24). DNA inserts were recovered from positive recombinant phages by restriction digestion and subcloned into the pBluescript SK vector. Nucleotide sequencing was performed by the chain termination method (68) using synthetic oligonucleotide primers derived from appropriate regions of the known ATFa cDNA sequence.
Mouse ATFa cDNA Fragment Cloning-A 300-bp cDNA fragment was amplified by reverse transcription-PCR from poly(A) ϩ RNA from mouse P19 cells (25). The oligonucleotide primers used for the reaction bracketed a region unique to ATFa (when compared with other members of the family), spanning positions ϩ803 to ϩ1102 of the human ATFa cDNA and located just 5Ј to the basic-leucine zipper motif (b-Zip). The resulting 300-bp fragment was inserted into the EcoRV site of pBluescript and sequenced, revealing a 92% identity to its human homolog at the nucleotide level.
Recombinant Plasmid Construction-Subcloning of the ATFa gene promoter region was performed as follows. An 11-kb SacI insert was recovered from a positive recombinant phage and cloned into pBluescript SK. A 6.5-kb ApaI fragment containing the first exon, upstream sequences, and part of the first intron of the ATFa gene was directly subcloned into the ApaI site of pBluescript SK. This recombinant plasmid was further processed by excising an XbaI fragment comprising the most 5Ј-portion of the ATFa gene upstream sequences and was mutagenized by oligonucleotide-directed mutagenesis to remove the intronic sequences and to create a SacI restriction site at the end of the first exon. A 2-kb SacI fragment containing the first exon and upstream sequences of the ATFa gene (between positions Ϫ1917 and ϩ83, with respect to the transcription start site) was then subcloned, in both orientations, in front of the chloramphenicol acetyltransferase (CAT) reporter gene of a promoterless vector (pBLCAT6) (26), generating the Ϫ1917/ϩ83 CAT and ϩ83/Ϫ1917 CAT recombinants. From the Ϫ1917/ ϩ83 CAT recombinant, a series of additional deletions were constructed by restriction digestion (giving rise to Ϫ62/ϩ83 CAT and Ϫ212/ϩ83 CAT), by PCR amplification of specific promoter fragments followed by subcloning into the same host vector (giving rise to Ϫ23/ϩ83 CAT and ϩ8/ϩ83 CAT), or by progressive exonuclease III-directed unilateral deletions of upstream sequences between positions Ϫ1917 and Ϫ797 (generating Ϫ1719/ϩ83 CAT, Ϫ1591/ϩ83 CAT, Ϫ1467/ϩ83 CAT, Ϫ1211/ϩ83 CAT, Ϫ1051/ϩ83 CAT, and Ϫ797/ϩ83 CAT). The abovementioned 2-kb SacI fragment was also subcloned in both orientations into a blunt-ended XbaI site at position Ϫ9 of a rabbit ␤-globin reporter gene of the PAL4 vector (27), generating the Ϫ1917/ϩ83 Glob and ϩ83/Ϫ1917 Glob recombinants.
The GST-ATFa derivatives were constructed as described previously (18) by inserting the ATFa3 cDNA or selected fragments thereof into the pBC expression vector (17). The GST-SVTNLS recombinant, which contains a 30-bp oligonucleotide encoding the NLS element of the SV40 large T antigen (PKKKRKVEDP) (28), was kindly provided by Dr. J. Ménissier de Murcia (CNRS, Strasbourg, France).
Transfections and CAT Assay-HeLa cells were grown in Dulbecco's modified Eagle's medium supplemented with 2.5% fetal calf serum and 2.5% newborn calf serum. 16 h after plating, the cells were transfected by calcium phosphate coprecipitation (29) with the amounts of supercoiled recombinant DNA indicated in the figure legends, adjusted to 13 g/9-cm Petri dish with double-stranded carrier DNA (pBluescript). The medium was changed after 15 h, and cells were harvested after an additional 24 h. Extracts were prepared, and aliquots (normalized by protein concentration) were assayed for CAT activity as described previously (30,31). Each transfection experiment was repeated at least three times with different plasmid preparations. Percent acetylation of chloramphenicol was determined by thin-layer chromatography and quantitated with a Bioimaging analyzer (Fuji Photo Film Co., Ltd.).
