Multiple transcripts for rat nucleoside diphosphate kinase alpha isoform are structurally categorized into two groups that exhibit cell-specific expression and distinct translation potential.

Rat nucleoside diphosphate (NDP) kinase is composed of two isoforms (α and β) encoded by independent genes. The mRNAs are expressed ubiquitously; however, the level of expression is tissue-dependent and is also up- or down-regulated under certain conditions, including growth stimulation, differentiation, and tumor metastasis. To address the regulatory mechanisms of gene expression for the rat NDP kinase major isoform α (an nm23-H2/PuF homologue), we identified the transcription initiation sites in detail by RNase protection and 5′-rapid amplification of DNA ends and located the core promoter region by chloramphenicol acetyltransferase assay. The transcripts, initiated from an extraordinarily wide range of sites, were categorized into two groups; one transcribed from an upstream region was spliced in the untranslated region (group 1), whereas the other initiated in the downstream region was not (group 2). RNase protection demonstrated that the group 1 mRNA was the dominant form present in all tissues except heart and skeletal muscle. In situ hybridization revealed cell-specific expression of these mRNA species. Furthermore, they differed in the translational efficiency (the group 2 α > β > the group 1 α). These findings suggest that the regulation of the NDP kinase expression at both transcriptional and posttranscriptional steps could be fundamentally governed by the selection of transcription initiation sites.

Rat nucleoside diphosphate (NDP) kinase is composed of two isoforms (␣ and ␤) encoded by independent genes. The mRNAs are expressed ubiquitously; however, the level of expression is tissue-dependent and is also up-or down-regulated under certain conditions, including growth stimulation, differentiation, and tumor metastasis. To address the regulatory mechanisms of gene expression for the rat NDP kinase major isoform ␣ (an nm23-H2/PuF homologue), we identified the transcription initiation sites in detail by RNase protection and 5-rapid amplification of DNA ends and located the core promoter region by chloramphenicol acetyltransferase assay. The transcripts, initiated from an extraordinarily wide range of sites, were categorized into two groups; one transcribed from an upstream region was spliced in the untranslated region (group 1), whereas the other initiated in the downstream region was not (group 2). RNase protection demonstrated that the group 1 mRNA was the dominant form present in all tissues except heart and skeletal muscle. In situ hybridization revealed cell-specific expression of these mRNA species. Furthermore, they differed in the translational efficiency (the group 2 ␣ > ␤ > the group 1 ␣). These findings suggest that the regulation of the NDP kinase expression at both transcriptional and posttranscriptional steps could be fundamentally governed by the selection of transcription initiation sites.
Nucleoside diphosphate (NDP) 1 kinase (EC 2.7.4.6) plays a pivotal role in maintaining intracellular levels of triphosphate nucleotides at the expense of ATP. It had been thought to be a typical housekeeping enzyme (1), but recent studies have revealed that the NDP kinase protein may have multiple regulatory functions beside the phosphotransferase activity and that it behaves as a tumor metastasis suppressor (nm23; Refs. 2 and 3, reviewed in Ref. 4), a differentiation-inhibiting factor (5), and a transcription factor for human c-myc (PuF; Ref. 6). These functions are reportedly unrelated to NDP kinase en-zyme activity per se (7)(8)(9). NDP kinase forms a gene family in various organisms. In mammalia, two isoforms of NDP kinase have been reported from humans, mice, and rats (10 -13). The rat isoforms, ␣ and ␤, are encoded by distinct genes arranged in a tandem array (13). The two NDP kinase isoforms display only small functional differences in their activity as phosphotransferases to the extent examined (14). However, recent studies have demonstrated that each of the two isoforms (␣ and ␤) may possess its own specific functions in addition to their common enzymatic properties: they are distinguished on the basis of one being a homologue to the transcription factor for the human c-myc gene (PuF/nm23-H2) and the other the candidate tumor metastasis suppressor (nm23-H1).
NDP kinase expression is reportedly increased at specific stages of life of several organisms and in various types of cells under certain circumstances; for example, during formation of the imaginal discs at the larval stage of Drosophila melanogaster (15), in murine systemic organs at their organogenesis stages (16), in concanavalin A-stimulated human T lymphocytes (17), and in human diploid fibroblasts when immortalized by SV40 transformation and 60 Co irradiation (18). Decreased expression of NDP kinase has been reported also in the slime mold Dictyostelium discoideum when aggregation and development of the multicellular organization was triggered by starvation (19) and in the case of tumor metastases in experimental animal systems and in certain human tumors (3,4).
Beginning at the very early stage of the cloning and characterization of the rat NDP kinase mRNA, it was noticed that the quantities of mRNA are not directly related to those of the protein or to enzymatic activity: the NDP kinase mRNA levels vary more than 10-fold among different organs tested, whereas levels of the protein or enzymatic activity vary at most by 2-fold (20). Similar enigmatic phenomena were also observed with cell immortalization: the mRNA was increased severalfold, whereas the protein amount only increased at most 1.5-fold (18). These observations may suggest possible posttranscriptional regulations of the NDP kinase genes. Although several factors (or motifs) in mRNA have been reported as regulatory elements responsible for the posttranscriptional regulation of gene expression (reviewed in Refs. 21 and 22), such structural analyses of NDP kinase mRNA are totally lacking.
Previously we reported genomic structure and transcription start sites for the major isoform (␣ isoform) of rat NDP kinase and assigned four exons (23). Since then we have found another type of cDNA clone coding for the ␣ isoform, which is composed of five exons, including an additional 5Ј-untranslated exon, and spliced differently, which we have termed the long form (now renamed as group 1 type) mRNA, and the previously reported type of the mRNA is termed the short form (now renamed as group 2 type) (13). These observations have raised questions as to the relative abundance of the two types of ␣ isoform transcripts, their exact transcription initiation sites, and regulatory mechanism of their expression.
To address these questions we sought to identify the transcription start sites for rat NDP kinase ␣ isoform mRNA in detail by RNase protection in combination with the 5Ј-RACE method, and have identified numerous initiation sites. Furthermore, RNase protection assay and in situ hybridization analyses have revealed the tissue-and cell-specific expression of these mRNA species. These data, together with CAT assay data, which show a core promoter region for the ␣ gene ( Fig. 1), may provide a perspective of the complicated transcriptional regulatory mechanisms of the NDP kinase genes. A possible mechanism of the posttranscriptional regulation for NDP kinase expression is discussed based on the finding that the rates at which the transcripts are translated under in vitro conditions are different among the heterogeneous forms.

RNA Isolation
Systemic organs were harvested from Wistar rats aged between 12 and 16 weeks. Total RNA was isolated by the acid-phenol extraction method (24) with modifications. Briefly, samples were extracted twice in a premixed phenol and guanidinium isothiocyanate solution (ISO-GEN; Nippon Gene Inc., Tokyo, Japan) and chloroform. Poly(A) ϩ RNA was further purified using an oligo(dT)-cellulose spin column (Pharmacia Biotech Inc.).

Ribonuclease Protection Assay
RNase protection assays were performed as described previously with modifications (23). Briefly, four RNA probes were synthesized. To identify transcription initiation sites we made two probes; ␣-l and ␣-s, which correspond to the rat NDP kinase ␣ gene from Ϫ209 to Ϫ453 and from Ϫ9 to Ϫ204, respectively (see Fig. 2). To quantify the two subtypes of the ␣ mRNA, group 1 and group 2, two probes were generated from two RACE clones; ␣-1 is generated from clone 35A and complementary to the 39 nucleotides of the 3Ј-end of the first exon (Ϫ244 to Ϫ206) and 46 nucleotides of the 5Ј-end of the second exon (Ϫ4 to ϩ42), and ␣-2 is generated from a group 2 type clone and corresponding to a fragment 70 nucleotides in length (Ϫ28 to ϩ42) (see Fig. 4). The numbering system used in this study to identify the nucleotide location starts with the translation initiation site (23). Full-length synthetic RNAs that represented group 1 and group 2 types of ␣ and ␤ mRNA were used as control samples. The processed materials were electrophoresed in denaturing gels supplemented with 40% formamide.