RNA Analysis-Total RNA from various cell lines and tissues was prepared using the single step guanidinium isothiocyanate/phenol procedure (25), and poly(A) ϩ RNA was subsequently purified on oligo(dT)cellulose (Pharmacia Biotech Inc.). For Northern blot analysis, this RNA was separated by electrophoresis on 1% formaldehyde-agarose gels, transferred to nitrocellulose filters, and hybridized with randomprimed 32 P-labeled ATFa3 or actin cDNA probes at 42°C in buffer containing 5 ϫ SSC and 50% formamide. Filters were washed at 25°C in 0.1 ϫ SSC and 0.1% SDS and exposed for autoradiography. Transcription start sites on the ATFa gene were mapped by RNase protection, reverse transcriptase primer extension, and S1 nuclease mapping essentially as described (32). RNase protection mapping was carried out by hybridizing HeLa cell poly(A) ϩ RNA to a uniformly 32 P-labeled antisense riboprobe produced as a T3 polymerase transcript of a pBluescript recombinant harboring a PCR-generated Ϫ153/ϩ83 fragment of the ATFa gene. Reverse transcriptase primer extension and S1 nuclease mapping were carried out with total RNA from transfected HeLa cells. In the case of reverse transcriptase primer extension mapping, a synthetic oligonucleotide spanning the ATFa transcribed strand from positions ϩ51 to ϩ81 was used as primer for the reverse transcriptase reaction. The same oligonucleotide was also used to sequence the Ϫ1917/ϩ83 CAT template, providing a size reference sequence ladder. For S1 nuclease mapping, the probe was a single-stranded DNA fragment that was produced by extension of the 5Ј-end-labeled ϩ51/ϩ81 oligonucleotide with DNA polymerase I (Klenow fragment) using the heat-denatured Ϫ1917/ϩ83 CAT recombinant plasmid as template. After cleavage with XmnI (position Ϫ427), the resulting single-stranded product (Ϫ427/ϩ81) was recovered and purified by gel electrophoresis.
DNase I Footprinting-HeLa whole cell extracts (10 g) were mixed with the PCR-generated Ϫ153/ϩ83 fragment of the ATFa gene ( 32 P-5Јend-labeled on either the transcribed or nontranscribed strand; ϳ10,000 cpm) in a final volume of 20 l containing 100 ng of poly(dI-dC), 0.05% Nonidet P-40, 5 mM MgCl 2 , 50 mM Tris-HCl, pH 7.9, 0.2 mM dithiothreitol, 15% glycerol, and 40 mM ammonium sulfate. After preincubation for 10 min at 25°C, 2 l of DNase I solutions (at 100, 40, 20, or 10 g/ml) were added to the mixture and further incubated for 2 min at 25°C. The reaction was stopped by addition of 0.4 ml of a mixture containing 0.5% SDS, 50 mM sodium acetate, pH 5, and 50 g/ml Escherichia coli tRNA. The DNase-resistant fragments were purified and separated on denaturing polyacrylamide gels.
Immunoblot (Western Blot) Analysis-Whole cell extracts from adult mouse tissues and nuclear extracts from 11.5-17.5-day-old mouse embryos were prepared (33,34). After SDS-polyacrylamide gel electrophoresis, the ATFa proteins were revealed by immunoblotting essentially as described (10) with a monoclonal antibody directed against bacterially expressed ATFa3 (monoclonal antibody 1A7; used at 1:5000). Specific protein-antibody complexes were revealed using a peroxidase-conjugated rabbit anti-mouse -light chain secondary antibody (Jackson ImmunoResearch Laboratories, Inc.), followed by treatment with the ECL detection system (Amersham Corp.) according to the manufacturer's instructions.
In Situ Hybridization-The organs collected from adult mice were frozen in 2-metylbutane maintained on dry ice. A 300-bp mouse ATFa cDNA fragment cloned into pBluescript (see above) was transcribed in vitro using T7 RNA polymerase in the presence of 35 S-CTP to generate an antisense riboprobe. Hybridization was carried out on cryosections as described (35), providing a higher signal-to-noise ratio, although with a somewhat poorer histology. Emulsion autoradiography was for 4 months.