5Ј-RACE Method
To ensure the ␣ isoform-specific cDNA synthesis, we used primer 4S (5Ј-CCGAAGGAACTTCATGG-3Ј, ϩ126 to ϩ110), corresponding to the 3Ј-terminal stretch of the second exon of the ␣ gene, where the two homologous (␣ and ␤) genes denote the most divergent sequences between them. To obtain full-length cDNA in high efficiency, we applied the two-step reverse transcription reaction. Briefly, 2.5 g of poly(A) ϩ RNA and 2 mM primer in water were heated at 85°C for 10 min and then annealed at 65°C for 15 min, followed by the addition of reverse transcription buffer and reverse transcriptase (SuperScript II; Life Technologies, Inc.) and ribonuclease inhibitor (RNasin, Promega), and then incubated at 50°C for 45 min. After heat denaturation at 85°C for 10 min, a second extension reaction was performed by further addition of fresh reverse transcriptase. Next, excess amounts of RNA and the single-stranded portion of RNA were eliminated by ribonuclease A and T1 (Ambion) digestion, and the resulting degraded RNA and excess amounts of primer were removed by an S-300 MicroSpin column (Pharmacia), which reportedly adsorbs single-stranded RNA and synthetic oligonucleotides (25). The samples enriched with heteroduplex forms of RNA and cDNA were treated with ribonuclease H (Life Technologies), purified by a Sephadex G-50 spin column (Boehringer Manheim) and then ethanol precipitation. To make templates for PCR, we adopted the single strand ligation to single-stranded cDNA method (26) using a 5Ј-AmpliFINDER RACE kit (Clontech) following the manufacturer's instruction. The RACE methods (27) were carried out with Taq and/or Pfu (Stratagene) polymerases. Representative conditions were as follows. The reaction mixture contained 2 mM each of an anchor-specific primer (5Ј-CCTCTGAAGGTTCCAGAATCGATAG-3Ј) and gene-specific primer 2 (5Ј-ATCTGGCTTGATGGCAATGAAGGTAC-3Ј, ϩ42 to ϩ17), or 6L (5Ј-AGAAGCAAGAAGTGTAGTCGATG-3Ј, Ϫ9 to Ϫ31), 10% Me 2 SO, 0.2 mM dNTP, and 2.5 units of Pfu polymerase in a low magnesium buffer (Idaho Technology). Ten l of the mixture in a glass tube was incubated at 97°C for 60 s, 52°C for 10 s, 75°C for 120 s for 40 cycles, and additionally at 75°C for 10 min in an Idaho Technology 1605 air thermocycler. A 5-l aliquot of each reaction was applied to a 2.5% agarose gel (see Fig. 3, A and B). The residual 1-l aliquot was used as a template for a second PCR in the same reaction mixture described above but containing Taq instead of Pfu polymerase. The second reaction was carried out at 95°C for 60 s, 52°C for 10 s, 72°C for 120 s for 10 cycles, and additionally at 72°C for 10 min. The PCR products were subjected to direct subcloning using a TA cloning kit (Invitrogen). The established clones were sequenced using a Sequenase sequencing kit (U. S. Biochemical Corp.).

Plasmid Constructions
Full-length cDNAs-The full-length cDNA clones were constructed with the 5Ј-RACE products described above and cDNA clones for the ␣ isoform previously reported (20). We used two RACE clones, 50B and 28H, as representatives of group 1 type ␣ and group 2 type ␣, respectively. They were cut out from the vectors by EcoRI and then digested with XhoI. The cap site to XhoI site fragments from the RACE products and the XhoI site to poly(A) fragments from cDNA were ligated. The reconstructed fragments were subcloned into pBluescripts (Stratagene) and/or pGEM (Promega) plasmids. A full-length cDNA for the ␤ isoform was also constructed and used for a control (13; data not shown).
Reporters-Several expression plasmids were constructed in the CAT gene containing plasmid pKK232-8 (Pharmacia). Six sense primers, 239 (5Ј-GAAGGATCCGGGTACCCCAGAGCAGAGAGT-3Ј), 238 (5Ј-GAGAGGATCCCAGGGAAAGGTGAATGCAGATG-3Ј), 237 (5Ј-GAGA-GGATCCTGCCTCACAGCCCTCCGT-3Ј), 227 (5Ј-GAGAGTCGATC-GCTCTCCGCTGGCACCAGCC-3Ј), 236 (5Ј-CTCAGG(C 3 A)TCCC-GCGGTCTCCTTTC-3Ј), and 228 (5Ј-AAA(T 3 G)ATCTGGAAAGCC-ACGTGTGTCCT-3Ј), and one antisense primer, 235 (5Ј-GAGAAGCT-TGCAGAAGCAAGAAGTGTAGTCGA-3Ј), were used to amplify seven overlapping genomic fragments of the 5Ј-regulatory region for the ␣ isoform. To subclone the PCR fragments into the BamHI-HindIII cloning site of the CAT vector, underlined sequences were added to or altered from gene-specific fragments to generate appropriate restriction enzyme sites denoted by boldface letters. Standard PCRs were performed to amplify the gene fragments, and then the products were digested with BamHI or Sau3AI or BglII and HindIII and then sub-FIG. 1. Nucleotide sequence of the rat NDP kinase ␣ isoform gene promoter region. The 5Ј-upstream region relative to the translation initiation site is expressed by negative numbers. Putative Sp1binding sequences and E box sequences are boxed by solid and dotted lines, respectively. The primers used are indicated by arrows. The transcription initiation sites determined by the 5Ј-RACE clones, which start from a G residue, are denoted by arrowheads, whereas clones in parentheses indicate those with ambiguity as to whether the G residues of the 5Ј-end were derived from the cap G or genomic G residue. Letters added to each clone's name: A, aorta; B, brain; H, heart; T, testis. cloned into pKK232-8.