Immunocytochemistry-Monolayer COS-7 cells (grown on glass coverslips) were transfected by calcium phosphate coprecipitation with 1 g of recombinant vector or pBC parental vector, together with 3.5 g of double-stranded pBluescript carrier DNA/tube. The medium was changed after 20 h, and the day after, the cells were washed in PBS and treated for 4 min at room temperature with 2% formaldehyde in PBS. After fixation, the cells were permeabilized by two treatments with PT buffer (PBS, 0.1% Triton X-100) for 10 min. After a 30-min treatment with blocking solution (2% nonfat dry milk in PBS), the cells were incubated for 1 h at room temperature with an anti-GST primary monoclonal antibody (EuroMedex, Paris, France) diluted in PT buffer, washed three times in PT buffer, incubated with a Cy3-conjugated goat anti-mouse secondary antibody for 1 h at room temperature, and washed in PT buffer. The preparations were mounted in glycerol/PBS (4:1) containing 5% propyl gallate, and observations were made with a fluorescence microscope (Nikon Microphot FXA).

RESULTS
Chromosomal Localization of the ATFa Gene-To determine the genomic localization of the ATFa sequences, in situ hybridizations on human metaphase chromosomes were performed (Fig. 1). In the 100 metaphase cells examined after hybridization, 199 silver grains were associated with chromosomes, and 67 of these (33.6%) were located on chromosome 12. The grain distribution was not random since 89.5% (60/67) of the grains counted on chromosome 12 mapped to position 12q12-14, with a maximum in band 12q13. It is noteworthy that the gene encoding the related ATF-1 factor (36) also maps within this locus.
The ATFa Gene-A human genomic library was screened at elevated stringency with the largest cDNA of the ATFa family. A number of positive clones were picked, characterized by Southern blot analysis, and subcloned. Partial sequence determination of the genomic clones and comparison with the previously established cDNA sequence (14) allowed us to position the introns and to establish the gene structure. As depicted in respectively. The lengths of the 11 introns have not been precisely determined, but were estimated to have a mean size of ϳ4 kb each. The sequences at the exon/intron junctions were in very good agreement with the consensus sequences around both splice donor and acceptor sites (Fig. 3A).
Four ATFa cDNA variants were originally isolated from human cDNA libraries (14 -16), differing from each other by specific sequence elements within the coding region (Fig. 3B): ATFa3 corresponds to the largest cDNA resulting from the juxtaposition of all elements boxed in Fig. 2; ATFa2 lacks a 33-bp element (element d); ATFa1 lacks both element d and an additional 63-bp element (element e); and ATFa0 lacks element d and exons F-I. The existence of cytoplasmic transcripts corresponding to these ATFa variants was confirmed by PCRmediated analysis of HeLa cell poly(A) ϩ RNA (16), 2 demonstrating the physiological relevance of these variants. These observations indicate that the various isoforms most likely result from differential splicing of a unique ATFa primary transcript. Thus, exons F-I correspond to alternative exons. Furthermore, examination of the nucleotide sequences at the 5Ј-border of elements d and e revealed strong homologies to the consensus splice donor sequence (Fig. 3C), suggesting that these sequences constitute alternative splice donor sites (DS1) that can be used in place of the donor sites located at the 3Ј-border of these elements (DS2). Nothing is known at present about the control of this splice donor site usage and the underlying mechanism.
The Promoter of the ATFa Gene-As a first step toward the delineation of the promoter region of the ATFa gene, we determined the transcription start sites. An RNase protection experiment performed with HeLa cell poly(A) ϩ RNA (Fig. 4B) clearly identified one minor and two major start sites, arbitrarily denoted Ϫ43, Ϫ22, and ϩ1, respectively, within the 5Ј-region of the ATFa leader (exon A). Transcription initiation around position Ϫ22 seems to be less precise, as indicated by the slight scattering of the start sites in this area. The position of this initiation region is compatible with the coordinates of the most 5Ј-extending ATFa cDNA originally isolated (14), starting at position ϩ8.
Examination of the nucleotide sequences located upstream of the transcription start sites did not reveal a typical TATA box. Instead, a significantly higher GC content (70%) was apparent between positions ϩ1 and Ϫ140, which, together with the relative spreading of initiation sites, is a characteristic of promoters of housekeeping genes (37). To test for the presence of promoter sequences within the sequences located upstream of the ATFa leader, a vector harboring ATFa sequences between positions Ϫ1917 and ϩ83 was transfected into HeLa cells. To ensure stable RNA synthesis, the ATFa sequences were linked to a promoterless rabbit ␤-globin gene. Specific transcription was investigated by S1 nuclease mapping and reverse transcriptase primer extension of total RNA prepared from the transfected HeLa cells. As shown in  (d and e) are represented as open and closed boxes, respectively, with corresponding lengths (in bp) above (exon L has not been sequenced entirely, but its size has been estimated electrophoretically). Transcription initiation sites, as determined in Fig. 4, are indicated. The coordinates (with respect to the ϩ1 transcription start site) of the translation initiation and termination codons as well as those of the extremities of each exon within the largest cDNA (ATFa3), are also given. initiated at positions ϩ1, Ϫ22, and Ϫ43 were detected in cells transfected with the ATFa/globin vector in which the ATFa sequences were in the sense orientation ( lanes 5 and 7). No transcription could be detected from a vector harboring the ATFa sequences in the antisense orientation with respect to the globin sequences (lanes 6 and 8). These results not only confirm the positions of the start sites of the ATFa gene, but indicate that the isolated ATFa 2-kb upstream sequences carry sufficient information to direct efficient transcription from the transfected construct.
To more precisely delineate the essential promoter region of the ATFa gene, larger deletions were created at the 5Ј-side of the Ϫ1917/ϩ83 ATFa gene fragment. The truncated segments were inserted in front of the bacterial CAT coding sequences, and their capacity to direct CAT transcription was assessed by measuring ATFa sequence-dependent CAT activity of cells transfected with the recombinant ATFa/CAT vectors (Fig. 5). In agreement with the S1 nuclease and reverse transcriptase mapping experiments (Fig. 4C), no CAT activity could be detected after transfection of a recombinant in which the Ϫ1917/ ϩ83 ATFa fragment was linked in the antisense orientation to the CAT sequences. By contrast, substantial enzymatic activity was measured when the ATFa promoter region was inserted in the sense orientation. Deletion of ATFa sequences between positions Ϫ1917 and Ϫ212 resulted in only modest effects on promoter activity, as reflected by the small fluctuations of relative CAT activities from one recombinant to the other: these activities varied by no more than 1.5-fold above or below the activity of the recombinant carrying the largest ATFa fragment. Further deletion to position Ϫ62 reduced ATFa promoter activity ϳ5-6-fold compared with that of the Ϫ1917/ϩ83 or Ϫ212/ϩ83 fragment. Finally, removal of the next 40 bp (to position Ϫ23) had the most dramatic effect on promoter activity, reducing it to undetectable levels. Together, these results clearly indicate that important ATFa promoter elements are located essentially between positions Ϫ212 and Ϫ62 and between positions Ϫ62 and Ϫ23.
Putative protein-binding sites within the minimal ATFa promoter fragment were mapped by the DNase I footprinting assay with HeLa cell extracts (Fig. 6). Areas protected against the nuclease as well as hypersensitive sites were observed on both DNA strands (Fig. 6). Interestingly, one cluster of protected and hypersensitive sites was located between positions Ϫ131 and Ϫ83 (i.e. within the Ϫ212/Ϫ62 element defined by deletion analysis), and a second, weaker area of interactions was identified between positions Ϫ56 and Ϫ33 (i.e. within the Ϫ62/Ϫ23 element defined above). The nature of the proteins interacting with these sequence elements is presently unknown, but their crucial contribution to promoter activity is clearly illustrated by the deleterious effect of the successive deletion of the corresponding recognition sites (Fig. 5). Interestingly, a computer search for known protein-binding sequences revealed potential target sites for several factors within the Ϫ131/Ϫ83 upstream protected region (see Fig. 6). On the other hand, no obvious candidate was found for the Ϫ56/Ϫ33 area, a region that is flanked by two putative AP2-  (14,15) or by PCR amplification (ATFa0) (16), is depicted. Carets span excised sequences. Areas encoding the zincbinding domain (ZF) and the b-Zip motif are marked. C, the exon polymorphism generated by two alternative splice donor sites (DS1 and DS2) on both sides of elements d and e is illustrated. Nucleotide sequence of both alternative elements is given, and conserved nucleotides with respect to the consensus splice donor site sequence are underlined.