In Situ Hybridization
Rat organ specimens embedded in Optimum Cutting Temperature compound (Miles Laboratories) were frozen in liquid nitrogen. In situ hybridization of frozen sections (8 m) was performed following the procedure described elsewhere (28). We used three kinds of riboprobes labeled with digoxigenin-11-UTP (Boehringer Mannheim). They were complementary to: 1) the 5Ј-untranslated region of the group 1 type ␣ mRNA (at position Ϫ206 to Ϫ384); 2) the 5Ј-untranslated region of the group 2 type ␣ mRNA (at position Ϫ4 to Ϫ85); and 3) the 5Ј-untranslated region for ␤ mRNA. The signals were detected using a digoxigenin detection kit (Boehringer Mannheim).

Cell Lines, Transfection, and CAT Assay
Rat fibroblast cell line 3Y1 and an osteosarcoma cell line, UMR106, (29) were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Transfections were performed by the lipofection method using N- [1-(2,3-dioleoxyloxy)propyl]-N,N,N-trimethylammonium methylsulfate (Boehringer Manheim), following the manufacturer's instruction. Briefly, cells at a 50 -60% confluent state in a dish 10 cm in diameter were transfected with a total of 5 g of DNA and incubated for 12 h, and then the medium was replaced with fresh medium and incubated for a further 48 h. Cells were harvested, the cell lysates were adjusted to equivalent protein concentrations, and CAT activity was measured following a standard method (30). The Rous sarcoma virus-CAT plasmid and purified CAT enzyme were used for positive controls. The conversion rate was calculated by using Quantityone software (PDI Imageware Systems, Huntington Station, NY).

In Vitro Translation Assay
Synthetic RNAs were generated from the template plasmids containing either full-length cDNA described above. These synthetic RNAs were capped by using the mRNA capping kit (mCAP TM ; Stratagene), and then 1 g of each synthetic RNA was subjected to an in vitro translation assay using the ECL in vitro translation system (Amersham Corp.). SDS-polyacrylamide gel electrophoresis aliquots of the biotinylated translation products were electrophoretically blotted onto a polyvinylidene difluoride membrane (Bio-Rad) and detected using streptavidin conjugated with horseradish peroxidase and an ECL reagent (Amersham). Densitometric quantification was carried out by using PDQuest software (PDI).

Reagents and Chemicals
Unless otherwise indicated, restriction enzymes and modified enzymes used in this study were purchased from Toyobo (Osaka, Japan).

Numerous Transcription Initiation Sites for the ␣ Isoform
Are Distributed in an Extraordinarily Wide Range-To determine the transcription initiation sites of the two types of ␣ isoform mRNA, we performed RNase protection assays using two kinds of synthetic RNA probes, ␣-l and ␣-s. When we used ␣-l, which is complementary to the first exon and a further upstream region of the rat NDP kinase ␣ gene from Ϫ209 to Ϫ453, extraordinarily numerous protected bands were observed between the positions 185 and 30 judging from the DNA sequencing ladder (Fig. 2A). The major protected bands migrated between positions 185 and 150, 100 and 85, and 60 and 30. On the other hand, the ␣-s probe, which is complementary to the first intron of the gene from Ϫ9 to Ϫ204, again provided multiple protected bands between positions 100 and 40 (Fig.  2B). Interestingly, protected band patterns of the ␣-l probe by the total RNA from the liver and heart were essentially identical, although some subtle differences of the signal intensity were observed ( Fig. 2A, lanes 1 and 2). The probe ␣-s was also protected by the liver total RNA in the same way as observed for the heart RNA (Fig. 2B, lane 1) with a weaker intensity (data not shown). Under the reaction conditions used, the positive control had a single dominant band, and the negative control had no measurable background, leading to the conclusion that reaction conditions were optimized. However, there existed slight ambiguities in the data obtained; synthetic RNA samples used as positive controls provided longer protected fragments than expected by 6 bases. Such differences could be ascribed in part to the fact that each full-length synthetic RNA possessed an artificially added linker sequence at the 5Ј-end of the genuine sequence. In addition to that possibility, RNA reportedly has slightly slower mobility compared with a DNA molecule of similar size. Thus we tentatively speculate that the major transcription initiation sites are located between positions Ϫ50 and Ϫ105, Ϫ240 and Ϫ270, Ϫ295 and Ϫ315, and Ϫ365 and Ϫ390. To confirm the data, we used other methods as described below.
5Ј-RACE Determined Numerous Capped mRNAs, Which Are Initiated in a Wide Range, and Defined a Splice Site for the ␣ mRNA Initiated in the Upstream Region-To identify the transcription initiation site of the rat NDP kinase ␣ isoform, the 5Ј-RACE method was performed. When we compared amplification efficiency of Pfu with the most conventional thermostable polymerase, Taq, under the same conditions, Pfu generated longer products with higher efficiency than Taq (Fig. 3A). Therefore, we performed PCR using Pfu under relatively high stringent conditions thereafter. Representative crude PCR products obtained with primer 2 and anchor primers are shown in Fig. 3B. Considering the anchor primer size (25 mer), the PCR by primer 6L and the anchor primer generated a broad band ranging from 50 to 130 bp and around 200 bp in net length, and PCR by primer 2 and the anchor primer produced a broad band from around 100 to 250-bp. These profiles of the gross PCR products generated with Pfu polymerase agreed well with the gross pattern of the transcription start sites predicted by the RNase protection assays. A number of the PCR products generated with primer 2 were subcloned and verified by sequencing. All of the sequenced clones were found to code for the genuine NDP kinase gene segments without exception. Interestingly, most of the clones appeared to have an additional guanine nucleotide at their 5Ј-ends. Representative clones are shown in Fig. 3, C and D. Recently, Hirzmann et al. (31) have reported that the reverse transcriptase can occasionally read the 5Ј-cap structure as G during cDNA synthesis, resulting in the addition of a G residue to about half of their RACE clones (31). In our study, 29 of 34 (85%) informative clones possessed a G residue at the 5Ј-end. Furthermore, 21 of the 29 clones had an unpaired G residue, suggesting that these clones are generated from the full-length RNA templates with the cap-G structure (summarized in Fig. 1). The higher incidence of the reverse transcription of the cap-G residue in our experiments reflects not only high integrity of our template mRNA but also reliability of the techniques we used. The location of the 5Јtermini of the NDP kinase ␣ mRNA identified by the RACE method agreed with most of the putative major transcription initiation sites suggested by the RNase protection assay within a couple of nucleotides (e.g. Ϫ50, Ϫ58, Ϫ78, Ϫ85, Ϫ250, Ϫ252, Ϫ257, Ϫ290, Ϫ299, Ϫ300, Ϫ366, Ϫ368, Ϫ380, and Ϫ384). Accumulated data of the RACE clones further verified the existence of the two different types of the transcripts for the NDP kinase ␣ isoform. All the clones except one (clone 20H), which started further upstream than position Ϫ241 (total, 12 clones), were spliced at Ϫ206 and accepted at Ϫ4 and categorized as the group 1 type mRNA. On the contrary, all clones initiated at any points downstream from position Ϫ206 were not spliced and bore various sizes of the 5Ј-untranslated stretch continuing from the cap sites to the identical translation initiation site at position ϩ1. These clones were categorized as the group 2 type. The splicing donor and acceptor sites do not merely obey the Chambon rule, but the adjacent sequence fits conservative consensus sequences (32). The exceptional clone 20H could have been generated from a nascent unprocessed mRNA template.
In the course of this study we could not find any typical TATA boxes but found heavily GC-rich stretches alongside of the expected binding sites of the basic transcriptional machinery.