Nuclear Localization of the ATFa Protein-Immunofluorescent staining experiments with ATFa-specific antibodies show that ATFa proteins localize to the nucleus when overexpressed in transfected HeLa cells. 3 In an attempt to map the peptidic elements (NLS) mediating ATFa nuclear localization, a selected series of ATFa fragments fused to GST were expressed in COS-7 cells after transfection of the corresponding vectors (Fig.  7A). The GST fusion proteins were revealed by indirect immunofluorescence with a monoclonal antibody directed against the GST moiety (Fig. 7B). Control experiments indicated that the GST protein was by itself essentially cytoplasmic (Fig. 7B,  panel a), whereas it was clearly directed to the nucleus when fused to the NLS of the SV40 large T antigen (panel b). Similarly, the full-length ATFa3 protein relocalized GST to the nucleus (Fig. 7B, panel 1), while deletion of the ATFa sequence C-terminal to residue 293 abolished nuclear accumulation of the chimeric protein (panel 2). A truncated ATFa version lacking the 326 N-terminal residues retained full nuclearization activity (Fig. 7B, panel 3), but derivatives with deletions extending to positions 348 and 368 partially or completely lost this activity (panels 4 and 5, respectively). In agreement with the ability of an ATFa segment spanning residues 327-377 to direct GST to the nucleus (Fig. 7B, panel 6), these results indicate that elements located between residues 327 and 368 are critically involved in the nuclear transportation of the ATFa protein. A comparison with reported NLS sequence compilations points to several elements (underlined in Fig. 7C) located within or flanking the basic region of the ATFa b-Zip motif and roughly spanning residues 327-368.
Tissue Distribution of ATFa Gene Expression-We first examined the expression of mRNAs from various members of the ATF family. As shown in Fig. 8, Northern blot analysis of HeLa cell transcripts (using specific probes of equivalent radioactivities) revealed that the ATFa-specific RNAs (ϳ9.5 kb in size) were the least abundant compared with the levels of the ATF-1, CRE-BP1/ATF-2, and CREB transcripts (ϳ2.9, 5, and 5.5 kb in size, respectively, in agreement with reported data) (39 -41). The relative intensities of the corresponding signals were ϳ1: 20:5:20, respectively. If translation efficiencies and protein stabilities are comparable in each case, this suggests that the ATFa and CRE-BP1/ATF-2 proteins are minor species within HeLa cells. A survey of a series of transformed human cell lines (originating from colon, cervix, breast, promyelocytes, chorion, endometrium, or fetal kidney) confirmed that the steady-state levels of ATFa transcripts were reproducibly low in each cell line tested (Fig. 9A, lanes 1-8).
To determine ATFa expression in normal cells, we chose to examine the level of ATFa-specific RNAs in adult mouse organs. As in human cells, a 9.5-kb transcript was detected with similar intensities in various organs (Fig. 9A, lanes 9 -12; and data not shown), suggesting that there is no striking tissue specificity of ATFa expression in adults.
To detect ATFa transcripts at the cellular level, in situ hybridization was performed on mouse cryosections (Fig. 10) using a 35 S-labeled antisense riboprobe. A control probe derived from the same DNA template, but transcribed in the sense orientation, was used on adjacent histological sections and showed no preferential labeling (data not shown). Weak and diffuse signals were detected in several adult organs (ovary, testis, liver, gut, kidney, adrenal gland, spleen, lymph node, thymus, skeletal muscle, and blood vessels) (data not shown). Interestingly, however, enhanced labeling was detected in several squamous epithelia such as those of the vagina (Fig. 10E), the left part of the stomach (Fig. 10D), or the tongue (not shown). Note that the labeling was restricted to the basal and suprabasal cell layers, i.e. the two layers adjacent to the basement membrane. Increased labeling was also seen in pseudostratified epithelia such as those of the epididymis (Fig.  10F), ductus deferens (Fig. 10G), seminal vesicle (Fig. 10H), and large bronchi (not shown). Strong expression was detected in the myocardium of the ventricles and in the atria (Fig. 10B) as well as in laryngeal muscle (Fig. 10C) and tracheal cartilage (not shown). On brain sections, specific signals were seen in the Purkinje cells of the cerebellum as well as in the limbic lobe (hippocampus and dentate gyrus) and piriform cortex (Fig. 10A).
In situ hybridization on embryo sections at various developmental stages showed weak and ubiquitous ATFa signals throughout the embryo. However, expression above the basal level was clearly detected in the seminiferous tubules of the fetal testis (at 14.5 days post-coitum) and cranial nerve and dorsal root ganglia (at 14.5 and 16.5 days post-coitum, respectively) and also in the squamous to glandular transition area of the stomach epithelium (at 16.5 days post-coitum) (data not shown).
Finally, the relative levels of ATFa protein in mouse organs 3 B. Chatton, unpublished observation.

FIG. 4. Mapping of the ATFa gene transcription start site.