The Group 1 and 2 Types of ␣ mRNA Are Expressed in Different Manners-Considering the huge heterogeneity of the transcription initiation sites for both the group 1 and group 2 type ␣ mRNA, it would be very difficult to quantify them as they are. Therefore, we tried to evaluate the two groups of ␣ isoform transcripts by RNase protection assay using relatively short RNA probes, which were expected to cover most of the transcripts at their common trunks. The RNA probes ␣-1 and ␣-2, schematized in Fig. 4, were accordingly used. When the probe ␣-1 is used, the group 1 and group 2 type ␣ mRNA are expected to produce 85-nucleotide (Fr 1) and 46-nucleotide (Fr 2) protected fragments, respectively. Synthetic standard RNA samples representing the group 1 and group 2 mRNA actually FIG. 3. Determination of rat NDP kinase ␣ mRNA 5-end by the RACE method. A, each cDNA synthesized from cerebrum and testis was subjected to the RACE protocol by the anchor (A. Primer) and 6L primers. PCR reactions were performed using Taq or Pfu polymerase, and then the reaction products were analyzed by electrophoresis, followed by ethidium bromide staining. Marker lanes contain a 100-bp ladder. The cDNA sources and the polymerase used are shown at the top. B, PCR with the anchor primer and primer 2 was performed using heart-derived cDNA and then applied to an agarose gel. C, nucleotide sequence of a representative RACE clone, 37A, for the group 1 type ␣ NDP kinase by primer 125. ‫,ء‬ extra G residue between the linker and cDNA sequences. On the right side, the corresponding genomic sequence is shown for comparison. D, nucleotide sequence of a representative RACE clone, 23H, for the group 2 type ␣ NDP kinase by primer 6L. An extra G and the corresponding genomic sequence are denoted as in C. produced such protected fragments, although the protected sizes were somewhat longer than the expected ones, due to complementarity of the adjacent vector sequences to the probe. Total RNA samples extracted from adult rat tissues, including cerebrum, spleen, heart, lung, liver, kidney, small intestine, testis, and skeletal muscle, and two cell lines, UMR106 and 3Y1, were examined and found to produce two to three major protected bands that were concordant with the signals corresponding to the Fr 1 and Fr 2 of the standard samples. The Fr 1 signals corresponding to the group 1 type ␣ mRNA were constantly expressed in these samples except those from the cell lines, in which the signal was extremely increased. Of interest was the finding that a decreased amount of the group 1 mRNA, the dominant form, was observed in a highly metastatic rat mammary adenocarcinoma cell line (MTLn3) compared with a low metastatic sib line (MTC) (data not shown; see ref. 33). Although the signal corresponding to the group 2 mRNA (Fr 2) was strong in the heart and skeletal muscle, it was weak in the liver and kidney and in only trace amounts in other samples, including cultured cell lines (Fig. 4A). Similar experiments were performed using probe ␣-2 (Fig. 4B) and provided "mirror image" protected fragments compared with those done using probe ␣-1. Furthermore, this probe demonstrated that there exist no remarkable initiation sites in the proximal region between position Ϫ24 and Ϫ4. Thus, the previously reported predominant transcription start site (position Ϫ3 in ref. 23) should be redefined as a splicing acceptor site (correctly assigned to position Ϫ4) of the second exon for the group 1 type mRNA.

In Situ Hybridization Demonstrated That Choice of the Transcription Initiation Sites and Their Quantity Are Regulated in
Cell-specific Manners-To analyze the expression of these transcripts at the cell level, we performed in situ hybridization using specific probes for group 1 type ␣, group 2 type ␣, and ␤. Generally, the group 1 type ␣ was ubiquitously expressed, and the expression was more intense than for the group 2 type ␣, as observed by the RNase protection analyses. It is worth noting that both types of ␣ mRNA coexisted in some kinds of cells, whereas there was no cell type that exhibited strong group 2 mRNA expression alone. One of the representative cases, the stomach of adult rat, is shown in Fig. 5. A large amount of the group 1 type ␣ mRNA was expressed in both gastric pit epithelial cells and fundic gland cells, whereas the group 2 type ␣ mRNA was expressed strongly only in the chief cells in the fundic glands, the pattern being similar to that for the ␤ mRNA.
Universally Active Core Promoter Region of the ␣ Isoform Gene-To identify promoter activity of the ␣ isoform gene, we made several CAT constructs (schematized in Fig. 6) and transfected them into UMR106 cells. The strongest CAT activity was observed in the cell lysates prepared from the cells transfected with CAT constructs 7 and 6. A reduced but countable amount of the activity was detected in the cell lysate transfected with plasmid 5, and only a trace amount was detected in the lysate from the transformant with construct 4. No significant CAT activity was detected in the lysates of the cells transfected with plasmids containing shorter genomic fragments than clone 3. Essentially similar results were obtained when we used a fibroblast cell line, 3Y1, as a host (data not shown). The data indicate that one of the strongest core promoter activities seems to reside in the region between Ϫ568 and Ϫ452, where the most distal putative Sp1 binding consensus sequence GT box, an aberrant form of the GC box, is located (34). Weak but significant promoter activity may be present in the region between Ϫ452 and Ϫ366 in which another GC box is included. The proximal region from position Ϫ366 has no significant promoter activity in these cells. It should be noted that the region (between Ϫ568 and Ϫ452) that provided the strongest promoter activity corresponds to the most distal universally active transcription initiation sites. However, this region might be too far from downstream initiation sites for the group 2 type ␣ mRNA to operate. Thus, it could be reasonable to postulate another promoter region for the downstream transcription initiation sites. Unfortunately, however, because of the trace amount of the group 2 type ␣ transcripts in the cells used as the host (Fig. 4), it would be very difficult to detect downstream FIG. 5. In situ hybridization analysis of NDP kinase mRNA expression in rat gastric mucosa. Photomicrographs were taken of serial thin sections of stomach (pars glandularis) derived from a 9-month-old Wistar rat. The hybridization was performed using three kinds of RNA probes, which are complementary to the group 1 type ␣ 5Ј-untranslated region (A), the group 2 type ␣ 5Ј-untranslated region (B), and the ␤ untranslated region (C). GP, gastric pit; FG, fundic glands. ϫ125. promoter activities in these cells.