A, a diagram of the probes and primers used for S1 nuclease mapping, RNase protection (RP), and reverse transcriptase primer extension (RT) or sequencing (see "Experimental Procedures" for details) is presented. Coordinates were given a posteriori with respect to the ϩ1 start site. B, RNase protection was carried out with 2.5 g of HeLa cell poly(A) ϩ RNA, adjusted to 10 g with E. coli tRNA (ϩ; lane 2). A control reaction performed in the absence of HeLa cell RNA (Ϫ; lane 1) reveals nonspecific signals (asterisks). Positions of DNA size markers (in bp) are given on the left. Arrows with coordinates (on the right) refer to the end point positions of protected RNA fragments. C, S1 nuclease mapping (lanes 5 and 6) and reverse transcriptase primer extension mapping (lanes 7 and 8) were carried out with 10 g of total RNA from HeLa cells that were transfected with 7.5 g of the Ϫ1917/ϩ83 Glob (ϩ; lanes 5 and 7) or ϩ83/Ϫ1917 Glob (Ϫ; lanes 6 and 8) recombinant plasmid. Sequence reactions primed with the ϩ81/ϩ51 oligonucleotide on the Ϫ1917/ϩ83 CAT template were run in parallel (lanes 1-4). The resulting ladders were used to position S1 nuclease mapping and reverse transcriptase primer extension signals: relevant nucleotide sequences (transcribed strand) are given on the left, with initiation sites highlighted. and in staged whole mouse embryos were determined by Western blot analysis (Fig. 9B). The monoclonal antibody (1A7) used in these experiments was directed against a peptide spanning residues 296 -307, a region unique to the ATFa proteins and sharing no significant homology with ATF-2, the most closely related protein. This antibody revealed weak but specific signals in the various adult mouse organs tested, reflecting low levels of ATFa protein (Fig. 9B) and therefore confirming the rather ubiquitous expression suggested by the RNA distribution studies (Figs. 9A and 10). Strikingly, although a weak signal could be detected on longer exposures, the ATFa protein content of the heart (Fig. 9B, lane 11) did not seem to parallel the specific RNA signal observed by in situ hybridization (Fig.  10B). Whether this reflects a reduced translation of the ATFa mRNAs in this tissue remains to be established. More important, overall ATFa expression was maximal in mouse embryos younger than 13.5 days and progressively decreased from day 14.5 onward (Fig. 9B, lanes 1-6). This observation suggests that the ATFa proteins may play essential roles in early embryonic development.

DISCUSSION
The ATFa Sequences Localize within a "Hot" Genomic Area-The human genomic sequences encoding the ATFa proteins have been localized on chromosome 12. Interestingly, these sequences map within a locus (12q12-14) that harbors a num-ber of genes playing essential functions in the control of cell homeostasis and proliferation: the hoxC cluster (42), an adenyl cyclase gene (43), the cdk2 and cdk4 genes (44), a member of the ras proto-oncogene family (rap1B) (45), the MDM2 gene (46), the ATF-1 gene (36), the C/EBP-related CHOP or GADD153 gene (47,48), as well as the genes of the retinoic acid receptor-␥ (49), vitamin D receptor (50), and Sp1 (51) transcription factors. Whether the clustering of genes encoding such regulatory proteins is fortuitous or unique in the human genome remains to be established. Strikingly, a number of chromosomal abnormalities within this portion of chromosome 12, mainly translocations, have been found in a variety of human neoplasia, including melanoma, liposarcoma, lipoma, bladder carcinoma, hemangiopericytoma, lung lymphoma, uterine leiomyoma, myelodisplasia, acute non-lymphoblastic leukemia, hairy cell leukemia, fibrosarcoma, and breast tumors. One translocation (t(12; 22)(q13; q12)) (52) that is recurrently found in malignant melanoma of soft tissues (or clear cell sarcoma) has been shown to result in the fusion of the b-Zip domain of ATF-1 (12q13) to the N-terminal part of the RNA-binding protein EWS. Another translocation (t(12; 16)(q13; p11)) (53) has been characterized in myxoid liposarcoma, where the fulllength coding sequence of the CHOP gene (12q13) is linked to the sequences encoding the N-terminal portion of the FUS gene (also designated as TLS), which also encodes an RNA-binding protein (see Ref. 54 for a recent review). Finally, intragenic rearrangements of the HMGI-C gene (12q14 -15), which encodes a member of the high mobility group family of DNAbinding proteins, have been associated with the genesis of pulmonary chondroid hamartomas (55). It will be of interest to determine the molecular basis of the other observed malignant transformation phenotypes and to examine whether they are linked to alterations of the regulatory proteins encoded within this portion of chromosome 12. In any case, the multiplicity of alterations in the q12-14 region of chromosome 12 suggests that this region is particularly recombinogenic or mutation-prone.