A Group 2 type ␣ mRNA Is More Efficiently Translated than a Group 1 Type ␣ mRNA and ␤ mRNA-Although the numerous different forms of transcripts for the ␣ isoform have been defined, the physiological meaning of them is largely unknown. We have examined several possibilities, including structural polymorphism at the peptide level for both types of the ␣ gene products. But so far these attempts have been unsuccessful, except in vitro translation analyses. The in vitro translation analysis using synthetic RNAs generated from each representative of the group 1 and 2 types of ␣ cDNA clones and the ␤ cDNA clone revealed remarkable differences in translation efficiency: the group 2 type ␣ RNA was most effectively translated among them, the group 1 type ␣ RNA was minimal; and the ␤ RNA was intermediate. Relative efficiency rates were approximately 100, 14, and 36%, respectively (Fig. 7). DISCUSSION The prevailing notion that NDP kinase is an essential enzyme for nucleotide metabolism in the cell has been proved in recent studies that have included the cloning of cDNA, in which transcripts coding for the enzyme from various organisms were found to be highly conserved. The compiled data promptly led to the realization that a single ancestral gene of NDP kinase could have been conserved from unicellular organisms to mammalian species. On the other hand, the discovery of nm23 (2) and its identification as an NDP kinase have opened a new era in NDP kinase research. As a consequence of the findings that the NDP kinase molecules may carry out multiple functions besides the enzymatic activity, as in the case of PuF (6,9) and inhibiting factor (5,8), it becomes apparent that NDP kinase may play a pivotal, multifunctional role in various organisms. Furthermore, based on compiled observations, it can be envisioned that the expression of NDP kinase is governed by binary regulatory mechanisms, that is, inducible and constitutive ones.
To get a bird's eye view of the NDP kinase genes in terms of evolutional divergence of their organization and function, we examined the exon-intron organization of the genes of various origins as a genomic evolution parameter. The intron numbers have increased as the species evolution has proceeded: 1) unicellular organisms such as Escherichia coli and Saccharomyces cerevisiae possess one undivided gene (35,36); 2) D. melanogaster also has a single but separated gene that is divided by an intron (37,38); 3) a slime mold D. discoideum bears two genes coding for cytosolic and mitochondrial type enzymes, both of which have two intervening sequences at identical locations with an additional two introns for the latter gene (39,40); and 4) the mammalian species examined have two isoforms of NDP kinase. The two rat genes have four introns at completely conserved locations (13). The partial human gene structures for one isoform (nm23-H1) reported from a couple of laboratories are similar but not identical (41,42); one shows exactly the same gene organization as the rat gene, whereas the other does not. Since the awd locus of the fruit fly is coding for a single NDP kinase gene that has one intronic sequence exactly corresponding to the location of the second intron of the rat genes, it is reasonable to suppose that the awd gene might represent an archetype of the mammalian NDP kinase gene, and duplication of the ancestral gene could have occurred after the arthropod and chordate ancestor's divergence. On the other hand, two NDP kinase genes of slime mold contain two intervening sequences, the locations of which are concordantly preserved between them but completely discordant with those of the mammalian species, making it likely that a single ancestral gene obtained species-specific intervening sequences after the phylum divergence from a common ancestor of the animal, then was duplicated and developed to the present forms.
Second, we compared CpG islands of the fly and rat NDP kinase genes to characterize the gross outline of the 5Ј-regulatory regions. The awd gene has a characteristic CpG island that covers continuously the 5Ј-flanking to the 5Ј-untranslated region (nine CpG islands in 115 bp) and the coding region (40 CpG islands in 459-bp). In the case of the rat ␣ gene, CG dinucleotides were distributed from the 5Ј-regulatory region to the second exon (50 CpG islands in 561 bp, between Ϫ435 and ϩ126). On the contrary, the ␤ gene has a sparsely distributed CpG cluster that covers the first exon and disappears in the 5Ј-portion of the first intron (data not shown). It follows, therefore, that the ␣ gene may be representative of the direct descendant of the archetype NDP kinase gene.
Although structures and regulatory mechanisms of NDP kinase genes of other mammalian species are largely unknown, comparison of the two human cDNA clones, nm23-H2 and PuF, allow us to speculate the existence of the two groups of transcripts as observed for the rat NDP kinase ␣ isoform gene: 1) nm23-H2 and PuF have the same coding sequence, demonstrating that they are products of an identical gene; 2) the PuF clone has an AG sequence, a candidate splicing acceptor site, at position Ϫ5, whereas the nm23-H2 clone does not have the dinucleotide and contains a completely different 5Ј-untranslated sequence; and 3) the PuF clone has multiple CG dinucleotides in the short 5Ј-untranslated stretch, which may represent part of a CpG island. These data suggest that the nm23-H2 clone may be representative of the rat group 1 type ␣ homologue, whereas the PuF clone may belong to the group 2 type ␣ homologue.
Heterogeneous transcription initiation can be interpreted mainly from two biological aspects; different transcription initiation sites: 1) make different translated products (peptides), and 2) are regulated under distinct elements. Regarding the rat NDP kinase isoforms, despite the huge heterogeneity of the 5Ј-regions of the transcripts, there exist stop codons between an upstream ATG and the authentic initiation codon of the adequate open reading frame of the enzyme, thus ruling out the former possibility. On the other hand, our RNase protection analyses demonstrated characteristic differences in the expression of the group 1 and 2 types of NDP kinase ␣ mRNA in tissues and cell lines. The upstream regulatory region may play an essential role in operating constitutive and predominant expression of the group 1 type ␣ mRNA in most cells, whereas the downstream regulatory region may be responsible for high expression of group 2 type ␣ mRNA in certain tissues or cells such as muscles. Furthermore, the observations that the group 1 type ␣ mRNA, the major component of the rat NDP kinase mRNA, was increased on immortalization of the cells (Fig. 4)   FIG. 7. In vitro translation analysis of the synthetic RNA for two groups of the ␣ and ␤ mRNA. From representative full-length clones for each group 1 and group 2 type ␣ and ␤ NDP kinase cDNA, three kinds of RNA were synthesized and subjected to an in vitro translation assay. An autoradiogram displays profiles of translated products. and decreased in a subline with higher metastatic potential (33) imply possible roles for the upstream region in tumorigenic processes. In situ hybridization analyses provided informative data in support of the idea of the basic mechanisms of the gene expression obtained by the RNase protection assay: 1) predominant expression of the group 1 mRNA was confirmed in most cells examined; 2) the group 2 mRNA was expressed in some of these cells; in other words, there were no cell types that exhibited the group 2 mRNA alone; and 3) it should be noted that in some cells the group 1 mRNA was almost exclusively expressed. Considering the CAT assay data (Fig. 6) in combination with these observations, the core promoter activity localized in the region between Ϫ568 and Ϫ452 might be necessary not only for the expression of the group 1 type transcripts but also essential for the expression of the group 2 type ones that initiate far downstream from the candidate general promoter region. It follows that, in cooperation with the universal core promoter activity, additional promoter activities could be generated by downstream cis-elements and trans-acting factors, such as putative E box consensus sequences (see Fig. 1) and the muscle-specific transcription factors (reviewed in Ref. 43) for the preferential induction of the group 2 transcripts in muscle cells.
Regarding the translational control mechanisms (reviewed in Refs. 21 and 22), there have been no definitive data indicating regulatory mechanisms of NDP kinase expression at the posttranscriptional steps. The data presented in this study suggest the possible biological significance of the multiple forms of the 5Ј-untranslated region; the group 1 type ␣ mRNA, which is constitutively expressed, may work for an "idling" state at a minimal translating rate, whereas the "induced" form (the group 2 type mRNA) could be used at a higher rate in case of need. Hence these differential translation rates seem to guarantee homeostasis in maintaining triphosphate nucleotide pools in response to different circumstances. The predicted conservation of these multiple forms of NDP kinase mRNA for the ␣ isoform among mammalian species could also give a rationale for its biological significance. Recently, the meaning of the multiple transcription initiation sites has been reported in the case of an E. coli enzyme, phosphoenolpyruvate (44). The heterogeneity of the 5Ј-untranslated region provides the mRNA with distinct stabilities; whereas the constitutively synthesized transcript is stable, the stimulus-induced one is rather labile. It follows that the two types of mRNA expression enable a quick adaptation by executing fine control of the enzyme amount, resulting in maintenance of metabolic homeostasis.
Although this study indicated that heterogeneity of the 5Јuntranslated region may determine the translational efficiency for rat NDP kinase mRNA, the question of how one or some of the multiple transcriptional initiation sites can be chosen in response to various milieus is crucially important for understanding the posttranscriptional as well as transcriptional regulation. We anticipate that our observations and concepts on NDP kinase gene expression will shed light not merely on the enzyme expression per se but also on the housekeeping enzymes in general.