The ATFa Isoforms: Structure and Function-Analysis of the ATFa genomic sequence revealed that two of the ATFa isoforms (ATFa1 and ATFa2) differ from the largest variant (ATFa3) by selection of alternative splice donor sites, one being located between elements D and d and the other between elements E and e (Fig. 3). None of these three variants exhibits transcriptional activity when assayed as intact proteins on adenoviral or artificial promoters (10,15), although some activity has been observed on the E-selectin promoter (16). In fact, deletion experiments revealed that a potent activation domain of the ATFa proteins, which is comprised within the 80 most N-terminal residues of the protein (10), is masked by the 100 most C-terminal residues, located beyond the b-Zip domain (10). The possibility therefore exists that the ATFa factors may  1-6). B, ϳ36 h following transfection of these vectors into COS-7 cells, localization of the chimeric proteins was assessed by immunofluorescence microscopy. Bars (20 m) refer to the four upper and lower micrographs, respectively. C, the sequence element (residues 327-377 of ATFa3) exhibiting nuclear localization properties is shown, with conserved NLS motifs underlined (38,60). The basic region and first leucine of the b-Zip element are shaded.
be activated in the cell upon unmasking of this activation domain by specific protein interactions.
No functional difference between ATFa1, ATFa2, and ATFa3 has so far been detected despite the rather peculiar amino acid composition of the peptides encoded by the d (rather basic, ARSRTVAKKLV) and e (Ser/Pro-rich, VDSSPPDSPASSPC-SPPLKEK) alternative elements as deduced from their nucleotide sequence (Fig. 3C). All three ATFa variants have previously been shown (i) to mediate the E1a responsiveness of specific genes by recruiting the E1a product to the corresponding promoters (15), (ii) to heterodimerize with members of the Jun/Fos family and thereby to modulate their DNA binding and trans-activation properties (10), and (iii) to strongly interact with a protein kinase closely related to JNK2 (18).
ATFa0, another variant of the ATFa family, is similar to ATFa2, but lacks residues 144 -320. This variant has been reported to act as a dominant inhibitor on the E-selectin promoter (16). Although this result is difficult to explain on the basis of the ATFa functional organization (see above), it may be related to potential interactions between ATFa0 and NF-B on this particular promoter (56).
The ATFa NLS Domain-Deletion and fusion protein analyses allowed the identification of the NLS element involved in the nuclear transportation of the ATFa proteins. This element, located between residues 327 and 368 (see Fig. 7C), appears to be composed of two subdomains, which cooperate in ATFa nuclearization. One is located within the basic region of the b-Zip domain (RNR----RCR-KRK) and is found at similar positions in several members of the AP1 family, including c-Jun, c-Fos, and the Epstein-Barr viral factor Zta (57,58). Interestingly, the invariant cysteine of this element has been suggested to participate in a redox-mediated control of AP1 activity (57,59). The other element, which is less conserved, is situated just N-terminal to the basic region, and its structure follows the rule initially proposed (60). This latter element, which is by itself unable to direct a heterologous protein to the nucleus, increases the efficiency of the second element. Similar sequences are also found in members of the Fos family (c-Fos, Fra1, and Fra2) and in the Zta protein, but not in c-Jun (58,61).
Tissue-specific Expression of ATFa-ATFa transcripts were detected at low levels in all human cell lines and mouse organs tested (Fig. 9A). Furthermore, ATFa expression was confirmed at the protein level (Fig. 9B) in mouse embryo (from stage 11.5 days post-coitum onward) and in most adult organs studied. The results of in situ hybridization studies (Fig. 10) were in close accordance with the low level ubiquitous pattern of expression of ATFa. However, expression above the basal level was observed in some tissues of both embryos and adult animals, most particularly in squamous (stomach and vagina) or pseudostratified (epididymis, ductus deferens, and seminal vesicle) epithelia and in specific areas of the brain (Purkinje cells, hippocampus, dentate gyrus, and piriform cortex). Together, these results suggest that the ATFa proteins could play a critical role during embryogenesis and in adult organ function, although their precise contribution remains unclear. The preferential expression of ATFa in specific epithelial structures may explain the tropism of adenovirus infection that is selectively directed toward epithelial cells of the respiratory tract.
Ubiquitous expression has also been noticed for the related CRE-BP1/ATF-2 protein, with increased levels in the central nervous system (41). Interestingly, the strongest expression was observed in the hippocampus and in the dentate gyrus, as in the case of the ATFa proteins, suggesting that both types of factors are involved in signal transduction in brain. Previous studies (62) showed that members of the Jun family are expressed in the piriform cortex and the hippocampus (c-Jun, JunB, and JunD) and in the Purkinje cell layer of the cerebellum (JunB). Such a colocalization of Jun and ATFa proteins is interesting, particularly in view of their capacity to heterodimerize (10,17).
Homozygous mutations have been introduced in the genes encoding the related CREB (63) and ATF-2 (64) proteins, with drastically different phenotypes: deficiency in long-term FIG. 9. Expression of the ATFa gene. A, Northern blot analysis (as described in the legend to Fig. 8) of ATFa-specific transcripts was carried out (see "Experimental Procedures") with 20 g of poly(A) ϩ RNA from the series of transformed human cell lines (lanes 1-8) and mouse organs (lanes 9 -12) listed. The same blot was subsequently probed with a human actin probe, used as an internal control (lanes 1-8). For lanes 9 -12, the actin level was checked separately and was found to be constant (data not shown). B, Western blot analysis of extracts (20 g) prepared from staged mouse embryo (between 11.5 and 17.5 days of development; lanes 1-6) and various adult mouse organs (lanes 7-14) was carried out as described under "Experimental Procedures." As a positive control for the antibodies, an extract from 293 cells that had been infected with a recombinant vaccinia virus expressing the ATFa1 protein (v-ATFa1) was run and probed in parallel (lane 15). Molecular size markers (in kDa) are shown on the right.
FIG. 8. Relative abundance of transcripts from selected members of the ATF family. HeLa cell poly(A) ϩ RNA (10 and 50 g in oddand even-numbered lanes, respectively) was separated by formaldehyde-agarose gel electrophoresis, transferred to nitrocellulose, and hybridized in 50% formamide with 5Ј-end-labeled (ϳ0.5 ϫ 10 6 cpm/pmol) synthetic 30-nucleotide-long oligonucleotides specific to ATFa, ATF-1, CRE-BP1/ATF-2, and CREB. After incubation at 37°C overnight, the blots were extensively washed in 0.1 ϫ SSPE and 0.1% SDS at 25°C and exposed for 4 days. The positions of DNA size markers (in kb) run in parallel are indicated on the right. memory for the CREB mutants and skeletal and central nervous system developmental abnormalities in the case of the ATF-2-disrupted mice. These results, indicating that particular members of the ATF family have specific effects, suggest that defects in these proteins are not or are only partially compensated by other members of the family. Using this approach, discrimination between each ATFa variant might perhaps be possible.
The ATFa Promoter Region-Examination of the 1.9-kb 5Јflanking sequences of the ATFa gene showed putative target sites for several transcription regulators, which included Sp1, AP1, AP2, PEA3, E2F, Myc, and NF-B. Transfection experiments in HeLa cells revealed an ϳ3-fold stimulation of ATFa promoter activity by c-Jun and c-Fos expression vectors (data not shown). In agreement with these results, 12-O-tetradecanoylphorbol-13-acetate or UV light treatments of HeLa cells resulted in similar levels of ATFa promoter activation, suggesting that AP1 may contribute, at least to some extent, to the control of ATFa expression. Our observation that expression of c-Ets or v-Ets vectors had no significant effects on ATFa promoter activity (data not shown) ruled out any major involvement of the PEA3 recognition element. In contrast to the promoter region of the ATF-3 gene (65), which harbors a consensus TATA box and an ATF/CRE site, the ATFa promoter exhibits neither of these elements. It is therefore unlikely that ATFa expression would be synergistically stimulated by cotransfection of c-Jun and ATF-2 vectors, as is the case for ATF-3 (65). By contrast, in keeping with their similar expression pattern, the ATFa promoter region more closely resembles that of the CRE-BP1/ATF-2 gene, which contains no canonical TATA box, but several Sp1-binding sites (66). Further experiments, including site-directed mutagenesis of specific elements, will help to clarify the physiological relevance of each of the potential ATFa promoter-binding sites.